godotengine
/
godot
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add FidelityFX Super Resolution 2.2 (FSR 2.2.1) support. 

Introduces support for FSR2 as a new upscaler option available from the project settings. Also introduces an specific render list for surfaces that require motion and the ability to derive motion vectors from depth buffer and camera motion.
This commit is contained in:
Dario 2023-09-22 18:38:02 -03:00
parent df0a822323
commit 057367bf4f
102 changed files with 22108 additions and 149 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
COPYRIGHT.txt
View File

@ -141,6 +141,11 @@ Comment: AMD FidelityFX Super Resolution
Copyright: 2021, Advanced Micro Devices, Inc.
License: Expat
Files: ./thirdparty/amd-fsr2/
Comment: AMD FidelityFX Super Resolution 2
Copyright: 2022-2023, Advanced Micro Devices, Inc.
License: Expat
Files: ./thirdparty/angle/
Comment: ANGLE
Copyright: 2018, The ANGLE Project Authors.

13
doc/classes/RenderSceneBuffersRD.xml
View File

@ -130,6 +130,19 @@
Returns the texture size of a given slice of a cached texture.
</description>
</method>
<method name="get_texture_slice_view">
<return type="RID" />
<param index="0" name="context" type="StringName" />
<param index="1" name="name" type="StringName" />
<param index="2" name="layer" type="int" />
<param index="3" name="mipmap" type="int" />
<param index="4" name="layers" type="int" />
<param index="5" name="mipmaps" type="int" />
<param index="6" name="view" type="RDTextureView" />
<description>
Returns a specific view of a slice (layer or mipmap) for a cached texture.
</description>
</method>
<method name="get_use_taa" qualifiers="const">
<return type="bool" />
<description>

8
doc/classes/RenderingServer.xml
View File

@ -4507,7 +4507,10 @@
<constant name="VIEWPORT_SCALING_3D_MODE_FSR" value="1" enum="ViewportScaling3DMode">
Use AMD FidelityFX Super Resolution 1.0 upscaling for the viewport's 3D buffer. The amount of scaling can be set using [member Viewport.scaling_3d_scale]. Values less than [code]1.0[/code] will be result in the viewport being upscaled using FSR. Values greater than [code]1.0[/code] are not supported and bilinear downsampling will be used instead. A value of [code]1.0[/code] disables scaling.
</constant>
<constant name="VIEWPORT_SCALING_3D_MODE_MAX" value="2" enum="ViewportScaling3DMode">
<constant name="VIEWPORT_SCALING_3D_MODE_FSR2" value="2" enum="ViewportScaling3DMode">
Use AMD FidelityFX Super Resolution 2.2 upscaling for the viewport's 3D buffer. The amount of scaling can be set using [member Viewport.scaling_3d_scale]. Values less than [code]1.0[/code] will be result in the viewport being upscaled using FSR2. Values greater than [code]1.0[/code] are not supported and bilinear downsampling will be used instead. A value of [code]1.0[/code] will use FSR2 at native resolution as a TAA solution.
</constant>
<constant name="VIEWPORT_SCALING_3D_MODE_MAX" value="3" enum="ViewportScaling3DMode">
Represents the size of the [enum ViewportScaling3DMode] enum.
</constant>
<constant name="VIEWPORT_UPDATE_DISABLED" value="0" enum="ViewportUpdateMode">
@ -4708,6 +4711,9 @@
<constant name="VIEWPORT_DEBUG_DRAW_MOTION_VECTORS" value="25" enum="ViewportDebugDraw">
Draws the motion vectors buffer. This is used by temporal antialiasing to correct for motion that occurs during gameplay.
</constant>
<constant name="VIEWPORT_DEBUG_DRAW_INTERNAL_BUFFER" value="26" enum="ViewportDebugDraw">
Internal buffer is drawn instead of regular scene so you can see the per-pixel output that will be used by post-processing effects.
</constant>
<constant name="VIEWPORT_VRS_DISABLED" value="0" enum="ViewportVRSMode">
Variable rate shading is disabled.
</constant>

8
doc/classes/Viewport.xml
View File

@ -443,7 +443,10 @@
<constant name="SCALING_3D_MODE_FSR" value="1" enum="Scaling3DMode">
Use AMD FidelityFX Super Resolution 1.0 upscaling for the viewport's 3D buffer. The amount of scaling can be set using [member scaling_3d_scale]. Values less than [code]1.0[/code] will be result in the viewport being upscaled using FSR. Values greater than [code]1.0[/code] are not supported and bilinear downsampling will be used instead. A value of [code]1.0[/code] disables scaling.
</constant>
<constant name="SCALING_3D_MODE_MAX" value="2" enum="Scaling3DMode">
<constant name="SCALING_3D_MODE_FSR2" value="2" enum="Scaling3DMode">
Use AMD FidelityFX Super Resolution 2.2 upscaling for the viewport's 3D buffer. The amount of scaling can be set using [member Viewport.scaling_3d_scale]. Values less than [code]1.0[/code] will be result in the viewport being upscaled using FSR2. Values greater than [code]1.0[/code] are not supported and bilinear downsampling will be used instead. A value of [code]1.0[/code] will use FSR2 at native resolution as a TAA solution.
</constant>
<constant name="SCALING_3D_MODE_MAX" value="3" enum="Scaling3DMode">
Represents the size of the [enum Scaling3DMode] enum.
</constant>
<constant name="MSAA_DISABLED" value="0" enum="MSAA">
@ -553,6 +556,9 @@
</constant>
<constant name="DEBUG_DRAW_MOTION_VECTORS" value="25" enum="DebugDraw">
</constant>
<constant name="DEBUG_DRAW_INTERNAL_BUFFER" value="26" enum="DebugDraw">
Draws the internal resolution buffer of the scene before post-processing is applied.
</constant>
<constant name="DEFAULT_CANVAS_ITEM_TEXTURE_FILTER_NEAREST" value="0" enum="DefaultCanvasItemTextureFilter">
The texture filter reads from the nearest pixel only. The simplest and fastest method of filtering, but the texture will look pixelized.
</constant>

12
drivers/vulkan/rendering_device_vulkan.cpp
View File

@ -5750,8 +5750,8 @@ RID RenderingDeviceVulkan::uniform_set_create(const Vector<Uniform> &p_uniforms,
Buffer *buffer = uniform_buffer_owner.get_or_null(uniform.get_id(0));
ERR_FAIL_NULL_V_MSG(buffer, RID(), "Uniform buffer supplied (binding: " + itos(uniform.binding) + ") is invalid.");
ERR_FAIL_COND_V_MSG(buffer->size != (uint32_t)set_uniform.length, RID(),
"Uniform buffer supplied (binding: " + itos(uniform.binding) + ") size (" + itos(buffer->size) + " does not match size of shader uniform: (" + itos(set_uniform.length) + ").");
ERR_FAIL_COND_V_MSG(buffer->size < (uint32_t)set_uniform.length, RID(),
"Uniform buffer supplied (binding: " + itos(uniform.binding) + ") size (" + itos(buffer->size) + " is smaller than size of shader uniform: (" + itos(set_uniform.length) + ").");
write.dstArrayElement = 0;
write.descriptorCount = 1;
@ -9562,6 +9562,14 @@ uint64_t RenderingDeviceVulkan::limit_get(Limit p_limit) const {
VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities();
return subgroup_capabilities.size;
}
case LIMIT_SUBGROUP_MIN_SIZE: {
VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities();
return subgroup_capabilities.min_size;
}
case LIMIT_SUBGROUP_MAX_SIZE: {
VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities();
return subgroup_capabilities.max_size;
}
case LIMIT_SUBGROUP_IN_SHADERS: {
VulkanContext::SubgroupCapabilities subgroup_capabilities = context->get_subgroup_capabilities();
return subgroup_capabilities.supported_stages_flags_rd();

23
drivers/vulkan/vulkan_context.cpp
View File

@ -504,6 +504,7 @@ Error VulkanContext::_initialize_device_extensions() {
register_requested_device_extension(VK_KHR_IMAGE_FORMAT_LIST_EXTENSION_NAME, false);
register_requested_device_extension(VK_KHR_MAINTENANCE_2_EXTENSION_NAME, false);
register_requested_device_extension(VK_EXT_PIPELINE_CREATION_CACHE_CONTROL_EXTENSION_NAME, false);
register_requested_device_extension(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME, false);
if (Engine::get_singleton()->is_generate_spirv_debug_info_enabled()) {
register_requested_device_extension(VK_KHR_SHADER_NON_SEMANTIC_INFO_EXTENSION_NAME, true);
@ -739,9 +740,12 @@ Error VulkanContext::_check_capabilities() {
multiview_capabilities.max_view_count = 0;
multiview_capabilities.max_instance_count = 0;
subgroup_capabilities.size = 0;
subgroup_capabilities.min_size = 0;
subgroup_capabilities.max_size = 0;
subgroup_capabilities.supportedStages = 0;
subgroup_capabilities.supportedOperations = 0;
subgroup_capabilities.quadOperationsInAllStages = false;
subgroup_capabilities.size_control_is_supported = false;
shader_capabilities.shader_float16_is_supported = false;
shader_capabilities.shader_int8_is_supported = false;
storage_buffer_capabilities.storage_buffer_16_bit_access_is_supported = false;
@ -886,6 +890,7 @@ Error VulkanContext::_check_capabilities() {
VkPhysicalDeviceFragmentShadingRatePropertiesKHR vrsProperties{};
VkPhysicalDeviceMultiviewProperties multiviewProperties{};
VkPhysicalDeviceSubgroupProperties subgroupProperties{};
VkPhysicalDeviceSubgroupSizeControlProperties subgroupSizeControlProperties = {};
VkPhysicalDeviceProperties2 physicalDeviceProperties{};
void *nextptr = nullptr;
@ -894,6 +899,15 @@ Error VulkanContext::_check_capabilities() {
subgroupProperties.pNext = nextptr;
nextptr = &subgroupProperties;
subgroup_capabilities.size_control_is_supported = is_device_extension_enabled(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME);
if (subgroup_capabilities.size_control_is_supported) {
subgroupSizeControlProperties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES;
subgroupSizeControlProperties.pNext = nextptr;
nextptr = &subgroupSizeControlProperties;
}
}
if (multiview_capabilities.is_supported) {
@ -916,6 +930,8 @@ Error VulkanContext::_check_capabilities() {
device_properties_func(gpu, &physicalDeviceProperties);
subgroup_capabilities.size = subgroupProperties.subgroupSize;
subgroup_capabilities.min_size = subgroupProperties.subgroupSize;
subgroup_capabilities.max_size = subgroupProperties.subgroupSize;
subgroup_capabilities.supportedStages = subgroupProperties.supportedStages;
subgroup_capabilities.supportedOperations = subgroupProperties.supportedOperations;
// Note: quadOperationsInAllStages will be true if:
@ -923,6 +939,11 @@ Error VulkanContext::_check_capabilities() {
// - supportedOperations has VK_SUBGROUP_FEATURE_QUAD_BIT.
subgroup_capabilities.quadOperationsInAllStages = subgroupProperties.quadOperationsInAllStages;
if (subgroup_capabilities.size_control_is_supported && (subgroupSizeControlProperties.requiredSubgroupSizeStages & VK_SHADER_STAGE_COMPUTE_BIT)) {
subgroup_capabilities.min_size = subgroupSizeControlProperties.minSubgroupSize;
subgroup_capabilities.max_size = subgroupSizeControlProperties.maxSubgroupSize;
}
if (vrs_capabilities.pipeline_vrs_supported || vrs_capabilities.primitive_vrs_supported || vrs_capabilities.attachment_vrs_supported) {
print_verbose("- Vulkan Variable Rate Shading supported:");
if (vrs_capabilities.pipeline_vrs_supported) {
@ -962,6 +983,8 @@ Error VulkanContext::_check_capabilities() {
print_verbose("- Vulkan subgroup:");
print_verbose(" size: " + itos(subgroup_capabilities.size));
print_verbose(" min size: " + itos(subgroup_capabilities.min_size));
print_verbose(" max size: " + itos(subgroup_capabilities.max_size));
print_verbose(" stages: " + subgroup_capabilities.supported_stages_desc());
print_verbose(" supported ops: " + subgroup_capabilities.supported_operations_desc());
if (subgroup_capabilities.quadOperationsInAllStages) {

3
drivers/vulkan/vulkan_context.h
View File

@ -52,9 +52,12 @@ class VulkanContext {
public:
struct SubgroupCapabilities {
uint32_t size;
uint32_t min_size;
uint32_t max_size;
VkShaderStageFlags supportedStages;
VkSubgroupFeatureFlags supportedOperations;
VkBool32 quadOperationsInAllStages;
bool size_control_is_supported;
uint32_t supported_stages_flags_rd() const;
String supported_stages_desc() const;

6
editor/plugins/node_3d_editor_plugin.cpp
View File

@ -3472,7 +3472,8 @@ void Node3DEditorViewport::_menu_option(int p_option) {
case VIEW_DISPLAY_DEBUG_CLUSTER_DECALS:
case VIEW_DISPLAY_DEBUG_CLUSTER_REFLECTION_PROBES:
case VIEW_DISPLAY_DEBUG_OCCLUDERS:
case VIEW_DISPLAY_MOTION_VECTORS: {
case VIEW_DISPLAY_MOTION_VECTORS:
case VIEW_DISPLAY_INTERNAL_BUFFER: {
static const int display_options[] = {
VIEW_DISPLAY_NORMAL,
VIEW_DISPLAY_WIREFRAME,
@ -3500,6 +3501,7 @@ void Node3DEditorViewport::_menu_option(int p_option) {
VIEW_DISPLAY_DEBUG_CLUSTER_REFLECTION_PROBES,
VIEW_DISPLAY_DEBUG_OCCLUDERS,
VIEW_DISPLAY_MOTION_VECTORS,
VIEW_DISPLAY_INTERNAL_BUFFER,
VIEW_MAX
};
static const Viewport::DebugDraw debug_draw_modes[] = {
@ -3529,6 +3531,7 @@ void Node3DEditorViewport::_menu_option(int p_option) {
Viewport::DEBUG_DRAW_CLUSTER_REFLECTION_PROBES,
Viewport::DEBUG_DRAW_OCCLUDERS,
Viewport::DEBUG_DRAW_MOTION_VECTORS,
Viewport::DEBUG_DRAW_INTERNAL_BUFFER,
};
for (int idx = 0; display_options[idx] != VIEW_MAX; idx++) {
@ -5112,6 +5115,7 @@ Node3DEditorViewport::Node3DEditorViewport(Node3DEditor *p_spatial_editor, int p
display_submenu->add_radio_check_item(TTR("ReflectionProbe Cluster"), VIEW_DISPLAY_DEBUG_CLUSTER_REFLECTION_PROBES);
display_submenu->add_radio_check_item(TTR("Occlusion Culling Buffer"), VIEW_DISPLAY_DEBUG_OCCLUDERS);
display_submenu->add_radio_check_item(TTR("Motion Vectors"), VIEW_DISPLAY_MOTION_VECTORS);
display_submenu->add_radio_check_item(TTR("Internal Buffer"), VIEW_DISPLAY_INTERNAL_BUFFER);
display_submenu->set_name("display_advanced");
view_menu->get_popup()->add_submenu_item(TTR("Display Advanced..."), "display_advanced", VIEW_DISPLAY_ADVANCED);

1
editor/plugins/node_3d_editor_plugin.h
View File

@ -156,6 +156,7 @@ class Node3DEditorViewport : public Control {
VIEW_DISPLAY_DEBUG_CLUSTER_REFLECTION_PROBES,
VIEW_DISPLAY_DEBUG_OCCLUDERS,
VIEW_DISPLAY_MOTION_VECTORS,
VIEW_DISPLAY_INTERNAL_BUFFER,
VIEW_DISPLAY_MAX,
// > Keep in sync with menu.

4
scene/main/viewport.cpp
View File

@ -4505,7 +4505,7 @@ void Viewport::_bind_methods() {
#ifndef _3D_DISABLED
ADD_GROUP("Scaling 3D", "");
ADD_PROPERTY(PropertyInfo(Variant::INT, "scaling_3d_mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast)"), "set_scaling_3d_mode", "get_scaling_3d_mode");
ADD_PROPERTY(PropertyInfo(Variant::INT, "scaling_3d_mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast),FSR 2.2 (Slow)"), "set_scaling_3d_mode", "get_scaling_3d_mode");
ADD_PROPERTY(PropertyInfo(Variant::FLOAT, "scaling_3d_scale", PROPERTY_HINT_RANGE, "0.25,2.0,0.01"), "set_scaling_3d_scale", "get_scaling_3d_scale");
ADD_PROPERTY(PropertyInfo(Variant::FLOAT, "texture_mipmap_bias", PROPERTY_HINT_RANGE, "-2,2,0.001"), "set_texture_mipmap_bias", "get_texture_mipmap_bias");
ADD_PROPERTY(PropertyInfo(Variant::FLOAT, "fsr_sharpness", PROPERTY_HINT_RANGE, "0,2,0.1"), "set_fsr_sharpness", "get_fsr_sharpness");
@ -4556,6 +4556,7 @@ void Viewport::_bind_methods() {
BIND_ENUM_CONSTANT(SCALING_3D_MODE_BILINEAR);
BIND_ENUM_CONSTANT(SCALING_3D_MODE_FSR);
BIND_ENUM_CONSTANT(SCALING_3D_MODE_FSR2);
BIND_ENUM_CONSTANT(SCALING_3D_MODE_MAX);
BIND_ENUM_CONSTANT(MSAA_DISABLED);
@ -4603,6 +4604,7 @@ void Viewport::_bind_methods() {
BIND_ENUM_CONSTANT(DEBUG_DRAW_CLUSTER_REFLECTION_PROBES);
BIND_ENUM_CONSTANT(DEBUG_DRAW_OCCLUDERS)
BIND_ENUM_CONSTANT(DEBUG_DRAW_MOTION_VECTORS)
BIND_ENUM_CONSTANT(DEBUG_DRAW_INTERNAL_BUFFER);
BIND_ENUM_CONSTANT(DEFAULT_CANVAS_ITEM_TEXTURE_FILTER_NEAREST);
BIND_ENUM_CONSTANT(DEFAULT_CANVAS_ITEM_TEXTURE_FILTER_LINEAR);

2
scene/main/viewport.h
View File

@ -98,6 +98,7 @@ public:
enum Scaling3DMode {
SCALING_3D_MODE_BILINEAR,
SCALING_3D_MODE_FSR,
SCALING_3D_MODE_FSR2,
SCALING_3D_MODE_MAX
};
@ -167,6 +168,7 @@ public:
DEBUG_DRAW_CLUSTER_REFLECTION_PROBES,
DEBUG_DRAW_OCCLUDERS,
DEBUG_DRAW_MOTION_VECTORS,
DEBUG_DRAW_INTERNAL_BUFFER,
};
enum DefaultCanvasItemTextureFilter {

31
servers/rendering/renderer_rd/effects/SCsub
View File

@ -2,4 +2,33 @@
Import("env")
env.add_source_files(env.servers_sources, "*.cpp")
env_effects = env.Clone()
# Thirdparty source files
thirdparty_obj = []
thirdparty_dir = "#thirdparty/amd-fsr2/"
thirdparty_sources = ["ffx_assert.cpp", "ffx_fsr2.cpp"]
thirdparty_sources = [thirdparty_dir + file for file in thirdparty_sources]
env_effects.Prepend(CPPPATH=[thirdparty_dir])
# This flag doesn't actually control anything GCC specific in FSR2. It determines
# if symbols should be exported, which is not required for Godot.
env_effects.Append(CPPDEFINES=["FFX_GCC"])
env_thirdparty = env_effects.Clone()
env_thirdparty.disable_warnings()
env_thirdparty.add_source_files(thirdparty_obj, thirdparty_sources)
env.servers_sources += thirdparty_obj
# Godot source files
module_obj = []
env_effects.add_source_files(module_obj, "*.cpp")
env.servers_sources += module_obj
# Needed to force rebuilding the module files when the thirdparty library is updated.
env.Depends(module_obj, thirdparty_obj)

4
servers/rendering/renderer_rd/effects/copy_effects.cpp
View File

@ -281,8 +281,8 @@ CopyEffects::CopyEffects(bool p_prefer_raster_effects) {
ba.enable_blend = true;
ba.src_color_blend_factor = RD::BLEND_FACTOR_ONE;
ba.dst_color_blend_factor = RD::BLEND_FACTOR_ONE;
ba.src_alpha_blend_factor = RD::BLEND_FACTOR_ONE;
ba.dst_alpha_blend_factor = RD::BLEND_FACTOR_ONE;
ba.src_alpha_blend_factor = RD::BLEND_FACTOR_ZERO;
ba.dst_alpha_blend_factor = RD::BLEND_FACTOR_ZERO;
ba.color_blend_op = RD::BLEND_OP_ADD;
ba.alpha_blend_op = RD::BLEND_OP_ADD;

23
servers/rendering/renderer_rd/effects/debug_effects.cpp
View File

@ -340,25 +340,38 @@ void DebugEffects::draw_shadow_frustum(RID p_light, const Projection &p_cam_proj
}
}
void DebugEffects::draw_motion_vectors(RID p_velocity, RID p_dest_fb, Size2i p_velocity_size) {
void DebugEffects::draw_motion_vectors(RID p_velocity, RID p_depth, RID p_dest_fb, const Projection &p_current_projection, const Transform3D &p_current_transform, const Projection &p_previous_projection, const Transform3D &p_previous_transform, Size2i p_resolution) {
MaterialStorage *material_storage = MaterialStorage::get_singleton();
ERR_FAIL_NULL(material_storage);
UniformSetCacheRD *uniform_set_cache = UniformSetCacheRD::get_singleton();
ERR_FAIL_NULL(uniform_set_cache);
RID default_sampler = material_storage->sampler_rd_get_default(RS::CANVAS_ITEM_TEXTURE_FILTER_LINEAR, RS::CANVAS_ITEM_TEXTURE_REPEAT_DISABLED);
RID default_sampler = material_storage->sampler_rd_get_default(RS::CANVAS_ITEM_TEXTURE_FILTER_NEAREST, RS::CANVAS_ITEM_TEXTURE_REPEAT_DISABLED);
RD::Uniform u_source_velocity(RD::UNIFORM_TYPE_SAMPLER_WITH_TEXTURE, 0, Vector<RID>({ default_sampler, p_velocity }));
RD::Uniform u_source_depth(RD::UNIFORM_TYPE_SAMPLER_WITH_TEXTURE, 1, Vector<RID>({ default_sampler, p_depth }));
RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(p_dest_fb, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_DROP, RD::FINAL_ACTION_DISCARD);
RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, motion_vectors.pipeline.get_render_pipeline(RD::INVALID_ID, RD::get_singleton()->framebuffer_get_format(p_dest_fb), false, RD::get_singleton()->draw_list_get_current_pass()));
motion_vectors.push_constant.velocity_resolution[0] = p_velocity_size.width;
motion_vectors.push_constant.velocity_resolution[1] = p_velocity_size.height;
Projection reprojection = p_previous_projection.flipped_y() * p_previous_transform.affine_inverse() * p_current_transform * p_current_projection.flipped_y().inverse();
RendererRD::MaterialStorage::store_camera(reprojection, motion_vectors.push_constant.reprojection_matrix);
motion_vectors.push_constant.resolution[0] = p_resolution.width;
motion_vectors.push_constant.resolution[1] = p_resolution.height;
motion_vectors.push_constant.force_derive_from_depth = false;
RID shader = motion_vectors.shader.version_get_shader(motion_vectors.shader_version, 0);
RD::get_singleton()->draw_list_bind_uniform_set(draw_list, uniform_set_cache->get_cache(shader, 0, u_source_velocity), 0);
RD::get_singleton()->draw_list_bind_uniform_set(draw_list, uniform_set_cache->get_cache(shader, 0, u_source_velocity, u_source_depth), 0);
RD::get_singleton()->draw_list_set_push_constant(draw_list, &motion_vectors.push_constant, sizeof(MotionVectorsPushConstant));
RD::get_singleton()->draw_list_draw(draw_list, false, 1u, 3u);
#ifdef DRAW_DERIVATION_FROM_DEPTH_ON_TOP
motion_vectors.push_constant.force_derive_from_depth = true;
RD::get_singleton()->draw_list_set_push_constant(draw_list, &motion_vectors.push_constant, sizeof(MotionVectorsPushConstant));
RD::get_singleton()->draw_list_draw(draw_list, false, 1u, 3u);
#endif
RD::get_singleton()->draw_list_end();
}

8
servers/rendering/renderer_rd/effects/debug_effects.h
View File

@ -72,8 +72,10 @@ private:
} shadow_frustum;
struct MotionVectorsPushConstant {
float velocity_resolution[2];
float pad[2];
float reprojection_matrix[16];
float resolution[2];
uint32_t force_derive_from_depth;
float pad;
};
struct {
@ -91,7 +93,7 @@ public:
~DebugEffects();
void draw_shadow_frustum(RID p_light, const Projection &p_cam_projection, const Transform3D &p_cam_transform, RID p_dest_fb, const Rect2 p_rect);
void draw_motion_vectors(RID p_velocity, RID p_dest_fb, Size2i p_velocity_size);
void draw_motion_vectors(RID p_velocity, RID p_depth, RID p_dest_fb, const Projection &p_current_projection, const Transform3D &p_current_transform, const Projection &p_previous_projection, const Transform3D &p_previous_transform, Size2i p_resolution);
};
} // namespace RendererRD

889
servers/rendering/renderer_rd/effects/fsr2.cpp Normal file
View File

@ -0,0 +1,889 @@
/**************************************************************************/
/* fsr2.cpp */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
#include "fsr2.h"
#include "../storage_rd/material_storage.h"
#include "../uniform_set_cache_rd.h"
using namespace RendererRD;
#ifndef _MSC_VER
#include <wchar.h>
#define wcscpy_s wcscpy
#endif
static RD::TextureType ffx_resource_type_to_rd_texture_type(FfxResourceType p_type) {
switch (p_type) {
case FFX_RESOURCE_TYPE_TEXTURE1D:
return RD::TEXTURE_TYPE_1D;
case FFX_RESOURCE_TYPE_TEXTURE2D:
return RD::TEXTURE_TYPE_2D;
case FFX_RESOURCE_TYPE_TEXTURE3D:
return RD::TEXTURE_TYPE_3D;
default:
return RD::TEXTURE_TYPE_MAX;
}
}
static FfxResourceType rd_texture_type_to_ffx_resource_type(RD::TextureType p_type) {
switch (p_type) {
case RD::TEXTURE_TYPE_1D:
return FFX_RESOURCE_TYPE_TEXTURE1D;
case RD::TEXTURE_TYPE_2D:
return FFX_RESOURCE_TYPE_TEXTURE2D;
case RD::TEXTURE_TYPE_3D:
return FFX_RESOURCE_TYPE_TEXTURE3D;
default:
return FFX_RESOURCE_TYPE_BUFFER;
}
}
static RD::DataFormat ffx_surface_format_to_rd_format(FfxSurfaceFormat p_format) {
switch (p_format) {
case FFX_SURFACE_FORMAT_R32G32B32A32_TYPELESS:
return RD::DATA_FORMAT_R32G32B32A32_SFLOAT;
case FFX_SURFACE_FORMAT_R32G32B32A32_FLOAT:
return RD::DATA_FORMAT_R32G32B32A32_SFLOAT;
case FFX_SURFACE_FORMAT_R16G16B16A16_FLOAT:
return RD::DATA_FORMAT_R16G16B16A16_SFLOAT;
case FFX_SURFACE_FORMAT_R16G16B16A16_UNORM:
return RD::DATA_FORMAT_R16G16B16A16_UNORM;
case FFX_SURFACE_FORMAT_R32G32_FLOAT:
return RD::DATA_FORMAT_R32G32_SFLOAT;
case FFX_SURFACE_FORMAT_R32_UINT:
return RD::DATA_FORMAT_R32_UINT;
case FFX_SURFACE_FORMAT_R8G8B8A8_TYPELESS:
return RD::DATA_FORMAT_R8G8B8A8_UNORM;
case FFX_SURFACE_FORMAT_R8G8B8A8_UNORM:
return RD::DATA_FORMAT_R8G8B8A8_UNORM;
case FFX_SURFACE_FORMAT_R11G11B10_FLOAT:
return RD::DATA_FORMAT_B10G11R11_UFLOAT_PACK32;
case FFX_SURFACE_FORMAT_R16G16_FLOAT:
return RD::DATA_FORMAT_R16G16_SFLOAT;
case FFX_SURFACE_FORMAT_R16G16_UINT:
return RD::DATA_FORMAT_R16G16_UINT;
case FFX_SURFACE_FORMAT_R16_FLOAT:
return RD::DATA_FORMAT_R16_SFLOAT;
case FFX_SURFACE_FORMAT_R16_UINT:
return RD::DATA_FORMAT_R16_UINT;
case FFX_SURFACE_FORMAT_R16_UNORM:
return RD::DATA_FORMAT_R16_UNORM;
case FFX_SURFACE_FORMAT_R16_SNORM:
return RD::DATA_FORMAT_R16_SNORM;
case FFX_SURFACE_FORMAT_R8_UNORM:
return RD::DATA_FORMAT_R8_UNORM;
case FFX_SURFACE_FORMAT_R8_UINT:
return RD::DATA_FORMAT_R8_UINT;
case FFX_SURFACE_FORMAT_R8G8_UNORM:
return RD::DATA_FORMAT_R8G8_UNORM;
case FFX_SURFACE_FORMAT_R32_FLOAT:
return RD::DATA_FORMAT_R32_SFLOAT;
default:
return RD::DATA_FORMAT_MAX;
}
}
static FfxSurfaceFormat rd_format_to_ffx_surface_format(RD::DataFormat p_format) {
switch (p_format) {
case RD::DATA_FORMAT_R32G32B32A32_SFLOAT:
return FFX_SURFACE_FORMAT_R32G32B32A32_FLOAT;
case RD::DATA_FORMAT_R16G16B16A16_SFLOAT:
return FFX_SURFACE_FORMAT_R16G16B16A16_FLOAT;
case RD::DATA_FORMAT_R16G16B16A16_UNORM:
return FFX_SURFACE_FORMAT_R16G16B16A16_UNORM;
case RD::DATA_FORMAT_R32G32_SFLOAT:
return FFX_SURFACE_FORMAT_R32G32_FLOAT;
case RD::DATA_FORMAT_R32_UINT:
return FFX_SURFACE_FORMAT_R32_UINT;
case RD::DATA_FORMAT_R8G8B8A8_UNORM:
return FFX_SURFACE_FORMAT_R8G8B8A8_UNORM;
case RD::DATA_FORMAT_B10G11R11_UFLOAT_PACK32:
return FFX_SURFACE_FORMAT_R11G11B10_FLOAT;
case RD::DATA_FORMAT_R16G16_SFLOAT:
return FFX_SURFACE_FORMAT_R16G16_FLOAT;
case RD::DATA_FORMAT_R16G16_UINT:
return FFX_SURFACE_FORMAT_R16G16_UINT;
case RD::DATA_FORMAT_R16_SFLOAT:
return FFX_SURFACE_FORMAT_R16_FLOAT;
case RD::DATA_FORMAT_R16_UINT:
return FFX_SURFACE_FORMAT_R16_UINT;
case RD::DATA_FORMAT_R16_UNORM:
return FFX_SURFACE_FORMAT_R16_UNORM;
case RD::DATA_FORMAT_R16_SNORM:
return FFX_SURFACE_FORMAT_R16_SNORM;
case RD::DATA_FORMAT_R8_UNORM:
return FFX_SURFACE_FORMAT_R8_UNORM;
case RD::DATA_FORMAT_R8_UINT:
return FFX_SURFACE_FORMAT_R8_UINT;
case RD::DATA_FORMAT_R8G8_UNORM:
return FFX_SURFACE_FORMAT_R8G8_UNORM;
case RD::DATA_FORMAT_R32_SFLOAT:
return FFX_SURFACE_FORMAT_R32_FLOAT;
default:
return FFX_SURFACE_FORMAT_UNKNOWN;
}
}
static uint32_t ffx_usage_to_rd_usage_flags(uint32_t p_flags) {
uint32_t ret = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_UPDATE_BIT;
if (p_flags & FFX_RESOURCE_USAGE_RENDERTARGET) {
ret |= RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT;
}
if (p_flags & FFX_RESOURCE_USAGE_UAV) {
ret |= RD::TEXTURE_USAGE_STORAGE_BIT;
ret |= RD::TEXTURE_USAGE_CAN_COPY_FROM_BIT;
ret |= RD::TEXTURE_USAGE_CAN_COPY_TO_BIT;
}
return ret;
}
static FfxErrorCode create_backend_context_rd(FfxFsr2Interface *p_backend_interface, FfxDevice p_device) {
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
// Store pointer to the device common to all contexts.
scratch.device = p_device;
// Create a ring buffer of uniform buffers.
// FIXME: This could be optimized to be a single memory block if it was possible for RD to create views into a particular memory range of a UBO.
for (uint32_t i = 0; i < FSR2_UBO_RING_BUFFER_SIZE; i++) {
scratch.ubo_ring_buffer[i] = RD::get_singleton()->uniform_buffer_create(FFX_MAX_CONST_SIZE * sizeof(uint32_t));
ERR_FAIL_COND_V(scratch.ubo_ring_buffer[i].is_null(), FFX_ERROR_BACKEND_API_ERROR);
}
return FFX_OK;
}
static FfxErrorCode get_device_capabilities_rd(FfxFsr2Interface *p_backend_interface, FfxDeviceCapabilities *p_out_device_capabilities, FfxDevice p_device) {
FSR2Effect::Device &effect_device = *reinterpret_cast<FSR2Effect::Device *>(p_device);
*p_out_device_capabilities = effect_device.capabilities;
return FFX_OK;
}
static FfxErrorCode destroy_backend_context_rd(FfxFsr2Interface *p_backend_interface) {
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
for (uint32_t i = 0; i < FSR2_UBO_RING_BUFFER_SIZE; i++) {
RD::get_singleton()->free(scratch.ubo_ring_buffer[i]);
}
return FFX_OK;
}
static FfxErrorCode create_resource_rd(FfxFsr2Interface *p_backend_interface, const FfxCreateResourceDescription *p_create_resource_description, FfxResourceInternal *p_out_resource) {
// FSR2's base implementation won't issue a call to create a heap type that isn't just default on its own,
// so we can safely ignore it as RD does not expose this concept.
ERR_FAIL_COND_V(p_create_resource_description->heapType != FFX_HEAP_TYPE_DEFAULT, FFX_ERROR_INVALID_ARGUMENT);
RenderingDevice *rd = RD::get_singleton();
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
FfxResourceDescription res_desc = p_create_resource_description->resourceDescription;
// FSR2's base implementation never requests buffer creation.
ERR_FAIL_COND_V(res_desc.type != FFX_RESOURCE_TYPE_TEXTURE1D && res_desc.type != FFX_RESOURCE_TYPE_TEXTURE2D && res_desc.type != FFX_RESOURCE_TYPE_TEXTURE3D, FFX_ERROR_INVALID_ARGUMENT);
if (res_desc.mipCount == 0) {
// Mipmap count must be derived from the resource's dimensions.
res_desc.mipCount = uint32_t(1 + floor(log2(MAX(MAX(res_desc.width, res_desc.height), res_desc.depth))));
}
Vector<PackedByteArray> initial_data;
if (p_create_resource_description->initDataSize) {
PackedByteArray byte_array;
byte_array.resize(p_create_resource_description->initDataSize);
memcpy(byte_array.ptrw(), p_create_resource_description->initData, p_create_resource_description->initDataSize);
initial_data.push_back(byte_array);
}
RD::TextureFormat texture_format;
texture_format.texture_type = ffx_resource_type_to_rd_texture_type(res_desc.type);
texture_format.format = ffx_surface_format_to_rd_format(res_desc.format);
texture_format.usage_bits = ffx_usage_to_rd_usage_flags(p_create_resource_description->usage);
texture_format.width = res_desc.width;
texture_format.height = res_desc.height;
texture_format.depth = res_desc.depth;
texture_format.mipmaps = res_desc.mipCount;
RID texture = rd->texture_create(texture_format, RD::TextureView(), initial_data);
ERR_FAIL_COND_V(texture.is_null(), FFX_ERROR_BACKEND_API_ERROR);
rd->set_resource_name(texture, String(p_create_resource_description->name));
// Add the resource to the storage and use the internal index to reference it.
p_out_resource->internalIndex = scratch.resources.add(texture, false, p_create_resource_description->id, res_desc);
return FFX_OK;
}
static FfxErrorCode register_resource_rd(FfxFsr2Interface *p_backend_interface, const FfxResource *p_in_resource, FfxResourceInternal *p_out_resource) {
if (p_in_resource->resource == nullptr) {
// Null resource case.
p_out_resource->internalIndex = -1;
return FFX_OK;
}
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
const RID &rid = *reinterpret_cast<const RID *>(p_in_resource->resource);
ERR_FAIL_COND_V(rid.is_null(), FFX_ERROR_INVALID_ARGUMENT);
// Add the resource to the storage and use the internal index to reference it.
p_out_resource->internalIndex = scratch.resources.add(rid, true, FSR2Context::RESOURCE_ID_DYNAMIC, p_in_resource->description);
return FFX_OK;
}
static FfxErrorCode unregister_resources_rd(FfxFsr2Interface *p_backend_interface) {
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
LocalVector<uint32_t> dynamic_list_copy = scratch.resources.dynamic_list;
for (uint32_t i : dynamic_list_copy) {
scratch.resources.remove(i);
}
return FFX_OK;
}
static FfxResourceDescription get_resource_description_rd(FfxFsr2Interface *p_backend_interface, FfxResourceInternal p_resource) {
if (p_resource.internalIndex != -1) {
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
return scratch.resources.descriptions[p_resource.internalIndex];
} else {
return {};
}
}
static FfxErrorCode destroy_resource_rd(FfxFsr2Interface *p_backend_interface, FfxResourceInternal p_resource) {
if (p_resource.internalIndex != -1) {
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
if (scratch.resources.rids[p_resource.internalIndex].is_valid()) {
RD::get_singleton()->free(scratch.resources.rids[p_resource.internalIndex]);
scratch.resources.remove(p_resource.internalIndex);
}
}
return FFX_OK;
}
static FfxErrorCode create_pipeline_rd(FfxFsr2Interface *p_backend_interface, FfxFsr2Pass p_pass, const FfxPipelineDescription *p_pipeline_description, FfxPipelineState *p_out_pipeline) {
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
FSR2Effect::Device &device = *reinterpret_cast<FSR2Effect::Device *>(scratch.device);
FSR2Effect::Pass &effect_pass = device.passes[p_pass];
if (effect_pass.pipeline.pipeline_rid.is_null()) {
// Create pipeline for the device if it hasn't been created yet.
effect_pass.root_signature.shader_rid = effect_pass.shader->version_get_shader(effect_pass.shader_version, effect_pass.shader_variant);
ERR_FAIL_COND_V(effect_pass.root_signature.shader_rid.is_null(), FFX_ERROR_BACKEND_API_ERROR);
effect_pass.pipeline.pipeline_rid = RD::get_singleton()->compute_pipeline_create(effect_pass.root_signature.shader_rid);
ERR_FAIL_COND_V(effect_pass.pipeline.pipeline_rid.is_null(), FFX_ERROR_BACKEND_API_ERROR);
}
// While this is not their intended use, we use the pipeline and root signature pointers to store the
// RIDs to the pipeline and shader that RD needs for the compute pipeline.
p_out_pipeline->pipeline = reinterpret_cast<FfxPipeline>(&effect_pass.pipeline);
p_out_pipeline->rootSignature = reinterpret_cast<FfxRootSignature>(&effect_pass.root_signature);
p_out_pipeline->srvCount = effect_pass.sampled_bindings.size();
ERR_FAIL_COND_V(p_out_pipeline->srvCount > FFX_MAX_NUM_SRVS, FFX_ERROR_OUT_OF_RANGE);
memcpy(p_out_pipeline->srvResourceBindings, effect_pass.sampled_bindings.ptr(), sizeof(FfxResourceBinding) * p_out_pipeline->srvCount);
p_out_pipeline->uavCount = effect_pass.storage_bindings.size();
ERR_FAIL_COND_V(p_out_pipeline->uavCount > FFX_MAX_NUM_UAVS, FFX_ERROR_OUT_OF_RANGE);
memcpy(p_out_pipeline->uavResourceBindings, effect_pass.storage_bindings.ptr(), sizeof(FfxResourceBinding) * p_out_pipeline->uavCount);
p_out_pipeline->constCount = effect_pass.uniform_bindings.size();
ERR_FAIL_COND_V(p_out_pipeline->constCount > FFX_MAX_NUM_CONST_BUFFERS, FFX_ERROR_OUT_OF_RANGE);
memcpy(p_out_pipeline->cbResourceBindings, effect_pass.uniform_bindings.ptr(), sizeof(FfxResourceBinding) * p_out_pipeline->constCount);
bool low_resolution_mvs = (p_pipeline_description->contextFlags & FFX_FSR2_ENABLE_DISPLAY_RESOLUTION_MOTION_VECTORS) == 0;
if (p_pass == FFX_FSR2_PASS_ACCUMULATE || p_pass == FFX_FSR2_PASS_ACCUMULATE_SHARPEN) {
// Change the binding for motion vectors in this particular pass if low resolution MVs are used.
if (low_resolution_mvs) {
FfxResourceBinding &binding = p_out_pipeline->srvResourceBindings[2];
wcscpy_s(binding.name, L"r_dilated_motion_vectors");
}
}
return FFX_OK;
}
static FfxErrorCode destroy_pipeline_rd(FfxFsr2Interface *p_backend_interface, FfxPipelineState *p_pipeline) {
// We don't want to destroy pipelines when the FSR2 API deems it necessary as it'll do so whenever the context is destroyed.
return FFX_OK;
}
static FfxErrorCode schedule_gpu_job_rd(FfxFsr2Interface *p_backend_interface, const FfxGpuJobDescription *p_job) {
ERR_FAIL_NULL_V(p_backend_interface, FFX_ERROR_INVALID_ARGUMENT);
ERR_FAIL_NULL_V(p_job, FFX_ERROR_INVALID_ARGUMENT);
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
scratch.gpu_jobs.push_back(*p_job);
return FFX_OK;
}
static FfxErrorCode execute_gpu_job_clear_float_rd(FSR2Context::Scratch &p_scratch, const FfxClearFloatJobDescription &p_job) {
RID resource = p_scratch.resources.rids[p_job.target.internalIndex];
FfxResourceDescription &desc = p_scratch.resources.descriptions[p_job.target.internalIndex];
ERR_FAIL_COND_V(desc.type == FFX_RESOURCE_TYPE_BUFFER, FFX_ERROR_INVALID_ARGUMENT);
Color color(p_job.color[0], p_job.color[1], p_job.color[2], p_job.color[3]);
RD::get_singleton()->texture_clear(resource, color, 0, desc.mipCount, 0, 1);
return FFX_OK;
}
static FfxErrorCode execute_gpu_job_copy_rd(FSR2Context::Scratch &p_scratch, const FfxCopyJobDescription &p_job) {
RID src = p_scratch.resources.rids[p_job.src.internalIndex];
RID dst = p_scratch.resources.rids[p_job.dst.internalIndex];
FfxResourceDescription &src_desc = p_scratch.resources.descriptions[p_job.src.internalIndex];
FfxResourceDescription &dst_desc = p_scratch.resources.descriptions[p_job.dst.internalIndex];
ERR_FAIL_COND_V(src_desc.type == FFX_RESOURCE_TYPE_BUFFER, FFX_ERROR_INVALID_ARGUMENT);
ERR_FAIL_COND_V(dst_desc.type == FFX_RESOURCE_TYPE_BUFFER, FFX_ERROR_INVALID_ARGUMENT);
for (uint32_t mip_level = 0; mip_level < src_desc.mipCount; mip_level++) {
// Only push the barriers on the last copy.
// FIXME: This could be optimized if RenderingDevice was able to copy multiple mip levels in a single command.
BitField<RD::BarrierMask> post_barrier = (mip_level == (src_desc.mipCount - 1)) ? RD::BARRIER_MASK_ALL_BARRIERS : RD::BARRIER_MASK_NO_BARRIER;
RD::get_singleton()->texture_copy(src, dst, Vector3(0, 0, 0), Vector3(0, 0, 0), Vector3(src_desc.width, src_desc.height, src_desc.depth), mip_level, mip_level, 0, 0, post_barrier);
}
return FFX_OK;
}
static FfxErrorCode execute_gpu_job_compute_rd(FSR2Context::Scratch &p_scratch, const FfxComputeJobDescription &p_job) {
UniformSetCacheRD *uniform_set_cache = UniformSetCacheRD::get_singleton();
ERR_FAIL_NULL_V(uniform_set_cache, FFX_ERROR_BACKEND_API_ERROR);
FSR2Effect::RootSignature &root_signature = *reinterpret_cast<FSR2Effect::RootSignature *>(p_job.pipeline.rootSignature);
ERR_FAIL_COND_V(root_signature.shader_rid.is_null(), FFX_ERROR_INVALID_ARGUMENT);
FSR2Effect::Pipeline &backend_pipeline = *reinterpret_cast<FSR2Effect::Pipeline *>(p_job.pipeline.pipeline);
ERR_FAIL_COND_V(backend_pipeline.pipeline_rid.is_null(), FFX_ERROR_INVALID_ARGUMENT);
Vector<RD::Uniform> compute_uniforms;
for (uint32_t i = 0; i < p_job.pipeline.srvCount; i++) {
RID texture_rid = p_scratch.resources.rids[p_job.srvs[i].internalIndex];
RD::Uniform texture_uniform(RD::UNIFORM_TYPE_TEXTURE, p_job.pipeline.srvResourceBindings[i].slotIndex, texture_rid);
compute_uniforms.push_back(texture_uniform);
}
for (uint32_t i = 0; i < p_job.pipeline.uavCount; i++) {
RID image_rid = p_scratch.resources.rids[p_job.uavs[i].internalIndex];
RD::Uniform storage_uniform;
storage_uniform.uniform_type = RD::UNIFORM_TYPE_IMAGE;
storage_uniform.binding = p_job.pipeline.uavResourceBindings[i].slotIndex;
if (p_job.uavMip[i] > 0) {
LocalVector<RID> &mip_slice_rids = p_scratch.resources.mip_slice_rids[p_job.uavs[i].internalIndex];
if (mip_slice_rids.is_empty()) {
mip_slice_rids.resize(p_scratch.resources.descriptions[p_job.uavs[i].internalIndex].mipCount);
}
ERR_FAIL_COND_V(p_job.uavMip[i] >= mip_slice_rids.size(), FFX_ERROR_INVALID_ARGUMENT);
if (mip_slice_rids[p_job.uavMip[i]].is_null()) {
mip_slice_rids[p_job.uavMip[i]] = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), image_rid, 0, p_job.uavMip[i]);
}
ERR_FAIL_COND_V(mip_slice_rids[p_job.uavMip[i]].is_null(), FFX_ERROR_BACKEND_API_ERROR);
storage_uniform.append_id(mip_slice_rids[p_job.uavMip[i]]);
} else {
storage_uniform.append_id(image_rid);
}
compute_uniforms.push_back(storage_uniform);
}
for (uint32_t i = 0; i < p_job.pipeline.constCount; i++) {
RID buffer_rid = p_scratch.ubo_ring_buffer[p_scratch.ubo_ring_buffer_index];
p_scratch.ubo_ring_buffer_index = (p_scratch.ubo_ring_buffer_index + 1) % FSR2_UBO_RING_BUFFER_SIZE;
BitField<RD::BarrierMask> post_barrier = (i == (p_job.pipeline.constCount - 1)) ? RD::BARRIER_MASK_ALL_BARRIERS : RD::BARRIER_MASK_NO_BARRIER;
RD::get_singleton()->buffer_update(buffer_rid, 0, p_job.cbs[i].uint32Size * sizeof(uint32_t), p_job.cbs[i].data, post_barrier);
RD::Uniform buffer_uniform(RD::UNIFORM_TYPE_UNIFORM_BUFFER, p_job.pipeline.cbResourceBindings[i].slotIndex, buffer_rid);
compute_uniforms.push_back(buffer_uniform);
}
FSR2Effect::Device &device = *reinterpret_cast<FSR2Effect::Device *>(p_scratch.device);
RD::Uniform u_point_clamp_sampler(RD::UniformType::UNIFORM_TYPE_SAMPLER, 0, device.point_clamp_sampler);
RD::Uniform u_linear_clamp_sampler(RD::UniformType::UNIFORM_TYPE_SAMPLER, 1, device.linear_clamp_sampler);
RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin();
RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, backend_pipeline.pipeline_rid);
RD::get_singleton()->compute_list_bind_uniform_set(compute_list, uniform_set_cache->get_cache(root_signature.shader_rid, 0, u_point_clamp_sampler, u_linear_clamp_sampler), 0);
RD::get_singleton()->compute_list_bind_uniform_set(compute_list, uniform_set_cache->get_cache_vec(root_signature.shader_rid, 1, compute_uniforms), 1);
RD::get_singleton()->compute_list_dispatch(compute_list, p_job.dimensions[0], p_job.dimensions[1], p_job.dimensions[2]);
RD::get_singleton()->compute_list_end();
return FFX_OK;
}
static FfxErrorCode execute_gpu_jobs_rd(FfxFsr2Interface *p_backend_interface, FfxCommandList p_command_list) {
ERR_FAIL_NULL_V(p_backend_interface, FFX_ERROR_INVALID_ARGUMENT);
FSR2Context::Scratch &scratch = *reinterpret_cast<FSR2Context::Scratch *>(p_backend_interface->scratchBuffer);
FfxErrorCode error_code = FFX_OK;
for (const FfxGpuJobDescription &job : scratch.gpu_jobs) {
switch (job.jobType) {
case FFX_GPU_JOB_CLEAR_FLOAT: {
error_code = execute_gpu_job_clear_float_rd(scratch, job.clearJobDescriptor);
} break;
case FFX_GPU_JOB_COPY: {
error_code = execute_gpu_job_copy_rd(scratch, job.copyJobDescriptor);
} break;
case FFX_GPU_JOB_COMPUTE: {
error_code = execute_gpu_job_compute_rd(scratch, job.computeJobDescriptor);
} break;
default: {
error_code = FFX_ERROR_INVALID_ARGUMENT;
} break;
}
if (error_code != FFX_OK) {
scratch.gpu_jobs.clear();
return error_code;
}
}
scratch.gpu_jobs.clear();
return FFX_OK;
}
static FfxResource get_resource_rd(RID *p_rid, const wchar_t *p_name) {
FfxResource res = {};
if (p_rid->is_null()) {
return res;
}
wcscpy_s(res.name, p_name);
RD::TextureFormat texture_format = RD::get_singleton()->texture_get_format(*p_rid);
res.description.type = rd_texture_type_to_ffx_resource_type(texture_format.texture_type);
res.description.format = rd_format_to_ffx_surface_format(texture_format.format);
res.description.width = texture_format.width;
res.description.height = texture_format.height;
res.description.depth = texture_format.depth;
res.description.mipCount = texture_format.mipmaps;
res.description.flags = FFX_RESOURCE_FLAGS_NONE;
res.resource = reinterpret_cast<void *>(p_rid);
res.isDepth = texture_format.usage_bits & RD::TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
return res;
}
FSR2Context::~FSR2Context() {
ffxFsr2ContextDestroy(&fsr_context);
}
FSR2Effect::FSR2Effect() {
FfxDeviceCapabilities &capabilities = device.capabilities;
uint64_t default_subgroup_size = RD::get_singleton()->limit_get(RD::LIMIT_SUBGROUP_SIZE);
capabilities.minimumSupportedShaderModel = FFX_SHADER_MODEL_5_1;
capabilities.waveLaneCountMin = RD::get_singleton()->limit_get(RD::LIMIT_SUBGROUP_MIN_SIZE);
capabilities.waveLaneCountMax = RD::get_singleton()->limit_get(RD::LIMIT_SUBGROUP_MAX_SIZE);
capabilities.fp16Supported = RD::get_singleton()->has_feature(RD::Features::SUPPORTS_FSR_HALF_FLOAT);
capabilities.raytracingSupported = false;
bool force_wave_64 = default_subgroup_size == 32 && capabilities.waveLaneCountMax == 64;
bool use_lut = force_wave_64 || default_subgroup_size == 64;
String general_defines_base =
"\n#define FFX_GPU\n"
"\n#define FFX_GLSL 1\n"
"\n#define FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS 1\n"
"\n#define FFX_FSR2_OPTION_HDR_COLOR_INPUT 1\n"
"\n#define FFX_FSR2_OPTION_GODOT_REACTIVE_MASK_CLAMP 1\n"
"\n#define FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS 1\n";
if (use_lut) {
general_defines_base += "\n#define FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE 1\n";
}
String general_defines = general_defines_base;
if (capabilities.fp16Supported) {
general_defines += "\n#define FFX_HALF 1\n";
}
Vector<String> modes;
modes.push_back("");
// Since Godot currently lacks a shader reflection mechanism to persist the name of the bindings in the shader cache and
// there's also no mechanism to compile the shaders offline, the bindings are created manually by looking at the GLSL
// files included in FSR2 and mapping the macro bindings (#define FSR2_BIND_*) to their respective implementation names.
//
// It is not guaranteed these will remain consistent at all between versions of FSR2, so it'll be necessary to keep these
// bindings up to date whenever the library is updated. In such cases, it is very likely the validation layer will throw an
// error if the bindings do not match.
{
Pass &pass = device.passes[FFX_FSR2_PASS_DEPTH_CLIP];
pass.shader = &shaders.depth_clip;
pass.shader->initialize(modes, general_defines);
pass.shader_version = pass.shader->version_create();
pass.sampled_bindings = {
FfxResourceBinding{ 0, 0, L"r_reconstructed_previous_nearest_depth" },
FfxResourceBinding{ 1, 0, L"r_dilated_motion_vectors" },
FfxResourceBinding{ 2, 0, L"r_dilatedDepth" },
FfxResourceBinding{ 3, 0, L"r_reactive_mask" },
FfxResourceBinding{ 4, 0, L"r_transparency_and_composition_mask" },
FfxResourceBinding{ 5, 0, L"r_prepared_input_color" },
FfxResourceBinding{ 6, 0, L"r_previous_dilated_motion_vectors" },
FfxResourceBinding{ 7, 0, L"r_input_motion_vectors" },
FfxResourceBinding{ 8, 0, L"r_input_color_jittered" },
FfxResourceBinding{ 9, 0, L"r_input_depth" },
FfxResourceBinding{ 10, 0, L"r_input_exposure" }
};
pass.storage_bindings = {
// FSR2_BIND_UAV_DEPTH_CLIP (11) does not point to anything.
FfxResourceBinding{ 12, 0, L"rw_dilated_reactive_masks" },
FfxResourceBinding{ 13, 0, L"rw_prepared_input_color" }
};
pass.uniform_bindings = {
FfxResourceBinding{ 14, 0, L"cbFSR2" }
};
}
{
Pass &pass = device.passes[FFX_FSR2_PASS_RECONSTRUCT_PREVIOUS_DEPTH];
pass.shader = &shaders.reconstruct_previous_depth;
pass.shader->initialize(modes, general_defines);
pass.shader_version = pass.shader->version_create();
pass.sampled_bindings = {
FfxResourceBinding{ 0, 0, L"r_input_motion_vectors" },
FfxResourceBinding{ 1, 0, L"r_input_depth" },
FfxResourceBinding{ 2, 0, L"r_input_color_jittered" },
FfxResourceBinding{ 3, 0, L"r_input_exposure" },
FfxResourceBinding{ 4, 0, L"r_luma_history" }
};
pass.storage_bindings = {
FfxResourceBinding{ 5, 0, L"rw_reconstructed_previous_nearest_depth" },
FfxResourceBinding{ 6, 0, L"rw_dilated_motion_vectors" },
FfxResourceBinding{ 7, 0, L"rw_dilatedDepth" },
FfxResourceBinding{ 8, 0, L"rw_prepared_input_color" },
FfxResourceBinding{ 9, 0, L"rw_luma_history" },
// FSR2_BIND_UAV_LUMA_INSTABILITY (10) does not point to anything.
FfxResourceBinding{ 11, 0, L"rw_lock_input_luma" }
};
pass.uniform_bindings = {
FfxResourceBinding{ 12, 0, L"cbFSR2" }
};
}
{
Pass &pass = device.passes[FFX_FSR2_PASS_LOCK];
pass.shader = &shaders.lock;
pass.shader->initialize(modes, general_defines);
pass.shader_version = pass.shader->version_create();
pass.sampled_bindings = {
FfxResourceBinding{ 0, 0, L"r_lock_input_luma" }
};
pass.storage_bindings = {
FfxResourceBinding{ 1, 0, L"rw_new_locks" },
FfxResourceBinding{ 2, 0, L"rw_reconstructed_previous_nearest_depth" }
};
pass.uniform_bindings = {
FfxResourceBinding{ 3, 0, L"cbFSR2" }
};
}
{
Vector<String> accumulate_modes;
accumulate_modes.push_back("\n");
accumulate_modes.push_back("\n#define FFX_FSR2_OPTION_APPLY_SHARPENING 1\n");
String general_defines_accumulate;
if (RD::get_singleton()->get_device_vendor_name() == "NVIDIA") {
// Workaround: Disable FP16 path for the accumulate pass on NVIDIA due to reduced occupancy and high VRAM throughput.
general_defines_accumulate = general_defines_base;
} else {
general_defines_accumulate = general_defines;
}
Pass &pass = device.passes[FFX_FSR2_PASS_ACCUMULATE];
pass.shader = &shaders.accumulate;
pass.shader->initialize(accumulate_modes, general_defines_accumulate);
pass.shader_version = pass.shader->version_create();
pass.sampled_bindings = {
FfxResourceBinding{ 0, 0, L"r_input_exposure" },
FfxResourceBinding{ 1, 0, L"r_dilated_reactive_masks" },
FfxResourceBinding{ 2, 0, L"r_input_motion_vectors" },
FfxResourceBinding{ 3, 0, L"r_internal_upscaled_color" },
FfxResourceBinding{ 4, 0, L"r_lock_status" },
FfxResourceBinding{ 5, 0, L"r_input_depth" },
FfxResourceBinding{ 6, 0, L"r_prepared_input_color" },
// FSR2_BIND_SRV_LUMA_INSTABILITY(7) does not point to anything.
FfxResourceBinding{ 8, 0, L"r_lanczos_lut" },
FfxResourceBinding{ 9, 0, L"r_upsample_maximum_bias_lut" },
FfxResourceBinding{ 10, 0, L"r_imgMips" },
FfxResourceBinding{ 11, 0, L"r_auto_exposure" },
FfxResourceBinding{ 12, 0, L"r_luma_history" }
};
pass.storage_bindings = {
FfxResourceBinding{ 13, 0, L"rw_internal_upscaled_color" },
FfxResourceBinding{ 14, 0, L"rw_lock_status" },
FfxResourceBinding{ 15, 0, L"rw_upscaled_output" },
FfxResourceBinding{ 16, 0, L"rw_new_locks" },
FfxResourceBinding{ 17, 0, L"rw_luma_history" }
};
pass.uniform_bindings = {
FfxResourceBinding{ 18, 0, L"cbFSR2" }
};
// Sharpen pass is a clone of the accumulate pass.
Pass &sharpen_pass = device.passes[FFX_FSR2_PASS_ACCUMULATE_SHARPEN];
sharpen_pass = pass;
sharpen_pass.shader_variant = 1;
}
{
Pass &pass = device.passes[FFX_FSR2_PASS_RCAS];
pass.shader = &shaders.rcas;
pass.shader->initialize(modes, general_defines_base);
pass.shader_version = pass.shader->version_create();
pass.sampled_bindings = {
FfxResourceBinding{ 0, 0, L"r_input_exposure" },
FfxResourceBinding{ 1, 0, L"r_rcas_input" }
};
pass.storage_bindings = {
FfxResourceBinding{ 2, 0, L"rw_upscaled_output" }
};
pass.uniform_bindings = {
FfxResourceBinding{ 3, 0, L"cbFSR2" },
FfxResourceBinding{ 4, 0, L"cbRCAS" }
};
}
{
Pass &pass = device.passes[FFX_FSR2_PASS_COMPUTE_LUMINANCE_PYRAMID];
pass.shader = &shaders.compute_luminance_pyramid;
pass.shader->initialize(modes, general_defines_base);
pass.shader_version = pass.shader->version_create();
pass.sampled_bindings = {
FfxResourceBinding{ 0, 0, L"r_input_color_jittered" }
};
pass.storage_bindings = {
FfxResourceBinding{ 1, 0, L"rw_spd_global_atomic" },
FfxResourceBinding{ 2, 0, L"rw_img_mip_shading_change" },
FfxResourceBinding{ 3, 0, L"rw_img_mip_5" },
FfxResourceBinding{ 4, 0, L"rw_auto_exposure" }
};
pass.uniform_bindings = {
FfxResourceBinding{ 5, 0, L"cbFSR2" },
FfxResourceBinding{ 6, 0, L"cbSPD" }
};
}
{
Pass &pass = device.passes[FFX_FSR2_PASS_GENERATE_REACTIVE];
pass.shader = &shaders.autogen_reactive;
pass.shader->initialize(modes, general_defines);
pass.shader_version = pass.shader->version_create();
pass.sampled_bindings = {
FfxResourceBinding{ 0, 0, L"r_input_opaque_only" },
FfxResourceBinding{ 1, 0, L"r_input_color_jittered" }
};
pass.storage_bindings = {
FfxResourceBinding{ 2, 0, L"rw_output_autoreactive" }
};
pass.uniform_bindings = {
FfxResourceBinding{ 3, 0, L"cbGenerateReactive" },
FfxResourceBinding{ 4, 0, L"cbFSR2" }
};
}
{
Pass &pass = device.passes[FFX_FSR2_PASS_TCR_AUTOGENERATE];
pass.shader = &shaders.tcr_autogen;
pass.shader->initialize(modes, general_defines);
pass.shader_version = pass.shader->version_create();
pass.sampled_bindings = {
FfxResourceBinding{ 0, 0, L"r_input_opaque_only" },
FfxResourceBinding{ 1, 0, L"r_input_color_jittered" },
FfxResourceBinding{ 2, 0, L"r_input_motion_vectors" },
FfxResourceBinding{ 3, 0, L"r_input_prev_color_pre_alpha" },
FfxResourceBinding{ 4, 0, L"r_input_prev_color_post_alpha" },
FfxResourceBinding{ 5, 0, L"r_reactive_mask" },
FfxResourceBinding{ 6, 0, L"r_transparency_and_composition_mask" },
FfxResourceBinding{ 13, 0, L"r_input_depth" }
};
pass.storage_bindings = {
FfxResourceBinding{ 7, 0, L"rw_output_autoreactive" },
FfxResourceBinding{ 8, 0, L"rw_output_autocomposition" },
FfxResourceBinding{ 9, 0, L"rw_output_prev_color_pre_alpha" },
FfxResourceBinding{ 10, 0, L"rw_output_prev_color_post_alpha" }
};
pass.uniform_bindings = {
FfxResourceBinding{ 11, 0, L"cbFSR2" },
FfxResourceBinding{ 12, 0, L"cbGenerateReactive" }
};
}
RD::SamplerState state;
state.mag_filter = RD::SAMPLER_FILTER_NEAREST;
state.min_filter = RD::SAMPLER_FILTER_NEAREST;
state.repeat_u = RD::SAMPLER_REPEAT_MODE_CLAMP_TO_EDGE;
state.repeat_v = RD::SAMPLER_REPEAT_MODE_CLAMP_TO_EDGE;
state.repeat_w = RD::SAMPLER_REPEAT_MODE_CLAMP_TO_EDGE;
state.min_lod = -1000.0f;
state.max_lod = 1000.0f;
state.anisotropy_max = 1.0;
device.point_clamp_sampler = RD::get_singleton()->sampler_create(state);
ERR_FAIL_COND(device.point_clamp_sampler.is_null());
state.mag_filter = RD::SAMPLER_FILTER_LINEAR;
state.min_filter = RD::SAMPLER_FILTER_LINEAR;
device.linear_clamp_sampler = RD::get_singleton()->sampler_create(state);
ERR_FAIL_COND(device.linear_clamp_sampler.is_null());
}
FSR2Effect::~FSR2Effect() {
RD::get_singleton()->free(device.point_clamp_sampler);
RD::get_singleton()->free(device.linear_clamp_sampler);
for (uint32_t i = 0; i < FFX_FSR2_PASS_COUNT; i++) {
RD::get_singleton()->free(device.passes[i].pipeline.pipeline_rid);
device.passes[i].shader->version_free(device.passes[i].shader_version);
}
}
FSR2Context *FSR2Effect::create_context(Size2i p_internal_size, Size2i p_target_size) {
FSR2Context *context = memnew(RendererRD::FSR2Context);
context->fsr_desc.flags = FFX_FSR2_ENABLE_HIGH_DYNAMIC_RANGE;
context->fsr_desc.maxRenderSize.width = p_internal_size.x;
context->fsr_desc.maxRenderSize.height = p_internal_size.y;
context->fsr_desc.displaySize.width = p_target_size.x;
context->fsr_desc.displaySize.height = p_target_size.y;
context->fsr_desc.device = &device;
FfxFsr2Interface &functions = context->fsr_desc.callbacks;
functions.fpCreateBackendContext = create_backend_context_rd;
functions.fpGetDeviceCapabilities = get_device_capabilities_rd;
functions.fpDestroyBackendContext = destroy_backend_context_rd;
functions.fpCreateResource = create_resource_rd;
functions.fpRegisterResource = register_resource_rd;
functions.fpUnregisterResources = unregister_resources_rd;
functions.fpGetResourceDescription = get_resource_description_rd;
functions.fpDestroyResource = destroy_resource_rd;
functions.fpCreatePipeline = create_pipeline_rd;
functions.fpDestroyPipeline = destroy_pipeline_rd;
functions.fpScheduleGpuJob = schedule_gpu_job_rd;
functions.fpExecuteGpuJobs = execute_gpu_jobs_rd;
functions.scratchBuffer = &context->scratch;
functions.scratchBufferSize = sizeof(context->scratch);
FfxErrorCode result = ffxFsr2ContextCreate(&context->fsr_context, &context->fsr_desc);
if (result == FFX_OK) {
return context;
} else {
memdelete(context);
return nullptr;
}
}
void FSR2Effect::upscale(const Parameters &p_params) {
// TODO: Transparency & Composition mask is not implemented.
FfxFsr2DispatchDescription dispatch_desc = {};
RID color = p_params.color;
RID depth = p_params.depth;
RID velocity = p_params.velocity;
RID reactive = p_params.reactive;
RID exposure = p_params.exposure;
RID output = p_params.output;
dispatch_desc.commandList = nullptr;
dispatch_desc.color = get_resource_rd(&color, L"color");
dispatch_desc.depth = get_resource_rd(&depth, L"depth");
dispatch_desc.motionVectors = get_resource_rd(&velocity, L"velocity");
dispatch_desc.reactive = get_resource_rd(&reactive, L"reactive");
dispatch_desc.exposure = get_resource_rd(&exposure, L"exposure");
dispatch_desc.transparencyAndComposition = {};
dispatch_desc.output = get_resource_rd(&output, L"output");
dispatch_desc.colorOpaqueOnly = {};
dispatch_desc.jitterOffset.x = p_params.jitter.x;
dispatch_desc.jitterOffset.y = p_params.jitter.y;
dispatch_desc.motionVectorScale.x = float(p_params.internal_size.width);
dispatch_desc.motionVectorScale.y = float(p_params.internal_size.height);
dispatch_desc.reset = p_params.reset_accumulation;
dispatch_desc.renderSize.width = p_params.internal_size.width;
dispatch_desc.renderSize.height = p_params.internal_size.height;
dispatch_desc.enableSharpening = (p_params.sharpness > 1e-6f);
dispatch_desc.sharpness = p_params.sharpness;
dispatch_desc.frameTimeDelta = p_params.delta_time;
dispatch_desc.preExposure = 1.0f;
dispatch_desc.cameraNear = p_params.z_near;
dispatch_desc.cameraFar = p_params.z_far;
dispatch_desc.cameraFovAngleVertical = p_params.fovy;
dispatch_desc.viewSpaceToMetersFactor = 1.0f;
dispatch_desc.enableAutoReactive = false;
dispatch_desc.autoTcThreshold = 1.0f;
dispatch_desc.autoTcScale = 1.0f;
dispatch_desc.autoReactiveScale = 1.0f;
dispatch_desc.autoReactiveMax = 1.0f;
RendererRD::MaterialStorage::store_camera(p_params.reprojection, dispatch_desc.reprojectionMatrix);
FfxErrorCode result = ffxFsr2ContextDispatch(&p_params.context->fsr_context, &dispatch_desc);
ERR_FAIL_COND(result != FFX_OK);
}

199
servers/rendering/renderer_rd/effects/fsr2.h Normal file
View File

@ -0,0 +1,199 @@
/**************************************************************************/
/* fsr2.h */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
#ifndef FSR2_RD_H
#define FSR2_RD_H
#include "servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_accumulate_pass.glsl.gen.h"
#include "servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_autogen_reactive_pass.glsl.gen.h"
#include "servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_compute_luminance_pyramid_pass.glsl.gen.h"
#include "servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_depth_clip_pass.glsl.gen.h"
#include "servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_lock_pass.glsl.gen.h"
#include "servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_rcas_pass.glsl.gen.h"
#include "servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_reconstruct_previous_depth_pass.glsl.gen.h"
#include "servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_tcr_autogen_pass.glsl.gen.h"
// This flag doesn't actually control anything GCC specific in FSR2. It determines
// if symbols should be exported, which is not required for Godot.
#ifndef FFX_GCC
#define FFX_GCC
#endif
#include "thirdparty/amd-fsr2/ffx_fsr2.h"
#define FSR2_MAX_QUEUED_FRAMES (4)
#define FSR2_MAX_UNIFORM_BUFFERS (4)
#define FSR2_MAX_BUFFERED_DESCRIPTORS (FFX_FSR2_PASS_COUNT * FSR2_MAX_QUEUED_FRAMES)
#define FSR2_UBO_RING_BUFFER_SIZE (FSR2_MAX_BUFFERED_DESCRIPTORS * FSR2_MAX_UNIFORM_BUFFERS)
namespace RendererRD {
class FSR2Context {
public:
enum ResourceID : uint32_t {
RESOURCE_ID_DYNAMIC = 0xFFFFFFFF
};
struct Resources {
LocalVector<RID> rids;
LocalVector<LocalVector<RID>> mip_slice_rids;
LocalVector<uint32_t> ids;
LocalVector<FfxResourceDescription> descriptions;
LocalVector<uint32_t> dynamic_list;
LocalVector<uint32_t> free_list;
uint32_t add(RID p_rid, bool p_dynamic, uint32_t p_id, FfxResourceDescription p_description) {
uint32_t ret_index;
if (free_list.is_empty()) {
ret_index = rids.size();
uint32_t new_size = ret_index + 1;
rids.resize(new_size);
mip_slice_rids.resize(new_size);
ids.resize(new_size);
descriptions.resize(new_size);
} else {
uint32_t end_index = free_list.size() - 1;
ret_index = free_list[end_index];
free_list.resize(end_index);
}
rids[ret_index] = p_rid;
mip_slice_rids[ret_index].clear();
ids[ret_index] = p_id;
descriptions[ret_index] = p_description;
if (p_dynamic) {
dynamic_list.push_back(ret_index);
}
return ret_index;
}
void remove(uint32_t p_index) {
DEV_ASSERT(p_index < rids.size());
free_list.push_back(p_index);
rids[p_index] = RID();
mip_slice_rids[p_index].clear();
ids[p_index] = 0;
descriptions[p_index] = {};
dynamic_list.erase(p_index);
}
uint32_t size() const {
return rids.size();
}
};
struct Scratch {
Resources resources;
LocalVector<FfxGpuJobDescription> gpu_jobs;
RID ubo_ring_buffer[FSR2_UBO_RING_BUFFER_SIZE];
uint32_t ubo_ring_buffer_index = 0;
FfxDevice device = nullptr;
};
Scratch scratch;
FfxFsr2Context fsr_context;
FfxFsr2ContextDescription fsr_desc;
~FSR2Context();
};
class FSR2Effect {
public:
struct RootSignature {
// Proxy structure to store the shader required by RD that uses the terminology used by the FSR2 API.
RID shader_rid;
};
struct Pipeline {
RID pipeline_rid;
};
struct Pass {
ShaderRD *shader;
RID shader_version;
RootSignature root_signature;
uint32_t shader_variant = 0;
Pipeline pipeline;
Vector<FfxResourceBinding> sampled_bindings;
Vector<FfxResourceBinding> storage_bindings;
Vector<FfxResourceBinding> uniform_bindings;
};
struct Device {
Pass passes[FFX_FSR2_PASS_COUNT];
FfxDeviceCapabilities capabilities;
RID point_clamp_sampler;
RID linear_clamp_sampler;
};
struct Parameters {
FSR2Context *context;
Size2i internal_size;
RID color;
RID depth;
RID velocity;
RID reactive;
RID exposure;
RID output;
float z_near = 0.0f;
float z_far = 0.0f;
float fovy = 0.0f;
Vector2 jitter;
float delta_time = 0.0f;
float sharpness = 0.0f;
bool reset_accumulation = false;
Projection reprojection;
};
FSR2Effect();
~FSR2Effect();
FSR2Context *create_context(Size2i p_internal_size, Size2i p_target_size);
void upscale(const Parameters &p_params);
private:
struct {
Fsr2DepthClipPassShaderRD depth_clip;
Fsr2ReconstructPreviousDepthPassShaderRD reconstruct_previous_depth;
Fsr2LockPassShaderRD lock;
Fsr2AccumulatePassShaderRD accumulate;
Fsr2AccumulatePassShaderRD accumulate_sharpen;
Fsr2RcasPassShaderRD rcas;
Fsr2ComputeLuminancePyramidPassShaderRD compute_luminance_pyramid;
Fsr2AutogenReactivePassShaderRD autogen_reactive;
Fsr2TcrAutogenPassShaderRD tcr_autogen;
} shaders;
Device device;
};
} // namespace RendererRD
#endif // FSR2_RD_H

14
servers/rendering/renderer_rd/effects/taa.cpp
View File

@ -47,20 +47,6 @@ TAA::~TAA() {
taa_shader.version_free(shader_version);
}
void TAA::msaa_resolve(Ref<RenderSceneBuffersRD> p_render_buffers) {
if (!p_render_buffers->has_velocity_buffer(true)) {
// nothing to resolve
return;
}
for (uint32_t v = 0; v < p_render_buffers->get_view_count(); v++) {
RID velocity_buffer_msaa = p_render_buffers->get_velocity_buffer(true, v);
RID velocity_buffer = p_render_buffers->get_velocity_buffer(false, v);
RD::get_singleton()->texture_resolve_multisample(velocity_buffer_msaa, velocity_buffer);
}
}
void TAA::resolve(RID p_frame, RID p_temp, RID p_depth, RID p_velocity, RID p_prev_velocity, RID p_history, Size2 p_resolution, float p_z_near, float p_z_far) {
UniformSetCacheRD *uniform_set_cache = UniformSetCacheRD::get_singleton();
ERR_FAIL_NULL(uniform_set_cache);

1
servers/rendering/renderer_rd/effects/taa.h
View File

@ -45,7 +45,6 @@ public:
TAA();
~TAA();
void msaa_resolve(Ref<RenderSceneBuffersRD> p_render_buffers);
void process(Ref<RenderSceneBuffersRD> p_render_buffers, RD::DataFormat p_format, float p_z_near, float p_z_far);
private:

11
servers/rendering/renderer_rd/environment/sky.cpp
View File

@ -973,7 +973,7 @@ SkyRD::~SkyRD() {
}
}
void SkyRD::setup_sky(RID p_env, Ref<RenderSceneBuffersRD> p_render_buffers, const PagedArray<RID> &p_lights, RID p_camera_attributes, uint32_t p_view_count, const Projection *p_view_projections, const Vector3 *p_view_eye_offsets, const Transform3D &p_cam_transform, const Projection &p_cam_projection, const Size2i p_screen_size, RendererSceneRenderRD *p_scene_render) {
void SkyRD::setup_sky(RID p_env, Ref<RenderSceneBuffersRD> p_render_buffers, const PagedArray<RID> &p_lights, RID p_camera_attributes, uint32_t p_view_count, const Projection *p_view_projections, const Vector3 *p_view_eye_offsets, const Transform3D &p_cam_transform, const Projection &p_cam_projection, const Size2i p_screen_size, Vector2 p_jitter, RendererSceneRenderRD *p_scene_render) {
RendererRD::LightStorage *light_storage = RendererRD::LightStorage::get_singleton();
RendererRD::MaterialStorage *material_storage = RendererRD::MaterialStorage::get_singleton();
ERR_FAIL_COND(p_env.is_null());
@ -1173,18 +1173,21 @@ void SkyRD::setup_sky(RID p_env, Ref<RenderSceneBuffersRD> p_render_buffers, con
}
}
Projection correction;
correction.add_jitter_offset(p_jitter);
sky_scene_state.view_count = p_view_count;
sky_scene_state.cam_transform = p_cam_transform;
sky_scene_state.cam_projection = p_cam_projection; // We only use this when rendering a single view.
sky_scene_state.cam_projection = correction * p_cam_projection; // We only use this when rendering a single view.
// Our info in our UBO is only used if we're rendering stereo.
for (uint32_t i = 0; i < p_view_count; i++) {
Projection view_inv_projection = p_view_projections[i].inverse();
Projection view_inv_projection = (correction * p_view_projections[i]).inverse();
if (p_view_count > 1) {
RendererRD::MaterialStorage::store_camera(p_cam_projection * view_inv_projection, sky_scene_state.ubo.combined_reprojection[i]);
} else {
Projection ident;
RendererRD::MaterialStorage::store_camera(ident, sky_scene_state.ubo.combined_reprojection[i]);
RendererRD::MaterialStorage::store_camera(correction, sky_scene_state.ubo.combined_reprojection[i]);
}
RendererRD::MaterialStorage::store_camera(view_inv_projection, sky_scene_state.ubo.view_inv_projections[i]);

2
servers/rendering/renderer_rd/environment/sky.h
View File

@ -294,7 +294,7 @@ public:
void set_texture_format(RD::DataFormat p_texture_format);
~SkyRD();
void setup_sky(RID p_env, Ref<RenderSceneBuffersRD> p_render_buffers, const PagedArray<RID> &p_lights, RID p_camera_attributes, uint32_t p_view_count, const Projection *p_view_projections, const Vector3 *p_view_eye_offsets, const Transform3D &p_cam_transform, const Projection &p_cam_projection, const Size2i p_screen_size, RendererSceneRenderRD *p_scene_render);
void setup_sky(RID p_env, Ref<RenderSceneBuffersRD> p_render_buffers, const PagedArray<RID> &p_lights, RID p_camera_attributes, uint32_t p_view_count, const Projection *p_view_projections, const Vector3 *p_view_eye_offsets, const Transform3D &p_cam_transform, const Projection &p_cam_projection, const Size2i p_screen_size, Vector2 p_jitter, RendererSceneRenderRD *p_scene_render);
void update_radiance_buffers(Ref<RenderSceneBuffersRD> p_render_buffers, RID p_env, const Vector3 &p_global_pos, double p_time, float p_luminance_multiplier = 1.0);
void update_res_buffers(Ref<RenderSceneBuffersRD> p_render_buffers, RID p_env, double p_time, float p_luminance_multiplier = 1.0);
void draw_sky(RD::DrawListID p_draw_list, Ref<RenderSceneBuffersRD> p_render_buffers, RID p_env, RID p_fb, double p_time, float p_luminance_multiplier = 1.0);

226
servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp
View File

@ -105,6 +105,12 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::ensure_voxelgi()
}
}
void RenderForwardClustered::RenderBufferDataForwardClustered::ensure_fsr2(RendererRD::FSR2Effect *p_effect) {
if (fsr2_context == nullptr) {
fsr2_context = p_effect->create_context(render_buffers->get_internal_size(), render_buffers->get_target_size());
}
}
void RenderForwardClustered::RenderBufferDataForwardClustered::free_data() {
// JIC, should already have been cleared
if (render_buffers) {
@ -120,6 +126,11 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::free_data() {
cluster_builder = nullptr;
}
if (fsr2_context) {
memdelete(fsr2_context);
fsr2_context = nullptr;
}
if (!render_sdfgi_uniform_set.is_null() && RD::get_singleton()->uniform_set_is_valid(render_sdfgi_uniform_set)) {
RD::get_singleton()->free(render_sdfgi_uniform_set);
}
@ -230,6 +241,14 @@ RID RenderForwardClustered::RenderBufferDataForwardClustered::get_specular_only_
return FramebufferCacheRD::get_singleton()->get_cache_multiview(render_buffers->get_view_count(), specular);
}
RID RenderForwardClustered::RenderBufferDataForwardClustered::get_velocity_only_fb() {
bool use_msaa = render_buffers->get_msaa_3d() != RS::VIEWPORT_MSAA_DISABLED;
RID velocity = render_buffers->get_texture(RB_SCOPE_BUFFERS, use_msaa ? RB_TEX_VELOCITY_MSAA : RB_TEX_VELOCITY);
return FramebufferCacheRD::get_singleton()->get_cache_multiview(render_buffers->get_view_count(), velocity);
}
void RenderForwardClustered::setup_render_buffer_data(Ref<RenderSceneBuffersRD> p_render_buffers) {
Ref<RenderBufferDataForwardClustered> data;
data.instantiate();
@ -285,8 +304,10 @@ void RenderForwardClustered::_render_list_template(RenderingDevice::DrawListID p
const GeometryInstanceSurfaceDataCache *surf = p_params->elements[i];
const RenderElementInfo &element_info = p_params->element_info[i];
if ((p_pass_mode == PASS_MODE_COLOR && !(p_color_pass_flags & COLOR_PASS_FLAG_TRANSPARENT)) && !(surf->flags & GeometryInstanceSurfaceDataCache::FLAG_PASS_OPAQUE)) {
continue; // Objects with "Depth-prepass" transparency are included in both render lists, but should only be rendered in the transparent pass
if (p_pass_mode == PASS_MODE_COLOR && surf->color_pass_inclusion_mask && (p_color_pass_flags & surf->color_pass_inclusion_mask) == 0) {
// Some surfaces can be repeated in multiple render lists. We exclude them from being rendered on the color pass based on the
// features supported by the pass compared to the exclusion mask.
continue;
}
if (surf->owner->instance_count == 0) {
@ -582,7 +603,7 @@ void RenderForwardClustered::_render_list_with_threads(RenderListParameters *p_p
}
}
void RenderForwardClustered::_setup_environment(const RenderDataRD *p_render_data, bool p_no_fog, const Size2i &p_screen_size, bool p_flip_y, const Color &p_default_bg_color, bool p_opaque_render_buffers, bool p_pancake_shadows, int p_index) {
void RenderForwardClustered::_setup_environment(const RenderDataRD *p_render_data, bool p_no_fog, const Size2i &p_screen_size, bool p_flip_y, const Color &p_default_bg_color, bool p_opaque_render_buffers, bool p_apply_alpha_multiplier, bool p_pancake_shadows, int p_index) {
RendererRD::LightStorage *light_storage = RendererRD::LightStorage::get_singleton();
Ref<RenderSceneBuffersRD> rd = p_render_data->render_buffers;
@ -598,7 +619,7 @@ void RenderForwardClustered::_setup_environment(const RenderDataRD *p_render_dat
}
}
p_render_data->scene_data->update_ubo(scene_state.uniform_buffers[p_index], get_debug_draw_mode(), env, reflection_probe_instance, p_render_data->camera_attributes, p_flip_y, p_pancake_shadows, p_screen_size, p_default_bg_color, _render_buffers_get_luminance_multiplier(), p_opaque_render_buffers);
p_render_data->scene_data->update_ubo(scene_state.uniform_buffers[p_index], get_debug_draw_mode(), env, reflection_probe_instance, p_render_data->camera_attributes, p_flip_y, p_pancake_shadows, p_screen_size, p_default_bg_color, _render_buffers_get_luminance_multiplier(), p_opaque_render_buffers, p_apply_alpha_multiplier);
// now do implementation UBO
@ -775,8 +796,9 @@ _FORCE_INLINE_ static uint32_t _indices_to_primitives(RS::PrimitiveType p_primit
static const uint32_t subtractor[RS::PRIMITIVE_MAX] = { 0, 0, 1, 0, 1 };
return (p_indices - subtractor[p_primitive]) / divisor[p_primitive];
}
void RenderForwardClustered::_fill_render_list(RenderListType p_render_list, const RenderDataRD *p_render_data, PassMode p_pass_mode, uint32_t p_color_pass_flags = 0, bool p_using_sdfgi, bool p_using_opaque_gi, bool p_append) {
void RenderForwardClustered::_fill_render_list(RenderListType p_render_list, const RenderDataRD *p_render_data, PassMode p_pass_mode, bool p_using_sdfgi, bool p_using_opaque_gi, bool p_using_motion_pass, bool p_append) {
RendererRD::MeshStorage *mesh_storage = RendererRD::MeshStorage::get_singleton();
uint64_t frame = RSG::rasterizer->get_frame_number();
if (p_render_list == RENDER_LIST_OPAQUE) {
scene_state.used_sss = false;
@ -797,7 +819,9 @@ void RenderForwardClustered::_fill_render_list(RenderListType p_render_list, con
if (!p_append) {
rl->clear();
if (p_render_list == RENDER_LIST_OPAQUE) {
render_list[RENDER_LIST_ALPHA].clear(); //opaque fills alpha too
// Opaque fills motion and alpha lists.
render_list[RENDER_LIST_MOTION].clear();
render_list[RENDER_LIST_ALPHA].clear();
}
}
@ -827,6 +851,7 @@ void RenderForwardClustered::_fill_render_list(RenderListType p_render_list, con
}
bool uses_lightmap = false;
bool uses_gi = false;
bool uses_motion = false;
float fade_alpha = 1.0;
if (inst->fade_near || inst->fade_far) {
@ -914,6 +939,14 @@ void RenderForwardClustered::_fill_render_list(RenderListType p_render_list, con
inst->gi_offset_cache = 0xFFFFFFFF;
}
}
if (p_pass_mode == PASS_MODE_COLOR && p_using_motion_pass) {
bool transform_changed = inst->prev_transform_change_frame == frame;
bool has_mesh_instance = inst->mesh_instance.is_valid();
bool uses_particles = inst->base_flags & INSTANCE_DATA_FLAG_PARTICLES;
bool is_multimesh_with_motion = !uses_particles && (inst->base_flags & INSTANCE_DATA_FLAG_MULTIMESH) && mesh_storage->_multimesh_uses_motion_vectors_offsets(inst->data->base);
uses_motion = transform_changed || has_mesh_instance || uses_particles || is_multimesh_with_motion;
}
}
inst->flags_cache = flags;
@ -990,11 +1023,18 @@ void RenderForwardClustered::_fill_render_list(RenderListType p_render_list, con
if (!force_alpha && (surf->flags & (GeometryInstanceSurfaceDataCache::FLAG_PASS_DEPTH | GeometryInstanceSurfaceDataCache::FLAG_PASS_OPAQUE))) {
rl->add_element(surf);
}
if (force_alpha || (surf->flags & GeometryInstanceSurfaceDataCache::FLAG_PASS_ALPHA)) {
surf->color_pass_inclusion_mask = COLOR_PASS_FLAG_TRANSPARENT;
render_list[RENDER_LIST_ALPHA].add_element(surf);
if (uses_gi) {
surf->sort.uses_forward_gi = 1;
}
} else if (p_using_motion_pass && (uses_motion || (surf->flags & GeometryInstanceSurfaceDataCache::FLAG_USES_MOTION_VECTOR))) {
surf->color_pass_inclusion_mask = COLOR_PASS_FLAG_MOTION_VECTORS;
render_list[RENDER_LIST_MOTION].add_element(surf);
} else {
surf->color_pass_inclusion_mask = 0;
}
if (uses_lightmap) {
@ -1580,16 +1620,24 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
RENDER_TIMESTAMP("Setup 3D Scene");
bool using_debug_mvs = get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_MOTION_VECTORS;
bool using_taa = rb->get_use_taa();
bool using_fsr2 = rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR2;
// check if we need motion vectors
if (get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_MOTION_VECTORS) {
p_render_data->scene_data->calculate_motion_vectors = true;
} else if (!is_reflection_probe && rb->get_use_taa()) {
p_render_data->scene_data->calculate_motion_vectors = true;
bool motion_vectors_required;
if (using_debug_mvs) {
motion_vectors_required = true;
} else if (!is_reflection_probe && using_taa) {
motion_vectors_required = true;
} else if (!is_reflection_probe && using_fsr2) {
motion_vectors_required = true;
} else {
p_render_data->scene_data->calculate_motion_vectors = false;
motion_vectors_required = false;
}
//p_render_data->scene_data->subsurface_scatter_width = subsurface_scatter_size;
p_render_data->scene_data->calculate_motion_vectors = motion_vectors_required;
p_render_data->scene_data->directional_light_count = 0;
p_render_data->scene_data->opaque_prepass_threshold = 0.99f;
@ -1607,6 +1655,7 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
bool using_voxelgi = false;
bool reverse_cull = p_render_data->scene_data->cam_transform.basis.determinant() < 0;
bool using_ssil = !is_reflection_probe && p_render_data->environment.is_valid() && environment_get_ssil_enabled(p_render_data->environment);
bool using_motion_pass = rb_data.is_valid() && using_fsr2;
if (is_reflection_probe) {
uint32_t resolution = light_storage->reflection_probe_instance_get_resolution(p_render_data->reflection_probe);
@ -1625,7 +1674,7 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
} else {
screen_size = rb->get_internal_size();
if (rb->get_use_taa() || get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_MOTION_VECTORS) {
if (p_render_data->scene_data->calculate_motion_vectors) {
color_pass_flags |= COLOR_PASS_FLAG_MOTION_VECTORS;
scene_shader.enable_advanced_shader_group();
}
@ -1663,12 +1712,16 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
_setup_voxelgis(*p_render_data->voxel_gi_instances);
_setup_environment(p_render_data, is_reflection_probe, screen_size, !is_reflection_probe, p_default_bg_color, false);
_update_render_base_uniform_set(rb->get_samplers()); //may have changed due to the above (light buffer enlarged, as an example)
_update_render_base_uniform_set(rb->get_samplers()); // May have changed due to the above (light buffer enlarged, as an example).
_fill_render_list(RENDER_LIST_OPAQUE, p_render_data, PASS_MODE_COLOR, color_pass_flags, using_sdfgi, using_sdfgi || using_voxelgi);
_fill_render_list(RENDER_LIST_OPAQUE, p_render_data, PASS_MODE_COLOR, using_sdfgi, using_sdfgi || using_voxelgi, using_motion_pass);
render_list[RENDER_LIST_OPAQUE].sort_by_key();
render_list[RENDER_LIST_MOTION].sort_by_key();
render_list[RENDER_LIST_ALPHA].sort_by_reverse_depth_and_priority();
_fill_instance_data(RENDER_LIST_OPAQUE, p_render_data->render_info ? p_render_data->render_info->info[RS::VIEWPORT_RENDER_INFO_TYPE_VISIBLE] : (int *)nullptr);
int *render_info = p_render_data->render_info ? p_render_data->render_info->info[RS::VIEWPORT_RENDER_INFO_TYPE_VISIBLE] : (int *)nullptr;
_fill_instance_data(RENDER_LIST_OPAQUE, render_info);
_fill_instance_data(RENDER_LIST_MOTION, render_info);
_fill_instance_data(RENDER_LIST_ALPHA);
RD::get_singleton()->draw_command_end_label();
@ -1792,9 +1845,9 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
correction.set_depth_correction(true);
Projection projection = correction * p_render_data->scene_data->cam_projection;
sky.setup_sky(p_render_data->environment, rb, *p_render_data->lights, p_render_data->camera_attributes, 1, &projection, &eye_offset, p_render_data->scene_data->cam_transform, projection, screen_size, this);
sky.setup_sky(p_render_data->environment, rb, *p_render_data->lights, p_render_data->camera_attributes, 1, &projection, &eye_offset, p_render_data->scene_data->cam_transform, projection, screen_size, Vector2(0.0f, 0.0f), this);
} else {
sky.setup_sky(p_render_data->environment, rb, *p_render_data->lights, p_render_data->camera_attributes, p_render_data->scene_data->view_count, p_render_data->scene_data->view_projection, p_render_data->scene_data->view_eye_offset, p_render_data->scene_data->cam_transform, p_render_data->scene_data->cam_projection, screen_size, this);
sky.setup_sky(p_render_data->environment, rb, *p_render_data->lights, p_render_data->camera_attributes, p_render_data->scene_data->view_count, p_render_data->scene_data->view_projection, p_render_data->scene_data->view_eye_offset, p_render_data->scene_data->cam_transform, p_render_data->scene_data->cam_projection, screen_size, p_render_data->scene_data->taa_jitter, this);
}
sky_energy_multiplier *= bg_energy_multiplier;
@ -1892,37 +1945,71 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
// Shadow pass can change the base uniform set samplers.
_update_render_base_uniform_set(rb->get_samplers());
_setup_environment(p_render_data, is_reflection_probe, screen_size, !is_reflection_probe, p_default_bg_color, true);
_setup_environment(p_render_data, is_reflection_probe, screen_size, !is_reflection_probe, p_default_bg_color, true, using_motion_pass);
RENDER_TIMESTAMP("Render Opaque Pass");
RID rp_uniform_set = _setup_render_pass_uniform_set(RENDER_LIST_OPAQUE, p_render_data, radiance_texture, true);
bool can_continue_color = !scene_state.used_screen_texture && !using_ssr && !using_sss;
bool can_continue_depth = !(scene_state.used_depth_texture || scene_state.used_normal_texture) && !using_ssr && !using_sss;
{
bool render_motion_pass = !render_list[RENDER_LIST_MOTION].elements.is_empty();
bool will_continue_color = (can_continue_color || draw_sky || draw_sky_fog_only || debug_voxelgis || debug_sdfgi_probes);
bool will_continue_depth = (can_continue_depth || draw_sky || draw_sky_fog_only || debug_voxelgis || debug_sdfgi_probes);
RD::FinalAction final_color_action = will_continue_color ? RD::FINAL_ACTION_CONTINUE : RD::FINAL_ACTION_READ;
RD::FinalAction final_depth_action = will_continue_depth ? RD::FINAL_ACTION_CONTINUE : RD::FINAL_ACTION_READ;
Vector<Color> c;
{
Color cc = clear_color.srgb_to_linear();
if (using_separate_specular || rb_data.is_valid()) {
cc.a = 0; //subsurf scatter must be 0
}
c.push_back(cc);
Vector<Color> c;
{
Color cc = clear_color.srgb_to_linear();
if (using_separate_specular || rb_data.is_valid()) {
// Effects that rely on separate specular, like subsurface scattering, must clear the alpha to zero.
cc.a = 0;
}
c.push_back(cc);
if (rb_data.is_valid()) {
c.push_back(Color(0, 0, 0, 0)); // Separate specular
c.push_back(Color(0, 0, 0, 0)); // Motion vectors
if (rb_data.is_valid()) {
c.push_back(Color(0, 0, 0, 0)); // Separate specular.
c.push_back(Color(0, 0, 0, 0)); // Motion vector. Pushed to the clear color vector even if the framebuffer isn't bound.
}
}
uint32_t opaque_color_pass_flags = using_motion_pass ? (color_pass_flags & ~COLOR_PASS_FLAG_MOTION_VECTORS) : color_pass_flags;
RID opaque_framebuffer = using_motion_pass ? rb_data->get_color_pass_fb(opaque_color_pass_flags) : color_framebuffer;
RenderListParameters render_list_params(render_list[RENDER_LIST_OPAQUE].elements.ptr(), render_list[RENDER_LIST_OPAQUE].element_info.ptr(), render_list[RENDER_LIST_OPAQUE].elements.size(), reverse_cull, PASS_MODE_COLOR, opaque_color_pass_flags, rb_data.is_null(), p_render_data->directional_light_soft_shadows, rp_uniform_set, get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_WIREFRAME, Vector2(), p_render_data->scene_data->lod_distance_multiplier, p_render_data->scene_data->screen_mesh_lod_threshold, p_render_data->scene_data->view_count);
_render_list_with_threads(&render_list_params, opaque_framebuffer, keep_color ? RD::INITIAL_ACTION_KEEP : RD::INITIAL_ACTION_CLEAR, render_motion_pass ? RD::FINAL_ACTION_CONTINUE : final_color_action, depth_pre_pass ? (continue_depth ? RD::INITIAL_ACTION_CONTINUE : RD::INITIAL_ACTION_KEEP) : RD::INITIAL_ACTION_CLEAR, render_motion_pass ? RD::FINAL_ACTION_CONTINUE : final_depth_action, c, 1.0, 0);
}
RD::get_singleton()->draw_command_end_label();
if (using_motion_pass) {
Vector<Color> motion_vector_clear_colors;
motion_vector_clear_colors.push_back(Color(-1, -1, 0, 0));
RD::get_singleton()->draw_list_begin(rb_data->get_velocity_only_fb(), RD::INITIAL_ACTION_CLEAR, render_motion_pass ? RD::FINAL_ACTION_CONTINUE : RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_CONTINUE, RD::FINAL_ACTION_CONTINUE, motion_vector_clear_colors);
RD::get_singleton()->draw_list_end();
}
if (render_motion_pass) {
RD::get_singleton()->draw_command_begin_label("Render Motion Pass");
RENDER_TIMESTAMP("Render Motion Pass");
rp_uniform_set = _setup_render_pass_uniform_set(RENDER_LIST_MOTION, p_render_data, radiance_texture, true);
RenderListParameters render_list_params(render_list[RENDER_LIST_MOTION].elements.ptr(), render_list[RENDER_LIST_MOTION].element_info.ptr(), render_list[RENDER_LIST_MOTION].elements.size(), reverse_cull, PASS_MODE_COLOR, color_pass_flags, rb_data.is_null(), p_render_data->directional_light_soft_shadows, rp_uniform_set, get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_WIREFRAME, Vector2(), p_render_data->scene_data->lod_distance_multiplier, p_render_data->scene_data->screen_mesh_lod_threshold, p_render_data->scene_data->view_count);
_render_list_with_threads(&render_list_params, color_framebuffer, RD::INITIAL_ACTION_CONTINUE, final_color_action, RD::INITIAL_ACTION_CONTINUE, final_depth_action);
if (will_continue_color) {
// Close the motion vectors framebuffer as it'll no longer be used.
RD::get_singleton()->draw_list_begin(rb_data->get_velocity_only_fb(), RD::INITIAL_ACTION_CONTINUE, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_CONTINUE, RD::FINAL_ACTION_CONTINUE);
RD::get_singleton()->draw_list_end();
}
}
RenderListParameters render_list_params(render_list[RENDER_LIST_OPAQUE].elements.ptr(), render_list[RENDER_LIST_OPAQUE].element_info.ptr(), render_list[RENDER_LIST_OPAQUE].elements.size(), reverse_cull, PASS_MODE_COLOR, color_pass_flags, rb_data.is_null(), p_render_data->directional_light_soft_shadows, rp_uniform_set, get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_WIREFRAME, Vector2(), p_render_data->scene_data->lod_distance_multiplier, p_render_data->scene_data->screen_mesh_lod_threshold, p_render_data->scene_data->view_count);
_render_list_with_threads(&render_list_params, color_framebuffer, keep_color ? RD::INITIAL_ACTION_KEEP : RD::INITIAL_ACTION_CLEAR, will_continue_color ? RD::FINAL_ACTION_CONTINUE : RD::FINAL_ACTION_READ, depth_pre_pass ? (continue_depth ? RD::INITIAL_ACTION_CONTINUE : RD::INITIAL_ACTION_KEEP) : RD::INITIAL_ACTION_CLEAR, will_continue_depth ? RD::FINAL_ACTION_CONTINUE : RD::FINAL_ACTION_READ, c, 1.0, 0);
if (will_continue_color && using_separate_specular) {
// close the specular framebuffer, as it's no longer used
// Close the specular framebuffer as it'll no longer be used.
RD::get_singleton()->draw_list_begin(rb_data->get_specular_only_fb(), RD::INITIAL_ACTION_CONTINUE, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_CONTINUE, RD::FINAL_ACTION_CONTINUE);
RD::get_singleton()->draw_list_end();
}
@ -2052,6 +2139,11 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
{
uint32_t transparent_color_pass_flags = (color_pass_flags | COLOR_PASS_FLAG_TRANSPARENT) & ~(COLOR_PASS_FLAG_SEPARATE_SPECULAR);
if (using_motion_pass) {
// Motion vectors on transparent draw calls are not required when using the reactive mask.
transparent_color_pass_flags &= ~(COLOR_PASS_FLAG_MOTION_VECTORS);
}
RID alpha_framebuffer = rb_data.is_valid() ? rb_data->get_color_pass_fb(transparent_color_pass_flags) : color_only_framebuffer;
RenderListParameters render_list_params(render_list[RENDER_LIST_ALPHA].elements.ptr(), render_list[RENDER_LIST_ALPHA].element_info.ptr(), render_list[RENDER_LIST_ALPHA].elements.size(), false, PASS_MODE_COLOR, transparent_color_pass_flags, rb_data.is_null(), p_render_data->directional_light_soft_shadows, rp_uniform_set, get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_WIREFRAME, Vector2(), p_render_data->scene_data->lod_distance_multiplier, p_render_data->scene_data->screen_mesh_lod_threshold, p_render_data->scene_data->view_count);
_render_list_with_threads(&render_list_params, alpha_framebuffer, can_continue_color ? RD::INITIAL_ACTION_CONTINUE : RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, can_continue_depth ? RD::INITIAL_ACTION_CONTINUE : RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ);
@ -2064,12 +2156,14 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
RD::get_singleton()->draw_command_begin_label("Resolve");
if (rb_data.is_valid() && rb->get_msaa_3d() != RS::VIEWPORT_MSAA_DISABLED) {
bool resolve_velocity_buffer = (using_taa || using_fsr2) && rb->has_velocity_buffer(true);
for (uint32_t v = 0; v < rb->get_view_count(); v++) {
RD::get_singleton()->texture_resolve_multisample(rb->get_color_msaa(v), rb->get_internal_texture(v));
resolve_effects->resolve_depth(rb->get_depth_msaa(v), rb->get_depth_texture(v), rb->get_internal_size(), texture_multisamples[rb->get_msaa_3d()]);
}
if (taa && rb->get_use_taa()) {
taa->msaa_resolve(rb);
if (resolve_velocity_buffer) {
RD::get_singleton()->texture_resolve_multisample(rb->get_velocity_buffer(true, v), rb->get_velocity_buffer(false, v));
}
}
}
@ -2082,9 +2176,51 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
}
RD::get_singleton()->draw_command_end_label();
if (rb_data.is_valid() && taa && rb->get_use_taa()) {
RENDER_TIMESTAMP("TAA")
taa->process(rb, _render_buffers_get_color_format(), p_render_data->scene_data->z_near, p_render_data->scene_data->z_far);
if (rb_data.is_valid() && (using_fsr2 || using_taa)) {
if (using_fsr2) {
rb->ensure_upscaled();
rb_data->ensure_fsr2(fsr2_effect);
RID exposure;
if (RSG::camera_attributes->camera_attributes_uses_auto_exposure(p_render_data->camera_attributes)) {
exposure = luminance->get_current_luminance_buffer(rb);
}
RENDER_TIMESTAMP("FSR2");
for (uint32_t v = 0; v < rb->get_view_count(); v++) {
real_t fov = p_render_data->scene_data->cam_projection.get_fov();
real_t aspect = p_render_data->scene_data->cam_projection.get_aspect();
real_t fovy = p_render_data->scene_data->cam_projection.get_fovy(fov, aspect);
Vector2 jitter = p_render_data->scene_data->taa_jitter * Vector2(rb->get_internal_size()) * 0.5f;
RendererRD::FSR2Effect::Parameters params;
params.context = rb_data->get_fsr2_context();
params.internal_size = rb->get_internal_size();
params.sharpness = CLAMP(1.0f - (rb->get_fsr_sharpness() / 2.0f), 0.0f, 1.0f);
params.color = rb->get_internal_texture(v);
params.depth = rb->get_depth_texture(v);
params.velocity = rb->get_velocity_buffer(false, v);
params.reactive = rb->get_internal_texture_reactive(v);
params.exposure = exposure;
params.output = rb->get_upscaled_texture(v);
params.z_near = p_render_data->scene_data->z_near;
params.z_far = p_render_data->scene_data->z_far;
params.fovy = fovy;
params.jitter = jitter;
params.delta_time = float(time_step);
params.reset_accumulation = false; // FIXME: The engine does not provide a way to reset the accumulation.
const Projection &prev_proj = p_render_data->scene_data->prev_cam_projection;
const Projection &cur_proj = p_render_data->scene_data->cam_projection;
const Transform3D &prev_transform = p_render_data->scene_data->prev_cam_transform;
const Transform3D &cur_transform = p_render_data->scene_data->cam_transform;
params.reprojection = prev_proj.flipped_y() * prev_transform.affine_inverse() * cur_transform * cur_proj.flipped_y().inverse();
fsr2_effect->upscale(params);
}
} else if (using_taa) {
RENDER_TIMESTAMP("TAA");
taa->process(rb, _render_buffers_get_color_format(), p_render_data->scene_data->z_near, p_render_data->scene_data->z_far);
}
}
if (rb_data.is_valid()) {
@ -2357,7 +2493,7 @@ void RenderForwardClustered::_render_shadow_append(RID p_framebuffer, const Page
render_data.instances = &p_instances;
render_data.render_info = p_render_info;
_setup_environment(&render_data, true, Vector2(1, 1), !p_flip_y, Color(), false, p_use_pancake, shadow_pass_index);
_setup_environment(&render_data, true, Vector2(1, 1), !p_flip_y, Color(), false, false, p_use_pancake, shadow_pass_index);
if (get_debug_draw_mode() == RS::VIEWPORT_DEBUG_DRAW_DISABLE_LOD) {
scene_data.screen_mesh_lod_threshold = 0.0;
@ -2368,7 +2504,7 @@ void RenderForwardClustered::_render_shadow_append(RID p_framebuffer, const Page
PassMode pass_mode = p_use_dp ? PASS_MODE_SHADOW_DP : PASS_MODE_SHADOW;
uint32_t render_list_from = render_list[RENDER_LIST_SECONDARY].elements.size();
_fill_render_list(RENDER_LIST_SECONDARY, &render_data, pass_mode, 0, false, false, true);
_fill_render_list(RENDER_LIST_SECONDARY, &render_data, pass_mode, false, false, false, true);
uint32_t render_list_size = render_list[RENDER_LIST_SECONDARY].elements.size() - render_list_from;
render_list[RENDER_LIST_SECONDARY].sort_by_key_range(render_list_from, render_list_size);
_fill_instance_data(RENDER_LIST_SECONDARY, p_render_info ? p_render_info->info[RS::VIEWPORT_RENDER_INFO_TYPE_SHADOW] : (int *)nullptr, render_list_from, render_list_size, false);
@ -2453,7 +2589,7 @@ void RenderForwardClustered::_render_particle_collider_heightfield(RID p_fb, con
_update_render_base_uniform_set(RendererRD::MaterialStorage::get_singleton()->samplers_rd_get_default());
_setup_environment(&render_data, true, Vector2(1, 1), true, Color(), false, false);
_setup_environment(&render_data, true, Vector2(1, 1), true, Color(), false, false, false);
PassMode pass_mode = PASS_MODE_SHADOW;
@ -3475,6 +3611,10 @@ void RenderForwardClustered::_geometry_instance_add_surface_with_material(Geomet
flags |= GeometryInstanceSurfaceDataCache::FLAG_USES_PARTICLE_TRAILS;
}
if (p_material->shader_data->is_animated()) {
flags |= GeometryInstanceSurfaceDataCache::FLAG_USES_MOTION_VECTOR;
}
SceneShaderForwardClustered::MaterialData *material_shadow = nullptr;
void *surface_shadow = nullptr;
if (!p_material->shader_data->uses_particle_trails && !p_material->shader_data->writes_modelview_or_projection && !p_material->shader_data->uses_vertex && !p_material->shader_data->uses_position && !p_material->shader_data->uses_discard && !p_material->shader_data->uses_depth_prepass_alpha && !p_material->shader_data->uses_alpha_clip && !p_material->shader_data->uses_alpha_antialiasing && p_material->shader_data->cull_mode == SceneShaderForwardClustered::ShaderData::CULL_BACK && !p_material->shader_data->uses_point_size) {
@ -3982,6 +4122,7 @@ RenderForwardClustered::RenderForwardClustered() {
resolve_effects = memnew(RendererRD::Resolve());
taa = memnew(RendererRD::TAA);
fsr2_effect = memnew(RendererRD::FSR2Effect);
ss_effects = memnew(RendererRD::SSEffects);
}
@ -3996,6 +4137,11 @@ RenderForwardClustered::~RenderForwardClustered() {
taa = nullptr;
}
if (fsr2_effect) {
memdelete(fsr2_effect);
fsr2_effect = nullptr;
}
if (resolve_effects != nullptr) {
memdelete(resolve_effects);
resolve_effects = nullptr;

14
servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h
View File

@ -33,6 +33,7 @@
#include "core/templates/paged_allocator.h"
#include "servers/rendering/renderer_rd/cluster_builder_rd.h"
#include "servers/rendering/renderer_rd/effects/fsr2.h"
#include "servers/rendering/renderer_rd/effects/resolve.h"
#include "servers/rendering/renderer_rd/effects/ss_effects.h"
#include "servers/rendering/renderer_rd/effects/taa.h"
@ -84,6 +85,7 @@ class RenderForwardClustered : public RendererSceneRenderRD {
enum RenderListType {
RENDER_LIST_OPAQUE, //used for opaque objects
RENDER_LIST_MOTION, //used for opaque objects with motion
RENDER_LIST_ALPHA, //used for transparent objects
RENDER_LIST_SECONDARY, //used for shadows and other objects
RENDER_LIST_MAX
@ -100,6 +102,7 @@ class RenderForwardClustered : public RendererSceneRenderRD {
private:
RenderSceneBuffersRD *render_buffers = nullptr;
RendererRD::FSR2Context *fsr2_context = nullptr;
public:
ClusterBuilderRD *cluster_builder = nullptr;
@ -140,10 +143,14 @@ class RenderForwardClustered : public RendererSceneRenderRD {
RID get_voxelgi(uint32_t p_layer) { return render_buffers->get_texture_slice(RB_SCOPE_FORWARD_CLUSTERED, RB_TEX_VOXEL_GI, p_layer, 0); }
RID get_voxelgi_msaa(uint32_t p_layer) { return render_buffers->get_texture_slice(RB_SCOPE_FORWARD_CLUSTERED, RB_TEX_VOXEL_GI_MSAA, p_layer, 0); }
void ensure_fsr2(RendererRD::FSR2Effect *p_effect);
RendererRD::FSR2Context *get_fsr2_context() const { return fsr2_context; }
RID get_color_only_fb();
RID get_color_pass_fb(uint32_t p_color_pass_flags);
RID get_depth_fb(DepthFrameBufferType p_type = DEPTH_FB);
RID get_specular_only_fb();
RID get_velocity_only_fb();
virtual void configure(RenderSceneBuffersRD *p_render_buffers) override;
virtual void free_data() override;
@ -345,7 +352,7 @@ class RenderForwardClustered : public RendererSceneRenderRD {
static RenderForwardClustered *singleton;
void _setup_environment(const RenderDataRD *p_render_data, bool p_no_fog, const Size2i &p_screen_size, bool p_flip_y, const Color &p_default_bg_color, bool p_opaque_render_buffers = false, bool p_pancake_shadows = false, int p_index = 0);
void _setup_environment(const RenderDataRD *p_render_data, bool p_no_fog, const Size2i &p_screen_size, bool p_flip_y, const Color &p_default_bg_color, bool p_opaque_render_buffers = false, bool p_apply_alpha_multiplier = false, bool p_pancake_shadows = false, int p_index = 0);
void _setup_voxelgis(const PagedArray<RID> &p_voxelgis);
void _setup_lightmaps(const RenderDataRD *p_render_data, const PagedArray<RID> &p_lightmaps, const Transform3D &p_cam_transform);
@ -372,7 +379,7 @@ class RenderForwardClustered : public RendererSceneRenderRD {
void _update_instance_data_buffer(RenderListType p_render_list);
void _fill_instance_data(RenderListType p_render_list, int *p_render_info = nullptr, uint32_t p_offset = 0, int32_t p_max_elements = -1, bool p_update_buffer = true);
void _fill_render_list(RenderListType p_render_list, const RenderDataRD *p_render_data, PassMode p_pass_mode, uint32_t p_color_pass_flags, bool p_using_sdfgi = false, bool p_using_opaque_gi = false, bool p_append = false);
void _fill_render_list(RenderListType p_render_list, const RenderDataRD *p_render_data, PassMode p_pass_mode, bool p_using_sdfgi = false, bool p_using_opaque_gi = false, bool p_using_motion_pass = false, bool p_append = false);
HashMap<Size2i, RID> sdfgi_framebuffer_size_cache;
@ -397,6 +404,7 @@ class RenderForwardClustered : public RendererSceneRenderRD {
FLAG_USES_NORMAL_TEXTURE = 16384,
FLAG_USES_DOUBLE_SIDED_SHADOWS = 32768,
FLAG_USES_PARTICLE_TRAILS = 65536,
FLAG_USES_MOTION_VECTOR = 131072,
};
union {
@ -424,6 +432,7 @@ class RenderForwardClustered : public RendererSceneRenderRD {
RS::PrimitiveType primitive = RS::PRIMITIVE_MAX;
uint32_t flags = 0;
uint32_t surface_index = 0;
uint32_t color_pass_inclusion_mask = 0;
void *surface = nullptr;
RID material_uniform_set;
@ -563,6 +572,7 @@ class RenderForwardClustered : public RendererSceneRenderRD {
RendererRD::Resolve *resolve_effects = nullptr;
RendererRD::TAA *taa = nullptr;
RendererRD::FSR2Effect *fsr2_effect = nullptr;
RendererRD::SSEffects *ss_effects = nullptr;
/* Cluster builder */

6
servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp
View File

@ -823,9 +823,9 @@ void RenderForwardMobile::_render_scene(RenderDataRD *p_render_data, const Color
correction.set_depth_correction(true);
Projection projection = correction * p_render_data->scene_data->cam_projection;
sky.setup_sky(p_render_data->environment, p_render_data->render_buffers, *p_render_data->lights, p_render_data->camera_attributes, 1, &projection, &eye_offset, p_render_data->scene_data->cam_transform, projection, screen_size, this);
sky.setup_sky(p_render_data->environment, p_render_data->render_buffers, *p_render_data->lights, p_render_data->camera_attributes, 1, &projection, &eye_offset, p_render_data->scene_data->cam_transform, projection, screen_size, Vector2(0.0f, 0.0f), this);
} else {
sky.setup_sky(p_render_data->environment, p_render_data->render_buffers, *p_render_data->lights, p_render_data->camera_attributes, p_render_data->scene_data->view_count, p_render_data->scene_data->view_projection, p_render_data->scene_data->view_eye_offset, p_render_data->scene_data->cam_transform, p_render_data->scene_data->cam_projection, screen_size, this);
sky.setup_sky(p_render_data->environment, p_render_data->render_buffers, *p_render_data->lights, p_render_data->camera_attributes, p_render_data->scene_data->view_count, p_render_data->scene_data->view_projection, p_render_data->scene_data->view_eye_offset, p_render_data->scene_data->cam_transform, p_render_data->scene_data->cam_projection, screen_size, p_render_data->scene_data->taa_jitter, this);
}
sky_energy_multiplier *= bg_energy_multiplier;
@ -1908,7 +1908,7 @@ void RenderForwardMobile::_setup_environment(const RenderDataRD *p_render_data,
}
}
p_render_data->scene_data->update_ubo(scene_state.uniform_buffers[p_index], get_debug_draw_mode(), env, reflection_probe_instance, p_render_data->camera_attributes, p_flip_y, p_pancake_shadows, p_screen_size, p_default_bg_color, _render_buffers_get_luminance_multiplier(), p_opaque_render_buffers);
p_render_data->scene_data->update_ubo(scene_state.uniform_buffers[p_index], get_debug_draw_mode(), env, reflection_probe_instance, p_render_data->camera_attributes, p_flip_y, p_pancake_shadows, p_screen_size, p_default_bg_color, _render_buffers_get_luminance_multiplier(), p_opaque_render_buffers, false);
}
void RenderForwardMobile::_fill_element_info(RenderListType p_render_list, uint32_t p_offset, int32_t p_max_elements) {

37
servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
View File

@ -340,14 +340,16 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende
// Glow, auto exposure and DoF (if enabled).
Size2i internal_size = rb->get_internal_size();
Size2i target_size = rb->get_target_size();
bool can_use_effects = target_size.x >= 8 && target_size.y >= 8; // FIXME I think this should check internal size, we do all our post processing at this size...
bool can_use_storage = _render_buffers_can_be_storage();
bool use_fsr = fsr && can_use_effects && rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR;
bool use_upscaled_texture = rb->has_upscaled_texture() && rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR2;
RID render_target = rb->get_render_target();
RID internal_texture = rb->get_internal_texture();
RID color_texture = use_upscaled_texture ? rb->get_upscaled_texture() : rb->get_internal_texture();
Size2i color_size = use_upscaled_texture ? target_size : rb->get_internal_size();
if (can_use_effects && RSG::camera_attributes->camera_attributes_uses_dof(p_render_data->camera_attributes)) {
RENDER_TIMESTAMP("Depth of Field");
@ -358,14 +360,14 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende
RendererRD::BokehDOF::BokehBuffers buffers;
// Textures we use
buffers.base_texture_size = rb->get_internal_size();
buffers.base_texture_size = color_size;
buffers.secondary_texture = rb->get_texture_slice(RB_SCOPE_BUFFERS, RB_TEX_BLUR_0, 0, 0);
buffers.half_texture[0] = rb->get_texture_slice(RB_SCOPE_BUFFERS, RB_TEX_BLUR_1, 0, 0);
buffers.half_texture[1] = rb->get_texture_slice(RB_SCOPE_BUFFERS, RB_TEX_BLUR_0, 0, 1);
if (can_use_storage) {
for (uint32_t i = 0; i < rb->get_view_count(); i++) {
buffers.base_texture = rb->get_internal_texture(i);
buffers.base_texture = use_upscaled_texture ? rb->get_upscaled_texture(i) : rb->get_internal_texture(i);
buffers.depth_texture = rb->get_depth_texture(i);
// In stereo p_render_data->z_near and p_render_data->z_far can be offset for our combined frustum.
@ -387,7 +389,7 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende
buffers.base_weight_fb = rb->weight_buffers[0].fb;
for (uint32_t i = 0; i < rb->get_view_count(); i++) {
buffers.base_texture = rb->get_internal_texture(i);
buffers.base_texture = use_upscaled_texture ? rb->get_upscaled_texture(i) : rb->get_internal_texture(i);
buffers.depth_texture = rb->get_depth_texture(i);
buffers.base_fb = FramebufferCacheRD::get_singleton()->get_cache(buffers.base_texture); // TODO move this into bokeh_dof_raster, we can do this internally
@ -416,7 +418,7 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende
double step = RSG::camera_attributes->camera_attributes_get_auto_exposure_adjust_speed(p_render_data->camera_attributes) * time_step;
float auto_exposure_min_sensitivity = RSG::camera_attributes->camera_attributes_get_auto_exposure_min_sensitivity(p_render_data->camera_attributes);
float auto_exposure_max_sensitivity = RSG::camera_attributes->camera_attributes_get_auto_exposure_max_sensitivity(p_render_data->camera_attributes);
luminance->luminance_reduction(internal_texture, internal_size, luminance_buffers, auto_exposure_min_sensitivity, auto_exposure_max_sensitivity, step, set_immediate);
luminance->luminance_reduction(color_texture, color_size, luminance_buffers, auto_exposure_min_sensitivity, auto_exposure_max_sensitivity, step, set_immediate);
// Swap final reduce with prev luminance.
@ -525,7 +527,7 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende
}
tonemap.use_debanding = rb->get_use_debanding();
tonemap.texture_size = Vector2i(rb->get_internal_size().x, rb->get_internal_size().y);
tonemap.texture_size = Vector2i(color_size.x, color_size.y);
if (p_render_data->environment.is_valid()) {
tonemap.tonemap_mode = environment_get_tone_mapper(p_render_data->environment);
@ -555,7 +557,8 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende
tonemap.convert_to_srgb = !texture_storage->render_target_is_using_hdr(render_target);
RID dest_fb;
if (fsr && can_use_effects && rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR) {
bool use_intermediate_fb = use_fsr;
if (use_intermediate_fb) {
// If we use FSR to upscale we need to write our result into an intermediate buffer.
// Note that this is cached so we only create the texture the first time.
RID dest_texture = rb->create_texture(SNAME("Tonemapper"), SNAME("destination"), _render_buffers_get_color_format(), RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT | RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT);
@ -567,12 +570,12 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende
dest_fb = texture_storage->render_target_get_rd_framebuffer(render_target);
}
tone_mapper->tonemapper(internal_texture, dest_fb, tonemap);
tone_mapper->tonemapper(color_texture, dest_fb, tonemap);
RD::get_singleton()->draw_command_end_label();
}
if (fsr && can_use_effects && rb->get_scaling_3d_mode() == RS::VIEWPORT_SCALING_3D_MODE_FSR) {
if (use_fsr) {
RD::get_singleton()->draw_command_begin_label("FSR 1.0 Upscale");
for (uint32_t v = 0; v < rb->get_view_count(); v++) {
@ -732,6 +735,11 @@ void RendererSceneRenderRD::_render_buffers_debug_draw(const RenderDataRD *p_ren
}
}
if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_INTERNAL_BUFFER) {
Size2 rtsize = texture_storage->render_target_get_size(render_target);
copy_effects->copy_to_fb_rect(rb->get_internal_texture(), texture_storage->render_target_get_rd_framebuffer(render_target), Rect2(Vector2(), rtsize), false, false);
}
if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_NORMAL_BUFFER && _render_buffers_get_normal_texture(rb).is_valid()) {
Size2 rtsize = texture_storage->render_target_get_size(render_target);
copy_effects->copy_to_fb_rect(_render_buffers_get_normal_texture(rb), texture_storage->render_target_get_rd_framebuffer(render_target), Rect2(Vector2(), rtsize), false, false);
@ -745,7 +753,12 @@ void RendererSceneRenderRD::_render_buffers_debug_draw(const RenderDataRD *p_ren
}
if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_MOTION_VECTORS && _render_buffers_get_velocity_texture(rb).is_valid()) {
debug_effects->draw_motion_vectors(_render_buffers_get_velocity_texture(rb), texture_storage->render_target_get_rd_framebuffer(render_target), rb->get_internal_size());
RID velocity = _render_buffers_get_velocity_texture(rb);
RID depth = rb->get_depth_texture();
RID dest_fb = texture_storage->render_target_get_rd_framebuffer(render_target);
Size2i resolution = rb->get_internal_size();
debug_effects->draw_motion_vectors(velocity, depth, dest_fb, p_render_data->scene_data->cam_projection, p_render_data->scene_data->cam_transform, p_render_data->scene_data->prev_cam_projection, p_render_data->scene_data->prev_cam_transform, resolution);
}
}

2
servers/rendering/renderer_rd/shaders/effects/SCsub
View File

@ -15,3 +15,5 @@ if "RD_GLSL" in env["BUILDERS"]:
# compile shaders
for glsl_file in glsl_files:
env.RD_GLSL(glsl_file)
SConscript("fsr2/SCsub")

17
servers/rendering/renderer_rd/shaders/effects/fsr2/SCsub Normal file
View File

@ -0,0 +1,17 @@
#!/usr/bin/env python
Import("env")
if "RD_GLSL" in env["BUILDERS"]:
# find all include files
gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + [str(f) for f in Glob("../*_inc.glsl")]
# find all shader code(all glsl files excluding our include files)
glsl_files = [str(f) for f in Glob("*.glsl") if str(f) not in gl_include_files]
# make sure we recompile shaders if include files change
env.Depends([f + ".gen.h" for f in glsl_files], gl_include_files + ["#glsl_builders.py"])
# compile shaders
for glsl_file in glsl_files:
env.RD_GLSL(glsl_file)

8
servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_accumulate_pass.glsl Normal file
View File

@ -0,0 +1,8 @@
#[compute]
#version 450
#VERSION_DEFINES
#include "../motion_vector_inc.glsl"
#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate_pass.glsl"

8
servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_autogen_reactive_pass.glsl Normal file
View File

@ -0,0 +1,8 @@
#[compute]
#version 450
#VERSION_DEFINES
#include "../motion_vector_inc.glsl"
#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_autogen_reactive_pass.glsl"

7
servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_compute_luminance_pyramid_pass.glsl Normal file
View File

@ -0,0 +1,7 @@
#[compute]
#version 450
#VERSION_DEFINES
#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl"

8
servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_depth_clip_pass.glsl Normal file
View File

@ -0,0 +1,8 @@
#[compute]
#version 450
#VERSION_DEFINES
#include "../motion_vector_inc.glsl"
#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip_pass.glsl"

7
servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_lock_pass.glsl Normal file
View File

@ -0,0 +1,7 @@
#[compute]
#version 450
#VERSION_DEFINES
#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_lock_pass.glsl"

7
servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_rcas_pass.glsl Normal file
View File

@ -0,0 +1,7 @@
#[compute]
#version 450
#VERSION_DEFINES
#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas_pass.glsl"

8
servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_reconstruct_previous_depth_pass.glsl Normal file
View File

@ -0,0 +1,8 @@
#[compute]
#version 450
#VERSION_DEFINES
#include "../motion_vector_inc.glsl"
#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl"

8
servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_tcr_autogen_pass.glsl Normal file
View File

@ -0,0 +1,8 @@
#[compute]
#version 450
#VERSION_DEFINES
#include "../motion_vector_inc.glsl"
#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen_pass.glsl"

6
servers/rendering/renderer_rd/shaders/effects/motion_vector_inc.glsl Normal file
View File

@ -0,0 +1,6 @@
vec2 derive_motion_vector(vec2 uv, float depth, mat4 reprojection_matrix) {
vec4 previous_pos_ndc = reprojection_matrix * vec4(uv * 2.0f - 1.0f, depth * 2.0f - 1.0f, 1.0f);
return 0.5f + (previous_pos_ndc.xy / previous_pos_ndc.w) * 0.5f - uv;
}
#define FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS_FUNCTION(i, j, k) derive_motion_vector(i, j, k)

19
servers/rendering/renderer_rd/shaders/effects/motion_vectors.glsl
View File

@ -18,14 +18,19 @@ void main() {
#VERSION_DEFINES
#include "motion_vector_inc.glsl"
layout(location = 0) in vec2 uv_interp;
layout(set = 0, binding = 0) uniform sampler2D source_velocity;
layout(set = 0, binding = 1) uniform sampler2D source_depth;
layout(location = 0) out vec4 frag_color;
layout(push_constant, std430) uniform Params {
highp mat4 reprojection_matrix;
vec2 resolution;
bool force_derive_from_depth;
}
params;
@ -49,7 +54,14 @@ void main() {
vec2 pos_pixel = uv_interp * params.resolution;
vec2 cell_pos_pixel = floor(pos_pixel / cell_size) * cell_size + (cell_size * 0.5f);
vec2 cell_pos_uv = cell_pos_pixel / params.resolution;
vec2 cell_pos_previous_uv = cell_pos_uv + textureLod(source_velocity, cell_pos_uv, 0.0f).xy;
vec2 cell_pos_velocity = textureLod(source_velocity, cell_pos_uv, 0.0f).xy;
bool derive_velocity = params.force_derive_from_depth || all(lessThanEqual(cell_pos_velocity, vec2(-1.0f, -1.0f)));
if (derive_velocity) {
float depth = textureLod(source_depth, cell_pos_uv, 0.0f).x;
cell_pos_velocity = derive_motion_vector(cell_pos_uv, depth, params.reprojection_matrix);
}
vec2 cell_pos_previous_uv = cell_pos_uv + cell_pos_velocity;
// Draw the shapes.
float epsilon = 1e-6f;
@ -76,5 +88,10 @@ void main() {
alpha = 0.0f;
}
if (derive_velocity) {
color = vec3(1.0f, 1.0f, 1.0f) - color;
alpha *= 0.5f;
}
frag_color = vec4(color, alpha);
}

2
servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl
View File

@ -2280,6 +2280,8 @@ void fragment_shader(in SceneData scene_data) {
#else //MODE_SEPARATE_SPECULAR
alpha *= scene_data.pass_alpha_multiplier;
#ifdef MODE_UNSHADED
frag_color = vec4(albedo, alpha);
#else

2
servers/rendering/renderer_rd/shaders/scene_data_inc.glsl
View File

@ -64,6 +64,6 @@ struct SceneData {
bool pancake_shadows;
uint camera_visible_layers;
uint pad2;
float pass_alpha_multiplier;
uint pad3;
};

12
servers/rendering/renderer_rd/storage_rd/mesh_storage.cpp
View File

@ -1392,12 +1392,18 @@ void MeshStorage::_multimesh_get_motion_vectors_offsets(RID p_multimesh, uint32_
MultiMesh *multimesh = multimesh_owner.get_or_null(p_multimesh);
ERR_FAIL_COND(!multimesh);
r_current_offset = multimesh->motion_vectors_current_offset;
if (RSG::rasterizer->get_frame_number() - multimesh->motion_vectors_last_change >= 2) {
if (!_multimesh_uses_motion_vectors(multimesh)) {
multimesh->motion_vectors_previous_offset = multimesh->motion_vectors_current_offset;
}
r_prev_offset = multimesh->motion_vectors_previous_offset;
}
bool MeshStorage::_multimesh_uses_motion_vectors_offsets(RID p_multimesh) {
MultiMesh *multimesh = multimesh_owner.get_or_null(p_multimesh);
ERR_FAIL_NULL_V(multimesh, false);
return _multimesh_uses_motion_vectors(multimesh);
}
int MeshStorage::multimesh_get_instance_count(RID p_multimesh) const {
MultiMesh *multimesh = multimesh_owner.get_or_null(p_multimesh);
ERR_FAIL_COND_V(!multimesh, 0);
@ -1500,6 +1506,10 @@ void MeshStorage::_multimesh_update_motion_vectors_data_cache(MultiMesh *multime
}
}
bool MeshStorage::_multimesh_uses_motion_vectors(MultiMesh *multimesh) {
return (RSG::rasterizer->get_frame_number() - multimesh->motion_vectors_last_change) < 2;
}
void MeshStorage::_multimesh_mark_dirty(MultiMesh *multimesh, int p_index, bool p_aabb) {
uint32_t region_index = p_index / MULTIMESH_DIRTY_REGION_SIZE;
#ifdef DEBUG_ENABLED

3
servers/rendering/renderer_rd/storage_rd/mesh_storage.h
View File

@ -244,6 +244,7 @@ private:
_FORCE_INLINE_ void _multimesh_make_local(MultiMesh *multimesh) const;
_FORCE_INLINE_ void _multimesh_enable_motion_vectors(MultiMesh *multimesh);
_FORCE_INLINE_ void _multimesh_update_motion_vectors_data_cache(MultiMesh *multimesh);
_FORCE_INLINE_ bool _multimesh_uses_motion_vectors(MultiMesh *multimesh);
_FORCE_INLINE_ void _multimesh_mark_dirty(MultiMesh *multimesh, int p_index, bool p_aabb);
_FORCE_INLINE_ void _multimesh_mark_all_dirty(MultiMesh *multimesh, bool p_data, bool p_aabb);
_FORCE_INLINE_ void _multimesh_re_create_aabb(MultiMesh *multimesh, const float *p_data, int p_instances);
@ -622,6 +623,8 @@ public:
void _update_dirty_multimeshes();
void _multimesh_get_motion_vectors_offsets(RID p_multimesh, uint32_t &r_current_offset, uint32_t &r_prev_offset);
bool _multimesh_uses_motion_vectors_offsets(RID p_multimesh);
bool _multimesh_uses_motion_vectors(RID p_multimesh);
_FORCE_INLINE_ RS::MultimeshTransformFormat multimesh_get_transform_format(RID p_multimesh) const {
MultiMesh *multimesh = multimesh_owner.get_or_null(p_multimesh);

60
servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.cpp
View File

@ -52,6 +52,7 @@ void RenderSceneBuffersRD::_bind_methods() {
ClassDB::bind_method(D_METHOD("get_texture", "context", "name"), &RenderSceneBuffersRD::get_texture);
ClassDB::bind_method(D_METHOD("get_texture_format", "context", "name"), &RenderSceneBuffersRD::_get_texture_format);
ClassDB::bind_method(D_METHOD("get_texture_slice", "context", "name", "layer", "mipmap", "layers", "mipmaps"), &RenderSceneBuffersRD::get_texture_slice);
ClassDB::bind_method(D_METHOD("get_texture_slice_view", "context", "name", "layer", "mipmap", "layers", "mipmaps", "view"), &RenderSceneBuffersRD::_get_texture_slice_view);
ClassDB::bind_method(D_METHOD("get_texture_slice_size", "context", "name", "mipmap"), &RenderSceneBuffersRD::get_texture_slice_size);
ClassDB::bind_method(D_METHOD("clear_context", "context"), &RenderSceneBuffersRD::clear_context);
@ -95,8 +96,8 @@ void RenderSceneBuffersRD::free_named_texture(NamedTexture &p_named_texture) {
void RenderSceneBuffersRD::update_samplers() {
float computed_mipmap_bias = texture_mipmap_bias;
if (use_taa) {
// Use negative mipmap LOD bias when TAA is enabled to compensate for loss of sharpness.
if (use_taa || (scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR2)) {
// Use negative mipmap LOD bias when TAA or FSR2 is enabled to compensate for loss of sharpness.
// This restores sharpness in still images to be roughly at the same level as without TAA,
// but moving scenes will still be blurrier.
computed_mipmap_bias -= 0.5;
@ -388,6 +389,15 @@ Ref<RDTextureFormat> RenderSceneBuffersRD::_get_texture_format(const StringName
return tf;
}
RID RenderSceneBuffersRD::_get_texture_slice_view(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers, const uint32_t p_mipmaps, const Ref<RDTextureView> p_view) {
RD::TextureView texture_view;
if (p_view.is_valid()) {
texture_view = p_view->base;
}
return get_texture_slice_view(p_context, p_texture_name, p_layer, p_mipmap, p_layers, p_mipmaps, texture_view);
}
const RD::TextureFormat RenderSceneBuffersRD::get_texture_format(const StringName &p_context, const StringName &p_texture_name) const {
NTKey key(p_context, p_texture_name);
@ -397,6 +407,10 @@ const RD::TextureFormat RenderSceneBuffersRD::get_texture_format(const StringNam
}
RID RenderSceneBuffersRD::get_texture_slice(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers, const uint32_t p_mipmaps) {
return get_texture_slice_view(p_context, p_texture_name, p_layer, p_mipmap, p_layers, p_mipmaps, RD::TextureView());
}
RID RenderSceneBuffersRD::get_texture_slice_view(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers, const uint32_t p_mipmaps, RD::TextureView p_view) {
NTKey key(p_context, p_texture_name);
// check if this is a known texture
@ -413,19 +427,20 @@ RID RenderSceneBuffersRD::get_texture_slice(const StringName &p_context, const S
ERR_FAIL_COND_V(p_mipmap + p_mipmaps > named_texture.format.mipmaps, RID());
// asking the whole thing? just return the original
if (p_layer == 0 && p_mipmap == 0 && named_texture.format.array_layers == p_layers && named_texture.format.mipmaps == p_mipmaps) {
RD::TextureView default_view = RD::TextureView();
if (p_layer == 0 && p_mipmap == 0 && named_texture.format.array_layers == p_layers && named_texture.format.mipmaps == p_mipmaps && p_view == default_view) {
return named_texture.texture;
}
// see if we have this
NTSliceKey slice_key(p_layer, p_layers, p_mipmap, p_mipmaps);
NTSliceKey slice_key(p_layer, p_layers, p_mipmap, p_mipmaps, p_view);
if (named_texture.slices.has(slice_key)) {
return named_texture.slices[slice_key];
}
// create our slice
RID &slice = named_texture.slices[slice_key];
slice = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), named_texture.texture, p_layer, p_mipmap, p_mipmaps, p_layers > 1 ? RD::TEXTURE_SLICE_2D_ARRAY : RD::TEXTURE_SLICE_2D, p_layers);
slice = RD::get_singleton()->texture_create_shared_from_slice(p_view, named_texture.texture, p_layer, p_mipmap, p_mipmaps, p_layers > 1 ? RD::TEXTURE_SLICE_2D_ARRAY : RD::TEXTURE_SLICE_2D, p_layers);
Array arr;
arr.push_back(p_context);
@ -434,7 +449,12 @@ RID RenderSceneBuffersRD::get_texture_slice(const StringName &p_context, const S
arr.push_back(itos(p_layers));
arr.push_back(itos(p_mipmap));
arr.push_back(itos(p_mipmaps));
RD::get_singleton()->set_resource_name(slice, String("RenderBuffer {0}/{1}, layer {2}/{3}, mipmap {4}/{5}").format(arr));
arr.push_back(itos(p_view.format_override));
arr.push_back(itos(p_view.swizzle_r));
arr.push_back(itos(p_view.swizzle_g));
arr.push_back(itos(p_view.swizzle_b));
arr.push_back(itos(p_view.swizzle_a));
RD::get_singleton()->set_resource_name(slice, String("RenderBuffer {0}/{1}, layer {2}/{3}, mipmap {4}/{5}, view {6}/{7}/{8}/{9}/{10}").format(arr));
// and return our slice
return slice;
@ -479,7 +499,13 @@ void RenderSceneBuffersRD::allocate_blur_textures() {
return;
}
uint32_t mipmaps_required = Image::get_image_required_mipmaps(internal_size.x, internal_size.y, Image::FORMAT_RGBAH);
Size2i blur_size = internal_size;
if (scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR2) {
// The blur texture should be as big as the target size when using an upscaler.
blur_size = target_size;
}
uint32_t mipmaps_required = Image::get_image_required_mipmaps(blur_size.x, blur_size.y, Image::FORMAT_RGBAH);
uint32_t usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_COPY_TO_BIT;
if (can_be_storage) {
@ -488,12 +514,12 @@ void RenderSceneBuffersRD::allocate_blur_textures() {
usage_bits += RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT;
}
create_texture(RB_SCOPE_BUFFERS, RB_TEX_BLUR_0, base_data_format, usage_bits, RD::TEXTURE_SAMPLES_1, internal_size, view_count, mipmaps_required);
create_texture(RB_SCOPE_BUFFERS, RB_TEX_BLUR_1, base_data_format, usage_bits, RD::TEXTURE_SAMPLES_1, Size2i(internal_size.x >> 1, internal_size.y >> 1), view_count, mipmaps_required - 1);
create_texture(RB_SCOPE_BUFFERS, RB_TEX_BLUR_0, base_data_format, usage_bits, RD::TEXTURE_SAMPLES_1, blur_size, view_count, mipmaps_required);
create_texture(RB_SCOPE_BUFFERS, RB_TEX_BLUR_1, base_data_format, usage_bits, RD::TEXTURE_SAMPLES_1, Size2i(blur_size.x >> 1, blur_size.y >> 1), view_count, mipmaps_required - 1);
// if !can_be_storage we need a half width version
if (!can_be_storage) {
create_texture(RB_SCOPE_BUFFERS, RB_TEX_HALF_BLUR, base_data_format, usage_bits, RD::TEXTURE_SAMPLES_1, Size2i(internal_size.x >> 1, internal_size.y), 1, mipmaps_required);
create_texture(RB_SCOPE_BUFFERS, RB_TEX_HALF_BLUR, base_data_format, usage_bits, RD::TEXTURE_SAMPLES_1, Size2i(blur_size.x >> 1, blur_size.y), 1, mipmaps_required);
}
// TODO redo this:
@ -502,8 +528,8 @@ void RenderSceneBuffersRD::allocate_blur_textures() {
RD::TextureFormat tf;
tf.format = RD::DATA_FORMAT_R16_SFLOAT; // We could probably use DATA_FORMAT_R8_SNORM if we don't pre-multiply by blur_size but that depends on whether we can remove DEPTH_GAP
tf.width = internal_size.x;
tf.height = internal_size.y;
tf.width = blur_size.x;
tf.height = blur_size.y;
tf.texture_type = RD::TEXTURE_TYPE_2D;
tf.array_layers = 1; // Our DOF effect handles one eye per turn
tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_COPY_TO_BIT | RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT;
@ -603,6 +629,16 @@ RID RenderSceneBuffersRD::get_depth_texture(const uint32_t p_layer) {
}
}
// Upscaled texture.
void RenderSceneBuffersRD::ensure_upscaled() {
if (!has_upscaled_texture()) {
uint32_t usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | (can_be_storage ? RD::TEXTURE_USAGE_STORAGE_BIT : 0) | RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT;
usage_bits |= RD::TEXTURE_USAGE_INPUT_ATTACHMENT_BIT;
create_texture(RB_SCOPE_BUFFERS, RB_TEX_COLOR_UPSCALED, base_data_format, usage_bits, RD::TEXTURE_SAMPLES_1, target_size);
}
}
// Velocity texture.
void RenderSceneBuffersRD::ensure_velocity() {

36
servers/rendering/renderer_rd/storage_rd/render_scene_buffers_rd.h
View File

@ -31,6 +31,7 @@
#ifndef RENDER_SCENE_BUFFERS_RD_H
#define RENDER_SCENE_BUFFERS_RD_H
#include "../effects/fsr2.h"
#include "../effects/vrs.h"
#include "../framebuffer_cache_rd.h"
#include "core/templates/hash_map.h"
@ -47,6 +48,7 @@
#define RB_TEXTURE SNAME("texture")
#define RB_TEX_COLOR SNAME("color")
#define RB_TEX_COLOR_MSAA SNAME("color_msaa")
#define RB_TEX_COLOR_UPSCALED SNAME("color_upscaled")
#define RB_TEX_DEPTH SNAME("depth")
#define RB_TEX_DEPTH_MSAA SNAME("depth_msaa")
#define RB_TEX_VELOCITY SNAME("velocity")
@ -114,9 +116,10 @@ private:
uint32_t layers;
uint32_t mipmap;
uint32_t mipmaps;
RD::TextureView texture_view;
bool operator==(const NTSliceKey &p_val) const {
return (layer == p_val.layer) && (layers == p_val.layers) && (mipmap == p_val.mipmap) && (mipmaps == p_val.mipmaps);
return (layer == p_val.layer) && (layers == p_val.layers) && (mipmap == p_val.mipmap) && (mipmaps == p_val.mipmaps) && (texture_view == p_val.texture_view);
}
static uint32_t hash(const NTSliceKey &p_val) {
@ -124,15 +127,21 @@ private:
h = hash_murmur3_one_32(p_val.layers, h);
h = hash_murmur3_one_32(p_val.mipmap, h);
h = hash_murmur3_one_32(p_val.mipmaps, h);
h = hash_murmur3_one_32(p_val.texture_view.format_override);
h = hash_murmur3_one_32(p_val.texture_view.swizzle_r, h);
h = hash_murmur3_one_32(p_val.texture_view.swizzle_g, h);
h = hash_murmur3_one_32(p_val.texture_view.swizzle_b, h);
h = hash_murmur3_one_32(p_val.texture_view.swizzle_a, h);
return hash_fmix32(h);
}
NTSliceKey() {}
NTSliceKey(uint32_t p_layer, uint32_t p_layers, uint32_t p_mipmap, uint32_t p_mipmaps) {
NTSliceKey(uint32_t p_layer, uint32_t p_layers, uint32_t p_mipmap, uint32_t p_mipmaps, RD::TextureView p_texture_view) {
layer = p_layer;
layers = p_layers;
mipmap = p_mipmap;
mipmaps = p_mipmaps;
texture_view = p_texture_view;
}
};
@ -190,6 +199,7 @@ public:
RID get_texture(const StringName &p_context, const StringName &p_texture_name) const;
const RD::TextureFormat get_texture_format(const StringName &p_context, const StringName &p_texture_name) const;
RID get_texture_slice(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers = 1, const uint32_t p_mipmaps = 1);
RID get_texture_slice_view(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers = 1, const uint32_t p_mipmaps = 1, RD::TextureView p_view = RD::TextureView());
Size2i get_texture_slice_size(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_mipmap);
void clear_context(const StringName &p_context);
@ -230,6 +240,14 @@ public:
_FORCE_INLINE_ RID get_internal_texture(const uint32_t p_layer) {
return get_texture_slice(RB_SCOPE_BUFFERS, RB_TEX_COLOR, p_layer, 0);
}
_FORCE_INLINE_ RID get_internal_texture_reactive(const uint32_t p_layer) {
RD::TextureView alpha_only_view;
alpha_only_view.swizzle_r = RD::TEXTURE_SWIZZLE_A;
alpha_only_view.swizzle_g = RD::TEXTURE_SWIZZLE_A;
alpha_only_view.swizzle_b = RD::TEXTURE_SWIZZLE_A;
alpha_only_view.swizzle_a = RD::TEXTURE_SWIZZLE_A;
return get_texture_slice_view(RB_SCOPE_BUFFERS, RB_TEX_COLOR, p_layer, 0, 1, 1, alpha_only_view);
}
_FORCE_INLINE_ RID get_color_msaa() const {
return get_texture(RB_SCOPE_BUFFERS, RB_TEX_COLOR_MSAA);
}
@ -251,6 +269,19 @@ public:
// back buffer (color)
RID get_back_buffer_texture() const { return has_texture(RB_SCOPE_BUFFERS, RB_TEX_BLUR_0) ? get_texture(RB_SCOPE_BUFFERS, RB_TEX_BLUR_0) : RID(); } // We (re)use our blur texture here.
// Upscaled.
void ensure_upscaled();
_FORCE_INLINE_ bool has_upscaled_texture() const {
return has_texture(RB_SCOPE_BUFFERS, RB_TEX_COLOR_UPSCALED);
}
_FORCE_INLINE_ RID get_upscaled_texture() const {
return get_texture(RB_SCOPE_BUFFERS, RB_TEX_COLOR_UPSCALED);
}
_FORCE_INLINE_ RID get_upscaled_texture(const uint32_t p_layer) {
return get_texture_slice(RB_SCOPE_BUFFERS, RB_TEX_COLOR_UPSCALED, p_layer, 0);
}
// Velocity, currently only used by TAA (Clustered) but we'll be using this in other places soon too.
void ensure_velocity();
@ -271,6 +302,7 @@ private:
RID _create_texture_from_format(const StringName &p_context, const StringName &p_texture_name, const Ref<RDTextureFormat> &p_texture_format, const Ref<RDTextureView> &p_view = Ref<RDTextureView>(), bool p_unique = true);
RID _create_texture_view(const StringName &p_context, const StringName &p_texture_name, const StringName p_view_name, const Ref<RDTextureView> p_view = Ref<RDTextureView>());
Ref<RDTextureFormat> _get_texture_format(const StringName &p_context, const StringName &p_texture_name) const;
RID _get_texture_slice_view(const StringName &p_context, const StringName &p_texture_name, const uint32_t p_layer, const uint32_t p_mipmap, const uint32_t p_layers = 1, const uint32_t p_mipmaps = 1, const Ref<RDTextureView> p_view = Ref<RDTextureView>());
// For color and depth as exposed to extensions, we return the buffer that we're rendering into.
// Resolving happens after effects etc. are run.

3
servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.cpp
View File

@ -38,7 +38,7 @@ RID RenderSceneDataRD::create_uniform_buffer() {
return RD::get_singleton()->uniform_buffer_create(sizeof(UBODATA));
}
void RenderSceneDataRD::update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p_debug_mode, RID p_env, RID p_reflection_probe_instance, RID p_camera_attributes, bool p_flip_y, bool p_pancake_shadows, const Size2i &p_screen_size, const Color &p_default_bg_color, float p_luminance_multiplier, bool p_opaque_render_buffers) {
void RenderSceneDataRD::update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p_debug_mode, RID p_env, RID p_reflection_probe_instance, RID p_camera_attributes, bool p_flip_y, bool p_pancake_shadows, const Size2i &p_screen_size, const Color &p_default_bg_color, float p_luminance_multiplier, bool p_opaque_render_buffers, bool p_apply_alpha_multiplier) {
RendererSceneRenderRD *render_scene_render = RendererSceneRenderRD::get_singleton();
UBODATA ubo_data;
@ -89,6 +89,7 @@ void RenderSceneDataRD::update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p
RendererRD::MaterialStorage::store_soft_shadow_kernel(render_scene_render->penumbra_shadow_kernel_get(), ubo.penumbra_shadow_kernel);
RendererRD::MaterialStorage::store_soft_shadow_kernel(render_scene_render->soft_shadow_kernel_get(), ubo.soft_shadow_kernel);
ubo.camera_visible_layers = camera_visible_layers;
ubo.pass_alpha_multiplier = p_opaque_render_buffers && p_apply_alpha_multiplier ? 0.0f : 1.0f;
ubo.viewport_size[0] = p_screen_size.x;
ubo.viewport_size[1] = p_screen_size.y;

4
servers/rendering/renderer_rd/storage_rd/render_scene_data_rd.h
View File

@ -77,7 +77,7 @@ public:
float time_step;
RID create_uniform_buffer();
void update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p_debug_mode, RID p_env, RID p_reflection_probe_instance, RID p_camera_attributes, bool p_flip_y, bool p_pancake_shadows, const Size2i &p_screen_size, const Color &p_default_bg_color, float p_luminance_multiplier, bool p_opaque_render_buffers);
void update_ubo(RID p_uniform_buffer, RS::ViewportDebugDraw p_debug_mode, RID p_env, RID p_reflection_probe_instance, RID p_camera_attributes, bool p_flip_y, bool p_pancake_shadows, const Size2i &p_screen_size, const Color &p_default_bg_color, float p_luminance_multiplier, bool p_opaque_render_buffers, bool p_apply_alpha_multiplier);
RID get_uniform_buffer();
private:
@ -144,7 +144,7 @@ private:
uint32_t pancake_shadows;
uint32_t camera_visible_layers;
uint32_t pad2;
float pass_alpha_multiplier;
uint32_t pad3;
};

49
servers/rendering/renderer_scene_cull.cpp
View File

@ -37,6 +37,21 @@
#include <new>
/* HALTON SEQUENCE */
#ifndef _3D_DISABLED
static float get_halton_value(int p_index, int p_base) {
float f = 1;
float r = 0;
while (p_index > 0) {
f = f / static_cast<float>(p_base);
r = r + f * (p_index % p_base);
p_index = p_index / p_base;
}
return r * 2.0f - 1.0f;
}
#endif // _3D_DISABLED
/* CAMERA API */
RID RendererSceneCull::camera_allocate() {
@ -2498,15 +2513,26 @@ bool RendererSceneCull::_light_instance_update_shadow(Instance *p_instance, cons
return animated_material_found;
}
void RendererSceneCull::render_camera(const Ref<RenderSceneBuffers> &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, bool p_use_taa, float p_screen_mesh_lod_threshold, RID p_shadow_atlas, Ref<XRInterface> &p_xr_interface, RenderInfo *r_render_info) {
void RendererSceneCull::render_camera(const Ref<RenderSceneBuffers> &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, uint32_t p_jitter_phase_count, float p_screen_mesh_lod_threshold, RID p_shadow_atlas, Ref<XRInterface> &p_xr_interface, RenderInfo *r_render_info) {
#ifndef _3D_DISABLED
Camera *camera = camera_owner.get_or_null(p_camera);
ERR_FAIL_COND(!camera);
Vector2 jitter;
if (p_use_taa) {
jitter = taa_jitter_array[RSG::rasterizer->get_frame_number() % TAA_JITTER_COUNT] / p_viewport_size;
if (p_jitter_phase_count > 0) {
uint32_t current_jitter_count = camera_jitter_array.size();
if (p_jitter_phase_count != current_jitter_count) {
// Resize the jitter array and fill it with the pre-computed Halton sequence.
camera_jitter_array.resize(p_jitter_phase_count);
for (uint32_t i = current_jitter_count; i < p_jitter_phase_count; i++) {
camera_jitter_array[i].x = get_halton_value(i, 2);
camera_jitter_array[i].y = get_halton_value(i, 3);
}
}
jitter = camera_jitter_array[RSG::rasterizer->get_frame_number() % p_jitter_phase_count] / p_viewport_size;
}
RendererSceneRender::CameraData camera_data;
@ -4113,17 +4139,6 @@ void RendererSceneCull::set_scene_render(RendererSceneRender *p_scene_render) {
geometry_instance_pair_mask = scene_render->geometry_instance_get_pair_mask();
}
float get_halton_value(int index, int base) {
float f = 1;
float r = 0;
while (index > 0) {
f = f / static_cast<float>(base);
r = r + f * (index % base);
index = index / base;
}
return r * 2.0f - 1.0f;
};
RendererSceneCull::RendererSceneCull() {
render_pass = 1;
singleton = this;
@ -4148,12 +4163,6 @@ RendererSceneCull::RendererSceneCull() {
thread_cull_threshold = GLOBAL_GET("rendering/limits/spatial_indexer/threaded_cull_minimum_instances");
thread_cull_threshold = MAX(thread_cull_threshold, (uint32_t)WorkerThreadPool::get_singleton()->get_thread_count()); //make sure there is at least one thread per CPU
taa_jitter_array.resize(TAA_JITTER_COUNT);
for (int i = 0; i < TAA_JITTER_COUNT; i++) {
taa_jitter_array[i].x = get_halton_value(i, 2);
taa_jitter_array[i].y = get_halton_value(i, 3);
}
dummy_occlusion_culling = memnew(RendererSceneOcclusionCull);
}

5
servers/rendering/renderer_scene_cull.h
View File

@ -954,8 +954,7 @@ public:
uint32_t geometry_instance_pair_mask = 0; // used in traditional forward, unnecessary on clustered
const int TAA_JITTER_COUNT = 16;
LocalVector<Vector2> taa_jitter_array;
LocalVector<Vector2> camera_jitter_array;
virtual RID instance_allocate();
virtual void instance_initialize(RID p_rid);
@ -1089,7 +1088,7 @@ public:
void _render_scene(const RendererSceneRender::CameraData *p_camera_data, const Ref<RenderSceneBuffers> &p_render_buffers, RID p_environment, RID p_force_camera_attributes, uint32_t p_visible_layers, RID p_scenario, RID p_viewport, RID p_shadow_atlas, RID p_reflection_probe, int p_reflection_probe_pass, float p_screen_mesh_lod_threshold, bool p_using_shadows = true, RenderInfo *r_render_info = nullptr);
void render_empty_scene(const Ref<RenderSceneBuffers> &p_render_buffers, RID p_scenario, RID p_shadow_atlas);
void render_camera(const Ref<RenderSceneBuffers> &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, bool p_use_taa, float p_screen_mesh_lod_threshold, RID p_shadow_atlas, Ref<XRInterface> &p_xr_interface, RenderingMethod::RenderInfo *r_render_info = nullptr);
void render_camera(const Ref<RenderSceneBuffers> &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, uint32_t p_jitter_phase_count, float p_screen_mesh_lod_threshold, RID p_shadow_atlas, Ref<XRInterface> &p_xr_interface, RenderingMethod::RenderInfo *r_render_info = nullptr);
void update_dirty_instances();
void render_particle_colliders();

59
servers/rendering/renderer_viewport.cpp
View File

@ -118,22 +118,29 @@ void RendererViewport::_configure_3d_render_buffers(Viewport *p_viewport) {
} else {
float scaling_3d_scale = p_viewport->scaling_3d_scale;
RS::ViewportScaling3DMode scaling_3d_mode = p_viewport->scaling_3d_mode;
bool scaling_3d_is_fsr = (scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR) || (scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR2);
bool use_taa = p_viewport->use_taa;
if ((scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR) && (scaling_3d_scale > 1.0)) {
if (scaling_3d_is_fsr && (scaling_3d_scale > 1.0)) {
// FSR is not designed for downsampling.
// Fall back to bilinear scaling.
WARN_PRINT_ONCE("FSR 3D resolution scaling is not designed for downsampling. Falling back to bilinear 3D resolution scaling.");
scaling_3d_mode = RS::VIEWPORT_SCALING_3D_MODE_BILINEAR;
}
if ((scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR) && !p_viewport->fsr_enabled) {
bool upscaler_available = p_viewport->fsr_enabled;
if (scaling_3d_is_fsr && !upscaler_available) {
// FSR is not actually available.
// Fall back to bilinear scaling.
WARN_PRINT_ONCE("FSR 1.0 3D resolution scaling is not available. Falling back to bilinear 3D resolution scaling.");
WARN_PRINT_ONCE("FSR 3D resolution scaling is not available. Falling back to bilinear 3D resolution scaling.");
scaling_3d_mode = RS::VIEWPORT_SCALING_3D_MODE_BILINEAR;
}
if (scaling_3d_scale == 1.0) {
scaling_3d_mode = RS::VIEWPORT_SCALING_3D_MODE_OFF;
if (use_taa && scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR2) {
// FSR2 can't be used with TAA.
// Turn it off and prefer using FSR2.
WARN_PRINT_ONCE("FSR 2 is not compatible with TAA. Disabling TAA internally.");
use_taa = false;
}
int width;
@ -151,6 +158,7 @@ void RendererViewport::_configure_3d_render_buffers(Viewport *p_viewport) {
render_height = height;
break;
case RS::VIEWPORT_SCALING_3D_MODE_FSR:
case RS::VIEWPORT_SCALING_3D_MODE_FSR2:
width = p_viewport->size.width;
height = p_viewport->size.height;
render_width = MAX(width * scaling_3d_scale, 1.0); // width / (width * scaling)
@ -174,7 +182,17 @@ void RendererViewport::_configure_3d_render_buffers(Viewport *p_viewport) {
break;
}
uint32_t jitter_phase_count = 0;
if (scaling_3d_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR2) {
// Implementation has been copied from ffxFsr2GetJitterPhaseCount.
jitter_phase_count = uint32_t(8.0f * pow(float(width) / render_width, 2.0f));
} else if (use_taa) {
// Default jitter count for TAA.
jitter_phase_count = 16;
}
p_viewport->internal_size = Size2(render_width, render_height);
p_viewport->jitter_phase_count = jitter_phase_count;
// At resolution scales lower than 1.0, use negative texture mipmap bias
// to compensate for the loss of sharpness.
@ -190,7 +208,7 @@ void RendererViewport::_configure_3d_render_buffers(Viewport *p_viewport) {
rb_config.set_screen_space_aa(p_viewport->screen_space_aa);
rb_config.set_fsr_sharpness(p_viewport->fsr_sharpness);
rb_config.set_texture_mipmap_bias(texture_mipmap_bias);
rb_config.set_use_taa(p_viewport->use_taa);
rb_config.set_use_taa(use_taa);
p_viewport->render_buffers->configure(&rb_config);
}
@ -221,7 +239,7 @@ void RendererViewport::_draw_3d(Viewport *p_viewport) {
}
float screen_mesh_lod_threshold = p_viewport->mesh_lod_threshold / float(p_viewport->size.width);
RSG::scene->render_camera(p_viewport->render_buffers, p_viewport->camera, p_viewport->scenario, p_viewport->self, p_viewport->internal_size, p_viewport->use_taa, screen_mesh_lod_threshold, p_viewport->shadow_atlas, xr_interface, &p_viewport->render_info);
RSG::scene->render_camera(p_viewport->render_buffers, p_viewport->camera, p_viewport->scenario, p_viewport->self, p_viewport->internal_size, p_viewport->jitter_phase_count, screen_mesh_lod_threshold, p_viewport->shadow_atlas, xr_interface, &p_viewport->render_info);
RENDER_TIMESTAMP("< Render 3D Scene");
}
@ -825,8 +843,20 @@ void RendererViewport::viewport_set_use_xr(RID p_viewport, bool p_use_xr) {
void RendererViewport::viewport_set_scaling_3d_mode(RID p_viewport, RS::ViewportScaling3DMode p_mode) {
Viewport *viewport = viewport_owner.get_or_null(p_viewport);
ERR_FAIL_COND(!viewport);
ERR_FAIL_COND_EDMSG(p_mode == RS::VIEWPORT_SCALING_3D_MODE_FSR2 && OS::get_singleton()->get_current_rendering_method() != "forward_plus", "FSR2 is only available when using the Forward+ renderer.");
if (viewport->scaling_3d_mode == p_mode) {
return;
}
bool motion_vectors_before = _viewport_requires_motion_vectors(viewport);
viewport->scaling_3d_mode = p_mode;
bool motion_vectors_after = _viewport_requires_motion_vectors(viewport);
if (motion_vectors_before != motion_vectors_after) {
num_viewports_with_motion_vectors += motion_vectors_after ? 1 : -1;
}
_configure_3d_render_buffers(viewport);
}
@ -888,6 +918,10 @@ void RendererViewport::_viewport_set_size(Viewport *p_viewport, int p_width, int
}
}
bool RendererViewport::_viewport_requires_motion_vectors(Viewport *p_viewport) {
return p_viewport->use_taa || p_viewport->scaling_3d_mode == RenderingServer::VIEWPORT_SCALING_3D_MODE_FSR2;
}
void RendererViewport::viewport_set_active(RID p_viewport, bool p_active) {
Viewport *viewport = viewport_owner.get_or_null(p_viewport);
ERR_FAIL_COND(!viewport);
@ -1193,8 +1227,15 @@ void RendererViewport::viewport_set_use_taa(RID p_viewport, bool p_use_taa) {
if (viewport->use_taa == p_use_taa) {
return;
}
bool motion_vectors_before = _viewport_requires_motion_vectors(viewport);
viewport->use_taa = p_use_taa;
num_viewports_with_motion_vectors += p_use_taa ? 1 : -1;
bool motion_vectors_after = _viewport_requires_motion_vectors(viewport);
if (motion_vectors_before != motion_vectors_after) {
num_viewports_with_motion_vectors += motion_vectors_after ? 1 : -1;
}
_configure_3d_render_buffers(viewport);
}
@ -1379,7 +1420,7 @@ bool RendererViewport::free(RID p_rid) {
RendererSceneOcclusionCull::get_singleton()->remove_buffer(p_rid);
}
if (viewport->use_taa) {
if (_viewport_requires_motion_vectors(viewport)) {
num_viewports_with_motion_vectors--;
}

2
servers/rendering/renderer_viewport.h
View File

@ -63,6 +63,7 @@ public:
float fsr_sharpness = 0.2f;
float texture_mipmap_bias = 0.0f;
bool fsr_enabled = false;
uint32_t jitter_phase_count = 0;
RS::ViewportUpdateMode update_mode = RenderingServer::VIEWPORT_UPDATE_WHEN_VISIBLE;
RID render_target;
RID render_target_texture;
@ -203,6 +204,7 @@ public:
private:
Vector<Viewport *> _sort_active_viewports();
void _viewport_set_size(Viewport *p_viewport, int p_width, int p_height, uint32_t p_view_count);
bool _viewport_requires_motion_vectors(Viewport *p_viewport);
void _configure_3d_render_buffers(Viewport *p_viewport);
void _draw_3d(Viewport *p_viewport);
void _draw_viewport(Viewport *p_viewport);

18
servers/rendering/rendering_device.h
View File

@ -518,6 +518,22 @@ public:
TextureSwizzle swizzle_b;
TextureSwizzle swizzle_a;
bool operator==(const TextureView &p_view) const {
if (format_override != p_view.format_override) {
return false;
} else if (swizzle_r != p_view.swizzle_r) {
return false;
} else if (swizzle_g != p_view.swizzle_g) {
return false;
} else if (swizzle_b != p_view.swizzle_b) {
return false;
} else if (swizzle_a != p_view.swizzle_a) {
return false;
} else {
return true;
}
}
TextureView() {
format_override = DATA_FORMAT_MAX; //means, use same as format
swizzle_r = TEXTURE_SWIZZLE_R;
@ -1270,6 +1286,8 @@ public:
LIMIT_MAX_VIEWPORT_DIMENSIONS_X,
LIMIT_MAX_VIEWPORT_DIMENSIONS_Y,
LIMIT_SUBGROUP_SIZE,
LIMIT_SUBGROUP_MIN_SIZE,
LIMIT_SUBGROUP_MAX_SIZE,
LIMIT_SUBGROUP_IN_SHADERS, // Set flags using SHADER_STAGE_VERTEX_BIT, SHADER_STAGE_FRAGMENT_BIT, etc.
LIMIT_SUBGROUP_OPERATIONS,
LIMIT_VRS_TEXEL_WIDTH,

2
servers/rendering/rendering_method.h
View File

@ -301,7 +301,7 @@ public:
int info[RS::VIEWPORT_RENDER_INFO_TYPE_MAX][RS::VIEWPORT_RENDER_INFO_MAX] = {};
};
virtual void render_camera(const Ref<RenderSceneBuffers> &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, bool p_use_taa, float p_mesh_lod_threshold, RID p_shadow_atlas, Ref<XRInterface> &p_xr_interface, RenderInfo *r_render_info = nullptr) = 0;
virtual void render_camera(const Ref<RenderSceneBuffers> &p_render_buffers, RID p_camera, RID p_scenario, RID p_viewport, Size2 p_viewport_size, uint32_t p_jitter_phase_count, float p_mesh_lod_threshold, RID p_shadow_atlas, Ref<XRInterface> &p_xr_interface, RenderInfo *r_render_info = nullptr) = 0;
virtual void update() = 0;
virtual void render_probes() = 0;

2
servers/rendering/storage/render_scene_buffers.cpp
View File

@ -49,7 +49,7 @@ void RenderSceneBuffersConfiguration::_bind_methods() {
ClassDB::bind_method(D_METHOD("get_scaling_3d_mode"), &RenderSceneBuffersConfiguration::get_scaling_3d_mode);
ClassDB::bind_method(D_METHOD("set_scaling_3d_mode", "scaling_3d_mode"), &RenderSceneBuffersConfiguration::set_scaling_3d_mode);
ADD_PROPERTY(PropertyInfo(Variant::INT, "scaling_3d_mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast)"), "set_scaling_3d_mode", "get_scaling_3d_mode"); // TODO VIEWPORT_SCALING_3D_MODE_OFF is possible here too, but we can't specify an enum string for it.
ADD_PROPERTY(PropertyInfo(Variant::INT, "scaling_3d_mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast),FSR 2.2 (Slow)"), "set_scaling_3d_mode", "get_scaling_3d_mode"); // TODO VIEWPORT_SCALING_3D_MODE_OFF is possible here too, but we can't specify an enum string for it.
ClassDB::bind_method(D_METHOD("get_msaa_3d"), &RenderSceneBuffersConfiguration::get_msaa_3d);
ClassDB::bind_method(D_METHOD("set_msaa_3d", "msaa_3d"), &RenderSceneBuffersConfiguration::set_msaa_3d);

4
servers/rendering_server.cpp
View File

@ -2252,6 +2252,7 @@ void RenderingServer::_bind_methods() {
BIND_ENUM_CONSTANT(VIEWPORT_SCALING_3D_MODE_BILINEAR);
BIND_ENUM_CONSTANT(VIEWPORT_SCALING_3D_MODE_FSR);
BIND_ENUM_CONSTANT(VIEWPORT_SCALING_3D_MODE_FSR2);
BIND_ENUM_CONSTANT(VIEWPORT_SCALING_3D_MODE_MAX);
BIND_ENUM_CONSTANT(VIEWPORT_UPDATE_DISABLED);
@ -2329,6 +2330,7 @@ void RenderingServer::_bind_methods() {
BIND_ENUM_CONSTANT(VIEWPORT_DEBUG_DRAW_CLUSTER_REFLECTION_PROBES);
BIND_ENUM_CONSTANT(VIEWPORT_DEBUG_DRAW_OCCLUDERS);
BIND_ENUM_CONSTANT(VIEWPORT_DEBUG_DRAW_MOTION_VECTORS);
BIND_ENUM_CONSTANT(VIEWPORT_DEBUG_DRAW_INTERNAL_BUFFER);
BIND_ENUM_CONSTANT(VIEWPORT_VRS_DISABLED);
BIND_ENUM_CONSTANT(VIEWPORT_VRS_TEXTURE);
@ -2959,7 +2961,7 @@ void RenderingServer::init() {
GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/anti_aliasing/screen_space_roughness_limiter/amount", PROPERTY_HINT_RANGE, "0.01,4.0,0.01"), 0.25);
GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/anti_aliasing/screen_space_roughness_limiter/limit", PROPERTY_HINT_RANGE, "0.01,1.0,0.01"), 0.18);
GLOBAL_DEF(PropertyInfo(Variant::INT, "rendering/scaling_3d/mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast)"), 0);
GLOBAL_DEF(PropertyInfo(Variant::INT, "rendering/scaling_3d/mode", PROPERTY_HINT_ENUM, "Bilinear (Fastest),FSR 1.0 (Fast),FSR 2.2 (Slow)"), 0);
GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/scaling_3d/scale", PROPERTY_HINT_RANGE, "0.25,2.0,0.01"), 1.0);
GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/scaling_3d/fsr_sharpness", PROPERTY_HINT_RANGE, "0,2,0.1"), 0.2f);
GLOBAL_DEF(PropertyInfo(Variant::FLOAT, "rendering/textures/default_filters/texture_mipmap_bias", PROPERTY_HINT_RANGE, "-2,2,0.001"), 0.0f);

2
servers/rendering_server.h
View File

@ -807,6 +807,7 @@ public:
enum ViewportScaling3DMode {
VIEWPORT_SCALING_3D_MODE_BILINEAR,
VIEWPORT_SCALING_3D_MODE_FSR,
VIEWPORT_SCALING_3D_MODE_FSR2,
VIEWPORT_SCALING_3D_MODE_MAX,
VIEWPORT_SCALING_3D_MODE_OFF = 255, // for internal use only
};
@ -971,6 +972,7 @@ public:
VIEWPORT_DEBUG_DRAW_CLUSTER_REFLECTION_PROBES,
VIEWPORT_DEBUG_DRAW_OCCLUDERS,
VIEWPORT_DEBUG_DRAW_MOTION_VECTORS,
VIEWPORT_DEBUG_DRAW_INTERNAL_BUFFER,
};
virtual void viewport_set_debug_draw(RID p_viewport, ViewportDebugDraw p_draw) = 0;

15
thirdparty/README.md vendored
View File

@ -17,6 +17,21 @@ Files extracted from upstream source:
- `license.txt`
## amd-fsr2
- Upstream: https://github.com/GPUOpen-Effects/FidelityFX-FSR2
- Version: 2.2.1 (1680d1edd5c034f88ebbbb793d8b88f8842cf804, 2023)
- License: MIT
Files extracted from upstream source:
- `ffx_*.cpp` and `ffx_*.h` from `src/ffx-fsr2-api`
- `shaders` folder from `src/ffx-fsr2-api` with `ffx_*.hlsl` files excluded
- `LICENSE.txt`
Apply `patches` to add the new options required by Godot and general compilation fixes.
## angle
- Upstream: https://chromium.googlesource.com/angle/angle/

21
thirdparty/amd-fsr2/LICENSE.txt vendored Normal file
View File

@ -0,0 +1,21 @@
FidelityFX Super Resolution 2.2
=================================
Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.

81
thirdparty/amd-fsr2/ffx_assert.cpp vendored Normal file
View File

@ -0,0 +1,81 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#include "ffx_assert.h"
#include <stdlib.h> // for malloc()
#ifdef _WIN32
#ifndef WIN32_LEAN_AND_MEAN
#define WIN32_LEAN_AND_MEAN
#endif
#include <windows.h> // required for OutputDebugString()
#include <stdio.h> // required for sprintf_s
#endif // #ifndef _WIN32
static FfxAssertCallback s_assertCallback;
// set the printing callback function
void ffxAssertSetPrintingCallback(FfxAssertCallback callback)
{
s_assertCallback = callback;
return;
}
// implementation of assert reporting
bool ffxAssertReport(const char* file, int32_t line, const char* condition, const char* message)
{
if (!file) {
return true;
}
#ifdef _WIN32
// form the final assertion string and output to the TTY.
const size_t bufferSize = static_cast<size_t>(snprintf(nullptr, 0, "%s(%d): ASSERTION FAILED. %s\n", file, line, message ? message : condition)) + 1;
char* tempBuf = static_cast<char*>(malloc(bufferSize));
if (!tempBuf) {
return true;
}
if (!message) {
sprintf_s(tempBuf, bufferSize, "%s(%d): ASSERTION FAILED. %s\n", file, line, condition);
} else {
sprintf_s(tempBuf, bufferSize, "%s(%d): ASSERTION FAILED. %s\n", file, line, message);
}
if (!s_assertCallback) {
OutputDebugStringA(tempBuf);
} else {
s_assertCallback(tempBuf);
}
// free the buffer.
free(tempBuf);
#else
FFX_UNUSED(line);
FFX_UNUSED(condition);
FFX_UNUSED(message);
#endif
return true;
}

132
thirdparty/amd-fsr2/ffx_assert.h vendored Normal file
View File

@ -0,0 +1,132 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#pragma once
#include "ffx_types.h"
#include "ffx_util.h"
#ifdef __cplusplus
extern "C" {
#endif // #ifdef __cplusplus
#ifdef _DEBUG
#ifdef _WIN32
#ifdef DISABLE_FFX_DEBUG_BREAK
#define FFX_DEBUG_BREAK \
{ \
}
#else
/// Macro to force the debugger to break at this point in the code.
#define FFX_DEBUG_BREAK __debugbreak();
#endif
#else
#define FFX_DEBUG_BREAK \
{ \
}
#endif
#else
// don't allow debug break in release builds.
#define FFX_DEBUG_BREAK
#endif
/// A typedef for the callback function for assert printing.
///
/// This can be used to re-route printing of assert messages from the FFX backend
/// to another destination. For example instead of the default behaviour of printing
/// the assert messages to the debugger's TTY the message can be re-routed to a
/// MessageBox in a GUI application.
///
/// @param [in] message The message generated by the assert.
///
typedef void (*FfxAssertCallback)(const char* message);
/// Function to report an assert.
///
/// @param [in] file The name of the file as a string.
/// @param [in] line The index of the line in the file.
/// @param [in] condition The boolean condition that was tested.
/// @param [in] msg The optional message to print.
///
/// @returns
/// Always returns true.
///
FFX_API bool ffxAssertReport(const char* file, int32_t line, const char* condition, const char* msg);
/// Provides the ability to set a callback for assert messages.
///
/// @param [in] callback The callback function that will receive assert messages.
///
FFX_API void ffxAssertSetPrintingCallback(FfxAssertCallback callback);
#ifdef _DEBUG
/// Standard assert macro.
#define FFX_ASSERT(condition) \
do \
{ \
if (!(condition) && ffxAssertReport(__FILE__, __LINE__, #condition, NULL)) \
FFX_DEBUG_BREAK \
} while (0)
/// Assert macro with message.
#define FFX_ASSERT_MESSAGE(condition, msg) \
do \
{ \
if (!(condition) && ffxAssertReport(__FILE__, __LINE__, #condition, msg)) \
FFX_DEBUG_BREAK \
} while (0)
/// Assert macro that always fails.
#define FFX_ASSERT_FAIL(message) \
do \
{ \
ffxAssertReport(__FILE__, __LINE__, NULL, message); \
FFX_DEBUG_BREAK \
} while (0)
#else
// asserts disabled
#define FFX_ASSERT(condition) \
do \
{ \
FFX_UNUSED(condition); \
} while (0)
#define FFX_ASSERT_MESSAGE(condition, message) \
do \
{ \
FFX_UNUSED(condition); \
FFX_UNUSED(message); \
} while (0)
#define FFX_ASSERT_FAIL(message) \
do \
{ \
FFX_UNUSED(message); \
} while (0)
#endif // #if _DEBUG
/// Simple static assert.
#define FFX_STATIC_ASSERT(condition) static_assert(condition, #condition)
#ifdef __cplusplus
}
#endif // #ifdef __cplusplus

59
thirdparty/amd-fsr2/ffx_error.h vendored Normal file
View File

@ -0,0 +1,59 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#pragma once
#include "ffx_types.h"
/// Typedef for error codes returned from functions in the FidelityFX SDK.
typedef int32_t FfxErrorCode;
static const FfxErrorCode FFX_OK = 0; ///< The operation completed successfully.
static const FfxErrorCode FFX_ERROR_INVALID_POINTER = 0x80000000; ///< The operation failed due to an invalid pointer.
static const FfxErrorCode FFX_ERROR_INVALID_ALIGNMENT = 0x80000001; ///< The operation failed due to an invalid alignment.
static const FfxErrorCode FFX_ERROR_INVALID_SIZE = 0x80000002; ///< The operation failed due to an invalid size.
static const FfxErrorCode FFX_EOF = 0x80000003; ///< The end of the file was encountered.
static const FfxErrorCode FFX_ERROR_INVALID_PATH = 0x80000004; ///< The operation failed because the specified path was invalid.
static const FfxErrorCode FFX_ERROR_EOF = 0x80000005; ///< The operation failed because end of file was reached.
static const FfxErrorCode FFX_ERROR_MALFORMED_DATA = 0x80000006; ///< The operation failed because of some malformed data.
static const FfxErrorCode FFX_ERROR_OUT_OF_MEMORY = 0x80000007; ///< The operation failed because it ran out memory.
static const FfxErrorCode FFX_ERROR_INCOMPLETE_INTERFACE = 0x80000008; ///< The operation failed because the interface was not fully configured.
static const FfxErrorCode FFX_ERROR_INVALID_ENUM = 0x80000009; ///< The operation failed because of an invalid enumeration value.
static const FfxErrorCode FFX_ERROR_INVALID_ARGUMENT = 0x8000000a; ///< The operation failed because an argument was invalid.
static const FfxErrorCode FFX_ERROR_OUT_OF_RANGE = 0x8000000b; ///< The operation failed because a value was out of range.
static const FfxErrorCode FFX_ERROR_NULL_DEVICE = 0x8000000c; ///< The operation failed because a device was null.
static const FfxErrorCode FFX_ERROR_BACKEND_API_ERROR = 0x8000000d; ///< The operation failed because the backend API returned an error code.
static const FfxErrorCode FFX_ERROR_INSUFFICIENT_MEMORY = 0x8000000e; ///< The operation failed because there was not enough memory.
/// Helper macro to return error code y from a function when a specific condition, x, is not met.
#define FFX_RETURN_ON_ERROR(x, y) \
if (!(x)) \
{ \
return (y); \
}
/// Helper macro to return error code x from a function when it is not FFX_OK.
#define FFX_VALIDATE(x) \
{ \
FfxErrorCode ret = x; \
FFX_RETURN_ON_ERROR(ret == FFX_OK, ret); \
}

1373
thirdparty/amd-fsr2/ffx_fsr2.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

458
thirdparty/amd-fsr2/ffx_fsr2.h vendored Normal file
View File

@ -0,0 +1,458 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
// @defgroup FSR2
#pragma once
// Include the interface for the backend of the FSR2 API.
#include "ffx_fsr2_interface.h"
/// FidelityFX Super Resolution 2 major version.
///
/// @ingroup FSR2
#define FFX_FSR2_VERSION_MAJOR (2)
/// FidelityFX Super Resolution 2 minor version.
///
/// @ingroup FSR2
#define FFX_FSR2_VERSION_MINOR (2)
/// FidelityFX Super Resolution 2 patch version.
///
/// @ingroup FSR2
#define FFX_FSR2_VERSION_PATCH (1)
/// The size of the context specified in 32bit values.
///
/// @ingroup FSR2
#define FFX_FSR2_CONTEXT_SIZE (16536)
#if defined(__cplusplus)
extern "C" {
#endif // #if defined(__cplusplus)
/// An enumeration of all the quality modes supported by FidelityFX Super
/// Resolution 2 upscaling.
///
/// In order to provide a consistent user experience across multiple
/// applications which implement FSR2. It is strongly recommended that the
/// following preset scaling factors are made available through your
/// application's user interface.
///
/// If your application does not expose the notion of preset scaling factors
/// for upscaling algorithms (perhaps instead implementing a fixed ratio which
/// is immutable) or implementing a more dynamic scaling scheme (such as
/// dynamic resolution scaling), then there is no need to use these presets.
///
/// Please note that <c><i>FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE</i></c> is
/// an optional mode which may introduce significant quality degradation in the
/// final image. As such it is recommended that you evaluate the final results
/// of using this scaling mode before deciding if you should include it in your
/// application.
///
/// @ingroup FSR2
typedef enum FfxFsr2QualityMode {
FFX_FSR2_QUALITY_MODE_QUALITY = 1, ///< Perform upscaling with a per-dimension upscaling ratio of 1.5x.
FFX_FSR2_QUALITY_MODE_BALANCED = 2, ///< Perform upscaling with a per-dimension upscaling ratio of 1.7x.
FFX_FSR2_QUALITY_MODE_PERFORMANCE = 3, ///< Perform upscaling with a per-dimension upscaling ratio of 2.0x.
FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE = 4 ///< Perform upscaling with a per-dimension upscaling ratio of 3.0x.
} FfxFsr2QualityMode;
/// An enumeration of bit flags used when creating a
/// <c><i>FfxFsr2Context</i></c>. See <c><i>FfxFsr2ContextDescription</i></c>.
///
/// @ingroup FSR2
typedef enum FfxFsr2InitializationFlagBits {
FFX_FSR2_ENABLE_HIGH_DYNAMIC_RANGE = (1<<0), ///< A bit indicating if the input color data provided is using a high-dynamic range.
FFX_FSR2_ENABLE_DISPLAY_RESOLUTION_MOTION_VECTORS = (1<<1), ///< A bit indicating if the motion vectors are rendered at display resolution.
FFX_FSR2_ENABLE_MOTION_VECTORS_JITTER_CANCELLATION = (1<<2), ///< A bit indicating that the motion vectors have the jittering pattern applied to them.
FFX_FSR2_ENABLE_DEPTH_INVERTED = (1<<3), ///< A bit indicating that the input depth buffer data provided is inverted [1..0].
FFX_FSR2_ENABLE_DEPTH_INFINITE = (1<<4), ///< A bit indicating that the input depth buffer data provided is using an infinite far plane.
FFX_FSR2_ENABLE_AUTO_EXPOSURE = (1<<5), ///< A bit indicating if automatic exposure should be applied to input color data.
FFX_FSR2_ENABLE_DYNAMIC_RESOLUTION = (1<<6), ///< A bit indicating that the application uses dynamic resolution scaling.
FFX_FSR2_ENABLE_TEXTURE1D_USAGE = (1<<7), ///< A bit indicating that the backend should use 1D textures.
FFX_FSR2_ENABLE_DEBUG_CHECKING = (1<<8), ///< A bit indicating that the runtime should check some API values and report issues.
} FfxFsr2InitializationFlagBits;
/// A structure encapsulating the parameters required to initialize FidelityFX
/// Super Resolution 2 upscaling.
///
/// @ingroup FSR2
typedef struct FfxFsr2ContextDescription {
uint32_t flags; ///< A collection of <c><i>FfxFsr2InitializationFlagBits</i></c>.
FfxDimensions2D maxRenderSize; ///< The maximum size that rendering will be performed at.
FfxDimensions2D displaySize; ///< The size of the presentation resolution targeted by the upscaling process.
FfxFsr2Interface callbacks; ///< A set of pointers to the backend implementation for FSR 2.0.
FfxDevice device; ///< The abstracted device which is passed to some callback functions.
FfxFsr2Message fpMessage; ///< A pointer to a function that can recieve messages from the runtime.
} FfxFsr2ContextDescription;
/// A structure encapsulating the parameters for dispatching the various passes
/// of FidelityFX Super Resolution 2.
///
/// @ingroup FSR2
typedef struct FfxFsr2DispatchDescription {
FfxCommandList commandList; ///< The <c><i>FfxCommandList</i></c> to record FSR2 rendering commands into.
FfxResource color; ///< A <c><i>FfxResource</i></c> containing the color buffer for the current frame (at render resolution).
FfxResource depth; ///< A <c><i>FfxResource</i></c> containing 32bit depth values for the current frame (at render resolution).
FfxResource motionVectors; ///< A <c><i>FfxResource</i></c> containing 2-dimensional motion vectors (at render resolution if <c><i>FFX_FSR2_ENABLE_DISPLAY_RESOLUTION_MOTION_VECTORS</i></c> is not set).
FfxResource exposure; ///< A optional <c><i>FfxResource</i></c> containing a 1x1 exposure value.
FfxResource reactive; ///< A optional <c><i>FfxResource</i></c> containing alpha value of reactive objects in the scene.
FfxResource transparencyAndComposition; ///< A optional <c><i>FfxResource</i></c> containing alpha value of special objects in the scene.
FfxResource output; ///< A <c><i>FfxResource</i></c> containing the output color buffer for the current frame (at presentation resolution).
FfxFloatCoords2D jitterOffset; ///< The subpixel jitter offset applied to the camera.
FfxFloatCoords2D motionVectorScale; ///< The scale factor to apply to motion vectors.
FfxDimensions2D renderSize; ///< The resolution that was used for rendering the input resources.
bool enableSharpening; ///< Enable an additional sharpening pass.
float sharpness; ///< The sharpness value between 0 and 1, where 0 is no additional sharpness and 1 is maximum additional sharpness.
float frameTimeDelta; ///< The time elapsed since the last frame (expressed in milliseconds).
float preExposure; ///< The pre exposure value (must be > 0.0f)
bool reset; ///< A boolean value which when set to true, indicates the camera has moved discontinuously.
float cameraNear; ///< The distance to the near plane of the camera.
float cameraFar; ///< The distance to the far plane of the camera.
float cameraFovAngleVertical; ///< The camera angle field of view in the vertical direction (expressed in radians).
float viewSpaceToMetersFactor; ///< The scale factor to convert view space units to meters
// EXPERIMENTAL reactive mask generation parameters
bool enableAutoReactive; ///< A boolean value to indicate internal reactive autogeneration should be used
FfxResource colorOpaqueOnly; ///< A <c><i>FfxResource</i></c> containing the opaque only color buffer for the current frame (at render resolution).
float autoTcThreshold; ///< Cutoff value for TC
float autoTcScale; ///< A value to scale the transparency and composition mask
float autoReactiveScale; ///< A value to scale the reactive mask
float autoReactiveMax; ///< A value to clamp the reactive mask
// -- GODOT start --
float reprojectionMatrix[16]; ///< The matrix used for reprojecting pixels with invalid motion vectors by using the depth.
// -- GODOT end --
} FfxFsr2DispatchDescription;
/// A structure encapsulating the parameters for automatic generation of a reactive mask
///
/// @ingroup FSR2
typedef struct FfxFsr2GenerateReactiveDescription {
FfxCommandList commandList; ///< The <c><i>FfxCommandList</i></c> to record FSR2 rendering commands into.
FfxResource colorOpaqueOnly; ///< A <c><i>FfxResource</i></c> containing the opaque only color buffer for the current frame (at render resolution).
FfxResource colorPreUpscale; ///< A <c><i>FfxResource</i></c> containing the opaque+translucent color buffer for the current frame (at render resolution).
FfxResource outReactive; ///< A <c><i>FfxResource</i></c> containing the surface to generate the reactive mask into.
FfxDimensions2D renderSize; ///< The resolution that was used for rendering the input resources.
float scale; ///< A value to scale the output
float cutoffThreshold; ///< A threshold value to generate a binary reactive mask
float binaryValue; ///< A value to set for the binary reactive mask
uint32_t flags; ///< Flags to determine how to generate the reactive mask
} FfxFsr2GenerateReactiveDescription;
/// A structure encapsulating the FidelityFX Super Resolution 2 context.
///
/// This sets up an object which contains all persistent internal data and
/// resources that are required by FSR2.
///
/// The <c><i>FfxFsr2Context</i></c> object should have a lifetime matching
/// your use of FSR2. Before destroying the FSR2 context care should be taken
/// to ensure the GPU is not accessing the resources created or used by FSR2.
/// It is therefore recommended that the GPU is idle before destroying the
/// FSR2 context.
///
/// @ingroup FSR2
typedef struct FfxFsr2Context {
uint32_t data[FFX_FSR2_CONTEXT_SIZE]; ///< An opaque set of <c>uint32_t</c> which contain the data for the context.
} FfxFsr2Context;
/// Create a FidelityFX Super Resolution 2 context from the parameters
/// programmed to the <c><i>FfxFsr2CreateParams</i></c> structure.
///
/// The context structure is the main object used to interact with the FSR2
/// API, and is responsible for the management of the internal resources used
/// by the FSR2 algorithm. When this API is called, multiple calls will be
/// made via the pointers contained in the <c><i>callbacks</i></c> structure.
/// These callbacks will attempt to retreive the device capabilities, and
/// create the internal resources, and pipelines required by FSR2's
/// frame-to-frame function. Depending on the precise configuration used when
/// creating the <c><i>FfxFsr2Context</i></c> a different set of resources and
/// pipelines might be requested via the callback functions.
///
/// The flags included in the <c><i>flags</i></c> field of
/// <c><i>FfxFsr2Context</i></c> how match the configuration of your
/// application as well as the intended use of FSR2. It is important that these
/// flags are set correctly (as well as a correct programmed
/// <c><i>FfxFsr2DispatchDescription</i></c>) to ensure correct operation. It is
/// recommended to consult the overview documentation for further details on
/// how FSR2 should be integerated into an application.
///
/// When the <c><i>FfxFsr2Context</i></c> is created, you should use the
/// <c><i>ffxFsr2ContextDispatch</i></c> function each frame where FSR2
/// upscaling should be applied. See the documentation of
/// <c><i>ffxFsr2ContextDispatch</i></c> for more details.
///
/// The <c><i>FfxFsr2Context</i></c> should be destroyed when use of it is
/// completed, typically when an application is unloaded or FSR2 upscaling is
/// disabled by a user. To destroy the FSR2 context you should call
/// <c><i>ffxFsr2ContextDestroy</i></c>.
///
/// @param [out] context A pointer to a <c><i>FfxFsr2Context</i></c> structure to populate.
/// @param [in] contextDescription A pointer to a <c><i>FfxFsr2ContextDescription</i></c> structure.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// FFX_ERROR_CODE_NULL_POINTER The operation failed because either <c><i>context</i></c> or <c><i>contextDescription</i></c> was <c><i>NULL</i></c>.
/// @retval
/// FFX_ERROR_INCOMPLETE_INTERFACE The operation failed because the <c><i>FfxFsr2ContextDescription.callbacks</i></c> was not fully specified.
/// @retval
/// FFX_ERROR_BACKEND_API_ERROR The operation failed because of an error returned from the backend.
///
/// @ingroup FSR2
FFX_API FfxErrorCode ffxFsr2ContextCreate(FfxFsr2Context* context, const FfxFsr2ContextDescription* contextDescription);
/// Dispatch the various passes that constitute FidelityFX Super Resolution 2.
///
/// FSR2 is a composite effect, meaning that it is compromised of multiple
/// constituent passes (implemented as one or more clears, copies and compute
/// dispatches). The <c><i>ffxFsr2ContextDispatch</i></c> function is the
/// function which (via the use of the functions contained in the
/// <c><i>callbacks</i></c> field of the <c><i>FfxFsr2Context</i></c>
/// structure) utlimately generates the sequence of graphics API calls required
/// each frame.
///
/// As with the creation of the <c><i>FfxFsr2Context</i></c> correctly
/// programming the <c><i>FfxFsr2DispatchDescription</i></c> is key to ensuring
/// the correct operation of FSR2. It is particularly important to ensure that
/// camera jitter is correctly applied to your application's projection matrix
/// (or camera origin for raytraced applications). FSR2 provides the
/// <c><i>ffxFsr2GetJitterPhaseCount</i></c> and
/// <c><i>ffxFsr2GetJitterOffset</i></c> entry points to help applications
/// correctly compute the camera jitter. Whatever jitter pattern is used by the
/// application it should be correctly programmed to the
/// <c><i>jitterOffset</i></c> field of the <c><i>dispatchDescription</i></c>
/// structure. For more guidance on camera jitter please consult the
/// documentation for <c><i>ffxFsr2GetJitterOffset</i></c> as well as the
/// accompanying overview documentation for FSR2.
///
/// @param [in] context A pointer to a <c><i>FfxFsr2Context</i></c> structure.
/// @param [in] dispatchDescription A pointer to a <c><i>FfxFsr2DispatchDescription</i></c> structure.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// FFX_ERROR_CODE_NULL_POINTER The operation failed because either <c><i>context</i></c> or <c><i>dispatchDescription</i></c> was <c><i>NULL</i></c>.
/// @retval
/// FFX_ERROR_OUT_OF_RANGE The operation failed because <c><i>dispatchDescription.renderSize</i></c> was larger than the maximum render resolution.
/// @retval
/// FFX_ERROR_NULL_DEVICE The operation failed because the device inside the context was <c><i>NULL</i></c>.
/// @retval
/// FFX_ERROR_BACKEND_API_ERROR The operation failed because of an error returned from the backend.
///
/// @ingroup FSR2
FFX_API FfxErrorCode ffxFsr2ContextDispatch(FfxFsr2Context* context, const FfxFsr2DispatchDescription* dispatchDescription);
/// A helper function generate a Reactive mask from an opaque only texure and one containing translucent objects.
///
/// @param [in] context A pointer to a <c><i>FfxFsr2Context</i></c> structure.
/// @param [in] params A pointer to a <c><i>FfxFsr2GenerateReactiveDescription</i></c> structure
///
/// @retval
/// FFX_OK The operation completed successfully.
///
/// @ingroup FSR2
FFX_API FfxErrorCode ffxFsr2ContextGenerateReactiveMask(FfxFsr2Context* context, const FfxFsr2GenerateReactiveDescription* params);
/// Destroy the FidelityFX Super Resolution context.
///
/// @param [out] context A pointer to a <c><i>FfxFsr2Context</i></c> structure to destroy.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// FFX_ERROR_CODE_NULL_POINTER The operation failed because either <c><i>context</i></c> was <c><i>NULL</i></c>.
///
/// @ingroup FSR2
FFX_API FfxErrorCode ffxFsr2ContextDestroy(FfxFsr2Context* context);
/// Get the upscale ratio from the quality mode.
///
/// The following table enumerates the mapping of the quality modes to
/// per-dimension scaling ratios.
///
/// Quality preset | Scale factor
/// ----------------------------------------------------- | -------------
/// <c><i>FFX_FSR2_QUALITY_MODE_QUALITY</i></c> | 1.5x
/// <c><i>FFX_FSR2_QUALITY_MODE_BALANCED</i></c> | 1.7x
/// <c><i>FFX_FSR2_QUALITY_MODE_PERFORMANCE</i></c> | 2.0x
/// <c><i>FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE</i></c> | 3.0x
///
/// Passing an invalid <c><i>qualityMode</i></c> will return 0.0f.
///
/// @param [in] qualityMode The quality mode preset.
///
/// @returns
/// The upscaling the per-dimension upscaling ratio for
/// <c><i>qualityMode</i></c> according to the table above.
///
/// @ingroup FSR2
FFX_API float ffxFsr2GetUpscaleRatioFromQualityMode(FfxFsr2QualityMode qualityMode);
/// A helper function to calculate the rendering resolution from a target
/// resolution and desired quality level.
///
/// This function applies the scaling factor returned by
/// <c><i>ffxFsr2GetUpscaleRatioFromQualityMode</i></c> to each dimension.
///
/// @param [out] renderWidth A pointer to a <c>uint32_t</c> which will hold the calculated render resolution width.
/// @param [out] renderHeight A pointer to a <c>uint32_t</c> which will hold the calculated render resolution height.
/// @param [in] displayWidth The target display resolution width.
/// @param [in] displayHeight The target display resolution height.
/// @param [in] qualityMode The desired quality mode for FSR 2 upscaling.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// FFX_ERROR_INVALID_POINTER Either <c><i>renderWidth</i></c> or <c><i>renderHeight</i></c> was <c>NULL</c>.
/// @retval
/// FFX_ERROR_INVALID_ENUM An invalid quality mode was specified.
///
/// @ingroup FSR2
FFX_API FfxErrorCode ffxFsr2GetRenderResolutionFromQualityMode(
uint32_t* renderWidth,
uint32_t* renderHeight,
uint32_t displayWidth,
uint32_t displayHeight,
FfxFsr2QualityMode qualityMode);
/// A helper function to calculate the jitter phase count from display
/// resolution.
///
/// For more detailed information about the application of camera jitter to
/// your application's rendering please refer to the
/// <c><i>ffxFsr2GetJitterOffset</i></c> function.
///
/// The table below shows the jitter phase count which this function
/// would return for each of the quality presets.
///
/// Quality preset | Scale factor | Phase count
/// ----------------------------------------------------- | ------------- | ---------------
/// <c><i>FFX_FSR2_QUALITY_MODE_QUALITY</i></c> | 1.5x | 18
/// <c><i>FFX_FSR2_QUALITY_MODE_BALANCED</i></c> | 1.7x | 23
/// <c><i>FFX_FSR2_QUALITY_MODE_PERFORMANCE</i></c> | 2.0x | 32
/// <c><i>FFX_FSR2_QUALITY_MODE_ULTRA_PERFORMANCE</i></c> | 3.0x | 72
/// Custom | [1..n]x | ceil(8*n^2)
///
/// @param [in] renderWidth The render resolution width.
/// @param [in] displayWidth The display resolution width.
///
/// @returns
/// The jitter phase count for the scaling factor between <c><i>renderWidth</i></c> and <c><i>displayWidth</i></c>.
///
/// @ingroup FSR2
FFX_API int32_t ffxFsr2GetJitterPhaseCount(int32_t renderWidth, int32_t displayWidth);
/// A helper function to calculate the subpixel jitter offset.
///
/// FSR2 relies on the application to apply sub-pixel jittering while rendering.
/// This is typically included in the projection matrix of the camera. To make
/// the application of camera jitter simple, the FSR2 API provides a small set
/// of utility function which computes the sub-pixel jitter offset for a
/// particular frame within a sequence of separate jitter offsets. To begin, the
/// index within the jitter phase must be computed. To calculate the
/// sequence's length, you can call the <c><i>ffxFsr2GetJitterPhaseCount</i></c>
/// function. The index should be a value which is incremented each frame modulo
/// the length of the sequence computed by <c><i>ffxFsr2GetJitterPhaseCount</i></c>.
/// The index within the jitter phase is passed to
/// <c><i>ffxFsr2GetJitterOffset</i></c> via the <c><i>index</i></c> parameter.
///
/// This function uses a Halton(2,3) sequence to compute the jitter offset.
/// The ultimate index used for the sequence is <c><i>index</i></c> %
/// <c><i>phaseCount</i></c>.
///
/// It is important to understand that the values returned from the
/// <c><i>ffxFsr2GetJitterOffset</i></c> function are in unit pixel space, and
/// in order to composite this correctly into a projection matrix we must
/// convert them into projection offsets. This is done as per the pseudo code
/// listing which is shown below.
///
/// const int32_t jitterPhaseCount = ffxFsr2GetJitterPhaseCount(renderWidth, displayWidth);
///
/// float jitterX = 0;
/// float jitterY = 0;
/// ffxFsr2GetJitterOffset(&jitterX, &jitterY, index, jitterPhaseCount);
///
/// const float jitterX = 2.0f * jitterX / (float)renderWidth;
/// const float jitterY = -2.0f * jitterY / (float)renderHeight;
/// const Matrix4 jitterTranslationMatrix = translateMatrix(Matrix3::identity, Vector3(jitterX, jitterY, 0));
/// const Matrix4 jitteredProjectionMatrix = jitterTranslationMatrix * projectionMatrix;
///
/// Jitter should be applied to all rendering. This includes opaque, alpha
/// transparent, and raytraced objects. For rasterized objects, the sub-pixel
/// jittering values calculated by the <c><i>iffxFsr2GetJitterOffset</i></c>
/// function can be applied to the camera projection matrix which is ultimately
/// used to perform transformations during vertex shading. For raytraced
/// rendering, the sub-pixel jitter should be applied to the ray's origin,
/// often the camera's position.
///
/// Whether you elect to use the <c><i>ffxFsr2GetJitterOffset</i></c> function
/// or your own sequence generator, you must program the
/// <c><i>jitterOffset</i></c> field of the
/// <c><i>FfxFsr2DispatchParameters</i></c> structure in order to inform FSR2
/// of the jitter offset that has been applied in order to render each frame.
///
/// If not using the recommended <c><i>ffxFsr2GetJitterOffset</i></c> function,
/// care should be taken that your jitter sequence never generates a null vector;
/// that is value of 0 in both the X and Y dimensions.
///
/// @param [out] outX A pointer to a <c>float</c> which will contain the subpixel jitter offset for the x dimension.
/// @param [out] outY A pointer to a <c>float</c> which will contain the subpixel jitter offset for the y dimension.
/// @param [in] index The index within the jitter sequence.
/// @param [in] phaseCount The length of jitter phase. See <c><i>ffxFsr2GetJitterPhaseCount</i></c>.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// FFX_ERROR_INVALID_POINTER Either <c><i>outX</i></c> or <c><i>outY</i></c> was <c>NULL</c>.
/// @retval
/// FFX_ERROR_INVALID_ARGUMENT Argument <c><i>phaseCount</i></c> must be greater than 0.
///
/// @ingroup FSR2
FFX_API FfxErrorCode ffxFsr2GetJitterOffset(float* outX, float* outY, int32_t index, int32_t phaseCount);
/// A helper function to check if a resource is
/// <c><i>FFX_FSR2_RESOURCE_IDENTIFIER_NULL</i></c>.
///
/// @param [in] resource A <c><i>FfxResource</i></c>.
///
/// @returns
/// true The <c><i>resource</i></c> was not <c><i>FFX_FSR2_RESOURCE_IDENTIFIER_NULL</i></c>.
/// @returns
/// false The <c><i>resource</i></c> was <c><i>FFX_FSR2_RESOURCE_IDENTIFIER_NULL</i></c>.
///
/// @ingroup FSR2
FFX_API bool ffxFsr2ResourceIsNull(FfxResource resource);
#if defined(__cplusplus)
}
#endif // #if defined(__cplusplus)

395
thirdparty/amd-fsr2/ffx_fsr2_interface.h vendored Normal file
View File

@ -0,0 +1,395 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#pragma once
#include "ffx_assert.h"
#include "ffx_types.h"
#include "ffx_error.h"
// Include the FSR2 resources defined in the HLSL code. This shared here to avoid getting out of sync.
#define FFX_CPU
#include "shaders/ffx_fsr2_resources.h"
#include "shaders/ffx_fsr2_common.h"
#if defined(__cplusplus)
extern "C" {
#endif // #if defined(__cplusplus)
FFX_FORWARD_DECLARE(FfxFsr2Interface);
/// An enumeration of all the passes which constitute the FSR2 algorithm.
///
/// FSR2 is implemented as a composite of several compute passes each
/// computing a key part of the final result. Each call to the
/// <c><i>FfxFsr2ScheduleGpuJobFunc</i></c> callback function will
/// correspond to a single pass included in <c><i>FfxFsr2Pass</i></c>. For a
/// more comprehensive description of each pass, please refer to the FSR2
/// reference documentation.
///
/// Please note in some cases e.g.: <c><i>FFX_FSR2_PASS_ACCUMULATE</i></c>
/// and <c><i>FFX_FSR2_PASS_ACCUMULATE_SHARPEN</i></c> either one pass or the
/// other will be used (they are mutually exclusive). The choice of which will
/// depend on the way the <c><i>FfxFsr2Context</i></c> is created and the
/// precise contents of <c><i>FfxFsr2DispatchParamters</i></c> each time a call
/// is made to <c><i>ffxFsr2ContextDispatch</i></c>.
///
/// @ingroup FSR2
typedef enum FfxFsr2Pass {
FFX_FSR2_PASS_DEPTH_CLIP = 0, ///< A pass which performs depth clipping.
FFX_FSR2_PASS_RECONSTRUCT_PREVIOUS_DEPTH = 1, ///< A pass which performs reconstruction of previous frame's depth.
FFX_FSR2_PASS_LOCK = 2, ///< A pass which calculates pixel locks.
FFX_FSR2_PASS_ACCUMULATE = 3, ///< A pass which performs upscaling.
FFX_FSR2_PASS_ACCUMULATE_SHARPEN = 4, ///< A pass which performs upscaling when sharpening is used.
FFX_FSR2_PASS_RCAS = 5, ///< A pass which performs sharpening.
FFX_FSR2_PASS_COMPUTE_LUMINANCE_PYRAMID = 6, ///< A pass which generates the luminance mipmap chain for the current frame.
FFX_FSR2_PASS_GENERATE_REACTIVE = 7, ///< An optional pass to generate a reactive mask
FFX_FSR2_PASS_TCR_AUTOGENERATE = 8, ///< An optional pass to generate a texture-and-composition and reactive masks
FFX_FSR2_PASS_COUNT ///< The number of passes performed by FSR2.
} FfxFsr2Pass;
typedef enum FfxFsr2MsgType {
FFX_FSR2_MESSAGE_TYPE_ERROR = 0,
FFX_FSR2_MESSAGE_TYPE_WARNING = 1,
FFX_FSR2_MESSAGE_TYPE_COUNT
} FfxFsr2MsgType;
/// Create and initialize the backend context.
///
/// The callback function sets up the backend context for rendering.
/// It will create or reference the device and create required internal data structures.
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [in] device The FfxDevice obtained by ffxGetDevice(DX12/VK/...).
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode (*FfxFsr2CreateBackendContextFunc)(
FfxFsr2Interface* backendInterface,
FfxDevice device);
/// Get a list of capabilities of the device.
///
/// When creating an <c><i>FfxFsr2Context</i></c> it is desirable for the FSR2
/// core implementation to be aware of certain characteristics of the platform
/// that is being targetted. This is because some optimizations which FSR2
/// attempts to perform are more effective on certain classes of hardware than
/// others, or are not supported by older hardware. In order to avoid cases
/// where optimizations actually have the effect of decreasing performance, or
/// reduce the breadth of support provided by FSR2, FSR2 queries the
/// capabilities of the device to make such decisions.
///
/// For target platforms with fixed hardware support you need not implement
/// this callback function by querying the device, but instead may hardcore
/// what features are available on the platform.
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [out] outDeviceCapabilities The device capabilities structure to fill out.
/// @param [in] device The device to query for capabilities.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode(*FfxFsr2GetDeviceCapabilitiesFunc)(
FfxFsr2Interface* backendInterface,
FfxDeviceCapabilities* outDeviceCapabilities,
FfxDevice device);
/// Destroy the backend context and dereference the device.
///
/// This function is called when the <c><i>FfxFsr2Context</i></c> is destroyed.
///
/// @param [in] backendInterface A pointer to the backend interface.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode(*FfxFsr2DestroyBackendContextFunc)(
FfxFsr2Interface* backendInterface);
/// Create a resource.
///
/// This callback is intended for the backend to create internal resources.
///
/// Please note: It is also possible that the creation of resources might
/// itself cause additional resources to be created by simply calling the
/// <c><i>FfxFsr2CreateResourceFunc</i></c> function pointer again. This is
/// useful when handling the initial creation of resources which must be
/// initialized. The flow in such a case would be an initial call to create the
/// CPU-side resource, another to create the GPU-side resource, and then a call
/// to schedule a copy render job to move the data between the two. Typically
/// this type of function call flow is only seen during the creation of an
/// <c><i>FfxFsr2Context</i></c>.
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [in] createResourceDescription A pointer to a <c><i>FfxCreateResourceDescription</i></c>.
/// @param [out] outResource A pointer to a <c><i>FfxResource</i></c> object.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode (*FfxFsr2CreateResourceFunc)(
FfxFsr2Interface* backendInterface,
const FfxCreateResourceDescription* createResourceDescription,
FfxResourceInternal* outResource);
/// Register a resource in the backend for the current frame.
///
/// Since FSR2 and the backend are not aware how many different
/// resources will get passed to FSR2 over time, it's not safe
/// to register all resources simultaneously in the backend.
/// Also passed resources may not be valid after the dispatch call.
/// As a result it's safest to register them as FfxResourceInternal
/// and clear them at the end of the dispatch call.
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [in] inResource A pointer to a <c><i>FfxResource</i></c>.
/// @param [out] outResource A pointer to a <c><i>FfxResourceInternal</i></c> object.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode(*FfxFsr2RegisterResourceFunc)(
FfxFsr2Interface* backendInterface,
const FfxResource* inResource,
FfxResourceInternal* outResource);
/// Unregister all temporary FfxResourceInternal from the backend.
///
/// Unregister FfxResourceInternal referencing resources passed to
/// a function as a parameter.
///
/// @param [in] backendInterface A pointer to the backend interface.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode(*FfxFsr2UnregisterResourcesFunc)(
FfxFsr2Interface* backendInterface);
/// Retrieve a <c><i>FfxResourceDescription</i></c> matching a
/// <c><i>FfxResource</i></c> structure.
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [in] resource A pointer to a <c><i>FfxResource</i></c> object.
///
/// @returns
/// A description of the resource.
///
/// @ingroup FSR2
typedef FfxResourceDescription (*FfxFsr2GetResourceDescriptionFunc)(
FfxFsr2Interface* backendInterface,
FfxResourceInternal resource);
/// Destroy a resource
///
/// This callback is intended for the backend to release an internal resource.
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [in] resource A pointer to a <c><i>FfxResource</i></c> object.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode (*FfxFsr2DestroyResourceFunc)(
FfxFsr2Interface* backendInterface,
FfxResourceInternal resource);
/// Create a render pipeline.
///
/// A rendering pipeline contains the shader as well as resource bindpoints
/// and samplers.
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [in] pass The identifier for the pass.
/// @param [in] pipelineDescription A pointer to a <c><i>FfxPipelineDescription</i></c> describing the pipeline to be created.
/// @param [out] outPipeline A pointer to a <c><i>FfxPipelineState</i></c> structure which should be populated.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode (*FfxFsr2CreatePipelineFunc)(
FfxFsr2Interface* backendInterface,
FfxFsr2Pass pass,
const FfxPipelineDescription* pipelineDescription,
FfxPipelineState* outPipeline);
/// Destroy a render pipeline.
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [out] pipeline A pointer to a <c><i>FfxPipelineState</i></c> structure which should be released.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode (*FfxFsr2DestroyPipelineFunc)(
FfxFsr2Interface* backendInterface,
FfxPipelineState* pipeline);
/// Schedule a render job to be executed on the next call of
/// <c><i>FfxFsr2ExecuteGpuJobsFunc</i></c>.
///
/// Render jobs can perform one of three different tasks: clear, copy or
/// compute dispatches.
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [in] job A pointer to a <c><i>FfxGpuJobDescription</i></c> structure.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode (*FfxFsr2ScheduleGpuJobFunc)(
FfxFsr2Interface* backendInterface,
const FfxGpuJobDescription* job);
/// Execute scheduled render jobs on the <c><i>comandList</i></c> provided.
///
/// The recording of the graphics API commands should take place in this
/// callback function, the render jobs which were previously enqueued (via
/// callbacks made to <c><i>FfxFsr2ScheduleGpuJobFunc</i></c>) should be
/// processed in the order they were received. Advanced users might choose to
/// reorder the rendering jobs, but should do so with care to respect the
/// resource dependencies.
///
/// Depending on the precise contents of <c><i>FfxFsr2DispatchDescription</i></c> a
/// different number of render jobs might have previously been enqueued (for
/// example if sharpening is toggled on and off).
///
/// @param [in] backendInterface A pointer to the backend interface.
/// @param [in] commandList A pointer to a <c><i>FfxCommandList</i></c> structure.
///
/// @retval
/// FFX_OK The operation completed successfully.
/// @retval
/// Anything else The operation failed.
///
/// @ingroup FSR2
typedef FfxErrorCode (*FfxFsr2ExecuteGpuJobsFunc)(
FfxFsr2Interface* backendInterface,
FfxCommandList commandList);
/// Pass a string message
///
/// Used for debug messages.
///
/// @param [in] type The type of message.
/// @param [in] message A string message to pass.
///
///
/// @ingroup FSR2
typedef void(*FfxFsr2Message)(
FfxFsr2MsgType type,
const wchar_t* message);
/// A structure encapsulating the interface between the core implentation of
/// the FSR2 algorithm and any graphics API that it should ultimately call.
///
/// This set of functions serves as an abstraction layer between FSR2 and the
/// API used to implement it. While FSR2 ships with backends for DirectX12 and
/// Vulkan, it is possible to implement your own backend for other platforms or
/// which sits ontop of your engine's own abstraction layer. For details on the
/// expectations of what each function should do you should refer the
/// description of the following function pointer types:
///
/// <c><i>FfxFsr2CreateDeviceFunc</i></c>
/// <c><i>FfxFsr2GetDeviceCapabilitiesFunc</i></c>
/// <c><i>FfxFsr2DestroyDeviceFunc</i></c>
/// <c><i>FfxFsr2CreateResourceFunc</i></c>
/// <c><i>FfxFsr2GetResourceDescriptionFunc</i></c>
/// <c><i>FfxFsr2DestroyResourceFunc</i></c>
/// <c><i>FfxFsr2CreatePipelineFunc</i></c>
/// <c><i>FfxFsr2DestroyPipelineFunc</i></c>
/// <c><i>FfxFsr2ScheduleGpuJobFunc</i></c>
/// <c><i>FfxFsr2ExecuteGpuJobsFunc</i></c>
///
/// Depending on the graphics API that is abstracted by the backend, it may be
/// required that the backend is to some extent stateful. To ensure that
/// applications retain full control to manage the memory used by FSR2, the
/// <c><i>scratchBuffer</i></c> and <c><i>scratchBufferSize</i></c> fields are
/// provided. A backend should provide a means of specifying how much scratch
/// memory is required for its internal implementation (e.g: via a function
/// or constant value). The application is that responsible for allocating that
/// memory and providing it when setting up the FSR2 backend. Backends provided
/// with FSR2 do not perform dynamic memory allocations, and instead
/// suballocate all memory from the scratch buffers provided.
///
/// The <c><i>scratchBuffer</i></c> and <c><i>scratchBufferSize</i></c> fields
/// should be populated according to the requirements of each backend. For
/// example, if using the DirectX 12 backend you should call the
/// <c><i>ffxFsr2GetScratchMemorySizeDX12</i></c> function. It is not required
/// that custom backend implementations use a scratch buffer.
///
/// @ingroup FSR2
typedef struct FfxFsr2Interface {
FfxFsr2CreateBackendContextFunc fpCreateBackendContext; ///< A callback function to create and initialize the backend context.
FfxFsr2GetDeviceCapabilitiesFunc fpGetDeviceCapabilities; ///< A callback function to query device capabilites.
FfxFsr2DestroyBackendContextFunc fpDestroyBackendContext; ///< A callback function to destroy the backendcontext. This also dereferences the device.
FfxFsr2CreateResourceFunc fpCreateResource; ///< A callback function to create a resource.
FfxFsr2RegisterResourceFunc fpRegisterResource; ///< A callback function to register an external resource.
FfxFsr2UnregisterResourcesFunc fpUnregisterResources; ///< A callback function to unregister external resource.
FfxFsr2GetResourceDescriptionFunc fpGetResourceDescription; ///< A callback function to retrieve a resource description.
FfxFsr2DestroyResourceFunc fpDestroyResource; ///< A callback function to destroy a resource.
FfxFsr2CreatePipelineFunc fpCreatePipeline; ///< A callback function to create a render or compute pipeline.
FfxFsr2DestroyPipelineFunc fpDestroyPipeline; ///< A callback function to destroy a render or compute pipeline.
FfxFsr2ScheduleGpuJobFunc fpScheduleGpuJob; ///< A callback function to schedule a render job.
FfxFsr2ExecuteGpuJobsFunc fpExecuteGpuJobs; ///< A callback function to execute all queued render jobs.
void* scratchBuffer; ///< A preallocated buffer for memory utilized internally by the backend.
size_t scratchBufferSize; ///< Size of the buffer pointed to by <c><i>scratchBuffer</i></c>.
} FfxFsr2Interface;
#if defined(__cplusplus)
}
#endif // #if defined(__cplusplus)

46
thirdparty/amd-fsr2/ffx_fsr2_maximum_bias.h vendored Normal file
View File

@ -0,0 +1,46 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
// @internal
#pragma once
static const int FFX_FSR2_MAXIMUM_BIAS_TEXTURE_WIDTH = 16;
static const int FFX_FSR2_MAXIMUM_BIAS_TEXTURE_HEIGHT = 16;
static const float ffxFsr2MaximumBias[] = {
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.876f, 1.809f, 1.772f, 1.753f, 1.748f,
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.869f, 1.801f, 1.764f, 1.745f, 1.739f,
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.976f, 1.841f, 1.774f, 1.737f, 1.716f, 1.71f,
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.914f, 1.784f, 1.716f, 1.673f, 1.649f, 1.641f,
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.793f, 1.676f, 1.604f, 1.562f, 1.54f, 1.533f,
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.802f, 1.619f, 1.536f, 1.492f, 1.467f, 1.454f, 1.449f,
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.812f, 1.575f, 1.496f, 1.456f, 1.432f, 1.416f, 1.408f, 1.405f,
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.555f, 1.479f, 1.438f, 1.413f, 1.398f, 1.387f, 1.381f, 1.379f,
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.812f, 1.555f, 1.474f, 1.43f, 1.404f, 1.387f, 1.376f, 1.368f, 1.363f, 1.362f,
2.0f, 2.0f, 2.0f, 2.0f, 2.0f, 1.802f, 1.575f, 1.479f, 1.43f, 1.401f, 1.382f, 1.369f, 1.36f, 1.354f, 1.351f, 1.35f,
2.0f, 2.0f, 1.976f, 1.914f, 1.793f, 1.619f, 1.496f, 1.438f, 1.404f, 1.382f, 1.367f, 1.357f, 1.349f, 1.344f, 1.341f, 1.34f,
1.876f, 1.869f, 1.841f, 1.784f, 1.676f, 1.536f, 1.456f, 1.413f, 1.387f, 1.369f, 1.357f, 1.347f, 1.341f, 1.336f, 1.333f, 1.332f,
1.809f, 1.801f, 1.774f, 1.716f, 1.604f, 1.492f, 1.432f, 1.398f, 1.376f, 1.36f, 1.349f, 1.341f, 1.335f, 1.33f, 1.328f, 1.327f,
1.772f, 1.764f, 1.737f, 1.673f, 1.562f, 1.467f, 1.416f, 1.387f, 1.368f, 1.354f, 1.344f, 1.336f, 1.33f, 1.326f, 1.323f, 1.323f,
1.753f, 1.745f, 1.716f, 1.649f, 1.54f, 1.454f, 1.408f, 1.381f, 1.363f, 1.351f, 1.341f, 1.333f, 1.328f, 1.323f, 1.321f, 1.32f,
1.748f, 1.739f, 1.71f, 1.641f, 1.533f, 1.449f, 1.405f, 1.379f, 1.362f, 1.35f, 1.34f, 1.332f, 1.327f, 1.323f, 1.32f, 1.319f,
};

86
thirdparty/amd-fsr2/ffx_fsr2_private.h vendored Normal file
View File

@ -0,0 +1,86 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#pragma once
// Constants for FSR2 DX12 dispatches. Must be kept in sync with cbFSR2 in ffx_fsr2_callbacks_hlsl.h
typedef struct Fsr2Constants {
int32_t renderSize[2];
int32_t maxRenderSize[2];
int32_t displaySize[2];
int32_t inputColorResourceDimensions[2];
int32_t lumaMipDimensions[2];
int32_t lumaMipLevelToUse;
int32_t frameIndex;
float deviceToViewDepth[4];
float jitterOffset[2];
float motionVectorScale[2];
float downscaleFactor[2];
float motionVectorJitterCancellation[2];
float preExposure;
float previousFramePreExposure;
float tanHalfFOV;
float jitterPhaseCount;
float deltaTime;
float dynamicResChangeFactor;
float viewSpaceToMetersFactor;
// -- GODOT start --
float pad;
float reprojectionMatrix[16];
// -- GODOT end --
} Fsr2Constants;
struct FfxFsr2ContextDescription;
struct FfxDeviceCapabilities;
struct FfxPipelineState;
struct FfxResource;
// FfxFsr2Context_Private
// The private implementation of the FSR2 context.
typedef struct FfxFsr2Context_Private {
FfxFsr2ContextDescription contextDescription;
Fsr2Constants constants;
FfxDevice device;
FfxDeviceCapabilities deviceCapabilities;
FfxPipelineState pipelineDepthClip;
FfxPipelineState pipelineReconstructPreviousDepth;
FfxPipelineState pipelineLock;
FfxPipelineState pipelineAccumulate;
FfxPipelineState pipelineAccumulateSharpen;
FfxPipelineState pipelineRCAS;
FfxPipelineState pipelineComputeLuminancePyramid;
FfxPipelineState pipelineGenerateReactive;
FfxPipelineState pipelineTcrAutogenerate;
// 2 arrays of resources, as e.g. FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS will use different resources when bound as SRV vs when bound as UAV
FfxResourceInternal srvResources[FFX_FSR2_RESOURCE_IDENTIFIER_COUNT];
FfxResourceInternal uavResources[FFX_FSR2_RESOURCE_IDENTIFIER_COUNT];
bool firstExecution;
bool refreshPipelineStates;
uint32_t resourceFrameIndex;
float previousJitterOffset[2];
int32_t jitterPhaseCountRemaining;
} FfxFsr2Context_Private;

367
thirdparty/amd-fsr2/ffx_types.h vendored Normal file
View File

@ -0,0 +1,367 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#pragma once
#include <stdint.h>
// -- GODOT start --
#include <stdlib.h>
// -- GODOT end --
#if defined (FFX_GCC)
/// FidelityFX exported functions
#define FFX_API
#else
/// FidelityFX exported functions
#define FFX_API __declspec(dllexport)
#endif // #if defined (FFX_GCC)
/// Maximum supported number of simultaneously bound SRVs.
#define FFX_MAX_NUM_SRVS 16
/// Maximum supported number of simultaneously bound UAVs.
#define FFX_MAX_NUM_UAVS 8
/// Maximum number of constant buffers bound.
#define FFX_MAX_NUM_CONST_BUFFERS 2
/// Maximum size of bound constant buffers.
#define FFX_MAX_CONST_SIZE 64
/// Off by default warnings
#if defined(_MSC_VER)
#pragma warning(disable : 4365 4710 4820 5039)
#elif defined(__clang__)
#pragma clang diagnostic ignored "-Wunused-parameter"
#pragma clang diagnostic ignored "-Wmissing-field-initializers"
#pragma clang diagnostic ignored "-Wsign-compare"
#pragma clang diagnostic ignored "-Wunused-function"
#pragma clang diagnostic ignored "-Wignored-qualifiers"
#elif defined(__GNUC__)
#pragma GCC diagnostic ignored "-Wunused-function"
#endif
#ifdef __cplusplus
extern "C" {
#endif // #ifdef __cplusplus
/// An enumeration of surface formats.
typedef enum FfxSurfaceFormat {
FFX_SURFACE_FORMAT_UNKNOWN, ///< Unknown format
FFX_SURFACE_FORMAT_R32G32B32A32_TYPELESS, ///< 32 bit per channel, 4 channel typeless format
FFX_SURFACE_FORMAT_R32G32B32A32_FLOAT, ///< 32 bit per channel, 4 channel float format
FFX_SURFACE_FORMAT_R16G16B16A16_FLOAT, ///< 16 bit per channel, 4 channel float format
FFX_SURFACE_FORMAT_R16G16B16A16_UNORM, ///< 16 bit per channel, 4 channel unsigned normalized format
FFX_SURFACE_FORMAT_R32G32_FLOAT, ///< 32 bit per channel, 2 channel float format
FFX_SURFACE_FORMAT_R32_UINT, ///< 32 bit per channel, 1 channel float format
FFX_SURFACE_FORMAT_R8G8B8A8_TYPELESS, ///< 8 bit per channel, 4 channel float format
FFX_SURFACE_FORMAT_R8G8B8A8_UNORM, ///< 8 bit per channel, 4 channel unsigned normalized format
FFX_SURFACE_FORMAT_R11G11B10_FLOAT, ///< 32 bit 3 channel float format
FFX_SURFACE_FORMAT_R16G16_FLOAT, ///< 16 bit per channel, 2 channel float format
FFX_SURFACE_FORMAT_R16G16_UINT, ///< 16 bit per channel, 2 channel unsigned int format
FFX_SURFACE_FORMAT_R16_FLOAT, ///< 16 bit per channel, 1 channel float format
FFX_SURFACE_FORMAT_R16_UINT, ///< 16 bit per channel, 1 channel unsigned int format
FFX_SURFACE_FORMAT_R16_UNORM, ///< 16 bit per channel, 1 channel unsigned normalized format
FFX_SURFACE_FORMAT_R16_SNORM, ///< 16 bit per channel, 1 channel signed normalized format
FFX_SURFACE_FORMAT_R8_UNORM, ///< 8 bit per channel, 1 channel unsigned normalized format
FFX_SURFACE_FORMAT_R8_UINT, ///< 8 bit per channel, 1 channel unsigned int format
FFX_SURFACE_FORMAT_R8G8_UNORM, ///< 8 bit per channel, 2 channel unsigned normalized format
FFX_SURFACE_FORMAT_R32_FLOAT ///< 32 bit per channel, 1 channel float format
} FfxSurfaceFormat;
/// An enumeration of resource usage.
typedef enum FfxResourceUsage {
FFX_RESOURCE_USAGE_READ_ONLY = 0, ///< No usage flags indicate a resource is read only.
FFX_RESOURCE_USAGE_RENDERTARGET = (1<<0), ///< Indicates a resource will be used as render target.
FFX_RESOURCE_USAGE_UAV = (1<<1), ///< Indicates a resource will be used as UAV.
} FfxResourceUsage;
/// An enumeration of resource states.
typedef enum FfxResourceStates {
FFX_RESOURCE_STATE_UNORDERED_ACCESS = (1<<0), ///< Indicates a resource is in the state to be used as UAV.
FFX_RESOURCE_STATE_COMPUTE_READ = (1 << 1), ///< Indicates a resource is in the state to be read by compute shaders.
FFX_RESOURCE_STATE_COPY_SRC = (1 << 2), ///< Indicates a resource is in the state to be used as source in a copy command.
FFX_RESOURCE_STATE_COPY_DEST = (1 << 3), ///< Indicates a resource is in the state to be used as destination in a copy command.
FFX_RESOURCE_STATE_GENERIC_READ = (FFX_RESOURCE_STATE_COPY_SRC | FFX_RESOURCE_STATE_COMPUTE_READ), ///< Indicates a resource is in generic (slow) read state.
} FfxResourceStates;
/// An enumeration of surface dimensions.
typedef enum FfxResourceDimension {
FFX_RESOURCE_DIMENSION_TEXTURE_1D, ///< A resource with a single dimension.
FFX_RESOURCE_DIMENSION_TEXTURE_2D, ///< A resource with two dimensions.
} FfxResourceDimension;
/// An enumeration of surface dimensions.
typedef enum FfxResourceFlags {
FFX_RESOURCE_FLAGS_NONE = 0, ///< No flags.
FFX_RESOURCE_FLAGS_ALIASABLE = (1<<0), ///< A bit indicating a resource does not need to persist across frames.
} FfxResourceFlags;
/// An enumeration of all resource view types.
typedef enum FfxResourceViewType {
FFX_RESOURCE_VIEW_UNORDERED_ACCESS, ///< The resource view is an unordered access view (UAV).
FFX_RESOURCE_VIEW_SHADER_READ, ///< The resource view is a shader resource view (SRV).
} FfxResourceViewType;
/// The type of filtering to perform when reading a texture.
typedef enum FfxFilterType {
FFX_FILTER_TYPE_POINT, ///< Point sampling.
FFX_FILTER_TYPE_LINEAR ///< Sampling with interpolation.
} FfxFilterType;
/// An enumeration of all supported shader models.
typedef enum FfxShaderModel {
FFX_SHADER_MODEL_5_1, ///< Shader model 5.1.
FFX_SHADER_MODEL_6_0, ///< Shader model 6.0.
FFX_SHADER_MODEL_6_1, ///< Shader model 6.1.
FFX_SHADER_MODEL_6_2, ///< Shader model 6.2.
FFX_SHADER_MODEL_6_3, ///< Shader model 6.3.
FFX_SHADER_MODEL_6_4, ///< Shader model 6.4.
FFX_SHADER_MODEL_6_5, ///< Shader model 6.5.
FFX_SHADER_MODEL_6_6, ///< Shader model 6.6.
FFX_SHADER_MODEL_6_7, ///< Shader model 6.7.
} FfxShaderModel;
// An enumeration for different resource types
typedef enum FfxResourceType {
FFX_RESOURCE_TYPE_BUFFER, ///< The resource is a buffer.
FFX_RESOURCE_TYPE_TEXTURE1D, ///< The resource is a 1-dimensional texture.
FFX_RESOURCE_TYPE_TEXTURE2D, ///< The resource is a 2-dimensional texture.
FFX_RESOURCE_TYPE_TEXTURE3D, ///< The resource is a 3-dimensional texture.
} FfxResourceType;
/// An enumeration for different heap types
typedef enum FfxHeapType {
FFX_HEAP_TYPE_DEFAULT = 0, ///< Local memory.
FFX_HEAP_TYPE_UPLOAD ///< Heap used for uploading resources.
} FfxHeapType;
/// An enumberation for different render job types
typedef enum FfxGpuJobType {
FFX_GPU_JOB_CLEAR_FLOAT = 0, ///< The GPU job is performing a floating-point clear.
FFX_GPU_JOB_COPY = 1, ///< The GPU job is performing a copy.
FFX_GPU_JOB_COMPUTE = 2, ///< The GPU job is performing a compute dispatch.
} FfxGpuJobType;
/// A typedef representing the graphics device.
typedef void* FfxDevice;
/// A typedef representing a command list or command buffer.
typedef void* FfxCommandList;
/// A typedef for a root signature.
typedef void* FfxRootSignature;
/// A typedef for a pipeline state object.
typedef void* FfxPipeline;
/// A structure encapasulating a collection of device capabilities.
typedef struct FfxDeviceCapabilities {
FfxShaderModel minimumSupportedShaderModel; ///< The minimum shader model supported by the device.
uint32_t waveLaneCountMin; ///< The minimum supported wavefront width.
uint32_t waveLaneCountMax; ///< The maximum supported wavefront width.
bool fp16Supported; ///< The device supports FP16 in hardware.
bool raytracingSupported; ///< The device supports raytracing.
} FfxDeviceCapabilities;
/// A structure encapsulating a 2-dimensional point, using 32bit unsigned integers.
typedef struct FfxDimensions2D {
uint32_t width; ///< The width of a 2-dimensional range.
uint32_t height; ///< The height of a 2-dimensional range.
} FfxDimensions2D;
/// A structure encapsulating a 2-dimensional point,
typedef struct FfxIntCoords2D {
int32_t x; ///< The x coordinate of a 2-dimensional point.
int32_t y; ///< The y coordinate of a 2-dimensional point.
} FfxIntCoords2D;
/// A structure encapsulating a 2-dimensional set of floating point coordinates.
typedef struct FfxFloatCoords2D {
float x; ///< The x coordinate of a 2-dimensional point.
float y; ///< The y coordinate of a 2-dimensional point.
} FfxFloatCoords2D;
/// A structure describing a resource.
typedef struct FfxResourceDescription {
FfxResourceType type; ///< The type of the resource.
FfxSurfaceFormat format; ///< The surface format.
uint32_t width; ///< The width of the resource.
uint32_t height; ///< The height of the resource.
uint32_t depth; ///< The depth of the resource.
uint32_t mipCount; ///< Number of mips (or 0 for full mipchain).
FfxResourceFlags flags; ///< A set of <c><i>FfxResourceFlags</i></c> flags.
} FfxResourceDescription;
/// An outward facing structure containing a resource
typedef struct FfxResource {
void* resource; ///< pointer to the resource.
wchar_t name[64];
FfxResourceDescription description;
FfxResourceStates state;
bool isDepth;
uint64_t descriptorData;
} FfxResource;
/// An internal structure containing a handle to a resource and resource views
typedef struct FfxResourceInternal {
int32_t internalIndex; ///< The index of the resource.
} FfxResourceInternal;
/// A structure defining a resource bind point
typedef struct FfxResourceBinding
{
uint32_t slotIndex;
uint32_t resourceIdentifier;
wchar_t name[64];
}FfxResourceBinding;
/// A structure encapsulating a single pass of an algorithm.
typedef struct FfxPipelineState {
FfxRootSignature rootSignature; ///< The pipelines rootSignature
FfxPipeline pipeline; ///< The pipeline object
uint32_t uavCount; ///< Count of UAVs used in this pipeline
uint32_t srvCount; ///< Count of SRVs used in this pipeline
uint32_t constCount; ///< Count of constant buffers used in this pipeline
FfxResourceBinding uavResourceBindings[FFX_MAX_NUM_UAVS]; ///< Array of ResourceIdentifiers bound as UAVs
FfxResourceBinding srvResourceBindings[FFX_MAX_NUM_SRVS]; ///< Array of ResourceIdentifiers bound as SRVs
FfxResourceBinding cbResourceBindings[FFX_MAX_NUM_CONST_BUFFERS]; ///< Array of ResourceIdentifiers bound as CBs
} FfxPipelineState;
/// A structure containing the data required to create a resource.
typedef struct FfxCreateResourceDescription {
FfxHeapType heapType; ///< The heap type to hold the resource, typically <c><i>FFX_HEAP_TYPE_DEFAULT</i></c>.
FfxResourceDescription resourceDescription; ///< A resource description.
FfxResourceStates initalState; ///< The initial resource state.
uint32_t initDataSize; ///< Size of initial data buffer.
void* initData; ///< Buffer containing data to fill the resource.
const wchar_t* name; ///< Name of the resource.
FfxResourceUsage usage; ///< Resource usage flags.
uint32_t id; ///< Internal resource ID.
} FfxCreateResourceDescription;
/// A structure containing the description used to create a
/// <c><i>FfxPipeline</i></c> structure.
///
/// A pipeline is the name given to a shader and the collection of state that
/// is required to dispatch it. In the context of FSR2 and its architecture
/// this means that a <c><i>FfxPipelineDescription</i></c> will map to either a
/// monolithic object in an explicit API (such as a
/// <c><i>PipelineStateObject</i></c> in DirectX 12). Or a shader and some
/// ancillary API objects (in something like DirectX 11).
///
/// The <c><i>contextFlags</i></c> field contains a copy of the flags passed
/// to <c><i>ffxFsr2ContextCreate</i></c> via the <c><i>flags</i></c> field of
/// the <c><i>FfxFsr2InitializationParams</i></c> structure. These flags are
/// used to determine which permutation of a pipeline for a specific
/// <c><i>FfxFsr2Pass</i></c> should be used to implement the features required
/// by each application, as well as to acheive the best performance on specific
/// target hardware configurations.
///
/// When using one of the provided backends for FSR2 (such as DirectX 12 or
/// Vulkan) the data required to create a pipeline is compiled offline and
/// included into the backend library that you are using. For cases where the
/// backend interface is overriden by providing custom callback function
/// implementations care should be taken to respect the contents of the
/// <c><i>contextFlags</i></c> field in order to correctly support the options
/// provided by FSR2, and acheive best performance.
///
/// @ingroup FSR2
typedef struct FfxPipelineDescription {
uint32_t contextFlags; ///< A collection of <c><i>FfxFsr2InitializationFlagBits</i></c> which were passed to the context.
FfxFilterType* samplers; ///< Array of static samplers.
size_t samplerCount; ///< The number of samples contained inside <c><i>samplers</i></c>.
const uint32_t* rootConstantBufferSizes; ///< Array containing the sizes of the root constant buffers (count of 32 bit elements).
uint32_t rootConstantBufferCount; ///< The number of root constants contained within <c><i>rootConstantBufferSizes</i></c>.
} FfxPipelineDescription;
/// A structure containing a constant buffer.
typedef struct FfxConstantBuffer {
uint32_t uint32Size; ///< Size of 32 bit chunks used in the constant buffer
uint32_t data[FFX_MAX_CONST_SIZE]; ///< Constant buffer data
}FfxConstantBuffer;
/// A structure describing a clear render job.
typedef struct FfxClearFloatJobDescription {
float color[4]; ///< The clear color of the resource.
FfxResourceInternal target; ///< The resource to be cleared.
} FfxClearFloatJobDescription;
/// A structure describing a compute render job.
typedef struct FfxComputeJobDescription {
FfxPipelineState pipeline; ///< Compute pipeline for the render job.
uint32_t dimensions[3]; ///< Dispatch dimensions.
FfxResourceInternal srvs[FFX_MAX_NUM_SRVS]; ///< SRV resources to be bound in the compute job.
wchar_t srvNames[FFX_MAX_NUM_SRVS][64];
FfxResourceInternal uavs[FFX_MAX_NUM_UAVS]; ///< UAV resources to be bound in the compute job.
uint32_t uavMip[FFX_MAX_NUM_UAVS]; ///< Mip level of UAV resources to be bound in the compute job.
wchar_t uavNames[FFX_MAX_NUM_UAVS][64];
FfxConstantBuffer cbs[FFX_MAX_NUM_CONST_BUFFERS]; ///< Constant buffers to be bound in the compute job.
wchar_t cbNames[FFX_MAX_NUM_CONST_BUFFERS][64];
uint32_t cbSlotIndex[FFX_MAX_NUM_CONST_BUFFERS]; ///< Slot index in the descriptor table
} FfxComputeJobDescription;
/// A structure describing a copy render job.
typedef struct FfxCopyJobDescription
{
FfxResourceInternal src; ///< Source resource for the copy.
FfxResourceInternal dst; ///< Destination resource for the copy.
} FfxCopyJobDescription;
/// A structure describing a single render job.
typedef struct FfxGpuJobDescription{
FfxGpuJobType jobType; ///< Type of the job.
union {
FfxClearFloatJobDescription clearJobDescriptor; ///< Clear job descriptor. Valid when <c><i>jobType</i></c> is <c><i>FFX_RENDER_JOB_CLEAR_FLOAT</i></c>.
FfxCopyJobDescription copyJobDescriptor; ///< Copy job descriptor. Valid when <c><i>jobType</i></c> is <c><i>FFX_RENDER_JOB_COPY</i></c>.
FfxComputeJobDescription computeJobDescriptor; ///< Compute job descriptor. Valid when <c><i>jobType</i></c> is <c><i>FFX_RENDER_JOB_COMPUTE</i></c>.
};
} FfxGpuJobDescription;
#ifdef __cplusplus
}
#endif // #ifdef __cplusplus

78
thirdparty/amd-fsr2/ffx_util.h vendored Normal file
View File

@ -0,0 +1,78 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#pragma once
#include "ffx_types.h"
/// The value of Pi.
const float FFX_PI = 3.141592653589793f;
/// An epsilon value for floating point numbers.
const float FFX_EPSILON = 1e-06f;
/// Helper macro to create the version number.
#define FFX_MAKE_VERSION(major, minor, patch) ((major << 22) | (minor << 12) | patch)
///< Use this to specify no version.
#define FFX_UNSPECIFIED_VERSION 0xFFFFAD00
/// Helper macro to avoid warnings about unused variables.
#define FFX_UNUSED(x) ((void)(x))
/// Helper macro to align an integer to the specified power of 2 boundary
#define FFX_ALIGN_UP(x, y) (((x) + ((y)-1)) & ~((y)-1))
/// Helper macro to check if a value is aligned.
#define FFX_IS_ALIGNED(x) (((x) != 0) && ((x) & ((x)-1)))
/// Helper macro to stringify a value.
#define FFX_STR(s) FFX_XSTR(s)
#define FFX_XSTR(s) #s
/// Helper macro to forward declare a structure.
#define FFX_FORWARD_DECLARE(x) typedef struct x x
/// Helper macro to return the maximum of two values.
#define FFX_MAXIMUM(x, y) (((x) > (y)) ? (x) : (y))
/// Helper macro to return the minimum of two values.
#define FFX_MINIMUM(x, y) (((x) < (y)) ? (x) : (y))
/// Helper macro to do safe free on a pointer.
#define FFX_SAFE_FREE(x) \
if (x) \
free(x)
/// Helper macro to return the abs of an integer value.
#define FFX_ABSOLUTE(x) (((x) < 0) ? (-(x)) : (x))
/// Helper macro to return sign of a value.
#define FFX_SIGN(x) (((x) < 0) ? -1 : 1)
/// Helper macro to work out the number of elements in an array.
#define FFX_ARRAY_ELEMENTS(x) (int32_t)((sizeof(x) / sizeof(0 [x])) / ((size_t)(!(sizeof(x) % sizeof(0 [x])))))
/// The maximum length of a path that can be specified to the FidelityFX API.
#define FFX_MAXIMUM_PATH (260)
/// Helper macro to check if the specified key is set in a bitfield.
#define FFX_CONTAINS_FLAG(options, key) ((options & key) == key)

BIN
thirdparty/amd-fsr2/patches/godot-changes.patch vendored Normal file
View File

Binary file not shown.

429
thirdparty/amd-fsr2/shaders/ffx_common_types.h vendored Normal file
View File

@ -0,0 +1,429 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#ifndef FFX_COMMON_TYPES_H
#define FFX_COMMON_TYPES_H
#if defined(FFX_CPU)
#define FFX_PARAMETER_IN
#define FFX_PARAMETER_OUT
#define FFX_PARAMETER_INOUT
#elif defined(FFX_HLSL)
#define FFX_PARAMETER_IN in
#define FFX_PARAMETER_OUT out
#define FFX_PARAMETER_INOUT inout
#elif defined(FFX_GLSL)
#define FFX_PARAMETER_IN in
#define FFX_PARAMETER_OUT out
#define FFX_PARAMETER_INOUT inout
#endif // #if defined(FFX_CPU)
#if defined(FFX_CPU)
/// A typedef for a boolean value.
///
/// @ingroup CPU
typedef bool FfxBoolean;
/// A typedef for a unsigned 8bit integer.
///
/// @ingroup CPU
typedef uint8_t FfxUInt8;
/// A typedef for a unsigned 16bit integer.
///
/// @ingroup CPU
typedef uint16_t FfxUInt16;
/// A typedef for a unsigned 32bit integer.
///
/// @ingroup CPU
typedef uint32_t FfxUInt32;
/// A typedef for a unsigned 64bit integer.
///
/// @ingroup CPU
typedef uint64_t FfxUInt64;
/// A typedef for a signed 8bit integer.
///
/// @ingroup CPU
typedef int8_t FfxInt8;
/// A typedef for a signed 16bit integer.
///
/// @ingroup CPU
typedef int16_t FfxInt16;
/// A typedef for a signed 32bit integer.
///
/// @ingroup CPU
typedef int32_t FfxInt32;
/// A typedef for a signed 64bit integer.
///
/// @ingroup CPU
typedef int64_t FfxInt64;
/// A typedef for a floating point value.
///
/// @ingroup CPU
typedef float FfxFloat32;
/// A typedef for a 2-dimensional floating point value.
///
/// @ingroup CPU
typedef float FfxFloat32x2[2];
/// A typedef for a 3-dimensional floating point value.
///
/// @ingroup CPU
typedef float FfxFloat32x3[3];
/// A typedef for a 4-dimensional floating point value.
///
/// @ingroup CPU
typedef float FfxFloat32x4[4];
/// A typedef for a 2-dimensional 32bit unsigned integer.
///
/// @ingroup CPU
typedef uint32_t FfxUInt32x2[2];
/// A typedef for a 3-dimensional 32bit unsigned integer.
///
/// @ingroup CPU
typedef uint32_t FfxUInt32x3[3];
/// A typedef for a 4-dimensional 32bit unsigned integer.
///
/// @ingroup CPU
typedef uint32_t FfxUInt32x4[4];
#endif // #if defined(FFX_CPU)
#if defined(FFX_HLSL)
/// A typedef for a boolean value.
///
/// @ingroup GPU
typedef bool FfxBoolean;
#if FFX_HLSL_6_2
typedef float32_t FfxFloat32;
typedef float32_t2 FfxFloat32x2;
typedef float32_t3 FfxFloat32x3;
typedef float32_t4 FfxFloat32x4;
/// A typedef for a unsigned 32bit integer.
///
/// @ingroup GPU
typedef uint32_t FfxUInt32;
typedef uint32_t2 FfxUInt32x2;
typedef uint32_t3 FfxUInt32x3;
typedef uint32_t4 FfxUInt32x4;
typedef int32_t FfxInt32;
typedef int32_t2 FfxInt32x2;
typedef int32_t3 FfxInt32x3;
typedef int32_t4 FfxInt32x4;
#else
#define FfxFloat32 float
#define FfxFloat32x2 float2
#define FfxFloat32x3 float3
#define FfxFloat32x4 float4
/// A typedef for a unsigned 32bit integer.
///
/// @ingroup GPU
typedef uint FfxUInt32;
typedef uint2 FfxUInt32x2;
typedef uint3 FfxUInt32x3;
typedef uint4 FfxUInt32x4;
typedef int FfxInt32;
typedef int2 FfxInt32x2;
typedef int3 FfxInt32x3;
typedef int4 FfxInt32x4;
#endif // #if defined(FFX_HLSL_6_2)
#if FFX_HALF
#if FFX_HLSL_6_2
typedef float16_t FfxFloat16;
typedef float16_t2 FfxFloat16x2;
typedef float16_t3 FfxFloat16x3;
typedef float16_t4 FfxFloat16x4;
/// A typedef for an unsigned 16bit integer.
///
/// @ingroup GPU
typedef uint16_t FfxUInt16;
typedef uint16_t2 FfxUInt16x2;
typedef uint16_t3 FfxUInt16x3;
typedef uint16_t4 FfxUInt16x4;
/// A typedef for a signed 16bit integer.
///
/// @ingroup GPU
typedef int16_t FfxInt16;
typedef int16_t2 FfxInt16x2;
typedef int16_t3 FfxInt16x3;
typedef int16_t4 FfxInt16x4;
#else
typedef min16float FfxFloat16;
typedef min16float2 FfxFloat16x2;
typedef min16float3 FfxFloat16x3;
typedef min16float4 FfxFloat16x4;
/// A typedef for an unsigned 16bit integer.
///
/// @ingroup GPU
typedef min16uint FfxUInt16;
typedef min16uint2 FfxUInt16x2;
typedef min16uint3 FfxUInt16x3;
typedef min16uint4 FfxUInt16x4;
/// A typedef for a signed 16bit integer.
///
/// @ingroup GPU
typedef min16int FfxInt16;
typedef min16int2 FfxInt16x2;
typedef min16int3 FfxInt16x3;
typedef min16int4 FfxInt16x4;
#endif // FFX_HLSL_6_2
#endif // FFX_HALF
#endif // #if defined(FFX_HLSL)
#if defined(FFX_GLSL)
/// A typedef for a boolean value.
///
/// @ingroup GPU
#define FfxBoolean bool
#define FfxFloat32 float
#define FfxFloat32x2 vec2
#define FfxFloat32x3 vec3
#define FfxFloat32x4 vec4
#define FfxUInt32 uint
#define FfxUInt32x2 uvec2
#define FfxUInt32x3 uvec3
#define FfxUInt32x4 uvec4
#define FfxInt32 int
#define FfxInt32x2 ivec2
#define FfxInt32x3 ivec3
#define FfxInt32x4 ivec4
#if FFX_HALF
#define FfxFloat16 float16_t
#define FfxFloat16x2 f16vec2
#define FfxFloat16x3 f16vec3
#define FfxFloat16x4 f16vec4
#define FfxUInt16 uint16_t
#define FfxUInt16x2 u16vec2
#define FfxUInt16x3 u16vec3
#define FfxUInt16x4 u16vec4
#define FfxInt16 int16_t
#define FfxInt16x2 i16vec2
#define FfxInt16x3 i16vec3
#define FfxInt16x4 i16vec4
#endif // FFX_HALF
#endif // #if defined(FFX_GLSL)
// Global toggles:
// #define FFX_HALF (1)
// #define FFX_HLSL_6_2 (1)
#if FFX_HALF
#if FFX_HLSL_6_2
#define FFX_MIN16_SCALAR( TypeName, BaseComponentType ) typedef BaseComponentType##16_t TypeName;
#define FFX_MIN16_VECTOR( TypeName, BaseComponentType, COL ) typedef vector<BaseComponentType##16_t, COL> TypeName;
#define FFX_MIN16_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix<BaseComponentType##16_t, ROW, COL> TypeName;
#define FFX_16BIT_SCALAR( TypeName, BaseComponentType ) typedef BaseComponentType##16_t TypeName;
#define FFX_16BIT_VECTOR( TypeName, BaseComponentType, COL ) typedef vector<BaseComponentType##16_t, COL> TypeName;
#define FFX_16BIT_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix<BaseComponentType##16_t, ROW, COL> TypeName;
#else //FFX_HLSL_6_2
#define FFX_MIN16_SCALAR( TypeName, BaseComponentType ) typedef min16##BaseComponentType TypeName;
#define FFX_MIN16_VECTOR( TypeName, BaseComponentType, COL ) typedef vector<min16##BaseComponentType, COL> TypeName;
#define FFX_MIN16_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix<min16##BaseComponentType, ROW, COL> TypeName;
#define FFX_16BIT_SCALAR( TypeName, BaseComponentType ) FFX_MIN16_SCALAR( TypeName, BaseComponentType );
#define FFX_16BIT_VECTOR( TypeName, BaseComponentType, COL ) FFX_MIN16_VECTOR( TypeName, BaseComponentType, COL );
#define FFX_16BIT_MATRIX( TypeName, BaseComponentType, ROW, COL ) FFX_MIN16_MATRIX( TypeName, BaseComponentType, ROW, COL );
#endif //FFX_HLSL_6_2
#else //FFX_HALF
#define FFX_MIN16_SCALAR( TypeName, BaseComponentType ) typedef BaseComponentType TypeName;
#define FFX_MIN16_VECTOR( TypeName, BaseComponentType, COL ) typedef vector<BaseComponentType, COL> TypeName;
#define FFX_MIN16_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix<BaseComponentType, ROW, COL> TypeName;
#define FFX_16BIT_SCALAR( TypeName, BaseComponentType ) typedef BaseComponentType TypeName;
#define FFX_16BIT_VECTOR( TypeName, BaseComponentType, COL ) typedef vector<BaseComponentType, COL> TypeName;
#define FFX_16BIT_MATRIX( TypeName, BaseComponentType, ROW, COL ) typedef matrix<BaseComponentType, ROW, COL> TypeName;
#endif //FFX_HALF
#if defined(FFX_GPU)
// Common typedefs:
#if defined(FFX_HLSL)
FFX_MIN16_SCALAR( FFX_MIN16_F , float );
FFX_MIN16_VECTOR( FFX_MIN16_F2, float, 2 );
FFX_MIN16_VECTOR( FFX_MIN16_F3, float, 3 );
FFX_MIN16_VECTOR( FFX_MIN16_F4, float, 4 );
FFX_MIN16_SCALAR( FFX_MIN16_I, int );
FFX_MIN16_VECTOR( FFX_MIN16_I2, int, 2 );
FFX_MIN16_VECTOR( FFX_MIN16_I3, int, 3 );
FFX_MIN16_VECTOR( FFX_MIN16_I4, int, 4 );
FFX_MIN16_SCALAR( FFX_MIN16_U, uint );
FFX_MIN16_VECTOR( FFX_MIN16_U2, uint, 2 );
FFX_MIN16_VECTOR( FFX_MIN16_U3, uint, 3 );
FFX_MIN16_VECTOR( FFX_MIN16_U4, uint, 4 );
FFX_16BIT_SCALAR( FFX_F16_t , float );
FFX_16BIT_VECTOR( FFX_F16_t2, float, 2 );
FFX_16BIT_VECTOR( FFX_F16_t3, float, 3 );
FFX_16BIT_VECTOR( FFX_F16_t4, float, 4 );
FFX_16BIT_SCALAR( FFX_I16_t, int );
FFX_16BIT_VECTOR( FFX_I16_t2, int, 2 );
FFX_16BIT_VECTOR( FFX_I16_t3, int, 3 );
FFX_16BIT_VECTOR( FFX_I16_t4, int, 4 );
FFX_16BIT_SCALAR( FFX_U16_t, uint );
FFX_16BIT_VECTOR( FFX_U16_t2, uint, 2 );
FFX_16BIT_VECTOR( FFX_U16_t3, uint, 3 );
FFX_16BIT_VECTOR( FFX_U16_t4, uint, 4 );
#define TYPEDEF_MIN16_TYPES(Prefix) \
typedef FFX_MIN16_F Prefix##_F; \
typedef FFX_MIN16_F2 Prefix##_F2; \
typedef FFX_MIN16_F3 Prefix##_F3; \
typedef FFX_MIN16_F4 Prefix##_F4; \
typedef FFX_MIN16_I Prefix##_I; \
typedef FFX_MIN16_I2 Prefix##_I2; \
typedef FFX_MIN16_I3 Prefix##_I3; \
typedef FFX_MIN16_I4 Prefix##_I4; \
typedef FFX_MIN16_U Prefix##_U; \
typedef FFX_MIN16_U2 Prefix##_U2; \
typedef FFX_MIN16_U3 Prefix##_U3; \
typedef FFX_MIN16_U4 Prefix##_U4;
#define TYPEDEF_16BIT_TYPES(Prefix) \
typedef FFX_16BIT_F Prefix##_F; \
typedef FFX_16BIT_F2 Prefix##_F2; \
typedef FFX_16BIT_F3 Prefix##_F3; \
typedef FFX_16BIT_F4 Prefix##_F4; \
typedef FFX_16BIT_I Prefix##_I; \
typedef FFX_16BIT_I2 Prefix##_I2; \
typedef FFX_16BIT_I3 Prefix##_I3; \
typedef FFX_16BIT_I4 Prefix##_I4; \
typedef FFX_16BIT_U Prefix##_U; \
typedef FFX_16BIT_U2 Prefix##_U2; \
typedef FFX_16BIT_U3 Prefix##_U3; \
typedef FFX_16BIT_U4 Prefix##_U4;
#define TYPEDEF_FULL_PRECISION_TYPES(Prefix) \
typedef FfxFloat32 Prefix##_F; \
typedef FfxFloat32x2 Prefix##_F2; \
typedef FfxFloat32x3 Prefix##_F3; \
typedef FfxFloat32x4 Prefix##_F4; \
typedef FfxInt32 Prefix##_I; \
typedef FfxInt32x2 Prefix##_I2; \
typedef FfxInt32x3 Prefix##_I3; \
typedef FfxInt32x4 Prefix##_I4; \
typedef FfxUInt32 Prefix##_U; \
typedef FfxUInt32x2 Prefix##_U2; \
typedef FfxUInt32x3 Prefix##_U3; \
typedef FfxUInt32x4 Prefix##_U4;
#endif // #if defined(FFX_HLSL)
#if defined(FFX_GLSL)
#if FFX_HALF
#define FFX_MIN16_F float16_t
#define FFX_MIN16_F2 f16vec2
#define FFX_MIN16_F3 f16vec3
#define FFX_MIN16_F4 f16vec4
#define FFX_MIN16_I int16_t
#define FFX_MIN16_I2 i16vec2
#define FFX_MIN16_I3 i16vec3
#define FFX_MIN16_I4 i16vec4
#define FFX_MIN16_U uint16_t
#define FFX_MIN16_U2 u16vec2
#define FFX_MIN16_U3 u16vec3
#define FFX_MIN16_U4 u16vec4
#define FFX_16BIT_F float16_t
#define FFX_16BIT_F2 f16vec2
#define FFX_16BIT_F3 f16vec3
#define FFX_16BIT_F4 f16vec4
#define FFX_16BIT_I int16_t
#define FFX_16BIT_I2 i16vec2
#define FFX_16BIT_I3 i16vec3
#define FFX_16BIT_I4 i16vec4
#define FFX_16BIT_U uint16_t
#define FFX_16BIT_U2 u16vec2
#define FFX_16BIT_U3 u16vec3
#define FFX_16BIT_U4 u16vec4
#else // FFX_HALF
#define FFX_MIN16_F float
#define FFX_MIN16_F2 vec2
#define FFX_MIN16_F3 vec3
#define FFX_MIN16_F4 vec4
#define FFX_MIN16_I int
#define FFX_MIN16_I2 ivec2
#define FFX_MIN16_I3 ivec3
#define FFX_MIN16_I4 ivec4
#define FFX_MIN16_U uint
#define FFX_MIN16_U2 uvec2
#define FFX_MIN16_U3 uvec3
#define FFX_MIN16_U4 uvec4
#define FFX_16BIT_F float
#define FFX_16BIT_F2 vec2
#define FFX_16BIT_F3 vec3
#define FFX_16BIT_F4 vec4
#define FFX_16BIT_I int
#define FFX_16BIT_I2 ivec2
#define FFX_16BIT_I3 ivec3
#define FFX_16BIT_I4 ivec4
#define FFX_16BIT_U uint
#define FFX_16BIT_U2 uvec2
#define FFX_16BIT_U3 uvec3
#define FFX_16BIT_U4 uvec4
#endif // FFX_HALF
#endif // #if defined(FFX_GLSL)
#endif // #if defined(FFX_GPU)
#endif // #ifndef FFX_COMMON_TYPES_H

52
thirdparty/amd-fsr2/shaders/ffx_core.h vendored Normal file
View File

@ -0,0 +1,52 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
/// @defgroup Core
/// @defgroup HLSL
/// @defgroup GLSL
/// @defgroup GPU
/// @defgroup CPU
/// @defgroup CAS
/// @defgroup FSR1
#if !defined(FFX_CORE_H)
#define FFX_CORE_H
#include "ffx_common_types.h"
#if defined(FFX_CPU)
#include "ffx_core_cpu.h"
#endif // #if defined(FFX_CPU)
#if defined(FFX_GLSL) && defined(FFX_GPU)
#include "ffx_core_glsl.h"
#endif // #if defined(FFX_GLSL) && defined(FFX_GPU)
#if defined(FFX_HLSL) && defined(FFX_GPU)
#include "ffx_core_hlsl.h"
#endif // #if defined(FFX_HLSL) && defined(FFX_GPU)
#if defined(FFX_GPU)
#include "ffx_core_gpu_common.h"
#include "ffx_core_gpu_common_half.h"
#include "ffx_core_portability.h"
#endif // #if defined(FFX_GPU)
#endif // #if !defined(FFX_CORE_H)

332
thirdparty/amd-fsr2/shaders/ffx_core_cpu.h vendored Normal file
View File

@ -0,0 +1,332 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
/// A define for a true value in a boolean expression.
///
/// @ingroup CPU
#define FFX_TRUE (1)
/// A define for a false value in a boolean expression.
///
/// @ingroup CPU
#define FFX_FALSE (0)
#if !defined(FFX_STATIC)
/// A define to abstract declaration of static variables and functions.
///
/// @ingroup CPU
#define FFX_STATIC static
#endif // #if !defined(FFX_STATIC)
#ifdef __clang__
#pragma clang diagnostic ignored "-Wunused-variable"
#endif
/// Interpret the bit layout of an IEEE-754 floating point value as an unsigned integer.
///
/// @param [in] x A 32bit floating value.
///
/// @returns
/// An unsigned 32bit integer value containing the bit pattern of <c><i>x</i></c>.
///
/// @ingroup CPU
FFX_STATIC FfxUInt32 ffxAsUInt32(FfxFloat32 x)
{
union
{
FfxFloat32 f;
FfxUInt32 u;
} bits;
bits.f = x;
return bits.u;
}
FFX_STATIC FfxFloat32 ffxDot2(FfxFloat32x2 a, FfxFloat32x2 b)
{
return a[0] * b[0] + a[1] * b[1];
}
FFX_STATIC FfxFloat32 ffxDot3(FfxFloat32x3 a, FfxFloat32x3 b)
{
return a[0] * b[0] + a[1] * b[1] + a[2] * b[2];
}
FFX_STATIC FfxFloat32 ffxDot4(FfxFloat32x4 a, FfxFloat32x4 b)
{
return a[0] * b[0] + a[1] * b[1] + a[2] * b[2] + a[3] * b[3];
}
/// Compute the linear interopation between two values.
///
/// Implemented by calling the GLSL <c><i>mix</i></c> instrinsic function. Implements the
/// following math:
///
/// (1 - t) * x + t * y
///
/// @param [in] x The first value to lerp between.
/// @param [in] y The second value to lerp between.
/// @param [in] t The value to determine how much of <c><i>x</i></c> and how much of <c><i>y</i></c>.
///
/// @returns
/// A linearly interpolated value between <c><i>x</i></c> and <c><i>y</i></c> according to <c><i>t</i></c>.
///
/// @ingroup CPU
FFX_STATIC FfxFloat32 ffxLerp(FfxFloat32 x, FfxFloat32 y, FfxFloat32 t)
{
return y * t + (-x * t + x);
}
/// Compute the reciprocal of a value.
///
/// @param [in] x The value to compute the reciprocal for.
///
/// @returns
/// The reciprocal value of <c><i>x</i></c>.
///
/// @ingroup CPU
FFX_STATIC FfxFloat32 ffxReciprocal(FfxFloat32 a)
{
return 1.0f / a;
}
/// Compute the square root of a value.
///
/// @param [in] x The first value to compute the min of.
///
/// @returns
/// The the square root of <c><i>x</i></c>.
///
/// @ingroup CPU
FFX_STATIC FfxFloat32 ffxSqrt(FfxFloat32 x)
{
return sqrt(x);
}
FFX_STATIC FfxUInt32 AShrSU1(FfxUInt32 a, FfxUInt32 b)
{
return FfxUInt32(FfxInt32(a) >> FfxInt32(b));
}
/// Compute the factional part of a decimal value.
///
/// This function calculates <c><i>x - floor(x)</i></c>.
///
/// @param [in] x The value to compute the fractional part from.
///
/// @returns
/// The fractional part of <c><i>x</i></c>.
///
/// @ingroup CPU
FFX_STATIC FfxFloat32 ffxFract(FfxFloat32 a)
{
return a - floor(a);
}
/// Compute the reciprocal square root of a value.
///
/// @param [in] x The value to compute the reciprocal for.
///
/// @returns
/// The reciprocal square root value of <c><i>x</i></c>.
///
/// @ingroup CPU
FFX_STATIC FfxFloat32 rsqrt(FfxFloat32 a)
{
return ffxReciprocal(ffxSqrt(a));
}
FFX_STATIC FfxFloat32 ffxMin(FfxFloat32 x, FfxFloat32 y)
{
return x < y ? x : y;
}
FFX_STATIC FfxUInt32 ffxMin(FfxUInt32 x, FfxUInt32 y)
{
return x < y ? x : y;
}
FFX_STATIC FfxFloat32 ffxMax(FfxFloat32 x, FfxFloat32 y)
{
return x > y ? x : y;
}
FFX_STATIC FfxUInt32 ffxMax(FfxUInt32 x, FfxUInt32 y)
{
return x > y ? x : y;
}
/// Clamp a value to a [0..1] range.
///
/// @param [in] x The value to clamp to [0..1] range.
///
/// @returns
/// The clamped version of <c><i>x</i></c>.
///
/// @ingroup CPU
FFX_STATIC FfxFloat32 ffxSaturate(FfxFloat32 a)
{
return ffxMin(1.0f, ffxMax(0.0f, a));
}
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
FFX_STATIC void opAAddOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b)
{
d[0] = a[0] + b;
d[1] = a[1] + b;
d[2] = a[2] + b;
return;
}
FFX_STATIC void opACpyF3(FfxFloat32x3 d, FfxFloat32x3 a)
{
d[0] = a[0];
d[1] = a[1];
d[2] = a[2];
return;
}
FFX_STATIC void opAMulF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32x3 b)
{
d[0] = a[0] * b[0];
d[1] = a[1] * b[1];
d[2] = a[2] * b[2];
return;
}
FFX_STATIC void opAMulOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b)
{
d[0] = a[0] * b;
d[1] = a[1] * b;
d[2] = a[2] * b;
return;
}
FFX_STATIC void opARcpF3(FfxFloat32x3 d, FfxFloat32x3 a)
{
d[0] = ffxReciprocal(a[0]);
d[1] = ffxReciprocal(a[1]);
d[2] = ffxReciprocal(a[2]);
return;
}
/// Convert FfxFloat32 to half (in lower 16-bits of output).
///
/// This function implements the same fast technique that is documented here: ftp://ftp.fox-toolkit.org/pub/fasthalffloatconversion.pdf
///
/// The function supports denormals.
///
/// Some conversion rules are to make computations possibly "safer" on the GPU,
/// -INF & -NaN -> -65504
/// +INF & +NaN -> +65504
///
/// @param [in] f The 32bit floating point value to convert.
///
/// @returns
/// The closest 16bit floating point value to <c><i>f</i></c>.
///
/// @ingroup CPU
FFX_STATIC FfxUInt32 f32tof16(FfxFloat32 f)
{
static FfxUInt16 base[512] = {
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, 0x0100, 0x0200, 0x0400,
0x0800, 0x0c00, 0x1000, 0x1400, 0x1800, 0x1c00, 0x2000, 0x2400, 0x2800, 0x2c00, 0x3000, 0x3400, 0x3800, 0x3c00, 0x4000, 0x4400, 0x4800, 0x4c00, 0x5000,
0x5400, 0x5800, 0x5c00, 0x6000, 0x6400, 0x6800, 0x6c00, 0x7000, 0x7400, 0x7800, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff,
0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff,
0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff,
0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff,
0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff,
0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff,
0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x7bff, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8001, 0x8002,
0x8004, 0x8008, 0x8010, 0x8020, 0x8040, 0x8080, 0x8100, 0x8200, 0x8400, 0x8800, 0x8c00, 0x9000, 0x9400, 0x9800, 0x9c00, 0xa000, 0xa400, 0xa800, 0xac00,
0xb000, 0xb400, 0xb800, 0xbc00, 0xc000, 0xc400, 0xc800, 0xcc00, 0xd000, 0xd400, 0xd800, 0xdc00, 0xe000, 0xe400, 0xe800, 0xec00, 0xf000, 0xf400, 0xf800,
0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff,
0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff,
0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff,
0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff,
0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff,
0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff, 0xfbff
};
static FfxUInt8 shift[512] = {
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x17, 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d,
0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x17, 0x16, 0x15, 0x14, 0x13, 0x12, 0x11, 0x10, 0x0f, 0x0e, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d,
0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x0d, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18,
0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18, 0x18
};
union
{
FfxFloat32 f;
FfxUInt32 u;
} bits;
bits.f = f;
FfxUInt32 u = bits.u;
FfxUInt32 i = u >> 23;
return (FfxUInt32)(base[i]) + ((u & 0x7fffff) >> shift[i]);
}
/// Pack 2x32-bit floating point values in a single 32bit value.
///
/// This function first converts each component of <c><i>value</i></c> into their nearest 16-bit floating
/// point representation, and then stores the X and Y components in the lower and upper 16 bits of the
/// 32bit unsigned integer respectively.
///
/// @param [in] value A 2-dimensional floating point value to convert and pack.
///
/// @returns
/// A packed 32bit value containing 2 16bit floating point values.
///
/// @ingroup CPU
FFX_STATIC FfxUInt32 packHalf2x16(FfxFloat32x2 a)
{
return f32tof16(a[0]) + (f32tof16(a[1]) << 16);
}

1669
thirdparty/amd-fsr2/shaders/ffx_core_glsl.h vendored Normal file
View File

File diff suppressed because it is too large Load Diff

2784
thirdparty/amd-fsr2/shaders/ffx_core_gpu_common.h vendored Normal file
View File

File diff suppressed because it is too large Load Diff

2978
thirdparty/amd-fsr2/shaders/ffx_core_gpu_common_half.h vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1502
thirdparty/amd-fsr2/shaders/ffx_core_hlsl.h vendored Normal file
View File

File diff suppressed because it is too large Load Diff

50
thirdparty/amd-fsr2/shaders/ffx_core_portability.h vendored Normal file
View File

@ -0,0 +1,50 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
FfxFloat32x3 opAAddOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b)
{
d = a + ffxBroadcast3(b);
return d;
}
FfxFloat32x3 opACpyF3(FfxFloat32x3 d, FfxFloat32x3 a)
{
d = a;
return d;
}
FfxFloat32x3 opAMulF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32x3 b)
{
d = a * b;
return d;
}
FfxFloat32x3 opAMulOneF3(FfxFloat32x3 d, FfxFloat32x3 a, FfxFloat32 b)
{
d = a * ffxBroadcast3(b);
return d;
}
FfxFloat32x3 opARcpF3(FfxFloat32x3 d, FfxFloat32x3 a)
{
d = rcp(a);
return d;
}

1250
thirdparty/amd-fsr2/shaders/ffx_fsr1.h vendored Normal file
View File

File diff suppressed because it is too large Load Diff

295
thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate.h vendored Normal file
View File

@ -0,0 +1,295 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#ifndef FFX_FSR2_ACCUMULATE_H
#define FFX_FSR2_ACCUMULATE_H
FfxFloat32 GetPxHrVelocity(FfxFloat32x2 fMotionVector)
{
return length(fMotionVector * DisplaySize());
}
#if FFX_HALF
FFX_MIN16_F GetPxHrVelocity(FFX_MIN16_F2 fMotionVector)
{
return length(fMotionVector * FFX_MIN16_F2(DisplaySize()));
}
#endif
void Accumulate(const AccumulationPassCommonParams params, FFX_PARAMETER_INOUT FfxFloat32x3 fHistoryColor, FfxFloat32x3 fAccumulation, FFX_PARAMETER_IN FfxFloat32x4 fUpsampledColorAndWeight)
{
// Aviod invalid values when accumulation and upsampled weight is 0
fAccumulation = ffxMax(FSR2_EPSILON.xxx, fAccumulation + fUpsampledColorAndWeight.www);
#if FFX_FSR2_OPTION_HDR_COLOR_INPUT
//YCoCg -> RGB -> Tonemap -> YCoCg (Use RGB tonemapper to avoid color desaturation)
fUpsampledColorAndWeight.xyz = RGBToYCoCg(Tonemap(YCoCgToRGB(fUpsampledColorAndWeight.xyz)));
fHistoryColor = RGBToYCoCg(Tonemap(YCoCgToRGB(fHistoryColor)));
#endif
const FfxFloat32x3 fAlpha = fUpsampledColorAndWeight.www / fAccumulation;
fHistoryColor = ffxLerp(fHistoryColor, fUpsampledColorAndWeight.xyz, fAlpha);
fHistoryColor = YCoCgToRGB(fHistoryColor);
#if FFX_FSR2_OPTION_HDR_COLOR_INPUT
fHistoryColor = InverseTonemap(fHistoryColor);
#endif
}
void RectifyHistory(
const AccumulationPassCommonParams params,
RectificationBox clippingBox,
FFX_PARAMETER_INOUT FfxFloat32x3 fHistoryColor,
FFX_PARAMETER_INOUT FfxFloat32x3 fAccumulation,
FfxFloat32 fLockContributionThisFrame,
FfxFloat32 fTemporalReactiveFactor,
FfxFloat32 fLumaInstabilityFactor)
{
FfxFloat32 fScaleFactorInfluence = ffxMin(20.0f, ffxPow(FfxFloat32(1.0f / length(DownscaleFactor().x * DownscaleFactor().y)), 3.0f));
const FfxFloat32 fVecolityFactor = ffxSaturate(params.fHrVelocity / 20.0f);
const FfxFloat32 fBoxScaleT = ffxMax(params.fDepthClipFactor, ffxMax(params.fAccumulationMask, fVecolityFactor));
FfxFloat32 fBoxScale = ffxLerp(fScaleFactorInfluence, 1.0f, fBoxScaleT);
FfxFloat32x3 fScaledBoxVec = clippingBox.boxVec * fBoxScale;
FfxFloat32x3 boxMin = clippingBox.boxCenter - fScaledBoxVec;
FfxFloat32x3 boxMax = clippingBox.boxCenter + fScaledBoxVec;
FfxFloat32x3 boxCenter = clippingBox.boxCenter;
FfxFloat32 boxVecSize = length(clippingBox.boxVec);
boxMin = ffxMax(clippingBox.aabbMin, boxMin);
boxMax = ffxMin(clippingBox.aabbMax, boxMax);
if (any(FFX_GREATER_THAN(boxMin, fHistoryColor)) || any(FFX_GREATER_THAN(fHistoryColor, boxMax))) {
const FfxFloat32x3 fClampedHistoryColor = clamp(fHistoryColor, boxMin, boxMax);
FfxFloat32x3 fHistoryContribution = ffxMax(fLumaInstabilityFactor, fLockContributionThisFrame).xxx;
const FfxFloat32 fReactiveFactor = params.fDilatedReactiveFactor;
const FfxFloat32 fReactiveContribution = 1.0f - ffxPow(fReactiveFactor, 1.0f / 2.0f);
fHistoryContribution *= fReactiveContribution;
// Scale history color using rectification info, also using accumulation mask to avoid potential invalid color protection
fHistoryColor = ffxLerp(fClampedHistoryColor, fHistoryColor, ffxSaturate(fHistoryContribution));
// Scale accumulation using rectification info
const FfxFloat32x3 fAccumulationMin = ffxMin(fAccumulation, FFX_BROADCAST_FLOAT32X3(0.1f));
fAccumulation = ffxLerp(fAccumulationMin, fAccumulation, ffxSaturate(fHistoryContribution));
}
}
void WriteUpscaledOutput(FfxInt32x2 iPxHrPos, FfxFloat32x3 fUpscaledColor)
{
StoreUpscaledOutput(iPxHrPos, fUpscaledColor);
}
void FinalizeLockStatus(const AccumulationPassCommonParams params, FfxFloat32x2 fLockStatus, FfxFloat32 fUpsampledWeight)
{
// we expect similar motion for next frame
// kill lock if that location is outside screen, avoid locks to be clamped to screen borders
FfxFloat32x2 fEstimatedUvNextFrame = params.fHrUv - params.fMotionVector;
if (IsUvInside(fEstimatedUvNextFrame) == false) {
KillLock(fLockStatus);
}
else {
// Decrease lock lifetime
const FfxFloat32 fLifetimeDecreaseLanczosMax = FfxFloat32(JitterSequenceLength()) * FfxFloat32(fAverageLanczosWeightPerFrame);
const FfxFloat32 fLifetimeDecrease = FfxFloat32(fUpsampledWeight / fLifetimeDecreaseLanczosMax);
fLockStatus[LOCK_LIFETIME_REMAINING] = ffxMax(FfxFloat32(0), fLockStatus[LOCK_LIFETIME_REMAINING] - fLifetimeDecrease);
}
StoreLockStatus(params.iPxHrPos, fLockStatus);
}
FfxFloat32x3 ComputeBaseAccumulationWeight(const AccumulationPassCommonParams params, FfxFloat32 fThisFrameReactiveFactor, FfxBoolean bInMotionLastFrame, FfxFloat32 fUpsampledWeight, LockState lockState)
{
// Always assume max accumulation was reached
FfxFloat32 fBaseAccumulation = fMaxAccumulationLanczosWeight * FfxFloat32(params.bIsExistingSample) * (1.0f - fThisFrameReactiveFactor) * (1.0f - params.fDepthClipFactor);
fBaseAccumulation = ffxMin(fBaseAccumulation, ffxLerp(fBaseAccumulation, fUpsampledWeight * 10.0f, ffxMax(FfxFloat32(bInMotionLastFrame), ffxSaturate(params.fHrVelocity * FfxFloat32(10)))));
fBaseAccumulation = ffxMin(fBaseAccumulation, ffxLerp(fBaseAccumulation, fUpsampledWeight, ffxSaturate(params.fHrVelocity / FfxFloat32(20))));
return fBaseAccumulation.xxx;
}
FfxFloat32 ComputeLumaInstabilityFactor(const AccumulationPassCommonParams params, RectificationBox clippingBox, FfxFloat32 fThisFrameReactiveFactor, FfxFloat32 fLuminanceDiff)
{
const FfxFloat32 fUnormThreshold = 1.0f / 255.0f;
const FfxInt32 N_MINUS_1 = 0;
const FfxInt32 N_MINUS_2 = 1;
const FfxInt32 N_MINUS_3 = 2;
const FfxInt32 N_MINUS_4 = 3;
FfxFloat32 fCurrentFrameLuma = clippingBox.boxCenter.x;
#if FFX_FSR2_OPTION_HDR_COLOR_INPUT
fCurrentFrameLuma = fCurrentFrameLuma / (1.0f + ffxMax(0.0f, fCurrentFrameLuma));
#endif
fCurrentFrameLuma = round(fCurrentFrameLuma * 255.0f) / 255.0f;
const FfxBoolean bSampleLumaHistory = (ffxMax(ffxMax(params.fDepthClipFactor, params.fAccumulationMask), fLuminanceDiff) < 0.1f) && (params.bIsNewSample == false);
FfxFloat32x4 fCurrentFrameLumaHistory = bSampleLumaHistory ? SampleLumaHistory(params.fReprojectedHrUv) : FFX_BROADCAST_FLOAT32X4(0.0f);
FfxFloat32 fLumaInstability = 0.0f;
FfxFloat32 fDiffs0 = (fCurrentFrameLuma - fCurrentFrameLumaHistory[N_MINUS_1]);
FfxFloat32 fMin = abs(fDiffs0);
if (fMin >= fUnormThreshold)
{
for (int i = N_MINUS_2; i <= N_MINUS_4; i++) {
FfxFloat32 fDiffs1 = (fCurrentFrameLuma - fCurrentFrameLumaHistory[i]);
if (sign(fDiffs0) == sign(fDiffs1)) {
// Scale difference to protect historically similar values
const FfxFloat32 fMinBias = 1.0f;
fMin = ffxMin(fMin, abs(fDiffs1) * fMinBias);
}
}
const FfxFloat32 fBoxSize = clippingBox.boxVec.x;
const FfxFloat32 fBoxSizeFactor = ffxPow(ffxSaturate(fBoxSize / 0.1f), 6.0f);
fLumaInstability = FfxFloat32(fMin != abs(fDiffs0)) * fBoxSizeFactor;
fLumaInstability = FfxFloat32(fLumaInstability > fUnormThreshold);
fLumaInstability *= 1.0f - ffxMax(params.fAccumulationMask, ffxPow(fThisFrameReactiveFactor, 1.0f / 6.0f));
}
//shift history
fCurrentFrameLumaHistory[N_MINUS_4] = fCurrentFrameLumaHistory[N_MINUS_3];
fCurrentFrameLumaHistory[N_MINUS_3] = fCurrentFrameLumaHistory[N_MINUS_2];
fCurrentFrameLumaHistory[N_MINUS_2] = fCurrentFrameLumaHistory[N_MINUS_1];
fCurrentFrameLumaHistory[N_MINUS_1] = fCurrentFrameLuma;
StoreLumaHistory(params.iPxHrPos, fCurrentFrameLumaHistory);
return fLumaInstability * FfxFloat32(fCurrentFrameLumaHistory[N_MINUS_4] != 0);
}
FfxFloat32 ComputeTemporalReactiveFactor(const AccumulationPassCommonParams params, FfxFloat32 fTemporalReactiveFactor)
{
FfxFloat32 fNewFactor = ffxMin(0.99f, fTemporalReactiveFactor);
fNewFactor = ffxMax(fNewFactor, ffxLerp(fNewFactor, 0.4f, ffxSaturate(params.fHrVelocity)));
fNewFactor = ffxMax(fNewFactor * fNewFactor, ffxMax(params.fDepthClipFactor * 0.1f, params.fDilatedReactiveFactor));
// Force reactive factor for new samples
fNewFactor = params.bIsNewSample ? 1.0f : fNewFactor;
if (ffxSaturate(params.fHrVelocity * 10.0f) >= 1.0f) {
fNewFactor = ffxMax(FSR2_EPSILON, fNewFactor) * -1.0f;
}
return fNewFactor;
}
AccumulationPassCommonParams InitParams(FfxInt32x2 iPxHrPos)
{
AccumulationPassCommonParams params;
params.iPxHrPos = iPxHrPos;
const FfxFloat32x2 fHrUv = (iPxHrPos + 0.5f) / DisplaySize();
params.fHrUv = fHrUv;
const FfxFloat32x2 fLrUvJittered = fHrUv + Jitter() / RenderSize();
params.fLrUv_HwSampler = ClampUv(fLrUvJittered, RenderSize(), MaxRenderSize());
params.fMotionVector = GetMotionVector(iPxHrPos, fHrUv);
params.fHrVelocity = GetPxHrVelocity(params.fMotionVector);
ComputeReprojectedUVs(params, params.fReprojectedHrUv, params.bIsExistingSample);
params.fDepthClipFactor = ffxSaturate(SampleDepthClip(params.fLrUv_HwSampler));
const FfxFloat32x2 fDilatedReactiveMasks = SampleDilatedReactiveMasks(params.fLrUv_HwSampler);
params.fDilatedReactiveFactor = fDilatedReactiveMasks.x;
params.fAccumulationMask = fDilatedReactiveMasks.y;
params.bIsResetFrame = (0 == FrameIndex());
params.bIsNewSample = (params.bIsExistingSample == false || params.bIsResetFrame);
return params;
}
void Accumulate(FfxInt32x2 iPxHrPos)
{
const AccumulationPassCommonParams params = InitParams(iPxHrPos);
FfxFloat32x3 fHistoryColor = FfxFloat32x3(0, 0, 0);
FfxFloat32x2 fLockStatus;
InitializeNewLockSample(fLockStatus);
FfxFloat32 fTemporalReactiveFactor = 0.0f;
FfxBoolean bInMotionLastFrame = FFX_FALSE;
LockState lockState = { FFX_FALSE , FFX_FALSE };
if (params.bIsExistingSample && !params.bIsResetFrame) {
ReprojectHistoryColor(params, fHistoryColor, fTemporalReactiveFactor, bInMotionLastFrame);
lockState = ReprojectHistoryLockStatus(params, fLockStatus);
}
FfxFloat32 fThisFrameReactiveFactor = ffxMax(params.fDilatedReactiveFactor, fTemporalReactiveFactor);
FfxFloat32 fLuminanceDiff = 0.0f;
FfxFloat32 fLockContributionThisFrame = 0.0f;
UpdateLockStatus(params, fThisFrameReactiveFactor, lockState, fLockStatus, fLockContributionThisFrame, fLuminanceDiff);
// Load upsampled input color
RectificationBox clippingBox;
FfxFloat32x4 fUpsampledColorAndWeight = ComputeUpsampledColorAndWeight(params, clippingBox, fThisFrameReactiveFactor);
const FfxFloat32 fLumaInstabilityFactor = ComputeLumaInstabilityFactor(params, clippingBox, fThisFrameReactiveFactor, fLuminanceDiff);
FfxFloat32x3 fAccumulation = ComputeBaseAccumulationWeight(params, fThisFrameReactiveFactor, bInMotionLastFrame, fUpsampledColorAndWeight.w, lockState);
if (params.bIsNewSample) {
fHistoryColor = YCoCgToRGB(fUpsampledColorAndWeight.xyz);
}
else {
RectifyHistory(params, clippingBox, fHistoryColor, fAccumulation, fLockContributionThisFrame, fThisFrameReactiveFactor, fLumaInstabilityFactor);
Accumulate(params, fHistoryColor, fAccumulation, fUpsampledColorAndWeight);
}
fHistoryColor = UnprepareRgb(fHistoryColor, Exposure());
FinalizeLockStatus(params, fLockStatus, fUpsampledColorAndWeight.w);
// Get new temporal reactive factor
fTemporalReactiveFactor = ComputeTemporalReactiveFactor(params, fThisFrameReactiveFactor);
StoreInternalColorAndWeight(iPxHrPos, FfxFloat32x4(fHistoryColor, fTemporalReactiveFactor));
// Output final color when RCAS is disabled
#if FFX_FSR2_OPTION_APPLY_SHARPENING == 0
WriteUpscaledOutput(iPxHrPos, fHistoryColor);
#endif
StoreNewLocks(iPxHrPos, 0);
}
#endif // FFX_FSR2_ACCUMULATE_H

92
thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate_pass.glsl vendored Normal file
View File

@ -0,0 +1,92 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#extension GL_GOOGLE_include_directive : require
#extension GL_EXT_samplerless_texture_functions : require
// Needed for rw_upscaled_output declaration
#extension GL_EXT_shader_image_load_formatted : require
#define FSR2_BIND_SRV_INPUT_EXPOSURE 0
#define FSR2_BIND_SRV_DILATED_REACTIVE_MASKS 1
#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
#define FSR2_BIND_SRV_DILATED_MOTION_VECTORS 2
#else
#define FSR2_BIND_SRV_INPUT_MOTION_VECTORS 2
#endif
#define FSR2_BIND_SRV_INTERNAL_UPSCALED 3
#define FSR2_BIND_SRV_LOCK_STATUS 4
#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 6
#define FSR2_BIND_SRV_LUMA_INSTABILITY 7
#define FSR2_BIND_SRV_LANCZOS_LUT 8
#define FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT 9
#define FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS 10
#define FSR2_BIND_SRV_AUTO_EXPOSURE 11
#define FSR2_BIND_SRV_LUMA_HISTORY 12
#define FSR2_BIND_UAV_INTERNAL_UPSCALED 13
#define FSR2_BIND_UAV_LOCK_STATUS 14
#define FSR2_BIND_UAV_UPSCALED_OUTPUT 15
#define FSR2_BIND_UAV_NEW_LOCKS 16
#define FSR2_BIND_UAV_LUMA_HISTORY 17
#define FSR2_BIND_CB_FSR2 18
// -- GODOT start --
#if FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS
#define FSR2_BIND_SRV_INPUT_DEPTH 5
#endif
// -- GODOT end --
#include "ffx_fsr2_callbacks_glsl.h"
#include "ffx_fsr2_common.h"
#include "ffx_fsr2_sample.h"
#include "ffx_fsr2_upsample.h"
#include "ffx_fsr2_postprocess_lock_status.h"
#include "ffx_fsr2_reproject.h"
#include "ffx_fsr2_accumulate.h"
#ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#define FFX_FSR2_THREAD_GROUP_WIDTH 8
#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#define FFX_FSR2_THREAD_GROUP_HEIGHT 8
#endif // FFX_FSR2_THREAD_GROUP_HEIGHT
#ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#define FFX_FSR2_THREAD_GROUP_DEPTH 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#ifndef FFX_FSR2_NUM_THREADS
#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in;
#endif // #ifndef FFX_FSR2_NUM_THREADS
FFX_FSR2_NUM_THREADS
void main()
{
uvec2 uGroupId = gl_WorkGroupID.xy;
const uint GroupRows = (uint(DisplaySize().y) + FFX_FSR2_THREAD_GROUP_HEIGHT - 1) / FFX_FSR2_THREAD_GROUP_HEIGHT;
uGroupId.y = GroupRows - uGroupId.y - 1;
uvec2 uDispatchThreadId = uGroupId * uvec2(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT) + gl_LocalInvocationID.xy;
Accumulate(ivec2(uDispatchThreadId));
}

93
thirdparty/amd-fsr2/shaders/ffx_fsr2_autogen_reactive_pass.glsl vendored Normal file
View File

@ -0,0 +1,93 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#extension GL_GOOGLE_include_directive : require
#extension GL_EXT_samplerless_texture_functions : require
#define FSR2_BIND_SRV_INPUT_OPAQUE_ONLY 0
#define FSR2_BIND_SRV_INPUT_COLOR 1
#define FSR2_BIND_UAV_AUTOREACTIVE 2
#define FSR2_BIND_CB_REACTIVE 3
#define FSR2_BIND_CB_FSR2 4
#include "ffx_fsr2_callbacks_glsl.h"
#include "ffx_fsr2_common.h"
// layout (set = 1, binding = FSR2_BIND_SRV_PRE_ALPHA_COLOR) uniform texture2D r_input_color_pre_alpha;
// layout (set = 1, binding = FSR2_BIND_SRV_POST_ALPHA_COLOR) uniform texture2D r_input_color_post_alpha;
// layout (set = 1, binding = FSR2_BIND_UAV_REACTIVE, r8) uniform image2D rw_output_reactive_mask;
#ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#define FFX_FSR2_THREAD_GROUP_WIDTH 8
#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#define FFX_FSR2_THREAD_GROUP_HEIGHT 8
#endif // FFX_FSR2_THREAD_GROUP_HEIGHT
#ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#define FFX_FSR2_THREAD_GROUP_DEPTH 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#ifndef FFX_FSR2_NUM_THREADS
#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in;
#endif // #ifndef FFX_FSR2_NUM_THREADS
#if defined(FSR2_BIND_CB_REACTIVE)
layout (set = 1, binding = FSR2_BIND_CB_REACTIVE, std140) uniform cbGenerateReactive_t
{
float scale;
float threshold;
float binaryValue;
uint flags;
} cbGenerateReactive;
#endif
FFX_FSR2_NUM_THREADS
void main()
{
FfxUInt32x2 uDispatchThreadId = gl_GlobalInvocationID.xy;
FfxFloat32x3 ColorPreAlpha = LoadOpaqueOnly(FFX_MIN16_I2(uDispatchThreadId)).rgb;
FfxFloat32x3 ColorPostAlpha = LoadInputColor(FFX_MIN16_I2(uDispatchThreadId)).rgb;
if ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_TONEMAP) != 0)
{
ColorPreAlpha = Tonemap(ColorPreAlpha);
ColorPostAlpha = Tonemap(ColorPostAlpha);
}
if ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_INVERSETONEMAP) != 0)
{
ColorPreAlpha = InverseTonemap(ColorPreAlpha);
ColorPostAlpha = InverseTonemap(ColorPostAlpha);
}
FfxFloat32 out_reactive_value = 0.f;
FfxFloat32x3 delta = abs(ColorPostAlpha - ColorPreAlpha);
out_reactive_value = ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_USE_COMPONENTS_MAX)!=0) ? max(delta.x, max(delta.y, delta.z)) : length(delta);
out_reactive_value *= cbGenerateReactive.scale;
out_reactive_value = ((cbGenerateReactive.flags & FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_THRESHOLD)!=0) ? ((out_reactive_value < cbGenerateReactive.threshold) ? 0 : cbGenerateReactive.binaryValue) : out_reactive_value;
imageStore(rw_output_autoreactive, FfxInt32x2(uDispatchThreadId), vec4(out_reactive_value));
}

704
thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_glsl.h vendored Normal file
View File

@ -0,0 +1,704 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#include "ffx_fsr2_resources.h"
#if defined(FFX_GPU)
#include "ffx_core.h"
#endif // #if defined(FFX_GPU)
#if defined(FFX_GPU)
#ifndef FFX_FSR2_PREFER_WAVE64
#define FFX_FSR2_PREFER_WAVE64
#endif // #if defined(FFX_GPU)
#if defined(FSR2_BIND_CB_FSR2)
layout (set = 1, binding = FSR2_BIND_CB_FSR2, std140) uniform cbFSR2_t
{
FfxInt32x2 iRenderSize;
FfxInt32x2 iMaxRenderSize;
FfxInt32x2 iDisplaySize;
FfxInt32x2 iInputColorResourceDimensions;
FfxInt32x2 iLumaMipDimensions;
FfxInt32 iLumaMipLevelToUse;
FfxInt32 iFrameIndex;
FfxFloat32x4 fDeviceToViewDepth;
FfxFloat32x2 fJitter;
FfxFloat32x2 fMotionVectorScale;
FfxFloat32x2 fDownscaleFactor;
FfxFloat32x2 fMotionVectorJitterCancellation;
FfxFloat32 fPreExposure;
FfxFloat32 fPreviousFramePreExposure;
FfxFloat32 fTanHalfFOV;
FfxFloat32 fJitterSequenceLength;
FfxFloat32 fDeltaTime;
FfxFloat32 fDynamicResChangeFactor;
FfxFloat32 fViewSpaceToMetersFactor;
// -- GODOT start --
FfxFloat32 fPad;
mat4 mReprojectionMatrix;
// -- GODOT end --
} cbFSR2;
#endif
FfxInt32x2 RenderSize()
{
return cbFSR2.iRenderSize;
}
FfxInt32x2 MaxRenderSize()
{
return cbFSR2.iMaxRenderSize;
}
FfxInt32x2 DisplaySize()
{
return cbFSR2.iDisplaySize;
}
FfxInt32x2 InputColorResourceDimensions()
{
return cbFSR2.iInputColorResourceDimensions;
}
FfxInt32x2 LumaMipDimensions()
{
return cbFSR2.iLumaMipDimensions;
}
FfxInt32 LumaMipLevelToUse()
{
return cbFSR2.iLumaMipLevelToUse;
}
FfxInt32 FrameIndex()
{
return cbFSR2.iFrameIndex;
}
FfxFloat32x4 DeviceToViewSpaceTransformFactors()
{
return cbFSR2.fDeviceToViewDepth;
}
FfxFloat32x2 Jitter()
{
return cbFSR2.fJitter;
}
FfxFloat32x2 MotionVectorScale()
{
return cbFSR2.fMotionVectorScale;
}
FfxFloat32x2 DownscaleFactor()
{
return cbFSR2.fDownscaleFactor;
}
FfxFloat32x2 MotionVectorJitterCancellation()
{
return cbFSR2.fMotionVectorJitterCancellation;
}
FfxFloat32 PreExposure()
{
return cbFSR2.fPreExposure;
}
FfxFloat32 PreviousFramePreExposure()
{
return cbFSR2.fPreviousFramePreExposure;
}
FfxFloat32 TanHalfFoV()
{
return cbFSR2.fTanHalfFOV;
}
FfxFloat32 JitterSequenceLength()
{
return cbFSR2.fJitterSequenceLength;
}
FfxFloat32 DeltaTime()
{
return cbFSR2.fDeltaTime;
}
FfxFloat32 DynamicResChangeFactor()
{
return cbFSR2.fDynamicResChangeFactor;
}
FfxFloat32 ViewSpaceToMetersFactor()
{
return cbFSR2.fViewSpaceToMetersFactor;
}
layout (set = 0, binding = 0) uniform sampler s_PointClamp;
layout (set = 0, binding = 1) uniform sampler s_LinearClamp;
// SRVs
#if defined(FSR2_BIND_SRV_INPUT_OPAQUE_ONLY)
layout (set = 1, binding = FSR2_BIND_SRV_INPUT_OPAQUE_ONLY) uniform texture2D r_input_opaque_only;
#endif
#if defined(FSR2_BIND_SRV_INPUT_COLOR)
layout (set = 1, binding = FSR2_BIND_SRV_INPUT_COLOR) uniform texture2D r_input_color_jittered;
#endif
#if defined(FSR2_BIND_SRV_INPUT_MOTION_VECTORS)
layout (set = 1, binding = FSR2_BIND_SRV_INPUT_MOTION_VECTORS) uniform texture2D r_input_motion_vectors;
#endif
#if defined(FSR2_BIND_SRV_INPUT_DEPTH)
layout (set = 1, binding = FSR2_BIND_SRV_INPUT_DEPTH) uniform texture2D r_input_depth;
#endif
#if defined(FSR2_BIND_SRV_INPUT_EXPOSURE)
layout (set = 1, binding = FSR2_BIND_SRV_INPUT_EXPOSURE) uniform texture2D r_input_exposure;
#endif
#if defined(FSR2_BIND_SRV_AUTO_EXPOSURE)
layout(set = 1, binding = FSR2_BIND_SRV_AUTO_EXPOSURE) uniform texture2D r_auto_exposure;
#endif
#if defined(FSR2_BIND_SRV_REACTIVE_MASK)
layout (set = 1, binding = FSR2_BIND_SRV_REACTIVE_MASK) uniform texture2D r_reactive_mask;
#endif
#if defined(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK)
layout (set = 1, binding = FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) uniform texture2D r_transparency_and_composition_mask;
#endif
#if defined(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH)
layout (set = 1, binding = FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH) uniform utexture2D r_reconstructed_previous_nearest_depth;
#endif
#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS)
layout (set = 1, binding = FSR2_BIND_SRV_DILATED_MOTION_VECTORS) uniform texture2D r_dilated_motion_vectors;
#endif
#if defined (FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS)
layout(set = 1, binding = FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS) uniform texture2D r_previous_dilated_motion_vectors;
#endif
#if defined(FSR2_BIND_SRV_DILATED_DEPTH)
layout (set = 1, binding = FSR2_BIND_SRV_DILATED_DEPTH) uniform texture2D r_dilatedDepth;
#endif
#if defined(FSR2_BIND_SRV_INTERNAL_UPSCALED)
layout (set = 1, binding = FSR2_BIND_SRV_INTERNAL_UPSCALED) uniform texture2D r_internal_upscaled_color;
#endif
#if defined(FSR2_BIND_SRV_LOCK_STATUS)
layout (set = 1, binding = FSR2_BIND_SRV_LOCK_STATUS) uniform texture2D r_lock_status;
#endif
#if defined(FSR2_BIND_SRV_LOCK_INPUT_LUMA)
layout (set = 1, binding = FSR2_BIND_SRV_LOCK_INPUT_LUMA) uniform texture2D r_lock_input_luma;
#endif
#if defined(FSR2_BIND_SRV_NEW_LOCKS)
layout(set = 1, binding = FSR2_BIND_SRV_NEW_LOCKS) uniform texture2D r_new_locks;
#endif
#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR)
layout (set = 1, binding = FSR2_BIND_SRV_PREPARED_INPUT_COLOR) uniform texture2D r_prepared_input_color;
#endif
#if defined(FSR2_BIND_SRV_LUMA_HISTORY)
layout (set = 1, binding = FSR2_BIND_SRV_LUMA_HISTORY) uniform texture2D r_luma_history;
#endif
#if defined(FSR2_BIND_SRV_RCAS_INPUT)
layout (set = 1, binding = FSR2_BIND_SRV_RCAS_INPUT) uniform texture2D r_rcas_input;
#endif
#if defined(FSR2_BIND_SRV_LANCZOS_LUT)
layout (set = 1, binding = FSR2_BIND_SRV_LANCZOS_LUT) uniform texture2D r_lanczos_lut;
#endif
#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS)
layout (set = 1, binding = FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS) uniform texture2D r_imgMips;
#endif
#if defined(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT)
layout (set = 1, binding = FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT) uniform texture2D r_upsample_maximum_bias_lut;
#endif
#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS)
layout (set = 1, binding = FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) uniform texture2D r_dilated_reactive_masks;
#endif
#if defined(FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR)
layout(set = 1, binding = FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR) uniform texture2D r_input_prev_color_pre_alpha;
#endif
#if defined(FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR)
layout(set = 1, binding = FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR) uniform texture2D r_input_prev_color_post_alpha;
#endif
// UAV
#if defined FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH
layout (set = 1, binding = FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH, r32ui) uniform uimage2D rw_reconstructed_previous_nearest_depth;
#endif
#if defined FSR2_BIND_UAV_DILATED_MOTION_VECTORS
layout (set = 1, binding = FSR2_BIND_UAV_DILATED_MOTION_VECTORS, rg16f) writeonly uniform image2D rw_dilated_motion_vectors;
#endif
#if defined FSR2_BIND_UAV_DILATED_DEPTH
layout (set = 1, binding = FSR2_BIND_UAV_DILATED_DEPTH, r16f) writeonly uniform image2D rw_dilatedDepth;
#endif
#if defined FSR2_BIND_UAV_INTERNAL_UPSCALED
layout (set = 1, binding = FSR2_BIND_UAV_INTERNAL_UPSCALED, rgba16f) writeonly uniform image2D rw_internal_upscaled_color;
#endif
#if defined FSR2_BIND_UAV_LOCK_STATUS
layout (set = 1, binding = FSR2_BIND_UAV_LOCK_STATUS, rg16f) uniform image2D rw_lock_status;
#endif
#if defined(FSR2_BIND_UAV_LOCK_INPUT_LUMA)
layout(set = 1, binding = FSR2_BIND_UAV_LOCK_INPUT_LUMA, r16f) writeonly uniform image2D rw_lock_input_luma;
#endif
#if defined FSR2_BIND_UAV_NEW_LOCKS
layout(set = 1, binding = FSR2_BIND_UAV_NEW_LOCKS, r8) uniform image2D rw_new_locks;
#endif
#if defined FSR2_BIND_UAV_PREPARED_INPUT_COLOR
layout (set = 1, binding = FSR2_BIND_UAV_PREPARED_INPUT_COLOR, rgba16) writeonly uniform image2D rw_prepared_input_color;
#endif
#if defined FSR2_BIND_UAV_LUMA_HISTORY
layout (set = 1, binding = FSR2_BIND_UAV_LUMA_HISTORY, rgba8) uniform image2D rw_luma_history;
#endif
#if defined FSR2_BIND_UAV_UPSCALED_OUTPUT
layout (set = 1, binding = FSR2_BIND_UAV_UPSCALED_OUTPUT /* app controlled format */) writeonly uniform image2D rw_upscaled_output;
#endif
#if defined FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE
layout (set = 1, binding = FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE, r16f) coherent uniform image2D rw_img_mip_shading_change;
#endif
#if defined FSR2_BIND_UAV_EXPOSURE_MIP_5
layout (set = 1, binding = FSR2_BIND_UAV_EXPOSURE_MIP_5, r16f) coherent uniform image2D rw_img_mip_5;
#endif
#if defined FSR2_BIND_UAV_DILATED_REACTIVE_MASKS
layout (set = 1, binding = FSR2_BIND_UAV_DILATED_REACTIVE_MASKS, rg8) writeonly uniform image2D rw_dilated_reactive_masks;
#endif
#if defined FSR2_BIND_UAV_EXPOSURE
layout (set = 1, binding = FSR2_BIND_UAV_EXPOSURE, rg32f) uniform image2D rw_exposure;
#endif
#if defined FSR2_BIND_UAV_AUTO_EXPOSURE
layout(set = 1, binding = FSR2_BIND_UAV_AUTO_EXPOSURE, rg32f) uniform image2D rw_auto_exposure;
#endif
#if defined FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC
layout (set = 1, binding = FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC, r32ui) coherent uniform uimage2D rw_spd_global_atomic;
#endif
#if defined FSR2_BIND_UAV_AUTOREACTIVE
layout(set = 1, binding = FSR2_BIND_UAV_AUTOREACTIVE, r32f) uniform image2D rw_output_autoreactive;
#endif
#if defined FSR2_BIND_UAV_AUTOCOMPOSITION
layout(set = 1, binding = FSR2_BIND_UAV_AUTOCOMPOSITION, r32f) uniform image2D rw_output_autocomposition;
#endif
#if defined FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR
layout(set = 1, binding = FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR, r11f_g11f_b10f) uniform image2D rw_output_prev_color_pre_alpha;
#endif
#if defined FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR
layout(set = 1, binding = FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR, r11f_g11f_b10f) uniform image2D rw_output_prev_color_post_alpha;
#endif
#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS)
FfxFloat32 LoadMipLuma(FfxInt32x2 iPxPos, FfxInt32 mipLevel)
{
return texelFetch(r_imgMips, iPxPos, FfxInt32(mipLevel)).r;
}
#endif
#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS)
FfxFloat32 SampleMipLuma(FfxFloat32x2 fUV, FfxInt32 mipLevel)
{
return textureLod(sampler2D(r_imgMips, s_LinearClamp), fUV, FfxFloat32(mipLevel)).r;
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_DEPTH)
FfxFloat32 LoadInputDepth(FfxInt32x2 iPxPos)
{
return texelFetch(r_input_depth, iPxPos, 0).r;
}
#endif
#if defined(FSR2_BIND_SRV_REACTIVE_MASK)
FfxFloat32 LoadReactiveMask(FfxInt32x2 iPxPos)
{
// -- GODOT start --
#if FFX_FSR2_OPTION_GODOT_REACTIVE_MASK_CLAMP
return min(texelFetch(r_reactive_mask, FfxInt32x2(iPxPos), 0).r, 0.9f);
#else
return texelFetch(r_reactive_mask, FfxInt32x2(iPxPos), 0).r;
#endif
// -- GODOT end --
}
#endif
#if defined(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK)
FfxFloat32 LoadTransparencyAndCompositionMask(FfxUInt32x2 iPxPos)
{
return texelFetch(r_transparency_and_composition_mask, FfxInt32x2(iPxPos), 0).r;
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_COLOR)
FfxFloat32x3 LoadInputColor(FfxInt32x2 iPxPos)
{
return texelFetch(r_input_color_jittered, iPxPos, 0).rgb;
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_COLOR)
FfxFloat32x3 SampleInputColor(FfxFloat32x2 fUV)
{
return textureLod(sampler2D(r_input_color_jittered, s_LinearClamp), fUV, 0.0f).rgb;
}
#endif
#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR)
FfxFloat32x3 LoadPreparedInputColor(FfxInt32x2 iPxPos)
{
return texelFetch(r_prepared_input_color, iPxPos, 0).xyz;
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_MOTION_VECTORS)
FfxFloat32x2 LoadInputMotionVector(FfxInt32x2 iPxDilatedMotionVectorPos)
{
FfxFloat32x2 fSrcMotionVector = texelFetch(r_input_motion_vectors, iPxDilatedMotionVectorPos, 0).xy;
// -- GODOT start --
#if FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS
bool bInvalidMotionVector = all(lessThanEqual(fSrcMotionVector, vec2(-1.0f, -1.0f)));
if (bInvalidMotionVector)
{
FfxFloat32 fSrcDepth = LoadInputDepth(iPxDilatedMotionVectorPos);
FfxFloat32x2 fUv = (iPxDilatedMotionVectorPos + FfxFloat32(0.5)) / RenderSize();
fSrcMotionVector = FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS_FUNCTION(fUv, fSrcDepth, cbFSR2.mReprojectionMatrix);
}
#endif
// -- GODOT end --
FfxFloat32x2 fUvMotionVector = fSrcMotionVector * MotionVectorScale();
#if FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
fUvMotionVector -= MotionVectorJitterCancellation();
#endif
return fUvMotionVector;
}
#endif
#if defined(FSR2_BIND_SRV_INTERNAL_UPSCALED)
FfxFloat32x4 LoadHistory(FfxInt32x2 iPxHistory)
{
return texelFetch(r_internal_upscaled_color, iPxHistory, 0);
}
#endif
#if defined(FSR2_BIND_UAV_LUMA_HISTORY)
void StoreLumaHistory(FfxInt32x2 iPxPos, FfxFloat32x4 fLumaHistory)
{
imageStore(rw_luma_history, FfxInt32x2(iPxPos), fLumaHistory);
}
#endif
#if defined(FSR2_BIND_SRV_LUMA_HISTORY)
FfxFloat32x4 SampleLumaHistory(FfxFloat32x2 fUV)
{
return textureLod(sampler2D(r_luma_history, s_LinearClamp), fUV, 0.0f);
}
#endif
#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED)
void StoreReprojectedHistory(FfxInt32x2 iPxHistory, FfxFloat32x4 fHistory)
{
imageStore(rw_internal_upscaled_color, iPxHistory, fHistory);
}
#endif
#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED)
void StoreInternalColorAndWeight(FfxInt32x2 iPxPos, FfxFloat32x4 fColorAndWeight)
{
imageStore(rw_internal_upscaled_color, FfxInt32x2(iPxPos), fColorAndWeight);
}
#endif
#if defined(FSR2_BIND_UAV_UPSCALED_OUTPUT)
void StoreUpscaledOutput(FfxInt32x2 iPxPos, FfxFloat32x3 fColor)
{
imageStore(rw_upscaled_output, FfxInt32x2(iPxPos), FfxFloat32x4(fColor, 1.f));
}
#endif
#if defined(FSR2_BIND_SRV_LOCK_STATUS)
FfxFloat32x2 LoadLockStatus(FfxInt32x2 iPxPos)
{
FfxFloat32x2 fLockStatus = texelFetch(r_lock_status, iPxPos, 0).rg;
return fLockStatus;
}
#endif
#if defined(FSR2_BIND_UAV_LOCK_STATUS)
void StoreLockStatus(FfxInt32x2 iPxPos, FfxFloat32x2 fLockstatus)
{
imageStore(rw_lock_status, iPxPos, vec4(fLockstatus, 0.0f, 0.0f));
}
#endif
#if defined(FSR2_BIND_SRV_LOCK_INPUT_LUMA)
FfxFloat32 LoadLockInputLuma(FfxInt32x2 iPxPos)
{
return texelFetch(r_lock_input_luma, iPxPos, 0).r;
}
#endif
#if defined(FSR2_BIND_UAV_LOCK_INPUT_LUMA)
void StoreLockInputLuma(FfxInt32x2 iPxPos, FfxFloat32 fLuma)
{
imageStore(rw_lock_input_luma, iPxPos, vec4(fLuma, 0, 0, 0));
}
#endif
#if defined(FSR2_BIND_SRV_NEW_LOCKS)
FfxFloat32 LoadNewLocks(FfxInt32x2 iPxPos)
{
return texelFetch(r_new_locks, iPxPos, 0).r;
}
#endif
#if defined(FSR2_BIND_UAV_NEW_LOCKS)
FfxFloat32 LoadRwNewLocks(FfxInt32x2 iPxPos)
{
return imageLoad(rw_new_locks, iPxPos).r;
}
#endif
#if defined(FSR2_BIND_UAV_NEW_LOCKS)
void StoreNewLocks(FfxInt32x2 iPxPos, FfxFloat32 newLock)
{
imageStore(rw_new_locks, iPxPos, vec4(newLock, 0, 0, 0));
}
#endif
#if defined(FSR2_BIND_UAV_PREPARED_INPUT_COLOR)
void StorePreparedInputColor(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x4 fTonemapped)
{
imageStore(rw_prepared_input_color, iPxPos, fTonemapped);
}
#endif
#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR)
FfxFloat32 SampleDepthClip(FfxFloat32x2 fUV)
{
return textureLod(sampler2D(r_prepared_input_color, s_LinearClamp), fUV, 0.0f).w;
}
#endif
#if defined(FSR2_BIND_SRV_LOCK_STATUS)
FfxFloat32x2 SampleLockStatus(FfxFloat32x2 fUV)
{
FfxFloat32x2 fLockStatus = textureLod(sampler2D(r_lock_status, s_LinearClamp), fUV, 0.0f).rg;
return fLockStatus;
}
#endif
#if defined(FSR2_BIND_SRV_DEPTH)
FfxFloat32 LoadSceneDepth(FfxInt32x2 iPxInput)
{
return texelFetch(r_input_depth, iPxInput, 0).r;
}
#endif
#if defined(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH)
FfxFloat32 LoadReconstructedPrevDepth(FfxInt32x2 iPxPos)
{
return uintBitsToFloat(texelFetch(r_reconstructed_previous_nearest_depth, iPxPos, 0).r);
}
#endif
#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH)
void StoreReconstructedDepth(FfxInt32x2 iPxSample, FfxFloat32 fDepth)
{
FfxUInt32 uDepth = floatBitsToUint(fDepth);
#if FFX_FSR2_OPTION_INVERTED_DEPTH
imageAtomicMax(rw_reconstructed_previous_nearest_depth, iPxSample, uDepth);
#else
imageAtomicMin(rw_reconstructed_previous_nearest_depth, iPxSample, uDepth); // min for standard, max for inverted depth
#endif
}
#endif
#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH)
void SetReconstructedDepth(FfxInt32x2 iPxSample, FfxUInt32 uValue)
{
imageStore(rw_reconstructed_previous_nearest_depth, iPxSample, uvec4(uValue, 0, 0, 0));
}
#endif
#if defined(FSR2_BIND_UAV_DILATED_DEPTH)
void StoreDilatedDepth(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32 fDepth)
{
//FfxUInt32 uDepth = f32tof16(fDepth);
imageStore(rw_dilatedDepth, iPxPos, vec4(fDepth, 0.0f, 0.0f, 0.0f));
}
#endif
#if defined(FSR2_BIND_UAV_DILATED_MOTION_VECTORS)
void StoreDilatedMotionVector(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fMotionVector)
{
imageStore(rw_dilated_motion_vectors, iPxPos, vec4(fMotionVector, 0.0f, 0.0f));
}
#endif
#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS)
FfxFloat32x2 LoadDilatedMotionVector(FfxInt32x2 iPxInput)
{
return texelFetch(r_dilated_motion_vectors, iPxInput, 0).rg;
}
#endif
#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS)
FfxFloat32x2 SampleDilatedMotionVector(FfxFloat32x2 fUV)
{
return textureLod(sampler2D(r_dilated_motion_vectors, s_LinearClamp), fUV, 0.0f).rg;
}
#endif
#if defined(FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS)
FfxFloat32x2 LoadPreviousDilatedMotionVector(FfxInt32x2 iPxInput)
{
return texelFetch(r_previous_dilated_motion_vectors, iPxInput, 0).rg;
}
FfxFloat32x2 SamplePreviousDilatedMotionVector(FfxFloat32x2 fUV)
{
return textureLod(sampler2D(r_previous_dilated_motion_vectors, s_LinearClamp), fUV, 0.0f).xy;
}
#endif
#if defined(FSR2_BIND_SRV_DILATED_DEPTH)
FfxFloat32 LoadDilatedDepth(FfxInt32x2 iPxInput)
{
return texelFetch(r_dilatedDepth, iPxInput, 0).r;
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_EXPOSURE)
FfxFloat32 Exposure()
{
FfxFloat32 exposure = texelFetch(r_input_exposure, FfxInt32x2(0, 0), 0).x;
if (exposure == 0.0f) {
exposure = 1.0f;
}
return exposure;
}
#endif
#if defined(FSR2_BIND_SRV_AUTO_EXPOSURE)
FfxFloat32 AutoExposure()
{
FfxFloat32 exposure = texelFetch(r_auto_exposure, FfxInt32x2(0, 0), 0).x;
if (exposure == 0.0f) {
exposure = 1.0f;
}
return exposure;
}
#endif
FfxFloat32 SampleLanczos2Weight(FfxFloat32 x)
{
#if defined(FSR2_BIND_SRV_LANCZOS_LUT)
return textureLod(sampler2D(r_lanczos_lut, s_LinearClamp), FfxFloat32x2(x / 2.0f, 0.5f), 0.0f).x;
#else
return 0.f;
#endif
}
#if defined(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT)
FfxFloat32 SampleUpsampleMaximumBias(FfxFloat32x2 uv)
{
// Stored as a SNORM, so make sure to multiply by 2 to retrieve the actual expected range.
return FfxFloat32(2.0f) * FfxFloat32(textureLod(sampler2D(r_upsample_maximum_bias_lut, s_LinearClamp), abs(uv) * 2.0f, 0.0f).r);
}
#endif
#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS)
FfxFloat32x2 SampleDilatedReactiveMasks(FfxFloat32x2 fUV)
{
return textureLod(sampler2D(r_dilated_reactive_masks, s_LinearClamp), fUV, 0.0f).rg;
}
#endif
#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS)
FfxFloat32x2 LoadDilatedReactiveMasks(FFX_PARAMETER_IN FfxInt32x2 iPxPos)
{
return texelFetch(r_dilated_reactive_masks, iPxPos, 0).rg;
}
#endif
#if defined(FSR2_BIND_UAV_DILATED_REACTIVE_MASKS)
void StoreDilatedReactiveMasks(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fDilatedReactiveMasks)
{
imageStore(rw_dilated_reactive_masks, iPxPos, vec4(fDilatedReactiveMasks, 0.0f, 0.0f));
}
#endif
#if defined(FFX_INTERNAL)
FfxFloat32x4 SampleDebug(FfxFloat32x2 fUV)
{
return textureLod(sampler2D(r_debug_out, s_LinearClamp), fUV, 0.0f).rgba;
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_OPAQUE_ONLY)
FfxFloat32x3 LoadOpaqueOnly(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos)
{
return texelFetch(r_input_opaque_only, iPxPos, 0).xyz;
}
#endif
#if defined(FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR)
FfxFloat32x3 LoadPrevPreAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos)
{
return texelFetch(r_input_prev_color_pre_alpha, iPxPos, 0).xyz;
}
#endif
#if defined(FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR)
FfxFloat32x3 LoadPrevPostAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos)
{
return texelFetch(r_input_prev_color_post_alpha, iPxPos, 0).xyz;
}
#endif
#if defined(FSR2_BIND_UAV_AUTOREACTIVE)
#if defined(FSR2_BIND_UAV_AUTOCOMPOSITION)
void StoreAutoReactive(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F2 fReactive)
{
imageStore(rw_output_autoreactive, iPxPos, vec4(FfxFloat32(fReactive.x), 0.0f, 0.0f, 0.0f));
imageStore(rw_output_autocomposition, iPxPos, vec4(FfxFloat32(fReactive.y), 0.0f, 0.0f, 0.0f));
}
#endif
#endif
#if defined(FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR)
void StorePrevPreAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F3 color)
{
imageStore(rw_output_prev_color_pre_alpha, iPxPos, vec4(color, 0.0f));
}
#endif
#if defined(FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR)
void StorePrevPostAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F3 color)
{
imageStore(rw_output_prev_color_post_alpha, iPxPos, vec4(color, 0.0f));
}
#endif
#endif // #if defined(FFX_GPU)

799
thirdparty/amd-fsr2/shaders/ffx_fsr2_callbacks_hlsl.h vendored Normal file
View File

@ -0,0 +1,799 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#include "ffx_fsr2_resources.h"
#if defined(FFX_GPU)
#ifdef __hlsl_dx_compiler
#pragma dxc diagnostic push
#pragma dxc diagnostic ignored "-Wambig-lit-shift"
#endif //__hlsl_dx_compiler
#include "ffx_core.h"
#ifdef __hlsl_dx_compiler
#pragma dxc diagnostic pop
#endif //__hlsl_dx_compiler
#endif // #if defined(FFX_GPU)
#if defined(FFX_GPU)
#ifndef FFX_FSR2_PREFER_WAVE64
#define FFX_FSR2_PREFER_WAVE64
#endif // #if defined(FFX_GPU)
#if defined(FFX_GPU)
#pragma warning(disable: 3205) // conversion from larger type to smaller
#endif // #if defined(FFX_GPU)
#define DECLARE_SRV_REGISTER(regIndex) t##regIndex
#define DECLARE_UAV_REGISTER(regIndex) u##regIndex
#define DECLARE_CB_REGISTER(regIndex) b##regIndex
#define FFX_FSR2_DECLARE_SRV(regIndex) register(DECLARE_SRV_REGISTER(regIndex))
#define FFX_FSR2_DECLARE_UAV(regIndex) register(DECLARE_UAV_REGISTER(regIndex))
#define FFX_FSR2_DECLARE_CB(regIndex) register(DECLARE_CB_REGISTER(regIndex))
#if defined(FSR2_BIND_CB_FSR2) || defined(FFX_INTERNAL)
cbuffer cbFSR2 : FFX_FSR2_DECLARE_CB(FSR2_BIND_CB_FSR2)
{
FfxInt32x2 iRenderSize;
FfxInt32x2 iMaxRenderSize;
FfxInt32x2 iDisplaySize;
FfxInt32x2 iInputColorResourceDimensions;
FfxInt32x2 iLumaMipDimensions;
FfxInt32 iLumaMipLevelToUse;
FfxInt32 iFrameIndex;
FfxFloat32x4 fDeviceToViewDepth;
FfxFloat32x2 fJitter;
FfxFloat32x2 fMotionVectorScale;
FfxFloat32x2 fDownscaleFactor;
FfxFloat32x2 fMotionVectorJitterCancellation;
FfxFloat32 fPreExposure;
FfxFloat32 fPreviousFramePreExposure;
FfxFloat32 fTanHalfFOV;
FfxFloat32 fJitterSequenceLength;
FfxFloat32 fDeltaTime;
FfxFloat32 fDynamicResChangeFactor;
FfxFloat32 fViewSpaceToMetersFactor;
};
#define FFX_FSR2_CONSTANT_BUFFER_1_SIZE (sizeof(cbFSR2) / 4) // Number of 32-bit values. This must be kept in sync with the cbFSR2 size.
#endif
#if defined(FFX_GPU)
#define FFX_FSR2_ROOTSIG_STRINGIFY(p) FFX_FSR2_ROOTSIG_STR(p)
#define FFX_FSR2_ROOTSIG_STR(p) #p
#define FFX_FSR2_ROOTSIG [RootSignature( "DescriptorTable(UAV(u0, numDescriptors = " FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_RESOURCE_IDENTIFIER_COUNT) ")), " \
"DescriptorTable(SRV(t0, numDescriptors = " FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_RESOURCE_IDENTIFIER_COUNT) ")), " \
"RootConstants(num32BitConstants=" FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_CONSTANT_BUFFER_1_SIZE) ", b0), " \
"StaticSampler(s0, filter = FILTER_MIN_MAG_MIP_POINT, " \
"addressU = TEXTURE_ADDRESS_CLAMP, " \
"addressV = TEXTURE_ADDRESS_CLAMP, " \
"addressW = TEXTURE_ADDRESS_CLAMP, " \
"comparisonFunc = COMPARISON_NEVER, " \
"borderColor = STATIC_BORDER_COLOR_TRANSPARENT_BLACK), " \
"StaticSampler(s1, filter = FILTER_MIN_MAG_MIP_LINEAR, " \
"addressU = TEXTURE_ADDRESS_CLAMP, " \
"addressV = TEXTURE_ADDRESS_CLAMP, " \
"addressW = TEXTURE_ADDRESS_CLAMP, " \
"comparisonFunc = COMPARISON_NEVER, " \
"borderColor = STATIC_BORDER_COLOR_TRANSPARENT_BLACK)" )]
#define FFX_FSR2_CONSTANT_BUFFER_2_SIZE 6 // Number of 32-bit values. This must be kept in sync with max( cbRCAS , cbSPD) size.
#define FFX_FSR2_CB2_ROOTSIG [RootSignature( "DescriptorTable(UAV(u0, numDescriptors = " FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_RESOURCE_IDENTIFIER_COUNT) ")), " \
"DescriptorTable(SRV(t0, numDescriptors = " FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_RESOURCE_IDENTIFIER_COUNT) ")), " \
"RootConstants(num32BitConstants=" FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_CONSTANT_BUFFER_1_SIZE) ", b0), " \
"RootConstants(num32BitConstants=" FFX_FSR2_ROOTSIG_STRINGIFY(FFX_FSR2_CONSTANT_BUFFER_2_SIZE) ", b1), " \
"StaticSampler(s0, filter = FILTER_MIN_MAG_MIP_POINT, " \
"addressU = TEXTURE_ADDRESS_CLAMP, " \
"addressV = TEXTURE_ADDRESS_CLAMP, " \
"addressW = TEXTURE_ADDRESS_CLAMP, " \
"comparisonFunc = COMPARISON_NEVER, " \
"borderColor = STATIC_BORDER_COLOR_TRANSPARENT_BLACK), " \
"StaticSampler(s1, filter = FILTER_MIN_MAG_MIP_LINEAR, " \
"addressU = TEXTURE_ADDRESS_CLAMP, " \
"addressV = TEXTURE_ADDRESS_CLAMP, " \
"addressW = TEXTURE_ADDRESS_CLAMP, " \
"comparisonFunc = COMPARISON_NEVER, " \
"borderColor = STATIC_BORDER_COLOR_TRANSPARENT_BLACK)" )]
#if defined(FFX_FSR2_EMBED_ROOTSIG)
#define FFX_FSR2_EMBED_ROOTSIG_CONTENT FFX_FSR2_ROOTSIG
#define FFX_FSR2_EMBED_CB2_ROOTSIG_CONTENT FFX_FSR2_CB2_ROOTSIG
#else
#define FFX_FSR2_EMBED_ROOTSIG_CONTENT
#define FFX_FSR2_EMBED_CB2_ROOTSIG_CONTENT
#endif // #if FFX_FSR2_EMBED_ROOTSIG
#endif // #if defined(FFX_GPU)
/* Define getter functions in the order they are defined in the CB! */
FfxInt32x2 RenderSize()
{
return iRenderSize;
}
FfxInt32x2 MaxRenderSize()
{
return iMaxRenderSize;
}
FfxInt32x2 DisplaySize()
{
return iDisplaySize;
}
FfxInt32x2 InputColorResourceDimensions()
{
return iInputColorResourceDimensions;
}
FfxInt32x2 LumaMipDimensions()
{
return iLumaMipDimensions;
}
FfxInt32 LumaMipLevelToUse()
{
return iLumaMipLevelToUse;
}
FfxInt32 FrameIndex()
{
return iFrameIndex;
}
FfxFloat32x2 Jitter()
{
return fJitter;
}
FfxFloat32x4 DeviceToViewSpaceTransformFactors()
{
return fDeviceToViewDepth;
}
FfxFloat32x2 MotionVectorScale()
{
return fMotionVectorScale;
}
FfxFloat32x2 DownscaleFactor()
{
return fDownscaleFactor;
}
FfxFloat32x2 MotionVectorJitterCancellation()
{
return fMotionVectorJitterCancellation;
}
FfxFloat32 PreExposure()
{
return fPreExposure;
}
FfxFloat32 PreviousFramePreExposure()
{
return fPreviousFramePreExposure;
}
FfxFloat32 TanHalfFoV()
{
return fTanHalfFOV;
}
FfxFloat32 JitterSequenceLength()
{
return fJitterSequenceLength;
}
FfxFloat32 DeltaTime()
{
return fDeltaTime;
}
FfxFloat32 DynamicResChangeFactor()
{
return fDynamicResChangeFactor;
}
FfxFloat32 ViewSpaceToMetersFactor()
{
return fViewSpaceToMetersFactor;
}
SamplerState s_PointClamp : register(s0);
SamplerState s_LinearClamp : register(s1);
// SRVs
#if defined(FFX_INTERNAL)
Texture2D<FfxFloat32x4> r_input_opaque_only : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_OPAQUE_ONLY);
Texture2D<FfxFloat32x4> r_input_color_jittered : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR);
Texture2D<FfxFloat32x4> r_input_motion_vectors : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_MOTION_VECTORS);
Texture2D<FfxFloat32> r_input_depth : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_DEPTH);
Texture2D<FfxFloat32x2> r_input_exposure : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_EXPOSURE);
Texture2D<FfxFloat32x2> r_auto_exposure : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE);
Texture2D<FfxFloat32> r_reactive_mask : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK);
Texture2D<FfxFloat32> r_transparency_and_composition_mask : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_TRANSPARENCY_AND_COMPOSITION_MASK);
Texture2D<FfxUInt32> r_reconstructed_previous_nearest_depth : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_RECONSTRUCTED_PREVIOUS_NEAREST_DEPTH);
Texture2D<FfxFloat32x2> r_dilated_motion_vectors : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS);
Texture2D<FfxFloat32x2> r_previous_dilated_motion_vectors : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREVIOUS_DILATED_MOTION_VECTORS);
Texture2D<FfxFloat32> r_dilatedDepth : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_DEPTH);
Texture2D<FfxFloat32x4> r_internal_upscaled_color : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR);
Texture2D<unorm FfxFloat32x2> r_lock_status : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS);
Texture2D<FfxFloat32> r_lock_input_luma : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_INPUT_LUMA);
Texture2D<unorm FfxFloat32> r_new_locks : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_NEW_LOCKS);
Texture2D<FfxFloat32x4> r_prepared_input_color : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR);
Texture2D<FfxFloat32x4> r_luma_history : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY);
Texture2D<FfxFloat32x4> r_rcas_input : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_RCAS_INPUT);
Texture2D<FfxFloat32> r_lanczos_lut : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_LANCZOS_LUT);
Texture2D<FfxFloat32> r_imgMips : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE);
Texture2D<FfxFloat32> r_upsample_maximum_bias_lut : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTITIER_UPSAMPLE_MAXIMUM_BIAS_LUT);
Texture2D<unorm FfxFloat32x2> r_dilated_reactive_masks : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_REACTIVE_MASKS);
Texture2D<float3> r_input_prev_color_pre_alpha : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR);
Texture2D<float3> r_input_prev_color_post_alpha : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR);
Texture2D<FfxFloat32x4> r_debug_out : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_DEBUG_OUTPUT);
// UAV declarations
RWTexture2D<FfxUInt32> rw_reconstructed_previous_nearest_depth : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_RECONSTRUCTED_PREVIOUS_NEAREST_DEPTH);
RWTexture2D<FfxFloat32x2> rw_dilated_motion_vectors : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS);
RWTexture2D<FfxFloat32> rw_dilatedDepth : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_DEPTH);
RWTexture2D<FfxFloat32x4> rw_internal_upscaled_color : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR);
RWTexture2D<unorm FfxFloat32x2> rw_lock_status : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS);
RWTexture2D<FfxFloat32> rw_lock_input_luma : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_INPUT_LUMA);
RWTexture2D<unorm FfxFloat32> rw_new_locks : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_NEW_LOCKS);
RWTexture2D<FfxFloat32x4> rw_prepared_input_color : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR);
RWTexture2D<FfxFloat32x4> rw_luma_history : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY);
RWTexture2D<FfxFloat32x4> rw_upscaled_output : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_UPSCALED_OUTPUT);
globallycoherent RWTexture2D<FfxFloat32> rw_img_mip_shading_change : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_SHADING_CHANGE);
globallycoherent RWTexture2D<FfxFloat32> rw_img_mip_5 : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_5);
RWTexture2D<unorm FfxFloat32x2> rw_dilated_reactive_masks : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_REACTIVE_MASKS);
RWTexture2D<FfxFloat32x2> rw_auto_exposure : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE);
globallycoherent RWTexture2D<FfxUInt32> rw_spd_global_atomic : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_SPD_ATOMIC_COUNT);
RWTexture2D<FfxFloat32x4> rw_debug_out : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_DEBUG_OUTPUT);
RWTexture2D<float> rw_output_autoreactive : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE);
RWTexture2D<float> rw_output_autocomposition : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_AUTOCOMPOSITION);
RWTexture2D<float3> rw_output_prev_color_pre_alpha : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR);
RWTexture2D<float3> rw_output_prev_color_post_alpha : FFX_FSR2_DECLARE_UAV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR);
#else // #if defined(FFX_INTERNAL)
#if defined FSR2_BIND_SRV_INPUT_COLOR
Texture2D<FfxFloat32x4> r_input_color_jittered : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_COLOR);
#endif
#if defined FSR2_BIND_SRV_INPUT_OPAQUE_ONLY
Texture2D<FfxFloat32x4> r_input_opaque_only : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_OPAQUE_ONLY);
#endif
#if defined FSR2_BIND_SRV_INPUT_MOTION_VECTORS
Texture2D<FfxFloat32x4> r_input_motion_vectors : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_MOTION_VECTORS);
#endif
#if defined FSR2_BIND_SRV_INPUT_DEPTH
Texture2D<FfxFloat32> r_input_depth : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_DEPTH);
#endif
#if defined FSR2_BIND_SRV_INPUT_EXPOSURE
Texture2D<FfxFloat32x2> r_input_exposure : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INPUT_EXPOSURE);
#endif
#if defined FSR2_BIND_SRV_AUTO_EXPOSURE
Texture2D<FfxFloat32x2> r_auto_exposure : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_AUTO_EXPOSURE);
#endif
#if defined FSR2_BIND_SRV_REACTIVE_MASK
Texture2D<FfxFloat32> r_reactive_mask : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_REACTIVE_MASK);
#endif
#if defined FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK
Texture2D<FfxFloat32> r_transparency_and_composition_mask : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK);
#endif
#if defined FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH
Texture2D<FfxUInt32> r_reconstructed_previous_nearest_depth : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH);
#endif
#if defined FSR2_BIND_SRV_DILATED_MOTION_VECTORS
Texture2D<FfxFloat32x2> r_dilated_motion_vectors : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_DILATED_MOTION_VECTORS);
#endif
#if defined FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS
Texture2D<FfxFloat32x2> r_previous_dilated_motion_vectors : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS);
#endif
#if defined FSR2_BIND_SRV_DILATED_DEPTH
Texture2D<FfxFloat32> r_dilatedDepth : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_DILATED_DEPTH);
#endif
#if defined FSR2_BIND_SRV_INTERNAL_UPSCALED
Texture2D<FfxFloat32x4> r_internal_upscaled_color : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_INTERNAL_UPSCALED);
#endif
#if defined FSR2_BIND_SRV_LOCK_STATUS
Texture2D<unorm FfxFloat32x2> r_lock_status : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_LOCK_STATUS);
#endif
#if defined FSR2_BIND_SRV_LOCK_INPUT_LUMA
Texture2D<FfxFloat32> r_lock_input_luma : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_LOCK_INPUT_LUMA);
#endif
#if defined FSR2_BIND_SRV_NEW_LOCKS
Texture2D<unorm FfxFloat32> r_new_locks : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_NEW_LOCKS);
#endif
#if defined FSR2_BIND_SRV_PREPARED_INPUT_COLOR
Texture2D<FfxFloat32x4> r_prepared_input_color : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_PREPARED_INPUT_COLOR);
#endif
#if defined FSR2_BIND_SRV_LUMA_HISTORY
Texture2D<unorm FfxFloat32x4> r_luma_history : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_LUMA_HISTORY);
#endif
#if defined FSR2_BIND_SRV_RCAS_INPUT
Texture2D<FfxFloat32x4> r_rcas_input : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_RCAS_INPUT);
#endif
#if defined FSR2_BIND_SRV_LANCZOS_LUT
Texture2D<FfxFloat32> r_lanczos_lut : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_LANCZOS_LUT);
#endif
#if defined FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS
Texture2D<FfxFloat32> r_imgMips : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS);
#endif
#if defined FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT
Texture2D<FfxFloat32> r_upsample_maximum_bias_lut : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT);
#endif
#if defined FSR2_BIND_SRV_DILATED_REACTIVE_MASKS
Texture2D<unorm FfxFloat32x2> r_dilated_reactive_masks : FFX_FSR2_DECLARE_SRV(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS);
#endif
#if defined FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR
Texture2D<float3> r_input_prev_color_pre_alpha : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR);
#endif
#if defined FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR
Texture2D<float3> r_input_prev_color_post_alpha : FFX_FSR2_DECLARE_SRV(FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR);
#endif
// UAV declarations
#if defined FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH
RWTexture2D<FfxUInt32> rw_reconstructed_previous_nearest_depth : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH);
#endif
#if defined FSR2_BIND_UAV_DILATED_MOTION_VECTORS
RWTexture2D<FfxFloat32x2> rw_dilated_motion_vectors : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_DILATED_MOTION_VECTORS);
#endif
#if defined FSR2_BIND_UAV_DILATED_DEPTH
RWTexture2D<FfxFloat32> rw_dilatedDepth : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_DILATED_DEPTH);
#endif
#if defined FSR2_BIND_UAV_INTERNAL_UPSCALED
RWTexture2D<FfxFloat32x4> rw_internal_upscaled_color : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_INTERNAL_UPSCALED);
#endif
#if defined FSR2_BIND_UAV_LOCK_STATUS
RWTexture2D<unorm FfxFloat32x2> rw_lock_status : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_LOCK_STATUS);
#endif
#if defined FSR2_BIND_UAV_LOCK_INPUT_LUMA
RWTexture2D<FfxFloat32> rw_lock_input_luma : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_LOCK_INPUT_LUMA);
#endif
#if defined FSR2_BIND_UAV_NEW_LOCKS
RWTexture2D<unorm FfxFloat32> rw_new_locks : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_NEW_LOCKS);
#endif
#if defined FSR2_BIND_UAV_PREPARED_INPUT_COLOR
RWTexture2D<FfxFloat32x4> rw_prepared_input_color : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_PREPARED_INPUT_COLOR);
#endif
#if defined FSR2_BIND_UAV_LUMA_HISTORY
RWTexture2D<FfxFloat32x4> rw_luma_history : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_LUMA_HISTORY);
#endif
#if defined FSR2_BIND_UAV_UPSCALED_OUTPUT
RWTexture2D<FfxFloat32x4> rw_upscaled_output : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_UPSCALED_OUTPUT);
#endif
#if defined FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE
globallycoherent RWTexture2D<FfxFloat32> rw_img_mip_shading_change : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE);
#endif
#if defined FSR2_BIND_UAV_EXPOSURE_MIP_5
globallycoherent RWTexture2D<FfxFloat32> rw_img_mip_5 : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_EXPOSURE_MIP_5);
#endif
#if defined FSR2_BIND_UAV_DILATED_REACTIVE_MASKS
RWTexture2D<unorm FfxFloat32x2> rw_dilated_reactive_masks : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_DILATED_REACTIVE_MASKS);
#endif
#if defined FSR2_BIND_UAV_EXPOSURE
RWTexture2D<FfxFloat32x2> rw_exposure : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_EXPOSURE);
#endif
#if defined FSR2_BIND_UAV_AUTO_EXPOSURE
RWTexture2D<FfxFloat32x2> rw_auto_exposure : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_AUTO_EXPOSURE);
#endif
#if defined FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC
globallycoherent RWTexture2D<FfxUInt32> rw_spd_global_atomic : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC);
#endif
#if defined FSR2_BIND_UAV_AUTOREACTIVE
RWTexture2D<float> rw_output_autoreactive : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_AUTOREACTIVE);
#endif
#if defined FSR2_BIND_UAV_AUTOCOMPOSITION
RWTexture2D<float> rw_output_autocomposition : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_AUTOCOMPOSITION);
#endif
#if defined FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR
RWTexture2D<float3> rw_output_prev_color_pre_alpha : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR);
#endif
#if defined FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR
RWTexture2D<float3> rw_output_prev_color_post_alpha : FFX_FSR2_DECLARE_UAV(FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR);
#endif
#endif // #if defined(FFX_INTERNAL)
#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS) || defined(FFX_INTERNAL)
FfxFloat32 LoadMipLuma(FfxUInt32x2 iPxPos, FfxUInt32 mipLevel)
{
return r_imgMips.mips[mipLevel][iPxPos];
}
#endif
#if defined(FSR2_BIND_SRV_SCENE_LUMINANCE_MIPS) || defined(FFX_INTERNAL)
FfxFloat32 SampleMipLuma(FfxFloat32x2 fUV, FfxUInt32 mipLevel)
{
return r_imgMips.SampleLevel(s_LinearClamp, fUV, mipLevel);
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_DEPTH) || defined(FFX_INTERNAL)
FfxFloat32 LoadInputDepth(FfxUInt32x2 iPxPos)
{
return r_input_depth[iPxPos];
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_DEPTH) || defined(FFX_INTERNAL)
FfxFloat32 SampleInputDepth(FfxFloat32x2 fUV)
{
return r_input_depth.SampleLevel(s_LinearClamp, fUV, 0).x;
}
#endif
#if defined(FSR2_BIND_SRV_REACTIVE_MASK) || defined(FFX_INTERNAL)
FfxFloat32 LoadReactiveMask(FfxUInt32x2 iPxPos)
{
return r_reactive_mask[iPxPos];
}
#endif
#if defined(FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK) || defined(FFX_INTERNAL)
FfxFloat32 LoadTransparencyAndCompositionMask(FfxUInt32x2 iPxPos)
{
return r_transparency_and_composition_mask[iPxPos];
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_COLOR) || defined(FFX_INTERNAL)
FfxFloat32x3 LoadInputColor(FfxUInt32x2 iPxPos)
{
return r_input_color_jittered[iPxPos].rgb;
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_COLOR) || defined(FFX_INTERNAL)
FfxFloat32x3 SampleInputColor(FfxFloat32x2 fUV)
{
return r_input_color_jittered.SampleLevel(s_LinearClamp, fUV, 0).rgb;
}
#endif
#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) || defined(FFX_INTERNAL)
FfxFloat32x3 LoadPreparedInputColor(FfxUInt32x2 iPxPos)
{
return r_prepared_input_color[iPxPos].xyz;
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_MOTION_VECTORS) || defined(FFX_INTERNAL)
FfxFloat32x2 LoadInputMotionVector(FfxUInt32x2 iPxDilatedMotionVectorPos)
{
FfxFloat32x2 fSrcMotionVector = r_input_motion_vectors[iPxDilatedMotionVectorPos].xy;
FfxFloat32x2 fUvMotionVector = fSrcMotionVector * MotionVectorScale();
#if FFX_FSR2_OPTION_JITTERED_MOTION_VECTORS
fUvMotionVector -= MotionVectorJitterCancellation();
#endif
return fUvMotionVector;
}
#endif
#if defined(FSR2_BIND_SRV_INTERNAL_UPSCALED) || defined(FFX_INTERNAL)
FfxFloat32x4 LoadHistory(FfxUInt32x2 iPxHistory)
{
return r_internal_upscaled_color[iPxHistory];
}
#endif
#if defined(FSR2_BIND_UAV_LUMA_HISTORY) || defined(FFX_INTERNAL)
void StoreLumaHistory(FfxUInt32x2 iPxPos, FfxFloat32x4 fLumaHistory)
{
rw_luma_history[iPxPos] = fLumaHistory;
}
#endif
#if defined(FSR2_BIND_SRV_LUMA_HISTORY) || defined(FFX_INTERNAL)
FfxFloat32x4 SampleLumaHistory(FfxFloat32x2 fUV)
{
return r_luma_history.SampleLevel(s_LinearClamp, fUV, 0);
}
#endif
#if defined(FFX_INTERNAL)
FfxFloat32x4 SampleDebug(FfxFloat32x2 fUV)
{
return r_debug_out.SampleLevel(s_LinearClamp, fUV, 0).w;
}
#endif
#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED) || defined(FFX_INTERNAL)
void StoreReprojectedHistory(FfxUInt32x2 iPxHistory, FfxFloat32x4 fHistory)
{
rw_internal_upscaled_color[iPxHistory] = fHistory;
}
#endif
#if defined(FSR2_BIND_UAV_INTERNAL_UPSCALED) || defined(FFX_INTERNAL)
void StoreInternalColorAndWeight(FfxUInt32x2 iPxPos, FfxFloat32x4 fColorAndWeight)
{
rw_internal_upscaled_color[iPxPos] = fColorAndWeight;
}
#endif
#if defined(FSR2_BIND_UAV_UPSCALED_OUTPUT) || defined(FFX_INTERNAL)
void StoreUpscaledOutput(FfxUInt32x2 iPxPos, FfxFloat32x3 fColor)
{
rw_upscaled_output[iPxPos] = FfxFloat32x4(fColor, 1.f);
}
#endif
//LOCK_LIFETIME_REMAINING == 0
//Should make LockInitialLifetime() return a const 1.0f later
#if defined(FSR2_BIND_SRV_LOCK_STATUS) || defined(FFX_INTERNAL)
FfxFloat32x2 LoadLockStatus(FfxUInt32x2 iPxPos)
{
return r_lock_status[iPxPos];
}
#endif
#if defined(FSR2_BIND_UAV_LOCK_STATUS) || defined(FFX_INTERNAL)
void StoreLockStatus(FfxUInt32x2 iPxPos, FfxFloat32x2 fLockStatus)
{
rw_lock_status[iPxPos] = fLockStatus;
}
#endif
#if defined(FSR2_BIND_SRV_LOCK_INPUT_LUMA) || defined(FFX_INTERNAL)
FfxFloat32 LoadLockInputLuma(FfxUInt32x2 iPxPos)
{
return r_lock_input_luma[iPxPos];
}
#endif
#if defined(FSR2_BIND_UAV_LOCK_INPUT_LUMA) || defined(FFX_INTERNAL)
void StoreLockInputLuma(FfxUInt32x2 iPxPos, FfxFloat32 fLuma)
{
rw_lock_input_luma[iPxPos] = fLuma;
}
#endif
#if defined(FSR2_BIND_SRV_NEW_LOCKS) || defined(FFX_INTERNAL)
FfxFloat32 LoadNewLocks(FfxUInt32x2 iPxPos)
{
return r_new_locks[iPxPos];
}
#endif
#if defined(FSR2_BIND_UAV_NEW_LOCKS) || defined(FFX_INTERNAL)
FfxFloat32 LoadRwNewLocks(FfxUInt32x2 iPxPos)
{
return rw_new_locks[iPxPos];
}
#endif
#if defined(FSR2_BIND_UAV_NEW_LOCKS) || defined(FFX_INTERNAL)
void StoreNewLocks(FfxUInt32x2 iPxPos, FfxFloat32 newLock)
{
rw_new_locks[iPxPos] = newLock;
}
#endif
#if defined(FSR2_BIND_UAV_PREPARED_INPUT_COLOR) || defined(FFX_INTERNAL)
void StorePreparedInputColor(FFX_PARAMETER_IN FfxUInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x4 fTonemapped)
{
rw_prepared_input_color[iPxPos] = fTonemapped;
}
#endif
#if defined(FSR2_BIND_SRV_PREPARED_INPUT_COLOR) || defined(FFX_INTERNAL)
FfxFloat32 SampleDepthClip(FfxFloat32x2 fUV)
{
return r_prepared_input_color.SampleLevel(s_LinearClamp, fUV, 0).w;
}
#endif
#if defined(FSR2_BIND_SRV_LOCK_STATUS) || defined(FFX_INTERNAL)
FfxFloat32x2 SampleLockStatus(FfxFloat32x2 fUV)
{
FfxFloat32x2 fLockStatus = r_lock_status.SampleLevel(s_LinearClamp, fUV, 0);
return fLockStatus;
}
#endif
#if defined(FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH) || defined(FFX_INTERNAL)
FfxFloat32 LoadReconstructedPrevDepth(FfxUInt32x2 iPxPos)
{
return asfloat(r_reconstructed_previous_nearest_depth[iPxPos]);
}
#endif
#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH) || defined(FFX_INTERNAL)
void StoreReconstructedDepth(FfxUInt32x2 iPxSample, FfxFloat32 fDepth)
{
FfxUInt32 uDepth = asuint(fDepth);
#if FFX_FSR2_OPTION_INVERTED_DEPTH
InterlockedMax(rw_reconstructed_previous_nearest_depth[iPxSample], uDepth);
#else
InterlockedMin(rw_reconstructed_previous_nearest_depth[iPxSample], uDepth); // min for standard, max for inverted depth
#endif
}
#endif
#if defined(FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH) || defined(FFX_INTERNAL)
void SetReconstructedDepth(FfxUInt32x2 iPxSample, const FfxUInt32 uValue)
{
rw_reconstructed_previous_nearest_depth[iPxSample] = uValue;
}
#endif
#if defined(FSR2_BIND_UAV_DILATED_DEPTH) || defined(FFX_INTERNAL)
void StoreDilatedDepth(FFX_PARAMETER_IN FfxUInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32 fDepth)
{
rw_dilatedDepth[iPxPos] = fDepth;
}
#endif
#if defined(FSR2_BIND_UAV_DILATED_MOTION_VECTORS) || defined(FFX_INTERNAL)
void StoreDilatedMotionVector(FFX_PARAMETER_IN FfxUInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fMotionVector)
{
rw_dilated_motion_vectors[iPxPos] = fMotionVector;
}
#endif
#if defined(FSR2_BIND_SRV_DILATED_MOTION_VECTORS) || defined(FFX_INTERNAL)
FfxFloat32x2 LoadDilatedMotionVector(FfxUInt32x2 iPxInput)
{
return r_dilated_motion_vectors[iPxInput].xy;
}
#endif
#if defined(FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS) || defined(FFX_INTERNAL)
FfxFloat32x2 LoadPreviousDilatedMotionVector(FfxUInt32x2 iPxInput)
{
return r_previous_dilated_motion_vectors[iPxInput].xy;
}
FfxFloat32x2 SamplePreviousDilatedMotionVector(FfxFloat32x2 uv)
{
return r_previous_dilated_motion_vectors.SampleLevel(s_LinearClamp, uv, 0).xy;
}
#endif
#if defined(FSR2_BIND_SRV_DILATED_DEPTH) || defined(FFX_INTERNAL)
FfxFloat32 LoadDilatedDepth(FfxUInt32x2 iPxInput)
{
return r_dilatedDepth[iPxInput];
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_EXPOSURE) || defined(FFX_INTERNAL)
FfxFloat32 Exposure()
{
FfxFloat32 exposure = r_input_exposure[FfxUInt32x2(0, 0)].x;
if (exposure == 0.0f) {
exposure = 1.0f;
}
return exposure;
}
#endif
#if defined(FSR2_BIND_SRV_AUTO_EXPOSURE) || defined(FFX_INTERNAL)
FfxFloat32 AutoExposure()
{
FfxFloat32 exposure = r_auto_exposure[FfxUInt32x2(0, 0)].x;
if (exposure == 0.0f) {
exposure = 1.0f;
}
return exposure;
}
#endif
FfxFloat32 SampleLanczos2Weight(FfxFloat32 x)
{
#if defined(FSR2_BIND_SRV_LANCZOS_LUT) || defined(FFX_INTERNAL)
return r_lanczos_lut.SampleLevel(s_LinearClamp, FfxFloat32x2(x / 2, 0.5f), 0);
#else
return 0.f;
#endif
}
#if defined(FSR2_BIND_SRV_UPSCALE_MAXIMUM_BIAS_LUT) || defined(FFX_INTERNAL)
FfxFloat32 SampleUpsampleMaximumBias(FfxFloat32x2 uv)
{
// Stored as a SNORM, so make sure to multiply by 2 to retrieve the actual expected range.
return FfxFloat32(2.0) * r_upsample_maximum_bias_lut.SampleLevel(s_LinearClamp, abs(uv) * 2.0, 0);
}
#endif
#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) || defined(FFX_INTERNAL)
FfxFloat32x2 SampleDilatedReactiveMasks(FfxFloat32x2 fUV)
{
return r_dilated_reactive_masks.SampleLevel(s_LinearClamp, fUV, 0);
}
#endif
#if defined(FSR2_BIND_SRV_DILATED_REACTIVE_MASKS) || defined(FFX_INTERNAL)
FfxFloat32x2 LoadDilatedReactiveMasks(FFX_PARAMETER_IN FfxUInt32x2 iPxPos)
{
return r_dilated_reactive_masks[iPxPos];
}
#endif
#if defined(FSR2_BIND_UAV_DILATED_REACTIVE_MASKS) || defined(FFX_INTERNAL)
void StoreDilatedReactiveMasks(FFX_PARAMETER_IN FfxUInt32x2 iPxPos, FFX_PARAMETER_IN FfxFloat32x2 fDilatedReactiveMasks)
{
rw_dilated_reactive_masks[iPxPos] = fDilatedReactiveMasks;
}
#endif
#if defined(FSR2_BIND_SRV_INPUT_OPAQUE_ONLY) || defined(FFX_INTERNAL)
FfxFloat32x3 LoadOpaqueOnly(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos)
{
return r_input_opaque_only[iPxPos].xyz;
}
#endif
#if defined(FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR) || defined(FFX_INTERNAL)
FfxFloat32x3 LoadPrevPreAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos)
{
return r_input_prev_color_pre_alpha[iPxPos];
}
#endif
#if defined(FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR) || defined(FFX_INTERNAL)
FfxFloat32x3 LoadPrevPostAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos)
{
return r_input_prev_color_post_alpha[iPxPos];
}
#endif
#if defined(FSR2_BIND_UAV_AUTOREACTIVE) || defined(FFX_INTERNAL)
#if defined(FSR2_BIND_UAV_AUTOCOMPOSITION) || defined(FFX_INTERNAL)
void StoreAutoReactive(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F2 fReactive)
{
rw_output_autoreactive[iPxPos] = fReactive.x;
rw_output_autocomposition[iPxPos] = fReactive.y;
}
#endif
#endif
#if defined(FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR) || defined(FFX_INTERNAL)
void StorePrevPreAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F3 color)
{
rw_output_prev_color_pre_alpha[iPxPos] = color;
}
#endif
#if defined(FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR) || defined(FFX_INTERNAL)
void StorePrevPostAlpha(FFX_PARAMETER_IN FFX_MIN16_I2 iPxPos, FFX_PARAMETER_IN FFX_MIN16_F3 color)
{
rw_output_prev_color_post_alpha[iPxPos] = color;
}
#endif
#endif // #if defined(FFX_GPU)

565
thirdparty/amd-fsr2/shaders/ffx_fsr2_common.h vendored Normal file
View File

@ -0,0 +1,565 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#if !defined(FFX_FSR2_COMMON_H)
#define FFX_FSR2_COMMON_H
#if defined(FFX_CPU) || defined(FFX_GPU)
//Locks
#define LOCK_LIFETIME_REMAINING 0
#define LOCK_TEMPORAL_LUMA 1
#endif // #if defined(FFX_CPU) || defined(FFX_GPU)
#if defined(FFX_GPU)
FFX_STATIC const FfxFloat32 FSR2_FP16_MIN = 6.10e-05f;
FFX_STATIC const FfxFloat32 FSR2_FP16_MAX = 65504.0f;
FFX_STATIC const FfxFloat32 FSR2_EPSILON = 1e-03f;
FFX_STATIC const FfxFloat32 FSR2_TONEMAP_EPSILON = 1.0f / FSR2_FP16_MAX;
FFX_STATIC const FfxFloat32 FSR2_FLT_MAX = 3.402823466e+38f;
FFX_STATIC const FfxFloat32 FSR2_FLT_MIN = 1.175494351e-38f;
// treat vector truncation warnings as errors
#pragma warning(error: 3206)
// suppress warnings
#pragma warning(disable: 3205) // conversion from larger type to smaller
#pragma warning(disable: 3571) // in ffxPow(f, e), f could be negative
// Reconstructed depth usage
FFX_STATIC const FfxFloat32 fReconstructedDepthBilinearWeightThreshold = 0.01f;
// Accumulation
FFX_STATIC const FfxFloat32 fUpsampleLanczosWeightScale = 1.0f / 12.0f;
FFX_STATIC const FfxFloat32 fMaxAccumulationLanczosWeight = 1.0f;
FFX_STATIC const FfxFloat32 fAverageLanczosWeightPerFrame = 0.74f * fUpsampleLanczosWeightScale; // Average lanczos weight for jitter accumulated samples
FFX_STATIC const FfxFloat32 fAccumulationMaxOnMotion = 3.0f * fUpsampleLanczosWeightScale;
// Auto exposure
FFX_STATIC const FfxFloat32 resetAutoExposureAverageSmoothing = 1e8f;
struct AccumulationPassCommonParams
{
FfxInt32x2 iPxHrPos;
FfxFloat32x2 fHrUv;
FfxFloat32x2 fLrUv_HwSampler;
FfxFloat32x2 fMotionVector;
FfxFloat32x2 fReprojectedHrUv;
FfxFloat32 fHrVelocity;
FfxFloat32 fDepthClipFactor;
FfxFloat32 fDilatedReactiveFactor;
FfxFloat32 fAccumulationMask;
FfxBoolean bIsResetFrame;
FfxBoolean bIsExistingSample;
FfxBoolean bIsNewSample;
};
struct LockState
{
FfxBoolean NewLock; //Set for both unique new and re-locked new
FfxBoolean WasLockedPrevFrame; //Set to identify if the pixel was already locked (relock)
};
void InitializeNewLockSample(FFX_PARAMETER_OUT FfxFloat32x2 fLockStatus)
{
fLockStatus = FfxFloat32x2(0, 0);
}
#if FFX_HALF
void InitializeNewLockSample(FFX_PARAMETER_OUT FFX_MIN16_F2 fLockStatus)
{
fLockStatus = FFX_MIN16_F2(0, 0);
}
#endif
void KillLock(FFX_PARAMETER_INOUT FfxFloat32x2 fLockStatus)
{
fLockStatus[LOCK_LIFETIME_REMAINING] = 0;
}
#if FFX_HALF
void KillLock(FFX_PARAMETER_INOUT FFX_MIN16_F2 fLockStatus)
{
fLockStatus[LOCK_LIFETIME_REMAINING] = FFX_MIN16_F(0);
}
#endif
struct RectificationBox
{
FfxFloat32x3 boxCenter;
FfxFloat32x3 boxVec;
FfxFloat32x3 aabbMin;
FfxFloat32x3 aabbMax;
FfxFloat32 fBoxCenterWeight;
};
#if FFX_HALF
struct RectificationBoxMin16
{
FFX_MIN16_F3 boxCenter;
FFX_MIN16_F3 boxVec;
FFX_MIN16_F3 aabbMin;
FFX_MIN16_F3 aabbMax;
FFX_MIN16_F fBoxCenterWeight;
};
#endif
void RectificationBoxReset(FFX_PARAMETER_INOUT RectificationBox rectificationBox)
{
rectificationBox.fBoxCenterWeight = FfxFloat32(0);
rectificationBox.boxCenter = FfxFloat32x3(0, 0, 0);
rectificationBox.boxVec = FfxFloat32x3(0, 0, 0);
rectificationBox.aabbMin = FfxFloat32x3(FSR2_FLT_MAX, FSR2_FLT_MAX, FSR2_FLT_MAX);
rectificationBox.aabbMax = -FfxFloat32x3(FSR2_FLT_MAX, FSR2_FLT_MAX, FSR2_FLT_MAX);
}
#if FFX_HALF
void RectificationBoxReset(FFX_PARAMETER_INOUT RectificationBoxMin16 rectificationBox)
{
rectificationBox.fBoxCenterWeight = FFX_MIN16_F(0);
rectificationBox.boxCenter = FFX_MIN16_F3(0, 0, 0);
rectificationBox.boxVec = FFX_MIN16_F3(0, 0, 0);
rectificationBox.aabbMin = FFX_MIN16_F3(FSR2_FP16_MAX, FSR2_FP16_MAX, FSR2_FP16_MAX);
rectificationBox.aabbMax = -FFX_MIN16_F3(FSR2_FP16_MAX, FSR2_FP16_MAX, FSR2_FP16_MAX);
}
#endif
void RectificationBoxAddInitialSample(FFX_PARAMETER_INOUT RectificationBox rectificationBox, const FfxFloat32x3 colorSample, const FfxFloat32 fSampleWeight)
{
rectificationBox.aabbMin = colorSample;
rectificationBox.aabbMax = colorSample;
FfxFloat32x3 weightedSample = colorSample * fSampleWeight;
rectificationBox.boxCenter = weightedSample;
rectificationBox.boxVec = colorSample * weightedSample;
rectificationBox.fBoxCenterWeight = fSampleWeight;
}
void RectificationBoxAddSample(FfxBoolean bInitialSample, FFX_PARAMETER_INOUT RectificationBox rectificationBox, const FfxFloat32x3 colorSample, const FfxFloat32 fSampleWeight)
{
if (bInitialSample) {
RectificationBoxAddInitialSample(rectificationBox, colorSample, fSampleWeight);
} else {
rectificationBox.aabbMin = ffxMin(rectificationBox.aabbMin, colorSample);
rectificationBox.aabbMax = ffxMax(rectificationBox.aabbMax, colorSample);
FfxFloat32x3 weightedSample = colorSample * fSampleWeight;
rectificationBox.boxCenter += weightedSample;
rectificationBox.boxVec += colorSample * weightedSample;
rectificationBox.fBoxCenterWeight += fSampleWeight;
}
}
#if FFX_HALF
void RectificationBoxAddInitialSample(FFX_PARAMETER_INOUT RectificationBoxMin16 rectificationBox, const FFX_MIN16_F3 colorSample, const FFX_MIN16_F fSampleWeight)
{
rectificationBox.aabbMin = colorSample;
rectificationBox.aabbMax = colorSample;
FFX_MIN16_F3 weightedSample = colorSample * fSampleWeight;
rectificationBox.boxCenter = weightedSample;
rectificationBox.boxVec = colorSample * weightedSample;
rectificationBox.fBoxCenterWeight = fSampleWeight;
}
void RectificationBoxAddSample(FfxBoolean bInitialSample, FFX_PARAMETER_INOUT RectificationBoxMin16 rectificationBox, const FFX_MIN16_F3 colorSample, const FFX_MIN16_F fSampleWeight)
{
if (bInitialSample) {
RectificationBoxAddInitialSample(rectificationBox, colorSample, fSampleWeight);
} else {
rectificationBox.aabbMin = ffxMin(rectificationBox.aabbMin, colorSample);
rectificationBox.aabbMax = ffxMax(rectificationBox.aabbMax, colorSample);
FFX_MIN16_F3 weightedSample = colorSample * fSampleWeight;
rectificationBox.boxCenter += weightedSample;
rectificationBox.boxVec += colorSample * weightedSample;
rectificationBox.fBoxCenterWeight += fSampleWeight;
}
}
#endif
void RectificationBoxComputeVarianceBoxData(FFX_PARAMETER_INOUT RectificationBox rectificationBox)
{
rectificationBox.fBoxCenterWeight = (abs(rectificationBox.fBoxCenterWeight) > FfxFloat32(FSR2_EPSILON) ? rectificationBox.fBoxCenterWeight : FfxFloat32(1.f));
rectificationBox.boxCenter /= rectificationBox.fBoxCenterWeight;
rectificationBox.boxVec /= rectificationBox.fBoxCenterWeight;
FfxFloat32x3 stdDev = sqrt(abs(rectificationBox.boxVec - rectificationBox.boxCenter * rectificationBox.boxCenter));
rectificationBox.boxVec = stdDev;
}
#if FFX_HALF
void RectificationBoxComputeVarianceBoxData(FFX_PARAMETER_INOUT RectificationBoxMin16 rectificationBox)
{
rectificationBox.fBoxCenterWeight = (abs(rectificationBox.fBoxCenterWeight) > FFX_MIN16_F(FSR2_EPSILON) ? rectificationBox.fBoxCenterWeight : FFX_MIN16_F(1.f));
rectificationBox.boxCenter /= rectificationBox.fBoxCenterWeight;
rectificationBox.boxVec /= rectificationBox.fBoxCenterWeight;
FFX_MIN16_F3 stdDev = sqrt(abs(rectificationBox.boxVec - rectificationBox.boxCenter * rectificationBox.boxCenter));
rectificationBox.boxVec = stdDev;
}
#endif
FfxFloat32x3 SafeRcp3(FfxFloat32x3 v)
{
return (all(FFX_NOT_EQUAL(v, FfxFloat32x3(0, 0, 0)))) ? (FfxFloat32x3(1, 1, 1) / v) : FfxFloat32x3(0, 0, 0);
}
#if FFX_HALF
FFX_MIN16_F3 SafeRcp3(FFX_MIN16_F3 v)
{
return (all(FFX_NOT_EQUAL(v, FFX_MIN16_F3(0, 0, 0)))) ? (FFX_MIN16_F3(1, 1, 1) / v) : FFX_MIN16_F3(0, 0, 0);
}
#endif
FfxFloat32 MinDividedByMax(const FfxFloat32 v0, const FfxFloat32 v1)
{
const FfxFloat32 m = ffxMax(v0, v1);
return m != 0 ? ffxMin(v0, v1) / m : 0;
}
#if FFX_HALF
FFX_MIN16_F MinDividedByMax(const FFX_MIN16_F v0, const FFX_MIN16_F v1)
{
const FFX_MIN16_F m = ffxMax(v0, v1);
return m != FFX_MIN16_F(0) ? ffxMin(v0, v1) / m : FFX_MIN16_F(0);
}
#endif
FfxFloat32x3 YCoCgToRGB(FfxFloat32x3 fYCoCg)
{
FfxFloat32x3 fRgb;
fRgb = FfxFloat32x3(
fYCoCg.x + fYCoCg.y - fYCoCg.z,
fYCoCg.x + fYCoCg.z,
fYCoCg.x - fYCoCg.y - fYCoCg.z);
return fRgb;
}
#if FFX_HALF
FFX_MIN16_F3 YCoCgToRGB(FFX_MIN16_F3 fYCoCg)
{
FFX_MIN16_F3 fRgb;
fRgb = FFX_MIN16_F3(
fYCoCg.x + fYCoCg.y - fYCoCg.z,
fYCoCg.x + fYCoCg.z,
fYCoCg.x - fYCoCg.y - fYCoCg.z);
return fRgb;
}
#endif
FfxFloat32x3 RGBToYCoCg(FfxFloat32x3 fRgb)
{
FfxFloat32x3 fYCoCg;
fYCoCg = FfxFloat32x3(
0.25f * fRgb.r + 0.5f * fRgb.g + 0.25f * fRgb.b,
0.5f * fRgb.r - 0.5f * fRgb.b,
-0.25f * fRgb.r + 0.5f * fRgb.g - 0.25f * fRgb.b);
return fYCoCg;
}
#if FFX_HALF
FFX_MIN16_F3 RGBToYCoCg(FFX_MIN16_F3 fRgb)
{
FFX_MIN16_F3 fYCoCg;
fYCoCg = FFX_MIN16_F3(
0.25 * fRgb.r + 0.5 * fRgb.g + 0.25 * fRgb.b,
0.5 * fRgb.r - 0.5 * fRgb.b,
-0.25 * fRgb.r + 0.5 * fRgb.g - 0.25 * fRgb.b);
return fYCoCg;
}
#endif
FfxFloat32 RGBToLuma(FfxFloat32x3 fLinearRgb)
{
return dot(fLinearRgb, FfxFloat32x3(0.2126f, 0.7152f, 0.0722f));
}
#if FFX_HALF
FFX_MIN16_F RGBToLuma(FFX_MIN16_F3 fLinearRgb)
{
return dot(fLinearRgb, FFX_MIN16_F3(0.2126f, 0.7152f, 0.0722f));
}
#endif
FfxFloat32 RGBToPerceivedLuma(FfxFloat32x3 fLinearRgb)
{
FfxFloat32 fLuminance = RGBToLuma(fLinearRgb);
FfxFloat32 fPercievedLuminance = 0;
if (fLuminance <= 216.0f / 24389.0f) {
fPercievedLuminance = fLuminance * (24389.0f / 27.0f);
}
else {
fPercievedLuminance = ffxPow(fLuminance, 1.0f / 3.0f) * 116.0f - 16.0f;
}
return fPercievedLuminance * 0.01f;
}
#if FFX_HALF
FFX_MIN16_F RGBToPerceivedLuma(FFX_MIN16_F3 fLinearRgb)
{
FFX_MIN16_F fLuminance = RGBToLuma(fLinearRgb);
FFX_MIN16_F fPercievedLuminance = FFX_MIN16_F(0);
if (fLuminance <= FFX_MIN16_F(216.0f / 24389.0f)) {
fPercievedLuminance = fLuminance * FFX_MIN16_F(24389.0f / 27.0f);
}
else {
fPercievedLuminance = ffxPow(fLuminance, FFX_MIN16_F(1.0f / 3.0f)) * FFX_MIN16_F(116.0f) - FFX_MIN16_F(16.0f);
}
return fPercievedLuminance * FFX_MIN16_F(0.01f);
}
#endif
FfxFloat32x3 Tonemap(FfxFloat32x3 fRgb)
{
return fRgb / (ffxMax(ffxMax(0.f, fRgb.r), ffxMax(fRgb.g, fRgb.b)) + 1.f).xxx;
}
FfxFloat32x3 InverseTonemap(FfxFloat32x3 fRgb)
{
return fRgb / ffxMax(FSR2_TONEMAP_EPSILON, 1.f - ffxMax(fRgb.r, ffxMax(fRgb.g, fRgb.b))).xxx;
}
#if FFX_HALF
FFX_MIN16_F3 Tonemap(FFX_MIN16_F3 fRgb)
{
return fRgb / (ffxMax(ffxMax(FFX_MIN16_F(0.f), fRgb.r), ffxMax(fRgb.g, fRgb.b)) + FFX_MIN16_F(1.f)).xxx;
}
FFX_MIN16_F3 InverseTonemap(FFX_MIN16_F3 fRgb)
{
return fRgb / ffxMax(FFX_MIN16_F(FSR2_TONEMAP_EPSILON), FFX_MIN16_F(1.f) - ffxMax(fRgb.r, ffxMax(fRgb.g, fRgb.b))).xxx;
}
#endif
FfxInt32x2 ClampLoad(FfxInt32x2 iPxSample, FfxInt32x2 iPxOffset, FfxInt32x2 iTextureSize)
{
FfxInt32x2 result = iPxSample + iPxOffset;
result.x = (iPxOffset.x < 0) ? ffxMax(result.x, 0) : result.x;
result.x = (iPxOffset.x > 0) ? ffxMin(result.x, iTextureSize.x - 1) : result.x;
result.y = (iPxOffset.y < 0) ? ffxMax(result.y, 0) : result.y;
result.y = (iPxOffset.y > 0) ? ffxMin(result.y, iTextureSize.y - 1) : result.y;
return result;
// return ffxMed3(iPxSample + iPxOffset, FfxInt32x2(0, 0), iTextureSize - FfxInt32x2(1, 1));
}
#if FFX_HALF
FFX_MIN16_I2 ClampLoad(FFX_MIN16_I2 iPxSample, FFX_MIN16_I2 iPxOffset, FFX_MIN16_I2 iTextureSize)
{
FFX_MIN16_I2 result = iPxSample + iPxOffset;
result.x = (iPxOffset.x < 0) ? ffxMax(result.x, FFX_MIN16_I(0)) : result.x;
result.x = (iPxOffset.x > 0) ? ffxMin(result.x, iTextureSize.x - FFX_MIN16_I(1)) : result.x;
result.y = (iPxOffset.y < 0) ? ffxMax(result.y, FFX_MIN16_I(0)) : result.y;
result.y = (iPxOffset.y > 0) ? ffxMin(result.y, iTextureSize.y - FFX_MIN16_I(1)) : result.y;
return result;
// return ffxMed3Half(iPxSample + iPxOffset, FFX_MIN16_I2(0, 0), iTextureSize - FFX_MIN16_I2(1, 1));
}
#endif
FfxFloat32x2 ClampUv(FfxFloat32x2 fUv, FfxInt32x2 iTextureSize, FfxInt32x2 iResourceSize)
{
const FfxFloat32x2 fSampleLocation = fUv * iTextureSize;
const FfxFloat32x2 fClampedLocation = ffxMax(FfxFloat32x2(0.5f, 0.5f), ffxMin(fSampleLocation, FfxFloat32x2(iTextureSize) - FfxFloat32x2(0.5f, 0.5f)));
const FfxFloat32x2 fClampedUv = fClampedLocation / FfxFloat32x2(iResourceSize);
return fClampedUv;
}
FfxBoolean IsOnScreen(FfxInt32x2 pos, FfxInt32x2 size)
{
return all(FFX_LESS_THAN(FfxUInt32x2(pos), FfxUInt32x2(size)));
}
#if FFX_HALF
FfxBoolean IsOnScreen(FFX_MIN16_I2 pos, FFX_MIN16_I2 size)
{
return all(FFX_LESS_THAN(FFX_MIN16_U2(pos), FFX_MIN16_U2(size)));
}
#endif
FfxFloat32 ComputeAutoExposureFromLavg(FfxFloat32 Lavg)
{
Lavg = exp(Lavg);
const FfxFloat32 S = 100.0f; //ISO arithmetic speed
const FfxFloat32 K = 12.5f;
FfxFloat32 ExposureISO100 = log2((Lavg * S) / K);
const FfxFloat32 q = 0.65f;
FfxFloat32 Lmax = (78.0f / (q * S)) * ffxPow(2.0f, ExposureISO100);
return 1 / Lmax;
}
#if FFX_HALF
FFX_MIN16_F ComputeAutoExposureFromLavg(FFX_MIN16_F Lavg)
{
Lavg = exp(Lavg);
const FFX_MIN16_F S = FFX_MIN16_F(100.0f); //ISO arithmetic speed
const FFX_MIN16_F K = FFX_MIN16_F(12.5f);
const FFX_MIN16_F ExposureISO100 = log2((Lavg * S) / K);
const FFX_MIN16_F q = FFX_MIN16_F(0.65f);
const FFX_MIN16_F Lmax = (FFX_MIN16_F(78.0f) / (q * S)) * ffxPow(FFX_MIN16_F(2.0f), ExposureISO100);
return FFX_MIN16_F(1) / Lmax;
}
#endif
FfxInt32x2 ComputeHrPosFromLrPos(FfxInt32x2 iPxLrPos)
{
FfxFloat32x2 fSrcJitteredPos = FfxFloat32x2(iPxLrPos) + 0.5f - Jitter();
FfxFloat32x2 fLrPosInHr = (fSrcJitteredPos / RenderSize()) * DisplaySize();
FfxInt32x2 iPxHrPos = FfxInt32x2(floor(fLrPosInHr));
return iPxHrPos;
}
#if FFX_HALF
FFX_MIN16_I2 ComputeHrPosFromLrPos(FFX_MIN16_I2 iPxLrPos)
{
FFX_MIN16_F2 fSrcJitteredPos = FFX_MIN16_F2(iPxLrPos) + FFX_MIN16_F(0.5f) - FFX_MIN16_F2(Jitter());
FFX_MIN16_F2 fLrPosInHr = (fSrcJitteredPos / FFX_MIN16_F2(RenderSize())) * FFX_MIN16_F2(DisplaySize());
FFX_MIN16_I2 iPxHrPos = FFX_MIN16_I2(floor(fLrPosInHr));
return iPxHrPos;
}
#endif
FfxFloat32x2 ComputeNdc(FfxFloat32x2 fPxPos, FfxInt32x2 iSize)
{
return fPxPos / FfxFloat32x2(iSize) * FfxFloat32x2(2.0f, -2.0f) + FfxFloat32x2(-1.0f, 1.0f);
}
FfxFloat32 GetViewSpaceDepth(FfxFloat32 fDeviceDepth)
{
const FfxFloat32x4 fDeviceToViewDepth = DeviceToViewSpaceTransformFactors();
// fDeviceToViewDepth details found in ffx_fsr2.cpp
return (fDeviceToViewDepth[1] / (fDeviceDepth - fDeviceToViewDepth[0]));
}
FfxFloat32 GetViewSpaceDepthInMeters(FfxFloat32 fDeviceDepth)
{
return GetViewSpaceDepth(fDeviceDepth) * ViewSpaceToMetersFactor();
}
FfxFloat32x3 GetViewSpacePosition(FfxInt32x2 iViewportPos, FfxInt32x2 iViewportSize, FfxFloat32 fDeviceDepth)
{
const FfxFloat32x4 fDeviceToViewDepth = DeviceToViewSpaceTransformFactors();
const FfxFloat32 Z = GetViewSpaceDepth(fDeviceDepth);
const FfxFloat32x2 fNdcPos = ComputeNdc(iViewportPos, iViewportSize);
const FfxFloat32 X = fDeviceToViewDepth[2] * fNdcPos.x * Z;
const FfxFloat32 Y = fDeviceToViewDepth[3] * fNdcPos.y * Z;
return FfxFloat32x3(X, Y, Z);
}
FfxFloat32x3 GetViewSpacePositionInMeters(FfxInt32x2 iViewportPos, FfxInt32x2 iViewportSize, FfxFloat32 fDeviceDepth)
{
return GetViewSpacePosition(iViewportPos, iViewportSize, fDeviceDepth) * ViewSpaceToMetersFactor();
}
FfxFloat32 GetMaxDistanceInMeters()
{
#if FFX_FSR2_OPTION_INVERTED_DEPTH
return GetViewSpaceDepth(0.0f) * ViewSpaceToMetersFactor();
#else
return GetViewSpaceDepth(1.0f) * ViewSpaceToMetersFactor();
#endif
}
FfxFloat32x3 PrepareRgb(FfxFloat32x3 fRgb, FfxFloat32 fExposure, FfxFloat32 fPreExposure)
{
fRgb /= fPreExposure;
fRgb *= fExposure;
fRgb = clamp(fRgb, 0.0f, FSR2_FP16_MAX);
return fRgb;
}
FfxFloat32x3 UnprepareRgb(FfxFloat32x3 fRgb, FfxFloat32 fExposure)
{
fRgb /= fExposure;
fRgb *= PreExposure();
return fRgb;
}
struct BilinearSamplingData
{
FfxInt32x2 iOffsets[4];
FfxFloat32 fWeights[4];
FfxInt32x2 iBasePos;
};
BilinearSamplingData GetBilinearSamplingData(FfxFloat32x2 fUv, FfxInt32x2 iSize)
{
BilinearSamplingData data;
FfxFloat32x2 fPxSample = (fUv * iSize) - FfxFloat32x2(0.5f, 0.5f);
data.iBasePos = FfxInt32x2(floor(fPxSample));
FfxFloat32x2 fPxFrac = ffxFract(fPxSample);
data.iOffsets[0] = FfxInt32x2(0, 0);
data.iOffsets[1] = FfxInt32x2(1, 0);
data.iOffsets[2] = FfxInt32x2(0, 1);
data.iOffsets[3] = FfxInt32x2(1, 1);
data.fWeights[0] = (1 - fPxFrac.x) * (1 - fPxFrac.y);
data.fWeights[1] = (fPxFrac.x) * (1 - fPxFrac.y);
data.fWeights[2] = (1 - fPxFrac.x) * (fPxFrac.y);
data.fWeights[3] = (fPxFrac.x) * (fPxFrac.y);
return data;
}
struct PlaneData
{
FfxFloat32x3 fNormal;
FfxFloat32 fDistanceFromOrigin;
};
PlaneData GetPlaneFromPoints(FfxFloat32x3 fP0, FfxFloat32x3 fP1, FfxFloat32x3 fP2)
{
PlaneData plane;
FfxFloat32x3 v0 = fP0 - fP1;
FfxFloat32x3 v1 = fP0 - fP2;
plane.fNormal = normalize(cross(v0, v1));
plane.fDistanceFromOrigin = -dot(fP0, plane.fNormal);
return plane;
}
FfxFloat32 PointToPlaneDistance(PlaneData plane, FfxFloat32x3 fPoint)
{
return abs(dot(plane.fNormal, fPoint) + plane.fDistanceFromOrigin);
}
#endif // #if defined(FFX_GPU)
#endif //!defined(FFX_FSR2_COMMON_H)

189
thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid.h vendored Normal file
View File

@ -0,0 +1,189 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
FFX_GROUPSHARED FfxUInt32 spdCounter;
#ifndef SPD_PACKED_ONLY
FFX_GROUPSHARED FfxFloat32 spdIntermediateR[16][16];
FFX_GROUPSHARED FfxFloat32 spdIntermediateG[16][16];
FFX_GROUPSHARED FfxFloat32 spdIntermediateB[16][16];
FFX_GROUPSHARED FfxFloat32 spdIntermediateA[16][16];
FfxFloat32x4 SpdLoadSourceImage(FfxFloat32x2 tex, FfxUInt32 slice)
{
FfxFloat32x2 fUv = (tex + 0.5f + Jitter()) / RenderSize();
fUv = ClampUv(fUv, RenderSize(), InputColorResourceDimensions());
FfxFloat32x3 fRgb = SampleInputColor(fUv);
fRgb /= PreExposure();
//compute log luma
const FfxFloat32 fLogLuma = log(ffxMax(FSR2_EPSILON, RGBToLuma(fRgb)));
// Make sure out of screen pixels contribute no value to the end result
const FfxFloat32 result = all(FFX_LESS_THAN(tex, RenderSize())) ? fLogLuma : 0.0f;
return FfxFloat32x4(result, 0, 0, 0);
}
FfxFloat32x4 SpdLoad(FfxInt32x2 tex, FfxUInt32 slice)
{
return SPD_LoadMipmap5(tex);
}
void SpdStore(FfxInt32x2 pix, FfxFloat32x4 outValue, FfxUInt32 index, FfxUInt32 slice)
{
if (index == LumaMipLevelToUse() || index == 5)
{
SPD_SetMipmap(pix, index, outValue.r);
}
if (index == MipCount() - 1) { //accumulate on 1x1 level
if (all(FFX_EQUAL(pix, FfxInt32x2(0, 0))))
{
FfxFloat32 prev = SPD_LoadExposureBuffer().y;
FfxFloat32 result = outValue.r;
if (prev < resetAutoExposureAverageSmoothing) // Compare Lavg, so small or negative values
{
FfxFloat32 rate = 1.0f;
result = prev + (result - prev) * (1 - exp(-DeltaTime() * rate));
}
FfxFloat32x2 spdOutput = FfxFloat32x2(ComputeAutoExposureFromLavg(result), result);
SPD_SetExposureBuffer(spdOutput);
}
}
}
void SpdIncreaseAtomicCounter(FfxUInt32 slice)
{
SPD_IncreaseAtomicCounter(spdCounter);
}
FfxUInt32 SpdGetAtomicCounter()
{
return spdCounter;
}
void SpdResetAtomicCounter(FfxUInt32 slice)
{
SPD_ResetAtomicCounter();
}
FfxFloat32x4 SpdLoadIntermediate(FfxUInt32 x, FfxUInt32 y)
{
return FfxFloat32x4(
spdIntermediateR[x][y],
spdIntermediateG[x][y],
spdIntermediateB[x][y],
spdIntermediateA[x][y]);
}
void SpdStoreIntermediate(FfxUInt32 x, FfxUInt32 y, FfxFloat32x4 value)
{
spdIntermediateR[x][y] = value.x;
spdIntermediateG[x][y] = value.y;
spdIntermediateB[x][y] = value.z;
spdIntermediateA[x][y] = value.w;
}
FfxFloat32x4 SpdReduce4(FfxFloat32x4 v0, FfxFloat32x4 v1, FfxFloat32x4 v2, FfxFloat32x4 v3)
{
return (v0 + v1 + v2 + v3) * 0.25f;
}
#endif
// define fetch and store functions Packed
#if FFX_HALF
#error Callback must be implemented
FFX_GROUPSHARED FfxFloat16x2 spdIntermediateRG[16][16];
FFX_GROUPSHARED FfxFloat16x2 spdIntermediateBA[16][16];
FfxFloat16x4 SpdLoadSourceImageH(FfxFloat32x2 tex, FfxUInt32 slice)
{
return FfxFloat16x4(imgDst[0][FfxFloat32x3(tex, slice)]);
}
FfxFloat16x4 SpdLoadH(FfxInt32x2 p, FfxUInt32 slice)
{
return FfxFloat16x4(imgDst6[FfxUInt32x3(p, slice)]);
}
void SpdStoreH(FfxInt32x2 p, FfxFloat16x4 value, FfxUInt32 mip, FfxUInt32 slice)
{
if (index == LumaMipLevelToUse() || index == 5)
{
imgDst6[FfxUInt32x3(p, slice)] = FfxFloat32x4(value);
return;
}
imgDst[mip + 1][FfxUInt32x3(p, slice)] = FfxFloat32x4(value);
}
void SpdIncreaseAtomicCounter(FfxUInt32 slice)
{
InterlockedAdd(rw_spd_global_atomic[FfxInt16x2(0, 0)].counter[slice], 1, spdCounter);
}
FfxUInt32 SpdGetAtomicCounter()
{
return spdCounter;
}
void SpdResetAtomicCounter(FfxUInt32 slice)
{
rw_spd_global_atomic[FfxInt16x2(0, 0)].counter[slice] = 0;
}
FfxFloat16x4 SpdLoadIntermediateH(FfxUInt32 x, FfxUInt32 y)
{
return FfxFloat16x4(
spdIntermediateRG[x][y].x,
spdIntermediateRG[x][y].y,
spdIntermediateBA[x][y].x,
spdIntermediateBA[x][y].y);
}
void SpdStoreIntermediateH(FfxUInt32 x, FfxUInt32 y, FfxFloat16x4 value)
{
spdIntermediateRG[x][y] = value.xy;
spdIntermediateBA[x][y] = value.zw;
}
FfxFloat16x4 SpdReduce4H(FfxFloat16x4 v0, FfxFloat16x4 v1, FfxFloat16x4 v2, FfxFloat16x4 v3)
{
return (v0 + v1 + v2 + v3) * FfxFloat16(0.25);
}
#endif
#include "ffx_spd.h"
void ComputeAutoExposure(FfxUInt32x3 WorkGroupId, FfxUInt32 LocalThreadIndex)
{
#if FFX_HALF
SpdDownsampleH(
FfxUInt32x2(WorkGroupId.xy),
FfxUInt32(LocalThreadIndex),
FfxUInt32(MipCount()),
FfxUInt32(NumWorkGroups()),
FfxUInt32(WorkGroupId.z),
FfxUInt32x2(WorkGroupOffset()));
#else
SpdDownsample(
FfxUInt32x2(WorkGroupId.xy),
FfxUInt32(LocalThreadIndex),
FfxUInt32(MipCount()),
FfxUInt32(NumWorkGroups()),
FfxUInt32(WorkGroupId.z),
FfxUInt32x2(WorkGroupOffset()));
#endif
}

134
thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl vendored Normal file
View File

@ -0,0 +1,134 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#extension GL_GOOGLE_include_directive : require
#extension GL_EXT_samplerless_texture_functions : require
#define FSR2_BIND_SRV_INPUT_COLOR 0
#define FSR2_BIND_UAV_SPD_GLOBAL_ATOMIC 1
#define FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE 2
#define FSR2_BIND_UAV_EXPOSURE_MIP_5 3
#define FSR2_BIND_UAV_AUTO_EXPOSURE 4
#define FSR2_BIND_CB_FSR2 5
#define FSR2_BIND_CB_SPD 6
#include "ffx_fsr2_callbacks_glsl.h"
#include "ffx_fsr2_common.h"
#if defined(FSR2_BIND_CB_SPD)
layout (set = 1, binding = FSR2_BIND_CB_SPD, std140) uniform cbSPD_t
{
uint mips;
uint numWorkGroups;
uvec2 workGroupOffset;
uvec2 renderSize;
} cbSPD;
uint MipCount()
{
return cbSPD.mips;
}
uint NumWorkGroups()
{
return cbSPD.numWorkGroups;
}
uvec2 WorkGroupOffset()
{
return cbSPD.workGroupOffset;
}
uvec2 SPD_RenderSize()
{
return cbSPD.renderSize;
}
#endif
vec2 SPD_LoadExposureBuffer()
{
return imageLoad(rw_auto_exposure, ivec2(0,0)).xy;
}
void SPD_SetExposureBuffer(vec2 value)
{
imageStore(rw_auto_exposure, ivec2(0,0), vec4(value, 0.0f, 0.0f));
}
vec4 SPD_LoadMipmap5(ivec2 iPxPos)
{
return vec4(imageLoad(rw_img_mip_5, iPxPos).x, 0.0f, 0.0f, 0.0f);
}
void SPD_SetMipmap(ivec2 iPxPos, uint slice, float value)
{
switch (slice)
{
case FFX_FSR2_SHADING_CHANGE_MIP_LEVEL:
imageStore(rw_img_mip_shading_change, iPxPos, vec4(value, 0.0f, 0.0f, 0.0f));
break;
case 5:
imageStore(rw_img_mip_5, iPxPos, vec4(value, 0.0f, 0.0f, 0.0f));
break;
default:
// avoid flattened side effect
#if defined(FSR2_BIND_UAV_EXPOSURE_MIP_LUMA_CHANGE)
imageStore(rw_img_mip_shading_change, iPxPos, vec4(imageLoad(rw_img_mip_shading_change, iPxPos).x, 0.0f, 0.0f, 0.0f));
#elif defined(FSR2_BIND_UAV_EXPOSURE_MIP_5)
imageStore(rw_img_mip_5, iPxPos, vec4(imageLoad(rw_img_mip_5, iPxPos).x, 0.0f, 0.0f, 0.0f));
#endif
break;
}
}
void SPD_IncreaseAtomicCounter(inout uint spdCounter)
{
spdCounter = imageAtomicAdd(rw_spd_global_atomic, ivec2(0,0), 1);
}
void SPD_ResetAtomicCounter()
{
imageStore(rw_spd_global_atomic, ivec2(0,0), uvec4(0));
}
#include "ffx_fsr2_compute_luminance_pyramid.h"
#ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#define FFX_FSR2_THREAD_GROUP_WIDTH 256
#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#define FFX_FSR2_THREAD_GROUP_HEIGHT 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#define FFX_FSR2_THREAD_GROUP_DEPTH 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#ifndef FFX_FSR2_NUM_THREADS
#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in;
#endif // #ifndef FFX_FSR2_NUM_THREADS
FFX_FSR2_NUM_THREADS
void main()
{
ComputeAutoExposure(gl_WorkGroupID.xyz, gl_LocalInvocationIndex);
}

258
thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip.h vendored Normal file
View File

@ -0,0 +1,258 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#ifndef FFX_FSR2_DEPTH_CLIP_H
#define FFX_FSR2_DEPTH_CLIP_H
FFX_STATIC const FfxFloat32 DepthClipBaseScale = 4.0f;
FfxFloat32 ComputeDepthClip(FfxFloat32x2 fUvSample, FfxFloat32 fCurrentDepthSample)
{
FfxFloat32 fCurrentDepthViewSpace = GetViewSpaceDepth(fCurrentDepthSample);
BilinearSamplingData bilinearInfo = GetBilinearSamplingData(fUvSample, RenderSize());
FfxFloat32 fDilatedSum = 0.0f;
FfxFloat32 fDepth = 0.0f;
FfxFloat32 fWeightSum = 0.0f;
for (FfxInt32 iSampleIndex = 0; iSampleIndex < 4; iSampleIndex++) {
const FfxInt32x2 iOffset = bilinearInfo.iOffsets[iSampleIndex];
const FfxInt32x2 iSamplePos = bilinearInfo.iBasePos + iOffset;
if (IsOnScreen(iSamplePos, RenderSize())) {
const FfxFloat32 fWeight = bilinearInfo.fWeights[iSampleIndex];
if (fWeight > fReconstructedDepthBilinearWeightThreshold) {
const FfxFloat32 fPrevDepthSample = LoadReconstructedPrevDepth(iSamplePos);
const FfxFloat32 fPrevNearestDepthViewSpace = GetViewSpaceDepth(fPrevDepthSample);
const FfxFloat32 fDepthDiff = fCurrentDepthViewSpace - fPrevNearestDepthViewSpace;
if (fDepthDiff > 0.0f) {
#if FFX_FSR2_OPTION_INVERTED_DEPTH
const FfxFloat32 fPlaneDepth = ffxMin(fPrevDepthSample, fCurrentDepthSample);
#else
const FfxFloat32 fPlaneDepth = ffxMax(fPrevDepthSample, fCurrentDepthSample);
#endif
const FfxFloat32x3 fCenter = GetViewSpacePosition(FfxInt32x2(RenderSize() * 0.5f), RenderSize(), fPlaneDepth);
const FfxFloat32x3 fCorner = GetViewSpacePosition(FfxInt32x2(0, 0), RenderSize(), fPlaneDepth);
const FfxFloat32 fHalfViewportWidth = length(FfxFloat32x2(RenderSize()));
const FfxFloat32 fDepthThreshold = ffxMax(fCurrentDepthViewSpace, fPrevNearestDepthViewSpace);
const FfxFloat32 Ksep = 1.37e-05f;
const FfxFloat32 Kfov = length(fCorner) / length(fCenter);
const FfxFloat32 fRequiredDepthSeparation = Ksep * Kfov * fHalfViewportWidth * fDepthThreshold;
const FfxFloat32 fResolutionFactor = ffxSaturate(length(FfxFloat32x2(RenderSize())) / length(FfxFloat32x2(1920.0f, 1080.0f)));
const FfxFloat32 fPower = ffxLerp(1.0f, 3.0f, fResolutionFactor);
fDepth += ffxPow(ffxSaturate(FfxFloat32(fRequiredDepthSeparation / fDepthDiff)), fPower) * fWeight;
fWeightSum += fWeight;
}
}
}
}
return (fWeightSum > 0) ? ffxSaturate(1.0f - fDepth / fWeightSum) : 0.0f;
}
FfxFloat32 ComputeMotionDivergence(FfxInt32x2 iPxPos, FfxInt32x2 iPxInputMotionVectorSize)
{
FfxFloat32 minconvergence = 1.0f;
FfxFloat32x2 fMotionVectorNucleus = LoadInputMotionVector(iPxPos);
FfxFloat32 fNucleusVelocityLr = length(fMotionVectorNucleus * RenderSize());
FfxFloat32 fMaxVelocityUv = length(fMotionVectorNucleus);
const FfxFloat32 MotionVectorVelocityEpsilon = 1e-02f;
if (fNucleusVelocityLr > MotionVectorVelocityEpsilon) {
for (FfxInt32 y = -1; y <= 1; ++y) {
for (FfxInt32 x = -1; x <= 1; ++x) {
FfxInt32x2 sp = ClampLoad(iPxPos, FfxInt32x2(x, y), iPxInputMotionVectorSize);
FfxFloat32x2 fMotionVector = LoadInputMotionVector(sp);
FfxFloat32 fVelocityUv = length(fMotionVector);
fMaxVelocityUv = ffxMax(fVelocityUv, fMaxVelocityUv);
fVelocityUv = ffxMax(fVelocityUv, fMaxVelocityUv);
minconvergence = ffxMin(minconvergence, dot(fMotionVector / fVelocityUv, fMotionVectorNucleus / fVelocityUv));
}
}
}
return ffxSaturate(1.0f - minconvergence) * ffxSaturate(fMaxVelocityUv / 0.01f);
}
FfxFloat32 ComputeDepthDivergence(FfxInt32x2 iPxPos)
{
const FfxFloat32 fMaxDistInMeters = GetMaxDistanceInMeters();
FfxFloat32 fDepthMax = 0.0f;
FfxFloat32 fDepthMin = fMaxDistInMeters;
FfxInt32 iMaxDistFound = 0;
for (FfxInt32 y = -1; y < 2; y++) {
for (FfxInt32 x = -1; x < 2; x++) {
const FfxInt32x2 iOffset = FfxInt32x2(x, y);
const FfxInt32x2 iSamplePos = iPxPos + iOffset;
const FfxFloat32 fOnScreenFactor = IsOnScreen(iSamplePos, RenderSize()) ? 1.0f : 0.0f;
FfxFloat32 fDepth = GetViewSpaceDepthInMeters(LoadDilatedDepth(iSamplePos)) * fOnScreenFactor;
iMaxDistFound |= FfxInt32(fMaxDistInMeters == fDepth);
fDepthMin = ffxMin(fDepthMin, fDepth);
fDepthMax = ffxMax(fDepthMax, fDepth);
}
}
return (1.0f - fDepthMin / fDepthMax) * (FfxBoolean(iMaxDistFound) ? 0.0f : 1.0f);
}
FfxFloat32 ComputeTemporalMotionDivergence(FfxInt32x2 iPxPos)
{
const FfxFloat32x2 fUv = FfxFloat32x2(iPxPos + 0.5f) / RenderSize();
FfxFloat32x2 fMotionVector = LoadDilatedMotionVector(iPxPos);
FfxFloat32x2 fReprojectedUv = fUv + fMotionVector;
fReprojectedUv = ClampUv(fReprojectedUv, RenderSize(), MaxRenderSize());
FfxFloat32x2 fPrevMotionVector = SamplePreviousDilatedMotionVector(fReprojectedUv);
float fPxDistance = length(fMotionVector * DisplaySize());
return fPxDistance > 1.0f ? ffxLerp(0.0f, 1.0f - ffxSaturate(length(fPrevMotionVector) / length(fMotionVector)), ffxSaturate(ffxPow(fPxDistance / 20.0f, 3.0f))) : 0;
}
void PreProcessReactiveMasks(FfxInt32x2 iPxLrPos, FfxFloat32 fMotionDivergence)
{
// Compensate for bilinear sampling in accumulation pass
FfxFloat32x3 fReferenceColor = LoadInputColor(iPxLrPos).xyz;
FfxFloat32x2 fReactiveFactor = FfxFloat32x2(0.0f, fMotionDivergence);
float fMasksSum = 0.0f;
FfxFloat32x3 fColorSamples[9];
FfxFloat32 fReactiveSamples[9];
FfxFloat32 fTransparencyAndCompositionSamples[9];
FFX_UNROLL
for (FfxInt32 y = -1; y < 2; y++) {
FFX_UNROLL
for (FfxInt32 x = -1; x < 2; x++) {
const FfxInt32x2 sampleCoord = ClampLoad(iPxLrPos, FfxInt32x2(x, y), FfxInt32x2(RenderSize()));
FfxInt32 sampleIdx = (y + 1) * 3 + x + 1;
FfxFloat32x3 fColorSample = LoadInputColor(sampleCoord).xyz;
FfxFloat32 fReactiveSample = LoadReactiveMask(sampleCoord);
FfxFloat32 fTransparencyAndCompositionSample = LoadTransparencyAndCompositionMask(sampleCoord);
fColorSamples[sampleIdx] = fColorSample;
fReactiveSamples[sampleIdx] = fReactiveSample;
fTransparencyAndCompositionSamples[sampleIdx] = fTransparencyAndCompositionSample;
fMasksSum += (fReactiveSample + fTransparencyAndCompositionSample);
}
}
if (fMasksSum > 0)
{
for (FfxInt32 sampleIdx = 0; sampleIdx < 9; sampleIdx++)
{
FfxFloat32x3 fColorSample = fColorSamples[sampleIdx];
FfxFloat32 fReactiveSample = fReactiveSamples[sampleIdx];
FfxFloat32 fTransparencyAndCompositionSample = fTransparencyAndCompositionSamples[sampleIdx];
const FfxFloat32 fMaxLenSq = ffxMax(dot(fReferenceColor, fReferenceColor), dot(fColorSample, fColorSample));
const FfxFloat32 fSimilarity = dot(fReferenceColor, fColorSample) / fMaxLenSq;
// Increase power for non-similar samples
const FfxFloat32 fPowerBiasMax = 6.0f;
const FfxFloat32 fSimilarityPower = 1.0f + (fPowerBiasMax - fSimilarity * fPowerBiasMax);
const FfxFloat32 fWeightedReactiveSample = ffxPow(fReactiveSample, fSimilarityPower);
const FfxFloat32 fWeightedTransparencyAndCompositionSample = ffxPow(fTransparencyAndCompositionSample, fSimilarityPower);
fReactiveFactor = ffxMax(fReactiveFactor, FfxFloat32x2(fWeightedReactiveSample, fWeightedTransparencyAndCompositionSample));
}
}
StoreDilatedReactiveMasks(iPxLrPos, fReactiveFactor);
}
FfxFloat32x3 ComputePreparedInputColor(FfxInt32x2 iPxLrPos)
{
//We assume linear data. if non-linear input (sRGB, ...),
//then we should convert to linear first and back to sRGB on output.
FfxFloat32x3 fRgb = ffxMax(FfxFloat32x3(0, 0, 0), LoadInputColor(iPxLrPos));
fRgb = PrepareRgb(fRgb, Exposure(), PreExposure());
const FfxFloat32x3 fPreparedYCoCg = RGBToYCoCg(fRgb);
return fPreparedYCoCg;
}
FfxFloat32 EvaluateSurface(FfxInt32x2 iPxPos, FfxFloat32x2 fMotionVector)
{
FfxFloat32 d0 = GetViewSpaceDepth(LoadReconstructedPrevDepth(iPxPos + FfxInt32x2(0, -1)));
FfxFloat32 d1 = GetViewSpaceDepth(LoadReconstructedPrevDepth(iPxPos + FfxInt32x2(0, 0)));
FfxFloat32 d2 = GetViewSpaceDepth(LoadReconstructedPrevDepth(iPxPos + FfxInt32x2(0, 1)));
return 1.0f - FfxFloat32(((d0 - d1) > (d1 * 0.01f)) && ((d1 - d2) > (d2 * 0.01f)));
}
void DepthClip(FfxInt32x2 iPxPos)
{
FfxFloat32x2 fDepthUv = (iPxPos + 0.5f) / RenderSize();
FfxFloat32x2 fMotionVector = LoadDilatedMotionVector(iPxPos);
// Discard tiny mvs
fMotionVector *= FfxFloat32(length(fMotionVector * DisplaySize()) > 0.01f);
const FfxFloat32x2 fDilatedUv = fDepthUv + fMotionVector;
const FfxFloat32 fDilatedDepth = LoadDilatedDepth(iPxPos);
const FfxFloat32 fCurrentDepthViewSpace = GetViewSpaceDepth(LoadInputDepth(iPxPos));
// Compute prepared input color and depth clip
FfxFloat32 fDepthClip = ComputeDepthClip(fDilatedUv, fDilatedDepth) * EvaluateSurface(iPxPos, fMotionVector);
FfxFloat32x3 fPreparedYCoCg = ComputePreparedInputColor(iPxPos);
StorePreparedInputColor(iPxPos, FfxFloat32x4(fPreparedYCoCg, fDepthClip));
// Compute dilated reactive mask
#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
FfxInt32x2 iSamplePos = iPxPos;
#else
FfxInt32x2 iSamplePos = ComputeHrPosFromLrPos(iPxPos);
#endif
FfxFloat32 fMotionDivergence = ComputeMotionDivergence(iSamplePos, RenderSize());
FfxFloat32 fTemporalMotionDifference = ffxSaturate(ComputeTemporalMotionDivergence(iPxPos) - ComputeDepthDivergence(iPxPos));
PreProcessReactiveMasks(iPxPos, ffxMax(fTemporalMotionDifference, fMotionDivergence));
}
#endif //!defined( FFX_FSR2_DEPTH_CLIPH )

67
thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip_pass.glsl vendored Normal file
View File

@ -0,0 +1,67 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#extension GL_GOOGLE_include_directive : require
#extension GL_EXT_samplerless_texture_functions : require
#define FSR2_BIND_SRV_RECONSTRUCTED_PREV_NEAREST_DEPTH 0
#define FSR2_BIND_SRV_DILATED_MOTION_VECTORS 1
#define FSR2_BIND_SRV_DILATED_DEPTH 2
#define FSR2_BIND_SRV_REACTIVE_MASK 3
#define FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK 4
#define FSR2_BIND_SRV_PREPARED_INPUT_COLOR 5
#define FSR2_BIND_SRV_PREVIOUS_DILATED_MOTION_VECTORS 6
#define FSR2_BIND_SRV_INPUT_MOTION_VECTORS 7
#define FSR2_BIND_SRV_INPUT_COLOR 8
#define FSR2_BIND_SRV_INPUT_DEPTH 9
#define FSR2_BIND_SRV_INPUT_EXPOSURE 10
#define FSR2_BIND_UAV_DEPTH_CLIP 11
#define FSR2_BIND_UAV_DILATED_REACTIVE_MASKS 12
#define FSR2_BIND_UAV_PREPARED_INPUT_COLOR 13
#define FSR2_BIND_CB_FSR2 14
#include "ffx_fsr2_callbacks_glsl.h"
#include "ffx_fsr2_common.h"
#include "ffx_fsr2_sample.h"
#include "ffx_fsr2_depth_clip.h"
#ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#define FFX_FSR2_THREAD_GROUP_WIDTH 8
#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#define FFX_FSR2_THREAD_GROUP_HEIGHT 8
#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#define FFX_FSR2_THREAD_GROUP_DEPTH 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#ifndef FFX_FSR2_NUM_THREADS
#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in;
#endif // #ifndef FFX_FSR2_NUM_THREADS
FFX_FSR2_NUM_THREADS
void main()
{
DepthClip(ivec2(gl_GlobalInvocationID.xy));
}

1
thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_begin.h vendored Normal file
View File

@ -0,0 +1 @@
// This file doesn't exist in this version of FSR.

1
thirdparty/amd-fsr2/shaders/ffx_fsr2_force16_end.h vendored Normal file
View File

@ -0,0 +1 @@
// This file doesn't exist in this version of FSR.

115
thirdparty/amd-fsr2/shaders/ffx_fsr2_lock.h vendored Normal file
View File

@ -0,0 +1,115 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#ifndef FFX_FSR2_LOCK_H
#define FFX_FSR2_LOCK_H
void ClearResourcesForNextFrame(in FfxInt32x2 iPxHrPos)
{
if (all(FFX_LESS_THAN(iPxHrPos, FfxInt32x2(RenderSize()))))
{
#if FFX_FSR2_OPTION_INVERTED_DEPTH
const FfxUInt32 farZ = 0x0;
#else
const FfxUInt32 farZ = 0x3f800000;
#endif
SetReconstructedDepth(iPxHrPos, farZ);
}
}
FfxBoolean ComputeThinFeatureConfidence(FfxInt32x2 pos)
{
const FfxInt32 RADIUS = 1;
FfxFloat32 fNucleus = LoadLockInputLuma(pos);
FfxFloat32 similar_threshold = 1.05f;
FfxFloat32 dissimilarLumaMin = FSR2_FLT_MAX;
FfxFloat32 dissimilarLumaMax = 0;
/*
0 1 2
3 4 5
6 7 8
*/
#define SETBIT(x) (1U << x)
FfxUInt32 mask = SETBIT(4); //flag fNucleus as similar
const FfxUInt32 uNumRejectionMasks = 4;
const FfxUInt32 uRejectionMasks[uNumRejectionMasks] = {
SETBIT(0) | SETBIT(1) | SETBIT(3) | SETBIT(4), //Upper left
SETBIT(1) | SETBIT(2) | SETBIT(4) | SETBIT(5), //Upper right
SETBIT(3) | SETBIT(4) | SETBIT(6) | SETBIT(7), //Lower left
SETBIT(4) | SETBIT(5) | SETBIT(7) | SETBIT(8), //Lower right
};
FfxInt32 idx = 0;
FFX_UNROLL
for (FfxInt32 y = -RADIUS; y <= RADIUS; y++) {
FFX_UNROLL
for (FfxInt32 x = -RADIUS; x <= RADIUS; x++, idx++) {
if (x == 0 && y == 0) continue;
FfxInt32x2 samplePos = ClampLoad(pos, FfxInt32x2(x, y), FfxInt32x2(RenderSize()));
FfxFloat32 sampleLuma = LoadLockInputLuma(samplePos);
FfxFloat32 difference = ffxMax(sampleLuma, fNucleus) / ffxMin(sampleLuma, fNucleus);
if (difference > 0 && (difference < similar_threshold)) {
mask |= SETBIT(idx);
} else {
dissimilarLumaMin = ffxMin(dissimilarLumaMin, sampleLuma);
dissimilarLumaMax = ffxMax(dissimilarLumaMax, sampleLuma);
}
}
}
FfxBoolean isRidge = fNucleus > dissimilarLumaMax || fNucleus < dissimilarLumaMin;
if (FFX_FALSE == isRidge) {
return false;
}
FFX_UNROLL
for (FfxInt32 i = 0; i < 4; i++) {
if ((mask & uRejectionMasks[i]) == uRejectionMasks[i]) {
return false;
}
}
return true;
}
void ComputeLock(FfxInt32x2 iPxLrPos)
{
if (ComputeThinFeatureConfidence(iPxLrPos))
{
StoreNewLocks(ComputeHrPosFromLrPos(iPxLrPos), 1.f);
}
ClearResourcesForNextFrame(iPxLrPos);
}
#endif // FFX_FSR2_LOCK_H

56
thirdparty/amd-fsr2/shaders/ffx_fsr2_lock_pass.glsl vendored Normal file
View File

@ -0,0 +1,56 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#extension GL_GOOGLE_include_directive : require
#extension GL_EXT_samplerless_texture_functions : require
#define FSR2_BIND_SRV_LOCK_INPUT_LUMA 0
#define FSR2_BIND_UAV_NEW_LOCKS 1
#define FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH 2
#define FSR2_BIND_CB_FSR2 3
#include "ffx_fsr2_callbacks_glsl.h"
#include "ffx_fsr2_common.h"
#include "ffx_fsr2_sample.h"
#include "ffx_fsr2_lock.h"
#ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#define FFX_FSR2_THREAD_GROUP_WIDTH 8
#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#define FFX_FSR2_THREAD_GROUP_HEIGHT 8
#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#define FFX_FSR2_THREAD_GROUP_DEPTH 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#ifndef FFX_FSR2_NUM_THREADS
#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in;
#endif // #ifndef FFX_FSR2_NUM_THREADS
FFX_FSR2_NUM_THREADS
void main()
{
uvec2 uDispatchThreadId = gl_WorkGroupID.xy * uvec2(FFX_FSR2_THREAD_GROUP_WIDTH, FFX_FSR2_THREAD_GROUP_HEIGHT) + gl_LocalInvocationID.xy;
ComputeLock(ivec2(uDispatchThreadId));
}

106
thirdparty/amd-fsr2/shaders/ffx_fsr2_postprocess_lock_status.h vendored Normal file
View File

@ -0,0 +1,106 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#ifndef FFX_FSR2_POSTPROCESS_LOCK_STATUS_H
#define FFX_FSR2_POSTPROCESS_LOCK_STATUS_H
FfxFloat32x4 WrapShadingChangeLuma(FfxInt32x2 iPxSample)
{
return FfxFloat32x4(LoadMipLuma(iPxSample, LumaMipLevelToUse()), 0, 0, 0);
}
#if FFX_HALF
FFX_MIN16_F4 WrapShadingChangeLuma(FFX_MIN16_I2 iPxSample)
{
return FFX_MIN16_F4(LoadMipLuma(iPxSample, LumaMipLevelToUse()), 0, 0, 0);
}
#endif
#if FFX_FSR2_OPTION_POSTPROCESSLOCKSTATUS_SAMPLERS_USE_DATA_HALF && FFX_HALF
DeclareCustomFetchBilinearSamplesMin16(FetchShadingChangeLumaSamples, WrapShadingChangeLuma)
#else
DeclareCustomFetchBicubicSamples(FetchShadingChangeLumaSamples, WrapShadingChangeLuma)
#endif
DeclareCustomTextureSample(ShadingChangeLumaSample, Lanczos2, FetchShadingChangeLumaSamples)
FfxFloat32 GetShadingChangeLuma(FfxInt32x2 iPxHrPos, FfxFloat32x2 fUvCoord)
{
FfxFloat32 fShadingChangeLuma = 0;
#if 0
fShadingChangeLuma = Exposure() * exp(ShadingChangeLumaSample(fUvCoord, LumaMipDimensions()).x);
#else
const FfxFloat32 fDiv = FfxFloat32(2 << LumaMipLevelToUse());
FfxInt32x2 iMipRenderSize = FfxInt32x2(RenderSize() / fDiv);
fUvCoord = ClampUv(fUvCoord, iMipRenderSize, LumaMipDimensions());
fShadingChangeLuma = Exposure() * exp(FfxFloat32(SampleMipLuma(fUvCoord, LumaMipLevelToUse())));
#endif
fShadingChangeLuma = ffxPow(fShadingChangeLuma, 1.0f / 6.0f);
return fShadingChangeLuma;
}
void UpdateLockStatus(AccumulationPassCommonParams params,
FFX_PARAMETER_INOUT FfxFloat32 fReactiveFactor, LockState state,
FFX_PARAMETER_INOUT FfxFloat32x2 fLockStatus,
FFX_PARAMETER_OUT FfxFloat32 fLockContributionThisFrame,
FFX_PARAMETER_OUT FfxFloat32 fLuminanceDiff) {
const FfxFloat32 fShadingChangeLuma = GetShadingChangeLuma(params.iPxHrPos, params.fHrUv);
//init temporal shading change factor, init to -1 or so in reproject to know if "true new"?
fLockStatus[LOCK_TEMPORAL_LUMA] = (fLockStatus[LOCK_TEMPORAL_LUMA] == FfxFloat32(0.0f)) ? fShadingChangeLuma : fLockStatus[LOCK_TEMPORAL_LUMA];
FfxFloat32 fPreviousShadingChangeLuma = fLockStatus[LOCK_TEMPORAL_LUMA];
fLuminanceDiff = 1.0f - MinDividedByMax(fPreviousShadingChangeLuma, fShadingChangeLuma);
if (state.NewLock) {
fLockStatus[LOCK_TEMPORAL_LUMA] = fShadingChangeLuma;
fLockStatus[LOCK_LIFETIME_REMAINING] = (fLockStatus[LOCK_LIFETIME_REMAINING] != 0.0f) ? 2.0f : 1.0f;
}
else if(fLockStatus[LOCK_LIFETIME_REMAINING] <= 1.0f) {
fLockStatus[LOCK_TEMPORAL_LUMA] = ffxLerp(fLockStatus[LOCK_TEMPORAL_LUMA], FfxFloat32(fShadingChangeLuma), 0.5f);
}
else {
if (fLuminanceDiff > 0.1f) {
KillLock(fLockStatus);
}
}
fReactiveFactor = ffxMax(fReactiveFactor, ffxSaturate((fLuminanceDiff - 0.1f) * 10.0f));
fLockStatus[LOCK_LIFETIME_REMAINING] *= (1.0f - fReactiveFactor);
fLockStatus[LOCK_LIFETIME_REMAINING] *= ffxSaturate(1.0f - params.fAccumulationMask);
fLockStatus[LOCK_LIFETIME_REMAINING] *= FfxFloat32(params.fDepthClipFactor < 0.1f);
// Compute this frame lock contribution
const FfxFloat32 fLifetimeContribution = ffxSaturate(fLockStatus[LOCK_LIFETIME_REMAINING] - 1.0f);
const FfxFloat32 fShadingChangeContribution = ffxSaturate(MinDividedByMax(fLockStatus[LOCK_TEMPORAL_LUMA], fShadingChangeLuma));
fLockContributionThisFrame = ffxSaturate(ffxSaturate(fLifetimeContribution * 4.0f) * fShadingChangeContribution);
}
#endif //!defined( FFX_FSR2_POSTPROCESS_LOCK_STATUS_H )

67
thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas.h vendored Normal file
View File

@ -0,0 +1,67 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#define GROUP_SIZE 8
#define FSR_RCAS_DENOISE 1
void WriteUpscaledOutput(FFX_MIN16_U2 iPxHrPos, FfxFloat32x3 fUpscaledColor)
{
StoreUpscaledOutput(FFX_MIN16_I2(iPxHrPos), fUpscaledColor);
}
#define FSR_RCAS_F
FfxFloat32x4 FsrRcasLoadF(FfxInt32x2 p)
{
FfxFloat32x4 fColor = LoadRCAS_Input(p);
fColor.rgb = PrepareRgb(fColor.rgb, Exposure(), PreExposure());
return fColor;
}
void FsrRcasInputF(inout FfxFloat32 r, inout FfxFloat32 g, inout FfxFloat32 b) {}
#include "ffx_fsr1.h"
void CurrFilter(FFX_MIN16_U2 pos)
{
FfxFloat32x3 c;
FsrRcasF(c.r, c.g, c.b, pos, RCASConfig());
c = UnprepareRgb(c, Exposure());
WriteUpscaledOutput(pos, c);
}
void RCAS(FfxUInt32x3 LocalThreadId, FfxUInt32x3 WorkGroupId, FfxUInt32x3 Dtid)
{
// Do remapping of local xy in workgroup for a more PS-like swizzle pattern.
FfxUInt32x2 gxy = ffxRemapForQuad(LocalThreadId.x) + FfxUInt32x2(WorkGroupId.x << 4u, WorkGroupId.y << 4u);
CurrFilter(FFX_MIN16_U2(gxy));
gxy.x += 8u;
CurrFilter(FFX_MIN16_U2(gxy));
gxy.y += 8u;
CurrFilter(FFX_MIN16_U2(gxy));
gxy.x -= 8u;
CurrFilter(FFX_MIN16_U2(gxy));
}

80
thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas_pass.glsl vendored Normal file
View File

@ -0,0 +1,80 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#extension GL_GOOGLE_include_directive : require
#extension GL_EXT_samplerless_texture_functions : require
// Needed for rw_upscaled_output declaration
#extension GL_EXT_shader_image_load_formatted : require
#define FSR2_BIND_SRV_INPUT_EXPOSURE 0
#define FSR2_BIND_SRV_RCAS_INPUT 1
#define FSR2_BIND_UAV_UPSCALED_OUTPUT 2
#define FSR2_BIND_CB_FSR2 3
#define FSR2_BIND_CB_RCAS 4
#include "ffx_fsr2_callbacks_glsl.h"
#include "ffx_fsr2_common.h"
//Move to prototype shader!
#if defined(FSR2_BIND_CB_RCAS)
layout (set = 1, binding = FSR2_BIND_CB_RCAS, std140) uniform cbRCAS_t
{
uvec4 rcasConfig;
} cbRCAS;
uvec4 RCASConfig()
{
return cbRCAS.rcasConfig;
}
#else
uvec4 RCASConfig()
{
return uvec4(0);
}
#endif
vec4 LoadRCAS_Input(FfxInt32x2 iPxPos)
{
return texelFetch(r_rcas_input, iPxPos, 0);
}
#include "ffx_fsr2_rcas.h"
#ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#define FFX_FSR2_THREAD_GROUP_WIDTH 64
#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#define FFX_FSR2_THREAD_GROUP_HEIGHT 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#define FFX_FSR2_THREAD_GROUP_DEPTH 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#ifndef FFX_FSR2_NUM_THREADS
#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in;
#endif // #ifndef FFX_FSR2_NUM_THREADS
FFX_FSR2_NUM_THREADS
void main()
{
RCAS(gl_LocalInvocationID.xyz, gl_WorkGroupID.xyz, gl_GlobalInvocationID.xyz);
}

145
thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h vendored Normal file
View File

@ -0,0 +1,145 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#ifndef FFX_FSR2_RECONSTRUCT_DILATED_VELOCITY_AND_PREVIOUS_DEPTH_H
#define FFX_FSR2_RECONSTRUCT_DILATED_VELOCITY_AND_PREVIOUS_DEPTH_H
void ReconstructPrevDepth(FfxInt32x2 iPxPos, FfxFloat32 fDepth, FfxFloat32x2 fMotionVector, FfxInt32x2 iPxDepthSize)
{
fMotionVector *= FfxFloat32(length(fMotionVector * DisplaySize()) > 0.1f);
FfxFloat32x2 fUv = (iPxPos + FfxFloat32(0.5)) / iPxDepthSize;
FfxFloat32x2 fReprojectedUv = fUv + fMotionVector;
BilinearSamplingData bilinearInfo = GetBilinearSamplingData(fReprojectedUv, RenderSize());
// Project current depth into previous frame locations.
// Push to all pixels having some contribution if reprojection is using bilinear logic.
for (FfxInt32 iSampleIndex = 0; iSampleIndex < 4; iSampleIndex++) {
const FfxInt32x2 iOffset = bilinearInfo.iOffsets[iSampleIndex];
FfxFloat32 fWeight = bilinearInfo.fWeights[iSampleIndex];
if (fWeight > fReconstructedDepthBilinearWeightThreshold) {
FfxInt32x2 iStorePos = bilinearInfo.iBasePos + iOffset;
if (IsOnScreen(iStorePos, iPxDepthSize)) {
StoreReconstructedDepth(iStorePos, fDepth);
}
}
}
}
void FindNearestDepth(FFX_PARAMETER_IN FfxInt32x2 iPxPos, FFX_PARAMETER_IN FfxInt32x2 iPxSize, FFX_PARAMETER_OUT FfxFloat32 fNearestDepth, FFX_PARAMETER_OUT FfxInt32x2 fNearestDepthCoord)
{
const FfxInt32 iSampleCount = 9;
const FfxInt32x2 iSampleOffsets[iSampleCount] = {
FfxInt32x2(+0, +0),
FfxInt32x2(+1, +0),
FfxInt32x2(+0, +1),
FfxInt32x2(+0, -1),
FfxInt32x2(-1, +0),
FfxInt32x2(-1, +1),
FfxInt32x2(+1, +1),
FfxInt32x2(-1, -1),
FfxInt32x2(+1, -1),
};
// pull out the depth loads to allow SC to batch them
FfxFloat32 depth[9];
FfxInt32 iSampleIndex = 0;
FFX_UNROLL
for (iSampleIndex = 0; iSampleIndex < iSampleCount; ++iSampleIndex) {
FfxInt32x2 iPos = iPxPos + iSampleOffsets[iSampleIndex];
depth[iSampleIndex] = LoadInputDepth(iPos);
}
// find closest depth
fNearestDepthCoord = iPxPos;
fNearestDepth = depth[0];
FFX_UNROLL
for (iSampleIndex = 1; iSampleIndex < iSampleCount; ++iSampleIndex) {
FfxInt32x2 iPos = iPxPos + iSampleOffsets[iSampleIndex];
if (IsOnScreen(iPos, iPxSize)) {
FfxFloat32 fNdDepth = depth[iSampleIndex];
#if FFX_FSR2_OPTION_INVERTED_DEPTH
if (fNdDepth > fNearestDepth) {
#else
if (fNdDepth < fNearestDepth) {
#endif
fNearestDepthCoord = iPos;
fNearestDepth = fNdDepth;
}
}
}
}
FfxFloat32 ComputeLockInputLuma(FfxInt32x2 iPxLrPos)
{
//We assume linear data. if non-linear input (sRGB, ...),
//then we should convert to linear first and back to sRGB on output.
FfxFloat32x3 fRgb = ffxMax(FfxFloat32x3(0, 0, 0), LoadInputColor(iPxLrPos));
// Use internal auto exposure for locking logic
fRgb /= PreExposure();
fRgb *= Exposure();
#if FFX_FSR2_OPTION_HDR_COLOR_INPUT
fRgb = Tonemap(fRgb);
#endif
//compute luma used to lock pixels, if used elsewhere the ffxPow must be moved!
const FfxFloat32 fLockInputLuma = ffxPow(RGBToPerceivedLuma(fRgb), FfxFloat32(1.0 / 6.0));
return fLockInputLuma;
}
void ReconstructAndDilate(FfxInt32x2 iPxLrPos)
{
FfxFloat32 fDilatedDepth;
FfxInt32x2 iNearestDepthCoord;
FindNearestDepth(iPxLrPos, RenderSize(), fDilatedDepth, iNearestDepthCoord);
#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
FfxInt32x2 iSamplePos = iPxLrPos;
FfxInt32x2 iMotionVectorPos = iNearestDepthCoord;
#else
FfxInt32x2 iSamplePos = ComputeHrPosFromLrPos(iPxLrPos);
FfxInt32x2 iMotionVectorPos = ComputeHrPosFromLrPos(iNearestDepthCoord);
#endif
FfxFloat32x2 fDilatedMotionVector = LoadInputMotionVector(iMotionVectorPos);
StoreDilatedDepth(iPxLrPos, fDilatedDepth);
StoreDilatedMotionVector(iPxLrPos, fDilatedMotionVector);
ReconstructPrevDepth(iPxLrPos, fDilatedDepth, fDilatedMotionVector, RenderSize());
FfxFloat32 fLockInputLuma = ComputeLockInputLuma(iPxLrPos);
StoreLockInputLuma(iPxLrPos, fLockInputLuma);
}
#endif //!defined( FFX_FSR2_RECONSTRUCT_DILATED_VELOCITY_AND_PREVIOUS_DEPTH_H )

65
thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl vendored Normal file
View File

@ -0,0 +1,65 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#extension GL_GOOGLE_include_directive : require
#extension GL_EXT_samplerless_texture_functions : require
#define FSR2_BIND_SRV_INPUT_MOTION_VECTORS 0
#define FSR2_BIND_SRV_INPUT_DEPTH 1
#define FSR2_BIND_SRV_INPUT_COLOR 2
#define FSR2_BIND_SRV_INPUT_EXPOSURE 3
#define FSR2_BIND_SRV_LUMA_HISTORY 4
#define FSR2_BIND_UAV_RECONSTRUCTED_PREV_NEAREST_DEPTH 5
#define FSR2_BIND_UAV_DILATED_MOTION_VECTORS 6
#define FSR2_BIND_UAV_DILATED_DEPTH 7
#define FSR2_BIND_UAV_PREPARED_INPUT_COLOR 8
#define FSR2_BIND_UAV_LUMA_HISTORY 9
#define FSR2_BIND_UAV_LUMA_INSTABILITY 10
#define FSR2_BIND_UAV_LOCK_INPUT_LUMA 11
#define FSR2_BIND_CB_FSR2 12
#include "ffx_fsr2_callbacks_glsl.h"
#include "ffx_fsr2_common.h"
#include "ffx_fsr2_sample.h"
#include "ffx_fsr2_reconstruct_dilated_velocity_and_previous_depth.h"
#ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#define FFX_FSR2_THREAD_GROUP_WIDTH 8
#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#define FFX_FSR2_THREAD_GROUP_HEIGHT 8
#endif // #ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#define FFX_FSR2_THREAD_GROUP_DEPTH 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#ifndef FFX_FSR2_NUM_THREADS
#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in;
#endif // #ifndef FFX_FSR2_NUM_THREADS
FFX_FSR2_NUM_THREADS
void main()
{
ReconstructAndDilate(FFX_MIN16_I2(gl_GlobalInvocationID.xy));
}

136
thirdparty/amd-fsr2/shaders/ffx_fsr2_reproject.h vendored Normal file
View File

@ -0,0 +1,136 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#ifndef FFX_FSR2_REPROJECT_H
#define FFX_FSR2_REPROJECT_H
#ifndef FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE
#define FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE 0 // Reference
#endif
FfxFloat32x4 WrapHistory(FfxInt32x2 iPxSample)
{
return LoadHistory(iPxSample);
}
#if FFX_HALF
FFX_MIN16_F4 WrapHistory(FFX_MIN16_I2 iPxSample)
{
return FFX_MIN16_F4(LoadHistory(iPxSample));
}
#endif
#if FFX_FSR2_OPTION_REPROJECT_SAMPLERS_USE_DATA_HALF && FFX_HALF
DeclareCustomFetchBicubicSamplesMin16(FetchHistorySamples, WrapHistory)
DeclareCustomTextureSampleMin16(HistorySample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchHistorySamples)
#else
DeclareCustomFetchBicubicSamples(FetchHistorySamples, WrapHistory)
DeclareCustomTextureSample(HistorySample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchHistorySamples)
#endif
FfxFloat32x4 WrapLockStatus(FfxInt32x2 iPxSample)
{
FfxFloat32x4 fSample = FfxFloat32x4(LoadLockStatus(iPxSample), 0.0f, 0.0f);
return fSample;
}
#if FFX_HALF
FFX_MIN16_F4 WrapLockStatus(FFX_MIN16_I2 iPxSample)
{
FFX_MIN16_F4 fSample = FFX_MIN16_F4(LoadLockStatus(iPxSample), 0.0, 0.0);
return fSample;
}
#endif
#if 1
#if FFX_FSR2_OPTION_REPROJECT_SAMPLERS_USE_DATA_HALF && FFX_HALF
DeclareCustomFetchBilinearSamplesMin16(FetchLockStatusSamples, WrapLockStatus)
DeclareCustomTextureSampleMin16(LockStatusSample, Bilinear, FetchLockStatusSamples)
#else
DeclareCustomFetchBilinearSamples(FetchLockStatusSamples, WrapLockStatus)
DeclareCustomTextureSample(LockStatusSample, Bilinear, FetchLockStatusSamples)
#endif
#else
#if FFX_FSR2_OPTION_REPROJECT_SAMPLERS_USE_DATA_HALF && FFX_HALF
DeclareCustomFetchBicubicSamplesMin16(FetchLockStatusSamples, WrapLockStatus)
DeclareCustomTextureSampleMin16(LockStatusSample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchLockStatusSamples)
#else
DeclareCustomFetchBicubicSamples(FetchLockStatusSamples, WrapLockStatus)
DeclareCustomTextureSample(LockStatusSample, FFX_FSR2_GET_LANCZOS_SAMPLER1D(FFX_FSR2_OPTION_REPROJECT_USE_LANCZOS_TYPE), FetchLockStatusSamples)
#endif
#endif
FfxFloat32x2 GetMotionVector(FfxInt32x2 iPxHrPos, FfxFloat32x2 fHrUv)
{
#if FFX_FSR2_OPTION_LOW_RESOLUTION_MOTION_VECTORS
FfxFloat32x2 fDilatedMotionVector = LoadDilatedMotionVector(FFX_MIN16_I2(fHrUv * RenderSize()));
#else
FfxFloat32x2 fDilatedMotionVector = LoadInputMotionVector(iPxHrPos);
#endif
return fDilatedMotionVector;
}
FfxBoolean IsUvInside(FfxFloat32x2 fUv)
{
return (fUv.x >= 0.0f && fUv.x <= 1.0f) && (fUv.y >= 0.0f && fUv.y <= 1.0f);
}
void ComputeReprojectedUVs(const AccumulationPassCommonParams params, FFX_PARAMETER_OUT FfxFloat32x2 fReprojectedHrUv, FFX_PARAMETER_OUT FfxBoolean bIsExistingSample)
{
fReprojectedHrUv = params.fHrUv + params.fMotionVector;
bIsExistingSample = IsUvInside(fReprojectedHrUv);
}
void ReprojectHistoryColor(const AccumulationPassCommonParams params, FFX_PARAMETER_OUT FfxFloat32x3 fHistoryColor, FFX_PARAMETER_OUT FfxFloat32 fTemporalReactiveFactor, FFX_PARAMETER_OUT FfxBoolean bInMotionLastFrame)
{
FfxFloat32x4 fHistory = HistorySample(params.fReprojectedHrUv, DisplaySize());
fHistoryColor = PrepareRgb(fHistory.rgb, Exposure(), PreviousFramePreExposure());
fHistoryColor = RGBToYCoCg(fHistoryColor);
//Compute temporal reactivity info
fTemporalReactiveFactor = ffxSaturate(abs(fHistory.w));
bInMotionLastFrame = (fHistory.w < 0.0f);
}
LockState ReprojectHistoryLockStatus(const AccumulationPassCommonParams params, FFX_PARAMETER_OUT FfxFloat32x2 fReprojectedLockStatus)
{
LockState state = { FFX_FALSE, FFX_FALSE };
const FfxFloat32 fNewLockIntensity = LoadRwNewLocks(params.iPxHrPos);
state.NewLock = fNewLockIntensity > (127.0f / 255.0f);
FfxFloat32 fInPlaceLockLifetime = state.NewLock ? fNewLockIntensity : 0;
fReprojectedLockStatus = SampleLockStatus(params.fReprojectedHrUv);
if (fReprojectedLockStatus[LOCK_LIFETIME_REMAINING] != FfxFloat32(0.0f)) {
state.WasLockedPrevFrame = true;
}
return state;
}
#endif //!defined( FFX_FSR2_REPROJECT_H )

105
thirdparty/amd-fsr2/shaders/ffx_fsr2_resources.h vendored Normal file
View File

@ -0,0 +1,105 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#ifndef FFX_FSR2_RESOURCES_H
#define FFX_FSR2_RESOURCES_H
#if defined(FFX_CPU) || defined(FFX_GPU)
#define FFX_FSR2_RESOURCE_IDENTIFIER_NULL 0
#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_OPAQUE_ONLY 1
#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_COLOR 2
#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_MOTION_VECTORS 3
#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_DEPTH 4
#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_EXPOSURE 5
#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_REACTIVE_MASK 6
#define FFX_FSR2_RESOURCE_IDENTIFIER_INPUT_TRANSPARENCY_AND_COMPOSITION_MASK 7
#define FFX_FSR2_RESOURCE_IDENTIFIER_RECONSTRUCTED_PREVIOUS_NEAREST_DEPTH 8
#define FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_MOTION_VECTORS 9
#define FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_DEPTH 10
#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR 11
#define FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS 12
#define FFX_FSR2_RESOURCE_IDENTIFIER_NEW_LOCKS 13
#define FFX_FSR2_RESOURCE_IDENTIFIER_PREPARED_INPUT_COLOR 14
#define FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY 15
#define FFX_FSR2_RESOURCE_IDENTIFIER_DEBUG_OUTPUT 16
#define FFX_FSR2_RESOURCE_IDENTIFIER_LANCZOS_LUT 17
#define FFX_FSR2_RESOURCE_IDENTIFIER_SPD_ATOMIC_COUNT 18
#define FFX_FSR2_RESOURCE_IDENTIFIER_UPSCALED_OUTPUT 19
#define FFX_FSR2_RESOURCE_IDENTIFIER_RCAS_INPUT 20
#define FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_1 21
#define FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_STATUS_2 22
#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_1 23
#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_UPSCALED_COLOR_2 24
#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_REACTIVITY 25
#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_TRANSPARENCY_AND_COMPOSITION 26
#define FFX_FSR2_RESOURCE_IDENTITIER_UPSAMPLE_MAXIMUM_BIAS_LUT 27
#define FFX_FSR2_RESOURCE_IDENTIFIER_DILATED_REACTIVE_MASKS 28
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE 29 // same as FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_0
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_0 29
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_1 30
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_2 31
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_3 32
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_4 33
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_5 34
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_6 35
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_7 36
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_8 37
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_9 38
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_10 39
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_11 40
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_12 41
#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DEFAULT_EXPOSURE 42
#define FFX_FSR2_RESOURCE_IDENTIFIER_AUTO_EXPOSURE 43
#define FFX_FSR2_RESOURCE_IDENTIFIER_AUTOREACTIVE 44
#define FFX_FSR2_RESOURCE_IDENTIFIER_AUTOCOMPOSITION 45
#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR 46
#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR 47
#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_1 48
#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_1 49
#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_PRE_ALPHA_COLOR_2 50
#define FFX_FSR2_RESOURCE_IDENTIFIER_PREV_POST_ALPHA_COLOR_2 51
#define FFX_FSR2_RESOURCE_IDENTIFIER_PREVIOUS_DILATED_MOTION_VECTORS 52
#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_1 53
#define FFX_FSR2_RESOURCE_IDENTIFIER_INTERNAL_DILATED_MOTION_VECTORS_2 54
#define FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_1 55
#define FFX_FSR2_RESOURCE_IDENTIFIER_LUMA_HISTORY_2 56
#define FFX_FSR2_RESOURCE_IDENTIFIER_LOCK_INPUT_LUMA 57
// Shading change detection mip level setting, value must be in the range [FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_0, FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_12]
#define FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_SHADING_CHANGE FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_4
#define FFX_FSR2_SHADING_CHANGE_MIP_LEVEL (FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE_MIPMAP_SHADING_CHANGE - FFX_FSR2_RESOURCE_IDENTIFIER_SCENE_LUMINANCE)
#define FFX_FSR2_RESOURCE_IDENTIFIER_COUNT 58
#define FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_FSR2 0
#define FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_SPD 1
#define FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_RCAS 2
#define FFX_FSR2_CONSTANTBUFFER_IDENTIFIER_GENREACTIVE 3
#define FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_TONEMAP 1
#define FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_INVERSETONEMAP 2
#define FFX_FSR2_AUTOREACTIVEFLAGS_APPLY_THRESHOLD 4
#define FFX_FSR2_AUTOREACTIVEFLAGS_USE_COMPONENTS_MAX 8
#endif // #if defined(FFX_CPU) || defined(FFX_GPU)
#endif //!defined( FFX_FSR2_RESOURCES_H )

605
thirdparty/amd-fsr2/shaders/ffx_fsr2_sample.h vendored Normal file
View File

@ -0,0 +1,605 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#ifndef FFX_FSR2_SAMPLE_H
#define FFX_FSR2_SAMPLE_H
// suppress warnings
#ifdef FFX_HLSL
#pragma warning(disable: 4008) // potentially divide by zero
#endif //FFX_HLSL
struct FetchedBilinearSamples {
FfxFloat32x4 fColor00;
FfxFloat32x4 fColor10;
FfxFloat32x4 fColor01;
FfxFloat32x4 fColor11;
};
struct FetchedBicubicSamples {
FfxFloat32x4 fColor00;
FfxFloat32x4 fColor10;
FfxFloat32x4 fColor20;
FfxFloat32x4 fColor30;
FfxFloat32x4 fColor01;
FfxFloat32x4 fColor11;
FfxFloat32x4 fColor21;
FfxFloat32x4 fColor31;
FfxFloat32x4 fColor02;
FfxFloat32x4 fColor12;
FfxFloat32x4 fColor22;
FfxFloat32x4 fColor32;
FfxFloat32x4 fColor03;
FfxFloat32x4 fColor13;
FfxFloat32x4 fColor23;
FfxFloat32x4 fColor33;
};
#if FFX_HALF
struct FetchedBilinearSamplesMin16 {
FFX_MIN16_F4 fColor00;
FFX_MIN16_F4 fColor10;
FFX_MIN16_F4 fColor01;
FFX_MIN16_F4 fColor11;
};
struct FetchedBicubicSamplesMin16 {
FFX_MIN16_F4 fColor00;
FFX_MIN16_F4 fColor10;
FFX_MIN16_F4 fColor20;
FFX_MIN16_F4 fColor30;
FFX_MIN16_F4 fColor01;
FFX_MIN16_F4 fColor11;
FFX_MIN16_F4 fColor21;
FFX_MIN16_F4 fColor31;
FFX_MIN16_F4 fColor02;
FFX_MIN16_F4 fColor12;
FFX_MIN16_F4 fColor22;
FFX_MIN16_F4 fColor32;
FFX_MIN16_F4 fColor03;
FFX_MIN16_F4 fColor13;
FFX_MIN16_F4 fColor23;
FFX_MIN16_F4 fColor33;
};
#else //FFX_HALF
#define FetchedBicubicSamplesMin16 FetchedBicubicSamples
#define FetchedBilinearSamplesMin16 FetchedBilinearSamples
#endif //FFX_HALF
FfxFloat32x4 Linear(FfxFloat32x4 A, FfxFloat32x4 B, FfxFloat32 t)
{
return A + (B - A) * t;
}
FfxFloat32x4 Bilinear(FetchedBilinearSamples BilinearSamples, FfxFloat32x2 fPxFrac)
{
FfxFloat32x4 fColorX0 = Linear(BilinearSamples.fColor00, BilinearSamples.fColor10, fPxFrac.x);
FfxFloat32x4 fColorX1 = Linear(BilinearSamples.fColor01, BilinearSamples.fColor11, fPxFrac.x);
FfxFloat32x4 fColorXY = Linear(fColorX0, fColorX1, fPxFrac.y);
return fColorXY;
}
#if FFX_HALF
FFX_MIN16_F4 Linear(FFX_MIN16_F4 A, FFX_MIN16_F4 B, FFX_MIN16_F t)
{
return A + (B - A) * t;
}
FFX_MIN16_F4 Bilinear(FetchedBilinearSamplesMin16 BilinearSamples, FFX_MIN16_F2 fPxFrac)
{
FFX_MIN16_F4 fColorX0 = Linear(BilinearSamples.fColor00, BilinearSamples.fColor10, fPxFrac.x);
FFX_MIN16_F4 fColorX1 = Linear(BilinearSamples.fColor01, BilinearSamples.fColor11, fPxFrac.x);
FFX_MIN16_F4 fColorXY = Linear(fColorX0, fColorX1, fPxFrac.y);
return fColorXY;
}
#endif
FfxFloat32 Lanczos2NoClamp(FfxFloat32 x)
{
const FfxFloat32 PI = 3.141592653589793f; // TODO: share SDK constants
return abs(x) < FSR2_EPSILON ? 1.f : (sin(PI * x) / (PI * x)) * (sin(0.5f * PI * x) / (0.5f * PI * x));
}
FfxFloat32 Lanczos2(FfxFloat32 x)
{
x = ffxMin(abs(x), 2.0f);
return Lanczos2NoClamp(x);
}
#if FFX_HALF
#if 0
FFX_MIN16_F Lanczos2NoClamp(FFX_MIN16_F x)
{
const FFX_MIN16_F PI = FFX_MIN16_F(3.141592653589793f); // TODO: share SDK constants
return abs(x) < FFX_MIN16_F(FSR2_EPSILON) ? FFX_MIN16_F(1.f) : (sin(PI * x) / (PI * x)) * (sin(FFX_MIN16_F(0.5f) * PI * x) / (FFX_MIN16_F(0.5f) * PI * x));
}
#endif
FFX_MIN16_F Lanczos2(FFX_MIN16_F x)
{
x = ffxMin(abs(x), FFX_MIN16_F(2.0f));
return FFX_MIN16_F(Lanczos2NoClamp(x));
}
#endif //FFX_HALF
// FSR1 lanczos approximation. Input is x*x and must be <= 4.
FfxFloat32 Lanczos2ApproxSqNoClamp(FfxFloat32 x2)
{
FfxFloat32 a = (2.0f / 5.0f) * x2 - 1;
FfxFloat32 b = (1.0f / 4.0f) * x2 - 1;
return ((25.0f / 16.0f) * a * a - (25.0f / 16.0f - 1)) * (b * b);
}
#if FFX_HALF
FFX_MIN16_F Lanczos2ApproxSqNoClamp(FFX_MIN16_F x2)
{
FFX_MIN16_F a = FFX_MIN16_F(2.0f / 5.0f) * x2 - FFX_MIN16_F(1);
FFX_MIN16_F b = FFX_MIN16_F(1.0f / 4.0f) * x2 - FFX_MIN16_F(1);
return (FFX_MIN16_F(25.0f / 16.0f) * a * a - FFX_MIN16_F(25.0f / 16.0f - 1)) * (b * b);
}
#endif //FFX_HALF
FfxFloat32 Lanczos2ApproxSq(FfxFloat32 x2)
{
x2 = ffxMin(x2, 4.0f);
return Lanczos2ApproxSqNoClamp(x2);
}
#if FFX_HALF
FFX_MIN16_F Lanczos2ApproxSq(FFX_MIN16_F x2)
{
x2 = ffxMin(x2, FFX_MIN16_F(4.0f));
return Lanczos2ApproxSqNoClamp(x2);
}
#endif //FFX_HALF
FfxFloat32 Lanczos2ApproxNoClamp(FfxFloat32 x)
{
return Lanczos2ApproxSqNoClamp(x * x);
}
#if FFX_HALF
FFX_MIN16_F Lanczos2ApproxNoClamp(FFX_MIN16_F x)
{
return Lanczos2ApproxSqNoClamp(x * x);
}
#endif //FFX_HALF
FfxFloat32 Lanczos2Approx(FfxFloat32 x)
{
return Lanczos2ApproxSq(x * x);
}
#if FFX_HALF
FFX_MIN16_F Lanczos2Approx(FFX_MIN16_F x)
{
return Lanczos2ApproxSq(x * x);
}
#endif //FFX_HALF
FfxFloat32 Lanczos2_UseLUT(FfxFloat32 x)
{
return SampleLanczos2Weight(abs(x));
}
#if FFX_HALF
FFX_MIN16_F Lanczos2_UseLUT(FFX_MIN16_F x)
{
return FFX_MIN16_F(SampleLanczos2Weight(abs(x)));
}
#endif //FFX_HALF
FfxFloat32x4 Lanczos2_UseLUT(FfxFloat32x4 fColor0, FfxFloat32x4 fColor1, FfxFloat32x4 fColor2, FfxFloat32x4 fColor3, FfxFloat32 t)
{
FfxFloat32 fWeight0 = Lanczos2_UseLUT(-1.f - t);
FfxFloat32 fWeight1 = Lanczos2_UseLUT(-0.f - t);
FfxFloat32 fWeight2 = Lanczos2_UseLUT(+1.f - t);
FfxFloat32 fWeight3 = Lanczos2_UseLUT(+2.f - t);
return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3);
}
#if FFX_HALF
FFX_MIN16_F4 Lanczos2_UseLUT(FFX_MIN16_F4 fColor0, FFX_MIN16_F4 fColor1, FFX_MIN16_F4 fColor2, FFX_MIN16_F4 fColor3, FFX_MIN16_F t)
{
FFX_MIN16_F fWeight0 = Lanczos2_UseLUT(FFX_MIN16_F(-1.f) - t);
FFX_MIN16_F fWeight1 = Lanczos2_UseLUT(FFX_MIN16_F(-0.f) - t);
FFX_MIN16_F fWeight2 = Lanczos2_UseLUT(FFX_MIN16_F(+1.f) - t);
FFX_MIN16_F fWeight3 = Lanczos2_UseLUT(FFX_MIN16_F(+2.f) - t);
return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3);
}
#endif
FfxFloat32x4 Lanczos2(FfxFloat32x4 fColor0, FfxFloat32x4 fColor1, FfxFloat32x4 fColor2, FfxFloat32x4 fColor3, FfxFloat32 t)
{
FfxFloat32 fWeight0 = Lanczos2(-1.f - t);
FfxFloat32 fWeight1 = Lanczos2(-0.f - t);
FfxFloat32 fWeight2 = Lanczos2(+1.f - t);
FfxFloat32 fWeight3 = Lanczos2(+2.f - t);
return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3);
}
FfxFloat32x4 Lanczos2(FetchedBicubicSamples Samples, FfxFloat32x2 fPxFrac)
{
FfxFloat32x4 fColorX0 = Lanczos2(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x);
FfxFloat32x4 fColorX1 = Lanczos2(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x);
FfxFloat32x4 fColorX2 = Lanczos2(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x);
FfxFloat32x4 fColorX3 = Lanczos2(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x);
FfxFloat32x4 fColorXY = Lanczos2(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y);
// Deringing
// TODO: only use 4 by checking jitter
const FfxInt32 iDeringingSampleCount = 4;
const FfxFloat32x4 fDeringingSamples[4] = {
Samples.fColor11,
Samples.fColor21,
Samples.fColor12,
Samples.fColor22,
};
FfxFloat32x4 fDeringingMin = fDeringingSamples[0];
FfxFloat32x4 fDeringingMax = fDeringingSamples[0];
FFX_UNROLL
for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) {
fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]);
fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]);
}
fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax);
return fColorXY;
}
#if FFX_HALF
FFX_MIN16_F4 Lanczos2(FFX_MIN16_F4 fColor0, FFX_MIN16_F4 fColor1, FFX_MIN16_F4 fColor2, FFX_MIN16_F4 fColor3, FFX_MIN16_F t)
{
FFX_MIN16_F fWeight0 = Lanczos2(FFX_MIN16_F(-1.f) - t);
FFX_MIN16_F fWeight1 = Lanczos2(FFX_MIN16_F(-0.f) - t);
FFX_MIN16_F fWeight2 = Lanczos2(FFX_MIN16_F(+1.f) - t);
FFX_MIN16_F fWeight3 = Lanczos2(FFX_MIN16_F(+2.f) - t);
return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3);
}
FFX_MIN16_F4 Lanczos2(FetchedBicubicSamplesMin16 Samples, FFX_MIN16_F2 fPxFrac)
{
FFX_MIN16_F4 fColorX0 = Lanczos2(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x);
FFX_MIN16_F4 fColorX1 = Lanczos2(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x);
FFX_MIN16_F4 fColorX2 = Lanczos2(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x);
FFX_MIN16_F4 fColorX3 = Lanczos2(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x);
FFX_MIN16_F4 fColorXY = Lanczos2(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y);
// Deringing
// TODO: only use 4 by checking jitter
const FfxInt32 iDeringingSampleCount = 4;
const FFX_MIN16_F4 fDeringingSamples[4] = {
Samples.fColor11,
Samples.fColor21,
Samples.fColor12,
Samples.fColor22,
};
FFX_MIN16_F4 fDeringingMin = fDeringingSamples[0];
FFX_MIN16_F4 fDeringingMax = fDeringingSamples[0];
FFX_UNROLL
for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex)
{
fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]);
fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]);
}
fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax);
return fColorXY;
}
#endif //FFX_HALF
FfxFloat32x4 Lanczos2LUT(FetchedBicubicSamples Samples, FfxFloat32x2 fPxFrac)
{
FfxFloat32x4 fColorX0 = Lanczos2_UseLUT(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x);
FfxFloat32x4 fColorX1 = Lanczos2_UseLUT(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x);
FfxFloat32x4 fColorX2 = Lanczos2_UseLUT(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x);
FfxFloat32x4 fColorX3 = Lanczos2_UseLUT(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x);
FfxFloat32x4 fColorXY = Lanczos2_UseLUT(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y);
// Deringing
// TODO: only use 4 by checking jitter
const FfxInt32 iDeringingSampleCount = 4;
const FfxFloat32x4 fDeringingSamples[4] = {
Samples.fColor11,
Samples.fColor21,
Samples.fColor12,
Samples.fColor22,
};
FfxFloat32x4 fDeringingMin = fDeringingSamples[0];
FfxFloat32x4 fDeringingMax = fDeringingSamples[0];
FFX_UNROLL
for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex) {
fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]);
fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]);
}
fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax);
return fColorXY;
}
#if FFX_HALF
FFX_MIN16_F4 Lanczos2LUT(FetchedBicubicSamplesMin16 Samples, FFX_MIN16_F2 fPxFrac)
{
FFX_MIN16_F4 fColorX0 = Lanczos2_UseLUT(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x);
FFX_MIN16_F4 fColorX1 = Lanczos2_UseLUT(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x);
FFX_MIN16_F4 fColorX2 = Lanczos2_UseLUT(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x);
FFX_MIN16_F4 fColorX3 = Lanczos2_UseLUT(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x);
FFX_MIN16_F4 fColorXY = Lanczos2_UseLUT(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y);
// Deringing
// TODO: only use 4 by checking jitter
const FfxInt32 iDeringingSampleCount = 4;
const FFX_MIN16_F4 fDeringingSamples[4] = {
Samples.fColor11,
Samples.fColor21,
Samples.fColor12,
Samples.fColor22,
};
FFX_MIN16_F4 fDeringingMin = fDeringingSamples[0];
FFX_MIN16_F4 fDeringingMax = fDeringingSamples[0];
FFX_UNROLL
for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex)
{
fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]);
fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]);
}
fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax);
return fColorXY;
}
#endif //FFX_HALF
FfxFloat32x4 Lanczos2Approx(FfxFloat32x4 fColor0, FfxFloat32x4 fColor1, FfxFloat32x4 fColor2, FfxFloat32x4 fColor3, FfxFloat32 t)
{
FfxFloat32 fWeight0 = Lanczos2ApproxNoClamp(-1.f - t);
FfxFloat32 fWeight1 = Lanczos2ApproxNoClamp(-0.f - t);
FfxFloat32 fWeight2 = Lanczos2ApproxNoClamp(+1.f - t);
FfxFloat32 fWeight3 = Lanczos2ApproxNoClamp(+2.f - t);
return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3);
}
#if FFX_HALF
FFX_MIN16_F4 Lanczos2Approx(FFX_MIN16_F4 fColor0, FFX_MIN16_F4 fColor1, FFX_MIN16_F4 fColor2, FFX_MIN16_F4 fColor3, FFX_MIN16_F t)
{
FFX_MIN16_F fWeight0 = Lanczos2ApproxNoClamp(FFX_MIN16_F(-1.f) - t);
FFX_MIN16_F fWeight1 = Lanczos2ApproxNoClamp(FFX_MIN16_F(-0.f) - t);
FFX_MIN16_F fWeight2 = Lanczos2ApproxNoClamp(FFX_MIN16_F(+1.f) - t);
FFX_MIN16_F fWeight3 = Lanczos2ApproxNoClamp(FFX_MIN16_F(+2.f) - t);
return (fWeight0 * fColor0 + fWeight1 * fColor1 + fWeight2 * fColor2 + fWeight3 * fColor3) / (fWeight0 + fWeight1 + fWeight2 + fWeight3);
}
#endif //FFX_HALF
FfxFloat32x4 Lanczos2Approx(FetchedBicubicSamples Samples, FfxFloat32x2 fPxFrac)
{
FfxFloat32x4 fColorX0 = Lanczos2Approx(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x);
FfxFloat32x4 fColorX1 = Lanczos2Approx(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x);
FfxFloat32x4 fColorX2 = Lanczos2Approx(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x);
FfxFloat32x4 fColorX3 = Lanczos2Approx(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x);
FfxFloat32x4 fColorXY = Lanczos2Approx(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y);
// Deringing
// TODO: only use 4 by checking jitter
const FfxInt32 iDeringingSampleCount = 4;
const FfxFloat32x4 fDeringingSamples[4] = {
Samples.fColor11,
Samples.fColor21,
Samples.fColor12,
Samples.fColor22,
};
FfxFloat32x4 fDeringingMin = fDeringingSamples[0];
FfxFloat32x4 fDeringingMax = fDeringingSamples[0];
FFX_UNROLL
for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex)
{
fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]);
fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]);
}
fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax);
return fColorXY;
}
#if FFX_HALF
FFX_MIN16_F4 Lanczos2Approx(FetchedBicubicSamplesMin16 Samples, FFX_MIN16_F2 fPxFrac)
{
FFX_MIN16_F4 fColorX0 = Lanczos2Approx(Samples.fColor00, Samples.fColor10, Samples.fColor20, Samples.fColor30, fPxFrac.x);
FFX_MIN16_F4 fColorX1 = Lanczos2Approx(Samples.fColor01, Samples.fColor11, Samples.fColor21, Samples.fColor31, fPxFrac.x);
FFX_MIN16_F4 fColorX2 = Lanczos2Approx(Samples.fColor02, Samples.fColor12, Samples.fColor22, Samples.fColor32, fPxFrac.x);
FFX_MIN16_F4 fColorX3 = Lanczos2Approx(Samples.fColor03, Samples.fColor13, Samples.fColor23, Samples.fColor33, fPxFrac.x);
FFX_MIN16_F4 fColorXY = Lanczos2Approx(fColorX0, fColorX1, fColorX2, fColorX3, fPxFrac.y);
// Deringing
// TODO: only use 4 by checking jitter
const FfxInt32 iDeringingSampleCount = 4;
const FFX_MIN16_F4 fDeringingSamples[4] = {
Samples.fColor11,
Samples.fColor21,
Samples.fColor12,
Samples.fColor22,
};
FFX_MIN16_F4 fDeringingMin = fDeringingSamples[0];
FFX_MIN16_F4 fDeringingMax = fDeringingSamples[0];
FFX_UNROLL
for (FfxInt32 iSampleIndex = 1; iSampleIndex < iDeringingSampleCount; ++iSampleIndex)
{
fDeringingMin = ffxMin(fDeringingMin, fDeringingSamples[iSampleIndex]);
fDeringingMax = ffxMax(fDeringingMax, fDeringingSamples[iSampleIndex]);
}
fColorXY = clamp(fColorXY, fDeringingMin, fDeringingMax);
return fColorXY;
}
#endif
// Clamp by offset direction. Assuming iPxSample is already in range and iPxOffset is compile time constant.
FfxInt32x2 ClampCoord(FfxInt32x2 iPxSample, FfxInt32x2 iPxOffset, FfxInt32x2 iTextureSize)
{
FfxInt32x2 result = iPxSample + iPxOffset;
result.x = (iPxOffset.x < 0) ? ffxMax(result.x, 0) : result.x;
result.x = (iPxOffset.x > 0) ? ffxMin(result.x, iTextureSize.x - 1) : result.x;
result.y = (iPxOffset.y < 0) ? ffxMax(result.y, 0) : result.y;
result.y = (iPxOffset.y > 0) ? ffxMin(result.y, iTextureSize.y - 1) : result.y;
return result;
}
#if FFX_HALF
FFX_MIN16_I2 ClampCoord(FFX_MIN16_I2 iPxSample, FFX_MIN16_I2 iPxOffset, FFX_MIN16_I2 iTextureSize)
{
FFX_MIN16_I2 result = iPxSample + iPxOffset;
result.x = (iPxOffset.x < FFX_MIN16_I(0)) ? ffxMax(result.x, FFX_MIN16_I(0)) : result.x;
result.x = (iPxOffset.x > FFX_MIN16_I(0)) ? ffxMin(result.x, iTextureSize.x - FFX_MIN16_I(1)) : result.x;
result.y = (iPxOffset.y < FFX_MIN16_I(0)) ? ffxMax(result.y, FFX_MIN16_I(0)) : result.y;
result.y = (iPxOffset.y > FFX_MIN16_I(0)) ? ffxMin(result.y, iTextureSize.y - FFX_MIN16_I(1)) : result.y;
return result;
}
#endif //FFX_HALF
#define DeclareCustomFetchBicubicSamplesWithType(SampleType, TextureType, AddrType, Name, LoadTexture) \
SampleType Name(AddrType iPxSample, AddrType iTextureSize) \
{ \
SampleType Samples; \
\
Samples.fColor00 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, -1), iTextureSize))); \
Samples.fColor10 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, -1), iTextureSize))); \
Samples.fColor20 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, -1), iTextureSize))); \
Samples.fColor30 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, -1), iTextureSize))); \
\
Samples.fColor01 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, +0), iTextureSize))); \
Samples.fColor11 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +0), iTextureSize))); \
Samples.fColor21 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +0), iTextureSize))); \
Samples.fColor31 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, +0), iTextureSize))); \
\
Samples.fColor02 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, +1), iTextureSize))); \
Samples.fColor12 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +1), iTextureSize))); \
Samples.fColor22 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +1), iTextureSize))); \
Samples.fColor32 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, +1), iTextureSize))); \
\
Samples.fColor03 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(-1, +2), iTextureSize))); \
Samples.fColor13 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +2), iTextureSize))); \
Samples.fColor23 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +2), iTextureSize))); \
Samples.fColor33 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+2, +2), iTextureSize))); \
\
return Samples; \
}
#define DeclareCustomFetchBicubicSamples(Name, LoadTexture) \
DeclareCustomFetchBicubicSamplesWithType(FetchedBicubicSamples, FfxFloat32x4, FfxInt32x2, Name, LoadTexture)
#define DeclareCustomFetchBicubicSamplesMin16(Name, LoadTexture) \
DeclareCustomFetchBicubicSamplesWithType(FetchedBicubicSamplesMin16, FFX_MIN16_F4, FfxInt32x2, Name, LoadTexture)
#define DeclareCustomFetchBilinearSamplesWithType(SampleType, TextureType,AddrType, Name, LoadTexture) \
SampleType Name(AddrType iPxSample, AddrType iTextureSize) \
{ \
SampleType Samples; \
Samples.fColor00 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +0), iTextureSize))); \
Samples.fColor10 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +0), iTextureSize))); \
Samples.fColor01 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+0, +1), iTextureSize))); \
Samples.fColor11 = TextureType(LoadTexture(ClampCoord(iPxSample, AddrType(+1, +1), iTextureSize))); \
return Samples; \
}
#define DeclareCustomFetchBilinearSamples(Name, LoadTexture) \
DeclareCustomFetchBilinearSamplesWithType(FetchedBilinearSamples, FfxFloat32x4, FfxInt32x2, Name, LoadTexture)
#define DeclareCustomFetchBilinearSamplesMin16(Name, LoadTexture) \
DeclareCustomFetchBilinearSamplesWithType(FetchedBilinearSamplesMin16, FFX_MIN16_F4, FfxInt32x2, Name, LoadTexture)
// BE CAREFUL: there is some precision issues and (3253, 125) leading to (3252.9989778, 125.001102)
// is common, so iPxSample can "jitter"
#define DeclareCustomTextureSample(Name, InterpolateSamples, FetchSamples) \
FfxFloat32x4 Name(FfxFloat32x2 fUvSample, FfxInt32x2 iTextureSize) \
{ \
FfxFloat32x2 fPxSample = (fUvSample * FfxFloat32x2(iTextureSize)) - FfxFloat32x2(0.5f, 0.5f); \
/* Clamp base coords */ \
fPxSample.x = ffxMax(0.0f, ffxMin(FfxFloat32(iTextureSize.x), fPxSample.x)); \
fPxSample.y = ffxMax(0.0f, ffxMin(FfxFloat32(iTextureSize.y), fPxSample.y)); \
/* */ \
FfxInt32x2 iPxSample = FfxInt32x2(floor(fPxSample)); \
FfxFloat32x2 fPxFrac = ffxFract(fPxSample); \
FfxFloat32x4 fColorXY = FfxFloat32x4(InterpolateSamples(FetchSamples(iPxSample, iTextureSize), fPxFrac)); \
return fColorXY; \
}
#define DeclareCustomTextureSampleMin16(Name, InterpolateSamples, FetchSamples) \
FFX_MIN16_F4 Name(FfxFloat32x2 fUvSample, FfxInt32x2 iTextureSize) \
{ \
FfxFloat32x2 fPxSample = (fUvSample * FfxFloat32x2(iTextureSize)) - FfxFloat32x2(0.5f, 0.5f); \
/* Clamp base coords */ \
fPxSample.x = ffxMax(0.0f, ffxMin(FfxFloat32(iTextureSize.x), fPxSample.x)); \
fPxSample.y = ffxMax(0.0f, ffxMin(FfxFloat32(iTextureSize.y), fPxSample.y)); \
/* */ \
FfxInt32x2 iPxSample = FfxInt32x2(floor(fPxSample)); \
FFX_MIN16_F2 fPxFrac = FFX_MIN16_F2(ffxFract(fPxSample)); \
FFX_MIN16_F4 fColorXY = FFX_MIN16_F4(InterpolateSamples(FetchSamples(iPxSample, iTextureSize), fPxFrac)); \
return fColorXY; \
}
#define FFX_FSR2_CONCAT_ID(x, y) x ## y
#define FFX_FSR2_CONCAT(x, y) FFX_FSR2_CONCAT_ID(x, y)
#define FFX_FSR2_SAMPLER_1D_0 Lanczos2
#define FFX_FSR2_SAMPLER_1D_1 Lanczos2LUT
#define FFX_FSR2_SAMPLER_1D_2 Lanczos2Approx
#define FFX_FSR2_GET_LANCZOS_SAMPLER1D(x) FFX_FSR2_CONCAT(FFX_FSR2_SAMPLER_1D_, x)
#endif //!defined( FFX_FSR2_SAMPLE_H )

250
thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen.h vendored Normal file
View File

@ -0,0 +1,250 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#define USE_YCOCG 1
#define fAutogenEpsilon 0.01f
// EXPERIMENTAL
FFX_MIN16_F ComputeAutoTC_01(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx)
{
FfxFloat32x3 colorPreAlpha = LoadOpaqueOnly(uDispatchThreadId);
FfxFloat32x3 colorPostAlpha = LoadInputColor(uDispatchThreadId);
FfxFloat32x3 colorPrevPreAlpha = LoadPrevPreAlpha(iPrevIdx);
FfxFloat32x3 colorPrevPostAlpha = LoadPrevPostAlpha(iPrevIdx);
#if USE_YCOCG
colorPreAlpha = RGBToYCoCg(colorPreAlpha);
colorPostAlpha = RGBToYCoCg(colorPostAlpha);
colorPrevPreAlpha = RGBToYCoCg(colorPrevPreAlpha);
colorPrevPostAlpha = RGBToYCoCg(colorPrevPostAlpha);
#endif
FfxFloat32x3 colorDeltaCurr = colorPostAlpha - colorPreAlpha;
FfxFloat32x3 colorDeltaPrev = colorPrevPostAlpha - colorPrevPreAlpha;
bool hasAlpha = any(FFX_GREATER_THAN(abs(colorDeltaCurr), FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon)));
bool hadAlpha = any(FFX_GREATER_THAN(abs(colorDeltaPrev), FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon)));
FfxFloat32x3 X = colorPreAlpha;
FfxFloat32x3 Y = colorPostAlpha;
FfxFloat32x3 Z = colorPrevPreAlpha;
FfxFloat32x3 W = colorPrevPostAlpha;
FFX_MIN16_F retVal = FFX_MIN16_F(ffxSaturate(dot(abs(abs(Y - X) - abs(W - Z)), FfxFloat32x3(1, 1, 1))));
// cleanup very small values
retVal = (retVal < getTcThreshold()) ? FFX_MIN16_F(0.0f) : FFX_MIN16_F(1.f);
return retVal;
}
// works ok: thin edges
FFX_MIN16_F ComputeAutoTC_02(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx)
{
FfxFloat32x3 colorPreAlpha = LoadOpaqueOnly(uDispatchThreadId);
FfxFloat32x3 colorPostAlpha = LoadInputColor(uDispatchThreadId);
FfxFloat32x3 colorPrevPreAlpha = LoadPrevPreAlpha(iPrevIdx);
FfxFloat32x3 colorPrevPostAlpha = LoadPrevPostAlpha(iPrevIdx);
#if USE_YCOCG
colorPreAlpha = RGBToYCoCg(colorPreAlpha);
colorPostAlpha = RGBToYCoCg(colorPostAlpha);
colorPrevPreAlpha = RGBToYCoCg(colorPrevPreAlpha);
colorPrevPostAlpha = RGBToYCoCg(colorPrevPostAlpha);
#endif
FfxFloat32x3 colorDelta = colorPostAlpha - colorPreAlpha;
FfxFloat32x3 colorPrevDelta = colorPrevPostAlpha - colorPrevPreAlpha;
bool hasAlpha = any(FFX_GREATER_THAN(abs(colorDelta), FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon)));
bool hadAlpha = any(FFX_GREATER_THAN(abs(colorPrevDelta), FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon)));
FfxFloat32x3 delta = colorPostAlpha - colorPreAlpha; //prev+1*d = post => d = color, alpha =
FfxFloat32x3 deltaPrev = colorPrevPostAlpha - colorPrevPreAlpha;
FfxFloat32x3 X = colorPrevPreAlpha;
FfxFloat32x3 N = colorPreAlpha - colorPrevPreAlpha;
FfxFloat32x3 YAminusXA = colorPrevPostAlpha - colorPrevPreAlpha;
FfxFloat32x3 NminusNA = colorPostAlpha - colorPrevPostAlpha;
FfxFloat32x3 A = (hasAlpha || hadAlpha) ? NminusNA / max(FfxFloat32x3(fAutogenEpsilon, fAutogenEpsilon, fAutogenEpsilon), N) : FfxFloat32x3(0, 0, 0);
FFX_MIN16_F retVal = FFX_MIN16_F( max(max(A.x, A.y), A.z) );
// only pixels that have significantly changed in color shuold be considered
retVal = ffxSaturate(retVal * FFX_MIN16_F(length(colorPostAlpha - colorPrevPostAlpha)) );
return retVal;
}
// This function computes the TransparencyAndComposition mask:
// This mask indicates pixels that should discard locks and apply color clamping.
//
// Typically this is the case for translucent pixels (that don't write depth values) or pixels where the correctness of
// the MVs can not be guaranteed (e.g. procedutal movement or vegetation that does not have MVs to reduce the cost during rasterization)
// Also, large changes in color due to changed lighting should be marked to remove locks on pixels with "old" lighting.
//
// This function takes a opaque only and a final texture and uses internal copies of those textures from the last frame.
// The function tries to determine where the color changes between opaque only and final image to determine the pixels that use transparency.
// Also it uses the previous frames and detects where the use of transparency changed to mark those pixels.
// Additionally it marks pixels where the color changed significantly in the opaque only image, e.g. due to lighting or texture animation.
//
// In the final step it stores the current textures in internal textures for the next frame
FFX_MIN16_F ComputeTransparencyAndComposition(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx)
{
FFX_MIN16_F retVal = ComputeAutoTC_02(uDispatchThreadId, iPrevIdx);
// [branch]
if (retVal > FFX_MIN16_F(0.01f))
{
retVal = ComputeAutoTC_01(uDispatchThreadId, iPrevIdx);
}
return retVal;
}
float computeSolidEdge(FFX_MIN16_I2 curPos, FFX_MIN16_I2 prevPos)
{
float lum[9];
int i = 0;
for (int y = -1; y < 2; ++y)
{
for (int x = -1; x < 2; ++x)
{
FfxFloat32x3 curCol = LoadOpaqueOnly(curPos + FFX_MIN16_I2(x, y)).rgb;
FfxFloat32x3 prevCol = LoadPrevPreAlpha(prevPos + FFX_MIN16_I2(x, y)).rgb;
lum[i++] = length(curCol - prevCol);
}
}
//float gradX = abs(lum[3] - lum[4]) + abs(lum[5] - lum[4]);
//float gradY = abs(lum[1] - lum[4]) + abs(lum[7] - lum[4]);
//return sqrt(gradX * gradX + gradY * gradY);
float gradX = abs(lum[3] - lum[4]) * abs(lum[5] - lum[4]);
float gradY = abs(lum[1] - lum[4]) * abs(lum[7] - lum[4]);
return sqrt(sqrt(gradX * gradY));
}
float computeAlphaEdge(FFX_MIN16_I2 curPos, FFX_MIN16_I2 prevPos)
{
float lum[9];
int i = 0;
for (int y = -1; y < 2; ++y)
{
for (int x = -1; x < 2; ++x)
{
FfxFloat32x3 curCol = abs(LoadInputColor(curPos + FFX_MIN16_I2(x, y)).rgb - LoadOpaqueOnly(curPos + FFX_MIN16_I2(x, y)).rgb);
FfxFloat32x3 prevCol = abs(LoadPrevPostAlpha(prevPos + FFX_MIN16_I2(x, y)).rgb - LoadPrevPreAlpha(prevPos + FFX_MIN16_I2(x, y)).rgb);
lum[i++] = length(curCol - prevCol);
}
}
//float gradX = abs(lum[3] - lum[4]) + abs(lum[5] - lum[4]);
//float gradY = abs(lum[1] - lum[4]) + abs(lum[7] - lum[4]);
//return sqrt(gradX * gradX + gradY * gradY);
float gradX = abs(lum[3] - lum[4]) * abs(lum[5] - lum[4]);
float gradY = abs(lum[1] - lum[4]) * abs(lum[7] - lum[4]);
return sqrt(sqrt(gradX * gradY));
}
FFX_MIN16_F ComputeAabbOverlap(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx)
{
FFX_MIN16_F retVal = FFX_MIN16_F(0.f);
FfxFloat32x2 fMotionVector = LoadInputMotionVector(uDispatchThreadId);
FfxFloat32x3 colorPreAlpha = LoadOpaqueOnly(uDispatchThreadId);
FfxFloat32x3 colorPostAlpha = LoadInputColor(uDispatchThreadId);
FfxFloat32x3 colorPrevPreAlpha = LoadPrevPreAlpha(iPrevIdx);
FfxFloat32x3 colorPrevPostAlpha = LoadPrevPostAlpha(iPrevIdx);
#if USE_YCOCG
colorPreAlpha = RGBToYCoCg(colorPreAlpha);
colorPostAlpha = RGBToYCoCg(colorPostAlpha);
colorPrevPreAlpha = RGBToYCoCg(colorPrevPreAlpha);
colorPrevPostAlpha = RGBToYCoCg(colorPrevPostAlpha);
#endif
FfxFloat32x3 minPrev = FFX_MIN16_F3(+1000.f, +1000.f, +1000.f);
FfxFloat32x3 maxPrev = FFX_MIN16_F3(-1000.f, -1000.f, -1000.f);
for (int y = -1; y < 2; ++y)
{
for (int x = -1; x < 2; ++x)
{
FfxFloat32x3 W = LoadPrevPostAlpha(iPrevIdx + FFX_MIN16_I2(x, y));
#if USE_YCOCG
W = RGBToYCoCg(W);
#endif
minPrev = min(minPrev, W);
maxPrev = max(maxPrev, W);
}
}
// instead of computing the overlap: simply count how many samples are outside
// set reactive based on that
FFX_MIN16_F count = FFX_MIN16_F(0.f);
for (int y = -1; y < 2; ++y)
{
for (int x = -1; x < 2; ++x)
{
FfxFloat32x3 Y = LoadInputColor(uDispatchThreadId + FFX_MIN16_I2(x, y));
#if USE_YCOCG
Y = RGBToYCoCg(Y);
#endif
count += ((Y.x < minPrev.x) || (Y.x > maxPrev.x)) ? FFX_MIN16_F(1.f) : FFX_MIN16_F(0.f);
count += ((Y.y < minPrev.y) || (Y.y > maxPrev.y)) ? FFX_MIN16_F(1.f) : FFX_MIN16_F(0.f);
count += ((Y.z < minPrev.z) || (Y.z > maxPrev.z)) ? FFX_MIN16_F(1.f) : FFX_MIN16_F(0.f);
}
}
retVal = count / FFX_MIN16_F(27.f);
return retVal;
}
// This function computes the Reactive mask:
// We want pixels marked where the alpha portion of the frame changes a lot between neighbours
// Those pixels are expected to change quickly between frames, too. (e.g. small particles, reflections on curved surfaces...)
// As a result history would not be trustworthy.
// On the other hand we don't want pixels marked where pre-alpha has a large differnce, since those would profit from accumulation
// For mirrors we may assume the pre-alpha is pretty uniform color.
//
// This works well generally, but also marks edge pixels
FFX_MIN16_F ComputeReactive(FFX_MIN16_I2 uDispatchThreadId, FFX_MIN16_I2 iPrevIdx)
{
// we only get here if alpha has a significant contribution and has changed since last frame.
FFX_MIN16_F retVal = FFX_MIN16_F(0.f);
// mark pixels with huge variance in alpha as reactive
FFX_MIN16_F alphaEdge = FFX_MIN16_F(computeAlphaEdge(uDispatchThreadId, iPrevIdx));
FFX_MIN16_F opaqueEdge = FFX_MIN16_F(computeSolidEdge(uDispatchThreadId, iPrevIdx));
retVal = ffxSaturate(alphaEdge - opaqueEdge);
// the above also marks edge pixels due to jitter, so we need to cancel those out
return retVal;
}

122
thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen_pass.glsl vendored Normal file
View File

@ -0,0 +1,122 @@
// This file is part of the FidelityFX SDK.
//
// Copyright (c) 2022-2023 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
#extension GL_GOOGLE_include_directive : require
#extension GL_EXT_samplerless_texture_functions : require
#define FSR2_BIND_SRV_INPUT_OPAQUE_ONLY 0
#define FSR2_BIND_SRV_INPUT_COLOR 1
#define FSR2_BIND_SRV_INPUT_MOTION_VECTORS 2
#define FSR2_BIND_SRV_PREV_PRE_ALPHA_COLOR 3
#define FSR2_BIND_SRV_PREV_POST_ALPHA_COLOR 4
#define FSR2_BIND_SRV_REACTIVE_MASK 5
#define FSR2_BIND_SRV_TRANSPARENCY_AND_COMPOSITION_MASK 6
#define FSR2_BIND_UAV_AUTOREACTIVE 7
#define FSR2_BIND_UAV_AUTOCOMPOSITION 8
#define FSR2_BIND_UAV_PREV_PRE_ALPHA_COLOR 9
#define FSR2_BIND_UAV_PREV_POST_ALPHA_COLOR 10
#define FSR2_BIND_CB_FSR2 11
#define FSR2_BIND_CB_REACTIVE 12
// -- GODOT start --
#if FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS
#define FSR2_BIND_SRV_INPUT_DEPTH 13
#endif
// -- GODOT end --
#include "ffx_fsr2_callbacks_glsl.h"
#include "ffx_fsr2_common.h"
#ifdef FSR2_BIND_CB_REACTIVE
layout (set = 1, binding = FSR2_BIND_CB_REACTIVE, std140) uniform cbGenerateReactive_t
{
float fTcThreshold; // 0.1 is a good starting value, lower will result in more TC pixels
float fTcScale;
float fReactiveScale;
float fReactiveMax;
} cbGenerateReactive;
float getTcThreshold()
{
return cbGenerateReactive.fTcThreshold;
}
#else
float getTcThreshold()
{
return 0.05f;
}
#endif
#include "ffx_fsr2_tcr_autogen.h"
#ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#define FFX_FSR2_THREAD_GROUP_WIDTH 8
#endif // #ifndef FFX_FSR2_THREAD_GROUP_WIDTH
#ifndef FFX_FSR2_THREAD_GROUP_HEIGHT
#define FFX_FSR2_THREAD_GROUP_HEIGHT 8
#endif // FFX_FSR2_THREAD_GROUP_HEIGHT
#ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#define FFX_FSR2_THREAD_GROUP_DEPTH 1
#endif // #ifndef FFX_FSR2_THREAD_GROUP_DEPTH
#ifndef FFX_FSR2_NUM_THREADS
#define FFX_FSR2_NUM_THREADS layout (local_size_x = FFX_FSR2_THREAD_GROUP_WIDTH, local_size_y = FFX_FSR2_THREAD_GROUP_HEIGHT, local_size_z = FFX_FSR2_THREAD_GROUP_DEPTH) in;
#endif // #ifndef FFX_FSR2_NUM_THREADS
FFX_FSR2_NUM_THREADS
void main()
{
FFX_MIN16_I2 uDispatchThreadId = FFX_MIN16_I2(gl_GlobalInvocationID.xy);
// ToDo: take into account jitter (i.e. add delta of previous jitter and current jitter to previous UV
// fetch pre- and post-alpha color values
FFX_MIN16_F2 fUv = ( FFX_MIN16_F2(uDispatchThreadId) + FFX_MIN16_F2(0.5f, 0.5f) ) / FFX_MIN16_F2( RenderSize() );
FFX_MIN16_F2 fPrevUV = fUv + FFX_MIN16_F2( LoadInputMotionVector(uDispatchThreadId) );
FFX_MIN16_I2 iPrevIdx = FFX_MIN16_I2(fPrevUV * FFX_MIN16_F2(RenderSize()) - 0.5f);
FFX_MIN16_F3 colorPreAlpha = FFX_MIN16_F3( LoadOpaqueOnly( uDispatchThreadId ) );
FFX_MIN16_F3 colorPostAlpha = FFX_MIN16_F3( LoadInputColor( uDispatchThreadId ) );
FFX_MIN16_F2 outReactiveMask = FFX_MIN16_F2( 0.f, 0.f );
outReactiveMask.y = ComputeTransparencyAndComposition(uDispatchThreadId, iPrevIdx);
if (outReactiveMask.y > 0.5f)
{
outReactiveMask.x = ComputeReactive(uDispatchThreadId, iPrevIdx);
outReactiveMask.x *= FFX_MIN16_F(cbGenerateReactive.fReactiveScale);
outReactiveMask.x = outReactiveMask.x < cbGenerateReactive.fReactiveMax ? outReactiveMask.x : FFX_MIN16_F( cbGenerateReactive.fReactiveMax );
}
outReactiveMask.y *= FFX_MIN16_F(cbGenerateReactive.fTcScale);
outReactiveMask.x = ffxMax(outReactiveMask.x, FFX_MIN16_F(LoadReactiveMask(uDispatchThreadId)));
outReactiveMask.y = ffxMax(outReactiveMask.y, FFX_MIN16_F(LoadTransparencyAndCompositionMask(uDispatchThreadId)));
StoreAutoReactive(uDispatchThreadId, outReactiveMask);
StorePrevPreAlpha(uDispatchThreadId, colorPreAlpha);
StorePrevPostAlpha(uDispatchThreadId, colorPostAlpha);
}

Some files were not shown because too many files have changed in this diff Show More