Add shader cache to gles3

This commit is contained in:
ChibiDenDen 2023-04-15 17:07:51 +03:00
parent a7276f1ce0
commit 4adbfcfbb2
3 changed files with 152 additions and 86 deletions

View File

@ -34,6 +34,7 @@
#ifdef GLES3_ENABLED #ifdef GLES3_ENABLED
#include "core/config/project_settings.h" #include "core/config/project_settings.h"
#include "core/io/dir_access.h"
#include "core/os/os.h" #include "core/os/os.h"
#include "storage/texture_storage.h" #include "storage/texture_storage.h"
@ -257,6 +258,36 @@ RasterizerGLES3::RasterizerGLES3() {
#endif // GLES_OVER_GL #endif // GLES_OVER_GL
#endif // CAN_DEBUG #endif // CAN_DEBUG
{
String shader_cache_dir = Engine::get_singleton()->get_shader_cache_path();
if (shader_cache_dir.is_empty()) {
shader_cache_dir = "user://";
}
Ref<DirAccess> da = DirAccess::open(shader_cache_dir);
if (da.is_null()) {
ERR_PRINT("Can't create shader cache folder, no shader caching will happen: " + shader_cache_dir);
} else {
Error err = da->change_dir("shader_cache");
if (err != OK) {
err = da->make_dir("shader_cache");
}
if (err != OK) {
ERR_PRINT("Can't create shader cache folder, no shader caching will happen: " + shader_cache_dir);
} else {
shader_cache_dir = shader_cache_dir.path_join("shader_cache");
bool shader_cache_enabled = GLOBAL_GET("rendering/shader_compiler/shader_cache/enabled");
if (!Engine::get_singleton()->is_editor_hint() && !shader_cache_enabled) {
shader_cache_dir = String(); //disable only if not editor
}
if (!shader_cache_dir.is_empty()) {
ShaderGLES3::set_shader_cache_dir(shader_cache_dir);
}
}
}
}
// OpenGL needs to be initialized before initializing the Rasterizers // OpenGL needs to be initialized before initializing the Rasterizers
config = memnew(GLES3::Config); config = memnew(GLES3::Config);
utilities = memnew(GLES3::Utilities); utilities = memnew(GLES3::Utilities);

View File

@ -242,6 +242,41 @@ static void _display_error_with_code(const String &p_error, const String &p_code
ERR_PRINT(p_error); ERR_PRINT(p_error);
} }
void ShaderGLES3::_get_uniform_locations(Version::Specialization &spec, Version *p_version) {
glUseProgram(spec.id);
spec.uniform_location.resize(uniform_count);
for (int i = 0; i < uniform_count; i++) {
spec.uniform_location[i] = glGetUniformLocation(spec.id, uniform_names[i]);
}
for (int i = 0; i < texunit_pair_count; i++) {
GLint loc = glGetUniformLocation(spec.id, texunit_pairs[i].name);
if (loc >= 0) {
if (texunit_pairs[i].index < 0) {
glUniform1i(loc, max_image_units + texunit_pairs[i].index);
} else {
glUniform1i(loc, texunit_pairs[i].index);
}
}
}
for (int i = 0; i < ubo_count; i++) {
GLint loc = glGetUniformBlockIndex(spec.id, ubo_pairs[i].name);
if (loc >= 0) {
glUniformBlockBinding(spec.id, loc, ubo_pairs[i].index);
}
}
// textures
for (int i = 0; i < p_version->texture_uniforms.size(); i++) {
String native_uniform_name = _mkid(p_version->texture_uniforms[i]);
GLint location = glGetUniformLocation(spec.id, (native_uniform_name).ascii().get_data());
glUniform1i(location, i + base_texture_index);
}
glUseProgram(0);
}
void ShaderGLES3::_compile_specialization(Version::Specialization &spec, uint32_t p_variant, Version *p_version, uint64_t p_specialization) { void ShaderGLES3::_compile_specialization(Version::Specialization &spec, uint32_t p_variant, Version *p_version, uint64_t p_specialization) {
spec.id = glCreateProgram(); spec.id = glCreateProgram();
spec.ok = false; spec.ok = false;
@ -402,40 +437,8 @@ void ShaderGLES3::_compile_specialization(Version::Specialization &spec, uint32_
ERR_FAIL(); ERR_FAIL();
} }
// get uniform locations _get_uniform_locations(spec, p_version);
glUseProgram(spec.id);
spec.uniform_location.resize(uniform_count);
for (int i = 0; i < uniform_count; i++) {
spec.uniform_location[i] = glGetUniformLocation(spec.id, uniform_names[i]);
}
for (int i = 0; i < texunit_pair_count; i++) {
GLint loc = glGetUniformLocation(spec.id, texunit_pairs[i].name);
if (loc >= 0) {
if (texunit_pairs[i].index < 0) {
glUniform1i(loc, max_image_units + texunit_pairs[i].index);
} else {
glUniform1i(loc, texunit_pairs[i].index);
}
}
}
for (int i = 0; i < ubo_count; i++) {
GLint loc = glGetUniformBlockIndex(spec.id, ubo_pairs[i].name);
if (loc >= 0) {
glUniformBlockBinding(spec.id, loc, ubo_pairs[i].index);
}
}
// textures
for (int i = 0; i < p_version->texture_uniforms.size(); i++) {
String native_uniform_name = _mkid(p_version->texture_uniforms[i]);
GLint location = glGetUniformLocation(spec.id, (native_uniform_name).ascii().get_data());
glUniform1i(location, i + base_texture_index);
}
glUseProgram(0);
spec.ok = true; spec.ok = true;
} }
@ -504,11 +507,15 @@ String ShaderGLES3::_version_get_sha1(Version *p_version) const {
return hash_build.as_string().sha1_text(); return hash_build.as_string().sha1_text();
} }
//static const char *shader_file_header = "GLSC"; #ifndef GLES_OVER_GL
//static const uint32_t cache_file_version = 2; static const char *shader_file_header = "GLSC";
static const uint32_t cache_file_version = 3;
#endif
bool ShaderGLES3::_load_from_cache(Version *p_version) { bool ShaderGLES3::_load_from_cache(Version *p_version) {
#if 0 #ifdef GLES_OVER_GL // glGetProgramBinary and glProgramBinary require GL version 4.1 while GLAD is only configured for 3.3 right now
return false;
#else
String sha1 = _version_get_sha1(p_version); String sha1 = _version_get_sha1(p_version);
String path = shader_cache_dir.path_join(name).path_join(base_sha256).path_join(sha1) + ".cache"; String path = shader_cache_dir.path_join(name).path_join(base_sha256).path_join(sha1) + ".cache";
@ -517,7 +524,7 @@ bool ShaderGLES3::_load_from_cache(Version *p_version) {
return false; return false;
} }
char header[5] = { 0, 0, 0, 0, 0 }; char header[5] = {};
f->get_buffer((uint8_t *)header, 4); f->get_buffer((uint8_t *)header, 4);
ERR_FAIL_COND_V(header != String(shader_file_header), false); ERR_FAIL_COND_V(header != String(shader_file_header), false);
@ -526,16 +533,20 @@ bool ShaderGLES3::_load_from_cache(Version *p_version) {
return false; // wrong version return false; // wrong version
} }
uint32_t variant_count = f->get_32(); int variant_count = static_cast<int>(f->get_32());
ERR_FAIL_COND_V_MSG(variant_count != this->variant_count, false, "shader cache variant count mismatch, expected " + itos(this->variant_count) + " got " + itos(variant_count)); //should not happen but check
ERR_FAIL_COND_V(variant_count != (uint32_t)variant_count, false); //should not happen but check LocalVector<OAHashMap<uint64_t, Version::Specialization>> variants;
for (int i = 0; i < variant_count; i++) {
for (uint32_t i = 0; i < variant_count; i++) { uint32_t specialization_count = f->get_32();
OAHashMap<uint64_t, Version::Specialization> variant;
for (uint32_t j = 0; j < specialization_count; j++) {
uint64_t specialization_key = f->get_64();
uint32_t variant_size = f->get_32(); uint32_t variant_size = f->get_32();
ERR_FAIL_COND_V(variant_size == 0 && variants_enabled[i], false); if (variant_size == 0) {
if (!variants_enabled[i]) {
continue; continue;
} }
uint32_t variant_format = f->get_32();
Vector<uint8_t> variant_bytes; Vector<uint8_t> variant_bytes;
variant_bytes.resize(variant_size); variant_bytes.resize(variant_size);
@ -543,51 +554,70 @@ bool ShaderGLES3::_load_from_cache(Version *p_version) {
ERR_FAIL_COND_V(br != variant_size, false); ERR_FAIL_COND_V(br != variant_size, false);
p_version->variant_data[i] = variant_bytes; Version::Specialization specialization;
}
for (uint32_t i = 0; i < variant_count; i++) { specialization.id = glCreateProgram();
if (!variants_enabled[i]) { glProgramBinary(specialization.id, variant_format, variant_bytes.ptr(), variant_bytes.size());
MutexLock lock(variant_set_mutex);
p_version->variants[i] = RID(); _get_uniform_locations(specialization, p_version);
continue;
} specialization.ok = true;
RID shader = GLES3::get_singleton()->shader_create_from_bytecode(p_version->variant_data[i]);
if (shader.is_null()) { variant.insert(specialization_key, specialization);
for (uint32_t j = 0; j < i; j++) { }
GLES3::get_singleton()->free(p_version->variants[i]); variants.push_back(variant);
} }
ERR_FAIL_COND_V(shader.is_null(), false); p_version->variants = variants;
}
{
MutexLock lock(variant_set_mutex);
p_version->variants[i] = shader;
}
}
memdelete_arr(p_version->variant_data); //clear stages
p_version->variant_data = nullptr;
p_version->valid = true;
return true; return true;
#endif #endif
return false;
} }
void ShaderGLES3::_save_to_cache(Version *p_version) { void ShaderGLES3::_save_to_cache(Version *p_version) {
#if 0 #ifdef GLES_OVER_GL // glGetProgramBinary and glProgramBinary require GL version 4.1 while GLAD is only configured for 3.3 right now
return;
#else
String sha1 = _version_get_sha1(p_version); String sha1 = _version_get_sha1(p_version);
String path = shader_cache_dir.path_join(name).path_join(base_sha256).path_join(sha1) + ".cache"; String path = shader_cache_dir.path_join(name).path_join(base_sha256).path_join(sha1) + ".cache";
Ref<FileAccess> f = FileAccess::open(path, FileAccess::WRITE); Error error;
Ref<FileAccess> f = FileAccess::open(path, FileAccess::WRITE, &error);
ERR_FAIL_COND(f.is_null()); ERR_FAIL_COND(f.is_null());
f->store_buffer((const uint8_t *)shader_file_header, 4); f->store_buffer((const uint8_t *)shader_file_header, 4);
f->store_32(cache_file_version); //file version f->store_32(cache_file_version);
uint32_t variant_count = variant_count; f->store_32(variant_count);
f->store_32(variant_count); //variant count
for (uint32_t i = 0; i < variant_count; i++) { for (int i = 0; i < variant_count; i++) {
f->store_32(p_version->variant_data[i].size()); //stage count int specialization_count = p_version->variants[i].get_num_elements();
f->store_buffer(p_version->variant_data[i].ptr(), p_version->variant_data[i].size()); f->store_32(specialization_count);
for (OAHashMap<uint64_t, ShaderGLES3::Version::Specialization>::Iterator it = p_version->variants[i].iter(); it.valid; it = p_version->variants[i].next_iter(it)) {
const uint64_t specialization_key = *it.key;
f->store_64(specialization_key);
const Version::Specialization *specialization = it.value;
if (specialization == nullptr) {
f->store_32(0);
continue;
}
GLint program_size = 0;
glGetProgramiv(specialization->id, GL_PROGRAM_BINARY_LENGTH, &program_size);
if (program_size == 0) {
f->store_32(0);
continue;
}
PackedByteArray compiled_program;
compiled_program.resize(program_size);
GLenum binary_format = 0;
glGetProgramBinary(specialization->id, program_size, nullptr, &binary_format, compiled_program.ptrw());
if (program_size != compiled_program.size()) {
f->store_32(0);
continue;
}
f->store_32(program_size);
f->store_32(binary_format);
f->store_buffer(compiled_program.ptr(), compiled_program.size());
}
} }
#endif #endif
} }
@ -613,6 +643,9 @@ void ShaderGLES3::_clear_version(Version *p_version) {
void ShaderGLES3::_initialize_version(Version *p_version) { void ShaderGLES3::_initialize_version(Version *p_version) {
ERR_FAIL_COND(p_version->variants.size() > 0); ERR_FAIL_COND(p_version->variants.size() > 0);
if (_load_from_cache(p_version)) {
return;
}
p_version->variants.reserve(variant_count); p_version->variants.reserve(variant_count);
for (int i = 0; i < variant_count; i++) { for (int i = 0; i < variant_count; i++) {
OAHashMap<uint64_t, Version::Specialization> variant; OAHashMap<uint64_t, Version::Specialization> variant;
@ -621,6 +654,7 @@ void ShaderGLES3::_initialize_version(Version *p_version) {
_compile_specialization(spec, i, p_version, specialization_default_mask); _compile_specialization(spec, i, p_version, specialization_default_mask);
p_version->variants[i].insert(specialization_default_mask, spec); p_version->variants[i].insert(specialization_default_mask, spec);
} }
_save_to_cache(p_version);
} }
void ShaderGLES3::version_set_code(RID p_version, const HashMap<String, String> &p_code, const String &p_uniforms, const String &p_vertex_globals, const String &p_fragment_globals, const Vector<String> &p_custom_defines, const Vector<StringName> &p_texture_uniforms, bool p_initialize) { void ShaderGLES3::version_set_code(RID p_version, const HashMap<String, String> &p_code, const String &p_uniforms, const String &p_vertex_globals, const String &p_fragment_globals, const Vector<String> &p_custom_defines, const Vector<StringName> &p_texture_uniforms, bool p_initialize) {

View File

@ -98,7 +98,6 @@ private:
GLuint frag_id; GLuint frag_id;
LocalVector<GLint> uniform_location; LocalVector<GLint> uniform_location;
LocalVector<GLint> texture_uniform_locations; LocalVector<GLint> texture_uniform_locations;
HashMap<StringName, GLint> custom_uniform_locations;
bool build_queued = false; bool build_queued = false;
bool ok = false; bool ok = false;
Specialization() { Specialization() {
@ -113,6 +112,7 @@ private:
Mutex variant_set_mutex; Mutex variant_set_mutex;
void _get_uniform_locations(Version::Specialization &spec, Version *p_version);
void _compile_specialization(Version::Specialization &spec, uint32_t p_variant, Version *p_version, uint64_t p_specialization); void _compile_specialization(Version::Specialization &spec, uint32_t p_variant, Version *p_version, uint64_t p_specialization);
void _clear_version(Version *p_version); void _clear_version(Version *p_version);
@ -209,6 +209,7 @@ protected:
_compile_specialization(s, p_variant, version, p_specialization); _compile_specialization(s, p_variant, version, p_specialization);
version->variants[p_variant].insert(p_specialization, s); version->variants[p_variant].insert(p_specialization, s);
spec = version->variants[p_variant].lookup_ptr(p_specialization); spec = version->variants[p_variant].lookup_ptr(p_specialization);
_save_to_cache(version);
} }
} else if (spec->build_queued) { } else if (spec->build_queued) {
// Still queued, wait // Still queued, wait