parent
104b06e967
commit
25007bdb61
|
@ -213,7 +213,7 @@ Godot-made change marked with `// -- GODOT --` comments.
|
||||||
## libwebp
|
## libwebp
|
||||||
|
|
||||||
- Upstream: https://chromium.googlesource.com/webm/libwebp/
|
- Upstream: https://chromium.googlesource.com/webm/libwebp/
|
||||||
- Version: 1.0.2
|
- Version: 1.0.3
|
||||||
- License: BSD-3-Clause
|
- License: BSD-3-Clause
|
||||||
|
|
||||||
Files extracted from upstream source:
|
Files extracted from upstream source:
|
||||||
|
|
|
@ -61,12 +61,17 @@ static const uint16_t kAcTable[128] = {
|
||||||
|
|
||||||
void VP8ParseQuant(VP8Decoder* const dec) {
|
void VP8ParseQuant(VP8Decoder* const dec) {
|
||||||
VP8BitReader* const br = &dec->br_;
|
VP8BitReader* const br = &dec->br_;
|
||||||
const int base_q0 = VP8GetValue(br, 7);
|
const int base_q0 = VP8GetValue(br, 7, "global-header");
|
||||||
const int dqy1_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
|
const int dqy1_dc = VP8Get(br, "global-header") ?
|
||||||
const int dqy2_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
|
VP8GetSignedValue(br, 4, "global-header") : 0;
|
||||||
const int dqy2_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
|
const int dqy2_dc = VP8Get(br, "global-header") ?
|
||||||
const int dquv_dc = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
|
VP8GetSignedValue(br, 4, "global-header") : 0;
|
||||||
const int dquv_ac = VP8Get(br) ? VP8GetSignedValue(br, 4) : 0;
|
const int dqy2_ac = VP8Get(br, "global-header") ?
|
||||||
|
VP8GetSignedValue(br, 4, "global-header") : 0;
|
||||||
|
const int dquv_dc = VP8Get(br, "global-header") ?
|
||||||
|
VP8GetSignedValue(br, 4, "global-header") : 0;
|
||||||
|
const int dquv_ac = VP8Get(br, "global-header") ?
|
||||||
|
VP8GetSignedValue(br, 4, "global-header") : 0;
|
||||||
|
|
||||||
const VP8SegmentHeader* const hdr = &dec->segment_hdr_;
|
const VP8SegmentHeader* const hdr = &dec->segment_hdr_;
|
||||||
int i;
|
int i;
|
||||||
|
|
|
@ -296,20 +296,21 @@ static void ParseIntraMode(VP8BitReader* const br,
|
||||||
// to decode more than 1 keyframe.
|
// to decode more than 1 keyframe.
|
||||||
if (dec->segment_hdr_.update_map_) {
|
if (dec->segment_hdr_.update_map_) {
|
||||||
// Hardcoded tree parsing
|
// Hardcoded tree parsing
|
||||||
block->segment_ = !VP8GetBit(br, dec->proba_.segments_[0])
|
block->segment_ = !VP8GetBit(br, dec->proba_.segments_[0], "segments")
|
||||||
? VP8GetBit(br, dec->proba_.segments_[1])
|
? VP8GetBit(br, dec->proba_.segments_[1], "segments")
|
||||||
: 2 + VP8GetBit(br, dec->proba_.segments_[2]);
|
: VP8GetBit(br, dec->proba_.segments_[2], "segments") + 2;
|
||||||
} else {
|
} else {
|
||||||
block->segment_ = 0; // default for intra
|
block->segment_ = 0; // default for intra
|
||||||
}
|
}
|
||||||
if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_);
|
if (dec->use_skip_proba_) block->skip_ = VP8GetBit(br, dec->skip_p_, "skip");
|
||||||
|
|
||||||
block->is_i4x4_ = !VP8GetBit(br, 145); // decide for B_PRED first
|
block->is_i4x4_ = !VP8GetBit(br, 145, "block-size");
|
||||||
if (!block->is_i4x4_) {
|
if (!block->is_i4x4_) {
|
||||||
// Hardcoded 16x16 intra-mode decision tree.
|
// Hardcoded 16x16 intra-mode decision tree.
|
||||||
const int ymode =
|
const int ymode =
|
||||||
VP8GetBit(br, 156) ? (VP8GetBit(br, 128) ? TM_PRED : H_PRED)
|
VP8GetBit(br, 156, "pred-modes") ?
|
||||||
: (VP8GetBit(br, 163) ? V_PRED : DC_PRED);
|
(VP8GetBit(br, 128, "pred-modes") ? TM_PRED : H_PRED) :
|
||||||
|
(VP8GetBit(br, 163, "pred-modes") ? V_PRED : DC_PRED);
|
||||||
block->imodes_[0] = ymode;
|
block->imodes_[0] = ymode;
|
||||||
memset(top, ymode, 4 * sizeof(*top));
|
memset(top, ymode, 4 * sizeof(*top));
|
||||||
memset(left, ymode, 4 * sizeof(*left));
|
memset(left, ymode, 4 * sizeof(*left));
|
||||||
|
@ -323,22 +324,25 @@ static void ParseIntraMode(VP8BitReader* const br,
|
||||||
const uint8_t* const prob = kBModesProba[top[x]][ymode];
|
const uint8_t* const prob = kBModesProba[top[x]][ymode];
|
||||||
#if (USE_GENERIC_TREE == 1)
|
#if (USE_GENERIC_TREE == 1)
|
||||||
// Generic tree-parsing
|
// Generic tree-parsing
|
||||||
int i = kYModesIntra4[VP8GetBit(br, prob[0])];
|
int i = kYModesIntra4[VP8GetBit(br, prob[0], "pred-modes")];
|
||||||
while (i > 0) {
|
while (i > 0) {
|
||||||
i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i])];
|
i = kYModesIntra4[2 * i + VP8GetBit(br, prob[i], "pred-modes")];
|
||||||
}
|
}
|
||||||
ymode = -i;
|
ymode = -i;
|
||||||
#else
|
#else
|
||||||
// Hardcoded tree parsing
|
// Hardcoded tree parsing
|
||||||
ymode = !VP8GetBit(br, prob[0]) ? B_DC_PRED :
|
ymode = !VP8GetBit(br, prob[0], "pred-modes") ? B_DC_PRED :
|
||||||
!VP8GetBit(br, prob[1]) ? B_TM_PRED :
|
!VP8GetBit(br, prob[1], "pred-modes") ? B_TM_PRED :
|
||||||
!VP8GetBit(br, prob[2]) ? B_VE_PRED :
|
!VP8GetBit(br, prob[2], "pred-modes") ? B_VE_PRED :
|
||||||
!VP8GetBit(br, prob[3]) ?
|
!VP8GetBit(br, prob[3], "pred-modes") ?
|
||||||
(!VP8GetBit(br, prob[4]) ? B_HE_PRED :
|
(!VP8GetBit(br, prob[4], "pred-modes") ? B_HE_PRED :
|
||||||
(!VP8GetBit(br, prob[5]) ? B_RD_PRED : B_VR_PRED)) :
|
(!VP8GetBit(br, prob[5], "pred-modes") ? B_RD_PRED
|
||||||
(!VP8GetBit(br, prob[6]) ? B_LD_PRED :
|
: B_VR_PRED)) :
|
||||||
(!VP8GetBit(br, prob[7]) ? B_VL_PRED :
|
(!VP8GetBit(br, prob[6], "pred-modes") ? B_LD_PRED :
|
||||||
(!VP8GetBit(br, prob[8]) ? B_HD_PRED : B_HU_PRED)));
|
(!VP8GetBit(br, prob[7], "pred-modes") ? B_VL_PRED :
|
||||||
|
(!VP8GetBit(br, prob[8], "pred-modes") ? B_HD_PRED
|
||||||
|
: B_HU_PRED))
|
||||||
|
);
|
||||||
#endif // USE_GENERIC_TREE
|
#endif // USE_GENERIC_TREE
|
||||||
top[x] = ymode;
|
top[x] = ymode;
|
||||||
}
|
}
|
||||||
|
@ -348,9 +352,9 @@ static void ParseIntraMode(VP8BitReader* const br,
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
// Hardcoded UVMode decision tree
|
// Hardcoded UVMode decision tree
|
||||||
block->uvmode_ = !VP8GetBit(br, 142) ? DC_PRED
|
block->uvmode_ = !VP8GetBit(br, 142, "pred-modes-uv") ? DC_PRED
|
||||||
: !VP8GetBit(br, 114) ? V_PRED
|
: !VP8GetBit(br, 114, "pred-modes-uv") ? V_PRED
|
||||||
: VP8GetBit(br, 183) ? TM_PRED : H_PRED;
|
: VP8GetBit(br, 183, "pred-modes-uv") ? TM_PRED : H_PRED;
|
||||||
}
|
}
|
||||||
|
|
||||||
int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec) {
|
int VP8ParseIntraModeRow(VP8BitReader* const br, VP8Decoder* const dec) {
|
||||||
|
@ -514,8 +518,10 @@ void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
|
||||||
for (b = 0; b < NUM_BANDS; ++b) {
|
for (b = 0; b < NUM_BANDS; ++b) {
|
||||||
for (c = 0; c < NUM_CTX; ++c) {
|
for (c = 0; c < NUM_CTX; ++c) {
|
||||||
for (p = 0; p < NUM_PROBAS; ++p) {
|
for (p = 0; p < NUM_PROBAS; ++p) {
|
||||||
const int v = VP8GetBit(br, CoeffsUpdateProba[t][b][c][p]) ?
|
const int v =
|
||||||
VP8GetValue(br, 8) : CoeffsProba0[t][b][c][p];
|
VP8GetBit(br, CoeffsUpdateProba[t][b][c][p], "global-header") ?
|
||||||
|
VP8GetValue(br, 8, "global-header") :
|
||||||
|
CoeffsProba0[t][b][c][p];
|
||||||
proba->bands_[t][b].probas_[c][p] = v;
|
proba->bands_[t][b].probas_[c][p] = v;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -524,9 +530,8 @@ void VP8ParseProba(VP8BitReader* const br, VP8Decoder* const dec) {
|
||||||
proba->bands_ptr_[t][b] = &proba->bands_[t][kBands[b]];
|
proba->bands_ptr_[t][b] = &proba->bands_[t][kBands[b]];
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
dec->use_skip_proba_ = VP8Get(br);
|
dec->use_skip_proba_ = VP8Get(br, "global-header");
|
||||||
if (dec->use_skip_proba_) {
|
if (dec->use_skip_proba_) {
|
||||||
dec->skip_p_ = VP8GetValue(br, 8);
|
dec->skip_p_ = VP8GetValue(br, 8, "global-header");
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
|
@ -161,23 +161,26 @@ static int ParseSegmentHeader(VP8BitReader* br,
|
||||||
VP8SegmentHeader* hdr, VP8Proba* proba) {
|
VP8SegmentHeader* hdr, VP8Proba* proba) {
|
||||||
assert(br != NULL);
|
assert(br != NULL);
|
||||||
assert(hdr != NULL);
|
assert(hdr != NULL);
|
||||||
hdr->use_segment_ = VP8Get(br);
|
hdr->use_segment_ = VP8Get(br, "global-header");
|
||||||
if (hdr->use_segment_) {
|
if (hdr->use_segment_) {
|
||||||
hdr->update_map_ = VP8Get(br);
|
hdr->update_map_ = VP8Get(br, "global-header");
|
||||||
if (VP8Get(br)) { // update data
|
if (VP8Get(br, "global-header")) { // update data
|
||||||
int s;
|
int s;
|
||||||
hdr->absolute_delta_ = VP8Get(br);
|
hdr->absolute_delta_ = VP8Get(br, "global-header");
|
||||||
for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
|
for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
|
||||||
hdr->quantizer_[s] = VP8Get(br) ? VP8GetSignedValue(br, 7) : 0;
|
hdr->quantizer_[s] = VP8Get(br, "global-header") ?
|
||||||
|
VP8GetSignedValue(br, 7, "global-header") : 0;
|
||||||
}
|
}
|
||||||
for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
|
for (s = 0; s < NUM_MB_SEGMENTS; ++s) {
|
||||||
hdr->filter_strength_[s] = VP8Get(br) ? VP8GetSignedValue(br, 6) : 0;
|
hdr->filter_strength_[s] = VP8Get(br, "global-header") ?
|
||||||
|
VP8GetSignedValue(br, 6, "global-header") : 0;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
if (hdr->update_map_) {
|
if (hdr->update_map_) {
|
||||||
int s;
|
int s;
|
||||||
for (s = 0; s < MB_FEATURE_TREE_PROBS; ++s) {
|
for (s = 0; s < MB_FEATURE_TREE_PROBS; ++s) {
|
||||||
proba->segments_[s] = VP8Get(br) ? VP8GetValue(br, 8) : 255u;
|
proba->segments_[s] = VP8Get(br, "global-header") ?
|
||||||
|
VP8GetValue(br, 8, "global-header") : 255u;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
|
@ -205,7 +208,7 @@ static VP8StatusCode ParsePartitions(VP8Decoder* const dec,
|
||||||
size_t last_part;
|
size_t last_part;
|
||||||
size_t p;
|
size_t p;
|
||||||
|
|
||||||
dec->num_parts_minus_one_ = (1 << VP8GetValue(br, 2)) - 1;
|
dec->num_parts_minus_one_ = (1 << VP8GetValue(br, 2, "global-header")) - 1;
|
||||||
last_part = dec->num_parts_minus_one_;
|
last_part = dec->num_parts_minus_one_;
|
||||||
if (size < 3 * last_part) {
|
if (size < 3 * last_part) {
|
||||||
// we can't even read the sizes with sz[]! That's a failure.
|
// we can't even read the sizes with sz[]! That's a failure.
|
||||||
|
@ -229,21 +232,21 @@ static VP8StatusCode ParsePartitions(VP8Decoder* const dec,
|
||||||
// Paragraph 9.4
|
// Paragraph 9.4
|
||||||
static int ParseFilterHeader(VP8BitReader* br, VP8Decoder* const dec) {
|
static int ParseFilterHeader(VP8BitReader* br, VP8Decoder* const dec) {
|
||||||
VP8FilterHeader* const hdr = &dec->filter_hdr_;
|
VP8FilterHeader* const hdr = &dec->filter_hdr_;
|
||||||
hdr->simple_ = VP8Get(br);
|
hdr->simple_ = VP8Get(br, "global-header");
|
||||||
hdr->level_ = VP8GetValue(br, 6);
|
hdr->level_ = VP8GetValue(br, 6, "global-header");
|
||||||
hdr->sharpness_ = VP8GetValue(br, 3);
|
hdr->sharpness_ = VP8GetValue(br, 3, "global-header");
|
||||||
hdr->use_lf_delta_ = VP8Get(br);
|
hdr->use_lf_delta_ = VP8Get(br, "global-header");
|
||||||
if (hdr->use_lf_delta_) {
|
if (hdr->use_lf_delta_) {
|
||||||
if (VP8Get(br)) { // update lf-delta?
|
if (VP8Get(br, "global-header")) { // update lf-delta?
|
||||||
int i;
|
int i;
|
||||||
for (i = 0; i < NUM_REF_LF_DELTAS; ++i) {
|
for (i = 0; i < NUM_REF_LF_DELTAS; ++i) {
|
||||||
if (VP8Get(br)) {
|
if (VP8Get(br, "global-header")) {
|
||||||
hdr->ref_lf_delta_[i] = VP8GetSignedValue(br, 6);
|
hdr->ref_lf_delta_[i] = VP8GetSignedValue(br, 6, "global-header");
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
for (i = 0; i < NUM_MODE_LF_DELTAS; ++i) {
|
for (i = 0; i < NUM_MODE_LF_DELTAS; ++i) {
|
||||||
if (VP8Get(br)) {
|
if (VP8Get(br, "global-header")) {
|
||||||
hdr->mode_lf_delta_[i] = VP8GetSignedValue(br, 6);
|
hdr->mode_lf_delta_[i] = VP8GetSignedValue(br, 6, "global-header");
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -352,8 +355,8 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
|
||||||
buf_size -= frm_hdr->partition_length_;
|
buf_size -= frm_hdr->partition_length_;
|
||||||
|
|
||||||
if (frm_hdr->key_frame_) {
|
if (frm_hdr->key_frame_) {
|
||||||
pic_hdr->colorspace_ = VP8Get(br);
|
pic_hdr->colorspace_ = VP8Get(br, "global-header");
|
||||||
pic_hdr->clamp_type_ = VP8Get(br);
|
pic_hdr->clamp_type_ = VP8Get(br, "global-header");
|
||||||
}
|
}
|
||||||
if (!ParseSegmentHeader(br, &dec->segment_hdr_, &dec->proba_)) {
|
if (!ParseSegmentHeader(br, &dec->segment_hdr_, &dec->proba_)) {
|
||||||
return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
|
return VP8SetError(dec, VP8_STATUS_BITSTREAM_ERROR,
|
||||||
|
@ -378,7 +381,7 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
|
||||||
"Not a key frame.");
|
"Not a key frame.");
|
||||||
}
|
}
|
||||||
|
|
||||||
VP8Get(br); // ignore the value of update_proba_
|
VP8Get(br, "global-header"); // ignore the value of update_proba_
|
||||||
|
|
||||||
VP8ParseProba(br, dec);
|
VP8ParseProba(br, dec);
|
||||||
|
|
||||||
|
@ -403,28 +406,28 @@ static const uint8_t kZigzag[16] = {
|
||||||
// See section 13-2: http://tools.ietf.org/html/rfc6386#section-13.2
|
// See section 13-2: http://tools.ietf.org/html/rfc6386#section-13.2
|
||||||
static int GetLargeValue(VP8BitReader* const br, const uint8_t* const p) {
|
static int GetLargeValue(VP8BitReader* const br, const uint8_t* const p) {
|
||||||
int v;
|
int v;
|
||||||
if (!VP8GetBit(br, p[3])) {
|
if (!VP8GetBit(br, p[3], "coeffs")) {
|
||||||
if (!VP8GetBit(br, p[4])) {
|
if (!VP8GetBit(br, p[4], "coeffs")) {
|
||||||
v = 2;
|
v = 2;
|
||||||
} else {
|
} else {
|
||||||
v = 3 + VP8GetBit(br, p[5]);
|
v = 3 + VP8GetBit(br, p[5], "coeffs");
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
if (!VP8GetBit(br, p[6])) {
|
if (!VP8GetBit(br, p[6], "coeffs")) {
|
||||||
if (!VP8GetBit(br, p[7])) {
|
if (!VP8GetBit(br, p[7], "coeffs")) {
|
||||||
v = 5 + VP8GetBit(br, 159);
|
v = 5 + VP8GetBit(br, 159, "coeffs");
|
||||||
} else {
|
} else {
|
||||||
v = 7 + 2 * VP8GetBit(br, 165);
|
v = 7 + 2 * VP8GetBit(br, 165, "coeffs");
|
||||||
v += VP8GetBit(br, 145);
|
v += VP8GetBit(br, 145, "coeffs");
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
const uint8_t* tab;
|
const uint8_t* tab;
|
||||||
const int bit1 = VP8GetBit(br, p[8]);
|
const int bit1 = VP8GetBit(br, p[8], "coeffs");
|
||||||
const int bit0 = VP8GetBit(br, p[9 + bit1]);
|
const int bit0 = VP8GetBit(br, p[9 + bit1], "coeffs");
|
||||||
const int cat = 2 * bit1 + bit0;
|
const int cat = 2 * bit1 + bit0;
|
||||||
v = 0;
|
v = 0;
|
||||||
for (tab = kCat3456[cat]; *tab; ++tab) {
|
for (tab = kCat3456[cat]; *tab; ++tab) {
|
||||||
v += v + VP8GetBit(br, *tab);
|
v += v + VP8GetBit(br, *tab, "coeffs");
|
||||||
}
|
}
|
||||||
v += 3 + (8 << cat);
|
v += 3 + (8 << cat);
|
||||||
}
|
}
|
||||||
|
@ -438,24 +441,24 @@ static int GetCoeffsFast(VP8BitReader* const br,
|
||||||
int ctx, const quant_t dq, int n, int16_t* out) {
|
int ctx, const quant_t dq, int n, int16_t* out) {
|
||||||
const uint8_t* p = prob[n]->probas_[ctx];
|
const uint8_t* p = prob[n]->probas_[ctx];
|
||||||
for (; n < 16; ++n) {
|
for (; n < 16; ++n) {
|
||||||
if (!VP8GetBit(br, p[0])) {
|
if (!VP8GetBit(br, p[0], "coeffs")) {
|
||||||
return n; // previous coeff was last non-zero coeff
|
return n; // previous coeff was last non-zero coeff
|
||||||
}
|
}
|
||||||
while (!VP8GetBit(br, p[1])) { // sequence of zero coeffs
|
while (!VP8GetBit(br, p[1], "coeffs")) { // sequence of zero coeffs
|
||||||
p = prob[++n]->probas_[0];
|
p = prob[++n]->probas_[0];
|
||||||
if (n == 16) return 16;
|
if (n == 16) return 16;
|
||||||
}
|
}
|
||||||
{ // non zero coeff
|
{ // non zero coeff
|
||||||
const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
|
const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
|
||||||
int v;
|
int v;
|
||||||
if (!VP8GetBit(br, p[2])) {
|
if (!VP8GetBit(br, p[2], "coeffs")) {
|
||||||
v = 1;
|
v = 1;
|
||||||
p = p_ctx[1];
|
p = p_ctx[1];
|
||||||
} else {
|
} else {
|
||||||
v = GetLargeValue(br, p);
|
v = GetLargeValue(br, p);
|
||||||
p = p_ctx[2];
|
p = p_ctx[2];
|
||||||
}
|
}
|
||||||
out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
|
out[kZigzag[n]] = VP8GetSigned(br, v, "coeffs") * dq[n > 0];
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
return 16;
|
return 16;
|
||||||
|
@ -468,24 +471,24 @@ static int GetCoeffsAlt(VP8BitReader* const br,
|
||||||
int ctx, const quant_t dq, int n, int16_t* out) {
|
int ctx, const quant_t dq, int n, int16_t* out) {
|
||||||
const uint8_t* p = prob[n]->probas_[ctx];
|
const uint8_t* p = prob[n]->probas_[ctx];
|
||||||
for (; n < 16; ++n) {
|
for (; n < 16; ++n) {
|
||||||
if (!VP8GetBitAlt(br, p[0])) {
|
if (!VP8GetBitAlt(br, p[0], "coeffs")) {
|
||||||
return n; // previous coeff was last non-zero coeff
|
return n; // previous coeff was last non-zero coeff
|
||||||
}
|
}
|
||||||
while (!VP8GetBitAlt(br, p[1])) { // sequence of zero coeffs
|
while (!VP8GetBitAlt(br, p[1], "coeffs")) { // sequence of zero coeffs
|
||||||
p = prob[++n]->probas_[0];
|
p = prob[++n]->probas_[0];
|
||||||
if (n == 16) return 16;
|
if (n == 16) return 16;
|
||||||
}
|
}
|
||||||
{ // non zero coeff
|
{ // non zero coeff
|
||||||
const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
|
const VP8ProbaArray* const p_ctx = &prob[n + 1]->probas_[0];
|
||||||
int v;
|
int v;
|
||||||
if (!VP8GetBitAlt(br, p[2])) {
|
if (!VP8GetBitAlt(br, p[2], "coeffs")) {
|
||||||
v = 1;
|
v = 1;
|
||||||
p = p_ctx[1];
|
p = p_ctx[1];
|
||||||
} else {
|
} else {
|
||||||
v = GetLargeValue(br, p);
|
v = GetLargeValue(br, p);
|
||||||
p = p_ctx[2];
|
p = p_ctx[2];
|
||||||
}
|
}
|
||||||
out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
|
out[kZigzag[n]] = VP8GetSigned(br, v, "coeffs") * dq[n > 0];
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
return 16;
|
return 16;
|
||||||
|
|
|
@ -32,7 +32,7 @@ extern "C" {
|
||||||
// version numbers
|
// version numbers
|
||||||
#define DEC_MAJ_VERSION 1
|
#define DEC_MAJ_VERSION 1
|
||||||
#define DEC_MIN_VERSION 0
|
#define DEC_MIN_VERSION 0
|
||||||
#define DEC_REV_VERSION 2
|
#define DEC_REV_VERSION 3
|
||||||
|
|
||||||
// YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
|
// YUV-cache parameters. Cache is 32-bytes wide (= one cacheline).
|
||||||
// Constraints are: We need to store one 16x16 block of luma samples (y),
|
// Constraints are: We need to store one 16x16 block of luma samples (y),
|
||||||
|
|
|
@ -362,12 +362,8 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
|
||||||
VP8LMetadata* const hdr = &dec->hdr_;
|
VP8LMetadata* const hdr = &dec->hdr_;
|
||||||
uint32_t* huffman_image = NULL;
|
uint32_t* huffman_image = NULL;
|
||||||
HTreeGroup* htree_groups = NULL;
|
HTreeGroup* htree_groups = NULL;
|
||||||
// When reading htrees, some might be unused, as the format allows it.
|
|
||||||
// We will still read them but put them in this htree_group_bogus.
|
|
||||||
HTreeGroup htree_group_bogus;
|
|
||||||
HuffmanCode* huffman_tables = NULL;
|
HuffmanCode* huffman_tables = NULL;
|
||||||
HuffmanCode* huffman_tables_bogus = NULL;
|
HuffmanCode* huffman_table = NULL;
|
||||||
HuffmanCode* next = NULL;
|
|
||||||
int num_htree_groups = 1;
|
int num_htree_groups = 1;
|
||||||
int num_htree_groups_max = 1;
|
int num_htree_groups_max = 1;
|
||||||
int max_alphabet_size = 0;
|
int max_alphabet_size = 0;
|
||||||
|
@ -418,12 +414,6 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
|
||||||
if (*mapped_group == -1) *mapped_group = num_htree_groups++;
|
if (*mapped_group == -1) *mapped_group = num_htree_groups++;
|
||||||
huffman_image[i] = *mapped_group;
|
huffman_image[i] = *mapped_group;
|
||||||
}
|
}
|
||||||
huffman_tables_bogus = (HuffmanCode*)WebPSafeMalloc(
|
|
||||||
table_size, sizeof(*huffman_tables_bogus));
|
|
||||||
if (huffman_tables_bogus == NULL) {
|
|
||||||
dec->status_ = VP8_STATUS_OUT_OF_MEMORY;
|
|
||||||
goto Error;
|
|
||||||
}
|
|
||||||
} else {
|
} else {
|
||||||
num_htree_groups = num_htree_groups_max;
|
num_htree_groups = num_htree_groups_max;
|
||||||
}
|
}
|
||||||
|
@ -453,63 +443,71 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
|
||||||
goto Error;
|
goto Error;
|
||||||
}
|
}
|
||||||
|
|
||||||
next = huffman_tables;
|
huffman_table = huffman_tables;
|
||||||
for (i = 0; i < num_htree_groups_max; ++i) {
|
for (i = 0; i < num_htree_groups_max; ++i) {
|
||||||
// If the index "i" is unused in the Huffman image, read the coefficients
|
// If the index "i" is unused in the Huffman image, just make sure the
|
||||||
// but store them to a bogus htree_group.
|
// coefficients are valid but do not store them.
|
||||||
const int is_bogus = (mapping != NULL && mapping[i] == -1);
|
if (mapping != NULL && mapping[i] == -1) {
|
||||||
HTreeGroup* const htree_group =
|
for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
|
||||||
is_bogus ? &htree_group_bogus :
|
int alphabet_size = kAlphabetSize[j];
|
||||||
&htree_groups[(mapping == NULL) ? i : mapping[i]];
|
if (j == 0 && color_cache_bits > 0) {
|
||||||
HuffmanCode** const htrees = htree_group->htrees;
|
alphabet_size += (1 << color_cache_bits);
|
||||||
HuffmanCode* huffman_tables_i = is_bogus ? huffman_tables_bogus : next;
|
}
|
||||||
int size;
|
// Passing in NULL so that nothing gets filled.
|
||||||
int total_size = 0;
|
if (!ReadHuffmanCode(alphabet_size, dec, code_lengths, NULL)) {
|
||||||
int is_trivial_literal = 1;
|
goto Error;
|
||||||
int max_bits = 0;
|
|
||||||
for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
|
|
||||||
int alphabet_size = kAlphabetSize[j];
|
|
||||||
htrees[j] = huffman_tables_i;
|
|
||||||
if (j == 0 && color_cache_bits > 0) {
|
|
||||||
alphabet_size += 1 << color_cache_bits;
|
|
||||||
}
|
|
||||||
size =
|
|
||||||
ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_tables_i);
|
|
||||||
if (size == 0) {
|
|
||||||
goto Error;
|
|
||||||
}
|
|
||||||
if (is_trivial_literal && kLiteralMap[j] == 1) {
|
|
||||||
is_trivial_literal = (huffman_tables_i->bits == 0);
|
|
||||||
}
|
|
||||||
total_size += huffman_tables_i->bits;
|
|
||||||
huffman_tables_i += size;
|
|
||||||
if (j <= ALPHA) {
|
|
||||||
int local_max_bits = code_lengths[0];
|
|
||||||
int k;
|
|
||||||
for (k = 1; k < alphabet_size; ++k) {
|
|
||||||
if (code_lengths[k] > local_max_bits) {
|
|
||||||
local_max_bits = code_lengths[k];
|
|
||||||
}
|
|
||||||
}
|
}
|
||||||
max_bits += local_max_bits;
|
|
||||||
}
|
}
|
||||||
}
|
} else {
|
||||||
if (!is_bogus) next = huffman_tables_i;
|
HTreeGroup* const htree_group =
|
||||||
htree_group->is_trivial_literal = is_trivial_literal;
|
&htree_groups[(mapping == NULL) ? i : mapping[i]];
|
||||||
htree_group->is_trivial_code = 0;
|
HuffmanCode** const htrees = htree_group->htrees;
|
||||||
if (is_trivial_literal) {
|
int size;
|
||||||
const int red = htrees[RED][0].value;
|
int total_size = 0;
|
||||||
const int blue = htrees[BLUE][0].value;
|
int is_trivial_literal = 1;
|
||||||
const int alpha = htrees[ALPHA][0].value;
|
int max_bits = 0;
|
||||||
htree_group->literal_arb = ((uint32_t)alpha << 24) | (red << 16) | blue;
|
for (j = 0; j < HUFFMAN_CODES_PER_META_CODE; ++j) {
|
||||||
if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) {
|
int alphabet_size = kAlphabetSize[j];
|
||||||
htree_group->is_trivial_code = 1;
|
htrees[j] = huffman_table;
|
||||||
htree_group->literal_arb |= htrees[GREEN][0].value << 8;
|
if (j == 0 && color_cache_bits > 0) {
|
||||||
|
alphabet_size += (1 << color_cache_bits);
|
||||||
|
}
|
||||||
|
size = ReadHuffmanCode(alphabet_size, dec, code_lengths, huffman_table);
|
||||||
|
if (size == 0) {
|
||||||
|
goto Error;
|
||||||
|
}
|
||||||
|
if (is_trivial_literal && kLiteralMap[j] == 1) {
|
||||||
|
is_trivial_literal = (huffman_table->bits == 0);
|
||||||
|
}
|
||||||
|
total_size += huffman_table->bits;
|
||||||
|
huffman_table += size;
|
||||||
|
if (j <= ALPHA) {
|
||||||
|
int local_max_bits = code_lengths[0];
|
||||||
|
int k;
|
||||||
|
for (k = 1; k < alphabet_size; ++k) {
|
||||||
|
if (code_lengths[k] > local_max_bits) {
|
||||||
|
local_max_bits = code_lengths[k];
|
||||||
|
}
|
||||||
|
}
|
||||||
|
max_bits += local_max_bits;
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
htree_group->is_trivial_literal = is_trivial_literal;
|
||||||
|
htree_group->is_trivial_code = 0;
|
||||||
|
if (is_trivial_literal) {
|
||||||
|
const int red = htrees[RED][0].value;
|
||||||
|
const int blue = htrees[BLUE][0].value;
|
||||||
|
const int alpha = htrees[ALPHA][0].value;
|
||||||
|
htree_group->literal_arb = ((uint32_t)alpha << 24) | (red << 16) | blue;
|
||||||
|
if (total_size == 0 && htrees[GREEN][0].value < NUM_LITERAL_CODES) {
|
||||||
|
htree_group->is_trivial_code = 1;
|
||||||
|
htree_group->literal_arb |= htrees[GREEN][0].value << 8;
|
||||||
|
}
|
||||||
|
}
|
||||||
|
htree_group->use_packed_table =
|
||||||
|
!htree_group->is_trivial_code && (max_bits < HUFFMAN_PACKED_BITS);
|
||||||
|
if (htree_group->use_packed_table) BuildPackedTable(htree_group);
|
||||||
}
|
}
|
||||||
htree_group->use_packed_table =
|
|
||||||
!htree_group->is_trivial_code && (max_bits < HUFFMAN_PACKED_BITS);
|
|
||||||
if (htree_group->use_packed_table) BuildPackedTable(htree_group);
|
|
||||||
}
|
}
|
||||||
ok = 1;
|
ok = 1;
|
||||||
|
|
||||||
|
@ -521,7 +519,6 @@ static int ReadHuffmanCodes(VP8LDecoder* const dec, int xsize, int ysize,
|
||||||
|
|
||||||
Error:
|
Error:
|
||||||
WebPSafeFree(code_lengths);
|
WebPSafeFree(code_lengths);
|
||||||
WebPSafeFree(huffman_tables_bogus);
|
|
||||||
WebPSafeFree(mapping);
|
WebPSafeFree(mapping);
|
||||||
if (!ok) {
|
if (!ok) {
|
||||||
WebPSafeFree(huffman_image);
|
WebPSafeFree(huffman_image);
|
||||||
|
|
|
@ -25,7 +25,7 @@
|
||||||
|
|
||||||
#define DMUX_MAJ_VERSION 1
|
#define DMUX_MAJ_VERSION 1
|
||||||
#define DMUX_MIN_VERSION 0
|
#define DMUX_MIN_VERSION 0
|
||||||
#define DMUX_REV_VERSION 2
|
#define DMUX_REV_VERSION 3
|
||||||
|
|
||||||
typedef struct {
|
typedef struct {
|
||||||
size_t start_; // start location of the data
|
size_t start_; // start location of the data
|
||||||
|
|
|
@ -214,7 +214,7 @@ static void ApplyAlphaMultiply_SSE2(uint8_t* rgba, int alpha_first,
|
||||||
// Alpha detection
|
// Alpha detection
|
||||||
|
|
||||||
static int HasAlpha8b_SSE2(const uint8_t* src, int length) {
|
static int HasAlpha8b_SSE2(const uint8_t* src, int length) {
|
||||||
const __m128i all_0xff = _mm_set1_epi8(0xff);
|
const __m128i all_0xff = _mm_set1_epi8((char)0xff);
|
||||||
int i = 0;
|
int i = 0;
|
||||||
for (; i + 16 <= length; i += 16) {
|
for (; i + 16 <= length; i += 16) {
|
||||||
const __m128i v = _mm_loadu_si128((const __m128i*)(src + i));
|
const __m128i v = _mm_loadu_si128((const __m128i*)(src + i));
|
||||||
|
@ -228,7 +228,7 @@ static int HasAlpha8b_SSE2(const uint8_t* src, int length) {
|
||||||
|
|
||||||
static int HasAlpha32b_SSE2(const uint8_t* src, int length) {
|
static int HasAlpha32b_SSE2(const uint8_t* src, int length) {
|
||||||
const __m128i alpha_mask = _mm_set1_epi32(0xff);
|
const __m128i alpha_mask = _mm_set1_epi32(0xff);
|
||||||
const __m128i all_0xff = _mm_set1_epi8(0xff);
|
const __m128i all_0xff = _mm_set1_epi8((char)0xff);
|
||||||
int i = 0;
|
int i = 0;
|
||||||
// We don't know if we can access the last 3 bytes after the last alpha
|
// We don't know if we can access the last 3 bytes after the last alpha
|
||||||
// value 'src[4 * length - 4]' (because we don't know if alpha is the first
|
// value 'src[4 * length - 4]' (because we don't know if alpha is the first
|
||||||
|
|
|
@ -173,8 +173,8 @@ static int AndroidCPUInfo(CPUFeature feature) {
|
||||||
const AndroidCpuFamily cpu_family = android_getCpuFamily();
|
const AndroidCpuFamily cpu_family = android_getCpuFamily();
|
||||||
const uint64_t cpu_features = android_getCpuFeatures();
|
const uint64_t cpu_features = android_getCpuFeatures();
|
||||||
if (feature == kNEON) {
|
if (feature == kNEON) {
|
||||||
return (cpu_family == ANDROID_CPU_FAMILY_ARM &&
|
return cpu_family == ANDROID_CPU_FAMILY_ARM &&
|
||||||
0 != (cpu_features & ANDROID_CPU_ARM_FEATURE_NEON));
|
(cpu_features & ANDROID_CPU_ARM_FEATURE_NEON) != 0;
|
||||||
}
|
}
|
||||||
return 0;
|
return 0;
|
||||||
}
|
}
|
||||||
|
|
|
@ -326,7 +326,7 @@ static WEBP_INLINE void Update2Pixels_SSE2(__m128i* const pi, __m128i* const qi,
|
||||||
const __m128i a1_lo = _mm_srai_epi16(*a0_lo, 7);
|
const __m128i a1_lo = _mm_srai_epi16(*a0_lo, 7);
|
||||||
const __m128i a1_hi = _mm_srai_epi16(*a0_hi, 7);
|
const __m128i a1_hi = _mm_srai_epi16(*a0_hi, 7);
|
||||||
const __m128i delta = _mm_packs_epi16(a1_lo, a1_hi);
|
const __m128i delta = _mm_packs_epi16(a1_lo, a1_hi);
|
||||||
const __m128i sign_bit = _mm_set1_epi8(0x80);
|
const __m128i sign_bit = _mm_set1_epi8((char)0x80);
|
||||||
*pi = _mm_adds_epi8(*pi, delta);
|
*pi = _mm_adds_epi8(*pi, delta);
|
||||||
*qi = _mm_subs_epi8(*qi, delta);
|
*qi = _mm_subs_epi8(*qi, delta);
|
||||||
FLIP_SIGN_BIT2(*pi, *qi);
|
FLIP_SIGN_BIT2(*pi, *qi);
|
||||||
|
@ -338,9 +338,9 @@ static WEBP_INLINE void NeedsFilter_SSE2(const __m128i* const p1,
|
||||||
const __m128i* const q0,
|
const __m128i* const q0,
|
||||||
const __m128i* const q1,
|
const __m128i* const q1,
|
||||||
int thresh, __m128i* const mask) {
|
int thresh, __m128i* const mask) {
|
||||||
const __m128i m_thresh = _mm_set1_epi8(thresh);
|
const __m128i m_thresh = _mm_set1_epi8((char)thresh);
|
||||||
const __m128i t1 = MM_ABS(*p1, *q1); // abs(p1 - q1)
|
const __m128i t1 = MM_ABS(*p1, *q1); // abs(p1 - q1)
|
||||||
const __m128i kFE = _mm_set1_epi8(0xFE);
|
const __m128i kFE = _mm_set1_epi8((char)0xFE);
|
||||||
const __m128i t2 = _mm_and_si128(t1, kFE); // set lsb of each byte to zero
|
const __m128i t2 = _mm_and_si128(t1, kFE); // set lsb of each byte to zero
|
||||||
const __m128i t3 = _mm_srli_epi16(t2, 1); // abs(p1 - q1) / 2
|
const __m128i t3 = _mm_srli_epi16(t2, 1); // abs(p1 - q1) / 2
|
||||||
|
|
||||||
|
@ -360,7 +360,7 @@ static WEBP_INLINE void DoFilter2_SSE2(__m128i* const p1, __m128i* const p0,
|
||||||
__m128i* const q0, __m128i* const q1,
|
__m128i* const q0, __m128i* const q1,
|
||||||
int thresh) {
|
int thresh) {
|
||||||
__m128i a, mask;
|
__m128i a, mask;
|
||||||
const __m128i sign_bit = _mm_set1_epi8(0x80);
|
const __m128i sign_bit = _mm_set1_epi8((char)0x80);
|
||||||
// convert p1/q1 to int8_t (for GetBaseDelta_SSE2)
|
// convert p1/q1 to int8_t (for GetBaseDelta_SSE2)
|
||||||
const __m128i p1s = _mm_xor_si128(*p1, sign_bit);
|
const __m128i p1s = _mm_xor_si128(*p1, sign_bit);
|
||||||
const __m128i q1s = _mm_xor_si128(*q1, sign_bit);
|
const __m128i q1s = _mm_xor_si128(*q1, sign_bit);
|
||||||
|
@ -380,7 +380,7 @@ static WEBP_INLINE void DoFilter4_SSE2(__m128i* const p1, __m128i* const p0,
|
||||||
const __m128i* const mask,
|
const __m128i* const mask,
|
||||||
int hev_thresh) {
|
int hev_thresh) {
|
||||||
const __m128i zero = _mm_setzero_si128();
|
const __m128i zero = _mm_setzero_si128();
|
||||||
const __m128i sign_bit = _mm_set1_epi8(0x80);
|
const __m128i sign_bit = _mm_set1_epi8((char)0x80);
|
||||||
const __m128i k64 = _mm_set1_epi8(64);
|
const __m128i k64 = _mm_set1_epi8(64);
|
||||||
const __m128i k3 = _mm_set1_epi8(3);
|
const __m128i k3 = _mm_set1_epi8(3);
|
||||||
const __m128i k4 = _mm_set1_epi8(4);
|
const __m128i k4 = _mm_set1_epi8(4);
|
||||||
|
@ -427,7 +427,7 @@ static WEBP_INLINE void DoFilter6_SSE2(__m128i* const p2, __m128i* const p1,
|
||||||
const __m128i* const mask,
|
const __m128i* const mask,
|
||||||
int hev_thresh) {
|
int hev_thresh) {
|
||||||
const __m128i zero = _mm_setzero_si128();
|
const __m128i zero = _mm_setzero_si128();
|
||||||
const __m128i sign_bit = _mm_set1_epi8(0x80);
|
const __m128i sign_bit = _mm_set1_epi8((char)0x80);
|
||||||
__m128i a, not_hev;
|
__m128i a, not_hev;
|
||||||
|
|
||||||
// compute hev mask
|
// compute hev mask
|
||||||
|
@ -941,7 +941,7 @@ static void VR4_SSE2(uint8_t* dst) { // Vertical-Right
|
||||||
const __m128i ABCD0 = _mm_srli_si128(XABCD, 1);
|
const __m128i ABCD0 = _mm_srli_si128(XABCD, 1);
|
||||||
const __m128i abcd = _mm_avg_epu8(XABCD, ABCD0);
|
const __m128i abcd = _mm_avg_epu8(XABCD, ABCD0);
|
||||||
const __m128i _XABCD = _mm_slli_si128(XABCD, 1);
|
const __m128i _XABCD = _mm_slli_si128(XABCD, 1);
|
||||||
const __m128i IXABCD = _mm_insert_epi16(_XABCD, I | (X << 8), 0);
|
const __m128i IXABCD = _mm_insert_epi16(_XABCD, (short)(I | (X << 8)), 0);
|
||||||
const __m128i avg1 = _mm_avg_epu8(IXABCD, ABCD0);
|
const __m128i avg1 = _mm_avg_epu8(IXABCD, ABCD0);
|
||||||
const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
|
const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
|
||||||
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
|
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
|
||||||
|
|
|
@ -777,7 +777,7 @@ static WEBP_INLINE void VR4_SSE2(uint8_t* dst,
|
||||||
const __m128i ABCD0 = _mm_srli_si128(XABCD, 1);
|
const __m128i ABCD0 = _mm_srli_si128(XABCD, 1);
|
||||||
const __m128i abcd = _mm_avg_epu8(XABCD, ABCD0);
|
const __m128i abcd = _mm_avg_epu8(XABCD, ABCD0);
|
||||||
const __m128i _XABCD = _mm_slli_si128(XABCD, 1);
|
const __m128i _XABCD = _mm_slli_si128(XABCD, 1);
|
||||||
const __m128i IXABCD = _mm_insert_epi16(_XABCD, I | (X << 8), 0);
|
const __m128i IXABCD = _mm_insert_epi16(_XABCD, (short)(I | (X << 8)), 0);
|
||||||
const __m128i avg1 = _mm_avg_epu8(IXABCD, ABCD0);
|
const __m128i avg1 = _mm_avg_epu8(IXABCD, ABCD0);
|
||||||
const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
|
const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
|
||||||
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
|
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
|
||||||
|
|
|
@ -33,9 +33,9 @@ static WEBP_INLINE void PredictLine_C(const uint8_t* src, const uint8_t* pred,
|
||||||
uint8_t* dst, int length, int inverse) {
|
uint8_t* dst, int length, int inverse) {
|
||||||
int i;
|
int i;
|
||||||
if (inverse) {
|
if (inverse) {
|
||||||
for (i = 0; i < length; ++i) dst[i] = src[i] + pred[i];
|
for (i = 0; i < length; ++i) dst[i] = (uint8_t)(src[i] + pred[i]);
|
||||||
} else {
|
} else {
|
||||||
for (i = 0; i < length; ++i) dst[i] = src[i] - pred[i];
|
for (i = 0; i < length; ++i) dst[i] = (uint8_t)(src[i] - pred[i]);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -155,7 +155,7 @@ static WEBP_INLINE void DoGradientFilter_C(const uint8_t* in,
|
||||||
const int pred = GradientPredictor_C(preds[w - 1],
|
const int pred = GradientPredictor_C(preds[w - 1],
|
||||||
preds[w - stride],
|
preds[w - stride],
|
||||||
preds[w - stride - 1]);
|
preds[w - stride - 1]);
|
||||||
out[w] = in[w] + (inverse ? pred : -pred);
|
out[w] = (uint8_t)(in[w] + (inverse ? pred : -pred));
|
||||||
}
|
}
|
||||||
++row;
|
++row;
|
||||||
preds += stride;
|
preds += stride;
|
||||||
|
@ -194,7 +194,7 @@ static void HorizontalUnfilter_C(const uint8_t* prev, const uint8_t* in,
|
||||||
uint8_t pred = (prev == NULL) ? 0 : prev[0];
|
uint8_t pred = (prev == NULL) ? 0 : prev[0];
|
||||||
int i;
|
int i;
|
||||||
for (i = 0; i < width; ++i) {
|
for (i = 0; i < width; ++i) {
|
||||||
out[i] = pred + in[i];
|
out[i] = (uint8_t)(pred + in[i]);
|
||||||
pred = out[i];
|
pred = out[i];
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -206,7 +206,7 @@ static void VerticalUnfilter_C(const uint8_t* prev, const uint8_t* in,
|
||||||
HorizontalUnfilter_C(NULL, in, out, width);
|
HorizontalUnfilter_C(NULL, in, out, width);
|
||||||
} else {
|
} else {
|
||||||
int i;
|
int i;
|
||||||
for (i = 0; i < width; ++i) out[i] = prev[i] + in[i];
|
for (i = 0; i < width; ++i) out[i] = (uint8_t)(prev[i] + in[i]);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
#endif // !WEBP_NEON_OMIT_C_CODE
|
#endif // !WEBP_NEON_OMIT_C_CODE
|
||||||
|
@ -220,7 +220,7 @@ static void GradientUnfilter_C(const uint8_t* prev, const uint8_t* in,
|
||||||
int i;
|
int i;
|
||||||
for (i = 0; i < width; ++i) {
|
for (i = 0; i < width; ++i) {
|
||||||
top = prev[i]; // need to read this first, in case prev==out
|
top = prev[i]; // need to read this first, in case prev==out
|
||||||
left = in[i] + GradientPredictor_C(left, top, top_left);
|
left = (uint8_t)(in[i] + GradientPredictor_C(left, top, top_left));
|
||||||
top_left = top;
|
top_left = top;
|
||||||
out[i] = left;
|
out[i] = left;
|
||||||
}
|
}
|
||||||
|
|
|
@ -163,7 +163,8 @@ static void GradientPredictDirect_SSE2(const uint8_t* const row,
|
||||||
_mm_storel_epi64((__m128i*)(out + i), H);
|
_mm_storel_epi64((__m128i*)(out + i), H);
|
||||||
}
|
}
|
||||||
for (; i < length; ++i) {
|
for (; i < length; ++i) {
|
||||||
out[i] = row[i] - GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
|
const int delta = GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
|
||||||
|
out[i] = (uint8_t)(row[i] - delta);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -188,7 +189,7 @@ static WEBP_INLINE void DoGradientFilter_SSE2(const uint8_t* in,
|
||||||
|
|
||||||
// Filter line-by-line.
|
// Filter line-by-line.
|
||||||
while (row < last_row) {
|
while (row < last_row) {
|
||||||
out[0] = in[0] - in[-stride];
|
out[0] = (uint8_t)(in[0] - in[-stride]);
|
||||||
GradientPredictDirect_SSE2(in + 1, in + 1 - stride, out + 1, width - 1);
|
GradientPredictDirect_SSE2(in + 1, in + 1 - stride, out + 1, width - 1);
|
||||||
++row;
|
++row;
|
||||||
in += stride;
|
in += stride;
|
||||||
|
@ -223,7 +224,7 @@ static void HorizontalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
|
||||||
uint8_t* out, int width) {
|
uint8_t* out, int width) {
|
||||||
int i;
|
int i;
|
||||||
__m128i last;
|
__m128i last;
|
||||||
out[0] = in[0] + (prev == NULL ? 0 : prev[0]);
|
out[0] = (uint8_t)(in[0] + (prev == NULL ? 0 : prev[0]));
|
||||||
if (width <= 1) return;
|
if (width <= 1) return;
|
||||||
last = _mm_set_epi32(0, 0, 0, out[0]);
|
last = _mm_set_epi32(0, 0, 0, out[0]);
|
||||||
for (i = 1; i + 8 <= width; i += 8) {
|
for (i = 1; i + 8 <= width; i += 8) {
|
||||||
|
@ -238,7 +239,7 @@ static void HorizontalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
|
||||||
_mm_storel_epi64((__m128i*)(out + i), A7);
|
_mm_storel_epi64((__m128i*)(out + i), A7);
|
||||||
last = _mm_srli_epi64(A7, 56);
|
last = _mm_srli_epi64(A7, 56);
|
||||||
}
|
}
|
||||||
for (; i < width; ++i) out[i] = in[i] + out[i - 1];
|
for (; i < width; ++i) out[i] = (uint8_t)(in[i] + out[i - 1]);
|
||||||
}
|
}
|
||||||
|
|
||||||
static void VerticalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
|
static void VerticalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
|
||||||
|
@ -259,7 +260,7 @@ static void VerticalUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
|
||||||
_mm_storeu_si128((__m128i*)&out[i + 0], C0);
|
_mm_storeu_si128((__m128i*)&out[i + 0], C0);
|
||||||
_mm_storeu_si128((__m128i*)&out[i + 16], C1);
|
_mm_storeu_si128((__m128i*)&out[i + 16], C1);
|
||||||
}
|
}
|
||||||
for (; i < width; ++i) out[i] = in[i] + prev[i];
|
for (; i < width; ++i) out[i] = (uint8_t)(in[i] + prev[i]);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -296,7 +297,8 @@ static void GradientPredictInverse_SSE2(const uint8_t* const in,
|
||||||
_mm_storel_epi64((__m128i*)&row[i], out);
|
_mm_storel_epi64((__m128i*)&row[i], out);
|
||||||
}
|
}
|
||||||
for (; i < length; ++i) {
|
for (; i < length; ++i) {
|
||||||
row[i] = in[i] + GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
|
const int delta = GradientPredictor_SSE2(row[i - 1], top[i], top[i - 1]);
|
||||||
|
row[i] = (uint8_t)(in[i] + delta);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -306,7 +308,7 @@ static void GradientUnfilter_SSE2(const uint8_t* prev, const uint8_t* in,
|
||||||
if (prev == NULL) {
|
if (prev == NULL) {
|
||||||
HorizontalUnfilter_SSE2(NULL, in, out, width);
|
HorizontalUnfilter_SSE2(NULL, in, out, width);
|
||||||
} else {
|
} else {
|
||||||
out[0] = in[0] + prev[0]; // predict from above
|
out[0] = (uint8_t)(in[0] + prev[0]); // predict from above
|
||||||
GradientPredictInverse_SSE2(in + 1, prev + 1, out + 1, width - 1);
|
GradientPredictInverse_SSE2(in + 1, prev + 1, out + 1, width - 1);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
|
@ -270,14 +270,14 @@ void VP8LTransformColorInverse_C(const VP8LMultipliers* const m,
|
||||||
int i;
|
int i;
|
||||||
for (i = 0; i < num_pixels; ++i) {
|
for (i = 0; i < num_pixels; ++i) {
|
||||||
const uint32_t argb = src[i];
|
const uint32_t argb = src[i];
|
||||||
const uint32_t green = argb >> 8;
|
const int8_t green = (int8_t)(argb >> 8);
|
||||||
const uint32_t red = argb >> 16;
|
const uint32_t red = argb >> 16;
|
||||||
int new_red = red & 0xff;
|
int new_red = red & 0xff;
|
||||||
int new_blue = argb & 0xff;
|
int new_blue = argb & 0xff;
|
||||||
new_red += ColorTransformDelta(m->green_to_red_, green);
|
new_red += ColorTransformDelta(m->green_to_red_, green);
|
||||||
new_red &= 0xff;
|
new_red &= 0xff;
|
||||||
new_blue += ColorTransformDelta(m->green_to_blue_, green);
|
new_blue += ColorTransformDelta(m->green_to_blue_, green);
|
||||||
new_blue += ColorTransformDelta(m->red_to_blue_, new_red);
|
new_blue += ColorTransformDelta(m->red_to_blue_, (int8_t)new_red);
|
||||||
new_blue &= 0xff;
|
new_blue &= 0xff;
|
||||||
dst[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue);
|
dst[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue);
|
||||||
}
|
}
|
||||||
|
|
|
@ -515,13 +515,17 @@ static WEBP_INLINE int ColorTransformDelta(int8_t color_pred, int8_t color) {
|
||||||
return ((int)color_pred * color) >> 5;
|
return ((int)color_pred * color) >> 5;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
static WEBP_INLINE int8_t U32ToS8(uint32_t v) {
|
||||||
|
return (int8_t)(v & 0xff);
|
||||||
|
}
|
||||||
|
|
||||||
void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,
|
void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,
|
||||||
int num_pixels) {
|
int num_pixels) {
|
||||||
int i;
|
int i;
|
||||||
for (i = 0; i < num_pixels; ++i) {
|
for (i = 0; i < num_pixels; ++i) {
|
||||||
const uint32_t argb = data[i];
|
const uint32_t argb = data[i];
|
||||||
const uint32_t green = argb >> 8;
|
const int8_t green = U32ToS8(argb >> 8);
|
||||||
const uint32_t red = argb >> 16;
|
const int8_t red = U32ToS8(argb >> 16);
|
||||||
int new_red = red & 0xff;
|
int new_red = red & 0xff;
|
||||||
int new_blue = argb & 0xff;
|
int new_blue = argb & 0xff;
|
||||||
new_red -= ColorTransformDelta(m->green_to_red_, green);
|
new_red -= ColorTransformDelta(m->green_to_red_, green);
|
||||||
|
@ -535,7 +539,7 @@ void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,
|
||||||
|
|
||||||
static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
|
static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
|
||||||
uint32_t argb) {
|
uint32_t argb) {
|
||||||
const uint32_t green = argb >> 8;
|
const int8_t green = U32ToS8(argb >> 8);
|
||||||
int new_red = argb >> 16;
|
int new_red = argb >> 16;
|
||||||
new_red -= ColorTransformDelta(green_to_red, green);
|
new_red -= ColorTransformDelta(green_to_red, green);
|
||||||
return (new_red & 0xff);
|
return (new_red & 0xff);
|
||||||
|
@ -544,9 +548,9 @@ static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
|
||||||
static WEBP_INLINE uint8_t TransformColorBlue(uint8_t green_to_blue,
|
static WEBP_INLINE uint8_t TransformColorBlue(uint8_t green_to_blue,
|
||||||
uint8_t red_to_blue,
|
uint8_t red_to_blue,
|
||||||
uint32_t argb) {
|
uint32_t argb) {
|
||||||
const uint32_t green = argb >> 8;
|
const int8_t green = U32ToS8(argb >> 8);
|
||||||
const uint32_t red = argb >> 16;
|
const int8_t red = U32ToS8(argb >> 16);
|
||||||
uint8_t new_blue = argb;
|
uint8_t new_blue = argb & 0xff;
|
||||||
new_blue -= ColorTransformDelta(green_to_blue, green);
|
new_blue -= ColorTransformDelta(green_to_blue, green);
|
||||||
new_blue -= ColorTransformDelta(red_to_blue, red);
|
new_blue -= ColorTransformDelta(red_to_blue, red);
|
||||||
return (new_blue & 0xff);
|
return (new_blue & 0xff);
|
||||||
|
@ -558,7 +562,7 @@ void VP8LCollectColorRedTransforms_C(const uint32_t* argb, int stride,
|
||||||
while (tile_height-- > 0) {
|
while (tile_height-- > 0) {
|
||||||
int x;
|
int x;
|
||||||
for (x = 0; x < tile_width; ++x) {
|
for (x = 0; x < tile_width; ++x) {
|
||||||
++histo[TransformColorRed(green_to_red, argb[x])];
|
++histo[TransformColorRed((uint8_t)green_to_red, argb[x])];
|
||||||
}
|
}
|
||||||
argb += stride;
|
argb += stride;
|
||||||
}
|
}
|
||||||
|
@ -571,7 +575,8 @@ void VP8LCollectColorBlueTransforms_C(const uint32_t* argb, int stride,
|
||||||
while (tile_height-- > 0) {
|
while (tile_height-- > 0) {
|
||||||
int x;
|
int x;
|
||||||
for (x = 0; x < tile_width; ++x) {
|
for (x = 0; x < tile_width; ++x) {
|
||||||
++histo[TransformColorBlue(green_to_blue, red_to_blue, argb[x])];
|
++histo[TransformColorBlue((uint8_t)green_to_blue, (uint8_t)red_to_blue,
|
||||||
|
argb[x])];
|
||||||
}
|
}
|
||||||
argb += stride;
|
argb += stride;
|
||||||
}
|
}
|
||||||
|
|
|
@ -363,7 +363,7 @@ static void BundleColorMap_SSE2(const uint8_t* const row, int width, int xbits,
|
||||||
assert(xbits <= 3);
|
assert(xbits <= 3);
|
||||||
switch (xbits) {
|
switch (xbits) {
|
||||||
case 0: {
|
case 0: {
|
||||||
const __m128i ff = _mm_set1_epi16(0xff00);
|
const __m128i ff = _mm_set1_epi16((short)0xff00);
|
||||||
const __m128i zero = _mm_setzero_si128();
|
const __m128i zero = _mm_setzero_si128();
|
||||||
// Store 0xff000000 | (row[x] << 8).
|
// Store 0xff000000 | (row[x] << 8).
|
||||||
for (x = 0; x + 16 <= width; x += 16, dst += 16) {
|
for (x = 0; x + 16 <= width; x += 16, dst += 16) {
|
||||||
|
@ -382,7 +382,7 @@ static void BundleColorMap_SSE2(const uint8_t* const row, int width, int xbits,
|
||||||
break;
|
break;
|
||||||
}
|
}
|
||||||
case 1: {
|
case 1: {
|
||||||
const __m128i ff = _mm_set1_epi16(0xff00);
|
const __m128i ff = _mm_set1_epi16((short)0xff00);
|
||||||
const __m128i mul = _mm_set1_epi16(0x110);
|
const __m128i mul = _mm_set1_epi16(0x110);
|
||||||
for (x = 0; x + 16 <= width; x += 16, dst += 8) {
|
for (x = 0; x + 16 <= width; x += 16, dst += 8) {
|
||||||
// 0a0b | (where a/b are 4 bits).
|
// 0a0b | (where a/b are 4 bits).
|
||||||
|
|
|
@ -51,9 +51,9 @@ static void CollectColorBlueTransforms_SSE41(const uint32_t* argb, int stride,
|
||||||
int histo[]) {
|
int histo[]) {
|
||||||
const __m128i mults_r = _mm_set1_epi16(CST_5b(red_to_blue));
|
const __m128i mults_r = _mm_set1_epi16(CST_5b(red_to_blue));
|
||||||
const __m128i mults_g = _mm_set1_epi16(CST_5b(green_to_blue));
|
const __m128i mults_g = _mm_set1_epi16(CST_5b(green_to_blue));
|
||||||
const __m128i mask_g = _mm_set1_epi16(0xff00); // green mask
|
const __m128i mask_g = _mm_set1_epi16((short)0xff00); // green mask
|
||||||
const __m128i mask_gb = _mm_set1_epi32(0xffff); // green/blue mask
|
const __m128i mask_gb = _mm_set1_epi32(0xffff); // green/blue mask
|
||||||
const __m128i mask_b = _mm_set1_epi16(0x00ff); // blue mask
|
const __m128i mask_b = _mm_set1_epi16(0x00ff); // blue mask
|
||||||
const __m128i shuffler_lo = _mm_setr_epi8(-1, 2, -1, 6, -1, 10, -1, 14, -1,
|
const __m128i shuffler_lo = _mm_setr_epi8(-1, 2, -1, 6, -1, 10, -1, 14, -1,
|
||||||
-1, -1, -1, -1, -1, -1, -1);
|
-1, -1, -1, -1, -1, -1, -1);
|
||||||
const __m128i shuffler_hi = _mm_setr_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1,
|
const __m128i shuffler_hi = _mm_setr_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1,
|
||||||
|
|
|
@ -10,6 +10,8 @@
|
||||||
#ifndef WEBP_DSP_QUANT_H_
|
#ifndef WEBP_DSP_QUANT_H_
|
||||||
#define WEBP_DSP_QUANT_H_
|
#define WEBP_DSP_QUANT_H_
|
||||||
|
|
||||||
|
#include <string.h>
|
||||||
|
|
||||||
#include "src/dsp/dsp.h"
|
#include "src/dsp/dsp.h"
|
||||||
#include "src/webp/types.h"
|
#include "src/webp/types.h"
|
||||||
|
|
||||||
|
@ -67,4 +69,17 @@ static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
|
||||||
#endif // defined(WEBP_USE_NEON) && !defined(WEBP_ANDROID_NEON) &&
|
#endif // defined(WEBP_USE_NEON) && !defined(WEBP_ANDROID_NEON) &&
|
||||||
// !defined(WEBP_HAVE_NEON_RTCD)
|
// !defined(WEBP_HAVE_NEON_RTCD)
|
||||||
|
|
||||||
|
static WEBP_INLINE int IsFlatSource16(const uint8_t* src) {
|
||||||
|
const uint32_t v = src[0] * 0x01010101u;
|
||||||
|
int i;
|
||||||
|
for (i = 0; i < 16; ++i) {
|
||||||
|
if (memcmp(src + 0, &v, 4) || memcmp(src + 4, &v, 4) ||
|
||||||
|
memcmp(src + 8, &v, 4) || memcmp(src + 12, &v, 4)) {
|
||||||
|
return 0;
|
||||||
|
}
|
||||||
|
src += BPS;
|
||||||
|
}
|
||||||
|
return 1;
|
||||||
|
}
|
||||||
|
|
||||||
#endif // WEBP_DSP_QUANT_H_
|
#endif // WEBP_DSP_QUANT_H_
|
||||||
|
|
|
@ -109,8 +109,7 @@ void WebPRescalerExportRowExpand_C(WebPRescaler* const wrk) {
|
||||||
for (x_out = 0; x_out < x_out_max; ++x_out) {
|
for (x_out = 0; x_out < x_out_max; ++x_out) {
|
||||||
const uint32_t J = frow[x_out];
|
const uint32_t J = frow[x_out];
|
||||||
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
|
const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
|
||||||
|
@ -120,8 +119,7 @@ void WebPRescalerExportRowExpand_C(WebPRescaler* const wrk) {
|
||||||
+ (uint64_t)B * irow[x_out];
|
+ (uint64_t)B * irow[x_out];
|
||||||
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
||||||
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -138,17 +136,15 @@ void WebPRescalerExportRowShrink_C(WebPRescaler* const wrk) {
|
||||||
assert(!wrk->y_expand);
|
assert(!wrk->y_expand);
|
||||||
if (yscale) {
|
if (yscale) {
|
||||||
for (x_out = 0; x_out < x_out_max; ++x_out) {
|
for (x_out = 0; x_out < x_out_max; ++x_out) {
|
||||||
const uint32_t frac = (uint32_t)MULT_FIX(frow[x_out], yscale);
|
const uint32_t frac = (uint32_t)MULT_FIX_FLOOR(frow[x_out], yscale);
|
||||||
const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
|
const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
irow[x_out] = frac; // new fractional start
|
irow[x_out] = frac; // new fractional start
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
for (x_out = 0; x_out < x_out_max; ++x_out) {
|
for (x_out = 0; x_out < x_out_max; ++x_out) {
|
||||||
const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
|
const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
irow[x_out] = 0;
|
irow[x_out] = 0;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
|
@ -107,10 +107,9 @@ static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
|
||||||
);
|
);
|
||||||
}
|
}
|
||||||
for (i = 0; i < (x_out_max & 0x3); ++i) {
|
for (i = 0; i < (x_out_max & 0x3); ++i) {
|
||||||
const uint32_t frac = (uint32_t)MULT_FIX(*frow++, yscale);
|
const uint32_t frac = (uint32_t)MULT_FIX_FLOOR(*frow++, yscale);
|
||||||
const int v = (int)MULT_FIX_FLOOR(*irow - frac, wrk->fxy_scale);
|
const int v = (int)MULT_FIX(*irow - frac, wrk->fxy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
*dst++ = (v > 255) ? 255u : (uint8_t)v;
|
||||||
*dst++ = v;
|
|
||||||
*irow++ = frac; // new fractional start
|
*irow++ = frac; // new fractional start
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
|
@ -157,8 +156,7 @@ static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
|
||||||
}
|
}
|
||||||
for (i = 0; i < (x_out_max & 0x3); ++i) {
|
for (i = 0; i < (x_out_max & 0x3); ++i) {
|
||||||
const int v = (int)MULT_FIX_FLOOR(*irow, wrk->fxy_scale);
|
const int v = (int)MULT_FIX_FLOOR(*irow, wrk->fxy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
*dst++ = (v > 255) ? 255u : (uint8_t)v;
|
||||||
*dst++ = v;
|
|
||||||
*irow++ = 0;
|
*irow++ = 0;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -219,8 +217,7 @@ static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
|
||||||
for (i = 0; i < (x_out_max & 0x3); ++i) {
|
for (i = 0; i < (x_out_max & 0x3); ++i) {
|
||||||
const uint32_t J = *frow++;
|
const uint32_t J = *frow++;
|
||||||
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
*dst++ = (v > 255) ? 255u : (uint8_t)v;
|
||||||
*dst++ = v;
|
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
|
const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
|
||||||
|
@ -291,8 +288,7 @@ static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
|
||||||
+ (uint64_t)B * *irow++;
|
+ (uint64_t)B * *irow++;
|
||||||
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
||||||
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
*dst++ = (v > 255) ? 255u : (uint8_t)v;
|
||||||
*dst++ = v;
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
|
@ -166,8 +166,7 @@ static WEBP_INLINE void ExportRowExpand_0(const uint32_t* frow, uint8_t* dst,
|
||||||
for (x_out = 0; x_out < length; ++x_out) {
|
for (x_out = 0; x_out < length; ++x_out) {
|
||||||
const uint32_t J = frow[x_out];
|
const uint32_t J = frow[x_out];
|
||||||
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -241,8 +240,7 @@ static WEBP_INLINE void ExportRowExpand_1(const uint32_t* frow, uint32_t* irow,
|
||||||
+ (uint64_t)B * irow[x_out];
|
+ (uint64_t)B * irow[x_out];
|
||||||
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
||||||
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -342,10 +340,9 @@ static WEBP_INLINE void ExportRowShrink_0(const uint32_t* frow, uint32_t* irow,
|
||||||
length -= 4;
|
length -= 4;
|
||||||
}
|
}
|
||||||
for (x_out = 0; x_out < length; ++x_out) {
|
for (x_out = 0; x_out < length; ++x_out) {
|
||||||
const uint32_t frac = (uint32_t)MULT_FIX(frow[x_out], yscale);
|
const uint32_t frac = (uint32_t)MULT_FIX_FLOOR(frow[x_out], yscale);
|
||||||
const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
|
const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
irow[x_out] = frac;
|
irow[x_out] = frac;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -406,8 +403,7 @@ static WEBP_INLINE void ExportRowShrink_1(uint32_t* irow, uint8_t* dst,
|
||||||
}
|
}
|
||||||
for (x_out = 0; x_out < length; ++x_out) {
|
for (x_out = 0; x_out < length; ++x_out) {
|
||||||
const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
|
const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
irow[x_out] = 0;
|
irow[x_out] = 0;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
|
@ -81,14 +81,13 @@ static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
|
||||||
const uint32x4_t B1 = MULT_FIX(A1, fy_scale_half);
|
const uint32x4_t B1 = MULT_FIX(A1, fy_scale_half);
|
||||||
const uint16x4_t C0 = vmovn_u32(B0);
|
const uint16x4_t C0 = vmovn_u32(B0);
|
||||||
const uint16x4_t C1 = vmovn_u32(B1);
|
const uint16x4_t C1 = vmovn_u32(B1);
|
||||||
const uint8x8_t D = vmovn_u16(vcombine_u16(C0, C1));
|
const uint8x8_t D = vqmovn_u16(vcombine_u16(C0, C1));
|
||||||
vst1_u8(dst + x_out, D);
|
vst1_u8(dst + x_out, D);
|
||||||
}
|
}
|
||||||
for (; x_out < x_out_max; ++x_out) {
|
for (; x_out < x_out_max; ++x_out) {
|
||||||
const uint32_t J = frow[x_out];
|
const uint32_t J = frow[x_out];
|
||||||
const int v = (int)MULT_FIX_C(J, fy_scale);
|
const int v = (int)MULT_FIX_C(J, fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
|
const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
|
||||||
|
@ -102,7 +101,7 @@ static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
|
||||||
const uint32x4_t D1 = MULT_FIX(C1, fy_scale_half);
|
const uint32x4_t D1 = MULT_FIX(C1, fy_scale_half);
|
||||||
const uint16x4_t E0 = vmovn_u32(D0);
|
const uint16x4_t E0 = vmovn_u32(D0);
|
||||||
const uint16x4_t E1 = vmovn_u32(D1);
|
const uint16x4_t E1 = vmovn_u32(D1);
|
||||||
const uint8x8_t F = vmovn_u16(vcombine_u16(E0, E1));
|
const uint8x8_t F = vqmovn_u16(vcombine_u16(E0, E1));
|
||||||
vst1_u8(dst + x_out, F);
|
vst1_u8(dst + x_out, F);
|
||||||
}
|
}
|
||||||
for (; x_out < x_out_max; ++x_out) {
|
for (; x_out < x_out_max; ++x_out) {
|
||||||
|
@ -110,8 +109,7 @@ static void RescalerExportRowExpand_NEON(WebPRescaler* const wrk) {
|
||||||
+ (uint64_t)B * irow[x_out];
|
+ (uint64_t)B * irow[x_out];
|
||||||
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
||||||
const int v = (int)MULT_FIX_C(J, fy_scale);
|
const int v = (int)MULT_FIX_C(J, fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -135,23 +133,22 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
|
||||||
for (x_out = 0; x_out < max_span; x_out += 8) {
|
for (x_out = 0; x_out < max_span; x_out += 8) {
|
||||||
LOAD_32x8(frow + x_out, in0, in1);
|
LOAD_32x8(frow + x_out, in0, in1);
|
||||||
LOAD_32x8(irow + x_out, in2, in3);
|
LOAD_32x8(irow + x_out, in2, in3);
|
||||||
const uint32x4_t A0 = MULT_FIX(in0, yscale_half);
|
const uint32x4_t A0 = MULT_FIX_FLOOR(in0, yscale_half);
|
||||||
const uint32x4_t A1 = MULT_FIX(in1, yscale_half);
|
const uint32x4_t A1 = MULT_FIX_FLOOR(in1, yscale_half);
|
||||||
const uint32x4_t B0 = vqsubq_u32(in2, A0);
|
const uint32x4_t B0 = vqsubq_u32(in2, A0);
|
||||||
const uint32x4_t B1 = vqsubq_u32(in3, A1);
|
const uint32x4_t B1 = vqsubq_u32(in3, A1);
|
||||||
const uint32x4_t C0 = MULT_FIX_FLOOR(B0, fxy_scale_half);
|
const uint32x4_t C0 = MULT_FIX(B0, fxy_scale_half);
|
||||||
const uint32x4_t C1 = MULT_FIX_FLOOR(B1, fxy_scale_half);
|
const uint32x4_t C1 = MULT_FIX(B1, fxy_scale_half);
|
||||||
const uint16x4_t D0 = vmovn_u32(C0);
|
const uint16x4_t D0 = vmovn_u32(C0);
|
||||||
const uint16x4_t D1 = vmovn_u32(C1);
|
const uint16x4_t D1 = vmovn_u32(C1);
|
||||||
const uint8x8_t E = vmovn_u16(vcombine_u16(D0, D1));
|
const uint8x8_t E = vqmovn_u16(vcombine_u16(D0, D1));
|
||||||
vst1_u8(dst + x_out, E);
|
vst1_u8(dst + x_out, E);
|
||||||
STORE_32x8(A0, A1, irow + x_out);
|
STORE_32x8(A0, A1, irow + x_out);
|
||||||
}
|
}
|
||||||
for (; x_out < x_out_max; ++x_out) {
|
for (; x_out < x_out_max; ++x_out) {
|
||||||
const uint32_t frac = (uint32_t)MULT_FIX_C(frow[x_out], yscale);
|
const uint32_t frac = (uint32_t)MULT_FIX_FLOOR_C(frow[x_out], yscale);
|
||||||
const int v = (int)MULT_FIX_FLOOR_C(irow[x_out] - frac, fxy_scale);
|
const int v = (int)MULT_FIX_C(irow[x_out] - frac, fxy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
irow[x_out] = frac; // new fractional start
|
irow[x_out] = frac; // new fractional start
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
|
@ -161,14 +158,13 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
|
||||||
const uint32x4_t A1 = MULT_FIX(in1, fxy_scale_half);
|
const uint32x4_t A1 = MULT_FIX(in1, fxy_scale_half);
|
||||||
const uint16x4_t B0 = vmovn_u32(A0);
|
const uint16x4_t B0 = vmovn_u32(A0);
|
||||||
const uint16x4_t B1 = vmovn_u32(A1);
|
const uint16x4_t B1 = vmovn_u32(A1);
|
||||||
const uint8x8_t C = vmovn_u16(vcombine_u16(B0, B1));
|
const uint8x8_t C = vqmovn_u16(vcombine_u16(B0, B1));
|
||||||
vst1_u8(dst + x_out, C);
|
vst1_u8(dst + x_out, C);
|
||||||
STORE_32x8(zero, zero, irow + x_out);
|
STORE_32x8(zero, zero, irow + x_out);
|
||||||
}
|
}
|
||||||
for (; x_out < x_out_max; ++x_out) {
|
for (; x_out < x_out_max; ++x_out) {
|
||||||
const int v = (int)MULT_FIX_C(irow[x_out], fxy_scale);
|
const int v = (int)MULT_FIX_C(irow[x_out], fxy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
irow[x_out] = 0;
|
irow[x_out] = 0;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
|
@ -225,35 +225,6 @@ static WEBP_INLINE void ProcessRow_SSE2(const __m128i* const A0,
|
||||||
_mm_storel_epi64((__m128i*)dst, G);
|
_mm_storel_epi64((__m128i*)dst, G);
|
||||||
}
|
}
|
||||||
|
|
||||||
static WEBP_INLINE void ProcessRow_Floor_SSE2(const __m128i* const A0,
|
|
||||||
const __m128i* const A1,
|
|
||||||
const __m128i* const A2,
|
|
||||||
const __m128i* const A3,
|
|
||||||
const __m128i* const mult,
|
|
||||||
uint8_t* const dst) {
|
|
||||||
const __m128i mask = _mm_set_epi32(0xffffffffu, 0, 0xffffffffu, 0);
|
|
||||||
const __m128i B0 = _mm_mul_epu32(*A0, *mult);
|
|
||||||
const __m128i B1 = _mm_mul_epu32(*A1, *mult);
|
|
||||||
const __m128i B2 = _mm_mul_epu32(*A2, *mult);
|
|
||||||
const __m128i B3 = _mm_mul_epu32(*A3, *mult);
|
|
||||||
const __m128i D0 = _mm_srli_epi64(B0, WEBP_RESCALER_RFIX);
|
|
||||||
const __m128i D1 = _mm_srli_epi64(B1, WEBP_RESCALER_RFIX);
|
|
||||||
#if (WEBP_RESCALER_RFIX < 32)
|
|
||||||
const __m128i D2 =
|
|
||||||
_mm_and_si128(_mm_slli_epi64(B2, 32 - WEBP_RESCALER_RFIX), mask);
|
|
||||||
const __m128i D3 =
|
|
||||||
_mm_and_si128(_mm_slli_epi64(B3, 32 - WEBP_RESCALER_RFIX), mask);
|
|
||||||
#else
|
|
||||||
const __m128i D2 = _mm_and_si128(B2, mask);
|
|
||||||
const __m128i D3 = _mm_and_si128(B3, mask);
|
|
||||||
#endif
|
|
||||||
const __m128i E0 = _mm_or_si128(D0, D2);
|
|
||||||
const __m128i E1 = _mm_or_si128(D1, D3);
|
|
||||||
const __m128i F = _mm_packs_epi32(E0, E1);
|
|
||||||
const __m128i G = _mm_packus_epi16(F, F);
|
|
||||||
_mm_storel_epi64((__m128i*)dst, G);
|
|
||||||
}
|
|
||||||
|
|
||||||
static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
|
static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
|
||||||
int x_out;
|
int x_out;
|
||||||
uint8_t* const dst = wrk->dst;
|
uint8_t* const dst = wrk->dst;
|
||||||
|
@ -274,8 +245,7 @@ static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
|
||||||
for (; x_out < x_out_max; ++x_out) {
|
for (; x_out < x_out_max; ++x_out) {
|
||||||
const uint32_t J = frow[x_out];
|
const uint32_t J = frow[x_out];
|
||||||
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
|
const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
|
||||||
|
@ -308,8 +278,7 @@ static void RescalerExportRowExpand_SSE2(WebPRescaler* const wrk) {
|
||||||
+ (uint64_t)B * irow[x_out];
|
+ (uint64_t)B * irow[x_out];
|
||||||
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
|
||||||
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
const int v = (int)MULT_FIX(J, wrk->fy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -328,20 +297,15 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
|
||||||
const int scale_xy = wrk->fxy_scale;
|
const int scale_xy = wrk->fxy_scale;
|
||||||
const __m128i mult_xy = _mm_set_epi32(0, scale_xy, 0, scale_xy);
|
const __m128i mult_xy = _mm_set_epi32(0, scale_xy, 0, scale_xy);
|
||||||
const __m128i mult_y = _mm_set_epi32(0, yscale, 0, yscale);
|
const __m128i mult_y = _mm_set_epi32(0, yscale, 0, yscale);
|
||||||
const __m128i rounder = _mm_set_epi32(0, ROUNDER, 0, ROUNDER);
|
|
||||||
for (x_out = 0; x_out + 8 <= x_out_max; x_out += 8) {
|
for (x_out = 0; x_out + 8 <= x_out_max; x_out += 8) {
|
||||||
__m128i A0, A1, A2, A3, B0, B1, B2, B3;
|
__m128i A0, A1, A2, A3, B0, B1, B2, B3;
|
||||||
LoadDispatchAndMult_SSE2(irow + x_out, NULL, &A0, &A1, &A2, &A3);
|
LoadDispatchAndMult_SSE2(irow + x_out, NULL, &A0, &A1, &A2, &A3);
|
||||||
LoadDispatchAndMult_SSE2(frow + x_out, &mult_y, &B0, &B1, &B2, &B3);
|
LoadDispatchAndMult_SSE2(frow + x_out, &mult_y, &B0, &B1, &B2, &B3);
|
||||||
{
|
{
|
||||||
const __m128i C0 = _mm_add_epi64(B0, rounder);
|
const __m128i D0 = _mm_srli_epi64(B0, WEBP_RESCALER_RFIX); // = frac
|
||||||
const __m128i C1 = _mm_add_epi64(B1, rounder);
|
const __m128i D1 = _mm_srli_epi64(B1, WEBP_RESCALER_RFIX);
|
||||||
const __m128i C2 = _mm_add_epi64(B2, rounder);
|
const __m128i D2 = _mm_srli_epi64(B2, WEBP_RESCALER_RFIX);
|
||||||
const __m128i C3 = _mm_add_epi64(B3, rounder);
|
const __m128i D3 = _mm_srli_epi64(B3, WEBP_RESCALER_RFIX);
|
||||||
const __m128i D0 = _mm_srli_epi64(C0, WEBP_RESCALER_RFIX); // = frac
|
|
||||||
const __m128i D1 = _mm_srli_epi64(C1, WEBP_RESCALER_RFIX);
|
|
||||||
const __m128i D2 = _mm_srli_epi64(C2, WEBP_RESCALER_RFIX);
|
|
||||||
const __m128i D3 = _mm_srli_epi64(C3, WEBP_RESCALER_RFIX);
|
|
||||||
const __m128i E0 = _mm_sub_epi64(A0, D0); // irow[x] - frac
|
const __m128i E0 = _mm_sub_epi64(A0, D0); // irow[x] - frac
|
||||||
const __m128i E1 = _mm_sub_epi64(A1, D1);
|
const __m128i E1 = _mm_sub_epi64(A1, D1);
|
||||||
const __m128i E2 = _mm_sub_epi64(A2, D2);
|
const __m128i E2 = _mm_sub_epi64(A2, D2);
|
||||||
|
@ -352,14 +316,13 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
|
||||||
const __m128i G1 = _mm_or_si128(D1, F3);
|
const __m128i G1 = _mm_or_si128(D1, F3);
|
||||||
_mm_storeu_si128((__m128i*)(irow + x_out + 0), G0);
|
_mm_storeu_si128((__m128i*)(irow + x_out + 0), G0);
|
||||||
_mm_storeu_si128((__m128i*)(irow + x_out + 4), G1);
|
_mm_storeu_si128((__m128i*)(irow + x_out + 4), G1);
|
||||||
ProcessRow_Floor_SSE2(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
|
ProcessRow_SSE2(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
for (; x_out < x_out_max; ++x_out) {
|
for (; x_out < x_out_max; ++x_out) {
|
||||||
const uint32_t frac = (int)MULT_FIX(frow[x_out], yscale);
|
const uint32_t frac = (int)MULT_FIX_FLOOR(frow[x_out], yscale);
|
||||||
const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
|
const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
irow[x_out] = frac; // new fractional start
|
irow[x_out] = frac; // new fractional start
|
||||||
}
|
}
|
||||||
} else {
|
} else {
|
||||||
|
@ -375,8 +338,7 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
|
||||||
}
|
}
|
||||||
for (; x_out < x_out_max; ++x_out) {
|
for (; x_out < x_out_max; ++x_out) {
|
||||||
const int v = (int)MULT_FIX(irow[x_out], scale);
|
const int v = (int)MULT_FIX(irow[x_out], scale);
|
||||||
assert(v >= 0 && v <= 255);
|
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
|
||||||
dst[x_out] = v;
|
|
||||||
irow[x_out] = 0;
|
irow[x_out] = 0;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
|
@ -191,13 +191,14 @@ void VP8LHashChainClear(VP8LHashChain* const p) {
|
||||||
|
|
||||||
// -----------------------------------------------------------------------------
|
// -----------------------------------------------------------------------------
|
||||||
|
|
||||||
#define HASH_MULTIPLIER_HI (0xc6a4a793ULL)
|
static const uint32_t kHashMultiplierHi = 0xc6a4a793u;
|
||||||
#define HASH_MULTIPLIER_LO (0x5bd1e996ULL)
|
static const uint32_t kHashMultiplierLo = 0x5bd1e996u;
|
||||||
|
|
||||||
static WEBP_INLINE uint32_t GetPixPairHash64(const uint32_t* const argb) {
|
static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
|
||||||
|
uint32_t GetPixPairHash64(const uint32_t* const argb) {
|
||||||
uint32_t key;
|
uint32_t key;
|
||||||
key = (argb[1] * HASH_MULTIPLIER_HI) & 0xffffffffu;
|
key = argb[1] * kHashMultiplierHi;
|
||||||
key += (argb[0] * HASH_MULTIPLIER_LO) & 0xffffffffu;
|
key += argb[0] * kHashMultiplierLo;
|
||||||
key = key >> (32 - HASH_BITS);
|
key = key >> (32 - HASH_BITS);
|
||||||
return key;
|
return key;
|
||||||
}
|
}
|
||||||
|
|
|
@ -929,9 +929,8 @@ static int HistogramCombineStochastic(VP8LHistogramSet* const image_histo,
|
||||||
}
|
}
|
||||||
|
|
||||||
mappings = (int*) WebPSafeMalloc(*num_used, sizeof(*mappings));
|
mappings = (int*) WebPSafeMalloc(*num_used, sizeof(*mappings));
|
||||||
if (mappings == NULL || !HistoQueueInit(&histo_queue, kHistoQueueSize)) {
|
if (mappings == NULL) return 0;
|
||||||
goto End;
|
if (!HistoQueueInit(&histo_queue, kHistoQueueSize)) goto End;
|
||||||
}
|
|
||||||
// Fill the initial mapping.
|
// Fill the initial mapping.
|
||||||
for (j = 0, iter = 0; iter < image_histo->size; ++iter) {
|
for (j = 0, iter = 0; iter < image_histo->size; ++iter) {
|
||||||
if (histograms[iter] == NULL) continue;
|
if (histograms[iter] == NULL) continue;
|
||||||
|
|
|
@ -202,7 +202,7 @@ static uint32_t NearLossless(uint32_t value, uint32_t predict,
|
||||||
}
|
}
|
||||||
if ((value >> 24) == 0 || (value >> 24) == 0xff) {
|
if ((value >> 24) == 0 || (value >> 24) == 0xff) {
|
||||||
// Preserve transparency of fully transparent or fully opaque pixels.
|
// Preserve transparency of fully transparent or fully opaque pixels.
|
||||||
a = NearLosslessDiff(value >> 24, predict >> 24);
|
a = NearLosslessDiff((value >> 24) & 0xff, (predict >> 24) & 0xff);
|
||||||
} else {
|
} else {
|
||||||
a = NearLosslessComponent(value >> 24, predict >> 24, 0xff, quantization);
|
a = NearLosslessComponent(value >> 24, predict >> 24, 0xff, quantization);
|
||||||
}
|
}
|
||||||
|
@ -215,12 +215,12 @@ static uint32_t NearLossless(uint32_t value, uint32_t predict,
|
||||||
// The amount by which green has been adjusted during quantization. It is
|
// The amount by which green has been adjusted during quantization. It is
|
||||||
// subtracted from red and blue for compensation, to avoid accumulating two
|
// subtracted from red and blue for compensation, to avoid accumulating two
|
||||||
// quantization errors in them.
|
// quantization errors in them.
|
||||||
green_diff = NearLosslessDiff(new_green, value >> 8);
|
green_diff = NearLosslessDiff(new_green, (value >> 8) & 0xff);
|
||||||
}
|
}
|
||||||
r = NearLosslessComponent(NearLosslessDiff(value >> 16, green_diff),
|
r = NearLosslessComponent(NearLosslessDiff((value >> 16) & 0xff, green_diff),
|
||||||
(predict >> 16) & 0xff, 0xff - new_green,
|
(predict >> 16) & 0xff, 0xff - new_green,
|
||||||
quantization);
|
quantization);
|
||||||
b = NearLosslessComponent(NearLosslessDiff(value, green_diff),
|
b = NearLosslessComponent(NearLosslessDiff(value & 0xff, green_diff),
|
||||||
predict & 0xff, 0xff - new_green, quantization);
|
predict & 0xff, 0xff - new_green, quantization);
|
||||||
return ((uint32_t)a << 24) | ((uint32_t)r << 16) | ((uint32_t)g << 8) | b;
|
return ((uint32_t)a << 24) | ((uint32_t)r << 16) | ((uint32_t)g << 8) | b;
|
||||||
}
|
}
|
||||||
|
@ -587,7 +587,7 @@ static void GetBestGreenToRed(
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
best_tx->green_to_red_ = green_to_red_best;
|
best_tx->green_to_red_ = (green_to_red_best & 0xff);
|
||||||
}
|
}
|
||||||
|
|
||||||
static float GetPredictionCostCrossColorBlue(
|
static float GetPredictionCostCrossColorBlue(
|
||||||
|
@ -666,8 +666,8 @@ static void GetBestGreenRedToBlue(
|
||||||
break; // out of iter-loop.
|
break; // out of iter-loop.
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
best_tx->green_to_blue_ = green_to_blue_best;
|
best_tx->green_to_blue_ = green_to_blue_best & 0xff;
|
||||||
best_tx->red_to_blue_ = red_to_blue_best;
|
best_tx->red_to_blue_ = red_to_blue_best & 0xff;
|
||||||
}
|
}
|
||||||
#undef kGreenRedToBlueMaxIters
|
#undef kGreenRedToBlueMaxIters
|
||||||
#undef kGreenRedToBlueNumAxis
|
#undef kGreenRedToBlueNumAxis
|
||||||
|
|
|
@ -33,7 +33,7 @@
|
||||||
|
|
||||||
// number of non-zero coeffs below which we consider the block very flat
|
// number of non-zero coeffs below which we consider the block very flat
|
||||||
// (and apply a penalty to complex predictions)
|
// (and apply a penalty to complex predictions)
|
||||||
#define FLATNESS_LIMIT_I16 10 // I16 mode
|
#define FLATNESS_LIMIT_I16 0 // I16 mode (special case)
|
||||||
#define FLATNESS_LIMIT_I4 3 // I4 mode
|
#define FLATNESS_LIMIT_I4 3 // I4 mode
|
||||||
#define FLATNESS_LIMIT_UV 2 // UV mode
|
#define FLATNESS_LIMIT_UV 2 // UV mode
|
||||||
#define FLATNESS_PENALTY 140 // roughly ~1bit per block
|
#define FLATNESS_PENALTY 140 // roughly ~1bit per block
|
||||||
|
@ -988,6 +988,7 @@ static void PickBestIntra16(VP8EncIterator* const it, VP8ModeScore* rd) {
|
||||||
VP8ModeScore* rd_cur = &rd_tmp;
|
VP8ModeScore* rd_cur = &rd_tmp;
|
||||||
VP8ModeScore* rd_best = rd;
|
VP8ModeScore* rd_best = rd;
|
||||||
int mode;
|
int mode;
|
||||||
|
int is_flat = IsFlatSource16(it->yuv_in_ + Y_OFF_ENC);
|
||||||
|
|
||||||
rd->mode_i16 = -1;
|
rd->mode_i16 = -1;
|
||||||
for (mode = 0; mode < NUM_PRED_MODES; ++mode) {
|
for (mode = 0; mode < NUM_PRED_MODES; ++mode) {
|
||||||
|
@ -1003,10 +1004,14 @@ static void PickBestIntra16(VP8EncIterator* const it, VP8ModeScore* rd) {
|
||||||
tlambda ? MULT_8B(tlambda, VP8TDisto16x16(src, tmp_dst, kWeightY)) : 0;
|
tlambda ? MULT_8B(tlambda, VP8TDisto16x16(src, tmp_dst, kWeightY)) : 0;
|
||||||
rd_cur->H = VP8FixedCostsI16[mode];
|
rd_cur->H = VP8FixedCostsI16[mode];
|
||||||
rd_cur->R = VP8GetCostLuma16(it, rd_cur);
|
rd_cur->R = VP8GetCostLuma16(it, rd_cur);
|
||||||
if (mode > 0 &&
|
if (is_flat) {
|
||||||
IsFlat(rd_cur->y_ac_levels[0], kNumBlocks, FLATNESS_LIMIT_I16)) {
|
// refine the first impression (which was in pixel space)
|
||||||
// penalty to avoid flat area to be mispredicted by complex mode
|
is_flat = IsFlat(rd_cur->y_ac_levels[0], kNumBlocks, FLATNESS_LIMIT_I16);
|
||||||
rd_cur->R += FLATNESS_PENALTY * kNumBlocks;
|
if (is_flat) {
|
||||||
|
// Block is very flat. We put emphasis on the distortion being very low!
|
||||||
|
rd_cur->D *= 2;
|
||||||
|
rd_cur->SD *= 2;
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
// Since we always examine Intra16 first, we can overwrite *rd directly.
|
// Since we always examine Intra16 first, we can overwrite *rd directly.
|
||||||
|
@ -1087,7 +1092,8 @@ static int PickBestIntra4(VP8EncIterator* const it, VP8ModeScore* const rd) {
|
||||||
: 0;
|
: 0;
|
||||||
rd_tmp.H = mode_costs[mode];
|
rd_tmp.H = mode_costs[mode];
|
||||||
|
|
||||||
// Add flatness penalty
|
// Add flatness penalty, to avoid flat area to be mispredicted
|
||||||
|
// by a complex mode.
|
||||||
if (mode > 0 && IsFlat(tmp_levels, kNumBlocks, FLATNESS_LIMIT_I4)) {
|
if (mode > 0 && IsFlat(tmp_levels, kNumBlocks, FLATNESS_LIMIT_I4)) {
|
||||||
rd_tmp.R = FLATNESS_PENALTY * kNumBlocks;
|
rd_tmp.R = FLATNESS_PENALTY * kNumBlocks;
|
||||||
} else {
|
} else {
|
||||||
|
@ -1242,11 +1248,19 @@ static void RefineUsingDistortion(VP8EncIterator* const it,
|
||||||
if (mode > 0 && VP8FixedCostsI16[mode] > bit_limit) {
|
if (mode > 0 && VP8FixedCostsI16[mode] > bit_limit) {
|
||||||
continue;
|
continue;
|
||||||
}
|
}
|
||||||
|
|
||||||
if (score < best_score) {
|
if (score < best_score) {
|
||||||
best_mode = mode;
|
best_mode = mode;
|
||||||
best_score = score;
|
best_score = score;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
if (it->x_ == 0 || it->y_ == 0) {
|
||||||
|
// avoid starting a checkerboard resonance from the border. See bug #432.
|
||||||
|
if (IsFlatSource16(src)) {
|
||||||
|
best_mode = (it->x_ == 0) ? 0 : 2;
|
||||||
|
try_both_modes = 0; // stick to i16
|
||||||
|
}
|
||||||
|
}
|
||||||
VP8SetIntra16Mode(it, best_mode);
|
VP8SetIntra16Mode(it, best_mode);
|
||||||
// we'll reconstruct later, if i16 mode actually gets selected
|
// we'll reconstruct later, if i16 mode actually gets selected
|
||||||
}
|
}
|
||||||
|
|
|
@ -32,7 +32,7 @@ extern "C" {
|
||||||
// version numbers
|
// version numbers
|
||||||
#define ENC_MAJ_VERSION 1
|
#define ENC_MAJ_VERSION 1
|
||||||
#define ENC_MIN_VERSION 0
|
#define ENC_MIN_VERSION 0
|
||||||
#define ENC_REV_VERSION 2
|
#define ENC_REV_VERSION 3
|
||||||
|
|
||||||
enum { MAX_LF_LEVELS = 64, // Maximum loop filter level
|
enum { MAX_LF_LEVELS = 64, // Maximum loop filter level
|
||||||
MAX_VARIABLE_LEVEL = 67, // last (inclusive) level with variable cost
|
MAX_VARIABLE_LEVEL = 67, // last (inclusive) level with variable cost
|
||||||
|
|
|
@ -29,7 +29,7 @@ extern "C" {
|
||||||
|
|
||||||
#define MUX_MAJ_VERSION 1
|
#define MUX_MAJ_VERSION 1
|
||||||
#define MUX_MIN_VERSION 0
|
#define MUX_MIN_VERSION 0
|
||||||
#define MUX_REV_VERSION 2
|
#define MUX_REV_VERSION 3
|
||||||
|
|
||||||
// Chunk object.
|
// Chunk object.
|
||||||
typedef struct WebPChunk WebPChunk;
|
typedef struct WebPChunk WebPChunk;
|
||||||
|
|
|
@ -104,7 +104,8 @@ void VP8LoadNewBytes(VP8BitReader* const br) {
|
||||||
}
|
}
|
||||||
|
|
||||||
// Read a bit with proba 'prob'. Speed-critical function!
|
// Read a bit with proba 'prob'. Speed-critical function!
|
||||||
static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) {
|
static WEBP_INLINE int VP8GetBit(VP8BitReader* const br,
|
||||||
|
int prob, const char label[]) {
|
||||||
// Don't move this declaration! It makes a big speed difference to store
|
// Don't move this declaration! It makes a big speed difference to store
|
||||||
// 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
|
// 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
|
||||||
// alter br->range_ value.
|
// alter br->range_ value.
|
||||||
|
@ -129,13 +130,14 @@ static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) {
|
||||||
br->bits_ -= shift;
|
br->bits_ -= shift;
|
||||||
}
|
}
|
||||||
br->range_ = range - 1;
|
br->range_ = range - 1;
|
||||||
|
BT_TRACK(br);
|
||||||
return bit;
|
return bit;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
// simplified version of VP8GetBit() for prob=0x80 (note shift is always 1 here)
|
// simplified version of VP8GetBit() for prob=0x80 (note shift is always 1 here)
|
||||||
static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
|
static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
|
||||||
int VP8GetSigned(VP8BitReader* const br, int v) {
|
int VP8GetSigned(VP8BitReader* const br, int v, const char label[]) {
|
||||||
if (br->bits_ < 0) {
|
if (br->bits_ < 0) {
|
||||||
VP8LoadNewBytes(br);
|
VP8LoadNewBytes(br);
|
||||||
}
|
}
|
||||||
|
@ -148,11 +150,13 @@ int VP8GetSigned(VP8BitReader* const br, int v) {
|
||||||
br->range_ += mask;
|
br->range_ += mask;
|
||||||
br->range_ |= 1;
|
br->range_ |= 1;
|
||||||
br->value_ -= (bit_t)((split + 1) & mask) << pos;
|
br->value_ -= (bit_t)((split + 1) & mask) << pos;
|
||||||
|
BT_TRACK(br);
|
||||||
return (v ^ mask) - mask;
|
return (v ^ mask) - mask;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br, int prob) {
|
static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br,
|
||||||
|
int prob, const char label[]) {
|
||||||
// Don't move this declaration! It makes a big speed difference to store
|
// Don't move this declaration! It makes a big speed difference to store
|
||||||
// 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
|
// 'range' *before* calling VP8LoadNewBytes(), even if this function doesn't
|
||||||
// alter br->range_ value.
|
// alter br->range_ value.
|
||||||
|
@ -179,6 +183,7 @@ static WEBP_INLINE int VP8GetBitAlt(VP8BitReader* const br, int prob) {
|
||||||
br->bits_ -= shift;
|
br->bits_ -= shift;
|
||||||
}
|
}
|
||||||
br->range_ = range;
|
br->range_ = range;
|
||||||
|
BT_TRACK(br);
|
||||||
return bit;
|
return bit;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
|
@ -109,17 +109,18 @@ void VP8LoadFinalBytes(VP8BitReader* const br) {
|
||||||
//------------------------------------------------------------------------------
|
//------------------------------------------------------------------------------
|
||||||
// Higher-level calls
|
// Higher-level calls
|
||||||
|
|
||||||
uint32_t VP8GetValue(VP8BitReader* const br, int bits) {
|
uint32_t VP8GetValue(VP8BitReader* const br, int bits, const char label[]) {
|
||||||
uint32_t v = 0;
|
uint32_t v = 0;
|
||||||
while (bits-- > 0) {
|
while (bits-- > 0) {
|
||||||
v |= VP8GetBit(br, 0x80) << bits;
|
v |= VP8GetBit(br, 0x80, label) << bits;
|
||||||
}
|
}
|
||||||
return v;
|
return v;
|
||||||
}
|
}
|
||||||
|
|
||||||
int32_t VP8GetSignedValue(VP8BitReader* const br, int bits) {
|
int32_t VP8GetSignedValue(VP8BitReader* const br, int bits,
|
||||||
const int value = VP8GetValue(br, bits);
|
const char label[]) {
|
||||||
return VP8Get(br) ? -value : value;
|
const int value = VP8GetValue(br, bits, label);
|
||||||
|
return VP8Get(br, label) ? -value : value;
|
||||||
}
|
}
|
||||||
|
|
||||||
//------------------------------------------------------------------------------
|
//------------------------------------------------------------------------------
|
||||||
|
@ -227,3 +228,78 @@ uint32_t VP8LReadBits(VP8LBitReader* const br, int n_bits) {
|
||||||
}
|
}
|
||||||
|
|
||||||
//------------------------------------------------------------------------------
|
//------------------------------------------------------------------------------
|
||||||
|
// Bit-tracing tool
|
||||||
|
|
||||||
|
#if (BITTRACE > 0)
|
||||||
|
|
||||||
|
#include <stdlib.h> // for atexit()
|
||||||
|
#include <stdio.h>
|
||||||
|
#include <string.h>
|
||||||
|
|
||||||
|
#define MAX_NUM_LABELS 32
|
||||||
|
static struct {
|
||||||
|
const char* label;
|
||||||
|
int size;
|
||||||
|
int count;
|
||||||
|
} kLabels[MAX_NUM_LABELS];
|
||||||
|
|
||||||
|
static int last_label = 0;
|
||||||
|
static int last_pos = 0;
|
||||||
|
static const uint8_t* buf_start = NULL;
|
||||||
|
static int init_done = 0;
|
||||||
|
|
||||||
|
static void PrintBitTraces(void) {
|
||||||
|
int i;
|
||||||
|
int scale = 1;
|
||||||
|
int total = 0;
|
||||||
|
const char* units = "bits";
|
||||||
|
#if (BITTRACE == 2)
|
||||||
|
scale = 8;
|
||||||
|
units = "bytes";
|
||||||
|
#endif
|
||||||
|
for (i = 0; i < last_label; ++i) total += kLabels[i].size;
|
||||||
|
if (total < 1) total = 1; // avoid rounding errors
|
||||||
|
printf("=== Bit traces ===\n");
|
||||||
|
for (i = 0; i < last_label; ++i) {
|
||||||
|
const int skip = 16 - (int)strlen(kLabels[i].label);
|
||||||
|
const int value = (kLabels[i].size + scale - 1) / scale;
|
||||||
|
assert(skip > 0);
|
||||||
|
printf("%s \%*s: %6d %s \t[%5.2f%%] [count: %7d]\n",
|
||||||
|
kLabels[i].label, skip, "", value, units,
|
||||||
|
100.f * kLabels[i].size / total,
|
||||||
|
kLabels[i].count);
|
||||||
|
}
|
||||||
|
total = (total + scale - 1) / scale;
|
||||||
|
printf("Total: %d %s\n", total, units);
|
||||||
|
}
|
||||||
|
|
||||||
|
void BitTrace(const struct VP8BitReader* const br, const char label[]) {
|
||||||
|
int i, pos;
|
||||||
|
if (!init_done) {
|
||||||
|
memset(kLabels, 0, sizeof(kLabels));
|
||||||
|
atexit(PrintBitTraces);
|
||||||
|
buf_start = br->buf_;
|
||||||
|
init_done = 1;
|
||||||
|
}
|
||||||
|
pos = (int)(br->buf_ - buf_start) * 8 - br->bits_;
|
||||||
|
// if there's a too large jump, we've changed partition -> reset counter
|
||||||
|
if (abs(pos - last_pos) > 32) {
|
||||||
|
buf_start = br->buf_;
|
||||||
|
pos = 0;
|
||||||
|
last_pos = 0;
|
||||||
|
}
|
||||||
|
if (br->range_ >= 0x7f) pos += kVP8Log2Range[br->range_ - 0x7f];
|
||||||
|
for (i = 0; i < last_label; ++i) {
|
||||||
|
if (!strcmp(label, kLabels[i].label)) break;
|
||||||
|
}
|
||||||
|
if (i == MAX_NUM_LABELS) abort(); // overflow!
|
||||||
|
kLabels[i].label = label;
|
||||||
|
kLabels[i].size += pos - last_pos;
|
||||||
|
kLabels[i].count += 1;
|
||||||
|
if (i == last_label) ++last_label;
|
||||||
|
last_pos = pos;
|
||||||
|
}
|
||||||
|
|
||||||
|
#endif // BITTRACE > 0
|
||||||
|
|
||||||
|
//------------------------------------------------------------------------------
|
||||||
|
|
|
@ -21,6 +21,27 @@
|
||||||
#endif
|
#endif
|
||||||
#include "src/webp/types.h"
|
#include "src/webp/types.h"
|
||||||
|
|
||||||
|
// Warning! This macro triggers quite some MACRO wizardry around func signature!
|
||||||
|
#if !defined(BITTRACE)
|
||||||
|
#define BITTRACE 0 // 0 = off, 1 = print bits, 2 = print bytes
|
||||||
|
#endif
|
||||||
|
|
||||||
|
#if (BITTRACE > 0)
|
||||||
|
struct VP8BitReader;
|
||||||
|
extern void BitTrace(const struct VP8BitReader* const br, const char label[]);
|
||||||
|
#define BT_TRACK(br) BitTrace(br, label)
|
||||||
|
#define VP8Get(BR, L) VP8GetValue(BR, 1, L)
|
||||||
|
#else
|
||||||
|
#define BT_TRACK(br)
|
||||||
|
// We'll REMOVE the 'const char label[]' from all signatures and calls (!!):
|
||||||
|
#define VP8GetValue(BR, N, L) VP8GetValue(BR, N)
|
||||||
|
#define VP8Get(BR, L) VP8GetValue(BR, 1, L)
|
||||||
|
#define VP8GetSignedValue(BR, N, L) VP8GetSignedValue(BR, N)
|
||||||
|
#define VP8GetBit(BR, P, L) VP8GetBit(BR, P)
|
||||||
|
#define VP8GetBitAlt(BR, P, L) VP8GetBitAlt(BR, P)
|
||||||
|
#define VP8GetSigned(BR, V, L) VP8GetSigned(BR, V)
|
||||||
|
#endif
|
||||||
|
|
||||||
#ifdef __cplusplus
|
#ifdef __cplusplus
|
||||||
extern "C" {
|
extern "C" {
|
||||||
#endif
|
#endif
|
||||||
|
@ -102,17 +123,15 @@ void VP8BitReaderSetBuffer(VP8BitReader* const br,
|
||||||
void VP8RemapBitReader(VP8BitReader* const br, ptrdiff_t offset);
|
void VP8RemapBitReader(VP8BitReader* const br, ptrdiff_t offset);
|
||||||
|
|
||||||
// return the next value made of 'num_bits' bits
|
// return the next value made of 'num_bits' bits
|
||||||
uint32_t VP8GetValue(VP8BitReader* const br, int num_bits);
|
uint32_t VP8GetValue(VP8BitReader* const br, int num_bits, const char label[]);
|
||||||
static WEBP_INLINE uint32_t VP8Get(VP8BitReader* const br) {
|
|
||||||
return VP8GetValue(br, 1);
|
|
||||||
}
|
|
||||||
|
|
||||||
// return the next value with sign-extension.
|
// return the next value with sign-extension.
|
||||||
int32_t VP8GetSignedValue(VP8BitReader* const br, int num_bits);
|
int32_t VP8GetSignedValue(VP8BitReader* const br, int num_bits,
|
||||||
|
const char label[]);
|
||||||
|
|
||||||
// bit_reader_inl.h will implement the following methods:
|
// bit_reader_inl.h will implement the following methods:
|
||||||
// static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob)
|
// static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob, ...)
|
||||||
// static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v)
|
// static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v, ...)
|
||||||
// and should be included by the .c files that actually need them.
|
// and should be included by the .c files that actually need them.
|
||||||
// This is to avoid recompiling the whole library whenever this file is touched,
|
// This is to avoid recompiling the whole library whenever this file is touched,
|
||||||
// and also allowing platform-specific ad-hoc hacks.
|
// and also allowing platform-specific ad-hoc hacks.
|
||||||
|
|
|
@ -70,7 +70,7 @@ static void Flush(VP8BitWriter* const bw) {
|
||||||
const int value = (bits & 0x100) ? 0x00 : 0xff;
|
const int value = (bits & 0x100) ? 0x00 : 0xff;
|
||||||
for (; bw->run_ > 0; --bw->run_) bw->buf_[pos++] = value;
|
for (; bw->run_ > 0; --bw->run_) bw->buf_[pos++] = value;
|
||||||
}
|
}
|
||||||
bw->buf_[pos++] = bits;
|
bw->buf_[pos++] = bits & 0xff;
|
||||||
bw->pos_ = pos;
|
bw->pos_ = pos;
|
||||||
} else {
|
} else {
|
||||||
bw->run_++; // delay writing of bytes 0xff, pending eventual carry.
|
bw->run_++; // delay writing of bytes 0xff, pending eventual carry.
|
||||||
|
|
|
@ -17,6 +17,7 @@
|
||||||
|
|
||||||
#include <assert.h>
|
#include <assert.h>
|
||||||
|
|
||||||
|
#include "src/dsp/dsp.h"
|
||||||
#include "src/webp/types.h"
|
#include "src/webp/types.h"
|
||||||
|
|
||||||
#ifdef __cplusplus
|
#ifdef __cplusplus
|
||||||
|
@ -30,10 +31,11 @@ typedef struct {
|
||||||
int hash_bits_;
|
int hash_bits_;
|
||||||
} VP8LColorCache;
|
} VP8LColorCache;
|
||||||
|
|
||||||
static const uint64_t kHashMul = 0x1e35a7bdull;
|
static const uint32_t kHashMul = 0x1e35a7bdu;
|
||||||
|
|
||||||
static WEBP_INLINE int VP8LHashPix(uint32_t argb, int shift) {
|
static WEBP_UBSAN_IGNORE_UNSIGNED_OVERFLOW WEBP_INLINE
|
||||||
return (int)(((argb * kHashMul) & 0xffffffffu) >> shift);
|
int VP8LHashPix(uint32_t argb, int shift) {
|
||||||
|
return (int)((argb * kHashMul) >> shift);
|
||||||
}
|
}
|
||||||
|
|
||||||
static WEBP_INLINE uint32_t VP8LColorCacheLookup(
|
static WEBP_INLINE uint32_t VP8LColorCacheLookup(
|
||||||
|
|
|
@ -91,7 +91,8 @@ static int BuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
|
||||||
|
|
||||||
assert(code_lengths_size != 0);
|
assert(code_lengths_size != 0);
|
||||||
assert(code_lengths != NULL);
|
assert(code_lengths != NULL);
|
||||||
assert(root_table != NULL);
|
assert((root_table != NULL && sorted != NULL) ||
|
||||||
|
(root_table == NULL && sorted == NULL));
|
||||||
assert(root_bits > 0);
|
assert(root_bits > 0);
|
||||||
|
|
||||||
// Build histogram of code lengths.
|
// Build histogram of code lengths.
|
||||||
|
@ -120,16 +121,22 @@ static int BuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
|
||||||
for (symbol = 0; symbol < code_lengths_size; ++symbol) {
|
for (symbol = 0; symbol < code_lengths_size; ++symbol) {
|
||||||
const int symbol_code_length = code_lengths[symbol];
|
const int symbol_code_length = code_lengths[symbol];
|
||||||
if (code_lengths[symbol] > 0) {
|
if (code_lengths[symbol] > 0) {
|
||||||
sorted[offset[symbol_code_length]++] = symbol;
|
if (sorted != NULL) {
|
||||||
|
sorted[offset[symbol_code_length]++] = symbol;
|
||||||
|
} else {
|
||||||
|
offset[symbol_code_length]++;
|
||||||
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
// Special case code with only one value.
|
// Special case code with only one value.
|
||||||
if (offset[MAX_ALLOWED_CODE_LENGTH] == 1) {
|
if (offset[MAX_ALLOWED_CODE_LENGTH] == 1) {
|
||||||
HuffmanCode code;
|
if (sorted != NULL) {
|
||||||
code.bits = 0;
|
HuffmanCode code;
|
||||||
code.value = (uint16_t)sorted[0];
|
code.bits = 0;
|
||||||
ReplicateValue(table, 1, total_size, code);
|
code.value = (uint16_t)sorted[0];
|
||||||
|
ReplicateValue(table, 1, total_size, code);
|
||||||
|
}
|
||||||
return total_size;
|
return total_size;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -151,6 +158,7 @@ static int BuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
|
||||||
if (num_open < 0) {
|
if (num_open < 0) {
|
||||||
return 0;
|
return 0;
|
||||||
}
|
}
|
||||||
|
if (root_table == NULL) continue;
|
||||||
for (; count[len] > 0; --count[len]) {
|
for (; count[len] > 0; --count[len]) {
|
||||||
HuffmanCode code;
|
HuffmanCode code;
|
||||||
code.bits = (uint8_t)len;
|
code.bits = (uint8_t)len;
|
||||||
|
@ -169,6 +177,7 @@ static int BuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
|
||||||
if (num_open < 0) {
|
if (num_open < 0) {
|
||||||
return 0;
|
return 0;
|
||||||
}
|
}
|
||||||
|
if (root_table == NULL) continue;
|
||||||
for (; count[len] > 0; --count[len]) {
|
for (; count[len] > 0; --count[len]) {
|
||||||
HuffmanCode code;
|
HuffmanCode code;
|
||||||
if ((key & mask) != low) {
|
if ((key & mask) != low) {
|
||||||
|
@ -206,7 +215,10 @@ int VP8LBuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
|
||||||
const int code_lengths[], int code_lengths_size) {
|
const int code_lengths[], int code_lengths_size) {
|
||||||
int total_size;
|
int total_size;
|
||||||
assert(code_lengths_size <= MAX_CODE_LENGTHS_SIZE);
|
assert(code_lengths_size <= MAX_CODE_LENGTHS_SIZE);
|
||||||
if (code_lengths_size <= SORTED_SIZE_CUTOFF) {
|
if (root_table == NULL) {
|
||||||
|
total_size = BuildHuffmanTable(NULL, root_bits,
|
||||||
|
code_lengths, code_lengths_size, NULL);
|
||||||
|
} else if (code_lengths_size <= SORTED_SIZE_CUTOFF) {
|
||||||
// use local stack-allocated array.
|
// use local stack-allocated array.
|
||||||
uint16_t sorted[SORTED_SIZE_CUTOFF];
|
uint16_t sorted[SORTED_SIZE_CUTOFF];
|
||||||
total_size = BuildHuffmanTable(root_table, root_bits,
|
total_size = BuildHuffmanTable(root_table, root_bits,
|
||||||
|
|
|
@ -78,6 +78,8 @@ void VP8LHtreeGroupsFree(HTreeGroup* const htree_groups);
|
||||||
// the huffman table.
|
// the huffman table.
|
||||||
// Returns built table size or 0 in case of error (invalid tree or
|
// Returns built table size or 0 in case of error (invalid tree or
|
||||||
// memory error).
|
// memory error).
|
||||||
|
// If root_table is NULL, it returns 0 if a lookup cannot be built, something
|
||||||
|
// > 0 otherwise (but not the table size).
|
||||||
int VP8LBuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
|
int VP8LBuildHuffmanTable(HuffmanCode* const root_table, int root_bits,
|
||||||
const int code_lengths[], int code_lengths_size);
|
const int code_lengths[], int code_lengths_size);
|
||||||
|
|
||||||
|
|
|
@ -84,14 +84,14 @@ int WebPRescalerGetScaledDimensions(int src_width, int src_height,
|
||||||
int height = *scaled_height;
|
int height = *scaled_height;
|
||||||
|
|
||||||
// if width is unspecified, scale original proportionally to height ratio.
|
// if width is unspecified, scale original proportionally to height ratio.
|
||||||
if (width == 0) {
|
if (width == 0 && src_height > 0) {
|
||||||
width =
|
width =
|
||||||
(int)(((uint64_t)src_width * height + src_height / 2) / src_height);
|
(int)(((uint64_t)src_width * height + src_height - 1) / src_height);
|
||||||
}
|
}
|
||||||
// if height is unspecified, scale original proportionally to width ratio.
|
// if height is unspecified, scale original proportionally to width ratio.
|
||||||
if (height == 0) {
|
if (height == 0 && src_width > 0) {
|
||||||
height =
|
height =
|
||||||
(int)(((uint64_t)src_height * width + src_width / 2) / src_width);
|
(int)(((uint64_t)src_height * width + src_width - 1) / src_width);
|
||||||
}
|
}
|
||||||
// Check if the overall dimensions still make sense.
|
// Check if the overall dimensions still make sense.
|
||||||
if (width <= 0 || height <= 0) {
|
if (width <= 0 || height <= 0) {
|
||||||
|
|
|
@ -217,8 +217,12 @@ static THREADFN ThreadLoop(void* ptr) {
|
||||||
done = 1;
|
done = 1;
|
||||||
}
|
}
|
||||||
// signal to the main thread that we're done (for Sync())
|
// signal to the main thread that we're done (for Sync())
|
||||||
pthread_cond_signal(&impl->condition_);
|
// Note the associated mutex does not need to be held when signaling the
|
||||||
|
// condition. Unlocking the mutex first may improve performance in some
|
||||||
|
// implementations, avoiding the case where the waiting thread can't
|
||||||
|
// reacquire the mutex when woken.
|
||||||
pthread_mutex_unlock(&impl->mutex_);
|
pthread_mutex_unlock(&impl->mutex_);
|
||||||
|
pthread_cond_signal(&impl->condition_);
|
||||||
}
|
}
|
||||||
return THREAD_RETURN(NULL); // Thread is finished
|
return THREAD_RETURN(NULL); // Thread is finished
|
||||||
}
|
}
|
||||||
|
@ -240,7 +244,13 @@ static void ChangeState(WebPWorker* const worker, WebPWorkerStatus new_status) {
|
||||||
// assign new status and release the working thread if needed
|
// assign new status and release the working thread if needed
|
||||||
if (new_status != OK) {
|
if (new_status != OK) {
|
||||||
worker->status_ = new_status;
|
worker->status_ = new_status;
|
||||||
|
// Note the associated mutex does not need to be held when signaling the
|
||||||
|
// condition. Unlocking the mutex first may improve performance in some
|
||||||
|
// implementations, avoiding the case where the waiting thread can't
|
||||||
|
// reacquire the mutex when woken.
|
||||||
|
pthread_mutex_unlock(&impl->mutex_);
|
||||||
pthread_cond_signal(&impl->condition_);
|
pthread_cond_signal(&impl->condition_);
|
||||||
|
return;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
pthread_mutex_unlock(&impl->mutex_);
|
pthread_mutex_unlock(&impl->mutex_);
|
||||||
|
|
|
@ -92,14 +92,14 @@ static WEBP_INLINE uint32_t GetLE32(const uint8_t* const data) {
|
||||||
// Store 16, 24 or 32 bits in little-endian order.
|
// Store 16, 24 or 32 bits in little-endian order.
|
||||||
static WEBP_INLINE void PutLE16(uint8_t* const data, int val) {
|
static WEBP_INLINE void PutLE16(uint8_t* const data, int val) {
|
||||||
assert(val < (1 << 16));
|
assert(val < (1 << 16));
|
||||||
data[0] = (val >> 0);
|
data[0] = (val >> 0) & 0xff;
|
||||||
data[1] = (val >> 8);
|
data[1] = (val >> 8) & 0xff;
|
||||||
}
|
}
|
||||||
|
|
||||||
static WEBP_INLINE void PutLE24(uint8_t* const data, int val) {
|
static WEBP_INLINE void PutLE24(uint8_t* const data, int val) {
|
||||||
assert(val < (1 << 24));
|
assert(val < (1 << 24));
|
||||||
PutLE16(data, val & 0xffff);
|
PutLE16(data, val & 0xffff);
|
||||||
data[2] = (val >> 16);
|
data[2] = (val >> 16) & 0xff;
|
||||||
}
|
}
|
||||||
|
|
||||||
static WEBP_INLINE void PutLE32(uint8_t* const data, uint32_t val) {
|
static WEBP_INLINE void PutLE32(uint8_t* const data, uint32_t val) {
|
||||||
|
|
|
@ -62,6 +62,10 @@ WEBP_EXTERN size_t WebPEncodeBGRA(const uint8_t* bgra,
|
||||||
// These functions are the equivalent of the above, but compressing in a
|
// These functions are the equivalent of the above, but compressing in a
|
||||||
// lossless manner. Files are usually larger than lossy format, but will
|
// lossless manner. Files are usually larger than lossy format, but will
|
||||||
// not suffer any compression loss.
|
// not suffer any compression loss.
|
||||||
|
// Note these functions, like the lossy versions, use the library's default
|
||||||
|
// settings. For lossless this means 'exact' is disabled. RGB values in
|
||||||
|
// transparent areas will be modified to improve compression. To avoid this,
|
||||||
|
// use WebPEncode() and set WebPConfig::exact to 1.
|
||||||
WEBP_EXTERN size_t WebPEncodeLosslessRGB(const uint8_t* rgb,
|
WEBP_EXTERN size_t WebPEncodeLosslessRGB(const uint8_t* rgb,
|
||||||
int width, int height, int stride,
|
int width, int height, int stride,
|
||||||
uint8_t** output);
|
uint8_t** output);
|
||||||
|
|
Loading…
Reference in New Issue