summaryrefslogtreecommitdiff
path: root/thirdparty/libwebp/src/dsp
diff options
context:
space:
mode:
Diffstat (limited to 'thirdparty/libwebp/src/dsp')
-rw-r--r--thirdparty/libwebp/src/dsp/dsp.h6
-rw-r--r--thirdparty/libwebp/src/dsp/enc.c6
-rw-r--r--thirdparty/libwebp/src/dsp/enc_avx2.c21
-rw-r--r--thirdparty/libwebp/src/dsp/lossless.c2
-rw-r--r--thirdparty/libwebp/src/dsp/lossless.h14
-rw-r--r--thirdparty/libwebp/src/dsp/lossless_enc.c86
-rw-r--r--thirdparty/libwebp/src/dsp/lossless_enc_mips32.c79
-rw-r--r--thirdparty/libwebp/src/dsp/lossless_enc_sse2.c44
-rw-r--r--thirdparty/libwebp/src/dsp/msa_macro.h2
-rw-r--r--thirdparty/libwebp/src/dsp/rescaler.c4
-rw-r--r--thirdparty/libwebp/src/dsp/rescaler_mips32.c4
-rw-r--r--thirdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c10
-rw-r--r--thirdparty/libwebp/src/dsp/rescaler_msa.c7
-rw-r--r--thirdparty/libwebp/src/dsp/rescaler_neon.c18
-rw-r--r--thirdparty/libwebp/src/dsp/rescaler_sse2.c35
-rw-r--r--thirdparty/libwebp/src/dsp/yuv.h2
16 files changed, 170 insertions, 170 deletions
diff --git a/thirdparty/libwebp/src/dsp/dsp.h b/thirdparty/libwebp/src/dsp/dsp.h
index 4ab77a5130..fafc2d05d3 100644
--- a/thirdparty/libwebp/src/dsp/dsp.h
+++ b/thirdparty/libwebp/src/dsp/dsp.h
@@ -76,10 +76,6 @@ extern "C" {
#define WEBP_USE_SSE41
#endif
-#if defined(__AVX2__) || defined(WEBP_HAVE_AVX2)
-#define WEBP_USE_AVX2
-#endif
-
// The intrinsics currently cause compiler errors with arm-nacl-gcc and the
// inline assembly would need to be modified for use with Native Client.
#if (defined(__ARM_NEON__) || \
@@ -679,4 +675,4 @@ void VP8FiltersInit(void);
} // extern "C"
#endif
-#endif /* WEBP_DSP_DSP_H_ */
+#endif // WEBP_DSP_DSP_H_
diff --git a/thirdparty/libwebp/src/dsp/enc.c b/thirdparty/libwebp/src/dsp/enc.c
index fa23b40a30..2fddbc4c52 100644
--- a/thirdparty/libwebp/src/dsp/enc.c
+++ b/thirdparty/libwebp/src/dsp/enc.c
@@ -734,7 +734,6 @@ VP8BlockCopy VP8Copy16x8;
extern void VP8EncDspInitSSE2(void);
extern void VP8EncDspInitSSE41(void);
-extern void VP8EncDspInitAVX2(void);
extern void VP8EncDspInitNEON(void);
extern void VP8EncDspInitMIPS32(void);
extern void VP8EncDspInitMIPSdspR2(void);
@@ -784,11 +783,6 @@ WEBP_DSP_INIT_FUNC(VP8EncDspInit) {
#endif
}
#endif
-#if defined(WEBP_USE_AVX2)
- if (VP8GetCPUInfo(kAVX2)) {
- VP8EncDspInitAVX2();
- }
-#endif
#if defined(WEBP_USE_MIPS32)
if (VP8GetCPUInfo(kMIPS32)) {
VP8EncDspInitMIPS32();
diff --git a/thirdparty/libwebp/src/dsp/enc_avx2.c b/thirdparty/libwebp/src/dsp/enc_avx2.c
deleted file mode 100644
index 8bc5798fee..0000000000
--- a/thirdparty/libwebp/src/dsp/enc_avx2.c
+++ /dev/null
@@ -1,21 +0,0 @@
-// Copyright 2014 Google Inc. All Rights Reserved.
-//
-// Use of this source code is governed by a BSD-style license
-// that can be found in the COPYING file in the root of the source
-// tree. An additional intellectual property rights grant can be found
-// in the file PATENTS. All contributing project authors may
-// be found in the AUTHORS file in the root of the source tree.
-// -----------------------------------------------------------------------------
-//
-// AVX2 version of speed-critical encoding functions.
-
-#include "src/dsp/dsp.h"
-
-#if defined(WEBP_USE_AVX2)
-
-#endif // WEBP_USE_AVX2
-
-//------------------------------------------------------------------------------
-// Entry point
-
-WEBP_DSP_INIT_STUB(VP8EncDspInitAVX2)
diff --git a/thirdparty/libwebp/src/dsp/lossless.c b/thirdparty/libwebp/src/dsp/lossless.c
index f9b3c182d3..d21aa6a0a0 100644
--- a/thirdparty/libwebp/src/dsp/lossless.c
+++ b/thirdparty/libwebp/src/dsp/lossless.c
@@ -23,8 +23,6 @@
#include "src/dsp/lossless.h"
#include "src/dsp/lossless_common.h"
-#define MAX_DIFF_COST (1e30f)
-
//------------------------------------------------------------------------------
// Image transforms.
diff --git a/thirdparty/libwebp/src/dsp/lossless.h b/thirdparty/libwebp/src/dsp/lossless.h
index b2bbdfc93c..f709cc86b2 100644
--- a/thirdparty/libwebp/src/dsp/lossless.h
+++ b/thirdparty/libwebp/src/dsp/lossless.h
@@ -163,7 +163,7 @@ extern VP8LCostCombinedFunc VP8LExtraCostCombined;
extern VP8LCombinedShannonEntropyFunc VP8LCombinedShannonEntropy;
typedef struct { // small struct to hold counters
- int counts[2]; // index: 0=zero steak, 1=non-zero streak
+ int counts[2]; // index: 0=zero streak, 1=non-zero streak
int streaks[2][2]; // [zero/non-zero][streak<3 / streak>=3]
} VP8LStreaks;
@@ -194,10 +194,14 @@ extern VP8LGetEntropyUnrefinedFunc VP8LGetEntropyUnrefined;
void VP8LBitsEntropyUnrefined(const uint32_t* const array, int n,
VP8LBitEntropy* const entropy);
-typedef void (*VP8LHistogramAddFunc)(const VP8LHistogram* const a,
- const VP8LHistogram* const b,
- VP8LHistogram* const out);
-extern VP8LHistogramAddFunc VP8LHistogramAdd;
+typedef void (*VP8LAddVectorFunc)(const uint32_t* a, const uint32_t* b,
+ uint32_t* out, int size);
+extern VP8LAddVectorFunc VP8LAddVector;
+typedef void (*VP8LAddVectorEqFunc)(const uint32_t* a, uint32_t* out, int size);
+extern VP8LAddVectorEqFunc VP8LAddVectorEq;
+void VP8LHistogramAdd(const VP8LHistogram* const a,
+ const VP8LHistogram* const b,
+ VP8LHistogram* const out);
// -----------------------------------------------------------------------------
// PrefixEncode()
diff --git a/thirdparty/libwebp/src/dsp/lossless_enc.c b/thirdparty/libwebp/src/dsp/lossless_enc.c
index d608326fef..1408fbf580 100644
--- a/thirdparty/libwebp/src/dsp/lossless_enc.c
+++ b/thirdparty/libwebp/src/dsp/lossless_enc.c
@@ -632,38 +632,67 @@ static double ExtraCostCombined_C(const uint32_t* X, const uint32_t* Y,
//------------------------------------------------------------------------------
-static void HistogramAdd_C(const VP8LHistogram* const a,
- const VP8LHistogram* const b,
- VP8LHistogram* const out) {
+static void AddVector_C(const uint32_t* a, const uint32_t* b, uint32_t* out,
+ int size) {
+ int i;
+ for (i = 0; i < size; ++i) out[i] = a[i] + b[i];
+}
+
+static void AddVectorEq_C(const uint32_t* a, uint32_t* out, int size) {
+ int i;
+ for (i = 0; i < size; ++i) out[i] += a[i];
+}
+
+#define ADD(X, ARG, LEN) do { \
+ if (a->is_used_[X]) { \
+ if (b->is_used_[X]) { \
+ VP8LAddVector(a->ARG, b->ARG, out->ARG, (LEN)); \
+ } else { \
+ memcpy(&out->ARG[0], &a->ARG[0], (LEN) * sizeof(out->ARG[0])); \
+ } \
+ } else if (b->is_used_[X]) { \
+ memcpy(&out->ARG[0], &b->ARG[0], (LEN) * sizeof(out->ARG[0])); \
+ } else { \
+ memset(&out->ARG[0], 0, (LEN) * sizeof(out->ARG[0])); \
+ } \
+} while (0)
+
+#define ADD_EQ(X, ARG, LEN) do { \
+ if (a->is_used_[X]) { \
+ if (out->is_used_[X]) { \
+ VP8LAddVectorEq(a->ARG, out->ARG, (LEN)); \
+ } else { \
+ memcpy(&out->ARG[0], &a->ARG[0], (LEN) * sizeof(out->ARG[0])); \
+ } \
+ } \
+} while (0)
+
+void VP8LHistogramAdd(const VP8LHistogram* const a,
+ const VP8LHistogram* const b, VP8LHistogram* const out) {
int i;
const int literal_size = VP8LHistogramNumCodes(a->palette_code_bits_);
assert(a->palette_code_bits_ == b->palette_code_bits_);
+
if (b != out) {
- for (i = 0; i < literal_size; ++i) {
- out->literal_[i] = a->literal_[i] + b->literal_[i];
- }
- for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
- out->distance_[i] = a->distance_[i] + b->distance_[i];
- }
- for (i = 0; i < NUM_LITERAL_CODES; ++i) {
- out->red_[i] = a->red_[i] + b->red_[i];
- out->blue_[i] = a->blue_[i] + b->blue_[i];
- out->alpha_[i] = a->alpha_[i] + b->alpha_[i];
+ ADD(0, literal_, literal_size);
+ ADD(1, red_, NUM_LITERAL_CODES);
+ ADD(2, blue_, NUM_LITERAL_CODES);
+ ADD(3, alpha_, NUM_LITERAL_CODES);
+ ADD(4, distance_, NUM_DISTANCE_CODES);
+ for (i = 0; i < 5; ++i) {
+ out->is_used_[i] = (a->is_used_[i] | b->is_used_[i]);
}
} else {
- for (i = 0; i < literal_size; ++i) {
- out->literal_[i] += a->literal_[i];
- }
- for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
- out->distance_[i] += a->distance_[i];
- }
- for (i = 0; i < NUM_LITERAL_CODES; ++i) {
- out->red_[i] += a->red_[i];
- out->blue_[i] += a->blue_[i];
- out->alpha_[i] += a->alpha_[i];
- }
+ ADD_EQ(0, literal_, literal_size);
+ ADD_EQ(1, red_, NUM_LITERAL_CODES);
+ ADD_EQ(2, blue_, NUM_LITERAL_CODES);
+ ADD_EQ(3, alpha_, NUM_LITERAL_CODES);
+ ADD_EQ(4, distance_, NUM_DISTANCE_CODES);
+ for (i = 0; i < 5; ++i) out->is_used_[i] |= a->is_used_[i];
}
}
+#undef ADD
+#undef ADD_EQ
//------------------------------------------------------------------------------
// Image transforms.
@@ -848,7 +877,8 @@ VP8LCombinedShannonEntropyFunc VP8LCombinedShannonEntropy;
VP8LGetEntropyUnrefinedFunc VP8LGetEntropyUnrefined;
VP8LGetCombinedEntropyUnrefinedFunc VP8LGetCombinedEntropyUnrefined;
-VP8LHistogramAddFunc VP8LHistogramAdd;
+VP8LAddVectorFunc VP8LAddVector;
+VP8LAddVectorEqFunc VP8LAddVectorEq;
VP8LVectorMismatchFunc VP8LVectorMismatch;
VP8LBundleColorMapFunc VP8LBundleColorMap;
@@ -885,7 +915,8 @@ WEBP_DSP_INIT_FUNC(VP8LEncDspInit) {
VP8LGetEntropyUnrefined = GetEntropyUnrefined_C;
VP8LGetCombinedEntropyUnrefined = GetCombinedEntropyUnrefined_C;
- VP8LHistogramAdd = HistogramAdd_C;
+ VP8LAddVector = AddVector_C;
+ VP8LAddVectorEq = AddVectorEq_C;
VP8LVectorMismatch = VectorMismatch_C;
VP8LBundleColorMap = VP8LBundleColorMap_C;
@@ -971,7 +1002,8 @@ WEBP_DSP_INIT_FUNC(VP8LEncDspInit) {
assert(VP8LCombinedShannonEntropy != NULL);
assert(VP8LGetEntropyUnrefined != NULL);
assert(VP8LGetCombinedEntropyUnrefined != NULL);
- assert(VP8LHistogramAdd != NULL);
+ assert(VP8LAddVector != NULL);
+ assert(VP8LAddVectorEq != NULL);
assert(VP8LVectorMismatch != NULL);
assert(VP8LBundleColorMap != NULL);
assert(VP8LPredictorsSub[0] != NULL);
diff --git a/thirdparty/libwebp/src/dsp/lossless_enc_mips32.c b/thirdparty/libwebp/src/dsp/lossless_enc_mips32.c
index e7b58f4e8c..0412a093cf 100644
--- a/thirdparty/libwebp/src/dsp/lossless_enc_mips32.c
+++ b/thirdparty/libwebp/src/dsp/lossless_enc_mips32.c
@@ -344,65 +344,29 @@ static void GetCombinedEntropyUnrefined_MIPS32(const uint32_t X[],
ASM_END_COMMON_0 \
ASM_END_COMMON_1
-#define ADD_VECTOR(A, B, OUT, SIZE, EXTRA_SIZE) do { \
- const uint32_t* pa = (const uint32_t*)(A); \
- const uint32_t* pb = (const uint32_t*)(B); \
- uint32_t* pout = (uint32_t*)(OUT); \
- const uint32_t* const LoopEnd = pa + (SIZE); \
- assert((SIZE) % 4 == 0); \
- ASM_START \
- ADD_TO_OUT(0, 4, 8, 12, 1, pa, pb, pout) \
- ASM_END_0 \
- if ((EXTRA_SIZE) > 0) { \
- const int last = (EXTRA_SIZE); \
- int i; \
- for (i = 0; i < last; ++i) pout[i] = pa[i] + pb[i]; \
- } \
-} while (0)
-
-#define ADD_VECTOR_EQ(A, OUT, SIZE, EXTRA_SIZE) do { \
- const uint32_t* pa = (const uint32_t*)(A); \
- uint32_t* pout = (uint32_t*)(OUT); \
- const uint32_t* const LoopEnd = pa + (SIZE); \
- assert((SIZE) % 4 == 0); \
- ASM_START \
- ADD_TO_OUT(0, 4, 8, 12, 0, pa, pout, pout) \
- ASM_END_1 \
- if ((EXTRA_SIZE) > 0) { \
- const int last = (EXTRA_SIZE); \
- int i; \
- for (i = 0; i < last; ++i) pout[i] += pa[i]; \
- } \
-} while (0)
-
-static void HistogramAdd_MIPS32(const VP8LHistogram* const a,
- const VP8LHistogram* const b,
- VP8LHistogram* const out) {
+static void AddVector_MIPS32(const uint32_t* pa, const uint32_t* pb,
+ uint32_t* pout, int size) {
uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
- const int extra_cache_size = VP8LHistogramNumCodes(a->palette_code_bits_)
- - (NUM_LITERAL_CODES + NUM_LENGTH_CODES);
- assert(a->palette_code_bits_ == b->palette_code_bits_);
-
- if (b != out) {
- ADD_VECTOR(a->literal_, b->literal_, out->literal_,
- NUM_LITERAL_CODES + NUM_LENGTH_CODES, extra_cache_size);
- ADD_VECTOR(a->distance_, b->distance_, out->distance_,
- NUM_DISTANCE_CODES, 0);
- ADD_VECTOR(a->red_, b->red_, out->red_, NUM_LITERAL_CODES, 0);
- ADD_VECTOR(a->blue_, b->blue_, out->blue_, NUM_LITERAL_CODES, 0);
- ADD_VECTOR(a->alpha_, b->alpha_, out->alpha_, NUM_LITERAL_CODES, 0);
- } else {
- ADD_VECTOR_EQ(a->literal_, out->literal_,
- NUM_LITERAL_CODES + NUM_LENGTH_CODES, extra_cache_size);
- ADD_VECTOR_EQ(a->distance_, out->distance_, NUM_DISTANCE_CODES, 0);
- ADD_VECTOR_EQ(a->red_, out->red_, NUM_LITERAL_CODES, 0);
- ADD_VECTOR_EQ(a->blue_, out->blue_, NUM_LITERAL_CODES, 0);
- ADD_VECTOR_EQ(a->alpha_, out->alpha_, NUM_LITERAL_CODES, 0);
- }
+ const uint32_t end = ((size) / 4) * 4;
+ const uint32_t* const LoopEnd = pa + end;
+ int i;
+ ASM_START
+ ADD_TO_OUT(0, 4, 8, 12, 1, pa, pb, pout)
+ ASM_END_0
+ for (i = end; i < size; ++i) pout[i] = pa[i] + pb[i];
+}
+
+static void AddVectorEq_MIPS32(const uint32_t* pa, uint32_t* pout, int size) {
+ uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
+ const uint32_t end = ((size) / 4) * 4;
+ const uint32_t* const LoopEnd = pa + end;
+ int i;
+ ASM_START
+ ADD_TO_OUT(0, 4, 8, 12, 0, pa, pout, pout)
+ ASM_END_1
+ for (i = end; i < size; ++i) pout[i] += pa[i];
}
-#undef ADD_VECTOR_EQ
-#undef ADD_VECTOR
#undef ASM_END_1
#undef ASM_END_0
#undef ASM_END_COMMON_1
@@ -422,7 +386,8 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitMIPS32(void) {
VP8LExtraCostCombined = ExtraCostCombined_MIPS32;
VP8LGetEntropyUnrefined = GetEntropyUnrefined_MIPS32;
VP8LGetCombinedEntropyUnrefined = GetCombinedEntropyUnrefined_MIPS32;
- VP8LHistogramAdd = HistogramAdd_MIPS32;
+ VP8LAddVector = AddVector_MIPS32;
+ VP8LAddVectorEq = AddVectorEq_MIPS32;
}
#else // !WEBP_USE_MIPS32
diff --git a/thirdparty/libwebp/src/dsp/lossless_enc_sse2.c b/thirdparty/libwebp/src/dsp/lossless_enc_sse2.c
index f84a9909e1..36478c4912 100644
--- a/thirdparty/libwebp/src/dsp/lossless_enc_sse2.c
+++ b/thirdparty/libwebp/src/dsp/lossless_enc_sse2.c
@@ -170,12 +170,13 @@ static void CollectColorRedTransforms_SSE2(const uint32_t* argb, int stride,
//------------------------------------------------------------------------------
+// Note we are adding uint32_t's as *signed* int32's (using _mm_add_epi32). But
+// that's ok since the histogram values are less than 1<<28 (max picture size).
#define LINE_SIZE 16 // 8 or 16
static void AddVector_SSE2(const uint32_t* a, const uint32_t* b, uint32_t* out,
int size) {
int i;
- assert(size % LINE_SIZE == 0);
- for (i = 0; i < size; i += LINE_SIZE) {
+ for (i = 0; i + LINE_SIZE <= size; i += LINE_SIZE) {
const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i + 0]);
const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i + 4]);
#if (LINE_SIZE == 16)
@@ -195,12 +196,14 @@ static void AddVector_SSE2(const uint32_t* a, const uint32_t* b, uint32_t* out,
_mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
#endif
}
+ for (; i < size; ++i) {
+ out[i] = a[i] + b[i];
+ }
}
static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {
int i;
- assert(size % LINE_SIZE == 0);
- for (i = 0; i < size; i += LINE_SIZE) {
+ for (i = 0; i + LINE_SIZE <= size; i += LINE_SIZE) {
const __m128i a0 = _mm_loadu_si128((const __m128i*)&a[i + 0]);
const __m128i a1 = _mm_loadu_si128((const __m128i*)&a[i + 4]);
#if (LINE_SIZE == 16)
@@ -220,35 +223,11 @@ static void AddVectorEq_SSE2(const uint32_t* a, uint32_t* out, int size) {
_mm_storeu_si128((__m128i*)&out[i + 12], _mm_add_epi32(a3, b3));
#endif
}
-}
-#undef LINE_SIZE
-
-// Note we are adding uint32_t's as *signed* int32's (using _mm_add_epi32). But
-// that's ok since the histogram values are less than 1<<28 (max picture size).
-static void HistogramAdd_SSE2(const VP8LHistogram* const a,
- const VP8LHistogram* const b,
- VP8LHistogram* const out) {
- int i;
- const int literal_size = VP8LHistogramNumCodes(a->palette_code_bits_);
- assert(a->palette_code_bits_ == b->palette_code_bits_);
- if (b != out) {
- AddVector_SSE2(a->literal_, b->literal_, out->literal_, NUM_LITERAL_CODES);
- AddVector_SSE2(a->red_, b->red_, out->red_, NUM_LITERAL_CODES);
- AddVector_SSE2(a->blue_, b->blue_, out->blue_, NUM_LITERAL_CODES);
- AddVector_SSE2(a->alpha_, b->alpha_, out->alpha_, NUM_LITERAL_CODES);
- } else {
- AddVectorEq_SSE2(a->literal_, out->literal_, NUM_LITERAL_CODES);
- AddVectorEq_SSE2(a->red_, out->red_, NUM_LITERAL_CODES);
- AddVectorEq_SSE2(a->blue_, out->blue_, NUM_LITERAL_CODES);
- AddVectorEq_SSE2(a->alpha_, out->alpha_, NUM_LITERAL_CODES);
- }
- for (i = NUM_LITERAL_CODES; i < literal_size; ++i) {
- out->literal_[i] = a->literal_[i] + b->literal_[i];
- }
- for (i = 0; i < NUM_DISTANCE_CODES; ++i) {
- out->distance_[i] = a->distance_[i] + b->distance_[i];
+ for (; i < size; ++i) {
+ out[i] += a[i];
}
}
+#undef LINE_SIZE
//------------------------------------------------------------------------------
// Entropy
@@ -675,7 +654,8 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE2(void) {
VP8LTransformColor = TransformColor_SSE2;
VP8LCollectColorBlueTransforms = CollectColorBlueTransforms_SSE2;
VP8LCollectColorRedTransforms = CollectColorRedTransforms_SSE2;
- VP8LHistogramAdd = HistogramAdd_SSE2;
+ VP8LAddVector = AddVector_SSE2;
+ VP8LAddVectorEq = AddVectorEq_SSE2;
VP8LCombinedShannonEntropy = CombinedShannonEntropy_SSE2;
VP8LVectorMismatch = VectorMismatch_SSE2;
VP8LBundleColorMap = BundleColorMap_SSE2;
diff --git a/thirdparty/libwebp/src/dsp/msa_macro.h b/thirdparty/libwebp/src/dsp/msa_macro.h
index dfacda6ccd..de026a1d9e 100644
--- a/thirdparty/libwebp/src/dsp/msa_macro.h
+++ b/thirdparty/libwebp/src/dsp/msa_macro.h
@@ -1389,4 +1389,4 @@ static WEBP_INLINE uint32_t func_hadd_uh_u32(v8u16 in) {
} while (0)
#define AVER_UB2_UB(...) AVER_UB2(v16u8, __VA_ARGS__)
-#endif /* WEBP_DSP_MSA_MACRO_H_ */
+#endif // WEBP_DSP_MSA_MACRO_H_
diff --git a/thirdparty/libwebp/src/dsp/rescaler.c b/thirdparty/libwebp/src/dsp/rescaler.c
index f307d35056..753f84fcf4 100644
--- a/thirdparty/libwebp/src/dsp/rescaler.c
+++ b/thirdparty/libwebp/src/dsp/rescaler.c
@@ -21,6 +21,7 @@
#define ROUNDER (WEBP_RESCALER_ONE >> 1)
#define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
//------------------------------------------------------------------------------
// Row import
@@ -138,7 +139,7 @@ void WebPRescalerExportRowShrink_C(WebPRescaler* const wrk) {
if (yscale) {
for (x_out = 0; x_out < x_out_max; ++x_out) {
const uint32_t frac = (uint32_t)MULT_FIX(frow[x_out], yscale);
- const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
+ const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
assert(v >= 0 && v <= 255);
dst[x_out] = v;
irow[x_out] = frac; // new fractional start
@@ -153,6 +154,7 @@ void WebPRescalerExportRowShrink_C(WebPRescaler* const wrk) {
}
}
+#undef MULT_FIX_FLOOR
#undef MULT_FIX
#undef ROUNDER
diff --git a/thirdparty/libwebp/src/dsp/rescaler_mips32.c b/thirdparty/libwebp/src/dsp/rescaler_mips32.c
index 542f7e5970..61f63c616c 100644
--- a/thirdparty/libwebp/src/dsp/rescaler_mips32.c
+++ b/thirdparty/libwebp/src/dsp/rescaler_mips32.c
@@ -209,6 +209,7 @@ static void ExportRowExpand_MIPS32(WebPRescaler* const wrk) {
}
}
+#if 0 // disabled for now. TODO(skal): make match the C-code
static void ExportRowShrink_MIPS32(WebPRescaler* const wrk) {
const int x_out_max = wrk->dst_width * wrk->num_channels;
uint8_t* dst = wrk->dst;
@@ -273,6 +274,7 @@ static void ExportRowShrink_MIPS32(WebPRescaler* const wrk) {
);
}
}
+#endif // 0
//------------------------------------------------------------------------------
// Entry point
@@ -283,7 +285,7 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMIPS32(void) {
WebPRescalerImportRowExpand = ImportRowExpand_MIPS32;
WebPRescalerImportRowShrink = ImportRowShrink_MIPS32;
WebPRescalerExportRowExpand = ExportRowExpand_MIPS32;
- WebPRescalerExportRowShrink = ExportRowShrink_MIPS32;
+// WebPRescalerExportRowShrink = ExportRowShrink_MIPS32;
}
#else // !WEBP_USE_MIPS32
diff --git a/thirdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c b/thirdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c
index b78aac15e6..ce9e64862e 100644
--- a/thirdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c
+++ b/thirdparty/libwebp/src/dsp/rescaler_mips_dsp_r2.c
@@ -20,10 +20,12 @@
#define ROUNDER (WEBP_RESCALER_ONE >> 1)
#define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
//------------------------------------------------------------------------------
// Row export
+#if 0 // disabled for now. TODO(skal): make match the C-code
static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
int i;
const int x_out_max = wrk->dst_width * wrk->num_channels;
@@ -106,7 +108,7 @@ static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
}
for (i = 0; i < (x_out_max & 0x3); ++i) {
const uint32_t frac = (uint32_t)MULT_FIX(*frow++, yscale);
- const int v = (int)MULT_FIX(*irow - frac, wrk->fxy_scale);
+ const int v = (int)MULT_FIX_FLOOR(*irow - frac, wrk->fxy_scale);
assert(v >= 0 && v <= 255);
*dst++ = v;
*irow++ = frac; // new fractional start
@@ -154,13 +156,14 @@ static void ExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
);
}
for (i = 0; i < (x_out_max & 0x3); ++i) {
- const int v = (int)MULT_FIX(*irow, wrk->fxy_scale);
+ const int v = (int)MULT_FIX_FLOOR(*irow, wrk->fxy_scale);
assert(v >= 0 && v <= 255);
*dst++ = v;
*irow++ = 0;
}
}
}
+#endif // 0
static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
int i;
@@ -294,6 +297,7 @@ static void ExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
}
}
+#undef MULT_FIX_FLOOR
#undef MULT_FIX
#undef ROUNDER
@@ -304,7 +308,7 @@ extern void WebPRescalerDspInitMIPSdspR2(void);
WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMIPSdspR2(void) {
WebPRescalerExportRowExpand = ExportRowExpand_MIPSdspR2;
- WebPRescalerExportRowShrink = ExportRowShrink_MIPSdspR2;
+// WebPRescalerExportRowShrink = ExportRowShrink_MIPSdspR2;
}
#else // !WEBP_USE_MIPS_DSP_R2
diff --git a/thirdparty/libwebp/src/dsp/rescaler_msa.c b/thirdparty/libwebp/src/dsp/rescaler_msa.c
index f3bc99f1cd..c559254836 100644
--- a/thirdparty/libwebp/src/dsp/rescaler_msa.c
+++ b/thirdparty/libwebp/src/dsp/rescaler_msa.c
@@ -22,6 +22,7 @@
#define ROUNDER (WEBP_RESCALER_ONE >> 1)
#define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
#define CALC_MULT_FIX_16(in0, in1, in2, in3, scale, shift, dst) do { \
v4u32 tmp0, tmp1, tmp2, tmp3; \
@@ -262,6 +263,7 @@ static void RescalerExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
}
}
+#if 0 // disabled for now. TODO(skal): make match the C-code
static WEBP_INLINE void ExportRowShrink_0(const uint32_t* frow, uint32_t* irow,
uint8_t* dst, int length,
const uint32_t yscale,
@@ -341,7 +343,7 @@ static WEBP_INLINE void ExportRowShrink_0(const uint32_t* frow, uint32_t* irow,
}
for (x_out = 0; x_out < length; ++x_out) {
const uint32_t frac = (uint32_t)MULT_FIX(frow[x_out], yscale);
- const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
+ const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
assert(v >= 0 && v <= 255);
dst[x_out] = v;
irow[x_out] = frac;
@@ -426,6 +428,7 @@ static void RescalerExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
ExportRowShrink_1(irow, dst, x_out_max, wrk);
}
}
+#endif // 0
//------------------------------------------------------------------------------
// Entry point
@@ -434,7 +437,7 @@ extern void WebPRescalerDspInitMSA(void);
WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMSA(void) {
WebPRescalerExportRowExpand = RescalerExportRowExpand_MIPSdspR2;
- WebPRescalerExportRowShrink = RescalerExportRowShrink_MIPSdspR2;
+// WebPRescalerExportRowShrink = RescalerExportRowShrink_MIPSdspR2;
}
#else // !WEBP_USE_MSA
diff --git a/thirdparty/libwebp/src/dsp/rescaler_neon.c b/thirdparty/libwebp/src/dsp/rescaler_neon.c
index 3eff9fbaf4..a553f06f79 100644
--- a/thirdparty/libwebp/src/dsp/rescaler_neon.c
+++ b/thirdparty/libwebp/src/dsp/rescaler_neon.c
@@ -22,6 +22,7 @@
#define ROUNDER (WEBP_RESCALER_ONE >> 1)
#define MULT_FIX_C(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR_C(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
#define LOAD_32x4(SRC, DST) const uint32x4_t DST = vld1q_u32((SRC))
#define LOAD_32x8(SRC, DST0, DST1) \
@@ -35,8 +36,11 @@
#if (WEBP_RESCALER_RFIX == 32)
#define MAKE_HALF_CST(C) vdupq_n_s32((int32_t)((C) >> 1))
-#define MULT_FIX(A, B) /* note: B is actualy scale>>1. See MAKE_HALF_CST */ \
+// note: B is actualy scale>>1. See MAKE_HALF_CST
+#define MULT_FIX(A, B) \
vreinterpretq_u32_s32(vqrdmulhq_s32(vreinterpretq_s32_u32((A)), (B)))
+#define MULT_FIX_FLOOR(A, B) \
+ vreinterpretq_u32_s32(vqdmulhq_s32(vreinterpretq_s32_u32((A)), (B)))
#else
#error "MULT_FIX/WEBP_RESCALER_RFIX need some more work"
#endif
@@ -135,8 +139,8 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
const uint32x4_t A1 = MULT_FIX(in1, yscale_half);
const uint32x4_t B0 = vqsubq_u32(in2, A0);
const uint32x4_t B1 = vqsubq_u32(in3, A1);
- const uint32x4_t C0 = MULT_FIX(B0, fxy_scale_half);
- const uint32x4_t C1 = MULT_FIX(B1, fxy_scale_half);
+ const uint32x4_t C0 = MULT_FIX_FLOOR(B0, fxy_scale_half);
+ const uint32x4_t C1 = MULT_FIX_FLOOR(B1, fxy_scale_half);
const uint16x4_t D0 = vmovn_u32(C0);
const uint16x4_t D1 = vmovn_u32(C1);
const uint8x8_t E = vmovn_u16(vcombine_u16(D0, D1));
@@ -145,7 +149,7 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
}
for (; x_out < x_out_max; ++x_out) {
const uint32_t frac = (uint32_t)MULT_FIX_C(frow[x_out], yscale);
- const int v = (int)MULT_FIX_C(irow[x_out] - frac, wrk->fxy_scale);
+ const int v = (int)MULT_FIX_FLOOR_C(irow[x_out] - frac, fxy_scale);
assert(v >= 0 && v <= 255);
dst[x_out] = v;
irow[x_out] = frac; // new fractional start
@@ -170,6 +174,12 @@ static void RescalerExportRowShrink_NEON(WebPRescaler* const wrk) {
}
}
+#undef MULT_FIX_FLOOR_C
+#undef MULT_FIX_C
+#undef MULT_FIX_FLOOR
+#undef MULT_FIX
+#undef ROUNDER
+
//------------------------------------------------------------------------------
extern void WebPRescalerDspInitNEON(void);
diff --git a/thirdparty/libwebp/src/dsp/rescaler_sse2.c b/thirdparty/libwebp/src/dsp/rescaler_sse2.c
index 64c50deab5..f7461a452c 100644
--- a/thirdparty/libwebp/src/dsp/rescaler_sse2.c
+++ b/thirdparty/libwebp/src/dsp/rescaler_sse2.c
@@ -25,6 +25,7 @@
#define ROUNDER (WEBP_RESCALER_ONE >> 1)
#define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
+#define MULT_FIX_FLOOR(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
// input: 8 bytes ABCDEFGH -> output: A0E0B0F0C0G0D0H0
static void LoadTwoPixels_SSE2(const uint8_t* const src, __m128i* out) {
@@ -224,6 +225,35 @@ static WEBP_INLINE void ProcessRow_SSE2(const __m128i* const A0,
_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) {
int x_out;
uint8_t* const dst = wrk->dst;
@@ -322,12 +352,12 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
const __m128i G1 = _mm_or_si128(D1, F3);
_mm_storeu_si128((__m128i*)(irow + x_out + 0), G0);
_mm_storeu_si128((__m128i*)(irow + x_out + 4), G1);
- ProcessRow_SSE2(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
+ ProcessRow_Floor_SSE2(&E0, &E1, &E2, &E3, &mult_xy, dst + x_out);
}
}
for (; x_out < x_out_max; ++x_out) {
const uint32_t frac = (int)MULT_FIX(frow[x_out], yscale);
- const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
+ const int v = (int)MULT_FIX_FLOOR(irow[x_out] - frac, wrk->fxy_scale);
assert(v >= 0 && v <= 255);
dst[x_out] = v;
irow[x_out] = frac; // new fractional start
@@ -352,6 +382,7 @@ static void RescalerExportRowShrink_SSE2(WebPRescaler* const wrk) {
}
}
+#undef MULT_FIX_FLOOR
#undef MULT_FIX
#undef ROUNDER
diff --git a/thirdparty/libwebp/src/dsp/yuv.h b/thirdparty/libwebp/src/dsp/yuv.h
index eb787270d2..c12be1d094 100644
--- a/thirdparty/libwebp/src/dsp/yuv.h
+++ b/thirdparty/libwebp/src/dsp/yuv.h
@@ -207,4 +207,4 @@ static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
} // extern "C"
#endif
-#endif /* WEBP_DSP_YUV_H_ */
+#endif // WEBP_DSP_YUV_H_
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-05 06:44:16 +0000