godotengine/godot
1
0
Fork 0
mirror of https://github.com/godotengine/godot.git synced 2025-11-24 15:26:15 +00:00
Code Releases Wiki Activity

libwebp: Sync with upstream 1.3.0

Browse Source
This commit is contained in:
DeeJayLSP
2023-01-25 10:24:01 -03:00
parent 6369196b96
commit d8e8517d11
47 changed files with 377 additions and 268 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
thirdparty/libwebp/src/dsp/alpha_processing_sse2.c vendored
View File

@@ -26,8 +26,8 @@ static int DispatchAlpha_SSE2(const uint8_t* WEBP_RESTRICT alpha,
uint32_t alpha_and = 0xff;
int i, j;
const __m128i zero = _mm_setzero_si128();
const __m128i rgb_mask = _mm_set1_epi32(0xffffff00u); // to preserve RGB
const __m128i all_0xff = _mm_set_epi32(0, 0, ~0u, ~0u);
const __m128i rgb_mask = _mm_set1_epi32((int)0xffffff00); // to preserve RGB
const __m128i all_0xff = _mm_set_epi32(0, 0, ~0, ~0);
__m128i all_alphas = all_0xff;
// We must be able to access 3 extra bytes after the last written byte
@@ -106,8 +106,8 @@ static int ExtractAlpha_SSE2(const uint8_t* WEBP_RESTRICT argb, int argb_stride,
// value is not 0xff if any of the alpha[] is not equal to 0xff.
uint32_t alpha_and = 0xff;
int i, j;
const __m128i a_mask = _mm_set1_epi32(0xffu); // to preserve alpha
const __m128i all_0xff = _mm_set_epi32(0, 0, ~0u, ~0u);
const __m128i a_mask = _mm_set1_epi32(0xff); // to preserve alpha
const __m128i all_0xff = _mm_set_epi32(0, 0, ~0, ~0);
__m128i all_alphas = all_0xff;
// We must be able to access 3 extra bytes after the last written byte
@@ -178,7 +178,7 @@ static int ExtractAlpha_SSE2(const uint8_t* WEBP_RESTRICT argb, int argb_stride,
static void ApplyAlphaMultiply_SSE2(uint8_t* rgba, int alpha_first,
int w, int h, int stride) {
const __m128i zero = _mm_setzero_si128();
const __m128i kMult = _mm_set1_epi16(0x8081u);
const __m128i kMult = _mm_set1_epi16((short)0x8081);
const __m128i kMask = _mm_set_epi16(0, 0xff, 0xff, 0, 0, 0xff, 0xff, 0);
const int kSpan = 4;
while (h-- > 0) {
@@ -267,7 +267,7 @@ static int HasAlpha32b_SSE2(const uint8_t* src, int length) {
}
static void AlphaReplace_SSE2(uint32_t* src, int length, uint32_t color) {
const __m128i m_color = _mm_set1_epi32(color);
const __m128i m_color = _mm_set1_epi32((int)color);
const __m128i zero = _mm_setzero_si128();
int i = 0;
for (; i + 8 <= length; i += 8) {

2
thirdparty/libwebp/src/dsp/alpha_processing_sse41.c vendored
View File

@@ -26,7 +26,7 @@ static int ExtractAlpha_SSE41(const uint8_t* WEBP_RESTRICT argb,
// value is not 0xff if any of the alpha[] is not equal to 0xff.
uint32_t alpha_and = 0xff;
int i, j;
const __m128i all_0xff = _mm_set1_epi32(~0u);
const __m128i all_0xff = _mm_set1_epi32(~0);
__m128i all_alphas = all_0xff;
// We must be able to access 3 extra bytes after the last written byte

2
thirdparty/libwebp/src/dsp/cpu.c vendored
View File

@@ -212,7 +212,7 @@ VP8CPUInfo VP8GetCPUInfo = wasmCPUInfo;
#elif defined(WEBP_HAVE_NEON)
// In most cases this function doesn't check for NEON support (it's assumed by
// the configuration), but enables turning off NEON at runtime, for testing
// purposes, by setting VP8DecGetCPUInfo = NULL.
// purposes, by setting VP8GetCPUInfo = NULL.
static int armCPUInfo(CPUFeature feature) {
if (feature != kNEON) return 0;
#if defined(__linux__) && defined(WEBP_HAVE_NEON_RTCD)

2
thirdparty/libwebp/src/dsp/cpu.h vendored
View File

@@ -14,6 +14,8 @@
#ifndef WEBP_DSP_CPU_H_
#define WEBP_DSP_CPU_H_
#include <stddef.h>
#ifdef HAVE_CONFIG_H
#include "src/webp/config.h"
#endif

93
thirdparty/libwebp/src/dsp/dec_sse2.c vendored
View File

@@ -158,10 +158,10 @@ static void Transform_SSE2(const int16_t* in, uint8_t* dst, int do_two) {
dst3 = _mm_loadl_epi64((__m128i*)(dst + 3 * BPS));
} else {
// Load four bytes/pixels per line.
dst0 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 0 * BPS));
dst1 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 1 * BPS));
dst2 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 2 * BPS));
dst3 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 3 * BPS));
dst0 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 0 * BPS));
dst1 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 1 * BPS));
dst2 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 2 * BPS));
dst3 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 3 * BPS));
}
// Convert to 16b.
dst0 = _mm_unpacklo_epi8(dst0, zero);
@@ -187,10 +187,10 @@ static void Transform_SSE2(const int16_t* in, uint8_t* dst, int do_two) {
_mm_storel_epi64((__m128i*)(dst + 3 * BPS), dst3);
} else {
// Store four bytes/pixels per line.
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(dst0));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(dst1));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(dst2));
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(dst3));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(dst0));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(dst1));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(dst2));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(dst3));
}
}
}
@@ -213,10 +213,10 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
const __m128i m3 = _mm_subs_epi16(B, d4);
const __m128i zero = _mm_setzero_si128();
// Load the source pixels.
__m128i dst0 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 0 * BPS));
__m128i dst1 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 1 * BPS));
__m128i dst2 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 2 * BPS));
__m128i dst3 = _mm_cvtsi32_si128(WebPMemToUint32(dst + 3 * BPS));
__m128i dst0 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 0 * BPS));
__m128i dst1 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 1 * BPS));
__m128i dst2 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 2 * BPS));
__m128i dst3 = _mm_cvtsi32_si128(WebPMemToInt32(dst + 3 * BPS));
// Convert to 16b.
dst0 = _mm_unpacklo_epi8(dst0, zero);
dst1 = _mm_unpacklo_epi8(dst1, zero);
@@ -233,10 +233,10 @@ static void TransformAC3(const int16_t* in, uint8_t* dst) {
dst2 = _mm_packus_epi16(dst2, dst2);
dst3 = _mm_packus_epi16(dst3, dst3);
// Store the results.
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(dst0));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(dst1));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(dst2));
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(dst3));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(dst0));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(dst1));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(dst2));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(dst3));
}
#undef MUL
#endif // USE_TRANSFORM_AC3
@@ -477,11 +477,11 @@ static WEBP_INLINE void Load8x4_SSE2(const uint8_t* const b, int stride,
// A0 = 63 62 61 60 23 22 21 20 43 42 41 40 03 02 01 00
// A1 = 73 72 71 70 33 32 31 30 53 52 51 50 13 12 11 10
const __m128i A0 = _mm_set_epi32(
WebPMemToUint32(&b[6 * stride]), WebPMemToUint32(&b[2 * stride]),
WebPMemToUint32(&b[4 * stride]), WebPMemToUint32(&b[0 * stride]));
WebPMemToInt32(&b[6 * stride]), WebPMemToInt32(&b[2 * stride]),
WebPMemToInt32(&b[4 * stride]), WebPMemToInt32(&b[0 * stride]));
const __m128i A1 = _mm_set_epi32(
WebPMemToUint32(&b[7 * stride]), WebPMemToUint32(&b[3 * stride]),
WebPMemToUint32(&b[5 * stride]), WebPMemToUint32(&b[1 * stride]));
WebPMemToInt32(&b[7 * stride]), WebPMemToInt32(&b[3 * stride]),
WebPMemToInt32(&b[5 * stride]), WebPMemToInt32(&b[1 * stride]));
// B0 = 53 43 52 42 51 41 50 40 13 03 12 02 11 01 10 00
// B1 = 73 63 72 62 71 61 70 60 33 23 32 22 31 21 30 20
@@ -540,7 +540,7 @@ static WEBP_INLINE void Store4x4_SSE2(__m128i* const x,
uint8_t* dst, int stride) {
int i;
for (i = 0; i < 4; ++i, dst += stride) {
WebPUint32ToMem(dst, _mm_cvtsi128_si32(*x));
WebPInt32ToMem(dst, _mm_cvtsi128_si32(*x));
*x = _mm_srli_si128(*x, 4);
}
}
@@ -908,10 +908,10 @@ static void VE4_SSE2(uint8_t* dst) { // vertical
const __m128i lsb = _mm_and_si128(_mm_xor_si128(ABCDEFGH, CDEFGH00), one);
const __m128i b = _mm_subs_epu8(a, lsb);
const __m128i avg = _mm_avg_epu8(b, BCDEFGH0);
const uint32_t vals = _mm_cvtsi128_si32(avg);
const int vals = _mm_cvtsi128_si32(avg);
int i;
for (i = 0; i < 4; ++i) {
WebPUint32ToMem(dst + i * BPS, vals);
WebPInt32ToMem(dst + i * BPS, vals);
}
}
@@ -925,10 +925,10 @@ static void LD4_SSE2(uint8_t* dst) { // Down-Left
const __m128i lsb = _mm_and_si128(_mm_xor_si128(ABCDEFGH, CDEFGHH0), one);
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
const __m128i abcdefg = _mm_avg_epu8(avg2, BCDEFGH0);
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( abcdefg ));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( abcdefg ));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
}
static void VR4_SSE2(uint8_t* dst) { // Vertical-Right
@@ -946,10 +946,10 @@ static void VR4_SSE2(uint8_t* dst) { // Vertical-Right
const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
const __m128i efgh = _mm_avg_epu8(avg2, XABCD);
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( abcd ));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32( efgh ));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(abcd, 1)));
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(efgh, 1)));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( abcd ));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32( efgh ));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(abcd, 1)));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(efgh, 1)));
// these two are hard to implement in SSE2, so we keep the C-version:
DST(0, 2) = AVG3(J, I, X);
@@ -970,11 +970,12 @@ static void VL4_SSE2(uint8_t* dst) { // Vertical-Left
const __m128i abbc = _mm_or_si128(ab, bc);
const __m128i lsb2 = _mm_and_si128(abbc, lsb1);
const __m128i avg4 = _mm_subs_epu8(avg3, lsb2);
const uint32_t extra_out = _mm_cvtsi128_si32(_mm_srli_si128(avg4, 4));
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( avg1 ));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32( avg4 ));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg1, 1)));
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg4, 1)));
const uint32_t extra_out =
(uint32_t)_mm_cvtsi128_si32(_mm_srli_si128(avg4, 4));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( avg1 ));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32( avg4 ));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg1, 1)));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg4, 1)));
// these two are hard to get and irregular
DST(3, 2) = (extra_out >> 0) & 0xff;
@@ -990,7 +991,7 @@ static void RD4_SSE2(uint8_t* dst) { // Down-right
const uint32_t K = dst[-1 + 2 * BPS];
const uint32_t L = dst[-1 + 3 * BPS];
const __m128i LKJI_____ =
_mm_cvtsi32_si128(L | (K << 8) | (J << 16) | (I << 24));
_mm_cvtsi32_si128((int)(L | (K << 8) | (J << 16) | (I << 24)));
const __m128i LKJIXABCD = _mm_or_si128(LKJI_____, ____XABCD);
const __m128i KJIXABCD_ = _mm_srli_si128(LKJIXABCD, 1);
const __m128i JIXABCD__ = _mm_srli_si128(LKJIXABCD, 2);
@@ -998,10 +999,10 @@ static void RD4_SSE2(uint8_t* dst) { // Down-right
const __m128i lsb = _mm_and_si128(_mm_xor_si128(JIXABCD__, LKJIXABCD), one);
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
const __m128i abcdefg = _mm_avg_epu8(avg2, KJIXABCD_);
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32( abcdefg ));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32( abcdefg ));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
}
#undef DST
@@ -1015,13 +1016,13 @@ static WEBP_INLINE void TrueMotion_SSE2(uint8_t* dst, int size) {
const __m128i zero = _mm_setzero_si128();
int y;
if (size == 4) {
const __m128i top_values = _mm_cvtsi32_si128(WebPMemToUint32(top));
const __m128i top_values = _mm_cvtsi32_si128(WebPMemToInt32(top));
const __m128i top_base = _mm_unpacklo_epi8(top_values, zero);
for (y = 0; y < 4; ++y, dst += BPS) {
const int val = dst[-1] - top[-1];
const __m128i base = _mm_set1_epi16(val);
const __m128i out = _mm_packus_epi16(_mm_add_epi16(base, top_base), zero);
WebPUint32ToMem(dst, _mm_cvtsi128_si32(out));
WebPInt32ToMem(dst, _mm_cvtsi128_si32(out));
}
} else if (size == 8) {
const __m128i top_values = _mm_loadl_epi64((const __m128i*)top);
@@ -1062,7 +1063,7 @@ static void VE16_SSE2(uint8_t* dst) {
static void HE16_SSE2(uint8_t* dst) { // horizontal
int j;
for (j = 16; j > 0; --j) {
const __m128i values = _mm_set1_epi8(dst[-1]);
const __m128i values = _mm_set1_epi8((char)dst[-1]);
_mm_storeu_si128((__m128i*)dst, values);
dst += BPS;
}
@@ -1070,7 +1071,7 @@ static void HE16_SSE2(uint8_t* dst) { // horizontal
static WEBP_INLINE void Put16_SSE2(uint8_t v, uint8_t* dst) {
int j;
const __m128i values = _mm_set1_epi8(v);
const __m128i values = _mm_set1_epi8((char)v);
for (j = 0; j < 16; ++j) {
_mm_storeu_si128((__m128i*)(dst + j * BPS), values);
}
@@ -1130,7 +1131,7 @@ static void VE8uv_SSE2(uint8_t* dst) { // vertical
// helper for chroma-DC predictions
static WEBP_INLINE void Put8x8uv_SSE2(uint8_t v, uint8_t* dst) {
int j;
const __m128i values = _mm_set1_epi8(v);
const __m128i values = _mm_set1_epi8((char)v);
for (j = 0; j < 8; ++j) {
_mm_storel_epi64((__m128i*)(dst + j * BPS), values);
}

2
thirdparty/libwebp/src/dsp/dec_sse41.c vendored
View File

@@ -23,7 +23,7 @@ static void HE16_SSE41(uint8_t* dst) { // horizontal
int j;
const __m128i kShuffle3 = _mm_set1_epi8(3);
for (j = 16; j > 0; --j) {
const __m128i in = _mm_cvtsi32_si128(WebPMemToUint32(dst - 4));
const __m128i in = _mm_cvtsi32_si128(WebPMemToInt32(dst - 4));
const __m128i values = _mm_shuffle_epi8(in, kShuffle3);
_mm_storeu_si128((__m128i*)dst, values);
dst += BPS;

9
thirdparty/libwebp/src/dsp/enc_neon.c vendored
View File

@@ -764,9 +764,14 @@ static WEBP_INLINE void AccumulateSSE16_NEON(const uint8_t* const a,
// Horizontal sum of all four uint32_t values in 'sum'.
static int SumToInt_NEON(uint32x4_t sum) {
#if defined(__aarch64__)
return (int)vaddvq_u32(sum);
#else
const uint64x2_t sum2 = vpaddlq_u32(sum);
const uint64_t sum3 = vgetq_lane_u64(sum2, 0) + vgetq_lane_u64(sum2, 1);
return (int)sum3;
const uint32x2_t sum3 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(sum2)),
vreinterpret_u32_u64(vget_high_u64(sum2)));
return (int)vget_lane_u32(sum3, 0);
#endif
}
static int SSE16x16_NEON(const uint8_t* a, const uint8_t* b) {

67
thirdparty/libwebp/src/dsp/enc_sse2.c vendored
View File

@@ -156,10 +156,10 @@ static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
ref3 = _mm_loadl_epi64((const __m128i*)&ref[3 * BPS]);
} else {
// Load four bytes/pixels per line.
ref0 = _mm_cvtsi32_si128(WebPMemToUint32(&ref[0 * BPS]));
ref1 = _mm_cvtsi32_si128(WebPMemToUint32(&ref[1 * BPS]));
ref2 = _mm_cvtsi32_si128(WebPMemToUint32(&ref[2 * BPS]));
ref3 = _mm_cvtsi32_si128(WebPMemToUint32(&ref[3 * BPS]));
ref0 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[0 * BPS]));
ref1 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[1 * BPS]));
ref2 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[2 * BPS]));
ref3 = _mm_cvtsi32_si128(WebPMemToInt32(&ref[3 * BPS]));
}
// Convert to 16b.
ref0 = _mm_unpacklo_epi8(ref0, zero);
@@ -185,10 +185,10 @@ static void ITransform_SSE2(const uint8_t* ref, const int16_t* in, uint8_t* dst,
_mm_storel_epi64((__m128i*)&dst[3 * BPS], ref3);
} else {
// Store four bytes/pixels per line.
WebPUint32ToMem(&dst[0 * BPS], _mm_cvtsi128_si32(ref0));
WebPUint32ToMem(&dst[1 * BPS], _mm_cvtsi128_si32(ref1));
WebPUint32ToMem(&dst[2 * BPS], _mm_cvtsi128_si32(ref2));
WebPUint32ToMem(&dst[3 * BPS], _mm_cvtsi128_si32(ref3));
WebPInt32ToMem(&dst[0 * BPS], _mm_cvtsi128_si32(ref0));
WebPInt32ToMem(&dst[1 * BPS], _mm_cvtsi128_si32(ref1));
WebPInt32ToMem(&dst[2 * BPS], _mm_cvtsi128_si32(ref2));
WebPInt32ToMem(&dst[3 * BPS], _mm_cvtsi128_si32(ref3));
}
}
}
@@ -481,7 +481,7 @@ static void CollectHistogram_SSE2(const uint8_t* ref, const uint8_t* pred,
// helper for chroma-DC predictions
static WEBP_INLINE void Put8x8uv_SSE2(uint8_t v, uint8_t* dst) {
int j;
const __m128i values = _mm_set1_epi8(v);
const __m128i values = _mm_set1_epi8((char)v);
for (j = 0; j < 8; ++j) {
_mm_storel_epi64((__m128i*)(dst + j * BPS), values);
}
@@ -489,7 +489,7 @@ static WEBP_INLINE void Put8x8uv_SSE2(uint8_t v, uint8_t* dst) {
static WEBP_INLINE void Put16_SSE2(uint8_t v, uint8_t* dst) {
int j;
const __m128i values = _mm_set1_epi8(v);
const __m128i values = _mm_set1_epi8((char)v);
for (j = 0; j < 16; ++j) {
_mm_store_si128((__m128i*)(dst + j * BPS), values);
}
@@ -540,7 +540,7 @@ static WEBP_INLINE void VerticalPred_SSE2(uint8_t* dst,
static WEBP_INLINE void HE8uv_SSE2(uint8_t* dst, const uint8_t* left) {
int j;
for (j = 0; j < 8; ++j) {
const __m128i values = _mm_set1_epi8(left[j]);
const __m128i values = _mm_set1_epi8((char)left[j]);
_mm_storel_epi64((__m128i*)dst, values);
dst += BPS;
}
@@ -549,7 +549,7 @@ static WEBP_INLINE void HE8uv_SSE2(uint8_t* dst, const uint8_t* left) {
static WEBP_INLINE void HE16_SSE2(uint8_t* dst, const uint8_t* left) {
int j;
for (j = 0; j < 16; ++j) {
const __m128i values = _mm_set1_epi8(left[j]);
const __m128i values = _mm_set1_epi8((char)left[j]);
_mm_store_si128((__m128i*)dst, values);
dst += BPS;
}
@@ -722,10 +722,10 @@ static WEBP_INLINE void VE4_SSE2(uint8_t* dst,
const __m128i lsb = _mm_and_si128(_mm_xor_si128(ABCDEFGH, CDEFGH00), one);
const __m128i b = _mm_subs_epu8(a, lsb);
const __m128i avg = _mm_avg_epu8(b, BCDEFGH0);
const uint32_t vals = _mm_cvtsi128_si32(avg);
const int vals = _mm_cvtsi128_si32(avg);
int i;
for (i = 0; i < 4; ++i) {
WebPUint32ToMem(dst + i * BPS, vals);
WebPInt32ToMem(dst + i * BPS, vals);
}
}
@@ -760,10 +760,10 @@ static WEBP_INLINE void LD4_SSE2(uint8_t* dst,
const __m128i lsb = _mm_and_si128(_mm_xor_si128(ABCDEFGH, CDEFGHH0), one);
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
const __m128i abcdefg = _mm_avg_epu8(avg2, BCDEFGH0);
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( abcdefg ));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( abcdefg ));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
}
static WEBP_INLINE void VR4_SSE2(uint8_t* dst,
@@ -782,10 +782,10 @@ static WEBP_INLINE void VR4_SSE2(uint8_t* dst,
const __m128i lsb = _mm_and_si128(_mm_xor_si128(IXABCD, ABCD0), one);
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
const __m128i efgh = _mm_avg_epu8(avg2, XABCD);
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( abcd ));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32( efgh ));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(abcd, 1)));
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(efgh, 1)));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( abcd ));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32( efgh ));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(abcd, 1)));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_slli_si128(efgh, 1)));
// these two are hard to implement in SSE2, so we keep the C-version:
DST(0, 2) = AVG3(J, I, X);
@@ -807,11 +807,12 @@ static WEBP_INLINE void VL4_SSE2(uint8_t* dst,
const __m128i abbc = _mm_or_si128(ab, bc);
const __m128i lsb2 = _mm_and_si128(abbc, lsb1);
const __m128i avg4 = _mm_subs_epu8(avg3, lsb2);
const uint32_t extra_out = _mm_cvtsi128_si32(_mm_srli_si128(avg4, 4));
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( avg1 ));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32( avg4 ));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg1, 1)));
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg4, 1)));
const uint32_t extra_out =
(uint32_t)_mm_cvtsi128_si32(_mm_srli_si128(avg4, 4));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32( avg1 ));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32( avg4 ));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg1, 1)));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(avg4, 1)));
// these two are hard to get and irregular
DST(3, 2) = (extra_out >> 0) & 0xff;
@@ -829,10 +830,10 @@ static WEBP_INLINE void RD4_SSE2(uint8_t* dst,
const __m128i lsb = _mm_and_si128(_mm_xor_si128(JIXABCD__, LKJIXABCD), one);
const __m128i avg2 = _mm_subs_epu8(avg1, lsb);
const __m128i abcdefg = _mm_avg_epu8(avg2, KJIXABCD_);
WebPUint32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32( abcdefg ));
WebPUint32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
WebPUint32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
WebPUint32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
WebPInt32ToMem(dst + 3 * BPS, _mm_cvtsi128_si32( abcdefg ));
WebPInt32ToMem(dst + 2 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 1)));
WebPInt32ToMem(dst + 1 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 2)));
WebPInt32ToMem(dst + 0 * BPS, _mm_cvtsi128_si32(_mm_srli_si128(abcdefg, 3)));
}
static WEBP_INLINE void HU4_SSE2(uint8_t* dst, const uint8_t* top) {
@@ -875,14 +876,14 @@ static WEBP_INLINE void HD4_SSE2(uint8_t* dst, const uint8_t* top) {
static WEBP_INLINE void TM4_SSE2(uint8_t* dst, const uint8_t* top) {
const __m128i zero = _mm_setzero_si128();
const __m128i top_values = _mm_cvtsi32_si128(WebPMemToUint32(top));
const __m128i top_values = _mm_cvtsi32_si128(WebPMemToInt32(top));
const __m128i top_base = _mm_unpacklo_epi8(top_values, zero);
int y;
for (y = 0; y < 4; ++y, dst += BPS) {
const int val = top[-2 - y] - top[-1];
const __m128i base = _mm_set1_epi16(val);
const __m128i out = _mm_packus_epi16(_mm_add_epi16(base, top_base), zero);
WebPUint32ToMem(dst, _mm_cvtsi128_si32(out));
WebPInt32ToMem(dst, _mm_cvtsi128_si32(out));
}
}

12
thirdparty/libwebp/src/dsp/lossless.c vendored
View File

@@ -49,7 +49,7 @@ static WEBP_INLINE uint32_t Clip255(uint32_t a) {
}
static WEBP_INLINE int AddSubtractComponentFull(int a, int b, int c) {
return Clip255(a + b - c);
return Clip255((uint32_t)(a + b - c));
}
static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
@@ -66,7 +66,7 @@ static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1,
}
static WEBP_INLINE int AddSubtractComponentHalf(int a, int b) {
return Clip255(a + (a - b) / 2);
return Clip255((uint32_t)(a + (a - b) / 2));
}
static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1,
@@ -293,10 +293,10 @@ void VP8LTransformColorInverse_C(const VP8LMultipliers* const m,
const uint32_t red = argb >> 16;
int new_red = red & 0xff;
int new_blue = argb & 0xff;
new_red += ColorTransformDelta(m->green_to_red_, green);
new_red += ColorTransformDelta((int8_t)m->green_to_red_, green);
new_red &= 0xff;
new_blue += ColorTransformDelta(m->green_to_blue_, green);
new_blue += ColorTransformDelta(m->red_to_blue_, (int8_t)new_red);
new_blue += ColorTransformDelta((int8_t)m->green_to_blue_, green);
new_blue += ColorTransformDelta((int8_t)m->red_to_blue_, (int8_t)new_red);
new_blue &= 0xff;
dst[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue);
}
@@ -395,7 +395,7 @@ void VP8LInverseTransform(const VP8LTransform* const transform,
assert(row_start < row_end);
assert(row_end <= transform->ysize_);
switch (transform->type_) {
case SUBTRACT_GREEN:
case SUBTRACT_GREEN_TRANSFORM:
VP8LAddGreenToBlueAndRed(in, (row_end - row_start) * width, out);
break;
case PREDICTOR_TRANSFORM:

18
thirdparty/libwebp/src/dsp/lossless_enc.c vendored
View File

@@ -522,11 +522,11 @@ static void GetCombinedEntropyUnrefined_C(const uint32_t X[],
void VP8LSubtractGreenFromBlueAndRed_C(uint32_t* argb_data, int num_pixels) {
int i;
for (i = 0; i < num_pixels; ++i) {
const int argb = argb_data[i];
const int argb = (int)argb_data[i];
const int green = (argb >> 8) & 0xff;
const uint32_t new_r = (((argb >> 16) & 0xff) - green) & 0xff;
const uint32_t new_b = (((argb >> 0) & 0xff) - green) & 0xff;
argb_data[i] = (argb & 0xff00ff00u) | (new_r << 16) | new_b;
argb_data[i] = ((uint32_t)argb & 0xff00ff00u) | (new_r << 16) | new_b;
}
}
@@ -547,10 +547,10 @@ void VP8LTransformColor_C(const VP8LMultipliers* const m, uint32_t* data,
const int8_t red = U32ToS8(argb >> 16);
int new_red = red & 0xff;
int new_blue = argb & 0xff;
new_red -= ColorTransformDelta(m->green_to_red_, green);
new_red -= ColorTransformDelta((int8_t)m->green_to_red_, green);
new_red &= 0xff;
new_blue -= ColorTransformDelta(m->green_to_blue_, green);
new_blue -= ColorTransformDelta(m->red_to_blue_, red);
new_blue -= ColorTransformDelta((int8_t)m->green_to_blue_, green);
new_blue -= ColorTransformDelta((int8_t)m->red_to_blue_, red);
new_blue &= 0xff;
data[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue);
}
@@ -560,7 +560,7 @@ static WEBP_INLINE uint8_t TransformColorRed(uint8_t green_to_red,
uint32_t argb) {
const int8_t green = U32ToS8(argb >> 8);
int new_red = argb >> 16;
new_red -= ColorTransformDelta(green_to_red, green);
new_red -= ColorTransformDelta((int8_t)green_to_red, green);
return (new_red & 0xff);
}
@@ -569,9 +569,9 @@ static WEBP_INLINE uint8_t TransformColorBlue(uint8_t green_to_blue,
uint32_t argb) {
const int8_t green = U32ToS8(argb >> 8);
const int8_t red = U32ToS8(argb >> 16);
uint8_t new_blue = argb & 0xff;
new_blue -= ColorTransformDelta(green_to_blue, green);
new_blue -= ColorTransformDelta(red_to_blue, red);
int new_blue = argb & 0xff;
new_blue -= ColorTransformDelta((int8_t)green_to_blue, green);
new_blue -= ColorTransformDelta((int8_t)red_to_blue, red);
return (new_blue & 0xff);
}

8
thirdparty/libwebp/src/dsp/lossless_enc_sse2.c vendored
View File

@@ -54,8 +54,8 @@ static void TransformColor_SSE2(const VP8LMultipliers* const m,
const __m128i mults_rb = MK_CST_16(CST_5b(m->green_to_red_),
CST_5b(m->green_to_blue_));
const __m128i mults_b2 = MK_CST_16(CST_5b(m->red_to_blue_), 0);
const __m128i mask_ag = _mm_set1_epi32(0xff00ff00); // alpha-green masks
const __m128i mask_rb = _mm_set1_epi32(0x00ff00ff); // red-blue masks
const __m128i mask_ag = _mm_set1_epi32((int)0xff00ff00); // alpha-green masks
const __m128i mask_rb = _mm_set1_epi32(0x00ff00ff); // red-blue masks
int i;
for (i = 0; i + 4 <= num_pixels; i += 4) {
const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]); // argb
@@ -376,7 +376,7 @@ static void BundleColorMap_SSE2(const uint8_t* const row, int width, int xbits,
break;
}
case 2: {
const __m128i mask_or = _mm_set1_epi32(0xff000000);
const __m128i mask_or = _mm_set1_epi32((int)0xff000000);
const __m128i mul_cst = _mm_set1_epi16(0x0104);
const __m128i mask_mul = _mm_set1_epi16(0x0f00);
for (x = 0; x + 16 <= width; x += 16, dst += 4) {
@@ -427,7 +427,7 @@ static WEBP_INLINE void Average2_m128i(const __m128i* const a0,
static void PredictorSub0_SSE2(const uint32_t* in, const uint32_t* upper,
int num_pixels, uint32_t* out) {
int i;
const __m128i black = _mm_set1_epi32(ARGB_BLACK);
const __m128i black = _mm_set1_epi32((int)ARGB_BLACK);
for (i = 0; i + 4 <= num_pixels; i += 4) {
const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
const __m128i res = _mm_sub_epi8(src, black);

88
thirdparty/libwebp/src/dsp/lossless_sse2.c vendored
View File

@@ -27,23 +27,22 @@ static WEBP_INLINE uint32_t ClampedAddSubtractFull_SSE2(uint32_t c0,
uint32_t c1,
uint32_t c2) {
const __m128i zero = _mm_setzero_si128();
const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
const __m128i C2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero);
const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c0), zero);
const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c1), zero);
const __m128i C2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c2), zero);
const __m128i V1 = _mm_add_epi16(C0, C1);
const __m128i V2 = _mm_sub_epi16(V1, C2);
const __m128i b = _mm_packus_epi16(V2, V2);
const uint32_t output = _mm_cvtsi128_si32(b);
return output;
return (uint32_t)_mm_cvtsi128_si32(b);
}
static WEBP_INLINE uint32_t ClampedAddSubtractHalf_SSE2(uint32_t c0,
uint32_t c1,
uint32_t c2) {
const __m128i zero = _mm_setzero_si128();
const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
const __m128i B0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero);
const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c0), zero);
const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c1), zero);
const __m128i B0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)c2), zero);
const __m128i avg = _mm_add_epi16(C1, C0);
const __m128i A0 = _mm_srli_epi16(avg, 1);
const __m128i A1 = _mm_sub_epi16(A0, B0);
@@ -52,16 +51,15 @@ static WEBP_INLINE uint32_t ClampedAddSubtractHalf_SSE2(uint32_t c0,
const __m128i A3 = _mm_srai_epi16(A2, 1);
const __m128i A4 = _mm_add_epi16(A0, A3);
const __m128i A5 = _mm_packus_epi16(A4, A4);
const uint32_t output = _mm_cvtsi128_si32(A5);
return output;
return (uint32_t)_mm_cvtsi128_si32(A5);
}
static WEBP_INLINE uint32_t Select_SSE2(uint32_t a, uint32_t b, uint32_t c) {
int pa_minus_pb;
const __m128i zero = _mm_setzero_si128();
const __m128i A0 = _mm_cvtsi32_si128(a);
const __m128i B0 = _mm_cvtsi32_si128(b);
const __m128i C0 = _mm_cvtsi32_si128(c);
const __m128i A0 = _mm_cvtsi32_si128((int)a);
const __m128i B0 = _mm_cvtsi32_si128((int)b);
const __m128i C0 = _mm_cvtsi32_si128((int)c);
const __m128i AC0 = _mm_subs_epu8(A0, C0);
const __m128i CA0 = _mm_subs_epu8(C0, A0);
const __m128i BC0 = _mm_subs_epu8(B0, C0);
@@ -94,8 +92,8 @@ static WEBP_INLINE void Average2_uint32_SSE2(const uint32_t a0,
__m128i* const avg) {
// (a + b) >> 1 = ((a + b + 1) >> 1) - ((a ^ b) & 1)
const __m128i ones = _mm_set1_epi8(1);
const __m128i A0 = _mm_cvtsi32_si128(a0);
const __m128i A1 = _mm_cvtsi32_si128(a1);
const __m128i A0 = _mm_cvtsi32_si128((int)a0);
const __m128i A1 = _mm_cvtsi32_si128((int)a1);
const __m128i avg1 = _mm_avg_epu8(A0, A1);
const __m128i one = _mm_and_si128(_mm_xor_si128(A0, A1), ones);
*avg = _mm_sub_epi8(avg1, one);
@@ -103,8 +101,8 @@ static WEBP_INLINE void Average2_uint32_SSE2(const uint32_t a0,
static WEBP_INLINE __m128i Average2_uint32_16_SSE2(uint32_t a0, uint32_t a1) {
const __m128i zero = _mm_setzero_si128();
const __m128i A0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a0), zero);
const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
const __m128i A0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)a0), zero);
const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)a1), zero);
const __m128i sum = _mm_add_epi16(A1, A0);
return _mm_srli_epi16(sum, 1);
}
@@ -112,19 +110,18 @@ static WEBP_INLINE __m128i Average2_uint32_16_SSE2(uint32_t a0, uint32_t a1) {
static WEBP_INLINE uint32_t Average2_SSE2(uint32_t a0, uint32_t a1) {
__m128i output;
Average2_uint32_SSE2(a0, a1, &output);
return _mm_cvtsi128_si32(output);
return (uint32_t)_mm_cvtsi128_si32(output);
}
static WEBP_INLINE uint32_t Average3_SSE2(uint32_t a0, uint32_t a1,
uint32_t a2) {
const __m128i zero = _mm_setzero_si128();
const __m128i avg1 = Average2_uint32_16_SSE2(a0, a2);
const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128((int)a1), zero);
const __m128i sum = _mm_add_epi16(avg1, A1);
const __m128i avg2 = _mm_srli_epi16(sum, 1);
const __m128i A2 = _mm_packus_epi16(avg2, avg2);
const uint32_t output = _mm_cvtsi128_si32(A2);
return output;
return (uint32_t)_mm_cvtsi128_si32(A2);
}
static WEBP_INLINE uint32_t Average4_SSE2(uint32_t a0, uint32_t a1,
@@ -134,8 +131,7 @@ static WEBP_INLINE uint32_t Average4_SSE2(uint32_t a0, uint32_t a1,
const __m128i sum = _mm_add_epi16(avg2, avg1);
const __m128i avg3 = _mm_srli_epi16(sum, 1);
const __m128i A0 = _mm_packus_epi16(avg3, avg3);
const uint32_t output = _mm_cvtsi128_si32(A0);
return output;
return (uint32_t)_mm_cvtsi128_si32(A0);
}
static uint32_t Predictor5_SSE2(const uint32_t* const left,
@@ -192,7 +188,7 @@ static uint32_t Predictor13_SSE2(const uint32_t* const left,
static void PredictorAdd0_SSE2(const uint32_t* in, const uint32_t* upper,
int num_pixels, uint32_t* out) {
int i;
const __m128i black = _mm_set1_epi32(ARGB_BLACK);
const __m128i black = _mm_set1_epi32((int)ARGB_BLACK);
for (i = 0; i + 4 <= num_pixels; i += 4) {
const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
const __m128i res = _mm_add_epi8(src, black);
@@ -208,7 +204,7 @@ static void PredictorAdd0_SSE2(const uint32_t* in, const uint32_t* upper,
static void PredictorAdd1_SSE2(const uint32_t* in, const uint32_t* upper,
int num_pixels, uint32_t* out) {
int i;
__m128i prev = _mm_set1_epi32(out[-1]);
__m128i prev = _mm_set1_epi32((int)out[-1]);
for (i = 0; i + 4 <= num_pixels; i += 4) {
// a | b | c | d
const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
@@ -285,12 +281,12 @@ GENERATE_PREDICTOR_2(9, upper[i + 1])
#undef GENERATE_PREDICTOR_2
// Predictor10: average of (average of (L,TL), average of (T, TR)).
#define DO_PRED10(OUT) do { \
__m128i avgLTL, avg; \
Average2_m128i(&L, &TL, &avgLTL); \
Average2_m128i(&avgTTR, &avgLTL, &avg); \
L = _mm_add_epi8(avg, src); \
out[i + (OUT)] = _mm_cvtsi128_si32(L); \
#define DO_PRED10(OUT) do { \
__m128i avgLTL, avg; \
Average2_m128i(&L, &TL, &avgLTL); \
Average2_m128i(&avgTTR, &avgLTL, &avg); \
L = _mm_add_epi8(avg, src); \
out[i + (OUT)] = (uint32_t)_mm_cvtsi128_si32(L); \
} while (0)
#define DO_PRED10_SHIFT do { \
@@ -303,7 +299,7 @@ GENERATE_PREDICTOR_2(9, upper[i + 1])
static void PredictorAdd10_SSE2(const uint32_t* in, const uint32_t* upper,
int num_pixels, uint32_t* out) {
int i;
__m128i L = _mm_cvtsi32_si128(out[-1]);
__m128i L = _mm_cvtsi32_si128((int)out[-1]);
for (i = 0; i + 4 <= num_pixels; i += 4) {
__m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
__m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]);
@@ -336,7 +332,7 @@ static void PredictorAdd10_SSE2(const uint32_t* in, const uint32_t* upper,
const __m128i B = _mm_andnot_si128(mask, T); \
const __m128i pred = _mm_or_si128(A, B); /* pred = (pa > b)? L : T*/ \
L = _mm_add_epi8(src, pred); \
out[i + (OUT)] = _mm_cvtsi128_si32(L); \
out[i + (OUT)] = (uint32_t)_mm_cvtsi128_si32(L); \
} while (0)
#define DO_PRED11_SHIFT do { \
@@ -351,7 +347,7 @@ static void PredictorAdd11_SSE2(const uint32_t* in, const uint32_t* upper,
int num_pixels, uint32_t* out) {
int i;
__m128i pa;
__m128i L = _mm_cvtsi32_si128(out[-1]);
__m128i L = _mm_cvtsi32_si128((int)out[-1]);
for (i = 0; i + 4 <= num_pixels; i += 4) {
__m128i T = _mm_loadu_si128((const __m128i*)&upper[i]);
__m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]);
@@ -384,12 +380,12 @@ static void PredictorAdd11_SSE2(const uint32_t* in, const uint32_t* upper,
#undef DO_PRED11_SHIFT
// Predictor12: ClampedAddSubtractFull.
#define DO_PRED12(DIFF, LANE, OUT) do { \
const __m128i all = _mm_add_epi16(L, (DIFF)); \
const __m128i alls = _mm_packus_epi16(all, all); \
const __m128i res = _mm_add_epi8(src, alls); \
out[i + (OUT)] = _mm_cvtsi128_si32(res); \
L = _mm_unpacklo_epi8(res, zero); \
#define DO_PRED12(DIFF, LANE, OUT) do { \
const __m128i all = _mm_add_epi16(L, (DIFF)); \
const __m128i alls = _mm_packus_epi16(all, all); \
const __m128i res = _mm_add_epi8(src, alls); \
out[i + (OUT)] = (uint32_t)_mm_cvtsi128_si32(res); \
L = _mm_unpacklo_epi8(res, zero); \
} while (0)
#define DO_PRED12_SHIFT(DIFF, LANE) do { \
@@ -402,7 +398,7 @@ static void PredictorAdd12_SSE2(const uint32_t* in, const uint32_t* upper,
int num_pixels, uint32_t* out) {
int i;
const __m128i zero = _mm_setzero_si128();
const __m128i L8 = _mm_cvtsi32_si128(out[-1]);
const __m128i L8 = _mm_cvtsi32_si128((int)out[-1]);
__m128i L = _mm_unpacklo_epi8(L8, zero);
for (i = 0; i + 4 <= num_pixels; i += 4) {
// Load 4 pixels at a time.
@@ -468,7 +464,7 @@ static void TransformColorInverse_SSE2(const VP8LMultipliers* const m,
const __m128i mults_b2 = MK_CST_16(CST(red_to_blue_), 0);
#undef MK_CST_16
#undef CST
const __m128i mask_ag = _mm_set1_epi32(0xff00ff00); // alpha-green masks
const __m128i mask_ag = _mm_set1_epi32((int)0xff00ff00); // alpha-green masks
int i;
for (i = 0; i + 4 <= num_pixels; i += 4) {
const __m128i in = _mm_loadu_si128((const __m128i*)&src[i]); // argb
@@ -532,7 +528,7 @@ static void ConvertBGRAToRGB_SSE2(const uint32_t* src, int num_pixels,
static void ConvertBGRAToRGBA_SSE2(const uint32_t* src,
int num_pixels, uint8_t* dst) {
const __m128i red_blue_mask = _mm_set1_epi32(0x00ff00ffu);
const __m128i red_blue_mask = _mm_set1_epi32(0x00ff00ff);
const __m128i* in = (const __m128i*)src;
__m128i* out = (__m128i*)dst;
while (num_pixels >= 8) {
@@ -561,7 +557,7 @@ static void ConvertBGRAToRGBA_SSE2(const uint32_t* src,
static void ConvertBGRAToRGBA4444_SSE2(const uint32_t* src,
int num_pixels, uint8_t* dst) {
const __m128i mask_0x0f = _mm_set1_epi8(0x0f);
const __m128i mask_0xf0 = _mm_set1_epi8(0xf0);
const __m128i mask_0xf0 = _mm_set1_epi8((char)0xf0);
const __m128i* in = (const __m128i*)src;
__m128i* out = (__m128i*)dst;
while (num_pixels >= 8) {
@@ -596,8 +592,8 @@ static void ConvertBGRAToRGBA4444_SSE2(const uint32_t* src,
static void ConvertBGRAToRGB565_SSE2(const uint32_t* src,
int num_pixels, uint8_t* dst) {
const __m128i mask_0xe0 = _mm_set1_epi8(0xe0);
const __m128i mask_0xf8 = _mm_set1_epi8(0xf8);
const __m128i mask_0xe0 = _mm_set1_epi8((char)0xe0);
const __m128i mask_0xf8 = _mm_set1_epi8((char)0xf8);
const __m128i mask_0x07 = _mm_set1_epi8(0x07);
const __m128i* in = (const __m128i*)src;
__m128i* out = (__m128i*)dst;

7
thirdparty/libwebp/src/dsp/lossless_sse41.c vendored
View File

@@ -25,11 +25,12 @@ static void TransformColorInverse_SSE41(const VP8LMultipliers* const m,
int num_pixels, uint32_t* dst) {
// sign-extended multiplying constants, pre-shifted by 5.
#define CST(X) (((int16_t)(m->X << 8)) >> 5) // sign-extend
const __m128i mults_rb = _mm_set1_epi32((uint32_t)CST(green_to_red_) << 16 |
(CST(green_to_blue_) & 0xffff));
const __m128i mults_rb =
_mm_set1_epi32((int)((uint32_t)CST(green_to_red_) << 16 |
(CST(green_to_blue_) & 0xffff)));
const __m128i mults_b2 = _mm_set1_epi32(CST(red_to_blue_));
#undef CST
const __m128i mask_ag = _mm_set1_epi32(0xff00ff00);
const __m128i mask_ag = _mm_set1_epi32((int)0xff00ff00);
const __m128i perm1 = _mm_setr_epi8(-1, 1, -1, 1, -1, 5, -1, 5,
-1, 9, -1, 9, -1, 13, -1, 13);
const __m128i perm2 = _mm_setr_epi8(-1, 2, -1, -1, -1, 6, -1, -1,

13
thirdparty/libwebp/src/dsp/quant.h vendored
View File

@@ -21,10 +21,15 @@
#define IsFlat IsFlat_NEON
static uint32x2_t horizontal_add_uint32x4(const uint32x4_t a) {
static uint32_t horizontal_add_uint32x4(const uint32x4_t a) {
#if defined(__aarch64__)
return vaddvq_u32(a);
#else
const uint64x2_t b = vpaddlq_u32(a);
return vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
vreinterpret_u32_u64(vget_high_u64(b)));
const uint32x2_t c = vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
vreinterpret_u32_u64(vget_high_u64(b)));
return vget_lane_u32(c, 0);
#endif
}
static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
@@ -45,7 +50,7 @@ static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
levels += 16;
}
return thresh >= (int32_t)vget_lane_u32(horizontal_add_uint32x4(sum), 0);
return thresh >= (int)horizontal_add_uint32x4(sum);
}
#else

6
thirdparty/libwebp/src/dsp/rescaler_sse2.c vendored
View File

@@ -85,7 +85,7 @@ static void RescalerImportRowExpand_SSE2(WebPRescaler* const wrk,
const __m128i mult = _mm_cvtsi32_si128(((x_add - accum) << 16) | accum);
const __m128i out = _mm_madd_epi16(cur_pixels, mult);
assert(sizeof(*frow) == sizeof(uint32_t));
WebPUint32ToMem((uint8_t*)frow, _mm_cvtsi128_si32(out));
WebPInt32ToMem((uint8_t*)frow, _mm_cvtsi128_si32(out));
frow += 1;
if (frow >= frow_end) break;
accum -= wrk->x_sub;
@@ -132,7 +132,7 @@ static void RescalerImportRowShrink_SSE2(WebPRescaler* const wrk,
__m128i base = zero;
accum += wrk->x_add;
while (accum > 0) {
const __m128i A = _mm_cvtsi32_si128(WebPMemToUint32(src));
const __m128i A = _mm_cvtsi32_si128(WebPMemToInt32(src));
src += 4;
base = _mm_unpacklo_epi8(A, zero);
// To avoid overflow, we need: base * x_add / x_sub < 32768
@@ -198,7 +198,7 @@ static WEBP_INLINE void ProcessRow_SSE2(const __m128i* const A0,
const __m128i* const mult,
uint8_t* const dst) {
const __m128i rounder = _mm_set_epi32(0, ROUNDER, 0, ROUNDER);
const __m128i mask = _mm_set_epi32(0xffffffffu, 0, 0xffffffffu, 0);
const __m128i mask = _mm_set_epi32(~0, 0, ~0, 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);

2
thirdparty/libwebp/src/dsp/upsampling_sse2.c vendored
View File

@@ -121,7 +121,7 @@ static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y, \
int uv_pos, pos; \
/* 16byte-aligned array to cache reconstructed u and v */ \
uint8_t uv_buf[14 * 32 + 15] = { 0 }; \
uint8_t* const r_u = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~15); \
uint8_t* const r_u = (uint8_t*)((uintptr_t)(uv_buf + 15) & ~(uintptr_t)15); \
uint8_t* const r_v = r_u + 32; \
\
assert(top_y != NULL); \

13
thirdparty/libwebp/src/dsp/yuv_sse2.c vendored
View File

@@ -15,10 +15,12 @@
#if defined(WEBP_USE_SSE2)
#include "src/dsp/common_sse2.h"
#include <stdlib.h>
#include <emmintrin.h>
#include "src/dsp/common_sse2.h"
#include "src/utils/utils.h"
//-----------------------------------------------------------------------------
// Convert spans of 32 pixels to various RGB formats for the fancy upsampler.
@@ -74,7 +76,7 @@ static WEBP_INLINE __m128i Load_HI_16_SSE2(const uint8_t* src) {
// Load and replicate the U/V samples
static WEBP_INLINE __m128i Load_UV_HI_8_SSE2(const uint8_t* src) {
const __m128i zero = _mm_setzero_si128();
const __m128i tmp0 = _mm_cvtsi32_si128(*(const uint32_t*)src);
const __m128i tmp0 = _mm_cvtsi32_si128(WebPMemToInt32(src));
const __m128i tmp1 = _mm_unpacklo_epi8(zero, tmp0);
return _mm_unpacklo_epi16(tmp1, tmp1); // replicate samples
}
@@ -130,7 +132,7 @@ static WEBP_INLINE void PackAndStore4444_SSE2(const __m128i* const R,
const __m128i rg0 = _mm_packus_epi16(*B, *A);
const __m128i ba0 = _mm_packus_epi16(*R, *G);
#endif
const __m128i mask_0xf0 = _mm_set1_epi8(0xf0);
const __m128i mask_0xf0 = _mm_set1_epi8((char)0xf0);
const __m128i rb1 = _mm_unpacklo_epi8(rg0, ba0); // rbrbrbrbrb...
const __m128i ga1 = _mm_unpackhi_epi8(rg0, ba0); // gagagagaga...
const __m128i rb2 = _mm_and_si128(rb1, mask_0xf0);
@@ -147,9 +149,10 @@ static WEBP_INLINE void PackAndStore565_SSE2(const __m128i* const R,
const __m128i r0 = _mm_packus_epi16(*R, *R);
const __m128i g0 = _mm_packus_epi16(*G, *G);
const __m128i b0 = _mm_packus_epi16(*B, *B);
const __m128i r1 = _mm_and_si128(r0, _mm_set1_epi8(0xf8));
const __m128i r1 = _mm_and_si128(r0, _mm_set1_epi8((char)0xf8));
const __m128i b1 = _mm_and_si128(_mm_srli_epi16(b0, 3), _mm_set1_epi8(0x1f));
const __m128i g1 = _mm_srli_epi16(_mm_and_si128(g0, _mm_set1_epi8(0xe0)), 5);
const __m128i g1 =
_mm_srli_epi16(_mm_and_si128(g0, _mm_set1_epi8((char)0xe0)), 5);
const __m128i g2 = _mm_slli_epi16(_mm_and_si128(g0, _mm_set1_epi8(0x1c)), 3);
const __m128i rg = _mm_or_si128(r1, g1);
const __m128i gb = _mm_or_si128(g2, b1);

6
thirdparty/libwebp/src/dsp/yuv_sse41.c vendored
View File

@@ -15,10 +15,12 @@
#if defined(WEBP_USE_SSE41)
#include "src/dsp/common_sse41.h"
#include <stdlib.h>
#include <smmintrin.h>
#include "src/dsp/common_sse41.h"
#include "src/utils/utils.h"
//-----------------------------------------------------------------------------
// Convert spans of 32 pixels to various RGB formats for the fancy upsampler.
@@ -74,7 +76,7 @@ static WEBP_INLINE __m128i Load_HI_16_SSE41(const uint8_t* src) {
// Load and replicate the U/V samples
static WEBP_INLINE __m128i Load_UV_HI_8_SSE41(const uint8_t* src) {
const __m128i zero = _mm_setzero_si128();
const __m128i tmp0 = _mm_cvtsi32_si128(*(const uint32_t*)src);
const __m128i tmp0 = _mm_cvtsi32_si128(WebPMemToInt32(src));
const __m128i tmp1 = _mm_unpacklo_epi8(zero, tmp0);
return _mm_unpacklo_epi16(tmp1, tmp1); // replicate samples
}