dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2024-11-20 04:18:26 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

Call the C function to finish off lossless SSE loops only when necessary.

Browse Source 
Change-Id: I4e221d80879dc9c90c24d69a40bc5811d73787ad
This commit is contained in:
Vincent Rabaud 2016-12-21 14:25:54 +01:00
parent 875fafc191
commit 1cb638010c
3 changed files with 75 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
src/dsp/lossless_enc_sse2.c
View File

@ -37,8 +37,10 @@ static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
_mm_storeu_si128((__m128i*)&argb_data[i], out);
}
// fallthrough and finish off with plain-C
if (i != num_pixels) {
VP8LSubtractGreenFromBlueAndRed_C(argb_data + i, num_pixels - i);
}
}
//------------------------------------------------------------------------------
// Color Transform
@ -71,8 +73,10 @@ static void TransformColor(const VP8LMultipliers* const m,
_mm_storeu_si128((__m128i*)&argb_data[i], out);
}
// fallthrough and finish off with plain-C
if (i != num_pixels) {
VP8LTransformColor_C(m, argb_data + i, num_pixels - i);
}
}
//------------------------------------------------------------------------------
#define SPAN 8
@ -477,8 +481,10 @@ static void PredictorSub0_SSE2(const uint32_t* in, const uint32_t* upper,
const __m128i res = _mm_sub_epi8(src, black);
_mm_storeu_si128((__m128i*)&out[i], res);
}
if (i != num_pixels) {
VP8LPredictorsSub_C[0](in + i, upper + i, num_pixels - i, out + i);
}
}
#define GENERATE_PREDICTOR_1(X, IN) \
static void PredictorSub##X##_SSE2(const uint32_t* in, const uint32_t* upper, \
@ -490,7 +496,9 @@ static void PredictorSub##X##_SSE2(const uint32_t* in, const uint32_t* upper, \
const __m128i res = _mm_sub_epi8(src, pred); \
_mm_storeu_si128((__m128i*)&out[i], res); \
} \
if (i != num_pixels) { \
VP8LPredictorsSub_C[(X)](in + i, upper + i, num_pixels - i, out + i); \
} \
}
GENERATE_PREDICTOR_1(1, in[i - 1]) // Predictor1: L
@ -514,8 +522,10 @@ static void PredictorSub5_SSE2(const uint32_t* in, const uint32_t* upper,
res = _mm_sub_epi8(src, pred);
_mm_storeu_si128((__m128i*)&out[i], res);
}
if (i != num_pixels) {
VP8LPredictorsSub_C[5](in + i, upper + i, num_pixels - i, out + i);
}
}
#define GENERATE_PREDICTOR_2(X, A, B) \
static void PredictorSub##X##_SSE2(const uint32_t* in, const uint32_t* upper, \
@ -530,7 +540,9 @@ static void PredictorSub##X##_SSE2(const uint32_t* in, const uint32_t* upper, \
res = _mm_sub_epi8(src, pred); \
_mm_storeu_si128((__m128i*)&out[i], res); \
} \
if (i != num_pixels) { \
VP8LPredictorsSub_C[(X)](in + i, upper + i, num_pixels - i, out + i); \
} \
}
GENERATE_PREDICTOR_2(6, in[i - 1], upper[i - 1]) // Predictor6: avg(L, TL)
@ -556,8 +568,10 @@ static void PredictorSub10_SSE2(const uint32_t* in, const uint32_t* upper,
res = _mm_sub_epi8(src, avg);
_mm_storeu_si128((__m128i*)&out[i], res);
}
if (i != num_pixels) {
VP8LPredictorsSub_C[10](in + i, upper + i, num_pixels - i, out + i);
}
}
// Predictor11: select.
static void GetSumAbsDiff32(const __m128i* const A, const __m128i* const B,
@ -593,8 +607,10 @@ static void PredictorSub11_SSE2(const uint32_t* in, const uint32_t* upper,
_mm_storeu_si128((__m128i*)&out[i], res);
}
}
if (i != num_pixels) {
VP8LPredictorsSub_C[11](in + i, upper + i, num_pixels - i, out + i);
}
}
// Predictor12: ClampedSubSubtractFull.
static void PredictorSub12_SSE2(const uint32_t* in, const uint32_t* upper,
@ -620,8 +636,10 @@ static void PredictorSub12_SSE2(const uint32_t* in, const uint32_t* upper,
const __m128i res = _mm_sub_epi8(src, pred);
_mm_storeu_si128((__m128i*)&out[i], res);
}
if (i != num_pixels) {
VP8LPredictorsSub_C[12](in + i, upper + i, num_pixels - i, out + i);
}
}
// Predictors13: ClampedAddSubtractHalf
static void PredictorSub13_SSE2(const uint32_t* in, const uint32_t* upper,
@ -648,8 +666,10 @@ static void PredictorSub13_SSE2(const uint32_t* in, const uint32_t* upper,
const __m128i res = _mm_sub_epi8(src, pred);
_mm_storel_epi64((__m128i*)&out[i], res);
}
if (i != num_pixels) {
VP8LPredictorsSub_C[13](in + i, upper + i, num_pixels - i, out + i);
}
}
//------------------------------------------------------------------------------
// Entry point

2
src/dsp/lossless_enc_sse41.c
View File

@ -32,8 +32,10 @@ static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) {
_mm_storeu_si128((__m128i*)&argb_data[i], out);
}
// fallthrough and finish off with plain-C
if (i != num_pixels) {
VP8LSubtractGreenFromBlueAndRed_C(argb_data + i, num_pixels - i);
}
}
//------------------------------------------------------------------------------
// Entry point

28
src/dsp/lossless_sse2.c
View File

@ -186,8 +186,10 @@ static void PredictorAdd0_SSE2(const uint32_t* in, const uint32_t* upper,
const __m128i res = _mm_add_epi8(src, black);
_mm_storeu_si128((__m128i*)&out[i], res);
}
if (i != num_pixels) {
VP8LPredictorsAdd_C[0](in + i, upper + i, num_pixels - i, out + i);
}
}
// Predictor1: left.
static void PredictorAdd1_SSE2(const uint32_t* in, const uint32_t* upper,
@ -210,8 +212,10 @@ static void PredictorAdd1_SSE2(const uint32_t* in, const uint32_t* upper,
// replicate prev output on the four lanes
prev = _mm_shuffle_epi32(res, (3 << 0) | (3 << 2) | (3 << 4) | (3 << 6));
}
if (i != num_pixels) {
VP8LPredictorsAdd_C[1](in + i, upper + i, num_pixels - i, out + i);
}
}
// Macro that adds 32-bit integers from IN using mod 256 arithmetic
// per 8 bit channel.
@ -225,7 +229,9 @@ static void PredictorAdd##X##_SSE2(const uint32_t* in, const uint32_t* upper, \
const __m128i res = _mm_add_epi8(src, other); \
_mm_storeu_si128((__m128i*)&out[i], res); \
} \
if (i != num_pixels) { \
VP8LPredictorsAdd_C[(X)](in + i, upper + i, num_pixels - i, out + i); \
} \
}
// Predictor2: Top.
@ -255,7 +261,9 @@ static void PredictorAdd##X##_SSE2(const uint32_t* in, const uint32_t* upper, \
res = _mm_add_epi8(avg, src); \
_mm_storeu_si128((__m128i*)&out[i], res); \
} \
if (i != num_pixels) { \
VP8LPredictorsAdd_C[(X)](in + i, upper + i, num_pixels - i, out + i); \
} \
}
// Predictor8: average TL T.
GENERATE_PREDICTOR_2(8, upper[i - 1])
@ -287,8 +295,10 @@ static void PredictorAdd10_SSE2(const uint32_t* in, const uint32_t* upper,
src = _mm_srli_si128(src, 4);
}
}
if (i != num_pixels) {
VP8LPredictorsAdd_C[10](in + i, upper + i, num_pixels - i, out + i);
}
}
// Predictor11: select.
static void GetSumAbsDiff32(const __m128i* const A, const __m128i* const B,
@ -331,8 +341,10 @@ static void PredictorAdd11_SSE2(const uint32_t* in, const uint32_t* upper,
pa = _mm_srli_si128(pa, 4);
}
}
if (i != num_pixels) {
VP8LPredictorsAdd_C[11](in + i, upper + i, num_pixels - i, out + i);
}
}
// Predictor12: ClampedAddSubtractFull.
#define DO_PRED12(DIFF, LANE, OUT) \
@ -369,8 +381,10 @@ static void PredictorAdd12_SSE2(const uint32_t* in, const uint32_t* upper,
DO_PRED12(diff_hi, 0, 2);
DO_PRED12(diff_hi, 1, 3);
}
if (i != num_pixels) {
VP8LPredictorsAdd_C[12](in + i, upper + i, num_pixels - i, out + i);
}
}
#undef DO_PRED12
// Due to averages with integers, values cannot be accumulated in parallel for
@ -392,8 +406,10 @@ static void AddGreenToBlueAndRed(const uint32_t* const src, int num_pixels,
_mm_storeu_si128((__m128i*)&dst[i], out);
}
// fallthrough and finish off with plain-C
if (i != num_pixels) {
VP8LAddGreenToBlueAndRed_C(src + i, num_pixels - i, dst + i);
}
}
//------------------------------------------------------------------------------
// Color Transform
@ -430,8 +446,10 @@ static void TransformColorInverse(const VP8LMultipliers* const m,
_mm_storeu_si128((__m128i*)&dst[i], out);
}
// Fall-back to C-version for left-overs.
if (i != num_pixels) {
VP8LTransformColorInverse_C(m, src + i, num_pixels - i, dst + i);
}
}
//------------------------------------------------------------------------------
// Color-space conversion functions
@ -467,8 +485,10 @@ static void ConvertBGRAToRGB(const uint32_t* src, int num_pixels,
num_pixels -= 32;
}
// left-overs
if (num_pixels > 0) {
VP8LConvertBGRAToRGB_C((const uint32_t*)in, num_pixels, (uint8_t*)out);
}
}
static void ConvertBGRAToRGBA(const uint32_t* src,
int num_pixels, uint8_t* dst) {
@ -494,8 +514,10 @@ static void ConvertBGRAToRGBA(const uint32_t* src,
num_pixels -= 8;
}
// left-overs
if (num_pixels > 0) {
VP8LConvertBGRAToRGBA_C((const uint32_t*)in, num_pixels, (uint8_t*)out);
}
}
static void ConvertBGRAToRGBA4444(const uint32_t* src,
int num_pixels, uint8_t* dst) {
@ -528,8 +550,10 @@ static void ConvertBGRAToRGBA4444(const uint32_t* src,
num_pixels -= 8;
}
// left-overs
if (num_pixels > 0) {
VP8LConvertBGRAToRGBA4444_C((const uint32_t*)in, num_pixels, (uint8_t*)out);
}
}
static void ConvertBGRAToRGB565(const uint32_t* src,
int num_pixels, uint8_t* dst) {
@ -567,8 +591,10 @@ static void ConvertBGRAToRGB565(const uint32_t* src,
num_pixels -= 8;
}
// left-overs
if (num_pixels > 0) {
VP8LConvertBGRAToRGB565_C((const uint32_t*)in, num_pixels, (uint8_t*)out);
}
}
static void ConvertBGRAToBGR(const uint32_t* src,
int num_pixels, uint8_t* dst) {
@ -598,8 +624,10 @@ static void ConvertBGRAToBGR(const uint32_t* src,
num_pixels -= 8;
}
// left-overs
if (num_pixels > 0) {
VP8LConvertBGRAToBGR_C((const uint32_t*)in, num_pixels, dst);
}
}
//------------------------------------------------------------------------------
// Entry point