dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2024-12-27 06:08:21 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

dsp/enc_sse2: add chroma intra predictors

Browse Source 
VP8EncPredChroma8 improvements over ~20M pixels
left/top: ~67%
left-only: ~52%
top-only: ~57%
none: ~61%

based on dec_sse2 versions with minor changes to benefit from the linear
storage of the left boundary

Change-Id: Iee7e387fb2570b4eb5af5bfd123e9c2e9ea49c76
This commit is contained in:
James Zern 2015-04-09 20:24:57 -07:00
parent aee021bbb1
commit 040b11bdf6
1 changed files with 101 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

105
src/dsp/enc_sse2.c
View File

@ -490,6 +490,15 @@ static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
//------------------------------------------------------------------------------
// Intra predictions
// helper for chroma-DC predictions
static WEBP_INLINE void Put8x8uv(uint8_t v, uint8_t* dst) {
int j;
const __m128i values = _mm_set1_epi8(v);
for (j = 0; j < 8; ++j) {
_mm_storel_epi64((__m128i*)(dst + j * BPS), values);
}
}
static WEBP_INLINE void Put16(uint8_t v, uint8_t* dst) {
int j;
const __m128i values = _mm_set1_epi8(v);
@ -502,12 +511,20 @@ static WEBP_INLINE void Fill(uint8_t* dst, int value, int size) {
if (size == 4) {
// TODO
} else if (size == 8) {
// TODO
Put8x8uv(value, dst);
} else {
Put16(value, dst);
}
}
static WEBP_INLINE void VE8uv(uint8_t* dst, const uint8_t* top) {
int j;
const __m128i top_values = _mm_loadl_epi64((const __m128i*)top);
for (j = 0; j < 8; ++j) {
_mm_storel_epi64((__m128i*)(dst + j * BPS), top_values);
}
}
static WEBP_INLINE void VE16(uint8_t* dst, const uint8_t* top) {
const __m128i top_values = _mm_load_si128((const __m128i*)top);
int j;
@ -520,7 +537,7 @@ static WEBP_INLINE void VerticalPred(uint8_t* dst,
const uint8_t* top, int size) {
if (top != NULL) {
if (size == 8) {
// TODO
VE8uv(dst, top);
} else {
VE16(dst, top);
}
@ -529,6 +546,15 @@ static WEBP_INLINE void VerticalPred(uint8_t* dst,
}
}
static WEBP_INLINE void HE8uv(uint8_t* dst, const uint8_t* left) {
int j;
for (j = 0; j < 8; ++j) {
const __m128i values = _mm_set1_epi8(left[j]);
_mm_storel_epi64((__m128i*)dst, values);
dst += BPS;
}
}
static WEBP_INLINE void HE16(uint8_t* dst, const uint8_t* left) {
int j;
for (j = 0; j < 16; ++j) {
@ -542,7 +568,7 @@ static WEBP_INLINE void HorizontalPred(uint8_t* dst,
const uint8_t* left, int size) {
if (left != NULL) {
if (size == 8) {
// TODO
HE8uv(dst, left);
} else {
HE16(dst, left);
}
@ -556,7 +582,14 @@ static WEBP_INLINE void TM(uint8_t* dst, const uint8_t* left,
const __m128i zero = _mm_setzero_si128();
int y;
if (size == 8) {
// TODO
const __m128i top_values = _mm_loadl_epi64((const __m128i*)top);
const __m128i top_base = _mm_unpacklo_epi8(top_values, zero);
for (y = 0; y < 8; ++y, dst += BPS) {
const int val = left[y] - left[-1];
const __m128i base = _mm_set1_epi16(val);
const __m128i out = _mm_packus_epi16(_mm_add_epi16(base, top_base), zero);
_mm_storel_epi64((__m128i*)dst, out);
}
} else {
const __m128i top_values = _mm_load_si128((const __m128i*)top);
const __m128i top_base_0 = _mm_unpacklo_epi8(top_values, zero);
@ -593,6 +626,49 @@ static WEBP_INLINE void TrueMotion(uint8_t* dst, const uint8_t* left,
}
}
static WEBP_INLINE void DC8uv(uint8_t* dst, const uint8_t* left,
const uint8_t* top) {
const __m128i zero = _mm_setzero_si128();
const __m128i top_values = _mm_loadl_epi64((const __m128i*)top);
const __m128i left_values = _mm_loadl_epi64((const __m128i*)left);
const __m128i sum_top = _mm_sad_epu8(top_values, zero);
const __m128i sum_left = _mm_sad_epu8(left_values, zero);
const int DC = _mm_cvtsi128_si32(sum_top) + _mm_cvtsi128_si32(sum_left) + 8;
Put8x8uv(DC >> 4, dst);
}
static WEBP_INLINE void DC8uvNoLeft(uint8_t* dst, const uint8_t* top) {
const __m128i zero = _mm_setzero_si128();
const __m128i top_values = _mm_loadl_epi64((const __m128i*)top);
const __m128i sum = _mm_sad_epu8(top_values, zero);
const int DC = _mm_cvtsi128_si32(sum) + 4;
Put8x8uv(DC >> 3, dst);
}
static WEBP_INLINE void DC8uvNoTop(uint8_t* dst, const uint8_t* left) {
// 'left' is contiguous so we can reuse the top summation.
DC8uvNoLeft(dst, left);
}
static WEBP_INLINE void DC8uvNoTopLeft(uint8_t* dst) {
Put8x8uv(0x80, dst);
}
static WEBP_INLINE void DC8uvMode(uint8_t* dst, const uint8_t* left,
const uint8_t* top) {
if (top != NULL) {
if (left != NULL) { // top and left present
DC8uv(dst, left, top);
} else { // top, but no left
DC8uvNoLeft(dst, top);
}
} else if (left != NULL) { // left but no top
DC8uvNoTop(dst, left);
} else { // no top, no left, nothing.
DC8uvNoTopLeft(dst);
}
}
static WEBP_INLINE void DC16(uint8_t* dst, const uint8_t* left,
const uint8_t* top) {
const __m128i zero = _mm_setzero_si128();
@ -643,6 +719,26 @@ static WEBP_INLINE void DC16Mode(uint8_t* dst, const uint8_t* left,
}
}
//------------------------------------------------------------------------------
// Chroma 8x8 prediction (paragraph 12.2)
static void IntraChromaPreds(uint8_t* dst, const uint8_t* left,
const uint8_t* top) {
// U block
DC8uvMode(C8DC8 + dst, left, top);
VerticalPred(C8VE8 + dst, top, 8);
HorizontalPred(C8HE8 + dst, left, 8);
TrueMotion(C8TM8 + dst, left, top, 8);
// V block
dst += 8;
if (top != NULL) top += 8;
if (left != NULL) left += 16;
DC8uvMode(C8DC8 + dst, left, top);
VerticalPred(C8VE8 + dst, top, 8);
HorizontalPred(C8HE8 + dst, left, 8);
TrueMotion(C8TM8 + dst, left, top, 8);
}
//------------------------------------------------------------------------------
// luma 16x16 prediction (paragraph 12.3)
@ -1089,6 +1185,7 @@ extern void VP8EncDspInitSSE2(void);
WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspInitSSE2(void) {
VP8CollectHistogram = CollectHistogram;
VP8EncPredLuma16 = Intra16Preds;
VP8EncPredChroma8 = IntraChromaPreds;
VP8EncQuantizeBlock = QuantizeBlock;
VP8EncQuantize2Blocks = Quantize2Blocks;
VP8EncQuantizeBlockWHT = QuantizeBlockWHT;