dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2024-12-28 14:38:21 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge changes I6eac17e5,I32d2b514

Browse Source 
* changes:
  dec_neon: add TM8uv
  dsp: initialize VP8PredChroma8 in VP8DspInit()
This commit is contained in:
James Zern 2014-11-25 15:32:14 -08:00 committed by Gerrit Code Review
commit f399d30764
3 changed files with 54 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
src/dsp/dec.c
View File

@ -463,10 +463,7 @@ const VP8PredFunc VP8PredLuma16[NUM_B_DC_MODES] = {
DC16NoTop, DC16NoLeft, DC16NoTopLeft DC16NoTop, DC16NoLeft, DC16NoTopLeft
}; };
const VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES] = { VP8PredFunc VP8PredChroma8[NUM_B_DC_MODES];
DC8uv, TM8uv, VE8uv, HE8uv,
DC8uvNoTop, DC8uvNoLeft, DC8uvNoTopLeft
};
//------------------------------------------------------------------------------ //------------------------------------------------------------------------------
// Edge filtering functions // Edge filtering functions
@ -721,6 +718,14 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8DspInit(void) {
VP8PredLuma4[8] = HD4; VP8PredLuma4[8] = HD4;
VP8PredLuma4[9] = HU4; VP8PredLuma4[9] = HU4;
VP8PredChroma8[0] = DC8uv;
VP8PredChroma8[1] = TM8uv;
VP8PredChroma8[2] = VE8uv;
VP8PredChroma8[3] = HE8uv;
VP8PredChroma8[4] = DC8uvNoTop;
VP8PredChroma8[5] = DC8uvNoLeft;
VP8PredChroma8[6] = DC8uvNoTopLeft;
// If defined, use CPUInfo() to overwrite some pointers with faster versions. // If defined, use CPUInfo() to overwrite some pointers with faster versions.
if (VP8GetCPUInfo != NULL) { if (VP8GetCPUInfo != NULL) {
#if defined(WEBP_USE_SSE2) #if defined(WEBP_USE_SSE2)

26
src/dsp/dec_neon.c
View File

@ -1281,11 +1281,15 @@ static void DC4(uint8_t* dst) { // DC
} }
} }
static void TM4(uint8_t* dst) { // TrueMotion // TrueMotion (4x4 + 8x8)
static WEBP_INLINE void TrueMotion(uint8_t* dst, int size) {
const uint8x8_t TL = vdup_n_u8(dst[-BPS - 1]); // top-left pixel 'A[-1]' const uint8x8_t TL = vdup_n_u8(dst[-BPS - 1]); // top-left pixel 'A[-1]'
const uint8x8_t T = vld1_u8(dst - BPS); // top row 'A[0..3]' const uint8x8_t T = vld1_u8(dst - BPS); // top row 'A[0..3]'
const int16x8_t d = vreinterpretq_s16_u16(vsubl_u8(T, TL)); // A[c] - A[-1] const int16x8_t d = vreinterpretq_s16_u16(vsubl_u8(T, TL)); // A[c] - A[-1]
const int16x8_t l0 = ConvertU8ToS16(vld1_u8(dst + 0 * BPS - 1)); // left edge int y;
for (y = 0; y < size; y += 4) {
// left edge
const int16x8_t l0 = ConvertU8ToS16(vld1_u8(dst + 0 * BPS - 1));
const int16x8_t l1 = ConvertU8ToS16(vld1_u8(dst + 1 * BPS - 1)); const int16x8_t l1 = ConvertU8ToS16(vld1_u8(dst + 1 * BPS - 1));
const int16x8_t l2 = ConvertU8ToS16(vld1_u8(dst + 2 * BPS - 1)); const int16x8_t l2 = ConvertU8ToS16(vld1_u8(dst + 2 * BPS - 1));
const int16x8_t l3 = ConvertU8ToS16(vld1_u8(dst + 3 * BPS - 1)); const int16x8_t l3 = ConvertU8ToS16(vld1_u8(dst + 3 * BPS - 1));
@ -1302,12 +1306,23 @@ static void TM4(uint8_t* dst) { // TrueMotion
const uint32x2_t r1_u32 = vreinterpret_u32_u8(vqmovun_s16(r1)); const uint32x2_t r1_u32 = vreinterpret_u32_u8(vqmovun_s16(r1));
const uint32x2_t r2_u32 = vreinterpret_u32_u8(vqmovun_s16(r2)); const uint32x2_t r2_u32 = vreinterpret_u32_u8(vqmovun_s16(r2));
const uint32x2_t r3_u32 = vreinterpret_u32_u8(vqmovun_s16(r3)); const uint32x2_t r3_u32 = vreinterpret_u32_u8(vqmovun_s16(r3));
if (size == 4) {
vst1_lane_u32((uint32_t*)(dst + 0 * BPS), r0_u32, 0); vst1_lane_u32((uint32_t*)(dst + 0 * BPS), r0_u32, 0);
vst1_lane_u32((uint32_t*)(dst + 1 * BPS), r1_u32, 0); vst1_lane_u32((uint32_t*)(dst + 1 * BPS), r1_u32, 0);
vst1_lane_u32((uint32_t*)(dst + 2 * BPS), r2_u32, 0); vst1_lane_u32((uint32_t*)(dst + 2 * BPS), r2_u32, 0);
vst1_lane_u32((uint32_t*)(dst + 3 * BPS), r3_u32, 0); vst1_lane_u32((uint32_t*)(dst + 3 * BPS), r3_u32, 0);
} else {
vst1_u32((uint32_t*)(dst + 0 * BPS), r0_u32);
vst1_u32((uint32_t*)(dst + 1 * BPS), r1_u32);
vst1_u32((uint32_t*)(dst + 2 * BPS), r2_u32);
vst1_u32((uint32_t*)(dst + 3 * BPS), r3_u32);
}
dst += 4 * BPS;
}
} }
static void TM4(uint8_t* dst) { return TrueMotion(dst, 4); }
static void VE4(uint8_t* dst) { // vertical static void VE4(uint8_t* dst) { // vertical
// NB: avoid vld1_u64 here as an alignment hint may be added -> SIGBUS. // NB: avoid vld1_u64 here as an alignment hint may be added -> SIGBUS.
const uint64x1_t A0 = vreinterpret_u64_u8(vld1_u8(dst - BPS - 1)); // top row const uint64x1_t A0 = vreinterpret_u64_u8(vld1_u8(dst - BPS - 1)); // top row
@ -1371,6 +1386,11 @@ static void LD4(uint8_t* dst) { // Down-left
vst1_lane_u32((uint32_t*)(dst + 3 * BPS), r3, 0); vst1_lane_u32((uint32_t*)(dst + 3 * BPS), r3, 0);
} }
//------------------------------------------------------------------------------
// Chroma
static void TM8uv(uint8_t* dst) { return TrueMotion(dst, 8); }
#endif // WEBP_USE_NEON #endif // WEBP_USE_NEON
//------------------------------------------------------------------------------ //------------------------------------------------------------------------------
@ -1407,5 +1427,7 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8DspInitNEON(void) {
VP8PredLuma4[2] = VE4; VP8PredLuma4[2] = VE4;
VP8PredLuma4[4] = RD4; VP8PredLuma4[4] = RD4;
VP8PredLuma4[6] = LD4; VP8PredLuma4[6] = LD4;
VP8PredChroma8[1] = TM8uv;
#endif // WEBP_USE_NEON #endif // WEBP_USE_NEON
} }

2
src/dsp/dsp.h
View File

@ -176,7 +176,7 @@ extern VP8WHT VP8TransformWHT;
// assumed accessible when needed. // assumed accessible when needed.
typedef void (*VP8PredFunc)(uint8_t* dst); typedef void (*VP8PredFunc)(uint8_t* dst);
extern const VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */]; extern const VP8PredFunc VP8PredLuma16[/* NUM_B_DC_MODES */];
extern const VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */]; extern VP8PredFunc VP8PredChroma8[/* NUM_B_DC_MODES */];
extern VP8PredFunc VP8PredLuma4[/* NUM_BMODES */]; extern VP8PredFunc VP8PredLuma4[/* NUM_BMODES */];
// clipping tables (for filtering) // clipping tables (for filtering)