Skip to content

F
ffmpeg.wasm-core
Commit 9ba9c340 authored by Christophe GISQUET's avatar Christophe GISQUET Committed by Janne Grunau
Browse files
Options

rv34: 1-pass inter MB reconstruction 

Implement 1-pass inverse transform and reconstruction for inter blocks.
parent ffa0923e
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 130 additions and 240 deletions
libavcodec/arm/rv34dsp_init_neon.c
View file @ 9ba9c340
...@@ -23,16 +23,12 @@ ...@@ -23,16 +23,12 @@
#include "libavcodec/avcodec.h" #include "libavcodec/avcodec.h"
#include "libavcodec/rv34dsp.h" #include "libavcodec/rv34dsp.h"
void ff_rv34_inv_transform_neon(DCTELEM *block);
void ff_rv34_inv_transform_noround_neon(DCTELEM *block); void ff_rv34_inv_transform_noround_neon(DCTELEM *block);
void ff_rv34_inv_transform_dc_neon(DCTELEM *block);
void ff_rv34_inv_transform_noround_dc_neon(DCTELEM *block); void ff_rv34_inv_transform_noround_dc_neon(DCTELEM *block);
void ff_rv34dsp_init_neon(RV34DSPContext *c, DSPContext* dsp) void ff_rv34dsp_init_neon(RV34DSPContext *c, DSPContext* dsp)
{ {
c->rv34_inv_transform_tab[0] = ff_rv34_inv_transform_neon; c->rv34_inv_transform = ff_rv34_inv_transform_noround_neon;
c->rv34_inv_transform_tab[1] = ff_rv34_inv_transform_noround_neon; c->rv34_inv_transform_dc = ff_rv34_inv_transform_noround_dc_neon;
c->rv34_inv_transform_dc_tab[0] = ff_rv34_inv_transform_dc_neon;
c->rv34_inv_transform_dc_tab[1] = ff_rv34_inv_transform_noround_dc_neon;
} }
libavcodec/arm/rv34dsp_neon.S
View file @ 9ba9c340
...@@ -21,11 +21,7 @@ ...@@ -21,11 +21,7 @@
#include "asm.S" #include "asm.S"
.macro rv34_inv_transform .macro rv34_inv_transform
mov r1, #16 vld1.16 {q14-q15}, [r0,:128]
vld1.16 {d28}, [r0,:64], r1 @ block[i+8*0]
vld1.16 {d29}, [r0,:64], r1 @ block[i+8*1]
vld1.16 {d30}, [r0,:64], r1 @ block[i+8*2]
vld1.16 {d31}, [r0,:64], r1 @ block[i+8*3]
vmov.s16 d0, #13 vmov.s16 d0, #13
vshll.s16 q12, d29, #3 vshll.s16 q12, d29, #3
vshll.s16 q13, d29, #4 vshll.s16 q13, d29, #4
...@@ -35,12 +31,12 @@ ...@@ -35,12 +31,12 @@
vmlal.s16 q10, d30, d0 vmlal.s16 q10, d30, d0
vmull.s16 q11, d28, d0 vmull.s16 q11, d28, d0
vmlsl.s16 q11, d30, d0 vmlsl.s16 q11, d30, d0
vsubw.s16 q12, q12, d29 @ z2 = block[i+8*1]*7 vsubw.s16 q12, q12, d29 @ z2 = block[i+4*1]*7
vaddw.s16 q13, q13, d29 @ z3 = block[i+8*1]*17 vaddw.s16 q13, q13, d29 @ z3 = block[i+4*1]*17
vsubw.s16 q9, q9, d31 vsubw.s16 q9, q9, d31
vaddw.s16 q1, q1, d31 vaddw.s16 q1, q1, d31
vadd.s32 q13, q13, q9 @ z3 = 17*block[i+8*1] + 7*block[i+8*3] vadd.s32 q13, q13, q9 @ z3 = 17*block[i+4*1] + 7*block[i+4*3]
vsub.s32 q12, q12, q1 @ z2 = 7*block[i+8*1] - 17*block[i+8*3] vsub.s32 q12, q12, q1 @ z2 = 7*block[i+4*1] - 17*block[i+4*3]
vadd.s32 q1, q10, q13 @ z0 + z3 vadd.s32 q1, q10, q13 @ z0 + z3
vadd.s32 q2, q11, q12 @ z1 + z2 vadd.s32 q2, q11, q12 @ z1 + z2
vsub.s32 q8, q10, q13 @ z0 - z3 vsub.s32 q8, q10, q13 @ z0 - z3
...@@ -70,24 +66,8 @@ ...@@ -70,24 +66,8 @@
vsub.s32 q15, q14, q9 @ z0 - z3 vsub.s32 q15, q14, q9 @ z0 - z3
.endm .endm
/* void ff_rv34_inv_transform_neon(DCTELEM *block); */
function ff_rv34_inv_transform_neon, export=1
mov r2, r0
rv34_inv_transform
vrshrn.s32 d1, q2, #10 @ (z1 + z2) >> 10
vrshrn.s32 d0, q1, #10 @ (z0 + z3) >> 10
vrshrn.s32 d2, q3, #10 @ (z1 - z2) >> 10
vrshrn.s32 d3, q15, #10 @ (z0 - z3) >> 10
vst4.16 {d0[0], d1[0], d2[0], d3[0]}, [r2,:64], r1
vst4.16 {d0[1], d1[1], d2[1], d3[1]}, [r2,:64], r1
vst4.16 {d0[2], d1[2], d2[2], d3[2]}, [r2,:64], r1
vst4.16 {d0[3], d1[3], d2[3], d3[3]}, [r2,:64], r1
bx lr
endfunc
/* void rv34_inv_transform_noround_neon(DCTELEM *block); */ /* void rv34_inv_transform_noround_neon(DCTELEM *block); */
function ff_rv34_inv_transform_noround_neon, export=1 function ff_rv34_inv_transform_noround_neon, export=1
mov r2, r0
rv34_inv_transform rv34_inv_transform
vshl.s32 q11, q2, #1 vshl.s32 q11, q2, #1
vshl.s32 q10, q1, #1 vshl.s32 q10, q1, #1
...@@ -101,38 +81,23 @@ function ff_rv34_inv_transform_noround_neon, export=1 ...@@ -101,38 +81,23 @@ function ff_rv34_inv_transform_noround_neon, export=1
vshrn.s32 d1, q11, #11 @ (z1 + z2)*3 >> 11 vshrn.s32 d1, q11, #11 @ (z1 + z2)*3 >> 11
vshrn.s32 d2, q12, #11 @ (z1 - z2)*3 >> 11 vshrn.s32 d2, q12, #11 @ (z1 - z2)*3 >> 11
vshrn.s32 d3, q13, #11 @ (z0 - z3)*3 >> 11 vshrn.s32 d3, q13, #11 @ (z0 - z3)*3 >> 11
vst4.16 {d0[0], d1[0], d2[0], d3[0]}, [r2,:64], r1 vst4.16 {d0[0], d1[0], d2[0], d3[0]}, [r0,:64]!
vst4.16 {d0[1], d1[1], d2[1], d3[1]}, [r2,:64], r1 vst4.16 {d0[1], d1[1], d2[1], d3[1]}, [r0,:64]!
vst4.16 {d0[2], d1[2], d2[2], d3[2]}, [r2,:64], r1 vst4.16 {d0[2], d1[2], d2[2], d3[2]}, [r0,:64]!
vst4.16 {d0[3], d1[3], d2[3], d3[3]}, [r2,:64], r1 vst4.16 {d0[3], d1[3], d2[3], d3[3]}, [r0,:64]!
bx lr bx lr
endfunc endfunc
/* void rv34_inv_transform_dc_c(DCTELEM *block) */
function ff_rv34_inv_transform_dc_neon, export=1
vld1.16 {d28[]}, [r0,:16] @ block[0]
vmov.i16 d4, #169
mov r1, #16
vmull.s16 q3, d28, d4
vrshrn.s32 d0, q3, #10
vst1.16 {d0}, [r0,:64], r1
vst1.16 {d0}, [r0,:64], r1
vst1.16 {d0}, [r0,:64], r1
vst1.16 {d0}, [r0,:64], r1
bx lr
endfunc
/* void rv34_inv_transform_dc_noround_c(DCTELEM *block) */ /* void rv34_inv_transform_dc_noround_c(DCTELEM *block) */
function ff_rv34_inv_transform_noround_dc_neon, export=1 function ff_rv34_inv_transform_noround_dc_neon, export=1
vld1.16 {d28[]}, [r0,:16] @ block[0] vld1.16 {d28[]}, [r0,:16] @ block[0]
vmov.i16 d4, #251 vmov.i16 d4, #251
vorr.s16 d4, #256 @ 13^2 * 3 vorr.s16 d4, #256 @ 13^2 * 3
mov r1, #16
vmull.s16 q3, d28, d4 vmull.s16 q3, d28, d4
vshrn.s32 d0, q3, #11 vshrn.s32 d0, q3, #11
vst1.64 {d0}, [r0,:64], r1 vmov.i16 d1, d0
vst1.64 {d0}, [r0,:64], r1 vst1.64 {q0}, [r0,:128]!
vst1.64 {d0}, [r0,:64], r1 vst1.64 {q0}, [r0,:128]!
vst1.64 {d0}, [r0,:64], r1
bx lr bx lr
endfunc endfunc
libavcodec/rv34.c
View file @ 9ba9c340
This diff is collapsed.
libavcodec/rv34dsp.c
View file @ 9ba9c340
...@@ -32,15 +32,15 @@ ...@@ -32,15 +32,15 @@
* @{ * @{
*/ */
static av_always_inline void rv34_row_transform(int temp[16], const DCTELEM *block) static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
{ {
int i; int i;
for(i = 0; i < 4; i++){ for(i = 0; i < 4; i++){
const int z0 = 13*(block[i+8*0] + block[i+8*2]); const int z0 = 13*(block[i+4*0] + block[i+4*2]);
const int z1 = 13*(block[i+8*0] - block[i+8*2]); const int z1 = 13*(block[i+4*0] - block[i+4*2]);
const int z2 = 7* block[i+8*1] - 17*block[i+8*3]; const int z2 = 7* block[i+4*1] - 17*block[i+4*3];
const int z3 = 17* block[i+8*1] + 7*block[i+8*3]; const int z3 = 17* block[i+4*1] + 7*block[i+4*3];
temp[4*i+0] = z0 + z3; temp[4*i+0] = z0 + z3;
temp[4*i+1] = z1 + z2; temp[4*i+1] = z1 + z2;
...@@ -49,39 +49,17 @@ static av_always_inline void rv34_row_transform(int temp[16], const DCTELEM *blo ...@@ -49,39 +49,17 @@ static av_always_inline void rv34_row_transform(int temp[16], const DCTELEM *blo
} }
} }
/**
* Real Video 3.0/4.0 inverse transform
* Code is almost the same as in SVQ3, only scaling is different.
*/
static void rv34_inv_transform_c(DCTELEM *block){
int temp[16];
int i;
rv34_row_transform(temp, block);
for(i = 0; i < 4; i++){
const int z0 = 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
const int z1 = 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
const int z2 = 7* temp[4*1+i] - 17*temp[4*3+i];
const int z3 = 17* temp[4*1+i] + 7*temp[4*3+i];
block[i*8+0] = (z0 + z3) >> 10;
block[i*8+1] = (z1 + z2) >> 10;
block[i*8+2] = (z1 - z2) >> 10;
block[i*8+3] = (z0 - z3) >> 10;
}
}
/** /**
* Real Video 3.0/4.0 inverse transform + sample reconstruction * Real Video 3.0/4.0 inverse transform + sample reconstruction
* Code is almost the same as in SVQ3, only scaling is different. * Code is almost the same as in SVQ3, only scaling is different.
*/ */
static void rv34_idct_add_c(uint8_t *dst, int stride, const DCTELEM *block){ static void rv34_idct_add_c(uint8_t *dst, int stride, DCTELEM *block){
int temp[16]; int temp[16];
uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
int i; int i;
rv34_row_transform(temp, block); rv34_row_transform(temp, block);
memset(block, 0, 16*sizeof(DCTELEM));
for(i = 0; i < 4; i++){ for(i = 0; i < 4; i++){
const int z0 = 13*(temp[4*0+i] + temp[4*2+i]) + 0x200; const int z0 = 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
...@@ -116,10 +94,10 @@ static void rv34_inv_transform_noround_c(DCTELEM *block){ ...@@ -116,10 +94,10 @@ static void rv34_inv_transform_noround_c(DCTELEM *block){
const int z2 = 7* temp[4*1+i] - 17*temp[4*3+i]; const int z2 = 7* temp[4*1+i] - 17*temp[4*3+i];
const int z3 = 17* temp[4*1+i] + 7*temp[4*3+i]; const int z3 = 17* temp[4*1+i] + 7*temp[4*3+i];
block[i*8+0] = ((z0 + z3) * 3) >> 11; block[i*4+0] = ((z0 + z3) * 3) >> 11;
block[i*8+1] = ((z1 + z2) * 3) >> 11; block[i*4+1] = ((z1 + z2) * 3) >> 11;
block[i*8+2] = ((z1 - z2) * 3) >> 11; block[i*4+2] = ((z1 - z2) * 3) >> 11;
block[i*8+3] = ((z0 - z3) * 3) >> 11; block[i*4+3] = ((z0 - z3) * 3) >> 11;
} }
} }
...@@ -139,22 +117,12 @@ static void rv34_idct_dc_add_c(uint8_t *dst, int stride, int dc) ...@@ -139,22 +117,12 @@ static void rv34_idct_dc_add_c(uint8_t *dst, int stride, int dc)
} }
} }
static void rv34_inv_transform_dc_c(DCTELEM *block)
{
DCTELEM dc = (13 * 13 * block[0] + 0x200) >> 10;
int i, j;
for (i = 0; i < 4; i++, block += 8)
for (j = 0; j < 4; j++)
block[j] = dc;
}
static void rv34_inv_transform_dc_noround_c(DCTELEM *block) static void rv34_inv_transform_dc_noround_c(DCTELEM *block)
{ {
DCTELEM dc = (13 * 13 * 3 * block[0]) >> 11; DCTELEM dc = (13 * 13 * 3 * block[0]) >> 11;
int i, j; int i, j;
for (i = 0; i < 4; i++, block += 8) for (i = 0; i < 4; i++, block += 4)
for (j = 0; j < 4; j++) for (j = 0; j < 4; j++)
block[j] = dc; block[j] = dc;
} }
...@@ -163,10 +131,8 @@ static void rv34_inv_transform_dc_noround_c(DCTELEM *block) ...@@ -163,10 +131,8 @@ static void rv34_inv_transform_dc_noround_c(DCTELEM *block)
av_cold void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp) { av_cold void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp) {
c->rv34_inv_transform_tab[0] = rv34_inv_transform_c; c->rv34_inv_transform = rv34_inv_transform_noround_c;
c->rv34_inv_transform_tab[1] = rv34_inv_transform_noround_c; c->rv34_inv_transform_dc = rv34_inv_transform_dc_noround_c;
c->rv34_inv_transform_dc_tab[0] = rv34_inv_transform_dc_c;
c->rv34_inv_transform_dc_tab[1] = rv34_inv_transform_dc_noround_c;
c->rv34_idct_add = rv34_idct_add_c; c->rv34_idct_add = rv34_idct_add_c;
c->rv34_idct_dc_add = rv34_idct_dc_add_c; c->rv34_idct_dc_add = rv34_idct_dc_add_c;
......
libavcodec/rv34dsp.h
View file @ 9ba9c340
...@@ -36,8 +36,7 @@ typedef void (*rv40_weight_func)(uint8_t *dst/*align width (8 or 16)*/, ...@@ -36,8 +36,7 @@ typedef void (*rv40_weight_func)(uint8_t *dst/*align width (8 or 16)*/,
typedef void (*rv34_inv_transform_func)(DCTELEM *block); typedef void (*rv34_inv_transform_func)(DCTELEM *block);
typedef void (*rv34_idct_add_func)(uint8_t *dst, int stride, typedef void (*rv34_idct_add_func)(uint8_t *dst, int stride, DCTELEM *block);
const DCTELEM *block);
typedef void (*rv34_idct_dc_add_func)(uint8_t *dst, int stride, typedef void (*rv34_idct_dc_add_func)(uint8_t *dst, int stride,
int dc); int dc);
...@@ -60,8 +59,8 @@ typedef struct RV34DSPContext { ...@@ -60,8 +59,8 @@ typedef struct RV34DSPContext {
h264_chroma_mc_func put_chroma_pixels_tab[3]; h264_chroma_mc_func put_chroma_pixels_tab[3];
h264_chroma_mc_func avg_chroma_pixels_tab[3]; h264_chroma_mc_func avg_chroma_pixels_tab[3];
rv40_weight_func rv40_weight_pixels_tab[2]; rv40_weight_func rv40_weight_pixels_tab[2];
rv34_inv_transform_func rv34_inv_transform_tab[2]; rv34_inv_transform_func rv34_inv_transform;
void (*rv34_inv_transform_dc_tab[2])(DCTELEM *block); rv34_inv_transform_func rv34_inv_transform_dc;
rv34_idct_add_func rv34_idct_add; rv34_idct_add_func rv34_idct_add;
rv34_idct_dc_add_func rv34_idct_dc_add; rv34_idct_dc_add_func rv34_idct_dc_add;
rv40_weak_loop_filter_func rv40_weak_loop_filter[2]; rv40_weak_loop_filter_func rv40_weak_loop_filter[2];
......
libavcodec/x86/rv34dsp.asm
View file @ 9ba9c340
...@@ -42,9 +42,9 @@ cglobal rv34_idct_%1_mmx2, 1, 2, 0 ...@@ -42,9 +42,9 @@ cglobal rv34_idct_%1_mmx2, 1, 2, 0
movd m0, r1 movd m0, r1
pshufw m0, m0, 0 pshufw m0, m0, 0
movq [r0+ 0], m0 movq [r0+ 0], m0
movq [r0+ 8], m0
movq [r0+16], m0 movq [r0+16], m0
movq [r0+32], m0 movq [r0+24], m0
movq [r0+48], m0
REP_RET REP_RET
%endmacro %endmacro
......
libavcodec/x86/rv34dsp_init.c
View file @ 9ba9c340
...@@ -37,8 +37,7 @@ av_cold void ff_rv34dsp_init_x86(RV34DSPContext* c, DSPContext *dsp) ...@@ -37,8 +37,7 @@ av_cold void ff_rv34dsp_init_x86(RV34DSPContext* c, DSPContext *dsp)
if (mm_flags & AV_CPU_FLAG_MMX) if (mm_flags & AV_CPU_FLAG_MMX)
c->rv34_idct_dc_add = ff_rv34_idct_dc_add_mmx; c->rv34_idct_dc_add = ff_rv34_idct_dc_add_mmx;
if (mm_flags & AV_CPU_FLAG_MMX2) { if (mm_flags & AV_CPU_FLAG_MMX2) {
c->rv34_inv_transform_dc_tab[0] = ff_rv34_idct_dc_mmx2; c->rv34_inv_transform_dc = ff_rv34_idct_dc_noround_mmx2;
c->rv34_inv_transform_dc_tab[1] = ff_rv34_idct_dc_noround_mmx2;
} }
if (mm_flags & AV_CPU_FLAG_SSE4) if (mm_flags & AV_CPU_FLAG_SSE4)
c->rv34_idct_dc_add = ff_rv34_idct_dc_add_sse4; c->rv34_idct_dc_add = ff_rv34_idct_dc_add_sse4;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment