har0ke/FFmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

huffyuvdsp: move functions only used by huffyuv from lossless_videodsp

Browse Source 
Signed-off-by: James Almer <jamrial@gmail.com>
This commit is contained in:
James Almer 2017-01-08 11:48:05 -03:00
parent cf9ef83960
commit 47f212329e
14 changed files with 208 additions and 204 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
libavcodec/huffyuvdec.c
View File

@ -297,8 +297,8 @@ static av_cold int decode_init(AVCodecContext *avctx)
if (ret < 0) if (ret < 0)
return ret; return ret;
ff_huffyuvdsp_init(&s->hdsp); ff_huffyuvdsp_init(&s->hdsp, avctx);
ff_llviddsp_init(&s->llviddsp, avctx); ff_llviddsp_init(&s->llviddsp);
memset(s->vlc, 0, 4 * sizeof(VLC)); memset(s->vlc, 0, 4 * sizeof(VLC));
s->interlaced = avctx->height > 288; s->interlaced = avctx->height > 288;
@ -891,7 +891,7 @@ static void add_bytes(HYuvContext *s, uint8_t *dst, uint8_t *src, int w)
if (s->bps <= 8) { if (s->bps <= 8) {
s->llviddsp.add_bytes(dst, src, w); s->llviddsp.add_bytes(dst, src, w);
} else { } else {
s->llviddsp.add_int16((uint16_t*)dst, (const uint16_t*)src, s->n - 1, w); s->hdsp.add_int16((uint16_t*)dst, (const uint16_t*)src, s->n - 1, w);
} }
} }
@ -900,7 +900,7 @@ static void add_median_prediction(HYuvContext *s, uint8_t *dst, const uint8_t *s
if (s->bps <= 8) { if (s->bps <= 8) {
s->llviddsp.add_median_pred(dst, src, diff, w, left, left_top); s->llviddsp.add_median_pred(dst, src, diff, w, left, left_top);
} else { } else {
s->llviddsp.add_hfyu_median_pred_int16((uint16_t *)dst, (const uint16_t *)src, (const uint16_t *)diff, s->n-1, w, left, left_top); s->hdsp.add_hfyu_median_pred_int16((uint16_t *)dst, (const uint16_t *)src, (const uint16_t *)diff, s->n-1, w, left, left_top);
} }
} }
static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame, static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame,

36
libavcodec/huffyuvdsp.c
View File

@ -23,6 +23,36 @@
#include "mathops.h" #include "mathops.h"
#include "huffyuvdsp.h" #include "huffyuvdsp.h"
static void add_int16_c(uint16_t *dst, const uint16_t *src, unsigned mask, int w){
long i;
unsigned long pw_lsb = (mask >> 1) * 0x0001000100010001ULL;
unsigned long pw_msb = pw_lsb + 0x0001000100010001ULL;
for (i = 0; i <= w - (int)sizeof(long)/2; i += sizeof(long)/2) {
long a = *(long*)(src+i);
long b = *(long*)(dst+i);
*(long*)(dst+i) = ((a&pw_lsb) + (b&pw_lsb)) ^ ((a^b)&pw_msb);
}
for(; i<w; i++)
dst[i] = (dst[i] + src[i]) & mask;
}
static void add_hfyu_median_pred_int16_c(uint16_t *dst, const uint16_t *src, const uint16_t *diff, unsigned mask, int w, int *left, int *left_top){
int i;
uint16_t l, lt;
l = *left;
lt = *left_top;
for(i=0; i<w; i++){
l = (mid_pred(l, src[i], (l + src[i] - lt) & mask) + diff[i]) & mask;
lt = src[i];
dst[i] = l;
}
*left = l;
*left_top = lt;
}
static void add_hfyu_left_pred_bgr32_c(uint8_t *dst, const uint8_t *src, static void add_hfyu_left_pred_bgr32_c(uint8_t *dst, const uint8_t *src,
intptr_t w, uint8_t *left) intptr_t w, uint8_t *left)
{ {
@ -47,10 +77,12 @@ static void add_hfyu_left_pred_bgr32_c(uint8_t *dst, const uint8_t *src,
left[A] = a; left[A] = a;
} }
av_cold void ff_huffyuvdsp_init(HuffYUVDSPContext *c) av_cold void ff_huffyuvdsp_init(HuffYUVDSPContext *c, AVCodecContext *avctx)
{ {
c->add_int16 = add_int16_c;
c->add_hfyu_median_pred_int16 = add_hfyu_median_pred_int16_c;
c->add_hfyu_left_pred_bgr32 = add_hfyu_left_pred_bgr32_c; c->add_hfyu_left_pred_bgr32 = add_hfyu_left_pred_bgr32_c;
if (ARCH_X86) if (ARCH_X86)
ff_huffyuvdsp_init_x86(c); ff_huffyuvdsp_init_x86(c, avctx);
} }

13
libavcodec/huffyuvdsp.h
View File

@ -21,6 +21,7 @@
#include <stdint.h> #include <stdint.h>
#include "config.h" #include "config.h"
#include "avcodec.h"
#if HAVE_BIGENDIAN #if HAVE_BIGENDIAN
#define B 3 #define B 3
@ -35,12 +36,18 @@
#endif #endif
typedef struct HuffYUVDSPContext { typedef struct HuffYUVDSPContext {
void (*add_int16)(uint16_t *dst/*align 16*/, const uint16_t *src/*align 16*/,
unsigned mask, int w);
void (*add_hfyu_median_pred_int16)(uint16_t *dst, const uint16_t *top,
const uint16_t *diff, unsigned mask,
int w, int *left, int *left_top);
void (*add_hfyu_left_pred_bgr32)(uint8_t *dst, const uint8_t *src, void (*add_hfyu_left_pred_bgr32)(uint8_t *dst, const uint8_t *src,
intptr_t w, uint8_t *left); intptr_t w, uint8_t *left);
} HuffYUVDSPContext; } HuffYUVDSPContext;
void ff_huffyuvdsp_init(HuffYUVDSPContext *c); void ff_huffyuvdsp_init(HuffYUVDSPContext *c, AVCodecContext *avctx);
void ff_huffyuvdsp_init_ppc(HuffYUVDSPContext *c); void ff_huffyuvdsp_init_ppc(HuffYUVDSPContext *c, AVCodecContext *avctx);
void ff_huffyuvdsp_init_x86(HuffYUVDSPContext *c); void ff_huffyuvdsp_init_x86(HuffYUVDSPContext *c, AVCodecContext *avctx);
#endif /* AVCODEC_HUFFYUVDSP_H */ #endif /* AVCODEC_HUFFYUVDSP_H */

2
libavcodec/lagarith.c
View File

@ -725,7 +725,7 @@ static av_cold int lag_decode_init(AVCodecContext *avctx)
LagarithContext *l = avctx->priv_data; LagarithContext *l = avctx->priv_data;
l->avctx = avctx; l->avctx = avctx;
ff_llviddsp_init(&l->llviddsp, avctx); ff_llviddsp_init(&l->llviddsp);
return 0; return 0;
} }

36
libavcodec/lossless_videodsp.c
View File

@ -79,36 +79,6 @@ static int add_left_pred_c(uint8_t *dst, const uint8_t *src, intptr_t w,
return acc; return acc;
} }
static void add_int16_c(uint16_t *dst, const uint16_t *src, unsigned mask, int w){
long i;
unsigned long pw_lsb = (mask >> 1) * 0x0001000100010001ULL;
unsigned long pw_msb = pw_lsb + 0x0001000100010001ULL;
for (i = 0; i <= w - (int)sizeof(long)/2; i += sizeof(long)/2) {
long a = *(long*)(src+i);
long b = *(long*)(dst+i);
*(long*)(dst+i) = ((a&pw_lsb) + (b&pw_lsb)) ^ ((a^b)&pw_msb);
}
for(; i<w; i++)
dst[i] = (dst[i] + src[i]) & mask;
}
static void add_hfyu_median_pred_int16_c(uint16_t *dst, const uint16_t *src, const uint16_t *diff, unsigned mask, int w, int *left, int *left_top){
int i;
uint16_t l, lt;
l = *left;
lt = *left_top;
for(i=0; i<w; i++){
l = (mid_pred(l, src[i], (l + src[i] - lt) & mask) + diff[i]) & mask;
lt = src[i];
dst[i] = l;
}
*left = l;
*left_top = lt;
}
static int add_hfyu_left_pred_int16_c(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc){ static int add_hfyu_left_pred_int16_c(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc){
int i; int i;
@ -129,16 +99,14 @@ static int add_hfyu_left_pred_int16_c(uint16_t *dst, const uint16_t *src, unsign
} }
void ff_llviddsp_init(LLVidDSPContext *c, AVCodecContext *avctx) void ff_llviddsp_init(LLVidDSPContext *c)
{ {
c->add_bytes = add_bytes_c; c->add_bytes = add_bytes_c;
c->add_median_pred = add_median_pred_c; c->add_median_pred = add_median_pred_c;
c->add_left_pred = add_left_pred_c; c->add_left_pred = add_left_pred_c;
c->add_int16 = add_int16_c;
c->add_hfyu_left_pred_int16 = add_hfyu_left_pred_int16_c; c->add_hfyu_left_pred_int16 = add_hfyu_left_pred_int16_c;
c->add_hfyu_median_pred_int16 = add_hfyu_median_pred_int16_c;
if (ARCH_X86) if (ARCH_X86)
ff_llviddsp_init_x86(c, avctx); ff_llviddsp_init_x86(c);
} }

9
libavcodec/lossless_videodsp.h
View File

@ -34,14 +34,11 @@ typedef struct LLVidDSPContext {
int (*add_left_pred)(uint8_t *dst, const uint8_t *src, int (*add_left_pred)(uint8_t *dst, const uint8_t *src,
intptr_t w, int left); intptr_t w, int left);
void (*add_int16)(uint16_t *dst/*align 16*/, const uint16_t *src/*align 16*/, unsigned mask, int w);
void (*add_hfyu_median_pred_int16)(uint16_t *dst, const uint16_t *top, const uint16_t *diff, unsigned mask, int w, int *left, int *left_top);
int (*add_hfyu_left_pred_int16)(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned left); int (*add_hfyu_left_pred_int16)(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned left);
} LLVidDSPContext; } LLVidDSPContext;
void ff_llviddsp_init(LLVidDSPContext *llviddsp, AVCodecContext *avctx); void ff_llviddsp_init(LLVidDSPContext *llviddsp);
void ff_llviddsp_init_x86(LLVidDSPContext *llviddsp, AVCodecContext *avctx); void ff_llviddsp_init_x86(LLVidDSPContext *llviddsp);
void ff_llviddsp_init_ppc(LLVidDSPContext *llviddsp, AVCodecContext *avctx); void ff_llviddsp_init_ppc(LLVidDSPContext *llviddsp);
#endif //AVCODEC_LOSSLESS_VIDEODSP_H #endif //AVCODEC_LOSSLESS_VIDEODSP_H

2
libavcodec/magicyuv.c
View File

@ -697,7 +697,7 @@ static int magy_init_thread_copy(AVCodecContext *avctx)
static av_cold int magy_decode_init(AVCodecContext *avctx) static av_cold int magy_decode_init(AVCodecContext *avctx)
{ {
MagicYUVContext *s = avctx->priv_data; MagicYUVContext *s = avctx->priv_data;
ff_llviddsp_init(&s->llviddsp, avctx); ff_llviddsp_init(&s->llviddsp);
return 0; return 0;
} }

2
libavcodec/ppc/lossless_videodsp_altivec.c
View File

@ -51,7 +51,7 @@ static void add_bytes_altivec(uint8_t *dst, uint8_t *src, intptr_t w)
} }
#endif /* HAVE_ALTIVEC */ #endif /* HAVE_ALTIVEC */
av_cold void ff_llviddsp_init_ppc(LLVidDSPContext *c, AVCodecContext *avctx) av_cold void ff_llviddsp_init_ppc(LLVidDSPContext *c)
{ {
#if HAVE_ALTIVEC #if HAVE_ALTIVEC
if (!PPC_ALTIVEC(av_get_cpu_flags())) if (!PPC_ALTIVEC(av_get_cpu_flags()))

2
libavcodec/utvideodec.c
View File

@ -827,7 +827,7 @@ static av_cold int decode_init(AVCodecContext *avctx)
c->avctx = avctx; c->avctx = avctx;
ff_bswapdsp_init(&c->bdsp); ff_bswapdsp_init(&c->bdsp);
ff_llviddsp_init(&c->llviddsp, avctx); ff_llviddsp_init(&c->llviddsp);
if (avctx->extradata_size >= 16) { if (avctx->extradata_size >= 16) {
av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n", av_log(avctx, AV_LOG_DEBUG, "Encoder version %d.%d.%d.%d\n",

2
libavcodec/vble.c
View File

@ -185,7 +185,7 @@ static av_cold int vble_decode_init(AVCodecContext *avctx)
/* Stash for later use */ /* Stash for later use */
ctx->avctx = avctx; ctx->avctx = avctx;
ff_llviddsp_init(&ctx->llviddsp, avctx); ff_llviddsp_init(&ctx->llviddsp);
avctx->pix_fmt = AV_PIX_FMT_YUV420P; avctx->pix_fmt = AV_PIX_FMT_YUV420P;
avctx->bits_per_raw_sample = 8; avctx->bits_per_raw_sample = 8;

137
libavcodec/x86/huffyuvdsp.asm
View File

@ -24,6 +24,78 @@
SECTION .text SECTION .text
%macro INT16_LOOP 2 ; %1 = a/u (aligned/unaligned), %2 = add/sub
movd m4, maskd
SPLATW m4, m4
add wd, wd
test wq, 2*mmsize - 1
jz %%.tomainloop
push tmpq
%%.wordloop:
sub wq, 2
%ifidn %2, add
mov tmpw, [srcq+wq]
add tmpw, [dstq+wq]
%else
mov tmpw, [src1q+wq]
sub tmpw, [src2q+wq]
%endif
and tmpw, maskw
mov [dstq+wq], tmpw
test wq, 2*mmsize - 1
jnz %%.wordloop
pop tmpq
%%.tomainloop:
%ifidn %2, add
add srcq, wq
%else
add src1q, wq
add src2q, wq
%endif
add dstq, wq
neg wq
jz %%.end
%%.loop:
%ifidn %2, add
mov%1 m0, [srcq+wq]
mov%1 m1, [dstq+wq]
mov%1 m2, [srcq+wq+mmsize]
mov%1 m3, [dstq+wq+mmsize]
%else
mov%1 m0, [src1q+wq]
mov%1 m1, [src2q+wq]
mov%1 m2, [src1q+wq+mmsize]
mov%1 m3, [src2q+wq+mmsize]
%endif
p%2w m0, m1
p%2w m2, m3
pand m0, m4
pand m2, m4
mov%1 [dstq+wq] , m0
mov%1 [dstq+wq+mmsize], m2
add wq, 2*mmsize
jl %%.loop
%%.end:
RET
%endmacro
%if ARCH_X86_32
INIT_MMX mmx
cglobal add_int16, 4,4,5, dst, src, mask, w, tmp
INT16_LOOP a, add
%endif
INIT_XMM sse2
cglobal add_int16, 4,4,5, dst, src, mask, w, tmp
test srcq, mmsize-1
jnz .unaligned
test dstq, mmsize-1
jnz .unaligned
INT16_LOOP a, add
.unaligned:
INT16_LOOP u, add
; void add_hfyu_left_pred_bgr32(uint8_t *dst, const uint8_t *src, ; void add_hfyu_left_pred_bgr32(uint8_t *dst, const uint8_t *src,
; intptr_t w, uint8_t *left) ; intptr_t w, uint8_t *left)
%macro LEFT_BGR32 0 %macro LEFT_BGR32 0
@ -63,3 +135,68 @@ LEFT_BGR32
%endif %endif
INIT_XMM sse2 INIT_XMM sse2
LEFT_BGR32 LEFT_BGR32
; void add_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int mask, int w, int *left, int *left_top)
INIT_MMX mmxext
cglobal add_hfyu_median_pred_int16, 7,7,0, dst, top, diff, mask, w, left, left_top
add wd, wd
movd mm6, maskd
SPLATW mm6, mm6
movq mm0, [topq]
movq mm2, mm0
movd mm4, [left_topq]
psllq mm2, 16
movq mm1, mm0
por mm4, mm2
movd mm3, [leftq]
psubw mm0, mm4 ; t-tl
add dstq, wq
add topq, wq
add diffq, wq
neg wq
jmp .skip
.loop:
movq mm4, [topq+wq]
movq mm0, mm4
psllq mm4, 16
por mm4, mm1
movq mm1, mm0 ; t
psubw mm0, mm4 ; t-tl
.skip:
movq mm2, [diffq+wq]
%assign i 0
%rep 4
movq mm4, mm0
paddw mm4, mm3 ; t-tl+l
pand mm4, mm6
movq mm5, mm3
pmaxsw mm3, mm1
pminsw mm5, mm1
pminsw mm3, mm4
pmaxsw mm3, mm5 ; median
paddw mm3, mm2 ; +residual
pand mm3, mm6
%if i==0
movq mm7, mm3
psllq mm7, 48
%else
movq mm4, mm3
psrlq mm7, 16
psllq mm4, 48
por mm7, mm4
%endif
%if i<3
psrlq mm0, 16
psrlq mm1, 16
psrlq mm2, 16
%endif
%assign i i+1
%endrep
movq [dstq+wq], mm7
add wq, 8
jl .loop
movzx r2d, word [dstq-2]
mov [leftq], r2d
movzx r2d, word [topq-2]
mov [left_topq], r2d
RET

13
libavcodec/x86/huffyuvdsp_init.c
View File

@ -21,24 +21,35 @@
#include "config.h" #include "config.h"
#include "libavutil/attributes.h" #include "libavutil/attributes.h"
#include "libavutil/cpu.h" #include "libavutil/cpu.h"
#include "libavutil/pixdesc.h"
#include "libavutil/x86/asm.h" #include "libavutil/x86/asm.h"
#include "libavutil/x86/cpu.h" #include "libavutil/x86/cpu.h"
#include "libavcodec/huffyuvdsp.h" #include "libavcodec/huffyuvdsp.h"
void ff_add_int16_mmx(uint16_t *dst, const uint16_t *src, unsigned mask, int w);
void ff_add_int16_sse2(uint16_t *dst, const uint16_t *src, unsigned mask, int w);
void ff_add_hfyu_left_pred_bgr32_mmx(uint8_t *dst, const uint8_t *src, void ff_add_hfyu_left_pred_bgr32_mmx(uint8_t *dst, const uint8_t *src,
intptr_t w, uint8_t *left); intptr_t w, uint8_t *left);
void ff_add_hfyu_left_pred_bgr32_sse2(uint8_t *dst, const uint8_t *src, void ff_add_hfyu_left_pred_bgr32_sse2(uint8_t *dst, const uint8_t *src,
intptr_t w, uint8_t *left); intptr_t w, uint8_t *left);
void ff_add_hfyu_median_pred_int16_mmxext(uint16_t *dst, const uint16_t *top, const uint16_t *diff, unsigned mask, int w, int *left, int *left_top);
av_cold void ff_huffyuvdsp_init_x86(HuffYUVDSPContext *c) av_cold void ff_huffyuvdsp_init_x86(HuffYUVDSPContext *c, AVCodecContext *avctx)
{ {
int cpu_flags = av_get_cpu_flags(); int cpu_flags = av_get_cpu_flags();
const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(avctx->pix_fmt);
if (ARCH_X86_32 && EXTERNAL_MMX(cpu_flags)) { if (ARCH_X86_32 && EXTERNAL_MMX(cpu_flags)) {
c->add_hfyu_left_pred_bgr32 = ff_add_hfyu_left_pred_bgr32_mmx; c->add_hfyu_left_pred_bgr32 = ff_add_hfyu_left_pred_bgr32_mmx;
c->add_int16 = ff_add_int16_mmx;
}
if (EXTERNAL_MMXEXT(cpu_flags) && pix_desc && pix_desc->comp[0].depth<16) {
c->add_hfyu_median_pred_int16 = ff_add_hfyu_median_pred_int16_mmxext;
} }
if (EXTERNAL_SSE2(cpu_flags)) { if (EXTERNAL_SSE2(cpu_flags)) {
c->add_int16 = ff_add_int16_sse2;
c->add_hfyu_left_pred_bgr32 = ff_add_hfyu_left_pred_bgr32_sse2; c->add_hfyu_left_pred_bgr32 = ff_add_hfyu_left_pred_bgr32_sse2;
} }
} }

136
libavcodec/x86/lossless_videodsp.asm
View File

@ -217,77 +217,6 @@ ADD_BYTES
INIT_XMM sse2 INIT_XMM sse2
ADD_BYTES ADD_BYTES
%macro INT16_LOOP 2 ; %1 = a/u (aligned/unaligned), %2 = add/sub
movd m4, maskd
SPLATW m4, m4
add wd, wd
test wq, 2*mmsize - 1
jz %%.tomainloop
push tmpq
%%.wordloop:
sub wq, 2
%ifidn %2, add
mov tmpw, [srcq+wq]
add tmpw, [dstq+wq]
%else
mov tmpw, [src1q+wq]
sub tmpw, [src2q+wq]
%endif
and tmpw, maskw
mov [dstq+wq], tmpw
test wq, 2*mmsize - 1
jnz %%.wordloop
pop tmpq
%%.tomainloop:
%ifidn %2, add
add srcq, wq
%else
add src1q, wq
add src2q, wq
%endif
add dstq, wq
neg wq
jz %%.end
%%.loop:
%ifidn %2, add
mov%1 m0, [srcq+wq]
mov%1 m1, [dstq+wq]
mov%1 m2, [srcq+wq+mmsize]
mov%1 m3, [dstq+wq+mmsize]
%else
mov%1 m0, [src1q+wq]
mov%1 m1, [src2q+wq]
mov%1 m2, [src1q+wq+mmsize]
mov%1 m3, [src2q+wq+mmsize]
%endif
p%2w m0, m1
p%2w m2, m3
pand m0, m4
pand m2, m4
mov%1 [dstq+wq] , m0
mov%1 [dstq+wq+mmsize], m2
add wq, 2*mmsize
jl %%.loop
%%.end:
RET
%endmacro
%if ARCH_X86_32
INIT_MMX mmx
cglobal add_int16, 4,4,5, dst, src, mask, w, tmp
INT16_LOOP a, add
%endif
INIT_XMM sse2
cglobal add_int16, 4,4,5, dst, src, mask, w, tmp
test srcq, mmsize-1
jnz .unaligned
test dstq, mmsize-1
jnz .unaligned
INT16_LOOP a, add
.unaligned:
INT16_LOOP u, add
%macro ADD_HFYU_LEFT_LOOP_INT16 2 ; %1 = dst alignment (a/u), %2 = src alignment (a/u) %macro ADD_HFYU_LEFT_LOOP_INT16 2 ; %1 = dst alignment (a/u), %2 = src alignment (a/u)
add wd, wd add wd, wd
add srcq, wq add srcq, wq
@ -359,68 +288,3 @@ cglobal add_hfyu_left_pred_int16, 4,4,8, dst, src, mask, w, left
ADD_HFYU_LEFT_LOOP_INT16 u, a ADD_HFYU_LEFT_LOOP_INT16 u, a
.src_unaligned: .src_unaligned:
ADD_HFYU_LEFT_LOOP_INT16 u, u ADD_HFYU_LEFT_LOOP_INT16 u, u
; void add_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int mask, int w, int *left, int *left_top)
INIT_MMX mmxext
cglobal add_hfyu_median_pred_int16, 7,7,0, dst, top, diff, mask, w, left, left_top
add wd, wd
movd mm6, maskd
SPLATW mm6, mm6
movq mm0, [topq]
movq mm2, mm0
movd mm4, [left_topq]
psllq mm2, 16
movq mm1, mm0
por mm4, mm2
movd mm3, [leftq]
psubw mm0, mm4 ; t-tl
add dstq, wq
add topq, wq
add diffq, wq
neg wq
jmp .skip
.loop:
movq mm4, [topq+wq]
movq mm0, mm4
psllq mm4, 16
por mm4, mm1
movq mm1, mm0 ; t
psubw mm0, mm4 ; t-tl
.skip:
movq mm2, [diffq+wq]
%assign i 0
%rep 4
movq mm4, mm0
paddw mm4, mm3 ; t-tl+l
pand mm4, mm6
movq mm5, mm3
pmaxsw mm3, mm1
pminsw mm5, mm1
pminsw mm3, mm4
pmaxsw mm3, mm5 ; median
paddw mm3, mm2 ; +residual
pand mm3, mm6
%if i==0
movq mm7, mm3
psllq mm7, 48
%else
movq mm4, mm3
psrlq mm7, 16
psllq mm4, 48
por mm7, mm4
%endif
%if i<3
psrlq mm0, 16
psrlq mm1, 16
psrlq mm2, 16
%endif
%assign i i+1
%endrep
movq [dstq+wq], mm7
add wq, 8
jl .loop
movzx r2d, word [dstq-2]
mov [leftq], r2d
movzx r2d, word [topq-2]
mov [left_topq], r2d
RET

14
libavcodec/x86/lossless_videodsp_init.c
View File

@ -21,7 +21,6 @@
#include "config.h" #include "config.h"
#include "libavutil/x86/asm.h" #include "libavutil/x86/asm.h"
#include "../lossless_videodsp.h" #include "../lossless_videodsp.h"
#include "libavutil/pixdesc.h"
#include "libavutil/x86/cpu.h" #include "libavutil/x86/cpu.h"
void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, intptr_t w); void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, intptr_t w);
@ -39,11 +38,8 @@ int ff_add_left_pred_ssse3(uint8_t *dst, const uint8_t *src,
int ff_add_left_pred_sse4(uint8_t *dst, const uint8_t *src, int ff_add_left_pred_sse4(uint8_t *dst, const uint8_t *src,
intptr_t w, int left); intptr_t w, int left);
void ff_add_int16_mmx(uint16_t *dst, const uint16_t *src, unsigned mask, int w);
void ff_add_int16_sse2(uint16_t *dst, const uint16_t *src, unsigned mask, int w);
int ff_add_hfyu_left_pred_int16_ssse3(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc); int ff_add_hfyu_left_pred_int16_ssse3(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc);
int ff_add_hfyu_left_pred_int16_sse4(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc); int ff_add_hfyu_left_pred_int16_sse4(uint16_t *dst, const uint16_t *src, unsigned mask, int w, unsigned acc);
void ff_add_hfyu_median_pred_int16_mmxext(uint16_t *dst, const uint16_t *top, const uint16_t *diff, unsigned mask, int w, int *left, int *left_top);
#if HAVE_INLINE_ASM && HAVE_7REGS && ARCH_X86_32 #if HAVE_INLINE_ASM && HAVE_7REGS && ARCH_X86_32
static void add_median_pred_cmov(uint8_t *dst, const uint8_t *top, static void add_median_pred_cmov(uint8_t *dst, const uint8_t *top,
@ -83,10 +79,9 @@ static void add_median_pred_cmov(uint8_t *dst, const uint8_t *top,
} }
#endif #endif
void ff_llviddsp_init_x86(LLVidDSPContext *c, AVCodecContext *avctx) void ff_llviddsp_init_x86(LLVidDSPContext *c)
{ {
int cpu_flags = av_get_cpu_flags(); int cpu_flags = av_get_cpu_flags();
const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(avctx->pix_fmt);
#if HAVE_INLINE_ASM && HAVE_7REGS && ARCH_X86_32 #if HAVE_INLINE_ASM && HAVE_7REGS && ARCH_X86_32
if (cpu_flags & AV_CPU_FLAG_CMOV) if (cpu_flags & AV_CPU_FLAG_CMOV)
@ -95,7 +90,6 @@ void ff_llviddsp_init_x86(LLVidDSPContext *c, AVCodecContext *avctx)
if (ARCH_X86_32 && EXTERNAL_MMX(cpu_flags)) { if (ARCH_X86_32 && EXTERNAL_MMX(cpu_flags)) {
c->add_bytes = ff_add_bytes_mmx; c->add_bytes = ff_add_bytes_mmx;
c->add_int16 = ff_add_int16_mmx;
} }
if (ARCH_X86_32 && EXTERNAL_MMXEXT(cpu_flags)) { if (ARCH_X86_32 && EXTERNAL_MMXEXT(cpu_flags)) {
@ -104,15 +98,9 @@ void ff_llviddsp_init_x86(LLVidDSPContext *c, AVCodecContext *avctx)
c->add_median_pred = ff_add_median_pred_mmxext; c->add_median_pred = ff_add_median_pred_mmxext;
} }
if (EXTERNAL_MMXEXT(cpu_flags) && pix_desc && pix_desc->comp[0].depth<16) {
c->add_hfyu_median_pred_int16 = ff_add_hfyu_median_pred_int16_mmxext;
}
if (EXTERNAL_SSE2(cpu_flags)) { if (EXTERNAL_SSE2(cpu_flags)) {
c->add_bytes = ff_add_bytes_sse2; c->add_bytes = ff_add_bytes_sse2;
c->add_median_pred = ff_add_median_pred_sse2; c->add_median_pred = ff_add_median_pred_sse2;
c->add_int16 = ff_add_int16_sse2;
} }
if (EXTERNAL_SSSE3(cpu_flags)) { if (EXTERNAL_SSSE3(cpu_flags)) {