summaryrefslogtreecommitdiff
path: root/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c')
-rw-r--r--src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c297
1 files changed, 250 insertions, 47 deletions
diff --git a/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c b/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c
index d2f477b7b..1c0e9f5ae 100644
--- a/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c
+++ b/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c
@@ -31,31 +31,92 @@ static const unsigned long long int mm_wabs __attribute__ ((aligned(8))) = 0xfff
static const unsigned long long int mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL;
-static void dct_unquantize_h263_mmx(MpegEncContext *s,
+static void dct_unquantize_h263_intra_mmx(MpegEncContext *s,
DCTELEM *block, int n, int qscale)
{
int level, qmul, qadd, nCoeffs;
qmul = qscale << 1;
- qadd = (qscale - 1) | 1;
- assert(s->block_last_index[n]>=0);
+ assert(s->block_last_index[n]>=0 || s->h263_aic);
- if (s->mb_intra) {
- if (!s->h263_aic) {
- if (n < 4)
- level = block[0] * s->y_dc_scale;
- else
- level = block[0] * s->c_dc_scale;
- }else{
- qadd = 0;
- level= block[0];
- }
- nCoeffs=63;
- } else {
- nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
- level = 0;/* keep gcc quiet */
+ if (!s->h263_aic) {
+ if (n < 4)
+ level = block[0] * s->y_dc_scale;
+ else
+ level = block[0] * s->c_dc_scale;
+ qadd = (qscale - 1) | 1;
+ }else{
+ qadd = 0;
+ level= block[0];
}
+ if(s->ac_pred)
+ nCoeffs=63;
+ else
+ nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
+//printf("%d %d ", qmul, qadd);
+asm volatile(
+ "movd %1, %%mm6 \n\t" //qmul
+ "packssdw %%mm6, %%mm6 \n\t"
+ "packssdw %%mm6, %%mm6 \n\t"
+ "movd %2, %%mm5 \n\t" //qadd
+ "pxor %%mm7, %%mm7 \n\t"
+ "packssdw %%mm5, %%mm5 \n\t"
+ "packssdw %%mm5, %%mm5 \n\t"
+ "psubw %%mm5, %%mm7 \n\t"
+ "pxor %%mm4, %%mm4 \n\t"
+ ".balign 16\n\t"
+ "1: \n\t"
+ "movq (%0, %3), %%mm0 \n\t"
+ "movq 8(%0, %3), %%mm1 \n\t"
+
+ "pmullw %%mm6, %%mm0 \n\t"
+ "pmullw %%mm6, %%mm1 \n\t"
+
+ "movq (%0, %3), %%mm2 \n\t"
+ "movq 8(%0, %3), %%mm3 \n\t"
+
+ "pcmpgtw %%mm4, %%mm2 \n\t" // block[i] < 0 ? -1 : 0
+ "pcmpgtw %%mm4, %%mm3 \n\t" // block[i] < 0 ? -1 : 0
+
+ "pxor %%mm2, %%mm0 \n\t"
+ "pxor %%mm3, %%mm1 \n\t"
+
+ "paddw %%mm7, %%mm0 \n\t"
+ "paddw %%mm7, %%mm1 \n\t"
+
+ "pxor %%mm0, %%mm2 \n\t"
+ "pxor %%mm1, %%mm3 \n\t"
+
+ "pcmpeqw %%mm7, %%mm0 \n\t" // block[i] == 0 ? -1 : 0
+ "pcmpeqw %%mm7, %%mm1 \n\t" // block[i] == 0 ? -1 : 0
+
+ "pandn %%mm2, %%mm0 \n\t"
+ "pandn %%mm3, %%mm1 \n\t"
+
+ "movq %%mm0, (%0, %3) \n\t"
+ "movq %%mm1, 8(%0, %3) \n\t"
+
+ "addl $16, %3 \n\t"
+ "jng 1b \n\t"
+ ::"r" (block+nCoeffs), "g"(qmul), "g" (qadd), "r" (2*(-nCoeffs))
+ : "memory"
+ );
+ block[0]= level;
+}
+
+
+static void dct_unquantize_h263_inter_mmx(MpegEncContext *s,
+ DCTELEM *block, int n, int qscale)
+{
+ int level, qmul, qadd, nCoeffs;
+
+ qmul = qscale << 1;
+ qadd = (qscale - 1) | 1;
+
+ assert(s->block_last_index[n]>=0 || s->h263_aic);
+
+ nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
//printf("%d %d ", qmul, qadd);
asm volatile(
"movd %1, %%mm6 \n\t" //qmul
@@ -104,8 +165,6 @@ asm volatile(
::"r" (block+nCoeffs), "g"(qmul), "g" (qadd), "r" (2*(-nCoeffs))
: "memory"
);
- if(s->mb_intra)
- block[0]= level;
}
@@ -138,24 +197,23 @@ asm volatile(
high3:low3 = low1*low2
high3 += tlow1
*/
-static void dct_unquantize_mpeg1_mmx(MpegEncContext *s,
+static void dct_unquantize_mpeg1_intra_mmx(MpegEncContext *s,
DCTELEM *block, int n, int qscale)
{
int nCoeffs;
const uint16_t *quant_matrix;
+ int block0;
assert(s->block_last_index[n]>=0);
nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
- if (s->mb_intra) {
- int block0;
- if (n < 4)
- block0 = block[0] * s->y_dc_scale;
- else
- block0 = block[0] * s->c_dc_scale;
- /* XXX: only mpeg1 */
- quant_matrix = s->intra_matrix;
+ if (n < 4)
+ block0 = block[0] * s->y_dc_scale;
+ else
+ block0 = block[0] * s->c_dc_scale;
+ /* XXX: only mpeg1 */
+ quant_matrix = s->intra_matrix;
asm volatile(
"pcmpeqw %%mm7, %%mm7 \n\t"
"psrlw $15, %%mm7 \n\t"
@@ -205,9 +263,19 @@ asm volatile(
::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "g" (qscale), "g" (-2*nCoeffs)
: "%eax", "memory"
);
- block[0]= block0;
+ block[0]= block0;
+}
+
+static void dct_unquantize_mpeg1_inter_mmx(MpegEncContext *s,
+ DCTELEM *block, int n, int qscale)
+{
+ int nCoeffs;
+ const uint16_t *quant_matrix;
+
+ assert(s->block_last_index[n]>=0);
+
+ nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
- } else {
quant_matrix = s->inter_matrix;
asm volatile(
"pcmpeqw %%mm7, %%mm7 \n\t"
@@ -262,28 +330,25 @@ asm volatile(
::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "g" (qscale), "g" (-2*nCoeffs)
: "%eax", "memory"
);
- }
-
}
-static void dct_unquantize_mpeg2_mmx(MpegEncContext *s,
+static void dct_unquantize_mpeg2_intra_mmx(MpegEncContext *s,
DCTELEM *block, int n, int qscale)
{
int nCoeffs;
const uint16_t *quant_matrix;
+ int block0;
assert(s->block_last_index[n]>=0);
if(s->alternate_scan) nCoeffs= 63; //FIXME
else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
- if (s->mb_intra) {
- int block0;
- if (n < 4)
- block0 = block[0] * s->y_dc_scale;
- else
- block0 = block[0] * s->c_dc_scale;
- quant_matrix = s->intra_matrix;
+ if (n < 4)
+ block0 = block[0] * s->y_dc_scale;
+ else
+ block0 = block[0] * s->c_dc_scale;
+ quant_matrix = s->intra_matrix;
asm volatile(
"pcmpeqw %%mm7, %%mm7 \n\t"
"psrlw $15, %%mm7 \n\t"
@@ -329,10 +394,21 @@ asm volatile(
::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "g" (qscale), "g" (-2*nCoeffs)
: "%eax", "memory"
);
- block[0]= block0;
+ block[0]= block0;
//Note, we dont do mismatch control for intra as errors cannot accumulate
+}
+
+static void dct_unquantize_mpeg2_inter_mmx(MpegEncContext *s,
+ DCTELEM *block, int n, int qscale)
+{
+ int nCoeffs;
+ const uint16_t *quant_matrix;
+
+ assert(s->block_last_index[n]>=0);
+
+ if(s->alternate_scan) nCoeffs= 63; //FIXME
+ else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
- } else {
quant_matrix = s->inter_matrix;
asm volatile(
"pcmpeqw %%mm7, %%mm7 \n\t"
@@ -397,7 +473,6 @@ asm volatile(
::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "g" (qscale), "r" (-2*nCoeffs)
: "%eax", "memory"
);
- }
}
/* draw the edges of width 'w' of an image of size width, height
@@ -488,13 +563,130 @@ static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w)
}
}
+static void denoise_dct_mmx(MpegEncContext *s, DCTELEM *block){
+ const int intra= s->mb_intra;
+ int *sum= s->dct_error_sum[intra];
+ uint16_t *offset= s->dct_offset[intra];
+
+ s->dct_count[intra]++;
+
+ asm volatile(
+ "pxor %%mm7, %%mm7 \n\t"
+ "1: \n\t"
+ "pxor %%mm0, %%mm0 \n\t"
+ "pxor %%mm1, %%mm1 \n\t"
+ "movq (%0), %%mm2 \n\t"
+ "movq 8(%0), %%mm3 \n\t"
+ "pcmpgtw %%mm2, %%mm0 \n\t"
+ "pcmpgtw %%mm3, %%mm1 \n\t"
+ "pxor %%mm0, %%mm2 \n\t"
+ "pxor %%mm1, %%mm3 \n\t"
+ "psubw %%mm0, %%mm2 \n\t"
+ "psubw %%mm1, %%mm3 \n\t"
+ "movq %%mm2, %%mm4 \n\t"
+ "movq %%mm3, %%mm5 \n\t"
+ "psubusw (%2), %%mm2 \n\t"
+ "psubusw 8(%2), %%mm3 \n\t"
+ "pxor %%mm0, %%mm2 \n\t"
+ "pxor %%mm1, %%mm3 \n\t"
+ "psubw %%mm0, %%mm2 \n\t"
+ "psubw %%mm1, %%mm3 \n\t"
+ "movq %%mm2, (%0) \n\t"
+ "movq %%mm3, 8(%0) \n\t"
+ "movq %%mm4, %%mm2 \n\t"
+ "movq %%mm5, %%mm3 \n\t"
+ "punpcklwd %%mm7, %%mm4 \n\t"
+ "punpckhwd %%mm7, %%mm2 \n\t"
+ "punpcklwd %%mm7, %%mm5 \n\t"
+ "punpckhwd %%mm7, %%mm3 \n\t"
+ "paddd (%1), %%mm4 \n\t"
+ "paddd 8(%1), %%mm2 \n\t"
+ "paddd 16(%1), %%mm5 \n\t"
+ "paddd 24(%1), %%mm3 \n\t"
+ "movq %%mm4, (%1) \n\t"
+ "movq %%mm2, 8(%1) \n\t"
+ "movq %%mm5, 16(%1) \n\t"
+ "movq %%mm3, 24(%1) \n\t"
+ "addl $16, %0 \n\t"
+ "addl $32, %1 \n\t"
+ "addl $16, %2 \n\t"
+ "cmpl %3, %0 \n\t"
+ " jb 1b \n\t"
+ : "+r" (block), "+r" (sum), "+r" (offset)
+ : "r"(block+64)
+ );
+}
+
+static void denoise_dct_sse2(MpegEncContext *s, DCTELEM *block){
+ const int intra= s->mb_intra;
+ int *sum= s->dct_error_sum[intra];
+ uint16_t *offset= s->dct_offset[intra];
+
+ s->dct_count[intra]++;
+
+ asm volatile(
+ "pxor %%xmm7, %%xmm7 \n\t"
+ "1: \n\t"
+ "pxor %%xmm0, %%xmm0 \n\t"
+ "pxor %%xmm1, %%xmm1 \n\t"
+ "movdqa (%0), %%xmm2 \n\t"
+ "movdqa 16(%0), %%xmm3 \n\t"
+ "pcmpgtw %%xmm2, %%xmm0 \n\t"
+ "pcmpgtw %%xmm3, %%xmm1 \n\t"
+ "pxor %%xmm0, %%xmm2 \n\t"
+ "pxor %%xmm1, %%xmm3 \n\t"
+ "psubw %%xmm0, %%xmm2 \n\t"
+ "psubw %%xmm1, %%xmm3 \n\t"
+ "movdqa %%xmm2, %%xmm4 \n\t"
+ "movdqa %%xmm3, %%xmm5 \n\t"
+ "psubusw (%2), %%xmm2 \n\t"
+ "psubusw 16(%2), %%xmm3 \n\t"
+ "pxor %%xmm0, %%xmm2 \n\t"
+ "pxor %%xmm1, %%xmm3 \n\t"
+ "psubw %%xmm0, %%xmm2 \n\t"
+ "psubw %%xmm1, %%xmm3 \n\t"
+ "movdqa %%xmm2, (%0) \n\t"
+ "movdqa %%xmm3, 16(%0) \n\t"
+ "movdqa %%xmm4, %%xmm6 \n\t"
+ "movdqa %%xmm5, %%xmm0 \n\t"
+ "punpcklwd %%xmm7, %%xmm4 \n\t"
+ "punpckhwd %%xmm7, %%xmm6 \n\t"
+ "punpcklwd %%xmm7, %%xmm5 \n\t"
+ "punpckhwd %%xmm7, %%xmm0 \n\t"
+ "paddd (%1), %%xmm4 \n\t"
+ "paddd 16(%1), %%xmm6 \n\t"
+ "paddd 32(%1), %%xmm5 \n\t"
+ "paddd 48(%1), %%xmm0 \n\t"
+ "movdqa %%xmm4, (%1) \n\t"
+ "movdqa %%xmm6, 16(%1) \n\t"
+ "movdqa %%xmm5, 32(%1) \n\t"
+ "movdqa %%xmm0, 48(%1) \n\t"
+ "addl $32, %0 \n\t"
+ "addl $64, %1 \n\t"
+ "addl $32, %2 \n\t"
+ "cmpl %3, %0 \n\t"
+ " jb 1b \n\t"
+ : "+r" (block), "+r" (sum), "+r" (offset)
+ : "r"(block+64)
+ );
+}
+
#undef HAVE_MMX2
#define RENAME(a) a ## _MMX
+#define RENAMEl(a) a ## _mmx
#include "mpegvideo_mmx_template.c"
#define HAVE_MMX2
#undef RENAME
+#undef RENAMEl
#define RENAME(a) a ## _MMX2
+#define RENAMEl(a) a ## _mmx2
+#include "mpegvideo_mmx_template.c"
+
+#undef RENAME
+#undef RENAMEl
+#define RENAME(a) a ## _SSE2
+#define RENAMEl(a) a ## _sse2
#include "mpegvideo_mmx_template.c"
void MPV_common_init_mmx(MpegEncContext *s)
@@ -502,14 +694,25 @@ void MPV_common_init_mmx(MpegEncContext *s)
if (mm_flags & MM_MMX) {
const int dct_algo = s->avctx->dct_algo;
- s->dct_unquantize_h263 = dct_unquantize_h263_mmx;
- s->dct_unquantize_mpeg1 = dct_unquantize_mpeg1_mmx;
- s->dct_unquantize_mpeg2 = dct_unquantize_mpeg2_mmx;
+ s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_mmx;
+ s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_mmx;
+ s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_mmx;
+ s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_mmx;
+ s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_mmx;
+ s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_mmx;
draw_edges = draw_edges_mmx;
+
+ if (mm_flags & MM_SSE2) {
+ s->denoise_dct= denoise_dct_sse2;
+ } else {
+ s->denoise_dct= denoise_dct_mmx;
+ }
if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){
- if(mm_flags & MM_MMXEXT){
+ if(mm_flags & MM_SSE2){
+ s->dct_quantize= dct_quantize_SSE2;
+ } else if(mm_flags & MM_MMXEXT){
s->dct_quantize= dct_quantize_MMX2;
} else {
s->dct_quantize= dct_quantize_MMX;
generated by cgit v1.2.3 (git 2.25.1) at 2025-04-27 14:07:15 +0000