summaryrefslogtreecommitdiff
path: root/ffmpeg/libavcodec/x86/mpegvideo.c
diff options
context:
space:
mode:
Diffstat (limited to 'ffmpeg/libavcodec/x86/mpegvideo.c')
-rw-r--r--ffmpeg/libavcodec/x86/mpegvideo.c155
1 files changed, 9 insertions, 146 deletions
diff --git a/ffmpeg/libavcodec/x86/mpegvideo.c b/ffmpeg/libavcodec/x86/mpegvideo.c
index 903ad62..b2ce680 100644
--- a/ffmpeg/libavcodec/x86/mpegvideo.c
+++ b/ffmpeg/libavcodec/x86/mpegvideo.c
@@ -22,11 +22,12 @@
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
#include "libavutil/x86/asm.h"
+#include "libavutil/x86/cpu.h"
#include "libavcodec/avcodec.h"
#include "libavcodec/mpegvideo.h"
-#include "dsputil_mmx.h"
+#include "dsputil_x86.h"
-#if HAVE_INLINE_ASM
+#if HAVE_MMX_INLINE
static void dct_unquantize_h263_intra_mmx(MpegEncContext *s,
int16_t *block, int n, int qscale)
@@ -111,7 +112,7 @@ static void dct_unquantize_h263_inter_mmx(MpegEncContext *s,
qmul = qscale << 1;
qadd = (qscale - 1) | 1;
- assert(s->block_last_index[n]>=0 || s->h263_aic);
+ av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
@@ -164,28 +165,6 @@ __asm__ volatile(
);
}
-
-/*
- We can suppose that result of two multiplications can't be greater than 0xFFFF
- i.e. is 16-bit, so we use here only PMULLW instruction and can avoid
- a complex multiplication.
-=====================================================
- Full formula for multiplication of 2 integer numbers
- which are represent as high:low words:
- input: value1 = high1:low1
- value2 = high2:low2
- output: value3 = value1*value2
- value3=high3:low3 (on overflow: modulus 2^32 wrap-around)
- this mean that for 0x123456 * 0x123456 correct result is 0x766cb0ce4
- but this algorithm will compute only 0x66cb0ce4
- this limited by 16-bit size of operands
- ---------------------------------
- tlow1 = high1*low2
- tlow2 = high2*low1
- tlow1 = tlow1 + tlow2
- high3:low3 = low1*low2
- high3 += tlow1
-*/
static void dct_unquantize_mpeg1_intra_mmx(MpegEncContext *s,
int16_t *block, int n, int qscale)
{
@@ -464,124 +443,14 @@ __asm__ volatile(
);
}
-static void denoise_dct_mmx(MpegEncContext *s, int16_t *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"
- "add $16, %0 \n\t"
- "add $32, %1 \n\t"
- "add $16, %2 \n\t"
- "cmp %3, %0 \n\t"
- " jb 1b \n\t"
- : "+r" (block), "+r" (sum), "+r" (offset)
- : "r"(block+64)
- );
-}
-
-static void denoise_dct_sse2(MpegEncContext *s, int16_t *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"
- "add $32, %0 \n\t"
- "add $64, %1 \n\t"
- "add $32, %2 \n\t"
- "cmp %3, %0 \n\t"
- " jb 1b \n\t"
- : "+r" (block), "+r" (sum), "+r" (offset)
- : "r"(block+64)
- XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
- "%xmm4", "%xmm5", "%xmm6", "%xmm7")
- );
-}
-
-#endif /* HAVE_INLINE_ASM */
+#endif /* HAVE_MMX_INLINE */
av_cold void ff_MPV_common_init_x86(MpegEncContext *s)
{
-#if HAVE_INLINE_ASM
- int mm_flags = av_get_cpu_flags();
+#if HAVE_MMX_INLINE
+ int cpu_flags = av_get_cpu_flags();
- if (mm_flags & AV_CPU_FLAG_MMX) {
+ if (INLINE_MMX(cpu_flags)) {
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;
@@ -589,12 +458,6 @@ av_cold void ff_MPV_common_init_x86(MpegEncContext *s)
if(!(s->flags & CODEC_FLAG_BITEXACT))
s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_mmx;
s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_mmx;
-
- if (mm_flags & AV_CPU_FLAG_SSE2) {
- s->denoise_dct= denoise_dct_sse2;
- } else {
- s->denoise_dct= denoise_dct_mmx;
- }
}
-#endif /* HAVE_INLINE_ASM */
+#endif /* HAVE_MMX_INLINE */
}
generated by cgit v1.2.3 (git 2.39.1) at 2025-12-06 13:57:21 +0000