virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-11-23 21:54:53 +02:00
Code Issues Releases Activity

avcodec/x86/mpegvideoenc: Port denoise_dct_sse2 to external assembly

Browse Source 
Reviewed-by: Lynne <dev@lynne.ee>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This commit is contained in:
Andreas Rheinhardt
2025-11-15 17:32:29 +01:00
parent 2cfef7031c
commit d633fa0433
2 changed files with 54 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files

59
libavcodec/x86/mpegvideoenc.c
View File

@@ -57,8 +57,10 @@ DECLARE_ALIGNED(16, static const uint16_t, inv_zigzag_direct16)[64] = {
#endif /* HAVE_6REGS */
#if HAVE_INLINE_ASM
#if HAVE_SSE2_INLINE
#if HAVE_SSE2_EXTERNAL
void ff_mpv_denoise_dct_sse2(int16_t block[64], int dct_error_sum[64],
const uint16_t dct_offset[64]);
static void denoise_dct_sse2(MPVEncContext *const s, int16_t block[])
{
const int intra = s->c.mb_intra;
@@ -67,56 +69,9 @@ static void denoise_dct_sse2(MPVEncContext *const s, int16_t block[])
s->dct_count[intra]++;
__asm__ volatile(
"pxor %%xmm6, %%xmm6 \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, %%xmm2 \n\t"
"movdqa %%xmm5, %%xmm0 \n\t"
"punpcklwd %%xmm6, %%xmm4 \n\t"
"punpckhwd %%xmm6, %%xmm2 \n\t"
"punpcklwd %%xmm6, %%xmm5 \n\t"
"punpckhwd %%xmm6, %%xmm0 \n\t"
"paddd (%1), %%xmm4 \n\t"
"paddd 16(%1), %%xmm2 \n\t"
"paddd 32(%1), %%xmm5 \n\t"
"paddd 48(%1), %%xmm0 \n\t"
"movdqa %%xmm4, (%1) \n\t"
"movdqa %%xmm2, 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")
);
ff_mpv_denoise_dct_sse2(block, sum, offset);
}
#endif /* HAVE_SSE2_INLINE */
#endif /* HAVE_INLINE_ASM */
#endif /* HAVE_SSE2_EXTERNAL */
av_cold void ff_dct_encode_init_x86(MPVEncContext *const s)
{
@@ -129,7 +84,9 @@ av_cold void ff_dct_encode_init_x86(MPVEncContext *const s)
#if HAVE_6REGS
s->dct_quantize = dct_quantize_sse2;
#endif
#if HAVE_SSE2_EXTERNAL
s->denoise_dct = denoise_dct_sse2;
#endif
}
#if HAVE_6REGS && HAVE_SSSE3_INLINE
if (INLINE_SSSE3(cpu_flags))

46
libavcodec/x86/mpegvideoencdsp.asm
View File

@@ -24,6 +24,52 @@
%include "libavutil/x86/x86util.asm"
SECTION .text
INIT_XMM sse2
cglobal mpv_denoise_dct, 3, 4, 7, block, sum, offset
pxor m6, m6
lea r3, [sumq+256]
.loop:
mova m2, [blockq]
mova m3, [blockq+16]
mova m0, m6
mova m1, m6
pcmpgtw m0, m2
pcmpgtw m1, m3
pxor m2, m0
pxor m3, m1
psubw m2, m0
psubw m3, m1
psubusw m4, m2, [offsetq]
psubusw m5, m3, [offsetq+16]
pxor m4, m0
pxor m5, m1
add offsetq, 32
psubw m4, m0
psubw m5, m1
mova [blockq], m4
mova [blockq+16], m5
mova m0, m2
mova m1, m3
add blockq, 32
punpcklwd m0, m6
punpckhwd m2, m6
punpcklwd m1, m6
punpckhwd m3, m6
paddd m0, [sumq]
paddd m2, [sumq+16]
paddd m1, [sumq+32]
paddd m3, [sumq+48]
mova [sumq], m0
mova [sumq+16], m2
mova [sumq+32], m1
mova [sumq+48], m3
add sumq, 64
cmp sumq, r3
jb .loop
RET
; int ff_pix_sum16(const uint8_t *pix, ptrdiff_t line_size)
; %1 = number of loops
; %2 = number of GPRs used