Skip to content

F
ffmpeg-mt
Commit 4fa72c3d authored by bcoudurier's avatar bcoudurier
Browse files
Options

convert fdct_mmx to plain asm 

git-svn-id: file:///var/local/repositories/ffmpeg/trunk@15819 9553f0bf-9b14-0410-a0b8-cfaf0461ba5b
parent b8434539
Hide whitespace changes
Inline Side-by-side
Showing with 170 additions and 159 deletions
libavcodec/i386/fdct_mmx.c
View file @ 4fa72c3d
...@@ -32,7 +32,6 @@ ...@@ -32,7 +32,6 @@
#include "libavutil/common.h" #include "libavutil/common.h"
#include "libavcodec/dsputil.h" #include "libavcodec/dsputil.h"
#include "mmx.h"
#define ATTR_ALIGN(align) __attribute__ ((__aligned__ (align))) #define ATTR_ALIGN(align) __attribute__ ((__aligned__ (align)))
...@@ -286,84 +285,90 @@ TABLE_SSE2 ...@@ -286,84 +285,90 @@ TABLE_SSE2
TABLE_SSE2 TABLE_SSE2
}}; }};
#define S(s) AV_TOSTRING(s) //AV_STRINGIFY is too long
#define FDCT_COL(cpu, mm, mov)\ #define FDCT_COL(cpu, mm, mov)\
static av_always_inline void fdct_col_##cpu(const int16_t *in, int16_t *out, int offset)\ static av_always_inline void fdct_col_##cpu(const int16_t *in, int16_t *out, int offset)\
{\ {\
mov##_m2r(*(in + offset + 1 * 8), mm##0);\ __asm__ volatile (\
mov##_m2r(*(in + offset + 6 * 8), mm##1);\ #mov" 16(%0), %%"#mm"0 \n\t" \
mov##_r2r(mm##0, mm##2);\ #mov" 96(%0), %%"#mm"1 \n\t" \
mov##_m2r(*(in + offset + 2 * 8), mm##3);\ #mov" %%"#mm"0, %%"#mm"2 \n\t" \
paddsw_r2r(mm##1, mm##0);\ #mov" 32(%0), %%"#mm"3 \n\t" \
mov##_m2r(*(in + offset + 5 * 8), mm##4);\ "paddsw %%"#mm"1, %%"#mm"0 \n\t" \
psllw_i2r(SHIFT_FRW_COL, mm##0);\ #mov" 80(%0), %%"#mm"4 \n\t" \
mov##_m2r(*(in + offset + 0 * 8), mm##5);\ "psllw $"S(SHIFT_FRW_COL)", %%"#mm"0 \n\t" \
paddsw_r2r(mm##3, mm##4);\ #mov" (%0), %%"#mm"5 \n\t" \
paddsw_m2r(*(in + offset + 7 * 8), mm##5);\ "paddsw %%"#mm"3, %%"#mm"4 \n\t" \
psllw_i2r(SHIFT_FRW_COL, mm##4);\ "paddsw 112(%0), %%"#mm"5 \n\t" \
mov##_r2r(mm##0, mm##6);\ "psllw $"S(SHIFT_FRW_COL)", %%"#mm"4 \n\t" \
psubsw_r2r(mm##1, mm##2);\ #mov" %%"#mm"0, %%"#mm"6 \n\t" \
mov##_m2r(*(fdct_tg_all_16 + 8), mm##1);\ "psubsw %%"#mm"1, %%"#mm"2 \n\t" \
psubsw_r2r(mm##4, mm##0);\ #mov" 16(%1), %%"#mm"1 \n\t" \
mov##_m2r(*(in + offset + 3 * 8), mm##7);\ "psubsw %%"#mm"4, %%"#mm"0 \n\t" \
pmulhw_r2r(mm##0, mm##1);\ #mov" 48(%0), %%"#mm"7 \n\t" \
paddsw_m2r(*(in + offset + 4 * 8), mm##7);\ "pmulhw %%"#mm"0, %%"#mm"1 \n\t" \
psllw_i2r(SHIFT_FRW_COL, mm##5);\ "paddsw 64(%0), %%"#mm"7 \n\t" \
paddsw_r2r(mm##4, mm##6);\ "psllw $"S(SHIFT_FRW_COL)", %%"#mm"5 \n\t" \
psllw_i2r(SHIFT_FRW_COL, mm##7);\ "paddsw %%"#mm"4, %%"#mm"6 \n\t" \
mov##_r2r(mm##5, mm##4);\ "psllw $"S(SHIFT_FRW_COL)", %%"#mm"7 \n\t" \
psubsw_r2r(mm##7, mm##5);\ #mov" %%"#mm"5, %%"#mm"4 \n\t" \
paddsw_r2r(mm##5, mm##1);\ "psubsw %%"#mm"7, %%"#mm"5 \n\t" \
paddsw_r2r(mm##7, mm##4);\ "paddsw %%"#mm"5, %%"#mm"1 \n\t" \
por_m2r(*fdct_one_corr, mm##1);\ "paddsw %%"#mm"7, %%"#mm"4 \n\t" \
psllw_i2r(SHIFT_FRW_COL + 1, mm##2);\ "por (%2), %%"#mm"1 \n\t" \
pmulhw_m2r(*(fdct_tg_all_16 + 8), mm##5);\ "psllw $"S(SHIFT_FRW_COL)"+1, %%"#mm"2 \n\t" \
mov##_r2r(mm##4, mm##7);\ "pmulhw 16(%1), %%"#mm"5 \n\t" \
psubsw_m2r(*(in + offset + 5 * 8), mm##3);\ #mov" %%"#mm"4, %%"#mm"7 \n\t" \
psubsw_r2r(mm##6, mm##4);\ "psubsw 80(%0), %%"#mm"3 \n\t" \
mov##_r2m(mm##1, *(out + offset + 2 * 8));\ "psubsw %%"#mm"6, %%"#mm"4 \n\t" \
paddsw_r2r(mm##6, mm##7);\ #mov" %%"#mm"1, 32(%3) \n\t" \
mov##_m2r(*(in + offset + 3 * 8), mm##1);\ "paddsw %%"#mm"6, %%"#mm"7 \n\t" \
psllw_i2r(SHIFT_FRW_COL + 1, mm##3);\ #mov" 48(%0), %%"#mm"1 \n\t" \
psubsw_m2r(*(in + offset + 4 * 8), mm##1);\ "psllw $"S(SHIFT_FRW_COL)"+1, %%"#mm"3 \n\t" \
mov##_r2r(mm##2, mm##6);\ "psubsw 64(%0), %%"#mm"1 \n\t" \
mov##_r2m(mm##4, *(out + offset + 4 * 8));\ #mov" %%"#mm"2, %%"#mm"6 \n\t" \
paddsw_r2r(mm##3, mm##2);\ #mov" %%"#mm"4, 64(%3) \n\t" \
pmulhw_m2r(*ocos_4_16, mm##2);\ "paddsw %%"#mm"3, %%"#mm"2 \n\t" \
psubsw_r2r(mm##3, mm##6);\ "pmulhw (%4), %%"#mm"2 \n\t" \
pmulhw_m2r(*ocos_4_16, mm##6);\ "psubsw %%"#mm"3, %%"#mm"6 \n\t" \
psubsw_r2r(mm##0, mm##5);\ "pmulhw (%4), %%"#mm"6 \n\t" \
por_m2r(*fdct_one_corr, mm##5);\ "psubsw %%"#mm"0, %%"#mm"5 \n\t" \
psllw_i2r(SHIFT_FRW_COL, mm##1);\ "por (%2), %%"#mm"5 \n\t" \
por_m2r(*fdct_one_corr, mm##2);\ "psllw $"S(SHIFT_FRW_COL)", %%"#mm"1 \n\t" \
mov##_r2r(mm##1, mm##4);\ "por (%2), %%"#mm"2 \n\t" \
mov##_m2r(*(in + offset + 0 * 8), mm##3);\ #mov" %%"#mm"1, %%"#mm"4 \n\t" \
paddsw_r2r(mm##6, mm##1);\ #mov" (%0), %%"#mm"3 \n\t" \
psubsw_m2r(*(in + offset + 7 * 8), mm##3);\ "paddsw %%"#mm"6, %%"#mm"1 \n\t" \
psubsw_r2r(mm##6, mm##4);\ "psubsw 112(%0), %%"#mm"3 \n\t" \
mov##_m2r(*(fdct_tg_all_16 + 0), mm##0);\ "psubsw %%"#mm"6, %%"#mm"4 \n\t" \
psllw_i2r(SHIFT_FRW_COL, mm##3);\ #mov" (%1), %%"#mm"0 \n\t" \
mov##_m2r(*(fdct_tg_all_16 + 16), mm##6);\ "psllw $"S(SHIFT_FRW_COL)", %%"#mm"3 \n\t" \
pmulhw_r2r(mm##1, mm##0);\ #mov" 32(%1), %%"#mm"6 \n\t" \
mov##_r2m(mm##7, *(out + offset + 0 * 8));\ "pmulhw %%"#mm"1, %%"#mm"0 \n\t" \
pmulhw_r2r(mm##4, mm##6);\ #mov" %%"#mm"7, (%3) \n\t" \
mov##_r2m(mm##5, *(out + offset + 6 * 8));\ "pmulhw %%"#mm"4, %%"#mm"6 \n\t" \
mov##_r2r(mm##3, mm##7);\ #mov" %%"#mm"5, 96(%3) \n\t" \
mov##_m2r(*(fdct_tg_all_16 + 16), mm##5);\ #mov" %%"#mm"3, %%"#mm"7 \n\t" \
psubsw_r2r(mm##2, mm##7);\ #mov" 32(%1), %%"#mm"5 \n\t" \
paddsw_r2r(mm##2, mm##3);\ "psubsw %%"#mm"2, %%"#mm"7 \n\t" \
pmulhw_r2r(mm##7, mm##5);\ "paddsw %%"#mm"2, %%"#mm"3 \n\t" \
paddsw_r2r(mm##3, mm##0);\ "pmulhw %%"#mm"7, %%"#mm"5 \n\t" \
paddsw_r2r(mm##4, mm##6);\ "paddsw %%"#mm"3, %%"#mm"0 \n\t" \
pmulhw_m2r(*(fdct_tg_all_16 + 0), mm##3);\ "paddsw %%"#mm"4, %%"#mm"6 \n\t" \
por_m2r(*fdct_one_corr, mm##0);\ "pmulhw (%1), %%"#mm"3 \n\t" \
paddsw_r2r(mm##7, mm##5);\ "por (%2), %%"#mm"0 \n\t" \
psubsw_r2r(mm##6, mm##7);\ "paddsw %%"#mm"7, %%"#mm"5 \n\t" \
mov##_r2m(mm##0, *(out + offset + 1 * 8));\ "psubsw %%"#mm"6, %%"#mm"7 \n\t" \
paddsw_r2r(mm##4, mm##5);\ #mov" %%"#mm"0, 16(%3) \n\t" \
mov##_r2m(mm##7, *(out + offset + 3 * 8));\ "paddsw %%"#mm"4, %%"#mm"5 \n\t" \
psubsw_r2r(mm##1, mm##3);\ #mov" %%"#mm"7, 48(%3) \n\t" \
mov##_r2m(mm##5, *(out + offset + 5 * 8));\ "psubsw %%"#mm"1, %%"#mm"3 \n\t" \
mov##_r2m(mm##3, *(out + offset + 7 * 8));\ #mov" %%"#mm"5, 80(%3) \n\t" \
#mov" %%"#mm"3, 112(%3) \n\t" \
: \
: "r" (in + offset), "r" (fdct_tg_all_16), "r" (fdct_one_corr), \
"r" (out + offset), "r" (ocos_4_16)); \
} }
FDCT_COL(mmx, mm, movq) FDCT_COL(mmx, mm, movq)
...@@ -433,93 +438,99 @@ static av_always_inline void fdct_row_sse2(const int16_t *in, int16_t *out) ...@@ -433,93 +438,99 @@ static av_always_inline void fdct_row_sse2(const int16_t *in, int16_t *out)
static av_always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const int16_t *table) static av_always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const int16_t *table)
{ {
pshufw_m2r(*(in + 4), mm5, 0x1B); __asm__ volatile (
movq_m2r(*(in + 0), mm0); "pshufw $0x1B, 8(%0), %%mm5 \n\t"
movq_r2r(mm0, mm1); "movq (%0), %%mm0 \n\t"
paddsw_r2r(mm5, mm0); "movq %%mm0, %%mm1 \n\t"
psubsw_r2r(mm5, mm1); "paddsw %%mm5, %%mm0 \n\t"
movq_r2r(mm0, mm2); "psubsw %%mm5, %%mm1 \n\t"
punpckldq_r2r(mm1, mm0); "movq %%mm0, %%mm2 \n\t"
punpckhdq_r2r(mm1, mm2); "punpckldq %%mm1, %%mm0 \n\t"
movq_m2r(*(table + 0), mm1); "punpckhdq %%mm1, %%mm2 \n\t"
movq_m2r(*(table + 4), mm3); "movq (%1), %%mm1 \n\t"
movq_m2r(*(table + 8), mm4); "movq 8(%1), %%mm3 \n\t"
movq_m2r(*(table + 12), mm5); "movq 16(%1), %%mm4 \n\t"
movq_m2r(*(table + 16), mm6); "movq 24(%1), %%mm5 \n\t"
movq_m2r(*(table + 20), mm7); "movq 32(%1), %%mm6 \n\t"
pmaddwd_r2r(mm0, mm1); "movq 40(%1), %%mm7 \n\t"
pmaddwd_r2r(mm2, mm3); "pmaddwd %%mm0, %%mm1 \n\t"
pmaddwd_r2r(mm0, mm4); "pmaddwd %%mm2, %%mm3 \n\t"
pmaddwd_r2r(mm2, mm5); "pmaddwd %%mm0, %%mm4 \n\t"
pmaddwd_r2r(mm0, mm6); "pmaddwd %%mm2, %%mm5 \n\t"
pmaddwd_r2r(mm2, mm7); "pmaddwd %%mm0, %%mm6 \n\t"
pmaddwd_m2r(*(table + 24), mm0); "pmaddwd %%mm2, %%mm7 \n\t"
pmaddwd_m2r(*(table + 28), mm2); "pmaddwd 48(%1), %%mm0 \n\t"
paddd_r2r(mm1, mm3); "pmaddwd 56(%1), %%mm2 \n\t"
paddd_r2r(mm4, mm5); "paddd %%mm1, %%mm3 \n\t"
paddd_r2r(mm6, mm7); "paddd %%mm4, %%mm5 \n\t"
paddd_r2r(mm0, mm2); "paddd %%mm6, %%mm7 \n\t"
movq_m2r(*fdct_r_row, mm0); "paddd %%mm0, %%mm2 \n\t"
paddd_r2r(mm0, mm3); "movq (%2), %%mm0 \n\t"
paddd_r2r(mm0, mm5); "paddd %%mm0, %%mm3 \n\t"
paddd_r2r(mm0, mm7); "paddd %%mm0, %%mm5 \n\t"
paddd_r2r(mm0, mm2); "paddd %%mm0, %%mm7 \n\t"
psrad_i2r(SHIFT_FRW_ROW, mm3); "paddd %%mm0, %%mm2 \n\t"
psrad_i2r(SHIFT_FRW_ROW, mm5); "psrad $"S(SHIFT_FRW_ROW)", %%mm3 \n\t"
psrad_i2r(SHIFT_FRW_ROW, mm7); "psrad $"S(SHIFT_FRW_ROW)", %%mm5 \n\t"
psrad_i2r(SHIFT_FRW_ROW, mm2); "psrad $"S(SHIFT_FRW_ROW)", %%mm7 \n\t"
packssdw_r2r(mm5, mm3); "psrad $"S(SHIFT_FRW_ROW)", %%mm2 \n\t"
packssdw_r2r(mm2, mm7); "packssdw %%mm5, %%mm3 \n\t"
movq_r2m(mm3, *(out + 0)); "packssdw %%mm2, %%mm7 \n\t"
movq_r2m(mm7, *(out + 4)); "movq %%mm3, (%3) \n\t"
"movq %%mm7, 8(%3) \n\t"
:
: "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
} }
static av_always_inline void fdct_row_mmx(const int16_t *in, int16_t *out, const int16_t *table) static av_always_inline void fdct_row_mmx(const int16_t *in, int16_t *out, const int16_t *table)
{ {
//FIXME reorder (I do not have an old MMX-only CPU here to benchmark ...) //FIXME reorder (I do not have an old MMX-only CPU here to benchmark ...)
movd_m2r(*(in + 6), mm1); __asm__ volatile(
punpcklwd_m2r(*(in + 4), mm1); "movd 12(%0), %%mm1 \n\t"
movq_r2r(mm1, mm2); "punpcklwd 8(%0), %%mm1 \n\t"
psrlq_i2r(0x20, mm1); "movq %%mm1, %%mm2 \n\t"
movq_m2r(*(in + 0), mm0); "psrlq $0x20, %%mm1 \n\t"
punpcklwd_r2r(mm2, mm1); "movq 0(%0), %%mm0 \n\t"
movq_r2r(mm0, mm5); "punpcklwd %%mm2, %%mm1 \n\t"
paddsw_r2r(mm1, mm0); "movq %%mm0, %%mm5 \n\t"
psubsw_r2r(mm1, mm5); "paddsw %%mm1, %%mm0 \n\t"
movq_r2r(mm0, mm2); "psubsw %%mm1, %%mm5 \n\t"
punpckldq_r2r(mm5, mm0); "movq %%mm0, %%mm2 \n\t"
punpckhdq_r2r(mm5, mm2); "punpckldq %%mm5, %%mm0 \n\t"
movq_m2r(*(table + 0), mm1); "punpckhdq %%mm5, %%mm2 \n\t"
movq_m2r(*(table + 4), mm3); "movq 0(%1), %%mm1 \n\t"
movq_m2r(*(table + 8), mm4); "movq 8(%1), %%mm3 \n\t"
movq_m2r(*(table + 12), mm5); "movq 16(%1), %%mm4 \n\t"
movq_m2r(*(table + 16), mm6); "movq 24(%1), %%mm5 \n\t"
movq_m2r(*(table + 20), mm7); "movq 32(%1), %%mm6 \n\t"
pmaddwd_r2r(mm0, mm1); "movq 40(%1), %%mm7 \n\t"
pmaddwd_r2r(mm2, mm3); "pmaddwd %%mm0, %%mm1 \n\t"
pmaddwd_r2r(mm0, mm4); "pmaddwd %%mm2, %%mm3 \n\t"
pmaddwd_r2r(mm2, mm5); "pmaddwd %%mm0, %%mm4 \n\t"
pmaddwd_r2r(mm0, mm6); "pmaddwd %%mm2, %%mm5 \n\t"
pmaddwd_r2r(mm2, mm7); "pmaddwd %%mm0, %%mm6 \n\t"
pmaddwd_m2r(*(table + 24), mm0); "pmaddwd %%mm2, %%mm7 \n\t"
pmaddwd_m2r(*(table + 28), mm2); "pmaddwd 48(%1), %%mm0 \n\t"
paddd_r2r(mm1, mm3); "pmaddwd 56(%1), %%mm2 \n\t"
paddd_r2r(mm4, mm5); "paddd %%mm1, %%mm3 \n\t"
paddd_r2r(mm6, mm7); "paddd %%mm4, %%mm5 \n\t"
paddd_r2r(mm0, mm2); "paddd %%mm6, %%mm7 \n\t"
movq_m2r(*fdct_r_row, mm0); "paddd %%mm0, %%mm2 \n\t"
paddd_r2r(mm0, mm3); "movq (%2), %%mm0 \n\t"
paddd_r2r(mm0, mm5); "paddd %%mm0, %%mm3 \n\t"
paddd_r2r(mm0, mm7); "paddd %%mm0, %%mm5 \n\t"
paddd_r2r(mm0, mm2); "paddd %%mm0, %%mm7 \n\t"
psrad_i2r(SHIFT_FRW_ROW, mm3); "paddd %%mm0, %%mm2 \n\t"
psrad_i2r(SHIFT_FRW_ROW, mm5); "psrad $"S(SHIFT_FRW_ROW)", %%mm3 \n\t"
psrad_i2r(SHIFT_FRW_ROW, mm7); "psrad $"S(SHIFT_FRW_ROW)", %%mm5 \n\t"
psrad_i2r(SHIFT_FRW_ROW, mm2); "psrad $"S(SHIFT_FRW_ROW)", %%mm7 \n\t"
packssdw_r2r(mm5, mm3); "psrad $"S(SHIFT_FRW_ROW)", %%mm2 \n\t"
packssdw_r2r(mm2, mm7); "packssdw %%mm5, %%mm3 \n\t"
movq_r2m(mm3, *(out + 0)); "packssdw %%mm2, %%mm7 \n\t"
movq_r2m(mm7, *(out + 4)); "movq %%mm3, 0(%3) \n\t"
"movq %%mm7, 8(%3) \n\t"
:
: "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
} }
void ff_fdct_mmx(int16_t *block) void ff_fdct_mmx(int16_t *block)
......
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