diff options
Diffstat (limited to 'libavcodec/i386/fdct_mmx.c')
| -rw-r--r-- | libavcodec/i386/fdct_mmx.c | 130 |
1 files changed, 65 insertions, 65 deletions
diff --git a/libavcodec/i386/fdct_mmx.c b/libavcodec/i386/fdct_mmx.c index f3023549a2..f6150c83c9 100644 --- a/libavcodec/i386/fdct_mmx.c +++ b/libavcodec/i386/fdct_mmx.c @@ -30,21 +30,21 @@ // ////////////////////////////////////////////////////////////////////// -#define BITS_FRW_ACC 3 //; 2 or 3 for accuracy -#define SHIFT_FRW_COL BITS_FRW_ACC -#define SHIFT_FRW_ROW (BITS_FRW_ACC + 17 - 3) -#define RND_FRW_ROW (1 << (SHIFT_FRW_ROW-1)) -//#define RND_FRW_COL (1 << (SHIFT_FRW_COL-1)) +#define BITS_FRW_ACC 3 //; 2 or 3 for accuracy +#define SHIFT_FRW_COL BITS_FRW_ACC +#define SHIFT_FRW_ROW (BITS_FRW_ACC + 17 - 3) +#define RND_FRW_ROW (1 << (SHIFT_FRW_ROW-1)) +//#define RND_FRW_COL (1 << (SHIFT_FRW_COL-1)) //concatenated table, for forward DCT transformation static const int16_t fdct_tg_all_16[] ATTR_ALIGN(8) = { - 13036, 13036, 13036, 13036, // tg * (2<<16) + 0.5 - 27146, 27146, 27146, 27146, // tg * (2<<16) + 0.5 - -21746, -21746, -21746, -21746, // tg * (2<<16) + 0.5 + 13036, 13036, 13036, 13036, // tg * (2<<16) + 0.5 + 27146, 27146, 27146, 27146, // tg * (2<<16) + 0.5 + -21746, -21746, -21746, -21746, // tg * (2<<16) + 0.5 }; static const int16_t ocos_4_16[4] ATTR_ALIGN(8) = { - 23170, 23170, 23170, 23170, //cos * (2<<15) + 0.5 + 23170, 23170, 23170, 23170, //cos * (2<<15) + 0.5 }; static const int64_t fdct_one_corr ATTR_ALIGN(8) = 0x0001000100010001LL; @@ -351,62 +351,62 @@ static always_inline void fdct_col(const int16_t *in, int16_t *out, int offset) static always_inline void fdct_row_sse2(const int16_t *in, int16_t *out) { asm volatile( - ".macro FDCT_ROW_SSE2_H1 i t \n\t" - "movq \\i(%0), %%xmm2 \n\t" - "movq \\i+8(%0), %%xmm0 \n\t" - "movdqa \\t+32(%1), %%xmm3 \n\t" - "movdqa \\t+48(%1), %%xmm7 \n\t" - "movdqa \\t(%1), %%xmm4 \n\t" - "movdqa \\t+16(%1), %%xmm5 \n\t" - ".endm \n\t" - ".macro FDCT_ROW_SSE2_H2 i t \n\t" - "movq \\i(%0), %%xmm2 \n\t" - "movq \\i+8(%0), %%xmm0 \n\t" - "movdqa \\t+32(%1), %%xmm3 \n\t" - "movdqa \\t+48(%1), %%xmm7 \n\t" - ".endm \n\t" - ".macro FDCT_ROW_SSE2 i \n\t" - "movq %%xmm2, %%xmm1 \n\t" - "pshuflw $27, %%xmm0, %%xmm0 \n\t" - "paddsw %%xmm0, %%xmm1 \n\t" - "psubsw %%xmm0, %%xmm2 \n\t" - "punpckldq %%xmm2, %%xmm1 \n\t" - "pshufd $78, %%xmm1, %%xmm2 \n\t" - "pmaddwd %%xmm2, %%xmm3 \n\t" - "pmaddwd %%xmm1, %%xmm7 \n\t" - "pmaddwd %%xmm5, %%xmm2 \n\t" - "pmaddwd %%xmm4, %%xmm1 \n\t" - "paddd %%xmm7, %%xmm3 \n\t" - "paddd %%xmm2, %%xmm1 \n\t" - "paddd %%xmm6, %%xmm3 \n\t" - "paddd %%xmm6, %%xmm1 \n\t" - "psrad %3, %%xmm3 \n\t" - "psrad %3, %%xmm1 \n\t" - "packssdw %%xmm3, %%xmm1 \n\t" - "movdqa %%xmm1, \\i(%4) \n\t" - ".endm \n\t" - "movdqa (%2), %%xmm6 \n\t" - "FDCT_ROW_SSE2_H1 0 0 \n\t" - "FDCT_ROW_SSE2 0 \n\t" - "FDCT_ROW_SSE2_H2 64 0 \n\t" - "FDCT_ROW_SSE2 64 \n\t" - - "FDCT_ROW_SSE2_H1 16 64 \n\t" - "FDCT_ROW_SSE2 16 \n\t" - "FDCT_ROW_SSE2_H2 112 64 \n\t" - "FDCT_ROW_SSE2 112 \n\t" - - "FDCT_ROW_SSE2_H1 32 128 \n\t" - "FDCT_ROW_SSE2 32 \n\t" - "FDCT_ROW_SSE2_H2 96 128 \n\t" - "FDCT_ROW_SSE2 96 \n\t" - - "FDCT_ROW_SSE2_H1 48 192 \n\t" - "FDCT_ROW_SSE2 48 \n\t" - "FDCT_ROW_SSE2_H2 80 192 \n\t" - "FDCT_ROW_SSE2 80 \n\t" - : - : "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2), "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out) + ".macro FDCT_ROW_SSE2_H1 i t \n\t" + "movq \\i(%0), %%xmm2 \n\t" + "movq \\i+8(%0), %%xmm0 \n\t" + "movdqa \\t+32(%1), %%xmm3 \n\t" + "movdqa \\t+48(%1), %%xmm7 \n\t" + "movdqa \\t(%1), %%xmm4 \n\t" + "movdqa \\t+16(%1), %%xmm5 \n\t" + ".endm \n\t" + ".macro FDCT_ROW_SSE2_H2 i t \n\t" + "movq \\i(%0), %%xmm2 \n\t" + "movq \\i+8(%0), %%xmm0 \n\t" + "movdqa \\t+32(%1), %%xmm3 \n\t" + "movdqa \\t+48(%1), %%xmm7 \n\t" + ".endm \n\t" + ".macro FDCT_ROW_SSE2 i \n\t" + "movq %%xmm2, %%xmm1 \n\t" + "pshuflw $27, %%xmm0, %%xmm0 \n\t" + "paddsw %%xmm0, %%xmm1 \n\t" + "psubsw %%xmm0, %%xmm2 \n\t" + "punpckldq %%xmm2, %%xmm1 \n\t" + "pshufd $78, %%xmm1, %%xmm2 \n\t" + "pmaddwd %%xmm2, %%xmm3 \n\t" + "pmaddwd %%xmm1, %%xmm7 \n\t" + "pmaddwd %%xmm5, %%xmm2 \n\t" + "pmaddwd %%xmm4, %%xmm1 \n\t" + "paddd %%xmm7, %%xmm3 \n\t" + "paddd %%xmm2, %%xmm1 \n\t" + "paddd %%xmm6, %%xmm3 \n\t" + "paddd %%xmm6, %%xmm1 \n\t" + "psrad %3, %%xmm3 \n\t" + "psrad %3, %%xmm1 \n\t" + "packssdw %%xmm3, %%xmm1 \n\t" + "movdqa %%xmm1, \\i(%4) \n\t" + ".endm \n\t" + "movdqa (%2), %%xmm6 \n\t" + "FDCT_ROW_SSE2_H1 0 0 \n\t" + "FDCT_ROW_SSE2 0 \n\t" + "FDCT_ROW_SSE2_H2 64 0 \n\t" + "FDCT_ROW_SSE2 64 \n\t" + + "FDCT_ROW_SSE2_H1 16 64 \n\t" + "FDCT_ROW_SSE2 16 \n\t" + "FDCT_ROW_SSE2_H2 112 64 \n\t" + "FDCT_ROW_SSE2 112 \n\t" + + "FDCT_ROW_SSE2_H1 32 128 \n\t" + "FDCT_ROW_SSE2 32 \n\t" + "FDCT_ROW_SSE2_H2 96 128 \n\t" + "FDCT_ROW_SSE2 96 \n\t" + + "FDCT_ROW_SSE2_H1 48 192 \n\t" + "FDCT_ROW_SSE2 48 \n\t" + "FDCT_ROW_SSE2_H2 80 192 \n\t" + "FDCT_ROW_SSE2 80 \n\t" + : + : "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2), "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out) ); } |