diff --git a/libavcodec/i386/mpegvideo_mmx.c b/libavcodec/i386/mpegvideo_mmx.c index 549d019f48..e797cc311d 100644 --- a/libavcodec/i386/mpegvideo_mmx.c +++ b/libavcodec/i386/mpegvideo_mmx.c @@ -65,6 +65,86 @@ void MPV_frame_start(MpegEncContext *s) static const unsigned long long int mm_wabs __attribute__ ((aligned(8))) = 0xffffffffffffffffULL; static const unsigned long long int mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL; + +static void dct_unquantize_h263_mmx(MpegEncContext *s, + DCTELEM *block, int n, int qscale) +{ + int i, level, qmul, qadd; + + qmul = s->qscale << 1; + qadd = (s->qscale - 1) | 1; + + if (s->mb_intra) { + if (n < 4) + block[0] = block[0] * s->y_dc_scale; + else + block[0] = block[0] * s->c_dc_scale; + + for(i=1; i<8; i++) { + level = block[i]; + if (level) { + if (level < 0) { + level = level * qmul - qadd; + } else { + level = level * qmul + qadd; + } + block[i] = level; + } + } + } else { + i = 0; + } + +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" + "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" + "cmpl $128, %3 \n\t" + "jb 1b \n\t" + ::"r" (block), "g"(qmul), "g" (qadd), "r" (2*i) + : "memory" + ); +} + + /* NK: Note: looking at PARANOID: @@ -142,9 +222,9 @@ static void dct_unquantize_mpeg1_mmx(MpegEncContext *s, "pcmpgtw %%mm6, %%mm2\n\t" "pmullw %2, %%mm1\n\t" "pandn %%mm4, %%mm2\n\t" - "por %%mm5, %%mm2\n\t" + "por %%mm5, %%mm2\n\t" "pmullw %%mm2, %%mm0\n\t" /* mm0 = abs(block[i]). */ - + "pcmpeqw %%mm6, %%mm3\n\t" "pmullw %%mm0, %%mm1\n\t" "psraw $3, %%mm1\n\t" @@ -164,7 +244,7 @@ static void dct_unquantize_mpeg1_mmx(MpegEncContext *s, unquant_even: quant_matrix = s->non_intra_matrix; /* Align on 4 elements boundary */ - while(i&3) + while(i&7) { level = block[i]; if (level) { @@ -177,6 +257,61 @@ static void dct_unquantize_mpeg1_mmx(MpegEncContext *s, } i++; } + +asm volatile( + "pcmpeqw %%mm7, %%mm7 \n\t" + "psrlw $15, %%mm7 \n\t" + "movd %2, %%mm6 \n\t" + "packssdw %%mm6, %%mm6 \n\t" + "packssdw %%mm6, %%mm6 \n\t" + "1: \n\t" + "movq (%0, %3), %%mm0 \n\t" + "movq 8(%0, %3), %%mm1 \n\t" + "movq (%1, %3), %%mm4 \n\t" + "movq 8(%1, %3), %%mm5 \n\t" + "pmullw %%mm6, %%mm4 \n\t" // q=qscale*quant_matrix[i] + "pmullw %%mm6, %%mm5 \n\t" // q=qscale*quant_matrix[i] + "pxor %%mm2, %%mm2 \n\t" + "pxor %%mm3, %%mm3 \n\t" + "pcmpgtw %%mm0, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 + "pcmpgtw %%mm1, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 + "pxor %%mm2, %%mm0 \n\t" + "pxor %%mm3, %%mm1 \n\t" + "psubw %%mm2, %%mm0 \n\t" // abs(block[i]) + "psubw %%mm3, %%mm1 \n\t" // abs(block[i]) + "paddw %%mm0, %%mm0 \n\t" // abs(block[i])*2 + "paddw %%mm1, %%mm1 \n\t" // abs(block[i])*2 + "paddw %%mm7, %%mm0 \n\t" // abs(block[i])*2 + 1 + "paddw %%mm7, %%mm1 \n\t" // abs(block[i])*2 + 1 + "pmullw %%mm4, %%mm0 \n\t" // (abs(block[i])*2 + 1)*q + "pmullw %%mm5, %%mm1 \n\t" // (abs(block[i])*2 + 1)*q + "pxor %%mm4, %%mm4 \n\t" + "pxor %%mm5, %%mm5 \n\t" // FIXME slow + "pcmpeqw (%0, %3), %%mm4 \n\t" // block[i] == 0 ? -1 : 0 + "pcmpeqw 8(%0, %3), %%mm5 \n\t" // block[i] == 0 ? -1 : 0 + "psraw $4, %%mm0 \n\t" + "psraw $4, %%mm1 \n\t" + "psubw %%mm7, %%mm0 \n\t" + "psubw %%mm7, %%mm1 \n\t" + "por %%mm7, %%mm0 \n\t" + "por %%mm7, %%mm1 \n\t" + "pxor %%mm2, %%mm0 \n\t" + "pxor %%mm3, %%mm1 \n\t" + "psubw %%mm2, %%mm0 \n\t" + "psubw %%mm3, %%mm1 \n\t" + "pandn %%mm0, %%mm4 \n\t" + "pandn %%mm1, %%mm5 \n\t" + "movq %%mm4, (%0, %3) \n\t" + "movq %%mm5, 8(%0, %3) \n\t" + + "addl $16, %3 \n\t" + "cmpl $128, %3 \n\t" + "jb 1b \n\t" + ::"r" (block), "r"(quant_matrix), "g" (qscale), "r" (2*i) + : "memory" + ); + +#if 0 __asm __volatile( "movd %0, %%mm6\n\t" /* mm6 = qscale | 0 */ "punpckldq %%mm6, %%mm6\n\t" /* mm6 = qscale | qscale */ @@ -185,7 +320,7 @@ static void dct_unquantize_mpeg1_mmx(MpegEncContext *s, "movq %1, %%mm5\n\t" "packssdw %%mm6, %%mm7\n\t" /* mm7 = qscale | qscale | qscale | qscale */ "pxor %%mm6, %%mm6\n\t" - ::"g"(qscale),"m"(mm_wone),"m"(mm_wabs):"memory"); + ::"g"(qscale),"m"(mm_wone),"m"(mm_wabs)); for(;i<64;i+=4) { __asm __volatile( "movq %1, %%mm0\n\t" @@ -195,11 +330,11 @@ static void dct_unquantize_mpeg1_mmx(MpegEncContext *s, "pcmpgtw %%mm6, %%mm2\n\t" "pmullw %2, %%mm1\n\t" "pandn %%mm4, %%mm2\n\t" - "por %%mm5, %%mm2\n\t" + "por %%mm5, %%mm2\n\t" "pmullw %%mm2, %%mm0\n\t" /* mm0 = abs(block[i]). */ "psllw $1, %%mm0\n\t" /* block[i] <<= 1 */ "paddw %%mm5, %%mm0\n\t" /* block[i] ++ */ - + "pmullw %%mm0, %%mm1\n\t" "psraw $4, %%mm1\n\t" "pcmpeqw %%mm6, %%mm3\n\t" @@ -211,17 +346,18 @@ static void dct_unquantize_mpeg1_mmx(MpegEncContext *s, "pand %%mm3, %%mm1\n\t" /* nullify if was zero */ "movq %%mm1, %0" :"=m"(block[i]) - :"m"(block[i]), "m"(quant_matrix[i]) - :"memory"); + :"m"(block[i]), "m"(quant_matrix[i])); } +#endif } } void MPV_common_init_mmx(MpegEncContext *s) { if (mm_flags & MM_MMX) { - /* XXX: should include h263 optimization too. It would go even - faster! */ - s->dct_unquantize = dct_unquantize_mpeg1_mmx; + if (s->out_format == FMT_H263) + s->dct_unquantize = dct_unquantize_h263_mmx; + else + s->dct_unquantize = dct_unquantize_mpeg1_mmx; } }