Attachment #233045 for bug #290741

Lines 1138-1144 Link Here

(-)libavcodec/x86/h264dsp_mmx.c.ori (-12 / +10 lines)
1138	int h=8;\	1138	int h=8;\
1139	__asm__ volatile(\	1139	__asm__ volatile(\
1140	"pxor %%mm7, %%mm7 \n\t"\	1140	"pxor %%mm7, %%mm7 \n\t"\
1141	"movq %5, %%mm6 \n\t"\	1141	"movq "MANGLE(ff_pw_5) ", %%mm6\n\t"\
1142	"1: \n\t"\	1142	"1: \n\t"\
1143	"movq (%0), %%mm0 \n\t"\	1143	"movq (%0), %%mm0 \n\t"\
1144	"movq 1(%0), %%mm2 \n\t"\	1144	"movq 1(%0), %%mm2 \n\t"\
Lines 1172-1178 Link Here
1172	"punpcklbw %%mm7, %%mm5 \n\t"\	1172	"punpcklbw %%mm7, %%mm5 \n\t"\
1173	"paddw %%mm3, %%mm2 \n\t"\	1173	"paddw %%mm3, %%mm2 \n\t"\
1174	"paddw %%mm5, %%mm4 \n\t"\	1174	"paddw %%mm5, %%mm4 \n\t"\
1175	"movq %6, %%mm5 \n\t"\	1175	"movq "MANGLE(ff_pw_16) ", %%mm5\n\t"\
1176	"paddw %%mm5, %%mm2 \n\t"\	1176	"paddw %%mm5, %%mm2 \n\t"\
1177	"paddw %%mm5, %%mm4 \n\t"\	1177	"paddw %%mm5, %%mm4 \n\t"\
1178	"paddw %%mm2, %%mm0 \n\t"\	1178	"paddw %%mm2, %%mm0 \n\t"\
Lines 1186-1192 Link Here
1186	"decl %2 \n\t"\	1186	"decl %2 \n\t"\
1187	" jnz 1b \n\t"\	1187	" jnz 1b \n\t"\
1188	: "+a"(src), "+c"(dst), "+g"(h)\	1188	: "+a"(src), "+c"(dst), "+g"(h)\
1189	: "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\	1189	: "d"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
1190	: "memory"\	1190	: "memory"\
1191	);\	1191	);\
1192	}\	1192	}\
Lines 1593-1600 Link Here
1593	int h=8;\	1593	int h=8;\
1594	__asm__ volatile(\	1594	__asm__ volatile(\
1595	"pxor %%xmm7, %%xmm7 \n\t"\	1595	"pxor %%xmm7, %%xmm7 \n\t"\
1596	"movdqa %0, %%xmm6 \n\t"\	1596	"movdqa "MANGLE(ff_pw_5) ", %%xmm6 \n\t"\
1597	:: "m"(ff_pw_5)\	1597	::\
1598	);\	1598	);\
1599	do{\	1599	do{\
1600	__asm__ volatile(\	1600	__asm__ volatile(\
Lines 1617-1623 Link Here
1617	"psllw $2, %%xmm2 \n\t"\	1617	"psllw $2, %%xmm2 \n\t"\
1618	"movq (%2), %%xmm3 \n\t"\	1618	"movq (%2), %%xmm3 \n\t"\
1619	"psubw %%xmm1, %%xmm2 \n\t"\	1619	"psubw %%xmm1, %%xmm2 \n\t"\
1620	"paddw %5, %%xmm0 \n\t"\	1620	"paddw "MANGLE(ff_pw_16)", %%xmm0 \n\t"\
1621	"pmullw %%xmm6, %%xmm2 \n\t"\	1621	"pmullw %%xmm6, %%xmm2 \n\t"\
1622	"paddw %%xmm0, %%xmm2 \n\t"\	1622	"paddw %%xmm0, %%xmm2 \n\t"\
1623	"psraw $5, %%xmm2 \n\t"\	1623	"psraw $5, %%xmm2 \n\t"\
Lines 1628-1635 Link Here
1628	"add %4, %1 \n\t"\	1628	"add %4, %1 \n\t"\
1629	"add %3, %2 \n\t"\	1629	"add %3, %2 \n\t"\
1630	: "+a"(src), "+c"(dst), "+d"(src2)\	1630	: "+a"(src), "+c"(dst), "+d"(src2)\
1631	: "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride),\	1631	: "D"((x86_reg)src2Stride), "S"((x86_reg)dstStride)\
1632	"m"(ff_pw_16)\
1633	: "memory"\	1632	: "memory"\
1634	);\	1633	);\
1635	}while(--h);\	1634	}while(--h);\
Lines 1640-1646 Link Here
1640	int h=8;\	1639	int h=8;\
1641	__asm__ volatile(\	1640	__asm__ volatile(\
1642	"pxor %%xmm7, %%xmm7 \n\t"\	1641	"pxor %%xmm7, %%xmm7 \n\t"\
1643	"movdqa %5, %%xmm6 \n\t"\	1642	"movdqa "MANGLE(ff_pw_5)", %%xmm6\n\t"\
1644	"1: \n\t"\	1643	"1: \n\t"\
1645	"lddqu -2(%0), %%xmm1 \n\t"\	1644	"lddqu -2(%0), %%xmm1 \n\t"\
1646	"movdqa %%xmm1, %%xmm0 \n\t"\	1645	"movdqa %%xmm1, %%xmm0 \n\t"\
Lines 1660-1666 Link Here
1660	"paddw %%xmm4, %%xmm1 \n\t"\	1659	"paddw %%xmm4, %%xmm1 \n\t"\
1661	"psllw $2, %%xmm2 \n\t"\	1660	"psllw $2, %%xmm2 \n\t"\
1662	"psubw %%xmm1, %%xmm2 \n\t"\	1661	"psubw %%xmm1, %%xmm2 \n\t"\
1663	"paddw %6, %%xmm0 \n\t"\	1662	"paddw "MANGLE(ff_pw_16)", %%xmm0 \n\t"\
1664	"pmullw %%xmm6, %%xmm2 \n\t"\	1663	"pmullw %%xmm6, %%xmm2 \n\t"\
1665	"paddw %%xmm0, %%xmm2 \n\t"\	1664	"paddw %%xmm0, %%xmm2 \n\t"\
1666	"psraw $5, %%xmm2 \n\t"\	1665	"psraw $5, %%xmm2 \n\t"\
Lines 1671-1678 Link Here
1671	"decl %2 \n\t"\	1670	"decl %2 \n\t"\
1672	" jnz 1b \n\t"\	1671	" jnz 1b \n\t"\
1673	: "+a"(src), "+c"(dst), "+g"(h)\	1672	: "+a"(src), "+c"(dst), "+g"(h)\
1674	: "D"((x86_reg)srcStride), "S"((x86_reg)dstStride),\	1673	: "D"((x86_reg)srcStride), "S"((x86_reg)dstStride)\
1675	"m"(ff_pw_5), "m"(ff_pw_16)\
1676	: "memory"\	1674	: "memory"\
1677	);\	1675	);\
1678	}\	1676	}\




}

static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){
    uint32_t l1 = *(uint32_t*)(src + 0*src_stride);
    uint32_t l2 = *(uint32_t*)(src + 1*src_stride);
    uint32_t l3 = *(uint32_t*)(src + 2*src_stride);
    uint32_t l4 = *(uint32_t*)(src + 3*src_stride);
 
    __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...
        "movd  %0, %%mm0                \n\t"
        "movd  %1, %%mm1                \n\t"
        "movd  %2, %%mm2                \n\t"
        "movd  %3, %%mm3                \n\t"
        "punpcklbw %%mm1, %%mm0         \n\t"
        "punpcklbw %%mm3, %%mm2         \n\t"
        "movq %%mm0, %%mm1              \n\t"

        "punpckhdq %%mm1, %%mm1         \n\t"
        "movd  %%mm1, %3                \n\t"

        : "+m" (l1),
          "+m" (l2),
          "+m" (l3),
          "+m" (l4)
        :  "m" (*(uint32_t*)(src + 0*src_stride)),
           "m" (*(uint32_t*)(src + 1*src_stride)),
           "m" (*(uint32_t*)(src + 2*src_stride)),
           "m" (*(uint32_t*)(src + 3*src_stride))
    );

    *(uint32_t*)(dst + 0*dst_stride) = l1;
    *(uint32_t*)(dst + 1*dst_stride) = l2;
    *(uint32_t*)(dst + 2*dst_stride) = l3;
    *(uint32_t*)(dst + 3*dst_stride) = l4;
}

static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){

Return to bug 290741

Lines 724-734 Link Here

(-)libavcodec/x86/dsputil_mmx.c.ori (-12 / +18 lines)
724	}	724	}
725		725
726	static inline void transpose4x4(uint8_t dst, uint8_t src, int dst_stride, int src_stride){	726	static inline void transpose4x4(uint8_t dst, uint8_t src, int dst_stride, int src_stride){
		727	uint32_t l1 = (uint32_t)(src + 0*src_stride);
		728	uint32_t l2 = (uint32_t)(src + 1*src_stride);
		729	uint32_t l3 = (uint32_t)(src + 2*src_stride);
		730	uint32_t l4 = (uint32_t)(src + 3*src_stride);
		731
727	__asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...	732	__asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...
728	"movd %4, %%mm0 \n\t"	733	"movd %0, %%mm0 \n\t"
729	"movd %5, %%mm1 \n\t"	734	"movd %1, %%mm1 \n\t"
730	"movd %6, %%mm2 \n\t"	735	"movd %2, %%mm2 \n\t"
731	"movd %7, %%mm3 \n\t"	736	"movd %3, %%mm3 \n\t"
732	"punpcklbw %%mm1, %%mm0 \n\t"	737	"punpcklbw %%mm1, %%mm0 \n\t"
733	"punpcklbw %%mm3, %%mm2 \n\t"	738	"punpcklbw %%mm3, %%mm2 \n\t"
734	"movq %%mm0, %%mm1 \n\t"	739	"movq %%mm0, %%mm1 \n\t"
Lines 741-755 Link Here
741	"punpckhdq %%mm1, %%mm1 \n\t"	746	"punpckhdq %%mm1, %%mm1 \n\t"
742	"movd %%mm1, %3 \n\t"	747	"movd %%mm1, %3 \n\t"
743		748
744	: "=m" ((uint32_t)(dst + 0*dst_stride)),	749	: "+m" (l1),
745	"=m" ((uint32_t)(dst + 1*dst_stride)),	750	"+m" (l2),
746	"=m" ((uint32_t)(dst + 2*dst_stride)),	751	"+m" (l3),
747	"=m" ((uint32_t)(dst + 3*dst_stride))	752	"+m" (l4)
748	: "m" ((uint32_t)(src + 0*src_stride)),
749	"m" ((uint32_t)(src + 1*src_stride)),
750	"m" ((uint32_t)(src + 2*src_stride)),
751	"m" ((uint32_t)(src + 3*src_stride))
752	);	753	);
		754
		755	(uint32_t)(dst + 0*dst_stride) = l1;
		756	(uint32_t)(dst + 1*dst_stride) = l2;
		757	(uint32_t)(dst + 2*dst_stride) = l3;
		758	(uint32_t)(dst + 3*dst_stride) = l4;
753	}	759	}
754		760
755	static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){	761	static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){