Attachment #233283 for bug #290741




}

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){

Lines 1136-1144 Link Here

(-)libavcodec/x86/h264dsp_mmx.c.ori (-6 / +11 lines)
1136	\	1136	\
1137	static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t dst, uint8_t src, int dstStride, int srcStride){\	1137	static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t dst, uint8_t src, int dstStride, int srcStride){\
1138	int h=8;\	1138	int h=8;\
		1139	const xmm_reg xmm_consts[2] __attribute__((aligned (16))) = { ff_pw_5, ff_pw_16 }; \
		1140	\
1139	__asm__ volatile(\	1141	__asm__ volatile(\
1140	"pxor %%mm7, %%mm7 \n\t"\	1142	"pxor %%mm7, %%mm7 \n\t"\
1141	"movq %5, %%mm6 \n\t"\	1143	"movq (%5), %%mm6 \n\t"\
		1144	"add $16, %5 \n\t"\
1142	"1: \n\t"\	1145	"1: \n\t"\
1143	"movq (%0), %%mm0 \n\t"\	1146	"movq (%0), %%mm0 \n\t"\
1144	"movq 1(%0), %%mm2 \n\t"\	1147	"movq 1(%0), %%mm2 \n\t"\
Lines 1172-1178 Link Here
1172	"punpcklbw %%mm7, %%mm5 \n\t"\	1175	"punpcklbw %%mm7, %%mm5 \n\t"\
1173	"paddw %%mm3, %%mm2 \n\t"\	1176	"paddw %%mm3, %%mm2 \n\t"\
1174	"paddw %%mm5, %%mm4 \n\t"\	1177	"paddw %%mm5, %%mm4 \n\t"\
1175	"movq %6, %%mm5 \n\t"\	1178	"movq (%5), %%mm5 \n\t"\
1176	"paddw %%mm5, %%mm2 \n\t"\	1179	"paddw %%mm5, %%mm2 \n\t"\
1177	"paddw %%mm5, %%mm4 \n\t"\	1180	"paddw %%mm5, %%mm4 \n\t"\
1178	"paddw %%mm2, %%mm0 \n\t"\	1181	"paddw %%mm2, %%mm0 \n\t"\
Lines 1186-1192 Link Here
1186	"decl %2 \n\t"\	1189	"decl %2 \n\t"\
1187	" jnz 1b \n\t"\	1190	" jnz 1b \n\t"\
1188	: "+a"(src), "+c"(dst), "+g"(h)\	1191	: "+a"(src), "+c"(dst), "+g"(h)\
1189	: "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\	1192	: "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "r"(xmm_consts)\
1190	: "memory"\	1193	: "memory"\
1191	);\	1194	);\
1192	}\	1195	}\
Lines 1638-1646 Link Here
1638	\	1641	\
1639	static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t dst, uint8_t src, int dstStride, int srcStride){\	1642	static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t dst, uint8_t src, int dstStride, int srcStride){\
1640	int h=8;\	1643	int h=8;\
		1644	const xmm_reg xmm_consts[2] __attribute__((aligned (16))) = { ff_pw_5, ff_pw_16 }; \
1641	__asm__ volatile(\	1645	__asm__ volatile(\
1642	"pxor %%xmm7, %%xmm7 \n\t"\	1646	"pxor %%xmm7, %%xmm7 \n\t"\
1643	"movdqa %5, %%xmm6 \n\t"\	1647	"movdqa (%5), %%xmm6 \n\t"\
		1648	"add $16, %5 \n\t"\
1644	"1: \n\t"\	1649	"1: \n\t"\
1645	"lddqu -2(%0), %%xmm1 \n\t"\	1650	"lddqu -2(%0), %%xmm1 \n\t"\
1646	"movdqa %%xmm1, %%xmm0 \n\t"\	1651	"movdqa %%xmm1, %%xmm0 \n\t"\
Lines 1660-1666 Link Here
1660	"paddw %%xmm4, %%xmm1 \n\t"\	1665	"paddw %%xmm4, %%xmm1 \n\t"\
1661	"psllw $2, %%xmm2 \n\t"\	1666	"psllw $2, %%xmm2 \n\t"\
1662	"psubw %%xmm1, %%xmm2 \n\t"\	1667	"psubw %%xmm1, %%xmm2 \n\t"\
1663	"paddw %6, %%xmm0 \n\t"\	1668	"paddw (%5), %%xmm0 \n\t"\
1664	"pmullw %%xmm6, %%xmm2 \n\t"\	1669	"pmullw %%xmm6, %%xmm2 \n\t"\
1665	"paddw %%xmm0, %%xmm2 \n\t"\	1670	"paddw %%xmm0, %%xmm2 \n\t"\
1666	"psraw $5, %%xmm2 \n\t"\	1671	"psraw $5, %%xmm2 \n\t"\
Lines 1672-1678 Link Here
1672	" jnz 1b \n\t"\	1677	" jnz 1b \n\t"\
1673	: "+a"(src), "+c"(dst), "+g"(h)\	1678	: "+a"(src), "+c"(dst), "+g"(h)\
1674	: "D"((x86_reg)srcStride), "S"((x86_reg)dstStride),\	1679	: "D"((x86_reg)srcStride), "S"((x86_reg)dstStride),\
1675	"m"(ff_pw_5), "m"(ff_pw_16)\	1680	"r"(xmm_consts)\
1676	: "memory"\	1681	: "memory"\
1677	);\	1682	);\
1678	}\	1683	}\

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){