}
}
+static void yuv2yuvX_sse3(const int16_t *filter, int filterSize,
+ const int16_t **src, uint8_t *dest, int dstW,
+ const uint8_t *dither, int offset)
+{
+ if (offset) {
+ __asm__ volatile("movq (%0), %%xmm3\n\t"
+ "movdqa %%xmm3, %%xmm4\n\t"
+ "psrlq $24, %%xmm3\n\t"
+ "psllq $40, %%xmm4\n\t"
+ "por %%xmm4, %%xmm3\n\t"
+ :: "r"(dither)
+ );
+ } else {
+ __asm__ volatile("movq (%0), %%xmm3\n\t"
+ :: "r"(dither)
+ );
+ }
+ __asm__ volatile(
+ "pxor %%xmm0, %%xmm0\n\t"
+ "punpcklbw %%xmm0, %%xmm3\n\t"
+ "psraw $4, %%xmm3\n\t"
+ "movdqa %%xmm3, %%xmm4\n\t"
+ "movdqa %%xmm3, %%xmm7\n\t"
+ "movl %3, %%ecx\n\t"
+ "mov %0, %%"REG_d" \n\t"\
+ "mov (%%"REG_d"), %%"REG_S" \n\t"\
+ ".p2align 4 \n\t" /* FIXME Unroll? */\
+ "1: \n\t"\
+ "movddup 8(%%"REG_d"), %%xmm0 \n\t" /* filterCoeff */\
+ "movdqa (%%"REG_S", %%"REG_c", 2), %%xmm2 \n\t" /* srcData */\
+ "movdqa 16(%%"REG_S", %%"REG_c", 2), %%xmm5 \n\t" /* srcData */\
+ "add $16, %%"REG_d" \n\t"\
+ "mov (%%"REG_d"), %%"REG_S" \n\t"\
+ "test %%"REG_S", %%"REG_S" \n\t"\
+ "pmulhw %%xmm0, %%xmm2 \n\t"\
+ "pmulhw %%xmm0, %%xmm5 \n\t"\
+ "paddw %%xmm2, %%xmm3 \n\t"\
+ "paddw %%xmm5, %%xmm4 \n\t"\
+ " jnz 1b \n\t"\
+ "psraw $3, %%xmm3 \n\t"\
+ "psraw $3, %%xmm4 \n\t"\
+ "packuswb %%xmm4, %%xmm3 \n\t"
+ "movntdq %%xmm3, (%1, %%"REG_c")\n\t"
+ "add $16, %%"REG_c" \n\t"\
+ "cmp %2, %%"REG_c" \n\t"\
+ "movdqa %%xmm7, %%xmm3\n\t"
+ "movdqa %%xmm7, %%xmm4\n\t"
+ "mov %0, %%"REG_d" \n\t"\
+ "mov (%%"REG_d"), %%"REG_S" \n\t"\
+ "jb 1b \n\t"\
+ :: "g" (filter),
+ "r" (dest-offset), "g" ((x86_reg)(dstW+offset)), "m" (offset)
+ : "%"REG_d, "%"REG_S, "%"REG_c
+ );
+}
+
#define SCALE_FUNC(filter_n, from_bpc, to_bpc, opt) \
extern void ff_hscale ## from_bpc ## to ## to_bpc ## _ ## filter_n ## _ ## opt( \
SwsContext *c, int16_t *data, \
SCALE_FUNCS_SSE(ssse3);
SCALE_FUNCS_SSE(sse4);
+#define VSCALEX_FUNC(size, opt) \
+extern void ff_yuv2planeX_ ## size ## _ ## opt(const int16_t *filter, int filterSize, \
+ const int16_t **src, uint8_t *dest, int dstW, \
+ const uint8_t *dither, int offset)
+#define VSCALEX_FUNCS(opt1, opt2) \
+ VSCALEX_FUNC(8, opt1); \
+ VSCALEX_FUNC(9, opt2); \
+ VSCALEX_FUNC(10, opt2)
+
+#if ARCH_X86_32
+VSCALEX_FUNCS(mmx, mmx2);
+#endif
+VSCALEX_FUNCS(sse2, sse2);
+VSCALEX_FUNCS(sse4, sse4);
+VSCALEX_FUNC(16, sse4);
+VSCALEX_FUNCS(avx, avx);
+
void ff_sws_init_swScale_mmx(SwsContext *c)
{
int cpu_flags = av_get_cpu_flags();
#if HAVE_MMX2
if (cpu_flags & AV_CPU_FLAG_MMX2)
sws_init_swScale_MMX2(c);
+ if (cpu_flags & AV_CPU_FLAG_SSE3){
+ if(c->use_mmx_vfilter && !(c->flags & SWS_ACCURATE_RND))
+ c->yuv2planeX = yuv2yuvX_sse3;
+ }
#endif
#if HAVE_YASM
case 8: ASSIGN_SCALE_FUNC2(hscalefn, 8, opt1, opt2); break; \
default: ASSIGN_SCALE_FUNC2(hscalefn, X, opt1, opt2); break; \
}
+#define ASSIGN_VSCALEX_FUNC(vscalefn, opt1, opt2, opt2chk, do_16_case) \
+switch(c->dstBpc){ \
+ case 16: /*do_16_case;*/ break; \
+ case 10: if (!isBE(c->dstFormat) && opt2chk) /*vscalefn = ff_yuv2planeX_10_ ## opt2;*/ break; \
+ case 9: if (!isBE(c->dstFormat) && opt2chk) /*vscalefn = ff_yuv2planeX_9_ ## opt2;*/ break; \
+ default: /*vscalefn = ff_yuv2planeX_8_ ## opt1;*/ break; \
+ }
#if ARCH_X86_32
if (cpu_flags & AV_CPU_FLAG_MMX) {
ASSIGN_MMX_SCALE_FUNC(c->hyScale, c->hLumFilterSize, mmx, mmx);
ASSIGN_MMX_SCALE_FUNC(c->hcScale, c->hChrFilterSize, mmx, mmx);
+ ASSIGN_VSCALEX_FUNC(c->yuv2planeX, mmx, mmx2, cpu_flags & AV_CPU_FLAG_MMX2,);
}
#endif
#define ASSIGN_SSE_SCALE_FUNC(hscalefn, filtersize, opt1, opt2) \
if (cpu_flags & AV_CPU_FLAG_SSE2) {
ASSIGN_SSE_SCALE_FUNC(c->hyScale, c->hLumFilterSize, sse2, sse2);
ASSIGN_SSE_SCALE_FUNC(c->hcScale, c->hChrFilterSize, sse2, sse2);
+ ASSIGN_VSCALEX_FUNC(c->yuv2planeX, sse2, sse2, 1,);
}
if (cpu_flags & AV_CPU_FLAG_SSSE3) {
ASSIGN_SSE_SCALE_FUNC(c->hyScale, c->hLumFilterSize, ssse3, ssse3);
/* Xto15 don't need special sse4 functions */
ASSIGN_SSE_SCALE_FUNC(c->hyScale, c->hLumFilterSize, sse4, ssse3);
ASSIGN_SSE_SCALE_FUNC(c->hcScale, c->hChrFilterSize, sse4, ssse3);
+ ASSIGN_VSCALEX_FUNC(c->yuv2planeX, sse4, sse4, 1,
+ if (!isBE(c->dstFormat)) c->yuv2planeX = ff_yuv2planeX_16_sse4);
+ }
+
+ if (cpu_flags & AV_CPU_FLAG_AVX) {
+ ASSIGN_VSCALEX_FUNC(c->yuv2planeX, avx, avx, 1,);
}
#endif
}