static void frame_i2f_sse(u_char *src,float *dst,int l) { int i; pxor_r2r(mm7,mm7); for( i=0; i<l; i+=8 ) { movq_m2r(*src,mm0); movq_r2r(mm0, mm2); punpcklbw_r2r(mm7, mm0); punpckhbw_r2r(mm7, mm2); movq_r2r(mm0, mm1); movq_r2r(mm2, mm3); punpcklwd_r2r(mm7, mm0); punpckhwd_r2r(mm7, mm1); punpcklwd_r2r(mm7, mm2); punpckhwd_r2r(mm7, mm3); cvtpi2ps_r2r(mm0,xmm0); cvtpi2ps_r2r(mm1,xmm1); cvtpi2ps_r2r(mm2,xmm2); cvtpi2ps_r2r(mm3,xmm3); movlps_r2m(xmm0,dst[0]); movlps_r2m(xmm1,dst[2]); movlps_r2m(xmm2,dst[4]); movlps_r2m(xmm3,dst[6]); src+=8; dst+=8; } emms(); }
static inline void mmx_unpack_32rgb (uint8_t * image, const int cpu) { /* * convert RGB plane to RGB packed format, * mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0, * mm4 -> GB, mm5 -> AR pixel 4-7, * mm6 -> GB, mm7 -> AR pixel 0-3 */ pxor_r2r (mm3, mm3); movq_r2r (mm0, mm6); movq_r2r (mm1, mm7); movq_r2r (mm0, mm4); movq_r2r (mm1, mm5); punpcklbw_r2r (mm2, mm6); punpcklbw_r2r (mm3, mm7); punpcklwd_r2r (mm7, mm6); movntq (mm6, *image); movq_r2r (mm0, mm6); punpcklbw_r2r (mm2, mm6); punpckhwd_r2r (mm7, mm6); movntq (mm6, *(image+8)); punpckhbw_r2r (mm2, mm4); punpckhbw_r2r (mm3, mm5); punpcklwd_r2r (mm5, mm4); movntq (mm4, *(image+16)); movq_r2r (mm0, mm4); punpckhbw_r2r (mm2, mm4); punpckhwd_r2r (mm5, mm4); movntq (mm4, *(image+24)); }
static void frame_f2i_sse(float *src,u_char *dst,int l) { int i; // put 128 in all 4 words of mm7 movd_g2r(128,mm7); punpcklwd_r2r(mm7,mm7); punpckldq_r2r(mm7,mm7); // put 128 in all 8 bytes of mm6 movd_g2r(128,mm6); punpcklbw_r2r(mm6,mm6); punpcklwd_r2r(mm6,mm6); punpckldq_r2r(mm6,mm6); for( i=0; i<l; i+=8 ) { movaps_m2r(src[0],xmm0); movaps_m2r(src[4],xmm2); movhlps_r2r(xmm0,xmm1); cvtps2pi_r2r(xmm0,mm0); cvtps2pi_r2r(xmm1,mm1); movhlps_r2r(xmm2,xmm3); cvtps2pi_r2r(xmm2,mm2); cvtps2pi_r2r(xmm3,mm3); packssdw_r2r(mm1,mm0); packssdw_r2r(mm3,mm2); psubw_r2r(mm7,mm0); psubw_r2r(mm7,mm2); packsswb_r2r(mm2, mm0); paddb_r2r(mm6, mm0); movq_r2m(mm0,dst[0]); src+=8; dst+=8; } emms(); }
static __inline__ void mmx_sum_4_word_accs( mmx_t *accs, int32_t *res ) { movq_m2r( *accs, mm1 ); movq_r2r( mm1, mm3 ); movq_r2r( mm1, mm2 ); /* Generate sign extensions for mm1 words! */ psraw_i2r( 15, mm3 ); punpcklwd_r2r( mm3, mm1 ); punpckhwd_r2r( mm3, mm2 ); paddd_r2r( mm1, mm2 ); movq_r2r( mm2, mm3); psrlq_i2r( 32, mm2); paddd_r2r( mm2, mm3); movd_r2m( mm3, *res ); }
static void _evas_yv12torgb_sse(unsigned char **yuv, unsigned char *rgb, int w, int h) { #ifdef BUILD_MMX int xx, yy; register unsigned char *yp1, *up, *vp; unsigned char *dp1; /* destination pointers */ dp1 = rgb; for (yy = 0; yy < h; yy++) { /* plane pointers */ yp1 = yuv[yy]; up = yuv[h + (yy / 2)]; vp = yuv[h + (h / 2) + (yy / 2)]; for (xx = 0; xx < (w - 7); xx += 8) { movd_m2r(*up, mm3); movd_m2r(*vp, mm2); movq_m2r(*yp1, mm0); pxor_r2r(mm7, mm7); punpcklbw_r2r(mm7, mm2); punpcklbw_r2r(mm7, mm3); movq_r2r(mm0, mm1); psrlw_i2r(8, mm0); psllw_i2r(8, mm1); psrlw_i2r(8, mm1); movq_m2r(CONST_16, mm4); psubsw_r2r(mm4, mm0); psubsw_r2r(mm4, mm1); movq_m2r(CONST_128, mm5); psubsw_r2r(mm5, mm2); psubsw_r2r(mm5, mm3); movq_m2r(CONST_YMUL, mm4); pmullw_r2r(mm4, mm0); pmullw_r2r(mm4, mm1); movq_m2r(CONST_CRVCRV, mm7); pmullw_r2r(mm3, mm7); movq_m2r(CONST_CBUCBU, mm6); pmullw_r2r(mm2, mm6); movq_m2r(CONST_CGUCGU, mm5); pmullw_r2r(mm2, mm5); movq_m2r(CONST_CGVCGV, mm4); pmullw_r2r(mm3, mm4); movq_r2r(mm0, mm2); paddsw_r2r(mm7, mm2); paddsw_r2r(mm1, mm7); psraw_i2r(RES, mm2); psraw_i2r(RES, mm7); packuswb_r2r(mm7, mm2); pxor_r2r(mm7, mm7); movq_r2r(mm2, mm3); punpckhbw_r2r(mm7, mm2); punpcklbw_r2r(mm3, mm7); por_r2r(mm7, mm2); movq_r2r(mm0, mm3); psubsw_r2r(mm5, mm3); psubsw_r2r(mm4, mm3); paddsw_m2r(CONST_32, mm3); movq_r2r(mm1, mm7); psubsw_r2r(mm5, mm7); psubsw_r2r(mm4, mm7); paddsw_m2r(CONST_32, mm7); psraw_i2r(RES, mm3); psraw_i2r(RES, mm7); packuswb_r2r(mm7, mm3); pxor_r2r(mm7, mm7); movq_r2r(mm3, mm4); punpckhbw_r2r(mm7, mm3); punpcklbw_r2r(mm4, mm7); por_r2r(mm7, mm3); movq_m2r(CONST_32, mm4); paddsw_r2r(mm6, mm0); paddsw_r2r(mm6, mm1); paddsw_r2r(mm4, mm0); paddsw_r2r(mm4, mm1); psraw_i2r(RES, mm0); psraw_i2r(RES, mm1); packuswb_r2r(mm1, mm0); pxor_r2r(mm7, mm7); movq_r2r(mm0, mm5); punpckhbw_r2r(mm7, mm0); punpcklbw_r2r(mm5, mm7); por_r2r(mm7, mm0); movq_m2r(CONST_FF, mm1); movq_r2r(mm0, mm5); movq_r2r(mm3, mm6); movq_r2r(mm2, mm7); punpckhbw_r2r(mm3, mm2); punpcklbw_r2r(mm6, mm7); punpckhbw_r2r(mm1, mm0); punpcklbw_r2r(mm1, mm5); movq_r2r(mm7, mm1); punpckhwd_r2r(mm5, mm7); punpcklwd_r2r(mm5, mm1); movq_r2r(mm2, mm4); punpckhwd_r2r(mm0, mm2); punpcklwd_r2r(mm0, mm4); movntq_r2m(mm1, *(dp1)); movntq_r2m(mm7, *(dp1 + 8)); movntq_r2m(mm4, *(dp1 + 16)); movntq_r2m(mm2, *(dp1 + 24)); yp1 += 8; up += 4; vp += 4; dp1 += 8 * 4; } /* cleanup pixles that arent a multiple of 8 pixels wide */ if (xx < w) { int y, u, v, r, g, b; for (; xx < w; xx += 2) { u = (*up++) - 128; v = (*vp++) - 128; y = RZ(YMUL) * ((*yp1++) - 16); r = LUT_CLIP((y + (_crv * v)) >> RES); g = LUT_CLIP((y - (_cgu * u) - (_cgv * v) + RZ(OFF)) >> RES); b = LUT_CLIP((y + (_cbu * u) + RZ(OFF)) >> RES); *((DATA32 *) dp1) = 0xff000000 + RGB_JOIN(r,g,b); dp1 += 4; y = RZ(YMUL) * ((*yp1++) - 16); r = LUT_CLIP((y + (_crv * v)) >> RES); g = LUT_CLIP((y - (_cgu * u) - (_cgv * v) + RZ(OFF)) >> RES); b = LUT_CLIP((y + (_cbu * u) + RZ(OFF)) >> RES); *((DATA32 *) dp1) = 0xff000000 + RGB_JOIN(r,g,b); dp1 += 4; } } }
int bsad_mmx(uint8_t *pf, uint8_t *pb, uint8_t *p2, int lx, int hxf, int hyf, int hxb, int hyb, int h) { uint8_t *pfa,*pfb,*pfc,*pba,*pbb,*pbc; int s, s1, s2; pfa = pf + hxf; pfb = pf + lx * hyf; pfc = pfb + hxf; pba = pb + hxb; pbb = pb + lx * hyb; pbc = pbb + hxb; s = 0; /* the accumulator */ if (h > 0) { pxor_r2r(mm7, mm7); pxor_r2r(mm6, mm6); pcmpeqw_r2r(mm5, mm5); psubw_r2r(mm5, mm6); psllw_i2r(1, mm6); do { BSAD_LOAD(pf[0],mm0,mm1); BSAD_LOAD_ACC(pfa[0],mm2,mm3,mm0,mm1); BSAD_LOAD_ACC(pfb[0],mm2,mm3,mm0,mm1); BSAD_LOAD_ACC(pfc[0],mm2,mm3,mm0,mm1); paddw_r2r(mm6, mm0); paddw_r2r(mm6, mm1); psrlw_i2r(2, mm0); psrlw_i2r(2, mm1); BSAD_LOAD(pb[0],mm2,mm3); BSAD_LOAD_ACC(pba[0],mm4,mm5,mm2,mm3); BSAD_LOAD_ACC(pbb[0],mm4,mm5,mm2,mm3); BSAD_LOAD_ACC(pbc[0],mm4,mm5,mm2,mm3); paddw_r2r(mm6, mm2); paddw_r2r(mm6, mm3); psrlw_i2r(2, mm2); psrlw_i2r(2, mm3); paddw_r2r(mm2, mm0); paddw_r2r(mm3, mm1); psrlw_i2r(1, mm6); paddw_r2r(mm6, mm0); paddw_r2r(mm6, mm1); psllw_i2r(1, mm6); psrlw_i2r(1, mm0); psrlw_i2r(1, mm1); packuswb_r2r(mm1, mm0); movq_m2r(p2[0], mm1); movq_r2r(mm0, mm2); psubusb_r2r(mm1, mm0); psubusb_r2r(mm2, mm1); por_r2r(mm1, mm0); movq_r2r(mm0, mm1); punpcklbw_r2r(mm7, mm0); punpckhbw_r2r(mm7, mm1); paddw_r2r(mm1, mm0); movq_r2r(mm0, mm1); punpcklwd_r2r(mm7, mm0); punpckhwd_r2r(mm7, mm1); paddd_r2r(mm1, mm0); movd_r2g(mm0, s1); psrlq_i2r(32, mm0); movd_r2g(mm0, s2); s += s1 + s2; BSAD_LOAD(pf[8],mm0,mm1); BSAD_LOAD_ACC(pfa[8],mm2,mm3,mm0,mm1); BSAD_LOAD_ACC(pfb[8],mm2,mm3,mm0,mm1); BSAD_LOAD_ACC(pfc[8],mm2,mm3,mm0,mm1); paddw_r2r(mm6, mm0); paddw_r2r(mm6, mm1); psrlw_i2r(2, mm0); psrlw_i2r(2, mm1); BSAD_LOAD(pb[8],mm2,mm3); BSAD_LOAD_ACC(pba[8],mm4,mm5,mm2,mm3); BSAD_LOAD_ACC(pbb[8],mm4,mm5,mm2,mm3); BSAD_LOAD_ACC(pbc[8],mm4,mm5,mm2,mm3); paddw_r2r(mm6, mm2); paddw_r2r(mm6, mm3); psrlw_i2r(2, mm2); psrlw_i2r(2, mm3); paddw_r2r(mm2, mm0); paddw_r2r(mm3, mm1); psrlw_i2r(1, mm6); paddw_r2r(mm6, mm0); paddw_r2r(mm6, mm1); psllw_i2r(1, mm6); psrlw_i2r(1, mm0); psrlw_i2r(1, mm1); packuswb_r2r(mm1, mm0); movq_m2r(p2[8], mm1); movq_r2r(mm0, mm2); psubusb_r2r(mm1, mm0); psubusb_r2r(mm2, mm1); por_r2r(mm1, mm0); movq_r2r(mm0, mm1); punpcklbw_r2r(mm7, mm0); punpckhbw_r2r(mm7, mm1); paddw_r2r(mm1, mm0); movq_r2r(mm0, mm1); punpcklwd_r2r(mm7, mm0); punpckhwd_r2r(mm7, mm1); paddd_r2r(mm1, mm0); movd_r2g(mm0, s1); psrlq_i2r(32, mm0); movd_r2g(mm0, s2); s += s1 + s2; p2 += lx; pf += lx; pfa += lx; pfb += lx; pfc += lx; pb += lx; pba += lx; pbb += lx; pbc += lx; h--; } while (h > 0); } emms(); return s; }
void yuv411planar_to_rgb_mmx (const unsigned char *yuv, unsigned char *rgb, unsigned int w, unsigned int h) { unsigned int xx, yy; register const unsigned char *yp1, *up, *vp; unsigned char *dp1; /* plane pointers */ yp1 = yuv; up = yuv + (w * h); vp = up + (w * (h / 4)); /* destination pointers */ dp1 = rgb; yp1 = yuv; up = yuv + (w * h); vp = up + ((w / 2) * (h / 2)); dp1 = rgb; for (yy = 0; yy < h; yy++) { for (xx = 0; xx < w; xx += 8) { movq_m2r(*yp1, mm0); movq_r2r(mm0, mm1); psrlw_i2r(8, mm0); psllw_i2r(8, mm1); psrlw_i2r(8, mm1); pxor_r2r(mm7, mm7); movd_m2r(*up, mm3); movd_m2r(*vp, mm2); punpcklbw_r2r(mm7, mm2); punpcklbw_r2r(mm7, mm3); movq_m2r(CONST_16, mm4); psubsw_r2r(mm4, mm0); psubsw_r2r(mm4, mm1); movq_m2r(CONST_128, mm5); psubsw_r2r(mm5, mm2); psubsw_r2r(mm5, mm3); movq_m2r(CONST_YMUL, mm4); pmullw_r2r(mm4, mm0); pmullw_r2r(mm4, mm1); movq_m2r(CONST_CRVCRV, mm7); pmullw_r2r(mm3, mm7); movq_m2r(CONST_CBUCBU, mm6); pmullw_r2r(mm2, mm6); movq_m2r(CONST_CGUCGU, mm5); pmullw_r2r(mm2, mm5); movq_m2r(CONST_CGVCGV, mm4); pmullw_r2r(mm3, mm4); movq_r2r(mm0, mm2); paddsw_r2r(mm7, mm2); paddsw_r2r(mm1, mm7); psraw_i2r(RES, mm2); psraw_i2r(RES, mm7); packuswb_r2r(mm7, mm2); pxor_r2r(mm7, mm7); movq_r2r(mm2, mm3); punpckhbw_r2r(mm7, mm2); punpcklbw_r2r(mm3, mm7); por_r2r(mm7, mm2); movq_r2r(mm0, mm3); psubsw_r2r(mm5, mm3); psubsw_r2r(mm4, mm3); paddsw_m2r(CONST_32, mm3); movq_r2r(mm1, mm7); psubsw_r2r(mm5, mm7); psubsw_r2r(mm4, mm7); paddsw_m2r(CONST_32, mm7); psraw_i2r(RES, mm3); psraw_i2r(RES, mm7); packuswb_r2r(mm7, mm3); pxor_r2r(mm7, mm7); movq_r2r(mm3, mm4); punpckhbw_r2r(mm7, mm3); punpcklbw_r2r(mm4, mm7); por_r2r(mm7, mm3); movq_m2r(CONST_32, mm4); paddsw_r2r(mm6, mm0); paddsw_r2r(mm6, mm1); paddsw_r2r(mm4, mm0); paddsw_r2r(mm4, mm1); psraw_i2r(RES, mm0); psraw_i2r(RES, mm1); packuswb_r2r(mm1, mm0); pxor_r2r(mm7, mm7); movq_r2r(mm0, mm5); punpckhbw_r2r(mm7, mm0); punpcklbw_r2r(mm5, mm7); por_r2r(mm7, mm0); pxor_r2r(mm1, mm1); movq_r2r(mm0, mm5); movq_r2r(mm3, mm6); movq_r2r(mm2, mm7); punpckhbw_r2r(mm3, mm2); punpcklbw_r2r(mm6, mm7); punpckhbw_r2r(mm1, mm0); punpcklbw_r2r(mm1, mm5); movq_r2r(mm7, mm1); punpckhwd_r2r(mm5, mm7); punpcklwd_r2r(mm5, mm1); movq_r2r(mm2, mm4); punpckhwd_r2r(mm0, mm2); punpcklwd_r2r(mm0, mm4); movntq_r2m(mm1, *(dp1)); movntq_r2m(mm7, *(dp1 + 8)); movntq_r2m(mm4, *(dp1 + 16)); movntq_r2m(mm2, *(dp1 + 24)); yp1 += 8; up += 4; vp += 4; dp1 += 8 * 4; } if (yy & 0x1) { up -= w / 2; vp -= w / 2; } } emms(); }