unsigned int interpolhline_mmx_2(unsigned char* image){
__m64 mm_A = _mm_set_pi16(image[1],image[0],image[-1],image[-2]);
__m64 mm_B = _mm_set_pi16(image[2],image[1],image[0],image[-1]);
__m64 mm_C = _mm_set_pi16(image[3],image[2],image[1],image[0]);
__m64 mm_D = _mm_set_pi16(image[4],image[3],image[2],image[1]);
__m64 mm_E = _mm_set_pi16(image[5],image[4],image[3],image[2]);
__m64 mm_F = _mm_set_pi16(image[6],image[5],image[4],image[3]);
__m64 mm_AF = _mm_add_pi16(mm_A,mm_F);//A + F
__m64 mm_inter0 = _mm_add_pi16(mm_AF,_mm_set_pi16(16,16,16,16));//A + F + 16
__m64 mm_BE = _mm_add_pi16(mm_B,mm_E);//B + E
__m64 mm_CD = _mm_add_pi16(mm_C,mm_D);//C + D
__m64 mm_CDS = _mm_slli_pi16(mm_CD,2);//(C + D) << 2
__m64 mm_inter1 = _mm_sub_pi16(mm_CDS,mm_BE);//((C + D) << 2)-(B + E)
__m64 mm_5 = _mm_set_pi16(5,5,5,5);
__m64 mm_inter_3 = _mm_mullo_pi16(mm_inter1, mm_5);//(((C + D) << 2)-(B + E))*5
__m64 mm_result = _mm_add_pi16(mm_inter_3,mm_inter0);//A + F + 16 + (((C + D) << 2)-(B + E))*5
__m64 mm_zero = _mm_setzero_si64();
__m64 mm_clip = _mm_max_pi16(mm_result,mm_zero);//Clip with 0
__m64 mm_ret = _mm_srai_pi16(mm_clip,5);
__m64 mm_clip1 = _mm_min_pi16(mm_ret,_mm_set_pi16(255,255,255,255)); //Clip with 255
__m64 result =_mm_packs_pu16(mm_clip1,mm_zero);
unsigned int ret = _mm_cvtsi64_si32(result);
empty();
return ret;
}
unsigned int interpolvline_mmx_3(unsigned char* image, int PicWidthInPix){
__m64 mm_A = _mm_set_pi16(image[-2 * PicWidthInPix + 3],image[-2 * PicWidthInPix + 2],image[-2 * PicWidthInPix + 1],image[-2 * PicWidthInPix]);
__m64 mm_B = _mm_set_pi16(image[-1 * PicWidthInPix + 3],image[-1 * PicWidthInPix + 2],image[-1 * PicWidthInPix + 1],image[-1 * PicWidthInPix]);
__m64 mm_C = _mm_set_pi16(image[3],image[2],image[1],image[0]);
__m64 mm_D = _mm_set_pi16(image[1 * PicWidthInPix + 3],image[1 * PicWidthInPix + 2],image[1 * PicWidthInPix + 1],image[1 * PicWidthInPix]);
__m64 mm_E = _mm_set_pi16(image[2 * PicWidthInPix + 3],image[2 * PicWidthInPix + 2],image[2 * PicWidthInPix + 1],image[2 * PicWidthInPix]);
__m64 mm_F = _mm_set_pi16(image[3 * PicWidthInPix + 3],image[3 * PicWidthInPix + 2],image[3 * PicWidthInPix + 1],image[3 * PicWidthInPix]);
__m64 mm_AF = _mm_add_pi16(mm_A,mm_F);//A + F
__m64 mm_inter0 = _mm_add_pi16(mm_AF,_mm_set_pi16(16,16,16,16));//A + F + 16
__m64 mm_BE = _mm_add_pi16(mm_B,mm_E);//B + E
__m64 mm_CD = _mm_add_pi16(mm_C,mm_D);//C + D
__m64 mm_CDS = _mm_slli_pi16(mm_CD,2);//(C + D) << 2
__m64 mm_inter1 = _mm_sub_pi16(mm_CDS,mm_BE);//((C + D) << 2)-(B + E)
__m64 mm_5 = _mm_set_pi16(5,5,5,5);
__m64 mm_inter_3 = _mm_mullo_pi16(mm_inter1, mm_5);//(((C + D) << 2)-(B + E))*5
__m64 mm_result = _mm_add_pi16(mm_inter_3,mm_inter0);//A + F + 16 + (((C + D) << 2)-(B + E))*5
__m64 mm_zero = _mm_setzero_si64();
__m64 mm_clip = _mm_max_pi16(mm_result,mm_zero);//Clip with 0
__m64 mm_ret = _mm_srai_pi16(mm_clip,5);
__m64 mm_clip1 = _mm_min_pi16(mm_ret,_mm_set_pi16(255,255,255,255)); //Clip with 255
__m64 test = _mm_avg_pu8(mm_clip1,mm_D);//(ptr_img[0] + ptr_rf[0] + 1) >> 1
__m64 test1 =_mm_packs_pu16(test,mm_zero);
unsigned int ret = _mm_cvtsi64_si32(test1);
empty();
return ret;
}
__m64 interpolhline64_3_mmx(__m64* temp){
__m64 res,res1;
__m64 ptr = _mm_setzero_si64();
__m64 mm_16 = _mm_set_pi16(16,16,16,16);
short A = _mm_extract_pi16(temp[0],0);
short B = _mm_extract_pi16(temp[1],0);
short C = _mm_extract_pi16(temp[2],0);
short D = _mm_extract_pi16(temp[3],0);
short E = _mm_extract_pi16(temp[4],0);
short F = _mm_extract_pi16(temp[5],0);
unsigned int result = A + F - 5 * (short)(B + E) + 20 * (short)(C + D) + 512;
ptr = _mm_insert_pi16(ptr,CLIP255_16(result >> 10),0);
A = _mm_extract_pi16(temp[0],1);
B = _mm_extract_pi16(temp[1],1);
C = _mm_extract_pi16(temp[2],1);
D = _mm_extract_pi16(temp[3],1);
E = _mm_extract_pi16(temp[4],1);
F = _mm_extract_pi16(temp[5],1);
result = A + F - 5 * (short)(B + E) + 20 * (short)(C + D) + 512;
ptr = _mm_insert_pi16(ptr,CLIP255_16(result >> 10),1);
A = _mm_extract_pi16(temp[0],2);
B = _mm_extract_pi16(temp[1],2);
C = _mm_extract_pi16(temp[2],2);
D = _mm_extract_pi16(temp[3],2);
E = _mm_extract_pi16(temp[4],2);
F = _mm_extract_pi16(temp[5],2);
result = A + F - 5 * (short)(B + E) + 20 * (short)(C + D) + 512;
ptr = _mm_insert_pi16(ptr,CLIP255_16(result >> 10),2);
A = _mm_extract_pi16(temp[0],3);
B = _mm_extract_pi16(temp[1],3);
C = _mm_extract_pi16(temp[2],3);
D = _mm_extract_pi16(temp[3],3);
E = _mm_extract_pi16(temp[4],3);
F = _mm_extract_pi16(temp[5],3);
result = A + F - 5 * (short)(B + E) + 20 * (short)(C + D) + 512;
ptr = _mm_insert_pi16(ptr,CLIP255_16(result >> 10),3);
res = _mm_add_pi16(temp[3],mm_16);
res1 = _mm_srai_pi16(res,5);
res1 = _mm_max_pi16(res1,_mm_set_pi16(0,0,0,0));
res1 = _mm_min_pi16(res1,_mm_set_pi16(255,255,255,255)); //Clip
res = _mm_avg_pu16(ptr,res1);//(ptr_img[0] + ptr_rf[0] + 1) >> 1*/
return res;
}
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Convert YV12 to RGB16.
VOID Yv12ToRgb16_mmx(PBYTE pbDstX, INT iDstXStride,
PBYTE pbSrcY, PBYTE pbSrcU, PBYTE pbSrcV, INT iSrcYStride, INT iSrcUvStride,
UINT uWidth, INT iHeight)
{
UINT x;
INT y;
INT iDstXDif;
INT iSrcYDif;
INT iSrcUvDif;
INT yy, bu, guv, rv, r, g, b;
M64 y0, y1, u0, v0, mz;
M64 r0, g0, b0, r1, g1, b1;
M64 bu0, gu0, gv0, rv0, bu1, rv1, guv0, guv1;
if (iHeight < 0)
{
iHeight = -iHeight;
pbDstX += (iHeight - 1) * iDstXStride;
iDstXStride = -iDstXStride;
}
iDstXDif = iDstXStride - (uWidth * 2);
iSrcYDif = iSrcYStride - uWidth;
iSrcUvDif = iSrcUvStride - (uWidth / 2);
mz = _mm_setzero_si64();
for (y = iHeight / 2; y; y--)
{
for (x = uWidth / 8; x; x--)
{
// Calculate coefficient.
u0 = _mm_cvtsi32_si64(*((PDWORD) pbSrcU)); // [ | u3 u2 u1 u0]
v0 = _mm_cvtsi32_si64(*((PDWORD) pbSrcV)); // [ | v3 v2 v1 v0]
u0 = _mm_unpacklo_pi8(u0, mz); // u3 u2 u1 u0
v0 = _mm_unpacklo_pi8(v0, mz); // v3 v2 v1 v0
u0 = _mm_subs_pi16(u0, g_mSub80);
v0 = _mm_subs_pi16(v0, g_mSub80);
gu0 = _mm_mullo_pi16(u0, g_mUGMul);
gv0 = _mm_mullo_pi16(v0, g_mVGMul);
bu0 = _mm_mullo_pi16(u0, g_mUBMul);
rv0 = _mm_mullo_pi16(v0, g_mVRMul);
guv0 = _mm_adds_pi16(gu0, gv0);
guv1 = _mm_unpackhi_pi16(guv0, guv0); // guv3 guv3 guv2 guv2
guv0 = _mm_unpacklo_pi16(guv0, guv0); // guv1 guv1 guv0 guv0
bu1 = _mm_unpackhi_pi16(bu0, bu0); // bu3 bu3 bu2 bu2
bu0 = _mm_unpacklo_pi16(bu0, bu0); // bu1 bu1 bu0 bu0
rv1 = _mm_unpackhi_pi16(rv0, rv0); // rv3 rv3 rv2 rv2
rv0 = _mm_unpacklo_pi16(rv0, rv0); // rv1 rv1 rv0 rv0
// Process for row 0.
y0 = *((PM64) pbSrcY); // [YYYY | YYYY]; row 0
y1 = _mm_unpackhi_pi8(y0, mz); // y7 y6 y5 y4
y0 = _mm_unpacklo_pi8(y0, mz); // y3 y2 y1 y0
y1 = _mm_subs_pi16(y1, g_mSub10);
y0 = _mm_subs_pi16(y0, g_mSub10);
y1 = _mm_mullo_pi16(y1, g_mYYMul);
y0 = _mm_mullo_pi16(y0, g_mYYMul);
b1 = _mm_adds_pi16(y1, bu1);
b0 = _mm_adds_pi16(y0, bu0);
b1 = _mm_srai_pi16(b1, SCALEBITS); // 8 bits (0 - 7)
b0 = _mm_srai_pi16(b0, SCALEBITS);
b1 = _mm_packs_pu16(b1, mz); // 0 0 0 0 b7 b6 b5 b4
b0 = _mm_packs_pu16(b0, mz); // 0 0 0 0 b3 b2 b1 b0
b1 = _mm_unpacklo_pi8(b1, mz); // 0 b7 0b 6 0 b5 0 b4
b0 = _mm_unpacklo_pi8(b0, mz);
b1 = _mm_srli_pi16(b1, 3);
b0 = _mm_srli_pi16(b0, 3); // 5 bits (0 - 4)
g1 = _mm_subs_pi16(y1, guv1); // g7 g6 g5 g4
g0 = _mm_subs_pi16(y0, guv0); // g3 g2 g1 g0
g1 = _mm_srai_pi16(g1, SCALEBITS); // 8 bits (0 - 7)
g0 = _mm_srai_pi16(g0, SCALEBITS);
g1 = _mm_packs_pu16(g1, mz); // 0 0 0 0 g7 g6 g5 g4
g0 = _mm_packs_pu16(g0, mz); // 0 0 0 0 g3 g2 g1 g0
g1 = _mm_unpacklo_pi8(g1, mz); // 0 g7 0 g6 0 g5 0 g4
g0 = _mm_unpacklo_pi8(g0, mz);
g1 = _mm_srli_pi16(g1, 2); // 6 bits (0 - 5)
g0 = _mm_srli_pi16(g0, 2);
g1 = _mm_slli_pi16(g1, 5); // 6 bits (5 - 10)
g0 = _mm_slli_pi16(g0, 5); // 6 bits (5 - 10)
r1 = _mm_adds_pi16(y1, rv1);
r0 = _mm_adds_pi16(y0, rv0);
r1 = _mm_srai_pi16(r1, SCALEBITS);
r0 = _mm_srai_pi16(r0, SCALEBITS);
r1 = _mm_packs_pu16(r1, mz); // 0 0 0 0 r7 r6 r5 r4
r0 = _mm_packs_pu16(r0, mz); // 0 0 0 0 r3 r2 r1 r0
r1 = _mm_unpacklo_pi8(r1, mz); // 0 r7 0 r6 0 r5 0 r4
r0 = _mm_unpacklo_pi8(r0, mz);
r1 = _mm_srli_pi16(r1, 3); // 5 bits (0 - 4)
r0 = _mm_srli_pi16(r0, 3);
r1 = _mm_slli_pi16(r1, 11); // 5 bits (11 - 15)
r0 = _mm_slli_pi16(r0, 11); // 5 bits (11 - 15)
// Combine RGB.
b0 = _mm_or_si64(g0, b0);
b0 = _mm_or_si64(r0, b0); // 16 bits rgb
b1 = _mm_or_si64(g1, b1);
b1 = _mm_or_si64(r1, b1); // 16 bits rgb
// Write out row 0.
((PM64) pbDstX)[0] = b0;
((PM64) pbDstX)[1] = b1;
// Process for row 1.
y0 = *((PM64) (pbSrcY + iSrcYStride)); // [YYYY | YYYY]; row 0
y1 = _mm_unpackhi_pi8(y0, mz); // y7 y6 y5 y4
y0 = _mm_unpacklo_pi8(y0, mz); // y3 y2 y1 y0
y1 = _mm_subs_pi16(y1, g_mSub10);
y0 = _mm_subs_pi16(y0, g_mSub10);
y1 = _mm_mullo_pi16(y1, g_mYYMul);
y0 = _mm_mullo_pi16(y0, g_mYYMul);
b1 = _mm_adds_pi16(y1, bu1);
b0 = _mm_adds_pi16(y0, bu0);
b1 = _mm_srai_pi16(b1, SCALEBITS); // 8 bits (0 - 7)
b0 = _mm_srai_pi16(b0, SCALEBITS);
b1 = _mm_packs_pu16(b1, mz); // 0 0 0 0 b7 b6 b5 b4
b0 = _mm_packs_pu16(b0, mz); // 0 0 0 0 b3 b2 b1 b0
b1 = _mm_unpacklo_pi8(b1, mz); // 0 b7 0b 6 0 b5 0 b4
b0 = _mm_unpacklo_pi8(b0, mz);
b1 = _mm_srli_pi16(b1, 3);
b0 = _mm_srli_pi16(b0, 3); // 5 bits (0 - 4)
g1 = _mm_subs_pi16(y1, guv1); // g7 g6 g5 g4
g0 = _mm_subs_pi16(y0, guv0); // g3 g2 g1 g0
g1 = _mm_srai_pi16(g1, SCALEBITS); // 8 bits (0 - 7)
g0 = _mm_srai_pi16(g0, SCALEBITS);
g1 = _mm_packs_pu16(g1, mz); // 0 0 0 0 g7 g6 g5 g4
g0 = _mm_packs_pu16(g0, mz); // 0 0 0 0 g3 g2 g1 g0
g1 = _mm_unpacklo_pi8(g1, mz); // 0 g7 0 g6 0 g5 0 g4
g0 = _mm_unpacklo_pi8(g0, mz);
g1 = _mm_srli_pi16(g1, 2); // 6 bits (0 - 5)
g0 = _mm_srli_pi16(g0, 2);
g1 = _mm_slli_pi16(g1, 5); // 6 bits (5 - 10)
g0 = _mm_slli_pi16(g0, 5); // 6 bits (5 - 10)
r1 = _mm_adds_pi16(y1, rv1);
r0 = _mm_adds_pi16(y0, rv0);
r1 = _mm_srai_pi16(r1, SCALEBITS);
r0 = _mm_srai_pi16(r0, SCALEBITS);
r1 = _mm_packs_pu16(r1, mz); // 0 0 0 0 r7 r6 r5 r4
r0 = _mm_packs_pu16(r0, mz); // 0 0 0 0 r3 r2 r1 r0
r1 = _mm_unpacklo_pi8(r1, mz); // 0 r7 0 r6 0 r5 0 r4
r0 = _mm_unpacklo_pi8(r0, mz);
r1 = _mm_srli_pi16(r1, 3); // 5 bits (0 - 4)
r0 = _mm_srli_pi16(r0, 3);
r1 = _mm_slli_pi16(r1, 11); // 5 bits (11 - 15)
r0 = _mm_slli_pi16(r0, 11); // 5 bits (11 - 15)
// Combine RGB.
b0 = _mm_or_si64(g0, b0);
b0 = _mm_or_si64(r0, b0); // 16 bits rgb
b1 = _mm_or_si64(g1, b1);
b1 = _mm_or_si64(r1, b1); // 16 bits rgb
// Write out row 1.
((PM64) (pbDstX + iDstXStride))[0] = b0;
((PM64) (pbDstX + iDstXStride))[1] = b1;
pbDstX += 16;
pbSrcY += 8;
pbSrcU += 4;
pbSrcV += 4;
}
for (x = (uWidth & 7) / 2; x; x--)
{
bu = g_iBUTab[pbSrcU[0]];
guv = g_iGUTab[pbSrcU[0]] + g_iGVTab[pbSrcV[0]];
rv = g_iRVTab[pbSrcV[0]];
yy = g_iYYTab[pbSrcY[0]];
b = _Clip(((yy + bu) >> SCALEBITS_OUT));
g = _Clip(((yy - guv) >> SCALEBITS_OUT));
r = _Clip(((yy + rv) >> SCALEBITS_OUT));
((PWORD) pbDstX)[0] = _MakeRgb16(r, g, b);
yy = g_iYYTab[pbSrcY[1]];
b = _Clip(((yy + bu) >> SCALEBITS_OUT));
g = _Clip(((yy - guv) >> SCALEBITS_OUT));
r = _Clip(((yy + rv) >> SCALEBITS_OUT));
((PWORD) pbDstX)[1] = _MakeRgb16(r, g, b);
yy = g_iYYTab[pbSrcY[iSrcYStride]];
b = _Clip(((yy + bu) >> SCALEBITS_OUT));
g = _Clip(((yy - guv) >> SCALEBITS_OUT));
r = _Clip(((yy + rv) >> SCALEBITS_OUT));
((PWORD) (pbDstX + iDstXStride))[0] = _MakeRgb16(r, g, b);
yy = g_iYYTab[pbSrcY[iSrcYStride + 1]];
b = _Clip(((yy + bu) >> SCALEBITS_OUT));
g = _Clip(((yy - guv) >> SCALEBITS_OUT));
r = _Clip(((yy + rv) >> SCALEBITS_OUT));
((PWORD) (pbDstX + iDstXStride))[1] = _MakeRgb16(r, g, b);
pbDstX += 4;
pbSrcY += 2;
pbSrcU++;
pbSrcV++;
}
pbDstX += iDstXDif + iDstXStride;
pbSrcY += iSrcYDif + iSrcYStride;
pbSrcU += iSrcUvDif;
pbSrcV += iSrcUvDif;
}
_mm_empty();
}
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Convert YV12 to RGB24.
VOID Yv12ToRgb24_mmx(PBYTE pbDstX, INT iDstXStride,
PBYTE pbSrcY, PBYTE pbSrcU, PBYTE pbSrcV, INT iSrcYStride, INT iSrcUvStride,
UINT uWidth, INT iHeight)
{
UINT x;
INT y;
INT iDstXDif;
INT iSrcYDif;
INT iSrcUvDif;
INT yy, bu, guv, rv;
M64 y0, y1, u0, v0, mz;
M64 r0, g0, b0, r1, g1, b1;
M64 rgb0, rgb1, rgb2, rgb3;
M64 bu0, gu0, gv0, rv0, bu1, rv1, guv0, guv1;
if (iHeight < 0)
{
iHeight = -iHeight;
pbDstX += (iHeight - 1) * iDstXStride;
iDstXStride = -iDstXStride;
}
iDstXDif = iDstXStride - (uWidth * 3);
iSrcYDif = iSrcYStride - uWidth;
iSrcUvDif = iSrcUvStride - (uWidth / 2);
mz = _mm_setzero_si64();
for (y = iHeight / 2; y; y--)
{
for (x = uWidth / 8; x; x--)
{
// Calculate coefficient.
u0 = _mm_cvtsi32_si64(*((PDWORD) pbSrcU)); // [ | u3 u2 u1 u0]
v0 = _mm_cvtsi32_si64(*((PDWORD) pbSrcV)); // [ | v3 v2 v1 v0]
u0 = _mm_unpacklo_pi8(u0, mz); // u3 u2 u1 u0
v0 = _mm_unpacklo_pi8(v0, mz); // v3 v2 v1 v0
u0 = _mm_subs_pi16(u0, g_mSub80);
v0 = _mm_subs_pi16(v0, g_mSub80);
gu0 = _mm_mullo_pi16(u0, g_mUGMul);
gv0 = _mm_mullo_pi16(v0, g_mVGMul);
bu0 = _mm_mullo_pi16(u0, g_mUBMul);
rv0 = _mm_mullo_pi16(v0, g_mVRMul);
guv0 = _mm_adds_pi16(gu0, gv0);
guv1 = _mm_unpackhi_pi16(guv0, guv0); // guv3 guv3 guv2 guv2
guv0 = _mm_unpacklo_pi16(guv0, guv0); // guv1 guv1 guv0 guv0
bu1 = _mm_unpackhi_pi16(bu0, bu0); // bu3 bu3 bu2 bu2
bu0 = _mm_unpacklo_pi16(bu0, bu0); // bu1 bu1 bu0 bu0
rv1 = _mm_unpackhi_pi16(rv0, rv0); // rv3 rv3 rv2 rv2
rv0 = _mm_unpacklo_pi16(rv0, rv0); // rv1 rv1 rv0 rv0
// Process for row 0.
y0 = *((PM64) pbSrcY); // [YYYY | YYYY]; row 0
y1 = _mm_unpackhi_pi8(y0, mz); // y7 y6 y5 y4
y0 = _mm_unpacklo_pi8(y0, mz); // y3 y2 y1 y0
y1 = _mm_subs_pi16(y1, g_mSub10);
y0 = _mm_subs_pi16(y0, g_mSub10);
y1 = _mm_mullo_pi16(y1, g_mYYMul);
y0 = _mm_mullo_pi16(y0, g_mYYMul);
g1 = _mm_subs_pi16(y1, guv1); // g7 g6 g5 g4
g0 = _mm_subs_pi16(y0, guv0); // g3 g2 g1 g0
g1 = _mm_srai_pi16(g1, SCALEBITS);
g0 = _mm_srai_pi16(g0, SCALEBITS);
g0 = _mm_packs_pu16(g0, g1); // g7 g6 ...g1 g0
b1 = _mm_adds_pi16(y1, bu1);
b0 = _mm_adds_pi16(y0, bu0);
b1 = _mm_srai_pi16(b1, SCALEBITS);
b0 = _mm_srai_pi16(b0, SCALEBITS);
b0 = _mm_packs_pu16(b0, b1);
r1 = _mm_adds_pi16(y1, rv1);
r0 = _mm_adds_pi16(y0, rv0);
r1 = _mm_srai_pi16(r1, SCALEBITS);
r0 = _mm_srai_pi16(r0, SCALEBITS);
r0 = _mm_packs_pu16(r0, r1);
r1 = _mm_unpackhi_pi8(b0, r0); // r7 b7 r6 b6 r5 b5 r4 b4
r0 = _mm_unpacklo_pi8(b0, r0); // r3 b3 r2 b2 r1 b1 r0 b0
g1 = _mm_unpackhi_pi8(g0, mz); // 0 g7 0 g6 0 g5 0 g4
g0 = _mm_unpacklo_pi8(g0, mz); // 0 g3 0 g2 0 g1 0 g0
rgb0 = _mm_unpacklo_pi8(r0, g0); // 0 r1 g1 b1 0 r0 g0 b0
rgb1 = _mm_unpackhi_pi8(r0, g0); // 0 r3 g3 b3 0 r2 g2 b2
rgb2 = _mm_unpacklo_pi8(r1, g1); // 0 r5 g5 b5 0 r4 g4 b4
rgb3 = _mm_unpackhi_pi8(r1, g1); // 0 r7 g7 b7 0 r6 g6 b6
// Write out row 0.
*((PDWORD) (pbDstX + 0)) = _mm_cvtsi64_si32(rgb0); rgb0 = _mm_srli_si64(rgb0, 32);
*((PDWORD) (pbDstX + 3)) = _mm_cvtsi64_si32(rgb0);
*((PDWORD) (pbDstX + 6)) = _mm_cvtsi64_si32(rgb1); rgb1 = _mm_srli_si64(rgb1, 32);
*((PDWORD) (pbDstX + 9)) = _mm_cvtsi64_si32(rgb1);
*((PDWORD) (pbDstX + 12)) = _mm_cvtsi64_si32(rgb2); rgb2 = _mm_srli_si64(rgb2, 32);
*((PDWORD) (pbDstX + 15)) = _mm_cvtsi64_si32(rgb2);
*((PDWORD) (pbDstX + 18)) = _mm_cvtsi64_si32(rgb3); rgb3 = _mm_srli_si64(rgb3, 32);
*((PDWORD) (pbDstX + 21)) = _mm_cvtsi64_si32(rgb3);
// Process for row 1.
y0 = *((PM64) (pbSrcY + iSrcYStride)); // [YYYY | YYYY]; row 1
y1 = _mm_unpackhi_pi8(y0, mz); // y7 y6 y5 y4
y0 = _mm_unpacklo_pi8(y0, mz); // y3 y2 y1 y0
y1 = _mm_subs_pi16(y1, g_mSub10);
y0 = _mm_subs_pi16(y0, g_mSub10);
y1 = _mm_mullo_pi16(y1, g_mYYMul);
y0 = _mm_mullo_pi16(y0, g_mYYMul);
g1 = _mm_subs_pi16(y1, guv1); // g7 g6 g5 g4
g0 = _mm_subs_pi16(y0, guv0); // g3 g2 g1 g0
g1 = _mm_srai_pi16(g1, SCALEBITS);
g0 = _mm_srai_pi16(g0, SCALEBITS);
g0 = _mm_packs_pu16(g0, g1); // g7 g6 ...g1 g0
b1 = _mm_adds_pi16(y1, bu1);
b0 = _mm_adds_pi16(y0, bu0);
b1 = _mm_srai_pi16(b1, SCALEBITS);
b0 = _mm_srai_pi16(b0, SCALEBITS);
b0 = _mm_packs_pu16(b0, b1);
r1 = _mm_adds_pi16(y1, rv1);
r0 = _mm_adds_pi16(y0, rv0);
r1 = _mm_srai_pi16(r1, SCALEBITS);
r0 = _mm_srai_pi16(r0, SCALEBITS);
r0 = _mm_packs_pu16(r0, r1);
r1 = _mm_unpackhi_pi8(b0, r0); // r7 b7 r6 b6 r5 b5 r4 b4
r0 = _mm_unpacklo_pi8(b0, r0); // r3 b3 r2 b2 r1 b1 r0 b0
g1 = _mm_unpackhi_pi8(g0, mz); // 0 g7 0 g6 0 g5 0 g4
g0 = _mm_unpacklo_pi8(g0, mz); // 0 g3 0 g2 0 g1 0 g0
rgb0 = _mm_unpacklo_pi8(r0, g0); // 0 r1 g1 b1 0 r0 g0 b0
rgb1 = _mm_unpackhi_pi8(r0, g0); // 0 r3 g3 b3 0 r2 g2 b2
rgb2 = _mm_unpacklo_pi8(r1, g1); // 0 r5 g5 b5 0 r4 g4 b4
rgb3 = _mm_unpackhi_pi8(r1, g1); // 0 r7 g7 b7 0 r6 g6 b6
// Write out row 1.
*((PDWORD) (pbDstX + iDstXStride + 0)) = _mm_cvtsi64_si32(rgb0); rgb0 = _mm_srli_si64(rgb0, 32);
*((PDWORD) (pbDstX + iDstXStride + 3)) = _mm_cvtsi64_si32(rgb0);
*((PDWORD) (pbDstX + iDstXStride + 6)) = _mm_cvtsi64_si32(rgb1); rgb1 = _mm_srli_si64(rgb1, 32);
*((PDWORD) (pbDstX + iDstXStride + 9)) = _mm_cvtsi64_si32(rgb1);
*((PDWORD) (pbDstX + iDstXStride + 12)) = _mm_cvtsi64_si32(rgb2); rgb2 = _mm_srli_si64(rgb2, 32);
*((PDWORD) (pbDstX + iDstXStride + 15)) = _mm_cvtsi64_si32(rgb2);
*((PDWORD) (pbDstX + iDstXStride + 18)) = _mm_cvtsi64_si32(rgb3); rgb3 = _mm_srli_si64(rgb3, 32);
*((PDWORD) (pbDstX + iDstXStride + 21)) = _mm_cvtsi64_si32(rgb3);
pbDstX += 24;
pbSrcY += 8;
pbSrcU += 4;
pbSrcV += 4;
}
for (x = (uWidth & 7) / 2; x; x--)
{
bu = g_iBUTab[pbSrcU[0]];
guv = g_iGUTab[pbSrcU[0]] + g_iGVTab[pbSrcV[0]];
rv = g_iRVTab[pbSrcV[0]];
yy = g_iYYTab[pbSrcY[0]];
pbDstX[0] = _Clip((yy + bu) >> SCALEBITS_OUT);
pbDstX[1] = _Clip((yy - guv) >> SCALEBITS_OUT);
pbDstX[2] = _Clip((yy + rv) >> SCALEBITS_OUT);
yy = g_iYYTab[pbSrcY[1]];
pbDstX[3] = _Clip((yy + bu) >> SCALEBITS_OUT);
pbDstX[4] = _Clip((yy - guv) >> SCALEBITS_OUT);
pbDstX[5] = _Clip((yy + rv) >> SCALEBITS_OUT);
yy = g_iYYTab[pbSrcY[iSrcYStride]];
pbDstX[iDstXStride + 0] = _Clip((yy + bu) >> SCALEBITS_OUT);
pbDstX[iDstXStride + 1] = _Clip((yy - guv) >> SCALEBITS_OUT);
pbDstX[iDstXStride + 2] = _Clip((yy + rv) >> SCALEBITS_OUT);
yy = g_iYYTab[pbSrcY[iSrcYStride + 1]];
pbDstX[iDstXStride + 3] = _Clip((yy + bu) >> SCALEBITS_OUT);
pbDstX[iDstXStride + 4] = _Clip((yy - guv) >> SCALEBITS_OUT);
pbDstX[iDstXStride + 5] = _Clip((yy + rv) >> SCALEBITS_OUT);
pbDstX += 6;
pbSrcY += 2;
pbSrcU++;
pbSrcV++;
}
pbDstX += iDstXDif + iDstXStride;
pbSrcY += iSrcYDif + iSrcYStride;
pbSrcU += iSrcUvDif;
pbSrcV += iSrcUvDif;
}
_mm_empty();
}
__m64 test73(__m64 a) {
// CHECK: psraw
return _mm_srai_pi16(a, 3);
}
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Convert YUY2 to RGB24.
VOID Yuy2ToRgb24_mmx(PBYTE pbDstX, INT iDstXStride, PBYTE pbSrcX, INT iSrcXStride, UINT uWidth, INT iHeight)
{
UINT x;
INT y;
INT iDstXDif;
INT iSrcXDif;
INT yy, bu, guv, rv;
M64 y0, y1, u0, v0, uv_temp1, uv_temp2, mz;
M64 r0, g0, b0, r1, g1, b1;
M64 rgb0, rgb1, rgb2, rgb3;
M64 bu0, gu0, gv0, rv0, bu1, rv1, guv0, guv1;
if (iHeight < 0)
{
iHeight = -iHeight;
pbSrcX += (iHeight - 1) * iSrcXStride;
iSrcXStride = -iSrcXStride;
}
iDstXDif = iDstXStride - (uWidth * 3);
iSrcXDif = iSrcXStride - (uWidth * 2);
mz = _mm_setzero_si64();
for (y = iHeight; y; y--)
{
for (x = uWidth / 8; x; x--)
{
y0 = ((PM64) pbSrcX)[0];
y1 = ((PM64) pbSrcX)[1];
u0 = y0;
v0 = y1;
y0 = _mm_and_si64(y0, g_mWord00FF);
y1 = _mm_and_si64(y1, g_mWord00FF);
u0 = _mm_srli_pi16(u0, 8);
v0 = _mm_srli_pi16(v0, 8);
uv_temp1 = _mm_srli_pi32(u0, 16);
u0 = _mm_slli_pi32(u0, 16);
u0 = _mm_srli_pi32(u0, 16);
uv_temp2 = _mm_srli_pi32(v0, 16);
v0 = _mm_slli_pi32(v0, 16);
v0 = _mm_srli_pi32(v0, 16);
u0 = _mm_packs_pi32(u0, v0);
v0 = _mm_packs_pi32(uv_temp1, uv_temp2);
// Calculate coefficient.
u0 = _mm_subs_pi16(u0, g_mSub80);
v0 = _mm_subs_pi16(v0, g_mSub80);
gu0 = _mm_mullo_pi16(u0, g_mUGMul);
gv0 = _mm_mullo_pi16(v0, g_mVGMul);
bu0 = _mm_mullo_pi16(u0, g_mUBMul);
rv0 = _mm_mullo_pi16(v0, g_mVRMul);
guv0 = _mm_adds_pi16(gu0, gv0);
guv1 = _mm_unpackhi_pi16(guv0, guv0); // guv3 guv3 guv2 guv2
guv0 = _mm_unpacklo_pi16(guv0, guv0); // guv1 guv1 guv0 guv0
bu1 = _mm_unpackhi_pi16(bu0, bu0); // bu3 bu3 bu2 bu2
bu0 = _mm_unpacklo_pi16(bu0, bu0); // bu1 bu1 bu0 bu0
rv1 = _mm_unpackhi_pi16(rv0, rv0); // rv3 rv3 rv2 rv2
rv0 = _mm_unpacklo_pi16(rv0, rv0); // rv1 rv1 rv0 rv0
// Process for row 0.
y1 = _mm_subs_pi16(y1, g_mSub10);
y0 = _mm_subs_pi16(y0, g_mSub10);
y1 = _mm_mullo_pi16(y1, g_mYYMul);
y0 = _mm_mullo_pi16(y0, g_mYYMul);
g1 = _mm_subs_pi16(y1, guv1); // g7 g6 g5 g4
g0 = _mm_subs_pi16(y0, guv0); // g3 g2 g1 g0
g1 = _mm_srai_pi16(g1, SCALEBITS);
g0 = _mm_srai_pi16(g0, SCALEBITS);
g0 = _mm_packs_pu16(g0, g1); // g7 g6 ...g1 g0
b1 = _mm_adds_pi16(y1, bu1);
b0 = _mm_adds_pi16(y0, bu0);
b1 = _mm_srai_pi16(b1, SCALEBITS);
b0 = _mm_srai_pi16(b0, SCALEBITS);
b0 = _mm_packs_pu16(b0, b1);
r1 = _mm_adds_pi16(y1, rv1);
r0 = _mm_adds_pi16(y0, rv0);
r1 = _mm_srai_pi16(r1, SCALEBITS);
r0 = _mm_srai_pi16(r0, SCALEBITS);
r0 = _mm_packs_pu16(r0, r1);
r1 = _mm_unpackhi_pi8(b0, r0); // r7 b7 r6 b6 r5 b5 r4 b4
r0 = _mm_unpacklo_pi8(b0, r0); // r3 b3 r2 b2 r1 b1 r0 b0
g1 = _mm_unpackhi_pi8(g0, mz); // 0 g7 0 g6 0 g5 0 g4
g0 = _mm_unpacklo_pi8(g0, mz); // 0 g3 0 g2 0 g1 0 g0
rgb0 = _mm_unpacklo_pi8(r0, g0); // 0 r1 g1 b1 0 r0 g0 b0
rgb1 = _mm_unpackhi_pi8(r0, g0); // 0 r3 g3 b3 0 r2 g2 b2
rgb2 = _mm_unpacklo_pi8(r1, g1); // 0 r5 g5 b5 0 r4 g4 b4
rgb3 = _mm_unpackhi_pi8(r1, g1); // 0 r7 g7 b7 0 r6 g6 b6
// Write out row 0.
*((PDWORD) (pbDstX + 0)) = _mm_cvtsi64_si32(rgb0); rgb0 = _mm_srli_si64(rgb0, 32);
*((PDWORD) (pbDstX + 3)) = _mm_cvtsi64_si32(rgb0);
*((PDWORD) (pbDstX + 6)) = _mm_cvtsi64_si32(rgb1); rgb1 = _mm_srli_si64(rgb1, 32);
*((PDWORD) (pbDstX + 9)) = _mm_cvtsi64_si32(rgb1);
*((PDWORD) (pbDstX + 12)) = _mm_cvtsi64_si32(rgb2); rgb2 = _mm_srli_si64(rgb2, 32);
*((PDWORD) (pbDstX + 15)) = _mm_cvtsi64_si32(rgb2);
*((PDWORD) (pbDstX + 18)) = _mm_cvtsi64_si32(rgb3); rgb3 = _mm_srli_si64(rgb3, 32);
*((PDWORD) (pbDstX + 21)) = _mm_cvtsi64_si32(rgb3);
pbDstX += 24;
pbSrcX += 16;
}
for (x = (uWidth & 7) / 2; x; x--)
{
bu = g_iBUTab[pbSrcX[1]];
guv = g_iGUTab[pbSrcX[1]] + g_iGVTab[pbSrcX[3]];
rv = g_iRVTab[pbSrcX[3]];
yy = g_iYYTab[pbSrcX[0]];
pbDstX[0] = _Clip((yy + bu) >> SCALEBITS_OUT);
pbDstX[1] = _Clip((yy - guv) >> SCALEBITS_OUT);
pbDstX[2] = _Clip((yy + rv) >> SCALEBITS_OUT);
yy = g_iYYTab[pbSrcX[2]];
pbDstX[3] = _Clip((yy + bu) >> SCALEBITS_OUT);
pbDstX[4] = _Clip((yy - guv) >> SCALEBITS_OUT);
pbDstX[5] = _Clip((yy + rv) >> SCALEBITS_OUT);
pbDstX += 6;
pbSrcX += 4;
}
pbDstX += iDstXDif;
pbSrcX += iSrcXDif;
}
_mm_empty();
}
void reverb::comb_allpass4(signed short *sp,
signed short *dp,
const comb_param &comb_delay,
const int comb_gain,
const int allpass_delay,
const int allpass_gain,
const int *rvol,
const unsigned int sz)
{
#ifdef use_intrinsics
__m64 cg=_mm_set1_pi16(comb_gain),
ag=_mm_set1_pi16(allpass_gain),
rv[2];
rv[0]=_mm_set1_pi16(rvol[0]);
rv[1]=_mm_set1_pi16(rvol[1]);
for (unsigned int i=0; i<(sz>>4); i++, sp+=2<<2, dp+=2<<2)
{
__m64 dv[2];
for (int c=0; c<2; c++)
{
// Comb
__m64 v=_mm_setzero_si64();
for (int f=0; f<4; f++)
{
int yck=(yp-comb_delay[c][f])&(max_delay-1);
__m64 xv=*(__m64 *)(&x[c][yck]),
yv=*(__m64 *)(&y[c][f][yck]);
yv=_mm_mulhi_pi16(yv,cg);
yv=_mm_adds_pi16(yv,yv);
yv=_mm_adds_pi16(xv,yv);
*((__m64 *)&y[c][f][yp])=yv;
yv=_mm_srai_pi16(yv,2);
v=_mm_adds_pi16(v,yv);
}
// Allpass
if (allpass_delay)
{
*((__m64 *)&ax[c][yp])=v;
int ypa=(yp-allpass_delay)&(max_delay-1);
__m64 ayv=*(__m64 *)&ay[c][ypa],
xv=*(__m64 *)&x[c][yp],
axv=*(__m64 *)&ax[c][ypa];
ayv=_mm_subs_pi16(ayv,xv);
ayv=_mm_mulhi_pi16(ayv,ag);
ayv=_mm_adds_pi16(ayv,ayv);
v=_mm_adds_pi16(ayv,axv);
*((__m64 *)&ay[c][yp])=v;
}
// Output
dv[c]=_mm_mulhi_pi16(v,rv[c]);
dv[c]=_mm_adds_pi16(dv[c],dv[c]);
}
__m64 dv1=_mm_unpacklo_pi16(dv[0],dv[1]),
dv2=_mm_unpackhi_pi16(dv[0],dv[1]),
d1=*(__m64 *)&dp[0],
d2=*(__m64 *)&dp[4],
s1=*(__m64 *)&sp[0],
s2=*(__m64 *)&sp[4];
d1=_mm_adds_pi16(d1,s1);
d2=_mm_adds_pi16(d2,s2);
d1=_mm_adds_pi16(d1,dv1);
d2=_mm_adds_pi16(d2,dv2);
*(__m64 *)&dp[0]=d1;
*(__m64 *)&dp[4]=d2;
yp=(yp+4)&(max_delay-1);
}
_mm_empty();
#endif
}
void TimgFilterDCT::multiply(void)
{
const char * const factors8 = (const char*)&factors[0][0];
*(__m64*)(pWorkArea + 0 * 8 + 0) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 0 * 8 + 0), *(__m64*)(factors8 + 0 * 16)), 3);
*(__m64*)(pWorkArea + 0 * 8 + 4) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 0 * 8 + 4), *(__m64*)(factors8 + 0 * 16 + 8)), 3);
*(__m64*)(pWorkArea + 1 * 8 + 0) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 1 * 8 + 0), *(__m64*)(factors8 + 1 * 16)), 3);
*(__m64*)(pWorkArea + 1 * 8 + 4) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 1 * 8 + 4), *(__m64*)(factors8 + 1 * 16 + 8)), 3);
*(__m64*)(pWorkArea + 2 * 8 + 0) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 2 * 8 + 0), *(__m64*)(factors8 + 2 * 16)), 3);
*(__m64*)(pWorkArea + 2 * 8 + 4) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 2 * 8 + 4), *(__m64*)(factors8 + 2 * 16 + 8)), 3);
*(__m64*)(pWorkArea + 3 * 8 + 0) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 3 * 8 + 0), *(__m64*)(factors8 + 3 * 16)), 3);
*(__m64*)(pWorkArea + 3 * 8 + 4) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 3 * 8 + 4), *(__m64*)(factors8 + 3 * 16 + 8)), 3);
*(__m64*)(pWorkArea + 4 * 8 + 0) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 4 * 8 + 0), *(__m64*)(factors8 + 4 * 16)), 3);
*(__m64*)(pWorkArea + 4 * 8 + 4) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 4 * 8 + 4), *(__m64*)(factors8 + 4 * 16 + 8)), 3);
*(__m64*)(pWorkArea + 5 * 8 + 0) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 5 * 8 + 0), *(__m64*)(factors8 + 5 * 16)), 3);
*(__m64*)(pWorkArea + 5 * 8 + 4) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 5 * 8 + 4), *(__m64*)(factors8 + 5 * 16 + 8)), 3);
*(__m64*)(pWorkArea + 6 * 8 + 0) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 6 * 8 + 0), *(__m64*)(factors8 + 6 * 16)), 3);
*(__m64*)(pWorkArea + 6 * 8 + 4) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 6 * 8 + 4), *(__m64*)(factors8 + 6 * 16 + 8)), 3);
*(__m64*)(pWorkArea + 7 * 8 + 0) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 7 * 8 + 0), *(__m64*)(factors8 + 7 * 16)), 3);
*(__m64*)(pWorkArea + 7 * 8 + 4) = _mm_srai_pi16(_mm_mullo_pi16(*(__m64*)(pWorkArea + 7 * 8 + 4), *(__m64*)(factors8 + 7 * 16 + 8)), 3);
}
__m64 test_mm_srai_pi16(__m64 a) {
// CHECK-LABEL: test_mm_srai_pi16
// CHECK: call x86_mmx @llvm.x86.mmx.psrai.w
return _mm_srai_pi16(a, 3);
}
