void pix_invert :: processYUVAltivec(imageStruct &image) { int h,w,width; width = image.xsize/8; union{ unsigned char c[16]; vector unsigned char v; }charBuffer; vector unsigned char offset; vector unsigned char *inData = (vector unsigned char*) image.data; charBuffer.c[0] = 255; offset = charBuffer.v; offset = (vector unsigned char) vec_splat(offset,0); #ifndef PPC970 UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( inData, prefetchSize, 0 ); #endif for ( h=0; h<image.ysize; h++){ for (w=0; w<width; w++) { #ifndef PPC970 vec_dst( inData, prefetchSize, 0 ); #endif inData[0]=vec_subs(offset,inData[0]); inData++; } #ifndef PPC970 vec_dss( 0 ); #endif } /*end of working altivec function */ }
void pix_add :: processRGBA_Altivec(imageStruct &image, imageStruct &right) { int h,w,width; width = image.xsize/4; vector unsigned char *inData = (vector unsigned char*) image.data; vector unsigned char *rightData = (vector unsigned char*) right.data; #ifndef PPC970 UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); #endif for ( h=0; h<image.ysize; h++){ for (w=0; w<width; w++) { #ifndef PPC970 vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); #endif inData[0] = vec_adds(inData[0], rightData[0]); inData++; rightData++; } #ifndef PPC970 vec_dss( 0 ); vec_dss( 1 ); #endif } /*end of working altivec function */ }
void pix_diff :: processRGBA_Altivec(imageStruct &image, imageStruct &right) { int datasize = image.xsize * image.ysize / 4; vector signed short hiImage, loImage, hiRight, loRight; vector unsigned char zero = vec_splat_u8(0); vector unsigned char *inData = (vector unsigned char *)image.data; vector unsigned char *rightData = (vector unsigned char *)right.data; #ifndef PPC970 UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); vec_dst( inData+256, prefetchSize, 2 ); vec_dst( rightData+256, prefetchSize, 3 ); #endif do { #ifndef PPC970 vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); vec_dst( inData+256, prefetchSize, 2 ); vec_dst( rightData+256, prefetchSize, 3 ); #endif hiImage = (vector signed short)vec_mergeh(zero,inData[0]); loImage = (vector signed short)vec_mergel(zero,inData[0]); hiRight = (vector signed short)vec_mergeh(zero,rightData[0]); loRight = (vector signed short)vec_mergel(zero,rightData[0]); hiImage = vec_subs(hiImage,hiRight); loImage = vec_subs(loImage,loRight); hiImage = vec_abs(hiImage); loImage = vec_abs(loImage); inData[0] = vec_packsu(hiImage,loImage); inData++; rightData++; } while (--datasize); #ifndef PPC970 vec_dss( 0 ); vec_dss( 1 ); vec_dss( 2 ); vec_dss( 3 ); #endif }
void pix_background :: processYUVAltivec(imageStruct &image) { register int h,w,i,j,width; int pixsize = image.xsize * image.ysize * image.csize; h = image.ysize; w = image.xsize/8; width = image.xsize/8; //check to see if the buffer isn't 16byte aligned (highly unlikely) if (image.ysize*image.xsize % 16 != 0){ error("image not properly aligned for Altivec - try something SD or HD maybe?"); return; } union{ unsigned short s[8]; vector unsigned short v; }shortBuffer; if(m_savedImage.xsize!=image.xsize || m_savedImage.ysize!=image.ysize || m_savedImage.format!=image.format)m_reset=1; m_savedImage.xsize=image.xsize; m_savedImage.ysize=image.ysize; m_savedImage.setCsizeByFormat(image.format); m_savedImage.reallocate(); if (m_reset){ memcpy(m_savedImage.data,image.data,pixsize); m_reset = 0; } register vector unsigned short UVres1, Yres1, UVres2, Yres2;//interleave; register vector unsigned short hiImage, loImage; register vector unsigned short Yrange, UVrange, Yblank,UVblank,blank; register vector bool short Ymasklo,Ymaskhi, UVmaskhi; register vector unsigned short Yhi,Ylo,UVhi,UVlo; register vector unsigned char one = vec_splat_u8(1); register vector unsigned short sone = vec_splat_u16(1); register vector unsigned int Uhi, Ulo, Vhi, Vlo,Ures,Vres; register vector bool int Umasklo, Umaskhi, Vmaskhi, Vmasklo; vector unsigned char *inData = (vector unsigned char*) image.data; vector unsigned char *rightData = (vector unsigned char*) m_savedImage.data; shortBuffer.s[0] = m_Yrange; Yrange = shortBuffer.v; Yrange = vec_splat(Yrange,0); shortBuffer.s[0] = 128; shortBuffer.s[1] = 0; shortBuffer.s[2] = 128; shortBuffer.s[3] = 0; shortBuffer.s[4] = 128; shortBuffer.s[5] = 0; shortBuffer.s[6] = 128; shortBuffer.s[7] = 0; blank = shortBuffer.v; shortBuffer.s[0] = 0; Yblank = shortBuffer.v; Yblank = vec_splat(Yblank,0); shortBuffer.s[0] = 128; UVblank = shortBuffer.v; UVblank = vec_splat(UVblank,0); shortBuffer.s[0] = m_Urange; shortBuffer.s[1] = m_Vrange; shortBuffer.s[2] = m_Urange; shortBuffer.s[3] = m_Vrange; shortBuffer.s[4] = m_Urange; shortBuffer.s[5] = m_Vrange; shortBuffer.s[6] = m_Urange; shortBuffer.s[7] = m_Vrange; UVrange = shortBuffer.v; //setup the cache prefetch -- A MUST!!! UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); #ifndef PPC970 vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); vec_dst( inData+32, prefetchSize, 2 ); vec_dst( rightData+32, prefetchSize, 3 ); #endif //PPC970 for ( i=0; i<h; i++){ for (j=0; j<w; j++) { #ifndef PPC970 //this function is probably memory bound on most G4's -- what else is new? vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); vec_dst( inData+32, prefetchSize, 2 ); vec_dst( rightData+32, prefetchSize, 3 ); #endif //separate the U and V from Y UVres1 = (vector unsigned short)vec_mule(one,inData[0]); UVres2 = (vector unsigned short)vec_mule(one,rightData[0]); //vec_mulo Y * 1 to short vector Y Y Y Y shorts Yres1 = (vector unsigned short)vec_mulo(one,inData[0]); Yres2 = (vector unsigned short)vec_mulo(one,rightData[0]); Yhi = vec_adds(Yres2,Yrange); Ylo = vec_subs(Yres2,Yrange); //go to ints for comparison UVhi = vec_adds(UVres2,UVrange); UVlo = vec_subs(UVres2,UVrange); Uhi = vec_mule(sone,UVhi); Ulo = vec_mule(sone,UVlo); Vhi = vec_mulo(sone,UVhi); Vlo = vec_mulo(sone,UVlo); Ures = vec_mule(sone,UVres1); Vres = vec_mulo(sone,UVres1); Umasklo = vec_cmpgt(Ures,Ulo); Umaskhi = vec_cmplt(Ures,Uhi); Vmasklo = vec_cmpgt(Vres,Vlo); Vmaskhi = vec_cmplt(Vres,Vhi); Umaskhi = vec_and(Umaskhi,Umasklo); Vmaskhi = vec_and(Vmaskhi,Vmasklo); Umasklo = vec_and(Umaskhi,Vmaskhi); Vmasklo = vec_and(Umaskhi,Vmaskhi); hiImage = (vector unsigned short)vec_mergeh(Umasklo,Vmasklo); loImage = (vector unsigned short)vec_mergel(Umasklo,Vmasklo); //pack it back down to bool short UVmaskhi = (vector bool short)vec_packsu(hiImage,loImage); Ymasklo = vec_cmpgt(Yres1,Ylo); Ymaskhi = vec_cmplt(Yres1,Yhi); Ymaskhi = vec_and(Ymaskhi,Ymasklo); Ymaskhi = vec_and(Ymaskhi,UVmaskhi); UVmaskhi = vec_and(Ymaskhi,UVmaskhi); //bitwise comparison and move using the result of the comparison as a mask Yres1 = vec_sel(Yres1,Yblank,Ymaskhi); //UVres1 = vec_sel(UVres1,UVres2,UVmaskhi); UVres1 = vec_sel(UVres1,UVblank,UVmaskhi); //merge the Y and UV back together hiImage = vec_mergeh(UVres1,Yres1); loImage = vec_mergel(UVres1,Yres1); //pack it back down to unsigned char to store inData[0] = vec_packsu(hiImage,loImage); inData++; rightData++; } #ifndef PPC970 vec_dss(0); vec_dss(1); vec_dss(2); vec_dss(3); #endif } }
void pix_add :: processYUV_Altivec(imageStruct &image, imageStruct &right) { int h,w,width; width = image.xsize/8; //format is U Y V Y union { //unsigned int i; short elements[8]; //vector signed char v; vector signed short v; }shortBuffer; union { //unsigned int i; unsigned char elements[16]; //vector signed char v; vector unsigned char v; }charBuffer; //vector unsigned char c; register vector signed short d, hiImage, loImage, YRight, UVRight, YImage, UVImage, UVTemp, YTemp; // vector unsigned char zero = vec_splat_u8(0); register vector unsigned char c,one; // vector signed short zshort = vec_splat_s16(0); vector unsigned char *inData = (vector unsigned char*) image.data; vector unsigned char *rightData = (vector unsigned char*) right.data; //Write the pixel (pair) to the transfer buffer charBuffer.elements[0] = 2; charBuffer.elements[1] = 1; charBuffer.elements[2] = 2; charBuffer.elements[3] = 1; charBuffer.elements[4] = 2; charBuffer.elements[5] = 1; charBuffer.elements[6] = 2; charBuffer.elements[7] = 1; charBuffer.elements[8] = 2; charBuffer.elements[9] = 1; charBuffer.elements[10] = 2; charBuffer.elements[11] = 1; charBuffer.elements[12] = 2; charBuffer.elements[13] = 1; charBuffer.elements[14] = 2; charBuffer.elements[15] = 1; //Load it into the vector unit c = charBuffer.v; one = vec_splat_u8( 1 ); shortBuffer.elements[0] = 255; //Load it into the vector unit d = shortBuffer.v; d = static_cast<vector signed short>(vec_splat(static_cast<vector signed short>(d),0)); #ifndef PPC970 UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); #endif for ( h=0; h<image.ysize; h++){ for (w=0; w<width; w++) { #ifndef PPC970 vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); #endif //interleaved U Y V Y chars //vec_mule UV * 2 to short vector U V U V shorts UVImage = static_cast<vector signed short>(vec_mule(one,inData[0])); UVRight = static_cast<vector signed short>(vec_mule(c,rightData[0])); //vec_mulo Y * 1 to short vector Y Y Y Y shorts YImage = static_cast<vector signed short>(vec_mulo(c,inData[0])); YRight = static_cast<vector signed short>(vec_mulo(c,rightData[0])); //vel_subs UV - 255 UVRight = static_cast<vector signed short>(vec_subs(UVRight, d)); //vec_adds UV UVTemp = vec_adds(UVImage,UVRight); //vec_adds Y YTemp = vec_adds(YImage,YRight); hiImage = vec_mergeh(UVTemp,YTemp); loImage = vec_mergel(UVTemp,YTemp); //vec_mergel + vec_mergeh Y and UV inData[0] = vec_packsu(hiImage, loImage); inData++; rightData++; } #ifndef PPC970 vec_dss( 0 ); vec_dss( 1 ); #endif } /*end of working altivec function */ }
/* more optimized version - unrolled and load-hoisted */ void pix_offset :: processYUVAltivec(imageStruct &image) { register int h,w,width,height; width = image.xsize/16; //for altivec height = image.ysize; //format is U Y V Y // start of working altivec function union { short elements[8]; vector signed short v; } transferBuffer; register vector signed short c, hi, lo; register vector signed short hi1, lo1; register vector signed short loadhi, loadhi1, loadlo, loadlo1; register vector unsigned char zero = vec_splat_u8(0); register vector unsigned char *inData = (vector unsigned char*) image.data; //Write the pixel (pair) to the transfer buffer //transferBuffer.i = (U << 24) | (Y << 16) | (V << 8 ) | Y; transferBuffer.elements[0] = U; transferBuffer.elements[1] = Y; transferBuffer.elements[2] = V; transferBuffer.elements[3] = Y; transferBuffer.elements[4] = U; transferBuffer.elements[5] = Y; transferBuffer.elements[6] = V; transferBuffer.elements[7] = Y; //Load it into the vector unit c = transferBuffer.v; #ifndef PPC970 UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( inData, prefetchSize, 0 ); vec_dst( inData+16, prefetchSize, 1 ); vec_dst( inData+32, prefetchSize, 2 ); vec_dst( inData+64, prefetchSize, 3 ); #endif //expand the UInt8's to short's loadhi = (vector signed short) vec_mergeh( zero, inData[0] ); loadlo = (vector signed short) vec_mergel( zero, inData[0] ); loadhi1 = (vector signed short) vec_mergeh( zero, inData[1] ); loadlo1 = (vector signed short) vec_mergel( zero, inData[1] ); \ for ( h=0; h<height; h++) { for (w=0; w<width; w++) { #ifndef PPC970 vec_dst( inData, prefetchSize, 0 ); vec_dst( inData+16, prefetchSize, 1 ); vec_dst( inData+32, prefetchSize, 2 ); vec_dst( inData+64, prefetchSize, 3 ); #endif //add the constant to it hi = vec_add( loadhi, c ); lo = vec_add( loadlo, c ); hi1 = vec_add( loadhi1, c ); lo1 = vec_add( loadlo1, c ); //expand the UInt8's to short's loadhi = (vector signed short) vec_mergeh( zero, inData[2] ); loadlo = (vector signed short) vec_mergel( zero, inData[2] ); loadhi1 = (vector signed short) vec_mergeh( zero, inData[3] ); loadlo1 = (vector signed short) vec_mergel( zero, inData[3] ); //pack the result back down, with saturation inData[0] = vec_packsu( hi, lo ); inData++; inData[0] = vec_packsu( hi1, lo1 ); inData++; } } // // finish the last iteration after the loop // hi = vec_add( loadhi, c ); lo = vec_add( loadlo, c ); hi1 = vec_add( loadhi1, c ); lo1 = vec_add( loadlo1, c ); //pack the result back down, with saturation inData[0] = vec_packsu( hi, lo ); inData++; inData[0] = vec_packsu( hi1, lo1 ); inData++; #ifndef PPC970 vec_dss( 0 ); vec_dss( 1 ); vec_dss( 2 ); vec_dss( 3 ); //end of working altivec function #endif }
void pix_movement :: processYUVAltivec(imageStruct &image) { if (image.xsize*image.ysize != buffer.xsize*buffer.ysize){ buffer.xsize = image.xsize; buffer.ysize = image.ysize; buffer.reallocate(buffer.xsize*buffer.ysize*2); } int pixsize = image.ysize * image.xsize/8; union{ signed short c[8]; vector signed short v; }shortBuffer; union{ unsigned short c[8]; vector unsigned short v; }ushortBuffer; int i; vector signed short thresh; shortBuffer.c[0] = threshold; thresh = shortBuffer.v; thresh = (vector signed short)vec_splat(thresh,0); vector unsigned char *rp = (vector unsigned char *) image.data; // read pointer vector unsigned char *wp = (vector unsigned char *) buffer.data; // write pointer to the copy vector unsigned char grey0,grey1; vector unsigned char one = vec_splat_u8(1); vector unsigned short Y0,Ywp0,hiImage0,loImage0; vector unsigned short Y1,Ywp1,hiImage1,loImage1; vector unsigned short UVwp0,UVwp1; vector signed short temp0,temp1; ushortBuffer.c[0]=127; vector unsigned short UV0= (vector unsigned short)vec_splat(ushortBuffer.v, 0); vector unsigned short UV1= (vector unsigned short)vec_splat(ushortBuffer.v, 0); #ifndef PPC970 //setup the cache prefetch -- A MUST!!! UInt32 prefetchSize = GetPrefetchConstant( 16, 0, 256 ); vec_dst( rp, prefetchSize, 0 ); vec_dst( wp, prefetchSize, 1 ); #endif int j = 16; pixsize/=2; for (i=0; i < pixsize; i++) { # ifndef PPC970 //setup the cache prefetch -- A MUST!!! UInt32 prefetchSize = GetPrefetchConstant( j, 0, j * 16 ); vec_dst( rp, prefetchSize, 0 ); vec_dst( wp, prefetchSize, 1 ); vec_dst( rp+16, prefetchSize, 2 ); vec_dst( wp+16, prefetchSize, 3 ); # endif grey0 = rp[0]; grey1 = rp[1]; // rp[Y0]=255*(abs(grey0-*wp)>thresh); // UV0= (vector unsigned short)vec_mule(grey0,one); Y0 = (vector unsigned short)vec_mulo(grey0,one); // UV1= (vector unsigned short)vec_mule(grey1,one); Y1 = (vector unsigned short)vec_mulo(grey1,one); //wp is actually 1/2 the size of the image because it is only Y?? //here the full U Y V Y is stored // UVwp0= (vector unsigned short)vec_mule(wp[0],one); Ywp0 = (vector unsigned short)vec_mulo(wp[0],one); // UVwp1= (vector unsigned short)vec_mule(wp[1],one); Ywp1 = (vector unsigned short)vec_mulo(wp[1],one); //store the current pixels as the history for next time wp[0]=grey0; wp++; wp[0]=grey1; wp++; temp0 = vec_abs(vec_sub((vector signed short)Y0,(vector signed short)Ywp0)); Y0 = (vector unsigned short)vec_cmpgt(temp0,thresh); temp1 = vec_abs(vec_sub((vector signed short)Y1,(vector signed short)Ywp1)); Y1 = (vector unsigned short)vec_cmpgt(temp1,thresh); hiImage0 = vec_mergeh(UV0,Y0); loImage0 = vec_mergel(UV0,Y0); hiImage1 = vec_mergeh(UV1,Y1); loImage1 = vec_mergel(UV1,Y1); grey0 = vec_packsu(hiImage0,loImage0); grey1 = vec_packsu(hiImage1,loImage1); rp[0]=grey0; rp++; rp[0]=grey1; rp++; // grey = rp[0]; // rp[Y1]=255*(abs(grey-*wp)>thresh); // *wp++=grey; // rp+=4; // rp++; } # ifndef PPC970 vec_dss(0); vec_dss(1); vec_dss(2); vec_dss(3); # endif }
void pix_diff :: processYUV_Altivec(imageStruct &image, imageStruct &right) { long h,w,width; width = image.xsize/8; //format is U Y V Y union { //unsigned int i; short elements[8]; //vector signed char v; vector short v; }shortBuffer; vector signed short d, hiImage, loImage,hiRight, loRight;//, YRight, UVRight, YImage, UVImage, UVTemp, YTemp; vector unsigned char zero = vec_splat_u8(0); vector unsigned char *inData = (vector unsigned char*) image.data; vector unsigned char *rightData = (vector unsigned char*) right.data; shortBuffer.elements[0] = 128; shortBuffer.elements[1] = 0; shortBuffer.elements[2] = 128; shortBuffer.elements[3] = 0; shortBuffer.elements[4] = 128; shortBuffer.elements[5] = 0; shortBuffer.elements[6] = 128; shortBuffer.elements[7] = 0; //Load it into the vector unit d = shortBuffer.v; #ifndef PPC970 UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); #endif for ( h=0; h<image.ysize; h++){ for (w=0; w<width; w++) { #ifndef PPC970 vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); #endif //interleaved U Y V Y chars //break out to unsigned shorts hiImage = (vector signed short) vec_mergeh( zero, inData[0] ); loImage = (vector signed short) vec_mergel( zero, inData[0] ); hiRight = (vector signed short) vec_mergeh( zero, rightData[0] ); loRight = (vector signed short) vec_mergel( zero, rightData[0] ); //subtract the 128 offset for UV hiImage = vec_subs(hiImage,d); loImage = vec_subs(loImage,d); hiRight = vec_subs(hiRight,d); loRight = vec_subs(loRight,d); hiImage = vec_subs(hiImage,hiRight); loImage = vec_subs(loImage,loRight); hiImage = vec_adds(hiImage,d); loImage = vec_adds(loImage,d); hiImage = vec_abs(hiImage); loImage = vec_abs(loImage); inData[0] = vec_packsu(hiImage, loImage); inData++; rightData++; } #ifndef PPC970 vec_dss( 0 ); vec_dss( 1 ); #endif } /*end of working altivec function */ }
void pix_compare :: processYUV_Altivec(imageStruct &image, imageStruct &right) { register int h,w,i,j,width; h = image.ysize; w = image.xsize/8; width = image.xsize/8; //check to see if the buffer isn't 16byte aligned (highly unlikely) if (image.ysize*image.xsize % 16 != 0){ error("image not properly aligned for Altivec"); return; } register vector unsigned short UVres1, Yres1, UVres2, Yres2;//interleave; register vector unsigned short hiImage, loImage; register vector bool short Ymask1; register vector unsigned char one = vec_splat_u8(1); vector unsigned char *inData = (vector unsigned char*) image.data; vector unsigned char *rightData = (vector unsigned char*) right.data; #ifndef PPC970 //setup the cache prefetch -- A MUST!!! UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); #endif if (m_direction) { for ( i=0; i<h; i++){ for (j=0; j<w; j++) { #ifndef PPC970 //this function is probably memory bound on most G4's -- what else is new? vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); #endif //separate the U and V from Y UVres1 = (vector unsigned short)vec_mule(one,inData[0]); UVres2 = (vector unsigned short)vec_mule(one,rightData[0]); //vec_mulo Y * 1 to short vector Y Y Y Y shorts Yres1 = (vector unsigned short)vec_mulo(one,inData[0]); Yres2 = (vector unsigned short)vec_mulo(one,rightData[0]); //compare the Y values Ymask1 = vec_cmpgt(Yres1,Yres2); //bitwise comparison and move using the result of the comparison as a mask Yres1 = vec_sel(Yres2,Yres1,Ymask1); UVres1 = vec_sel(UVres2,UVres1,Ymask1); //merge the Y and UV back together hiImage = vec_mergeh(UVres1,Yres1); loImage = vec_mergel(UVres1,Yres1); //pack it back down to unsigned char to store inData[0] = vec_packsu(hiImage,loImage); inData++; rightData++; } #ifndef PPC970 vec_dss(1); vec_dss(0); #endif } }else{ for ( i=0; i<h; i++){ for (j=0; j<w; j++) { #ifndef PPC970 vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); #endif UVres1 = (vector unsigned short)vec_mule(one,inData[0]); UVres2 = (vector unsigned short)vec_mule(one,rightData[0]); //vec_mulo Y * 1 to short vector Y Y Y Y shorts Yres1 = (vector unsigned short)vec_mulo(one,inData[0]); Yres2 = (vector unsigned short)vec_mulo(one,rightData[0]); Ymask1 = vec_cmplt(Yres1,Yres2); Yres1 = vec_sel(Yres2,Yres1,Ymask1); UVres1 = vec_sel(UVres2,UVres1,Ymask1); hiImage = vec_mergeh(UVres1,Yres1); loImage = vec_mergel(UVres1,Yres1); inData[0] = vec_packsu(hiImage,loImage); inData++; rightData++; } #ifndef PPC970 vec_dss(1); vec_dss(0); #endif } } }
/* start of optimized motionblur */ void pix_motionblur :: processYUVAltivec(imageStruct &image) { int h,w,width; signed short rightGain,imageGain; unsigned char *saved = m_savedImage.data; m_savedImage.xsize=image.xsize; m_savedImage.ysize=image.ysize; m_savedImage.setCsizeByFormat(image.format); m_savedImage.reallocate(); if(saved!=m_savedImage.data) { m_savedImage.setBlack(); } saved=m_savedImage.data; width = image.xsize/8; /* // hmm: why does it read 235 ? rightGain = (signed short)(235. * m_motionblur); imageGain = (signed short) (255. - (235. * m_motionblur)); */ rightGain = m_blur1; imageGain = m_blur0; union { signed short elements[8]; vector signed short v; } shortBuffer; union { unsigned int elements[4]; vector unsigned int v; } bitBuffer; register vector signed short gainAdd, hiImage, loImage,hiRight,loRight, YImage, UVImage; // register vector signed short loadhiImage, loadloImage,loadhiRight,loadloRight; register vector unsigned char loadImage, loadRight; register vector unsigned char zero = vec_splat_u8(0); register vector signed int UVhi,UVlo,Yhi,Ylo; register vector signed int UVhiR,UVloR,YhiR,YloR; register vector signed short gainSub,gain,gainR;//,d; register vector unsigned int bitshift; vector unsigned char *inData = (vector unsigned char*) image.data; vector unsigned char *rightData = (vector unsigned char*) saved; shortBuffer.elements[0] = 128; shortBuffer.elements[1] = 0; shortBuffer.elements[2] = 128; shortBuffer.elements[3] = 0; shortBuffer.elements[4] = 128; shortBuffer.elements[5] = 0; shortBuffer.elements[6] = 128; shortBuffer.elements[7] = 0; gainSub = shortBuffer.v; shortBuffer.elements[0] = imageGain; gain = shortBuffer.v; gain = vec_splat(gain, 0 ); shortBuffer.elements[0] = rightGain; gainR = shortBuffer.v; gainR = vec_splat(gainR, 0 ); bitBuffer.elements[0] = 8; //Load it into the vector unit bitshift = bitBuffer.v; bitshift = vec_splat(bitshift,0); shortBuffer.elements[0] = 128; //Load it into the vector unit gainAdd = shortBuffer.v; gainAdd = (vector signed short)vec_splat((vector signed short)gainAdd,0); # ifndef PPC970 UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); vec_dst( inData+32, prefetchSize, 2 ); vec_dst( rightData+32, prefetchSize, 3 ); # endif loadImage = inData[0]; loadRight = rightData[0]; for ( h=0; h<image.ysize; h++) { for (w=0; w<width; w++) { # ifndef PPC970 vec_dst( inData, prefetchSize, 0 ); vec_dst( rightData, prefetchSize, 1 ); vec_dst( inData+32, prefetchSize, 2 ); vec_dst( rightData+32, prefetchSize, 3 ); # endif //interleaved U Y V Y chars hiImage = (vector signed short) vec_mergeh( zero, loadImage ); loImage = (vector signed short) vec_mergel( zero, loadImage ); hiRight = (vector signed short) vec_mergeh( zero, loadRight ); loRight = (vector signed short) vec_mergel( zero, loadRight ); //hoist that load!! loadImage = inData[1]; loadRight = rightData[1]; //subtract 128 from UV hiImage = vec_subs(hiImage,gainSub); loImage = vec_subs(loImage,gainSub); hiRight = vec_subs(hiRight,gainSub); loRight = vec_subs(loRight,gainSub); //now vec_mule the UV into two vector ints //change sone to gain UVhi = vec_mule(gain,hiImage); UVlo = vec_mule(gain,loImage); UVhiR = vec_mule(gainR,hiRight); UVloR = vec_mule(gainR,loRight); //now vec_mulo the Y into two vector ints Yhi = vec_mulo(gain,hiImage); Ylo = vec_mulo(gain,loImage); YhiR = vec_mulo(gainR,hiRight); YloR = vec_mulo(gainR,loRight); //this is where to do the add and bitshift due to the resolution //add UV UVhi = vec_adds(UVhi,UVhiR); UVlo = vec_adds(UVlo,UVloR); Yhi = vec_adds(Yhi,YhiR); Ylo = vec_adds(Ylo,YloR); //bitshift UV UVhi = vec_sra(UVhi,bitshift); UVlo = vec_sra(UVlo,bitshift); Yhi = vec_sra(Yhi,bitshift); Ylo = vec_sra(Ylo,bitshift); //pack the UV into a single short vector UVImage = vec_packs(UVhi,UVlo); //pack the Y into a single short vector YImage = vec_packs(Yhi,Ylo); //vec_mergel + vec_mergeh Y and UV hiImage = vec_mergeh(UVImage,YImage); loImage = vec_mergel(UVImage,YImage); //add 128 offset back hiImage = vec_adds(hiImage,gainSub); loImage = vec_adds(loImage,gainSub); //vec_mergel + vec_mergeh Y and UV rightData[0] = (vector unsigned char)vec_packsu(hiImage, loImage); inData[0] = (vector unsigned char)vec_packsu(hiImage, loImage); inData++; rightData++; } } # ifndef PPC970 //stop the cache streams vec_dss( 0 ); vec_dss( 1 ); vec_dss( 2 ); vec_dss( 3 ); # endif }/* end of working altivec function */
void YUV422_to_YV12_altivec(short*pY, short*pY2, short*pU, short*pV, const unsigned char *gem_image, int xsize, int ysize) { // UYVY UYVY UYVY UYVY const vector unsigned char *pixels1=reinterpret_cast<const vector unsigned char *>(gem_image); const vector unsigned char *pixels2=reinterpret_cast<const vector unsigned char *>(gem_image+(xsize*2)); // PDP packet to be filled: // first Y plane vector signed short *py1 = reinterpret_cast<vector signed short *>(pY); // 2nd Y pixel vector signed short *py2 = reinterpret_cast<vector signed short *>(pY2); // U plane vector signed short *pCr = reinterpret_cast<vector signed short *>(pU); // V plane vector signed short *pCb = reinterpret_cast<vector signed short *>(pV); vector signed short uvSub = static_cast<vector signed short>( 128, 128, 128, 128, 128, 128, 128, 128 ); vector unsigned short yShift = static_cast<vector unsigned short>( 7, 7, 7, 7, 7, 7, 7, 7 ); vector unsigned short uvShift = static_cast<vector unsigned short>( 8, 8, 8, 8, 8, 8, 8, 8 ); vector signed short tempY1, tempY2, tempY3, tempY4, tempUV1, tempUV2, tempUV3, tempUV4, tempUV5, tempUV6; vector unsigned char uvPerm = static_cast<vector unsigned char>( 16, 0, 17, 4, 18, 8, 19, 12, // u0..u3 20, 2, 21, 6, 22, 10, 23, 14 ); // v0..v3 vector unsigned char uPerm = static_cast<vector unsigned char>( 0, 1, 2, 3, 4, 5, 6, 7, 16,17,18,19,20,21,22,23); vector unsigned char vPerm = static_cast<vector unsigned char>( 8, 9, 10,11,12,13,14,15, 24,25,26,27,28,29,30,31); vector unsigned char yPerm = static_cast<vector unsigned char>( 16, 1, 17, 3, 18, 5, 19, 7, // y0..y3 20, 9, 21, 11, 23, 13, 25, 15);// y4..y7 vector unsigned char zeroVec = static_cast<vector unsigned char>(0); int row=ysize>>1; int cols=xsize>>4; #if 0 # ifndef PPC970 UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( pu, prefetchSize, 0 ); vec_dst( pv, prefetchSize, 0 ); vec_dst( py1, prefetchSize, 0 ); vec_dst( py2, prefetchSize, 0 ); # endif #endif while(row--){ int col=cols; while(col--){ #if 0 # ifndef PPC970 vec_dst( ); # endif #endif tempUV1 = static_cast<vector signed short>(vec_perm( *pixels1, zeroVec, uvPerm)); tempY1 = static_cast<vector signed short>(vec_perm( *pixels1, zeroVec, yPerm)); tempY2 = static_cast<vector signed short>(vec_perm( *pixels2, zeroVec, yPerm)); pixels1++;pixels2++; tempUV2 = static_cast<vector signed short>(vec_perm( *pixels1, zeroVec, uvPerm)); tempY3 = static_cast<vector signed short>(vec_perm( *pixels1, zeroVec, yPerm)); tempY4 = static_cast<vector signed short>(vec_perm( *pixels2, zeroVec, yPerm)); pixels1++;pixels2++; tempUV3 = vec_sub( tempUV1, uvSub ); tempUV4 = vec_sub( tempUV2, uvSub ); tempUV5 = vec_sl( tempUV3, uvShift ); tempUV6 = vec_sl( tempUV4, uvShift ); *pCb = vec_perm( tempUV5, tempUV6, uPerm ); *pCr = vec_perm( tempUV5, tempUV6, vPerm ); pCr++; pCb++; *py1++ = vec_sl( tempY1, yShift); *py2++ = vec_sl( tempY2, yShift); *py1++ = vec_sl( tempY3, yShift); *py2++ = vec_sl( tempY4, yShift); } py1+=(xsize>>3); py2+=(xsize>>3); pixels1+=(xsize*2)>>4; pixels2+=(xsize*2)>>4; } }
void YV12_to_YUV422_altivec(const short*Y, const short*U, const short*V, unsigned char *data, int xsize, int ysize) { // from [email protected], 3/15/2005 // #1. Don't use the pointers. Use vec_ld with an index that you increment (by 16) instead. vector unsigned char *pixels1=reinterpret_cast<vector unsigned char *>(data); vector unsigned char *pixels2=reinterpret_cast<vector unsigned char *>(data+(xsize*2)); const vector unsigned short *py1 = reinterpret_cast<const vector unsigned short *>(Y); const vector unsigned short *py2 = reinterpret_cast<const vector unsigned short *>(Y + xsize ); const vector unsigned short *pu = reinterpret_cast<const vector unsigned short *>(U); const vector unsigned short *pv = reinterpret_cast<const vector unsigned short *>(V); vector unsigned short uvAdd = static_cast<vector unsigned short>( 128, 128, 128, 128, 128, 128, 128, 128 ); vector unsigned short yShift = static_cast<vector unsigned short>( 7, 7, 7, 7, 7, 7, 7, 7 ); vector unsigned short uvShift = static_cast<vector unsigned short>( 8, 8, 8, 8, 8, 8, 8, 8 ); vector unsigned short tempU, tempV, doneU, doneV, tempY1, tempY2, tempY3, tempY4, uv1, uv2, out1, out2, out3, out4, out5, out6, out7, out8; vector unsigned char Perm1 = static_cast<vector unsigned char>( 0, 1, 16, 17, 2, 3, 18, 19, 4, 5, 20, 21, 6, 7, 22, 23 ); vector unsigned char Perm2 = static_cast<vector unsigned char>( 8, 9, 24, 25, 10, 11, 26, 27, 12, 13, 28, 29, 14, 15, 30, 31 ); int row=ysize>>1; int cols=xsize>>4; #if 0 # ifndef PPC970 UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 ); vec_dst( pu, prefetchSize, 0 ); vec_dst( pv, prefetchSize, 0 ); vec_dst( py1, prefetchSize, 0 ); vec_dst( py2, prefetchSize, 0 ); # endif #endif while(row--){ int col=cols; while(col--){ #if 0 # ifndef PPC970 vec_dst( ); # endif #endif tempU = vec_sra( (*pu++), uvShift ); tempV = vec_sra( (*pv++), uvShift ); doneU = vec_add( tempU, uvAdd ); doneV = vec_add( tempV, uvAdd ); uv1 = vec_perm( doneU, doneV, Perm1 ); // uvuvuvuv uvuvuvuv uv2 = vec_perm( doneU, doneV, Perm2 ); tempY1 = vec_sra( (*py1++), yShift ); tempY2 = vec_sra( (*py2++), yShift ); out1 = vec_perm( uv1, tempY1, Perm1 ); //fill Y's, U's & V's out2 = vec_perm( uv1, tempY1, Perm2 ); out3 = vec_perm( uv1, tempY2, Perm1 ); //fill 2nd Y's, U's & V's out4 = vec_perm( uv1, tempY2, Perm2 ); *pixels1 = vec_packsu( out1, out2 ); *pixels2 = vec_packsu( out3, out4 ); pixels1++; pixels2++; tempY3 = vec_sra( (*py1++), yShift ); // load second set of Y's tempY4 = vec_sra( (*py2++), yShift ); out5 = vec_perm( uv2, tempY3, Perm1 ); out6 = vec_perm( uv2, tempY3, Perm2 ); out7 = vec_perm( uv2, tempY4, Perm1 ); out8 = vec_perm( uv2, tempY4, Perm2 ); *pixels1 = vec_packsu( out5, out6 ); *pixels2 = vec_packsu( out7, out8 ); pixels1++; pixels2++; } pixels1+=(xsize*2)>>4; pixels2+=(xsize*2)>>4; py1+=xsize>>3; py2+=xsize>>3; } }