void gimp_composite_multiply_rgba8_rgba8_rgba8_altivec (GimpCompositeContext *ctx) { const guchar *A = ctx->A; const guchar *B = ctx->B; guchar *D = ctx->D; guint length = ctx->n_pixels; vector unsigned char a,b,d,alpha_a,alpha_b,alpha; vector unsigned short al,ah; while (length >= 4) { a=LoadUnaligned(A); b=LoadUnaligned(B); al=vec_mule(a,b); al=vec_add(al,ox0080); ah=vec_mulo(a,b); ah=vec_add(ah,ox0080); al=vec_add(al,vec_sr(al,ox0008)); ah=vec_add(ah,vec_sr(ah,ox0008)); d=vec_perm((vector unsigned char)al,(vector unsigned char)ah,combine_high_bytes); alpha_a=vec_and(a, alphamask); alpha_b=vec_and(b, alphamask); alpha=vec_min(alpha_a, alpha_b); d=vec_andc(d, alphamask); d=vec_or(d, alpha); StoreUnaligned(d, D); A+=16; B+=16; D+=16; length-=4; } /* process last pixels */ length = length*4; a=LoadUnalignedLess(A, length); b=LoadUnalignedLess(B, length); al=vec_mule(a,b); al=vec_add(al,ox0080); ah=vec_mulo(a,b); ah=vec_add(ah,ox0080); al=vec_add(al,vec_sr(al,ox0008)); ah=vec_add(ah,vec_sr(ah,ox0008)); d=vec_perm((vector unsigned char)al,(vector unsigned char)ah,combine_high_bytes); alpha_a=vec_and(a, alphamask); alpha_b=vec_and(b, alphamask); alpha=vec_min(alpha_a, alpha_b); d=vec_andc(d, alphamask); d=vec_or(d, alpha); StoreUnalignedLess(d, D, length); }
SIMD_INLINE v128_u32 BgraToGray32(v128_u8 bgra) { const v128_u16 _b_r = vec_mule(bgra, K8_01); const v128_u16 _g_a = vec_mulo(bgra, K8_01); const v128_u32 weightedSum = vec_add(vec_mule(_g_a, K16_GREEN_0000), vec_add(vec_mule(_b_r, K16_BLUE_RED), vec_mulo(_b_r, K16_BLUE_RED))); return vec_sr(vec_add(weightedSum, K32_ROUND_TERM), K32_SHIFT); }
SIMD_INLINE v128_u8 Average8(const v128_u8 & s00, const v128_u8 & s01, const v128_u8 & s10, const v128_u8 & s11) { v128_u16 lo = Average16( vec_mule(s00, K8_01), vec_mulo(s00, K8_01), vec_mule(s10, K8_01), vec_mulo(s10, K8_01)); v128_u16 hi = Average16( vec_mule(s01, K8_01), vec_mulo(s01, K8_01), vec_mule(s11, K8_01), vec_mulo(s11, K8_01)); return vec_pack(lo, hi); }
static void predict_16x16_p_altivec( uint8_t *src ) { int16_t a, b, c, i; int H = 0; int V = 0; int16_t i00; for( i = 1; i <= 8; i++ ) { H += i * ( src[7+i - FDEC_STRIDE ] - src[7-i - FDEC_STRIDE ] ); V += i * ( src[(7+i)*FDEC_STRIDE -1] - src[(7-i)*FDEC_STRIDE -1] ); } a = 16 * ( src[15*FDEC_STRIDE -1] + src[15 - FDEC_STRIDE] ); b = ( 5 * H + 32 ) >> 6; c = ( 5 * V + 32 ) >> 6; i00 = a - b * 7 - c * 7 + 16; vect_sshort_u i00_u, b_u, c_u; i00_u.s[0] = i00; b_u.s[0] = b; c_u.s[0] = c; vec_u16_t val5_v = vec_splat_u16(5); vec_s16_t i00_v, b_v, c_v; i00_v = vec_splat(i00_u.v, 0); b_v = vec_splat(b_u.v, 0); c_v = vec_splat(c_u.v, 0); vec_s16_t induc_v = (vec_s16_t) CV(0, 1, 2, 3, 4, 5, 6, 7); vec_s16_t b8_v = vec_sl(b_v, vec_splat_u16(3)); vec_s32_t mule_b_v = vec_mule(induc_v, b_v); vec_s32_t mulo_b_v = vec_mulo(induc_v, b_v); vec_s16_t mul_b_induc0_v = vec_pack(vec_mergeh(mule_b_v, mulo_b_v), vec_mergel(mule_b_v, mulo_b_v)); vec_s16_t add_i0_b_0v = vec_adds(i00_v, mul_b_induc0_v); vec_s16_t add_i0_b_8v = vec_adds(b8_v, add_i0_b_0v); int y; for( y = 0; y < 16; y++ ) { vec_s16_t shift_0_v = vec_sra(add_i0_b_0v, val5_v); vec_s16_t shift_8_v = vec_sra(add_i0_b_8v, val5_v); vec_u8_t com_sat_v = vec_packsu(shift_0_v, shift_8_v); vec_st( com_sat_v, 0, &src[0]); src += FDEC_STRIDE; i00 += c; add_i0_b_0v = vec_adds(add_i0_b_0v, c_v); add_i0_b_8v = vec_adds(add_i0_b_8v, c_v); } }
void rgbaint_t::blend(const rgbaint_t& other, UINT8 factor) { const VECU32 shift = vec_splat_u32(-16); const VECS32 scale1 = { factor, factor, factor, factor }; const VECS32 scale2 = { 0x100 - factor, 0x100 - factor, 0x100 - factor, 0x100 - factor, }; VECU32 temp = vec_msum(VECU16(m_value), VECU16(vec_rl(scale1, shift)), vec_splat_u32(0)); temp = vec_msum(VECU16(other.m_value), VECU16(vec_rl(scale2, shift)), temp); #if defined __LITTLE_ENDIAN__ m_value = VECS32(vec_msum(VECU16(m_value), VECU16(scale1), vec_mule(VECU16(other.m_value), VECU16(scale2)))); #else m_value = VECS32(vec_msum(VECU16(m_value), VECU16(scale1), vec_mulo(VECU16(other.m_value), VECU16(scale2)))); #endif m_value = VECS32(vec_add(vec_sl(temp, shift), VECU32(m_value))); sra_imm(8); }
void S_PaintChannelFrom16_altivec( portable_samplepair_t paintbuffer[PAINTBUFFER_SIZE], int snd_vol, channel_t *ch, const sfx_t *sc, int count, int sampleOffset, int bufferOffset ) { int data, aoff, boff; int leftvol, rightvol; int i, j; portable_samplepair_t *samp; sndBuffer *chunk; short *samples; float ooff, fdata[2], fdiv, fleftvol, frightvol; if (sc->soundChannels <= 0) { return; } samp = &paintbuffer[ bufferOffset ]; if (ch->doppler) { sampleOffset = sampleOffset*ch->oldDopplerScale; } if ( sc->soundChannels == 2 ) { sampleOffset *= sc->soundChannels; if ( sampleOffset & 1 ) { sampleOffset &= ~1; } } chunk = sc->soundData; while (sampleOffset>=SND_CHUNK_SIZE) { chunk = chunk->next; sampleOffset -= SND_CHUNK_SIZE; if (!chunk) { chunk = sc->soundData; } } if (!ch->doppler || ch->dopplerScale==1.0f) { vector signed short volume_vec; vector unsigned int volume_shift; int vectorCount, samplesLeft, chunkSamplesLeft; leftvol = ch->leftvol*snd_vol; rightvol = ch->rightvol*snd_vol; samples = chunk->sndChunk; ((short *)&volume_vec)[0] = leftvol; ((short *)&volume_vec)[1] = leftvol; ((short *)&volume_vec)[4] = leftvol; ((short *)&volume_vec)[5] = leftvol; ((short *)&volume_vec)[2] = rightvol; ((short *)&volume_vec)[3] = rightvol; ((short *)&volume_vec)[6] = rightvol; ((short *)&volume_vec)[7] = rightvol; volume_shift = vec_splat_u32(8); i = 0; while(i < count) { /* Try to align destination to 16-byte boundary */ while(i < count && (((unsigned long)&samp[i] & 0x1f) || ((count-i) < 8) || ((SND_CHUNK_SIZE - sampleOffset) < 8))) { data = samples[sampleOffset++]; samp[i].left += (data * leftvol)>>8; if ( sc->soundChannels == 2 ) { data = samples[sampleOffset++]; } samp[i].right += (data * rightvol)>>8; if (sampleOffset == SND_CHUNK_SIZE) { chunk = chunk->next; samples = chunk->sndChunk; sampleOffset = 0; } i++; } /* Destination is now aligned. Process as many 8-sample chunks as we can before we run out of room from the current sound chunk. We do 8 per loop to avoid extra source data reads. */ samplesLeft = count - i; chunkSamplesLeft = SND_CHUNK_SIZE - sampleOffset; if(samplesLeft > chunkSamplesLeft) samplesLeft = chunkSamplesLeft; vectorCount = samplesLeft / 8; if(vectorCount) { vector unsigned char tmp; vector short s0, s1, sampleData0, sampleData1; vector signed int merge0, merge1; vector signed int d0, d1, d2, d3; vector unsigned char samplePermute0 = VECCONST_UINT8(0, 1, 4, 5, 0, 1, 4, 5, 2, 3, 6, 7, 2, 3, 6, 7); vector unsigned char samplePermute1 = VECCONST_UINT8(8, 9, 12, 13, 8, 9, 12, 13, 10, 11, 14, 15, 10, 11, 14, 15); vector unsigned char loadPermute0, loadPermute1; // Rather than permute the vectors after we load them to do the sample // replication and rearrangement, we permute the alignment vector so // we do everything in one step below and avoid data shuffling. tmp = vec_lvsl(0,&samples[sampleOffset]); loadPermute0 = vec_perm(tmp,tmp,samplePermute0); loadPermute1 = vec_perm(tmp,tmp,samplePermute1); s0 = *(vector short *)&samples[sampleOffset]; while(vectorCount) { /* Load up source (16-bit) sample data */ s1 = *(vector short *)&samples[sampleOffset+7]; /* Load up destination sample data */ d0 = *(vector signed int *)&samp[i]; d1 = *(vector signed int *)&samp[i+2]; d2 = *(vector signed int *)&samp[i+4]; d3 = *(vector signed int *)&samp[i+6]; sampleData0 = vec_perm(s0,s1,loadPermute0); sampleData1 = vec_perm(s0,s1,loadPermute1); merge0 = vec_mule(sampleData0,volume_vec); merge0 = vec_sra(merge0,volume_shift); /* Shift down to proper range */ merge1 = vec_mulo(sampleData0,volume_vec); merge1 = vec_sra(merge1,volume_shift); d0 = vec_add(merge0,d0); d1 = vec_add(merge1,d1); merge0 = vec_mule(sampleData1,volume_vec); merge0 = vec_sra(merge0,volume_shift); /* Shift down to proper range */ merge1 = vec_mulo(sampleData1,volume_vec); merge1 = vec_sra(merge1,volume_shift); d2 = vec_add(merge0,d2); d3 = vec_add(merge1,d3); /* Store destination sample data */ *(vector signed int *)&samp[i] = d0; *(vector signed int *)&samp[i+2] = d1; *(vector signed int *)&samp[i+4] = d2; *(vector signed int *)&samp[i+6] = d3; i += 8; vectorCount--; s0 = s1; sampleOffset += 8; } if (sampleOffset == SND_CHUNK_SIZE) { chunk = chunk->next; samples = chunk->sndChunk; sampleOffset = 0; } } } } else {
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 */ }
/* this code assume stride % 16 == 0 *and* tmp is properly aligned */ static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp, uint8_t * src, int dstStride, int tmpStride, int srcStride) { POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_hv_lowpass_num, 1); POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1); register int i; const vector signed int vzero = vec_splat_s32(0); const vector unsigned char permM2 = vec_lvsl(-2, src); const vector unsigned char permM1 = vec_lvsl(-1, src); const vector unsigned char permP0 = vec_lvsl(+0, src); const vector unsigned char permP1 = vec_lvsl(+1, src); const vector unsigned char permP2 = vec_lvsl(+2, src); const vector unsigned char permP3 = vec_lvsl(+3, src); const vector signed short v20ss = (const vector signed short)AVV(20); const vector unsigned int v10ui = vec_splat_u32(10); const vector signed short v5ss = vec_splat_s16(5); const vector signed short v1ss = vec_splat_s16(1); const vector signed int v512si = (const vector signed int)AVV(512); const vector unsigned int v16ui = (const vector unsigned int)AVV(16); register int align = ((((unsigned long)src) - 2) % 16); src -= (2 * srcStride); for (i = 0 ; i < 21 ; i ++) { vector unsigned char srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; vector unsigned char srcR1 = vec_ld(-2, src); vector unsigned char srcR2 = vec_ld(14, src); switch (align) { default: { srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = vec_perm(srcR1, srcR2, permP0); srcP1 = vec_perm(srcR1, srcR2, permP1); srcP2 = vec_perm(srcR1, srcR2, permP2); srcP3 = vec_perm(srcR1, srcR2, permP3); } break; case 11: { srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = vec_perm(srcR1, srcR2, permP0); srcP1 = vec_perm(srcR1, srcR2, permP1); srcP2 = vec_perm(srcR1, srcR2, permP2); srcP3 = srcR2; } break; case 12: { vector unsigned char srcR3 = vec_ld(30, src); srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = vec_perm(srcR1, srcR2, permP0); srcP1 = vec_perm(srcR1, srcR2, permP1); srcP2 = srcR2; srcP3 = vec_perm(srcR2, srcR3, permP3); } break; case 13: { vector unsigned char srcR3 = vec_ld(30, src); srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = vec_perm(srcR1, srcR2, permP0); srcP1 = srcR2; srcP2 = vec_perm(srcR2, srcR3, permP2); srcP3 = vec_perm(srcR2, srcR3, permP3); } break; case 14: { vector unsigned char srcR3 = vec_ld(30, src); srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = srcR2; srcP1 = vec_perm(srcR2, srcR3, permP1); srcP2 = vec_perm(srcR2, srcR3, permP2); srcP3 = vec_perm(srcR2, srcR3, permP3); } break; case 15: { vector unsigned char srcR3 = vec_ld(30, src); srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = srcR2; srcP0 = vec_perm(srcR2, srcR3, permP0); srcP1 = vec_perm(srcR2, srcR3, permP1); srcP2 = vec_perm(srcR2, srcR3, permP2); srcP3 = vec_perm(srcR2, srcR3, permP3); } break; } const vector signed short srcP0A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0); const vector signed short srcP0B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0); const vector signed short srcP1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1); const vector signed short srcP1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1); const vector signed short srcP2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2); const vector signed short srcP2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2); const vector signed short srcP3A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3); const vector signed short srcP3B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3); const vector signed short srcM1A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1); const vector signed short srcM1B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1); const vector signed short srcM2A = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2); const vector signed short srcM2B = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2); const vector signed short sum1A = vec_adds(srcP0A, srcP1A); const vector signed short sum1B = vec_adds(srcP0B, srcP1B); const vector signed short sum2A = vec_adds(srcM1A, srcP2A); const vector signed short sum2B = vec_adds(srcM1B, srcP2B); const vector signed short sum3A = vec_adds(srcM2A, srcP3A); const vector signed short sum3B = vec_adds(srcM2B, srcP3B); const vector signed short pp1A = vec_mladd(sum1A, v20ss, sum3A); const vector signed short pp1B = vec_mladd(sum1B, v20ss, sum3B); const vector signed short pp2A = vec_mladd(sum2A, v5ss, (vector signed short)vzero); const vector signed short pp2B = vec_mladd(sum2B, v5ss, (vector signed short)vzero); const vector signed short psumA = vec_sub(pp1A, pp2A); const vector signed short psumB = vec_sub(pp1B, pp2B); vec_st(psumA, 0, tmp); vec_st(psumB, 16, tmp); src += srcStride; tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */ } const vector unsigned char dstperm = vec_lvsr(0, dst); const vector unsigned char neg1 = (const vector unsigned char)vec_splat_s8(-1); const vector unsigned char dstmask = vec_perm((const vector unsigned char)vzero, neg1, dstperm); const vector unsigned char mperm = (const vector unsigned char) AVV(0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B, 0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F); int16_t *tmpbis = tmp - (tmpStride * 21); vector signed short tmpM2ssA = vec_ld(0, tmpbis); vector signed short tmpM2ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; vector signed short tmpM1ssA = vec_ld(0, tmpbis); vector signed short tmpM1ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; vector signed short tmpP0ssA = vec_ld(0, tmpbis); vector signed short tmpP0ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; vector signed short tmpP1ssA = vec_ld(0, tmpbis); vector signed short tmpP1ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; vector signed short tmpP2ssA = vec_ld(0, tmpbis); vector signed short tmpP2ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; for (i = 0 ; i < 16 ; i++) { const vector signed short tmpP3ssA = vec_ld(0, tmpbis); const vector signed short tmpP3ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; const vector signed short sum1A = vec_adds(tmpP0ssA, tmpP1ssA); const vector signed short sum1B = vec_adds(tmpP0ssB, tmpP1ssB); const vector signed short sum2A = vec_adds(tmpM1ssA, tmpP2ssA); const vector signed short sum2B = vec_adds(tmpM1ssB, tmpP2ssB); const vector signed short sum3A = vec_adds(tmpM2ssA, tmpP3ssA); const vector signed short sum3B = vec_adds(tmpM2ssB, tmpP3ssB); tmpM2ssA = tmpM1ssA; tmpM2ssB = tmpM1ssB; tmpM1ssA = tmpP0ssA; tmpM1ssB = tmpP0ssB; tmpP0ssA = tmpP1ssA; tmpP0ssB = tmpP1ssB; tmpP1ssA = tmpP2ssA; tmpP1ssB = tmpP2ssB; tmpP2ssA = tmpP3ssA; tmpP2ssB = tmpP3ssB; const vector signed int pp1Ae = vec_mule(sum1A, v20ss); const vector signed int pp1Ao = vec_mulo(sum1A, v20ss); const vector signed int pp1Be = vec_mule(sum1B, v20ss); const vector signed int pp1Bo = vec_mulo(sum1B, v20ss); const vector signed int pp2Ae = vec_mule(sum2A, v5ss); const vector signed int pp2Ao = vec_mulo(sum2A, v5ss); const vector signed int pp2Be = vec_mule(sum2B, v5ss); const vector signed int pp2Bo = vec_mulo(sum2B, v5ss); const vector signed int pp3Ae = vec_sra((vector signed int)sum3A, v16ui); const vector signed int pp3Ao = vec_mulo(sum3A, v1ss); const vector signed int pp3Be = vec_sra((vector signed int)sum3B, v16ui); const vector signed int pp3Bo = vec_mulo(sum3B, v1ss); const vector signed int pp1cAe = vec_add(pp1Ae, v512si); const vector signed int pp1cAo = vec_add(pp1Ao, v512si); const vector signed int pp1cBe = vec_add(pp1Be, v512si); const vector signed int pp1cBo = vec_add(pp1Bo, v512si); const vector signed int pp32Ae = vec_sub(pp3Ae, pp2Ae); const vector signed int pp32Ao = vec_sub(pp3Ao, pp2Ao); const vector signed int pp32Be = vec_sub(pp3Be, pp2Be); const vector signed int pp32Bo = vec_sub(pp3Bo, pp2Bo); const vector signed int sumAe = vec_add(pp1cAe, pp32Ae); const vector signed int sumAo = vec_add(pp1cAo, pp32Ao); const vector signed int sumBe = vec_add(pp1cBe, pp32Be); const vector signed int sumBo = vec_add(pp1cBo, pp32Bo); const vector signed int ssumAe = vec_sra(sumAe, v10ui); const vector signed int ssumAo = vec_sra(sumAo, v10ui); const vector signed int ssumBe = vec_sra(sumBe, v10ui); const vector signed int ssumBo = vec_sra(sumBo, v10ui); const vector signed short ssume = vec_packs(ssumAe, ssumBe); const vector signed short ssumo = vec_packs(ssumAo, ssumBo); const vector unsigned char sumv = vec_packsu(ssume, ssumo); const vector unsigned char sum = vec_perm(sumv, sumv, mperm); const vector unsigned char dst1 = vec_ld(0, dst); const vector unsigned char dst2 = vec_ld(16, dst); const vector unsigned char vdst = vec_perm(dst1, dst2, vec_lvsl(0, dst)); vector unsigned char fsum; OP_U8_ALTIVEC(fsum, sum, vdst); const vector unsigned char rsum = vec_perm(fsum, fsum, dstperm); const vector unsigned char fdst1 = vec_sel(dst1, rsum, dstmask); const vector unsigned char fdst2 = vec_sel(rsum, dst2, dstmask); vec_st(fdst1, 0, dst); vec_st(fdst2, 16, dst); dst += dstStride; } POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_hv_lowpass_num, 1); }
void iquant_intra_m1_altivec(IQUANT_INTRA_PDECL) { int i; vector signed short vsrc; uint16_t *qmat; vector unsigned short vqmat; vector unsigned short vmquant; vector bool short eqzero, ltzero; vector signed short val, t0; vector signed short zero, one; vector unsigned int four; vector signed short min, max; int offset, offset2; int16_t dst0; union { vector unsigned short vu16; unsigned short mquant; vector signed int vs32; struct { signed int pad[3]; signed int sum; } s; } vu; #ifdef ALTIVEC_DST DataStreamControl dsc; #endif #ifdef ALTIVEC_VERIFY /* {{{ */ if (NOT_VECTOR_ALIGNED(wsp->intra_q_mat)) mjpeg_error_exit1("iquant_intra_m1: wsp->intra_q_mat %% 16 != 0, (%d)", wsp->intra_q_mat); if (NOT_VECTOR_ALIGNED(src)) mjpeg_error_exit1("iquant_intra_m1: src %% 16 != 0, (%d)", src); if (NOT_VECTOR_ALIGNED(dst)) mjpeg_error_exit1("iquant_intra_m1: dst %% 16 != 0, (%d)", dst); for (i = 0; i < 64; i++) if (src[i] < -256 || src[i] > 255) mjpeg_error_exit1("iquant_intra_m2: -256 > src[%i] > 255, (%d)", i, src[i]); #endif /* }}} */ AMBER_START; dst0 = src[0] << (3 - dc_prec); qmat = (uint16_t*)wsp->intra_q_mat; #ifdef ALTIVEC_DST dsc.control = DATA_STREAM_CONTROL(64/8,1,0); vec_dst(src, dsc.control, 0); vec_dst(qmat, dsc.control, 1); #endif /* vmquant = (vector unsigned short)(mquant); */ vu.mquant = (unsigned short)mquant; vmquant = vec_splat(vu.vu16, 0); zero = vec_splat_s16(0); one = vec_splat_s16(1); four = vec_splat_u32(4); /* max = (2047); min = (-2048); {{{ */ vu8(max) = vec_splat_u8(0x7); t0 = vec_splat_s16(-1); /* 0xffff */ vu8(max) = vec_mergeh(vu8(max), vu8(t0)); /* 0x07ff == 2047 */ min = vec_sub(t0, max); /* }}} */ offset = 0; #if 1 vsrc = vec_ld(offset, (signed short*)src); vqmat = vec_ld(offset, (unsigned short*)qmat); i = (64/8) - 1; do { /* intra_q[i] * mquant */ vu16(vqmat) = vec_mulo(vu8(vqmat), vu8(vmquant)); /* save sign */ ltzero = vec_cmplt(vsrc, zero); eqzero = vec_cmpeq(vsrc, zero); /* val = abs(src) */ t0 = vec_sub(zero, vsrc); val = vec_max(t0, vsrc); /* val = (src * quant) >> 4 */ vs32(t0) = vec_mule(val, vs16(vqmat)); vs32(val) = vec_mulo(val, vs16(vqmat)); vs32(t0) = vec_sra(vs32(t0), four); vs16(t0) = vec_pack(vs32(t0), vs32(t0)); vs32(val) = vec_sra(vs32(val), four); vs16(val) = vec_pack(vs32(val), vs32(val)); val = vec_mergeh(vs16(t0), vs16(val)); offset2 = offset; offset += 8*sizeof(int16_t); vsrc = vec_ld(offset, (signed short*)src); vqmat = vec_ld(offset, (unsigned short*)qmat); /* val = val - 1&~(val|val==0) */ t0 = vec_or(val, eqzero); t0 = vec_andc(one, t0); val = vec_sub(val, t0); /* restore sign */ t0 = vec_sub(zero, val); val = vec_sel(val, t0, ltzero); /* val = (val > 2047) ? ((val < -2048) ? -2048 : val); */ val = vec_min(val, max); val = vec_max(val, min); vec_st(val, offset2, dst); } while (--i); /* intra_q[i] * mquant */ vu16(vqmat) = vec_mulo(vu8(vqmat), vu8(vmquant)); /* save sign */ ltzero = vec_cmplt(vsrc, zero); eqzero = vec_cmpeq(vsrc, zero); /* val = abs(src) */ t0 = vec_sub(zero, vsrc); val = vec_max(t0, vsrc); /* val = (src * quant) >> 4 */ vs32(t0) = vec_mule(val, vs16(vqmat)); vs32(val) = vec_mulo(val, vs16(vqmat)); vs32(t0) = vec_sra(vs32(t0), four); vs16(t0) = vec_pack(vs32(t0), vs32(t0)); vs32(val) = vec_sra(vs32(val), four); vs16(val) = vec_pack(vs32(val), vs32(val)); val = vec_mergeh(vs16(t0), vs16(val)); /* val = val - 1&~(val|val==0) */ t0 = vec_or(val, eqzero); t0 = vec_andc(one, t0); val = vec_sub(val, t0); /* restore sign */ t0 = vec_sub(zero, val); val = vec_sel(val, t0, ltzero); /* val = (val > 2047) ? ((val < -2048) ? -2048 : val); */ val = vec_min(val, max); val = vec_max(val, min); vec_st(val, offset, dst); #else /* {{{ */ i = (64/8); do { vsrc = vec_ld(offset, (signed short*)src); vqmat = vec_ld(offset, (unsigned short*)qmat); /* intra_q[i] * mquant */ vu16(vqmat) = vec_mulo(vu8(vqmat), vu8(vmquant)); /* save sign */ ltzero = vec_cmplt(vsrc, zero); eqzero = vec_cmpeq(vsrc, zero); /* val = abs(src) */ t0 = vec_sub(zero, vsrc); val = vec_max(t0, vsrc); /* val = (src * quant) >> 4 */ vs32(t0) = vec_mule(val, vs16(vqmat)); vs32(val) = vec_mulo(val, vs16(vqmat)); vs32(t0) = vec_sra(vs32(t0), four); vs16(t0) = vec_pack(vs32(t0), vs32(t0)); vs32(val) = vec_sra(vs32(val), four); vs16(val) = vec_pack(vs32(val), vs32(val)); val = vec_mergeh(vs16(t0), vs16(val)); /* val = val - 1&~(val|val==0) */ t0 = vec_or(val, eqzero); t0 = vec_andc(one, t0); val = vec_sub(val, t0); /* restore sign */ t0 = vec_sub(zero, val); val = vec_sel(val, t0, ltzero); /* val = (val > 2047) ? ((val < -2048) ? -2048 : val); */ val = vec_min(val, max); val = vec_max(val, min); vec_st(val, offset, dst); offset += 8*sizeof(int16_t); } while (--i); /* }}} */ #endif dst[0] = dst0; AMBER_STOP; }
void gimp_composite_blend_rgba8_rgba8_rgba8_altivec (GimpCompositeContext *ctx) { const guchar *A = ctx->A; const guchar *B = ctx->B; guchar *D = ctx->D; guint length = ctx->n_pixels; guchar blend = ctx->blend.blend; union { vector unsigned char v; unsigned char u8[16]; } vblend; vector unsigned char vblendc; vector unsigned char a,b,d; vector unsigned short al,ah,bl,bh,one=vec_splat_u16(1); guchar tmp; for (tmp=0; tmp<16; tmp++ ) vblend.u8[tmp]=blend; vblendc=vec_nor(vblend.v,vblend.v); while (length >= 4) { a=LoadUnaligned(A); b=LoadUnaligned(B); /* dest[b] = (src1[b] * blend2 + src2[b] * blend) / 255; * to divide by 255 we use ((n+1)+(n+1)>>8)>>8 * It works for all value but 0xffff * happily blending formula can't give this value */ al=vec_mule(a,vblendc); ah=vec_mulo(a,vblendc); bl=vec_mule(b,vblend.v); bh=vec_mulo(b,vblend.v); al=vec_add(al,bl); al=vec_add(al,one); al=vec_add(al,vec_sr(al,ox0008)); ah=vec_add(ah,bh); ah=vec_add(ah,one); ah=vec_add(ah,vec_sr(ah,ox0008)); d=vec_perm((vector unsigned char)al,(vector unsigned char)ah,combine_high_bytes); StoreUnaligned(d, D); A+=16; B+=16; D+=16; length-=4; } /* process last pixels */ length = length*4; a=LoadUnalignedLess(A, length); b=LoadUnalignedLess(B, length); al=vec_mule(a,vblendc); ah=vec_mulo(a,vblendc); bl=vec_mule(b,vblend.v); bh=vec_mulo(b,vblend.v); al=vec_add(al,bl); al=vec_add(al,one); al=vec_add(al,vec_sr(al,ox0008)); ah=vec_add(ah,bh); ah=vec_add(ah,one); ah=vec_add(ah,vec_sr(ah,ox0008)); d=vec_perm((vector unsigned char)al,(vector unsigned char)ah,combine_high_bytes); StoreUnalignedLess(d, D, length); }
void YUV422_to_BGRA_altivec(const unsigned char *yuvdata, size_t pixelnum, unsigned char *output) { const vector unsigned char *UYVY_ptr=reinterpret_cast<const vector unsigned char *>(yuvdata); vector unsigned char *BGRA_ptr=reinterpret_cast<vector unsigned char *>(output); vector unsigned int vShift; vector signed short tempU, tempV, tempY, tempUV, out1, out2, out3, out4; vector signed short v16, v128, a255, szero, one; vector unsigned char zero; vector signed short t0, t1, t2, tempGB1, tempGB2, tempRA1, tempRA2; vector signed short vU_G, vV_G, vU_B, vU_R, y0, hiImage, loImage; vector unsigned int uv_rEven, uv_rOdd, uv_rHi, uv_rLo, uv_gUEven, uv_gVEven, uv_gUOdd, uv_gVOdd, uv_gHi, uv_gLo, uv_bEven, uv_bOdd; vector signed int tempUhi, tempUlo, tempVhi, tempVlo; vector signed int yEven, yOdd; vector unsigned int t0Even, t0Odd, t1Even, t1Odd, t2Even, t2Odd; /* Load the equation constants. */ vector signed short vConst = static_cast<vector signed short>(298, 519, 409, 16, 128, 255, -100, -210 ); vector unsigned char vPerm1 = static_cast<vector unsigned char>( 0, 1, 16, 17, 8, 9, 24, 25, 2, 3, 18, 19, 10, 11, 26, 27 ); vector unsigned char vPerm2 = static_cast<vector unsigned char>( 4, 5, 20, 21, 12, 13, 28, 29, 6, 7, 22, 23, 14, 15, 30, 31 ); vector unsigned char vPermY = static_cast<vector unsigned char>( 2, 3, 6, 7, 10, 11, 14, 15, 18, 19, 22, 23, 26, 27, 30, 31 ); vector unsigned char vPermU = static_cast<vector unsigned char>( 0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29 ); vector unsigned char vPermV = static_cast<vector unsigned char>( 2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31 ); vector unsigned char vOutPerm1 = static_cast<vector unsigned char>( 0, 1, 2, 3, 16, 17, 18, 19, 4, 5, 6, 7, 20, 21, 22, 23 ); vector unsigned char vOutPerm2 = static_cast<vector unsigned char>( 8, 9, 10, 11, 24, 25, 26, 27, 12, 13, 14, 15, 28, 29, 30, 31 ); vector unsigned char uvPerm = static_cast<vector unsigned char>( 0, 1, 4, 5, 8, 9, 12, 13, 16, 17, 20, 21, 24, 25, 28, 29 ); zero = vec_splat_u8(0); szero = vec_splat_s16(0); one = vec_splat_s16(1); vShift = vec_splat_u32(8); a255 = vec_splat( vConst, 5 ); // alpha channel = 255 vU_G = vec_splat( vConst, 6 ); // -100 vV_G = vec_splat( vConst, 7 ); // -210 vU_B = vec_splat( vConst, 1 ); // 519 vU_R = vec_splat( vConst, 2 ); // 409 y0 = vec_splat( vConst, 0 ); // 298 v16 = vec_splat( vConst, 3 ); // 16 v128 = vec_splat( vConst, 4 ); // 128 for ( unsigned int i = 0; i < (pixelnum/sizeof(vector unsigned char)); i++ ) { // Load UYUV input vector const vector unsigned char *vec1 = UYVY_ptr++; //expand the UInt8's to short's hiImage = static_cast<vector signed short>(vec_mergeh( zero, *vec1 )); loImage = static_cast<vector signed short>(vec_mergel( zero, *vec1 )); tempUV = static_cast<vector signed short>(vec_perm( hiImage, loImage, uvPerm )); tempY = static_cast<vector signed short>(vec_perm( hiImage, loImage, vPermY )); // subtract UV_OFFSET from UV's (should this be saturated?) tempUV = static_cast<vector signed short>(vec_sub( tempUV, v128 )); // subtract Y-OFFSET from Y's (should this be saturated?) tempY = static_cast<vector signed short>(vec_sub( tempY, v16 )); // expand to UUUU UUUU and VVVV VVVV tempU = vec_perm(tempUV, tempUV, vPermU); tempV = vec_perm(tempUV, tempUV, vPermV); //below: // //error: cannot convert `vector int' to `vector unsigned int' in assignment tempUhi = vec_mule( tempU, one ); // unsigned int = vec_mule( signed short, signed short ) // should be // signed int = vec_mule( signed short, signed short ) tempUlo = vec_mulo( tempU, one ); tempVhi = vec_mule( tempV, one ); tempVlo = vec_mulo( tempV, one ); // uv_r = YUV2RGB_12*u + YUV2RGB_13*v // uv_r = (-1)*u + 409*v (or "409*V - U") uv_rEven = vec_mule( tempV, vU_R ); uv_rOdd = vec_mulo( tempV, vU_R ); uv_rHi = vec_sub( uv_rEven, tempUhi ); uv_rLo = vec_sub( uv_rOdd, tempUlo ); // uv_g = YUV2RGB_22*u + YUV2RGB_23*v // uv_g = -100*u + (-210)*v // multiply U by -100 uv_gUEven = vec_mule( tempU, vU_G ); uv_gUOdd = vec_mulo( tempU, vU_G ); // multiply V by -210 uv_gVEven = vec_mule( tempV, vV_G ); uv_gVOdd = vec_mulo( tempV, vV_G ); // add U & V products uv_gHi = vec_add( uv_gUEven, uv_gVEven ); uv_gLo = vec_add( uv_gUOdd, uv_gVOdd ); // uv_b = YUV2RGB_32*u + YUV2RGB_33*v // uv_b = 519*u + 0*v uv_bEven = vec_mule( tempU, vU_B ); uv_bOdd = vec_mulo( tempU, vU_B ); // y = YUV2RGB_11 * tempY // y = 298* (tempY - 16) yEven = vec_mule( tempY, y0 ); yOdd = vec_mulo( tempY, y0 ); // add while int's t0Even = vec_add( yEven, uv_bEven ); t0Odd = vec_add( yOdd, uv_bOdd ); t1Even = vec_add( yEven, uv_gHi ); t1Odd = vec_add( yOdd, uv_gLo ); t2Even = vec_add( yEven, uv_rHi ); t2Odd = vec_add( yOdd, uv_rLo ); // shift while int's t0Even = vec_sra( t0Even, vShift ); t0Odd = vec_sra( t0Odd, vShift ); t1Even = vec_sra( t1Even, vShift ); t1Odd = vec_sra( t1Odd, vShift ); t2Even = vec_sra( t2Even, vShift ); t2Odd = vec_sra( t2Odd, vShift ); // pack down to shorts t0 = vec_packs( t0Even, t0Odd ); t1 = vec_packs( t1Even, t1Odd ); t2 = vec_packs( t2Even, t2Odd ); // Permute to GBGBGBGB GBGBGBGB + re-interleave even & odd tempGB1 = vec_perm( t1, t0, vPerm1 ); tempGB2 = vec_perm( t1, t0, vPerm2 ); // Permute to ARARARAR ARARARAR + re-interleave even & odd tempRA1 = vec_perm( a255, t2, vPerm1 ); tempRA2 = vec_perm( a255, t2, vPerm2 ); // Permute to ARGB's out1 = vec_perm( tempRA1, tempGB1, vOutPerm1 ); out2 = vec_perm( tempRA1, tempGB1, vOutPerm2 ); out3 = vec_perm( tempRA2, tempGB2, vOutPerm1 ); out4 = vec_perm( tempRA2, tempGB2, vOutPerm2 ); // pack down to char's *BGRA_ptr = vec_packsu( out1, out2 ); BGRA_ptr++; *BGRA_ptr = vec_packsu( out3, out4 ); BGRA_ptr++; } }
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 } } }
// these functions do (a*C)>>16 // things are tricky because a is signed, but C unsigned. // M15 is used if C fits in 15 bit unsigned (C6,C7) // M16 is used if C requires 16 bits unsigned static inline vec_s16 M15(vec_s16 a, vec_s16 C) { return (vec_s16)vec_perm(vec_mule(a,C), vec_mulo(a,C), interleave_high); }
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 } }
template<bool align> SIMD_INLINE v128_u16 ReduceColBody(const uint8_t * p) { const v128_u8 t0 = Load<false>(p - 1); const v128_u8 t1 = Load<align>(p); return BinomialSum(vec_mule(t0, K8_01), vec_mule(t1, K8_01), vec_mulo(t1, K8_01)); }
template<bool align> SIMD_INLINE v128_u16 ReduceColNose(const uint8_t * p) { const v128_u8 t = Load<align>(p); return BinomialSum(vec_mule(LoadBeforeFirst<1>(t), K8_01), vec_mule(t, K8_01), vec_mulo(t, K8_01)); }
// CHECK-LABEL: define void @test1 void test1() { /* vec_cmpeq */ res_vbll = vec_cmpeq(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpequd // CHECK-LE: @llvm.ppc.altivec.vcmpequd // CHECK-PPC: error: call to 'vec_cmpeq' is ambiguous res_vbll = vec_cmpeq(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpequd // CHECK-LE: @llvm.ppc.altivec.vcmpequd // CHECK-PPC: error: call to 'vec_cmpeq' is ambiguous /* vec_cmpgt */ res_vbll = vec_cmpgt(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtsd // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd // CHECK-PPC: error: call to 'vec_cmpgt' is ambiguous res_vbll = vec_cmpgt(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud // CHECK-LE: @llvm.ppc.altivec.vcmpgtud // CHECK-PPC: error: call to 'vec_cmpgt' is ambiguous /* ----------------------- predicates --------------------------- */ /* vec_all_eq */ res_i = vec_all_eq(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_eq' is ambiguous res_i = vec_all_eq(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_eq' is ambiguous res_i = vec_all_eq(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_eq' is ambiguous res_i = vec_all_eq(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_eq' is ambiguous res_i = vec_all_eq(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_eq' is ambiguous res_i = vec_all_eq(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_eq' is ambiguous res_i = vec_all_eq(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_eq' is ambiguous /* vec_all_ne */ res_i = vec_all_ne(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_ne' is ambiguous res_i = vec_all_ne(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_ne' is ambiguous res_i = vec_all_ne(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_ne' is ambiguous res_i = vec_all_ne(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_ne' is ambiguous res_i = vec_all_ne(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_ne' is ambiguous res_i = vec_all_ne(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_ne' is ambiguous res_i = vec_all_ne(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_all_ne' is ambiguous /* vec_any_eq */ res_i = vec_any_eq(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_eq' is ambiguous res_i = vec_any_eq(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_eq' is ambiguous res_i = vec_any_eq(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_eq' is ambiguous res_i = vec_any_eq(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_eq' is ambiguous res_i = vec_any_eq(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_eq' is ambiguous res_i = vec_any_eq(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_eq' is ambiguous res_i = vec_any_eq(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_eq' is ambiguous /* vec_any_ne */ res_i = vec_any_ne(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_ne' is ambiguous res_i = vec_any_ne(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_ne' is ambiguous res_i = vec_any_ne(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_ne' is ambiguous res_i = vec_any_ne(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_ne' is ambiguous res_i = vec_any_ne(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_ne' is ambiguous res_i = vec_any_ne(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_ne' is ambiguous res_i = vec_any_ne(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpequd.p // CHECK-LE: @llvm.ppc.altivec.vcmpequd.p // CHECK-PPC: error: call to 'vec_any_ne' is ambiguous /* vec_all_ge */ res_i = vec_all_ge(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_all_ge' is ambiguous res_i = vec_all_ge(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_all_ge' is ambiguous res_i = vec_all_ge(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_ge' is ambiguous res_i = vec_all_ge(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_ge' is ambiguous res_i = vec_all_ge(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_ge' is ambiguous res_i = vec_all_ge(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_ge' is ambiguous res_i = vec_all_ge(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_ge' is ambiguous /* vec_all_gt */ res_i = vec_all_gt(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_all_gt' is ambiguous res_i = vec_all_gt(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_all_gt' is ambiguous res_i = vec_all_gt(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_gt' is ambiguous res_i = vec_all_gt(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_gt' is ambiguous res_i = vec_all_gt(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_gt' is ambiguous res_i = vec_all_gt(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_gt' is ambiguous res_i = vec_all_gt(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_gt' is ambiguous /* vec_all_le */ res_i = vec_all_le(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_all_le' is ambiguous res_i = vec_all_le(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_all_le' is ambiguous res_i = vec_all_le(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_le' is ambiguous res_i = vec_all_le(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_le' is ambiguous res_i = vec_all_le(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_le' is ambiguous res_i = vec_all_le(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_le' is ambiguous res_i = vec_all_le(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_le' is ambiguous /* vec_all_lt */ res_i = vec_all_lt(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_all_lt' is ambiguous res_i = vec_all_lt(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_all_lt' is ambiguous res_i = vec_all_lt(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_lt' is ambiguous res_i = vec_all_lt(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_lt' is ambiguous res_i = vec_all_lt(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_lt' is ambiguous res_i = vec_all_lt(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_lt' is ambiguous res_i = vec_all_lt(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_all_lt' is ambiguous /* vec_any_ge */ res_i = vec_any_ge(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_any_ge' is ambiguous res_i = vec_any_ge(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_any_ge' is ambiguous res_i = vec_any_ge(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_ge' is ambiguous res_i = vec_any_ge(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_ge' is ambiguous res_i = vec_any_ge(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_ge' is ambiguous res_i = vec_any_ge(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_ge' is ambiguous res_i = vec_any_ge(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_ge' is ambiguous /* vec_any_gt */ res_i = vec_any_gt(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_any_gt' is ambiguous res_i = vec_any_gt(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_any_gt' is ambiguous res_i = vec_any_gt(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_gt' is ambiguous res_i = vec_any_gt(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_gt' is ambiguous res_i = vec_any_gt(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_gt' is ambiguous res_i = vec_any_gt(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_gt' is ambiguous res_i = vec_any_gt(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_gt' is ambiguous /* vec_any_le */ res_i = vec_any_le(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_any_le' is ambiguous res_i = vec_any_le(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_any_le' is ambiguous res_i = vec_any_le(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_le' is ambiguous res_i = vec_any_le(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_le' is ambiguous res_i = vec_any_le(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_le' is ambiguous res_i = vec_any_le(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_le' is ambiguous res_i = vec_any_le(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_le' is ambiguous /* vec_any_lt */ res_i = vec_any_lt(vsll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_any_lt' is ambiguous res_i = vec_any_lt(vsll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtsd.p // CHECK-PPC: error: call to 'vec_any_lt' is ambiguous res_i = vec_any_lt(vull, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_lt' is ambiguous res_i = vec_any_lt(vull, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_lt' is ambiguous res_i = vec_any_lt(vbll, vsll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_lt' is ambiguous res_i = vec_any_lt(vbll, vull); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_lt' is ambiguous res_i = vec_any_lt(vbll, vbll); // CHECK: @llvm.ppc.altivec.vcmpgtud.p // CHECK-LE: @llvm.ppc.altivec.vcmpgtud.p // CHECK-PPC: error: call to 'vec_any_lt' is ambiguous /* vec_max */ res_vsll = vec_max(vsll, vsll); // CHECK: @llvm.ppc.altivec.vmaxsd // CHECK-LE: @llvm.ppc.altivec.vmaxsd // CHECK-PPC: error: call to 'vec_max' is ambiguous res_vsll = vec_max(vbll, vsll); // CHECK: @llvm.ppc.altivec.vmaxsd // CHECK-LE: @llvm.ppc.altivec.vmaxsd // CHECK-PPC: error: call to 'vec_max' is ambiguous res_vsll = vec_max(vsll, vbll); // CHECK: @llvm.ppc.altivec.vmaxsd // CHECK-LE: @llvm.ppc.altivec.vmaxsd // CHECK-PPC: error: call to 'vec_max' is ambiguous res_vull = vec_max(vull, vull); // CHECK: @llvm.ppc.altivec.vmaxud // CHECK-LE: @llvm.ppc.altivec.vmaxud // CHECK-PPC: error: call to 'vec_max' is ambiguous res_vull = vec_max(vbll, vull); // CHECK: @llvm.ppc.altivec.vmaxud // CHECK-LE: @llvm.ppc.altivec.vmaxud // CHECK-PPC: error: call to 'vec_max' is ambiguous res_vull = vec_max(vull, vbll); // CHECK: @llvm.ppc.altivec.vmaxud // CHECK-LE: @llvm.ppc.altivec.vmaxud // CHECK-PPC: error: call to 'vec_max' is ambiguous /* vec_min */ res_vsll = vec_min(vsll, vsll); // CHECK: @llvm.ppc.altivec.vminsd // CHECK-LE: @llvm.ppc.altivec.vminsd // CHECK-PPC: error: call to 'vec_min' is ambiguous res_vsll = vec_min(vbll, vsll); // CHECK: @llvm.ppc.altivec.vminsd // CHECK-LE: @llvm.ppc.altivec.vminsd // CHECK-PPC: error: call to 'vec_min' is ambiguous res_vsll = vec_min(vsll, vbll); // CHECK: @llvm.ppc.altivec.vminsd // CHECK-LE: @llvm.ppc.altivec.vminsd // CHECK-PPC: error: call to 'vec_min' is ambiguous res_vull = vec_min(vull, vull); // CHECK: @llvm.ppc.altivec.vminud // CHECK-LE: @llvm.ppc.altivec.vminud // CHECK-PPC: error: call to 'vec_min' is ambiguous res_vull = vec_min(vbll, vull); // CHECK: @llvm.ppc.altivec.vminud // CHECK-LE: @llvm.ppc.altivec.vminud // CHECK-PPC: error: call to 'vec_min' is ambiguous res_vull = vec_min(vull, vbll); // CHECK: @llvm.ppc.altivec.vminud // CHECK-LE: @llvm.ppc.altivec.vminud // CHECK-PPC: error: call to 'vec_min' is ambiguous /* vec_mule */ res_vsll = vec_mule(vi, vi); // CHECK: @llvm.ppc.altivec.vmulesw // CHECK-LE: @llvm.ppc.altivec.vmulosw // CHECK-PPC: error: call to 'vec_mule' is ambiguous res_vull = vec_mule(vui , vui); // CHECK: @llvm.ppc.altivec.vmuleuw // CHECK-LE: @llvm.ppc.altivec.vmulouw // CHECK-PPC: error: call to 'vec_mule' is ambiguous /* vec_mulo */ res_vsll = vec_mulo(vi, vi); // CHECK: @llvm.ppc.altivec.vmulosw // CHECK-LE: @llvm.ppc.altivec.vmulesw // CHECK-PPC: error: call to 'vec_mulo' is ambiguous res_vull = vec_mulo(vui, vui); // CHECK: @llvm.ppc.altivec.vmulouw // CHECK-LE: @llvm.ppc.altivec.vmuleuw // CHECK-PPC: error: call to 'vec_mulo' is ambiguous /* vec_packs */ res_vi = vec_packs(vsll, vsll); // CHECK: @llvm.ppc.altivec.vpksdss // CHECK-LE: @llvm.ppc.altivec.vpksdss // CHECK-PPC: error: call to 'vec_packs' is ambiguous res_vui = vec_packs(vull, vull); // CHECK: @llvm.ppc.altivec.vpkudus // CHECK-LE: @llvm.ppc.altivec.vpkudus // CHECK-PPC: error: call to 'vec_packs' is ambiguous /* vec_packsu */ res_vui = vec_packsu(vsll, vsll); // CHECK: @llvm.ppc.altivec.vpksdus // CHECK-LE: @llvm.ppc.altivec.vpksdus // CHECK-PPC: error: call to 'vec_packsu' is ambiguous res_vui = vec_packsu(vull, vull); // CHECK: @llvm.ppc.altivec.vpkudus // CHECK-LE: @llvm.ppc.altivec.vpkudus // CHECK-PPC: error: call to 'vec_packsu' is ambiguous /* vec_rl */ res_vsll = vec_rl(vsll, vull); // CHECK: @llvm.ppc.altivec.vrld // CHECK-LE: @llvm.ppc.altivec.vrld // CHECK-PPC: error: call to 'vec_rl' is ambiguous res_vull = vec_rl(vull, vull); // CHECK: @llvm.ppc.altivec.vrld // CHECK-LE: @llvm.ppc.altivec.vrld // CHECK-PPC: error: call to 'vec_rl' is ambiguous /* vec_sl */ res_vsll = vec_sl(vsll, vull); // CHECK: shl <2 x i64> // CHECK-LE: shl <2 x i64> // CHECK-PPC: error: call to 'vec_sl' is ambiguous res_vull = vec_sl(vull, vull); // CHECK: shl <2 x i64> // CHECK-LE: shl <2 x i64> // CHECK-PPC: error: call to 'vec_sl' is ambiguous /* vec_sr */ res_vsll = vec_sr(vsll, vull); // CHECK: ashr <2 x i64> // CHECK-LE: ashr <2 x i64> // CHECK-PPC: error: call to 'vec_sr' is ambiguous res_vull = vec_sr(vull, vull); // CHECK: lshr <2 x i64> // CHECK-LE: lshr <2 x i64> // CHECK-PPC: error: call to 'vec_sr' is ambiguous /* vec_sra */ res_vsll = vec_sra(vsll, vull); // CHECK: ashr <2 x i64> // CHECK-LE: ashr <2 x i64> // CHECK-PPC: error: call to 'vec_sra' is ambiguous res_vull = vec_sra(vull, vull); // CHECK: ashr <2 x i64> // CHECK-LE: ashr <2 x i64> // CHECK-PPC: error: call to 'vec_sra' is ambiguous /* vec_unpackh */ res_vsll = vec_unpackh(vi); // CHECK: llvm.ppc.altivec.vupkhsw // CHECK-LE: llvm.ppc.altivec.vupklsw // CHECK-PPC: error: call to 'vec_unpackh' is ambiguous res_vbll = vec_unpackh(vbi); // CHECK: llvm.ppc.altivec.vupkhsw // CHECK-LE: llvm.ppc.altivec.vupklsw // CHECK-PPC: error: call to 'vec_unpackh' is ambiguous /* vec_unpackl */ res_vsll = vec_unpackl(vi); // CHECK: llvm.ppc.altivec.vupklsw // CHECK-LE: llvm.ppc.altivec.vupkhsw // CHECK-PPC: error: call to 'vec_unpackl' is ambiguous res_vbll = vec_unpackl(vbi); // CHECK: llvm.ppc.altivec.vupklsw // CHECK-LE: llvm.ppc.altivec.vupkhsw // CHECK-PPC: error: call to 'vec_unpackl' is ambiguous /* vec_vpksdss */ res_vi = vec_vpksdss(vsll, vsll); // CHECK: llvm.ppc.altivec.vpksdss // CHECK-LE: llvm.ppc.altivec.vpksdss // CHECK-PPC: warning: implicit declaration of function 'vec_vpksdss' /* vec_vpksdus */ res_vui = vec_vpksdus(vsll, vsll); // CHECK: llvm.ppc.altivec.vpksdus // CHECK-LE: llvm.ppc.altivec.vpksdus // CHECK-PPC: warning: implicit declaration of function 'vec_vpksdus' /* vec_vpkudum */ res_vi = vec_vpkudum(vsll, vsll); // CHECK: vperm // CHECK-LE: vperm // CHECK-PPC: warning: implicit declaration of function 'vec_vpkudum' res_vui = vec_vpkudum(vull, vull); // CHECK: vperm // CHECK-LE: vperm res_vui = vec_vpkudus(vull, vull); // CHECK: llvm.ppc.altivec.vpkudus // CHECK-LE: llvm.ppc.altivec.vpkudus // CHECK-PPC: warning: implicit declaration of function 'vec_vpkudus' /* vec_vupkhsw */ res_vsll = vec_vupkhsw(vi); // CHECK: llvm.ppc.altivec.vupkhsw // CHECK-LE: llvm.ppc.altivec.vupklsw // CHECK-PPC: warning: implicit declaration of function 'vec_vupkhsw' res_vbll = vec_vupkhsw(vbi); // CHECK: llvm.ppc.altivec.vupkhsw // CHECK-LE: llvm.ppc.altivec.vupklsw /* vec_vupklsw */ res_vsll = vec_vupklsw(vi); // CHECK: llvm.ppc.altivec.vupklsw // CHECK-LE: llvm.ppc.altivec.vupkhsw // CHECK-PPC: warning: implicit declaration of function 'vec_vupklsw' res_vbll = vec_vupklsw(vbi); // CHECK: llvm.ppc.altivec.vupklsw // CHECK-LE: llvm.ppc.altivec.vupkhsw /* vec_max */ res_vsll = vec_max(vsll, vsll); // CHECK: @llvm.ppc.altivec.vmaxsd // CHECK-LE: @llvm.ppc.altivec.vmaxsd res_vsll = vec_max(vbll, vsll); // CHECK: @llvm.ppc.altivec.vmaxsd // CHECK-LE: @llvm.ppc.altivec.vmaxsd res_vsll = vec_max(vsll, vbll); // CHECK: @llvm.ppc.altivec.vmaxsd // CHECK-LE: @llvm.ppc.altivec.vmaxsd res_vull = vec_max(vull, vull); // CHECK: @llvm.ppc.altivec.vmaxud // CHECK-LE: @llvm.ppc.altivec.vmaxud res_vull = vec_max(vbll, vull); // CHECK: @llvm.ppc.altivec.vmaxud // CHECK-LE: @llvm.ppc.altivec.vmaxud /* vec_min */ res_vsll = vec_min(vsll, vsll); // CHECK: @llvm.ppc.altivec.vminsd // CHECK-LE: @llvm.ppc.altivec.vminsd res_vsll = vec_min(vbll, vsll); // CHECK: @llvm.ppc.altivec.vminsd // CHECK-LE: @llvm.ppc.altivec.vminsd res_vsll = vec_min(vsll, vbll); // CHECK: @llvm.ppc.altivec.vminsd // CHECK-LE: @llvm.ppc.altivec.vminsd res_vull = vec_min(vull, vull); // CHECK: @llvm.ppc.altivec.vminud // CHECK-LE: @llvm.ppc.altivec.vminud res_vull = vec_min(vbll, vull); // CHECK: @llvm.ppc.altivec.vminud // CHECK-LE: @llvm.ppc.altivec.vminud }
SIMD_INLINE v128_s16 Average(const v128_u8 & s0, const v128_u8 & s1) { return (v128_s16)vec_sr(vec_add(vec_add( vec_add(vec_mule(s0, K8_01), vec_mulo(s0, K8_01)), vec_add(vec_mule(s1, K8_01), vec_mulo(s1, K8_01))), K16_0002), K16_0002); }
SIMD_INLINE v128_s16 Average(const v128_u8 & s) { return (v128_s16)vec_sr(vec_add(vec_add(vec_mule(s, K8_01), vec_mulo(s, K8_01)), K16_0001), K16_0001); }
tmpIndex = vec_sub(tmpIndex, (((vector unsigned char){32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32})) ); tmp1 = vec_perm(table[6], table[7], tmpIndex); stmp1 = vec_perm(slope_cos[6], slope_cos[7], tmpIndex); select = (vector unsigned short)vec_cmpgt(PerIndex, (((vector unsigned char){95,95,95,95,95,95,95,95,95,95,95,95,95,95,95,95})) ); tmp3 = vec_sel(tmp0, tmp1, select); stmp3 = vec_sel(stmp0, stmp1, select); select = (vector unsigned short)vec_cmpgt(PerIndex, (((vector unsigned char){63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63})) ); table1 = vec_sel(tmp2, tmp3, select); slope1 = vec_sel(stmp2, stmp3, select); L_tmp0 = vec_sra(vec_mule(slope0, offset0), (((vector unsigned int){12,12,12,12})) ); L_tmp1 = vec_sra(vec_mulo(slope0, offset0), (((vector unsigned int){12,12,12,12})) ); L_tmp2 = vec_sra(vec_mule(slope1, offset1), (((vector unsigned int){12,12,12,12})) ); L_tmp3 = vec_sra(vec_mulo(slope1, offset1), (((vector unsigned int){12,12,12,12})) ); tmp0 = vec_packs(L_tmp0, L_tmp2); tmp1 = vec_packs(L_tmp1, L_tmp3); tmp2 = vec_mergeh(tmp0, tmp1); tmp3 = vec_mergel(tmp0, tmp1); lspq[0] = vec_adds(table0, tmp2); lspq[1] = vec_adds(table1, tmp3); return;
/* 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 */
static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp, uint8_t * src, int dstStride, int tmpStride, int srcStride) { register int i; LOAD_ZERO; const vec_u8 permM2 = vec_lvsl(-2, src); const vec_u8 permM1 = vec_lvsl(-1, src); const vec_u8 permP0 = vec_lvsl(+0, src); const vec_u8 permP1 = vec_lvsl(+1, src); const vec_u8 permP2 = vec_lvsl(+2, src); const vec_u8 permP3 = vec_lvsl(+3, src); const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); const vec_u32 v10ui = vec_splat_u32(10); const vec_s16 v5ss = vec_splat_s16(5); const vec_s16 v1ss = vec_splat_s16(1); const vec_s32 v512si = vec_sl(vec_splat_s32(1),vec_splat_u32(9)); const vec_u32 v16ui = vec_sl(vec_splat_u32(1),vec_splat_u32(4)); register int align = ((((unsigned long)src) - 2) % 16); vec_s16 srcP0A, srcP0B, srcP1A, srcP1B, srcP2A, srcP2B, srcP3A, srcP3B, srcM1A, srcM1B, srcM2A, srcM2B, sum1A, sum1B, sum2A, sum2B, sum3A, sum3B, pp1A, pp1B, pp2A, pp2B, psumA, psumB; const vec_u8 mperm = (const vec_u8) {0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B, 0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F}; int16_t *tmpbis = tmp; vec_s16 tmpM1ssA, tmpM1ssB, tmpM2ssA, tmpM2ssB, tmpP0ssA, tmpP0ssB, tmpP1ssA, tmpP1ssB, tmpP2ssA, tmpP2ssB; vec_s32 pp1Ae, pp1Ao, pp1Be, pp1Bo, pp2Ae, pp2Ao, pp2Be, pp2Bo, pp3Ae, pp3Ao, pp3Be, pp3Bo, pp1cAe, pp1cAo, pp1cBe, pp1cBo, pp32Ae, pp32Ao, pp32Be, pp32Bo, sumAe, sumAo, sumBe, sumBo, ssumAe, ssumAo, ssumBe, ssumBo; vec_u8 fsum, sumv, sum; vec_s16 ssume, ssumo; src -= (2 * srcStride); for (i = 0 ; i < 21 ; i ++) { vec_u8 srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; vec_u8 srcR1 = vec_ld(-2, src); vec_u8 srcR2 = vec_ld(14, src); switch (align) { default: { srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = vec_perm(srcR1, srcR2, permP0); srcP1 = vec_perm(srcR1, srcR2, permP1); srcP2 = vec_perm(srcR1, srcR2, permP2); srcP3 = vec_perm(srcR1, srcR2, permP3); } break; case 11: { srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = vec_perm(srcR1, srcR2, permP0); srcP1 = vec_perm(srcR1, srcR2, permP1); srcP2 = vec_perm(srcR1, srcR2, permP2); srcP3 = srcR2; } break; case 12: { vec_u8 srcR3 = vec_ld(30, src); srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = vec_perm(srcR1, srcR2, permP0); srcP1 = vec_perm(srcR1, srcR2, permP1); srcP2 = srcR2; srcP3 = vec_perm(srcR2, srcR3, permP3); } break; case 13: { vec_u8 srcR3 = vec_ld(30, src); srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = vec_perm(srcR1, srcR2, permP0); srcP1 = srcR2; srcP2 = vec_perm(srcR2, srcR3, permP2); srcP3 = vec_perm(srcR2, srcR3, permP3); } break; case 14: { vec_u8 srcR3 = vec_ld(30, src); srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = vec_perm(srcR1, srcR2, permM1); srcP0 = srcR2; srcP1 = vec_perm(srcR2, srcR3, permP1); srcP2 = vec_perm(srcR2, srcR3, permP2); srcP3 = vec_perm(srcR2, srcR3, permP3); } break; case 15: { vec_u8 srcR3 = vec_ld(30, src); srcM2 = vec_perm(srcR1, srcR2, permM2); srcM1 = srcR2; srcP0 = vec_perm(srcR2, srcR3, permP0); srcP1 = vec_perm(srcR2, srcR3, permP1); srcP2 = vec_perm(srcR2, srcR3, permP2); srcP3 = vec_perm(srcR2, srcR3, permP3); } break; } srcP0A = (vec_s16) vec_mergeh(zero_u8v, srcP0); srcP0B = (vec_s16) vec_mergel(zero_u8v, srcP0); srcP1A = (vec_s16) vec_mergeh(zero_u8v, srcP1); srcP1B = (vec_s16) vec_mergel(zero_u8v, srcP1); srcP2A = (vec_s16) vec_mergeh(zero_u8v, srcP2); srcP2B = (vec_s16) vec_mergel(zero_u8v, srcP2); srcP3A = (vec_s16) vec_mergeh(zero_u8v, srcP3); srcP3B = (vec_s16) vec_mergel(zero_u8v, srcP3); srcM1A = (vec_s16) vec_mergeh(zero_u8v, srcM1); srcM1B = (vec_s16) vec_mergel(zero_u8v, srcM1); srcM2A = (vec_s16) vec_mergeh(zero_u8v, srcM2); srcM2B = (vec_s16) vec_mergel(zero_u8v, srcM2); sum1A = vec_adds(srcP0A, srcP1A); sum1B = vec_adds(srcP0B, srcP1B); sum2A = vec_adds(srcM1A, srcP2A); sum2B = vec_adds(srcM1B, srcP2B); sum3A = vec_adds(srcM2A, srcP3A); sum3B = vec_adds(srcM2B, srcP3B); pp1A = vec_mladd(sum1A, v20ss, sum3A); pp1B = vec_mladd(sum1B, v20ss, sum3B); pp2A = vec_mladd(sum2A, v5ss, zero_s16v); pp2B = vec_mladd(sum2B, v5ss, zero_s16v); psumA = vec_sub(pp1A, pp2A); psumB = vec_sub(pp1B, pp2B); vec_st(psumA, 0, tmp); vec_st(psumB, 16, tmp); src += srcStride; tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */ } tmpM2ssA = vec_ld(0, tmpbis); tmpM2ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; tmpM1ssA = vec_ld(0, tmpbis); tmpM1ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; tmpP0ssA = vec_ld(0, tmpbis); tmpP0ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; tmpP1ssA = vec_ld(0, tmpbis); tmpP1ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; tmpP2ssA = vec_ld(0, tmpbis); tmpP2ssB = vec_ld(16, tmpbis); tmpbis += tmpStride; for (i = 0 ; i < 16 ; i++) { const vec_s16 tmpP3ssA = vec_ld(0, tmpbis); const vec_s16 tmpP3ssB = vec_ld(16, tmpbis); const vec_s16 sum1A = vec_adds(tmpP0ssA, tmpP1ssA); const vec_s16 sum1B = vec_adds(tmpP0ssB, tmpP1ssB); const vec_s16 sum2A = vec_adds(tmpM1ssA, tmpP2ssA); const vec_s16 sum2B = vec_adds(tmpM1ssB, tmpP2ssB); const vec_s16 sum3A = vec_adds(tmpM2ssA, tmpP3ssA); const vec_s16 sum3B = vec_adds(tmpM2ssB, tmpP3ssB); tmpbis += tmpStride; tmpM2ssA = tmpM1ssA; tmpM2ssB = tmpM1ssB; tmpM1ssA = tmpP0ssA; tmpM1ssB = tmpP0ssB; tmpP0ssA = tmpP1ssA; tmpP0ssB = tmpP1ssB; tmpP1ssA = tmpP2ssA; tmpP1ssB = tmpP2ssB; tmpP2ssA = tmpP3ssA; tmpP2ssB = tmpP3ssB; pp1Ae = vec_mule(sum1A, v20ss); pp1Ao = vec_mulo(sum1A, v20ss); pp1Be = vec_mule(sum1B, v20ss); pp1Bo = vec_mulo(sum1B, v20ss); pp2Ae = vec_mule(sum2A, v5ss); pp2Ao = vec_mulo(sum2A, v5ss); pp2Be = vec_mule(sum2B, v5ss); pp2Bo = vec_mulo(sum2B, v5ss); pp3Ae = vec_sra((vec_s32)sum3A, v16ui); pp3Ao = vec_mulo(sum3A, v1ss); pp3Be = vec_sra((vec_s32)sum3B, v16ui); pp3Bo = vec_mulo(sum3B, v1ss); pp1cAe = vec_add(pp1Ae, v512si); pp1cAo = vec_add(pp1Ao, v512si); pp1cBe = vec_add(pp1Be, v512si); pp1cBo = vec_add(pp1Bo, v512si); pp32Ae = vec_sub(pp3Ae, pp2Ae); pp32Ao = vec_sub(pp3Ao, pp2Ao); pp32Be = vec_sub(pp3Be, pp2Be); pp32Bo = vec_sub(pp3Bo, pp2Bo); sumAe = vec_add(pp1cAe, pp32Ae); sumAo = vec_add(pp1cAo, pp32Ao); sumBe = vec_add(pp1cBe, pp32Be); sumBo = vec_add(pp1cBo, pp32Bo); ssumAe = vec_sra(sumAe, v10ui); ssumAo = vec_sra(sumAo, v10ui); ssumBe = vec_sra(sumBe, v10ui); ssumBo = vec_sra(sumBo, v10ui); ssume = vec_packs(ssumAe, ssumBe); ssumo = vec_packs(ssumAo, ssumBo); sumv = vec_packsu(ssume, ssumo); sum = vec_perm(sumv, sumv, mperm); ASSERT_ALIGNED(dst); OP_U8_ALTIVEC(fsum, sum, vec_ld(0, dst)); vec_st(fsum, 0, dst); dst += dstStride; } }
vector signed short test_even (vector signed char x, vector signed char y) { return vec_mule (x, y); }