[Sumover-dev] [svn commit] r3886 - vic/branches/mpeg4/postproc
sumover-dev at cs.ucl.ac.uk
sumover-dev at cs.ucl.ac.uk
Wed Oct 4 18:05:53 BST 2006
Author: barz
Date: Wed Oct 4 18:05:29 2006
New Revision: 3886
Added:
vic/branches/mpeg4/postproc/asmalign.h
vic/branches/mpeg4/postproc/swscale_altivec_template.c
vic/branches/mpeg4/postproc/yuv2rgb_altivec.c
Log:
postproc fixed
Added: vic/branches/mpeg4/postproc/asmalign.h
==============================================================================
--- (empty file)
+++ vic/branches/mpeg4/postproc/asmalign.h Wed Oct 4 18:05:29 2006
@@ -0,0 +1,7 @@
+#ifdef SYS_DARWIN
+#define ASMALIGN8 ".align 3\n\t"
+#define ASMALIGN16 ".align 4\n\t"
+#else
+#define ASMALIGN8 ".balign 8\n\t"
+#define ASMALIGN16 ".balign 16\n\t"
+#endif
Added: vic/branches/mpeg4/postproc/swscale_altivec_template.c
==============================================================================
--- (empty file)
+++ vic/branches/mpeg4/postproc/swscale_altivec_template.c Wed Oct 4 18:05:29 2006
@@ -0,0 +1,541 @@
+/*
+ AltiVec-enhanced yuv2yuvX
+
+ Copyright (C) 2004 Romain Dolbeau <romain at dolbeau.org>
+ based on the equivalent C code in "postproc/swscale.c"
+
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 2 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program; if not, write to the Free Software
+ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+*/
+
+#ifdef CONFIG_DARWIN
+#define AVV(x...) (x)
+#else
+#define AVV(x...) {x}
+#endif
+
+#define vzero vec_splat_s32(0)
+
+static inline void
+altivec_packIntArrayToCharArray(int *val, uint8_t* dest, int dstW) {
+ register int i;
+ vector unsigned int altivec_vectorShiftInt19 =
+ vec_add(vec_splat_u32(10),vec_splat_u32(9));
+ if ((unsigned long)dest % 16) {
+ /* badly aligned store, we force store alignement */
+ /* and will handle load misalignement on val w/ vec_perm */
+ for (i = 0 ; (i < dstW) &&
+ (((unsigned long)dest + i) % 16) ; i++) {
+ int t = val[i] >> 19;
+ dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
+ }
+ vector unsigned char perm1 = vec_lvsl(i << 2, val);
+ vector signed int v1 = vec_ld(i << 2, val);
+ for ( ; i < (dstW - 15); i+=16) {
+ int offset = i << 2;
+ vector signed int v2 = vec_ld(offset + 16, val);
+ vector signed int v3 = vec_ld(offset + 32, val);
+ vector signed int v4 = vec_ld(offset + 48, val);
+ vector signed int v5 = vec_ld(offset + 64, val);
+ vector signed int v12 = vec_perm(v1,v2,perm1);
+ vector signed int v23 = vec_perm(v2,v3,perm1);
+ vector signed int v34 = vec_perm(v3,v4,perm1);
+ vector signed int v45 = vec_perm(v4,v5,perm1);
+
+ vector signed int vA = vec_sra(v12, altivec_vectorShiftInt19);
+ vector signed int vB = vec_sra(v23, altivec_vectorShiftInt19);
+ vector signed int vC = vec_sra(v34, altivec_vectorShiftInt19);
+ vector signed int vD = vec_sra(v45, altivec_vectorShiftInt19);
+ vector unsigned short vs1 = vec_packsu(vA, vB);
+ vector unsigned short vs2 = vec_packsu(vC, vD);
+ vector unsigned char vf = vec_packsu(vs1, vs2);
+ vec_st(vf, i, dest);
+ v1 = v5;
+ }
+ } else { // dest is properly aligned, great
+ for (i = 0; i < (dstW - 15); i+=16) {
+ int offset = i << 2;
+ vector signed int v1 = vec_ld(offset, val);
+ vector signed int v2 = vec_ld(offset + 16, val);
+ vector signed int v3 = vec_ld(offset + 32, val);
+ vector signed int v4 = vec_ld(offset + 48, val);
+ vector signed int v5 = vec_sra(v1, altivec_vectorShiftInt19);
+ vector signed int v6 = vec_sra(v2, altivec_vectorShiftInt19);
+ vector signed int v7 = vec_sra(v3, altivec_vectorShiftInt19);
+ vector signed int v8 = vec_sra(v4, altivec_vectorShiftInt19);
+ vector unsigned short vs1 = vec_packsu(v5, v6);
+ vector unsigned short vs2 = vec_packsu(v7, v8);
+ vector unsigned char vf = vec_packsu(vs1, vs2);
+ vec_st(vf, i, dest);
+ }
+ }
+ for ( ; i < dstW ; i++) {
+ int t = val[i] >> 19;
+ dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
+ }
+}
+
+static inline void
+yuv2yuvX_altivec_real(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
+ int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
+ uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
+{
+ const vector signed int vini = {(1 << 18), (1 << 18), (1 << 18), (1 << 18)};
+ register int i, j;
+ {
+ int __attribute__ ((aligned (16))) val[dstW];
+
+ for (i = 0; i < (dstW -7); i+=4) {
+ vec_st(vini, i << 2, val);
+ }
+ for (; i < dstW; i++) {
+ val[i] = (1 << 18);
+ }
+
+ for (j = 0; j < lumFilterSize; j++) {
+ vector signed short vLumFilter = vec_ld(j << 1, lumFilter);
+ vector unsigned char perm0 = vec_lvsl(j << 1, lumFilter);
+ vLumFilter = vec_perm(vLumFilter, vLumFilter, perm0);
+ vLumFilter = vec_splat(vLumFilter, 0); // lumFilter[j] is loaded 8 times in vLumFilter
+
+ vector unsigned char perm = vec_lvsl(0, lumSrc[j]);
+ vector signed short l1 = vec_ld(0, lumSrc[j]);
+
+ for (i = 0; i < (dstW - 7); i+=8) {
+ int offset = i << 2;
+ vector signed short l2 = vec_ld((i << 1) + 16, lumSrc[j]);
+
+ vector signed int v1 = vec_ld(offset, val);
+ vector signed int v2 = vec_ld(offset + 16, val);
+
+ vector signed short ls = vec_perm(l1, l2, perm); // lumSrc[j][i] ... lumSrc[j][i+7]
+
+ vector signed int i1 = vec_mule(vLumFilter, ls);
+ vector signed int i2 = vec_mulo(vLumFilter, ls);
+
+ vector signed int vf1 = vec_mergeh(i1, i2);
+ vector signed int vf2 = vec_mergel(i1, i2); // lumSrc[j][i] * lumFilter[j] ... lumSrc[j][i+7] * lumFilter[j]
+
+ vector signed int vo1 = vec_add(v1, vf1);
+ vector signed int vo2 = vec_add(v2, vf2);
+
+ vec_st(vo1, offset, val);
+ vec_st(vo2, offset + 16, val);
+
+ l1 = l2;
+ }
+ for ( ; i < dstW; i++) {
+ val[i] += lumSrc[j][i] * lumFilter[j];
+ }
+ }
+ altivec_packIntArrayToCharArray(val,dest,dstW);
+ }
+ if (uDest != 0) {
+ int __attribute__ ((aligned (16))) u[chrDstW];
+ int __attribute__ ((aligned (16))) v[chrDstW];
+
+ for (i = 0; i < (chrDstW -7); i+=4) {
+ vec_st(vini, i << 2, u);
+ vec_st(vini, i << 2, v);
+ }
+ for (; i < chrDstW; i++) {
+ u[i] = (1 << 18);
+ v[i] = (1 << 18);
+ }
+
+ for (j = 0; j < chrFilterSize; j++) {
+ vector signed short vChrFilter = vec_ld(j << 1, chrFilter);
+ vector unsigned char perm0 = vec_lvsl(j << 1, chrFilter);
+ vChrFilter = vec_perm(vChrFilter, vChrFilter, perm0);
+ vChrFilter = vec_splat(vChrFilter, 0); // chrFilter[j] is loaded 8 times in vChrFilter
+
+ vector unsigned char perm = vec_lvsl(0, chrSrc[j]);
+ vector signed short l1 = vec_ld(0, chrSrc[j]);
+ vector signed short l1_V = vec_ld(2048 << 1, chrSrc[j]);
+
+ for (i = 0; i < (chrDstW - 7); i+=8) {
+ int offset = i << 2;
+ vector signed short l2 = vec_ld((i << 1) + 16, chrSrc[j]);
+ vector signed short l2_V = vec_ld(((i + 2048) << 1) + 16, chrSrc[j]);
+
+ vector signed int v1 = vec_ld(offset, u);
+ vector signed int v2 = vec_ld(offset + 16, u);
+ vector signed int v1_V = vec_ld(offset, v);
+ vector signed int v2_V = vec_ld(offset + 16, v);
+
+ vector signed short ls = vec_perm(l1, l2, perm); // chrSrc[j][i] ... chrSrc[j][i+7]
+ vector signed short ls_V = vec_perm(l1_V, l2_V, perm); // chrSrc[j][i+2048] ... chrSrc[j][i+2055]
+
+ vector signed int i1 = vec_mule(vChrFilter, ls);
+ vector signed int i2 = vec_mulo(vChrFilter, ls);
+ vector signed int i1_V = vec_mule(vChrFilter, ls_V);
+ vector signed int i2_V = vec_mulo(vChrFilter, ls_V);
+
+ vector signed int vf1 = vec_mergeh(i1, i2);
+ vector signed int vf2 = vec_mergel(i1, i2); // chrSrc[j][i] * chrFilter[j] ... chrSrc[j][i+7] * chrFilter[j]
+ vector signed int vf1_V = vec_mergeh(i1_V, i2_V);
+ vector signed int vf2_V = vec_mergel(i1_V, i2_V); // chrSrc[j][i] * chrFilter[j] ... chrSrc[j][i+7] * chrFilter[j]
+
+ vector signed int vo1 = vec_add(v1, vf1);
+ vector signed int vo2 = vec_add(v2, vf2);
+ vector signed int vo1_V = vec_add(v1_V, vf1_V);
+ vector signed int vo2_V = vec_add(v2_V, vf2_V);
+
+ vec_st(vo1, offset, u);
+ vec_st(vo2, offset + 16, u);
+ vec_st(vo1_V, offset, v);
+ vec_st(vo2_V, offset + 16, v);
+
+ l1 = l2;
+ l1_V = l2_V;
+ }
+ for ( ; i < chrDstW; i++) {
+ u[i] += chrSrc[j][i] * chrFilter[j];
+ v[i] += chrSrc[j][i + 2048] * chrFilter[j];
+ }
+ }
+ altivec_packIntArrayToCharArray(u,uDest,chrDstW);
+ altivec_packIntArrayToCharArray(v,vDest,chrDstW);
+ }
+}
+
+static inline void hScale_altivec_real(int16_t *dst, int dstW, uint8_t *src, int srcW, int xInc, int16_t *filter, int16_t *filterPos, int filterSize) {
+ register int i;
+ int __attribute__ ((aligned (16))) tempo[4];
+
+ if (filterSize % 4) {
+ for(i=0; i<dstW; i++) {
+ register int j;
+ register int srcPos = filterPos[i];
+ register int val = 0;
+ for(j=0; j<filterSize; j++) {
+ val += ((int)src[srcPos + j])*filter[filterSize*i + j];
+ }
+ dst[i] = MIN(MAX(0, val>>7), (1<<15)-1);
+ }
+ }
+ else
+ switch (filterSize) {
+ case 4:
+ {
+ for(i=0; i<dstW; i++) {
+ register int j;
+ register int srcPos = filterPos[i];
+
+ vector unsigned char src_v0 = vec_ld(srcPos, src);
+ vector unsigned char src_v1;
+ if ((((int)src + srcPos)% 16) > 12) {
+ src_v1 = vec_ld(srcPos + 16, src);
+ }
+ vector unsigned char src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
+
+ vector signed short src_v = // vec_unpackh sign-extends...
+ (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
+ // now put our elements in the even slots
+ src_v = vec_mergeh(src_v, (vector signed short)vzero);
+
+ vector signed short filter_v = vec_ld(i << 3, filter);
+ // the 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2)
+
+ // the neat trick : we only care for half the elements,
+ // high or low depending on (i<<3)%16 (it's 0 or 8 here),
+ // and we're going to use vec_mule, so we chose
+ // carefully how to "unpack" the elements into the even slots
+ if ((i << 3) % 16)
+ filter_v = vec_mergel(filter_v,(vector signed short)vzero);
+ else
+ filter_v = vec_mergeh(filter_v,(vector signed short)vzero);
+
+ vector signed int val_vEven = vec_mule(src_v, filter_v);
+ vector signed int val_s = vec_sums(val_vEven, vzero);
+ vec_st(val_s, 0, tempo);
+ dst[i] = MIN(MAX(0, tempo[3]>>7), (1<<15)-1);
+ }
+ }
+ break;
+
+ case 8:
+ {
+ for(i=0; i<dstW; i++) {
+ register int srcPos = filterPos[i];
+
+ vector unsigned char src_v0 = vec_ld(srcPos, src);
+ vector unsigned char src_v1;
+ if ((((int)src + srcPos)% 16) > 8) {
+ src_v1 = vec_ld(srcPos + 16, src);
+ }
+ vector unsigned char src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
+
+ vector signed short src_v = // vec_unpackh sign-extends...
+ (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
+ vector signed short filter_v = vec_ld(i << 4, filter);
+ // the 4 above is 3 (filterSize == 8) + 1 (sizeof(short) == 2)
+
+ vector signed int val_v = vec_msums(src_v, filter_v, (vector signed int)vzero);
+ vector signed int val_s = vec_sums(val_v, vzero);
+ vec_st(val_s, 0, tempo);
+ dst[i] = MIN(MAX(0, tempo[3]>>7), (1<<15)-1);
+ }
+ }
+ break;
+
+ case 16:
+ {
+ for(i=0; i<dstW; i++) {
+ register int srcPos = filterPos[i];
+
+ vector unsigned char src_v0 = vec_ld(srcPos, src);
+ vector unsigned char src_v1 = vec_ld(srcPos + 16, src);
+ vector unsigned char src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
+
+ vector signed short src_vA = // vec_unpackh sign-extends...
+ (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
+ vector signed short src_vB = // vec_unpackh sign-extends...
+ (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
+
+ vector signed short filter_v0 = vec_ld(i << 5, filter);
+ vector signed short filter_v1 = vec_ld((i << 5) + 16, filter);
+ // the 5 above are 4 (filterSize == 16) + 1 (sizeof(short) == 2)
+
+ vector signed int val_acc = vec_msums(src_vA, filter_v0, (vector signed int)vzero);
+ vector signed int val_v = vec_msums(src_vB, filter_v1, val_acc);
+
+ vector signed int val_s = vec_sums(val_v, vzero);
+
+ vec_st(val_s, 0, tempo);
+ dst[i] = MIN(MAX(0, tempo[3]>>7), (1<<15)-1);
+ }
+ }
+ break;
+
+ default:
+ {
+ for(i=0; i<dstW; i++) {
+ register int j;
+ register int srcPos = filterPos[i];
+
+ vector signed int val_v = (vector signed int)vzero;
+ vector signed short filter_v0R = vec_ld(i * 2 * filterSize, filter);
+ vector unsigned char permF = vec_lvsl((i * 2 * filterSize), filter);
+
+ vector unsigned char src_v0 = vec_ld(srcPos, src);
+ vector unsigned char permS = vec_lvsl(srcPos, src);
+
+ for (j = 0 ; j < filterSize - 15; j += 16) {
+ vector unsigned char src_v1 = vec_ld(srcPos + j + 16, src);
+ vector unsigned char src_vF = vec_perm(src_v0, src_v1, permS);
+
+ vector signed short src_vA = // vec_unpackh sign-extends...
+ (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
+ vector signed short src_vB = // vec_unpackh sign-extends...
+ (vector signed short)(vec_mergel((vector unsigned char)vzero, src_vF));
+
+ vector signed short filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
+ vector signed short filter_v2R = vec_ld((i * 2 * filterSize) + (j * 2) + 32, filter);
+ vector signed short filter_v0 = vec_perm(filter_v0R, filter_v1R, permF);
+ vector signed short filter_v1 = vec_perm(filter_v1R, filter_v2R, permF);
+
+ vector signed int val_acc = vec_msums(src_vA, filter_v0, val_v);
+ val_v = vec_msums(src_vB, filter_v1, val_acc);
+
+ filter_v0R = filter_v2R;
+ src_v0 = src_v1;
+ }
+
+ if (j < (filterSize-7)) {
+ // loading src_v0 is useless, it's already done above
+ //vector unsigned char src_v0 = vec_ld(srcPos + j, src);
+ vector unsigned char src_v1;
+ if ((((int)src + srcPos)% 16) > 8) {
+ src_v1 = vec_ld(srcPos + j + 16, src);
+ }
+ vector unsigned char src_vF = vec_perm(src_v0, src_v1, permS);
+
+ vector signed short src_v = // vec_unpackh sign-extends...
+ (vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
+ // loading filter_v0R is useless, it's already done above
+ //vector signed short filter_v0R = vec_ld((i * 2 * filterSize) + j, filter);
+ vector signed short filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
+ vector signed short filter_v = vec_perm(filter_v0R, filter_v1R, permF);
+
+ val_v = vec_msums(src_v, filter_v, val_v);
+ }
+
+ vector signed int val_s = vec_sums(val_v, vzero);
+
+ vec_st(val_s, 0, tempo);
+ dst[i] = MIN(MAX(0, tempo[3]>>7), (1<<15)-1);
+ }
+
+ }
+ }
+}
+
+static inline int yv12toyuy2_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
+ int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
+ uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
+ // yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
+ uint8_t *ysrc = src[0];
+ uint8_t *usrc = src[1];
+ uint8_t *vsrc = src[2];
+ const int width = c->srcW;
+ const int height = srcSliceH;
+ const int lumStride = srcStride[0];
+ const int chromStride = srcStride[1];
+ const int dstStride = dstStride_a[0];
+ const vector unsigned char yperm = vec_lvsl(0, ysrc);
+ const int vertLumPerChroma = 2;
+ register unsigned int y;
+
+ if(width&15){
+ yv12toyuy2( ysrc, usrc, vsrc, dst,c->srcW,srcSliceH, lumStride, chromStride, dstStride);
+ return srcSliceH;
+ }
+
+ /* this code assume:
+
+ 1) dst is 16 bytes-aligned
+ 2) dstStride is a multiple of 16
+ 3) width is a multiple of 16
+ 4) lum&chrom stride are multiple of 8
+ */
+
+ for(y=0; y<height; y++)
+ {
+ int i;
+ for (i = 0; i < width - 31; i+= 32) {
+ const unsigned int j = i >> 1;
+ vector unsigned char v_yA = vec_ld(i, ysrc);
+ vector unsigned char v_yB = vec_ld(i + 16, ysrc);
+ vector unsigned char v_yC = vec_ld(i + 32, ysrc);
+ vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
+ vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
+ vector unsigned char v_uA = vec_ld(j, usrc);
+ vector unsigned char v_uB = vec_ld(j + 16, usrc);
+ vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
+ vector unsigned char v_vA = vec_ld(j, vsrc);
+ vector unsigned char v_vB = vec_ld(j + 16, vsrc);
+ vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
+ vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
+ vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
+ vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
+ vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
+ vector unsigned char v_yuy2_2 = vec_mergeh(v_y2, v_uv_b);
+ vector unsigned char v_yuy2_3 = vec_mergel(v_y2, v_uv_b);
+ vec_st(v_yuy2_0, (i << 1), dst);
+ vec_st(v_yuy2_1, (i << 1) + 16, dst);
+ vec_st(v_yuy2_2, (i << 1) + 32, dst);
+ vec_st(v_yuy2_3, (i << 1) + 48, dst);
+ }
+ if (i < width) {
+ const unsigned int j = i >> 1;
+ vector unsigned char v_y1 = vec_ld(i, ysrc);
+ vector unsigned char v_u = vec_ld(j, usrc);
+ vector unsigned char v_v = vec_ld(j, vsrc);
+ vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
+ vector unsigned char v_yuy2_0 = vec_mergeh(v_y1, v_uv_a);
+ vector unsigned char v_yuy2_1 = vec_mergel(v_y1, v_uv_a);
+ vec_st(v_yuy2_0, (i << 1), dst);
+ vec_st(v_yuy2_1, (i << 1) + 16, dst);
+ }
+ if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
+ {
+ usrc += chromStride;
+ vsrc += chromStride;
+ }
+ ysrc += lumStride;
+ dst += dstStride;
+ }
+
+ return srcSliceH;
+}
+
+static inline int yv12touyvy_unscaled_altivec(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
+ int srcSliceH, uint8_t* dstParam[], int dstStride_a[]) {
+ uint8_t *dst=dstParam[0] + dstStride_a[0]*srcSliceY;
+ // yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
+ uint8_t *ysrc = src[0];
+ uint8_t *usrc = src[1];
+ uint8_t *vsrc = src[2];
+ const int width = c->srcW;
+ const int height = srcSliceH;
+ const int lumStride = srcStride[0];
+ const int chromStride = srcStride[1];
+ const int dstStride = dstStride_a[0];
+ const int vertLumPerChroma = 2;
+ const vector unsigned char yperm = vec_lvsl(0, ysrc);
+ register unsigned int y;
+
+ if(width&15){
+ yv12touyvy( ysrc, usrc, vsrc, dst,c->srcW,srcSliceH, lumStride, chromStride, dstStride);
+ return srcSliceH;
+ }
+
+ /* this code assume:
+
+ 1) dst is 16 bytes-aligned
+ 2) dstStride is a multiple of 16
+ 3) width is a multiple of 16
+ 4) lum&chrom stride are multiple of 8
+ */
+
+ for(y=0; y<height; y++)
+ {
+ int i;
+ for (i = 0; i < width - 31; i+= 32) {
+ const unsigned int j = i >> 1;
+ vector unsigned char v_yA = vec_ld(i, ysrc);
+ vector unsigned char v_yB = vec_ld(i + 16, ysrc);
+ vector unsigned char v_yC = vec_ld(i + 32, ysrc);
+ vector unsigned char v_y1 = vec_perm(v_yA, v_yB, yperm);
+ vector unsigned char v_y2 = vec_perm(v_yB, v_yC, yperm);
+ vector unsigned char v_uA = vec_ld(j, usrc);
+ vector unsigned char v_uB = vec_ld(j + 16, usrc);
+ vector unsigned char v_u = vec_perm(v_uA, v_uB, vec_lvsl(j, usrc));
+ vector unsigned char v_vA = vec_ld(j, vsrc);
+ vector unsigned char v_vB = vec_ld(j + 16, vsrc);
+ vector unsigned char v_v = vec_perm(v_vA, v_vB, vec_lvsl(j, vsrc));
+ vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
+ vector unsigned char v_uv_b = vec_mergel(v_u, v_v);
+ vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
+ vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
+ vector unsigned char v_uyvy_2 = vec_mergeh(v_uv_b, v_y2);
+ vector unsigned char v_uyvy_3 = vec_mergel(v_uv_b, v_y2);
+ vec_st(v_uyvy_0, (i << 1), dst);
+ vec_st(v_uyvy_1, (i << 1) + 16, dst);
+ vec_st(v_uyvy_2, (i << 1) + 32, dst);
+ vec_st(v_uyvy_3, (i << 1) + 48, dst);
+ }
+ if (i < width) {
+ const unsigned int j = i >> 1;
+ vector unsigned char v_y1 = vec_ld(i, ysrc);
+ vector unsigned char v_u = vec_ld(j, usrc);
+ vector unsigned char v_v = vec_ld(j, vsrc);
+ vector unsigned char v_uv_a = vec_mergeh(v_u, v_v);
+ vector unsigned char v_uyvy_0 = vec_mergeh(v_uv_a, v_y1);
+ vector unsigned char v_uyvy_1 = vec_mergel(v_uv_a, v_y1);
+ vec_st(v_uyvy_0, (i << 1), dst);
+ vec_st(v_uyvy_1, (i << 1) + 16, dst);
+ }
+ if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
+ {
+ usrc += chromStride;
+ vsrc += chromStride;
+ }
+ ysrc += lumStride;
+ dst += dstStride;
+ }
+ return srcSliceH;
+}
Added: vic/branches/mpeg4/postproc/yuv2rgb_altivec.c
==============================================================================
--- (empty file)
+++ vic/branches/mpeg4/postproc/yuv2rgb_altivec.c Wed Oct 4 18:05:29 2006
@@ -0,0 +1,953 @@
+/*
+ marc.hoffman at analog.com March 8, 2004
+
+ Altivec Acceleration for Color Space Conversion revision 0.2
+
+ convert I420 YV12 to RGB in various formats,
+ it rejects images that are not in 420 formats
+ it rejects images that don't have widths of multiples of 16
+ it rejects images that don't have heights of multiples of 2
+ reject defers to C simulation codes.
+
+ lots of optimizations to be done here
+
+ 1. need to fix saturation code, I just couldn't get it to fly with packs and adds.
+ so we currently use max min to clip
+
+ 2. the inefficient use of chroma loading needs a bit of brushing up
+
+ 3. analysis of pipeline stalls needs to be done, use shark to identify pipeline stalls
+
+
+ MODIFIED to calculate coeffs from currently selected color space.
+ MODIFIED core to be a macro which you spec the output format.
+ ADDED UYVY conversion which is never called due to some thing in SWSCALE.
+ CORRECTED algorithim selection to be strict on input formats.
+ ADDED runtime detection of altivec.
+
+ ADDED altivec_yuv2packedX vertical scl + RGB converter
+
+ March 27,2004
+ PERFORMANCE ANALYSIS
+
+ The C version use 25% of the processor or ~250Mips for D1 video rawvideo used as test
+ The ALTIVEC version uses 10% of the processor or ~100Mips for D1 video same sequence
+
+ 720*480*30 ~10MPS
+
+ so we have roughly 10clocks per pixel this is too high something has to be wrong.
+
+ OPTIMIZED clip codes to utilize vec_max and vec_packs removing the need for vec_min.
+
+ OPTIMIZED DST OUTPUT cache/dma controls. we are pretty much
+ guaranteed to have the input video frame it was just decompressed so
+ it probably resides in L1 caches. However we are creating the
+ output video stream this needs to use the DSTST instruction to
+ optimize for the cache. We couple this with the fact that we are
+ not going to be visiting the input buffer again so we mark it Least
+ Recently Used. This shaves 25% of the processor cycles off.
+
+ Now MEMCPY is the largest mips consumer in the system, probably due
+ to the inefficient X11 stuff.
+
+ GL libraries seem to be very slow on this machine 1.33Ghz PB running
+ Jaguar, this is not the case for my 1Ghz PB. I thought it might be
+ a versioning issues, however i have libGL.1.2.dylib for both
+ machines. ((We need to figure this out now))
+
+ GL2 libraries work now with patch for RGB32
+
+ NOTE quartz vo driver ARGB32_to_RGB24 consumes 30% of the processor
+
+ Integrated luma prescaling adjustment for saturation/contrast/brightness adjustment.
+
+*/
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <inttypes.h>
+#include <assert.h>
+#include "config.h"
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif
+#include "rgb2rgb.h"
+#include "swscale.h"
+#include "swscale_internal.h"
+#include "mangle.h"
+#include "img_format.h" //FIXME try to reduce dependency of such stuff
+
+#undef PROFILE_THE_BEAST
+#undef INC_SCALING
+
+typedef unsigned char ubyte;
+typedef signed char sbyte;
+
+
+/* RGB interleaver, 16 planar pels 8-bit samples per channel in
+ homogeneous vector registers x0,x1,x2 are interleaved with the
+ following technique:
+
+ o0 = vec_mergeh (x0,x1);
+ o1 = vec_perm (o0, x2, perm_rgb_0);
+ o2 = vec_perm (o0, x2, perm_rgb_1);
+ o3 = vec_mergel (x0,x1);
+ o4 = vec_perm (o3,o2,perm_rgb_2);
+ o5 = vec_perm (o3,o2,perm_rgb_3);
+
+ perm_rgb_0: o0(RG).h v1(B) --> o1*
+ 0 1 2 3 4
+ rgbr|gbrg|brgb|rgbr
+ 0010 0100 1001 0010
+ 0102 3145 2673 894A
+
+ perm_rgb_1: o0(RG).h v1(B) --> o2
+ 0 1 2 3 4
+ gbrg|brgb|bbbb|bbbb
+ 0100 1001 1111 1111
+ B5CD 6EF7 89AB CDEF
+
+ perm_rgb_2: o3(RG).l o2(rgbB.l) --> o4*
+ 0 1 2 3 4
+ gbrg|brgb|rgbr|gbrg
+ 1111 1111 0010 0100
+ 89AB CDEF 0182 3945
+
+ perm_rgb_2: o3(RG).l o2(rgbB.l) ---> o5*
+ 0 1 2 3 4
+ brgb|rgbr|gbrg|brgb
+ 1001 0010 0100 1001
+ a67b 89cA BdCD eEFf
+
+*/
+static
+const vector unsigned char
+ perm_rgb_0 = (const vector unsigned char)AVV(0x00,0x01,0x10,0x02,0x03,0x11,0x04,0x05,
+ 0x12,0x06,0x07,0x13,0x08,0x09,0x14,0x0a),
+ perm_rgb_1 = (const vector unsigned char)AVV(0x0b,0x15,0x0c,0x0d,0x16,0x0e,0x0f,0x17,
+ 0x18,0x19,0x1a,0x1b,0x1c,0x1d,0x1e,0x1f),
+ perm_rgb_2 = (const vector unsigned char)AVV(0x10,0x11,0x12,0x13,0x14,0x15,0x16,0x17,
+ 0x00,0x01,0x18,0x02,0x03,0x19,0x04,0x05),
+ perm_rgb_3 = (const vector unsigned char)AVV(0x1a,0x06,0x07,0x1b,0x08,0x09,0x1c,0x0a,
+ 0x0b,0x1d,0x0c,0x0d,0x1e,0x0e,0x0f,0x1f);
+
+#define vec_merge3(x2,x1,x0,y0,y1,y2) \
+do { \
+ typeof(x0) o0,o2,o3; \
+ o0 = vec_mergeh (x0,x1); \
+ y0 = vec_perm (o0, x2, perm_rgb_0);\
+ o2 = vec_perm (o0, x2, perm_rgb_1);\
+ o3 = vec_mergel (x0,x1); \
+ y1 = vec_perm (o3,o2,perm_rgb_2); \
+ y2 = vec_perm (o3,o2,perm_rgb_3); \
+} while(0)
+
+#define vec_mstbgr24(x0,x1,x2,ptr) \
+do { \
+ typeof(x0) _0,_1,_2; \
+ vec_merge3 (x0,x1,x2,_0,_1,_2); \
+ vec_st (_0, 0, ptr++); \
+ vec_st (_1, 0, ptr++); \
+ vec_st (_2, 0, ptr++); \
+} while (0);
+
+#define vec_mstrgb24(x0,x1,x2,ptr) \
+do { \
+ typeof(x0) _0,_1,_2; \
+ vec_merge3 (x2,x1,x0,_0,_1,_2); \
+ vec_st (_0, 0, ptr++); \
+ vec_st (_1, 0, ptr++); \
+ vec_st (_2, 0, ptr++); \
+} while (0);
+
+/* pack the pixels in rgb0 format
+ msb R
+ lsb 0
+*/
+#define vec_mstrgb32(T,x0,x1,x2,x3,ptr) \
+do { \
+ T _0,_1,_2,_3; \
+ _0 = vec_mergeh (x0,x1); \
+ _1 = vec_mergeh (x2,x3); \
+ _2 = (T)vec_mergeh ((vector unsigned short)_0,(vector unsigned short)_1); \
+ _3 = (T)vec_mergel ((vector unsigned short)_0,(vector unsigned short)_1); \
+ vec_st (_2, 0*16, (T *)ptr); \
+ vec_st (_3, 1*16, (T *)ptr); \
+ _0 = vec_mergel (x0,x1); \
+ _1 = vec_mergel (x2,x3); \
+ _2 = (T)vec_mergeh ((vector unsigned short)_0,(vector unsigned short)_1); \
+ _3 = (T)vec_mergel ((vector unsigned short)_0,(vector unsigned short)_1); \
+ vec_st (_2, 2*16, (T *)ptr); \
+ vec_st (_3, 3*16, (T *)ptr); \
+ ptr += 4; \
+} while (0);
+
+/*
+
+ | 1 0 1.4021 | | Y |
+ | 1 -0.3441 -0.7142 |x| Cb|
+ | 1 1.7718 0 | | Cr|
+
+
+ Y: [-128 127]
+ Cb/Cr : [-128 127]
+
+ typical yuv conversion work on Y: 0-255 this version has been optimized for jpeg decode.
+
+*/
+
+
+
+
+#define vec_unh(x) \
+ (vector signed short) \
+ vec_perm(x,(typeof(x))AVV(0),\
+ (vector unsigned char)AVV(0x10,0x00,0x10,0x01,0x10,0x02,0x10,0x03,\
+ 0x10,0x04,0x10,0x05,0x10,0x06,0x10,0x07))
+#define vec_unl(x) \
+ (vector signed short) \
+ vec_perm(x,(typeof(x))AVV(0),\
+ (vector unsigned char)AVV(0x10,0x08,0x10,0x09,0x10,0x0A,0x10,0x0B,\
+ 0x10,0x0C,0x10,0x0D,0x10,0x0E,0x10,0x0F))
+
+#define vec_clip_s16(x) \
+ vec_max (vec_min (x, (vector signed short)AVV(235,235,235,235,235,235,235,235)),\
+ (vector signed short)AVV(16, 16, 16, 16, 16, 16, 16, 16 ))
+
+#define vec_packclp(x,y) \
+ (vector unsigned char)vec_packs \
+ ((vector unsigned short)vec_max (x,(vector signed short) AVV(0)), \
+ (vector unsigned short)vec_max (y,(vector signed short) AVV(0)))
+
+//#define out_pixels(a,b,c,ptr) vec_mstrgb32(typeof(a),((typeof (a))AVV(0)),a,a,a,ptr)
+
+
+static inline void cvtyuvtoRGB (SwsContext *c,
+ vector signed short Y, vector signed short U, vector signed short V,
+ vector signed short *R, vector signed short *G, vector signed short *B)
+{
+ vector signed short vx,ux,uvx;
+
+ Y = vec_mradds (Y, c->CY, c->OY);
+ U = vec_sub (U,(vector signed short)
+ vec_splat((vector signed short)AVV(128),0));
+ V = vec_sub (V,(vector signed short)
+ vec_splat((vector signed short)AVV(128),0));
+
+ // ux = (CBU*(u<<c->CSHIFT)+0x4000)>>15;
+ ux = vec_sl (U, c->CSHIFT);
+ *B = vec_mradds (ux, c->CBU, Y);
+
+ // vx = (CRV*(v<<c->CSHIFT)+0x4000)>>15;
+ vx = vec_sl (V, c->CSHIFT);
+ *R = vec_mradds (vx, c->CRV, Y);
+
+ // uvx = ((CGU*u) + (CGV*v))>>15;
+ uvx = vec_mradds (U, c->CGU, Y);
+ *G = vec_mradds (V, c->CGV, uvx);
+}
+
+
+/*
+ ------------------------------------------------------------------------------
+ CS converters
+ ------------------------------------------------------------------------------
+*/
+
+
+#define DEFCSP420_CVT(name,out_pixels) \
+static int altivec_##name (SwsContext *c, \
+ unsigned char **in, int *instrides, \
+ int srcSliceY, int srcSliceH, \
+ unsigned char **oplanes, int *outstrides) \
+{ \
+ int w = c->srcW; \
+ int h = srcSliceH; \
+ int i,j; \
+ int instrides_scl[3]; \
+ vector unsigned char y0,y1; \
+ \
+ vector signed char u,v; \
+ \
+ vector signed short Y0,Y1,Y2,Y3; \
+ vector signed short U,V; \
+ vector signed short vx,ux,uvx; \
+ vector signed short vx0,ux0,uvx0; \
+ vector signed short vx1,ux1,uvx1; \
+ vector signed short R0,G0,B0; \
+ vector signed short R1,G1,B1; \
+ vector unsigned char R,G,B; \
+ \
+ vector unsigned char *y1ivP, *y2ivP, *uivP, *vivP; \
+ vector unsigned char align_perm; \
+ \
+ vector signed short \
+ lCY = c->CY, \
+ lOY = c->OY, \
+ lCRV = c->CRV, \
+ lCBU = c->CBU, \
+ lCGU = c->CGU, \
+ lCGV = c->CGV; \
+ \
+ vector unsigned short lCSHIFT = c->CSHIFT; \
+ \
+ ubyte *y1i = in[0]; \
+ ubyte *y2i = in[0]+instrides[0]; \
+ ubyte *ui = in[1]; \
+ ubyte *vi = in[2]; \
+ \
+ vector unsigned char *oute \
+ = (vector unsigned char *) \
+ (oplanes[0]+srcSliceY*outstrides[0]); \
+ vector unsigned char *outo \
+ = (vector unsigned char *) \
+ (oplanes[0]+srcSliceY*outstrides[0]+outstrides[0]); \
+ \
+ \
+ instrides_scl[0] = instrides[0]*2-w; /* the loop moves y{1,2}i by w */ \
+ instrides_scl[1] = instrides[1]-w/2; /* the loop moves ui by w/2 */ \
+ instrides_scl[2] = instrides[2]-w/2; /* the loop moves vi by w/2 */ \
+ \
+ \
+ for (i=0;i<h/2;i++) { \
+ vec_dstst (outo, (0x02000002|(((w*3+32)/32)<<16)), 0); \
+ vec_dstst (oute, (0x02000002|(((w*3+32)/32)<<16)), 1); \
+ \
+ for (j=0;j<w/16;j++) { \
+ \
+ y1ivP = (vector unsigned char *)y1i; \
+ y2ivP = (vector unsigned char *)y2i; \
+ uivP = (vector unsigned char *)ui; \
+ vivP = (vector unsigned char *)vi; \
+ \
+ align_perm = vec_lvsl (0, y1i); \
+ y0 = (vector unsigned char)vec_perm (y1ivP[0], y1ivP[1], align_perm);\
+ \
+ align_perm = vec_lvsl (0, y2i); \
+ y1 = (vector unsigned char)vec_perm (y2ivP[0], y2ivP[1], align_perm);\
+ \
+ align_perm = vec_lvsl (0, ui); \
+ u = (vector signed char)vec_perm (uivP[0], uivP[1], align_perm); \
+ \
+ align_perm = vec_lvsl (0, vi); \
+ v = (vector signed char)vec_perm (vivP[0], vivP[1], align_perm); \
+ \
+ u = (vector signed char) \
+ vec_sub (u,(vector signed char) \
+ vec_splat((vector signed char)AVV(128),0));\
+ v = (vector signed char) \
+ vec_sub (v,(vector signed char) \
+ vec_splat((vector signed char)AVV(128),0));\
+ \
+ U = vec_unpackh (u); \
+ V = vec_unpackh (v); \
+ \
+ \
+ Y0 = vec_unh (y0); \
+ Y1 = vec_unl (y0); \
+ Y2 = vec_unh (y1); \
+ Y3 = vec_unl (y1); \
+ \
+ Y0 = vec_mradds (Y0, lCY, lOY); \
+ Y1 = vec_mradds (Y1, lCY, lOY); \
+ Y2 = vec_mradds (Y2, lCY, lOY); \
+ Y3 = vec_mradds (Y3, lCY, lOY); \
+ \
+ /* ux = (CBU*(u<<CSHIFT)+0x4000)>>15 */ \
+ ux = vec_sl (U, lCSHIFT); \
+ ux = vec_mradds (ux, lCBU, (vector signed short)AVV(0)); \
+ ux0 = vec_mergeh (ux,ux); \
+ ux1 = vec_mergel (ux,ux); \
+ \
+ /* vx = (CRV*(v<<CSHIFT)+0x4000)>>15; */ \
+ vx = vec_sl (V, lCSHIFT); \
+ vx = vec_mradds (vx, lCRV, (vector signed short)AVV(0)); \
+ vx0 = vec_mergeh (vx,vx); \
+ vx1 = vec_mergel (vx,vx); \
+ \
+ /* uvx = ((CGU*u) + (CGV*v))>>15 */ \
+ uvx = vec_mradds (U, lCGU, (vector signed short)AVV(0)); \
+ uvx = vec_mradds (V, lCGV, uvx); \
+ uvx0 = vec_mergeh (uvx,uvx); \
+ uvx1 = vec_mergel (uvx,uvx); \
+ \
+ R0 = vec_add (Y0,vx0); \
+ G0 = vec_add (Y0,uvx0); \
+ B0 = vec_add (Y0,ux0); \
+ R1 = vec_add (Y1,vx1); \
+ G1 = vec_add (Y1,uvx1); \
+ B1 = vec_add (Y1,ux1); \
+ \
+ R = vec_packclp (R0,R1); \
+ G = vec_packclp (G0,G1); \
+ B = vec_packclp (B0,B1); \
+ \
+ out_pixels(R,G,B,oute); \
+ \
+ R0 = vec_add (Y2,vx0); \
+ G0 = vec_add (Y2,uvx0); \
+ B0 = vec_add (Y2,ux0); \
+ R1 = vec_add (Y3,vx1); \
+ G1 = vec_add (Y3,uvx1); \
+ B1 = vec_add (Y3,ux1); \
+ R = vec_packclp (R0,R1); \
+ G = vec_packclp (G0,G1); \
+ B = vec_packclp (B0,B1); \
+ \
+ \
+ out_pixels(R,G,B,outo); \
+ \
+ y1i += 16; \
+ y2i += 16; \
+ ui += 8; \
+ vi += 8; \
+ \
+ } \
+ \
+ outo += (outstrides[0])>>4; \
+ oute += (outstrides[0])>>4; \
+ \
+ ui += instrides_scl[1]; \
+ vi += instrides_scl[2]; \
+ y1i += instrides_scl[0]; \
+ y2i += instrides_scl[0]; \
+ } \
+ return srcSliceH; \
+}
+
+
+#define out_abgr(a,b,c,ptr) vec_mstrgb32(typeof(a),((typeof (a))AVV(0)),c,b,a,ptr)
+#define out_bgra(a,b,c,ptr) vec_mstrgb32(typeof(a),c,b,a,((typeof (a))AVV(0)),ptr)
+#define out_rgba(a,b,c,ptr) vec_mstrgb32(typeof(a),a,b,c,((typeof (a))AVV(0)),ptr)
+#define out_argb(a,b,c,ptr) vec_mstrgb32(typeof(a),((typeof (a))AVV(0)),a,b,c,ptr)
+#define out_rgb24(a,b,c,ptr) vec_mstrgb24(a,b,c,ptr)
+#define out_bgr24(a,b,c,ptr) vec_mstbgr24(a,b,c,ptr)
+
+DEFCSP420_CVT (yuv2_abgr, out_abgr)
+#if 1
+DEFCSP420_CVT (yuv2_bgra, out_bgra)
+#else
+static int altivec_yuv2_bgra32 (SwsContext *c,
+ unsigned char **in, int *instrides,
+ int srcSliceY, int srcSliceH,
+ unsigned char **oplanes, int *outstrides)
+{
+ int w = c->srcW;
+ int h = srcSliceH;
+ int i,j;
+ int instrides_scl[3];
+ vector unsigned char y0,y1;
+
+ vector signed char u,v;
+
+ vector signed short Y0,Y1,Y2,Y3;
+ vector signed short U,V;
+ vector signed short vx,ux,uvx;
+ vector signed short vx0,ux0,uvx0;
+ vector signed short vx1,ux1,uvx1;
+ vector signed short R0,G0,B0;
+ vector signed short R1,G1,B1;
+ vector unsigned char R,G,B;
+
+ vector unsigned char *uivP, *vivP;
+ vector unsigned char align_perm;
+
+ vector signed short
+ lCY = c->CY,
+ lOY = c->OY,
+ lCRV = c->CRV,
+ lCBU = c->CBU,
+ lCGU = c->CGU,
+ lCGV = c->CGV;
+
+ vector unsigned short lCSHIFT = c->CSHIFT;
+
+ ubyte *y1i = in[0];
+ ubyte *y2i = in[0]+w;
+ ubyte *ui = in[1];
+ ubyte *vi = in[2];
+
+ vector unsigned char *oute
+ = (vector unsigned char *)
+ (oplanes[0]+srcSliceY*outstrides[0]);
+ vector unsigned char *outo
+ = (vector unsigned char *)
+ (oplanes[0]+srcSliceY*outstrides[0]+outstrides[0]);
+
+
+ instrides_scl[0] = instrides[0];
+ instrides_scl[1] = instrides[1]-w/2; /* the loop moves ui by w/2 */
+ instrides_scl[2] = instrides[2]-w/2; /* the loop moves vi by w/2 */
+
+
+ for (i=0;i<h/2;i++) {
+ vec_dstst (outo, (0x02000002|(((w*3+32)/32)<<16)), 0);
+ vec_dstst (oute, (0x02000002|(((w*3+32)/32)<<16)), 1);
+
+ for (j=0;j<w/16;j++) {
+
+ y0 = vec_ldl (0,y1i);
+ y1 = vec_ldl (0,y2i);
+ uivP = (vector unsigned char *)ui;
+ vivP = (vector unsigned char *)vi;
+
+ align_perm = vec_lvsl (0, ui);
+ u = (vector signed char)vec_perm (uivP[0], uivP[1], align_perm);
+
+ align_perm = vec_lvsl (0, vi);
+ v = (vector signed char)vec_perm (vivP[0], vivP[1], align_perm);
+ u = (vector signed char)
+ vec_sub (u,(vector signed char)
+ vec_splat((vector signed char)AVV(128),0));
+
+ v = (vector signed char)
+ vec_sub (v, (vector signed char)
+ vec_splat((vector signed char)AVV(128),0));
+
+ U = vec_unpackh (u);
+ V = vec_unpackh (v);
+
+
+ Y0 = vec_unh (y0);
+ Y1 = vec_unl (y0);
+ Y2 = vec_unh (y1);
+ Y3 = vec_unl (y1);
+
+ Y0 = vec_mradds (Y0, lCY, lOY);
+ Y1 = vec_mradds (Y1, lCY, lOY);
+ Y2 = vec_mradds (Y2, lCY, lOY);
+ Y3 = vec_mradds (Y3, lCY, lOY);
+
+ /* ux = (CBU*(u<<CSHIFT)+0x4000)>>15 */
+ ux = vec_sl (U, lCSHIFT);
+ ux = vec_mradds (ux, lCBU, (vector signed short)AVV(0));
+ ux0 = vec_mergeh (ux,ux);
+ ux1 = vec_mergel (ux,ux);
+
+ /* vx = (CRV*(v<<CSHIFT)+0x4000)>>15; */
+ vx = vec_sl (V, lCSHIFT);
+ vx = vec_mradds (vx, lCRV, (vector signed short)AVV(0));
+ vx0 = vec_mergeh (vx,vx);
+ vx1 = vec_mergel (vx,vx);
+ /* uvx = ((CGU*u) + (CGV*v))>>15 */
+ uvx = vec_mradds (U, lCGU, (vector signed short)AVV(0));
+ uvx = vec_mradds (V, lCGV, uvx);
+ uvx0 = vec_mergeh (uvx,uvx);
+ uvx1 = vec_mergel (uvx,uvx);
+ R0 = vec_add (Y0,vx0);
+ G0 = vec_add (Y0,uvx0);
+ B0 = vec_add (Y0,ux0);
+ R1 = vec_add (Y1,vx1);
+ G1 = vec_add (Y1,uvx1);
+ B1 = vec_add (Y1,ux1);
+ R = vec_packclp (R0,R1);
+ G = vec_packclp (G0,G1);
+ B = vec_packclp (B0,B1);
+
+ out_argb(R,G,B,oute);
+ R0 = vec_add (Y2,vx0);
+ G0 = vec_add (Y2,uvx0);
+ B0 = vec_add (Y2,ux0);
+ R1 = vec_add (Y3,vx1);
+ G1 = vec_add (Y3,uvx1);
+ B1 = vec_add (Y3,ux1);
+ R = vec_packclp (R0,R1);
+ G = vec_packclp (G0,G1);
+ B = vec_packclp (B0,B1);
+
+ out_argb(R,G,B,outo);
+ y1i += 16;
+ y2i += 16;
+ ui += 8;
+ vi += 8;
+
+ }
+
+ outo += (outstrides[0])>>4;
+ oute += (outstrides[0])>>4;
+
+ ui += instrides_scl[1];
+ vi += instrides_scl[2];
+ y1i += instrides_scl[0];
+ y2i += instrides_scl[0];
+ }
+ return srcSliceH;
+}
+
+#endif
+
+
+DEFCSP420_CVT (yuv2_rgba, out_rgba)
+DEFCSP420_CVT (yuv2_argb, out_argb)
+DEFCSP420_CVT (yuv2_rgb24, out_rgb24)
+DEFCSP420_CVT (yuv2_bgr24, out_bgr24)
+
+
+// uyvy|uyvy|uyvy|uyvy
+// 0123 4567 89ab cdef
+static
+const vector unsigned char
+ demux_u = (const vector unsigned char)AVV(0x10,0x00,0x10,0x00,
+ 0x10,0x04,0x10,0x04,
+ 0x10,0x08,0x10,0x08,
+ 0x10,0x0c,0x10,0x0c),
+ demux_v = (const vector unsigned char)AVV(0x10,0x02,0x10,0x02,
+ 0x10,0x06,0x10,0x06,
+ 0x10,0x0A,0x10,0x0A,
+ 0x10,0x0E,0x10,0x0E),
+ demux_y = (const vector unsigned char)AVV(0x10,0x01,0x10,0x03,
+ 0x10,0x05,0x10,0x07,
+ 0x10,0x09,0x10,0x0B,
+ 0x10,0x0D,0x10,0x0F);
+
+/*
+ this is so I can play live CCIR raw video
+*/
+static int altivec_uyvy_rgb32 (SwsContext *c,
+ unsigned char **in, int *instrides,
+ int srcSliceY, int srcSliceH,
+ unsigned char **oplanes, int *outstrides)
+{
+ int w = c->srcW;
+ int h = srcSliceH;
+ int i,j;
+ vector unsigned char uyvy;
+ vector signed short Y,U,V;
+ vector signed short vx,ux,uvx;
+ vector signed short R0,G0,B0,R1,G1,B1;
+ vector unsigned char R,G,B;
+ vector unsigned char *out;
+ ubyte *img;
+
+ img = in[0];
+ out = (vector unsigned char *)(oplanes[0]+srcSliceY*outstrides[0]);
+
+ for (i=0;i<h;i++) {
+ for (j=0;j<w/16;j++) {
+ uyvy = vec_ld (0, img);
+ U = (vector signed short)
+ vec_perm (uyvy, (vector unsigned char)AVV(0), demux_u);
+
+ V = (vector signed short)
+ vec_perm (uyvy, (vector unsigned char)AVV(0), demux_v);
+
+ Y = (vector signed short)
+ vec_perm (uyvy, (vector unsigned char)AVV(0), demux_y);
+
+ cvtyuvtoRGB (c, Y,U,V,&R0,&G0,&B0);
+
+ uyvy = vec_ld (16, img);
+ U = (vector signed short)
+ vec_perm (uyvy, (vector unsigned char)AVV(0), demux_u);
+
+ V = (vector signed short)
+ vec_perm (uyvy, (vector unsigned char)AVV(0), demux_v);
+
+ Y = (vector signed short)
+ vec_perm (uyvy, (vector unsigned char)AVV(0), demux_y);
+
+ cvtyuvtoRGB (c, Y,U,V,&R1,&G1,&B1);
+
+ R = vec_packclp (R0,R1);
+ G = vec_packclp (G0,G1);
+ B = vec_packclp (B0,B1);
+
+ // vec_mstbgr24 (R,G,B, out);
+ out_rgba (R,G,B,out);
+
+ img += 32;
+ }
+ }
+ return srcSliceH;
+}
+
+
+
+/* Ok currently the acceleration routine only supports
+ inputs of widths a multiple of 16
+ and heights a multiple 2
+
+ So we just fall back to the C codes for this.
+*/
+SwsFunc yuv2rgb_init_altivec (SwsContext *c)
+{
+ if (!(c->flags & SWS_CPU_CAPS_ALTIVEC))
+ return NULL;
+
+ /*
+ and this seems not to matter too much I tried a bunch of
+ videos with abnormal widths and mplayer crashes else where.
+ mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv
+ boom with X11 bad match.
+
+ */
+ if ((c->srcW & 0xf) != 0) return NULL;
+
+ switch (c->srcFormat) {
+ case IMGFMT_YVU9:
+ case IMGFMT_IF09:
+ case IMGFMT_YV12:
+ case IMGFMT_I420:
+ case IMGFMT_IYUV:
+ case IMGFMT_CLPL:
+ case IMGFMT_Y800:
+ case IMGFMT_Y8:
+ case IMGFMT_NV12:
+ case IMGFMT_NV21:
+ if ((c->srcH & 0x1) != 0)
+ return NULL;
+
+ switch(c->dstFormat){
+ case IMGFMT_RGB24:
+ MSG_WARN("ALTIVEC: Color Space RGB24\n");
+ return altivec_yuv2_rgb24;
+ case IMGFMT_BGR24:
+ MSG_WARN("ALTIVEC: Color Space BGR24\n");
+ return altivec_yuv2_bgr24;
+ case IMGFMT_ARGB:
+ MSG_WARN("ALTIVEC: Color Space ARGB\n");
+ return altivec_yuv2_argb;
+ case IMGFMT_ABGR:
+ MSG_WARN("ALTIVEC: Color Space ABGR\n");
+ return altivec_yuv2_abgr;
+ case IMGFMT_RGBA:
+ MSG_WARN("ALTIVEC: Color Space RGBA\n");
+ return altivec_yuv2_rgba;
+ case IMGFMT_BGRA:
+ MSG_WARN("ALTIVEC: Color Space BGRA\n");
+ return altivec_yuv2_bgra;
+ default: return NULL;
+ }
+ break;
+
+ case IMGFMT_UYVY:
+ switch(c->dstFormat){
+ case IMGFMT_RGB32:
+ MSG_WARN("ALTIVEC: Color Space UYVY -> RGB32\n");
+ return altivec_uyvy_rgb32;
+ default: return NULL;
+ }
+ break;
+
+ }
+ return NULL;
+}
+
+static uint16_t roundToInt16(int64_t f){
+ int r= (f + (1<<15))>>16;
+ if(r<-0x7FFF) return 0x8000;
+ else if(r> 0x7FFF) return 0x7FFF;
+ else return r;
+}
+
+void yuv2rgb_altivec_init_tables (SwsContext *c, const int inv_table[4],int brightness,int contrast, int saturation)
+{
+ union {
+ signed short tmp[8] __attribute__ ((aligned(16)));
+ vector signed short vec;
+ } buf;
+
+ buf.tmp[0] = ( (0xffffLL) * contrast>>8 )>>9; //cy
+ buf.tmp[1] = -256*brightness; //oy
+ buf.tmp[2] = (inv_table[0]>>3) *(contrast>>16)*(saturation>>16); //crv
+ buf.tmp[3] = (inv_table[1]>>3) *(contrast>>16)*(saturation>>16); //cbu
+ buf.tmp[4] = -((inv_table[2]>>1)*(contrast>>16)*(saturation>>16)); //cgu
+ buf.tmp[5] = -((inv_table[3]>>1)*(contrast>>16)*(saturation>>16)); //cgv
+
+
+ c->CSHIFT = (vector unsigned short)vec_splat_u16(2);
+ c->CY = vec_splat ((vector signed short)buf.vec, 0);
+ c->OY = vec_splat ((vector signed short)buf.vec, 1);
+ c->CRV = vec_splat ((vector signed short)buf.vec, 2);
+ c->CBU = vec_splat ((vector signed short)buf.vec, 3);
+ c->CGU = vec_splat ((vector signed short)buf.vec, 4);
+ c->CGV = vec_splat ((vector signed short)buf.vec, 5);
+#if 0
+{
+int i;
+char *v[6]={"cy","oy","crv","cbu","cgu","cgv"};
+for (i=0; i<6;i++)
+ printf("%s %d ", v[i],buf.tmp[i] );
+ printf("\n");
+}
+#endif
+ return;
+}
+
+
+void
+altivec_yuv2packedX (SwsContext *c,
+ int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
+ int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
+ uint8_t *dest, int dstW, int dstY)
+{
+ int i,j;
+ short *f;
+ vector signed short X,X0,X1,Y0,U0,V0,Y1,U1,V1,U,V;
+ vector signed short R0,G0,B0,R1,G1,B1;
+
+ vector unsigned char R,G,B,pels[3];
+ vector unsigned char *out,*nout;
+
+ vector signed short RND = vec_splat_s16(1<<3);
+ vector unsigned short SCL = vec_splat_u16(4);
+ unsigned long scratch[16] __attribute__ ((aligned (16)));
+
+ vector signed short *YCoeffs, *CCoeffs;
+
+ YCoeffs = c->vYCoeffsBank+dstY*lumFilterSize;
+ CCoeffs = c->vCCoeffsBank+dstY*chrFilterSize;
+
+ out = (vector unsigned char *)dest;
+
+ for(i=0; i<dstW; i+=16){
+ Y0 = RND;
+ Y1 = RND;
+ /* extract 16 coeffs from lumSrc */
+ for(j=0; j<lumFilterSize; j++) {
+ X0 = vec_ld (0, &lumSrc[j][i]);
+ X1 = vec_ld (16, &lumSrc[j][i]);
+ Y0 = vec_mradds (X0, YCoeffs[j], Y0);
+ Y1 = vec_mradds (X1, YCoeffs[j], Y1);
+ }
+
+ U = RND;
+ V = RND;
+ /* extract 8 coeffs from U,V */
+ for(j=0; j<chrFilterSize; j++) {
+ X = vec_ld (0, &chrSrc[j][i/2]);
+ U = vec_mradds (X, CCoeffs[j], U);
+ X = vec_ld (0, &chrSrc[j][i/2+2048]);
+ V = vec_mradds (X, CCoeffs[j], V);
+ }
+
+ /* scale and clip signals */
+ Y0 = vec_sra (Y0, SCL);
+ Y1 = vec_sra (Y1, SCL);
+ U = vec_sra (U, SCL);
+ V = vec_sra (V, SCL);
+
+ Y0 = vec_clip_s16 (Y0);
+ Y1 = vec_clip_s16 (Y1);
+ U = vec_clip_s16 (U);
+ V = vec_clip_s16 (V);
+
+ /* now we have
+ Y0= y0 y1 y2 y3 y4 y5 y6 y7 Y1= y8 y9 y10 y11 y12 y13 y14 y15
+ U= u0 u1 u2 u3 u4 u5 u6 u7 V= v0 v1 v2 v3 v4 v5 v6 v7
+
+ Y0= y0 y1 y2 y3 y4 y5 y6 y7 Y1= y8 y9 y10 y11 y12 y13 y14 y15
+ U0= u0 u0 u1 u1 u2 u2 u3 u3 U1= u4 u4 u5 u5 u6 u6 u7 u7
+ V0= v0 v0 v1 v1 v2 v2 v3 v3 V1= v4 v4 v5 v5 v6 v6 v7 v7
+ */
+
+ U0 = vec_mergeh (U,U);
+ V0 = vec_mergeh (V,V);
+
+ U1 = vec_mergel (U,U);
+ V1 = vec_mergel (V,V);
+
+ cvtyuvtoRGB (c, Y0,U0,V0,&R0,&G0,&B0);
+ cvtyuvtoRGB (c, Y1,U1,V1,&R1,&G1,&B1);
+
+ R = vec_packclp (R0,R1);
+ G = vec_packclp (G0,G1);
+ B = vec_packclp (B0,B1);
+
+ switch(c->dstFormat) {
+ case IMGFMT_ABGR: out_abgr (R,G,B,out); break;
+ case IMGFMT_BGRA: out_bgra (R,G,B,out); break;
+ case IMGFMT_RGBA: out_rgba (R,G,B,out); break;
+ case IMGFMT_ARGB: out_argb (R,G,B,out); break;
+ case IMGFMT_RGB24: out_rgb24 (R,G,B,out); break;
+ case IMGFMT_BGR24: out_bgr24 (R,G,B,out); break;
+ default:
+ {
+ /* If this is reached, the caller should have called yuv2packedXinC
+ instead. */
+ static int printed_error_message;
+ if(!printed_error_message) {
+ MSG_ERR("altivec_yuv2packedX doesn't support %s output\n",
+ vo_format_name(c->dstFormat));
+ printed_error_message=1;
+ }
+ return;
+ }
+ }
+ }
+
+ if (i < dstW) {
+ i -= 16;
+
+ Y0 = RND;
+ Y1 = RND;
+ /* extract 16 coeffs from lumSrc */
+ for(j=0; j<lumFilterSize; j++) {
+ X0 = vec_ld (0, &lumSrc[j][i]);
+ X1 = vec_ld (16, &lumSrc[j][i]);
+ Y0 = vec_mradds (X0, YCoeffs[j], Y0);
+ Y1 = vec_mradds (X1, YCoeffs[j], Y1);
+ }
+
+ U = RND;
+ V = RND;
+ /* extract 8 coeffs from U,V */
+ for(j=0; j<chrFilterSize; j++) {
+ X = vec_ld (0, &chrSrc[j][i/2]);
+ U = vec_mradds (X, CCoeffs[j], U);
+ X = vec_ld (0, &chrSrc[j][i/2+2048]);
+ V = vec_mradds (X, CCoeffs[j], V);
+ }
+
+ /* scale and clip signals */
+ Y0 = vec_sra (Y0, SCL);
+ Y1 = vec_sra (Y1, SCL);
+ U = vec_sra (U, SCL);
+ V = vec_sra (V, SCL);
+
+ Y0 = vec_clip_s16 (Y0);
+ Y1 = vec_clip_s16 (Y1);
+ U = vec_clip_s16 (U);
+ V = vec_clip_s16 (V);
+
+ /* now we have
+ Y0= y0 y1 y2 y3 y4 y5 y6 y7 Y1= y8 y9 y10 y11 y12 y13 y14 y15
+ U= u0 u1 u2 u3 u4 u5 u6 u7 V= v0 v1 v2 v3 v4 v5 v6 v7
+
+ Y0= y0 y1 y2 y3 y4 y5 y6 y7 Y1= y8 y9 y10 y11 y12 y13 y14 y15
+ U0= u0 u0 u1 u1 u2 u2 u3 u3 U1= u4 u4 u5 u5 u6 u6 u7 u7
+ V0= v0 v0 v1 v1 v2 v2 v3 v3 V1= v4 v4 v5 v5 v6 v6 v7 v7
+ */
+
+ U0 = vec_mergeh (U,U);
+ V0 = vec_mergeh (V,V);
+
+ U1 = vec_mergel (U,U);
+ V1 = vec_mergel (V,V);
+
+ cvtyuvtoRGB (c, Y0,U0,V0,&R0,&G0,&B0);
+ cvtyuvtoRGB (c, Y1,U1,V1,&R1,&G1,&B1);
+
+ R = vec_packclp (R0,R1);
+ G = vec_packclp (G0,G1);
+ B = vec_packclp (B0,B1);
+
+ nout = (vector unsigned char *)scratch;
+ switch(c->dstFormat) {
+ case IMGFMT_ABGR: out_abgr (R,G,B,nout); break;
+ case IMGFMT_BGRA: out_bgra (R,G,B,nout); break;
+ case IMGFMT_RGBA: out_rgba (R,G,B,nout); break;
+ case IMGFMT_ARGB: out_argb (R,G,B,nout); break;
+ case IMGFMT_RGB24: out_rgb24 (R,G,B,nout); break;
+ case IMGFMT_BGR24: out_bgr24 (R,G,B,nout); break;
+ default:
+ /* Unreachable, I think. */
+ MSG_ERR("altivec_yuv2packedX doesn't support %s output\n",
+ vo_format_name(c->dstFormat));
+ return;
+ }
+
+ memcpy (&((uint32_t*)dest)[i], scratch, (dstW-i)/4);
+ }
+
+}
More information about the Sumover-dev
mailing list