swscale: K&R formatting cosmetics for PowerPC code (part I/II)

This commit is contained in:
Diego Biurrun 2012-02-21 19:58:48 +01:00
parent 3f486e0dae
commit 04217de4d6
2 changed files with 315 additions and 304 deletions

View File

@ -22,6 +22,7 @@
*/
#include <inttypes.h>
#include "config.h"
#include "libswscale/swscale.h"
#include "libswscale/swscale_internal.h"
@ -30,8 +31,8 @@
#define vzero vec_splat_s32(0)
static inline void
altivec_packIntArrayToCharArray(int *val, uint8_t* dest, int dstW)
static inline void altivec_packIntArrayToCharArray(int *val, uint8_t *dest,
int dstW)
{
register int i;
vector unsigned int altivec_vectorShiftInt19 =
@ -41,106 +42,104 @@ altivec_packIntArrayToCharArray(int *val, uint8_t* dest, int dstW)
/* and will handle load misalignment on val w/ vec_perm */
vector unsigned char perm1;
vector signed int v1;
for (i = 0 ; (i < dstW) &&
(((uintptr_t)dest + i) % 16) ; i++) {
int t = val[i] >> 19;
dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
for (i = 0; (i < dstW) &&
(((uintptr_t)dest + i) % 16); i++) {
int t = val[i] >> 19;
dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
}
perm1 = vec_lvsl(i << 2, val);
v1 = vec_ld(i << 2, val);
for ( ; i < (dstW - 15); i+=16) {
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 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 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);
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) {
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 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);
vector unsigned char vf = vec_packsu(vs1, vs2);
vec_st(vf, i, dest);
}
}
for ( ; i < dstW ; i++) {
for (; i < dstW; i++) {
int t = val[i] >> 19;
dest[i] = (t < 0) ? 0 : ((t > 255) ? 255 : t);
}
}
//FIXME remove the usage of scratch buffers.
static void
yuv2planeX_altivec(const int16_t *filter, int filterSize,
const int16_t **src, uint8_t *dest, int dstW,
const uint8_t *dither, int offset)
// FIXME remove the usage of scratch buffers.
static void yuv2planeX_altivec(const int16_t *filter, int filterSize,
const int16_t **src, uint8_t *dest, int dstW,
const uint8_t *dither, int offset)
{
register int i, j;
DECLARE_ALIGNED(16, int, val)[dstW];
DECLARE_ALIGNED(16, int, val)[dstW];
for (i=0; i<dstW; i++)
val[i] = dither[(i + offset) & 7] << 12;
for (i = 0; i < dstW; i++)
val[i] = dither[(i + offset) & 7] << 12;
for (j = 0; j < filterSize; j++) {
vector signed short l1, vLumFilter = vec_ld(j << 1, filter);
vector unsigned char perm, perm0 = vec_lvsl(j << 1, filter);
vLumFilter = vec_perm(vLumFilter, vLumFilter, perm0);
vLumFilter = vec_splat(vLumFilter, 0); // lumFilter[j] is loaded 8 times in vLumFilter
for (j = 0; j < filterSize; j++) {
vector signed short l1, vLumFilter = vec_ld(j << 1, filter);
vector unsigned char perm, perm0 = vec_lvsl(j << 1, filter);
vLumFilter = vec_perm(vLumFilter, vLumFilter, perm0);
vLumFilter = vec_splat(vLumFilter, 0); // lumFilter[j] is loaded 8 times in vLumFilter
perm = vec_lvsl(0, src[j]);
l1 = vec_ld(0, src[j]);
perm = vec_lvsl(0, src[j]);
l1 = vec_ld(0, src[j]);
for (i = 0; i < (dstW - 7); i+=8) {
int offset = i << 2;
vector signed short l2 = vec_ld((i << 1) + 16, src[j]);
for (i = 0; i < (dstW - 7); i += 8) {
int offset = i << 2;
vector signed short l2 = vec_ld((i << 1) + 16, src[j]);
vector signed int v1 = vec_ld(offset, val);
vector signed int v2 = vec_ld(offset + 16, val);
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 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 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 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);
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);
vec_st(vo1, offset, val);
vec_st(vo2, offset + 16, val);
l1 = l2;
}
for ( ; i < dstW; i++) {
val[i] += src[j][i] * filter[j];
}
l1 = l2;
}
altivec_packIntArrayToCharArray(val, dest, dstW);
for (; i < dstW; i++)
val[i] += src[j][i] * filter[j];
}
altivec_packIntArrayToCharArray(val, dest, dstW);
}
static void hScale_altivec_real(SwsContext *c, int16_t *dst, int dstW,
@ -151,167 +150,164 @@ static void hScale_altivec_real(SwsContext *c, int16_t *dst, int dstW,
DECLARE_ALIGNED(16, int, tempo)[4];
if (filterSize % 4) {
for (i=0; i<dstW; i++) {
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];
register int val = 0;
for (j = 0; j < filterSize; j++)
val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
dst[i] = FFMIN(val >> 7, (1 << 15) - 1);
}
} else
switch (filterSize) {
case 4:
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, src_vF;
vector signed short src_v, filter_v;
vector signed int val_vEven, val_s;
if ((((uintptr_t)src + srcPos) % 16) > 12) {
src_v1 = vec_ld(srcPos + 16, src);
}
src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
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);
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 choose
// 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);
val_vEven = vec_mule(src_v, filter_v);
val_s = vec_sums(val_vEven, vzero);
vec_st(val_s, 0, tempo);
dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1);
}
dst[i] = FFMIN(val>>7, (1<<15)-1);
}
}
else
switch (filterSize) {
case 4:
for (i=0; i<dstW; i++) {
register int srcPos = filterPos[i];
break;
vector unsigned char src_v0 = vec_ld(srcPos, src);
vector unsigned char src_v1, src_vF;
vector signed short src_v, filter_v;
vector signed int val_vEven, val_s;
if ((((uintptr_t)src + srcPos) % 16) > 12) {
src_v1 = vec_ld(srcPos + 16, src);
}
src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
case 8:
for (i = 0; i < dstW; i++) {
register int srcPos = filterPos[i];
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 unsigned char src_v0 = vec_ld(srcPos, src);
vector unsigned char src_v1, src_vF;
vector signed short src_v, filter_v;
vector signed int val_v, val_s;
if ((((uintptr_t)src + srcPos) % 16) > 8) {
src_v1 = vec_ld(srcPos + 16, src);
}
src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
filter_v = vec_ld(i << 3, filter);
// The 3 above is 2 (filterSize == 4) + 1 (sizeof(short) == 2).
src_v = // vec_unpackh sign-extends...
(vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
filter_v = vec_ld(i << 4, filter);
// the 4 above is 3 (filterSize == 8) + 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 choose
// 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);
val_vEven = vec_mule(src_v, filter_v);
val_s = vec_sums(val_vEven, vzero);
vec_st(val_s, 0, tempo);
dst[i] = FFMIN(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, src_vF;
vector signed short src_v, filter_v;
vector signed int val_v, val_s;
if ((((uintptr_t)src + srcPos) % 16) > 8) {
src_v1 = vec_ld(srcPos + 16, src);
}
src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
src_v = // vec_unpackh sign-extends...
(vector signed short)(vec_mergeh((vector unsigned char)vzero, src_vF));
filter_v = vec_ld(i << 4, filter);
// the 4 above is 3 (filterSize == 8) + 1 (sizeof(short) == 2)
val_v = vec_msums(src_v, filter_v, (vector signed int)vzero);
val_s = vec_sums(val_v, vzero);
vec_st(val_s, 0, tempo);
dst[i] = FFMIN(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] = FFMIN(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_s, 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, src_vF;
vector signed short src_v, filter_v1R, filter_v;
if ((((uintptr_t)src + srcPos) % 16) > 8) {
src_v1 = vec_ld(srcPos + j + 16, src);
val_v = vec_msums(src_v, filter_v, (vector signed int)vzero);
val_s = vec_sums(val_v, vzero);
vec_st(val_s, 0, tempo);
dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1);
}
src_vF = vec_perm(src_v0, src_v1, permS);
break;
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);
filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
filter_v = vec_perm(filter_v0R, filter_v1R, permF);
case 16:
for (i = 0; i < dstW; i++) {
register int srcPos = filterPos[i];
val_v = vec_msums(src_v, filter_v, val_v);
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] = FFMIN(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_s, 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, src_vF;
vector signed short src_v, filter_v1R, filter_v;
if ((((uintptr_t)src + srcPos) % 16) > 8) {
src_v1 = vec_ld(srcPos + j + 16, src);
}
src_vF = vec_perm(src_v0, src_v1, permS);
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);
filter_v1R = vec_ld((i * 2 * filterSize) + (j * 2) + 16, filter);
filter_v = vec_perm(filter_v0R, filter_v1R, permF);
val_v = vec_msums(src_v, filter_v, val_v);
}
val_s = vec_sums(val_v, vzero);
vec_st(val_s, 0, tempo);
dst[i] = FFMIN(tempo[3] >> 7, (1 << 15) - 1);
}
}
val_s = vec_sums(val_v, vzero);
vec_st(val_s, 0, tempo);
dst[i] = FFMIN(tempo[3]>>7, (1<<15)-1);
}
}
}
void ff_sws_init_swScale_altivec(SwsContext *c)
@ -334,12 +330,24 @@ void ff_sws_init_swScale_altivec(SwsContext *c)
* match what's found in the body of ff_yuv2packedX_altivec() */
if (!(c->flags & (SWS_BITEXACT | SWS_FULL_CHR_H_INT)) && !c->alpPixBuf) {
switch (c->dstFormat) {
case PIX_FMT_ABGR: c->yuv2packedX = ff_yuv2abgr_X_altivec; break;
case PIX_FMT_BGRA: c->yuv2packedX = ff_yuv2bgra_X_altivec; break;
case PIX_FMT_ARGB: c->yuv2packedX = ff_yuv2argb_X_altivec; break;
case PIX_FMT_RGBA: c->yuv2packedX = ff_yuv2rgba_X_altivec; break;
case PIX_FMT_BGR24: c->yuv2packedX = ff_yuv2bgr24_X_altivec; break;
case PIX_FMT_RGB24: c->yuv2packedX = ff_yuv2rgb24_X_altivec; break;
case PIX_FMT_ABGR:
c->yuv2packedX = ff_yuv2abgr_X_altivec;
break;
case PIX_FMT_BGRA:
c->yuv2packedX = ff_yuv2bgra_X_altivec;
break;
case PIX_FMT_ARGB:
c->yuv2packedX = ff_yuv2argb_X_altivec;
break;
case PIX_FMT_RGBA:
c->yuv2packedX = ff_yuv2rgba_X_altivec;
break;
case PIX_FMT_BGR24:
c->yuv2packedX = ff_yuv2bgr24_X_altivec;
break;
case PIX_FMT_RGB24:
c->yuv2packedX = ff_yuv2rgb24_X_altivec;
break;
}
}
}

View File

@ -22,55 +22,57 @@
*/
#include <inttypes.h>
#include "config.h"
#include "libswscale/swscale.h"
#include "libswscale/swscale_internal.h"
#include "libavutil/cpu.h"
static int yv12toyuy2_unscaled_altivec(SwsContext *c, const uint8_t* src[],
static int yv12toyuy2_unscaled_altivec(SwsContext *c, const uint8_t *src[],
int srcStride[], int srcSliceY,
int srcSliceH, uint8_t* dstParam[],
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]);
const uint8_t *ysrc = src[0];
const uint8_t *usrc = src[1];
const uint8_t *vsrc = src[2];
const int width = c->srcW;
const int height = srcSliceH;
const int lumStride = srcStride[0];
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]);
const uint8_t *ysrc = src[0];
const uint8_t *usrc = src[1];
const 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 dstStride = dstStride_a[0];
const vector unsigned char yperm = vec_lvsl(0, ysrc);
const int vertLumPerChroma = 2;
const int vertLumPerChroma = 2;
register unsigned int y;
/* This code assumes:
*
* 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 multiples of 8
*/
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 multiples of 8
*/
for (y=0; y<height; y++) {
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);
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);
@ -81,71 +83,72 @@ static int yv12toyuy2_unscaled_altivec(SwsContext *c, const uint8_t* src[],
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);
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) {
if ((y & (vertLumPerChroma - 1)) == vertLumPerChroma - 1) {
usrc += chromStride;
vsrc += chromStride;
}
ysrc += lumStride;
dst += dstStride;
dst += dstStride;
}
return srcSliceH;
}
static int yv12touyvy_unscaled_altivec(SwsContext *c, const uint8_t* src[],
static int yv12touyvy_unscaled_altivec(SwsContext *c, const uint8_t *src[],
int srcStride[], int srcSliceY,
int srcSliceH, uint8_t* dstParam[],
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]);
const uint8_t *ysrc = src[0];
const uint8_t *usrc = src[1];
const 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;
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]);
const uint8_t *ysrc = src[0];
const uint8_t *usrc = src[1];
const 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;
/* This code assumes:
*
* 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 multiples of 8
*/
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 multiples of 8
*/
for (y=0; y<height; y++) {
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);
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);
@ -156,22 +159,22 @@ static int yv12touyvy_unscaled_altivec(SwsContext *c, const uint8_t* src[],
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);
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) {
if ((y & (vertLumPerChroma - 1)) == vertLumPerChroma - 1) {
usrc += chromStride;
vsrc += chromStride;
}
ysrc += lumStride;
dst += dstStride;
dst += dstStride;
}
return srcSliceH;
}
@ -184,8 +187,8 @@ void ff_swscale_get_unscaled_altivec(SwsContext *c)
// unscaled YV12 -> packed YUV, we want speed
if (dstFormat == PIX_FMT_YUYV422)
c->swScale= yv12toyuy2_unscaled_altivec;
c->swScale = yv12toyuy2_unscaled_altivec;
else if (dstFormat == PIX_FMT_UYVY422)
c->swScale= yv12touyvy_unscaled_altivec;
c->swScale = yv12touyvy_unscaled_altivec;
}
}