Skip to content

F
ffmpeg.wasm-core
Commit 484267f3 authored by Diego Biurrun's avatar Diego Biurrun
Browse files
Options

Do not mix declarations and statements. 

Originally committed as revision 20600 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale
parent 65342b2b
Hide whitespace changes
Inline Side-by-side
Showing with 37 additions and 30 deletions
libswscale/swscale_altivec_template.c
View file @ 484267f3
......@@ -37,13 +37,15 @@ altivec_packIntArrayToCharArray(int *val, uint8_t* dest, int dstW) {
if ((unsigned long)dest % 16) {
/* badly aligned store, we force store alignement */
/* and will handle load misalignement on val w/ vec_perm */
vector unsigned char perm1;
vector signed int v1;
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);
perm1 = vec_lvsl(i << 2, val);
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);
......@@ -106,13 +108,13 @@ yuv2yuvX_altivec_real(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
}
for (j = 0; j < lumFilterSize; j++) {
vector signed short vLumFilter = vec_ld(j << 1, lumFilter);
vector unsigned char perm0 = vec_lvsl(j << 1, lumFilter);
vector signed short l1, vLumFilter = vec_ld(j << 1, lumFilter);
vector unsigned char perm, 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]);
perm = vec_lvsl(0, lumSrc[j]);
l1 = vec_ld(0, lumSrc[j]);
for (i = 0; i < (dstW - 7); i+=8) {
int offset = i << 2;
......@@ -157,14 +159,14 @@ yuv2yuvX_altivec_real(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
}
for (j = 0; j < chrFilterSize; j++) {
vector signed short vChrFilter = vec_ld(j << 1, chrFilter);
vector unsigned char perm0 = vec_lvsl(j << 1, chrFilter);
vector signed short l1, l1_V, vChrFilter = vec_ld(j << 1, chrFilter);
vector unsigned char perm, 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]);
perm = vec_lvsl(0, chrSrc[j]);
l1 = vec_ld(0, chrSrc[j]);
l1_V = vec_ld(2048 << 1, chrSrc[j]);
for (i = 0; i < (chrDstW - 7); i+=8) {
int offset = i << 2;
......@@ -235,18 +237,20 @@ static inline void hScale_altivec_real(int16_t *dst, int dstW, uint8_t *src, int
register int srcPos = filterPos[i];
vector unsigned char src_v0 = vec_ld(srcPos, src);
vector unsigned char src_v1;
vector unsigned char src_v1, src_vF;
vector signed short src_v, filter_v;
vector signed int val_vEven, val_s;
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));
src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
vector signed short src_v = // vec_unpackh sign-extends...
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);
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,
......@@ -258,8 +262,8 @@ static inline void hScale_altivec_real(int16_t *dst, int dstW, uint8_t *src, int
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);
val_vEven = vec_mule(src_v, filter_v);
val_s = vec_sums(val_vEven, vzero);
vec_st(val_s, 0, tempo);
dst[i] = FFMIN(FFMAX(0, tempo[3]>>7), (1<<15)-1);
}
......@@ -272,19 +276,21 @@ static inline void hScale_altivec_real(int16_t *dst, int dstW, uint8_t *src, int
register int srcPos = filterPos[i];
vector unsigned char src_v0 = vec_ld(srcPos, src);
vector unsigned char src_v1;
vector unsigned char src_v1, src_vF;
vector signed short src_v, filter_v;
vector signed int val_v, val_s;
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));
src_vF = vec_perm(src_v0, src_v1, vec_lvsl(srcPos, src));
vector signed short src_v = // vec_unpackh sign-extends...
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);
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);
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(FFMAX(0, tempo[3]>>7), (1<<15)-1);
}
......@@ -326,7 +332,7 @@ static inline void hScale_altivec_real(int16_t *dst, int dstW, uint8_t *src, int
register int j;
register int srcPos = filterPos[i];
vector signed int val_v = (vector signed int)vzero;
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);
......@@ -357,23 +363,24 @@ static inline void hScale_altivec_real(int16_t *dst, int dstW, uint8_t *src, int
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;
vector unsigned char src_v1, src_vF;
vector signed short src_v, filter_v1R, filter_v;
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);
src_vF = vec_perm(src_v0, src_v1, permS);
vector signed short src_v = // vec_unpackh sign-extends...
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);
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);
}
vector signed int val_s = vec_sums(val_v, vzero);
val_s = vec_sums(val_v, vzero);
vec_st(val_s, 0, tempo);
dst[i] = FFMIN(FFMAX(0, tempo[3]>>7), (1<<15)-1);
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment