static int pix_norm1_altivec(uint8_t *pix, int line_size)
{
int i;
int s;
__vector zero = __vzero();
/*
vector unsigned char *tv;
vector unsigned char pixv;
vector unsigned int sv;
vector signed int sum;
*/
__vector *tv;
__vector pixv;
__vector sv;
__vector sum;
sv = __vzero();
s = 0;
for (i = 0; i < 16; i++) {
/* Read in the potentially unaligned pixels */
//tv = (vector unsigned char *) pix;
tv = (__vector*) pix;
//pixv = vec_perm(tv[0], tv[1], vec_lvsl(0, pix));
pixv = __vperm(tv[0], tv[1], __lvsl(pix,0));
/* Square the values, and add them to our sum */
sv = vec_msum(pixv, pixv, sv);
pix += line_size;
}
/* Sum up the four partial sums, and put the result into s */
sum = vec_sums((vector signed int) sv, (vector signed int) zero);
sum = vec_splat(sum, 3);
vec_ste(sum, 0, &s);
return s;
}
static force_inline __vector4
pix_multiply (__vector4 p, __vector4 a)
{
__vector4 hi, lo, mod;
__vector4 hiLow, hiHigh, modLow, modHigh, loLow, loHigh;
__vector4 hiLowFP, hiHighFP, modLowFP, modHighFP, loLowFP, loHighFP;
__vector4 himodLow, himodHigh, lomodLow, lomodHigh;
__vector4 zeroVector = *(__vector4*)(&zeroVectori);
__vector4 zeroEightVector = *(__vector4*)(&zeroEightVectori);
__vector4 permVecLo = *(__vector4*)(&permVecLoi);
__vector4 permVecHi = *(__vector4*)(&permVecHii);
__vector4 permVec1Vec2 = *(__vector4*)(&permVec1Vec2i);
/* unpack to short */
hi = __vmrghb(zeroVector,p);
mod = __vmrghb(zeroVector,a);
//+ What we want to do here is to multiply 8 unsigned shorts of the hi with 8 unsigned shorts of mod.
/* Extract the hi vector into 4 Unsigned int by using 4 Lower unsigned shorts*/
hiLow = __vperm(hi,zeroVector, permVecLo);
/* Extract the hi vector into 4 Unsigned int by using 4 Upper unsigned shorts*/
hiHigh = __vperm(hi,zeroVector, permVecHi);
/* Extract the mod vector into 4 Unsigned int by using 4 Lower unsigned shorts*/
modLow = __vperm(mod,zeroVector, permVecLo);
/* Extract the mod vector into 4 Unsigned int by using 4 Upper unsigned shorts*/
modHigh = __vperm(mod,zeroVector, permVecHi);
/* Convert the 4 unsigned ints to floating point by treating them as Fixed point*/
hiLowFP = __vcfux(hiLow,0);
hiHighFP = __vcfux(hiHigh,0);
modLowFP = __vcfux(modLow,0);
modHighFP = __vcfux(modHigh,0);
/* Multiply the floating points */
himodLow = __vmaddfp(hiLowFP, modLowFP,zeroVector);
himodHigh = __vmaddfp(hiHighFP, modHighFP,zeroVector);
/* Convert the floating points to Fixed Point with zero digits after radix point - Effectively an unsigned int*/
himodLow = __vctuxs(himodLow,0);
himodHigh = __vctuxs(himodHigh,0);
/* Fuse the multiplication together to get the final product*/
hi = __vperm(himodLow,himodHigh,permVec1Vec2);
//-
hi = __vadduhm(hi,zeroEightVector);
hi = __vadduhs(hi, __vsrh(hi, __vspltish (8)));
hi = __vsrh (hi, __vspltish (8));
/* unpack to short */
lo = __vmrglb(zeroVector,p);
mod = __vmrglb(zeroVector,a);
//+ Comments from few lines above applicable here.
loLow = __vperm(lo,zeroVector,permVecLo);
loHigh = __vperm(lo,zeroVector, permVecHi);
modLow = __vperm(mod,zeroVector, permVecLo);
modHigh = __vperm(mod,zeroVector, permVecHi);
loLowFP = __vcfux(loLow,0);
loHighFP = __vcfux(loHigh,0);
modLowFP = __vcfux(modLow,0);
modHighFP = __vcfux(modHigh,0);
lomodLow = __vmaddfp(loLowFP, modLowFP,zeroVector);
lomodHigh = __vmaddfp(loHighFP, modHighFP,zeroVector);
lomodLow = __vctuxs(lomodLow,0);
lomodHigh = __vctuxs(lomodHigh,0);
lo = __vperm(lomodLow,lomodHigh,permVec1Vec2);
//-
lo = __vadduhm(lo,zeroEightVector);
lo = __vadduhs (lo, __vsrh (lo, __vspltish (8)));
lo = __vsrh (lo, __vspltish (8));
return __vpkuhus (hi, lo);
}
static force_inline __vector4
splat_alpha (__vector4 pix)
{
return __vperm (pix, pix, *(__vector4*)(&vmx128i_splat_alpha_vector));
}