/* Store a vector to an unaligned location in memory */
static inline void
StoreUnaligned (vector unsigned char v,
const guchar *where)
{
if ((unsigned long)where & 0x0f)
{
/* Load the surrounding area */
vector unsigned char low = vec_ld(0, where);
vector unsigned char high = vec_ld(16, where);
/* Prepare the constants that we need */
vector unsigned char permuteVector = vec_lvsr(0, where);
vector signed char oxFF = vec_splat_s8(-1);
vector signed char ox00 = vec_splat_s8(0);
/* Make a mask for which parts of the vectors to swap out */
vector unsigned char mask = (vector unsigned char)vec_perm(ox00, oxFF, permuteVector);
v = vec_perm(v, v, permuteVector);
/* Insert our data into the low and high vectors */
low = vec_sel(low, v, mask);
high = vec_sel(v, high, mask);
/* Store the two aligned result vectors */
vec_st(low, 0, CONST_BUFFER(where));
vec_st(high, 16, CONST_BUFFER(where));
}
else
{ /* prevent overflow */
vec_st(v, 0, CONST_BUFFER(where));
}
}
void foo (vector bool long long *vblr,
vector double *vdr)
{
*vblr++ = vec_andc (vbla, vblb);
*vdr++ = vec_double (vsla);
*vdr++ = vec_double (vula);
*vblr++ = vec_mergeh (vbla, vblb);
*vblr++ = vec_mergel (vbla, vblb);
*vblr++ = vec_nor (vbla, vblb);
*vblr++ = vec_or (vbla, vblb);
*vblr++ = vec_sel (vbla, vblb, vblc);
*vblr++ = vec_sel (vbla, vblb, vulc);
*vblr++ = vec_xor (vbla, vblb);
}
void test_select(void)
{
vector float firstvector = {1.0, 2.0, 3.0, 4.0};
vector float secondvector = {100.0, 200.0, 300.0, 400.0};
vector unsigned int selector = {0, 0, 0, 0xffffffff};
vector float outputvector;
float printfloat[FLOAT_ARRAYSIZE];
int i;
fprintf(stderr, "--- function %s ------\n", __FUNCTION__);
/* i'm expecting the output here to be {1.0, 2.0, 3.0, 400.0} */
/* select selector has zeros for all elements of the first three */
/* vectors, they should get the bits from the firstvector. since */
/* selector is all ones for the fourth element, the fourth element */
/* of outputvector should get the fourth element of secondvector */
outputvector = vec_sel(firstvector, secondvector, selector);
vec_st(outputvector, 0, printfloat);
fprintf(stderr, "selected vector: ");
for(i = 0; i < FLOAT_ARRAYSIZE; i++)
{
fprintf(stderr, "%f ", printfloat[i]);
}
fprintf(stderr, "\n");
} /* test_select */
__SIMDd _SIMD_sel_pd(__SIMDd a, __SIMDd b, void** resultPtr)
{
#ifdef USE_SSE
__SIMDd* result = (__SIMDd*) (*resultPtr);
return _mm_or_pd(_mm_andnot_pd(*result,a),_mm_and_pd(*result,b));
#elif defined USE_AVX
__SIMDd* result = (__SIMDd*) resultPtr;
return _mm256_or_pd(_mm256_andnot_pd(*result,a),_mm256_and_pd(*result,b));
#elif defined USE_IBM
return vec_sel(a,b,c);
#endif
}
void foo( float scalar)
{
unsigned long width;
unsigned long x;
vector float vColor;
vector unsigned int selectMask;
vColor = vec_perm( vec_ld( 0, &scalar), vec_ld( 3, &scalar), vec_lvsl( 0, &scalar) );
float *destRow;
vector float store, load0;
for( ; x < width; x++)
{
load0 = vec_sel( vColor, load0, selectMask );
vec_st( store, 0, destRow );
store = load0;
}
}
int main(int argc, char **argv)
{
int i;
__vector float *vin = (__vector float *) in;
__vector float *vout = (__vector float *) out;
__vector float vin_negative;
__vector unsigned int vpat;
__vector float vzero = (__vector float) { 0.0f, 0.0f, 0.0f, 0.0f };
__vector float vminus = (__vector float) { -1.0f, -1.0f, -1.0f, -1.0f };
for (i = 0; i < SIZE/4; i++) {
vpat = vec_cmpgt(vin[i], vzero);
vin_negative = vec_madd(vin[i], vminus, vzero);
vout[i] = vec_sel(vin_negative, vin[i], vpat);
}
for (i = 0; i < SIZE; i++) {
printf("out[%02d]=%0.0f\n", i, out[i]);
}
return 0;
}
static vector unsigned char permute_128(vector unsigned char input)
{
vector unsigned char result, new_bit;
/* and now the code */
result = vec_vperm(input, input, control1);
result = vec_rl(result, rotate1);
new_bit = vec_vperm(input, input, control2);
new_bit = vec_rl(new_bit, rotate2);
result = vec_sel(result, new_bit, select2);
new_bit = vec_vperm(input, input, control3);
new_bit = vec_rl(new_bit, rotate3);
result = vec_sel(result, new_bit, select3);
new_bit = vec_vperm(input, input, control4);
new_bit = vec_rl(new_bit, rotate4);
result = vec_sel(result, new_bit, select4);
new_bit = vec_vperm(input, input, control5);
new_bit = vec_rl(new_bit, rotate5);
result = vec_sel(result, new_bit, select5);
new_bit = vec_vperm(input, input, control6);
new_bit = vec_rl(new_bit, rotate6);
result = vec_sel(result, new_bit, select6);
new_bit = vec_vperm(input, input, control7);
new_bit = vec_rl(new_bit, rotate7);
result = vec_sel(result, new_bit, select7);
new_bit = vec_vperm(input, input, control8);
new_bit = vec_rl(new_bit, rotate8);
result = vec_sel(result, new_bit, select8);
return result;
}
/* this code assume stride % 16 == 0 */
static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_v_lowpass_num, 1);
POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_v_lowpass_num, 1);
register int i;
const vector signed int vzero = vec_splat_s32(0);
const vector unsigned char perm = vec_lvsl(0, src);
const vector signed short v20ss = (const vector signed short)AVV(20);
const vector unsigned short v5us = vec_splat_u16(5);
const vector signed short v5ss = vec_splat_s16(5);
const vector signed short v16ss = (const vector signed short)AVV(16);
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);
uint8_t *srcbis = src - (srcStride * 2);
const vector unsigned char srcM2a = vec_ld(0, srcbis);
const vector unsigned char srcM2b = vec_ld(16, srcbis);
const vector unsigned char srcM2 = vec_perm(srcM2a, srcM2b, perm);
srcbis += srcStride;
const vector unsigned char srcM1a = vec_ld(0, srcbis);
const vector unsigned char srcM1b = vec_ld(16, srcbis);
const vector unsigned char srcM1 = vec_perm(srcM1a, srcM1b, perm);
srcbis += srcStride;
const vector unsigned char srcP0a = vec_ld(0, srcbis);
const vector unsigned char srcP0b = vec_ld(16, srcbis);
const vector unsigned char srcP0 = vec_perm(srcP0a, srcP0b, perm);
srcbis += srcStride;
const vector unsigned char srcP1a = vec_ld(0, srcbis);
const vector unsigned char srcP1b = vec_ld(16, srcbis);
const vector unsigned char srcP1 = vec_perm(srcP1a, srcP1b, perm);
srcbis += srcStride;
const vector unsigned char srcP2a = vec_ld(0, srcbis);
const vector unsigned char srcP2b = vec_ld(16, srcbis);
const vector unsigned char srcP2 = vec_perm(srcP2a, srcP2b, perm);
srcbis += srcStride;
vector signed short srcM2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM2);
vector signed short srcM2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM2);
vector signed short srcM1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcM1);
vector signed short srcM1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcM1);
vector signed short srcP0ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP0);
vector signed short srcP0ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP0);
vector signed short srcP1ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP1);
vector signed short srcP1ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP1);
vector signed short srcP2ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP2);
vector signed short srcP2ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP2);
for (i = 0 ; i < 16 ; i++) {
const vector unsigned char srcP3a = vec_ld(0, srcbis);
const vector unsigned char srcP3b = vec_ld(16, srcbis);
const vector unsigned char srcP3 = vec_perm(srcP3a, srcP3b, perm);
const vector signed short srcP3ssA = (vector signed short)vec_mergeh((vector unsigned char)vzero, srcP3);
const vector signed short srcP3ssB = (vector signed short)vec_mergel((vector unsigned char)vzero, srcP3);
srcbis += srcStride;
const vector signed short sum1A = vec_adds(srcP0ssA, srcP1ssA);
const vector signed short sum1B = vec_adds(srcP0ssB, srcP1ssB);
const vector signed short sum2A = vec_adds(srcM1ssA, srcP2ssA);
const vector signed short sum2B = vec_adds(srcM1ssB, srcP2ssB);
const vector signed short sum3A = vec_adds(srcM2ssA, srcP3ssA);
const vector signed short sum3B = vec_adds(srcM2ssB, srcP3ssB);
srcM2ssA = srcM1ssA;
srcM2ssB = srcM1ssB;
srcM1ssA = srcP0ssA;
srcM1ssB = srcP0ssB;
srcP0ssA = srcP1ssA;
srcP0ssB = srcP1ssB;
srcP1ssA = srcP2ssA;
srcP1ssB = srcP2ssB;
srcP2ssA = srcP3ssA;
srcP2ssB = srcP3ssB;
const vector signed short pp1A = vec_mladd(sum1A, v20ss, v16ss);
const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
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 pp3A = vec_add(sum3A, pp1A);
const vector signed short pp3B = vec_add(sum3B, pp1B);
const vector signed short psumA = vec_sub(pp3A, pp2A);
const vector signed short psumB = vec_sub(pp3B, pp2B);
const vector signed short sumA = vec_sra(psumA, v5us);
const vector signed short sumB = vec_sra(psumB, v5us);
const vector unsigned char sum = vec_packsu(sumA, sumB);
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_v_lowpass_num, 1);
}
/* this code assume stride % 16 == 0 */
static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_h_lowpass_num, 1);
POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_h_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 short v5us = vec_splat_u16(5);
const vector signed short v5ss = vec_splat_s16(5);
const vector signed short v16ss = (const vector signed short)AVV(16);
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);
register int align = ((((unsigned long)src) - 2) % 16);
for (i = 0 ; i < 16 ; 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, v16ss);
const vector signed short pp1B = vec_mladd(sum1B, v20ss, v16ss);
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 pp3A = vec_add(sum3A, pp1A);
const vector signed short pp3B = vec_add(sum3B, pp1B);
const vector signed short psumA = vec_sub(pp3A, pp2A);
const vector signed short psumB = vec_sub(pp3B, pp2B);
const vector signed short sumA = vec_sra(psumA, v5us);
const vector signed short sumB = vec_sra(psumB, v5us);
const vector unsigned char sum = vec_packsu(sumA, sumB);
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);
src += srcStride;
dst += dstStride;
}
POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
}
/* GEMV with the following assumptions:
* 1) alpha = 1
* 2) beta = 0 or 1 (controlled by BETA0/BETA1)
* 3) incX = 1 and incY = 1
* 4) Column-major storage of A
* 4) A is stored column major but is transposed.
*
* y = [0,1]*y + A*x, len(X) = M, len(Y) = N
* At is MxN and is the transpose of A (NxM)
*/
void ATL_UGEMV(
const int M, const int N,
const TYPE *At, const int lda,
const TYPE *X, TYPE *Y)
{
int i, j;
const long int mu = VEC_STEP*3; /* Unrolling used for the X vector */
const long int nu = 8; /* Unrolling used for the Y vector */
vector TYPE vzero = vec_splats( ((TYPE) 0.0) );
long int M1 = (M/mu)*mu;
long int N1 = (N/nu)*nu;
TYPE *py = &Y[0];
for (i=0; i < N1; i+=nu) {
vector TYPE vy0, vy1, vy2, vy3;
vector TYPE vy4, vy5, vy6, vy7;
{
#if defined BETA0
vy0 = vzero;
vy1 = vzero;
vy2 = vzero;
vy3 = vzero;
vy4 = vzero;
vy5 = vzero;
vy6 = vzero;
vy7 = vzero;
#else /* BETA1 */
vector TYPE vt0, vt1;
vt0 = *((vector TYPE *)(py+0*VEC_STEP));
vt1 = *((vector TYPE *)(py+1*VEC_STEP));
#ifdef DREAL
vector TYPE vt2, vt3;
vt2 = *((vector TYPE *)(py+2*VEC_STEP));
vt3 = *((vector TYPE *)(py+3*VEC_STEP));
#endif
#ifdef DREAL
vy0 = vec_mergeh(vt0, vzero);
vy1 = vec_mergel(vt0, vzero);
vy2 = vec_mergeh(vt1, vzero);
vy3 = vec_mergel(vt1, vzero);
vy4 = vec_mergeh(vt2, vzero);
vy5 = vec_mergel(vt2, vzero);
vy6 = vec_mergeh(vt3, vzero);
vy7 = vec_mergel(vt3, vzero);
#else /* SREAL */
vy0 = vec_sel(vt0, vzero, ELEM0);
vy1 = vec_sel(vt0, vzero, ELEM1);
vy2 = vec_sel(vt0, vzero, ELEM2);
vy3 = vec_sel(vt0, vzero, ELEM3);
vy4 = vec_sel(vt1, vzero, ELEM0);
vy5 = vec_sel(vt1, vzero, ELEM1);
vy6 = vec_sel(vt1, vzero, ELEM2);
vy7 = vec_sel(vt1, vzero, ELEM3);
#endif
#endif
}
const TYPE *pa0 = &At[(i+0)*lda];
const TYPE *pa1 = &At[(i+1)*lda];
const TYPE *pa2 = &At[(i+2)*lda];
const TYPE *pa3 = &At[(i+3)*lda];
const TYPE *pa4 = &At[(i+4)*lda];
const TYPE *pa5 = &At[(i+5)*lda];
const TYPE *pa6 = &At[(i+6)*lda];
const TYPE *pa7 = &At[(i+7)*lda];
const TYPE *px = &X[0];
for (j=0; j < M1; j+=mu) {
vector TYPE va00, va01, va02, va03;
vector TYPE va04, va05, va06, va07;
vector TYPE va10, va11, va12, va13;
vector TYPE va14, va15, va16, va17;
vector TYPE va20, va21, va22, va23;
vector TYPE va24, va25, va26, va27;
vector TYPE vx0, vx1, vx2;
vx0 = *((vector TYPE *)( px+0*VEC_STEP ));
vx1 = *((vector TYPE *)( px+1*VEC_STEP ));
vx2 = *((vector TYPE *)( px+2*VEC_STEP ));
px += mu;
va00 = *((vector TYPE *)( pa0+0*VEC_STEP ));
va01 = *((vector TYPE *)( pa1+0*VEC_STEP ));
va02 = *((vector TYPE *)( pa2+0*VEC_STEP ));
va03 = *((vector TYPE *)( pa3+0*VEC_STEP ));
va04 = *((vector TYPE *)( pa4+0*VEC_STEP ));
va05 = *((vector TYPE *)( pa5+0*VEC_STEP ));
va06 = *((vector TYPE *)( pa6+0*VEC_STEP ));
va07 = *((vector TYPE *)( pa7+0*VEC_STEP ));
vy0 = vec_madd(va00, vx0, vy0);
vy1 = vec_madd(va01, vx0, vy1);
vy2 = vec_madd(va02, vx0, vy2);
vy3 = vec_madd(va03, vx0, vy3);
vy4 = vec_madd(va04, vx0, vy4);
vy5 = vec_madd(va05, vx0, vy5);
vy6 = vec_madd(va06, vx0, vy6);
vy7 = vec_madd(va07, vx0, vy7);
va10 = *((vector TYPE *)( pa0+1*VEC_STEP ));
va11 = *((vector TYPE *)( pa1+1*VEC_STEP ));
va12 = *((vector TYPE *)( pa2+1*VEC_STEP ));
va13 = *((vector TYPE *)( pa3+1*VEC_STEP ));
va14 = *((vector TYPE *)( pa4+1*VEC_STEP ));
va15 = *((vector TYPE *)( pa5+1*VEC_STEP ));
va16 = *((vector TYPE *)( pa6+1*VEC_STEP ));
va17 = *((vector TYPE *)( pa7+1*VEC_STEP ));
vy0 = vec_madd(va10, vx1, vy0);
vy1 = vec_madd(va11, vx1, vy1);
vy2 = vec_madd(va12, vx1, vy2);
vy3 = vec_madd(va13, vx1, vy3);
vy4 = vec_madd(va14, vx1, vy4);
vy5 = vec_madd(va15, vx1, vy5);
vy6 = vec_madd(va16, vx1, vy6);
vy7 = vec_madd(va17, vx1, vy7);
va20 = *((vector TYPE *)( pa0+2*VEC_STEP ));
va21 = *((vector TYPE *)( pa1+2*VEC_STEP ));
va22 = *((vector TYPE *)( pa2+2*VEC_STEP ));
va23 = *((vector TYPE *)( pa3+2*VEC_STEP ));
va24 = *((vector TYPE *)( pa4+2*VEC_STEP ));
va25 = *((vector TYPE *)( pa5+2*VEC_STEP ));
va26 = *((vector TYPE *)( pa6+2*VEC_STEP ));
va27 = *((vector TYPE *)( pa7+2*VEC_STEP ));
vy0 = vec_madd(va20, vx2, vy0);
vy1 = vec_madd(va21, vx2, vy1);
vy2 = vec_madd(va22, vx2, vy2);
vy3 = vec_madd(va23, vx2, vy3);
vy4 = vec_madd(va24, vx2, vy4);
vy5 = vec_madd(va25, vx2, vy5);
vy6 = vec_madd(va26, vx2, vy6);
vy7 = vec_madd(va27, vx2, vy7);
pa0 += mu;
pa1 += mu;
pa2 += mu;
pa3 += mu;
pa4 += mu;
pa5 += mu;
pa6 += mu;
pa7 += mu;
}
{
#ifdef DREAL
vector TYPE vt0, vt1, vt2, vt3;
vector TYPE vt4, vt5, vt6, vt7;
vt0 = vec_mergeh(vy0, vy1);
vt1 = vec_mergel(vy0, vy1);
vt2 = vec_mergeh(vy2, vy3);
vt3 = vec_mergel(vy2, vy3);
vt4 = vec_mergeh(vy4, vy5);
vt5 = vec_mergel(vy4, vy5);
vt6 = vec_mergeh(vy6, vy7);
vt7 = vec_mergel(vy6, vy7);
vy0 = vec_add(vt0, vt1);
vy2 = vec_add(vt2, vt3);
vy4 = vec_add(vt4, vt5);
vy6 = vec_add(vt6, vt7);
*((vector TYPE *)(py+0*VEC_STEP)) = vy0;
*((vector TYPE *)(py+1*VEC_STEP)) = vy2;
*((vector TYPE *)(py+2*VEC_STEP)) = vy4;
*((vector TYPE *)(py+3*VEC_STEP)) = vy6;
#else /* SREAL */
vy0 = vec_reduce(vy0);
vy1 = vec_reduce(vy1);
vy2 = vec_reduce(vy2);
vy3 = vec_reduce(vy3);
vy4 = vec_reduce(vy4);
vy5 = vec_reduce(vy5);
vy6 = vec_reduce(vy6);
vy7 = vec_reduce(vy7);
vec_ste(vy0,0,py+0);
vec_ste(vy1,0,py+1);
vec_ste(vy2,0,py+2);
vec_ste(vy3,0,py+3);
vec_ste(vy4,0,py+4);
vec_ste(vy5,0,py+5);
vec_ste(vy6,0,py+6);
vec_ste(vy7,0,py+7);
#endif
}
{
register TYPE y0, y1, y2, y3;
register TYPE y4, y5, y6, y7;
y0 = *(py+0);
y1 = *(py+1);
y2 = *(py+2);
y3 = *(py+3);
y4 = *(py+4);
y5 = *(py+5);
y6 = *(py+6);
y7 = *(py+7);
for (j=M1; j < M; j++) {
y0 += *pa0 * X[j];
y1 += *pa1 * X[j];
y2 += *pa2 * X[j];
y3 += *pa3 * X[j];
y4 += *pa4 * X[j];
y5 += *pa5 * X[j];
y6 += *pa6 * X[j];
y7 += *pa7 * X[j];
pa0++;
pa1++;
pa2++;
pa3++;
pa4++;
pa5++;
pa6++;
pa7++;
}
*(py+0) = y0;
*(py+1) = y1;
*(py+2) = y2;
*(py+3) = y3;
*(py+4) = y4;
*(py+5) = y5;
*(py+6) = y6;
*(py+7) = y7;
}
py += nu;
}
for (i=N1; i < N; i++)
{
register TYPE y0;
#if defined BETA0
y0 = 0.0;
#else /* BETA1 */
y0 = Y[i];
#endif
for (j=0; j < M; j++)
y0 += At[j+i*lda] * X[j];
Y[i] = y0;
}
}
static inline vector unsigned int Maj(const vector unsigned int b, const vector unsigned int c, const vector unsigned int d) {
return vec_sel(b,c, vec_xor(b,d));
}
/* a vectorized version of the Voigt function using Altivec / VMX instructions */
void my_voigt(const float *damping, const float *frequency_offset, float *voigt_value, int N)
{
// coefficients of the rational approximation formula
// to the complementary error function
const vector float A0 = (vector float) (122.607931777104326f);
const vector float A1 = (vector float) (214.382388694706425f);
const vector float A2 = (vector float) (181.928533092181549f);
const vector float A3 = (vector float) (93.155580458138441f);
const vector float A4 = (vector float) (30.180142196210589f);
const vector float A5 = (vector float) (5.912626209773153f);
const vector float A6 = (vector float) (0.564189583562615f);
const vector float B0 = (vector float) (122.60793177387535f);
const vector float B1 = (vector float) (352.730625110963558f);
const vector float B2 = (vector float) (457.334478783897737f);
const vector float B3 = (vector float) (348.703917719495792f);
const vector float B4 = (vector float) (170.354001821091472f);
const vector float B5 = (vector float) (53.992906912940207f);
const vector float B6 = (vector float) (10.479857114260399f);
vector float ivsigno;
vector float V;
vector float Z1_real;
vector float Z1_imag;
vector float Z2_real;
vector float Z2_imag;
vector float Z3_real;
vector float Z3_imag;
vector float Z4_real;
vector float Z4_imag;
vector float Z5_real;
vector float Z5_imag;
vector float Z6_real;
vector float Z6_imag;
vector float ZZ1_real;
vector float ZZ1_imag;
vector float ZZ2_real;
vector float ZZ2_imag;
vector float ZZ3_real;
vector float ZZ3_imag;
vector float ZZ4_real;
vector float ZZ4_imag;
vector float ZZ5_real;
vector float ZZ5_imag;
vector float ZZ6_real;
vector float ZZ6_imag;
vector float ZZ7_real;
vector float ZZ7_imag;
vector float division_factor;
vector float ZZZ_real;
vector bool int mask;
const vector float one = (vector float) (1.0f);
const vector float zero = (vector float) (-0.0f);
const vector float mone = (vector float) (-1.0f);
vector float damp;
vector float offs;
for(int i=0; i<N; i+=4){
damp = vec_ld(0,(float *) &damping[i]);
offs = vec_ld(0,(float *) &frequency_offset[i]);
mask = vec_cmplt(offs,zero);
ivsigno = vec_sel(mone, one, mask);
//ivsigno = (vector float) (1.0f);
V = vec_madd(ivsigno, offs, zero);
Z1_real = vec_madd(A6, damp, A5);
Z1_imag = vec_nmsub(A6, V, zero);
Z2_real = vec_add(vec_madd(Z1_real,damp,zero),vec_madd(Z1_imag,V,A4));
Z2_imag = vec_add(vec_nmsub(Z1_real,V,zero),vec_madd(Z1_imag,damp,zero));
Z3_real = vec_add(vec_madd(Z2_real,damp,zero),vec_madd(Z2_imag,V,A3));
Z3_imag = vec_add(vec_nmsub(Z2_real,V,zero),vec_madd(Z2_imag,damp,zero));
Z4_real = vec_add(vec_madd(Z3_real,damp,zero),vec_madd(Z3_imag,V,A2));
Z4_imag = vec_add(vec_nmsub(Z3_real,V,zero),vec_madd(Z3_imag,damp,zero));
Z5_real = vec_add(vec_madd(Z4_real,damp,zero),vec_madd(Z4_imag,V,A1));
Z5_imag = vec_add(vec_nmsub(Z4_real,V,zero),vec_madd(Z4_imag,damp,zero));
Z6_real = vec_add(vec_madd(Z5_real,damp,zero),vec_madd(Z5_imag,V,A0));
Z6_imag = vec_add(vec_nmsub(Z5_real,V,zero),vec_madd(Z5_imag,damp,zero));
ZZ1_real = vec_add(damp,B6);
ZZ1_imag = vec_madd(mone,V,zero);
ZZ2_real = vec_add(vec_madd(ZZ1_real,damp,zero),vec_madd(ZZ1_imag,V,B5));
ZZ2_imag = vec_add(vec_nmsub(ZZ1_real,V,zero),vec_madd(ZZ1_imag,damp,zero));
ZZ3_real = vec_add(vec_madd(ZZ2_real,damp,zero),vec_madd(ZZ2_imag,V,B4));
ZZ3_imag = vec_add(vec_nmsub(ZZ2_real,V,zero),vec_madd(ZZ2_imag,damp,zero));
ZZ4_real = vec_add(vec_madd(ZZ3_real,damp,zero),vec_madd(ZZ3_imag,V,B3));
ZZ4_imag = vec_add(vec_nmsub(ZZ3_real,V,zero),vec_madd(ZZ3_imag,damp,zero));
ZZ5_real = vec_add(vec_madd(ZZ4_real,damp,zero),vec_madd(ZZ4_imag,V,B2));
ZZ5_imag = vec_add(vec_nmsub(ZZ4_real,V,zero),vec_madd(ZZ4_imag,damp,zero));
ZZ6_real = vec_add(vec_madd(ZZ5_real,damp,zero),vec_madd(ZZ5_imag,V,B1));
ZZ6_imag = vec_add(vec_nmsub(ZZ5_real,V,zero),vec_madd(ZZ5_imag,damp,zero));
ZZ7_real = vec_add(vec_madd(ZZ6_real,damp,zero),vec_madd(ZZ6_imag,V,B0));
ZZ7_imag = vec_add(vec_nmsub(ZZ6_real,V,zero),vec_madd(ZZ6_imag,damp,zero));
division_factor = vec_div(one,vec_madd(ZZ7_real,ZZ7_real,vec_madd(ZZ7_imag,ZZ7_imag,zero)));
ZZZ_real = vec_madd(vec_madd(Z6_real,ZZ7_real,vec_madd(Z6_imag,ZZ7_imag,zero)),division_factor,zero);
vec_st(ZZZ_real,0,(float *)&voigt_value[i]);
}
}
static int dct_quantize_altivec(MpegEncContext* s,
DCTELEM* data, int n,
int qscale, int* overflow)
{
int lastNonZero;
vector float row0, row1, row2, row3, row4, row5, row6, row7;
vector float alt0, alt1, alt2, alt3, alt4, alt5, alt6, alt7;
const vector float zero = (const vector float)FOUROF(0.);
// used after quantize step
int oldBaseValue = 0;
// Load the data into the row/alt vectors
{
vector signed short data0, data1, data2, data3, data4, data5, data6, data7;
data0 = vec_ld(0, data);
data1 = vec_ld(16, data);
data2 = vec_ld(32, data);
data3 = vec_ld(48, data);
data4 = vec_ld(64, data);
data5 = vec_ld(80, data);
data6 = vec_ld(96, data);
data7 = vec_ld(112, data);
// Transpose the data before we start
TRANSPOSE8(data0, data1, data2, data3, data4, data5, data6, data7);
// load the data into floating point vectors. We load
// the high half of each row into the main row vectors
// and the low half into the alt vectors.
row0 = vec_ctf(vec_unpackh(data0), 0);
alt0 = vec_ctf(vec_unpackl(data0), 0);
row1 = vec_ctf(vec_unpackh(data1), 0);
alt1 = vec_ctf(vec_unpackl(data1), 0);
row2 = vec_ctf(vec_unpackh(data2), 0);
alt2 = vec_ctf(vec_unpackl(data2), 0);
row3 = vec_ctf(vec_unpackh(data3), 0);
alt3 = vec_ctf(vec_unpackl(data3), 0);
row4 = vec_ctf(vec_unpackh(data4), 0);
alt4 = vec_ctf(vec_unpackl(data4), 0);
row5 = vec_ctf(vec_unpackh(data5), 0);
alt5 = vec_ctf(vec_unpackl(data5), 0);
row6 = vec_ctf(vec_unpackh(data6), 0);
alt6 = vec_ctf(vec_unpackl(data6), 0);
row7 = vec_ctf(vec_unpackh(data7), 0);
alt7 = vec_ctf(vec_unpackl(data7), 0);
}
// The following block could exist as a separate an altivec dct
// function. However, if we put it inline, the DCT data can remain
// in the vector local variables, as floats, which we'll use during the
// quantize step...
{
const vector float vec_0_298631336 = (vector float)FOUROF(0.298631336f);
const vector float vec_0_390180644 = (vector float)FOUROF(-0.390180644f);
const vector float vec_0_541196100 = (vector float)FOUROF(0.541196100f);
const vector float vec_0_765366865 = (vector float)FOUROF(0.765366865f);
const vector float vec_0_899976223 = (vector float)FOUROF(-0.899976223f);
const vector float vec_1_175875602 = (vector float)FOUROF(1.175875602f);
const vector float vec_1_501321110 = (vector float)FOUROF(1.501321110f);
const vector float vec_1_847759065 = (vector float)FOUROF(-1.847759065f);
const vector float vec_1_961570560 = (vector float)FOUROF(-1.961570560f);
const vector float vec_2_053119869 = (vector float)FOUROF(2.053119869f);
const vector float vec_2_562915447 = (vector float)FOUROF(-2.562915447f);
const vector float vec_3_072711026 = (vector float)FOUROF(3.072711026f);
int whichPass, whichHalf;
for(whichPass = 1; whichPass<=2; whichPass++) {
for(whichHalf = 1; whichHalf<=2; whichHalf++) {
vector float tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
vector float tmp10, tmp11, tmp12, tmp13;
vector float z1, z2, z3, z4, z5;
tmp0 = vec_add(row0, row7); // tmp0 = dataptr[0] + dataptr[7];
tmp7 = vec_sub(row0, row7); // tmp7 = dataptr[0] - dataptr[7];
tmp3 = vec_add(row3, row4); // tmp3 = dataptr[3] + dataptr[4];
tmp4 = vec_sub(row3, row4); // tmp4 = dataptr[3] - dataptr[4];
tmp1 = vec_add(row1, row6); // tmp1 = dataptr[1] + dataptr[6];
tmp6 = vec_sub(row1, row6); // tmp6 = dataptr[1] - dataptr[6];
tmp2 = vec_add(row2, row5); // tmp2 = dataptr[2] + dataptr[5];
tmp5 = vec_sub(row2, row5); // tmp5 = dataptr[2] - dataptr[5];
tmp10 = vec_add(tmp0, tmp3); // tmp10 = tmp0 + tmp3;
tmp13 = vec_sub(tmp0, tmp3); // tmp13 = tmp0 - tmp3;
tmp11 = vec_add(tmp1, tmp2); // tmp11 = tmp1 + tmp2;
tmp12 = vec_sub(tmp1, tmp2); // tmp12 = tmp1 - tmp2;
// dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << PASS1_BITS);
row0 = vec_add(tmp10, tmp11);
// dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
row4 = vec_sub(tmp10, tmp11);
// z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
z1 = vec_madd(vec_add(tmp12, tmp13), vec_0_541196100, (vector float)zero);
// dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
// CONST_BITS-PASS1_BITS);
row2 = vec_madd(tmp13, vec_0_765366865, z1);
// dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
// CONST_BITS-PASS1_BITS);
row6 = vec_madd(tmp12, vec_1_847759065, z1);
z1 = vec_add(tmp4, tmp7); // z1 = tmp4 + tmp7;
z2 = vec_add(tmp5, tmp6); // z2 = tmp5 + tmp6;
z3 = vec_add(tmp4, tmp6); // z3 = tmp4 + tmp6;
z4 = vec_add(tmp5, tmp7); // z4 = tmp5 + tmp7;
// z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
z5 = vec_madd(vec_add(z3, z4), vec_1_175875602, (vector float)zero);
// z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z3 = vec_madd(z3, vec_1_961570560, z5);
// z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z4 = vec_madd(z4, vec_0_390180644, z5);
// The following adds are rolled into the multiplies above
// z3 = vec_add(z3, z5); // z3 += z5;
// z4 = vec_add(z4, z5); // z4 += z5;
// z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
// Wow! It's actually more efficient to roll this multiply
// into the adds below, even thought the multiply gets done twice!
// z2 = vec_madd(z2, vec_2_562915447, (vector float)zero);
// z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
// Same with this one...
// z1 = vec_madd(z1, vec_0_899976223, (vector float)zero);
// tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
// dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-PASS1_BITS);
row7 = vec_madd(tmp4, vec_0_298631336, vec_madd(z1, vec_0_899976223, z3));
// tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
// dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-PASS1_BITS);
row5 = vec_madd(tmp5, vec_2_053119869, vec_madd(z2, vec_2_562915447, z4));
// tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
// dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-PASS1_BITS);
row3 = vec_madd(tmp6, vec_3_072711026, vec_madd(z2, vec_2_562915447, z3));
// tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
// dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-PASS1_BITS);
row1 = vec_madd(z1, vec_0_899976223, vec_madd(tmp7, vec_1_501321110, z4));
// Swap the row values with the alts. If this is the first half,
// this sets up the low values to be acted on in the second half.
// If this is the second half, it puts the high values back in
// the row values where they are expected to be when we're done.
SWAP(row0, alt0);
SWAP(row1, alt1);
SWAP(row2, alt2);
SWAP(row3, alt3);
SWAP(row4, alt4);
SWAP(row5, alt5);
SWAP(row6, alt6);
SWAP(row7, alt7);
}
if (whichPass == 1) {
// transpose the data for the second pass
// First, block transpose the upper right with lower left.
SWAP(row4, alt0);
SWAP(row5, alt1);
SWAP(row6, alt2);
SWAP(row7, alt3);
// Now, transpose each block of four
TRANSPOSE4(row0, row1, row2, row3);
TRANSPOSE4(row4, row5, row6, row7);
TRANSPOSE4(alt0, alt1, alt2, alt3);
TRANSPOSE4(alt4, alt5, alt6, alt7);
}
}
}
// perform the quantize step, using the floating point data
// still in the row/alt registers
{
const int* biasAddr;
const vector signed int* qmat;
vector float bias, negBias;
if (s->mb_intra) {
vector signed int baseVector;
// We must cache element 0 in the intra case
// (it needs special handling).
baseVector = vec_cts(vec_splat(row0, 0), 0);
vec_ste(baseVector, 0, &oldBaseValue);
qmat = (vector signed int*)s->q_intra_matrix[qscale];
biasAddr = &(s->intra_quant_bias);
} else {
qmat = (vector signed int*)s->q_inter_matrix[qscale];
biasAddr = &(s->inter_quant_bias);
}
// Load the bias vector (We add 0.5 to the bias so that we're
// rounding when we convert to int, instead of flooring.)
{
vector signed int biasInt;
const vector float negOneFloat = (vector float)FOUROF(-1.0f);
LOAD4(biasInt, biasAddr);
bias = vec_ctf(biasInt, QUANT_BIAS_SHIFT);
negBias = vec_madd(bias, negOneFloat, zero);
}
{
vector float q0, q1, q2, q3, q4, q5, q6, q7;
q0 = vec_ctf(qmat[0], QMAT_SHIFT);
q1 = vec_ctf(qmat[2], QMAT_SHIFT);
q2 = vec_ctf(qmat[4], QMAT_SHIFT);
q3 = vec_ctf(qmat[6], QMAT_SHIFT);
q4 = vec_ctf(qmat[8], QMAT_SHIFT);
q5 = vec_ctf(qmat[10], QMAT_SHIFT);
q6 = vec_ctf(qmat[12], QMAT_SHIFT);
q7 = vec_ctf(qmat[14], QMAT_SHIFT);
row0 = vec_sel(vec_madd(row0, q0, negBias), vec_madd(row0, q0, bias),
vec_cmpgt(row0, zero));
row1 = vec_sel(vec_madd(row1, q1, negBias), vec_madd(row1, q1, bias),
vec_cmpgt(row1, zero));
row2 = vec_sel(vec_madd(row2, q2, negBias), vec_madd(row2, q2, bias),
vec_cmpgt(row2, zero));
row3 = vec_sel(vec_madd(row3, q3, negBias), vec_madd(row3, q3, bias),
vec_cmpgt(row3, zero));
row4 = vec_sel(vec_madd(row4, q4, negBias), vec_madd(row4, q4, bias),
vec_cmpgt(row4, zero));
row5 = vec_sel(vec_madd(row5, q5, negBias), vec_madd(row5, q5, bias),
vec_cmpgt(row5, zero));
row6 = vec_sel(vec_madd(row6, q6, negBias), vec_madd(row6, q6, bias),
vec_cmpgt(row6, zero));
row7 = vec_sel(vec_madd(row7, q7, negBias), vec_madd(row7, q7, bias),
vec_cmpgt(row7, zero));
q0 = vec_ctf(qmat[1], QMAT_SHIFT);
q1 = vec_ctf(qmat[3], QMAT_SHIFT);
q2 = vec_ctf(qmat[5], QMAT_SHIFT);
q3 = vec_ctf(qmat[7], QMAT_SHIFT);
q4 = vec_ctf(qmat[9], QMAT_SHIFT);
q5 = vec_ctf(qmat[11], QMAT_SHIFT);
q6 = vec_ctf(qmat[13], QMAT_SHIFT);
q7 = vec_ctf(qmat[15], QMAT_SHIFT);
alt0 = vec_sel(vec_madd(alt0, q0, negBias), vec_madd(alt0, q0, bias),
vec_cmpgt(alt0, zero));
alt1 = vec_sel(vec_madd(alt1, q1, negBias), vec_madd(alt1, q1, bias),
vec_cmpgt(alt1, zero));
alt2 = vec_sel(vec_madd(alt2, q2, negBias), vec_madd(alt2, q2, bias),
vec_cmpgt(alt2, zero));
alt3 = vec_sel(vec_madd(alt3, q3, negBias), vec_madd(alt3, q3, bias),
vec_cmpgt(alt3, zero));
alt4 = vec_sel(vec_madd(alt4, q4, negBias), vec_madd(alt4, q4, bias),
vec_cmpgt(alt4, zero));
alt5 = vec_sel(vec_madd(alt5, q5, negBias), vec_madd(alt5, q5, bias),
vec_cmpgt(alt5, zero));
alt6 = vec_sel(vec_madd(alt6, q6, negBias), vec_madd(alt6, q6, bias),
vec_cmpgt(alt6, zero));
alt7 = vec_sel(vec_madd(alt7, q7, negBias), vec_madd(alt7, q7, bias),
vec_cmpgt(alt7, zero));
}
}
// Store the data back into the original block
{
vector signed short data0, data1, data2, data3, data4, data5, data6, data7;
data0 = vec_pack(vec_cts(row0, 0), vec_cts(alt0, 0));
data1 = vec_pack(vec_cts(row1, 0), vec_cts(alt1, 0));
data2 = vec_pack(vec_cts(row2, 0), vec_cts(alt2, 0));
data3 = vec_pack(vec_cts(row3, 0), vec_cts(alt3, 0));
data4 = vec_pack(vec_cts(row4, 0), vec_cts(alt4, 0));
data5 = vec_pack(vec_cts(row5, 0), vec_cts(alt5, 0));
data6 = vec_pack(vec_cts(row6, 0), vec_cts(alt6, 0));
data7 = vec_pack(vec_cts(row7, 0), vec_cts(alt7, 0));
{
// Clamp for overflow
vector signed int max_q_int, min_q_int;
vector signed short max_q, min_q;
LOAD4(max_q_int, &(s->max_qcoeff));
LOAD4(min_q_int, &(s->min_qcoeff));
max_q = vec_pack(max_q_int, max_q_int);
min_q = vec_pack(min_q_int, min_q_int);
data0 = vec_max(vec_min(data0, max_q), min_q);
data1 = vec_max(vec_min(data1, max_q), min_q);
data2 = vec_max(vec_min(data2, max_q), min_q);
data4 = vec_max(vec_min(data4, max_q), min_q);
data5 = vec_max(vec_min(data5, max_q), min_q);
data6 = vec_max(vec_min(data6, max_q), min_q);
data7 = vec_max(vec_min(data7, max_q), min_q);
}
{
vector bool char zero_01, zero_23, zero_45, zero_67;
vector signed char scanIndexes_01, scanIndexes_23, scanIndexes_45, scanIndexes_67;
vector signed char negOne = vec_splat_s8(-1);
vector signed char* scanPtr =
(vector signed char*)(s->intra_scantable.inverse);
signed char lastNonZeroChar;
// Determine the largest non-zero index.
zero_01 = vec_pack(vec_cmpeq(data0, (vector signed short)zero),
vec_cmpeq(data1, (vector signed short)zero));
zero_23 = vec_pack(vec_cmpeq(data2, (vector signed short)zero),
vec_cmpeq(data3, (vector signed short)zero));
zero_45 = vec_pack(vec_cmpeq(data4, (vector signed short)zero),
vec_cmpeq(data5, (vector signed short)zero));
zero_67 = vec_pack(vec_cmpeq(data6, (vector signed short)zero),
vec_cmpeq(data7, (vector signed short)zero));
// 64 biggest values
scanIndexes_01 = vec_sel(scanPtr[0], negOne, zero_01);
scanIndexes_23 = vec_sel(scanPtr[1], negOne, zero_23);
scanIndexes_45 = vec_sel(scanPtr[2], negOne, zero_45);
scanIndexes_67 = vec_sel(scanPtr[3], negOne, zero_67);
// 32 largest values
scanIndexes_01 = vec_max(scanIndexes_01, scanIndexes_23);
scanIndexes_45 = vec_max(scanIndexes_45, scanIndexes_67);
// 16 largest values
scanIndexes_01 = vec_max(scanIndexes_01, scanIndexes_45);
// 8 largest values
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
// 4 largest values
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
// 2 largest values
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
// largest value
scanIndexes_01 = vec_max(vec_mergeh(scanIndexes_01, negOne),
vec_mergel(scanIndexes_01, negOne));
scanIndexes_01 = vec_splat(scanIndexes_01, 0);
vec_ste(scanIndexes_01, 0, &lastNonZeroChar);
lastNonZero = lastNonZeroChar;
// While the data is still in vectors we check for the transpose IDCT permute
// and handle it using the vector unit if we can. This is the permute used
// by the altivec idct, so it is common when using the altivec dct.
if ((lastNonZero > 0) && (s->dsp.idct_permutation_type == FF_TRANSPOSE_IDCT_PERM)) {
TRANSPOSE8(data0, data1, data2, data3, data4, data5, data6, data7);
}
vec_st(data0, 0, data);
vec_st(data1, 16, data);
vec_st(data2, 32, data);
vec_st(data3, 48, data);
vec_st(data4, 64, data);
vec_st(data5, 80, data);
vec_st(data6, 96, data);
vec_st(data7, 112, data);
}
}
// special handling of block[0]
if (s->mb_intra) {
if (!s->h263_aic) {
if (n < 4)
oldBaseValue /= s->y_dc_scale;
else
oldBaseValue /= s->c_dc_scale;
}
// Divide by 8, rounding the result
data[0] = (oldBaseValue + 4) >> 3;
}
// We handled the transpose permutation above and we don't
// need to permute the "no" permutation case.
if ((lastNonZero > 0) &&
(s->dsp.idct_permutation_type != FF_TRANSPOSE_IDCT_PERM) &&
(s->dsp.idct_permutation_type != FF_NO_IDCT_PERM)) {
ff_block_permute(data, s->dsp.idct_permutation,
s->intra_scantable.scantable, lastNonZero);
}
return lastNonZero;
}
void test1() {
// CHECK-LABEL: define void @test1
// CHECK-LE-LABEL: define void @test1
res_vf = vec_abs(vf);
// CHECK: call <4 x float> @llvm.fabs.v4f32(<4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x float> @llvm.fabs.v4f32(<4 x float> %{{[0-9]*}})
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_add(vd, vd);
// CHECK: fadd <2 x double>
// CHECK-LE: fadd <2 x double>
res_vd = vec_and(vbll, vd);
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
res_vd = vec_and(vd, vbll);
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
res_vd = vec_and(vd, vd);
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_andc(vbll, vd);
// CHECK: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK-LE: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_andc(vd, vbll);
// CHECK: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
// CHECK-LE: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK-LE: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK-LE: and <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
res_vd = vec_andc(vd, vd);
// CHECK: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_ceil(vd);
// CHECK: call <2 x double> @llvm.ceil.v2f64(<2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x double> @llvm.ceil.v2f64(<2 x double> %{{[0-9]*}})
res_vf = vec_ceil(vf);
// CHECK: call <4 x float> @llvm.ceil.v4f32(<4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x float> @llvm.ceil.v4f32(<4 x float> %{{[0-9]*}})
res_vbll = vec_cmpeq(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpeqdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpeqdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmpeq(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpeqsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpeqsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmpge(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmpge(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmpgt(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmpgt(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmple(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmple(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmplt(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
// CHECK-LE: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmplt(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
// CHECK-LE: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
/* vec_cpsgn */
res_vf = vec_cpsgn(vf, vf);
// CHECK: call <4 x float> @llvm.copysign.v4f32(<4 x float> %{{.+}}, <4 x float> %{{.+}})
// CHECK-LE: call <4 x float> @llvm.copysign.v4f32(<4 x float> %{{.+}}, <4 x float> %{{.+}})
res_vd = vec_cpsgn(vd, vd);
// CHECK: call <2 x double> @llvm.copysign.v2f64(<2 x double> %{{.+}}, <2 x double> %{{.+}})
// CHECK-LE: call <2 x double> @llvm.copysign.v2f64(<2 x double> %{{.+}}, <2 x double> %{{.+}})
/* vec_div */
res_vsll = vec_div(vsll, vsll);
// CHECK: sdiv <2 x i64>
// CHECK-LE: sdiv <2 x i64>
res_vull = vec_div(vull, vull);
// CHECK: udiv <2 x i64>
// CHECK-LE: udiv <2 x i64>
res_vf = vec_div(vf, vf);
// CHECK: fdiv <4 x float>
// CHECK-LE: fdiv <4 x float>
res_vd = vec_div(vd, vd);
// CHECK: fdiv <2 x double>
// CHECK-LE: fdiv <2 x double>
/* vec_max */
res_vf = vec_max(vf, vf);
// CHECK: @llvm.ppc.vsx.xvmaxsp
// CHECK-LE: @llvm.ppc.vsx.xvmaxsp
res_vd = vec_max(vd, vd);
// CHECK: @llvm.ppc.vsx.xvmaxdp
// CHECK-LE: @llvm.ppc.vsx.xvmaxdp
res_vf = vec_vmaxfp(vf, vf);
// CHECK: @llvm.ppc.vsx.xvmaxsp
// CHECK-LE: @llvm.ppc.vsx.xvmaxsp
/* vec_min */
res_vf = vec_min(vf, vf);
// CHECK: @llvm.ppc.vsx.xvminsp
// CHECK-LE: @llvm.ppc.vsx.xvminsp
res_vd = vec_min(vd, vd);
// CHECK: @llvm.ppc.vsx.xvmindp
// CHECK-LE: @llvm.ppc.vsx.xvmindp
res_vf = vec_vminfp(vf, vf);
// CHECK: @llvm.ppc.vsx.xvminsp
// CHECK-LE: @llvm.ppc.vsx.xvminsp
res_d = __builtin_vsx_xsmaxdp(d, d);
// CHECK: @llvm.ppc.vsx.xsmaxdp
// CHECK-LE: @llvm.ppc.vsx.xsmaxdp
res_d = __builtin_vsx_xsmindp(d, d);
// CHECK: @llvm.ppc.vsx.xsmindp
// CHECK-LE: @llvm.ppc.vsx.xsmindp
/* vec_perm */
res_vsll = vec_perm(vsll, vsll, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_perm(vull, vull, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vbll = vec_perm(vbll, vbll, vuc);
// CHECK: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vf = vec_round(vf);
// CHECK: call <4 x float> @llvm.round.v4f32(<4 x float>
// CHECK-LE: call <4 x float> @llvm.round.v4f32(<4 x float>
res_vd = vec_round(vd);
// CHECK: call <2 x double> @llvm.round.v2f64(<2 x double>
// CHECK-LE: call <2 x double> @llvm.round.v2f64(<2 x double>
res_vd = vec_perm(vd, vd, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vd = vec_splat(vd, 1);
// CHECK: [[T1:%.+]] = bitcast <2 x double> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x double> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x double> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x double> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vbll = vec_splat(vbll, 1);
// CHECK: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vsll = vec_splat(vsll, 1);
// CHECK: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vull = vec_splat(vull, 1);
// CHECK: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
// CHECK-LE: xor <16 x i8>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: [[T2:%.+]] = bitcast <2 x i64> {{.+}} to <4 x i32>
// CHECK-LE: call <4 x i32> @llvm.ppc.altivec.vperm(<4 x i32> [[T1]], <4 x i32> [[T2]], <16 x i8>
res_vsi = vec_pack(vsll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vui = vec_pack(vull, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vbi = vec_pack(vbll, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_vperm(vsll, vsll, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_vperm(vull, vull, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vd = vec_vperm(vd, vd, vuc);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
/* vec_vsx_ld */
res_vsi = vec_vsx_ld(0, &vsi);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vui = vec_vsx_ld(0, &vui);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vf = vec_vsx_ld (0, &vf);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vsll = vec_vsx_ld(0, &vsll);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vull = vec_vsx_ld(0, &vull);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vd = vec_vsx_ld(0, &vd);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vull = vec_vsx_ld(0, &vull);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vd = vec_vsx_ld(0, &vd);
// CHECK: @llvm.ppc.vsx.lxvd2x
// CHECK-LE: @llvm.ppc.vsx.lxvd2x
res_vss = vec_vsx_ld(0, &vss);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vss = vec_vsx_ld(0, &ss);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vus = vec_vsx_ld(0, &vus);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vus = vec_vsx_ld(0, &us);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vbc = vec_vsx_ld(0, &vbc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vsc = vec_vsx_ld(0, &vsc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vuc = vec_vsx_ld(0, &vuc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vsc = vec_vsx_ld(0, &sc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
res_vuc = vec_vsx_ld(0, &uc);
// CHECK: @llvm.ppc.vsx.lxvw4x
// CHECK-LE: @llvm.ppc.vsx.lxvw4x
/* vec_vsx_st */
vec_vsx_st(vsi, 0, &res_vsi);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsi, 0, &res_si);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vui, 0, &res_vui);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vui, 0, &res_ui);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vf, 0, &res_vf);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsll, 0, &res_vsll);
// CHECK: @llvm.ppc.vsx.stxvd2x
// CHECK-LE: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vull, 0, &res_vull);
// CHECK: @llvm.ppc.vsx.stxvd2x
// CHECK-LE: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vd, 0, &res_vd);
// CHECK: @llvm.ppc.vsx.stxvd2x
// CHECK-LE: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vss, 0, &res_vss);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vss, 0, &res_ss);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vus, 0, &res_vus);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vus, 0, &res_us);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsc, 0, &res_vsc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsc, 0, &res_sc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vuc, 0, &res_vuc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vuc, 0, &res_uc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vbc, 0, &res_vbc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vbc, 0, &res_sc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vbc, 0, &res_uc);
// CHECK: @llvm.ppc.vsx.stxvw4x
// CHECK-LE: @llvm.ppc.vsx.stxvw4x
/* vec_and */
res_vsll = vec_and(vsll, vsll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_and(vbll, vsll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_and(vsll, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_and(vull, vull);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_and(vbll, vull);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_and(vull, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vbll = vec_and(vbll, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
/* vec_vand */
res_vsll = vec_vand(vsll, vsll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_vand(vbll, vsll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_vand(vsll, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_vand(vull, vull);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_vand(vbll, vull);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_vand(vull, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
res_vbll = vec_vand(vbll, vbll);
// CHECK: and <2 x i64>
// CHECK-LE: and <2 x i64>
/* vec_andc */
res_vsll = vec_andc(vsll, vsll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_andc(vbll, vsll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vsll = vec_andc(vsll, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_andc(vull, vull);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_andc(vbll, vull);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vull = vec_andc(vull, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vbll = vec_andc(vbll, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
// CHECK-LE: xor <2 x i64>
// CHECK-LE: and <2 x i64>
res_vf = vec_floor(vf);
// CHECK: call <4 x float> @llvm.floor.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.floor.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_floor(vd);
// CHECK: call <2 x double> @llvm.floor.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.floor.v2f64(<2 x double> %{{[0-9]+}})
res_vf = vec_madd(vf, vf, vf);
// CHECK: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
res_vd = vec_madd(vd, vd, vd);
// CHECK: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
/* vec_mergeh */
res_vsll = vec_mergeh(vsll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_mergeh(vsll, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_mergeh(vbll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergeh(vull, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergeh(vull, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergeh(vbll, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
/* vec_mergel */
res_vsll = vec_mergel(vsll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_mergel(vsll, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vsll = vec_mergel(vbll, vsll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergel(vull, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergel(vull, vbll);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
res_vull = vec_mergel(vbll, vull);
// CHECK: @llvm.ppc.altivec.vperm
// CHECK-LE: @llvm.ppc.altivec.vperm
/* vec_msub */
res_vf = vec_msub(vf, vf, vf);
// CHECK: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
// CHECK-LE: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
res_vd = vec_msub(vd, vd, vd);
// CHECK: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
// CHECK-LE: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE-NEXT: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
res_vsll = vec_mul(vsll, vsll);
// CHECK: mul <2 x i64>
// CHECK-LE: mul <2 x i64>
res_vull = vec_mul(vull, vull);
// CHECK: mul <2 x i64>
// CHECK-LE: mul <2 x i64>
res_vf = vec_mul(vf, vf);
// CHECK: fmul <4 x float> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: fmul <4 x float> %{{[0-9]+}}, %{{[0-9]+}}
res_vd = vec_mul(vd, vd);
// CHECK: fmul <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: fmul <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
res_vf = vec_nearbyint(vf);
// CHECK: call <4 x float> @llvm.round.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.round.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_nearbyint(vd);
// CHECK: call <2 x double> @llvm.round.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.round.v2f64(<2 x double> %{{[0-9]+}})
res_vf = vec_nmadd(vf, vf, vf);
// CHECK: [[FM:[0-9]+]] = call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
// CHECK-NEXT: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %[[FM]]
// CHECK-LE: [[FM:[0-9]+]] = call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
// CHECK-LE-NEXT: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %[[FM]]
res_vd = vec_nmadd(vd, vd, vd);
// CHECK: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
// CHECK-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
// CHECK-LE: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
// CHECK-LE-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
res_vf = vec_nmsub(vf, vf, vf);
// CHECK: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
// CHECK: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
// CHECK-LE: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
res_vd = vec_nmsub(vd, vd, vd);
// CHECK: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
// CHECK-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
// CHECK-LE: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-LE-NEXT: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
// CHECK-LE-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
/* vec_nor */
res_vsll = vec_nor(vsll, vsll);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
// CHECK-LE: or <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_nor(vull, vull);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
// CHECK-LE: or <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_nor(vbll, vbll);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
// CHECK-LE: or <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vd = vec_nor(vd, vd);
// CHECK: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: [[OR:%.+]] = or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-NEXT: xor <2 x i64> [[OR]], <i64 -1, i64 -1>
// CHECK-LE: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK-LE: [[OR:%.+]] = or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE-NEXT: xor <2 x i64> [[OR]], <i64 -1, i64 -1>
/* vec_or */
res_vsll = vec_or(vsll, vsll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vsll = vec_or(vbll, vsll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vsll = vec_or(vsll, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_or(vull, vull);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_or(vbll, vull);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_or(vull, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vbll = vec_or(vbll, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vd = vec_or(vd, vd);
// CHECK: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK-LE: or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
res_vd = vec_or(vbll, vd);
// CHECK: [[T1:%.+]] = bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: [[T2:%.+]] = or <2 x i64> %{{[0-9]+}}, [[T1]]
// CHECK: bitcast <2 x i64> [[T2]] to <2 x double>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK-LE: [[T2:%.+]] = or <2 x i64> %{{[0-9]+}}, [[T1]]
// CHECK-LE: bitcast <2 x i64> [[T2]] to <2 x double>
res_vd = vec_or(vd, vbll);
// CHECK: [[T1:%.+]] = bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: [[T2:%.+]] = or <2 x i64> [[T1]], %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[T2]] to <2 x double>
// CHECK-LE: [[T1:%.+]] = bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK-LE: [[T2:%.+]] = or <2 x i64> [[T1]], %{{[0-9]+}}
// CHECK-LE: bitcast <2 x i64> [[T2]] to <2 x double>
res_vf = vec_re(vf);
// CHECK: call <4 x float> @llvm.ppc.vsx.xvresp(<4 x float>
// CHECK-LE: call <4 x float> @llvm.ppc.vsx.xvresp(<4 x float>
res_vd = vec_re(vd);
// CHECK: call <2 x double> @llvm.ppc.vsx.xvredp(<2 x double>
// CHECK-LE: call <2 x double> @llvm.ppc.vsx.xvredp(<2 x double>
res_vf = vec_rint(vf);
// CHECK: call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_rint(vd);
// CHECK: call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %{{[0-9]+}})
res_vf = vec_rsqrte(vf);
// CHECK: call <4 x float> @llvm.ppc.vsx.xvrsqrtesp(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.ppc.vsx.xvrsqrtesp(<4 x float> %{{[0-9]+}})
res_vd = vec_rsqrte(vd);
// CHECK: call <2 x double> @llvm.ppc.vsx.xvrsqrtedp(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.ppc.vsx.xvrsqrtedp(<2 x double> %{{[0-9]+}})
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vf = vec_sel(vd, vd, vbll);
// CHECK: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64> %{{[0-9]+}},
// CHECK: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: or <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
// CHECK-LE: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK-LE: and <2 x i64> %{{[0-9]+}},
// CHECK-LE: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: or <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_sel(vd, vd, vull);
// CHECK: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64> %{{[0-9]+}},
// CHECK: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: or <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
// CHECK-LE: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK-LE: and <2 x i64> %{{[0-9]+}},
// CHECK-LE: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: or <2 x i64>
// CHECK-LE: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
res_vf = vec_sqrt(vf);
// CHECK: call <4 x float> @llvm.sqrt.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.sqrt.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_sqrt(vd);
// CHECK: call <2 x double> @llvm.sqrt.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.sqrt.v2f64(<2 x double> %{{[0-9]+}})
res_vd = vec_sub(vd, vd);
// CHECK: fsub <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: fsub <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
res_vf = vec_trunc(vf);
// CHECK: call <4 x float> @llvm.trunc.v4f32(<4 x float> %{{[0-9]+}})
// CHECK-LE: call <4 x float> @llvm.trunc.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_trunc(vd);
// CHECK: call <2 x double> @llvm.trunc.v2f64(<2 x double> %{{[0-9]+}})
// CHECK-LE: call <2 x double> @llvm.trunc.v2f64(<2 x double> %{{[0-9]+}})
/* vec_vor */
res_vsll = vec_vor(vsll, vsll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vsll = vec_vor(vbll, vsll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vsll = vec_vor(vsll, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_vor(vull, vull);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_vor(vbll, vull);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vull = vec_vor(vull, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
res_vbll = vec_vor(vbll, vbll);
// CHECK: or <2 x i64>
// CHECK-LE: or <2 x i64>
/* vec_xor */
res_vsll = vec_xor(vsll, vsll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_xor(vbll, vsll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_xor(vsll, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_xor(vull, vull);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_xor(vbll, vull);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_xor(vull, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vbll = vec_xor(vbll, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_xor(vd, vd);
// CHECK: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[X1]] to <2 x double>
// CHECK-LE: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: bitcast <2 x i64> [[X1]] to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_xor(vd, vbll);
// CHECK: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[X1]] to <2 x double>
// CHECK-LE: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: bitcast <2 x i64> [[X1]] to <2 x double>
dummy();
// CHECK: call void @dummy()
// CHECK-LE: call void @dummy()
res_vd = vec_xor(vbll, vd);
// CHECK: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[X1]] to <2 x double>
// CHECK-LE: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-LE: bitcast <2 x i64> [[X1]] to <2 x double>
/* vec_vxor */
res_vsll = vec_vxor(vsll, vsll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_vxor(vbll, vsll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_vxor(vsll, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_vxor(vull, vull);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_vxor(vbll, vull);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vull = vec_vxor(vull, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vbll = vec_vxor(vbll, vbll);
// CHECK: xor <2 x i64>
// CHECK-LE: xor <2 x i64>
res_vsll = vec_cts(vd, 0);
// CHECK: fmul <2 x double>
// CHECK: fptosi <2 x double> %{{.*}} to <2 x i64>
// CHECK-LE: fmul <2 x double>
// CHECK-LE: fptosi <2 x double> %{{.*}} to <2 x i64>
res_vsll = vec_cts(vd, 31);
// CHECK: fmul <2 x double>
// CHECK: fptosi <2 x double> %{{.*}} to <2 x i64>
// CHECK-LE: fmul <2 x double>
// CHECK-LE: fptosi <2 x double> %{{.*}} to <2 x i64>
res_vsll = vec_ctu(vd, 0);
// CHECK: fmul <2 x double>
// CHECK: fptoui <2 x double> %{{.*}} to <2 x i64>
// CHECK-LE: fmul <2 x double>
// CHECK-LE: fptoui <2 x double> %{{.*}} to <2 x i64>
res_vsll = vec_ctu(vd, 31);
// CHECK: fmul <2 x double>
// CHECK: fptoui <2 x double> %{{.*}} to <2 x i64>
// CHECK-LE: fmul <2 x double>
// CHECK-LE: fptoui <2 x double> %{{.*}} to <2 x i64>
res_vd = vec_ctf(vsll, 0);
// CHECK: sitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK: fmul <2 x double>
// CHECK-LE: sitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK-LE: fmul <2 x double>
res_vd = vec_ctf(vsll, 31);
// CHECK: sitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK: fmul <2 x double>
// CHECK-LE: sitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK-LE: fmul <2 x double>
res_vd = vec_ctf(vull, 0);
// CHECK: uitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK: fmul <2 x double>
// CHECK-LE: uitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK-LE: fmul <2 x double>
res_vd = vec_ctf(vull, 31);
// CHECK: uitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK: fmul <2 x double>
// CHECK-LE: uitofp <2 x i64> %{{.*}} to <2 x double>
// CHECK-LE: fmul <2 x double>
}
static inline
vec_uint4 vec_Ch(vec_uint4 x, vec_uint4 y, vec_uint4 z)
{
return vec_sel(z, y, x);
}
static inline
vec_uint4 vec_Maj(vec_uint4 x, vec_uint4 y, vec_uint4 z)
{
return vec_sel(x, y, vec_xor(x, z));
}
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
}
}
}
SIMD_INLINE void ConditionalFill(const uint8_t * src, size_t offset, const v128_u8 & threshold, const v128_u8 & value, uint8_t * dst)
{
v128_u8 _src = Load<align>(src + offset);
v128_u8 _dst = Load<align>(dst + offset);
Store<align>(dst + offset, vec_sel(_dst, value, Compare8u<compareType>(_src, threshold)));
}
void init_particles(particle* system){
/*
* particles will be placed randomly 4 quadrants of 3d space
* q1: x>0, y>0
* q2: x<0, y>0
* q3: x<0, y<0
* q4: x>0, y<0
*
*/
int i = 0;
srand(time(NULL));
// q1 x>0, y>0
for(; i < NO_OF_PARTICLES/4; i++){
system[i].position = (__vector float){rand()%INIT_BOUNDING_BOX,
rand()%INIT_BOUNDING_BOX,
(rand()%INIT_BOUNDING_BOX*2)-INIT_BOUNDING_BOX, 0};
system[i].velocity = VECZERO;
system[i].acceleration = VECZERO;
}
// q2 x<0, y>0
for(; i < NO_OF_PARTICLES/2; i++){
system[i].position = (__vector float){-rand()%INIT_BOUNDING_BOX,
rand()%INIT_BOUNDING_BOX,
(rand()%INIT_BOUNDING_BOX*2)-INIT_BOUNDING_BOX, 0};
system[i].velocity = VECZERO;
system[i].acceleration = VECZERO;
}
// q3 x<0, y<0
for(; i < 3*NO_OF_PARTICLES/4; i++){
system[i].position = (__vector float){-rand()%INIT_BOUNDING_BOX,
-rand()%INIT_BOUNDING_BOX,
(rand()%INIT_BOUNDING_BOX*2)-INIT_BOUNDING_BOX, 0};
system[i].velocity = VECZERO;
system[i].acceleration = VECZERO;
}
// q4 x>0, y<0
for(; i < NO_OF_PARTICLES; i++){
system[i].position = (__vector float){rand()%INIT_BOUNDING_BOX,
-rand()%INIT_BOUNDING_BOX,
(rand()%INIT_BOUNDING_BOX*2)-INIT_BOUNDING_BOX, 0};
system[i].velocity = VECZERO;
system[i].acceleration = VECZERO;
}
}
void compute_interaction(particle* i , particle* j){
__vector float radius, radius_sqr, s_vector, displ, accel, distSqr,
distSixth, invDistCube;
/*compute acceleration of particle i*/
radius = vec_sub(j->position,i->position);
radius_sqr = vec_madd(radius,radius, VECZERO);
distSqr = vec_add(vec_splat(radius_sqr,0),vec_splat(radius_sqr,1));
distSqr = vec_add(vec_splat(radius_sqr,2),distSqr);
distSqr = vec_add(EPS2_VECTOR,distSqr);
distSixth = vec_madd(distSqr,distSqr,VECZERO);
distSixth = vec_madd(distSixth,distSqr,VEC3ZERO);
invDistCube = vec_rsqrte(distSixth);
s_vector = vec_madd(MASS_VECTOR,invDistCube,VECZERO);
i->acceleration = vec_madd(radius,s_vector,i->acceleration);
/*compute new position & velocity of particle i*/
displ = vec_madd(i->velocity,TIME_STEP_VECTOR,i->position);
accel = vec_madd(VECHALF,i->acceleration, VECZERO);
i->position = vec_madd(accel,TIME_SQUARED, displ);
i->velocity = vec_madd(i->acceleration,TIME_STEP_VECTOR, i->velocity);
}
void update_particles(particle* system){
int i, j;
//Thread 1 ?
for(i = 0;i<NO_OF_PARTICLES/4;i++){
for(j = 0;j<NO_OF_PARTICLES;j++){
compute_interaction(&system[i],&system[j]);
}
}
//Thread 2 ?
for(i = NO_OF_PARTICLES/4;i<NO_OF_PARTICLES/2;i++){
for(j = 0;j<NO_OF_PARTICLES;j++){
compute_interaction(&system[i],&system[j]);
}
}
//Thread 3 ?
for(i = NO_OF_PARTICLES/2;i<3*NO_OF_PARTICLES/4;i++){
for(j = 0;j<NO_OF_PARTICLES;j++){
compute_interaction(&system[i],&system[j]);
}
}
//Thread 4 ?
for(i = 3*NO_OF_PARTICLES/4;i<NO_OF_PARTICLES;i++){
for(j = 0;j<NO_OF_PARTICLES;j++){
compute_interaction(&system[i],&system[j]);
}
}
}
__vector int get_quadrant_count(particle* system){
__vector int quad_count = VECINTZERO;
__vector int quad_mask = VECINTZERO;
int i;
for(i = 0; i < NO_OF_PARTICLES ; i++){
__vector int top2 = (__vector int){1,1,0,0};
__vector int bottom2 = (__vector int){0,0,1,1};
__vector int left2 = (__vector int){0,1,0,1};
__vector int right2 = (__vector int){1,0,1,0};
__vector int mask1, mask2;
__vector float vx = vec_splat(system[i].position,0);
__vector float vy = vec_splat(system[i].position,1);
mask1 = vec_sel(right2, left2, vec_cmpgt(vx,VECZERO));
mask2 = vec_sel(top2, bottom2, vec_cmpgt(vy,VECZERO));
quad_mask = vec_and(mask1,mask2);
quad_count = vec_add(quad_count,quad_mask);
}
return quad_count;
}
void render(particle* system){
int i = 0;
for(; i < NO_OF_PARTICLES; i++){
float *pos = (float*) &system[i].position;
float *vel = (float*) &system[i].velocity;
float *acc = (float*) &system[i].acceleration;
printf("position : %f %f %f ", pos[0], pos[1], pos[2]);
printf("velocity : %f %f %f ", vel[0], vel[1], vel[2]);
printf("acceleration : %f %f %f \n", acc[0], acc[1], acc[2]);
}
}
int main ()
{
/* create particle system --> array of particles */
particle particle_system[NO_OF_PARTICLES] __attribute__((aligned(64)));
/* vector that tracks the no. of particles in each quadrant */
__vector int quad_count;
int * qc;
/* place particles in 4 quadrants */
init_particles(particle_system);
/* run simulation */
float simulationTime = 0.0;
int iterations = COMPUTE_ITERATIONS;
printf("----------------------------------------------");
printf("----------------------------------------------\n");
printf("Running Simulation with %d particles & %d iterations with %f seconds time steps\n",
NO_OF_PARTICLES, COMPUTE_ITERATIONS, TIME_STEP);
printf("----------------------------------------------");
printf("----------------------------------------------\n");
while(iterations > 0){
/* Compute */
update_particles(particle_system);
/* Display */
//render(particle_system);
/* Update Time */
simulationTime = simulationTime + TIME_STEP;
printf("----------------------------------");
printf("Simulation Time: %f |",simulationTime);
quad_count = get_quadrant_count(particle_system);
qc = (int*)&quad_count;
printf(" q1:%d q2:%d q3:%d q4:%d",qc[0], qc[1], qc[2], qc[3]);
printf("----------------------------------\n");
iterations --;
}
return 0;
}
/* 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);
}
/* AltiVec version of dct_unquantize_h263
this code assumes `block' is 16 bytes-aligned */
static void dct_unquantize_h263_altivec(MpegEncContext *s,
DCTELEM *block, int n, int qscale)
{
int i, level, qmul, qadd;
int nCoeffs;
assert(s->block_last_index[n]>=0);
qadd = (qscale - 1) | 1;
qmul = qscale << 1;
if (s->mb_intra) {
if (!s->h263_aic) {
if (n < 4)
block[0] = block[0] * s->y_dc_scale;
else
block[0] = block[0] * s->c_dc_scale;
}else
qadd = 0;
i = 1;
nCoeffs= 63; //does not always use zigzag table
} else {
i = 0;
nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
}
{
register const vector signed short vczero = (const vector signed short)vec_splat_s16(0);
DECLARE_ALIGNED(16, short, qmul8) = qmul;
DECLARE_ALIGNED(16, short, qadd8) = qadd;
register vector signed short blockv, qmulv, qaddv, nqaddv, temp1;
register vector bool short blockv_null, blockv_neg;
register short backup_0 = block[0];
register int j = 0;
qmulv = vec_splat((vec_s16)vec_lde(0, &qmul8), 0);
qaddv = vec_splat((vec_s16)vec_lde(0, &qadd8), 0);
nqaddv = vec_sub(vczero, qaddv);
#if 0 // block *is* 16 bytes-aligned, it seems.
// first make sure block[j] is 16 bytes-aligned
for(j = 0; (j <= nCoeffs) && ((((unsigned long)block) + (j << 1)) & 0x0000000F) ; j++) {
level = block[j];
if (level) {
if (level < 0) {
level = level * qmul - qadd;
} else {
level = level * qmul + qadd;
}
block[j] = level;
}
}
#endif
// vectorize all the 16 bytes-aligned blocks
// of 8 elements
for(; (j + 7) <= nCoeffs ; j+=8) {
blockv = vec_ld(j << 1, block);
blockv_neg = vec_cmplt(blockv, vczero);
blockv_null = vec_cmpeq(blockv, vczero);
// choose between +qadd or -qadd as the third operand
temp1 = vec_sel(qaddv, nqaddv, blockv_neg);
// multiply & add (block{i,i+7} * qmul [+-] qadd)
temp1 = vec_mladd(blockv, qmulv, temp1);
// put 0 where block[{i,i+7} used to have 0
blockv = vec_sel(temp1, blockv, blockv_null);
vec_st(blockv, j << 1, block);
}
// if nCoeffs isn't a multiple of 8, finish the job
// using good old scalar units.
// (we could do it using a truncated vector,
// but I'm not sure it's worth the hassle)
for(; j <= nCoeffs ; j++) {
level = block[j];
if (level) {
if (level < 0) {
level = level * qmul - qadd;
} else {
level = level * qmul + qadd;
}
block[j] = level;
}
}
if (i == 1) {
// cheat. this avoid special-casing the first iteration
block[0] = backup_0;
}
}
}
tmp0 = vec_sl(index0, (vector unsigned short)((((vector signed short){1,1,1,1,1,1,1,1})) ));
PerIndex = (vector unsigned char)vec_packs(tmp0, vec_add(tmp0, (((vector signed short){1,1,1,1,1,1,1,1})) ));
PerIndex = vec_perm(PerIndex, PerIndex, perm1);
tmp0 = vec_perm(table[0], table[1], PerIndex);
stmp0 = vec_perm(slope_cos[0], slope_cos[1], PerIndex);
tmpIndex = vec_sub(PerIndex, (((vector unsigned char){32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32})) );
tmp1 = vec_perm(table[2], table[3], tmpIndex);
stmp1 = vec_perm(slope_cos[2], slope_cos[3], tmpIndex);
select = (vector unsigned short)vec_cmpgt(PerIndex, (((vector unsigned char){31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31})) );
tmp2 = vec_sel(tmp0, tmp1, select);
stmp2 = vec_sel(stmp0, stmp1, select);
tmpIndex = vec_sub(tmpIndex, (((vector unsigned char){32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32})) );
tmp0 = vec_perm(table[4], table[5], tmpIndex);
stmp0 = vec_perm(slope_cos[4], slope_cos[5], tmpIndex);
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})) );
static void vector_fmul_add_add_altivec(float *dst, const float *src0,
const float *src1, const float *src2,
int src3, int len, int step)
{
int i;
vector float d, s0, s1, s2, t0, t1, edges;
vector unsigned char align = vec_lvsr(0,dst),
mask = vec_lvsl(0, dst);
#if 0 //FIXME: there is still something wrong
if (step == 2) {
int y;
vector float d0, d1, s3, t2;
vector unsigned int sel =
vec_mergeh(vec_splat_u32(-1), vec_splat_u32(0));
t1 = vec_ld(16, dst);
for (i=0,y=0; i<len-3; i+=4,y+=8) {
s0 = vec_ld(0,src0+i);
s1 = vec_ld(0,src1+i);
s2 = vec_ld(0,src2+i);
// t0 = vec_ld(0, dst+y); //[x x x|a]
// t1 = vec_ld(16, dst+y); //[b c d|e]
t2 = vec_ld(31, dst+y); //[f g h|x]
d = vec_madd(s0,s1,s2); // [A B C D]
// [A A B B]
// [C C D D]
d0 = vec_perm(t0, t1, mask); // [a b c d]
d0 = vec_sel(vec_mergeh(d, d), d0, sel); // [A b B d]
edges = vec_perm(t1, t0, mask);
t0 = vec_perm(edges, d0, align); // [x x x|A]
t1 = vec_perm(d0, edges, align); // [b B d|e]
vec_stl(t0, 0, dst+y);
d1 = vec_perm(t1, t2, mask); // [e f g h]
d1 = vec_sel(vec_mergel(d, d), d1, sel); // [C f D h]
edges = vec_perm(t2, t1, mask);
t1 = vec_perm(edges, d1, align); // [b B d|C]
t2 = vec_perm(d1, edges, align); // [f D h|x]
vec_stl(t1, 16, dst+y);
t0 = t1;
vec_stl(t2, 31, dst+y);
t1 = t2;
}
} else
#endif
if (step == 1 && src3 == 0)
for (i=0; i<len-3; i+=4) {
t0 = vec_ld(0, dst+i);
t1 = vec_ld(15, dst+i);
s0 = vec_ld(0, src0+i);
s1 = vec_ld(0, src1+i);
s2 = vec_ld(0, src2+i);
edges = vec_perm(t1 ,t0, mask);
d = vec_madd(s0,s1,s2);
t1 = vec_perm(d, edges, align);
t0 = vec_perm(edges, d, align);
vec_st(t1, 15, dst+i);
vec_st(t0, 0, dst+i);
}
else
ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
}
static void ProjectDlightTexture_altivec( void ) {
int i, l;
vec_t origin0, origin1, origin2;
float texCoords0, texCoords1;
vector float floatColorVec0, floatColorVec1;
vector float modulateVec, colorVec, zero;
vector short colorShort;
vector signed int colorInt;
vector unsigned char floatColorVecPerm, modulatePerm, colorChar;
vector unsigned char vSel = VECCONST_UINT8(0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff);
float *texCoords;
byte *colors;
int *intColors;
byte clipBits[SHADER_MAX_VERTEXES];
float texCoordsArray[SHADER_MAX_VERTEXES][2];
byte colorArray[SHADER_MAX_VERTEXES][4];
glIndex_t hitIndexes[SHADER_MAX_INDEXES];
int numIndexes;
float scale;
float radius;
float radiusInverseCubed;
float intensity, remainder;
vec3_t floatColor;
float modulate = 0.0f;
qboolean vertexLight;
if ( !backEnd.refdef.num_dlights ) {
return;
}
// There has to be a better way to do this so that floatColor
// and/or modulate are already 16-byte aligned.
floatColorVecPerm = vec_lvsl(0,(float *)floatColor);
modulatePerm = vec_lvsl(0,(float *)&modulate);
modulatePerm = (vector unsigned char)vec_splat((vector unsigned int)modulatePerm,0);
zero = (vector float)vec_splat_s8(0);
for ( l = 0 ; l < backEnd.refdef.num_dlights ; l++ ) {
dlight_t *dl;
if ( !( tess.dlightBits & ( 1 << l ) ) ) {
continue; // this surface definately doesn't have any of this light
}
// clear colors
Com_Memset( colorArray, 0, sizeof( colorArray ) );
texCoords = texCoordsArray[0];
colors = colorArray[0];
dl = &backEnd.refdef.dlights[l];
origin0 = dl->transformed[0];
origin1 = dl->transformed[1];
origin2 = dl->transformed[2];
radius = dl->radius;
scale = 1.0f / radius;
radiusInverseCubed = dl->radiusInverseCubed;
intensity = dl->intensity;
vertexLight = ( ( dl->flags & REF_DIRECTED_DLIGHT ) || ( dl->flags & REF_VERTEX_DLIGHT ) );
// directional lights have max intensity and washout remainder intensity
if ( dl->flags & REF_DIRECTED_DLIGHT ) {
remainder = intensity * 0.125;
} else {
remainder = 0.0f;
}
if(r_greyscale->integer)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = floatColor[1] = floatColor[2] = luminance;
}
else if(r_greyscale->value)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = LERP(dl->color[0] * 255.0f, luminance, r_greyscale->value);
floatColor[1] = LERP(dl->color[1] * 255.0f, luminance, r_greyscale->value);
floatColor[2] = LERP(dl->color[2] * 255.0f, luminance, r_greyscale->value);
}
else
{
floatColor[0] = dl->color[0] * 255.0f;
floatColor[1] = dl->color[1] * 255.0f;
floatColor[2] = dl->color[2] * 255.0f;
}
floatColorVec0 = vec_ld(0, floatColor);
floatColorVec1 = vec_ld(11, floatColor);
floatColorVec0 = vec_perm(floatColorVec0,floatColorVec0,floatColorVecPerm);
for ( i = 0 ; i < tess.numVertexes ; i++, texCoords += 2, colors += 4 ) {
int clip = 0;
vec_t dist0, dist1, dist2;
dist0 = origin0 - tess.xyz[i][0];
dist1 = origin1 - tess.xyz[i][1];
dist2 = origin2 - tess.xyz[i][2];
backEnd.pc.c_dlightVertexes++;
// directional dlight, origin is a directional normal
if ( dl->flags & REF_DIRECTED_DLIGHT ) {
// twosided surfaces use absolute value of the calculated lighting
modulate = intensity * DotProduct( dl->origin, tess.normal[ i ] );
if ( tess.shader->cullType == CT_TWO_SIDED ) {
modulate = fabs( modulate );
}
modulate += remainder;
}
// spherical vertex lit dlight
else if ( dl->flags & REF_VERTEX_DLIGHT )
{
vec3_t dir;
dir[ 0 ] = radius - fabs( dist0 );
if ( dir[ 0 ] <= 0.0f ) {
continue;
}
dir[ 1 ] = radius - fabs( dist1 );
if ( dir[ 1 ] <= 0.0f ) {
continue;
}
dir[ 2 ] = radius - fabs( dist2 );
if ( dir[ 2 ] <= 0.0f ) {
continue;
}
modulate = intensity * dir[ 0 ] * dir[ 1 ] * dir[ 2 ] * radiusInverseCubed;
}
// vertical cylinder dlight
else
{
texCoords0 = 0.5f + dist0 * scale;
texCoords1 = 0.5f + dist1 * scale;
if( !r_dlightBacks->integer &&
// dist . tess.normal[i]
( dist0 * tess.normal[i][0] +
dist1 * tess.normal[i][1] +
dist2 * tess.normal[i][2] ) < 0.0f ) {
clip = 63;
} else {
if ( texCoords0 < 0.0f ) {
clip |= 1;
} else if ( texCoords0 > 1.0f ) {
clip |= 2;
}
if ( texCoords1 < 0.0f ) {
clip |= 4;
} else if ( texCoords1 > 1.0f ) {
clip |= 8;
}
texCoords[0] = texCoords0;
texCoords[1] = texCoords1;
// modulate the strength based on the height and color
if ( dist2 > radius ) {
clip |= 16;
modulate = 0.0f;
} else if ( dist2 < -radius ) {
clip |= 32;
modulate = 0.0f;
} else {
dist2 = Q_fabs(dist2);
if ( dist2 < radius * 0.5f ) {
modulate = intensity;
} else {
modulate = intensity * 2.0f * (radius - dist2) * scale;
}
}
}
}
clipBits[i] = clip;
// optimizations
if ( vertexLight && modulate < ( 1.0f / 128.0f ) ) {
continue;
} else if ( modulate > 1.0f ) {
modulate = 1.0f;
}
// ZTM: FIXME: should probably clamp to 0-255 range before converting to char,
// but I don't know how to do altvec stuff or if it's even used anymore
modulateVec = vec_ld(0,(float *)&modulate);
modulateVec = vec_perm(modulateVec,modulateVec,modulatePerm);
colorVec = vec_madd(floatColorVec0,modulateVec,zero);
colorInt = vec_cts(colorVec,0); // RGBx
colorShort = vec_pack(colorInt,colorInt); // RGBxRGBx
colorChar = vec_packsu(colorShort,colorShort); // RGBxRGBxRGBxRGBx
colorChar = vec_sel(colorChar,vSel,vSel); // RGBARGBARGBARGBA replace alpha with 255
vec_ste((vector unsigned int)colorChar,0,(unsigned int *)colors); // store color
}
// build a list of triangles that need light
intColors = (int*) colorArray;
numIndexes = 0;
for ( i = 0 ; i < tess.numIndexes ; i += 3 ) {
int a, b, c;
a = tess.indexes[i];
b = tess.indexes[i+1];
c = tess.indexes[i+2];
if ( vertexLight ) {
if ( !( intColors[ a ] | intColors[ b ] | intColors[ c ] ) ) {
continue;
}
} else {
if ( clipBits[a] & clipBits[b] & clipBits[c] ) {
continue; // not lighted
}
}
hitIndexes[numIndexes] = a;
hitIndexes[numIndexes+1] = b;
hitIndexes[numIndexes+2] = c;
numIndexes += 3;
}
if ( !numIndexes ) {
continue;
}
if ( !vertexLight ) {
qglEnableClientState( GL_TEXTURE_COORD_ARRAY );
qglTexCoordPointer( 2, GL_FLOAT, 0, texCoordsArray[0] );
} else {
qglDisableClientState( GL_TEXTURE_COORD_ARRAY );
}
qglEnableClientState( GL_COLOR_ARRAY );
qglColorPointer( 4, GL_UNSIGNED_BYTE, 0, colorArray );
if ( dl->dlshader ) {
shader_t *dls = dl->dlshader;
for ( i = 0; i < dls->numUnfoggedPasses; i++ ) {
shaderStage_t *stage = dls->stages[i];
R_BindAnimatedImage( &dls->stages[i]->bundle[0] );
GL_State( stage->stateBits | GLS_DEPTHFUNC_EQUAL );
R_DrawElements( numIndexes, hitIndexes );
backEnd.pc.c_totalIndexes += numIndexes;
backEnd.pc.c_dlightIndexes += numIndexes;
}
} else {
R_FogOff();
if ( !vertexLight ) {
GL_Bind( tr.dlightImage );
} else {
GL_Bind( tr.whiteImage );
}
// include GLS_DEPTHFUNC_EQUAL so alpha tested surfaces don't add light
// where they aren't rendered
if ( dl->flags & REF_ADDITIVE_DLIGHT ) {
GL_State( GLS_SRCBLEND_ONE | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
else {
GL_State( GLS_SRCBLEND_DST_COLOR | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
R_DrawElements( numIndexes, hitIndexes );
backEnd.pc.c_totalIndexes += numIndexes;
backEnd.pc.c_dlightIndexes += numIndexes;
RB_FogOn();
}
}
}
int main ()
{
vector float fa = {1.0, 2.0, 3.0, -4.0};
vector float fb = {-2.0, -3.0, -4.0, -5.0};
vector float fc = vec_cpsgn (fa, fb);
vector long long la = {5L, 14L};
vector long long lb = {3L, 86L};
vector long long lc = vec_and (la, lb);
vector bool long long ld = {0, -1};
vector long long le = vec_and (la, ld);
vector long long lf = vec_and (ld, lb);
vector unsigned long long ua = {5L, 14L};
vector unsigned long long ub = {3L, 86L};
vector unsigned long long uc = vec_and (ua, ub);
vector bool long long ud = {0, -1};
vector unsigned long long ue = vec_and (ua, ud);
vector unsigned long long uf = vec_and (ud, ub);
vector long long lg = vec_andc (la, lb);
vector long long lh = vec_andc (la, ld);
vector long long li = vec_andc (ld, lb);
vector unsigned long long ug = vec_andc (ua, ub);
vector unsigned long long uh = vec_andc (ua, ud);
vector unsigned long long ui = vec_andc (ud, ub);
vector double da = {1.0, -4.0};
vector double db = {-2.0, 5.0};
vector double dc = vec_cpsgn (da, db);
vector long long lj = vec_mergeh (la, lb);
vector long long lk = vec_mergeh (la, ld);
vector long long ll = vec_mergeh (ld, la);
vector unsigned long long uj = vec_mergeh (ua, ub);
vector unsigned long long uk = vec_mergeh (ua, ud);
vector unsigned long long ul = vec_mergeh (ud, ua);
vector long long lm = vec_mergel (la, lb);
vector long long ln = vec_mergel (la, ld);
vector long long lo = vec_mergel (ld, la);
vector unsigned long long um = vec_mergel (ua, ub);
vector unsigned long long un = vec_mergel (ua, ud);
vector unsigned long long uo = vec_mergel (ud, ua);
vector long long lp = vec_nor (la, lb);
vector long long lq = vec_nor (la, ld);
vector long long lr = vec_nor (ld, la);
vector unsigned long long up = vec_nor (ua, ub);
vector unsigned long long uq = vec_nor (ua, ud);
vector unsigned long long ur = vec_nor (ud, ua);
vector long long ls = vec_or (la, lb);
vector long long lt = vec_or (la, ld);
vector long long lu = vec_or (ld, la);
vector unsigned long long us = vec_or (ua, ub);
vector unsigned long long ut = vec_or (ua, ud);
vector unsigned long long uu = vec_or (ud, ua);
vector unsigned char ca = {0,4,8,1,5,9,2,6,10,3,7,11,15,12,14,13};
vector long long lv = vec_perm (la, lb, ca);
vector unsigned long long uv = vec_perm (ua, ub, ca);
vector long long lw = vec_sel (la, lb, lc);
vector long long lx = vec_sel (la, lb, uc);
vector long long ly = vec_sel (la, lb, ld);
vector unsigned long long uw = vec_sel (ua, ub, lc);
vector unsigned long long ux = vec_sel (ua, ub, uc);
vector unsigned long long uy = vec_sel (ua, ub, ld);
vector long long lz = vec_xor (la, lb);
vector long long l0 = vec_xor (la, ld);
vector long long l1 = vec_xor (ld, la);
vector unsigned long long uz = vec_xor (ua, ub);
vector unsigned long long u0 = vec_xor (ua, ud);
vector unsigned long long u1 = vec_xor (ud, ua);
int ia = vec_all_eq (ua, ub);
int ib = vec_all_ge (ua, ub);
int ic = vec_all_gt (ua, ub);
int id = vec_all_le (ua, ub);
int ie = vec_all_lt (ua, ub);
int ig = vec_all_ne (ua, ub);
int ih = vec_any_eq (ua, ub);
int ii = vec_any_ge (ua, ub);
int ij = vec_any_gt (ua, ub);
int ik = vec_any_le (ua, ub);
int il = vec_any_lt (ua, ub);
int im = vec_any_ne (ua, ub);
vector int sia = {9, 16, 25, 36};
vector int sib = {-8, -27, -64, -125};
vector int sic = vec_mergee (sia, sib);
vector int sid = vec_mergeo (sia, sib);
vector unsigned int uia = {9, 16, 25, 36};
vector unsigned int uib = {8, 27, 64, 125};
vector unsigned int uic = vec_mergee (uia, uib);
vector unsigned int uid = vec_mergeo (uia, uib);
vector bool int bia = {0, -1, -1, 0};
vector bool int bib = {-1, -1, 0, -1};
vector bool int bic = vec_mergee (bia, bib);
vector bool int bid = vec_mergeo (bia, bib);
vector unsigned int uie = vec_packsu (ua, ub);
vector long long l2 = vec_cntlz (la);
vector unsigned long long u2 = vec_cntlz (ua);
vector int sie = vec_cntlz (sia);
vector unsigned int uif = vec_cntlz (uia);
vector short ssa = {20, -40, -60, 80, 100, -120, -140, 160};
vector short ssb = vec_cntlz (ssa);
vector unsigned short usa = {81, 72, 63, 54, 45, 36, 27, 18};
vector unsigned short usb = vec_cntlz (usa);
vector signed char sca = {-4, 3, -9, 15, -31, 31, 0, 0,
1, 117, -36, 99, 98, 97, 96, 95};
vector signed char scb = vec_cntlz (sca);
vector unsigned char cb = vec_cntlz (ca);
vector double dd = vec_xl (0, &y);
vec_xst (dd, 0, &z);
vector double de = vec_round (dd);
vector double df = vec_splat (de, 0);
vector double dg = vec_splat (de, 1);
vector long long l3 = vec_splat (l2, 0);
vector long long l4 = vec_splat (l2, 1);
vector unsigned long long u3 = vec_splat (u2, 0);
vector unsigned long long u4 = vec_splat (u2, 1);
vector bool long long l5 = vec_splat (ld, 0);
vector bool long long l6 = vec_splat (ld, 1);
vector long long l7 = vec_div (l3, l4);
vector unsigned long long u5 = vec_div (u3, u4);
vector long long l8 = vec_mul (l3, l4);
vector unsigned long long u6 = vec_mul (u3, u4);
vector double dh = vec_ctf (la, -2);
vector double di = vec_ctf (ua, 2);
vector long long l9 = vec_cts (dh, -2);
vector unsigned long long u7 = vec_ctu (di, 2);
return 0;
}
void test1() {
// CHECK-LABEL: define void @test1
res_vd = vec_add(vd, vd);
// CHECK: fadd <2 x double>
res_vd = vec_and(vbll, vd);
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
res_vd = vec_and(vd, vbll);
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
res_vd = vec_and(vd, vd);
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
res_vd = vec_andc(vbll, vd);
// CHECK: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
res_vd = vec_andc(vd, vbll);
// CHECK: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
res_vd = vec_andc(vd, vd);
// CHECK: bitcast <2 x double> %{{[0-9]*}} to <2 x i64>
// CHECK: xor <2 x i64> %{{[0-9]*}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]*}} to <2 x double>
dummy();
// CHECK: call void @dummy()
res_vd = vec_ceil(vd);
// CHECK: call <2 x double> @llvm.ceil.v2f64(<2 x double> %{{[0-9]*}})
res_vf = vec_ceil(vf);
// CHECK: call <4 x float> @llvm.ceil.v4f32(<4 x float> %{{[0-9]*}})
res_vbll = vec_cmpeq(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpeqdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmpeq(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpeqsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmpge(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmpge(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmpgt(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmpgt(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmple(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgedp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmple(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgesp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
res_vbll = vec_cmplt(vd, vd);
// CHECK: call <2 x i64> @llvm.ppc.vsx.xvcmpgtdp(<2 x double> %{{[0-9]*}}, <2 x double> %{{[0-9]*}})
res_vbi = vec_cmplt(vf, vf);
// CHECK: call <4 x i32> @llvm.ppc.vsx.xvcmpgtsp(<4 x float> %{{[0-9]*}}, <4 x float> %{{[0-9]*}})
/* vec_div */
res_vf = vec_div(vf, vf);
// CHECK: @llvm.ppc.vsx.xvdivsp
res_vd = vec_div(vd, vd);
// CHECK: @llvm.ppc.vsx.xvdivdp
/* vec_max */
res_vf = vec_max(vf, vf);
// CHECK: @llvm.ppc.vsx.xvmaxsp
res_vd = vec_max(vd, vd);
// CHECK: @llvm.ppc.vsx.xvmaxdp
res_vf = vec_vmaxfp(vf, vf);
// CHECK: @llvm.ppc.vsx.xvmaxsp
/* vec_min */
res_vf = vec_min(vf, vf);
// CHECK: @llvm.ppc.vsx.xvminsp
res_vd = vec_min(vd, vd);
// CHECK: @llvm.ppc.vsx.xvmindp
res_vf = vec_vminfp(vf, vf);
// CHECK: @llvm.ppc.vsx.xvminsp
res_d = __builtin_vsx_xsmaxdp(d, d);
// CHECK: @llvm.ppc.vsx.xsmaxdp
res_d = __builtin_vsx_xsmindp(d, d);
// CHECK: @llvm.ppc.vsx.xsmindp
/* vec_perm */
res_vsll = vec_perm(vsll, vsll, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vull = vec_perm(vull, vull, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vd = vec_perm(vd, vd, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vsll = vec_vperm(vsll, vsll, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vull = vec_vperm(vull, vull, vuc);
// CHECK: @llvm.ppc.altivec.vperm
res_vd = vec_vperm(vd, vd, vuc);
// CHECK: @llvm.ppc.altivec.vperm
/* vec_vsx_ld */
res_vsi = vec_vsx_ld(0, &vsi);
// CHECK: @llvm.ppc.vsx.lxvw4x
res_vui = vec_vsx_ld(0, &vui);
// CHECK: @llvm.ppc.vsx.lxvw4x
res_vf = vec_vsx_ld (0, &vf);
// CHECK: @llvm.ppc.vsx.lxvw4x
res_vsll = vec_vsx_ld(0, &vsll);
// CHECK: @llvm.ppc.vsx.lxvd2x
res_vull = vec_vsx_ld(0, &vull);
// CHECK: @llvm.ppc.vsx.lxvd2x
res_vd = vec_vsx_ld(0, &vd);
// CHECK: @llvm.ppc.vsx.lxvd2x
/* vec_vsx_st */
vec_vsx_st(vsi, 0, &res_vsi);
// CHECK: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vui, 0, &res_vui);
// CHECK: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vf, 0, &res_vf);
// CHECK: @llvm.ppc.vsx.stxvw4x
vec_vsx_st(vsll, 0, &res_vsll);
// CHECK: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vull, 0, &res_vull);
// CHECK: @llvm.ppc.vsx.stxvd2x
vec_vsx_st(vd, 0, &res_vd);
// CHECK: @llvm.ppc.vsx.stxvd2x
/* vec_and */
res_vsll = vec_and(vsll, vsll);
// CHECK: and <2 x i64>
res_vsll = vec_and(vbll, vsll);
// CHECK: and <2 x i64>
res_vsll = vec_and(vsll, vbll);
// CHECK: and <2 x i64>
res_vull = vec_and(vull, vull);
// CHECK: and <2 x i64>
res_vull = vec_and(vbll, vull);
// CHECK: and <2 x i64>
res_vull = vec_and(vull, vbll);
// CHECK: and <2 x i64>
res_vbll = vec_and(vbll, vbll);
// CHECK: and <2 x i64>
/* vec_vand */
res_vsll = vec_vand(vsll, vsll);
// CHECK: and <2 x i64>
res_vsll = vec_vand(vbll, vsll);
// CHECK: and <2 x i64>
res_vsll = vec_vand(vsll, vbll);
// CHECK: and <2 x i64>
res_vull = vec_vand(vull, vull);
// CHECK: and <2 x i64>
res_vull = vec_vand(vbll, vull);
// CHECK: and <2 x i64>
res_vull = vec_vand(vull, vbll);
// CHECK: and <2 x i64>
res_vbll = vec_vand(vbll, vbll);
// CHECK: and <2 x i64>
/* vec_andc */
res_vsll = vec_andc(vsll, vsll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vsll = vec_andc(vbll, vsll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vsll = vec_andc(vsll, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vull = vec_andc(vull, vull);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vull = vec_andc(vbll, vull);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vull = vec_andc(vull, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vbll = vec_andc(vbll, vbll);
// CHECK: xor <2 x i64>
// CHECK: and <2 x i64>
res_vf = vec_floor(vf);
// CHECK: call <4 x float> @llvm.floor.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_floor(vd);
// CHECK: call <2 x double> @llvm.floor.v2f64(<2 x double> %{{[0-9]+}})
res_vf = vec_madd(vf, vf, vf);
// CHECK: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
res_vd = vec_madd(vd, vd, vd);
// CHECK: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
res_vf = vec_msub(vf, vf, vf);
// CHECK: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
res_vd = vec_msub(vd, vd, vd);
// CHECK: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
res_vf = vec_mul(vf, vf);
// CHECK: fmul <4 x float> %{{[0-9]+}}, %{{[0-9]+}}
res_vd = vec_mul(vd, vd);
// CHECK: fmul <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
res_vf = vec_nearbyint(vf);
// CHECK: call <4 x float> @llvm.round.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_nearbyint(vd);
// CHECK: call <2 x double> @llvm.round.v2f64(<2 x double> %{{[0-9]+}})
res_vf = vec_nmadd(vf, vf, vf);
// CHECK: [[FM:[0-9]+]] = call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}})
// CHECK-NEXT: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %[[FM]]
res_vd = vec_nmadd(vd, vd, vd);
// CHECK: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}})
// CHECK-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
res_vf = vec_nmsub(vf, vf, vf);
// CHECK: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: call <4 x float> @llvm.fma.v4f32(<4 x float> %{{[0-9]+}}, <4 x float> %{{[0-9]+}}, <4 x float>
// CHECK: fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %{{[0-9]+}}
res_vd = vec_nmsub(vd, vd, vd);
// CHECK: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %{{[0-9]+}}
// CHECK-NEXT: [[FM:[0-9]+]] = call <2 x double> @llvm.fma.v2f64(<2 x double> %{{[0-9]+}}, <2 x double> %{{[0-9]+}}, <2 x double>
// CHECK-NEXT: fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %[[FM]]
/* vec_nor */
res_vsll = vec_nor(vsll, vsll);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
res_vull = vec_nor(vull, vull);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
res_vull = vec_nor(vbll, vbll);
// CHECK: or <2 x i64>
// CHECK: xor <2 x i64>
res_vd = vec_nor(vd, vd);
// CHECK: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: [[OR:%.+]] = or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK-NEXT: xor <2 x i64> [[OR]], <i64 -1, i64 -1>
/* vec_or */
res_vsll = vec_or(vsll, vsll);
// CHECK: or <2 x i64>
res_vsll = vec_or(vbll, vsll);
// CHECK: or <2 x i64>
res_vsll = vec_or(vsll, vbll);
// CHECK: or <2 x i64>
res_vull = vec_or(vull, vull);
// CHECK: or <2 x i64>
res_vull = vec_or(vbll, vull);
// CHECK: or <2 x i64>
res_vull = vec_or(vull, vbll);
// CHECK: or <2 x i64>
res_vbll = vec_or(vbll, vbll);
// CHECK: or <2 x i64>
res_vd = vec_or(vd, vd);
// CHECK: bitcast <2 x double> %{{[0-9]+}} to <2 x i64>
// CHECK: or <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
res_vf = vec_rint(vf);
// CHECK: call <4 x float> @llvm.nearbyint.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_rint(vd);
// CHECK: call <2 x double> @llvm.nearbyint.v2f64(<2 x double> %{{[0-9]+}})
res_vf = vec_rsqrte(vf);
// CHECK: call <4 x float> @llvm.ppc.vsx.xvrsqrtesp(<4 x float> %{{[0-9]+}})
res_vd = vec_rsqrte(vd);
// CHECK: call <2 x double> @llvm.ppc.vsx.xvrsqrtedp(<2 x double> %{{[0-9]+}})
dummy();
// CHECK: call void @dummy()
res_vf = vec_sel(vd, vd, vbll);
// CHECK: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64> %{{[0-9]+}},
// CHECK: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: or <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
dummy();
// CHECK: call void @dummy()
res_vd = vec_sel(vd, vd, vull);
// CHECK: xor <2 x i64> %{{[0-9]+}}, <i64 -1, i64 -1>
// CHECK: and <2 x i64> %{{[0-9]+}},
// CHECK: and <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: or <2 x i64>
// CHECK: bitcast <2 x i64> %{{[0-9]+}} to <2 x double>
res_vf = vec_sqrt(vf);
// CHECK: call <4 x float> @llvm.sqrt.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_sqrt(vd);
// CHECK: call <2 x double> @llvm.sqrt.v2f64(<2 x double> %{{[0-9]+}})
res_vd = vec_sub(vd, vd);
// CHECK: fsub <2 x double> %{{[0-9]+}}, %{{[0-9]+}}
res_vf = vec_trunc(vf);
// CHECK: call <4 x float> @llvm.trunc.v4f32(<4 x float> %{{[0-9]+}})
res_vd = vec_trunc(vd);
// CHECK: call <2 x double> @llvm.trunc.v2f64(<2 x double> %{{[0-9]+}})
/* vec_vor */
res_vsll = vec_vor(vsll, vsll);
// CHECK: or <2 x i64>
res_vsll = vec_vor(vbll, vsll);
// CHECK: or <2 x i64>
res_vsll = vec_vor(vsll, vbll);
// CHECK: or <2 x i64>
res_vull = vec_vor(vull, vull);
// CHECK: or <2 x i64>
res_vull = vec_vor(vbll, vull);
// CHECK: or <2 x i64>
res_vull = vec_vor(vull, vbll);
// CHECK: or <2 x i64>
res_vbll = vec_vor(vbll, vbll);
// CHECK: or <2 x i64>
/* vec_xor */
res_vsll = vec_xor(vsll, vsll);
// CHECK: xor <2 x i64>
res_vsll = vec_xor(vbll, vsll);
// CHECK: xor <2 x i64>
res_vsll = vec_xor(vsll, vbll);
// CHECK: xor <2 x i64>
res_vull = vec_xor(vull, vull);
// CHECK: xor <2 x i64>
res_vull = vec_xor(vbll, vull);
// CHECK: xor <2 x i64>
res_vull = vec_xor(vull, vbll);
// CHECK: xor <2 x i64>
res_vbll = vec_xor(vbll, vbll);
// CHECK: xor <2 x i64>
dummy();
// CHECK: call void @dummy()
res_vd = vec_xor(vd, vd);
// CHECK: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[X1]] to <2 x double>
dummy();
// CHECK: call void @dummy()
res_vd = vec_xor(vd, vbll);
// CHECK: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[X1]] to <2 x double>
dummy();
// CHECK: call void @dummy()
res_vd = vec_xor(vbll, vd);
// CHECK: [[X1:%.+]] = xor <2 x i64> %{{[0-9]+}}, %{{[0-9]+}}
// CHECK: bitcast <2 x i64> [[X1]] to <2 x double>
/* vec_vxor */
res_vsll = vec_vxor(vsll, vsll);
// CHECK: xor <2 x i64>
res_vsll = vec_vxor(vbll, vsll);
// CHECK: xor <2 x i64>
res_vsll = vec_vxor(vsll, vbll);
// CHECK: xor <2 x i64>
res_vull = vec_vxor(vull, vull);
// CHECK: xor <2 x i64>
res_vull = vec_vxor(vbll, vull);
// CHECK: xor <2 x i64>
res_vull = vec_vxor(vull, vbll);
// CHECK: xor <2 x i64>
res_vbll = vec_vxor(vbll, vbll);
// CHECK: xor <2 x i64>
}
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
}
}
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;
}
static void ProjectDlightTexture_altivec( void ) {
int i, l;
vec_t origin0, origin1, origin2;
float texCoords0, texCoords1;
vector float floatColorVec0, floatColorVec1;
vector float modulateVec, colorVec, zero;
vector short colorShort;
vector signed int colorInt;
vector unsigned char floatColorVecPerm, modulatePerm, colorChar;
vector unsigned char vSel = VECCONST_UINT8(0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff,
0x00, 0x00, 0x00, 0xff);
float *texCoords;
byte *colors;
byte clipBits[SHADER_MAX_VERTEXES];
float texCoordsArray[SHADER_MAX_VERTEXES][2];
byte colorArray[SHADER_MAX_VERTEXES][4];
unsigned hitIndexes[SHADER_MAX_INDEXES];
int numIndexes;
float scale;
float radius;
vec3_t floatColor;
float modulate = 0.0f;
if ( !backEnd.refdef.num_dlights ) {
return;
}
// There has to be a better way to do this so that floatColor
// and/or modulate are already 16-byte aligned.
floatColorVecPerm = vec_lvsl(0,(float *)floatColor);
modulatePerm = vec_lvsl(0,(float *)&modulate);
modulatePerm = (vector unsigned char)vec_splat((vector unsigned int)modulatePerm,0);
zero = (vector float)vec_splat_s8(0);
for ( l = 0 ; l < backEnd.refdef.num_dlights ; l++ ) {
dlight_t *dl;
if ( !( tess.dlightBits & ( 1 << l ) ) ) {
continue; // this surface definately doesn't have any of this light
}
texCoords = texCoordsArray[0];
colors = colorArray[0];
dl = &backEnd.refdef.dlights[l];
origin0 = dl->transformed[0];
origin1 = dl->transformed[1];
origin2 = dl->transformed[2];
radius = dl->radius;
scale = 1.0f / radius;
if(r_greyscale->integer)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = floatColor[1] = floatColor[2] = luminance;
}
else if(r_greyscale->value)
{
float luminance;
luminance = LUMA(dl->color[0], dl->color[1], dl->color[2]) * 255.0f;
floatColor[0] = LERP(dl->color[0] * 255.0f, luminance, r_greyscale->value);
floatColor[1] = LERP(dl->color[1] * 255.0f, luminance, r_greyscale->value);
floatColor[2] = LERP(dl->color[2] * 255.0f, luminance, r_greyscale->value);
}
else
{
floatColor[0] = dl->color[0] * 255.0f;
floatColor[1] = dl->color[1] * 255.0f;
floatColor[2] = dl->color[2] * 255.0f;
}
floatColorVec0 = vec_ld(0, floatColor);
floatColorVec1 = vec_ld(11, floatColor);
floatColorVec0 = vec_perm(floatColorVec0,floatColorVec0,floatColorVecPerm);
for ( i = 0 ; i < tess.numVertexes ; i++, texCoords += 2, colors += 4 ) {
int clip = 0;
vec_t dist0, dist1, dist2;
dist0 = origin0 - tess.xyz[i][0];
dist1 = origin1 - tess.xyz[i][1];
dist2 = origin2 - tess.xyz[i][2];
backEnd.pc.c_dlightVertexes++;
texCoords0 = 0.5f + dist0 * scale;
texCoords1 = 0.5f + dist1 * scale;
if( !r_dlightBacks->integer &&
// dist . tess.normal[i]
( dist0 * tess.normal[i][0] +
dist1 * tess.normal[i][1] +
dist2 * tess.normal[i][2] ) < 0.0f ) {
clip = 63;
} else {
if ( texCoords0 < 0.0f ) {
clip |= 1;
} else if ( texCoords0 > 1.0f ) {
clip |= 2;
}
if ( texCoords1 < 0.0f ) {
clip |= 4;
} else if ( texCoords1 > 1.0f ) {
clip |= 8;
}
texCoords[0] = texCoords0;
texCoords[1] = texCoords1;
// modulate the strength based on the height and color
if ( dist2 > radius ) {
clip |= 16;
modulate = 0.0f;
} else if ( dist2 < -radius ) {
clip |= 32;
modulate = 0.0f;
} else {
dist2 = Q_fabs(dist2);
if ( dist2 < radius * 0.5f ) {
modulate = 1.0f;
} else {
modulate = 2.0f * (radius - dist2) * scale;
}
}
}
clipBits[i] = clip;
modulateVec = vec_ld(0,(float *)&modulate);
modulateVec = vec_perm(modulateVec,modulateVec,modulatePerm);
colorVec = vec_madd(floatColorVec0,modulateVec,zero);
colorInt = vec_cts(colorVec,0); // RGBx
colorShort = vec_pack(colorInt,colorInt); // RGBxRGBx
colorChar = vec_packsu(colorShort,colorShort); // RGBxRGBxRGBxRGBx
colorChar = vec_sel(colorChar,vSel,vSel); // RGBARGBARGBARGBA replace alpha with 255
vec_ste((vector unsigned int)colorChar,0,(unsigned int *)colors); // store color
}
// build a list of triangles that need light
numIndexes = 0;
for ( i = 0 ; i < tess.numIndexes ; i += 3 ) {
int a, b, c;
a = tess.indexes[i];
b = tess.indexes[i+1];
c = tess.indexes[i+2];
if ( clipBits[a] & clipBits[b] & clipBits[c] ) {
continue; // not lighted
}
hitIndexes[numIndexes] = a;
hitIndexes[numIndexes+1] = b;
hitIndexes[numIndexes+2] = c;
numIndexes += 3;
}
if ( !numIndexes ) {
continue;
}
qglEnableClientState( GL_TEXTURE_COORD_ARRAY );
qglTexCoordPointer( 2, GL_FLOAT, 0, texCoordsArray[0] );
qglEnableClientState( GL_COLOR_ARRAY );
qglColorPointer( 4, GL_UNSIGNED_BYTE, 0, colorArray );
GL_Bind( tr.dlightImage );
// include GLS_DEPTHFUNC_EQUAL so alpha tested surfaces don't add light
// where they aren't rendered
if ( dl->additive ) {
GL_State( GLS_SRCBLEND_ONE | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
else {
GL_State( GLS_SRCBLEND_DST_COLOR | GLS_DSTBLEND_ONE | GLS_DEPTHFUNC_EQUAL );
}
R_DrawElements( numIndexes, hitIndexes );
backEnd.pc.c_totalIndexes += numIndexes;
backEnd.pc.c_dlightIndexes += numIndexes;
}
}
static inline vector unsigned int Ch(const vector unsigned int b, const vector unsigned int c, const vector unsigned int d) {
return vec_sel(d,c,b);
}
