int
test (v4si *x, v4si *y)
{
v4si m = { 2, 3, 6, 5 };
v4si z = __builtin_shuffle (*x, *y, m);
return z[2];
}
/* { dg-do compile { target { ! { ia32 } } } } */
/* { dg-options "-mno-sse -Wvector-operation-performance" } */
#define vector(elcount, type) \
__attribute__((vector_size((elcount)*sizeof(type)))) type
int main (int argc, char *argv[])
{
vector (4, int) v0 = {argc, 1, 15, 38};
vector (4, int) v1 = {-4, argc, 2, 11};
vector (4, int) res[] =
{
v0 + v1, /* { dg-warning "expanded piecewise" } */
v0 - v1, /* { dg-warning "expanded piecewise" } */
v0 > v1, /* { dg-warning "expanded piecewise" } */
v0 & v1, /* { dg-warning "expanded in parallel" } */
__builtin_shuffle (v0, v1), /* { dg-warning "expanded piecewise" } */
__builtin_shuffle (v0, v1, v1) /* { dg-warning "expanded piecewise" } */
};
return res[argc][argc];
}
v4si vs (v4si a, v4si b)
{
return __builtin_shuffle (a, b, (v4si) {0, 4, 1, 5});
}
static void Scaler_DrawLine(v4hi *dst, Uint32 *src, int ww, int repeat, int pitch)
{
int xx;
int yy;
int yrep2;
int yrep3;
int blank;
register v4hi *b2p;
register v4hi r1, r2;
v4hi *d0;
register v4hi *b;
int pitch2;
register int ip;
v4hi r3v[5 * 80];
#if AG_BIG_ENDIAN != 1
const v4ui bb = {0xff000000, 0xff000000, 0xff000000, 0xff000000};
const v4ui order3 = (v4ui){3, 3, 4, 4};
#else
const v4ui bb = {0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff};
const v4ui order3 = (v4ui){3, 3, 4, 4};
#endif
if(repeat <= 0) return;
b = (v4hi *)src;
b2p = dst;
pitch2 = pitch / sizeof(v4hi);
// _prefetch_data_read_l2((void *)src, sizeof(Uint32) * ww);
_prefetch_data_write_l1((void *)r3v, sizeof(r3v));
if(__builtin_expect(((bFullScan) || (repeat < 2)), 1)) {
ip = 0;
for(xx = 0; xx < ww; xx += 8) {
b2p = dst;
r1 = b[0];
r2 = b[1];
// 76543210 -> 77666554443322211000
r3v[ip + 0].uv = __builtin_ia32_pshufd(r1.uv, 0b01000000);
r3v[ip + 1].uv = __builtin_ia32_pshufd(r1.uv, 0b10101001);
r3v[ip + 2] = (v4hi)__builtin_shuffle(r1.uv, r2.uv, order3);
r3v[ip + 3].uv = __builtin_ia32_pshufd(r2.uv, 0b10010100);
r3v[ip + 4].uv = __builtin_ia32_pshufd(r2.uv, 0b11111010);
ip += 5;
b += 2;
}
for(yy = 0; yy < repeat; yy++) {
// _prefetch_data_write_l2((void *)b2p, sizeof(v4hi) * 5);
memcpy((void *)b2p, (void *)r3v, sizeof(v4hi) * ip);
b2p = b2p + pitch2;
}
} else {
ip = 0;
for(xx = 0; xx < ww; xx += 8) {
yy = 0;
// b2p = dst;
r1 = b[0];
r2 = b[1];
// 76543210 -> 77666554443322211000
r3v[ip + 0].uv = __builtin_ia32_pshufd(r1.uv, 0b01000000);
r3v[ip + 1].uv = __builtin_ia32_pshufd(r1.uv, 0b10101001);
r3v[ip + 2] = (v4hi)__builtin_shuffle(r1.uv, r2.uv, order3);
r3v[ip + 3].uv = __builtin_ia32_pshufd(r2.uv, 0b10010100);
r3v[ip + 4].uv = __builtin_ia32_pshufd(r2.uv, 0b11111010);
ip += 5;
b += 2;
}
b2p = dst;
for(yy = 0; yy < repeat - 1; yy++) {
// _prefetch_data_write_l2((void *)b2p, sizeof(v4hi) * 5);
memcpy((void *)b2p, (void *)r3v, ip * sizeof(v4hi));
b2p = b2p + pitch2;
}
// _prefetch_data_write_l2((void *)b2p, sizeof(v4hi) * 5);
for(xx = 0; xx < ip; xx++) b2p[xx].uv = bb;
}
}
long f (long d, long e)
{
vec x = { d, e };
vec m = { 1, 0 };
return __builtin_shuffle (x, m) [1];
}
v2df foo1 (v2df x, v2df y)
{
v2df tem0 = x - y;
v2df tem1 = x + y;
return __builtin_shuffle (tem0, tem1, (v2di) { 0, 3 });
}
v2df foo4 (v2df x, v2df y)
{
v2df tem0 = y + x;
v2df tem1 = x - y;
return __builtin_shuffle (tem0, tem1, (v2di) { 2, 1 });
}
