Instruction *tex, *add;
Value *zero = bld.loadImm(bld.getSSA(), 0);
int l, c;
- const int dim = i->tex.target.getDim();
+ const int dim = i->tex.target.getDim() + i->tex.target.isCube();
const int array = i->tex.target.isArray();
i->op = OP_TEX; // no need to clone dPdx/dPdy later
tmp = bld.getScratch();
for (l = 0; l < 4; ++l) {
+ Value *src[3], *val;
// mov coordinates from lane l to all lanes
bld.mkOp(OP_QUADON, TYPE_NONE, NULL);
for (c = 0; c < dim; ++c) {
add->lanes = 1; /* abused for .ndv */
}
+ // normalize cube coordinates if necessary
+ if (i->tex.target.isCube()) {
+ for (c = 0; c < 3; ++c)
+ src[c] = bld.mkOp1v(OP_ABS, TYPE_F32, bld.getSSA(), crd[c]);
+ val = bld.getScratch();
+ bld.mkOp2(OP_MAX, TYPE_F32, val, src[0], src[1]);
+ bld.mkOp2(OP_MAX, TYPE_F32, val, src[2], val);
+ bld.mkOp1(OP_RCP, TYPE_F32, val, val);
+ for (c = 0; c < 3; ++c)
+ src[c] = bld.mkOp2v(OP_MUL, TYPE_F32, bld.getSSA(), crd[c], val);
+ } else {
+ for (c = 0; c < dim; ++c)
+ src[c] = crd[c];
+ }
+
// texture
bld.insert(tex = cloneForward(func, i));
for (c = 0; c < dim; ++c)
- tex->setSrc(c + array, crd[c]);
+ tex->setSrc(c + array, src[c]);
bld.mkOp(OP_QUADPOP, TYPE_NONE, NULL);
// save results