gallivm: Use lp_build_alloca instead of LLVMBuildAlloca on the loop limiter.

To ensure that the alloca is at the top of the function body, otherwise
LLVM will not eliminate them, causing stack misalignment on 32bits.

Reviewed-by: James Benton <jbenton@vmware.com>
This commit is contained in:
José Fonseca 2012-04-25 16:30:06 +01:00
parent 9f45093433
commit 914244e59d

View file

@ -79,10 +79,7 @@ static void lp_exec_mask_init(struct lp_exec_mask *mask, struct lp_build_context
mask->exec_mask = mask->ret_mask = mask->break_mask = mask->cont_mask = mask->cond_mask =
LLVMConstAllOnes(mask->int_vec_type);
mask->loop_limiter = LLVMBuildAlloca(
builder,
int_type,
"looplimiter");
mask->loop_limiter = lp_build_alloca(bld->gallivm, int_type, "looplimiter");
LLVMBuildStore(
builder,