diff options
| author | Roland Scheidegger <sroland@vmware.com> | 2013-07-17 18:13:41 +0200 |
|---|---|---|
| committer | Roland Scheidegger <sroland@vmware.com> | 2013-07-17 18:16:34 +0200 |
| commit | 7fd30a8621ccb10b1b6df54ca2afb73bb8e959e8 (patch) | |
| tree | 91a27ec799ab08354a795373af3e9aab40cea7ac | |
| parent | f0f9fb59c326a7b911bed8491edb0f5fd5966360 (diff) | |
gallivm: (trivial) simplify lp_build_cos/lp_build_sin a tiny bit
Use "or" instead of "add" (this is a classic select sequence, which at
least newer llvm versions can actually recognize (3.2+?), and the "add"
might prevent that - and we really don't want an add instead of an or with
avx if it isn't recognized (even without avx logic ops might be cheaper)).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
| -rw-r--r-- | src/gallium/auxiliary/gallivm/lp_bld_arit.c | 13 |
1 files changed, 6 insertions, 7 deletions
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c index e7955aa6bf..74b4e9fce2 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c @@ -2602,15 +2602,14 @@ lp_build_sin(struct lp_build_context *bld, * xmm3 = poly_mask; * y2 = _mm_and_ps(xmm3, y2); //, xmm3); * y = _mm_andnot_ps(xmm3, y); - * y = _mm_add_ps(y,y2); + * y = _mm_or_ps(y,y2); */ LLVMValueRef y2_i = LLVMBuildBitCast(b, y2_9, bld->int_vec_type, "y2_i"); LLVMValueRef y_i = LLVMBuildBitCast(b, y_10, bld->int_vec_type, "y_i"); LLVMValueRef y2_and = LLVMBuildAnd(b, y2_i, poly_mask, "y2_and"); - LLVMValueRef inv = lp_build_const_int_vec(gallivm, bld->type, ~0); - LLVMValueRef poly_mask_inv = LLVMBuildXor(b, poly_mask, inv, "poly_mask_inv"); + LLVMValueRef poly_mask_inv = LLVMBuildNot(b, poly_mask, "poly_mask_inv"); LLVMValueRef y_and = LLVMBuildAnd(b, y_i, poly_mask_inv, "y_and"); - LLVMValueRef y_combine = LLVMBuildAdd(b, y_and, y2_and, "y_combine"); + LLVMValueRef y_combine = LLVMBuildOr(b, y_and, y2_and, "y_combine"); /* * update the sign @@ -2820,14 +2819,14 @@ lp_build_cos(struct lp_build_context *bld, * xmm3 = poly_mask; * y2 = _mm_and_ps(xmm3, y2); //, xmm3); * y = _mm_andnot_ps(xmm3, y); - * y = _mm_add_ps(y,y2); + * y = _mm_or_ps(y,y2); */ LLVMValueRef y2_i = LLVMBuildBitCast(b, y2_9, bld->int_vec_type, "y2_i"); LLVMValueRef y_i = LLVMBuildBitCast(b, y_10, bld->int_vec_type, "y_i"); LLVMValueRef y2_and = LLVMBuildAnd(b, y2_i, poly_mask, "y2_and"); - LLVMValueRef poly_mask_inv = LLVMBuildXor(b, poly_mask, inv, "poly_mask_inv"); + LLVMValueRef poly_mask_inv = LLVMBuildNot(b, poly_mask, "poly_mask_inv"); LLVMValueRef y_and = LLVMBuildAnd(b, y_i, poly_mask_inv, "y_and"); - LLVMValueRef y_combine = LLVMBuildAdd(b, y_and, y2_and, "y_combine"); + LLVMValueRef y_combine = LLVMBuildOr(b, y_and, y2_and, "y_combine"); /* * update the sign |