aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Added masked load optimization pass.

Browse Source 
This pass handles the "all on" and "all off" mask cases appropriately.

Also renamed load_masked stuff in built-ins to masked_load for consistency with
masked_store.
This commit is contained in:
Matt Pharr
2012-01-04 11:51:26 -08:00
parent 75f18c7c66
commit 562d61caff
9 changed files with 152 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
builtins/target-avx-x2.ll
View File

@@ -385,13 +385,13 @@ load_and_broadcast(16, i32, 32)
load_and_broadcast(16, i64, 64)
; no masked load instruction for i8 and i16 types??
load_masked(16, i8, 8, 1)
load_masked(16, i16, 16, 2)
masked_load(16, i8, 8, 1)
masked_load(16, i16, 16, 2)
declare <8 x float> @llvm.x86.avx.maskload.ps.256(i8 *, <8 x float> %mask)
declare <4 x double> @llvm.x86.avx.maskload.pd.256(i8 *, <4 x double> %mask)
define <16 x i32> @__load_masked_32(i8 *, <16 x i32> %mask) nounwind alwaysinline {
define <16 x i32> @__masked_load_32(i8 *, <16 x i32> %mask) nounwind alwaysinline {
%floatmask = bitcast <16 x i32> %mask to <16 x float>
%mask0 = shufflevector <16 x float> %floatmask, <16 x float> undef,
<8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -409,7 +409,7 @@ define <16 x i32> @__load_masked_32(i8 *, <16 x i32> %mask) nounwind alwaysinlin
}
define <16 x i64> @__load_masked_64(i8 *, <16 x i32> %mask) nounwind alwaysinline {
define <16 x i64> @__masked_load_64(i8 *, <16 x i32> %mask) nounwind alwaysinline {
; double up masks, bitcast to doubles
%mask0 = shufflevector <16 x i32> %mask, <16 x i32> undef,
<8 x i32> <i32 0, i32 0, i32 1, i32 1, i32 2, i32 2, i32 3, i32 3>