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

Explicitly call the PBLENDVB intrinsic for i8 blending with sse4-8.

Browse Source 
This is slightly cleaner than trunc-ing the i8 mask to i1 and using
a vector select.  (And is probably more safe in terms of good code.)
This commit is contained in:
Matt Pharr
2013-07-24 15:10:08 -07:00
parent bba84f247c
commit 2d063925a1
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
builtins/target-sse4-8.ll
View File

@@ -398,11 +398,13 @@ define void @__masked_store_blend_i16(<16 x i16>* nocapture, <16 x i16>,
ret void
}
declare <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone
define void @__masked_store_blend_i8(<16 x i8>* nocapture, <16 x i8>,
<16 x MASK> %mask) nounwind alwaysinline {
%mask_as_i1 = trunc <16 x MASK> %mask to <16 x i1>
%old = load <16 x i8>* %0, align 4
%blend = select <16 x i1> %mask_as_i1, <16 x i8> %1, <16 x i8> %old
%blend = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %old, <16 x i8> %1,
<16 x i8> %mask)
store <16 x i8> %blend, <16 x i8>* %0, align 4
ret void
}