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

Added the following mask tests: __any(), __all(), __none() for all supported targets.

Browse Source 
This allows for more efficient code generation of KNC.
This commit is contained in:
Jean-Luc Duprat
2012-09-14 11:06:18 -07:00
parent 4ecdbe4bd9
commit f0b0618484
15 changed files with 355 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
builtins/target-sse4.ll
View File

@@ -278,6 +278,27 @@ define i64 @__movmsk(<4 x i32>) nounwind readnone alwaysinline {
ret i64 %v64
}
define i1 @__any(<4 x i32>) nounwind readnone alwaysinline {
%floatmask = bitcast <4 x i32> %0 to <4 x float>
%v = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %floatmask) nounwind readnone
%cmp = icmp ne i32 %v, 0
ret i1 %cmp
}
define i1 @__all(<4 x i32>) nounwind readnone alwaysinline {
%floatmask = bitcast <4 x i32> %0 to <4 x float>
%v = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %floatmask) nounwind readnone
%cmp = icmp eq i32 %v, 15
ret i1 %cmp
}
define i1 @__none(<4 x i32>) nounwind readnone alwaysinline {
%floatmask = bitcast <4 x i32> %0 to <4 x float>
%v = call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %floatmask) nounwind readnone
%cmp = icmp eq i32 %v, 0
ret i1 %cmp
}
declare <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float>, <4 x float>) nounwind readnone
define float @__reduce_add_float(<4 x float>) nounwind readonly alwaysinline {