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

Represent MOVMSK'ed masks with int64s rather than int32s.

Browse Source 
This allows us to scale up to 64-wide execution.
This commit is contained in:
Matt Pharr
2012-05-25 11:48:08 -07:00
parent 38cea6dc71
commit 90db01d038
20 changed files with 137 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
builtins/target-generic-common.ll
View File

@@ -1,4 +1,4 @@
;; Copyright (c) 2010-2011, Intel Corporation
;; Copyright (c) 2010-2012, Intel Corporation
;; All rights reserved.
;;
;; Redistribution and use in source and binary forms, with or without
@@ -201,7 +201,7 @@ declare <WIDTH x float> @__svml_pow(<WIDTH x float>, <WIDTH x float>)
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; reductions
declare i32 @__movmsk(<WIDTH x i1>) nounwind readnone
declare i64 @__movmsk(<WIDTH x i1>) nounwind readnone
declare float @__reduce_add_float(<WIDTH x float>) nounwind readnone
declare float @__reduce_min_float(<WIDTH x float>) nounwind readnone