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

Add comment re popcnt on SSE2 target.

Browse Source
This commit is contained in:
Matt Pharr
2011-07-06 07:37:35 +01:00
parent 92106e866e
commit 5cc750ecee
1 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
stdlib-sse2.ll
View File

@@ -244,6 +244,14 @@ define internal i32 @__max_uniform_uint32(i32, i32) nounwind readonly alwaysinli
; FIXME: this is very inefficient, loops over all 32 bits... ; FIXME: this is very inefficient, loops over all 32 bits...
; we could use the LLVM intrinsic declare i32 @llvm.ctpop.i32(i32),
; although that currently ends up generating a POPCNT instruction even
; if we give --target=sse2 on the command line. We probably need to
; pipe through the 'sse2' request to LLVM via the 'features' string
; at codegen time... (If e.g. --cpu=penryn is also passed along, then
; it does generate non-POPCNT code and in particular better code than
; the below does.)
define internal i32 @__popcnt(i32) nounwind readonly alwaysinline { define internal i32 @__popcnt(i32) nounwind readonly alwaysinline {
entry: entry:
br label %loop br label %loop