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

added common_gpu makefile helper

Browse Source
This commit is contained in:
Evghenii
2014-01-05 10:00:44 +01:00
parent 034e6b9a0b
commit fd429e4fda
14 changed files with 114 additions and 1578 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
builtins/util_ptx.m4
View File

@@ -2924,11 +2924,11 @@ if.then: ; preds = %entry
if.end: ; preds = %if.then, %entry
%ptr.0 = phi i64 [ %phitmp, %if.then ], [ undef, %entry ]
%val.sroa.0.0.extract.trunc = trunc i64 %ptr.0 to i32
%call3 = tail call i32 @__shfl_i32(i32 %val.sroa.0.0.extract.trunc, i32 0)
%call3 = tail call i32 @__shfl_i32_nvptx(i32 %val.sroa.0.0.extract.trunc, i32 0)
%val.sroa.0.0.insert.ext = zext i32 %call3 to i64
%val.sroa.0.4.extract.shift = lshr i64 %ptr.0, 32
%val.sroa.0.4.extract.trunc = trunc i64 %val.sroa.0.4.extract.shift to i32
%call8 = tail call i32 @__shfl_i32(i32 %val.sroa.0.4.extract.trunc, i32 0)
%call8 = tail call i32 @__shfl_i32_nvptx(i32 %val.sroa.0.4.extract.trunc, i32 0)
%val.sroa.0.4.insert.ext = zext i32 %call8 to i64
%val.sroa.0.4.insert.shift = shl nuw i64 %val.sroa.0.4.insert.ext, 32
%val.sroa.0.4.insert.insert = or i64 %val.sroa.0.4.insert.shift, %val.sroa.0.0.insert.ext