RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-26 04:32:44 +01:00
Code Issues Projects Releases Wiki Activity

[DAGCombine] Fix for shuffle to vector extend for non power 2 vectors

Browse Source 
Summary:
See https://llvm.org/PR33743 for more details

It seems that for non-power of 2 vector sizes, the algorithm can produce
non-matching sizes for input and result causing an assert.

This usually isn't a problem as the isAnyExtend check will weed these out, but
in some cases (most often with lots of undefined values for the mask indices) it
can pass this check for non power of 2 vectors.

Adding in an extra check that ensures that bit size will match for the result
and input (as required)

Subscribers: nhaehnle

Differential Revision: https://reviews.llvm.org/D35241

llvm-svn: 315307
This commit is contained in:
David Stuttard 2017-10-10 12:45:45 +00:00
parent 39a050b107
commit ed11cfd166
2 changed files with 35 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
lib/CodeGen/SelectionDAG/DAGCombiner.cpp
View File

@ -15566,6 +15566,9 @@ static SDValue combineShuffleToVectorExtend(ShuffleVectorSDNode *SVN,
// Attempt to match a '*_extend_vector_inreg' shuffle, we just search for
// power-of-2 extensions as they are the most likely.
for (unsigned Scale = 2; Scale < NumElts; Scale *= 2) {
// Check for non power of 2 vector sizes
if (NumElts % Scale != 0)
continue;
if (!isAnyExtend(Scale))
continue;

32
test/CodeGen/AMDGPU/dagcomb-shuffle-vecextend-non2.ll Normal file
View File

@ -0,0 +1,32 @@
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
; We are only checking that instruction selection can succeed in this case. This
; cut down test results in no instructions, but that's fine.
;
; See https://llvm.org/PR33743 for details of the bug being addressed
;
; Checking that shufflevector with 3-vec mask is handled in
; combineShuffleToVectorExtend
;
; GCN: s_endpgm
define amdgpu_ps void @main(i32 %in1) local_unnamed_addr {
.entry:
br i1 undef, label %bb12, label %bb
bb:
%__llpc_global_proxy_r5.12.vec.insert = insertelement <4 x i32> undef, i32 %in1, i32 3
%tmp3 = shufflevector <4 x i32> %__llpc_global_proxy_r5.12.vec.insert, <4 x i32> undef, <3 x i32> <i32 undef, i32 undef, i32 1>
%tmp4 = bitcast <3 x i32> %tmp3 to <3 x float>
%a2.i123 = extractelement <3 x float> %tmp4, i32 2
%tmp5 = bitcast float %a2.i123 to i32
%__llpc_global_proxy_r2.0.vec.insert196 = insertelement <4 x i32> undef, i32 %tmp5, i32 0
br label %bb12
bb12:
%__llpc_global_proxy_r2.0 = phi <4 x i32> [ %__llpc_global_proxy_r2.0.vec.insert196, %bb ], [ undef, %.entry ]
%tmp6 = shufflevector <4 x i32> %__llpc_global_proxy_r2.0, <4 x i32> undef, <3 x i32> <i32 1, i32 2, i32 3>
%tmp7 = bitcast <3 x i32> %tmp6 to <3 x float>
%a0.i = extractelement <3 x float> %tmp7, i32 0
ret void
}