RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-12-04 17:56:53 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

AVX: Fix a bug in getTargetVShiftNode. The shift amount has to be a 128bit vector with the same element type as the input vector.

Browse Source 
This is needed because of the patterns we have for the VP[SLL/SRA/SRL][W/D/Q] instructions.

llvm-svn: 160222
This commit is contained in:
Nadav Rotem 2012-07-14 22:26:05 +00:00
parent a6dfd89cd0
commit c40d85dda5
2 changed files with 38 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
lib/Target/X86/X86ISelLowering.cpp
View File

@ -9458,7 +9458,13 @@ static SDValue getTargetVShiftNode(unsigned Opc, DebugLoc dl, EVT VT,
ShOps[2] = DAG.getUNDEF(MVT::i32);
ShOps[3] = DAG.getUNDEF(MVT::i32);
ShAmt = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, &ShOps[0], 4);
ShAmt = DAG.getNode(ISD::BITCAST, dl, VT, ShAmt);
// The return type has to be a 128-bit type with the same element
// type as the input type.
MVT EltVT = VT.getVectorElementType().getSimpleVT();
EVT ShVT = MVT::getVectorVT(EltVT, 128/EltVT.getSizeInBits());
ShAmt = DAG.getNode(ISD::BITCAST, dl, ShVT, ShAmt);
return DAG.getNode(Opc, dl, VT, SrcOp, ShAmt);
}

31
test/CodeGen/X86/2012-07-15-vshl.ll Normal file
View File

@ -0,0 +1,31 @@
; RUN: llc < %s -march=x86 -mcpu=corei7 -mattr=+avx
; PR13352
declare <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float>, <8 x float>, <8 x float>) nounwind readnone
define void @f_f() nounwind {
allocas:
br label %for_loop29
for_loop29: ; preds = %safe_if_after_true, %allocas
%indvars.iv596 = phi i64 [ %indvars.iv.next597, %safe_if_after_true ], [ 0, %allocas ]
%0 = trunc i64 %indvars.iv596 to i32
%smear.15 = insertelement <16 x i32> undef, i32 %0, i32 15
%bitop = lshr <16 x i32> zeroinitializer, %smear.15
%bitop35 = and <16 x i32> %bitop, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
%bitop35_to_bool = icmp ne <16 x i32> %bitop35, zeroinitializer
%val_to_boolvec32 = sext <16 x i1> %bitop35_to_bool to <16 x i32>
%floatmask.i526 = bitcast <16 x i32> %val_to_boolvec32 to <16 x float>
%mask1.i529 = shufflevector <16 x float> %floatmask.i526, <16 x float> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
%"internal_mask&function_mask41_any" = icmp eq i32 undef, 0
br i1 %"internal_mask&function_mask41_any", label %safe_if_after_true, label %safe_if_run_true
safe_if_after_true: ; preds = %for_loop29
%indvars.iv.next597 = add i64 %indvars.iv596, 1
br label %for_loop29
safe_if_run_true: ; preds = %for_loop29
%blend1.i583 = call <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float> undef, <8 x float> undef, <8 x float> %mask1.i529) nounwind
unreachable
}