[llvm] r330981 - DAG: Fix not legalizing vector fcanonicalizes
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Thu Apr 26 12:21:37 PDT 2018
Author: arsenm
Date: Thu Apr 26 12:21:37 2018
New Revision: 330981
URL: http://llvm.org/viewvc/llvm-project?rev=330981&view=rev
Log:
DAG: Fix not legalizing vector fcanonicalizes
If an fcanoncialize was done on a vector type that was legal,
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
llvm/trunk/test/CodeGen/AMDGPU/fcanonicalize.ll
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp?rev=330981&r1=330980&r2=330981&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp Thu Apr 26 12:21:37 2018
@@ -366,6 +366,7 @@ SDValue VectorLegalizer::LegalizeOp(SDVa
case ISD::UMAX:
case ISD::SMUL_LOHI:
case ISD::UMUL_LOHI:
+ case ISD::FCANONICALIZE:
QueryType = Node->getValueType(0);
break;
case ISD::FP_ROUND_INREG:
Modified: llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp?rev=330981&r1=330980&r2=330981&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp Thu Apr 26 12:21:37 2018
@@ -479,6 +479,7 @@ SITargetLowering::SITargetLowering(const
setOperationAction(ISD::FMA, MVT::v2f16, Legal);
setOperationAction(ISD::FMINNUM, MVT::v2f16, Legal);
setOperationAction(ISD::FMAXNUM, MVT::v2f16, Legal);
+ setOperationAction(ISD::FCANONICALIZE, MVT::v2f16, Legal);
// This isn't really legal, but this avoids the legalizer unrolling it (and
// allows matching fneg (fabs x) patterns)
Modified: llvm/trunk/test/CodeGen/AMDGPU/fcanonicalize.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fcanonicalize.ll?rev=330981&r1=330980&r2=330981&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fcanonicalize.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fcanonicalize.ll Thu Apr 26 12:21:37 2018
@@ -7,6 +7,7 @@ declare double @llvm.canonicalize.f64(do
declare half @llvm.canonicalize.f16(half) #0
declare <2 x half> @llvm.canonicalize.v2f16(<2 x half>) #0
declare i32 @llvm.amdgcn.workitem.id.x() #0
+declare <2 x double> @llvm.canonicalize.v2f64(<2 x double>) #0
; GCN-LABEL: {{^}}v_test_canonicalize_var_f32:
; GCN: v_mul_f32_e32 [[REG:v[0-9]+]], 1.0, {{v[0-9]+}}
@@ -528,6 +529,18 @@ define amdgpu_kernel void @test_canonica
ret void
}
+; GCN-LABEL: {{^}}v_test_canonicalize_var_v2f64:
+; GCN: v_max_f64
+; GCN: v_max_f64
+define amdgpu_kernel void @v_test_canonicalize_var_v2f64(<2 x double> addrspace(1)* %out) #1 {
+ %tid = call i32 @llvm.amdgcn.workitem.id.x()
+ %gep = getelementptr <2 x double>, <2 x double> addrspace(1)* %out, i32 %tid
+ %val = load <2 x double>, <2 x double> addrspace(1)* %gep
+ %canonicalized = call <2 x double> @llvm.canonicalize.v2f64(<2 x double> %val)
+ store <2 x double> %canonicalized, <2 x double> addrspace(1)* %out
+ ret void
+}
+
attributes #0 = { nounwind readnone }
attributes #1 = { nounwind }
attributes #2 = { nounwind "target-features"="-fp32-denormals,-fp64-fp16-denormals" }
More information about the llvm-commits
mailing list