forked from OSchip/llvm-project
[SystemZ] Do not emit VEXTEND or VROUND nodes without vector support.
Review: Ulrich Weigand https://reviews.llvm.org/D58240 llvm-svn: 354039
This commit is contained in:
parent
a2a6acf3f0
commit
aa0b77d339
|
@ -5505,6 +5505,10 @@ SDValue SystemZTargetLowering::combineJOIN_DWORDS(
|
|||
|
||||
SDValue SystemZTargetLowering::combineFP_ROUND(
|
||||
SDNode *N, DAGCombinerInfo &DCI) const {
|
||||
|
||||
if (!Subtarget.hasVector())
|
||||
return SDValue();
|
||||
|
||||
// (fpround (extract_vector_elt X 0))
|
||||
// (fpround (extract_vector_elt X 1)) ->
|
||||
// (extract_vector_elt (VROUND X) 0)
|
||||
|
@ -5552,6 +5556,10 @@ SDValue SystemZTargetLowering::combineFP_ROUND(
|
|||
|
||||
SDValue SystemZTargetLowering::combineFP_EXTEND(
|
||||
SDNode *N, DAGCombinerInfo &DCI) const {
|
||||
|
||||
if (!Subtarget.hasVector())
|
||||
return SDValue();
|
||||
|
||||
// (fpextend (extract_vector_elt X 0))
|
||||
// (fpextend (extract_vector_elt X 2)) ->
|
||||
// (extract_vector_elt (VEXTEND X) 0)
|
||||
|
|
|
@ -0,0 +1,50 @@
|
|||
# RUN: llc -mtriple=s390x-linux-gnu -mcpu=z10 -start-before=codegenprepare %s \
|
||||
# RUN: -o - | FileCheck %s
|
||||
|
||||
# Test that VEXTEND or VROUND nodes are not emitted without vector support.
|
||||
|
||||
--- |
|
||||
|
||||
@.str = external dso_local unnamed_addr constant [21 x i8], align 2
|
||||
|
||||
define void @fun1() #0 {
|
||||
bb:
|
||||
%tmp = load <4 x float>, <4 x float>* undef, align 16
|
||||
%tmp1 = extractelement <4 x float> %tmp, i32 0
|
||||
%tmp2 = fpext float %tmp1 to double
|
||||
%tmp3 = extractelement <4 x float> %tmp, i32 2
|
||||
%tmp4 = fpext float %tmp3 to double
|
||||
tail call void (i8*, ...) @printf(i8* getelementptr inbounds ([21 x i8], [21 x i8]* @.str, i64 0, i64 0), double %tmp2, double undef, double %tmp4, double undef)
|
||||
ret void
|
||||
}
|
||||
|
||||
define void @fun2() #0 {
|
||||
bb:
|
||||
%tmp = load <2 x double>, <2 x double>* undef, align 16
|
||||
%tmp1 = extractelement <2 x double> %tmp, i32 0
|
||||
%tmp2 = fptrunc double %tmp1 to float
|
||||
%tmp3 = extractelement <2 x double> %tmp, i32 1
|
||||
%tmp4 = fptrunc double %tmp3 to float
|
||||
tail call void (i8*, ...) @printf(i8* getelementptr inbounds ([21 x i8], [21 x i8]* @.str, i64 0, i64 0), float %tmp2, float undef, float %tmp4, float undef)
|
||||
ret void
|
||||
}
|
||||
|
||||
declare dso_local void @printf(i8*, ...) #0
|
||||
|
||||
...
|
||||
|
||||
# CHECK-LABEL: fun1:
|
||||
# CHECK: ldeb
|
||||
# CHECK-LABEL: fun2:
|
||||
# CHECK: ledbr
|
||||
|
||||
---
|
||||
name: fun1
|
||||
alignment: 4
|
||||
tracksRegLiveness: true
|
||||
...
|
||||
---
|
||||
name: fun2
|
||||
alignment: 4
|
||||
tracksRegLiveness: true
|
||||
...
|
Loading…
Reference in New Issue