[PATCH] Don't generate 64-bit movd after cmpneqsd in 32-bit mode (PR19059)
Hans Wennborg
hans at chromium.org
Mon Mar 10 18:15:11 PDT 2014
Actually uploading the new patch...
Hi craig.topper,
http://llvm-reviews.chandlerc.com/D3009
CHANGE SINCE LAST DIFF
http://llvm-reviews.chandlerc.com/D3009?vs=7725&id=7728#toc
Files:
lib/Target/X86/X86ISelLowering.cpp
test/CodeGen/X86/isint.ll
Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -18056,9 +18056,25 @@
SDValue OnesOrZeroesF = DAG.getNode(X86ISD::FSETCC, DL,
CMP00.getValueType(), CMP00, CMP01,
DAG.getConstant(x86cc, MVT::i8));
- MVT IntVT = (is64BitFP ? MVT::i64 : MVT::i32);
- SDValue OnesOrZeroesI = DAG.getNode(ISD::BITCAST, DL, IntVT,
- OnesOrZeroesF);
+
+ MVT IntVT = is64BitFP ? MVT::i64 : MVT::i32;
+
+ if (is64BitFP && !Subtarget->is64Bit()) {
+ // On a 32-bit target, we cannot bitcast the 64-bit float to a
+ // 64-bit integer, since that's not a legal type. Since
+ // OnesOrZeroesF is all ones of all zeroes, we don't need all the
+ // bits, but can do this little dance to extract the lowest 32 bits
+ // and work with those going forward.
+ SDValue Vector64 = DAG.getNode(ISD::SCALAR_TO_VECTOR, DL, MVT::v2f64,
+ OnesOrZeroesF);
+ SDValue Vector32 = DAG.getNode(ISD::BITCAST, DL, MVT::v4f32,
+ Vector64);
+ OnesOrZeroesF = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f32,
+ Vector32, DAG.getIntPtrConstant(0));
+ IntVT = MVT::i32;
+ }
+
+ SDValue OnesOrZeroesI = DAG.getNode(ISD::BITCAST, DL, IntVT, OnesOrZeroesF);
SDValue ANDed = DAG.getNode(ISD::AND, DL, IntVT, OnesOrZeroesI,
DAG.getConstant(1, IntVT));
SDValue OneBitOfTruth = DAG.getNode(ISD::TRUNCATE, DL, MVT::i8, ANDed);
Index: test/CodeGen/X86/isint.ll
===================================================================
--- test/CodeGen/X86/isint.ll
+++ test/CodeGen/X86/isint.ll
@@ -1,22 +1,46 @@
-; RUN: llc < %s -march=x86 -mattr=+sse2 -mcpu=penryn| FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-pc-unknown -mattr=+sse2 -mcpu=penryn | FileCheck %s
+; RUN: llc < %s -mtriple=i686-pc-unknown -mattr=+sse2 -mcpu=penryn | FileCheck %s
+
+; PR19059
+; RUN: llc < %s -mtriple=i686-pc-unknown -mattr=+sse2 -mcpu=penryn | FileCheck -check-prefix=CHECK32 %s
define i32 @isint_return(double %d) nounwind {
+; CHECK-LABEL: isint_return:
; CHECK-NOT: xor
; CHECK: cvt
%i = fptosi double %d to i32
; CHECK-NEXT: cvt
%e = sitofp i32 %i to double
; CHECK: cmpeqsd
%c = fcmp oeq double %d, %e
+; CHECK32-NOT: movd {{.*}}, %r{{.*}}
+; CHECK32-NOT: andq
+; CHECK-NEXT: movd
+; CHECK-NEXT: andl
+ %z = zext i1 %c to i32
+ ret i32 %z
+}
+
+define i32 @isint_float_return(float %f) nounwind {
+; CHECK-LABEL: isint_float_return:
+; CHECK-NOT: xor
+; CHECK: cvt
+ %i = fptosi float %f to i32
+; CHECK-NEXT: cvt
+ %g = sitofp i32 %i to float
+; CHECK: cmpeqss
+ %c = fcmp oeq float %f, %g
+; CHECK-NOT: movd {{.*}}, %r{{.*}}
; CHECK-NEXT: movd
-; CHECK-NEXT: andq
+; CHECK-NEXT: andl
%z = zext i1 %c to i32
ret i32 %z
}
declare void @foo()
define void @isint_branch(double %d) nounwind {
+; CHECK-LABEL: isint_branch:
; CHECK: cvt
%i = fptosi double %d to i32
; CHECK-NEXT: cvt
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D3009.4.patch
Type: text/x-patch
Size: 3375 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140310/c248823a/attachment.bin>
More information about the llvm-commits
mailing list