[x86] avoid infinite loop from SoftenFloatOperand (PR34866)

Legalization of fp128 assumes things that we should have asserts for,
so that's another potential improvement.

Differential Revision: https://reviews.llvm.org/D38771

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@315485 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index f95603e9fefb8f81af3aa6b48b8db81b86e925eb..28c4968247af428eb5c7ee3fe250d88db0134e94 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -35214,6 +35214,11 @@ static SDValue combineVectorSizedSetCCEquality(SDNode *SetCC, SelectionDAG &DAG,
   if (!OpVT.isScalarInteger() || OpSize < 128 || isNullConstant(Y))
     return SDValue();
 
+  // Bail out if we know that this is not really just an oversized integer.
+  if (peekThroughBitcasts(X).getValueType() == MVT::f128 ||
+      peekThroughBitcasts(Y).getValueType() == MVT::f128)
+    return SDValue();
+
   // TODO: Use PXOR + PTEST for SSE4.1 or later?
   // TODO: Add support for AVX-512.
   EVT VT = SetCC->getValueType(0);

diff --git a/test/CodeGen/X86/fp128-cast.ll b/test/CodeGen/X86/fp128-cast.ll
index 6543292c08b4c2295036e5b1e00ccafe74f24b49..560892485d89955a7a5a3b7cecffad721ee64839 100644 (file)
--- a/test/CodeGen/X86/fp128-cast.ll
+++ b/test/CodeGen/X86/fp128-cast.ll
@@ -1,5 +1,7 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -O2 -mtriple=x86_64-linux-android -mattr=+mmx | FileCheck %s --check-prefix=X64
 ; RUN: llc < %s -O2 -mtriple=x86_64-linux-gnu -mattr=+mmx | FileCheck %s --check-prefix=X64
+; RUN: llc < %s -O2 -mtriple=x86_64-linux-gnu -mattr=-mmx | FileCheck %s --check-prefix=X64_NO_MMX
 ; RUN: llc < %s -O2 -mtriple=i686-linux-gnu -mattr=+mmx | FileCheck %s --check-prefix=X32
 
 ; Check soft floating point conversion function calls.
@@ -359,6 +361,69 @@ cleanup:                                          ; preds = %entry, %if.then
 ; X64:       retq
 }
 
+define i1 @PR34866(i128 %x) {
+; X64-LABEL: PR34866:
+; X64:       # BB#0:
+; X64-NEXT:    movaps {{.*}}(%rip), %xmm0
+; X64-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    xorq -{{[0-9]+}}(%rsp), %rsi
+; X64-NEXT:    xorq -{{[0-9]+}}(%rsp), %rdi
+; X64-NEXT:    orq %rsi, %rdi
+; X64-NEXT:    sete %al
+; X64-NEXT:    retq
+;
+; X64_NO_MMX-LABEL: PR34866:
+; X64_NO_MMX:       # BB#0:
+; X64_NO_MMX-NEXT:    orq %rsi, %rdi
+; X64_NO_MMX-NEXT:    sete %al
+; X64_NO_MMX-NEXT:    retq
+;
+; X32-LABEL: PR34866:
+; X32:       # BB#0:
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    orl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    orl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    orl %ecx, %eax
+; X32-NEXT:    sete %al
+; X32-NEXT:    retl
+  %bc_mmx = bitcast fp128 0xL00000000000000000000000000000000 to i128
+  %cmp = icmp eq i128 %bc_mmx, %x
+  ret i1 %cmp
+}
+
+define i1 @PR34866_commute(i128 %x) {
+; X64-LABEL: PR34866_commute:
+; X64:       # BB#0:
+; X64-NEXT:    movaps {{.*}}(%rip), %xmm0
+; X64-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    xorq -{{[0-9]+}}(%rsp), %rsi
+; X64-NEXT:    xorq -{{[0-9]+}}(%rsp), %rdi
+; X64-NEXT:    orq %rsi, %rdi
+; X64-NEXT:    sete %al
+; X64-NEXT:    retq
+;
+; X64_NO_MMX-LABEL: PR34866_commute:
+; X64_NO_MMX:       # BB#0:
+; X64_NO_MMX-NEXT:    orq %rsi, %rdi
+; X64_NO_MMX-NEXT:    sete %al
+; X64_NO_MMX-NEXT:    retq
+;
+; X32-LABEL: PR34866_commute:
+; X32:       # BB#0:
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    orl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    orl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    orl %ecx, %eax
+; X32-NEXT:    sete %al
+; X32-NEXT:    retl
+  %bc_mmx = bitcast fp128 0xL00000000000000000000000000000000 to i128
+  %cmp = icmp eq i128 %x, %bc_mmx
+  ret i1 %cmp
+}
+
+
 declare double @copysign(double, double) #1
 
 attributes #2 = { nounwind readnone }