[PATCH] D23974: [TargetLowering] remove fdiv and frem from canOpTrap() (PR29114)

[Sanjay Patel via llvm-commits]

spatel created this revision.
spatel added reviewers: hfinkel, scanon, RKSimon.
spatel added a subscriber: llvm-commits.
Herald added a subscriber: mcrosier.

Assuming the default FP env, we should not treat fdiv and frem any differently in terms of trapping behavior than any other FP op. Ie, FP ops do not trap with the default FP env.

This matches how we treat these ops in IR with isSafeToSpeculativelyExecute(). If this patch is correct, there's a similar bug in Constant::canTrap().

This bug manifests in PR29114:
https://llvm.org/bugs/show_bug.cgi?id=29114
...as a sequence of scalar divisions instead of a vector division on x86 for a <3 x float> type.

https://reviews.llvm.org/D23974

Files:
  include/llvm/Target/TargetLowering.h
  lib/CodeGen/TargetLoweringBase.cpp
  test/CodeGen/X86/vec3.ll

Index: test/CodeGen/X86/vec3.ll
===================================================================
--- test/CodeGen/X86/vec3.ll
+++ test/CodeGen/X86/vec3.ll
@@ -19,16 +19,8 @@
 define <3 x float> @fdiv(<3 x float> %v, float %d) {
 ; CHECK-LABEL: fdiv:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movaps %xmm1, %xmm2
-; CHECK-NEXT:    movaps %xmm0, %xmm3
-; CHECK-NEXT:    movaps %xmm1, %xmm4
-; CHECK-NEXT:    divss %xmm0, %xmm1
-; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; CHECK-NEXT:    divss %xmm0, %xmm2
-; CHECK-NEXT:    movhlps {{.*#+}} xmm3 = xmm3[1,1]
-; CHECK-NEXT:    divss %xmm3, %xmm4
-; CHECK-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1]
-; CHECK-NEXT:    unpcklps {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; CHECK-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0,0,3]
+; CHECK-NEXT:    divps %xmm0, %xmm1
 ; CHECK-NEXT:    movaps %xmm1, %xmm0
 ; CHECK-NEXT:    retq
 ;
Index: lib/CodeGen/TargetLoweringBase.cpp
===================================================================
--- lib/CodeGen/TargetLoweringBase.cpp
+++ lib/CodeGen/TargetLoweringBase.cpp
@@ -955,15 +955,11 @@
   return getScalarShiftAmountTy(DL, LHSTy);
 }
 
-/// canOpTrap - Returns true if the operation can trap for the value type.
-/// VT must be a legal type.
 bool TargetLoweringBase::canOpTrap(unsigned Op, EVT VT) const {
   assert(isTypeLegal(VT));
   switch (Op) {
   default:
     return false;
-  case ISD::FDIV:
-  case ISD::FREM:
   case ISD::SDIV:
   case ISD::UDIV:
   case ISD::SREM:
Index: include/llvm/Target/TargetLowering.h
===================================================================
--- include/llvm/Target/TargetLowering.h
+++ include/llvm/Target/TargetLowering.h
@@ -592,7 +592,7 @@
   /// Returns true if the operation can trap for the value type.
   ///
   /// VT must be a legal type. By default, we optimistically assume most
-  /// operations don't trap except for divide and remainder.
+  /// operations don't trap except for integer divide and remainder.
   virtual bool canOpTrap(unsigned Op, EVT VT) const;
 
   /// Similar to isShuffleMaskLegal. This is used by Targets can use this to


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D23974.69516.patch
Type: text/x-patch
Size: 2144 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160828/6c1799a1/attachment.bin>