[llvm] r235295 - [X86][FastIsel] Fix assertion failure when selecting int-to-double conversion (PR23273).

[Andrea Di Biagio]

Author: adibiagio
Date: Mon Apr 20 06:56:59 2015
New Revision: 235295

URL: http://llvm.org/viewvc/llvm-project?rev=235295&view=rev
Log:
[X86][FastIsel] Fix assertion failure when selecting int-to-double conversion (PR23273).

This fixes a regression introduced at revision 231243.
The target-independent selection algorithm in FastISel knows how to select
a SINT_TO_FP if the target is SSE but not AVX. That is because on X86, the
tablegen'd 'fastEmit' functions know how to select CVTSI2SSrr and CVTSI2SDrr.

Method X86FastISel::X86SelectSIToFP was therefore working under the
wrong assumption that the target was AVX. That assumption was incorrect since
we can have a target that is neither AVX nor SSE.

So, rather than asserting for the presence of AVX, we should have had an
early exit from 'X86SelectSIToFP' if the target was not AVX.
This patch fixes the issue replacing the invalid assertion with an early exit.

Thanks to Dimitry Andric for reporting this problem and for providing a small
reproducible testcase. Added test pr23273.ll.

Added:
    llvm/trunk/test/CodeGen/X86/pr23273.ll
Modified:
    llvm/trunk/lib/Target/X86/X86FastISel.cpp

Modified: llvm/trunk/lib/Target/X86/X86FastISel.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86FastISel.cpp?rev=235295&r1=235294&r2=235295&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86FastISel.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86FastISel.cpp Mon Apr 20 06:56:59 2015
@@ -2065,6 +2065,12 @@ bool X86FastISel::X86SelectSelect(const
 }
 
 bool X86FastISel::X86SelectSIToFP(const Instruction *I) {
+  // The target-independent selection algorithm in FastISel already knows how
+  // to select a SINT_TO_FP if the target is SSE but not AVX.
+  // Early exit if the subtarget doesn't have AVX.
+  if (!Subtarget->hasAVX())
+    return false;
+
   if (!I->getOperand(0)->getType()->isIntegerTy(32))
     return false;
 
@@ -2087,11 +2093,6 @@ bool X86FastISel::X86SelectSIToFP(const
   } else
     return false;
 
-  // The target-independent selection algorithm in FastISel already knows how
-  // to select a SINT_TO_FP if the target is SSE but not AVX. This code is only
-  // reachable if the subtarget has AVX.
-  assert(Subtarget->hasAVX() && "Expected a subtarget with AVX!");
-
   unsigned ImplicitDefReg = createResultReg(RC);
   BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
           TII.get(TargetOpcode::IMPLICIT_DEF), ImplicitDefReg);

Added: llvm/trunk/test/CodeGen/X86/pr23273.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pr23273.ll?rev=235295&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/pr23273.ll (added)
+++ llvm/trunk/test/CodeGen/X86/pr23273.ll Mon Apr 20 06:56:59 2015
@@ -0,0 +1,17 @@
+; RUN: llc -mtriple=i386-unknown-unknown -mcpu=generic -march=x86 -mattr=-sse2 -fast-isel < %s
+
+; Verify that the backend doesn't crash during fast-isel with an assertion
+; failure when selecting a int-to-double conversion. The fast selection routine
+; for SINT_TO_FP wrongly assumed that the target had at least SSE2.
+
+ at a = common global i32 0, align 4
+
+define i32 @pr23273() {
+entry:
+  %0 = load i32, i32* @a, align 4
+  %conv = sitofp i32 %0 to double
+  %call = call i32 @fn1(double %conv)
+  ret i32 0
+}
+
+declare i32 @fn1(double) #1