RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-04-03 00:01:39 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[SystemZ] Fix LLVM crash on unoptimized code

Browse Source 
Compiling the following function with -O0 would crash, since LLVM would
hit an assertion in getTestUnderMaskCond:

  int test(unsigned long x)
  {
    return x >= 0 && x <= 15;
  }

Fixed by detecting the case in the caller of getTestUnderMaskCond.

llvm-svn: 233541
This commit is contained in:
Ulrich Weigand 2015-03-30 13:46:59 +00:00
parent 9ef437c907
commit fb9da3c0ff
2 changed files with 32 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
lib/Target/SystemZ/SystemZISelLowering.cpp
View File

@ -1530,6 +1530,8 @@ static void adjustForTestUnderMask(SelectionDAG &DAG, Comparison &C) {
MaskVal = -(CmpVal & -CmpVal);
NewC.ICmpType = SystemZICMP::UnsignedOnly;
}
if (!MaskVal)
return;
// Check whether the combination of mask, comparison value and comparison
// type are suitable.

30
test/CodeGen/SystemZ/int-cmp-50.ll Normal file
View File

@ -0,0 +1,30 @@
; Verify that we do not crash on always-true conditions
;
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 -O0
;
; This test was compiled using clang -O0 from the following source code:
;
; int test(unsigned long x)
; {
; return x >= 0 && x <= 15;
; }
define signext i32 @test(i64 %x) {
entry:
%x.addr = alloca i64, align 8
store i64 %x, i64* %x.addr, align 8
%0 = load i64, i64* %x.addr, align 8
%cmp = icmp uge i64 %0, 0
br i1 %cmp, label %land.rhs, label %land.end
land.rhs: ; preds = %entry
%1 = load i64, i64* %x.addr, align 8
%cmp1 = icmp ule i64 %1, 15
br label %land.end
land.end: ; preds = %land.rhs, %entry
%2 = phi i1 [ false, %entry ], [ %cmp1, %land.rhs ]
%land.ext = zext i1 %2 to i32
ret i32 %land.ext
}