https://llvm.org/bugs/show_bug.cgi?id=26559
Bug ID: 26559
Summary: Crash on converting int128 to float
Product: libraries
Version: trunk
Hardware: Other
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: Backend: SystemZ
Assignee: unassignedb...@nondot.org
Reporter: koria...@0x04.net
CC: llvm-bugs@lists.llvm.org
Classification: Unclassified
$ cat bug.c
typedef unsigned tu_int __attribute__ ((mode (TI)));
tu_int a = 1;
float b;
int main() {
b = a;
return 0;
}
$ clang bug.c -O3
$ ./a.out
Segmentation fault (core dumped)
(-O3 is not necessary, but simplifies the emitted assembly)
This is a backend bug - clang simply emits a uitofp:
; Function Attrs: norecurse nounwind
define signext i32 @main() #0 {
entry:
%0 = load i128, i128* @a, align 16, !tbaa !1
%conv = uitofp i128 %0 to float
store float %conv, float* @b, align 4, !tbaa !5
ret i32 0
}
It seems __floatuntis is incorrectly called with the argument in %r2:%r3, when
it should be called with pointer to argument in %r2:
main: # @main
# BB#0: # %entry
stmg %r14, %r15, 112(%r15)
aghi %r15, -160
lgrl %r2, a
lgrl %r3, a+8
brasl %r14, __floatuntisf@PLT
larl %r1, b
ste %f0, 0(%r1)
lghi %r2, 0
lmg %r14, %r15, 272(%r15)
br %r14
gcc gets that right:
main:
.LFB0:
.cfi_startproc
stmg %r14,%r15,112(%r15)
larl %r1,a
lmg %r4,%r5,0(%r1)
.cfi_offset 14, -48
.cfi_offset 15, -40
lay %r15,-176(%r15)
.cfi_def_cfa_offset 336
la %r2,160(%r15)
stmg %r4,%r5,160(%r15)
brasl %r14,__floatuntisf
lghi %r2,0
larl %r5,b
lmg %r14,%r15,288(%r15)
.cfi_restore 15
.cfi_restore 14
.cfi_def_cfa_offset 160
ste %f0,0(%r5)
br %r14
.cfi_endproc
Found when attempting to port UBSan to SystemZ
(compiler-rt/test/ubsan/TestCases/Float/cast-overflow.cpp triggers it).
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
