[llvm-bugs] [Bug 45677] New: [arm] After 186155b89c2d, fatal error: error in backend: Cannot select: t28: i32 = ARMISD::QADD8b t33, t35

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=45677

            Bug ID: 45677
           Summary: [arm] After 186155b89c2d, fatal error: error in
                    backend: Cannot select: t28: i32 = ARMISD::QADD8b t33,
                    t35
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: dimitry at andric.com
                CC: htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org

Recently I got the following fatal backend error when building
https://github.com/freebsd/freebsd/blob/master/sys/dev/sound/pcm/feeder_mixer.c
for armv5 (on FreeBSD this apparently defaults to -target-cpu arm1022e):

fatal error: error in backend: Cannot select: t37: i32 = ARMISD::QADD8b t43,
t44
  t43: i32,ch = load<(load 1 from %ir.8, !tbaa !65), sext from i8> t0, t5,
undef:i32
    t5: i32 = add t2, t4
      t2: i32,ch = CopyFromReg t0, Register:i32 %1
        t1: i32 = Register %1
      t4: i32,ch = CopyFromReg t0, Register:i32 %2
        t3: i32 = Register %2
    t14: i32 = undef
  t44: i32,ch = load<(load 1 from %ir.11, !tbaa !65), sext from i8> t0, t8,
undef:i32
    t8: i32 = add t7, t4
      t7: i32,ch = CopyFromReg t0, Register:i32 %0
        t6: i32 = Register %0
      t4: i32,ch = CopyFromReg t0, Register:i32 %2
        t3: i32 = Register %2
    t14: i32 = undef
In function: feed_mixer_S8NE

The full test case reduces to:

// clang -cc1 -triple armv5e--- -S -target-cpu arm1022e -O1 feeder_mixer-min.c
char *a;
char b;
int c;
void d() {
  int e, f = *(signed char *)a;
  c = *(signed char *)1;
  e = f + c;
  f = e > 127 ? 127 : e < -0x80 ? 128 : e;
  b = f;
}

Interestingly, this regressed with the *very* last Subversion commit to trunk,
https://reviews.llvm.org/rL375505 ("[InstCombine] Signed saturation patterns"),
a.k.a. https://reviews.llvm.org/rG186155b89c2d, and
https://reviews.llvm.org/D68651.

It still reproduces with recent llvm master.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20200426/11f30a93/attachment.html>