[RISC-V] Miscompile at -O2

[patrick-rivos]

Testcase:

int printf(const char *, ...);
long a;
short b = 2;
char c(unsigned char d, unsigned char e) { return e == 0 ?: d % e; }
int main() {
  char f;
  a = 1;
  f = c(0 <= 7, b);
  a ^= f;
  printf("%lX\n", a);
}

Commands:

> clang -O2 red.c -o red.out
> qemu-riscv64 red.out
FFFFFFFFFFFFFFFE
> clang red.c -o red.out
> qemu-riscv64 red.out
0

-opt-bisect-limit points to Type Promotion on function (main)

This is a recent regression.
Known good hash: 083d8aa
Known bad hash: 8f79cdd

Found via fuzzer

[topperc]

Can you check if #84232 fixes this?

[llvmbot]

@llvm/issue-subscribers-backend-risc-v

Author: Patrick O'Neill (patrick-rivos)

Testcase: ```c int printf(const char *, ...); long a; short b = 2; char c(unsigned char d, unsigned char e) { return e == 0 ?: d % e; } int main() { char f; a = 1; f = c(0 <= 7, b); a ^= f; printf("%lX\n", a); } ```

Commands:

> /scratch/tc-testing/tc-mar-5-llvm-slp-ice-2/build/bin/clang -O2 red.c -o red.out
> /scratch/tc-testing/tc-mar-4-llvm/build/bin/qemu-riscv64 red.out
FFFFFFFFFFFFFFFE
> /scratch/tc-testing/tc-mar-5-llvm-slp-ice-2/build/bin/clang red.c -o red.out
> /scratch/tc-testing/tc-mar-4-llvm/build/bin/qemu-riscv64 red.out
0

-opt-bisect-limit points to Type Promotion on function (main)

This is a recent regression.
Known good hash: 083d8aa
Known bad hash: 8f79cdd

Found via fuzzer

[patrick-rivos]

I checked out the head of the PR branch at #84232 and this issue does not exist there.

> clang -O3 red.c -o red.out
> qemu-riscv64 red.out
0

Thanks!

[topperc]

Thanks for confirming. I'm trying to mitigate some of the fallout from that fix before committing it.

[dtcxzyw]

Should be fixed by #84232.