[PowerPC] Wrong code at Os

[XChy]

Generated by Csmith:

#include "csmith.h"
struct S0 {
  int32_t f1
};
struct {
  struct S0 f3
} g_4, g_1048;
int8_t g_470;
int main() {
  int print_hash_value, i, j;
  crc32_gentab();
  int32_t *l_7 = &g_4;
  int32_t *l_10 = &g_4;
  uint64_t l_18[9][3];
  i = 0;
  for (; i < 9; i++)
    for (; j < 3; j++)
      l_18[i][j] = 9055118;
  for (g_1048.f3.f1 = 0; g_1048.f3.f1 != 8;
       g_1048.f3.f1 = safe_sub_func_uint64_t_u_u(g_1048.f3.f1, 6))
    *l_7 |= l_18[6][0];
  g_470 = safe_rshift_func_uint8_t_u_u(
      safe_sub_func_int64_t_s_s(safe_lshift_func_uint64_t_u_s(*l_10, 0), 0), 0);
  transparent_crc(g_470, "", print_hash_value);
  platform_main_end(crc32_context, print_hash_value);
}

Compile with -Os and run on qemu-ppc64le:

checksum = 368DB106

On X86 it outputs:

checksum = DEBB20E3

[llvmbot]

@llvm/issue-subscribers-backend-powerpc

Author: Hongyu Chen (XChy)

Generated by Csmith: ```c #include "csmith.h" struct S0 { int32_t f1 }; struct { struct S0 f3 } g_4, g_1048; int8_t g_470; int main() { int print_hash_value, i, j; crc32_gentab(); int32_t *l_7 = &g_4; int32_t *l_10 = &g_4; uint64_t l_18[9][3]; i = 0; for (; i < 9; i++) for (; j < 3; j++) l_18[i][j] = 9055118; for (g_1048.f3.f1 = 0; g_1048.f3.f1 != 8; g_1048.f3.f1 = safe_sub_func_uint64_t_u_u(g_1048.f3.f1, 6)) *l_7 |= l_18[6][0]; g_470 = safe_rshift_func_uint8_t_u_u( safe_sub_func_int64_t_s_s(safe_lshift_func_uint64_t_u_s(*l_10, 0), 0), 0); transparent_crc(g_470, "", print_hash_value); platform_main_end(crc32_context, print_hash_value); } ```

Compile with -Os and run on qemu-ppc64le:

checksum = 368DB106

On X86 it outputs:

checksum = DEBB20E3

[nemanjai]

@lei137 I'm not sure if this is on your radar - seems fairly low priority since it's synthetic CSmith code, but thought I'd point it out anyway.

@XChy It would probably really help if rather than C code that requires inclusion of csmith.h, you attached either a pre-processed file or better yet, an LLVM IR file.

[XChy]

@nemanjai, sure, it may take a few hours or days to reduce it again due to some updates.
If possible, could you please also take a look at #159116 #156826? Many thanks! And looks like the PowerPC backend is inactive these days.

[XChy]

Hmm, it cannot be reproduced on the trunk clang now. Looks like a fixed regression now.