| Issue |
176647
|
| Summary |
[LoongArch] LLVM 22 miscompiles Zig's `std.math.big.int` code
|
| Labels |
miscompilation,
backend:loongarch,
regression:22
|
| Assignees |
|
| Reporter |
alexrp
|
With LLVM 22, this Zig test is failing (which didn't previously):
```
test-std
└─ run test std-loongarch64-linux-gnu-la64v1_0-Debug-libc 64 pass, 1 crash (65 total)
error: 'math.big.int_test.test.bitAnd #10932' terminated with signal ABRT with stderr:
thread 522312 panic: reached unreachable code
/home/alexrp/Sources/ziglang/zig-llvm22/lib/std/debug.zig:419:14: 0x106579f in assert (test)
if (!ok) unreachable; // assertion failure
^
/home/alexrp/Sources/ziglang/zig-llvm22/lib/std/math/big/int.zig:391:23: 0x122906f in setString (test)
assert(!llmulLimb(.add, self.limbs[0..len], self.limbs[0..len], constants.big_bases[base] - 1));
^
/home/alexrp/Sources/ziglang/zig-llvm22/lib/std/math/big/int.zig:2924:24: 0x1228707 in setString (test)
try m.setString(base, value);
^
/home/alexrp/Sources/ziglang/zig-llvm22/lib/std/math/big/int_test.zig:1938:20: 0x122a8a7 in test.bitAnd #10932 (test)
try b.setString(10, "55000000000915215865915724129619485917228346934191537590366734850266784978214506142389798064826139649163838075568111457203909393174933092857416500785632012953993352521899237655507306575657169267399324107627651067352600878339870446048204062696260567762088867991835386857942106708741836433444432529637331429212430394179472179237695833247299409249810963487516399177133175950185719220422442438098353430605822151595560743492661038899294517012784306863064670126197566982968906306814338148792888550378533207318063660581924736840687332023636827401670268933229183389040490792300121030647791095178823932734160000000000000000000000000000000000000555555550000000000000000000000000000000000000000000000000800000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000");
^
```
Still trying to figure out a way to reduce this; will update the issue with more info soon.
(Just to clarify, we define `la64v1_0` as `32s+64bit+f+d+lsx+ual`.)
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
