wasm2c/spec/float_exprs.txt test fails with GCC 15 on ppc64le and s390x #2542
Comments
|
at first glance, this looks like a GCC bug. |
|
Ok, gcc bug opened: https://bugzilla.redhat.com/show_bug.cgi?id=2346233 . Could you help prepare a minimum piece of code to reproduce the issue and help gcc maintainers fix it? |
|
test/wasm2c/spec/float_exprs.txt points to https://github.com/WebAssembly/testsuite/blob/eeb6dac81be81151da0958ed955aba91230c1f25/float_exprs.wast#L2269 for the offending code, but how this is translated to a C code that GCC consumes is magic to me :-) Although maybe I am getting somewhere ... |
|
with reduced |
|
does this exhibit the problem? /* min.c */
typedef uint32_t u32;
typedef float f32;
u32 w2c_float__exprs__88__wasm_f320x2Eno_trichotomy_lt_0(void* instance, f32 var_p0, f32 var_p1) {
u32 var_i0, var_i1;
f32 var_f0, var_f1, var_f2;
var_f0 = var_p0;
var_f1 = var_p1;
var_i0 = var_f0 < var_f1;
var_f1 = var_p0;
var_f2 = var_p1;
var_i1 = var_f1 >= var_f2;
var_i0 |= var_i1;
return var_i0;
}/* main.c */
#include <string.h>
typedef uint32_t u32;
typedef float f32;
u32 w2c_float__exprs__88__wasm_f320x2Eno_trichotomy_lt_0(void* instance, f32 var_p0, f32 var_p1);
static f32 make_nan_f32(u32 x) {
x |= 0x7f800000;
f32 res;
memcpy(&res, &x, sizeof(res));
return res;
}
int main(int argc, char **argv) {
return w2c_float__exprs__88__wasm_f320x2Eno_trichotomy_lt_0(NULL, 0.f, make_nan_f32(0x400000));
}(compile with same flags and whatnot ofc, feel free to rename the function the name clearly doesn't matter) |
|
It does not AFAICT, but often there is more code required "surrounding" the minimal code snippets, so the compiler makes different decisions. I believe I should be able to reduce my example above to have something useful for the toolchains team. |
|
we assume you did check the exit status? it doesn't output anything other than the exit status, but that might have been unclear, sorry. |
|
sorry, I haven't read the code :-) Is return code |
|
indeed, it's supposed to return 0. |
|
it can be minimized further, but you can probably do that a lot easier now. |
and compiled with |
|
@SoniEx2 your assistance is much appreciated! Could you take a look at other failed test issues I opened? |
|
I am getting this on s390x, too. |
rathann
changed the title
|
btw, does |
it does And I am confirming the s390x failure. |
|
GCC, that's not how floats work 🤦🏻♀️ thanks for confirming! |
When building wabt-1.0.36 with GCC 15 (Fedora 42 and rawhide) and running the testsuite on ppc64le
This does not happen with GCC 14.x and is not reproducible on x86_64 or aarch64.
The text was updated successfully, but these errors were encountered: