[Bug target/118409] [15 regression] "Unsupported broadcast" assemble failure with -march=x86-64-v4
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118409 Sam James changed: What|Removed |Added Resolution|MOVED |--- Status|RESOLVED|UNCONFIRMED --- Comment #8 from Sam James --- OK, so far: * tamar is innocent, sorry tamar * PR118097 is unfixed (ipa-cp bug) but it's not related * gas is miscompiled (works with -fdisable-tree-ifcombine, still fails with -fno-ipa-cp) I'll reduce gas now and go from there.
[Bug target/118409] [15 regression] "Unsupported broadcast" assemble failure with -march=x86-64-v4
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118409 Sam James changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED See Also||https://sourceware.org/bugz ||illa/show_bug.cgi?id=32544 Resolution|--- |MOVED --- Comment #7 from Sam James --- We think it's a gas bug: https://sourceware.org/bugzilla/show_bug.cgi?id=32544
[Bug target/118409] [15 regression] "Unsupported broadcast" assemble failure with -march=x86-64-v4
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118409
--- Comment #6 from Sam James ---
```
int ResetLimiterBands_workLimiterBandTable_0, ResetLimiterBands_noPatches,
ResetLimiterBands_k;
char ResetLimiterBands_freqBandTable_0;
int *ResetLimiterBands_xOverQmf;
void ResetLimiterBands(char *limiterBandTable) {
int i;
int patchBorders[6];
if (ResetLimiterBands_xOverQmf)
for (; i < ResetLimiterBands_noPatches; i++)
patchBorders[i] =
ResetLimiterBands_xOverQmf[i] - ResetLimiterBands_freqBandTable_0;
ResetLimiterBands_workLimiterBandTable_0 = patchBorders[0];
limiterBandTable[ResetLimiterBands_k] =
ResetLimiterBands_workLimiterBandTable_0;
}
```
This fails with:
```
$ g++ env_calc.ii -m32 -O2 -march=sapphirerapids
/tmp/ccA3dqYi.s: Assembler messages:
/tmp/ccA3dqYi.s:46: Error: unsupported broadcast for `vpsubd'
```
[Bug target/118409] [15 regression] "Unsupported broadcast" assemble failure with -march=x86-64-v4
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118409 Sam James changed: What|Removed |Added Target Milestone|--- |15.0 CC||tnfchris at gcc dot gnu.org --- Comment #5 from Sam James --- I'm guessing it's a latent issue but exposed by tamar/alex's vectorisation improvements. works: gcc version 15.0.0 20250110 (experimental) 9193641d1695293006ed0b818bb4161a1b6fbed2 (Gentoo Hardened 15.0. p, commit 93975113f3b353a27dea263c60ee62b27894be9b) fails: gcc version 15.0.0 20250110 (experimental) 06c4cf398947b53b4bfc65752f9f879bb2d07924 (Gentoo 15.0. p, commit 93975113f3b353a27dea263c60ee62b27894be9b)
[Bug target/118409] [15 regression] "Unsupported broadcast" assemble failure with -march=x86-64-v4
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118409 --- Comment #4 from Sam James --- I see tonnes of these. I just picked fdk-aac as the first example.
[Bug target/118409] [15 regression] "Unsupported broadcast" assemble failure with -march=x86-64-v4
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118409 --- Comment #3 from Sam James --- Created attachment 60099 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60099&action=edit env_calc.s.xz
[Bug target/118409] [15 regression] "Unsupported broadcast" assemble failure with -march=x86-64-v4
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118409 --- Comment #2 from Sam James --- $ as --version GNU assembler (Gentoo p1) 2.43.50.20250110 [...]
[Bug target/118409] [15 regression] "Unsupported broadcast" assemble failure with -march=x86-64-v4
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118409 --- Comment #1 from Sam James --- Created attachment 60098 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60098&action=edit env_calc.ii.xz
