| Issue |
79249
|
| Summary |
Inquiry Regarding Exclusion of Super Alignment for DynLDS in D115488
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
Origami404
|
Dear @JonChesterfield,
I hope this issue finds you well. I am an enthusiast developer interested in LLVM AMDGPU backend, and I noticed that in your patch [D115488](https://reviews.llvm.org/D115488) submitted to LLVM, you excluded handling super alignment for LDS. I am curious about the rationale behind this change and would appreciate if you could provide some insights into your decision.
More precisely, I want to understand why you included this code: (still existing in LLVM at [AMDGPULowerModuleLDSPass.cpp#L1339](https://github.com/llvm/llvm-project/blob/baba7e4175b6ca21e83b1cf8229f29dbba02e979/llvm/lib/Target/AMDGPU/AMDGPULowerModuleLDSPass.cpp#L1339) currently)
```cpp
if (!GV.hasInitializer()) {
// cuda/hip extern __shared__ variable, leave alignment alone
continue;
}
```
(Since LLVM Phabricator has been closed, I can only raise this question in Github issue. )
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
