diff options
| author | Princeton Ferro <pferro@nvidia.com> | 2025-06-24 16:02:13 -0400 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-06-24 16:02:13 -0400 |
| commit | e1cd450c8f7ca1815364b6b0a275be71eea39860 (patch) | |
| tree | fdc0857847aa504e7fb247472f590a9c88a91c9e /clang/lib/CodeGen/CodeGenModule.cpp | |
| parent | 3187d4da24adad5f31589c25b88f5e34f97dd52b (diff) | |
| download | llvm-e1cd450c8f7ca1815364b6b0a275be71eea39860.zip llvm-e1cd450c8f7ca1815364b6b0a275be71eea39860.tar.gz llvm-e1cd450c8f7ca1815364b6b0a275be71eea39860.tar.bz2 | |
[NVPTX] fold movs into loads and stores (#144581)
Fold movs into loads and stores by increasing the number of return
values or operands. For example:
```
L: v2f16,ch = Load [p]
e0 = extractelt L, 0
e1 = extractelt L, 1
consume(e0, e1)
```
...becomes...
```
L: f16,f16,ch = LoadV2 [p]
consume(L:0, L:1)
```
Diffstat (limited to 'clang/lib/CodeGen/CodeGenModule.cpp')
0 files changed, 0 insertions, 0 deletions