Details
- Reviewers
rampitec arsenm - Commits
- rG62bcfcb5a588: [AMDGPU] Add llvm.amdgcn.s.setprio intrinsic
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
LGTM.
| llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.s.setprio.ll | ||
|---|---|---|
| 1 ↗ | (On Diff #412946) | I'd suggest just adding -global-isel run lines to the normal test. No need for this file. |
| llvm/test/CodeGen/AMDGPU/llvm.amdgcn.s.setprio.ll | ||
|---|---|---|
| 20 | Should test the maximum value (and beyond it too) | |
| llvm/test/CodeGen/AMDGPU/llvm.amdgcn.s.setprio.ll | ||
|---|---|---|
| 20 | This is printing an unencodable value, should have sign extended or something | |
| llvm/test/CodeGen/AMDGPU/llvm.amdgcn.s.setprio.ll | ||
|---|---|---|
| 20 | Do you mean the test should have a sign extend or that it should not select with an unencodable value? | |
| llvm/test/CodeGen/AMDGPU/llvm.amdgcn.s.setprio.ll | ||
|---|---|---|
| 20 | Ideally the frontend would diagnose, but I would expect discarding the extra bits on select | |
LGTM. We were using i32 in the intrinsics everywhere else for immediate fields but I guess it doesn't really matter
| llvm/test/CodeGen/AMDGPU/llvm.amdgcn.s.setprio.ll | ||
|---|---|---|
| 20 | Surprisingly, it is the same for other SOP intrinsics. | |
Should test the maximum value (and beyond it too)