[Mlir-commits] [mlir] [MLIR][NVGPU] Remove Memref Rank vs. Coordinates `tma.async.load` (PR #69584)

[llvmlistbot at llvm.org]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-mlir-gpu

Author: Guray Ozen (grypp)

<details>
<summary>Changes</summary>

Previously, a verifier to check for mismatches between memref rank and number of coordinates was introduced. I noticed that it is very strict. Let's take following IR snippet where the verifier complains about mismatches (2 coordinates (%c1,%c2) != memref rank 3).

```
nvgpu.tma.async.load %0[%c1, %c2], %1 to %2 : ... -> memref<1x64x128xf16, ..., 3>
```

This PR relax the verifier.

---
Full diff: https://github.com/llvm/llvm-project/pull/69584.diff


1 Files Affected:

- (modified) mlir/lib/Dialect/NVGPU/IR/NVGPUDialect.cpp (-6) 


``````````diff
diff --git a/mlir/lib/Dialect/NVGPU/IR/NVGPUDialect.cpp b/mlir/lib/Dialect/NVGPU/IR/NVGPUDialect.cpp
index f5b02fe1b515591..2a280df68371d17 100644
--- a/mlir/lib/Dialect/NVGPU/IR/NVGPUDialect.cpp
+++ b/mlir/lib/Dialect/NVGPU/IR/NVGPUDialect.cpp
@@ -347,12 +347,6 @@ LogicalResult TmaAsyncLoadOp::verify() {
   if (getCoordinates().size() > 5) {
     return emitError() << "Maximum 5 coordinates are supported.";
   }
-  if (getCoordinates().size() != size_t(dstMemref.getRank())) {
-    return emitError() << "Destination memref rank is "
-                       << size_t(dstMemref.getRank()) << " but there are  "
-                       << getCoordinates().size()
-                       << " coordinates. They must match.";
-  }
   return success();
 }
 

``````````

</details>


https://github.com/llvm/llvm-project/pull/69584