[mlir][linalg] Update docs for linalg.generic(NFC)

[CoTinker]

The mixed tensor/buffer semantics has been disallowed in #80660. Closes #124090.

[llvmbot]

@llvm/pr-subscribers-mlir

Author: Longsheng Mou (CoTinker)

Changes

The mixed tensor/buffer semantics has been disallowed in #80660. Closed #124090.

---

Full diff: https://github.com/llvm/llvm-project/pull/127178.diff

1 Files Affected:

• (modified) mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td (-13)

diff --git a/mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td b/mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td
index 29cb8035b583b..6a439bfb09078 100644
--- a/mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td
+++ b/mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td
@@ -138,19 +138,6 @@ def GenericOp : LinalgStructuredBase_Op<"generic", [
       }
     }
     ```
-
-    To allow progressive lowering from the value world (a.k.a tensor values) to
-    the buffer world (a.k.a memref values), a `linalg.generic` op allows mixing
-    tensors and buffers operands and tensor results.
-
-    ```mlir
-    %C = linalg.generic #trait_attribute
-      ins(%A, %B : tensor<?x?xf32>, memref<?x?xf32, stride_specification>)
-      outs(%C : tensor<?x?xf32>)
-      {other-optional-attributes}
-      {region}
-      -> (tensor<?x?xf32>)
-    ```
   }];
 
   let arguments = (ins Variadic<AnyType>:$inputs,

[llvmbot]

@llvm/pr-subscribers-mlir-linalg

Author: Longsheng Mou (CoTinker)

Changes

The mixed tensor/buffer semantics has been disallowed in #80660. Closed #124090.

---

Full diff: https://github.com/llvm/llvm-project/pull/127178.diff

1 Files Affected:

• (modified) mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td (-13)

diff --git a/mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td b/mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td
index 29cb8035b583b..6a439bfb09078 100644
--- a/mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td
+++ b/mlir/include/mlir/Dialect/Linalg/IR/LinalgStructuredOps.td
@@ -138,19 +138,6 @@ def GenericOp : LinalgStructuredBase_Op<"generic", [
       }
     }
     ```
-
-    To allow progressive lowering from the value world (a.k.a tensor values) to
-    the buffer world (a.k.a memref values), a `linalg.generic` op allows mixing
-    tensors and buffers operands and tensor results.
-
-    ```mlir
-    %C = linalg.generic #trait_attribute
-      ins(%A, %B : tensor<?x?xf32>, memref<?x?xf32, stride_specification>)
-      outs(%C : tensor<?x?xf32>)
-      {other-optional-attributes}
-      {region}
-      -> (tensor<?x?xf32>)
-    ```
   }];
 
   let arguments = (ins Variadic<AnyType>:$inputs,

[CoTinker]

I'm uncertain whether we should delete the following code:

llvm-project/mlir/lib/Dialect/Linalg/IR/LinalgOps.cpp

Lines 1296 to 1298 in d8b2e43

	// Mixed semantics is not supported yet.
	if (!linalgOp.hasPureTensorSemantics())
	return failure();

[matthias-springer]

I thought that DestinationStyleOpInterface already rejects mixed tensor/memref ops, but it does not. If we merge this PR, we should also improve the Linalg op verifier to reject mixed tensor/memref.

Maybe somebody who's been working more closely with Linalg can chime in on mixed tensor/memref support. @nicolasvasilache @MaheshRavishankar @rengolin

[rengolin]

Maybe somebody who's been working more closely with Linalg can chime in on mixed tensor/memref support.

We have been working on it for the past two years with single semantics: bufferization needs to convert all operands in one go. If there are places still allowing the old semantics, I'd fix them in situ.

[CoTinker]

Okay, thanks.