[Mlir-commits] [mlir] [mlir][linalg] Update createWriteOrMaskedWrite (PR #174810)

[Andrzej Warzyński]

banach-space wrote:

Thanks for reviewing!

> How does the load and store look like for the example in the description? 
BEFORE
```mlir
    %c0_i32 = arith.constant 0 : i32
    %c0 = arith.constant 0 : index
    %c0_0 = arith.constant 0 : index
    %c5 = arith.constant 5 : index
    %c1 = arith.constant 1 : index
    %1 = vector.create_mask %c5, %c1 : vector<8x1xi1>
    %2 = vector.mask %1 { vector.transfer_read %extracted_slice[%c0_0, %c0_0], %c0_i32 {in_bounds = [true, true]} : tensor<5x1xi32>, vector<8x1xi32> } : vector<8x1xi1> -> vector<8x1xi32>
    %c0_1 = arith.constant 0 : index
    %c0_2 = arith.constant 0 : index
    %dim = tensor.dim %0, %c0_2 : tensor<?x3xi32>
    %c3 = arith.constant 3 : index
    %3 = vector.create_mask %dim, %c3 : vector<8x1xi1>
    %4 = vector.mask %3 { vector.transfer_write %2, %0[%c0_1, %c2] {in_bounds = [true, true]} : vector<8x1xi32>, tensor<?x3xi32> } : vector<8x1xi1> -> tensor<?x3xi32>
```

AFTER:
```mlir
    %c0_i32 = arith.constant 0 : i32
    %c0 = arith.constant 0 : index
    %c0_0 = arith.constant 0 : index
    %c5 = arith.constant 5 : index
    %c1 = arith.constant 1 : index
    %1 = vector.create_mask %c5, %c1 : vector<8x1xi1>
    %2 = vector.mask %1 { vector.transfer_read %extracted_slice[%c0_0, %c0_0], %c0_i32 {in_bounds = [true, true]} : tensor<5x1xi32>, vector<8x1xi32> } : vector<8x1xi1> -> vector<8x1xi32>
    %c0_1 = arith.constant 0 : index
    %c0_2 = arith.constant 0 : index
    %dim = tensor.dim %0, %c0_2 : tensor<?x3xi32>
    %c3 = arith.constant 3 : index
    %c1_3 = arith.constant 1 : index
    %3 = vector.create_mask %dim, %c1_3 : vector<8x1xi1>
    %4 = vector.mask %3 { vector.transfer_write %2, %0[%c0_1, %c2] {in_bounds = [true, true]} : vector<8x1xi32>, tensor<?x3xi32> } : vector<8x1xi1> -> tensor<?x3xi32>
  }
```

> The indices should be part of the "base" address computation, right?

Hm, what _address_ and what _computation_? 😅 

> Also, does insert slice define OOB insertions?

I didn't find anything: https://mlir.llvm.org/docs/Dialects/TensorOps/#tensorinsert_slice-tensorinsertsliceop

Btw, I haven't looked 

https://github.com/llvm/llvm-project/pull/174810