gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[Triton-MLIR] Replace triton.extract_slice with tensor.extract_slice and support more general tensor slicing (#837)

Browse Source 
## Features

- Allow taking a block of tensor slice, as long as each dimension is
contiguous (unit stride).
- Fix some problems in `insert_slice_async`'s semantic.
- More general verification for ops that return shared layout encoding.

## Known Limitations

- `insert_slice_async` still uses the old semantic. May submit another
PR later to support similar semantic like `tensor.extract_slice`.
- No encoding verification for `tensor.extract_slice`.
- 3d tensor ops are broken.
- Strided accesses are not allowed.
- May cause a little performance slowdown since we are passing strides
as values but not constants (e.g., int).
It would be difficult to pass strides as attributes when we have control
flows. A block argument is possible to accept tensors with different
strides.
This commit is contained in:
Keren Zhou
2022-11-06 22:59:03 -08:00
committed by GitHub
parent a4ff0c362c
commit fdd59900f7
26 changed files with 507 additions and 339 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
test/TritonGPU/combine.mlir
View File

@@ -62,7 +62,7 @@ func @remat(%arg0: i32) -> tensor<1024xi32, #layout1> {
// CHECK-LABEL: transpose
func @transpose(%arg0: !tt.ptr<f32> {tt.divisibility = 16 : i32}, %arg1: i32 {tt.divisibility = 16 : i32}, %arg2: !tt.ptr<f32> {tt.divisibility = 16 : i32}, %arg3: i32 {tt.divisibility = 16 : i32}) {
// CHECK-NOT: triton_gpu.convert_layout
// CHECK: [[loaded_val:%.*]] = tt.load {{.*}}, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<64x64xf32, [[row_layout]]>
// CHECK: [[loaded_val:%.*]] = tt.load {{.*}}, %cst, %cst_0 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<64x64xf32, [[row_layout]]>
// CHECK: [[cvt_val:%.*]] = triton_gpu.convert_layout [[loaded_val]] : (tensor<64x64xf32, [[row_layout]]>) -> tensor<64x64xf32, [[col_layout]]>
// CHECK: tt.store {{.*}}, [[cvt_val]], %cst_1 : tensor<64x64xf32, [[col_layout]]>
// CHECK: return
@@ -91,7 +91,7 @@ func @transpose(%arg0: !tt.ptr<f32> {tt.divisibility = 16 : i32}, %arg1: i32 {tt
%19 = triton_gpu.convert_layout %10 : (tensor<64x64x!tt.ptr<f32>, #blocked1>) -> tensor<64x64x!tt.ptr<f32>, #blocked3>
%20 = triton_gpu.convert_layout %cst_0 : (tensor<64x64xi1, #blocked1>) -> tensor<64x64xi1, #blocked3>
%21 = triton_gpu.convert_layout %cst : (tensor<64x64xf32, #blocked1>) -> tensor<64x64xf32, #blocked3>
%22 = tt.load %19, %20, %21 {cache = 1 : i32, evict = 1 : i32, isVolatile = false, isOtherUnspecified = false} : tensor<64x64xf32, #blocked3>
%22 = tt.load %19, %20, %21 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<64x64xf32, #blocked3>
%23 = triton_gpu.convert_layout %22 : (tensor<64x64xf32, #blocked3>) -> tensor<64x64xf32, #blocked1>
%24 = triton_gpu.convert_layout %18 : (tensor<64x64x!tt.ptr<f32>, #blocked1>) -> tensor<64x64x!tt.ptr<f32>, #blocked4>
%25 = triton_gpu.convert_layout %23 : (tensor<64x64xf32, #blocked1>) -> tensor<64x64xf32, #blocked4>
@@ -133,7 +133,7 @@ func @loop(%arg0: !tt.ptr<f32>, %arg1: i32, %arg2: !tt.ptr<f32>, %arg3: i32, %ar
%23 = triton_gpu.convert_layout %arg7 : (tensor<64x64x!tt.ptr<f32>, #blocked1>) -> tensor<64x64x!tt.ptr<f32>, #blocked3>
%24 = triton_gpu.convert_layout %cst : (tensor<64x64xi1, #blocked1>) -> tensor<64x64xi1, #blocked3>
%25 = triton_gpu.convert_layout %cst_1 : (tensor<64x64xf32, #blocked1>) -> tensor<64x64xf32, #blocked3>
%26 = tt.load %23, %24, %25 {cache = 1 : i32, evict = 1 : i32, isOtherUnspecified = false, isVolatile = false} : tensor<64x64xf32, #blocked3>
%26 = tt.load %23, %24, %25 {cache = 1 : i32, evict = 1 : i32, isVolatile = false} : tensor<64x64xf32, #blocked3>
%27 = triton_gpu.convert_layout %26 : (tensor<64x64xf32, #blocked3>) -> tensor<64x64xf32, #blocked1>
%28 = arith.addf %arg6, %27 : tensor<64x64xf32, #blocked1>
%29 = tt.addptr %arg7, %cst_0 : tensor<64x64x!tt.ptr<f32>, #blocked1>