| 1 |
// RUN: mlir-opt %s -affine-super-vectorize="virtual-vector-size=32,256 test-fastest-varying=1,0 vectorize-reductions=true" -verify-diagnostics |
| 2 |
|
| 3 |
// TODO: Vectorization of reduction loops along the reduction dimension is not |
| 4 |
// supported for higher-rank vectors yet, so we are just checking that an |
| 5 |
// error message is produced. |
| 6 |
|
| 7 |
// expected-error@+1 {{Vectorizing reductions is supported only for 1-D vectors}} |
| 8 |
func.func @vecdim_reduction_2d(%in: memref<256x512x1024xf32>, %out: memref<256xf32>) { |
| 9 |
%cst = arith.constant 0.000000e+00 : f32 |
| 10 |
affine.for %i = 0 to 256 { |
| 11 |
%sum_j = affine.for %j = 0 to 512 iter_args(%red_iter_j = %cst) -> (f32) { |
| 12 |
%sum_k = affine.for %k = 0 to 1024 iter_args(%red_iter_k = %cst) -> (f32) { |
| 13 |
%ld = affine.load %in[%i, %j, %k] : memref<256x512x1024xf32> |
| 14 |
%add = arith.addf %red_iter_k, %ld : f32 |
| 15 |
affine.yield %add : f32 |
| 16 |
} |
| 17 |
%add = arith.addf %red_iter_j, %sum_k : f32 |
| 18 |
affine.yield %add : f32 |
| 19 |
} |
| 20 |
affine.store %sum_j, %out[%i] : memref<256xf32> |
| 21 |
} |
| 22 |
return |
| 23 |
} |
| 24 |
|