| func.func @winograd_output() { |
| %input = util.unfoldable_constant dense<1.0> : tensor<8x8x1x1x1x1xf32> |
| |
| %init = tensor.empty() : tensor<1x6x6x1xf32> |
| %1 = iree_linalg_ext.winograd.output_transform |
| output_tile_size(6) kernel_size(3) image_dimensions([1, 2]) |
| ins(%input : tensor<8x8x1x1x1x1xf32>) |
| outs(%init : tensor<1x6x6x1xf32>) -> tensor<1x6x6x1xf32> |
| %2 = flow.tensor.reshape %1 : tensor<1x6x6x1xf32> -> tensor<6x6xf32> |
| |
| check.expect_almost_eq_const( |
| %2, |
| dense<[[ 49.00000, 0.00000, 73.50000, 0.00000, 238.87500, 7.00000], |
| [ 0.00000, 0.00000, 0.00000, 0.00000, 0.00000, 0.00000], |
| [ 73.50000, 0.00000, 110.25000, 0.00000, 358.31250, 10.50000], |
| [ 0.00000, 0.00000, 0.00000, 0.00000, 0.00000, 0.00000], |
| [ 238.87500, 0.00000, 358.31250, 0.00000, 1164.51562, 34.12500], |
| [ 7.00000, 0.00000, 10.50000, 0.00000, 34.12500, 1.00000]]> : tensor<6x6xf32> |
| ) : tensor<6x6xf32> |
| |
| return |
| } |
| |
| func.func @winograd_output_nchw() { |
| %input = util.unfoldable_constant dense<1.0> : tensor<8x8x1x1x1x1xf32> |
| |
| %init = tensor.empty() : tensor<1x1x6x6xf32> |
| %1 = iree_linalg_ext.winograd.output_transform |
| output_tile_size(6) kernel_size(3) image_dimensions([2, 3]) |
| ins(%input : tensor<8x8x1x1x1x1xf32>) |
| outs(%init : tensor<1x1x6x6xf32>) -> tensor<1x1x6x6xf32> |
| %2 = flow.tensor.reshape %1 : tensor<1x1x6x6xf32> -> tensor<6x6xf32> |
| |
| check.expect_almost_eq_const( |
| %2, |
| dense<[[ 49.00000, 0.00000, 73.50000, 0.00000, 238.87500, 7.00000], |
| [ 0.00000, 0.00000, 0.00000, 0.00000, 0.00000, 0.00000], |
| [ 73.50000, 0.00000, 110.25000, 0.00000, 358.31250, 10.50000], |
| [ 0.00000, 0.00000, 0.00000, 0.00000, 0.00000, 0.00000], |
| [ 238.87500, 0.00000, 358.31250, 0.00000, 1164.51562, 34.12500], |
| [ 7.00000, 0.00000, 10.50000, 0.00000, 34.12500, 1.00000]]> : tensor<6x6xf32> |
| ) : tensor<6x6xf32> |
| |
| return |
| } |
