diff options
| author | Vidhya Sudhan Loganathan <vidhyasudhan.loganathan@arm.com> | 2018-11-20 15:38:13 +0000 |
|---|---|---|
| committer | Anthony Barbier <Anthony.barbier@arm.com> | 2018-11-20 17:40:03 +0000 |
| commit | 38d93bdda73f9b1024c6b4b834b382f7f25aae19 (patch) | |
| tree | d2372cd8f6aafabaee500beadfab969516b1c524 /tests/benchmark | |
| parent | 8f2cbfa15bfb0e49ca6a334a220f0e36964289d6 (diff) | |
| download | ComputeLibrary-38d93bdda73f9b1024c6b4b834b382f7f25aae19.tar.gz | |
COMPMID-1801 : (Nightly) CLWinogradConvolutionLayer FP16 mismatches
FP mixed precision support added to GEMM kernel used for fp16 winograd conv on Midgard GPUs
Change-Id: I1619beb025fc484a1ac9d3e528d785edabbc7ee6
Diffstat (limited to 'tests/benchmark')
| -rw-r--r-- | tests/benchmark/fixtures/WinogradConvolutionLayerFixture.h | 6 |
1 files changed, 5 insertions, 1 deletions
diff --git a/tests/benchmark/fixtures/WinogradConvolutionLayerFixture.h b/tests/benchmark/fixtures/WinogradConvolutionLayerFixture.h index 5f44517817..86ea9c4dc1 100644 --- a/tests/benchmark/fixtures/WinogradConvolutionLayerFixture.h +++ b/tests/benchmark/fixtures/WinogradConvolutionLayerFixture.h @@ -60,7 +60,11 @@ public: dst = create_tensor<TensorType>(dst_shape, data_type, 1); // Create and configure function - conv_layer.configure(&src, &weights, &biases, &dst, info, act_info); + if(data_type == DataType::F16){ + conv_layer.configure(&src, &weights, &biases, &dst, info, act_info, true); + }else{ + conv_layer.configure(&src, &weights, &biases, &dst, info, act_info); + } // Allocate tensors src.allocator()->allocate(); |