[dnn] better specification of recompilation key

This commit is contained in:
Philippe Tillet
2019-08-02 17:42:48 -07:00
parent 3b92ddf7e6
commit d9945692a9
31 changed files with 418 additions and 428 deletions

View File

@@ -30,7 +30,7 @@ torch::Tensor conv_common(
stride_d, stride_h, stride_w,
pad_d, pad_h, pad_w,
1, 1, 1,
"fp32", "fp32", ty, has_bias);
"float", "float", ty, has_bias);
// Bind memory
triton::driver::cu_buffer a(ctx, (CUdeviceptr)torcha.storage().data(), false);
triton::driver::cu_buffer b(ctx, (CUdeviceptr)torchb.storage().data(), false);