This is an automated email from the ASF dual-hosted git repository. masahi pushed a commit to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git
The following commit(s) were added to refs/heads/main by this push: new 706a5b2dd5 [Target][Minor] Add A6000 Target Tag (#14762) 706a5b2dd5 is described below commit 706a5b2dd5ab8f22a88b84eeb4fd94f318f1e9a5 Author: Xiyou Zhou <xi...@octoml.ai> AuthorDate: Thu May 4 19:44:34 2023 -0700 [Target][Minor] Add A6000 Target Tag (#14762) Add a6000 tag and sources. --- src/target/tag.cc | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/src/target/tag.cc b/src/target/tag.cc index bd2b5958c5..037d2e5937 100644 --- a/src/target/tag.cc +++ b/src/target/tag.cc @@ -105,6 +105,12 @@ TVM_REGISTER_TARGET_TAG("nvidia/jetson-agx-xavier") {"registers_per_block", Integer(RegPerBlock)}, \ }); +// Naming convention for CUDA tags see https://developer.nvidia.com/cuda-gpus +// Parameters see Table 15. Technical Specifications per Compute Capability +// https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html +// Check `Maximum y- or z-dimension of a grid of thread blocks` for max threads per block +// Check `Maximum amount of shared memory per thread block` for max shared memory per block +// Note that above 48 KB requires dynamic shared memory TVM_REGISTER_CUDA_TAG("nvidia/tesla-k80", "sm_37", 49152, 65536); TVM_REGISTER_CUDA_TAG("nvidia/tesla-k40", "sm_35", 49152, 65536); TVM_REGISTER_CUDA_TAG("nvidia/tesla-k20", "sm_35", 49152, 65536); @@ -129,6 +135,7 @@ TVM_REGISTER_CUDA_TAG("nvidia/tesla-k80", "sm_37", 49152, 65536); TVM_REGISTER_CUDA_TAG("nvidia/tesla-k40", "sm_35", 49152, 65536); TVM_REGISTER_CUDA_TAG("nvidia/tesla-k20", "sm_35", 49152, 65536); TVM_REGISTER_CUDA_TAG("nvidia/tesla-k10", "sm_30", 49152, 65536); +TVM_REGISTER_CUDA_TAG("nvidia/rtx-a6000", "sm_86", 49152, 65536); TVM_REGISTER_CUDA_TAG("nvidia/quadro-rtx-8000", "sm_75", 49152, 65536); TVM_REGISTER_CUDA_TAG("nvidia/quadro-rtx-6000", "sm_75", 49152, 65536); TVM_REGISTER_CUDA_TAG("nvidia/quadro-rtx-5000", "sm_75", 49152, 65536);