================ @@ -0,0 +1,71 @@ +//===--- NVPTXTargetParser.def - NVPTX target data -----------*- C++ -*-===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// +// +// This file is the single source of truth for the NVPTX (CUDA) GPU list. Each +// row describes one virtual architecture (sm_XX). Adding a new target is a +// single row here. +// +// NVPTX_GPU(NAME, KIND, VIRTUAL, SM_ID, MIN_VER, MAX_VER, SUFFIX) +// NAME - Canonical processor name string, e.g. "sm_90". +// KIND - GPUKind enumerator suffix; the enumerator is GK_<KIND> +// VIRTUAL - Virtual (compute_) arch name string, e.g. "compute_90". +// SM_ID - Numeric compute-capability id (sm_90 -> 900). +// MIN_VER - Earliest supporting CudaVersion +// MAX_VER - Latest supporting CudaVersion. +// SUFFIX - Arch suffix class: NONE, ACCELERATED (sm_90a), FAMILY (sm_90f). +// +//===----------------------------------------------------------------------===// + +#ifndef NVPTX_GPU +#define NVPTX_GPU(NAME, KIND, VIRTUAL, SM_ID, MIN_VER, MAX_VER, SUFFIX) ---------------- durga4github wrote:
We have the `getMinPTXVersionForSM()` function in NVPTXSubtarget.cpp that maintains yet another table, mapping each SM to its minimum required PTX version. Would it be reasonable to extend this mapping, with another column for MIN_PTX_VER (as a follow-up)? Just want to check if it aligns with the direction you have in mind... https://github.com/llvm/llvm-project/pull/212734 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
