leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 24 Packages Projects Releases Wiki Activity

New updates for 2.11 (#775)

Browse Source 
* New updates.

* Minor profiler updates

Co-authored-by: Aniket Shivam <ashivam@nvidia.com>
This commit is contained in:
ANIKET SHIVAM
2023-01-20 16:32:57 -05:00
committed by GitHub
co-authored by Aniket Shivam
parent d49bef88f9
commit 66d9cddc83
1341 changed files with 2003 additions and 1732 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cutlass/cutlass.h
+26 -16
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017 - 2022 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* Copyright (c) 2017 - 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: BSD-3-Clause
*
* Redistribution and use in source and binary forms, with or without
@@ -38,12 +38,35 @@
////////////////////////////////////////////////////////////////////////////////////////////////////
#ifdef CUTLASS_NAMESPACE
#define cutlass CUTLASS_NAMESPACE
#define concat_tok(a, b) a ## b
#define mkcutlassnamespace(pre, ns) concat_tok(pre, ns)
#define cutlass mkcutlassnamespace(cutlass_, CUTLASS_NAMESPACE)
#endif
////////////////////////////////////////////////////////////////////////////////////////////////////
#define CUTLASS_UNUSED(expr) do { ; } while (&expr != &expr)
#if defined(__NVCC__) || (defined(__clang__) && defined(__CUDA__))
#define CUTLASS_HOST_DEVICE __forceinline__ __device__ __host__
#define CUTLASS_DEVICE __forceinline__ __device__
#elif defined(__CUDACC_RTC__)
#define CUTLASS_HOST_DEVICE __forceinline__ __device__
#define CUTLASS_DEVICE __forceinline__ __device__
#else
#define CUTLASS_HOST_DEVICE inline
#define CUTLASS_DEVICE inline
#endif
////////////////////////////////////////////////////////////////////////////////////////////////////
template<typename T>
CUTLASS_HOST_DEVICE void __CUTLASS_UNUSED(T const &)
{ }
#if defined(__GNUC__)
#define CUTLASS_UNUSED(expr) __CUTLASS_UNUSED(expr)
#else
#define CUTLASS_UNUSED(expr) do { ; } while (&expr != &expr)
#endif
#if !defined(__CUDACC_RTC__)
@@ -69,19 +92,6 @@
namespace cutlass {
////////////////////////////////////////////////////////////////////////////////////////////////////
#if defined(__NVCC__) || (defined(__clang__) && defined(__CUDA__))
#define CUTLASS_HOST_DEVICE __forceinline__ __device__ __host__
#define CUTLASS_DEVICE __forceinline__ __device__
#elif defined(__CUDACC_RTC__)
#define CUTLASS_HOST_DEVICE __forceinline__ __device__
#define CUTLASS_DEVICE __forceinline__ __device__
#else
#define CUTLASS_HOST_DEVICE inline
#define CUTLASS_DEVICE inline
#endif
/// Status code returned by CUTLASS operations
enum class Status {
kSuccess, ///< Operation was successful.