leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 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 13:32:57 -08:00
committed by GitHub
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

47
include/cutlass/subbyte_reference.h
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
@@ -294,8 +294,18 @@ public:
template <
typename Element_, /// CUTLASS numeric element type.
typename Storage_ = uint8_t /// Underlying storage type. Must be able to hold an integer
typename Storage_ = /// Underlying storage type. Must be able to hold an integer
/// number of objects of type Element.
#if defined(__CUDA_ARCH__) /// Default size depends on width of atomicCas() overloads.
#if (__CUDA_ARCH__ >= 700) ///
uint16_t
#else
uint32_t
#endif
#else
uint8_t
#endif
>
class SubbyteReference {
public:
@@ -387,14 +397,41 @@ public:
CUTLASS_HOST_DEVICE
SubbyteReference & set(Element const &x) {
Storage item = (reinterpret_cast<Storage const &>(x) & kMask);
Storage item = (reinterpret_cast<Storage const &>(x) & kMask);
Storage kUpdateMask = Storage(~(kMask << (offset_ * cutlass::sizeof_bits<Element>::value)));
Storage new_bits = Storage(item << (offset_ * cutlass::sizeof_bits<Element>::value));
Storage kUpdateMask = Storage(~(kMask << (offset_ * sizeof_bits<Element>::value)));
*ptr_ = Storage((*ptr_ & kUpdateMask) | Storage(item << (offset_ * sizeof_bits<Element>::value)));
#if defined(__CUDA_ARCH__)
//
// Homebrew read-modify-write
//
Storage original;
Storage updated;
do {
original = (*ptr_);
updated = Storage((original & kUpdateMask) | new_bits);
original = atomicCAS(ptr_, original, updated);
} while (updated != original);
#else
Storage original = (*ptr_);
Storage updated = Storage((original & kUpdateMask) | new_bits);
*ptr_ = updated;
#endif
return *this;
}
////
/// Unpacks an element from memory
CUTLASS_HOST_DEVICE
operator Element() const {