leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 28 Packages Projects Releases Wiki Activity

Cutlass 1.3 Release (#42)

Browse Source 
CUTLASS 1.3 Release
- Efficient GEMM kernel targeting Volta Tensor Cores via mma.sync instruction added in CUDA 10.1.
This commit is contained in:
Andrew Kerr
2019-03-20 10:49:17 -07:00
committed by GitHub
parent 19a9d64e3c
commit 877bdcace6
256 changed files with 16930 additions and 802 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/util/command_line.h
+1 -1
View File
@@ -1,5 +1,5 @@
/******************************************************************************
* Copyright (c) 2011-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2011-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are not permitted.
tools/util/device_memory.h
+1 -1
View File
@@ -1,5 +1,5 @@
/******************************************************************************
* Copyright (c) 2011-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2011-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are not permitted.
tools/util/distribution.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/exceptions.h
+1 -1
View File
@@ -1,5 +1,5 @@
/******************************************************************************
* Copyright (c) 2011-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2011-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are not permitted.
tools/util/half.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/host_matrix.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/host_matrix_view.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/host_tensor.h
+14 -14
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
@@ -114,29 +114,29 @@ class HostTensor : public HostTensorView<
typedef typename DeviceTensorView::ConstTensorView ConstDeviceTensorView;
/// Tensor reference to host memory
typedef typename Base::TensorRef TensorRef;
typedef typename Base::TensorRef_t TensorRef_t;
/// Tensor view to host memory
typedef TensorView<
typedef HostTensorView<
typename TypeTraits<T>::host_type,
Rank_,
MapFunc_,
StorageRank_,
Index_,
LongIndex_> HostTensorView;
LongIndex_> HostTensorView_t;
/// Tensor view to host memory
typedef typename HostTensorView::ConstTensorView ConstHostTensorView;
typedef typename HostTensorView_t::ConstTensorView ConstHostTensorView;
/// Coordinate in logical tensor space
typedef typename TensorRef::TensorCoord TensorCoord;
typedef typename TensorRef_t::TensorCoord TensorCoord;
/// Coordinate in storage n-D array
typedef typename TensorRef::StorageCoord StorageCoord;
typedef typename TensorRef_t::StorageCoord StorageCoord;
/// Stride vector in storage coordinate space
/// Least significant stride is = 1 and not stored
typedef typename TensorRef::StrideVector StrideVector;
typedef typename TensorRef_t::StrideVector StrideVector;
/// Rank of internal storage.
static int const kStorageRank = Base::kStorageRank;
@@ -216,14 +216,14 @@ class HostTensor : public HostTensorView<
host_.resize(_capacity);
device_.reset(_device_memory, _capacity);
Base::reset(TensorRef(host_.data(), stride), size);
Base::reset(TensorRef_t(host_.data(), stride), size);
}
/// Accesses the tensor reference pointing to data
TensorRef host_ref() { return Base::ref(); }
TensorRef_t host_ref() { return Base::ref(); }
/// Accesses the tensor reference pointing to data
TensorRef host_ref() const { return Base::ref(); }
TensorRef_t host_ref() const { return Base::ref(); }
/// Accesses the tensor reference pointing to data
DeviceTensorRef device_ref() const {
@@ -231,13 +231,13 @@ class HostTensor : public HostTensorView<
}
/// Accesses the tensor reference pointing to data
HostTensorView host_view() {
return HostTensorView(host_data(), this->stride(), this->size());
HostTensorView_t host_view() {
return HostTensorView_t(host_data(), this->stride(), this->size());
}
/// Accesses the tensor reference pointing to data
ConstHostTensorView host_view() const {
return HostTensorView(host_data(), this->stride(), this->size());
return HostTensorView_t(host_data(), this->stride(), this->size());
}
/// Accesses the tensor reference pointing to data
tools/util/host_tensor_view.h
+15 -15
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
@@ -61,21 +61,21 @@ class HostTensorView :
/// Storage type
typedef typename Base::Storage Storage;
/// Alias for underlying TensorRef
typedef typename Base::TensorRef TensorRef;
/// Alias for underlying TensorRef_t
typedef typename Base::TensorRef_t TensorRef_t;
/// Index type
typedef typename Base::Index Index;
/// Coordinate in logical tensor space
typedef typename TensorRef::TensorCoord TensorCoord;
typedef typename TensorRef_t::TensorCoord TensorCoord;
/// Coordinate in storage n-D array
typedef typename TensorRef::StorageCoord StorageCoord;
typedef typename TensorRef_t::StorageCoord StorageCoord;
/// Stride vector in storage coordinate space
/// Least significant stride is = 1 and not stored
typedef typename TensorRef::StrideVector StrideVector;
typedef typename TensorRef_t::StrideVector StrideVector;
/// Long index type for pointer offsets
typedef typename Base::LongIndex LongIndex;
@@ -121,18 +121,18 @@ class HostTensorView :
Storage_ *_ptr,
StrideVector const &_stride,
TensorCoord const& _size
) : Base(TensorRef(_ptr, _stride), _size) {}
) : Base(TensorRef_t(_ptr, _stride), _size) {}
/// Helper to construct from pointer, stride, and size
HostTensorView(
Storage_ *_ptr,
StorageCoord const &_stride,
TensorCoord const& _size
) : Base(TensorRef(_ptr, _stride), _size) {}
) : Base(TensorRef_t(_ptr, _stride), _size) {}
/// Constructs a Tensor_view from a TensorRef and size assuming dense packing
/// Constructs a Tensor_view from a TensorRef_t and size assuming dense packing
HostTensorView(
TensorRef const& _ref,
TensorRef_t const& _ref,
TensorCoord const& _size) : Base(_ref, _size) {}
/// Assigns a tensor view
@@ -149,22 +149,22 @@ class HostTensorView :
return result;
}
/// Returns a TensorRef offset by a given amount
/// Returns a TensorRef_t offset by a given amount
CUTLASS_HOST_DEVICE
HostTensorView& operator+=(TensorCoord const& b) {
this->add_pointer_offset(this->offset(b));
return *this;
}
/// Returns a TensorRef offset by a given amount
/// Returns a TensorRef_t offset by a given amount
CUTLASS_HOST_DEVICE
HostTensorView operator-(TensorCoord const& b) const {
TensorRef result(*this);
TensorRef_t result(*this);
result.add_pointer_offset(-this->offset(b));
return result;
}
/// Returns a TensorRef offset by a given amount
/// Returns a TensorRef_t offset by a given amount
CUTLASS_HOST_DEVICE
HostTensorView& operator-=(TensorCoord const& b) {
this->add_pointer_offset(-this->offset(b));
@@ -474,7 +474,7 @@ class HostTensorView :
void operator()(Storage const& element) {
double value(element);
double conj(element); // TODO - conjugates for complex
double conj(element);
sum += value * conj;
}
tools/util/reference/detail/inner_product.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/gemm.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/kernel/gemm.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/kernel/split_complex_gemm.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/kernel/tensor_elementwise.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/kernel/tensor_foreach.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/split_complex_gemm.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/tensor_elementwise.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/tensor_foreach.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/thread/gemm.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/device/thread/split_complex_gemm.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/host/gemm.h
+2 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
@@ -219,6 +219,7 @@ void BatchedGemm(
////////////////////////////////////////////////////////////////////////////////////////////////////
} // namespace host
} // namespace reference
} // namespace cutlass
tools/util/reference/host/split_complex_gemm.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/host/tensor_elementwise.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/reference/host/tensor_foreach.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/tensor_view_io.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2018-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
tools/util/type_traits.h
+1 -1
View File
@@ -1,5 +1,5 @@
/***************************************************************************************************
* Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met: