CUTLASS 2.0 (#62)

CUTLASS 2.0

Substantially refactored for

- Better performance, particularly for native Turing Tensor Cores
- Robust and durable templates spanning the design space
- Encapsulated functionality embodying modern C++11 programming techniques
- Optimized containers and data types for efficient, generic, portable device code

Updates to:
- Quick start guide
- Documentation
- Utilities
- CUTLASS Profiler

Native Turing Tensor Cores
- Efficient GEMM kernels targeting Turing Tensor Cores
- Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands

Coverage of existing CUTLASS functionality:
- GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs
- Volta Tensor Cores through native mma.sync and through WMMA API
- Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions
- Batched GEMM operations
- Complex-valued GEMMs

Note: this commit and all that follow require a host compiler supporting C++11 or greater.

include/cutlass/tensor_coord.h

@@ -0,0 +1,168 @@
+/***************************************************************************************************
+ * 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:
+ *     * Redistributions of source code must retain the above copyright notice, this list of
+ *       conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright notice, this list of
+ *       conditions and the following disclaimer in the documentation and/or other materials
+ *       provided with the distribution.
+ *     * Neither the name of the NVIDIA CORPORATION nor the names of its contributors may be used
+ *       to endorse or promote products derived from this software without specific prior written
+ *       permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NVIDIA CORPORATION BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
+ * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TOR (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ **************************************************************************************************/
+/*! \file
+    \brief Defines a canonical coordinate for rank=4 tensors offering named indices.
+*/
+#pragma once
+
+#include "cutlass/cutlass.h"
+#include "cutlass/coord.h"
+
+namespace cutlass {
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+/// Defines a canonical 4D coordinate used by tensor operations.
+struct Tensor4DCoord : public Coord<4> {
+
+  /// Base class
+  using Base = Coord<4>;
+
+  /// Index type
+  using Index = typename Base::Index;
+
+  /// LongIndex type
+  using LongIndex = typename Base::LongIndex;
+
+  /// Batch dimension
+  static int const kN = 0;
+
+  /// Height dimension
+  static int const kH = 1;
+
+  /// Width dimension
+  static int const kW = 2;
+
+  /// Channels dimension
+  static int const kC = 3;
+
+  //
+  // Methods
+  //
+
+  /// Default ctor
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord() { }
+
+  /// Constructs from Coord<4>
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord(Coord<4> const &coord): Base(coord) { }
+
+  /// Helper to construct from N, H, W, and C.
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord(Index n, Index h, Index w, Index c): Base(make_Coord(n, h, w, c)) { }
+
+  /// Returns the batch of the coordinate
+  CUTLASS_HOST_DEVICE
+  Index const & n() const { return this->at(kN); }
+
+  /// Returns the batch of the coordinate
+  CUTLASS_HOST_DEVICE
+  Index & n() { return this->at(kN); }
+
+  /// Returns the row of the coordinate
+  CUTLASS_HOST_DEVICE
+  Index const & h() const { return this->at(kH); }
+
+  /// Returns the row of the coordinate
+  CUTLASS_HOST_DEVICE
+  Index & h() { return this->at(kH); }
+
+  /// Returns the column of the coordinate
+  CUTLASS_HOST_DEVICE
+  Index const & w() const { return this->at(kW); }
+
+  /// Returns the column of the coordinate
+  CUTLASS_HOST_DEVICE
+  Index & w() { return this->at(kW); }
+
+  /// Returns the channel of the coordinate
+  CUTLASS_HOST_DEVICE
+  Index const & c() const { return this->at(kC); }
+
+  /// Returns the channel of the coordinate
+  CUTLASS_HOST_DEVICE
+  Index & c() { return this->at(kC); }
+
+  //
+  // Coord operators
+  //
+
+  /// Element-wise addition
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord operator+(Base const& b) const {
+    return Tensor4DCoord(Base::operator+(b));
+  }
+
+  /// Element-wise subtraction
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord operator-(Base const& b) const {
+    return Tensor4DCoord(Base::operator-(b));
+  }
+
+  /// Element-wise multiplication
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord operator*(Base const& b) const {
+    return Tensor4DCoord(Base::operator*(b));
+  }
+
+  /// Element-wise division
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord operator/(Base const& b) const {
+    return Tensor4DCoord(Base::operator/(b));
+  }
+
+  /// In-place addition
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord& operator+=(Base const& b) {
+    Base::operator+=(b);
+    return *this;
+  }
+
+  /// In-place subtraction
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord& operator-=(Base const& b) {
+    Base::operator-=(b);
+    return *this;
+  }
+
+  /// In-place multiplication
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord& operator*=(Base const& b) {
+    Base::operator*=(b);
+    return *this;
+  }
+
+  /// In-place division
+  CUTLASS_HOST_DEVICE
+  Tensor4DCoord& operator/=(Base const& b) {
+    Base::operator/=(b);
+    return *this;
+  }
+};
+
+////////////////////////////////////////////////////////////////////////////////////////////////////
+
+} // namespace cutlass