
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.
198 lines
5.9 KiB
C++
198 lines
5.9 KiB
C++
/***************************************************************************************************
|
|
* 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 math function
|
|
*/
|
|
|
|
#pragma once
|
|
|
|
#include <vector>
|
|
#include <string>
|
|
#include <memory>
|
|
#include <unordered_map>
|
|
|
|
// CUTLASS Library includes
|
|
#include "cutlass/library/library.h"
|
|
#include "cutlass/library/manifest.h"
|
|
|
|
// Profiler includes
|
|
#include "options.h"
|
|
#include "device_context.h"
|
|
#include "operation_profiler.h"
|
|
#include "performance_result.h"
|
|
#include "problem_space.h"
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
namespace cutlass {
|
|
namespace profiler {
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
/// Abstract base class for each math function
|
|
class GemmOperationProfiler : public OperationProfiler {
|
|
public:
|
|
|
|
/// Problem structure obtained from problem space
|
|
struct GemmProblem {
|
|
int64_t m;
|
|
int64_t n;
|
|
int64_t k;
|
|
int64_t lda;
|
|
int64_t ldb;
|
|
int64_t ldc;
|
|
std::vector<uint8_t> alpha;
|
|
std::vector<uint8_t> beta;
|
|
int64_t split_k_slices;
|
|
int64_t batch_count;
|
|
|
|
//
|
|
// Methods
|
|
//
|
|
|
|
GemmProblem():
|
|
m(16), n(16), k(16), lda(0), ldb(0), ldc(0), split_k_slices(1), batch_count(1) { }
|
|
};
|
|
|
|
/// Workspace used
|
|
struct GemmWorkspace {
|
|
|
|
DeviceAllocation *A;
|
|
DeviceAllocation *B;
|
|
DeviceAllocation *C;
|
|
DeviceAllocation *Computed;
|
|
DeviceAllocation *Reference;
|
|
|
|
library::GemmConfiguration configuration;
|
|
library::GemmArguments arguments;
|
|
|
|
/// Buffer used for the operation's host workspace
|
|
std::vector<uint8_t> host_workspace;
|
|
|
|
/// Buffer used for the operations' device workspace
|
|
DeviceAllocation device_workspace;
|
|
|
|
//
|
|
// Methods
|
|
//
|
|
|
|
GemmWorkspace():
|
|
A(nullptr), B(nullptr), C(nullptr), Computed(nullptr), Reference(nullptr) { }
|
|
};
|
|
|
|
protected:
|
|
|
|
//
|
|
// Data members
|
|
//
|
|
|
|
/// GEMM problem obtained from problem space
|
|
GemmProblem problem_;
|
|
|
|
/// Device memory allocations
|
|
GemmWorkspace gemm_workspace_;
|
|
|
|
|
|
public:
|
|
//
|
|
// Methods
|
|
//
|
|
|
|
/// Ctor
|
|
GemmOperationProfiler();
|
|
|
|
/// Destructor
|
|
virtual ~GemmOperationProfiler();
|
|
|
|
/// Prints usage statement for the math function
|
|
virtual void print_usage(std::ostream &out) const;
|
|
|
|
/// Prints examples
|
|
virtual void print_examples(std::ostream &out) const;
|
|
|
|
/// Extracts the problem dimensions
|
|
virtual Status initialize_configuration(
|
|
Options const &options,
|
|
PerformanceReport &report,
|
|
DeviceContext &device_context,
|
|
library::Operation const *operation,
|
|
ProblemSpace const &problem_space,
|
|
ProblemSpace::Problem const &problem);
|
|
|
|
/// Initializes workspace
|
|
virtual Status initialize_workspace(
|
|
Options const &options,
|
|
PerformanceReport &report,
|
|
DeviceContext &device_context,
|
|
library::Operation const *operation,
|
|
ProblemSpace const &problem_space,
|
|
ProblemSpace::Problem const &problem);
|
|
|
|
/// Verifies CUTLASS against references
|
|
virtual bool verify_cutlass(
|
|
Options const &options,
|
|
PerformanceReport &report,
|
|
DeviceContext &device_context,
|
|
library::Operation const *operation,
|
|
ProblemSpace const &problem_space,
|
|
ProblemSpace::Problem const &problem);
|
|
|
|
/// Measures performance results
|
|
virtual bool profile(
|
|
Options const &options,
|
|
PerformanceReport &report,
|
|
DeviceContext &device_context,
|
|
library::Operation const *operation,
|
|
ProblemSpace const &problem_space,
|
|
ProblemSpace::Problem const &problem);
|
|
|
|
protected:
|
|
|
|
/// Initializes the performance result
|
|
void initialize_result_(
|
|
PerformanceResult &result,
|
|
Options const &options,
|
|
library::GemmDescription const &operation_desc,
|
|
ProblemSpace const &problem_space);
|
|
|
|
/// Verifies CUTLASS against references
|
|
bool verify_with_cublas_(
|
|
Options const &options,
|
|
PerformanceReport &report,
|
|
DeviceContext &device_context,
|
|
library::Operation const *operation,
|
|
ProblemSpace const &problem_space,
|
|
ProblemSpace::Problem const &problem);
|
|
|
|
};
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
} // namespace profiler
|
|
} // namespace cutlass
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|