Files
composable_kernel/profiler/src/profile_transpose.cpp
arai713 aa3e2d7967 Transpose profiler fix (#1114)
* added working example for 5D input using 1D kernel

* example with 5D input tensor and 2d kernel - not working: issues with arguments

* added updated version of 3d device op - changed descriptors/dims

* added example file to check kernel

* fixed descriptor and isSupportedArgument stride problem

* added and modified kernel for 3d - updated tids/loop

* adding some more 5d example files

* fixed some issues

* changes made for testing

* working version: fixed error in stride for A, still a bit inefficient

* cleaned up formatting/comments

* updating formatting

* more formatting fixes

* fixing cmake, adding back gpu targets in cmake script

* adding client example

* added instances for client example

* fixed errors in client example

* implemented client ex with device_elementwise.hpp and device_elementwise_3d_impl.hpp

* removed extra files

* minor formatting and naming fixes

* adding test files and profiler

* fixing minor error

* minor fix

* removed unneccesary comments, renamed files

* updated instance list for client example, added different layout example

* removing instances

* fixed error in instance generation

* remove comments

* update profiler and client example tensor layouts

* fixed errors in test/profiler

* updated vector dim access to enable vector load

* updated test/profiler files

* updated example with 1d kernel

* updating profiler

* renamed files

* disabled device op for MI300

* skip  elementwise_permute_2d on gfx94x

* Update CMakeLists.txt

* fixing CMake - disabling some GPU targets

* added transpose profiler to CMake

* fixed transpose profiler errors

* fixed instances for tests/profiler

* cleaned up code in transpose profiler source code

* added some comments, updated copyright

* made function arguments const where possible

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
Co-authored-by: Jing Zhang <jizhan@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
2024-01-04 10:33:19 -06:00

113 lines
3.2 KiB
C++

// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include "profiler/profile_transpose_impl.hpp"
#include "profiler_operation_registry.hpp"
enum struct DataType
{
F32_F32_F32_F32_F32, // 0
F16_F16_F16_F16_F16, // 1
};
#define OP_NAME "transpose"
#define OP_DESC "Transpose"
struct TransposeArgParser
{
std::unordered_map<std::string, std::vector<int>> long_opts = {{"lengths", {}}};
bool parse_opt(const int argc, char* argv[], const std::string& key, int i)
{
if(std::string("--") + key == argv[i])
{
const int pos = i;
while(++i < argc && argv[i][0] != '-') {}
int end = i;
for(int j = pos + 1; j < end; j++)
{
long_opts[key].push_back(std::stoi(argv[j]));
}
return true;
}
return false;
}
void operator()(int argc, char* argv[])
{
for(auto& kv : long_opts)
{
for(int i = 1; i < argc; i++)
{
if(parse_opt(argc, argv, kv.first, i))
break;
}
}
}
};
static void print_helper_msg()
{
printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
printf("arg2: data type (0: fp32; 1: fp16)\n");
printf("arg3: verification (0: no; 1: yes)\n");
printf("arg4: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg5: print tensor value (0: no; 1: yes)\n");
printf("arg6: time kernel (0=no, 1=yes)\n");
printf("arg7: --lengths: N, C, D, H, W\n");
}
int profile_transpose(int argc, char* argv[])
{
if(argc != 7)
{
print_helper_msg();
exit(1);
}
TransposeArgParser arg_parser;
const auto data_type = static_cast<DataType>(std::stoi(argv[2]));
const bool do_verification = std::stoi(argv[3]);
const int init_method = std::stoi(argv[4]);
const bool do_log = std::stoi(argv[5]);
const bool time_kernel = std::stoi(argv[6]);
arg_parser(argc, argv);
const std::vector<ck::index_t> lengths = arg_parser.long_opts["lengths"];
using F32 = float;
using F16 = ck::half_t;
auto profile = [&](auto a_type, auto b_type) {
using ADataType = decltype(a_type);
using BDataType = decltype(b_type);
constexpr ck::index_t NumDim = 5;
bool pass = ck::profiler::profile_transpose_impl<ADataType, BDataType, NumDim>(
do_verification, init_method, do_log, time_kernel, lengths);
return pass ? 0 : 1;
};
if(data_type == DataType::F32_F32_F32_F32_F32)
{
return profile(F32{}, F32{});
}
else if(data_type == DataType::F16_F16_F16_F16_F16)
{
return profile(F16{}, F16{});
}
else
{
std::cout << "this data_type & layout is not implemented" << std::endl;
return 1;
}
}
REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_transpose);