Files
composable_kernel/profiler/src/profile_grouped_conv_bwd_data_tile.cpp
Ville Pietilä 78d657c4f7 [rocm-libraries] ROCm/rocm-libraries#7284 (commit e7d25b2)
[CK_TILE] Integrate CK Tile Dispatcher code generation into
 CK Tile Profiler (#7284)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## Motivation

CK Tile is going to be delivered to hipDNN via CK Dispatcher. Currently
the CK Tile Profiler using CK Builder for generating the profiled
instances from the configuration files that identify the instances that
old CK exposes. We need to replace this instance generation with the CK
Tile Dispatcher codegen.

## Technical Details
The old CK Profiler config files are converted to JSON files that the CK
Tile Dispatcher can digest. The conversion script for configurations is
stored to source control in case we need to update the JSON
configurations later. The dispatcher generates instance libraries per
conv direction (fwd, bwd data, and bwd weight) that are linked to the CK
Profiler executable. I also implemented codegne for the stream-K and
depthwise conv instances. The proposed solution replaces the CK Builder
codegen with the CK Tile Dispatcher codegen.

There are two new methods that are exposed via the dispatcher backend

- `is_supported` - required to enabled the profiler workflow where we
check the applicability of the kernel instance before running it.
- `get_instance_string` - this mainly for verification. This provide the
CK Builder instance string for verifying that the old CK Builder based
profiler and the new CK Tile Dispatcher based profiler have the same
instances.

The rules that limit the generated instances are now collected to a
single location under the dispacther. The CK Builder codegen uses these,
which ensures that the two codegen pipelines are in sync. The next step
(different PR) is to remove the CK Builder codegen pipeline altogether.

## Test Plan

Verified that the old CK Builder based profiler and the new CK Tile
Dispatcher based profiler have the same instances, that is, the
Dispatcher based codgen can generate the same instances as the old CK
Builder.

## Submission Checklist

- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
2026-05-28 21:03:37 +00:00

233 lines
8.6 KiB
C++

// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <string>
#include "ck_tile/builder/testing/conv/ck_tile.hpp"
#include "ck_tile/host/device_prop.hpp"
#ifdef CK_TILE_DISPATCHER
#include "profiler/grouped_convolution_backward_data_tile_dispatcher_algs.hpp"
#else
#include "profiler/grouped_convolution_backward_data_tile_algs.hpp"
#endif
#include "profiler/tile_profiler_utils.hpp"
#include "profiler/profiler_arg_utils.hpp"
#include "profiler_operation_registry.hpp"
namespace {
enum struct ConvLayout
{
GNHWC_GKYXC_GNHWK, // 0
NHWGC_GKYXC_NHWGK, // 1
NGCHW_GKYXC_NGKHW, // 2
NGCHW_GKCYX_NGKHW, // 3
};
enum struct ConvDataType
{
F32_F32_F32, // 0
F16_F16_F16, // 1
BF16_BF16_BF16, // 2
F32_F32_F32_TF32, // 3
};
#define OP_NAME "grouped_conv_bwd_data_tile"
#define OP_DESC "Grouped Convolution Backward Data (CK Tile)"
static void print_helper_msg()
{
std::cout
// clang-format off
<< "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
<< "arg2: data type (0: Output fp32, Weight fp32, Input fp32\n"
<< " 1: Output fp16, Weight fp16, Input fp16\n"
<< " 2: Output bf16, Weight bf16, Input bf16\n"
<< " 3: Output fp32, Weight fp32, Input fp32, Compute tf32)\n"
<< "arg3: tensor layout (0: Output[G, N, Ho, Wo, C], Weight[G, K, Y, X, C], Input[G, N, Hi, Wi, K]\n"
<< " 1: Output[N, Ho, Wo, G, C], Weight[G, K, Y, X, C], Input[N, Hi, Wi, G, K])\n"
<< " 2: Output[N, G, C, Ho, Wo], Weight[G, K, Y, X, C], Input[N, G, K, Hi, Wi])\n"
<< " 3: Output[N, G, C, Ho, Wo], Weight[G, K, C, Y, X], Input[N, G, K, Hi, Wi])\n"
<< "arg4: verification (0: no, 1: yes)\n"
<< "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
<< "arg6: print tensor value (0: no; 1: yes)\n"
<< "arg7: time kernel (0: no, 1: yes)\n"
<< ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl
<< "Last argument: split-K (0: internally computed split-K value; 1, 2, 4, 8, 16, 32, 64, 128: set k batches explicitly)\n"
<< "\nOptional arguments:\n"
<< " --instance <id> Run only the specified instance (0-indexed among valid instances)\n";
// clang-format on
}
namespace ckb = ck_tile::builder;
namespace ckt = ck_tile::builder::test;
namespace ckp = ck_tile::builder::profiling;
template <auto SIGNATURE>
int call_profiler(const ckt::Args<SIGNATURE>& args,
const std::string& split_k,
bool do_verification,
bool time_kernel,
ck_tile::index_t instance_index)
{
auto inputs = ckt::alloc_inputs(args);
auto outputs = ckt::alloc_outputs(args);
ckt::init_inputs(args, inputs.get());
std::cout << args.make_input_descriptor() << std::endl;
std::cout << args.make_weight_descriptor() << std::endl;
std::cout << args.make_output_descriptor() << std::endl;
auto&& [valid, avg_time, op_name, best_split_k, best_instance_index] =
ckp::run_grouped_conv_backward_data_tile_algs(
args,
split_k,
instance_index,
inputs.get(),
outputs.get(),
ck_tile::stream_config{nullptr,
time_kernel,
0 /*log_level*/,
5 /*cold_iters*/,
50 /*nrepeat_*/,
true /*is_gpu_timer_*/,
time_kernel /*flush_cache*/},
do_verification);
if(time_kernel)
{
std::cout << "\nBest configuration parameters:" << "\n\tname: " << op_name << " (instance "
<< best_instance_index << ")" << "\n\tavg_time: " << avg_time << ", SplitK "
<< best_split_k << std::endl;
}
return !valid;
}
} // namespace
int profile_grouped_conv_bwd_data_tile(int argc, char* argv[])
{
// Parse optional named arguments first
ck_tile::index_t instance_index = -1;
bool dummy;
ck::profiler::parse_named_args(argc, argv, instance_index, dummy);
const int named_arg_count = ck::profiler::count_named_args(argc, argv);
// Adjust argc for positional argument checking
const int positional_argc = argc - named_arg_count;
// 8 for control, 1 for num_dim_spatial
if(positional_argc < 9)
{
print_helper_msg();
return 1;
}
const auto data_type = static_cast<ConvDataType>(std::stoi(argv[2]));
const auto layout = static_cast<ConvLayout>(std::stoi(argv[3]));
const bool do_verification = std::stoi(argv[4]);
const bool time_kernel = std::stoi(argv[7]);
const int num_dim_spatial = std::stoi(argv[8]);
// 8 for control, 1 for num_dim_spatial, 4 for G/N/K/C, and 6 * num_dim_spatial, 1 for split-K
if(positional_argc != 8 + 1 + 4 + 6 * num_dim_spatial + 1)
{
print_helper_msg();
return 1;
}
constexpr ck_tile::index_t conv_params_start_idx = 9;
const auto params =
ck::utils::conv::parse_conv_param(num_dim_spatial, conv_params_start_idx, argv);
std::cout << params << std::endl;
auto split_k = std::string(argv[8 + 1 + 4 + 6 * num_dim_spatial]);
// The bwd data profiler in old CK uses -1 to loop over all split-K values.
// We want to have the same API for backward compatibility, but we need to convert it to "all"
// for the new API.
if(split_k == "-1")
{
split_k = "all";
}
if(layout == ConvLayout::NHWGC_GKYXC_NHWGK)
{
if(num_dim_spatial == 2)
{
if(data_type == ConvDataType::F16_F16_F16)
{
constexpr auto SIGNATURE = ckp::SIGNATURE_NHWGC_FP16_BWD_DATA;
return call_profiler<SIGNATURE>(
ckp::parse_conv_args<SIGNATURE>(conv_params_start_idx, argv),
split_k,
do_verification,
time_kernel,
instance_index);
}
else if(data_type == ConvDataType::BF16_BF16_BF16)
{
constexpr auto SIGNATURE = ckp::SIGNATURE_NHWGC_BF16_BWD_DATA;
return call_profiler<SIGNATURE>(
ckp::parse_conv_args<SIGNATURE>(conv_params_start_idx, argv),
split_k,
do_verification,
time_kernel,
instance_index);
}
else if(data_type == ConvDataType::F32_F32_F32)
{
constexpr auto SIGNATURE = ckp::SIGNATURE_NHWGC_FP32_BWD_DATA;
return call_profiler<SIGNATURE>(
ckp::parse_conv_args<SIGNATURE>(conv_params_start_idx, argv),
split_k,
do_verification,
time_kernel,
instance_index);
}
}
else if(num_dim_spatial == 3)
{
if(data_type == ConvDataType::F16_F16_F16)
{
constexpr auto SIGNATURE = ckp::SIGNATURE_NDHWGC_FP16_BWD_DATA;
return call_profiler<SIGNATURE>(
ckp::parse_conv_args<SIGNATURE>(conv_params_start_idx, argv),
split_k,
do_verification,
time_kernel,
instance_index);
}
else if(data_type == ConvDataType::BF16_BF16_BF16)
{
constexpr auto SIGNATURE = ckp::SIGNATURE_NDHWGC_BF16_BWD_DATA;
return call_profiler<SIGNATURE>(
ckp::parse_conv_args<SIGNATURE>(conv_params_start_idx, argv),
split_k,
do_verification,
time_kernel,
instance_index);
}
else if(data_type == ConvDataType::F32_F32_F32)
{
constexpr auto SIGNATURE = ckp::SIGNATURE_NDHWGC_FP32_BWD_DATA;
return call_profiler<SIGNATURE>(
ckp::parse_conv_args<SIGNATURE>(conv_params_start_idx, argv),
split_k,
do_verification,
time_kernel,
instance_index);
}
}
}
std::cout << "this data_type & layout is not implemented" << std::endl;
return 1;
}
REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_grouped_conv_bwd_data_tile);