ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-02 20:51:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_BUILDER] Add grouped conv fwd ck tile profiler (#3518)

Browse Source 
* [BULDER] Add grouped conv fwd ck tile profiler

* [CK TILE] Fix grouped conv kernels splitk and double lds

* Updates

* Fixes

* Move to ckProfiler

* Fixes

* fix

* fix

* Change instances to empty list by default

* fix

* fix

* Update grouped_convolution_signatures.hpp

* Update grouped_convolution_forward_tile_algs.hpp

* [CK TILE] Add grouped convolution forward tests (#3556)

* [CK TILE] Add grouped convolution forward tests

* fix jenkins

* fixes

* comments fixes

* unit test

* unit test fix

* Move instances outside builder

* fix includes

* clang format fix

* readme fix

* fix includes

* fixes
This commit is contained in:
Bartłomiej Kocot
2026-01-20 06:29:01 +01:00
committed by GitHub
parent 0517d43d31
commit 0727e85e52
44 changed files with 3083 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
experimental/grouped_convolution_tile_instances/instances/grouped_convolution_forward_tile.cpp.in Normal file
View File

@@ -0,0 +1,19 @@
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT
#include "../instance_includes.inc"
namespace ck_tile::builder::profiling {
constexpr auto SIGNATURE = gen_signature;
std::tuple<bool, float, std::string> run_gen_instance_name(const ckt::Args<SIGNATURE>& args,
const ckt::Inputs<SIGNATURE>& inputs,
const ckt::Outputs<SIGNATURE>& outputs,
const ck_tile::stream_config& s_conf)
{
constexpr auto ALGORITHM = cku::ConvAlgorithm_Tile_GroupedConvolutionKernel{}
.with_tile_specializations(gen_specialization)
.with_tile_thread_block(gen_thread_block)
.with_tile_block_gemm(gen_block_gemm_desc)
.with_tile_transfer(gen_block_transfer)
.with_tile_optimizations(gen_optimizations);
#include "../instance_run.inc"
}
} // namespace ck_tile::builder::profiling

64
experimental/grouped_convolution_tile_instances/instances/instance_includes.inc Normal file
View File

@@ -0,0 +1,64 @@
#include "../../builder/test/utils/ckb_conv_tile_test_configs.hpp"
#include "ck_tile/builder/testing/conv_fwd_ck_tile.hpp"
namespace ckb = ck_tile::builder;
namespace ckt = ck_tile::builder::test;
namespace cku = ck_tile::builder::test_utils;
namespace ck_tile::builder::profiling {
constexpr auto SIGNATURE_NHWGC_FP32_FWD =
ckt::ConvSignature{.spatial_dim = 2,
.direction = ckb::ConvDirection::FORWARD,
.data_type = ckb::DataType::FP32,
.accumulation_data_type = ckb::DataType::FP32,
.input = {.config = {.layout = ckb::TensorLayout::NHWGC}},
.weight = {.config = {.layout = ckb::TensorLayout::GKYXC}},
.output = {.config = {.layout = ckb::TensorLayout::NHWGK}}};
constexpr auto SIGNATURE_NHWGC_BF16_FWD =
ckt::ConvSignature{.spatial_dim = 2,
.direction = ckb::ConvDirection::FORWARD,
.data_type = ckb::DataType::BF16,
.accumulation_data_type = ckb::DataType::FP32,
.input = {.config = {.layout = ckb::TensorLayout::NHWGC}},
.weight = {.config = {.layout = ckb::TensorLayout::GKYXC}},
.output = {.config = {.layout = ckb::TensorLayout::NHWGK}}};
constexpr auto SIGNATURE_NHWGC_FP16_FWD =
ckt::ConvSignature{.spatial_dim = 2,
.direction = ckb::ConvDirection::FORWARD,
.data_type = ckb::DataType::FP16,
.accumulation_data_type = ckb::DataType::FP32,
.input = {.config = {.layout = ckb::TensorLayout::NHWGC}},
.weight = {.config = {.layout = ckb::TensorLayout::GKYXC}},
.output = {.config = {.layout = ckb::TensorLayout::NHWGK}}};
constexpr auto SIGNATURE_NDHWGC_FP32_FWD =
ckt::ConvSignature{.spatial_dim = 3,
.direction = ckb::ConvDirection::FORWARD,
.data_type = ckb::DataType::FP32,
.accumulation_data_type = ckb::DataType::FP32,
.input = {.config = {.layout = ckb::TensorLayout::NDHWGC}},
.weight = {.config = {.layout = ckb::TensorLayout::GKZYXC}},
.output = {.config = {.layout = ckb::TensorLayout::NDHWGK}}};
constexpr auto SIGNATURE_NDHWGC_BF16_FWD =
ckt::ConvSignature{.spatial_dim = 3,
.direction = ckb::ConvDirection::FORWARD,
.data_type = ckb::DataType::BF16,
.accumulation_data_type = ckb::DataType::FP32,
.input = {.config = {.layout = ckb::TensorLayout::NDHWGC}},
.weight = {.config = {.layout = ckb::TensorLayout::GKZYXC}},
.output = {.config = {.layout = ckb::TensorLayout::NDHWGK}}};
constexpr auto SIGNATURE_NDHWGC_FP16_FWD =
ckt::ConvSignature{.spatial_dim = 3,
.direction = ckb::ConvDirection::FORWARD,
.data_type = ckb::DataType::FP16,
.accumulation_data_type = ckb::DataType::FP32,
.input = {.config = {.layout = ckb::TensorLayout::NDHWGC}},
.weight = {.config = {.layout = ckb::TensorLayout::GKZYXC}},
.output = {.config = {.layout = ckb::TensorLayout::NDHWGK}}};
} // namespace ck_tile::builder::profiling

9
experimental/grouped_convolution_tile_instances/instances/instance_run.inc Normal file
View File

@@ -0,0 +1,9 @@
using Builder = ckb::ConvBuilder<SIGNATURE, ALGORITHM>;
using Instance = Builder::Instance;
auto conv = Instance{};
bool is_supported;
float avg_time;
std::tie(is_supported, avg_time) = ckt::run(conv, args, inputs, outputs, s_conf);
return std::make_tuple(is_supported, avg_time, conv.GetInstanceString());