mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-06-29 11:16:59 +00:00
e826b2eb7e
=?UTF-8?q?[CK=20TILE]=20Unification=20Work=20=E2=80=93=20?= =?UTF-8?q?Add=20MFMA=20specialisations=20for=20`tf32=5Ft`=20(#6768)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit ## Motivation This PR adds two specialisations related to `tf32_t`. ## Technical Details This change treats `tf32_t` as a concrete type rather than an empty `struct`. It also adds two new specialisations for MFMA dense builtins and resolves existing circular include issues. ## Test Plan All the new wrappers were added to the test suite in test_amdgcn_mma_layout.inc. ## Test Result Test should pass. ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
182 lines
8.2 KiB
C++
182 lines
8.2 KiB
C++
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
|
// SPDX-License-Identifier: MIT
|
|
#pragma once
|
|
#include "gemm_utils.hpp"
|
|
|
|
struct BasicInvoker
|
|
{
|
|
template <typename GemmConfig,
|
|
typename ADataType,
|
|
typename BDataType,
|
|
typename DsDataType,
|
|
typename AccDataType,
|
|
typename CDataType,
|
|
typename ALayout,
|
|
typename BLayout,
|
|
typename DsLayout,
|
|
typename CLayout,
|
|
bool Persistent,
|
|
typename CDEElementWise>
|
|
static float gemm(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config& s)
|
|
{
|
|
if constexpr(std::is_same_v<ADataType, ck_tile::tf32_t>)
|
|
{
|
|
static_assert(std::is_same_v<ADataType, BDataType>,
|
|
"ADataType and BDataType must be the same");
|
|
}
|
|
|
|
if constexpr(Persistent)
|
|
{
|
|
std::cout << "WARNING: Ignoring persistent kernel option for basic gemm." << std::endl;
|
|
}
|
|
|
|
constexpr bool is_fp32_or_tf32_input =
|
|
std::is_same_v<ADataType, float> || std::is_same_v<ADataType, ck_tile::tf32_t>;
|
|
constexpr bool is_tf32_compute = std::is_same_v<ADataType, ck_tile::tf32_t>;
|
|
|
|
// This part comes from the Codegen
|
|
constexpr ck_tile::index_t M_Tile = is_fp32_or_tf32_input ? 128 : 256;
|
|
constexpr ck_tile::index_t N_Tile = is_fp32_or_tf32_input ? 128 : 256;
|
|
constexpr ck_tile::index_t K_Tile = 64;
|
|
|
|
#if CK_TILE_USE_WMMA
|
|
constexpr ck_tile::index_t M_Warp = 4;
|
|
constexpr ck_tile::index_t N_Warp = 2;
|
|
constexpr ck_tile::index_t K_Warp = 1;
|
|
|
|
constexpr ck_tile::index_t M_Warp_Tile = 16;
|
|
constexpr ck_tile::index_t N_Warp_Tile = 16;
|
|
constexpr ck_tile::index_t K_Warp_Tile =
|
|
ck_tile::get_k_warp_tile<ADataType, M_Warp_Tile, true>();
|
|
ck_tile::ignore = is_tf32_compute;
|
|
#else
|
|
// gfx950: fp32 uses 16x16x16 tile (native MFMA)
|
|
// tf32 uses 32x32x16 tile (3x bf16 32x32x16 MFMA emulation)
|
|
constexpr ck_tile::index_t M_Warp = (is_fp32_or_tf32_input && !is_tf32_compute) ? 4 : 2;
|
|
constexpr ck_tile::index_t N_Warp = (is_fp32_or_tf32_input && !is_tf32_compute) ? 4 : 2;
|
|
constexpr ck_tile::index_t K_Warp = 1;
|
|
|
|
constexpr ck_tile::index_t M_Warp_Tile =
|
|
(is_fp32_or_tf32_input && !is_tf32_compute) ? 16 : 32;
|
|
constexpr ck_tile::index_t N_Warp_Tile =
|
|
(is_fp32_or_tf32_input && !is_tf32_compute) ? 16 : 32;
|
|
constexpr ck_tile::index_t K_Warp_Tile = 16;
|
|
#endif
|
|
|
|
using CodegenGemmShape =
|
|
ck_tile::TileGemmShape<ck_tile::sequence<M_Tile, N_Tile, K_Tile>,
|
|
ck_tile::sequence<M_Warp, N_Warp, K_Warp>,
|
|
ck_tile::sequence<M_Warp_Tile, N_Warp_Tile, K_Warp_Tile>>;
|
|
|
|
using TilePartitioner = ck_tile::GemmTile1DPartitioner<CodegenGemmShape>;
|
|
|
|
using CodegenGemmTraits = ck_tile::TileGemmTraits<GemmConfig::kPadM,
|
|
GemmConfig::kPadN,
|
|
GemmConfig::kPadK,
|
|
ALayout,
|
|
BLayout,
|
|
CLayout>;
|
|
|
|
using AComputeDataType =
|
|
std::conditional_t<std::is_same_v<ADataType, ck_tile::pk_int4_t>, BDataType, ADataType>;
|
|
using BComputeDataType =
|
|
std::conditional_t<std::is_same_v<BDataType, ck_tile::pk_int4_t> ||
|
|
std::is_same_v<BDataType, ck_tile::pk_fp4_raw_t>,
|
|
ADataType,
|
|
BDataType>;
|
|
using CodegenPipelineProblem = ck_tile::GemmPipelineProblem<ADataType,
|
|
BDataType,
|
|
AccDataType,
|
|
CodegenGemmShape,
|
|
CodegenGemmTraits,
|
|
AComputeDataType,
|
|
BComputeDataType>;
|
|
|
|
using CodegenGemmPipeline = ck_tile::GemmPipelineAGmemBGmemCRegV1<CodegenPipelineProblem>;
|
|
|
|
using GemmEpilogue = ck_tile::CShuffleEpilogue<
|
|
ck_tile::CShuffleEpilogueProblem<ADataType,
|
|
BDataType,
|
|
ck_tile::tuple<>,
|
|
AccDataType,
|
|
CDataType,
|
|
ck_tile::tuple<>,
|
|
CLayout,
|
|
ck_tile::element_wise::PassThrough,
|
|
TilePartitioner::MPerBlock,
|
|
TilePartitioner::NPerBlock,
|
|
M_Warp,
|
|
N_Warp,
|
|
M_Warp_Tile,
|
|
N_Warp_Tile,
|
|
K_Warp_Tile,
|
|
CodegenPipelineProblem::TransposeC>>;
|
|
|
|
// ToDo: Will add the codegen part to test different pipeline policies in GEMM.
|
|
// Now we only use the BlockGemmASmemBSmemCRegV1DefaultPolicy.
|
|
using Kernel = ck_tile::GemmKernel<TilePartitioner, CodegenGemmPipeline, GemmEpilogue>;
|
|
auto kargs = Kernel::MakeKernelArgs(args);
|
|
|
|
const dim3 grids = Kernel::GridSize(args.M, args.N, args.k_batch);
|
|
const dim3 blocks = Kernel::BlockSize();
|
|
|
|
if(!Kernel::IsSupportedArgument(kargs))
|
|
{
|
|
throw std::runtime_error("Wrong! Arguments not supported! Skipping gemm!\n");
|
|
}
|
|
|
|
if(s.log_level_ > 0)
|
|
{
|
|
std::cout << "Launching kernel with args: " << Kernel::GetName() << '\n'
|
|
<< "shape: " << CodegenGemmShape::GetName() << '\n'
|
|
<< "problem: " << CodegenPipelineProblem::GetName() << '\n'
|
|
<< "pipeline: " << CodegenGemmPipeline::GetName() << '\n'
|
|
<< "grid: {" << grids.x << ", " << grids.y << ", " << grids.z << "}"
|
|
<< ", blocks: {" << blocks.x << ", " << blocks.y << ", " << blocks.z << "}"
|
|
<< std::endl;
|
|
}
|
|
|
|
// Declare rotating_mem_ptr here so it stays in scope until it is needed
|
|
std::unique_ptr<ck_tile::RotatingMemWrapper<ADataType, BDataType>> rotating_mem_ptr;
|
|
std::function<void()> preprocess;
|
|
|
|
auto clear_gemm_output = [&]() {
|
|
if(args.k_batch > 1)
|
|
hipGetErrorString(hipMemsetAsync(
|
|
args.e_ptr, 0, args.M * args.N * sizeof(CDataType), s.stream_id_));
|
|
};
|
|
|
|
if(s.flush_cache_)
|
|
{
|
|
std::cout << "Flushing cache..." << std::endl;
|
|
|
|
ck_tile::HostTensor<ADataType> a_m(ck_tile::host_tensor_descriptor(
|
|
args.M, args.K, args.stride_A, is_row_major(ALayout{})));
|
|
ck_tile::HostTensor<BDataType> b_n(ck_tile::host_tensor_descriptor(
|
|
args.K, args.N, args.stride_B, is_row_major(BLayout{})));
|
|
|
|
auto size_a_buffer = a_m.get_element_space_size_in_bytes();
|
|
auto size_b_buffer = b_n.get_element_space_size_in_bytes();
|
|
|
|
rotating_mem_ptr = std::make_unique<ck_tile::RotatingMemWrapper<ADataType, BDataType>>(
|
|
kargs.as_ptr[0], kargs.bs_ptr[0], s.rotating_count_, size_a_buffer, size_b_buffer);
|
|
rotating_mem_ptr->Print();
|
|
|
|
preprocess = [&]() {
|
|
ck_tile::flush_icache();
|
|
rotating_mem_ptr->Next();
|
|
clear_gemm_output();
|
|
};
|
|
}
|
|
else
|
|
{
|
|
preprocess = clear_gemm_output;
|
|
}
|
|
|
|
return ck_tile::launch_kernel_time_mask(
|
|
s,
|
|
preprocess,
|
|
ck_tile::make_kernel<GemmConfig::kBlockPerCu>(Kernel{}, grids, blocks, 0, kargs));
|
|
}
|
|
};
|