mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-26 16:04:58 +00:00
[CK-Tile] Refactor base pipeline usage (#3251)
* initial poc
* factor out common parts in operator()
* cv4
* rest of the universal gemm pipelines
* fix test
* remove boilerplate from tile engine
* fix example
* fix example
* format
* fix tests build for gemm
* remove base pipeline codegen from gemm instance builder
* unify v3 logic with the rest of universal gemm pipelines
* fix build for multi abd test
* fix test gemm multi d
* fix build for weight preshuffle
* fix grouped gemm test
* fix grouped gemm multi d test
* fix grouped gemm preshuffle
* fix grouped gemm example except for quant
* fix gemm preshuffle
* fix splitk 2 stage example
* fix batched gemm example
* fix multid example
* fix multiabd example
* fix batched gemm test
* fixup
* fix examples build
* fix grouped gemm test build
* fix smoke builder
[ROCm/composable_kernel commit: d184eed823]
This commit is contained in:
Max Podkorytov
@@ -59,7 +59,6 @@ float batched_gemm(const ck_tile::BatchedGemmHostArgs& args, const ck_tile::stre
|
||||
using TilePartitioner = ck_tile::
|
||||
GemmSpatiallyLocalTilePartitioner<GemmShape, TileParitionerGroupNum, TileParitionerM01>;
|
||||
|
||||
using Traits = ck_tile::TileGemmTraits<kPadM, kPadN, kPadK, ALayout, BLayout, CLayout>;
|
||||
using GemmUniversalTraits = ck_tile::TileGemmUniversalTraits<kPadM,
|
||||
kPadN,
|
||||
kPadK,
|
||||
@@ -68,105 +67,74 @@ float batched_gemm(const ck_tile::BatchedGemmHostArgs& args, const ck_tile::stre
|
||||
BLayout,
|
||||
CLayout,
|
||||
TransposeC>;
|
||||
using GemmPipelineProblem =
|
||||
ck_tile::GemmPipelineProblem<ADataType, BDataType, AccDataType, GemmShape, Traits>;
|
||||
constexpr auto scheduler = GemmConfig::Scheduler;
|
||||
|
||||
using BaseGemmPipeline = typename PipelineTypeTraits<
|
||||
GemmConfig::Pipeline>::template UniversalGemmPipeline<GemmPipelineProblem>;
|
||||
using UniversalGemmProblem = ck_tile::UniversalGemmPipelineProblem<ADataType,
|
||||
BDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
scheduler>;
|
||||
|
||||
const ck_tile::index_t k_grain = args.k_batch * K_Tile;
|
||||
const ck_tile::index_t K_split = (args.K + k_grain - 1) / k_grain * K_Tile;
|
||||
const ck_tile::index_t num_loop = TilePartitioner::GetLoopNum(K_split);
|
||||
const bool has_hot_loop = BaseGemmPipeline::BlockHasHotloop(num_loop);
|
||||
const ck_tile::TailNumber tail_num = BaseGemmPipeline::GetBlockLoopTailNum(num_loop);
|
||||
using GemmPipeline = typename PipelineTypeTraits<GemmConfig::Pipeline>::template GemmPipeline<
|
||||
UniversalGemmProblem>;
|
||||
const auto Run = [&](const auto memory_operation_) {
|
||||
constexpr auto memory_operation = memory_operation_.value;
|
||||
|
||||
float ave_time{0};
|
||||
using GemmEpilogue = ck_tile::CShuffleEpilogue<
|
||||
ck_tile::CShuffleEpilogueProblem<ADataType,
|
||||
BDataType,
|
||||
DsDataType,
|
||||
AccDataType,
|
||||
CDataType,
|
||||
DsLayout,
|
||||
CLayout,
|
||||
CDEElementWise,
|
||||
TilePartitioner::MPerBlock,
|
||||
TilePartitioner::NPerBlock,
|
||||
M_Warp,
|
||||
N_Warp,
|
||||
M_Warp_Tile,
|
||||
N_Warp_Tile,
|
||||
K_Warp_Tile,
|
||||
UniversalGemmProblem::TransposeC,
|
||||
memory_operation>>;
|
||||
|
||||
const auto Run =
|
||||
[&](const auto has_hot_loop_, const auto tail_number_, const auto memory_operation_) {
|
||||
constexpr bool has_hot_loop_v = has_hot_loop_.value;
|
||||
constexpr auto tail_number_v = tail_number_.value;
|
||||
constexpr auto scheduler = GemmConfig::Scheduler;
|
||||
constexpr auto memory_operation = memory_operation_.value;
|
||||
using Kernel = ck_tile::BatchedGemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue>;
|
||||
auto kargs = Kernel::MakeKernelArgs(args);
|
||||
|
||||
using UniversalGemmProblem = ck_tile::UniversalGemmPipelineProblem<ADataType,
|
||||
BDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
scheduler,
|
||||
has_hot_loop_v,
|
||||
tail_number_v>;
|
||||
const dim3 grids = Kernel::GridSize(args.M, args.N, args.k_batch, args.batch_count);
|
||||
const dim3 blocks = Kernel::BlockSize();
|
||||
|
||||
using GemmPipeline = typename PipelineTypeTraits<
|
||||
GemmConfig::Pipeline>::template GemmPipeline<UniversalGemmProblem>;
|
||||
|
||||
using GemmEpilogue = ck_tile::CShuffleEpilogue<
|
||||
ck_tile::CShuffleEpilogueProblem<ADataType,
|
||||
BDataType,
|
||||
DsDataType,
|
||||
AccDataType,
|
||||
CDataType,
|
||||
DsLayout,
|
||||
CLayout,
|
||||
CDEElementWise,
|
||||
TilePartitioner::MPerBlock,
|
||||
TilePartitioner::NPerBlock,
|
||||
M_Warp,
|
||||
N_Warp,
|
||||
M_Warp_Tile,
|
||||
N_Warp_Tile,
|
||||
K_Warp_Tile,
|
||||
UniversalGemmProblem::TransposeC,
|
||||
memory_operation>>;
|
||||
|
||||
using Kernel = ck_tile::BatchedGemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue>;
|
||||
auto kargs = Kernel::MakeKernelArgs(args);
|
||||
|
||||
const dim3 grids = Kernel::GridSize(args.M, args.N, args.k_batch, args.batch_count);
|
||||
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: " << GemmShape::GetName() << '\n'
|
||||
<< "problem: " << GemmPipelineProblem::GetName() << '\n'
|
||||
<< "pipeline: " << GemmPipeline::GetName() << '\n'
|
||||
<< "grid: {" << grids.x << ", " << grids.y << ", " << grids.z << "}"
|
||||
<< ", blocks: {" << blocks.x << ", " << blocks.y << ", " << blocks.z
|
||||
<< "}" << std::endl;
|
||||
}
|
||||
|
||||
ave_time = ck_tile::launch_kernel(
|
||||
s, ck_tile::make_kernel<kBlockPerCu>(Kernel{}, grids, blocks, 0, kargs));
|
||||
return ave_time;
|
||||
};
|
||||
|
||||
const auto RunSplitk = [&](const auto has_hot_loop_, const auto tail_number_) {
|
||||
if(args.k_batch == 1)
|
||||
if(!Kernel::IsSupportedArgument(kargs))
|
||||
{
|
||||
Run(has_hot_loop_,
|
||||
tail_number_,
|
||||
ck_tile::integral_constant<ck_tile::memory_operation_enum,
|
||||
ck_tile::memory_operation_enum::set>{});
|
||||
throw std::runtime_error("Wrong! Arguments not supported! Skipping gemm!\n");
|
||||
}
|
||||
else
|
||||
|
||||
if(s.log_level_ > 0)
|
||||
{
|
||||
Run(has_hot_loop_,
|
||||
tail_number_,
|
||||
ck_tile::integral_constant<ck_tile::memory_operation_enum,
|
||||
ck_tile::memory_operation_enum::atomic_add>{});
|
||||
std::cout << "Launching kernel with args: " << Kernel::GetName() << '\n'
|
||||
<< "shape: " << GemmShape::GetName() << '\n'
|
||||
<< "pipeline: " << GemmPipeline::GetName() << '\n'
|
||||
<< "grid: {" << grids.x << ", " << grids.y << ", " << grids.z << "}"
|
||||
<< ", blocks: {" << blocks.x << ", " << blocks.y << ", " << blocks.z << "}"
|
||||
<< std::endl;
|
||||
}
|
||||
|
||||
return ck_tile::launch_kernel(
|
||||
s, ck_tile::make_kernel<kBlockPerCu>(Kernel{}, grids, blocks, 0, kargs));
|
||||
};
|
||||
|
||||
BaseGemmPipeline::TailHandler(RunSplitk, has_hot_loop, tail_num);
|
||||
|
||||
return ave_time;
|
||||
if(args.k_batch == 1)
|
||||
{
|
||||
return Run(ck_tile::integral_constant<ck_tile::memory_operation_enum,
|
||||
ck_tile::memory_operation_enum::set>{});
|
||||
}
|
||||
else
|
||||
{
|
||||
return Run(ck_tile::integral_constant<ck_tile::memory_operation_enum,
|
||||
ck_tile::memory_operation_enum::atomic_add>{});
|
||||
}
|
||||
}
|
||||
|
||||
#include "run_batched_gemm_example.inc"
|
||||
|
||||
Reference in New Issue
Block a user