mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-05 06:01:23 +00:00
[CK-Tile] move out memory operation from cshuffle epilogue class (#3359)
* 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 * hacky poc * fix tile engine * kill the lambda * maybe fix test build * more fixes * clang-format * save temp * clang-format * mostly fix examples * clang-format * remove dead code * more cleanup * fix fmha bwd build (default epilogue set/add appears to be broken) * fix default epilogue tests but not correctness * clang-format * fix bquant * clang-format * cleanup dead code * rearrange make windows for readability * restore changes to IsSupportedArgument * fix smoke-builder * clang-format * fixup rename class * build fixes * clang-format * fix builder * fixup * remove set from builder tests * fix test * clang-format * re-refactor the kernels * clang-format * fix header license * remove memory operation from conv bwd test * clang-format * clang-format example,include * clang-format test * build fixes * clang-format * solve compilation error * fix the CI * solve compilation error * clang format * solve merge conflict * solve merge conflict * solve the gfx11 error * solve test error * moar build fixes * remove AtomicAddRequiresKBatchGreaterThanOne test since the property is removed from the kernel scope --------- Co-authored-by: Thomas Ning <Thomas.Ning@amd.com>
This commit is contained in:
Max Podkorytov
@@ -96,7 +96,7 @@ class TestCkTileGroupedGemmMultiD : public ::testing::Test
|
||||
const ck_tile::stream_config& s,
|
||||
void* kargs_ptr)
|
||||
{
|
||||
|
||||
EXPECT_TRUE(gemm_descs[0].k_batch == 1);
|
||||
using GemmShape = ck_tile::TileGemmShape<
|
||||
ck_tile::sequence<Config::M_Tile_, Config::N_Tile_, Config::K_Tile_>,
|
||||
ck_tile::sequence<Config::M_Warp_, Config::N_Warp_, Config::K_Warp_>,
|
||||
@@ -134,74 +134,56 @@ class TestCkTileGroupedGemmMultiD : public ::testing::Test
|
||||
ck_tile::GemmPipelineAgBgCrCompV3<UniversalGemmProblem>,
|
||||
ck_tile::GemmPipelineAgBgCrCompV4<UniversalGemmProblem>>>;
|
||||
|
||||
const auto Run = [&](const auto memory_operation_) {
|
||||
constexpr auto memory_operation = memory_operation_.value;
|
||||
using GemmEpilogue = ck_tile::CShuffleEpilogue<
|
||||
ck_tile::CShuffleEpilogueProblem<ADataType,
|
||||
BDataType,
|
||||
DsDataType,
|
||||
AccDataType,
|
||||
EDataType,
|
||||
DsLayout,
|
||||
ELayout,
|
||||
MultiplyMultiply,
|
||||
TilePartitioner::MPerBlock,
|
||||
TilePartitioner::NPerBlock,
|
||||
Config::M_Warp_,
|
||||
Config::N_Warp_,
|
||||
Config::M_Warp_Tile_,
|
||||
Config::N_Warp_Tile_,
|
||||
Config::K_Warp_Tile_,
|
||||
UniversalGemmProblem::TransposeC>>;
|
||||
using Kernel = ck_tile::GroupedGemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue>;
|
||||
auto kargs = Kernel::MakeKargs(gemm_descs);
|
||||
EXPECT_TRUE(Kernel::IsSupportedArgument(kargs));
|
||||
const dim3 grids = Kernel::GridSize(gemm_descs);
|
||||
const dim3 blocks = Kernel::BlockSize();
|
||||
|
||||
using GemmEpilogue = ck_tile::CShuffleEpilogue<
|
||||
ck_tile::CShuffleEpilogueProblem<ADataType,
|
||||
BDataType,
|
||||
DsDataType,
|
||||
AccDataType,
|
||||
EDataType,
|
||||
DsLayout,
|
||||
ELayout,
|
||||
MultiplyMultiply,
|
||||
TilePartitioner::MPerBlock,
|
||||
TilePartitioner::NPerBlock,
|
||||
Config::M_Warp_,
|
||||
Config::N_Warp_,
|
||||
Config::M_Warp_Tile_,
|
||||
Config::N_Warp_Tile_,
|
||||
Config::K_Warp_Tile_,
|
||||
UniversalGemmProblem::TransposeC,
|
||||
memory_operation>>;
|
||||
using Kernel = ck_tile::GroupedGemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue>;
|
||||
auto kargs = Kernel::MakeKargs(gemm_descs);
|
||||
EXPECT_TRUE(Kernel::IsSupportedArgument(kargs));
|
||||
const dim3 grids = Kernel::GridSize(gemm_descs);
|
||||
const dim3 blocks = Kernel::BlockSize();
|
||||
|
||||
if(s.log_level_ > 0)
|
||||
{
|
||||
std::cout << "Launching kernel: " << Kernel::GetName()
|
||||
<< " with args:" << " grid: {" << grids.x << ", " << grids.y << ", "
|
||||
<< grids.z << "}" << ", blocks: {" << blocks.x << ", " << blocks.y << ", "
|
||||
<< blocks.z << "}" << std::endl;
|
||||
}
|
||||
|
||||
ck_tile::hip_check_error(hipMemcpyWithStream(kargs_ptr,
|
||||
kargs.data(),
|
||||
get_workspace_size(gemm_descs),
|
||||
hipMemcpyHostToDevice,
|
||||
s.stream_id_));
|
||||
|
||||
return ck_tile::launch_kernel(
|
||||
s,
|
||||
ck_tile::make_kernel<Config::BlockPerCu_>(
|
||||
Kernel{},
|
||||
grids,
|
||||
blocks,
|
||||
0,
|
||||
ck_tile::cast_pointer_to_constant_address_space(kargs_ptr),
|
||||
gemm_descs.size()));
|
||||
};
|
||||
|
||||
if(gemm_descs[0].k_batch == 1)
|
||||
if(s.log_level_ > 0)
|
||||
{
|
||||
Run(ck_tile::integral_constant<ck_tile::memory_operation_enum,
|
||||
ck_tile::memory_operation_enum::set>{});
|
||||
}
|
||||
else
|
||||
{
|
||||
// EXPECT TO FAIL because splitk is not supported
|
||||
EXPECT_FALSE(true);
|
||||
std::cout << "Launching kernel: " << Kernel::GetName() << " with args:" << " grid: {"
|
||||
<< grids.x << ", " << grids.y << ", " << grids.z << "}" << ", blocks: {"
|
||||
<< blocks.x << ", " << blocks.y << ", " << blocks.z << "}" << std::endl;
|
||||
}
|
||||
|
||||
ck_tile::hip_check_error(hipMemcpyWithStream(kargs_ptr,
|
||||
kargs.data(),
|
||||
get_workspace_size(gemm_descs),
|
||||
hipMemcpyHostToDevice,
|
||||
s.stream_id_));
|
||||
|
||||
ck_tile::ignore =
|
||||
ck_tile::launch_kernel(s,
|
||||
ck_tile::make_kernel<Config::BlockPerCu_>(
|
||||
Kernel{},
|
||||
grids,
|
||||
blocks,
|
||||
0,
|
||||
ck_tile::cast_pointer_to_constant_address_space(kargs_ptr),
|
||||
gemm_descs.size()));
|
||||
}
|
||||
|
||||
void invoke_grouped_gemm_persistent(const ck_tile::stream_config& s,
|
||||
const ck_tile::index_t num_groups,
|
||||
void* kargs_ptr,
|
||||
bool splitk)
|
||||
void* kargs_ptr)
|
||||
{
|
||||
using GemmShape = ck_tile::TileGemmShape<
|
||||
ck_tile::sequence<Config::M_Tile_, Config::N_Tile_, Config::K_Tile_>,
|
||||
@@ -218,78 +200,58 @@ class TestCkTileGroupedGemmMultiD : public ::testing::Test
|
||||
BLayout,
|
||||
ELayout>;
|
||||
|
||||
float ave_time{0};
|
||||
// We create the GEMM pipeline without specifying hotloop or tailnumber.
|
||||
// These are automatically run inside the kernel based on the given input data.
|
||||
using UniversalGemmProblem = ck_tile::UniversalGemmPipelineProblem<ADataType,
|
||||
BDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
Config::Scheduler_>;
|
||||
|
||||
const auto Run = [&](const auto memory_operation_) {
|
||||
constexpr auto memory_operation = memory_operation_.value;
|
||||
using GemmPipeline = std::conditional_t<
|
||||
Config::Pipeline_ == (PipelineType::Memory),
|
||||
ck_tile::GemmPipelineAgBgCrMem<UniversalGemmProblem>,
|
||||
std::conditional_t<Config::Pipeline_ == (PipelineType::CompV3),
|
||||
ck_tile::GemmPipelineAgBgCrCompV3<UniversalGemmProblem>,
|
||||
ck_tile::GemmPipelineAgBgCrCompV4<UniversalGemmProblem>>>;
|
||||
using GemmEpilogue = ck_tile::CShuffleEpilogue<
|
||||
ck_tile::CShuffleEpilogueProblem<ADataType,
|
||||
BDataType,
|
||||
DsDataType,
|
||||
AccDataType,
|
||||
EDataType,
|
||||
DsLayout,
|
||||
ELayout,
|
||||
MultiplyMultiply,
|
||||
TilePartitioner::MPerBlock,
|
||||
TilePartitioner::NPerBlock,
|
||||
Config::M_Warp_,
|
||||
Config::N_Warp_,
|
||||
Config::M_Warp_Tile_,
|
||||
Config::N_Warp_Tile_,
|
||||
Config::K_Warp_Tile_,
|
||||
UniversalGemmProblem::TransposeC>>;
|
||||
using Kernel = ck_tile::GroupedGemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue>;
|
||||
const dim3 blocks = Kernel::BlockSize();
|
||||
const dim3 grids = Kernel::MaxOccupancyGridSize(s);
|
||||
|
||||
// We create the GEMM pipeline without specifying hotloop or tailnumber.
|
||||
// These are automatically run inside the kernel based on the given input data.
|
||||
using UniversalGemmProblem = ck_tile::UniversalGemmPipelineProblem<ADataType,
|
||||
BDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
Config::Scheduler_>;
|
||||
|
||||
using GemmPipeline = std::conditional_t<
|
||||
Config::Pipeline_ == (PipelineType::Memory),
|
||||
ck_tile::GemmPipelineAgBgCrMem<UniversalGemmProblem>,
|
||||
std::conditional_t<Config::Pipeline_ == (PipelineType::CompV3),
|
||||
ck_tile::GemmPipelineAgBgCrCompV3<UniversalGemmProblem>,
|
||||
ck_tile::GemmPipelineAgBgCrCompV4<UniversalGemmProblem>>>;
|
||||
using GemmEpilogue = ck_tile::CShuffleEpilogue<
|
||||
ck_tile::CShuffleEpilogueProblem<ADataType,
|
||||
BDataType,
|
||||
DsDataType,
|
||||
AccDataType,
|
||||
EDataType,
|
||||
DsLayout,
|
||||
ELayout,
|
||||
MultiplyMultiply,
|
||||
TilePartitioner::MPerBlock,
|
||||
TilePartitioner::NPerBlock,
|
||||
Config::M_Warp_,
|
||||
Config::N_Warp_,
|
||||
Config::M_Warp_Tile_,
|
||||
Config::N_Warp_Tile_,
|
||||
Config::K_Warp_Tile_,
|
||||
UniversalGemmProblem::TransposeC,
|
||||
memory_operation>>;
|
||||
using Kernel = ck_tile::GroupedGemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue>;
|
||||
const dim3 blocks = Kernel::BlockSize();
|
||||
const dim3 grids = Kernel::MaxOccupancyGridSize(s);
|
||||
|
||||
if(s.log_level_ > 0)
|
||||
{
|
||||
std::cout << "Launching kernel: " << Kernel::GetName()
|
||||
<< " with args:" << " 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<Config::BlockPerCu_>(
|
||||
Kernel{},
|
||||
grids,
|
||||
blocks,
|
||||
0,
|
||||
ck_tile::cast_pointer_to_constant_address_space(kargs_ptr),
|
||||
num_groups));
|
||||
|
||||
return ave_time;
|
||||
};
|
||||
if(!splitk)
|
||||
if(s.log_level_ > 0)
|
||||
{
|
||||
Run(ck_tile::integral_constant<ck_tile::memory_operation_enum,
|
||||
ck_tile::memory_operation_enum::set>{});
|
||||
}
|
||||
else
|
||||
{
|
||||
Run(ck_tile::integral_constant<ck_tile::memory_operation_enum,
|
||||
ck_tile::memory_operation_enum::atomic_add>{});
|
||||
std::cout << "Launching kernel: " << Kernel::GetName() << " with args:" << " grid: {"
|
||||
<< grids.x << ", " << grids.y << ", " << grids.z << "}" << ", blocks: {"
|
||||
<< blocks.x << ", " << blocks.y << ", " << blocks.z << "}" << std::endl;
|
||||
}
|
||||
|
||||
ck_tile::ignore =
|
||||
ck_tile::launch_kernel(s,
|
||||
ck_tile::make_kernel<Config::BlockPerCu_>(
|
||||
Kernel{},
|
||||
grids,
|
||||
blocks,
|
||||
0,
|
||||
ck_tile::cast_pointer_to_constant_address_space(kargs_ptr),
|
||||
num_groups));
|
||||
}
|
||||
|
||||
public:
|
||||
@@ -445,8 +407,7 @@ class TestCkTileGroupedGemmMultiD : public ::testing::Test
|
||||
if constexpr(Config::Persistent_)
|
||||
{
|
||||
std::vector<ck_tile::GemmTransKernelArg<DsDataType::size()>> kargs;
|
||||
void* kargs_ptr = gemm_workspace.GetDeviceBuffer();
|
||||
const bool splitk = gemm_descs[0].k_batch > 1;
|
||||
void* kargs_ptr = gemm_workspace.GetDeviceBuffer();
|
||||
for(const auto& arg : gemm_descs)
|
||||
{
|
||||
kargs.emplace_back(
|
||||
@@ -471,7 +432,7 @@ class TestCkTileGroupedGemmMultiD : public ::testing::Test
|
||||
hipMemcpyHostToDevice,
|
||||
stream.stream_id_));
|
||||
|
||||
invoke_grouped_gemm_persistent(stream, group_count, kargs_ptr, splitk);
|
||||
invoke_grouped_gemm_persistent(stream, group_count, kargs_ptr);
|
||||
}
|
||||
else
|
||||
{
|
||||
|
||||
Reference in New Issue
Block a user