mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-03 13:11:25 +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
@@ -148,10 +148,9 @@ class TestCkTileGroupedGemmQuant : public ::testing::Test
|
||||
float ave_time{0};
|
||||
|
||||
const auto Run = [&](const auto has_hot_loop_, const auto tail_number_) {
|
||||
constexpr bool has_hot_loop_v = has_hot_loop_.value;
|
||||
constexpr auto tail_number_v = tail_number_.value;
|
||||
constexpr auto scheduler = ck_tile::GemmPipelineScheduler::Intrawave;
|
||||
constexpr auto memory_operation = ck_tile::memory_operation_enum::set;
|
||||
constexpr bool has_hot_loop_v = has_hot_loop_.value;
|
||||
constexpr auto tail_number_v = tail_number_.value;
|
||||
constexpr auto scheduler = ck_tile::GemmPipelineScheduler::Intrawave;
|
||||
|
||||
using QuantGemmProblem = std::conditional_t<
|
||||
UseGroupedQuant,
|
||||
@@ -217,8 +216,7 @@ class TestCkTileGroupedGemmQuant : public ::testing::Test
|
||||
GroupedGemKernelParam::M_Warp_Tile,
|
||||
GroupedGemKernelParam::N_Warp_Tile,
|
||||
GroupedGemKernelParam::K_Warp_Tile,
|
||||
QuantGemmProblem::TransposeC,
|
||||
memory_operation>>;
|
||||
QuantGemmProblem::TransposeC>>;
|
||||
|
||||
using Kernel = ck_tile::QuantGroupedGemmKernel<TilePartitioner,
|
||||
GemmPipeline,
|
||||
@@ -287,99 +285,92 @@ class TestCkTileGroupedGemmQuant : public ::testing::Test
|
||||
using TilePartitioner = ck_tile::
|
||||
GemmSpatiallyLocalTilePartitioner<GemmShape, TileParitionerGroupNum, TileParitionerM01>;
|
||||
|
||||
using GemmUniversalTraits = ck_tile::TileGemmQuantTraits<GroupedGemKernelParam::kPadM,
|
||||
GroupedGemKernelParam::kPadN,
|
||||
GroupedGemKernelParam::kPadK,
|
||||
false,
|
||||
PreshuffleB,
|
||||
ALayout,
|
||||
BLayout,
|
||||
CLayout,
|
||||
QuantType,
|
||||
AQLayout,
|
||||
BQLayout,
|
||||
TransposeC,
|
||||
DoubleSmemBuffer,
|
||||
Persistent>;
|
||||
using GemmUniversalTraits = ck_tile::TileGemmQuantTraits<GroupedGemKernelParam::kPadM,
|
||||
GroupedGemKernelParam::kPadN,
|
||||
GroupedGemKernelParam::kPadK,
|
||||
false,
|
||||
PreshuffleB,
|
||||
ALayout,
|
||||
BLayout,
|
||||
CLayout,
|
||||
QuantType,
|
||||
AQLayout,
|
||||
BQLayout,
|
||||
TransposeC,
|
||||
DoubleSmemBuffer,
|
||||
Persistent>;
|
||||
constexpr auto scheduler = ck_tile::GemmPipelineScheduler::Intrawave;
|
||||
constexpr bool UseGroupedQuant = QuantType == ck_tile::QuantType::AQuantGrouped ||
|
||||
QuantType == ck_tile::QuantType::BQuantGrouped;
|
||||
using QuantGemmProblem = std::conditional_t<
|
||||
UseGroupedQuant,
|
||||
std::conditional_t<QuantType == ck_tile::QuantType::AQuantGrouped,
|
||||
ck_tile::GemmAQuantPipelineProblem<ADataType,
|
||||
AQDataType,
|
||||
BDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
QuantGroupSize,
|
||||
TransposeC>,
|
||||
ck_tile::GemmBQuantPipelineProblem<ADataType,
|
||||
BDataType,
|
||||
BQDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
QuantGroupSize>>,
|
||||
ck_tile::GemmRowColTensorQuantPipelineProblem<ADataType,
|
||||
BDataType,
|
||||
AccDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
TransposeC,
|
||||
BDataType,
|
||||
scheduler>>;
|
||||
|
||||
const auto Run = [&](const auto memory_operation_) {
|
||||
constexpr auto scheduler = ck_tile::GemmPipelineScheduler::Intrawave;
|
||||
constexpr auto memory_operation = memory_operation_.value;
|
||||
// We create the GEMM pipeline without specifying hotloop or tailnumber.
|
||||
// These are automatically run inside the kernel based on the given input data.
|
||||
using GemmPipeline = std::conditional_t<
|
||||
UseGroupedQuant,
|
||||
std::conditional_t<
|
||||
QuantType == ck_tile::QuantType::AQuantGrouped,
|
||||
ck_tile::AQuantGemmPipelineAgBgCrCompV3<QuantGemmProblem>,
|
||||
std::conditional_t<PreshuffleB == true,
|
||||
ck_tile::WPQuantBPipelineAgBgCrV2<QuantGemmProblem>,
|
||||
ck_tile::BQuantGemmPipelineAgBgCrCompV3<QuantGemmProblem>>>,
|
||||
ck_tile::GemmPipelineAgBgCrCompV3<QuantGemmProblem>>;
|
||||
using GemmEpilogue = ck_tile::CShuffleEpilogue<
|
||||
ck_tile::CShuffleEpilogueProblem<ADataType,
|
||||
BDataType,
|
||||
DsDataType,
|
||||
AccDataType,
|
||||
CDataType,
|
||||
DsLayout,
|
||||
CLayout,
|
||||
ck_tile::element_wise::PassThrough,
|
||||
TilePartitioner::MPerBlock,
|
||||
TilePartitioner::NPerBlock,
|
||||
GroupedGemKernelParam::M_Warp,
|
||||
GroupedGemKernelParam::N_Warp,
|
||||
GroupedGemKernelParam::M_Warp_Tile,
|
||||
GroupedGemKernelParam::N_Warp_Tile,
|
||||
GroupedGemKernelParam::K_Warp_Tile,
|
||||
QuantGemmProblem::TransposeC>>;
|
||||
using Kernel = ck_tile::QuantGroupedGemmKernel<TilePartitioner,
|
||||
GemmPipeline,
|
||||
GemmEpilogue,
|
||||
GemmUniversalTraits::kQuantType>;
|
||||
const dim3 blocks = Kernel::BlockSize();
|
||||
const dim3 grids = Kernel::MaxOccupancyGridSize(s);
|
||||
|
||||
constexpr bool UseGroupedQuant = QuantType == ck_tile::QuantType::AQuantGrouped ||
|
||||
QuantType == ck_tile::QuantType::BQuantGrouped;
|
||||
using QuantGemmProblem = std::conditional_t<
|
||||
UseGroupedQuant,
|
||||
std::conditional_t<QuantType == ck_tile::QuantType::AQuantGrouped,
|
||||
ck_tile::GemmAQuantPipelineProblem<ADataType,
|
||||
AQDataType,
|
||||
BDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
QuantGroupSize,
|
||||
TransposeC>,
|
||||
ck_tile::GemmBQuantPipelineProblem<ADataType,
|
||||
BDataType,
|
||||
BQDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
QuantGroupSize>>,
|
||||
ck_tile::GemmRowColTensorQuantPipelineProblem<ADataType,
|
||||
BDataType,
|
||||
AccDataType,
|
||||
AccDataType,
|
||||
GemmShape,
|
||||
GemmUniversalTraits,
|
||||
TransposeC,
|
||||
BDataType,
|
||||
scheduler>>;
|
||||
|
||||
using GemmPipeline = std::conditional_t<
|
||||
UseGroupedQuant,
|
||||
std::conditional_t<
|
||||
QuantType == ck_tile::QuantType::AQuantGrouped,
|
||||
ck_tile::AQuantGemmPipelineAgBgCrCompV3<QuantGemmProblem>,
|
||||
std::conditional_t<PreshuffleB == true,
|
||||
ck_tile::WPQuantBPipelineAgBgCrV2<QuantGemmProblem>,
|
||||
ck_tile::BQuantGemmPipelineAgBgCrCompV3<QuantGemmProblem>>>,
|
||||
ck_tile::GemmPipelineAgBgCrCompV3<QuantGemmProblem>>;
|
||||
using GemmEpilogue = ck_tile::CShuffleEpilogue<
|
||||
ck_tile::CShuffleEpilogueProblem<ADataType,
|
||||
BDataType,
|
||||
DsDataType,
|
||||
AccDataType,
|
||||
CDataType,
|
||||
DsLayout,
|
||||
CLayout,
|
||||
ck_tile::element_wise::PassThrough,
|
||||
TilePartitioner::MPerBlock,
|
||||
TilePartitioner::NPerBlock,
|
||||
GroupedGemKernelParam::M_Warp,
|
||||
GroupedGemKernelParam::N_Warp,
|
||||
GroupedGemKernelParam::M_Warp_Tile,
|
||||
GroupedGemKernelParam::N_Warp_Tile,
|
||||
GroupedGemKernelParam::K_Warp_Tile,
|
||||
QuantGemmProblem::TransposeC,
|
||||
memory_operation>>;
|
||||
using Kernel = ck_tile::QuantGroupedGemmKernel<TilePartitioner,
|
||||
GemmPipeline,
|
||||
GemmEpilogue,
|
||||
GemmUniversalTraits::kQuantType>;
|
||||
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;
|
||||
}
|
||||
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::ignore =
|
||||
ck_tile::launch_kernel(s,
|
||||
ck_tile::make_kernel<GroupedGemKernelParam::kBlockPerCu>(
|
||||
Kernel{},
|
||||
@@ -388,10 +379,6 @@ class TestCkTileGroupedGemmQuant : public ::testing::Test
|
||||
0,
|
||||
ck_tile::cast_pointer_to_constant_address_space(kargs_ptr),
|
||||
num_groups));
|
||||
};
|
||||
|
||||
Run(ck_tile::integral_constant<ck_tile::memory_operation_enum,
|
||||
ck_tile::memory_operation_enum::set>{});
|
||||
}
|
||||
|
||||
template <typename Layout>
|
||||
|
||||
Reference in New Issue
Block a user