Skip to content

GitLab

Commit cdec576c authored by Adam Osewski's avatar Adam Osewski
Browse files

Remove old splitk impl and replace it with tile looping one.

parent 50540084
Hide whitespace changes
Inline Side-by-side
Showing with 348 additions and 971 deletions
include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
View file @ cdec576c
......@@ -5,11 +5,13 @@
#include <iostream>
#include <sstream>
#include <tuple>
#include "ck/ck.hpp"
#include "ck/host_utility/device_prop.hpp"
#include "ck/host_utility/kernel_launch.hpp"
#include "ck/host_utility/hip_check_error.hpp"
#include "ck/host_utility/stream_utility.hpp"
#include "ck/utility/common_header.hpp"
#include "ck/utility/tuple.hpp"
#include "ck/tensor_description/tensor_descriptor.hpp"
......@@ -23,8 +25,28 @@ namespace ck {
namespace tensor_operation {
namespace device {
///
/// @brief Entry point kernel for device-wide Grouped GEMM operation.
///
/// @param[in] gemm_descs_const The pointer to the array of GEMM descriptor structures.
/// @param[in] tile_count The overall number of output tiles we divided all groups
/// into.
/// @param[in] k_batch The number of batches we split the K dimension into.
///
/// @tparam GridwiseGemm The specific GridwiseGEMM algorithm implementation.
/// @tparam GemmDesc The structure holding all necessary descriptors and
/// other data needed for groupd gemm calculation and work
/// distribution.
/// @tparam HasMainKBlockLoop Flag indicating whether all GEMM problem configurations
/// need to loop over tiles in K dimension.
/// @tparam CGlobalMemoryDataOperation The functor used to store data in output C matrix.
/// In example could be: AtomicAdd or Store.
///
template <typename GridwiseGemm,
typename GemmDesc,
typename FloatA,
typename FloatB,
typename FloatC,
bool HasMainKBlockLoop,
InMemoryDataOperationEnum CGlobalMemoryDataOperation>
__global__ void
......@@ -32,44 +54,99 @@ __global__ void
__launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
#endif
kernel_grouped_gemm_xdl_splitk(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
const index_t group_count)
const index_t tile_count,
const index_t k_batch)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
__shared__ uint8_t p_shared[shared_size];
const index_t block_id = get_block_1d_id();
index_t tile_id = get_block_1d_id();
const index_t grid_size = get_grid_size();
const auto gemm_desc_ptr =
reinterpret_cast<const GemmDesc*>(cast_pointer_to_generic_address_space(gemm_descs_const));
index_t left = 0;
index_t right = group_count;
index_t group_id = index_t((left + right) / 2);
while((!(block_id >= gemm_desc_ptr[group_id].block_start_ &&
block_id < gemm_desc_ptr[group_id].block_end_)) &&
left <= right)
static constexpr index_t MPerBlock = GridwiseGemm::GetMPerBlock();
static constexpr index_t NPerBlock = GridwiseGemm::GetNPerBlock();
static constexpr index_t B2E_M01 = 8;
using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
using Block2ETileMapKSplit =
BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
index_t group_id = 0;
index_t offset = 0;
auto M = gemm_desc_ptr[group_id].M;
auto N = gemm_desc_ptr[group_id].N;
auto StrideC = gemm_desc_ptr[group_id].StrideC;
auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, StrideC);
auto b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, k_batch};
index_t grid_size_grp = b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
index_t gemm_tile_id_start = 0;
index_t gemm_tile_id_end = grid_size_grp;
while(tile_id < tile_count)
{
if(block_id < gemm_desc_ptr[group_id].block_start_)
{
right = group_id;
}
else
// Find corresponding GEMM group for out tile
while(!(tile_id >= gemm_tile_id_start && tile_id < gemm_tile_id_end))
{
left = group_id;
offset += grid_size_grp;
group_id++;
M = gemm_desc_ptr[group_id].M;
N = gemm_desc_ptr[group_id].N;
StrideC = gemm_desc_ptr[group_id].StrideC;
c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, StrideC);
b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, k_batch};
grid_size_grp = b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
gemm_tile_id_start = offset;
gemm_tile_id_end = offset + grid_size_grp;
}
group_id = index_t((left + right) / 2);
}
LocalBlockToCTileMap<typename GemmDesc::B2CType> local_b2c{
gemm_desc_ptr[group_id].block_2_ctile_map_,
block_id - gemm_desc_ptr[group_id].block_start_};
const auto p_a_grid = reinterpret_cast<const FloatA*>(gemm_desc_ptr[group_id].p_a_grid);
const auto p_b_grid = reinterpret_cast<const FloatB*>(gemm_desc_ptr[group_id].p_b_grid);
const auto p_c_grid = reinterpret_cast<FloatC*>(gemm_desc_ptr[group_id].p_c_grid);
const auto K = gemm_desc_ptr[group_id].K;
const auto StrideA = gemm_desc_ptr[group_id].StrideA;
const auto StrideB = gemm_desc_ptr[group_id].StrideB;
const auto MPadded = GridwiseGemm::CalculateMPadded(M);
const auto NPadded = GridwiseGemm::CalculateNPadded(N);
const auto KPadded = GridwiseGemm::CalculateKPadded(K, k_batch);
const auto K0 = GridwiseGemm::CalculateK0(K, k_batch);
LocalBlockToCTileMap<Block2ETileMapKSplit> local_b2c{b2c_tile_map, tile_id - offset};
GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
p_a_grid,
p_b_grid,
p_c_grid,
M,
N,
K,
StrideA,
StrideB,
StrideC,
MPadded,
NPadded,
KPadded,
K0,
k_batch,
static_cast<void*>(p_shared),
local_b2c);
tile_id += grid_size;
}
GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
gemm_desc_ptr[group_id].karg_, static_cast<void*>(p_shared), local_b2c);
#else
ignore = gemm_descs_const;
ignore = group_count;
ignore = tile_count;
ignore = k_batch;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
......@@ -186,35 +263,13 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
LoopSched,
PipelineVersion::v2>;
using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
using GridwiseGemmArg = typename GridwiseGemm::Argument;
using KernelArguments = GroupedGemmKernelArguments;
using Block2ETileMapKSplit =
BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
// Block2CTileMap configuration parameter.
static constexpr index_t B2E_M01 = 8;
// using GroupedGemmBlock2ETileMap = LocalBlockToCTileMap<Block2ETileMapKSplit>;
using KernelArgument = typename GridwiseGemm::Argument;
struct GemmTransKernelArg
{
using B2CType = Block2ETileMapKSplit;
KernelArgument karg_;
Block2ETileMapKSplit block_2_ctile_map_;
index_t block_start_, block_end_;
GemmTransKernelArg() = default;
GemmTransKernelArg(KernelArgument&& karg,
Block2ETileMapKSplit&& b2c_map,
index_t block_start,
index_t block_end)
: karg_{karg},
block_2_ctile_map_{b2c_map},
block_start_{block_start},
block_end_{block_end}
{
}
};
static constexpr index_t B2E_M01 = 8;
static constexpr index_t DefaultKBatch = 1;
// Argument
......@@ -227,7 +282,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
std::vector<GemmDesc>& gemm_descs)
: Argument(p_As, p_Bs, p_Es, gemm_descs, DefaultKBatch)
{
// TODO: use occupancy api to calculate appropriate batch size.
}
Argument(std::vector<const void*>& p_As,
......@@ -235,9 +289,8 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
std::vector<void*>& p_Es,
std::vector<GemmDesc>& gemm_descs,
index_t kbatch)
: K_BATCH{kbatch}
: K_BATCH{kbatch}, group_count_{0}, skipped_group_count_{0}, grid_size_{0}
{
grid_size_ = 0;
group_count_ = ck::type_convert<ck::index_t>(gemm_descs.size());
if(!(group_count_ == ck::type_convert<ck::index_t>(p_As.size()) &&
......@@ -249,8 +302,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
gemm_kernel_args_.reserve(group_count_);
skipped_group_count_ = 0;
for(std::size_t i = 0; i < gemm_descs.size(); ++i)
{
const index_t M = gemm_descs[i].M_;
......@@ -267,46 +318,29 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
const index_t stride_b = gemm_descs[i].stride_B_;
const index_t stride_c = gemm_descs[i].stride_C_;
const index_t m_padded = GridwiseGemm::CalculateMPadded(M);
const index_t n_padded = GridwiseGemm::CalculateNPadded(N);
const index_t k_padded = GridwiseGemm::CalculateKPadded(K, K_BATCH);
const index_t k0 = GridwiseGemm::CalculateK0(K, K_BATCH);
const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, stride_c);
auto local_b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
const index_t block_start = grid_size_;
const index_t block_end = grid_size_ + grid_size_grp;
grid_size_ += grid_size_grp;
auto karg = KernelArgument{type_convert<const ADataType*>(p_As[i]),
type_convert<const BDataType*>(p_Bs[i]),
type_convert<EDataType*>(p_Es[i]),
M,
N,
K,
stride_a,
stride_b,
stride_c,
m_padded,
n_padded,
k_padded,
k0,
K_BATCH};
gemm_kernel_args_.emplace_back(
std::move(karg), std::move(local_b2c_tile_map), block_start, block_end);
gemm_kernel_args_.emplace_back(type_convert<const ADataType*>(p_As[i]),
type_convert<const BDataType*>(p_Bs[i]),
type_convert<EDataType*>(p_Es[i]),
M,
N,
K,
stride_a,
stride_b,
stride_c);
}
}
/**
* @brief Recalculate group grid size for all gemms and update B2C maps.
*
* @param[in] kbatch The new splitK parameter value.
*/
///
/// @brief Set new kbatch value.
///
/// @param[in] kbatch The new splitK parameter value.
///
void UpdateKBatch(index_t kbatch)
{
K_BATCH = kbatch;
......@@ -315,28 +349,14 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
for(std::size_t i = 0; i < gemm_kernel_args_.size(); ++i)
{
auto& karg = gemm_kernel_args_[i].karg_;
const index_t k_padded = GridwiseGemm::CalculateKPadded(karg.K, K_BATCH);
const index_t k0 = GridwiseGemm::CalculateK0(karg.K, K_BATCH);
auto& gemm_arg = gemm_kernel_args_[i];
const auto c_grid_desc_m_n =
GridwiseGemm::MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
GridwiseGemm::MakeCGridDescriptor_M_N(gemm_arg.M, gemm_arg.N, gemm_arg.StrideC);
auto local_b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
const index_t block_start = grid_size_;
const index_t block_end = grid_size_ + grid_size_grp;
grid_size_ += grid_size_grp;
karg.KPadded = k_padded;
karg.K0 = k0;
karg.k_batch = K_BATCH;
gemm_kernel_args_[i].block_2_ctile_map_ = local_b2c_tile_map;
gemm_kernel_args_[i].block_start_ = block_start;
gemm_kernel_args_[i].block_end_ = block_end;
}
}
......@@ -344,31 +364,165 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
index_t K_BATCH;
index_t group_count_;
index_t skipped_group_count_;
std::vector<GemmTransKernelArg> gemm_kernel_args_;
// The overall number of output tiles to be processed.
index_t grid_size_;
const void* p_dev_gemm_args_;
std::vector<KernelArguments> gemm_kernel_args_;
};
// Invoker
struct Invoker : public BaseInvoker
{
// The oversubscription factor for the number of blocks that can simultaneously reside on
// GPU.
static constexpr int BLOCK_SUBSCRIPTION_FACTOR = 1;
static constexpr int BLOCK_WAVES = BlockSize / get_warp_size();
static constexpr int CU_SIMDS = 4;
// Assume we want to have at most 2 waves per SIMD
static constexpr int CU_BLOCKS = math::integer_divide_floor(2 * CU_SIMDS, BLOCK_WAVES);
///
/// @brief Launch Grouped Gemm kernel.
///
/// @note This function overload is using user provided device buffer for kernel
/// arguments.
///
/// @param[in] arg The structure containing kernel arguments (in host memory).
/// @param[in] dev_gemm_args The point to device memory with kernel arguments.
/// @param[in] stream_config The device stream configuration.
///
/// @return The average kernel execution time (if time measurement is enabled.)
///
float Run(const Argument& arg,
const void* dev_gemm_args,
const StreamConfig& stream_config = StreamConfig{})
{
auto [all_have_kbatch_gt_one, all_have_main_k0_block_loop] =
CheckArgument(arg, stream_config);
if(dev_gemm_args == nullptr)
{
std::ostringstream err;
err << "The gemm arguments workspace buffer is not allocated!"
<< " In " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
throw std::runtime_error(err.str());
}
if(all_have_kbatch_gt_one)
{
for(const auto& gemm_arg : arg.gemm_kernel_args_)
{
hip_check_error(hipMemset(
gemm_arg.p_c_grid, 0, gemm_arg.M * gemm_arg.N * sizeof(EDataType)));
}
}
float ave_time = 0;
if(all_have_main_k0_block_loop)
{
if(all_have_kbatch_gt_one)
{
ave_time = DispatchKernel<InMemoryDataOperationEnum::AtomicAdd, true>(
arg, dev_gemm_args, stream_config);
}
else
{
ave_time = DispatchKernel<InMemoryDataOperationEnum::Set, true>(
arg, dev_gemm_args, stream_config);
}
}
else
{
if(all_have_kbatch_gt_one)
{
ave_time = DispatchKernel<InMemoryDataOperationEnum::AtomicAdd, false>(
arg, dev_gemm_args, stream_config);
}
else
{
ave_time = DispatchKernel<InMemoryDataOperationEnum::Set, false>(
arg, dev_gemm_args, stream_config);
}
}
return ave_time;
}
///
/// @brief Launch Grouped Gemm kernel.
///
/// @note This function overload is using device workspace buffer for kernel
/// arguments. The user should call @see GetWorkSpaceSize and @see
/// SetWorkSpacePointer on arg parameter to properly allocate this buffer.
///
/// @param[in] arg The structure containing kernel arguments (in host memory).
/// @param[in] stream_config The device stream configuration.
///
/// @return The average kernel execution time (if time measurement is enabled.)
///
float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
{
index_t K0 = arg.gemm_kernel_args_[0].karg_.K0;
bool all_have_kbatch_gt_one = arg.gemm_kernel_args_[0].karg_.k_batch > 1;
if(arg.p_workspace_ != nullptr)
{
hip_check_error(
hipMemcpyWithStream(arg.p_workspace_,
arg.gemm_kernel_args_.data(),
arg.gemm_kernel_args_.size() * sizeof(KernelArguments),
hipMemcpyHostToDevice,
stream_config.stream_id_));
}
else
{
std::ostringstream err;
err << "The gemm arguments workspace buffer is not allocated!"
<< " In " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
throw std::runtime_error(err.str());
}
return Run(arg, arg.p_workspace_, stream_config);
}
float Run(const BaseArgument* p_arg,
const StreamConfig& stream_config = StreamConfig{}) override
{
return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
}
private:
auto CheckArgument(const Argument& arg, const StreamConfig& stream_config) const
{
index_t K0 = GridwiseGemm::CalculateK0(arg.gemm_kernel_args_[0].K, arg.K_BATCH);
bool all_have_kbatch_gt_one = arg.K_BATCH > 1;
bool all_have_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
{
const auto& karg = arg.gemm_kernel_args_[i].karg_;
const auto& gemm_arg = arg.gemm_kernel_args_[i];
if(stream_config.log_level_ > 0)
{
karg.Print();
gemm_arg.Print();
}
auto kbatch = karg.k_batch;
if(!GridwiseGemm::CheckValidity(karg))
// Currently all groups use same kbatch value.
auto kbatch = arg.K_BATCH;
K0 = GridwiseGemm::CalculateK0(arg.gemm_kernel_args_[i].K, arg.K_BATCH);
if(!GridwiseGemm::CheckValidity(GridwiseGemmArg{nullptr,
nullptr,
nullptr,
gemm_arg.M,
gemm_arg.N,
gemm_arg.K,
gemm_arg.StrideA,
gemm_arg.StrideB,
gemm_arg.StrideC,
0, // MPadded
0, // NPadded
0, // KPadded
K0,
kbatch}))
{
std::ostringstream err;
err << "Group id: " << i << " has invalid GridwiseGemm settings!" << __FILE__
......@@ -376,7 +530,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
throw std::runtime_error(err.str());
}
K0 = karg.K0;
bool not_all_have_main_k0_block_loop_same =
all_have_main_k0_block_loop xor GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
bool not_all_have_kbatch_value_same = all_have_kbatch_gt_one xor (kbatch > 1);
......@@ -394,97 +547,75 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
std::ostringstream err;
err << "Not all gemms have same kbatch value (=1 or >1)! "
<< "group [" << i << "], kbatch: " << kbatch
<< ", group [0], kbatch: " << arg.gemm_kernel_args_[0].karg_.k_batch
<< " in " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
<< ", group [0], kbatch: " << arg.K_BATCH << " in " << __FILE__ << ":"
<< __LINE__ << ", in function: " << __func__;
throw std::runtime_error(err.str());
}
}
return std::make_tuple(all_have_kbatch_gt_one, all_have_main_k0_block_loop);
}
hip_check_error(
hipMemcpyWithStream(arg.p_workspace_,
arg.gemm_kernel_args_.data(),
arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
hipMemcpyHostToDevice,
stream_config.stream_id_));
float ave_time = 0;
const auto Run = [&](const auto& kernel) {
if(all_have_kbatch_gt_one)
{
for(const auto& trans_arg : arg.gemm_kernel_args_)
{
const auto& karg = trans_arg.karg_;
hip_check_error(
hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(EDataType)));
}
}
ave_time =
launch_and_time_kernel(stream_config,
kernel,
dim3(arg.grid_size_),
dim3(BlockSize),
0,
cast_pointer_to_constant_address_space(arg.p_workspace_),
arg.gemm_kernel_args_.size());
};
template <InMemoryDataOperationEnum CGlobalMemoryDataOperation, bool HasMainKBlockLoop>
float DispatchKernel(const Argument& arg,
const void* dev_gemm_args,
const StreamConfig& stream_config) const
{
const auto kernel = kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
KernelArguments,
ADataType,
BDataType,
EDataType,
HasMainKBlockLoop,
CGlobalMemoryDataOperation>;
return LaunchKernel(kernel, arg, dev_gemm_args, stream_config);
}
if(all_have_main_k0_block_loop)
{
if(all_have_kbatch_gt_one)
{
const auto kernel =
kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
GemmTransKernelArg,
true,
InMemoryDataOperationEnum::AtomicAdd>;
template <typename KernelFunction>
int CalculateMaxOccupancyGridSize(const KernelFunction& kernel,
const StreamConfig& stream_config) const
{
// Calculate max number of workgroups that can simultaneously reside on the CU.
int num_blocks = 0;
size_t dyn_shared_mem_per_blk = 0;
hip_check_error(hipOccupancyMaxActiveBlocksPerMultiprocessor(
&num_blocks, kernel, BlockSize, dyn_shared_mem_per_blk));
Run(kernel);
}
else
{
const auto kernel =
kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
GemmTransKernelArg,
true,
InMemoryDataOperationEnum::Set>;
int cu_count = getAvailableComputeUnitCount(stream_config);
Run(kernel);
}
}
else
if(stream_config.log_level_ > 0)
{
if(all_have_kbatch_gt_one)
{
const auto kernel =
kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
GemmTransKernelArg,
false,
InMemoryDataOperationEnum::AtomicAdd>;
Run(kernel);
}
else
{
const auto kernel =
kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
GemmTransKernelArg,
false,
InMemoryDataOperationEnum::Set>;
Run(kernel);
}
std::cout << "MaxActiveBlocksPerCU: " << num_blocks
<< ", available CUs count: " << cu_count << ", occup. grid size: "
<< ck::math::min(num_blocks, CU_BLOCKS) * cu_count *
BLOCK_SUBSCRIPTION_FACTOR
<< std::endl;
}
return ave_time;
return cu_count * ck::math::min(num_blocks, CU_BLOCKS) * BLOCK_SUBSCRIPTION_FACTOR;
}
// polymorphic
float Run(const BaseArgument* p_arg,
const StreamConfig& stream_config = StreamConfig{}) override
template <typename KernelFunction>
float LaunchKernel(const KernelFunction& kernel,
const Argument& arg,
const void* dev_gemm_args,
const StreamConfig& stream_config) const
{
return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
int max_occupancy_grid_size = CalculateMaxOccupancyGridSize(kernel, stream_config);
// We launch the smaller number of workgroups from acutally needed tiles and the
// number of workgroups that maximize the GPU occupancy. That is because for some tile
// configuration the first is smaller than the latter. Launching too many workgroups
// mean some of them will have to iterate through all gemm problem descriptors just to
// find out they have nothing to do which is of course waste of GPU cycles.
return launch_and_time_kernel(
stream_config,
kernel,
dim3(ck::math::min(arg.grid_size_, max_occupancy_grid_size)),
dim3(BlockSize),
0,
cast_pointer_to_constant_address_space(dev_gemm_args),
arg.grid_size_,
arg.K_BATCH);
}
};
......@@ -496,11 +627,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
static bool IsSupportedArgument(const Argument& arg)
{
if(!ck::is_xdl_supported())
{
return false;
}
if((ck::type_convert<ck::index_t>(arg.gemm_kernel_args_.size()) +
arg.skipped_group_count_) != arg.group_count_)
{
......@@ -515,14 +641,28 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
bool supported = true;
for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
{
const auto& a = arg.gemm_kernel_args_[i].karg_;
bool group_arg_valid = GridwiseGemm::CheckValidity(a);
const auto& gemm_arg = arg.gemm_kernel_args_[i];
const auto K0 = GridwiseGemm::CalculateK0(gemm_arg.K, arg.K_BATCH);
bool group_arg_valid = GridwiseGemm::CheckValidity(GridwiseGemmArg{nullptr,
nullptr,
nullptr,
gemm_arg.M,
gemm_arg.N,
gemm_arg.K,
gemm_arg.StrideA,
gemm_arg.StrideB,
gemm_arg.StrideC,
0, // MPadded
0, // NPadded
0, // KPadded
K0,
arg.K_BATCH});
if(not group_arg_valid)
{
#if DEBUG_LOG
std::cout << "[" << __func__ << "] group id: " << i
<< " has invalid GridwiseGemm settings!" << std::endl;
a.Print();
gemm_arg.Print();
#endif // DEBUG_LOG
}
supported = supported && group_arg_valid;
......@@ -530,7 +670,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
return supported;
}
// polymorphic
bool IsSupportedArgument(const BaseArgument* p_arg) override
{
return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
......@@ -550,7 +689,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
static auto MakeInvoker() { return Invoker{}; }
// polymorphic
std::unique_ptr<BaseArgument>
MakeArgumentPointer(std::vector<const void*>& p_As,
std::vector<const void*>& p_Bs,
......@@ -564,19 +702,17 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
return std::make_unique<Argument>(p_As, p_Bs, p_Es, gemm_descs);
}
// polymorphic
std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
{
return std::make_unique<Invoker>(Invoker{});
}
// polymorphic
std::string GetTypeString() const override
{
auto str = std::stringstream();
// clang-format off
str << "DeviceGroupedGemm_XdlSplitK"
str << "DeviceGroupedGemm_XdlSplitKTileLoop"
<< "<"
<< std::string(ALayout::name)[0] << ","
<< std::string(BLayout::name)[0] << ","
......@@ -595,6 +731,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
<< BBlockTransferSrcScalarPerVector << ", "
<< CShuffleMXdlPerWavePerShuffle << ", "
<< CShuffleNXdlPerWavePerShuffle << ", "
<< ABlockTransferThreadClusterLengths_K0_M_K1{} << ", "
<< getGemmSpecializationString(GemmSpec)
<< ">";
// clang-format on
......@@ -605,16 +742,24 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
size_t GetWorkSpaceSize(const BaseArgument* p_arg) const override
{
return dynamic_cast<const Argument*>(p_arg)->gemm_kernel_args_.size() *
sizeof(GemmTransKernelArg);
sizeof(KernelArguments);
}
static void SetKBatchSize(Argument& arg, index_t kbatch) { arg.UpdateKBatch(kbatch); }
static void SetDeviceKernelArgs(Argument& arg, const void* p_dev_kernel_args)
{
arg.p_dev_gemm_args_ = p_dev_kernel_args;
}
// polymorphic
void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const override
{
return SetKBatchSize(*dynamic_cast<Argument*>(p_arg), kbatch);
}
void SetDeviceKernelArgs(BaseArgument* p_arg, const void* p_dev_kernel_args) const override
{
return SetDeviceKernelArgs(*dynamic_cast<Argument*>(p_arg), p_dev_kernel_args);
}
};
} // namespace device
......
include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle_tile_loop.hpp deleted 100644 → 0
View file @ 50540084
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <iostream>
#include <sstream>
#include <tuple>
#include "ck/ck.hpp"
#include "ck/host_utility/device_prop.hpp"
#include "ck/host_utility/kernel_launch.hpp"
#include "ck/host_utility/hip_check_error.hpp"
#include "ck/host_utility/stream_utility.hpp"
#include "ck/utility/common_header.hpp"
#include "ck/utility/tuple.hpp"
#include "ck/tensor_description/tensor_descriptor.hpp"
#include "ck/tensor_description/tensor_descriptor_helper.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
///
/// @brief Entry point kernel for device-wide Grouped GEMM operation.
///
/// @param[in] gemm_descs_const The pointer to the array of GEMM descriptor structures.
/// @param[in] tile_count The overall number of output tiles we divided all groups
/// into.
/// @param[in] k_batch The number of batches we split the K dimension into.
///
/// @tparam GridwiseGemm The specific GridwiseGEMM algorithm implementation.
/// @tparam GemmDesc The structure holding all necessary descriptors and
/// other data needed for groupd gemm calculation and work
/// distribution.
/// @tparam HasMainKBlockLoop Flag indicating whether all GEMM problem configurations
/// need to loop over tiles in K dimension.
/// @tparam CGlobalMemoryDataOperation The functor used to store data in output C matrix.
/// In example could be: AtomicAdd or Store.
///
template <typename GridwiseGemm,
typename GemmDesc,
typename FloatA,
typename FloatB,
typename FloatC,
bool HasMainKBlockLoop,
InMemoryDataOperationEnum CGlobalMemoryDataOperation>
__global__ void
#if CK_USE_LAUNCH_BOUNDS
__launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
#endif
kernel_grouped_gemm_xdl_splitk(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
const index_t tile_count,
const index_t k_batch)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
__shared__ uint8_t p_shared[shared_size];
index_t tile_id = get_block_1d_id();
const index_t grid_size = get_grid_size();
const auto gemm_desc_ptr =
reinterpret_cast<const GemmDesc*>(cast_pointer_to_generic_address_space(gemm_descs_const));
static constexpr index_t MPerBlock = GridwiseGemm::GetMPerBlock();
static constexpr index_t NPerBlock = GridwiseGemm::GetNPerBlock();
static constexpr index_t B2E_M01 = 8;
using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
using Block2ETileMapKSplit =
BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
index_t group_id = 0;
index_t offset = 0;
auto M = gemm_desc_ptr[group_id].M;
auto N = gemm_desc_ptr[group_id].N;
auto StrideC = gemm_desc_ptr[group_id].StrideC;
auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, StrideC);
auto b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, k_batch};
index_t grid_size_grp = b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
index_t gemm_tile_id_start = 0;
index_t gemm_tile_id_end = grid_size_grp;
while(tile_id < tile_count)
{
// Find corresponding GEMM group for out tile
while(!(tile_id >= gemm_tile_id_start && tile_id < gemm_tile_id_end))
{
offset += grid_size_grp;
group_id++;
M = gemm_desc_ptr[group_id].M;
N = gemm_desc_ptr[group_id].N;
StrideC = gemm_desc_ptr[group_id].StrideC;
c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, StrideC);
b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, k_batch};
grid_size_grp = b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
gemm_tile_id_start = offset;
gemm_tile_id_end = offset + grid_size_grp;
}
const auto p_a_grid = reinterpret_cast<const FloatA*>(gemm_desc_ptr[group_id].p_a_grid);
const auto p_b_grid = reinterpret_cast<const FloatB*>(gemm_desc_ptr[group_id].p_b_grid);
const auto p_c_grid = reinterpret_cast<FloatC*>(gemm_desc_ptr[group_id].p_c_grid);
const auto K = gemm_desc_ptr[group_id].K;
const auto StrideA = gemm_desc_ptr[group_id].StrideA;
const auto StrideB = gemm_desc_ptr[group_id].StrideB;
const auto MPadded = GridwiseGemm::CalculateMPadded(M);
const auto NPadded = GridwiseGemm::CalculateNPadded(N);
const auto KPadded = GridwiseGemm::CalculateKPadded(K, k_batch);
const auto K0 = GridwiseGemm::CalculateK0(K, k_batch);
LocalBlockToCTileMap<Block2ETileMapKSplit> local_b2c{b2c_tile_map, tile_id - offset};
GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
p_a_grid,
p_b_grid,
p_c_grid,
M,
N,
K,
StrideA,
StrideB,
StrideC,
MPadded,
NPadded,
KPadded,
K0,
k_batch,
static_cast<void*>(p_shared),
local_b2c);
tile_id += grid_size;
}
#else
ignore = gemm_descs_const;
ignore = tile_count;
ignore = k_batch;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
typename ADataType,
typename BDataType,
typename AccDataType,
typename CShuffleDataType,
typename DsDataType,
typename EDataType,
typename AElementwiseOperation,
typename BElementwiseOperation,
typename CDEElementwiseOperation,
GemmSpecialization GemmSpec,
ck::index_t NumGemmKPrefetchStage,
ck::index_t BlockSize,
ck::index_t MPerBlock,
ck::index_t NPerBlock,
ck::index_t KPerBlock,
ck::index_t AK1,
ck::index_t BK1,
ck::index_t MPerXDL,
ck::index_t NPerXDL,
ck::index_t MXdlPerWave,
ck::index_t NXdlPerWave,
typename ABlockTransferThreadClusterLengths_K0_M_K1,
typename ABlockTransferThreadClusterArrangeOrder,
typename ABlockTransferSrcAccessOrder,
ck::index_t ABlockTransferSrcVectorDim,
ck::index_t ABlockTransferSrcScalarPerVector,
ck::index_t ABlockTransferDstScalarPerVector_K1,
bool ABlockLdsExtraM,
typename BBlockTransferThreadClusterLengths_K0_N_K1,
typename BBlockTransferThreadClusterArrangeOrder,
typename BBlockTransferSrcAccessOrder,
ck::index_t BBlockTransferSrcVectorDim,
ck::index_t BBlockTransferSrcScalarPerVector,
ck::index_t BBlockTransferDstScalarPerVector_K1,
bool BBlockLdsExtraN,
index_t CShuffleMXdlPerWavePerShuffle,
index_t CShuffleNXdlPerWavePerShuffle,
typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
index_t CDEBlockTransferScalarPerVector_NPerBlock,
LoopScheduler LoopSched = make_default_loop_scheduler(),
// Current implementation does not support multiple D fusions.
enable_if_t<AK1 == BK1 && is_same_v<DsLayout, ck::Tuple<>> &&
is_same_v<DsDataType, ck::Tuple<>>,
bool> = false>
struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayout,
BLayout,
DsLayout,
ELayout,
ADataType,
BDataType,
DsDataType,
EDataType,
AElementwiseOperation,
BElementwiseOperation,
CDEElementwiseOperation>
{
static constexpr index_t NumDTensor = DsDataType::Size();
static constexpr auto I0 = Number<0>{};
static constexpr auto I1 = Number<1>{};
static constexpr auto I2 = Number<2>{};
static constexpr auto I3 = Number<3>{};
static_assert(KPerBlock % AK1 == 0);
static constexpr index_t K0PerBlock = KPerBlock / AK1;
using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
BlockSize,
ADataType, // TODO: distinguish A/B datatype
AccDataType,
EDataType,
ALayout,
BLayout,
ELayout,
AElementwiseOperation,
BElementwiseOperation,
CDEElementwiseOperation,
GemmSpec,
NumGemmKPrefetchStage,
MPerBlock,
NPerBlock,
K0PerBlock,
MPerXDL,
NPerXDL,
AK1,
MXdlPerWave,
NXdlPerWave,
ABlockTransferThreadClusterLengths_K0_M_K1,
ABlockTransferThreadClusterArrangeOrder,
ABlockTransferSrcAccessOrder,
ABlockTransferSrcVectorDim,
ABlockTransferSrcScalarPerVector,
ABlockTransferDstScalarPerVector_K1,
false, // AThreadTransferSrcResetCoordinateAfterRun,
ABlockLdsExtraM,
BBlockTransferThreadClusterLengths_K0_N_K1,
BBlockTransferThreadClusterArrangeOrder,
BBlockTransferSrcAccessOrder,
BBlockTransferSrcVectorDim,
BBlockTransferSrcScalarPerVector,
BBlockTransferDstScalarPerVector_K1,
false, // BThreadTransferSrcResetCoordinateAfterRun,
BBlockLdsExtraN,
CShuffleMXdlPerWavePerShuffle,
CShuffleNXdlPerWavePerShuffle,
CDEBlockTransferScalarPerVector_NPerBlock,
CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
LoopSched,
PipelineVersion::v2>;
using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
using GridwiseGemmArg = typename GridwiseGemm::Argument;
using KernelArguments = GroupedGemmKernelArguments;
using Block2ETileMapKSplit =
BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
// Block2CTileMap configuration parameter.
static constexpr index_t B2E_M01 = 8;
static constexpr index_t DefaultKBatch = 1;
// Argument
struct Argument : public BaseArgument
{
Argument(std::vector<const void*>& p_As,
std::vector<const void*>& p_Bs,
std::vector<void*>& p_Es,
std::vector<GemmDesc>& gemm_descs)
: Argument(p_As, p_Bs, p_Es, gemm_descs, DefaultKBatch)
{
}
Argument(std::vector<const void*>& p_As,
std::vector<const void*>& p_Bs,
std::vector<void*>& p_Es,
std::vector<GemmDesc>& gemm_descs,
index_t kbatch)
: K_BATCH{kbatch}, group_count_{0}, skipped_group_count_{0}, grid_size_{0}
{
group_count_ = ck::type_convert<ck::index_t>(gemm_descs.size());
if(!(group_count_ == ck::type_convert<ck::index_t>(p_As.size()) &&
group_count_ == ck::type_convert<ck::index_t>(p_Bs.size()) &&
group_count_ == ck::type_convert<ck::index_t>(p_Es.size())))
{
throw std::runtime_error("wrong! group_count_ != p_As/b/c.size");
}
gemm_kernel_args_.reserve(group_count_);
for(std::size_t i = 0; i < gemm_descs.size(); ++i)
{
const index_t M = gemm_descs[i].M_;
const index_t N = gemm_descs[i].N_;
const index_t K = gemm_descs[i].K_;
if(M == 0)
{
skipped_group_count_++;
continue;
}
const index_t stride_a = gemm_descs[i].stride_A_;
const index_t stride_b = gemm_descs[i].stride_B_;
const index_t stride_c = gemm_descs[i].stride_C_;
const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, stride_c);
auto local_b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
grid_size_ += grid_size_grp;
gemm_kernel_args_.emplace_back(type_convert<const ADataType*>(p_As[i]),
type_convert<const BDataType*>(p_Bs[i]),
type_convert<EDataType*>(p_Es[i]),
M,
N,
K,
stride_a,
stride_b,
stride_c);
}
}
/**
* @brief Set new kbatch value.
*
* @param[in] kbatch The new splitK parameter value.
*/
void UpdateKBatch(index_t kbatch)
{
K_BATCH = kbatch;
grid_size_ = 0;
for(std::size_t i = 0; i < gemm_kernel_args_.size(); ++i)
{
auto& gemm_arg = gemm_kernel_args_[i];
const auto c_grid_desc_m_n =
GridwiseGemm::MakeCGridDescriptor_M_N(gemm_arg.M, gemm_arg.N, gemm_arg.StrideC);
auto local_b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
grid_size_ += grid_size_grp;
}
}
// private:
index_t K_BATCH;
index_t group_count_;
index_t skipped_group_count_;
// The overall number of output tiles to be processed.
index_t grid_size_;
const void* p_dev_gemm_args_;
std::vector<KernelArguments> gemm_kernel_args_;
};
// Invoker
struct Invoker : public BaseInvoker
{
// The oversubscription factor for the number of blocks that can simultaneously reside on
// GPU.
static constexpr int BLOCK_SUBSCRIPTION_FACTOR = 1;
static constexpr int BLOCK_WAVES = BlockSize / get_warp_size();
static constexpr int CU_SIMDS = 4;
// Assume we want to have at most 2 waves per SIMD
static constexpr int CU_BLOCKS = math::integer_divide_floor(2 * CU_SIMDS, BLOCK_WAVES);
///
/// @brief Launch Grouped Gemm kernel.
///
/// @note This function overload is using user provided device buffer for kernel
/// arguments.
///
/// @param[in] arg The structure containing kernel arguments (in host memory).
/// @param[in] dev_gemm_args The point to device memory with kernel arguments.
/// @param[in] stream_config The device stream configuration.
///
/// @return The average kernel execution time (if time measurement is enabled.)
///
float Run(const Argument& arg,
const void* dev_gemm_args,
const StreamConfig& stream_config = StreamConfig{})
{
auto [all_have_kbatch_gt_one, all_have_main_k0_block_loop] =
CheckArgument(arg, stream_config);
if(dev_gemm_args == nullptr)
{
std::ostringstream err;
err << "The gemm arguments workspace buffer is not allocated!"
<< " In " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
throw std::runtime_error(err.str());
}
if(all_have_kbatch_gt_one)
{
for(const auto& gemm_arg : arg.gemm_kernel_args_)
{
hip_check_error(hipMemset(
gemm_arg.p_c_grid, 0, gemm_arg.M * gemm_arg.N * sizeof(EDataType)));
}
}
float ave_time = 0;
if(all_have_main_k0_block_loop)
{
if(all_have_kbatch_gt_one)
{
ave_time = DispatchKernel<InMemoryDataOperationEnum::AtomicAdd, true>(
arg, dev_gemm_args, stream_config);
}
else
{
ave_time = DispatchKernel<InMemoryDataOperationEnum::Set, true>(
arg, dev_gemm_args, stream_config);
}
}
else
{
if(all_have_kbatch_gt_one)
{
ave_time = DispatchKernel<InMemoryDataOperationEnum::AtomicAdd, false>(
arg, dev_gemm_args, stream_config);
}
else
{
ave_time = DispatchKernel<InMemoryDataOperationEnum::Set, false>(
arg, dev_gemm_args, stream_config);
}
}
return ave_time;
}
///
/// @brief Launch Grouped Gemm kernel.
///
/// @note This function overload is using device workspace buffer for kernel
/// arguments. The user should call @see GetWorkSpaceSize and @see
/// SetWorkSpacePointer on arg parameter to properly allocate this buffer.
///
/// @param[in] arg The structure containing kernel arguments (in host memory).
/// @param[in] stream_config The device stream configuration.
///
/// @return The average kernel execution time (if time measurement is enabled.)
///
float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
{
if(arg.p_workspace_ != nullptr)
{
hip_check_error(
hipMemcpyWithStream(arg.p_workspace_,
arg.gemm_kernel_args_.data(),
arg.gemm_kernel_args_.size() * sizeof(KernelArguments),
hipMemcpyHostToDevice,
stream_config.stream_id_));
}
else
{
std::ostringstream err;
err << "The gemm arguments workspace buffer is not allocated!"
<< " In " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
throw std::runtime_error(err.str());
}
return Run(arg, arg.p_workspace_, stream_config);
}
float Run(const BaseArgument* p_arg,
const StreamConfig& stream_config = StreamConfig{}) override
{
return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
}
private:
auto CheckArgument(const Argument& arg, const StreamConfig& stream_config) const
{
index_t K0 = GridwiseGemm::CalculateK0(arg.gemm_kernel_args_[0].K, arg.K_BATCH);
bool all_have_kbatch_gt_one = arg.K_BATCH > 1;
bool all_have_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
{
const auto& gemm_arg = arg.gemm_kernel_args_[i];
if(stream_config.log_level_ > 0)
{
// gemm_arg.Print();
}
// Currently all groups use same kbatch value.
auto kbatch = arg.K_BATCH;
K0 = GridwiseGemm::CalculateK0(arg.gemm_kernel_args_[i].K, arg.K_BATCH);
if(!GridwiseGemm::CheckValidity(GridwiseGemmArg{nullptr,
nullptr,
nullptr,
gemm_arg.M,
gemm_arg.N,
gemm_arg.K,
gemm_arg.StrideA,
gemm_arg.StrideB,
gemm_arg.StrideC,
0, // MPadded
0, // NPadded
0, // KPadded
K0,
kbatch}))
{
std::ostringstream err;
err << "Group id: " << i << " has invalid GridwiseGemm settings!" << __FILE__
<< ":" << __LINE__ << ", in function: " << __func__;
throw std::runtime_error(err.str());
}
bool not_all_have_main_k0_block_loop_same =
all_have_main_k0_block_loop xor GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
bool not_all_have_kbatch_value_same = all_have_kbatch_gt_one xor (kbatch > 1);
if(not_all_have_main_k0_block_loop_same)
{
std::ostringstream err;
err << "Not all gemms have same value for main_k0_block_loop! in " << __FILE__
<< ":" << __LINE__ << ", in function: " << __func__;
throw std::runtime_error(err.str());
}
if(not_all_have_kbatch_value_same)
{
std::ostringstream err;
err << "Not all gemms have same kbatch value (=1 or >1)! "
<< "group [" << i << "], kbatch: " << kbatch
<< ", group [0], kbatch: " << arg.K_BATCH << " in " << __FILE__ << ":"
<< __LINE__ << ", in function: " << __func__;
throw std::runtime_error(err.str());
}
}
return std::make_tuple(all_have_kbatch_gt_one, all_have_main_k0_block_loop);
}
template <InMemoryDataOperationEnum CGlobalMemoryDataOperation, bool HasMainKBlockLoop>
float DispatchKernel(const Argument& arg,
const void* dev_gemm_args,
const StreamConfig& stream_config) const
{
const auto kernel = kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
KernelArguments,
ADataType,
BDataType,
EDataType,
HasMainKBlockLoop,
CGlobalMemoryDataOperation>;
return LaunchKernel(kernel, arg, dev_gemm_args, stream_config);
}
template <typename KernelFunction>
int CalculateMaxOccupancyGridSize(const KernelFunction& kernel,
const StreamConfig& stream_config) const
{
// Calculate max number of workgroups that can simultaneously reside on the CU.
int num_blocks = 0;
size_t dyn_shared_mem_per_blk = 0;
hip_check_error(hipOccupancyMaxActiveBlocksPerMultiprocessor(
&num_blocks, kernel, BlockSize, dyn_shared_mem_per_blk));
int cu_count = getAvailableComputeUnitCount(stream_config);
if(stream_config.log_level_ > 0)
{
std::cout << "MaxActiveBlocksPerCU: " << num_blocks
<< ", available CUs count: " << cu_count << ", grid size: "
<< ck::math::min(num_blocks, CU_BLOCKS) * cu_count *
BLOCK_SUBSCRIPTION_FACTOR
<< std::endl;
}
return cu_count * ck::math::min(num_blocks, CU_BLOCKS) * BLOCK_SUBSCRIPTION_FACTOR;
}
template <typename KernelFunction>
float LaunchKernel(const KernelFunction& kernel,
const Argument& arg,
const void* dev_gemm_args,
const StreamConfig& stream_config) const
{
int max_occupancy_grid_size = CalculateMaxOccupancyGridSize(kernel, stream_config);
// We launch the smaller number of workgroups from acutally needed tiles and the
// number of workgroups that maximize the GPU occupancy. That is because for some tile
// configuration the first is smaller than the latter. Launching too many workgroups
// mean some of them will have to iterate through all gemm problem descriptors just to
// find out they have nothing to do which is of course waste of GPU cycles.
return launch_and_time_kernel(
stream_config,
kernel,
dim3(ck::math::min(arg.grid_size_, max_occupancy_grid_size)),
dim3(BlockSize),
0,
cast_pointer_to_constant_address_space(dev_gemm_args),
arg.grid_size_,
arg.K_BATCH);
}
};
static constexpr bool IsValidCompilationParameter()
{
// TODO: properly implement this check
return true;
}
static bool IsSupportedArgument(const Argument& arg)
{
if((ck::type_convert<ck::index_t>(arg.gemm_kernel_args_.size()) +
arg.skipped_group_count_) != arg.group_count_)
{
#if DEBUG_LOG
std::cout << "The group count is not equal to sum of skipped groups "
"and kernel args size!"
<< std::endl;
#endif // DEBUG_LOG
return false;
}
bool supported = true;
for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
{
const auto& gemm_arg = arg.gemm_kernel_args_[i];
const auto K0 = GridwiseGemm::CalculateK0(gemm_arg.K, arg.K_BATCH);
bool group_arg_valid = GridwiseGemm::CheckValidity(GridwiseGemmArg{nullptr,
nullptr,
nullptr,
gemm_arg.M,
gemm_arg.N,
gemm_arg.K,
gemm_arg.StrideA,
gemm_arg.StrideB,
gemm_arg.StrideC,
0, // MPadded
0, // NPadded
0, // KPadded
K0,
arg.K_BATCH});
if(not group_arg_valid)
{
#if DEBUG_LOG
std::cout << "[" << __func__ << "] group id: " << i
<< " has invalid GridwiseGemm settings!" << std::endl;
gemm_arg.Print();
#endif // DEBUG_LOG
}
supported = supported && group_arg_valid;
}
return supported;
}
bool IsSupportedArgument(const BaseArgument* p_arg) override
{
return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
}
static auto MakeArgument(std::vector<const void*>& p_As,
std::vector<const void*>& p_Bs,
std::vector<std::array<const void*, NumDTensor>>&,
std::vector<void*>& p_Es,
std::vector<GemmDesc> gemm_descs,
AElementwiseOperation,
BElementwiseOperation,
CDEElementwiseOperation)
{
return Argument{p_As, p_Bs, p_Es, gemm_descs};
}
static auto MakeInvoker() { return Invoker{}; }
std::unique_ptr<BaseArgument>
MakeArgumentPointer(std::vector<const void*>& p_As,
std::vector<const void*>& p_Bs,
std::vector<std::array<const void*, NumDTensor>>&,
std::vector<void*>& p_Es,
std::vector<GemmDesc>& gemm_descs,
AElementwiseOperation,
BElementwiseOperation,
CDEElementwiseOperation) override
{
return std::make_unique<Argument>(p_As, p_Bs, p_Es, gemm_descs);
}
std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
{
return std::make_unique<Invoker>(Invoker{});
}
std::string GetTypeString() const override
{
auto str = std::stringstream();
// clang-format off
str << "DeviceGroupedGemm_XdlSplitKTileLoop"
<< "<"
<< std::string(ALayout::name)[0] << ","
<< std::string(BLayout::name)[0] << ","
<< std::string(ELayout::name)[0] << ","
<< BlockSize << ", "
<< MPerBlock << ", "
<< NPerBlock << ", "
<< KPerBlock << ", "
<< AK1 << ", "
<< BK1 << ", "
<< MPerXDL << ", "
<< NPerXDL << ", "
<< MXdlPerWave << ", "
<< NXdlPerWave << ", "
<< ABlockTransferSrcScalarPerVector << ", "
<< BBlockTransferSrcScalarPerVector << ", "
<< CShuffleMXdlPerWavePerShuffle << ", "
<< CShuffleNXdlPerWavePerShuffle << ", "
<< ABlockTransferThreadClusterLengths_K0_M_K1{} << ", "
<< getGemmSpecializationString(GemmSpec)
<< ">";
// clang-format on
return str.str();
}
size_t GetWorkSpaceSize(const BaseArgument* p_arg) const override
{
return dynamic_cast<const Argument*>(p_arg)->gemm_kernel_args_.size() *
sizeof(KernelArguments);
}
static void SetKBatchSize(Argument& arg, index_t kbatch) { arg.UpdateKBatch(kbatch); }
static void SetDeviceKernelArgs(Argument& arg, const void* p_dev_kernel_args)
{
arg.p_dev_gemm_args_ = p_dev_kernel_args;
}
void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const override
{
return SetKBatchSize(*dynamic_cast<Argument*>(p_arg), kbatch);
}
void SetDeviceKernelArgs(BaseArgument* p_arg, const void* p_dev_kernel_args) const override
{
return SetDeviceKernelArgs(*dynamic_cast<Argument*>(p_arg), p_dev_kernel_args);
}
};
} // namespace device
} // namespace tensor_operation
} // namespace ck
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment