Skip to content

GitLab

"vscode:/vscode.git/clone" did not exist on "e1cbf68c61896b7e50a4114e535bccd1ad49fe97"
Commit e92395d9 authored by coderfeli's avatar coderfeli
Browse files

Merge remote-tracking branch 'origin/cka8w8_devtimer' into update_cka8w8_uc

parents 842d910e 7efafa11
Hide whitespace changes
Inline Side-by-side
Showing with 51 additions and 51 deletions
test/ck_tile/gemm/test_gemm_mem_pipeline_util.hpp → test/ck_tile/gemm/test_gemm_pipeline_util.hpp
View file @ e92395d9
...@@ -11,36 +11,28 @@ ...@@ -11,36 +11,28 @@
#include "ck_tile/ops/epilogue.hpp" #include "ck_tile/ops/epilogue.hpp"
#include "ck_tile/ops/gemm.hpp" #include "ck_tile/ops/gemm.hpp"
enum struct GemmPipelineType
{
Mem,
Comp
};
template <typename Tuple> template <typename Tuple>
class TestCkTileGemmMemPipeline : public ::testing::Test class TestCkTileGemmPipeline : public ::testing::Test
{ {
protected: protected:
using ALayout = std::tuple_element_t<0, Tuple>; using ALayout = std::tuple_element_t<0, Tuple>;
using BLayout = std::tuple_element_t<1, Tuple>; using BLayout = std::tuple_element_t<1, Tuple>;
using CLayout = std::tuple_element_t<2, Tuple>; using CLayout = std::tuple_element_t<2, Tuple>;
using ADataType = std::tuple_element_t<3, Tuple>; using ADataType = std::tuple_element_t<3, Tuple>;
using BDataType = std::tuple_element_t<4, Tuple>; using BDataType = std::tuple_element_t<4, Tuple>;
using AccDataType = std::tuple_element_t<5, Tuple>; using AccDataType = std::tuple_element_t<5, Tuple>;
using CDataType = std::tuple_element_t<6, Tuple>; using CDataType = std::tuple_element_t<6, Tuple>;
static constexpr auto Scheduler = std::tuple_element_t<7, Tuple>::value; static constexpr auto Scheduler = std::tuple_element_t<7, Tuple>::value;
static constexpr auto PipelineType = std::tuple_element_t<8, Tuple>::value;
// TODO: expose tile size through test t-param ? // TODO: expose tile size through test t-param ?
struct gemm_args
{
const void* p_a;
const void* p_b;
void* p_c;
ck_tile::index_t kbatch;
ck_tile::index_t M;
ck_tile::index_t N;
ck_tile::index_t K;
ck_tile::index_t stride_A;
ck_tile::index_t stride_B;
ck_tile::index_t stride_C;
};
template <bool PadM, bool PadN, bool PadK> template <bool PadM, bool PadN, bool PadK>
void invoke_gemm(const gemm_args& args, const ck_tile::stream_config& s) void invoke_gemm(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config& s)
{ {
// TODO: This should be parameterized in tests // TODO: This should be parameterized in tests
constexpr ck_tile::index_t M_Tile = 128; constexpr ck_tile::index_t M_Tile = 128;
...@@ -74,8 +66,13 @@ class TestCkTileGemmMemPipeline : public ::testing::Test ...@@ -74,8 +66,13 @@ class TestCkTileGemmMemPipeline : public ::testing::Test
using Traits = ck_tile::TileGemmTraits<kPadM, kPadN, kPadK, ALayout, BLayout, CLayout>; using Traits = ck_tile::TileGemmTraits<kPadM, kPadN, kPadK, ALayout, BLayout, CLayout>;
using BaseGemmPipeline = ck_tile::BaseGemmPipelineAgBgCrMem< using BaseGemmPipeline = std::conditional_t<
ck_tile::GemmPipelineProblem<ADataType, BDataType, AccDataType, GemmShape, Traits>>; PipelineType == GemmPipelineType::Mem,
ck_tile::BaseGemmPipelineAgBgCrMem<
ck_tile::GemmPipelineProblem<ADataType, BDataType, AccDataType, GemmShape, Traits>>,
ck_tile::BaseGemmPipelineAgBgCrCompV3<
ck_tile::
GemmPipelineProblem<ADataType, BDataType, AccDataType, GemmShape, Traits>>>;
const ck_tile::index_t num_loop = TilePartitioner::GetLoopNum(args.K); const ck_tile::index_t num_loop = TilePartitioner::GetLoopNum(args.K);
const bool has_hot_loop = BaseGemmPipeline::BlockHasHotloop(num_loop); const bool has_hot_loop = BaseGemmPipeline::BlockHasHotloop(num_loop);
...@@ -85,27 +82,30 @@ class TestCkTileGemmMemPipeline : public ::testing::Test ...@@ -85,27 +82,30 @@ class TestCkTileGemmMemPipeline : public ::testing::Test
constexpr bool has_hot_loop_v = has_hot_loop_.value; constexpr bool has_hot_loop_v = has_hot_loop_.value;
constexpr auto tail_number_v = tail_number_.value; constexpr auto tail_number_v = tail_number_.value;
using GemmPipeline = ck_tile::GemmPipelineAgBgCrMem< using GemmPipeline =
ck_tile::UniversalGemmPipelineProblem<ADataType, std::conditional_t<PipelineType == GemmPipelineType::Mem,
BDataType, ck_tile::GemmPipelineAgBgCrMem<
AccDataType, ck_tile::UniversalGemmPipelineProblem<ADataType,
GemmShape, BDataType,
Traits, AccDataType,
Scheduler, GemmShape,
has_hot_loop_v, Traits,
tail_number_v>>; Scheduler,
has_hot_loop_v,
tail_number_v>>,
ck_tile::GemmPipelineAgBgCrCompV3<
ck_tile::UniversalGemmPipelineProblem<ADataType,
BDataType,
AccDataType,
GemmShape,
Traits,
Scheduler,
has_hot_loop_v,
tail_number_v>>>;
using Kernel = ck_tile::GemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue>; using Kernel = ck_tile::GemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue>;
auto kargs = Kernel::MakeKargs(args.p_a, auto kargs = Kernel::MakeKernelArgs(args);
args.p_b,
args.p_c, const dim3 grids = Kernel::GridSize(args.M, args.N, args.k_batch);
args.M,
args.N,
args.K,
args.stride_A,
args.stride_B,
args.stride_C);
const dim3 grids = Kernel::GridSize(args.M, args.N, args.kbatch);
constexpr dim3 blocks = Kernel::BlockSize(); constexpr dim3 blocks = Kernel::BlockSize();
if(!Kernel::IsSupportedArgument(kargs)) if(!Kernel::IsSupportedArgument(kargs))
...@@ -297,11 +297,11 @@ class TestCkTileGemmMemPipeline : public ::testing::Test ...@@ -297,11 +297,11 @@ class TestCkTileGemmMemPipeline : public ::testing::Test
c_m_n_dev_buf.SetZero(); c_m_n_dev_buf.SetZero();
c_m_n_dev_result.SetZero(); c_m_n_dev_result.SetZero();
gemm_args args; ck_tile::GemmHostArgs args;
args.p_a = a_m_k_dev_buf.GetDeviceBuffer(); args.a_ptr = a_m_k_dev_buf.GetDeviceBuffer();
args.p_b = b_k_n_dev_buf.GetDeviceBuffer(); args.b_ptr = b_k_n_dev_buf.GetDeviceBuffer();
args.p_c = c_m_n_dev_buf.GetDeviceBuffer(); args.c_ptr = c_m_n_dev_buf.GetDeviceBuffer();
args.kbatch = kbatch; args.k_batch = kbatch;
args.M = M; args.M = M;
args.N = N; args.N = N;
args.K = K; args.K = K;
......
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