Skip to content

GitLab

"...composable_kernel_rocm.git" did not exist on "1857761a298fa03393aee624134e0017d8a1147b"
Unverified Commit 5bb041ba authored by jakpiase's avatar jakpiase Committed by GitHub
Browse files

add vectorloads on non-k dim for memory pipelines (#1856)

parent 2bef5501
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 251 additions and 72 deletions
include/ck_tile/ops/gemm/pipeline/gemm_universal_pipeline_ag_bg_cr_policy.hpp
View file @ 5bb041ba
......@@ -519,7 +519,7 @@ struct UniversalGemmPipelineAgBgCrPolicy
using ALayout = remove_cvref_t<typename Problem::ALayout>;
static_assert(std::is_same_v<ALayout, ck_tile::tensor_layout::gemm::ColumnMajor>);
constexpr index_t BlockSize = Problem::kBlockSize;
constexpr index_t MPerBlock = Problem::BlockGemmShape::kN;
constexpr index_t MPerBlock = Problem::BlockGemmShape::kM;
constexpr index_t KPerBlock = Problem::BlockGemmShape::kK;
constexpr index_t VecLoadSize = GetVectorSizeA<Problem>();
......
test/ck_tile/gemm/test_gemm_pipeline.cpp
View file @ 5bb041ba
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
#include <tuple>
......@@ -14,28 +14,26 @@ using Row = ck_tile::tensor_layout::gemm::RowMajor;
using Col = ck_tile::tensor_layout::gemm::ColumnMajor;
using Intrawave = ck_tile::integral_constant<ck_tile::GemmPipelineScheduler,
ck_tile::GemmPipelineScheduler::Intrawave>;
// using Interwave = ck_tile::integral_constant<ck_tile::GemmPipelineScheduler,
// ck_tile::GemmPipelineScheduler::Interwave>;
// using Mem = ck_tile::integral_constant<GemmPipelineType, GemmPipelineType::Mem>;
using Comp = ck_tile::integral_constant<GemmPipelineType, GemmPipelineType::Comp>;
// TODO: Enable Memory pipeline, when it would be updated for vector loads on non-K major tensors.
using Interwave = ck_tile::integral_constant<ck_tile::GemmPipelineScheduler,
ck_tile::GemmPipelineScheduler::Interwave>;
using Mem = ck_tile::integral_constant<GemmPipelineType, GemmPipelineType::Mem>;
using Comp = ck_tile::integral_constant<GemmPipelineType, GemmPipelineType::Comp>;
// clang-format off
using KernelTypes = ::testing::Types<
// ALayout, BLayout, CLayout, ADataType, BDataType, AccDataType, CDataType, GemmPipelineScheduler, PipelineType
// std::tuple< Row, Row, Row, F16, F16, F32, F16, Intrawave, Mem>,
std::tuple< Row, Row, Row, F16, F16, F32, F16, Intrawave, Mem>,
std::tuple< Row, Row, Row, F16, F16, F32, F16, Intrawave, Comp>,
// std::tuple< Row, Row, Row, F16, F16, F32, F16, Interwave, Mem>,
// std::tuple< Row, Col, Row, F16, F16, F32, F16, Intrawave, Mem>,
std::tuple< Row, Row, Row, F16, F16, F32, F16, Interwave, Mem>,
std::tuple< Row, Col, Row, F16, F16, F32, F16, Intrawave, Mem>,
std::tuple< Row, Col, Row, F16, F16, F32, F16, Intrawave, Comp>,
// std::tuple< Row, Col, Row, F16, F16, F32, F16, Interwave, Mem>,
// std::tuple< Col, Row, Row, F16, F16, F32, F16, Intrawave, Mem>,
std::tuple< Row, Col, Row, F16, F16, F32, F16, Interwave, Mem>,
std::tuple< Col, Row, Row, F16, F16, F32, F16, Intrawave, Mem>,
std::tuple< Col, Row, Row, F16, F16, F32, F16, Intrawave, Comp>,
// std::tuple< Col, Row, Row, F16, F16, F32, F16, Interwave, Mem>,
// std::tuple< Col, Col, Row, F16, F16, F32, F16, Intrawave, Mem>,
std::tuple< Col, Col, Row, F16, F16, F32, F16, Intrawave, Comp>
// std::tuple< Col, Col, Row, F16, F16, F32, F16, Interwave, Mem>
std::tuple< Col, Row, Row, F16, F16, F32, F16, Interwave, Mem>,
std::tuple< Col, Col, Row, F16, F16, F32, F16, Intrawave, Mem>,
std::tuple< Col, Col, Row, F16, F16, F32, F16, Intrawave, Comp>,
std::tuple< Col, Col, Row, F16, F16, F32, F16, Interwave, Mem>
>;
// clang-format on
......
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