Skip to content

GitLab

"tests/git@developer.sourcefind.cn:OpenDAS/apex.git" did not exist on "59e992da71b678fbdbeb6181f8dd8fd81bdc495b"
Commit dcdaeb2e authored by Bartlomiej Kocot's avatar Bartlomiej Kocot
Browse files

Fixes

parent e28aa8dd
Hide whitespace changes
Inline Side-by-side
Showing with 49 additions and 47 deletions
include/ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp
View file @ dcdaeb2e
...@@ -263,19 +263,18 @@ struct DeviceColumnToImageImpl ...@@ -263,19 +263,18 @@ struct DeviceColumnToImageImpl
decltype(BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, InputGridDesc>( decltype(BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, InputGridDesc>(
InputGridDesc{}))>; InputGridDesc{}))>;
using GridwiseTensorRearrangeKernel = using GridwiseTensorRearrangeKernel = GridwiseTensorRearrange<InputGridDesc,
GridwiseTensorRearrange<InputGridDesc, InputDataType,
InputDataType, OutputGridDesc,
OutputGridDesc, OutputDataType,
OutputDataType, BlockSize,
BlockSize, MPerBlock,
MPerBlock, KPerBlock,
KPerBlock, ThreadClusterLengths,
ThreadClusterLengths, ScalarPerVector,
ScalarPerVector, InMemoryDataOperationEnum::Add,
InMemoryDataOperationEnum::Add, Block2ETileMap,
Block2ETileMap, ComputePtrOffsetOfStridedBatch<>>;
ComputePtrOffsetOfStridedBatch<I0>>;
struct Argument : public BaseArgument struct Argument : public BaseArgument
{ {
...@@ -453,7 +452,7 @@ struct DeviceColumnToImageImpl ...@@ -453,7 +452,7 @@ struct DeviceColumnToImageImpl
std::vector<const InputDataType*> p_in_container_; std::vector<const InputDataType*> p_in_container_;
std::vector<OutputDataType*> p_out_container_; std::vector<OutputDataType*> p_out_container_;
ComputePtrOffsetOfStridedBatch<I0> compute_ptr_offset_of_batch_; ComputePtrOffsetOfStridedBatch<> compute_ptr_offset_of_batch_;
}; };
struct Invoker : public BaseInvoker struct Invoker : public BaseInvoker
...@@ -471,7 +470,7 @@ struct DeviceColumnToImageImpl ...@@ -471,7 +470,7 @@ struct DeviceColumnToImageImpl
OutputGridDesc, OutputGridDesc,
OutputDataType, OutputDataType,
Block2ETileMap, Block2ETileMap,
ComputePtrOffsetOfStridedBatch<I0>, ComputePtrOffsetOfStridedBatch<>,
GridwiseTensorRearrangeKernel>; GridwiseTensorRearrangeKernel>;
// Execute each set of independent filters // Execute each set of independent filters
......
include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
View file @ dcdaeb2e
...@@ -385,9 +385,11 @@ struct DeviceContractionMultipleABD_Xdl_CShuffle ...@@ -385,9 +385,11 @@ struct DeviceContractionMultipleABD_Xdl_CShuffle
// desc for blockwise copy // desc for blockwise copy
using AsGridDesc_AK0_M_AK1 = using AsGridDesc_AK0_M_AK1 =
remove_cvref_t<decltype(GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(AsGridDesc_M_K{}))>; remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAsGridDescriptor_AK0_M_AK1(
AsGridDesc_M_K{}))>;
using BsGridDesc_BK0_N_BK1 = using BsGridDesc_BK0_N_BK1 =
remove_cvref_t<decltype(GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(BsGridDesc_N_K{}))>; remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBsGridDescriptor_BK0_N_BK1(
BsGridDesc_N_K{}))>;
using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t< using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock( decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
DsGridDesc_M_N{}))>; DsGridDesc_M_N{}))>;
...@@ -397,7 +399,7 @@ struct DeviceContractionMultipleABD_Xdl_CShuffle ...@@ -397,7 +399,7 @@ struct DeviceContractionMultipleABD_Xdl_CShuffle
// block-to-e-tile map // block-to-e-tile map
using Block2ETileMap = using Block2ETileMap =
remove_cvref_t<decltype(GridwiseGemm::MakeBlock2ETileMap(EGridDesc_M_N{}))>; remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBlock2ETileMap(EGridDesc_M_N{}))>;
// Argument // Argument
struct Argument : public BaseArgument struct Argument : public BaseArgument
...@@ -429,7 +431,7 @@ struct DeviceContractionMultipleABD_Xdl_CShuffle ...@@ -429,7 +431,7 @@ struct DeviceContractionMultipleABD_Xdl_CShuffle
bs_grid_desc_bk0_n_bk1_{}, bs_grid_desc_bk0_n_bk1_{},
ds_grid_desc_mblock_mperblock_nblock_nperblock_{}, ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
e_grid_desc_mblock_mperblock_nblock_nperblock_{}, e_grid_desc_mblock_mperblock_nblock_nperblock_{},
block_2_etile_map_{GridwiseGemm::MakeBlock2ETileMap(e_grid_desc_m_n_)}, block_2_etile_map_{GridwiseGemm::MakeDefaultBlock2ETileMap(e_grid_desc_m_n_)},
a_element_op_{a_element_op}, a_element_op_{a_element_op},
b_element_op_{b_element_op}, b_element_op_{b_element_op},
cde_element_op_{cde_element_op} cde_element_op_{cde_element_op}
...@@ -481,10 +483,10 @@ struct DeviceContractionMultipleABD_Xdl_CShuffle ...@@ -481,10 +483,10 @@ struct DeviceContractionMultipleABD_Xdl_CShuffle
block_2_etile_map_)) block_2_etile_map_))
{ {
as_grid_desc_ak0_m_ak1_ = as_grid_desc_ak0_m_ak1_ =
GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k_); GridwiseGemm::MakeDefaultAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k_);
bs_grid_desc_bk0_n_bk1_ = bs_grid_desc_bk0_n_bk1_ =
GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k_); GridwiseGemm::MakeDefaultBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k_);
ds_grid_desc_mblock_mperblock_nblock_nperblock_ = ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock( GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
......
include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp
View file @ dcdaeb2e
...@@ -305,9 +305,11 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou ...@@ -305,9 +305,11 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
// desc for blockwise copy // desc for blockwise copy
using AsGridDesc_AK0_M_AK1 = using AsGridDesc_AK0_M_AK1 =
remove_cvref_t<decltype(GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(AsGridDesc_M_K{}))>; remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAsGridDescriptor_AK0_M_AK1(
AsGridDesc_M_K{}))>;
using BsGridDesc_BK0_N_BK1 = using BsGridDesc_BK0_N_BK1 =
remove_cvref_t<decltype(GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(BsGridDesc_N_K{}))>; remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBsGridDescriptor_BK0_N_BK1(
BsGridDesc_N_K{}))>;
using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t< using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock( decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
DsGridDesc_M_N{}))>; DsGridDesc_M_N{}))>;
...@@ -317,7 +319,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou ...@@ -317,7 +319,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
// block-to-e-tile map // block-to-e-tile map
using Block2ETileMap = using Block2ETileMap =
remove_cvref_t<decltype(GridwiseGemm::MakeBlock2ETileMap(EGridDesc_M_N{}))>; remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBlock2ETileMap(EGridDesc_M_N{}))>;
// Argument // Argument
struct Argument : public BaseArgument struct Argument : public BaseArgument
...@@ -349,7 +351,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou ...@@ -349,7 +351,7 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
bs_grid_desc_bk0_n_bk1_{}, bs_grid_desc_bk0_n_bk1_{},
ds_grid_desc_mblock_mperblock_nblock_nperblock_{}, ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
e_grid_desc_mblock_mperblock_nblock_nperblock_{}, e_grid_desc_mblock_mperblock_nblock_nperblock_{},
block_2_etile_map_{GridwiseGemm::MakeBlock2ETileMap(e_grid_desc_m_n_)}, block_2_etile_map_{GridwiseGemm::MakeDefaultBlock2ETileMap(e_grid_desc_m_n_)},
a_element_op_{a_element_op}, a_element_op_{a_element_op},
b_element_op_{b_element_op}, b_element_op_{b_element_op},
cde_element_op_{cde_element_op}, cde_element_op_{cde_element_op},
...@@ -407,10 +409,10 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou ...@@ -407,10 +409,10 @@ struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayou
block_2_etile_map_)) block_2_etile_map_))
{ {
as_grid_desc_ak0_m_ak1_ = as_grid_desc_ak0_m_ak1_ =
GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k_); GridwiseGemm::MakeDefaultAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k_);
bs_grid_desc_bk0_n_bk1_ = bs_grid_desc_bk0_n_bk1_ =
GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k_); GridwiseGemm::MakeDefaultBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k_);
ds_grid_desc_mblock_mperblock_nblock_nperblock_ = ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock( GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
......
include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp
View file @ dcdaeb2e
...@@ -79,7 +79,7 @@ template <index_t NumATensor, index_t NumBTensor, index_t NumDTensor> ...@@ -79,7 +79,7 @@ template <index_t NumATensor, index_t NumBTensor, index_t NumDTensor>
struct ComputePtrOffsetOfStridedBatch<NumATensor, struct ComputePtrOffsetOfStridedBatch<NumATensor,
NumBTensor, NumBTensor,
NumDTensor, NumDTensor,
ck::enable_if_t<(NumATensor == 1 || NumBTensor == 1)>> ck::enable_if_t<(NumATensor == 1 && NumBTensor == 1)>>
{ {
ComputePtrOffsetOfStridedBatch() = default; ComputePtrOffsetOfStridedBatch() = default;
......
include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
View file @ dcdaeb2e
...@@ -142,19 +142,18 @@ struct DeviceImageToColumnImpl ...@@ -142,19 +142,18 @@ struct DeviceImageToColumnImpl
decltype(BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>( decltype(BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(
OutputGridDesc{}))>; OutputGridDesc{}))>;
using GridwiseTensorRearrangeKernel = using GridwiseTensorRearrangeKernel = GridwiseTensorRearrange<InputGridDesc,
GridwiseTensorRearrange<InputGridDesc, InputDataType,
InputDataType, OutputGridDesc,
OutputGridDesc, OutputDataType,
OutputDataType, BlockSize,
BlockSize, MPerBlock,
MPerBlock, KPerBlock,
KPerBlock, ThreadClusterLengths,
ThreadClusterLengths, ScalarPerVector,
ScalarPerVector, InMemoryDataOperationEnum::Set,
InMemoryDataOperationEnum::Set, Block2ETileMap,
Block2ETileMap, ComputePtrOffsetOfStridedBatch<>>;
ComputePtrOffsetOfStridedBatch<I0>>;
struct Argument : public BaseArgument struct Argument : public BaseArgument
{ {
...@@ -224,7 +223,7 @@ struct DeviceImageToColumnImpl ...@@ -224,7 +223,7 @@ struct DeviceImageToColumnImpl
InputGridDesc in_grid_desc_m_k_; InputGridDesc in_grid_desc_m_k_;
OutputGridDesc out_grid_desc_m_k_; OutputGridDesc out_grid_desc_m_k_;
ComputePtrOffsetOfStridedBatch<I0> compute_ptr_offset_of_batch_; ComputePtrOffsetOfStridedBatch<> compute_ptr_offset_of_batch_;
}; };
struct Invoker : public BaseInvoker struct Invoker : public BaseInvoker
...@@ -246,7 +245,7 @@ struct DeviceImageToColumnImpl ...@@ -246,7 +245,7 @@ struct DeviceImageToColumnImpl
OutputGridDesc, OutputGridDesc,
OutputDataType, OutputDataType,
Block2ETileMap, Block2ETileMap,
ComputePtrOffsetOfStridedBatch<I0>, ComputePtrOffsetOfStridedBatch<>,
GridwiseTensorRearrangeKernel>; GridwiseTensorRearrangeKernel>;
float elapsed_time = launch_and_time_kernel(stream_config, float elapsed_time = launch_and_time_kernel(stream_config,
......
include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
View file @ dcdaeb2e
...@@ -222,7 +222,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle ...@@ -222,7 +222,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
MakeDefaultAsGridDescriptor_AK0_M_AK1(const AsGridDesc_M_K& as_grid_desc_m_k) MakeDefaultAsGridDescriptor_AK0_M_AK1(const AsGridDesc_M_K& as_grid_desc_m_k)
{ {
return generate_tuple( return generate_tuple(
[&](auto i) { return MakeAGridDescriptor_AK0_M_AK1(as_grid_desc_m_k[i]); }, [&](auto i) { return MakeDefaultAGridDescriptor_AK0_M_AK1(as_grid_desc_m_k[i]); },
Number<NumATensor>{}); Number<NumATensor>{});
} }
...@@ -248,7 +248,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle ...@@ -248,7 +248,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
MakeDefaultBsGridDescriptor_BK0_N_BK1(const BsGridDesc_N_K& bs_grid_desc_n_k) MakeDefaultBsGridDescriptor_BK0_N_BK1(const BsGridDesc_N_K& bs_grid_desc_n_k)
{ {
return generate_tuple( return generate_tuple(
[&](auto i) { return MakeBGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k[i]); }, [&](auto i) { return MakeDefaultBGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k[i]); },
Number<NumBTensor>{}); Number<NumBTensor>{});
} }
...@@ -1011,9 +1011,9 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle ...@@ -1011,9 +1011,9 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
const auto e_grid_desc_m_n = MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE); const auto e_grid_desc_m_n = MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE);
// tensor descriptors for block/thread-wise copy // tensor descriptors for block/thread-wise copy
const auto as_grid_desc_ak0_m_ak1 = MakeAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k); const auto as_grid_desc_ak0_m_ak1 = MakeDefaultAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k);
const auto bs_grid_desc_bk0_n_bk1 = MakeBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k); const auto bs_grid_desc_bk0_n_bk1 = MakeDefaultBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k);
const auto ds_grid_desc_mblock_mperblock_nblock_nperblock = const auto ds_grid_desc_mblock_mperblock_nblock_nperblock =
MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n); MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n);
......
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