Skip to content

GitLab

Commit 6e28a8ac authored by aska-0096's avatar aska-0096
Browse files

format

parent 708fd81f
Show whitespace changes
Inline Side-by-side
Showing with 77 additions and 100 deletions
include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp
View file @ 6e28a8ac
......@@ -167,7 +167,7 @@ struct DeviceBatchedContractionMultipleD_Wmma_CShuffle
// lengths for K0, K1, ...
const auto kLengths = get_container_subset(a_ms_ks_lengths, kDimIds);
const auto a_grid_desc_m_k = [&](){
const auto a_grid_desc_m_k = [&]() {
if constexpr(ASpec == TensorSpecialization::Packed)
{
auto M = container_reduce(mLengths, math::multiplies{}, Number<1>{});
......@@ -256,7 +256,7 @@ struct DeviceBatchedContractionMultipleD_Wmma_CShuffle
// lengths for N0, N1, ...
const auto nLengths = get_container_subset(b_ns_ks_lengths, nDimIds);
const auto b_grid_desc_n_k = [&](){
const auto b_grid_desc_n_k = [&]() {
if constexpr(BSpec == TensorSpecialization::Packed)
{
auto N = container_reduce(nLengths, math::multiplies{}, Number<1>{});
......@@ -522,8 +522,8 @@ struct DeviceBatchedContractionMultipleD_Wmma_CShuffle
EGridDesc_G_M_N e_grid_desc_g_m_n_;
};
using AGridDesc = decltype(DeviceOp::MakeAGridDescriptor({},{}));
using BGridDesc = decltype(DeviceOp::MakeBGridDescriptor({},{}));
using AGridDesc = decltype(DeviceOp::MakeAGridDescriptor({}, {}));
using BGridDesc = decltype(DeviceOp::MakeBGridDescriptor({}, {}));
// GridwiseOp
using GridwiseOp = GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle<
......@@ -648,7 +648,6 @@ struct DeviceBatchedContractionMultipleD_Wmma_CShuffle
e_grid_desc_m_n_ =
DeviceOp::MakeEGridDescriptor_M_N(e_gs_ms_ns_lengths, e_gs_ms_ns_strides);
block_2_ctile_map_ = GridwiseOp::MakeDefaultBlock2CTileMap(e_grid_desc_m_n_, M01, N01);
ds_grid_desc_mblock_mperblock_nblock_nperblock =
......@@ -686,7 +685,6 @@ struct DeviceBatchedContractionMultipleD_Wmma_CShuffle
DsGridDesc_G_M_N ds_grid_desc_g_m_n_;
EGridDesc_G_M_N e_grid_desc_g_m_n_;
typename GridwiseOp::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
ds_grid_desc_mblock_mperblock_nblock_nperblock;
typename GridwiseOp::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
......
include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp
View file @ 6e28a8ac
......@@ -18,7 +18,6 @@
#include "ck/host_utility/kernel_launch.hpp"
#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
......@@ -163,16 +162,14 @@ struct DeviceGemmMultipleD_Wmma_CShuffle : public DeviceGemmMultipleD<ALayout,
if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
{
const auto b_grid_desc_nraw_kraw =
make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
make_tuple(I1, StrideB));
make_naive_tensor_descriptor(make_tuple(NRaw, KRaw), make_tuple(I1, StrideB));
return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
}
else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
{
const auto b_grid_desc_nraw_kraw =
make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
make_tuple(StrideB, I1));
make_naive_tensor_descriptor(make_tuple(NRaw, KRaw), make_tuple(StrideB, I1));
return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
}
......
include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
View file @ 6e28a8ac
......@@ -153,16 +153,14 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
{
const auto b_grid_desc_nraw_kraw =
make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
make_tuple(I1, StrideB));
make_naive_tensor_descriptor(make_tuple(NRaw, KRaw), make_tuple(I1, StrideB));
return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
}
else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
{
const auto b_grid_desc_nraw_kraw =
make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
make_tuple(StrideB, I1));
make_naive_tensor_descriptor(make_tuple(NRaw, KRaw), make_tuple(StrideB, I1));
return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
}
......@@ -304,8 +302,7 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
c_element_op_{c_element_op}
{
a_grid_desc_ = DeviceGemmWmma_CShuffle::MakeAGridDescriptor(M, K, StrideA);
b_grid_desc_k0_n_k1_ =
DeviceGemmWmma_CShuffle::MakeBGridDescriptor(K, N, StrideB);
b_grid_desc_k0_n_k1_ = DeviceGemmWmma_CShuffle::MakeBGridDescriptor(K, N, StrideB);
c_grid_desc_m_n_ = DeviceGemmWmma_CShuffle::MakeCGridDescriptor_M_N(M, N, StrideC);
block_2_ctile_map_ =
......
include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
View file @ 6e28a8ac
......@@ -184,8 +184,7 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, KPerBlock};
template <typename ALay>
static auto
MakeAGridDescriptor(const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
static auto MakeAGridDescriptor(const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
......@@ -244,8 +243,7 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
}
template <typename BLay>
static auto
MakeBGridDescriptor(const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
static auto MakeBGridDescriptor(const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides)
{
const auto wei_gemmnraw_gemmkraw_desc =
......@@ -320,7 +318,8 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
using EGridDesc_M_N = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<ELayout>({}, {}))>;
using AGridDesc = decltype(DeviceOp::MakeAGridDescriptor<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}));
using AGridDesc =
decltype(DeviceOp::MakeAGridDescriptor<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}));
using BGridDesc = decltype(DeviceOp::MakeBGridDescriptor<BLayout>({}, {}));
// GridwiseOp
......@@ -423,8 +422,8 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
conv_filter_dilations,
input_left_pads,
input_right_pads)},
b_grid_desc_{DeviceOp::MakeBGridDescriptor<BLayout>(b_g_k_c_xs_lengths,
b_g_k_c_xs_strides)},
b_grid_desc_{
DeviceOp::MakeBGridDescriptor<BLayout>(b_g_k_c_xs_lengths, b_g_k_c_xs_strides)},
ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
e_grid_desc_mblock_mperblock_nblock_nperblock_{},
block_2_etile_map_{GridwiseOp::MakeDefaultBlock2CTileMap(e_grid_desc_m_n_, M01, N01)},
......
include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
View file @ 6e28a8ac
......@@ -513,8 +513,7 @@ struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
}
template <typename BBlockDesc_>
__host__ __device__ static constexpr auto
MakeBWaveDescriptor(const BBlockDesc_&)
__host__ __device__ static constexpr auto MakeBWaveDescriptor(const BBlockDesc_&)
{
constexpr auto b_wave_desc = [&]() {
if constexpr(BEnableLds)
......@@ -595,8 +594,7 @@ struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
// block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
template <typename Block2CTileMap>
__host__ __device__ static constexpr bool
CheckValidity(const AGridDesc& a_grid_desc,
__host__ __device__ static constexpr bool CheckValidity(const AGridDesc& a_grid_desc,
const BGridDesc& b_grid_desc,
const DsGridDesc_M_N& ds_grid_desc_m_n,
const EGridDesc_M_N& e_grid_desc_m_n,
......@@ -628,13 +626,11 @@ struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
if constexpr(BEnableLds)
{
return make_tuple(b_grid_desc.GetLength(I1),
b_grid_desc.GetLength(I0) *
b_grid_desc.GetLength(I2));
b_grid_desc.GetLength(I0) * b_grid_desc.GetLength(I2));
}
else
{
return make_tuple(
b_grid_desc.GetLength(I1) * b_grid_desc.GetLength(I2) *
return make_tuple(b_grid_desc.GetLength(I1) * b_grid_desc.GetLength(I2) *
b_grid_desc.GetLength(I4),
b_grid_desc.GetLength(I0) * b_grid_desc.GetLength(I3) *
b_grid_desc.GetLength(I5));
......@@ -747,8 +743,7 @@ struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
max_lds_align)
: 0;
static constexpr auto b_block_space_size_aligned =
BEnableLds ? math::integer_least_multiple(
MakeBBlockDescriptor().GetElementSpaceSize(),
BEnableLds ? math::integer_least_multiple(MakeBBlockDescriptor().GetElementSpaceSize(),
max_lds_align)
: 0;
......
include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp
View file @ 6e28a8ac
......@@ -327,8 +327,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_wmma
}
template <typename BBlockDesc_>
__host__ __device__ static constexpr auto
MakeBWaveDescriptor(const BBlockDesc_&)
__host__ __device__ static constexpr auto MakeBWaveDescriptor(const BBlockDesc_&)
{
constexpr auto b_wave_desc = [&]() {
if constexpr(BEnableLds)
......@@ -394,8 +393,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_wmma
// block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
template <typename Block2CTileMap>
__host__ __device__ static constexpr bool
CheckValidity(const AGridDesc& a_grid_desc,
__host__ __device__ static constexpr bool CheckValidity(const AGridDesc& a_grid_desc,
const BGridDesc& b_grid_desc,
const CGridDesc_M_N& c_grid_desc_m_n,
const Block2CTileMap& block_2_ctile_map)
......@@ -426,13 +424,11 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_wmma
if constexpr(BEnableLds)
{
return make_tuple(b_grid_desc.GetLength(I1),
b_grid_desc.GetLength(I0) *
b_grid_desc.GetLength(I2));
b_grid_desc.GetLength(I0) * b_grid_desc.GetLength(I2));
}
else
{
return make_tuple(
b_grid_desc.GetLength(I1) * b_grid_desc.GetLength(I2) *
return make_tuple(b_grid_desc.GetLength(I1) * b_grid_desc.GetLength(I2) *
b_grid_desc.GetLength(I4),
b_grid_desc.GetLength(I0) * b_grid_desc.GetLength(I3) *
b_grid_desc.GetLength(I5));
......
include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
View file @ 6e28a8ac
......@@ -1399,12 +1399,7 @@ struct ThreadwiseTensorSliceTransfer_StaticToStatic_InterRow
if constexpr(IntraRowSwizzlePerm)
{
temp = __builtin_amdgcn_permlane16(
temp,
type_convert<int>(v_this_row),
0xb3a29180,
0xf7e6d5c4,
1,
0);
temp, type_convert<int>(v_this_row), 0xb3a29180, 0xf7e6d5c4, 1, 0);
v_this_row = type_convert<SrcData>(temp);
}
......
script/clang-format-overwrite.sh
View file @ 6e28a8ac
#find . -name deps -prune -o -name build -prune -o -iname '*.h' -o -iname '*.hpp' -o -iname '*.cpp' -o -iname '*.h.in' -o -iname '*.hpp.in' -o -iname '*.cpp.in' -o -iname '*.cl' -o -iname '*.cuh' -o -iname '*.cu' -o -iname '*.inc' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-10 -i -style=file {}'
# find . -name deps -prune -o -name build -prune -o -iname '*.h' -o -iname '*.hpp' -o -iname '*.cpp' -o -iname '*.h.in' -o -iname '*.hpp.in' -o -iname '*.cpp.in' -o -iname '*.cl' -o -iname '*.cuh' -o -iname '*.cu' -o -iname '*.inc' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-10 -i -style=file {}'
git status --porcelain | awk '$1 != "D" && (match($2, "\\.cpp|hpp|inc")) {print $2}' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-10 -i -style=file {}'
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