Skip to content

GitLab

Commit b3054fea authored by Adam Osewski's avatar Adam Osewski
Browse files

Merge branch 'develop' into aosewski/ck_tile_gemm_policy

parents 7cbc1492 052a7265
Show whitespace changes
Inline Side-by-side
Showing with 475 additions and 63 deletions
client_example/07_grouped_convnd_fwd/CMakeLists.txt
View file @ b3054fea
...@@ -22,4 +22,7 @@ if(GPU_TARGETS MATCHES "gfx9") ...@@ -22,4 +22,7 @@ if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_grouped_conv3d_fwd_bf8_fp8 grouped_conv3d_fwd_bf8_fp8.cpp) add_executable(client_grouped_conv3d_fwd_bf8_fp8 grouped_conv3d_fwd_bf8_fp8.cpp)
target_link_libraries(client_grouped_conv3d_fwd_bf8_fp8 PRIVATE composable_kernel::device_conv_operations) target_link_libraries(client_grouped_conv3d_fwd_bf8_fp8 PRIVATE composable_kernel::device_conv_operations)
endif() endif()
add_executable(grouped_conv2d_fwd_ngchw grouped_conv2d_fwd_ngchw.cpp)
target_link_libraries(grouped_conv2d_fwd_ngchw PRIVATE composable_kernel::device_conv_operations)
endif() endif()
client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd_ngchw.cpp 0 → 100644
View file @ b3054fea
// SPDX-License-Identifier: MIT
// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
#include <tuple>
#include <cstdlib>
#include <iomanip>
#include <iostream>
#include <iterator>
#include <numeric>
#include <vector>
#include "ck/utility/data_type.hpp"
#include "ck/utility/tuple.hpp"
#include "ck/ck.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
using InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using OutDataType = ck::half_t;
// Use std tuple instead of ck tuple to avoid clang
// implicit instantiation of undefined template error.
using DDataTypes = std::tuple<ck::half_t>;
using InLayout = ck::tensor_layout::convolution::NGCHW;
using WeiLayout = ck::tensor_layout::convolution::GKYXC;
using OutLayout = ck::tensor_layout::convolution::NGKHW;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr ck::index_t NumDimSpatial = 2;
static constexpr ck::index_t G = 32;
static constexpr ck::index_t N = 64; // batch size
static constexpr ck::index_t K = 64; // output channel
static constexpr ck::index_t C = 32; // input channel (per group)
static constexpr ck::index_t Y = 3; // filter H
static constexpr ck::index_t X = 3; // filter W
static constexpr ck::index_t Hi = 14; // input H
static constexpr ck::index_t Wi = 14; // input W
static constexpr ck::index_t Ho = 14; // output H
static constexpr ck::index_t Wo = 14; // output W
struct SimpleDeviceMem
{
SimpleDeviceMem() = delete;
SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
{
(void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
}
void* GetDeviceBuffer() { return p_mem_; }
~SimpleDeviceMem() { (void)hipFree(p_mem_); }
void* p_mem_;
};
int execute_conv_fwd()
{
std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
std::array<ck::index_t, 5> in_strides{C * Hi * Wi, G * C * Hi * Wi, Hi * Wi, Wi, 1};
std::array<ck::index_t, 5> wei_lengths{G, K, C, Y, X};
std::array<ck::index_t, 5> wei_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
std::array<ck::index_t, 5> out_strides{K * Ho * Wo, G * K * Ho * Wo, Ho * Wo, Wo, 1};
std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
InLayout,
WeiLayout,
ck::Tuple<>,
OutLayout,
InDataType,
WeiDataType,
ck::Tuple<>,
OutDataType,
PassThrough,
PassThrough,
PassThrough>;
// get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances();
std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
std::string best_op_name;
int best_op_id = -1;
float best_avg_time = std::numeric_limits<float>::max();
float best_gb_per_sec = 0;
float best_tflops = 0;
// profile device operation instances
std::cout << "Run all instances and do timing" << std::endl;
for(int i = 0; i < op_ptrs.size(); ++i)
{
auto& op_ptr = op_ptrs[i];
auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
wei.GetDeviceBuffer(),
{},
out.GetDeviceBuffer(),
in_lengths,
in_strides,
wei_lengths,
wei_strides,
{},
{},
out_lengths,
out_strides,
filter_strides,
filter_dilations,
input_left_pads,
input_right_pads,
PassThrough{},
PassThrough{},
PassThrough{});
auto invoker_ptr = op_ptr->MakeInvokerPointer();
std::string op_name = op_ptr->GetTypeString();
// workspace_sz will be equal to 0 for other layout than NGCHW
const std::size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
SimpleDeviceMem workspace_dev(workspace_sz);
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
std::size_t flop =
std::size_t(2) * G * N * K * C * Ho * Wo * Y * X + 3 * N * Ho * Wo * G * K;
std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * G * C +
sizeof(WeiDataType) * G * K * Y * X * C +
sizeof(OutDataType) * 2 * N * Ho * Wo * G * K;
float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
float gb_per_sec = num_bytes / 1.E6 / avg_time;
std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
<< gb_per_sec << " GB/s, " << op_name << std::endl;
if(tflops > best_tflops)
{
best_op_id = i;
best_op_name = op_name;
best_avg_time = avg_time;
best_gb_per_sec = gb_per_sec;
best_tflops = tflops;
}
}
else
{
std::cerr << op_name << " does not support this problem" << std::endl;
}
}
if(best_op_id < 0)
{
std::cerr << "no suitable instance" << std::endl;
return EXIT_FAILURE;
}
std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
<< " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
// run the best intance
{
auto& op_ptr = op_ptrs[best_op_id];
std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
<< std::endl;
auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
wei.GetDeviceBuffer(),
{},
out.GetDeviceBuffer(),
in_lengths,
in_strides,
wei_lengths,
wei_strides,
{},
{},
out_lengths,
out_strides,
filter_strides,
filter_dilations,
input_left_pads,
input_right_pads,
PassThrough{},
PassThrough{},
PassThrough{});
const std::size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
SimpleDeviceMem workspace_dev(workspace_sz);
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
auto invoker_ptr = op_ptr->MakeInvokerPointer();
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
}
std::cout << "Done" << std::endl;
}
return 0;
}
int main() { return execute_conv_fwd(); }
docs/sphinx/requirements.in
View file @ b3054fea
rocm-docs-core==1.13.0 rocm-docs-core==1.14.1
sphinxcontrib-bibtex==2.6.3 sphinxcontrib-bibtex==2.6.3
docs/sphinx/requirements.txt
View file @ b3054fea
...@@ -8,6 +8,13 @@ accessible-pygments==0.0.5 ...@@ -8,6 +8,13 @@ accessible-pygments==0.0.5
# via pydata-sphinx-theme # via pydata-sphinx-theme
alabaster==0.7.16 alabaster==0.7.16
# via sphinx # via sphinx
asttokens==3.0.0
# via stack-data
attrs==24.3.0
# via
# jsonschema
# jupyter-cache
# referencing
babel==2.15.0 babel==2.15.0
# via # via
# pydata-sphinx-theme # pydata-sphinx-theme
...@@ -25,9 +32,17 @@ cffi==1.16.0 ...@@ -25,9 +32,17 @@ cffi==1.16.0
charset-normalizer==3.3.2 charset-normalizer==3.3.2
# via requests # via requests
click==8.1.7 click==8.1.7
# via sphinx-external-toc # via
# jupyter-cache
# sphinx-external-toc
comm==0.2.2
# via ipykernel
cryptography==43.0.0 cryptography==43.0.0
# via pyjwt # via pyjwt
debugpy==1.8.12
# via ipykernel
decorator==5.1.1
# via ipython
deprecated==1.2.14 deprecated==1.2.14
# via pygithub # via pygithub
docutils==0.21.2 docutils==0.21.2
...@@ -38,20 +53,56 @@ docutils==0.21.2 ...@@ -38,20 +53,56 @@ docutils==0.21.2
# pydata-sphinx-theme # pydata-sphinx-theme
# sphinx # sphinx
# sphinxcontrib-bibtex # sphinxcontrib-bibtex
exceptiongroup==1.2.2
# via ipython
executing==2.1.0
# via stack-data
fastjsonschema==2.20.0 fastjsonschema==2.20.0
# via rocm-docs-core # via
# nbformat
# rocm-docs-core
gitdb==4.0.11 gitdb==4.0.11
# via gitpython # via gitpython
gitpython==3.1.43 gitpython==3.1.43
# via rocm-docs-core # via rocm-docs-core
greenlet==3.1.1
# via sqlalchemy
idna==3.7 idna==3.7
# via requests # via requests
imagesize==1.4.1 imagesize==1.4.1
# via sphinx # via sphinx
importlib-metadata==8.6.1
# via
# jupyter-cache
# myst-nb
ipykernel==6.29.5
# via myst-nb
ipython==8.31.0
# via
# ipykernel
# myst-nb
jedi==0.19.2
# via ipython
jinja2==3.1.4 jinja2==3.1.4
# via # via
# myst-parser # myst-parser
# sphinx # sphinx
jsonschema==4.23.0
# via nbformat
jsonschema-specifications==2024.10.1
# via jsonschema
jupyter-cache==1.0.1
# via myst-nb
jupyter-client==8.6.3
# via
# ipykernel
# nbclient
jupyter-core==5.7.2
# via
# ipykernel
# jupyter-client
# nbclient
# nbformat
latexcodec==3.0.0 latexcodec==3.0.0
# via pybtex # via pybtex
markdown-it-py==3.0.0 markdown-it-py==3.0.0
...@@ -60,16 +111,48 @@ markdown-it-py==3.0.0 ...@@ -60,16 +111,48 @@ markdown-it-py==3.0.0
# myst-parser # myst-parser
markupsafe==2.1.5 markupsafe==2.1.5
# via jinja2 # via jinja2
matplotlib-inline==0.1.7
# via
# ipykernel
# ipython
mdit-py-plugins==0.4.1 mdit-py-plugins==0.4.1
# via myst-parser # via myst-parser
mdurl==0.1.2 mdurl==0.1.2
# via markdown-it-py # via markdown-it-py
myst-parser==3.0.1 myst-nb==1.1.2
# via rocm-docs-core # via rocm-docs-core
myst-parser==3.0.1
# via myst-nb
nbclient==0.10.2
# via
# jupyter-cache
# myst-nb
nbformat==5.10.4
# via
# jupyter-cache
# myst-nb
# nbclient
nest-asyncio==1.6.0
# via ipykernel
packaging==24.1 packaging==24.1
# via # via
# ipykernel
# pydata-sphinx-theme # pydata-sphinx-theme
# sphinx # sphinx
parso==0.8.4
# via jedi
pexpect==4.9.0
# via ipython
platformdirs==4.3.6
# via jupyter-core
prompt-toolkit==3.0.50
# via ipython
psutil==6.1.1
# via ipykernel
ptyprocess==0.7.0
# via pexpect
pure-eval==0.2.3
# via stack-data
pybtex==0.24.0 pybtex==0.24.0
# via # via
# pybtex-docutils # pybtex-docutils
...@@ -87,26 +170,45 @@ pygithub==2.3.0 ...@@ -87,26 +170,45 @@ pygithub==2.3.0
pygments==2.18.0 pygments==2.18.0
# via # via
# accessible-pygments # accessible-pygments
# ipython
# pydata-sphinx-theme # pydata-sphinx-theme
# sphinx # sphinx
pyjwt[crypto]==2.8.0 pyjwt[crypto]==2.8.0
# via pygithub # via pygithub
pynacl==1.5.0 pynacl==1.5.0
# via pygithub # via pygithub
python-dateutil==2.9.0.post0
# via jupyter-client
pyyaml==6.0.1 pyyaml==6.0.1
# via # via
# jupyter-cache
# myst-nb
# myst-parser # myst-parser
# pybtex # pybtex
# rocm-docs-core # rocm-docs-core
# sphinx-external-toc # sphinx-external-toc
pyzmq==26.2.0
# via
# ipykernel
# jupyter-client
referencing==0.36.1
# via
# jsonschema
# jsonschema-specifications
requests==2.32.3 requests==2.32.3
# via # via
# pygithub # pygithub
# sphinx # sphinx
rocm-docs-core==1.13.0 rocm-docs-core==1.14.1
# via -r requirements.in # via -r requirements.in
rpds-py==0.22.3
# via
# jsonschema
# referencing
six==1.16.0 six==1.16.0
# via pybtex # via
# pybtex
# python-dateutil
smmap==5.0.1 smmap==5.0.1
# via gitdb # via gitdb
snowballstemmer==2.2.0 snowballstemmer==2.2.0
...@@ -116,6 +218,7 @@ soupsieve==2.5 ...@@ -116,6 +218,7 @@ soupsieve==2.5
sphinx==7.4.7 sphinx==7.4.7
# via # via
# breathe # breathe
# myst-nb
# myst-parser # myst-parser
# pydata-sphinx-theme # pydata-sphinx-theme
# rocm-docs-core # rocm-docs-core
...@@ -149,15 +252,43 @@ sphinxcontrib-qthelp==2.0.0 ...@@ -149,15 +252,43 @@ sphinxcontrib-qthelp==2.0.0
# via sphinx # via sphinx
sphinxcontrib-serializinghtml==2.0.0 sphinxcontrib-serializinghtml==2.0.0
# via sphinx # via sphinx
sqlalchemy==2.0.37
# via jupyter-cache
stack-data==0.6.3
# via ipython
tabulate==0.9.0
# via jupyter-cache
tomli==2.0.1 tomli==2.0.1
# via sphinx # via sphinx
tornado==6.4.2
# via
# ipykernel
# jupyter-client
traitlets==5.14.3
# via
# comm
# ipykernel
# ipython
# jupyter-client
# jupyter-core
# matplotlib-inline
# nbclient
# nbformat
typing-extensions==4.12.2 typing-extensions==4.12.2
# via # via
# ipython
# myst-nb
# pydata-sphinx-theme # pydata-sphinx-theme
# pygithub # pygithub
# referencing
# sqlalchemy
urllib3==2.2.2 urllib3==2.2.2
# via # via
# pygithub # pygithub
# requests # requests
wcwidth==0.2.13
# via prompt-toolkit
wrapt==1.16.0 wrapt==1.16.0
# via deprecated # via deprecated
zipp==3.21.0
# via importlib-metadata
example/ck_tile/03_gemm/gemm_basic.cpp
View file @ b3054fea
// SPDX-License-Identifier: MIT // 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 <hip/hip_runtime.h> #include <hip/hip_runtime.h>
...@@ -49,7 +49,7 @@ float gemm_calc(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config& ...@@ -49,7 +49,7 @@ float gemm_calc(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config&
ck_tile::sequence<M_Warp, N_Warp, K_Warp>, ck_tile::sequence<M_Warp, N_Warp, K_Warp>,
ck_tile::sequence<M_Warp_Tile, N_Warp_Tile, K_Warp_Tile>>; ck_tile::sequence<M_Warp_Tile, N_Warp_Tile, K_Warp_Tile>>;
using TilePartitioner = ck_tile::GemmTilePartitioner<CodegenGemmShape>; using TilePartitioner = ck_tile::GemmTile2DPartitioner<CodegenGemmShape>;
using GemmEpilogue = std::conditional_t< using GemmEpilogue = std::conditional_t<
CShuffleEpilogue, CShuffleEpilogue,
...@@ -61,8 +61,8 @@ float gemm_calc(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config& ...@@ -61,8 +61,8 @@ float gemm_calc(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config&
kOutputRank, kOutputRank,
1, 1,
0, 0,
TilePartitioner::kM, TilePartitioner::MPerBlock,
TilePartitioner::kN>>, TilePartitioner::NPerBlock>>,
ck_tile::Default2DEpilogue< ck_tile::Default2DEpilogue<
ck_tile::Default2DEpilogueProblem<AccDataType, CDataType, kPadM, kPadN>>>; ck_tile::Default2DEpilogueProblem<AccDataType, CDataType, kPadM, kPadN>>>;
......
example/ck_tile/03_gemm/universal_gemm.cpp
View file @ b3054fea
...@@ -58,7 +58,7 @@ float gemm_calc(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config& ...@@ -58,7 +58,7 @@ float gemm_calc(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config&
ck_tile::TileGemmShape<ck_tile::sequence<M_Tile, N_Tile, K_Tile>, ck_tile::TileGemmShape<ck_tile::sequence<M_Tile, N_Tile, K_Tile>,
ck_tile::sequence<M_Warp, N_Warp, K_Warp>, ck_tile::sequence<M_Warp, N_Warp, K_Warp>,
ck_tile::sequence<M_Warp_Tile, N_Warp_Tile, K_Warp_Tile>>; ck_tile::sequence<M_Warp_Tile, N_Warp_Tile, K_Warp_Tile>>;
using TilePartitioner = ck_tile::GemmTilePartitioner<GemmShape>; using TilePartitioner = ck_tile::GemmTile2DPartitioner<GemmShape>;
using GemmEpilogue = ck_tile::Default2DEpilogue< using GemmEpilogue = ck_tile::Default2DEpilogue<
ck_tile::Default2DEpilogueProblem<AccDataType, CDataType, kPadM, kPadN>>; ck_tile::Default2DEpilogueProblem<AccDataType, CDataType, kPadM, kPadN>>;
......
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n1024_instance.cpp
View file @ b3054fea
...@@ -15,8 +15,13 @@ template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 16, 4, 64, 1, true ...@@ -15,8 +15,13 @@ template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 16, 4, 64, 1, true
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 1, 1, 256, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 1, 1, 256, 4, true, false>>(const S&, A);
#endif #endif
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 1, 2, 128, 8, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 1, 2, 128, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 2, 128, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 2, 2, 128, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 2, 128, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 4, 2, 128, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 1, 256, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 4, 1, 256, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 1, 2, 128, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 2, 2, 128, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 4, 2, 128, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 4, 1, 256, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n1536_instance.cpp
View file @ b3054fea
...@@ -6,8 +6,13 @@ ...@@ -6,8 +6,13 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 3, 4, 64, 8, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 3, 4, 64, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 3, 2, 128, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 3, 2, 128, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 3, 1, 256, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 3, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 6, 1, 256, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 6, 1, 256, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 3, 4, 64, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 3, 2, 128, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 3, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 6, 1, 256, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n2048_instance.cpp
View file @ b3054fea
...@@ -6,9 +6,14 @@ ...@@ -6,9 +6,14 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 1, 1, 256, 8, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 1, 1, 256, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 1, 256, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 2, 1, 256, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 1, 256, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 4, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 8, 1, 256, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 8, 1, 256, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 1, 1, 256, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 2, 1, 256, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 4, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 8, 1, 256, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n256_instance.cpp
View file @ b3054fea
...@@ -6,7 +6,11 @@ ...@@ -6,7 +6,11 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 1, 4, 64, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 1, 4, 64, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 4, 64, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 2, 4, 64, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 4, 64, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 4, 4, 64, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 1, 4, 64, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 2, 4, 64, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 4, 4, 64, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n3072_instance.cpp
View file @ b3054fea
...@@ -6,9 +6,13 @@ ...@@ -6,9 +6,13 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 3, 1, 128, 8, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 3, 1, 128, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 3, 1, 256, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 3, 1, 256, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 6, 1, 256, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 6, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 3, 1, 1024, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 3, 1, 1024, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 3, 1, 128, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 3, 1, 256, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 6, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 3, 1, 1024, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n4096_instance.cpp
View file @ b3054fea
...@@ -6,9 +6,13 @@ ...@@ -6,9 +6,13 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 1, 256, 8, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 2, 1, 256, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 1, 256, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 4, 1, 256, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 1, 1024, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 2, 1, 1024, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 1, 1024, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 4, 1, 1024, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 2, 1, 256, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 4, 1, 256, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 2, 1, 1024, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 4, 1, 1024, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n4096_tp_instance.cpp
View file @ b3054fea
...@@ -6,9 +6,13 @@ ...@@ -6,9 +6,13 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 1, 256, 8, true, true>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 2, 1, 256, 8, true, true>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 1, 256, 4, true, true>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 4, 1, 256, 4, true, true>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 1, 1024, 2, true, true>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 2, 1, 1024, 2, true, true>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 1, 1024, 1, true, true>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 4, 1, 1024, 1, true, true>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 2, 1, 256, 8, true, true>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 4, 1, 256, 4, true, true>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 2, 1, 1024, 2, true, true>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 4, 1, 1024, 1, true, true>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n512_instance.cpp
View file @ b3054fea
...@@ -6,8 +6,13 @@ ...@@ -6,8 +6,13 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 1, 4, 64, 8, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 1, 4, 64, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 4, 64, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 2, 4, 64, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 4, 4, 64, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 4, 4, 64, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 8, 4, 64, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 8, 4, 64, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 1, 4, 64, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 2, 4, 64, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 4, 4, 64, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 8, 4, 64, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n64_n128_instance.cpp
View file @ b3054fea
...@@ -6,7 +6,11 @@ ...@@ -6,7 +6,11 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 1, 4, 64, 1, true , false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 1, 4, 64, 1, true , false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 1, 4, 64, 2, true , false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 1, 4, 64, 2, true , false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 2, 4, 64, 1, true , false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 2, 4, 64, 1, true , false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 1, 4, 64, 1, true , false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 1, 4, 64, 2, true , false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 2, 4, 64, 1, true , false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_bf16_n768_instance.cpp
View file @ b3054fea
...@@ -6,7 +6,11 @@ ...@@ -6,7 +6,11 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 3, 4, 64, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 3, 4, 64, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 6, 4, 64, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 6, 4, 64, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, 1, 12, 4, 64, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::int8_t, 1, 12, 4, 64, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 3, 4, 64, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 6, 4, 64, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::bf16_t, ck_tile::fp8_t, 1, 12, 4, 64, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_fp16_n1024_instance.cpp
View file @ b3054fea
...@@ -15,8 +15,13 @@ template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 16, 4, 64, 1, true ...@@ -15,8 +15,13 @@ template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 16, 4, 64, 1, true
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 1, 1, 256, 4, true ,false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 1, 1, 256, 4, true ,false>>(const S&, A);
#endif #endif
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 1, 2, 128, 8, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 1, 2, 128, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 2, 2, 128, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 2, 2, 128, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 4, 2, 128, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 4, 2, 128, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 4, 1, 256, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 4, 1, 256, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 1, 2, 128, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 2, 2, 128, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 4, 2, 128, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 4, 1, 256, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_fp16_n1536_instance.cpp
View file @ b3054fea
...@@ -6,8 +6,13 @@ ...@@ -6,8 +6,13 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 3, 4, 64, 8, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 3, 4, 64, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 3, 2, 128, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 3, 2, 128, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 3, 1, 256, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 3, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 6, 1, 256, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 6, 1, 256, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 3, 4, 64, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 3, 2, 128, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 3, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 6, 1, 256, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_fp16_n2048_instance.cpp
View file @ b3054fea
...@@ -6,9 +6,13 @@ ...@@ -6,9 +6,13 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 1, 1, 256, 8, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 1, 1, 256, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 2, 1, 256, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 2, 1, 256, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 4, 1, 256, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 4, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 8, 1, 256, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 8, 1, 256, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 1, 1, 256, 8, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 2, 1, 256, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 4, 1, 256, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 8, 1, 256, 1, true, false>>(const S&, A);
// clang-format on // clang-format on
example/ck_tile/14_moe_smoothquant/instances/moe_smoothquant_fp16_n256_instance.cpp
View file @ b3054fea
...@@ -6,7 +6,11 @@ ...@@ -6,7 +6,11 @@
// clang-format off // clang-format off
// rm rn tm tn vn pd 2p // rm rn tm tn vn pd 2p
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 1, 4, 64, 4, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 1, 4, 64, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 2, 4, 64, 2, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 2, 4, 64, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, 1, 4, 4, 64, 1, true, false>>(const S&, A); template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::int8_t, 1, 4, 4, 64, 1, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 1, 4, 64, 4, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 2, 4, 64, 2, true, false>>(const S&, A);
template float moe_smoothquant_<trait_<ck_tile::fp16_t, ck_tile::fp8_t, 1, 4, 4, 64, 1, true, false>>(const S&, A);
// clang-format on // 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