Skip to content

GitLab

"test/batched_gemm/batched_gemm_fp16.cpp" did not exist on "485ea46a40f6ed9310443a33541b494d042c57a8"
Unverified Commit 189ea3b9 authored by Illia Silin's avatar Illia Silin Committed by GitHub
Browse files

Add mechanism to build CK for select data types, add Navi3x CI. (#790) 

* allow building CK for specific data types

* add CI build and test stage on Naiv3x without some int8 instances

* add missing gemm fp16 instances

* add the changes to the missed cmake file

* add empty lines at end of source files

* Do not build quantization client example on navi3 in CI

* disable batched_gemm_multi_d_int8 instances with DTYPES

* disable device_conv2d_bwd_data_instance with DTYPES

* fix ckprofiler for conv_bwd_data for int8

* properly isolate the conv_bwd_data int8 instances

* remove empty line
parent 4867db42
Hide whitespace changes
Inline Side-by-side
Showing with 56 additions and 22 deletions
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
View file @ 189ea3b9
......@@ -8,7 +8,7 @@
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef __int8__
namespace ck {
namespace tensor_operation {
namespace device {
......@@ -80,3 +80,4 @@ void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
View file @ 189ea3b9
......@@ -8,7 +8,7 @@
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef __int8__
namespace ck {
namespace tensor_operation {
namespace device {
......@@ -78,3 +78,4 @@ void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
View file @ 189ea3b9
......@@ -8,7 +8,7 @@
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef __int8__
namespace ck {
namespace tensor_operation {
namespace device {
......@@ -66,3 +66,4 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
View file @ 189ea3b9
......@@ -8,7 +8,7 @@
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef __int8__
namespace ck {
namespace tensor_operation {
namespace device {
......@@ -66,3 +66,4 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
View file @ 189ea3b9
......@@ -8,7 +8,7 @@
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef __int8__
namespace ck {
namespace tensor_operation {
namespace device {
......@@ -66,3 +66,4 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
View file @ 189ea3b9
......@@ -8,7 +8,7 @@
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#ifdef __int8__
namespace ck {
namespace tensor_operation {
namespace device {
......@@ -63,3 +63,4 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
} // namespace device
} // namespace tensor_operation
} // namespace ck
#endif
library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
View file @ 189ea3b9
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
set(CONV2D_PERLAYER_QUANT_SRC
conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
......@@ -36,3 +37,4 @@ add_instance_library(device_quantization_instance
${CONV2D_BIAS_PERCHANNEL_QUANT_SRC}
${GEMM_QUANT_SRC}
)
endif()
\ No newline at end of file
profiler/src/profile_batched_gemm_multi_d.cpp
View file @ 189ea3b9
......@@ -70,8 +70,10 @@ int profile_batched_gemm_multi_d(int argc, char* argv[])
const int BatchCount = std::stoi(argv[17]);
using F16 = ck::half_t;
using F16 = ck::half_t;
#ifdef __int8__
using INT8 = int8_t;
#endif
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
......@@ -163,6 +165,7 @@ int profile_batched_gemm_multi_d(int argc, char* argv[])
{
return profile(F16{}, F16{}, F16{}, Col{}, Col{}, Row{});
}
#ifdef __int8__
else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_KN_MN)
{
return profile(INT8{}, INT8{}, INT8{}, Row{}, Row{}, Row{});
......@@ -179,6 +182,7 @@ int profile_batched_gemm_multi_d(int argc, char* argv[])
{
return profile(INT8{}, INT8{}, INT8{}, Col{}, Col{}, Row{});
}
#endif
else
{
std::cout << "this data_type & layout is not implemented" << std::endl;
......
profiler/src/profile_conv_bwd_data.cpp
View file @ 189ea3b9
......@@ -77,7 +77,9 @@ int profile_conv_bwd_data(int argc, char* argv[])
using F32 = float;
using F16 = ck::half_t;
using BF16 = ck::bhalf_t;
#ifdef __int8__
using INT8 = int8_t;
#endif
using NWC = ck::tensor_layout::convolution::NWC;
using NHWC = ck::tensor_layout::convolution::NHWC;
......@@ -138,10 +140,12 @@ int profile_conv_bwd_data(int argc, char* argv[])
{
return profile(I1, NWC{}, KXC{}, NWK{}, BF16{}, BF16{}, BF16{});
}
#ifdef __int8__
else if(data_type == ConvDataType::INT8_INT8_INT8)
{
return profile(I1, NWC{}, KXC{}, NWK{}, INT8{}, INT8{}, INT8{});
}
#endif
}
else if(num_dim_spatial == 2 && layout == ConvLayout::NHWC_KYXC_NHWK)
{
......@@ -157,10 +161,12 @@ int profile_conv_bwd_data(int argc, char* argv[])
{
return profile(I2, NHWC{}, KYXC{}, NHWK{}, BF16{}, BF16{}, BF16{});
}
#ifdef __int8__
else if(data_type == ConvDataType::INT8_INT8_INT8)
{
return profile(I2, NHWC{}, KYXC{}, NHWK{}, INT8{}, INT8{}, INT8{});
}
#endif
}
else if(num_dim_spatial == 3 && layout == ConvLayout::NHWC_KYXC_NHWK)
{
......@@ -176,10 +182,12 @@ int profile_conv_bwd_data(int argc, char* argv[])
{
return profile(I3, NDHWC{}, KZYXC{}, NDHWK{}, BF16{}, BF16{}, BF16{});
}
#ifdef __int8__
else if(data_type == ConvDataType::INT8_INT8_INT8)
{
return profile(I3, NDHWC{}, KZYXC{}, NDHWK{}, INT8{}, INT8{}, INT8{});
}
#endif
}
std::cout << "this data_type & layout is not implemented" << std::endl;
......
profiler/src/profile_gemm.cpp
View file @ 189ea3b9
......@@ -67,11 +67,15 @@ int profile_gemm(int argc, char* argv[])
const int StrideB = std::stoi(argv[12]);
const int StrideC = std::stoi(argv[13]);
using F32 = float;
using F16 = ck::half_t;
using BF16 = ck::bhalf_t;
using F32 = float;
using F16 = ck::half_t;
#ifdef __bf16__
using BF16 = ck::bhalf_t;
#endif
#ifdef __int8__
using INT8 = int8_t;
using INT32 = int32_t;
#endif
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
......@@ -149,6 +153,7 @@ int profile_gemm(int argc, char* argv[])
{
return profile(Col{}, Col{}, Row{}, F16{}, F16{}, F32{}, F16{});
}
#ifdef __bf16__
else if(data_type == GemmDataType::BF16_BF16_BF16 && layout == GemmMatrixLayout::MK_KN_MN)
{
return profile(Row{}, Row{}, Row{}, BF16{}, BF16{}, F32{}, BF16{});
......@@ -165,6 +170,8 @@ int profile_gemm(int argc, char* argv[])
{
return profile(Col{}, Col{}, Row{}, BF16{}, BF16{}, F32{}, BF16{});
}
#endif
#ifdef __int8__
else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_KN_MN)
{
return profile(Row{}, Row{}, Row{}, INT8{}, INT8{}, INT32{}, INT8{});
......@@ -181,6 +188,7 @@ int profile_gemm(int argc, char* argv[])
{
return profile(Col{}, Col{}, Row{}, INT8{}, INT8{}, INT32{}, INT8{});
}
#endif
else
{
std::cout << "this data_type & layout is not implemented" << std::endl;
......
test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
View file @ 189ea3b9
......@@ -68,7 +68,9 @@ using KernelTypes = ::testing::Types<std::tuple<Row, Row, Row>,
} // namespace
TYPED_TEST_SUITE(TestBatchedGemmMultiD, KernelTypes);
#ifdef __fp16
TYPED_TEST(TestBatchedGemmMultiD, f16) { this->template Run<F16>(); }
#endif
#ifdef __int8__
TYPED_TEST(TestBatchedGemmMultiD, int8) { this->template Run<int8_t>(); }
#endif
test/gemm/CMakeLists.txt
View file @ 189ea3b9
if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
add_test_executable(test_gemm_fp32 gemm_fp32.cpp)
target_link_libraries(test_gemm_fp32 PRIVATE utility)
target_link_libraries(test_gemm_fp32 PRIVATE device_gemm_instance)
endif()
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
add_test_executable(test_gemm_fp16 gemm_fp16.cpp)
target_link_libraries(test_gemm_fp16 PRIVATE utility)
target_link_libraries(test_gemm_fp16 PRIVATE device_gemm_instance)
add_test_executable(test_gemm_bf16 gemm_bf16.cpp)
target_link_libraries(test_gemm_bf16 PRIVATE utility)
target_link_libraries(test_gemm_bf16 PRIVATE device_gemm_instance)
add_test_executable(test_gemm_int8 gemm_int8.cpp)
target_link_libraries(test_gemm_int8 PRIVATE utility)
target_link_libraries(test_gemm_int8 PRIVATE device_gemm_instance)
add_library(gemm_standalone_xdl_fp16_instances STATIC
instance/gemm_f16_nn_instance.cpp
instance/gemm_f16_nt_instance.cpp
......@@ -24,3 +17,14 @@ add_library(gemm_standalone_xdl_fp16_instances STATIC
add_test_executable(test_gemm_standalone_xdl_fp16 gemm_standalone_xdl_fp16.cpp)
target_link_libraries(test_gemm_standalone_xdl_fp16 PRIVATE gemm_standalone_xdl_fp16_instances utility)
target_include_directories(test_gemm_standalone_xdl_fp16 PRIVATE instance/)
endif()
if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
add_test_executable(test_gemm_bf16 gemm_bf16.cpp)
target_link_libraries(test_gemm_bf16 PRIVATE utility)
target_link_libraries(test_gemm_bf16 PRIVATE device_gemm_instance)
endif()
if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
add_test_executable(test_gemm_int8 gemm_int8.cpp)
target_link_libraries(test_gemm_int8 PRIVATE utility)
target_link_libraries(test_gemm_int8 PRIVATE device_gemm_instance)
endif()
\ No newline at end of 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