Skip to content

GitLab

Unverified Commit 48c85879 authored by rocking5566's avatar rocking5566 Committed by GitHub
Browse files

Merge branch 'develop' into conv_quant_int8

parents aa71a478 1b62bfaa
Hide whitespace changes
Inline Side-by-side
Showing with 71 additions and 93 deletions
profiler/src/profile_layernorm.cpp
View file @ 48c85879
......@@ -12,8 +12,7 @@ using ck::index_t;
struct LayernormArgParser
{
std::unordered_map<std::string, std::vector<int>> long_opts = {
{"length", {}}, {"strideXY", {}}, {"strideGamma", {}}, {"strideBeta", {}}};
std::unordered_map<std::string, std::vector<int>> long_opts = {{"length", {}}};
bool parse_opt(int argc, char* argv[], const std::string& key, int i)
{
......@@ -52,9 +51,6 @@ void print_help_layernorm()
<< "arg4: print tensor value (0: no; 1: yes)\n"
<< "arg5: time kernel (0=no, 1=yes)\n"
<< "--length: tensor extents (e.g, --length 1024 1024) \n"
<< "--strideXY: tensor strides (e.g, --strideXY 1024 1)\n"
<< "--strideGamma: tensor strides (e.g, --strideGamma 1)\n"
<< "--strideBeta: tensor strides (e.g, --strideBeta 1)\n"
<< std::endl;
}
......@@ -77,10 +73,7 @@ int profile_layernorm(int argc, char* argv[])
// parse the long options
arg_parser(argc, argv);
const std::vector<index_t> length = arg_parser.long_opts["length"];
const std::vector<index_t> strideXY = arg_parser.long_opts["strideXY"];
const std::vector<index_t> strideGamma = arg_parser.long_opts["strideGamma"];
const std::vector<index_t> strideBeta = arg_parser.long_opts["strideBeta"];
const std::vector<index_t> length = arg_parser.long_opts["length"];
using F16 = ck::half_t;
using F32 = float;
......@@ -88,25 +81,13 @@ int profile_layernorm(int argc, char* argv[])
if(data_type == ck::DataTypeEnum::Half)
{
ck::profiler::profile_layernorm_impl<F16, F16, F16, F32, F16, rank>(do_verification,
init_method,
do_log,
time_kernel,
length,
strideXY,
strideGamma,
strideBeta);
ck::profiler::profile_layernorm_impl<F16, F16, F16, F32, F16, rank>(
do_verification, init_method, do_log, time_kernel, length);
}
else if(data_type == ck::DataTypeEnum::Float)
{
ck::profiler::profile_layernorm_impl<F32, F32, F32, F32, F32, rank>(do_verification,
init_method,
do_log,
time_kernel,
length,
strideXY,
strideGamma,
strideBeta);
ck::profiler::profile_layernorm_impl<F32, F32, F32, F32, F32, rank>(
do_verification, init_method, do_log, time_kernel, length);
}
else
{
......
profiler/src/profile_normalization.cpp → profiler/src/profile_softmax.cpp
View file @ 48c85879
......@@ -5,7 +5,7 @@
#include <vector>
#include <unordered_map>
#include "profiler/include/profile_normalization_impl.hpp"
#include "profiler/include/profile_softmax_impl.hpp"
using ck::index_t;
using ck::profiler::NormDataType;
......@@ -95,30 +95,29 @@ int profile_normalization(int argc, char* argv[])
{
if(data_type == NormDataType::F16_F16)
{
ck::profiler::profile_normalization_impl<ck::half_t, float, ck::half_t, 3>(
do_verification,
init_method,
do_log,
time_kernel,
length,
stride,
reduce,
float(alpha),
float(beta),
norm_type);
ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 3>(do_verification,
init_method,
do_log,
time_kernel,
length,
stride,
reduce,
float(alpha),
float(beta),
norm_type);
}
else if(data_type == NormDataType::F32_F32)
{
ck::profiler::profile_normalization_impl<float, float, float, 3>(do_verification,
init_method,
do_log,
time_kernel,
length,
stride,
reduce,
float(alpha),
float(beta),
norm_type);
ck::profiler::profile_softmax_impl<float, float, float, 3>(do_verification,
init_method,
do_log,
time_kernel,
length,
stride,
reduce,
float(alpha),
float(beta),
norm_type);
}
else
{
......@@ -129,30 +128,29 @@ int profile_normalization(int argc, char* argv[])
{
if(data_type == NormDataType::F16_F16)
{
ck::profiler::profile_normalization_impl<ck::half_t, float, ck::half_t, 4>(
do_verification,
init_method,
do_log,
time_kernel,
length,
stride,
reduce,
float(alpha),
float(beta),
norm_type);
ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4>(do_verification,
init_method,
do_log,
time_kernel,
length,
stride,
reduce,
float(alpha),
float(beta),
norm_type);
}
else if(data_type == NormDataType::F32_F32)
{
ck::profiler::profile_normalization_impl<float, float, float, 4>(do_verification,
init_method,
do_log,
time_kernel,
length,
stride,
reduce,
float(alpha),
float(beta),
norm_type);
ck::profiler::profile_softmax_impl<float, float, float, 4>(do_verification,
init_method,
do_log,
time_kernel,
length,
stride,
reduce,
float(alpha),
float(beta),
norm_type);
}
else
{
......
test/CMakeLists.txt
View file @ 48c85879
......@@ -6,11 +6,10 @@ include(googletest)
add_custom_target(tests)
function(add_test_executable TEST_NAME)
message("adding test ${TEST_NAME}")
add_executable(${TEST_NAME} ${ARGN})
add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}> )
add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}>)
add_dependencies(tests ${TEST_NAME})
add_dependencies(check ${TEST_NAME})
rocm_install(TARGETS ${TEST_NAME} COMPONENT tests)
......@@ -23,6 +22,7 @@ function(add_gtest_executable TEST_NAME)
add_executable(${TEST_NAME} ${ARGN})
add_dependencies(tests ${TEST_NAME})
add_dependencies(check ${TEST_NAME})
# suppress gtest warnings
target_compile_options(${TEST_NAME} PRIVATE -Wno-global-constructors -Wno-undef)
target_link_libraries(${TEST_NAME} PRIVATE gtest_main)
......@@ -30,7 +30,6 @@ function(add_gtest_executable TEST_NAME)
rocm_install(TARGETS ${TEST_NAME} COMPONENT tests)
endfunction(add_gtest_executable TEST_NAME)
add_subdirectory(magic_number_division)
add_subdirectory(space_filling_curve)
add_subdirectory(conv_util)
......@@ -51,5 +50,5 @@ add_subdirectory(convnd_bwd_data)
add_subdirectory(grouped_convnd_fwd)
add_subdirectory(block_to_ctile_map)
add_subdirectory(softmax)
add_subdirectory(layernorm)
add_subdirectory(normalization)
add_subdirectory(data_type)
test/layernorm/test_layernorm2d_util.hpp → test/normalization/test_layernorm2d_util.hpp
View file @ 48c85879
......@@ -9,7 +9,7 @@
#include "ck/ck.hpp"
#include "ck/utility/number.hpp"
#include "ck/tensor_operation/gpu/device/device_layernorm_impl.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization_impl.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/host_tensor.hpp"
......@@ -65,26 +65,26 @@ class TestLayernorm2d : public ::testing::Test
Rank,
NumReduceDim>;
using DeviceInstance = tensor_operation::device::DeviceLayernormImpl<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
YDataType,
PassThrough,
Rank,
NumReduceDim,
BlockSize,
MThreadClusterSize,
KThreadClusterSize,
MThreadSliceSize,
KThreadSliceSize,
XYSrcVectorDim,
XSrcVectorSize,
GammaSrcVectorDim,
GammaSrcVectorSize,
BetaSrcVectorDim,
BetaSrcVectorSize,
YDstVectorSize>;
using DeviceInstance = tensor_operation::device::DeviceNormalizationImpl<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
YDataType,
PassThrough,
Rank,
NumReduceDim,
BlockSize,
MThreadClusterSize,
KThreadClusterSize,
MThreadSliceSize,
KThreadSliceSize,
XYSrcVectorDim,
XSrcVectorSize,
GammaSrcVectorDim,
GammaSrcVectorSize,
BetaSrcVectorDim,
BetaSrcVectorSize,
YDstVectorSize>;
TestLayernorm2d() : ref_instance_invoker_(ReferenceInstance{}.MakeInvoker()) {}
......
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