File reorganization
Move the tracer_tool from the 'test' directory to the 'src' directory. Change-Id: I13768b9610cd359f78a66147f0255ab1e4c657e9
このコミットが含まれているのは:
Laurent Morichetti
+9
-22
@@ -32,16 +32,16 @@ set(GEN_INC_DIR ${PROJECT_BINARY_DIR}/inc)
|
||||
set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "/opt/rocm/hip/cmake")
|
||||
find_package(HIP REQUIRED MODULE)
|
||||
|
||||
set_source_files_properties(MatrixTranspose/MatrixTranspose.cpp MatrixTranspose_test/MatrixTranspose.cpp
|
||||
set_source_files_properties(hip/MatrixTranspose.cpp app/MatrixTranspose_test.cpp
|
||||
PROPERTIES HIP_SOURCE_PROPERTY_FORMAT 1)
|
||||
|
||||
hip_add_executable(MatrixTranspose EXCLUDE_FROM_ALL MatrixTranspose/MatrixTranspose.cpp)
|
||||
hip_add_executable(MatrixTranspose EXCLUDE_FROM_ALL hip/MatrixTranspose.cpp)
|
||||
target_include_directories(MatrixTranspose PRIVATE ${PROJECT_SOURCE_DIR}/inc)
|
||||
target_link_libraries(MatrixTranspose PRIVATE roctracer roctx)
|
||||
add_dependencies(mytest MatrixTranspose)
|
||||
|
||||
function(build_matrix_transpose_test OUTPUT_FILE DEFINITIONS)
|
||||
hip_add_executable(${OUTPUT_FILE} EXCLUDE_FROM_ALL MatrixTranspose_test/MatrixTranspose.cpp)
|
||||
hip_add_executable(${OUTPUT_FILE} EXCLUDE_FROM_ALL app/MatrixTranspose_test.cpp)
|
||||
target_compile_definitions(${OUTPUT_FILE} PRIVATE ITERATIONS=100 HIP_TEST=1 ${DEFINITIONS})
|
||||
target_include_directories(${OUTPUT_FILE} PRIVATE ${PROJECT_SOURCE_DIR}/inc ${GEN_INC_DIR})
|
||||
target_link_libraries(${OUTPUT_FILE} PRIVATE roctracer roctx)
|
||||
@@ -54,7 +54,7 @@ build_matrix_transpose_test(MatrixTranspose_hipaact_test HIP_API_ACTIVITY_ON=1)
|
||||
build_matrix_transpose_test(MatrixTranspose_mgpu MGPU_TEST=1)
|
||||
|
||||
add_custom_command(OUTPUT MatrixTranspose.c
|
||||
COMMAND ${CMAKE_COMMAND} -E create_symlink ${CMAKE_CURRENT_SOURCE_DIR}/MatrixTranspose_test/MatrixTranspose.cpp MatrixTranspose.c)
|
||||
COMMAND ${CMAKE_COMMAND} -E create_symlink ${CMAKE_CURRENT_SOURCE_DIR}/app/MatrixTranspose_test.cpp MatrixTranspose.c)
|
||||
|
||||
hip_add_executable(MatrixTranspose_ctest EXCLUDE_FROM_ALL MatrixTranspose.c)
|
||||
target_compile_definitions(MatrixTranspose_ctest PRIVATE HIP_TEST=0 __HIP_PLATFORM_HCC__)
|
||||
@@ -62,27 +62,14 @@ target_include_directories(MatrixTranspose_ctest PRIVATE ${PROJECT_SOURCE_DIR}/i
|
||||
target_link_libraries(MatrixTranspose_ctest PRIVATE roctracer roctx)
|
||||
add_dependencies(mytest MatrixTranspose_ctest)
|
||||
|
||||
file(GLOB files "${CMAKE_CURRENT_SOURCE_DIR}/golden_traces/*_trace.txt")
|
||||
file(GLOB files RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} "golden_traces/tests_trace_cmp_levels.txt" "golden_traces/*_trace.txt")
|
||||
foreach(file ${files})
|
||||
execute_process(COMMAND ${CMAKE_COMMAND} -E copy ${file} ${PROJECT_BINARY_DIR}/test/)
|
||||
configure_file(${file} ${PROJECT_BINARY_DIR}/test/${file} COPYONLY)
|
||||
endforeach()
|
||||
execute_process(COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/golden_traces/tests_trace_cmp_levels.txt ${PROJECT_BINARY_DIR}/test/)
|
||||
|
||||
## Build HSA test
|
||||
add_subdirectory(hsa/test ${PROJECT_BINARY_DIR}/test/hsa)
|
||||
|
||||
if(DEFINED ROCTRACER_TARGET)
|
||||
## Build the tracer_tool library
|
||||
file(GLOB TRACER_TOOL_SOURCES "tool/*.cpp" "${PROJECT_SOURCE_DIR}/src/util/*.cpp")
|
||||
add_library(roctracer_tool SHARED ${TRACER_TOOL_SOURCES})
|
||||
target_compile_definitions(roctracer_tool PRIVATE HIP_PROF_HIP_API_STRING=1 __HIP_PLATFORM_HCC__)
|
||||
target_include_directories(roctracer_tool PRIVATE hsa/test ${PROJECT_SOURCE_DIR} ${PROJECT_SOURCE_DIR}/inc ${HIP_INCLUDE_DIRECTORIES} ${GEN_INC_DIR})
|
||||
target_link_libraries(roctracer_tool ${ROCTRACER_TARGET} hsa-runtime64::hsa-runtime64 Threads::Threads atomic dl)
|
||||
set_target_properties(roctracer_tool PROPERTIES CXX_VISIBILITY_PRESET hidden LINK_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/tool/exportmap)
|
||||
target_link_options(roctracer_tool PRIVATE -Wl,--version-script=${CMAKE_CURRENT_SOURCE_DIR}/tool/exportmap -Wl,--no-undefined)
|
||||
install(TARGETS roctracer_tool LIBRARY DESTINATION lib/${ROCTRACER_NAME})
|
||||
endif ()
|
||||
|
||||
## Build hsaco_test.cpp referenc test
|
||||
add_library(hsaco_test SHARED EXCLUDE_FROM_ALL app/hsaco_test.cpp)
|
||||
target_compile_definitions(hsaco_test PRIVATE AMD_INTERNAL_BUILD)
|
||||
@@ -97,12 +84,12 @@ add_dependencies(mytest codeobj_test)
|
||||
|
||||
## Build the trace_buffer test
|
||||
add_executable(trace_buffer EXCLUDE_FROM_ALL directed/trace_buffer.cpp)
|
||||
target_include_directories(trace_buffer PRIVATE ${PROJECT_SOURCE_DIR}/test/tool)
|
||||
target_include_directories(trace_buffer PRIVATE ${PROJECT_SOURCE_DIR}/src/tracer_tool)
|
||||
target_link_libraries(trace_buffer Threads::Threads atomic)
|
||||
add_dependencies(mytest trace_buffer)
|
||||
|
||||
## copying run script
|
||||
execute_process(COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/run.sh ${PROJECT_BINARY_DIR})
|
||||
configure_file(run.sh ${PROJECT_BINARY_DIR} COPYONLY)
|
||||
execute_process(COMMAND ${CMAKE_COMMAND} -E create_symlink run.sh ${PROJECT_BINARY_DIR}/run_ci.sh)
|
||||
## copying tests output check script
|
||||
execute_process(COMMAND ${CMAKE_COMMAND} -E copy ${PROJECT_SOURCE_DIR}/script/check_trace.py ${PROJECT_BINARY_DIR}/test/.)
|
||||
configure_file(${PROJECT_SOURCE_DIR}/script/check_trace.py ${PROJECT_BINARY_DIR}/test/check_trace.py COPYONLY)
|
||||
|
||||
@@ -1,97 +0,0 @@
|
||||
## Writing first HIP program ###
|
||||
|
||||
This tutorial shows how to get write simple HIP application. We will write the simplest Matrix Transpose program.
|
||||
|
||||
## HIP Introduction:
|
||||
|
||||
HIP is a C++ runtime API and kernel language that allows developers to create portable applications that can run on AMD and other GPU’s. Our goal was to rise above the lowest-common-denominator paths and deliver a solution that allows you, the developer, to use essential hardware features and maximize your application’s performance on GPU hardware.
|
||||
|
||||
## Requirement:
|
||||
For hardware requirement and software installation [Installation](https://github.com/ROCm-Developer-Tools/HIP/INSTALL.md)
|
||||
|
||||
## prerequiste knowledge:
|
||||
|
||||
Programmers familiar with CUDA, OpenCL will be able to quickly learn and start coding with the HIP API. In case you are not, don't worry. You choose to start with the best one. We'll be explaining everything assuming you are completely new to gpgpu programming.
|
||||
|
||||
## Simple Matrix Transpose
|
||||
|
||||
Here is simple example showing how to write your first program in HIP.
|
||||
In order to use the HIP framework, we need to add the "hip_runtime.h" header file. SInce its c++ api you can add any header file you have been using earlier while writing your c/c++ program. For gpgpu programming, we have host(microprocessor) and the device(gpu).
|
||||
|
||||
## Device-side code
|
||||
We will work on device side code first, Here is simple example showing a snippet of HIP device side code:
|
||||
|
||||
`__global__ void matrixTranspose(float *out, `
|
||||
` float *in, `
|
||||
` const int width, `
|
||||
` const int height) `
|
||||
`{ `
|
||||
` int x = hipBlockDim_x * hipBlockIdx_x + hipThreadIdx_x; `
|
||||
` int y = hipBlockDim_y * hipBlockIdx_y + hipThreadIdx_y; `
|
||||
` `
|
||||
` out[y * width + x] = in[x * height + y]; `
|
||||
`} `
|
||||
|
||||
`__global__` keyword is the Function-Type Qualifiers, it is used with functions that are executed on device and are called/launched from the hosts.
|
||||
other function-type qualifiers are:
|
||||
`__device__` functions are Executed on the device and Called from the device only
|
||||
`__host__` functions are Executed on the host and Called from the host
|
||||
|
||||
`__host__` can combine with `__device__`, in which case the function compiles for both the host and device. These functions cannot use the HIP grid coordinate functions (for example, "hipThreadIdx_x", will talk about it latter). A possible workaround is to pass the necessary coordinate info as an argument to the function.
|
||||
`__host__` cannot combine with `__global__`.
|
||||
|
||||
`__global__` functions are often referred to as *kernels*, and calling one is termed *launching the kernel*.
|
||||
|
||||
Next keyword is `void`. HIP `__global__` functions must have a `void` return type. Global functions require the caller to specify an "execution configuration" that includes the grid and block dimensions. The execution configuration can also include other information for the launch, such as the amount of additional shared memory to allocate and the stream where the kernel should execute.
|
||||
|
||||
The kernel function begins with
|
||||
` int x = hipBlockDim_x * hipBlockIdx_x + hipThreadIdx_x;`
|
||||
` int y = hipBlockDim_y * hipBlockIdx_y + hipThreadIdx_y;`
|
||||
here the keyword hipBlockIdx_x, hipBlockIdx_y and hipBlockIdx_z(not used here) are the built-in functions to identify the threads in a block. The keyword hipBlockDim_x, hipBlockDim_y and hipBlockDim_z(not used here) are to identify the dimensions of the block.
|
||||
|
||||
We are familiar with rest of the code on device-side.
|
||||
|
||||
## Host-side code
|
||||
|
||||
Now, we'll see how to call the kernel from the host. Inside the main() function, we first defined the pointers(for both, the host-side as well as device). The declaration of device pointer is similar to that of the host. Next, we have `hipDeviceProp_t`, it is the pre-defined struct for hip device properties. This is followed by `hipGetDeviceProperties(&devProp, 0)` It is used to extract the device information. The first parameter is the struct, second parameter is the device number to get properties for. Next line print the name of the device.
|
||||
|
||||
We allocated memory to the Matrix on host side by using malloc and initiallized it. While in order to allocate memory on device side we will be using `hipMalloc`, it's quiet similar to that of malloc instruction. After this, we will copy the data to the allocated memory on device-side using `hipMemcpy`.
|
||||
` hipMemcpy(gpuMatrix, Matrix, NUM*sizeof(float), hipMemcpyHostToDevice);`
|
||||
here the first parameter is the destination pointer, second is the source pointer, third is the size of memory copy and the last specify the direction on memory copy(which is in this case froom host to device). While in order to transfer memory from device to host, use `hipMemcpyDeviceToHost` and for device to device memory copy use `hipMemcpyDeviceToDevice`.
|
||||
|
||||
Now, we'll see how to launch the kernel.
|
||||
` hipLaunchKernelGGL(matrixTranspose, `
|
||||
` dim3(WIDTH/THREADS_PER_BLOCK_X, HEIGHT/THREADS_PER_BLOCK_Y), `
|
||||
` dim3(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y), `
|
||||
` 0, 0, `
|
||||
` gpuTransposeMatrix , gpuMatrix, WIDTH ,HEIGHT); `
|
||||
|
||||
HIP introduces a standard C++ calling convention to pass the execution configuration to the kernel (this convention replaces the `Cuda <<< >>>` syntax). In HIP,
|
||||
- Kernels launch with the `"hipLaunchKernelGGL"` function
|
||||
- The first five parameters to hipLaunchKernelGGL are the following:
|
||||
- **symbol kernelName**: the name of the kernel to launch. To support template kernels which contains "," use the HIP_KERNEL_NAME macro. In current application it's "matrixTranspose".
|
||||
- **dim3 gridDim**: 3D-grid dimensions specifying the number of blocks to launch. In MatrixTranspose sample, it's "dim3(WIDTH/THREADS_PER_BLOCK_X, HEIGHT/THREADS_PER_BLOCK_Y)".
|
||||
- **dim3 blockDim**: 3D-block dimensions specifying the number of threads in each block.In MatrixTranspose sample, it's "dim3(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y)".
|
||||
- **size_t dynamicShared**: amount of additional shared memory to allocate when launching the kernel. In MatrixTranspose sample, it's '0'.
|
||||
- **hipStream_t**: stream where the kernel should execute. A value of 0 corresponds to the NULL stream.In MatrixTranspose sample, it's '0'.
|
||||
- Kernel arguments follow these first five parameters. Here, these are "gpuTransposeMatrix , gpuMatrix, WIDTH ,HEIGHT".
|
||||
|
||||
Next, we'll copy the computed values/data back to the device using the `hipMemcpy`. Here the last parameter will be `hipMemcpyDeviceToHost`
|
||||
|
||||
After, copying the data from device to memory, we will verify it with the one we computed with the cpu reference funtion.
|
||||
|
||||
Finally, we will free the memory allocated earlier by using free() for host while for devices we will use `hipFree`.
|
||||
|
||||
## How to build and run:
|
||||
Use the make command and execute it using ./exe
|
||||
Use hipcc to build the application, which is using hcc on AMD and nvcc on nvidia.
|
||||
|
||||
## More Info:
|
||||
- [HIP FAQ](https://github.com/ROCm-Developer-Tools/HIP/docs/markdown/hip_faq.md)
|
||||
- [HIP Kernel Language](https://github.com/ROCm-Developer-Tools/HIP/docs/markdown/hip_kernel_language.md)
|
||||
- [HIP Runtime API (Doxygen)](http://rocm-developer-tools.github.io/HIP)
|
||||
- [HIP Porting Guide](https://github.com/ROCm-Developer-Tools/HIP/docs/markdown/hip_porting_guide.md)
|
||||
- [HIP Terminology](https://github.com/ROCm-Developer-Tools/HIP/docs/markdown/hip_terms.md) (including Rosetta Stone of GPU computing terms across CUDA/HIP/HC/AMP/OpenL)
|
||||
- [hipify-clang](https://github.com/ROCm-Developer-Tools/HIP/hipify-clang/README.md)
|
||||
- [Developer/CONTRIBUTING Info](https://github.com/ROCm-Developer-Tools/HIP/CONTRIBUTING.md)
|
||||
- [Release Notes](https://github.com/ROCm-Developer-Tools/HIP/RELEASE.md)
|
||||
@@ -1,101 +0,0 @@
|
||||
## Writing first HIP program ###
|
||||
|
||||
This tutorial shows how to get write simple HIP application. We will write the simplest Matrix Transpose program.
|
||||
|
||||
## HIP Introduction:
|
||||
|
||||
HIP is a C++ runtime API and kernel language that allows developers to create portable applications that can run on AMD and other GPU’s. Our goal was to rise above the lowest-common-denominator paths and deliver a solution that allows you, the developer, to use essential hardware features and maximize your application’s performance on GPU hardware.
|
||||
|
||||
## Requirement:
|
||||
For hardware requirement and software installation [Installation](https://github.com/ROCm-Developer-Tools/HIP/INSTALL.md)
|
||||
|
||||
## prerequiste knowledge:
|
||||
|
||||
Programmers familiar with CUDA, OpenCL will be able to quickly learn and start coding with the HIP API. In case you are not, don't worry. You choose to start with the best one. We'll be explaining everything assuming you are completely new to gpgpu programming.
|
||||
|
||||
## Simple Matrix Transpose
|
||||
|
||||
Here is simple example showing how to write your first program in HIP.
|
||||
In order to use the HIP framework, we need to add the "hip_runtime.h" header file. SInce its c++ api you can add any header file you have been using earlier while writing your c/c++ program. For gpgpu programming, we have host(microprocessor) and the device(gpu).
|
||||
|
||||
## Device-side code
|
||||
We will work on device side code first, Here is simple example showing a snippet of HIP device side code:
|
||||
|
||||
```
|
||||
__global__ void matrixTranspose(float *out,
|
||||
float *in,
|
||||
const int width,
|
||||
const int height)
|
||||
{
|
||||
int x = hipBlockDim_x * hipBlockIdx_x + hipThreadIdx_x;
|
||||
int y = hipBlockDim_y * hipBlockIdx_y + hipThreadIdx_y;
|
||||
|
||||
out[y * width + x] = in[x * height + y];
|
||||
}
|
||||
```
|
||||
|
||||
`__global__` keyword is the Function-Type Qualifiers, it is used with functions that are executed on device and are called/launched from the hosts.
|
||||
other function-type qualifiers are:
|
||||
`__device__` functions are Executed on the device and Called from the device only
|
||||
`__host__` functions are Executed on the host and Called from the host
|
||||
|
||||
`__host__` can combine with `__device__`, in which case the function compiles for both the host and device. These functions cannot use the HIP grid coordinate functions (for example, "hipThreadIdx_x", will talk about it latter). A possible workaround is to pass the necessary coordinate info as an argument to the function.
|
||||
`__host__` cannot combine with `__global__`.
|
||||
|
||||
`__global__` functions are often referred to as *kernels*, and calling one is termed *launching the kernel*.
|
||||
|
||||
Next keyword is `void`. HIP `__global__` functions must have a `void` return type. Global functions require the caller to specify an "execution configuration" that includes the grid and block dimensions. The execution configuration can also include other information for the launch, such as the amount of additional shared memory to allocate and the stream where the kernel should execute.
|
||||
|
||||
The kernel function begins with
|
||||
` int x = hipBlockDim_x * hipBlockIdx_x + hipThreadIdx_x;`
|
||||
` int y = hipBlockDim_y * hipBlockIdx_y + hipThreadIdx_y;`
|
||||
here the keyword hipBlockIdx_x, hipBlockIdx_y and hipBlockIdx_z(not used here) are the built-in functions to identify the threads in a block. The keyword hipBlockDim_x, hipBlockDim_y and hipBlockDim_z(not used here) are to identify the dimensions of the block.
|
||||
|
||||
We are familiar with rest of the code on device-side.
|
||||
|
||||
## Host-side code
|
||||
|
||||
Now, we'll see how to call the kernel from the host. Inside the main() function, we first defined the pointers(for both, the host-side as well as device). The declaration of device pointer is similar to that of the host. Next, we have `hipDeviceProp_t`, it is the pre-defined struct for hip device properties. This is followed by `hipGetDeviceProperties(&devProp, 0)` It is used to extract the device information. The first parameter is the struct, second parameter is the device number to get properties for. Next line print the name of the device.
|
||||
|
||||
We allocated memory to the Matrix on host side by using malloc and initiallized it. While in order to allocate memory on device side we will be using `hipMalloc`, it's quiet similar to that of malloc instruction. After this, we will copy the data to the allocated memory on device-side using `hipMemcpy`.
|
||||
` hipMemcpy(gpuMatrix, Matrix, NUM*sizeof(float), hipMemcpyHostToDevice);`
|
||||
here the first parameter is the destination pointer, second is the source pointer, third is the size of memory copy and the last specify the direction on memory copy(which is in this case froom host to device). While in order to transfer memory from device to host, use `hipMemcpyDeviceToHost` and for device to device memory copy use `hipMemcpyDeviceToDevice`.
|
||||
|
||||
Now, we'll see how to launch the kernel.
|
||||
```
|
||||
hipLaunchKernelGGL(matrixTranspose,
|
||||
dim3(WIDTH/THREADS_PER_BLOCK_X, HEIGHT/THREADS_PER_BLOCK_Y),
|
||||
dim3(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y),
|
||||
0, 0,
|
||||
gpuTransposeMatrix , gpuMatrix, WIDTH ,HEIGHT);
|
||||
```
|
||||
|
||||
HIP introduces a standard C++ calling convention to pass the execution configuration to the kernel (this convention replaces the `Cuda <<< >>>` syntax). In HIP,
|
||||
- Kernels launch with the `"hipLaunchKernelGGL"` function
|
||||
- The first five parameters to hipLaunchKernelGGL are the following:
|
||||
- **symbol kernelName**: the name of the kernel to launch. To support template kernels which contains "," use the HIP_KERNEL_NAME macro. In current application it's "matrixTranspose".
|
||||
- **dim3 gridDim**: 3D-grid dimensions specifying the number of blocks to launch. In MatrixTranspose sample, it's "dim3(WIDTH/THREADS_PER_BLOCK_X, HEIGHT/THREADS_PER_BLOCK_Y)".
|
||||
- **dim3 blockDim**: 3D-block dimensions specifying the number of threads in each block.In MatrixTranspose sample, it's "dim3(THREADS_PER_BLOCK_X, THREADS_PER_BLOCK_Y)".
|
||||
- **size_t dynamicShared**: amount of additional shared memory to allocate when launching the kernel. In MatrixTranspose sample, it's '0'.
|
||||
- **hipStream_t**: stream where the kernel should execute. A value of 0 corresponds to the NULL stream.In MatrixTranspose sample, it's '0'.
|
||||
- Kernel arguments follow these first five parameters. Here, these are "gpuTransposeMatrix , gpuMatrix, WIDTH ,HEIGHT".
|
||||
|
||||
Next, we'll copy the computed values/data back to the device using the `hipMemcpy`. Here the last parameter will be `hipMemcpyDeviceToHost`
|
||||
|
||||
After, copying the data from device to memory, we will verify it with the one we computed with the cpu reference funtion.
|
||||
|
||||
Finally, we will free the memory allocated earlier by using free() for host while for devices we will use `hipFree`.
|
||||
|
||||
## How to build and run:
|
||||
Use the make command and execute it using ./exe
|
||||
Use hipcc to build the application, which is using hcc on AMD and nvcc on nvidia.
|
||||
|
||||
## More Info:
|
||||
- [HIP FAQ](https://github.com/ROCm-Developer-Tools/HIP/docs/markdown/hip_faq.md)
|
||||
- [HIP Kernel Language](https://github.com/ROCm-Developer-Tools/HIP/docs/markdown/hip_kernel_language.md)
|
||||
- [HIP Runtime API (Doxygen)](http://rocm-developer-tools.github.io/HIP)
|
||||
- [HIP Porting Guide](https://github.com/ROCm-Developer-Tools/HIP/docs/markdown/hip_porting_guide.md)
|
||||
- [HIP Terminology](https://github.com/ROCm-Developer-Tools/HIP/docs/markdown/hip_terms.md) (including Rosetta Stone of GPU computing terms across CUDA/HIP/HC/AMP/OpenL)
|
||||
- [hipify-clang](https://github.com/ROCm-Developer-Tools/HIP/hipify-clang/README.md)
|
||||
- [Developer/CONTRIBUTING Info](https://github.com/ROCm-Developer-Tools/HIP/CONTRIBUTING.md)
|
||||
- [Release Notes](https://github.com/ROCm-Developer-Tools/HIP/RELEASE.md)
|
||||
+9
-9
@@ -46,7 +46,7 @@ if [ -z "$ROCTRACER_LIB_PATH" ] ; then
|
||||
ROCTRACER_LIB_PATH="."
|
||||
fi
|
||||
if [ -z "$ROCTRACER_TOOL_PATH" ] ; then
|
||||
ROCTRACER_TOOL_PATH="./test"
|
||||
ROCTRACER_TOOL_PATH="."
|
||||
fi
|
||||
|
||||
# test filter input
|
||||
@@ -74,19 +74,19 @@ eval_test() {
|
||||
label=$1
|
||||
cmdline=$2
|
||||
test_name=$3
|
||||
test_trace=$test_name.txt
|
||||
|
||||
if [ $test_filter = -1 -o $test_filter = $test_number ] ; then
|
||||
echo "test $test_number: $test_name \"$label\""
|
||||
echo "CMD: \"$cmdline\""
|
||||
mkdir -p test/out
|
||||
test_runnum=$((test_runnum + 1))
|
||||
eval "$cmdline" 1>$test_trace 2>$test_name.err
|
||||
eval "$cmdline" 1>test/out/$test_name.out 2>test/out/$test_name.err
|
||||
is_failed=$?
|
||||
if [ $is_failed != 0 ] ; then
|
||||
echo "--- stdout ---"
|
||||
cat $test_trace
|
||||
cat test/out/$test_name.out
|
||||
echo "--- stderr ---"
|
||||
cat $test_name.err
|
||||
cat test/out/$test_name.err
|
||||
fi
|
||||
if [ $IS_CI = 1 ] ; then
|
||||
is_failed=0;
|
||||
@@ -138,8 +138,8 @@ eval_test "tool period test" "ROCP_CTRL_RATE=10:50000:500000 ./test/MatrixTransp
|
||||
eval_test "tool flushing test" "ROCP_FLUSH_RATE=100000 ./test/MatrixTranspose" MatrixTranspose_hip_flush_trace
|
||||
|
||||
#API records filtering
|
||||
echo "<trace name=\"HIP\"><parameters api=\"hipFree, hipMalloc, hipMemcpy\"></parameters></trace>" > input.xml
|
||||
export ROCP_INPUT=input.xml
|
||||
echo "<trace name=\"HIP\"><parameters api=\"hipFree, hipMalloc, hipMemcpy\"></parameters></trace>" > test/input.xml
|
||||
export ROCP_INPUT=test/input.xml
|
||||
eval_test "tool HIP test input" ./test/MatrixTranspose MatrixTranspose_hip_input_trace
|
||||
unset ROCP_INPUT
|
||||
|
||||
@@ -160,8 +160,8 @@ export ROCP_THRS=1
|
||||
|
||||
eval_test "tool HSA test" ./test/hsa/ctrl ctrl_hsa_trace
|
||||
|
||||
echo "<trace name=\"HSA\"><parameters api=\"hsa_agent_get_info, hsa_amd_memory_pool_allocate\"></parameters></trace>" > input.xml
|
||||
export ROCP_INPUT=input.xml
|
||||
echo "<trace name=\"HSA\"><parameters api=\"hsa_agent_get_info, hsa_amd_memory_pool_allocate\"></parameters></trace>" > test/input.xml
|
||||
export ROCP_INPUT=test/input.xml
|
||||
eval_test "tool HSA test input" ./test/hsa/ctrl ctrl_hsa_input_trace
|
||||
unset ROCP_INPUT
|
||||
|
||||
|
||||
@@ -1 +0,0 @@
|
||||
{ global: OnLoad; OnUnload; local: *; };
|
||||
@@ -1,270 +0,0 @@
|
||||
/* Copyright (c) 2018-2022 Advanced Micro Devices, Inc.
|
||||
|
||||
Permission is hereby granted, free of charge, to any person obtaining a copy
|
||||
of this software and associated documentation files (the "Software"), to deal
|
||||
in the Software without restriction, including without limitation the rights
|
||||
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||||
copies of the Software, and to permit persons to whom the Software is
|
||||
furnished to do so, subject to the following conditions:
|
||||
|
||||
The above copyright notice and this permission notice shall be included in
|
||||
all copies or substantial portions of the Software.
|
||||
|
||||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
||||
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||||
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||||
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||||
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||||
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
||||
THE SOFTWARE. */
|
||||
|
||||
#ifndef TOOL_TRACE_BUFFER_H_
|
||||
#define TOOL_TRACE_BUFFER_H_
|
||||
|
||||
#include <atomic>
|
||||
#include <cassert>
|
||||
#include <condition_variable>
|
||||
#include <functional>
|
||||
#include <future>
|
||||
#include <iostream>
|
||||
#include <list>
|
||||
#include <mutex>
|
||||
#include <optional>
|
||||
#include <sstream>
|
||||
#include <string>
|
||||
#include <thread>
|
||||
|
||||
namespace roctracer {
|
||||
|
||||
class TraceBufferBase {
|
||||
public:
|
||||
static void FlushAll() {
|
||||
std::lock_guard lock(mutex_);
|
||||
|
||||
for (auto* trace_buffer = head_; trace_buffer != nullptr; trace_buffer = trace_buffer->next_)
|
||||
trace_buffer->Flush();
|
||||
}
|
||||
|
||||
static void Push(TraceBufferBase* elem) {
|
||||
std::lock_guard lock(mutex_);
|
||||
|
||||
auto** prev_ptr = &head_;
|
||||
while (*prev_ptr != nullptr && elem->priority_ > (*prev_ptr)->priority_)
|
||||
prev_ptr = &(*prev_ptr)->next_;
|
||||
|
||||
elem->next_ = *prev_ptr;
|
||||
*prev_ptr = elem;
|
||||
}
|
||||
|
||||
TraceBufferBase(std::string name, int priority)
|
||||
: name_(std::move(name)), priority_(priority), next_(nullptr) {}
|
||||
|
||||
TraceBufferBase(const TraceBufferBase&) = delete;
|
||||
TraceBufferBase& operator=(const TraceBufferBase&) = delete;
|
||||
|
||||
virtual void Flush() = 0;
|
||||
|
||||
std::string name() && { return std::move(name_); }
|
||||
const std::string& name() const& { return name_; }
|
||||
|
||||
private:
|
||||
const std::string name_;
|
||||
const int priority_;
|
||||
TraceBufferBase* next_;
|
||||
|
||||
static TraceBufferBase* head_;
|
||||
static std::mutex mutex_;
|
||||
};
|
||||
|
||||
enum TraceEntryState { TRACE_ENTRY_INVALID = 0, TRACE_ENTRY_INIT = 1, TRACE_ENTRY_COMPLETE = 2 };
|
||||
|
||||
template <typename Entry, typename Allocator = std::allocator<Entry>>
|
||||
class TraceBuffer : protected TraceBufferBase {
|
||||
public:
|
||||
using callback_t = std::function<void(Entry*)>;
|
||||
|
||||
TraceBuffer(std::string name, uint64_t size, callback_t flush_callback, int priority = 0)
|
||||
: TraceBufferBase(std::move(name), priority),
|
||||
flush_callback_(std::move(flush_callback)),
|
||||
size_(size) {
|
||||
assert(size_ != 0 && "cannot create an empty trace buffer");
|
||||
|
||||
Entry* write_buffer = allocator_.allocate(size_);
|
||||
assert(write_buffer != nullptr);
|
||||
buffer_list_.push_back(write_buffer);
|
||||
|
||||
read_index_ = 0;
|
||||
write_index_ = {0, write_buffer};
|
||||
|
||||
AllocateFreeBuffer();
|
||||
|
||||
// Add this instance to the link list of all trace buffers in the process.
|
||||
TraceBufferBase::Push(this);
|
||||
}
|
||||
|
||||
~TraceBuffer() {
|
||||
// Flush the remaining records. After flushing, there should not be any records left in the
|
||||
// trace buffer.
|
||||
Flush();
|
||||
assert(read_index_ == write_index_.load().index);
|
||||
|
||||
// Acquire both the writer and worker lock as we are accessing shared variables they protect.
|
||||
std::unique_lock writer_lock(write_mutex_, std::defer_lock);
|
||||
std::unique_lock worker_lock(worker_mutex_, std::defer_lock);
|
||||
std::lock(writer_lock, worker_lock);
|
||||
|
||||
// Deallocate the buffers.
|
||||
allocator_.deallocate(write_index_.load().buffer, size_);
|
||||
allocator_.deallocate(free_buffer_, size_);
|
||||
|
||||
// Stop the worker thread. The worker thread loop checks the 'worker_thread_' std::optional
|
||||
// after waking up, and exits if it does not have a value.
|
||||
if (worker_thread_) {
|
||||
std::thread worker_thread = std::move(worker_thread_.value());
|
||||
{
|
||||
// Tell the worker thread loop to exit.
|
||||
worker_thread_.reset();
|
||||
free_buffer_ = nullptr;
|
||||
worker_cond_.notify_one();
|
||||
}
|
||||
// Release the worker lock to allow the worker thread to exit.
|
||||
worker_lock.unlock();
|
||||
worker_thread.join();
|
||||
}
|
||||
}
|
||||
|
||||
Entry* GetEntry() {
|
||||
auto current = write_index_.load(std::memory_order_relaxed);
|
||||
|
||||
while (true) {
|
||||
// If the pointer is at the end of the current buffer, switch to the available free buffer and
|
||||
// notify the worker thread to allocate a new buffer.
|
||||
if (current.index != 0 && current.index % size_ == 0) {
|
||||
std::lock_guard lock(write_mutex_);
|
||||
|
||||
// If the worker thread wasn't already started, start it now. This avoids starting a new
|
||||
// thread when the trace buffer is created.
|
||||
if (!worker_thread_) {
|
||||
std::promise<void> ready;
|
||||
auto future = ready.get_future();
|
||||
{
|
||||
std::lock_guard worker_lock(worker_mutex_);
|
||||
worker_thread_.emplace(&TraceBuffer::WorkerThreadLoop, this, std::move(ready));
|
||||
}
|
||||
future.wait();
|
||||
}
|
||||
|
||||
// Re-check the pointer overflow under the writer lock, another thread could have beaten us
|
||||
// to it and already bumped the write_index_.
|
||||
current = write_index_.load(std::memory_order_relaxed);
|
||||
if (current.index % size_ == 0) {
|
||||
std::unique_lock worker_lock(worker_mutex_);
|
||||
|
||||
// Wait for the free buffer to become available.
|
||||
worker_cond_.wait(worker_lock, [this]() { return free_buffer_ != nullptr; });
|
||||
|
||||
current.buffer = free_buffer_;
|
||||
buffer_list_.push_back(current.buffer);
|
||||
write_index_.store({current.index + 1, current.buffer}, std::memory_order_relaxed);
|
||||
|
||||
// Tell the worker thread to allocate a new free buffer.
|
||||
free_buffer_ = nullptr;
|
||||
worker_cond_.notify_one();
|
||||
|
||||
// We successfully allocated a new buffer, return the first element.
|
||||
return ¤t.buffer[0];
|
||||
}
|
||||
}
|
||||
|
||||
if (write_index_.compare_exchange_weak(current, {current.index + 1, current.buffer},
|
||||
std::memory_order_relaxed))
|
||||
return ¤t.buffer[current.index % size_];
|
||||
}
|
||||
}
|
||||
|
||||
// Flush all entries between read_pointer and write_pointer. read_pointer and write_pointer are
|
||||
// monotonically increasing indices, with read_pointer % size always indexing inside the first
|
||||
// buffer in the list. Stop flushing if an incomplete entry is found, it will be flushed with
|
||||
// the next invocation after changing its state to 'complete'.
|
||||
void Flush() override {
|
||||
std::lock_guard lock(write_mutex_);
|
||||
auto write_index = write_index_.load(std::memory_order_relaxed);
|
||||
|
||||
for (auto it = buffer_list_.begin(); it != buffer_list_.end();) {
|
||||
auto end_of_buffer = read_index_ - read_index_ % size_ + size_;
|
||||
|
||||
while (read_index_ < std::min(write_index.index, end_of_buffer)) {
|
||||
Entry* entry = &(*it)[read_index_ % size_];
|
||||
|
||||
// The entry is not yet complete, stop flushing here.
|
||||
if (entry->valid.load(std::memory_order_acquire) != TRACE_ENTRY_COMPLETE) return;
|
||||
|
||||
flush_callback_(entry);
|
||||
++read_index_;
|
||||
}
|
||||
|
||||
// The buffer is still in use or the read pointer did not reach the end of the buffer.
|
||||
if (*it == write_index.buffer || read_index_ != end_of_buffer) return;
|
||||
|
||||
// All entries in the current buffer are now processed. Destroy the buffer and move onto the
|
||||
// next buffer in the list.
|
||||
allocator_.deallocate(*it, size_);
|
||||
it = buffer_list_.erase(it);
|
||||
}
|
||||
}
|
||||
|
||||
private:
|
||||
void AllocateFreeBuffer() {
|
||||
assert(free_buffer_ == nullptr);
|
||||
|
||||
free_buffer_ = allocator_.allocate(size_);
|
||||
assert(free_buffer_ != nullptr);
|
||||
|
||||
for (size_t i = 0; i < size_; ++i)
|
||||
free_buffer_[i].valid.store(TRACE_ENTRY_INVALID, std::memory_order_relaxed);
|
||||
}
|
||||
|
||||
void WorkerThreadLoop(std::promise<void> ready) {
|
||||
std::unique_lock lock(worker_mutex_);
|
||||
|
||||
// This worker thread is now ready to accept work.
|
||||
ready.set_value();
|
||||
|
||||
while (true) {
|
||||
worker_cond_.wait(lock, [this]() { return free_buffer_ == nullptr; });
|
||||
if (!worker_thread_) break;
|
||||
AllocateFreeBuffer();
|
||||
worker_cond_.notify_one();
|
||||
}
|
||||
}
|
||||
|
||||
// The WriteIndex is used to store both the index and the buffer associated with that index (the
|
||||
// buffer contains the trace buffer records at [index - index % size, index - index % size_t +
|
||||
// size_ - 1]) in a single atomic variable.
|
||||
struct WriteIndex {
|
||||
uint64_t index;
|
||||
Entry* buffer;
|
||||
};
|
||||
|
||||
const callback_t flush_callback_;
|
||||
const uint64_t size_;
|
||||
|
||||
uint64_t read_index_; // The index of the next record to flush.
|
||||
std::atomic<WriteIndex> write_index_; // The index of the next record that could be written.
|
||||
Entry* free_buffer_{nullptr}; // The next available free buffer.
|
||||
|
||||
std::optional<std::thread> worker_thread_;
|
||||
std::mutex worker_mutex_;
|
||||
std::condition_variable worker_cond_;
|
||||
|
||||
std::mutex write_mutex_;
|
||||
std::list<Entry*> buffer_list_;
|
||||
Allocator allocator_;
|
||||
};
|
||||
} // namespace roctracer
|
||||
|
||||
#define TRACE_BUFFER_INSTANTIATE() \
|
||||
roctracer::TraceBufferBase* roctracer::TraceBufferBase::head_ = nullptr; \
|
||||
std::mutex roctracer::TraceBufferBase::mutex_;
|
||||
|
||||
#endif // TOOL_TRACE_BUFFER_H_
|
||||
ファイル差分が大きすぎるため省略します
差分を読み込み
新しいイシューから参照
ユーザーをブロックする