AI/rocm-systems
2
0
Çatalla 0
Kod Konular Değişiklik İstekleri İşlemler Paketler Projeler Sürüm Viki Aktivite

EXSWHTEC-306 - Implement tests for memory device functions memcpy/memset (#279)

Kaynağa Gözat 
Change-Id: I609efbd4f6ebaa48571e14e11ff97340dca1ab95
Bu işleme şunda yer alıyor:
Nives Vukovic
2023-12-08 07:24:30 +00:00
işlemeyi yapan: Rakesh Roy
ebeveyn 08a41abaad
işleme 9cadc3b875
11 değiştirilmiş dosya ile 613 ekleme ve 4 silme
Yama Dosyasını İndir Diff Dosyasını İndir Tüm dosyaları genişlet Tüm dosyaları daralt
catch/hipTestMain/config/config_amd_windows
+3
Dosyayı Görüntüle
@@ -332,6 +332,9 @@
"Unit_Printf_length_Sanity_Positive",
"Unit_Printf_Negative",
"Unit_Device__hip_hc_8pk_Negative",
"=== Below tests fail in external CI for PR https://github.com/ROCm-Developer-Tools/hip-tests/pull/279 ===",
"Unit_Device_memcpy_Negative",
"Unit_Device_memset_Negative",
#endif
"End of json"
]
catch/hipTestMain/config/config_nvidia_linux.json
+4 -1
Dosyayı Görüntüle
@@ -54,6 +54,9 @@
"Unit_atomicExch_system_Positive_Host_And_GPU - unsigned long long",
"Unit_atomicExch_system_Positive_Host_And_GPU - float",
"=== Below test fails in external CI for PR https://github.com/ROCm-Developer-Tools/hip-tests/pull/274 ===",
"Unit_Printf_Negative"
"Unit_Printf_Negative",
"=== Below tests fail in external CI for PR https://github.com/ROCm-Developer-Tools/hip-tests/pull/279 ===",
"Unit_Device_memcpy_Negative",
"Unit_Device_memset_Negative"
]
}
catch/unit/CMakeLists.txt
+1 -1
Dosyayı Görüntüle
@@ -37,7 +37,7 @@ add_subdirectory(errorHandling)
add_subdirectory(cooperativeGrps)
add_subdirectory(warp)
add_subdirectory(context)
add_subdirectory(warp)
add_subdirectory(device_memory)
add_subdirectory(dynamicLoading)
add_subdirectory(g++)
add_subdirectory(module)
catch/unit/compileAndCaptureOutput.py
+2 -2
Dosyayı Görüntüle
@@ -80,7 +80,7 @@ if __name__ == '__main__':
CompileAndCapture.platform = sys.argv[2]
except IndexError:
CompileAndCapture.platform = None
try:
CompileAndCapture.hip_path = sys.argv[3]
except IndexError:
@@ -95,7 +95,7 @@ if __name__ == '__main__':
CompileAndCapture.expected_error_count = int(sys.argv[5])
except IndexError:
CompileAndCapture.expected_error_count = 0
try:
CompileAndCapture.expected_warning_count = int(sys.argv[6])
except IndexError:
catch/unit/device_memory/CMakeLists.txt
+56
Dosyayı Görüntüle
@@ -0,0 +1,56 @@
# Copyright (c) 2022 Advanced Micro Devices, Inc. All Rights Reserved.
#
# 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.
# Common Tests - Test independent of all platforms
set(TEST_SRC
memcpy.cc
memset.cc
)
if(HIP_PLATFORM MATCHES "nvidia")
set(LINKER_LIBS nvrtc)
elseif(HIP_PLATFORM MATCHES "amd")
set(LINKER_LIBS hiprtc)
endif()
if(HIP_PLATFORM MATCHES "amd")
hip_add_exe_to_target(NAME DeviceMemoryTest
TEST_SRC ${TEST_SRC}
TEST_TARGET_NAME build_tests
LINKER_LIBS ${LINKER_LIBS}
PROPERTY CXX_STANDARD 17)
elseif (HIP_PLATFORM MATCHES "nvidia")
hip_add_exe_to_target(NAME DeviceMemoryTest
TEST_SRC ${TEST_SRC}
TEST_TARGET_NAME build_tests
LINKER_LIBS ${LINKER_LIBS}
COMPILE_OPTIONS -std=c++17)
endif()
add_test(NAME Unit_Device_memcpy_Negative
COMMAND python3 ${CMAKE_CURRENT_SOURCE_DIR}/../compileAndCaptureOutput.py
${CMAKE_CURRENT_SOURCE_DIR} ${HIP_PLATFORM} ${HIP_PATH}
memcpy_negative_kernels.cc 4)
add_test(NAME Unit_Device_memset_Negative
COMMAND python3 ${CMAKE_CURRENT_SOURCE_DIR}/../compileAndCaptureOutput.py
${CMAKE_CURRENT_SOURCE_DIR} ${HIP_PLATFORM} ${HIP_PATH}
memset_negative_kernels.cc 4)
catch/unit/device_memory/memcpy.cc
+249
Dosyayı Görüntüle
@@ -0,0 +1,249 @@
/*
Copyright (c) 2023 Advanced Micro Devices, Inc. All rights reserved.
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.
*/
#include "memcpy_negative_kernels_rtc.hh"
#include <hip_test_common.hh>
#include <resource_guards.hh>
#include <utils.hh>
#include <hip/hip_cooperative_groups.h>
/**
* @addtogroup memcpy memcpy
* @{
* @ingroup DeviceLanguageTest
* `memcpy(void* dst, const void* src, size_t size)` -
* copies device accessible data inside a kernel
*/
template <typename T> using kernel_sig = void (*)(T*, T*, const size_t);
template <typename T>
__global__ void memcpy_at_once_kernel(T* dst, T* src, const size_t alloc_size) {
memcpy(dst, src, alloc_size);
}
template <typename T> __global__ void memcpy_one_by_one_kernel(T* dst, T* src, const size_t N) {
const auto tid = cooperative_groups::this_grid().thread_rank();
const auto stride = cooperative_groups::this_grid().size();
for (auto i = tid; i < N; i += stride) {
memcpy(dst + tid, src + tid, sizeof(T));
}
}
template <typename T> void MemcpyDeviceToDeviceCommon(kernel_sig<T> memcpy_kernel) {
const auto allocation_size = GENERATE(kPageSize / 2, kPageSize, kPageSize * 2);
const auto element_count = allocation_size / sizeof(T);
LinearAllocGuard<T> input(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> result(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> src_allocation(LinearAllocs::hipMalloc, allocation_size);
LinearAllocGuard<T> dst_allocation(LinearAllocs::hipMalloc, allocation_size);
/* fill input data */
for (auto i = 0; i < element_count; i++) {
input.host_ptr()[i] = static_cast<T>(i);
}
/* Copy input data to device memory */
HIP_CHECK(
hipMemcpy(src_allocation.ptr(), input.host_ptr(), allocation_size, hipMemcpyHostToDevice));
/* Launch appropriate kernel*/
if (memcpy_kernel == &memcpy_at_once_kernel<T>) {
memcpy_at_once_kernel<T><<<1, 1>>>(dst_allocation.ptr(), src_allocation.ptr(), allocation_size);
} else {
constexpr auto thread_count = 1024;
const auto block_count = element_count / thread_count + 1;
memcpy_one_by_one_kernel<T>
<<<thread_count, block_count>>>(dst_allocation.ptr(), src_allocation.ptr(), element_count);
}
/* Copy filled device memory to result */
HIP_CHECK(
hipMemcpy(result.host_ptr(), dst_allocation.ptr(), allocation_size, hipMemcpyDeviceToHost));
ArrayMismatch(input.host_ptr(), result.host_ptr(), element_count);
}
template <typename T> void MemcpyPinnedToDeviceCommon(kernel_sig<T> memcpy_kernel) {
const auto allocation_size = GENERATE(kPageSize / 2, kPageSize, kPageSize * 2);
const auto element_count = allocation_size / sizeof(T);
LinearAllocGuard<T> input(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> result(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> dst_allocation(LinearAllocs::hipMalloc, allocation_size);
/* fill input data */
for (auto i = 0; i < element_count; i++) {
input.host_ptr()[i] = static_cast<T>(i);
}
/* Launch appropriate kernel*/
if (memcpy_kernel == &memcpy_at_once_kernel<T>) {
memcpy_at_once_kernel<T><<<1, 1>>>(dst_allocation.ptr(), input.host_ptr(), allocation_size);
} else {
constexpr auto thread_count = 1024;
const auto block_count = element_count / thread_count + 1;
memcpy_one_by_one_kernel<T>
<<<thread_count, block_count>>>(dst_allocation.ptr(), input.host_ptr(), element_count);
}
/* Copy filled device memory to result */
HIP_CHECK(
hipMemcpy(result.host_ptr(), dst_allocation.ptr(), allocation_size, hipMemcpyDeviceToHost));
ArrayMismatch(input.host_ptr(), result.host_ptr(), element_count);
}
template <typename T> void MemcpyDeviceToPinnedCommon(kernel_sig<T> memcpy_kernel) {
const auto allocation_size = GENERATE(kPageSize / 2, kPageSize, kPageSize * 2);
const auto element_count = allocation_size / sizeof(T);
LinearAllocGuard<T> input(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> result(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> src_allocation(LinearAllocs::hipMalloc, allocation_size);
/* fill input data */
for (auto i = 0; i < element_count; i++) {
input.host_ptr()[i] = static_cast<T>(i);
}
/* Copy input data to device memory */
HIP_CHECK(
hipMemcpy(src_allocation.ptr(), input.host_ptr(), allocation_size, hipMemcpyHostToDevice));
/* Launch appropriate kernel*/
if (memcpy_kernel == &memcpy_at_once_kernel<T>) {
memcpy_at_once_kernel<T><<<1, 1>>>(result.host_ptr(), src_allocation.ptr(), allocation_size);
} else {
constexpr auto thread_count = 1024;
const auto block_count = element_count / thread_count + 1;
memcpy_one_by_one_kernel<T>
<<<thread_count, block_count>>>(result.host_ptr(), src_allocation.ptr(), element_count);
}
HIP_CHECK(hipStreamSynchronize(nullptr));
ArrayMismatch(input.host_ptr(), result.host_ptr(), element_count);
}
template <typename T> void MemcpyPinnedToPinnedCommon(kernel_sig<T> memcpy_kernel) {
const auto allocation_size = GENERATE(kPageSize / 2, kPageSize, kPageSize * 2);
const auto element_count = allocation_size / sizeof(T);
LinearAllocGuard<T> input(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> result(LinearAllocs::hipHostMalloc, allocation_size);
/* fill input data */
for (auto i = 0; i < element_count; i++) {
input.host_ptr()[i] = static_cast<T>(i);
}
/* Launch appropriate kernel*/
if (memcpy_kernel == &memcpy_at_once_kernel<T>) {
memcpy_at_once_kernel<T><<<1, 1>>>(result.host_ptr(), input.host_ptr(), allocation_size);
} else {
constexpr auto thread_count = 1024;
const auto block_count = element_count / thread_count + 1;
memcpy_one_by_one_kernel<T>
<<<thread_count, block_count>>>(result.host_ptr(), input.host_ptr(), element_count);
}
HIP_CHECK(hipStreamSynchronize(nullptr));
ArrayMismatch(input.host_ptr(), result.host_ptr(), element_count);
}
template <typename T> void DeviceMemcpyCommon(kernel_sig<T> memcpy_kernel) {
SECTION("Device to Device memory") { MemcpyDeviceToDeviceCommon<T>(memcpy_kernel); }
SECTION("Pinned to Device memory") { MemcpyPinnedToDeviceCommon<T>(memcpy_kernel); }
SECTION("Device to Pinned memory") { MemcpyDeviceToPinnedCommon<T>(memcpy_kernel); }
SECTION("Pinned to Pinned memory") { MemcpyPinnedToPinnedCommon<T>(memcpy_kernel); }
}
/**
* Test Description
* ------------------------
* - Verifies basic test cases for copying device/pinned memory inside a kernel using various data
* types and memory sizes:
* -# Copies whole memory buffer in one thread
* -# Copies memory buffer elements one by one in multiple threads/blocks
* Test source
* ------------------------
* - unit/device_memory/memcpy.cc
* Test requirements
* ------------------------
* - HIP_VERSION >= 5.2
*/
TEMPLATE_TEST_CASE("Unit_Device_memcpy_Positive", "", char, int, unsigned int, long, unsigned long,
long long, unsigned long long, float, double) {
SECTION("Memcpy whole buffer in one thread") {
DeviceMemcpyCommon<TestType>(memcpy_at_once_kernel);
}
SECTION("Memcpy buffer in multiple threads/blocks") {
DeviceMemcpyCommon<TestType>(memcpy_one_by_one_kernel);
}
}
/**
* Test Description
* ------------------------
* - RTCs kernels that pass combinations of arguments of invalid types for memcpy
* Test source
* ------------------------
* - unit/device_memory/memcpy.cc
* Test requirements
* ------------------------
* - HIP_VERSION >= 5.2
*/
TEST_CASE("Unit_Device_memcpy_Negative_Parameters_RTC") {
hiprtcProgram program{};
const auto program_source = kMemcpyParam;
HIPRTC_CHECK(
hiprtcCreateProgram(&program, program_source, "memcpy_negative.cc", 0, nullptr, nullptr));
hiprtcResult result{hiprtcCompileProgram(program, 0, nullptr)};
// Get the compile log and count compiler error messages
size_t log_size{};
HIPRTC_CHECK(hiprtcGetProgramLogSize(program, &log_size));
std::string log(log_size, ' ');
HIPRTC_CHECK(hiprtcGetProgramLog(program, log.data()));
int error_count{0};
int expected_error_count{4};
std::string error_message{"error:"};
size_t n_pos = log.find(error_message, 0);
while (n_pos != std::string::npos) {
++error_count;
n_pos = log.find(error_message, n_pos + 1);
}
HIPRTC_CHECK(hiprtcDestroyProgram(&program));
HIPRTC_CHECK_ERROR(result, HIPRTC_ERROR_COMPILATION);
REQUIRE(error_count == expected_error_count);
}
catch/unit/device_memory/memcpy_negative_kernels.cc
+31
Dosyayı Görüntüle
@@ -0,0 +1,31 @@
/*
Copyright (c) 2023 Advanced Micro Devices, Inc. All rights reserved.
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.
*/
#include <hip_test_common.hh>
struct Dummy {
__device__ Dummy() {}
__device__ ~Dummy() {}
};
/*void* memcpy(void* dst, const void* src, size_t size)*/
__global__ void memcpy_n1(int* dst, const int src, size_t size) { memcpy(dst, src, size); }
__global__ void memcpy_n2(int dst, const int* src, size_t size) { memcpy(dst, src, size); }
__global__ void memcpy_n3(int* dst, const int* src, size_t* size) { memcpy(dst, src, size); }
__global__ void memcpy_n4(int* dst, const int* src, Dummy size) { memcpy(dst, src, size); }
catch/unit/device_memory/memcpy_negative_kernels_rtc.hh
+32
Dosyayı Görüntüle
@@ -0,0 +1,32 @@
/*
Copyright (c) 2023 Advanced Micro Devices, Inc. All rights reserved.
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.
*/
#pragma once
static constexpr auto kMemcpyParam{
R"(
struct Dummy {
__device__ Dummy() {}
__device__ ~Dummy() {}
};
__global__ void memcpy_n1(int* dst, const int src, size_t size) { memcpy(dst, src, size); }
__global__ void memcpy_n2(int dst, const int* src, size_t size) { memcpy(dst, src, size); }
__global__ void memcpy_n3(int* dst, const int* src, size_t* size) { memcpy(dst, src, size); }
__global__ void memcpy_n8(int* dst, const int* src, Dummy size) { memcpy(dst, src, size); }
)"};
catch/unit/device_memory/memset.cc
+172
Dosyayı Görüntüle
@@ -0,0 +1,172 @@
/*
Copyright (c) 2023 Advanced Micro Devices, Inc. All rights reserved.
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.
*/
#include "memset_negative_kernels_rtc.hh"
#include <hip_test_common.hh>
#include <resource_guards.hh>
#include <utils.hh>
#include <hip/hip_cooperative_groups.h>
/**
* @addtogroup memset memset
* @{
* @ingroup DeviceLanguageTest
* `memset(void* ptr, int val, size_t size)` -
* sets device accessible data inside a kernel
*/
template <typename T> using kernel_sig = void (*)(T*, int, const size_t);
template <typename T>
__global__ void memset_at_once_kernel(T* dst, int value, const size_t alloc_size) {
memset(dst, value, alloc_size);
}
template <typename T> __global__ void memset_one_by_one_kernel(T* dst, int value, const size_t N) {
const auto tid = cooperative_groups::this_grid().thread_rank();
const auto stride = cooperative_groups::this_grid().size();
for (auto i = tid; i < N; i += stride) {
memset(dst + tid, value, sizeof(T));
}
}
template <typename T> void MemsetDeviceCommon(kernel_sig<T> memset_kernel) {
const auto allocation_size = GENERATE(kPageSize / 2, kPageSize, kPageSize * 2);
const auto element_count = allocation_size / sizeof(T);
LinearAllocGuard<T> reference(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> result(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> dst_allocation(LinearAllocs::hipMalloc, allocation_size);
constexpr auto thread_count = 1024;
const auto block_count = element_count / thread_count + 1;
constexpr auto expected_value = 42;
memset(reference.host_ptr(), expected_value, allocation_size);
if (memset_kernel == &memset_at_once_kernel<T>) {
memset_at_once_kernel<T><<<1, 1>>>(dst_allocation.ptr(), expected_value, allocation_size);
} else {
memset_one_by_one_kernel<T>
<<<thread_count, block_count>>>(dst_allocation.ptr(), expected_value, element_count);
}
HIP_CHECK(
hipMemcpy(result.host_ptr(), dst_allocation.ptr(), allocation_size, hipMemcpyDeviceToHost));
ArrayMismatch(reference.host_ptr(), result.host_ptr(), element_count);
}
template <typename T> void MemsetPinnedCommon(kernel_sig<T> memset_kernel) {
const auto allocation_size = GENERATE(kPageSize / 2, kPageSize, kPageSize * 2);
const auto element_count = allocation_size / sizeof(T);
LinearAllocGuard<T> reference(LinearAllocs::hipHostMalloc, allocation_size);
LinearAllocGuard<T> result(LinearAllocs::hipHostMalloc, allocation_size);
constexpr auto thread_count = 1024;
const auto block_count = element_count / thread_count + 1;
constexpr auto expected_value = 42;
memset(reference.host_ptr(), expected_value, allocation_size);
if (memset_kernel == &memset_at_once_kernel<T>) {
memset_at_once_kernel<T><<<1, 1>>>(result.host_ptr(), expected_value, allocation_size);
} else {
memset_one_by_one_kernel<T>
<<<thread_count, block_count>>>(result.host_ptr(), expected_value, element_count);
}
HIP_CHECK(hipStreamSynchronize(nullptr));
ArrayMismatch(reference.host_ptr(), result.host_ptr(), element_count);
}
template <typename T> void DeviceMemsetCommon(kernel_sig<T> memset_kernel) {
SECTION("Set Device memory") { MemsetDeviceCommon<T>(memset_kernel); }
SECTION("Set Pinned memory") { MemsetPinnedCommon<T>(memset_kernel); }
}
/**
* Test Description
* ------------------------
* - Verifies basic test cases for setting device/pinned memory inside a kernel using various data
* types and memory sizes:
* -# Set whole memory buffer in one thread
* -# Set memory buffer elements one by one in multiple threads/blocks
* Test source
* ------------------------
* - unit/device_memory/memset.cc
* Test requirements
* ------------------------
* - HIP_VERSION >= 5.2
*/
TEMPLATE_TEST_CASE("Unit_Device_memset_Positive", "", char, int, unsigned int, long, unsigned long,
long long, unsigned long long, float, double) {
SECTION("Memset whole buffer in one thread") {
DeviceMemsetCommon<TestType>(memset_at_once_kernel);
}
SECTION("Memset buffer in multiple threads/blocks") {
DeviceMemsetCommon<TestType>(memset_one_by_one_kernel);
}
}
/**
* Test Description
* ------------------------
* - RTCs kernels that pass combinations of arguments of invalid types for memset
* Test source
* ------------------------
* - unit/device_memory/memset.cc
* Test requirements
* ------------------------
* - HIP_VERSION >= 5.2
*/
TEST_CASE("Unit_Device_memset_Negative_Parameters_RTC") {
hiprtcProgram program{};
const auto program_source = kMemsetParam;
HIPRTC_CHECK(
hiprtcCreateProgram(&program, program_source, "memset_negative.cc", 0, nullptr, nullptr));
hiprtcResult result{hiprtcCompileProgram(program, 0, nullptr)};
// Get the compile log and count compiler error messages
size_t log_size{};
HIPRTC_CHECK(hiprtcGetProgramLogSize(program, &log_size));
std::string log(log_size, ' ');
HIPRTC_CHECK(hiprtcGetProgramLog(program, log.data()));
int error_count{0};
int expected_error_count{4};
std::string error_message{"error:"};
size_t n_pos = log.find(error_message, 0);
while (n_pos != std::string::npos) {
++error_count;
n_pos = log.find(error_message, n_pos + 1);
}
HIPRTC_CHECK(hiprtcDestroyProgram(&program));
HIPRTC_CHECK_ERROR(result, HIPRTC_ERROR_COMPILATION);
REQUIRE(error_count == expected_error_count);
}
catch/unit/device_memory/memset_negative_kernels.cc
+31
Dosyayı Görüntüle
@@ -0,0 +1,31 @@
/*
Copyright (c) 2023 Advanced Micro Devices, Inc. All rights reserved.
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.
*/
#include <hip_test_common.hh>
struct Dummy {
__device__ Dummy() {}
__device__ ~Dummy() {}
};
/*void* memset(void* ptr, int val, size_t size)*/
__global__ void memset_n1(int* ptr, int* val, size_t size) { memset(ptr, val, size); }
__global__ void memset_n2(int ptr, int val, size_t size) { memset(ptr, val, size); }
__global__ void memset_n3(int* ptr, int val, size_t* size) { memset(ptr, val, size); }
__global__ void memset_n4(int* ptr, int val, Dummy size) { memset(ptr, val, size); }
catch/unit/device_memory/memset_negative_kernels_rtc.hh
+32
Dosyayı Görüntüle
@@ -0,0 +1,32 @@
/*
Copyright (c) 2023 Advanced Micro Devices, Inc. All rights reserved.
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.
*/
#pragma once
static constexpr auto kMemsetParam{
R"(
struct Dummy {
__device__ Dummy() {}
__device__ ~Dummy() {}
};
__global__ void memset_n1(int* ptr, int* val, size_t size) { memset(ptr, val, size); }
__global__ void memset_n2(int ptr, int val, size_t size) { memset(ptr, val, size); }
__global__ void memset_n3(int* ptr, int val, size_t* size) { memset(ptr, val, size); }
__global__ void memset_n4(int* ptr, int val, Dummy size) { memset(ptr, val, size); }
)"};