AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[SWDEV-562726] Fix clang + ASAN errors

Browse Source 
* Updates:
  - [ASAN] GCC does not support `-shared-libsan flags`, so removed this one
  - [Clang] Fixed refernces to local binding errors (name collision)
    & other strict scope/structure/lamda binding errors
  - [Clang] Fix rsmi_wrapper error: \"error: missing default argument on parameter \'args\'\"
  - [ASAN] Fixed stack-buffer-overflow found in
    `amdsmi_get_gpu_accelerator_partition_profile()`

Change-Id: I854007efb75d828dbb8088c0d56dbc125081f0f2
Signed-off-by: Charis Poag <Charis.Poag@amd.com>
This commit is contained in:
Charis Poag
2025-10-27 13:07:49 -05:00
committed by Arif, Maisam
parent ced7d12395
commit 00a04f5810
4 changed files with 48 additions and 58 deletions
Download Patch File Download Diff File Expand all files Collapse all files
cmake_modules/help_package.cmake
+5 -3
View File
@@ -40,12 +40,14 @@ function(generic_package)
set(ASAN_LINKER_FLAGS "-fsanitize=address")
if(BUILD_SHARED_LIBS)
set(ASAN_COMPILER_FLAGS "${ASAN_COMPILER_FLAGS} -shared-libsan")
set(ASAN_LINKER_FLAGS "${ASAN_LINKER_FLAGS} -shared-libsan")
# Clang-specific flag for shared ASAN library
if(CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
set(ASAN_COMPILER_FLAGS "${ASAN_COMPILER_FLAGS} -shared-libsan")
set(ASAN_LINKER_FLAGS "${ASAN_LINKER_FLAGS} -shared-libsan")
endif()
else()
set(ASAN_LINKER_FLAGS "${ASAN_LINKER_FLAGS} -static-libsan")
endif()
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${ASAN_COMPILER_FLAGS}" PARENT_SCOPE)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${ASAN_COMPILER_FLAGS}" PARENT_SCOPE)
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${ASAN_LINKER_FLAGS}" PARENT_SCOPE)
rocm_smi/src/rocm_smi_gpu_metrics.cc
+36 -49
View File
@@ -478,66 +478,53 @@ template<class T> struct is_vector : std::false_type {};
template<class U, class A> struct is_vector<std::vector<U,A>> : std::true_type {};
template<typename T>
AMDGpuDynamicMetricTblValues_t format_metric_row(const T& metric, const std::string& value_title)
{
AMDGpuDynamicMetricTblValues_t format_metric_row(const T& metric, const std::string& value_title) {
auto multi_values = AMDGpuDynamicMetricTblValues_t{};
auto get_data_type_info = [&]() {
auto data_type(AMDGpuMetricsDataType_t::kUInt64);
if constexpr (std::is_array_v<T>) {
const uint8_t kCheckUint8[]={1};
const uint16_t kCheckUint16[]={2};
const uint32_t kCheckUint32[]={3};
const uint64_t kCheckUint64[]={4};
if constexpr (std::is_same_v<decltype(metric[0]), decltype(kCheckUint8[0])>) {
data_type = AMDGpuMetricsDataType_t::kUInt8;
}
if constexpr (std::is_same_v<decltype(metric[0]), decltype(kCheckUint16[0])>) {
data_type = AMDGpuMetricsDataType_t::kUInt16;
}
if constexpr (std::is_same_v<decltype(metric[0]), decltype(kCheckUint32[0])>) {
data_type = AMDGpuMetricsDataType_t::kUInt32;
}
if constexpr (std::is_same_v<decltype(metric[0]), decltype(kCheckUint64[0])>) {
data_type = AMDGpuMetricsDataType_t::kUInt64;
}
return std::make_tuple(data_type, static_cast<uint16_t>(std::end(metric) - std::begin(metric)));
}
// Determine data type and count inline
AMDGpuMetricsDataType_t inferred_data_type = AMDGpuMetricsDataType_t::kUInt64;
uint16_t num_values = 1;
const uint16_t kSingleValue(1);
if constexpr (std::is_same_v<T, std::uint8_t>) {
data_type = AMDGpuMetricsDataType_t::kUInt8;
if constexpr (std::is_array_v<T>) {
using ElementType = std::remove_extent_t<T>;
if constexpr (std::is_same_v<ElementType, uint8_t>) {
inferred_data_type = AMDGpuMetricsDataType_t::kUInt8;
} else if constexpr (std::is_same_v<ElementType, uint16_t>) {
inferred_data_type = AMDGpuMetricsDataType_t::kUInt16;
} else if constexpr (std::is_same_v<ElementType, uint32_t>) {
inferred_data_type = AMDGpuMetricsDataType_t::kUInt32;
} else if constexpr (std::is_same_v<ElementType, uint64_t>) {
inferred_data_type = AMDGpuMetricsDataType_t::kUInt64;
}
if constexpr (std::is_same_v<T, std::uint16_t>) {
data_type = AMDGpuMetricsDataType_t::kUInt16;
num_values = static_cast<uint16_t>(std::extent_v<T>);
} else {
// Scalar case
if constexpr (std::is_same_v<T, uint8_t>) {
inferred_data_type = AMDGpuMetricsDataType_t::kUInt8;
} else if constexpr (std::is_same_v<T, uint16_t>) {
inferred_data_type = AMDGpuMetricsDataType_t::kUInt16;
} else if constexpr (std::is_same_v<T, uint32_t>) {
inferred_data_type = AMDGpuMetricsDataType_t::kUInt32;
} else if constexpr (std::is_same_v<T, uint64_t>) {
inferred_data_type = AMDGpuMetricsDataType_t::kUInt64;
}
if constexpr (std::is_same_v<T, std::uint32_t>) {
data_type = AMDGpuMetricsDataType_t::kUInt32;
}
if constexpr (std::is_same_v<T, std::uint64_t>) {
data_type = AMDGpuMetricsDataType_t::kUInt64;
}
return std::make_tuple(data_type, kSingleValue);
};
}
const auto [data_type, num_values] = get_data_type_info();
for (auto idx = uint16_t(0); idx < num_values; ++idx) {
auto value = uint64_t(0);
// Populate the values
for (uint16_t idx = 0; idx < num_values; ++idx) {
uint64_t value = 0;
if constexpr (std::is_array_v<T>) {
value = (metric[idx]);
value = static_cast<uint64_t>(metric[idx]);
} else {
value = (metric);
value = static_cast<uint64_t>(metric);
}
auto amdgpu_dynamic_metric_value = [&]() {
AMDGpuDynamicMetricsValue_t amdgpu_dynamic_metric_value_init{};
amdgpu_dynamic_metric_value_init.m_value = value;
amdgpu_dynamic_metric_value_init.m_info = (value_title + " : " + std::to_string(idx));
amdgpu_dynamic_metric_value_init.m_original_type = data_type;
return amdgpu_dynamic_metric_value_init;
}();
AMDGpuDynamicMetricsValue_t amdgpu_dynamic_metric_value_init{};
amdgpu_dynamic_metric_value_init.m_value = value;
amdgpu_dynamic_metric_value_init.m_info = (value_title + " : " + std::to_string(idx));
amdgpu_dynamic_metric_value_init.m_original_type = inferred_data_type;
multi_values.emplace_back(amdgpu_dynamic_metric_value);
multi_values.emplace_back(amdgpu_dynamic_metric_value_init);
}
return multi_values;
src/amd_smi/amd_smi.cc
+3 -3
View File
@@ -145,7 +145,7 @@ static amdsmi_status_t get_gpu_device_from_handle(amdsmi_processor_handle proces
template <typename F, typename ...Args>
amdsmi_status_t rsmi_wrapper(F && f,
amdsmi_processor_handle processor_handle, uint32_t increment_gpu_id = 0, Args &&... args) {
amdsmi_processor_handle processor_handle, uint32_t increment_gpu_id, Args &&... args) {
AMDSMI_CHECK_INIT();
@@ -3057,8 +3057,8 @@ amdsmi_get_gpu_accelerator_partition_profile(amdsmi_processor_handle processor_h
std::end(tokens),
amd::smi::make_ostream_joiner(&ss_1, ", "));
constexpr uint32_t kCurrentPartitionSize = 5;
char current_partition[kCurrentPartitionSize];
constexpr uint32_t kCurrentPartitionSize = 16;
char current_partition[kCurrentPartitionSize] = {0};
std::string current_partition_str = "N/A";
amdsmi_status_t compute_status = amdsmi_get_gpu_compute_partition(processor_handle,
current_partition, kCurrentPartitionSize);
tests/amd_smi_test/functional/dynamic_metrics_test.cc
+4 -3
View File
@@ -76,7 +76,8 @@ auto GetExpectedMetricVersionFlag(uint16_t major, uint16_t minor, bool is_partit
}
// pass a header we want to test against
auto BuildFakeMetricsBlob(amd::smi::AMDGpuMetricsHeader_v1_t new_header) -> std::vector<uint8_t> {
auto BuildFakeMetricsBlob(amd::smi::AMDGpuMetricsHeader_v1_t new_header)
-> std::vector<uint8_t> {
if (new_header.m_structure_size < sizeof(new_header)) {
throw std::runtime_error("Header size too small");
}
@@ -119,7 +120,7 @@ TEST(AmdSmiDynamicMetricTest, GPUMetricDynamicVersionSupported) {
const auto blob = BuildFakeMetricsBlob(amd::smi::AMDGpuMetricsHeader_v1_t{
.m_structure_size = sizeof(amd::smi::AMDGpuMetricsHeader_v1_t),
.m_format_revision = 1,
.m_content_revision = static_cast<uint16_t>(ver), // Known minor versions
.m_content_revision = static_cast<uint8_t>(ver), // Known minor versions
});
const auto fake_path =
WriteBlobToTempFile(blob, "amdsmi_fake_gpu_metrics_v1" + std::to_string(ver) + ".bin");
@@ -170,7 +171,7 @@ TEST(AmdSmiDynamicMetricTest, XCPMetricDynamicVersionSupported) {
const auto blob = BuildFakeMetricsBlob(amd::smi::AMDGpuMetricsHeader_v1_t{
.m_structure_size = sizeof(amd::smi::AMDGpuMetricsHeader_v1_t),
.m_format_revision = 1,
.m_content_revision = static_cast<uint16_t>(ver), // Known minor versions
.m_content_revision = static_cast<uint8_t>(ver), // Known minor versions
});
const auto fake_path =
WriteBlobToTempFile(blob, "amdsmi_fake_xcp_metrics_v1" + std::to_string(ver) + ".bin");