Skip to content

P
Paddle

Crayon鑫 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
提交 772e0956 编写于 作者: Z zhouwei25 提交者： liuwei1031
浏览文件
操作

Optimize the content of error reporting information, print error code and... 

Optimize the content of error reporting information, print error code and official document web sites (#18671)

 optimize the error reporting information of cuda related API
 index on develop: 130ac177 Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into develop
上级 ae58afc5
隐藏空白更改
内联 并排
Showing with 83 addition and 41 deletion
paddle/fluid/platform/gpu_info.cc
浏览文件 @ 772e0956
...@@ -79,6 +79,12 @@ DEFINE_string(selected_gpus, "", ...@@ -79,6 +79,12 @@ DEFINE_string(selected_gpus, "",
namespace paddle { namespace paddle {
namespace platform { namespace platform {
inline std::string CudaErrorWebsite() {
return "Please see detail in https://docs.nvidia.com/cuda/cuda-runtime-api"
"/group__CUDART__TYPES.html#group__CUDART__TYPES_1g3f51e3575c217824"
"6db0a94a430e0038";
}
static int GetCUDADeviceCountImpl() { static int GetCUDADeviceCountImpl() {
const auto *cuda_visible_devices = std::getenv("CUDA_VISIBLE_DEVICES"); const auto *cuda_visible_devices = std::getenv("CUDA_VISIBLE_DEVICES");
if (cuda_visible_devices != nullptr) { if (cuda_visible_devices != nullptr) {
...@@ -92,9 +98,12 @@ static int GetCUDADeviceCountImpl() { ...@@ -92,9 +98,12 @@ static int GetCUDADeviceCountImpl() {
} }
int count; int count;
auto error_code = cudaGetDeviceCount(&count);
PADDLE_ENFORCE( PADDLE_ENFORCE(
cudaGetDeviceCount(&count), error_code,
"cudaGetDeviceCount failed in paddle::platform::GetCUDADeviceCount"); "cudaGetDeviceCount failed in "
"paddle::platform::GetCUDADeviceCountImpl, error code : %d, %s",
error_code, CudaErrorWebsite());
return count; return count;
} }
...@@ -107,28 +116,33 @@ int GetCUDAComputeCapability(int id) { ...@@ -107,28 +116,33 @@ int GetCUDAComputeCapability(int id) {
PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count"); PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count");
cudaDeviceProp device_prop; cudaDeviceProp device_prop;
auto error_code = cudaGetDeviceProperties(&device_prop, id); auto error_code = cudaGetDeviceProperties(&device_prop, id);
PADDLE_ENFORCE(error_code, PADDLE_ENFORCE(
"cudaGetDeviceProperties failed in " error_code,
"paddle::platform::GetCUDAComputeCapability, error code : %d", "cudaGetDeviceProperties failed in "
error_code); "paddle::platform::GetCUDAComputeCapability, error code : %d, %s",
error_code, CudaErrorWebsite());
return device_prop.major * 10 + device_prop.minor; return device_prop.major * 10 + device_prop.minor;
} }
int GetCUDARuntimeVersion(int id) { int GetCUDARuntimeVersion(int id) {
PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count"); PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count");
int runtime_version = 0; int runtime_version = 0;
PADDLE_ENFORCE(cudaRuntimeGetVersion(&runtime_version), auto error_code = cudaRuntimeGetVersion(&runtime_version);
PADDLE_ENFORCE(error_code,
"cudaRuntimeGetVersion failed in " "cudaRuntimeGetVersion failed in "
"paddle::platform::cudaRuntimeGetVersion"); "paddle::platform::GetCUDARuntimeVersion, error code : %d, %s",
error_code, CudaErrorWebsite());
return runtime_version; return runtime_version;
} }
int GetCUDADriverVersion(int id) { int GetCUDADriverVersion(int id) {
PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count"); PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count");
int driver_version = 0; int driver_version = 0;
PADDLE_ENFORCE(cudaDriverGetVersion(&driver_version), auto error_code = cudaDriverGetVersion(&driver_version);
PADDLE_ENFORCE(error_code,
"cudaDriverGetVersion failed in " "cudaDriverGetVersion failed in "
"paddle::platform::GetCUDADriverVersion"); "paddle::platform::GetCUDADriverVersion, error code : %d, %s",
error_code, CudaErrorWebsite());
return driver_version; return driver_version;
} }
...@@ -145,28 +159,35 @@ bool TensorCoreAvailable() { ...@@ -145,28 +159,35 @@ bool TensorCoreAvailable() {
int GetCUDAMultiProcessors(int id) { int GetCUDAMultiProcessors(int id) {
PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count"); PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count");
int count; int count;
PADDLE_ENFORCE( auto error_code =
cudaDeviceGetAttribute(&count, cudaDevAttrMultiProcessorCount, id), cudaDeviceGetAttribute(&count, cudaDevAttrMultiProcessorCount, id);
"cudaDeviceGetAttribute failed in " PADDLE_ENFORCE(error_code,
"paddle::platform::GetCUDAMultiProcessors"); "cudaDeviceGetAttribute failed in "
"paddle::platform::GetCUDAMultiProcess, error code : %d, %s",
error_code, CudaErrorWebsite());
return count; return count;
} }
int GetCUDAMaxThreadsPerMultiProcessor(int id) { int GetCUDAMaxThreadsPerMultiProcessor(int id) {
PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count"); PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count");
int count; int count;
PADDLE_ENFORCE(cudaDeviceGetAttribute( auto error_code = cudaDeviceGetAttribute(
&count, cudaDevAttrMaxThreadsPerMultiProcessor, id), &count, cudaDevAttrMaxThreadsPerMultiProcessor, id);
"cudaDeviceGetAttribute failed in " PADDLE_ENFORCE(
"paddle::platform::GetCUDAMaxThreadsPerMultiProcessor"); error_code,
"cudaDeviceGetAttribute failed in paddle::"
"platform::GetCUDAMaxThreadsPerMultiProcessor, error code : %d, %s",
error_code, CudaErrorWebsite());
return count; return count;
} }
int GetCurrentDeviceId() { int GetCurrentDeviceId() {
int device_id; int device_id;
PADDLE_ENFORCE( auto error_code = cudaGetDevice(&device_id);
cudaGetDevice(&device_id), PADDLE_ENFORCE(error_code,
"cudaGetDevice failed in paddle::platform::GetCurrentDeviceId"); "cudaGetDevice failed in "
"paddle::platform::GetCurrentDeviceId, error code : %d, %s",
error_code, CudaErrorWebsite());
return device_id; return device_id;
} }
...@@ -191,13 +212,19 @@ std::vector<int> GetSelectedDevices() { ...@@ -191,13 +212,19 @@ std::vector<int> GetSelectedDevices() {
void SetDeviceId(int id) { void SetDeviceId(int id) {
// TODO(qijun): find a better way to cache the cuda device count // TODO(qijun): find a better way to cache the cuda device count
PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count"); PADDLE_ENFORCE_LT(id, GetCUDADeviceCount(), "id must less than GPU count");
PADDLE_ENFORCE(cudaSetDevice(id), auto error_code = cudaSetDevice(id);
"cudaSetDevice failed in paddle::platform::SetDeviceId"); PADDLE_ENFORCE(error_code,
"cudaSetDevice failed in "
"paddle::platform::SetDeviced, error code : %d, %s",
error_code, CudaErrorWebsite());
} }
void GpuMemoryUsage(size_t *available, size_t *total) { void GpuMemoryUsage(size_t *available, size_t *total) {
PADDLE_ENFORCE(cudaMemGetInfo(available, total), auto error_code = cudaMemGetInfo(available, total);
"cudaMemGetInfo failed in paddle::platform::GetMemoryUsage"); PADDLE_ENFORCE(error_code,
"cudaMemGetInfo failed in "
"paddle::platform::GetMemoryUsage, error code : %d, %s",
error_code, CudaErrorWebsite());
} }
size_t GpuMaxAllocSize() { size_t GpuMaxAllocSize() {
...@@ -224,11 +251,13 @@ size_t GpuInitAllocSize() { ...@@ -224,11 +251,13 @@ size_t GpuInitAllocSize() {
size_t GpuReallocSize() { size_t GpuReallocSize() {
if (FLAGS_reallocate_gpu_memory_in_mb > 0ul) { if (FLAGS_reallocate_gpu_memory_in_mb > 0ul) {
// Additional memory will be allocated by FLAGS_reallocate_gpu_memory_in_mb // Additional memory will be allocated by
// FLAGS_reallocate_gpu_memory_in_mb
return static_cast<size_t>(FLAGS_reallocate_gpu_memory_in_mb << 20); return static_cast<size_t>(FLAGS_reallocate_gpu_memory_in_mb << 20);
} }
// FLAGS_reallocate_gpu_memory_in_mb is 0, additional memory will be allocated // FLAGS_reallocate_gpu_memory_in_mb is 0, additional memory will be
// allocated
// by fraction // by fraction
size_t total = 0; size_t total = 0;
size_t available = 0; size_t available = 0;
...@@ -268,37 +297,50 @@ size_t GpuMaxChunkSize() { ...@@ -268,37 +297,50 @@ size_t GpuMaxChunkSize() {
void GpuMemcpyAsync(void *dst, const void *src, size_t count, void GpuMemcpyAsync(void *dst, const void *src, size_t count,
enum cudaMemcpyKind kind, cudaStream_t stream) { enum cudaMemcpyKind kind, cudaStream_t stream) {
PADDLE_ENFORCE(cudaMemcpyAsync(dst, src, count, kind, stream), auto error_code = cudaMemcpyAsync(dst, src, count, kind, stream);
PADDLE_ENFORCE(error_code,
"cudaMemcpyAsync failed in paddle::platform::GpuMemcpyAsync " "cudaMemcpyAsync failed in paddle::platform::GpuMemcpyAsync "
"(%p -> %p, length: %d)", "(%p -> %p, length: %d) error code : %d, %s",
src, dst, static_cast<int>(count)); src, dst, static_cast<int>(count), error_code,
CudaErrorWebsite());
} }
void GpuMemcpySync(void *dst, const void *src, size_t count, void GpuMemcpySync(void *dst, const void *src, size_t count,
enum cudaMemcpyKind kind) { enum cudaMemcpyKind kind) {
PADDLE_ENFORCE(cudaMemcpy(dst, src, count, kind), auto error_code = cudaMemcpy(dst, src, count, kind);
"cudaMemcpy failed in paddle::platform::GpuMemcpySync (%p -> " PADDLE_ENFORCE(error_code,
"%p, length: %d)", "cudaMemcpy failed in paddle::platform::GpuMemcpySync "
src, dst, static_cast<int>(count)); "(%p -> %p, length: %d) error code : %d, %s",
src, dst, static_cast<int>(count), error_code,
CudaErrorWebsite());
} }
void GpuMemcpyPeerAsync(void *dst, int dst_device, const void *src, void GpuMemcpyPeerAsync(void *dst, int dst_device, const void *src,
int src_device, size_t count, cudaStream_t stream) { int src_device, size_t count, cudaStream_t stream) {
auto error_code =
cudaMemcpyPeerAsync(dst, dst_device, src, src_device, count, stream);
PADDLE_ENFORCE( PADDLE_ENFORCE(
cudaMemcpyPeerAsync(dst, dst_device, src, src_device, count, stream), error_code,
"cudaMemcpyPeerAsync failed in paddle::platform::GpuMemcpyPeerAsync"); "cudaMemcpyPeerAsync failed in paddle::platform::GpuMemcpyPeerAsync "
"error code : %d, %s",
error_code, CudaErrorWebsite());
} }
void GpuMemcpyPeerSync(void *dst, int dst_device, const void *src, void GpuMemcpyPeerSync(void *dst, int dst_device, const void *src,
int src_device, size_t count) { int src_device, size_t count) {
PADDLE_ENFORCE( auto error_code = cudaMemcpyPeer(dst, dst_device, src, src_device, count);
cudaMemcpyPeer(dst, dst_device, src, src_device, count), PADDLE_ENFORCE(error_code,
"cudaMemcpyPeer failed in paddle::platform::GpuMemcpyPeerSync"); "cudaMemcpyPeer failed in paddle::platform::GpuMemcpyPeerSync "
"error code : %d, %s",
error_code, CudaErrorWebsite());
} }
void GpuMemsetAsync(void *dst, int value, size_t count, cudaStream_t stream) { void GpuMemsetAsync(void *dst, int value, size_t count, cudaStream_t stream) {
PADDLE_ENFORCE(cudaMemsetAsync(dst, value, count, stream), auto error_code = cudaMemsetAsync(dst, value, count, stream);
"cudaMemsetAsync failed in paddle::platform::GpuMemsetAsync"); PADDLE_ENFORCE(error_code,
"cudaMemsetAsync failed in paddle::platform::GpuMemsetAsync "
"error code : %d, %s",
error_code, CudaErrorWebsite());
} }
} // namespace platform } // namespace platform
} // namespace paddle } // namespace paddle
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或