Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Various bug fixes (#2825)

Browse Source 
* Fatal error if GPU algorithm selected without GPU support compiled

* Resolve type conversion warnings

* Fix gpu unit test failure

* Fix compressed iterator edge case

* Fix python unit test failures due to flake8 update on pip
This commit is contained in:
Rory Mitchell
2017-10-25 14:45:01 +13:00
committed by GitHub
parent c71b62d48d
commit 13e7a2cff0
21 changed files with 163 additions and 180 deletions
Download Patch File Download Diff File Expand all files Collapse all files

75
src/common/device_helpers.cuh
View File

@@ -125,7 +125,7 @@ inline size_t available_memory(int device_idx) {
* \param device_idx Zero-based index of the device.
*/
inline int max_shared_memory(int device_idx) {
inline size_t max_shared_memory(int device_idx) {
cudaDeviceProp prop;
dh::safe_cuda(cudaGetDeviceProperties(&prop, device_idx));
return prop.sharedMemPerBlock;
@@ -241,8 +241,7 @@ inline void launch_n(int device_idx, size_t n, L lambda) {
}
safe_cuda(cudaSetDevice(device_idx));
// TODO: Template on n so GRID_SIZE always fits into int.
const int GRID_SIZE = div_round_up(n, ITEMS_PER_THREAD * BLOCK_THREADS);
const int GRID_SIZE = static_cast<int>(div_round_up(n, ITEMS_PER_THREAD * BLOCK_THREADS));
launch_n_kernel<<<GRID_SIZE, BLOCK_THREADS>>>(static_cast<size_t>(0), n,
lambda);
}
@@ -428,74 +427,66 @@ class bulk_allocator {
const int align = 256;
template <typename SizeT>
size_t align_round_up(SizeT n) {
size_t align_round_up(size_t n) const {
n = (n + align - 1) / align;
return n * align;
}
template <typename T, typename SizeT>
size_t get_size_bytes(dvec<T> *first_vec, SizeT first_size) {
return align_round_up<SizeT>(first_size * sizeof(T));
template <typename T>
size_t get_size_bytes(dvec<T> *first_vec, size_t first_size) {
return align_round_up(first_size * sizeof(T));
}
template <typename T, typename SizeT, typename... Args>
size_t get_size_bytes(dvec<T> *first_vec, SizeT first_size, Args... args) {
return get_size_bytes<T, SizeT>(first_vec, first_size) +
get_size_bytes(args...);
template <typename T, typename... Args>
size_t get_size_bytes(dvec<T> *first_vec, size_t first_size, Args... args) {
return get_size_bytes<T>(first_vec, first_size) + get_size_bytes(args...);
}
template <typename T, typename SizeT>
template <typename T>
void allocate_dvec(int device_idx, char *ptr, dvec<T> *first_vec,
SizeT first_size) {
size_t first_size) {
first_vec->external_allocate(device_idx, static_cast<void *>(ptr),
first_size);
}
template <typename T, typename SizeT, typename... Args>
template <typename T, typename... Args>
void allocate_dvec(int device_idx, char *ptr, dvec<T> *first_vec,
SizeT first_size, Args... args) {
first_vec->external_allocate(device_idx, static_cast<void *>(ptr),
first_size);
size_t first_size, Args... args) {
allocate_dvec<T>(device_idx, ptr, first_vec, first_size);
ptr += align_round_up(first_size * sizeof(T));
allocate_dvec(device_idx, ptr, args...);
}
// template <memory_type MemoryT>
char *allocate_device(int device_idx, size_t bytes, memory_type t) {
char *ptr;
if (t == memory_type::DEVICE) {
safe_cuda(cudaSetDevice(device_idx));
safe_cuda(cudaMalloc(&ptr, bytes));
} else {
safe_cuda(cudaMallocManaged(&ptr, bytes));
}
safe_cuda(cudaSetDevice(device_idx));
safe_cuda(cudaMalloc(&ptr, bytes));
return ptr;
}
template <typename T, typename SizeT>
size_t get_size_bytes(dvec2<T> *first_vec, SizeT first_size) {
template <typename T>
size_t get_size_bytes(dvec2<T> *first_vec, size_t first_size) {
return 2 * align_round_up(first_size * sizeof(T));
}
template <typename T, typename SizeT, typename... Args>
size_t get_size_bytes(dvec2<T> *first_vec, SizeT first_size, Args... args) {
return get_size_bytes<T, SizeT>(first_vec, first_size) +
template <typename T, typename... Args>
size_t get_size_bytes(dvec2<T> *first_vec, size_t first_size, Args... args) {
return get_size_bytes<T>(first_vec, first_size) +
get_size_bytes(args...);
}
template <typename T, typename SizeT>
template <typename T>
void allocate_dvec(int device_idx, char *ptr, dvec2<T> *first_vec,
SizeT first_size) {
size_t first_size) {
first_vec->external_allocate(
device_idx, static_cast<void *>(ptr),
static_cast<void *>(ptr + align_round_up(first_size * sizeof(T))),
first_size);
}
template <typename T, typename SizeT, typename... Args>
template <typename T, typename... Args>
void allocate_dvec(int device_idx, char *ptr, dvec2<T> *first_vec,
SizeT first_size, Args... args) {
allocate_dvec<T, SizeT>(device_idx, ptr, first_vec, first_size);
size_t first_size, Args... args) {
allocate_dvec<T>(device_idx, ptr, first_vec, first_size);
ptr += (align_round_up(first_size * sizeof(T)) * 2);
allocate_dvec(device_idx, ptr, args...);
}
@@ -544,14 +535,13 @@ struct CubMemory {
// Thrust
typedef char value_type;
CubMemory() : d_temp_storage(NULL), temp_storage_bytes(0) {}
CubMemory() : d_temp_storage(nullptr), temp_storage_bytes(0) {}
~CubMemory() { Free(); }
template <typename T>
T* Pointer()
{
return static_cast<T*>(d_temp_storage);
T *Pointer() {
return static_cast<T *>(d_temp_storage);
}
void Free() {
@@ -611,7 +601,7 @@ void print(const dvec<T> &v, size_t max_items = 10) {
template <typename coordinate_t, typename segments_t, typename offset_t>
void FindMergePartitions(int device_idx, coordinate_t *d_tile_coordinates,
int num_tiles, int tile_size, segments_t segments,
size_t num_tiles, int tile_size, segments_t segments,
offset_t num_rows, offset_t num_elements) {
dh::launch_n(device_idx, num_tiles + 1, [=] __device__(int idx) {
offset_t diagonal = idx * tile_size;
@@ -692,7 +682,8 @@ void SparseTransformLbs(int device_idx, dh::CubMemory *temp_memory,
const int BLOCK_THREADS = 256;
const int ITEMS_PER_THREAD = 1;
const int TILE_SIZE = BLOCK_THREADS * ITEMS_PER_THREAD;
int num_tiles = dh::div_round_up(count + num_segments, BLOCK_THREADS);
auto num_tiles = dh::div_round_up(count + num_segments, BLOCK_THREADS);
CHECK(num_tiles < std::numeric_limits<unsigned int>::max());
temp_memory->LazyAllocate(sizeof(coordinate_t) * (num_tiles + 1));
coordinate_t *tmp_tile_coordinates =
@@ -702,7 +693,7 @@ void SparseTransformLbs(int device_idx, dh::CubMemory *temp_memory,
BLOCK_THREADS, segments, num_segments, count);
LbsKernel<TILE_SIZE, ITEMS_PER_THREAD, BLOCK_THREADS, offset_t>
<<<num_tiles, BLOCK_THREADS>>>(tmp_tile_coordinates, segments + 1, f,
<<<uint32_t(num_tiles), BLOCK_THREADS>>>(tmp_tile_coordinates, segments + 1, f,
num_segments);
}