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

[GPU-Plugin] Fix gpu_hist to allow matrices with more than just 2^{32} elements. Also fixed CPU hist algorithm. (#2518)

Browse Source
This commit is contained in:
PSEUDOTENSOR / Jonathan McKinney
2017-07-17 16:19:27 -07:00
committed by Rory Mitchell
parent c85bf9859e
commit ca7fc9fda3
11 changed files with 413 additions and 283 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
plugin/updater_gpu/src/device_helpers.cuh
View File

@@ -441,7 +441,7 @@ class bulk_allocator {
public:
~bulk_allocator() {
for (int i = 0; i < d_ptr.size(); i++) {
for (size_t i = 0; i < d_ptr.size(); i++) {
if (!(d_ptr[i] == nullptr)) {
safe_cuda(cudaSetDevice(_device_idx[i]));
safe_cuda(cudaFree(d_ptr[i]));
@@ -522,7 +522,7 @@ inline size_t available_memory(int device_idx) {
template <typename T>
void print(const thrust::device_vector<T> &v, size_t max_items = 10) {
thrust::host_vector<T> h = v;
for (int i = 0; i < std::min(max_items, h.size()); i++) {
for (size_t i = 0; i < std::min(max_items, h.size()); i++) {
std::cout << " " << h[i];
}
std::cout << "\n";
@@ -531,7 +531,7 @@ void print(const thrust::device_vector<T> &v, size_t max_items = 10) {
template <typename T>
void print(const dvec<T> &v, size_t max_items = 10) {
std::vector<T> h = v.as_vector();
for (int i = 0; i < std::min(max_items, h.size()); i++) {
for (size_t i = 0; i < std::min(max_items, h.size()); i++) {
std::cout << " " << h[i];
}
std::cout << "\n";
@@ -539,10 +539,10 @@ void print(const dvec<T> &v, size_t max_items = 10) {
template <typename T>
void print(char *label, const thrust::device_vector<T> &v,
const char *format = "%d ", int max = 10) {
const char *format = "%d ", size_t max = 10) {
thrust::host_vector<T> h_v = v;
std::cout << label << ":\n";
for (int i = 0; i < std::min(static_cast<int>(h_v.size()), max); i++) {
for (size_t i = 0; i < std::min(static_cast<size_t>(h_v.size()), max); i++) {
printf(format, h_v[i]);
}
std::cout << "\n";
@@ -593,6 +593,7 @@ __global__ void launch_n_kernel(int device_idx, size_t begin, size_t end,
template <int ITEMS_PER_THREAD = 8, int BLOCK_THREADS = 256, typename L>
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);
#if defined(__CUDACC__)
launch_n_kernel<<<GRID_SIZE, BLOCK_THREADS>>>(static_cast<size_t>(0), n,
@@ -607,6 +608,7 @@ inline void multi_launch_n(size_t n, int n_devices, L lambda) {
CHECK_LE(n_devices, n_visible_devices()) << "Number of devices requested "
"needs to be less than equal to "
"number of visible devices.";
// TODO: Template on n so GRID_SIZE always fits into int.
const int GRID_SIZE = div_round_up(n, ITEMS_PER_THREAD * BLOCK_THREADS);
#if defined(__CUDACC__)
n_devices = n_devices > n ? n : n_devices;