Skip to content

Migrate metric ops kernels to FBGEMM_LAUNCH_KERNEL #4453

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 1 commit into from
Closed

Migrate metric ops kernels to FBGEMM_LAUNCH_KERNEL #4453

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
47 changes: 24 additions & 23 deletions fbgemm_gpu/src/metric_ops/metric_ops.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@
#include "fbgemm_gpu/utils/cuda_prelude.cuh"
#include "fbgemm_gpu/utils/dispatch_macros.h"
#include "fbgemm_gpu/utils/inclusive_sum_scan.cuh"
#include "fbgemm_gpu/utils/kernel_launcher.cuh"
#include "metric_ops.h"

constexpr int MAX_ENTRIES_PER_BLOCK = 512;
Expand Down Expand Up @@ -251,28 +252,29 @@ at::Tensor batch_auc(
auto max_smem_size =
at::cuda::getCurrentDeviceProperties()->sharedMemPerBlock;

#define LAUNCH_AUC_KERNEL(pad) \
typedef cub::BlockScan<acc_t, NUM_THREADS_PER_BLOCK> BlockScan; \
TORCH_CHECK( \
sizeof(BlockScan::TempStorage) + \
((MAX_ENTRIES_PER_BLOCK * 2 + 3) * sizeof(acc_t)) <= \
max_smem_size) \
auc_kernel<index_t, label_t, scalar_t, acc_t, pad> \
<<<dim3(grid_size), \
dim3(NUM_THREADS_PER_BLOCK), \
0, \
at::cuda::getCurrentCUDAStream()>>>( \
output.data_ptr<acc_t>(), \
indices.data_ptr<index_t>(), \
labels.data_ptr<label_t>(), \
weights.data_ptr<scalar_t>(), \
num_blocks > 1 ? block_flags.data_ptr<int>() : nullptr, \
num_blocks > 1 ? block_sums.data_ptr<acc_t>() : nullptr, \
num_entries, \
last_block_num_entries, \
padded_num_entries_per_block, \
num_blocks); \
C10_CUDA_KERNEL_LAUNCH_CHECK();
#define LAUNCH_AUC_KERNEL(pad) \
typedef cub::BlockScan<acc_t, NUM_THREADS_PER_BLOCK> BlockScan; \
TORCH_CHECK( \
sizeof(BlockScan::TempStorage) + \
((MAX_ENTRIES_PER_BLOCK * 2 + 3) * sizeof(acc_t)) <= \
max_smem_size) \
\
FBGEMM_LAUNCH_KERNEL( \
(auc_kernel<index_t, label_t, scalar_t, acc_t, pad>), \
dim3(grid_size), \
dim3(NUM_THREADS_PER_BLOCK), \
0, \
at::cuda::getCurrentCUDAStream(), \
output.data_ptr<acc_t>(), \
indices.data_ptr<index_t>(), \
labels.data_ptr<label_t>(), \
weights.data_ptr<scalar_t>(), \
num_blocks > 1 ? block_flags.data_ptr<int>() : nullptr, \
num_blocks > 1 ? block_sums.data_ptr<acc_t>() : nullptr, \
num_entries, \
last_block_num_entries, \
padded_num_entries_per_block, \
num_blocks);

AT_DISPATCH_INDEX_TYPES(indices.scalar_type(), "auc_wrapper_1", [&] {
FBGEMM_DISPATCH_ALL_TYPES(labels.scalar_type(), "auc_wrapper_2", [&] {
Expand All @@ -285,7 +287,6 @@ at::Tensor batch_auc(
} else {
LAUNCH_AUC_KERNEL(2)
}
C10_CUDA_KERNEL_LAUNCH_CHECK();
});
});
});
Expand Down
Loading