Skip to content

Fix the MultiStages for intranode combine and the offset_int4 for int… #527

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

Open
cyhdmjzzy wants to merge 1 commit into
base: main
Choose a base branch
from
+10 −5
Open

Fix the MultiStages for intranode combine and the offset_int4 for int… #527

Show file tree
Hide file tree
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
2 changes: 1 addition & 1 deletion csrc/kernels/internode_ll.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -866,7 +866,7 @@ __global__ __launch_bounds__(1024, 1) void combine(void* combined_x,
const int& next_stage_idx = (iter_idx + 1) % kNumStages;
if (iter_idx + 1 < kNumIters and elect_one_sync()) {
tma_store_wait<kNumStages - kNumPrefetch - 1>();
const auto& offset_int4 = i + 32 * kNumSendUnrolls;
const auto& offset_int4 = (iter_idx + 1) * 32 * kNumSendUnrolls;
tma_load_and_arrive(next_stage_idx, cpy_src_int4_ptr + offset_int4, get_num_tma_bytes(offset_int4));
}
__syncwarp();
Expand Down
13 changes: 9 additions & 4 deletions csrc/kernels/intranode.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -983,8 +983,9 @@ __global__ void __launch_bounds__(kNumThreads, 1) combine(dtype_t* recv_x,
#ifndef DISABLE_SM90_FEATURES
if (i < hidden_int4_aligned) {
// Wait TMA arrival
tma_store_wait<kNumStages - 1>();
__syncwarp();
if (lane_id < kNumStages) {
tma_store_wait<0>();
}

// Write into TMA buffer
auto tma_stage_idx = (i / 32) % kNumStages;
Expand All @@ -993,7 +994,7 @@ __global__ void __launch_bounds__(kNumThreads, 1) combine(dtype_t* recv_x,
// Issue TMA
tma_store_fence();
__syncwarp();
if (elect_one_sync()) {
if (lane_id < kNumStages) {
auto tma_bytes = min(32, hidden_int4 - i) * static_cast<int>(sizeof(int4));
tma_store_1d(reinterpret_cast<int4*>(tma_buffer) + tma_stage_idx * 32,
recv_int4 + token_idx * hidden_int4 + i,
Expand All @@ -1005,7 +1006,11 @@ __global__ void __launch_bounds__(kNumThreads, 1) combine(dtype_t* recv_x,
#endif
recv_int4[token_idx * hidden_int4 + i] = out_int4;
#ifndef DISABLE_SM90_FEATURES
}
}
}
// Flush all stores
tma_store_wait<0>();
__syncwarp();
#endif
}

Expand Down