Skip to content

fix: prevent MXFP8 amax aliasing with dSoftmaxOffset in bwd #4

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
vedaanta wants to merge 1 commit into
base: add_mxfp8
Choose a base branch
from
Open

fix: prevent MXFP8 amax aliasing with dSoftmaxOffset in bwd #4

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
31 changes: 27 additions & 4 deletions transformer_engine/common/fused_attn/fused_attn_fp8.cu
100644 → 100755
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2898,11 +2898,25 @@ void fused_attn_fp8_bwd(
void* devPtrdK = output_dK->data.dptr;
void* devPtrdV = output_dV->data.dptr;
void* devPtrAmaxdQ = output_dQ->amax.dptr;
void* devPtrAmaxdK = output_dQ->amax.dptr;
void* devPtrAmaxdV = output_dQ->amax.dptr;
void* devPtrAmaxdK = output_dK->amax.dptr;
void* devPtrAmaxdV = output_dV->amax.dptr;
void* devPtrScaledQ = output_dQ->scale.dptr;
void* devPtrScaledK = output_dQ->scale.dptr;
void* devPtrScaledV = output_dQ->scale.dptr;
void* devPtrScaledK = output_dK->scale.dptr;
void* devPtrScaledV = output_dV->scale.dptr;
// For MXFP8, amax pointers are unused but cuDNN still writes to them.
// They may alias other output tensors (e.g. dSoftmaxOffset) since the allocator
// can reuse memory for these dummy scalars. Use scratch space to prevent stomping.
float* amax_dqkv_scratch = nullptr;
if (input_Q->scaling_mode == NVTE_MXFP8_1D_SCALING && softmax_type != NVTE_VANILLA_SOFTMAX &&
output_dSoftmaxOffset->data.dptr != nullptr &&
(devPtrAmaxdQ == output_dSoftmaxOffset->data.dptr ||
devPtrAmaxdK == output_dSoftmaxOffset->data.dptr ||
devPtrAmaxdV == output_dSoftmaxOffset->data.dptr)) {
NVTE_CHECK_CUDA(cudaMallocAsync(&amax_dqkv_scratch, 3 * sizeof(float), stream));
devPtrAmaxdQ = &amax_dqkv_scratch[0];
devPtrAmaxdK = &amax_dqkv_scratch[1];
devPtrAmaxdV = &amax_dqkv_scratch[2];
}

void* devPtrcuSeqlensQ =
reinterpret_cast<void*>(reinterpret_cast<int32_t*>(cu_seqlens_q->data.dptr));
Expand Down Expand Up @@ -2954,13 +2968,22 @@ void fused_attn_fp8_bwd(
if (workspace->data.dptr == nullptr) {
workspace->data.shape = {workspace_size};
workspace->data.dtype = DType::kByte;
if (amax_dqkv_scratch != nullptr) {
NVTE_CHECK_CUDA(cudaFreeAsync(amax_dqkv_scratch, stream));
}
return;
}
} else if (workspace_size == 0) {
workspace->data.shape = {1};
workspace->data.dtype = DType::kByte;
if (amax_dqkv_scratch != nullptr) {
NVTE_CHECK_CUDA(cudaFreeAsync(amax_dqkv_scratch, stream));
}
return;
}
if (amax_dqkv_scratch != nullptr) {
NVTE_CHECK_CUDA(cudaFreeAsync(amax_dqkv_scratch, stream));
}
}
#endif // end of CUDNN>=8900
} // namespace transformer_engine