Skip to content

Upstream rocm guards #111

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
wants to merge 1 commit into
base: shbiswas/upstream_0815
Choose a base branch
from
Open

Upstream rocm guards #111

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
15 changes: 14 additions & 1 deletion fbgemm_gpu/codegen/training/backward/embedding_backward_split_indice_weights_template.cu
100644 → 100755
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,10 @@
#include "fbgemm_gpu/utils/assert_macros.h"
#include "fbgemm_gpu/utils/kernel_launcher.cuh"

{%- if is_rocm %}
#include "fbgemm_gpu/rocm/cdna_guard.h"
{%- endif %}

using Tensor = at::Tensor;
using namespace fbgemm_gpu;

Expand Down Expand Up @@ -359,7 +363,16 @@ Tensor {{ mdesc }}_embedding_codegen_grad_indice_weights{{ vdesc }}_cuda(
auto aligned_grad_output = aligned_grad_output_tensor_for_cuda_backwards(grad_output);

CUDA_DEVICE_GUARD(dev_weights);

#ifdef USE_ROCM
if (!rocm::is_supported_cdna()) {
TORCH_WARN_ONCE("Running on non-CDNA architecture. Performance may be suboptimal.");
}
else {
// Ensure we're running on a supported CDNA architecture (including MI350)
TORCH_WARN_ONCE("Running on CDNA architecture");
}
#endif

const auto T = D_offsets.size(0) - 1;
TORCH_CHECK_GT(T, 0);
// offsets = [B x T + 1]
Expand Down
10 changes: 10 additions & 0 deletions fbgemm_gpu/codegen/training/backward/embedding_backward_split_template.cu
100644 → 100755
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -652,6 +652,16 @@ Tensor {{ embedding_cuda_op }}(

CUDA_DEVICE_GUARD(dev_weights);

#ifdef USE_ROCM
if (!rocm::is_supported_cdna()) {
TORCH_WARN_ONCE("Running on non-CDNA architecture. Performance may be suboptimal.");
}
else {
// Ensure we're running on a supported CDNA architecture (including MI350)
TORCH_WARN_ONCE("Running on CDNA architecture");
}
#endif

{%- if nobag and not is_index_select %}
auto max_D = D;
{%- endif %}
Expand Down
14 changes: 14 additions & 0 deletions fbgemm_gpu/codegen/training/forward/embedding_forward_split_template.cu
100644 → 100755
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,10 @@
#include "fbgemm_gpu/utils/dispatch_macros.h"
{%- endif %}

{%- if is_rocm %}
#include "fbgemm_gpu/rocm/cdna_guard.h"
{%- endif %}

{%- if not is_index_select %}
////////////////////////////////////////////////////////////////////////////////
// Required for op registrations
Expand Down Expand Up @@ -454,6 +458,16 @@ batch_index_select_dim0_codegen_forward_cuda(

CUDA_DEVICE_GUARD(dev_weights);

#ifdef USE_ROCM
if (!rocm::is_supported_cdna()) {
TORCH_WARN_ONCE("Running on non-CDNA architecture. Performance may be suboptimal.");
}
else {
// Ensure we're running on a supported CDNA architecture (including MI350)
TORCH_WARN_ONCE("Running on CDNA architecture");
}
#endif

{%- if not nobag %}
int32_t T = D_offsets.numel() - 1;
{%- else %}
Expand Down