Skip to content

CUDA: use registers instead of smem in topk-moe #16647

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

Merged
merged 1 commit into from
Oct 18, 2025
+23 −19
Merged

CUDA: use registers instead of smem in topk-moe #16647

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
42 changes: 23 additions & 19 deletions ggml/src/ggml-cuda/topk-moe.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,8 +73,7 @@ __launch_bounds__(4 * WARP_SIZE, 1) __global__ void topk_moe_cuda(const float *

float wt_sum = 0.f;

extern __shared__ float data_topk_shared[];
float * wt_shared_ptr = data_topk_shared + threadIdx.y * n_expert_used;
float output_weights[experts_per_thread];

for (int k = 0; k < n_expert_used; k++) {
float max_val = wt[0];
Expand All @@ -99,11 +98,14 @@ __launch_bounds__(4 * WARP_SIZE, 1) __global__ void topk_moe_cuda(const float *
}
}

if ((k & (WARP_SIZE - 1)) == threadIdx.x) {
output_weights[k / WARP_SIZE] = max_val;
}

if ((max_expert & (WARP_SIZE - 1)) == threadIdx.x) {
wt[max_expert / WARP_SIZE] = -INFINITY;

wt_shared_ptr[k] = max_val;
ids[k] = max_expert;
ids[k] = max_expert;
if constexpr (with_norm) {
wt_sum += max_val;
}
Expand All @@ -115,12 +117,16 @@ __launch_bounds__(4 * WARP_SIZE, 1) __global__ void topk_moe_cuda(const float *
const float inv_sum = 1.0f / wt_sum;

for (int i = threadIdx.x; i < n_expert_used; i += WARP_SIZE) {
wt_shared_ptr[i] = wt_shared_ptr[i] * inv_sum;
output_weights[i] *= inv_sum;
}
}

for (int i = threadIdx.x; i < n_expert_used; i += WARP_SIZE) {
weights[i] = wt_shared_ptr[i];
#pragma unroll
for (int i = 0; i < experts_per_thread; i++) {
const int idx = i * WARP_SIZE + threadIdx.x;
if (idx < n_expert_used) {
weights[idx] = output_weights[i];
}
}
}

Expand All @@ -137,48 +143,46 @@ static void launch_topk_moe_cuda(ggml_backend_cuda_context & ctx,
dim3 block_dims(WARP_SIZE, rows_per_block, 1);
cudaStream_t stream = ctx.stream();

const int nbytes_shared = n_expert_used * rows_per_block * sizeof(float);

switch (n_expert) {
case 1:
topk_moe_cuda<1, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
case 2:
topk_moe_cuda<2, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
case 4:
topk_moe_cuda<4, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
case 8:
topk_moe_cuda<8, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
case 16:
topk_moe_cuda<16, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
case 32:
topk_moe_cuda<32, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
case 64:
topk_moe_cuda<64, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
case 128:
topk_moe_cuda<128, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
case 256:
topk_moe_cuda<256, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
case 512:
topk_moe_cuda<512, with_norm>
<<<grid_dims, block_dims, nbytes_shared, stream>>>(logits, weights, ids, n_rows, n_expert_used);
<<<grid_dims, block_dims, 0, stream>>>(logits, weights, ids, n_rows, n_expert_used);
break;
default:
GGML_ASSERT(false && "fatal error");
Expand Down