[Refactor] Remove dead code multiple places (#61843)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2026-10-16 15:01:07 +00:00
committed by GitHub
parent 8a6c0449b3
commit 8e25bbca19
10 changed files with 14 additions and 168 deletions
-27
View File
@@ -1059,33 +1059,6 @@ void situ_and_mul(torch::stable::Tensor& out, // [..., d]
});
}
void masked_situ_and_mul(torch::stable::Tensor& out, // [E, T, d]
torch::stable::Tensor& input, // [E, T, 2 * d]
const torch::stable::Tensor& expert_num_tokens,
double beta, double linear_beta) {
int num_experts = input.size(0);
int max_num_tokens = input.size(1);
int d = input.size(2) / 2;
if (num_experts == 0 || max_num_tokens == 0 || d == 0) {
return;
}
constexpr int block_size = 256;
dim3 grid((d + block_size - 1) / block_size, num_experts);
dim3 block(block_size);
const torch::stable::accelerator::DeviceGuard device_guard(
input.get_device_index());
const cudaStream_t stream = get_current_cuda_stream();
VLLM_STABLE_DISPATCH_FLOATING_TYPES(
input.scalar_type(), "masked_situ_and_mul_kernel", [&] {
vllm::masked_situ_and_mul_kernel<scalar_t, true>
<<<grid, block, 0, stream>>>(
out.mutable_data_ptr<scalar_t>(),
input.const_data_ptr<scalar_t>(),
expert_num_tokens.const_data_ptr<int>(), max_num_tokens, d,
(float)beta, (float)linear_beta);
});
}
#define LAUNCH_MASKED_ACT_AND_MUL(KERNEL, ACT_FIRST, HAS_CLAMP, LIMIT, ALPHA, \
BETA) \
vllm::masked_act_and_mul_kernel<scalar_t, KERNEL<scalar_t>, ACT_FIRST, \
-4
View File
@@ -579,10 +579,6 @@ void situ_and_mul_quant(
int64_t group_size = 0,
std::optional<torch::stable::Tensor> num_valid_tokens = std::nullopt,
int64_t topk = 1);
void masked_situ_and_mul(torch::stable::Tensor& out,
torch::stable::Tensor& input,
const torch::stable::Tensor& expert_num_tokens,
double beta = 1.0, double linear_beta = -1.0);
void masked_moe_activation(torch::stable::Tensor& out,
torch::stable::Tensor& input,
const torch::stable::Tensor& valid_token_counts,
-4
View File
@@ -640,9 +640,6 @@ STABLE_TORCH_LIBRARY_FRAGMENT(_C, ops) {
"situ_and_mul_quant(Tensor! out, Tensor! scale, Tensor input, "
"float beta=1.0, float linear_beta=-1.0, int group_size=0, "
"Tensor? num_valid_tokens=None, int topk=1) -> ()");
ops.def(
"masked_situ_and_mul(Tensor! out, Tensor input, Tensor "
"expert_num_tokens, float beta=1.0, float linear_beta=-1.0) -> ()");
ops.def(
"masked_moe_activation(Tensor! out, Tensor input, Tensor "
"valid_token_counts, str activation, float clamp_limit=0.0, float "
@@ -875,7 +872,6 @@ STABLE_TORCH_LIBRARY_IMPL(_C, CUDA, ops) {
ops.impl("swigluoai_and_mul", TORCH_BOX(&swigluoai_and_mul));
ops.impl("situ_and_mul", TORCH_BOX(&situ_and_mul));
ops.impl("situ_and_mul_quant", TORCH_BOX(&situ_and_mul_quant));
ops.impl("masked_situ_and_mul", TORCH_BOX(&masked_situ_and_mul));
ops.impl("masked_moe_activation", TORCH_BOX(&masked_moe_activation));
ops.impl("gelu_new", TORCH_BOX(&gelu_new));
ops.impl("gelu_fast", TORCH_BOX(&gelu_fast));