696dcc9265
<!-- Thanks for sending a pull request! BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing/overview.html --> ### What this PR does / why we need it? <!-- - Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. - Please clarify why the changes are needed. For instance, the use case and bug description. - Fixes # --> The triton kernels in sample encounter some problems, scenarios are shown below: 1. 【expand_kernel/ rejection_random_sample_kernel/ prepare_inputs_padded_kernel】, these three operations will use ‘tl.load(prt + offsets -1, mask)’ in their implementations, but triton compiler reports that the masks in these scenarios are not static and contiguous. As a result, compiler will first access this memory and apply the mask. Therefore, I modified the code to ‘tl.load(prt +tl.maximum(offsets - 1, 0), mask)’ to ensure no -1 reads. 2. 【sample_recovered_tokens_kernel/ rejection_random_sample_kernel】, this kernel uses draft_token_id as an address offset for the load operation. In the PD separation scenario, if the pad token is -1, illegal memory reads and writes can occur. Therefore, i modified the kernel and so they can do well with -1 token. ### Does this PR introduce _any_ user-facing change? <!-- Note that it means *any* user-facing change including all aspects such as API, interface or other behavior changes. Documentation-only updates are not considered user-facing changes. --> ### How was this patch tested? <!-- CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. --> Signed-off-by: ppppeng <zepengliu912@qq.com> Co-authored-by: zepengliu912@qq.com <root@localhost.localdomain>
66 lines
2.7 KiB
Python
66 lines
2.7 KiB
Python
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
# This file is a part of the vllm-ascend project.
|
|
#
|
|
# Adapted from https://github.com/vllm-project/vllm/blob/main/vllm/v1/spec_decode/utils.py
|
|
|
|
from vllm.triton_utils import tl, triton
|
|
|
|
|
|
@triton.jit(do_not_specialize=["num_reqs"])
|
|
def prepare_inputs_padded_kernel(
|
|
cu_num_draft_tokens_ptr, # [num_reqs]
|
|
valid_sampled_tokens_count_ptr, # [num_reqs]
|
|
query_start_loc_gpu_ptr, # [num_reqs + 1]
|
|
token_indices_to_sample_ptr, # [num_reqs] (output)
|
|
num_rejected_tokens_gpu_ptr,
|
|
num_reqs, # tl.int32
|
|
BLOCK_SIZE: tl.constexpr,
|
|
):
|
|
pid = tl.program_id(axis=0)
|
|
num_programs = tl.num_programs(axis=0)
|
|
|
|
# Grid-Stride Loop:
|
|
block_start_step = num_programs * BLOCK_SIZE
|
|
|
|
for block_start in tl.range(pid * BLOCK_SIZE, num_reqs, block_start_step):
|
|
offsets = block_start + tl.arange(0, BLOCK_SIZE)
|
|
mask = offsets < num_reqs
|
|
|
|
# Calculate num_draft_tokens from cu_num_draft_tokens, which is an inclusive
|
|
# cumulative sum (first entry is the first value, not zero).
|
|
cu_draft_curr = tl.load(cu_num_draft_tokens_ptr + offsets, mask=mask)
|
|
|
|
prev_indices = tl.maximum(offsets - 1, 0)
|
|
has_prev = offsets > 0
|
|
cu_draft_prev = tl.load(
|
|
cu_num_draft_tokens_ptr + prev_indices,
|
|
mask=mask & has_prev,
|
|
other=0,
|
|
)
|
|
|
|
num_draft_tokens = tl.where(has_prev, cu_draft_curr - cu_draft_prev, cu_draft_curr)
|
|
|
|
valid_count = tl.load(valid_sampled_tokens_count_ptr + offsets, mask=mask)
|
|
num_rejected = num_draft_tokens + 1 - valid_count
|
|
num_rejected = tl.where(num_draft_tokens > 0, num_rejected, 0)
|
|
|
|
# query_start_loc[req_idx + 1] is the start position of the next request,
|
|
# which is one past the last token of this request.
|
|
q_last_tok_idx = tl.load(query_start_loc_gpu_ptr + offsets + 1, mask=mask) - 1
|
|
|
|
index_to_sample = q_last_tok_idx - num_rejected
|
|
tl.store(token_indices_to_sample_ptr + offsets, index_to_sample, mask=mask)
|
|
tl.store(num_rejected_tokens_gpu_ptr + offsets, num_rejected, mask=mask)
|