mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
134 lines
4.7 KiB
C++
134 lines
4.7 KiB
C++
/*
|
|
* Copyright (c) 2020-2023, NVIDIA CORPORATION. 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.
|
|
*/
|
|
|
|
#pragma once
|
|
|
|
#include <cassert>
|
|
#include <cstring>
|
|
#include <iostream>
|
|
#include <memory>
|
|
#include <tuple>
|
|
#include <vector>
|
|
|
|
#include <cublas_v2.h>
|
|
#include <cuda_fp16.h>
|
|
#include <cuda_runtime.h>
|
|
|
|
#include "fused_multihead_attention_common.h"
|
|
#include "tensorrt_llm/common/cudaUtils.h"
|
|
#include "tmaDescriptor.h"
|
|
|
|
namespace tensorrt_llm
|
|
{
|
|
namespace kernels
|
|
{
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
class MHARunner
|
|
{
|
|
public:
|
|
MHARunner(
|
|
Data_type const dataType, bool const pagedKVFMHA, int const numHeads, int const headSize, float const qScaling);
|
|
|
|
MHARunner() = default;
|
|
|
|
virtual ~MHARunner() = default;
|
|
|
|
virtual void setup(int const b, int const s_q, int const s_kv, int const blocks_per_context_sequence,
|
|
int const tokens_per_kv_block, int const sliding_window_size, int const total_seqlen,
|
|
bool const has_alibi = false, bool const scale_alibi = false, int const tp_size = 1, int const tp_rank = 0)
|
|
= 0;
|
|
|
|
// Keep this for bert attention plugin.
|
|
virtual void setup(int const b, int const s, int const sliding_window_size, int const total_seqlen) = 0;
|
|
|
|
static bool fmha_supported(int const headSize, int const sm);
|
|
|
|
virtual bool fmha_supported() = 0;
|
|
|
|
virtual void setup_flags(bool const force_fp32_acc, bool const is_s_padded, bool const causal_mask,
|
|
int const num_kv_heads /* MQA or GQA */)
|
|
= 0;
|
|
|
|
virtual void run(void const* qPtr, void const* pagedKVBlockOffsetsOnHost, KVBlockArray const& pagedKVCache,
|
|
void const* cuQSeqlenPtr, void const* cuKVSeqlenPtr, uint32_t* tileCounterPtr, float const* scaleBmm2Ptr,
|
|
void* outputPtr, cudaStream_t stream)
|
|
= 0;
|
|
|
|
// Keep this for bert attention plugin.
|
|
virtual void run(void const* inputPtr, void const* cuQSeqlenPtr, uint32_t* tileCounterPtr,
|
|
float const* scaleBmm2Ptr, void* outputPtr, cudaStream_t stream)
|
|
= 0;
|
|
|
|
virtual bool isValid(int s) const = 0;
|
|
};
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// Workflow of fmha runner:
|
|
// 1. check if FMHA kernels are supported statically.
|
|
// 2. construct FMHA runner object.
|
|
// 3. setup_flags (used by all kernels).
|
|
// 4. setup runtime parameters (used by this specific case).
|
|
// 5. run the kernel (with all needed device pointers).
|
|
|
|
class FusedMHARunnerV2 : public MHARunner
|
|
{
|
|
public:
|
|
FusedMHARunnerV2(
|
|
Data_type const dataType, bool const pagedKVFMHA, int const numHeads, int const headSize, float const qScaling);
|
|
|
|
~FusedMHARunnerV2(); // for pimpl
|
|
|
|
void setup(int const b, int const s_q, int const s_kv, int const blocks_per_context_sequence,
|
|
int const tokens_per_kv_block, int const sliding_window_size, int const total_seqlen,
|
|
bool const has_alibi = false, bool const scale_alibi = false, int const tp_size = 1,
|
|
int const tp_rank = 0) override;
|
|
|
|
// Keep this for bert attention plugin.
|
|
void setup(int const b, int const s, int const sliding_window_size, int const total_seqlen)
|
|
{
|
|
setup(b, s, s, 0, 0, sliding_window_size, total_seqlen);
|
|
}
|
|
|
|
bool fmha_supported() override;
|
|
|
|
void run(void const* qPtr, void const* pagedKVBlockOffsetsOnHost, KVBlockArray const& pagedKVCache,
|
|
void const* cuQSeqlenPtr, void const* cuKVSeqlenPtr, uint32_t* tileCounterPtr, float const* scaleBmm2Ptr,
|
|
void* outputPtr, cudaStream_t stream) override;
|
|
|
|
// Keep this for bert attention plugin.
|
|
void run(void const* inputPtr, void const* cuQSeqlenPtr, uint32_t* tileCounterPtr, float const* scaleBmm2Ptr,
|
|
void* outputPtr, cudaStream_t stream)
|
|
{
|
|
run(inputPtr, nullptr, KVBlockArray(), cuQSeqlenPtr, cuQSeqlenPtr, tileCounterPtr, scaleBmm2Ptr, outputPtr,
|
|
stream);
|
|
}
|
|
|
|
void setup_flags(bool const force_fp32_acc, bool const is_s_padded, bool const causal_mask,
|
|
int const num_kv_heads /* MQA or GQA */) override;
|
|
|
|
bool isValid(int s) const override;
|
|
|
|
private:
|
|
class mhaImpl;
|
|
std::unique_ptr<mhaImpl> pimpl;
|
|
};
|
|
|
|
} // namespace kernels
|
|
} // namespace tensorrt_llm
|