TensorRT-LLMs/cpp/tensorrt_llm/kernels/allReduceFusionKernels.h
Yibin Li 32ae1564bd
update FP4 quantize layout (#3045)
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
2025-04-03 13:13:54 -04:00

110 lines
3.6 KiB
C++

/*
* Copyright (c) 2022-2024, 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 <NvInferRuntime.h>
#include <cuda_bf16.h>
#include <cuda_fp16.h>
#include "tensorrt_llm/common/assert.h"
#include "tensorrt_llm/common/cudaUtils.h"
#include "tensorrt_llm/kernels/quantization.h"
#include "tensorrt_llm/runtime/ipcUtils.h"
namespace tensorrt_llm::kernels::ar_fusion
{
static constexpr int kElemsPerAccess = 8;
static constexpr int kOneShotMaxToken = 128;
static constexpr int kBarrierFlagCount = 256;
// DS R1
// pattern1: AR+Add_RMS+Quant
// [m, 7168] bf16 allreduce_in, [m, 7168] bf16 residual_in
// [m, 7168] bf16 residual_out, [m, 7168] fp4 quant_out
// pattern2: AR+AddRMS
// [m, 7168] bf16 allreduce_in, [m, 7168] bf16 residual_in
// [m, 7168] bf16 norm_out
struct AllReduceFusionParams
{
int nranks;
int rank;
nvinfer1::DataType dtype;
// size = token_num * hidden_dim
int size;
int hidden_dim;
void** workspace;
void* allreduce_in;
void* residual_in;
void* residual_out;
void* norm_out;
void* quant_out;
void* scale_out;
void* rms_gamma;
float rms_eps;
float* scale_factor;
FP4QuantizationSFLayout layout = FP4QuantizationSFLayout::SWIZZLED;
cudaStream_t stream;
};
void allreduce_fusion_op(AllReduceFusionParams const& params);
class Workspace
{
public:
Workspace(int rank, int tp_size, int max_token_num, int hidden_dim,
std::shared_ptr<tensorrt_llm::runtime::CudaStream> stream_ptr);
~Workspace();
void** get_workspace();
private:
tensorrt_llm::runtime::WorldConfig m_world_config;
std::shared_ptr<tensorrt_llm::runtime::BufferManager> m_buffer_mgr;
std::vector<tensorrt_llm::runtime::IpcMemory> m_ipc_mem_handles;
void* m_workspace;
std::shared_ptr<tensorrt_llm::runtime::CudaStream> m_cuda_stream;
void* m_flag_d_ptr;
};
void lamport_initialize(void* ptr, int bytes, cudaStream_t stream);
/////////////////////////////////////////////////////////////////
// * MoE Reduction Fusion * //
/////////////////////////////////////////////////////////////////
// Fuse MoE Reduction before AR + RMS
// pattern1: MoE Reduction + Add Residual + AR + ADD_RMS + Quant
// pattern2: MoE Reduction + Add Residual + AR + ADD_RMS
// [device_num_experts, m, 7168] bf16 moe_reduction_active_experts_token_input
// [m, 7168] bf16 moe_reduction_token_input
// [device_num_experts, m] moe_reduction_scale_input
struct MoeReductionAllReduceFusionParams : public AllReduceFusionParams
{
// * moe reduction specific params
// Refer to kernel implementation on layout of those params
// number of active experts on current device
int* moe_reduction_device_num_experts = nullptr;
// per token per expert fp32 scale
float* moe_reduction_scale_input = nullptr;
// per token per expert input
void* moe_reduction_active_experts_token_input = nullptr;
// per token input
void* moe_reduction_token_input = nullptr;
};
void moereduction_allreduce_fusion_op(MoeReductionAllReduceFusionParams const& params);
} // namespace tensorrt_llm::kernels::ar_fusion