lars_momentum_op.cu 22.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
/* Copyright (c) 2016 PaddlePaddle Authors. 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. */

#include "paddle/fluid/framework/op_registry.h"
16
#include "paddle/fluid/operators/amp/fp16_type_traits.h"
17
#include "paddle/fluid/operators/math/math_cuda_utils.h"
W
Wu Yi 已提交
18
#include "paddle/fluid/operators/optimizers/lars_momentum_op.h"
19 20
#include "paddle/fluid/platform/fast_divmod.h"

L
limingshu 已提交
21
#if CUDA_VERSION >= 11000
22 23 24 25 26 27 28 29
#include <cooperative_groups.h>
#endif

#ifdef __HIPCC__
#define LARS_BLOCK_SIZE 256
#else
#define LARS_BLOCK_SIZE 512
#endif
30

Z
Zeng Jinle 已提交
31
#define LARS_MAX_MERGED_OPS 60
L
limingshu 已提交
32

33 34 35 36
namespace paddle {
namespace operators {

template <typename T>
37 38
using MultiPrecisionType = typename details::MPTypeTrait<T>::Type;

39 40 41 42 43 44 45 46 47
__device__ __forceinline__ float Sqrt(float x) { return sqrtf(x); }
__device__ __forceinline__ double Sqrt(double x) { return sqrt(x); }
__device__ __forceinline__ float Fma(float x, float y, float z) {
  return fmaf(x, y, z);
}
__device__ __forceinline__ double Fma(double x, double y, double z) {
  return fma(x, y, z);
}

L
limingshu 已提交
48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
template <typename T>
class LarsThreadConfig {
 public:
  int grid_for_norm;
  int grid_for_lars;
#if CUDA_VERSION >= 11000

 private:
  int grid_stride;

 public:
  explicit LarsThreadConfig(int64_t numel, int sm_num, int num_blocks_per_sm) {
    int grid = (numel + LARS_BLOCK_SIZE - 1) / LARS_BLOCK_SIZE;
    grid_for_lars =
        std::min(std::min(sm_num * num_blocks_per_sm, grid), LARS_BLOCK_SIZE);
    grid_stride = LARS_BLOCK_SIZE * grid_for_lars;
  }

  int GetRepeatTimes(int64_t numel) {
    return (numel + grid_stride - 1) / grid_stride - 1;
  }
#else
  int repeat_times;
  explicit LarsThreadConfig(const int64_t numel) {
    int grid = (numel + LARS_BLOCK_SIZE - 1) / LARS_BLOCK_SIZE;
    grid_for_norm = std::min(grid, LARS_BLOCK_SIZE);
    const int grid_stride = grid_for_norm * LARS_BLOCK_SIZE;
    repeat_times = (numel + grid_stride - 1) / grid_stride - 1;
    // Determine to read 4 fp16 or float data once, but 2 double data once.
    grid_for_lars =
        std::is_same<double, T>::value
            ? (numel + (LARS_BLOCK_SIZE << 1) - 1) / (LARS_BLOCK_SIZE << 1)
            : (numel + (LARS_BLOCK_SIZE << 2) - 1) / (LARS_BLOCK_SIZE << 2);
  }
#endif
};

85 86
template <typename T, typename MT, int VecSize, bool IsAmp = false>
__device__ inline void VectorizeLarsUpdate(
87 88
    const T* __restrict__ grad, const MT* param, const MT* velocity,
    T* param_out, MT* velocity_out, const MT mu, MT local_lr,
89
    const MT lars_weight_decay, const MT rescale_grad, const int tid,
90
    const int grid_stride, const int numel, MT* master_param_out = nullptr) {
91 92 93 94 95
  using VecType = paddle::platform::AlignedVector<T, VecSize>;
  using VecMType = paddle::platform::AlignedVector<MT, VecSize>;
  int main = numel >> (VecSize >> 1);
  int tail_offset = main * VecSize;

96 97 98
  const VecType* grad_vec = reinterpret_cast<const VecType*>(grad);
  const VecMType* param_vec = reinterpret_cast<const VecMType*>(param);
  const VecMType* velocity_vec = reinterpret_cast<const VecMType*>(velocity);
99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142
  VecType* param_out_vec = reinterpret_cast<VecType*>(param_out);
  VecMType* velocity_out_vec = reinterpret_cast<VecMType*>(velocity_out);

  VecMType* master_param_out_vec;
  if (IsAmp) {
    master_param_out_vec = reinterpret_cast<VecMType*>(master_param_out);
  }

  for (int i = tid; i < main; i += grid_stride) {
    VecType param_out_tmp;
    VecMType velocity_tmp, param_tmp;
    VecType grad_data = grad_vec[i];
    VecMType param_data = param_vec[i];
    VecMType velocity_data = velocity_vec[i];
#pragma unroll
    for (int j = 0; j < VecSize; ++j) {
      MT grad_val = static_cast<MT>(grad_data[j]) * rescale_grad;
      velocity_tmp[j] =
          Fma(velocity_data[j], mu,
              local_lr * Fma(lars_weight_decay, param_data[j], grad_val));
      param_tmp[j] = param_data[j] - velocity_tmp[j];
      param_out_tmp[j] = static_cast<T>(param_tmp[j]);
    }
    param_out_vec[i] = param_out_tmp;
    velocity_out_vec[i] = velocity_tmp;
    if (IsAmp) {
      master_param_out_vec[i] = param_tmp;
    }
  }

  for (int i = tid + tail_offset; i < numel; i += grid_stride) {
    MT grad_val = static_cast<MT>(grad[i]) * rescale_grad;
    MT param_val = param[i];
    MT velocity_tmp = Fma(velocity[i], mu, local_lr * Fma(lars_weight_decay,
                                                          param_val, grad_val));
    MT param_tmp = param_val - velocity_tmp;
    param_out[i] = static_cast<T>(param_tmp);
    velocity_out[i] = velocity_tmp;
    if (IsAmp) {
      master_param_out[i] = param_tmp;
    }
  }
}

L
limingshu 已提交
143 144 145 146 147 148 149 150 151 152
#if CUDA_VERSION >= 11000
/* Once CUDA_VERSION is beyond 11, cooperative_groups can be involved in without
  --rdc=true compile flag, then L2_norm kernel can be set with __device__ and
  cooperative_groups::grid_group also can be involved. Otherwise, adding this
  flag may affect much, L2_norm kernel shall be set with __global__.*/
// TODO(limingshu): declaration of cooperative_groups wapper is invalid in host.
template <typename T, typename MT>
__forceinline__ __device__ void L2NormKernel(
    const cooperative_groups::grid_group* cg,
#else
153
template <typename T, typename MT>
L
limingshu 已提交
154 155
__global__ void L2NormKernel(
#endif
156 157 158 159
    const T* p_data, const T* __restrict__ g_data, MT* __restrict__ p_buffer,
    MT* __restrict__ g_buffer, const int64_t numel, const int repeat_times,
    const MT rescale_grad, const int thresh = 0, MT* __restrict__ p_n = nullptr,
    MT* __restrict__ g_n = nullptr) {
L
limingshu 已提交
160
  __shared__ MT s_buffer[2];
161 162
  int tid = threadIdx.x + blockDim.x * blockIdx.x;
  int grid_stride = LARS_BLOCK_SIZE * gridDim.x;
163

L
limingshu 已提交
164 165
  MT p_tmp = static_cast<MT>(0);
  MT g_tmp = static_cast<MT>(0);
166 167 168 169 170 171
  while (tid < numel) {
    MT tmp0 = static_cast<MT>(p_data[tid]);
    MT tmp1 = static_cast<MT>(g_data[tid]);
    p_tmp += (tmp0 * tmp0);
    g_tmp += (tmp1 * tmp1);
    tid += grid_stride;
172
  }
173 174
  p_tmp = math::blockReduceSum<MT>(p_tmp, FINAL_MASK);
  g_tmp = math::blockReduceSum<MT>(g_tmp, FINAL_MASK);
175 176

  if (threadIdx.x == 0) {
177 178
    p_buffer[blockIdx.x] = p_tmp;
    g_buffer[blockIdx.x] = g_tmp;
179 180
  }
#if CUDA_VERSION >= 11000
L
limingshu 已提交
181 182 183
  cg->sync();  // Grid sync for writring partial result to gloabl memory
  MT p_part_sum = threadIdx.x < gridDim.x ? p_buffer[threadIdx.x] : 0;
  MT g_part_sum = threadIdx.x < gridDim.x ? g_buffer[threadIdx.x] : 0;
Z
Zeng Jinle 已提交
184 185 186 187 188 189 190 191
  MT tmp0 = math::blockReduceSum<MT>(p_part_sum, FINAL_MASK);
  MT tmp1 = math::blockReduceSum<MT>(g_part_sum, FINAL_MASK);
  if (threadIdx.x == 0) {
    s_buffer[0] = tmp0;
    s_buffer[1] = tmp1;
  }
  __syncthreads();
  *p_n = Sqrt(s_buffer[0]);
Z
Zeng Jinle 已提交
192
  *g_n = rescale_grad * Sqrt(s_buffer[1]);
193 194
#endif
}
195

196
template <typename T, typename MT>
L
limingshu 已提交
197
__forceinline__ __device__ void MomentumUpdate(
198 199 200
    const T* param, const T* __restrict__ grad, const MT* velocity,
    T* param_out, MT* velocity_out, const MT* master_param,
    MT* master_param_out, const MT* __restrict__ learning_rate, const MT mu,
L
limingshu 已提交
201 202 203 204
    const MT lars_weight_decay, const MT lars_coeff, const MT epsilon,
    const MT rescale_grad, const MT param_norm, const MT grad_norm,
    const int tid, const int grid_stride, const int64_t numel,
    const bool is_amp) {
205 206 207 208
  const MT lr = learning_rate[0];
  MT local_lr = lr;
  if (lars_weight_decay > static_cast<MT>(0)) {
    local_lr = lr * lars_coeff * param_norm /
L
limingshu 已提交
209
               (fma(lars_weight_decay, param_norm, grad_norm) + epsilon);
210
  }
L
limingshu 已提交
211 212 213 214 215
  if (is_amp) {
    VectorizeLarsUpdate<T, MT, /*VecSize=*/4, /*IsAmp=*/true>(
        grad, master_param, velocity, param_out, velocity_out, mu, local_lr,
        lars_weight_decay, rescale_grad, tid, grid_stride, numel,
        master_param_out);
216 217 218
  } else {
    if (std::is_same<T, float>::value ||
        std::is_same<T, paddle::platform::float16>::value) {
L
limingshu 已提交
219 220
      /* TODO(limingshu): pointer cast may damage memory accessing for fp16 */
      VectorizeLarsUpdate<T, MT, /*VecSize=*/4, /*IsAmp=*/false>(
221 222 223 224
          grad, reinterpret_cast<const MT*>(param), velocity, param_out,
          velocity_out, mu, local_lr, lars_weight_decay, rescale_grad, tid,
          grid_stride, numel);
    } else {
L
limingshu 已提交
225
      VectorizeLarsUpdate<T, MT, /*VecSize=*/2, /*IsAmp=*/false>(
226 227 228 229
          grad, reinterpret_cast<const MT*>(param), velocity, param_out,
          velocity_out, mu, local_lr, lars_weight_decay, rescale_grad, tid,
          grid_stride, numel);
    }
230 231 232
  }
}

L
limingshu 已提交
233 234 235 236 237 238 239 240 241 242 243 244 245 246
#if CUDA_VERSION >= 11000
template <typename T, typename MT>
struct LarsParamWarpper {
  int64_t numel_arr[LARS_MAX_MERGED_OPS];
  int repeat_arr[LARS_MAX_MERGED_OPS];
  const T* __restrict__ g_arr[LARS_MAX_MERGED_OPS];
  const MT* __restrict__ lr_arr[LARS_MAX_MERGED_OPS];
  T* __restrict__ p_out_arr[LARS_MAX_MERGED_OPS];
  MT* __restrict__ v_out_arr[LARS_MAX_MERGED_OPS];
  MT* __restrict__ master_p_out_arr[LARS_MAX_MERGED_OPS];
  MT weight_decay_arr[LARS_MAX_MERGED_OPS];
};

template <typename T, typename MT>
Z
Zeng Jinle 已提交
247
__global__ void MergedMomentumLarsKernel(LarsParamWarpper<T, MT> lars_warpper,
L
limingshu 已提交
248 249 250 251 252 253 254 255 256 257
                                         MT* __restrict__ p_buffer,
                                         MT* __restrict__ g_buffer,
                                         const int op_num, const MT mu,
                                         const MT lars_coeff, const MT epsilon,
                                         const MT rescale_grad,
                                         const bool is_amp) {
  int grid_stride = gridDim.x * LARS_BLOCK_SIZE;
  int tid = threadIdx.x + blockIdx.x * blockDim.x;
  const cooperative_groups::grid_group cg = cooperative_groups::this_grid();
  for (int i = 0; i < op_num; ++i) {
Z
Zeng Jinle 已提交
258
    int numel = lars_warpper.numel_arr[i];
L
limingshu 已提交
259 260
    MT param_norm = static_cast<MT>(0);
    MT grad_norm = static_cast<MT>(0);
Z
Zeng Jinle 已提交
261 262
    L2NormKernel<T, MT>(&cg, lars_warpper.p_out_arr[i], lars_warpper.g_arr[i],
                        p_buffer, g_buffer, numel, lars_warpper.repeat_arr[i],
L
limingshu 已提交
263 264
                        rescale_grad, 0, &param_norm, &grad_norm);
    MomentumUpdate<T, MT>(
Z
Zeng Jinle 已提交
265 266 267 268 269
        lars_warpper.p_out_arr[i], lars_warpper.g_arr[i],
        lars_warpper.v_out_arr[i], lars_warpper.p_out_arr[i],
        lars_warpper.v_out_arr[i], lars_warpper.master_p_out_arr[i],
        lars_warpper.master_p_out_arr[i], lars_warpper.lr_arr[i], mu,
        lars_warpper.weight_decay_arr[i], lars_coeff, epsilon, rescale_grad,
L
limingshu 已提交
270 271 272 273 274 275 276
        param_norm, grad_norm, tid, grid_stride, numel, is_amp);
  }
}
#endif

template <typename T, typename MT>
__global__ void MomentumLarsKernel(
277 278 279 280 281 282 283
    const T* param, const T* __restrict__ grad, const MT* velocity,
    T* param_out, MT* velocity_out, const MT* master_param,
    MT* master_param_out, const MT* __restrict__ learning_rate,
    MT* __restrict__ p_buffer, MT* __restrict__ g_buffer, const MT mu,
    const MT lars_coeff, const MT lars_weight_decay, const MT epsilon,
    const MT rescale_grad, const int repeat_times, const int thresh,
    const int64_t numel, const bool is_amp) {
L
limingshu 已提交
284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
  int tid = threadIdx.x + blockIdx.x * blockDim.x;
  int grid_stride = gridDim.x * LARS_BLOCK_SIZE;
#if CUDA_VERSION >= 11000
  const cooperative_groups::grid_group cg = cooperative_groups::this_grid();
  MT param_norm = static_cast<MT>(0);
  MT grad_norm = static_cast<MT>(0);
  L2NormKernel<T, MT>(&cg, param, grad, p_buffer, g_buffer, numel, repeat_times,
                      rescale_grad, gridDim.x, &param_norm, &grad_norm);
#else
  const MT rescale_grad_pow = rescale_grad * rescale_grad;
  MT param_part_norm = threadIdx.x < thresh ? p_buffer[threadIdx.x] : 0;
  MT grad_part_norm = threadIdx.x < thresh ? g_buffer[threadIdx.x] : 0;
  __syncthreads();
  MT param_norm = Sqrt(math::blockReduceSum<MT>(param_part_norm, FINAL_MASK));
  MT grad_norm = Sqrt(rescale_grad_pow *
                      math::blockReduceSum<MT>(grad_part_norm, FINAL_MASK));
#endif
  MomentumUpdate<T, MT>(param, grad, velocity, param_out, velocity_out,
                        master_param, master_param_out, learning_rate, mu,
                        lars_weight_decay, lars_coeff, epsilon, rescale_grad,
                        param_norm, grad_norm, tid, grid_stride, numel, is_amp);
}

template <typename T, typename MT>
inline void SeparatedLarsMomentumOpCUDAKernel(
    const platform::CUDADeviceContext& cuda_ctx, const T* param_data,
    T* param_out_data, const MT* velocity_data, MT* velocity_out_data,
    const T* grad_data, const MT* lr, MT* p_buffer, MT* g_buffer, const MT mu,
    const MT lars_coeff, const MT weight_decay, const MT epsilon,
    const MT rescale_grad, const int64_t numel, const MT* master_param_data,
    MT* master_out_data, const bool is_amp) {
  LarsThreadConfig<T> lars_thread_config(numel);
  L2NormKernel<T, MT><<<lars_thread_config.grid_for_norm, LARS_BLOCK_SIZE, 0,
                        cuda_ctx.stream()>>>(
      param_data, grad_data, p_buffer, g_buffer, numel,
      lars_thread_config.repeat_times, rescale_grad);

  MomentumLarsKernel<T, MT><<<lars_thread_config.grid_for_lars, LARS_BLOCK_SIZE,
                              0, cuda_ctx.stream()>>>(
      param_data, grad_data, velocity_data, param_out_data, velocity_out_data,
      master_param_data, master_out_data, lr, p_buffer, g_buffer, mu,
      lars_coeff, weight_decay, epsilon, rescale_grad, 0,
      lars_thread_config.grid_for_norm, numel, is_amp);
}

329 330
template <typename DeviceContext, typename T>
class LarsMomentumOpCUDAKernel : public framework::OpKernel<T> {
331
  using MT = MultiPrecisionType<T>;
332

333 334
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
335
    int num_blocks_per_sm = 0;
L
limingshu 已提交
336 337
    bool multi_precision = ctx.Attr<bool>("multi_precision");
    auto& cuda_ctx = ctx.template device_context<platform::CUDADeviceContext>();
338 339 340 341 342 343 344
    int sm_num = cuda_ctx.GetSMCount();
    framework::Tensor tmp_buffer_t =
        ctx.AllocateTmpTensor<MT, platform::CUDADeviceContext>(
            {LARS_BLOCK_SIZE << 1}, cuda_ctx);
    auto* p_buffer = tmp_buffer_t.mutable_data<MT>(ctx.GetPlace());
    auto* g_buffer = p_buffer + LARS_BLOCK_SIZE;

L
limingshu 已提交
345 346 347 348
    MT mu = static_cast<MT>(ctx.Attr<float>("mu"));
    MT lars_coeff = static_cast<MT>(ctx.Attr<float>("lars_coeff"));
    MT epsilon = static_cast<MT>(ctx.Attr<float>("epsilon"));
    MT rescale_grad = static_cast<MT>(ctx.Attr<float>("rescale_grad"));
349

L
limingshu 已提交
350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394
    auto weight_decay_arr = ctx.Attr<std::vector<float>>("lars_weight_decay");
    auto grad = ctx.MultiInput<framework::LoDTensor>("Grad");
    auto param = ctx.MultiInput<framework::LoDTensor>("Param");
    auto velocity = ctx.MultiInput<framework::LoDTensor>("Velocity");
    auto param_out = ctx.MultiOutput<framework::LoDTensor>("ParamOut");
    auto velocity_out = ctx.MultiOutput<framework::LoDTensor>("VelocityOut");
    auto learning_rate = ctx.MultiInput<framework::LoDTensor>("LearningRate");
    auto master_param = ctx.MultiInput<framework::LoDTensor>("MasterParam");
    auto master_param_out =
        ctx.MultiOutput<framework::LoDTensor>("MasterParamOut");

    int op_num = grad.size();
#if CUDA_VERSION >= 11000
    if (op_num > 1) {
      LarsParamWarpper<T, MT> lars_warpper;
      PADDLE_ENFORCE_LT(
          op_num, LARS_MAX_MERGED_OPS,
          platform::errors::InvalidArgument(
              "The maximum number of merged-ops supported is (%d), but"
              "lars op required for trainning this model is (%d)\n",
              LARS_MAX_MERGED_OPS, op_num));

      /* Implementation of lars optimizer consists of following two steps:
        1. Figure out the L2 norm statistic result of grad data and param data.
        2. Update param and velocity with usage of L2 norm statistic result.
      Step1 and step2 can be merged with api provided by nvida
        cudaLaunchCooperativeKernel:
        - The thread quantity shall less than pyhsical SM limited threads
        - Launche as thread-block can synchronizlly execute. */
      cudaOccupancyMaxActiveBlocksPerMultiprocessor(
          &num_blocks_per_sm, MergedMomentumLarsKernel<T, MT>, LARS_BLOCK_SIZE,
          sizeof(MT) << 1);

      size_t total_numel = 0;
      for (int i = 0; i < op_num; ++i) {
        size_t temp_numel = param[i]->numel();
        total_numel += temp_numel;
        lars_warpper.numel_arr[i] = temp_numel;
        lars_warpper.g_arr[i] = grad[i]->data<T>();
        lars_warpper.lr_arr[i] = learning_rate[i]->data<MT>();
        lars_warpper.p_out_arr[i] =
            param_out[i]->mutable_data<T>(ctx.GetPlace());
        lars_warpper.v_out_arr[i] =
            velocity_out[i]->mutable_data<MT>(ctx.GetPlace());
        lars_warpper.weight_decay_arr[i] = static_cast<MT>(weight_decay_arr[i]);
Z
Zeng Jinle 已提交
395 396 397 398 399 400 401 402
        PADDLE_ENFORCE_EQ(
            param[i]->data<T>(), lars_warpper.p_out_arr[i],
            platform::errors::InvalidArgument(
                "Input(Param) and Output(ParamOut) must be the same Tensors."));
        PADDLE_ENFORCE_EQ(velocity[i]->data<MT>(), lars_warpper.v_out_arr[i],
                          platform::errors::InvalidArgument(
                              "Input(Velocity) and Output(VelocityOut) must be "
                              "the same Tensors."));
L
limingshu 已提交
403 404 405 406 407 408 409 410 411 412 413 414
      }
      int64_t avg_numel = total_numel / op_num;
      LarsThreadConfig<float> lars_thread_config(avg_numel, sm_num,
                                                 num_blocks_per_sm);
      for (int i = 0; i < op_num; ++i) {
        lars_warpper.repeat_arr[i] =
            lars_thread_config.GetRepeatTimes(lars_warpper.numel_arr[i]);
      }
      if (multi_precision) {
        for (int i = 0; i < op_num; ++i) {
          lars_warpper.master_p_out_arr[i] =
              master_param_out[i]->mutable_data<MT>(ctx.GetPlace());
Z
Zeng Jinle 已提交
415 416 417 418 419
          PADDLE_ENFORCE_EQ(master_param[i]->data<MT>(),
                            lars_warpper.master_p_out_arr[i],
                            platform::errors::InvalidArgument(
                                "Input(MasterParam) and Output(MasterParamOut) "
                                "must be the same Tensors."));
L
limingshu 已提交
420 421
        }
      }
Z
Zeng Jinle 已提交
422
      void* cuda_param[] = {reinterpret_cast<void*>(&lars_warpper),
L
limingshu 已提交
423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503
                            reinterpret_cast<void*>(&p_buffer),
                            reinterpret_cast<void*>(&g_buffer),
                            reinterpret_cast<void*>(&op_num),
                            reinterpret_cast<void*>(&mu),
                            reinterpret_cast<void*>(&lars_coeff),
                            reinterpret_cast<void*>(&epsilon),
                            reinterpret_cast<void*>(&rescale_grad),
                            reinterpret_cast<void*>(&multi_precision)};
      // Lanuch all sm theads, and thead of each block synchronizedly cooperate.
      cudaLaunchCooperativeKernel(
          reinterpret_cast<void*>(MergedMomentumLarsKernel<T, MT>),
          lars_thread_config.grid_for_lars, LARS_BLOCK_SIZE, cuda_param, 0,
          cuda_ctx.stream());
    } else {
      auto* param_data = param[0]->data<T>();
      auto* grad_data = grad[0]->data<T>();
      auto* velocity_data = velocity[0]->data<MT>();
      auto* lr = learning_rate[0]->data<MT>();
      auto* param_out_data = param_out[0]->mutable_data<T>(ctx.GetPlace());
      auto* velocity_out_data =
          velocity_out[0]->mutable_data<MT>(ctx.GetPlace());
      const MT* master_param_data =
          multi_precision ? master_param[0]->data<MT>() : nullptr;
      MT* master_param_out_data =
          multi_precision
              ? master_param_out[0]->mutable_data<MT>(ctx.GetPlace())
              : nullptr;
      int64_t numel = param[0]->numel();
      MT lars_weight_decay = weight_decay_arr[0];

      // Figure out how many blocks can be active in each sm.
      cudaOccupancyMaxActiveBlocksPerMultiprocessor(
          &num_blocks_per_sm, MomentumLarsKernel<T, MT>, LARS_BLOCK_SIZE,
          sizeof(MT) << 1);
      LarsThreadConfig<float> lars_thread_config(numel, sm_num,
                                                 num_blocks_per_sm);
      int repeat_times = lars_thread_config.GetRepeatTimes(numel);
      int thresh = 0;
      void* cuda_param[] = {
          reinterpret_cast<void*>(&param_data),
          reinterpret_cast<void*>(&grad_data),
          reinterpret_cast<void*>(&velocity_data),
          reinterpret_cast<void*>(&param_out_data),
          reinterpret_cast<void*>(&velocity_out_data),
          reinterpret_cast<void*>(&master_param_data),
          reinterpret_cast<void*>(&master_param_out_data),
          reinterpret_cast<void*>(&lr),
          reinterpret_cast<void*>(&p_buffer),
          reinterpret_cast<void*>(&g_buffer),
          reinterpret_cast<void*>(&mu),
          reinterpret_cast<void*>(&lars_coeff),
          reinterpret_cast<void*>(&lars_weight_decay),
          reinterpret_cast<void*>(&epsilon),
          reinterpret_cast<void*>(&rescale_grad),
          reinterpret_cast<void*>(&repeat_times),
          reinterpret_cast<void*>(&thresh),  // Just a placeholder
          reinterpret_cast<void*>(&numel),
          reinterpret_cast<void*>(&multi_precision)};
      // Lanuch all sm theads.
      cudaLaunchCooperativeKernel(
          reinterpret_cast<void*>(MomentumLarsKernel<T, MT>),
          lars_thread_config.grid_for_lars, LARS_BLOCK_SIZE, cuda_param, 0,
          cuda_ctx.stream());
    }
#else
    for (int i = 0; i < op_num; ++i) {
      const MT* master_param_data =
          multi_precision ? master_param[i]->data<MT>() : nullptr;
      MT* master_param_out_data =
          multi_precision
              ? master_param_out[i]->mutable_data<MT>(ctx.GetPlace())
              : nullptr;
      SeparatedLarsMomentumOpCUDAKernel<T, MT>(
          cuda_ctx, param[i]->data<T>(),
          param_out[i]->mutable_data<T>(ctx.GetPlace()),
          velocity[i]->data<MT>(),
          velocity_out[i]->mutable_data<MT>(ctx.GetPlace()), grad[i]->data<T>(),
          learning_rate[i]->data<MT>(), p_buffer, g_buffer, mu, lars_coeff,
          weight_decay_arr[i], epsilon, rescale_grad, param[i]->numel(),
          master_param_data, master_param_out_data, multi_precision);
    }
504
#endif
505 506 507 508 509 510 511 512 513 514
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
REGISTER_OP_CUDA_KERNEL(
    lars_momentum,
    ops::LarsMomentumOpCUDAKernel<paddle::platform::CUDADeviceContext, float>,
515 516 517
    ops::LarsMomentumOpCUDAKernel<paddle::platform::CUDADeviceContext, double>,
    ops::LarsMomentumOpCUDAKernel<paddle::platform::CUDADeviceContext,
                                  paddle::platform::float16>);