derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
1348 lines
54 KiB
C++
1348 lines
54 KiB
C++
/*******************************************************************************
|
|
*
|
|
* MIT License
|
|
*
|
|
* Copyright (c) 2025 Advanced Micro Devices, Inc.
|
|
*
|
|
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
* of this software and associated documentation files (the "Software"), to deal
|
|
* in the Software without restriction, including without limitation the rights
|
|
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
* copies of the Software, and to permit persons to whom the Software is
|
|
* furnished to do so, subject to the following conditions:
|
|
*
|
|
* The above copyright notice and this permission notice shall be included in all
|
|
* copies or substantial portions of the Software.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
* SOFTWARE.
|
|
*
|
|
*******************************************************************************/
|
|
#ifndef GUARD_MIOPEN_RNN_VERIFY_GEMM_HPP
|
|
#define GUARD_MIOPEN_RNN_VERIFY_GEMM_HPP
|
|
|
|
#include "dropout_gpu_emulator.hpp"
|
|
|
|
#include <../test/rnn_util.hpp>
|
|
|
|
#include <algorithm>
|
|
#include <cassert>
|
|
#include <cmath>
|
|
|
|
template <typename Tgpu, typename Tref>
|
|
void RunRNNForwardGEMMCPUVerify(miopenHandle_t handle,
|
|
std::vector<Tgpu>& in,
|
|
std::vector<Tgpu>& wei, // [ input_state_weight_trans
|
|
// hidden_state_weight0_trans input1_trans
|
|
// hidden1_trans ... output_weight;
|
|
// bidirectional reversed weights ]
|
|
std::vector<Tref>& hy_host, // current/final hidden state
|
|
std::vector<Tgpu>& hx, // initial hidden state
|
|
std::vector<Tref>& out_host,
|
|
std::vector<int>& in_n, // input batch size
|
|
int in_h, // input data length
|
|
int seqLength, // Number of iterations to unroll over
|
|
bool bidirection, // whether using bidirectional net
|
|
bool biased, // whether using bias
|
|
int hy_d, // 1 by numlayer (number of stacks of hidden layers) for
|
|
// unidirection, 2 by numlayer for bidirection
|
|
int hy_n, // equal to input batch size in_n[0]
|
|
int hy_h, // hidden state number
|
|
int out_h, // 1 by hy_h related function for unidirection, 2 by hy_h
|
|
// related function for bidirection
|
|
int squash,
|
|
int inputMode,
|
|
std::vector<Tref>& rsvspace_host,
|
|
bool use_dropout,
|
|
miopenDropoutDescriptor_t dropoutDesc,
|
|
bool hx_is_null = false)
|
|
{
|
|
// printf("FWD TRAIN CPU:\n");
|
|
// printf("seqLen: %d, in_h: %d, hy_d: %d, hy_n: %d, hy_h: %d, out_h: %d\n", seqLength, in_h,
|
|
// hy_d, hy_n, hy_h, out_h);
|
|
// printf("dirmode: %d, hx size: %d, hy_host size: %d, reserveSpace: %d\n", bidirection ? 2 :
|
|
// 1, hx.size(), hy_host.size(), rsvspace_host.size());
|
|
// printf("input size: %d\n", in.size());
|
|
// printf("output size: %d\n", out_host.size());
|
|
int batch_n = sumvc(in_n);
|
|
std::vector<Tref> hid_state(hy_d * batch_n * hy_h, static_cast<Tref>(0));
|
|
std::vector<Tref> wk_state(hy_d * batch_n * hy_h, static_cast<Tref>(0));
|
|
std::vector<Tref> out_state(batch_n * out_h, static_cast<Tref>(0));
|
|
|
|
int numlayer = bidirection ? hy_d / 2 : hy_d;
|
|
int bacc, baccbi; // accumulation of batch
|
|
int bi = bidirection ? 2 : 1;
|
|
|
|
int in_stride = in_h;
|
|
int hy_stride = hy_h * bi;
|
|
int out_stride = out_h;
|
|
int uni_stride = hy_h;
|
|
int bi_stride = hy_h * bi;
|
|
|
|
// initial input
|
|
std::vector<Tref> in_state(batch_n * in_h, static_cast<Tref>(0));
|
|
for(int h = 0; h < batch_n; h++)
|
|
{
|
|
for(int w = 0; w < in_h; w++)
|
|
{
|
|
in_state.at(h * in_h + w) = in.at(h * in_h + w);
|
|
}
|
|
}
|
|
|
|
// initial hidden states
|
|
std::vector<Tref> hy_state(hy_d * hy_n * hy_h, static_cast<Tref>(0));
|
|
std::vector<Tref> hx_state(hy_d * hy_n * hy_h, static_cast<Tref>(0));
|
|
for(int h = 0; h < hy_d * hy_n * hy_h; h++)
|
|
{
|
|
hx_state.at(h) = hx.at(h);
|
|
}
|
|
|
|
if(inputMode == 1)
|
|
{
|
|
if(in_h != hy_h)
|
|
{
|
|
printf("Verification cannot be completed: The input tensor size must equal to the "
|
|
"hidden state size of the network in SKIP_INPUT mode!\n");
|
|
return;
|
|
}
|
|
in_h = 0;
|
|
}
|
|
|
|
// initial weights
|
|
int wei_len = (bi * (in_h + hy_h) + (numlayer - 1) * bi * (bi + 1) * hy_h) * hy_h;
|
|
if(biased)
|
|
{
|
|
int in_bias = 2;
|
|
wei_len += (bi * in_bias + (numlayer - 1) * bi * 2) * hy_h;
|
|
}
|
|
|
|
std::vector<Tref> wei_state(wei_len, static_cast<Tref>(0));
|
|
for(int h = 0; h < wei_len; h++)
|
|
{
|
|
wei_state.at(h) = wei[h];
|
|
}
|
|
|
|
int wei_shift_bias = ((in_h + hy_h) * bi + (bi * hy_h + hy_h) * bi * (numlayer - 1)) * hy_h;
|
|
|
|
// initial dropoput
|
|
std::vector<rocrand_state_xorwow> dropout_states_host;
|
|
std::vector<unsigned char> dropout_reservespace_host;
|
|
std::vector<Tref> dropout_hid_state;
|
|
miopenTensorDescriptor_t dropout_inputTensor{}, dropout_outputTensor{};
|
|
if(use_dropout)
|
|
{
|
|
size_t statesSizeInBytes = 0;
|
|
miopenDropoutGetStatesSize(handle, &statesSizeInBytes);
|
|
size_t states_size = statesSizeInBytes / sizeof(rocrand_state_xorwow);
|
|
dropout_states_host = std::vector<rocrand_state_xorwow>(states_size);
|
|
InitKernelStateEmulator(dropout_states_host, dropoutDesc);
|
|
|
|
std::array<int, 2> drop_in_len = {{batch_n, hy_h * bi}};
|
|
std::array<int, 2> drop_in_str = {{hy_stride, 1}};
|
|
std::array<int, 2> drop_out_str = {{hy_h * bi, 1}};
|
|
miopenCreateTensorDescriptor(&dropout_inputTensor);
|
|
miopenCreateTensorDescriptor(&dropout_outputTensor);
|
|
miopenSetTensorDescriptor(
|
|
dropout_inputTensor, miopenFloat, 2, drop_in_len.data(), drop_in_str.data());
|
|
miopenSetTensorDescriptor(
|
|
dropout_outputTensor, miopenFloat, 2, drop_in_len.data(), drop_out_str.data());
|
|
|
|
size_t reserveSpaceSizeInBytes = 0;
|
|
miopenDropoutGetReserveSpaceSize(dropout_inputTensor, &reserveSpaceSizeInBytes);
|
|
size_t reserve_size = reserveSpaceSizeInBytes / sizeof(unsigned char);
|
|
dropout_reservespace_host = std::vector<unsigned char>(reserve_size * (numlayer - 1),
|
|
static_cast<unsigned char>(1));
|
|
|
|
dropout_hid_state =
|
|
std::vector<Tref>((numlayer - 1) * batch_n * hy_h * bi, static_cast<Tref>(0));
|
|
}
|
|
|
|
// forward emulator
|
|
for(int li = 0; li < numlayer; li++)
|
|
{
|
|
int hid_shift = li * batch_n * hy_h * bi;
|
|
int hx_shift = li * bi * in_n.at(0) * hy_h;
|
|
|
|
// from input
|
|
if(li == 0)
|
|
{
|
|
if(inputMode == 1)
|
|
{
|
|
for(int bs = 0; bs < batch_n; bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
hid_state.at(hid_shift + bs * hy_stride + h) +=
|
|
in_state.at(bs * in_stride + h);
|
|
if(bidirection)
|
|
{
|
|
hid_state.at(hid_shift + bs * hy_stride + hy_h + h) +=
|
|
in_state.at(bs * in_stride + h);
|
|
}
|
|
}
|
|
}
|
|
|
|
// from bias
|
|
if(biased)
|
|
{
|
|
for(int bs = 0; bs < batch_n; bs++)
|
|
{
|
|
for(int h = 0; h < hy_stride; h++)
|
|
{
|
|
hid_state.at(hid_shift + bs * hy_stride + h) +=
|
|
wei.at(wei_shift_bias + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
gemm_cpu<Tref>(in_state.data(),
|
|
in_h,
|
|
batch_n,
|
|
in_stride,
|
|
false,
|
|
wei_state.data(),
|
|
in_h,
|
|
hy_h * bi,
|
|
in_stride,
|
|
true,
|
|
&hid_state[hid_shift],
|
|
hy_h * bi,
|
|
batch_n,
|
|
hy_stride,
|
|
1,
|
|
1);
|
|
|
|
// from bias
|
|
if(biased)
|
|
{
|
|
for(int bs = 0; bs < batch_n; bs++)
|
|
{
|
|
for(int h = 0; h < hy_stride; h++)
|
|
{
|
|
hid_state.at(hid_shift + bs * hy_stride + h) +=
|
|
wei.at(wei_shift_bias + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
int wei_shift = bi * (in_h + hy_h) * hy_h + (li - 1) * bi * (bi * hy_h + hy_h) * hy_h;
|
|
int prelayer_shift = (li - 1) * batch_n * hy_h * bi;
|
|
if(use_dropout)
|
|
{
|
|
auto dropout_states_tmp = dropout_states_host;
|
|
size_t drop_out_offset = (li - 1) * batch_n * hy_h * bi;
|
|
|
|
RunDropoutForwardEmulator<Tref>(handle,
|
|
dropoutDesc,
|
|
dropout_inputTensor,
|
|
dropout_inputTensor,
|
|
wk_state,
|
|
dropout_outputTensor,
|
|
dropout_hid_state,
|
|
dropout_reservespace_host,
|
|
dropout_states_tmp,
|
|
prelayer_shift,
|
|
drop_out_offset,
|
|
drop_out_offset);
|
|
|
|
prelayer_shift = drop_out_offset;
|
|
}
|
|
|
|
gemm_cpu<Tref>(use_dropout ? &dropout_hid_state[prelayer_shift]
|
|
: &wk_state[prelayer_shift],
|
|
hy_h * bi,
|
|
batch_n,
|
|
use_dropout ? hy_h * bi : hy_stride,
|
|
false,
|
|
&wei_state[wei_shift],
|
|
hy_h * bi,
|
|
hy_h * bi,
|
|
bi_stride,
|
|
true,
|
|
&hid_state[hid_shift],
|
|
hy_h * bi,
|
|
batch_n,
|
|
hy_stride,
|
|
1,
|
|
1);
|
|
|
|
// from bias
|
|
if(biased)
|
|
{
|
|
int wei_shift_bias_temp = wei_shift_bias + bi * li * 2 * hy_h;
|
|
|
|
for(int bs = 0; bs < batch_n; bs++)
|
|
{
|
|
for(int h = 0; h < hy_stride; h++)
|
|
{
|
|
hid_state.at(hid_shift + bs * hy_stride + h) +=
|
|
wei.at(wei_shift_bias_temp + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// from hidden state
|
|
bacc = 0;
|
|
baccbi = batch_n;
|
|
for(int ti = 0; ti < seqLength; ti++)
|
|
{
|
|
baccbi -= in_n.at(seqLength - 1 - ti);
|
|
|
|
int wei_shift =
|
|
li == 0 ? (in_h * hy_h * bi)
|
|
: (bi * (in_h + hy_h) * hy_h + (li - 1) * bi * (bi * hy_h + hy_h) * hy_h +
|
|
bi * hy_h * hy_stride);
|
|
|
|
if(ti == 0)
|
|
{
|
|
if(!hx_is_null)
|
|
{
|
|
gemm_cpu<Tref>(&hx_state[hx_shift],
|
|
hy_h,
|
|
in_n[ti],
|
|
uni_stride,
|
|
false,
|
|
&wei_state[wei_shift],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
true,
|
|
&hid_state[hid_shift + bacc * hy_stride],
|
|
hy_h,
|
|
in_n[ti],
|
|
hy_stride,
|
|
1,
|
|
1);
|
|
|
|
// from bias
|
|
if(biased)
|
|
{
|
|
int wei_shift_bias_temp = wei_shift_bias + bi * (li * 2 + 1) * hy_h;
|
|
|
|
for(int bs = 0; bs < in_n[ti]; bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
hid_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h) +=
|
|
wei.at(wei_shift_bias_temp + h);
|
|
}
|
|
}
|
|
}
|
|
|
|
if(bidirection)
|
|
{
|
|
gemm_cpu<Tref>(&hx_state[hx_shift + hy_n * hy_h],
|
|
hy_h,
|
|
in_n[seqLength - 1 - ti],
|
|
uni_stride,
|
|
false,
|
|
&wei_state[wei_shift + hy_h * uni_stride],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
true,
|
|
&hid_state[hid_shift + baccbi * hy_stride + hy_h],
|
|
hy_h,
|
|
in_n[seqLength - 1 - ti],
|
|
hy_stride,
|
|
1,
|
|
1);
|
|
|
|
// from bias
|
|
if(biased)
|
|
{
|
|
int wei_shift_bias_temp = wei_shift_bias + bi * (li * 2 + 1) * hy_h;
|
|
|
|
for(int bs = 0; bs < in_n.at(seqLength - 1 - ti); bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
hid_state.at(hid_shift + baccbi * hy_stride + hy_h +
|
|
bs * hy_stride + h) +=
|
|
wei.at(wei_shift_bias_temp + hy_h + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
gemm_cpu<Tref>(&hy_state[hx_shift],
|
|
hy_h,
|
|
in_n[ti],
|
|
uni_stride,
|
|
false,
|
|
&wei_state[wei_shift],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
true,
|
|
&hid_state[hid_shift + bacc * hy_stride],
|
|
hy_h,
|
|
in_n[ti],
|
|
hy_stride,
|
|
1,
|
|
1);
|
|
|
|
// from bias
|
|
if(biased)
|
|
{
|
|
int wei_shift_bias_temp = wei_shift_bias + bi * (li * 2 + 1) * hy_h;
|
|
|
|
for(int bs = 0; bs < in_n[ti]; bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
hid_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h) +=
|
|
wei.at(wei_shift_bias_temp + h);
|
|
}
|
|
}
|
|
}
|
|
|
|
if(bidirection)
|
|
{
|
|
|
|
if(!hx_is_null && in_n.at(seqLength - 1 - ti) > in_n.at(seqLength - ti))
|
|
{
|
|
gemm_cpu<Tref>(
|
|
&hx_state[hx_shift + hy_n * hy_h + in_n.at(seqLength - ti) * hy_h],
|
|
hy_h,
|
|
(in_n.at(seqLength - 1 - ti) - in_n.at(seqLength - ti)),
|
|
uni_stride,
|
|
false,
|
|
&wei_state[wei_shift + hy_h * uni_stride],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
true,
|
|
&hid_state[hid_shift + (baccbi + in_n.at(seqLength - ti)) * hy_stride +
|
|
hy_h],
|
|
hy_h,
|
|
(in_n.at(seqLength - 1 - ti) - in_n.at(seqLength - ti)),
|
|
hy_stride,
|
|
1,
|
|
1);
|
|
|
|
// from bias
|
|
if(biased)
|
|
{
|
|
int wei_shift_bias_temp = wei_shift_bias + bi * (li * 2 + 1) * hy_h;
|
|
|
|
for(int bs = in_n.at(seqLength - ti); bs < in_n.at(seqLength - 1 - ti);
|
|
bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
hid_state.at(hid_shift + baccbi * hy_stride + hy_h +
|
|
bs * hy_stride + h) +=
|
|
wei.at(wei_shift_bias_temp + hy_h + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
gemm_cpu<Tref>(&hy_state[hx_shift + hy_n * hy_h],
|
|
hy_h,
|
|
in_n[seqLength - ti],
|
|
uni_stride,
|
|
false,
|
|
&wei_state[wei_shift + hy_h * uni_stride],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
true,
|
|
&hid_state[hid_shift + baccbi * hy_stride + hy_h],
|
|
hy_h,
|
|
in_n[seqLength - ti],
|
|
hy_stride,
|
|
1,
|
|
1);
|
|
|
|
// from bias
|
|
if(biased)
|
|
{
|
|
int wei_shift_bias_temp = wei_shift_bias + bi * (li * 2 + 1) * hy_h;
|
|
|
|
for(int bs = 0; bs < in_n.at(seqLength - ti); bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
hid_state.at(hid_shift + baccbi * hy_stride + hy_h +
|
|
bs * hy_stride + h) +=
|
|
wei.at(wei_shift_bias_temp + hy_h + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
for(int bs = 0; bs < in_n[ti]; bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
wk_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h) =
|
|
activfunc(hid_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h),
|
|
squash); // squash_func
|
|
hy_state.at(hx_shift + bs * uni_stride + h) =
|
|
wk_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h);
|
|
|
|
rsvspace_host.at(hid_shift + bacc * hy_stride + bs * hy_stride + h) =
|
|
hid_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h);
|
|
|
|
rsvspace_host.at(hid_shift + bacc * hy_stride + bs * hy_stride + h +
|
|
numlayer * batch_n * hy_h * bi) =
|
|
activfunc(hid_state[hid_shift + bacc * hy_stride + bs * hy_stride + h],
|
|
squash);
|
|
|
|
hy_host.at(hx_shift + bs * uni_stride + h) =
|
|
hy_state.at(hx_shift + bs * uni_stride + h);
|
|
}
|
|
}
|
|
|
|
if(bidirection)
|
|
{
|
|
for(int bs = 0; bs < in_n.at(seqLength - 1 - ti); bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
wk_state.at(hid_shift + baccbi * hy_stride + hy_h + bs * hy_stride + h) =
|
|
activfunc(hid_state[hid_shift + baccbi * hy_stride + hy_h +
|
|
bs * hy_stride + h],
|
|
squash); // squash_func
|
|
|
|
hy_state.at(hx_shift + hy_n * hy_h + bs * uni_stride + h) =
|
|
wk_state.at(hid_shift + baccbi * hy_stride + hy_h + bs * hy_stride + h);
|
|
|
|
rsvspace_host.at(hid_shift + baccbi * hy_stride + hy_h + bs * hy_stride +
|
|
h) = hid_state.at(hid_shift + baccbi * hy_stride + hy_h +
|
|
bs * hy_stride + h);
|
|
|
|
rsvspace_host.at(hid_shift + baccbi * hy_stride + hy_h + bs * hy_stride +
|
|
h + numlayer * batch_n * hy_h * bi) =
|
|
activfunc(hid_state[hid_shift + baccbi * hy_stride + hy_h +
|
|
bs * hy_stride + h],
|
|
squash);
|
|
|
|
hy_host.at(hx_shift + hy_n * hy_h + bs * uni_stride + h) =
|
|
hy_state.at(hx_shift + hy_n * hy_h + bs * uni_stride + h);
|
|
}
|
|
}
|
|
}
|
|
|
|
bacc += in_n.at(ti);
|
|
}
|
|
}
|
|
|
|
// output
|
|
int prelayer_shift = (numlayer - 1) * batch_n * hy_h * bi;
|
|
|
|
if(use_dropout)
|
|
{
|
|
for(int i = 0; i < (numlayer - 1) * batch_n * hy_h * bi; i++)
|
|
{
|
|
rsvspace_host.at(numlayer * batch_n * hy_stride * 2 + i) = dropout_hid_state.at(i);
|
|
}
|
|
auto p_drop_rsv =
|
|
reinterpret_cast<unsigned char*>(&rsvspace_host[numlayer * batch_n * hy_stride * 2 +
|
|
(numlayer - 1) * batch_n * hy_h * bi]);
|
|
for(int i = 0; i < (numlayer - 1) * batch_n * hy_h * bi; i++)
|
|
{
|
|
*(p_drop_rsv + i) = dropout_reservespace_host.at(i);
|
|
}
|
|
}
|
|
for(int bs = 0; bs < batch_n; bs++)
|
|
{
|
|
for(int h = 0; h < out_h; h++)
|
|
{
|
|
assert(!std::isnan(wk_state.at(prelayer_shift + bs * hy_stride + h)));
|
|
out_host.at(bs * out_stride + h) = wk_state.at(prelayer_shift + bs * hy_stride + h);
|
|
// printf("out_host[%d]: %f\n", bs * out_stride + h, out_host.at(bs * out_stride + h));
|
|
}
|
|
}
|
|
}
|
|
|
|
template <typename Tgpu, typename Tref>
|
|
void RunRNNBackwardDataGEMMCPUVerify(std::vector<Tref>& din_host,
|
|
std::vector<Tgpu>& wei, // [ input_state_weight_trans
|
|
// hidden_state_weight0_trans input1_trans
|
|
// hidden1_trans ... output_weight;
|
|
// bidirectional reversed weights ]
|
|
std::vector<Tgpu>& dhy, // current/final hidden state
|
|
std::vector<Tref>& dhx_host,
|
|
std::vector<Tgpu>& hx, // initial hidden state
|
|
std::vector<Tgpu>& dout,
|
|
std::vector<int>& in_n, // input batch size
|
|
int in_h, // input data length
|
|
int seqLength, // Number of iterations to unroll over
|
|
bool bidirection, // whether using bidirectional net
|
|
bool biased, // whether using bias
|
|
int hy_d, // 1 by numlayer (number of stacks of hidden layers)
|
|
// for unidirection, 2 by numlayer for bidirection
|
|
int hy_n, // equal to input batch size in_n[0]
|
|
int hy_h, // hidden state number
|
|
int out_h, // 1 by hy_h related function for unidirection, 2 by
|
|
// hy_h related function for bidirection
|
|
int squash,
|
|
int inputMode,
|
|
std::vector<Tref>& rsvspace_host,
|
|
std::vector<Tref>& wkspace_host,
|
|
bool use_dropout,
|
|
miopenDropoutDescriptor_t dropoutDesc,
|
|
bool dhy_is_null = false)
|
|
{
|
|
/*
|
|
printf("BWD DATA CPU driver:\n");
|
|
printf("seqLen: %d, in_h: %d, hy_d: %d, hy_n: %d, hy_h: %d, out_h: %d\n", seqLength, in_h,
|
|
hy_d, hy_n, hy_h, out_h);
|
|
printf("hx size: %d, dhx size: %d, dhy size: %d, reserveSpace: %d, workSpace: %d\n",
|
|
hx.size(), dhx_host.size(), dhy.size(), rsvspace_host.size(),wkspace_host.size());
|
|
printf("dinput size: %d\n", din_host.size());
|
|
*/
|
|
int batch_n = sumvc(in_n);
|
|
std::vector<Tref> dh_state(hy_d * batch_n * hy_h, static_cast<Tref>(0));
|
|
|
|
std::vector<Tref> din_state(batch_n * in_h, static_cast<Tref>(0));
|
|
|
|
int numlayer = bidirection ? hy_d / 2 : hy_d;
|
|
int bacc, baccbi; // accumulation of batch
|
|
int bi = bidirection ? 2 : 1;
|
|
|
|
int in_stride = in_h;
|
|
int hy_stride = hy_h * bi;
|
|
int out_stride = out_h;
|
|
int uni_stride = hy_h;
|
|
int bi_stride = hy_h * bi;
|
|
|
|
(void)hx;
|
|
|
|
// initial dout
|
|
std::vector<Tref> dout_state(batch_n * out_h, static_cast<Tref>(0));
|
|
for(int h = 0; h < batch_n; h++)
|
|
{
|
|
for(int w = 0; w < out_h; w++)
|
|
{
|
|
dout_state.at(h * out_h + w) = dout.at(h * out_h + w);
|
|
}
|
|
}
|
|
|
|
// initial hidden states
|
|
std::vector<Tref> dhx_state(hy_d * hy_n * hy_h, static_cast<Tref>(0));
|
|
std::vector<Tref> dhy_state(hy_d * hy_n * hy_h, static_cast<Tref>(0));
|
|
for(int h = 0; h < hy_d * hy_n * hy_h; h++)
|
|
{
|
|
dhy_state.at(h) = dhy.at(h);
|
|
}
|
|
|
|
if(inputMode == 1)
|
|
{
|
|
if(in_h != hy_h)
|
|
{
|
|
printf("Verification cannot be completed: The input tensor size must equal to the "
|
|
"hidden state size of the network in SKIP_INPUT mode!\n");
|
|
return;
|
|
}
|
|
in_h = 0;
|
|
}
|
|
|
|
// initial weights
|
|
int wei_len = (bi * (in_h + hy_h) + (numlayer - 1) * bi * (bi + 1) * hy_h) * hy_h;
|
|
if(biased)
|
|
{
|
|
int in_bias = 2;
|
|
wei_len += (bi * in_bias + (numlayer - 1) * bi * 2) * hy_h;
|
|
}
|
|
|
|
std::vector<Tref> wei_state(wei_len, static_cast<Tref>(0));
|
|
for(int h = 0; h < wei_len; h++)
|
|
{
|
|
wei_state.at(h) = wei.at(h);
|
|
}
|
|
|
|
// initial dropoput
|
|
miopenTensorDescriptor_t dropout_inputTensor{};
|
|
std::vector<unsigned char> dropout_reservespace_host;
|
|
if(use_dropout)
|
|
{
|
|
std::array<int, 2> drop_in_len = {{batch_n, hy_h * bi}};
|
|
std::array<int, 2> drop_in_str = {{hy_stride, 1}};
|
|
miopenCreateTensorDescriptor(&dropout_inputTensor);
|
|
miopenSetTensorDescriptor(
|
|
dropout_inputTensor, miopenFloat, 2, drop_in_len.data(), drop_in_str.data());
|
|
|
|
size_t reserveSpaceSizeInBytes = 0;
|
|
miopenDropoutGetReserveSpaceSize(dropout_inputTensor, &reserveSpaceSizeInBytes);
|
|
size_t reserve_size = reserveSpaceSizeInBytes / sizeof(unsigned char);
|
|
dropout_reservespace_host = std::vector<unsigned char>(reserve_size * (numlayer - 1),
|
|
static_cast<unsigned char>(0));
|
|
|
|
auto p_drop_rsv =
|
|
reinterpret_cast<unsigned char*>(&rsvspace_host[numlayer * batch_n * hy_stride * 2 +
|
|
(numlayer - 1) * batch_n * hy_h * bi]);
|
|
for(int i = 0; i < (numlayer - 1) * batch_n * hy_h * bi; i++)
|
|
{
|
|
dropout_reservespace_host.at(i) = *(p_drop_rsv + i);
|
|
}
|
|
}
|
|
|
|
// bwd data emulator
|
|
for(int li = numlayer - 1; li >= 0; li--)
|
|
{
|
|
int wei_shift = bi * (in_h + hy_h) * hy_h + li * bi * (bi * hy_h + hy_h) * hy_h;
|
|
int hid_shift = li * batch_n * hy_h * bi;
|
|
int hx_shift = li * bi * in_n.at(0) * hy_h;
|
|
|
|
if(li == numlayer - 1)
|
|
{
|
|
for(int bs = 0; bs < batch_n; bs++)
|
|
{
|
|
for(int h = 0; h < out_h; h++)
|
|
{
|
|
dh_state.at(hid_shift + bs * hy_stride + h) +=
|
|
dout_state.at(bs * out_stride + h);
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
int prelayer_shift = (li + 1) * batch_n * hy_h * bi;
|
|
|
|
gemm_cpu<Tref>(&dh_state[prelayer_shift],
|
|
hy_h * bi,
|
|
batch_n,
|
|
hy_stride,
|
|
false,
|
|
&wei_state[wei_shift],
|
|
hy_h * bi,
|
|
hy_h * bi,
|
|
bi_stride,
|
|
false,
|
|
&dh_state[hid_shift],
|
|
hy_h * bi,
|
|
batch_n,
|
|
hy_stride,
|
|
1,
|
|
1);
|
|
|
|
if(use_dropout)
|
|
{
|
|
RunDropoutBackwardEmulator<Tref>(dropoutDesc,
|
|
dropout_inputTensor,
|
|
dh_state,
|
|
dropout_inputTensor,
|
|
dh_state,
|
|
dropout_reservespace_host,
|
|
hid_shift,
|
|
hid_shift,
|
|
li * batch_n * hy_h * bi);
|
|
}
|
|
}
|
|
|
|
bacc = batch_n;
|
|
baccbi = 0;
|
|
for(int ti = seqLength - 1; ti >= 0; ti--)
|
|
{
|
|
bacc -= in_n.at(ti);
|
|
|
|
// from post state
|
|
if(ti == seqLength - 1)
|
|
{
|
|
if(!dhy_is_null)
|
|
{
|
|
for(int bs = 0; bs < in_n.at(ti); bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
dh_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h) +=
|
|
dhy_state.at(hx_shift + bs * uni_stride + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
if(!dhy_is_null && in_n.at(ti) > in_n.at(ti + 1))
|
|
{
|
|
for(int bs = in_n.at(ti + 1); bs < in_n.at(ti); bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
dh_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h) +=
|
|
dhy_state.at(hx_shift + bs * uni_stride + h);
|
|
}
|
|
}
|
|
}
|
|
|
|
for(int bs = 0; bs < in_n.at(ti + 1); bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
dh_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h) +=
|
|
dhx_state.at(hx_shift + bs * uni_stride + h);
|
|
}
|
|
}
|
|
}
|
|
|
|
for(int bs = 0; bs < in_n.at(ti); bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
dh_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h) *= dervactivfunc(
|
|
rsvspace_host.at(hid_shift + bacc * hy_stride + bs * hy_stride + h),
|
|
squash);
|
|
wkspace_host.at(hid_shift + bacc * hy_stride + bs * hy_stride + h) =
|
|
dh_state.at(hid_shift + bacc * hy_stride + bs * hy_stride + h);
|
|
}
|
|
}
|
|
|
|
if(ti < seqLength - 1)
|
|
{
|
|
for(int bs = 0; bs < in_n.at(ti + 1); bs++)
|
|
{
|
|
memset(&dhx_state[hx_shift + bs * uni_stride], 0, hy_h * sizeof(Tref));
|
|
}
|
|
}
|
|
|
|
wei_shift = li == 0
|
|
? (in_h * hy_stride)
|
|
: (bi * (in_h + hy_h) * hy_h +
|
|
(li - 1) * bi * (bi * hy_h + hy_h) * hy_h + bi * hy_h * hy_stride);
|
|
|
|
gemm_cpu<Tref>(&dh_state[hid_shift + bacc * hy_stride],
|
|
hy_h,
|
|
in_n.at(ti),
|
|
hy_stride,
|
|
false,
|
|
&wei_state[wei_shift],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
false,
|
|
&dhx_state[hx_shift],
|
|
hy_h,
|
|
in_n.at(ti),
|
|
uni_stride,
|
|
1,
|
|
1);
|
|
|
|
if(bidirection)
|
|
{
|
|
for(int bs = 0; bs < in_n.at(seqLength - 1 - ti); bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
// from post state
|
|
if(ti == seqLength - 1)
|
|
{
|
|
if(!dhy_is_null)
|
|
{
|
|
dh_state.at(hid_shift + baccbi * hy_stride + hy_h + bs * hy_stride +
|
|
h) +=
|
|
dhy_state.at(hx_shift + hy_n * hy_h + bs * uni_stride + h);
|
|
}
|
|
}
|
|
else
|
|
{
|
|
dh_state.at(hid_shift + baccbi * hy_stride + hy_h + bs * hy_stride +
|
|
h) +=
|
|
dhx_state.at(hx_shift + hy_n * hy_h + bs * uni_stride + h);
|
|
}
|
|
|
|
dh_state.at(hid_shift + baccbi * hy_stride + hy_h + bs * hy_stride + h) *=
|
|
dervactivfunc(rsvspace_host.at(hid_shift + baccbi * hy_stride + hy_h +
|
|
bs * hy_stride + h),
|
|
squash);
|
|
wkspace_host.at(hid_shift + baccbi * hy_stride + hy_h + bs * hy_stride +
|
|
h) =
|
|
dh_state.at(hid_shift + baccbi * hy_stride + hy_h + bs * hy_stride + h);
|
|
}
|
|
}
|
|
|
|
if(ti < seqLength - 1)
|
|
{
|
|
for(int bs = 0; bs < in_n.at(seqLength - 1 - ti); bs++)
|
|
{
|
|
memset(&dhx_state[hx_shift + bs * uni_stride + hy_n * hy_h],
|
|
0,
|
|
hy_h * sizeof(Tref));
|
|
}
|
|
}
|
|
|
|
gemm_cpu<Tref>(&dh_state[hid_shift + baccbi * hy_stride + hy_h],
|
|
hy_h,
|
|
in_n.at(seqLength - 1 - ti),
|
|
hy_stride,
|
|
false,
|
|
&wei_state[wei_shift + hy_h * uni_stride],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
false,
|
|
&dhx_state[hx_shift + hy_n * hy_h],
|
|
hy_h,
|
|
in_n.at(seqLength - 1 - ti),
|
|
uni_stride,
|
|
1,
|
|
1);
|
|
}
|
|
|
|
baccbi += in_n.at(seqLength - 1 - ti);
|
|
}
|
|
}
|
|
|
|
// dinput
|
|
if(inputMode == 1)
|
|
{
|
|
for(int bs = 0; bs < batch_n; bs++)
|
|
{
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
din_state.at(bs * in_stride + h) += dh_state.at(bs * hy_stride + h);
|
|
if(bidirection)
|
|
{
|
|
din_state.at(bs * in_stride + h) += dh_state.at(bs * hy_stride + hy_h + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
gemm_cpu<Tref>(dh_state.data(),
|
|
hy_h * bi,
|
|
batch_n,
|
|
hy_stride,
|
|
false,
|
|
wei_state.data(),
|
|
in_h,
|
|
hy_h * bi,
|
|
in_stride,
|
|
false,
|
|
din_state.data(),
|
|
in_h,
|
|
batch_n,
|
|
in_stride,
|
|
1,
|
|
1);
|
|
}
|
|
|
|
for(int bs = 0; bs < batch_n; bs++)
|
|
{
|
|
for(int w = 0; w < in_stride; w++)
|
|
{
|
|
din_host.at(bs * in_stride + w) = din_state.at(bs * in_stride + w);
|
|
}
|
|
}
|
|
|
|
for(int h = 0; h < hy_d * hy_n * hy_h; h++)
|
|
{
|
|
dhx_host.at(h) = dhx_state.at(h);
|
|
// printf("dhx_host[%d]: %f\n", h, dhx_host.at(h));
|
|
}
|
|
}
|
|
|
|
template <typename Tgpu, typename Tref>
|
|
void RunRNNBackwardWeightGEMMCPUVerify(std::vector<Tgpu>& in,
|
|
std::vector<Tref>& dwei_host, // [ input_state_weight_trans
|
|
// hidden_state_weight0_trans
|
|
// input1_trans hidden1_trans ...
|
|
// output_weight; bidirectional
|
|
// reversed weights ]
|
|
std::vector<Tgpu>& hx, // initial hidden state
|
|
std::vector<Tgpu>& dout,
|
|
std::vector<int>& in_n, // input batch size
|
|
int in_h, // input data length
|
|
int seqLength, // Number of iterations to unroll over
|
|
bool bidirection, // whether using bidirectional net
|
|
bool biased, // whether using bias
|
|
int hy_d, // 1 by numlayer (number of stacks of hidden
|
|
// layers) for unidirection, 2 by numlayer for
|
|
// bidirection
|
|
int hy_n, // equal to input batch size in_n[0]
|
|
int hy_h, // hidden state number
|
|
int out_h, // 1 by hy_h related function for unidirection, 2
|
|
// by hy_h related function for bidirection
|
|
int squash,
|
|
int inputMode,
|
|
std::vector<Tref>& rsvspace_host,
|
|
std::vector<Tref>& wkspace_host,
|
|
bool use_dropout,
|
|
bool hx_is_null = false)
|
|
{
|
|
|
|
// printf("BWD WEGIHTS CPU driver:\n");
|
|
// printf("seqLen: %d, in_h: %d, hy_d: %d, hy_n: %d, hy_h: %d, out_h: %d\n", seqLength, in_h,
|
|
// hy_d, hy_n, hy_h, out_h);
|
|
// printf("dirmode: %d, hx size: %d, dout size: %d, reserveSpace: %d, workSpace: %d\n",
|
|
// bidirection ? 2 : 1, hx.size(), dout.size(), rsvspace_host.size(),wkspace_host.size());
|
|
// printf("input size: %d\n", in.size());
|
|
int batch_n = sumvc(in_n);
|
|
int numlayer = bidirection ? hy_d / 2 : hy_d;
|
|
int bacc; // accumulation of batch
|
|
int bi = bidirection ? 2 : 1;
|
|
|
|
int in_stride = in_h;
|
|
int hy_stride = hy_h * bi;
|
|
int uni_stride = hy_h;
|
|
int bi_stride = hy_h * bi;
|
|
|
|
(void)hy_n;
|
|
|
|
// initial input
|
|
std::vector<Tref> in_state(batch_n * in_h, static_cast<Tref>(0));
|
|
for(int h = 0; h < batch_n; h++)
|
|
{
|
|
for(int w = 0; w < in_h; w++)
|
|
{
|
|
in_state.at(h * in_h + w) = in.at(h * in_h + w);
|
|
}
|
|
}
|
|
|
|
// initial output difference
|
|
std::vector<Tref> dout_state(batch_n * out_h, static_cast<Tref>(0));
|
|
for(int h = 0; h < batch_n; h++)
|
|
{
|
|
for(int w = 0; w < out_h; w++)
|
|
{
|
|
dout_state.at(h * out_h + w) = dout.at(h * out_h + w);
|
|
}
|
|
}
|
|
|
|
// initial saved data
|
|
std::vector<Tref> wkspace_state(hy_d * batch_n * hy_h, static_cast<Tref>(0));
|
|
std::vector<Tref> rsvspace_state(rsvspace_host.size(), static_cast<Tref>(0));
|
|
for(int h = 0; h < hy_d * batch_n * hy_h; h++)
|
|
{
|
|
rsvspace_state.at(h) = activfunc(rsvspace_host.at(h), squash);
|
|
wkspace_state.at(h) = wkspace_host.at(h);
|
|
}
|
|
if(use_dropout)
|
|
{
|
|
for(int h = hy_d * batch_n * hy_h; h < rsvspace_host.size(); h++)
|
|
{
|
|
rsvspace_state.at(h) = rsvspace_host.at(h);
|
|
}
|
|
}
|
|
|
|
// initial hidden states
|
|
std::vector<Tref> hx_state(hy_d * hy_n * hy_h, static_cast<Tref>(0));
|
|
for(int h = 0; h < hy_d * hy_n * hy_h; h++)
|
|
{
|
|
hx_state.at(h) = hx.at(h);
|
|
}
|
|
|
|
if(inputMode == 1)
|
|
{
|
|
if(in_h != hy_h)
|
|
{
|
|
printf("Verification cannot be completed: The input tensor size must equal to the "
|
|
"hidden state size of the network in SKIP_INPUT mode!\n");
|
|
return;
|
|
}
|
|
in_h = 0;
|
|
}
|
|
|
|
int wei_len = (bi * (in_h + hy_h) + (numlayer - 1) * bi * (bi + 1) * hy_h) * hy_h;
|
|
int wei_shift_bias = wei_len;
|
|
if(biased)
|
|
{
|
|
int in_bias = 2;
|
|
wei_len += (bi * in_bias + (numlayer - 1) * bi * 2) * hy_h;
|
|
}
|
|
|
|
// initial dwei
|
|
std::vector<Tref> dwei_state(wei_len, static_cast<Tref>(0));
|
|
|
|
// bwd weights emulator
|
|
for(int li = 0; li < numlayer; li++)
|
|
{
|
|
// between layers
|
|
if(li == 0)
|
|
{
|
|
if(inputMode != 1)
|
|
{
|
|
gemm_cpu<Tref>(wkspace_state.data(),
|
|
hy_h * bi,
|
|
batch_n,
|
|
hy_stride,
|
|
true,
|
|
in_state.data(),
|
|
in_h,
|
|
batch_n,
|
|
in_stride,
|
|
false,
|
|
dwei_state.data(),
|
|
in_h,
|
|
hy_h * bi,
|
|
in_stride,
|
|
1,
|
|
1);
|
|
}
|
|
if(biased)
|
|
{
|
|
for(int h = 0; h < hy_stride; h++)
|
|
{
|
|
for(int w = 0; w < batch_n; w++)
|
|
{
|
|
dwei_state.at(wei_shift_bias + h) += wkspace_host.at(w * hy_stride + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
int prelayer_shift = (use_dropout ? 2 * numlayer * batch_n * hy_stride : 0) +
|
|
(li - 1) * batch_n * hy_h * bi;
|
|
int hid_shift = li * bi * batch_n * hy_h;
|
|
int wei_shift = bi * (in_h + hy_h) * hy_h + (li - 1) * bi * (bi * hy_h + hy_h) * hy_h;
|
|
|
|
gemm_cpu<Tref>(&wkspace_state[hid_shift],
|
|
hy_h * bi,
|
|
batch_n,
|
|
hy_stride,
|
|
true,
|
|
&rsvspace_state[prelayer_shift],
|
|
hy_h * bi,
|
|
batch_n,
|
|
hy_stride,
|
|
false,
|
|
&dwei_state[wei_shift],
|
|
hy_h * bi,
|
|
hy_h * bi,
|
|
bi_stride,
|
|
1,
|
|
1);
|
|
|
|
if(biased)
|
|
{
|
|
wei_shift = wei_shift_bias + li * bi * 2 * hy_h;
|
|
|
|
for(int h = 0; h < hy_stride; h++)
|
|
{
|
|
for(int w = 0; w < batch_n; w++)
|
|
{
|
|
dwei_state.at(wei_shift + h) +=
|
|
wkspace_host.at(hid_shift + w * hy_stride + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
bacc = 0;
|
|
for(int ti = 0; ti < seqLength; ti++)
|
|
{
|
|
int hid_shift = li * bi * batch_n * hy_h + bacc * hy_stride;
|
|
int hx_shift = li * bi * in_n.at(0) * hy_h;
|
|
int wei_shift;
|
|
int pretime_shift;
|
|
|
|
wei_shift = li == 0
|
|
? (in_h * hy_stride)
|
|
: (bi * (in_h + hy_h) * hy_h +
|
|
(li - 1) * bi * (bi * hy_h + hy_h) * hy_h + bi * hy_h * hy_stride);
|
|
|
|
// between time
|
|
if(ti == 0)
|
|
{
|
|
if(!hx_is_null)
|
|
{
|
|
gemm_cpu<Tref>(&wkspace_state[hid_shift],
|
|
hy_h,
|
|
in_n.at(ti),
|
|
hy_stride,
|
|
true,
|
|
&hx_state[hx_shift],
|
|
hy_h,
|
|
in_n.at(ti),
|
|
uni_stride,
|
|
false,
|
|
&dwei_state[wei_shift],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
1,
|
|
1);
|
|
|
|
if(biased)
|
|
{
|
|
int bias_shift = wei_shift_bias + li * bi * 2 * hy_h + bi * hy_h;
|
|
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
for(int w = 0; w < in_n.at(ti); w++)
|
|
{
|
|
dwei_state.at(bias_shift + h) +=
|
|
wkspace_host.at(hid_shift + w * hy_stride + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
pretime_shift = li * bi * batch_n * hy_h + (bacc - in_n.at(ti - 1)) * hy_stride;
|
|
|
|
gemm_cpu<Tref>(&wkspace_state[hid_shift],
|
|
hy_h,
|
|
in_n.at(ti),
|
|
hy_stride,
|
|
true,
|
|
&rsvspace_state[pretime_shift],
|
|
hy_h,
|
|
in_n.at(ti),
|
|
hy_stride,
|
|
false,
|
|
&dwei_state[wei_shift],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
1,
|
|
1);
|
|
|
|
if(biased)
|
|
{
|
|
int bias_shift = wei_shift_bias + li * bi * 2 * hy_h + bi * hy_h;
|
|
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
for(int w = 0; w < in_n.at(ti); w++)
|
|
{
|
|
dwei_state.at(bias_shift + h) +=
|
|
wkspace_host.at(hid_shift + w * hy_stride + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if(bidirection)
|
|
{
|
|
if(ti == seqLength - 1)
|
|
{
|
|
if(!hx_is_null)
|
|
{
|
|
gemm_cpu<Tref>(&wkspace_state[hid_shift + hy_h],
|
|
hy_h,
|
|
in_n.at(ti),
|
|
hy_stride,
|
|
true,
|
|
&hx_state[hx_shift + hy_n * hy_h],
|
|
hy_h,
|
|
in_n.at(ti),
|
|
uni_stride,
|
|
false,
|
|
&dwei_state[wei_shift + hy_h * uni_stride],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
1,
|
|
1);
|
|
|
|
if(biased)
|
|
{
|
|
int bias_shift = wei_shift_bias + li * bi * 2 * hy_h + bi * hy_h;
|
|
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
for(int w = 0; w < in_n.at(ti); w++)
|
|
{
|
|
dwei_state.at(bias_shift + hy_h + h) +=
|
|
wkspace_host.at(hid_shift + w * hy_stride + hy_h + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
if(!hx_is_null && in_n.at(ti) > in_n.at(ti + 1))
|
|
{
|
|
gemm_cpu<Tref>(
|
|
&wkspace_state[hid_shift + hy_h + in_n.at(ti + 1) * hy_stride],
|
|
hy_h,
|
|
(in_n.at(ti) - in_n.at(ti + 1)),
|
|
hy_stride,
|
|
true,
|
|
&hx_state[hx_shift + hy_n * hy_h + in_n.at(ti + 1) * hy_h],
|
|
hy_h,
|
|
(in_n.at(ti) - in_n.at(ti + 1)),
|
|
uni_stride,
|
|
false,
|
|
&dwei_state[wei_shift + hy_h * uni_stride],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
1,
|
|
1);
|
|
|
|
if(biased)
|
|
{
|
|
int bias_shift = wei_shift_bias + li * bi * 2 * hy_h + bi * hy_h;
|
|
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
for(int w = in_n.at(ti + 1); w < in_n.at(ti); w++)
|
|
{
|
|
dwei_state.at(bias_shift + hy_h + h) +=
|
|
wkspace_host.at(hid_shift + w * hy_stride + hy_h + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
pretime_shift = li * bi * batch_n * hy_h + (bacc + in_n.at(ti)) * hy_stride;
|
|
|
|
gemm_cpu<Tref>(const_cast<Tref*>(&wkspace_state[hid_shift + hy_h]),
|
|
hy_h,
|
|
in_n.at(ti + 1),
|
|
hy_stride,
|
|
true,
|
|
&rsvspace_state[pretime_shift + hy_h],
|
|
hy_h,
|
|
in_n.at(ti + 1),
|
|
hy_stride,
|
|
false,
|
|
&dwei_state[wei_shift + hy_h * uni_stride],
|
|
hy_h,
|
|
hy_h,
|
|
uni_stride,
|
|
1,
|
|
1);
|
|
|
|
if(biased)
|
|
{
|
|
int bias_shift = wei_shift_bias + li * bi * 2 * hy_h + bi * hy_h;
|
|
|
|
for(int h = 0; h < hy_h; h++)
|
|
{
|
|
for(int w = 0; w < in_n.at(ti + 1); w++)
|
|
{
|
|
dwei_state.at(bias_shift + hy_h + h) +=
|
|
wkspace_host.at(hid_shift + w * hy_stride + hy_h + h);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
bacc += in_n.at(ti);
|
|
}
|
|
}
|
|
|
|
for(int i = 0; i < wei_len; i++)
|
|
{
|
|
dwei_host.at(i) = dwei_state.at(i);
|
|
}
|
|
}
|
|
|
|
#endif // GUARD_MIOPEN_RNN_VERIFY_GEMM_HPP
|