davrot/pytorch-sbs
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
pytorch-sbs/network/CPP_Cuda_new_preview/HDynamicCNNManyIP.cu

967 lines
38 KiB
Text
Raw Permalink Normal View History

Add files via upload
2023-01-13 21:33:57 +01:00
#include <omp.h>
#include <stdio.h>
#include <string.h>
#include <algorithm>
#include <cassert>
#include <iostream>
#include "HDynamicCNNManyIP.h"
#include "approximation_multiplication_function.h"
#include "kernel_approximation_multiplication.h"
#include "kernel_phxy_fill_with_h.h"
#include "kernel_phxy_fill_with_spike_selected_w.h"
#include "kernel_phxy_one_over_sum_into_pxy.h"
#include "kernel_phxy_plus_phxy.h"
#include "kernel_phxy_plus_pxy.h"
#include "kernel_phxy_times_phxy_equals_phxy.h"
#include "kernel_phxy_times_pxy.h"
#include "kernel_pxy_plus_v.h"
#include "kernel_pxy_reciprocal.h"
#include "kernel_pxy_set_to_v.h"
#include "kernel_pxy_time_pxy.h"
#include "kernel_pxy_times_spike_selected_sxy.h"
#include "kernel_pxy_times_v.h"
HDynamicCNNManyIP::HDynamicCNNManyIP(){
};
HDynamicCNNManyIP::~HDynamicCNNManyIP(){
};
bool HDynamicCNNManyIP::update_entrypoint(
int64_t h_pointer_addr, int64_t h_dim_0, int64_t h_dim_1, int64_t h_dim_2,
int64_t h_dim_3, int64_t epsilon_xy_pointer_addr, int64_t epsilon_xy_dim_0,
int64_t epsilon_xy_dim_1, int64_t epsilon_xy_dim_2,
int64_t epsilon_t_pointer_addr, int64_t epsilon_t_dim_0,
int64_t weights_pointer_addr, int64_t weights_dim_0, int64_t weights_dim_1,
int64_t input_pointer_addr, int64_t input_dim_0, int64_t input_dim_1,
int64_t input_dim_2, int64_t input_dim_3, int64_t init_vector_pointer_addr,
int64_t init_vector_dim_0, int64_t number_of_processes,
float forgetting_offset, int64_t gpu_tuning_factor
// ,bool approximation_multiplication_enable, uint64_t
// number_of_frac_bits, bool approximation_enable,
// uint64_t number_of_trunc_bits
) {
bool approximation_multiplication_enable = false;
uint64_t number_of_frac_bits = 1;
bool approximation_enable = false;
uint64_t number_of_trunc_bits = false;
uint32_t ap_mask = static_cast<uint64_t>(pow(2, number_of_trunc_bits)) - 1;
size_t number_of_pattern = input_dim_0;
size_t h_dim = init_vector_dim_0;
float* h_init_ptr = (float*)init_vector_pointer_addr;
assert((h_init_ptr != nullptr));
assert((h_dim > 0));
float* h_pointer = (float*)h_pointer_addr;
assert((h_pointer != nullptr));
assert((h_dim_0 > 0));
assert((h_dim_1 > 0));
assert((h_dim_2 > 0));
assert((h_dim_3 > 0));
size_t h_dim_c0 = h_dim_1 * h_dim_2 * h_dim_3;
size_t h_dim_c1 = h_dim_2 * h_dim_3;
size_t h_dim_c2 = h_dim_3;
float* epsilon_xy_pointer = (float*)epsilon_xy_pointer_addr;
assert((epsilon_xy_pointer != nullptr));
assert((epsilon_xy_dim_0 > 0));
assert((epsilon_xy_dim_1 > 0));
size_t epsilon_xy_dim_c0 = epsilon_xy_dim_2 * epsilon_xy_dim_1;
size_t epsilon_xy_dim_c1 = epsilon_xy_dim_2;
float* epsilon_t_pointer = (float*)epsilon_t_pointer_addr;
assert((epsilon_t_pointer != nullptr));
assert((epsilon_t_dim_0 > 0));
float* weights_pointer = (float*)weights_pointer_addr;
assert((weights_pointer != nullptr));
assert((weights_dim_0 > 0));
assert((weights_dim_1 > 0));
size_t weights_dim_c0 = weights_dim_1;
int64_t* input_pointer = (int64_t*)input_pointer_addr;
assert((input_pointer != nullptr));
assert((input_dim_0 > 0));
assert((input_dim_1 > 0));
assert((input_dim_2 > 0));
assert((input_dim_3 > 0));
size_t input_dim_c0 = input_dim_1 * input_dim_2 * input_dim_3;
size_t input_dim_c1 = input_dim_2 * input_dim_3;
size_t input_dim_c2 = input_dim_3;
assert((h_dim == weights_dim_1));
size_t number_of_spikes = input_dim_1;
size_t dim_x = input_dim_2;
size_t dim_y = input_dim_3;
float forgetting_offset_local = forgetting_offset / static_cast<float>(h_dim);
// --------------------
if (number_of_processes > 0) {
omp_set_num_threads(number_of_processes);
size_t pattern_id;
#pragma omp parallel for
for (pattern_id = 0; pattern_id < number_of_pattern; pattern_id++) {
update(h_init_ptr, h_pointer, h_dim_c0, h_dim_c1, h_dim_c2, h_dim,
epsilon_xy_pointer, epsilon_xy_dim_c0, epsilon_xy_dim_c1,
epsilon_t_pointer, weights_pointer, weights_dim_c0, input_pointer,
input_dim_c0, input_dim_c1, input_dim_c2, number_of_spikes, dim_x,
dim_y, forgetting_offset, forgetting_offset_local, pattern_id,
approximation_multiplication_enable, number_of_frac_bits,
approximation_enable, number_of_trunc_bits, ap_mask);
}
} else {
gpu_update(h_init_ptr, h_pointer, h_dim_c0, h_dim_c1, h_dim_c2, h_dim,
epsilon_xy_pointer, epsilon_xy_dim_c0, epsilon_xy_dim_c1,
epsilon_t_pointer, weights_pointer, weights_dim_c0,
input_pointer, input_dim_c0, input_dim_c1, input_dim_c2,
number_of_spikes, dim_x, dim_y, forgetting_offset,
forgetting_offset_local, number_of_pattern, gpu_tuning_factor,
approximation_multiplication_enable, number_of_frac_bits,
approximation_enable, number_of_trunc_bits, ap_mask);
}
return true;
};
bool HDynamicCNNManyIP::update(
float* h_init_ptr, float* h_pointer, size_t h_dim_c0, size_t h_dim_c1,
size_t h_dim_c2, size_t h_dim, float* epsilon_xy_pointer,
size_t epsilon_xy_dim_c0, size_t epsilon_xy_dim_c1,
float* epsilon_t_pointer, float* weights_pointer, size_t weights_dim_c0,
int64_t* input_pointer, size_t input_dim_c0, size_t input_dim_c1,
size_t input_dim_c2, size_t number_of_spikes, size_t dim_x, size_t dim_y,
float forgetting_offset, float forgetting_offset_local, size_t pattern_id,
bool approximation_multiplication_enable, uint64_t number_of_frac_bits,
bool approximation_enable, uint64_t number_of_trunc_bits,
uint32_t ap_mask) {
float* h_ptr;
float* epsilon_xy_ptr;
int64_t* input_ptr;
size_t counter_x;
size_t counter_y;
for (counter_x = 0; counter_x < dim_x; counter_x++) {
for (counter_y = 0; counter_y < dim_y; counter_y++) {
epsilon_xy_ptr =
epsilon_xy_pointer + counter_x * epsilon_xy_dim_c1 + counter_y;
h_ptr =
h_pointer + pattern_id * h_dim_c0 + counter_x * h_dim_c2 + counter_y;
input_ptr = input_pointer + pattern_id * input_dim_c0 +
counter_x * input_dim_c2 + counter_y;
if (approximation_multiplication_enable == false) {
update_one_ip(h_init_ptr, h_ptr, h_dim_c1, h_dim, weights_pointer,
weights_dim_c0, input_ptr, input_dim_c1, epsilon_xy_ptr,
epsilon_xy_dim_c0, epsilon_t_pointer, number_of_spikes,
forgetting_offset, forgetting_offset_local);
} else {
update_one_ip_approx(
h_init_ptr, h_ptr, h_dim_c1, h_dim, weights_pointer, weights_dim_c0,
input_ptr, input_dim_c1, epsilon_xy_ptr, epsilon_xy_dim_c0,
epsilon_t_pointer, number_of_spikes, forgetting_offset,
forgetting_offset_local, approximation_multiplication_enable,
number_of_frac_bits, approximation_enable, number_of_trunc_bits,
ap_mask);
}
}
}
return true;
};
void HDynamicCNNManyIP::update_one_ip_approx(
float* h_init_ptr, float* h_pointer, size_t h_dim_c1, size_t h_dim,
float* weights_pointer, size_t weights_dim_c0, int64_t* input_pointer,
size_t input_dim_c1, float* epsilon_xy_pointer, size_t epsilon_xy_dim_c0,
float* epsilon_t_pointer, size_t number_of_spikes, float forgetting_offset,
float forgetting_offset_local, bool approximation_multiplication_enable,
uint64_t number_of_frac_bits, bool approximation_enable,
uint64_t number_of_trunc_bits, uint32_t ap_mask) {
float* h_temp = new float[h_dim];
float* h_subsegment = new float[h_dim];
memcpy(h_subsegment, h_init_ptr, sizeof(float) * h_dim);
size_t counter_spike;
size_t counter;
float h_temp_sum;
float temp_value;
float epsilon_subsegment;
float epsilon_scale = 1.0;
int64_t* spike;
float* w_ptr;
// ---------------
// Approx...
uint64_t pattern_size = h_dim;
std::vector<float> ap_h_vector;
ap_h_vector.resize(pattern_size);
float* ap_h_ptr = ap_h_vector.data();
std::vector<uint32_t> ap_x_vector;
ap_x_vector.resize(pattern_size);
uint32_t* ap_x_ptr = ap_x_vector.data();
std::vector<uint32_t> ap_y_vector;
ap_y_vector.resize(pattern_size);
uint32_t* ap_y_ptr = ap_y_vector.data();
std::vector<uint32_t> ap_x_exponent_vector;
ap_x_exponent_vector.resize(pattern_size);
uint32_t* ap_x_exponent_ptr = ap_x_exponent_vector.data();
std::vector<uint32_t> ap_y_exponent_vector;
ap_y_exponent_vector.resize(pattern_size);
uint32_t* ap_y_exponent_ptr = ap_y_exponent_vector.data();
std::vector<uint32_t> ap_h_exponent_vector;
ap_h_exponent_vector.resize(pattern_size);
uint32_t* ap_h_exponent_ptr = ap_h_exponent_vector.data();
std::vector<uint64_t> ap_res_vector;
ap_res_vector.resize(pattern_size);
uint64_t* ap_res_ptr = ap_res_vector.data();
std::vector<uint32_t> sign_temp_vector;
sign_temp_vector.resize(pattern_size);
uint32_t* sign_temp_ptr = sign_temp_vector.data();
// --------------
for (counter_spike = 0; counter_spike < number_of_spikes; counter_spike++) {
if (epsilon_scale > 1E10) {
temp_value = 1.0 / epsilon_scale;
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_subsegment[counter] *= temp_value;
}
epsilon_scale = 1.0;
}
spike = input_pointer + counter_spike * input_dim_c1;
if (*spike >= 0) {
epsilon_subsegment = epsilon_xy_pointer[*spike * epsilon_xy_dim_c0] *
epsilon_t_pointer[counter_spike];
w_ptr = weights_pointer + *spike * weights_dim_c0;
memcpy(h_temp, h_subsegment, sizeof(float) * h_dim);
approximation_multiplication_function(
ap_h_ptr, w_ptr, pattern_size, number_of_trunc_bits,
number_of_frac_bits, ap_x_ptr, ap_y_ptr, ap_x_exponent_ptr,
ap_y_exponent_ptr, ap_h_exponent_ptr, ap_mask, ap_res_ptr,
sign_temp_ptr, approximation_enable);
// --------------------------
h_temp_sum = 0.0;
#pragma omp simd reduction(+ : h_temp_sum)
for (counter = 0; counter < h_dim; counter++) {
h_temp_sum += h_temp[counter];
}
if (h_temp_sum > 1E-10) {
temp_value = epsilon_scale * epsilon_subsegment / h_temp_sum;
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_temp[counter] *= temp_value;
}
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_subsegment[counter] += h_temp[counter];
}
if (forgetting_offset_local > 0.0) {
temp_value =
epsilon_scale * epsilon_subsegment * forgetting_offset_local;
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_subsegment[counter] += temp_value;
}
epsilon_scale *= 1.0 + epsilon_subsegment * (1.0 + forgetting_offset);
} else {
epsilon_scale *= 1.0 + epsilon_subsegment * 1.0;
}
}
}
}
temp_value = 1.0 / epsilon_scale;
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_pointer[counter * h_dim_c1] = h_subsegment[counter] * temp_value;
}
delete[] h_temp;
delete[] h_subsegment;
return;
};
void HDynamicCNNManyIP::update_one_ip(
float* h_init_ptr, float* h_pointer, size_t h_dim_c1, size_t h_dim,
float* weights_pointer, size_t weights_dim_c0, int64_t* input_pointer,
size_t input_dim_c1, float* epsilon_xy_pointer, size_t epsilon_xy_dim_c0,
float* epsilon_t_pointer, size_t number_of_spikes, float forgetting_offset,
float forgetting_offset_local) {
float* h_temp = new float[h_dim];
float* h_subsegment = new float[h_dim];
memcpy(h_subsegment, h_init_ptr, sizeof(float) * h_dim);
size_t counter_spike;
size_t counter;
float h_temp_sum;
float temp_value;
float epsilon_subsegment;
float epsilon_scale = 1.0;
int64_t* spike;
float* w_ptr;
// --------------
for (counter_spike = 0; counter_spike < number_of_spikes; counter_spike++) {
if (epsilon_scale > 1E10) {
temp_value = 1.0 / epsilon_scale;
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_subsegment[counter] *= temp_value;
}
epsilon_scale = 1.0;
}
spike = input_pointer + counter_spike * input_dim_c1;
if (*spike >= 0) {
epsilon_subsegment = epsilon_xy_pointer[*spike * epsilon_xy_dim_c0] *
epsilon_t_pointer[counter_spike];
w_ptr = weights_pointer + *spike * weights_dim_c0;
memcpy(h_temp, h_subsegment, sizeof(float) * h_dim);
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_temp[counter] *= w_ptr[counter];
}
h_temp_sum = 0.0;
#pragma omp simd reduction(+ : h_temp_sum)
for (counter = 0; counter < h_dim; counter++) {
h_temp_sum += h_temp[counter];
}
if (h_temp_sum > 1E-10) {
temp_value = epsilon_scale * epsilon_subsegment / h_temp_sum;
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_temp[counter] *= temp_value;
}
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_subsegment[counter] += h_temp[counter];
}
if (forgetting_offset_local > 0.0) {
temp_value =
epsilon_scale * epsilon_subsegment * forgetting_offset_local;
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_subsegment[counter] += temp_value;
}
epsilon_scale *= 1.0 + epsilon_subsegment * (1.0 + forgetting_offset);
} else {
epsilon_scale *= 1.0 + epsilon_subsegment * 1.0;
}
}
}
}
temp_value = 1.0 / epsilon_scale;
#pragma omp simd
for (counter = 0; counter < h_dim; counter++) {
h_pointer[counter * h_dim_c1] = h_subsegment[counter] * temp_value;
}
delete[] h_temp;
delete[] h_subsegment;
return;
};
// ------------------------------------------------
void HDynamicCNNManyIP::gpu_occupancy_measure(size_t dim_x, size_t dim_y,
size_t number_of_pattern,
size_t h_dim) {
grid_and_thread_calculated = false;
assert((dim_x < 65535));
assert((dim_y < 65535));
grid_and_thread_settings.resize(14);
occupancy_kernel_phxy_plus_phxy(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PHXY], display_debug);
occupancy_kernel_pxy_plus_v(dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PXY_PLUS_V],
display_debug);
occupancy_kernel_pxy_times_v(dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_V],
display_debug);
occupancy_kernel_phxy_fill_with_h(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_H], display_debug);
occupancy_kernel_phxy_plus_pxy(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PXY], display_debug);
occupancy_kernel_pxy_reciprocal(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PXY_RECIPROCAL], display_debug);
occupancy_kernel_phxy_fill_with_spike_selected_w(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_SPIKE_SELECTED_W],
display_debug);
occupancy_kernel_phxy_times_phxy_equals_phxy(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PHXY_EQUALS_PHXY],
display_debug);
occupancy_kernel_pxy_set_to_v(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V], display_debug);
occupancy_kernel_phxy_one_over_sum_into_pxy(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PHXY_ONE_OVER_SUM_INTO_PXY],
display_debug);
occupancy_kernel_phxy_times_pxy(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY], display_debug);
occupancy_kernel_pxy_time_pxy(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY], display_debug);
occupancy_kernel_approximation_pure_multiplication(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_APPROXIMATION_MULTIPLICATION],
display_debug);
occupancy_kernel_pxy_times_spike_selected_sxy(
dim_x, dim_y, number_of_pattern, h_dim,
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_SPIKE_SELECTED_SXY],
display_debug);
grid_and_thread_calculated = true;
return;
};
void HDynamicCNNManyIP::gpu_occupancy_export(
size_t dim_x, size_t dim_y, size_t number_of_pattern, size_t h_dim,
int64_t setting_memory_addr, size_t setting_dim_0, size_t setting_dim_1) {
int64_t* setting_memory = (int64_t*)setting_memory_addr;
assert((setting_memory != nullptr));
assert((setting_dim_1 == H_DYNAMIC_NUMBER_OF_KERNELS_PARAMETERS));
gpu_occupancy_measure(dim_x, dim_y, number_of_pattern, h_dim);
assert((grid_and_thread_calculated == true));
assert((setting_dim_0 == grid_and_thread_settings.size()));
for (size_t counter_0 = 0; counter_0 < setting_dim_0; counter_0++) {
for (size_t counter_1 = 0; counter_1 < setting_dim_1; counter_1++) {
setting_memory[counter_0 * setting_dim_1 + counter_1] =
grid_and_thread_settings[counter_0][counter_1];
}
}
};
void HDynamicCNNManyIP::gpu_occupancy_import(int64_t setting_memory_addr,
size_t setting_dim_0,
size_t setting_dim_1) {
grid_and_thread_calculated = false;
int64_t* setting_memory = (int64_t*)setting_memory_addr;
assert((setting_memory != nullptr));
assert((setting_dim_1 == H_DYNAMIC_NUMBER_OF_KERNELS_PARAMETERS));
assert((setting_dim_0 == H_DYNAMIC_NUMBER_OF_KERNELS));
grid_and_thread_settings.resize(H_DYNAMIC_NUMBER_OF_KERNELS);
for (size_t counter_0 = 0; counter_0 < setting_dim_0; counter_0++) {
grid_and_thread_settings[counter_0].resize(
H_DYNAMIC_NUMBER_OF_KERNELS_PARAMETERS);
for (size_t counter_1 = 0; counter_1 < setting_dim_1; counter_1++) {
grid_and_thread_settings[counter_0][counter_1] =
setting_memory[counter_0 * setting_dim_1 + counter_1];
}
}
grid_and_thread_calculated = true;
};
bool HDynamicCNNManyIP::gpu_update(
float* h_init_ptr, float* h_pointer, size_t h_dim_c0, size_t h_dim_c1,
size_t h_dim_c2, size_t h_dim, float* epsilon_xy_pointer,
size_t epsilon_xy_dim_c0, size_t epsilon_xy_dim_c1,
float* epsilon_t_pointer, float* weights_pointer, size_t weights_dim_c0,
int64_t* input_pointer, size_t input_dim_c0, size_t input_dim_c1,
size_t input_dim_c2, size_t number_of_spikes, size_t dim_x, size_t dim_y,
float forgetting_offset, float forgetting_offset_local,
size_t number_of_pattern, size_t gpu_tuning_factor,
bool approximation_multiplication_enable, uint64_t number_of_frac_bits,
bool approximation_enable, uint64_t number_of_trunc_bits,
uint32_t ap_mask) {
if (grid_and_thread_calculated == false) {
gpu_occupancy_measure(dim_x, dim_y, number_of_pattern, h_dim);
}
assert((grid_and_thread_calculated == true));
cudaError_t status;
size_t h_sum_dim_c0 = dim_x * dim_y;
size_t h_sum_dim_c1 = dim_y;
size_t phxy_block_dim_c0 = h_dim * dim_x * dim_y;
size_t phxy_block_dim_c1 = dim_x * dim_y;
size_t phxy_block_dim_c2 = dim_y;
size_t pxy_block_dim_c0 = dim_x * dim_y;
size_t pxy_block_dim_c1 = dim_y;
float* w_memory = nullptr;
status = cudaMalloc((void**)&w_memory, number_of_pattern * h_dim * dim_x *
dim_y * sizeof(float));
assert((status == cudaSuccess));
float* h_temp_memory = nullptr;
status =
cudaMalloc((void**)&h_temp_memory,
number_of_pattern * h_dim * dim_x * dim_y * sizeof(float));
assert((status == cudaSuccess));
float* h_sum_memory = nullptr;
status = cudaMalloc((void**)&h_sum_memory,
number_of_pattern * dim_x * dim_y * sizeof(float));
assert((status == cudaSuccess));
float* epsilon_subsegment_memory = nullptr;
status = cudaMalloc((void**)&epsilon_subsegment_memory,
number_of_pattern * dim_x * dim_y * sizeof(float));
assert((status == cudaSuccess));
float* epsilon_scale_memory = nullptr;
status = cudaMalloc((void**)&epsilon_scale_memory,
number_of_pattern * dim_x * dim_y * sizeof(float));
assert((status == cudaSuccess));
float* forget_memory = nullptr;
status = cudaMalloc((void**)&forget_memory,
number_of_pattern * dim_x * dim_y * sizeof(float));
assert((status == cudaSuccess));
// ---
// Initialize h
kernel_phxy_fill_with_h<<<
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_H][0],
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_H][1],
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_H][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_H][3],
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_H][4],
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_H][5])>>>(
h_init_ptr, h_pointer, h_dim_c0, h_dim_c1, h_dim_c2, h_dim,
phxy_block_dim_c0, phxy_block_dim_c1, phxy_block_dim_c2,
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_H][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// Set epsilon memory scale to 1.0
kernel_pxy_set_to_v<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][0],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][1],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][3],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][4],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][5])>>>(
epsilon_scale_memory, 1.0,
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
for (size_t counter_spike = 0; counter_spike < number_of_spikes;
counter_spike++) {
// Get epsilon_t from gpu memory
float epsilon_t;
status = cudaMemcpy(&epsilon_t, &epsilon_t_pointer[counter_spike],
sizeof(float), cudaMemcpyDeviceToHost);
assert((status == cudaSuccess));
// Set epsilon memory subsegment to epsilon(t)
kernel_pxy_set_to_v<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][0],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][1],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][3],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][4],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][5])>>>(
epsilon_subsegment_memory, epsilon_t,
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// Set epsilon memory subsegment to forgetting_offset_local
kernel_pxy_set_to_v<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][0],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][1],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][3],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][4],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][5])>>>(
forget_memory, forgetting_offset_local,
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// if (*spike >= 0) {
// epsilon_subsegment = *epsilon_xy_pointer[*spike *
// epsilon_xy_dim_c0]
kernel_pxy_times_spike_selected_sxy<<<
dim3(
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_SPIKE_SELECTED_SXY][0],
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_SPIKE_SELECTED_SXY][1],
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_SPIKE_SELECTED_SXY]
[2]),
dim3(
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_SPIKE_SELECTED_SXY][3],
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_SPIKE_SELECTED_SXY][4],
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_SPIKE_SELECTED_SXY]
[5])>>>(
epsilon_subsegment_memory, epsilon_xy_pointer, input_pointer,
counter_spike, input_dim_c0, input_dim_c1, input_dim_c2,
epsilon_xy_dim_c0, epsilon_xy_dim_c1, epsilon_xy_dim_c0,
epsilon_xy_dim_c1, pxy_block_dim_c0, pxy_block_dim_c1,
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_SPIKE_SELECTED_SXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// Get the weight vectors according the spikes
kernel_phxy_fill_with_spike_selected_w<<<
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_SPIKE_SELECTED_W]
[0],
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_SPIKE_SELECTED_W]
[1],
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_SPIKE_SELECTED_W]
[2]),
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_SPIKE_SELECTED_W]
[3],
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_SPIKE_SELECTED_W]
[4],
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_SPIKE_SELECTED_W]
[5])>>>(
w_memory, weights_pointer, input_pointer, counter_spike, weights_dim_c0,
input_dim_c0, input_dim_c1, input_dim_c2, h_dim_c0, h_dim_c1, h_dim_c2,
h_dim, phxy_block_dim_c0, phxy_block_dim_c1, phxy_block_dim_c2,
grid_and_thread_settings[ID_KERNEL_PHXY_FILL_WITH_SPIKE_SELECTED_W][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// h_temp = h * w
if (approximation_multiplication_enable == false) {
kernel_phxy_times_phxy_equals_phxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PHXY_EQUALS_PHXY]
[0],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PHXY_EQUALS_PHXY]
[1],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PHXY_EQUALS_PHXY]
[2]),
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PHXY_EQUALS_PHXY]
[3],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PHXY_EQUALS_PHXY]
[4],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PHXY_EQUALS_PHXY]
[5])>>>(
h_pointer, w_memory, h_temp_memory,
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PHXY_EQUALS_PHXY][6]);
} else {
kernel_approximation_pure_multiplication<<<
dim3(grid_and_thread_settings[ID_KERNEL_APPROXIMATION_MULTIPLICATION]
[0],
grid_and_thread_settings[ID_KERNEL_APPROXIMATION_MULTIPLICATION]
[1],
grid_and_thread_settings[ID_KERNEL_APPROXIMATION_MULTIPLICATION]
[2]),
dim3(grid_and_thread_settings[ID_KERNEL_APPROXIMATION_MULTIPLICATION]
[3],
grid_and_thread_settings[ID_KERNEL_APPROXIMATION_MULTIPLICATION]
[4],
grid_and_thread_settings[ID_KERNEL_APPROXIMATION_MULTIPLICATION]
[5])>>>(
h_pointer, w_memory, h_temp_memory, number_of_frac_bits,
approximation_enable, number_of_trunc_bits, ap_mask,
grid_and_thread_settings[ID_KERNEL_APPROXIMATION_MULTIPLICATION][6]);
}
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// 1 / sum h_temp
kernel_phxy_one_over_sum_into_pxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_ONE_OVER_SUM_INTO_PXY][0],
grid_and_thread_settings[ID_KERNEL_PHXY_ONE_OVER_SUM_INTO_PXY][1],
grid_and_thread_settings[ID_KERNEL_PHXY_ONE_OVER_SUM_INTO_PXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_ONE_OVER_SUM_INTO_PXY][3],
grid_and_thread_settings[ID_KERNEL_PHXY_ONE_OVER_SUM_INTO_PXY][4],
grid_and_thread_settings[ID_KERNEL_PHXY_ONE_OVER_SUM_INTO_PXY]
[5])>>>(
h_temp_memory, h_sum_memory, h_dim_c0, h_dim_c1, h_dim_c2, h_dim,
h_sum_dim_c0, h_sum_dim_c1, pxy_block_dim_c0, pxy_block_dim_c1,
grid_and_thread_settings[ID_KERNEL_PHXY_ONE_OVER_SUM_INTO_PXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// epsilon_scale / sum h_temp
kernel_pxy_time_pxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][0],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][1],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][3],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][4],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][5])>>>(
h_sum_memory, epsilon_scale_memory,
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// epsilon_subsegment * epsilon_scale / sum h_temp
kernel_pxy_time_pxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][0],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][1],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][3],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][4],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][5])>>>(
h_sum_memory, epsilon_subsegment_memory,
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// epsilon_scale * forget_memory which contains forgetting_offset_local
kernel_pxy_time_pxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][0],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][1],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][3],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][4],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][5])>>>(
forget_memory, epsilon_scale_memory,
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// delta_forget = epsilon_subsegment * epsilon_scale * forget_memory
kernel_pxy_time_pxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][0],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][1],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][3],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][4],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][5])>>>(
forget_memory, epsilon_subsegment_memory,
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// delta_h = h_temp_memory * epsilon_subsegment * epsilon_scale / sum h
kernel_phxy_times_pxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][0],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][1],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][3],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][4],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][5])>>>(
h_temp_memory, h_sum_memory, h_dim_c0, h_dim_c1, h_dim_c2, h_dim,
h_sum_dim_c0, h_sum_dim_c1, phxy_block_dim_c0, phxy_block_dim_c1,
phxy_block_dim_c2,
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// h + delta_h
kernel_phxy_plus_phxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PHXY][0],
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PHXY][1],
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PHXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PHXY][3],
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PHXY][4],
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PHXY][5])>>>(
h_pointer, h_temp_memory,
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PHXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// h + delta_h + delta_forget
kernel_phxy_plus_pxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PXY][0],
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PXY][1],
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PXY][3],
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PXY][4],
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PXY][5])>>>(
h_pointer, forget_memory, h_dim_c0, h_dim_c1, h_dim_c2, h_dim,
h_sum_dim_c0, h_sum_dim_c1, phxy_block_dim_c0, phxy_block_dim_c1,
phxy_block_dim_c2,
grid_and_thread_settings[ID_KERNEL_PHXY_PLUS_PXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
kernel_pxy_times_v<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIMES_V][0],
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_V][1],
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_V][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIMES_V][3],
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_V][4],
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_V][5])>>>(
epsilon_subsegment_memory, (1.0 + forgetting_offset),
grid_and_thread_settings[ID_KERNEL_PXY_TIMES_V][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
kernel_pxy_plus_v<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_PLUS_V][0],
grid_and_thread_settings[ID_KERNEL_PXY_PLUS_V][1],
grid_and_thread_settings[ID_KERNEL_PXY_PLUS_V][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_PLUS_V][3],
grid_and_thread_settings[ID_KERNEL_PXY_PLUS_V][4],
grid_and_thread_settings[ID_KERNEL_PXY_PLUS_V][5])>>>(
epsilon_subsegment_memory, 1.0,
grid_and_thread_settings[ID_KERNEL_PXY_PLUS_V][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// epsilon_scale * epsilon_subsegment
kernel_pxy_time_pxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][0],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][1],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][3],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][4],
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][5])>>>(
epsilon_scale_memory, epsilon_subsegment_memory,
grid_and_thread_settings[ID_KERNEL_PXY_TIME_PXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
if (((counter_spike > 0) && (counter_spike % 1000 == 0)) ||
(counter_spike + 1 == number_of_spikes)) {
kernel_pxy_reciprocal<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_RECIPROCAL][0],
grid_and_thread_settings[ID_KERNEL_PXY_RECIPROCAL][1],
grid_and_thread_settings[ID_KERNEL_PXY_RECIPROCAL][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_RECIPROCAL][3],
grid_and_thread_settings[ID_KERNEL_PXY_RECIPROCAL][4],
grid_and_thread_settings[ID_KERNEL_PXY_RECIPROCAL][5])>>>(
epsilon_scale_memory,
grid_and_thread_settings[ID_KERNEL_PXY_RECIPROCAL][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
kernel_phxy_times_pxy<<<
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][0],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][1],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][3],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][4],
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][5])>>>(
h_pointer, epsilon_scale_memory, h_dim_c0, h_dim_c1, h_dim_c2, h_dim,
h_sum_dim_c0, h_sum_dim_c1, phxy_block_dim_c0, phxy_block_dim_c1,
phxy_block_dim_c2,
grid_and_thread_settings[ID_KERNEL_PHXY_TIMES_PXY][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
// Set epsilon memory scale to 1.0
kernel_pxy_set_to_v<<<
dim3(grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][0],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][1],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][2]),
dim3(grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][3],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][4],
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][5])>>>(
epsilon_scale_memory, 1.0,
grid_and_thread_settings[ID_KERNEL_PXY_SET_TO_V][6]);
status = cudaDeviceSynchronize();
assert((status == cudaSuccess));
}
}
// ------------
status = cudaFree(w_memory);
assert((status == cudaSuccess));
status = cudaFree(h_temp_memory);
assert((status == cudaSuccess));
status = cudaFree(h_sum_memory);
assert((status == cudaSuccess));
status = cudaFree(epsilon_subsegment_memory);
assert((status == cudaSuccess));
status = cudaFree(epsilon_scale_memory);
assert((status == cudaSuccess));
status = cudaFree(forget_memory);
assert((status == cudaSuccess));
return true;
};
Reference in a new issue Copy permalink