73fe848e07
In particular DEBUG, but prefix all of them to be sure. Pull Request: https://projects.blender.org/blender/blender/pulls/141749
102 lines
2.6 KiB
C++
102 lines
2.6 KiB
C++
/* SPDX-FileCopyrightText: 2011-2022 Blender Foundation
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0 */
|
|
|
|
#include <iomanip>
|
|
|
|
#include "device/kernel.h"
|
|
#include "device/queue.h"
|
|
|
|
#include "util/algorithm.h"
|
|
#include "util/log.h"
|
|
#include "util/time.h"
|
|
|
|
CCL_NAMESPACE_BEGIN
|
|
|
|
DeviceQueue::DeviceQueue(Device *device) : device(device)
|
|
{
|
|
DCHECK_NE(device, nullptr);
|
|
is_per_kernel_performance_ = getenv("CYCLES_DEBUG_PER_KERNEL_PERFORMANCE");
|
|
}
|
|
|
|
DeviceQueue::~DeviceQueue()
|
|
{
|
|
if (LOG_IS_ON(LOG_LEVEL_STATS)) {
|
|
/* Print kernel execution times sorted by time. */
|
|
vector<pair<DeviceKernelMask, double>> stats_sorted;
|
|
for (const auto &stat : stats_kernel_time_) {
|
|
stats_sorted.push_back(stat);
|
|
}
|
|
|
|
sort(stats_sorted.begin(),
|
|
stats_sorted.end(),
|
|
[](const pair<DeviceKernelMask, double> &a, const pair<DeviceKernelMask, double> &b) {
|
|
return a.second > b.second;
|
|
});
|
|
|
|
LOG_STATS << "GPU queue stats:";
|
|
double total_time = 0.0;
|
|
for (const auto &[mask, time] : stats_sorted) {
|
|
total_time += time;
|
|
LOG_STATS << " " << std::setfill(' ') << std::setw(10) << std::fixed << std::setprecision(5)
|
|
<< std::right << time << "s: " << device_kernel_mask_as_string(mask);
|
|
}
|
|
|
|
if (is_per_kernel_performance_) {
|
|
LOG_STATS << "GPU queue total time: " << std::fixed << std::setprecision(5) << total_time;
|
|
}
|
|
}
|
|
}
|
|
|
|
void DeviceQueue::debug_init_execution()
|
|
{
|
|
if (LOG_IS_ON(LOG_LEVEL_STATS)) {
|
|
last_sync_time_ = time_dt();
|
|
}
|
|
|
|
last_kernels_enqueued_.reset();
|
|
}
|
|
|
|
void DeviceQueue::debug_enqueue_begin(DeviceKernel kernel, const int work_size)
|
|
{
|
|
if (LOG_IS_ON(LOG_LEVEL_STATS)) {
|
|
LOG_STATS << "GPU queue launch " << device_kernel_as_string(kernel) << ", work_size "
|
|
<< work_size;
|
|
}
|
|
|
|
last_kernels_enqueued_.set(kernel, true);
|
|
}
|
|
|
|
void DeviceQueue::debug_enqueue_end()
|
|
{
|
|
if (LOG_IS_ON(LOG_LEVEL_STATS) && is_per_kernel_performance_) {
|
|
synchronize();
|
|
}
|
|
}
|
|
|
|
void DeviceQueue::debug_synchronize()
|
|
{
|
|
if (LOG_IS_ON(LOG_LEVEL_STATS)) {
|
|
const double new_time = time_dt();
|
|
const double elapsed_time = new_time - last_sync_time_;
|
|
LOG_STATS << "GPU queue synchronize, elapsed " << std::setw(10) << elapsed_time << "s";
|
|
|
|
/* There is no sense to have an entries in the performance data
|
|
* container without related kernel information. */
|
|
if (last_kernels_enqueued_.any()) {
|
|
stats_kernel_time_[last_kernels_enqueued_] += elapsed_time;
|
|
}
|
|
|
|
last_sync_time_ = new_time;
|
|
}
|
|
|
|
last_kernels_enqueued_.reset();
|
|
}
|
|
|
|
string DeviceQueue::debug_active_kernels()
|
|
{
|
|
return device_kernel_mask_as_string(last_kernels_enqueued_);
|
|
}
|
|
|
|
CCL_NAMESPACE_END
|