NVIDIA/nvbench
1
0
Fork 0
mirror of https://github.com/NVIDIA/nvbench.git synced 2026-04-20 14:58:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add a cuda stream member to nvbench::state

Browse Source
This commit is contained in:
Yunsong Wang
2022-02-04 13:51:30 -05:00
parent 15f2e92fdf
commit 8aea3e467e
3 changed files with 23 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
nvbench/cuda_stream.cuh
View File

@@ -29,14 +29,14 @@ namespace nvbench
struct cuda_stream
{
cuda_stream()
: m_owning(true)
: m_owning{true}
{
NVBENCH_CUDA_CALL(cudaStreamCreate(&m_stream));
}
cuda_stream(cuda_stream stream, bool owning)
: m_stream(stream)
, m_owning(owning)
: m_stream{stream}
, m_owning{owning}
{}
// destroy the stream if it's owning
@@ -55,8 +55,8 @@ struct cuda_stream
cuda_stream &operator=(const cuda_stream &) = delete;
cuda_stream(cuda_stream &&other)
: m_stream(other.get_stream())
, m_owning(other.is_owning())
: m_stream{other.get_stream()}
, m_owning{other.is_owning()}
{
other.destroy();
}

19
nvbench/state.cuh
View File

@@ -18,6 +18,7 @@
#pragma once
#include <nvbench/cuda_stream.cuh>
#include <nvbench/device_info.cuh>
#include <nvbench/exec_tag.cuh>
#include <nvbench/named_values.cuh>
@@ -62,6 +63,15 @@ struct state
state &operator=(const state &) = delete;
state &operator=(state &&) = default;
[[nodiscard]] const nvbench::cuda_stream &get_cuda_stream() const
{
return m_cuda_stream;
}
void set_cuda_stream(nvbench::cuda_stream &&stream)
{
m_cuda_stream = std::move(stream);
}
/// The CUDA device associated with with this benchmark state. May be
/// nullopt for CPU-only benchmarks.
[[nodiscard]] const std::optional<nvbench::device_info> &get_device() const
@@ -259,11 +269,9 @@ struct state
[[nodiscard]] bool is_cupti_required() const
{
return is_l2_hit_rate_collected()
|| is_l1_hit_rate_collected()
|| is_stores_efficiency_collected()
|| is_loads_efficiency_collected()
|| is_dram_throughput_collected();
return is_l2_hit_rate_collected() || is_l1_hit_rate_collected() ||
is_stores_efficiency_collected() ||
is_loads_efficiency_collected() || is_dram_throughput_collected();
}
summary &add_summary(std::string summary_tag);
@@ -303,6 +311,7 @@ private:
std::optional<nvbench::device_info> device,
std::size_t type_config_index);
nvbench::cuda_stream m_cuda_stream;
std::reference_wrapper<const nvbench::benchmark_base> m_benchmark;
nvbench::named_values m_axis_values;
std::optional<nvbench::device_info> m_device;

6
nvbench/state.cxx
View File

@@ -33,7 +33,8 @@ namespace nvbench
{
state::state(const benchmark_base &bench)
: m_benchmark{bench}
: m_cuda_stream{}
, m_benchmark{bench}
, m_run_once{bench.get_run_once()}
, m_min_samples{bench.get_min_samples()}
, m_min_time{bench.get_min_time()}
@@ -46,7 +47,8 @@ state::state(const benchmark_base &bench,
nvbench::named_values values,
std::optional<nvbench::device_info> device,
std::size_t type_config_index)
: m_benchmark{bench}
: m_cuda_stream{}
, m_benchmark{bench}
, m_axis_values{std::move(values)}
, m_device{std::move(device)}
, m_type_config_index{type_config_index}