mirror/trigdx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add allocate_memory and free_memory

Browse Source
This commit is contained in:
Bram Veenboer
2025-09-02 11:55:33 +02:00
parent 716f323b26
commit 8df4bbf54e
4 changed files with 59 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
benchmarks/benchmark_utils.hpp
View File

@@ -7,8 +7,8 @@
#include <benchmark/benchmark.h>
void init_x(std::vector<float> &x) {
for (size_t i = 0; i < x.size(); ++i) {
void init_x(float *x, size_t n) {
for (size_t i = 0; i < n; ++i) {
x[i] = (i % 360) * 0.0174533f; // degrees to radians
}
}
@@ -16,24 +16,31 @@ void init_x(std::vector<float> &x) {
template <typename Backend>
static void benchmark_sinf(benchmark::State &state) {
const size_t N = static_cast<size_t>(state.range(0));
std::vector<float> x(N), s(N);
init_x(x);
Backend backend;
auto start = std::chrono::high_resolution_clock::now();
backend.init(N);
float *x =
reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
float *s =
reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
auto end = std::chrono::high_resolution_clock::now();
state.counters["init_ms"] =
std::chrono::duration_cast<std::chrono::microseconds>(end - start)
.count() /
1.e3;
init_x(x, N);
for (auto _ : state) {
backend.compute_sinf(N, x.data(), s.data());
backend.compute_sinf(N, x, s);
benchmark::DoNotOptimize(s);
}
backend.free_memory(x);
backend.free_memory(s);
state.SetItemsProcessed(static_cast<int64_t>(state.iterations()) *
static_cast<int64_t>(N));
}
@@ -41,24 +48,31 @@ static void benchmark_sinf(benchmark::State &state) {
template <typename Backend>
static void benchmark_cosf(benchmark::State &state) {
const size_t N = static_cast<size_t>(state.range(0));
std::vector<float> x(N), c(N);
init_x(x);
Backend backend;
auto start = std::chrono::high_resolution_clock::now();
backend.init(N);
float *x =
reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
float *c =
reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
auto end = std::chrono::high_resolution_clock::now();
state.counters["init_ms"] =
std::chrono::duration_cast<std::chrono::microseconds>(end - start)
.count() /
1.e3;
init_x(x, N);
for (auto _ : state) {
backend.compute_cosf(N, x.data(), c.data());
backend.compute_cosf(N, x, c);
benchmark::DoNotOptimize(c);
}
backend.free_memory(x);
backend.free_memory(c);
state.SetItemsProcessed(static_cast<int64_t>(state.iterations()) *
static_cast<int64_t>(N));
}
@@ -66,25 +80,35 @@ static void benchmark_cosf(benchmark::State &state) {
template <typename Backend>
static void benchmark_sincosf(benchmark::State &state) {
const size_t N = static_cast<size_t>(state.range(0));
std::vector<float> x(N), s(N), c(N);
init_x(x);
Backend backend;
auto start = std::chrono::high_resolution_clock::now();
backend.init(N);
float *x =
reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
float *s =
reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
float *c =
reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
auto end = std::chrono::high_resolution_clock::now();
state.counters["init_ms"] =
std::chrono::duration_cast<std::chrono::microseconds>(end - start)
.count() /
1.e3;
init_x(x, N);
for (auto _ : state) {
backend.compute_sincosf(N, x.data(), s.data(), c.data());
backend.compute_sincosf(N, x, s, c);
benchmark::DoNotOptimize(s);
benchmark::DoNotOptimize(c);
}
backend.free_memory(x);
backend.free_memory(s);
backend.free_memory(c);
state.SetItemsProcessed(static_cast<int64_t>(state.iterations()) *
static_cast<int64_t>(N));
}