happyz
/
gemma.cpp
mirror of https://github.com/google/gemma.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Only temporarily enable spinning in threading benchmark 

PiperOrigin-RevId: 727114863
This commit is contained in:
Jan Wassenberg 2025-02-14 17:15:10 -08:00 committed by Copybara-Service
parent 06c70dccd9
commit bdf5d25e97
2 changed files with 22 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
util/threading.h
View File

@ -226,6 +226,8 @@ class NestedPools {
BoundedSlice cluster_slice = BoundedSlice(), BoundedSlice cluster_slice = BoundedSlice(),
BoundedSlice lp_slice = BoundedSlice()); BoundedSlice lp_slice = BoundedSlice());
bool AllPinned() const { return all_pinned_; }
// Subject to `use_spinning`, enables spin waits with the goal of reducing the // Subject to `use_spinning`, enables spin waits with the goal of reducing the
// latency of barrier synchronization. We only spin during Generate to avoid // latency of barrier synchronization. We only spin during Generate to avoid
// wasting energy during long waits. If `use_spinning` is kDefault, we first // wasting energy during long waits. If `use_spinning` is kDefault, we first

30
util/threading_test.cc
View File

@ -276,16 +276,27 @@ TEST(ThreadingTest, BenchJoin) {
constexpr size_t kInputs = 1; constexpr size_t kInputs = 1;
static hwy::FuncInput inputs[kInputs]; static hwy::FuncInput inputs[kInputs];
const auto measure = [&](hwy::ThreadPool& pool, const char* caption) { const auto measure = [&](hwy::ThreadPool& pool, bool spin,
const char* caption) {
inputs[0] = inputs[0] =
static_cast<hwy::FuncInput>(hwy::Unpredictable1() * pool.NumWorkers()); static_cast<hwy::FuncInput>(hwy::Unpredictable1() * pool.NumWorkers());
hwy::Result results[kInputs]; hwy::Result results[kInputs];
hwy::Params params; hwy::Params params;
params.verbose = false; params.verbose = false;
params.max_evals = kMaxEvals; params.max_evals = kMaxEvals;
// Only spin for the duration of the benchmark to avoid wasting energy and
// interfering with the other pools.
if (spin) {
pool.SetWaitMode(hwy::PoolWaitMode::kSpin);
}
const size_t num_results = const size_t num_results =
Measure(&ForkJoin, reinterpret_cast<const uint8_t*>(&pool), inputs, Measure(&ForkJoin, reinterpret_cast<const uint8_t*>(&pool), inputs,
kInputs, results, params); kInputs, results, params);
if (spin) {
pool.SetWaitMode(hwy::PoolWaitMode::kBlock);
}
for (size_t i = 0; i < num_results; ++i) { for (size_t i = 0; i < num_results; ++i) {
printf("%-20s: %5d: %6.2f us; MAD=%4.2f%%\n", caption, printf("%-20s: %5d: %6.2f us; MAD=%4.2f%%\n", caption,
static_cast<int>(results[i].input), static_cast<int>(results[i].input),
@ -303,20 +314,19 @@ TEST(ThreadingTest, BenchJoin) {
}; };
NestedPools pools(0); NestedPools pools(0);
measure(pools.AllPackages(), "block packages"); measure(pools.AllPackages(), false, "block packages");
if (pools.AllClusters(0).NumWorkers() > 1) { if (pools.AllClusters(0).NumWorkers() > 1) {
measure(pools.AllClusters(0), "block clusters"); measure(pools.AllClusters(0), false, "block clusters");
} }
measure(pools.Cluster(0, 0), "block in_cluster"); measure(pools.Cluster(0, 0), false, "block in_cluster");
Tristate use_spinning = Tristate::kDefault; if (pools.AllPinned()) {
pools.MaybeStartSpinning(use_spinning); const bool kSpin = true;
if (use_spinning == Tristate::kTrue) { measure(pools.AllPackages(), kSpin, "spin packages");
measure(pools.AllPackages(), "spin packages");
if (pools.AllClusters(0).NumWorkers() > 1) { if (pools.AllClusters(0).NumWorkers() > 1) {
measure(pools.AllClusters(0), "spin clusters"); measure(pools.AllClusters(0), kSpin, "spin clusters");
} }
measure(pools.Cluster(0, 0), "spin in_cluster"); measure(pools.Cluster(0, 0), kSpin, "spin in_cluster");
} }
} }