Skip to content

Split ARM64 prefetches at the cache line size #9599

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
alexreinking merged 1 commit into from
Oct 26, 2026
+35 −10
Merged

Split ARM64 prefetches at the cache line size #9599

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 11 additions & 2 deletions src/Prefetch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -433,9 +433,18 @@ Stmt reduce_prefetch_dimension(Stmt stmt, const Target &t) {
// two dimension. Other architectures generate one prefetch per cache line.
if (t.has_feature(Target::HVX)) {
max_dim = 2;
} else if (t.arch == Target::ARM && t.bits == 64 &&
(t.os == Target::MacOS || t.os == Target::IOS)) {
// Apple's 64-bit ARM cores have 128-byte cache lines.
max_dim = 1;
max_byte_size = 128;

@slomp slomp Oct 23, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Out of curiosity, have you tried 256 instead of 128?

Context: not long ago I wrote a benchmark, and found out that even though the cache line size of Apple Silicon is 128 bytes long, for false-sharing scenarios, using 256 bytes gave better performance still.

Curious to see if it would have an effect here.

} else if (t.arch == Target::ARM && t.bits == 64) {
// 64-bit ARM cores have cache lines of at least 64 bytes.
max_dim = 1;
max_byte_size = 64;
} else if (t.arch == Target::ARM) {
// ARM's cache line size can be 32 or 64 bytes and it can switch the
// size at runtime. To be safe, we just use 32 bytes.
// 32-bit ARM's cache line size can be 32 or 64 bytes and it can
// switch the size at runtime. To be safe, we just use 32 bytes.
max_dim = 1;
max_byte_size = 32;
} else {
Expand Down
32 changes: 24 additions & 8 deletions test/correctness/prefetch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,6 +59,11 @@ Expr get_max_byte_size(const Target &t) {
Expr max_byte_size;
if (t.has_feature(Target::HVX)) {
max_byte_size = Expr();
} else if (t.arch == Target::ARM && t.bits == 64 &&
(t.os == Target::MacOS || t.os == Target::IOS)) {
max_byte_size = 128;
} else if (t.arch == Target::ARM && t.bits == 64) {
max_byte_size = 64;
} else if (t.arch == Target::ARM) {
max_byte_size = 32;
} else {
Expand Down Expand Up @@ -472,18 +477,29 @@ int test12(const Target &t) {
} // anonymous namespace

int main(int argc, char **argv) {
Target t = get_jit_target_from_environment();
std::cout << "Testing target: " << t << "\n";
// These tests only lower, so we can also check the prefetch splitting of
// targets other than the host.
std::vector<Target> targets = {
get_jit_target_from_environment(),
Target("x86-64-linux"),
Target("arm-32-linux"),
Target("arm-64-linux"),
Target("arm-64-macos"),
Target("arm-64-ios"),
};

using Fn = int (*)(const Target &t);
std::vector<Fn> tests = {test1, test2, test3, test4, test5, test6, test7, test8, test9, test10, test11, test12};

for (size_t i = 0; i < tests.size(); i++) {
printf("Running prefetch test %d\n", (int)i + 1);
int result = tests[i](t);
if (result != 0) {
printf(" prefetch test %d failed!\n", (int)i + 1);
return result;
for (const Target &t : targets) {
std::cout << "Testing target: " << t << "\n";
for (size_t i = 0; i < tests.size(); i++) {
printf("Running prefetch test %d\n", (int)i + 1);
int result = tests[i](t);
if (result != 0) {
printf(" prefetch test %d failed!\n", (int)i + 1);
return result;
}
}
}

Expand Down
Loading