Skip to content
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

Hopefully fix flaky mullapudi reorder test #8542

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Hopefully fix flaky mullapudi reorder test #8542

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e1dd026
Hopefully fix flaky test
abadams Dec 24, 2024
eac3a29
clang-format
abadams Dec 24, 2024
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
23 changes: 13 additions & 10 deletions test/autoschedulers/mullapudi2016/reorder.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,10 +7,16 @@ using namespace Halide::Tools;
double run_test_1(bool auto_schedule) {
Var x("x"), y("y"), dx("dx"), dy("dy"), c("c");

int W = 1024;
int H = 1920;
int search_area = 7;

Buffer<uint32_t> im(2048);
im.fill(17);

Func f("f");
f(x, y, dx, dy) = x + y + dx + dy;
f(x, y, dx, dy) = im(x) + im(y + 1) + im(dx + search_area / 2) + im(dy + search_area / 2);

int search_area = 7;
RDom dom(-search_area / 2, search_area, -search_area / 2, search_area, "dom");

// If 'f' is inlined into 'r', the only storage layout that the auto scheduler
Expand All @@ -23,23 +29,20 @@ double run_test_1(bool auto_schedule) {

if (auto_schedule) {
// Provide estimates on the pipeline output
r.set_estimates({{0, 1024}, {0, 1024}, {0, 3}});
r.set_estimates({{0, W}, {0, H}, {0, 3}});
// Auto-schedule the pipeline
p.apply_autoscheduler(target, {"Mullapudi2016"});
} else {
/*
Var par;
r.update(0).fuse(c, y, par).parallel(par).reorder(x, dom.x, dom.y).vectorize(x, 4);
r.fuse(c, y, par).parallel(par).vectorize(x, 4); */

// The sequential schedule in this case seems to perform best which is
// odd have to investigate this further.
r.fuse(c, y, par).parallel(par).vectorize(x, 4);
}

// Inspect the schedule (only for debugging))
// r.print_loop_nest();

// Run the schedule
Buffer<int> out(1024, 1024, 3);
Buffer<int> out(W, H, 3);
double t = benchmark(3, 10, [&]() {
p.realize(out);
});
Expand Down Expand Up @@ -154,7 +157,7 @@ int main(int argc, char **argv) {
double manual_time = run_test_1(false);
double auto_time = run_test_1(true);

const double slowdown_factor = 15.0; // TODO: whoa
const double slowdown_factor = 2.0;
if (!get_jit_target_from_environment().has_gpu_feature() && auto_time > manual_time * slowdown_factor) {
std::cerr << "Autoscheduler time (1) is slower than expected:\n"
<< "======================\n"
Expand Down