Skip to content

[SYCL] Default work-group sizes based on max #952

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
merged 2 commits into from
Mar 25, 2020
Merged

[SYCL] Default work-group sizes based on max #952

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3044baf
Default work-group sizes based on max
nyalloc Dec 19, 2019
646c65c
Updated formatting
nyalloc Mar 24, 2020
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
15 changes: 10 additions & 5 deletions sycl/source/detail/scheduler/commands.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1417,14 +1417,19 @@ static void adjustNDRangePerKernel(NDRDescT &NDR, RT::PiKernel Kernel,

if (WGSize[0] == 0) {
Copy link
Contributor

Choose a reason for hiding this comment

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

Is this condition valid to represent all "default work-group size" cases?

Copy link
Contributor

@romanovvlad romanovvlad Mar 2, 2020 •
edited
Loading

Choose a reason for hiding this comment

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

WGSize[0] == 0 means user hasn't asked for a specific local size using attribute((reqd_work_group_size(X, Y, Z)) only.

// kernel does not request specific workgroup shape - set one
// TODO maximum work group size as the local size might not be the best
// choice for CPU or FPGA devices
id<3> MaxWGSizes =
get_device_info<id<3>, cl::sycl::info::device::max_work_item_sizes>::
get(DeviceImpl.getHandleRef(), DeviceImpl.getPlugin());

size_t WGSize1D = get_kernel_work_group_info<
size_t, cl::sycl::info::kernel_work_group::work_group_size>::
get(Kernel, DeviceImpl.getHandleRef(), DeviceImpl.getPlugin());
assert(WGSize1D != 0);
// TODO implement better default for 2D/3D case:
WGSize = {WGSize1D, 1, 1};

assert(MaxWGSizes[2] != 0);

// Set default work-group size in the Z-direction to either the max
// number of work-items or the maximum work-group size in the Z-direction.
WGSize = {1, 1, min(WGSize1D, MaxWGSizes[2])};
}
NDR.set(NDR.Dims, nd_range<3>(NDR.NumWorkGroups * WGSize, WGSize));
}
Expand Down