Skip to content

[SYCL][PI] Fix PI_KERNEL_MAX_SUB_GROUP_SIZE query for devices without sub groups #9543

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 6 commits into from
May 30, 2023
Merged

[SYCL][PI] Fix PI_KERNEL_MAX_SUB_GROUP_SIZE query for devices without sub groups #9543

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
3c64a2c
[SYCL] Fix `PI_KERNEL_MAX_SUB_GROUP_SIZE` query for devices without s…
0x12CC May 19, 2023
a745102
Add error check for second `piDeviceGetInfo` call
0x12CC May 19, 2023
3bbc51c
Add `piDeviceGetInfo` comment
0x12CC May 25, 2023
110e606
Merge branch 'sycl' into kernel_max_sg_size
0x12CC May 25, 2023
21a74d8
Add size assertion
0x12CC May 30, 2023
e7f4f52
Merge branch 'sycl' into kernel_max_sg_size
0x12CC May 30, 2023
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
25 changes: 18 additions & 7 deletions sycl/plugins/opencl/pi_opencl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1486,14 +1486,25 @@ pi_result piKernelGetSubGroupInfo(pi_kernel kernel, pi_device device,
ret_val = 0; // Not specified by kernel
ret_err = CL_SUCCESS;
} else if (param_name == PI_KERNEL_MAX_SUB_GROUP_SIZE) {
// Return the maximum work group size for the kernel
size_t kernel_work_group_size = 0;
pi_result pi_ret_err = piKernelGetGroupInfo(
kernel, device, PI_KERNEL_GROUP_INFO_WORK_GROUP_SIZE, sizeof(size_t),
&kernel_work_group_size, nullptr);
if (pi_ret_err != PI_SUCCESS)
// Return the maximum sub group size for the device
size_t result_size = 0;
// Two calls to piDeviceGetInfo are needed: the first determines the size
// required to store the result, and the second returns the actual size
// values.
pi_result pi_ret_err =
piDeviceGetInfo(device, PI_DEVICE_INFO_SUB_GROUP_SIZES_INTEL, 0,
nullptr, &result_size);
if (pi_ret_err != PI_SUCCESS) {
return pi_ret_err;
ret_val = kernel_work_group_size;
}
assert(result_size % sizeof(size_t) == 0);
std::vector<size_t> result(result_size / sizeof(size_t));
pi_ret_err = piDeviceGetInfo(device, PI_DEVICE_INFO_SUB_GROUP_SIZES_INTEL,
result_size, result.data(), nullptr);
if (pi_ret_err != PI_SUCCESS) {
return pi_ret_err;
}
ret_val = *std::max_element(result.begin(), result.end());
ret_err = CL_SUCCESS;
} else if (param_name == PI_KERNEL_COMPILE_SUB_GROUP_SIZE_INTEL) {
ret_val = 0; // Not specified by kernel
Expand Down