SYCL/oneMath backend#46

Open

constracktor wants to merge 24 commits into

SC-SGS:mainfrom

MarcelGraf0710:sycl_backend

Commits on May 12, 2026

Fix redundant tile copies
constracktor
committed
Fix spelling error
constracktor
committed
GPyTorch and GPflow work on AMD

MarcelGraf0710
authored and
constracktor
committed
Adapted scripts such that installation of GPflow and GPyTorch acutally works on AMD and Intel

MarcelGraf0710
authored and
constracktor
committed
Adapted paths such that Intel variant compiles

MarcelGraf0710
authored and
constracktor
committed
Major update to the PyTorch reference

MarcelGraf0710
authored and
constracktor
committed
Adapted and fixed GPyTorch and GPflow reference implementations

MarcelGraf0710
authored and
constracktor
committed
Adapted GPRat benchmark

MarcelGraf0710
authored and
constracktor
committed
Improvements to compile architecture and correctness test.

MarcelGraf0710
authored and
constracktor
committed
Add predict_with_full_cov for GPflow and GPyTorch, and minor improvement for GPRat Python script

MarcelGraf0710
authored and
constracktor
committed
Created shell scripts for benchmarking and minor adjustments to GPRat Python script

MarcelGraf0710
authored and
constracktor
committed
Adapt execute.cpp

MarcelGraf0710
authored and
constracktor
committed
execute.cpp now tests different CPU counts on GPU

MarcelGraf0710
authored and
constracktor
committed
Update all Python benchmarks to test multiple tile and core counts

MarcelGraf0710
authored and
constracktor
committed
Set common benchmark baseline

MarcelGraf0710
authored and
constracktor
committed
Thread binding for benchmarks with HPX

MarcelGraf0710
authored and
constracktor
committed
Adapt boilerplate

MarcelGraf0710
authored and
constracktor
committed
Apply formating
constracktor
committed
Remove parallel building
constracktor
authored

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

SYCL/oneMath backend#46

SYCL/oneMath backend#46
constracktor wants to merge 24 commits into
SC-SGS:mainfrom
MarcelGraf0710:sycl_backend

Commits on May 12, 2026

Fix redundant tile copies

Fix spelling error

GPyTorch and GPflow work on AMD

Adapted scripts such that installation of GPflow and GPyTorch acutally works on AMD and Intel

Adapted paths such that Intel variant compiles

Major update to the PyTorch reference

Adapted and fixed GPyTorch and GPflow reference implementations

Adapted GPRat benchmark

Improvements to compile architecture and correctness test.

Add predict_with_full_cov for GPflow and GPyTorch, and minor improvement for GPRat Python script

Created shell scripts for benchmarking and minor adjustments to GPRat Python script

Adapt execute.cpp

execute.cpp now tests different CPU counts on GPU

Update all Python benchmarks to test multiple tile and core counts

Set common benchmark baseline

Thread binding for benchmarks with HPX

Adapt boilerplate

Apply formating

Remove parallel building

Commits on May 24, 2026

Revert to canonical test data

Remove commented out code for unused optimization-related functionality

Assimilate hyperparam stubs of SYCL to those of CUDA

Comment corrections and code hygiene

Fix formatting