Skip to content

Releases: intel/llvm

DPC++ daily 2022-08-25

25 Aug 16:20
e286166

Choose a tag to compare

Pre-release
[SYCL] Make builtins accept half pointers (#6596)

Some SYCL math builtins with pointer arguments, such as modf and sincos,
do not currently accept pointers to halfs due to the conversion to
OpenCL types not propagating through pointers. This commit fixes this
by making a special case for pointers, applying the type conversion to
the underlying types.

Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>

DPC++ daily 2022-08-24

24 Aug 16:23
351b123

Choose a tag to compare

Pre-release
[SYCL] Remove and deprecate address spaces (#6606)

DPC++ has a selection of extension address spaces with deprecated
namings. This commit removes these old namings for the address spaces.
Additionally, this PR adds a deprecation warning for
sycl::access::address_space::constant_space as it is deprecated in
SYCL2020.

Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>

DPC++ daily 2022-08-23

23 Aug 16:21
7805aa3

Choose a tag to compare

Pre-release
[SYCL][DOC] Add noexcept specifier for has_property() (#6632)

As discussed here https://github.com/intel/llvm/pull/6614 has_property() version for non compile-time-properties should be marked as `noexcept`.

DPC++ daily 2022-08-22

22 Aug 16:22
1372120

Choose a tag to compare

Pre-release
[SYCL] Enable discard_events mode for the Level Zero (#6533)

Teach the Level Zero plugin piEnqueue* functions to accept nullptr instead of a pointer to the output event. In this case event is created internally and is not visible externally.
Introduce RefCountExternal for pi_event which allows to track external references to an event. It allows to do some optimizations if we know that event is not externally visible. These optimizations are going to be implemented in subsequent PRs.
Don't create proxy event for batch if there are no externally visible events in the batch.

DPC++ daily 2022-08-20

20 Aug 16:20
770f540

Choose a tag to compare

Pre-release
[SYCL][Matrix] Add documentation about new matrix features (#6157)

The new two features are joint_matrix_fill and get_wi_data for piece-wise operations.

DPC++ daily 2022-08-19

19 Aug 16:20
36e7587

Choose a tag to compare

Pre-release
[SYCL] Move simple event_impl constructor to header (#6611)

On certain systems unittests that uses the event_impl constructor with
no arguments may disagree with the event_impl source file on the layout
of the default std::optional argument. This can cause unexpected memory
accesses when performing operations on the passed argument. This commit
works around this issue by moving the body of the constructor to the
header.

Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>

DPC++ daily 2022-08-18

18 Aug 16:19
5b2cfe2

Choose a tag to compare

Pre-release
[SYCL] Emit llvm.compiler.used for internal device globals (#5958)

This prevents early optimizations from removing device globals with
internal linkage.

DPC++ daily 2022-08-17

17 Aug 16:20
54655a2

Choose a tag to compare

Pre-release
[SYCL][ABI-break] Promote extended CG/handler members (#6555)

Several new members were added to CG/handler via the extended member
workaround. This patch promotes them to actual fields of those classes
now that the ABI can be broken.

DPC++ daily 2022-08-16

16 Aug 16:21
3323da6

Choose a tag to compare

Pre-release
[SYCL] Improve range reduction performance on CPU (#6164)

The performance improvement is the result of two complementary changes:

Using an alternative heuristic to select work-group size on the CPU.
Keeping work-groups small simplifies combination of partial results
and reduces the number of temporary variables.

Adjusting the mapping of the range to an ND-range.
Breaking the range into contiguous chunks that are assigned to each
results in streaming patterns that are better-suited to prefetching
hardware.

Signed-off-by: John Pennycook john.pennycook@intel.com

DPC++ daily 2022-08-13

13 Aug 16:20
bfc7e98

Choose a tag to compare

Pre-release
sycl-nightly/20220813

[SYCL] SYCL 2020 standalone device selectors ( gpu_selector_v and fri…