Skip to content

[SYCL][WIP] Reduce compile time builtin#21688

Draft
koparasy wants to merge 2 commits intointel:syclfrom
koparasy:reduce-compile-time-builtin
Draft

[SYCL][WIP] Reduce compile time builtin#21688
koparasy wants to merge 2 commits intointel:syclfrom
koparasy:reduce-compile-time-builtin

Conversation

@koparasy
Copy link
Copy Markdown
Contributor

@koparasy koparasy commented Apr 6, 2026

No description provided.

koparasy added 2 commits April 5, 2026 14:36
…lper

customization points so <sycl/builtins.hpp> no longer needs to include the
full <sycl/multi_ptr.hpp> definition. Preserve multi_ptr builtin behavior once
multi_ptr.hpp is included, and update include-deps / regression coverage.

On a TU including only <sycl/builtins.hpp>:

Host-only Frontend: 601.943 ms -> 572.657 ms (-4.87%)
Device-only Frontend: 592.140 ms -> 457.361 ms (-22.76%)
multi_ptr.hpp parse time dropped from 18.708 ms -> 0 ms on host
multi_ptr.hpp parse time dropped from 108.662 ms -> 0 ms on device
…uiltins.hpp`.

This keeps the relational builtins path self-contained by introducing a
narrow relational_mask_widen helper for relation-mask widening, and
replaces the old transitive dependencies with the direct headers that
builtins actually needs (generic_type_traits.hpp, half_type.hpp).
It also adds the missing direct exception.hpp include for
ext/oneapi/bf16_storage_builtins.hpp and updates the affected
include-deps / IR checks.

Measured with measure_builtin.cpp and -ftime-trace
against the base branch:

host Total ExecuteCompiler: 604.853 ms -> 533.452 ms (-71.401 ms, -11.80%)
host Total Frontend: 600.196 ms -> 529.366 ms (-70.830 ms, -11.80%)
device Total ExecuteCompiler: 598.557 ms -> 420.885 ms (-177.672 ms, -29.68%)
device Total Frontend: 593.798 ms -> 416.060 ms (-177.738 ms, -29.93%)

changes
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant