aneshlya
aneshlya
It is not related to XE_ENABLED builds. Seems that it was a bug in ISPC code for ages which can be only discoverable by using `-D_GLIBCXX_ASSERTIONS`. Proposed fix https://github.com/ispc/ispc/pull/2353.
Yes, the work to support opaque pointers is in progress.
Related PRs so far: https://github.com/ispc/ispc/pull/2357 https://github.com/ispc/ispc/pull/2360 https://github.com/ispc/ispc/pull/2363 https://github.com/ispc/ispc/pull/2365 https://github.com/ispc/ispc/pull/2368 https://github.com/ispc/ispc/pull/2374 https://github.com/ispc/ispc/pull/2377
I believe we agreed here that the tests are correct and the problem is with alignment. Closing this PR.
Support of opaque pointers for CPU build was completed. Adjustments for Xe-enabled built will be done when dependencies are ready.
The issue seems outdated, printing of bools works fine for several releases already. We have tests for checking this functionality: - https://github.com/ispc/ispc/blob/main/tests/print_bool.ispc - https://github.com/ispc/ispc/blob/main/tests/print_bool_simd.ispc
It seems that in this case ISPC has the same behavior as Clang: https://godbolt.org/z/o74oo1eE1
Fixed by PR #2501 and PR #2531.
Function specializations are added in [#2595](https://github.com/ispc/ispc/pull/2595)
UPDATE There are 3 issues described in this issue: 1. The loads are not fused ``` ld1r { v1.4s }, [x8], #4 ldp q2, q3, [x1, #16] ldr s0, [x8]...