On posix run tests in parallel using a thread pool with fork-per-test
isolation (ported from edge).
On all platforms add progress output showing suite name and test count,
as well as elapsed time in the summary.
Copy src1 into dest before the masked vrangeps so that unmasked lanes
(normal/inf/NaN values) are preserved via merge-masking. Previously,
when the register allocator assigned different physical registers for
dest and src1, unmasked lanes would retain stale values from dest.
Tested with Intel SDE and AVX-512 cpu to verify.
Also shift unused FpRangeSign constants into bits [3:2] of the vrangeps
immediate, matching the Intel encoding. Only one used in the code was 0
so it was already correct by accident but using the others in the future
would produce incorrect results.
Use the `vptestnmd`-instruction to quickly test if the exponent bits of
each element is zero and put these results into a mask register. This
mask register can then be used to write `+0.0` or `-0.0` values
depending on the original value's sign-bit.
A masked `vrangeps` instruction is used to move zero-values into the effected
elements while preserving the sign-bit without having to touch memory.
Adds support for systems with host pages > 4KB (some Linux distros,
MacOS) where guest 4KB page operations can't map 1:1 to host mprotect calls.
No-op on windows and posix systems with 4KB page size.
Co-authored-by: Will Martin <wmarti@bu.edu>
When emit_inline_mmio_checks is enabled, the x64 backend emits
explicit address range checks (0x7FC00000–0x7FFFFFFF) before I32 memory
ops and routes hits through MMIOAwareLoad/Store directly, avoiding the
cost of trapping access violations for MMIO regions. The MMIO page
commit in AddVirtualMappedRange is skipped in this mode since accesses
are handled inline rather than via exception.
version.h is now found via CMAKE_BINARY_DIR on the include path instead
of a hardcoded "build/version.h" path. This allows out-of-tree and
multi-config builds to each generate and find their own version header
without colliding.
When SuspendThread is called on an already-suspended thread, it would send
another pthread_kill signal, which would nest signal handlers and create
multiple outstanding sem_wait calls while Resume only posts once when count
reaches 0. Instead we increment suspend count and skip redundant signal.
- add SIGTRAP handling
- remove per-test dots and print per suite instead
- refactor test loop from collecting all tests to iterating suites
- fix logging up output
d37c22aa replaced skylander with the new portal HID library but
missed updating three CMakeLists.txt files. No replacement needed
since xenia-hid already links xenia-hid-portal transitively.
For 4xMSAA 32bpp, samples 2/3 add a +4 byte horizontal offset to the
EDRAM address. The old typed buffer code (Buffer<uint4>) implicitly
truncated this offset via integer division (>> 2), and the .yw swizzle
compensated by selecting odd elements within the aligned uint4 load.
After the ByteAddressBuffer conversion, Load4 uses the exact byte
address including the +4 offset, shifting the load window by one dword.
The .yw swizzle then picks samples from the wrong pixels, causing
visual artifacts like vertical streaks in Resident Evil 5.
Fix by aligning the load address down to 16 bytes and selecting the
swizzle based on whether the address was offset.
When two threads fault on the same watched page, the second thread may
find the watch already cleared by the first. TriggerCallbacks returned
false in this case, leaving the SIGSEGV unhandled. Return true instead
so the instruction retries — the page is already unprotected.
This is the signal-safe equivalent of the QueryProtect check that
handles the same race on Windows. Fixes a regression from dbd58b.
