AUDIT-059 round 15 — diagnostic. New cvar `audit_jit_prolog_r3_bytes`
(default 64 = existing behaviour, capped at 256, rounded up to 16B
multiple) controls how many bytes are dumped at host(r3) when the
audit_jit_prolog_pc probe fires. Set to 80 to capture audit-051's
stack-local struct at sub_82452DC0's r31+96.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Round-14 of AUDIT-2BF (singleton-dump). Pairs with xenia-rs'
--audit-mem-read-hex to emit one comparable XELOGKERNEL line resolving
the bctrl target at sub_822F1AA8+0x90 (PC 0x822F1B4C):
[0x828E1F08] -> singleton instance ptr
[singleton+0] -> vtable
[vtable+0] -> vtable[0] (= first virtual method, bctrl tgt)
[vtable+24] -> vtable[24] (= slot 6, silph chain target)
Two complementary hooks:
1. src/xenia/cpu/backend/x64/x64_emitter.cc: extend
AuditLogJitPrologArgs. New cvar `audit_jit_prolog_mem_dump` (uint32).
When non-zero and an `audit_jit_prolog_pc` fire happens, the host
side dereferences the VA 3 levels deep and emits one
AUDIT-MEM-READ line in the same format ours emits. Defensive
per-level null + VA-range checks.
2. src/xenia/kernel/kernel_state.cc: one-shot dump in
EmulateCPInterruptDPC of the same chain (hard-coded to
0x828E1F08). Useful when audit_jit_prolog_pc isn't set; fires the
first time the CP interrupt path runs (after the singleton ctor
has had time to populate).
Read-only. Both gates default-off; no impact when cvars unset.
~65 LOC total across the two files.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two cvar-gated AUDIT-HLC probes for thread-spawn lineage attribution:
1. ExCreateThread_entry shim logs (start_address, start_context, xapi,
flags, kernel lr, guest_lr). guest_lr is recovered via the same
one-frame-up PPC back-chain walk used by the NtWaitForSingleObjectEx
probe (lr saved at [back2 - 8]).
2. XThread::Execute emits (tid, start_address, start_context, xapi) at
the very top so each running thread can be matched back to its
creating ExCreateThread entry by (start_address, start_context).
Gated on the existing cvars::audit_handle_lifecycle; no new cvar.
Pulled in xenia/kernel/kernel_flags.h from xthread.cc to expose the
cvar declaration there.
Used by audit-059 round 12 to enumerate the 23 thread spawns in
canary's 35 s boot, attribute each to its guest_lr caller, and cross-
reference against ours' 10 spawns to identify the 7 missing
spawner functions (including sub_824F7800 which spawns the 4 silph
PKEVENT workers at entry=0x82506528/58/88/B8 via guest_lr=0x824F7B24).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replace the round-7 hardcoded `current_guest_function_ == 0x824F7800`
gate with a runtime cvar `audit_jit_prolog_pc` (uint32, 0 = disabled).
The cvar names the guest entry PC at which the x64 emitter inserts a
CallNative to the prolog probe; the probe now dumps the active PC in
each log line so multiple instrumentation campaigns can share output.
Renames `audit_log_sub824F7800_args` (cvar) and `AuditLogSub824F7800Args`
(host function) — the round-7 sub_824F7800-specific cvar and function
are deleted, not left dormant. The `audit_handle_lifecycle` cvar and
its existing AUDIT-HLC probe sites in xboxkrnl_threading.cc /
kernel_state.cc are untouched.
Validated by audit-059 round 9: with `--audit_jit_prolog_pc=0x821B55D8`
the silph WorkerCtx-init chain entry fires 1× in canary at tid=6
lr=0x82172D8C (call site at sub_82172BA0+0x1E8 `bctrl`, the virtual
dispatch through vtable slot 6 that statically-blind xref-walkers
cannot enumerate).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds cvar audit_log_sub824F7800_args (off by default). When set, the x64
JIT emits one CallNative to AuditLogSub824F7800Args at the start of the
JIT-compiled body of guest function 0x824F7800 (silph::WorkerCtx ctor's
immediate caller per audit-058). The hook reads r3..r10, LR, and 64
bytes at host(r3) from PPCContext and writes them to XELOGKERNEL.
The hook is placed *after* the JIT prolog (stack push, GUEST_RET_ADDR /
GUEST_CALL_RET_ADDR setup, optional trace-functions block) and *before*
the first HIR body instruction, so r3..r10 and lr in PPCContext still
reflect the caller's args (no LOAD/STORE_CONTEXT has executed yet inside
the callee). The compile-time gate `current_guest_function_ == 0x824F7800`
makes this a zero-cost no-op for all other functions and a single extra
CallNative for the one target.
Audit-059 round 7 probe captures r3..r10 + LR for a-prime synthetic
replay of sub_824F7800 from a host hook in xenia-rs. Run with:
--audit_handle_lifecycle=true --audit_log_sub824F7800_args=true
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds a 41-LOC cvar-gated probe (audit_handle_lifecycle) that emits a one-shot
0x300-byte hexdump of the silph::WorkerCtx context the first time KeSetEvent
fires into the silph UI PKEVENT cluster (0xBCE25200..0xBCE25300 in current
builds). Recovers ctx_base by anchoring on the canonical layout (events at
ctx+0x54/+0x64/+0x74/+0x84 with 16-byte stride; ctx_base = ev_addr - (ev_addr
- 0xBCE251C0)). Also emits an explicit per-slot summary of the 8 candidate
KEVENT headers at +0x54..+0xC4 for sanity.
Produces 48 DUMP rows + 8 slot rows in canary.log on the first triggering
KeSetEvent. Guarded by std::atomic_bool so subsequent fires are silent. Used
by iterate 2.BF context-replication in ours: we need the live [ctx+0] vtable
pointer and the 4 quiet/4 active KEVENT slot config to synthesize a matching
WorkerCtx and spawn the four sub_82506xxx entries.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Extend the AUDIT-HLC probes with guest LR capture:
- NtCreateEvent, NtSetEvent, KeSetEvent: log immediate guest LR
(cpu::ThreadState::Get()->context()->lr) so callers can be
attributed to specific guest functions.
- NtWaitForSingleObjectEx entry + _done: also walk one PPC stack
frame up to recover the guest_lr of the wait wrapper's caller
(Xbox 360 EABI: saved LR at [prev_sp - 8], see xenia-rs's
walk_guest_back_chain). lr_enter alone gives only the kernel
wait wrapper return address (e.g. 0x824AC578); guest_lr surfaces
the actual call site.
Still cvar-gated on audit_handle_lifecycle. Used in audit-059 round 4
to map canary handles F8000xxx to ours' 0x12xx wedges via guest LR
matching against silph::UImpl/GamePart cluster.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New cvar `audit_handle_lifecycle` (Auditing group, default false).
When enabled, emits one-line XELOGKERNEL traces tagged AUDIT-HLC at:
NtCreateEvent return, NtSetEvent entry, KeSetEvent entry,
NtWaitForSingleObjectEx entry + completion,
EmulateCPInterruptDPC entry.
Observation-only: zero-overhead when cvar off (single branch + flag
read). Intended for handle disambiguation between xenia-rs and
canary under the Wine cross-build oracle.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
New CMake preset `cross-win-clangcl` for Ninja Multi-Config on
non-Windows hosts. Toolchain: clang-cl (MSVC-ABI), lld-link, xwin SDK/CRT
splat. Shader compilation routes through wine fxc.exe with FXC_PATH
env forwarding and unix→Windows path translation via `winepath -w`.
Plus per-file build fixes (constexpr→const, llvm-rc forward-slash,
zlib-ng AVX guard, /RTCsu MSVC-only). third_party/snappy bumped to
fabi/sylpheed-crossbuild for the target-aware POSIX-gate header.
Produces `xenia_canary.exe` (Debug MSVC) suitable for use as the
audit oracle under Wine for xenia-rs work. See
docs/CROSS_BUILD_SETUP.md for the full reproduction recipe.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Thanks to latest change there is no need to use hardcoded nanosleep, so we can go back to previous stuff.
Hopefully it will fix some random issues with audio introduced lately
Integer num_format fetches now get their scale CPU-side instead of trying to guess in the shader. This is authoritative; no cvar. Both translators now use texture_integer_scale_bits after signs/gamma and before exponent bias. This alone fixes black screens and a bunch of rendering bugs across at least several dozen titles.
This new information lives in the updated FormatInfo table, including fixed component widths.
Resolve also has two new fixes.
8_8_8_8_GAMMA EDRAM sources can now decode through the PWL curve while still in linear space, before MSAA resolve or format conversion, then re-encoded for gamma destinations. This is also now default enabled via gamma_decode_pwl_resolve and has improved blowout in at least 4 titles, with no obvious regressions thus far.
Full resolve can also now pack fixed destinations according to copy_dest_number.
Check shared memory validity before marking textures up to date and installing watches. If the backing range was invalidated, leave the texture outdated so it can be reloaded later.
Use new archive input to rework artifacts.
Add runs-on and llvm_version inputs and ubuntu_base output to Lint so it can be passed to the Linux job.
Remove broken config input.
Upgrade Windows to VS2026.
Replace the x64 STVL/STVR lowering with explicit byte stores that preserve the
guest-visible vector byte order.
The previous vector shuffle/blend path could write incorrect bytes for partial
vector stores, which broke optimized guest memcpy implementations using
stvlx/stvrx for unaligned heads and tails. In 4156081C this corrupted skinned
model render command data and eventually caused render-thread crashes.
Add regression coverage for StoreVectorLeft/StoreVectorRight across common
unaligned offsets, plus a memcpy-head style case matching the affected guest
instruction pattern.
Implements EVENT_WRITE_ZPD handling across shared CommandProcessor, D3D12, and Vulkan backends.
This adds a shared report lifecycle. Reports can span submissions/render passes, split into multiple segments, and asynchronously retire. Reports can also survive same-slot reuse, which is critical for titles that aggressively recycle against a small pool.
RTV/FBO use their respective host occlusion queries, whereas ROV/FSI use counter buffers instead, with shader helpers that are called from the translated PS, accumulating surviving MSAA sample coverage into active slots. Regardless of approach, results are copied to small, dedicated readbacks and fed through shared retirement.
Fast mode writes speculative values and patches in resolved results later. Strict mode waits for real results with a small retire backstop so it doesn't spin forever when a report gets weird. Fake mode's fundamentals are unchanged, it just now uses XenosZPDReport memory helpers like all the modes do.
Also adds optional sample count saturation for instances when attenuation isn't a one-for-one with real hardware, an optional alternate fast behavior that preserves cached zeroes (that can improve flare-heavy titles but might regress occlusion culling), and normalization for upscaling.
QueryBatch is still unsupported other than a fake path that's enabled when a range is set via cvar.
