BaseHeap::Dispose() walks the page table and calls
DeallocFixed(addr, 0, kRelease) for every allocated page region.
These addresses are within file-backed view mappings (i.e. guest memory).
On Windows, VirtualFree on file-mapped pages silently returns FALSE.
On POSIX, the code hit assert_always() and crashed. Changed to return false
to match Windows behavior.
Memory leaks from munmap(addr, 0):
Callers passed length=0 to DeallocFixed(kRelease). On Windows this works because
VirtualFree(addr, 0, MEM_RELEASE) means "release the entire region" — Windows
ignores the length parameter for MEM_RELEASE and always frees the whole
allocation. On POSIX, munmap(addr, 0) fails with EINVAL, so the memory was
silently leaked every time.
On Linux, these instructions appear to use A format (FRA, FRB, FRT)
instead of the standard X format (RA, RB, RT). Need to verify if this
is needed on Windows or is Linux specific.
Some games persist BGM volume through their own save files but never
call XMPSetVolume. Add a TitlePatch system that intercepts NtReadFile/
NtWriteFile to apply game-specific logic as well as logic to identify
the memory location that the volume data is read into and monitor that
memory location for changes, updating XMP player volume if it changes
during gameplay.
Includes patches for known games with this issue:
Dead or Alive Xtreme 2 (544307D2), PGR4 (4D5307F9) and PGR3 (4D5307D1)
File-scope static initialization order is undefined across translation
units. Code that acquires the global critical region during early startup
could run before the mutex constructor, operating on uninitialized state.
A function-local static is constructed on first use, guaranteeing the
mutex is initialized before any caller can access it.
The decoder gate required a full frame of free blocks (4 mono / 8 stereo +
padding) before allowing a decode pass, but Consume() only writes
subframe_decode_count blocks per call. Games like TGM Ace with
subframe_decode_count=2 on a small ring buffer could never meet the old
threshold, causing a permanent decoder stall.
Gamma stored as R16G16B16A16_UNORM is 64bpp in host memory but
represents a single 32bpp pixel, unlike true 64bpp formats which pack
two 32bpp pixels together. Separate dest_is_gamma_unorm16 from
dest_is_64bpp to avoid incorrect coordinate transforms.
Add output paths for gamma_unorm16 destinations:
- gamma→gamma: direct passthrough (already linear in R16G16B16A16)
- k_8_8_8_8→gamma: convert gamma-encoded source to linear
- Cross-format packed EDRAM: midpoint encoding to survive UNORM16
quantization round-trip through PreSaturatedLinearToPWLGamma
Fixes red bleeding in Halo Reach, image halving in Just Cause 2,
Portal not rendering, and black rock/plant textures in Halo Reach.
The original commit (2eea146b) correctly added kXenosFormat_16_16_EDRAM
case labels and fixed the unpack clamp from -1.0 to -32.0, but
incorrectly changed the resolve pack functions from UNorm to signed EDRAM
encoding. The resolve destination is a regular texture sampled as UNorm
[0,1], not EDRAM. Using EDRAM packing double-encodes the [-32,32]
scaling, crushing all values by ~64x and caused visual issues
particularly in UE3 titles.
The bytecode submitted with the change was actually correct, but the
source had additional incorrect pieces that only manifested when shaders
were rebuilt.
The /all_resources_bound flag alters the generated DXBC bytecode by changing dcl_globalFlags, register allocation and instruction scheduling. This causes rendering artifacts (blue lightning/flash glitches) at scaled draw resolutions. The other FXC flags (/O3, /Qstrip_*, /Gfp) are retained as they do not affect the executable shader bytecode.
