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.
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.
Adds async_shader_compilation cvar (default true) that forces new
pipelines to be created async from the main thread, skips draws entirely
on D3D12 and on vulkan replaces pixel shader with placeholder until the
real one is ready. Causes some visual artifacts on first load but
greatly reduces load times and stutter.
With render target HLE, directly store linear values as R16G16B16A16_UNORM
without gamma conversion, as this format provides more than enough bits
(need at least 11 per component due to the maximum scale being 2^3 in the
piecewise linear gamma curve) to represent linear values without precision
loss.
This makes blending work correctly in linear space, improving quality of
transparency, lighting passes, and fixing issues such as transparent parts
of impact and footstep decals in 4D5307E6 being bright instead.
The new behavior is enabled by default, as it hugely improves the accuracy
of emulation of this format, that is pretty commonplace in Xbox 360 games,
with likely just a small GPU memory and bandwidth usage increase, compared
to the alternatives that were previously available on the HLE RB path.
It's currently implemented only on Direct3D 12, as most of the current GPU
emulation code is planned to be phased out and redone, and no methods other
than 8-bit with pre-conversion were implemented on Vulkan previously.
To implement on Vulkan later, same conversion as in the Direct3D 12
implementation will need to be done in ownership transfer and resolve
shaders. Currently it's somewhat inconvenient to decouple the conversion
functions in `SpirvShaderTranslator` from an instance of the translator due
to vector constant usage. Later, simpler SPIR-V generation functions may be
added (`spv::Builder` usage in general is overly verbose).
The previously default method (8-bit storage with pre-conversion in shaders
and incorrect blending) can be re-enabled by setting the
"gamma_render_target_as_unorm16" configuration option to `false`. This may
be useful if the game, for instance, switches between 8_8_8_8_GAMMA and
8_8_8_8 formats for the same data frequently, as switching will result in
EDRAM range ownership transfer data copying now. Also, the old path is
preserved for Vulkan devices not supporting R16G16B16A16_UNORM with
blending.
The other workaround that was available previously, replacing the PWL
encoding with host hardware sRGB with linear-space blending in render
target management and in texture fetching, was also inherently inaccurate
in many ways (especially when games have their own PWL encoding math, like
4541080F that displayed incorrect colors on the loading screen), and
required tracking of the encoding needed for ranges in the memory.
The sRGB workaround therefore was deleted in this commit, greatly
simplifying the code in the parts of render target, texture and memory
management and shader generation that were involved in it.
Replace the `SubmissionTracker`s with new `GPUCompletionTimeline`s with a
more unified interface (using a base class), and without the internal logic
for queue ownership transfers since that idea was scrapped during the
development of the `Presenter`.
Also use this fence management logic for GPU emulation, though without
architectural reworks for now, just on the bottom level.
Still very messy, but can be cleaned up in further GPU command processor
and presenter reworks.
Introduce "fast" readback resolve that reads from previous frame's
resolve buffers, quick swap between buffers each frame to avoid
copies so minimal performance impact and mostly correct bahavior.
Checks if previous frame had a buffer for the current resolve and
falls through to the slow path, which also allows to support
"screenshot" features in the games that do that without stalling
on normal resolve operations.
Re-enabled readback_memexport as separate feature, was previously
bundled with readback_resolve (probably not intentionally) and
ensures destination address is writeable to avoid memory access
related crashes and unnecessary work.
readback_resolve cvar changes from bool to string ternary with
"none", "fast" and "full" options, defaulting to the new "fast"
mode.
Largely similar to D3D12 implementation but more simple buffer
management and no mips scaling. Includes both FBO and FSI
rendering paths.
Tested on Linux with 2x2 and 3x3 running smoothly.
Enable portability subset physical device enumeration.
Don't use Vulkan 1.1+ logical devices on Vulkan 1.0 instances due to the
VkApplicationInfo::apiVersion specification.
Make sure all extension dependencies are enabled when creating a device.
Prefer exposing feature support over extension support via the device
interface to avoid causing confusion with regard to promoted extensions
(especially those that required some features as extensions, but had those
features made optional when they were promoted).
Allow creating presentation-only devices, not demanding any optional
features beyond the basic Vulkan 1.0, for use cases such as internal tools
or CPU rendering.
Require the independentBlend feature for GPU emulation as working around is
complicated, while support is almost ubiquitous.
Move the graphics system initialization fatal error message to xenia_main
after attempting to initialize all implementations, for automatic fallback
to other implementations in the future.
Log Vulkan driver info.
Improve Vulkan debug message logging, enabled by default.
Refactor code, with simplified logic for enabling extensions and layers.
VS/PS_NUM_REG is 6-bit on Adreno 200, and games aren't seen using the
bit 7 to indicate that no GPRs are used. It's not clear why Freedreno
configures it this way.
Some texture fetch fields were deprecated or moved during the development
of the Xenos, reflect that in the comments.
Add definitions of the registers configuring the conversion of vertex
positions to fixed-point. Although there isn't much that can be done with
it when emulating using PC GPU APIs, there are some places in Xenia that
wrongly (though sometimes deliberately, for results closer to the behavior
of the host GPU) assume that the conversion works like in Direct3D 10+,
however the Xenos supports only up to 4 subpixel bits rather than 8. The
effects of this difference are largely negligible, though.
Also add more detailed info about register references and differences from
other ATI/AMD GPUs for potential future contributors.
Hopefully prevents some potential #1971-like situations.
WAIT_REG_MEM's implementation also allowed the compiler to load the value
only once, which caused an infinite loop with the other changes in the
commit (even in debug builds), so it's now accessed as volatile. Possibly
it would be even better to replace it with some (acquire/release?) atomic
load/store some day at least for the registers actually seen as
participating in those waits.
Also fixes the endianness being handled only on the first wait iteration in
WAIT_REG_MEM.
Functional changes:
- Enable only actually used features, as drivers may take more optimal
paths when certain features are disabled.
- Support VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE.
- Fix the separateStencilMaskRef check doing the opposite.
- Support shaderRoundingModeRTEFloat32.
- Fix vkGetDeviceBufferMemoryRequirements pointer not passed to the Vulkan
Memory Allocator.
Stylistic changes:
- Move all device extensions, properties and features to one structure,
especially simplifying portability subset feature checks, and also making
it easier to request new extension functionality in the future.
- Remove extension suffixes from usage of promoted extensions.
There's no limit on the number of memory exports in a shader on the real
Xenos, and exports can be done anywhere, including in loops. Now, instead
of deferring the exports to the end of the shader, and assuming that export
allocs are executed only once, Xenia flushes exports when it reaches an
alloc (allocs terminate memory exports on Xenos, as well as individual ALU
instructions with `serialize`, but not handling this case for simplicity,
it's only truly mandatory to flush memory exports before starting a new
one), the end of the shader, or a pixel with outstanding exports is killed.
To know which eM# registers need to be flushed to the memory, traversing
the successors of each exec potentially writing any eM#, and specifying
that certain eM# registers might have potentially been written before each
reached control flow instruction, until a flush point or the end of the
shader is reached.
Also, some games export to sub-32bpp formats. These are now supported via
atomic AND clearing the bits of the dword to replace followed by an atomic
OR inserting the new byte/short.
The "close window" keyboard hotkey (Guide-B) now toggles between loglevel -1 and the loglevel set in your config.
Added LoggerBatch class, which accumulates strings into the threads scratch buffer. This is only intended to be used for very high frequency debug logging. if it exhausts the thread buffer, it just silently stops.
Cleaned nearly 8 years of dust off of the pm4 packet disassembler code, now supports all packets that the command processor supports.
Added extremely verbose logging for gpu register writes. This is not compiled in outside of debug builds, requires LogLevel::Debug and log_guest_driven_gpu_register_written_values = true.
Added full logging of all PM4 packets in the cp. This is not compiled in outside of debug builds, requires LogLevel::Debug and disassemble_pm4.
Piggybacked an implementation of guest callstack backtraces using the stackpoints from enable_host_guest_stack_synchronization. If enable_host_guest_stack_synchronization = false, no backtraces can be obtained.
Added log_ringbuffer_kickoff_initiator_bts. when a thread updates the cp's read pointer, it dumps the backtrace of that thread
Changed the names of the gpu registers CALLBACK_ADDRESS and CALLBACK_CONTEXT to the correct names.
Added a note about CP_PROG_COUNTER
Added CP_RB_WPTR to the gpu register table
Added notes about CP_RB_CNTL and CP_RB_RPTR_ADDR. Both aren't necessary for HLE
Changed name of UNKNOWN_0E00 gpu register to TC_CNTL_STATUS. Games only seem to write 1 to it (L2 invalidate)
Also fixes addressing of MSAA samples 2 and 3 for 64bpp color render targets in the ROV RB implementation on Direct3D 12.
Additionally, with FSI/ROV, alpha test and alpha to coverage are done only if the render target 0 was dynamically written to (according to the Direct3D 9 rules for writing to color render targets, though not sure if they actually apply to the alpha tests on Direct3D 9, but for safety).
There is also some code cleanup for things spotted during the development of the feature.
PM4 buffer handling made a virtual member of commandprocessor, place the implementation/declaration into reusable macro files. this is probably the biggest boost here.
Optimized SET_CONSTANT/ LOAD_CONSTANT pm4 ops based on the register range they start writing at, this was also a nice boost
Expose X64 extension flags to code outside of x64 backend, so we can detect and use things like avx512, xop, avx2, etc in normal code
Add freelists for HIR structures to try to reduce the number of last level cache misses during optimization (currently disabled... fixme later)
Analyzed PGO feedback and reordered branches, uninlined functions, moved code out into different functions based on info from it in the PM4 functions, this gave like a 2% boost at best.
Added support for the db16cyc opcode, which is used often in xb360 spinlocks. before it was just being translated to nop, now on x64 we translate it to _mm_pause but may change that in the future to reduce cpu time wasted
texture util - all our divisors were powers of 2, instead we look up a shift. this made texture scaling slightly faster, more so on intel processors which seem to be worse at int divs. GetGuestTextureLayout is now a little faster, although it is still one of the heaviest functions in the emulator when scaling is on.
xe_unlikely_mutex was not a good choice for the guest clock lock, (running theory) on intel processors another thread may take a significant time to update the clock? maybe because of the uint64 division? really not sure, but switched it to xe_mutex. This fixed audio stutter that i had introduced to 1 or 2 games, fixed performance on that n64 rare game with the monkeys.
Took another crack at DMA implementation, another failure.
Instead of passing as a parameter, keep the ringbuffer reader as the first member of commandprocessor so it can be accessed through this
Added macro for noalias
Applied noalias to Memory::LookupHeap. This reduced the size of the executable by 7 kb.
Reworked kernel shim template, this shaved like 100kb off the exe and eliminated the indirect calls from the shim to the actual implementation. We still unconditionally generate string representations of kernel calls though :(, unless it is kHighFrequency
Add nvapi extensions support, currently unused. Will use CPUVISIBLE memory at some point
Inserted prefetches in a few places based on feedback from vtune.
Add native implementation of SHA int8 if all elements are the same
Vectorized comparisons for SetViewport, SetScissorRect
Vectorized ranged comparisons for WriteRegister
Add XE_MSVC_ASSUME
Move FormatInfo::name out of the structure, instead look up the name in a different table. Debug related data and critical runtime data are best kept apart
Templated UpdateSystemConstantValues based on ROV/RTV and primitive_polygonal
Add ArchFloatMask functions, these are for storing the results of floating point comparisons without doing costly float->int pipeline transfers (vucomiss/setb)
Use floatmasks in UpdateSystemConstantValues for checking if dirty, only transfer to int at end of function.
Instead of dirty |= (x == y) in UpdateSystemConstantValues, now we do dirty_u32 |= (x^y). if any of them are not equal, dirty_u32 will be nz, else if theyre all equal it will be zero. This is more friendly to register renaming and the lack of dependencies on EFLAGS lets the compiler reorder better
Add PrefetchSamplerParameters to D3D12TextureCache
use PrefetchSamplerParameters in UpdateBindings to eliminate cache misses that vtune detected
Add PrefetchTextureBinding to D3D12TextureCache
Prefetch texture bindings to get rid of more misses vtune detected (more accesses out of order with random strides)
Rewrote DMAC, still terrible though and have disabled it for now.
Replace tiny memcmp of 6 U64 in render_target_cache with inline loop, msvc fails to make it a loop and instead does a thunk to their memcmp function, which is optimized for larger sizes
PrefetchTextureBinding in AreActiveTextureSRVKeysUpToDate
Replace memcmp calls for pipelinedescription with handwritten cmp
Directly write some registers that dont have special handling in PM4 functions
Changed EstimateMaxY to try to eliminate mispredictions that vtune was reporting, msvc ended up turning the changed code into a series of blends
in ExecutePacketType3_EVENT_WRITE_EXT, instead of writing extents to an array on the stack and then doing xe_copy_and_swap_16 of the data to its dest, pre-swap each constant and then store those. msvc manages to unroll that into wider stores
stop logging XE_SWAP every time we receive XE_SWAP, stop logging the start and end of each viz query
Prefetch watch nodes in FireWatches based on feedback from vtune
Removed dead code from texture_info.cc
NOINLINE on GpuSwap, PGO builds did it so we should too.
add split_map class for mapping keys to values in a way that optimizes for frequent searches and infrequent insertions/removals
remove jump table implementation of GetColorRenderTargetFormatComponentCount, it was appearing relatively high in profiles. instead pack the component counts into a single 32 bit word, which is indexed by shifting
Add cvar to align all basic blocks to a boundary
Add mmio aware load paths
liberally apply XE_RESTRICT in ringbuffer related code
Removed the IS_TRUE and IS_FALSE opcodes, they were pointless duplicates of COMPARE_EQ/COMPARE_NE and i want to simplify our set of opcodes for future backends
More work on LVSR/LVSL/STVR/STVL opcodes
Optimized X64 translated code emission, now only compute instrkey once
Add code for pre-computing integer division magic numbers
Optimized GetHostViewportInfo a little
Move args for GetHostViewportInfo into a class, cache the result and compare for future queries. moved GetHostViewportInfo far lower on the profile
Add (currently not functional, and very racy) asynchronous memcpy code. will improve it and actually use it in future commits.
Add non-temporal memcpy function for huge page-aligned allocations. Used for copying to shared memory/readback
hoist are_accumulated_render_targets_valid_ check out of loop in render_target_cache already bound check.
Add stosb/movsb code for small constant memcpys/memsets that arent worth the overhead of memcpy/memset
