Drivers and hardware may provide more optimal paths for untyped buffers
compared to typed.
Also, ByteAddressBuffer may be bound via a root descriptor in Direct3D 12,
greatly simplifying the binding logic.
ByteAddressBuffer also doesn't have the 128 * 2^20 element count limit that
typed and structured buffers have, and doesn't require the structure stride
to be specified at resource creation time in Direct3D 11.
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.
Switch between even and odd 16-byte element sequences along X by simply
flipping a bit rather than going to a different resolution-scaled group of
pixels, by increasing the size of the group within the constraints imposed
by tiling.
Reorganized SystemPageFlags for sharedmemory, each field now goes into its own array, the three arrays are page aligned in a single virtual allocation
Refactored sharedmemory a bit, use tzcnt if available when finding ranges (faster on pre-zen4 amd cpus)
Made commandprocessor GetcurrentRingReadcount inline, it was made noinline to match PGO decisions but i think PGO can make extra reg allocation decisions that make this inlining choice yield gains, whereas if we do it manually we lose a tiny bit of performance
Working on a more compact vectorized version of GetScissor to save icache on cmd processor thread
Add WriteRegisterForceinline, will probably end up amending to remove it
add ERMS path for vastcpy
Adding WriteRegistersFromMemCommonSense (name will be changed later), name is because i realized my approach with optimizing writeregisters has been backwards, instead of handling all checks more quickly within the loop that writes the registers, we need a different loop for each range with unique handling. we also manage to hoist a lot of the logic out of the loops
Use 100ns delay for MaybeYield, noticed we often return almost immediately from the syscall so we end up wasting some cpu, instead we give up the cpu for the min waitable time (on my system, this is 0.5 ms)
Added a note about affinity mask/dynamic process affinity updates to threading_win
Add TextureFetchConstantsWritten
add xe::clear_lowest_bit, use it in place of shift-andnot in some bit iteration code
make is_allocated_ and is_enabled_ volatile in xma_context
preallocate avpacket buffer in XMAContext::Setup, the reallocations of the buffer in ffmpeg were showing up on profiles
check is_enabled and is_allocated BEFORE locking an xmacontext. XMA worker was spending most of its time locking and unlocking contexts
Removed XeDMAC, dma:: namespace. It was a bad idea and I couldn't make it work in the end. Kept vastcpy and moved it to the memory namespace instead
Made the rest of global_critical_region's members static. They never needed an instance.
Removed ifdef'ed out code from ring_buffer.h
Added EventInfo struct to threading, added Event::Query to aid with implementing NtQueryEvent.
Removed vector from WaitMultiple, instead use a fixed array of 64 handles that we populate. WaitForMultipleObjects cannot handle more than 64 objects.
Remove XE_MSVC_OPTIMIZE_SMALL() use in x64_sequences, x64 backend is now always size optimized because of premake
Make global_critical_region_ static constexpr in shared_memory.h to get rid of wasteage of 8 bytes (empty class=1byte, +alignment for next member=8)
Move trace-related data to the tail of SharedMemory to keep more important data together
In IssueDraw build an array of fetch constant addresses/sizes, then pre-lock the global lock before doing requestrange for each instead of individually locking within requestrange for each of them
Consistent access specifier protected for pm4_command_processor_declare
Devirtualize WriteOneRegisterFromRing.
Move ExecutePacket and ExecutePrimaryBuffer to pm4_command_buffer_x
Remove many redundant header inclusions access xenia-gpu
Minor microoptimization of ExecutePacketType0
Add TextureCache::RequestTextures for batch invocation of LoadTexturesData
Add TextureCache::LoadTexturesData for reducing the number of times we release and reacquire the global lock.
Ideally you should hold the global lock for as little time as possible, but if you are constantly acquiring and releasing it you are actually more likely to have contention
Add already_locked param to ObjectTable::LookupObject to help with reducing lock acquire/release pairs
Add missing checks to XAudioRegisterRenderDriverClient_entry. this is unlikely to fix anything, it was just an easy thing to do
Add NtQueryEvent system call implementation. I don't actually know of any games that need it.
Instead of using std::vector + push_back in KeWaitForMultipleObjects and xeNtWaitForMultipleObjectsEx use a fixed size array of 64 and track the count. More than 64 objects is not permitted by the kernel. The repeated reallocations from push_back were appearing unusually high on the profiler, but were masked until now by waitformultipleobjects natural overhead
Pre-lock the global lock before looking up each handle for xeNtWaitForMultipleObjectsEx and KeWaitForMultipleObjects.
Pre-lock before looking up the signal and waiter in NtSignalAndWaitForSingleObjectEx
add missing checks to NtWaitForMultipleObjectsEx
Support pre-locking in XObject::GetNativeObject
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
