- If a stale reference is left lying around (e.g the texture bound to
depth has been deleted and we attach a color image) no operations
actually take place. glCheckFramebufferStatus also does not catch this
problem.
- Sometimes program-point-size is enabled, but the vs does not actually
write to the point size register. In this case, pass the incoming point
size along instead of the default register init.
- Separate displayed statistics from actual backend statistics.
Allows asynchronous flipping to work correctly as it just uses display stats.
The real stats are used by the frame scope marker to determine behavior like engaging the FIFO optimizer or skipping draw calls correctly.
- Avoid silly broken tests due to queue_tag being called before pitch is initialized.
- Return actual memory range covered and exclude trailing padding.
- Coordinates in src are to be calculated with src_pitch, not required_pitch.
- This allows creating buffers with no MAP bits set which should ensure they are created for VRAM usage only
- TODO: Implement compute kernels to avoid software fallback mode for pack/unpack operations
- Fix 2D coordinate sampling of W coordinate.
W is actually HPOS.w and not 1. Z is however always 0.
- Optimize register usage a bit
Disassembling compiled SPV shows that global declaration results in less ops than using inout modifiers. Modifiers generate extra mov instructions.
- Fix reading of varying registers in FP
Different registers have different behavior
- Always write to varying registers. If a register is not written to, it is initialized to (0, 0, 0, 1)
- Reimplements two-sided lighting correctly without hacks
- Also bumps shader cache version
- Do not allow offloader to handle its own faults. Serialize them on RSX instead.
This approach introduces a GPU race condition that should be avoided with improved synchronization.
- TODO: Use proper GPU-side synchronization to avoid this situation
- Avoids memory appearing older when used for depth test without depth write
The write_barrier before the call will inherit new data but the tag will not update as no new information is added.
- Properly commit orphaned blocks not invalidating existing cache structures
- Do not ignore overwritten objects when commiting as unprotected fbo. Avoids stale references to invalidated surface objects.
- Load into memory as straightforward BGRA
- Fixes a bug in vulkan caused by byte shuffling in blit engine vs shader access
- Removes the need for memory shuffling when transferring into a rendertarget
- Implements render target data load (aka Read Color Buffer/Read Depth Buffer)
- Refactors vulkan surface barrier to be much cleaner.
- Removes redundant surface barrier invocations after doing a merged load
from surface cache.
- Adds explicit access modes when gathering surfaces from cache.
- Further improve aliased data preservation by unconditionally scanning.
Its is possible for cache aliasing to occur when doing memory split.
- Also sets up for RCB/RDB implementation
- Merge viewport raster window and scissor into one clipping region
- Viewport raster clip is different from viewport geometry clipping in
hardware as the latter is configurable separately
vm::spu max address was overflowing resulting in issues, so cast to u64 where needed. Fixes#6145.
Use vm::get_addr instead of manually substructing vm::base(0) from pointer in texture cache code.
Prefer std::atomic_thread_fence over _mm_?fence(), adjust usage to be more correct.
Used sequantially consistent ordering in semaphore_release for TSX path as well.
Improved memory ordering for sys_rsx_context_iounmap/map.
Fixed sync bugs in HLE gcm because of not using atomic instructions.
Use release memory barrier in lwsync for PPU LLVM, according to this xbox360 programming guide lwsync is a hw release memory barrier.
Also use release barrier where lwsync was originally used in liblv2 sys_lwmutex and cellSync.
Use acquire barrier for isync instruction, see https://devblogs.microsoft.com/oldnewthing/20180814-00/?p=99485
- When multithreaded RSX is enabled, the vertex cache just lowers performance
- The small cost of upload is paid by the asynchronous thread, allowing RSX to work optimally
- Multiple header files where missing #includes to other headers that
where used in the header. Correct header was included in correct
order in source files which caused everything to compile.
- Added missing #includes so header files correctly include all their
dependencies and fixes problems with IDEs being unable to parse
headers correctly due to missing symbols
- Removes a lot of wm_event code that was used to perform window management and is no longer needed.
- Significantly simplifies the vulkan code.
- Implements resource management when vulkan window is minimized to allow resources to be freed.
- Allows render targets to behave like stacked 3D views same as shader inputs are resolved
- Basically implements most of 'Read Color/Depth Buffers" option for 'free'.
- Allows splitting RTV/DSV resources if they are superceded by a partial surface
- Also allows intersecting new resources through the surface cache for proper inheritance from other scattered data
- TODO: Refactor bind_surface_as_rtt and bind_surface_as_ds to reduce asinine code duplication
- The hw generates inaccurate values when doing perspective-correct
interpolation of vertex output attributes and makes the comparison (a ==
b) fail even when they are a fixed constant value.
- Increase equality tolerance when doing comparisons in fragment
shaders for NV cards only to work around this issue.
- Teepo fix
- The fixed-point D24S8 format does special Z clamping during compare which matches PS3 behaviour
- D32S8 is a floating point format and comparison with Dref > 1 always fails causing black edges/borders
- Improve support for float16_t by minimizing mixed inputs to functions
(ambiguous overloads)
- Minimize amount of downcasts in code by using opcode flags
- Re-enable float16_t support for vulkan
- Emulating f16 with f32 is not ideal and requires a lot of value clamping
- Using native data type can significantly improve performance and accuracy
- With openGL, check for the compatible extensions NV_gpu_shader5 and
AMD_gpu_shader_half_float
- With Vulkan, enable this functionality in the deviceFeatures if
applicable. (VK_KHR_shader_float16_int8 extension)
- Temporarily disable hw fp16 for vulkan
- Removes CPU-only transforms that broke GPU-side code.
-- Channels in GPU compute are laid out in cell-order, but CPU was uploading in favorable order and compensating with swizzles.
-- This leads to 2 different layouts depending on the location of the data (CPU vs GPU)
- Implement R8G8_R8B8 interleaved format decode
- General improvements
- Also refactors some bpp handling code
- Simplify texture intersection test to use a normalized/uniform coordinate space
- Fix broken bounds checking as well
- Batch dma transfers whenever possible and do them in one go
- vk: Always ensure that queued dma transfers are visible to the GPU before they are needed by the host
Requires a little refactoring to allow proper communication of the commandbuffer state
- vk: Code cleanup, the simplified mechanism makes it so that its not necessary to pass tons of args to methods
- vk: Fixup - do not forcefully do dma transfers on sections in an invalidation zone! They may have been speculated correctly already
- Properly wait for the buffer transfer operation to finish before map/readback!
- Change vkFence to vkEvent which works more like a GL fence which is what is needed.
- Implement supporting methods and functions
- Do not destroy fence by immediately waiting after copying to dma buffer
- Blit operations do format conversion automatically which is NOT what we want!
- Scale onto temp buffer with similar format before performing data cast.
- gl: Properly initialize and manage sampler states
- gl/vk: Snap overlay elements to pixel grid by aligning to pixel centers
- overlays: Disable grid snapping in stb since its now handled in the backend
- Disable DEPTH<->RGBA typeless transfers for now as they require a lot more work to work for all vendors
- Do not allow switching layouts to UNDEFINED/PREINITIALIZED formats
- Apply dither to edges that almost fail the straight-up alpha test
- Significantly improves alpha tested geometry far from the camera
- Also removes blend factor overrides/hacks as they give incorrect results due to background bleeding
- Index offset is ignored anyway and only used to calculate vertex attribute divisor index
- Specialized optimization for untouched xfer without primitive restart
- Avoid tagging and rely on read/write barriers and the dirty flag mechanism. Testing is done with a weak 8-byte memory test
- Introducing new data when tagging breaks applications with race conditions where tags can overwrite flushed data
- D24S8 targets have 2 aspects that are dealt with separately; Forcefully initialize the remaining data if a partial init is done. Its 'free' anyway
- It seems that the stencil mask matters when clearing unlike the depth mask and color mask
- gl: Include an execution state wrapper to ensure state changes are consistent. Also removes a lot of required 'cleanup' for helper methods
- texture_cache: Make execition context a mandatory field as it is required for all operations. Also removes a lot of situations where duplicate argument is added in for both fixed and vararg fields
- Explicit read/write barrier for framebuffer resources depending on
usage. Allows for operations like optional memory initialization before
reading
- Implicitly invoke a memory barrier if actively reading from an unsynchronized texture
- Simplify memory transfer operations
- Should allow more games to work without strict mode
- Do not bind companion framebuffer when clearing single aspect; let the
contest mechanism sort it out instead
- Do not prematurely tag framebuffers, instead only do so at
write-confirmation time. Should avoid false tagging if setup does not
allow a render to occur.
- Implements a mirror view of D24S8 data that accesses the stencil components.
Finishes the implementation of TEX2D_DEPTH_RGBA as the stencil component was previously missing from the reconstructed data
- Add a few missing destructors
Image classes are inherited a lot and I forgot to make the dtors virtual
Fixes VRAM leaks and incorrect destruction of resources, which could
lead to drivers crashes.
Additionally, lock_memory_region is now able to flush superseded
sections. However, due to the potential performance impact of this
for little gain, a new debug setting ("Strict Flushing") has been
added to config.yaml
- Ignore barriers inserted after BEGIN but before any draw commands are emitted
- Properly process tail barriers inserted before END but after draw commands are submitted
- Ignore execution barriers with no effect (same register value written)
- Tries to detect when FIFO preprocessing is beneficial and only enables optimizations if the benefit outweighs the cost
- Current threshold is at least 500 draw calls saved at over 2000 draw calls to justify the overhead
- TODO: More tuning for other CPUs
- Improve vertex attribute layout format. Allows for full 16-bit attribute divisor
- Use actual pitch when declaring framebuffer rsx pitch instead of register value in case of swizzle? rendering
- Replace a few more vectors with simple_array<T>
- Avoid unnecessary string comparisons in backends. We already know referenced textures from the program analysers!
- Also fix visual corruption when using disjoint indexed draws
- Refactor draw call emit again (vk)
- Improve execution barrier resolve
- Allow vertex/index rebase inside begin/end pair
- Add ALPHA_TEST to list of excluded methods [TODO: defer raster state]
- gl bringup
- Simplify
- using the simple_array gets back a few more fps :)
- Matching attachments with resource id fails because drivers are reusing
handles!
- Properly sets up stale fbo ref counting and removal
- Properly sets up resource reference test with subsequent removal to
avoid using a broken fbo entry
- Orders flushing to preserve memory at all cost
- Avoids false positive where flushing overlapping sections can falsely invalidate another with head/tail test
- Forcefully downloads and reuploads data from the CPU in case of unexpected overlaps
- Properly detect correct size of newly created blit targets
- Remember to clear any existing views when changing the default component map!
* CMake: Refactor build to multiple libraries
- Refactor CMake build system by creating separate libraries for
different components
- Create interface libraries for most dependencies and add 3rdparty::*
ALIAS targets for ease of use and use them to try specifying correct
dependencies for each target
- Prefer 3rdparty:: ALIAS when linking dependencies
- Exclude xxHash subdirectory from ALL build target
- Add USE_SYSTEM_ZLIB option to select between using included ZLib and
the ZLib in CMake search path
* Add cstring include to Log.cpp
* CMake: Add 3rdparty::glew interface target
* Add Visual Studio CMakeSettings.json to gitignore
* CMake: Move building and finding LLVM to 3rdparty/llvm.cmake script
- LLVM is now built under 3rdparty/ directory in the binary directory
* CMake: Move finding Qt5 to 3rdparty/qt5.cmake script
- Script has to be included in rpcs3/CMakeLists.txt because it defines
Qt5::moc target which isn't available in that folder if it is
included in 3rdparty directory
- Set AUTOMOC and AUTOUIC properties for targets requiring them (rpcs3
and rpcs3_ui) instead of setting CMAKE_AUTOMOC and CMAKE_AUTOUIC so
those properties are not defined for all targets under rpcs3 dir
* CMake: Remove redundant code from rpcs3/CMakeLists.txt
* CMake: Add BUILD_LLVM_SUBMODULE option instead of hardcoded check
- Add BUILD_LLVM_SUBMODULE option (defaults to ON) to allow controlling
usage of the LLVM submodule.
- Move option definitions to root CMakeLists
* CMake: Remove separate Emu subtargets
- Based on discussion in pull request #5032, I decided to combine
subtargets under Emu folder back to a single rpcs3_emu target
* CMake: Remove utilities, loader and crypto targets: merge them to Emu
- Removed separate targets and merged them into rpcs3_emu target as
recommended in pull request (#5032) conversations. Separating targets
probably later in a separate pull request
* Fix relative includes in pad_thread.cpp
* Fix Travis-CI cloning all submodules needlessly
- Retag resources reprotected under flush_always rules
- Properly check for blit resource fitting taking into account format
mismatch, pitch mismatch and typeless transfers
- Do not do a full sync on a texture read barrier
- Avoid calling zcull sync in FIFO spin wait
- Do not flush memory to cache from the renderer side; this method is now obsolete
- Defer compilation process to worker threads
- vulkan: Fixup for graphics_pipeline_state.
Never use struct assignment operator on vk** structs due to padding after sType member (4 bytes)
- Adds proper support for vertex textures, including dimensions other than 2D textures
- Minor analyser fixup, removes spurious 'analyser failed' errors
- Minor optimizations for program state tracking
- Adds dead code elimination
- Fix absolute branch target addresses to take base address into account
- Patch branch targets relative to base address to improve hash matching
- Bumps shader cache version
- Enables shader logging option to write out vertex program binary,
helpful when debugging problems.
- Avoid re-locking memory if there is no reason to do so (no draws issued)
- Actively bound regions should always get written to the backing cache
- Forcefully read memory during download if writes to the target have occured since last sync event
- Region pitch of 64 (disabled) can be used to indicate packed contents - do not assume it is the actual pitch!
- Also fixes interaction of AA factors with lockable_region size
- Use names for overlay command config and vertex data instead of std::pair.
- Make a couple of compiled_resource constructors explicitly named functions.
- Removes the old depth scaling using an overlay.
It was never going to work properly due to per-pixel stencil writes being unavailable
- TODO: Preserve stencil buffer during ARGB8->D32S8 shader conversion pass
- Some applications (e.g Backbreaker) use an evil hack to resolve MSAA.
The application respecifies a formerly AA region as a region with no AA then performs a framebuffer feedback lookup.
The old memory keeps AA during read, but writes back to itself with AA resolved.
This is evil on several levels but it just happens to work on PS3
- The benefits of FIFO optimizations are huge in some cases.
The optimizations also do not break any tested applications so no need to disable with strict mode
- A debug option is provided to disable this behaviour for testing
- Improves cleanup code to consist of 2 parts, remove then dispose. Remove
does not deallocate the item until dispose is called on it, allowing the
backends to first deallocate external references.
- Caller is responsible for managing list locking and tracking disposable list
of items when external references have been cleaned up before using
dispose method.
1. rsx: Rework section synchronization using the new memory mirrors
2. rsx: Tweaks
- Simplify peeking into the current rsx::thread instance.
Use a simple rsx::get_current_renderer instead of asking fxm for the same
- Fix global rsx super memory shm block management
3. rsx: Improve memory validation. test_framebuffer() and
tag_framebuffer() are simplified due to mirror support
4. rsx: Only write back confirmed memory range to avoid overapproximation errors in blit engine
5. rsx: Explicitly mark clobbered flushable sections as dirty to have them
removed
6. rsx: Cumulative fixes
- Reimplement rsx::buffered_section management routines
- blit engine subsections are not hit-tested against confirmed/committed memory range
Not all applications are 'honest' about region bounds, making the real cpu range useless for blit ops
- Reimplements the AMD workaround using an identity buffer to avoid the performance hit of doing multiple glDrawArrays for every single compiled set
- Reimplements first/count allocation using a scratch buffer to reduce allocation overhead when large number of draw calls is used
- Improve dirty state tracking affecting program state
- vk: Refactor out transform constants upload into a separate channel to avoid if possible
transform data uploads are quite expensive
- Introduces a gpu program analyser step to examine shader contents before attempting compilation or cache search
- Avoids detecting shader as being different because of unused textures having state changes
- Adds better program size detection for vertex programs
- Improved vertex program decompiler
- Properly support CAL type instructions
- Support jumping over instructions marked with a termination marker with BRA/CAL class opcodes
- Fix SRC checks and abort
- Fix CC register initialization
- NOTE: Even unused SRC registers have to be valid (usually referencing in.POS)
- Readback does not work at all with float textures on AMD openGL
Driver throws a bogus OUT_OF_MEMORY error regardless of amount of VRAM and system RAM available
- vk: Clear dirty textures before copying 'old contents' in case the old data does not fill the new region
- rsx: Properly decode border color - seems to be in BGRA format
- vk: better approximation of border color to better choose between the presets
- vk: Individually clear color images outside render pass and without scissor
- vk: Fix renderpass selection for clear overlay pass
- vk: Include scissor region when emulating clear mask
NOTES:
- vk: Completely avoid using vkClearXXXXimage - its 'broken' on nvidia drivers
Spec is vague about the function so its not an actual bug
ClearAttachment is clearly defined as bypassing bound state which works correctly
- TODO: Implement memory sampling to simulate loading precleared memory if cell used memset to preinitialize the framebuffer
Autoclear depth to 1|255 and color to 0 is hacky!
- gl/vk: Fix subresource copy/blit
- gl/vk: Fix default_component_map reading
- vk: Reimplement cell readback path and improve software channel decoder
- Properly name the subresource layout field - its in blocks not bytes!
- Implement d24s8 upload from memory correctly
- Do not ignore DEPTH_FLOAT textures - they are depth textures and abide by the depth compare rules
- NOTE: Redirection of 16-bit textures is not implemented yet
Primary:
- Fix SET_SURFACE_CLEAR channel mask - it has been wrong for all these
years! Layout is RGBA not ARGB/BGRA like other registers
Other Fixes:
- vk: Implement subchannel clears using overla pass
- vk: Simplify and clean up state management
- gl: Fix nullptr deref in case of failed subresource copy
- vk/gl: Ignore float buffer clears as hardware seems to do
- Ignore unlocked blit sections [TODO]
- Do not attempt blit on hw if bytesize is unsupported
- gl: Implement typeless memory transfers
Uses pbo to handle type-agnostic memory transfer
- Mainly affected are colormasks and read swizzles
NOTES:
- Writes to G write to the second and fourth component (YW)
- Writes to B write to first and third component (XZ)
- This means the actual format layout is BGBG (RGBA) making RG mapping actually GR
- Clear does not seem to have any intended effect on this format (TLOU)
Fixes the following issues on Tales of Vesperia which requires SRM.
- Blacked out scene after the sleeping dog now renders correctly
- Ghosting effect. The ghosting was most noticeable as a delay between the character rendering and the cell shading around the character. This appears to be gone with this change.
- GL queries share the target binding (not asynchronous!)
- Discard active queries by closing them, leave closed queries alone (nothing to be done for discard op)
- Reimplements render target views used for sampling
- Optimizes access using an encoded control token
- Adds proper encoding for 24-bit textures (DRGB8 -> ORGB/OBGR)
- Adds proper encoding for ABGR textures (ABGR8 -> ARGB8)
- Silence some compiler warnings as well
- TODO: Real texture views for OGL current method is a hack
- Separate TXB from TXL: They are completely different!
- Properly perform TMU emulation in the fragment shader. Implemens SRGB conversion and alphakill at the moment
- Properly perform ROP emulation in the fragment shader. Implements FRAMEBUFFER_SRGB. While support on the chip looks to be incomplete (and wierd), it does work
- Document some more bits in SHADER_CONTROL register
gl/vk: Bump shader cache version
gl/vk: Disable anisotropic override when strict mode enabled as it is proven to alter some games negatively
gl: Clamp buffer view range to not exceed the backing buffer size. Also add assert for the same condition
- Track asynchronous operations in RSX core
- Add read barriers to force pending writes to finish.
Fixes zcull delay flicker in all UE3 titles without forcing hard stall
- Increase zcull latency as all writes should be synchronized now
- ZCULL unit emulation rewritten
- ZCULL reports are now deferred avoiding pipeline stalls
- Minor optimizations; replaced std::mutex with shared_mutex where contention is rare
- Silence unnecessary error message
- Small improvement to out of memory handling for vulkan and slightly bump vertex buffer heap
According to the NV_fragment_program spec, its not feasible to have 16-bit depth wries
NOTE: NV_fragement_program precedes NV_fragment_program2 which is very
close to what RSX consumes. It is hardware from that era afterall
- Forces Bitcast of texture data if input format cannot possibly be the
same as the existing texture format
- rsx: Other minor improvements to texture cache :-
- remove obsolete blit engine incompatibility warning. The texture will be re-uploaded if it is indeed incompatible
- Implement warn_once and err_once to avoid spamming the log with systemic errors
- Track mispredicted flushes
- Reswizzle bitcasted texture data to native layout
TODO: Also needs reshuffle according to input remap vector
- A new step is added between decompilation and pipeline object creation allowing for properties to be updated based on shader contents
- Allos masking off attachment writes that are unmodified in the shader
- gl: Do not call makeCurrent every flip - it is already called in set_current()
- gl: Improve ring buffer behaviour; use sliding window to view buffers larger than maximum viewable hardware range
NV hardware can only view 128M at a time
- gl/vk: Bump transform constant heap size When lots of draw calls are issued, the heap is exhaused very fast (8k per draw)
- gl: Remove CLIENT_STORAGE_BIT from ring buffers. Performance is marginally better without this flag (at least on windows)
- Identify depth textures reaching the gpu via shader_read upload path
- Use correct timestamp counter for opengl
- inline draw_state::test_property because msvc doesnt do it for us
