2 Commits

Author	SHA1	Message	Date
chss95cs@gmail.com	d372d8d5e3	nasty commit with a bunch of test code left in, will clean up and pr ... Remove the logger_ != nullptr check from shouldlog, it will nearly always be true except on initialization and gets checked later anyway, this shrinks the size of the generated code for some Select specialized vastcpy for current cpu, for now only have paths for MOVDIR64B and generic avx1 Add XE_UNLIKELY/LIKELY if, they map better to the c++ unlikely/likely attributes which we will need to use soon Finished reimplementing STVL/STVR/LVL/LVR as their own opcodes. we now generate far less code for these instructions. this also means optimization passes can be written to simplify/remove/replace these instructions in some cases. Found that a good deal of the X86 we were emitting for these instructions was dead code or redundant. the reduction in generated HIR/x86 should help a lot with compilation times and make function precompilation more feasible as a default Don't static assert in default prefetch impl, in c++20 the assertion will be triggered even without an instantiation Reorder some if/else to prod msvc into ordering the branches optimally. it somewhat worked... Added some notes about which opcodes should be removed/refactored Dispatch in WriteRegister via vector compares for the bounds. still not very optimal, we ought to be checking whether any register in a range may be special A lot of work on trying to optimize writeregister, moved wraparound path into a noinline function based on profiling info Hoist the IsUcodeAnalyzed check out of AnalyzeShader, instead check it before each call. Profiler recorded many hits in the stack frame setup of the function, but none in the actual body of it, so the check is often true but the stack frame setup is run unconditionally Pre-check whether we're about to write a single register from a ring Replace more jump tables from draw_util/texture_info with popcnt based sparse indexing/bit tables/shuffle lookups Place the GPU register file on its own VAD/virtual allocation, it is no longer a member of graphics system	2022-09-04 22:42:51 -05:00
chss95cs@gmail.com	f31869092c	Fixed a bug with readback_resolve and readback_memexport that was responsible for a large portion of their overhead. readback_memexport and resolve are now usable for games, depending on your hardware. in my case games that were slideshows now run at like 20-30 fps, and my hardware isnt the best for xenia. ... 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	2022-08-28 14:24:25 -07:00