We do this in order to be able to actually fit in all possible values from AmdGpu::NumberConversion.
Fixes gcc compiler warnings:
warning: ‘Shader::PsColorBuffer::num_conversion’ is too small to hold all values of ‘enum class AmdGpu::NumberConversion’
* shader_recompiler: Add lowering pass for when 64-bit float is unsupported.
* shader_recompiler: Fix PackDouble2x32/UnpackDouble2x32 type.
* shader_recompiler: Remove extra bit cast implementations.
* Implementing DS_SUB_U32, DS_INC_U32, DS_DEC_U32, DS_WRITE_SRC2_B32, DS_WRITE_SRC2_B64.
* Added ir instructions for new opcodes.
Removing Write implementations.
Maping operation S_BFE_I32 as it was added in translate but wasnt pointing to anything.
* Suggestions
* control_flow_graph: Improve divergence handling
* recompiler: Simplify optimization passes
Removes a redudant constant propagation and cleans up the passes a little
* ir_passes: Add new readlane elimination pass
The algorithm has grown complex enough where it deserves its own pass. The old implementation could only handle a single phi level properly,
however this one should be able to eliminate vast majority of lane cases remaining. It first performs a traversal of the phi tree to ensure
that all phi sources can be rewritten into an expected value and then performs elimintation by recursively duplicating the phi nodes at each step,
in order to preserve control flow.
* clang format
* control_flow_graph: Remove debug code
* Specialize vertex attributes on dst_sel
* compare vs attrib specs by default, ignore NumberFmt when vertex input dynamic state is supported
* specialize data_format when attribute uses step rates
* use num_components in data fmt instead of fmt itself
-dont do trivial phi removal during SRT pass, that's now done in
ssa_rewrite
-remove unused variable when walking tess attributes
-fix some tess comments
* ir_passes: Add barrier at end of block too
* vk_platform: Always assign names to resources
* texture_cache: Better overlap handling
* liverpool: Avoid resuming ce_task when its finished
* spirv_quad_rect: Skip default attributes
Fixes some crashes
* memory: Improve buffer size clamping
* liverpool: Relax binary header validity check
* liverpool: Stub SetPredication with a warning
* Better than outright crash
* emit_spirv: Implement round to zero mode
* liverpool: queue::pop takes the front element
* image_info: Remove obsolete assert
The old code assumed the mip only had 1 layer thus a right overlap could not return mip 0. But with the new path we handle images that are both mip-mapped and multi-layer, thus this can happen
* tile_manager: Fix size calculation
* spirv_quad_rect: Skip default attributes
---------
Co-authored-by: poly <47796739+polybiusproxy@users.noreply.github.com>
Co-authored-by: squidbus <175574877+squidbus@users.noreply.github.com>
* shader_recompiler: Implement AMD buffer bounds checking behavior.
* shader_recompiler: Use SRT flatbuf for bounds check size.
* shader_recompiler: Fix buffer atomic bounds check.
* buffer_cache: Prevent false image-to-buffer sync.
Lowering vertex fetch to formatted buffer surfaced an issue where a CPU modified range may be overwritten with stale GPU modified image data.
* Address review comments.
* shader_recompiler: Move shared mem lowering into emitter
* IR can be quite verbose during first stages of translation, before ssa and constant prop passes have run that drastically simplify it. This lowering can also be done during emission so why not do it then to save some compilation time
* runtime_info: Pack PsColorBuffer into 8 bytes
* Drops the size of the total structure by half from 396 to 204 bytes. Also should make comparison of the array a bit faster, since its a hot path done every draw
* emit_spirv_context: Add infrastructure for buffer aliases
* Splits out the buffer creation function so it can be reused when defining multiple type aliases
* shader_recompiler: Merge srt_flatbuf into buffers list
* Its no longer a special case, yay
* shader_recompiler: Complete buffer aliasing support
* Add a bunch more types into buffers, such as F32 for float reads/writes and 8/16 bit integer types for formatted buffers
* shader_recompiler: Remove existing shared memory emulation
* The current impl relies on backend side implementaton and hooking into every shared memory access. It also doesnt handle atomics. Will be replaced by an IR pass that solves these issues
* shader_recompiler: Reintroduce shared memory on ssbo emulation
* Now it is performed with an IR pass, and combined with the previous commit cleanup, is fully transparent from the backend, other than requiring workgroup_index be provided as an attribute (computing this on every shared memory access is gonna be too verbose
* clang format
* buffer_cache: Reduce buffer sizes
* vk_rasterizer: Cleanup resource binding code
* Reduce noise in the functions, also remove some arguments which are class members
* Fix gcc
the shift value should be extracted from the 5 least significant bits of the second operand (S1.u[4:0]), to ensure that the shift is limited to values from 0 to 31, suitable for 32-bit operations
Instruction S_LSHR_B32
Description D.u = S0.u >> S1.u[4:0]. SCC = 1 if result is non-zero.
* shader_recompiler: Replace texel buffers with in-shader buffer format interpretation
* shader_recompiler: Move 10/11-bit float conversion to functions and address some comments.
* vulkan: Remove VK_KHR_maintenance5 as it is no longer needed for buffer views.
* shader_recompiler: Add helpers for composites and bitfields in pack/unpack.
* shader_recompiler: Use initializer_list for bitfield insert helper.
* shader_recompiler: Account for instruction array flag in image type.
* shader_recompiler: Check da flag for all mimg instructions.
* shader_recompiler: Convert cube images into 2D arrays.
* shader_recompiler: Move image resource functions into sharp type.
* shader_recompiler: Use native AMD cube instructions when possible.
* specialization: Fix buffer storage mistake.
* video_core: Implement conversion for uncommon/unsupported number formats.
* shader_recompiler: Reinterpret image sample output as well.
* liverpool_to_vk: Remove mappings for remapped number formats.
These were poorly supported by drivers anyway.
* resource_tracking_pass: Fix image write swizzle mistake.
* amdgpu: Add missing specialization and move format mapping data to types
* reinterpret: Fix U/SToF input type.
* shader_recompiler: Add swizzle support for unsupported formats.
* renderer_vulkan: Rework MRT swizzles and add unsupported format swizzle support.
* shader_recompiler: Clean up swizzle handling and handle ImageRead storage swizzle.
* shader_recompiler: Fix type errors
* liverpool_to_vk: Remove redundant clear color swizzles.
* shader_recompiler: Reduce CompositeConstruct to constants where possible.
* shader_recompiler: Fix ImageRead/Write and StoreBufferFormatF32 types.
* amdgpu: Add a few more unsupported format remaps.
* Implement IMAGE_GATHER4_O
Used by The Last of Us Remastered.
* Fix type on IMAGE_GATHER4_C_LZ
Had a different set of types compared to the other IMAGE_GATHER4 ops.
* IMAGE_GATHER4
* Fixes artifacts in TLG when using NVIDIA gpus. When LDS is written and read in the same basic block, the barrier pass wont handle it properly, so insert a barrier before reads
* ir: Add heuristic based LDS barrier pass
* Attempts to insert barriers after zero-depth divergant conditional blocks in shaders that use shared memory
* lds_barriers: Limit to nvidia
* Intel has historically had problems with cs barriers, will debug other time
* Add ExecLo case to S_SAVEEXEC_B64
Seen in CUSA38209
* S_BCNT1_I32_B32
Turtle said our implementation of S_BCNT1_I32_B64 was meant to be for S_BCNT1_I32_B32, so renaming the opcode is the fix.
* coroutine code prettification
* asc queues submission refactoring
* better asc ring context handling
* final touches and review notes
* even more simplification for context saving
