| Age | Commit message (Collapse) | Author |
|---|
|
Add support for missing errno values needed by TOTK:
- Add BUSY (16) for "Device or resource busy" errors
- Add NOTSOCK (88) for "Socket operation on non-socket" errors
Improvements:
- Update TranslateNativeError on both Windows and Unix to handle new error codes
- Change socket error logging for NOTSOCK from WARNING to DEBUG level
- Fix formatting in Unix errno translation code
- Update shader storage buffer tracking range to accommodate TOTK buffers
- Add hex format to storage buffer logging for easier comparison with bias range
- Change storage buffer tracking log level from WARNING to DEBUG
These changes help prevent error messages in games
that use network features not fully implemented in the emulator yet.
Signed-off-by: Zephyron <zephyron@citron-emu.org>
|
|
Improve GPU storage buffer detection and memory access patterns:
- Expand NVN bias address range (0x100-0x800 vs 0x110-0x610)
- Increase alignment from 16 to 32 bytes for optimal memory access
- Raise default alignment from 8 to 16 bytes for non-biased addresses
- Refactor bias handling code for better readability
- Add detailed performance-related comments
These changes help identify more storage buffers within shaders and
ensure memory accesses are better aligned, which improves overall
shader compilation and execution performance.
Signed-off-by: Zephyron <zephyron@citron-emu.org>
|
|
Add proper handling of input topologies in the Geometry stage for all three
shader backends (GLASM, GLSL, SPIRV). This implementation uses a lookup table
approach to determine vertex counts based on input topology type (Points,
Lines, LinesAdjacency, Triangles, TrianglesAdjacency) and shifts the vertex
count by 16 bits as required by the invocation info format.
Additional changes:
- Fixed TessellationControl and TessellationEval stages to properly break
after emitting code
- Added proper header include for runtime_info.h in GLASM backend
- Improved code documentation with clear commenting patterns
This change ensures accurate geometry shader behavior across all backends,
improving compatibility with games that rely on proper vertex count reporting.
Signed-off-by: Zephyron <zephyron@citron-emu.org>
|
|
Update fmt library to version 11.0.2 and make necessary adjustments:
- Replace fmt/format.h includes with fmt/ranges.h
- Add const qualifiers to formatter::format functions
- Update CMake to require fmt version 11
Additional dependency updates:
- Update SDL2 bundled version from 2.28.2 to 2.32.0
- Update catch2 to version 3.7.1
- Update vcpkg baseline to ca846b21276c9a3171074ac8d2b4f6516894a7d0
|
|
- Add proper invocation info handling for tessellation and fragment stages
- Return patch vertices info shifted by 16 bits for tessellation stages
- Return sample mask shifted by 16 bits for fragment stage
- Return standard format (0x00ff0000) for compute and other stages
- Implement consistently across SPIRV, GLSL, and GLASM backends
- Remove stubbed warning message
|
|
Add stub implementations for previously unhandled flow test conditions in the
shader recompiler's IR emitter. These conditions were previously throwing
"Not Implemented" exceptions when encountered.
The following flow test cases are now stubbed:
- FCSM_TR (Fragment Shader Coarse/Fine Mode Test and Reject)
- CSM_TA (Coarse/Fine Mode Test Accept)
- CSM_TR (Coarse/Fine Mode Test and Reject)
- CSM_MX (Coarse/Fine Mode Maximum)
- FCSM_TA (Fragment Shader Coarse/Fine Mode Test Accept)
- FCSM_MX (Fragment Shader Coarse/Fine Mode Maximum)
Currently these stubs:
1. Return false (ir.Imm1(false)) as a placeholder value
2. Log a warning message indicating the stub
3. Allow shaders using these conditions to compile rather than fail
This is a step toward proper implementation of coarse/fine mode shader
operations. The stubs prevent crashes while making it clear which code paths
need full implementation.
Technical notes:
- Removed the previous FCSM_TR implementation that used flag operations
- Added consistent warning messages for tracking stub usage
- Kept within the existing GetFlowTest switch statement structure
|
|
- Simplify ISBERD instruction to handle register-to-register moves
- Remove incorrect CompositeConstruct usage
- Replace with direct register value passing
- Fix compilation errors in internal stage buffer handling
|
|
Simplifies the negative reciprocal calculation in FSWZADD by using the
dedicated FPRecip operation instead of manually constructing a division.
This change:
- Replaces FPDiv(Imm32(f32(1.0f)), src_b) with FPRecip(src_b)
- Results in more efficient code for calculating 1.0/x
- Fixes build errors from undefined IR emitter methods
|
|
Fixes incorrect method names in the floating point swizzled add implementation:
- FNeg -> FPNeg
- FDiv -> FPDiv
- FImm32(1.0f) -> Imm32(ir.f32(1.0f))
These changes align with the correct IR emitter interface naming conventions,
where floating-point operations use the 'FP' prefix and immediate values are
properly constructed using f32().
|
|
Implements the Internal Stage Buffer Entry Read (ISBERD) instruction in the
Maxwell shader recompiler. This replaces the previous stubbed implementation
with actual buffer reading functionality.
The implementation:
- Validates unsupported features (skew, o, mode, shift)
- Performs buffer read using IR::InternalStageBufferRead
- Stores the read value to the destination register
This removes the "(STUBBED) called" warning messages that were previously
being logged during shader compilation.
|
|
Implements the ISBERD (Internal Stage Buffer Entry Read) instruction's
mode and shift options that were previously throwing NotImplemented
exceptions. This includes:
- Patch mode for reading patch data
- Prim mode for reading primitive data
- Attr mode for reading attribute data
- U16 shift for 16-bit unsigned values
- B32 shift for 32-bit values
The implementation follows Maxwell's ISA specification for handling
different buffer read modes and data shifts.
|
|
- Return to original NVN bias values which better cover the full range of storage buffer accesses:
- offset_begin: 0x110 -> 0x100
- offset_end: 0x610 -> 0x700
The wider range ensures we catch all valid storage buffer accesses
while maintaining the same alignment requirements.
ref: https://git.citron-emu.org/Citron/Citron/commit/2564e721df8b11e81df66c3fd5b04ae760e9550b
|
|
- Update EmitInvocationInfo in SPIRV backend to handle geometry shader stage
ref: https://git.citron-emu.org/Citron/Citron/commit/1266fad68128b3779dcd9bbb6c5f77ec1055e2e6#diff-4cbf73a98ed4ad649a4a6f38b569bc31fcf68cf7
|
|
|
|
|
|
|
|
|
|
Reviewed-on: http://y2nlvhmmk5jnsvechppxnbyzmmv3vbl7dvzn6ltwcdbpgxixp3clkgqd.onion/darktux/torzu/pulls/10
Co-authored-by: darktux <darktux@y2nlvhmmk5jnsvechppxnbyzmmv3vbl7dvzn6ltwcdbpgxixp3clkgqd.onion>
Co-committed-by: darktux <darktux@y2nlvhmmk5jnsvechppxnbyzmmv3vbl7dvzn6ltwcdbpgxixp3clkgqd.onion>
|
|
Pulled in and cleaned up from Sudachi
Reviewed-on: http://y2nlvhmmk5jnsvechppxnbyzmmv3vbl7dvzn6ltwcdbpgxixp3clkgqd.onion/darktux/torzu/pulls/5
Co-authored-by: darktux <darktux@y2nlvhmmk5jnsvechppxnbyzmmv3vbl7dvzn6ltwcdbpgxixp3clkgqd.onion>
Co-committed-by: darktux <darktux@y2nlvhmmk5jnsvechppxnbyzmmv3vbl7dvzn6ltwcdbpgxixp3clkgqd.onion>
|
|
|
|
shader_recompiler: throw on missing geometry streams in geometry shaders
|
|
|
|
|
|
This reverts commit f296a9ce9a1a144d322d54d4628dba6f8a800cb7.
|
|
|
|
shader_recompiler: fix Offset operand usage for non-OpImage*Gather
|
|
|
|
|
|
|
|
|
|
|
|
OpenGL: Fixes and workaround updates for AMD
|
|
|
|
|
|
shader_recompiler: ensure derivatives for textureGrad are f32
|
|
|
|
|
|
|
|
|
|
|
|
|
|
GLSL: Use known cbuf sizes when possible
|
|
|
|
|
|
|
|
shader_recompiler: add byteswap pattern workaround for Nvidia
|
|
|
|
|
|
|
|
Co-Authored-By: Billy Laws <blaws05@gmail.com>
|
