Mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-05-25 04:45:00 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

shader_recompiler: Implement most integer image atomics, workgroup barriers and shared memory load/store (#231)

Browse source 
* shader_recompiler: Add LDEXP

* shader_recompiler: Add most image integer atomic ops

* shader_recompiler: Implement shared memory load/store

* shader_recompiler: More image atomics

* externals: Update sirit

* clang format

* cmake: Add missing files

* shader_recompiler: Fix some atomic bugs

* shader_recompiler: Vs outputs

* shader_recompiler: Shared mem has side-effects, fix format component order

* shader_recompiler: Inline constant buffer impl

* video_core: Fix regressions

* Work

* Fixup a few things
This commit is contained in:
TheTurtle 2024-07-05 00:15:44 +03:00 • committed by GitHub
parent af3bbc33e9
commit 6ceab6dfac
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
69 changed files with 1597 additions and 310 deletions
Download patch file Download diff file Expand all files Collapse all files

39
src/shader_recompiler/ir/opcodes.inc
View file

@ -19,6 +19,25 @@ OPCODE(ReadConst, U32, U32x
OPCODE(ReadConstBuffer, F32, Opaque, U32, )
OPCODE(ReadConstBufferU32, U32, Opaque, U32, )
// Barriers
OPCODE(Barrier, Void, )
OPCODE(WorkgroupMemoryBarrier, Void, )
OPCODE(DeviceMemoryBarrier, Void, )
// Shared memory operations
OPCODE(LoadSharedU8, U32, U32, )
OPCODE(LoadSharedS8, U32, U32, )
OPCODE(LoadSharedU16, U32, U32, )
OPCODE(LoadSharedS16, U32, U32, )
OPCODE(LoadSharedU32, U32, U32, )
OPCODE(LoadSharedU64, U32x2, U32, )
OPCODE(LoadSharedU128, U32x4, U32, )
OPCODE(WriteSharedU8, Void, U32, U32, )
OPCODE(WriteSharedU16, Void, U32, U32, )
OPCODE(WriteSharedU32, Void, U32, U32, )
OPCODE(WriteSharedU64, Void, U32, U32x2, )
OPCODE(WriteSharedU128, Void, U32, U32x4, )
// Context getters/setters
OPCODE(GetUserData, U32, ScalarReg, )
OPCODE(GetThreadBitScalarReg, U1, ScalarReg, )
@ -37,11 +56,13 @@ OPCODE(SetAttribute, Void, Attr
OPCODE(GetScc, U1, Void, )
OPCODE(GetExec, U1, Void, )
OPCODE(GetVcc, U1, Void, )
OPCODE(GetSccLo, U32, Void, )
OPCODE(GetVccLo, U32, Void, )
OPCODE(GetVccHi, U32, Void, )
OPCODE(SetScc, Void, U1, )
OPCODE(SetExec, Void, U1, )
OPCODE(SetVcc, Void, U1, )
OPCODE(SetSccLo, Void, U32, )
OPCODE(SetVccLo, Void, U32, )
OPCODE(SetVccHi, Void, U32, )
@ -148,6 +169,7 @@ OPCODE(FPRecipSqrt64, F64, F64,
OPCODE(FPSqrt, F32, F32, )
OPCODE(FPSin, F32, F32, )
OPCODE(FPExp2, F32, F32, )
OPCODE(FPLdexp, F32, F32, U32, )
OPCODE(FPCos, F32, F32, )
OPCODE(FPLog2, F32, F32, )
OPCODE(FPSaturate32, F32, F32, )
@ -190,10 +212,13 @@ OPCODE(FPUnordGreaterThanEqual32, U1, F32,
OPCODE(FPUnordGreaterThanEqual64, U1, F64, F64, )
OPCODE(FPIsNan32, U1, F32, )
OPCODE(FPIsNan64, U1, F64, )
OPCODE(FPIsInf32, U1, F32, )
OPCODE(FPIsInf64, U1, F64, )
// Integer operations
OPCODE(IAdd32, U32, U32, U32, )
OPCODE(IAdd64, U64, U64, U64, )
OPCODE(IAddCary32, U32x2, U32, U32, )
OPCODE(ISub32, U32, U32, U32, )
OPCODE(ISub64, U64, U64, U64, )
OPCODE(IMul32, U32, U32, U32, )
@ -258,6 +283,7 @@ OPCODE(ConvertF32U32, F32, U32,
OPCODE(ConvertF64S32, F64, U32, )
OPCODE(ConvertF64U32, F64, U32, )
OPCODE(ConvertF32U16, F32, U16, )
OPCODE(ConvertU16U32, U16, U32, )
// Image operations
OPCODE(ImageSampleImplicitLod, F32x4, Opaque, Opaque, Opaque, Opaque, )
@ -273,6 +299,19 @@ OPCODE(ImageGradient, F32x4, Opaq
OPCODE(ImageRead, U32x4, Opaque, Opaque, )
OPCODE(ImageWrite, Void, Opaque, Opaque, U32x4, )
// Image atomic operations
OPCODE(ImageAtomicIAdd32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicSMin32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicUMin32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicSMax32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicUMax32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicInc32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicDec32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicAnd32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicOr32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicXor32, U32, Opaque, Opaque, U32, )
OPCODE(ImageAtomicExchange32, U32, Opaque, Opaque, U32, )
// Warp operations
OPCODE(LaneId, U32, )
OPCODE(QuadShuffle, U32, U32, U32 )