shader_recompiler: BUFFER_ATOMIC & DS_* Opcodes (#428)

* BUFFER_ATOMIC | DS_MINMAX_U32

- Emission of BufferAtomicU32
- Addition of Buffer opcodes to IR
- Translator for BUFFER_ATOMIC Opcode
- Translators for DS_MAXMIN_U32 Opcodes

* Clang Format & UNREACHABLE_MSG

* clang

* no crash on compile

* clang

* Shared Atomics

* reuse

* rm vscode

* resolve

* opcodes

* side effects

* attempt fix shader comp

* failed attempt to fix

* clang

* do correct vdata set (still fails)

* clang

* fixed BUFFER_ATOMIC_ADD, DS_ADD_U32 fails

* data share should work

* clang

* resource tracking for buffer atomic

* clang

* distinguish RTN opcodes

* clean IsBufferInstruction

---------

Co-authored-by: microsoftv <6063922+microsoftv@users.noreply.github.com>
This commit is contained in:
Lizardy 2024-08-17 15:06:06 -04:00 committed by GitHub
parent 3be2e4b2b8
commit 63938ba8dd
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
11 changed files with 375 additions and 27 deletions

View file

@ -12,6 +12,25 @@ std::pair<Id, Id> AtomicArgs(EmitContext& ctx) {
return {scope, semantics};
}
Id SharedAtomicU32(EmitContext& ctx, Id offset, Id value,
Id (Sirit::Module::*atomic_func)(Id, Id, Id, Id, Id)) {
const Id shift_id{ctx.ConstU32(2U)};
const Id index{ctx.OpShiftRightArithmetic(ctx.U32[1], offset, shift_id)};
const Id pointer{ctx.OpAccessChain(ctx.shared_u32, ctx.shared_memory_u32, index)};
const auto [scope, semantics]{AtomicArgs(ctx)};
return (ctx.*atomic_func)(ctx.U32[1], pointer, scope, semantics, value);
}
Id BufferAtomicU32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value,
Id (Sirit::Module::*atomic_func)(Id, Id, Id, Id, Id)) {
auto& buffer = ctx.buffers[handle];
address = ctx.OpIAdd(ctx.U32[1], address, buffer.offset);
const Id index = ctx.OpShiftRightLogical(ctx.U32[1], address, ctx.ConstU32(2u));
const Id ptr = ctx.OpAccessChain(buffer.pointer_type, buffer.id, ctx.u32_zero_value, index);
const auto [scope, semantics]{AtomicArgs(ctx)};
return (ctx.*atomic_func)(ctx.U32[1], ptr, scope, semantics, value);
}
Id ImageAtomicU32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value,
Id (Sirit::Module::*atomic_func)(Id, Id, Id, Id, Id)) {
const auto& texture = ctx.images[handle & 0xFFFF];
@ -21,6 +40,72 @@ Id ImageAtomicU32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id va
}
} // Anonymous namespace
Id EmitSharedAtomicIAdd32(EmitContext& ctx, Id offset, Id value) {
return SharedAtomicU32(ctx, offset, value, &Sirit::Module::OpAtomicIAdd);
}
Id EmitSharedAtomicUMax32(EmitContext& ctx, Id offset, Id value) {
return SharedAtomicU32(ctx, offset, value, &Sirit::Module::OpAtomicUMax);
}
Id EmitSharedAtomicSMax32(EmitContext& ctx, Id offset, Id value) {
return SharedAtomicU32(ctx, offset, value, &Sirit::Module::OpAtomicSMax);
}
Id EmitSharedAtomicUMin32(EmitContext& ctx, Id offset, Id value) {
return SharedAtomicU32(ctx, offset, value, &Sirit::Module::OpAtomicUMin);
}
Id EmitSharedAtomicSMin32(EmitContext& ctx, Id offset, Id value) {
return SharedAtomicU32(ctx, offset, value, &Sirit::Module::OpAtomicSMin);
}
Id EmitBufferAtomicIAdd32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) {
return BufferAtomicU32(ctx, inst, handle, address, value, &Sirit::Module::OpAtomicIAdd);
}
Id EmitBufferAtomicSMin32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) {
return BufferAtomicU32(ctx, inst, handle, address, value, &Sirit::Module::OpAtomicSMin);
}
Id EmitBufferAtomicUMin32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) {
return BufferAtomicU32(ctx, inst, handle, address, value, &Sirit::Module::OpAtomicUMin);
}
Id EmitBufferAtomicSMax32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) {
return BufferAtomicU32(ctx, inst, handle, address, value, &Sirit::Module::OpAtomicSMax);
}
Id EmitBufferAtomicUMax32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) {
return BufferAtomicU32(ctx, inst, handle, address, value, &Sirit::Module::OpAtomicUMax);
}
Id EmitBufferAtomicInc32(EmitContext&, IR::Inst*, u32, Id, Id) {
// TODO
UNREACHABLE_MSG("Unsupported BUFFER_ATOMIC opcode: ", IR::Opcode::BufferAtomicInc32);
}
Id EmitBufferAtomicDec32(EmitContext&, IR::Inst*, u32, Id, Id) {
// TODO
UNREACHABLE_MSG("Unsupported BUFFER_ATOMIC opcode: ", IR::Opcode::BufferAtomicDec32);
}
Id EmitBufferAtomicAnd32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) {
return BufferAtomicU32(ctx, inst, handle, address, value, &Sirit::Module::OpAtomicAnd);
}
Id EmitBufferAtomicOr32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) {
return BufferAtomicU32(ctx, inst, handle, address, value, &Sirit::Module::OpAtomicOr);
}
Id EmitBufferAtomicXor32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) {
return BufferAtomicU32(ctx, inst, handle, address, value, &Sirit::Module::OpAtomicXor);
}
Id EmitBufferAtomicExchange32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) {
return BufferAtomicU32(ctx, inst, handle, address, value, &Sirit::Module::OpAtomicExchange);
}
Id EmitImageAtomicIAdd32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id value) {
return ImageAtomicU32(ctx, inst, handle, coords, value, &Sirit::Module::OpAtomicIAdd);
}

View file

@ -495,6 +495,7 @@ static void EmitStoreBufferFormatF32xN(EmitContext& ctx, u32 handle, Id address,
case AmdGpu::DataFormat::Format8_8_8_8:
case AmdGpu::DataFormat::Format16:
case AmdGpu::DataFormat::Format32:
case AmdGpu::DataFormat::Format32_32:
case AmdGpu::DataFormat::Format32_32_32_32: {
ASSERT(N == AmdGpu::NumComponents(format));

View file

@ -81,6 +81,17 @@ void EmitStoreBufferFormatF32x2(EmitContext& ctx, IR::Inst* inst, u32 handle, Id
void EmitStoreBufferFormatF32x3(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
void EmitStoreBufferFormatF32x4(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
void EmitStoreBufferU32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicIAdd32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicSMin32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicUMin32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicSMax32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicUMax32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicInc32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicDec32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicAnd32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicOr32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicXor32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitBufferAtomicExchange32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value);
Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, u32 comp);
Id EmitGetAttributeU32(EmitContext& ctx, IR::Attribute attr, u32 comp);
void EmitSetAttribute(EmitContext& ctx, IR::Attribute attr, Id value, u32 comp);
@ -103,6 +114,11 @@ Id EmitLoadSharedU128(EmitContext& ctx, Id offset);
void EmitWriteSharedU32(EmitContext& ctx, Id offset, Id value);
void EmitWriteSharedU64(EmitContext& ctx, Id offset, Id value);
void EmitWriteSharedU128(EmitContext& ctx, Id offset, Id value);
Id EmitSharedAtomicIAdd32(EmitContext& ctx, Id offset, Id value);
Id EmitSharedAtomicUMax32(EmitContext& ctx, Id offset, Id value);
Id EmitSharedAtomicSMax32(EmitContext& ctx, Id offset, Id value);
Id EmitSharedAtomicUMin32(EmitContext& ctx, Id offset, Id value);
Id EmitSharedAtomicSMin32(EmitContext& ctx, Id offset, Id value);
Id EmitCompositeConstructU32x2(EmitContext& ctx, Id e1, Id e2);
Id EmitCompositeConstructU32x3(EmitContext& ctx, Id e1, Id e2, Id e3);
Id EmitCompositeConstructU32x4(EmitContext& ctx, Id e1, Id e2, Id e3, Id e4);