hipnosuer
/
Zuma
forked from ShuriZma/suyu
1
0
Fork 0
Code Pull Requests Activity

shader/memory: Implement RED.E.ADD 

Implements a reduction operation. It's an atomic operation that doesn't
return a value.

This commit introduces another primitive because some shading languages
might have a primitive for reduction operations.
This commit is contained in:
ReinUsesLisp 2020-04-06 02:24:47 -03:00
parent fd0a2b5151
commit 3185245845
5 changed files with 98 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/video_core/engines/shader_bytecode.h
View File

@ -988,6 +988,12 @@ union Instruction {
BitField<46, 2, u64> cache_mode;
} stg;
union {
BitField<23, 3, AtomicOp> operation;
BitField<48, 1, u64> extended;
BitField<20, 3, GlobalAtomicType> type;
} red;
union {
BitField<52, 4, AtomicOp> operation;
BitField<49, 3, GlobalAtomicType> type;
@ -1733,6 +1739,7 @@ public:
ST_S,
ST, // Store in generic memory
STG, // Store in global memory
RED, // Reduction operation
ATOM, // Atomic operation on global memory
ATOMS, // Atomic operation on shared memory
AL2P, // Transforms attribute memory into physical memory
@ -2039,6 +2046,7 @@ private:
INST("1110111101010---", Id::ST_L, Type::Memory, "ST_L"),
INST("101-------------", Id::ST, Type::Memory, "ST"),
INST("1110111011011---", Id::STG, Type::Memory, "STG"),
INST("1110101111111---", Id::RED, Type::Memory, "RED"),
INST("11101101--------", Id::ATOM, Type::Memory, "ATOM"),
INST("11101100--------", Id::ATOMS, Type::Memory, "ATOMS"),
INST("1110111110100---", Id::AL2P, Type::Memory, "AL2P"),

24
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View File

@ -2119,8 +2119,14 @@ private:
return {};
}
return {fmt::format("atomic{}({}, {})", opname, Visit(operation[0]).GetCode(),
Visit(operation[1]).As(type)),
type};
Visit(operation[1]).AsUint()),
Type::Uint};
}
template <const std::string_view& opname, Type type>
Expression Reduce(Operation operation) {
code.AddLine("{};", Atomic<opname, type>(operation).GetCode());
return {};
}
Expression Branch(Operation operation) {
@ -2479,6 +2485,20 @@ private:
&GLSLDecompiler::Atomic<Func::Or, Type::Int>,
&GLSLDecompiler::Atomic<Func::Xor, Type::Int>,
&GLSLDecompiler::Reduce<Func::Add, Type::Uint>,
&GLSLDecompiler::Reduce<Func::Min, Type::Uint>,
&GLSLDecompiler::Reduce<Func::Max, Type::Uint>,
&GLSLDecompiler::Reduce<Func::And, Type::Uint>,
&GLSLDecompiler::Reduce<Func::Or, Type::Uint>,
&GLSLDecompiler::Reduce<Func::Xor, Type::Uint>,
&GLSLDecompiler::Reduce<Func::Add, Type::Int>,
&GLSLDecompiler::Reduce<Func::Min, Type::Int>,
&GLSLDecompiler::Reduce<Func::Max, Type::Int>,
&GLSLDecompiler::Reduce<Func::And, Type::Int>,
&GLSLDecompiler::Reduce<Func::Or, Type::Int>,
&GLSLDecompiler::Reduce<Func::Xor, Type::Int>,
&GLSLDecompiler::Branch,
&GLSLDecompiler::BranchIndirect,
&GLSLDecompiler::PushFlowStack,

65
src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
View File

@ -1941,11 +1941,8 @@ private:
return {};
}
template <Id (Module::*func)(Id, Id, Id, Id, Id), Type result_type,
Type value_type = result_type>
template <Id (Module::*func)(Id, Id, Id, Id, Id)>
Expression Atomic(Operation operation) {
const Id type_def = GetTypeDefinition(result_type);
Id pointer;
if (const auto smem = std::get_if<SmemNode>(&*operation[0])) {
pointer = GetSharedMemoryPointer(*smem);
@ -1953,15 +1950,19 @@ private:
pointer = GetGlobalMemoryPointer(*gmem);
} else {
UNREACHABLE();
return {Constant(type_def, 0), result_type};
return {v_float_zero, Type::Float};
}
const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
const Id semantics = Constant(t_uint, 0);
const Id value = AsUint(Visit(operation[1]));
return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint};
}
const Id value = As(Visit(operation[1]), value_type);
const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
const Id semantics = Constant(type_def, 0);
return {(this->*func)(type_def, pointer, scope, semantics, value), result_type};
template <Id (Module::*func)(Id, Id, Id, Id, Id)>
Expression Reduce(Operation operation) {
Atomic<func>(operation);
return {};
}
Expression Branch(Operation operation) {
@ -2550,21 +2551,35 @@ private:
&SPIRVDecompiler::AtomicImageXor,
&SPIRVDecompiler::AtomicImageExchange,
&SPIRVDecompiler::Atomic<&Module::OpAtomicExchange, Type::Uint>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd, Type::Uint>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicUMin, Type::Uint>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicUMax, Type::Uint>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicAnd, Type::Uint>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicOr, Type::Uint>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicXor, Type::Uint>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicExchange>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicUMin>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicUMax>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicAnd>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicOr>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicXor>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicExchange, Type::Int>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd, Type::Int>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicSMin, Type::Int>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicSMax, Type::Int>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicAnd, Type::Int>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicOr, Type::Int>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicXor, Type::Int>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicExchange>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicSMin>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicSMax>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicAnd>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicOr>,
&SPIRVDecompiler::Atomic<&Module::OpAtomicXor>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicIAdd>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicUMin>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicUMax>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicAnd>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicOr>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicXor>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicIAdd>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicSMin>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicSMax>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicAnd>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicOr>,
&SPIRVDecompiler::Reduce<&Module::OpAtomicXor>,
&SPIRVDecompiler::Branch,
&SPIRVDecompiler::BranchIndirect,

16
src/video_core/shader/decode/memory.cpp
View File

@ -378,13 +378,27 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
if (IsUnaligned(type)) {
const u32 mask = GetUnalignedMask(type);
value = InsertUnaligned(gmem, std::move(value), real_address, mask, size);
value = InsertUnaligned(gmem, move(value), real_address, mask, size);
}
bb.push_back(Operation(OperationCode::Assign, gmem, value));
}
break;
}
case OpCode::Id::RED: {
UNIMPLEMENTED_IF_MSG(instr.red.type != GlobalAtomicType::U32);
UNIMPLEMENTED_IF_MSG(instr.red.operation != AtomicOp::Add);
const auto [real_address, base_address, descriptor] =
TrackGlobalMemory(bb, instr, true, true);
if (!real_address || !base_address) {
// Tracking failed, skip atomic.
break;
}
Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
Node value = GetRegister(instr.gpr0);
bb.push_back(Operation(OperationCode::ReduceIAdd, move(gmem), move(value)));
break;
}
case OpCode::Id::ATOM: {
UNIMPLEMENTED_IF_MSG(instr.atom.operation == AtomicOp::Inc ||
instr.atom.operation == AtomicOp::Dec ||

14
src/video_core/shader/node.h
View File

@ -178,6 +178,20 @@ enum class OperationCode {
AtomicIOr, /// (memory, int) -> int
AtomicIXor, /// (memory, int) -> int
ReduceUAdd, /// (memory, uint) -> void
ReduceUMin, /// (memory, uint) -> void
ReduceUMax, /// (memory, uint) -> void
ReduceUAnd, /// (memory, uint) -> void
ReduceUOr, /// (memory, uint) -> void
ReduceUXor, /// (memory, uint) -> void
ReduceIAdd, /// (memory, int) -> void
ReduceIMin, /// (memory, int) -> void
ReduceIMax, /// (memory, int) -> void
ReduceIAnd, /// (memory, int) -> void
ReduceIOr, /// (memory, int) -> void
ReduceIXor, /// (memory, int) -> void
Branch, /// (uint branch_target) -> void
BranchIndirect, /// (uint branch_target) -> void
PushFlowStack, /// (uint branch_target) -> void