raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-25 04:16:55 +01:00
Code Issues Packages Projects Releases Wiki Activity

shader: fix LDS counter bounds

Browse Source
This commit is contained in:
nmzik
2026-09-27 21:42:59 +04:00
parent e68b9b8d37
commit ab4b10cf48
4 changed files with 129 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemory.cpp
+24 -15
View File
@@ -1104,16 +1104,27 @@ uint32_t EmitAppendConsume(ValueEmitContext& ctx, const IR::Inst& inst) {
if (ctx.half == 1) {
return ctx.other_half->Def(IR::Value(const_cast<IR::Inst*>(&inst)));
}
const auto m0 = ctx.Arg(inst, 0);
const auto base =
Binary(state, spv::OpShiftRightLogical, TypeU32(state), m0, ConstantU32(state, 16));
const auto size =
Binary(state, spv::OpBitwiseAnd, TypeU32(state), m0, ConstantU32(state, 0xffffu));
const auto address = Binary(state, spv::OpIAdd, TypeU32(state), base,
ConstantU32(state, ctx.Memory(inst).offset));
const auto mem = ctx.Memory(inst);
const auto offset = mem.offset & 0xfffcu;
auto address = ConstantU32(state, offset);
uint32_t region_bounds = 0;
if (mem.kind == IR::ResourceKind::Gds) {
const auto m0 = ctx.Arg(inst, 0);
const auto base =
Binary(state, spv::OpShiftRightLogical, TypeU32(state), m0, ConstantU32(state, 16));
const auto size =
Binary(state, spv::OpBitwiseAnd, TypeU32(state), m0, ConstantU32(state, 0xffffu));
address = Binary(state, spv::OpIAdd, TypeU32(state), base, address);
// The entire M0 region must fit the PS5's 48 KiB GDS partition.
region_bounds = AndCondition(
state, Binary(state, spv::OpULessThan, TypeBool(state),
ConstantU32(state, offset + 3u), size),
Binary(state, spv::OpULessThanEqual, TypeBool(state),
Binary(state, spv::OpIAdd, TypeU32(state), base, size),
ConstantU32(state, 0xc000u)));
}
const auto raw_index =
Binary(state, spv::OpShiftRightLogical, TypeU32(state), address, ConstantU32(state, 2));
const auto mem = ctx.Memory(inst);
const auto access = PrepareMemoryResourceAccess(state, mem);
const auto index = EmitMemoryElementIndex(state, access, raw_index);
const auto exec = ctx.Arg(inst, 1);
@@ -1132,19 +1143,17 @@ uint32_t EmitAppendConsume(ValueEmitContext& ctx, const IR::Inst& inst) {
: first;
const auto is_first = Binary(state, spv::OpIEqual, TypeBool(state),
EmitSubgroupLocalInvocationId(state), source_lane);
const auto storage_bounds = EmitMemoryElementInBounds(state, access, index);
const auto m0_bounds =
mem.kind == IR::ResourceKind::Gds
? Binary(state, spv::OpINotEqual, TypeBool(state), size, ConstantU32(state, 0))
: Binary(state, spv::OpULessThan, TypeBool(state),
ConstantU32(state, ctx.Memory(inst).offset + 3u), size);
auto bounds = EmitMemoryElementInBounds(state, access, index);
if (mem.kind == IR::ResourceKind::Gds) {
bounds = AndCondition(state, bounds, region_bounds);
}
const auto condition = AndCondition(
state, is_first,
AndCondition(state,
state.lane_count == 2 ? Binary(state, spv::OpINotEqual, TypeBool(state), count,
ConstantU32(state, 0))
: exec,
AndCondition(state, storage_bounds, m0_bounds)));
bounds));
const auto atomic = EmitValueOrZeroIfCondition(state, condition, [&]() {
const auto value = state.builder.AllocateId();
state.builder.AddFunction(append ? spv::OpAtomicIAdd : spv::OpAtomicISub, TypeU32(state),
src/graphics/shader/recompiler/frontend/translate/Memory.cpp
+2 -2
View File
@@ -235,6 +235,7 @@ Decoder::Operand MemorySourceAt(const Decoder::Instruction& decoded, uint32_t in
return index == 0u ? decoded.src0 : MakeImmediate(decoded.offset & 0xffffu);
case Decoder::Opcode::DS_CONSUME:
case Decoder::Opcode::DS_APPEND:
return decoded.gds ? MakeM0Operand() : MakeImmediate(0);
case Decoder::Opcode::DS_READ_ADDTID_B32: return MakeM0Operand();
case Decoder::Opcode::DS_WRITE_ADDTID_B32:
return index == 0u ? decoded.src1 : MakeM0Operand();
@@ -845,8 +846,7 @@ void Translator::DS_MINMAX_F32(const Decoder::Instruction& inst, IR::ValueOpcode
void Translator::DS_APPEND_CONSUME(const Decoder::Instruction& inst, IR::ValueOpcode opcode) {
const auto memory = MemoryInfoFromDecoded(inst);
WriteOperand(inst.dst, ir.Emit(opcode,
{ReadU32(MemorySourceAt(inst, 0)), ir.GetExec(), ir.GetExecLo(),
ir.GetExecHi()},
{ReadU32(MemorySourceAt(inst, 0)), ir.GetExec()},
AddMemoryInfo(memory, inst.pc)));
}
src/graphics/shader/recompiler/ir/opcodes/ValueOpcodes.inc
+2 -2
View File
@@ -275,8 +275,8 @@ VALUE_OPCODE(SharedAtomicUMax32, U32, U32, U32, U1)
VALUE_OPCODE(SharedAtomicAnd32, U32, U32, U32, U1)
VALUE_OPCODE(SharedAtomicOr32, U32, U32, U32, U1)
VALUE_OPCODE(SharedAtomicXor32, U32, U32, U32, U1)
VALUE_OPCODE(DataAppend, U32, U32, U1, U32, U32)
VALUE_OPCODE(DataConsume, U32, U32, U1, U32, U32)
VALUE_OPCODE(DataAppend, U32, U32, U1)
VALUE_OPCODE(DataConsume, U32, U32, U1)
VALUE_OPCODE(SwizzleU32, U32, U32, U32, U1)
VALUE_OPCODE(ImageQueryDimensions, U32x4, ImageResource, ImageAddress)