shader: add BUFFER_LOAD_FORMAT_D16_X opcode

This commit is contained in:
nmzik
2026-10-07 09:22:14 +04:00
parent 5af370c257
commit 26cce9909a
6 changed files with 90 additions and 2 deletions
@@ -716,10 +716,21 @@ uint32_t FormattedLoad(ValueEmitContext& ctx, const IR::Inst& inst, const IR::Me
return LoadWordPrepared(ctx, inst, RebaseRawComponent(mem, 0u), resource);
}
const auto plan = PrepareFormattedMemory(ctx, inst, mem, resource, info);
return EmitValueOrDefaultIfCondition(
const auto value = EmitValueOrDefaultIfCondition(
ctx.state, plan.in_bounds, TypeU32(ctx.state),
FormattedOutOfBoundsValue(ctx, mem, plan, 1u),
[&]() { return LoadFormattedInBounds(ctx, mem, plan, 0u); });
if (mem.data_bits != 16u) {
return value;
}
if (info.type == Format::ComponentType::Uint) {
return EmitUMin32(ctx.state, value, ConstantU32(ctx.state, 0xffffu));
}
if (info.type == Format::ComponentType::Sint) {
const auto lower = EmitSMax32(ctx.state, value, ConstantU32(ctx.state, 0xffff8000u));
return EmitSMin32(ctx.state, lower, ConstantU32(ctx.state, 0x7fffu));
}
return EmitConvertF16F32(ctx.state, EmitBitCastF32U32(ctx.state, value));
});
}
@@ -63,6 +63,7 @@ constexpr MemoryOpcodeInfo MUBUF_OPCODE_LIST[] = {
{0x50u, Opcode::BUFFER_ATOMIC_SWAP_X2, 2, 32},
{0x59u, Opcode::BUFFER_ATOMIC_AND_X2, 2, 32},
{0x5au, Opcode::BUFFER_ATOMIC_OR_X2, 2, 32},
{0x80u, Opcode::BUFFER_LOAD_FORMAT_D16_X, 1, 16, false, false, true},
{0x84u, Opcode::BUFFER_STORE_FORMAT_D16_X, 1, 16, false, false, true},
};
@@ -297,6 +298,9 @@ void DecodeMubuf(uint32_t pc, std::span<const uint32_t> code, uint32_t word_inde
}
DecodeVectorGpr(vdata, inst.dst);
if (inst.opcode == Opcode::BUFFER_LOAD_FORMAT_D16_X) {
inst.dst.sdwa_sel = 4u;
}
DecodeVectorGpr(vaddr, inst.src0);
DecodeScalarSource(srsrc * 4u, pc, inst.src1);
DecodeScalarSource(soffset, pc, inst.src2);
@@ -618,6 +618,7 @@ std::string InstructionToString(const Instruction& inst) {
case Opcode::BUFFER_LOAD_FORMAT_XY:
case Opcode::BUFFER_LOAD_FORMAT_XYZ:
case Opcode::BUFFER_LOAD_FORMAT_XYZW:
case Opcode::BUFFER_LOAD_FORMAT_D16_X:
case Opcode::BUFFER_STORE_FORMAT_X:
case Opcode::BUFFER_STORE_FORMAT_XY:
case Opcode::BUFFER_STORE_FORMAT_XYZ:
@@ -505,6 +505,7 @@ enum class Opcode {
BUFFER_LOAD_FORMAT_XY,
BUFFER_LOAD_FORMAT_XYZ,
BUFFER_LOAD_FORMAT_XYZW,
BUFFER_LOAD_FORMAT_D16_X,
BUFFER_STORE_FORMAT_X,
BUFFER_STORE_FORMAT_XY,
BUFFER_STORE_FORMAT_XYZ,
@@ -427,7 +427,7 @@ void Translator::S_LOAD(const Decoder::Instruction& inst, bool raw) {
void Translator::BUFFER_LOAD(const Decoder::Instruction& inst) {
const auto memory = MemoryInfoFromDecoded(inst);
IR::ValueOpcode opcode;
const auto bits = memory.data_bits;
const auto bits = memory.formatted ? 32u : memory.data_bits;
const auto sign = memory.data_signed;
switch (bits) {
case 8u: opcode = IR::ValueOpcode::LoadBufferU8; break;
@@ -981,6 +981,7 @@ void Translator::EmitMemory(const Decoder::Instruction& inst) {
case Decoder::Opcode::BUFFER_LOAD_FORMAT_XY:
case Decoder::Opcode::BUFFER_LOAD_FORMAT_XYZ:
case Decoder::Opcode::BUFFER_LOAD_FORMAT_XYZW:
case Decoder::Opcode::BUFFER_LOAD_FORMAT_D16_X:
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_X:
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_XY:
case Decoder::Opcode::TBUFFER_LOAD_FORMAT_XYZ: