shader: decode unsigned byte loads into preserved low halves

Co-authored-by: fxpw <fxpw@outlook.com>
This commit is contained in:
nmzik
2026-10-14 18:21:58 +02:00
parent 302240da70
commit e0586cc85a
6 changed files with 71 additions and 13 deletions
@@ -48,6 +48,7 @@ constexpr MemoryOpcodeInfo MUBUF_OPCODE_LIST[] = {
{0x1du, Opcode::BUFFER_STORE_DWORDX2, 2, 32},
{0x1eu, Opcode::BUFFER_STORE_DWORDX4, 4, 32},
{0x1fu, Opcode::BUFFER_STORE_DWORDX3, 3, 32},
{0x20u, Opcode::BUFFER_LOAD_UBYTE_D16, 1, 8},
{0x24u, Opcode::BUFFER_LOAD_SHORT_D16, 1, 16},
{0x25u, Opcode::BUFFER_LOAD_SHORT_D16_HI, 1, 16},
{0x30u, Opcode::BUFFER_ATOMIC_SWAP, 1, 32},
@@ -310,6 +311,7 @@ void DecodeMubuf(uint32_t pc, std::span<const uint32_t> code, uint32_t word_inde
inst.data_components = 4u;
}
if (inst.opcode == Opcode::BUFFER_LOAD_FORMAT_D16_X ||
inst.opcode == Opcode::BUFFER_LOAD_UBYTE_D16 ||
inst.opcode == Opcode::BUFFER_LOAD_SHORT_D16 ||
inst.opcode == Opcode::BUFFER_LOAD_SHORT_D16_HI) {
inst.dst.sdwa_sel = inst.opcode == Opcode::BUFFER_LOAD_SHORT_D16_HI ? 5u : 4u;
@@ -638,6 +638,7 @@ std::string InstructionToString(const Instruction& inst) {
case Opcode::BUFFER_STORE_FORMAT_XYZW:
case Opcode::BUFFER_STORE_FORMAT_D16_X:
case Opcode::BUFFER_LOAD_UBYTE:
case Opcode::BUFFER_LOAD_UBYTE_D16:
case Opcode::BUFFER_LOAD_USHORT:
case Opcode::BUFFER_LOAD_SHORT_D16:
case Opcode::BUFFER_LOAD_SHORT_D16_HI:
@@ -554,6 +554,7 @@ enum class Opcode {
BUFFER_STORE_FORMAT_XYZW,
BUFFER_STORE_FORMAT_D16_X,
BUFFER_LOAD_UBYTE,
BUFFER_LOAD_UBYTE_D16,
BUFFER_LOAD_SBYTE,
BUFFER_LOAD_USHORT,
BUFFER_LOAD_SSHORT,
@@ -973,6 +973,7 @@ void Translator::EmitMemory(const Decoder::Instruction& inst) {
}
case Decoder::Opcode::BUFFER_LOAD_UBYTE:
case Decoder::Opcode::BUFFER_LOAD_UBYTE_D16:
case Decoder::Opcode::BUFFER_LOAD_SBYTE:
case Decoder::Opcode::BUFFER_LOAD_USHORT:
case Decoder::Opcode::BUFFER_LOAD_SSHORT: