shader: link multiple compute helpers without synthetic return loops

This commit is contained in:
nmzik
2026-10-18 23:19:53 +02:00
parent b2d3ce6dfc
commit ab8566591f
8 changed files with 363 additions and 109 deletions
@@ -164,9 +164,23 @@ void DecodeSop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
switch (inst.opcode) {
case Opcode::S_SWAPPC_B64:
inst.branch_target = UINT32_MAX;
inst.data_dwords = 2;
[[fallthrough]];
case Opcode::S_MOV_B64:
case Opcode::S_CMOV_B64:
case Opcode::S_NOT_B64:
case Opcode::S_WQM_B64:
case Opcode::S_BREV_B64:
case Opcode::S_BITSET0_B64:
case Opcode::S_BITSET1_B64:
case Opcode::S_AND_SAVEEXEC_B64:
case Opcode::S_ORN2_SAVEEXEC_B64:
case Opcode::S_QUADMASK_B64:
case Opcode::S_ANDN1_SAVEEXEC_B64:
case Opcode::S_BITREPLICATE_B64_B32:
inst.data_dwords = 2u;
break;
case Opcode::S_GETPC_B64:
inst.data_dwords = 2u;
inst.src_count = 0;
DecodeScalarDestination(sdst, pc, inst.dst);
return;
@@ -206,6 +220,8 @@ void DecodeSop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
DecodeScalarDestination(sdst, pc, inst.dst);
DecodeBinarySources(pc, code, word_index, inst, ssrc0, ssrc1);
if (opcode == 0x0bu || (opcode >= 0x0fu && opcode <= 0x25u && (opcode & 1u)) || opcode == 0x29u)
inst.data_dwords = 2u;
}
void DecodeSopk(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,