Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

perf: Group basic block end opcodes to eliminate slower pattern matching #347

Merged
merged 1 commit into from
Jun 7, 2023
Merged

perf: Group basic block end opcodes to eliminate slower pattern matching #347

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
302 changes: 153 additions & 149 deletions definitions/src/instructions.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,158 +86,159 @@ macro_rules! __for_each_inst_inner {
(ADDW, 0x14),
(AND, 0x15),
(ANDI, 0x16),
(AUIPC, 0x17),
(BEQ, 0x18),
(BGE, 0x19),
(BGEU, 0x1a),
(BLT, 0x1b),
(BLTU, 0x1c),
(BNE, 0x1d),
(DIV, 0x1e),
(DIVU, 0x1f),
(DIVUW, 0x20),
(DIVW, 0x21),
(EBREAK, 0x22),
(ECALL, 0x23),
(FENCE, 0x24),
(FENCEI, 0x25),
(JAL, 0x26),
(JALR_VERSION0, 0x27),
(JALR_VERSION1, 0x28),
(LB_VERSION0, 0x29),
(LB_VERSION1, 0x2a),
(LBU_VERSION0, 0x2b),
(LBU_VERSION1, 0x2c),
(LD_VERSION0, 0x2d),
(LD_VERSION1, 0x2e),
(LH_VERSION0, 0x2f),
(LH_VERSION1, 0x30),
(LHU_VERSION0, 0x31),
(LHU_VERSION1, 0x32),
(LUI, 0x33),
(LW_VERSION0, 0x34),
(LW_VERSION1, 0x35),
(LWU_VERSION0, 0x36),
(LWU_VERSION1, 0x37),
(MUL, 0x38),
(MULH, 0x39),
(MULHSU, 0x3a),
(MULHU, 0x3b),
(MULW, 0x3c),
(OR, 0x3d),
(ORI, 0x3e),
(REM, 0x3f),
(REMU, 0x40),
(REMUW, 0x41),
(REMW, 0x42),
(SB, 0x43),
(SD, 0x44),
(SH, 0x45),
(SLL, 0x46),
(SLLI, 0x47),
(SLLIW, 0x48),
(SLLW, 0x49),
(SLT, 0x4a),
(SLTI, 0x4b),
(SLTIU, 0x4c),
(SLTU, 0x4d),
(SRA, 0x4e),
(SRAI, 0x4f),
(SRAIW, 0x50),
(SRAW, 0x51),
(SRL, 0x52),
(SRLI, 0x53),
(SRLIW, 0x54),
(SRLW, 0x55),
(SUB, 0x56),
(SUBW, 0x57),
(SW, 0x58),
(XOR, 0x59),
(XORI, 0x5a),
(DIV, 0x17),
(DIVU, 0x18),
(DIVUW, 0x19),
(DIVW, 0x1a),
(LB_VERSION0, 0x1b),
(LB_VERSION1, 0x1c),
(LBU_VERSION0, 0x1d),
(LBU_VERSION1, 0x1e),
(LD_VERSION0, 0x1f),
(LD_VERSION1, 0x20),
(LH_VERSION0, 0x21),
(LH_VERSION1, 0x22),
(LHU_VERSION0, 0x23),
(LHU_VERSION1, 0x24),
(LUI, 0x25),
(LW_VERSION0, 0x26),
(LW_VERSION1, 0x27),
(LWU_VERSION0, 0x28),
(LWU_VERSION1, 0x29),
(MUL, 0x2a),
(MULH, 0x2b),
(MULHSU, 0x2c),
(MULHU, 0x2d),
(MULW, 0x2e),
(OR, 0x2f),
(ORI, 0x30),
(REM, 0x31),
(REMU, 0x32),
(REMUW, 0x33),
(REMW, 0x34),
(SB, 0x35),
(SD, 0x36),
(SH, 0x37),
(SLL, 0x38),
(SLLI, 0x39),
(SLLIW, 0x3a),
(SLLW, 0x3b),
(SLT, 0x3c),
(SLTI, 0x3d),
(SLTIU, 0x3e),
(SLTU, 0x3f),
(SRA, 0x40),
(SRAI, 0x41),
(SRAIW, 0x42),
(SRAW, 0x43),
(SRL, 0x44),
(SRLI, 0x45),
(SRLIW, 0x46),
(SRLW, 0x47),
(SUB, 0x48),
(SUBW, 0x49),
(SW, 0x4a),
(XOR, 0x4b),
(XORI, 0x4c),
// A
(LR_W, 0x5b),
(SC_W, 0x5c),
(AMOSWAP_W, 0x5d),
(AMOADD_W, 0x5e),
(AMOXOR_W, 0x5f),
(AMOAND_W, 0x60),
(AMOOR_W, 0x61),
(AMOMIN_W, 0x62),
(AMOMAX_W, 0x63),
(AMOMINU_W, 0x64),
(AMOMAXU_W, 0x65),
(LR_D, 0x66),
(SC_D, 0x67),
(AMOSWAP_D, 0x68),
(AMOADD_D, 0x69),
(AMOXOR_D, 0x6a),
(AMOAND_D, 0x6b),
(AMOOR_D, 0x6c),
(AMOMIN_D, 0x6d),
(AMOMAX_D, 0x6e),
(AMOMINU_D, 0x6f),
(AMOMAXU_D, 0x70),
(LR_W, 0x4d),
(SC_W, 0x4e),
(AMOSWAP_W, 0x4f),
(AMOADD_W, 0x50),
(AMOXOR_W, 0x51),
(AMOAND_W, 0x52),
(AMOOR_W, 0x53),
(AMOMIN_W, 0x54),
(AMOMAX_W, 0x55),
(AMOMINU_W, 0x56),
(AMOMAXU_W, 0x57),
(LR_D, 0x58),
(SC_D, 0x59),
(AMOSWAP_D, 0x5a),
(AMOADD_D, 0x5b),
(AMOXOR_D, 0x5c),
(AMOAND_D, 0x5d),
(AMOOR_D, 0x5e),
(AMOMIN_D, 0x5f),
(AMOMAX_D, 0x60),
(AMOMINU_D, 0x61),
(AMOMAXU_D, 0x62),
// B
(ADDUW, 0x71),
(ANDN, 0x72),
(BCLR, 0x73),
(BCLRI, 0x74),
(BEXT, 0x75),
(BEXTI, 0x76),
(BINV, 0x77),
(BINVI, 0x78),
(BSET, 0x79),
(BSETI, 0x7a),
(CLMUL, 0x7b),
(CLMULH, 0x7c),
(CLMULR, 0x7d),
(CLZ, 0x7e),
(CLZW, 0x7f),
(CPOP, 0x80),
(CPOPW, 0x81),
(CTZ, 0x82),
(CTZW, 0x83),
(MAX, 0x84),
(MAXU, 0x85),
(MIN, 0x86),
(MINU, 0x87),
(ORCB, 0x88),
(ORN, 0x89),
(REV8, 0x8a),
(ROL, 0x8b),
(ROLW, 0x8c),
(ROR, 0x8d),
(RORI, 0x8e),
(RORIW, 0x8f),
(RORW, 0x90),
(SEXTB, 0x91),
(SEXTH, 0x92),
(SH1ADD, 0x93),
(SH1ADDUW, 0x94),
(SH2ADD, 0x95),
(SH2ADDUW, 0x96),
(SH3ADD, 0x97),
(SH3ADDUW, 0x98),
(SLLIUW, 0x99),
(XNOR, 0x9a),
(ZEXTH, 0x9b),
(ADDUW, 0x63),
(ANDN, 0x64),
(BCLR, 0x65),
(BCLRI, 0x66),
(BEXT, 0x67),
(BEXTI, 0x68),
(BINV, 0x69),
(BINVI, 0x6a),
(BSET, 0x6b),
(BSETI, 0x6c),
(CLMUL, 0x6d),
(CLMULH, 0x6e),
(CLMULR, 0x6f),
(CLZ, 0x70),
(CLZW, 0x71),
(CPOP, 0x72),
(CPOPW, 0x73),
(CTZ, 0x74),
(CTZW, 0x75),
(MAX, 0x76),
(MAXU, 0x77),
(MIN, 0x78),
(MINU, 0x79),
(ORCB, 0x7a),
(ORN, 0x7b),
(REV8, 0x7c),
(ROL, 0x7d),
(ROLW, 0x7e),
(ROR, 0x7f),
(RORI, 0x80),
(RORIW, 0x81),
(RORW, 0x82),
(SEXTB, 0x83),
(SEXTH, 0x84),
(SH1ADD, 0x85),
(SH1ADDUW, 0x86),
(SH2ADD, 0x87),
(SH2ADDUW, 0x88),
(SH3ADD, 0x89),
(SH3ADDUW, 0x8a),
(SLLIUW, 0x8b),
(XNOR, 0x8c),
(ZEXTH, 0x8d),
// Mop
(WIDE_MUL, 0x9c),
(WIDE_MULU, 0x9d),
(WIDE_MULSU, 0x9e),
(WIDE_DIV, 0x9f),
(WIDE_DIVU, 0xa0),
(FAR_JUMP_REL, 0xa1),
(FAR_JUMP_ABS, 0xa2),
(ADC, 0xa3),
(SBB, 0xa4),
(ADCS, 0xa5),
(SBBS, 0xa6),
(ADD3A, 0xa7),
(ADD3B, 0xa8),
(ADD3C, 0xa9),
(CUSTOM_LOAD_UIMM, 0xaa),
(CUSTOM_LOAD_IMM, 0xab),
(WIDE_MUL, 0x8e),
(WIDE_MULU, 0x8f),
(WIDE_MULSU, 0x90),
(WIDE_DIV, 0x91),
(WIDE_DIVU, 0x92),
(ADC, 0x93),
(SBB, 0x94),
(ADCS, 0x95),
(SBBS, 0x96),
(ADD3A, 0x97),
(ADD3B, 0x98),
(ADD3C, 0x99),
(CUSTOM_LOAD_UIMM, 0x9a),
(CUSTOM_LOAD_IMM, 0x9b),
// All branches
(AUIPC, 0x9c),
(BEQ, 0x9d),
(BGE, 0x9e),
(BGEU, 0x9f),
(BLT, 0xa0),
(BLTU, 0xa1),
(BNE, 0xa2),
(EBREAK, 0xa3),
(ECALL, 0xa4),
(FENCE, 0xa5),
mohanson marked this conversation as resolved.
Show resolved Hide resolved
(FENCEI, 0xa6),
(JAL, 0xa7),
(JALR_VERSION0, 0xa8),
(JALR_VERSION1, 0xa9),
(FAR_JUMP_REL, 0xaa),
(FAR_JUMP_ABS, 0xab),
(CUSTOM_TRACE_END, 0xac)
);
};
Expand Down Expand Up @@ -296,6 +297,9 @@ for_each_inst!(define_instruction);
pub const MINIMAL_OPCODE: InstructionOpcode = OP_UNLOADED;
pub const MAXIMUM_OPCODE: InstructionOpcode = OP_CUSTOM_TRACE_END;

pub const MINIMAL_BASIC_BLOCK_END_OPCODE: InstructionOpcode = OP_AUIPC;
pub const MAXIMUM_BASIC_BLOCK_END_OPCODE: InstructionOpcode = OP_FAR_JUMP_ABS;

macro_rules! inst_real_name {
($name:ident, $real_name:ident, $code:expr) => {
stringify!($real_name)
Expand Down
43 changes: 25 additions & 18 deletions src/instructions/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,8 @@ pub use self::register::Register;
use super::Error;
pub use ckb_vm_definitions::{
instructions::{
self as insts, instruction_opcode_name, Instruction, InstructionOpcode, MINIMAL_OPCODE,
self as insts, instruction_opcode_name, Instruction, InstructionOpcode,
MAXIMUM_BASIC_BLOCK_END_OPCODE, MINIMAL_BASIC_BLOCK_END_OPCODE, MINIMAL_OPCODE,
},
registers::REGISTER_ABI_NAMES,
};
Expand Down Expand Up @@ -410,23 +411,9 @@ pub fn is_slowpath_instruction(i: Instruction) -> bool {
}

pub fn is_basic_block_end_instruction(i: Instruction) -> bool {
matches!(
extract_opcode(i),
insts::OP_AUIPC
| insts::OP_JALR_VERSION0
| insts::OP_JALR_VERSION1
| insts::OP_BEQ
| insts::OP_BNE
| insts::OP_BLT
| insts::OP_BGE
| insts::OP_BLTU
| insts::OP_BGEU
| insts::OP_ECALL
| insts::OP_EBREAK
| insts::OP_JAL
| insts::OP_FAR_JUMP_ABS
| insts::OP_FAR_JUMP_REL
) | is_slowpath_instruction(i)
let opcode = extract_opcode(i);
(MINIMAL_BASIC_BLOCK_END_OPCODE..=MAXIMUM_BASIC_BLOCK_END_OPCODE).contains(&opcode)
|| is_slowpath_instruction(i)
}

#[inline(always)]
Expand Down Expand Up @@ -506,4 +493,24 @@ mod tests {
for_each_inst_fold!(update_max_opcode, o);
assert_eq!(MAXIMUM_OPCODE, o);
}

#[test]
fn test_basic_block_end_opcode_is_in_range() {
for o in MINIMAL_OPCODE..=MAXIMUM_OPCODE {
if is_basic_block_end_instruction(blank_instruction(o)) {
assert!(
o >= MINIMAL_BASIC_BLOCK_END_OPCODE,
"Opcode {} ({}) is smaller than minimal basic block end opcode!",
o,
instruction_opcode_name(o)
);
assert!(
o <= MAXIMUM_BASIC_BLOCK_END_OPCODE,
"Opcode {} ({}) is bigger than maximum basic block end opcode!",
o,
instruction_opcode_name(o)
);
}
}
}
}
Loading