Binary Translation型のエミュレータの開発続き。今回は久しぶりに浮動小数点に戻って、FCVT命令を実装する。
FCVT命令は、すべてSoftfloatでの実装となる。FCVT.xx命令をデコードすると、すべてFCVTのヘルパー関数にジャンプし、そこで処理を行うことになる。
という訳で、簡単にTCGを記述してヘルパー関数にジャンプする実装を追加した。
pub fn translate_fcvt_w_s (&mut self, inst: &InstrInfo) -> Vec<TCGOp> { let mut tcg_lists = vec![]; let rs1 = TCGv::new_reg(get_rs1_addr!(inst.inst) as u64); let rd = TCGv::new_reg(get_rd_addr!(inst.inst) as u64); let fcvt_helper_idx = TCGv::new_imm(CallFcvtIdx::W_S as u64); tcg_lists.push(TCGOp::new_helper_call_arg3(CALL_HELPER_IDX::CALL_FCVT_IDX as usize, fcvt_helper_idx, rd, rs1)); tcg_lists }
このヘルパー関数は、Rustの実装で処理される。具体的には以下のようになる。
pub fn helper_func_fcvt(emu: &mut EmuEnv, call_idx: u64, rd: u64, rs1: u64, _: u64) -> usize { let mut flag = ExceptionFlags::default(); flag.set(); let helper_idx = CallFcvtIdx::from_u64(call_idx); match helper_idx { CallFcvtIdx::W_S => { let to_data = F32::from_bits(emu.m_fregs[rs1 as usize] as u32).to_i32(RoundingMode::TowardZero, true); emu.m_iregs[rd as usize] = to_data as u64; }, CallFcvtIdx::WU_S => { let to_data = F32::from_bits(emu.m_fregs[rs1 as usize] as u32).to_u32(RoundingMode::TowardZero, true); emu.m_iregs[rd as usize] = to_data as i32 as u64; }, CallFcvtIdx::S_W => { let to_data = F32::from_i32 (emu.m_iregs[rs1 as usize] as i32, RoundingMode::TowardZero) ; emu.m_fregs[rd as usize] = to_data.bits() as u64; }, CallFcvtIdx::S_WU => { let to_data = F32::from_u32 (emu.m_iregs[rs1 as usize] as u32, RoundingMode::TiesToEven) ; emu.m_fregs[rd as usize] = to_data.bits() as u64; }, CallFcvtIdx::S_D => { let to_data = F64::from_bits(emu.m_fregs[rs1 as usize] as u64).to_f32(RoundingMode::TowardZero); emu.m_fregs[rd as usize] = to_data.bits() as u64; }, CallFcvtIdx::D_S => { let to_data = F32::from_bits(emu.m_fregs[rs1 as usize] as u32).to_f64(RoundingMode::TowardZero); emu.m_fregs[rd as usize] = to_data.bits() as u64; }, CallFcvtIdx::W_D => { let to_data = F64::from_bits(emu.m_fregs[rs1 as usize] as u64).to_i32(RoundingMode::TowardZero, true); emu.m_iregs[rd as usize] = to_data as u64; }, CallFcvtIdx::WU_D => { let to_data = F64::from_bits(emu.m_fregs[rs1 as usize] as u64).to_u32(RoundingMode::TowardZero, true); emu.m_iregs[rd as usize] = to_data as i32 as u64; }, CallFcvtIdx::D_W => { let to_data = F64::from_i32 (emu.m_iregs[rs1 as usize] as i32, RoundingMode::TowardZero) ; emu.m_fregs[rd as usize] = to_data.bits() as u64; }, CallFcvtIdx::D_WU => { let to_data = F64::from_u32 (emu.m_iregs[rs1 as usize] as u32, RoundingMode::TowardZero) ; emu.m_fregs[rd as usize] = to_data.bits() as u64; }, CallFcvtIdx::L_S => { let to_data = F32::from_bits(emu.m_fregs[rs1 as usize] as u32).to_i64(RoundingMode::TowardZero, true); emu.m_iregs[rd as usize] = to_data as u64; }, CallFcvtIdx::LU_S => { let to_data = F32::from_bits(emu.m_fregs[rs1 as usize] as u32).to_u64(RoundingMode::TowardZero, true); emu.m_iregs[rd as usize] = to_data as u64; }, CallFcvtIdx::S_L => { let to_data = F32::from_i64 (emu.m_iregs[rs1 as usize] as i64, RoundingMode::TowardZero) ; emu.m_fregs[rd as usize] = to_data.bits() as u64; }, CallFcvtIdx::S_LU => { let to_data = F32::from_u64 (emu.m_iregs[rs1 as usize] as u64, RoundingMode::TiesToEven) ; emu.m_fregs[rd as usize] = to_data.bits() as u64; }, CallFcvtIdx::L_D => { let to_data = F64::from_bits(emu.m_fregs[rs1 as usize] as u64).to_i64(RoundingMode::TowardZero, true); emu.m_iregs[rd as usize] = to_data as u64; }, CallFcvtIdx::LU_D => { let to_data = F64::from_bits(emu.m_fregs[rs1 as usize] as u64).to_u64(RoundingMode::TowardZero, true); emu.m_iregs[rd as usize] = to_data as u64; }, CallFcvtIdx::D_L => { let to_data = F64::from_i64 (emu.m_iregs[rs1 as usize] as i64, RoundingMode::TowardZero) ; emu.m_fregs[rd as usize] = to_data.bits() as u64; }, CallFcvtIdx::D_LU => { let to_data = F64::from_u64 (emu.m_iregs[rs1 as usize] as u64, RoundingMode::TiesToEven) ; emu.m_fregs[rd as usize] = to_data.bits() as u64; }, }; flag.get(); let ret_flag = flag.bits(); println!("ret_flags = {:x}", ret_flag); emu.m_csr.csrrw(CsrAddr::FFlags, ret_flag as i64); return 0; }
途中でSoftFloatのラッパーにオプションが必要だったので、PRを投げるとすぐに受け入れてくれた。ありがたい。
これだけの実装で、FCVT命令は動くようになった。テストパタンの動作も問題無いようだ。
$ cargo run -- --step --dump-gpr --dump-fpr --dump-guest --dump-host --elf-file rv64uf-p-fcvt
========= BLOCK START ========= 85: Guest PC Address = 800001ac 00000000800001ac:00000000800001ac Hostcode d0257053 : fcvt.s.l ft0, a0 00007FD7A9770000 48BF28B449F9FF7F0000 movabs $0x7FFF_F949_B428,%rdi 00007FD7A977000A 48BE0C00000000000000 movabs $0xC,%rsi 00007FD7A9770014 48BA0000000000000000 movabs $0,%rdx 00007FD7A977001E 48B90A00000000000000 movabs $0xA,%rcx 00007FD7A9770028 FF95E0040000 callq *0x4E0(%rbp) 00007FD7A977002E E9DCFF4E00 jmp 0x0000_7FD7_A9C6_000F ret_flags = 0 x00(zero ) = 0000000000000000 x01(ra ) = 0000000000000000 x02(sp ) = 0000000000000000 x03(gp ) = 0000000000000006 x04(tp ) = 0000000000000000 x05(t0 ) = 0000000080000108 x06(t1 ) = 0000000000000000 x07(t2 ) = 0000000000000000 x08(s0/fp) = 0000000000000000 x09(s1 ) = 0000000000000000 x10(a0 ) = 0000000000000002 x11(a1 ) = 0000000000000000 x12(a2 ) = 0000000000000000 x13(a3 ) = 0000000040000000 x14(a4 ) = 0000000000000000 x15(a5 ) = 0000000000000000 x16(a6 ) = 0000000000000000 x17(a7 ) = 0000000000000000 x18(s2 ) = 0000000000000000 x19(s3 ) = 0000000000000000 x20(s4 ) = 0000000000000000 x21(s5 ) = 0000000000000000 x22(s6 ) = 0000000000000000 x23(s7 ) = 0000000000000000 x24(s8 ) = 0000000000000000 x25(s9 ) = 0000000000000000 x26(s10 ) = 0000000000000000 x27(s11 ) = 0000000000000000 x28(t3 ) = 0000000000000000 x29(t4 ) = 0000000000000000 x30(t5 ) = 0000000000000000 x31(t6 ) = 0000000000000000 f00(ft0 ) = 0000000040000000 f01(ft1 ) = 0000000000000000 f02(ft2 ) = 0000000000000000 f03(ft3 ) = 0000000000000000 f04(ft4 ) = 0000000000000000 f05(ft5 ) = 0000000000000000 f06(ft6 ) = 0000000000000000 f07(ft7 ) = 0000000000000000 f08(fs0 ) = 0000000000000000 f09(fs1 ) = 0000000000000000 f10(fa0 ) = 0000000000000000 f11(fa1 ) = 0000000000000000 f12(fa2 ) = 0000000000000000 f13(fa3 ) = 0000000000000000 f14(fa4 ) = 0000000000000000 f15(fa5 ) = 0000000000000000 f16(fa6 ) = 0000000000000000 f17(fa7 ) = 0000000000000000 f18(fs2 ) = 0000000000000000 f19(fs3 ) = 0000000000000000 f20(fs4 ) = 0000000000000000 f21(fs5 ) = 0000000000000000 f22(fs6 ) = 0000000000000000 f23(fs7 ) = 0000000000000000 f24(fs8 ) = 0000000000000000 f25(fs9 ) = 0000000000000000 f26(fs10 ) = 0000000000000000 f27(fs11 ) = 0000000000000000 f28(ft8 ) = 0000000000000000 f29(ft9 ) = 0000000000000000 f30(ft10 ) = 0000000000000000 f31(ft11 ) = 0000000000000000 ========= BLOCK START =========
変換されたx86命令としては少なめ。全部ヘルパー関数による処理になってしまっている。
