diff options
author | Jin Guojie <jinguojie@loongson.cn> | 2017-01-05 12:57:49 +0800 |
---|---|---|
committer | Richard Henderson <rth@twiddle.net> | 2017-01-06 10:03:54 -0800 |
commit | 7f54eaa3b78d71cb57e45a719980f9b5ff06d21c (patch) | |
tree | ed8a7f68d11704f5aa62521d528b02333bf458d6 /tcg/mips/tcg-target.inc.c | |
parent | 0119b1927d531f3fac22b9b4da01dafc23644973 (diff) |
tcg-mips: Add bswap32u and bswap64
Without the mips32r2 instructions to perform swapping, bswap is quite large,
dominating the size of each reverse-endian qemu_ld/qemu_st operation.
Create two subroutines in the prologue block. The subroutines require extra
reserved registers (TCG_TMP[2, 3]). Using these within qemu_ld means that
we need not place additional restrictions on the qemu_ld outputs.
Tested-by: Aurelien Jarno <aurelien@aurel32.net>
Tested-by: James Hogan <james.hogan@imgtec.com>
Tested-by: YunQiang Su <wzssyqa@gmail.com>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Jin Guojie <jinguojie@loongson.cn>
Message-Id: <1483592275-4496-5-git-send-email-jinguojie@loongson.cn>
Diffstat (limited to 'tcg/mips/tcg-target.inc.c')
-rw-r--r-- | tcg/mips/tcg-target.inc.c | 102 |
1 files changed, 99 insertions, 3 deletions
diff --git a/tcg/mips/tcg-target.inc.c b/tcg/mips/tcg-target.inc.c index 2d91d0cff1..ec139cdfeb 100644 --- a/tcg/mips/tcg-target.inc.c +++ b/tcg/mips/tcg-target.inc.c @@ -125,6 +125,8 @@ static const TCGReg tcg_target_call_oarg_regs[2] = { static tcg_insn_unit *tb_ret_addr; static tcg_insn_unit *bswap32_addr; +static tcg_insn_unit *bswap32u_addr; +static tcg_insn_unit *bswap64_addr; static inline uint32_t reloc_pc16_val(tcg_insn_unit *pc, tcg_insn_unit *target) { @@ -622,7 +624,10 @@ static void tcg_out_bswap32u(TCGContext *s, TCGReg ret, TCGReg arg) tcg_out_opc_reg(s, OPC_DSHD, ret, 0, ret); tcg_out_dsrl(s, ret, ret, 32); } else { - abort(); + tcg_out_bswap_subr(s, bswap32u_addr); + /* delay slot -- never omit the insn, like tcg_out_mov might. */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP0, arg, TCG_REG_ZERO); + tcg_out_mov(s, TCG_TYPE_I32, ret, TCG_TMP3); } } @@ -632,7 +637,10 @@ static void tcg_out_bswap64(TCGContext *s, TCGReg ret, TCGReg arg) tcg_out_opc_reg(s, OPC_DSBH, ret, 0, arg); tcg_out_opc_reg(s, OPC_DSHD, ret, 0, ret); } else { - abort(); + tcg_out_bswap_subr(s, bswap64_addr); + /* delay slot -- never omit the insn, like tcg_out_mov might. */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP0, arg, TCG_REG_ZERO); + tcg_out_mov(s, TCG_TYPE_I32, ret, TCG_TMP3); } } @@ -2279,7 +2287,7 @@ static void tcg_target_qemu_prologue(TCGContext *s) return; } - /* Bswap subroutine: Input in TCG_TMP0, output in TCG_TMP3; + /* Bswap subroutines: Input in TCG_TMP0, output in TCG_TMP3; clobbers TCG_TMP1, TCG_TMP2. */ /* @@ -2305,6 +2313,94 @@ static void tcg_target_qemu_prologue(TCGContext *s) tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_RA, 0); /* t3 = dcba -- delay slot */ tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); + + if (TCG_TARGET_REG_BITS == 32) { + return; + } + + /* + * bswap32u -- unsigned 32-bit swap. a0 = ....abcd. + */ + bswap32u_addr = align_code_ptr(s); + /* t1 = (0000)000d */ + tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP0, 0xff); + /* t3 = 000a */ + tcg_out_opc_sa(s, OPC_SRL, TCG_TMP3, TCG_TMP0, 24); + /* t1 = (0000)d000 */ + tcg_out_dsll(s, TCG_TMP1, TCG_TMP1, 24); + /* t2 = 00c0 */ + tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP2, TCG_TMP0, 0xff00); + /* t3 = d00a */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); + /* t1 = 0abc */ + tcg_out_opc_sa(s, OPC_SRL, TCG_TMP1, TCG_TMP0, 8); + /* t2 = 0c00 */ + tcg_out_opc_sa(s, OPC_SLL, TCG_TMP2, TCG_TMP2, 8); + /* t1 = 00b0 */ + tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP1, 0xff00); + /* t3 = dc0a */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP2); + tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_RA, 0); + /* t3 = dcba -- delay slot */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); + + /* + * bswap64 -- 64-bit swap. a0 = abcdefgh + */ + bswap64_addr = align_code_ptr(s); + /* t3 = h0000000 */ + tcg_out_dsll(s, TCG_TMP3, TCG_TMP0, 56); + /* t1 = 0000000a */ + tcg_out_dsrl(s, TCG_TMP1, TCG_TMP0, 56); + + /* t2 = 000000g0 */ + tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP2, TCG_TMP0, 0xff00); + /* t3 = h000000a */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); + /* t1 = 00000abc */ + tcg_out_dsrl(s, TCG_TMP1, TCG_TMP0, 40); + /* t2 = 0g000000 */ + tcg_out_dsll(s, TCG_TMP2, TCG_TMP2, 40); + /* t1 = 000000b0 */ + tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP1, 0xff00); + + /* t3 = hg00000a */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP2); + /* t2 = 0000abcd */ + tcg_out_dsrl(s, TCG_TMP2, TCG_TMP0, 32); + /* t3 = hg0000ba */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); + + /* t1 = 000000c0 */ + tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP2, 0xff00); + /* t2 = 0000000d */ + tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP2, TCG_TMP2, 0x00ff); + /* t1 = 00000c00 */ + tcg_out_dsll(s, TCG_TMP1, TCG_TMP1, 8); + /* t2 = 0000d000 */ + tcg_out_dsll(s, TCG_TMP2, TCG_TMP2, 24); + + /* t3 = hg000cba */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); + /* t1 = 00abcdef */ + tcg_out_dsrl(s, TCG_TMP1, TCG_TMP0, 16); + /* t3 = hg00dcba */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP2); + + /* t2 = 0000000f */ + tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP2, TCG_TMP1, 0x00ff); + /* t1 = 000000e0 */ + tcg_out_opc_imm(s, OPC_ANDI, TCG_TMP1, TCG_TMP1, 0xff00); + /* t2 = 00f00000 */ + tcg_out_dsll(s, TCG_TMP2, TCG_TMP2, 40); + /* t1 = 000e0000 */ + tcg_out_dsll(s, TCG_TMP1, TCG_TMP1, 24); + + /* t3 = hgf0dcba */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP2); + tcg_out_opc_reg(s, OPC_JR, 0, TCG_REG_RA, 0); + /* t3 = hgfedcba -- delay slot */ + tcg_out_opc_reg(s, OPC_OR, TCG_TMP3, TCG_TMP3, TCG_TMP1); } static void tcg_target_init(TCGContext *s) |