tcg-aarch64 improvements
v4 cover tcg-aarch64 improvements
Changes since v3:
* Using TCGType instead of bool for the ext variables.
* Marginal tidying of the tcg_fmt_* functions to match fields names with the architecture document now that it's released. In particular, the ext variables incoming to them have been renamed "sf". I have not propagated that change all the way back up the call chain.
* Split up the movi improvements into 3 separate patches. I also fixed the bug therein that affected i386-softmmu booting. I've also significantly improved the documentation in that function.
* I came up with a significantly faster way to actually perform the MOVN check, although I didn't confuse the issue by actually implementing it, since it requires using the neon unit, and probably real hardware to see what kind of inter-unit slow down there is. It's left as a comment. I actually suspect that the parallelism of the A57 is high enough to support performing 3 tests at once, deciding which of the 3 alternatives at once at the end.
* Split up the movcond addition and setcond cleanup. I did not find a problem in either patch, as suggested during the previous review.
* Rebase vs head, 2d1fe18.
Tested bios boots of i386-softmmu, alpha-softmmu, sparc-softmmu. Tested a full run of Paulo's sieve{32,64}.flat. Takes around an hour to run under the Foundation model. Ouch.
r~
Richard Henderson (33):
tcg-aarch64: Change all ext variables to TCGType
tcg-aarch64: Set ext based on TCG_OPF_64BIT
tcg-aarch64: Don't handle mov/movi in tcg_out_op
tcg-aarch64: Hoist common argument loads in tcg_out_op
tcg-aarch64: Change enum aarch64_arith_opc to AArch64Insn
tcg-aarch64: Merge enum aarch64_srr_opc with AArch64Insn
tcg-aarch64: Remove the shift_imm parameter from tcg_out_cmp
tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
tcg-aarch64: Introduce tcg_fmt_Rdn_aimm
tcg-aarch64: Implement mov with tcg_fmt_* functions
tcg-aarch64: Handle constant operands to add, sub, and compare
tcg-aarch64: Handle constant operands to and, or, xor
tcg-aarch64: Support andc, orc, eqv, not
tcg-aarch64: Handle zero as first argument to sub
tcg-aarch64: Support movcond
tcg-aarch64: Use tcg_fmt_Rdnm_cond for setcond
tcg-aarch64: Support deposit
tcg-aarch64: Support add2, sub2
tcg-aarch64: Support muluh, mulsh
tcg-aarch64: Support div, rem
tcg-aarch64: Introduce tcg_fmt_Rd_uimm
tcg-aarch64: Use MOVN in tcg_out_movi
tcg-aarch64: Use ORRI in tcg_out_movi
tcg-aarch64: Special case small constants in tcg_out_movi
tcg-aarch64: Use adrp in tcg_out_movi
tcg-aarch64: Avoid add with zero in tlb load
tcg-aarch64: Pass return address to load/store helpers directly.
tcg-aarch64: Use tcg_out_call for qemu_ld/st
tcg-aarch64: Use symbolic names for branches
tcg-aarch64: Implement tcg_register_jit
tcg-aarch64: Reuse FP and LR in translated code
tcg-aarch64: Introduce tcg_out_ldst_pair
tcg-aarch64: Remove redundant CPU_TLB_ENTRY_BITS check
include/exec/exec-all.h | 11 - tcg/aarch64/tcg-target.c | 1352 ++++++++++++++++++++++++++++++---------------- tcg/aarch64/tcg-target.h | 76 +-- 3 files changed, 940 insertions(+), 499 deletions(-)
Signature · 1 line
1.8.3.1
PATCH v4 1/33 tcg-aarch64: Change all ext variables to TCGType
We assert that the values for _I32 and _I64 are 0 and 1 respectively. This will make a couple of functions declared by tcg.c cleaner.
--- tcg/aarch64/tcg-target.c | 64 ++++++++++++++++++++++++++++-------------------- 1 file changed, 37 insertions(+), 27 deletions(-)
tcg/aarch64/tcg-target.c +37 -27
@@ -12,6 +12,11 @@#include "qemu/bitops.h" +/* We're going to re-use TCGType in setting of the SF bit, which controls +the size of the operation performed. If we know the values match, it +makes things much cleaner. */ +QEMU_BUILD_BUG_ON (TCG_TYPE_I32 !=0 ||TCG_TYPE_I64 !=1 ); +#ifndef NDEBUG static const char *const tcg_target_reg_names[TCG_TARGET_NB_REGS ] = { "%x0", "%x1", "%x2", "%x3", "%x4", "%x5", "%x6", "%x7",
@@ -326,7 +331,8 @@ static inline void tcg_out_ldst_12(TCGContext *s, | op_type << 20 | scaled_uimm << 10 | rn << 5 | rd); } -static inline void tcg_out_movr (TCGContext *s,int ext,TCGReg rd,TCGReg src) +static inline void tcg_out_movr (TCGContext *s,TCGType ext, +TCGReg rd,TCGReg src) {/* register to register move using MOV (shifted register with no shift) */ /* using MOV 0x2a0003e0 | (shift).. */
@@ -407,7 +413,8 @@ static inline void tcg_out_ldst(TCGContext *s, enum aarch64_ldst_op_data data, }/* mov alias implemented with add immediate, useful to move to/from SP */ -static inline void tcg_out_movr_sp (TCGContext *s,int ext,TCGReg rd,TCGReg rn) +static inline void tcg_out_movr_sp (TCGContext *s,TCGType ext, +TCGReg rd,TCGReg rn) {/* using ADD 0x11000000 | (ext) | rn << 5 | rd */ unsigned int base = ext ?0x91000000 :0x11000000 ;
@@ -437,7 +444,7 @@ static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, }static inline void tcg_out_arith (TCGContext *s,enum aarch64_arith_opc opc, -int ext,TCGReg rd,TCGReg rn,TCGReg rm, +TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm,int shift_imm) {/* Using shifted register arithmetic operations */
@@ -453,7 +460,7 @@ static inline void tcg_out_arith(TCGContext *s, enum aarch64_arith_opc opc,tcg_out32 (s, base | rm <<16 | shift | rn <<5 | rd); } -static inline void tcg_out_mul (TCGContext *s,int ext, +static inline void tcg_out_mul (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm) {/* Using MADD 0x1b000000 with Ra = wzr alias MUL 0x1b007c00 */
@@ -462,7 +469,7 @@ static inline void tcg_out_mul(TCGContext *s, int ext, }static inline void tcg_out_shiftrot_reg (TCGContext *s, -enum aarch64_srr_opc opc,int ext, +enum aarch64_srr_opc opc,TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm) {/* using 2-source data processing instructions 0x1ac02000 */
@@ -470,23 +477,23 @@ static inline void tcg_out_shiftrot_reg(TCGContext *s,tcg_out32 (s, base | rm <<16 | opc <<8 | rn <<5 | rd); } -static inline void tcg_out_ubfm (TCGContext *s,int ext,TCGReg rd,TCGReg rn, -unsigned int a,unsigned int b) +static inline void tcg_out_ubfm (TCGContext *s,TCGType ext,TCGReg rd, +TCGReg rn,unsigned int a,unsigned int b) {/* Using UBFM 0x53000000 Wd, Wn, a, b */ unsigned int base = ext ?0xd3400000 :0x53000000 ;tcg_out32 (s, base | a <<16 | b <<10 | rn <<5 | rd); } -static inline void tcg_out_sbfm (TCGContext *s,int ext,TCGReg rd,TCGReg rn, -unsigned int a,unsigned int b) +static inline void tcg_out_sbfm (TCGContext *s,TCGType ext,TCGReg rd, +TCGReg rn,unsigned int a,unsigned int b) {/* Using SBFM 0x13000000 Wd, Wn, a, b */ unsigned int base = ext ?0x93400000 :0x13000000 ;tcg_out32 (s, base | a <<16 | b <<10 | rn <<5 | rd); } -static inline void tcg_out_extr (TCGContext *s,int ext,TCGReg rd, +static inline void tcg_out_extr (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm,unsigned int a) {/* Using EXTR 0x13800000 Wd, Wn, Wm, a */
@@ -494,7 +501,7 @@ static inline void tcg_out_extr(TCGContext *s, int ext, TCGReg rd,tcg_out32 (s, base | rm <<16 | a <<10 | rn <<5 | rd); } -static inline void tcg_out_shl (TCGContext *s,int ext, +static inline void tcg_out_shl (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int m) {int bits, max;
@@ -503,28 +510,28 @@ static inline void tcg_out_shl(TCGContext *s, int ext,tcg_out_ubfm (s, ext, rd, rn, bits - (m & max), max - (m & max)); } -static inline void tcg_out_shr (TCGContext *s,int ext, +static inline void tcg_out_shr (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int m) {int max = ext ?63 :31 ;tcg_out_ubfm (s, ext, rd, rn, m & max, max); } -static inline void tcg_out_sar (TCGContext *s,int ext, +static inline void tcg_out_sar (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int m) {int max = ext ?63 :31 ;tcg_out_sbfm (s, ext, rd, rn, m & max, max); } -static inline void tcg_out_rotr (TCGContext *s,int ext, +static inline void tcg_out_rotr (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int m) {int max = ext ?63 :31 ;tcg_out_extr (s, ext, rd, rn, rn, m & max); } -static inline void tcg_out_rotl (TCGContext *s,int ext, +static inline void tcg_out_rotl (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int m) {int bits, max;
@@ -533,14 +540,15 @@ static inline void tcg_out_rotl(TCGContext *s, int ext,tcg_out_extr (s, ext, rd, rn, rn, bits - (m & max)); } -static inline void tcg_out_cmp (TCGContext *s,int ext,TCGReg rn,TCGReg rm, -int shift_imm) +static inline void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg rn, +TCGReg rm,int shift_imm) {/* Using CMP alias SUBS wzr, Wn, Wm */ tcg_out_arith (s,ARITH_SUBS , ext,TCG_REG_XZR , rn, rm, shift_imm); } -static inline void tcg_out_cset (TCGContext *s,int ext,TCGReg rd,TCGCond c) +static inline void tcg_out_cset (TCGContext *s,TCGType ext, +TCGReg rd,TCGCond c) {/* Using CSET alias of CSINC 0x1a800400 Xd, XZR, XZR, invert(cond) */ unsigned int base = ext ?0x9a9f07e0 :0x1a9f07e0 ;
@@ -637,7 +645,7 @@ aarch64_limm(unsigned int m, unsigned int r)to test a 32bit reg against 0xff000000, pass M = 8, R = 8. to test a 32bit reg against 0xff0000ff, pass M = 16, R = 8. */ -static inline void tcg_out_tst (TCGContext *s,int ext,TCGReg rn, +static inline void tcg_out_tst (TCGContext *s,TCGType ext,TCGReg rn,unsigned int m,unsigned int r) {/* using TST alias of ANDS XZR, Xn,#bimm64 0x7200001f */
@@ -646,8 +654,8 @@ static inline void tcg_out_tst(TCGContext *s, int ext, TCGReg rn, }/* and a register with a bit pattern, similarly to TST, no flags change */ -static inline void tcg_out_andi (TCGContext *s,int ext,TCGReg rd,TCGReg rn, -unsigned int m,unsigned int r) +static inline void tcg_out_andi (TCGContext *s,TCGType ext,TCGReg rd, +TCGReg rn,unsigned int m,unsigned int r) {/* using AND 0x12000000 */ unsigned int base = ext ?0x92400000 :0x12000000 ;
@@ -700,21 +708,23 @@ static inline void tcg_out_goto_label_cond(TCGContext *s, } } -static inline void tcg_out_rev (TCGContext *s,int ext,TCGReg rd,TCGReg rm) +static inline void tcg_out_rev (TCGContext *s,TCGType ext, +TCGReg rd,TCGReg rm) {/* using REV 0x5ac00800 */ unsigned int base = ext ?0xdac00c00 :0x5ac00800 ;tcg_out32 (s, base | rm <<5 | rd); } -static inline void tcg_out_rev16 (TCGContext *s,int ext,TCGReg rd,TCGReg rm) +static inline void tcg_out_rev16 (TCGContext *s,TCGType ext, +TCGReg rd,TCGReg rm) {/* using REV16 0x5ac00400 */ unsigned int base = ext ?0xdac00400 :0x5ac00400 ;tcg_out32 (s, base | rm <<5 | rd); } -static inline void tcg_out_sxt (TCGContext *s,int ext,int s_bits, +static inline void tcg_out_sxt (TCGContext *s,TCGType ext,int s_bits,TCGReg rd,TCGReg rn) {/* using ALIASes SXTB 0x13001c00, SXTH 0x13003c00, SXTW 0x93407c00
@@ -732,7 +742,7 @@ static inline void tcg_out_uxt(TCGContext *s, int s_bits,tcg_out_ubfm (s,0 , rd, rn,0 , bits); } -static inline void tcg_out_addi (TCGContext *s,int ext, +static inline void tcg_out_addi (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int aimm) {/* add immediate aimm unsigned 12bit value (with LSL 0 or 12) */
@@ -752,7 +762,7 @@ static inline void tcg_out_addi(TCGContext *s, int ext,tcg_out32 (s, base | aimm | (rn <<5 ) | rd); } -static inline void tcg_out_subi (TCGContext *s,int ext, +static inline void tcg_out_subi (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int aimm) {/* sub immediate aimm unsigned 12bit value (with LSL 0 or 12) */
@@ -1107,7 +1117,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, {/* ext will be set in the switch below, which will fall through to the common code. It triggers the use of extended regs where appropriate. */ -int ext =0 ; +TCGType ext =0 ;switch (opc) {case INDEX_op_exit_tb:
Signature · 1 line
1.8.3.1
PATCH v4 2/33 tcg-aarch64: Set ext based on TCG_OPF_64BIT
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 28 +++++++--------------------- 1 file changed, 7 insertions(+), 21 deletions(-)
tcg/aarch64/tcg-target.c +7 -21
@@ -1115,9 +1115,9 @@ static inline void tcg_out_load_pair(TCGContext *s, TCGReg addr,static void tcg_out_op (TCGContext *s,TCGOpcode opc,const TCGArg *args,const int *const_args) { -/* ext will be set in the switch below, which will fall through to the -common code. It triggers the use of extended regs where appropriate. */ -TCGType ext =0 ; +/* 99% of the time, we can signal the use of extension registers +by looking to see if the opcode handles 64-bit data. */ +TCGType ext = (tcg_op_defs[opc].flags &TCG_OPF_64BIT ) !=0 ;switch (opc) {case INDEX_op_exit_tb:
@@ -1173,7 +1173,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,break ;case INDEX_op_mov_i64: - ext =1 ;/* fall through */ case INDEX_op_mov_i32:tcg_out_movr (s, ext, args[0 ], args[1 ]);break ;
@@ -1186,43 +1185,36 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,break ;case INDEX_op_add_i64: - ext =1 ;/* fall through */ case INDEX_op_add_i32:tcg_out_arith (s,ARITH_ADD , ext, args[0 ], args[1 ], args[2 ],0 );break ;case INDEX_op_sub_i64: - ext =1 ;/* fall through */ case INDEX_op_sub_i32:tcg_out_arith (s,ARITH_SUB , ext, args[0 ], args[1 ], args[2 ],0 );break ;case INDEX_op_and_i64: - ext =1 ;/* fall through */ case INDEX_op_and_i32:tcg_out_arith (s,ARITH_AND , ext, args[0 ], args[1 ], args[2 ],0 );break ;case INDEX_op_or_i64: - ext =1 ;/* fall through */ case INDEX_op_or_i32:tcg_out_arith (s,ARITH_OR , ext, args[0 ], args[1 ], args[2 ],0 );break ;case INDEX_op_xor_i64: - ext =1 ;/* fall through */ case INDEX_op_xor_i32:tcg_out_arith (s,ARITH_XOR , ext, args[0 ], args[1 ], args[2 ],0 );break ;case INDEX_op_mul_i64: - ext =1 ;/* fall through */ case INDEX_op_mul_i32:tcg_out_mul (s, ext, args[0 ], args[1 ], args[2 ]);break ;case INDEX_op_shl_i64: - ext =1 ;/* fall through */ case INDEX_op_shl_i32:if (const_args[2 ]) {/* LSL / UBFM Wd, Wn, (32 - m) */ tcg_out_shl (s, ext, args[0 ], args[1 ], args[2 ]);
@@ -1232,7 +1224,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,break ;case INDEX_op_shr_i64: - ext =1 ;/* fall through */ case INDEX_op_shr_i32:if (const_args[2 ]) {/* LSR / UBFM Wd, Wn, m, 31 */ tcg_out_shr (s, ext, args[0 ], args[1 ], args[2 ]);
@@ -1242,7 +1233,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,break ;case INDEX_op_sar_i64: - ext =1 ;/* fall through */ case INDEX_op_sar_i32:if (const_args[2 ]) {/* ASR / SBFM Wd, Wn, m, 31 */ tcg_out_sar (s, ext, args[0 ], args[1 ], args[2 ]);
@@ -1252,7 +1242,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,break ;case INDEX_op_rotr_i64: - ext =1 ;/* fall through */ case INDEX_op_rotr_i32:if (const_args[2 ]) {/* ROR / EXTR Wd, Wm, Wm, m */ tcg_out_rotr (s, ext, args[0 ], args[1 ], args[2 ]);
@@ -1262,7 +1251,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,break ;case INDEX_op_rotl_i64: - ext =1 ;/* fall through */ case INDEX_op_rotl_i32:/* same as rotate right by (32 - m) */ if (const_args[2 ]) {/* ROR / EXTR Wd, Wm, Wm, 32 - m */ tcg_out_rotl (s, ext, args[0 ], args[1 ], args[2 ]);
@@ -1275,14 +1263,12 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,break ;case INDEX_op_brcond_i64: - ext =1 ;/* fall through */ case INDEX_op_brcond_i32:/* CMP 0, 1, cond(2), label 3 */ tcg_out_cmp (s, ext, args[0 ], args[1 ],0 );tcg_out_goto_label_cond (s, args[2 ], args[3 ]);break ;case INDEX_op_setcond_i64: - ext =1 ;/* fall through */ case INDEX_op_setcond_i32:tcg_out_cmp (s, ext, args[1 ], args[2 ],0 );tcg_out_cset (s,0 , args[0 ], args[3 ]);
@@ -1325,9 +1311,11 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,tcg_out_qemu_st (s, args,3 );break ; -case INDEX_op_bswap64_i64: - ext =1 ;/* fall through */ case INDEX_op_bswap32_i64: +/* Despite the _i64, this is a 32-bit bswap. */ + ext =0 ; +/* FALLTHRU */ +case INDEX_op_bswap64_i64:case INDEX_op_bswap32_i32:tcg_out_rev (s, ext, args[0 ], args[1 ]);break ;
@@ -1337,12 +1325,10 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,break ;case INDEX_op_ext8s_i64: - ext =1 ;/* fall through */ case INDEX_op_ext8s_i32:tcg_out_sxt (s, ext,0 , args[0 ], args[1 ]);break ;case INDEX_op_ext16s_i64: - ext =1 ;/* fall through */ case INDEX_op_ext16s_i32:tcg_out_sxt (s, ext,1 , args[0 ], args[1 ]);break ;
Signature · 1 line
1.8.3.1
PATCH v4 3/33 tcg-aarch64: Don't handle mov/movi in tcg_out_op
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 20 +++++++------------- 1 file changed, 7 insertions(+), 13 deletions(-)
tcg/aarch64/tcg-target.c +7 -13
@@ -1172,18 +1172,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, args[0], args[1], args[2]);break ; -case INDEX_op_mov_i64: -case INDEX_op_mov_i32: -tcg_out_movr (s, ext, args[0 ], args[1 ]); -break ; - -case INDEX_op_movi_i64: -tcg_out_movi (s,TCG_TYPE_I64 , args[0 ], args[1 ]); -break ; -case INDEX_op_movi_i32: -tcg_out_movi (s,TCG_TYPE_I32 , args[0 ], args[1 ]); -break ; -case INDEX_op_add_i64:case INDEX_op_add_i32:tcg_out_arith (s,ARITH_ADD , ext, args[0 ], args[1 ], args[2 ],0 );
@@ -1347,8 +1335,14 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,tcg_out_movr (s,0 , args[0 ], args[1 ]);break ; +case INDEX_op_mov_i64: +case INDEX_op_mov_i32: +case INDEX_op_movi_i64: +case INDEX_op_movi_i32: +/* Always implemented with tcg_out_mov/i, never with tcg_out_op. */ default : -tcg_abort ();/* opcode not implemented */ +/* Opcode not implemented. */ +tcg_abort (); } }
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 03/33] tcg-aarch64: Don't handle mov/movi in tcg_out_op
Quoted 29 lines · Richard Henderson
Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 20 +++++++------------- 1 file changed, 7 insertions(+), 13 deletions(-)tcg/aarch64/tcg-target.c +0
-12@@ -1172,18 +1172,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, args[0], args[1], args[2]);break ; -case INDEX_op_mov_i64: -case INDEX_op_mov_i32: -tcg_out_movr (s, ext, args[0 ], args[1 ]); -break ; - -case INDEX_op_movi_i64: -tcg_out_movi (s,TCG_TYPE_I64 , args[0 ], args[1 ]); -break ; -case INDEX_op_movi_i32: -tcg_out_movi (s,TCG_TYPE_I32 , args[0 ], args[1 ]); -break ; -case INDEX_op_add_i64:case INDEX_op_add_i32:tcg_out_arith (s,ARITH_ADD , ext, args[0 ], args[1 ], args[2 ],0 );
tcg/aarch64/tcg-target.c static void tcg_out_op(TCGContext *s, TCGOpcode opc, · line 1342
Quoted 8 lines
tcg/aarch64/tcg-target.c +5
-0@@ -1347,8 +1335,14 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,tcg_out_movr (s,0 , args[0 ], args[1 ]);break ; +case INDEX_op_mov_i64: +case INDEX_op_mov_i32: +case INDEX_op_movi_i64: +case INDEX_op_movi_i32: +/* Always implemented with tcg_out_mov/i, never with tcg_out_op. */
Ok (as in, "I am ok with moving these labels here").
tcg/aarch64/tcg-target.c static void tcg_out_op(TCGContext *s, TCGOpcode opc, · line 1347
Quoted 6 lines
tcg/aarch64/tcg-target.c +2
-1default : -tcg_abort ();/* opcode not implemented */ +/* Opcode not implemented. */ +tcg_abort (); } }
This change above seems unnecessary.
C.
Re: [Qemu-devel] [PATCH v4 03/33] tcg-aarch64: Don't handle mov/movi in tcg_out_op
Quoted 7 lines · Claudio Fontana
Quoted 6 lines
tcg/aarch64/tcg-target.c static void tcg_out_op(TCGContext *s, TCGOpcode opc, · line 1347
Quoted 6 lines
default:
tcg/aarch64/tcg-target.c +2
-1-tcg_abort ();/* opcode not implemented */ +/* Opcode not implemented. */ +tcg_abort (); } }This change above seems unnecessary.
Perhaps qemu doesn't have the same "comments are sentences" rule that gcc does, but I still aim to follow it.
r~
Re: [Qemu-devel] [PATCH v4 03/33] tcg-aarch64: Don't handle mov/movi in tcg_out_op
Quoted 7 lines · Richard Henderson
Quoted 6 lines · Claudio Fontana
Quoted 6 lines
Quoted 6 lines
default:
tcg/aarch64/tcg-target.c +2
-1-tcg_abort ();/* opcode not implemented */ +/* Opcode not implemented. */ +tcg_abort (); } }
Quoted 5 lines
Quoted 1 line
This change above seems unnecessary.
Perhaps qemu doesn't have the same "comments are sentences" rule that gcc does, but I still aim to follow it.
It is wasteful of y space for no reason. This comment is not a sentence, when you read the code as you would prose. It is a comment _on_ tcg_abort() : tcg_abort(), because the opcode is not implemented.
In other cases, the comment alone on a line is a good idea, but not in this one.
C.
PATCH v4 4/33 tcg-aarch64: Hoist common argument loads in tcg_out_op
In reply to Richard HendersonThis reduces the code size of the function significantly.
--- tcg/aarch64/tcg-target.c | 95 +++++++++++++++++++++++++----------------------- 1 file changed, 50 insertions(+), 45 deletions(-)
tcg/aarch64/tcg-target.c +50 -45
@@ -1113,15 +1113,22 @@ static inline void tcg_out_load_pair(TCGContext *s, TCGReg addr, }static void tcg_out_op (TCGContext *s,TCGOpcode opc, -const TCGArg *args,const int *const_args) +const TCGArg args[TCG_MAX_OP_ARGS ], +const int const_args[TCG_MAX_OP_ARGS ]) {/* 99% of the time, we can signal the use of extension registers by looking to see if the opcode handles 64-bit data. */ TCGType ext = (tcg_op_defs[opc].flags &TCG_OPF_64BIT ) !=0 ; +/* Hoist the loads of the most common arguments. */ +TCGArg a0 = args[0 ]; +TCGArg a1 = args[1 ]; +TCGArg a2 = args[2 ]; +int c2 = const_args[2 ]; +switch (opc) {case INDEX_op_exit_tb: -tcg_out_movi (s,TCG_TYPE_I64 ,TCG_REG_X0 , args[0 ]); +tcg_out_movi (s,TCG_TYPE_I64 ,TCG_REG_X0 , a0);tcg_out_goto (s, (tcg_target_long )tb_ret_addr);break ;
@@ -1130,23 +1137,23 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,#error "USE_DIRECT_JUMP required for aarch64"#endif assert (s->tb_jmp_offset !=NULL );/* consistency for USE_DIRECT_JUMP */ - s->tb_jmp_offset[args[0 ]] = s->code_ptr - s->code_buf; + s->tb_jmp_offset[a0] = s->code_ptr - s->code_buf;/* actual branch destination will be patched by aarch64_tb_set_jmp_target later, beware retranslation. */ tcg_out_goto_noaddr (s); - s->tb_next_offset[args[0 ]] = s->code_ptr - s->code_buf; + s->tb_next_offset[a0] = s->code_ptr - s->code_buf;break ;case INDEX_op_call:if (const_args[0 ]) { -tcg_out_call (s, args[0 ]); +tcg_out_call (s, a0); }else { -tcg_out_callr (s, args[0 ]); +tcg_out_callr (s, a0); }break ;case INDEX_op_br: -tcg_out_goto_label (s, args[0 ]); +tcg_out_goto_label (s, a0);break ;case INDEX_op_ld_i32:
@@ -1169,97 +1176,95 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_st16_i64:case INDEX_op_st32_i64:tcg_out_ldst (s,aarch64_ldst_get_data (opc),aarch64_ldst_get_type (opc), - args[0 ], args[1 ], args[2 ]); + a0, a1, a2);break ;case INDEX_op_add_i64:case INDEX_op_add_i32: -tcg_out_arith (s,ARITH_ADD , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_ADD , ext, a0, a1, a2,0 );break ;case INDEX_op_sub_i64:case INDEX_op_sub_i32: -tcg_out_arith (s,ARITH_SUB , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_SUB , ext, a0, a1, a2,0 );break ;case INDEX_op_and_i64:case INDEX_op_and_i32: -tcg_out_arith (s,ARITH_AND , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_AND , ext, a0, a1, a2,0 );break ;case INDEX_op_or_i64:case INDEX_op_or_i32: -tcg_out_arith (s,ARITH_OR , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_OR , ext, a0, a1, a2,0 );break ;case INDEX_op_xor_i64:case INDEX_op_xor_i32: -tcg_out_arith (s,ARITH_XOR , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_XOR , ext, a0, a1, a2,0 );break ;case INDEX_op_mul_i64:case INDEX_op_mul_i32: -tcg_out_mul (s, ext, args[0 ], args[1 ], args[2 ]); +tcg_out_mul (s, ext, a0, a1, a2);break ;case INDEX_op_shl_i64:case INDEX_op_shl_i32: -if (const_args[2 ]) {/* LSL / UBFM Wd, Wn, (32 - m) */ -tcg_out_shl (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* LSL / UBFM Wd, Wn, (32 - m) */ +tcg_out_shl (s, ext, a0, a1, a2); }else {/* LSL / LSLV */ -tcg_out_shiftrot_reg (s,SRR_SHL , ext, args[0 ], args[1 ], args[2 ]); +tcg_out_shiftrot_reg (s,SRR_SHL , ext, a0, a1, a2); }break ;case INDEX_op_shr_i64:case INDEX_op_shr_i32: -if (const_args[2 ]) {/* LSR / UBFM Wd, Wn, m, 31 */ -tcg_out_shr (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* LSR / UBFM Wd, Wn, m, 31 */ +tcg_out_shr (s, ext, a0, a1, a2); }else {/* LSR / LSRV */ -tcg_out_shiftrot_reg (s,SRR_SHR , ext, args[0 ], args[1 ], args[2 ]); +tcg_out_shiftrot_reg (s,SRR_SHR , ext, a0, a1, a2); }break ;case INDEX_op_sar_i64:case INDEX_op_sar_i32: -if (const_args[2 ]) {/* ASR / SBFM Wd, Wn, m, 31 */ -tcg_out_sar (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* ASR / SBFM Wd, Wn, m, 31 */ +tcg_out_sar (s, ext, a0, a1, a2); }else {/* ASR / ASRV */ -tcg_out_shiftrot_reg (s,SRR_SAR , ext, args[0 ], args[1 ], args[2 ]); +tcg_out_shiftrot_reg (s,SRR_SAR , ext, a0, a1, a2); }break ;case INDEX_op_rotr_i64:case INDEX_op_rotr_i32: -if (const_args[2 ]) {/* ROR / EXTR Wd, Wm, Wm, m */ -tcg_out_rotr (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* ROR / EXTR Wd, Wm, Wm, m */ +tcg_out_rotr (s, ext, a0, a1, a2); }else {/* ROR / RORV */ -tcg_out_shiftrot_reg (s,SRR_ROR , ext, args[0 ], args[1 ], args[2 ]); +tcg_out_shiftrot_reg (s,SRR_ROR , ext, a0, a1, a2); }break ;case INDEX_op_rotl_i64:case INDEX_op_rotl_i32:/* same as rotate right by (32 - m) */ -if (const_args[2 ]) {/* ROR / EXTR Wd, Wm, Wm, 32 - m */ -tcg_out_rotl (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* ROR / EXTR Wd, Wm, Wm, 32 - m */ +tcg_out_rotl (s, ext, a0, a1, a2); }else { -tcg_out_arith (s,ARITH_SUB ,0 , -TCG_REG_TMP ,TCG_REG_XZR , args[2 ],0 ); -tcg_out_shiftrot_reg (s,SRR_ROR , ext, - args[0 ], args[1 ],TCG_REG_TMP ); +tcg_out_arith (s,ARITH_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2,0 ); +tcg_out_shiftrot_reg (s,SRR_ROR , ext, a0, a1,TCG_REG_TMP ); }break ;case INDEX_op_brcond_i64: -case INDEX_op_brcond_i32:/* CMP 0, 1, cond(2), label 3 */ -tcg_out_cmp (s, ext, args[0 ], args[1 ],0 ); -tcg_out_goto_label_cond (s, args[2 ], args[3 ]); +case INDEX_op_brcond_i32: +tcg_out_cmp (s, ext, a0, a1,0 ); +tcg_out_goto_label_cond (s, a2, args[3 ]);break ;case INDEX_op_setcond_i64:case INDEX_op_setcond_i32: -tcg_out_cmp (s, ext, args[1 ], args[2 ],0 ); -tcg_out_cset (s,0 , args[0 ], args[3 ]); +tcg_out_cmp (s, ext, a1, a2,0 ); +tcg_out_cset (s,0 , a0, args[3 ]);break ;case INDEX_op_qemu_ld8u:
@@ -1305,34 +1310,34 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,/* FALLTHRU */ case INDEX_op_bswap64_i64:case INDEX_op_bswap32_i32: -tcg_out_rev (s, ext, args[0 ], args[1 ]); +tcg_out_rev (s, ext, a0, a1);break ;case INDEX_op_bswap16_i64:case INDEX_op_bswap16_i32: -tcg_out_rev16 (s,0 , args[0 ], args[1 ]); +tcg_out_rev16 (s,0 , a0, a1);break ;case INDEX_op_ext8s_i64:case INDEX_op_ext8s_i32: -tcg_out_sxt (s, ext,0 , args[0 ], args[1 ]); +tcg_out_sxt (s, ext,0 , a0, a1);break ;case INDEX_op_ext16s_i64:case INDEX_op_ext16s_i32: -tcg_out_sxt (s, ext,1 , args[0 ], args[1 ]); +tcg_out_sxt (s, ext,1 , a0, a1);break ;case INDEX_op_ext32s_i64: -tcg_out_sxt (s,1 ,2 , args[0 ], args[1 ]); +tcg_out_sxt (s,1 ,2 , a0, a1);break ;case INDEX_op_ext8u_i64:case INDEX_op_ext8u_i32: -tcg_out_uxt (s,0 , args[0 ], args[1 ]); +tcg_out_uxt (s,0 , a0, a1);break ;case INDEX_op_ext16u_i64:case INDEX_op_ext16u_i32: -tcg_out_uxt (s,1 , args[0 ], args[1 ]); +tcg_out_uxt (s,1 , a0, a1);break ;case INDEX_op_ext32u_i64: -tcg_out_movr (s,0 , args[0 ], args[1 ]); +tcg_out_movr (s,0 , a0, a1);break ;case INDEX_op_mov_i64:
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 04/33] tcg-aarch64: Hoist common argument loads in tcg_out_op
Hello Richard,
Quoted 19 lines · Richard Henderson
This reduces the code size of the function significantly.
Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 95 +++++++++++++++++++++++++----------------------- 1 file changed, 50 insertions(+), 45 deletions(-)tcg/aarch64/tcg-target.c +2
-1@@ -1113,15 +1113,22 @@ static inline void tcg_out_load_pair(TCGContext *s, TCGReg addr, }static void tcg_out_op (TCGContext *s,TCGOpcode opc, -const TCGArg *args,const int *const_args) +const TCGArg args[TCG_MAX_OP_ARGS ], +const int const_args[TCG_MAX_OP_ARGS ]) {
tcg/aarch64/tcg-target.c static inline void tcg_out_load_pair(TCGContext *s, TCGReg addr, · line 1128
Quoted 10 lines
tcg/aarch64/tcg-target.c +6
-0/* 99% of the time, we can signal the use of extension registers by looking to see if the opcode handles 64-bit data. */ TCGType ext = (tcg_op_defs[opc].flags &TCG_OPF_64BIT ) !=0 ; +/* Hoist the loads of the most common arguments. */ +TCGArg a0 = args[0 ]; +TCGArg a1 = args[1 ]; +TCGArg a2 = args[2 ]; +int c2 = const_args[2 ]; +
Either all or none (add c0, c1), I would expect the compiler not to generate code for the paths that don't use C[n].
Btw, if the compiler generates bloated code without this, we should notify the projects working on gcc for aarch64.
Quoted 196 lines
tcg/aarch64/tcg-target.c +40
-42switch (opc) {case INDEX_op_exit_tb: -tcg_out_movi (s,TCG_TYPE_I64 ,TCG_REG_X0 , args[0 ]); +tcg_out_movi (s,TCG_TYPE_I64 ,TCG_REG_X0 , a0);tcg_out_goto (s, (tcg_target_long )tb_ret_addr);break ;@@ -1130,23 +1137,23 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,#error "USE_DIRECT_JUMP required for aarch64"#endif assert (s->tb_jmp_offset !=NULL );/* consistency for USE_DIRECT_JUMP */ - s->tb_jmp_offset[args[0 ]] = s->code_ptr - s->code_buf; + s->tb_jmp_offset[a0] = s->code_ptr - s->code_buf;/* actual branch destination will be patched by aarch64_tb_set_jmp_target later, beware retranslation. */ tcg_out_goto_noaddr (s); - s->tb_next_offset[args[0 ]] = s->code_ptr - s->code_buf; + s->tb_next_offset[a0] = s->code_ptr - s->code_buf;break ;case INDEX_op_call:if (const_args[0 ]) { -tcg_out_call (s, args[0 ]); +tcg_out_call (s, a0); }else { -tcg_out_callr (s, args[0 ]); +tcg_out_callr (s, a0); }break ;case INDEX_op_br: -tcg_out_goto_label (s, args[0 ]); +tcg_out_goto_label (s, a0);break ;case INDEX_op_ld_i32:@@ -1169,97 +1176,95 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_st16_i64:case INDEX_op_st32_i64:tcg_out_ldst (s,aarch64_ldst_get_data (opc),aarch64_ldst_get_type (opc), - args[0 ], args[1 ], args[2 ]); + a0, a1, a2);break ;case INDEX_op_add_i64:case INDEX_op_add_i32: -tcg_out_arith (s,ARITH_ADD , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_ADD , ext, a0, a1, a2,0 );break ;case INDEX_op_sub_i64:case INDEX_op_sub_i32: -tcg_out_arith (s,ARITH_SUB , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_SUB , ext, a0, a1, a2,0 );break ;case INDEX_op_and_i64:case INDEX_op_and_i32: -tcg_out_arith (s,ARITH_AND , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_AND , ext, a0, a1, a2,0 );break ;case INDEX_op_or_i64:case INDEX_op_or_i32: -tcg_out_arith (s,ARITH_OR , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_OR , ext, a0, a1, a2,0 );break ;case INDEX_op_xor_i64:case INDEX_op_xor_i32: -tcg_out_arith (s,ARITH_XOR , ext, args[0 ], args[1 ], args[2 ],0 ); +tcg_out_arith (s,ARITH_XOR , ext, a0, a1, a2,0 );break ;case INDEX_op_mul_i64:case INDEX_op_mul_i32: -tcg_out_mul (s, ext, args[0 ], args[1 ], args[2 ]); +tcg_out_mul (s, ext, a0, a1, a2);break ;case INDEX_op_shl_i64:case INDEX_op_shl_i32: -if (const_args[2 ]) {/* LSL / UBFM Wd, Wn, (32 - m) */ -tcg_out_shl (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* LSL / UBFM Wd, Wn, (32 - m) */ +tcg_out_shl (s, ext, a0, a1, a2); }else {/* LSL / LSLV */ -tcg_out_shiftrot_reg (s,SRR_SHL , ext, args[0 ], args[1 ], args[2 ]); +tcg_out_shiftrot_reg (s,SRR_SHL , ext, a0, a1, a2); }break ;case INDEX_op_shr_i64:case INDEX_op_shr_i32: -if (const_args[2 ]) {/* LSR / UBFM Wd, Wn, m, 31 */ -tcg_out_shr (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* LSR / UBFM Wd, Wn, m, 31 */ +tcg_out_shr (s, ext, a0, a1, a2); }else {/* LSR / LSRV */ -tcg_out_shiftrot_reg (s,SRR_SHR , ext, args[0 ], args[1 ], args[2 ]); +tcg_out_shiftrot_reg (s,SRR_SHR , ext, a0, a1, a2); }break ;case INDEX_op_sar_i64:case INDEX_op_sar_i32: -if (const_args[2 ]) {/* ASR / SBFM Wd, Wn, m, 31 */ -tcg_out_sar (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* ASR / SBFM Wd, Wn, m, 31 */ +tcg_out_sar (s, ext, a0, a1, a2); }else {/* ASR / ASRV */ -tcg_out_shiftrot_reg (s,SRR_SAR , ext, args[0 ], args[1 ], args[2 ]); +tcg_out_shiftrot_reg (s,SRR_SAR , ext, a0, a1, a2); }break ;case INDEX_op_rotr_i64:case INDEX_op_rotr_i32: -if (const_args[2 ]) {/* ROR / EXTR Wd, Wm, Wm, m */ -tcg_out_rotr (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* ROR / EXTR Wd, Wm, Wm, m */ +tcg_out_rotr (s, ext, a0, a1, a2); }else {/* ROR / RORV */ -tcg_out_shiftrot_reg (s,SRR_ROR , ext, args[0 ], args[1 ], args[2 ]); +tcg_out_shiftrot_reg (s,SRR_ROR , ext, a0, a1, a2); }break ;case INDEX_op_rotl_i64:case INDEX_op_rotl_i32:/* same as rotate right by (32 - m) */ -if (const_args[2 ]) {/* ROR / EXTR Wd, Wm, Wm, 32 - m */ -tcg_out_rotl (s, ext, args[0 ], args[1 ], args[2 ]); +if (c2) {/* ROR / EXTR Wd, Wm, Wm, 32 - m */ +tcg_out_rotl (s, ext, a0, a1, a2); }else { -tcg_out_arith (s,ARITH_SUB ,0 , -TCG_REG_TMP ,TCG_REG_XZR , args[2 ],0 ); -tcg_out_shiftrot_reg (s,SRR_ROR , ext, - args[0 ], args[1 ],TCG_REG_TMP ); +tcg_out_arith (s,ARITH_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2,0 ); +tcg_out_shiftrot_reg (s,SRR_ROR , ext, a0, a1,TCG_REG_TMP ); }break ;case INDEX_op_brcond_i64: -case INDEX_op_brcond_i32:/* CMP 0, 1, cond(2), label 3 */ -tcg_out_cmp (s, ext, args[0 ], args[1 ],0 ); -tcg_out_goto_label_cond (s, args[2 ], args[3 ]); +case INDEX_op_brcond_i32: +tcg_out_cmp (s, ext, a0, a1,0 ); +tcg_out_goto_label_cond (s, a2, args[3 ]);break ;case INDEX_op_setcond_i64:case INDEX_op_setcond_i32: -tcg_out_cmp (s, ext, args[1 ], args[2 ],0 ); -tcg_out_cset (s,0 , args[0 ], args[3 ]); +tcg_out_cmp (s, ext, a1, a2,0 ); +tcg_out_cset (s,0 , a0, args[3 ]);break ;case INDEX_op_qemu_ld8u:@@ -1305,34 +1310,34 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,/* FALLTHRU */ case INDEX_op_bswap64_i64:case INDEX_op_bswap32_i32: -tcg_out_rev (s, ext, args[0 ], args[1 ]); +tcg_out_rev (s, ext, a0, a1);break ;case INDEX_op_bswap16_i64:case INDEX_op_bswap16_i32: -tcg_out_rev16 (s,0 , args[0 ], args[1 ]); +tcg_out_rev16 (s,0 , a0, a1);break ;case INDEX_op_ext8s_i64:case INDEX_op_ext8s_i32: -tcg_out_sxt (s, ext,0 , args[0 ], args[1 ]); +tcg_out_sxt (s, ext,0 , a0, a1);break ;case INDEX_op_ext16s_i64:case INDEX_op_ext16s_i32: -tcg_out_sxt (s, ext,1 , args[0 ], args[1 ]); +tcg_out_sxt (s, ext,1 , a0, a1);break ;case INDEX_op_ext32s_i64: -tcg_out_sxt (s,1 ,2 , args[0 ], args[1 ]); +tcg_out_sxt (s,1 ,2 , a0, a1);break ;case INDEX_op_ext8u_i64:case INDEX_op_ext8u_i32: -tcg_out_uxt (s,0 , args[0 ], args[1 ]); +tcg_out_uxt (s,0 , a0, a1);break ;case INDEX_op_ext16u_i64:
tcg/aarch64/tcg-target.c static void tcg_out_op(TCGContext *s, TCGOpcode opc, · line 1343
Quoted 10 lines
tcg/aarch64/tcg-target.c +2
-2case INDEX_op_ext16u_i32: -tcg_out_uxt (s,1 , args[0 ], args[1 ]); +tcg_out_uxt (s,1 , a0, a1);break ;case INDEX_op_ext32u_i64: -tcg_out_movr (s,0 , args[0 ], args[1 ]); +tcg_out_movr (s,0 , a0, a1);break ;case INDEX_op_mov_i64:
Claudio
Re: [Qemu-devel] [PATCH v4 04/33] tcg-aarch64: Hoist common argument loads in tcg_out_op
Quoted 8 lines · Claudio Fontana
Quoted 6 lines
+ /* Hoist the loads of the most common arguments. */
Quoted 5 lines
tcg/aarch64/tcg-target.c +5
-0+TCGArg a0 = args[0 ]; +TCGArg a1 = args[1 ]; +TCGArg a2 = args[2 ]; +int c2 = const_args[2 ]; +Either all or none (add c0, c1), I would expect the compiler not to generate code for the paths that don't use C[n].
I chose the most common. Those used in 90% of all of the cases.
Quoted 3 lines
Btw, if the compiler generates bloated code without this, we should notify the projects working on gcc for aarch64.
It's not the compiler's fault. After parameter decomposition, the arrays become pointers, and the compiler can't tell that it's always safe to perform the loads. So in general it can't hoist the loads higher than the first explicit reference that proves the pointers must be non-null.
Now that I think about it, we might actually do better, generically, to package all of these arguments up into a struct. The compiler can more easily reason about the collective safety of structure access...
r~
Re: [Qemu-devel] [PATCH v4 04/33] tcg-aarch64: Hoist common argument loads in tcg_out_op
Quoted 11 lines · Richard Henderson
Quoted 8 lines · Claudio Fontana
Quoted 6 lines
+ /* Hoist the loads of the most common arguments. */
Quoted 5 lines
tcg/aarch64/tcg-target.c +5
-0+TCGArg a0 = args[0 ]; +TCGArg a1 = args[1 ]; +TCGArg a2 = args[2 ]; +int c2 = const_args[2 ]; +Either all or none (add c0, c1), I would expect the compiler not to generate code for the paths that don't use C[n].
I chose the most common. Those used in 90% of all of the cases.
What you did is clear, does not change the fact that the mixing of use of the variables and args[] is confusing. So either we add int c0 and int c1, and use the hoisted variables exclusively later, or we don't do it.
Quoted 11 lines
Quoted 2 lines
Btw, if the compiler generates bloated code without this, we should notify the projects working on gcc for aarch64.
It's not the compiler's fault. After parameter decomposition, the arrays become pointers, and the compiler can't tell that it's always safe to perform the loads. So in general it can't hoist the loads higher than the first explicit reference that proves the pointers must be non-null.
Now that I think about it, we might actually do better, generically, to package all of these arguments up into a struct. The compiler can more easily reason about the collective safety of structure access...
I don't have anything against it in principle, but just adding c0 and c1, which iirc should cover all uses, would be fine by me.
Claudio
Re: [Qemu-devel] [PATCH v4 04/33] tcg-aarch64: Hoist common argument loads in tcg_out_op
Quoted 2 lines · Claudio Fontana
I don't have anything against it in principle, but just adding c0 and c1, which iirc should cover all uses, would be fine by me.
Not really.
There are 6 potential args[] values, 5 of which might be const_args[].
r~
Re: [Qemu-devel] [PATCH v4 04/33] tcg-aarch64: Hoist common argument loads in tcg_out_op
Quoted 8 lines · Richard Henderson
Quoted 2 lines · Claudio Fontana
I don't have anything against it in principle, but just adding c0 and c1, which iirc should cover all uses, would be fine by me.
Not really.
There are 6 potential args[] values, 5 of which might be const_args[].
Current uses. Like the args[] that are actually used right now.
Re: [Qemu-devel] [PATCH v4 04/33] tcg-aarch64: Hoist common argument loads in tcg_out_op
Quoted 11 lines · Claudio Fontana
Quoted 8 lines · Richard Henderson
Quoted 2 lines · Claudio Fontana
I don't have anything against it in principle, but just adding c0 and c1, which iirc should cover all uses, would be fine by me.
Not really.
There are 6 potential args[] values, 5 of which might be const_args[].
Current uses. Like the args[] that are actually used right now.
That's what I'm talking about. E.g. movcond: 6 args[], 1-4 are inputs that might have const_args[].
r~
Re: [Qemu-devel] [PATCH v4 04/33] tcg-aarch64: Hoist common argument loads in tcg_out_op
In reply to Richard HendersonQuoted 10 lines · Richard Henderson
Quoted 2 lines · Claudio Fontana
I don't have anything against it in principle, but just adding c0 and c1, which iirc should cover all uses, would be fine by me.
Not really.
There are 6 potential args[] values, 5 of which might be const_args[].
r~
Ok adding all of those would be a waste, did not keep the movcond changes into consideration. Your initial proposal was indeed the best tradeoff then, lets just hoist the most commonly used args then.
C.
PATCH v4 5/33 tcg-aarch64: Change enum aarch64_arith_opc to AArch64Insn
In reply to Richard HendersonAnd since we're no longer talking about opcodes, change the values to be shifted into the opcode field, avoiding a shift at runtime.
--- tcg/aarch64/tcg-target.c | 43 +++++++++++++++++++++++-------------------- 1 file changed, 23 insertions(+), 20 deletions(-)
tcg/aarch64/tcg-target.c +23 -20
@@ -204,16 +204,19 @@ enum aarch64_ldst_op_type { /* type of operation */ LDST_LD_S_W = 0xc, /* load and sign-extend into Wt */ }; -enum aarch64_arith_opc { -ARITH_AND =0x0a , -ARITH_ADD =0x0b , -ARITH_OR =0x2a , -ARITH_ADDS =0x2b , -ARITH_XOR =0x4a , -ARITH_SUB =0x4b , -ARITH_ANDS =0x6a , -ARITH_SUBS =0x6b , -}; +typedef enum { +/* Logical shifted register instructions */ +INSN_AND =0x0a000000 , +INSN_ORR =0x2a000000 , +INSN_EOR =0x4a000000 , +INSN_ANDS =0x6a000000 , + +/* Add/subtract shifted register instructions */ +INSN_ADD =0x0b000000 , +INSN_ADDS =0x2b000000 , +INSN_SUB =0x4b000000 , +INSN_SUBS =0x6b000000 , +}AArch64Insn ;enum aarch64_srr_opc {SRR_SHL =0x0 ,
@@ -443,13 +446,13 @@ static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, arg, arg1, arg2); } -static inline void tcg_out_arith (TCGContext *s,enum aarch64_arith_opc opc, +static inline void tcg_out_arith (TCGContext *s,AArch64Insn insn,TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm,int shift_imm) {/* Using shifted register arithmetic operations */ /* if extended register operation (64bit) just OR with 0x80 << 24 */ -unsigned int shift, base = ext ? (0x80 | opc) <<24 : opc <<24 ; +unsigned int shift, base = insn | (ext ?0x80000000 :0 );if (shift_imm ==0 ) { shift =0 ; }else if (shift_imm >0 ) {
@@ -544,7 +547,7 @@ static inline void tcg_out_cmp(TCGContext *s, TCGType ext, TCGReg rn, TCGReg rm, int shift_imm) {/* Using CMP alias SUBS wzr, Wn, Wm */ -tcg_out_arith (s ,ARITH_SUBS ,ext ,TCG_REG_XZR ,rn ,rm ,shift_imm ); +tcg_out_arith (s ,INSN_SUBS ,ext ,TCG_REG_XZR ,rn ,rm ,shift_imm ); } static inline void tcg_out_cset(TCGContext *s, TCGType ext,
@@ -904,7 +907,7 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg,tcg_out_addi (s,1 ,TCG_REG_X2 , base, tlb_offset &0xfff000 );/* Merge the tlb index contribution into X2. X2 = X2 + (X0 << CPU_TLB_ENTRY_BITS) */ -tcg_out_arith (s,ARITH_ADD ,1 ,TCG_REG_X2 ,TCG_REG_X2 , +tcg_out_arith (s,INSN_ADD ,1 ,TCG_REG_X2 ,TCG_REG_X2 ,TCG_REG_X0 , -CPU_TLB_ENTRY_BITS );/* Merge "low bits" from tlb offset, load the tlb comparator into X0. X0 = load [X2 + (tlb_offset & 0x000fff)] */
@@ -1181,27 +1184,27 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_add_i64:case INDEX_op_add_i32: -tcg_out_arith (s,ARITH_ADD , ext, a0, a1, a2,0 ); +tcg_out_arith (s,INSN_ADD , ext, a0, a1, a2,0 );break ;case INDEX_op_sub_i64:case INDEX_op_sub_i32: -tcg_out_arith (s,ARITH_SUB , ext, a0, a1, a2,0 ); +tcg_out_arith (s,INSN_SUB , ext, a0, a1, a2,0 );break ;case INDEX_op_and_i64:case INDEX_op_and_i32: -tcg_out_arith (s,ARITH_AND , ext, a0, a1, a2,0 ); +tcg_out_arith (s,INSN_AND , ext, a0, a1, a2,0 );break ;case INDEX_op_or_i64:case INDEX_op_or_i32: -tcg_out_arith (s,ARITH_OR , ext, a0, a1, a2,0 ); +tcg_out_arith (s,INSN_ORR , ext, a0, a1, a2,0 );break ;case INDEX_op_xor_i64:case INDEX_op_xor_i32: -tcg_out_arith (s,ARITH_XOR , ext, a0, a1, a2,0 ); +tcg_out_arith (s,INSN_EOR , ext, a0, a1, a2,0 );break ;case INDEX_op_mul_i64:
@@ -1250,7 +1253,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {/* ROR / EXTR Wd, Wm, Wm, 32 - m */ tcg_out_rotl (s, ext, a0, a1, a2); }else { -tcg_out_arith (s,ARITH_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2,0 ); +tcg_out_arith (s,INSN_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2,0 );tcg_out_shiftrot_reg (s,SRR_ROR , ext, a0, a1,TCG_REG_TMP ); }break ;
Signature · 1 line
1.8.3.1
PATCH v4 6/33 tcg-aarch64: Merge enum aarch64_srr_opc with AArch64Insn
In reply to Richard HendersonAnd since we're no longer talking about opcodes, merge the 0x1ac02000 data2 primary opcode with the shift subcode to create the full insn.
--- tcg/aarch64/tcg-target.c | 49 ++++++++++++++++++++++++------------------------ 1 file changed, 24 insertions(+), 25 deletions(-)
tcg/aarch64/tcg-target.c +24 -25
@@ -216,14 +216,13 @@ typedef enum {INSN_ADDS =0x2b000000 ,INSN_SUB =0x4b000000 ,INSN_SUBS =0x6b000000 , -} AArch64Insn; -enum aarch64_srr_opc { -SRR_SHL =0x0 , -SRR_SHR =0x4 , -SRR_SAR =0x8 , -SRR_ROR =0xc -}; +/* Data-processing (2 source) instructions */ +INSN_LSLV =0x1ac02000 , +INSN_LSRV =0x1ac02400 , +INSN_ASRV =0x1ac02800 , +INSN_RORV =0x1ac02c00 , +} AArch64Insn;static inline enum aarch64_ldst_op_data aarch64_ldst_get_data (TCGOpcode tcg_op)
@@ -472,12 +471,12 @@ static inline void tcg_out_mul(TCGContext *s, TCGType ext, }static inline void tcg_out_shiftrot_reg (TCGContext *s, -enum aarch64_srr_opc opc,TCGType ext, +AArch64Insn insn,TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm) {/* using 2-source data processing instructions 0x1ac02000 */ -unsigned int base = ext ?0x9ac02000 :0x1ac02000 ; -tcg_out32 (s, base | rm <<16 | opc <<8 | rn <<5 | rd); +unsigned int base = insn | (ext ?0x80000000 :0 ); +tcg_out32 (s, base | rm <<16 | rn <<5 | rd); } static inline void tcg_out_ubfm(TCGContext *s, TCGType ext, TCGReg rd,
@@ -1214,47 +1213,47 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_shl_i64:case INDEX_op_shl_i32: -if (c2) {/* LSL / UBFM Wd, Wn, (32 - m) */ +if (c2) {tcg_out_shl (s, ext, a0, a1, a2); - }else {/* LSL / LSLV */ -tcg_out_shiftrot_reg (s,SRR_SHL , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_LSLV , ext, a0, a1, a2); }break ;case INDEX_op_shr_i64:case INDEX_op_shr_i32: -if (c2) {/* LSR / UBFM Wd, Wn, m, 31 */ +if (c2) {tcg_out_shr (s, ext, a0, a1, a2); - }else {/* LSR / LSRV */ -tcg_out_shiftrot_reg (s,SRR_SHR , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_LSRV , ext, a0, a1, a2); }break ;case INDEX_op_sar_i64:case INDEX_op_sar_i32: -if (c2) {/* ASR / SBFM Wd, Wn, m, 31 */ +if (c2) {tcg_out_sar (s, ext, a0, a1, a2); - }else {/* ASR / ASRV */ -tcg_out_shiftrot_reg (s,SRR_SAR , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_ASRV , ext, a0, a1, a2); }break ;case INDEX_op_rotr_i64:case INDEX_op_rotr_i32: -if (c2) {/* ROR / EXTR Wd, Wm, Wm, m */ +if (c2) {tcg_out_rotr (s, ext, a0, a1, a2); - }else {/* ROR / RORV */ -tcg_out_shiftrot_reg (s,SRR_ROR , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_RORV , ext, a0, a1, a2); }break ;case INDEX_op_rotl_i64: -case INDEX_op_rotl_i32:/* same as rotate right by (32 - m) */ -if (c2) {/* ROR / EXTR Wd, Wm, Wm, 32 - m */ +case INDEX_op_rotl_i32: +if (c2) {tcg_out_rotl (s, ext, a0, a1, a2); }else {tcg_out_arith (s,INSN_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2,0 ); -tcg_out_shiftrot_reg (s,SRR_ROR , ext, a0, a1,TCG_REG_TMP ); +tcg_out_shiftrot_reg (s,INSN_RORV , ext, a0, a1,TCG_REG_TMP ); }break ;
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 06/33] tcg-aarch64: Merge enum aarch64_srr_opc with AArch64Insn
Quoted 52 lines · Richard Henderson
And since we're no longer talking about opcodes, merge the
0x1ac02000data2 primary opcode with the shift subcode to create the full insn.Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 49 ++++++++++++++++++++++++------------------------ 1 file changed, 24 insertions(+), 25 deletions(-)tcg/aarch64/tcg-target.c +9
-10@@ -216,14 +216,13 @@ typedef enum {INSN_ADDS =0x2b000000 ,INSN_SUB =0x4b000000 ,INSN_SUBS =0x6b000000 , -} AArch64Insn; -enum aarch64_srr_opc { -SRR_SHL =0x0 , -SRR_SHR =0x4 , -SRR_SAR =0x8 , -SRR_ROR =0xc -}; +/* Data-processing (2 source) instructions */ +INSN_LSLV =0x1ac02000 , +INSN_LSRV =0x1ac02400 , +INSN_ASRV =0x1ac02800 , +INSN_RORV =0x1ac02c00 , +} AArch64Insn;static inline enum aarch64_ldst_op_data aarch64_ldst_get_data (TCGOpcode tcg_op)@@ -472,12 +471,12 @@ static inline void tcg_out_mul(TCGContext *s, TCGType ext, }static inline void tcg_out_shiftrot_reg (TCGContext *s, -enum aarch64_srr_opc opc,TCGType ext, +AArch64Insn insn,TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm) {/* using 2-source data processing instructions 0x1ac02000 */ -unsigned int base = ext ?0x9ac02000 :0x1ac02000 ; -tcg_out32 (s, base | rm <<16 | opc <<8 | rn <<5 | rd); +unsigned int base = insn | (ext ?0x80000000 :0 ); +tcg_out32 (s, base | rm <<16 | rn <<5 | rd); } static inline void tcg_out_ubfm(TCGContext *s, TCGType ext, TCGReg rd,@@ -1214,47 +1213,47 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_shl_i64:
tcg/aarch64/tcg-target.c static void tcg_out_op(TCGContext *s, TCGOpcode opc, · line 1221
Quoted 10 lines
tcg/aarch64/tcg-target.c +3
-3case INDEX_op_shl_i32: -if (c2) {/* LSL / UBFM Wd, Wn, (32 - m) */ +if (c2) {tcg_out_shl (s, ext, a0, a1, a2); - }else {/* LSL / LSLV */ -tcg_out_shiftrot_reg (s,SRR_SHL , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_LSLV , ext, a0, a1, a2); }break ;
Any reason you strip all comments out?
They are supposed to hint the reader about how the tcg operation is implemented.
Quoted 50 lines
tcg/aarch64/tcg-target.c +12
-12case INDEX_op_shr_i64:case INDEX_op_shr_i32: -if (c2) {/* LSR / UBFM Wd, Wn, m, 31 */ +if (c2) {tcg_out_shr (s, ext, a0, a1, a2); - }else {/* LSR / LSRV */ -tcg_out_shiftrot_reg (s,SRR_SHR , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_LSRV , ext, a0, a1, a2); }break ;case INDEX_op_sar_i64:case INDEX_op_sar_i32: -if (c2) {/* ASR / SBFM Wd, Wn, m, 31 */ +if (c2) {tcg_out_sar (s, ext, a0, a1, a2); - }else {/* ASR / ASRV */ -tcg_out_shiftrot_reg (s,SRR_SAR , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_ASRV , ext, a0, a1, a2); }break ;case INDEX_op_rotr_i64:case INDEX_op_rotr_i32: -if (c2) {/* ROR / EXTR Wd, Wm, Wm, m */ +if (c2) {tcg_out_rotr (s, ext, a0, a1, a2); - }else {/* ROR / RORV */ -tcg_out_shiftrot_reg (s,SRR_ROR , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_RORV , ext, a0, a1, a2); }break ;case INDEX_op_rotl_i64: -case INDEX_op_rotl_i32:/* same as rotate right by (32 - m) */ -if (c2) {/* ROR / EXTR Wd, Wm, Wm, 32 - m */ +case INDEX_op_rotl_i32: +if (c2) {tcg_out_rotl (s, ext, a0, a1, a2); }else {tcg_out_arith (s,INSN_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2,0 ); -tcg_out_shiftrot_reg (s,SRR_ROR , ext, a0, a1,TCG_REG_TMP ); +tcg_out_shiftrot_reg (s,INSN_RORV , ext, a0, a1,TCG_REG_TMP ); }break ;
Re: [Qemu-devel] [PATCH v4 06/33] tcg-aarch64: Merge enum aarch64_srr_opc with AArch64Insn
Quoted 11 lines · Claudio Fontana
Quoted 11 lines
tcg/aarch64/tcg-target.c static void tcg_out_op(TCGContext *s, TCGOpcode opc, · line 1221
Quoted 11 lines
case INDEX_op_shl_i64: case INDEX_op_shl_i32:tcg/aarch64/tcg-target.c +3
-3-if (c2) {/* LSL / UBFM Wd, Wn, (32 - m) */ +if (c2) {tcg_out_shl (s, ext, a0, a1, a2); - }else {/* LSL / LSLV */ -tcg_out_shiftrot_reg (s,SRR_SHL , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_LSLV , ext, a0, a1, a2); }break ;
Quoted 3 lines
Any reason you strip all comments out?
They are supposed to hint the reader about how the tcg operation is implemented.
Well, LSLV is now there in the code in the form of INSN_LSLV. The comment does nothing additional to hint at what's going on; indeed, "LSL / LSLV" is more confusing than not.
As for tcg_out_shl, comments about its implementation are with that function.
r~
Re: [Qemu-devel] [PATCH v4 06/33] tcg-aarch64: Merge enum aarch64_srr_opc with AArch64Insn
Quoted 16 lines · Richard Henderson
Quoted 11 lines · Claudio Fontana
Quoted 11 lines
tcg/aarch64/tcg-target.c static void tcg_out_op(TCGContext *s, TCGOpcode opc, · line 1221
Quoted 11 lines
case INDEX_op_shl_i64: case INDEX_op_shl_i32:tcg/aarch64/tcg-target.c +3
-3-if (c2) {/* LSL / UBFM Wd, Wn, (32 - m) */ +if (c2) {tcg_out_shl (s, ext, a0, a1, a2); - }else {/* LSL / LSLV */ -tcg_out_shiftrot_reg (s,SRR_SHL , ext, a0, a1, a2); + }else { +tcg_out_shiftrot_reg (s,INSN_LSLV , ext, a0, a1, a2); }break ;Quoted 3 lines
Any reason you strip all comments out?
They are supposed to hint the reader about how the tcg operation is implemented.
Quoted 7 lines
Well, LSLV is now there in the code in the form of
INSN_LSLV. The comment does nothing additional to hint at what's going on; indeed, "LSL / LSLV" is more confusing than not.As for
tcg_out_shl, comments about its implementation are with that function.
Fair enough.
C.
PATCH v4 7/33 tcg-aarch64: Remove the shift_imm parameter from tcg_out_cmp
In reply to Richard HendersonIt was unused. Let's not overcomplicate things before we need them.
--- tcg/aarch64/tcg-target.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
tcg/aarch64/tcg-target.c +5 -5
@@ -543,10 +543,10 @@ static inline void tcg_out_rotl(TCGContext *s, TCGType ext, }static inline void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg rn, -TCGReg rm,int shift_imm) +TCGReg rm) {/* Using CMP alias SUBS wzr, Wn, Wm */ -tcg_out_arith (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm, shift_imm); +tcg_out_arith (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm,0 ); } static inline void tcg_out_cset(TCGContext *s, TCGType ext,
@@ -920,7 +920,7 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, (is_read ? offsetof(CPUTLBEntry , addr_read) : offsetof(CPUTLBEntry, addr_write)));/* Perform the address comparison. */ -tcg_out_cmp (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 ,TCG_REG_X3 ,0 ); +tcg_out_cmp (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 ,TCG_REG_X3 ); *label_ptr = s->code_ptr;/* If not equal, we jump to the slow path. */ tcg_out_goto_cond_noaddr (s,TCG_COND_NE );
@@ -1259,13 +1259,13 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_brcond_i64:case INDEX_op_brcond_i32: -tcg_out_cmp (s, ext, a0, a1,0 ); +tcg_out_cmp (s, ext, a0, a1);tcg_out_goto_label_cond (s, a2, args[3 ]);break ;case INDEX_op_setcond_i64:case INDEX_op_setcond_i32: -tcg_out_cmp (s, ext, a1, a2,0 ); +tcg_out_cmp (s, ext, a1, a2);tcg_out_cset (s,0 , a0, args[3 ]);break ;
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 07/33] tcg-aarch64: Remove the shift_imm parameter from tcg_out_cmp
Quoted 40 lines · Richard Henderson
It was unused. Let's not overcomplicate things before we need them.
Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)tcg/aarch64/tcg-target.c +4
-4@@ -543,10 +543,10 @@ static inline void tcg_out_rotl(TCGContext *s, TCGType ext, }static inline void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg rn, -TCGReg rm,int shift_imm) +TCGReg rm) {/* Using CMP alias SUBS wzr, Wn, Wm */ -tcg_out_arith (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm, shift_imm); +tcg_out_arith (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm,0 ); } static inline void tcg_out_cset(TCGContext *s, TCGType ext,@@ -920,7 +920,7 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, (is_read ? offsetof(CPUTLBEntry , addr_read) : offsetof(CPUTLBEntry, addr_write)));/* Perform the address comparison. */ -tcg_out_cmp (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 ,TCG_REG_X3 ,0 ); +tcg_out_cmp (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 ,TCG_REG_X3 ); *label_ptr = s->code_ptr;/* If not equal, we jump to the slow path. */ tcg_out_goto_cond_noaddr (s,TCG_COND_NE );@@ -1259,13 +1259,13 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_brcond_i64:case INDEX_op_brcond_i32: -tcg_out_cmp (s, ext, a0, a1,0 ); +tcg_out_cmp (s, ext, a0, a1);
tcg/aarch64/tcg-target.c static void tcg_out_op(TCGContext *s, TCGOpcode opc, · line 1271
Quoted 10 lines
tcg/aarch64/tcg-target.c +1
-1tcg_out_goto_label_cond (s, a2, args[3 ]);break ;case INDEX_op_setcond_i64:case INDEX_op_setcond_i32: -tcg_out_cmp (s, ext, a1, a2,0 ); +tcg_out_cmp (s, ext, a1, a2);tcg_out_cset (s,0 , a0, args[3 ]);break ;
agreed. This is an artifact from a previous implementation of the tlb lookup.
PATCH v4 8/33 tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
In reply to Richard HendersonNow that we've converted opcode fields to pre-shifted insns, we can merge the implementation of arithmetic and shift insns.
Simplify the left/right shift parameter to just the left shift needed by tcg_out_tlb_read.
--- tcg/aarch64/tcg-target.c | 78 +++++++++++++++++++++++------------------------- 1 file changed, 38 insertions(+), 40 deletions(-)
tcg/aarch64/tcg-target.c +38 -40
@@ -302,6 +302,30 @@ static inline uint32_t tcg_in32(TCGContext *s)return v; } +/* +* Encode various formats. +*/ + +/* This function can be used for both Arithmetic and Logical (shifted register) +type insns. Since we don't actually use the other available shifts, we only +support LSL here. */ +static inline void tcg_fmt_Rdnm_lsl (TCGContext *s,AArch64Insn insn, +TCGType sf,TCGReg rd,TCGReg rn, +TCGReg rm,int imm6) +{ +/* Note that LSL is bits {23,22} = 0. */ +tcg_out32 (s, insn | sf <<31 | imm6 <<10 | rm <<16 | rn <<5 | rd); +} + +/* This function can be used for most insns with 2 input registers and one +output register. This includes Arithmetic (shifted register, sans shift), +Logical, Shift, Multiply, Divide, and Bit operation. */ +static inline void tcg_fmt_Rdnm (TCGContext *s,AArch64Insn insn,TCGType sf, +TCGReg rd,TCGReg rn,TCGReg rm) +{ +tcg_out32 (s, insn | sf <<31 | rm <<16 | rn <<5 | rd); +} + static inline void tcg_out_ldst_9(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type,
@@ -445,23 +469,6 @@ static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, arg, arg1, arg2); } -static inline void tcg_out_arith (TCGContext *s,AArch64Insn insn, -TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm, -int shift_imm) -{ -/* Using shifted register arithmetic operations */ -/* if extended register operation (64bit) just OR with 0x80 << 24 */ -unsigned int shift, base = insn | (ext ?0x80000000 :0 ); -if (shift_imm ==0 ) { - shift =0 ; - }else if (shift_imm >0 ) { - shift = shift_imm <<10 |1 <<22 ; - }else /* (shift_imm < 0) */ { - shift = (-shift_imm) <<10 ; - } -tcg_out32 (s, base | rm <<16 | shift | rn <<5 | rd); -} -static inline void tcg_out_mul (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm) {
@@ -470,15 +477,6 @@ static inline void tcg_out_mul(TCGContext *s, TCGType ext,tcg_out32 (s, base | rm <<16 | rn <<5 | rd); } -static inline void tcg_out_shiftrot_reg (TCGContext *s, -AArch64Insn insn,TCGType ext, -TCGReg rd,TCGReg rn,TCGReg rm) -{ -/* using 2-source data processing instructions 0x1ac02000 */ -unsigned int base = insn | (ext ?0x80000000 :0 ); -tcg_out32 (s, base | rm <<16 | rn <<5 | rd); -} -static inline void tcg_out_ubfm (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int a,unsigned int b) {
@@ -546,7 +544,7 @@ static inline void tcg_out_cmp(TCGContext *s, TCGType ext, TCGReg rn, TCGReg rm) {/* Using CMP alias SUBS wzr, Wn, Wm */ -tcg_out_arith (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm,0 ); +tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm); } static inline void tcg_out_cset(TCGContext *s, TCGType ext,
@@ -906,8 +904,8 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg,tcg_out_addi (s,1 ,TCG_REG_X2 , base, tlb_offset &0xfff000 );/* Merge the tlb index contribution into X2. X2 = X2 + (X0 << CPU_TLB_ENTRY_BITS) */ -tcg_out_arith (s,INSN_ADD ,1 ,TCG_REG_X2 ,TCG_REG_X2 , -TCG_REG_X0 , -CPU_TLB_ENTRY_BITS ); +tcg_fmt_Rdnm_lsl (s,INSN_ADD ,1 ,TCG_REG_X2 ,TCG_REG_X2 , +TCG_REG_X0 ,CPU_TLB_ENTRY_BITS );/* Merge "low bits" from tlb offset, load the tlb comparator into X0. X0 = load [X2 + (tlb_offset & 0x000fff)] */ tcg_out_ldst(s, TARGET_LONG_BITS == 64 ? LDST_64 : LDST_32,
@@ -1183,27 +1181,27 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_add_i64:case INDEX_op_add_i32: -tcg_out_arith (s,INSN_ADD , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_ADD , ext, a0, a1, a2);break ;case INDEX_op_sub_i64:case INDEX_op_sub_i32: -tcg_out_arith (s,INSN_SUB , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_SUB , ext, a0, a1, a2);break ;case INDEX_op_and_i64:case INDEX_op_and_i32: -tcg_out_arith (s,INSN_AND , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_AND , ext, a0, a1, a2);break ;case INDEX_op_or_i64:case INDEX_op_or_i32: -tcg_out_arith (s,INSN_ORR , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_ORR , ext, a0, a1, a2);break ;case INDEX_op_xor_i64:case INDEX_op_xor_i32: -tcg_out_arith (s,INSN_EOR , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_EOR , ext, a0, a1, a2);break ;case INDEX_op_mul_i64:
@@ -1216,7 +1214,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_shl (s, ext, a0, a1, a2); }else { -tcg_out_shiftrot_reg (s,INSN_LSLV , ext, a0, a1, a2); +tcg_fmt_Rdnm (s,INSN_LSLV , ext, a0, a1, a2); }break ;
@@ -1225,7 +1223,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_shr (s, ext, a0, a1, a2); }else { -tcg_out_shiftrot_reg (s,INSN_LSRV , ext, a0, a1, a2); +tcg_fmt_Rdnm (s,INSN_LSRV , ext, a0, a1, a2); }break ;
@@ -1234,7 +1232,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_sar (s, ext, a0, a1, a2); }else { -tcg_out_shiftrot_reg (s,INSN_ASRV , ext, a0, a1, a2); +tcg_fmt_Rdnm (s,INSN_ASRV , ext, a0, a1, a2); }break ;
@@ -1243,7 +1241,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_rotr (s, ext, a0, a1, a2); }else { -tcg_out_shiftrot_reg (s,INSN_RORV , ext, a0, a1, a2); +tcg_fmt_Rdnm (s,INSN_RORV , ext, a0, a1, a2); }break ;
@@ -1252,8 +1250,8 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_rotl (s, ext, a0, a1, a2); }else { -tcg_out_arith (s,INSN_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2,0 ); -tcg_out_shiftrot_reg (s,INSN_RORV , ext, a0, a1,TCG_REG_TMP ); +tcg_fmt_Rdnm (s,INSN_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2); +tcg_fmt_Rdnm (s,INSN_RORV , ext, a0, a1,TCG_REG_TMP ); }break ;
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 08/33] tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
Quoted 33 lines · Richard Henderson
Now that we've converted opcode fields to pre-shifted insns, we can merge the implementation of arithmetic and shift insns.
Simplify the left/right shift parameter to just the left shift needed by
tcg_out_tlb_read.Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 78 +++++++++++++++++++++++------------------------- 1 file changed, 38 insertions(+), 40 deletions(-)tcg/aarch64/tcg-target.c +14
-0@@ -302,6 +302,30 @@ static inline uint32_t tcg_in32(TCGContext *s)return v; } +/* +* Encode various formats. +*/ + +/* This function can be used for both Arithmetic and Logical (shifted register) +type insns. Since we don't actually use the other available shifts, we only +support LSL here. */ +static inline void tcg_fmt_Rdnm_lsl (TCGContext *s,AArch64Insn insn, +TCGType sf,TCGReg rd,TCGReg rn, +TCGReg rm,int imm6) +{ +/* Note that LSL is bits {23,22} = 0. */ +tcg_out32 (s, insn | sf <<31 | imm6 <<10 | rm <<16 | rn <<5 | rd); +}
tcg/aarch64/tcg-target.c static inline uint32_t tcg_in32(TCGContext *s) · line 328
Quoted 10 lines
tcg/aarch64/tcg-target.c +10
-0+ +/* This function can be used for most insns with 2 input registers and one +output register. This includes Arithmetic (shifted register, sans shift), +Logical, Shift, Multiply, Divide, and Bit operation. */ +static inline void tcg_fmt_Rdnm (TCGContext *s,AArch64Insn insn,TCGType sf, +TCGReg rd,TCGReg rn,TCGReg rm) +{ +tcg_out32 (s, insn | sf <<31 | rm <<16 | rn <<5 | rd); +} +
The name of the function should reflect the fact that we are actually emitting instructions, not only formatting them. Also I despise mixed case in functions.
So theoretically, tcg_out_rdnm.
I'd still rather have a name of the function that expresses the meaning of what we are trying to do (tcg_out_arith seems a good name, you can merge with shiftrot_reg if you want), rather than how we are doing it, if the model fits.
I guess the question would be, are all instructions formatted exactly that way arithmetic and logical shifted register instructions of some sort? If so, I'd go with tcg_out_arith or similar. If not, we can say tcg_out_rdnm.
Also we lose a couple things here.
The previous implementation made it impossible to pass wrong opcodes to the function, since the opcode for the arith was a separate type. It made it obvious to the reader in which cases the function can be used.
We would lose this with this change here (combined with the INSN change).
Also we lose the ability to do right-shifted arithmetic operations, which I feel we should provide for completeness and to reduce the pain for the programmer who will eventually need them.
Quoted 144 lines
tcg/aarch64/tcg-target.c +14
-40static inline void tcg_out_ldst_9(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type,@@ -445,23 +469,6 @@ static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, arg, arg1, arg2); } -static inline void tcg_out_arith (TCGContext *s,AArch64Insn insn, -TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm, -int shift_imm) -{ -/* Using shifted register arithmetic operations */ -/* if extended register operation (64bit) just OR with 0x80 << 24 */ -unsigned int shift, base = insn | (ext ?0x80000000 :0 ); -if (shift_imm ==0 ) { - shift =0 ; - }else if (shift_imm >0 ) { - shift = shift_imm <<10 |1 <<22 ; - }else /* (shift_imm < 0) */ { - shift = (-shift_imm) <<10 ; - } -tcg_out32 (s, base | rm <<16 | shift | rn <<5 | rd); -} -static inline void tcg_out_mul (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm) {@@ -470,15 +477,6 @@ static inline void tcg_out_mul(TCGContext *s, TCGType ext,tcg_out32 (s, base | rm <<16 | rn <<5 | rd); } -static inline void tcg_out_shiftrot_reg (TCGContext *s, -AArch64Insn insn,TCGType ext, -TCGReg rd,TCGReg rn,TCGReg rm) -{ -/* using 2-source data processing instructions 0x1ac02000 */ -unsigned int base = insn | (ext ?0x80000000 :0 ); -tcg_out32 (s, base | rm <<16 | rn <<5 | rd); -} -static inline void tcg_out_ubfm (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int a,unsigned int b) {@@ -546,7 +544,7 @@ static inline void tcg_out_cmp(TCGContext *s, TCGType ext, TCGReg rn, TCGReg rm) {/* Using CMP alias SUBS wzr, Wn, Wm */ -tcg_out_arith (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm,0 ); +tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm); } static inline void tcg_out_cset(TCGContext *s, TCGType ext,@@ -906,8 +904,8 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg,tcg_out_addi (s,1 ,TCG_REG_X2 , base, tlb_offset &0xfff000 );/* Merge the tlb index contribution into X2. X2 = X2 + (X0 << CPU_TLB_ENTRY_BITS) */ -tcg_out_arith (s,INSN_ADD ,1 ,TCG_REG_X2 ,TCG_REG_X2 , -TCG_REG_X0 , -CPU_TLB_ENTRY_BITS ); +tcg_fmt_Rdnm_lsl (s,INSN_ADD ,1 ,TCG_REG_X2 ,TCG_REG_X2 , +TCG_REG_X0 ,CPU_TLB_ENTRY_BITS );/* Merge "low bits" from tlb offset, load the tlb comparator into X0. X0 = load [X2 + (tlb_offset & 0x000fff)] */ tcg_out_ldst(s, TARGET_LONG_BITS == 64 ? LDST_64 : LDST_32,@@ -1183,27 +1181,27 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_add_i64:case INDEX_op_add_i32: -tcg_out_arith (s,INSN_ADD , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_ADD , ext, a0, a1, a2);break ;case INDEX_op_sub_i64:case INDEX_op_sub_i32: -tcg_out_arith (s,INSN_SUB , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_SUB , ext, a0, a1, a2);break ;case INDEX_op_and_i64:case INDEX_op_and_i32: -tcg_out_arith (s,INSN_AND , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_AND , ext, a0, a1, a2);break ;case INDEX_op_or_i64:case INDEX_op_or_i32: -tcg_out_arith (s,INSN_ORR , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_ORR , ext, a0, a1, a2);break ;case INDEX_op_xor_i64:case INDEX_op_xor_i32: -tcg_out_arith (s,INSN_EOR , ext, a0, a1, a2,0 ); +tcg_fmt_Rdnm (s,INSN_EOR , ext, a0, a1, a2);break ;case INDEX_op_mul_i64:@@ -1216,7 +1214,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_shl (s, ext, a0, a1, a2); }else { -tcg_out_shiftrot_reg (s,INSN_LSLV , ext, a0, a1, a2); +tcg_fmt_Rdnm (s,INSN_LSLV , ext, a0, a1, a2); }break ;@@ -1225,7 +1223,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_shr (s, ext, a0, a1, a2); }else { -tcg_out_shiftrot_reg (s,INSN_LSRV , ext, a0, a1, a2); +tcg_fmt_Rdnm (s,INSN_LSRV , ext, a0, a1, a2); }break ;@@ -1234,7 +1232,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_sar (s, ext, a0, a1, a2); }else { -tcg_out_shiftrot_reg (s,INSN_ASRV , ext, a0, a1, a2); +tcg_fmt_Rdnm (s,INSN_ASRV , ext, a0, a1, a2); }break ;@@ -1243,7 +1241,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_rotr (s, ext, a0, a1, a2); }else { -tcg_out_shiftrot_reg (s,INSN_RORV , ext, a0, a1, a2); +tcg_fmt_Rdnm (s,INSN_RORV , ext, a0, a1, a2); }break ;@@ -1252,8 +1250,8 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,if (c2) {tcg_out_rotl (s, ext, a0, a1, a2); }else { -tcg_out_arith (s,INSN_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2,0 ); -tcg_out_shiftrot_reg (s,INSN_RORV , ext, a0, a1,TCG_REG_TMP ); +tcg_fmt_Rdnm (s,INSN_SUB ,0 ,TCG_REG_TMP ,TCG_REG_XZR , a2); +tcg_fmt_Rdnm (s,INSN_RORV , ext, a0, a1,TCG_REG_TMP ); }break ;
Re: [Qemu-devel] [PATCH v4 08/33] tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
Quoted 40 lines · Claudio Fontana
Quoted 39 lines · Richard Henderson
Now that we've converted opcode fields to pre-shifted insns, we can merge the implementation of arithmetic and shift insns.
Simplify the left/right shift parameter to just the left shift needed by
tcg_out_tlb_read.Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 78 +++++++++++++++++++++++------------------------- 1 file changed, 38 insertions(+), 40 deletions(-)tcg/aarch64/tcg-target.c +20
-0@@ -302,6 +302,30 @@ static inline uint32_t tcg_in32(TCGContext *s)return v; } +/* +* Encode various formats. +*/ + +/* This function can be used for both Arithmetic and Logical (shifted register) +type insns. Since we don't actually use the other available shifts, we only +support LSL here. */ +static inline void tcg_fmt_Rdnm_lsl (TCGContext *s,AArch64Insn insn, +TCGType sf,TCGReg rd,TCGReg rn, +TCGReg rm,int imm6) +{ +/* Note that LSL is bits {23,22} = 0. */ +tcg_out32 (s, insn | sf <<31 | imm6 <<10 | rm <<16 | rn <<5 | rd); +} + +/* This function can be used for most insns with 2 input registers and one +output register. This includes Arithmetic (shifted register, sans shift), +Logical, Shift, Multiply, Divide, and Bit operation. */ +static inline void tcg_fmt_Rdnm (TCGContext *s,AArch64Insn insn,TCGType sf, +TCGReg rd,TCGReg rn,TCGReg rm)
Quoted 8 lines
tcg/aarch64/tcg-target.c static inline uint32_t tcg_in32(TCGContext *s) · line 328
Quoted 4 lines
tcg/aarch64/tcg-target.c +4
-0+{ +tcg_out32 (s, insn | sf <<31 | rm <<16 | rn <<5 | rd); +} +The name of the function should reflect the fact that we are actually emitting instructions, not only formatting them. Also I despise mixed case in functions.
So theoretically,tcg_out_rdnm.
Ok.
Quoted 2 lines
I guess the question would be, are all instructions formatted exactly that way arithmetic and logical shifted register instructions of some sort?
The functino comment lists the insn groups from the manual to which the function may be applied: Arithemetic, Logical, Shift, Multiply, Divide, Bit operation.
Quoted 1 line
If so, I'd go with
tcg_out_arithor similar. If not, we can saytcg_out_rdnm.
rdmn it is then.
Quoted 2 lines
The previous implementation made it impossible to pass wrong opcodes to the function, since the opcode for the arith was a separate type.
No, this isn't C++. Enumeration checks like that don't happen for C.
Quoted 2 lines
It made it obvious to the reader in which cases the function can be used. We would lose this with this change here (combined with the INSN change).
Perhaps, perhaps not.
It would have been handy if ARM had officially assigned identifiers to the formats, like Power, S390, and ia64 do. Then one can build in the format ids into both the function and enumeration names and use the preprocessor for typechecking (c.f. the tcg_out_insn macro in tcg/s390/tcg-target.c).
But without those format ids being official, inventing a set of format names may be more confusing than not. I'm not sure.
Quoted 3 lines
Also we lose the ability to do right-shifted arithmetic operations, which I feel we should provide for completeness and to reduce the pain for the programmer who will eventually need them.
Nor do we provide ASR or ROR shifts; should we provide those too? Please think about what situations in which those would be useful. Also think about the one operation at a time nature of TCG.
My guess is that, beyond the one explicit use in the tlb, we could only make use of shifted operations if TCG grew some sort of peephole optimizer so that we can look across single operations. And I don't ever see that happening.
Therefore I think adding LSR, ASR and ROR shifts is both a waste of time and bloats the backend.
r~
Re: [Qemu-devel] [PATCH v4 08/33] tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
Quoted 10 lines · Richard Henderson
Nor do we provide ASR or ROR shifts; should we provide those too? Please think about what situations in which those would be useful. Also think about the one operation at a time nature of TCG.
My guess is that, beyond the one explicit use in the tlb, we could only make use of shifted operations if TCG grew some sort of peephole optimizer so that we can look across single operations. And I don't ever see that happening.
Therefore I think adding LSR, ASR and ROR shifts is both a waste of time and bloats the backend.
Unless: one decides NOT to use the alias definitions in the manual:
LSR d, n, shift => UBFM d, n, shift, 63
LSL d, n, shift => UBFM d, n, 64 - shift, 63 - shift ASR d, n, shift => SBFN d, n, shift, 63
ROR d, n, shift => EXTR d, n, n, shift
and instead use
LSR d, n, shift => ORR d, xzr, n, lsr #shift
LSL d, n, shift => ORR d, xzr, n, lsl #shift
ASR d, n, shift => ORR d, xzr, n, asr #shift
ROR d, n, shift => ORR d, xzr, n, ror #shift
It's not implausible that using the same (base) insn for all of these could result in more code sharing and therefore smaller code size within the tcg backend.
r~
PS: All rather academic, but considering that AArch64 ORR (shifted register) is the preferred implementation for MOV, and AArch32 MOV (shifted register) is the preferred implementation for shift immediates, I'm actually surprised that:
there is no AArch64 MOV (shifted register) alias for AArch32 compatibility, and that ORR isn't the preferred implementation of shift immediates for AArch64.
Re: [Qemu-devel] [PATCH v4 08/33] tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
In reply to Richard HendersonQuoted 55 lines · Richard Henderson
Quoted 40 lines · Claudio Fontana
Quoted 39 lines · Richard Henderson
Now that we've converted opcode fields to pre-shifted insns, we can merge the implementation of arithmetic and shift insns.
Simplify the left/right shift parameter to just the left shift needed by
tcg_out_tlb_read.Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 78 +++++++++++++++++++++++------------------------- 1 file changed, 38 insertions(+), 40 deletions(-)tcg/aarch64/tcg-target.c +20
-0@@ -302,6 +302,30 @@ static inline uint32_t tcg_in32(TCGContext *s)return v; } +/* +* Encode various formats. +*/ + +/* This function can be used for both Arithmetic and Logical (shifted register) +type insns. Since we don't actually use the other available shifts, we only +support LSL here. */ +static inline void tcg_fmt_Rdnm_lsl (TCGContext *s,AArch64Insn insn, +TCGType sf,TCGReg rd,TCGReg rn, +TCGReg rm,int imm6) +{ +/* Note that LSL is bits {23,22} = 0. */ +tcg_out32 (s, insn | sf <<31 | imm6 <<10 | rm <<16 | rn <<5 | rd); +} + +/* This function can be used for most insns with 2 input registers and one +output register. This includes Arithmetic (shifted register, sans shift), +Logical, Shift, Multiply, Divide, and Bit operation. */ +static inline void tcg_fmt_Rdnm (TCGContext *s,AArch64Insn insn,TCGType sf, +TCGReg rd,TCGReg rn,TCGReg rm)Quoted 8 lines
tcg/aarch64/tcg-target.c static inline uint32_t tcg_in32(TCGContext *s) · line 328
Quoted 4 lines
tcg/aarch64/tcg-target.c +4
-0+{ +tcg_out32 (s, insn | sf <<31 | rm <<16 | rn <<5 | rd); +} +The name of the function should reflect the fact that we are actually emitting instructions, not only formatting them. Also I despise mixed case in functions.
So theoretically,tcg_out_rdnm.Ok.
Quoted 2 lines
I guess the question would be, are all instructions formatted exactly that way arithmetic and logical shifted register instructions of some sort?
Quoted 7 lines
The functino comment lists the insn groups from the manual to which the function may be applied: Arithemetic, Logical, Shift, Multiply, Divide, Bit operation.
Quoted 1 line
If so, I'd go with
tcg_out_arithor similar. If not, we can saytcg_out_rdnm.rdmn it is then.
Ok.
Quoted 5 lines
Quoted 3 lines
The previous implementation made it impossible to pass wrong opcodes to the function, since the opcode for the arith was a separate type.
No, this isn't C++. Enumeration checks like that don't happen for C.
I know, I did not express it well. What I meant is that it is impossible to misunderstand what is supposed to be passed to the function. Not that it is impossible to willingly do so.
Quoted 10 lines
Quoted 3 lines
It made it obvious to the reader in which cases the function can be used. We would lose this with this change here (combined with the INSN change).
Perhaps, perhaps not.
It would have been handy if ARM had officially assigned identifiers to the formats, like Power, S390, and ia64 do. Then one can build in the format ids into both the function and enumeration names and use the preprocessor for typechecking (c.f. the
tcg_out_insnmacro intcg/s390/tcg-target.c).
No need to do force explicit typechecking like that.
That kind of use of the preprocessor really hurts.
The only thing that needs to be addressed is to ensure that the programmer calling the function can quickly know for sure which instructions are ok to pass and which not.
Maybe we can categorize the instructions by order of appearance in the enum, coupled with an appropriate prefix. For example INSN_RDNM_ADD, INSN_RDNM_SUB, ...
Quoted 9 lines
But without those format ids being official, inventing a set of format names may be more confusing than not. I'm not sure.
Quoted 3 lines
Also we lose the ability to do right-shifted arithmetic operations, which I feel we should provide for completeness and to reduce the pain for the programmer who will eventually need them.
Nor do we provide ASR or ROR shifts; should we provide those too?
No, not yet.
Quoted 10 lines
Please think
about what situations in which those would be useful. Also think about the one operation at a time nature of TCG.My guess is that, beyond the one explicit use in the tlb, we could only make use of shifted operations if TCG grew some sort of peephole optimizer so that we can look across single operations. And I don't ever see that happening.
Therefore I think adding LSR, ASR and ROR shifts is both a waste of time and bloats the backend.
I agree, lets just keep only left shift and right shift. Distinguishing the two costs one comparison per call, I think we can survive it.
C.
Re: [Qemu-devel] [PATCH v4 08/33] tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
Quoted 7 lines · Claudio Fontana
Quoted 4 lines
It would have been handy if ARM had officially assigned identifiers to the formats, like Power, S390, and ia64 do. Then one can build in the format ids into both the function and enumeration names and use the preprocessor for typechecking (c.f. the
tcg_out_insnmacro intcg/s390/tcg-target.c).No need to do force explicit typechecking like that. That kind of use of the preprocessor really hurts.
Why do you believe this? Have you browsed through the s390 backend? I think it's a remarkably clean solution -- one we ought to have used in the ia64 backend, which has even more format codes.
Quoted 5 lines
Quoted 2 lines
Therefore I think adding LSR, ASR and ROR shifts is both a waste of time and bloats the backend.
I agree, lets just keep only left shift and right shift. Distinguishing the two costs one comparison per call, I think we can survive it.
I don't understand you at all.
You agree that removing the unused shift from cmp is sensible. You agree that not adding unused asr/ror shifts is sensible. But you insist that the unused lsr shift should be retained?
You complain about wasting Y space in my positioning of comments. But you insist on wasting X space, and allowing the possibility of mismatch, by requiring format names to be duplicated?
r~
Re: [Qemu-devel] [PATCH v4 08/33] tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
Quoted 6 lines · Richard Henderson
Quoted 5 lines · Claudio Fontana
Quoted 4 lines
It would have been handy if ARM had officially assigned identifiers to the formats, like Power, S390, and ia64 do. Then one can build in the format ids into both the function and enumeration names and use the preprocessor for typechecking (c.f. the
tcg_out_insnmacro intcg/s390/tcg-target.c).
Quoted 6 lines
Quoted 2 lines
No need to do force explicit typechecking like that. That kind of use of the preprocessor really hurts.
Why do you believe this? Have you browsed through the s390 backend? I think it's a remarkably clean solution -- one we ought to have used in the ia64 backend, which has even more format codes.
Why do I believe this? Because my experience tells me to let this kind of stuff go,
in order to allow developers that are not familiar with the code base yet to trace their way through the calls. It keeps the code base discoverable, by hand and by tools.
Quoted 9 lines
Quoted 5 lines
Quoted 2 lines
Therefore I think adding LSR, ASR and ROR shifts is both a waste of time and bloats the backend.
I agree, lets just keep only left shift and right shift. Distinguishing the two costs one comparison per call, I think we can survive it.
I don't understand you at all.
You agree that removing the unused shift from cmp is sensible.
Yes. The additional parameter to cmp (0) can be confusing, therefore it seems more sensible to remove it until there is actual demonstrable use for the shifted version of cmp.
Quoted 1 line
You agree that not adding unused asr/ror shifts is sensible.
Yep.
Quoted 1 line
But you insist that the unused lsr shift should be retained?
Yes. It's a balance thing, it's not black and white.
In this case the drawback of keeping the right shift is negligible.
No additional parameter is needed, the existing code just looks at the sign of the immediate to decide the direction to shift. It's one comparison only, and I find that the tradeoff is acceptable.
If you _really_ want to strip the right shift functionality away for some reason, could you state it? Is it for that one comparision? Or for consistency with something else?
Quoted 3 lines
You complain about wasting Y space in my positioning of comments. But you insist on wasting X space,
They are different resources. Y space gives us context and is a more scarse and precious resource than X space. Of course it's a tradeoff.
We don't want the whole code on one line, we have 80 char limits and we do one thing per line.
Quoted 2 lines
and allowing the possibility of
mismatch, by requiring format names to be duplicated?
I don't understand this one.
C.
Re: [Qemu-devel] [PATCH v4 08/33] tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
Quoted 1 line · Claudio Fontana
If you _really_ want to strip the right shift functionality away for some reason, could you state it?
(1) It will never never ever be used.
(2) I do not find + for LSR, - for LSL intuitive at all. If we're to have them at all, better to pass that separately from the shift count, an enumeration that exactly maps to the "shift" field in the insn.
(3) It will never never ever be used.
r~
Re: [Qemu-devel] [PATCH v4 08/33] tcg-aarch64: Introduce tcg_fmt_Rdnm and tcg_fmt_Rdnm_lsl
Quoted 11 lines · Richard Henderson
Quoted 1 line · Claudio Fontana
If you _really_ want to strip the right shift functionality away for some reason, could you state it?
(1) It will never never ever be used.
(2) I do not find + for LSR, - for LSL intuitive at all. If we're to have them at all, better to pass that separately from the shift count, an enumeration that exactly maps to the "shift" field in the insn.
(3) It will never never ever be used.
You seem to feel strongly about it, and I don't, so lets indeed drop it then.
C.
PATCH v4 9/33 tcg-aarch64: Introduce tcg_fmt_Rdn_aimm
In reply to Richard HendersonThis merges the implementation of tcg_out_addi and tcg_out_subi.
--- tcg/aarch64/tcg-target.c | 79 +++++++++++++++++++----------------------------- 1 file changed, 31 insertions(+), 48 deletions(-)
tcg/aarch64/tcg-target.c +31 -48
@@ -211,6 +211,12 @@ typedef enum {INSN_EOR =0x4a000000 ,INSN_ANDS =0x6a000000 , +/* Add/subtract immediate instructions */ +INSN_ADDI =0x11000000 , +INSN_ADDSI =0x31000000 , +INSN_SUBI =0x51000000 , +INSN_SUBSI =0x71000000 , +/* Add/subtract shifted register instructions */ INSN_ADD =0x0b000000 , INSN_ADDS = 0x2b000000,
@@ -326,6 +332,22 @@ static inline void tcg_fmt_Rdnm(TCGContext *s, AArch64Insn insn, TCGType sf,tcg_out32 (s, insn | sf <<31 | rm <<16 | rn <<5 | rd); } +/* This function is used for the Arithmetic (immediate) instruction group. +The value of AIMM must be appropriate for encoding in the shift+imm12 +fields. */ +static inline void tcg_fmt_Rdn_aimm (TCGContext *s,AArch64Insn insn, +TCGType sf,TCGReg rd,TCGReg rn, +unsigned int aimm) +{ +if (aimm >0xfff ) { +assert ((aimm &0xfff ) ==0 ); + aimm >>=12 ; +assert (aimm <=0xfff ); + aimm |=1 <<12 ;/* apply LSL 12 */ + } +tcg_out32 (s, insn | sf <<31 | aimm <<10 | rn <<5 | rd); +} + static inline void tcg_out_ldst_9(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type,
@@ -742,46 +764,6 @@ static inline void tcg_out_uxt(TCGContext *s, int s_bits,tcg_out_ubfm (s,0 , rd, rn,0 , bits); } -static inline void tcg_out_addi (TCGContext *s,TCGType ext, -TCGReg rd,TCGReg rn,unsigned int aimm) -{ -/* add immediate aimm unsigned 12bit value (with LSL 0 or 12) */ -/* using ADD 0x11000000 | (ext) | (aimm << 10) | (rn << 5) | rd */ -unsigned int base = ext ?0x91000000 :0x11000000 ; - -if (aimm <=0xfff ) { - aimm <<=10 ; - }else { -/* we can only shift left by 12, on assert we cannot represent */ -assert (!(aimm &0xfff )); -assert (aimm <=0xfff000 ); - base |=1 <<22 ;/* apply LSL 12 */ - aimm >>=2 ; - } - -tcg_out32 (s, base | aimm | (rn <<5 ) | rd); -} - -static inline void tcg_out_subi (TCGContext *s,TCGType ext, -TCGReg rd,TCGReg rn,unsigned int aimm) -{ -/* sub immediate aimm unsigned 12bit value (with LSL 0 or 12) */ -/* using SUB 0x51000000 | (ext) | (aimm << 10) | (rn << 5) | rd */ -unsigned int base = ext ?0xd1000000 :0x51000000 ; - -if (aimm <=0xfff ) { - aimm <<=10 ; - }else { -/* we can only shift left by 12, on assert we cannot represent */ -assert (!(aimm &0xfff )); -assert (aimm <=0xfff000 ); - base |=1 <<22 ;/* apply LSL 12 */ - aimm >>=2 ; - } - -tcg_out32 (s, base | aimm | (rn <<5 ) | rd); -} -static inline void tcg_out_nop (TCGContext *s) {tcg_out32 (s,0xd503201f );
@@ -899,9 +881,9 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, (TARGET_LONG_BITS -TARGET_PAGE_BITS ) + s_bits, (TARGET_LONG_BITS - TARGET_PAGE_BITS));/* Add any "high bits" from the tlb offset to the env address into X2, -to take advantage of the LSL12 form of the addi instruction. +to take advantage of the LSL12 form of the ADDI instruction. X2 = env + (tlb_offset & 0xfff000) */ -tcg_out_addi (s,1 ,TCG_REG_X2 , base, tlb_offset &0xfff000 ); +tcg_fmt_Rdn_aimm (s,INSN_ADDI ,1 ,TCG_REG_X2 , base, tlb_offset &0xfff000 );/* Merge the tlb index contribution into X2. X2 = X2 + (X0 << CPU_TLB_ENTRY_BITS) */ tcg_fmt_Rdnm_lsl(s, INSN_ADD, 1, TCG_REG_X2, TCG_REG_X2,
@@ -1510,9 +1492,10 @@ static void tcg_target_qemu_prologue(TCGContext *s)tcg_out_store_pair (s,TCG_REG_FP , r, r +1 , idx); } -/* make stack space for TCG locals */ -tcg_out_subi (s,1 ,TCG_REG_SP ,TCG_REG_SP , - frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN ); +/* Make stack space for TCG locals. */ +tcg_fmt_Rdn_aimm (s,INSN_SUBI ,1 ,TCG_REG_SP ,TCG_REG_SP , + frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN ); +/* inform TCG about how to find TCG locals with register, offset, size */ tcg_set_frame (s,TCG_REG_SP ,TCG_STATIC_CALL_ARGS_SIZE ,CPU_TEMP_BUF_NLONGS *sizeof (long ));
@@ -1529,9 +1512,9 @@ static void tcg_target_qemu_prologue(TCGContext *s) tb_ret_addr = s->code_ptr; -/* remove TCG locals stack space */ -tcg_out_addi (s,1 ,TCG_REG_SP ,TCG_REG_SP , - frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN ); +/* Remove TCG locals stack space. */ +tcg_fmt_Rdn_aimm (s,INSN_ADDI ,1 ,TCG_REG_SP ,TCG_REG_SP , + frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN );/* restore registers x19..x28. FP must be preserved, so it still points to callee_saved area */
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 09/33] tcg-aarch64: Introduce tcg_fmt_Rdn_aimm
Quoted 25 lines · Richard Henderson
This merges the implementation of
tcg_out_addiandtcg_out_subi.Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 79 +++++++++++++++++++----------------------------- 1 file changed, 31 insertions(+), 48 deletions(-)tcg/aarch64/tcg-target.c +6
-0@@ -211,6 +211,12 @@ typedef enum {INSN_EOR =0x4a000000 ,INSN_ANDS =0x6a000000 , +/* Add/subtract immediate instructions */ +INSN_ADDI =0x11000000 , +INSN_ADDSI =0x31000000 , +INSN_SUBI =0x51000000 , +INSN_SUBSI =0x71000000 , +/* Add/subtract shifted register instructions */ INSN_ADD =0x0b000000 , INSN_ADDS = 0x2b000000,
tcg/aarch64/tcg-target.c static inline void tcg_fmt_Rdnm(TCGContext *s, AArch64Insn insn, TCGType sf, · line 340
Quoted 9 lines
tcg/aarch64/tcg-target.c +6
-0@@ -326,6 +332,22 @@ static inline void tcg_fmt_Rdnm(TCGContext *s, AArch64Insn insn, TCGType sf,tcg_out32 (s, insn | sf <<31 | rm <<16 | rn <<5 | rd); } +/* This function is used for the Arithmetic (immediate) instruction group. +The value of AIMM must be appropriate for encoding in the shift+imm12 +fields. */ +static inline void tcg_fmt_Rdn_aimm (TCGContext *s,AArch64Insn insn, +TCGType sf,TCGReg rd,TCGReg rn, +unsigned int aimm)
Since we are emitting, tcg_out_* seems better to me. 'tcg_out_arith_imm' ?
Also here we have the same issue where it might be not immediately obvious which AArch64Insn values are ok to pass.
Quoted 100 lines
tcg/aarch64/tcg-target.c +19
-48+{ +if (aimm >0xfff ) { +assert ((aimm &0xfff ) ==0 ); + aimm >>=12 ; +assert (aimm <=0xfff ); + aimm |=1 <<12 ;/* apply LSL 12 */ + } +tcg_out32 (s, insn | sf <<31 | aimm <<10 | rn <<5 | rd); +} + static inline void tcg_out_ldst_9(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type,@@ -742,46 +764,6 @@ static inline void tcg_out_uxt(TCGContext *s, int s_bits,tcg_out_ubfm (s,0 , rd, rn,0 , bits); } -static inline void tcg_out_addi (TCGContext *s,TCGType ext, -TCGReg rd,TCGReg rn,unsigned int aimm) -{ -/* add immediate aimm unsigned 12bit value (with LSL 0 or 12) */ -/* using ADD 0x11000000 | (ext) | (aimm << 10) | (rn << 5) | rd */ -unsigned int base = ext ?0x91000000 :0x11000000 ; - -if (aimm <=0xfff ) { - aimm <<=10 ; - }else { -/* we can only shift left by 12, on assert we cannot represent */ -assert (!(aimm &0xfff )); -assert (aimm <=0xfff000 ); - base |=1 <<22 ;/* apply LSL 12 */ - aimm >>=2 ; - } - -tcg_out32 (s, base | aimm | (rn <<5 ) | rd); -} - -static inline void tcg_out_subi (TCGContext *s,TCGType ext, -TCGReg rd,TCGReg rn,unsigned int aimm) -{ -/* sub immediate aimm unsigned 12bit value (with LSL 0 or 12) */ -/* using SUB 0x51000000 | (ext) | (aimm << 10) | (rn << 5) | rd */ -unsigned int base = ext ?0xd1000000 :0x51000000 ; - -if (aimm <=0xfff ) { - aimm <<=10 ; - }else { -/* we can only shift left by 12, on assert we cannot represent */ -assert (!(aimm &0xfff )); -assert (aimm <=0xfff000 ); - base |=1 <<22 ;/* apply LSL 12 */ - aimm >>=2 ; - } - -tcg_out32 (s, base | aimm | (rn <<5 ) | rd); -} -static inline void tcg_out_nop (TCGContext *s) {tcg_out32 (s,0xd503201f );@@ -899,9 +881,9 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, (TARGET_LONG_BITS -TARGET_PAGE_BITS ) + s_bits, (TARGET_LONG_BITS - TARGET_PAGE_BITS));/* Add any "high bits" from the tlb offset to the env address into X2, -to take advantage of the LSL12 form of the addi instruction. +to take advantage of the LSL12 form of the ADDI instruction. X2 = env + (tlb_offset & 0xfff000) */ -tcg_out_addi (s,1 ,TCG_REG_X2 , base, tlb_offset &0xfff000 ); +tcg_fmt_Rdn_aimm (s,INSN_ADDI ,1 ,TCG_REG_X2 , base, tlb_offset &0xfff000 );/* Merge the tlb index contribution into X2. X2 = X2 + (X0 << CPU_TLB_ENTRY_BITS) */ tcg_fmt_Rdnm_lsl(s, INSN_ADD, 1, TCG_REG_X2, TCG_REG_X2,@@ -1510,9 +1492,10 @@ static void tcg_target_qemu_prologue(TCGContext *s)tcg_out_store_pair (s,TCG_REG_FP , r, r +1 , idx); } -/* make stack space for TCG locals */ -tcg_out_subi (s,1 ,TCG_REG_SP ,TCG_REG_SP , - frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN ); +/* Make stack space for TCG locals. */ +tcg_fmt_Rdn_aimm (s,INSN_SUBI ,1 ,TCG_REG_SP ,TCG_REG_SP , + frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN ); +/* inform TCG about how to find TCG locals with register, offset, size */ tcg_set_frame (s,TCG_REG_SP ,TCG_STATIC_CALL_ARGS_SIZE ,CPU_TEMP_BUF_NLONGS *sizeof (long ));@@ -1529,9 +1512,9 @@ static void tcg_target_qemu_prologue(TCGContext *s) tb_ret_addr = s->code_ptr; -/* remove TCG locals stack space */ -tcg_out_addi (s,1 ,TCG_REG_SP ,TCG_REG_SP , - frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN ); +/* Remove TCG locals stack space. */ +tcg_fmt_Rdn_aimm (s,INSN_ADDI ,1 ,TCG_REG_SP ,TCG_REG_SP , + frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN );/* restore registers x19..x28. FP must be preserved, so it still points to callee_saved area */
PATCH v4 10/33 tcg-aarch64: Implement mov with tcg_fmt_* functions
In reply to Richard HendersonAvoid the magic numbers in the current implementation.
--- tcg/aarch64/tcg-target.c | 24 ++++++++++-------------- 1 file changed, 10 insertions(+), 14 deletions(-)
tcg/aarch64/tcg-target.c +10 -14
@@ -379,13 +379,18 @@ static inline void tcg_out_ldst_12(TCGContext *s, | op_type << 20 | scaled_uimm << 10 | rn << 5 | rd); } +/* Register to register move using ORR (shifted register with no shift). */ static inline void tcg_out_movr (TCGContext *s,TCGType ext, -TCGReg rd,TCGReg src) +TCGReg rd,TCGReg rm) { -/* register to register move using MOV (shifted register with no shift) */ -/* using MOV 0x2a0003e0 | (shift).. */ -unsigned int base = ext ?0xaa0003e0 :0x2a0003e0 ; -tcg_out32 (s, base | src <<16 | rd); +tcg_fmt_Rdnm (s,INSN_ORR , ext, rd,TCG_REG_XZR , rm); +} + +/* Register to register move using ADDI (move to/from SP). */ +static inline void tcg_out_movr_sp (TCGContext *s,TCGType ext, +TCGReg rd,TCGReg rn) +{ +tcg_fmt_Rdn_aimm (s,INSN_ADDI , ext, rd, rn,0 ); } static inline void tcg_out_movi_aux(TCGContext *s,
@@ -460,15 +465,6 @@ static inline void tcg_out_ldst(TCGContext *s, enum aarch64_ldst_op_data data,tcg_out_ldst_r (s, data, type, rd, rn,TCG_REG_TMP ); } -/* mov alias implemented with add immediate, useful to move to/from SP */ -static inline void tcg_out_movr_sp (TCGContext *s,TCGType ext, -TCGReg rd,TCGReg rn) -{ -/* using ADD 0x11000000 | (ext) | rn << 5 | rd */ -unsigned int base = ext ?0x91000000 :0x11000000 ; -tcg_out32 (s, base | rn <<5 | rd); -} -static inline void tcg_out_mov (TCGContext *s,TCGType type,TCGReg ret,TCGReg arg) {
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 10/33] tcg-aarch64: Implement mov with tcg_fmt_* functions
Quoted 36 lines · Richard Henderson
Avoid the magic numbers in the current implementation.
Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 24 ++++++++++-------------- 1 file changed, 10 insertions(+), 14 deletions(-)tcg/aarch64/tcg-target.c +10
-5@@ -379,13 +379,18 @@ static inline void tcg_out_ldst_12(TCGContext *s, | op_type << 20 | scaled_uimm << 10 | rn << 5 | rd); } +/* Register to register move using ORR (shifted register with no shift). */ static inline void tcg_out_movr (TCGContext *s,TCGType ext, -TCGReg rd,TCGReg src) +TCGReg rd,TCGReg rm) { -/* register to register move using MOV (shifted register with no shift) */ -/* using MOV 0x2a0003e0 | (shift).. */ -unsigned int base = ext ?0xaa0003e0 :0x2a0003e0 ; -tcg_out32 (s, base | src <<16 | rd); +tcg_fmt_Rdnm (s,INSN_ORR , ext, rd,TCG_REG_XZR , rm); +} + +/* Register to register move using ADDI (move to/from SP). */ +static inline void tcg_out_movr_sp (TCGContext *s,TCGType ext, +TCGReg rd,TCGReg rn) +{ +tcg_fmt_Rdn_aimm (s,INSN_ADDI , ext, rd, rn,0 ); } static inline void tcg_out_movi_aux(TCGContext *s,
tcg/aarch64/tcg-target.c static inline void tcg_out_ldst(TCGContext *s, enum aarch64_ldst_op_data data, · line 467
Quoted 3 lines
tcg/aarch64/tcg-target.c +0
-0@@ -460,15 +465,6 @@ static inline void tcg_out_ldst(TCGContext *s, enum aarch64_ldst_op_data data,tcg_out_ldst_r (s, data, type, rd, rn,TCG_REG_TMP ); }
I am ok with reusing the existing functions of course, this is an improvement. See the comments about the tcg_fmt_* functions in previous patches for the related issues.
Quoted 13 lines
tcg/aarch64/tcg-target.c +0
-9-/* mov alias implemented with add immediate, useful to move to/from SP */ -static inline void tcg_out_movr_sp (TCGContext *s,TCGType ext, -TCGReg rd,TCGReg rn) -{ -/* using ADD 0x11000000 | (ext) | rn << 5 | rd */ -unsigned int base = ext ?0x91000000 :0x11000000 ; -tcg_out32 (s, base | rn <<5 | rd); -} -static inline void tcg_out_mov (TCGContext *s,TCGType type,TCGReg ret,TCGReg arg) {
PATCH v4 11/33 tcg-aarch64: Handle constant operands to add, sub, and compare
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 103 ++++++++++++++++++++++++++++++++++++----------- 1 file changed, 80 insertions(+), 23 deletions(-)
tcg/aarch64/tcg-target.c +80 -23
@@ -111,6 +111,9 @@ static inline void patch_reloc(uint8_t *code_ptr, int type, } } +#define TCG_CT_CONST_IS32 0x100 +#define TCG_CT_CONST_AIMM 0x200 +/* parse target specific constraints */ static int target_parse_constraint (TCGArgConstraint *ct,const char **pct_str)
@@ -134,6 +137,12 @@ static int target_parse_constraint(TCGArgConstraint *ct,tcg_regset_reset_reg (ct->u.regs,TCG_REG_X3 );#endif break ; +case 'w' :/* The operand should be considered 32-bit. */ + ct->ct |=TCG_CT_CONST_IS32 ; +break ; +case 'A' :/* Valid for arithmetic immediate (positive or negative). */ + ct->ct |=TCG_CT_CONST_AIMM ; +break ;default :return -1 ; }
@@ -143,14 +152,25 @@ static int target_parse_constraint(TCGArgConstraint *ct,return 0 ; } -static inline int tcg_target_const_match (tcg_target_long val, -const TCGArgConstraint *arg_ct) +static inline bool is_aimm (uint64_t val) +{ +return (val & ~0xfff ) ==0 || (val & ~0xfff000 ) ==0 ; +} + +static int tcg_target_const_match (tcg_target_long val, +const TCGArgConstraint *arg_ct) {int ct = arg_ct->ct;if (ct &TCG_CT_CONST ) {return 1 ; } +if (ct &TCG_CT_CONST_IS32 ) { + val = (int32_t )val; + } +if ((ct &TCG_CT_CONST_AIMM ) && (is_aimm (val) ||is_aimm (-val))) { +return 1 ; + }return 0 ; }
@@ -558,11 +578,21 @@ static inline void tcg_out_rotl(TCGContext *s, TCGType ext,tcg_out_extr (s, ext, rd, rn, rn, bits - (m & max)); } -static inline void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg rn, -TCGReg rm) +static void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg a, +tcg_target_long b,bool const_b) { -/* Using CMP alias SUBS wzr, Wn, Wm */ -tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm); +if (const_b) { +/* Using CMP or CMN aliases. */ +AArch64Insn insn =INSN_SUBSI ; +if (b <0 ) { + insn =INSN_ADDSI ; + b = -b; + } +tcg_fmt_Rdn_aimm (s, insn, ext,TCG_REG_XZR , a, b); + }else { +/* Using CMP alias SUBS wzr, Wn, Wm */ +tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , a, b); + } } static inline void tcg_out_cset(TCGContext *s, TCGType ext,
@@ -760,6 +790,17 @@ static inline void tcg_out_uxt(TCGContext *s, int s_bits,tcg_out_ubfm (s,0 , rd, rn,0 , bits); } +static void tcg_out_addsubi (TCGContext *s,int ext,TCGReg rd, +TCGReg rn,int aimm) +{ +AArch64Insn insn =INSN_ADDI ; +if (aimm <0 ) { + insn =INSN_SUBI ; + aimm = -aimm; + } +tcg_fmt_Rdn_aimm (s, insn, ext, rd, rn, aimm); +} +static inline void tcg_out_nop (TCGContext *s) {tcg_out32 (s,0xd503201f );
@@ -896,7 +937,7 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, (is_read ? offsetof(CPUTLBEntry , addr_read) : offsetof(CPUTLBEntry, addr_write)));/* Perform the address comparison. */ -tcg_out_cmp (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 ,TCG_REG_X3 ); +tcg_out_cmp (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 ,TCG_REG_X3 ,0 ); *label_ptr = s->code_ptr;/* If not equal, we jump to the slow path. */ tcg_out_goto_cond_noaddr (s,TCG_COND_NE );
@@ -1157,14 +1198,26 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, a0, a1, a2);break ; -case INDEX_op_add_i64:case INDEX_op_add_i32: -tcg_fmt_Rdnm (s,INSN_ADD , ext, a0, a1, a2); + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_add_i64: +if (c2) { +tcg_out_addsubi (s, ext, a0, a1, a2); + }else { +tcg_fmt_Rdnm (s,INSN_ADD , ext, a0, a1, a2); + }break ; -case INDEX_op_sub_i64:case INDEX_op_sub_i32: -tcg_fmt_Rdnm (s,INSN_SUB , ext, a0, a1, a2); + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_sub_i64: +if (c2) { +tcg_out_addsubi (s, ext, a0, a1, -a2); + }else { +tcg_fmt_Rdnm (s,INSN_SUB , ext, a0, a1, a2); + }break ;case INDEX_op_and_i64:
@@ -1233,15 +1286,19 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, }break ; -case INDEX_op_brcond_i64:case INDEX_op_brcond_i32: -tcg_out_cmp (s, ext, a0, a1); + a1 = (int32_t )a1; +/* FALLTHRU */ +case INDEX_op_brcond_i64: +tcg_out_cmp (s, ext, a0, a1, const_args[1 ]);tcg_out_goto_label_cond (s, a2, args[3 ]);break ; -case INDEX_op_setcond_i64:case INDEX_op_setcond_i32: -tcg_out_cmp (s, ext, a1, a2); + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_setcond_i64: +tcg_out_cmp (s, ext, a1, a2, c2);tcg_out_cset (s,0 , a0, args[3 ]);break ;
@@ -1362,10 +1419,10 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_st32_i64, { "r", "r" } }, { INDEX_op_st_i64, { "r", "r" } }, - { INDEX_op_add_i32, { "r", "r", "r" } }, - { INDEX_op_add_i64, { "r", "r", "r" } }, - { INDEX_op_sub_i32, { "r", "r", "r" } }, - { INDEX_op_sub_i64, { "r", "r", "r" } }, + { INDEX_op_add_i32, { "r", "r", "rwA" } }, + { INDEX_op_add_i64, { "r", "r", "rA" } }, + { INDEX_op_sub_i32, { "r", "r", "rwA" } }, + { INDEX_op_sub_i64, { "r", "r", "rA" } }, { INDEX_op_mul_i32, { "r", "r", "r" } }, { INDEX_op_mul_i64, { "r", "r", "r" } }, { INDEX_op_and_i32, { "r", "r", "r" } },
@@ -1386,10 +1443,10 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_rotl_i64, { "r", "r", "ri" } }, { INDEX_op_rotr_i64, { "r", "r", "ri" } }, - { INDEX_op_brcond_i32, { "r", "r" } }, - { INDEX_op_setcond_i32, { "r", "r", "r" } }, - { INDEX_op_brcond_i64, { "r", "r" } }, - { INDEX_op_setcond_i64, { "r", "r", "r" } }, + { INDEX_op_brcond_i32, { "r", "rwA" } }, + { INDEX_op_brcond_i64, { "r", "rA" } }, + { INDEX_op_setcond_i32, { "r", "r", "rwA" } }, + { INDEX_op_setcond_i64, { "r", "r", "rA" } }, { INDEX_op_qemu_ld8u, { "r", "l" } }, { INDEX_op_qemu_ld8s, { "r", "l" } },
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 11/33] tcg-aarch64: Handle constant operands to add, sub, and compare
Quoted 74 lines · Richard Henderson
Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 103 ++++++++++++++++++++++++++++++++++++----------- 1 file changed, 80 insertions(+), 23 deletions(-)tcg/aarch64/tcg-target.c +27
-6@@ -111,6 +111,9 @@ static inline void patch_reloc(uint8_t *code_ptr, int type, } } +#define TCG_CT_CONST_IS32 0x100 +#define TCG_CT_CONST_AIMM 0x200 +/* parse target specific constraints */ static int target_parse_constraint (TCGArgConstraint *ct,const char **pct_str)@@ -134,6 +137,12 @@ static int target_parse_constraint(TCGArgConstraint *ct,tcg_regset_reset_reg (ct->u.regs,TCG_REG_X3 );#endif break ; +case 'w' :/* The operand should be considered 32-bit. */ + ct->ct |=TCG_CT_CONST_IS32 ; +break ; +case 'A' :/* Valid for arithmetic immediate (positive or negative). */ + ct->ct |=TCG_CT_CONST_AIMM ; +break ;default :return -1 ; }@@ -143,14 +152,25 @@ static int target_parse_constraint(TCGArgConstraint *ct,return 0 ; } -static inline int tcg_target_const_match (tcg_target_long val, -const TCGArgConstraint *arg_ct) +static inline bool is_aimm (uint64_t val) +{ +return (val & ~0xfff ) ==0 || (val & ~0xfff000 ) ==0 ; +} + +static int tcg_target_const_match (tcg_target_long val, +const TCGArgConstraint *arg_ct) {int ct = arg_ct->ct;if (ct &TCG_CT_CONST ) {return 1 ; } +if (ct &TCG_CT_CONST_IS32 ) { + val = (int32_t )val; + } +if ((ct &TCG_CT_CONST_AIMM ) && (is_aimm (val) ||is_aimm (-val))) { +return 1 ; + }return 0 ; }@@ -558,11 +578,21 @@ static inline void tcg_out_rotl(TCGContext *s, TCGType ext,tcg_out_extr (s, ext, rd, rn, rn, bits - (m & max)); } -static inline void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg rn, -TCGReg rm) +static void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg a, +tcg_target_long b,bool const_b) { -/* Using CMP alias SUBS wzr, Wn, Wm */ -tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm); +if (const_b) { +/* Using CMP or CMN aliases. */ +AArch64Insn insn =INSN_SUBSI ;
tcg/aarch64/tcg-target.c static inline void tcg_out_rotl(TCGContext *s, TCGType ext, · line 596
Quoted 10 lines
tcg/aarch64/tcg-target.c +9
-0+if (b <0 ) { + insn =INSN_ADDSI ; + b = -b; + } +tcg_fmt_Rdn_aimm (s, insn, ext,TCG_REG_XZR , a, b); + }else { +/* Using CMP alias SUBS wzr, Wn, Wm */ +tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , a, b); + } }
What about splitting into tcg_out_cmp_imm and tcg_out_cmp_reg?
or tcg_out_cmp_i and tcg_out_cmp_r.
The function is an 'if else' anyway with no code sharing, and we would avoid sidestepping the TCGReg type check for b in the _r case, as well as the const_b additional param.
Quoted 7 lines
tcg/aarch64/tcg-target.c +1
-0static inline void tcg_out_cset(TCGContext *s, TCGType ext,@@ -760,6 +790,17 @@ static inline void tcg_out_uxt(TCGContext *s, int s_bits,tcg_out_ubfm (s,0 , rd, rn,0 , bits); } +static void tcg_out_addsubi(TCGContext *s, int ext, TCGReg rd,
tcg/aarch64/tcg-target.c static inline void tcg_out_uxt(TCGContext *s, int s_bits, · line 803
Quoted 10 lines
tcg/aarch64/tcg-target.c +10
-0+ TCGReg rn, int aimm) +{ +AArch64Insn insn =INSN_ADDI ; +if (aimm <0 ) { + insn =INSN_SUBI ; + aimm = -aimm; + } +tcg_fmt_Rdn_aimm (s, insn, ext, rd, rn, aimm); +} +
Could this be a tcg_out_arith_imm, in the similar way we would do tcg_out_arith? (tcg_out_arith_reg?)
Quoted 13 lines
tcg/aarch64/tcg-target.c +1
-1static inline void tcg_out_nop (TCGContext *s) {tcg_out32 (s,0xd503201f );@@ -896,7 +937,7 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, (is_read ? offsetof(CPUTLBEntry , addr_read) : offsetof(CPUTLBEntry, addr_write)));/* Perform the address comparison. */ -tcg_out_cmp (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 ,TCG_REG_X3 ); +tcg_out_cmp (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 ,TCG_REG_X3 ,0 ); *label_ptr = s->code_ptr;/* If not equal, we jump to the slow path. */ tcg_out_goto_cond_noaddr (s,TCG_COND_NE );
tcg/aarch64/tcg-target.c static void tcg_out_op(TCGContext *s, TCGOpcode opc, · line 1203
Quoted 8 lines
tcg/aarch64/tcg-target.c +2
-2@@ -1157,14 +1198,26 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, a0, a1, a2);break ; -case INDEX_op_add_i64:case INDEX_op_add_i32: -tcg_fmt_Rdnm (s,INSN_ADD , ext, a0, a1, a2); + a2 = (int32_t )a2; +/* FALLTHRU */
/* fall through */ is less ugly.
Quoted 77 lines
tcg/aarch64/tcg-target.c +30
-14+case INDEX_op_add_i64: +if (c2) { +tcg_out_addsubi (s, ext, a0, a1, a2); + }else { +tcg_fmt_Rdnm (s,INSN_ADD , ext, a0, a1, a2); + }break ; -case INDEX_op_sub_i64:case INDEX_op_sub_i32: -tcg_fmt_Rdnm (s,INSN_SUB , ext, a0, a1, a2); + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_sub_i64: +if (c2) { +tcg_out_addsubi (s, ext, a0, a1, -a2); + }else { +tcg_fmt_Rdnm (s,INSN_SUB , ext, a0, a1, a2); + }break ;case INDEX_op_and_i64:@@ -1233,15 +1286,19 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, }break ; -case INDEX_op_brcond_i64:case INDEX_op_brcond_i32: -tcg_out_cmp (s, ext, a0, a1); + a1 = (int32_t )a1; +/* FALLTHRU */ +case INDEX_op_brcond_i64: +tcg_out_cmp (s, ext, a0, a1, const_args[1 ]);tcg_out_goto_label_cond (s, a2, args[3 ]);break ; -case INDEX_op_setcond_i64:case INDEX_op_setcond_i32: -tcg_out_cmp (s, ext, a1, a2); + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_setcond_i64: +tcg_out_cmp (s, ext, a1, a2, c2);tcg_out_cset (s,0 , a0, args[3 ]);break ;@@ -1362,10 +1419,10 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_st32_i64, { "r", "r" } }, { INDEX_op_st_i64, { "r", "r" } }, - { INDEX_op_add_i32, { "r", "r", "r" } }, - { INDEX_op_add_i64, { "r", "r", "r" } }, - { INDEX_op_sub_i32, { "r", "r", "r" } }, - { INDEX_op_sub_i64, { "r", "r", "r" } }, + { INDEX_op_add_i32, { "r", "r", "rwA" } }, + { INDEX_op_add_i64, { "r", "r", "rA" } }, + { INDEX_op_sub_i32, { "r", "r", "rwA" } }, + { INDEX_op_sub_i64, { "r", "r", "rA" } }, { INDEX_op_mul_i32, { "r", "r", "r" } }, { INDEX_op_mul_i64, { "r", "r", "r" } }, { INDEX_op_and_i32, { "r", "r", "r" } },@@ -1386,10 +1443,10 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_rotl_i64, { "r", "r", "ri" } }, { INDEX_op_rotr_i64, { "r", "r", "ri" } }, - { INDEX_op_brcond_i32, { "r", "r" } }, - { INDEX_op_setcond_i32, { "r", "r", "r" } }, - { INDEX_op_brcond_i64, { "r", "r" } }, - { INDEX_op_setcond_i64, { "r", "r", "r" } }, + { INDEX_op_brcond_i32, { "r", "rwA" } }, + { INDEX_op_brcond_i64, { "r", "rA" } }, + { INDEX_op_setcond_i32, { "r", "r", "rwA" } }, + { INDEX_op_setcond_i64, { "r", "r", "rA" } }, { INDEX_op_qemu_ld8u, { "r", "l" } }, { INDEX_op_qemu_ld8s, { "r", "l" } },
Re: [Qemu-devel] [PATCH v4 11/33] tcg-aarch64: Handle constant operands to add, sub, and compare
Quoted 17 lines · Claudio Fontana
Quoted 17 lines
tcg/aarch64/tcg-target.c +12
-4-static inline void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg rn, -TCGReg rm) +static void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg a, +tcg_target_long b,bool const_b) { -/* Using CMP alias SUBS wzr, Wn, Wm */ -tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm); +if (const_b) { +/* Using CMP or CMN aliases. */ +AArch64Insn insn =INSN_SUBSI ; +if (b <0 ) { + insn =INSN_ADDSI ; + b = -b; + } +tcg_fmt_Rdn_aimm (s, insn, ext,TCG_REG_XZR , a, b); + }else { +/* Using CMP alias SUBS wzr, Wn, Wm */
Quoted 8 lines
tcg/aarch64/tcg-target.c static inline void tcg_out_rotl(TCGContext *s, TCGType ext, · line 596
Quoted 3 lines
tcg/aarch64/tcg-target.c +2
-0+tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , a, b); + } }What about splitting into
tcg_out_cmp_immandtcg_out_cmp_reg? ortcg_out_cmp_iandtcg_out_cmp_r. The function is an 'if else' anyway with no code sharing, and we would avoid sidestepping the TCGReg type check for b in the _r case, as well as theconst_badditional param.
This function is called once from tcg_out_tlb_read and three times from tcg_out_opc. I just thought since the majority of uses would have to perform this if then we might as well have it in the subroutine than force all of the callers to replicate it.
Quoted 6 lines
Quoted 6 lines
tcg/aarch64/tcg-target.c +6
-0+static void tcg_out_addsubi (TCGContext *s,int ext,TCGReg rd, +TCGReg rn,int aimm) +{ +AArch64Insn insn =INSN_ADDI ; +if (aimm <0 ) { + insn =INSN_SUBI ;
Quoted 8 lines
tcg/aarch64/tcg-target.c static inline void tcg_out_uxt(TCGContext *s, int s_bits, · line 803
Quoted 5 lines
tcg/aarch64/tcg-target.c +5
-0+ aimm = -aimm; + } +tcg_fmt_Rdn_aimm (s, insn, ext, rd, rn, aimm); +} +Could this be a
tcg_out_arith_imm, in the similar way we would dotcg_out_arith? (tcg_out_arith_reg?)
Which one gets renamed? You already proposed tcg_fmt_Rdn_aimm be named tcg_out_arith_imm. Now you want tcg_out_addsubi renamed to the same name?
I suppose we could merge the two if we add the S bit as a parameter. Then we don't have to distinguish between ADDI and ADDIS, and we could share code with comparisons above...
r~
Re: [Qemu-devel] [PATCH v4 11/33] tcg-aarch64: Handle constant operands to add, sub, and compare
Quoted 27 lines · Richard Henderson
Quoted 17 lines · Claudio Fontana
Quoted 17 lines
tcg/aarch64/tcg-target.c +12
-4-static inline void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg rn, -TCGReg rm) +static void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg a, +tcg_target_long b,bool const_b) { -/* Using CMP alias SUBS wzr, Wn, Wm */ -tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , rn, rm); +if (const_b) { +/* Using CMP or CMN aliases. */ +AArch64Insn insn =INSN_SUBSI ; +if (b <0 ) { + insn =INSN_ADDSI ; + b = -b; + } +tcg_fmt_Rdn_aimm (s, insn, ext,TCG_REG_XZR , a, b); + }else { +/* Using CMP alias SUBS wzr, Wn, Wm */ Quoted 8 lines
tcg/aarch64/tcg-target.c static inline void tcg_out_rotl(TCGContext *s, TCGType ext, · line 596
Quoted 3 lines
tcg/aarch64/tcg-target.c +2
-0+tcg_fmt_Rdnm (s,INSN_SUBS , ext,TCG_REG_XZR , a, b); + } }What about splitting into
tcg_out_cmp_immandtcg_out_cmp_reg? ortcg_out_cmp_iandtcg_out_cmp_r. The function is an 'if else' anyway with no code sharing, and we would avoid sidestepping the TCGReg type check for b in the _r case, as well as theconst_badditional param.
Quoted 4 lines
This function is called once from
tcg_out_tlb_readand three times fromtcg_out_opc. I just thought since the majority of uses would have to perform this if then we might as well have it in the subroutine than force all of the callers to replicate it.
Ok, that's a good point.
What about we keep the tcg_out_cmp and we have it do
if (const_b) {
tcg_out_cmp_i();
} else {
tcg_out_cmp_r();
}
so that code that wants to call cmp_r or cmp_i directly can do that? I realize that at them moment it would benefit only tcg_out_tlb_read's use.
Quoted 15 lines
Quoted 6 lines
Quoted 6 lines
tcg/aarch64/tcg-target.c +6
-0+static void tcg_out_addsubi (TCGContext *s,int ext,TCGReg rd, +TCGReg rn,int aimm) +{ +AArch64Insn insn =INSN_ADDI ; +if (aimm <0 ) { + insn =INSN_SUBI ;Quoted 8 lines
tcg/aarch64/tcg-target.c static inline void tcg_out_uxt(TCGContext *s, int s_bits, · line 803
Quoted 5 lines
tcg/aarch64/tcg-target.c +5
-0+ aimm = -aimm; + } +tcg_fmt_Rdn_aimm (s, insn, ext, rd, rn, aimm); +} +Could this be a
tcg_out_arith_imm, in the similar way we would dotcg_out_arith? (tcg_out_arith_reg?)
Quoted 2 lines
Which one gets renamed? You already proposed
tcg_fmt_Rdn_aimmbe namedtcg_out_arith_imm. Now you wanttcg_out_addsubirenamed to the same name?
This one confused me, possibly because I don't see the reason for addsubi.
Quoted 4 lines
I suppose we could merge the two if we add the S bit as a parameter. Then we don't have to distinguish between ADDI and ADDIS, and we could share code with comparisons above...
I am ok with keeping the two separate, distinguishing in principle a subtraction from a comparison.
Claudio
PATCH v4 12/33 tcg-aarch64: Handle constant operands to and, or, xor
In reply to Richard HendersonHandle a simplified set of logical immediates for the moment.
The way gcc and binutils do it, with 52k worth of tables, and a binary search depth of log2(5334) = 13, seems slow for the most common cases.
--- tcg/aarch64/tcg-target.c | 155 ++++++++++++++++++++++++++++++++--------------- 1 file changed, 106 insertions(+), 49 deletions(-)
tcg/aarch64/tcg-target.c +106 -49
@@ -113,6 +113,7 @@ static inline void patch_reloc(uint8_t *code_ptr, int type,#define TCG_CT_CONST_IS32 0x100#define TCG_CT_CONST_AIMM 0x200 +#define TCG_CT_CONST_LIMM 0x400/* parse target specific constraints */ static int target_parse_constraint(TCGArgConstraint *ct,
@@ -143,6 +144,9 @@ static int target_parse_constraint(TCGArgConstraint *ct,case 'A' :/* Valid for arithmetic immediate (positive or negative). */ ct->ct |=TCG_CT_CONST_AIMM ;break ; +case 'L' :/* Valid for logical immediate. */ + ct->ct |=TCG_CT_CONST_LIMM ; +break ;default :return -1 ; }
@@ -157,6 +161,26 @@ static inline bool is_aimm(uint64_t val)return (val & ~0xfff ) ==0 || (val & ~0xfff000 ) ==0 ; } +static inline bool is_limm (uint64_t val) +{ +/* Taking a simplified view of the logical immediates for now, ignoring +the replication that can happen across the field. Match bit patterns +of the forms +0....01....1 +0..01..10..0 +and their inverses. */ + +/* Make things easier below, by testing the form with msb clear. */ +if ((int64_t )val <0 ) { + val = ~val; + } +if (val ==0 ) { +return false; + } + val += val & -val; +return (val & (val -1 )) ==0 ; +} +static int tcg_target_const_match (tcg_target_long val,const TCGArgConstraint *arg_ct) {
@@ -171,6 +195,9 @@ static int tcg_target_const_match(tcg_target_long val,if ((ct &TCG_CT_CONST_AIMM ) && (is_aimm (val) ||is_aimm (-val))) {return 1 ; } +if ((ct &TCG_CT_CONST_LIMM ) &&is_limm (val)) { +return 1 ; + }return 0 ; }
@@ -225,6 +252,11 @@ enum aarch64_ldst_op_type { /* type of operation */ }; typedef enum { + /* Logical immediate instructions */ + INSN_ANDI = 0x12000000, + INSN_ORRI = 0x32000000, + INSN_EORI = 0x52000000, + /* Logical shifted register instructions */ INSN_AND = 0x0a000000, INSN_ORR = 0x2a000000,
@@ -368,6 +400,48 @@ static inline void tcg_fmt_Rdn_aimm(TCGContext *s, AArch64Insn insn,tcg_out32 (s, insn | sf <<31 | aimm <<10 | rn <<5 | rd); } +/* This function can be used for both Logical (immediate) and Bitfield +instruction groups, both of which have N, IMMR and IMMS fields, that +feed the DecodeBitMasks pseudo function in the reference manual. */ +static inline void tcg_fmt_Rdn_nrs (TCGContext *s,AArch64Insn insn, +TCGType sf,TCGReg rd,TCGReg rn, +int n,int immr,int imms) +{ +tcg_out32 (s, insn | sf <<31 | n <<22 | immr <<16 | imms <<10 + | rn <<5 | rd); +} + +/* This function is used for the Logical (immediate) instruction group. +The value of LIMM must satisfy IS_LIMM. See the comment above about +only supporting simplified logical immediates. */ +static void tcg_fmt_Rdn_limm (TCGContext *s,AArch64Insn insn,TCGType sf, +TCGReg rd,TCGReg rn,uint64_t limm) +{ +unsigned h, l, r, c; + +assert (is_limm (limm)); + + h =clz64 (limm); + l =ctz64 (limm); +if (l ==0 ) { + r =0 ;/* form 0....01....1 */ + c =ctz64 (~limm) -1 ; +if (h ==0 ) { + r =clz64 (~limm);/* form 1..10..01..1 */ + c += r; + } + }else { + r =64 - l;/* form 1....10....0 or 0..01..10..0 */ + c = r - h -1 ; + } +if (!sf) { + r &=31 ; + c &=31 ; + } + +tcg_fmt_Rdn_nrs (s, insn, sf, rd, rn, sf, r, c); +} + static inline void tcg_out_ldst_9(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type,
@@ -676,40 +750,6 @@ static inline void tcg_out_call(TCGContext *s, tcg_target_long target) } } -/* encode a logical immediate, mapping user parameter -M=set bits pattern length to S=M-1 */ -static inline unsigned int -aarch64_limm (unsigned int m,unsigned int r) -{ -assert (m >0 ); -return r <<16 | (m -1 ) <<10 ; -} - -/* test a register against an immediate bit pattern made of -M set bits rotated right by R. -Examples: -to test a 32/64 reg against 0x00000007, pass M = 3, R = 0. -to test a 32/64 reg against 0x000000ff, pass M = 8, R = 0. -to test a 32bit reg against 0xff000000, pass M = 8, R = 8. -to test a 32bit reg against 0xff0000ff, pass M = 16, R = 8. -*/ -static inline void tcg_out_tst (TCGContext *s,TCGType ext,TCGReg rn, -unsigned int m,unsigned int r) -{ -/* using TST alias of ANDS XZR, Xn,#bimm64 0x7200001f */ -unsigned int base = ext ?0xf240001f :0x7200001f ; -tcg_out32 (s, base |aarch64_limm (m, r) | rn <<5 ); -} - -/* and a register with a bit pattern, similarly to TST, no flags change */ -static inline void tcg_out_andi (TCGContext *s,TCGType ext,TCGReg rd, -TCGReg rn,unsigned int m,unsigned int r) -{ -/* using AND 0x12000000 */ -unsigned int base = ext ?0x92400000 :0x12000000 ; -tcg_out32 (s, base |aarch64_limm (m, r) | rn <<5 | rd); -} -static inline void tcg_out_ret (TCGContext *s) {/* emit RET { LR } */
@@ -914,9 +954,8 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg,/* Store the page mask part of the address and the low s_bits into X3. Later this allows checking for equality and alignment at the same time. X3 = addr_reg & (PAGE_MASK | ((1 << s_bits) - 1)) */ -tcg_out_andi (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X3 , addr_reg, - (TARGET_LONG_BITS -TARGET_PAGE_BITS ) + s_bits, - (TARGET_LONG_BITS -TARGET_PAGE_BITS )); +tcg_fmt_Rdn_limm (s,INSN_ANDI ,TARGET_LONG_BITS ==64 ,TCG_REG_X3 , + addr_reg,TARGET_PAGE_MASK | ((1 << s_bits) -1 ));/* Add any "high bits" from the tlb offset to the env address into X2, to take advantage of the LSL12 form of the ADDI instruction. X2 = env + (tlb_offset & 0xfff000) */
@@ -1220,19 +1259,37 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, }break ; -case INDEX_op_and_i64:case INDEX_op_and_i32: -tcg_fmt_Rdnm (s,INSN_AND , ext, a0, a1, a2); + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_and_i64: +if (c2) { +tcg_fmt_Rdn_limm (s,INSN_ANDI , ext, a0, a1, a2); + }else { +tcg_fmt_Rdnm (s,INSN_AND , ext, a0, a1, a2); + }break ; -case INDEX_op_or_i64:case INDEX_op_or_i32: -tcg_fmt_Rdnm (s,INSN_ORR , ext, a0, a1, a2); + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_or_i64: +if (c2) { +tcg_fmt_Rdn_limm (s,INSN_ORRI , ext, a0, a1, a2); + }else { +tcg_fmt_Rdnm (s,INSN_ORR , ext, a0, a1, a2); + }break ; -case INDEX_op_xor_i64:case INDEX_op_xor_i32: -tcg_fmt_Rdnm (s,INSN_EOR , ext, a0, a1, a2); + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_xor_i64: +if (c2) { +tcg_fmt_Rdn_limm (s,INSN_EORI , ext, a0, a1, a2); + }else { +tcg_fmt_Rdnm (s,INSN_EOR , ext, a0, a1, a2); + }break ;case INDEX_op_mul_i64:
@@ -1425,12 +1482,12 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_sub_i64, { "r", "r", "rA" } }, { INDEX_op_mul_i32, { "r", "r", "r" } }, { INDEX_op_mul_i64, { "r", "r", "r" } }, - { INDEX_op_and_i32, { "r", "r", "r" } }, - { INDEX_op_and_i64, { "r", "r", "r" } }, - { INDEX_op_or_i32, { "r", "r", "r" } }, - { INDEX_op_or_i64, { "r", "r", "r" } }, - { INDEX_op_xor_i32, { "r", "r", "r" } }, - { INDEX_op_xor_i64, { "r", "r", "r" } }, + { INDEX_op_and_i32, { "r", "r", "rwL" } }, + { INDEX_op_and_i64, { "r", "r", "rL" } }, + { INDEX_op_or_i32, { "r", "r", "rwL" } }, + { INDEX_op_or_i64, { "r", "r", "rL" } }, + { INDEX_op_xor_i32, { "r", "r", "rwL" } }, + { INDEX_op_xor_i64, { "r", "r", "rL" } }, { INDEX_op_shl_i32, { "r", "r", "ri" } }, { INDEX_op_shr_i32, { "r", "r", "ri" } },
Signature · 1 line
1.8.3.1
PATCH v4 13/33 tcg-aarch64: Support andc, orc, eqv, not
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 65 ++++++++++++++++++++++++++++++++++++++++++------ tcg/aarch64/tcg-target.h | 16 ++++++------ 2 files changed, 65 insertions(+), 16 deletions(-)
tcg/aarch64/tcg-target.c +57 -8
@@ -258,10 +258,12 @@ typedef enum {INSN_EORI =0x52000000 ,/* Logical shifted register instructions */ -INSN_AND =0x0a000000 , -INSN_ORR =0x2a000000 , -INSN_EOR =0x4a000000 , -INSN_ANDS =0x6a000000 , +INSN_AND =0x0a000000 , +INSN_BIC =0x0a200000 , +INSN_ORR =0x2a000000 , +INSN_ORN =0x2a200000 , +INSN_EOR =0x4a000000 , +INSN_EON =0x4a200000 ,/* Add/subtract immediate instructions */ INSN_ADDI = 0x11000000,
@@ -270,10 +272,10 @@ typedef enum {INSN_SUBSI =0x71000000 ,/* Add/subtract shifted register instructions */ -INSN_ADD =0x0b000000 , -INSN_ADDS =0x2b000000 , -INSN_SUB =0x4b000000 , -INSN_SUBS =0x6b000000 , +INSN_ADD =0x0b000000 , +INSN_ADDS =0x2b000000 , +INSN_SUB =0x4b000000 , +INSN_SUBS =0x6b000000 ,/* Data-processing (2 source) instructions */ INSN_LSLV = 0x1ac02000,
@@ -1270,6 +1272,17 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, }break ; +case INDEX_op_andc_i32: + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_andc_i64: +if (c2) { +tcg_fmt_Rdn_limm (s,INSN_ANDI , ext, a0, a1, ~a2); + }else { +tcg_fmt_Rdnm (s,INSN_BIC , ext, a0, a1, a2); + } +break ; +case INDEX_op_or_i32: a2 = (int32_t )a2;/* FALLTHRU */
@@ -1281,6 +1294,17 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, }break ; +case INDEX_op_orc_i32: + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_orc_i64: +if (c2) { +tcg_fmt_Rdn_limm (s,INSN_ORRI , ext, a0, a1, ~a2); + }else { +tcg_fmt_Rdnm (s,INSN_ORN , ext, a0, a1, a2); + } +break ; +case INDEX_op_xor_i32: a2 = (int32_t )a2;/* FALLTHRU */
@@ -1292,6 +1316,22 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, }break ; +case INDEX_op_eqv_i32: + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_eqv_i64: +if (c2) { +tcg_fmt_Rdn_limm (s,INSN_EORI , ext, a0, a1, ~a2); + }else { +tcg_fmt_Rdnm (s,INSN_EON , ext, a0, a1, a2); + } +break ; + +case INDEX_op_not_i64: +case INDEX_op_not_i32: +tcg_fmt_Rdnm (s,INSN_ORN , ext, a0,TCG_REG_XZR , a1); +break ; +case INDEX_op_mul_i64:case INDEX_op_mul_i32:tcg_out_mul (s, ext, a0, a1, a2);
@@ -1488,6 +1528,15 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_or_i64, { "r", "r", "rL" } }, { INDEX_op_xor_i32, { "r", "r", "rwL" } }, { INDEX_op_xor_i64, { "r", "r", "rL" } }, + { INDEX_op_andc_i32, { "r", "r", "rwL" } }, + { INDEX_op_andc_i64, { "r", "r", "rL" } }, + { INDEX_op_orc_i32, { "r", "r", "rwL" } }, + { INDEX_op_orc_i64, { "r", "r", "rL" } }, + { INDEX_op_eqv_i32, { "r", "r", "rwL" } }, + { INDEX_op_eqv_i64, { "r", "r", "rL" } }, + + { INDEX_op_not_i32, { "r", "r" } }, + { INDEX_op_not_i64, { "r", "r" } }, { INDEX_op_shl_i32, { "r", "r", "ri" } }, { INDEX_op_shr_i32, { "r", "r", "ri" } },
tcg/aarch64/tcg-target.h +8 -8
@@ -47,12 +47,12 @@ typedef enum {#define TCG_TARGET_HAS_ext16u_i32 1#define TCG_TARGET_HAS_bswap16_i32 1#define TCG_TARGET_HAS_bswap32_i32 1 -#define TCG_TARGET_HAS_not_i32 0 +#define TCG_TARGET_HAS_not_i32 1#define TCG_TARGET_HAS_neg_i32 0#define TCG_TARGET_HAS_rot_i32 1 -#define TCG_TARGET_HAS_andc_i32 0 -#define TCG_TARGET_HAS_orc_i32 0 -#define TCG_TARGET_HAS_eqv_i32 0 +#define TCG_TARGET_HAS_andc_i32 1 +#define TCG_TARGET_HAS_orc_i32 1 +#define TCG_TARGET_HAS_eqv_i32 1#define TCG_TARGET_HAS_nand_i32 0#define TCG_TARGET_HAS_nor_i32 0#define TCG_TARGET_HAS_deposit_i32 0
@@ -75,12 +75,12 @@ typedef enum {#define TCG_TARGET_HAS_bswap16_i64 1#define TCG_TARGET_HAS_bswap32_i64 1#define TCG_TARGET_HAS_bswap64_i64 1 -#define TCG_TARGET_HAS_not_i64 0 +#define TCG_TARGET_HAS_not_i64 1#define TCG_TARGET_HAS_neg_i64 0#define TCG_TARGET_HAS_rot_i64 1 -#define TCG_TARGET_HAS_andc_i64 0 -#define TCG_TARGET_HAS_orc_i64 0 -#define TCG_TARGET_HAS_eqv_i64 0 +#define TCG_TARGET_HAS_andc_i64 1 +#define TCG_TARGET_HAS_orc_i64 1 +#define TCG_TARGET_HAS_eqv_i64 1#define TCG_TARGET_HAS_nand_i64 0#define TCG_TARGET_HAS_nor_i64 0#define TCG_TARGET_HAS_deposit_i64 0
Signature · 1 line
1.8.3.1
PATCH v4 14/33 tcg-aarch64: Handle zero as first argument to sub
In reply to Richard HendersonIn order to properly handle neg, as generated by TCG generic code.
--- tcg/aarch64/tcg-target.c | 28 ++++++++++++++++++++++++---- 1 file changed, 24 insertions(+), 4 deletions(-)
tcg/aarch64/tcg-target.c +24 -4
@@ -114,6 +114,7 @@ static inline void patch_reloc(uint8_t *code_ptr, int type,#define TCG_CT_CONST_IS32 0x100#define TCG_CT_CONST_AIMM 0x200#define TCG_CT_CONST_LIMM 0x400 +#define TCG_CT_CONST_ZERO 0x800/* parse target specific constraints */ static int target_parse_constraint(TCGArgConstraint *ct,
@@ -147,6 +148,9 @@ static int target_parse_constraint(TCGArgConstraint *ct,case 'L' :/* Valid for logical immediate. */ ct->ct |=TCG_CT_CONST_LIMM ;break ; +case 'Z' :/* zero */ + ct->ct |=TCG_CT_CONST_ZERO ; +break ;default :return -1 ; }
@@ -198,6 +202,9 @@ static int tcg_target_const_match(tcg_target_long val,if ((ct &TCG_CT_CONST_LIMM ) &&is_limm (val)) {return 1 ; } +if ((ct &TCG_CT_CONST_ZERO ) && val ==0 ) { +return 1 ; + }return 0 ; }
@@ -1186,6 +1193,10 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,TCGArg a2 = args[2 ];int c2 = const_args[2 ]; +/* Some operands are defined with "rZ" constraint, a register or +the zero register. These need not actually test args[I] == 0. */ +#define REG0 (I ) (const_args[I] ? TCG_REG_XZR : (TCGReg)args[I]) +switch (opc) {case INDEX_op_exit_tb:tcg_out_movi (s,TCG_TYPE_I64 ,TCG_REG_X0 , a0);
@@ -1255,9 +1266,16 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,/* FALLTHRU */ case INDEX_op_sub_i64:if (c2) { -tcg_out_addsubi (s, ext, a0, a1, -a2); +/* Arithmetic immediate instructions use Xn|sp, and thus +we cannot encode the zero register if tcg optimization +is turned off and both A1 and A2 are constants. */ +if (const_args[1 ]) { +tcg_out_movi (s, ext ?TCG_TYPE_I64 :TCG_TYPE_I32 , a0, -a2); + }else { +tcg_out_addsubi (s, ext, a0, a1, -a2); + } }else { -tcg_fmt_Rdnm (s,INSN_SUB , ext, a0, a1, a2); +tcg_fmt_Rdnm (s,INSN_SUB , ext, a0,REG0 (1 ), a2); }break ;
@@ -1481,6 +1499,8 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,/* Opcode not implemented. */ tcg_abort (); } + +#undef REG0 }static const TCGTargetOpDef aarch64_op_defs[] = {
@@ -1518,8 +1538,8 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_add_i32, { "r", "r", "rwA" } }, { INDEX_op_add_i64, { "r", "r", "rA" } }, - { INDEX_op_sub_i32, { "r", "r", "rwA" } }, - { INDEX_op_sub_i64, { "r", "r", "rA" } }, + { INDEX_op_sub_i32, { "r", "rZ", "rwA" } }, + { INDEX_op_sub_i64, { "r", "rZ", "rA" } }, { INDEX_op_mul_i32, { "r", "r", "r" } }, { INDEX_op_mul_i64, { "r", "r", "r" } }, { INDEX_op_and_i32, { "r", "r", "rwL" } },
Signature · 1 line
1.8.3.1
PATCH v4 15/33 tcg-aarch64: Support movcond
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 24 +++++++++++++++++++++++- tcg/aarch64/tcg-target.h | 4 ++-- 2 files changed, 25 insertions(+), 3 deletions(-)
tcg/aarch64/tcg-target.c +23 -1
@@ -289,6 +289,10 @@ typedef enum {INSN_LSRV =0x1ac02400 ,INSN_ASRV =0x1ac02800 ,INSN_RORV =0x1ac02c00 , + +/* Conditional select instructions */ +INSN_CSEL =0x1a800000 , +INSN_CSINC =0x1a800400 , } AArch64Insn; static inline enum aarch64_ldst_op_data
@@ -451,6 +455,14 @@ static void tcg_fmt_Rdn_limm(TCGContext *s, AArch64Insn insn, TCGType sf,tcg_fmt_Rdn_nrs (s, insn, sf, rd, rn, sf, r, c); } +static inline void tcg_fmt_Rdnm_cond (TCGContext *s,AArch64Insn insn, +TCGType sf,TCGReg rd,TCGReg rn, +TCGReg rm,TCGCond c) +{ +tcg_out32 (s, insn | sf <<31 | rm <<16 | rn <<5 | rd + | tcg_cond_to_aarch64[c] <<12 ); +} + static inline void tcg_out_ldst_9(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type,
@@ -1414,7 +1426,15 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,/* FALLTHRU */ case INDEX_op_setcond_i64:tcg_out_cmp (s, ext, a1, a2, c2); -tcg_out_cset (s,0 , a0, args[3 ]); +tcg_out_cset (s, ext, a0, args[3 ]); +break ; + +case INDEX_op_movcond_i32: + a2 = (int32_t )a2; +/* FALLTHRU */ +case INDEX_op_movcond_i64: +tcg_out_cmp (s, ext, a1, a2, c2); +tcg_fmt_Rdnm_cond (s,INSN_CSEL , ext, a0,REG0 (3 ),REG0 (4 ), args[5 ]);break ;case INDEX_op_qemu_ld8u:
@@ -1573,6 +1593,8 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_brcond_i64, { "r", "rA" } }, { INDEX_op_setcond_i32, { "r", "r", "rwA" } }, { INDEX_op_setcond_i64, { "r", "r", "rA" } }, + { INDEX_op_movcond_i32, { "r", "r", "rwA", "rZ", "rZ" } }, + { INDEX_op_movcond_i64, { "r", "r", "rwA", "rZ", "rZ" } }, { INDEX_op_qemu_ld8u, { "r", "l" } }, { INDEX_op_qemu_ld8s, { "r", "l" } },
tcg/aarch64/tcg-target.h +2 -2
@@ -56,7 +56,7 @@ typedef enum {#define TCG_TARGET_HAS_nand_i32 0#define TCG_TARGET_HAS_nor_i32 0#define TCG_TARGET_HAS_deposit_i32 0 -#define TCG_TARGET_HAS_movcond_i32 0 +#define TCG_TARGET_HAS_movcond_i32 1#define TCG_TARGET_HAS_add2_i32 0#define TCG_TARGET_HAS_sub2_i32 0#define TCG_TARGET_HAS_mulu2_i32 0
@@ -84,7 +84,7 @@ typedef enum {#define TCG_TARGET_HAS_nand_i64 0#define TCG_TARGET_HAS_nor_i64 0#define TCG_TARGET_HAS_deposit_i64 0 -#define TCG_TARGET_HAS_movcond_i64 0 +#define TCG_TARGET_HAS_movcond_i64 1#define TCG_TARGET_HAS_add2_i64 0#define TCG_TARGET_HAS_sub2_i64 0#define TCG_TARGET_HAS_mulu2_i64 0
Signature · 1 line
1.8.3.1
PATCH v4 16/33 tcg-aarch64: Use tcg_fmt_Rdnm_cond for setcond
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 12 +++--------- 1 file changed, 3 insertions(+), 9 deletions(-)
tcg/aarch64/tcg-target.c +3 -9
@@ -690,14 +690,6 @@ static void tcg_out_cmp(TCGContext *s, TCGType ext, TCGReg a, } } -static inline void tcg_out_cset (TCGContext *s,TCGType ext, -TCGReg rd,TCGCond c) -{ -/* Using CSET alias of CSINC 0x1a800400 Xd, XZR, XZR, invert(cond) */ -unsigned int base = ext ?0x9a9f07e0 :0x1a9f07e0 ; -tcg_out32 (s, base | tcg_cond_to_aarch64[tcg_invert_cond (c)] <<12 | rd); -} -static inline void tcg_out_goto (TCGContext *s,tcg_target_long target) {tcg_target_long offset;
@@ -1426,7 +1418,9 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,/* FALLTHRU */ case INDEX_op_setcond_i64:tcg_out_cmp (s, ext, a1, a2, c2); -tcg_out_cset (s, ext, a0, args[3 ]); +/* Use CSET alias of CSINC Wd, WZR, WZR, invert(cond). */ +tcg_fmt_Rdnm_cond (s,INSN_CSINC ,0 , a0,TCG_REG_XZR ,TCG_REG_XZR , +tcg_invert_cond (args[3 ]));break ;case INDEX_op_movcond_i32:
Signature · 1 line
1.8.3.1
PATCH v4 17/33 tcg-aarch64: Support deposit
In reply to Richard HendersonAlso tidy the implementation of ubfm, sbfm, extr in order to share code.
--- tcg/aarch64/tcg-target.c | 49 +++++++++++++++++++++++++++++++++++------------- tcg/aarch64/tcg-target.h | 4 ++-- 2 files changed, 38 insertions(+), 15 deletions(-)
tcg/aarch64/tcg-target.c +36 -13
@@ -290,6 +290,12 @@ typedef enum {INSN_ASRV =0x1ac02800 ,INSN_RORV =0x1ac02c00 , +/* Bitfield instructions */ +INSN_BFM =0x33000000 , +INSN_SBFM =0x13000000 , +INSN_UBFM =0x53000000 , +INSN_EXTR =0x13800000 , +/* Conditional select instructions */ INSN_CSEL =0x1a800000 , INSN_CSINC = 0x1a800400,
@@ -610,28 +616,30 @@ static inline void tcg_out_mul(TCGContext *s, TCGType ext,tcg_out32 (s, base | rm <<16 | rn <<5 | rd); } +static inline void tcg_out_bfm (TCGContext *s,TCGType ext,TCGReg rd, +TCGReg rn,unsigned int a,unsigned int b) +{ +tcg_fmt_Rdn_nrs (s,INSN_BFM , ext, rd, rn, ext, a, b); +} +static inline void tcg_out_ubfm (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int a,unsigned int b) { -/* Using UBFM 0x53000000 Wd, Wn, a, b */ -unsigned int base = ext ?0xd3400000 :0x53000000 ; -tcg_out32 (s, base | a <<16 | b <<10 | rn <<5 | rd); +tcg_fmt_Rdn_nrs (s,INSN_UBFM , ext, rd, rn, ext, a, b); }static inline void tcg_out_sbfm (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int a,unsigned int b) { -/* Using SBFM 0x13000000 Wd, Wn, a, b */ -unsigned int base = ext ?0x93400000 :0x13000000 ; -tcg_out32 (s, base | a <<16 | b <<10 | rn <<5 | rd); +tcg_fmt_Rdn_nrs (s,INSN_SBFM , ext, rd, rn, ext, a, b); }static inline void tcg_out_extr (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,TCGReg rm,unsigned int a) { -/* Using EXTR 0x13800000 Wd, Wn, Wm, a */ -unsigned int base = ext ?0x93c00000 :0x13800000 ; -tcg_out32 (s, base | rm <<16 | a <<10 | rn <<5 | rd); +/* ??? Abuse of the fmt function, but RM field is at the same location +as the IMMR field and we do still need to set N = SF for IMMS. */ +tcg_fmt_Rdn_nrs (s,INSN_EXTR , ext, rd, rn, ext, rm, a); } static inline void tcg_out_shl(TCGContext *s, TCGType ext,
@@ -673,6 +681,15 @@ static inline void tcg_out_rotl(TCGContext *s, TCGType ext,tcg_out_extr (s, ext, rd, rn, rn, bits - (m & max)); } +static inline void tcg_out_dep (TCGContext *s,TCGType ext,TCGReg rd, +TCGReg rn,unsigned lsb,unsigned width) +{ +unsigned size = ext ?64 :32 ; +unsigned a = (size - lsb) & (size -1 ); +unsigned b = width -1 ; +tcg_out_bfm (s, ext, rd, rn, a, b); +} +static void tcg_out_cmp (TCGContext *s,TCGType ext,TCGReg a,tcg_target_long b,bool const_b) {
@@ -828,8 +845,7 @@ static inline void tcg_out_rev16(TCGContext *s, TCGType ext,static inline void tcg_out_sxt (TCGContext *s,TCGType ext,int s_bits,TCGReg rd,TCGReg rn) { -/* using ALIASes SXTB 0x13001c00, SXTH 0x13003c00, SXTW 0x93407c00 -of SBFM Xd, Xn, #0, #7|15|31 */ +/* Using ALIASes SXTB, SXTH, SXTW, of SBFM Xd, Xn, #0, #7|15|31 */ int bits =8 * (1 << s_bits) -1 ;tcg_out_sbfm (s, ext, rd, rn,0 , bits); }
@@ -837,8 +853,7 @@ static inline void tcg_out_sxt(TCGContext *s, TCGType ext, int s_bits,static inline void tcg_out_uxt (TCGContext *s,int s_bits,TCGReg rd,TCGReg rn) { -/* using ALIASes UXTB 0x53001c00, UXTH 0x53003c00 -of UBFM Wd, Wn, #0, #7|15 */ +/* Using ALIASes UXTB, UXTH of UBFM Wd, Wn, #0, #7|15 */ int bits =8 * (1 << s_bits) -1 ;tcg_out_ubfm (s,0 , rd, rn,0 , bits); }
@@ -1504,6 +1519,11 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,tcg_out_movr (s,0 , a0, a1);break ; +case INDEX_op_deposit_i64: +case INDEX_op_deposit_i32: +tcg_out_dep (s, ext, a0,REG0 (2 ), args[3 ], args[4 ]); +break ; +case INDEX_op_mov_i64:case INDEX_op_mov_i32:case INDEX_op_movi_i64:
@@ -1623,6 +1643,9 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_ext16u_i64, { "r", "r" } }, { INDEX_op_ext32u_i64, { "r", "r" } }, + { INDEX_op_deposit_i32, { "r", "0", "rZ" } }, + { INDEX_op_deposit_i64, { "r", "0", "rZ" } }, + { -1 }, };
tcg/aarch64/tcg-target.h +2 -2
@@ -55,7 +55,7 @@ typedef enum {#define TCG_TARGET_HAS_eqv_i32 1#define TCG_TARGET_HAS_nand_i32 0#define TCG_TARGET_HAS_nor_i32 0 -#define TCG_TARGET_HAS_deposit_i32 0 +#define TCG_TARGET_HAS_deposit_i32 1#define TCG_TARGET_HAS_movcond_i32 1#define TCG_TARGET_HAS_add2_i32 0#define TCG_TARGET_HAS_sub2_i32 0
@@ -83,7 +83,7 @@ typedef enum {#define TCG_TARGET_HAS_eqv_i64 1#define TCG_TARGET_HAS_nand_i64 0#define TCG_TARGET_HAS_nor_i64 0 -#define TCG_TARGET_HAS_deposit_i64 0 +#define TCG_TARGET_HAS_deposit_i64 1#define TCG_TARGET_HAS_movcond_i64 1#define TCG_TARGET_HAS_add2_i64 0#define TCG_TARGET_HAS_sub2_i64 0
Signature · 1 line
1.8.3.1
PATCH v4 18/33 tcg-aarch64: Support add2, sub2
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 78 ++++++++++++++++++++++++++++++++++++++++++++++++ tcg/aarch64/tcg-target.h | 8 ++--- 2 files changed, 82 insertions(+), 4 deletions(-)
tcg/aarch64/tcg-target.c +78 -0
@@ -115,6 +115,7 @@ static inline void patch_reloc(uint8_t *code_ptr, int type,#define TCG_CT_CONST_AIMM 0x200#define TCG_CT_CONST_LIMM 0x400#define TCG_CT_CONST_ZERO 0x800 +#define TCG_CT_CONST_MONE 0x1000/* parse target specific constraints */ static int target_parse_constraint(TCGArgConstraint *ct,
@@ -148,6 +149,9 @@ static int target_parse_constraint(TCGArgConstraint *ct,case 'L' :/* Valid for logical immediate. */ ct->ct |=TCG_CT_CONST_LIMM ;break ; +case 'M' :/* minus one */ + ct->ct |=TCG_CT_CONST_MONE ; +break ;case 'Z' :/* zero */ ct->ct |=TCG_CT_CONST_ZERO ;break ;
@@ -205,6 +209,9 @@ static int tcg_target_const_match(tcg_target_long val,if ((ct &TCG_CT_CONST_ZERO ) && val ==0 ) {return 1 ; } +if ((ct &TCG_CT_CONST_MONE ) && val ==-1 ) { +return 1 ; + }return 0 ; }
@@ -284,6 +291,10 @@ typedef enum {INSN_SUB =0x4b000000 ,INSN_SUBS =0x6b000000 , +/* Add/subtract with carry instructions */ +INSN_ADC =0x1a000000 , +INSN_SBC =0x5a000000 , +/* Data-processing (2 source) instructions */ INSN_LSLV =0x1ac02000 , INSN_LSRV = 0x1ac02400,
@@ -869,6 +880,47 @@ static void tcg_out_addsubi(TCGContext *s, int ext, TCGReg rd,tcg_fmt_Rdn_aimm (s, insn, ext, rd, rn, aimm); } +static inline void tcg_out_addsub2 (TCGContext *s,int ext,TCGReg rl, +TCGReg rh,TCGReg al,TCGReg ah, +tcg_target_long bl,tcg_target_long bh, +bool const_bl,bool const_bh,bool sub) +{ +TCGReg orig_rl = rl; +AArch64Insn insn; + +if (rl == ah || (!const_bh && rl == bh)) { + rl =TCG_REG_TMP ; + } + +if (const_bl) { + insn =INSN_ADDSI ; +if ((bl <0 ) ^ sub) { + insn =INSN_SUBSI ; + bl = -bl; + } +tcg_fmt_Rdn_aimm (s, insn, ext, rl, al, bl); + }else { +tcg_fmt_Rdnm (s, sub ?INSN_SUBS :INSN_ADDS , ext, rl, al, bl); + } + + insn =INSN_ADC ; +if (const_bh) { +/* Note that the only two constants we support are 0 and -1, and +that SBC = rn + ~rm + c, so adc -1 is sbc 0, and vice-versa. */ +if ((bh !=0 ) ^ sub) { + insn =INSN_SBC ; + } + bh =TCG_REG_XZR ; + }else if (sub) { + insn =INSN_SBC ; + } +tcg_fmt_Rdnm (s, insn, ext, rh, ah, bh); + +if (rl != orig_rl) { +tcg_out_movr (s, ext, orig_rl, rl); + } +} +static inline void tcg_out_nop (TCGContext *s) {tcg_out32 (s,0xd503201f );
@@ -1524,6 +1576,27 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,tcg_out_dep (s, ext, a0,REG0 (2 ), args[3 ], args[4 ]);break ; +case INDEX_op_add2_i32: + a2 = (int32_t )args[4 ]; + c2 = false; + goto do_addsub2; +case INDEX_op_add2_i64: + a2 = args[4 ]; + c2 = false; + goto do_addsub2; +case INDEX_op_sub2_i32: + a2 = (int32_t )args[4 ]; + c2 = true; + goto do_addsub2; +case INDEX_op_sub2_i64: + a2 = args[4 ]; + c2 = true; + goto do_addsub2; + do_addsub2: +tcg_out_addsub2 (s, ext, a0, a1,REG0 (2 ),REG0 (3 ), a2, + args[5 ], const_args[4 ], const_args[5 ], c2); +break ; +case INDEX_op_mov_i64:case INDEX_op_mov_i32:case INDEX_op_movi_i64:
@@ -1646,6 +1719,11 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_deposit_i32, { "r", "0", "rZ" } }, { INDEX_op_deposit_i64, { "r", "0", "rZ" } }, + { INDEX_op_add2_i32, { "r", "r", "rZ", "rZ", "rwA", "rwMZ" } }, + { INDEX_op_add2_i64, { "r", "r", "rZ", "rZ", "rA", "rMZ" } }, + { INDEX_op_sub2_i32, { "r", "r", "rZ", "rZ", "rwA", "rwMZ" } }, + { INDEX_op_sub2_i64, { "r", "r", "rZ", "rZ", "rA", "rMZ" } }, + { -1 }, };
tcg/aarch64/tcg-target.h +4 -4
@@ -57,8 +57,8 @@ typedef enum {#define TCG_TARGET_HAS_nor_i32 0#define TCG_TARGET_HAS_deposit_i32 1#define TCG_TARGET_HAS_movcond_i32 1 -#define TCG_TARGET_HAS_add2_i32 0 -#define TCG_TARGET_HAS_sub2_i32 0 +#define TCG_TARGET_HAS_add2_i32 1 +#define TCG_TARGET_HAS_sub2_i32 1#define TCG_TARGET_HAS_mulu2_i32 0#define TCG_TARGET_HAS_muls2_i32 0#define TCG_TARGET_HAS_muluh_i32 0
@@ -85,8 +85,8 @@ typedef enum {#define TCG_TARGET_HAS_nor_i64 0#define TCG_TARGET_HAS_deposit_i64 1#define TCG_TARGET_HAS_movcond_i64 1 -#define TCG_TARGET_HAS_add2_i64 0 -#define TCG_TARGET_HAS_sub2_i64 0 +#define TCG_TARGET_HAS_add2_i64 1 +#define TCG_TARGET_HAS_sub2_i64 1#define TCG_TARGET_HAS_mulu2_i64 0#define TCG_TARGET_HAS_muls2_i64 0#define TCG_TARGET_HAS_muluh_i64 0
Signature · 1 line
1.8.3.1
PATCH v4 19/33 tcg-aarch64: Support muluh, mulsh
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 12 ++++++++++++ tcg/aarch64/tcg-target.h | 4 ++-- 2 files changed, 14 insertions(+), 2 deletions(-)
tcg/aarch64/tcg-target.c +12 -0
@@ -300,6 +300,8 @@ typedef enum {INSN_LSRV =0x1ac02400 ,INSN_ASRV =0x1ac02800 ,INSN_RORV =0x1ac02c00 , +INSN_SMULH =0x9b407c00 , +INSN_UMULH =0x9bc07c00 ,/* Bitfield instructions */ INSN_BFM = 0x33000000,
@@ -1597,6 +1599,13 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, args[5], const_args[4], const_args[5], c2);break ; +case INDEX_op_muluh_i64: +tcg_fmt_Rdnm (s,INSN_UMULH ,1 , a0, a1, a2); +break ; +case INDEX_op_mulsh_i64: +tcg_fmt_Rdnm (s,INSN_SMULH ,1 , a0, a1, a2); +break ; +case INDEX_op_mov_i64:case INDEX_op_mov_i32:case INDEX_op_movi_i64:
@@ -1724,6 +1733,9 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_sub2_i32, { "r", "r", "rZ", "rZ", "rwA", "rwMZ" } }, { INDEX_op_sub2_i64, { "r", "r", "rZ", "rZ", "rA", "rMZ" } }, + { INDEX_op_muluh_i64, { "r", "r", "r" } }, + { INDEX_op_mulsh_i64, { "r", "r", "r" } }, + { -1 }, };
tcg/aarch64/tcg-target.h +2 -2
@@ -89,8 +89,8 @@ typedef enum {#define TCG_TARGET_HAS_sub2_i64 1#define TCG_TARGET_HAS_mulu2_i64 0#define TCG_TARGET_HAS_muls2_i64 0 -#define TCG_TARGET_HAS_muluh_i64 0 -#define TCG_TARGET_HAS_mulsh_i64 0 +#define TCG_TARGET_HAS_muluh_i64 1 +#define TCG_TARGET_HAS_mulsh_i64 1enum {TCG_AREG0 =TCG_REG_X19 ,
Signature · 1 line
1.8.3.1
PATCH v4 20/33 tcg-aarch64: Support div, rem
In reply to Richard HendersonFor remainder, generic code will produce mul+sub,
whereas we can implement with msub.
--- tcg/aarch64/tcg-target.c | 51 +++++++++++++++++++++++++++++++++++++++--------- tcg/aarch64/tcg-target.h | 8 ++++---- 2 files changed, 46 insertions(+), 13 deletions(-)
tcg/aarch64/tcg-target.c +42 -9
@@ -302,6 +302,12 @@ typedef enum {INSN_RORV =0x1ac02c00 ,INSN_SMULH =0x9b407c00 ,INSN_UMULH =0x9bc07c00 , +INSN_UDIV =0x1ac00800 , +INSN_SDIV =0x1ac00c00 , + +/* Data-processing (3 source) instructions */ +INSN_MADD =0x1b000000 , +INSN_MSUB =0x1b008000 ,/* Bitfield instructions */ INSN_BFM = 0x33000000,
@@ -416,6 +422,13 @@ static inline void tcg_fmt_Rdnm(TCGContext *s, AArch64Insn insn, TCGType sf,tcg_out32 (s, insn | sf <<31 | rm <<16 | rn <<5 | rd); } +/* This function is used for the Multiply instruction group. */ +static inline void tcg_fmt_Rdnma (TCGContext *s,AArch64Insn insn,TCGType sf, +TCGReg rd,TCGReg rn,TCGReg rm,TCGReg ra) +{ +tcg_out32 (s, insn | sf <<31 | rm <<16 | ra <<10 | rn <<5 | rd); +} +/* This function is used for the Arithmetic (immediate) instruction group. The value of AIMM must be appropriate for encoding in the shift+imm12 fields. */
@@ -621,14 +634,6 @@ static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, arg, arg1, arg2); } -static inline void tcg_out_mul (TCGContext *s,TCGType ext, -TCGReg rd,TCGReg rn,TCGReg rm) -{ -/* Using MADD 0x1b000000 with Ra = wzr alias MUL 0x1b007c00 */ -unsigned int base = ext ?0x9b007c00 :0x1b007c00 ; -tcg_out32 (s, base | rm <<16 | rn <<5 | rd); -} -static inline void tcg_out_bfm (TCGContext *s,TCGType ext,TCGReg rd,TCGReg rn,unsigned int a,unsigned int b) {
@@ -1425,7 +1430,27 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,case INDEX_op_mul_i64:case INDEX_op_mul_i32: -tcg_out_mul (s, ext, a0, a1, a2); +tcg_fmt_Rdnma (s,INSN_MADD , ext, a0, a1, a2,TCG_REG_XZR ); +break ; + +case INDEX_op_div_i64: +case INDEX_op_div_i32: +tcg_fmt_Rdnm (s,INSN_SDIV , ext, a0, a1, a2); +break ; +case INDEX_op_divu_i64: +case INDEX_op_divu_i32: +tcg_fmt_Rdnm (s,INSN_UDIV , ext, a0, a1, a2); +break ; + +case INDEX_op_rem_i64: +case INDEX_op_rem_i32: +tcg_fmt_Rdnm (s,INSN_SDIV , ext,TCG_REG_TMP , a1, a2); +tcg_fmt_Rdnma (s,INSN_MSUB , ext, a0,TCG_REG_TMP , a2, a1); +break ; +case INDEX_op_remu_i64: +case INDEX_op_remu_i32: +tcg_fmt_Rdnm (s,INSN_UDIV , ext,TCG_REG_TMP , a1, a2); +tcg_fmt_Rdnma (s,INSN_MSUB , ext, a0,TCG_REG_TMP , a2, a1);break ;case INDEX_op_shl_i64:
@@ -1658,6 +1683,14 @@ static const TCGTargetOpDef aarch64_op_defs[] = { { INDEX_op_sub_i64, { "r", "rZ", "rA" } }, { INDEX_op_mul_i32, { "r", "r", "r" } }, { INDEX_op_mul_i64, { "r", "r", "r" } }, + { INDEX_op_div_i32, { "r", "r", "r" } }, + { INDEX_op_div_i64, { "r", "r", "r" } }, + { INDEX_op_divu_i32, { "r", "r", "r" } }, + { INDEX_op_divu_i64, { "r", "r", "r" } }, + { INDEX_op_rem_i32, { "r", "r", "r" } }, + { INDEX_op_rem_i64, { "r", "r", "r" } }, + { INDEX_op_remu_i32, { "r", "r", "r" } }, + { INDEX_op_remu_i64, { "r", "r", "r" } }, { INDEX_op_and_i32, { "r", "r", "rwL" } }, { INDEX_op_and_i64, { "r", "r", "rL" } }, { INDEX_op_or_i32, { "r", "r", "rwL" } },
tcg/aarch64/tcg-target.h +4 -4
@@ -39,8 +39,8 @@ typedef enum {#define TCG_TARGET_CALL_STACK_OFFSET 0/* optional instructions */ -#define TCG_TARGET_HAS_div_i32 0 -#define TCG_TARGET_HAS_rem_i32 0 +#define TCG_TARGET_HAS_div_i32 1 +#define TCG_TARGET_HAS_rem_i32 1#define TCG_TARGET_HAS_ext8s_i32 1#define TCG_TARGET_HAS_ext16s_i32 1#define TCG_TARGET_HAS_ext8u_i32 1
@@ -64,8 +64,8 @@ typedef enum {#define TCG_TARGET_HAS_muluh_i32 0#define TCG_TARGET_HAS_mulsh_i32 0 -#define TCG_TARGET_HAS_div_i64 0 -#define TCG_TARGET_HAS_rem_i64 0 +#define TCG_TARGET_HAS_div_i64 1 +#define TCG_TARGET_HAS_rem_i64 1#define TCG_TARGET_HAS_ext8s_i64 1#define TCG_TARGET_HAS_ext16s_i64 1#define TCG_TARGET_HAS_ext32s_i64 1
Signature · 1 line
1.8.3.1
PATCH v4 21/33 tcg-aarch64: Introduce tcg_fmt_Rd_uimm
In reply to Richard HendersonCleaning up the implementation of tcg_out_movi at the same time.
--- tcg/aarch64/tcg-target.c | 49 +++++++++++++++++++++++++++--------------------- 1 file changed, 28 insertions(+), 21 deletions(-)
tcg/aarch64/tcg-target.c +28 -21
@@ -279,6 +279,11 @@ typedef enum {INSN_EOR =0x4a000000 ,INSN_EON =0x4a200000 , +/* Move wide immediate instructions */ +INSN_MOVN =0x12800000 , +INSN_MOVZ =0x52800000 , +INSN_MOVK =0x72800000 , +/* Add/subtract immediate instructions */ INSN_ADDI =0x11000000 , INSN_ADDSI = 0x31000000,
@@ -495,6 +500,16 @@ static inline void tcg_fmt_Rdnm_cond(TCGContext *s, AArch64Insn insn, | tcg_cond_to_aarch64[c] << 12); } +/* This function is used for the Move (wide immediate) instruction group. +Note that SHIFT is a full shift count, not the 2 bit HW field. */ +static inline void tcg_fmt_Rd_uimm (TCGContext *s,AArch64Insn insn, +TCGType sf,TCGReg rd,uint16_t half, +unsigned shift) +{ +assert ((shift & ~0x30 ) ==0 ); +tcg_out32 (s, insn | sf <<31 | shift << (21 -4 ) | half <<5 | rd); +} + static inline void tcg_out_ldst_9(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type,
@@ -540,38 +555,30 @@ static inline void tcg_out_movr_sp(TCGContext *s, TCGType ext,tcg_fmt_Rdn_aimm (s,INSN_ADDI , ext, rd, rn,0 ); } -static inline void tcg_out_movi_aux (TCGContext *s, -TCGReg rd,uint64_t value) +static void tcg_out_movi (TCGContext *s,TCGType type,TCGReg rd, +tcg_target_long value) { -uint32_t half, base, shift, movk =0 ; -/* construct halfwords of the immediate with MOVZ/MOVK with LSL */ -/* using MOVZ 0x52800000 | extended reg.. */ - base = (value >0xffffffff ) ?0xd2800000 :0x52800000 ; +AArch64Insn insn; + +if (type ==TCG_TYPE_I32 ) { + value = (uint32_t )value; + } +/* count trailing zeros in 16 bit steps, mapping 64 to 0. Emit the first MOVZ with the half-word immediate skipping the zeros, with a shift -(LSL) equal to this number. Then morph all next instructions into MOVKs. +(LSL) equal to this number. Then all next instructions use MOVKs. Zero the processed half-word in the value, continue until empty. We build the final result 16bits at a time with up to 4 instructions, but do not emit instructions for 16bit zero holes. */ + insn =INSN_MOVZ ;do { - shift =ctz64 (value) & (63 &-16 ); - half = (value >> shift) &0xffff ; -tcg_out32 (s, base | movk | shift <<17 | half <<5 | rd); - movk =0x20000000 ;/* morph next MOVZs into MOVKs */ +unsigned shift =ctz64 (value) & (63 &-16 ); +tcg_fmt_Rd_uimm (s, insn, shift >=32 , rd, value >> shift, shift); value &= ~(0xffffUL << shift); + insn =INSN_MOVK ; }while (value); } -static inline void tcg_out_movi (TCGContext *s,TCGType type, -TCGReg rd,tcg_target_long value) -{ -if (type ==TCG_TYPE_I64 ) { -tcg_out_movi_aux (s, rd, value); - }else { -tcg_out_movi_aux (s, rd, value &0xffffffff ); - } -} - static inline void tcg_out_ldst_r(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type,
Signature · 1 line
1.8.3.1
PATCH v4 22/33 tcg-aarch64: Use MOVN in tcg_out_movi
In reply to Richard HendersonWhen profitable, initialize the register with MOVN instead of MOVZ, before setting the remaining lanes with MOVK.
--- tcg/aarch64/tcg-target.c | 88 +++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 75 insertions(+), 13 deletions(-)
tcg/aarch64/tcg-target.c +75 -13
@@ -559,24 +559,86 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd, tcg_target_long value) {AArch64Insn insn; - -if (type ==TCG_TYPE_I32 ) { +int i, wantinv, shift; +tcg_target_long svalue = value; +tcg_target_long ivalue, imask; + +/* For 32-bit values, discard potential garbage in value. For 64-bit +values within [2**31, 2**32-1], we can create smaller sequences by +interpreting this as a negative 32-bit number, while ensuring that +the high 32 bits are cleared by setting SF=0. */ +if (type ==TCG_TYPE_I32 || (value & ~0xffffffffull ) ==0 ) { + svalue = (int32_t )value; value = (uint32_t )value; + type =TCG_TYPE_I32 ; + } + +/* Would it take fewer insns to begin with MOVN? For the value and its +inverse, count the number of 16-bit lanes that are 0. For the benefit +of 32-bit quantities, compare the zero-extended normal value vs the +sign-extended inverted value. For example, +v = 0x00000000f100ffff, zeros = 2 +~v = 0xffffffff0eff0000, zeros = 1 +~sv = 0x000000000eff0000, zeros = 3 +By using ~sv we see that 3 > 2, leading us to emit just a single insn +"movn ret, 0x0eff, lsl #16". */ + + ivalue = ~svalue; + imask =0 ; + wantinv =0 ; + +/* ??? This can be done in the simd unit without a loop: +// Move value and ivalue into V0 and V1 respectively. +mov v0.d[0], value +mov v1.d[0], ivalue +// Compare each 16-bit lane vs 0, producing -1 for true. +cmeq v0.4h, v0.4h, #0 +cmeq v1.4h, v1.4h, #0 +mov imask, v1.d[0] +// Sum the comparisons, producing 0 to -4. +addv h0, v0.4h +addv h1, v1.4h +// Subtract the two, forming a positive wantinv result. +sub v0.4h, v0.4h, v1.4h +smov wantinv, v0.h[0] +*/ +for (i =0 ; i <64 ; i +=16 ) { +tcg_target_long mask =0xffffull << i; +if ((value & mask) ==0 ) { + wantinv -=1 ; + } +if ((ivalue & mask) ==0 ) { + wantinv +=1 ; + imask |= mask; + } } -/* count trailing zeros in 16 bit steps, mapping 64 to 0. Emit the -first MOVZ with the half-word immediate skipping the zeros, with a shift -(LSL) equal to this number. Then all next instructions use MOVKs. -Zero the processed half-word in the value, continue until empty. -We build the final result 16bits at a time with up to 4 instructions, -but do not emit instructions for 16bit zero holes. */ +/* If we had more 0xffff than 0x0000, invert VALUE and use MOVN. */ insn =INSN_MOVZ ; -do { -unsigned shift =ctz64 (value) & (63 &-16 ); -tcg_fmt_Rd_uimm (s, insn, shift >=32 , rd, value >> shift, shift); +if (wantinv >0 ) { + value = ivalue; + insn =INSN_MOVN ; + } + +/* Find the lowest lane that is not 0x0000. */ + shift =ctz64 (value) & (63 &-16 ); +tcg_fmt_Rd_uimm (s, insn, type, rd, value >> shift, shift); + +if (wantinv >0 ) { +/* Re-invert the value, so MOVK sees non-inverted bits. */ + value = ~value; +/* Clear out all the 0xffff lanes. */ + value ^= imask; + } +/* Clear out the lane that we just set. */ + value &= ~(0xffffUL << shift); + +/* Iterate until all lanes have been set, and thus cleared from VALUE. */ +while (value) { + shift =ctz64 (value) & (63 &-16 ); +tcg_fmt_Rd_uimm (s,INSN_MOVK , type, rd, value >> shift, shift); value &= ~(0xffffUL << shift); - insn =INSN_MOVK ; - }while (value); + } } static inline void tcg_out_ldst_r(TCGContext *s,
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 22/33] tcg-aarch64: Use MOVN in tcg_out_movi
Quoted 103 lines · Richard Henderson
When profitable, initialize the register with MOVN instead of MOVZ, before setting the remaining lanes with MOVK.
Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 88 +++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 75 insertions(+), 13 deletions(-)tcg/aarch64/tcg-target.c +71
-11@@ -559,24 +559,86 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd, tcg_target_long value) {AArch64Insn insn; - -if (type ==TCG_TYPE_I32 ) { +int i, wantinv, shift; +tcg_target_long svalue = value; +tcg_target_long ivalue, imask; + +/* For 32-bit values, discard potential garbage in value. For 64-bit +values within [2**31, 2**32-1], we can create smaller sequences by +interpreting this as a negative 32-bit number, while ensuring that +the high 32 bits are cleared by setting SF=0. */ +if (type ==TCG_TYPE_I32 || (value & ~0xffffffffull ) ==0 ) { + svalue = (int32_t )value; value = (uint32_t )value; + type =TCG_TYPE_I32 ; + } + +/* Would it take fewer insns to begin with MOVN? For the value and its +inverse, count the number of 16-bit lanes that are 0. For the benefit +of 32-bit quantities, compare the zero-extended normal value vs the +sign-extended inverted value. For example, +v = 0x00000000f100ffff, zeros = 2 +~v = 0xffffffff0eff0000, zeros = 1 +~sv = 0x000000000eff0000, zeros = 3 +By using ~sv we see that 3 > 2, leading us to emit just a single insn +"movn ret, 0x0eff, lsl #16". */ + + ivalue = ~svalue; + imask =0 ; + wantinv =0 ; + +/* ??? This can be done in the simd unit without a loop: +// Move value and ivalue into V0 and V1 respectively. +mov v0.d[0], value +mov v1.d[0], ivalue +// Compare each 16-bit lane vs 0, producing -1 for true. +cmeq v0.4h, v0.4h, #0 +cmeq v1.4h, v1.4h, #0 +mov imask, v1.d[0] +// Sum the comparisons, producing 0 to -4. +addv h0, v0.4h +addv h1, v1.4h +// Subtract the two, forming a positive wantinv result. +sub v0.4h, v0.4h, v1.4h +smov wantinv, v0.h[0] +*/ +for (i =0 ; i <64 ; i +=16 ) { +tcg_target_long mask =0xffffull << i; +if ((value & mask) ==0 ) { + wantinv -=1 ; + } +if ((ivalue & mask) ==0 ) { + wantinv +=1 ; + imask |= mask; + } } -/* count trailing zeros in 16 bit steps, mapping 64 to 0. Emit the -first MOVZ with the half-word immediate skipping the zeros, with a shift -(LSL) equal to this number. Then all next instructions use MOVKs. -Zero the processed half-word in the value, continue until empty. -We build the final result 16bits at a time with up to 4 instructions, -but do not emit instructions for 16bit zero holes. */ +/* If we had more 0xffff than 0x0000, invert VALUE and use MOVN. */ insn =INSN_MOVZ ; - do { - unsigned shift = ctz64(value) & (63 & -16); - tcg_fmt_Rd_uimm(s, insn, shift >= 32, rd, value >> shift, shift); +if (wantinv >0 ) { + value = ivalue; + insn =INSN_MOVN ; + } + +/* Find the lowest lane that is not 0x0000. */ + shift =ctz64 (value) & (63 &-16 ); +tcg_fmt_Rd_uimm (s, insn, type, rd, value >> shift, shift); + +if (wantinv >0 ) { +/* Re-invert the value, so MOVK sees non-inverted bits. */ + value = ~value; +/* Clear out all the 0xffff lanes. */ + value ^= imask; + } +/* Clear out the lane that we just set. */ + value &= ~(0xffffUL << shift); + +/* Iterate until all lanes have been set, and thus cleared from VALUE. */
tcg/aarch64/tcg-target.c static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd, · line 644
Quoted 10 lines
tcg/aarch64/tcg-target.c +4
-2+while (value) { + shift =ctz64 (value) & (63 &-16 ); +tcg_fmt_Rd_uimm (s,INSN_MOVK , type, rd, value >> shift, shift); value &= ~(0xffffUL << shift); - insn =INSN_MOVK ; - }while (value); + } } static inline void tcg_out_ldst_r(TCGContext *s,
I agree in general with the approach "lets see if it is more convenient to start with MOVN". The existing implementation is, although not easy, leaner.
Can we make it a little this one a little bit leaner?
I'll think myself about it as well.
C.
Re: [Qemu-devel] [PATCH v4 22/33] tcg-aarch64: Use MOVN in tcg_out_movi
Quoted 3 lines · Claudio Fontana
I agree in general with the approach "lets see if it is more convenient to start with MOVN". The existing implementation is, although not easy, leaner.
Can we make it a little this one a little bit leaner?
This sentence is not well formed. What did you mean?
In what sense did you mean "lean"? Smaller or faster? If faster, see the comment about using neon insns. If smaller... um... why?
r~
Re: [Qemu-devel] [PATCH v4 22/33] tcg-aarch64: Use MOVN in tcg_out_movi
Quoted 10 lines · Richard Henderson
Quoted 3 lines · Claudio Fontana
I agree in general with the approach "lets see if it is more convenient to start with MOVN". The existing implementation is, although not easy, leaner.
Can we make it a little this one a little bit leaner?This sentence is not well formed. What did you mean?
In what sense did you mean "lean"? Smaller or faster? If faster, see the comment about using neon insns. If smaller... um... why?
I am not suggesting implementing the neon insns based thing.
I am suggesting looking at ways to reduce the size and complexity of the code needed to implement the same thing you just posted. If you don't see the why, there is probably little I can say to change that.
Re: [Qemu-devel] [PATCH v4 22/33] tcg-aarch64: Use MOVN in tcg_out_movi
Quoted 8 lines · Claudio Fontana
Quoted 7 lines · Richard Henderson
Quoted 3 lines · Claudio Fontana
I agree in general with the approach "lets see if it is more convenient to start with MOVN". The existing implementation is, although not easy, leaner.
Can we make it a little this one a little bit leaner?This sentence is not well formed. What did you mean?
Quoted 7 lines
Quoted 3 lines
In what sense did you mean "lean"? Smaller or faster? If faster, see the comment about using neon insns. If smaller... um... why?
I am not suggesting implementing the neon insns based thing.
I am suggesting looking at ways to reduce the size and complexity of the code needed to implement the same thing you just posted. If you don't see the why, there is probably little I can say to change that.
I interpreted you meaning akin to -Os, producing a smaller binary.
I obviously wrote it in the least complex way I could think to do the job. If you can find a simpler way, feel free.
r~
PATCH v4 23/33 tcg-aarch64: Use ORRI in tcg_out_movi
In reply to Richard HendersonThe subset of logical immediates that we support is quite quick to test, and such constants are quite common to want to load.
--- tcg/aarch64/tcg-target.c | 9 +++++++++ 1 file changed, 9 insertions(+)
tcg/aarch64/tcg-target.c +9 -0
@@ -573,6 +573,15 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd, type =TCG_TYPE_I32 ; } +/* Check for bitfield immediates. For the benefit of 32-bit quantities, +use the sign-extended value. That lets us match rotated values such +as 0xff0000ff with the same 64-bit logic matching 0xffffffffff0000ff. +The truncation happens inside tcg_fmt_Rdn_limm. */ +if (is_limm (svalue)) { +tcg_fmt_Rdn_limm (s,INSN_ORRI , type, rd,TCG_REG_XZR , svalue); +return ; + } +/* Would it take fewer insns to begin with MOVN? For the value and its inverse, count the number of 16-bit lanes that are 0. For the benefit of 32-bit quantities, compare the zero-extended normal value vs the
Signature · 1 line
1.8.3.1
PATCH v4 24/33 tcg-aarch64: Special case small constants in tcg_out_movi
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 11 +++++++++++ 1 file changed, 11 insertions(+)
tcg/aarch64/tcg-target.c +11 -0
@@ -573,6 +573,17 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd, type =TCG_TYPE_I32 ; } +/* Speed things up by handling the common case of small positive +and negative values specially. */ +if ((value & ~0xffffull ) ==0 ) { +tcg_fmt_Rd_uimm (s,INSN_MOVZ , type, rd, value,0 ); +return ; + } +if ((~svalue & ~0xffffull ) ==0 ) { +tcg_fmt_Rd_uimm (s,INSN_MOVN , type, rd, ~svalue,0 ); +return ; + } +/* Check for bitfield immediates. For the benefit of 32-bit quantities, use the sign-extended value. That lets us match rotated values such as 0xff0000ff with the same 64-bit logic matching 0xffffffffff0000ff.
Signature · 1 line
1.8.3.1
PATCH v4 25/33 tcg-aarch64: Use adrp in tcg_out_movi
In reply to Richard HendersonLoading an qemu pointer as an immediate happens often. E.g.
- exit_tb $0x7fa8140013
+ exit_tb $0x7f81ee0013
...
- : d2800260 mov x0, #0x13 - : f2b50280 movk x0, #0xa814, lsl #16 - : f2c00fe0 movk x0, #0x7f, lsl #32 + : 90ff1000 adrp x0, 0x7f81ee0000 + : 91004c00 add x0, x0, #0x13
--- tcg/aarch64/tcg-target.c | 23 ++++++++++++++++++++++- 1 file changed, 22 insertions(+), 1 deletion(-)
tcg/aarch64/tcg-target.c +22 -1
@@ -323,6 +323,10 @@ typedef enum {/* Conditional select instructions */ INSN_CSEL =0x1a800000 ,INSN_CSINC =0x1a800400 , + +/* PC relative addressing instructions */ +INSN_ADR =0x10000000 , +INSN_ADRP =0x90000000 , } AArch64Insn; static inline enum aarch64_ldst_op_data
@@ -510,6 +514,12 @@ static inline void tcg_fmt_Rd_uimm(TCGContext *s, AArch64Insn insn,tcg_out32 (s, insn | sf <<31 | shift << (21 -4 ) | half <<5 | rd); } +static inline void tcg_fmt_Rd_disp21 (TCGContext *s,AArch64Insn insn, +TCGReg rd,tcg_target_long disp) +{ +tcg_out32 (s, insn | (disp &3 ) <<29 | (disp &0x1ffffc ) << (5 -2 ) | rd); +} + static inline void tcg_out_ldst_9(TCGContext *s, enum aarch64_ldst_op_data op_data, enum aarch64_ldst_op_type op_type,
@@ -561,7 +571,7 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,AArch64Insn insn;int i, wantinv, shift;tcg_target_long svalue = value; -tcg_target_long ivalue, imask; +tcg_target_long ivalue, imask, disp;/* For 32-bit values, discard potential garbage in value. For 64-bit values within [2**31, 2**32-1], we can create smaller sequences by
@@ -593,6 +603,17 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,return ; } +/* Look for host pointer values within 4G of the PC. This happens +often when loading pointers to QEMU's own data structures. */ + disp = (value >>12 ) - ((intptr_t )s->code_ptr >>12 ); +if (disp ==sextract64 (disp,0 ,21 )) { +tcg_fmt_Rd_disp21 (s,INSN_ADRP , rd, disp); +if (value &0xfff ) { +tcg_fmt_Rdn_aimm (s,INSN_ADDI , type, rd, rd, value &0xfff ); + } +return ; + } +/* Would it take fewer insns to begin with MOVN? For the value and its inverse, count the number of 16-bit lanes that are 0. For the benefit of 32-bit quantities, compare the zero-extended normal value vs the
Signature · 1 line
1.8.3.1
PATCH v4 26/33 tcg-aarch64: Avoid add with zero in tlb load
In reply to Richard HendersonSome guest env are small enough to reach the tlb with only a 12-bit addition.
--- tcg/aarch64/tcg-target.c | 24 ++++++++++++++++++------ 1 file changed, 18 insertions(+), 6 deletions(-)
tcg/aarch64/tcg-target.c +18 -6
@@ -1136,46 +1136,58 @@ static void add_qemu_ldst_label(TCGContext *s, int is_ld, int opc,slow path for the failure case, which will be patched later when finalizing the slow path. Generated code returns the host addend in X1, clobbers X0,X2,X3,TMP. */ -static void tcg_out_tlb_read (TCGContext *s,TCGReg addr_reg, -int s_bits,uint8_t **label_ptr,int mem_index,int is_read) +static void tcg_out_tlb_read (TCGContext *s,TCGReg addr_reg,int s_bits, +uint8_t **label_ptr,int mem_index,int is_read) {TCGReg base =TCG_AREG0 ;int tlb_offset = is_read ? offsetof(CPUArchState, tlb_table[mem_index][0].addr_read) : offsetof(CPUArchState, tlb_table[mem_index][0].addr_write); +/* Extract the TLB index from the address into X0. X0<CPU_TLB_BITS:0> = addr_reg<TARGET_PAGE_BITS+CPU_TLB_BITS:TARGET_PAGE_BITS> */ tcg_out_ubfm (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 , addr_reg,TARGET_PAGE_BITS ,TARGET_PAGE_BITS +CPU_TLB_BITS ); +/* Store the page mask part of the address and the low s_bits into X3. Later this allows checking for equality and alignment at the same time. X3 = addr_reg & (PAGE_MASK | ((1 << s_bits) - 1)) */ tcg_fmt_Rdn_limm (s,INSN_ANDI ,TARGET_LONG_BITS ==64 ,TCG_REG_X3 , addr_reg,TARGET_PAGE_MASK | ((1 << s_bits) -1 )); +/* Add any "high bits" from the tlb offset to the env address into X2, to take advantage of the LSL12 form of the ADDI instruction. X2 = env + (tlb_offset & 0xfff000) */ -tcg_fmt_Rdn_aimm (s,INSN_ADDI ,1 ,TCG_REG_X2 , base, tlb_offset &0xfff000 ); +if (tlb_offset &0xfff000 ) { +tcg_fmt_Rdn_aimm (s,INSN_ADDI ,1 ,TCG_REG_X2 , base, + tlb_offset &0xfff000 ); + base =TCG_REG_X2 ; + } +/* Merge the tlb index contribution into X2. -X2 = X2 + (X0 << CPU_TLB_ENTRY_BITS) */ -tcg_fmt_Rdnm_lsl (s,INSN_ADD ,1 ,TCG_REG_X2 ,TCG_REG_X2 , +X2 = base + (X0 << CPU_TLB_ENTRY_BITS) */ +tcg_fmt_Rdnm_lsl (s,INSN_ADD ,1 ,TCG_REG_X2 , base,TCG_REG_X0 ,CPU_TLB_ENTRY_BITS ); +/* Merge "low bits" from tlb offset, load the tlb comparator into X0. X0 = load [X2 + (tlb_offset & 0x000fff)] */ tcg_out_ldst (s,TARGET_LONG_BITS ==64 ?LDST_64 :LDST_32 ,LDST_LD ,TCG_REG_X0 ,TCG_REG_X2 , (tlb_offset &0xfff )); +/* Load the tlb addend. Do that early to avoid stalling. X1 = load [X2 + (tlb_offset & 0xfff) + offsetof(addend)] */ tcg_out_ldst (s,LDST_64 ,LDST_LD ,TCG_REG_X1 ,TCG_REG_X2 , (tlb_offset &0xfff ) + (offsetof(CPUTLBEntry , addend)) - (is_read ? offsetof(CPUTLBEntry , addr_read) : offsetof(CPUTLBEntry , addr_write))); +/* Perform the address comparison. */ tcg_out_cmp (s, (TARGET_LONG_BITS ==64 ),TCG_REG_X0 ,TCG_REG_X3 ,0 ); - *label_ptr = s->code_ptr; +/* If not equal, we jump to the slow path. */ + *label_ptr = s->code_ptr;tcg_out_goto_cond_noaddr (s,TCG_COND_NE ); }
Signature · 1 line
1.8.3.1
PATCH v4 27/33 tcg-aarch64: Pass return address to load/store helpers directly.
In reply to Richard Henderson--- include/exec/exec-all.h | 11 ----------- tcg/aarch64/tcg-target.c | 44 ++++++++++++++++++++++++++------------------ 2 files changed, 26 insertions(+), 29 deletions(-)
include/exec/exec-all.h +0 -11
@@ -340,17 +340,6 @@ static inline uintptr_t tcg_getra_ldst(uintptr_t ra) ra += b;/* apply the displacement */ return ra; } -# elif defined(__aarch64__) -# define GETRA_LDST (RA ) tcg_getra_ldst(RA) -static inline uintptr_t tcg_getra_ldst (uintptr_t ra) -{ -int32_t b; - ra +=4 ;/* skip one instruction */ - b = *(int32_t *)ra;/* load the branch insn */ - b = (b <<6 ) >> (6 -2 );/* extract the displacement */ - ra += b;/* apply the displacement */ -return ra; -}# endif #endif /* CONFIG_QEMU_LDST_OPTIMIZATION */
tcg/aarch64/tcg-target.c +26 -18
76 lines not on this page · raw message
Signature · 1 line
1.8.3.1
PATCH v4 28/33 tcg-aarch64: Use tcg_out_call for qemu_ld/st
In reply to Richard HendersonIn some cases, a direct branch will be in range.
--- tcg/aarch64/tcg-target.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)
tcg/aarch64/tcg-target.c +2 -6
@@ -1074,9 +1074,7 @@ static void tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)tcg_out_movi (s,TCG_TYPE_I32 ,TCG_REG_X2 , lb->mem_index);tcg_out_adr (s,TCG_REG_X3 , (uintptr_t )lb->raddr); -tcg_out_movi (s,TCG_TYPE_I64 ,TCG_REG_TMP , - (tcg_target_long )qemu_ld_helpers[lb->opc &3 ]); -tcg_out_callr (s,TCG_REG_TMP ); +tcg_out_call (s, (tcg_target_long )qemu_ld_helpers[lb->opc &3 ]);if (lb->opc &0x04 ) {tcg_out_sxt (s,1 , lb->opc &3 , lb->datalo_reg,TCG_REG_X0 );
@@ -1097,9 +1095,7 @@ static void tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)tcg_out_movi (s,TCG_TYPE_I32 ,TCG_REG_X3 , lb->mem_index);tcg_out_adr (s,TCG_REG_X4 , (uintptr_t )lb->raddr); -tcg_out_movi (s,TCG_TYPE_I64 ,TCG_REG_TMP , - (tcg_target_long )qemu_st_helpers[lb->opc &3 ]); -tcg_out_callr (s,TCG_REG_TMP ); +tcg_out_call (s, (tcg_target_long )qemu_st_helpers[lb->opc &3 ]);tcg_out_goto (s, (tcg_target_long )lb->raddr); }
Signature · 1 line
1.8.3.1
PATCH v4 29/33 tcg-aarch64: Use symbolic names for branches
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 48 ++++++++++++++++++++++++------------------------ 1 file changed, 24 insertions(+), 24 deletions(-)
tcg/aarch64/tcg-target.c +24 -24
@@ -327,6 +327,14 @@ typedef enum {/* PC relative addressing instructions */ INSN_ADR =0x10000000 ,INSN_ADRP =0x90000000 , + +/* Branch instructions */ +INSN_B =0x14000000 , +INSN_BL =0x94000000 , +INSN_BR =0xd61f0000 , +INSN_BLR =0xd63f0000 , +INSN_RET =0xd65f0000 , +INSN_B_C =0x54000000 , } AArch64Insn; static inline enum aarch64_ldst_op_data
@@ -837,15 +845,14 @@ static void tcg_out_cmp(TCGContext *s, TCGType ext, TCGReg a,static inline void tcg_out_goto (TCGContext *s,tcg_target_long target) { -tcg_target_long offset; - offset = (target - (tcg_target_long )s->code_ptr) /4 ; +tcg_target_long offset = (target - (tcg_target_long )s->code_ptr) /4 ;if (offset <-0x02000000 || offset >=0x02000000 ) {/* out of 26bit range */ tcg_abort (); } -tcg_out32 (s,0x14000000 | (offset &0x03ffffff )); +tcg_out32 (s,INSN_B | (offset &0x03ffffff )); }static inline void tcg_out_goto_noaddr (TCGContext *s)
@@ -855,25 +862,21 @@ static inline void tcg_out_goto_noaddr(TCGContext *s)kept coherent during retranslation. Mask away possible garbage in the high bits for the first translation, while keeping the offset bits for retranslation. */ -uint32_t insn; - insn = (tcg_in32 (s) &0x03ffffff ) |0x14000000 ; -tcg_out32 (s, insn); +uint32_t old =tcg_in32 (s) &0x03ffffff ; +tcg_out32 (s,INSN_B | old); }static inline void tcg_out_goto_cond_noaddr (TCGContext *s,TCGCond c) { -/* see comments in tcg_out_goto_noaddr */ -uint32_t insn; - insn =tcg_in32 (s) & (0x07ffff <<5 ); - insn |=0x54000000 | tcg_cond_to_aarch64[c]; -tcg_out32 (s, insn); +/* See comments in tcg_out_goto_noaddr. */ +uint32_t old =tcg_in32 (s) & (0x07ffff <<5 ); +tcg_out32 (s,INSN_B_C | tcg_cond_to_aarch64[c] | old); }static inline void tcg_out_goto_cond (TCGContext *s,TCGCond c,tcg_target_long target) { -tcg_target_long offset; - offset = (target - (tcg_target_long )s->code_ptr) /4 ; +tcg_target_long offset = (target - (tcg_target_long )s->code_ptr) /4 ;if (offset <-0x40000 || offset >=0x40000 ) {/* out of 19bit range */
@@ -881,37 +884,34 @@ static inline void tcg_out_goto_cond(TCGContext *s, TCGCond c, } offset &=0x7ffff ; -tcg_out32 (s,0x54000000 | tcg_cond_to_aarch64[c] | offset <<5 ); +tcg_out32 (s,INSN_B_C | tcg_cond_to_aarch64[c] | offset <<5 ); }static inline void tcg_out_callr (TCGContext *s,TCGReg reg) { -tcg_out32 (s,0xd63f0000 | reg <<5 ); +tcg_out32 (s,INSN_BLR | reg <<5 ); }static inline void tcg_out_gotor (TCGContext *s,TCGReg reg) { -tcg_out32 (s,0xd61f0000 | reg <<5 ); +tcg_out32 (s,INSN_BR | reg <<5 ); }static inline void tcg_out_call (TCGContext *s,tcg_target_long target) { -tcg_target_long offset; - - offset = (target - (tcg_target_long )s->code_ptr) /4 ; +tcg_target_long offset = (target - (tcg_target_long )s->code_ptr) /4 ;if (offset <-0x02000000 || offset >=0x02000000 ) {/* out of 26bit rng */ tcg_out_movi (s,TCG_TYPE_I64 ,TCG_REG_TMP , target);tcg_out_callr (s,TCG_REG_TMP ); }else { -tcg_out32 (s,0x94000000 | (offset &0x03ffffff )); +tcg_out32 (s,INSN_BL | (offset &0x03ffffff )); } } -static inline void tcg_out_ret (TCGContext *s) +static inline void tcg_out_ret (TCGContext *s,TCGReg rn) { -/* emit RET { LR } */ -tcg_out32 (s,0xd65f03c0 ); +tcg_out32 (s,INSN_RET | rn <<5 ); }void aarch64_tb_set_jmp_target (uintptr_t jmp_addr,uintptr_t addr)
@@ -1993,5 +1993,5 @@ static void tcg_target_qemu_prologue(TCGContext *s)/* pop (FP, LR), restore SP to previous frame, return */ tcg_out_pop_pair (s,TCG_REG_SP ,TCG_REG_FP ,TCG_REG_LR , frame_size_callee_saved); -tcg_out_ret (s); +tcg_out_ret (s,TCG_REG_LR ); }
Signature · 1 line
1.8.3.1
PATCH v4 30/33 tcg-aarch64: Implement tcg_register_jit
In reply to Richard Henderson--- tcg/aarch64/tcg-target.c | 93 ++++++++++++++++++++++++++++++++++++------------ 1 file changed, 70 insertions(+), 23 deletions(-)
tcg/aarch64/tcg-target.c +70 -23
@@ -1930,30 +1930,27 @@ static void tcg_target_init(TCGContext *s)tcg_add_target_add_op_defs (aarch64_op_defs); } +/* Saving pairs: (X19, X20) .. (X27, X28), (X29(fp), X30(lr)). */ +#define PUSH_SIZE ((30 - 19 + 1) * 8) + +#define FRAME_SIZE \ + ((PUSH_SIZE \ + + TCG_STATIC_CALL_ARGS_SIZE \ + + CPU_TEMP_BUF_NLONGS * sizeof(long) \ + + TCG_TARGET_STACK_ALIGN - 1) \ + & ~(TCG_TARGET_STACK_ALIGN - 1)) +static void tcg_target_qemu_prologue (TCGContext *s) { -/* NB: frame sizes are in 16 byte stack units! */ -int frame_size_callee_saved, frame_size_tcg_locals;TCGReg r; -/* save pairs (FP, LR) and (X19, X20) .. (X27, X28) */ - frame_size_callee_saved = (1 ) + (TCG_REG_X28 -TCG_REG_X19 ) /2 +1 ; - -/* frame size requirement for TCG local variables */ - frame_size_tcg_locals =TCG_STATIC_CALL_ARGS_SIZE - +CPU_TEMP_BUF_NLONGS *sizeof (long ) - + (TCG_TARGET_STACK_ALIGN -1 ); - frame_size_tcg_locals &= ~(TCG_TARGET_STACK_ALIGN -1 ); - frame_size_tcg_locals /=TCG_TARGET_STACK_ALIGN ; - -/* push (FP, LR) and update sp */ -tcg_out_push_pair (s,TCG_REG_SP , -TCG_REG_FP ,TCG_REG_LR , frame_size_callee_saved); +/* Push (FP, LR) and allocate space for all saved registers. */ +tcg_out_push_pair (s,TCG_REG_SP ,TCG_REG_FP ,TCG_REG_LR ,PUSH_SIZE /16 );/* FP -> callee_saved */ tcg_out_movr_sp (s,1 ,TCG_REG_FP ,TCG_REG_SP ); -/* store callee-preserved regs x19..x28 using FP -> callee_saved */ +/* Store callee-preserved regs x19..x28. */ for (r =TCG_REG_X19 ; r <=TCG_REG_X27 ; r +=2 ) {int idx = (r -TCG_REG_X19 ) /2 +1 ;tcg_out_store_pair (s,TCG_REG_FP , r, r +1 , idx);
@@ -1961,7 +1958,7 @@ static void tcg_target_qemu_prologue(TCGContext *s)/* Make stack space for TCG locals. */ tcg_fmt_Rdn_aimm (s,INSN_SUBI ,1 ,TCG_REG_SP ,TCG_REG_SP , - frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN ); +FRAME_SIZE -PUSH_SIZE );/* inform TCG about how to find TCG locals with register, offset, size */ tcg_set_frame(s, TCG_REG_SP, TCG_STATIC_CALL_ARGS_SIZE,
@@ -1981,17 +1978,67 @@ static void tcg_target_qemu_prologue(TCGContext *s)/* Remove TCG locals stack space. */ tcg_fmt_Rdn_aimm (s,INSN_ADDI ,1 ,TCG_REG_SP ,TCG_REG_SP , - frame_size_tcg_locals *TCG_TARGET_STACK_ALIGN ); +FRAME_SIZE -PUSH_SIZE ); -/* restore registers x19..x28. -FP must be preserved, so it still points to callee_saved area */ +/* Restore callee-preserved registers x19..x28. */ for (r =TCG_REG_X19 ; r <=TCG_REG_X27 ; r +=2 ) {int idx = (r -TCG_REG_X19 ) /2 +1 ;tcg_out_load_pair (s,TCG_REG_FP , r, r +1 , idx); } -/* pop (FP, LR), restore SP to previous frame, return */ -tcg_out_pop_pair (s,TCG_REG_SP , -TCG_REG_FP ,TCG_REG_LR , frame_size_callee_saved); +/* Pop (FP, LR), restore SP to previous frame, return. */ +tcg_out_pop_pair (s,TCG_REG_SP ,TCG_REG_FP ,TCG_REG_LR ,PUSH_SIZE /16 );tcg_out_ret (s,TCG_REG_LR ); } + +typedef struct { +DebugFrameCIE cie; +DebugFrameFDEHeader fde; +uint8_t fde_def_cfa[4 ]; +uint8_t fde_reg_ofs[24 ]; +}DebugFrame ; + +/* We're expecting a 2 byte uleb128 encoded value. */ +QEMU_BUILD_BUG_ON (FRAME_SIZE >= (1 <<14 )); + +#define ELF_HOST_MACHINE EM_AARCH64 + +static DebugFrame debug_frame = { + .cie.len =sizeof (DebugFrameCIE)-4 ,/* length after .len member */ + .cie.id =-1 , + .cie.version =1 , + .cie.code_align =1 , + .cie.data_align =0x78 ,/* sleb128 -8 */ + .cie.return_column =16 , + +/* Total FDE size does not include the "len" member. */ + .fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, fde.cie_offset), + + .fde_def_cfa = { +12 ,7 ,/* DW_CFA_def_cfa %rsp, ... */ + (FRAME_SIZE &0x7f ) |0x80 ,/* ... uleb128 FRAME_SIZE */ + (FRAME_SIZE >>7 ) + }, + .fde_reg_ofs = { +0x80 +30 ,1 ,/* DW_CFA_offset, lr, -8 */ +0x80 +29 ,2 ,/* DW_CFA_offset, fp, -16 */ +0x80 +28 ,3 ,/* DW_CFA_offset, x28, -24 */ +0x80 +27 ,4 ,/* DW_CFA_offset, x27, -32 */ +0x80 +26 ,5 ,/* DW_CFA_offset, x26, -40 */ +0x80 +25 ,6 ,/* DW_CFA_offset, x25, -48 */ +0x80 +24 ,7 ,/* DW_CFA_offset, x24, -56 */ +0x80 +23 ,8 ,/* DW_CFA_offset, x23, -64 */ +0x80 +22 ,9 ,/* DW_CFA_offset, x22, -72 */ +0x80 +21 ,10 ,/* DW_CFA_offset, x21, -80 */ +0x80 +20 ,11 ,/* DW_CFA_offset, x20, -88 */ +0x80 +19 ,12 ,/* DW_CFA_offset, x1p, -96 */ + } +}; + +void tcg_register_jit (void *buf,size_t buf_size) +{ + debug_frame.fde.func_start = (tcg_target_long ) buf; + debug_frame.fde.func_len = buf_size; + +tcg_register_jit_int (buf, buf_size, &debug_frame,sizeof (debug_frame)); +}
Signature · 1 line
1.8.3.1
PATCH v4 31/33 tcg-aarch64: Reuse FP and LR in translated code
In reply to Richard HendersonWe don't need the FP within translated code, and the LR is otherwise unused.
--- tcg/aarch64/tcg-target.c | 36 +++++++++++++++--------------------- tcg/aarch64/tcg-target.h | 32 +++++++++++++++++--------------- 2 files changed, 32 insertions(+), 36 deletions(-)
tcg/aarch64/tcg-target.c +15 -21
@@ -22,10 +22,7 @@ static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {"%x0" ,"%x1" ,"%x2" ,"%x3" ,"%x4" ,"%x5" ,"%x6" ,"%x7" ,"%x8" ,"%x9" ,"%x10" ,"%x11" ,"%x12" ,"%x13" ,"%x14" ,"%x15" ,"%x16" ,"%x17" ,"%x18" ,"%x19" ,"%x20" ,"%x21" ,"%x22" ,"%x23" , -"%x24" ,"%x25" ,"%x26" ,"%x27" ,"%x28" , -"%fp" ,/* frame pointer */ -"%lr" ,/* link register */ - "%sp", /* stack pointer */ + "%x24", "%x25", "%x26", "%x27", "%x28", "%x29", "%x30", "%sp", };#endif /* NDEBUG */
@@ -38,18 +35,19 @@ static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {static const int tcg_target_reg_alloc_order[] = {TCG_REG_X20 ,TCG_REG_X21 ,TCG_REG_X22 ,TCG_REG_X23 ,TCG_REG_X24 ,TCG_REG_X25 ,TCG_REG_X26 ,TCG_REG_X27 , -TCG_REG_X28 ,/* we will reserve this for GUEST_BASE if configured */ +TCG_REG_X28 , +TCG_REG_X29 ,/* maybe used for TCG_REG_GUEST_BASE */ -TCG_REG_X9 ,TCG_REG_X10 ,TCG_REG_X11 ,TCG_REG_X12 , -TCG_REG_X13 ,TCG_REG_X14 ,TCG_REG_X15 , +TCG_REG_X8 ,TCG_REG_X9 ,TCG_REG_X10 ,TCG_REG_X11 , +TCG_REG_X12 ,TCG_REG_X13 ,TCG_REG_X14 ,TCG_REG_X15 ,TCG_REG_X16 ,TCG_REG_X17 , -TCG_REG_X18 ,TCG_REG_X19 ,/* will not use these, see tcg_target_init */ -TCG_REG_X0 ,TCG_REG_X1 ,TCG_REG_X2 ,TCG_REG_X3 ,TCG_REG_X4 ,TCG_REG_X5 ,TCG_REG_X6 ,TCG_REG_X7 , -TCG_REG_X8 ,/* will not use, see tcg_target_init */ +/* X18 reserved by system */ +/* X19 reserved for AREG0 */ +/* X30 reserved as temporary */ };static const int tcg_target_call_iarg_regs[8 ] = {
@@ -60,13 +58,13 @@ static const int tcg_target_call_oarg_regs[1] = { TCG_REG_X0 }; -#define TCG_REG_TMP TCG_REG_X8 +#define TCG_REG_TMP TCG_REG_X30#ifndef CONFIG_SOFTMMU -# if defined(CONFIG_USE_GUEST_BASE ) -# define TCG_REG_GUEST_BASE TCG_REG_X28 +# ifdef CONFIG_USE_GUEST_BASE +# define TCG_REG_GUEST_BASE TCG_REG_X29# else -# define TCG_REG_GUEST_BASE TCG_REG_XZR +# define TCG_REG_GUEST_BASE TCG_REG_XZR# endif #endif
@@ -1919,11 +1917,10 @@ static void tcg_target_init(TCGContext *s) (1 <<TCG_REG_X12 ) | (1 <<TCG_REG_X13 ) | (1 <<TCG_REG_X14 ) | (1 <<TCG_REG_X15 ) | (1 <<TCG_REG_X16 ) | (1 <<TCG_REG_X17 ) | - (1 << TCG_REG_X18)); + (1 << TCG_REG_X18) | (1 << TCG_REG_X30));tcg_regset_clear (s->reserved_regs);tcg_regset_set_reg (s->reserved_regs,TCG_REG_SP ); -tcg_regset_set_reg (s->reserved_regs,TCG_REG_FP );tcg_regset_set_reg (s->reserved_regs,TCG_REG_TMP );tcg_regset_set_reg (s->reserved_regs,TCG_REG_X18 );/* platform register */
@@ -1947,13 +1944,10 @@ static void tcg_target_qemu_prologue(TCGContext *s)/* Push (FP, LR) and allocate space for all saved registers. */ tcg_out_push_pair (s,TCG_REG_SP ,TCG_REG_FP ,TCG_REG_LR ,PUSH_SIZE /16 ); -/* FP -> callee_saved */ -tcg_out_movr_sp (s,1 ,TCG_REG_FP ,TCG_REG_SP ); -/* Store callee-preserved regs x19..x28. */ for (r =TCG_REG_X19 ; r <=TCG_REG_X27 ; r +=2 ) {int idx = (r -TCG_REG_X19 ) /2 +1 ; -tcg_out_store_pair (s,TCG_REG_FP , r, r +1 , idx); +tcg_out_store_pair (s,TCG_REG_SP , r, r +1 , idx); }/* Make stack space for TCG locals. */
@@ -1983,7 +1977,7 @@ static void tcg_target_qemu_prologue(TCGContext *s)/* Restore callee-preserved registers x19..x28. */ for (r =TCG_REG_X19 ; r <=TCG_REG_X27 ; r +=2 ) {int idx = (r -TCG_REG_X19 ) /2 +1 ; -tcg_out_load_pair (s,TCG_REG_FP , r, r +1 , idx); +tcg_out_load_pair (s,TCG_REG_SP , r, r +1 , idx); }/* Pop (FP, LR), restore SP to previous frame, return. */
tcg/aarch64/tcg-target.h +17 -15
@@ -17,17 +17,23 @@#undef TCG_TARGET_STACK_GROWSUPtypedef enum { -TCG_REG_X0 ,TCG_REG_X1 ,TCG_REG_X2 ,TCG_REG_X3 ,TCG_REG_X4 , -TCG_REG_X5 ,TCG_REG_X6 ,TCG_REG_X7 ,TCG_REG_X8 ,TCG_REG_X9 , -TCG_REG_X10 ,TCG_REG_X11 ,TCG_REG_X12 ,TCG_REG_X13 ,TCG_REG_X14 , -TCG_REG_X15 ,TCG_REG_X16 ,TCG_REG_X17 ,TCG_REG_X18 ,TCG_REG_X19 , -TCG_REG_X20 ,TCG_REG_X21 ,TCG_REG_X22 ,TCG_REG_X23 ,TCG_REG_X24 , -TCG_REG_X25 ,TCG_REG_X26 ,TCG_REG_X27 ,TCG_REG_X28 , -TCG_REG_FP ,/* frame pointer */ -TCG_REG_LR ,/* link register */ -TCG_REG_SP ,/* stack pointer or zero register */ -TCG_REG_XZR =TCG_REG_SP /* same register number */ -/* program counter is not directly accessible! */ +TCG_REG_X0 ,TCG_REG_X1 ,TCG_REG_X2 ,TCG_REG_X3 , +TCG_REG_X4 ,TCG_REG_X5 ,TCG_REG_X6 ,TCG_REG_X7 , +TCG_REG_X8 ,TCG_REG_X9 ,TCG_REG_X10 ,TCG_REG_X11 , +TCG_REG_X12 ,TCG_REG_X13 ,TCG_REG_X14 ,TCG_REG_X15 , +TCG_REG_X16 ,TCG_REG_X17 ,TCG_REG_X18 ,TCG_REG_X19 , +TCG_REG_X20 ,TCG_REG_X21 ,TCG_REG_X22 ,TCG_REG_X23 , +TCG_REG_X24 ,TCG_REG_X25 ,TCG_REG_X26 ,TCG_REG_X27 , +TCG_REG_X28 ,TCG_REG_X29 ,TCG_REG_X30 , + +/* X31 is either the stack pointer or zero, depending on context. */ +TCG_REG_SP =31 , +TCG_REG_XZR =31 , + +/* Aliases. */ +TCG_REG_FP =TCG_REG_X29 , +TCG_REG_LR =TCG_REG_X30 , +TCG_AREG0 =TCG_REG_X19 , }TCGReg ;#define TCG_TARGET_NB_REGS 32
@@ -92,10 +98,6 @@ typedef enum {#define TCG_TARGET_HAS_muluh_i64 1#define TCG_TARGET_HAS_mulsh_i64 1 -enum { -TCG_AREG0 =TCG_REG_X19 , -}; -static inline void flush_icache_range (uintptr_t start,uintptr_t stop) {__builtin___clear_cache ((char *)start, (char *)stop);
Signature · 1 line
1.8.3.1
PATCH v4 32/33 tcg-aarch64: Introduce tcg_out_ldst_pair
In reply to Richard HendersonCombines 4 other inline functions and tidies the prologue.
--- tcg/aarch64/tcg-target.c | 84 ++++++++++++++++-------------------------------- 1 file changed, 27 insertions(+), 57 deletions(-)
tcg/aarch64/tcg-target.c +27 -57
@@ -333,6 +333,10 @@ typedef enum {INSN_BLR =0xd63f0000 ,INSN_RET =0xd65f0000 ,INSN_B_C =0x54000000 , + +/* Load/store instructions */ +INSN_LDP =0x28400000 , +INSN_STP =0x28000000 , } AArch64Insn; static inline enum aarch64_ldst_op_data
@@ -1331,56 +1335,6 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, int opc)static uint8_t *tb_ret_addr; -/* callee stack use example: -stp x29, x30, [sp,#-32]! -mov x29, sp -stp x1, x2, [sp,#16] -... -ldp x1, x2, [sp,#16] -ldp x29, x30, [sp],#32 -ret -*/ - -/* push r1 and r2, and alloc stack space for a total of -alloc_n elements (1 element=16 bytes, must be between 1 and 31. */ -static inline void tcg_out_push_pair (TCGContext *s,TCGReg addr, -TCGReg r1,TCGReg r2,int alloc_n) -{ -/* using indexed scaled simm7 STP 0x28800000 | (ext) | 0x01000000 (pre-idx) -| alloc_n * (-1) << 16 | r2 << 10 | addr << 5 | r1 */ -assert (alloc_n >0 && alloc_n <0x20 ); - alloc_n = (-alloc_n) &0x3f ; -tcg_out32 (s,0xa9800000 | alloc_n <<16 | r2 <<10 | addr <<5 | r1); -} - -/* dealloc stack space for a total of alloc_n elements and pop r1, r2. */ -static inline void tcg_out_pop_pair (TCGContext *s,TCGReg addr, -TCGReg r1,TCGReg r2,int alloc_n) -{ -/* using indexed scaled simm7 LDP 0x28c00000 | (ext) | nothing (post-idx) -| alloc_n << 16 | r2 << 10 | addr << 5 | r1 */ -assert (alloc_n >0 && alloc_n <0x20 ); -tcg_out32 (s,0xa8c00000 | alloc_n <<16 | r2 <<10 | addr <<5 | r1); -} - -static inline void tcg_out_store_pair (TCGContext *s,TCGReg addr, -TCGReg r1,TCGReg r2,int idx) -{ -/* using register pair offset simm7 STP 0x29000000 | (ext) -| idx << 16 | r2 << 10 | addr << 5 | r1 */ -assert (idx >0 && idx <0x20 ); -tcg_out32 (s,0xa9000000 | idx <<16 | r2 <<10 | addr <<5 | r1); -} - -static inline void tcg_out_load_pair (TCGContext *s,TCGReg addr, -TCGReg r1,TCGReg r2,int idx) -{ -/* using register pair offset simm7 LDP 0x29400000 | (ext) -| idx << 16 | r2 << 10 | addr << 5 | r1 */ -assert (idx >0 && idx <0x20 ); -tcg_out32 (s,0xa9400000 | idx <<16 | r2 <<10 | addr <<5 | r1); -} -static void tcg_out_op (TCGContext *s,TCGOpcode opc,const TCGArg args[TCG_MAX_OP_ARGS ],const int const_args[TCG_MAX_OP_ARGS ])
@@ -1937,17 +1891,32 @@ static void tcg_target_init(TCGContext *s) + TCG_TARGET_STACK_ALIGN - 1) \ & ~(TCG_TARGET_STACK_ALIGN - 1)) +static void tcg_out_ldst_pair (TCGContext *s,AArch64Insn insn, +TCGReg r1,TCGReg r2,TCGReg base, +tcg_target_long ofs,bool pre,bool w) +{ + insn |=1u <<31 ;/* ext */ + insn |= pre <<24 ; + insn |= w <<23 ; + +assert (ofs >=-0x200 && ofs <0x200 && (ofs &7 ) ==0 ); + insn |= (ofs & (0x7f <<3 )) << (15 -3 ); + +tcg_out32 (s, insn | r2 <<10 | base <<5 | r1); +} +static void tcg_target_qemu_prologue (TCGContext *s) {TCGReg r;/* Push (FP, LR) and allocate space for all saved registers. */ -tcg_out_push_pair (s,TCG_REG_SP ,TCG_REG_FP ,TCG_REG_LR ,PUSH_SIZE /16 ); +tcg_out_ldst_pair (s,INSN_STP ,TCG_REG_FP ,TCG_REG_LR , +TCG_REG_SP , -PUSH_SIZE ,1 ,1 );/* Store callee-preserved regs x19..x28. */ for (r =TCG_REG_X19 ; r <=TCG_REG_X27 ; r +=2 ) { -int idx = (r -TCG_REG_X19 ) /2 +1 ; -tcg_out_store_pair (s,TCG_REG_SP , r, r +1 , idx); +int ofs = (r -TCG_REG_X19 +2 ) *8 ; +tcg_out_ldst_pair (s,INSN_STP , r, r +1 ,TCG_REG_SP , ofs,1 ,0 ); }/* Make stack space for TCG locals. */
@@ -1976,12 +1945,13 @@ static void tcg_target_qemu_prologue(TCGContext *s)/* Restore callee-preserved registers x19..x28. */ for (r =TCG_REG_X19 ; r <=TCG_REG_X27 ; r +=2 ) { -int idx = (r -TCG_REG_X19 ) /2 +1 ; -tcg_out_load_pair (s,TCG_REG_SP , r, r +1 , idx); +int ofs = (r -TCG_REG_X19 +2 ) *8 ; +tcg_out_ldst_pair (s,INSN_LDP , r, r +1 ,TCG_REG_SP , ofs,1 ,0 ); } -/* Pop (FP, LR), restore SP to previous frame, return. */ -tcg_out_pop_pair (s,TCG_REG_SP ,TCG_REG_FP ,TCG_REG_LR ,PUSH_SIZE /16 ); +/* Pop (FP, LR), restore SP to previous frame. */ +tcg_out_ldst_pair (s,INSN_LDP ,TCG_REG_FP ,TCG_REG_LR , +TCG_REG_SP ,PUSH_SIZE ,0 ,1 );tcg_out_ret (s,TCG_REG_LR ); }
Signature · 1 line
1.8.3.1
PATCH v4 33/33 tcg-aarch64: Remove redundant CPU_TLB_ENTRY_BITS check
In reply to Richard HendersonRemoved from other targets in 56bbc2f967ce185fa1c5c39e1aeb5b68b26242e9.
--- tcg/aarch64/tcg-target.c | 6 ------ 1 file changed, 6 deletions(-)
tcg/aarch64/tcg-target.c +0 -6
@@ -1852,12 +1852,6 @@ static const TCGTargetOpDef aarch64_op_defs[] = {static void tcg_target_init (TCGContext *s) { -#if !defined(CONFIG_USER_ONLY ) -/* fail safe */ -if ((1ULL <<CPU_TLB_ENTRY_BITS ) !=sizeof (CPUTLBEntry)) { -tcg_abort (); - } -#endif tcg_regset_set32 (tcg_target_available_regs[TCG_TYPE_I32 ],0 ,0xffffffff );tcg_regset_set32 (tcg_target_available_regs[TCG_TYPE_I64 ],0 ,0xffffffff );
Signature · 1 line
1.8.3.1
Re: [Qemu-devel] [PATCH v4 33/33] tcg-aarch64: Remove redundant CPU_TLB_ENTRY_BITS check
Quoted 15 lines · Richard Henderson
Removed from other targets in
56bbc2f967ce185fa1c5c39e1aeb5b68b26242e9.Signed-off-by Richard Henderson <[email protected]>--- tcg/aarch64/tcg-target.c | 6 ------ 1 file changed, 6 deletions(-)tcg/aarch64/tcg-target.c +0
-0@@ -1852,12 +1852,6 @@ static const TCGTargetOpDef aarch64_op_defs[] = {static void tcg_target_init (TCGContext *s)
tcg/aarch64/tcg-target.c static const TCGTargetOpDef aarch64_op_defs[] = { · line 1857
Quoted 10 lines
tcg/aarch64/tcg-target.c +0
-6{ -#if !defined(CONFIG_USER_ONLY ) -/* fail safe */ -if ((1ULL <<CPU_TLB_ENTRY_BITS ) !=sizeof (CPUTLBEntry)) { -tcg_abort (); - } -#endif tcg_regset_set32 (tcg_target_available_regs[TCG_TYPE_I32 ],0 ,0xffffffff );tcg_regset_set32 (tcg_target_available_regs[TCG_TYPE_I64 ],0 ,0xffffffff );