Modernize tcg/ppc64
v3 cover Modernize tcg/ppc64
Changes v2-v3:
* Incorporated feedback from Aurelien's review.
Tree updated at
git://github.com/rth7680/qemu.git tcg-ppc64
r~
Richard Henderson (27):
disas: Disassemble all ppc insns for the host
tcg-ppc64: Use TCGReg everywhere
tcg-ppc64: Introduce and use tcg_out_rlw
tcg-ppc64: Introduce and use tcg_out_ext32u
tcg-ppc64: Introduce and use tcg_out_shli64
tcg-ppc64: Introduce and use tcg_out_shri64
tcg-ppc64: Cleanup tcg_out_movi
tcg-ppc64: Introduce and use TAI and SAI
tcg-ppc64: Rearrange integer constant constraints
tcg-ppc64: Improve constant add and sub ops.
tcg-ppc64: Tidy or and xor patterns.
tcg-ppc64: Improve and_i32 with constant
tcg-ppc64: Improve and_i64 with constant
tcg-ppc64: Streamline qemu_ld/st insn selection
tcg-ppc64: Implement rotates
tcg-ppc64: Implement bswap16 and bswap32
tcg-ppc64: Implement bswap64
tcg-ppc64: Implement compound logicals
tcg-ppc64: Handle constant inputs for some compound logicals
tcg-ppc64: Implement deposit
tcg-ppc64: Use I constraint for mul
tcg-ppc64: Use TCGType throughout compares
tcg-ppc64: Rewrite setcond
tcg-ppc64: Implement movcond
tcg-ppc64: Use getauxval for ISA detection
tcg-ppc64: Implement add2/sub2_i64
tcg-ppc64: Implement mulu2/muls2_i64
configure | 18 + disas.c | 1 + tcg/ppc64/tcg-target.c | 1395 +++++++++++++++++++++++++++++++++--------------- tcg/ppc64/tcg-target.h | 60 ++- 4 files changed, 1000 insertions(+), 474 deletions(-)
Signature · 1 line
1.8.1.4
PATCH v3 1/27 disas: Disassemble all ppc insns for the host
--- disas.c | 1 + 1 file changed, 1 insertion(+)
disas.c +1 -0
@@ -325,6 +325,7 @@ void disas(FILE *out, void *code, unsigned long size) s.info.mach = bfd_mach_x86_64; print_insn = print_insn_i386;#elif defined(_ARCH_PPC) + s.info.disassembler_options = (char *)"any" ; print_insn = print_insn_ppc;#elif defined(__alpha__) print_insn = print_insn_alpha;
Signature · 1 line
1.8.1.4
PATCH v3 2/27 tcg-ppc64: Use TCGReg everywhere
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 36 +++++++++++++++++++++--------------- 1 file changed, 21 insertions(+), 15 deletions(-)
tcg/ppc64/tcg-target.c +21 -15
@@ -431,19 +431,21 @@ static const uint32_t tcg_to_bc[] = { [TCG_COND_GTU] = BC | BI (7, CR_GT) | BO_COND_TRUE, }; -static void tcg_out_mov (TCGContext *s,TCGType type,TCGReg ret,TCGReg arg) +static inline void tcg_out_mov (TCGContext *s,TCGType type, +TCGReg ret,TCGReg arg) {tcg_out32 (s,OR |SAB (arg, ret, arg)); } -static void tcg_out_rld (TCGContext *s,int op,int ra,int rs,int sh,int mb) +static inline void tcg_out_rld (TCGContext *s,int op,TCGReg ra,TCGReg rs, +int sh,int mb) { sh =SH (sh &0x1f ) | (((sh >>5 ) &1 ) <<1 ); mb =MB64 ((mb >>5 ) | ((mb <<1 ) &0x3f ));tcg_out32 (s, op |RA (ra) |RS (rs) | sh | mb); } -static void tcg_out_movi32 (TCGContext *s,int ret,int32_t arg) +static void tcg_out_movi32 (TCGContext *s,TCGReg ret,int32_t arg) {if (arg == (int16_t ) arg)tcg_out32 (s,ADDI |RT (ret) |RA (0 ) | (arg &0xffff ));
@@ -522,8 +524,8 @@ static void tcg_out_call (TCGContext *s, tcg_target_long arg, int const_arg)#endif } -static void tcg_out_ldst (TCGContext *s,int ret,int addr, -int offset,int op1,int op2) +static void tcg_out_ldst (TCGContext *s,TCGReg ret,TCGReg addr, +int offset,int op1,int op2) {if (offset == (int16_t ) offset)tcg_out32 (s, op1 |RT (ret) |RA (addr) | (offset &0xffff ));
@@ -533,8 +535,8 @@ static void tcg_out_ldst (TCGContext *s, int ret, int addr, } } -static void tcg_out_ldsta (TCGContext *s,int ret,int addr, -int offset,int op1,int op2) +static void tcg_out_ldsta (TCGContext *s,TCGReg ret,TCGReg addr, +int offset,int op1,int op2) {if (offset == (int16_t ) (offset & ~3 ))tcg_out32 (s, op1 |RT (ret) |RA (addr) | (offset &0xffff ));
@@ -566,8 +568,8 @@ static const void * const qemu_st_helpers[4] = { helper_stq_mmu, }; -static void tcg_out_tlb_read (TCGContext *s,int r0,int r1,int r2, -int addr_reg,int s_bits,int offset) +static void tcg_out_tlb_read (TCGContext *s,TCGReg r0,TCGReg r1,TCGReg r2, +TCGReg addr_reg,int s_bits,int offset) {#if TARGET_LONG_BITS ==32 tcg_out_rld (s,RLDICL , addr_reg, addr_reg,0 ,32 );
@@ -616,9 +618,11 @@ static void tcg_out_tlb_read (TCGContext *s, int r0, int r1, int r2,static void tcg_out_qemu_ld (TCGContext *s,const TCGArg *args,int opc) { -int addr_reg, data_reg, r0, r1, rbase, bswap; +TCGReg addr_reg, data_reg, r0, r1, rbase; +int bswap;#ifdef CONFIG_SOFTMMU -int r2, mem_index, s_bits, ir; +TCGReg r2, ir; +int mem_index, s_bits;void *label1_ptr, *label2_ptr;#endif
@@ -766,9 +770,11 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc)static void tcg_out_qemu_st (TCGContext *s,const TCGArg *args,int opc) { -int addr_reg, r0, r1, rbase, data_reg, bswap; +TCGReg addr_reg, r0, r1, rbase, data_reg; +int bswap;#ifdef CONFIG_SOFTMMU -int r2, mem_index, ir; +TCGReg r2, ir; +int mem_index;void *label1_ptr, *label2_ptr;#endif
@@ -954,7 +960,7 @@ static void tcg_out_st (TCGContext *s, TCGType type, TCGReg arg, TCGReg arg1,tcg_out_ldsta (s, arg, arg1, arg2,STD ,STDX ); } -static void ppc_addi32 (TCGContext *s,int rt,int ra,tcg_target_long si) +static void ppc_addi32 (TCGContext *s,TCGReg rt,TCGReg ra,tcg_target_long si) {if (!si && rt == ra)return ;
@@ -968,7 +974,7 @@ static void ppc_addi32 (TCGContext *s, int rt, int ra, tcg_target_long si) } } -static void ppc_addi64 (TCGContext *s,int rt,int ra,tcg_target_long si) +static void ppc_addi64 (TCGContext *s,TCGReg rt,TCGReg ra,tcg_target_long si) {/* XXX: suboptimal */ if (si == (int16_t) si
Signature · 1 line
1.8.1.4
PATCH v3 3/27 tcg-ppc64: Introduce and use tcg_out_rlw
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 72 ++++++++++++++------------------------------------ 1 file changed, 20 insertions(+), 52 deletions(-)
tcg/ppc64/tcg-target.c +20 -52
@@ -445,6 +445,12 @@ static inline void tcg_out_rld(TCGContext *s, int op, TCGReg ra, TCGReg rs,tcg_out32 (s, op |RA (ra) |RS (rs) | sh | mb); } +static inline void tcg_out_rlw (TCGContext *s,int op,TCGReg ra,TCGReg rs, +int sh,int mb,int me) +{ +tcg_out32 (s, op |RA (ra) |RS (rs) |SH (sh) |MB (mb) |ME (me)); +} +static void tcg_out_movi32 (TCGContext *s,TCGReg ret,int32_t arg) {if (arg == (int16_t ) arg)
@@ -574,24 +580,14 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg r0, TCGReg r1, TCGReg r2,#if TARGET_LONG_BITS ==32 tcg_out_rld (s,RLDICL , addr_reg, addr_reg,0 ,32 ); -tcg_out32 (s, (RLWINM - |RA (r0) - |RS (addr_reg) - |SH (32 - (TARGET_PAGE_BITS -CPU_TLB_ENTRY_BITS )) - |MB (32 - (CPU_TLB_BITS +CPU_TLB_ENTRY_BITS )) - |ME (31 -CPU_TLB_ENTRY_BITS ) - ) - ); +tcg_out_rlw (s,RLWINM , r0, addr_reg, +32 - (TARGET_PAGE_BITS -CPU_TLB_ENTRY_BITS ), +32 - (CPU_TLB_BITS +CPU_TLB_ENTRY_BITS ), +31 -CPU_TLB_ENTRY_BITS );tcg_out32 (s,ADD |RT (r0) |RA (r0) |RB (TCG_AREG0 ));tcg_out32 (s, (LWZU |RT (r1) |RA (r0) | offset)); -tcg_out32 (s, (RLWINM - |RA (r2) - |RS (addr_reg) - |SH (0 ) - |MB ((32 - s_bits) &31 ) - |ME (31 -TARGET_PAGE_BITS ) - ) - ); +tcg_out_rlw (s,RLWINM , r2, addr_reg,0 , + (32 - s_bits) &31 ,31 -TARGET_PAGE_BITS );#else tcg_out_rld (s, RLDICL, r0, addr_reg, 64 - TARGET_PAGE_BITS,
@@ -1093,14 +1089,7 @@ static void tcg_out_setcond (TCGContext *s, TCGType type, TCGCond cond, }else {tcg_out32 (s,CNTLZW |RS (arg) |RA (0 )); -tcg_out32 (s, (RLWINM - |RA (arg0) - |RS (0 ) - |SH (27 ) - |MB (5 ) - |ME (31 ) - ) - ); +tcg_out_rlw (s,RLWINM , arg0,0 ,27 ,5 ,31 ); }break ;
@@ -1161,14 +1150,7 @@ static void tcg_out_setcond (TCGContext *s, TCGType type, TCGCond cond,tcg_out_cmp (s, cond, arg1, arg2, const_arg2,7 , type ==TCG_TYPE_I64 );if (crop)tcg_out32 (s, crop);tcg_out32 (s,MFCR |RT (0 )); -tcg_out32 (s, (RLWINM - |RA (arg0) - |RS (0 ) - |SH (sh) - |MB (31 ) - |ME (31 ) - ) - ); +tcg_out_rlw (s,RLWINM , arg0,0 , sh,31 ,31 );break ;default :
@@ -1407,31 +1389,17 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,case INDEX_op_shl_i32:if (const_args[2 ]) { -tcg_out32 (s, (RLWINM - |RA (args[0 ]) - |RS (args[1 ]) - |SH (args[2 ]) - |MB (0 ) - |ME (31 - args[2 ]) - ) - ); - } -else +tcg_out_rlw (s,RLWINM , args[0 ], args[1 ], args[2 ],0 ,31 - args[2 ]); + }else {tcg_out32 (s,SLW |SAB (args[1 ], args[0 ], args[2 ])); + }break ;case INDEX_op_shr_i32:if (const_args[2 ]) { -tcg_out32 (s, (RLWINM - |RA (args[0 ]) - |RS (args[1 ]) - |SH (32 - args[2 ]) - |MB (args[2 ]) - |ME (31 ) - ) - ); - } -else +tcg_out_rlw (s,RLWINM , args[0 ], args[1 ],32 - args[2 ], args[2 ],31 ); + }else {tcg_out32 (s,SRW |SAB (args[1 ], args[0 ], args[2 ])); + }break ;case INDEX_op_sar_i32:if (const_args[2 ])
Signature · 1 line
1.8.1.4
PATCH v3 4/27 tcg-ppc64: Introduce and use tcg_out_ext32u
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)
tcg/ppc64/tcg-target.c +10 -5
@@ -451,6 +451,11 @@ static inline void tcg_out_rlw(TCGContext *s, int op, TCGReg ra, TCGReg rs,tcg_out32 (s, op |RA (ra) |RS (rs) |SH (sh) |MB (mb) |ME (me)); } +static inline void tcg_out_ext32u (TCGContext *s,TCGReg dst,TCGReg src) +{ +tcg_out_rld (s,RLDICL , dst, src,0 ,32 ); +} +static void tcg_out_movi32 (TCGContext *s,TCGReg ret,int32_t arg) {if (arg == (int16_t ) arg)
@@ -484,7 +489,7 @@ static void tcg_out_movi (TCGContext *s, TCGType type,else {tcg_out_movi32 (s, ret, arg32);if (arg32 <0 ) -tcg_out_rld (s,RLDICL , ret, ret,0 ,32 ); +tcg_out_ext32u (s, ret, ret); } } }
@@ -578,7 +583,7 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg r0, TCGReg r1, TCGReg r2, TCGReg addr_reg, int s_bits, int offset) { #if TARGET_LONG_BITS == 32 - tcg_out_rld (s, RLDICL, addr_reg, addr_reg, 0, 32); +tcg_out_ext32u (s ,addr_reg ,addr_reg ); tcg_out_rlw(s, RLWINM, r0, addr_reg, 32 - (TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS),
@@ -691,7 +696,7 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc)#else /* !CONFIG_SOFTMMU */ #if TARGET_LONG_BITS ==32 -tcg_out_rld (s,RLDICL , addr_reg, addr_reg,0 ,32 ); +tcg_out_ext32u (s, addr_reg, addr_reg);#endif r0 = addr_reg; r1 =3 ;
@@ -824,7 +829,7 @@ static void tcg_out_qemu_st (TCGContext *s, const TCGArg *args, int opc)#else /* !CONFIG_SOFTMMU */ #if TARGET_LONG_BITS ==32 -tcg_out_rld (s,RLDICL , addr_reg, addr_reg,0 ,32 ); +tcg_out_ext32u (s, addr_reg, addr_reg);#endif r1 =3 ; r0 = addr_reg;
@@ -1531,7 +1536,7 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_ext32u_i64: -tcg_out_rld (s,RLDICL , args[0 ], args[1 ],0 ,32 ); +tcg_out_ext32u (s, args[0 ], args[1 ]);break ;case INDEX_op_setcond_i32:
Signature · 1 line
1.8.1.4
PATCH v3 5/27 tcg-ppc64: Introduce and use tcg_out_shli64
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-)
tcg/ppc64/tcg-target.c +8 -5
@@ -456,6 +456,11 @@ static inline void tcg_out_ext32u(TCGContext *s, TCGReg dst, TCGReg src)tcg_out_rld (s,RLDICL , dst, src,0 ,32 ); } +static inline void tcg_out_shli64 (TCGContext *s,TCGReg dst,TCGReg src,int c) +{ +tcg_out_rld (s,RLDICR , dst, src, c,63 - c); +} +static void tcg_out_movi32 (TCGContext *s,TCGReg ret,int32_t arg) {if (arg == (int16_t ) arg)
@@ -482,7 +487,7 @@ static void tcg_out_movi (TCGContext *s, TCGType type,uint16_t l16 = arg;tcg_out_movi32 (s, ret, arg >>32 ); -tcg_out_rld (s,RLDICR , ret, ret,32 ,31 ); +tcg_out_shli64 (s, ret, ret,32 );if (h16)tcg_out32 (s,ORIS |RS (ret) |RA (ret) | h16);if (l16)tcg_out32 (s,ORI |RS (ret) |RA (ret) | l16); }
@@ -597,9 +602,7 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg r0, TCGReg r1, TCGReg r2,tcg_out_rld (s,RLDICL , r0, addr_reg,64 -TARGET_PAGE_BITS ,64 -CPU_TLB_BITS ); -tcg_out_rld (s,RLDICR , r0, r0, -CPU_TLB_ENTRY_BITS , -63 -CPU_TLB_ENTRY_BITS ); +tcg_out_shli64 (s, r0, r0,CPU_TLB_ENTRY_BITS );tcg_out32 (s,ADD |TAB (r0, r0,TCG_AREG0 ));tcg_out32 (s,LD_ADDR |RT (r1) |RA (r0) | offset);
@@ -1446,7 +1449,7 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,case INDEX_op_shl_i64:if (const_args[2 ]) -tcg_out_rld (s,RLDICR , args[0 ], args[1 ], args[2 ],63 - args[2 ]); +tcg_out_shli64 (s, args[0 ], args[1 ], args[2 ]);else tcg_out32 (s,SLD |SAB (args[1 ], args[0 ], args[2 ]));break ;
Signature · 1 line
1.8.1.4
PATCH v3 6/27 tcg-ppc64: Introduce and use tcg_out_shri64
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)
tcg/ppc64/tcg-target.c +7 -2
@@ -461,6 +461,11 @@ static inline void tcg_out_shli64(TCGContext *s, TCGReg dst, TCGReg src, int c)tcg_out_rld (s,RLDICR , dst, src, c,63 - c); } +static inline void tcg_out_shri64 (TCGContext *s,TCGReg dst,TCGReg src,int c) +{ +tcg_out_rld (s,RLDICL , dst, src,64 - c, c); +} +static void tcg_out_movi32 (TCGContext *s,TCGReg ret,int32_t arg) {if (arg == (int16_t ) arg)
@@ -864,7 +869,7 @@ static void tcg_out_qemu_st (TCGContext *s, const TCGArg *args, int opc)if (bswap) {tcg_out32 (s,STWBRX |SAB (data_reg, rbase, r0));tcg_out32 (s,ADDI |RT (r1) |RA (r0) |4 ); -tcg_out_rld (s,RLDICL ,0 , data_reg,32 ,0 ); +tcg_out_shri64 (s,0 , data_reg,32 );tcg_out32 (s,STWBRX |SAB (0 , rbase, r1)); }else tcg_out32 (s,STDX |SAB (data_reg, rbase, r0));
@@ -1455,7 +1460,7 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_shr_i64:if (const_args[2 ]) -tcg_out_rld (s,RLDICL , args[0 ], args[1 ],64 - args[2 ], args[2 ]); +tcg_out_shri64 (s, args[0 ], args[1 ], args[2 ]);else tcg_out32 (s,SRD |SAB (args[1 ], args[0 ], args[2 ]));break ;
Signature · 1 line
1.8.1.4
PATCH v3 7/27 tcg-ppc64: Cleanup tcg_out_movi
In reply to Richard HendersonThe test for using movi32 was sub-optimal for TCG_TYPE_I32, comparing a signed 32-bit quantity against an unsigned 32-bit quantity.
When possible, use addi+oris for 32-bit unsigned constants. Otherwise, standardize on addi+oris+ori instead of addis+ori+rldicl.
--- tcg/ppc64/tcg-target.c | 36 ++++++++++++++++-------------------- 1 file changed, 16 insertions(+), 20 deletions(-)
tcg/ppc64/tcg-target.c +16 -20
@@ -477,29 +477,25 @@ static void tcg_out_movi32(TCGContext *s, TCGReg ret, int32_t arg) } } -static void tcg_out_movi (TCGContext *s,TCGType type, -TCGReg ret,tcg_target_long arg) +static void tcg_out_movi (TCGContext *s,TCGType type,TCGReg ret, +tcg_target_long arg) { -int32_t arg32 = arg; - arg = type ==TCG_TYPE_I32 ? arg &0xffffffff : arg; - -if (arg == arg32) { -tcg_out_movi32 (s, ret, arg32); - } -else { -if ((uint64_t ) arg >>32 ) { -uint16_t h16 = arg >>16 ; -uint16_t l16 = arg; - -tcg_out_movi32 (s, ret, arg >>32 ); +if (type ==TCG_TYPE_I32 || arg == (int32_t )arg) { +tcg_out_movi32 (s, ret, arg); + }else if (arg == (uint32_t )arg && !(arg &0x8000 )) { +tcg_out32 (s,ADDI |RT (ret) |RA (0 ) | (arg &0xffff )); +tcg_out32 (s,ORIS |RT (ret) |RA (ret) | ((arg >>16 ) &0xffff )); + }else { +int32_t high = arg >>32 ; +tcg_out_movi32 (s, ret, high); +if (high) {tcg_out_shli64 (s, ret, ret,32 ); -if (h16)tcg_out32 (s,ORIS |RS (ret) |RA (ret) | h16); -if (l16)tcg_out32 (s,ORI |RS (ret) |RA (ret) | l16); } -else { -tcg_out_movi32 (s, ret, arg32); -if (arg32 <0 ) -tcg_out_ext32u (s, ret, ret); +if (arg &0xffff0000 ) { +tcg_out32 (s,ORIS |RS (ret) |RA (ret) | ((arg >>16 ) &0xffff )); + } +if (arg &0xffff ) { +tcg_out32 (s,ORI |RS (ret) |RA (ret) | (arg &0xffff )); } } }
Signature · 1 line
1.8.1.4
PATCH v3 8/27 tcg-ppc64: Introduce and use TAI and SAI
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 138 +++++++++++++++++++++++-------------------------- 1 file changed, 66 insertions(+), 72 deletions(-)
tcg/ppc64/tcg-target.c +66 -72
@@ -398,8 +398,10 @@ static int tcg_target_const_match (tcg_target_long val,#define LK 1 -#define TAB (t,a,b) (RT(t) | RA(a) | RB(b)) -#define SAB (s,a,b) (RS(s) | RA(a) | RB(b)) +#define TAB (t, a, b) (RT(t) | RA(a) | RB(b)) +#define SAB (s, a, b) (RS(s) | RA(a) | RB(b)) +#define TAI (s, a, i) (RT(s) | RA(a) | ((i) & 0xffff)) +#define SAI (s, a, i) (RS(s) | RA(a) | ((i) & 0xffff))#define BF (n) ((n)<<23)#define BI (n, c) (((c)+((n)*4))<<16)
@@ -468,12 +470,13 @@ static inline void tcg_out_shri64(TCGContext *s, TCGReg dst, TCGReg src, int c)static void tcg_out_movi32 (TCGContext *s,TCGReg ret,int32_t arg) { -if (arg == (int16_t ) arg) -tcg_out32 (s,ADDI |RT (ret) |RA (0 ) | (arg &0xffff )); -else { -tcg_out32 (s,ADDIS |RT (ret) |RA (0 ) | ((arg >>16 ) &0xffff )); -if (arg &0xffff ) -tcg_out32 (s,ORI |RS (ret) |RA (ret) | (arg &0xffff )); +if (arg == (int16_t ) arg) { +tcg_out32 (s,ADDI |TAI (ret,0 , arg)); + }else { +tcg_out32 (s,ADDIS |TAI (ret,0 , arg >>16 )); +if (arg &0xffff ) { +tcg_out32 (s,ORI |SAI (ret, ret, arg)); + } } }
@@ -483,8 +486,8 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,if (type ==TCG_TYPE_I32 || arg == (int32_t )arg) {tcg_out_movi32 (s, ret, arg); }else if (arg == (uint32_t )arg && !(arg &0x8000 )) { -tcg_out32 (s,ADDI |RT (ret) |RA (0 ) | (arg &0xffff )); -tcg_out32 (s,ORIS |RT (ret) |RA (ret) | ((arg >>16 ) &0xffff )); +tcg_out32 (s,ADDI |TAI (ret,0 , arg)); +tcg_out32 (s,ORIS |SAI (ret, ret, arg >>16 )); }else {int32_t high = arg >>32 ;tcg_out_movi32 (s, ret, high);
@@ -492,10 +495,10 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret,tcg_out_shli64 (s, ret, ret,32 ); }if (arg &0xffff0000 ) { -tcg_out32 (s,ORIS |RS (ret) |RA (ret) | ((arg >>16 ) &0xffff )); +tcg_out32 (s,ORIS |SAI (ret, ret, arg >>16 )); }if (arg &0xffff ) { -tcg_out32 (s,ORI |RS (ret) |RA (ret) | (arg &0xffff )); +tcg_out32 (s,ORI |SAI (ret, ret, arg)); } } }
@@ -544,22 +547,22 @@ static void tcg_out_call (TCGContext *s, tcg_target_long arg, int const_arg)static void tcg_out_ldst (TCGContext *s,TCGReg ret,TCGReg addr,int offset,int op1,int op2) { -if (offset == (int16_t ) offset) -tcg_out32 (s, op1 |RT (ret) |RA (addr) | (offset &0xffff )); -else { -tcg_out_movi (s,TCG_TYPE_I64 ,0 , offset); -tcg_out32 (s, op2 |RT (ret) |RA (addr) |RB (0 )); +if (offset == (int16_t ) offset) { +tcg_out32 (s, op1 |TAI (ret, addr, offset)); + }else { +tcg_out_movi (s,TCG_TYPE_I64 ,0 , offset); +tcg_out32 (s, op2 |TAB (ret, addr,0 )); } }static void tcg_out_ldsta (TCGContext *s,TCGReg ret,TCGReg addr,int offset,int op1,int op2) { -if (offset == (int16_t ) (offset & ~3 )) -tcg_out32 (s, op1 |RT (ret) |RA (addr) | (offset &0xffff )); -else { -tcg_out_movi (s,TCG_TYPE_I64 ,0 , offset); -tcg_out32 (s, op2 |RT (ret) |RA (addr) |RB (0 )); +if (offset == (int16_t ) (offset & ~3 )) { +tcg_out32 (s, op1 |TAI (ret, addr, offset)); + }else { +tcg_out_movi (s,TCG_TYPE_I64 ,0 , offset); +tcg_out32 (s, op2 |TAB (ret, addr,0 )); } }
@@ -595,8 +598,8 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg r0, TCGReg r1, TCGReg r2,32 - (TARGET_PAGE_BITS -CPU_TLB_ENTRY_BITS ),32 - (CPU_TLB_BITS +CPU_TLB_ENTRY_BITS ), 31 - CPU_TLB_ENTRY_BITS); -tcg_out32 (s,ADD |RT (r0) |RA (r0) |RB (TCG_AREG0 )); -tcg_out32 (s, (LWZU |RT (r1) |RA (r0) | offset)); +tcg_out32 (s,ADD |TAB (r0, r0,TCG_AREG0 )); +tcg_out32 (s,LWZU |TAI (r1, r0, offset));tcg_out_rlw (s,RLWINM , r2, addr_reg,0 , (32 - s_bits) &31 ,31 -TARGET_PAGE_BITS );#else
@@ -605,8 +608,8 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg r0, TCGReg r1, TCGReg r2, 64 - CPU_TLB_BITS);tcg_out_shli64 (s, r0, r0,CPU_TLB_ENTRY_BITS ); -tcg_out32 (s,ADD |TAB (r0, r0,TCG_AREG0 )); -tcg_out32 (s,LD_ADDR |RT (r1) |RA (r0) | offset); +tcg_out32 (s,ADD |TAB (r0, r0,TCG_AREG0 )); +tcg_out32 (s,LD_ADDR |TAI (r1, r0, offset));if (!s_bits) {tcg_out_rld (s,RLDICR , r2, addr_reg,0 ,63 -TARGET_PAGE_BITS );
@@ -688,14 +691,11 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc)#endif /* r0 now contains &env->tlb_table[mem_index][index].addr_read */ -tcg_out32 (s, (LD - |RT (r0) - |RA (r0) - | (offsetof (CPUTLBEntry , addend) - - offsetof (CPUTLBEntry , addr_read)) - )); +tcg_out32 (s,LD |TAI (r0, r0, + offsetof(CPUTLBEntry , addend) + - offsetof(CPUTLBEntry , addr_read)));/* r0 = env->tlb_table[mem_index][index].addend */ -tcg_out32 (s,ADD |RT (r0) |RA (r0) |RB (addr_reg)); +tcg_out32 (s,ADD |TAB (r0, r0, addr_reg));/* r0 = env->tlb_table[mem_index][index].addend + addr */ #else /* !CONFIG_SOFTMMU */
@@ -750,7 +750,7 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc)case 3 :#ifdef CONFIG_USE_GUEST_BASE if (bswap) { -tcg_out32 (s,ADDI |RT (r1) |RA (r0) |4 ); +tcg_out32 (s,ADDI |TAI (r1, r0,4 ));tcg_out32 (s,LWBRX |TAB (data_reg, rbase, r0));tcg_out32 (s,LWBRX |TAB ( r1, rbase, r1));tcg_out_rld (s,RLDIMI , data_reg, r1,32 ,0 );
@@ -828,7 +828,7 @@ static void tcg_out_qemu_st (TCGContext *s, const TCGArg *args, int opc) - offsetof (CPUTLBEntry, addr_write)) ));/* r0 = env->tlb_table[mem_index][index].addend */ -tcg_out32 (s,ADD |RT (r0) |RA (r0) |RB (addr_reg)); +tcg_out32 (s,ADD |TAB (r0, r0, addr_reg));/* r0 = env->tlb_table[mem_index][index].addend + addr */ #else /* !CONFIG_SOFTMMU */
@@ -864,7 +864,7 @@ static void tcg_out_qemu_st (TCGContext *s, const TCGArg *args, int opc)case 3 :if (bswap) {tcg_out32 (s,STWBRX |SAB (data_reg, rbase, r0)); -tcg_out32 (s,ADDI |RT (r1) |RA (r0) |4 ); +tcg_out32 (s,ADDI |TAI (r1, r0,4 ));tcg_out_shri64 (s,0 , data_reg,32 );tcg_out32 (s,STWBRX |SAB (0 , rbase, r1)); }
@@ -941,10 +941,10 @@ static void tcg_target_qemu_prologue (TCGContext *s) | (i * 8 + 48 + TCG_STATIC_CALL_ARGS_SIZE) ) ); -tcg_out32 (s,LD |RT (0 ) |RA (1 ) | (frame_size +16 )); -tcg_out32 (s,MTSPR |RS (0 ) |LR ); -tcg_out32 (s,ADDI |RT (1 ) |RA (1 ) | frame_size); -tcg_out32 (s,BCLR |BO_ALWAYS ); +tcg_out32 (s,LD |TAI (0 ,1 , frame_size +16 )); +tcg_out32 (s,MTSPR |RS (0 ) |LR ); +tcg_out32 (s,ADDI |TAI (1 ,1 , frame_size)); +tcg_out32 (s,BCLR |BO_ALWAYS ); } static void tcg_out_ld (TCGContext *s, TCGType type, TCGReg ret, TCGReg arg1,
@@ -971,11 +971,11 @@ static void ppc_addi32(TCGContext *s, TCGReg rt, TCGReg ra, tcg_target_long si)return ;if (si == (int16_t ) si) -tcg_out32 (s,ADDI |RT (rt) |RA (ra) | (si &0xffff )); +tcg_out32 (s,ADDI |TAI (rt, ra, si));else {uint16_t h = ((si >>16 ) &0xffff ) + ((uint16_t ) si >>15 ); -tcg_out32 (s,ADDIS |RT (rt) |RA (ra) | h); -tcg_out32 (s,ADDI |RT (rt) |RA (rt) | (si &0xffff )); +tcg_out32 (s,ADDIS |TAI (rt, ra, h)); +tcg_out32 (s,ADDI |TAI (rt, rt, si)); } }
@@ -987,7 +987,7 @@ static void ppc_addi64(TCGContext *s, TCGReg rt, TCGReg ra, tcg_target_long si)ppc_addi32 (s, rt, ra, si); else {tcg_out_movi (s,TCG_TYPE_I64 ,0 , si); -tcg_out32 (s,ADD |RT (rt) |RA (ra)); +tcg_out32 (s,ADD |TAB (rt, ra,0 )); } }
@@ -1079,7 +1079,7 @@ static void tcg_out_setcond (TCGContext *s, TCGType type, TCGCond cond,else { arg =0 ;if ((uint16_t ) arg2 == arg2) { -tcg_out32 (s,XORI |RS (arg1) |RA (0 ) | arg2); +tcg_out32 (s,XORI |SAI (arg1,0 , arg2)); }else {tcg_out_movi (s, type,0 , arg2);
@@ -1110,9 +1110,8 @@ static void tcg_out_setcond (TCGContext *s, TCGType type, TCGCond cond,else { arg =0 ;if ((uint16_t ) arg2 == arg2) { -tcg_out32 (s,XORI |RS (arg1) |RA (0 ) | arg2); - } -else { +tcg_out32 (s,XORI |SAI (arg1,0 , arg2)); + }else {tcg_out_movi (s, type,0 , arg2);tcg_out32 (s,XOR |SAB (arg1,0 ,0 )); }
@@ -1124,12 +1123,12 @@ static void tcg_out_setcond (TCGContext *s, TCGType type, TCGCond cond, }if (arg == arg1 && arg1 == arg0) { -tcg_out32 (s,ADDIC |RT (0 ) |RA (arg) |0xffff ); -tcg_out32 (s,SUBFE |TAB (arg0,0 , arg)); +tcg_out32 (s,ADDIC |TAI (0 , arg,-1 )); +tcg_out32 (s,SUBFE |TAB (arg0,0 , arg)); }else { -tcg_out32 (s,ADDIC |RT (arg0) |RA (arg) |0xffff ); -tcg_out32 (s,SUBFE |TAB (arg0, arg0, arg)); +tcg_out32 (s,ADDIC |TAI (arg0, arg,-1 )); +tcg_out32 (s,SUBFE |TAB (arg0, arg0, arg)); }break ;
@@ -1307,12 +1306,11 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,case INDEX_op_and_i64:case INDEX_op_and_i32:if (const_args[2 ]) { -if ((args[2 ] &0xffff ) == args[2 ]) -tcg_out32 (s,ANDI |RS (args[1 ]) |RA (args[0 ]) | args[2 ]); -else if ((args[2 ] &0xffff0000 ) == args[2 ]) -tcg_out32 (s,ANDIS |RS (args[1 ]) |RA (args[0 ]) - | ((args[2 ] >>16 ) &0xffff )); -else { +if ((args[2 ] &0xffff ) == args[2 ]) { +tcg_out32 (s,ANDI |SAI (args[1 ], args[0 ], args[2 ])); + }else if ((args[2 ] &0xffff0000 ) == args[2 ]) { +tcg_out32 (s,ANDIS |SAI (args[1 ], args[0 ], args[2 ] >>16 )); + }else { tcg_out_movi (s, (opc == INDEX_op_and_i32 ? TCG_TYPE_I32 : TCG_TYPE_I64),
@@ -1327,15 +1325,13 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,case INDEX_op_or_i32:if (const_args[2 ]) {if (args[2 ] &0xffff ) { -tcg_out32 (s,ORI |RS (args[1 ]) |RA (args[0 ]) - | (args[2 ] &0xffff )); -if (args[2 ] >>16 ) -tcg_out32 (s,ORIS |RS (args[0 ]) |RA (args[0 ]) - | ((args[2 ] >>16 ) &0xffff )); +tcg_out32 (s,ORI |SAI (args[1 ], args[0 ], args[2 ])); +if (args[2 ] >>16 ) { +tcg_out32 (s,ORIS |SAI (args[0 ], args[0 ], args[2 ] >>16 )); + } }else { -tcg_out32 (s,ORIS |RS (args[1 ]) |RA (args[0 ]) - | ((args[2 ] >>16 ) &0xffff )); +tcg_out32 (s,ORIS |SAI (args[1 ], args[0 ], args[2 ] >>16 )); } }else
@@ -1344,13 +1340,11 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,case INDEX_op_xor_i64:case INDEX_op_xor_i32:if (const_args[2 ]) { -if ((args[2 ] &0xffff ) == args[2 ]) -tcg_out32 (s,XORI |RS (args[1 ]) |RA (args[0 ]) - | (args[2 ] &0xffff )); -else if ((args[2 ] &0xffff0000 ) == args[2 ]) -tcg_out32 (s,XORIS |RS (args[1 ]) |RA (args[0 ]) - | ((args[2 ] >>16 ) &0xffff )); -else { +if ((args[2 ] &0xffff ) == args[2 ]) { +tcg_out32 (s,XORI |SAI (args[1 ], args[0 ], args[2 ])); + }else if ((args[2 ] &0xffff0000 ) == args[2 ]) { +tcg_out32 (s,XORIS |SAI (args[1 ], args[0 ], args[2 ] >>16 )); + }else { tcg_out_movi (s, (opc == INDEX_op_and_i32 ? TCG_TYPE_I32 : TCG_TYPE_I64),
Signature · 1 line
1.8.1.4
PATCH v3 9/27 tcg-ppc64: Rearrange integer constant constraints
In reply to Richard HendersonWe'll need a zero, and Z makes more sense for that. Make sure we have a full compliment of signed and unsigned 16 and 32-bit tests.
--- tcg/ppc64/tcg-target.c | 43 +++++++++++++++++++++++++++++++++---------- 1 file changed, 33 insertions(+), 10 deletions(-)
tcg/ppc64/tcg-target.c +33 -10
@@ -22,7 +22,11 @@* THE SOFTWARE. */ -#define TCG_CT_CONST_U32 0x100 +#define TCG_CT_CONST_S16 0x100 +#define TCG_CT_CONST_U16 0x200 +#define TCG_CT_CONST_S32 0x400 +#define TCG_CT_CONST_U32 0x800 +#define TCG_CT_CONST_ZERO 0x1000static uint8_t *tb_ret_addr;
@@ -242,9 +246,21 @@ static int target_parse_constraint (TCGArgConstraint *ct, const char **pct_str)tcg_regset_reset_reg (ct->u.regs,TCG_REG_R6 );#endif break ; -case 'Z' : +case 'I' : + ct->ct |=TCG_CT_CONST_S16 ; +break ; +case 'J' : + ct->ct |=TCG_CT_CONST_U16 ; +break ; +case 'T' : + ct->ct |=TCG_CT_CONST_S32 ; +break ; +case 'U' : ct->ct |=TCG_CT_CONST_U32 ;break ; +case 'Z' : + ct->ct |=TCG_CT_CONST_ZERO ; +break ;default :return -1 ; }
@@ -257,13 +273,20 @@ static int target_parse_constraint (TCGArgConstraint *ct, const char **pct_str)static int tcg_target_const_match (tcg_target_long val,const TCGArgConstraint *arg_ct) { -int ct; - - ct = arg_ct->ct; -if (ct &TCG_CT_CONST ) +int ct = arg_ct->ct; +if (ct &TCG_CT_CONST ) { +return 1 ; + }else if ((ct &TCG_CT_CONST_S16 ) && val == (int16_t )val) { +return 1 ; + }else if ((ct &TCG_CT_CONST_U16 ) && val == (uint16_t )val) {return 1 ; -else if ((ct &TCG_CT_CONST_U32 ) && (val == (uint32_t ) val)) + }else if ((ct &TCG_CT_CONST_S32 ) && val == (int32_t )val) {return 1 ; + }else if ((ct &TCG_CT_CONST_U32 ) && val == (uint32_t )val) { +return 1 ; + }else if ((ct &TCG_CT_CONST_ZERO ) && val ==0 ) { +return 1 ; + }return 0 ; }
@@ -1607,9 +1630,9 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_add_i64, { "r", "r", "ri" } }, { INDEX_op_sub_i64, { "r", "r", "ri" } }, - { INDEX_op_and_i64, { "r", "r", "rZ" } }, - { INDEX_op_or_i64, { "r", "r", "rZ" } }, - { INDEX_op_xor_i64, { "r", "r", "rZ" } }, + { INDEX_op_and_i64, { "r", "r", "rU" } }, + { INDEX_op_or_i64, { "r", "r", "rU" } }, + { INDEX_op_xor_i64, { "r", "r", "rU" } }, { INDEX_op_shl_i64, { "r", "r", "ri" } }, { INDEX_op_shr_i64, { "r", "r", "ri" } },
Signature · 1 line
1.8.1.4
PATCH v3 10/27 tcg-ppc64: Improve constant add and sub ops.
In reply to Richard HendersonUse SUBFIC to implement subtraction with constant op1. Improve constant addition -- previously we'd emit useless addis with 0. Use new constraints to force the driver to pull full 64-bit constants into a register.
--- tcg/ppc64/tcg-target.c | 117 ++++++++++++++++++++++++++++++------------------- 1 file changed, 72 insertions(+), 45 deletions(-)
tcg/ppc64/tcg-target.c +72 -45
@@ -329,6 +329,7 @@ static int tcg_target_const_match (tcg_target_long val,#define MULLI OPCD( 7)#define CMPLI OPCD( 10)#define CMPI OPCD( 11) +#define SUBFIC OPCD( 8)#define LWZU OPCD( 33)#define STWU OPCD( 37)
@@ -988,32 +989,6 @@ static void tcg_out_st (TCGContext *s, TCGType type, TCGReg arg, TCGReg arg1,tcg_out_ldsta (s, arg, arg1, arg2,STD ,STDX ); } -static void ppc_addi32 (TCGContext *s,TCGReg rt,TCGReg ra,tcg_target_long si) -{ -if (!si && rt == ra) -return ; - -if (si == (int16_t ) si) -tcg_out32 (s,ADDI |TAI (rt, ra, si)); -else { -uint16_t h = ((si >>16 ) &0xffff ) + ((uint16_t ) si >>15 ); -tcg_out32 (s,ADDIS |TAI (rt, ra, h)); -tcg_out32 (s,ADDI |TAI (rt, rt, si)); - } -} - -static void ppc_addi64 (TCGContext *s,TCGReg rt,TCGReg ra,tcg_target_long si) -{ -/* XXX: suboptimal */ -if (si == (int16_t ) si - || ((((uint64_t ) si >>31 ) ==0 ) && (si &0x8000 ) ==0 )) -ppc_addi32 (s, rt, ra, si); -else { -tcg_out_movi (s,TCG_TYPE_I64 ,0 , si); -tcg_out32 (s,ADD |TAB (rt, ra,0 )); - } -} -static void tcg_out_cmp (TCGContext *s,int cond,TCGArg arg1,TCGArg arg2,int const_arg2,int cr,int arch64) {
@@ -1226,6 +1201,7 @@ void ppc_tb_set_jmp_target (unsigned long jmp_addr, unsigned long addr)static void tcg_out_op (TCGContext *s,TCGOpcode opc,const TCGArg *args,const int *const_args) { +TCGArg a0, a1, a2;int c;switch (opc) {
@@ -1314,16 +1290,37 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_add_i32: -if (const_args[2 ]) -ppc_addi32 (s, args[0 ], args[1 ], args[2 ]); -else -tcg_out32 (s,ADD |TAB (args[0 ], args[1 ], args[2 ])); + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ]; +if (const_args[2 ]) { +int32_t l, h; + do_addi_32: + l = (int16_t )a2; + h = a2 - l; +if (h) { +tcg_out32 (s,ADDIS |TAI (a0, a1, h >>16 )); + a1 = a0; + } +if (l || a0 != a1) { +tcg_out32 (s,ADDI |TAI (a0, a1, l)); + } + }else { +tcg_out32 (s,ADD |TAB (a0, a1, a2)); + }break ;case INDEX_op_sub_i32: -if (const_args[2 ]) -ppc_addi32 (s, args[0 ], args[1 ], -args[2 ]); -else -tcg_out32 (s,SUBF |TAB (args[0 ], args[2 ], args[1 ])); + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ]; +if (const_args[1 ]) { +if (const_args[2 ]) { +tcg_out_movi (s,TCG_TYPE_I32 , a0, a1 - a2); + }else { +tcg_out32 (s,SUBFIC |TAI (a0, a2, a1)); + } + }else if (const_args[2 ]) { + a2 = -a2; + goto do_addi_32; + }else { +tcg_out32 (s,SUBF |TAB (a0, a2, a1)); + }break ;case INDEX_op_and_i64:
@@ -1453,16 +1450,46 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_add_i64: -if (const_args[2 ]) -ppc_addi64 (s, args[0 ], args[1 ], args[2 ]); -else -tcg_out32 (s,ADD |TAB (args[0 ], args[1 ], args[2 ])); + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ]; +if (const_args[2 ]) { +int32_t l0, h1, h2; + do_addi_64: +/* We can always split any 32-bit constant into +3 sign-extending pieces. */ + l0 = (int16_t )a2; + a2 -= l0; + h1 = ((int64_t )a2 >0 && (int32_t )a2 <0 ?0x40000000 :0 ); + h2 = a2 - h1; + +if (h2) { +tcg_out32 (s,ADDIS |TAI (a0, a1, h2 >>16 )); + a1 = a0; + } +if (h1) { +tcg_out32 (s,ADDIS |TAI (a0, a1, h1 >>16 )); + a1 = a0; + } +if (l0 || a0 != a1) { +tcg_out32 (s,ADDI |TAI (a0, a1, l0)); + } + }else { +tcg_out32 (s,ADD |TAB (a0, a1, a2)); + }break ;case INDEX_op_sub_i64: -if (const_args[2 ]) -ppc_addi64 (s, args[0 ], args[1 ], -args[2 ]); -else -tcg_out32 (s,SUBF |TAB (args[0 ], args[2 ], args[1 ])); + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ]; +if (const_args[1 ]) { +if (const_args[2 ]) { +tcg_out_movi (s,TCG_TYPE_I64 , a0, a1 - a2); + }else { +tcg_out32 (s,SUBFIC |TAI (a0, a2, a1)); + } + }else if (const_args[2 ]) { + a2 = -a2; + goto do_addi_64; + }else { +tcg_out32 (s,SUBF |TAB (a0, a2, a1)); + }break ;case INDEX_op_shl_i64:
@@ -1613,7 +1640,7 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_divu_i32, { "r", "r", "r" } }, { INDEX_op_rem_i32, { "r", "r", "r" } }, { INDEX_op_remu_i32, { "r", "r", "r" } }, - { INDEX_op_sub_i32, { "r", "r", "ri" } }, + { INDEX_op_sub_i32, { "r", "rI", "ri" } }, { INDEX_op_and_i32, { "r", "r", "ri" } }, { INDEX_op_or_i32, { "r", "r", "ri" } }, { INDEX_op_xor_i32, { "r", "r", "ri" } },
@@ -1628,8 +1655,8 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_neg_i32, { "r", "r" } }, { INDEX_op_not_i32, { "r", "r" } }, - { INDEX_op_add_i64, { "r", "r", "ri" } }, - { INDEX_op_sub_i64, { "r", "r", "ri" } }, + { INDEX_op_add_i64, { "r", "r", "rTU" } }, + { INDEX_op_sub_i64, { "r", "rI", "rTU" } }, { INDEX_op_and_i64, { "r", "r", "rU" } }, { INDEX_op_or_i64, { "r", "r", "rU" } }, { INDEX_op_xor_i64, { "r", "r", "rU" } },
Signature · 1 line
1.8.1.4
PATCH v3 11/27 tcg-ppc64: Tidy or and xor patterns.
In reply to Richard HendersonHandle constants in common code; we'll want to reuse that later.
--- tcg/ppc64/tcg-target.c | 55 ++++++++++++++++++++++++++++---------------------- 1 file changed, 31 insertions(+), 24 deletions(-)
tcg/ppc64/tcg-target.c +31 -24
@@ -527,6 +527,29 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret, } } +static void tcg_out_zori32 (TCGContext *s,TCGReg dst,TCGReg src,uint32_t c, +int op_lo,int op_hi) +{ +if (c >>16 ) { +tcg_out32 (s, op_hi |SAI (src, dst, c >>16 )); + src = dst; + } +if (c &0xffff ) { +tcg_out32 (s, op_lo |SAI (src, dst, c)); + src = dst; + } +} + +static void tcg_out_ori32 (TCGContext *s,TCGReg dst,TCGReg src,uint32_t c) +{ +tcg_out_zori32 (s, dst, src, c,ORI ,ORIS ); +} + +static void tcg_out_xori32 (TCGContext *s,TCGReg dst,TCGReg src,uint32_t c) +{ +tcg_out_zori32 (s, dst, src, c,XORI ,XORIS ); +} +static void tcg_out_b (TCGContext *s,int mask,tcg_target_long target) {tcg_target_long disp;
@@ -1343,37 +1366,21 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_or_i64:case INDEX_op_or_i32: + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ];if (const_args[2 ]) { -if (args[2 ] &0xffff ) { -tcg_out32 (s,ORI |SAI (args[1 ], args[0 ], args[2 ])); -if (args[2 ] >>16 ) { -tcg_out32 (s,ORIS |SAI (args[0 ], args[0 ], args[2 ] >>16 )); - } - } -else { -tcg_out32 (s,ORIS |SAI (args[1 ], args[0 ], args[2 ] >>16 )); - } +tcg_out_ori32 (s, a0, a1, a2); + }else { +tcg_out32 (s,OR |SAB (a1, a0, a2)); } -else -tcg_out32 (s,OR |SAB (args[1 ], args[0 ], args[2 ]));break ;case INDEX_op_xor_i64:case INDEX_op_xor_i32: + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ];if (const_args[2 ]) { -if ((args[2 ] &0xffff ) == args[2 ]) { -tcg_out32 (s,XORI |SAI (args[1 ], args[0 ], args[2 ])); - }else if ((args[2 ] &0xffff0000 ) == args[2 ]) { -tcg_out32 (s,XORIS |SAI (args[1 ], args[0 ], args[2 ] >>16 )); - }else { -tcg_out_movi (s, (opc == INDEX_op_and_i32 - ?TCG_TYPE_I32 - :TCG_TYPE_I64 ), -0 , args[2 ]); -tcg_out32 (s,XOR |SAB (args[1 ], args[0 ],0 )); - } +tcg_out_xori32 (s, a0, a1, a2); + }else { +tcg_out32 (s,XOR |SAB (a1, a0, a2)); } -else -tcg_out32 (s,XOR |SAB (args[1 ], args[0 ], args[2 ]));break ;case INDEX_op_mul_i32:
Signature · 1 line
1.8.1.4
PATCH v3 12/27 tcg-ppc64: Improve and_i32 with constant
In reply to Richard HendersonUse RLWINM
--- tcg/ppc64/tcg-target.c | 50 +++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 49 insertions(+), 1 deletion(-)
tcg/ppc64/tcg-target.c +49 -1
@@ -527,6 +527,48 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret, } } +static inline bool mask_operand (uint32_t c,int *mb,int *me) +{ +uint32_t lsb, test; + +/* Accept a bit pattern like: +0....01....1 +1....10....0 +0..01..10..0 +Keep track of the transitions. */ +if (c ==0 || c ==-1 ) { +return false; + } + test = c; + lsb = test & -test; + test += lsb; +if (test & (test -1 )) { +return false; + } + + *me =clz32 (lsb); + *mb = test ?clz32 (test & -test) +1 :0 ; +return true; +} + +static void tcg_out_andi32 (TCGContext *s,TCGReg dst,TCGReg src,uint32_t c) +{ +int mb, me; + +if ((c &0xffff ) == c) { +tcg_out32 (s,ANDI |SAI (src, dst, c)); +return ; + }else if ((c &0xffff0000 ) == c) { +tcg_out32 (s,ANDIS |SAI (src, dst, c >>16 )); +return ; + }else if (mask_operand (c, &mb, &me)) { +tcg_out_rlw (s,RLWINM , dst, src,0 , mb, me); + }else { +tcg_out_movi (s,TCG_TYPE_I32 ,0 , c); +tcg_out32 (s,AND |SAB (src, dst,0 )); + } +} +static void tcg_out_zori32 (TCGContext *s,TCGReg dst,TCGReg src,uint32_t c,int op_lo,int op_hi) {
@@ -1346,9 +1388,15 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, }break ; -case INDEX_op_and_i64:case INDEX_op_and_i32:if (const_args[2 ]) { +tcg_out_andi32 (s, args[0 ], args[1 ], args[2 ]); + }else { +tcg_out32 (s,AND |SAB (args[1 ], args[0 ], args[2 ])); + } +break ; +case INDEX_op_and_i64: +if (const_args[2 ]) {if ((args[2 ] &0xffff ) == args[2 ]) {tcg_out32 (s,ANDI |SAI (args[1 ], args[0 ], args[2 ])); }else if ((args[2 ] &0xffff0000 ) == args[2 ]) {
Signature · 1 line
1.8.1.4
PATCH v3 13/27 tcg-ppc64: Improve and_i64 with constant
In reply to Richard HendersonUse RLDICL and RLDICR.
--- tcg/ppc64/tcg-target.c | 64 +++++++++++++++++++++++++++++++++++++++----------- 1 file changed, 50 insertions(+), 14 deletions(-)
tcg/ppc64/tcg-target.c +50 -14
@@ -527,7 +527,7 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret, } } -static inline bool mask_operand (uint32_t c,int *mb,int *me) +static bool mask_operand (uint32_t c,int *mb,int *me) {uint32_t lsb, test;
@@ -551,6 +551,30 @@ static inline bool mask_operand(uint32_t c, int *mb, int *me)return true; } +static bool mask64_operand (uint64_t c,int *mb,int *me) +{ +uint64_t lsb; + +if (c ==0 ) { +return false; + } + + lsb = c & -c; +/* Accept 1..10..0. */ +if (c == -lsb) { + *mb =0 ; + *me =clz64 (lsb); +return true; + } +/* Accept 0..01..1. */ +if (lsb ==1 && (c & (c +1 )) ==0 ) { + *mb =clz64 (c +1 ) +1 ; + *me =63 ; +return true; + } +return false; +} +static void tcg_out_andi32 (TCGContext *s,TCGReg dst,TCGReg src,uint32_t c) {int mb, me;
@@ -569,6 +593,28 @@ static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c) } } +static void tcg_out_andi64 (TCGContext *s,TCGReg dst,TCGReg src,uint64_t c) +{ +int mb, me; + +if ((c &0xffff ) == c) { +tcg_out32 (s,ANDI |SAI (src, dst, c)); +return ; + }else if ((c &0xffff0000 ) == c) { +tcg_out32 (s,ANDIS |SAI (src, dst, c >>16 )); +return ; + }else if (mask64_operand (c, &mb, &me)) { +if (mb ==0 ) { +tcg_out_rld (s,RLDICR , dst, src,0 , me); + }else { +tcg_out_rld (s,RLDICL , dst, src,0 , mb); + } + }else { +tcg_out_movi (s,TCG_TYPE_I64 ,0 , c); +tcg_out32 (s,AND |SAB (src, dst,0 )); + } +} +static void tcg_out_zori32 (TCGContext *s,TCGReg dst,TCGReg src,uint32_t c,int op_lo,int op_hi) {
@@ -1397,20 +1443,10 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_and_i64:if (const_args[2 ]) { -if ((args[2 ] &0xffff ) == args[2 ]) { -tcg_out32 (s,ANDI |SAI (args[1 ], args[0 ], args[2 ])); - }else if ((args[2 ] &0xffff0000 ) == args[2 ]) { -tcg_out32 (s,ANDIS |SAI (args[1 ], args[0 ], args[2 ] >>16 )); - }else { -tcg_out_movi (s, (opc == INDEX_op_and_i32 - ?TCG_TYPE_I32 - :TCG_TYPE_I64 ), -0 , args[2 ]); -tcg_out32 (s,AND |SAB (args[1 ], args[0 ],0 )); - } +tcg_out_andi64 (s, args[0 ], args[1 ], args[2 ]); + }else { +tcg_out32 (s,AND |SAB (args[1 ], args[0 ], args[2 ])); } -else -tcg_out32 (s,AND |SAB (args[1 ], args[0 ], args[2 ]));break ;case INDEX_op_or_i64:case INDEX_op_or_i32:
Signature · 1 line
1.8.1.4
Re: [Qemu-devel] [PATCH v3 13/27] tcg-ppc64: Improve and_i64 with constant
Quoted 94 lines · Richard Henderson
Use RLDICL and RLDICR.
Signed-off-by Richard Henderson <[email protected]>--- tcg/ppc64/tcg-target.c | 64 +++++++++++++++++++++++++++++++++++++++----------- 1 file changed, 50 insertions(+), 14 deletions(-)tcg/ppc64/tcg-target.c +47
-11@@ -527,7 +527,7 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg ret, } } -static inline bool mask_operand (uint32_t c,int *mb,int *me) +static bool mask_operand (uint32_t c,int *mb,int *me) {uint32_t lsb, test;@@ -551,6 +551,30 @@ static inline bool mask_operand(uint32_t c, int *mb, int *me)return true; } +static bool mask64_operand (uint64_t c,int *mb,int *me) +{ +uint64_t lsb; + +if (c ==0 ) { +return false; + } + + lsb = c & -c; +/* Accept 1..10..0. */ +if (c == -lsb) { + *mb =0 ; + *me =clz64 (lsb); +return true; + } +/* Accept 0..01..1. */ +if (lsb ==1 && (c & (c +1 )) ==0 ) { + *mb =clz64 (c +1 ) +1 ; + *me =63 ; +return true; + } +return false; +} +static void tcg_out_andi32 (TCGContext *s,TCGReg dst,TCGReg src,uint32_t c) {int mb, me;@@ -569,6 +593,28 @@ static void tcg_out_andi32(TCGContext *s, TCGReg dst, TCGReg src, uint32_t c) } } +static void tcg_out_andi64 (TCGContext *s,TCGReg dst,TCGReg src,uint64_t c) +{ +int mb, me; + +if ((c &0xffff ) == c) { +tcg_out32 (s,ANDI |SAI (src, dst, c)); +return ; + }else if ((c &0xffff0000 ) == c) { +tcg_out32 (s,ANDIS |SAI (src, dst, c >>16 )); +return ; + }else if (mask64_operand (c, &mb, &me)) { +if (mb ==0 ) { +tcg_out_rld (s,RLDICR , dst, src,0 , me); + }else { +tcg_out_rld (s,RLDICL , dst, src,0 , mb); + } + }else { +tcg_out_movi (s,TCG_TYPE_I64 ,0 , c); +tcg_out32 (s,AND |SAB (src, dst,0 )); + } +} +static void tcg_out_zori32 (TCGContext *s,TCGReg dst,TCGReg src,uint32_t c,int op_lo,int op_hi) {@@ -1397,20 +1443,10 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_and_i64:if (const_args[2 ]) { -if ((args[2 ] &0xffff ) == args[2 ]) { -tcg_out32 (s,ANDI |SAI (args[1 ], args[0 ], args[2 ])); - }else if ((args[2 ] &0xffff0000 ) == args[2 ]) { -tcg_out32 (s,ANDIS |SAI (args[1 ], args[0 ], args[2 ] >>16 )); - }else { -tcg_out_movi (s, (opc == INDEX_op_and_i32 - ?TCG_TYPE_I32 - :TCG_TYPE_I64 ), -0 , args[2 ]); -tcg_out32 (s,AND |SAB (args[1 ], args[0 ],0 ));
tcg/ppc64/tcg-target.c static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, · line 1452
Quoted 10 lines
tcg/ppc64/tcg-target.c +3
-3- } +tcg_out_andi64 (s, args[0 ], args[1 ], args[2 ]); + }else { +tcg_out32 (s,AND |SAB (args[1 ], args[0 ], args[2 ])); } -else -tcg_out32 (s,AND |SAB (args[1 ], args[0 ], args[2 ]));break ;case INDEX_op_or_i64:case INDEX_op_or_i32:
Signature · 2 lines
Aurelien Jarno GPG: 1024D/F1BCDB73 [email protected] http://www.aurel32.net
PATCH v3 14/27 tcg-ppc64: Streamline qemu_ld/st insn selection
In reply to Richard HendersonUsing a table to look up insns of the right width and sign. Include support for the Power 2.05 LDBRX and STDBRX insns included in e.g. Power6.
--- tcg/ppc64/tcg-target.c | 166 +++++++++++++++++-------------------------------- 1 file changed, 56 insertions(+), 110 deletions(-)
tcg/ppc64/tcg-target.c +56 -110
@@ -44,6 +44,8 @@ static uint8_t *tb_ret_addr;#define GUEST_BASE 0#endif +#define HAVE_ISA_2_06 0 +#ifdef CONFIG_USE_GUEST_BASE #define TCG_GUEST_BASE_REG 30#else
@@ -368,8 +370,10 @@ static int tcg_target_const_match (tcg_target_long val,#define CMPL XO31( 32)#define LHBRX XO31(790)#define LWBRX XO31(534) +#define LDBRX XO31(532)#define STHBRX XO31(918)#define STWBRX XO31(662) +#define STDBRX XO31(660)#define MFSPR XO31(339)#define MTSPR XO31(467)#define SRAWI XO31(824)
@@ -759,22 +763,44 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg r0, TCGReg r1, TCGReg r2, }#endif +static const uint32_t qemu_ldx_opc[8 ] = { +#ifdef TARGET_WORDS_BIGENDIAN +LBZX ,LHZX ,LWZX ,LDX , + 0, LHAX, LWAX, LDX +#else +LBZX ,LHBRX ,LWBRX ,LDBRX , +0 ,0 ,0 ,LDBRX , +#endif +}; + +static const uint32_t qemu_stx_opc[4 ] = { +#ifdef TARGET_WORDS_BIGENDIAN + STBX, STHX, STWX, STDX +#else +STBX ,STHBRX ,STWBRX ,STDBRX , +#endif +}; + +static const uint32_t qemu_exts_opc[4 ] = { +EXTSB ,EXTSH ,EXTSW ,0 +}; +static void tcg_out_qemu_ld (TCGContext *s,const TCGArg *args,int opc) {TCGReg addr_reg, data_reg, r0, r1, rbase; -int bswap; +uint32_t insn, s_bits;#ifdef CONFIG_SOFTMMU TCGReg r2, ir; -int mem_index, s_bits; +int mem_index;void *label1_ptr, *label2_ptr;#endif data_reg = *args++; addr_reg = *args++; + s_bits = opc &3 ;#ifdef CONFIG_SOFTMMU mem_index = *args; - s_bits = opc &3 ; r0 =3 ; r1 =4 ;
@@ -799,23 +825,11 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc)tcg_out_call (s, (tcg_target_long ) qemu_ld_helpers[s_bits],1 ); -switch (opc) { -case 0 |4 : -tcg_out32 (s,EXTSB |RA (data_reg) |RS (3 )); -break ; -case 1 |4 : -tcg_out32 (s,EXTSH |RA (data_reg) |RS (3 )); -break ; -case 2 |4 : -tcg_out32 (s,EXTSW |RA (data_reg) |RS (3 )); -break ; -case 0 : -case 1 : -case 2 : -case 3 : -if (data_reg !=3 ) -tcg_out_mov (s,TCG_TYPE_I64 , data_reg,3 ); -break ; +if (opc &4 ) { + insn = qemu_exts_opc[s_bits]; +tcg_out32 (s, insn |RA (data_reg) |RS (3 )); + }else if (data_reg !=3 ) { +tcg_out_mov (s,TCG_TYPE_I64 , data_reg,3 ); } label2_ptr = s->code_ptr;tcg_out32 (s,B );
@@ -842,65 +856,19 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc) rbase =GUEST_BASE ?TCG_GUEST_BASE_REG :0 ;#endif -#ifdef TARGET_WORDS_BIGENDIAN - bswap =0 ; -#else - bswap =1 ; -#endif -switch (opc) { -default : -case 0 : -tcg_out32 (s,LBZX |TAB (data_reg, rbase, r0)); -break ; -case 0 |4 : -tcg_out32 (s,LBZX |TAB (data_reg, rbase, r0)); -tcg_out32 (s,EXTSB |RA (data_reg) |RS (data_reg)); -break ; -case 1 : -if (bswap) -tcg_out32 (s,LHBRX |TAB (data_reg, rbase, r0)); -else -tcg_out32 (s,LHZX |TAB (data_reg, rbase, r0)); -break ; -case 1 |4 : -if (bswap) { -tcg_out32 (s,LHBRX |TAB (data_reg, rbase, r0)); -tcg_out32 (s,EXTSH |RA (data_reg) |RS (data_reg)); - } -else tcg_out32 (s,LHAX |TAB (data_reg, rbase, r0)); -break ; -case 2 : -if (bswap) -tcg_out32 (s,LWBRX |TAB (data_reg, rbase, r0)); -else -tcg_out32 (s,LWZX |TAB (data_reg, rbase, r0)); -break ; -case 2 |4 : -if (bswap) { -tcg_out32 (s,LWBRX |TAB (data_reg, rbase, r0)); -tcg_out32 (s,EXTSW |RA (data_reg) |RS (data_reg)); - } -else tcg_out32 (s,LWAX |TAB (data_reg, rbase, r0)); -break ; -case 3 : -#ifdef CONFIG_USE_GUEST_BASE -if (bswap) { -tcg_out32 (s,ADDI |TAI (r1, r0,4 )); -tcg_out32 (s,LWBRX |TAB (data_reg, rbase, r0)); -tcg_out32 (s,LWBRX |TAB ( r1, rbase, r1)); -tcg_out_rld (s,RLDIMI , data_reg, r1,32 ,0 ); - } -else tcg_out32 (s,LDX |TAB (data_reg, rbase, r0)); -#else -if (bswap) { -tcg_out_movi32 (s,0 ,4 ); -tcg_out32 (s,LWBRX |RT (data_reg) |RB (r0)); -tcg_out32 (s,LWBRX |RT ( r1) |RA (r0)); -tcg_out_rld (s,RLDIMI , data_reg, r1,32 ,0 ); - } -else tcg_out32 (s,LD |RT (data_reg) |RA (r0)); -#endif -break ; + insn = qemu_ldx_opc[opc]; +if (!HAVE_ISA_2_06 && insn ==LDBRX ) { +tcg_out32 (s,ADDI |TAI (r1, r0,4 )); +tcg_out32 (s,LWBRX |TAB (data_reg, rbase, r0)); +tcg_out32 (s,LWBRX |TAB ( r1, rbase, r1)); +tcg_out_rld (s,RLDIMI , data_reg, r1,32 ,0 ); + }else if (insn) { +tcg_out32 (s, insn |TAB (data_reg, rbase, r0)); + }else { + insn = qemu_ldx_opc[s_bits]; +tcg_out32 (s, insn |TAB (data_reg, rbase, r0)); + insn = qemu_exts_opc[s_bits]; +tcg_out32 (s, insn |RA (data_reg) |RS (data_reg)); }#ifdef CONFIG_SOFTMMU
@@ -911,7 +879,7 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc)static void tcg_out_qemu_st (TCGContext *s,const TCGArg *args,int opc) {TCGReg addr_reg, r0, r1, rbase, data_reg; -int bswap; +uint32_t insn;#ifdef CONFIG_SOFTMMU TCGReg r2, ir;int mem_index;
@@ -975,36 +943,14 @@ static void tcg_out_qemu_st (TCGContext *s, const TCGArg *args, int opc) rbase =GUEST_BASE ?TCG_GUEST_BASE_REG :0 ;#endif -#ifdef TARGET_WORDS_BIGENDIAN - bswap =0 ; -#else - bswap =1 ; -#endif -switch (opc) { -case 0 : -tcg_out32 (s,STBX |SAB (data_reg, rbase, r0)); -break ; -case 1 : -if (bswap) -tcg_out32 (s,STHBRX |SAB (data_reg, rbase, r0)); -else -tcg_out32 (s,STHX |SAB (data_reg, rbase, r0)); -break ; -case 2 : -if (bswap) -tcg_out32 (s,STWBRX |SAB (data_reg, rbase, r0)); -else -tcg_out32 (s,STWX |SAB (data_reg, rbase, r0)); -break ; -case 3 : -if (bswap) { -tcg_out32 (s,STWBRX |SAB (data_reg, rbase, r0)); -tcg_out32 (s,ADDI |TAI (r1, r0,4 )); -tcg_out_shri64 (s,0 , data_reg,32 ); -tcg_out32 (s,STWBRX |SAB (0 , rbase, r1)); - } -else tcg_out32 (s,STDX |SAB (data_reg, rbase, r0)); -break ; + insn = qemu_stx_opc[opc]; +if (!HAVE_ISA_2_06 && insn ==STDBRX ) { +tcg_out32 (s,STWBRX |SAB (data_reg, rbase, r0)); +tcg_out32 (s,ADDI |TAI (r1, r0,4 )); +tcg_out_shri64 (s,0 , data_reg,32 ); +tcg_out32 (s,STWBRX |SAB (0 , rbase, r1)); + }else { +tcg_out32 (s, insn |SAB (data_reg, rbase, r0)); }#ifdef CONFIG_SOFTMMU
Signature · 1 line
1.8.1.4
Re: [Qemu-devel] [PATCH v3 14/27] tcg-ppc64: Streamline qemu_ld/st insn selection
Quoted 233 lines · Richard Henderson
Using a table to look up insns of the right width and sign. Include support for the Power 2.05 LDBRX and STDBRX insns included in e.g. Power6.
Signed-off-by Richard Henderson <[email protected]>--- tcg/ppc64/tcg-target.c | 166 +++++++++++++++++-------------------------------- 1 file changed, 56 insertions(+), 110 deletions(-)tcg/ppc64/tcg-target.c +49
-110@@ -44,6 +44,8 @@ static uint8_t *tb_ret_addr;#define GUEST_BASE 0#endif +#define HAVE_ISA_2_06 0 +#ifdef CONFIG_USE_GUEST_BASE #define TCG_GUEST_BASE_REG 30#else @@ -368,8 +370,10 @@ static int tcg_target_const_match (tcg_target_long val,#define CMPL XO31( 32)#define LHBRX XO31(790)#define LWBRX XO31(534) +#define LDBRX XO31(532)#define STHBRX XO31(918)#define STWBRX XO31(662) +#define STDBRX XO31(660)#define MFSPR XO31(339)#define MTSPR XO31(467)#define SRAWI XO31(824)@@ -759,22 +763,44 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg r0, TCGReg r1, TCGReg r2, }#endif +static const uint32_t qemu_ldx_opc[8 ] = { +#ifdef TARGET_WORDS_BIGENDIAN +LBZX ,LHZX ,LWZX ,LDX , + 0, LHAX, LWAX, LDX +#else +LBZX ,LHBRX ,LWBRX ,LDBRX , +0 ,0 ,0 ,LDBRX , +#endif +}; + +static const uint32_t qemu_stx_opc[4 ] = { +#ifdef TARGET_WORDS_BIGENDIAN + STBX, STHX, STWX, STDX +#else +STBX ,STHBRX ,STWBRX ,STDBRX , +#endif +}; + +static const uint32_t qemu_exts_opc[4 ] = { +EXTSB ,EXTSH ,EXTSW ,0 +}; +static void tcg_out_qemu_ld (TCGContext *s,const TCGArg *args,int opc) {TCGReg addr_reg, data_reg, r0, r1, rbase; -int bswap; +uint32_t insn, s_bits;#ifdef CONFIG_SOFTMMU TCGReg r2, ir; -int mem_index, s_bits; +int mem_index;void *label1_ptr, *label2_ptr;#endif data_reg = *args++; addr_reg = *args++; + s_bits = opc &3 ;#ifdef CONFIG_SOFTMMU mem_index = *args; - s_bits = opc &3 ; r0 =3 ; r1 =4 ;@@ -799,23 +825,11 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc)tcg_out_call (s, (tcg_target_long ) qemu_ld_helpers[s_bits],1 ); -switch (opc) { -case 0 |4 : -tcg_out32 (s,EXTSB |RA (data_reg) |RS (3 )); -break ; -case 1 |4 : -tcg_out32 (s,EXTSH |RA (data_reg) |RS (3 )); -break ; -case 2 |4 : -tcg_out32 (s,EXTSW |RA (data_reg) |RS (3 )); -break ; -case 0 : -case 1 : -case 2 : -case 3 : -if (data_reg !=3 ) -tcg_out_mov (s,TCG_TYPE_I64 , data_reg,3 ); -break ; +if (opc &4 ) { + insn = qemu_exts_opc[s_bits]; +tcg_out32 (s, insn |RA (data_reg) |RS (3 )); + }else if (data_reg !=3 ) { +tcg_out_mov (s,TCG_TYPE_I64 , data_reg,3 ); } label2_ptr = s->code_ptr;tcg_out32 (s,B );@@ -842,65 +856,19 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc) rbase =GUEST_BASE ?TCG_GUEST_BASE_REG :0 ;#endif -#ifdef TARGET_WORDS_BIGENDIAN - bswap =0 ; -#else - bswap =1 ; -#endif -switch (opc) { -default : -case 0 : -tcg_out32 (s,LBZX |TAB (data_reg, rbase, r0)); -break ; -case 0 |4 : -tcg_out32 (s,LBZX |TAB (data_reg, rbase, r0)); -tcg_out32 (s,EXTSB |RA (data_reg) |RS (data_reg)); -break ; -case 1 : -if (bswap) -tcg_out32 (s,LHBRX |TAB (data_reg, rbase, r0)); -else -tcg_out32 (s,LHZX |TAB (data_reg, rbase, r0)); -break ; -case 1 |4 : -if (bswap) { -tcg_out32 (s,LHBRX |TAB (data_reg, rbase, r0)); -tcg_out32 (s,EXTSH |RA (data_reg) |RS (data_reg)); - } -else tcg_out32 (s,LHAX |TAB (data_reg, rbase, r0)); -break ; -case 2 : -if (bswap) -tcg_out32 (s,LWBRX |TAB (data_reg, rbase, r0)); -else -tcg_out32 (s,LWZX |TAB (data_reg, rbase, r0)); -break ; -case 2 |4 : -if (bswap) { -tcg_out32 (s,LWBRX |TAB (data_reg, rbase, r0)); -tcg_out32 (s,EXTSW |RA (data_reg) |RS (data_reg)); - } -else tcg_out32 (s,LWAX |TAB (data_reg, rbase, r0)); -break ; -case 3 : -#ifdef CONFIG_USE_GUEST_BASE -if (bswap) { -tcg_out32 (s,ADDI |TAI (r1, r0,4 )); -tcg_out32 (s,LWBRX |TAB (data_reg, rbase, r0)); -tcg_out32 (s,LWBRX |TAB ( r1, rbase, r1)); -tcg_out_rld (s,RLDIMI , data_reg, r1,32 ,0 ); - } -else tcg_out32 (s,LDX |TAB (data_reg, rbase, r0)); -#else -if (bswap) { -tcg_out_movi32 (s,0 ,4 ); -tcg_out32 (s,LWBRX |RT (data_reg) |RB (r0)); -tcg_out32 (s,LWBRX |RT ( r1) |RA (r0)); -tcg_out_rld (s,RLDIMI , data_reg, r1,32 ,0 ); - } -else tcg_out32 (s,LD |RT (data_reg) |RA (r0)); -#endif -break ; + insn = qemu_ldx_opc[opc]; +if (!HAVE_ISA_2_06 && insn ==LDBRX ) { +tcg_out32 (s,ADDI |TAI (r1, r0,4 )); +tcg_out32 (s,LWBRX |TAB (data_reg, rbase, r0)); +tcg_out32 (s,LWBRX |TAB ( r1, rbase, r1)); +tcg_out_rld (s,RLDIMI , data_reg, r1,32 ,0 ); + }else if (insn) { +tcg_out32 (s, insn |TAB (data_reg, rbase, r0)); + }else { + insn = qemu_ldx_opc[s_bits]; +tcg_out32 (s, insn |TAB (data_reg, rbase, r0)); + insn = qemu_exts_opc[s_bits]; +tcg_out32 (s, insn |RA (data_reg) |RS (data_reg)); }#ifdef CONFIG_SOFTMMU @@ -911,7 +879,7 @@ static void tcg_out_qemu_ld (TCGContext *s, const TCGArg *args, int opc)static void tcg_out_qemu_st (TCGContext *s,const TCGArg *args,int opc) {TCGReg addr_reg, r0, r1, rbase, data_reg; -int bswap; +uint32_t insn;#ifdef CONFIG_SOFTMMU TCGReg r2, ir;int mem_index;@@ -975,36 +943,14 @@ static void tcg_out_qemu_st (TCGContext *s, const TCGArg *args, int opc) rbase =GUEST_BASE ?TCG_GUEST_BASE_REG :0 ;#endif -#ifdef TARGET_WORDS_BIGENDIAN - bswap =0 ; -#else - bswap =1 ; -#endif -switch (opc) { -case 0 : -tcg_out32 (s,STBX |SAB (data_reg, rbase, r0)); -break ; -case 1 : -if (bswap) -tcg_out32 (s,STHBRX |SAB (data_reg, rbase, r0)); -else -tcg_out32 (s,STHX |SAB (data_reg, rbase, r0)); -break ; -case 2 : -if (bswap) -tcg_out32 (s,STWBRX |SAB (data_reg, rbase, r0)); -else -tcg_out32 (s,STWX |SAB (data_reg, rbase, r0)); -break ; -case 3 : -if (bswap) { -tcg_out32 (s,STWBRX |SAB (data_reg, rbase, r0)); -tcg_out32 (s,ADDI |TAI (r1, r0,4 )); -tcg_out_shri64 (s,0 , data_reg,32 ); -tcg_out32 (s,STWBRX |SAB (0 , rbase, r1)); - } -else tcg_out32 (s,STDX |SAB (data_reg, rbase, r0)); -break ; + insn = qemu_stx_opc[opc];
tcg/ppc64/tcg-target.c static void tcg_out_qemu_st (TCGContext *s, const TCGArg *args, int opc) · line 956
Quoted 10 lines
tcg/ppc64/tcg-target.c +7
-0+if (!HAVE_ISA_2_06 && insn ==STDBRX ) { +tcg_out32 (s,STWBRX |SAB (data_reg, rbase, r0)); +tcg_out32 (s,ADDI |TAI (r1, r0,4 )); +tcg_out_shri64 (s,0 , data_reg,32 ); +tcg_out32 (s,STWBRX |SAB (0 , rbase, r1)); + }else { +tcg_out32 (s, insn |SAB (data_reg, rbase, r0)); }#ifdef CONFIG_SOFTMMU
Signature · 2 lines
Aurelien Jarno GPG: 1024D/F1BCDB73 [email protected] http://www.aurel32.net
PATCH v3 15/27 tcg-ppc64: Implement rotates
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 39 +++++++++++++++++++++++++++++++++++++++ tcg/ppc64/tcg-target.h | 4 ++-- 2 files changed, 41 insertions(+), 2 deletions(-)
tcg/ppc64/tcg-target.c +39 -0
@@ -336,11 +336,14 @@ static int tcg_target_const_match (tcg_target_long val,#define LWZU OPCD( 33)#define STWU OPCD( 37) +#define RLWIMI OPCD( 20)#define RLWINM OPCD( 21) +#define RLWNM OPCD( 23)#define RLDICL XO30( 0)#define RLDICR XO30( 1)#define RLDIMI XO30( 3) +#define RLDCL XO30( 8)#define BCLR XO19( 16)#define BCCTR XO19(528)
@@ -1467,6 +1470,23 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,else tcg_out32 (s,SRAW |SAB (args[1 ], args[0 ], args[2 ]));break ; +case INDEX_op_rotl_i32: +if (const_args[2 ]) { +tcg_out_rlw (s,RLWINM , args[0 ], args[1 ], args[2 ],0 ,31 ); + }else { +tcg_out32 (s,RLWNM |SAB (args[1 ], args[0 ], args[2 ]) + |MB (0 ) |ME (31 )); + } +break ; +case INDEX_op_rotr_i32: +if (const_args[2 ]) { +tcg_out_rlw (s,RLWINM , args[0 ], args[1 ],32 - args[2 ],0 ,31 ); + }else { +tcg_out32 (s,SUBFIC |TAI (0 , args[2 ],32 )); +tcg_out32 (s,RLWNM |SAB (args[1 ], args[0 ], args[2 ]) + |MB (0 ) |ME (31 )); + } +break ;case INDEX_op_brcond_i32:tcg_out_brcond (s, args[2 ], args[0 ], args[1 ], const_args[1 ], args[3 ],0 );
@@ -1549,6 +1569,21 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,else tcg_out32 (s,SRAD |SAB (args[1 ], args[0 ], args[2 ]));break ; +case INDEX_op_rotl_i64: +if (const_args[2 ]) { +tcg_out_rld (s,RLDICL , args[0 ], args[1 ], args[2 ],0 ); + }else { +tcg_out32 (s,RLDCL |SAB (args[1 ], args[0 ], args[2 ]) |MB64 (0 )); + } +break ; +case INDEX_op_rotr_i64: +if (const_args[2 ]) { +tcg_out_rld (s,RLDICL , args[0 ], args[1 ],64 - args[2 ],0 ); + }else { +tcg_out32 (s,SUBFIC |TAI (0 , args[2 ],64 )); +tcg_out32 (s,RLDCL |SAB (args[1 ], args[0 ],0 ) |MB64 (0 )); + } +break ;case INDEX_op_mul_i64:tcg_out32 (s,MULLD |TAB (args[0 ], args[1 ], args[2 ]));
@@ -1685,6 +1720,8 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_shl_i32, { "r", "r", "ri" } }, { INDEX_op_shr_i32, { "r", "r", "ri" } }, { INDEX_op_sar_i32, { "r", "r", "ri" } }, + { INDEX_op_rotl_i32, { "r", "r", "ri" } }, + { INDEX_op_rotr_i32, { "r", "r", "ri" } }, { INDEX_op_brcond_i32, { "r", "ri" } }, { INDEX_op_brcond_i64, { "r", "ri" } },
@@ -1701,6 +1738,8 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_shl_i64, { "r", "r", "ri" } }, { INDEX_op_shr_i64, { "r", "r", "ri" } }, { INDEX_op_sar_i64, { "r", "r", "ri" } }, + { INDEX_op_rotl_i64, { "r", "r", "ri" } }, + { INDEX_op_rotr_i64, { "r", "r", "ri" } }, { INDEX_op_mul_i64, { "r", "r", "r" } }, { INDEX_op_div_i64, { "r", "r", "r" } },
tcg/ppc64/tcg-target.h +2 -2
@@ -69,7 +69,7 @@ typedef enum {/* optional instructions */ #define TCG_TARGET_HAS_div_i32 1 -#define TCG_TARGET_HAS_rot_i32 0 +#define TCG_TARGET_HAS_rot_i32 1#define TCG_TARGET_HAS_ext8s_i32 1#define TCG_TARGET_HAS_ext16s_i32 1#define TCG_TARGET_HAS_ext8u_i32 0
@@ -91,7 +91,7 @@ typedef enum {#define TCG_TARGET_HAS_muls2_i32 0#define TCG_TARGET_HAS_div_i64 1 -#define TCG_TARGET_HAS_rot_i64 0 +#define TCG_TARGET_HAS_rot_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
1.8.1.4
PATCH v3 16/27 tcg-ppc64: Implement bswap16 and bswap32
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 43 +++++++++++++++++++++++++++++++++++++++++++ tcg/ppc64/tcg-target.h | 18 ++++++++++-------- 2 files changed, 53 insertions(+), 8 deletions(-)
tcg/ppc64/tcg-target.c +43 -0
@@ -1668,6 +1668,44 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, const_args[2]);break ; +case INDEX_op_bswap16_i32: +case INDEX_op_bswap16_i64: + a0 = args[0 ], a1 = args[1 ]; +/* a1 = abcd */ +if (a0 != a1) { +/* a0 = (a1 r<< 24) & 0xff # 000c */ +tcg_out_rlw (s,RLWINM , a0, a1,24 ,24 ,31 ); +/* a0 = (a0 & ~0xff00) | (a1 r<< 8) & 0xff00 # 00dc */ +tcg_out_rlw (s,RLWIMI , a0, a1,8 ,16 ,23 ); + }else { +/* r0 = (a1 r<< 8) & 0xff00 # 00d0 */ +tcg_out_rlw (s,RLWINM ,0 , a1,8 ,16 ,23 ); +/* a0 = (a1 r<< 24) & 0xff # 000c */ +tcg_out_rlw (s,RLWINM , a0, a1,24 ,24 ,31 ); +/* a0 = a0 | r0 # 00dc */ +tcg_out32 (s,OR |SAB (0 , a0, a0)); + } +break ; + +case INDEX_op_bswap32_i32: +case INDEX_op_bswap32_i64: +/* Stolen from gcc's builtin_bswap32 */ + a1 = args[1 ]; + a0 = args[0 ] == a1 ?0 : args[0 ]; + +/* a1 = args[1] # abcd */ +/* a0 = rotate_left (a1, 8) # bcda */ +tcg_out_rlw (s,RLWINM , a0, a1,8 ,0 ,31 ); +/* a0 = (a0 & ~0xff000000) | ((a1 r<< 24) & 0xff000000) # dcda */ +tcg_out_rlw (s,RLWIMI , a0, a1,24 ,0 ,7 ); +/* a0 = (a0 & ~0x0000ff00) | ((a1 r<< 24) & 0x0000ff00) # dcba */ +tcg_out_rlw (s,RLWIMI , a0, a1,24 ,16 ,23 ); + +if (!a0) { +tcg_out_mov (s,TCG_TYPE_I64 , args[0 ], a0); + } +break ; +default :tcg_dump_ops (s);tcg_abort ();
@@ -1774,6 +1812,11 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_setcond_i32, { "r", "r", "ri" } }, { INDEX_op_setcond_i64, { "r", "r", "ri" } }, + { INDEX_op_bswap16_i32, { "r", "r" } }, + { INDEX_op_bswap16_i64, { "r", "r" } }, + { INDEX_op_bswap32_i32, { "r", "r" } }, + { INDEX_op_bswap32_i64, { "r", "r" } }, + { -1 }, };
tcg/ppc64/tcg-target.h +10 -8
@@ -67,15 +67,19 @@ typedef enum {#define TCG_TARGET_STACK_ALIGN 16#define TCG_TARGET_CALL_STACK_OFFSET 48 +/* optional instructions automatically implemented */ +#define TCG_TARGET_HAS_ext8u_i32 0/* andi */ +#define TCG_TARGET_HAS_ext16u_i32 0 +#define TCG_TARGET_HAS_ext8u_i64 0 +#define TCG_TARGET_HAS_ext16u_i64 0 +/* optional instructions */ #define TCG_TARGET_HAS_div_i32 1#define TCG_TARGET_HAS_rot_i32 1#define TCG_TARGET_HAS_ext8s_i32 1#define TCG_TARGET_HAS_ext16s_i32 1 -#define TCG_TARGET_HAS_ext8u_i32 0 -#define TCG_TARGET_HAS_ext16u_i32 0 -#define TCG_TARGET_HAS_bswap16_i32 0 -#define TCG_TARGET_HAS_bswap32_i32 0 +#define TCG_TARGET_HAS_bswap16_i32 1 +#define TCG_TARGET_HAS_bswap32_i32 1#define TCG_TARGET_HAS_not_i32 1#define TCG_TARGET_HAS_neg_i32 1#define TCG_TARGET_HAS_andc_i32 0
@@ -95,11 +99,9 @@ typedef enum {#define TCG_TARGET_HAS_ext8s_i64 1#define TCG_TARGET_HAS_ext16s_i64 1#define TCG_TARGET_HAS_ext32s_i64 1 -#define TCG_TARGET_HAS_ext8u_i64 0 -#define TCG_TARGET_HAS_ext16u_i64 0#define TCG_TARGET_HAS_ext32u_i64 1 -#define TCG_TARGET_HAS_bswap16_i64 0 -#define TCG_TARGET_HAS_bswap32_i64 0 +#define TCG_TARGET_HAS_bswap16_i64 1 +#define TCG_TARGET_HAS_bswap32_i64 1#define TCG_TARGET_HAS_bswap64_i64 0#define TCG_TARGET_HAS_not_i64 1#define TCG_TARGET_HAS_neg_i64 1
Signature · 1 line
1.8.1.4
Re: [Qemu-devel] [PATCH v3 16/27] tcg-ppc64: Implement bswap16 and bswap32
Quoted 42 lines · Richard Henderson
Signed-off-by Richard Henderson <[email protected]>--- tcg/ppc64/tcg-target.c | 43 +++++++++++++++++++++++++++++++++++++++++++ tcg/ppc64/tcg-target.h | 18 ++++++++++-------- 2 files changed, 53 insertions(+), 8 deletions(-)tcg/ppc64/tcg-target.c +28
-0@@ -1668,6 +1668,44 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, const_args[2]);break ; +case INDEX_op_bswap16_i32: +case INDEX_op_bswap16_i64: + a0 = args[0 ], a1 = args[1 ]; +/* a1 = abcd */ +if (a0 != a1) { +/* a0 = (a1 r<< 24) & 0xff # 000c */ +tcg_out_rlw (s,RLWINM , a0, a1,24 ,24 ,31 ); +/* a0 = (a0 & ~0xff00) | (a1 r<< 8) & 0xff00 # 00dc */ +tcg_out_rlw (s,RLWIMI , a0, a1,8 ,16 ,23 ); + }else { +/* r0 = (a1 r<< 8) & 0xff00 # 00d0 */ +tcg_out_rlw (s,RLWINM ,0 , a1,8 ,16 ,23 ); +/* a0 = (a1 r<< 24) & 0xff # 000c */ +tcg_out_rlw (s,RLWINM , a0, a1,24 ,24 ,31 ); +/* a0 = a0 | r0 # 00dc */ +tcg_out32 (s,OR |SAB (0 , a0, a0)); + } +break ; + +case INDEX_op_bswap32_i32: +case INDEX_op_bswap32_i64: +/* Stolen from gcc's builtin_bswap32 */ + a1 = args[1 ]; + a0 = args[0 ] == a1 ?0 : args[0 ]; + +/* a1 = args[1] # abcd */ +/* a0 = rotate_left (a1, 8) # bcda */ +tcg_out_rlw (s,RLWINM , a0, a1,8 ,0 ,31 );
tcg/ppc64/tcg-target.c static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, · line 1708
Quoted 10 lines
tcg/ppc64/tcg-target.c +10
-0+/* a0 = (a0 & ~0xff000000) | ((a1 r<< 24) & 0xff000000) # dcda */ +tcg_out_rlw (s,RLWIMI , a0, a1,24 ,0 ,7 ); +/* a0 = (a0 & ~0x0000ff00) | ((a1 r<< 24) & 0x0000ff00) # dcba */ +tcg_out_rlw (s,RLWIMI , a0, a1,24 ,16 ,23 ); + +if (!a0) { +tcg_out_mov (s,TCG_TYPE_I64 , args[0 ], a0); + } +break ; +
It would be a lot easier to read if you had use TCG_REG_R0 instead of 0, especially for the if.
Quoted 20 lines
tcg/ppc64/tcg-target.c +5
-0default :tcg_dump_ops (s);tcg_abort ();@@ -1774,6 +1812,11 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_setcond_i32, { "r", "r", "ri" } }, { INDEX_op_setcond_i64, { "r", "r", "ri" } }, + { INDEX_op_bswap16_i32, { "r", "r" } }, + { INDEX_op_bswap16_i64, { "r", "r" } }, + { INDEX_op_bswap32_i32, { "r", "r" } }, + { INDEX_op_bswap32_i64, { "r", "r" } }, + { -1 }, };tcg/ppc64/tcg-target.h +0
-0
tcg/ppc64/tcg-target.h typedef enum { · line 75
Quoted 9 lines
tcg/ppc64/tcg-target.h +6
-0@@ -67,15 +67,19 @@ typedef enum {#define TCG_TARGET_STACK_ALIGN 16#define TCG_TARGET_CALL_STACK_OFFSET 48 +/* optional instructions automatically implemented */ +#define TCG_TARGET_HAS_ext8u_i32 0/* andi */ +#define TCG_TARGET_HAS_ext16u_i32 0 +#define TCG_TARGET_HAS_ext8u_i64 0 +#define TCG_TARGET_HAS_ext16u_i64 0 +
That should go in the andi patches, not this one.
Quoted 18 lines
tcg/ppc64/tcg-target.h +2
-4/* optional instructions */ #define TCG_TARGET_HAS_div_i32 1#define TCG_TARGET_HAS_rot_i32 1#define TCG_TARGET_HAS_ext8s_i32 1#define TCG_TARGET_HAS_ext16s_i32 1 -#define TCG_TARGET_HAS_ext8u_i32 0 -#define TCG_TARGET_HAS_ext16u_i32 0 -#define TCG_TARGET_HAS_bswap16_i32 0 -#define TCG_TARGET_HAS_bswap32_i32 0 +#define TCG_TARGET_HAS_bswap16_i32 1 +#define TCG_TARGET_HAS_bswap32_i32 1#define TCG_TARGET_HAS_not_i32 1#define TCG_TARGET_HAS_neg_i32 1#define TCG_TARGET_HAS_andc_i32 0@@ -95,11 +99,9 @@ typedef enum {#define TCG_TARGET_HAS_ext8s_i64 1#define TCG_TARGET_HAS_ext16s_i64 1#define TCG_TARGET_HAS_ext32s_i64 1
tcg/ppc64/tcg-target.h typedef enum { · line 107
Quoted 10 lines
tcg/ppc64/tcg-target.h +2
-4-#define TCG_TARGET_HAS_ext8u_i64 0 -#define TCG_TARGET_HAS_ext16u_i64 0#define TCG_TARGET_HAS_ext32u_i64 1 -#define TCG_TARGET_HAS_bswap16_i64 0 -#define TCG_TARGET_HAS_bswap32_i64 0 +#define TCG_TARGET_HAS_bswap16_i64 1 +#define TCG_TARGET_HAS_bswap32_i64 1#define TCG_TARGET_HAS_bswap64_i64 0#define TCG_TARGET_HAS_not_i64 1#define TCG_TARGET_HAS_neg_i64 1
Besides that the patch looks fine.
Signature · 2 lines
Aurelien Jarno GPG: 1024D/F1BCDB73 [email protected] http://www.aurel32.net
PATCH v3 17/27 tcg-ppc64: Implement bswap64
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 35 +++++++++++++++++++++++++++++++++++ tcg/ppc64/tcg-target.h | 2 +- 2 files changed, 36 insertions(+), 1 deletion(-)
tcg/ppc64/tcg-target.c +35 -0
@@ -1706,6 +1706,40 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, }break ; +case INDEX_op_bswap64_i64: + a0 = args[0 ], a1 = args[1 ], a2 =0 ; +if (a0 == a1) { + a0 =0 ; + a2 = a1; + } + +/* a1 = # abcd efgh */ +/* a0 = rl32(a1, 8) # 0000 fghe */ +tcg_out_rlw (s,RLWINM , a0, a1,8 ,0 ,31 ); +/* a0 = dep(a0, rl32(a1, 24), 0xff000000) # 0000 hghe */ +tcg_out_rlw (s,RLWIMI , a0, a1,24 ,0 ,7 ); +/* a0 = dep(a0, rl32(a1, 24), 0x0000ff00) # 0000 hgfe */ +tcg_out_rlw (s,RLWIMI , a0, a1,24 ,16 ,23 ); + +/* a0 = rl64(a0, 32) # hgfe 0000 */ +/* a2 = rl64(a1, 32) # efgh abcd */ +tcg_out_rld (s,RLDICL , a0, a0,32 ,0 ); +tcg_out_rld (s,RLDICL , a2, a1,32 ,0 ); + +/* a0 = dep(a0, rl32(a2, 8), 0xffffffff) # hgfe bcda */ +tcg_out_rlw (s,RLWIMI , a0, a2,8 ,0 ,31 ); +/* a0 = dep(a0, rl32(a2, 24), 0xff000000) # hgfe dcda */ +tcg_out_rlw (s,RLWIMI , a0, a2,24 ,0 ,7 ); +/* a0 = dep(a0, rl32(a2, 24), 0x0000ff00) # hgfe dcba */ +tcg_out_rlw (s,RLWIMI , a0, a2,24 ,16 ,23 ); + +if (a0 ==0 ) { +tcg_out_mov (s,TCG_TYPE_I64 , args[0 ], a0); +/* Revert the source rotate that we performed above. */ +tcg_out_rld (s,RLDICL , a1, a1,32 ,0 ); + } +break ; +default :tcg_dump_ops (s);tcg_abort ();
@@ -1816,6 +1850,7 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_bswap16_i64, { "r", "r" } }, { INDEX_op_bswap32_i32, { "r", "r" } }, { INDEX_op_bswap32_i64, { "r", "r" } }, + { INDEX_op_bswap64_i64, { "r", "r" } }, { -1 }, };
tcg/ppc64/tcg-target.h +1 -1
@@ -102,7 +102,7 @@ typedef enum {#define TCG_TARGET_HAS_ext32u_i64 1#define TCG_TARGET_HAS_bswap16_i64 1#define TCG_TARGET_HAS_bswap32_i64 1 -#define TCG_TARGET_HAS_bswap64_i64 0 +#define TCG_TARGET_HAS_bswap64_i64 1#define TCG_TARGET_HAS_not_i64 1#define TCG_TARGET_HAS_neg_i64 1#define TCG_TARGET_HAS_andc_i64 0
Signature · 1 line
1.8.1.4
Re: [Qemu-devel] [PATCH v3 17/27] tcg-ppc64: Implement bswap64
Quoted 2 lines · Richard Henderson
Reviewed-by Aurelien Jarno <[email protected]>Signed-off-by Richard Henderson <[email protected]>
Is this faster than a load/store with std/ldbrx?
Alex
Quoted 74 lines
---
tcg/ppc64/tcg-target.c| 35 +++++++++++++++++++++++++++++++++++tcg/ppc64/tcg-target.h| 2 +-2 files changed, 36 insertions(+), 1 deletion(-)tcg/ppc64/tcg-target.c +35
-0@@ -1706,6 +1706,40 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, }break ; +case INDEX_op_bswap64_i64: + a0 = args[0 ], a1 = args[1 ], a2 =0 ; +if (a0 == a1) { + a0 =0 ; + a2 = a1; + } + +/* a1 = # abcd efgh */ +/* a0 = rl32(a1, 8) # 0000 fghe */ +tcg_out_rlw (s,RLWINM , a0, a1,8 ,0 ,31 ); +/* a0 = dep(a0, rl32(a1, 24), 0xff000000) # 0000 hghe */ +tcg_out_rlw (s,RLWIMI , a0, a1,24 ,0 ,7 ); +/* a0 = dep(a0, rl32(a1, 24), 0x0000ff00) # 0000 hgfe */ +tcg_out_rlw (s,RLWIMI , a0, a1,24 ,16 ,23 ); + +/* a0 = rl64(a0, 32) # hgfe 0000 */ +/* a2 = rl64(a1, 32) # efgh abcd */ +tcg_out_rld (s,RLDICL , a0, a0,32 ,0 ); +tcg_out_rld (s,RLDICL , a2, a1,32 ,0 ); + +/* a0 = dep(a0, rl32(a2, 8), 0xffffffff) # hgfe bcda */ +tcg_out_rlw (s,RLWIMI , a0, a2,8 ,0 ,31 ); +/* a0 = dep(a0, rl32(a2, 24), 0xff000000) # hgfe dcda */ +tcg_out_rlw (s,RLWIMI , a0, a2,24 ,0 ,7 ); +/* a0 = dep(a0, rl32(a2, 24), 0x0000ff00) # hgfe dcba */ +tcg_out_rlw (s,RLWIMI , a0, a2,24 ,16 ,23 ); + +if (a0 ==0 ) { +tcg_out_mov (s,TCG_TYPE_I64 , args[0 ], a0); +/* Revert the source rotate that we performed above. */ +tcg_out_rld (s,RLDICL , a1, a1,32 ,0 ); + } +break ; +default :tcg_dump_ops (s);tcg_abort ();@@ -1816,6 +1850,7 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_bswap16_i64, { "r", "r" } }, { INDEX_op_bswap32_i32, { "r", "r" } }, { INDEX_op_bswap32_i64, { "r", "r" } }, + { INDEX_op_bswap64_i64, { "r", "r" } }, { -1 }, };tcg/ppc64/tcg-target.h +1
-1@@ -102,7 +102,7 @@ typedef enum {#define TCG_TARGET_HAS_ext32u_i64 1#define TCG_TARGET_HAS_bswap16_i64 1#define TCG_TARGET_HAS_bswap32_i64 1 -#define TCG_TARGET_HAS_bswap64_i64 0 +#define TCG_TARGET_HAS_bswap64_i64 1#define TCG_TARGET_HAS_not_i64 1#define TCG_TARGET_HAS_neg_i64 1#define TCG_TARGET_HAS_andc_i64 0Signature · 1 line
1.8.1.4
Re: [Qemu-devel] [PATCH v3 17/27] tcg-ppc64: Implement bswap64
Quoted 1 line · Alexander Graf
Is this faster than a load/store with std/ldbrx?
Hmm. Almost certainly not. And since we've got stack space allocated for function calls, we've got scratch space to do it in.
Probably similar for bswap32 too, eh?
I'll do a tiny bit o benchmarking for power7.
r~
Re: [Qemu-devel] [PATCH v3 17/27] tcg-ppc64: Implement bswap64
Am 02.04.2013 um 15:44 schrieb Richard Henderson <[email protected]>:
Quoted 7 lines · Alexander Graf
Quoted 1 line
Is this faster than a load/store with std/ldbrx?
Hmm. Almost certainly not. And since we've got stack space allocated for function calls, we've got scratch space to do it in.
Probably similar for bswap32 too, eh?
Depends - memory load/store doesn't come for free and bswap32 is quite short.
Quoted 2 lines
I'll do a tiny bit o benchmarking for power7.
Cool, thanks a bunch :)
Alex
Quoted 3 lines
r~
Re: [Qemu-devel] [PATCH v3 17/27] tcg-ppc64: Implement bswap64
Quoted 6 lines · Alexander Graf
Quoted 6 lines · Alexander Graf
Quoted 1 line
Is this faster than a load/store with std/ldbrx?
Hmm. Almost certainly not. And since we've got stack space allocated for function calls, we've got scratch space to do it in.
Quoted 8 lines
Quoted 1 line
Probably similar for bswap32 too, eh?
Depends - memory load/store doesn't come for free and bswap32 is quite short.
Quoted 2 lines
I'll do a tiny bit o benchmarking for power7.
Cool, thanks a bunch :)
Heh. "Almost certainly not" indeed. Unless I've made some silly mistake, going through memory stalls badly. No store buffer forwarding on power7?
With the following test case, time reports:
f1 2.967s
f2 8.930s
f3 7.071s
f4 7.166s
And note that f4 is a normal store/load pair, trying to determine what the store buffer forwarding delay might be.
r~
static long __attribute__((noinline)) f1(long x, long *mem) {
long r, t;
asm volatile (
"rlwinm %0,%1,8,0,31\n\ rlwimi %0,%1,24,0,7\n\ rlwimi %0,%1,24,16,23\n\ rldicl %0,%0,32,0\n\ rldicl %2,%1,32,0\n\ rlwimi %0,%2,8,0,31\n\ rlwimi %0,%2,24,0,7\n\ rlwimi %0,%2,24,16,23" : "=&r"(r), "=r"(t) : "r"(x));
return r;
}
static long __attribute__((noinline)) f2(long x, long *mem)
{
long r, t;
asm volatile ("std %1,0(%2); ldbrx %0,0,%2" : "=r"(r) : "r"(x), "b"(mem)); return r;
}
static long __attribute__((noinline)) f3(long x, long *mem)
{
long r, t;
asm volatile ("stdbrx %1,0,%2; ld %0,0(%2)" : "=r"(r) : "r"(x), "b"(mem)); return r;
}
static long __attribute__((noinline)) f4(long x, long *mem)
{
long r, t;
asm volatile ("std %1,0(%2); ld %0,0(%2)" : "=r"(r) : "r"(x), "b"(mem)); return r;
}
#define D1(x,y) x##y
#define DO(x) D1(f,x)
int main()
{
long tmp, i;
for (i = 0; i < 1000000000; ++i)
DO(N)(i, &tmp);
return 0;
}Re: [Qemu-devel] [PATCH v3 17/27] tcg-ppc64: Implement bswap64
Quoted 21 lines · Richard Henderson
Quoted 6 lines · Alexander Graf
Quoted 6 lines · Alexander Graf
Quoted 1 line
Is this faster than a load/store with std/ldbrx?
Hmm. Almost certainly not. And since we've got stack space allocated for function calls, we've got scratch space to do it in.
Quoted 8 lines
Quoted 1 line
Probably similar for bswap32 too, eh?
Depends - memory load/store doesn't come for free and bswap32 is quite short.
Quoted 2 lines
I'll do a tiny bit o benchmarking for power7.
Cool, thanks a bunch :)
Heh. "Almost certainly not" indeed. Unless I've made some silly mistake, going through memory stalls badly. No store buffer forwarding on power7?
With the following test case, time reports:
Quoted 7 lines
f1 2.967s f2 8.930s f3 7.071s f4 7.166sAnd note that f4 is a normal store/load pair, trying to determine what the store buffer forwarding delay might be.
Yeah, doesn't look like it makes any sense at all to do a load/store cycle then. What a shame :).
Keep in mind that this tests icache hot cycles. However, you might get bad icache penalties due to the long bswap64 sequence. So all the memory latency you see here might also affect the instruction stream when it gets executed. But then again we only care about performance of cache hot sequences in the first place....
Alex
PATCH v3 18/27 tcg-ppc64: Implement compound logicals
In reply to Richard HendersonMostly copied from the ppc32 port.
--- tcg/ppc64/tcg-target.c | 34 ++++++++++++++++++++++++++++++++++ tcg/ppc64/tcg-target.h | 20 ++++++++++---------- 2 files changed, 44 insertions(+), 10 deletions(-)
tcg/ppc64/tcg-target.c +34 -0
@@ -385,6 +385,10 @@ static int tcg_target_const_match (tcg_target_long val,#define NOR XO31(124)#define CNTLZW XO31( 26)#define CNTLZD XO31( 58) +#define ANDC XO31( 60) +#define ORC XO31(412) +#define EQV XO31(284) +#define NAND XO31(476)#define MULLD XO31(233)#define MULHD XO31( 73)
@@ -1415,6 +1419,26 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,tcg_out32 (s,XOR |SAB (a1, a0, a2)); }break ; +case INDEX_op_andc_i32: +case INDEX_op_andc_i64: +tcg_out32 (s,ANDC |SAB (args[1 ], args[0 ], args[2 ])); +break ; +case INDEX_op_orc_i32: +case INDEX_op_orc_i64: +tcg_out32 (s,ORC |SAB (args[1 ], args[0 ], args[2 ])); +break ; +case INDEX_op_eqv_i32: +case INDEX_op_eqv_i64: +tcg_out32 (s,EQV |SAB (args[1 ], args[0 ], args[2 ])); +break ; +case INDEX_op_nand_i32: +case INDEX_op_nand_i64: +tcg_out32 (s,NAND |SAB (args[1 ], args[0 ], args[2 ])); +break ; +case INDEX_op_nor_i32: +case INDEX_op_nor_i64: +tcg_out32 (s,NOR |SAB (args[1 ], args[0 ], args[2 ])); +break ;case INDEX_op_mul_i32:if (const_args[2 ]) {
@@ -1788,6 +1812,11 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_and_i32, { "r", "r", "ri" } }, { INDEX_op_or_i32, { "r", "r", "ri" } }, { INDEX_op_xor_i32, { "r", "r", "ri" } }, + { INDEX_op_andc_i32, { "r", "r", "r" } }, + { INDEX_op_orc_i32, { "r", "r", "r" } }, + { INDEX_op_eqv_i32, { "r", "r", "r" } }, + { INDEX_op_nand_i32, { "r", "r", "r" } }, + { INDEX_op_nor_i32, { "r", "r", "r" } }, { INDEX_op_shl_i32, { "r", "r", "ri" } }, { INDEX_op_shr_i32, { "r", "r", "ri" } },
@@ -1806,6 +1835,11 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_and_i64, { "r", "r", "rU" } }, { INDEX_op_or_i64, { "r", "r", "rU" } }, { INDEX_op_xor_i64, { "r", "r", "rU" } }, + { INDEX_op_andc_i64, { "r", "r", "r" } }, + { INDEX_op_orc_i64, { "r", "r", "r" } }, + { INDEX_op_eqv_i64, { "r", "r", "r" } }, + { INDEX_op_nand_i64, { "r", "r", "r" } }, + { INDEX_op_nor_i64, { "r", "r", "r" } }, { INDEX_op_shl_i64, { "r", "r", "ri" } }, { INDEX_op_shr_i64, { "r", "r", "ri" } },
tcg/ppc64/tcg-target.h +10 -10
@@ -82,11 +82,11 @@ typedef enum {#define TCG_TARGET_HAS_bswap32_i32 1#define TCG_TARGET_HAS_not_i32 1#define TCG_TARGET_HAS_neg_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_nand_i32 0 -#define TCG_TARGET_HAS_nor_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 1 +#define TCG_TARGET_HAS_nor_i32 1#define TCG_TARGET_HAS_deposit_i32 0#define TCG_TARGET_HAS_movcond_i32 0#define TCG_TARGET_HAS_add2_i32 0
@@ -105,11 +105,11 @@ typedef enum {#define TCG_TARGET_HAS_bswap64_i64 1#define TCG_TARGET_HAS_not_i64 1#define TCG_TARGET_HAS_neg_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_nand_i64 0 -#define TCG_TARGET_HAS_nor_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 1 +#define TCG_TARGET_HAS_nor_i64 1#define TCG_TARGET_HAS_deposit_i64 0#define TCG_TARGET_HAS_movcond_i64 0#define TCG_TARGET_HAS_add2_i64 0
Signature · 1 line
1.8.1.4
PATCH v3 19/27 tcg-ppc64: Handle constant inputs for some compound logicals
In reply to Richard HendersonSince we have special code to handle and/or/xor with a constant, apply the same to andc/orc/eqv with a constant.
--- tcg/ppc64/tcg-target.c | 44 ++++++++++++++++++++++++++++++++++---------- 1 file changed, 34 insertions(+), 10 deletions(-)
tcg/ppc64/tcg-target.c +34 -10
@@ -1388,17 +1388,19 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_and_i32: + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ];if (const_args[2 ]) { -tcg_out_andi32 (s, args[0 ], args[1 ], args[2 ]); +tcg_out_andi32 (s, a0, a1, a2); }else { -tcg_out32 (s,AND |SAB (args[1 ], args[0 ], args[2 ])); +tcg_out32 (s,AND |SAB (a1, a0, a2)); }break ;case INDEX_op_and_i64: + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ];if (const_args[2 ]) { -tcg_out_andi64 (s, args[0 ], args[1 ], args[2 ]); +tcg_out_andi64 (s, a0, a1, a2); }else { -tcg_out32 (s,AND |SAB (args[1 ], args[0 ], args[2 ])); +tcg_out32 (s,AND |SAB (a1, a0, a2)); }break ;case INDEX_op_or_i64:
@@ -1420,14 +1422,36 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, }break ;case INDEX_op_andc_i32: + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ]; +if (const_args[2 ]) { +tcg_out_andi32 (s, a0, a1, ~a2); + }else { +tcg_out32 (s,ANDC |SAB (a1, a0, a2)); + } +break ;case INDEX_op_andc_i64: -tcg_out32 (s,ANDC |SAB (args[1 ], args[0 ], args[2 ])); + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ]; +if (const_args[2 ]) { +tcg_out_andi64 (s, a0, a1, ~a2); + }else { +tcg_out32 (s,ANDC |SAB (a1, a0, a2)); + }break ;case INDEX_op_orc_i32: +if (const_args[2 ]) { +tcg_out_ori32 (s, args[0 ], args[1 ], ~args[2 ]); +break ; + } +/* FALLTHRU */ case INDEX_op_orc_i64:tcg_out32 (s,ORC |SAB (args[1 ], args[0 ], args[2 ]));break ;case INDEX_op_eqv_i32: +if (const_args[2 ]) { +tcg_out_xori32 (s, args[0 ], args[1 ], ~args[2 ]); +break ; + } +/* FALLTHRU */ case INDEX_op_eqv_i64:tcg_out32 (s,EQV |SAB (args[1 ], args[0 ], args[2 ]));break ;
@@ -1812,9 +1836,9 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_and_i32, { "r", "r", "ri" } }, { INDEX_op_or_i32, { "r", "r", "ri" } }, { INDEX_op_xor_i32, { "r", "r", "ri" } }, - { INDEX_op_andc_i32, { "r", "r", "r" } }, - { INDEX_op_orc_i32, { "r", "r", "r" } }, - { INDEX_op_eqv_i32, { "r", "r", "r" } }, + { INDEX_op_andc_i32, { "r", "r", "ri" } }, + { INDEX_op_orc_i32, { "r", "r", "ri" } }, + { INDEX_op_eqv_i32, { "r", "r", "ri" } }, { INDEX_op_nand_i32, { "r", "r", "r" } }, { INDEX_op_nor_i32, { "r", "r", "r" } },
@@ -1832,10 +1856,10 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_add_i64, { "r", "r", "rTU" } }, { INDEX_op_sub_i64, { "r", "rI", "rTU" } }, - { INDEX_op_and_i64, { "r", "r", "rU" } }, + { INDEX_op_and_i64, { "r", "r", "ri" } }, { INDEX_op_or_i64, { "r", "r", "rU" } }, { INDEX_op_xor_i64, { "r", "r", "rU" } }, - { INDEX_op_andc_i64, { "r", "r", "r" } }, + { INDEX_op_andc_i64, { "r", "r", "ri" } }, { INDEX_op_orc_i64, { "r", "r", "r" } }, { INDEX_op_eqv_i64, { "r", "r", "r" } }, { INDEX_op_nand_i64, { "r", "r", "r" } },
Signature · 1 line
1.8.1.4
PATCH v3 20/27 tcg-ppc64: Implement deposit
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 12 ++++++++++++ tcg/ppc64/tcg-target.h | 4 ++-- 2 files changed, 14 insertions(+), 2 deletions(-)
tcg/ppc64/tcg-target.c +12 -0
@@ -1788,6 +1788,15 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, }break ; +case INDEX_op_deposit_i32: +tcg_out_rlw (s,RLWIMI , args[0 ], args[2 ], args[3 ], +32 - args[3 ] - args[4 ],31 - args[3 ]); +break ; +case INDEX_op_deposit_i64: +tcg_out_rld (s,RLDIMI , args[0 ], args[2 ], args[3 ], +64 - args[3 ] - args[4 ]); +break ; +default :tcg_dump_ops (s);tcg_abort ();
@@ -1910,6 +1919,9 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_bswap32_i64, { "r", "r" } }, { INDEX_op_bswap64_i64, { "r", "r" } }, + { INDEX_op_deposit_i32, { "r", "0", "r" } }, + { INDEX_op_deposit_i64, { "r", "0", "r" } }, + { -1 }, };
tcg/ppc64/tcg-target.h +2 -2
@@ -87,7 +87,7 @@ typedef enum {#define TCG_TARGET_HAS_eqv_i32 1#define TCG_TARGET_HAS_nand_i32 1#define TCG_TARGET_HAS_nor_i32 1 -#define TCG_TARGET_HAS_deposit_i32 0 +#define TCG_TARGET_HAS_deposit_i32 1#define TCG_TARGET_HAS_movcond_i32 0#define TCG_TARGET_HAS_add2_i32 0#define TCG_TARGET_HAS_sub2_i32 0
@@ -110,7 +110,7 @@ typedef enum {#define TCG_TARGET_HAS_eqv_i64 1#define TCG_TARGET_HAS_nand_i64 1#define TCG_TARGET_HAS_nor_i64 1 -#define TCG_TARGET_HAS_deposit_i64 0 +#define TCG_TARGET_HAS_deposit_i64 1#define TCG_TARGET_HAS_movcond_i64 0#define TCG_TARGET_HAS_add2_i64 0#define TCG_TARGET_HAS_sub2_i64 0
Signature · 1 line
1.8.1.4
PATCH v3 21/27 tcg-ppc64: Use I constraint for mul
In reply to Richard HendersonThe mul_i32 pattern was loading non-16-bit constants into a register, when we can get the middle-end to do that for us. The mul_i64 pattern was not considering that MULLI takes 64-bit inputs.
--- tcg/ppc64/tcg-target.c | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-)
tcg/ppc64/tcg-target.c +12 -12
@@ -1465,17 +1465,12 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_mul_i32: + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ];if (const_args[2 ]) { -if (args[2 ] == (int16_t ) args[2 ]) -tcg_out32 (s,MULLI |RT (args[0 ]) |RA (args[1 ]) - | (args[2 ] &0xffff )); -else { -tcg_out_movi (s,TCG_TYPE_I32 ,0 , args[2 ]); -tcg_out32 (s,MULLW |TAB (args[0 ], args[1 ],0 )); - } +tcg_out32 (s,MULLI |TAI (a0, a1, a2)); + }else { +tcg_out32 (s,MULLW |TAB (a0, a1, a2)); } -else -tcg_out32 (s,MULLW |TAB (args[0 ], args[1 ], args[2 ]));break ;case INDEX_op_div_i32:
@@ -1634,7 +1629,12 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_mul_i64: -tcg_out32 (s,MULLD |TAB (args[0 ], args[1 ], args[2 ])); + a0 = args[0 ], a1 = args[1 ], a2 = args[2 ]; +if (const_args[2 ]) { +tcg_out32 (s,MULLI |TAI (a0, a1, a2)); + }else { +tcg_out32 (s,MULLD |TAB (a0, a1, a2)); + }break ;case INDEX_op_div_i64:tcg_out32 (s,DIVD |TAB (args[0 ], args[1 ], args[2 ]));
@@ -1836,7 +1836,7 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_ld32s_i64, { "r", "r" } }, { INDEX_op_add_i32, { "r", "r", "ri" } }, - { INDEX_op_mul_i32, { "r", "r", "ri" } }, + { INDEX_op_mul_i32, { "r", "r", "rI" } }, { INDEX_op_div_i32, { "r", "r", "r" } }, { INDEX_op_divu_i32, { "r", "r", "r" } }, { INDEX_op_rem_i32, { "r", "r", "r" } },
@@ -1880,7 +1880,7 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_rotl_i64, { "r", "r", "ri" } }, { INDEX_op_rotr_i64, { "r", "r", "ri" } }, - { INDEX_op_mul_i64, { "r", "r", "r" } }, + { INDEX_op_mul_i64, { "r", "r", "rI" } }, { INDEX_op_div_i64, { "r", "r", "r" } }, { INDEX_op_divu_i64, { "r", "r", "r" } }, { INDEX_op_rem_i64, { "r", "r", "r" } },
Signature · 1 line
1.8.1.4
PATCH v3 22/27 tcg-ppc64: Use TCGType throughout compares
In reply to Richard HendersonThe optimization/bug being fixed is that tcg_out_cmp was not applying the right type to loading a constant, in the case it can't be implemented directly. Rather than recomputing the TCGType enum from the arch64 bool, pass around the original TCGType throughout.
--- tcg/ppc64/tcg-target.c | 38 +++++++++++++++++++------------------- 1 file changed, 19 insertions(+), 19 deletions(-)
tcg/ppc64/tcg-target.c +19 -19
@@ -1053,8 +1053,8 @@ static void tcg_out_st (TCGContext *s, TCGType type, TCGReg arg, TCGReg arg1,tcg_out_ldsta (s, arg, arg1, arg2,STD ,STDX ); } -static void tcg_out_cmp (TCGContext *s,int cond,TCGArg arg1,TCGArg arg2, -int const_arg2,int cr,int arch64) +static void tcg_out_cmp (TCGContext *s,int cond,TCGArg arg1,TCGArg arg2, +int const_arg2,int cr,TCGType type) {int imm;uint32_t op;
@@ -1111,19 +1111,17 @@ static void tcg_out_cmp (TCGContext *s, int cond, TCGArg arg1, TCGArg arg2,default :tcg_abort (); } - op |=BF (cr) | (arch64 <<21 ); + op |=BF (cr) | ((type ==TCG_TYPE_I64 ) <<21 ); -if (imm) -tcg_out32 (s, op |RA (arg1) | (arg2 &0xffff )); -else { +if (imm) { +tcg_out32 (s, op |RA (arg1) | (arg2 &0xffff )); + }else {if (const_arg2) { -tcg_out_movi (s,TCG_TYPE_I64 ,0 , arg2); -tcg_out32 (s, op |RA (arg1) |RB (0 )); +tcg_out_movi (s, type,0 , arg2); + arg2 =0 ; } -else -tcg_out32 (s, op |RA (arg1) |RB (arg2)); +tcg_out32 (s, op |RA (arg1) |RB (arg2)); } - } static void tcg_out_setcond (TCGContext *s, TCGType type, TCGCond cond,
@@ -1217,7 +1215,7 @@ static void tcg_out_setcond (TCGContext *s, TCGType type, TCGCond cond, sh =31 ; crop =CRNOR |BT (7 ,CR_EQ ) |BA (7 ,CR_GT ) |BB (7 ,CR_GT ); crtest: -tcg_out_cmp (s, cond, arg1, arg2, const_arg2,7 , type ==TCG_TYPE_I64 ); +tcg_out_cmp (s, cond, arg1, arg2, const_arg2,7 , type);if (crop)tcg_out32 (s, crop);tcg_out32 (s,MFCR |RT (0 ));tcg_out_rlw (s,RLWINM , arg0,0 , sh,31 ,31 );
@@ -1243,12 +1241,12 @@ static void tcg_out_bc (TCGContext *s, int bc, int label_index) } } -static void tcg_out_brcond (TCGContext *s,TCGCond cond, -TCGArg arg1,TCGArg arg2,int const_arg2, -int label_index,int arch64) +static void tcg_out_brcond (TCGContext *s,TCGCond cond, +TCGArg arg1,TCGArg arg2,int const_arg2, +int label_index,TCGType type) { -tcg_out_cmp (s, cond, arg1, arg2, const_arg2,7 , arch64); -tcg_out_bc (s, tcg_to_bc[cond], label_index); +tcg_out_cmp (s, cond, arg1, arg2, const_arg2,7 , type); +tcg_out_bc (s, tcg_to_bc[cond], label_index); }void ppc_tb_set_jmp_target (unsigned long jmp_addr,unsigned long addr)
@@ -1532,11 +1530,13 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args,break ;case INDEX_op_brcond_i32: -tcg_out_brcond (s, args[2 ], args[0 ], args[1 ], const_args[1 ], args[3 ],0 ); +tcg_out_brcond (s, args[2 ], args[0 ], args[1 ], const_args[1 ], + args[3 ],TCG_TYPE_I32 );break ;case INDEX_op_brcond_i64: -tcg_out_brcond (s, args[2 ], args[0 ], args[1 ], const_args[1 ], args[3 ],1 ); +tcg_out_brcond (s, args[2 ], args[0 ], args[1 ], const_args[1 ], + args[3 ],TCG_TYPE_I64 );break ;case INDEX_op_neg_i32:
Signature · 1 line
1.8.1.4
PATCH v3 23/27 tcg-ppc64: Rewrite setcond
In reply to Richard HendersonNever use MFCR, as the latency is really high. Even MFOCRF, at half the latency of MFCR, isn't as fast as we can do with carry tricks.
The ADDIC+SUBFE trick only works for word-sized operands, as we need carry-out from bit 63. So for ppc64 we must extend 32-bit inputs.
Use ISEL if available.
--- tcg/ppc64/tcg-target.c | 264 +++++++++++++++++++++++++++++++++---------------- 1 file changed, 181 insertions(+), 83 deletions(-)
tcg/ppc64/tcg-target.c +181 -83
@@ -45,6 +45,7 @@ static uint8_t *tb_ret_addr;#endif #define HAVE_ISA_2_06 0 +#define HAVE_ISEL 0#ifdef CONFIG_USE_GUEST_BASE #define TCG_GUEST_BASE_REG 30
@@ -389,6 +390,7 @@ static int tcg_target_const_match (tcg_target_long val,#define ORC XO31(412)#define EQV XO31(284)#define NAND XO31(476) +#define ISEL XO31( 15)#define MULLD XO31(233)#define MULHD XO31( 73)
@@ -443,6 +445,7 @@ static int tcg_target_const_match (tcg_target_long val,#define BT (n, c) (((c)+((n)*4))<<21)#define BA (n, c) (((c)+((n)*4))<<16)#define BB (n, c) (((c)+((n)*4))<<11) +#define BC_ (n, c) (((c)+((n)*4))<<6)#define BO_COND_TRUE BO (12)#define BO_COND_FALSE BO ( 4)
@@ -468,6 +471,20 @@ static const uint32_t tcg_to_bc[] = { [TCG_COND_GTU] = BC | BI (7, CR_GT) | BO_COND_TRUE, }; +/* The low bit here is set if the RA and RB fields must be inverted. */ +static const uint32_t tcg_to_isel[] = { + [TCG_COND_EQ ] =ISEL |BC_ (7 ,CR_EQ ), + [TCG_COND_NE ] =ISEL |BC_ (7 ,CR_EQ ) |1 , + [TCG_COND_LT ] =ISEL |BC_ (7 ,CR_LT ), + [TCG_COND_GE ] =ISEL |BC_ (7 ,CR_LT ) |1 , + [TCG_COND_LE ] =ISEL |BC_ (7 ,CR_GT ) |1 , + [TCG_COND_GT ] =ISEL |BC_ (7 ,CR_GT ), + [TCG_COND_LTU ] =ISEL |BC_ (7 ,CR_LT ), + [TCG_COND_GEU ] =ISEL |BC_ (7 ,CR_LT ) |1 , + [TCG_COND_LEU ] =ISEL |BC_ (7 ,CR_GT ) |1 , + [TCG_COND_GTU ] =ISEL |BC_ (7 ,CR_GT ), +}; +static inline void tcg_out_mov (TCGContext *s,TCGType type,TCGReg ret,TCGReg arg) {
@@ -1124,105 +1141,186 @@ static void tcg_out_cmp(TCGContext *s, int cond, TCGArg arg1, TCGArg arg2, } } -static void tcg_out_setcond (TCGContext *s,TCGType type,TCGCond cond, -TCGArg arg0,TCGArg arg1,TCGArg arg2, -int const_arg2) +static void tcg_out_setcond_eq0 (TCGContext *s,TCGType type, +TCGReg dst,TCGReg src) { -int crop, sh, arg; +tcg_out32 (s, (type ==TCG_TYPE_I64 ?CNTLZD :CNTLZW ) |RS (dst) |RA (src)); +tcg_out_shri64 (s, dst, dst, type ==TCG_TYPE_I64 ?6 :5 ); +} -switch (cond) { -case TCG_COND_EQ : -if (const_arg2) { -if (!arg2) { - arg = arg1; - } -else { - arg =0 ; -if ((uint16_t ) arg2 == arg2) { -tcg_out32 (s,XORI |SAI (arg1,0 , arg2)); - } -else { -tcg_out_movi (s, type,0 , arg2); -tcg_out32 (s,XOR |SAB (arg1,0 ,0 )); - } - } - } -else { - arg =0 ; -tcg_out32 (s,XOR |SAB (arg1,0 , arg2)); - } +static void tcg_out_setcond_ne0 (TCGContext *s,TCGReg dst,TCGReg src) +{ +/* X != 0 implies X + -1 generates a carry. Extra addition +trickery means: R = X-1 + ~X + C = X-1 + (-X+1) + C = C. */ +if (dst != src) { +tcg_out32 (s,ADDIC |TAI (dst, src,-1 )); +tcg_out32 (s,SUBFE |TAB (dst, dst, src)); + }else { +tcg_out32 (s,ADDIC |TAI (0 , src,-1 )); +tcg_out32 (s,SUBFE |TAB (dst,0 , src)); + } +} -if (type ==TCG_TYPE_I64 ) { -tcg_out32 (s,CNTLZD |RS (arg) |RA (0 )); -tcg_out_rld (s,RLDICL , arg0,0 ,58 ,6 ); - } -else { -tcg_out32 (s,CNTLZW |RS (arg) |RA (0 )); -tcg_out_rlw (s,RLWINM , arg0,0 ,27 ,5 ,31 ); +static int tcg_gen_setcond_xor (TCGContext *s,TCGReg arg1,TCGArg arg2, +bool const_arg2) +{ +if (const_arg2) { +if ((uint32_t )arg2 == arg2) { +tcg_out_xori32 (s,0 , arg1, arg2); + }else { +tcg_out_movi (s,TCG_TYPE_I64 ,0 , arg2); +tcg_out32 (s,XOR |SAB (arg1,0 ,0 )); } -break ; + }else { +tcg_out32 (s,XOR |SAB (arg1,0 , arg2)); + } +return 0 ; +} -case TCG_COND_NE : -if (const_arg2) { -if (!arg2) { - arg = arg1; - } -else { - arg =0 ; -if ((uint16_t ) arg2 == arg2) { -tcg_out32 (s,XORI |SAI (arg1,0 , arg2)); - }else { -tcg_out_movi (s, type,0 , arg2); -tcg_out32 (s,XOR |SAB (arg1,0 ,0 )); - } +static void tcg_out_setcond (TCGContext *s,TCGType type,TCGCond cond, +TCGArg arg0,TCGArg arg1,TCGArg arg2, +int const_arg2) +{ +bool invert, swap; + +/* Handle common and trivial cases before handling anything else. */ +if (arg2 ==0 ) { +switch (cond) { +case TCG_COND_EQ : +tcg_out_setcond_eq0 (s, type, arg0, arg1); +return ; +case TCG_COND_NE : +if (type ==TCG_TYPE_I32 ) { +tcg_out_ext32u (s,0 , arg1); + arg1 =0 ; } +tcg_out_setcond_ne0 (s, arg0, arg1); +return ; +case TCG_COND_GE : +tcg_out32 (s,NOR |SAB (arg1, arg0, arg1)); + arg1 = arg0; +/* FALLTHRU */ +case TCG_COND_LT : +/* Extract the sign bit. */ +tcg_out_rld (s,RLDICL , arg0, arg1, + type ==TCG_TYPE_I64 ?1 :33 ,63 ); +return ; +default : +break ; } -else { - arg =0 ; -tcg_out32 (s,XOR |SAB (arg1,0 , arg2)); - } + } -if (arg == arg1 && arg1 == arg0) { -tcg_out32 (s,ADDIC |TAI (0 , arg,-1 )); -tcg_out32 (s,SUBFE |TAB (arg0,0 , arg)); +/* If we have ISEL, we can implement everything with 3 or 4 insns. +All other cases below are also at least 3 insns, so speed up the +code generator by not considering them and always using ISEL. */ +if (HAVE_ISEL ) { +int isel, tab; + +tcg_out_cmp (s, cond, arg1, arg2, const_arg2,7 , type); + + isel = tcg_to_isel[cond]; + +tcg_out_movi (s, type, arg0,1 ); +if (isel &1 ) { +/* arg0 = (bc ? 0 : 1) */ + tab =TAB (arg0,0 , arg0); + isel &= ~1 ; + }else { +/* arg0 = (bc ? 1 : 0) */ +tcg_out_movi (s, type,0 ,0 ); + tab =TAB (arg0, arg0,0 ); } -else { -tcg_out32 (s,ADDIC |TAI (arg0, arg,-1 )); -tcg_out32 (s,SUBFE |TAB (arg0, arg0, arg)); +tcg_out32 (s, isel | tab); +return ; + } + + invert = swap = false; +switch (cond) { +case TCG_COND_EQ : +/* Given that we can ignore the high bits in setcond_eq0, make +sure we go through the XORIS+XORI path in setcond_xor. */ +if (type ==TCG_TYPE_I32 ) { + arg2 = (uint32_t )arg2; } -break ; + arg1 =tcg_gen_setcond_xor (s, arg1, arg2, const_arg2); +tcg_out_setcond_eq0 (s, type, arg0, arg1); +return ; -case TCG_COND_GT : -case TCG_COND_GTU : - sh =30 ; - crop =0 ; - goto crtest; +case TCG_COND_NE : +if (type ==TCG_TYPE_I32 ) { +tcg_out32 (s,EXTSW |RS (arg1) |RA (0 )); + arg1 =0 ; +if (const_arg2) { + arg2 = (int32_t )arg2; + }else { +tcg_out32 (s,EXTSW |RS (arg2) |RA (arg0)); + arg2 = arg0; + } + } + arg1 =tcg_gen_setcond_xor (s, arg1, arg2, const_arg2); +tcg_out_setcond_ne0 (s, arg0, arg1); +return ; -case TCG_COND_LT : +/* Reduce the ordered relations to GT. */ case TCG_COND_LTU : - sh =29 ; - crop =0 ; - goto crtest; - -case TCG_COND_GE : -case TCG_COND_GEU : - sh =31 ; - crop =CRNOR |BT (7 ,CR_EQ ) |BA (7 ,CR_LT ) |BB (7 ,CR_LT ); - goto crtest; - -case TCG_COND_LE : +case TCG_COND_LT : + swap = true; +break ;case TCG_COND_LEU : - sh =31 ; - crop =CRNOR |BT (7 ,CR_EQ ) |BA (7 ,CR_GT ) |BB (7 ,CR_GT ); - crtest: -tcg_out_cmp (s, cond, arg1, arg2, const_arg2,7 , type); -if (crop)tcg_out32 (s, crop); -tcg_out32 (s,MFCR |RT (0 )); -tcg_out_rlw (s,RLWINM , arg0,0 , sh,31 ,31 ); +case TCG_COND_LE : + invert = true; +break ; +case TCG_COND_GEU : +case TCG_COND_GE : + swap = true, invert = true; +break ; +case TCG_COND_GTU : +case TCG_COND_GT :break ; -default : -tcg_abort (); +tcg_abort (); + } + +/* In 64-bit mode, carry-out is only generated from the 63-rd bit. +Thus 32-bit inputs must be sign-extended. */ +if (type ==TCG_TYPE_I32 ) { +tcg_out32 (s,EXTSW |RS (arg1) |RA (0 )); + arg1 =0 ; +if (const_arg2) { + arg2 = (int32_t )arg2; + }else { +tcg_out32 (s,EXTSW |RS (arg2) |RA (arg0)); + arg2 = arg0; + } + } +if (swap) { +if (const_arg2) { +tcg_out_movi (s,TCG_TYPE_I64 ,0 , arg2); + arg2 = arg1; + arg1 =0 ; + const_arg2 = false; + }else { +int t = arg1; + arg1 = arg2; + arg2 = t; + } + } + +/* X > Y implies Y - X generates a carry-out. This works for both +signed and unsigned comparisons. */ +if (const_arg2 && (int16_t )arg2 == arg2) { +tcg_out32 (s,SUBFIC |TAI (0 , arg1, arg2)); + }else { +if (const_arg2) { +tcg_out_movi (s, type,0 , arg2); + arg2 =0 ; + } +tcg_out32 (s,SUBFC |TAB (0 , arg1, arg2)); + } +tcg_out_movi (s, type, arg0,0 ); +tcg_out32 (s,ADDE |TAB (arg0, arg0, arg0)); +if (invert) { +tcg_out32 (s,XORI |SAI (arg0, arg0,1 )); } }
Signature · 1 line
1.8.1.4
PATCH v3 24/27 tcg-ppc64: Implement movcond
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 59 ++++++++++++++++++++++++++++++++++++++++++++++++++ tcg/ppc64/tcg-target.h | 4 ++-- 2 files changed, 61 insertions(+), 2 deletions(-)
tcg/ppc64/tcg-target.c +59 -0
@@ -1347,6 +1347,54 @@ static void tcg_out_brcond(TCGContext *s, TCGCond cond,tcg_out_bc (s, tcg_to_bc[cond], label_index); } +static void tcg_out_movcond (TCGContext *s,TCGType type,TCGCond cond, +TCGArg dest,TCGArg c1,TCGArg c2,TCGArg v1, +TCGArg v2,bool const_c2) +{ +/* If for some reason both inputs are zero, don't produce bad code. */ +if (v1 ==0 && v2 ==0 ) { +tcg_out_movi (s, type, dest,0 ); +return ; + } + +tcg_out_cmp (s, cond, c1, c2, const_c2,7 , type); + +if (HAVE_ISEL ) { +int isel = tcg_to_isel[cond]; + +/* Swap the V operands if the operation indicates inversion. */ +if (isel &1 ) { +int t = v1; + v1 = v2; + v2 = t; + isel &= ~1 ; + } +/* V1 == 0 is handled by isel; V2 == 0 must be handled by hand. */ +if (v2 ==0 ) { +tcg_out_movi (s, type,0 ,0 ); + } +tcg_out32 (s, isel |TAB (dest, v1, v2)); + }else { +if (dest == v2) { + cond =tcg_invert_cond (cond); + v2 = v1; + }else if (dest != v1) { +if (v1 ==0 ) { +tcg_out_movi (s, type, dest,0 ); + }else { +tcg_out_mov (s, type, dest, v1); + } + } +/* Branch forward over one insn */ +tcg_out32 (s, tcg_to_bc[cond] |8 ); +if (v2 ==0 ) { +tcg_out_movi (s, type, dest,0 ); + }else { +tcg_out_mov (s, type, dest, v2); + } + } +} +void ppc_tb_set_jmp_target (unsigned long jmp_addr,unsigned long addr) {TCGContext s;
@@ -1895,6 +1943,15 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, 64 - args[3] - args[4]);break ; +case INDEX_op_movcond_i32: +tcg_out_movcond (s,TCG_TYPE_I32 , args[5 ], args[0 ], args[1 ], args[2 ], + args[3 ], args[4 ], const_args[2 ]); +break ; +case INDEX_op_movcond_i64: +tcg_out_movcond (s,TCG_TYPE_I64 , args[5 ], args[0 ], args[1 ], args[2 ], + args[3 ], args[4 ], const_args[2 ]); +break ; +default :tcg_dump_ops (s);tcg_abort ();
@@ -2010,6 +2067,8 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_setcond_i32, { "r", "r", "ri" } }, { INDEX_op_setcond_i64, { "r", "r", "ri" } }, + { INDEX_op_movcond_i32, { "r", "r", "ri", "rZ", "rZ" } }, + { INDEX_op_movcond_i64, { "r", "r", "ri", "rZ", "rZ" } }, { INDEX_op_bswap16_i32, { "r", "r" } }, { INDEX_op_bswap16_i64, { "r", "r" } },
tcg/ppc64/tcg-target.h +2 -2
@@ -88,7 +88,7 @@ typedef enum {#define TCG_TARGET_HAS_nand_i32 1#define TCG_TARGET_HAS_nor_i32 1#define TCG_TARGET_HAS_deposit_i32 1 -#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
@@ -111,7 +111,7 @@ typedef enum {#define TCG_TARGET_HAS_nand_i64 1#define TCG_TARGET_HAS_nor_i64 1#define TCG_TARGET_HAS_deposit_i64 1 -#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
1.8.1.4
PATCH v3 25/27 tcg-ppc64: Use getauxval for ISA detection
In reply to Richard HendersonGlibc 2.16 includes an easy way to get feature bits previously buried in /proc or the program startup auxiliary vector. Use it.
--- configure | 18 ++++++++++++++++++ tcg/ppc64/tcg-target.c | 16 +++++++++++++++- 2 files changed, 33 insertions(+), 1 deletion(-)
configure +18 -0
@@ -3252,6 +3252,20 @@ if compile_prog "" "" ; then int128=yes fi +######################################## +# check if getauxval is available. + +getauxval=no +cat > $TMPC << EOF +#include <sys/auxv.h> +int main(void) { + return getauxval(AT_HWCAP) == 0; +} +EOF +if compile_prog "" "" ; then + getauxval=yes +fi + ########################################## # End of CC checks # After here, no more $cc or $ld runs
@@ -3809,6 +3823,10 @@ if test "$int128" = "yes" ; then echo "CONFIG_INT128=y" >> $config_host_mak fi +if test "$getauxval" = "yes" ; then + echo "CONFIG_GETAUXVAL=y" >> $config_host_mak +fi + if test "$glusterfs" = "yes" ; then echo "CONFIG_GLUSTERFS=y" >> $config_host_mak fi
tcg/ppc64/tcg-target.c +15 -1
@@ -44,8 +44,15 @@ static uint8_t *tb_ret_addr;#define GUEST_BASE 0#endif +#ifdef CONFIG_GETAUXVAL +#include <sys/auxv.h> +static bool have_isa_2_06; +#define HAVE_ISA_2_06 have_isa_2_06 +#define HAVE_ISEL have_isa_2_06 +#else #define HAVE_ISA_2_06 0 -#define HAVE_ISEL 0 +#define HAVE_ISA_ISEL 0 +#endif #ifdef CONFIG_USE_GUEST_BASE #define TCG_GUEST_BASE_REG 30
@@ -2084,6 +2091,13 @@ static const TCGTargetOpDef ppc_op_defs[] = {static void tcg_target_init (TCGContext *s) { +#ifdef CONFIG_GETAUXVAL +unsigned long hwcap =getauxval (AT_HWCAP ); +if (hwcap &PPC_FEATURE_ARCH_2_06 ) { + have_isa_2_06 = true; + } +#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 ); tcg_regset_set32 (tcg_target_call_clobber_regs, 0,
Signature · 1 line
1.8.1.4
Re: [Qemu-devel] [PATCH v3 25/27] tcg-ppc64: Use getauxval for ISA detection
Quoted 70 lines · Richard Henderson
Glibc 2.16 includes an easy way to get feature bits previously buried in /proc or the program startup auxiliary vector. Use it.
Signed-off-by Richard Henderson <[email protected]>--- configure | 18 ++++++++++++++++++ tcg/ppc64/tcg-target.c | 16 +++++++++++++++- 2 files changed, 33 insertions(+), 1 deletion(-)configure +18
-0@@ -3252,6 +3252,20 @@ if compile_prog "" "" ; then int128=yes fi +######################################## +# check if getauxval is available. + +getauxval=no +cat > $TMPC << EOF +#include <sys/auxv.h> +int main(void) { + return getauxval(AT_HWCAP) == 0; +} +EOF +if compile_prog "" "" ; then + getauxval=yes +fi + ########################################## # End of CC checks # After here, no more $cc or $ld runs@@ -3809,6 +3823,10 @@ if test "$int128" = "yes" ; then echo "CONFIG_INT128=y" >> $config_host_mak fi +if test "$getauxval" = "yes" ; then + echo "CONFIG_GETAUXVAL=y" >> $config_host_mak +fi + if test "$glusterfs" = "yes" ; then echo "CONFIG_GLUSTERFS=y" >> $config_host_mak fitcg/ppc64/tcg-target.c +8
-1@@ -44,8 +44,15 @@ static uint8_t *tb_ret_addr;#define GUEST_BASE 0#endif +#ifdef CONFIG_GETAUXVAL +#include <sys/auxv.h> +static bool have_isa_2_06; +#define HAVE_ISA_2_06 have_isa_2_06 +#define HAVE_ISEL have_isa_2_06 +#else #define HAVE_ISA_2_06 0 -#define HAVE_ISEL 0 +#define HAVE_ISA_ISEL 0 +#endif #ifdef CONFIG_USE_GUEST_BASE #define TCG_GUEST_BASE_REG 30@@ -2084,6 +2091,13 @@ static const TCGTargetOpDef ppc_op_defs[] = {static void tcg_target_init (TCGContext *s) {
tcg/ppc64/tcg-target.c static const TCGTargetOpDef ppc_op_defs[] = { · line 2103
Quoted 10 lines
tcg/ppc64/tcg-target.c +7
-0+#ifdef CONFIG_GETAUXVAL +unsigned long hwcap =getauxval (AT_HWCAP ); +if (hwcap &PPC_FEATURE_ARCH_2_06 ) { + have_isa_2_06 = true; + } +#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 ); tcg_regset_set32 (tcg_target_call_clobber_regs, 0,
Signature · 2 lines
Aurelien Jarno GPG: 1024D/F1BCDB73 [email protected] http://www.aurel32.net
PATCH v3 26/27 tcg-ppc64: Implement add2/sub2_i64
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++++ tcg/ppc64/tcg-target.h | 4 ++-- 2 files changed, 58 insertions(+), 2 deletions(-)
tcg/ppc64/tcg-target.c +56 -0
@@ -27,6 +27,7 @@#define TCG_CT_CONST_S32 0x400#define TCG_CT_CONST_U32 0x800#define TCG_CT_CONST_ZERO 0x1000 +#define TCG_CT_CONST_MONE 0x1000static uint8_t *tb_ret_addr;
@@ -262,6 +263,9 @@ static int target_parse_constraint (TCGArgConstraint *ct, const char **pct_str)case 'J' : ct->ct |=TCG_CT_CONST_U16 ;break ; +case 'M' : + ct->ct |=TCG_CT_CONST_MONE ; +break ;case 'T' : ct->ct |=TCG_CT_CONST_S32 ;break ;
@@ -296,6 +300,8 @@ static int tcg_target_const_match (tcg_target_long val,return 1 ; }else if ((ct &TCG_CT_CONST_ZERO ) && val ==0 ) {return 1 ; + }else if ((ct &TCG_CT_CONST_MONE ) && val ==-1 ) { +return 1 ; }return 0 ; }
@@ -366,11 +372,15 @@ static int tcg_target_const_match (tcg_target_long val,#define EXTSW XO31(986)#define ADD XO31(266)#define ADDE XO31(138) +#define ADDME XO31(234) +#define ADDZE XO31(202)#define ADDC XO31( 10)#define AND XO31( 28)#define SUBF XO31( 40)#define SUBFC XO31( 8)#define SUBFE XO31(136) +#define SUBFME XO31(232) +#define SUBFZE XO31(200)#define OR XO31(444)#define XOR XO31(316)#define MULLW XO31(235)
@@ -1959,6 +1969,49 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, args[3], args[4], const_args[2]);break ; +case INDEX_op_add2_i64: +/* Note that the CA bit is defined based on the word size of the +environment. So in 64-bit mode it's always carry-out of bit 63. +The fallback code using deposit works just as well for 32-bit. */ + a0 = args[0 ], a1 = args[1 ]; +if (a0 == args[4 ] || (!const_args[5 ] && a0 == args[5 ])) { + a0 =TCG_REG_R0 ; + } +if (const_args[3 ]) { +tcg_out32 (s,ADDIC |TAI (a0, args[2 ], args[3 ])); + }else { +tcg_out32 (s,ADDC |TAB (a0, args[2 ], args[3 ])); + } +if (const_args[5 ]) { +tcg_out32 (s, (args[5 ] ?ADDME :ADDZE ) |RT (a1) |RA (args[4 ])); + }else { +tcg_out32 (s,ADDE |TAB (a1, args[4 ], args[5 ])); + } +if (a0 != args[0 ]) { +tcg_out_mov (s,TCG_TYPE_I64 , args[0 ], a0); + } +break ; + +case INDEX_op_sub2_i64: + a0 = args[0 ], a1 = args[1 ]; +if (a0 == args[5 ] || (!const_args[4 ] && a0 == args[4 ])) { + a0 =TCG_REG_R0 ; + } +if (const_args[2 ]) { +tcg_out32 (s,SUBFIC |TAI (a0, args[3 ], args[2 ])); + }else { +tcg_out32 (s,SUBFC |TAB (a0, args[3 ], args[2 ])); + } +if (const_args[4 ]) { +tcg_out32 (s, (args[4 ] ?SUBFME :SUBFZE ) |RT (a1) |RA (args[5 ])); + }else { +tcg_out32 (s,SUBFE |TAB (a1, args[5 ], args[4 ])); + } +if (a0 != args[0 ]) { +tcg_out_mov (s,TCG_TYPE_I64 , args[0 ], a0); + } +break ; +default :tcg_dump_ops (s);tcg_abort ();
@@ -2086,6 +2139,9 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_deposit_i32, { "r", "0", "r" } }, { INDEX_op_deposit_i64, { "r", "0", "r" } }, + { INDEX_op_add2_i64, { "r", "r", "r", "rI", "r", "rZM" } }, + { INDEX_op_sub2_i64, { "r", "r", "rI", "r", "rZM", "r" } }, + { -1 }, };
tcg/ppc64/tcg-target.h +2 -2
@@ -112,8 +112,8 @@ typedef enum {#define TCG_TARGET_HAS_nor_i64 1#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
Signature · 1 line
1.8.1.4
PATCH v3 27/27 tcg-ppc64: Implement mulu2/muls2_i64
In reply to Richard Henderson--- tcg/ppc64/tcg-target.c | 27 +++++++++++++++++++++++++++ tcg/ppc64/tcg-target.h | 4 ++-- 2 files changed, 29 insertions(+), 2 deletions(-)
tcg/ppc64/tcg-target.c +27 -0
@@ -2012,6 +2012,31 @@ static void tcg_out_op (TCGContext *s, TCGOpcode opc, const TCGArg *args, }break ; +case INDEX_op_mulu2_i64: +case INDEX_op_muls2_i64: + { +int oph = (opc == INDEX_op_mulu2_i64 ?MULHDU :MULHD ); +TCGReg outl = args[0 ], outh = args[1 ]; + a0 = args[2 ], a1 = args[3 ]; + +if (outl == a0 || outl == a1) { +if (outh == a0 || outh == a1) { + outl =TCG_REG_R0 ; + }else { +tcg_out32 (s, oph |TAB (outh, a0, a1)); + oph =0 ; + } + } +tcg_out32 (s,MULLD |TAB (outl, a0, a1)); +if (oph !=0 ) { +tcg_out32 (s, oph |TAB (outh, a0, a1)); + } +if (outl != args[0 ]) { +tcg_out_mov (s,TCG_TYPE_I64 , args[0 ], outl); + } + } +break ; +default :tcg_dump_ops (s);tcg_abort ();
@@ -2141,6 +2166,8 @@ static const TCGTargetOpDef ppc_op_defs[] = { { INDEX_op_add2_i64, { "r", "r", "r", "rI", "r", "rZM" } }, { INDEX_op_sub2_i64, { "r", "r", "rI", "r", "rZM", "r" } }, + { INDEX_op_muls2_i64, { "r", "r", "r", "r" } }, + { INDEX_op_mulu2_i64, { "r", "r", "r", "r" } }, { -1 }, };
tcg/ppc64/tcg-target.h +2 -2
@@ -114,8 +114,8 @@ typedef enum {#define TCG_TARGET_HAS_movcond_i64 1#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_mulu2_i64 1 +#define TCG_TARGET_HAS_muls2_i64 1#define TCG_AREG0 TCG_REG_R27
Signature · 1 line
1.8.1.4
Quoted 6 lines · Richard Henderson
Changes v2-v3:
* Incorporated feedback from Aurelien's review.Tree updated at
git://github.com/rth7680/qemu.git tcg-ppc64
ppc/ppc64-on-ppc64:
Alex
Quoted 4 lines · Alexander Graf
Quoted 3 lines · Richard Henderson
Changes v2-v3:
* Incorporated feedback from Aurelien's review.
Quoted 8 lines
Quoted 3 lines
Tree updated at
git://github.com/rth7680/qemu.gittcg-ppc64ppc/ppc64-on-ppc64:
Tested-by Alexander Graf <[email protected]>
How have you tested that? I tried to boot x86-64 and mipsel guests in system mode, and they both fail. The symptoms are slightly different before and after your patches, but I can't say if it is better or worse. What is sure is that this TCG target is broken.
Signature · 2 lines
Aurelien Jarno GPG: 1024D/F1BCDB73 [email protected] http://www.aurel32.net
Quoted 8 lines · Aurelien Jarno
Quoted 7 lines · Alexander Graf
Quoted 6 lines · Richard Henderson
Changes v2-v3:
* Incorporated feedback from Aurelien's review.Tree updated at
git://github.com/rth7680/qemu.git tcg-ppc64
Quoted 8 lines
Quoted 4 lines
ppc/ppc64-on-ppc64:
Tested-by Alexander Graf<[email protected]>How have you tested that? I tried to boot x86-64 and mipsel guests in system mode, and they both fail. The symptoms are slightly different before and after your patches, but I can't say if it is better or worse. What is sure is that this TCG target is broken.
I ran my usual test suite: http://awardnbg.ath.cx/results/247-alex/clay/job_report.html
However, I only ever executed ppc guests on a G5 ppc64 host. Sorry if that wasn't clear from the "ppc/ppc64-on-ppc64" note above.
Alex
Quoted 40 lines · Richard Henderson
Changes v2-v3:
* Incorporated feedback from Aurelien's review.Tree updated at
git://github.com/rth7680/qemu.gittcg-ppc64r~
Richard Henderson (27):
disas: Disassemble all ppc insns for the host
tcg-ppc64: Use TCGReg everywhere
tcg-ppc64: Introduce and usetcg_out_rlw
tcg-ppc64: Introduce and usetcg_out_ext32u
tcg-ppc64: Introduce and usetcg_out_shli64
tcg-ppc64: Introduce and usetcg_out_shri64
tcg-ppc64: Cleanuptcg_out_movi
tcg-ppc64: Introduce and use TAI and SAI
tcg-ppc64: Rearrange integer constant constraints
tcg-ppc64: Improve constant add and sub ops.
tcg-ppc64: Tidy or and xor patterns.
tcg-ppc64: Improveand_i32with constant
tcg-ppc64: Improveand_i64with constant
tcg-ppc64: Streamline qemu_ld/st insn selection
tcg-ppc64: Implement rotates
tcg-ppc64: Implement bswap16 and bswap32
tcg-ppc64: Implement bswap64
tcg-ppc64: Implement compound logicals
tcg-ppc64: Handle constant inputs for some compound logicals
tcg-ppc64: Implement deposit
tcg-ppc64: Use I constraint for mul
tcg-ppc64: Use TCGType throughout compares
tcg-ppc64: Rewrite setcond
tcg-ppc64: Implement movcond
tcg-ppc64: Use getauxval for ISA detection
tcg-ppc64: Implement add2/sub2_i64
tcg-ppc64: Implement mulu2/muls2_i64
Quoted 6 lines
configure | 18 + disas.c | 1 + tcg/ppc64/tcg-target.c | 1395 +++++++++++++++++++++++++++++++++--------------- tcg/ppc64/tcg-target.h | 60 ++- 4 files changed, 1000 insertions(+), 474 deletions(-)
Please find the review in the individual patches. Except minor nitpicks in patch 16, the series looks fine for me.
I have tested it with arm, i386, mips(64), mips(64)el, powerpc, sh4 and sparc guests without problem. OTOH it didn't work for me for x86_64 guests, which hangs in the boot process (both Linux and FreeBSD). i386 guests using qemu-system-x86_64 works fine though.
So it's a huge improvement considering it was not working for half of these guests before. Thanks.
Signature · 2 lines
Aurelien Jarno GPG: 1024D/F1BCDB73 [email protected] http://www.aurel32.net