Alpha improvments, round 3
PATCH 2/5 target-alpha: Implement fp branch/cmov inline.
In reply to Richard HendersonThe old fcmov implementation had a typo:
- tcg_gen_mov_i64(cpu_fir[rc], cpu_fir[ra]);
which moved the condition, not the second source, to the destination.
But it's also easy to implement the simplified fp comparison inline.
--- target-alpha/helper.h | 7 -- target-alpha/op_helper.c | 31 ------- target-alpha/translate.c | 197 ++++++++++++++++++++++++++-------------------- 3 files changed, 110 insertions(+), 125 deletions(-)
target-alpha/helper.h +0 -7
@@ -77,13 +77,6 @@ DEF_HELPER_2(cmpgeq, i64, i64, i64) DEF_HELPER_2(cmpgle, i64, i64, i64) DEF_HELPER_2(cmpglt, i64, i64, i64) -DEF_HELPER_1(cmpfeq, i64, i64) -DEF_HELPER_1(cmpfne, i64, i64) -DEF_HELPER_1(cmpflt, i64, i64) -DEF_HELPER_1(cmpfle, i64, i64) -DEF_HELPER_1(cmpfgt, i64, i64) -DEF_HELPER_1(cmpfge, i64, i64) - DEF_HELPER_2(cpys, i64, i64, i64) DEF_HELPER_2(cpysn, i64, i64, i64)DEF_HELPER_2 (cpyse, i64, i64, i64)
target-alpha/op_helper.c +0 -31
@@ -884,37 +884,6 @@ uint64_t helper_cmpglt(uint64_t a, uint64_t b)return 0 ; } -uint64_t helper_cmpfeq (uint64_t a) -{ -return !(a &0x7FFFFFFFFFFFFFFFULL ); -} - -uint64_t helper_cmpfne (uint64_t a) -{ -return (a &0x7FFFFFFFFFFFFFFFULL ); -} - -uint64_t helper_cmpflt (uint64_t a) -{ -return (a &0x8000000000000000ULL ) && (a &0x7FFFFFFFFFFFFFFFULL ); -} - -uint64_t helper_cmpfle (uint64_t a) -{ -return (a &0x8000000000000000ULL ) || !(a &0x7FFFFFFFFFFFFFFFULL ); -} - -uint64_t helper_cmpfgt (uint64_t a) -{ -return !(a &0x8000000000000000ULL ) && (a &0x7FFFFFFFFFFFFFFFULL ); -} - -uint64_t helper_cmpfge (uint64_t a) -{ -return !(a &0x8000000000000000ULL ) || !(a &0x7FFFFFFFFFFFFFFFULL ); -} - -/* Floating point format conversion */ uint64_t helper_cvtts (uint64_t a) {
target-alpha/translate.c +111 -88
277 lines not on this page · raw message
Signature · 1 line
1.6.5.2
PATCH 3/5 target-alpha: Fix generic ctz64.
In reply to Richard Henderson--- host-utils.h | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
host-utils.h +1 -1
@@ -164,7 +164,7 @@ static inline int ctz64(uint64_t val) {#if QEMU_GNUC_PREREQ (3 ,4 )if (val) -return __builtin_ctz (val); +return __builtin_ctzll (val);else return 64 ;#else
Signature · 1 line
1.6.5.2
cover Alpha improvments, round 3
The first is the IMB patch, adjusted for Aurelien's comments.
The third is a bug in a generic header; I'm not sure what the proper tagging is for that sort of thing.
The rest are new bugs found in the Alpha backend via the gcc testsuite. We're down to a handfull of failures over native there. I hope to post an approach to the rounding mode issue shortly.
r~
Richard Henderson (5):
target-alpha: Add placeholders for missing userspace PALcalls.
target-alpha: Implement fp branch/cmov inline.
target-alpha: Fix generic ctz64.
target-alpha: Fix cvtlq.
target-alpha: Fix float32_to_s vs zero exponent.
host-utils.h | 2 +- hw/alpha_palcode.c | 32 +++++++- target-alpha/helper.h | 7 -- target-alpha/op_helper.c | 79 ++++++++---------- target-alpha/translate.c | 197 ++++++++++++++++++++++++++-------------------- 5 files changed, 176 insertions(+), 141 deletions(-)
PATCH 1/5 target-alpha: Add placeholders for missing userspace PALcalls.
--- hw/alpha_palcode.c | 32 ++++++++++++++++++++++++++++++-- 1 files changed, 30 insertions(+), 2 deletions(-)
hw/alpha_palcode.c +30 -2
@@ -1061,6 +1061,16 @@ void call_pal (CPUState *env, int palcode)target_long ret;switch (palcode) { +case 0x80 : +/* BPT */ +qemu_log ("BPT\n" ); +/* FIXME: Sends SIGTRAP, si_code=TRAP_BRKPT. */ +exit (1 ); +case 0x81 : +/* BUGCHK */ +qemu_log ("BUGCHK\n" ); +/* FIXME: Sends SIGTRAP, si_code=SI_FAULT. */ +exit (1 );case 0x83 :/* CALLSYS */ qemu_log ("CALLSYS n " TARGET_FMT_ld"\n" , env->ir[0 ]);
@@ -1075,6 +1085,14 @@ void call_pal (CPUState *env, int palcode) env->ir[IR_V0 ] = -ret; }break ; +case 0x86 : +/* IMB */ +qemu_log ("IMB\n" ); +/* ??? We can probably elide the code using page_unprotect that is +checking for self-modifying code. Instead we could simply call +tb_flush here. Until we work out the changes required to turn +off the extra write protection, this can be a no-op. */ +break ;case 0x9E :/* RDUNIQUE */ qemu_log ("RDUNIQUE: " TARGET_FMT_lx"\n" , env->unique);
@@ -1085,9 +1103,19 @@ void call_pal (CPUState *env, int palcode)qemu_log ("WRUNIQUE: " TARGET_FMT_lx"\n" , env->ir[IR_A0 ]);/* Handled in the translator for usermode. */ abort (); +case 0xAA : +/* GENTRAP */ +qemu_log ("GENTRAP: " TARGET_FMT_lx"\n" , env->ir[IR_A0 ]); +/* FIXME: This is supposed to send a signal: +SIGFPE: +GEN_INTOVF, GEN_INTDIV, GEN_FLTOVF, GEN_FLTDIV, +GEN_FLTUND, GEN_FLTINV, GEN_FLTINE, GEN_ROPRAND +SIGTRAP: +others +with various settings of si_code. */ +exit (1 );default : -qemu_log ("%s: unhandled palcode %02x\n" , - __func__, palcode); +qemu_log ("%s: unhandled palcode %02x\n" , __func__, palcode);exit (1 ); } }
Signature · 1 line
1.6.5.2
PATCH 4/5 target-alpha: Fix cvtlq.
In reply to Richard HendersonWe were missing the 0xc0000000 mask, leading to incorrect results.
--- target-alpha/op_helper.c | 4 +++- 1 files changed, 3 insertions(+), 1 deletions(-)
target-alpha/op_helper.c +3 -1
@@ -954,7 +954,9 @@ uint64_t helper_cvtqg (uint64_t a)uint64_t helper_cvtlq (uint64_t a) { -return (int64_t )((int32_t )((a >>32 ) | ((a >>29 ) &0x3FFFFFFF ))); +int32_t lo = a >>29 ; +int32_t hi = a >>32 ; +return (lo &0x3FFFFFFF ) | (hi &0xc0000000 ); }static inline uint64_t __helper_cvtql (uint64_t a,int s,int v)
Signature · 1 line
1.6.5.2
PATCH 5/5 target-alpha: Fix float32_to_s vs zero exponent.
In reply to Richard HendersonThere was a bug in float32_to_s that incorrectly mapped a zero exponent to 0x38. This meant 0.0f != 0. At the same time, fix a generic type punning bug in helper_memory_to_s and helper_s_to_memory.
--- target-alpha/op_helper.c | 44 ++++++++++++++++++++++++++++++++------------ 1 files changed, 32 insertions(+), 12 deletions(-)
target-alpha/op_helper.c +32 -12
@@ -625,37 +625,57 @@ uint64_t helper_sqrtg (uint64_t a)/* S floating (single) */ + +/* Taken from linux/arch/alpha/kernel/traps.c, s_mem_to_reg. */ +static inline uint64_t float32_to_s_int (uint32_t fi) +{ +uint32_t frac = fi &0x7fffff ; +uint32_t sign = fi >>31 ; +uint32_t exp_msb = (fi >>30 ) &1 ; +uint32_t exp_low = (fi >>23 ) &0x7f ; +uint32_t exp; + + exp = (exp_msb <<10 ) | exp_low; +if (exp_msb) { +if (exp_low ==0x7f ) + exp =0x7ff ; + }else { +if (exp_low !=0x00 ) + exp |=0x380 ; + } + +return (((uint64_t )sign <<63 ) + | ((uint64_t )exp <<52 ) + | ((uint64_t )frac <<29 )); +} +static inline uint64_t float32_to_s (float32 fa) {CPU_FloatU a; -uint64_t r; - a.f = fa; +return float32_to_s_int (a.l); +} - r = (((uint64_t )(a.l &0xc0000000 )) <<32 ) | (((uint64_t )(a.l &0x3fffffff )) <<29 ); -if (((a.l &0x7f800000 ) !=0x7f800000 ) && (!(a.l &0x40000000 ))) - r |=0x7ll <<59 ; -return r; +static inline uint32_t s_to_float32_int (uint64_t a) +{ +return ((a >>32 ) &0xc0000000 ) | ((a >>29 ) &0x3fffffff ); }static inline float32 s_to_float32 (uint64_t a) {CPU_FloatU r; - r.l = ((a >>32 ) &0xc0000000 ) | ((a >>29 ) &0x3fffffff ); + r.l =s_to_float32_int (a);return r.f; }uint32_t helper_s_to_memory (uint64_t a) { -/* Memory format is the same as float32 */ -float32 fa =s_to_float32 (a); -return *(uint32_t *)(&fa); +return s_to_float32_int (a); }uint64_t helper_memory_to_s (uint32_t a) { -/* Memory format is the same as float32 */ -return float32_to_s (*(float32 *)(&a)); +return float32_to_s_int (a); }uint64_t helper_adds (uint64_t a,uint64_t b)
Signature · 1 line
1.6.5.2