X-Git-Url: http://wien.tomnetworks.com/gitweb/?a=blobdiff_plain;f=mono%2Fmini%2Fmini-ppc.c;h=70bf728c803c53b5add949c790d030bf3e76446d;hb=102197b11d6ce990d2bdc2325ed944a05e780e40;hp=936a9a1b6135b0c80d723bee28e9b46f785ee343;hpb=61262b44a9f1c3d8a225664aba891592f2fba840;p=mono.git diff --git a/mono/mini/mini-ppc.c b/mono/mini/mini-ppc.c index 936a9a1b613..70bf728c803 100644 --- a/mono/mini/mini-ppc.c +++ b/mono/mini/mini-ppc.c @@ -14,9 +14,10 @@ #include #include +#include #include "mini-ppc.h" -#ifdef __mono_ppc64__ +#ifdef TARGET_POWERPC64 #include "cpu-ppc64.h" #else #include "cpu-ppc.h" @@ -26,6 +27,9 @@ #ifdef __APPLE__ #include #endif +#ifdef __linux__ +#include +#endif #define FORCE_INDIR_CALL 1 @@ -38,6 +42,20 @@ enum { TLS_MODE_DARWIN_G5 }; +/* cpu_hw_caps contains the flags defined below */ +static int cpu_hw_caps = 0; +static int cachelinesize = 0; +static int cachelineinc = 0; +enum { + PPC_ICACHE_SNOOP = 1 << 0, + PPC_MULTIPLE_LS_UNITS = 1 << 1, + PPC_SMP_CAPABLE = 1 << 2, + PPC_ISA_2X = 1 << 3, + PPC_ISA_64 = 1 << 4, + PPC_MOVE_FPR_GPR = 1 << 5, + PPC_HW_CAP_END +}; + /* This mutex protects architecture specific caches */ #define mono_mini_arch_lock() EnterCriticalSection (&mini_arch_mutex) #define mono_mini_arch_unlock() LeaveCriticalSection (&mini_arch_mutex) @@ -46,7 +64,6 @@ static CRITICAL_SECTION mini_arch_mutex; int mono_exc_esp_offset = 0; static int tls_mode = TLS_MODE_DETECT; static int lmf_pthread_key = -1; -static int monothread_key = -1; static int monodomain_key = -1; static int @@ -61,14 +78,14 @@ offsets_from_pthread_key (guint32 key, int *offset2) #define emit_linuxthreads_tls(code,dreg,key) do {\ int off1, off2; \ off1 = offsets_from_pthread_key ((key), &off2); \ - ppc_load_reg ((code), (dreg), off1, ppc_r2); \ - ppc_load_reg ((code), (dreg), off2, (dreg)); \ + ppc_ldptr ((code), (dreg), off1, ppc_r2); \ + ppc_ldptr ((code), (dreg), off2, (dreg)); \ } while (0); #define emit_darwing5_tls(code,dreg,key) do {\ int off1 = 0x48 + key * sizeof (gpointer); \ ppc_mfspr ((code), (dreg), 104); \ - ppc_load_reg ((code), (dreg), off1, (dreg)); \ + ppc_ldptr ((code), (dreg), off1, (dreg)); \ } while (0); /* FIXME: ensure the sc call preserves all but r3 */ @@ -81,9 +98,28 @@ offsets_from_pthread_key (guint32 key, int *offset2) if ((dreg) != ppc_r3) ppc_mr ((code), ppc_r3, ppc_r11); \ } while (0); +#ifdef PPC_THREAD_PTR_REG +#define emit_nptl_tls(code,dreg,key) do { \ + int off1 = key; \ + int off2 = key >> 15; \ + if ((off2 == 0) || (off2 == -1)) { \ + ppc_ldptr ((code), (dreg), off1, PPC_THREAD_PTR_REG); \ + } else { \ + int off3 = (off2 + 1) > 1; \ + ppc_addis ((code), ppc_r11, PPC_THREAD_PTR_REG, off3); \ + ppc_ldptr ((code), (dreg), off1, ppc_r11); \ + } \ + } while (0); +#else +#define emit_nptl_tls(code,dreg,key) do { \ + g_assert_not_reached (); \ + } while (0) +#endif + #define emit_tls_access(code,dreg,key) do { \ switch (tls_mode) { \ case TLS_MODE_LTHREADS: emit_linuxthreads_tls(code,dreg,key); break; \ + case TLS_MODE_NPTL: emit_nptl_tls(code,dreg,key); break; \ case TLS_MODE_DARWIN_G5: emit_darwing5_tls(code,dreg,key); break; \ case TLS_MODE_DARWIN_G4: emit_darwing4_tls(code,dreg,key); break; \ default: g_assert_not_reached (); \ @@ -91,7 +127,7 @@ offsets_from_pthread_key (guint32 key, int *offset2) } while (0) #define MONO_EMIT_NEW_LOAD_R8(cfg,dr,addr) do { \ - MonoInst *inst; \ + MonoInst *inst; \ MONO_INST_NEW ((cfg), (inst), OP_R8CONST); \ inst->type = STACK_R8; \ inst->dreg = (dr); \ @@ -137,7 +173,7 @@ emit_memcpy (guint8 *code, int size, int dreg, int doffset, int sreg, int soffse { /* unrolled, use the counter in big */ if (size > sizeof (gpointer) * 5) { - int shifted = size >> MONO_PPC_32_64_CASE (2, 3); + long shifted = size / SIZEOF_VOID_P; guint8 *copy_loop_start, *copy_loop_jump; ppc_load (code, ppc_r0, shifted); @@ -146,8 +182,8 @@ emit_memcpy (guint8 *code, int size, int dreg, int doffset, int sreg, int soffse ppc_addi (code, ppc_r12, dreg, (doffset - sizeof (gpointer))); ppc_addi (code, ppc_r11, sreg, (soffset - sizeof (gpointer))); copy_loop_start = code; - ppc_load_reg_update (code, ppc_r0, sizeof (gpointer), ppc_r11); - ppc_store_reg_update (code, ppc_r0, sizeof (gpointer), ppc_r12); + ppc_ldptr_update (code, ppc_r0, (unsigned int)sizeof (gpointer), ppc_r11); + ppc_stptr_update (code, ppc_r0, (unsigned int)sizeof (gpointer), ppc_r12); copy_loop_jump = code; ppc_bc (code, PPC_BR_DEC_CTR_NONZERO, 0, 0); ppc_patch (copy_loop_jump, copy_loop_start); @@ -156,13 +192,39 @@ emit_memcpy (guint8 *code, int size, int dreg, int doffset, int sreg, int soffse dreg = ppc_r12; } #ifdef __mono_ppc64__ + /* the hardware has multiple load/store units and the move is long + enough to use more then one regiester, then use load/load/store/store + to execute 2 instructions per cycle. */ + if ((cpu_hw_caps & PPC_MULTIPLE_LS_UNITS) && (dreg != ppc_r12) && (sreg != ppc_r12)) { + while (size >= 16) { + ppc_ldptr (code, ppc_r0, soffset, sreg); + ppc_ldptr (code, ppc_r12, soffset+8, sreg); + ppc_stptr (code, ppc_r0, doffset, dreg); + ppc_stptr (code, ppc_r12, doffset+8, dreg); + size -= 16; + soffset += 16; + doffset += 16; + } + } while (size >= 8) { - ppc_load_reg (code, ppc_r0, soffset, sreg); - ppc_store_reg (code, ppc_r0, doffset, dreg); + ppc_ldr (code, ppc_r0, soffset, sreg); + ppc_str (code, ppc_r0, doffset, dreg); size -= 8; soffset += 8; doffset += 8; } +#else + if ((cpu_hw_caps & PPC_MULTIPLE_LS_UNITS) && (dreg != ppc_r12) && (sreg != ppc_r12)) { + while (size >= 8) { + ppc_lwz (code, ppc_r0, soffset, sreg); + ppc_lwz (code, ppc_r12, soffset+4, sreg); + ppc_stw (code, ppc_r0, doffset, dreg); + ppc_stw (code, ppc_r12, doffset+4, dreg); + size -= 8; + soffset += 8; + doffset += 8; + } + } #endif while (size >= 4) { ppc_lwz (code, ppc_r0, soffset, sreg); @@ -300,11 +362,12 @@ mono_ppc_is_direct_call_sequence (guint32 *code) } gpointer -mono_arch_get_vcall_slot (guint8 *code_ptr, gpointer *regs, int *displacement) +mono_arch_get_vcall_slot (guint8 *code_ptr, mgreg_t *regs, int *displacement) { char *o = NULL; int reg, offset = 0; guint32* code = (guint32*)code_ptr; + mgreg_t *r = (mgreg_t*)regs; *displacement = 0; @@ -337,13 +400,7 @@ mono_arch_get_vcall_slot (guint8 *code_ptr, gpointer *regs, int *displacement) reg = (*code >> 16) & 0x1f; g_assert (reg != ppc_r1); /*g_print ("patching reg is %d\n", reg);*/ - if (reg >= 13) { - MonoLMF *lmf = (MonoLMF*)((char*)regs + (14 * sizeof (double)) + (13 * sizeof (gpointer))); - /* saved in the MonoLMF structure */ - o = (gpointer)lmf->iregs [reg - 13]; - } else { - o = regs [reg]; - } + o = (gpointer)(gsize)r [reg]; break; } } @@ -351,83 +408,50 @@ mono_arch_get_vcall_slot (guint8 *code_ptr, gpointer *regs, int *displacement) return o; } -gpointer* -mono_arch_get_vcall_slot_addr (guint8 *code, gpointer *regs) -{ - gpointer vt; - int displacement; - vt = mono_arch_get_vcall_slot (code, regs, &displacement); - if (!vt) - return NULL; - return (gpointer*)((char*)vt + displacement); -} - #define MAX_ARCH_DELEGATE_PARAMS 7 -gpointer -mono_arch_get_delegate_invoke_impl (MonoMethodSignature *sig, gboolean has_target) +static gpointer +get_delegate_invoke_impl (gboolean has_target, guint32 param_count, guint32 *code_len, gboolean aot) { guint8 *code, *start; - /* FIXME: Support more cases */ - if (MONO_TYPE_ISSTRUCT (sig->ret)) - return NULL; - if (has_target) { - static guint8* cached = NULL; - int size = MONO_PPC_32_64_CASE (16, 20); - mono_mini_arch_lock (); - if (cached) { - mono_mini_arch_unlock (); - return cached; - } - + int size = MONO_PPC_32_64_CASE (32, 32) + PPC_FTNPTR_SIZE; + start = code = mono_global_codeman_reserve (size); + if (!aot) + code = mono_ppc_create_pre_code_ftnptr (code); /* Replace the this argument with the target */ - ppc_load_reg (code, ppc_r0, G_STRUCT_OFFSET (MonoDelegate, method_ptr), ppc_r3); + ppc_ldptr (code, ppc_r0, G_STRUCT_OFFSET (MonoDelegate, method_ptr), ppc_r3); #ifdef PPC_USES_FUNCTION_DESCRIPTOR /* it's a function descriptor */ - ppc_ldx (code, ppc_r0, 0, ppc_r0); + /* Can't use ldptr as it doesn't work with r0 */ + ppc_ldptr_indexed (code, ppc_r0, 0, ppc_r0); #endif ppc_mtctr (code, ppc_r0); - ppc_load_reg (code, ppc_r3, G_STRUCT_OFFSET (MonoDelegate, target), ppc_r3); + ppc_ldptr (code, ppc_r3, G_STRUCT_OFFSET (MonoDelegate, target), ppc_r3); ppc_bcctr (code, PPC_BR_ALWAYS, 0); g_assert ((code - start) <= size); mono_arch_flush_icache (start, size); - cached = start; - mono_mini_arch_unlock (); - return cached; } else { - static guint8* cache [MAX_ARCH_DELEGATE_PARAMS + 1] = {NULL}; int size, i; - if (sig->param_count > MAX_ARCH_DELEGATE_PARAMS) - return NULL; - for (i = 0; i < sig->param_count; ++i) - if (!mono_is_regsize_var (sig->params [i])) - return NULL; - - mono_mini_arch_lock (); - code = cache [sig->param_count]; - if (code) { - mono_mini_arch_unlock (); - return code; - } - - size = MONO_PPC_32_64_CASE (12, 16) + sig->param_count * 4; + size = MONO_PPC_32_64_CASE (32, 32) + param_count * 4 + PPC_FTNPTR_SIZE; start = code = mono_global_codeman_reserve (size); + if (!aot) + code = mono_ppc_create_pre_code_ftnptr (code); - ppc_load_reg (code, ppc_r0, G_STRUCT_OFFSET (MonoDelegate, method_ptr), ppc_r3); + ppc_ldptr (code, ppc_r0, G_STRUCT_OFFSET (MonoDelegate, method_ptr), ppc_r3); #ifdef PPC_USES_FUNCTION_DESCRIPTOR /* it's a function descriptor */ - ppc_ldx (code, ppc_r0, 0, ppc_r0); + ppc_ldptr_indexed (code, ppc_r0, 0, ppc_r0); #endif ppc_mtctr (code, ppc_r0); /* slide down the arguments */ - for (i = 0; i < sig->param_count; ++i) { + for (i = 0; i < param_count; ++i) { ppc_mr (code, (ppc_r3 + i), (ppc_r3 + i + 1)); } ppc_bcctr (code, PPC_BR_ALWAYS, 0); @@ -435,28 +459,211 @@ mono_arch_get_delegate_invoke_impl (MonoMethodSignature *sig, gboolean has_targe g_assert ((code - start) <= size); mono_arch_flush_icache (start, size); + } + + if (code_len) + *code_len = code - start; + + return start; +} + +GSList* +mono_arch_get_delegate_invoke_impls (void) +{ + GSList *res = NULL; + guint8 *code; + guint32 code_len; + int i; + + code = get_delegate_invoke_impl (TRUE, 0, &code_len, TRUE); + res = g_slist_prepend (res, mono_aot_tramp_info_create (g_strdup ("delegate_invoke_impl_has_target"), code, code_len)); + + for (i = 0; i < MAX_ARCH_DELEGATE_PARAMS; ++i) { + code = get_delegate_invoke_impl (FALSE, i, &code_len, TRUE); + res = g_slist_prepend (res, mono_aot_tramp_info_create (g_strdup_printf ("delegate_invoke_impl_target_%d", i), code, code_len)); + } + + return res; +} + +gpointer +mono_arch_get_delegate_invoke_impl (MonoMethodSignature *sig, gboolean has_target) +{ + guint8 *code, *start; + + /* FIXME: Support more cases */ + if (MONO_TYPE_ISSTRUCT (sig->ret)) + return NULL; + + if (has_target) { + static guint8* cached = NULL; + + if (cached) + return cached; + + if (mono_aot_only) + start = mono_aot_get_named_code ("delegate_invoke_impl_has_target"); + else + start = get_delegate_invoke_impl (TRUE, 0, NULL, FALSE); + + mono_memory_barrier (); + + cached = start; + } else { + static guint8* cache [MAX_ARCH_DELEGATE_PARAMS + 1] = {NULL}; + int i; + + if (sig->param_count > MAX_ARCH_DELEGATE_PARAMS) + return NULL; + for (i = 0; i < sig->param_count; ++i) + if (!mono_is_regsize_var (sig->params [i])) + return NULL; + + + code = cache [sig->param_count]; + if (code) + return code; + + if (mono_aot_only) { + char *name = g_strdup_printf ("delegate_invoke_impl_target_%d", sig->param_count); + start = mono_aot_get_named_code (name); + g_free (name); + } else { + start = get_delegate_invoke_impl (FALSE, sig->param_count, NULL, FALSE); + } + + mono_memory_barrier (); + cache [sig->param_count] = start; - mono_mini_arch_unlock (); - return start; } - return NULL; + return start; } gpointer -mono_arch_get_this_arg_from_call (MonoGenericSharingContext *gsctx, MonoMethodSignature *sig, gssize *regs, guint8 *code) +mono_arch_get_this_arg_from_call (MonoGenericSharingContext *gsctx, MonoMethodSignature *sig, mgreg_t *regs, guint8 *code) { + mgreg_t *r = (mgreg_t*)regs; + /* FIXME: handle returning a struct */ if (MONO_TYPE_ISSTRUCT (sig->ret)) - return (gpointer)regs [ppc_r4]; - return (gpointer)regs [ppc_r3]; + return (gpointer)(gsize)r [ppc_r4]; + return (gpointer)(gsize)r [ppc_r3]; } +typedef struct { + long int type; + long int value; +} AuxVec; + +#ifdef USE_ENVIRON_HACK +static AuxVec* +linux_find_auxv (int *count) +{ + AuxVec *vec; + int c = 0; + char **result = __environ; + /* Scan over the env vector looking for the ending NULL */ + for (; *result != NULL; ++result) { + } + /* Bump the pointer one more step, which should be the auxv. */ + ++result; + vec = (AuxVec *)result; + if (vec->type != 22 /*AT_IGNOREPPC*/) { + *count = 0; + return NULL; + } + while (vec->type != 0 /*AT_NULL*/) { + vec++; + c++; + } + *count = c; + return (AuxVec *)result; +} +#endif + +#define MAX_AUX_ENTRIES 128 +/* + * PPC_FEATURE_POWER4, PPC_FEATURE_POWER5, PPC_FEATURE_POWER5_PLUS, PPC_FEATURE_CELL, + * PPC_FEATURE_PA6T, PPC_FEATURE_ARCH_2_05 are considered supporting 2X ISA features + */ +#define ISA_2X (0x00080000 | 0x00040000 | 0x00020000 | 0x00010000 | 0x00000800 | 0x00001000) + +/* define PPC_FEATURE_64 HWCAP for 64-bit category. */ +#define ISA_64 0x40000000 + +/* define PPC_FEATURE_POWER6_EXT HWCAP for power6x mffgpr/mftgpr instructions. */ +#define ISA_MOVE_FPR_GPR 0x00000200 /* * Initialize the cpu to execute managed code. */ void mono_arch_cpu_init (void) { +#ifdef __APPLE__ + int mib [3]; + size_t len; + mib [0] = CTL_HW; + mib [1] = HW_CACHELINE; + len = sizeof (cachelinesize); + if (sysctl (mib, 2, &cachelinesize, (size_t*)&len, NULL, 0) == -1) { + perror ("sysctl"); + cachelinesize = 128; + } else { + cachelineinc = cachelinesize; + } +#elif defined(__linux__) + AuxVec vec [MAX_AUX_ENTRIES]; + int i, vec_entries = 0; + /* sadly this will work only with 2.6 kernels... */ + FILE* f = fopen ("/proc/self/auxv", "rb"); + if (f) { + vec_entries = fread (&vec, sizeof (AuxVec), MAX_AUX_ENTRIES, f); + fclose (f); +#ifdef USE_ENVIRON_HACK + } else { + AuxVec *evec = linux_find_auxv (&vec_entries); + if (vec_entries) + memcpy (&vec, evec, sizeof (AuxVec) * MIN (vec_entries, MAX_AUX_ENTRIES)); +#endif + } + for (i = 0; i < vec_entries; i++) { + int type = vec [i].type; + if (type == 19) { /* AT_DCACHEBSIZE */ + cachelinesize = vec [i].value; + continue; + } else if (type == 16) { /* AT_HWCAP */ + if (vec [i].value & 0x00002000 /*PPC_FEATURE_ICACHE_SNOOP*/) + cpu_hw_caps |= PPC_ICACHE_SNOOP; + if (vec [i].value & ISA_2X) + cpu_hw_caps |= PPC_ISA_2X; + if (vec [i].value & ISA_64) + cpu_hw_caps |= PPC_ISA_64; + if (vec [i].value & ISA_MOVE_FPR_GPR) + cpu_hw_caps |= PPC_MOVE_FPR_GPR; + continue; + } else if (type == 15) { /* AT_PLATFORM */ + const char *arch = (char*)vec [i].value; + if (strcmp (arch, "ppc970") == 0 || + (strncmp (arch, "power", 5) == 0 && arch [5] >= '4' && arch [5] <= '7')) + cpu_hw_caps |= PPC_MULTIPLE_LS_UNITS; + /*printf ("cpu: %s\n", (char*)vec [i].value);*/ + continue; + } + } +#elif defined(G_COMPILER_CODEWARRIOR) + cachelinesize = 32; + cachelineinc = 32; +#elif defined(MONO_CROSS_COMPILE) +#else +//#error Need a way to get cache line size +#endif + if (!cachelinesize) + cachelinesize = 32; + if (!cachelineinc) + cachelineinc = cachelinesize; + + if (mono_cpu_count () > 1) + cpu_hw_caps |= PPC_SMP_CAPABLE; } /* @@ -585,62 +792,20 @@ mono_arch_regalloc_cost (MonoCompile *cfg, MonoMethodVar *vmv) return 2; } -typedef struct { - long int type; - long int value; -} AuxVec; - void mono_arch_flush_icache (guint8 *code, gint size) { +#ifdef MONO_CROSS_COMPILE +#else register guint8 *p; guint8 *endp, *start; - static int cachelinesize = 0; - static int cachelineinc = 16; - if (!cachelinesize) { -#ifdef __APPLE__ - int mib [3]; - size_t len; - mib [0] = CTL_HW; - mib [1] = HW_CACHELINE; - len = sizeof (cachelinesize); - if (sysctl(mib, 2, &cachelinesize, (size_t*)&len, NULL, 0) == -1) { - perror ("sysctl"); - cachelinesize = 128; - } else { - cachelineinc = cachelinesize; - /*g_print ("setting cl size to %d\n", cachelinesize);*/ - } -#elif defined(__linux__) - /* sadly this will work only with 2.6 kernels... */ - FILE* f = fopen ("/proc/self/auxv", "rb"); - if (f) { - AuxVec vec; - while (fread (&vec, sizeof (vec), 1, f) == 1) { - if (vec.type == 19) { - cachelinesize = vec.value; - break; - } - } - fclose (f); - } - if (!cachelinesize) - cachelinesize = 128; -#elif defined(G_COMPILER_CODEWARRIOR) - cachelinesize = 32; - cachelineinc = 32; -#else -#warning Need a way to get cache line size - cachelinesize = 128; -#endif - } p = start = code; endp = p + size; start = (guint8*)((gsize)start & ~(cachelinesize - 1)); /* use dcbf for smp support, later optimize for UP, see pem._64bit.d20030611.pdf page 211 */ #if defined(G_COMPILER_CODEWARRIOR) - if (1) { + if (cpu_hw_caps & PPC_SMP_CAPABLE) { for (p = start; p < endp; p += cachelineinc) { asm { dcbf 0, p }; } @@ -662,7 +827,19 @@ mono_arch_flush_icache (guint8 *code, gint size) isync } #else - if (1) { + /* For POWER5/6 with ICACHE_SNOOPing only one icbi in the range is required. + * The sync is required to insure that the store queue is completely empty. + * While the icbi performs no cache operations, icbi/isync is required to + * kill local prefetch. + */ + if (cpu_hw_caps & PPC_ICACHE_SNOOP) { + asm ("sync"); + asm ("icbi 0,%0;" : : "r"(code) : "memory"); + asm ("isync"); + return; + } + /* use dcbf for smp support, see pem._64bit.d20030611.pdf page 211 */ + if (cpu_hw_caps & PPC_SMP_CAPABLE) { for (p = start; p < endp; p += cachelineinc) { asm ("dcbf 0,%0;" : : "r"(p) : "memory"); } @@ -674,11 +851,21 @@ mono_arch_flush_icache (guint8 *code, gint size) asm ("sync"); p = code; for (p = start; p < endp; p += cachelineinc) { - asm ("icbi 0,%0; sync;" : : "r"(p) : "memory"); + /* for ISA2.0+ implementations we should not need any extra sync between the + * icbi instructions. Both the 2.0 PEM and the PowerISA-2.05 say this. + * So I am not sure which chip had this problem but its not an issue on + * of the ISA V2 chips. + */ + if (cpu_hw_caps & PPC_ISA_2X) + asm ("icbi 0,%0;" : : "r"(p) : "memory"); + else + asm ("icbi 0,%0; sync;" : : "r"(p) : "memory"); } - asm ("sync"); + if (!(cpu_hw_caps & PPC_ISA_2X)) + asm ("sync"); asm ("isync"); #endif +#endif } void @@ -712,6 +899,7 @@ typedef struct { gint32 offset; guint32 vtsize; /* in param area */ guint8 reg; + guint8 vtregs; /* number of registers used to pass a RegTypeStructByVal */ guint8 regtype : 4; /* 0 general, 1 basereg, 2 floating point register, see RegType* */ guint8 size : 4; /* 1, 2, 4, 8, or regs used by RegTypeStructByVal */ guint8 bytes : 4; /* size in bytes - only valid for @@ -874,14 +1062,19 @@ calculate_sizes (MonoMethodSignature *sig, gboolean is_pinvoke) break; } /* Fall through */ - case MONO_TYPE_VALUETYPE: { + case MONO_TYPE_VALUETYPE: + case MONO_TYPE_TYPEDBYREF: { gint size; MonoClass *klass; + klass = mono_class_from_mono_type (sig->params [i]); - if (is_pinvoke) + if (simpletype == MONO_TYPE_TYPEDBYREF) + size = sizeof (MonoTypedRef); + else if (is_pinvoke) size = mono_class_native_size (klass, NULL); else size = mono_class_value_size (klass, NULL); + #if defined(__APPLE__) || defined(__mono_ppc64__) if ((size == 4 || size == 8) && has_only_a_r48_field (klass)) { cinfo->args [n].size = size; @@ -907,76 +1100,41 @@ calculate_sizes (MonoMethodSignature *sig, gboolean is_pinvoke) #endif DEBUG(printf ("load %d bytes struct\n", mono_class_native_size (sig->params [i]->data.klass, NULL))); + #if PPC_PASS_STRUCTS_BY_VALUE { int align_size = size; - int nwords = 0; + int nregs = 0; int rest = PPC_LAST_ARG_REG - gr + 1; int n_in_regs; + align_size += (sizeof (gpointer) - 1); align_size &= ~(sizeof (gpointer) - 1); - nwords = (align_size + sizeof (gpointer) -1 ) / sizeof (gpointer); - n_in_regs = MIN (rest, nwords); - cinfo->args [n].regtype = RegTypeStructByVal; - if (gr > PPC_LAST_ARG_REG + nregs = (align_size + sizeof (gpointer) -1 ) / sizeof (gpointer); + n_in_regs = MIN (rest, nregs); + if (n_in_regs < 0) + n_in_regs = 0; #ifdef __APPLE__ - /* FIXME: check this */ - || (size >= 3 && size % 4 != 0) -#endif - ) { - cinfo->args [n].size = 0; - cinfo->args [n].vtsize = nwords; - } else { - cinfo->args [n].size = n_in_regs; - cinfo->args [n].vtsize = nwords - n_in_regs; - cinfo->args [n].reg = gr; - } -#ifdef __mono_ppc64__ - if (nwords == 1 && is_pinvoke) - cinfo->args [n].bytes = size; - else + /* FIXME: check this */ + if (size >= 3 && size % 4 != 0) + n_in_regs = 0; #endif - cinfo->args [n].bytes = 0; - gr += n_in_regs; - cinfo->args [n].offset = PPC_STACK_PARAM_OFFSET + stack_size; - /*g_print ("offset for arg %d at %d\n", n, PPC_STACK_PARAM_OFFSET + stack_size);*/ - stack_size += nwords * sizeof (gpointer); - } -#else - add_general (&gr, &stack_size, cinfo->args + n, TRUE); - cinfo->args [n].regtype = RegTypeStructByAddr; - cinfo->args [n].vtsize = size; -#endif - n++; - break; - } - case MONO_TYPE_TYPEDBYREF: { - int size = sizeof (MonoTypedRef); - /* keep in sync or merge with the valuetype case */ -#if PPC_PASS_STRUCTS_BY_VALUE - { - int nwords = (size + sizeof (gpointer) -1 ) / sizeof (gpointer); cinfo->args [n].regtype = RegTypeStructByVal; - if (gr <= PPC_LAST_ARG_REG) { - int rest = PPC_LAST_ARG_REG - gr + 1; - int n_in_regs = rest >= nwords? nwords: rest; - cinfo->args [n].size = n_in_regs; - cinfo->args [n].vtsize = nwords - n_in_regs; - cinfo->args [n].reg = gr; - gr += n_in_regs; - } else { - cinfo->args [n].size = 0; - cinfo->args [n].vtsize = nwords; - } + cinfo->args [n].vtregs = n_in_regs; + cinfo->args [n].size = n_in_regs; + cinfo->args [n].vtsize = nregs - n_in_regs; + cinfo->args [n].reg = gr; + #ifdef __mono_ppc64__ - if (nwords == 1 && is_pinvoke) + if (nregs == 1 && is_pinvoke) cinfo->args [n].bytes = size; else #endif cinfo->args [n].bytes = 0; + gr += n_in_regs; cinfo->args [n].offset = PPC_STACK_PARAM_OFFSET + stack_size; /*g_print ("offset for arg %d at %d\n", n, PPC_STACK_PARAM_OFFSET + stack_size);*/ - stack_size += nwords * sizeof (gpointer); + stack_size += nregs * sizeof (gpointer); } #else add_general (&gr, &stack_size, cinfo->args + n, TRUE); @@ -989,7 +1147,7 @@ calculate_sizes (MonoMethodSignature *sig, gboolean is_pinvoke) case MONO_TYPE_U8: case MONO_TYPE_I8: cinfo->args [n].size = 8; - add_general (&gr, &stack_size, cinfo->args + n, sizeof (gpointer) == 8); + add_general (&gr, &stack_size, cinfo->args + n, SIZEOF_REGISTER == 8); n++; break; case MONO_TYPE_R4: @@ -1001,12 +1159,12 @@ calculate_sizes (MonoMethodSignature *sig, gboolean is_pinvoke) cinfo->args [n].reg = fr; fr ++; FP_ALSO_IN_REG (gr ++); - ALWAYS_ON_STACK (stack_size += sizeof (gpointer)); + ALWAYS_ON_STACK (stack_size += SIZEOF_REGISTER); } else { cinfo->args [n].offset = PPC_STACK_PARAM_OFFSET + stack_size + MONO_PPC_32_64_CASE (0, 4); cinfo->args [n].regtype = RegTypeBase; cinfo->args [n].reg = ppc_sp; /* in the caller*/ - stack_size += sizeof (gpointer); + stack_size += SIZEOF_REGISTER; } n++; break; @@ -1017,7 +1175,7 @@ calculate_sizes (MonoMethodSignature *sig, gboolean is_pinvoke) cinfo->args [n].regtype = RegTypeFP; cinfo->args [n].reg = fr; fr ++; - FP_ALSO_IN_REG (gr += sizeof (double) / sizeof (gpointer)); + FP_ALSO_IN_REG (gr += sizeof (double) / SIZEOF_REGISTER); ALWAYS_ON_STACK (stack_size += 8); } else { cinfo->args [n].offset = PPC_STACK_PARAM_OFFSET + stack_size; @@ -1344,6 +1502,9 @@ emit_sig_cookie (MonoCompile *cfg, MonoCallInst *call, CallInfo *cinfo) { int sig_reg = mono_alloc_ireg (cfg); + /* FIXME: Add support for signature tokens to AOT */ + cfg->disable_aot = TRUE; + MONO_EMIT_NEW_ICONST (cfg, sig_reg, (gulong)call->signature); MONO_EMIT_NEW_STORE_MEMBASE (cfg, OP_STORE_MEMBASE_REG, ppc_r1, cinfo->sig_cookie.offset, sig_reg); @@ -1497,7 +1658,9 @@ mono_arch_emit_outarg_vt (MonoCompile *cfg, MonoInst *ins, MonoInst *src) int i, soffset, dreg; if (ainfo->regtype == RegTypeStructByVal) { +#ifdef __APPLE__ guint32 size = 0; +#endif soffset = 0; #ifdef __APPLE__ /* @@ -1518,7 +1681,7 @@ mono_arch_emit_outarg_vt (MonoCompile *cfg, MonoInst *ins, MonoInst *src) mono_call_inst_add_outarg_reg (cfg, call, dreg, ainfo->reg, FALSE); } else #endif - for (i = 0; i < ainfo->size; ++i) { + for (i = 0; i < ainfo->vtregs; ++i) { int antipadding = 0; if (ainfo->bytes) { g_assert (i == 0); @@ -1609,7 +1772,7 @@ mono_arch_instrument_prolog (MonoCompile *cfg, void *func, void *p, gboolean ena { guchar *code = p; - ppc_load (code, ppc_r3, cfg->method); + ppc_load_ptr (code, ppc_r3, cfg->method); ppc_li (code, ppc_r4, 0); /* NULL ebp for now */ ppc_load_func (code, ppc_r0, func); ppc_mtlr (code, ppc_r0); @@ -1626,7 +1789,7 @@ enum { }; void* -mono_arch_instrument_epilog (MonoCompile *cfg, void *func, void *p, gboolean enable_arguments) +mono_arch_instrument_epilog_full (MonoCompile *cfg, void *func, void *p, gboolean enable_arguments, gboolean preserve_argument_registers) { guchar *code = p; int save_mode = SAVE_NONE; @@ -1682,7 +1845,7 @@ mono_arch_instrument_epilog (MonoCompile *cfg, void *func, void *p, gboolean ena } break; case SAVE_ONE: - ppc_store_reg (code, ppc_r3, save_offset, cfg->frame_reg); + ppc_stptr (code, ppc_r3, save_offset, cfg->frame_reg); if (enable_arguments) { ppc_mr (code, ppc_r4, ppc_r3); } @@ -1708,7 +1871,7 @@ mono_arch_instrument_epilog (MonoCompile *cfg, void *func, void *p, gboolean ena break; } - ppc_load (code, ppc_r3, cfg->method); + ppc_load_ptr (code, ppc_r3, cfg->method); ppc_load_func (code, ppc_r0, func); ppc_mtlr (code, ppc_r0); ppc_blrl (code); @@ -1719,7 +1882,7 @@ mono_arch_instrument_epilog (MonoCompile *cfg, void *func, void *p, gboolean ena ppc_lwz (code, ppc_r4, save_offset + 4, cfg->frame_reg); break; case SAVE_ONE: - ppc_load_reg (code, ppc_r3, save_offset, cfg->frame_reg); + ppc_ldptr (code, ppc_r3, save_offset, cfg->frame_reg); break; case SAVE_FP: ppc_lfd (code, ppc_f1, save_offset, cfg->frame_reg); @@ -1750,31 +1913,22 @@ typedef struct { } MonoOvfJump; #define EMIT_COND_BRANCH_FLAGS(ins,b0,b1) \ -if (ins->flags & MONO_INST_BRLABEL) { \ - if (0 && ins->inst_i0->inst_c0) { \ - ppc_bc (code, (b0), (b1), (code - cfg->native_code + ins->inst_i0->inst_c0) & 0xffff); \ - } else { \ - mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_LABEL, ins->inst_i0); \ - ppc_bc (code, (b0), (b1), 0); \ - } \ +if (0 && ins->inst_true_bb->native_offset) { \ + ppc_bc (code, (b0), (b1), (code - cfg->native_code + ins->inst_true_bb->native_offset) & 0xffff); \ } else { \ - if (0 && ins->inst_true_bb->native_offset) { \ - ppc_bc (code, (b0), (b1), (code - cfg->native_code + ins->inst_true_bb->native_offset) & 0xffff); \ - } else { \ - int br_disp = ins->inst_true_bb->max_offset - offset; \ - if (!ppc_is_imm16 (br_disp + 1024) || ! ppc_is_imm16 (ppc_is_imm16 (br_disp - 1024))) { \ - MonoOvfJump *ovfj = mono_mempool_alloc (cfg->mempool, sizeof (MonoOvfJump)); \ - ovfj->data.bb = ins->inst_true_bb; \ - ovfj->ip_offset = 0; \ - ovfj->b0_cond = (b0); \ - ovfj->b1_cond = (b1); \ - mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_BB_OVF, ovfj); \ - ppc_b (code, 0); \ - } else { \ - mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_BB, ins->inst_true_bb); \ - ppc_bc (code, (b0), (b1), 0); \ - } \ - } \ + int br_disp = ins->inst_true_bb->max_offset - offset; \ + if (!ppc_is_imm16 (br_disp + 1024) || ! ppc_is_imm16 (ppc_is_imm16 (br_disp - 1024))) { \ + MonoOvfJump *ovfj = mono_mempool_alloc (cfg->mempool, sizeof (MonoOvfJump)); \ + ovfj->data.bb = ins->inst_true_bb; \ + ovfj->ip_offset = 0; \ + ovfj->b0_cond = (b0); \ + ovfj->b1_cond = (b1); \ + mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_BB_OVF, ovfj); \ + ppc_b (code, 0); \ + } else { \ + mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_BB, ins->inst_true_bb); \ + ppc_bc (code, (b0), (b1), 0); \ + } \ } #define EMIT_COND_BRANCH(ins,cond) EMIT_COND_BRANCH_FLAGS(ins, branch_b0_table [(cond)], branch_b1_table [(cond)]) @@ -1814,6 +1968,7 @@ static int normalize_opcode (int opcode) { switch (opcode) { +#ifndef __mono_ilp32__ case MONO_PPC_32_64_CASE (OP_LOADI4_MEMBASE, OP_LOADI8_MEMBASE): return OP_LOAD_MEMBASE; case MONO_PPC_32_64_CASE (OP_LOADI4_MEMINDEX, OP_LOADI8_MEMINDEX): @@ -1824,6 +1979,7 @@ normalize_opcode (int opcode) return OP_STORE_MEMBASE_IMM; case MONO_PPC_32_64_CASE (OP_STOREI4_MEMINDEX, OP_STOREI8_MEMINDEX): return OP_STORE_MEMINDEX; +#endif case MONO_PPC_32_64_CASE (OP_ISHR_IMM, OP_LSHR_IMM): return OP_SHR_IMM; case MONO_PPC_32_64_CASE (OP_ISHR_UN_IMM, OP_LSHR_UN_IMM): @@ -1998,28 +2154,34 @@ mono_arch_decompose_opts (MonoCompile *cfg, MonoInst *ins) #ifndef __mono_ppc64__ case OP_ICONV_TO_R4: case OP_ICONV_TO_R8: { - /* FIXME: change precision for CEE_CONV_R4 */ - static const guint64 adjust_val = 0x4330000080000000ULL; - int msw_reg = mono_alloc_ireg (cfg); - int xored = mono_alloc_ireg (cfg); - int adj_reg = mono_alloc_freg (cfg); - int tmp_reg = mono_alloc_freg (cfg); - int basereg = ppc_sp; - int offset = -8; - if (!ppc_is_imm16 (offset + 4)) { - basereg = mono_alloc_ireg (cfg); - MONO_EMIT_NEW_BIALU_IMM (cfg, OP_IADD_IMM, basereg, cfg->frame_reg, offset); + /* If we have a PPC_FEATURE_64 machine we can avoid + this and use the fcfid instruction. Otherwise + on an old 32-bit chip and we have to do this the + hard way. */ + if (!(cpu_hw_caps & PPC_ISA_64)) { + /* FIXME: change precision for CEE_CONV_R4 */ + static const guint64 adjust_val = 0x4330000080000000ULL; + int msw_reg = mono_alloc_ireg (cfg); + int xored = mono_alloc_ireg (cfg); + int adj_reg = mono_alloc_freg (cfg); + int tmp_reg = mono_alloc_freg (cfg); + int basereg = ppc_sp; + int offset = -8; + if (!ppc_is_imm16 (offset + 4)) { + basereg = mono_alloc_ireg (cfg); + MONO_EMIT_NEW_BIALU_IMM (cfg, OP_IADD_IMM, basereg, cfg->frame_reg, offset); + } + MONO_EMIT_NEW_ICONST (cfg, msw_reg, 0x43300000); + MONO_EMIT_NEW_STORE_MEMBASE (cfg, OP_STOREI4_MEMBASE_REG, basereg, offset, msw_reg); + MONO_EMIT_NEW_BIALU_IMM (cfg, OP_XOR_IMM, xored, ins->sreg1, 0x80000000); + MONO_EMIT_NEW_STORE_MEMBASE (cfg, OP_STOREI4_MEMBASE_REG, basereg, offset + 4, xored); + MONO_EMIT_NEW_LOAD_R8 (cfg, adj_reg, (gpointer)&adjust_val); + MONO_EMIT_NEW_LOAD_MEMBASE_OP (cfg, OP_LOADR8_MEMBASE, tmp_reg, basereg, offset); + MONO_EMIT_NEW_BIALU (cfg, OP_FSUB, ins->dreg, tmp_reg, adj_reg); + if (ins->opcode == OP_ICONV_TO_R4) + MONO_EMIT_NEW_UNALU (cfg, OP_FCONV_TO_R4, ins->dreg, ins->dreg); + ins->opcode = OP_NOP; } - MONO_EMIT_NEW_ICONST (cfg, msw_reg, 0x43300000); - MONO_EMIT_NEW_STORE_MEMBASE (cfg, OP_STOREI4_MEMBASE_REG, basereg, offset, msw_reg); - MONO_EMIT_NEW_BIALU_IMM (cfg, OP_XOR_IMM, xored, ins->sreg1, 0x80000000); - MONO_EMIT_NEW_STORE_MEMBASE (cfg, OP_STOREI4_MEMBASE_REG, basereg, offset + 4, xored); - MONO_EMIT_NEW_LOAD_R8 (cfg, adj_reg, (gpointer)&adjust_val); - MONO_EMIT_NEW_LOAD_MEMBASE_OP (cfg, OP_LOADR8_MEMBASE, tmp_reg, basereg, offset); - MONO_EMIT_NEW_BIALU (cfg, OP_FSUB, ins->dreg, tmp_reg, adj_reg); - if (ins->opcode == OP_ICONV_TO_R4) - MONO_EMIT_NEW_UNALU (cfg, OP_FCONV_TO_R4, ins->dreg, ins->dreg); - ins->opcode = OP_NOP; break; } #endif @@ -2059,6 +2221,48 @@ mono_arch_decompose_opts (MonoCompile *cfg, MonoInst *ins) } } +void +mono_arch_decompose_long_opts (MonoCompile *cfg, MonoInst *ins) +{ + switch (ins->opcode) { + case OP_LADD_OVF: + /* ADC sets the condition code */ + MONO_EMIT_NEW_BIALU (cfg, OP_ADDCC, ins->dreg + 1, ins->sreg1 + 1, ins->sreg2 + 1); + MONO_EMIT_NEW_BIALU (cfg, OP_ADD_OVF_CARRY, ins->dreg + 2, ins->sreg1 + 2, ins->sreg2 + 2); + NULLIFY_INS (ins); + break; + case OP_LADD_OVF_UN: + /* ADC sets the condition code */ + MONO_EMIT_NEW_BIALU (cfg, OP_ADDCC, ins->dreg + 1, ins->sreg1 + 1, ins->sreg2 + 1); + MONO_EMIT_NEW_BIALU (cfg, OP_ADD_OVF_UN_CARRY, ins->dreg + 2, ins->sreg1 + 2, ins->sreg2 + 2); + NULLIFY_INS (ins); + break; + case OP_LSUB_OVF: + /* SBB sets the condition code */ + MONO_EMIT_NEW_BIALU (cfg, OP_SUBCC, ins->dreg + 1, ins->sreg1 + 1, ins->sreg2 + 1); + MONO_EMIT_NEW_BIALU (cfg, OP_SUB_OVF_CARRY, ins->dreg + 2, ins->sreg1 + 2, ins->sreg2 + 2); + NULLIFY_INS (ins); + break; + case OP_LSUB_OVF_UN: + /* SBB sets the condition code */ + MONO_EMIT_NEW_BIALU (cfg, OP_SUBCC, ins->dreg + 1, ins->sreg1 + 1, ins->sreg2 + 1); + MONO_EMIT_NEW_BIALU (cfg, OP_SUB_OVF_UN_CARRY, ins->dreg + 2, ins->sreg1 + 2, ins->sreg2 + 2); + NULLIFY_INS (ins); + break; + case OP_LNEG: + /* This is the old version from inssel-long32.brg */ + MONO_EMIT_NEW_UNALU (cfg, OP_INOT, ins->dreg + 1, ins->sreg1 + 1); + MONO_EMIT_NEW_UNALU (cfg, OP_INOT, ins->dreg + 2, ins->sreg1 + 2); + /* ADC sets the condition codes */ + MONO_EMIT_NEW_BIALU_IMM (cfg, OP_ADC_IMM, ins->dreg + 1, ins->dreg + 1, 1); + MONO_EMIT_NEW_BIALU_IMM (cfg, OP_ADC_IMM, ins->dreg + 2, ins->dreg + 2, 0); + NULLIFY_INS (ins); + break; + default: + break; + } +} + /* * the branch_b0_table should maintain the order of these * opcodes. @@ -2293,7 +2497,7 @@ loop_start: case OP_XOR_IMM: { gboolean is_imm = ((ins->inst_imm & 0xffff0000) && (ins->inst_imm & 0xffff)); #ifdef __mono_ppc64__ - if (ins->inst_imm & 0xffffffff00000000UL) + if (ins->inst_imm & 0xffffffff00000000ULL) is_imm = TRUE; #endif if (is_imm) { @@ -2420,6 +2624,10 @@ loop_start: goto loop_start; /* make it handle the possibly big ins->inst_offset */ case OP_R8CONST: case OP_R4CONST: + if (cfg->compile_aot) { + /* Keep these in the aot case */ + break; + } NEW_INS (cfg, temp, OP_ICONST); temp->inst_c0 = (gulong)ins->inst_p0; temp->dreg = mono_alloc_ireg (cfg); @@ -2441,8 +2649,8 @@ loop_start: static guchar* emit_float_to_int (MonoCompile *cfg, guchar *code, int dreg, int sreg, int size, gboolean is_signed) { - int offset = cfg->arch.fp_conv_var_offset; - int sub_offset; + long offset = cfg->arch.fp_conv_var_offset; + long sub_offset; /* sreg is a float, dreg is an integer reg. ppc_f0 is used a scratch */ #ifdef __mono_ppc64__ if (size == 8) { @@ -2457,7 +2665,7 @@ emit_float_to_int (MonoCompile *cfg, guchar *code, int dreg, int sreg, int size, if (ppc_is_imm16 (offset + sub_offset)) { ppc_stfd (code, ppc_f0, offset, cfg->frame_reg); if (size == 8) - ppc_load_reg (code, dreg, offset + sub_offset, cfg->frame_reg); + ppc_ldr (code, dreg, offset + sub_offset, cfg->frame_reg); else ppc_lwz (code, dreg, offset + sub_offset, cfg->frame_reg); } else { @@ -2465,7 +2673,7 @@ emit_float_to_int (MonoCompile *cfg, guchar *code, int dreg, int sreg, int size, ppc_add (code, dreg, dreg, cfg->frame_reg); ppc_stfd (code, ppc_f0, 0, dreg); if (size == 8) - ppc_load_reg (code, dreg, sub_offset, dreg); + ppc_ldr (code, dreg, sub_offset, dreg); else ppc_lwz (code, dreg, sub_offset, dreg); } @@ -2577,12 +2785,12 @@ handle_thunk (int absolute, guchar *code, const guchar *target) { pdata.found = 0; mono_domain_lock (domain); - mono_code_manager_foreach (domain->code_mp, search_thunk_slot, &pdata); + mono_domain_code_foreach (domain, search_thunk_slot, &pdata); if (!pdata.found) { /* this uses the first available slot */ pdata.found = 2; - mono_code_manager_foreach (domain->code_mp, search_thunk_slot, &pdata); + mono_domain_code_foreach (domain, search_thunk_slot, &pdata); } mono_domain_unlock (domain); @@ -2594,14 +2802,14 @@ handle_thunk (int absolute, guchar *code, const guchar *target) { static void patch_ins (guint8 *code, guint32 ins) { - *(guint32*)code = ins; + *(guint32*)code = GUINT32_TO_BE (ins); mono_arch_flush_icache (code, 4); } void ppc_patch_full (guchar *code, const guchar *target, gboolean is_fd) { - guint32 ins = *(guint32*)code; + guint32 ins = GUINT32_FROM_BE (*(guint32*)code); guint32 prim = ins >> 26; guint32 ovf; @@ -2709,7 +2917,7 @@ ppc_patch_full (guchar *code, const guchar *target, gboolean is_fd) /* FIXME: make this thread safe */ /* FIXME: we're assuming we're using r11 here */ - ppc_load_sequence (code, ppc_r11, target); + ppc_load_ptr_sequence (code, ppc_r11, target); mono_arch_flush_icache ((guint8*)seq, 28); #else guint32 *seq; @@ -2740,6 +2948,12 @@ ppc_patch (guchar *code, const guchar *target) ppc_patch_full (code, target, FALSE); } +void +mono_ppc_patch (guchar *code, const guchar *target) +{ + ppc_patch (code, target); +} + static guint8* emit_move_return_value (MonoCompile *cfg, MonoInst *ins, guint8 *code) { @@ -2783,7 +2997,7 @@ emit_load_volatile_arguments (MonoCompile *cfg, guint8 *code) ArgInfo *ainfo = &cinfo->ret; inst = cfg->vret_addr; g_assert (ppc_is_imm16 (inst->inst_offset)); - ppc_load_reg (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); + ppc_ldptr (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); } for (i = 0; i < sig->param_count + sig->hasthis; ++i) { ArgInfo *ainfo = cinfo->args + i; @@ -2807,7 +3021,7 @@ emit_load_volatile_arguments (MonoCompile *cfg, guint8 *code) break; #endif default: - ppc_load_reg (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); + ppc_ldptr (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); break; } break; @@ -2835,8 +3049,8 @@ emit_load_volatile_arguments (MonoCompile *cfg, guint8 *code) #endif if (MONO_TYPE_IS_REFERENCE (type) || type->type == MONO_TYPE_I8) { - ppc_load_reg (code, ppc_r0, inst->inst_offset, inst->inst_basereg); - ppc_store_reg (code, ppc_r0, ainfo->offset, ainfo->reg); + ppc_ldptr (code, ppc_r0, inst->inst_offset, inst->inst_basereg); + ppc_stptr (code, ppc_r0, ainfo->offset, ainfo->reg); } else if (type->type == MONO_TYPE_I4) { ppc_lwz (code, ppc_r0, inst->inst_offset, inst->inst_basereg); ppc_stw (code, ppc_r0, ainfo->offset, ainfo->reg); @@ -2848,7 +3062,9 @@ emit_load_volatile_arguments (MonoCompile *cfg, guint8 *code) } case RegTypeStructByVal: { +#ifdef __APPLE__ guint32 size = 0; +#endif int j; /* FIXME: */ @@ -2866,8 +3082,8 @@ emit_load_volatile_arguments (MonoCompile *cfg, guint8 *code) NOT_IMPLEMENTED; } else #endif - for (j = 0; j < ainfo->size; ++j) { - ppc_load_reg (code, ainfo->reg + j, + for (j = 0; j < ainfo->vtregs; ++j) { + ppc_ldptr (code, ainfo->reg + j, inst->inst_offset + j * sizeof (gpointer), inst->inst_basereg); /* FIXME: shift to the right */ @@ -2882,7 +3098,7 @@ emit_load_volatile_arguments (MonoCompile *cfg, guint8 *code) g_assert (ppc_is_imm16 (addr->inst_offset)); g_assert (!ainfo->offset); - ppc_load_reg (code, ainfo->reg, addr->inst_offset, addr->inst_basereg); + ppc_ldptr (code, ainfo->reg, addr->inst_offset, addr->inst_basereg); struct_index++; break; @@ -2953,7 +3169,7 @@ ins_native_length (MonoCompile *cfg, MonoInst *ins) static guint8* emit_reserve_param_area (MonoCompile *cfg, guint8 *code) { - int size = cfg->param_area; + long size = cfg->param_area; size += MONO_ARCH_FRAME_ALIGNMENT - 1; size &= -MONO_ARCH_FRAME_ALIGNMENT; @@ -2961,12 +3177,12 @@ emit_reserve_param_area (MonoCompile *cfg, guint8 *code) if (!size) return code; - ppc_load_reg (code, ppc_r0, 0, ppc_sp); + ppc_ldptr (code, ppc_r0, 0, ppc_sp); if (ppc_is_imm16 (-size)) { - ppc_store_reg_update (code, ppc_r0, -size, ppc_sp); + ppc_stptr_update (code, ppc_r0, -size, ppc_sp); } else { ppc_load (code, ppc_r11, -size); - ppc_store_reg_update_indexed (code, ppc_r0, ppc_sp, ppc_r11); + ppc_stptr_update_indexed (code, ppc_r0, ppc_sp, ppc_r11); } return code; @@ -2975,7 +3191,7 @@ emit_reserve_param_area (MonoCompile *cfg, guint8 *code) static guint8* emit_unreserve_param_area (MonoCompile *cfg, guint8 *code) { - int size = cfg->param_area; + long size = cfg->param_area; size += MONO_ARCH_FRAME_ALIGNMENT - 1; size &= -MONO_ARCH_FRAME_ALIGNMENT; @@ -2983,12 +3199,12 @@ emit_unreserve_param_area (MonoCompile *cfg, guint8 *code) if (!size) return code; - ppc_load_reg (code, ppc_r0, 0, ppc_sp); + ppc_ldptr (code, ppc_r0, 0, ppc_sp); if (ppc_is_imm16 (size)) { - ppc_store_reg_update (code, ppc_r0, size, ppc_sp); + ppc_stptr_update (code, ppc_r0, size, ppc_sp); } else { ppc_load (code, ppc_r11, size); - ppc_store_reg_update_indexed (code, ppc_r0, ppc_sp, ppc_r11); + ppc_stptr_update_indexed (code, ppc_r0, ppc_sp, ppc_r11); } return code; @@ -3068,66 +3284,116 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) if (ppc_is_imm16 (ins->inst_offset)) { ppc_stb (code, ins->sreg1, ins->inst_offset, ins->inst_destbasereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_stbx (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset)) { + ppc_addis (code, ppc_r12, ins->inst_destbasereg, ppc_ha(ins->inst_offset)); + ppc_stb (code, ins->sreg1, ins->inst_offset, ppc_r12); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_stbx (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + } } break; case OP_STOREI2_MEMBASE_REG: if (ppc_is_imm16 (ins->inst_offset)) { ppc_sth (code, ins->sreg1, ins->inst_offset, ins->inst_destbasereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_sthx (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset)) { + ppc_addis (code, ppc_r12, ins->inst_destbasereg, ppc_ha(ins->inst_offset)); + ppc_sth (code, ins->sreg1, ins->inst_offset, ppc_r12); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_sthx (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + } } break; case OP_STORE_MEMBASE_REG: if (ppc_is_imm16 (ins->inst_offset)) { - ppc_store_reg (code, ins->sreg1, ins->inst_offset, ins->inst_destbasereg); + ppc_stptr (code, ins->sreg1, ins->inst_offset, ins->inst_destbasereg); + } else { + if (ppc_is_imm32 (ins->inst_offset)) { + ppc_addis (code, ppc_r12, ins->inst_destbasereg, ppc_ha(ins->inst_offset)); + ppc_stptr (code, ins->sreg1, ins->inst_offset, ppc_r12); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_stptr_indexed (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + } + } + break; +#ifdef __mono_ilp32__ + case OP_STOREI8_MEMBASE_REG: + if (ppc_is_imm16 (ins->inst_offset)) { + ppc_str (code, ins->sreg1, ins->inst_offset, ins->inst_destbasereg); } else { ppc_load (code, ppc_r0, ins->inst_offset); - ppc_store_reg_indexed (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + ppc_str_indexed (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); } break; +#endif case OP_STOREI1_MEMINDEX: - ppc_stbx (code, ins->sreg1, ins->sreg2, ins->inst_destbasereg); + ppc_stbx (code, ins->sreg1, ins->inst_destbasereg, ins->sreg2); break; case OP_STOREI2_MEMINDEX: - ppc_sthx (code, ins->sreg1, ins->sreg2, ins->inst_destbasereg); + ppc_sthx (code, ins->sreg1, ins->inst_destbasereg, ins->sreg2); break; case OP_STORE_MEMINDEX: - ppc_store_reg_indexed (code, ins->sreg1, ins->sreg2, ins->inst_destbasereg); + ppc_stptr_indexed (code, ins->sreg1, ins->inst_destbasereg, ins->sreg2); break; case OP_LOADU4_MEM: g_assert_not_reached (); break; case OP_LOAD_MEMBASE: if (ppc_is_imm16 (ins->inst_offset)) { - ppc_load_reg (code, ins->dreg, ins->inst_offset, ins->inst_basereg); + ppc_ldptr (code, ins->dreg, ins->inst_offset, ins->inst_basereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_load_reg_indexed (code, ins->dreg, ins->inst_basereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset) && (ins->dreg > 0)) { + ppc_addis (code, ins->dreg, ins->inst_basereg, ppc_ha(ins->inst_offset)); + ppc_ldptr (code, ins->dreg, ins->inst_offset, ins->dreg); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_ldptr_indexed (code, ins->dreg, ins->inst_basereg, ppc_r0); + } } break; case OP_LOADI4_MEMBASE: +#ifdef __mono_ppc64__ + if (ppc_is_imm16 (ins->inst_offset)) { + ppc_lwa (code, ins->dreg, ins->inst_offset, ins->inst_basereg); + } else { + if (ppc_is_imm32 (ins->inst_offset) && (ins->dreg > 0)) { + ppc_addis (code, ins->dreg, ins->inst_basereg, ppc_ha(ins->inst_offset)); + ppc_lwa (code, ins->dreg, ins->inst_offset, ins->dreg); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_lwax (code, ins->dreg, ins->inst_basereg, ppc_r0); + } + } + break; +#endif case OP_LOADU4_MEMBASE: if (ppc_is_imm16 (ins->inst_offset)) { ppc_lwz (code, ins->dreg, ins->inst_offset, ins->inst_basereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_lwzx (code, ins->dreg, ins->inst_basereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset) && (ins->dreg > 0)) { + ppc_addis (code, ins->dreg, ins->inst_basereg, ppc_ha(ins->inst_offset)); + ppc_lwz (code, ins->dreg, ins->inst_offset, ins->dreg); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_lwzx (code, ins->dreg, ins->inst_basereg, ppc_r0); + } } -#ifdef __mono_ppc64__ - if (ins->opcode == OP_LOADI4_MEMBASE) - ppc_extsw (code, ins->dreg, ins->dreg); -#endif break; case OP_LOADI1_MEMBASE: case OP_LOADU1_MEMBASE: if (ppc_is_imm16 (ins->inst_offset)) { ppc_lbz (code, ins->dreg, ins->inst_offset, ins->inst_basereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_lbzx (code, ins->dreg, ins->inst_basereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset) && (ins->dreg > 0)) { + ppc_addis (code, ins->dreg, ins->inst_basereg, ppc_ha(ins->inst_offset)); + ppc_lbz (code, ins->dreg, ins->inst_offset, ins->dreg); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_lbzx (code, ins->dreg, ins->inst_basereg, ppc_r0); + } } if (ins->opcode == OP_LOADI1_MEMBASE) ppc_extsb (code, ins->dreg, ins->dreg); @@ -3136,40 +3402,60 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) if (ppc_is_imm16 (ins->inst_offset)) { ppc_lhz (code, ins->dreg, ins->inst_offset, ins->inst_basereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_lhzx (code, ins->dreg, ins->inst_basereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset) && (ins->dreg > 0)) { + ppc_addis (code, ins->dreg, ins->inst_basereg, ppc_ha(ins->inst_offset)); + ppc_lhz (code, ins->dreg, ins->inst_offset, ins->dreg); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_lhzx (code, ins->dreg, ins->inst_basereg, ppc_r0); + } } break; case OP_LOADI2_MEMBASE: if (ppc_is_imm16 (ins->inst_offset)) { - ppc_lha (code, ins->dreg, ins->inst_basereg, ins->inst_offset); + ppc_lha (code, ins->dreg, ins->inst_offset, ins->inst_basereg); + } else { + if (ppc_is_imm32 (ins->inst_offset) && (ins->dreg > 0)) { + ppc_addis (code, ins->dreg, ins->inst_basereg, ppc_ha(ins->inst_offset)); + ppc_lha (code, ins->dreg, ins->inst_offset, ins->dreg); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_lhax (code, ins->dreg, ins->inst_basereg, ppc_r0); + } + } + break; +#ifdef __mono_ilp32__ + case OP_LOADI8_MEMBASE: + if (ppc_is_imm16 (ins->inst_offset)) { + ppc_ldr (code, ins->dreg, ins->inst_offset, ins->inst_basereg); } else { ppc_load (code, ppc_r0, ins->inst_offset); - ppc_lhax (code, ins->dreg, ins->inst_basereg, ppc_r0); + ppc_ldr_indexed (code, ins->dreg, ins->inst_basereg, ppc_r0); } break; +#endif case OP_LOAD_MEMINDEX: - ppc_load_reg_indexed (code, ins->dreg, ins->sreg2, ins->inst_basereg); + ppc_ldptr_indexed (code, ins->dreg, ins->inst_basereg, ins->sreg2); break; case OP_LOADI4_MEMINDEX: - case OP_LOADU4_MEMINDEX: - ppc_lwzx (code, ins->dreg, ins->sreg2, ins->inst_basereg); #ifdef __mono_ppc64__ - if (ins->opcode == OP_LOADI4_MEMINDEX) - ppc_extsb (code, ins->dreg, ins->dreg); + ppc_lwax (code, ins->dreg, ins->inst_basereg, ins->sreg2); + break; #endif + case OP_LOADU4_MEMINDEX: + ppc_lwzx (code, ins->dreg, ins->inst_basereg, ins->sreg2); break; case OP_LOADU2_MEMINDEX: - ppc_lhzx (code, ins->dreg, ins->sreg2, ins->inst_basereg); + ppc_lhzx (code, ins->dreg, ins->inst_basereg, ins->sreg2); break; case OP_LOADI2_MEMINDEX: - ppc_lhax (code, ins->dreg, ins->sreg2, ins->inst_basereg); + ppc_lhax (code, ins->dreg, ins->inst_basereg, ins->sreg2); break; case OP_LOADU1_MEMINDEX: - ppc_lbzx (code, ins->dreg, ins->sreg2, ins->inst_basereg); + ppc_lbzx (code, ins->dreg, ins->inst_basereg, ins->sreg2); break; case OP_LOADI1_MEMINDEX: - ppc_lbzx (code, ins->dreg, ins->sreg2, ins->inst_basereg); + ppc_lbzx (code, ins->dreg, ins->inst_basereg, ins->sreg2); ppc_extsb (code, ins->dreg, ins->dreg); break; case OP_ICONV_TO_I1: @@ -3191,7 +3477,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) case OP_COMPARE: case OP_ICOMPARE: CASE_PPC64 (OP_LCOMPARE) - L = (sizeof (gpointer) == 4 || ins->opcode == OP_ICOMPARE) ? 0 : 1; + L = (sizeof (mgreg_t) == 4 || ins->opcode == OP_ICOMPARE) ? 0 : 1; next = ins->next; if (next && compare_opcode_is_unsigned (next->opcode)) ppc_cmpl (code, 0, L, ins->sreg1, ins->sreg2); @@ -3201,7 +3487,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) case OP_COMPARE_IMM: case OP_ICOMPARE_IMM: CASE_PPC64 (OP_LCOMPARE_IMM) - L = (sizeof (gpointer) == 4 || ins->opcode == OP_ICOMPARE_IMM) ? 0 : 1; + L = (sizeof (mgreg_t) == 4 || ins->opcode == OP_ICOMPARE_IMM) ? 0 : 1; next = ins->next; if (next && compare_opcode_is_unsigned (next->opcode)) { if (ppc_is_uimm16 (ins->inst_imm)) { @@ -3512,9 +3798,25 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) ppc_multiply (code, ins->dreg, ins->sreg1, ins->sreg2); break; case OP_ICONST: - CASE_PPC64 (OP_I8CONST) ppc_load (code, ins->dreg, ins->inst_c0); break; + case OP_I8CONST: { + ppc_load (code, ins->dreg, ins->inst_l); + break; + } + case OP_LOAD_GOTADDR: + /* The PLT implementation depends on this */ + g_assert (ins->dreg == ppc_r30); + + code = mono_arch_emit_load_got_addr (cfg->native_code, code, cfg, NULL); + break; + case OP_GOT_ENTRY: + // FIXME: Fix max instruction length + mono_add_patch_info (cfg, offset, (MonoJumpInfoType)ins->inst_right->inst_i1, ins->inst_right->inst_p0); + /* arch_emit_got_access () patches this */ + ppc_load32 (code, ppc_r0, 0); + ppc_ldptr_indexed (code, ins->dreg, ins->inst_basereg, ppc_r0); + break; case OP_AOTCONST: mono_add_patch_info (cfg, offset, (MonoJumpInfoType)ins->inst_i1, ins->inst_p0); ppc_load_sequence (code, ins->dreg, 0); @@ -3554,11 +3856,12 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) * we're leaving the method. */ if (1 || cfg->flags & MONO_CFG_HAS_CALLS) { - if (ppc_is_imm16 (cfg->stack_usage + PPC_RET_ADDR_OFFSET)) { - ppc_load_reg (code, ppc_r0, cfg->stack_usage + PPC_RET_ADDR_OFFSET, cfg->frame_reg); + long ret_offset = cfg->stack_usage + PPC_RET_ADDR_OFFSET; + if (ppc_is_imm16 (ret_offset)) { + ppc_ldptr (code, ppc_r0, ret_offset, cfg->frame_reg); } else { - ppc_load (code, ppc_r11, cfg->stack_usage + PPC_RET_ADDR_OFFSET); - ppc_load_reg_indexed (code, ppc_r0, cfg->frame_reg, ppc_r11); + ppc_load (code, ppc_r11, ret_offset); + ppc_ldptr_indexed (code, ppc_r0, cfg->frame_reg, ppc_r11); } ppc_mtlr (code, ppc_r0); } @@ -3566,10 +3869,12 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) code = emit_load_volatile_arguments (cfg, code); if (ppc_is_imm16 (cfg->stack_usage)) { - ppc_addic (code, ppc_r11, cfg->frame_reg, cfg->stack_usage); + ppc_addi (code, ppc_r11, cfg->frame_reg, cfg->stack_usage); } else { - ppc_load (code, ppc_r11, cfg->stack_usage); - ppc_add (code, ppc_r11, cfg->frame_reg, ppc_r11); + /* cfg->stack_usage is an int, so we can use + * an addis/addi sequence here even in 64-bit. */ + ppc_addis (code, ppc_r11, cfg->frame_reg, ppc_ha(cfg->stack_usage)); + ppc_addi (code, ppc_r11, ppc_r11, cfg->stack_usage); } if (!cfg->method->save_lmf) { /*for (i = 31; i >= 14; --i) { @@ -3582,7 +3887,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) for (i = 31; i >= 13; --i) { if (cfg->used_int_regs & (1 << i)) { pos += sizeof (gpointer); - ppc_load_reg (code, i, -pos, ppc_r11); + ppc_ldptr (code, i, -pos, ppc_r11); } } } else { @@ -3590,21 +3895,35 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) } ppc_mr (code, ppc_sp, ppc_r11); mono_add_patch_info (cfg, (guint8*) code - cfg->native_code, MONO_PATCH_INFO_METHOD_JUMP, ins->inst_p0); - ppc_b (code, 0); + if (cfg->compile_aot) { + /* arch_emit_got_access () patches this */ + ppc_load32 (code, ppc_r0, 0); +#ifdef PPC_USES_FUNCTION_DESCRIPTOR + ppc_ldptr_indexed (code, ppc_r11, ppc_r30, ppc_r0); + ppc_ldptr (code, ppc_r0, 0, ppc_r11); +#else + ppc_ldptr_indexed (code, ppc_r0, ppc_r30, ppc_r0); +#endif + ppc_mtctr (code, ppc_r0); + ppc_bcctr (code, PPC_BR_ALWAYS, 0); + } else { + ppc_b (code, 0); + } break; } case OP_CHECK_THIS: /* ensure ins->sreg1 is not NULL */ - ppc_load_reg (code, ppc_r0, 0, ins->sreg1); + ppc_ldptr (code, ppc_r0, 0, ins->sreg1); break; case OP_ARGLIST: { - if (ppc_is_imm16 (cfg->sig_cookie + cfg->stack_usage)) { - ppc_addi (code, ppc_r0, cfg->frame_reg, cfg->sig_cookie + cfg->stack_usage); + long cookie_offset = cfg->sig_cookie + cfg->stack_usage; + if (ppc_is_imm16 (cookie_offset)) { + ppc_addi (code, ppc_r0, cfg->frame_reg, cookie_offset); } else { - ppc_load (code, ppc_r0, cfg->sig_cookie + cfg->stack_usage); + ppc_load (code, ppc_r0, cookie_offset); ppc_add (code, ppc_r0, cfg->frame_reg, ppc_r0); } - ppc_store_reg (code, ppc_r0, 0, ins->sreg1); + ppc_stptr (code, ppc_r0, 0, ins->sreg1); break; } case OP_FCALL: @@ -3618,7 +3937,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) mono_add_patch_info (cfg, offset, MONO_PATCH_INFO_METHOD, call->method); else mono_add_patch_info (cfg, offset, MONO_PATCH_INFO_ABS, call->fptr); - if (FORCE_INDIR_CALL || cfg->method->dynamic) { + if ((FORCE_INDIR_CALL || cfg->method->dynamic) && !cfg->compile_aot) { ppc_load_func (code, ppc_r0, 0); ppc_mtlr (code, ppc_r0); ppc_blrl (code); @@ -3635,10 +3954,10 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) case OP_VOIDCALL_REG: case OP_CALL_REG: #ifdef PPC_USES_FUNCTION_DESCRIPTOR - ppc_load_reg (code, ppc_r0, 0, ins->sreg1); + ppc_ldptr (code, ppc_r0, 0, ins->sreg1); /* FIXME: if we know that this is a method, we can omit this load */ - ppc_load_reg (code, ppc_r2, 8, ins->sreg1); + ppc_ldptr (code, ppc_r2, 8, ins->sreg1); ppc_mtlr (code, ppc_r0); #else ppc_mtlr (code, ins->sreg1); @@ -3653,7 +3972,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) case OP_VCALL2_MEMBASE: case OP_VOIDCALL_MEMBASE: case OP_CALL_MEMBASE: - ppc_load_reg (code, ppc_r0, ins->inst_offset, ins->sreg1); + ppc_ldptr (code, ppc_r0, ins->inst_offset, ins->sreg1); ppc_mtlr (code, ppc_r0); ppc_blrl (code); /* FIXME: this should be handled somewhere else in the new jit */ @@ -3679,9 +3998,9 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) ppc_shift_right_arith_imm (code, ppc_r0, ppc_r0, 2); ppc_mtctr (code, ppc_r0); } - ppc_load_reg (code, ppc_r0, 0, ppc_sp); + ppc_ldptr (code, ppc_r0, 0, ppc_sp); ppc_neg (code, ppc_r11, ppc_r11); - ppc_store_reg_update_indexed (code, ppc_r0, ppc_sp, ppc_r11); + ppc_stptr_update_indexed (code, ppc_r0, ppc_sp, ppc_r11); /* FIXME: make this loop work in 8 byte increments on PPC64 */ @@ -3706,7 +4025,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) ppc_mr (code, ppc_r3, ins->sreg1); mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_INTERNAL_METHOD, (gpointer)"mono_arch_throw_exception"); - if (FORCE_INDIR_CALL || cfg->method->dynamic) { + if ((FORCE_INDIR_CALL || cfg->method->dynamic) && !cfg->compile_aot) { ppc_load_func (code, ppc_r0, 0); ppc_mtlr (code, ppc_r0); ppc_blrl (code); @@ -3720,7 +4039,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) ppc_mr (code, ppc_r3, ins->sreg1); mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_INTERNAL_METHOD, (gpointer)"mono_arch_rethrow_exception"); - if (FORCE_INDIR_CALL || cfg->method->dynamic) { + if ((FORCE_INDIR_CALL || cfg->method->dynamic) && !cfg->compile_aot) { ppc_load_func (code, ppc_r0, 0); ppc_mtlr (code, ppc_r0); ppc_blrl (code); @@ -3735,10 +4054,10 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) code = emit_reserve_param_area (cfg, code); ppc_mflr (code, ppc_r0); if (ppc_is_imm16 (spvar->inst_offset)) { - ppc_store_reg (code, ppc_r0, spvar->inst_offset, spvar->inst_basereg); + ppc_stptr (code, ppc_r0, spvar->inst_offset, spvar->inst_basereg); } else { ppc_load (code, ppc_r11, spvar->inst_offset); - ppc_store_reg_indexed (code, ppc_r0, ppc_r11, spvar->inst_basereg); + ppc_stptr_indexed (code, ppc_r0, ppc_r11, spvar->inst_basereg); } break; } @@ -3749,10 +4068,10 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) if (ins->sreg1 != ppc_r3) ppc_mr (code, ppc_r3, ins->sreg1); if (ppc_is_imm16 (spvar->inst_offset)) { - ppc_load_reg (code, ppc_r0, spvar->inst_offset, spvar->inst_basereg); + ppc_ldptr (code, ppc_r0, spvar->inst_offset, spvar->inst_basereg); } else { ppc_load (code, ppc_r11, spvar->inst_offset); - ppc_load_reg_indexed (code, ppc_r0, spvar->inst_basereg, ppc_r11); + ppc_ldptr_indexed (code, ppc_r0, spvar->inst_basereg, ppc_r11); } ppc_mtlr (code, ppc_r0); ppc_blr (code); @@ -3762,7 +4081,7 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) MonoInst *spvar = mono_find_spvar_for_region (cfg, bb->region); g_assert (spvar->inst_basereg != ppc_sp); code = emit_unreserve_param_area (cfg, code); - ppc_load_reg (code, ppc_r0, spvar->inst_offset, spvar->inst_basereg); + ppc_ldptr (code, ppc_r0, spvar->inst_offset, spvar->inst_basereg); ppc_mtlr (code, ppc_r0); ppc_blr (code); break; @@ -3775,22 +4094,12 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) ins->inst_c0 = code - cfg->native_code; break; case OP_BR: - if (ins->flags & MONO_INST_BRLABEL) { - /*if (ins->inst_i0->inst_c0) { - ppc_b (code, 0); - //x86_jump_code (code, cfg->native_code + ins->inst_i0->inst_c0); - } else*/ { - mono_add_patch_info (cfg, offset, MONO_PATCH_INFO_LABEL, ins->inst_i0); - ppc_b (code, 0); - } - } else { - /*if (ins->inst_target_bb->native_offset) { - ppc_b (code, 0); - //x86_jump_code (code, cfg->native_code + ins->inst_target_bb->native_offset); - } else*/ { - mono_add_patch_info (cfg, offset, MONO_PATCH_INFO_BB, ins->inst_target_bb); - ppc_b (code, 0); - } + /*if (ins->inst_target_bb->native_offset) { + ppc_b (code, 0); + //x86_jump_code (code, cfg->native_code + ins->inst_target_bb->native_offset); + } else*/ { + mono_add_patch_info (cfg, offset, MONO_PATCH_INFO_BB, ins->inst_target_bb); + ppc_b (code, 0); } break; case OP_BR_REG: @@ -3863,22 +4172,43 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) /* floating point opcodes */ case OP_R8CONST: + g_assert (cfg->compile_aot); + + /* FIXME: Optimize this */ + ppc_bl (code, 1); + ppc_mflr (code, ppc_r11); + ppc_b (code, 3); + *(double*)code = *(double*)ins->inst_p0; + code += 8; + ppc_lfd (code, ins->dreg, 8, ppc_r11); + break; case OP_R4CONST: g_assert_not_reached (); + break; case OP_STORER8_MEMBASE_REG: if (ppc_is_imm16 (ins->inst_offset)) { ppc_stfd (code, ins->sreg1, ins->inst_offset, ins->inst_destbasereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_stfdx (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset)) { + ppc_addis (code, ppc_r12, ins->inst_destbasereg, ppc_ha(ins->inst_offset)); + ppc_stfd (code, ins->sreg1, ins->inst_offset, ppc_r12); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_stfdx (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + } } break; case OP_LOADR8_MEMBASE: if (ppc_is_imm16 (ins->inst_offset)) { ppc_lfd (code, ins->dreg, ins->inst_offset, ins->inst_basereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_lfdx (code, ins->dreg, ins->inst_destbasereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset)) { + ppc_addis (code, ppc_r12, ins->inst_destbasereg, ppc_ha(ins->inst_offset)); + ppc_lfd (code, ins->dreg, ins->inst_offset, ppc_r12); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_lfdx (code, ins->dreg, ins->inst_destbasereg, ppc_r0); + } } break; case OP_STORER4_MEMBASE_REG: @@ -3886,30 +4216,40 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) if (ppc_is_imm16 (ins->inst_offset)) { ppc_stfs (code, ins->sreg1, ins->inst_offset, ins->inst_destbasereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_stfsx (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset)) { + ppc_addis (code, ppc_r12, ins->inst_destbasereg, ppc_ha(ins->inst_offset)); + ppc_stfs (code, ins->sreg1, ins->inst_offset, ppc_r12); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_stfsx (code, ins->sreg1, ins->inst_destbasereg, ppc_r0); + } } break; case OP_LOADR4_MEMBASE: if (ppc_is_imm16 (ins->inst_offset)) { ppc_lfs (code, ins->dreg, ins->inst_offset, ins->inst_basereg); } else { - ppc_load (code, ppc_r0, ins->inst_offset); - ppc_lfsx (code, ins->dreg, ins->inst_destbasereg, ppc_r0); + if (ppc_is_imm32 (ins->inst_offset)) { + ppc_addis (code, ppc_r12, ins->inst_destbasereg, ppc_ha(ins->inst_offset)); + ppc_lfs (code, ins->dreg, ins->inst_offset, ppc_r12); + } else { + ppc_load (code, ppc_r0, ins->inst_offset); + ppc_lfsx (code, ins->dreg, ins->inst_destbasereg, ppc_r0); + } } break; case OP_LOADR4_MEMINDEX: - ppc_lfsx (code, ins->dreg, ins->sreg2, ins->inst_basereg); + ppc_lfsx (code, ins->dreg, ins->inst_basereg, ins->sreg2); break; case OP_LOADR8_MEMINDEX: - ppc_lfdx (code, ins->dreg, ins->sreg2, ins->inst_basereg); + ppc_lfdx (code, ins->dreg, ins->inst_basereg, ins->sreg2); break; case OP_STORER4_MEMINDEX: ppc_frsp (code, ins->sreg1, ins->sreg1); - ppc_stfsx (code, ins->sreg1, ins->sreg2, ins->inst_destbasereg); + ppc_stfsx (code, ins->sreg1, ins->inst_destbasereg, ins->sreg2); break; case OP_STORER8_MEMINDEX: - ppc_stfdx (code, ins->sreg1, ins->sreg2, ins->inst_destbasereg); + ppc_stfdx (code, ins->sreg1, ins->inst_destbasereg, ins->sreg2); break; case CEE_CONV_R_UN: case CEE_CONV_R4: /* FIXME: change precision */ @@ -4072,7 +4412,11 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) break; case OP_JUMP_TABLE: mono_add_patch_info (cfg, offset, (MonoJumpInfoType)ins->inst_i1, ins->inst_p0); - ppc_load_sequence (code, ins->dreg, (gulong)0x0f0f0f0f0f0f0f0fL); +#ifdef __mono_ppc64__ + ppc_load_sequence (code, ins->dreg, (guint64)0x0f0f0f0f0f0f0f0fLL); +#else + ppc_load_sequence (code, ins->dreg, (gulong)0x0f0f0f0fL); +#endif break; } @@ -4096,8 +4440,12 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) } else { tmp = ins->sreg1; } - ppc_store_reg (code, tmp, -8, ppc_r1); - ppc_lfd (code, ins->dreg, -8, ppc_r1); + if (cpu_hw_caps & PPC_MOVE_FPR_GPR) { + ppc_mffgpr (code, ins->dreg, tmp); + } else { + ppc_str (code, tmp, -8, ppc_r1); + ppc_lfd (code, ins->dreg, -8, ppc_r1); + } ppc_fcfid (code, ins->dreg, ins->dreg); if (ins->opcode == OP_ICONV_TO_R4 || ins->opcode == OP_LCONV_TO_R4) ppc_frsp (code, ins->dreg, ins->dreg); @@ -4178,7 +4526,54 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) ppc_mr (code, ins->dreg, ppc_r0); break; } +#else + case OP_ICONV_TO_R4: + case OP_ICONV_TO_R8: { + if (cpu_hw_caps & PPC_ISA_64) { + ppc_srawi(code, ppc_r0, ins->sreg1, 31); + ppc_stw (code, ppc_r0, -8, ppc_r1); + ppc_stw (code, ins->sreg1, -4, ppc_r1); + ppc_lfd (code, ins->dreg, -8, ppc_r1); + ppc_fcfid (code, ins->dreg, ins->dreg); + if (ins->opcode == OP_ICONV_TO_R4) + ppc_frsp (code, ins->dreg, ins->dreg); + } + break; + } #endif + case OP_ATOMIC_CAS_I4: + CASE_PPC64 (OP_ATOMIC_CAS_I8) { + int location = ins->sreg1; + int value = ins->sreg2; + int comparand = ins->sreg3; + guint8 *start, *not_equal, *lost_reservation; + + start = code; + if (ins->opcode == OP_ATOMIC_CAS_I4) + ppc_lwarx (code, ppc_r0, 0, location); +#ifdef __mono_ppc64__ + else + ppc_ldarx (code, ppc_r0, 0, location); +#endif + ppc_cmp (code, 0, ins->opcode == OP_ATOMIC_CAS_I4 ? 0 : 1, ppc_r0, comparand); + + not_equal = code; + ppc_bc (code, PPC_BR_FALSE, PPC_BR_EQ, 0); + if (ins->opcode == OP_ATOMIC_CAS_I4) + ppc_stwcxd (code, value, 0, location); +#ifdef __mono_ppc64__ + else + ppc_stdcxd (code, value, 0, location); +#endif + + lost_reservation = code; + ppc_bc (code, PPC_BR_FALSE, PPC_BR_EQ, 0); + ppc_patch (lost_reservation, start); + + ppc_patch (not_equal, code); + ppc_mr (code, ins->dreg, ppc_r0); + break; + } default: g_warning ("unknown opcode %s in %s()\n", mono_inst_name (ins->opcode), __FUNCTION__); @@ -4203,15 +4598,18 @@ mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb) void mono_arch_register_lowlevel_calls (void) { + /* The signature doesn't matter */ + mono_register_jit_icall (mono_ppc_throw_exception, "mono_ppc_throw_exception", mono_create_icall_signature ("void"), TRUE); } #ifdef __mono_ppc64__ #define patch_load_sequence(ip,val) do {\ guint16 *__load = (guint16*)(ip); \ - __load [1] = (((guint64)(val)) >> 48) & 0xffff; \ - __load [3] = (((guint64)(val)) >> 32) & 0xffff; \ - __load [7] = (((guint64)(val)) >> 16) & 0xffff; \ - __load [9] = ((guint64)(val)) & 0xffff; \ + g_assert (sizeof (val) == sizeof (gsize)); \ + __load [1] = (((guint64)(gsize)(val)) >> 48) & 0xffff; \ + __load [3] = (((guint64)(gsize)(val)) >> 32) & 0xffff; \ + __load [7] = (((guint64)(gsize)(val)) >> 16) & 0xffff; \ + __load [9] = ((guint64)(gsize)(val)) & 0xffff; \ } while (0) #else #define patch_load_sequence(ip,val) do {\ @@ -4225,6 +4623,7 @@ void mono_arch_patch_code (MonoMethod *method, MonoDomain *domain, guint8 *code, MonoJumpInfo *ji, gboolean run_cctors) { MonoJumpInfo *patch_info; + gboolean compile_aot = !run_cctors; for (patch_info = ji; patch_info; patch_info = patch_info->next) { unsigned char *ip = patch_info->ip.i + code; @@ -4233,6 +4632,17 @@ mono_arch_patch_code (MonoMethod *method, MonoDomain *domain, guint8 *code, Mono target = mono_resolve_patch_target (method, domain, code, patch_info, run_cctors); + if (compile_aot) { + switch (patch_info->type) { + case MONO_PATCH_INFO_BB: + case MONO_PATCH_INFO_LABEL: + break; + default: + /* No need to patch these */ + continue; + } + } + switch (patch_info->type) { case MONO_PATCH_INFO_IP: patch_load_sequence (ip, ip); @@ -4301,22 +4711,24 @@ mono_arch_patch_code (MonoMethod *method, MonoDomain *domain, guint8 *code, Mono * the instruction offset immediate for all the registers. */ static guint8* -save_registers (guint8* code, int pos, int base_reg, gboolean save_lmf, guint32 used_int_regs) +save_registers (MonoCompile *cfg, guint8* code, int pos, int base_reg, gboolean save_lmf, guint32 used_int_regs, int cfa_offset) { int i; if (!save_lmf) { for (i = 13; i <= 31; i++) { if (used_int_regs & (1 << i)) { - ppc_store_reg (code, i, pos, base_reg); - pos += sizeof (gulong); + ppc_str (code, i, pos, base_reg); + mono_emit_unwind_op_offset (cfg, code, i, pos - cfa_offset); + pos += sizeof (mgreg_t); } } } else { /* pos is the start of the MonoLMF structure */ int offset = pos + G_STRUCT_OFFSET (MonoLMF, iregs); for (i = 13; i <= 31; i++) { - ppc_store_reg (code, i, offset, base_reg); - offset += sizeof (gulong); + ppc_str (code, i, offset, base_reg); + mono_emit_unwind_op_offset (cfg, code, i, offset - cfa_offset); + offset += sizeof (mgreg_t); } offset = pos + G_STRUCT_OFFSET (MonoLMF, fregs); for (i = 14; i < 32; i++) { @@ -4352,7 +4764,8 @@ mono_arch_emit_prolog (MonoCompile *cfg) MonoBasicBlock *bb; MonoMethodSignature *sig; MonoInst *inst; - int alloc_size, pos, max_offset, i; + long alloc_size, pos, max_offset, cfa_offset; + int i; guint8 *code; CallInfo *cinfo; int tracing = 0; @@ -4366,9 +4779,15 @@ mono_arch_emit_prolog (MonoCompile *cfg) cfg->code_size = MONO_PPC_32_64_CASE (260, 384) + sig->param_count * 20; code = cfg->native_code = g_malloc (cfg->code_size); + cfa_offset = 0; + + /* We currently emit unwind info for aot, but don't use it */ + mono_emit_unwind_op_def_cfa (cfg, code, ppc_r1, 0); + if (1 || cfg->flags & MONO_CFG_HAS_CALLS) { ppc_mflr (code, ppc_r0); - ppc_store_reg (code, ppc_r0, PPC_RET_ADDR_OFFSET, ppc_sp); + ppc_str (code, ppc_r0, PPC_RET_ADDR_OFFSET, ppc_sp); + mono_emit_unwind_op_offset (cfg, code, ppc_lr, PPC_RET_ADDR_OFFSET); } alloc_size = cfg->stack_offset; @@ -4377,7 +4796,7 @@ mono_arch_emit_prolog (MonoCompile *cfg) if (!method->save_lmf) { for (i = 31; i >= 13; --i) { if (cfg->used_int_regs & (1 << i)) { - pos += sizeof (gulong); + pos += sizeof (mgreg_t); } } } else { @@ -4395,25 +4814,31 @@ mono_arch_emit_prolog (MonoCompile *cfg) g_assert ((alloc_size & (MONO_ARCH_FRAME_ALIGNMENT-1)) == 0); if (alloc_size) { if (ppc_is_imm16 (-alloc_size)) { - ppc_store_reg_update (code, ppc_sp, -alloc_size, ppc_sp); - code = save_registers (code, alloc_size - pos, ppc_sp, method->save_lmf, cfg->used_int_regs); + ppc_str_update (code, ppc_sp, -alloc_size, ppc_sp); + cfa_offset = alloc_size; + mono_emit_unwind_op_def_cfa_offset (cfg, code, alloc_size); + code = save_registers (cfg, code, alloc_size - pos, ppc_sp, method->save_lmf, cfg->used_int_regs, cfa_offset); } else { if (pos) ppc_addi (code, ppc_r11, ppc_sp, -pos); ppc_load (code, ppc_r0, -alloc_size); - ppc_store_reg_update_indexed (code, ppc_sp, ppc_sp, ppc_r0); - code = save_registers (code, 0, ppc_r11, method->save_lmf, cfg->used_int_regs); + ppc_str_update_indexed (code, ppc_sp, ppc_sp, ppc_r0); + cfa_offset = alloc_size; + mono_emit_unwind_op_def_cfa_offset (cfg, code, alloc_size); + code = save_registers (cfg, code, 0, ppc_r11, method->save_lmf, cfg->used_int_regs, cfa_offset); } } - if (cfg->frame_reg != ppc_sp) + if (cfg->frame_reg != ppc_sp) { ppc_mr (code, cfg->frame_reg, ppc_sp); + mono_emit_unwind_op_def_cfa_reg (cfg, code, cfg->frame_reg); + } /* store runtime generic context */ if (cfg->rgctx_var) { g_assert (cfg->rgctx_var->opcode == OP_REGOFFSET && (cfg->rgctx_var->inst_basereg == ppc_r1 || cfg->rgctx_var->inst_basereg == ppc_r31)); - ppc_store_reg (code, MONO_ARCH_RGCTX_REG, cfg->rgctx_var->inst_offset, cfg->rgctx_var->inst_basereg); + ppc_stptr (code, MONO_ARCH_RGCTX_REG, cfg->rgctx_var->inst_offset, cfg->rgctx_var->inst_basereg); } /* compute max_offset in order to use short forward jumps @@ -4444,10 +4869,10 @@ mono_arch_emit_prolog (MonoCompile *cfg) g_assert (inst); if (ppc_is_imm16 (inst->inst_offset)) { - ppc_store_reg (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); + ppc_stptr (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); } else { ppc_load (code, ppc_r11, inst->inst_offset); - ppc_store_reg_indexed (code, ainfo->reg, ppc_r11, inst->inst_basereg); + ppc_stptr_indexed (code, ainfo->reg, ppc_r11, inst->inst_basereg); } } @@ -4464,13 +4889,13 @@ mono_arch_emit_prolog (MonoCompile *cfg) else if (ainfo->regtype == RegTypeFP) ppc_fmr (code, inst->dreg, ainfo->reg); else if (ainfo->regtype == RegTypeBase) { - ppc_load_reg (code, ppc_r11, 0, ppc_sp); - ppc_load_reg (code, inst->dreg, ainfo->offset, ppc_r11); + ppc_ldr (code, ppc_r11, 0, ppc_sp); + ppc_ldptr (code, inst->dreg, ainfo->offset, ppc_r11); } else g_assert_not_reached (); if (cfg->verbose_level > 2) - g_print ("Argument %d assigned to register %s\n", pos, mono_arch_regname (inst->dreg)); + g_print ("Argument %ld assigned to register %s\n", pos, mono_arch_regname (inst->dreg)); } else { /* the argument should be put on the stack: FIXME handle size != word */ if (ainfo->regtype == RegTypeGeneral) { @@ -4479,25 +4904,48 @@ mono_arch_emit_prolog (MonoCompile *cfg) if (ppc_is_imm16 (inst->inst_offset)) { ppc_stb (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); } else { - ppc_load (code, ppc_r11, inst->inst_offset); - ppc_stbx (code, ainfo->reg, ppc_r11, inst->inst_basereg); + if (ppc_is_imm32 (inst->inst_offset)) { + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_stb (code, ainfo->reg, ppc_r11, inst->inst_offset); + } else { + ppc_load (code, ppc_r11, inst->inst_offset); + ppc_stbx (code, ainfo->reg, inst->inst_basereg, ppc_r11); + } } break; case 2: if (ppc_is_imm16 (inst->inst_offset)) { ppc_sth (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); } else { - ppc_load (code, ppc_r11, inst->inst_offset); - ppc_sthx (code, ainfo->reg, ppc_r11, inst->inst_basereg); + if (ppc_is_imm32 (inst->inst_offset)) { + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_sth (code, ainfo->reg, ppc_r11, inst->inst_offset); + } else { + ppc_load (code, ppc_r11, inst->inst_offset); + ppc_sthx (code, ainfo->reg, inst->inst_basereg, ppc_r11); + } } break; #ifdef __mono_ppc64__ case 4: if (ppc_is_imm16 (inst->inst_offset)) { ppc_stw (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); + } else { + if (ppc_is_imm32 (inst->inst_offset)) { + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_stw (code, ainfo->reg, ppc_r11, inst->inst_offset); + } else { + ppc_load (code, ppc_r11, inst->inst_offset); + ppc_stwx (code, ainfo->reg, inst->inst_basereg, ppc_r11); + } + } + break; + case 8: + if (ppc_is_imm16 (inst->inst_offset)) { + ppc_str (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); } else { ppc_load (code, ppc_r11, inst->inst_offset); - ppc_stwx (code, ainfo->reg, ppc_r11, inst->inst_basereg); + ppc_str_indexed (code, ainfo->reg, ppc_r11, inst->inst_basereg); } break; #else @@ -4506,8 +4954,8 @@ mono_arch_emit_prolog (MonoCompile *cfg) ppc_stw (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); ppc_stw (code, ainfo->reg + 1, inst->inst_offset + 4, inst->inst_basereg); } else { - ppc_load (code, ppc_r11, inst->inst_offset); - ppc_add (code, ppc_r11, ppc_r11, inst->inst_basereg); + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_addi (code, ppc_r11, ppc_r11, inst->inst_offset); ppc_stw (code, ainfo->reg, 0, ppc_r11); ppc_stw (code, ainfo->reg + 1, 4, ppc_r11); } @@ -4515,61 +4963,100 @@ mono_arch_emit_prolog (MonoCompile *cfg) #endif default: if (ppc_is_imm16 (inst->inst_offset)) { - ppc_store_reg (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); + ppc_stptr (code, ainfo->reg, inst->inst_offset, inst->inst_basereg); } else { - ppc_load (code, ppc_r11, inst->inst_offset); - ppc_store_reg_indexed (code, ainfo->reg, ppc_r11, inst->inst_basereg); + if (ppc_is_imm32 (inst->inst_offset)) { + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_stptr (code, ainfo->reg, ppc_r11, inst->inst_offset); + } else { + ppc_load (code, ppc_r11, inst->inst_offset); + ppc_stptr_indexed (code, ainfo->reg, inst->inst_basereg, ppc_r11); + } } break; } } else if (ainfo->regtype == RegTypeBase) { + g_assert (ppc_is_imm16 (ainfo->offset)); /* load the previous stack pointer in r11 */ - ppc_load_reg (code, ppc_r11, 0, ppc_sp); - ppc_load_reg (code, ppc_r0, ainfo->offset, ppc_r11); + ppc_ldr (code, ppc_r11, 0, ppc_sp); + ppc_ldptr (code, ppc_r0, ainfo->offset, ppc_r11); switch (ainfo->size) { case 1: if (ppc_is_imm16 (inst->inst_offset)) { ppc_stb (code, ppc_r0, inst->inst_offset, inst->inst_basereg); } else { - ppc_load (code, ppc_r11, inst->inst_offset); - ppc_stbx (code, ppc_r0, ppc_r11, inst->inst_basereg); + if (ppc_is_imm32 (inst->inst_offset)) { + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_stb (code, ppc_r0, ppc_r11, inst->inst_offset); + } else { + ppc_load (code, ppc_r11, inst->inst_offset); + ppc_stbx (code, ppc_r0, inst->inst_basereg, ppc_r11); + } } break; case 2: if (ppc_is_imm16 (inst->inst_offset)) { ppc_sth (code, ppc_r0, inst->inst_offset, inst->inst_basereg); } else { - ppc_load (code, ppc_r11, inst->inst_offset); - ppc_sthx (code, ppc_r0, ppc_r11, inst->inst_basereg); + if (ppc_is_imm32 (inst->inst_offset)) { + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_sth (code, ppc_r0, ppc_r11, inst->inst_offset); + } else { + ppc_load (code, ppc_r11, inst->inst_offset); + ppc_sthx (code, ppc_r0, inst->inst_basereg, ppc_r11); + } } break; #ifdef __mono_ppc64__ case 4: if (ppc_is_imm16 (inst->inst_offset)) { ppc_stw (code, ppc_r0, inst->inst_offset, inst->inst_basereg); + } else { + if (ppc_is_imm32 (inst->inst_offset)) { + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_stw (code, ppc_r0, ppc_r11, inst->inst_offset); + } else { + ppc_load (code, ppc_r11, inst->inst_offset); + ppc_stwx (code, ppc_r0, inst->inst_basereg, ppc_r11); + } + } + break; + case 8: + if (ppc_is_imm16 (inst->inst_offset)) { + ppc_str (code, ppc_r0, inst->inst_offset, inst->inst_basereg); } else { ppc_load (code, ppc_r11, inst->inst_offset); - ppc_stwx (code, ppc_r0, ppc_r11, inst->inst_basereg); + ppc_str_indexed (code, ppc_r0, ppc_r11, inst->inst_basereg); } break; #else case 8: + g_assert (ppc_is_imm16 (ainfo->offset + 4)); if (ppc_is_imm16 (inst->inst_offset + 4)) { ppc_stw (code, ppc_r0, inst->inst_offset, inst->inst_basereg); ppc_lwz (code, ppc_r0, ainfo->offset + 4, ppc_r11); ppc_stw (code, ppc_r0, inst->inst_offset + 4, inst->inst_basereg); } else { - /* FIXME */ - g_assert_not_reached (); + /* use r12 to load the 2nd half of the long before we clobber r11. */ + ppc_lwz (code, ppc_r12, ainfo->offset + 4, ppc_r11); + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_addi (code, ppc_r11, ppc_r11, inst->inst_offset); + ppc_stw (code, ppc_r0, 0, ppc_r11); + ppc_stw (code, ppc_r12, 4, ppc_r11); } break; #endif default: if (ppc_is_imm16 (inst->inst_offset)) { - ppc_store_reg (code, ppc_r0, inst->inst_offset, inst->inst_basereg); + ppc_stptr (code, ppc_r0, inst->inst_offset, inst->inst_basereg); } else { - ppc_load (code, ppc_r11, inst->inst_offset); - ppc_store_reg_indexed (code, ppc_r0, ppc_r11, inst->inst_basereg); + if (ppc_is_imm32 (inst->inst_offset)) { + ppc_addis (code, ppc_r11, inst->inst_basereg, ppc_ha(inst->inst_offset)); + ppc_stptr (code, ppc_r0, ppc_r11, inst->inst_offset); + } else { + ppc_load (code, ppc_r11, inst->inst_offset); + ppc_stptr_indexed (code, ppc_r0, inst->inst_basereg, ppc_r11); + } } break; } @@ -4587,11 +5074,11 @@ mono_arch_emit_prolog (MonoCompile *cfg) int cur_reg; int size = 0; g_assert (ppc_is_imm16 (inst->inst_offset)); - g_assert (ppc_is_imm16 (inst->inst_offset + ainfo->size * sizeof (gpointer))); + g_assert (ppc_is_imm16 (inst->inst_offset + ainfo->vtregs * sizeof (gpointer))); /* FIXME: what if there is no class? */ if (sig->pinvoke && mono_class_from_mono_type (inst->inst_vtype)) size = mono_class_native_size (mono_class_from_mono_type (inst->inst_vtype), NULL); - for (cur_reg = 0; cur_reg < ainfo->size; ++cur_reg) { + for (cur_reg = 0; cur_reg < ainfo->vtregs; ++cur_reg) { #if __APPLE__ /* * Darwin handles 1 and 2 byte @@ -4612,11 +5099,11 @@ mono_arch_emit_prolog (MonoCompile *cfg) g_assert (cur_reg == 0); ppc_sldi (code, ppc_r0, ainfo->reg, (sizeof (gpointer) - ainfo->bytes) * 8); - ppc_store_reg (code, ppc_r0, doffset, inst->inst_basereg); + ppc_stptr (code, ppc_r0, doffset, inst->inst_basereg); } else #endif { - ppc_store_reg (code, ainfo->reg + cur_reg, doffset, + ppc_stptr (code, ainfo->reg + cur_reg, doffset, inst->inst_basereg); } } @@ -4628,7 +5115,7 @@ mono_arch_emit_prolog (MonoCompile *cfg) if (ainfo->bytes) NOT_IMPLEMENTED; /* load the previous stack pointer in r11 (r0 gets overwritten by the memcpy) */ - ppc_load_reg (code, ppc_r11, 0, ppc_sp); + ppc_ldr (code, ppc_r11, 0, ppc_sp); if ((size & MONO_PPC_32_64_CASE (3, 7)) != 0) { code = emit_memcpy (code, size - soffset, inst->inst_basereg, doffset, @@ -4643,8 +5130,8 @@ mono_arch_emit_prolog (MonoCompile *cfg) /* if it was originally a RegTypeBase */ if (ainfo->offset) { /* load the previous stack pointer in r11 */ - ppc_load_reg (code, ppc_r11, 0, ppc_sp); - ppc_load_reg (code, ppc_r11, ainfo->offset, ppc_r11); + ppc_ldr (code, ppc_r11, 0, ppc_sp); + ppc_ldptr (code, ppc_r11, ainfo->offset, ppc_r11); } else { ppc_mr (code, ppc_r11, ainfo->reg); } @@ -4653,7 +5140,7 @@ mono_arch_emit_prolog (MonoCompile *cfg) MonoInst *addr = cfg->tailcall_valuetype_addrs [tailcall_struct_index]; g_assert (ppc_is_imm16 (addr->inst_offset)); - ppc_store_reg (code, ppc_r11, addr->inst_offset, addr->inst_basereg); + ppc_stptr (code, ppc_r11, addr->inst_offset, addr->inst_basereg); tailcall_struct_index++; } @@ -4668,13 +5155,14 @@ mono_arch_emit_prolog (MonoCompile *cfg) } if (method->wrapper_type == MONO_WRAPPER_NATIVE_TO_MANAGED) { - ppc_load (code, ppc_r3, cfg->domain); + ppc_load_ptr (code, ppc_r3, cfg->domain); mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_INTERNAL_METHOD, (gpointer)"mono_jit_thread_attach"); - if (FORCE_INDIR_CALL || cfg->method->dynamic) { + if ((FORCE_INDIR_CALL || cfg->method->dynamic) && !cfg->compile_aot) { ppc_load_func (code, ppc_r0, 0); ppc_mtlr (code, ppc_r0); ppc_blrl (code); } else { + // FIXME: AOT ppc_bl (code, 0); } } @@ -4682,12 +5170,16 @@ mono_arch_emit_prolog (MonoCompile *cfg) if (method->save_lmf) { if (lmf_pthread_key != -1) { emit_tls_access (code, ppc_r3, lmf_pthread_key); - if (G_STRUCT_OFFSET (MonoJitTlsData, lmf)) + if (tls_mode != TLS_MODE_NPTL && G_STRUCT_OFFSET (MonoJitTlsData, lmf)) ppc_addi (code, ppc_r3, ppc_r3, G_STRUCT_OFFSET (MonoJitTlsData, lmf)); } else { + if (cfg->compile_aot) { + /* Compute the got address which is needed by the PLT entry */ + code = mono_arch_emit_load_got_addr (cfg->native_code, code, cfg, NULL); + } mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_INTERNAL_METHOD, (gpointer)"mono_get_lmf_addr"); - if (FORCE_INDIR_CALL || cfg->method->dynamic) { + if ((FORCE_INDIR_CALL || cfg->method->dynamic) && !cfg->compile_aot) { ppc_load_func (code, ppc_r0, 0); ppc_mtlr (code, ppc_r0); ppc_blrl (code); @@ -4704,20 +5196,33 @@ mono_arch_emit_prolog (MonoCompile *cfg) */ ppc_addi (code, ppc_r11, ppc_sp, alloc_size - lmf_offset); /* ppc_r3 is the result from mono_get_lmf_addr () */ - ppc_store_reg (code, ppc_r3, G_STRUCT_OFFSET(MonoLMF, lmf_addr), ppc_r11); + ppc_stptr (code, ppc_r3, G_STRUCT_OFFSET(MonoLMF, lmf_addr), ppc_r11); /* new_lmf->previous_lmf = *lmf_addr */ - ppc_load_reg (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r3); - ppc_store_reg (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r11); + ppc_ldptr (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r3); + ppc_stptr (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r11); /* *(lmf_addr) = r11 */ - ppc_store_reg (code, ppc_r11, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r3); + ppc_stptr (code, ppc_r11, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r3); /* save method info */ - ppc_load (code, ppc_r0, method); - ppc_store_reg (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, method), ppc_r11); - ppc_store_reg (code, ppc_sp, G_STRUCT_OFFSET(MonoLMF, ebp), ppc_r11); + if (cfg->compile_aot) + // FIXME: + ppc_load (code, ppc_r0, 0); + else + ppc_load_ptr (code, ppc_r0, method); + ppc_stptr (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, method), ppc_r11); + ppc_stptr (code, ppc_sp, G_STRUCT_OFFSET(MonoLMF, ebp), ppc_r11); /* save the current IP */ - mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_IP, NULL); - ppc_load_sequence (code, ppc_r0, (gulong)0x0101010101010101L); - ppc_store_reg (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, eip), ppc_r11); + if (cfg->compile_aot) { + ppc_bl (code, 1); + ppc_mflr (code, ppc_r0); + } else { + mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_IP, NULL); +#ifdef __mono_ppc64__ + ppc_load_sequence (code, ppc_r0, (guint64)0x0101010101010101LL); +#else + ppc_load_sequence (code, ppc_r0, (gulong)0x01010101L); +#endif + } + ppc_stptr (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, eip), ppc_r11); } if (tracing) @@ -4771,16 +5276,16 @@ mono_arch_emit_epilog (MonoCompile *cfg) ppc_mr (code, ppc_r8, cfg->frame_reg); ppc_addi (code, ppc_r11, cfg->frame_reg, cfg->stack_usage - lmf_offset); /* r5 = previous_lmf */ - ppc_load_reg (code, ppc_r5, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r11); + ppc_ldptr (code, ppc_r5, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r11); /* r6 = lmf_addr */ - ppc_load_reg (code, ppc_r6, G_STRUCT_OFFSET(MonoLMF, lmf_addr), ppc_r11); + ppc_ldptr (code, ppc_r6, G_STRUCT_OFFSET(MonoLMF, lmf_addr), ppc_r11); /* *(lmf_addr) = previous_lmf */ - ppc_store_reg (code, ppc_r5, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r6); + ppc_stptr (code, ppc_r5, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r6); /* FIXME: speedup: there is no actual need to restore the registers if * we didn't actually change them (idea from Zoltan). */ /* restore iregs */ - ppc_load_multiple_regs (code, ppc_r13, ppc_r11, G_STRUCT_OFFSET(MonoLMF, iregs)); + ppc_ldr_multiple (code, ppc_r13, G_STRUCT_OFFSET(MonoLMF, iregs), ppc_r11); /* restore fregs */ /*for (i = 14; i < 32; i++) { ppc_lfd (code, i, G_STRUCT_OFFSET(MonoLMF, fregs) + ((i-14) * sizeof (gdouble)), ppc_r11); @@ -4788,17 +5293,18 @@ mono_arch_emit_epilog (MonoCompile *cfg) g_assert (ppc_is_imm16 (cfg->stack_usage + PPC_RET_ADDR_OFFSET)); /* use the saved copy of the frame reg in r8 */ if (1 || cfg->flags & MONO_CFG_HAS_CALLS) { - ppc_load_reg (code, ppc_r0, cfg->stack_usage + PPC_RET_ADDR_OFFSET, ppc_r8); + ppc_ldr (code, ppc_r0, cfg->stack_usage + PPC_RET_ADDR_OFFSET, ppc_r8); ppc_mtlr (code, ppc_r0); } ppc_addic (code, ppc_sp, ppc_r8, cfg->stack_usage); } else { if (1 || cfg->flags & MONO_CFG_HAS_CALLS) { - if (ppc_is_imm16 (cfg->stack_usage + PPC_RET_ADDR_OFFSET)) { - ppc_load_reg (code, ppc_r0, cfg->stack_usage + PPC_RET_ADDR_OFFSET, cfg->frame_reg); + long return_offset = cfg->stack_usage + PPC_RET_ADDR_OFFSET; + if (ppc_is_imm16 (return_offset)) { + ppc_ldr (code, ppc_r0, return_offset, cfg->frame_reg); } else { - ppc_load (code, ppc_r11, cfg->stack_usage + PPC_RET_ADDR_OFFSET); - ppc_load_reg_indexed (code, ppc_r0, cfg->frame_reg, ppc_r11); + ppc_load (code, ppc_r11, return_offset); + ppc_ldr_indexed (code, ppc_r0, cfg->frame_reg, ppc_r11); } ppc_mtlr (code, ppc_r0); } @@ -4806,29 +5312,29 @@ mono_arch_emit_epilog (MonoCompile *cfg) int offset = cfg->stack_usage; for (i = 13; i <= 31; i++) { if (cfg->used_int_regs & (1 << i)) - offset -= sizeof (gulong); + offset -= sizeof (mgreg_t); } if (cfg->frame_reg != ppc_sp) ppc_mr (code, ppc_r11, cfg->frame_reg); /* note r31 (possibly the frame register) is restored last */ for (i = 13; i <= 31; i++) { if (cfg->used_int_regs & (1 << i)) { - ppc_load_reg (code, i, offset, cfg->frame_reg); - offset += sizeof (gulong); + ppc_ldr (code, i, offset, cfg->frame_reg); + offset += sizeof (mgreg_t); } } if (cfg->frame_reg != ppc_sp) - ppc_addic (code, ppc_sp, ppc_r11, cfg->stack_usage); + ppc_addi (code, ppc_sp, ppc_r11, cfg->stack_usage); else - ppc_addic (code, ppc_sp, ppc_sp, cfg->stack_usage); + ppc_addi (code, ppc_sp, ppc_sp, cfg->stack_usage); } else { - ppc_load (code, ppc_r11, cfg->stack_usage); + ppc_load32 (code, ppc_r11, cfg->stack_usage); if (cfg->used_int_regs) { ppc_add (code, ppc_r11, cfg->frame_reg, ppc_r11); for (i = 31; i >= 13; --i) { if (cfg->used_int_regs & (1 << i)) { - pos += sizeof (gulong); - ppc_load_reg (code, i, -pos, ppc_r11); + pos += sizeof (mgreg_t); + ppc_ldr (code, i, -pos, ppc_r11); } } ppc_mr (code, ppc_sp, ppc_r11); @@ -4882,13 +5388,13 @@ mono_arch_emit_exceptions (MonoCompile *cfg) /* * make sure we have enough space for exceptions - * 24 is the simulated call to throw_exception_by_name + * 28 is the simulated call to throw_corlib_exception */ for (patch_info = cfg->patch_info; patch_info; patch_info = patch_info->next) { if (patch_info->type == MONO_PATCH_INFO_EXC) { i = exception_id_by_name (patch_info->data.target); if (!exc_throw_found [i]) { - max_epilog_size += 24; + max_epilog_size += 28; exc_throw_found [i] = TRUE; } } else if (patch_info->type == MONO_PATCH_INFO_BB_OVF) @@ -4897,7 +5403,7 @@ mono_arch_emit_exceptions (MonoCompile *cfg) MonoOvfJump *ovfj = (MonoOvfJump*)patch_info->data.target; i = exception_id_by_name (ovfj->data.exception); if (!exc_throw_found [i]) { - max_epilog_size += 24; + max_epilog_size += 28; exc_throw_found [i] = TRUE; } max_epilog_size += 8; @@ -4927,6 +5433,7 @@ mono_arch_emit_exceptions (MonoCompile *cfg) ppc_b (code, 0); ip = ovfj->data.bb->native_offset + cfg->native_code; ppc_patch (code - 4, ip); + patch_info->type = MONO_PATCH_INFO_NONE; break; } case MONO_PATCH_INFO_EXC_OVF: { @@ -4947,9 +5454,12 @@ mono_arch_emit_exceptions (MonoCompile *cfg) newji->data.target = ovfj->data.exception; newji->next = patch_info->next; patch_info->next = newji; + patch_info->type = MONO_PATCH_INFO_NONE; break; } case MONO_PATCH_INFO_EXC: { + MonoClass *exc_class; + unsigned char *ip = patch_info->ip.i + cfg->native_code; i = exception_id_by_name (patch_info->data.target); if (exc_throw_pos [i]) { @@ -4959,19 +5469,24 @@ mono_arch_emit_exceptions (MonoCompile *cfg) } else { exc_throw_pos [i] = code; } + + exc_class = mono_class_from_name (mono_defaults.corlib, "System", patch_info->data.name); + g_assert (exc_class); + ppc_patch (ip, code); /*mono_add_patch_info (cfg, code - cfg->native_code, MONO_PATCH_INFO_EXC_NAME, patch_info->data.target);*/ - ppc_load (code, ppc_r3, patch_info->data.target); - /* we got here from a conditional call, so the calling ip is set in lr already */ + ppc_load (code, ppc_r3, exc_class->type_token); + /* we got here from a conditional call, so the calling ip is set in lr */ + ppc_mflr (code, ppc_r4); patch_info->type = MONO_PATCH_INFO_INTERNAL_METHOD; - patch_info->data.name = "mono_arch_throw_exception_by_name"; + patch_info->data.name = "mono_arch_throw_corlib_exception"; patch_info->ip.i = code - cfg->native_code; if (FORCE_INDIR_CALL || cfg->method->dynamic) { ppc_load_func (code, ppc_r0, 0); ppc_mtctr (code, ppc_r0); ppc_bcctr (code, PPC_BR_ALWAYS, 0); } else { - ppc_b (code, 0); + ppc_bl (code, 0); } break; } @@ -4983,10 +5498,10 @@ mono_arch_emit_exceptions (MonoCompile *cfg) cfg->code_len = code - cfg->native_code; - g_assert (cfg->code_len < cfg->code_size); - + g_assert (cfg->code_len <= cfg->code_size); } +#if DEAD_CODE static int try_offset_access (void *value, guint32 idx) { @@ -5000,27 +5515,41 @@ try_offset_access (void *value, guint32 idx) return 0; return 1; } +#endif static void setup_tls_access (void) { -#ifdef __mono_ppc64__ - /* FIXME: implement */ - tls_mode = TLS_MODE_FAILED; - return; -#else guint32 ptk; + +#if defined(__linux__) && defined(_CS_GNU_LIBPTHREAD_VERSION) + size_t conf_size = 0; + char confbuf[128]; +#else + /* FIXME for darwin */ guint32 *ins, *code; guint32 cmplwi_1023, li_0x48, blr_ins; +#endif + +#ifdef TARGET_PS3 + tls_mode = TLS_MODE_FAILED; +#endif + if (tls_mode == TLS_MODE_FAILED) return; - if (g_getenv ("MONO_NO_TLS")) { tls_mode = TLS_MODE_FAILED; return; } - if (tls_mode == TLS_MODE_DETECT) { + if (tls_mode == TLS_MODE_DETECT) { +#if defined(__APPLE__) && defined(__mono_ppc__) && !defined(__mono_ppc64__) + tls_mode = TLS_MODE_DARWIN_G4; +#elif defined(__linux__) && defined(_CS_GNU_LIBPTHREAD_VERSION) + conf_size = confstr ( _CS_GNU_LIBPTHREAD_VERSION, confbuf, sizeof(confbuf)); + if ((conf_size > 4) && (strncmp (confbuf, "NPTL", 4) == 0)) + tls_mode = TLS_MODE_NPTL; +#elif !defined(TARGET_PS3) ins = (guint32*)pthread_getspecific; /* uncond branch to the real method */ if ((*ins >> 26) == 18) { @@ -5029,7 +5558,7 @@ setup_tls_access (void) val >>= 6; if (*ins & 2) { /* absolute */ - ins = (guint32*)val; + ins = (guint32*)(long)val; } else { ins = (guint32*) ((char*)ins + val); } @@ -5065,7 +5594,7 @@ setup_tls_access (void) val >>= 6; if (*ins & 2) { /* absolute */ - ins = (guint32*)val; + ins = (guint32*)(long)val; } else { ins = (guint32*) ((char*)ins + val); } @@ -5092,7 +5621,18 @@ setup_tls_access (void) tls_mode = TLS_MODE_FAILED; return; } +#endif } +#ifndef TARGET_PS3 + if (tls_mode == TLS_MODE_DETECT) + tls_mode = TLS_MODE_FAILED; + if (tls_mode == TLS_MODE_FAILED) + return; + if ((monodomain_key == -1) && (tls_mode == TLS_MODE_NPTL)) { + monodomain_key = mono_domain_get_tls_offset(); + } + /* if not TLS_MODE_NPTL or local dynamic (as indicated by + mono_domain_get_tls_offset returning -1) then use keyed access. */ if (monodomain_key == -1) { ptk = mono_domain_get_tls_key (); if (ptk < 1024) { @@ -5102,6 +5642,12 @@ setup_tls_access (void) } } } + + if ((lmf_pthread_key == -1) && (tls_mode == TLS_MODE_NPTL)) { + lmf_pthread_key = mono_get_lmf_addr_tls_offset(); + } + /* if not TLS_MODE_NPTL or local dynamic (as indicated by + mono_get_lmf_addr_tls_offset returning -1) then use keyed access. */ if (lmf_pthread_key == -1) { ptk = mono_pthread_key_for_tls (mono_jit_tls_id); if (ptk < 1024) { @@ -5113,18 +5659,6 @@ setup_tls_access (void) lmf_pthread_key = ptk; } } - if (monothread_key == -1) { - ptk = mono_thread_get_tls_key (); - if (ptk < 1024) { - ptk = mono_pthread_key_for_tls (ptk); - if (ptk < 1024) { - monothread_key = ptk; - /*g_print ("thread inited: %d\n", ptk);*/ - } - } else { - /*g_print ("thread not inited yet %d\n", ptk);*/ - } - } #endif } @@ -5165,13 +5699,15 @@ mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckI if (item->check_target_idx) { if (!item->compare_done) item->chunk_size += CMP_SIZE; - if (fail_tramp) + if (item->has_target_code) item->chunk_size += BR_SIZE + JUMP_IMM32_SIZE; else item->chunk_size += LOADSTORE_SIZE + BR_SIZE + JUMP_IMM_SIZE; } else { if (fail_tramp) { item->chunk_size += CMP_SIZE + BR_SIZE + JUMP_IMM32_SIZE * 2; + if (!item->has_target_code) + item->chunk_size += LOADSTORE_SIZE; } else { item->chunk_size += LOADSTORE_SIZE + JUMP_IMM_SIZE; #if ENABLE_WRONG_METHOD_CHECK @@ -5190,7 +5726,7 @@ mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckI } else { /* the initial load of the vtable address */ size += PPC_LOAD_SEQUENCE_LENGTH + LOADSTORE_SIZE; - code = mono_code_manager_reserve (domain->code_mp, size); + code = mono_domain_code_reserve (domain, size); } start = code; if (!fail_tramp) { @@ -5202,8 +5738,8 @@ mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckI * FIXME: Get rid of this by making sure that r11 is * not used as the vtable register in interface calls. */ - ppc_store_reg (code, ppc_r11, PPC_RET_ADDR_OFFSET, ppc_sp); - ppc_load (code, ppc_r11, (gulong)(& (vtable->vtable [0]))); + ppc_stptr (code, ppc_r11, PPC_RET_ADDR_OFFSET, ppc_sp); + ppc_load (code, ppc_r11, (gsize)(& (vtable->vtable [0]))); } for (i = 0; i < count; ++i) { MonoIMTCheckItem *item = imt_entries [i]; @@ -5211,16 +5747,16 @@ mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckI if (item->is_equals) { if (item->check_target_idx) { if (!item->compare_done) { - ppc_load (code, ppc_r0, (gulong)item->key); + ppc_load (code, ppc_r0, (gsize)item->key); ppc_compare_log (code, 0, MONO_ARCH_IMT_REG, ppc_r0); } item->jmp_code = code; ppc_bc (code, PPC_BR_FALSE, PPC_BR_EQ, 0); - if (fail_tramp) { - ppc_load (code, ppc_r0, item->value.target_code); + if (item->has_target_code) { + ppc_load_ptr (code, ppc_r0, item->value.target_code); } else { - ppc_load_reg (code, ppc_r0, (sizeof (gpointer) * item->value.vtable_slot), ppc_r11); - ppc_load_reg (code, ppc_r11, PPC_RET_ADDR_OFFSET, ppc_sp); + ppc_ldptr (code, ppc_r0, (sizeof (gpointer) * item->value.vtable_slot), ppc_r11); + ppc_ldptr (code, ppc_r11, PPC_RET_ADDR_OFFSET, ppc_sp); } ppc_mtctr (code, ppc_r0); ppc_bcctr (code, PPC_BR_ALWAYS, 0); @@ -5230,11 +5766,17 @@ mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckI ppc_compare_log (code, 0, MONO_ARCH_IMT_REG, ppc_r0); item->jmp_code = code; ppc_bc (code, PPC_BR_FALSE, PPC_BR_EQ, 0); - ppc_load (code, ppc_r0, item->value.target_code); + if (item->has_target_code) { + ppc_load_ptr (code, ppc_r0, item->value.target_code); + } else { + g_assert (vtable); + ppc_load_ptr (code, ppc_r0, & (vtable->vtable [item->value.vtable_slot])); + ppc_ldptr_indexed (code, ppc_r0, 0, ppc_r0); + } ppc_mtctr (code, ppc_r0); ppc_bcctr (code, PPC_BR_ALWAYS, 0); ppc_patch (item->jmp_code, code); - ppc_load (code, ppc_r0, fail_tramp); + ppc_load_ptr (code, ppc_r0, fail_tramp); ppc_mtctr (code, ppc_r0); ppc_bcctr (code, PPC_BR_ALWAYS, 0); item->jmp_code = NULL; @@ -5246,8 +5788,8 @@ mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckI item->jmp_code = code; ppc_bc (code, PPC_BR_FALSE, PPC_BR_EQ, 0); #endif - ppc_load_reg (code, ppc_r0, (sizeof (gpointer) * item->value.vtable_slot), ppc_r11); - ppc_load_reg (code, ppc_r11, PPC_RET_ADDR_OFFSET, ppc_sp); + ppc_ldptr (code, ppc_r0, (sizeof (gpointer) * item->value.vtable_slot), ppc_r11); + ppc_ldptr (code, ppc_r11, PPC_RET_ADDR_OFFSET, ppc_sp); ppc_mtctr (code, ppc_r0); ppc_bcctr (code, PPC_BR_ALWAYS, 0); #if ENABLE_WRONG_METHOD_CHECK @@ -5282,22 +5824,26 @@ mono_arch_build_imt_thunk (MonoVTable *vtable, MonoDomain *domain, MonoIMTCheckI } MonoMethod* -mono_arch_find_imt_method (gpointer *regs, guint8 *code) +mono_arch_find_imt_method (mgreg_t *regs, guint8 *code) { - return (MonoMethod*) regs [MONO_ARCH_IMT_REG]; + mgreg_t *r = (mgreg_t*)regs; + + return (MonoMethod*)(gsize) r [MONO_ARCH_IMT_REG]; } MonoObject* -mono_arch_find_this_argument (gpointer *regs, MonoMethod *method, MonoGenericSharingContext *gsctx) +mono_arch_find_this_argument (mgreg_t *regs, MonoMethod *method, MonoGenericSharingContext *gsctx) { - return mono_arch_get_this_arg_from_call (gsctx, mono_method_signature (method), (gssize*)regs, NULL); + return mono_arch_get_this_arg_from_call (gsctx, mono_method_signature (method), regs, NULL); } #endif MonoVTable* -mono_arch_find_static_call_vtable (gpointer *regs, guint8 *code) +mono_arch_find_static_call_vtable (mgreg_t *regs, guint8 *code) { - return (MonoVTable*) regs [MONO_ARCH_RGCTX_REG]; + mgreg_t *r = (mgreg_t*)regs; + + return (MonoVTable*)(gsize) r [MONO_ARCH_RGCTX_REG]; } MonoInst* @@ -5326,24 +5872,69 @@ MonoInst* mono_arch_get_domain_intrinsic (MonoCompile* cfg) return ins; } -MonoInst* -mono_arch_get_thread_intrinsic (MonoCompile* cfg) +gpointer +mono_arch_context_get_int_reg (MonoContext *ctx, int reg) { - MonoInst* ins; + if (reg == ppc_r1) + return MONO_CONTEXT_GET_SP (ctx); - setup_tls_access (); - if (monothread_key == -1) - return NULL; - - MONO_INST_NEW (cfg, ins, OP_TLS_GET); - ins->inst_offset = monothread_key; - return ins; + g_assert (reg >= ppc_r13); + + return (gpointer)(gsize)ctx->regs [reg - ppc_r13]; } -gpointer -mono_arch_context_get_int_reg (MonoContext *ctx, int reg) +guint32 +mono_arch_get_patch_offset (guint8 *code) { - g_assert (reg >= ppc_r13); + return 0; +} + +/* + * mono_aot_emit_load_got_addr: + * + * Emit code to load the got address. + * On PPC, the result is placed into r30. + */ +guint8* +mono_arch_emit_load_got_addr (guint8 *start, guint8 *code, MonoCompile *cfg, MonoJumpInfo **ji) +{ + ppc_bl (code, 1); + ppc_mflr (code, ppc_r30); + if (cfg) + mono_add_patch_info (cfg, code - start, MONO_PATCH_INFO_GOT_OFFSET, NULL); + else + *ji = mono_patch_info_list_prepend (*ji, code - start, MONO_PATCH_INFO_GOT_OFFSET, NULL); + /* arch_emit_got_address () patches this */ +#if defined(TARGET_POWERPC64) + ppc_nop (code); + ppc_nop (code); + ppc_nop (code); + ppc_nop (code); +#else + ppc_load32 (code, ppc_r0, 0); + ppc_add (code, ppc_r30, ppc_r30, ppc_r0); +#endif + + return code; +} - return (gpointer)ctx->regs [reg - ppc_r13]; +/* + * mono_ppc_emit_load_aotconst: + * + * Emit code to load the contents of the GOT slot identified by TRAMP_TYPE and + * TARGET from the mscorlib GOT in full-aot code. + * On PPC, the GOT address is assumed to be in r30, and the result is placed into + * r11. + */ +guint8* +mono_arch_emit_load_aotconst (guint8 *start, guint8 *code, MonoJumpInfo **ji, int tramp_type, gconstpointer target) +{ + /* Load the mscorlib got address */ + ppc_ldptr (code, ppc_r11, sizeof (gpointer), ppc_r30); + *ji = mono_patch_info_list_prepend (*ji, code - start, tramp_type, target); + /* arch_emit_got_access () patches this */ + ppc_load32 (code, ppc_r0, 0); + ppc_ldptr_indexed (code, ppc_r11, ppc_r11, ppc_r0); + + return code; }