* Authors:
* Paolo Molaro (lupus@ximian.com)
* Dietmar Maurer (dietmar@ximian.com)
+ * Andreas Faerber <andreas.faerber@web.de>
*
* (C) 2003 Ximian, Inc.
+ * (C) 2007-2008 Andreas Faerber
*/
#include "mini.h"
#include <string.h>
#include <mono/metadata/appdomain.h>
#include <mono/metadata/debug-helpers.h>
+#include <mono/utils/mono-proclib.h>
+#include <mono/utils/mono-mmap.h>
#include "mini-ppc.h"
+#ifdef TARGET_POWERPC64
+#include "cpu-ppc64.h"
+#else
#include "cpu-ppc.h"
+#endif
#include "trace.h"
#include "ir-emit.h"
#ifdef __APPLE__
#include <sys/sysctl.h>
#endif
+#ifdef __linux__
+#include <unistd.h>
+#endif
#define FORCE_INDIR_CALL 1
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
+};
+
+#define BREAKPOINT_SIZE (PPC_LOAD_SEQUENCE_LENGTH + 4)
+
/* This mutex protects architecture specific caches */
#define mono_mini_arch_lock() EnterCriticalSection (&mini_arch_mutex)
#define mono_mini_arch_unlock() LeaveCriticalSection (&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;
+/*
+ * The code generated for sequence points reads from this location, which is
+ * made read-only when single stepping is enabled.
+ */
+static gpointer ss_trigger_page;
+
+/* Enabled breakpoints read from this trigger page */
+static gpointer bp_trigger_page;
+
static int
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_lwz ((code), (dreg), off1, ppc_r2); \
- ppc_lwz ((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_lwz ((code), (dreg), off1, (dreg)); \
+ ppc_ldptr ((code), (dreg), off1, (dreg)); \
} while (0);
/* FIXME: ensure the sc call preserves all but r3 */
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 (); \
} 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); \
{
/* unrolled, use the counter in big */
if (size > sizeof (gpointer) * 5) {
- int shifted = size >> 2;
+ long shifted = size / SIZEOF_VOID_P;
guint8 *copy_loop_start, *copy_loop_jump;
ppc_load (code, ppc_r0, shifted);
ppc_mtctr (code, ppc_r0);
g_assert (sreg == ppc_r11);
- ppc_addi (code, ppc_r12, dreg, (doffset - 4));
- ppc_addi (code, ppc_r11, sreg, (soffset - 4));
+ ppc_addi (code, ppc_r12, dreg, (doffset - sizeof (gpointer)));
+ ppc_addi (code, ppc_r11, sreg, (soffset - sizeof (gpointer)));
copy_loop_start = code;
- ppc_lwzu (code, ppc_r0, ppc_r11, 4);
- ppc_stwu (code, ppc_r0, 4, 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);
- size -= shifted * 4;
+ size -= shifted * sizeof (gpointer);
doffset = soffset = 0;
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_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);
ppc_stw (code, ppc_r0, doffset, dreg);
int
mono_arch_get_argument_info (MonoMethodSignature *csig, int param_count, MonoJitArgumentInfo *arg_info)
{
+#ifdef __mono_ppc64__
+ NOT_IMPLEMENTED;
+ return -1;
+#else
int k, frame_size = 0;
int size, align, pad;
int offset = 8;
arg_info [k].pad = pad;
return frame_size;
+#endif
+}
+
+#ifdef __mono_ppc64__
+static gboolean
+is_load_sequence (guint32 *seq)
+{
+ return ppc_opcode (seq [0]) == 15 && /* lis */
+ ppc_opcode (seq [1]) == 24 && /* ori */
+ ppc_opcode (seq [2]) == 30 && /* sldi */
+ ppc_opcode (seq [3]) == 25 && /* oris */
+ ppc_opcode (seq [4]) == 24; /* ori */
}
+#define ppc_load_get_dest(l) (((l)>>21) & 0x1f)
+#define ppc_load_get_off(l) ((gint16)((l) & 0xffff))
+#endif
+
/* code must point to the blrl */
gboolean
mono_ppc_is_direct_call_sequence (guint32 *code)
{
+#ifdef __mono_ppc64__
+ g_assert(*code == 0x4e800021 || *code == 0x4e800020 || *code == 0x4e800420);
+
+ /* the thunk-less direct call sequence: lis/ori/sldi/oris/ori/mtlr/blrl */
+ if (ppc_opcode (code [-1]) == 31) { /* mtlr */
+ if (ppc_opcode (code [-2]) == 58 && ppc_opcode (code [-3]) == 58) { /* ld/ld */
+ if (!is_load_sequence (&code [-8]))
+ return FALSE;
+ /* one of the loads must be "ld r2,8(rX)" */
+ return (ppc_load_get_dest (code [-2]) == ppc_r2 && ppc_load_get_off (code [-2]) == 8) ||
+ (ppc_load_get_dest (code [-3]) == ppc_r2 && ppc_load_get_off (code [-3]) == 8);
+ }
+ if (ppc_opcode (code [-2]) == 24 && ppc_opcode (code [-3]) == 31) /* mr/nop */
+ return is_load_sequence (&code [-8]);
+ else
+ return is_load_sequence (&code [-6]);
+ }
+ return FALSE;
+#else
g_assert(*code == 0x4e800021);
/* the thunk-less direct call sequence: lis/ori/mtlr/blrl */
return ppc_opcode (code [-1]) == 31 &&
ppc_opcode (code [-2]) == 24 &&
ppc_opcode (code [-3]) == 15;
+#endif
}
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;
if (*code != 0x4e800021)
return NULL;
- /* the thunk-less direct call sequence: lis/ori/mtlr/blrl */
- if ((code [-1] >> 26) == 31 && (code [-2] >> 26) == 24 && (code [-3] >> 26) == 15) {
+ if (mono_ppc_is_direct_call_sequence (code))
return NULL;
- }
+ /* FIXME: more sanity checks here */
/* OK, we're now at the 'blrl' instruction. Now walk backwards
till we get to a 'mtlr rA' */
for (; --code;) {
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 (gulong)));
- /* saved in the MonoLMF structure */
- o = (gpointer)lmf->iregs [reg - 13];
- } else {
- o = regs [reg];
- }
+ o = (gpointer)(gsize)r [reg];
break;
}
}
return o;
}
-gpointer*
-mono_arch_get_vcall_slot_addr (guint8 *code, gpointer *regs)
+#define MAX_ARCH_DELEGATE_PARAMS 7
+
+static gpointer
+get_delegate_invoke_impl (gboolean has_target, guint32 param_count, guint32 *code_len, gboolean aot)
{
- gpointer vt;
- int displacement;
- vt = mono_arch_get_vcall_slot (code, regs, &displacement);
- if (!vt)
- return NULL;
- return (gpointer*)((char*)vt + displacement);
+ guint8 *code, *start;
+
+ if (has_target) {
+ 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_ldptr (code, ppc_r0, G_STRUCT_OFFSET (MonoDelegate, method_ptr), ppc_r3);
+#ifdef PPC_USES_FUNCTION_DESCRIPTOR
+ /* it's a function descriptor */
+ /* 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_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);
+ } else {
+ int size, i;
+
+ 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_ldptr (code, ppc_r0, G_STRUCT_OFFSET (MonoDelegate, method_ptr), ppc_r3);
+#ifdef PPC_USES_FUNCTION_DESCRIPTOR
+ /* it's a function descriptor */
+ ppc_ldptr_indexed (code, ppc_r0, 0, ppc_r0);
+#endif
+ ppc_mtctr (code, ppc_r0);
+ /* slide down the arguments */
+ for (i = 0; i < param_count; ++i) {
+ ppc_mr (code, (ppc_r3 + i), (ppc_r3 + i + 1));
+ }
+ ppc_bcctr (code, PPC_BR_ALWAYS, 0);
+
+ g_assert ((code - start) <= size);
+
+ mono_arch_flush_icache (start, size);
+ }
+
+ if (code_len)
+ *code_len = code - start;
+
+ return start;
}
-#define MAX_ARCH_DELEGATE_PARAMS 7
+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)
if (has_target) {
static guint8* cached = NULL;
- mono_mini_arch_lock ();
- if (cached) {
- mono_mini_arch_unlock ();
+
+ if (cached)
return cached;
- }
-
- start = code = mono_global_codeman_reserve (16);
- /* Replace the this argument with the target */
- ppc_lwz (code, ppc_r0, G_STRUCT_OFFSET (MonoDelegate, method_ptr), ppc_r3);
- ppc_mtctr (code, ppc_r0);
- ppc_lwz (code, ppc_r3, G_STRUCT_OFFSET (MonoDelegate, target), ppc_r3);
- ppc_bcctr (code, PPC_BR_ALWAYS, 0);
+ 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);
- g_assert ((code - start) <= 16);
+ mono_memory_barrier ();
- mono_arch_flush_icache (start, 16);
cached = start;
- mono_mini_arch_unlock ();
- return cached;
} else {
static guint8* cache [MAX_ARCH_DELEGATE_PARAMS + 1] = {NULL};
- int size, i;
+ int i;
if (sig->param_count > MAX_ARCH_DELEGATE_PARAMS)
return NULL;
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 ();
+ if (code)
return code;
- }
-
- size = 12 + sig->param_count * 4;
- start = code = mono_global_codeman_reserve (size);
- ppc_lwz (code, ppc_r0, G_STRUCT_OFFSET (MonoDelegate, method_ptr), ppc_r3);
- ppc_mtctr (code, ppc_r0);
- /* slide down the arguments */
- for (i = 0; i < sig->param_count; ++i) {
- ppc_mr (code, (ppc_r3 + i), (ppc_r3 + i + 1));
+ 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);
}
- ppc_bcctr (code, PPC_BR_ALWAYS, 0);
- g_assert ((code - start) <= size);
+ mono_memory_barrier ();
- mono_arch_flush_icache (start, size);
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;
}
/*
void
mono_arch_init (void)
{
- InitializeCriticalSection (&mini_arch_mutex);
+ InitializeCriticalSection (&mini_arch_mutex);
+
+ ss_trigger_page = mono_valloc (NULL, mono_pagesize (), MONO_MMAP_READ|MONO_MMAP_32BIT);
+ bp_trigger_page = mono_valloc (NULL, mono_pagesize (), MONO_MMAP_READ|MONO_MMAP_32BIT);
+ mono_mprotect (bp_trigger_page, mono_pagesize (), 0);
}
/*
return opts;
}
+#ifdef __mono_ppc64__
+#define CASE_PPC32(c)
+#define CASE_PPC64(c) case c:
+#else
+#define CASE_PPC32(c) case c:
+#define CASE_PPC64(c)
+#endif
+
static gboolean
is_regsize_var (MonoType *t) {
if (t->byref)
switch (t->type) {
case MONO_TYPE_I4:
case MONO_TYPE_U4:
+ CASE_PPC64 (MONO_TYPE_I8)
+ CASE_PPC64 (MONO_TYPE_U8)
case MONO_TYPE_I:
case MONO_TYPE_U:
case MONO_TYPE_PTR:
if (cfg->frame_reg != ppc_sp)
top = 31;
/* ppc_r13 is used by the system on PPC EABI */
- for (i = 14; i < top; ++i)
- regs = g_list_prepend (regs, GUINT_TO_POINTER (i));
+ for (i = 14; i < top; ++i) {
+ /*
+ * Reserve r29 for holding the vtable address for virtual calls in AOT mode,
+ * since the trampolines can clobber r11.
+ */
+ if (!(cfg->compile_aot && i == 29))
+ regs = g_list_prepend (regs, GUINT_TO_POINTER (i));
+ }
return regs;
}
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 };
}
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");
}
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
#define ALWAYS_ON_STACK(s) s
#define FP_ALSO_IN_REG(s) s
#else
+#ifdef __mono_ppc64__
+#define ALWAYS_ON_STACK(s) s
+#define FP_ALSO_IN_REG(s) s
+#else
#define ALWAYS_ON_STACK(s)
#define FP_ALSO_IN_REG(s)
+#endif
#define ALIGN_DOUBLES
#endif
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
+ RegTypeStructByVal if the struct fits
+ in one word, otherwise it's 0*/
} ArgInfo;
typedef struct {
static void inline
add_general (guint *gr, guint *stack_size, ArgInfo *ainfo, gboolean simple)
{
+#ifdef __mono_ppc64__
+ g_assert (simple);
+#endif
+
if (simple) {
if (*gr >= 3 + PPC_NUM_REG_ARGS) {
ainfo->offset = PPC_STACK_PARAM_OFFSET + *stack_size;
ainfo->reg = ppc_sp; /* in the caller */
ainfo->regtype = RegTypeBase;
- *stack_size += 4;
+ *stack_size += sizeof (gpointer);
} else {
- ALWAYS_ON_STACK (*stack_size += 4);
+ ALWAYS_ON_STACK (*stack_size += sizeof (gpointer));
ainfo->reg = *gr;
}
} else {
(*gr) ++;
}
-#if __APPLE__
+#if defined(__APPLE__) || defined(__mono_ppc64__)
static gboolean
has_only_a_r48_field (MonoClass *klass)
{
{
guint i, fr, gr;
int n = sig->hasthis + sig->param_count;
- guint32 simpletype;
+ MonoType *simpletype;
guint32 stack_size = 0;
CallInfo *cinfo = g_malloc0 (sizeof (CallInfo) + sizeof (ArgInfo) * n);
n++;
continue;
}
- simpletype = mini_type_get_underlying_type (NULL, sig->params [i])->type;
- switch (simpletype) {
+ simpletype = mini_type_get_underlying_type (NULL, sig->params [i]);
+ switch (simpletype->type) {
case MONO_TYPE_BOOLEAN:
case MONO_TYPE_I1:
case MONO_TYPE_U1:
n++;
break;
case MONO_TYPE_GENERICINST:
- if (!mono_type_generic_inst_is_valuetype (sig->params [i])) {
+ if (!mono_type_generic_inst_is_valuetype (simpletype)) {
cinfo->args [n].size = sizeof (gpointer);
add_general (&gr, &stack_size, cinfo->args + n, TRUE);
n++;
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->type == 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 __APPLE__
+
+#if defined(__APPLE__) || defined(__mono_ppc64__)
if ((size == 4 || size == 8) && has_only_a_r48_field (klass)) {
cinfo->args [n].size = size;
#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 = rest >= nwords? nwords: rest;
- cinfo->args [n].regtype = RegTypeStructByVal;
- if (gr > PPC_LAST_ARG_REG || (size >= 3 && size % 4 != 0)) {
- 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;
- }
- 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;
+ 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 */
+ if (size >= 3 && size % 4 != 0)
+ n_in_regs = 0;
#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 (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);
case MONO_TYPE_U8:
case MONO_TYPE_I8:
cinfo->args [n].size = 8;
- add_general (&gr, &stack_size, cinfo->args + n, FALSE);
+ add_general (&gr, &stack_size, cinfo->args + n, SIZEOF_REGISTER == 8);
n++;
break;
case MONO_TYPE_R4:
cinfo->args [n].reg = fr;
fr ++;
FP_ALSO_IN_REG (gr ++);
- ALWAYS_ON_STACK (stack_size += 4);
+ ALWAYS_ON_STACK (stack_size += SIZEOF_REGISTER);
} else {
- cinfo->args [n].offset = PPC_STACK_PARAM_OFFSET + stack_size;
+ 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 += 4;
+ stack_size += SIZEOF_REGISTER;
}
n++;
break;
cinfo->args [n].regtype = RegTypeFP;
cinfo->args [n].reg = fr;
fr ++;
- FP_ALSO_IN_REG (gr += 2);
+ 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;
}
{
- simpletype = mini_type_get_underlying_type (NULL, sig->ret)->type;
- switch (simpletype) {
+ simpletype = mini_type_get_underlying_type (NULL, sig->ret);
+ switch (simpletype->type) {
case MONO_TYPE_BOOLEAN:
case MONO_TYPE_I1:
case MONO_TYPE_U1:
cinfo->ret.regtype = RegTypeFP;
break;
case MONO_TYPE_GENERICINST:
- if (!mono_type_generic_inst_is_valuetype (sig->ret)) {
+ if (!mono_type_generic_inst_is_valuetype (simpletype)) {
cinfo->ret.reg = ppc_r3;
break;
}
if (m->method->wrapper_type == MONO_WRAPPER_MANAGED_TO_NATIVE)
m->param_area = MAX (m->param_area, sizeof (gpointer)*8);
- header = mono_method_get_header (m->method);
+ header = m->header;
/*
* We use the frame register also for any method that has
/* this is a global constant */
mono_exc_esp_offset = offset;
#endif
- if (sig->call_convention == MONO_CALL_VARARG) {
- m->sig_cookie = PPC_STACK_PARAM_OFFSET;
- }
if (MONO_TYPE_ISSTRUCT (sig->ret)) {
offset += sizeof(gpointer) - 1;
}
offset += sizeof(gpointer);
- if (sig->call_convention == MONO_CALL_VARARG)
- m->sig_cookie += sizeof (gpointer);
}
offsets = mono_allocate_stack_slots_full (m, FALSE, &locals_stack_size, &locals_stack_align);
offset &= ~(sizeof (gpointer) - 1);
inst->inst_offset = offset;
offset += sizeof (gpointer);
- if (sig->call_convention == MONO_CALL_VARARG)
- m->sig_cookie += sizeof (gpointer);
}
curinst++;
}
} else {
size = mono_type_size (sig->params [i], &align);
}
+ if (MONO_TYPE_ISSTRUCT (sig->params [i]) && size < sizeof (gpointer))
+ size = align = sizeof (gpointer);
offset += align - 1;
offset &= ~(align - 1);
inst->inst_offset = offset;
offset += size;
- if ((sig->call_convention == MONO_CALL_VARARG) && (i < sig->sentinelpos))
- m->sig_cookie += size;
}
curinst++;
}
{
int sig_reg = mono_alloc_ireg (cfg);
- MONO_EMIT_NEW_ICONST (cfg, sig_reg, (guint32)call->signature);
+ /* 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);
}
in = call->args [i];
if (ainfo->regtype == RegTypeGeneral) {
+#ifndef __mono_ppc64__
if (!t->byref && ((t->type == MONO_TYPE_I8) || (t->type == MONO_TYPE_U8))) {
MONO_INST_NEW (cfg, ins, OP_MOVE);
ins->dreg = mono_alloc_ireg (cfg);
ins->sreg1 = in->dreg + 2;
MONO_ADD_INS (cfg->cbb, ins);
mono_call_inst_add_outarg_reg (cfg, call, ins->dreg, ainfo->reg, FALSE);
- } else {
+ } else
+#endif
+ {
MONO_INST_NEW (cfg, ins, OP_MOVE);
ins->dreg = mono_alloc_ireg (cfg);
ins->sreg1 = in->dreg;
}
call->stack_usage = cinfo->stack_usage;
- cfg->param_area = MAX (cfg->param_area, cinfo->stack_usage);
+ cfg->param_area = MAX (PPC_MINIMAL_PARAM_AREA_SIZE, MAX (cfg->param_area, cinfo->stack_usage));
cfg->flags |= MONO_CFG_HAS_CALLS;
g_free (cinfo);
int i, soffset, dreg;
if (ainfo->regtype == RegTypeStructByVal) {
+#ifdef __APPLE__
guint32 size = 0;
+#endif
soffset = 0;
#ifdef __APPLE__
/*
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);
+ antipadding = sizeof (gpointer) - ainfo->bytes;
+ }
dreg = mono_alloc_ireg (cfg);
MONO_EMIT_NEW_LOAD_MEMBASE (cfg, dreg, src->dreg, soffset);
+ if (antipadding)
+ MONO_EMIT_NEW_BIALU_IMM (cfg, OP_SHR_UN_IMM, dreg, dreg, antipadding * 8);
mono_call_inst_add_outarg_reg (cfg, call, dreg, ainfo->reg + i, FALSE);
soffset += sizeof (gpointer);
}
mono_method_signature (method)->ret);
if (!ret->byref) {
+#ifndef __mono_ppc64__
if (ret->type == MONO_TYPE_I8 || ret->type == MONO_TYPE_U8) {
MonoInst *ins;
MONO_ADD_INS (cfg->cbb, ins);
return;
}
+#endif
if (ret->type == MONO_TYPE_R8 || ret->type == MONO_TYPE_R4) {
MONO_EMIT_NEW_UNALU (cfg, OP_FMOVE, cfg->ret->dreg, val->dreg);
return;
{
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 (code, ppc_r0, func);
+ ppc_load_func (code, ppc_r0, func);
ppc_mtlr (code, ppc_r0);
ppc_blrl (code);
return code;
};
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;
else
save_mode = SAVE_NONE;
break;
+#ifndef __mono_ppc64__
case MONO_TYPE_I8:
case MONO_TYPE_U8:
save_mode = SAVE_TWO;
break;
+#endif
case MONO_TYPE_R4:
case MONO_TYPE_R8:
save_mode = SAVE_FP;
}
break;
case SAVE_ONE:
- ppc_stw (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);
}
if (enable_arguments) {
/* FIXME: what reg? */
ppc_fmr (code, ppc_f3, ppc_f1);
+ /* FIXME: use 8 byte load on PPC64 */
ppc_lwz (code, ppc_r4, save_offset, cfg->frame_reg);
ppc_lwz (code, ppc_r5, save_offset + 4, cfg->frame_reg);
}
break;
}
- ppc_load (code, ppc_r3, cfg->method);
- ppc_load (code, ppc_r0, func);
+ ppc_load_ptr (code, ppc_r3, cfg->method);
+ ppc_load_func (code, ppc_r0, func);
ppc_mtlr (code, ppc_r0);
ppc_blrl (code);
ppc_lwz (code, ppc_r4, save_offset + 4, cfg->frame_reg);
break;
case SAVE_ONE:
- ppc_lwz (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);
} 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)])
{
}
+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):
+ return OP_LOAD_MEMINDEX;
+ case MONO_PPC_32_64_CASE (OP_STOREI4_MEMBASE_REG, OP_STOREI8_MEMBASE_REG):
+ return OP_STORE_MEMBASE_REG;
+ case MONO_PPC_32_64_CASE (OP_STOREI4_MEMBASE_IMM, OP_STOREI8_MEMBASE_IMM):
+ 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):
+ return OP_SHR_UN_IMM;
+ default:
+ return opcode;
+ }
+}
+
void
mono_arch_peephole_pass_2 (MonoCompile *cfg, MonoBasicBlock *bb)
{
MonoInst *ins, *n, *last_ins = NULL;
MONO_BB_FOR_EACH_INS_SAFE (bb, n, ins) {
- switch (ins->opcode) {
+ switch (normalize_opcode (ins->opcode)) {
case OP_MUL_IMM:
/* remove unnecessary multiplication with 1 */
if (ins->inst_imm == 1) {
}
break;
case OP_LOAD_MEMBASE:
- case OP_LOADI4_MEMBASE:
/*
* OP_STORE_MEMBASE_REG reg, offset(basereg)
* OP_LOAD_MEMBASE offset(basereg), reg
*/
- if (last_ins && (last_ins->opcode == OP_STOREI4_MEMBASE_REG
- || last_ins->opcode == OP_STORE_MEMBASE_REG) &&
+ if (last_ins && normalize_opcode (last_ins->opcode) == OP_STORE_MEMBASE_REG &&
ins->inst_basereg == last_ins->inst_destbasereg &&
ins->inst_offset == last_ins->inst_offset) {
if (ins->dreg == last_ins->sreg1) {
* OP_LOAD_MEMBASE offset(basereg), reg1
* OP_MOVE reg1, reg2
*/
- } if (last_ins && (last_ins->opcode == OP_LOADI4_MEMBASE
- || last_ins->opcode == OP_LOAD_MEMBASE) &&
+ } else if (last_ins && normalize_opcode (last_ins->opcode) == OP_LOAD_MEMBASE &&
ins->inst_basereg != last_ins->dreg &&
ins->inst_basereg == last_ins->inst_basereg &&
ins->inst_offset == last_ins->inst_offset) {
* OP_STORE_MEMBASE_IMM imm, offset(basereg)
* OP_ICONST reg, imm
*/
- } else if (last_ins && (last_ins->opcode == OP_STOREI4_MEMBASE_IMM
- || last_ins->opcode == OP_STORE_MEMBASE_IMM) &&
+ } else if (last_ins && normalize_opcode (last_ins->opcode) == OP_STORE_MEMBASE_IMM &&
ins->inst_basereg == last_ins->inst_destbasereg &&
ins->inst_offset == last_ins->inst_offset) {
//static int c = 0; printf ("MATCHX %s %d\n", cfg->method->name,c++);
ins->sreg1 = last_ins->sreg1;
}
break;
- case OP_MOVE:
- ins->opcode = OP_MOVE;
- /*
- * OP_MOVE reg, reg
+#ifdef __mono_ppc64__
+ case OP_LOADU4_MEMBASE:
+ case OP_LOADI4_MEMBASE:
+ if (last_ins && (last_ins->opcode == OP_STOREI4_MEMBASE_REG) &&
+ ins->inst_basereg == last_ins->inst_destbasereg &&
+ ins->inst_offset == last_ins->inst_offset) {
+ ins->opcode = (ins->opcode == OP_LOADI4_MEMBASE) ? OP_ICONV_TO_I4 : OP_ICONV_TO_U4;
+ ins->sreg1 = last_ins->sreg1;
+ }
+ break;
+#endif
+ case OP_MOVE:
+ ins->opcode = OP_MOVE;
+ /*
+ * OP_MOVE reg, reg
*/
if (ins->dreg == ins->sreg1) {
MONO_DELETE_INS (bb, ins);
ins->opcode = OP_NOP;
break;
}
+#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
case OP_CKFINITE: {
int msw_reg = mono_alloc_ireg (cfg);
int basereg = ppc_sp;
ins->opcode = OP_NOP;
break;
}
+#ifdef __mono_ppc64__
+ case OP_IADD_OVF:
+ case OP_IADD_OVF_UN:
+ case OP_ISUB_OVF: {
+ int shifted1_reg = mono_alloc_ireg (cfg);
+ int shifted2_reg = mono_alloc_ireg (cfg);
+ int result_shifted_reg = mono_alloc_ireg (cfg);
+
+ MONO_EMIT_NEW_BIALU_IMM (cfg, OP_SHL_IMM, shifted1_reg, ins->sreg1, 32);
+ MONO_EMIT_NEW_BIALU_IMM (cfg, OP_SHL_IMM, shifted2_reg, ins->sreg2, 32);
+ MONO_EMIT_NEW_BIALU (cfg, ins->opcode, result_shifted_reg, shifted1_reg, shifted2_reg);
+ if (ins->opcode == OP_IADD_OVF_UN)
+ MONO_EMIT_NEW_BIALU_IMM (cfg, OP_SHR_UN_IMM, ins->dreg, result_shifted_reg, 32);
+ else
+ MONO_EMIT_NEW_BIALU_IMM (cfg, OP_SHR_IMM, ins->dreg, result_shifted_reg, 32);
+ ins->opcode = OP_NOP;
+ }
+#endif
+ }
+}
+
+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;
}
}
return OP_COMPARE;
case OP_ICOMPARE_IMM:
return OP_ICOMPARE;
+ case OP_LCOMPARE_IMM:
+ return OP_LCOMPARE;
case OP_ADDCC_IMM:
return OP_IADDCC;
case OP_ADC_IMM:
return OP_LOADI4_MEMINDEX;
case OP_LOADU4_MEMBASE:
return OP_LOADU4_MEMINDEX;
+ case OP_LOADI8_MEMBASE:
+ return OP_LOADI8_MEMINDEX;
case OP_LOADU1_MEMBASE:
return OP_LOADU1_MEMINDEX;
case OP_LOADI2_MEMBASE:
return OP_STOREI2_MEMINDEX;
case OP_STOREI4_MEMBASE_REG:
return OP_STOREI4_MEMINDEX;
+ case OP_STOREI8_MEMBASE_REG:
+ return OP_STOREI8_MEMINDEX;
case OP_STORE_MEMBASE_REG:
return OP_STORE_MEMINDEX;
case OP_STORER4_MEMBASE_REG:
return OP_STOREI2_MEMBASE_REG;
case OP_STOREI4_MEMBASE_IMM:
return OP_STOREI4_MEMBASE_REG;
+ case OP_STOREI8_MEMBASE_IMM:
+ return OP_STOREI8_MEMBASE_REG;
}
return mono_op_imm_to_op (op);
}
#define compare_opcode_is_unsigned(opcode) \
(((opcode) >= CEE_BNE_UN && (opcode) <= CEE_BLT_UN) || \
- (((opcode) >= OP_IBNE_UN && (opcode) <= OP_IBLT_UN) || \
+ ((opcode) >= OP_IBNE_UN && (opcode) <= OP_IBLT_UN) || \
+ ((opcode) >= OP_LBNE_UN && (opcode) <= OP_LBLT_UN) || \
((opcode) >= OP_COND_EXC_NE_UN && (opcode) <= OP_COND_EXC_LT_UN) || \
((opcode) >= OP_COND_EXC_INE_UN && (opcode) <= OP_COND_EXC_ILT_UN) || \
- ((opcode) == OP_CLT_UN || (opcode) == OP_CGT_UN || (opcode) == OP_ICLT_UN || (opcode) == OP_ICGT_UN)))
+ ((opcode) == OP_CLT_UN || (opcode) == OP_CGT_UN || \
+ (opcode) == OP_ICLT_UN || (opcode) == OP_ICGT_UN || \
+ (opcode) == OP_LCLT_UN || (opcode) == OP_LCGT_UN))
+
/*
* Remove from the instruction list the instructions that can't be
* represented with very simple instructions with no register
/* handle rem separately */
goto loop_start;
case OP_IREM:
- case OP_IREM_UN: {
+ case OP_IREM_UN:
+ CASE_PPC64 (OP_LREM)
+ CASE_PPC64 (OP_LREM_UN) {
MonoInst *mul;
/* we change a rem dest, src1, src2 to
* div temp1, src1, src2
* mul temp2, temp1, src2
* sub dest, src1, temp2
*/
- NEW_INS (cfg, mul, OP_IMUL);
- NEW_INS (cfg, temp, ins->opcode == OP_IREM? OP_IDIV: OP_IDIV_UN);
+ if (ins->opcode == OP_IREM || ins->opcode == OP_IREM_UN) {
+ NEW_INS (cfg, mul, OP_IMUL);
+ NEW_INS (cfg, temp, ins->opcode == OP_IREM? OP_IDIV: OP_IDIV_UN);
+ ins->opcode = OP_ISUB;
+ } else {
+ NEW_INS (cfg, mul, OP_LMUL);
+ NEW_INS (cfg, temp, ins->opcode == OP_LREM? OP_LDIV: OP_LDIV_UN);
+ ins->opcode = OP_LSUB;
+ }
temp->sreg1 = ins->sreg1;
temp->sreg2 = ins->sreg2;
temp->dreg = mono_alloc_ireg (cfg);
mul->sreg1 = temp->dreg;
mul->sreg2 = ins->sreg2;
mul->dreg = mono_alloc_ireg (cfg);
- ins->opcode = OP_ISUB;
ins->sreg2 = mul->dreg;
break;
}
case OP_IADD_IMM:
+ CASE_PPC64 (OP_LADD_IMM)
case OP_ADD_IMM:
case OP_ADDCC_IMM:
if (!ppc_is_imm16 (ins->inst_imm)) {
}
break;
case OP_ISUB_IMM:
+ CASE_PPC64 (OP_LSUB_IMM)
case OP_SUB_IMM:
if (!ppc_is_imm16 (-ins->inst_imm)) {
NEW_INS (cfg, temp, OP_ICONST);
case OP_IAND_IMM:
case OP_IOR_IMM:
case OP_IXOR_IMM:
+ case OP_LAND_IMM:
+ case OP_LOR_IMM:
+ case OP_LXOR_IMM:
case OP_AND_IMM:
case OP_OR_IMM:
- case OP_XOR_IMM:
- if ((ins->inst_imm & 0xffff0000) && (ins->inst_imm & 0xffff)) {
+ case OP_XOR_IMM: {
+ gboolean is_imm = ((ins->inst_imm & 0xffff0000) && (ins->inst_imm & 0xffff));
+#ifdef __mono_ppc64__
+ if (ins->inst_imm & 0xffffffff00000000ULL)
+ is_imm = TRUE;
+#endif
+ if (is_imm) {
NEW_INS (cfg, temp, OP_ICONST);
temp->inst_c0 = ins->inst_imm;
temp->dreg = mono_alloc_ireg (cfg);
ins->opcode = map_to_reg_reg_op (ins->opcode);
}
break;
+ }
case OP_ISBB_IMM:
case OP_IADC_IMM:
case OP_SBB_IMM:
break;
case OP_COMPARE_IMM:
case OP_ICOMPARE_IMM:
+ CASE_PPC64 (OP_LCOMPARE_IMM)
next = ins->next;
/* Branch opts can eliminate the branch */
if (!next || (!(MONO_IS_COND_BRANCH_OP (next) || MONO_IS_COND_EXC (next) || MONO_IS_SETCC (next)))) {
break;
case OP_LOAD_MEMBASE:
case OP_LOADI4_MEMBASE:
+ CASE_PPC64 (OP_LOADI8_MEMBASE)
case OP_LOADU4_MEMBASE:
case OP_LOADI2_MEMBASE:
case OP_LOADU2_MEMBASE:
case OP_LOADR4_MEMBASE:
case OP_LOADR8_MEMBASE:
case OP_STORE_MEMBASE_REG:
+ CASE_PPC64 (OP_STOREI8_MEMBASE_REG)
case OP_STOREI4_MEMBASE_REG:
case OP_STOREI2_MEMBASE_REG:
case OP_STOREI1_MEMBASE_REG:
case OP_STOREI1_MEMBASE_IMM:
case OP_STOREI2_MEMBASE_IMM:
case OP_STOREI4_MEMBASE_IMM:
+ CASE_PPC64 (OP_STOREI8_MEMBASE_IMM)
NEW_INS (cfg, temp, OP_ICONST);
temp->inst_c0 = ins->inst_imm;
temp->dreg = mono_alloc_ireg (cfg);
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 = (guint32)ins->inst_p0;
+ temp->inst_c0 = (gulong)ins->inst_p0;
temp->dreg = mono_alloc_ireg (cfg);
ins->inst_basereg = temp->dreg;
ins->inst_offset = 0;
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;
+ 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 */
- ppc_fctiwz (code, ppc_f0, sreg);
- if (ppc_is_imm16 (offset + 4)) {
+#ifdef __mono_ppc64__
+ if (size == 8) {
+ ppc_fctidz (code, ppc_f0, sreg);
+ sub_offset = 0;
+ } else
+#endif
+ {
+ ppc_fctiwz (code, ppc_f0, sreg);
+ sub_offset = 4;
+ }
+ if (ppc_is_imm16 (offset + sub_offset)) {
ppc_stfd (code, ppc_f0, offset, cfg->frame_reg);
- ppc_lwz (code, dreg, offset + 4, cfg->frame_reg);
+ if (size == 8)
+ ppc_ldr (code, dreg, offset + sub_offset, cfg->frame_reg);
+ else
+ ppc_lwz (code, dreg, offset + sub_offset, cfg->frame_reg);
} else {
ppc_load (code, dreg, offset);
ppc_add (code, dreg, dreg, cfg->frame_reg);
ppc_stfd (code, ppc_f0, 0, dreg);
- ppc_lwz (code, dreg, 4, dreg);
+ if (size == 8)
+ ppc_ldr (code, dreg, sub_offset, dreg);
+ else
+ ppc_lwz (code, dreg, sub_offset, dreg);
}
if (!is_signed) {
if (size == 1)
ppc_andid (code, dreg, dreg, 0xff);
else if (size == 2)
ppc_andid (code, dreg, dreg, 0xffff);
+#ifdef __mono_ppc64__
+ else if (size == 4)
+ ppc_clrldi (code, dreg, dreg, 32);
+#endif
} else {
if (size == 1)
ppc_extsb (code, dreg, dreg);
else if (size == 2)
ppc_extsh (code, dreg, dreg);
+#ifdef __mono_ppc64__
+ else if (size == 4)
+ ppc_extsw (code, dreg, dreg);
+#endif
}
return code;
}
int found;
} PatchData;
-#define is_call_imm(diff) ((gint)(diff) >= -33554432 && (gint)(diff) <= 33554431)
+#define is_call_imm(diff) ((glong)(diff) >= -33554432 && (glong)(diff) <= 33554431)
static int
search_thunk_slot (void *data, int csize, int bsize, void *user_data) {
+#ifdef __mono_ppc64__
+ g_assert_not_reached ();
+#else
PatchData *pdata = (PatchData*)user_data;
guchar *code = data;
guint32 *thunks = data;
return 0;
templ = (guchar*)load;
- ppc_lis (templ, ppc_r0, (guint32)(pdata->target) >> 16);
- ppc_ori (templ, ppc_r0, ppc_r0, (guint32)(pdata->target) & 0xffff);
+ ppc_load_sequence (templ, ppc_r0, pdata->target);
//g_print ("thunk nentries: %d\n", ((char*)endthunks - (char*)thunks)/16);
if ((pdata->found == 2) || (pdata->code >= code && pdata->code <= code + csize)) {
//g_print ("looking for target: %p at %p (%08x-%08x)\n", pdata->target, thunks, thunks [0], thunks [1]);
if ((thunks [0] == load [0]) && (thunks [1] == load [1])) {
ppc_patch (pdata->code, (guchar*)thunks);
- mono_arch_flush_icache (pdata->code, 4);
pdata->found = 1;
/*{
static int num_thunks = 0;
} else if ((thunks [0] == 0) && (thunks [1] == 0)) {
/* found a free slot instead: emit thunk */
code = (guchar*)thunks;
- ppc_lis (code, ppc_r0, (guint32)(pdata->target) >> 16);
- ppc_ori (code, ppc_r0, ppc_r0, (guint32)(pdata->target) & 0xffff);
+ ppc_lis (code, ppc_r0, (gulong)(pdata->target) >> 16);
+ ppc_ori (code, ppc_r0, ppc_r0, (gulong)(pdata->target) & 0xffff);
ppc_mtctr (code, ppc_r0);
ppc_bcctr (code, PPC_BR_ALWAYS, 0);
mono_arch_flush_icache ((guchar*)thunks, 16);
ppc_patch (pdata->code, (guchar*)thunks);
- mono_arch_flush_icache (pdata->code, 4);
pdata->found = 1;
/*{
static int num_thunks = 0;
}
//g_print ("failed thunk lookup for %p from %p at %p (%d entries)\n", pdata->target, pdata->code, data, count);
}
+#endif
return 0;
}
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);
g_assert (pdata.found == 1);
}
+static void
+patch_ins (guint8 *code, guint32 ins)
+{
+ *(guint32*)code = GUINT32_TO_BE (ins);
+ mono_arch_flush_icache (code, 4);
+}
+
void
-ppc_patch (guchar *code, const guchar *target)
+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;
if (prim == 18) {
// prefer relative branches, they are more position independent (e.g. for AOT compilation).
gint diff = target - code;
+ g_assert (!is_fd);
if (diff >= 0){
if (diff <= 33554431){
ins = (18 << 26) | (diff) | (ins & 1);
- *(guint32*)code = ins;
+ patch_ins (code, ins);
return;
}
} else {
/* diff between 0 and -33554432 */
if (diff >= -33554432){
ins = (18 << 26) | (diff & ~0xfc000000) | (ins & 1);
- *(guint32*)code = ins;
+ patch_ins (code, ins);
return;
}
}
if ((glong)target >= 0){
if ((glong)target <= 33554431){
- ins = (18 << 26) | ((guint32) target) | (ins & 1) | 2;
- *(guint32*)code = ins;
+ ins = (18 << 26) | ((gulong) target) | (ins & 1) | 2;
+ patch_ins (code, ins);
return;
}
} else {
if ((glong)target >= -33554432){
- ins = (18 << 26) | (((guint32)target) & ~0xfc000000) | (ins & 1) | 2;
- *(guint32*)code = ins;
+ ins = (18 << 26) | (((gulong)target) & ~0xfc000000) | (ins & 1) | 2;
+ patch_ins (code, ins);
return;
}
}
if (prim == 16) {
+ g_assert (!is_fd);
// absolute address
if (ins & 2) {
- guint32 li = (guint32)target;
+ guint32 li = (gulong)target;
ins = (ins & 0xffff0000) | (ins & 3);
ovf = li & 0xffff0000;
if (ovf != 0 && ovf != 0xffff0000)
diff &= 0xffff;
ins |= diff;
}
- *(guint32*)code = ins;
+ patch_ins (code, ins);
return;
}
if (prim == 15 || ins == 0x4e800021 || ins == 0x4e800020 || ins == 0x4e800420) {
+#ifdef __mono_ppc64__
+ guint32 *seq = (guint32*)code;
+ guint32 *branch_ins;
+
+ /* the trampoline code will try to patch the blrl, blr, bcctr */
+ if (ins == 0x4e800021 || ins == 0x4e800020 || ins == 0x4e800420) {
+ branch_ins = seq;
+ if (ppc_opcode (seq [-3]) == 58 || ppc_opcode (seq [-3]) == 31) /* ld || mr */
+ code -= 32;
+ else
+ code -= 24;
+ } else {
+ if (ppc_opcode (seq [5]) == 58 || ppc_opcode (seq [5]) == 31) /* ld || mr */
+ branch_ins = seq + 8;
+ else
+ branch_ins = seq + 6;
+ }
+
+ seq = (guint32*)code;
+ /* this is the lis/ori/sldi/oris/ori/(ld/ld|mr/nop)/mtlr/blrl sequence */
+ g_assert (mono_ppc_is_direct_call_sequence (branch_ins));
+
+ if (ppc_opcode (seq [5]) == 58) { /* ld */
+ g_assert (ppc_opcode (seq [6]) == 58); /* ld */
+
+ if (!is_fd) {
+ guint8 *buf = (guint8*)&seq [5];
+ ppc_mr (buf, ppc_r0, ppc_r11);
+ ppc_nop (buf);
+ }
+ } else {
+ if (is_fd)
+ target = mono_get_addr_from_ftnptr ((gpointer)target);
+ }
+
+ /* FIXME: make this thread safe */
+ /* FIXME: we're assuming we're using r11 here */
+ ppc_load_ptr_sequence (code, ppc_r11, target);
+ mono_arch_flush_icache ((guint8*)seq, 28);
+#else
guint32 *seq;
/* the trampoline code will try to patch the blrl, blr, bcctr */
if (ins == 0x4e800021 || ins == 0x4e800020 || ins == 0x4e800420) {
ppc_lis (code, ppc_r0, (guint32)(target) >> 16);
ppc_ori (code, ppc_r0, ppc_r0, (guint32)(target) & 0xffff);
mono_arch_flush_icache (code - 8, 8);
+#endif
} else {
g_assert_not_reached ();
}
// g_print ("patched with 0x%08x\n", ins);
}
+void
+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)
{
guint32 i, pos;
int struct_index = 0;
- /* FIXME: Generate intermediate code instead */
-
sig = mono_method_signature (method);
/* This is the opposite of the code in emit_prolog */
ArgInfo *ainfo = &cinfo->ret;
inst = cfg->vret_addr;
g_assert (ppc_is_imm16 (inst->inst_offset));
- ppc_lwz (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;
case 2:
ppc_lhz (code, ainfo->reg, inst->inst_offset, inst->inst_basereg);
break;
- default:
+#ifdef __mono_ppc64__
+ case 4:
ppc_lwz (code, ainfo->reg, inst->inst_offset, inst->inst_basereg);
break;
+#endif
+ default:
+ ppc_ldptr (code, ainfo->reg, inst->inst_offset, inst->inst_basereg);
+ break;
}
break;
MonoType *type = mini_type_get_underlying_type (cfg->generic_sharing_context,
&inst->klass->byval_arg);
- if (!MONO_TYPE_IS_REFERENCE (type) && type->type != MONO_TYPE_I4)
+#ifndef __mono_ppc64__
+ if (type->type == MONO_TYPE_I8)
+ NOT_IMPLEMENTED;
+#endif
+
+ if (MONO_TYPE_IS_REFERENCE (type) || type->type == MONO_TYPE_I8) {
+ 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);
+ } else {
NOT_IMPLEMENTED;
+ }
- ppc_lwz (code, ppc_r0, inst->inst_offset, inst->inst_basereg);
- ppc_stw (code, ppc_r0, ainfo->offset, ainfo->reg);
break;
}
case RegTypeStructByVal: {
+#ifdef __APPLE__
guint32 size = 0;
+#endif
int j;
/* FIXME: */
NOT_IMPLEMENTED;
} else
#endif
- for (j = 0; j < ainfo->size; ++j) {
- ppc_lwz (code, ainfo->reg + j,
- inst->inst_offset + j * sizeof (gpointer), inst->inst_basereg);
+ 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 */
+ if (ainfo->bytes)
+ NOT_IMPLEMENTED;
}
break;
}
g_assert (ppc_is_imm16 (addr->inst_offset));
g_assert (!ainfo->offset);
- ppc_lwz (code, ainfo->reg, addr->inst_offset, addr->inst_basereg);
+ ppc_ldptr (code, ainfo->reg, addr->inst_offset, addr->inst_basereg);
struct_index++;
break;
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;
if (!size)
return code;
- ppc_lwz (code, ppc_r0, 0, ppc_sp);
+ ppc_ldptr (code, ppc_r0, 0, ppc_sp);
if (ppc_is_imm16 (-size)) {
- ppc_stwu (code, ppc_r0, -size, ppc_sp);
+ ppc_stptr_update (code, ppc_r0, -size, ppc_sp);
} else {
ppc_load (code, ppc_r11, -size);
- ppc_stwux (code, ppc_r0, ppc_sp, ppc_r11);
+ ppc_stptr_update_indexed (code, ppc_r0, ppc_sp, ppc_r11);
}
return 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;
if (!size)
return code;
- ppc_lwz (code, ppc_r0, 0, ppc_sp);
+ ppc_ldptr (code, ppc_r0, 0, ppc_sp);
if (ppc_is_imm16 (size)) {
- ppc_stwu (code, ppc_r0, size, ppc_sp);
+ ppc_stptr_update (code, ppc_r0, size, ppc_sp);
} else {
ppc_load (code, ppc_r11, size);
- ppc_stwux (code, ppc_r0, ppc_sp, ppc_r11);
+ ppc_stptr_update_indexed (code, ppc_r0, ppc_sp, ppc_r11);
}
return code;
}
+#define MASK_SHIFT_IMM(i) ((i) & MONO_PPC_32_64_CASE (0x1f, 0x3f))
+
void
mono_arch_output_basic_block (MonoCompile *cfg, MonoBasicBlock *bb)
{
MonoInst *last_ins = NULL;
guint last_offset = 0;
int max_len, cpos;
+ int L;
/* we don't align basic blocks of loops on ppc */
// g_print ("cil code\n");
mono_debug_record_line_number (cfg, ins, offset);
- switch (ins->opcode) {
+ switch (normalize_opcode (ins->opcode)) {
case OP_RELAXED_NOP:
case OP_NOP:
case OP_DUMMY_USE:
case OP_NOT_REACHED:
case OP_NOT_NULL:
break;
+ case OP_SEQ_POINT: {
+ int i;
+
+ if (cfg->compile_aot)
+ NOT_IMPLEMENTED;
+
+ /*
+ * Read from the single stepping trigger page. This will cause a
+ * SIGSEGV when single stepping is enabled.
+ * We do this _before_ the breakpoint, so single stepping after
+ * a breakpoint is hit will step to the next IL offset.
+ */
+ if (ins->flags & MONO_INST_SINGLE_STEP_LOC) {
+ ppc_load (code, ppc_r11, (gsize)ss_trigger_page);
+ ppc_ldptr (code, ppc_r11, 0, ppc_r11);
+ }
+
+ mono_add_seq_point (cfg, bb, ins, code - cfg->native_code);
+
+ /*
+ * A placeholder for a possible breakpoint inserted by
+ * mono_arch_set_breakpoint ().
+ */
+ for (i = 0; i < BREAKPOINT_SIZE / 4; ++i)
+ ppc_nop (code);
+ break;
+ }
case OP_TLS_GET:
emit_tls_access (code, ins->dreg, ins->inst_offset);
break;
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:
- case OP_STOREI4_MEMBASE_REG:
if (ppc_is_imm16 (ins->inst_offset)) {
- ppc_stw (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_stwx (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:
- case OP_STOREI4_MEMINDEX:
- ppc_stwx (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_ldptr (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_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);
+ }
}
break;
case OP_LOADI1_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);
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_ldptr_indexed (code, ins->dreg, ins->inst_basereg, ins->sreg2);
+ break;
case OP_LOADI4_MEMINDEX:
+#ifdef __mono_ppc64__
+ ppc_lwax (code, ins->dreg, ins->inst_basereg, ins->sreg2);
+ break;
+#endif
case OP_LOADU4_MEMINDEX:
- ppc_lwzx (code, ins->dreg, ins->sreg2, ins->inst_basereg);
+ 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:
+ CASE_PPC64 (OP_LCONV_TO_I1)
ppc_extsb (code, ins->dreg, ins->sreg1);
break;
case OP_ICONV_TO_I2:
+ CASE_PPC64 (OP_LCONV_TO_I2)
ppc_extsh (code, ins->dreg, ins->sreg1);
break;
case OP_ICONV_TO_U1:
- ppc_rlwinm (code, ins->dreg, ins->sreg1, 0, 24, 31);
+ CASE_PPC64 (OP_LCONV_TO_U1)
+ ppc_clrlwi (code, ins->dreg, ins->sreg1, 24);
break;
case OP_ICONV_TO_U2:
- ppc_rlwinm (code, ins->dreg, ins->sreg1, 0, 16, 31);
+ CASE_PPC64 (OP_LCONV_TO_U2)
+ ppc_clrlwi (code, ins->dreg, ins->sreg1, 16);
break;
case OP_COMPARE:
case OP_ICOMPARE:
+ CASE_PPC64 (OP_LCOMPARE)
+ 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, 0, ins->sreg1, ins->sreg2);
+ ppc_cmpl (code, 0, L, ins->sreg1, ins->sreg2);
else
- ppc_cmp (code, 0, 0, ins->sreg1, ins->sreg2);
+ ppc_cmp (code, 0, L, ins->sreg1, ins->sreg2);
break;
case OP_COMPARE_IMM:
case OP_ICOMPARE_IMM:
+ CASE_PPC64 (OP_LCOMPARE_IMM)
+ 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)) {
- ppc_cmpli (code, 0, 0, ins->sreg1, (ins->inst_imm & 0xffff));
+ ppc_cmpli (code, 0, L, ins->sreg1, (ins->inst_imm & 0xffff));
} else {
g_assert_not_reached ();
}
} else {
if (ppc_is_imm16 (ins->inst_imm)) {
- ppc_cmpi (code, 0, 0, ins->sreg1, (ins->inst_imm & 0xffff));
+ ppc_cmpi (code, 0, L, ins->sreg1, (ins->inst_imm & 0xffff));
} else {
g_assert_not_reached ();
}
break;
case OP_ADDCC:
case OP_IADDCC:
- ppc_addc (code, ins->dreg, ins->sreg1, ins->sreg2);
+ ppc_addco (code, ins->dreg, ins->sreg1, ins->sreg2);
break;
case OP_IADD:
+ CASE_PPC64 (OP_LADD)
ppc_add (code, ins->dreg, ins->sreg1, ins->sreg2);
break;
case OP_ADC:
break;
case OP_ADD_IMM:
case OP_IADD_IMM:
+ CASE_PPC64 (OP_LADD_IMM)
if (ppc_is_imm16 (ins->inst_imm)) {
ppc_addi (code, ins->dreg, ins->sreg1, ins->inst_imm);
} else {
EMIT_COND_SYSTEM_EXCEPTION_FLAGS (PPC_BR_FALSE, PPC_BR_EQ, "OverflowException");
break;
case OP_ISUB_OVF:
+ CASE_PPC64 (OP_LSUB_OVF)
/* check XER [0-3] (SO, OV, CA): we can't use mcrxr
*/
ppc_subfo (code, ins->dreg, ins->sreg2, ins->sreg1);
EMIT_COND_SYSTEM_EXCEPTION_FLAGS (PPC_BR_FALSE, PPC_BR_EQ, "OverflowException");
break;
case OP_ISUB_OVF_UN:
+ CASE_PPC64 (OP_LSUB_OVF_UN)
/* check XER [0-3] (SO, OV, CA): we can't use mcrxr
*/
ppc_subfc (code, ins->dreg, ins->sreg2, ins->sreg1);
break;
case OP_SUBCC:
case OP_ISUBCC:
- ppc_subfc (code, ins->dreg, ins->sreg2, ins->sreg1);
+ ppc_subfco (code, ins->dreg, ins->sreg2, ins->sreg1);
break;
case OP_ISUB:
+ CASE_PPC64 (OP_LSUB)
ppc_subf (code, ins->dreg, ins->sreg2, ins->sreg1);
break;
case OP_SBB:
break;
case OP_SUB_IMM:
case OP_ISUB_IMM:
+ CASE_PPC64 (OP_LSUB_IMM)
// we add the negated value
if (ppc_is_imm16 (-ins->inst_imm))
ppc_addi (code, ins->dreg, ins->sreg1, -ins->inst_imm);
ppc_subfze (code, ins->dreg, ins->sreg1);
break;
case OP_IAND:
+ CASE_PPC64 (OP_LAND)
/* FIXME: the ppc macros as inconsistent here: put dest as the first arg! */
ppc_and (code, ins->sreg1, ins->dreg, ins->sreg2);
break;
case OP_AND_IMM:
case OP_IAND_IMM:
+ CASE_PPC64 (OP_LAND_IMM)
if (!(ins->inst_imm & 0xffff0000)) {
ppc_andid (code, ins->sreg1, ins->dreg, ins->inst_imm);
} else if (!(ins->inst_imm & 0xffff)) {
g_assert_not_reached ();
}
break;
- case OP_IDIV: {
+ case OP_IDIV:
+ CASE_PPC64 (OP_LDIV) {
guint8 *divisor_is_m1;
/* XER format: SO, OV, CA, reserved [21 bits], count [8 bits]
*/
- ppc_cmpi (code, 0, 0, ins->sreg2, -1);
+ ppc_compare_reg_imm (code, 0, ins->sreg2, -1);
divisor_is_m1 = code;
ppc_bc (code, PPC_BR_FALSE | PPC_BR_LIKELY, PPC_BR_EQ, 0);
ppc_lis (code, ppc_r0, 0x8000);
- ppc_cmp (code, 0, 0, ins->sreg1, ppc_r0);
+#ifdef __mono_ppc64__
+ if (ins->opcode == OP_LDIV)
+ ppc_sldi (code, ppc_r0, ppc_r0, 32);
+#endif
+ ppc_compare (code, 0, ins->sreg1, ppc_r0);
EMIT_COND_SYSTEM_EXCEPTION_FLAGS (PPC_BR_TRUE, PPC_BR_EQ, "ArithmeticException");
ppc_patch (divisor_is_m1, code);
/* XER format: SO, OV, CA, reserved [21 bits], count [8 bits]
*/
- ppc_divwod (code, ins->dreg, ins->sreg1, ins->sreg2);
+ if (ins->opcode == OP_IDIV)
+ ppc_divwod (code, ins->dreg, ins->sreg1, ins->sreg2);
+#ifdef __mono_ppc64__
+ else
+ ppc_divdod (code, ins->dreg, ins->sreg1, ins->sreg2);
+#endif
ppc_mfspr (code, ppc_r0, ppc_xer);
ppc_andisd (code, ppc_r0, ppc_r0, (1<<14));
EMIT_COND_SYSTEM_EXCEPTION_FLAGS (PPC_BR_FALSE, PPC_BR_EQ, "DivideByZeroException");
break;
}
case OP_IDIV_UN:
- ppc_divwuod (code, ins->dreg, ins->sreg1, ins->sreg2);
+ CASE_PPC64 (OP_LDIV_UN)
+ if (ins->opcode == OP_IDIV_UN)
+ ppc_divwuod (code, ins->dreg, ins->sreg1, ins->sreg2);
+#ifdef __mono_ppc64__
+ else
+ ppc_divduod (code, ins->dreg, ins->sreg1, ins->sreg2);
+#endif
ppc_mfspr (code, ppc_r0, ppc_xer);
ppc_andisd (code, ppc_r0, ppc_r0, (1<<14));
EMIT_COND_SYSTEM_EXCEPTION_FLAGS (PPC_BR_FALSE, PPC_BR_EQ, "DivideByZeroException");
case OP_REM_IMM:
g_assert_not_reached ();
case OP_IOR:
+ CASE_PPC64 (OP_LOR)
ppc_or (code, ins->dreg, ins->sreg1, ins->sreg2);
break;
case OP_OR_IMM:
case OP_IOR_IMM:
+ CASE_PPC64 (OP_LOR_IMM)
if (!(ins->inst_imm & 0xffff0000)) {
ppc_ori (code, ins->sreg1, ins->dreg, ins->inst_imm);
} else if (!(ins->inst_imm & 0xffff)) {
}
break;
case OP_IXOR:
+ CASE_PPC64 (OP_LXOR)
ppc_xor (code, ins->dreg, ins->sreg1, ins->sreg2);
break;
case OP_IXOR_IMM:
case OP_XOR_IMM:
+ CASE_PPC64 (OP_LXOR_IMM)
if (!(ins->inst_imm & 0xffff0000)) {
ppc_xori (code, ins->sreg1, ins->dreg, ins->inst_imm);
} else if (!(ins->inst_imm & 0xffff)) {
}
break;
case OP_ISHL:
- ppc_slw (code, ins->sreg1, ins->dreg, ins->sreg2);
+ CASE_PPC64 (OP_LSHL)
+ ppc_shift_left (code, ins->dreg, ins->sreg1, ins->sreg2);
break;
case OP_SHL_IMM:
case OP_ISHL_IMM:
- ppc_rlwinm (code, ins->dreg, ins->sreg1, (ins->inst_imm & 0x1f), 0, (31 - (ins->inst_imm & 0x1f)));
+ CASE_PPC64 (OP_LSHL_IMM)
+ ppc_shift_left_imm (code, ins->dreg, ins->sreg1, MASK_SHIFT_IMM (ins->inst_imm));
break;
case OP_ISHR:
ppc_sraw (code, ins->dreg, ins->sreg1, ins->sreg2);
break;
case OP_SHR_IMM:
- case OP_ISHR_IMM:
- ppc_srawi (code, ins->dreg, ins->sreg1, (ins->inst_imm & 0x1f));
+ ppc_shift_right_arith_imm (code, ins->dreg, ins->sreg1, MASK_SHIFT_IMM (ins->inst_imm));
break;
case OP_SHR_UN_IMM:
- case OP_ISHR_UN_IMM:
- if (ins->inst_imm)
- ppc_rlwinm (code, ins->dreg, ins->sreg1, (32 - (ins->inst_imm & 0x1f)), (ins->inst_imm & 0x1f), 31);
+ if (MASK_SHIFT_IMM (ins->inst_imm))
+ ppc_shift_right_imm (code, ins->dreg, ins->sreg1, MASK_SHIFT_IMM (ins->inst_imm));
else
ppc_mr (code, ins->dreg, ins->sreg1);
break;
ppc_srw (code, ins->dreg, ins->sreg1, ins->sreg2);
break;
case OP_INOT:
+ CASE_PPC64 (OP_LNOT)
ppc_not (code, ins->dreg, ins->sreg1);
break;
case OP_INEG:
+ CASE_PPC64 (OP_LNEG)
ppc_neg (code, ins->dreg, ins->sreg1);
break;
case OP_IMUL:
- ppc_mullw (code, ins->dreg, ins->sreg1, ins->sreg2);
+ CASE_PPC64 (OP_LMUL)
+ ppc_multiply (code, ins->dreg, ins->sreg1, ins->sreg2);
break;
case OP_IMUL_IMM:
case OP_MUL_IMM:
+ CASE_PPC64 (OP_LMUL_IMM)
if (ppc_is_imm16 (ins->inst_imm)) {
ppc_mulli (code, ins->dreg, ins->sreg1, ins->inst_imm);
} else {
}
break;
case OP_IMUL_OVF:
+ CASE_PPC64 (OP_LMUL_OVF)
/* we annot use mcrxr, since it's not implemented on some processors
* XER format: SO, OV, CA, reserved [21 bits], count [8 bits]
*/
- ppc_mullwo (code, ins->dreg, ins->sreg1, ins->sreg2);
+ if (ins->opcode == OP_IMUL_OVF)
+ ppc_mullwo (code, ins->dreg, ins->sreg1, ins->sreg2);
+#ifdef __mono_ppc64__
+ else
+ ppc_mulldo (code, ins->dreg, ins->sreg1, ins->sreg2);
+#endif
ppc_mfspr (code, ppc_r0, ppc_xer);
ppc_andisd (code, ppc_r0, ppc_r0, (1<<14));
EMIT_COND_SYSTEM_EXCEPTION_FLAGS (PPC_BR_FALSE, PPC_BR_EQ, "OverflowException");
break;
case OP_IMUL_OVF_UN:
+ CASE_PPC64 (OP_LMUL_OVF_UN)
/* we first multiply to get the high word and compare to 0
* to set the flags, then the result is discarded and then
* we multiply to get the lower * bits result
*/
- ppc_mulhwu (code, ppc_r0, ins->sreg1, ins->sreg2);
+ if (ins->opcode == OP_IMUL_OVF_UN)
+ ppc_mulhwu (code, ppc_r0, ins->sreg1, ins->sreg2);
+#ifdef __mono_ppc64__
+ else
+ ppc_mulhdu (code, ppc_r0, ins->sreg1, ins->sreg2);
+#endif
ppc_cmpi (code, 0, 0, ppc_r0, 0);
EMIT_COND_SYSTEM_EXCEPTION (CEE_BNE_UN - CEE_BEQ, "OverflowException");
- ppc_mullw (code, ins->dreg, ins->sreg1, ins->sreg2);
+ ppc_multiply (code, ins->dreg, ins->sreg1, ins->sreg2);
break;
case OP_ICONST:
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_lis (code, ins->dreg, 0);
- ppc_ori (code, ins->dreg, ins->dreg, 0);
+ ppc_load_sequence (code, ins->dreg, 0);
break;
- case OP_ICONV_TO_I4:
- case OP_ICONV_TO_U4:
+ CASE_PPC32 (OP_ICONV_TO_I4)
+ CASE_PPC32 (OP_ICONV_TO_U4)
case OP_MOVE:
ppc_mr (code, ins->dreg, ins->sreg1);
break;
ppc_frsp (code, ins->dreg, ins->sreg1);
break;
case OP_JMP: {
- int i, pos = 0;
+ int i, pos;
/*
* Keep in sync with mono_arch_emit_epilog
* 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_lwz (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_lwzx (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);
}
code = emit_load_volatile_arguments (cfg, code);
if (ppc_is_imm16 (cfg->stack_usage)) {
- ppc_addic (code, ppc_sp, 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_sp, 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) {
ppc_lfd (code, i, -pos, cfg->frame_reg);
}
}*/
- /* FIXME: restore registers before changing ppc_sp */
+ pos = 0;
for (i = 31; i >= 13; --i) {
if (cfg->used_int_regs & (1 << i)) {
- pos += sizeof (gulong);
- ppc_lwz (code, i, -pos, ppc_sp);
+ pos += sizeof (gpointer);
+ ppc_ldptr (code, i, -pos, ppc_r11);
}
}
} else {
/* FIXME restore from MonoLMF: though this can't happen yet */
}
+ 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_lwz (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_stw (code, ppc_r0, 0, ins->sreg1);
+ ppc_stptr (code, ppc_r0, 0, ins->sreg1);
break;
}
case OP_FCALL:
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) {
- ppc_lis (code, ppc_r0, 0);
- ppc_ori (code, ppc_r0, ppc_r0, 0);
+ 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 {
case OP_VCALL2_REG:
case OP_VOIDCALL_REG:
case OP_CALL_REG:
+#ifdef PPC_USES_FUNCTION_DESCRIPTOR
+ ppc_ldptr (code, ppc_r0, 0, ins->sreg1);
+ /* FIXME: if we know that this is a method, we
+ can omit this load */
+ ppc_ldptr (code, ppc_r2, 8, ins->sreg1);
+ ppc_mtlr (code, ppc_r0);
+#else
ppc_mtlr (code, ins->sreg1);
+#endif
ppc_blrl (code);
/* FIXME: this should be handled somewhere else in the new jit */
code = emit_move_return_value (cfg, ins, code);
case OP_VCALL2_MEMBASE:
case OP_VOIDCALL_MEMBASE:
case OP_CALL_MEMBASE:
- ppc_lwz (code, ppc_r0, ins->inst_offset, ins->sreg1);
+ if (cfg->compile_aot && ins->sreg1 == ppc_r11) {
+ /* The trampolines clobber this */
+ ppc_mr (code, ppc_r29, ins->sreg1);
+ ppc_ldptr (code, ppc_r0, ins->inst_offset, ppc_r29);
+ } else {
+ 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 */
int area_offset = alloca_waste;
area_offset &= ~31;
ppc_addi (code, ppc_r11, ins->sreg1, alloca_waste + 31);
- ppc_rlwinm (code, ppc_r11, ppc_r11, 0, 0, 27);
+ /* FIXME: should be calculated from MONO_ARCH_FRAME_ALIGNMENT */
+ ppc_clear_right_imm (code, ppc_r11, ppc_r11, 4);
/* use ctr to store the number of words to 0 if needed */
if (ins->flags & MONO_INST_INIT) {
/* we zero 4 bytes at a time:
* it negative and iterate billions of times.
*/
ppc_addi (code, ppc_r0, ins->sreg1, 7);
- ppc_srawi (code, ppc_r0, ppc_r0, 2);
+ ppc_shift_right_arith_imm (code, ppc_r0, ppc_r0, 2);
ppc_mtctr (code, ppc_r0);
}
- ppc_lwz (code, ppc_r0, 0, ppc_sp);
+ ppc_ldptr (code, ppc_r0, 0, ppc_sp);
ppc_neg (code, ppc_r11, ppc_r11);
- ppc_stwux (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 */
if (ins->flags & MONO_INST_INIT) {
/* adjust the dest reg by -4 so we can use stwu */
/* we actually adjust -8 because we let the loop
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) {
- ppc_lis (code, ppc_r0, 0);
- ppc_ori (code, ppc_r0, ppc_r0, 0);
+ 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 {
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) {
- ppc_lis (code, ppc_r0, 0);
- ppc_ori (code, ppc_r0, ppc_r0, 0);
+ 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 {
code = emit_reserve_param_area (cfg, code);
ppc_mflr (code, ppc_r0);
if (ppc_is_imm16 (spvar->inst_offset)) {
- ppc_stw (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_stwx (code, ppc_r0, ppc_r11, spvar->inst_basereg);
+ ppc_stptr_indexed (code, ppc_r0, ppc_r11, spvar->inst_basereg);
}
break;
}
if (ins->sreg1 != ppc_r3)
ppc_mr (code, ppc_r3, ins->sreg1);
if (ppc_is_imm16 (spvar->inst_offset)) {
- ppc_lwz (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_lwzx (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);
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_lwz (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;
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:
break;
case OP_CEQ:
case OP_ICEQ:
+ CASE_PPC64 (OP_LCEQ)
ppc_li (code, ins->dreg, 0);
ppc_bc (code, PPC_BR_FALSE, PPC_BR_EQ, 2);
ppc_li (code, ins->dreg, 1);
case OP_CLT_UN:
case OP_ICLT:
case OP_ICLT_UN:
+ CASE_PPC64 (OP_LCLT)
+ CASE_PPC64 (OP_LCLT_UN)
ppc_li (code, ins->dreg, 1);
ppc_bc (code, PPC_BR_TRUE, PPC_BR_LT, 2);
ppc_li (code, ins->dreg, 0);
case OP_CGT_UN:
case OP_ICGT:
case OP_ICGT_UN:
+ CASE_PPC64 (OP_LCGT)
+ CASE_PPC64 (OP_LCGT_UN)
ppc_li (code, ins->dreg, 1);
ppc_bc (code, PPC_BR_TRUE, PPC_BR_GT, 2);
ppc_li (code, ins->dreg, 0);
case OP_COND_EXC_ILE_UN:
EMIT_COND_SYSTEM_EXCEPTION (ins->opcode - OP_COND_EXC_IEQ, ins->inst_p1);
break;
- case OP_COND_EXC_C:
- /* check XER [0-3] (SO, OV, CA): we can't use mcrxr
- */
- /*ppc_mfspr (code, ppc_r0, ppc_xer);
- ppc_andisd (code, ppc_r0, ppc_r0, (1<<14));
- EMIT_COND_SYSTEM_EXCEPTION_FLAGS (PPC_BR_FALSE, PPC_BR_EQ, "OverflowException");
- break;*/
- case OP_COND_EXC_OV:
- /*ppc_mcrxr (code, 0);
- EMIT_COND_SYSTEM_EXCEPTION (CEE_BGT - CEE_BEQ, ins->inst_p1);
- break;*/
- case OP_COND_EXC_NC:
- case OP_COND_EXC_NO:
- g_assert_not_reached ();
- break;
case OP_IBEQ:
case OP_IBNE_UN:
case OP_IBLT:
/* 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:
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 */
case OP_FCONV_TO_U:
code = emit_float_to_int (cfg, code, ins->dreg, ins->sreg1, 4, FALSE);
break;
- case OP_FCONV_TO_I8:
- case OP_FCONV_TO_U8:
- g_assert_not_reached ();
- /* Implemented as helper calls */
- break;
case OP_LCONV_TO_R_UN:
g_assert_not_reached ();
/* Implemented as helper calls */
break;
case OP_LCONV_TO_OVF_I4_2:
case OP_LCONV_TO_OVF_I: {
+#ifdef __mono_ppc64__
+ NOT_IMPLEMENTED;
+#else
guint8 *negative_branch, *msword_positive_branch, *msword_negative_branch, *ovf_ex_target;
// Check if its negative
ppc_cmpi (code, 0, 0, ins->sreg1, 0);
if (ins->dreg != ins->sreg1)
ppc_mr (code, ins->dreg, ins->sreg1);
break;
+#endif
}
case OP_SQRT:
ppc_fsqrtd (code, ins->dreg, ins->sreg1);
break;
case OP_JUMP_TABLE:
mono_add_patch_info (cfg, offset, (MonoJumpInfoType)ins->inst_i1, ins->inst_p0);
- ppc_load (code, ins->dreg, 0x0f0f0f0f);
+#ifdef __mono_ppc64__
+ ppc_load_sequence (code, ins->dreg, (guint64)0x0f0f0f0f0f0f0f0fLL);
+#else
+ ppc_load_sequence (code, ins->dreg, (gulong)0x0f0f0f0fL);
+#endif
break;
}
- default:
- g_warning ("unknown opcode %s in %s()\n", mono_inst_name (ins->opcode), __FUNCTION__);
- g_assert_not_reached ();
- }
- if ((cfg->opt & MONO_OPT_BRANCH) && ((code - cfg->native_code - offset) > max_len)) {
- g_warning ("wrong maximal instruction length of instruction %s (expected %d, got %d)",
- mono_inst_name (ins->opcode), max_len, code - cfg->native_code - offset);
- g_assert_not_reached ();
- }
-
+#ifdef __mono_ppc64__
+ case OP_ICONV_TO_I4:
+ case OP_SEXT_I4:
+ ppc_extsw (code, ins->dreg, ins->sreg1);
+ break;
+ case OP_ICONV_TO_U4:
+ case OP_ZEXT_I4:
+ ppc_clrldi (code, ins->dreg, ins->sreg1, 32);
+ break;
+ case OP_ICONV_TO_R4:
+ case OP_ICONV_TO_R8:
+ case OP_LCONV_TO_R4:
+ case OP_LCONV_TO_R8: {
+ int tmp;
+ if (ins->opcode == OP_ICONV_TO_R4 || ins->opcode == OP_ICONV_TO_R8) {
+ ppc_extsw (code, ppc_r0, ins->sreg1);
+ tmp = ppc_r0;
+ } else {
+ tmp = ins->sreg1;
+ }
+ 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);
+ break;
+ }
+ case OP_LSHR:
+ ppc_srad (code, ins->dreg, ins->sreg1, ins->sreg2);
+ break;
+ case OP_LSHR_UN:
+ ppc_srd (code, ins->dreg, ins->sreg1, ins->sreg2);
+ break;
+ case OP_COND_EXC_C:
+ /* check XER [0-3] (SO, OV, CA): we can't use mcrxr
+ */
+ ppc_mfspr (code, ppc_r0, ppc_xer);
+ ppc_andisd (code, ppc_r0, ppc_r0, (1 << 13)); /* CA */
+ EMIT_COND_SYSTEM_EXCEPTION_FLAGS (PPC_BR_FALSE, PPC_BR_EQ, ins->inst_p1);
+ break;
+ case OP_COND_EXC_OV:
+ ppc_mfspr (code, ppc_r0, ppc_xer);
+ ppc_andisd (code, ppc_r0, ppc_r0, (1 << 14)); /* OV */
+ EMIT_COND_SYSTEM_EXCEPTION_FLAGS (PPC_BR_FALSE, PPC_BR_EQ, ins->inst_p1);
+ break;
+ case OP_LBEQ:
+ case OP_LBNE_UN:
+ case OP_LBLT:
+ case OP_LBLT_UN:
+ case OP_LBGT:
+ case OP_LBGT_UN:
+ case OP_LBGE:
+ case OP_LBGE_UN:
+ case OP_LBLE:
+ case OP_LBLE_UN:
+ EMIT_COND_BRANCH (ins, ins->opcode - OP_LBEQ);
+ break;
+ case OP_FCONV_TO_I8:
+ code = emit_float_to_int (cfg, code, ins->dreg, ins->sreg1, 8, TRUE);
+ break;
+ case OP_FCONV_TO_U8:
+ code = emit_float_to_int (cfg, code, ins->dreg, ins->sreg1, 8, FALSE);
+ break;
+ case OP_STOREI4_MEMBASE_REG:
+ if (ppc_is_imm16 (ins->inst_offset)) {
+ ppc_stw (code, ins->sreg1, ins->inst_offset, ins->inst_destbasereg);
+ } else {
+ ppc_load (code, ppc_r0, ins->inst_offset);
+ ppc_stwx (code, ins->sreg1, ins->inst_destbasereg, ppc_r0);
+ }
+ break;
+ case OP_STOREI4_MEMINDEX:
+ ppc_stwx (code, ins->sreg1, ins->sreg2, ins->inst_destbasereg);
+ break;
+ case OP_ISHR_IMM:
+ ppc_srawi (code, ins->dreg, ins->sreg1, (ins->inst_imm & 0x1f));
+ break;
+ case OP_ISHR_UN_IMM:
+ if (ins->inst_imm & 0x1f)
+ ppc_srwi (code, ins->dreg, ins->sreg1, (ins->inst_imm & 0x1f));
+ else
+ ppc_mr (code, ins->dreg, ins->sreg1);
+ break;
+ case OP_ATOMIC_ADD_NEW_I4:
+ case OP_ATOMIC_ADD_NEW_I8: {
+ guint8 *loop = code, *branch;
+ g_assert (ins->inst_offset == 0);
+ if (ins->opcode == OP_ATOMIC_ADD_NEW_I4)
+ ppc_lwarx (code, ppc_r0, 0, ins->inst_basereg);
+ else
+ ppc_ldarx (code, ppc_r0, 0, ins->inst_basereg);
+ ppc_add (code, ppc_r0, ppc_r0, ins->sreg2);
+ if (ins->opcode == OP_ATOMIC_ADD_NEW_I4)
+ ppc_stwcxd (code, ppc_r0, 0, ins->inst_basereg);
+ else
+ ppc_stdcxd (code, ppc_r0, 0, ins->inst_basereg);
+ branch = code;
+ ppc_bc (code, PPC_BR_FALSE, PPC_BR_EQ, 0);
+ ppc_patch (branch, loop);
+ 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__);
+ g_assert_not_reached ();
+ }
+
+ if ((cfg->opt & MONO_OPT_BRANCH) && ((code - cfg->native_code - offset) > max_len)) {
+ g_warning ("wrong maximal instruction length of instruction %s (expected %d, got %ld)",
+ mono_inst_name (ins->opcode), max_len, (glong)(code - cfg->native_code - offset));
+ g_assert_not_reached ();
+ }
+
cpos += max_len;
last_ins = ins;
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);
}
-#define patch_lis_ori(ip,val) do {\
+#ifdef __mono_ppc64__
+#define patch_load_sequence(ip,val) do {\
+ guint16 *__load = (guint16*)(ip); \
+ 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 {\
guint16 *__lis_ori = (guint16*)(ip); \
- __lis_ori [1] = (((guint32)(val)) >> 16) & 0xffff; \
- __lis_ori [3] = ((guint32)(val)) & 0xffff; \
+ __lis_ori [1] = (((gulong)(val)) >> 16) & 0xffff; \
+ __lis_ori [3] = ((gulong)(val)) & 0xffff; \
} while (0)
+#endif
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;
unsigned char *target;
+ gboolean is_fd = FALSE;
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_lis_ori (ip, ip);
+ patch_load_sequence (ip, ip);
continue;
case MONO_PATCH_INFO_METHOD_REL:
g_assert_not_reached ();
gpointer *table = (gpointer *)patch_info->data.table->table;
int i;
- patch_lis_ori (ip, table);
+ patch_load_sequence (ip, table);
- for (i = 0; i < patch_info->data.table->table_size; i++) {
- table [i] = (int)patch_info->data.table->table [i] + code;
+ for (i = 0; i < patch_info->data.table->table_size; i++) {
+ table [i] = (glong)patch_info->data.table->table [i] + code;
}
/* we put into the table the absolute address, no need for ppc_patch in this case */
continue;
case MONO_PATCH_INFO_TYPE_FROM_HANDLE:
case MONO_PATCH_INFO_LDTOKEN:
/* from OP_AOTCONST : lis + ori */
- patch_lis_ori (ip, target);
+ patch_load_sequence (ip, target);
continue;
case MONO_PATCH_INFO_R4:
case MONO_PATCH_INFO_R8:
case MONO_PATCH_INFO_EXC_OVF:
/* everything is dealt with at epilog output time */
continue;
+#ifdef PPC_USES_FUNCTION_DESCRIPTOR
+ case MONO_PATCH_INFO_INTERNAL_METHOD:
+ case MONO_PATCH_INFO_ABS:
+ case MONO_PATCH_INFO_CLASS_INIT:
+ case MONO_PATCH_INFO_RGCTX_FETCH:
+ is_fd = TRUE;
+ break;
+#endif
default:
break;
}
- ppc_patch (ip, target);
+ ppc_patch_full (ip, target, is_fd);
}
}
* 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_stw (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_stw (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++) {
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;
tracing = 1;
sig = mono_method_signature (method);
- cfg->code_size = 256 + sig->param_count * 20;
+ 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_stw (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;
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 {
g_assert ((alloc_size & (MONO_ARCH_FRAME_ALIGNMENT-1)) == 0);
if (alloc_size) {
if (ppc_is_imm16 (-alloc_size)) {
- ppc_stwu (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_stwux (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_stw (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
g_assert (inst);
if (ppc_is_imm16 (inst->inst_offset)) {
- ppc_stw (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_stwx (code, ainfo->reg, ppc_r11, inst->inst_basereg);
+ ppc_stptr_indexed (code, ainfo->reg, ppc_r11, inst->inst_basereg);
}
}
else if (ainfo->regtype == RegTypeFP)
ppc_fmr (code, inst->dreg, ainfo->reg);
else if (ainfo->regtype == RegTypeBase) {
- ppc_lwz (code, ppc_r11, 0, ppc_sp);
- ppc_lwz (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) {
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 {
+ 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_sthx (code, ainfo->reg, ppc_r11, inst->inst_basereg);
+ ppc_str_indexed (code, ainfo->reg, ppc_r11, inst->inst_basereg);
}
break;
+#else
case 8:
if (ppc_is_imm16 (inst->inst_offset + 4)) {
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);
}
break;
+#endif
default:
if (ppc_is_imm16 (inst->inst_offset)) {
- ppc_stw (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_stwx (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_lwz (code, ppc_r11, 0, ppc_sp);
- ppc_lwz (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 {
+ 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_sthx (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_stw (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_stwx (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;
}
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
ppc_stb (code, ainfo->reg + cur_reg, doffset, inst->inst_basereg);
else
#endif
- ppc_stw (code, ainfo->reg + cur_reg, doffset, inst->inst_basereg);
+ {
+#ifdef __mono_ppc64__
+ if (ainfo->bytes) {
+ g_assert (cur_reg == 0);
+ ppc_sldi (code, ppc_r0, ainfo->reg,
+ (sizeof (gpointer) - ainfo->bytes) * 8);
+ ppc_stptr (code, ppc_r0, doffset, inst->inst_basereg);
+ } else
+#endif
+ {
+ ppc_stptr (code, ainfo->reg + cur_reg, doffset,
+ inst->inst_basereg);
+ }
+ }
soffset += sizeof (gpointer);
doffset += sizeof (gpointer);
}
if (ainfo->vtsize) {
+ /* FIXME: we need to do the shifting here, too */
+ if (ainfo->bytes)
+ NOT_IMPLEMENTED;
/* load the previous stack pointer in r11 (r0 gets overwritten by the memcpy) */
- ppc_lwz (code, ppc_r11, 0, ppc_sp);
- if ((size & 3) != 0) {
+ 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,
ppc_r11, ainfo->offset + soffset);
/* if it was originally a RegTypeBase */
if (ainfo->offset) {
/* load the previous stack pointer in r11 */
- ppc_lwz (code, ppc_r11, 0, ppc_sp);
- ppc_lwz (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);
}
MonoInst *addr = cfg->tailcall_valuetype_addrs [tailcall_struct_index];
g_assert (ppc_is_imm16 (addr->inst_offset));
- ppc_stw (code, ppc_r11, addr->inst_offset, addr->inst_basereg);
+ ppc_stptr (code, ppc_r11, addr->inst_offset, addr->inst_basereg);
tailcall_struct_index++;
}
}
if (method->wrapper_type == MONO_WRAPPER_NATIVE_TO_MANAGED) {
- ppc_load (code, ppc_r3, cfg->domain);
+ if (cfg->compile_aot)
+ /* AOT code is only used in the root domain */
+ ppc_load_ptr (code, ppc_r3, 0);
+ else
+ 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) {
- ppc_lis (code, ppc_r0, 0);
- ppc_ori (code, ppc_r0, ppc_r0, 0);
+ 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 {
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) {
- ppc_lis (code, ppc_r0, 0);
- ppc_ori (code, ppc_r0, ppc_r0, 0);
+ 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 {
*/
ppc_addi (code, ppc_r11, ppc_sp, alloc_size - lmf_offset);
/* ppc_r3 is the result from mono_get_lmf_addr () */
- ppc_stw (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_lwz (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, previous_lmf), ppc_r3);
- ppc_stw (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_stw (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_stw (code, ppc_r0, G_STRUCT_OFFSET(MonoLMF, method), ppc_r11);
- ppc_stw (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 (code, ppc_r0, 0x01010101);
- ppc_stw (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)
code = mono_arch_instrument_prolog (cfg, mono_trace_enter_method, code, TRUE);
cfg->code_len = code - cfg->native_code;
- g_assert (cfg->code_len < cfg->code_size);
+ g_assert (cfg->code_len <= cfg->code_size);
g_free (cinfo);
return code;
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_lwz (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_lwz (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_stw (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_lmw (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);
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_lwz (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_lwz (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_lwzx (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);
}
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_lwz (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_lwz (code, i, -pos, ppc_r11);
+ pos += sizeof (mgreg_t);
+ ppc_ldr (code, i, -pos, ppc_r11);
}
}
ppc_mr (code, ppc_sp, ppc_r11);
/*
* make sure we have enough space for exceptions
- * 24 is the simulated call to throw_exception_by_name
*/
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 += (2 * PPC_LOAD_SEQUENCE_LENGTH) + 5 * 4;
exc_throw_found [i] = TRUE;
}
} else if (patch_info->type == MONO_PATCH_INFO_BB_OVF)
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 += (2 * PPC_LOAD_SEQUENCE_LENGTH) + 5 * 4;
exc_throw_found [i] = TRUE;
}
max_epilog_size += 8;
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: {
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]) {
} 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_lis (code, ppc_r0, 0);
- ppc_ori (code, ppc_r0, ppc_r0, 0);
+ 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;
}
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)
{
return 0;
return 1;
}
+#endif
static void
setup_tls_access (void)
{
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) {
val >>= 6;
if (*ins & 2) {
/* absolute */
- ins = (guint32*)val;
+ ins = (guint32*)(long)val;
} else {
ins = (guint32*) ((char*)ins + val);
}
val >>= 6;
if (*ins & 2) {
/* absolute */
- ins = (guint32*)val;
+ ins = (guint32*)(long)val;
} else {
ins = (guint32*) ((char*)ins + val);
}
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) {
}
}
}
+
+ 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) {
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
}
void
#ifdef MONO_ARCH_HAVE_IMT
-#define CMP_SIZE 12
+#define CMP_SIZE (PPC_LOAD_SEQUENCE_LENGTH + 4)
#define BR_SIZE 4
+#define LOADSTORE_SIZE 4
#define JUMP_IMM_SIZE 12
-#define JUMP_IMM32_SIZE 16
+#define JUMP_IMM32_SIZE (PPC_LOAD_SEQUENCE_LENGTH + 8)
#define ENABLE_WRONG_METHOD_CHECK 0
/*
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 += BR_SIZE + JUMP_IMM_SIZE;
+ 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 += JUMP_IMM_SIZE;
+ item->chunk_size += LOADSTORE_SIZE + JUMP_IMM_SIZE;
#if ENABLE_WRONG_METHOD_CHECK
item->chunk_size += CMP_SIZE + BR_SIZE + 4;
#endif
code = mono_method_alloc_generic_virtual_thunk (domain, size);
} else {
/* the initial load of the vtable address */
- size += 8;
- code = mono_code_manager_reserve (domain->code_mp, size);
+ size += PPC_LOAD_SEQUENCE_LENGTH + LOADSTORE_SIZE;
+ code = mono_domain_code_reserve (domain, size);
}
start = code;
- if (!fail_tramp)
- ppc_load (code, ppc_r11, (guint32)(& (vtable->vtable [0])));
+ if (!fail_tramp) {
+ /*
+ * We need to save and restore r11 because it might be
+ * used by the caller as the vtable register, so
+ * clobbering it will trip up the magic trampoline.
+ *
+ * FIXME: Get rid of this by making sure that r11 is
+ * not used as the vtable register in interface calls.
+ */
+ 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];
item->code_target = code;
if (item->is_equals) {
if (item->check_target_idx) {
if (!item->compare_done) {
- ppc_load (code, ppc_r0, (guint32)item->key);
- ppc_cmpl (code, 0, 0, MONO_ARCH_IMT_REG, ppc_r0);
+ 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);
- else
- ppc_lwz (code, ppc_r0, (sizeof (gpointer) * item->value.vtable_slot), ppc_r11);
+ if (item->has_target_code) {
+ ppc_load_ptr (code, ppc_r0, item->value.target_code);
+ } else {
+ 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);
} else {
if (fail_tramp) {
- ppc_load (code, ppc_r0, (guint32)item->key);
- ppc_cmpl (code, 0, 0, MONO_ARCH_IMT_REG, ppc_r0);
+ ppc_load (code, ppc_r0, (gulong)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);
- 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;
/* enable the commented code to assert on wrong method */
#if ENABLE_WRONG_METHOD_CHECK
ppc_load (code, ppc_r0, (guint32)item->key);
- ppc_cmpl (code, 0, 0, MONO_ARCH_IMT_REG, ppc_r0);
+ 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);
#endif
- ppc_lwz (code, ppc_r0, (sizeof (gpointer) * item->value.vtable_slot), ppc_r11);
+ 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
}
}
} else {
- ppc_load (code, ppc_r0, (guint32)item->key);
- ppc_cmpl (code, 0, 0, MONO_ARCH_IMT_REG, ppc_r0);
+ ppc_load (code, ppc_r0, (gulong)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_LT, 0);
}
}
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;
-MonoObject*
-mono_arch_find_this_argument (gpointer *regs, MonoMethod *method, MonoGenericSharingContext *gsctx)
-{
- return mono_arch_get_this_arg_from_call (gsctx, mono_method_signature (method), (gssize*)regs, NULL);
+ return (MonoMethod*)(gsize) r [MONO_ARCH_IMT_REG];
}
#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*
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;
+ g_assert (reg >= ppc_r13);
+
+ return (gpointer)(gsize)ctx->regs [reg - ppc_r13];
+}
+
+guint32
+mono_arch_get_patch_offset (guint8 *code)
+{
+ 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;
+}
+
+/*
+ * 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;
+}
+
+/* Soft Debug support */
+#ifdef MONO_ARCH_SOFT_DEBUG_SUPPORTED
+
+/*
+ * BREAKPOINTS
+ */
+
+/*
+ * mono_arch_set_breakpoint:
+ *
+ * See mini-amd64.c for docs.
+ */
+void
+mono_arch_set_breakpoint (MonoJitInfo *ji, guint8 *ip)
+{
+ guint8 *code = ip;
+ guint8 *orig_code = code;
+
+ ppc_load_sequence (code, ppc_r11, (gsize)bp_trigger_page);
+ ppc_ldptr (code, ppc_r11, 0, ppc_r11);
+
+ g_assert (code - orig_code == BREAKPOINT_SIZE);
+
+ mono_arch_flush_icache (orig_code, code - orig_code);
+}
+
+/*
+ * mono_arch_clear_breakpoint:
+ *
+ * See mini-amd64.c for docs.
+ */
+void
+mono_arch_clear_breakpoint (MonoJitInfo *ji, guint8 *ip)
+{
+ guint8 *code = ip;
+ int i;
+
+ for (i = 0; i < BREAKPOINT_SIZE / 4; ++i)
+ ppc_nop (code);
+
+ mono_arch_flush_icache (ip, code - ip);
+}
+
+/*
+ * mono_arch_is_breakpoint_event:
+ *
+ * See mini-amd64.c for docs.
+ */
+gboolean
+mono_arch_is_breakpoint_event (void *info, void *sigctx)
+{
+ siginfo_t* sinfo = (siginfo_t*) info;
+ /* Sometimes the address is off by 4 */
+ if (sinfo->si_addr >= bp_trigger_page && (guint8*)sinfo->si_addr <= (guint8*)bp_trigger_page + 128)
+ return TRUE;
+ else
+ return FALSE;
+}
+
+/*
+ * mono_arch_get_ip_for_breakpoint:
+ *
+ * See mini-amd64.c for docs.
+ */
+guint8*
+mono_arch_get_ip_for_breakpoint (MonoJitInfo *ji, MonoContext *ctx)
+{
+ guint8 *ip = MONO_CONTEXT_GET_IP (ctx);
+
+ /* ip points at the ldptr instruction */
+ ip -= PPC_LOAD_SEQUENCE_LENGTH;
+
+ return ip;
+}
+
+/*
+ * mono_arch_skip_breakpoint:
+ *
+ * See mini-amd64.c for docs.
+ */
+void
+mono_arch_skip_breakpoint (MonoContext *ctx)
+{
+ /* skip the ldptr */
+ MONO_CONTEXT_SET_IP (ctx, (guint8*)MONO_CONTEXT_GET_IP (ctx) + 4);
+}
+
+/*
+ * SINGLE STEPPING
+ */
- MONO_INST_NEW (cfg, ins, OP_TLS_GET);
- ins->inst_offset = monothread_key;
- return ins;
+/*
+ * mono_arch_start_single_stepping:
+ *
+ * See mini-amd64.c for docs.
+ */
+void
+mono_arch_start_single_stepping (void)
+{
+ mono_mprotect (ss_trigger_page, mono_pagesize (), 0);
+}
+
+/*
+ * mono_arch_stop_single_stepping:
+ *
+ * See mini-amd64.c for docs.
+ */
+void
+mono_arch_stop_single_stepping (void)
+{
+ mono_mprotect (ss_trigger_page, mono_pagesize (), MONO_MMAP_READ);
}
-gpointer
-mono_arch_context_get_int_reg (MonoContext *ctx, int reg)
+/*
+ * mono_arch_is_single_step_event:
+ *
+ * See mini-amd64.c for docs.
+ */
+gboolean
+mono_arch_is_single_step_event (void *info, void *sigctx)
{
- g_assert (reg >= ppc_r13);
+ siginfo_t* sinfo = (siginfo_t*) info;
+ /* Sometimes the address is off by 4 */
+ if (sinfo->si_addr >= ss_trigger_page && (guint8*)sinfo->si_addr <= (guint8*)ss_trigger_page + 128)
+ return TRUE;
+ else
+ return FALSE;
+}
- return (gpointer)ctx->regs [reg - ppc_r13];
+/*
+ * mono_arch_get_ip_for_single_step:
+ *
+ * See mini-amd64.c for docs.
+ */
+guint8*
+mono_arch_get_ip_for_single_step (MonoJitInfo *ji, MonoContext *ctx)
+{
+ guint8 *ip = MONO_CONTEXT_GET_IP (ctx);
+
+ /* ip points after the ldptr instruction */
+ return ip;
+}
+
+/*
+ * mono_arch_skip_single_step:
+ *
+ * See mini-amd64.c for docs.
+ */
+void
+mono_arch_skip_single_step (MonoContext *ctx)
+{
+ /* skip the ldptr */
+ MONO_CONTEXT_SET_IP (ctx, (guint8*)MONO_CONTEXT_GET_IP (ctx) + 4);
+}
+
+/*
+ * mono_arch_create_seq_point_info:
+ *
+ * See mini-amd64.c for docs.
+ */
+gpointer
+mono_arch_get_seq_point_info (MonoDomain *domain, guint8 *code)
+{
+ NOT_IMPLEMENTED;
+ return NULL;
}
+
+#endif