Add qemu 2.4.0
[kvmfornfv.git] / qemu / tcg / tcg.c
diff --git a/qemu/tcg/tcg.c b/qemu/tcg/tcg.c
new file mode 100644 (file)
index 0000000..0892a9b
--- /dev/null
@@ -0,0 +1,2764 @@
+/*
+ * Tiny Code Generator for QEMU
+ *
+ * Copyright (c) 2008 Fabrice Bellard
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+/* define it to use liveness analysis (better code) */
+#define USE_LIVENESS_ANALYSIS
+#define USE_TCG_OPTIMIZATIONS
+
+#include "config.h"
+
+/* Define to jump the ELF file used to communicate with GDB.  */
+#undef DEBUG_JIT
+
+#if !defined(CONFIG_DEBUG_TCG) && !defined(NDEBUG)
+/* define it to suppress various consistency checks (faster) */
+#define NDEBUG
+#endif
+
+#include "qemu-common.h"
+#include "qemu/host-utils.h"
+#include "qemu/timer.h"
+
+/* Note: the long term plan is to reduce the dependencies on the QEMU
+   CPU definitions. Currently they are used for qemu_ld/st
+   instructions */
+#define NO_CPU_IO_DEFS
+#include "cpu.h"
+
+#include "tcg-op.h"
+
+#if UINTPTR_MAX == UINT32_MAX
+# define ELF_CLASS  ELFCLASS32
+#else
+# define ELF_CLASS  ELFCLASS64
+#endif
+#ifdef HOST_WORDS_BIGENDIAN
+# define ELF_DATA   ELFDATA2MSB
+#else
+# define ELF_DATA   ELFDATA2LSB
+#endif
+
+#include "elf.h"
+
+/* Forward declarations for functions declared in tcg-target.c and used here. */
+static void tcg_target_init(TCGContext *s);
+static void tcg_target_qemu_prologue(TCGContext *s);
+static void patch_reloc(tcg_insn_unit *code_ptr, int type,
+                        intptr_t value, intptr_t addend);
+
+/* The CIE and FDE header definitions will be common to all hosts.  */
+typedef struct {
+    uint32_t len __attribute__((aligned((sizeof(void *)))));
+    uint32_t id;
+    uint8_t version;
+    char augmentation[1];
+    uint8_t code_align;
+    uint8_t data_align;
+    uint8_t return_column;
+} DebugFrameCIE;
+
+typedef struct QEMU_PACKED {
+    uint32_t len __attribute__((aligned((sizeof(void *)))));
+    uint32_t cie_offset;
+    uintptr_t func_start;
+    uintptr_t func_len;
+} DebugFrameFDEHeader;
+
+typedef struct QEMU_PACKED {
+    DebugFrameCIE cie;
+    DebugFrameFDEHeader fde;
+} DebugFrameHeader;
+
+static void tcg_register_jit_int(void *buf, size_t size,
+                                 const void *debug_frame,
+                                 size_t debug_frame_size)
+    __attribute__((unused));
+
+/* Forward declarations for functions declared and used in tcg-target.c. */
+static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str);
+static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg1,
+                       intptr_t arg2);
+static void tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg);
+static void tcg_out_movi(TCGContext *s, TCGType type,
+                         TCGReg ret, tcg_target_long arg);
+static void tcg_out_op(TCGContext *s, TCGOpcode opc, const TCGArg *args,
+                       const int *const_args);
+static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, TCGReg arg1,
+                       intptr_t arg2);
+static void tcg_out_call(TCGContext *s, tcg_insn_unit *target);
+static int tcg_target_const_match(tcg_target_long val, TCGType type,
+                                  const TCGArgConstraint *arg_ct);
+static void tcg_out_tb_init(TCGContext *s);
+static void tcg_out_tb_finalize(TCGContext *s);
+
+
+TCGOpDef tcg_op_defs[] = {
+#define DEF(s, oargs, iargs, cargs, flags) { #s, oargs, iargs, cargs, iargs + oargs + cargs, flags },
+#include "tcg-opc.h"
+#undef DEF
+};
+const size_t tcg_op_defs_max = ARRAY_SIZE(tcg_op_defs);
+
+static TCGRegSet tcg_target_available_regs[2];
+static TCGRegSet tcg_target_call_clobber_regs;
+
+#if TCG_TARGET_INSN_UNIT_SIZE == 1
+static __attribute__((unused)) inline void tcg_out8(TCGContext *s, uint8_t v)
+{
+    *s->code_ptr++ = v;
+}
+
+static __attribute__((unused)) inline void tcg_patch8(tcg_insn_unit *p,
+                                                      uint8_t v)
+{
+    *p = v;
+}
+#endif
+
+#if TCG_TARGET_INSN_UNIT_SIZE <= 2
+static __attribute__((unused)) inline void tcg_out16(TCGContext *s, uint16_t v)
+{
+    if (TCG_TARGET_INSN_UNIT_SIZE == 2) {
+        *s->code_ptr++ = v;
+    } else {
+        tcg_insn_unit *p = s->code_ptr;
+        memcpy(p, &v, sizeof(v));
+        s->code_ptr = p + (2 / TCG_TARGET_INSN_UNIT_SIZE);
+    }
+}
+
+static __attribute__((unused)) inline void tcg_patch16(tcg_insn_unit *p,
+                                                       uint16_t v)
+{
+    if (TCG_TARGET_INSN_UNIT_SIZE == 2) {
+        *p = v;
+    } else {
+        memcpy(p, &v, sizeof(v));
+    }
+}
+#endif
+
+#if TCG_TARGET_INSN_UNIT_SIZE <= 4
+static __attribute__((unused)) inline void tcg_out32(TCGContext *s, uint32_t v)
+{
+    if (TCG_TARGET_INSN_UNIT_SIZE == 4) {
+        *s->code_ptr++ = v;
+    } else {
+        tcg_insn_unit *p = s->code_ptr;
+        memcpy(p, &v, sizeof(v));
+        s->code_ptr = p + (4 / TCG_TARGET_INSN_UNIT_SIZE);
+    }
+}
+
+static __attribute__((unused)) inline void tcg_patch32(tcg_insn_unit *p,
+                                                       uint32_t v)
+{
+    if (TCG_TARGET_INSN_UNIT_SIZE == 4) {
+        *p = v;
+    } else {
+        memcpy(p, &v, sizeof(v));
+    }
+}
+#endif
+
+#if TCG_TARGET_INSN_UNIT_SIZE <= 8
+static __attribute__((unused)) inline void tcg_out64(TCGContext *s, uint64_t v)
+{
+    if (TCG_TARGET_INSN_UNIT_SIZE == 8) {
+        *s->code_ptr++ = v;
+    } else {
+        tcg_insn_unit *p = s->code_ptr;
+        memcpy(p, &v, sizeof(v));
+        s->code_ptr = p + (8 / TCG_TARGET_INSN_UNIT_SIZE);
+    }
+}
+
+static __attribute__((unused)) inline void tcg_patch64(tcg_insn_unit *p,
+                                                       uint64_t v)
+{
+    if (TCG_TARGET_INSN_UNIT_SIZE == 8) {
+        *p = v;
+    } else {
+        memcpy(p, &v, sizeof(v));
+    }
+}
+#endif
+
+/* label relocation processing */
+
+static void tcg_out_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type,
+                          TCGLabel *l, intptr_t addend)
+{
+    TCGRelocation *r;
+
+    if (l->has_value) {
+        /* FIXME: This may break relocations on RISC targets that
+           modify instruction fields in place.  The caller may not have 
+           written the initial value.  */
+        patch_reloc(code_ptr, type, l->u.value, addend);
+    } else {
+        /* add a new relocation entry */
+        r = tcg_malloc(sizeof(TCGRelocation));
+        r->type = type;
+        r->ptr = code_ptr;
+        r->addend = addend;
+        r->next = l->u.first_reloc;
+        l->u.first_reloc = r;
+    }
+}
+
+static void tcg_out_label(TCGContext *s, TCGLabel *l, tcg_insn_unit *ptr)
+{
+    intptr_t value = (intptr_t)ptr;
+    TCGRelocation *r;
+
+    assert(!l->has_value);
+
+    for (r = l->u.first_reloc; r != NULL; r = r->next) {
+        patch_reloc(r->ptr, r->type, value, r->addend);
+    }
+
+    l->has_value = 1;
+    l->u.value_ptr = ptr;
+}
+
+TCGLabel *gen_new_label(void)
+{
+    TCGContext *s = &tcg_ctx;
+    TCGLabel *l = tcg_malloc(sizeof(TCGLabel));
+
+    *l = (TCGLabel){
+        .id = s->nb_labels++
+    };
+
+    return l;
+}
+
+#include "tcg-target.c"
+
+/* pool based memory allocation */
+void *tcg_malloc_internal(TCGContext *s, int size)
+{
+    TCGPool *p;
+    int pool_size;
+    
+    if (size > TCG_POOL_CHUNK_SIZE) {
+        /* big malloc: insert a new pool (XXX: could optimize) */
+        p = g_malloc(sizeof(TCGPool) + size);
+        p->size = size;
+        p->next = s->pool_first_large;
+        s->pool_first_large = p;
+        return p->data;
+    } else {
+        p = s->pool_current;
+        if (!p) {
+            p = s->pool_first;
+            if (!p)
+                goto new_pool;
+        } else {
+            if (!p->next) {
+            new_pool:
+                pool_size = TCG_POOL_CHUNK_SIZE;
+                p = g_malloc(sizeof(TCGPool) + pool_size);
+                p->size = pool_size;
+                p->next = NULL;
+                if (s->pool_current) 
+                    s->pool_current->next = p;
+                else
+                    s->pool_first = p;
+            } else {
+                p = p->next;
+            }
+        }
+    }
+    s->pool_current = p;
+    s->pool_cur = p->data + size;
+    s->pool_end = p->data + p->size;
+    return p->data;
+}
+
+void tcg_pool_reset(TCGContext *s)
+{
+    TCGPool *p, *t;
+    for (p = s->pool_first_large; p; p = t) {
+        t = p->next;
+        g_free(p);
+    }
+    s->pool_first_large = NULL;
+    s->pool_cur = s->pool_end = NULL;
+    s->pool_current = NULL;
+}
+
+typedef struct TCGHelperInfo {
+    void *func;
+    const char *name;
+    unsigned flags;
+    unsigned sizemask;
+} TCGHelperInfo;
+
+#include "exec/helper-proto.h"
+
+static const TCGHelperInfo all_helpers[] = {
+#include "exec/helper-tcg.h"
+};
+
+void tcg_context_init(TCGContext *s)
+{
+    int op, total_args, n, i;
+    TCGOpDef *def;
+    TCGArgConstraint *args_ct;
+    int *sorted_args;
+    GHashTable *helper_table;
+
+    memset(s, 0, sizeof(*s));
+    s->nb_globals = 0;
+    
+    /* Count total number of arguments and allocate the corresponding
+       space */
+    total_args = 0;
+    for(op = 0; op < NB_OPS; op++) {
+        def = &tcg_op_defs[op];
+        n = def->nb_iargs + def->nb_oargs;
+        total_args += n;
+    }
+
+    args_ct = g_malloc(sizeof(TCGArgConstraint) * total_args);
+    sorted_args = g_malloc(sizeof(int) * total_args);
+
+    for(op = 0; op < NB_OPS; op++) {
+        def = &tcg_op_defs[op];
+        def->args_ct = args_ct;
+        def->sorted_args = sorted_args;
+        n = def->nb_iargs + def->nb_oargs;
+        sorted_args += n;
+        args_ct += n;
+    }
+
+    /* Register helpers.  */
+    /* Use g_direct_hash/equal for direct pointer comparisons on func.  */
+    s->helpers = helper_table = g_hash_table_new(NULL, NULL);
+
+    for (i = 0; i < ARRAY_SIZE(all_helpers); ++i) {
+        g_hash_table_insert(helper_table, (gpointer)all_helpers[i].func,
+                            (gpointer)&all_helpers[i]);
+    }
+
+    tcg_target_init(s);
+}
+
+void tcg_prologue_init(TCGContext *s)
+{
+    /* init global prologue and epilogue */
+    s->code_buf = s->code_gen_prologue;
+    s->code_ptr = s->code_buf;
+    tcg_target_qemu_prologue(s);
+    flush_icache_range((uintptr_t)s->code_buf, (uintptr_t)s->code_ptr);
+
+#ifdef DEBUG_DISAS
+    if (qemu_loglevel_mask(CPU_LOG_TB_OUT_ASM)) {
+        size_t size = tcg_current_code_size(s);
+        qemu_log("PROLOGUE: [size=%zu]\n", size);
+        log_disas(s->code_buf, size);
+        qemu_log("\n");
+        qemu_log_flush();
+    }
+#endif
+}
+
+void tcg_set_frame(TCGContext *s, int reg, intptr_t start, intptr_t size)
+{
+    s->frame_start = start;
+    s->frame_end = start + size;
+    s->frame_reg = reg;
+}
+
+void tcg_func_start(TCGContext *s)
+{
+    tcg_pool_reset(s);
+    s->nb_temps = s->nb_globals;
+
+    /* No temps have been previously allocated for size or locality.  */
+    memset(s->free_temps, 0, sizeof(s->free_temps));
+
+    s->nb_labels = 0;
+    s->current_frame_offset = s->frame_start;
+
+#ifdef CONFIG_DEBUG_TCG
+    s->goto_tb_issue_mask = 0;
+#endif
+
+    s->gen_first_op_idx = 0;
+    s->gen_last_op_idx = -1;
+    s->gen_next_op_idx = 0;
+    s->gen_next_parm_idx = 0;
+
+    s->be = tcg_malloc(sizeof(TCGBackendData));
+}
+
+static inline void tcg_temp_alloc(TCGContext *s, int n)
+{
+    if (n > TCG_MAX_TEMPS)
+        tcg_abort();
+}
+
+static inline int tcg_global_reg_new_internal(TCGType type, int reg,
+                                              const char *name)
+{
+    TCGContext *s = &tcg_ctx;
+    TCGTemp *ts;
+    int idx;
+
+#if TCG_TARGET_REG_BITS == 32
+    if (type != TCG_TYPE_I32)
+        tcg_abort();
+#endif
+    if (tcg_regset_test_reg(s->reserved_regs, reg))
+        tcg_abort();
+    idx = s->nb_globals;
+    tcg_temp_alloc(s, s->nb_globals + 1);
+    ts = &s->temps[s->nb_globals];
+    ts->base_type = type;
+    ts->type = type;
+    ts->fixed_reg = 1;
+    ts->reg = reg;
+    ts->name = name;
+    s->nb_globals++;
+    tcg_regset_set_reg(s->reserved_regs, reg);
+    return idx;
+}
+
+TCGv_i32 tcg_global_reg_new_i32(int reg, const char *name)
+{
+    int idx;
+
+    idx = tcg_global_reg_new_internal(TCG_TYPE_I32, reg, name);
+    return MAKE_TCGV_I32(idx);
+}
+
+TCGv_i64 tcg_global_reg_new_i64(int reg, const char *name)
+{
+    int idx;
+
+    idx = tcg_global_reg_new_internal(TCG_TYPE_I64, reg, name);
+    return MAKE_TCGV_I64(idx);
+}
+
+static inline int tcg_global_mem_new_internal(TCGType type, int reg,
+                                              intptr_t offset,
+                                              const char *name)
+{
+    TCGContext *s = &tcg_ctx;
+    TCGTemp *ts;
+    int idx;
+
+    idx = s->nb_globals;
+#if TCG_TARGET_REG_BITS == 32
+    if (type == TCG_TYPE_I64) {
+        char buf[64];
+        tcg_temp_alloc(s, s->nb_globals + 2);
+        ts = &s->temps[s->nb_globals];
+        ts->base_type = type;
+        ts->type = TCG_TYPE_I32;
+        ts->fixed_reg = 0;
+        ts->mem_allocated = 1;
+        ts->mem_reg = reg;
+#ifdef HOST_WORDS_BIGENDIAN
+        ts->mem_offset = offset + 4;
+#else
+        ts->mem_offset = offset;
+#endif
+        pstrcpy(buf, sizeof(buf), name);
+        pstrcat(buf, sizeof(buf), "_0");
+        ts->name = strdup(buf);
+        ts++;
+
+        ts->base_type = type;
+        ts->type = TCG_TYPE_I32;
+        ts->fixed_reg = 0;
+        ts->mem_allocated = 1;
+        ts->mem_reg = reg;
+#ifdef HOST_WORDS_BIGENDIAN
+        ts->mem_offset = offset;
+#else
+        ts->mem_offset = offset + 4;
+#endif
+        pstrcpy(buf, sizeof(buf), name);
+        pstrcat(buf, sizeof(buf), "_1");
+        ts->name = strdup(buf);
+
+        s->nb_globals += 2;
+    } else
+#endif
+    {
+        tcg_temp_alloc(s, s->nb_globals + 1);
+        ts = &s->temps[s->nb_globals];
+        ts->base_type = type;
+        ts->type = type;
+        ts->fixed_reg = 0;
+        ts->mem_allocated = 1;
+        ts->mem_reg = reg;
+        ts->mem_offset = offset;
+        ts->name = name;
+        s->nb_globals++;
+    }
+    return idx;
+}
+
+TCGv_i32 tcg_global_mem_new_i32(int reg, intptr_t offset, const char *name)
+{
+    int idx = tcg_global_mem_new_internal(TCG_TYPE_I32, reg, offset, name);
+    return MAKE_TCGV_I32(idx);
+}
+
+TCGv_i64 tcg_global_mem_new_i64(int reg, intptr_t offset, const char *name)
+{
+    int idx = tcg_global_mem_new_internal(TCG_TYPE_I64, reg, offset, name);
+    return MAKE_TCGV_I64(idx);
+}
+
+static inline int tcg_temp_new_internal(TCGType type, int temp_local)
+{
+    TCGContext *s = &tcg_ctx;
+    TCGTemp *ts;
+    int idx, k;
+
+    k = type + (temp_local ? TCG_TYPE_COUNT : 0);
+    idx = find_first_bit(s->free_temps[k].l, TCG_MAX_TEMPS);
+    if (idx < TCG_MAX_TEMPS) {
+        /* There is already an available temp with the right type.  */
+        clear_bit(idx, s->free_temps[k].l);
+
+        ts = &s->temps[idx];
+        ts->temp_allocated = 1;
+        assert(ts->base_type == type);
+        assert(ts->temp_local == temp_local);
+    } else {
+        idx = s->nb_temps;
+#if TCG_TARGET_REG_BITS == 32
+        if (type == TCG_TYPE_I64) {
+            tcg_temp_alloc(s, s->nb_temps + 2);
+            ts = &s->temps[s->nb_temps];
+            ts->base_type = type;
+            ts->type = TCG_TYPE_I32;
+            ts->temp_allocated = 1;
+            ts->temp_local = temp_local;
+            ts->name = NULL;
+            ts++;
+            ts->base_type = type;
+            ts->type = TCG_TYPE_I32;
+            ts->temp_allocated = 1;
+            ts->temp_local = temp_local;
+            ts->name = NULL;
+            s->nb_temps += 2;
+        } else
+#endif
+        {
+            tcg_temp_alloc(s, s->nb_temps + 1);
+            ts = &s->temps[s->nb_temps];
+            ts->base_type = type;
+            ts->type = type;
+            ts->temp_allocated = 1;
+            ts->temp_local = temp_local;
+            ts->name = NULL;
+            s->nb_temps++;
+        }
+    }
+
+#if defined(CONFIG_DEBUG_TCG)
+    s->temps_in_use++;
+#endif
+    return idx;
+}
+
+TCGv_i32 tcg_temp_new_internal_i32(int temp_local)
+{
+    int idx;
+
+    idx = tcg_temp_new_internal(TCG_TYPE_I32, temp_local);
+    return MAKE_TCGV_I32(idx);
+}
+
+TCGv_i64 tcg_temp_new_internal_i64(int temp_local)
+{
+    int idx;
+
+    idx = tcg_temp_new_internal(TCG_TYPE_I64, temp_local);
+    return MAKE_TCGV_I64(idx);
+}
+
+static void tcg_temp_free_internal(int idx)
+{
+    TCGContext *s = &tcg_ctx;
+    TCGTemp *ts;
+    int k;
+
+#if defined(CONFIG_DEBUG_TCG)
+    s->temps_in_use--;
+    if (s->temps_in_use < 0) {
+        fprintf(stderr, "More temporaries freed than allocated!\n");
+    }
+#endif
+
+    assert(idx >= s->nb_globals && idx < s->nb_temps);
+    ts = &s->temps[idx];
+    assert(ts->temp_allocated != 0);
+    ts->temp_allocated = 0;
+
+    k = ts->base_type + (ts->temp_local ? TCG_TYPE_COUNT : 0);
+    set_bit(idx, s->free_temps[k].l);
+}
+
+void tcg_temp_free_i32(TCGv_i32 arg)
+{
+    tcg_temp_free_internal(GET_TCGV_I32(arg));
+}
+
+void tcg_temp_free_i64(TCGv_i64 arg)
+{
+    tcg_temp_free_internal(GET_TCGV_I64(arg));
+}
+
+TCGv_i32 tcg_const_i32(int32_t val)
+{
+    TCGv_i32 t0;
+    t0 = tcg_temp_new_i32();
+    tcg_gen_movi_i32(t0, val);
+    return t0;
+}
+
+TCGv_i64 tcg_const_i64(int64_t val)
+{
+    TCGv_i64 t0;
+    t0 = tcg_temp_new_i64();
+    tcg_gen_movi_i64(t0, val);
+    return t0;
+}
+
+TCGv_i32 tcg_const_local_i32(int32_t val)
+{
+    TCGv_i32 t0;
+    t0 = tcg_temp_local_new_i32();
+    tcg_gen_movi_i32(t0, val);
+    return t0;
+}
+
+TCGv_i64 tcg_const_local_i64(int64_t val)
+{
+    TCGv_i64 t0;
+    t0 = tcg_temp_local_new_i64();
+    tcg_gen_movi_i64(t0, val);
+    return t0;
+}
+
+#if defined(CONFIG_DEBUG_TCG)
+void tcg_clear_temp_count(void)
+{
+    TCGContext *s = &tcg_ctx;
+    s->temps_in_use = 0;
+}
+
+int tcg_check_temp_count(void)
+{
+    TCGContext *s = &tcg_ctx;
+    if (s->temps_in_use) {
+        /* Clear the count so that we don't give another
+         * warning immediately next time around.
+         */
+        s->temps_in_use = 0;
+        return 1;
+    }
+    return 0;
+}
+#endif
+
+/* Note: we convert the 64 bit args to 32 bit and do some alignment
+   and endian swap. Maybe it would be better to do the alignment
+   and endian swap in tcg_reg_alloc_call(). */
+void tcg_gen_callN(TCGContext *s, void *func, TCGArg ret,
+                   int nargs, TCGArg *args)
+{
+    int i, real_args, nb_rets, pi, pi_first;
+    unsigned sizemask, flags;
+    TCGHelperInfo *info;
+
+    info = g_hash_table_lookup(s->helpers, (gpointer)func);
+    flags = info->flags;
+    sizemask = info->sizemask;
+
+#if defined(__sparc__) && !defined(__arch64__) \
+    && !defined(CONFIG_TCG_INTERPRETER)
+    /* We have 64-bit values in one register, but need to pass as two
+       separate parameters.  Split them.  */
+    int orig_sizemask = sizemask;
+    int orig_nargs = nargs;
+    TCGv_i64 retl, reth;
+
+    TCGV_UNUSED_I64(retl);
+    TCGV_UNUSED_I64(reth);
+    if (sizemask != 0) {
+        TCGArg *split_args = __builtin_alloca(sizeof(TCGArg) * nargs * 2);
+        for (i = real_args = 0; i < nargs; ++i) {
+            int is_64bit = sizemask & (1 << (i+1)*2);
+            if (is_64bit) {
+                TCGv_i64 orig = MAKE_TCGV_I64(args[i]);
+                TCGv_i32 h = tcg_temp_new_i32();
+                TCGv_i32 l = tcg_temp_new_i32();
+                tcg_gen_extr_i64_i32(l, h, orig);
+                split_args[real_args++] = GET_TCGV_I32(h);
+                split_args[real_args++] = GET_TCGV_I32(l);
+            } else {
+                split_args[real_args++] = args[i];
+            }
+        }
+        nargs = real_args;
+        args = split_args;
+        sizemask = 0;
+    }
+#elif defined(TCG_TARGET_EXTEND_ARGS) && TCG_TARGET_REG_BITS == 64
+    for (i = 0; i < nargs; ++i) {
+        int is_64bit = sizemask & (1 << (i+1)*2);
+        int is_signed = sizemask & (2 << (i+1)*2);
+        if (!is_64bit) {
+            TCGv_i64 temp = tcg_temp_new_i64();
+            TCGv_i64 orig = MAKE_TCGV_I64(args[i]);
+            if (is_signed) {
+                tcg_gen_ext32s_i64(temp, orig);
+            } else {
+                tcg_gen_ext32u_i64(temp, orig);
+            }
+            args[i] = GET_TCGV_I64(temp);
+        }
+    }
+#endif /* TCG_TARGET_EXTEND_ARGS */
+
+    pi_first = pi = s->gen_next_parm_idx;
+    if (ret != TCG_CALL_DUMMY_ARG) {
+#if defined(__sparc__) && !defined(__arch64__) \
+    && !defined(CONFIG_TCG_INTERPRETER)
+        if (orig_sizemask & 1) {
+            /* The 32-bit ABI is going to return the 64-bit value in
+               the %o0/%o1 register pair.  Prepare for this by using
+               two return temporaries, and reassemble below.  */
+            retl = tcg_temp_new_i64();
+            reth = tcg_temp_new_i64();
+            s->gen_opparam_buf[pi++] = GET_TCGV_I64(reth);
+            s->gen_opparam_buf[pi++] = GET_TCGV_I64(retl);
+            nb_rets = 2;
+        } else {
+            s->gen_opparam_buf[pi++] = ret;
+            nb_rets = 1;
+        }
+#else
+        if (TCG_TARGET_REG_BITS < 64 && (sizemask & 1)) {
+#ifdef HOST_WORDS_BIGENDIAN
+            s->gen_opparam_buf[pi++] = ret + 1;
+            s->gen_opparam_buf[pi++] = ret;
+#else
+            s->gen_opparam_buf[pi++] = ret;
+            s->gen_opparam_buf[pi++] = ret + 1;
+#endif
+            nb_rets = 2;
+        } else {
+            s->gen_opparam_buf[pi++] = ret;
+            nb_rets = 1;
+        }
+#endif
+    } else {
+        nb_rets = 0;
+    }
+    real_args = 0;
+    for (i = 0; i < nargs; i++) {
+        int is_64bit = sizemask & (1 << (i+1)*2);
+        if (TCG_TARGET_REG_BITS < 64 && is_64bit) {
+#ifdef TCG_TARGET_CALL_ALIGN_ARGS
+            /* some targets want aligned 64 bit args */
+            if (real_args & 1) {
+                s->gen_opparam_buf[pi++] = TCG_CALL_DUMMY_ARG;
+                real_args++;
+            }
+#endif
+           /* If stack grows up, then we will be placing successive
+              arguments at lower addresses, which means we need to
+              reverse the order compared to how we would normally
+              treat either big or little-endian.  For those arguments
+              that will wind up in registers, this still works for
+              HPPA (the only current STACK_GROWSUP target) since the
+              argument registers are *also* allocated in decreasing
+              order.  If another such target is added, this logic may
+              have to get more complicated to differentiate between
+              stack arguments and register arguments.  */
+#if defined(HOST_WORDS_BIGENDIAN) != defined(TCG_TARGET_STACK_GROWSUP)
+            s->gen_opparam_buf[pi++] = args[i] + 1;
+            s->gen_opparam_buf[pi++] = args[i];
+#else
+            s->gen_opparam_buf[pi++] = args[i];
+            s->gen_opparam_buf[pi++] = args[i] + 1;
+#endif
+            real_args += 2;
+            continue;
+        }
+
+        s->gen_opparam_buf[pi++] = args[i];
+        real_args++;
+    }
+    s->gen_opparam_buf[pi++] = (uintptr_t)func;
+    s->gen_opparam_buf[pi++] = flags;
+
+    i = s->gen_next_op_idx;
+    tcg_debug_assert(i < OPC_BUF_SIZE);
+    tcg_debug_assert(pi <= OPPARAM_BUF_SIZE);
+
+    /* Set links for sequential allocation during translation.  */
+    s->gen_op_buf[i] = (TCGOp){
+        .opc = INDEX_op_call,
+        .callo = nb_rets,
+        .calli = real_args,
+        .args = pi_first,
+        .prev = i - 1,
+        .next = i + 1
+    };
+
+    /* Make sure the calli field didn't overflow.  */
+    tcg_debug_assert(s->gen_op_buf[i].calli == real_args);
+
+    s->gen_last_op_idx = i;
+    s->gen_next_op_idx = i + 1;
+    s->gen_next_parm_idx = pi;
+
+#if defined(__sparc__) && !defined(__arch64__) \
+    && !defined(CONFIG_TCG_INTERPRETER)
+    /* Free all of the parts we allocated above.  */
+    for (i = real_args = 0; i < orig_nargs; ++i) {
+        int is_64bit = orig_sizemask & (1 << (i+1)*2);
+        if (is_64bit) {
+            TCGv_i32 h = MAKE_TCGV_I32(args[real_args++]);
+            TCGv_i32 l = MAKE_TCGV_I32(args[real_args++]);
+            tcg_temp_free_i32(h);
+            tcg_temp_free_i32(l);
+        } else {
+            real_args++;
+        }
+    }
+    if (orig_sizemask & 1) {
+        /* The 32-bit ABI returned two 32-bit pieces.  Re-assemble them.
+           Note that describing these as TCGv_i64 eliminates an unnecessary
+           zero-extension that tcg_gen_concat_i32_i64 would create.  */
+        tcg_gen_concat32_i64(MAKE_TCGV_I64(ret), retl, reth);
+        tcg_temp_free_i64(retl);
+        tcg_temp_free_i64(reth);
+    }
+#elif defined(TCG_TARGET_EXTEND_ARGS) && TCG_TARGET_REG_BITS == 64
+    for (i = 0; i < nargs; ++i) {
+        int is_64bit = sizemask & (1 << (i+1)*2);
+        if (!is_64bit) {
+            TCGv_i64 temp = MAKE_TCGV_I64(args[i]);
+            tcg_temp_free_i64(temp);
+        }
+    }
+#endif /* TCG_TARGET_EXTEND_ARGS */
+}
+
+static void tcg_reg_alloc_start(TCGContext *s)
+{
+    int i;
+    TCGTemp *ts;
+    for(i = 0; i < s->nb_globals; i++) {
+        ts = &s->temps[i];
+        if (ts->fixed_reg) {
+            ts->val_type = TEMP_VAL_REG;
+        } else {
+            ts->val_type = TEMP_VAL_MEM;
+        }
+    }
+    for(i = s->nb_globals; i < s->nb_temps; i++) {
+        ts = &s->temps[i];
+        if (ts->temp_local) {
+            ts->val_type = TEMP_VAL_MEM;
+        } else {
+            ts->val_type = TEMP_VAL_DEAD;
+        }
+        ts->mem_allocated = 0;
+        ts->fixed_reg = 0;
+    }
+    for(i = 0; i < TCG_TARGET_NB_REGS; i++) {
+        s->reg_to_temp[i] = -1;
+    }
+}
+
+static char *tcg_get_arg_str_idx(TCGContext *s, char *buf, int buf_size,
+                                 int idx)
+{
+    TCGTemp *ts;
+
+    assert(idx >= 0 && idx < s->nb_temps);
+    ts = &s->temps[idx];
+    if (idx < s->nb_globals) {
+        pstrcpy(buf, buf_size, ts->name);
+    } else {
+        if (ts->temp_local) 
+            snprintf(buf, buf_size, "loc%d", idx - s->nb_globals);
+        else
+            snprintf(buf, buf_size, "tmp%d", idx - s->nb_globals);
+    }
+    return buf;
+}
+
+char *tcg_get_arg_str_i32(TCGContext *s, char *buf, int buf_size, TCGv_i32 arg)
+{
+    return tcg_get_arg_str_idx(s, buf, buf_size, GET_TCGV_I32(arg));
+}
+
+char *tcg_get_arg_str_i64(TCGContext *s, char *buf, int buf_size, TCGv_i64 arg)
+{
+    return tcg_get_arg_str_idx(s, buf, buf_size, GET_TCGV_I64(arg));
+}
+
+/* Find helper name.  */
+static inline const char *tcg_find_helper(TCGContext *s, uintptr_t val)
+{
+    const char *ret = NULL;
+    if (s->helpers) {
+        TCGHelperInfo *info = g_hash_table_lookup(s->helpers, (gpointer)val);
+        if (info) {
+            ret = info->name;
+        }
+    }
+    return ret;
+}
+
+static const char * const cond_name[] =
+{
+    [TCG_COND_NEVER] = "never",
+    [TCG_COND_ALWAYS] = "always",
+    [TCG_COND_EQ] = "eq",
+    [TCG_COND_NE] = "ne",
+    [TCG_COND_LT] = "lt",
+    [TCG_COND_GE] = "ge",
+    [TCG_COND_LE] = "le",
+    [TCG_COND_GT] = "gt",
+    [TCG_COND_LTU] = "ltu",
+    [TCG_COND_GEU] = "geu",
+    [TCG_COND_LEU] = "leu",
+    [TCG_COND_GTU] = "gtu"
+};
+
+static const char * const ldst_name[] =
+{
+    [MO_UB]   = "ub",
+    [MO_SB]   = "sb",
+    [MO_LEUW] = "leuw",
+    [MO_LESW] = "lesw",
+    [MO_LEUL] = "leul",
+    [MO_LESL] = "lesl",
+    [MO_LEQ]  = "leq",
+    [MO_BEUW] = "beuw",
+    [MO_BESW] = "besw",
+    [MO_BEUL] = "beul",
+    [MO_BESL] = "besl",
+    [MO_BEQ]  = "beq",
+};
+
+void tcg_dump_ops(TCGContext *s)
+{
+    char buf[128];
+    TCGOp *op;
+    int oi;
+
+    for (oi = s->gen_first_op_idx; oi >= 0; oi = op->next) {
+        int i, k, nb_oargs, nb_iargs, nb_cargs;
+        const TCGOpDef *def;
+        const TCGArg *args;
+        TCGOpcode c;
+
+        op = &s->gen_op_buf[oi];
+        c = op->opc;
+        def = &tcg_op_defs[c];
+        args = &s->gen_opparam_buf[op->args];
+
+        if (c == INDEX_op_debug_insn_start) {
+            uint64_t pc;
+#if TARGET_LONG_BITS > TCG_TARGET_REG_BITS
+            pc = ((uint64_t)args[1] << 32) | args[0];
+#else
+            pc = args[0];
+#endif
+            if (oi != s->gen_first_op_idx) {
+                qemu_log("\n");
+            }
+            qemu_log(" ---- 0x%" PRIx64, pc);
+        } else if (c == INDEX_op_call) {
+            /* variable number of arguments */
+            nb_oargs = op->callo;
+            nb_iargs = op->calli;
+            nb_cargs = def->nb_cargs;
+
+            /* function name, flags, out args */
+            qemu_log(" %s %s,$0x%" TCG_PRIlx ",$%d", def->name,
+                     tcg_find_helper(s, args[nb_oargs + nb_iargs]),
+                     args[nb_oargs + nb_iargs + 1], nb_oargs);
+            for (i = 0; i < nb_oargs; i++) {
+                qemu_log(",%s", tcg_get_arg_str_idx(s, buf, sizeof(buf),
+                                                   args[i]));
+            }
+            for (i = 0; i < nb_iargs; i++) {
+                TCGArg arg = args[nb_oargs + i];
+                const char *t = "<dummy>";
+                if (arg != TCG_CALL_DUMMY_ARG) {
+                    t = tcg_get_arg_str_idx(s, buf, sizeof(buf), arg);
+                }
+                qemu_log(",%s", t);
+            }
+        } else {
+            qemu_log(" %s ", def->name);
+
+            nb_oargs = def->nb_oargs;
+            nb_iargs = def->nb_iargs;
+            nb_cargs = def->nb_cargs;
+
+            k = 0;
+            for (i = 0; i < nb_oargs; i++) {
+                if (k != 0) {
+                    qemu_log(",");
+                }
+                qemu_log("%s", tcg_get_arg_str_idx(s, buf, sizeof(buf),
+                                                   args[k++]));
+            }
+            for (i = 0; i < nb_iargs; i++) {
+                if (k != 0) {
+                    qemu_log(",");
+                }
+                qemu_log("%s", tcg_get_arg_str_idx(s, buf, sizeof(buf),
+                                                   args[k++]));
+            }
+            switch (c) {
+            case INDEX_op_brcond_i32:
+            case INDEX_op_setcond_i32:
+            case INDEX_op_movcond_i32:
+            case INDEX_op_brcond2_i32:
+            case INDEX_op_setcond2_i32:
+            case INDEX_op_brcond_i64:
+            case INDEX_op_setcond_i64:
+            case INDEX_op_movcond_i64:
+                if (args[k] < ARRAY_SIZE(cond_name) && cond_name[args[k]]) {
+                    qemu_log(",%s", cond_name[args[k++]]);
+                } else {
+                    qemu_log(",$0x%" TCG_PRIlx, args[k++]);
+                }
+                i = 1;
+                break;
+            case INDEX_op_qemu_ld_i32:
+            case INDEX_op_qemu_st_i32:
+            case INDEX_op_qemu_ld_i64:
+            case INDEX_op_qemu_st_i64:
+                {
+                    TCGMemOpIdx oi = args[k++];
+                    TCGMemOp op = get_memop(oi);
+                    unsigned ix = get_mmuidx(oi);
+
+                    if (op & ~(MO_AMASK | MO_BSWAP | MO_SSIZE)) {
+                        qemu_log(",$0x%x,%u", op, ix);
+                    } else {
+                        const char *s_al = "", *s_op;
+                        if (op & MO_AMASK) {
+                            if ((op & MO_AMASK) == MO_ALIGN) {
+                                s_al = "al+";
+                            } else {
+                                s_al = "un+";
+                            }
+                        }
+                        s_op = ldst_name[op & (MO_BSWAP | MO_SSIZE)];
+                        qemu_log(",%s%s,%u", s_al, s_op, ix);
+                    }
+                    i = 1;
+                }
+                break;
+            default:
+                i = 0;
+                break;
+            }
+            switch (c) {
+            case INDEX_op_set_label:
+            case INDEX_op_br:
+            case INDEX_op_brcond_i32:
+            case INDEX_op_brcond_i64:
+            case INDEX_op_brcond2_i32:
+                qemu_log("%s$L%d", k ? "," : "", arg_label(args[k])->id);
+                i++, k++;
+                break;
+            default:
+                break;
+            }
+            for (; i < nb_cargs; i++, k++) {
+                qemu_log("%s$0x%" TCG_PRIlx, k ? "," : "", args[k]);
+            }
+        }
+        qemu_log("\n");
+    }
+}
+
+/* we give more priority to constraints with less registers */
+static int get_constraint_priority(const TCGOpDef *def, int k)
+{
+    const TCGArgConstraint *arg_ct;
+
+    int i, n;
+    arg_ct = &def->args_ct[k];
+    if (arg_ct->ct & TCG_CT_ALIAS) {
+        /* an alias is equivalent to a single register */
+        n = 1;
+    } else {
+        if (!(arg_ct->ct & TCG_CT_REG))
+            return 0;
+        n = 0;
+        for(i = 0; i < TCG_TARGET_NB_REGS; i++) {
+            if (tcg_regset_test_reg(arg_ct->u.regs, i))
+                n++;
+        }
+    }
+    return TCG_TARGET_NB_REGS - n + 1;
+}
+
+/* sort from highest priority to lowest */
+static void sort_constraints(TCGOpDef *def, int start, int n)
+{
+    int i, j, p1, p2, tmp;
+
+    for(i = 0; i < n; i++)
+        def->sorted_args[start + i] = start + i;
+    if (n <= 1)
+        return;
+    for(i = 0; i < n - 1; i++) {
+        for(j = i + 1; j < n; j++) {
+            p1 = get_constraint_priority(def, def->sorted_args[start + i]);
+            p2 = get_constraint_priority(def, def->sorted_args[start + j]);
+            if (p1 < p2) {
+                tmp = def->sorted_args[start + i];
+                def->sorted_args[start + i] = def->sorted_args[start + j];
+                def->sorted_args[start + j] = tmp;
+            }
+        }
+    }
+}
+
+void tcg_add_target_add_op_defs(const TCGTargetOpDef *tdefs)
+{
+    TCGOpcode op;
+    TCGOpDef *def;
+    const char *ct_str;
+    int i, nb_args;
+
+    for(;;) {
+        if (tdefs->op == (TCGOpcode)-1)
+            break;
+        op = tdefs->op;
+        assert((unsigned)op < NB_OPS);
+        def = &tcg_op_defs[op];
+#if defined(CONFIG_DEBUG_TCG)
+        /* Duplicate entry in op definitions? */
+        assert(!def->used);
+        def->used = 1;
+#endif
+        nb_args = def->nb_iargs + def->nb_oargs;
+        for(i = 0; i < nb_args; i++) {
+            ct_str = tdefs->args_ct_str[i];
+            /* Incomplete TCGTargetOpDef entry? */
+            assert(ct_str != NULL);
+            tcg_regset_clear(def->args_ct[i].u.regs);
+            def->args_ct[i].ct = 0;
+            if (ct_str[0] >= '0' && ct_str[0] <= '9') {
+                int oarg;
+                oarg = ct_str[0] - '0';
+                assert(oarg < def->nb_oargs);
+                assert(def->args_ct[oarg].ct & TCG_CT_REG);
+                /* TCG_CT_ALIAS is for the output arguments. The input
+                   argument is tagged with TCG_CT_IALIAS. */
+                def->args_ct[i] = def->args_ct[oarg];
+                def->args_ct[oarg].ct = TCG_CT_ALIAS;
+                def->args_ct[oarg].alias_index = i;
+                def->args_ct[i].ct |= TCG_CT_IALIAS;
+                def->args_ct[i].alias_index = oarg;
+            } else {
+                for(;;) {
+                    if (*ct_str == '\0')
+                        break;
+                    switch(*ct_str) {
+                    case 'i':
+                        def->args_ct[i].ct |= TCG_CT_CONST;
+                        ct_str++;
+                        break;
+                    default:
+                        if (target_parse_constraint(&def->args_ct[i], &ct_str) < 0) {
+                            fprintf(stderr, "Invalid constraint '%s' for arg %d of operation '%s'\n",
+                                    ct_str, i, def->name);
+                            exit(1);
+                        }
+                    }
+                }
+            }
+        }
+
+        /* TCGTargetOpDef entry with too much information? */
+        assert(i == TCG_MAX_OP_ARGS || tdefs->args_ct_str[i] == NULL);
+
+        /* sort the constraints (XXX: this is just an heuristic) */
+        sort_constraints(def, 0, def->nb_oargs);
+        sort_constraints(def, def->nb_oargs, def->nb_iargs);
+
+#if 0
+        {
+            int i;
+
+            printf("%s: sorted=", def->name);
+            for(i = 0; i < def->nb_oargs + def->nb_iargs; i++)
+                printf(" %d", def->sorted_args[i]);
+            printf("\n");
+        }
+#endif
+        tdefs++;
+    }
+
+#if defined(CONFIG_DEBUG_TCG)
+    i = 0;
+    for (op = 0; op < ARRAY_SIZE(tcg_op_defs); op++) {
+        const TCGOpDef *def = &tcg_op_defs[op];
+        if (def->flags & TCG_OPF_NOT_PRESENT) {
+            /* Wrong entry in op definitions? */
+            if (def->used) {
+                fprintf(stderr, "Invalid op definition for %s\n", def->name);
+                i = 1;
+            }
+        } else {
+            /* Missing entry in op definitions? */
+            if (!def->used) {
+                fprintf(stderr, "Missing op definition for %s\n", def->name);
+                i = 1;
+            }
+        }
+    }
+    if (i == 1) {
+        tcg_abort();
+    }
+#endif
+}
+
+void tcg_op_remove(TCGContext *s, TCGOp *op)
+{
+    int next = op->next;
+    int prev = op->prev;
+
+    if (next >= 0) {
+        s->gen_op_buf[next].prev = prev;
+    } else {
+        s->gen_last_op_idx = prev;
+    }
+    if (prev >= 0) {
+        s->gen_op_buf[prev].next = next;
+    } else {
+        s->gen_first_op_idx = next;
+    }
+
+    memset(op, -1, sizeof(*op));
+
+#ifdef CONFIG_PROFILER
+    s->del_op_count++;
+#endif
+}
+
+#ifdef USE_LIVENESS_ANALYSIS
+/* liveness analysis: end of function: all temps are dead, and globals
+   should be in memory. */
+static inline void tcg_la_func_end(TCGContext *s, uint8_t *dead_temps,
+                                   uint8_t *mem_temps)
+{
+    memset(dead_temps, 1, s->nb_temps);
+    memset(mem_temps, 1, s->nb_globals);
+    memset(mem_temps + s->nb_globals, 0, s->nb_temps - s->nb_globals);
+}
+
+/* liveness analysis: end of basic block: all temps are dead, globals
+   and local temps should be in memory. */
+static inline void tcg_la_bb_end(TCGContext *s, uint8_t *dead_temps,
+                                 uint8_t *mem_temps)
+{
+    int i;
+
+    memset(dead_temps, 1, s->nb_temps);
+    memset(mem_temps, 1, s->nb_globals);
+    for(i = s->nb_globals; i < s->nb_temps; i++) {
+        mem_temps[i] = s->temps[i].temp_local;
+    }
+}
+
+/* Liveness analysis : update the opc_dead_args array to tell if a
+   given input arguments is dead. Instructions updating dead
+   temporaries are removed. */
+static void tcg_liveness_analysis(TCGContext *s)
+{
+    uint8_t *dead_temps, *mem_temps;
+    int oi, oi_prev, nb_ops;
+
+    nb_ops = s->gen_next_op_idx;
+    s->op_dead_args = tcg_malloc(nb_ops * sizeof(uint16_t));
+    s->op_sync_args = tcg_malloc(nb_ops * sizeof(uint8_t));
+    
+    dead_temps = tcg_malloc(s->nb_temps);
+    mem_temps = tcg_malloc(s->nb_temps);
+    tcg_la_func_end(s, dead_temps, mem_temps);
+
+    for (oi = s->gen_last_op_idx; oi >= 0; oi = oi_prev) {
+        int i, nb_iargs, nb_oargs;
+        TCGOpcode opc_new, opc_new2;
+        bool have_opc_new2;
+        uint16_t dead_args;
+        uint8_t sync_args;
+        TCGArg arg;
+
+        TCGOp * const op = &s->gen_op_buf[oi];
+        TCGArg * const args = &s->gen_opparam_buf[op->args];
+        TCGOpcode opc = op->opc;
+        const TCGOpDef *def = &tcg_op_defs[opc];
+
+        oi_prev = op->prev;
+
+        switch (opc) {
+        case INDEX_op_call:
+            {
+                int call_flags;
+
+                nb_oargs = op->callo;
+                nb_iargs = op->calli;
+                call_flags = args[nb_oargs + nb_iargs + 1];
+
+                /* pure functions can be removed if their result is unused */
+                if (call_flags & TCG_CALL_NO_SIDE_EFFECTS) {
+                    for (i = 0; i < nb_oargs; i++) {
+                        arg = args[i];
+                        if (!dead_temps[arg] || mem_temps[arg]) {
+                            goto do_not_remove_call;
+                        }
+                    }
+                    goto do_remove;
+                } else {
+                do_not_remove_call:
+
+                    /* output args are dead */
+                    dead_args = 0;
+                    sync_args = 0;
+                    for (i = 0; i < nb_oargs; i++) {
+                        arg = args[i];
+                        if (dead_temps[arg]) {
+                            dead_args |= (1 << i);
+                        }
+                        if (mem_temps[arg]) {
+                            sync_args |= (1 << i);
+                        }
+                        dead_temps[arg] = 1;
+                        mem_temps[arg] = 0;
+                    }
+
+                    if (!(call_flags & TCG_CALL_NO_READ_GLOBALS)) {
+                        /* globals should be synced to memory */
+                        memset(mem_temps, 1, s->nb_globals);
+                    }
+                    if (!(call_flags & (TCG_CALL_NO_WRITE_GLOBALS |
+                                        TCG_CALL_NO_READ_GLOBALS))) {
+                        /* globals should go back to memory */
+                        memset(dead_temps, 1, s->nb_globals);
+                    }
+
+                    /* record arguments that die in this helper */
+                    for (i = nb_oargs; i < nb_iargs + nb_oargs; i++) {
+                        arg = args[i];
+                        if (arg != TCG_CALL_DUMMY_ARG) {
+                            if (dead_temps[arg]) {
+                                dead_args |= (1 << i);
+                            }
+                        }
+                    }
+                    /* input arguments are live for preceeding opcodes */
+                    for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) {
+                        arg = args[i];
+                        dead_temps[arg] = 0;
+                    }
+                    s->op_dead_args[oi] = dead_args;
+                    s->op_sync_args[oi] = sync_args;
+                }
+            }
+            break;
+        case INDEX_op_debug_insn_start:
+            break;
+        case INDEX_op_discard:
+            /* mark the temporary as dead */
+            dead_temps[args[0]] = 1;
+            mem_temps[args[0]] = 0;
+            break;
+
+        case INDEX_op_add2_i32:
+            opc_new = INDEX_op_add_i32;
+            goto do_addsub2;
+        case INDEX_op_sub2_i32:
+            opc_new = INDEX_op_sub_i32;
+            goto do_addsub2;
+        case INDEX_op_add2_i64:
+            opc_new = INDEX_op_add_i64;
+            goto do_addsub2;
+        case INDEX_op_sub2_i64:
+            opc_new = INDEX_op_sub_i64;
+        do_addsub2:
+            nb_iargs = 4;
+            nb_oargs = 2;
+            /* Test if the high part of the operation is dead, but not
+               the low part.  The result can be optimized to a simple
+               add or sub.  This happens often for x86_64 guest when the
+               cpu mode is set to 32 bit.  */
+            if (dead_temps[args[1]] && !mem_temps[args[1]]) {
+                if (dead_temps[args[0]] && !mem_temps[args[0]]) {
+                    goto do_remove;
+                }
+                /* Replace the opcode and adjust the args in place,
+                   leaving 3 unused args at the end.  */
+                op->opc = opc = opc_new;
+                args[1] = args[2];
+                args[2] = args[4];
+                /* Fall through and mark the single-word operation live.  */
+                nb_iargs = 2;
+                nb_oargs = 1;
+            }
+            goto do_not_remove;
+
+        case INDEX_op_mulu2_i32:
+            opc_new = INDEX_op_mul_i32;
+            opc_new2 = INDEX_op_muluh_i32;
+            have_opc_new2 = TCG_TARGET_HAS_muluh_i32;
+            goto do_mul2;
+        case INDEX_op_muls2_i32:
+            opc_new = INDEX_op_mul_i32;
+            opc_new2 = INDEX_op_mulsh_i32;
+            have_opc_new2 = TCG_TARGET_HAS_mulsh_i32;
+            goto do_mul2;
+        case INDEX_op_mulu2_i64:
+            opc_new = INDEX_op_mul_i64;
+            opc_new2 = INDEX_op_muluh_i64;
+            have_opc_new2 = TCG_TARGET_HAS_muluh_i64;
+            goto do_mul2;
+        case INDEX_op_muls2_i64:
+            opc_new = INDEX_op_mul_i64;
+            opc_new2 = INDEX_op_mulsh_i64;
+            have_opc_new2 = TCG_TARGET_HAS_mulsh_i64;
+            goto do_mul2;
+        do_mul2:
+            nb_iargs = 2;
+            nb_oargs = 2;
+            if (dead_temps[args[1]] && !mem_temps[args[1]]) {
+                if (dead_temps[args[0]] && !mem_temps[args[0]]) {
+                    /* Both parts of the operation are dead.  */
+                    goto do_remove;
+                }
+                /* The high part of the operation is dead; generate the low. */
+                op->opc = opc = opc_new;
+                args[1] = args[2];
+                args[2] = args[3];
+            } else if (have_opc_new2 && dead_temps[args[0]]
+                       && !mem_temps[args[0]]) {
+                /* The low part of the operation is dead; generate the high. */
+                op->opc = opc = opc_new2;
+                args[0] = args[1];
+                args[1] = args[2];
+                args[2] = args[3];
+            } else {
+                goto do_not_remove;
+            }
+            /* Mark the single-word operation live.  */
+            nb_oargs = 1;
+            goto do_not_remove;
+
+        default:
+            /* XXX: optimize by hardcoding common cases (e.g. triadic ops) */
+            nb_iargs = def->nb_iargs;
+            nb_oargs = def->nb_oargs;
+
+            /* Test if the operation can be removed because all
+               its outputs are dead. We assume that nb_oargs == 0
+               implies side effects */
+            if (!(def->flags & TCG_OPF_SIDE_EFFECTS) && nb_oargs != 0) {
+                for (i = 0; i < nb_oargs; i++) {
+                    arg = args[i];
+                    if (!dead_temps[arg] || mem_temps[arg]) {
+                        goto do_not_remove;
+                    }
+                }
+            do_remove:
+                tcg_op_remove(s, op);
+            } else {
+            do_not_remove:
+                /* output args are dead */
+                dead_args = 0;
+                sync_args = 0;
+                for (i = 0; i < nb_oargs; i++) {
+                    arg = args[i];
+                    if (dead_temps[arg]) {
+                        dead_args |= (1 << i);
+                    }
+                    if (mem_temps[arg]) {
+                        sync_args |= (1 << i);
+                    }
+                    dead_temps[arg] = 1;
+                    mem_temps[arg] = 0;
+                }
+
+                /* if end of basic block, update */
+                if (def->flags & TCG_OPF_BB_END) {
+                    tcg_la_bb_end(s, dead_temps, mem_temps);
+                } else if (def->flags & TCG_OPF_SIDE_EFFECTS) {
+                    /* globals should be synced to memory */
+                    memset(mem_temps, 1, s->nb_globals);
+                }
+
+                /* record arguments that die in this opcode */
+                for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) {
+                    arg = args[i];
+                    if (dead_temps[arg]) {
+                        dead_args |= (1 << i);
+                    }
+                }
+                /* input arguments are live for preceeding opcodes */
+                for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) {
+                    arg = args[i];
+                    dead_temps[arg] = 0;
+                }
+                s->op_dead_args[oi] = dead_args;
+                s->op_sync_args[oi] = sync_args;
+            }
+            break;
+        }
+    }
+}
+#else
+/* dummy liveness analysis */
+static void tcg_liveness_analysis(TCGContext *s)
+{
+    int nb_ops;
+    nb_ops = s->gen_opc_ptr - s->gen_opc_buf;
+
+    s->op_dead_args = tcg_malloc(nb_ops * sizeof(uint16_t));
+    memset(s->op_dead_args, 0, nb_ops * sizeof(uint16_t));
+    s->op_sync_args = tcg_malloc(nb_ops * sizeof(uint8_t));
+    memset(s->op_sync_args, 0, nb_ops * sizeof(uint8_t));
+}
+#endif
+
+#ifndef NDEBUG
+static void dump_regs(TCGContext *s)
+{
+    TCGTemp *ts;
+    int i;
+    char buf[64];
+
+    for(i = 0; i < s->nb_temps; i++) {
+        ts = &s->temps[i];
+        printf("  %10s: ", tcg_get_arg_str_idx(s, buf, sizeof(buf), i));
+        switch(ts->val_type) {
+        case TEMP_VAL_REG:
+            printf("%s", tcg_target_reg_names[ts->reg]);
+            break;
+        case TEMP_VAL_MEM:
+            printf("%d(%s)", (int)ts->mem_offset, tcg_target_reg_names[ts->mem_reg]);
+            break;
+        case TEMP_VAL_CONST:
+            printf("$0x%" TCG_PRIlx, ts->val);
+            break;
+        case TEMP_VAL_DEAD:
+            printf("D");
+            break;
+        default:
+            printf("???");
+            break;
+        }
+        printf("\n");
+    }
+
+    for(i = 0; i < TCG_TARGET_NB_REGS; i++) {
+        if (s->reg_to_temp[i] >= 0) {
+            printf("%s: %s\n", 
+                   tcg_target_reg_names[i], 
+                   tcg_get_arg_str_idx(s, buf, sizeof(buf), s->reg_to_temp[i]));
+        }
+    }
+}
+
+static void check_regs(TCGContext *s)
+{
+    int reg, k;
+    TCGTemp *ts;
+    char buf[64];
+
+    for(reg = 0; reg < TCG_TARGET_NB_REGS; reg++) {
+        k = s->reg_to_temp[reg];
+        if (k >= 0) {
+            ts = &s->temps[k];
+            if (ts->val_type != TEMP_VAL_REG ||
+                ts->reg != reg) {
+                printf("Inconsistency for register %s:\n", 
+                       tcg_target_reg_names[reg]);
+                goto fail;
+            }
+        }
+    }
+    for(k = 0; k < s->nb_temps; k++) {
+        ts = &s->temps[k];
+        if (ts->val_type == TEMP_VAL_REG &&
+            !ts->fixed_reg &&
+            s->reg_to_temp[ts->reg] != k) {
+                printf("Inconsistency for temp %s:\n", 
+                       tcg_get_arg_str_idx(s, buf, sizeof(buf), k));
+        fail:
+                printf("reg state:\n");
+                dump_regs(s);
+                tcg_abort();
+        }
+    }
+}
+#endif
+
+static void temp_allocate_frame(TCGContext *s, int temp)
+{
+    TCGTemp *ts;
+    ts = &s->temps[temp];
+#if !(defined(__sparc__) && TCG_TARGET_REG_BITS == 64)
+    /* Sparc64 stack is accessed with offset of 2047 */
+    s->current_frame_offset = (s->current_frame_offset +
+                               (tcg_target_long)sizeof(tcg_target_long) - 1) &
+        ~(sizeof(tcg_target_long) - 1);
+#endif
+    if (s->current_frame_offset + (tcg_target_long)sizeof(tcg_target_long) >
+        s->frame_end) {
+        tcg_abort();
+    }
+    ts->mem_offset = s->current_frame_offset;
+    ts->mem_reg = s->frame_reg;
+    ts->mem_allocated = 1;
+    s->current_frame_offset += sizeof(tcg_target_long);
+}
+
+/* sync register 'reg' by saving it to the corresponding temporary */
+static inline void tcg_reg_sync(TCGContext *s, int reg)
+{
+    TCGTemp *ts;
+    int temp;
+
+    temp = s->reg_to_temp[reg];
+    ts = &s->temps[temp];
+    assert(ts->val_type == TEMP_VAL_REG);
+    if (!ts->mem_coherent && !ts->fixed_reg) {
+        if (!ts->mem_allocated) {
+            temp_allocate_frame(s, temp);
+        }
+        tcg_out_st(s, ts->type, reg, ts->mem_reg, ts->mem_offset);
+    }
+    ts->mem_coherent = 1;
+}
+
+/* free register 'reg' by spilling the corresponding temporary if necessary */
+static void tcg_reg_free(TCGContext *s, int reg)
+{
+    int temp;
+
+    temp = s->reg_to_temp[reg];
+    if (temp != -1) {
+        tcg_reg_sync(s, reg);
+        s->temps[temp].val_type = TEMP_VAL_MEM;
+        s->reg_to_temp[reg] = -1;
+    }
+}
+
+/* Allocate a register belonging to reg1 & ~reg2 */
+static int tcg_reg_alloc(TCGContext *s, TCGRegSet reg1, TCGRegSet reg2)
+{
+    int i, reg;
+    TCGRegSet reg_ct;
+
+    tcg_regset_andnot(reg_ct, reg1, reg2);
+
+    /* first try free registers */
+    for(i = 0; i < ARRAY_SIZE(tcg_target_reg_alloc_order); i++) {
+        reg = tcg_target_reg_alloc_order[i];
+        if (tcg_regset_test_reg(reg_ct, reg) && s->reg_to_temp[reg] == -1)
+            return reg;
+    }
+
+    /* XXX: do better spill choice */
+    for(i = 0; i < ARRAY_SIZE(tcg_target_reg_alloc_order); i++) {
+        reg = tcg_target_reg_alloc_order[i];
+        if (tcg_regset_test_reg(reg_ct, reg)) {
+            tcg_reg_free(s, reg);
+            return reg;
+        }
+    }
+
+    tcg_abort();
+}
+
+/* mark a temporary as dead. */
+static inline void temp_dead(TCGContext *s, int temp)
+{
+    TCGTemp *ts;
+
+    ts = &s->temps[temp];
+    if (!ts->fixed_reg) {
+        if (ts->val_type == TEMP_VAL_REG) {
+            s->reg_to_temp[ts->reg] = -1;
+        }
+        if (temp < s->nb_globals || ts->temp_local) {
+            ts->val_type = TEMP_VAL_MEM;
+        } else {
+            ts->val_type = TEMP_VAL_DEAD;
+        }
+    }
+}
+
+/* sync a temporary to memory. 'allocated_regs' is used in case a
+   temporary registers needs to be allocated to store a constant. */
+static inline void temp_sync(TCGContext *s, int temp, TCGRegSet allocated_regs)
+{
+    TCGTemp *ts;
+
+    ts = &s->temps[temp];
+    if (!ts->fixed_reg) {
+        switch(ts->val_type) {
+        case TEMP_VAL_CONST:
+            ts->reg = tcg_reg_alloc(s, tcg_target_available_regs[ts->type],
+                                    allocated_regs);
+            ts->val_type = TEMP_VAL_REG;
+            s->reg_to_temp[ts->reg] = temp;
+            ts->mem_coherent = 0;
+            tcg_out_movi(s, ts->type, ts->reg, ts->val);
+            /* fallthrough*/
+        case TEMP_VAL_REG:
+            tcg_reg_sync(s, ts->reg);
+            break;
+        case TEMP_VAL_DEAD:
+        case TEMP_VAL_MEM:
+            break;
+        default:
+            tcg_abort();
+        }
+    }
+}
+
+/* save a temporary to memory. 'allocated_regs' is used in case a
+   temporary registers needs to be allocated to store a constant. */
+static inline void temp_save(TCGContext *s, int temp, TCGRegSet allocated_regs)
+{
+#ifdef USE_LIVENESS_ANALYSIS
+    /* The liveness analysis already ensures that globals are back
+       in memory. Keep an assert for safety. */
+    assert(s->temps[temp].val_type == TEMP_VAL_MEM || s->temps[temp].fixed_reg);
+#else
+    temp_sync(s, temp, allocated_regs);
+    temp_dead(s, temp);
+#endif
+}
+
+/* save globals to their canonical location and assume they can be
+   modified be the following code. 'allocated_regs' is used in case a
+   temporary registers needs to be allocated to store a constant. */
+static void save_globals(TCGContext *s, TCGRegSet allocated_regs)
+{
+    int i;
+
+    for(i = 0; i < s->nb_globals; i++) {
+        temp_save(s, i, allocated_regs);
+    }
+}
+
+/* sync globals to their canonical location and assume they can be
+   read by the following code. 'allocated_regs' is used in case a
+   temporary registers needs to be allocated to store a constant. */
+static void sync_globals(TCGContext *s, TCGRegSet allocated_regs)
+{
+    int i;
+
+    for (i = 0; i < s->nb_globals; i++) {
+#ifdef USE_LIVENESS_ANALYSIS
+        assert(s->temps[i].val_type != TEMP_VAL_REG || s->temps[i].fixed_reg ||
+               s->temps[i].mem_coherent);
+#else
+        temp_sync(s, i, allocated_regs);
+#endif
+    }
+}
+
+/* at the end of a basic block, we assume all temporaries are dead and
+   all globals are stored at their canonical location. */
+static void tcg_reg_alloc_bb_end(TCGContext *s, TCGRegSet allocated_regs)
+{
+    TCGTemp *ts;
+    int i;
+
+    for(i = s->nb_globals; i < s->nb_temps; i++) {
+        ts = &s->temps[i];
+        if (ts->temp_local) {
+            temp_save(s, i, allocated_regs);
+        } else {
+#ifdef USE_LIVENESS_ANALYSIS
+            /* The liveness analysis already ensures that temps are dead.
+               Keep an assert for safety. */
+            assert(ts->val_type == TEMP_VAL_DEAD);
+#else
+            temp_dead(s, i);
+#endif
+        }
+    }
+
+    save_globals(s, allocated_regs);
+}
+
+#define IS_DEAD_ARG(n) ((dead_args >> (n)) & 1)
+#define NEED_SYNC_ARG(n) ((sync_args >> (n)) & 1)
+
+static void tcg_reg_alloc_movi(TCGContext *s, const TCGArg *args,
+                               uint16_t dead_args, uint8_t sync_args)
+{
+    TCGTemp *ots;
+    tcg_target_ulong val;
+
+    ots = &s->temps[args[0]];
+    val = args[1];
+
+    if (ots->fixed_reg) {
+        /* for fixed registers, we do not do any constant
+           propagation */
+        tcg_out_movi(s, ots->type, ots->reg, val);
+    } else {
+        /* The movi is not explicitly generated here */
+        if (ots->val_type == TEMP_VAL_REG)
+            s->reg_to_temp[ots->reg] = -1;
+        ots->val_type = TEMP_VAL_CONST;
+        ots->val = val;
+    }
+    if (NEED_SYNC_ARG(0)) {
+        temp_sync(s, args[0], s->reserved_regs);
+    }
+    if (IS_DEAD_ARG(0)) {
+        temp_dead(s, args[0]);
+    }
+}
+
+static void tcg_reg_alloc_mov(TCGContext *s, const TCGOpDef *def,
+                              const TCGArg *args, uint16_t dead_args,
+                              uint8_t sync_args)
+{
+    TCGRegSet allocated_regs;
+    TCGTemp *ts, *ots;
+    TCGType otype, itype;
+
+    tcg_regset_set(allocated_regs, s->reserved_regs);
+    ots = &s->temps[args[0]];
+    ts = &s->temps[args[1]];
+
+    /* Note that otype != itype for no-op truncation.  */
+    otype = ots->type;
+    itype = ts->type;
+
+    /* If the source value is not in a register, and we're going to be
+       forced to have it in a register in order to perform the copy,
+       then copy the SOURCE value into its own register first.  That way
+       we don't have to reload SOURCE the next time it is used. */
+    if (((NEED_SYNC_ARG(0) || ots->fixed_reg) && ts->val_type != TEMP_VAL_REG)
+        || ts->val_type == TEMP_VAL_MEM) {
+        ts->reg = tcg_reg_alloc(s, tcg_target_available_regs[itype],
+                                allocated_regs);
+        if (ts->val_type == TEMP_VAL_MEM) {
+            tcg_out_ld(s, itype, ts->reg, ts->mem_reg, ts->mem_offset);
+            ts->mem_coherent = 1;
+        } else if (ts->val_type == TEMP_VAL_CONST) {
+            tcg_out_movi(s, itype, ts->reg, ts->val);
+            ts->mem_coherent = 0;
+        }
+        s->reg_to_temp[ts->reg] = args[1];
+        ts->val_type = TEMP_VAL_REG;
+    }
+
+    if (IS_DEAD_ARG(0) && !ots->fixed_reg) {
+        /* mov to a non-saved dead register makes no sense (even with
+           liveness analysis disabled). */
+        assert(NEED_SYNC_ARG(0));
+        /* The code above should have moved the temp to a register. */
+        assert(ts->val_type == TEMP_VAL_REG);
+        if (!ots->mem_allocated) {
+            temp_allocate_frame(s, args[0]);
+        }
+        tcg_out_st(s, otype, ts->reg, ots->mem_reg, ots->mem_offset);
+        if (IS_DEAD_ARG(1)) {
+            temp_dead(s, args[1]);
+        }
+        temp_dead(s, args[0]);
+    } else if (ts->val_type == TEMP_VAL_CONST) {
+        /* propagate constant */
+        if (ots->val_type == TEMP_VAL_REG) {
+            s->reg_to_temp[ots->reg] = -1;
+        }
+        ots->val_type = TEMP_VAL_CONST;
+        ots->val = ts->val;
+        if (IS_DEAD_ARG(1)) {
+            temp_dead(s, args[1]);
+        }
+    } else {
+        /* The code in the first if block should have moved the
+           temp to a register. */
+        assert(ts->val_type == TEMP_VAL_REG);
+        if (IS_DEAD_ARG(1) && !ts->fixed_reg && !ots->fixed_reg) {
+            /* the mov can be suppressed */
+            if (ots->val_type == TEMP_VAL_REG) {
+                s->reg_to_temp[ots->reg] = -1;
+            }
+            ots->reg = ts->reg;
+            temp_dead(s, args[1]);
+        } else {
+            if (ots->val_type != TEMP_VAL_REG) {
+                /* When allocating a new register, make sure to not spill the
+                   input one. */
+                tcg_regset_set_reg(allocated_regs, ts->reg);
+                ots->reg = tcg_reg_alloc(s, tcg_target_available_regs[otype],
+                                         allocated_regs);
+            }
+            tcg_out_mov(s, otype, ots->reg, ts->reg);
+        }
+        ots->val_type = TEMP_VAL_REG;
+        ots->mem_coherent = 0;
+        s->reg_to_temp[ots->reg] = args[0];
+        if (NEED_SYNC_ARG(0)) {
+            tcg_reg_sync(s, ots->reg);
+        }
+    }
+}
+
+static void tcg_reg_alloc_op(TCGContext *s, 
+                             const TCGOpDef *def, TCGOpcode opc,
+                             const TCGArg *args, uint16_t dead_args,
+                             uint8_t sync_args)
+{
+    TCGRegSet allocated_regs;
+    int i, k, nb_iargs, nb_oargs, reg;
+    TCGArg arg;
+    const TCGArgConstraint *arg_ct;
+    TCGTemp *ts;
+    TCGArg new_args[TCG_MAX_OP_ARGS];
+    int const_args[TCG_MAX_OP_ARGS];
+
+    nb_oargs = def->nb_oargs;
+    nb_iargs = def->nb_iargs;
+
+    /* copy constants */
+    memcpy(new_args + nb_oargs + nb_iargs, 
+           args + nb_oargs + nb_iargs, 
+           sizeof(TCGArg) * def->nb_cargs);
+
+    /* satisfy input constraints */ 
+    tcg_regset_set(allocated_regs, s->reserved_regs);
+    for(k = 0; k < nb_iargs; k++) {
+        i = def->sorted_args[nb_oargs + k];
+        arg = args[i];
+        arg_ct = &def->args_ct[i];
+        ts = &s->temps[arg];
+        if (ts->val_type == TEMP_VAL_MEM) {
+            reg = tcg_reg_alloc(s, arg_ct->u.regs, allocated_regs);
+            tcg_out_ld(s, ts->type, reg, ts->mem_reg, ts->mem_offset);
+            ts->val_type = TEMP_VAL_REG;
+            ts->reg = reg;
+            ts->mem_coherent = 1;
+            s->reg_to_temp[reg] = arg;
+        } else if (ts->val_type == TEMP_VAL_CONST) {
+            if (tcg_target_const_match(ts->val, ts->type, arg_ct)) {
+                /* constant is OK for instruction */
+                const_args[i] = 1;
+                new_args[i] = ts->val;
+                goto iarg_end;
+            } else {
+                /* need to move to a register */
+                reg = tcg_reg_alloc(s, arg_ct->u.regs, allocated_regs);
+                tcg_out_movi(s, ts->type, reg, ts->val);
+                ts->val_type = TEMP_VAL_REG;
+                ts->reg = reg;
+                ts->mem_coherent = 0;
+                s->reg_to_temp[reg] = arg;
+            }
+        }
+        assert(ts->val_type == TEMP_VAL_REG);
+        if (arg_ct->ct & TCG_CT_IALIAS) {
+            if (ts->fixed_reg) {
+                /* if fixed register, we must allocate a new register
+                   if the alias is not the same register */
+                if (arg != args[arg_ct->alias_index])
+                    goto allocate_in_reg;
+            } else {
+                /* if the input is aliased to an output and if it is
+                   not dead after the instruction, we must allocate
+                   a new register and move it */
+                if (!IS_DEAD_ARG(i)) {
+                    goto allocate_in_reg;
+                }
+                /* check if the current register has already been allocated
+                   for another input aliased to an output */
+                int k2, i2;
+                for (k2 = 0 ; k2 < k ; k2++) {
+                    i2 = def->sorted_args[nb_oargs + k2];
+                    if ((def->args_ct[i2].ct & TCG_CT_IALIAS) &&
+                        (new_args[i2] == ts->reg)) {
+                        goto allocate_in_reg;
+                    }
+                }
+            }
+        }
+        reg = ts->reg;
+        if (tcg_regset_test_reg(arg_ct->u.regs, reg)) {
+            /* nothing to do : the constraint is satisfied */
+        } else {
+        allocate_in_reg:
+            /* allocate a new register matching the constraint 
+               and move the temporary register into it */
+            reg = tcg_reg_alloc(s, arg_ct->u.regs, allocated_regs);
+            tcg_out_mov(s, ts->type, reg, ts->reg);
+        }
+        new_args[i] = reg;
+        const_args[i] = 0;
+        tcg_regset_set_reg(allocated_regs, reg);
+    iarg_end: ;
+    }
+    
+    /* mark dead temporaries and free the associated registers */
+    for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) {
+        if (IS_DEAD_ARG(i)) {
+            temp_dead(s, args[i]);
+        }
+    }
+
+    if (def->flags & TCG_OPF_BB_END) {
+        tcg_reg_alloc_bb_end(s, allocated_regs);
+    } else {
+        if (def->flags & TCG_OPF_CALL_CLOBBER) {
+            /* XXX: permit generic clobber register list ? */ 
+            for(reg = 0; reg < TCG_TARGET_NB_REGS; reg++) {
+                if (tcg_regset_test_reg(tcg_target_call_clobber_regs, reg)) {
+                    tcg_reg_free(s, reg);
+                }
+            }
+        }
+        if (def->flags & TCG_OPF_SIDE_EFFECTS) {
+            /* sync globals if the op has side effects and might trigger
+               an exception. */
+            sync_globals(s, allocated_regs);
+        }
+        
+        /* satisfy the output constraints */
+        tcg_regset_set(allocated_regs, s->reserved_regs);
+        for(k = 0; k < nb_oargs; k++) {
+            i = def->sorted_args[k];
+            arg = args[i];
+            arg_ct = &def->args_ct[i];
+            ts = &s->temps[arg];
+            if (arg_ct->ct & TCG_CT_ALIAS) {
+                reg = new_args[arg_ct->alias_index];
+            } else {
+                /* if fixed register, we try to use it */
+                reg = ts->reg;
+                if (ts->fixed_reg &&
+                    tcg_regset_test_reg(arg_ct->u.regs, reg)) {
+                    goto oarg_end;
+                }
+                reg = tcg_reg_alloc(s, arg_ct->u.regs, allocated_regs);
+            }
+            tcg_regset_set_reg(allocated_regs, reg);
+            /* if a fixed register is used, then a move will be done afterwards */
+            if (!ts->fixed_reg) {
+                if (ts->val_type == TEMP_VAL_REG) {
+                    s->reg_to_temp[ts->reg] = -1;
+                }
+                ts->val_type = TEMP_VAL_REG;
+                ts->reg = reg;
+                /* temp value is modified, so the value kept in memory is
+                   potentially not the same */
+                ts->mem_coherent = 0;
+                s->reg_to_temp[reg] = arg;
+            }
+        oarg_end:
+            new_args[i] = reg;
+        }
+    }
+
+    /* emit instruction */
+    tcg_out_op(s, opc, new_args, const_args);
+    
+    /* move the outputs in the correct register if needed */
+    for(i = 0; i < nb_oargs; i++) {
+        ts = &s->temps[args[i]];
+        reg = new_args[i];
+        if (ts->fixed_reg && ts->reg != reg) {
+            tcg_out_mov(s, ts->type, ts->reg, reg);
+        }
+        if (NEED_SYNC_ARG(i)) {
+            tcg_reg_sync(s, reg);
+        }
+        if (IS_DEAD_ARG(i)) {
+            temp_dead(s, args[i]);
+        }
+    }
+}
+
+#ifdef TCG_TARGET_STACK_GROWSUP
+#define STACK_DIR(x) (-(x))
+#else
+#define STACK_DIR(x) (x)
+#endif
+
+static void tcg_reg_alloc_call(TCGContext *s, int nb_oargs, int nb_iargs,
+                               const TCGArg * const args, uint16_t dead_args,
+                               uint8_t sync_args)
+{
+    int flags, nb_regs, i, reg;
+    TCGArg arg;
+    TCGTemp *ts;
+    intptr_t stack_offset;
+    size_t call_stack_size;
+    tcg_insn_unit *func_addr;
+    int allocate_args;
+    TCGRegSet allocated_regs;
+
+    func_addr = (tcg_insn_unit *)(intptr_t)args[nb_oargs + nb_iargs];
+    flags = args[nb_oargs + nb_iargs + 1];
+
+    nb_regs = ARRAY_SIZE(tcg_target_call_iarg_regs);
+    if (nb_regs > nb_iargs) {
+        nb_regs = nb_iargs;
+    }
+
+    /* assign stack slots first */
+    call_stack_size = (nb_iargs - nb_regs) * sizeof(tcg_target_long);
+    call_stack_size = (call_stack_size + TCG_TARGET_STACK_ALIGN - 1) & 
+        ~(TCG_TARGET_STACK_ALIGN - 1);
+    allocate_args = (call_stack_size > TCG_STATIC_CALL_ARGS_SIZE);
+    if (allocate_args) {
+        /* XXX: if more than TCG_STATIC_CALL_ARGS_SIZE is needed,
+           preallocate call stack */
+        tcg_abort();
+    }
+
+    stack_offset = TCG_TARGET_CALL_STACK_OFFSET;
+    for(i = nb_regs; i < nb_iargs; i++) {
+        arg = args[nb_oargs + i];
+#ifdef TCG_TARGET_STACK_GROWSUP
+        stack_offset -= sizeof(tcg_target_long);
+#endif
+        if (arg != TCG_CALL_DUMMY_ARG) {
+            ts = &s->temps[arg];
+            if (ts->val_type == TEMP_VAL_REG) {
+                tcg_out_st(s, ts->type, ts->reg, TCG_REG_CALL_STACK, stack_offset);
+            } else if (ts->val_type == TEMP_VAL_MEM) {
+                reg = tcg_reg_alloc(s, tcg_target_available_regs[ts->type], 
+                                    s->reserved_regs);
+                /* XXX: not correct if reading values from the stack */
+                tcg_out_ld(s, ts->type, reg, ts->mem_reg, ts->mem_offset);
+                tcg_out_st(s, ts->type, reg, TCG_REG_CALL_STACK, stack_offset);
+            } else if (ts->val_type == TEMP_VAL_CONST) {
+                reg = tcg_reg_alloc(s, tcg_target_available_regs[ts->type], 
+                                    s->reserved_regs);
+                /* XXX: sign extend may be needed on some targets */
+                tcg_out_movi(s, ts->type, reg, ts->val);
+                tcg_out_st(s, ts->type, reg, TCG_REG_CALL_STACK, stack_offset);
+            } else {
+                tcg_abort();
+            }
+        }
+#ifndef TCG_TARGET_STACK_GROWSUP
+        stack_offset += sizeof(tcg_target_long);
+#endif
+    }
+    
+    /* assign input registers */
+    tcg_regset_set(allocated_regs, s->reserved_regs);
+    for(i = 0; i < nb_regs; i++) {
+        arg = args[nb_oargs + i];
+        if (arg != TCG_CALL_DUMMY_ARG) {
+            ts = &s->temps[arg];
+            reg = tcg_target_call_iarg_regs[i];
+            tcg_reg_free(s, reg);
+            if (ts->val_type == TEMP_VAL_REG) {
+                if (ts->reg != reg) {
+                    tcg_out_mov(s, ts->type, reg, ts->reg);
+                }
+            } else if (ts->val_type == TEMP_VAL_MEM) {
+                tcg_out_ld(s, ts->type, reg, ts->mem_reg, ts->mem_offset);
+            } else if (ts->val_type == TEMP_VAL_CONST) {
+                /* XXX: sign extend ? */
+                tcg_out_movi(s, ts->type, reg, ts->val);
+            } else {
+                tcg_abort();
+            }
+            tcg_regset_set_reg(allocated_regs, reg);
+        }
+    }
+    
+    /* mark dead temporaries and free the associated registers */
+    for(i = nb_oargs; i < nb_iargs + nb_oargs; i++) {
+        if (IS_DEAD_ARG(i)) {
+            temp_dead(s, args[i]);
+        }
+    }
+    
+    /* clobber call registers */
+    for(reg = 0; reg < TCG_TARGET_NB_REGS; reg++) {
+        if (tcg_regset_test_reg(tcg_target_call_clobber_regs, reg)) {
+            tcg_reg_free(s, reg);
+        }
+    }
+
+    /* Save globals if they might be written by the helper, sync them if
+       they might be read. */
+    if (flags & TCG_CALL_NO_READ_GLOBALS) {
+        /* Nothing to do */
+    } else if (flags & TCG_CALL_NO_WRITE_GLOBALS) {
+        sync_globals(s, allocated_regs);
+    } else {
+        save_globals(s, allocated_regs);
+    }
+
+    tcg_out_call(s, func_addr);
+
+    /* assign output registers and emit moves if needed */
+    for(i = 0; i < nb_oargs; i++) {
+        arg = args[i];
+        ts = &s->temps[arg];
+        reg = tcg_target_call_oarg_regs[i];
+        assert(s->reg_to_temp[reg] == -1);
+
+        if (ts->fixed_reg) {
+            if (ts->reg != reg) {
+                tcg_out_mov(s, ts->type, ts->reg, reg);
+            }
+        } else {
+            if (ts->val_type == TEMP_VAL_REG) {
+                s->reg_to_temp[ts->reg] = -1;
+            }
+            ts->val_type = TEMP_VAL_REG;
+            ts->reg = reg;
+            ts->mem_coherent = 0;
+            s->reg_to_temp[reg] = arg;
+            if (NEED_SYNC_ARG(i)) {
+                tcg_reg_sync(s, reg);
+            }
+            if (IS_DEAD_ARG(i)) {
+                temp_dead(s, args[i]);
+            }
+        }
+    }
+}
+
+#ifdef CONFIG_PROFILER
+
+static int64_t tcg_table_op_count[NB_OPS];
+
+void tcg_dump_op_count(FILE *f, fprintf_function cpu_fprintf)
+{
+    int i;
+
+    for (i = 0; i < NB_OPS; i++) {
+        cpu_fprintf(f, "%s %" PRId64 "\n", tcg_op_defs[i].name,
+                    tcg_table_op_count[i]);
+    }
+}
+#else
+void tcg_dump_op_count(FILE *f, fprintf_function cpu_fprintf)
+{
+    cpu_fprintf(f, "[TCG profiler not compiled]\n");
+}
+#endif
+
+
+static inline int tcg_gen_code_common(TCGContext *s,
+                                      tcg_insn_unit *gen_code_buf,
+                                      long search_pc)
+{
+    int oi, oi_next;
+
+#ifdef DEBUG_DISAS
+    if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP))) {
+        qemu_log("OP:\n");
+        tcg_dump_ops(s);
+        qemu_log("\n");
+    }
+#endif
+
+#ifdef CONFIG_PROFILER
+    s->opt_time -= profile_getclock();
+#endif
+
+#ifdef USE_TCG_OPTIMIZATIONS
+    tcg_optimize(s);
+#endif
+
+#ifdef CONFIG_PROFILER
+    s->opt_time += profile_getclock();
+    s->la_time -= profile_getclock();
+#endif
+
+    tcg_liveness_analysis(s);
+
+#ifdef CONFIG_PROFILER
+    s->la_time += profile_getclock();
+#endif
+
+#ifdef DEBUG_DISAS
+    if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP_OPT))) {
+        qemu_log("OP after optimization and liveness analysis:\n");
+        tcg_dump_ops(s);
+        qemu_log("\n");
+    }
+#endif
+
+    tcg_reg_alloc_start(s);
+
+    s->code_buf = gen_code_buf;
+    s->code_ptr = gen_code_buf;
+
+    tcg_out_tb_init(s);
+
+    for (oi = s->gen_first_op_idx; oi >= 0; oi = oi_next) {
+        TCGOp * const op = &s->gen_op_buf[oi];
+        TCGArg * const args = &s->gen_opparam_buf[op->args];
+        TCGOpcode opc = op->opc;
+        const TCGOpDef *def = &tcg_op_defs[opc];
+        uint16_t dead_args = s->op_dead_args[oi];
+        uint8_t sync_args = s->op_sync_args[oi];
+
+        oi_next = op->next;
+#ifdef CONFIG_PROFILER
+        tcg_table_op_count[opc]++;
+#endif
+
+        switch (opc) {
+        case INDEX_op_mov_i32:
+        case INDEX_op_mov_i64:
+            tcg_reg_alloc_mov(s, def, args, dead_args, sync_args);
+            break;
+        case INDEX_op_movi_i32:
+        case INDEX_op_movi_i64:
+            tcg_reg_alloc_movi(s, args, dead_args, sync_args);
+            break;
+        case INDEX_op_debug_insn_start:
+            break;
+        case INDEX_op_discard:
+            temp_dead(s, args[0]);
+            break;
+        case INDEX_op_set_label:
+            tcg_reg_alloc_bb_end(s, s->reserved_regs);
+            tcg_out_label(s, arg_label(args[0]), s->code_ptr);
+            break;
+        case INDEX_op_call:
+            tcg_reg_alloc_call(s, op->callo, op->calli, args,
+                               dead_args, sync_args);
+            break;
+        default:
+            /* Sanity check that we've not introduced any unhandled opcodes. */
+            if (def->flags & TCG_OPF_NOT_PRESENT) {
+                tcg_abort();
+            }
+            /* Note: in order to speed up the code, it would be much
+               faster to have specialized register allocator functions for
+               some common argument patterns */
+            tcg_reg_alloc_op(s, def, opc, args, dead_args, sync_args);
+            break;
+        }
+        if (search_pc >= 0 && search_pc < tcg_current_code_size(s)) {
+            return oi;
+        }
+#ifndef NDEBUG
+        check_regs(s);
+#endif
+    }
+
+    /* Generate TB finalization at the end of block */
+    tcg_out_tb_finalize(s);
+    return -1;
+}
+
+int tcg_gen_code(TCGContext *s, tcg_insn_unit *gen_code_buf)
+{
+#ifdef CONFIG_PROFILER
+    {
+        int n;
+
+        n = s->gen_last_op_idx + 1;
+        s->op_count += n;
+        if (n > s->op_count_max) {
+            s->op_count_max = n;
+        }
+
+        n = s->nb_temps;
+        s->temp_count += n;
+        if (n > s->temp_count_max) {
+            s->temp_count_max = n;
+        }
+    }
+#endif
+
+    tcg_gen_code_common(s, gen_code_buf, -1);
+
+    /* flush instruction cache */
+    flush_icache_range((uintptr_t)s->code_buf, (uintptr_t)s->code_ptr);
+
+    return tcg_current_code_size(s);
+}
+
+/* Return the index of the micro operation such as the pc after is <
+   offset bytes from the start of the TB.  The contents of gen_code_buf must
+   not be changed, though writing the same values is ok.
+   Return -1 if not found. */
+int tcg_gen_code_search_pc(TCGContext *s, tcg_insn_unit *gen_code_buf,
+                           long offset)
+{
+    return tcg_gen_code_common(s, gen_code_buf, offset);
+}
+
+#ifdef CONFIG_PROFILER
+void tcg_dump_info(FILE *f, fprintf_function cpu_fprintf)
+{
+    TCGContext *s = &tcg_ctx;
+    int64_t tot;
+
+    tot = s->interm_time + s->code_time;
+    cpu_fprintf(f, "JIT cycles          %" PRId64 " (%0.3f s at 2.4 GHz)\n",
+                tot, tot / 2.4e9);
+    cpu_fprintf(f, "translated TBs      %" PRId64 " (aborted=%" PRId64 " %0.1f%%)\n", 
+                s->tb_count, 
+                s->tb_count1 - s->tb_count,
+                s->tb_count1 ? (double)(s->tb_count1 - s->tb_count) / s->tb_count1 * 100.0 : 0);
+    cpu_fprintf(f, "avg ops/TB          %0.1f max=%d\n", 
+                s->tb_count ? (double)s->op_count / s->tb_count : 0, s->op_count_max);
+    cpu_fprintf(f, "deleted ops/TB      %0.2f\n",
+                s->tb_count ? 
+                (double)s->del_op_count / s->tb_count : 0);
+    cpu_fprintf(f, "avg temps/TB        %0.2f max=%d\n",
+                s->tb_count ? 
+                (double)s->temp_count / s->tb_count : 0,
+                s->temp_count_max);
+    
+    cpu_fprintf(f, "cycles/op           %0.1f\n", 
+                s->op_count ? (double)tot / s->op_count : 0);
+    cpu_fprintf(f, "cycles/in byte      %0.1f\n", 
+                s->code_in_len ? (double)tot / s->code_in_len : 0);
+    cpu_fprintf(f, "cycles/out byte     %0.1f\n", 
+                s->code_out_len ? (double)tot / s->code_out_len : 0);
+    if (tot == 0)
+        tot = 1;
+    cpu_fprintf(f, "  gen_interm time   %0.1f%%\n", 
+                (double)s->interm_time / tot * 100.0);
+    cpu_fprintf(f, "  gen_code time     %0.1f%%\n", 
+                (double)s->code_time / tot * 100.0);
+    cpu_fprintf(f, "optim./code time    %0.1f%%\n",
+                (double)s->opt_time / (s->code_time ? s->code_time : 1)
+                * 100.0);
+    cpu_fprintf(f, "liveness/code time  %0.1f%%\n", 
+                (double)s->la_time / (s->code_time ? s->code_time : 1) * 100.0);
+    cpu_fprintf(f, "cpu_restore count   %" PRId64 "\n",
+                s->restore_count);
+    cpu_fprintf(f, "  avg cycles        %0.1f\n",
+                s->restore_count ? (double)s->restore_time / s->restore_count : 0);
+}
+#else
+void tcg_dump_info(FILE *f, fprintf_function cpu_fprintf)
+{
+    cpu_fprintf(f, "[TCG profiler not compiled]\n");
+}
+#endif
+
+#ifdef ELF_HOST_MACHINE
+/* In order to use this feature, the backend needs to do three things:
+
+   (1) Define ELF_HOST_MACHINE to indicate both what value to
+       put into the ELF image and to indicate support for the feature.
+
+   (2) Define tcg_register_jit.  This should create a buffer containing
+       the contents of a .debug_frame section that describes the post-
+       prologue unwind info for the tcg machine.
+
+   (3) Call tcg_register_jit_int, with the constructed .debug_frame.
+*/
+
+/* Begin GDB interface.  THE FOLLOWING MUST MATCH GDB DOCS.  */
+typedef enum {
+    JIT_NOACTION = 0,
+    JIT_REGISTER_FN,
+    JIT_UNREGISTER_FN
+} jit_actions_t;
+
+struct jit_code_entry {
+    struct jit_code_entry *next_entry;
+    struct jit_code_entry *prev_entry;
+    const void *symfile_addr;
+    uint64_t symfile_size;
+};
+
+struct jit_descriptor {
+    uint32_t version;
+    uint32_t action_flag;
+    struct jit_code_entry *relevant_entry;
+    struct jit_code_entry *first_entry;
+};
+
+void __jit_debug_register_code(void) __attribute__((noinline));
+void __jit_debug_register_code(void)
+{
+    asm("");
+}
+
+/* Must statically initialize the version, because GDB may check
+   the version before we can set it.  */
+struct jit_descriptor __jit_debug_descriptor = { 1, 0, 0, 0 };
+
+/* End GDB interface.  */
+
+static int find_string(const char *strtab, const char *str)
+{
+    const char *p = strtab + 1;
+
+    while (1) {
+        if (strcmp(p, str) == 0) {
+            return p - strtab;
+        }
+        p += strlen(p) + 1;
+    }
+}
+
+static void tcg_register_jit_int(void *buf_ptr, size_t buf_size,
+                                 const void *debug_frame,
+                                 size_t debug_frame_size)
+{
+    struct __attribute__((packed)) DebugInfo {
+        uint32_t  len;
+        uint16_t  version;
+        uint32_t  abbrev;
+        uint8_t   ptr_size;
+        uint8_t   cu_die;
+        uint16_t  cu_lang;
+        uintptr_t cu_low_pc;
+        uintptr_t cu_high_pc;
+        uint8_t   fn_die;
+        char      fn_name[16];
+        uintptr_t fn_low_pc;
+        uintptr_t fn_high_pc;
+        uint8_t   cu_eoc;
+    };
+
+    struct ElfImage {
+        ElfW(Ehdr) ehdr;
+        ElfW(Phdr) phdr;
+        ElfW(Shdr) shdr[7];
+        ElfW(Sym)  sym[2];
+        struct DebugInfo di;
+        uint8_t    da[24];
+        char       str[80];
+    };
+
+    struct ElfImage *img;
+
+    static const struct ElfImage img_template = {
+        .ehdr = {
+            .e_ident[EI_MAG0] = ELFMAG0,
+            .e_ident[EI_MAG1] = ELFMAG1,
+            .e_ident[EI_MAG2] = ELFMAG2,
+            .e_ident[EI_MAG3] = ELFMAG3,
+            .e_ident[EI_CLASS] = ELF_CLASS,
+            .e_ident[EI_DATA] = ELF_DATA,
+            .e_ident[EI_VERSION] = EV_CURRENT,
+            .e_type = ET_EXEC,
+            .e_machine = ELF_HOST_MACHINE,
+            .e_version = EV_CURRENT,
+            .e_phoff = offsetof(struct ElfImage, phdr),
+            .e_shoff = offsetof(struct ElfImage, shdr),
+            .e_ehsize = sizeof(ElfW(Shdr)),
+            .e_phentsize = sizeof(ElfW(Phdr)),
+            .e_phnum = 1,
+            .e_shentsize = sizeof(ElfW(Shdr)),
+            .e_shnum = ARRAY_SIZE(img->shdr),
+            .e_shstrndx = ARRAY_SIZE(img->shdr) - 1,
+#ifdef ELF_HOST_FLAGS
+            .e_flags = ELF_HOST_FLAGS,
+#endif
+#ifdef ELF_OSABI
+            .e_ident[EI_OSABI] = ELF_OSABI,
+#endif
+        },
+        .phdr = {
+            .p_type = PT_LOAD,
+            .p_flags = PF_X,
+        },
+        .shdr = {
+            [0] = { .sh_type = SHT_NULL },
+            /* Trick: The contents of code_gen_buffer are not present in
+               this fake ELF file; that got allocated elsewhere.  Therefore
+               we mark .text as SHT_NOBITS (similar to .bss) so that readers
+               will not look for contents.  We can record any address.  */
+            [1] = { /* .text */
+                .sh_type = SHT_NOBITS,
+                .sh_flags = SHF_EXECINSTR | SHF_ALLOC,
+            },
+            [2] = { /* .debug_info */
+                .sh_type = SHT_PROGBITS,
+                .sh_offset = offsetof(struct ElfImage, di),
+                .sh_size = sizeof(struct DebugInfo),
+            },
+            [3] = { /* .debug_abbrev */
+                .sh_type = SHT_PROGBITS,
+                .sh_offset = offsetof(struct ElfImage, da),
+                .sh_size = sizeof(img->da),
+            },
+            [4] = { /* .debug_frame */
+                .sh_type = SHT_PROGBITS,
+                .sh_offset = sizeof(struct ElfImage),
+            },
+            [5] = { /* .symtab */
+                .sh_type = SHT_SYMTAB,
+                .sh_offset = offsetof(struct ElfImage, sym),
+                .sh_size = sizeof(img->sym),
+                .sh_info = 1,
+                .sh_link = ARRAY_SIZE(img->shdr) - 1,
+                .sh_entsize = sizeof(ElfW(Sym)),
+            },
+            [6] = { /* .strtab */
+                .sh_type = SHT_STRTAB,
+                .sh_offset = offsetof(struct ElfImage, str),
+                .sh_size = sizeof(img->str),
+            }
+        },
+        .sym = {
+            [1] = { /* code_gen_buffer */
+                .st_info = ELF_ST_INFO(STB_GLOBAL, STT_FUNC),
+                .st_shndx = 1,
+            }
+        },
+        .di = {
+            .len = sizeof(struct DebugInfo) - 4,
+            .version = 2,
+            .ptr_size = sizeof(void *),
+            .cu_die = 1,
+            .cu_lang = 0x8001,  /* DW_LANG_Mips_Assembler */
+            .fn_die = 2,
+            .fn_name = "code_gen_buffer"
+        },
+        .da = {
+            1,          /* abbrev number (the cu) */
+            0x11, 1,    /* DW_TAG_compile_unit, has children */
+            0x13, 0x5,  /* DW_AT_language, DW_FORM_data2 */
+            0x11, 0x1,  /* DW_AT_low_pc, DW_FORM_addr */
+            0x12, 0x1,  /* DW_AT_high_pc, DW_FORM_addr */
+            0, 0,       /* end of abbrev */
+            2,          /* abbrev number (the fn) */
+            0x2e, 0,    /* DW_TAG_subprogram, no children */
+            0x3, 0x8,   /* DW_AT_name, DW_FORM_string */
+            0x11, 0x1,  /* DW_AT_low_pc, DW_FORM_addr */
+            0x12, 0x1,  /* DW_AT_high_pc, DW_FORM_addr */
+            0, 0,       /* end of abbrev */
+            0           /* no more abbrev */
+        },
+        .str = "\0" ".text\0" ".debug_info\0" ".debug_abbrev\0"
+               ".debug_frame\0" ".symtab\0" ".strtab\0" "code_gen_buffer",
+    };
+
+    /* We only need a single jit entry; statically allocate it.  */
+    static struct jit_code_entry one_entry;
+
+    uintptr_t buf = (uintptr_t)buf_ptr;
+    size_t img_size = sizeof(struct ElfImage) + debug_frame_size;
+    DebugFrameHeader *dfh;
+
+    img = g_malloc(img_size);
+    *img = img_template;
+
+    img->phdr.p_vaddr = buf;
+    img->phdr.p_paddr = buf;
+    img->phdr.p_memsz = buf_size;
+
+    img->shdr[1].sh_name = find_string(img->str, ".text");
+    img->shdr[1].sh_addr = buf;
+    img->shdr[1].sh_size = buf_size;
+
+    img->shdr[2].sh_name = find_string(img->str, ".debug_info");
+    img->shdr[3].sh_name = find_string(img->str, ".debug_abbrev");
+
+    img->shdr[4].sh_name = find_string(img->str, ".debug_frame");
+    img->shdr[4].sh_size = debug_frame_size;
+
+    img->shdr[5].sh_name = find_string(img->str, ".symtab");
+    img->shdr[6].sh_name = find_string(img->str, ".strtab");
+
+    img->sym[1].st_name = find_string(img->str, "code_gen_buffer");
+    img->sym[1].st_value = buf;
+    img->sym[1].st_size = buf_size;
+
+    img->di.cu_low_pc = buf;
+    img->di.cu_high_pc = buf + buf_size;
+    img->di.fn_low_pc = buf;
+    img->di.fn_high_pc = buf + buf_size;
+
+    dfh = (DebugFrameHeader *)(img + 1);
+    memcpy(dfh, debug_frame, debug_frame_size);
+    dfh->fde.func_start = buf;
+    dfh->fde.func_len = buf_size;
+
+#ifdef DEBUG_JIT
+    /* Enable this block to be able to debug the ELF image file creation.
+       One can use readelf, objdump, or other inspection utilities.  */
+    {
+        FILE *f = fopen("/tmp/qemu.jit", "w+b");
+        if (f) {
+            if (fwrite(img, img_size, 1, f) != img_size) {
+                /* Avoid stupid unused return value warning for fwrite.  */
+            }
+            fclose(f);
+        }
+    }
+#endif
+
+    one_entry.symfile_addr = img;
+    one_entry.symfile_size = img_size;
+
+    __jit_debug_descriptor.action_flag = JIT_REGISTER_FN;
+    __jit_debug_descriptor.relevant_entry = &one_entry;
+    __jit_debug_descriptor.first_entry = &one_entry;
+    __jit_debug_register_code();
+}
+#else
+/* No support for the feature.  Provide the entry point expected by exec.c,
+   and implement the internal function we declared earlier.  */
+
+static void tcg_register_jit_int(void *buf, size_t size,
+                                 const void *debug_frame,
+                                 size_t debug_frame_size)
+{
+}
+
+void tcg_register_jit(void *buf, size_t buf_size)
+{
+}
+#endif /* ELF_HOST_MACHINE */