* THE SOFTWARE.
*/
-/* define it to suppress various consistency checks (faster) */
-#define NDEBUG
-
/* define it to use liveness analysis (better code) */
#define USE_LIVENESS_ANALYSIS
-#include <assert.h>
+#include "config.h"
+
+#ifndef DEBUG_TCG
+/* define it to suppress various consistency checks (faster) */
+#define NDEBUG
+#endif
+
#include <stdarg.h>
#include <stdlib.h>
#include <stdio.h>
#ifdef _WIN32
#include <malloc.h>
#endif
+#ifdef _AIX
+#include <alloca.h>
+#endif
-#include "config.h"
#include "qemu-common.h"
+#include "cache-utils.h"
/* Note: the long term plan is to reduce the dependancies on the QEMU
CPU definitions. Currently they are used for qemu_ld/st
static void patch_reloc(uint8_t *code_ptr, int type,
tcg_target_long value, tcg_target_long addend);
-TCGOpDef tcg_op_defs[] = {
+static TCGOpDef tcg_op_defs[] = {
#define DEF(s, n, copy_size) { #s, 0, 0, n, n, 0, copy_size },
#define DEF2(s, iargs, oargs, cargs, flags) { #s, iargs, oargs, cargs, iargs + oargs + cargs, flags, 0 },
#include "tcg-opc.h"
#undef DEF2
};
-TCGRegSet tcg_target_available_regs[2];
-TCGRegSet tcg_target_call_clobber_regs;
+static TCGRegSet tcg_target_available_regs[2];
+static TCGRegSet tcg_target_call_clobber_regs;
/* XXX: move that inside the context */
uint16_t *gen_opc_ptr;
s->pool_current = NULL;
}
-/* free all the pool */
-void tcg_pool_free(TCGContext *s)
-{
- TCGPool *p, *p1;
-
- for(p = s->pool_first; p != NULL; p = p1) {
- p1 = p->next;
- qemu_free(p);
- }
- s->pool_first = NULL;
- s->pool_cur = s->pool_end = NULL;
-}
-
void tcg_context_init(TCGContext *s)
{
int op, total_args, n;
s->frame_reg = reg;
}
-void tcg_set_macro_func(TCGContext *s, TCGMacroFunc *func)
-{
- s->macro_func = func;
-}
-
void tcg_func_start(TCGContext *s)
{
int i;
tcg_pool_reset(s);
s->nb_temps = s->nb_globals;
- for(i = 0; i < TCG_TYPE_COUNT; i++)
+ for(i = 0; i < (TCG_TYPE_COUNT * 2); i++)
s->first_free_temp[i] = -1;
s->labels = tcg_malloc(sizeof(TCGLabel) * TCG_MAX_LABELS);
s->nb_labels = 0;
tcg_abort();
}
-TCGv tcg_global_reg_new(TCGType type, int reg, const char *name)
+static inline int tcg_global_reg_new_internal(TCGType type, int reg,
+ const char *name)
{
TCGContext *s = &tcg_ctx;
TCGTemp *ts;
ts->name = name;
s->nb_globals++;
tcg_regset_set_reg(s->reserved_regs, reg);
- return MAKE_TCGV(idx);
+ return idx;
}
-#if TCG_TARGET_REG_BITS == 32
-/* temporary hack to avoid register shortage for tcg_qemu_st64() */
-TCGv tcg_global_reg2_new_hack(TCGType type, int reg1, int reg2,
- const char *name)
+TCGv_i32 tcg_global_reg_new_i32(int reg, const char *name)
{
- TCGContext *s = &tcg_ctx;
- TCGTemp *ts;
int idx;
- char buf[64];
- if (type != TCG_TYPE_I64)
- tcg_abort();
- idx = s->nb_globals;
- 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 = 1;
- ts->reg = reg1;
- pstrcpy(buf, sizeof(buf), name);
- pstrcat(buf, sizeof(buf), "_0");
- ts->name = strdup(buf);
+ idx = tcg_global_reg_new_internal(TCG_TYPE_I32, reg, name);
+ return MAKE_TCGV_I32(idx);
+}
- ts++;
- ts->base_type = type;
- ts->type = TCG_TYPE_I32;
- ts->fixed_reg = 1;
- ts->reg = reg2;
- pstrcpy(buf, sizeof(buf), name);
- pstrcat(buf, sizeof(buf), "_1");
- ts->name = strdup(buf);
+TCGv_i64 tcg_global_reg_new_i64(int reg, const char *name)
+{
+ int idx;
- s->nb_globals += 2;
- return MAKE_TCGV(idx);
+ idx = tcg_global_reg_new_internal(TCG_TYPE_I64, reg, name);
+ return MAKE_TCGV_I64(idx);
}
-#endif
-TCGv tcg_global_mem_new(TCGType type, int reg, tcg_target_long offset,
- const char *name)
+static inline int tcg_global_mem_new_internal(TCGType type, int reg,
+ tcg_target_long offset,
+ const char *name)
{
TCGContext *s = &tcg_ctx;
TCGTemp *ts;
#if TCG_TARGET_REG_BITS == 32
if (type == TCG_TYPE_I64) {
char buf[64];
- tcg_temp_alloc(s, s->nb_globals + 1);
+ tcg_temp_alloc(s, s->nb_globals + 2);
ts = &s->temps[s->nb_globals];
ts->base_type = type;
ts->type = TCG_TYPE_I32;
ts->name = name;
s->nb_globals++;
}
- return MAKE_TCGV(idx);
+ return idx;
+}
+
+TCGv_i32 tcg_global_mem_new_i32(int reg, tcg_target_long offset,
+ const char *name)
+{
+ int idx;
+
+ 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, tcg_target_long offset,
+ const char *name)
+{
+ int idx;
+
+ idx = tcg_global_mem_new_internal(TCG_TYPE_I64, reg, offset, name);
+ return MAKE_TCGV_I64(idx);
}
-TCGv tcg_temp_new(TCGType type)
+static inline int tcg_temp_new_internal(TCGType type, int temp_local)
{
TCGContext *s = &tcg_ctx;
TCGTemp *ts;
- int idx;
+ int idx, k;
- idx = s->first_free_temp[type];
+ k = type;
+ if (temp_local)
+ k += TCG_TYPE_COUNT;
+ idx = s->first_free_temp[k];
if (idx != -1) {
/* There is already an available temp with the
right type */
ts = &s->temps[idx];
- s->first_free_temp[type] = ts->next_free_temp;
+ s->first_free_temp[k] = ts->next_free_temp;
ts->temp_allocated = 1;
+ 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 + 1);
+ 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 = TCG_TYPE_I32;
ts->type = TCG_TYPE_I32;
ts->temp_allocated = 1;
+ ts->temp_local = temp_local;
ts->name = NULL;
s->nb_temps += 2;
} else
ts->base_type = type;
ts->type = type;
ts->temp_allocated = 1;
+ ts->temp_local = temp_local;
ts->name = NULL;
s->nb_temps++;
}
}
- return MAKE_TCGV(idx);
+ 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);
}
-void tcg_temp_free(TCGv arg)
+static inline void tcg_temp_free_internal(int idx)
{
TCGContext *s = &tcg_ctx;
TCGTemp *ts;
- int idx = GET_TCGV(arg);
- TCGType type;
+ int k;
assert(idx >= s->nb_globals && idx < s->nb_temps);
ts = &s->temps[idx];
assert(ts->temp_allocated != 0);
ts->temp_allocated = 0;
- type = ts->base_type;
- ts->next_free_temp = s->first_free_temp[type];
- s->first_free_temp[type] = idx;
+ k = ts->base_type;
+ if (ts->temp_local)
+ k += TCG_TYPE_COUNT;
+ ts->next_free_temp = s->first_free_temp[k];
+ s->first_free_temp[k] = idx;
}
+void tcg_temp_free_i32(TCGv_i32 arg)
+{
+ tcg_temp_free_internal(GET_TCGV_I32(arg));
+}
-TCGv tcg_const_i32(int32_t val)
+void tcg_temp_free_i64(TCGv_i64 arg)
{
- TCGv t0;
- t0 = tcg_temp_new(TCG_TYPE_I32);
+ 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 tcg_const_i64(int64_t val)
+TCGv_i64 tcg_const_local_i64(int64_t val)
{
- TCGv t0;
- t0 = tcg_temp_new(TCG_TYPE_I64);
+ TCGv_i64 t0;
+ t0 = tcg_temp_local_new_i64();
tcg_gen_movi_i64(t0, val);
return t0;
}
s->nb_helpers++;
}
-static inline TCGType tcg_get_base_type(TCGContext *s, TCGv arg)
-{
- return s->temps[GET_TCGV(arg)].base_type;
-}
-
-static void tcg_gen_call_internal(TCGContext *s, TCGv func,
- unsigned int flags,
- unsigned int nb_rets, const TCGv *rets,
- unsigned int nb_params, const TCGv *params)
-{
- int i;
- *gen_opc_ptr++ = INDEX_op_call;
- *gen_opparam_ptr++ = (nb_rets << 16) | (nb_params + 1);
- for(i = 0; i < nb_rets; i++) {
- *gen_opparam_ptr++ = GET_TCGV(rets[i]);
- }
- for(i = 0; i < nb_params; i++) {
- *gen_opparam_ptr++ = GET_TCGV(params[i]);
- }
- *gen_opparam_ptr++ = GET_TCGV(func);
-
- *gen_opparam_ptr++ = flags;
- /* total parameters, needed to go backward in the instruction stream */
- *gen_opparam_ptr++ = 1 + nb_rets + nb_params + 3;
-}
-
-
-#if TCG_TARGET_REG_BITS < 64
/* 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_call(TCGContext *s, TCGv func, unsigned int flags,
- unsigned int nb_rets, const TCGv *rets,
- unsigned int nb_params, const TCGv *args1)
+void tcg_gen_callN(TCGContext *s, TCGv_ptr func, unsigned int flags,
+ int sizemask, TCGArg ret, int nargs, TCGArg *args)
{
- TCGv ret, *args2, rets_2[2], arg;
- int j, i, call_type;
-
- if (nb_rets == 1) {
- ret = rets[0];
- if (tcg_get_base_type(s, ret) == TCG_TYPE_I64) {
- nb_rets = 2;
+ int call_type;
+ int i;
+ int real_args;
+ int nb_rets;
+ TCGArg *nparam;
+ *gen_opc_ptr++ = INDEX_op_call;
+ nparam = gen_opparam_ptr++;
+ call_type = (flags & TCG_CALL_TYPE_MASK);
+ if (ret != TCG_CALL_DUMMY_ARG) {
+#if TCG_TARGET_REG_BITS < 64
+ if (sizemask & 1) {
#ifdef TCG_TARGET_WORDS_BIGENDIAN
- rets_2[0] = TCGV_HIGH(ret);
- rets_2[1] = ret;
+ *gen_opparam_ptr++ = ret + 1;
+ *gen_opparam_ptr++ = ret;
#else
- rets_2[0] = ret;
- rets_2[1] = TCGV_HIGH(ret);
+ *gen_opparam_ptr++ = ret;
+ *gen_opparam_ptr++ = ret + 1;
#endif
- rets = rets_2;
+ nb_rets = 2;
+ } else
+#endif
+ {
+ *gen_opparam_ptr++ = ret;
+ nb_rets = 1;
}
+ } else {
+ nb_rets = 0;
}
- args2 = alloca((nb_params * 3) * sizeof(TCGv));
- j = 0;
- call_type = (flags & TCG_CALL_TYPE_MASK);
- for(i = 0; i < nb_params; i++) {
- arg = args1[i];
- if (tcg_get_base_type(s, arg) == TCG_TYPE_I64) {
+ real_args = 0;
+ for (i = 0; i < nargs; i++) {
+#if TCG_TARGET_REG_BITS < 64
+ if (sizemask & (2 << i)) {
#ifdef TCG_TARGET_I386
/* REGPARM case: if the third parameter is 64 bit, it is
allocated on the stack */
- if (j == 2 && call_type == TCG_CALL_TYPE_REGPARM) {
+ if (i == 2 && call_type == TCG_CALL_TYPE_REGPARM) {
call_type = TCG_CALL_TYPE_REGPARM_2;
flags = (flags & ~TCG_CALL_TYPE_MASK) | call_type;
}
- args2[j++] = arg;
- args2[j++] = TCGV_HIGH(arg);
-#else
+#endif
#ifdef TCG_TARGET_CALL_ALIGN_ARGS
/* some targets want aligned 64 bit args */
- if (j & 1) {
- args2[j++] = TCG_CALL_DUMMY_ARG;
+ if (real_args & 1) {
+ *gen_opparam_ptr++ = TCG_CALL_DUMMY_ARG;
+ real_args++;
}
#endif
#ifdef TCG_TARGET_WORDS_BIGENDIAN
- args2[j++] = TCGV_HIGH(arg);
- args2[j++] = arg;
+ *gen_opparam_ptr++ = args[i] + 1;
+ *gen_opparam_ptr++ = args[i];
#else
- args2[j++] = arg;
- args2[j++] = TCGV_HIGH(arg);
+ *gen_opparam_ptr++ = args[i];
+ *gen_opparam_ptr++ = args[i] + 1;
#endif
+ real_args += 2;
+ } else
#endif
- } else {
- args2[j++] = arg;
+ {
+ *gen_opparam_ptr++ = args[i];
+ real_args++;
}
}
- tcg_gen_call_internal(s, func, flags,
- nb_rets, rets, j, args2);
-}
-#else
-void tcg_gen_call(TCGContext *s, TCGv func, unsigned int flags,
- unsigned int nb_rets, const TCGv *rets,
- unsigned int nb_params, const TCGv *args1)
-{
- tcg_gen_call_internal(s, func, flags,
- nb_rets, rets, nb_params, args1);
+ *gen_opparam_ptr++ = GET_TCGV_PTR(func);
+
+ *gen_opparam_ptr++ = flags;
+
+ *nparam = (nb_rets << 16) | (real_args + 1);
+
+ /* total parameters, needed to go backward in the instruction stream */
+ *gen_opparam_ptr++ = 1 + nb_rets + real_args + 3;
}
-#endif
#if TCG_TARGET_REG_BITS == 32
-void tcg_gen_shifti_i64(TCGv ret, TCGv arg1,
+void tcg_gen_shifti_i64(TCGv_i64 ret, TCGv_i64 arg1,
int c, int right, int arith)
{
if (c == 0) {
- tcg_gen_mov_i32(ret, arg1);
+ tcg_gen_mov_i32(TCGV_LOW(ret), TCGV_LOW(arg1));
tcg_gen_mov_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1));
} else if (c >= 32) {
c -= 32;
if (right) {
if (arith) {
- tcg_gen_sari_i32(ret, TCGV_HIGH(arg1), c);
+ tcg_gen_sari_i32(TCGV_LOW(ret), TCGV_HIGH(arg1), c);
tcg_gen_sari_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), 31);
} else {
- tcg_gen_shri_i32(ret, TCGV_HIGH(arg1), c);
+ tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_HIGH(arg1), c);
tcg_gen_movi_i32(TCGV_HIGH(ret), 0);
}
} else {
- tcg_gen_shli_i32(TCGV_HIGH(ret), arg1, c);
- tcg_gen_movi_i32(ret, 0);
+ tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_LOW(arg1), c);
+ tcg_gen_movi_i32(TCGV_LOW(ret), 0);
}
} else {
- TCGv t0, t1;
+ TCGv_i32 t0, t1;
- t0 = tcg_temp_new(TCG_TYPE_I32);
- t1 = tcg_temp_new(TCG_TYPE_I32);
+ t0 = tcg_temp_new_i32();
+ t1 = tcg_temp_new_i32();
if (right) {
tcg_gen_shli_i32(t0, TCGV_HIGH(arg1), 32 - c);
if (arith)
tcg_gen_sari_i32(t1, TCGV_HIGH(arg1), c);
- else
+ else
tcg_gen_shri_i32(t1, TCGV_HIGH(arg1), c);
- tcg_gen_shri_i32(ret, arg1, c);
- tcg_gen_or_i32(ret, ret, t0);
+ tcg_gen_shri_i32(TCGV_LOW(ret), TCGV_LOW(arg1), c);
+ tcg_gen_or_i32(TCGV_LOW(ret), TCGV_LOW(ret), t0);
tcg_gen_mov_i32(TCGV_HIGH(ret), t1);
} else {
- tcg_gen_shri_i32(t0, arg1, 32 - c);
+ tcg_gen_shri_i32(t0, TCGV_LOW(arg1), 32 - c);
/* Note: ret can be the same as arg1, so we use t1 */
- tcg_gen_shli_i32(t1, arg1, c);
+ tcg_gen_shli_i32(t1, TCGV_LOW(arg1), c);
tcg_gen_shli_i32(TCGV_HIGH(ret), TCGV_HIGH(arg1), c);
tcg_gen_or_i32(TCGV_HIGH(ret), TCGV_HIGH(ret), t0);
- tcg_gen_mov_i32(ret, t1);
+ tcg_gen_mov_i32(TCGV_LOW(ret), t1);
}
- tcg_temp_free(t0);
- tcg_temp_free(t1);
+ tcg_temp_free_i32(t0);
+ tcg_temp_free_i32(t1);
}
}
#endif
-void tcg_reg_alloc_start(TCGContext *s)
+static void tcg_reg_alloc_start(TCGContext *s)
{
int i;
TCGTemp *ts;
if (idx < s->nb_globals) {
pstrcpy(buf, buf_size, ts->name);
} else {
- snprintf(buf, buf_size, "tmp%d", idx - s->nb_globals);
+ 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(TCGContext *s, char *buf, int buf_size, TCGv arg)
+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(arg));
+ return tcg_get_arg_str_idx(s, buf, buf_size, GET_TCGV_I64(arg));
}
static int helper_cmp(const void *p1, const void *p2)
return NULL;
}
+static const char * const cond_name[] =
+{
+ [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"
+};
+
void tcg_dump_ops(TCGContext *s, FILE *outfile)
{
const uint16_t *opc_ptr;
val = args[1];
th = tcg_find_helper(s, val);
if (th) {
- fprintf(outfile, th->name);
+ fprintf(outfile, "%s", th->name);
} else {
if (c == INDEX_op_movi_i32)
fprintf(outfile, "0x%x", (uint32_t)val);
fprintf(outfile, "%s",
tcg_get_arg_str_idx(s, buf, sizeof(buf), args[k++]));
}
- for(i = 0; i < nb_cargs; i++) {
+ if (c == INDEX_op_brcond_i32
+#if TCG_TARGET_REG_BITS == 32
+ || c == INDEX_op_brcond2_i32
+#elif TCG_TARGET_REG_BITS == 64
+ || c == INDEX_op_brcond_i64
+#endif
+ ) {
+ if (args[k] < ARRAY_SIZE(cond_name) && cond_name[args[k]])
+ fprintf(outfile, ",%s", cond_name[args[k++]]);
+ else
+ fprintf(outfile, ",$0x%" TCG_PRIlx, args[k++]);
+ i = 1;
+ }
+ else
+ i = 0;
+ for(; i < nb_cargs; i++) {
if (k != 0)
fprintf(outfile, ",");
arg = args[k++];
}
}
-/* liveness analysis: end of basic block: globals are live, temps are dead */
-static inline void tcg_la_bb_end(TCGContext *s, uint8_t *dead_temps)
+/* liveness analysis: end of function: globals are live, temps are
+ dead. */
+/* XXX: at this stage, not used as there would be little gains because
+ most TBs end with a conditional jump. */
+static inline void tcg_la_func_end(TCGContext *s, uint8_t *dead_temps)
{
memset(dead_temps, 0, s->nb_globals);
memset(dead_temps + s->nb_globals, 1, s->nb_temps - s->nb_globals);
}
+/* liveness analysis: end of basic block: globals are live, temps are
+ dead, local temps are live. */
+static inline void tcg_la_bb_end(TCGContext *s, uint8_t *dead_temps)
+{
+ int i;
+ TCGTemp *ts;
+
+ memset(dead_temps, 0, s->nb_globals);
+ ts = &s->temps[s->nb_globals];
+ for(i = s->nb_globals; i < s->nb_temps; i++) {
+ if (ts->temp_local)
+ dead_temps[i] = 0;
+ else
+ dead_temps[i] = 1;
+ ts++;
+ }
+}
+
/* Liveness analysis : update the opc_dead_iargs array to tell if a
given input arguments is dead. Instructions updating dead
temporaries are removed. */
-void tcg_liveness_analysis(TCGContext *s)
+static void tcg_liveness_analysis(TCGContext *s)
{
int i, op_index, op, nb_args, nb_iargs, nb_oargs, arg, nb_ops;
TCGArg *args;
dead_temps[arg] = 1;
}
- /* globals are live (they may be used by the call) */
- memset(dead_temps, 0, s->nb_globals);
-
+ if (!(call_flags & TCG_CALL_CONST)) {
+ /* globals are live (they may be used by the call) */
+ memset(dead_temps, 0, s->nb_globals);
+ }
+
/* input args are live */
dead_iargs = 0;
for(i = 0; i < nb_iargs; i++) {
/* mark the temporary as dead */
dead_temps[args[0]] = 1;
break;
- case INDEX_op_macro_2:
- {
- int dead_args[2], macro_id;
- int saved_op_index, saved_arg_index;
- int macro_op_index, macro_arg_index;
- int macro_end_op_index, macro_end_arg_index;
- int last_nb_temps;
-
- nb_args = 3;
- args -= nb_args;
- dead_args[0] = dead_temps[args[0]];
- dead_args[1] = dead_temps[args[1]];
- macro_id = args[2];
-
- /* call the macro function which generate code
- depending on the live outputs */
- saved_op_index = op_index;
- saved_arg_index = args - gen_opparam_buf;
-
- /* add a macro start instruction */
- *gen_opc_ptr++ = INDEX_op_macro_start;
- *gen_opparam_ptr++ = saved_op_index;
- *gen_opparam_ptr++ = saved_arg_index;
-
- macro_op_index = gen_opc_ptr - gen_opc_buf;
- macro_arg_index = gen_opparam_ptr - gen_opparam_buf;
-
- last_nb_temps = s->nb_temps;
-
- s->macro_func(s, macro_id, dead_args);
-
- /* realloc temp info (XXX: make it faster) */
- if (s->nb_temps > last_nb_temps) {
- uint8_t *new_dead_temps;
-
- new_dead_temps = tcg_malloc(s->nb_temps);
- memcpy(new_dead_temps, dead_temps, last_nb_temps);
- memset(new_dead_temps + last_nb_temps, 1,
- s->nb_temps - last_nb_temps);
- dead_temps = new_dead_temps;
- }
-
- macro_end_op_index = gen_opc_ptr - gen_opc_buf;
- macro_end_arg_index = gen_opparam_ptr - gen_opparam_buf;
-
- /* end of macro: add a goto to the next instruction */
- *gen_opc_ptr++ = INDEX_op_macro_end;
- *gen_opparam_ptr++ = op_index + 1;
- *gen_opparam_ptr++ = saved_arg_index + nb_args;
-
- /* modify the macro operation to be a macro_goto */
- gen_opc_buf[op_index] = INDEX_op_macro_goto;
- args[0] = macro_op_index;
- args[1] = macro_arg_index;
- args[2] = 0; /* dummy third arg to match the
- macro parameters */
-
- /* set the next instruction to the end of the macro */
- op_index = macro_end_op_index;
- args = macro_end_arg_index + gen_opparam_buf;
- }
- break;
- case INDEX_op_macro_start:
- args -= 2;
- op_index = args[0];
- args = gen_opparam_buf + args[1];
- break;
- case INDEX_op_macro_goto:
- case INDEX_op_macro_end:
- tcg_abort(); /* should never happen in liveness analysis */
case INDEX_op_end:
break;
/* XXX: optimize by hardcoding common cases (e.g. triadic ops) */
default:
- if (op > INDEX_op_end) {
- args -= def->nb_args;
- nb_iargs = def->nb_iargs;
- nb_oargs = def->nb_oargs;
+ args -= def->nb_args;
+ 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])
- goto do_not_remove;
- }
- tcg_set_nop(s, gen_opc_buf + op_index, args, def->nb_args);
+ /* 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])
+ goto do_not_remove;
+ }
+ tcg_set_nop(s, gen_opc_buf + op_index, args, def->nb_args);
#ifdef CONFIG_PROFILER
- s->del_op_count++;
+ s->del_op_count++;
#endif
- } else {
- do_not_remove:
+ } else {
+ do_not_remove:
- /* output args are dead */
- for(i = 0; i < nb_oargs; i++) {
- arg = args[i];
- dead_temps[arg] = 1;
- }
-
- /* if end of basic block, update */
- if (def->flags & TCG_OPF_BB_END) {
- tcg_la_bb_end(s, dead_temps);
- } else if (def->flags & TCG_OPF_CALL_CLOBBER) {
- /* globals are live */
- memset(dead_temps, 0, s->nb_globals);
- }
-
- /* input args are live */
- dead_iargs = 0;
- for(i = 0; i < nb_iargs; i++) {
- arg = args[i + nb_oargs];
- if (dead_temps[arg]) {
- dead_iargs |= (1 << i);
- }
- dead_temps[arg] = 0;
+ /* output args are dead */
+ for(i = 0; i < nb_oargs; i++) {
+ arg = args[i];
+ dead_temps[arg] = 1;
+ }
+
+ /* if end of basic block, update */
+ if (def->flags & TCG_OPF_BB_END) {
+ tcg_la_bb_end(s, dead_temps);
+ } else if (def->flags & TCG_OPF_CALL_CLOBBER) {
+ /* globals are live */
+ memset(dead_temps, 0, s->nb_globals);
+ }
+
+ /* input args are live */
+ dead_iargs = 0;
+ for(i = 0; i < nb_iargs; i++) {
+ arg = args[i + nb_oargs];
+ if (dead_temps[arg]) {
+ dead_iargs |= (1 << i);
}
- s->op_dead_iargs[op_index] = dead_iargs;
+ dead_temps[arg] = 0;
}
- } else {
- /* legacy dyngen operations */
- args -= def->nb_args;
- /* mark end of basic block */
- tcg_la_bb_end(s, dead_temps);
+ s->op_dead_iargs[op_index] = dead_iargs;
}
break;
}
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 void temp_save(TCGContext *s, int temp, TCGRegSet allocated_regs)
+{
+ TCGTemp *ts;
+ int reg;
+
+ ts = &s->temps[temp];
+ if (!ts->fixed_reg) {
+ switch(ts->val_type) {
+ case TEMP_VAL_REG:
+ tcg_reg_free(s, ts->reg);
+ break;
+ case TEMP_VAL_DEAD:
+ ts->val_type = TEMP_VAL_MEM;
+ break;
+ case TEMP_VAL_CONST:
+ reg = tcg_reg_alloc(s, tcg_target_available_regs[ts->type],
+ allocated_regs);
+ if (!ts->mem_allocated)
+ temp_allocate_frame(s, temp);
+ tcg_out_movi(s, ts->type, reg, ts->val);
+ tcg_out_st(s, ts->type, reg, ts->mem_reg, ts->mem_offset);
+ ts->val_type = TEMP_VAL_MEM;
+ break;
+ case TEMP_VAL_MEM:
+ break;
+ default:
+ tcg_abort();
+ }
+ }
+}
+
/* save globals to their cannonical 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)
{
- TCGTemp *ts;
- int i, reg;
+ int i;
for(i = 0; i < s->nb_globals; i++) {
- ts = &s->temps[i];
- if (!ts->fixed_reg) {
- switch(ts->val_type) {
- case TEMP_VAL_REG:
- tcg_reg_free(s, ts->reg);
- break;
- case TEMP_VAL_DEAD:
- ts->val_type = TEMP_VAL_MEM;
- break;
- case TEMP_VAL_CONST:
- reg = tcg_reg_alloc(s, tcg_target_available_regs[ts->type],
- allocated_regs);
- tcg_out_movi(s, ts->type, reg, ts->val);
- tcg_out_st(s, ts->type, reg, ts->mem_reg, ts->mem_offset);
- ts->val_type = TEMP_VAL_MEM;
- break;
- case TEMP_VAL_MEM:
- break;
- default:
- tcg_abort();
- }
- }
+ temp_save(s, i, allocated_regs);
}
}
for(i = s->nb_globals; i < s->nb_temps; i++) {
ts = &s->temps[i];
- if (ts->val_type == TEMP_VAL_REG) {
- s->reg_to_temp[ts->reg] = -1;
+ if (ts->temp_local) {
+ temp_save(s, i, allocated_regs);
+ } else {
+ if (ts->val_type == TEMP_VAL_REG) {
+ s->reg_to_temp[ts->reg] = -1;
+ }
+ ts->val_type = TEMP_VAL_DEAD;
}
- ts->val_type = TEMP_VAL_DEAD;
}
save_globals(s, allocated_regs);
propagation */
tcg_out_movi(s, ots->type, ots->reg, val);
} else {
- /* The movi is not explicitely generated here */
+ /* 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;
/* store globals and free associated registers (we assume the call
can modify any global. */
- save_globals(s, allocated_regs);
+ if (!(flags & TCG_CALL_CONST)) {
+ save_globals(s, allocated_regs);
+ }
tcg_out_op(s, opc, &func_arg, &const_func_arg);
#ifdef CONFIG_PROFILER
-static int64_t dyngen_table_op_count[NB_OPS];
+static int64_t tcg_table_op_count[NB_OPS];
void dump_op_count(void)
{
int i;
FILE *f;
- f = fopen("/tmp/op1.log", "w");
- for(i = 0; i < INDEX_op_end; i++) {
- fprintf(f, "%s %" PRId64 "\n", tcg_op_defs[i].name, dyngen_table_op_count[i]);
- }
- fclose(f);
- f = fopen("/tmp/op2.log", "w");
+ f = fopen("/tmp/op.log", "w");
for(i = INDEX_op_end; i < NB_OPS; i++) {
- fprintf(f, "%s %" PRId64 "\n", tcg_op_defs[i].name, dyngen_table_op_count[i]);
+ fprintf(f, "%s %" PRId64 "\n", tcg_op_defs[i].name, tcg_table_op_count[i]);
}
fclose(f);
}
static inline int tcg_gen_code_common(TCGContext *s, uint8_t *gen_code_buf,
long search_pc)
{
- int opc, op_index, macro_op_index;
+ int opc, op_index;
const TCGOpDef *def;
unsigned int dead_iargs;
const TCGArg *args;
#ifdef DEBUG_DISAS
- if (unlikely(loglevel & CPU_LOG_TB_OP)) {
- fprintf(logfile, "OP:\n");
+ if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP))) {
+ qemu_log("OP:\n");
tcg_dump_ops(s, logfile);
- fprintf(logfile, "\n");
+ qemu_log("\n");
}
#endif
#endif
#ifdef DEBUG_DISAS
- if (unlikely(loglevel & CPU_LOG_TB_OP_OPT)) {
- fprintf(logfile, "OP after la:\n");
+ if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP_OPT))) {
+ qemu_log("OP after la:\n");
tcg_dump_ops(s, logfile);
- fprintf(logfile, "\n");
+ qemu_log("\n");
}
#endif
s->code_buf = gen_code_buf;
s->code_ptr = gen_code_buf;
- macro_op_index = -1;
args = gen_opparam_buf;
op_index = 0;
for(;;) {
opc = gen_opc_buf[op_index];
#ifdef CONFIG_PROFILER
- dyngen_table_op_count[opc]++;
+ tcg_table_op_count[opc]++;
#endif
def = &tcg_op_defs[opc];
#if 0
}
}
break;
- case INDEX_op_macro_goto:
- macro_op_index = op_index; /* only used for exceptions */
- op_index = args[0] - 1;
- args = gen_opparam_buf + args[1];
- goto next;
- case INDEX_op_macro_end:
- macro_op_index = -1; /* only used for exceptions */
- op_index = args[0] - 1;
- args = gen_opparam_buf + args[1];
- goto next;
- case INDEX_op_macro_start:
- /* must never happen here */
- tcg_abort();
case INDEX_op_set_label:
tcg_reg_alloc_bb_end(s, s->reserved_regs);
tcg_out_label(s, args[0], (long)s->code_ptr);
goto next;
case INDEX_op_end:
goto the_end;
-
-#ifdef CONFIG_DYNGEN_OP
- case 0 ... INDEX_op_end - 1:
- /* legacy dyngen ops */
-#ifdef CONFIG_PROFILER
- s->old_op_count++;
-#endif
- tcg_reg_alloc_bb_end(s, s->reserved_regs);
- if (search_pc >= 0) {
- s->code_ptr += def->copy_size;
- args += def->nb_args;
- } else {
- args = dyngen_op(s, opc, args);
- }
- goto next;
-#endif
default:
/* Note: in order to speed up the code, it would be much
faster to have specialized register allocator functions for
break;
}
args += def->nb_args;
- next: ;
+ next:
if (search_pc >= 0 && search_pc < s->code_ptr - gen_code_buf) {
- if (macro_op_index >= 0)
- return macro_op_index;
- else
- return op_index;
+ return op_index;
}
op_index++;
#ifndef NDEBUG
return -1;
}
-int dyngen_code(TCGContext *s, uint8_t *gen_code_buf)
+int tcg_gen_code(TCGContext *s, uint8_t *gen_code_buf)
{
#ifdef CONFIG_PROFILER
{
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 dyngen_code_search_pc(TCGContext *s, uint8_t *gen_code_buf, long offset)
+int tcg_gen_code_search_pc(TCGContext *s, uint8_t *gen_code_buf, long offset)
{
return tcg_gen_code_common(s, gen_code_buf, offset);
}
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, "old ops/total ops %0.1f%%\n",
- s->op_count ? (double)s->old_op_count / s->op_count * 100.0 : 0);
cpu_fprintf(f, "deleted ops/TB %0.2f\n",
s->tb_count ?
(double)s->del_op_count / s->tb_count : 0);
projects / qemu / blobdiff