qemu-e2k/tcg/optimize.c

1536 lines
49 KiB
C
Raw Normal View History

/*
* Optimizations for Tiny Code Generator for QEMU
*
* Copyright (c) 2010 Samsung Electronics.
* Contributed by Kirill Batuzov <batuzovk@ispras.ru>
*
* 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.
*/
#include "qemu/osdep.h"
#include "tcg-op.h"
#define CASE_OP_32_64(x) \
glue(glue(case INDEX_op_, x), _i32): \
glue(glue(case INDEX_op_, x), _i64)
#define CASE_OP_32_64_VEC(x) \
glue(glue(case INDEX_op_, x), _i32): \
glue(glue(case INDEX_op_, x), _i64): \
glue(glue(case INDEX_op_, x), _vec)
struct tcg_temp_info {
bool is_const;
TCGTemp *prev_copy;
TCGTemp *next_copy;
tcg_target_ulong val;
tcg_target_ulong mask;
};
static inline struct tcg_temp_info *ts_info(TCGTemp *ts)
{
return ts->state_ptr;
}
static inline struct tcg_temp_info *arg_info(TCGArg arg)
{
return ts_info(arg_temp(arg));
}
static inline bool ts_is_const(TCGTemp *ts)
{
return ts_info(ts)->is_const;
}
static inline bool arg_is_const(TCGArg arg)
{
return ts_is_const(arg_temp(arg));
}
static inline bool ts_is_copy(TCGTemp *ts)
{
return ts_info(ts)->next_copy != ts;
}
/* Reset TEMP's state, possibly removing the temp for the list of copies. */
static void reset_ts(TCGTemp *ts)
{
struct tcg_temp_info *ti = ts_info(ts);
struct tcg_temp_info *pi = ts_info(ti->prev_copy);
struct tcg_temp_info *ni = ts_info(ti->next_copy);
ni->prev_copy = ti->prev_copy;
pi->next_copy = ti->next_copy;
ti->next_copy = ts;
ti->prev_copy = ts;
ti->is_const = false;
ti->mask = -1;
}
static void reset_temp(TCGArg arg)
{
reset_ts(arg_temp(arg));
}
/* Initialize and activate a temporary. */
2017-07-19 20:32:24 +02:00
static void init_ts_info(struct tcg_temp_info *infos,
TCGTempSet *temps_used, TCGTemp *ts)
{
size_t idx = temp_idx(ts);
2017-07-19 20:32:24 +02:00
if (!test_bit(idx, temps_used->l)) {
struct tcg_temp_info *ti = &infos[idx];
ts->state_ptr = ti;
ti->next_copy = ts;
ti->prev_copy = ts;
ti->is_const = false;
ti->mask = -1;
2017-07-19 20:32:24 +02:00
set_bit(idx, temps_used->l);
}
}
2017-07-19 20:32:24 +02:00
static void init_arg_info(struct tcg_temp_info *infos,
TCGTempSet *temps_used, TCGArg arg)
{
2017-07-19 20:32:24 +02:00
init_ts_info(infos, temps_used, arg_temp(arg));
}
static TCGTemp *find_better_copy(TCGContext *s, TCGTemp *ts)
{
TCGTemp *i;
/* If this is already a global, we can't do better. */
if (ts->temp_global) {
return ts;
}
/* Search for a global first. */
for (i = ts_info(ts)->next_copy; i != ts; i = ts_info(i)->next_copy) {
if (i->temp_global) {
return i;
}
}
/* If it is a temp, search for a temp local. */
if (!ts->temp_local) {
for (i = ts_info(ts)->next_copy; i != ts; i = ts_info(i)->next_copy) {
if (ts->temp_local) {
return i;
}
}
}
/* Failure to find a better representation, return the same temp. */
return ts;
}
static bool ts_are_copies(TCGTemp *ts1, TCGTemp *ts2)
{
TCGTemp *i;
if (ts1 == ts2) {
return true;
}
if (!ts_is_copy(ts1) || !ts_is_copy(ts2)) {
return false;
}
for (i = ts_info(ts1)->next_copy; i != ts1; i = ts_info(i)->next_copy) {
if (i == ts2) {
return true;
}
}
return false;
}
static bool args_are_copies(TCGArg arg1, TCGArg arg2)
{
return ts_are_copies(arg_temp(arg1), arg_temp(arg2));
}
static void tcg_opt_gen_movi(TCGContext *s, TCGOp *op, TCGArg dst, TCGArg val)
{
const TCGOpDef *def;
TCGOpcode new_op;
tcg_target_ulong mask;
struct tcg_temp_info *di = arg_info(dst);
def = &tcg_op_defs[op->opc];
if (def->flags & TCG_OPF_VECTOR) {
new_op = INDEX_op_dupi_vec;
} else if (def->flags & TCG_OPF_64BIT) {
new_op = INDEX_op_movi_i64;
} else {
new_op = INDEX_op_movi_i32;
}
op->opc = new_op;
/* TCGOP_VECL and TCGOP_VECE remain unchanged. */
op->args[0] = dst;
op->args[1] = val;
reset_temp(dst);
di->is_const = true;
di->val = val;
mask = val;
if (TCG_TARGET_REG_BITS > 32 && new_op == INDEX_op_movi_i32) {
/* High bits of the destination are now garbage. */
mask |= ~0xffffffffull;
}
di->mask = mask;
}
static void tcg_opt_gen_mov(TCGContext *s, TCGOp *op, TCGArg dst, TCGArg src)
{
TCGTemp *dst_ts = arg_temp(dst);
TCGTemp *src_ts = arg_temp(src);
const TCGOpDef *def;
struct tcg_temp_info *di;
struct tcg_temp_info *si;
tcg_target_ulong mask;
TCGOpcode new_op;
if (ts_are_copies(dst_ts, src_ts)) {
tcg_op_remove(s, op);
return;
}
reset_ts(dst_ts);
di = ts_info(dst_ts);
si = ts_info(src_ts);
def = &tcg_op_defs[op->opc];
if (def->flags & TCG_OPF_VECTOR) {
new_op = INDEX_op_mov_vec;
} else if (def->flags & TCG_OPF_64BIT) {
new_op = INDEX_op_mov_i64;
} else {
new_op = INDEX_op_mov_i32;
}
op->opc = new_op;
/* TCGOP_VECL and TCGOP_VECE remain unchanged. */
op->args[0] = dst;
op->args[1] = src;
mask = si->mask;
if (TCG_TARGET_REG_BITS > 32 && new_op == INDEX_op_mov_i32) {
/* High bits of the destination are now garbage. */
mask |= ~0xffffffffull;
}
di->mask = mask;
if (src_ts->type == dst_ts->type) {
struct tcg_temp_info *ni = ts_info(si->next_copy);
di->next_copy = si->next_copy;
di->prev_copy = src_ts;
ni->prev_copy = dst_ts;
si->next_copy = dst_ts;
di->is_const = si->is_const;
di->val = si->val;
}
}
static TCGArg do_constant_folding_2(TCGOpcode op, TCGArg x, TCGArg y)
{
uint64_t l64, h64;
switch (op) {
CASE_OP_32_64(add):
return x + y;
CASE_OP_32_64(sub):
return x - y;
CASE_OP_32_64(mul):
return x * y;
CASE_OP_32_64(and):
return x & y;
CASE_OP_32_64(or):
return x | y;
CASE_OP_32_64(xor):
return x ^ y;
case INDEX_op_shl_i32:
return (uint32_t)x << (y & 31);
case INDEX_op_shl_i64:
return (uint64_t)x << (y & 63);
case INDEX_op_shr_i32:
return (uint32_t)x >> (y & 31);
case INDEX_op_shr_i64:
return (uint64_t)x >> (y & 63);
case INDEX_op_sar_i32:
return (int32_t)x >> (y & 31);
case INDEX_op_sar_i64:
return (int64_t)x >> (y & 63);
case INDEX_op_rotr_i32:
return ror32(x, y & 31);
case INDEX_op_rotr_i64:
return ror64(x, y & 63);
case INDEX_op_rotl_i32:
return rol32(x, y & 31);
case INDEX_op_rotl_i64:
return rol64(x, y & 63);
CASE_OP_32_64(not):
return ~x;
CASE_OP_32_64(neg):
return -x;
CASE_OP_32_64(andc):
return x & ~y;
CASE_OP_32_64(orc):
return x | ~y;
CASE_OP_32_64(eqv):
return ~(x ^ y);
CASE_OP_32_64(nand):
return ~(x & y);
CASE_OP_32_64(nor):
return ~(x | y);
case INDEX_op_clz_i32:
return (uint32_t)x ? clz32(x) : y;
case INDEX_op_clz_i64:
return x ? clz64(x) : y;
case INDEX_op_ctz_i32:
return (uint32_t)x ? ctz32(x) : y;
case INDEX_op_ctz_i64:
return x ? ctz64(x) : y;
case INDEX_op_ctpop_i32:
return ctpop32(x);
case INDEX_op_ctpop_i64:
return ctpop64(x);
CASE_OP_32_64(ext8s):
return (int8_t)x;
CASE_OP_32_64(ext16s):
return (int16_t)x;
CASE_OP_32_64(ext8u):
return (uint8_t)x;
CASE_OP_32_64(ext16u):
return (uint16_t)x;
CASE_OP_32_64(bswap16):
return bswap16(x);
CASE_OP_32_64(bswap32):
return bswap32(x);
case INDEX_op_bswap64_i64:
return bswap64(x);
case INDEX_op_ext_i32_i64:
case INDEX_op_ext32s_i64:
return (int32_t)x;
case INDEX_op_extu_i32_i64:
case INDEX_op_extrl_i64_i32:
case INDEX_op_ext32u_i64:
return (uint32_t)x;
case INDEX_op_extrh_i64_i32:
return (uint64_t)x >> 32;
case INDEX_op_muluh_i32:
return ((uint64_t)(uint32_t)x * (uint32_t)y) >> 32;
case INDEX_op_mulsh_i32:
return ((int64_t)(int32_t)x * (int32_t)y) >> 32;
case INDEX_op_muluh_i64:
mulu64(&l64, &h64, x, y);
return h64;
case INDEX_op_mulsh_i64:
muls64(&l64, &h64, x, y);
return h64;
case INDEX_op_div_i32:
/* Avoid crashing on divide by zero, otherwise undefined. */
return (int32_t)x / ((int32_t)y ? : 1);
case INDEX_op_divu_i32:
return (uint32_t)x / ((uint32_t)y ? : 1);
case INDEX_op_div_i64:
return (int64_t)x / ((int64_t)y ? : 1);
case INDEX_op_divu_i64:
return (uint64_t)x / ((uint64_t)y ? : 1);
case INDEX_op_rem_i32:
return (int32_t)x % ((int32_t)y ? : 1);
case INDEX_op_remu_i32:
return (uint32_t)x % ((uint32_t)y ? : 1);
case INDEX_op_rem_i64:
return (int64_t)x % ((int64_t)y ? : 1);
case INDEX_op_remu_i64:
return (uint64_t)x % ((uint64_t)y ? : 1);
default:
fprintf(stderr,
"Unrecognized operation %d in do_constant_folding.\n", op);
tcg_abort();
}
}
static TCGArg do_constant_folding(TCGOpcode op, TCGArg x, TCGArg y)
{
const TCGOpDef *def = &tcg_op_defs[op];
TCGArg res = do_constant_folding_2(op, x, y);
if (!(def->flags & TCG_OPF_64BIT)) {
tcg/optimize: fix constant signedness By convention, on a 64-bit host TCG internally stores 32-bit constants as sign-extended. This is not the case in the optimizer when a 32-bit constant is folded. This doesn't seem to have more consequences than suboptimal code generation. For instance the x86 backend assumes sign-extended constants, and in some rare cases uses a 32-bit unsigned immediate 0xffffffff instead of a 8-bit signed immediate 0xff for the constant -1. This is with a ppc guest: before ------ ---- 0x9f29cc movi_i32 tmp1,$0xffffffff movi_i32 tmp2,$0x0 add2_i32 tmp0,CA,CA,tmp2,r6,tmp2 add2_i32 tmp0,CA,tmp0,CA,tmp1,tmp2 mov_i32 r10,tmp0 0x7fd8c7dfe90c: xor %ebp,%ebp 0x7fd8c7dfe90e: mov %ebp,%r11d 0x7fd8c7dfe911: mov 0x18(%r14),%r9d 0x7fd8c7dfe915: add %r9d,%r10d 0x7fd8c7dfe918: adc %ebp,%r11d 0x7fd8c7dfe91b: add $0xffffffff,%r10d 0x7fd8c7dfe922: adc %ebp,%r11d 0x7fd8c7dfe925: mov %r11d,0x134(%r14) 0x7fd8c7dfe92c: mov %r10d,0x28(%r14) after ----- ---- 0x9f29cc movi_i32 tmp1,$0xffffffffffffffff movi_i32 tmp2,$0x0 add2_i32 tmp0,CA,CA,tmp2,r6,tmp2 add2_i32 tmp0,CA,tmp0,CA,tmp1,tmp2 mov_i32 r10,tmp0 0x7f37010d490c: xor %ebp,%ebp 0x7f37010d490e: mov %ebp,%r11d 0x7f37010d4911: mov 0x18(%r14),%r9d 0x7f37010d4915: add %r9d,%r10d 0x7f37010d4918: adc %ebp,%r11d 0x7f37010d491b: add $0xffffffffffffffff,%r10d 0x7f37010d491f: adc %ebp,%r11d 0x7f37010d4922: mov %r11d,0x134(%r14) 0x7f37010d4929: mov %r10d,0x28(%r14) Signed-off-by: Aurelien Jarno <aurelien@aurel32.net> Message-Id: <1436544211-2769-2-git-send-email-aurelien@aurel32.net> Signed-off-by: Richard Henderson <rth@twiddle.net>
2015-07-10 18:03:31 +02:00
res = (int32_t)res;
}
return res;
}
static bool do_constant_folding_cond_32(uint32_t x, uint32_t y, TCGCond c)
{
switch (c) {
case TCG_COND_EQ:
return x == y;
case TCG_COND_NE:
return x != y;
case TCG_COND_LT:
return (int32_t)x < (int32_t)y;
case TCG_COND_GE:
return (int32_t)x >= (int32_t)y;
case TCG_COND_LE:
return (int32_t)x <= (int32_t)y;
case TCG_COND_GT:
return (int32_t)x > (int32_t)y;
case TCG_COND_LTU:
return x < y;
case TCG_COND_GEU:
return x >= y;
case TCG_COND_LEU:
return x <= y;
case TCG_COND_GTU:
return x > y;
default:
tcg_abort();
}
}
static bool do_constant_folding_cond_64(uint64_t x, uint64_t y, TCGCond c)
{
switch (c) {
case TCG_COND_EQ:
return x == y;
case TCG_COND_NE:
return x != y;
case TCG_COND_LT:
return (int64_t)x < (int64_t)y;
case TCG_COND_GE:
return (int64_t)x >= (int64_t)y;
case TCG_COND_LE:
return (int64_t)x <= (int64_t)y;
case TCG_COND_GT:
return (int64_t)x > (int64_t)y;
case TCG_COND_LTU:
return x < y;
case TCG_COND_GEU:
return x >= y;
case TCG_COND_LEU:
return x <= y;
case TCG_COND_GTU:
return x > y;
default:
tcg_abort();
}
}
static bool do_constant_folding_cond_eq(TCGCond c)
{
switch (c) {
case TCG_COND_GT:
case TCG_COND_LTU:
case TCG_COND_LT:
case TCG_COND_GTU:
case TCG_COND_NE:
return 0;
case TCG_COND_GE:
case TCG_COND_GEU:
case TCG_COND_LE:
case TCG_COND_LEU:
case TCG_COND_EQ:
return 1;
default:
tcg_abort();
}
}
/* Return 2 if the condition can't be simplified, and the result
of the condition (0 or 1) if it can */
static TCGArg do_constant_folding_cond(TCGOpcode op, TCGArg x,
TCGArg y, TCGCond c)
{
tcg_target_ulong xv = arg_info(x)->val;
tcg_target_ulong yv = arg_info(y)->val;
if (arg_is_const(x) && arg_is_const(y)) {
const TCGOpDef *def = &tcg_op_defs[op];
tcg_debug_assert(!(def->flags & TCG_OPF_VECTOR));
if (def->flags & TCG_OPF_64BIT) {
return do_constant_folding_cond_64(xv, yv, c);
} else {
return do_constant_folding_cond_32(xv, yv, c);
}
} else if (args_are_copies(x, y)) {
return do_constant_folding_cond_eq(c);
} else if (arg_is_const(y) && yv == 0) {
switch (c) {
case TCG_COND_LTU:
return 0;
case TCG_COND_GEU:
return 1;
default:
return 2;
}
}
return 2;
}
/* Return 2 if the condition can't be simplified, and the result
of the condition (0 or 1) if it can */
static TCGArg do_constant_folding_cond2(TCGArg *p1, TCGArg *p2, TCGCond c)
{
TCGArg al = p1[0], ah = p1[1];
TCGArg bl = p2[0], bh = p2[1];
if (arg_is_const(bl) && arg_is_const(bh)) {
tcg_target_ulong blv = arg_info(bl)->val;
tcg_target_ulong bhv = arg_info(bh)->val;
uint64_t b = deposit64(blv, 32, 32, bhv);
if (arg_is_const(al) && arg_is_const(ah)) {
tcg_target_ulong alv = arg_info(al)->val;
tcg_target_ulong ahv = arg_info(ah)->val;
uint64_t a = deposit64(alv, 32, 32, ahv);
return do_constant_folding_cond_64(a, b, c);
}
if (b == 0) {
switch (c) {
case TCG_COND_LTU:
return 0;
case TCG_COND_GEU:
return 1;
default:
break;
}
}
}
if (args_are_copies(al, bl) && args_are_copies(ah, bh)) {
return do_constant_folding_cond_eq(c);
}
return 2;
}
static bool swap_commutative(TCGArg dest, TCGArg *p1, TCGArg *p2)
{
TCGArg a1 = *p1, a2 = *p2;
int sum = 0;
sum += arg_is_const(a1);
sum -= arg_is_const(a2);
/* Prefer the constant in second argument, and then the form
op a, a, b, which is better handled on non-RISC hosts. */
if (sum > 0 || (sum == 0 && dest == a2)) {
*p1 = a2;
*p2 = a1;
return true;
}
return false;
}
static bool swap_commutative2(TCGArg *p1, TCGArg *p2)
{
int sum = 0;
sum += arg_is_const(p1[0]);
sum += arg_is_const(p1[1]);
sum -= arg_is_const(p2[0]);
sum -= arg_is_const(p2[1]);
if (sum > 0) {
TCGArg t;
t = p1[0], p1[0] = p2[0], p2[0] = t;
t = p1[1], p1[1] = p2[1], p2[1] = t;
return true;
}
return false;
}
/* Propagate constants and copies, fold constant expressions. */
void tcg_optimize(TCGContext *s)
{
int nb_temps, nb_globals;
TCGOp *op, *op_next, *prev_mb = NULL;
2017-07-19 20:32:24 +02:00
struct tcg_temp_info *infos;
TCGTempSet temps_used;
/* Array VALS has an element for each temp.
If this temp holds a constant then its value is kept in VALS' element.
If this temp is a copy of other ones then the other copies are
available through the doubly linked circular list. */
nb_temps = s->nb_temps;
nb_globals = s->nb_globals;
2017-07-19 20:32:24 +02:00
bitmap_zero(temps_used.l, nb_temps);
infos = tcg_malloc(sizeof(struct tcg_temp_info) * nb_temps);
QTAILQ_FOREACH_SAFE(op, &s->ops, link, op_next) {
tcg_target_ulong mask, partmask, affected;
int nb_oargs, nb_iargs, i;
TCGArg tmp;
TCGOpcode opc = op->opc;
const TCGOpDef *def = &tcg_op_defs[opc];
/* Count the arguments, and initialize the temps that are
going to be used */
if (opc == INDEX_op_call) {
nb_oargs = TCGOP_CALLO(op);
nb_iargs = TCGOP_CALLI(op);
for (i = 0; i < nb_oargs + nb_iargs; i++) {
TCGTemp *ts = arg_temp(op->args[i]);
if (ts) {
2017-07-19 20:32:24 +02:00
init_ts_info(infos, &temps_used, ts);
}
}
} else {
nb_oargs = def->nb_oargs;
nb_iargs = def->nb_iargs;
for (i = 0; i < nb_oargs + nb_iargs; i++) {
2017-07-19 20:32:24 +02:00
init_arg_info(infos, &temps_used, op->args[i]);
}
}
/* Do copy propagation */
for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) {
TCGTemp *ts = arg_temp(op->args[i]);
if (ts && ts_is_copy(ts)) {
op->args[i] = temp_arg(find_better_copy(s, ts));
}
}
/* For commutative operations make constant second argument */
switch (opc) {
CASE_OP_32_64_VEC(add):
CASE_OP_32_64_VEC(mul):
CASE_OP_32_64_VEC(and):
CASE_OP_32_64_VEC(or):
CASE_OP_32_64_VEC(xor):
CASE_OP_32_64(eqv):
CASE_OP_32_64(nand):
CASE_OP_32_64(nor):
CASE_OP_32_64(muluh):
CASE_OP_32_64(mulsh):
swap_commutative(op->args[0], &op->args[1], &op->args[2]);
break;
CASE_OP_32_64(brcond):
if (swap_commutative(-1, &op->args[0], &op->args[1])) {
op->args[2] = tcg_swap_cond(op->args[2]);
}
break;
CASE_OP_32_64(setcond):
if (swap_commutative(op->args[0], &op->args[1], &op->args[2])) {
op->args[3] = tcg_swap_cond(op->args[3]);
}
break;
CASE_OP_32_64(movcond):
if (swap_commutative(-1, &op->args[1], &op->args[2])) {
op->args[5] = tcg_swap_cond(op->args[5]);
}
/* For movcond, we canonicalize the "false" input reg to match
the destination reg so that the tcg backend can implement
a "move if true" operation. */
if (swap_commutative(op->args[0], &op->args[4], &op->args[3])) {
op->args[5] = tcg_invert_cond(op->args[5]);
}
break;
CASE_OP_32_64(add2):
swap_commutative(op->args[0], &op->args[2], &op->args[4]);
swap_commutative(op->args[1], &op->args[3], &op->args[5]);
break;
CASE_OP_32_64(mulu2):
CASE_OP_32_64(muls2):
swap_commutative(op->args[0], &op->args[2], &op->args[3]);
break;
case INDEX_op_brcond2_i32:
if (swap_commutative2(&op->args[0], &op->args[2])) {
op->args[4] = tcg_swap_cond(op->args[4]);
}
break;
case INDEX_op_setcond2_i32:
if (swap_commutative2(&op->args[1], &op->args[3])) {
op->args[5] = tcg_swap_cond(op->args[5]);
}
break;
default:
break;
}
/* Simplify expressions for "shift/rot r, 0, a => movi r, 0",
and "sub r, 0, a => neg r, a" case. */
switch (opc) {
CASE_OP_32_64(shl):
CASE_OP_32_64(shr):
CASE_OP_32_64(sar):
CASE_OP_32_64(rotl):
CASE_OP_32_64(rotr):
if (arg_is_const(op->args[1])
&& arg_info(op->args[1])->val == 0) {
tcg_opt_gen_movi(s, op, op->args[0], 0);
continue;
}
break;
CASE_OP_32_64_VEC(sub):
{
TCGOpcode neg_op;
bool have_neg;
if (arg_is_const(op->args[2])) {
/* Proceed with possible constant folding. */
break;
}
if (opc == INDEX_op_sub_i32) {
neg_op = INDEX_op_neg_i32;
have_neg = TCG_TARGET_HAS_neg_i32;
} else if (opc == INDEX_op_sub_i64) {
neg_op = INDEX_op_neg_i64;
have_neg = TCG_TARGET_HAS_neg_i64;
} else if (TCG_TARGET_HAS_neg_vec) {
TCGType type = TCGOP_VECL(op) + TCG_TYPE_V64;
unsigned vece = TCGOP_VECE(op);
neg_op = INDEX_op_neg_vec;
have_neg = tcg_can_emit_vec_op(neg_op, type, vece) > 0;
} else {
break;
}
if (!have_neg) {
break;
}
if (arg_is_const(op->args[1])
&& arg_info(op->args[1])->val == 0) {
op->opc = neg_op;
reset_temp(op->args[0]);
op->args[1] = op->args[2];
continue;
}
}
break;
CASE_OP_32_64_VEC(xor):
CASE_OP_32_64(nand):
if (!arg_is_const(op->args[1])
&& arg_is_const(op->args[2])
&& arg_info(op->args[2])->val == -1) {
i = 1;
goto try_not;
}
break;
CASE_OP_32_64(nor):
if (!arg_is_const(op->args[1])
&& arg_is_const(op->args[2])
&& arg_info(op->args[2])->val == 0) {
i = 1;
goto try_not;
}
break;
CASE_OP_32_64_VEC(andc):
if (!arg_is_const(op->args[2])
&& arg_is_const(op->args[1])
&& arg_info(op->args[1])->val == -1) {
i = 2;
goto try_not;
}
break;
CASE_OP_32_64_VEC(orc):
CASE_OP_32_64(eqv):
if (!arg_is_const(op->args[2])
&& arg_is_const(op->args[1])
&& arg_info(op->args[1])->val == 0) {
i = 2;
goto try_not;
}
break;
try_not:
{
TCGOpcode not_op;
bool have_not;
if (def->flags & TCG_OPF_VECTOR) {
not_op = INDEX_op_not_vec;
have_not = TCG_TARGET_HAS_not_vec;
} else if (def->flags & TCG_OPF_64BIT) {
not_op = INDEX_op_not_i64;
have_not = TCG_TARGET_HAS_not_i64;
} else {
not_op = INDEX_op_not_i32;
have_not = TCG_TARGET_HAS_not_i32;
}
if (!have_not) {
break;
}
op->opc = not_op;
reset_temp(op->args[0]);
op->args[1] = op->args[i];
continue;
}
default:
break;
}
/* Simplify expression for "op r, a, const => mov r, a" cases */
switch (opc) {
CASE_OP_32_64_VEC(add):
CASE_OP_32_64_VEC(sub):
CASE_OP_32_64_VEC(or):
CASE_OP_32_64_VEC(xor):
CASE_OP_32_64_VEC(andc):
CASE_OP_32_64(shl):
CASE_OP_32_64(shr):
CASE_OP_32_64(sar):
CASE_OP_32_64(rotl):
CASE_OP_32_64(rotr):
if (!arg_is_const(op->args[1])
&& arg_is_const(op->args[2])
&& arg_info(op->args[2])->val == 0) {
tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
continue;
}
break;
CASE_OP_32_64_VEC(and):
CASE_OP_32_64_VEC(orc):
CASE_OP_32_64(eqv):
if (!arg_is_const(op->args[1])
&& arg_is_const(op->args[2])
&& arg_info(op->args[2])->val == -1) {
tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
continue;
}
break;
default:
break;
}
/* Simplify using known-zero bits. Currently only ops with a single
output argument is supported. */
mask = -1;
optimize: optimize using nonzero bits This adds two optimizations using the non-zero bit mask. In some cases involving shifts or ANDs the value can become zero, and can thus be optimized to a move of zero. Second, useless zero-extension or an AND with constant can be detected that would only zero bits that are already zero. The main advantage of this optimization is that it turns zero-extensions into moves, thus enabling much better copy propagation (around 1% code reduction). Here is for example a "test $0xff0000,%ecx + je" before optimization: mov_i64 tmp0,rcx movi_i64 tmp1,$0xff0000 discard cc_src and_i64 cc_dst,tmp0,tmp1 movi_i32 cc_op,$0x1c ext32u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,eq,$0x0 and after (without patch on the left, with on the right): movi_i64 tmp1,$0xff0000 movi_i64 tmp1,$0xff0000 discard cc_src discard cc_src and_i64 cc_dst,rcx,tmp1 and_i64 cc_dst,rcx,tmp1 movi_i32 cc_op,$0x1c movi_i32 cc_op,$0x1c ext32u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,eq,$0x0 brcond_i64 cc_dst,tmp12,eq,$0x0 Other similar cases: "test %eax, %eax + jne" where eax is already 32-bit (after optimization, without patch on the left, with on the right): discard cc_src discard cc_src mov_i64 cc_dst,rax mov_i64 cc_dst,rax movi_i32 cc_op,$0x1c movi_i32 cc_op,$0x1c ext32u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,ne,$0x0 brcond_i64 rax,tmp12,ne,$0x0 "test $0x1, %dl + je": movi_i64 tmp1,$0x1 movi_i64 tmp1,$0x1 discard cc_src discard cc_src and_i64 cc_dst,rdx,tmp1 and_i64 cc_dst,rdx,tmp1 movi_i32 cc_op,$0x1a movi_i32 cc_op,$0x1a ext8u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,eq,$0x0 brcond_i64 cc_dst,tmp12,eq,$0x0 In some cases TCG even outsmarts GCC. :) Here the input code has "and $0x2,%eax + movslq %eax,%rbx + test %rbx, %rbx" and the optimizer, thanks to copy propagation, does the following: movi_i64 tmp12,$0x2 movi_i64 tmp12,$0x2 and_i64 rax,rax,tmp12 and_i64 rax,rax,tmp12 mov_i64 cc_dst,rax mov_i64 cc_dst,rax ext32s_i64 tmp0,rax -> nop mov_i64 rbx,tmp0 -> mov_i64 rbx,cc_dst and_i64 cc_dst,rbx,rbx -> nop Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
2013-01-12 00:42:53 +01:00
affected = -1;
switch (opc) {
CASE_OP_32_64(ext8s):
if ((arg_info(op->args[1])->mask & 0x80) != 0) {
break;
}
CASE_OP_32_64(ext8u):
mask = 0xff;
goto and_const;
CASE_OP_32_64(ext16s):
if ((arg_info(op->args[1])->mask & 0x8000) != 0) {
break;
}
CASE_OP_32_64(ext16u):
mask = 0xffff;
goto and_const;
case INDEX_op_ext32s_i64:
if ((arg_info(op->args[1])->mask & 0x80000000) != 0) {
break;
}
case INDEX_op_ext32u_i64:
mask = 0xffffffffU;
goto and_const;
CASE_OP_32_64(and):
mask = arg_info(op->args[2])->mask;
if (arg_is_const(op->args[2])) {
and_const:
affected = arg_info(op->args[1])->mask & ~mask;
}
mask = arg_info(op->args[1])->mask & mask;
break;
case INDEX_op_ext_i32_i64:
if ((arg_info(op->args[1])->mask & 0x80000000) != 0) {
break;
}
case INDEX_op_extu_i32_i64:
/* We do not compute affected as it is a size changing op. */
mask = (uint32_t)arg_info(op->args[1])->mask;
break;
CASE_OP_32_64(andc):
/* Known-zeros does not imply known-ones. Therefore unless
op->args[2] is constant, we can't infer anything from it. */
if (arg_is_const(op->args[2])) {
mask = ~arg_info(op->args[2])->mask;
goto and_const;
}
/* But we certainly know nothing outside args[1] may be set. */
mask = arg_info(op->args[1])->mask;
break;
case INDEX_op_sar_i32:
if (arg_is_const(op->args[2])) {
tmp = arg_info(op->args[2])->val & 31;
mask = (int32_t)arg_info(op->args[1])->mask >> tmp;
}
break;
case INDEX_op_sar_i64:
if (arg_is_const(op->args[2])) {
tmp = arg_info(op->args[2])->val & 63;
mask = (int64_t)arg_info(op->args[1])->mask >> tmp;
}
break;
case INDEX_op_shr_i32:
if (arg_is_const(op->args[2])) {
tmp = arg_info(op->args[2])->val & 31;
mask = (uint32_t)arg_info(op->args[1])->mask >> tmp;
}
break;
case INDEX_op_shr_i64:
if (arg_is_const(op->args[2])) {
tmp = arg_info(op->args[2])->val & 63;
mask = (uint64_t)arg_info(op->args[1])->mask >> tmp;
}
break;
case INDEX_op_extrl_i64_i32:
mask = (uint32_t)arg_info(op->args[1])->mask;
break;
case INDEX_op_extrh_i64_i32:
mask = (uint64_t)arg_info(op->args[1])->mask >> 32;
break;
CASE_OP_32_64(shl):
if (arg_is_const(op->args[2])) {
tmp = arg_info(op->args[2])->val & (TCG_TARGET_REG_BITS - 1);
mask = arg_info(op->args[1])->mask << tmp;
}
break;
CASE_OP_32_64(neg):
/* Set to 1 all bits to the left of the rightmost. */
mask = -(arg_info(op->args[1])->mask
& -arg_info(op->args[1])->mask);
break;
CASE_OP_32_64(deposit):
mask = deposit64(arg_info(op->args[1])->mask,
op->args[3], op->args[4],
arg_info(op->args[2])->mask);
break;
CASE_OP_32_64(extract):
mask = extract64(arg_info(op->args[1])->mask,
op->args[2], op->args[3]);
if (op->args[2] == 0) {
affected = arg_info(op->args[1])->mask & ~mask;
}
break;
CASE_OP_32_64(sextract):
mask = sextract64(arg_info(op->args[1])->mask,
op->args[2], op->args[3]);
if (op->args[2] == 0 && (tcg_target_long)mask >= 0) {
affected = arg_info(op->args[1])->mask & ~mask;
}
break;
CASE_OP_32_64(or):
CASE_OP_32_64(xor):
mask = arg_info(op->args[1])->mask | arg_info(op->args[2])->mask;
break;
case INDEX_op_clz_i32:
case INDEX_op_ctz_i32:
mask = arg_info(op->args[2])->mask | 31;
break;
case INDEX_op_clz_i64:
case INDEX_op_ctz_i64:
mask = arg_info(op->args[2])->mask | 63;
break;
case INDEX_op_ctpop_i32:
mask = 32 | 31;
break;
case INDEX_op_ctpop_i64:
mask = 64 | 63;
break;
CASE_OP_32_64(setcond):
case INDEX_op_setcond2_i32:
mask = 1;
break;
CASE_OP_32_64(movcond):
mask = arg_info(op->args[3])->mask | arg_info(op->args[4])->mask;
break;
CASE_OP_32_64(ld8u):
mask = 0xff;
break;
CASE_OP_32_64(ld16u):
mask = 0xffff;
break;
case INDEX_op_ld32u_i64:
mask = 0xffffffffu;
break;
CASE_OP_32_64(qemu_ld):
{
TCGMemOpIdx oi = op->args[nb_oargs + nb_iargs];
MemOp mop = get_memop(oi);
if (!(mop & MO_SIGN)) {
mask = (2ULL << ((8 << (mop & MO_SIZE)) - 1)) - 1;
}
}
break;
default:
break;
}
/* 32-bit ops generate 32-bit results. For the result is zero test
below, we can ignore high bits, but for further optimizations we
need to record that the high bits contain garbage. */
partmask = mask;
if (!(def->flags & TCG_OPF_64BIT)) {
mask |= ~(tcg_target_ulong)0xffffffffu;
partmask &= 0xffffffffu;
affected &= 0xffffffffu;
}
if (partmask == 0) {
tcg_debug_assert(nb_oargs == 1);
tcg_opt_gen_movi(s, op, op->args[0], 0);
optimize: optimize using nonzero bits This adds two optimizations using the non-zero bit mask. In some cases involving shifts or ANDs the value can become zero, and can thus be optimized to a move of zero. Second, useless zero-extension or an AND with constant can be detected that would only zero bits that are already zero. The main advantage of this optimization is that it turns zero-extensions into moves, thus enabling much better copy propagation (around 1% code reduction). Here is for example a "test $0xff0000,%ecx + je" before optimization: mov_i64 tmp0,rcx movi_i64 tmp1,$0xff0000 discard cc_src and_i64 cc_dst,tmp0,tmp1 movi_i32 cc_op,$0x1c ext32u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,eq,$0x0 and after (without patch on the left, with on the right): movi_i64 tmp1,$0xff0000 movi_i64 tmp1,$0xff0000 discard cc_src discard cc_src and_i64 cc_dst,rcx,tmp1 and_i64 cc_dst,rcx,tmp1 movi_i32 cc_op,$0x1c movi_i32 cc_op,$0x1c ext32u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,eq,$0x0 brcond_i64 cc_dst,tmp12,eq,$0x0 Other similar cases: "test %eax, %eax + jne" where eax is already 32-bit (after optimization, without patch on the left, with on the right): discard cc_src discard cc_src mov_i64 cc_dst,rax mov_i64 cc_dst,rax movi_i32 cc_op,$0x1c movi_i32 cc_op,$0x1c ext32u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,ne,$0x0 brcond_i64 rax,tmp12,ne,$0x0 "test $0x1, %dl + je": movi_i64 tmp1,$0x1 movi_i64 tmp1,$0x1 discard cc_src discard cc_src and_i64 cc_dst,rdx,tmp1 and_i64 cc_dst,rdx,tmp1 movi_i32 cc_op,$0x1a movi_i32 cc_op,$0x1a ext8u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,eq,$0x0 brcond_i64 cc_dst,tmp12,eq,$0x0 In some cases TCG even outsmarts GCC. :) Here the input code has "and $0x2,%eax + movslq %eax,%rbx + test %rbx, %rbx" and the optimizer, thanks to copy propagation, does the following: movi_i64 tmp12,$0x2 movi_i64 tmp12,$0x2 and_i64 rax,rax,tmp12 and_i64 rax,rax,tmp12 mov_i64 cc_dst,rax mov_i64 cc_dst,rax ext32s_i64 tmp0,rax -> nop mov_i64 rbx,tmp0 -> mov_i64 rbx,cc_dst and_i64 cc_dst,rbx,rbx -> nop Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
2013-01-12 00:42:53 +01:00
continue;
}
if (affected == 0) {
tcg_debug_assert(nb_oargs == 1);
tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
optimize: optimize using nonzero bits This adds two optimizations using the non-zero bit mask. In some cases involving shifts or ANDs the value can become zero, and can thus be optimized to a move of zero. Second, useless zero-extension or an AND with constant can be detected that would only zero bits that are already zero. The main advantage of this optimization is that it turns zero-extensions into moves, thus enabling much better copy propagation (around 1% code reduction). Here is for example a "test $0xff0000,%ecx + je" before optimization: mov_i64 tmp0,rcx movi_i64 tmp1,$0xff0000 discard cc_src and_i64 cc_dst,tmp0,tmp1 movi_i32 cc_op,$0x1c ext32u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,eq,$0x0 and after (without patch on the left, with on the right): movi_i64 tmp1,$0xff0000 movi_i64 tmp1,$0xff0000 discard cc_src discard cc_src and_i64 cc_dst,rcx,tmp1 and_i64 cc_dst,rcx,tmp1 movi_i32 cc_op,$0x1c movi_i32 cc_op,$0x1c ext32u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,eq,$0x0 brcond_i64 cc_dst,tmp12,eq,$0x0 Other similar cases: "test %eax, %eax + jne" where eax is already 32-bit (after optimization, without patch on the left, with on the right): discard cc_src discard cc_src mov_i64 cc_dst,rax mov_i64 cc_dst,rax movi_i32 cc_op,$0x1c movi_i32 cc_op,$0x1c ext32u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,ne,$0x0 brcond_i64 rax,tmp12,ne,$0x0 "test $0x1, %dl + je": movi_i64 tmp1,$0x1 movi_i64 tmp1,$0x1 discard cc_src discard cc_src and_i64 cc_dst,rdx,tmp1 and_i64 cc_dst,rdx,tmp1 movi_i32 cc_op,$0x1a movi_i32 cc_op,$0x1a ext8u_i64 tmp0,cc_dst movi_i64 tmp12,$0x0 movi_i64 tmp12,$0x0 brcond_i64 tmp0,tmp12,eq,$0x0 brcond_i64 cc_dst,tmp12,eq,$0x0 In some cases TCG even outsmarts GCC. :) Here the input code has "and $0x2,%eax + movslq %eax,%rbx + test %rbx, %rbx" and the optimizer, thanks to copy propagation, does the following: movi_i64 tmp12,$0x2 movi_i64 tmp12,$0x2 and_i64 rax,rax,tmp12 and_i64 rax,rax,tmp12 mov_i64 cc_dst,rax mov_i64 cc_dst,rax ext32s_i64 tmp0,rax -> nop mov_i64 rbx,tmp0 -> mov_i64 rbx,cc_dst and_i64 cc_dst,rbx,rbx -> nop Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
2013-01-12 00:42:53 +01:00
continue;
}
/* Simplify expression for "op r, a, 0 => movi r, 0" cases */
switch (opc) {
CASE_OP_32_64_VEC(and):
CASE_OP_32_64_VEC(mul):
CASE_OP_32_64(muluh):
CASE_OP_32_64(mulsh):
if (arg_is_const(op->args[2])
&& arg_info(op->args[2])->val == 0) {
tcg_opt_gen_movi(s, op, op->args[0], 0);
continue;
}
break;
default:
break;
}
/* Simplify expression for "op r, a, a => mov r, a" cases */
switch (opc) {
CASE_OP_32_64_VEC(or):
CASE_OP_32_64_VEC(and):
if (args_are_copies(op->args[1], op->args[2])) {
tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
continue;
}
break;
default:
break;
}
/* Simplify expression for "op r, a, a => movi r, 0" cases */
switch (opc) {
CASE_OP_32_64_VEC(andc):
CASE_OP_32_64_VEC(sub):
CASE_OP_32_64_VEC(xor):
if (args_are_copies(op->args[1], op->args[2])) {
tcg_opt_gen_movi(s, op, op->args[0], 0);
continue;
}
break;
default:
break;
}
/* Propagate constants through copy operations and do constant
folding. Constants will be substituted to arguments by register
allocator where needed and possible. Also detect copies. */
switch (opc) {
CASE_OP_32_64_VEC(mov):
tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
break;
CASE_OP_32_64(movi):
case INDEX_op_dupi_vec:
tcg_opt_gen_movi(s, op, op->args[0], op->args[1]);
break;
case INDEX_op_dup_vec:
if (arg_is_const(op->args[1])) {
tmp = arg_info(op->args[1])->val;
tmp = dup_const(TCGOP_VECE(op), tmp);
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break;
}
goto do_default;
CASE_OP_32_64(not):
CASE_OP_32_64(neg):
CASE_OP_32_64(ext8s):
CASE_OP_32_64(ext8u):
CASE_OP_32_64(ext16s):
CASE_OP_32_64(ext16u):
CASE_OP_32_64(ctpop):
CASE_OP_32_64(bswap16):
CASE_OP_32_64(bswap32):
case INDEX_op_bswap64_i64:
case INDEX_op_ext32s_i64:
case INDEX_op_ext32u_i64:
case INDEX_op_ext_i32_i64:
case INDEX_op_extu_i32_i64:
case INDEX_op_extrl_i64_i32:
case INDEX_op_extrh_i64_i32:
if (arg_is_const(op->args[1])) {
tmp = do_constant_folding(opc, arg_info(op->args[1])->val, 0);
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break;
}
goto do_default;
CASE_OP_32_64(add):
CASE_OP_32_64(sub):
CASE_OP_32_64(mul):
CASE_OP_32_64(or):
CASE_OP_32_64(and):
CASE_OP_32_64(xor):
CASE_OP_32_64(shl):
CASE_OP_32_64(shr):
CASE_OP_32_64(sar):
CASE_OP_32_64(rotl):
CASE_OP_32_64(rotr):
CASE_OP_32_64(andc):
CASE_OP_32_64(orc):
CASE_OP_32_64(eqv):
CASE_OP_32_64(nand):
CASE_OP_32_64(nor):
CASE_OP_32_64(muluh):
CASE_OP_32_64(mulsh):
CASE_OP_32_64(div):
CASE_OP_32_64(divu):
CASE_OP_32_64(rem):
CASE_OP_32_64(remu):
if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) {
tmp = do_constant_folding(opc, arg_info(op->args[1])->val,
arg_info(op->args[2])->val);
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break;
}
goto do_default;
CASE_OP_32_64(clz):
CASE_OP_32_64(ctz):
if (arg_is_const(op->args[1])) {
TCGArg v = arg_info(op->args[1])->val;
if (v != 0) {
tmp = do_constant_folding(opc, v, 0);
tcg_opt_gen_movi(s, op, op->args[0], tmp);
} else {
tcg_opt_gen_mov(s, op, op->args[0], op->args[2]);
}
break;
}
goto do_default;
CASE_OP_32_64(deposit):
if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) {
tmp = deposit64(arg_info(op->args[1])->val,
op->args[3], op->args[4],
arg_info(op->args[2])->val);
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break;
}
goto do_default;
CASE_OP_32_64(extract):
if (arg_is_const(op->args[1])) {
tmp = extract64(arg_info(op->args[1])->val,
op->args[2], op->args[3]);
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break;
}
goto do_default;
CASE_OP_32_64(sextract):
if (arg_is_const(op->args[1])) {
tmp = sextract64(arg_info(op->args[1])->val,
op->args[2], op->args[3]);
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break;
}
goto do_default;
CASE_OP_32_64(extract2):
if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) {
TCGArg v1 = arg_info(op->args[1])->val;
TCGArg v2 = arg_info(op->args[2])->val;
if (opc == INDEX_op_extract2_i64) {
tmp = (v1 >> op->args[3]) | (v2 << (64 - op->args[3]));
} else {
tmp = (int32_t)(((uint32_t)v1 >> op->args[3]) |
((uint32_t)v2 << (32 - op->args[3])));
}
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break;
}
goto do_default;
CASE_OP_32_64(setcond):
tmp = do_constant_folding_cond(opc, op->args[1],
op->args[2], op->args[3]);
if (tmp != 2) {
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break;
}
goto do_default;
CASE_OP_32_64(brcond):
tmp = do_constant_folding_cond(opc, op->args[0],
op->args[1], op->args[2]);
if (tmp != 2) {
if (tmp) {
2017-07-19 20:32:24 +02:00
bitmap_zero(temps_used.l, nb_temps);
op->opc = INDEX_op_br;
op->args[0] = op->args[3];
} else {
tcg_op_remove(s, op);
}
break;
}
goto do_default;
CASE_OP_32_64(movcond):
tmp = do_constant_folding_cond(opc, op->args[1],
op->args[2], op->args[5]);
if (tmp != 2) {
tcg_opt_gen_mov(s, op, op->args[0], op->args[4-tmp]);
break;
}
if (arg_is_const(op->args[3]) && arg_is_const(op->args[4])) {
tcg_target_ulong tv = arg_info(op->args[3])->val;
tcg_target_ulong fv = arg_info(op->args[4])->val;
TCGCond cond = op->args[5];
if (fv == 1 && tv == 0) {
cond = tcg_invert_cond(cond);
} else if (!(tv == 1 && fv == 0)) {
goto do_default;
}
op->args[3] = cond;
op->opc = opc = (opc == INDEX_op_movcond_i32
? INDEX_op_setcond_i32
: INDEX_op_setcond_i64);
nb_iargs = 2;
}
goto do_default;
case INDEX_op_add2_i32:
case INDEX_op_sub2_i32:
if (arg_is_const(op->args[2]) && arg_is_const(op->args[3])
&& arg_is_const(op->args[4]) && arg_is_const(op->args[5])) {
uint32_t al = arg_info(op->args[2])->val;
uint32_t ah = arg_info(op->args[3])->val;
uint32_t bl = arg_info(op->args[4])->val;
uint32_t bh = arg_info(op->args[5])->val;
uint64_t a = ((uint64_t)ah << 32) | al;
uint64_t b = ((uint64_t)bh << 32) | bl;
TCGArg rl, rh;
TCGOp *op2 = tcg_op_insert_before(s, op, INDEX_op_movi_i32);
if (opc == INDEX_op_add2_i32) {
a += b;
} else {
a -= b;
}
rl = op->args[0];
rh = op->args[1];
tcg_opt_gen_movi(s, op, rl, (int32_t)a);
tcg_opt_gen_movi(s, op2, rh, (int32_t)(a >> 32));
break;
}
goto do_default;
case INDEX_op_mulu2_i32:
if (arg_is_const(op->args[2]) && arg_is_const(op->args[3])) {
uint32_t a = arg_info(op->args[2])->val;
uint32_t b = arg_info(op->args[3])->val;
uint64_t r = (uint64_t)a * b;
TCGArg rl, rh;
TCGOp *op2 = tcg_op_insert_before(s, op, INDEX_op_movi_i32);
rl = op->args[0];
rh = op->args[1];
tcg_opt_gen_movi(s, op, rl, (int32_t)r);
tcg_opt_gen_movi(s, op2, rh, (int32_t)(r >> 32));
break;
}
goto do_default;
case INDEX_op_brcond2_i32:
tmp = do_constant_folding_cond2(&op->args[0], &op->args[2],
op->args[4]);
if (tmp != 2) {
if (tmp) {
do_brcond_true:
2017-07-19 20:32:24 +02:00
bitmap_zero(temps_used.l, nb_temps);
op->opc = INDEX_op_br;
op->args[0] = op->args[5];
} else {
do_brcond_false:
tcg_op_remove(s, op);
}
} else if ((op->args[4] == TCG_COND_LT
|| op->args[4] == TCG_COND_GE)
&& arg_is_const(op->args[2])
&& arg_info(op->args[2])->val == 0
&& arg_is_const(op->args[3])
&& arg_info(op->args[3])->val == 0) {
/* Simplify LT/GE comparisons vs zero to a single compare
vs the high word of the input. */
do_brcond_high:
2017-07-19 20:32:24 +02:00
bitmap_zero(temps_used.l, nb_temps);
op->opc = INDEX_op_brcond_i32;
op->args[0] = op->args[1];
op->args[1] = op->args[3];
op->args[2] = op->args[4];
op->args[3] = op->args[5];
} else if (op->args[4] == TCG_COND_EQ) {
/* Simplify EQ comparisons where one of the pairs
can be simplified. */
tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
op->args[0], op->args[2],
TCG_COND_EQ);
if (tmp == 0) {
goto do_brcond_false;
} else if (tmp == 1) {
goto do_brcond_high;
}
tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
op->args[1], op->args[3],
TCG_COND_EQ);
if (tmp == 0) {
goto do_brcond_false;
} else if (tmp != 1) {
goto do_default;
}
do_brcond_low:
2017-07-19 20:32:24 +02:00
bitmap_zero(temps_used.l, nb_temps);
op->opc = INDEX_op_brcond_i32;
op->args[1] = op->args[2];
op->args[2] = op->args[4];
op->args[3] = op->args[5];
} else if (op->args[4] == TCG_COND_NE) {
/* Simplify NE comparisons where one of the pairs
can be simplified. */
tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
op->args[0], op->args[2],
TCG_COND_NE);
if (tmp == 0) {
goto do_brcond_high;
} else if (tmp == 1) {
goto do_brcond_true;
}
tmp = do_constant_folding_cond(INDEX_op_brcond_i32,
op->args[1], op->args[3],
TCG_COND_NE);
if (tmp == 0) {
goto do_brcond_low;
} else if (tmp == 1) {
goto do_brcond_true;
}
goto do_default;
} else {
goto do_default;
}
break;
case INDEX_op_setcond2_i32:
tmp = do_constant_folding_cond2(&op->args[1], &op->args[3],
op->args[5]);
if (tmp != 2) {
do_setcond_const:
tcg_opt_gen_movi(s, op, op->args[0], tmp);
} else if ((op->args[5] == TCG_COND_LT
|| op->args[5] == TCG_COND_GE)
&& arg_is_const(op->args[3])
&& arg_info(op->args[3])->val == 0
&& arg_is_const(op->args[4])
&& arg_info(op->args[4])->val == 0) {
/* Simplify LT/GE comparisons vs zero to a single compare
vs the high word of the input. */
do_setcond_high:
reset_temp(op->args[0]);
arg_info(op->args[0])->mask = 1;
op->opc = INDEX_op_setcond_i32;
op->args[1] = op->args[2];
op->args[2] = op->args[4];
op->args[3] = op->args[5];
} else if (op->args[5] == TCG_COND_EQ) {
/* Simplify EQ comparisons where one of the pairs
can be simplified. */
tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
op->args[1], op->args[3],
TCG_COND_EQ);
if (tmp == 0) {
goto do_setcond_const;
} else if (tmp == 1) {
goto do_setcond_high;
}
tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
op->args[2], op->args[4],
TCG_COND_EQ);
if (tmp == 0) {
goto do_setcond_high;
} else if (tmp != 1) {
goto do_default;
}
do_setcond_low:
reset_temp(op->args[0]);
arg_info(op->args[0])->mask = 1;
op->opc = INDEX_op_setcond_i32;
op->args[2] = op->args[3];
op->args[3] = op->args[5];
} else if (op->args[5] == TCG_COND_NE) {
/* Simplify NE comparisons where one of the pairs
can be simplified. */
tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
op->args[1], op->args[3],
TCG_COND_NE);
if (tmp == 0) {
goto do_setcond_high;
} else if (tmp == 1) {
goto do_setcond_const;
}
tmp = do_constant_folding_cond(INDEX_op_setcond_i32,
op->args[2], op->args[4],
TCG_COND_NE);
if (tmp == 0) {
goto do_setcond_low;
} else if (tmp == 1) {
goto do_setcond_const;
}
goto do_default;
} else {
goto do_default;
}
break;
case INDEX_op_call:
if (!(op->args[nb_oargs + nb_iargs + 1]
& (TCG_CALL_NO_READ_GLOBALS | TCG_CALL_NO_WRITE_GLOBALS))) {
for (i = 0; i < nb_globals; i++) {
if (test_bit(i, temps_used.l)) {
reset_ts(&s->temps[i]);
}
}
}
goto do_reset_output;
default:
do_default:
/* Default case: we know nothing about operation (or were unable
to compute the operation result) so no propagation is done.
We trash everything if the operation is the end of a basic
block, otherwise we only trash the output args. "mask" is
the non-zero bits mask for the first output arg. */
if (def->flags & TCG_OPF_BB_END) {
2017-07-19 20:32:24 +02:00
bitmap_zero(temps_used.l, nb_temps);
} else {
do_reset_output:
for (i = 0; i < nb_oargs; i++) {
reset_temp(op->args[i]);
/* Save the corresponding known-zero bits mask for the
first output argument (only one supported so far). */
if (i == 0) {
arg_info(op->args[i])->mask = mask;
}
}
}
break;
}
/* Eliminate duplicate and redundant fence instructions. */
if (prev_mb) {
switch (opc) {
case INDEX_op_mb:
/* Merge two barriers of the same type into one,
* or a weaker barrier into a stronger one,
* or two weaker barriers into a stronger one.
* mb X; mb Y => mb X|Y
* mb; strl => mb; st
* ldaq; mb => ld; mb
* ldaq; strl => ld; mb; st
* Other combinations are also merged into a strong
* barrier. This is stricter than specified but for
* the purposes of TCG is better than not optimizing.
*/
prev_mb->args[0] |= op->args[0];
tcg_op_remove(s, op);
break;
default:
/* Opcodes that end the block stop the optimization. */
if ((def->flags & TCG_OPF_BB_END) == 0) {
break;
}
/* fallthru */
case INDEX_op_qemu_ld_i32:
case INDEX_op_qemu_ld_i64:
case INDEX_op_qemu_st_i32:
case INDEX_op_qemu_st_i64:
case INDEX_op_call:
/* Opcodes that touch guest memory stop the optimization. */
prev_mb = NULL;
break;
}
} else if (opc == INDEX_op_mb) {
prev_mb = op;
}
}
}