yuzu/unicorn
1
0
Fork 0
mirror of https://github.com/yuzu-emu/unicorn.git synced 2025-07-15 23:37:31 +00:00
Code Issues Packages Projects Releases Wiki Activity

tcg: Propagate args to op->args in optimizer

Browse source 
Backports commit acd937019bdaf933fcf1a7b57679ba07119c89b7 from qemu
This commit is contained in:
Richard Henderson 2018-03-05 05:46:48 -05:00 committed by Lioncash
parent eb488f5bd6
commit 845cfc2ae9
No known key found for this signature in database
GPG key ID: 4E3C3CC1031BA9C7
1 changed files with 225 additions and 201 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

426
qemu/tcg/optimize.c
View file

@ -167,8 +167,7 @@ static bool temps_are_copies(TCGContext *s, TCGArg arg1, TCGArg arg2)
return false; return false;
} }
static void tcg_opt_gen_movi(TCGContext *s, TCGOp *op, TCGArg *args, static void tcg_opt_gen_movi(TCGContext *s, TCGOp *op, TCGArg dst, TCGArg val)
TCGArg dst, TCGArg val)
{ {
struct tcg_temp_info *temps = s->temps2; struct tcg_temp_info *temps = s->temps2;
TCGOpcode new_op = op_to_movi(s, op->opc); TCGOpcode new_op = op_to_movi(s, op->opc);
@ -187,12 +186,11 @@ static void tcg_opt_gen_movi(TCGContext *s, TCGOp *op, TCGArg *args,
} }
temps[dst].mask = mask; temps[dst].mask = mask;
args[0] = dst; op->args[0] = dst;
args[1] = val; op->args[1] = val;
} }
static void tcg_opt_gen_mov(TCGContext *s, TCGOp *op, TCGArg *args, static void tcg_opt_gen_mov(TCGContext *s, TCGOp *op, TCGArg dst, TCGArg src)
TCGArg dst, TCGArg src)
{ {
if (temps_are_copies(s, dst, src)) { if (temps_are_copies(s, dst, src)) {
tcg_op_remove(s, op); tcg_op_remove(s, op);
@ -222,8 +220,8 @@ static void tcg_opt_gen_mov(TCGContext *s, TCGOp *op, TCGArg *args,
temps[dst].val = temps[src].val; temps[dst].val = temps[src].val;
} }
args[0] = dst; op->args[0] = dst;
args[1] = src; op->args[1] = src;
} }
static TCGArg do_constant_folding_2(TCGOpcode op, TCGArg x, TCGArg y) static TCGArg do_constant_folding_2(TCGOpcode op, TCGArg x, TCGArg y)
@ -570,7 +568,7 @@ void tcg_optimize(TCGContext *s)
{ {
struct tcg_temp_info *temps = s->temps2; struct tcg_temp_info *temps = s->temps2;
int oi, oi_next, nb_temps, nb_globals; int oi, oi_next, nb_temps, nb_globals;
TCGArg *prev_mb_args = NULL; TCGOp *prev_mb = NULL;
/* Array VALS has an element for each temp. /* 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 holds a constant then its value is kept in VALS' element.
@ -587,7 +585,6 @@ void tcg_optimize(TCGContext *s)
TCGArg tmp; TCGArg tmp;
TCGOp * const op = &s->gen_op_buf[oi]; TCGOp * const op = &s->gen_op_buf[oi];
TCGArg * const args = op->args;
TCGOpcode opc = op->opc; TCGOpcode opc = op->opc;
const TCGOpDef *def = &s->tcg_op_defs[opc]; const TCGOpDef *def = &s->tcg_op_defs[opc];
@ -599,7 +596,7 @@ void tcg_optimize(TCGContext *s)
nb_oargs = op->callo; nb_oargs = op->callo;
nb_iargs = op->calli; nb_iargs = op->calli;
for (i = 0; i < nb_oargs + nb_iargs; i++) { for (i = 0; i < nb_oargs + nb_iargs; i++) {
tmp = args[i]; tmp = op->args[i];
if (tmp != TCG_CALL_DUMMY_ARG) { if (tmp != TCG_CALL_DUMMY_ARG) {
init_temp_info(s, tmp); init_temp_info(s, tmp);
} }
@ -608,14 +605,14 @@ void tcg_optimize(TCGContext *s)
nb_oargs = def->nb_oargs; nb_oargs = def->nb_oargs;
nb_iargs = def->nb_iargs; nb_iargs = def->nb_iargs;
for (i = 0; i < nb_oargs + nb_iargs; i++) { for (i = 0; i < nb_oargs + nb_iargs; i++) {
init_temp_info(s, args[i]); init_temp_info(s, op->args[i]);
} }
} }
/* Do copy propagation */ /* Do copy propagation */
for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) { for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) {
if (temp_is_copy(s, args[i])) { if (temp_is_copy(s, op->args[i])) {
args[i] = find_better_copy(s, args[i]); op->args[i] = find_better_copy(s, op->args[i]);
} }
} }
@ -631,45 +628,45 @@ void tcg_optimize(TCGContext *s)
CASE_OP_32_64(nor): CASE_OP_32_64(nor):
CASE_OP_32_64(muluh): CASE_OP_32_64(muluh):
CASE_OP_32_64(mulsh): CASE_OP_32_64(mulsh):
swap_commutative(s, args[0], &args[1], &args[2]); swap_commutative(s, op->args[0], &op->args[1], &op->args[2]);
break; break;
CASE_OP_32_64(brcond): CASE_OP_32_64(brcond):
if (swap_commutative(s, -1, &args[0], &args[1])) { if (swap_commutative(s, -1, &op->args[0], &op->args[1])) {
args[2] = tcg_swap_cond(args[2]); op->args[2] = tcg_swap_cond(op->args[2]);
} }
break; break;
CASE_OP_32_64(setcond): CASE_OP_32_64(setcond):
if (swap_commutative(s, args[0], &args[1], &args[2])) { if (swap_commutative(s, op->args[0], &op->args[1], &op->args[2])) {
args[3] = tcg_swap_cond(args[3]); op->args[3] = tcg_swap_cond(op->args[3]);
} }
break; break;
CASE_OP_32_64(movcond): CASE_OP_32_64(movcond):
if (swap_commutative(s, -1, &args[1], &args[2])) { if (swap_commutative(s, -1, &op->args[1], &op->args[2])) {
args[5] = tcg_swap_cond(args[5]); op->args[5] = tcg_swap_cond(op->args[5]);
} }
/* For movcond, we canonicalize the "false" input reg to match /* For movcond, we canonicalize the "false" input reg to match
the destination reg so that the tcg backend can implement the destination reg so that the tcg backend can implement
a "move if true" operation. */ a "move if true" operation. */
if (swap_commutative(s, args[0], &args[4], &args[3])) { if (swap_commutative(s, op->args[0], &op->args[4], &op->args[3])) {
args[5] = tcg_invert_cond(args[5]); op->args[5] = tcg_invert_cond(op->args[5]);
} }
break; break;
CASE_OP_32_64(add2): CASE_OP_32_64(add2):
swap_commutative(s, args[0], &args[2], &args[4]); swap_commutative(s, op->args[0], &op->args[2], &op->args[4]);
swap_commutative(s, args[1], &args[3], &args[5]); swap_commutative(s, op->args[1], &op->args[3], &op->args[5]);
break; break;
CASE_OP_32_64(mulu2): CASE_OP_32_64(mulu2):
CASE_OP_32_64(muls2): CASE_OP_32_64(muls2):
swap_commutative(s, args[0], &args[2], &args[3]); swap_commutative(s, op->args[0], &op->args[2], &op->args[3]);
break; break;
case INDEX_op_brcond2_i32: case INDEX_op_brcond2_i32:
if (swap_commutative2(s, &args[0], &args[2])) { if (swap_commutative2(s, &op->args[0], &op->args[2])) {
args[4] = tcg_swap_cond(args[4]); op->args[4] = tcg_swap_cond(op->args[4]);
} }
break; break;
case INDEX_op_setcond2_i32: case INDEX_op_setcond2_i32:
if (swap_commutative2(s, &args[1], &args[3])) { if (swap_commutative2(s, &op->args[1], &op->args[3])) {
args[5] = tcg_swap_cond(args[5]); op->args[5] = tcg_swap_cond(op->args[5]);
} }
break; break;
default: default:
@ -684,8 +681,8 @@ void tcg_optimize(TCGContext *s)
CASE_OP_32_64(sar): CASE_OP_32_64(sar):
CASE_OP_32_64(rotl): CASE_OP_32_64(rotl):
CASE_OP_32_64(rotr): CASE_OP_32_64(rotr):
if (temp_is_const(s, args[1]) && temps[args[1]].val == 0) { if (temp_is_const(s, op->args[1]) && temps[op->args[1]].val == 0) {
tcg_opt_gen_movi(s, op, args, args[0], 0); tcg_opt_gen_movi(s, op, op->args[0], 0);
continue; continue;
} }
break; break;
@ -694,7 +691,7 @@ void tcg_optimize(TCGContext *s)
TCGOpcode neg_op; TCGOpcode neg_op;
bool have_neg; bool have_neg;
if (temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[2])) {
/* Proceed with possible constant folding. */ /* Proceed with possible constant folding. */
break; break;
} }
@ -708,40 +705,45 @@ void tcg_optimize(TCGContext *s)
if (!have_neg) { if (!have_neg) {
break; break;
} }
if (temp_is_const(s, args[1]) && temps[args[1]].val == 0) { if (temp_is_const(s, op->args[1])
&& temps[op->args[1]].val == 0) {
op->opc = neg_op; op->opc = neg_op;
reset_temp(s, args[0]); reset_temp(s, op->args[0]);
args[1] = args[2]; op->args[1] = op->args[2];
continue; continue;
} }
} }
break; break;
CASE_OP_32_64(xor): CASE_OP_32_64(xor):
CASE_OP_32_64(nand): CASE_OP_32_64(nand):
if (!temp_is_const(s, args[1]) if (!temp_is_const(s, op->args[1])
&& temp_is_const(s, args[2]) && temps[args[2]].val == -1) { && temp_is_const(s, op->args[2])
&& temps[op->args[2]].val == -1) {
i = 1; i = 1;
goto try_not; goto try_not;
} }
break; break;
CASE_OP_32_64(nor): CASE_OP_32_64(nor):
if (!temp_is_const(s, args[1]) if (!temp_is_const(s, op->args[1])
&& temp_is_const(s, args[2]) && temps[args[2]].val == 0) { && temp_is_const(s, op->args[2])
&& temps[op->args[2]].val == 0) {
i = 1; i = 1;
goto try_not; goto try_not;
} }
break; break;
CASE_OP_32_64(andc): CASE_OP_32_64(andc):
if (!temp_is_const(s, args[2]) if (!temp_is_const(s, op->args[2])
&& temp_is_const(s, args[1]) && temps[args[1]].val == -1) { && temp_is_const(s, op->args[1])
&& temps[op->args[1]].val == -1) {
i = 2; i = 2;
goto try_not; goto try_not;
} }
break; break;
CASE_OP_32_64(orc): CASE_OP_32_64(orc):
CASE_OP_32_64(eqv): CASE_OP_32_64(eqv):
if (!temp_is_const(s, args[2]) if (!temp_is_const(s, op->args[2])
&& temp_is_const(s, args[1]) && temps[args[1]].val == 0) { && temp_is_const(s, op->args[1])
&& temps[op->args[1]].val == 0) {
i = 2; i = 2;
goto try_not; goto try_not;
} }
@ -762,8 +764,8 @@ void tcg_optimize(TCGContext *s)
break; break;
} }
op->opc = not_op; op->opc = not_op;
reset_temp(s, args[0]); reset_temp(s, op->args[0]);
args[1] = args[i]; op->args[1] = op->args[i];
continue; continue;
} }
default: default:
@ -782,18 +784,20 @@ void tcg_optimize(TCGContext *s)
CASE_OP_32_64(or): CASE_OP_32_64(or):
CASE_OP_32_64(xor): CASE_OP_32_64(xor):
CASE_OP_32_64(andc): CASE_OP_32_64(andc):
if (!temp_is_const(s, args[1]) if (!temp_is_const(s, op->args[1])
&& temp_is_const(s, args[2]) && temps[args[2]].val == 0) { && temp_is_const(s, op->args[2])
tcg_opt_gen_mov(s, op, args, args[0], args[1]); && temps[op->args[2]].val == 0) {
tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
continue; continue;
} }
break; break;
CASE_OP_32_64(and): CASE_OP_32_64(and):
CASE_OP_32_64(orc): CASE_OP_32_64(orc):
CASE_OP_32_64(eqv): CASE_OP_32_64(eqv):
if (!temp_is_const(s, args[1]) if (!temp_is_const(s, op->args[1])
&& temp_is_const(s, args[2]) && temps[args[2]].val == -1) { && temp_is_const(s, op->args[2])
tcg_opt_gen_mov(s, op, args, args[0], args[1]); && temps[op->args[2]].val == -1) {
tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
continue; continue;
} }
break; break;
@ -807,21 +811,21 @@ void tcg_optimize(TCGContext *s)
affected = -1; affected = -1;
switch (opc) { switch (opc) {
CASE_OP_32_64(ext8s): CASE_OP_32_64(ext8s):
if ((temps[args[1]].mask & 0x80) != 0) { if ((temps[op->args[1]].mask & 0x80) != 0) {
break; break;
} }
CASE_OP_32_64(ext8u): CASE_OP_32_64(ext8u):
mask = 0xff; mask = 0xff;
goto and_const; goto and_const;
CASE_OP_32_64(ext16s): CASE_OP_32_64(ext16s):
if ((temps[args[1]].mask & 0x8000) != 0) { if ((temps[op->args[1]].mask & 0x8000) != 0) {
break; break;
} }
CASE_OP_32_64(ext16u): CASE_OP_32_64(ext16u):
mask = 0xffff; mask = 0xffff;
goto and_const; goto and_const;
case INDEX_op_ext32s_i64: case INDEX_op_ext32s_i64:
if ((temps[args[1]].mask & 0x80000000) != 0) { if ((temps[op->args[1]].mask & 0x80000000) != 0) {
break; break;
} }
case INDEX_op_ext32u_i64: case INDEX_op_ext32u_i64:
@ -829,110 +833,111 @@ void tcg_optimize(TCGContext *s)
goto and_const; goto and_const;
CASE_OP_32_64(and): CASE_OP_32_64(and):
mask = temps[args[2]].mask; mask = temps[op->args[2]].mask;
if (temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[2])) {
and_const: and_const:
affected = temps[args[1]].mask & ~mask; affected = temps[op->args[1]].mask & ~mask;
} }
mask = temps[args[1]].mask & mask; mask = temps[op->args[1]].mask & mask;
break; break;
case INDEX_op_ext_i32_i64: case INDEX_op_ext_i32_i64:
if ((temps[args[1]].mask & 0x80000000) != 0) { if ((temps[op->args[1]].mask & 0x80000000) != 0) {
break; break;
} }
case INDEX_op_extu_i32_i64: case INDEX_op_extu_i32_i64:
/* We do not compute affected as it is a size changing op. */ /* We do not compute affected as it is a size changing op. */
mask = (uint32_t)temps[args[1]].mask; mask = (uint32_t)temps[op->args[1]].mask;
break; break;
CASE_OP_32_64(andc): CASE_OP_32_64(andc):
/* Known-zeros does not imply known-ones. Therefore unless /* Known-zeros does not imply known-ones. Therefore unless
args[2] is constant, we can't infer anything from it. */ args[2] is constant, we can't infer anything from it. */
if (temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[2])) {
mask = ~temps[args[2]].mask; mask = ~temps[op->args[2]].mask;
goto and_const; goto and_const;
} }
/* But we certainly know nothing outside args[1] may be set. */ /* But we certainly know nothing outside args[1] may be set. */
mask = temps[args[1]].mask; mask = temps[op->args[1]].mask;
break; break;
case INDEX_op_sar_i32: case INDEX_op_sar_i32:
if (temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[2])) {
tmp = temps[args[2]].val & 31; tmp = temps[op->args[2]].val & 31;
mask = (int32_t)temps[args[1]].mask >> tmp; mask = (int32_t)temps[op->args[1]].mask >> tmp;
} }
break; break;
case INDEX_op_sar_i64: case INDEX_op_sar_i64:
if (temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[2])) {
tmp = temps[args[2]].val & 63; tmp = temps[op->args[2]].val & 63;
mask = (int64_t)temps[args[1]].mask >> tmp; mask = (int64_t)temps[op->args[1]].mask >> tmp;
} }
break; break;
case INDEX_op_shr_i32: case INDEX_op_shr_i32:
if (temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[2])) {
tmp = temps[args[2]].val & 31; tmp = temps[op->args[2]].val & 31;
mask = (uint32_t)temps[args[1]].mask >> tmp; mask = (uint32_t)temps[op->args[1]].mask >> tmp;
} }
break; break;
case INDEX_op_shr_i64: case INDEX_op_shr_i64:
if (temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[2])) {
tmp = temps[args[2]].val & 63; tmp = temps[op->args[2]].val & 63;
mask = (uint64_t)temps[args[1]].mask >> tmp; mask = (uint64_t)temps[op->args[1]].mask >> tmp;
} }
break; break;
case INDEX_op_extrl_i64_i32: case INDEX_op_extrl_i64_i32:
mask = (uint32_t)temps[args[1]].mask; mask = (uint32_t)temps[op->args[1]].mask;
break; break;
case INDEX_op_extrh_i64_i32: case INDEX_op_extrh_i64_i32:
mask = (uint64_t)temps[args[1]].mask >> 32; mask = (uint64_t)temps[op->args[1]].mask >> 32;
break; break;
CASE_OP_32_64(shl): CASE_OP_32_64(shl):
if (temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[2])) {
tmp = temps[args[2]].val & (TCG_TARGET_REG_BITS - 1); tmp = temps[op->args[2]].val & (TCG_TARGET_REG_BITS - 1);
mask = temps[args[1]].mask << tmp; mask = temps[op->args[1]].mask << tmp;
} }
break; break;
CASE_OP_32_64(neg): CASE_OP_32_64(neg):
/* Set to 1 all bits to the left of the rightmost. */ /* Set to 1 all bits to the left of the rightmost. */
mask = 0-(temps[args[1]].mask & (0-temps[args[1]].mask)); mask = 0-(temps[op->args[1]].mask & (0-temps[op->args[1]].mask));
break; break;
CASE_OP_32_64(deposit): CASE_OP_32_64(deposit):
mask = (tcg_target_ulong)deposit64(temps[args[1]].mask, args[3], args[4], mask = deposit64(temps[op->args[1]].mask, op->args[3],
temps[args[2]].mask); op->args[4], temps[op->args[2]].mask);
break; break;
CASE_OP_32_64(extract): CASE_OP_32_64(extract):
mask = extract64(temps[args[1]].mask, args[2], args[3]); mask = extract64(temps[op->args[1]].mask, op->args[2], op->args[3]);
if (args[2] == 0) { if (op->args[2] == 0) {
affected = temps[args[1]].mask & ~mask; affected = temps[op->args[1]].mask & ~mask;
} }
break; break;
CASE_OP_32_64(sextract): CASE_OP_32_64(sextract):
mask = sextract64(temps[args[1]].mask, args[2], args[3]); mask = sextract64(temps[op->args[1]].mask,
if (args[2] == 0 && (tcg_target_long)mask >= 0) { op->args[2], op->args[3]);
affected = temps[args[1]].mask & ~mask; if (op->args[2] == 0 && (tcg_target_long)mask >= 0) {
affected = temps[op->args[1]].mask & ~mask;
} }
break; break;
CASE_OP_32_64(or): CASE_OP_32_64(or):
CASE_OP_32_64(xor): CASE_OP_32_64(xor):
mask = temps[args[1]].mask | temps[args[2]].mask; mask = temps[op->args[1]].mask | temps[op->args[2]].mask;
break; break;
case INDEX_op_clz_i32: case INDEX_op_clz_i32:
case INDEX_op_ctz_i32: case INDEX_op_ctz_i32:
mask = temps[args[2]].mask | 31; mask = temps[op->args[2]].mask | 31;
break; break;
case INDEX_op_clz_i64: case INDEX_op_clz_i64:
case INDEX_op_ctz_i64: case INDEX_op_ctz_i64:
mask = temps[args[2]].mask | 63; mask = temps[op->args[2]].mask | 63;
break; break;
case INDEX_op_ctpop_i32: case INDEX_op_ctpop_i32:
@ -948,7 +953,7 @@ void tcg_optimize(TCGContext *s)
break; break;
CASE_OP_32_64(movcond): CASE_OP_32_64(movcond):
mask = temps[args[3]].mask | temps[args[4]].mask; mask = temps[op->args[3]].mask | temps[op->args[4]].mask;
break; break;
CASE_OP_32_64(ld8u): CASE_OP_32_64(ld8u):
@ -963,7 +968,7 @@ void tcg_optimize(TCGContext *s)
CASE_OP_32_64(qemu_ld): CASE_OP_32_64(qemu_ld):
{ {
TCGMemOpIdx oi = args[nb_oargs + nb_iargs]; TCGMemOpIdx oi = op->args[nb_oargs + nb_iargs];
TCGMemOp mop = get_memop(oi); TCGMemOp mop = get_memop(oi);
if (!(mop & MO_SIGN)) { if (!(mop & MO_SIGN)) {
mask = (2ULL << ((8 << (mop & MO_SIZE)) - 1)) - 1; mask = (2ULL << ((8 << (mop & MO_SIZE)) - 1)) - 1;
@ -987,12 +992,12 @@ void tcg_optimize(TCGContext *s)
if (partmask == 0) { if (partmask == 0) {
tcg_debug_assert(nb_oargs == 1); tcg_debug_assert(nb_oargs == 1);
tcg_opt_gen_movi(s, op, args, args[0], 0); tcg_opt_gen_movi(s, op, op->args[0], 0);
continue; continue;
} }
if (affected == 0) { if (affected == 0) {
tcg_debug_assert(nb_oargs == 1); tcg_debug_assert(nb_oargs == 1);
tcg_opt_gen_mov(s, op, args, args[0], args[1]); tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
continue; continue;
} }
@ -1002,8 +1007,8 @@ void tcg_optimize(TCGContext *s)
CASE_OP_32_64(mul): CASE_OP_32_64(mul):
CASE_OP_32_64(muluh): CASE_OP_32_64(muluh):
CASE_OP_32_64(mulsh): CASE_OP_32_64(mulsh):
if ((temp_is_const(s, args[2]) && temps[args[2]].val == 0)) { if ((temp_is_const(s, op->args[2]) && temps[op->args[2]].val == 0)) {
tcg_opt_gen_movi(s, op, args, args[0], 0); tcg_opt_gen_movi(s, op, op->args[0], 0);
continue; continue;
} }
break; break;
@ -1015,8 +1020,8 @@ void tcg_optimize(TCGContext *s)
switch (opc) { switch (opc) {
CASE_OP_32_64(or): CASE_OP_32_64(or):
CASE_OP_32_64(and): CASE_OP_32_64(and):
if (temps_are_copies(s, args[1], args[2])) { if (temps_are_copies(s, op->args[1], op->args[2])) {
tcg_opt_gen_mov(s, op, args, args[0], args[1]); tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
continue; continue;
} }
break; break;
@ -1029,8 +1034,8 @@ void tcg_optimize(TCGContext *s)
CASE_OP_32_64(andc): CASE_OP_32_64(andc):
CASE_OP_32_64(sub): CASE_OP_32_64(sub):
CASE_OP_32_64(xor): CASE_OP_32_64(xor):
if (temps_are_copies(s, args[1], args[2])) { if (temps_are_copies(s, op->args[1], op->args[2])) {
tcg_opt_gen_movi(s, op, args, args[0], 0); tcg_opt_gen_movi(s, op, op->args[0], 0);
continue; continue;
} }
break; break;
@ -1043,10 +1048,10 @@ void tcg_optimize(TCGContext *s)
allocator where needed and possible. Also detect copies. */ allocator where needed and possible. Also detect copies. */
switch (opc) { switch (opc) {
CASE_OP_32_64(mov): CASE_OP_32_64(mov):
tcg_opt_gen_mov(s, op, args, args[0], args[1]); tcg_opt_gen_mov(s, op, op->args[0], op->args[1]);
break; break;
CASE_OP_32_64(movi): CASE_OP_32_64(movi):
tcg_opt_gen_movi(s, op, args, args[0], args[1]); tcg_opt_gen_movi(s, op, op->args[0], op->args[1]);
break; break;
CASE_OP_32_64(not): CASE_OP_32_64(not):
@ -1062,9 +1067,9 @@ void tcg_optimize(TCGContext *s)
case INDEX_op_extu_i32_i64: case INDEX_op_extu_i32_i64:
case INDEX_op_extrl_i64_i32: case INDEX_op_extrl_i64_i32:
case INDEX_op_extrh_i64_i32: case INDEX_op_extrh_i64_i32:
if (temp_is_const(s, args[1])) { if (temp_is_const(s, op->args[1])) {
tmp = do_constant_folding(s, opc, temps[args[1]].val, 0); tmp = do_constant_folding(s, opc, temps[op->args[1]].val, 0);
tcg_opt_gen_movi(s, op, args, args[0], tmp); tcg_opt_gen_movi(s, op, op->args[0], tmp);
break; break;
} }
goto do_default; goto do_default;
@ -1091,68 +1096,72 @@ void tcg_optimize(TCGContext *s)
CASE_OP_32_64(divu): CASE_OP_32_64(divu):
CASE_OP_32_64(rem): CASE_OP_32_64(rem):
CASE_OP_32_64(remu): CASE_OP_32_64(remu):
if (temp_is_const(s, args[1]) && temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[1]) && temp_is_const(s, op->args[2])) {
tmp = do_constant_folding(s, opc, temps[args[1]].val, tmp = do_constant_folding(s, opc, temps[op->args[1]].val,
temps[args[2]].val); temps[op->args[2]].val);
tcg_opt_gen_movi(s, op, args, args[0], tmp); tcg_opt_gen_movi(s, op, op->args[0], tmp);
break; break;
} }
goto do_default; goto do_default;
CASE_OP_32_64(clz): CASE_OP_32_64(clz):
CASE_OP_32_64(ctz): CASE_OP_32_64(ctz):
if (temp_is_const(s, args[1])) { if (temp_is_const(s, op->args[1])) {
TCGArg v = temps[args[1]].val; TCGArg v = temps[op->args[1]].val;
if (v != 0) { if (v != 0) {
tmp = do_constant_folding(s, opc, v, 0); tmp = do_constant_folding(s, opc, v, 0);
tcg_opt_gen_movi(s, op, args, args[0], tmp); tcg_opt_gen_movi(s, op, op->args[0], tmp);
} else { } else {
tcg_opt_gen_mov(s, op, args, args[0], args[2]); tcg_opt_gen_mov(s, op, op->args[0], op->args[2]);
} }
break; break;
} }
goto do_default; goto do_default;
CASE_OP_32_64(deposit): CASE_OP_32_64(deposit):
if (temp_is_const(s, args[1]) && temp_is_const(s, args[2])) { if (temp_is_const(s, op->args[1]) && temp_is_const(s, op->args[2])) {
tmp = (TCGArg)deposit64(temps[args[1]].val, args[3], args[4], tmp = deposit64(temps[op->args[1]].val, op->args[3],
temps[args[2]].val); op->args[4], temps[op->args[2]].val);
tcg_opt_gen_movi(s, op, args, args[0], tmp); tcg_opt_gen_movi(s, op, op->args[0], tmp);
break; break;
} }
goto do_default; goto do_default;
CASE_OP_32_64(extract): CASE_OP_32_64(extract):
if (temp_is_const(s, args[1])) { if (temp_is_const(s, op->args[1])) {
tmp = extract64(temps[args[1]].val, args[2], args[3]); tmp = extract64(temps[op->args[1]].val,
tcg_opt_gen_movi(s, op, args, args[0], tmp); op->args[2], op->args[3]);
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break; break;
} }
goto do_default; goto do_default;
CASE_OP_32_64(sextract): CASE_OP_32_64(sextract):
if (temp_is_const(s, args[1])) { if (temp_is_const(s, op->args[1])) {
tmp = sextract64(temps[args[1]].val, args[2], args[3]); tmp = sextract64(temps[op->args[1]].val,
tcg_opt_gen_movi(s, op, args, args[0], tmp); op->args[2], op->args[3]);
tcg_opt_gen_movi(s, op, op->args[0], tmp);
break; break;
} }
goto do_default; goto do_default;
CASE_OP_32_64(setcond): CASE_OP_32_64(setcond):
tmp = do_constant_folding_cond(s, opc, args[1], args[2], args[3]); tmp = do_constant_folding_cond(s, opc, op->args[1],
op->args[2], op->args[3]);
if (tmp != 2) { if (tmp != 2) {
tcg_opt_gen_movi(s, op, args, args[0], tmp); tcg_opt_gen_movi(s, op, op->args[0], tmp);
break; break;
} }
goto do_default; goto do_default;
CASE_OP_32_64(brcond): CASE_OP_32_64(brcond):
tmp = do_constant_folding_cond(s, opc, args[0], args[1], args[2]); tmp = do_constant_folding_cond(s, opc, op->args[0],
op->args[1], op->args[2]);
if (tmp != 2) { if (tmp != 2) {
if (tmp) { if (tmp) {
reset_all_temps(s, nb_temps); reset_all_temps(s, nb_temps);
op->opc = INDEX_op_br; op->opc = INDEX_op_br;
args[0] = args[3]; op->args[0] = op->args[3];
} else { } else {
tcg_op_remove(s, op); tcg_op_remove(s, op);
} }
@ -1161,21 +1170,22 @@ void tcg_optimize(TCGContext *s)
goto do_default; goto do_default;
CASE_OP_32_64(movcond): CASE_OP_32_64(movcond):
tmp = do_constant_folding_cond(s, opc, args[1], args[2], args[5]); tmp = do_constant_folding_cond(s, opc, op->args[1],
op->args[2], op->args[5]);
if (tmp != 2) { if (tmp != 2) {
tcg_opt_gen_mov(s, op, args, args[0], args[4-tmp]); tcg_opt_gen_mov(s, op, op->args[0], op->args[4-tmp]);
break; break;
} }
if (temp_is_const(s, args[3]) && temp_is_const(s, args[4])) { if (temp_is_const(s, op->args[3]) && temp_is_const(s, op->args[4])) {
tcg_target_ulong tv = temps[args[3]].val; tcg_target_ulong tv = temps[op->args[3]].val;
tcg_target_ulong fv = temps[args[4]].val; tcg_target_ulong fv = temps[op->args[4]].val;
TCGCond cond = args[5]; TCGCond cond = op->args[5];
if (fv == 1 && tv == 0) { if (fv == 1 && tv == 0) {
cond = tcg_invert_cond(cond); cond = tcg_invert_cond(cond);
} else if (!(tv == 1 && fv == 0)) { } else if (!(tv == 1 && fv == 0)) {
goto do_default; goto do_default;
} }
args[3] = cond; op->args[3] = cond;
op->opc = opc = (opc == INDEX_op_movcond_i32 op->opc = opc = (opc == INDEX_op_movcond_i32
? INDEX_op_setcond_i32 ? INDEX_op_setcond_i32
: INDEX_op_setcond_i64); : INDEX_op_setcond_i64);
@ -1185,17 +1195,16 @@ void tcg_optimize(TCGContext *s)
case INDEX_op_add2_i32: case INDEX_op_add2_i32:
case INDEX_op_sub2_i32: case INDEX_op_sub2_i32:
if (temp_is_const(s, args[2]) && temp_is_const(s, args[3]) if (temp_is_const(s, op->args[2]) && temp_is_const(s, op->args[3])
&& temp_is_const(s, args[4]) && temp_is_const(s, args[5])) { && temp_is_const(s, op->args[4]) && temp_is_const(s, op->args[5])) {
uint32_t al = temps[args[2]].val; uint32_t al = temps[op->args[2]].val;
uint32_t ah = temps[args[3]].val; uint32_t ah = temps[op->args[3]].val;
uint32_t bl = temps[args[4]].val; uint32_t bl = temps[op->args[4]].val;
uint32_t bh = temps[args[5]].val; uint32_t bh = temps[op->args[5]].val;
uint64_t a = ((uint64_t)ah << 32) | al; uint64_t a = ((uint64_t)ah << 32) | al;
uint64_t b = ((uint64_t)bh << 32) | bl; uint64_t b = ((uint64_t)bh << 32) | bl;
TCGArg rl, rh; TCGArg rl, rh;
TCGOp *op2 = tcg_op_insert_before(s, op, INDEX_op_movi_i32, 2); TCGOp *op2 = tcg_op_insert_before(s, op, INDEX_op_movi_i32, 2);
TCGArg *args2 = op2->args;
if (opc == INDEX_op_add2_i32) { if (opc == INDEX_op_add2_i32) {
a += b; a += b;
@ -1203,10 +1212,10 @@ void tcg_optimize(TCGContext *s)
a -= b; a -= b;
} }
rl = args[0]; rl = op->args[0];
rh = args[1]; rh = op->args[1];
tcg_opt_gen_movi(s, op, args, rl, (int32_t)a); tcg_opt_gen_movi(s, op, rl, (int32_t)a);
tcg_opt_gen_movi(s, op2, args2, rh, (int32_t)(a >> 32)); tcg_opt_gen_movi(s, op2, rh, (int32_t)(a >> 32));
/* We've done all we need to do with the movi. Skip it. */ /* We've done all we need to do with the movi. Skip it. */
oi_next = op2->next; oi_next = op2->next;
@ -1215,18 +1224,17 @@ void tcg_optimize(TCGContext *s)
goto do_default; goto do_default;
case INDEX_op_mulu2_i32: case INDEX_op_mulu2_i32:
if (temp_is_const(s, args[2]) && temp_is_const(s, args[3])) { if (temp_is_const(s, op->args[2]) && temp_is_const(s, op->args[3])) {
uint32_t a = temps[args[2]].val; uint32_t a = temps[op->args[2]].val;
uint32_t b = temps[args[3]].val; uint32_t b = temps[op->args[3]].val;
uint64_t r = (uint64_t)a * b; uint64_t r = (uint64_t)a * b;
TCGArg rl, rh; TCGArg rl, rh;
TCGOp *op2 = tcg_op_insert_before(s, op, INDEX_op_movi_i32, 2); TCGOp *op2 = tcg_op_insert_before(s, op, INDEX_op_movi_i32, 2);
TCGArg *args2 = op2->args;
rl = args[0]; rl = op->args[0];
rh = args[1]; rh = op->args[1];
tcg_opt_gen_movi(s, op, args, rl, (int32_t)r); tcg_opt_gen_movi(s, op, rl, (int32_t)r);
tcg_opt_gen_movi(s, op2, args2, rh, (int32_t)(r >> 32)); tcg_opt_gen_movi(s, op2, rh, (int32_t)(r >> 32));
/* We've done all we need to do with the movi. Skip it. */ /* We've done all we need to do with the movi. Skip it. */
oi_next = op2->next; oi_next = op2->next;
@ -1235,41 +1243,47 @@ void tcg_optimize(TCGContext *s)
goto do_default; goto do_default;
case INDEX_op_brcond2_i32: case INDEX_op_brcond2_i32:
tmp = do_constant_folding_cond2(s, &args[0], &args[2], args[4]); tmp = do_constant_folding_cond2(s, &op->args[0], &op->args[2],
op->args[4]);
if (tmp != 2) { if (tmp != 2) {
if (tmp) { if (tmp) {
do_brcond_true: do_brcond_true:
reset_all_temps(s, nb_temps); reset_all_temps(s, nb_temps);
op->opc = INDEX_op_br; op->opc = INDEX_op_br;
args[0] = args[5]; op->args[0] = op->args[5];
} else { } else {
do_brcond_false: do_brcond_false:
tcg_op_remove(s, op); tcg_op_remove(s, op);
} }
} else if ((args[4] == TCG_COND_LT || args[4] == TCG_COND_GE) } else if ((op->args[4] == TCG_COND_LT
&& temp_is_const(s, args[2]) && temps[args[2]].val == 0 || op->args[4] == TCG_COND_GE)
&& temp_is_const(s, args[3]) && temps[args[3]].val == 0) { && temp_is_const(s, op->args[2])
&& temps[op->args[2]].val == 0
&& temp_is_const(s, op->args[3])
&& temps[op->args[3]].val == 0) {
/* Simplify LT/GE comparisons vs zero to a single compare /* Simplify LT/GE comparisons vs zero to a single compare
vs the high word of the input. */ vs the high word of the input. */
do_brcond_high: do_brcond_high:
reset_all_temps(s, nb_temps); reset_all_temps(s, nb_temps);
op->opc = INDEX_op_brcond_i32; op->opc = INDEX_op_brcond_i32;
args[0] = args[1]; op->args[0] = op->args[1];
args[1] = args[3]; op->args[1] = op->args[3];
args[2] = args[4]; op->args[2] = op->args[4];
args[3] = args[5]; op->args[3] = op->args[5];
} else if (args[4] == TCG_COND_EQ) { } else if (op->args[4] == TCG_COND_EQ) {
/* Simplify EQ comparisons where one of the pairs /* Simplify EQ comparisons where one of the pairs
can be simplified. */ can be simplified. */
tmp = do_constant_folding_cond(s, INDEX_op_brcond_i32, tmp = do_constant_folding_cond(s, INDEX_op_brcond_i32,
args[0], args[2], TCG_COND_EQ); op->args[0], op->args[2],
TCG_COND_EQ);
if (tmp == 0) { if (tmp == 0) {
goto do_brcond_false; goto do_brcond_false;
} else if (tmp == 1) { } else if (tmp == 1) {
goto do_brcond_high; goto do_brcond_high;
} }
tmp = do_constant_folding_cond(s, INDEX_op_brcond_i32, tmp = do_constant_folding_cond(s, INDEX_op_brcond_i32,
args[1], args[3], TCG_COND_EQ); op->args[1], op->args[3],
TCG_COND_EQ);
if (tmp == 0) { if (tmp == 0) {
goto do_brcond_false; goto do_brcond_false;
} else if (tmp != 1) { } else if (tmp != 1) {
@ -1278,21 +1292,23 @@ void tcg_optimize(TCGContext *s)
do_brcond_low: do_brcond_low:
reset_all_temps(s, nb_temps); reset_all_temps(s, nb_temps);
op->opc = INDEX_op_brcond_i32; op->opc = INDEX_op_brcond_i32;
args[1] = args[2]; op->args[1] = op->args[2];
args[2] = args[4]; op->args[2] = op->args[4];
args[3] = args[5]; op->args[3] = op->args[5];
} else if (args[4] == TCG_COND_NE) { } else if (op->args[4] == TCG_COND_NE) {
/* Simplify NE comparisons where one of the pairs /* Simplify NE comparisons where one of the pairs
can be simplified. */ can be simplified. */
tmp = do_constant_folding_cond(s, INDEX_op_brcond_i32, tmp = do_constant_folding_cond(s, INDEX_op_brcond_i32,
args[0], args[2], TCG_COND_NE); op->args[0], op->args[2],
TCG_COND_NE);
if (tmp == 0) { if (tmp == 0) {
goto do_brcond_high; goto do_brcond_high;
} else if (tmp == 1) { } else if (tmp == 1) {
goto do_brcond_true; goto do_brcond_true;
} }
tmp = do_constant_folding_cond(s, INDEX_op_brcond_i32, tmp = do_constant_folding_cond(s, INDEX_op_brcond_i32,
args[1], args[3], TCG_COND_NE); op->args[1], op->args[3],
TCG_COND_NE);
if (tmp == 0) { if (tmp == 0) {
goto do_brcond_low; goto do_brcond_low;
} else if (tmp == 1) { } else if (tmp == 1) {
@ -1305,57 +1321,65 @@ void tcg_optimize(TCGContext *s)
break; break;
case INDEX_op_setcond2_i32: case INDEX_op_setcond2_i32:
tmp = do_constant_folding_cond2(s, &args[1], &args[3], args[5]); tmp = do_constant_folding_cond2(s, &op->args[1], &op->args[3],
op->args[5]);
if (tmp != 2) { if (tmp != 2) {
do_setcond_const: do_setcond_const:
tcg_opt_gen_movi(s, op, args, args[0], tmp); tcg_opt_gen_movi(s, op, op->args[0], tmp);
} else if ((args[5] == TCG_COND_LT || args[5] == TCG_COND_GE) } else if ((op->args[5] == TCG_COND_LT
&& temp_is_const(s, args[3]) && temps[args[3]].val == 0 || op->args[5] == TCG_COND_GE)
&& temp_is_const(s, args[4]) && temps[args[4]].val == 0) { && temp_is_const(s, op->args[3])
&& temps[op->args[3]].val == 0
&& temp_is_const(s, op->args[4])
&& temps[op->args[4]].val == 0) {
/* Simplify LT/GE comparisons vs zero to a single compare /* Simplify LT/GE comparisons vs zero to a single compare
vs the high word of the input. */ vs the high word of the input. */
do_setcond_high: do_setcond_high:
reset_temp(s, args[0]); reset_temp(s, op->args[0]);
temps[args[0]].mask = 1; temps[op->args[0]].mask = 1;
op->opc = INDEX_op_setcond_i32; op->opc = INDEX_op_setcond_i32;
args[1] = args[2]; op->args[1] = op->args[2];
args[2] = args[4]; op->args[2] = op->args[4];
args[3] = args[5]; op->args[3] = op->args[5];
} else if (args[5] == TCG_COND_EQ) { } else if (op->args[5] == TCG_COND_EQ) {
/* Simplify EQ comparisons where one of the pairs /* Simplify EQ comparisons where one of the pairs
can be simplified. */ can be simplified. */
tmp = do_constant_folding_cond(s, INDEX_op_setcond_i32, tmp = do_constant_folding_cond(s, INDEX_op_setcond_i32,
args[1], args[3], TCG_COND_EQ); op->args[1], op->args[3],
TCG_COND_EQ);
if (tmp == 0) { if (tmp == 0) {
goto do_setcond_const; goto do_setcond_const;
} else if (tmp == 1) { } else if (tmp == 1) {
goto do_setcond_high; goto do_setcond_high;
} }
tmp = do_constant_folding_cond(s, INDEX_op_setcond_i32, tmp = do_constant_folding_cond(s, INDEX_op_setcond_i32,
args[2], args[4], TCG_COND_EQ); op->args[2], op->args[4],
TCG_COND_EQ);
if (tmp == 0) { if (tmp == 0) {
goto do_setcond_high; goto do_setcond_high;
} else if (tmp != 1) { } else if (tmp != 1) {
goto do_default; goto do_default;
} }
do_setcond_low: do_setcond_low:
reset_temp(s, args[0]); reset_temp(s, op->args[0]);
temps[args[0]].mask = 1; temps[op->args[0]].mask = 1;
op->opc = INDEX_op_setcond_i32; op->opc = INDEX_op_setcond_i32;
args[2] = args[3]; op->args[2] = op->args[3];
args[3] = args[5]; op->args[3] = op->args[5];
} else if (args[5] == TCG_COND_NE) { } else if (op->args[5] == TCG_COND_NE) {
/* Simplify NE comparisons where one of the pairs /* Simplify NE comparisons where one of the pairs
can be simplified. */ can be simplified. */
tmp = do_constant_folding_cond(s, INDEX_op_setcond_i32, tmp = do_constant_folding_cond(s, INDEX_op_setcond_i32,
args[1], args[3], TCG_COND_NE); op->args[1], op->args[3],
TCG_COND_NE);
if (tmp == 0) { if (tmp == 0) {
goto do_setcond_high; goto do_setcond_high;
} else if (tmp == 1) { } else if (tmp == 1) {
goto do_setcond_const; goto do_setcond_const;
} }
tmp = do_constant_folding_cond(s, INDEX_op_setcond_i32, tmp = do_constant_folding_cond(s, INDEX_op_setcond_i32,
args[2], args[4], TCG_COND_NE); op->args[2], op->args[4],
TCG_COND_NE);
if (tmp == 0) { if (tmp == 0) {
goto do_setcond_low; goto do_setcond_low;
} else if (tmp == 1) { } else if (tmp == 1) {
@ -1368,7 +1392,7 @@ void tcg_optimize(TCGContext *s)
break; break;
case INDEX_op_call: case INDEX_op_call:
if (!(args[nb_oargs + nb_iargs + 1] if (!(op->args[nb_oargs + nb_iargs + 1]
& (TCG_CALL_NO_READ_GLOBALS | TCG_CALL_NO_WRITE_GLOBALS))) { & (TCG_CALL_NO_READ_GLOBALS | TCG_CALL_NO_WRITE_GLOBALS))) {
for (i = 0; i < nb_globals; i++) { for (i = 0; i < nb_globals; i++) {
if (test_bit(i, s->temps2_used.l)) { if (test_bit(i, s->temps2_used.l)) {
@ -1390,11 +1414,11 @@ void tcg_optimize(TCGContext *s)
} else { } else {
do_reset_output: do_reset_output:
for (i = 0; i < nb_oargs; i++) { for (i = 0; i < nb_oargs; i++) {
reset_temp(s, args[i]); reset_temp(s, op->args[i]);
/* Save the corresponding known-zero bits mask for the /* Save the corresponding known-zero bits mask for the
first output argument (only one supported so far). */ first output argument (only one supported so far). */
if (i == 0) { if (i == 0) {
temps[args[i]].mask = mask; temps[op->args[i]].mask = mask;
} }
} }
} }
@ -1402,7 +1426,7 @@ void tcg_optimize(TCGContext *s)
} }
/* Eliminate duplicate and redundant fence instructions. */ /* Eliminate duplicate and redundant fence instructions. */
if (prev_mb_args) { if (prev_mb) {
switch (opc) { switch (opc) {
case INDEX_op_mb: case INDEX_op_mb:
/* Merge two barriers of the same type into one, /* Merge two barriers of the same type into one,
@ -1416,7 +1440,7 @@ void tcg_optimize(TCGContext *s)
* barrier. This is stricter than specified but for * barrier. This is stricter than specified but for
* the purposes of TCG is better than not optimizing. * the purposes of TCG is better than not optimizing.
*/ */
prev_mb_args[0] |= args[0]; prev_mb->args[0] |= op->args[0];
tcg_op_remove(s, op); tcg_op_remove(s, op);
break; break;
@ -1432,11 +1456,11 @@ void tcg_optimize(TCGContext *s)
case INDEX_op_qemu_st_i64: case INDEX_op_qemu_st_i64:
case INDEX_op_call: case INDEX_op_call:
/* Opcodes that touch guest memory stop the optimization. */ /* Opcodes that touch guest memory stop the optimization. */
prev_mb_args = NULL; prev_mb = NULL;
break; break;
} }
} else if (opc == INDEX_op_mb) { } else if (opc == INDEX_op_mb) {
prev_mb_args = args; prev_mb = op;
} }
} }
} }