2 * Copyright (C) 1995-2007 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief Machine dependent Firm optimizations.
24 * @author Sebastian Hack, Michael Beck
27 * Implements "Strenght Reduction of Multiplications by Integer Constants" by Youfeng Wu.
28 * Implements Division and Modulo by Consts from "Hackers Delight",
41 #include "irgraph_t.h"
48 #include "dbginfo_t.h"
49 #include "iropt_dbg.h"
60 /* when we need verifying */
62 # define IRN_VRFY_IRG(res, irg)
64 # define IRN_VRFY_IRG(res, irg) irn_vrfy_irg(res, irg)
67 /** The params got from the factory in arch_dep_init(...). */
68 static const ir_settings_arch_dep_t *params = NULL;
70 /** The bit mask, which optimizations to apply. */
71 static arch_dep_opts_t opts;
74 * construct a Mulh: Mulh(a,b) = (a * b) >> w, w is the with in bits of a, b
77 new_rd_Mulh (dbg_info *db, ir_graph *irg, ir_node *block,
78 ir_node *op1, ir_node *op2, ir_mode *mode) {
84 res = new_ir_node(db, irg, block, op_Mulh, mode, 2, in);
85 res = optimize_node(res);
86 IRN_VRFY_IRG(res, irg);
90 void arch_dep_init(arch_dep_params_factory_t factory) {
97 void arch_dep_set_opts(arch_dep_opts_t the_opts) {
100 if (opts & arch_dep_mul_to_shift)
101 set_opt_arch_dep_running(1);
104 /** check, whether a mode allows a Mulh instruction. */
105 static int allow_Mulh(ir_mode *mode) {
106 if (get_mode_size_bits(mode) > params->max_bits_for_mulh)
108 return (mode_is_signed(mode) && params->allow_mulhs) || (!mode_is_signed(mode) && params->allow_mulhu);
114 typedef struct instruction instruction;
116 insn_kind kind; /**< the instruction kind */
117 instruction *in[2]; /**< the ins */
118 unsigned shift_count; /**< shift count for LEA and SHIFT */
119 ir_node *irn; /**< the generated node for this instruction if any. */
120 int costs; /**< the costs for this instruction */
124 * The environment for the strength reduction of multiplications.
126 typedef struct _mul_env {
127 struct obstack obst; /**< an obstack for local space. */
128 ir_mode *mode; /**< the mode of the multiplication constant */
129 unsigned bits; /**< number of bits in the mode */
130 unsigned max_S; /**< the maximum LEA shift value. */
131 instruction *root; /**< the root of the instruction tree */
132 ir_node *op; /**< the operand that is multiplied */
133 ir_node *blk; /**< the block where the new graph is built */
134 dbg_info *dbg; /**< the debug info for the new graph. */
135 ir_mode *shf_mode; /**< the (unsigned) mode for the shift constants */
136 int fail; /**< set to 1 if the instruction sequence fails the constraints */
137 int n_shift; /**< maximum number of allowed shift instructions */
139 evaluate_costs_func evaluate; /**< the evaluate callback */
143 * Some kind of default evaluator. Return the cost of
146 static int default_evaluate(insn_kind kind, tarval *tv) {
155 * emit a LEA (or an Add) instruction
157 static instruction *emit_LEA(mul_env *env, instruction *a, instruction *b, unsigned shift) {
158 instruction *res = obstack_alloc(&env->obst, sizeof(*res));
159 res->kind = shift > 0 ? LEA : ADD;
162 res->shift_count = shift;
169 * emit a SHIFT (or an Add or a Zero) instruction
171 static instruction *emit_SHIFT(mul_env *env, instruction *a, unsigned shift) {
172 instruction *res = obstack_alloc(&env->obst, sizeof(*res));
173 if (shift == env->bits) {
174 /* a 2^bits with bits resolution is a zero */
178 res->shift_count = 0;
179 } else if (shift != 1) {
183 res->shift_count = shift;
188 res->shift_count = 0;
196 * emit a SUB instruction
198 static instruction *emit_SUB(mul_env *env, instruction *a, instruction *b) {
199 instruction *res = obstack_alloc(&env->obst, sizeof(*res));
203 res->shift_count = 0;
210 * emit the ROOT instruction
212 static instruction *emit_ROOT(mul_env *env, ir_node *root_op) {
213 instruction *res = obstack_alloc(&env->obst, sizeof(*res));
217 res->shift_count = 0;
225 * Returns the condensed representation of the tarval tv
227 static unsigned char *value_to_condensed(mul_env *env, tarval *tv, int *pr) {
228 ir_mode *mode = get_tarval_mode(tv);
229 int bits = get_mode_size_bits(mode);
230 char *bitstr = get_tarval_bitpattern(tv);
232 unsigned char *R = obstack_alloc(&env->obst, bits);
235 for (i = 0; bitstr[i] != '\0'; ++i) {
236 if (bitstr[i] == '1') {
249 * Calculate the gain when using the generalized complementary technique
251 static int calculate_gain(unsigned char *R, int r) {
256 /* the gain for r == 1 */
258 for (i = 2; i < r; ++i) {
259 /* calculate the gain for r from the gain for r-1 */
260 gain += 2 - R[i - 1];
262 if (gain > max_gain) {
273 * Calculates the condensed complement of a given (R,r) tuple
275 static unsigned char *complement_condensed(mul_env *env, unsigned char *R, int r, int gain, int *prs) {
276 unsigned char *value = obstack_alloc(&env->obst, env->bits);
280 memset(value, 0, env->bits);
283 for (i = 0; i < gain; ++i) {
288 /* negate and propagate 1 */
290 for (i = 0; i <= j; ++i) {
291 unsigned char v = !value[i];
297 /* condense it again */
300 for (i = 0; i <= j; ++i) {
313 * creates a tarval from a condensed representation.
315 static tarval *condensed_to_value(mul_env *env, unsigned char *R, int r) {
320 tv = get_mode_one(env->mode);
322 for (i = 0; i < r; ++i) {
325 tarval *t = new_tarval_from_long(j, mode_Iu);
326 tv = tarval_shl(tv, t);
328 res = res ? tarval_add(res, tv) : tv;
334 static instruction *basic_decompose_mul(mul_env *env, unsigned char *R, int r, tarval *N);
337 * handle simple cases with up-to 2 bits set
339 static instruction *decompose_simple_cases(mul_env *env, unsigned char *R, int r, tarval *N) {
340 instruction *ins, *ins2;
344 return emit_SHIFT(env, env->root, R[0]);
350 ins = emit_SHIFT(env, ins, R[0]);
352 if (R[1] <= env->max_S)
353 return emit_LEA(env, ins, ins, R[1]);
355 ins2 = emit_SHIFT(env, env->root, R[0] + R[1]);
356 return emit_LEA(env, ins, ins2, 0);
361 * Main decompose driver.
363 static instruction *decompose_mul(mul_env *env, unsigned char *R, int r, tarval *N) {
368 return decompose_simple_cases(env, R, r, N);
370 if (params->also_use_subs) {
371 gain = calculate_gain(R, r);
373 instruction *instr1, *instr2;
374 unsigned char *R1, *R2;
377 R1 = complement_condensed(env, R, r, gain, &r1);
379 R2 = obstack_alloc(&env->obst, r2);
382 for (i = 0; i < gain; ++i) {
389 /* Two identical bits: normalize */
394 for (i = gain + 1; i < r; ++i) {
398 instr1 = decompose_mul(env, R1, r1, NULL);
399 instr2 = decompose_mul(env, R2, r2, NULL);
400 return emit_SUB(env, instr2, instr1);
405 N = condensed_to_value(env, R, r);
407 for (i = env->max_S; i > 0; --i) {
408 tarval *div_res, *mod_res;
409 tarval *tv = new_tarval_from_long((1 << i) + 1, env->mode);
411 div_res = tarval_divmod(N, tv, &mod_res);
412 if (mod_res == get_mode_null(env->mode)) {
416 Rs = value_to_condensed(env, div_res, &rs);
418 instruction *N1 = decompose_mul(env, Rs, rs, div_res);
419 return emit_LEA(env, N1, N1, i);
423 return basic_decompose_mul(env, R, r, N);
426 #define IMAX(a,b) ((a) > (b) ? (a) : (b))
429 * basic decomposition routine
431 static instruction *basic_decompose_mul(mul_env *env, unsigned char *R, int r, tarval *N) {
435 if (R[0] == 0) { /* Case 1 */
436 t = R[1] > IMAX(env->max_S, R[1]);
438 Ns = decompose_mul(env, &R[1], r - 1, N);
439 return emit_LEA(env, env->root, Ns, t);
440 } else if (R[0] <= env->max_S) { /* Case 2 */
443 Ns = decompose_mul(env, &R[1], r - 1, N);
444 return emit_LEA(env, Ns, env->root, t);
448 Ns = decompose_mul(env, R, r, N);
449 return emit_SHIFT(env, Ns, t);
454 * recursive build the graph form the instructions.
456 * @param env the environment
457 * @param inst the instruction
459 static ir_node *build_graph(mul_env *env, instruction *inst) {
465 switch (inst->kind) {
467 l = build_graph(env, inst->in[0]);
468 r = build_graph(env, inst->in[1]);
469 c = new_r_Const(current_ir_graph, env->blk, env->shf_mode, new_tarval_from_long(inst->shift_count, env->shf_mode));
470 r = new_rd_Shl(env->dbg, current_ir_graph, env->blk, r, c, env->mode);
471 return inst->irn = new_rd_Add(env->dbg, current_ir_graph, env->blk, l, r, env->mode);
473 l = build_graph(env, inst->in[0]);
474 c = new_r_Const(current_ir_graph, env->blk, env->shf_mode, new_tarval_from_long(inst->shift_count, env->shf_mode));
475 return inst->irn = new_rd_Shl(env->dbg, current_ir_graph, env->blk, l, c, env->mode);
477 l = build_graph(env, inst->in[0]);
478 r = build_graph(env, inst->in[1]);
479 return inst->irn = new_rd_Sub(env->dbg, current_ir_graph, env->blk, l, r, env->mode);
481 l = build_graph(env, inst->in[0]);
482 r = build_graph(env, inst->in[1]);
483 return inst->irn = new_rd_Add(env->dbg, current_ir_graph, env->blk, l, r, env->mode);
485 return inst->irn = new_r_Const(current_ir_graph, env->blk, env->mode, get_mode_null(env->mode));
493 * Calculate the costs for the given instruction sequence.
494 * Note that additional costs due to higher register pressure are NOT evaluated yet
496 static int evaluate_insn(mul_env *env, instruction *inst) {
499 if (inst->costs >= 0) {
500 /* was already evaluated */
504 switch (inst->kind) {
508 costs = evaluate_insn(env, inst->in[0]);
509 costs += evaluate_insn(env, inst->in[1]);
510 costs += env->evaluate(inst->kind, NULL);
514 if (inst->shift_count > params->highest_shift_amount)
516 if (env->n_shift <= 0)
520 costs = evaluate_insn(env, inst->in[0]);
521 costs += env->evaluate(inst->kind, NULL);
525 inst->costs = costs = env->evaluate(inst->kind, NULL);
534 * Evaluate the replacement instructions and build a new graph
535 * if faster than the Mul.
536 * returns the root of the new graph then or irn otherwise.
538 * @param irn the Mul operation
539 * @param operand the multiplication operand
540 * @param tv the multiplication constant
542 * @return the new graph
544 static ir_node *do_decomposition(ir_node *irn, ir_node *operand, tarval *tv) {
552 obstack_init(&env.obst);
553 env.mode = get_tarval_mode(tv);
554 env.bits = (unsigned)get_mode_size_bits(env.mode);
556 env.root = emit_ROOT(&env, operand);
558 env.n_shift = params->maximum_shifts;
559 env.evaluate = params->evaluate != NULL ? params->evaluate : default_evaluate;
561 R = value_to_condensed(&env, tv, &r);
562 inst = decompose_mul(&env, R, r, tv);
564 /* the paper suggests 70% here */
565 mul_costs = (env.evaluate(MUL, tv) * 7) / 10;
566 if (evaluate_insn(&env, inst) <= mul_costs && !env.fail) {
568 env.blk = get_nodes_block(irn);
569 env.dbg = get_irn_dbg_info(irn);
570 env.shf_mode = find_unsigned_mode(env.mode);
571 if (env.shf_mode == NULL)
572 env.shf_mode = mode_Iu;
574 res = build_graph(&env, inst);
576 obstack_free(&env.obst, NULL);
580 /* Replace Muls with Shifts and Add/Subs. */
581 ir_node *arch_dep_replace_mul_with_shifts(ir_node *irn) {
583 ir_mode *mode = get_irn_mode(irn);
585 /* If the architecture dependent optimizations were not initialized
586 or this optimization was not enabled. */
587 if (params == NULL || (opts & arch_dep_mul_to_shift) == 0)
590 if (is_Mul(irn) && mode_is_int(mode)) {
591 ir_node *left = get_binop_left(irn);
592 ir_node *right = get_binop_right(irn);
594 ir_node *operand = NULL;
596 /* Look, if one operand is a constant. */
597 if (is_Const(left)) {
598 tv = get_Const_tarval(left);
600 } else if (is_Const(right)) {
601 tv = get_Const_tarval(right);
606 res = do_decomposition(irn, operand, tv);
609 hook_arch_dep_replace_mul_with_shifts(irn);
619 * calculated the ld2 of a tarval if tarval is 2^n, else returns -1.
621 static int tv_ld2(tarval *tv, int bits) {
624 for (num = i = 0; i < bits; ++i) {
625 unsigned char v = get_tarval_sub_bits(tv, i);
630 for (j = 0; j < 8; ++j)
643 /* for shorter lines */
644 #define ABS(a) tarval_abs(a)
645 #define NEG(a) tarval_neg(a)
646 #define NOT(a) tarval_not(a)
647 #define SHL(a, b) tarval_shl(a, b)
648 #define SHR(a, b) tarval_shr(a, b)
649 #define ADD(a, b) tarval_add(a, b)
650 #define SUB(a, b) tarval_sub(a, b)
651 #define MUL(a, b) tarval_mul(a, b)
652 #define DIV(a, b) tarval_div(a, b)
653 #define MOD(a, b) tarval_mod(a, b)
654 #define CMP(a, b) tarval_cmp(a, b)
655 #define CNV(a, m) tarval_convert_to(a, m)
656 #define ONE(m) get_mode_one(m)
657 #define ZERO(m) get_mode_null(m)
659 /** The result of a the magic() function. */
661 tarval *M; /**< magic number */
662 int s; /**< shift amount */
663 int need_add; /**< an additional add is needed */
664 int need_sub; /**< an additional sub is needed */
668 * Signed division by constant d: calculate the Magic multiplier M and the shift amount s
670 * see Hacker's Delight: 10-6 Integer Division by Constants: Incorporation into a Compiler
672 static struct ms magic(tarval *d) {
673 ir_mode *mode = get_tarval_mode(d);
674 ir_mode *u_mode = find_unsigned_mode(mode);
675 int bits = get_mode_size_bits(u_mode);
677 tarval *ad, *anc, *delta, *q1, *r1, *q2, *r2, *t; /* unsigned */
680 tarval *bits_minus_1, *two_bits_1;
684 tarval_int_overflow_mode_t rem = tarval_get_integer_overflow_mode();
686 /* we need overflow mode to work correctly */
687 tarval_set_integer_overflow_mode(TV_OVERFLOW_WRAP);
690 bits_minus_1 = new_tarval_from_long(bits - 1, u_mode);
691 two_bits_1 = SHL(get_mode_one(u_mode), bits_minus_1);
693 ad = CNV(ABS(d), u_mode);
694 t = ADD(two_bits_1, SHR(CNV(d, u_mode), bits_minus_1));
695 anc = SUB(SUB(t, ONE(u_mode)), MOD(t, ad)); /* Absolute value of nc */
696 p = bits - 1; /* Init: p */
697 q1 = DIV(two_bits_1, anc); /* Init: q1 = 2^p/|nc| */
698 r1 = SUB(two_bits_1, MUL(q1, anc)); /* Init: r1 = rem(2^p, |nc|) */
699 q2 = DIV(two_bits_1, ad); /* Init: q2 = 2^p/|d| */
700 r2 = SUB(two_bits_1, MUL(q2, ad)); /* Init: r2 = rem(2^p, |d|) */
704 q1 = ADD(q1, q1); /* Update q1 = 2^p/|nc| */
705 r1 = ADD(r1, r1); /* Update r1 = rem(2^p, |nc|) */
707 if (CMP(r1, anc) & pn_Cmp_Ge) {
708 q1 = ADD(q1, ONE(u_mode));
712 q2 = ADD(q2, q2); /* Update q2 = 2^p/|d| */
713 r2 = ADD(r2, r2); /* Update r2 = rem(2^p, |d|) */
715 if (CMP(r2, ad) & pn_Cmp_Ge) {
716 q2 = ADD(q2, ONE(u_mode));
721 } while (CMP(q1, delta) & pn_Cmp_Lt || (CMP(q1, delta) & pn_Cmp_Eq && CMP(r1, ZERO(u_mode)) & pn_Cmp_Eq));
723 d_cmp = CMP(d, ZERO(mode));
725 if (d_cmp & pn_Cmp_Ge)
726 mag.M = ADD(CNV(q2, mode), ONE(mode));
728 mag.M = SUB(ZERO(mode), ADD(CNV(q2, mode), ONE(mode)));
730 M_cmp = CMP(mag.M, ZERO(mode));
734 /* need an add if d > 0 && M < 0 */
735 mag.need_add = d_cmp & pn_Cmp_Gt && M_cmp & pn_Cmp_Lt;
737 /* need a sub if d < 0 && M > 0 */
738 mag.need_sub = d_cmp & pn_Cmp_Lt && M_cmp & pn_Cmp_Gt;
740 tarval_set_integer_overflow_mode(rem);
745 /** The result of the magicu() function. */
747 tarval *M; /**< magic add constant */
748 int s; /**< shift amount */
749 int need_add; /**< add indicator */
753 * Unsigned division by constant d: calculate the Magic multiplier M and the shift amount s
755 * see Hacker's Delight: 10-10 Integer Division by Constants: Incorporation into a Compiler (Unsigned)
757 static struct mu magicu(tarval *d) {
758 ir_mode *mode = get_tarval_mode(d);
759 int bits = get_mode_size_bits(mode);
761 tarval *nc, *delta, *q1, *r1, *q2, *r2;
762 tarval *bits_minus_1, *two_bits_1, *seven_ff;
766 tarval_int_overflow_mode_t rem = tarval_get_integer_overflow_mode();
768 /* we need overflow mode to work correctly */
769 tarval_set_integer_overflow_mode(TV_OVERFLOW_WRAP);
771 bits_minus_1 = new_tarval_from_long(bits - 1, mode);
772 two_bits_1 = SHL(get_mode_one(mode), bits_minus_1);
773 seven_ff = SUB(two_bits_1, ONE(mode));
775 magu.need_add = 0; /* initialize the add indicator */
776 nc = SUB(NEG(ONE(mode)), MOD(NEG(d), d));
777 p = bits - 1; /* Init: p */
778 q1 = DIV(two_bits_1, nc); /* Init: q1 = 2^p/nc */
779 r1 = SUB(two_bits_1, MUL(q1, nc)); /* Init: r1 = rem(2^p, nc) */
780 q2 = DIV(seven_ff, d); /* Init: q2 = (2^p - 1)/d */
781 r2 = SUB(seven_ff, MUL(q2, d)); /* Init: r2 = rem(2^p - 1, d) */
785 if (CMP(r1, SUB(nc, r1)) & pn_Cmp_Ge) {
786 q1 = ADD(ADD(q1, q1), ONE(mode));
787 r1 = SUB(ADD(r1, r1), nc);
794 if (CMP(ADD(r2, ONE(mode)), SUB(d, r2)) & pn_Cmp_Ge) {
795 if (CMP(q2, seven_ff) & pn_Cmp_Ge)
798 q2 = ADD(ADD(q2, q2), ONE(mode));
799 r2 = SUB(ADD(ADD(r2, r2), ONE(mode)), d);
802 if (CMP(q2, two_bits_1) & pn_Cmp_Ge)
806 r2 = ADD(ADD(r2, r2), ONE(mode));
808 delta = SUB(SUB(d, ONE(mode)), r2);
809 } while (p < 2*bits &&
810 (CMP(q1, delta) & pn_Cmp_Lt || (CMP(q1, delta) & pn_Cmp_Eq && CMP(r1, ZERO(mode)) & pn_Cmp_Eq)));
812 magu.M = ADD(q2, ONE(mode)); /* Magic number */
813 magu.s = p - bits; /* and shift amount */
815 tarval_set_integer_overflow_mode(rem);
821 * Build the Mulh replacement code for n / tv.
823 * Note that 'div' might be a mod or DivMod operation as well
825 static ir_node *replace_div_by_mulh(ir_node *div, tarval *tv) {
826 dbg_info *dbg = get_irn_dbg_info(div);
827 ir_node *n = get_binop_left(div);
828 ir_node *block = get_irn_n(div, -1);
829 ir_mode *mode = get_irn_mode(n);
830 int bits = get_mode_size_bits(mode);
833 /* Beware: do not transform bad code */
834 if (is_Bad(n) || is_Bad(block))
837 if (mode_is_signed(mode)) {
838 struct ms mag = magic(tv);
840 /* generate the Mulh instruction */
841 c = new_r_Const(current_ir_graph, block, mode, mag.M);
842 q = new_rd_Mulh(dbg, current_ir_graph, block, n, c, mode);
844 /* do we need an Add or Sub */
846 q = new_rd_Add(dbg, current_ir_graph, block, q, n, mode);
847 else if (mag.need_sub)
848 q = new_rd_Sub(dbg, current_ir_graph, block, q, n, mode);
850 /* Do we need the shift */
852 c = new_r_Const_long(current_ir_graph, block, mode_Iu, mag.s);
853 q = new_rd_Shrs(dbg, current_ir_graph, block, q, c, mode);
857 c = new_r_Const_long(current_ir_graph, block, mode_Iu, bits-1);
858 t = new_rd_Shr(dbg, current_ir_graph, block, q, c, mode);
860 q = new_rd_Add(dbg, current_ir_graph, block, q, t, mode);
862 struct mu mag = magicu(tv);
865 /* generate the Mulh instruction */
866 c = new_r_Const(current_ir_graph, block, mode, mag.M);
867 q = new_rd_Mulh(dbg, current_ir_graph, block, n, c, mode);
871 /* use the GM scheme */
872 t = new_rd_Sub(dbg, current_ir_graph, block, n, q, mode);
874 c = new_r_Const(current_ir_graph, block, mode_Iu, get_mode_one(mode_Iu));
875 t = new_rd_Shr(dbg, current_ir_graph, block, t, c, mode);
877 t = new_rd_Add(dbg, current_ir_graph, block, t, q, mode);
879 c = new_r_Const_long(current_ir_graph, block, mode_Iu, mag.s-1);
880 q = new_rd_Shr(dbg, current_ir_graph, block, t, c, mode);
882 /* use the default scheme */
883 q = new_rd_Add(dbg, current_ir_graph, block, q, n, mode);
885 } else if (mag.s > 0) { /* default scheme, shift needed */
886 c = new_r_Const_long(current_ir_graph, block, mode_Iu, mag.s);
887 q = new_rd_Shr(dbg, current_ir_graph, block, q, c, mode);
893 /* Replace Divs with Shifts and Add/Subs and Mulh. */
894 ir_node *arch_dep_replace_div_by_const(ir_node *irn) {
897 /* If the architecture dependent optimizations were not initialized
898 or this optimization was not enabled. */
899 if (params == NULL || (opts & arch_dep_div_by_const) == 0)
902 if (get_irn_opcode(irn) == iro_Div) {
903 ir_node *c = get_Div_right(irn);
904 ir_node *block, *left;
911 if (get_irn_op(c) != op_Const)
914 tv = get_Const_tarval(c);
916 /* check for division by zero */
917 if (classify_tarval(tv) == TV_CLASSIFY_NULL)
920 left = get_Div_left(irn);
921 mode = get_irn_mode(left);
922 block = get_irn_n(irn, -1);
923 dbg = get_irn_dbg_info(irn);
925 bits = get_mode_size_bits(mode);
929 if (mode_is_signed(mode)) {
930 /* for signed divisions, the algorithm works for a / -2^k by negating the result */
931 ntv = tarval_neg(tv);
941 if (k >= 0) { /* division by 2^k or -2^k */
942 if (mode_is_signed(mode)) {
944 ir_node *curr = left;
947 k_node = new_r_Const_long(current_ir_graph, block, mode_Iu, k - 1);
948 curr = new_rd_Shrs(dbg, current_ir_graph, block, left, k_node, mode);
951 k_node = new_r_Const_long(current_ir_graph, block, mode_Iu, bits - k);
952 curr = new_rd_Shr(dbg, current_ir_graph, block, curr, k_node, mode);
954 curr = new_rd_Add(dbg, current_ir_graph, block, left, curr, mode);
956 k_node = new_r_Const_long(current_ir_graph, block, mode_Iu, k);
957 res = new_rd_Shrs(dbg, current_ir_graph, block, curr, k_node, mode);
959 if (n_flag) { /* negate the result */
962 k_node = new_r_Const(current_ir_graph, block, mode, get_mode_null(mode));
963 res = new_rd_Sub(dbg, current_ir_graph, block, k_node, res, mode);
965 } else { /* unsigned case */
968 k_node = new_r_Const_long(current_ir_graph, block, mode_Iu, k);
969 res = new_rd_Shr(dbg, current_ir_graph, block, left, k_node, mode);
973 if (allow_Mulh(mode))
974 res = replace_div_by_mulh(irn, tv);
979 hook_arch_dep_replace_division_by_const(irn);
984 /* Replace Mods with Shifts and Add/Subs and Mulh. */
985 ir_node *arch_dep_replace_mod_by_const(ir_node *irn) {
988 /* If the architecture dependent optimizations were not initialized
989 or this optimization was not enabled. */
990 if (params == NULL || (opts & arch_dep_mod_by_const) == 0)
993 if (get_irn_opcode(irn) == iro_Mod) {
994 ir_node *c = get_Mod_right(irn);
995 ir_node *block, *left;
1002 if (get_irn_op(c) != op_Const)
1005 tv = get_Const_tarval(c);
1007 /* check for division by zero */
1008 if (classify_tarval(tv) == TV_CLASSIFY_NULL)
1011 left = get_Mod_left(irn);
1012 mode = get_irn_mode(left);
1013 block = get_irn_n(irn, -1);
1014 dbg = get_irn_dbg_info(irn);
1015 bits = get_mode_size_bits(mode);
1019 if (mode_is_signed(mode)) {
1020 /* for signed divisions, the algorithm works for a / -2^k by negating the result */
1021 ntv = tarval_neg(tv);
1030 /* division by 2^k or -2^k:
1031 * we use "modulus" here, so x % y == x % -y that's why is no difference between the case 2^k and -2^k
1033 if (mode_is_signed(mode)) {
1035 ir_node *curr = left;
1038 k_node = new_r_Const_long(current_ir_graph, block, mode_Iu, k - 1);
1039 curr = new_rd_Shrs(dbg, current_ir_graph, block, left, k_node, mode);
1042 k_node = new_r_Const_long(current_ir_graph, block, mode_Iu, bits - k);
1043 curr = new_rd_Shr(dbg, current_ir_graph, block, curr, k_node, mode);
1045 curr = new_rd_Add(dbg, current_ir_graph, block, left, curr, mode);
1047 k_node = new_r_Const_long(current_ir_graph, block, mode, (-1) << k);
1048 curr = new_rd_And(dbg, current_ir_graph, block, curr, k_node, mode);
1050 res = new_rd_Sub(dbg, current_ir_graph, block, left, curr, mode);
1051 } else { /* unsigned case */
1054 k_node = new_r_Const_long(current_ir_graph, block, mode, (1 << k) - 1);
1055 res = new_rd_And(dbg, current_ir_graph, block, left, k_node, mode);
1058 /* other constant */
1059 if (allow_Mulh(mode)) {
1060 res = replace_div_by_mulh(irn, tv);
1062 res = new_rd_Mul(dbg, current_ir_graph, block, res, c, mode);
1064 /* res = arch_dep_mul_to_shift(res); */
1066 res = new_rd_Sub(dbg, current_ir_graph, block, left, res, mode);
1072 hook_arch_dep_replace_division_by_const(irn);
1077 /* Replace DivMods with Shifts and Add/Subs and Mulh. */
1078 void arch_dep_replace_divmod_by_const(ir_node **div, ir_node **mod, ir_node *irn) {
1081 /* If the architecture dependent optimizations were not initialized
1082 or this optimization was not enabled. */
1083 if (params == NULL ||
1084 ((opts & (arch_dep_div_by_const|arch_dep_mod_by_const)) != (arch_dep_div_by_const|arch_dep_mod_by_const)))
1087 if (get_irn_opcode(irn) == iro_DivMod) {
1088 ir_node *c = get_DivMod_right(irn);
1089 ir_node *block, *left;
1096 if (get_irn_op(c) != op_Const)
1099 tv = get_Const_tarval(c);
1101 /* check for division by zero */
1102 if (classify_tarval(tv) == TV_CLASSIFY_NULL)
1105 left = get_DivMod_left(irn);
1106 mode = get_irn_mode(left);
1107 block = get_irn_n(irn, -1);
1108 dbg = get_irn_dbg_info(irn);
1110 bits = get_mode_size_bits(mode);
1114 if (mode_is_signed(mode)) {
1115 /* for signed divisions, the algorithm works for a / -2^k by negating the result */
1116 ntv = tarval_neg(tv);
1126 if (k >= 0) { /* division by 2^k or -2^k */
1127 if (mode_is_signed(mode)) {
1128 ir_node *k_node, *c_k;
1129 ir_node *curr = left;
1132 k_node = new_r_Const_long(current_ir_graph, block, mode_Iu, k - 1);
1133 curr = new_rd_Shrs(dbg, current_ir_graph, block, left, k_node, mode);
1136 k_node = new_r_Const_long(current_ir_graph, block, mode_Iu, bits - k);
1137 curr = new_rd_Shr(dbg, current_ir_graph, block, curr, k_node, mode);
1139 curr = new_rd_Add(dbg, current_ir_graph, block, left, curr, mode);
1141 c_k = new_r_Const_long(current_ir_graph, block, mode_Iu, k);
1143 *div = new_rd_Shrs(dbg, current_ir_graph, block, curr, c_k, mode);
1145 if (n_flag) { /* negate the div result */
1148 k_node = new_r_Const(current_ir_graph, block, mode, get_mode_null(mode));
1149 *div = new_rd_Sub(dbg, current_ir_graph, block, k_node, *div, mode);
1152 k_node = new_r_Const_long(current_ir_graph, block, mode, (-1) << k);
1153 curr = new_rd_And(dbg, current_ir_graph, block, curr, k_node, mode);
1155 *mod = new_rd_Sub(dbg, current_ir_graph, block, left, curr, mode);
1156 } else { /* unsigned case */
1159 k_node = new_r_Const_long(current_ir_graph, block, mode_Iu, k);
1160 *div = new_rd_Shr(dbg, current_ir_graph, block, left, k_node, mode);
1162 k_node = new_r_Const_long(current_ir_graph, block, mode, (1 << k) - 1);
1163 *mod = new_rd_And(dbg, current_ir_graph, block, left, k_node, mode);
1166 /* other constant */
1167 if (allow_Mulh(mode)) {
1170 *div = replace_div_by_mulh(irn, tv);
1172 t = new_rd_Mul(dbg, current_ir_graph, block, *div, c, mode);
1174 /* t = arch_dep_mul_to_shift(t); */
1176 *mod = new_rd_Sub(dbg, current_ir_graph, block, left, t, mode);
1182 hook_arch_dep_replace_division_by_const(irn);
1186 static const ir_settings_arch_dep_t default_params = {
1187 1, /* also use subs */
1188 4, /* maximum shifts */
1189 31, /* maximum shift amount */
1190 default_evaluate, /* default evaluator */
1192 0, /* allow Mulhs */
1193 0, /* allow Mulus */
1194 32 /* Mulh allowed up to 32 bit */
1197 /* A default parameter factory for testing purposes. */
1198 const ir_settings_arch_dep_t *arch_dep_default_factory(void) {
1199 return &default_params;