ia32_intrinsics.c 21.2 KB
Newer Older
1
2
3
4
5
6
7
/**
 * This file implements the mapping of 64Bit intrinsic functions to
 * code or library calls.
 * @author Michael Beck
 * $Id$
 */

8
9
10
11
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

12
#include "irgmod.h"
13
14
15
16
17
18
19
20
21
#include "irop.h"
#include "irnode_t.h"
#include "ircons.h"
#include "irprog_t.h"
#include "lower_intrinsics.h"
#include "lower_dw.h"
#include "mangle.h"
#include "array.h"

22
#include "ia32_new_nodes.h"
23
24
#include "bearch_ia32_t.h"
#include "gen_ia32_regalloc_if.h"
25

26
27
28
29
30
31
/** The array of all intrinsics that must be mapped. */
static i_record *intrinsics;

/** An array to cache all entities */
static entity *i_ents[iro_MaxOpcode];

32
/*
33
34
35
36
37
38
39
40
41
42
43
44
45
46
 * Maps all intrinsic calls that the backend support
 * and map all instructions the backend did not support
 * to runtime calls.
 */
void ia32_handle_intrinsics(void) {
	if (intrinsics && ARR_LEN(intrinsics) > 0)
		lower_intrinsics(intrinsics, ARR_LEN(intrinsics));
}

#define BINOP_Left_Low   0
#define BINOP_Left_High  1
#define BINOP_Right_Low  2
#define BINOP_Right_High 3

47
48
static void resolve_call(ir_node *call, ir_node *l_res, ir_node *h_res, ir_graph *irg, ir_node *block) {
	ir_node *res, *in[2];
49
50
51

	in[0] = l_res;
	in[1] = h_res;
52
	res = new_r_Tuple(irg, block, h_res == NULL ? 1 : 2, in);
53
54
55
56
57

	turn_into_tuple(call, pn_Call_max);
	set_Tuple_pred(call, pn_Call_M_regular,        get_irg_no_mem(irg));
	set_Tuple_pred(call, pn_Call_X_except,         get_irg_bad(irg));
	set_Tuple_pred(call, pn_Call_T_result,         res);
Christian Würdig's avatar
Christian Würdig committed
58
	set_Tuple_pred(call, pn_Call_M_except,         get_irg_no_mem(irg));
59
	set_Tuple_pred(call, pn_Call_P_value_res_base, get_irg_bad(irg));
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
}

/**
 * Map an Add (a_l, a_h, b_l, b_h)
 */
static int map_Add(ir_node *call, void *ctx) {
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	ir_node  *a_l        = params[BINOP_Left_Low];
	ir_node  *a_h        = params[BINOP_Left_High];
	ir_node  *b_l        = params[BINOP_Right_Low];
	ir_node  *b_h        = params[BINOP_Right_High];
	ir_mode  *l_res_mode = get_type_mode(get_method_res_type(method, 0));
	ir_mode  *h_res_mode = get_type_mode(get_method_res_type(method, 1));
77
	ir_node  *l_res, *h_res, *add;
78
79
80

	/* l_res = a_l + b_l */
	/* h_res = a_h + b_h + carry */
81
82
83
84

	add   = new_rd_ia32_Add64Bit(dbg, irg, block, a_l, a_h, b_l, b_h);
	l_res = new_r_Proj(irg, block, add, l_res_mode, pn_ia32_Add64Bit_low_res);
	h_res = new_r_Proj(irg, block, add, h_res_mode, pn_ia32_Add64Bit_high_res);
85

86
	resolve_call(call, l_res, h_res, irg, block);
87
88
89
90
91
92
93
	return 1;
}

/**
 * Map a Sub (a_l, a_h, b_l, b_h)
 */
static int map_Sub(ir_node *call, void *ctx) {
94
95
96
97
98
99
100
101
102
103
104
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	ir_node  *a_l        = params[BINOP_Left_Low];
	ir_node  *a_h        = params[BINOP_Left_High];
	ir_node  *b_l        = params[BINOP_Right_Low];
	ir_node  *b_h        = params[BINOP_Right_High];
	ir_mode  *l_res_mode = get_type_mode(get_method_res_type(method, 0));
	ir_mode  *h_res_mode = get_type_mode(get_method_res_type(method, 1));
105
	ir_node  *l_res, *h_res, *res;
106
107

	/* l_res = a_l - b_l */
108
	/* h_res = a_h - b_h - carry */
109

110
111
112
	res   = new_rd_ia32_Sub64Bit(dbg, irg, block, a_l, a_h, b_l, b_h);
	l_res = new_r_Proj(irg, block, res, l_res_mode, pn_ia32_Sub64Bit_low_res);
	h_res = new_r_Proj(irg, block, res, h_res_mode, pn_ia32_Sub64Bit_high_res);
113

114
	resolve_call(call, l_res, h_res, irg, block);
115
116
117
	return 1;
}

118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
/**
 * Map a Shl (a_l, a_h, count)
 */
static int map_Shl(ir_node *call, void *ctx) {
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	ir_node  *a_l        = params[BINOP_Left_Low];
	ir_node  *a_h        = params[BINOP_Left_High];
	ir_node  *cnt        = params[BINOP_Right_Low];
	ir_mode  *l_res_mode = get_type_mode(get_method_res_type(method, 0));
	ir_mode  *h_res_mode = get_type_mode(get_method_res_type(method, 1));
	ir_node  *l_res, *h_res;

	/* h_res = SHLD a_h, a_l, cnt */
	l_res = new_rd_ia32_l_ShlD(dbg, irg, block, a_h, a_l, cnt, l_res_mode);

	/* l_res = SHL a_l, cnt */
	h_res = new_rd_ia32_l_Shl(dbg, irg, block, a_l, cnt, h_res_mode);
139

Sebastian Hack's avatar
Sebastian Hack committed
140
	add_irn_dep(h_res, l_res);
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166

	resolve_call(call, l_res, h_res, irg, block);
	return 1;
}

/**
 * Map a Shr (a_l, a_h, count)
 */
static int map_Shr(ir_node *call, void *ctx) {
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	ir_node  *a_l        = params[BINOP_Left_Low];
	ir_node  *a_h        = params[BINOP_Left_High];
	ir_node  *cnt        = params[BINOP_Right_Low];
	ir_mode  *l_res_mode = get_type_mode(get_method_res_type(method, 0));
	ir_mode  *h_res_mode = get_type_mode(get_method_res_type(method, 1));
	ir_node  *l_res, *h_res;

	/* l_res = SHRD a_l, a_h, cnt */
	l_res = new_rd_ia32_l_ShrD(dbg, irg, block, a_l, a_h, cnt, l_res_mode);

	/* h_res = SHR a_h, cnt */
	h_res = new_rd_ia32_l_Shr(dbg, irg, block, a_h, cnt, h_res_mode);
167

Sebastian Hack's avatar
Sebastian Hack committed
168
	add_irn_dep(h_res, l_res);
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194

	resolve_call(call, l_res, h_res, irg, block);
	return 1;
}

/**
 * Map a Shrs (a_l, a_h, count)
 */
static int map_Shrs(ir_node *call, void *ctx) {
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	ir_node  *a_l        = params[BINOP_Left_Low];
	ir_node  *a_h        = params[BINOP_Left_High];
	ir_node  *cnt        = params[BINOP_Right_Low];
	ir_mode  *l_res_mode = get_type_mode(get_method_res_type(method, 0));
	ir_mode  *h_res_mode = get_type_mode(get_method_res_type(method, 1));
	ir_node  *l_res, *h_res;

	/* l_res = SHRD a_l, a_h, cnt */
	l_res = new_rd_ia32_l_ShrD(dbg, irg, block, a_l, a_h, cnt, l_res_mode);

	/* h_res = SAR a_h, cnt */
	h_res = new_rd_ia32_l_Shrs(dbg, irg, block, a_h, cnt, h_res_mode);
195

Sebastian Hack's avatar
Sebastian Hack committed
196
	add_irn_dep(h_res, l_res);
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228

	resolve_call(call, l_res, h_res, irg, block);
	return 1;
}

/**
 * Map a Mul (a_l, a_h, b_l, b_h)
 */
static int map_Mul(ir_node *call, void *ctx) {
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	ir_node  *a_l        = params[BINOP_Left_Low];
	ir_node  *a_h        = params[BINOP_Left_High];
	ir_node  *b_l        = params[BINOP_Right_Low];
	ir_node  *b_h        = params[BINOP_Right_High];
	ir_mode  *l_res_mode = get_type_mode(get_method_res_type(method, 0));
	ir_mode  *h_res_mode = get_type_mode(get_method_res_type(method, 1));
	ir_node  *l_res, *h_res, *mul, *pEDX, *add;

	/*
		EDX:EAX = a_l * b_l
		l_res   = EAX

		t1 = b_l * a_h
		t2 = t1 + EDX
		t3 = a_l * b_h
		h_res = t2 + t3
	*/
	mul   = new_rd_ia32_l_MulS(dbg, irg, block, a_l, b_l);
229
	set_ia32_res_mode(mul, l_res_mode);
230
231
232
233
234
235
236
237
238
239
240
241
242
	pEDX  = new_rd_Proj(dbg, irg, block, mul, l_res_mode, pn_ia32_l_MulS_EDX);
	l_res = new_rd_Proj(dbg, irg, block, mul, l_res_mode, pn_ia32_l_MulS_EAX);

	mul   = new_rd_ia32_l_Mul(dbg, irg, block, a_h, b_l, h_res_mode);
	add   = new_rd_ia32_l_Add(dbg, irg, block, mul, pEDX, h_res_mode);
	mul   = new_rd_ia32_l_Mul(dbg, irg, block, a_l, b_h, h_res_mode);
	h_res = new_rd_ia32_l_Add(dbg, irg, block, add, mul, h_res_mode);

	resolve_call(call, l_res, h_res, irg, block);

	return 1;
}

243
244
245
246
247
248
249
250
251
252
253
254
255
/**
 * Map a Minus (a_l, a_h)
 */
static int map_Minus(ir_node *call, void *ctx) {
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	ir_node  *a_l        = params[BINOP_Left_Low];
	ir_node  *a_h        = params[BINOP_Left_High];
	ir_mode  *l_res_mode = get_type_mode(get_method_res_type(method, 0));
	ir_mode  *h_res_mode = get_type_mode(get_method_res_type(method, 1));
256
	ir_node  *l_res, *h_res, *cnst, *res;
257

258
259
	/* too bad: we need 0 in a register here */
	cnst  = new_Const_long(h_res_mode, 0);
260

261
	/* l_res = 0 - a_l */
262
263
	/* h_res = 0 - a_h - carry */

264
265
266
	res   = new_rd_ia32_Minus64Bit(dbg, irg, block, cnst, a_l, a_h);
	l_res = new_r_Proj(irg, block, res, l_res_mode, pn_ia32_Minus64Bit_low_res);
	h_res = new_r_Proj(irg, block, res, h_res_mode, pn_ia32_Minus64Bit_high_res);
267
268
269
270
271
272

	resolve_call(call, l_res, h_res, irg, block);

	return 1;
}

273
274
275
276
277
278
279
280
281
282
283
284
285
/**
 * Map a Abs (a_l, a_h)
 */
static int map_Abs(ir_node *call, void *ctx) {
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	ir_node  *a_l        = params[BINOP_Left_Low];
	ir_node  *a_h        = params[BINOP_Left_High];
	ir_mode  *l_res_mode = get_type_mode(get_method_res_type(method, 0));
	ir_mode  *h_res_mode = get_type_mode(get_method_res_type(method, 1));
286
	ir_node  *l_res, *h_res, *sign, *sub_l, *sub_h, *res;
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304

	/*
		Code inspired by gcc output :) (although gcc doubles the
		operation for t1 as t2 and uses t1 for operations with low part
		and t2 for operations with high part which is actually unnecessary
		because t1 and t2 represent the same value)

		t1    = SHRS a_h, 31
		t2    = a_l ^ t1
		t3    = a_h ^ t1
		l_res = t2 - t1
		h_res = t3 - t1 - carry

	*/

	sign  = new_rd_ia32_l_Shrs(dbg, irg, block, a_h, new_Const_long(h_res_mode, 31), h_res_mode);
	sub_l = new_rd_ia32_l_Eor(dbg, irg, block, a_l, sign, l_res_mode);
	sub_h = new_rd_ia32_l_Eor(dbg, irg, block, a_h, sign, h_res_mode);
305
306
307
	res   = new_rd_ia32_Sub64Bit(dbg, irg, block, sub_l, sub_h, sign, sign);
	l_res = new_r_Proj(irg, block, res, l_res_mode, pn_ia32_Sub64Bit_low_res);
	h_res = new_r_Proj(irg, block, res, h_res_mode, pn_ia32_Sub64Bit_high_res);
308
309
310
311
312
313

	resolve_call(call, l_res, h_res, irg, block);

	return 1;
}

314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
/**
 * Maps a Div/Mod (a_l, a_h, b_l, b_h)
 */
static int DivMod_mapper(ir_node *call, void *ctx, int need_mod) {
	ia32_intrinsic_env_t *env = ctx;
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	ir_node  *a_l        = params[BINOP_Left_Low];
	ir_node  *a_h        = params[BINOP_Left_High];
	ir_node  *b_l        = params[BINOP_Right_Low];
	ir_node  *b_h        = params[BINOP_Right_High];
	ir_mode  *l_res_mode = get_type_mode(get_method_res_type(method, 0));
	ir_mode  *h_res_mode = get_type_mode(get_method_res_type(method, 1));
	int      mode_bytes  = get_mode_size_bytes(ia32_reg_classes[CLASS_ia32_gp].mode);
	entity   *ent_a      = env->ll_div_op1;
	entity   *ent_b      = env->ll_div_op2;
	ir_node  *l_res, *h_res, *frame;
	ir_node  *store_l, *store_h;
	ir_node  *op_mem[2], *mem, *fa_mem, *fb_mem;
	ir_node  *fa, *fb, *fres;
	char     buf[3];

	/* allocate memory on frame to store args */

	if (! ent_a) {
		ent_a = env->ll_div_op1 =
			frame_alloc_area(get_irg_frame_type(irg), 2 * mode_bytes, 16, 0);
	}

	if (! ent_b) {
		ent_b = env->ll_div_op2 =
			frame_alloc_area(get_irg_frame_type(irg), 2 * mode_bytes, 16, 0);
	}

	snprintf(buf, sizeof(buf), "%d", mode_bytes);
	frame = get_irg_frame(irg);

	/* store first arg */
	store_l   = new_rd_ia32_l_Store(dbg, irg, block, frame, a_l, get_irg_no_mem(irg));
	set_ia32_frame_ent(store_l, ent_a);
	set_ia32_use_frame(store_l);
	set_ia32_ls_mode(store_l, get_irn_mode(a_l));
	op_mem[0] = new_r_Proj(irg, block, store_l, mode_M, pn_ia32_l_Store_M);

	store_h   = new_rd_ia32_l_Store(dbg, irg, block, frame, a_h, get_irg_no_mem(irg));
	set_ia32_frame_ent(store_h, ent_a);
	add_ia32_am_offs(store_h, buf);
	set_ia32_use_frame(store_h);
	set_ia32_ls_mode(store_h, get_irn_mode(a_h));
	op_mem[1] = new_r_Proj(irg, block, store_h, mode_M, pn_ia32_l_Store_M);

	mem = new_r_Sync(irg, block, 2, op_mem);

	/* load first arg into FPU */
	fa = new_rd_ia32_l_vfild(dbg, irg, block, frame, mem);
	set_ia32_frame_ent(fa, ent_a);
	set_ia32_use_frame(fa);
	set_ia32_ls_mode(fa, mode_D);
	fa_mem = new_r_Proj(irg, block, fa, mode_M, pn_ia32_l_vfild_M);
	fa     = new_r_Proj(irg, block, fa, mode_D, pn_ia32_l_vfild_res);

	/* store second arg */
	store_l   = new_rd_ia32_l_Store(dbg, irg, block, frame, b_l, get_irg_no_mem(irg));
	set_ia32_frame_ent(store_l, ent_b);
	set_ia32_use_frame(store_l);
	set_ia32_ls_mode(store_l, get_irn_mode(b_l));
	op_mem[0] = new_r_Proj(irg, block, store_l, mode_M, pn_ia32_l_Store_M);

	store_h   = new_rd_ia32_l_Store(dbg, irg, block, frame, b_h, get_irg_no_mem(irg));
	set_ia32_frame_ent(store_h, ent_b);
	add_ia32_am_offs(store_h, buf);
	set_ia32_use_frame(store_h);
	set_ia32_ls_mode(store_h, get_irn_mode(b_h));
	op_mem[1] = new_r_Proj(irg, block, store_h, mode_M, pn_ia32_l_Store_M);

	mem = new_r_Sync(irg, block, 2, op_mem);

	/* load second arg into FPU */
	fb = new_rd_ia32_l_vfild(dbg, irg, block, frame, mem);
	set_ia32_frame_ent(fb, ent_b);
	set_ia32_use_frame(fb);
	set_ia32_ls_mode(fb, mode_D);
	fb_mem = new_r_Proj(irg, block, fb, mode_M, pn_ia32_l_vfild_M);
	fb     = new_r_Proj(irg, block, fb, mode_D, pn_ia32_l_vfild_res);

	op_mem[0] = fa_mem;
	op_mem[1] = fb_mem;

	mem = new_r_Sync(irg, block, 2, op_mem);

	/* perform division */
	fres = new_rd_ia32_l_vfdiv(dbg, irg, block, fa, fb, mode_D);

	if (need_mod) {
		/* we need modulo: mod = a - b * res */

		fres = new_rd_ia32_l_vfmul(dbg, irg, block, fb, fres, mode_D);
		fres = new_rd_ia32_l_vfsub(dbg, irg, block, fa, fres, mode_D);
	}

	/* store back result, we use ent_a here */
	fres = new_rd_ia32_l_vfist(dbg, irg, block, frame, fres, mem);
	set_ia32_frame_ent(fres, ent_a);
	set_ia32_use_frame(fres);
	set_ia32_ls_mode(fres, mode_D);
	mem = new_r_Proj(irg, block, fres, mode_M, pn_ia32_l_vfist_M);

	/* load low part of the result */
	l_res = new_rd_ia32_l_Load(dbg, irg, block, frame, mem);
	set_ia32_frame_ent(l_res, ent_a);
	set_ia32_use_frame(l_res);
	set_ia32_ls_mode(l_res, l_res_mode);
	l_res = new_r_Proj(irg, block, l_res, l_res_mode, pn_ia32_l_Load_res);

	/* load hight part of the result */
	h_res = new_rd_ia32_l_Load(dbg, irg, block, frame, mem);
	set_ia32_frame_ent(h_res, ent_a);
	add_ia32_am_offs(h_res, buf);
	set_ia32_use_frame(h_res);
	set_ia32_ls_mode(h_res, h_res_mode);
	h_res = new_r_Proj(irg, block, h_res, h_res_mode, pn_ia32_l_Load_res);

	/* lower the call */
	resolve_call(call, l_res, h_res, irg, block);

	return 1;
}

static int map_Div(ir_node *call, void *ctx) {
	return DivMod_mapper(call, ctx, 0);
}

static int map_Mod(ir_node *call, void *ctx) {
	return DivMod_mapper(call, ctx, 1);
}

/**
 * Maps a Conv (a_l, a_h)
 */
static int map_Conv(ir_node *call, void *ctx) {
	ia32_intrinsic_env_t *env = ctx;
	ir_graph *irg        = current_ir_graph;
	dbg_info *dbg        = get_irn_dbg_info(call);
	ir_node  *block      = get_nodes_block(call);
	ir_node  **params    = get_Call_param_arr(call);
	ir_type  *method     = get_Call_type(call);
	int      n           = get_Call_n_params(call);
	int      gp_bytes    = get_mode_size_bytes(ia32_reg_classes[CLASS_ia32_gp].mode);
	entity   *ent;
	ir_node  *l_res, *h_res, *frame, *fres;
	ir_node  *store_l, *store_h;
	ir_node  *op_mem[2], *mem;
	char     buf[3];

	if (n == 1) {
		/* We have a Conv float -> long long here */
		ir_node *a_f        = params[0];
		ir_mode *l_res_mode = get_type_mode(get_method_res_type(method, 0));
		ir_mode *h_res_mode = get_type_mode(get_method_res_type(method, 1));

		assert(mode_is_float(get_irn_mode(a_f)) && "unexpected Conv call");

		/* allocate memory on frame to store args */
		ent = env->d_ll_conv;
		if (! ent) {
			ent = env->d_ll_conv = frame_alloc_area(get_irg_frame_type(irg), 2 * gp_bytes, 16, 0);
		}

		/* Store arg */
		snprintf(buf, sizeof(buf), "%d", gp_bytes);
		frame = get_irg_frame(irg);

		/*
			Now we create a node to move the value from a XMM register into
			x87 FPU because it is unknown here, which FPU is used.
			This node is killed in transformation phase when not needed.
			Otherwise it is split up into a movsd + fld
		*/
		a_f = new_rd_ia32_l_SSEtoX87(dbg, irg, block, frame, a_f, get_irg_no_mem(irg), mode_D);
		set_ia32_frame_ent(a_f, ent);
		set_ia32_use_frame(a_f);
		set_ia32_ls_mode(a_f, mode_D);

		/* store from FPU as Int */
		a_f = new_rd_ia32_l_vfist(dbg, irg, block, frame, a_f, get_irg_no_mem(irg));
		set_ia32_frame_ent(a_f, ent);
		set_ia32_use_frame(a_f);
		set_ia32_ls_mode(a_f, mode_D);
		mem = new_r_Proj(irg, block, a_f, mode_M, pn_ia32_l_vfist_M);

		/* load low part of the result */
		l_res = new_rd_ia32_l_Load(dbg, irg, block, frame, mem);
		set_ia32_frame_ent(l_res, ent);
		set_ia32_use_frame(l_res);
		set_ia32_ls_mode(l_res, l_res_mode);
		l_res = new_r_Proj(irg, block, l_res, l_res_mode, pn_ia32_l_Load_res);

		/* load hight part of the result */
		h_res = new_rd_ia32_l_Load(dbg, irg, block, frame, mem);
		set_ia32_frame_ent(h_res, ent);
		add_ia32_am_offs(h_res, buf);
		set_ia32_use_frame(h_res);
		set_ia32_ls_mode(h_res, h_res_mode);
		h_res = new_r_Proj(irg, block, h_res, h_res_mode, pn_ia32_l_Load_res);

		/* lower the call */
		resolve_call(call, l_res, h_res, irg, block);
	}
	else if (n == 2) {
		/* We have a Conv long long -> float here */
		ir_node *a_l       = params[BINOP_Left_Low];
		ir_node *a_h       = params[BINOP_Left_High];
		ir_mode *mode_a_l  = get_irn_mode(a_l);
		ir_mode *mode_a_h  = get_irn_mode(a_h);
		ir_mode *fres_mode = get_type_mode(get_method_res_type(method, 0));

		assert(! mode_is_float(mode_a_l) && ! mode_is_float(mode_a_h) && "unexpected Conv call");

		/* allocate memory on frame to store args */
		ent = env->ll_d_conv;
		if (! ent) {
			ent = env->ll_d_conv = frame_alloc_area(get_irg_frame_type(irg), 2 * gp_bytes, 16, 0);
		}

		/* Store arg */
		snprintf(buf, sizeof(buf), "%d", gp_bytes);
		frame = get_irg_frame(irg);

		/* store first arg (low part) */
		store_l   = new_rd_ia32_l_Store(dbg, irg, block, frame, a_l, get_irg_no_mem(irg));
		set_ia32_frame_ent(store_l, ent);
		set_ia32_use_frame(store_l);
		set_ia32_ls_mode(store_l, get_irn_mode(a_l));
		op_mem[0] = new_r_Proj(irg, block, store_l, mode_M, pn_ia32_l_Store_M);

		/* store second arg (high part) */
		store_h   = new_rd_ia32_l_Store(dbg, irg, block, frame, a_h, get_irg_no_mem(irg));
		set_ia32_frame_ent(store_h, ent);
		add_ia32_am_offs(store_h, buf);
		set_ia32_use_frame(store_h);
		set_ia32_ls_mode(store_h, get_irn_mode(a_h));
		op_mem[1] = new_r_Proj(irg, block, store_h, mode_M, pn_ia32_l_Store_M);

		mem = new_r_Sync(irg, block, 2, op_mem);

		/* Load arg into x87 FPU (implicit convert) */
		fres = new_rd_ia32_l_vfild(dbg, irg, block, frame, mem);
		set_ia32_frame_ent(fres, ent);
		set_ia32_use_frame(fres);
		set_ia32_ls_mode(fres, mode_D);
		mem  = new_r_Proj(irg, block, fres, mode_M, pn_ia32_l_vfild_M);
		fres = new_r_Proj(irg, block, fres, fres_mode, pn_ia32_l_vfild_res);

		/*
			Now we create a node to move the loaded value into a XMM
			register because it is unknown here, which FPU is used.
			This node is killed in transformation phase when not needed.
			Otherwise it is split up into a fst + movsd
		*/
		fres = new_rd_ia32_l_X87toSSE(dbg, irg, block, frame, fres, mem, fres_mode);
		set_ia32_frame_ent(fres, ent);
		set_ia32_use_frame(fres);
		set_ia32_ls_mode(fres, fres_mode);

		/* lower the call */
		resolve_call(call, fres, NULL, irg, block);
	}
	else {
		assert(0 && "unexpected Conv call");
	}

	return 1;
}

591
/* Ia32 implementation of intrinsic mapping. */
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
entity *ia32_create_intrinsic_fkt(ir_type *method, const ir_op *op,
                                  const ir_mode *imode, const ir_mode *omode,
                                  void *context)
{
	i_record      elt;
	entity        **ent = NULL;
	i_mapper_func mapper;

	if (! intrinsics)
		intrinsics = NEW_ARR_F(i_record, 0);

	switch (get_op_code(op)) {
	case iro_Add:
		ent    = &i_ents[iro_Add];
		mapper = map_Add;
		break;
	case iro_Sub:
		ent    = &i_ents[iro_Sub];
		mapper = map_Sub;
		break;
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
	case iro_Shl:
		ent    = &i_ents[iro_Shl];
		mapper = map_Shl;
		break;
	case iro_Shr:
		ent    = &i_ents[iro_Shr];
		mapper = map_Shr;
		break;
	case iro_Shrs:
		ent    = &i_ents[iro_Shrs];
		mapper = map_Shrs;
		break;
	case iro_Mul:
		ent    = &i_ents[iro_Mul];
		mapper = map_Mul;
		break;
628
629
630
631
632
633
634
635
	case iro_Minus:
		ent    = &i_ents[iro_Minus];
		mapper = map_Minus;
		break;
	case iro_Abs:
		ent    = &i_ents[iro_Abs];
		mapper = map_Abs;
		break;
636
637
638
639
640
641
642
643
644
645
646
647
	case iro_Div:
		ent    = &i_ents[iro_Div];
		mapper = map_Div;
		break;
	case iro_Mod:
		ent    = &i_ents[iro_Mod];
		mapper = map_Mod;
		break;
	case iro_Conv:
		ent    = &i_ents[iro_Conv];
		mapper = map_Conv;
		break;
648
	default:
Christian Würdig's avatar
Christian Würdig committed
649
		fprintf(stderr, "FIXME: unhandled op for ia32 intrinsic function %s\n", get_id_str(op->name));
650
651
652
653
654
655
		return def_create_intrinsic_fkt(method, op, imode, omode, context);
	}

	if (ent && ! *ent) {
#define IDENT(s)  new_id_from_chars(s, sizeof(s)-1)

Sebastian Hack's avatar
Sebastian Hack committed
656
		ident *id = mangle(IDENT("L"), get_op_ident(op));
657
658
659
660
661
662
		*ent = new_entity(get_glob_type(), id, method);
	}

	elt.i_call.kind     = INTRINSIC_CALL;
	elt.i_call.i_ent    = *ent;
	elt.i_call.i_mapper = mapper;
663
	elt.i_call.ctx      = context;
664
665
666
	elt.i_call.link     = NULL;

	ARR_APP1(i_record, intrinsics, elt);
Sebastian Hack's avatar
Sebastian Hack committed
667
	return *ent;
668
}