ia32_intrinsics.c 21.9 KB
Newer Older
Christian Würdig's avatar
Christian Würdig committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
/*
 * Copyright (C) 1995-2007 University of Karlsruhe.  All right reserved.
 *
 * This file is part of libFirm.
 *
 * This file may be distributed and/or modified under the terms of the
 * GNU General Public License version 2 as published by the Free Software
 * Foundation and appearing in the file LICENSE.GPL included in the
 * packaging of this file.
 *
 * Licensees holding valid libFirm Professional Edition licenses may use
 * this file in accordance with the libFirm Commercial License.
 * Agreement provided with the Software.
 *
 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE.
 */

20
/**
Christian Würdig's avatar
Christian Würdig committed
21
22
23
24
25
 * @file
 * @brief       This file implements the mapping of 64Bit intrinsic
 *              functions to code or library calls.
 * @author      Michael Beck
 * @version     $Id$
26
 */
27
28
29
30
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

31
#include "irgmod.h"
32
33
34
35
#include "irop.h"
#include "irnode_t.h"
#include "ircons.h"
#include "irprog_t.h"
36
#include "lowering.h"
37
38
#include "array.h"

39
#include "ia32_new_nodes.h"
40
41
#include "bearch_ia32_t.h"
#include "gen_ia32_regalloc_if.h"
42

43
44
45
46
/** The array of all intrinsics that must be mapped. */
static i_record *intrinsics;

/** An array to cache all entities */
47
static ir_entity *i_ents[iro_MaxOpcode];
48

49
/*
50
51
52
53
54
55
56
57
58
59
60
61
62
63
 * 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

64
65
66
/**
 * Replace a call be a tuple of l_res, h_res.
 */
67
68
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];
69
70
71

	in[0] = l_res;
	in[1] = h_res;
72
	res = new_r_Tuple(irg, block, h_res == NULL ? 1 : 2, in);
73
74
75

	turn_into_tuple(call, pn_Call_max);
	set_Tuple_pred(call, pn_Call_M_regular,        get_irg_no_mem(irg));
76
	set_Tuple_pred(call, pn_Call_X_regular,        new_r_Jmp(irg, block));
77
78
	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
79
	set_Tuple_pred(call, pn_Call_M_except,         get_irg_no_mem(irg));
80
	set_Tuple_pred(call, pn_Call_P_value_res_base, get_irg_bad(irg));
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
}

/**
 * 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));
98
	ir_node  *l_res, *h_res, *add;
99
100
101

	/* l_res = a_l + b_l */
	/* h_res = a_h + b_h + carry */
102
103
104
105

	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);
106

107
	resolve_call(call, l_res, h_res, irg, block);
108
109
110
111
112
113
114
	return 1;
}

/**
 * Map a Sub (a_l, a_h, b_l, b_h)
 */
static int map_Sub(ir_node *call, void *ctx) {
115
116
117
118
119
120
121
122
123
124
125
	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));
126
	ir_node  *l_res, *h_res, *res;
127
128

	/* l_res = a_l - b_l */
129
	/* h_res = a_h - b_h - carry */
130

131
132
133
	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);
134

135
	resolve_call(call, l_res, h_res, irg, block);
136
137
138
	return 1;
}

139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
/**
 * 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 */
Christian Würdig's avatar
Christian Würdig committed
156
	h_res = new_rd_ia32_l_ShlD(dbg, irg, block, a_h, a_l, cnt, l_res_mode);
157
158

	/* l_res = SHL a_l, cnt */
Christian Würdig's avatar
Christian Würdig committed
159
	l_res = new_rd_ia32_l_Shl(dbg, irg, block, a_l, cnt, h_res_mode);
160

161
	//add_irn_dep(l_res, h_res);
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187

	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);
188

189
	//add_irn_dep(h_res, l_res);
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214

	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 */
215
	h_res = new_rd_ia32_l_Sar(dbg, irg, block, a_h, cnt, h_res_mode);
216

217
	//add_irn_dep(h_res, l_res);
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248

	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
	*/
249
250
251
	mul   = new_rd_ia32_l_Mul(dbg, irg, block, a_l, b_l);
	pEDX  = new_rd_Proj(dbg, irg, block, mul, l_res_mode, pn_ia32_l_Mul_EDX);
	l_res = new_rd_Proj(dbg, irg, block, mul, l_res_mode, pn_ia32_l_Mul_EAX);
252

253
	mul   = new_rd_ia32_l_Mul(dbg, irg, block, a_h, b_l);
254
	add   = new_rd_ia32_l_Add(dbg, irg, block, mul, pEDX, h_res_mode);
255
	mul   = new_rd_ia32_l_Mul(dbg, irg, block, a_l, b_h);
256
257
258
259
260
261
262
	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;
}

263
264
265
266
267
268
269
270
271
272
273
274
275
/**
 * 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));
276
	ir_node  *l_res, *h_res, *cnst, *res;
277

278
279
	/* too bad: we need 0 in a register here */
	cnst  = new_Const_long(h_res_mode, 0);
280

281
	/* l_res = 0 - a_l */
282
283
	/* h_res = 0 - a_h - carry */

284
285
286
	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);
287
288
289
290
291
292

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

	return 1;
}

293
294
295
296
297
298
299
300
301
302
303
304
305
/**
 * 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));
306
	ir_node  *l_res, *h_res, *sign, *sub_l, *sub_h, *res;
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321

	/*
		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

	*/

322
323
324
	sign  = new_rd_ia32_l_Sar(dbg, irg, block, a_h, new_Const_long(h_res_mode, 31), h_res_mode);
	sub_l = new_rd_ia32_l_Xor(dbg, irg, block, a_l, sign, l_res_mode);
	sub_h = new_rd_ia32_l_Xor(dbg, irg, block, a_h, sign, h_res_mode);
325
326
327
	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);
328
329
330
331
332
333

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

	return 1;
}

Christian Würdig's avatar
Christian Würdig committed
334
335
336
337
338
typedef enum {
	IA32_INTRINSIC_DIV,
	IA32_INTRINSIC_MOD,
} ia32_intrinsic_divmod_t;

339
340
341
/**
 * Maps a Div/Mod (a_l, a_h, b_l, b_h)
 */
Christian Würdig's avatar
Christian Würdig committed
342
static int DivMod_mapper(ir_node *call, void *ctx, ia32_intrinsic_divmod_t dmtp) {
343
	ia32_intrinsic_env_t *env = ctx;
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
	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);
	ir_entity *ent_a      = env->irg == irg ? env->ll_div_op1 : NULL;
	ir_entity *ent_b      = env->irg == irg ? env->ll_div_op2 : NULL;
	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;
362
363
364
365
366

	/* 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);
367
		env->irg = irg;
368
369
370
371
372
	}

	if (! ent_b) {
		ent_b = env->ll_div_op2 =
			frame_alloc_area(get_irg_frame_type(irg), 2 * mode_bytes, 16, 0);
373
		env->irg = irg;
374
375
376
377
378
379
380
381
382
	}

	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));
383
	op_mem[0] = store_l;
384
385
386

	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);
387
	add_ia32_am_offs_int(store_h, mode_bytes);
388
389
	set_ia32_use_frame(store_h);
	set_ia32_ls_mode(store_h, get_irn_mode(a_h));
390
	op_mem[1] = store_h;
391
392
393
394
395
396
397
398
399

	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);
400
	fa     = new_r_Proj(irg, block, fa, mode_E, pn_ia32_l_vfild_res);
401
402
403
404
405
406

	/* 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));
407
	op_mem[0] = store_l;
408
409
410

	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);
411
	add_ia32_am_offs_int(store_h, mode_bytes);
412
413
	set_ia32_use_frame(store_h);
	set_ia32_ls_mode(store_h, get_irn_mode(b_h));
414
	op_mem[1] = store_h;
415
416
417
418
419
420
421
422
423

	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);
424
	fb     = new_r_Proj(irg, block, fb, mode_E, pn_ia32_l_vfild_res);
425
426
427
428
429
430
431

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

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

	/* perform division */
Christian Würdig's avatar
Christian Würdig committed
432
433
	switch (dmtp) {
		case IA32_INTRINSIC_DIV:
434
			fres = new_rd_ia32_l_vfdiv(dbg, irg, block, fa, fb);
435
			fres = new_rd_Proj(dbg, irg, block, fres, mode_E, pn_ia32_l_vfdiv_res);
Christian Würdig's avatar
Christian Würdig committed
436
437
			break;
		case IA32_INTRINSIC_MOD:
438
			fres = new_rd_ia32_l_vfprem(dbg, irg, block, fa, fb, mode_E);
Christian Würdig's avatar
Christian Würdig committed
439
440
441
			break;
		default:
			assert(0);
442
443
444
445
446
447
448
	}

	/* 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);
449
	mem = fres;
450
451
452
453
454
455
456
457
458
459
460

	/* 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);
461
	add_ia32_am_offs_int(h_res, mode_bytes);
462
463
464
465
466
467
468
469
470
471
472
	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) {
Christian Würdig's avatar
Christian Würdig committed
473
	return DivMod_mapper(call, ctx, IA32_INTRINSIC_DIV);
474
475
476
}

static int map_Mod(ir_node *call, void *ctx) {
Christian Würdig's avatar
Christian Würdig committed
477
	return DivMod_mapper(call, ctx, IA32_INTRINSIC_MOD);
478
479
480
481
482
483
484
}

/**
 * Maps a Conv (a_l, a_h)
 */
static int map_Conv(ir_node *call, void *ctx) {
	ia32_intrinsic_env_t *env = ctx;
485
486
487
488
489
490
491
492
493
494
495
	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);
	ir_entity *ent;
	ir_node   *l_res, *h_res, *frame, *fres;
	ir_node   *store_l, *store_h;
	ir_node   *op_mem[2], *mem;
496
497
498
499
500
501
502
503
504
505

	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 */
506
		ent = env->irg == irg ? env->d_ll_conv : NULL;
507
		if (! ent) {
508
509
			ent      = env->d_ll_conv = frame_alloc_area(get_irg_frame_type(irg), 2 * gp_bytes, 16, 0);
			env->irg = irg;
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
		}

		/* Store arg */
		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);
531
		mem = a_f;
532
533
534
535
536
537
538
539
540
541
542

		/* 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);
543
		add_ia32_am_offs_int(h_res, gp_bytes);
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
		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 */
562
		ent = env->irg == irg ? env->ll_d_conv : NULL;
563
564
		if (! ent) {
			ent = env->ll_d_conv = frame_alloc_area(get_irg_frame_type(irg), 2 * gp_bytes, 16, 0);
565
			env->irg = irg;
566
567
568
569
570
571
572
573
574
575
		}

		/* Store arg */
		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));
576
		op_mem[0] = store_l;
577
578
579
580

		/* 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);
581
		add_ia32_am_offs_int(store_h, gp_bytes);
582
583
		set_ia32_use_frame(store_h);
		set_ia32_ls_mode(store_h, get_irn_mode(a_h));
584
		op_mem[1] = store_h;
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616

		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;
}

617
/* Ia32 implementation of intrinsic mapping. */
618
619
620
ir_entity *ia32_create_intrinsic_fkt(ir_type *method, const ir_op *op,
                                     const ir_mode *imode, const ir_mode *omode,
                                     void *context)
621
622
{
	i_record      elt;
623
	ir_entity     **ent = NULL;
624
625
626
627
628
629
630
631
632
633
634
635
636
637
	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;
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
	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;
654
655
656
657
658
659
660
661
	case iro_Minus:
		ent    = &i_ents[iro_Minus];
		mapper = map_Minus;
		break;
	case iro_Abs:
		ent    = &i_ents[iro_Abs];
		mapper = map_Abs;
		break;
662
663
664
665
666
667
668
669
670
671
672
673
	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;
674
	default:
Christian Würdig's avatar
Christian Würdig committed
675
		fprintf(stderr, "FIXME: unhandled op for ia32 intrinsic function %s\n", get_id_str(op->name));
676
677
678
679
680
681
		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
682
		ident *id = mangle(IDENT("L"), get_op_ident(op));
683
684
685
686
687
688
		*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;
689
	elt.i_call.ctx      = context;
690
691
692
	elt.i_call.link     = NULL;

	ARR_APP1(i_record, intrinsics, elt);
Sebastian Hack's avatar
Sebastian Hack committed
693
	return *ent;
694
}