ia32_intrinsics.c 21.1 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
/** The array of all intrinsics that must be mapped. */
static i_record *intrinsics;

/** An array to cache all entities */
30
static ir_entity *i_ents[iro_MaxOpcode];
31

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
/**
 * 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
135
	h_res = new_rd_ia32_l_ShlD(dbg, irg, block, a_h, a_l, cnt, l_res_mode);
136
137

	/* l_res = SHL a_l, cnt */
Christian Würdig's avatar
Christian Würdig committed
138
	l_res = new_rd_ia32_l_Shl(dbg, irg, block, a_l, cnt, h_res_mode);
139

140
	//add_irn_dep(l_res, h_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

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

	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 */
194
	h_res = new_rd_ia32_l_Sar(dbg, irg, block, a_h, cnt, h_res_mode);
195

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

	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
	*/
228
229
230
	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);
231

232
	mul   = new_rd_ia32_l_Mul(dbg, irg, block, a_h, b_l);
233
	add   = new_rd_ia32_l_Add(dbg, irg, block, mul, pEDX, h_res_mode);
234
	mul   = new_rd_ia32_l_Mul(dbg, irg, block, a_l, b_h);
235
236
237
238
239
240
241
	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;
}

242
243
244
245
246
247
248
249
250
251
252
253
254
/**
 * 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));
255
	ir_node  *l_res, *h_res, *cnst, *res;
256

257
258
	/* too bad: we need 0 in a register here */
	cnst  = new_Const_long(h_res_mode, 0);
259

260
	/* l_res = 0 - a_l */
261
262
	/* h_res = 0 - a_h - carry */

263
264
265
	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);
266
267
268
269
270
271

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

	return 1;
}

272
273
274
275
276
277
278
279
280
281
282
283
284
/**
 * 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));
285
	ir_node  *l_res, *h_res, *sign, *sub_l, *sub_h, *res;
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300

	/*
		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

	*/

301
302
303
	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);
304
305
306
	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);
307
308
309
310
311
312

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

	return 1;
}

Christian Würdig's avatar
Christian Würdig committed
313
314
315
316
317
typedef enum {
	IA32_INTRINSIC_DIV,
	IA32_INTRINSIC_MOD,
} ia32_intrinsic_divmod_t;

318
319
320
/**
 * Maps a Div/Mod (a_l, a_h, b_l, b_h)
 */
Christian Würdig's avatar
Christian Würdig committed
321
static int DivMod_mapper(ir_node *call, void *ctx, ia32_intrinsic_divmod_t dmtp) {
322
	ia32_intrinsic_env_t *env = ctx;
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
	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;
341
342
343
344
345

	/* 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);
346
		env->irg = irg;
347
348
349
350
351
	}

	if (! ent_b) {
		ent_b = env->ll_div_op2 =
			frame_alloc_area(get_irg_frame_type(irg), 2 * mode_bytes, 16, 0);
352
		env->irg = irg;
353
354
355
356
357
358
359
360
361
	}

	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));
362
	op_mem[0] = store_l;
363
364
365

	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);
366
	add_ia32_am_offs_int(store_h, mode_bytes);
367
368
	set_ia32_use_frame(store_h);
	set_ia32_ls_mode(store_h, get_irn_mode(a_h));
369
	op_mem[1] = store_h;
370
371
372
373
374
375
376
377
378

	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);
379
	fa     = new_r_Proj(irg, block, fa, mode_E, pn_ia32_l_vfild_res);
380
381
382
383
384
385

	/* 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));
386
	op_mem[0] = store_l;
387
388
389

	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);
390
	add_ia32_am_offs_int(store_h, mode_bytes);
391
392
	set_ia32_use_frame(store_h);
	set_ia32_ls_mode(store_h, get_irn_mode(b_h));
393
	op_mem[1] = store_h;
394
395
396
397
398
399
400
401
402

	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);
403
	fb     = new_r_Proj(irg, block, fb, mode_E, pn_ia32_l_vfild_res);
404
405
406
407
408
409
410

	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
411
412
	switch (dmtp) {
		case IA32_INTRINSIC_DIV:
413
			fres = new_rd_ia32_l_vfdiv(dbg, irg, block, fa, fb);
414
			fres = new_rd_Proj(dbg, irg, block, fres, mode_E, pn_ia32_l_vfdiv_res);
Christian Würdig's avatar
Christian Würdig committed
415
416
			break;
		case IA32_INTRINSIC_MOD:
417
			fres = new_rd_ia32_l_vfprem(dbg, irg, block, fa, fb, mode_E);
Christian Würdig's avatar
Christian Würdig committed
418
419
420
			break;
		default:
			assert(0);
421
422
423
424
425
426
427
	}

	/* 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);
428
	mem = fres;
429
430
431
432
433
434
435
436
437
438
439

	/* 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);
440
	add_ia32_am_offs_int(h_res, mode_bytes);
441
442
443
444
445
446
447
448
449
450
451
	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
452
	return DivMod_mapper(call, ctx, IA32_INTRINSIC_DIV);
453
454
455
}

static int map_Mod(ir_node *call, void *ctx) {
Christian Würdig's avatar
Christian Würdig committed
456
	return DivMod_mapper(call, ctx, IA32_INTRINSIC_MOD);
457
458
459
460
461
462
463
}

/**
 * Maps a Conv (a_l, a_h)
 */
static int map_Conv(ir_node *call, void *ctx) {
	ia32_intrinsic_env_t *env = ctx;
464
465
466
467
468
469
470
471
472
473
474
	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;
475
476
477
478
479
480
481
482
483
484

	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 */
485
		ent = env->irg == irg ? env->d_ll_conv : NULL;
486
		if (! ent) {
487
488
			ent      = env->d_ll_conv = frame_alloc_area(get_irg_frame_type(irg), 2 * gp_bytes, 16, 0);
			env->irg = irg;
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
		}

		/* 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);
510
		mem = a_f;
511
512
513
514
515
516
517
518
519
520
521

		/* 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);
522
		add_ia32_am_offs_int(h_res, gp_bytes);
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
		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 */
541
		ent = env->irg == irg ? env->ll_d_conv : NULL;
542
543
		if (! ent) {
			ent = env->ll_d_conv = frame_alloc_area(get_irg_frame_type(irg), 2 * gp_bytes, 16, 0);
544
			env->irg = irg;
545
546
547
548
549
550
551
552
553
554
		}

		/* 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));
555
		op_mem[0] = store_l;
556
557
558
559

		/* 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);
560
		add_ia32_am_offs_int(store_h, gp_bytes);
561
562
		set_ia32_use_frame(store_h);
		set_ia32_ls_mode(store_h, get_irn_mode(a_h));
563
		op_mem[1] = store_h;
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
591
592
593
594
595

		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;
}

596
/* Ia32 implementation of intrinsic mapping. */
597
598
599
ir_entity *ia32_create_intrinsic_fkt(ir_type *method, const ir_op *op,
                                     const ir_mode *imode, const ir_mode *omode,
                                     void *context)
600
601
{
	i_record      elt;
602
	ir_entity     **ent = NULL;
603
604
605
606
607
608
609
610
611
612
613
614
615
616
	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;
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
	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;
633
634
635
636
637
638
639
640
	case iro_Minus:
		ent    = &i_ents[iro_Minus];
		mapper = map_Minus;
		break;
	case iro_Abs:
		ent    = &i_ents[iro_Abs];
		mapper = map_Abs;
		break;
641
642
643
644
645
646
647
648
649
650
651
652
	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;
653
	default:
Christian Würdig's avatar
Christian Würdig committed
654
		fprintf(stderr, "FIXME: unhandled op for ia32 intrinsic function %s\n", get_id_str(op->name));
655
656
657
658
659
660
		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
661
		ident *id = mangle(IDENT("L"), get_op_ident(op));
662
663
664
665
666
667
		*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;
668
	elt.i_call.ctx      = context;
669
670
671
	elt.i_call.link     = NULL;

	ARR_APP1(i_record, intrinsics, elt);
Sebastian Hack's avatar
Sebastian Hack committed
672
	return *ent;
673
}