re PR target/60610 (ICE in convert_regs_1, at reg-stack.c:3064)
[gcc.git] / gcc / config / i386 / i386.h
1 /* Definitions of target machine for GCC for IA-32.
2 Copyright (C) 1988-2014 Free Software Foundation, Inc.
3
4 This file is part of GCC.
5
6 GCC is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3, or (at your option)
9 any later version.
10
11 GCC is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
15
16 Under Section 7 of GPL version 3, you are granted additional
17 permissions described in the GCC Runtime Library Exception, version
18 3.1, as published by the Free Software Foundation.
19
20 You should have received a copy of the GNU General Public License and
21 a copy of the GCC Runtime Library Exception along with this program;
22 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
23 <http://www.gnu.org/licenses/>. */
24
25 /* The purpose of this file is to define the characteristics of the i386,
26 independent of assembler syntax or operating system.
27
28 Three other files build on this one to describe a specific assembler syntax:
29 bsd386.h, att386.h, and sun386.h.
30
31 The actual tm.h file for a particular system should include
32 this file, and then the file for the appropriate assembler syntax.
33
34 Many macros that specify assembler syntax are omitted entirely from
35 this file because they really belong in the files for particular
36 assemblers. These include RP, IP, LPREFIX, PUT_OP_SIZE, USE_STAR,
37 ADDR_BEG, ADDR_END, PRINT_IREG, PRINT_SCALE, PRINT_B_I_S, and many
38 that start with ASM_ or end in ASM_OP. */
39
40 /* Redefines for option macros. */
41
42 #define TARGET_64BIT TARGET_ISA_64BIT
43 #define TARGET_64BIT_P(x) TARGET_ISA_64BIT_P(x)
44 #define TARGET_MMX TARGET_ISA_MMX
45 #define TARGET_MMX_P(x) TARGET_ISA_MMX_P(x)
46 #define TARGET_3DNOW TARGET_ISA_3DNOW
47 #define TARGET_3DNOW_P(x) TARGET_ISA_3DNOW_P(x)
48 #define TARGET_3DNOW_A TARGET_ISA_3DNOW_A
49 #define TARGET_3DNOW_A_P(x) TARGET_ISA_3DNOW_A_P(x)
50 #define TARGET_SSE TARGET_ISA_SSE
51 #define TARGET_SSE_P(x) TARGET_ISA_SSE_P(x)
52 #define TARGET_SSE2 TARGET_ISA_SSE2
53 #define TARGET_SSE2_P(x) TARGET_ISA_SSE2_P(x)
54 #define TARGET_SSE3 TARGET_ISA_SSE3
55 #define TARGET_SSE3_P(x) TARGET_ISA_SSE3_P(x)
56 #define TARGET_SSSE3 TARGET_ISA_SSSE3
57 #define TARGET_SSSE3_P(x) TARGET_ISA_SSSE3_P(x)
58 #define TARGET_SSE4_1 TARGET_ISA_SSE4_1
59 #define TARGET_SSE4_1_P(x) TARGET_ISA_SSE4_1_P(x)
60 #define TARGET_SSE4_2 TARGET_ISA_SSE4_2
61 #define TARGET_SSE4_2_P(x) TARGET_ISA_SSE4_2_P(x)
62 #define TARGET_AVX TARGET_ISA_AVX
63 #define TARGET_AVX_P(x) TARGET_ISA_AVX_P(x)
64 #define TARGET_AVX2 TARGET_ISA_AVX2
65 #define TARGET_AVX2_P(x) TARGET_ISA_AVX2_P(x)
66 #define TARGET_AVX512F TARGET_ISA_AVX512F
67 #define TARGET_AVX512F_P(x) TARGET_ISA_AVX512F_P(x)
68 #define TARGET_AVX512PF TARGET_ISA_AVX512PF
69 #define TARGET_AVX512PF_P(x) TARGET_ISA_AVX512PF_P(x)
70 #define TARGET_AVX512ER TARGET_ISA_AVX512ER
71 #define TARGET_AVX512ER_P(x) TARGET_ISA_AVX512ER_P(x)
72 #define TARGET_AVX512CD TARGET_ISA_AVX512CD
73 #define TARGET_AVX512CD_P(x) TARGET_ISA_AVX512CD_P(x)
74 #define TARGET_FMA TARGET_ISA_FMA
75 #define TARGET_FMA_P(x) TARGET_ISA_FMA_P(x)
76 #define TARGET_SSE4A TARGET_ISA_SSE4A
77 #define TARGET_SSE4A_P(x) TARGET_ISA_SSE4A_P(x)
78 #define TARGET_FMA4 TARGET_ISA_FMA4
79 #define TARGET_FMA4_P(x) TARGET_ISA_FMA4_P(x)
80 #define TARGET_XOP TARGET_ISA_XOP
81 #define TARGET_XOP_P(x) TARGET_ISA_XOP_P(x)
82 #define TARGET_LWP TARGET_ISA_LWP
83 #define TARGET_LWP_P(x) TARGET_ISA_LWP_P(x)
84 #define TARGET_ROUND TARGET_ISA_ROUND
85 #define TARGET_ABM TARGET_ISA_ABM
86 #define TARGET_ABM_P(x) TARGET_ISA_ABM_P(x)
87 #define TARGET_BMI TARGET_ISA_BMI
88 #define TARGET_BMI_P(x) TARGET_ISA_BMI_P(x)
89 #define TARGET_BMI2 TARGET_ISA_BMI2
90 #define TARGET_BMI2_P(x) TARGET_ISA_BMI2_P(x)
91 #define TARGET_LZCNT TARGET_ISA_LZCNT
92 #define TARGET_LZCNT_P(x) TARGET_ISA_LZCNT_P(x)
93 #define TARGET_TBM TARGET_ISA_TBM
94 #define TARGET_TBM_P(x) TARGET_ISA_TBM_P(x)
95 #define TARGET_POPCNT TARGET_ISA_POPCNT
96 #define TARGET_POPCNT_P(x) TARGET_ISA_POPCNT_P(x)
97 #define TARGET_SAHF TARGET_ISA_SAHF
98 #define TARGET_SAHF_P(x) TARGET_ISA_SAHF_P(x)
99 #define TARGET_MOVBE TARGET_ISA_MOVBE
100 #define TARGET_MOVBE_P(x) TARGET_ISA_MOVBE_P(x)
101 #define TARGET_CRC32 TARGET_ISA_CRC32
102 #define TARGET_CRC32_P(x) TARGET_ISA_CRC32_P(x)
103 #define TARGET_AES TARGET_ISA_AES
104 #define TARGET_AES_P(x) TARGET_ISA_AES_P(x)
105 #define TARGET_SHA TARGET_ISA_SHA
106 #define TARGET_SHA_P(x) TARGET_ISA_SHA_P(x)
107 #define TARGET_PCLMUL TARGET_ISA_PCLMUL
108 #define TARGET_PCLMUL_P(x) TARGET_ISA_PCLMUL_P(x)
109 #define TARGET_CMPXCHG16B TARGET_ISA_CX16
110 #define TARGET_CMPXCHG16B_P(x) TARGET_ISA_CX16_P(x)
111 #define TARGET_FSGSBASE TARGET_ISA_FSGSBASE
112 #define TARGET_FSGSBASE_P(x) TARGET_ISA_FSGSBASE_P(x)
113 #define TARGET_RDRND TARGET_ISA_RDRND
114 #define TARGET_RDRND_P(x) TARGET_ISA_RDRND_P(x)
115 #define TARGET_F16C TARGET_ISA_F16C
116 #define TARGET_F16C_P(x) TARGET_ISA_F16C_P(x)
117 #define TARGET_RTM TARGET_ISA_RTM
118 #define TARGET_RTM_P(x) TARGET_ISA_RTM_P(x)
119 #define TARGET_HLE TARGET_ISA_HLE
120 #define TARGET_HLE_P(x) TARGET_ISA_HLE_P(x)
121 #define TARGET_RDSEED TARGET_ISA_RDSEED
122 #define TARGET_RDSEED_P(x) TARGET_ISA_RDSEED_P(x)
123 #define TARGET_PRFCHW TARGET_ISA_PRFCHW
124 #define TARGET_PRFCHW_P(x) TARGET_ISA_PRFCHW_P(x)
125 #define TARGET_ADX TARGET_ISA_ADX
126 #define TARGET_ADX_P(x) TARGET_ISA_ADX_P(x)
127 #define TARGET_FXSR TARGET_ISA_FXSR
128 #define TARGET_FXSR_P(x) TARGET_ISA_FXSR_P(x)
129 #define TARGET_XSAVE TARGET_ISA_XSAVE
130 #define TARGET_XSAVE_P(x) TARGET_ISA_XSAVE_P(x)
131 #define TARGET_XSAVEOPT TARGET_ISA_XSAVEOPT
132 #define TARGET_XSAVEOPT_P(x) TARGET_ISA_XSAVEOPT_P(x)
133 #define TARGET_PREFETCHWT1 TARGET_ISA_PREFETCHWT1
134 #define TARGET_PREFETCHWT1_P(x) TARGET_ISA_PREFETCHWT1_P(x)
135
136 #define TARGET_LP64 TARGET_ABI_64
137 #define TARGET_LP64_P(x) TARGET_ABI_64_P(x)
138 #define TARGET_X32 TARGET_ABI_X32
139 #define TARGET_X32_P(x) TARGET_ABI_X32_P(x)
140 #define TARGET_16BIT TARGET_CODE16
141 #define TARGET_16BIT_P(x) TARGET_CODE16_P(x)
142
143 /* SSE4.1 defines round instructions */
144 #define OPTION_MASK_ISA_ROUND OPTION_MASK_ISA_SSE4_1
145 #define TARGET_ISA_ROUND ((ix86_isa_flags & OPTION_MASK_ISA_ROUND) != 0)
146
147 #include "config/vxworks-dummy.h"
148
149 #include "config/i386/i386-opts.h"
150
151 #define MAX_STRINGOP_ALGS 4
152
153 /* Specify what algorithm to use for stringops on known size.
154 When size is unknown, the UNKNOWN_SIZE alg is used. When size is
155 known at compile time or estimated via feedback, the SIZE array
156 is walked in order until MAX is greater then the estimate (or -1
157 means infinity). Corresponding ALG is used then.
158 When NOALIGN is true the code guaranting the alignment of the memory
159 block is skipped.
160
161 For example initializer:
162 {{256, loop}, {-1, rep_prefix_4_byte}}
163 will use loop for blocks smaller or equal to 256 bytes, rep prefix will
164 be used otherwise. */
165 struct stringop_algs
166 {
167 const enum stringop_alg unknown_size;
168 const struct stringop_strategy {
169 const int max;
170 const enum stringop_alg alg;
171 int noalign;
172 } size [MAX_STRINGOP_ALGS];
173 };
174
175 /* Define the specific costs for a given cpu */
176
177 struct processor_costs {
178 const int add; /* cost of an add instruction */
179 const int lea; /* cost of a lea instruction */
180 const int shift_var; /* variable shift costs */
181 const int shift_const; /* constant shift costs */
182 const int mult_init[5]; /* cost of starting a multiply
183 in QImode, HImode, SImode, DImode, TImode*/
184 const int mult_bit; /* cost of multiply per each bit set */
185 const int divide[5]; /* cost of a divide/mod
186 in QImode, HImode, SImode, DImode, TImode*/
187 int movsx; /* The cost of movsx operation. */
188 int movzx; /* The cost of movzx operation. */
189 const int large_insn; /* insns larger than this cost more */
190 const int move_ratio; /* The threshold of number of scalar
191 memory-to-memory move insns. */
192 const int movzbl_load; /* cost of loading using movzbl */
193 const int int_load[3]; /* cost of loading integer registers
194 in QImode, HImode and SImode relative
195 to reg-reg move (2). */
196 const int int_store[3]; /* cost of storing integer register
197 in QImode, HImode and SImode */
198 const int fp_move; /* cost of reg,reg fld/fst */
199 const int fp_load[3]; /* cost of loading FP register
200 in SFmode, DFmode and XFmode */
201 const int fp_store[3]; /* cost of storing FP register
202 in SFmode, DFmode and XFmode */
203 const int mmx_move; /* cost of moving MMX register. */
204 const int mmx_load[2]; /* cost of loading MMX register
205 in SImode and DImode */
206 const int mmx_store[2]; /* cost of storing MMX register
207 in SImode and DImode */
208 const int sse_move; /* cost of moving SSE register. */
209 const int sse_load[3]; /* cost of loading SSE register
210 in SImode, DImode and TImode*/
211 const int sse_store[3]; /* cost of storing SSE register
212 in SImode, DImode and TImode*/
213 const int mmxsse_to_integer; /* cost of moving mmxsse register to
214 integer and vice versa. */
215 const int l1_cache_size; /* size of l1 cache, in kilobytes. */
216 const int l2_cache_size; /* size of l2 cache, in kilobytes. */
217 const int prefetch_block; /* bytes moved to cache for prefetch. */
218 const int simultaneous_prefetches; /* number of parallel prefetch
219 operations. */
220 const int branch_cost; /* Default value for BRANCH_COST. */
221 const int fadd; /* cost of FADD and FSUB instructions. */
222 const int fmul; /* cost of FMUL instruction. */
223 const int fdiv; /* cost of FDIV instruction. */
224 const int fabs; /* cost of FABS instruction. */
225 const int fchs; /* cost of FCHS instruction. */
226 const int fsqrt; /* cost of FSQRT instruction. */
227 /* Specify what algorithm
228 to use for stringops on unknown size. */
229 struct stringop_algs *memcpy, *memset;
230 const int scalar_stmt_cost; /* Cost of any scalar operation, excluding
231 load and store. */
232 const int scalar_load_cost; /* Cost of scalar load. */
233 const int scalar_store_cost; /* Cost of scalar store. */
234 const int vec_stmt_cost; /* Cost of any vector operation, excluding
235 load, store, vector-to-scalar and
236 scalar-to-vector operation. */
237 const int vec_to_scalar_cost; /* Cost of vect-to-scalar operation. */
238 const int scalar_to_vec_cost; /* Cost of scalar-to-vector operation. */
239 const int vec_align_load_cost; /* Cost of aligned vector load. */
240 const int vec_unalign_load_cost; /* Cost of unaligned vector load. */
241 const int vec_store_cost; /* Cost of vector store. */
242 const int cond_taken_branch_cost; /* Cost of taken branch for vectorizer
243 cost model. */
244 const int cond_not_taken_branch_cost;/* Cost of not taken branch for
245 vectorizer cost model. */
246 };
247
248 extern const struct processor_costs *ix86_cost;
249 extern const struct processor_costs ix86_size_cost;
250
251 #define ix86_cur_cost() \
252 (optimize_insn_for_size_p () ? &ix86_size_cost: ix86_cost)
253
254 /* Macros used in the machine description to test the flags. */
255
256 /* configure can arrange to change it. */
257
258 #ifndef TARGET_CPU_DEFAULT
259 #define TARGET_CPU_DEFAULT PROCESSOR_GENERIC
260 #endif
261
262 #ifndef TARGET_FPMATH_DEFAULT
263 #define TARGET_FPMATH_DEFAULT \
264 (TARGET_64BIT && TARGET_SSE ? FPMATH_SSE : FPMATH_387)
265 #endif
266
267 #ifndef TARGET_FPMATH_DEFAULT_P
268 #define TARGET_FPMATH_DEFAULT_P(x) \
269 (TARGET_64BIT_P(x) && TARGET_SSE_P(x) ? FPMATH_SSE : FPMATH_387)
270 #endif
271
272 #define TARGET_FLOAT_RETURNS_IN_80387 TARGET_FLOAT_RETURNS
273 #define TARGET_FLOAT_RETURNS_IN_80387_P(x) TARGET_FLOAT_RETURNS_P(x)
274
275 /* 64bit Sledgehammer mode. For libgcc2 we make sure this is a
276 compile-time constant. */
277 #ifdef IN_LIBGCC2
278 #undef TARGET_64BIT
279 #ifdef __x86_64__
280 #define TARGET_64BIT 1
281 #else
282 #define TARGET_64BIT 0
283 #endif
284 #else
285 #ifndef TARGET_BI_ARCH
286 #undef TARGET_64BIT
287 #undef TARGET_64BIT_P
288 #if TARGET_64BIT_DEFAULT
289 #define TARGET_64BIT 1
290 #define TARGET_64BIT_P(x) 1
291 #else
292 #define TARGET_64BIT 0
293 #define TARGET_64BIT_P(x) 0
294 #endif
295 #endif
296 #endif
297
298 #define HAS_LONG_COND_BRANCH 1
299 #define HAS_LONG_UNCOND_BRANCH 1
300
301 #define TARGET_386 (ix86_tune == PROCESSOR_I386)
302 #define TARGET_486 (ix86_tune == PROCESSOR_I486)
303 #define TARGET_PENTIUM (ix86_tune == PROCESSOR_PENTIUM)
304 #define TARGET_PENTIUMPRO (ix86_tune == PROCESSOR_PENTIUMPRO)
305 #define TARGET_GEODE (ix86_tune == PROCESSOR_GEODE)
306 #define TARGET_K6 (ix86_tune == PROCESSOR_K6)
307 #define TARGET_ATHLON (ix86_tune == PROCESSOR_ATHLON)
308 #define TARGET_PENTIUM4 (ix86_tune == PROCESSOR_PENTIUM4)
309 #define TARGET_K8 (ix86_tune == PROCESSOR_K8)
310 #define TARGET_ATHLON_K8 (TARGET_K8 || TARGET_ATHLON)
311 #define TARGET_NOCONA (ix86_tune == PROCESSOR_NOCONA)
312 #define TARGET_CORE2 (ix86_tune == PROCESSOR_CORE2)
313 #define TARGET_NEHALEM (ix86_tune == PROCESSOR_NEHALEM)
314 #define TARGET_SANDYBRIDGE (ix86_tune == PROCESSOR_SANDYBRIDGE)
315 #define TARGET_HASWELL (ix86_tune == PROCESSOR_HASWELL)
316 #define TARGET_BONNELL (ix86_tune == PROCESSOR_BONNELL)
317 #define TARGET_SILVERMONT (ix86_tune == PROCESSOR_SILVERMONT)
318 #define TARGET_INTEL (ix86_tune == PROCESSOR_INTEL)
319 #define TARGET_GENERIC (ix86_tune == PROCESSOR_GENERIC)
320 #define TARGET_AMDFAM10 (ix86_tune == PROCESSOR_AMDFAM10)
321 #define TARGET_BDVER1 (ix86_tune == PROCESSOR_BDVER1)
322 #define TARGET_BDVER2 (ix86_tune == PROCESSOR_BDVER2)
323 #define TARGET_BDVER3 (ix86_tune == PROCESSOR_BDVER3)
324 #define TARGET_BDVER4 (ix86_tune == PROCESSOR_BDVER4)
325 #define TARGET_BTVER1 (ix86_tune == PROCESSOR_BTVER1)
326 #define TARGET_BTVER2 (ix86_tune == PROCESSOR_BTVER2)
327
328 /* Feature tests against the various tunings. */
329 enum ix86_tune_indices {
330 #undef DEF_TUNE
331 #define DEF_TUNE(tune, name, selector) tune,
332 #include "x86-tune.def"
333 #undef DEF_TUNE
334 X86_TUNE_LAST
335 };
336
337 extern unsigned char ix86_tune_features[X86_TUNE_LAST];
338
339 #define TARGET_USE_LEAVE ix86_tune_features[X86_TUNE_USE_LEAVE]
340 #define TARGET_PUSH_MEMORY ix86_tune_features[X86_TUNE_PUSH_MEMORY]
341 #define TARGET_ZERO_EXTEND_WITH_AND \
342 ix86_tune_features[X86_TUNE_ZERO_EXTEND_WITH_AND]
343 #define TARGET_UNROLL_STRLEN ix86_tune_features[X86_TUNE_UNROLL_STRLEN]
344 #define TARGET_BRANCH_PREDICTION_HINTS \
345 ix86_tune_features[X86_TUNE_BRANCH_PREDICTION_HINTS]
346 #define TARGET_DOUBLE_WITH_ADD ix86_tune_features[X86_TUNE_DOUBLE_WITH_ADD]
347 #define TARGET_USE_SAHF ix86_tune_features[X86_TUNE_USE_SAHF]
348 #define TARGET_MOVX ix86_tune_features[X86_TUNE_MOVX]
349 #define TARGET_PARTIAL_REG_STALL ix86_tune_features[X86_TUNE_PARTIAL_REG_STALL]
350 #define TARGET_PARTIAL_FLAG_REG_STALL \
351 ix86_tune_features[X86_TUNE_PARTIAL_FLAG_REG_STALL]
352 #define TARGET_LCP_STALL \
353 ix86_tune_features[X86_TUNE_LCP_STALL]
354 #define TARGET_USE_HIMODE_FIOP ix86_tune_features[X86_TUNE_USE_HIMODE_FIOP]
355 #define TARGET_USE_SIMODE_FIOP ix86_tune_features[X86_TUNE_USE_SIMODE_FIOP]
356 #define TARGET_USE_MOV0 ix86_tune_features[X86_TUNE_USE_MOV0]
357 #define TARGET_USE_CLTD ix86_tune_features[X86_TUNE_USE_CLTD]
358 #define TARGET_USE_XCHGB ix86_tune_features[X86_TUNE_USE_XCHGB]
359 #define TARGET_SPLIT_LONG_MOVES ix86_tune_features[X86_TUNE_SPLIT_LONG_MOVES]
360 #define TARGET_READ_MODIFY_WRITE ix86_tune_features[X86_TUNE_READ_MODIFY_WRITE]
361 #define TARGET_READ_MODIFY ix86_tune_features[X86_TUNE_READ_MODIFY]
362 #define TARGET_PROMOTE_QImode ix86_tune_features[X86_TUNE_PROMOTE_QIMODE]
363 #define TARGET_FAST_PREFIX ix86_tune_features[X86_TUNE_FAST_PREFIX]
364 #define TARGET_SINGLE_STRINGOP ix86_tune_features[X86_TUNE_SINGLE_STRINGOP]
365 #define TARGET_MISALIGNED_MOVE_STRING_PRO_EPILOGUES \
366 ix86_tune_features[X86_TUNE_MISALIGNED_MOVE_STRING_PRO_EPILOGUES]
367 #define TARGET_QIMODE_MATH ix86_tune_features[X86_TUNE_QIMODE_MATH]
368 #define TARGET_HIMODE_MATH ix86_tune_features[X86_TUNE_HIMODE_MATH]
369 #define TARGET_PROMOTE_QI_REGS ix86_tune_features[X86_TUNE_PROMOTE_QI_REGS]
370 #define TARGET_PROMOTE_HI_REGS ix86_tune_features[X86_TUNE_PROMOTE_HI_REGS]
371 #define TARGET_SINGLE_POP ix86_tune_features[X86_TUNE_SINGLE_POP]
372 #define TARGET_DOUBLE_POP ix86_tune_features[X86_TUNE_DOUBLE_POP]
373 #define TARGET_SINGLE_PUSH ix86_tune_features[X86_TUNE_SINGLE_PUSH]
374 #define TARGET_DOUBLE_PUSH ix86_tune_features[X86_TUNE_DOUBLE_PUSH]
375 #define TARGET_INTEGER_DFMODE_MOVES \
376 ix86_tune_features[X86_TUNE_INTEGER_DFMODE_MOVES]
377 #define TARGET_PARTIAL_REG_DEPENDENCY \
378 ix86_tune_features[X86_TUNE_PARTIAL_REG_DEPENDENCY]
379 #define TARGET_SSE_PARTIAL_REG_DEPENDENCY \
380 ix86_tune_features[X86_TUNE_SSE_PARTIAL_REG_DEPENDENCY]
381 #define TARGET_SSE_UNALIGNED_LOAD_OPTIMAL \
382 ix86_tune_features[X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL]
383 #define TARGET_SSE_UNALIGNED_STORE_OPTIMAL \
384 ix86_tune_features[X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL]
385 #define TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL \
386 ix86_tune_features[X86_TUNE_SSE_PACKED_SINGLE_INSN_OPTIMAL]
387 #define TARGET_SSE_SPLIT_REGS ix86_tune_features[X86_TUNE_SSE_SPLIT_REGS]
388 #define TARGET_SSE_TYPELESS_STORES \
389 ix86_tune_features[X86_TUNE_SSE_TYPELESS_STORES]
390 #define TARGET_SSE_LOAD0_BY_PXOR ix86_tune_features[X86_TUNE_SSE_LOAD0_BY_PXOR]
391 #define TARGET_MEMORY_MISMATCH_STALL \
392 ix86_tune_features[X86_TUNE_MEMORY_MISMATCH_STALL]
393 #define TARGET_PROLOGUE_USING_MOVE \
394 ix86_tune_features[X86_TUNE_PROLOGUE_USING_MOVE]
395 #define TARGET_EPILOGUE_USING_MOVE \
396 ix86_tune_features[X86_TUNE_EPILOGUE_USING_MOVE]
397 #define TARGET_SHIFT1 ix86_tune_features[X86_TUNE_SHIFT1]
398 #define TARGET_USE_FFREEP ix86_tune_features[X86_TUNE_USE_FFREEP]
399 #define TARGET_INTER_UNIT_MOVES_TO_VEC \
400 ix86_tune_features[X86_TUNE_INTER_UNIT_MOVES_TO_VEC]
401 #define TARGET_INTER_UNIT_MOVES_FROM_VEC \
402 ix86_tune_features[X86_TUNE_INTER_UNIT_MOVES_FROM_VEC]
403 #define TARGET_INTER_UNIT_CONVERSIONS \
404 ix86_tune_features[X86_TUNE_INTER_UNIT_CONVERSIONS]
405 #define TARGET_FOUR_JUMP_LIMIT ix86_tune_features[X86_TUNE_FOUR_JUMP_LIMIT]
406 #define TARGET_SCHEDULE ix86_tune_features[X86_TUNE_SCHEDULE]
407 #define TARGET_USE_BT ix86_tune_features[X86_TUNE_USE_BT]
408 #define TARGET_USE_INCDEC ix86_tune_features[X86_TUNE_USE_INCDEC]
409 #define TARGET_PAD_RETURNS ix86_tune_features[X86_TUNE_PAD_RETURNS]
410 #define TARGET_PAD_SHORT_FUNCTION \
411 ix86_tune_features[X86_TUNE_PAD_SHORT_FUNCTION]
412 #define TARGET_EXT_80387_CONSTANTS \
413 ix86_tune_features[X86_TUNE_EXT_80387_CONSTANTS]
414 #define TARGET_AVOID_VECTOR_DECODE \
415 ix86_tune_features[X86_TUNE_AVOID_VECTOR_DECODE]
416 #define TARGET_TUNE_PROMOTE_HIMODE_IMUL \
417 ix86_tune_features[X86_TUNE_PROMOTE_HIMODE_IMUL]
418 #define TARGET_SLOW_IMUL_IMM32_MEM \
419 ix86_tune_features[X86_TUNE_SLOW_IMUL_IMM32_MEM]
420 #define TARGET_SLOW_IMUL_IMM8 ix86_tune_features[X86_TUNE_SLOW_IMUL_IMM8]
421 #define TARGET_MOVE_M1_VIA_OR ix86_tune_features[X86_TUNE_MOVE_M1_VIA_OR]
422 #define TARGET_NOT_UNPAIRABLE ix86_tune_features[X86_TUNE_NOT_UNPAIRABLE]
423 #define TARGET_NOT_VECTORMODE ix86_tune_features[X86_TUNE_NOT_VECTORMODE]
424 #define TARGET_USE_VECTOR_FP_CONVERTS \
425 ix86_tune_features[X86_TUNE_USE_VECTOR_FP_CONVERTS]
426 #define TARGET_USE_VECTOR_CONVERTS \
427 ix86_tune_features[X86_TUNE_USE_VECTOR_CONVERTS]
428 #define TARGET_FUSE_CMP_AND_BRANCH_32 \
429 ix86_tune_features[X86_TUNE_FUSE_CMP_AND_BRANCH_32]
430 #define TARGET_FUSE_CMP_AND_BRANCH_64 \
431 ix86_tune_features[X86_TUNE_FUSE_CMP_AND_BRANCH_64]
432 #define TARGET_FUSE_CMP_AND_BRANCH \
433 (TARGET_64BIT ? TARGET_FUSE_CMP_AND_BRANCH_64 \
434 : TARGET_FUSE_CMP_AND_BRANCH_32)
435 #define TARGET_FUSE_CMP_AND_BRANCH_SOFLAGS \
436 ix86_tune_features[X86_TUNE_FUSE_CMP_AND_BRANCH_SOFLAGS]
437 #define TARGET_FUSE_ALU_AND_BRANCH \
438 ix86_tune_features[X86_TUNE_FUSE_ALU_AND_BRANCH]
439 #define TARGET_OPT_AGU ix86_tune_features[X86_TUNE_OPT_AGU]
440 #define TARGET_AVOID_LEA_FOR_ADDR \
441 ix86_tune_features[X86_TUNE_AVOID_LEA_FOR_ADDR]
442 #define TARGET_VECTORIZE_DOUBLE \
443 ix86_tune_features[X86_TUNE_VECTORIZE_DOUBLE]
444 #define TARGET_SOFTWARE_PREFETCHING_BENEFICIAL \
445 ix86_tune_features[X86_TUNE_SOFTWARE_PREFETCHING_BENEFICIAL]
446 #define TARGET_AVX128_OPTIMAL \
447 ix86_tune_features[X86_TUNE_AVX128_OPTIMAL]
448 #define TARGET_REASSOC_INT_TO_PARALLEL \
449 ix86_tune_features[X86_TUNE_REASSOC_INT_TO_PARALLEL]
450 #define TARGET_REASSOC_FP_TO_PARALLEL \
451 ix86_tune_features[X86_TUNE_REASSOC_FP_TO_PARALLEL]
452 #define TARGET_GENERAL_REGS_SSE_SPILL \
453 ix86_tune_features[X86_TUNE_GENERAL_REGS_SSE_SPILL]
454 #define TARGET_AVOID_MEM_OPND_FOR_CMOVE \
455 ix86_tune_features[X86_TUNE_AVOID_MEM_OPND_FOR_CMOVE]
456 #define TARGET_SPLIT_MEM_OPND_FOR_FP_CONVERTS \
457 ix86_tune_features[X86_TUNE_SPLIT_MEM_OPND_FOR_FP_CONVERTS]
458 #define TARGET_ADJUST_UNROLL \
459 ix86_tune_features[X86_TUNE_ADJUST_UNROLL]
460
461 /* Feature tests against the various architecture variations. */
462 enum ix86_arch_indices {
463 X86_ARCH_CMOV,
464 X86_ARCH_CMPXCHG,
465 X86_ARCH_CMPXCHG8B,
466 X86_ARCH_XADD,
467 X86_ARCH_BSWAP,
468
469 X86_ARCH_LAST
470 };
471
472 extern unsigned char ix86_arch_features[X86_ARCH_LAST];
473
474 #define TARGET_CMOV ix86_arch_features[X86_ARCH_CMOV]
475 #define TARGET_CMPXCHG ix86_arch_features[X86_ARCH_CMPXCHG]
476 #define TARGET_CMPXCHG8B ix86_arch_features[X86_ARCH_CMPXCHG8B]
477 #define TARGET_XADD ix86_arch_features[X86_ARCH_XADD]
478 #define TARGET_BSWAP ix86_arch_features[X86_ARCH_BSWAP]
479
480 /* For sane SSE instruction set generation we need fcomi instruction.
481 It is safe to enable all CMOVE instructions. Also, RDRAND intrinsic
482 expands to a sequence that includes conditional move. */
483 #define TARGET_CMOVE (TARGET_CMOV || TARGET_SSE || TARGET_RDRND)
484
485 #define TARGET_FISTTP (TARGET_SSE3 && TARGET_80387)
486
487 extern unsigned char x86_prefetch_sse;
488 #define TARGET_PREFETCH_SSE x86_prefetch_sse
489
490 #define ASSEMBLER_DIALECT (ix86_asm_dialect)
491
492 #define TARGET_SSE_MATH ((ix86_fpmath & FPMATH_SSE) != 0)
493 #define TARGET_MIX_SSE_I387 \
494 ((ix86_fpmath & (FPMATH_SSE | FPMATH_387)) == (FPMATH_SSE | FPMATH_387))
495
496 #define TARGET_GNU_TLS (ix86_tls_dialect == TLS_DIALECT_GNU)
497 #define TARGET_GNU2_TLS (ix86_tls_dialect == TLS_DIALECT_GNU2)
498 #define TARGET_ANY_GNU_TLS (TARGET_GNU_TLS || TARGET_GNU2_TLS)
499 #define TARGET_SUN_TLS 0
500
501 #ifndef TARGET_64BIT_DEFAULT
502 #define TARGET_64BIT_DEFAULT 0
503 #endif
504 #ifndef TARGET_TLS_DIRECT_SEG_REFS_DEFAULT
505 #define TARGET_TLS_DIRECT_SEG_REFS_DEFAULT 0
506 #endif
507
508 #define TARGET_SSP_GLOBAL_GUARD (ix86_stack_protector_guard == SSP_GLOBAL)
509 #define TARGET_SSP_TLS_GUARD (ix86_stack_protector_guard == SSP_TLS)
510
511 /* Fence to use after loop using storent. */
512
513 extern tree x86_mfence;
514 #define FENCE_FOLLOWING_MOVNT x86_mfence
515
516 /* Once GDB has been enhanced to deal with functions without frame
517 pointers, we can change this to allow for elimination of
518 the frame pointer in leaf functions. */
519 #define TARGET_DEFAULT 0
520
521 /* Extra bits to force. */
522 #define TARGET_SUBTARGET_DEFAULT 0
523 #define TARGET_SUBTARGET_ISA_DEFAULT 0
524
525 /* Extra bits to force on w/ 32-bit mode. */
526 #define TARGET_SUBTARGET32_DEFAULT 0
527 #define TARGET_SUBTARGET32_ISA_DEFAULT 0
528
529 /* Extra bits to force on w/ 64-bit mode. */
530 #define TARGET_SUBTARGET64_DEFAULT 0
531 #define TARGET_SUBTARGET64_ISA_DEFAULT 0
532
533 /* Replace MACH-O, ifdefs by in-line tests, where possible.
534 (a) Macros defined in config/i386/darwin.h */
535 #define TARGET_MACHO 0
536 #define TARGET_MACHO_BRANCH_ISLANDS 0
537 #define MACHOPIC_ATT_STUB 0
538 /* (b) Macros defined in config/darwin.h */
539 #define MACHO_DYNAMIC_NO_PIC_P 0
540 #define MACHOPIC_INDIRECT 0
541 #define MACHOPIC_PURE 0
542
543 /* For the RDOS */
544 #define TARGET_RDOS 0
545
546 /* For the Windows 64-bit ABI. */
547 #define TARGET_64BIT_MS_ABI (TARGET_64BIT && ix86_cfun_abi () == MS_ABI)
548
549 /* For the Windows 32-bit ABI. */
550 #define TARGET_32BIT_MS_ABI (!TARGET_64BIT && ix86_cfun_abi () == MS_ABI)
551
552 /* This is re-defined by cygming.h. */
553 #define TARGET_SEH 0
554
555 /* This is re-defined by cygming.h. */
556 #define TARGET_PECOFF 0
557
558 /* The default abi used by target. */
559 #define DEFAULT_ABI SYSV_ABI
560
561 /* The default TLS segment register used by target. */
562 #define DEFAULT_TLS_SEG_REG (TARGET_64BIT ? SEG_FS : SEG_GS)
563
564 /* Subtargets may reset this to 1 in order to enable 96-bit long double
565 with the rounding mode forced to 53 bits. */
566 #define TARGET_96_ROUND_53_LONG_DOUBLE 0
567
568 /* -march=native handling only makes sense with compiler running on
569 an x86 or x86_64 chip. If changing this condition, also change
570 the condition in driver-i386.c. */
571 #if defined(__i386__) || defined(__x86_64__)
572 /* In driver-i386.c. */
573 extern const char *host_detect_local_cpu (int argc, const char **argv);
574 #define EXTRA_SPEC_FUNCTIONS \
575 { "local_cpu_detect", host_detect_local_cpu },
576 #define HAVE_LOCAL_CPU_DETECT
577 #endif
578
579 #if TARGET_64BIT_DEFAULT
580 #define OPT_ARCH64 "!m32"
581 #define OPT_ARCH32 "m32"
582 #else
583 #define OPT_ARCH64 "m64|mx32"
584 #define OPT_ARCH32 "m64|mx32:;"
585 #endif
586
587 /* Support for configure-time defaults of some command line options.
588 The order here is important so that -march doesn't squash the
589 tune or cpu values. */
590 #define OPTION_DEFAULT_SPECS \
591 {"tune", "%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}" }, \
592 {"tune_32", "%{" OPT_ARCH32 ":%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}}" }, \
593 {"tune_64", "%{" OPT_ARCH64 ":%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}}" }, \
594 {"cpu", "%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}" }, \
595 {"cpu_32", "%{" OPT_ARCH32 ":%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}}" }, \
596 {"cpu_64", "%{" OPT_ARCH64 ":%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}}" }, \
597 {"arch", "%{!march=*:-march=%(VALUE)}"}, \
598 {"arch_32", "%{" OPT_ARCH32 ":%{!march=*:-march=%(VALUE)}}"}, \
599 {"arch_64", "%{" OPT_ARCH64 ":%{!march=*:-march=%(VALUE)}}"},
600
601 /* Specs for the compiler proper */
602
603 #ifndef CC1_CPU_SPEC
604 #define CC1_CPU_SPEC_1 ""
605
606 #ifndef HAVE_LOCAL_CPU_DETECT
607 #define CC1_CPU_SPEC CC1_CPU_SPEC_1
608 #else
609 #define CC1_CPU_SPEC CC1_CPU_SPEC_1 \
610 "%{march=native:%>march=native %:local_cpu_detect(arch) \
611 %{!mtune=*:%>mtune=native %:local_cpu_detect(tune)}} \
612 %{mtune=native:%>mtune=native %:local_cpu_detect(tune)}"
613 #endif
614 #endif
615 \f
616 /* Target CPU builtins. */
617 #define TARGET_CPU_CPP_BUILTINS() ix86_target_macros ()
618
619 /* Target Pragmas. */
620 #define REGISTER_TARGET_PRAGMAS() ix86_register_pragmas ()
621
622 #ifndef CC1_SPEC
623 #define CC1_SPEC "%(cc1_cpu) "
624 #endif
625
626 /* This macro defines names of additional specifications to put in the
627 specs that can be used in various specifications like CC1_SPEC. Its
628 definition is an initializer with a subgrouping for each command option.
629
630 Each subgrouping contains a string constant, that defines the
631 specification name, and a string constant that used by the GCC driver
632 program.
633
634 Do not define this macro if it does not need to do anything. */
635
636 #ifndef SUBTARGET_EXTRA_SPECS
637 #define SUBTARGET_EXTRA_SPECS
638 #endif
639
640 #define EXTRA_SPECS \
641 { "cc1_cpu", CC1_CPU_SPEC }, \
642 SUBTARGET_EXTRA_SPECS
643 \f
644
645 /* Set the value of FLT_EVAL_METHOD in float.h. When using only the
646 FPU, assume that the fpcw is set to extended precision; when using
647 only SSE, rounding is correct; when using both SSE and the FPU,
648 the rounding precision is indeterminate, since either may be chosen
649 apparently at random. */
650 #define TARGET_FLT_EVAL_METHOD \
651 (TARGET_MIX_SSE_I387 ? -1 : TARGET_SSE_MATH ? 0 : 2)
652
653 /* Whether to allow x87 floating-point arithmetic on MODE (one of
654 SFmode, DFmode and XFmode) in the current excess precision
655 configuration. */
656 #define X87_ENABLE_ARITH(MODE) \
657 (flag_excess_precision == EXCESS_PRECISION_FAST || (MODE) == XFmode)
658
659 /* Likewise, whether to allow direct conversions from integer mode
660 IMODE (HImode, SImode or DImode) to MODE. */
661 #define X87_ENABLE_FLOAT(MODE, IMODE) \
662 (flag_excess_precision == EXCESS_PRECISION_FAST \
663 || (MODE) == XFmode \
664 || ((MODE) == DFmode && (IMODE) == SImode) \
665 || (IMODE) == HImode)
666
667 /* target machine storage layout */
668
669 #define SHORT_TYPE_SIZE 16
670 #define INT_TYPE_SIZE 32
671 #define LONG_TYPE_SIZE (TARGET_X32 ? 32 : BITS_PER_WORD)
672 #define POINTER_SIZE (TARGET_X32 ? 32 : BITS_PER_WORD)
673 #define LONG_LONG_TYPE_SIZE 64
674 #define FLOAT_TYPE_SIZE 32
675 #define DOUBLE_TYPE_SIZE 64
676 #define LONG_DOUBLE_TYPE_SIZE \
677 (TARGET_LONG_DOUBLE_64 ? 64 : (TARGET_LONG_DOUBLE_128 ? 128 : 80))
678
679 /* Define this to set long double type size to use in libgcc2.c, which can
680 not depend on target_flags. */
681 #ifdef __LONG_DOUBLE_64__
682 #define LIBGCC2_LONG_DOUBLE_TYPE_SIZE 64
683 #elif defined (__LONG_DOUBLE_128__)
684 #define LIBGCC2_LONG_DOUBLE_TYPE_SIZE 128
685 #else
686 #define LIBGCC2_LONG_DOUBLE_TYPE_SIZE 80
687 #endif
688
689 #define WIDEST_HARDWARE_FP_SIZE 80
690
691 #if defined (TARGET_BI_ARCH) || TARGET_64BIT_DEFAULT
692 #define MAX_BITS_PER_WORD 64
693 #else
694 #define MAX_BITS_PER_WORD 32
695 #endif
696
697 /* Define this if most significant byte of a word is the lowest numbered. */
698 /* That is true on the 80386. */
699
700 #define BITS_BIG_ENDIAN 0
701
702 /* Define this if most significant byte of a word is the lowest numbered. */
703 /* That is not true on the 80386. */
704 #define BYTES_BIG_ENDIAN 0
705
706 /* Define this if most significant word of a multiword number is the lowest
707 numbered. */
708 /* Not true for 80386 */
709 #define WORDS_BIG_ENDIAN 0
710
711 /* Width of a word, in units (bytes). */
712 #define UNITS_PER_WORD (TARGET_64BIT ? 8 : 4)
713
714 #ifndef IN_LIBGCC2
715 #define MIN_UNITS_PER_WORD 4
716 #endif
717
718 /* Allocation boundary (in *bits*) for storing arguments in argument list. */
719 #define PARM_BOUNDARY BITS_PER_WORD
720
721 /* Boundary (in *bits*) on which stack pointer should be aligned. */
722 #define STACK_BOUNDARY \
723 (TARGET_64BIT && ix86_abi == MS_ABI ? 128 : BITS_PER_WORD)
724
725 /* Stack boundary of the main function guaranteed by OS. */
726 #define MAIN_STACK_BOUNDARY (TARGET_64BIT ? 128 : 32)
727
728 /* Minimum stack boundary. */
729 #define MIN_STACK_BOUNDARY (TARGET_64BIT ? (TARGET_SSE ? 128 : 64) : 32)
730
731 /* Boundary (in *bits*) on which the stack pointer prefers to be
732 aligned; the compiler cannot rely on having this alignment. */
733 #define PREFERRED_STACK_BOUNDARY ix86_preferred_stack_boundary
734
735 /* It should be MIN_STACK_BOUNDARY. But we set it to 128 bits for
736 both 32bit and 64bit, to support codes that need 128 bit stack
737 alignment for SSE instructions, but can't realign the stack. */
738 #define PREFERRED_STACK_BOUNDARY_DEFAULT 128
739
740 /* 1 if -mstackrealign should be turned on by default. It will
741 generate an alternate prologue and epilogue that realigns the
742 runtime stack if nessary. This supports mixing codes that keep a
743 4-byte aligned stack, as specified by i386 psABI, with codes that
744 need a 16-byte aligned stack, as required by SSE instructions. */
745 #define STACK_REALIGN_DEFAULT 0
746
747 /* Boundary (in *bits*) on which the incoming stack is aligned. */
748 #define INCOMING_STACK_BOUNDARY ix86_incoming_stack_boundary
749
750 /* According to Windows x64 software convention, the maximum stack allocatable
751 in the prologue is 4G - 8 bytes. Furthermore, there is a limited set of
752 instructions allowed to adjust the stack pointer in the epilog, forcing the
753 use of frame pointer for frames larger than 2 GB. This theorical limit
754 is reduced by 256, an over-estimated upper bound for the stack use by the
755 prologue.
756 We define only one threshold for both the prolog and the epilog. When the
757 frame size is larger than this threshold, we allocate the area to save SSE
758 regs, then save them, and then allocate the remaining. There is no SEH
759 unwind info for this later allocation. */
760 #define SEH_MAX_FRAME_SIZE ((2U << 30) - 256)
761
762 /* Target OS keeps a vector-aligned (128-bit, 16-byte) stack. This is
763 mandatory for the 64-bit ABI, and may or may not be true for other
764 operating systems. */
765 #define TARGET_KEEPS_VECTOR_ALIGNED_STACK TARGET_64BIT
766
767 /* Minimum allocation boundary for the code of a function. */
768 #define FUNCTION_BOUNDARY 8
769
770 /* C++ stores the virtual bit in the lowest bit of function pointers. */
771 #define TARGET_PTRMEMFUNC_VBIT_LOCATION ptrmemfunc_vbit_in_pfn
772
773 /* Minimum size in bits of the largest boundary to which any
774 and all fundamental data types supported by the hardware
775 might need to be aligned. No data type wants to be aligned
776 rounder than this.
777
778 Pentium+ prefers DFmode values to be aligned to 64 bit boundary
779 and Pentium Pro XFmode values at 128 bit boundaries. */
780
781 #define BIGGEST_ALIGNMENT \
782 (TARGET_AVX512F ? 512 : (TARGET_AVX ? 256 : 128))
783
784 /* Maximum stack alignment. */
785 #define MAX_STACK_ALIGNMENT MAX_OFILE_ALIGNMENT
786
787 /* Alignment value for attribute ((aligned)). It is a constant since
788 it is the part of the ABI. We shouldn't change it with -mavx. */
789 #define ATTRIBUTE_ALIGNED_VALUE 128
790
791 /* Decide whether a variable of mode MODE should be 128 bit aligned. */
792 #define ALIGN_MODE_128(MODE) \
793 ((MODE) == XFmode || SSE_REG_MODE_P (MODE))
794
795 /* The published ABIs say that doubles should be aligned on word
796 boundaries, so lower the alignment for structure fields unless
797 -malign-double is set. */
798
799 /* ??? Blah -- this macro is used directly by libobjc. Since it
800 supports no vector modes, cut out the complexity and fall back
801 on BIGGEST_FIELD_ALIGNMENT. */
802 #ifdef IN_TARGET_LIBS
803 #ifdef __x86_64__
804 #define BIGGEST_FIELD_ALIGNMENT 128
805 #else
806 #define BIGGEST_FIELD_ALIGNMENT 32
807 #endif
808 #else
809 #define ADJUST_FIELD_ALIGN(FIELD, COMPUTED) \
810 x86_field_alignment (FIELD, COMPUTED)
811 #endif
812
813 /* If defined, a C expression to compute the alignment given to a
814 constant that is being placed in memory. EXP is the constant
815 and ALIGN is the alignment that the object would ordinarily have.
816 The value of this macro is used instead of that alignment to align
817 the object.
818
819 If this macro is not defined, then ALIGN is used.
820
821 The typical use of this macro is to increase alignment for string
822 constants to be word aligned so that `strcpy' calls that copy
823 constants can be done inline. */
824
825 #define CONSTANT_ALIGNMENT(EXP, ALIGN) ix86_constant_alignment ((EXP), (ALIGN))
826
827 /* If defined, a C expression to compute the alignment for a static
828 variable. TYPE is the data type, and ALIGN is the alignment that
829 the object would ordinarily have. The value of this macro is used
830 instead of that alignment to align the object.
831
832 If this macro is not defined, then ALIGN is used.
833
834 One use of this macro is to increase alignment of medium-size
835 data to make it all fit in fewer cache lines. Another is to
836 cause character arrays to be word-aligned so that `strcpy' calls
837 that copy constants to character arrays can be done inline. */
838
839 #define DATA_ALIGNMENT(TYPE, ALIGN) \
840 ix86_data_alignment ((TYPE), (ALIGN), true)
841
842 /* Similar to DATA_ALIGNMENT, but for the cases where the ABI mandates
843 some alignment increase, instead of optimization only purposes. E.g.
844 AMD x86-64 psABI says that variables with array type larger than 15 bytes
845 must be aligned to 16 byte boundaries.
846
847 If this macro is not defined, then ALIGN is used. */
848
849 #define DATA_ABI_ALIGNMENT(TYPE, ALIGN) \
850 ix86_data_alignment ((TYPE), (ALIGN), false)
851
852 /* If defined, a C expression to compute the alignment for a local
853 variable. TYPE is the data type, and ALIGN is the alignment that
854 the object would ordinarily have. The value of this macro is used
855 instead of that alignment to align the object.
856
857 If this macro is not defined, then ALIGN is used.
858
859 One use of this macro is to increase alignment of medium-size
860 data to make it all fit in fewer cache lines. */
861
862 #define LOCAL_ALIGNMENT(TYPE, ALIGN) \
863 ix86_local_alignment ((TYPE), VOIDmode, (ALIGN))
864
865 /* If defined, a C expression to compute the alignment for stack slot.
866 TYPE is the data type, MODE is the widest mode available, and ALIGN
867 is the alignment that the slot would ordinarily have. The value of
868 this macro is used instead of that alignment to align the slot.
869
870 If this macro is not defined, then ALIGN is used when TYPE is NULL,
871 Otherwise, LOCAL_ALIGNMENT will be used.
872
873 One use of this macro is to set alignment of stack slot to the
874 maximum alignment of all possible modes which the slot may have. */
875
876 #define STACK_SLOT_ALIGNMENT(TYPE, MODE, ALIGN) \
877 ix86_local_alignment ((TYPE), (MODE), (ALIGN))
878
879 /* If defined, a C expression to compute the alignment for a local
880 variable DECL.
881
882 If this macro is not defined, then
883 LOCAL_ALIGNMENT (TREE_TYPE (DECL), DECL_ALIGN (DECL)) will be used.
884
885 One use of this macro is to increase alignment of medium-size
886 data to make it all fit in fewer cache lines. */
887
888 #define LOCAL_DECL_ALIGNMENT(DECL) \
889 ix86_local_alignment ((DECL), VOIDmode, DECL_ALIGN (DECL))
890
891 /* If defined, a C expression to compute the minimum required alignment
892 for dynamic stack realignment purposes for EXP (a TYPE or DECL),
893 MODE, assuming normal alignment ALIGN.
894
895 If this macro is not defined, then (ALIGN) will be used. */
896
897 #define MINIMUM_ALIGNMENT(EXP, MODE, ALIGN) \
898 ix86_minimum_alignment (EXP, MODE, ALIGN)
899
900
901 /* Set this nonzero if move instructions will actually fail to work
902 when given unaligned data. */
903 #define STRICT_ALIGNMENT 0
904
905 /* If bit field type is int, don't let it cross an int,
906 and give entire struct the alignment of an int. */
907 /* Required on the 386 since it doesn't have bit-field insns. */
908 #define PCC_BITFIELD_TYPE_MATTERS 1
909 \f
910 /* Standard register usage. */
911
912 /* This processor has special stack-like registers. See reg-stack.c
913 for details. */
914
915 #define STACK_REGS
916
917 #define IS_STACK_MODE(MODE) \
918 (((MODE) == SFmode && !(TARGET_SSE && TARGET_SSE_MATH)) \
919 || ((MODE) == DFmode && !(TARGET_SSE2 && TARGET_SSE_MATH)) \
920 || (MODE) == XFmode)
921
922 /* Number of actual hardware registers.
923 The hardware registers are assigned numbers for the compiler
924 from 0 to just below FIRST_PSEUDO_REGISTER.
925 All registers that the compiler knows about must be given numbers,
926 even those that are not normally considered general registers.
927
928 In the 80386 we give the 8 general purpose registers the numbers 0-7.
929 We number the floating point registers 8-15.
930 Note that registers 0-7 can be accessed as a short or int,
931 while only 0-3 may be used with byte `mov' instructions.
932
933 Reg 16 does not correspond to any hardware register, but instead
934 appears in the RTL as an argument pointer prior to reload, and is
935 eliminated during reloading in favor of either the stack or frame
936 pointer. */
937
938 #define FIRST_PSEUDO_REGISTER 77
939
940 /* Number of hardware registers that go into the DWARF-2 unwind info.
941 If not defined, equals FIRST_PSEUDO_REGISTER. */
942
943 #define DWARF_FRAME_REGISTERS 17
944
945 /* 1 for registers that have pervasive standard uses
946 and are not available for the register allocator.
947 On the 80386, the stack pointer is such, as is the arg pointer.
948
949 REX registers are disabled for 32bit targets in
950 TARGET_CONDITIONAL_REGISTER_USAGE. */
951
952 #define FIXED_REGISTERS \
953 /*ax,dx,cx,bx,si,di,bp,sp,st,st1,st2,st3,st4,st5,st6,st7*/ \
954 { 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, \
955 /*arg,flags,fpsr,fpcr,frame*/ \
956 1, 1, 1, 1, 1, \
957 /*xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7*/ \
958 0, 0, 0, 0, 0, 0, 0, 0, \
959 /* mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7*/ \
960 0, 0, 0, 0, 0, 0, 0, 0, \
961 /* r8, r9, r10, r11, r12, r13, r14, r15*/ \
962 0, 0, 0, 0, 0, 0, 0, 0, \
963 /*xmm8,xmm9,xmm10,xmm11,xmm12,xmm13,xmm14,xmm15*/ \
964 0, 0, 0, 0, 0, 0, 0, 0, \
965 /*xmm16,xmm17,xmm18,xmm19,xmm20,xmm21,xmm22,xmm23*/ \
966 0, 0, 0, 0, 0, 0, 0, 0, \
967 /*xmm24,xmm25,xmm26,xmm27,xmm28,xmm29,xmm30,xmm31*/ \
968 0, 0, 0, 0, 0, 0, 0, 0, \
969 /* k0, k1, k2, k3, k4, k5, k6, k7*/ \
970 0, 0, 0, 0, 0, 0, 0, 0 }
971
972 /* 1 for registers not available across function calls.
973 These must include the FIXED_REGISTERS and also any
974 registers that can be used without being saved.
975 The latter must include the registers where values are returned
976 and the register where structure-value addresses are passed.
977 Aside from that, you can include as many other registers as you like.
978
979 Value is set to 1 if the register is call used unconditionally.
980 Bit one is set if the register is call used on TARGET_32BIT ABI.
981 Bit two is set if the register is call used on TARGET_64BIT ABI.
982 Bit three is set if the register is call used on TARGET_64BIT_MS_ABI.
983
984 Proper values are computed in TARGET_CONDITIONAL_REGISTER_USAGE. */
985
986 #define CALL_USED_REGISTERS \
987 /*ax,dx,cx,bx,si,di,bp,sp,st,st1,st2,st3,st4,st5,st6,st7*/ \
988 { 1, 1, 1, 0, 4, 4, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, \
989 /*arg,flags,fpsr,fpcr,frame*/ \
990 1, 1, 1, 1, 1, \
991 /*xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7*/ \
992 1, 1, 1, 1, 1, 1, 6, 6, \
993 /* mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7*/ \
994 1, 1, 1, 1, 1, 1, 1, 1, \
995 /* r8, r9, r10, r11, r12, r13, r14, r15*/ \
996 1, 1, 1, 1, 2, 2, 2, 2, \
997 /*xmm8,xmm9,xmm10,xmm11,xmm12,xmm13,xmm14,xmm15*/ \
998 6, 6, 6, 6, 6, 6, 6, 6, \
999 /*xmm16,xmm17,xmm18,xmm19,xmm20,xmm21,xmm22,xmm23*/ \
1000 6, 6, 6, 6, 6, 6, 6, 6, \
1001 /*xmm24,xmm25,xmm26,xmm27,xmm28,xmm29,xmm30,xmm31*/ \
1002 6, 6, 6, 6, 6, 6, 6, 6, \
1003 /* k0, k1, k2, k3, k4, k5, k6, k7*/ \
1004 1, 1, 1, 1, 1, 1, 1, 1 }
1005
1006 /* Order in which to allocate registers. Each register must be
1007 listed once, even those in FIXED_REGISTERS. List frame pointer
1008 late and fixed registers last. Note that, in general, we prefer
1009 registers listed in CALL_USED_REGISTERS, keeping the others
1010 available for storage of persistent values.
1011
1012 The ADJUST_REG_ALLOC_ORDER actually overwrite the order,
1013 so this is just empty initializer for array. */
1014
1015 #define REG_ALLOC_ORDER \
1016 { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,\
1017 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, \
1018 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, \
1019 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, \
1020 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76 }
1021
1022 /* ADJUST_REG_ALLOC_ORDER is a macro which permits reg_alloc_order
1023 to be rearranged based on a particular function. When using sse math,
1024 we want to allocate SSE before x87 registers and vice versa. */
1025
1026 #define ADJUST_REG_ALLOC_ORDER x86_order_regs_for_local_alloc ()
1027
1028
1029 #define OVERRIDE_ABI_FORMAT(FNDECL) ix86_call_abi_override (FNDECL)
1030
1031 /* Return number of consecutive hard regs needed starting at reg REGNO
1032 to hold something of mode MODE.
1033 This is ordinarily the length in words of a value of mode MODE
1034 but can be less for certain modes in special long registers.
1035
1036 Actually there are no two word move instructions for consecutive
1037 registers. And only registers 0-3 may have mov byte instructions
1038 applied to them. */
1039
1040 #define HARD_REGNO_NREGS(REGNO, MODE) \
1041 (STACK_REGNO_P (REGNO) || SSE_REGNO_P (REGNO) || MMX_REGNO_P (REGNO) \
1042 ? (COMPLEX_MODE_P (MODE) ? 2 : 1) \
1043 : ((MODE) == XFmode \
1044 ? (TARGET_64BIT ? 2 : 3) \
1045 : (MODE) == XCmode \
1046 ? (TARGET_64BIT ? 4 : 6) \
1047 : ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) / UNITS_PER_WORD)))
1048
1049 #define HARD_REGNO_NREGS_HAS_PADDING(REGNO, MODE) \
1050 ((TARGET_128BIT_LONG_DOUBLE && !TARGET_64BIT) \
1051 ? (STACK_REGNO_P (REGNO) || SSE_REGNO_P (REGNO) || MMX_REGNO_P (REGNO) \
1052 ? 0 \
1053 : ((MODE) == XFmode || (MODE) == XCmode)) \
1054 : 0)
1055
1056 #define HARD_REGNO_NREGS_WITH_PADDING(REGNO, MODE) ((MODE) == XFmode ? 4 : 8)
1057
1058 #define VALID_AVX256_REG_MODE(MODE) \
1059 ((MODE) == V32QImode || (MODE) == V16HImode || (MODE) == V8SImode \
1060 || (MODE) == V4DImode || (MODE) == V2TImode || (MODE) == V8SFmode \
1061 || (MODE) == V4DFmode)
1062
1063 #define VALID_AVX256_REG_OR_OI_MODE(MODE) \
1064 (VALID_AVX256_REG_MODE (MODE) || (MODE) == OImode)
1065
1066 #define VALID_AVX512F_SCALAR_MODE(MODE) \
1067 ((MODE) == DImode || (MODE) == DFmode || (MODE) == SImode \
1068 || (MODE) == SFmode)
1069
1070 #define VALID_AVX512F_REG_MODE(MODE) \
1071 ((MODE) == V8DImode || (MODE) == V8DFmode || (MODE) == V64QImode \
1072 || (MODE) == V16SImode || (MODE) == V16SFmode || (MODE) == V32HImode)
1073
1074 #define VALID_SSE2_REG_MODE(MODE) \
1075 ((MODE) == V16QImode || (MODE) == V8HImode || (MODE) == V2DFmode \
1076 || (MODE) == V2DImode || (MODE) == DFmode)
1077
1078 #define VALID_SSE_REG_MODE(MODE) \
1079 ((MODE) == V1TImode || (MODE) == TImode \
1080 || (MODE) == V4SFmode || (MODE) == V4SImode \
1081 || (MODE) == SFmode || (MODE) == TFmode)
1082
1083 #define VALID_MMX_REG_MODE_3DNOW(MODE) \
1084 ((MODE) == V2SFmode || (MODE) == SFmode)
1085
1086 #define VALID_MMX_REG_MODE(MODE) \
1087 ((MODE == V1DImode) || (MODE) == DImode \
1088 || (MODE) == V2SImode || (MODE) == SImode \
1089 || (MODE) == V4HImode || (MODE) == V8QImode)
1090
1091 #define VALID_DFP_MODE_P(MODE) \
1092 ((MODE) == SDmode || (MODE) == DDmode || (MODE) == TDmode)
1093
1094 #define VALID_FP_MODE_P(MODE) \
1095 ((MODE) == SFmode || (MODE) == DFmode || (MODE) == XFmode \
1096 || (MODE) == SCmode || (MODE) == DCmode || (MODE) == XCmode) \
1097
1098 #define VALID_INT_MODE_P(MODE) \
1099 ((MODE) == QImode || (MODE) == HImode || (MODE) == SImode \
1100 || (MODE) == DImode \
1101 || (MODE) == CQImode || (MODE) == CHImode || (MODE) == CSImode \
1102 || (MODE) == CDImode \
1103 || (TARGET_64BIT && ((MODE) == TImode || (MODE) == CTImode \
1104 || (MODE) == TFmode || (MODE) == TCmode)))
1105
1106 /* Return true for modes passed in SSE registers. */
1107 #define SSE_REG_MODE_P(MODE) \
1108 ((MODE) == V1TImode || (MODE) == TImode || (MODE) == V16QImode \
1109 || (MODE) == TFmode || (MODE) == V8HImode || (MODE) == V2DFmode \
1110 || (MODE) == V2DImode || (MODE) == V4SFmode || (MODE) == V4SImode \
1111 || (MODE) == V32QImode || (MODE) == V16HImode || (MODE) == V8SImode \
1112 || (MODE) == V4DImode || (MODE) == V8SFmode || (MODE) == V4DFmode \
1113 || (MODE) == V2TImode || (MODE) == V8DImode || (MODE) == V64QImode \
1114 || (MODE) == V16SImode || (MODE) == V32HImode || (MODE) == V8DFmode \
1115 || (MODE) == V16SFmode)
1116
1117 #define VALID_MASK_REG_MODE(MODE) ((MODE) == HImode || (MODE) == QImode)
1118
1119 /* Value is 1 if hard register REGNO can hold a value of machine-mode MODE. */
1120
1121 #define HARD_REGNO_MODE_OK(REGNO, MODE) \
1122 ix86_hard_regno_mode_ok ((REGNO), (MODE))
1123
1124 /* Value is 1 if it is a good idea to tie two pseudo registers
1125 when one has mode MODE1 and one has mode MODE2.
1126 If HARD_REGNO_MODE_OK could produce different values for MODE1 and MODE2,
1127 for any hard reg, then this must be 0 for correct output. */
1128
1129 #define MODES_TIEABLE_P(MODE1, MODE2) ix86_modes_tieable_p (MODE1, MODE2)
1130
1131 /* It is possible to write patterns to move flags; but until someone
1132 does it, */
1133 #define AVOID_CCMODE_COPIES
1134
1135 /* Specify the modes required to caller save a given hard regno.
1136 We do this on i386 to prevent flags from being saved at all.
1137
1138 Kill any attempts to combine saving of modes. */
1139
1140 #define HARD_REGNO_CALLER_SAVE_MODE(REGNO, NREGS, MODE) \
1141 (CC_REGNO_P (REGNO) ? VOIDmode \
1142 : (MODE) == VOIDmode && (NREGS) != 1 ? VOIDmode \
1143 : (MODE) == VOIDmode ? choose_hard_reg_mode ((REGNO), (NREGS), false) \
1144 : (MODE) == HImode && !(TARGET_PARTIAL_REG_STALL \
1145 || MASK_REGNO_P (REGNO)) ? SImode \
1146 : (MODE) == QImode && !(TARGET_64BIT || QI_REGNO_P (REGNO) \
1147 || MASK_REGNO_P (REGNO)) ? SImode \
1148 : (MODE))
1149
1150 /* The only ABI that saves SSE registers across calls is Win64 (thus no
1151 need to check the current ABI here), and with AVX enabled Win64 only
1152 guarantees that the low 16 bytes are saved. */
1153 #define HARD_REGNO_CALL_PART_CLOBBERED(REGNO, MODE) \
1154 (SSE_REGNO_P (REGNO) && GET_MODE_SIZE (MODE) > 16)
1155
1156 /* Specify the registers used for certain standard purposes.
1157 The values of these macros are register numbers. */
1158
1159 /* on the 386 the pc register is %eip, and is not usable as a general
1160 register. The ordinary mov instructions won't work */
1161 /* #define PC_REGNUM */
1162
1163 /* Register to use for pushing function arguments. */
1164 #define STACK_POINTER_REGNUM 7
1165
1166 /* Base register for access to local variables of the function. */
1167 #define HARD_FRAME_POINTER_REGNUM 6
1168
1169 /* Base register for access to local variables of the function. */
1170 #define FRAME_POINTER_REGNUM 20
1171
1172 /* First floating point reg */
1173 #define FIRST_FLOAT_REG 8
1174
1175 /* First & last stack-like regs */
1176 #define FIRST_STACK_REG FIRST_FLOAT_REG
1177 #define LAST_STACK_REG (FIRST_FLOAT_REG + 7)
1178
1179 #define FIRST_SSE_REG (FRAME_POINTER_REGNUM + 1)
1180 #define LAST_SSE_REG (FIRST_SSE_REG + 7)
1181
1182 #define FIRST_MMX_REG (LAST_SSE_REG + 1) /*29*/
1183 #define LAST_MMX_REG (FIRST_MMX_REG + 7)
1184
1185 #define FIRST_REX_INT_REG (LAST_MMX_REG + 1) /*37*/
1186 #define LAST_REX_INT_REG (FIRST_REX_INT_REG + 7)
1187
1188 #define FIRST_REX_SSE_REG (LAST_REX_INT_REG + 1) /*45*/
1189 #define LAST_REX_SSE_REG (FIRST_REX_SSE_REG + 7)
1190
1191 #define FIRST_EXT_REX_SSE_REG (LAST_REX_SSE_REG + 1) /*53*/
1192 #define LAST_EXT_REX_SSE_REG (FIRST_EXT_REX_SSE_REG + 15) /*68*/
1193
1194 #define FIRST_MASK_REG (LAST_EXT_REX_SSE_REG + 1) /*69*/
1195 #define LAST_MASK_REG (FIRST_MASK_REG + 7) /*76*/
1196
1197 /* Override this in other tm.h files to cope with various OS lossage
1198 requiring a frame pointer. */
1199 #ifndef SUBTARGET_FRAME_POINTER_REQUIRED
1200 #define SUBTARGET_FRAME_POINTER_REQUIRED 0
1201 #endif
1202
1203 /* Make sure we can access arbitrary call frames. */
1204 #define SETUP_FRAME_ADDRESSES() ix86_setup_frame_addresses ()
1205
1206 /* Base register for access to arguments of the function. */
1207 #define ARG_POINTER_REGNUM 16
1208
1209 /* Register to hold the addressing base for position independent
1210 code access to data items. We don't use PIC pointer for 64bit
1211 mode. Define the regnum to dummy value to prevent gcc from
1212 pessimizing code dealing with EBX.
1213
1214 To avoid clobbering a call-saved register unnecessarily, we renumber
1215 the pic register when possible. The change is visible after the
1216 prologue has been emitted. */
1217
1218 #define REAL_PIC_OFFSET_TABLE_REGNUM BX_REG
1219
1220 #define PIC_OFFSET_TABLE_REGNUM \
1221 ((TARGET_64BIT && (ix86_cmodel == CM_SMALL_PIC \
1222 || TARGET_PECOFF)) \
1223 || !flag_pic ? INVALID_REGNUM \
1224 : reload_completed ? REGNO (pic_offset_table_rtx) \
1225 : REAL_PIC_OFFSET_TABLE_REGNUM)
1226
1227 #define GOT_SYMBOL_NAME "_GLOBAL_OFFSET_TABLE_"
1228
1229 /* This is overridden by <cygwin.h>. */
1230 #define MS_AGGREGATE_RETURN 0
1231
1232 #define KEEP_AGGREGATE_RETURN_POINTER 0
1233 \f
1234 /* Define the classes of registers for register constraints in the
1235 machine description. Also define ranges of constants.
1236
1237 One of the classes must always be named ALL_REGS and include all hard regs.
1238 If there is more than one class, another class must be named NO_REGS
1239 and contain no registers.
1240
1241 The name GENERAL_REGS must be the name of a class (or an alias for
1242 another name such as ALL_REGS). This is the class of registers
1243 that is allowed by "g" or "r" in a register constraint.
1244 Also, registers outside this class are allocated only when
1245 instructions express preferences for them.
1246
1247 The classes must be numbered in nondecreasing order; that is,
1248 a larger-numbered class must never be contained completely
1249 in a smaller-numbered class.
1250
1251 For any two classes, it is very desirable that there be another
1252 class that represents their union.
1253
1254 It might seem that class BREG is unnecessary, since no useful 386
1255 opcode needs reg %ebx. But some systems pass args to the OS in ebx,
1256 and the "b" register constraint is useful in asms for syscalls.
1257
1258 The flags, fpsr and fpcr registers are in no class. */
1259
1260 enum reg_class
1261 {
1262 NO_REGS,
1263 AREG, DREG, CREG, BREG, SIREG, DIREG,
1264 AD_REGS, /* %eax/%edx for DImode */
1265 Q_REGS, /* %eax %ebx %ecx %edx */
1266 NON_Q_REGS, /* %esi %edi %ebp %esp */
1267 INDEX_REGS, /* %eax %ebx %ecx %edx %esi %edi %ebp */
1268 LEGACY_REGS, /* %eax %ebx %ecx %edx %esi %edi %ebp %esp */
1269 CLOBBERED_REGS, /* call-clobbered integer registers */
1270 GENERAL_REGS, /* %eax %ebx %ecx %edx %esi %edi %ebp %esp
1271 %r8 %r9 %r10 %r11 %r12 %r13 %r14 %r15 */
1272 FP_TOP_REG, FP_SECOND_REG, /* %st(0) %st(1) */
1273 FLOAT_REGS,
1274 SSE_FIRST_REG,
1275 SSE_REGS,
1276 EVEX_SSE_REGS,
1277 ALL_SSE_REGS,
1278 MMX_REGS,
1279 FP_TOP_SSE_REGS,
1280 FP_SECOND_SSE_REGS,
1281 FLOAT_SSE_REGS,
1282 FLOAT_INT_REGS,
1283 INT_SSE_REGS,
1284 FLOAT_INT_SSE_REGS,
1285 MASK_EVEX_REGS,
1286 MASK_REGS,
1287 ALL_REGS, LIM_REG_CLASSES
1288 };
1289
1290 #define N_REG_CLASSES ((int) LIM_REG_CLASSES)
1291
1292 #define INTEGER_CLASS_P(CLASS) \
1293 reg_class_subset_p ((CLASS), GENERAL_REGS)
1294 #define FLOAT_CLASS_P(CLASS) \
1295 reg_class_subset_p ((CLASS), FLOAT_REGS)
1296 #define SSE_CLASS_P(CLASS) \
1297 reg_class_subset_p ((CLASS), ALL_SSE_REGS)
1298 #define MMX_CLASS_P(CLASS) \
1299 ((CLASS) == MMX_REGS)
1300 #define MAYBE_INTEGER_CLASS_P(CLASS) \
1301 reg_classes_intersect_p ((CLASS), GENERAL_REGS)
1302 #define MAYBE_FLOAT_CLASS_P(CLASS) \
1303 reg_classes_intersect_p ((CLASS), FLOAT_REGS)
1304 #define MAYBE_SSE_CLASS_P(CLASS) \
1305 reg_classes_intersect_p ((CLASS), ALL_SSE_REGS)
1306 #define MAYBE_MMX_CLASS_P(CLASS) \
1307 reg_classes_intersect_p ((CLASS), MMX_REGS)
1308 #define MAYBE_MASK_CLASS_P(CLASS) \
1309 reg_classes_intersect_p ((CLASS), MASK_REGS)
1310
1311 #define Q_CLASS_P(CLASS) \
1312 reg_class_subset_p ((CLASS), Q_REGS)
1313
1314 #define MAYBE_NON_Q_CLASS_P(CLASS) \
1315 reg_classes_intersect_p ((CLASS), NON_Q_REGS)
1316
1317 /* Give names of register classes as strings for dump file. */
1318
1319 #define REG_CLASS_NAMES \
1320 { "NO_REGS", \
1321 "AREG", "DREG", "CREG", "BREG", \
1322 "SIREG", "DIREG", \
1323 "AD_REGS", \
1324 "Q_REGS", "NON_Q_REGS", \
1325 "INDEX_REGS", \
1326 "LEGACY_REGS", \
1327 "CLOBBERED_REGS", \
1328 "GENERAL_REGS", \
1329 "FP_TOP_REG", "FP_SECOND_REG", \
1330 "FLOAT_REGS", \
1331 "SSE_FIRST_REG", \
1332 "SSE_REGS", \
1333 "EVEX_SSE_REGS", \
1334 "ALL_SSE_REGS", \
1335 "MMX_REGS", \
1336 "FP_TOP_SSE_REGS", \
1337 "FP_SECOND_SSE_REGS", \
1338 "FLOAT_SSE_REGS", \
1339 "FLOAT_INT_REGS", \
1340 "INT_SSE_REGS", \
1341 "FLOAT_INT_SSE_REGS", \
1342 "MASK_EVEX_REGS", \
1343 "MASK_REGS", \
1344 "ALL_REGS" }
1345
1346 /* Define which registers fit in which classes. This is an initializer
1347 for a vector of HARD_REG_SET of length N_REG_CLASSES.
1348
1349 Note that CLOBBERED_REGS are calculated by
1350 TARGET_CONDITIONAL_REGISTER_USAGE. */
1351
1352 #define REG_CLASS_CONTENTS \
1353 { { 0x00, 0x0, 0x0 }, \
1354 { 0x01, 0x0, 0x0 }, /* AREG */ \
1355 { 0x02, 0x0, 0x0 }, /* DREG */ \
1356 { 0x04, 0x0, 0x0 }, /* CREG */ \
1357 { 0x08, 0x0, 0x0 }, /* BREG */ \
1358 { 0x10, 0x0, 0x0 }, /* SIREG */ \
1359 { 0x20, 0x0, 0x0 }, /* DIREG */ \
1360 { 0x03, 0x0, 0x0 }, /* AD_REGS */ \
1361 { 0x0f, 0x0, 0x0 }, /* Q_REGS */ \
1362 { 0x1100f0, 0x1fe0, 0x0 }, /* NON_Q_REGS */ \
1363 { 0x7f, 0x1fe0, 0x0 }, /* INDEX_REGS */ \
1364 { 0x1100ff, 0x0, 0x0 }, /* LEGACY_REGS */ \
1365 { 0x07, 0x0, 0x0 }, /* CLOBBERED_REGS */ \
1366 { 0x1100ff, 0x1fe0, 0x0 }, /* GENERAL_REGS */ \
1367 { 0x100, 0x0, 0x0 }, /* FP_TOP_REG */ \
1368 { 0x0200, 0x0, 0x0 }, /* FP_SECOND_REG */ \
1369 { 0xff00, 0x0, 0x0 }, /* FLOAT_REGS */ \
1370 { 0x200000, 0x0, 0x0 }, /* SSE_FIRST_REG */ \
1371 { 0x1fe00000, 0x1fe000, 0x0 }, /* SSE_REGS */ \
1372 { 0x0,0xffe00000, 0x1f }, /* EVEX_SSE_REGS */ \
1373 { 0x1fe00000,0xffffe000, 0x1f }, /* ALL_SSE_REGS */ \
1374 { 0xe0000000, 0x1f, 0x0 }, /* MMX_REGS */ \
1375 { 0x1fe00100,0xffffe000, 0x1f }, /* FP_TOP_SSE_REG */ \
1376 { 0x1fe00200,0xffffe000, 0x1f }, /* FP_SECOND_SSE_REG */ \
1377 { 0x1fe0ff00,0xffffe000, 0x1f }, /* FLOAT_SSE_REGS */ \
1378 { 0x11ffff, 0x1fe0, 0x0 }, /* FLOAT_INT_REGS */ \
1379 { 0x1ff100ff,0xffffffe0, 0x1f }, /* INT_SSE_REGS */ \
1380 { 0x1ff1ffff,0xffffffe0, 0x1f }, /* FLOAT_INT_SSE_REGS */ \
1381 { 0x0, 0x0,0x1fc0 }, /* MASK_EVEX_REGS */ \
1382 { 0x0, 0x0,0x1fe0 }, /* MASK_REGS */ \
1383 { 0xffffffff,0xffffffff,0x1fff } \
1384 }
1385
1386 /* The same information, inverted:
1387 Return the class number of the smallest class containing
1388 reg number REGNO. This could be a conditional expression
1389 or could index an array. */
1390
1391 #define REGNO_REG_CLASS(REGNO) (regclass_map[REGNO])
1392
1393 /* When this hook returns true for MODE, the compiler allows
1394 registers explicitly used in the rtl to be used as spill registers
1395 but prevents the compiler from extending the lifetime of these
1396 registers. */
1397 #define TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P hook_bool_mode_true
1398
1399 #define QI_REG_P(X) (REG_P (X) && QI_REGNO_P (REGNO (X)))
1400 #define QI_REGNO_P(N) IN_RANGE ((N), AX_REG, BX_REG)
1401
1402 #define GENERAL_REG_P(X) \
1403 (REG_P (X) && GENERAL_REGNO_P (REGNO (X)))
1404 #define GENERAL_REGNO_P(N) \
1405 (IN_RANGE ((N), AX_REG, SP_REG) || REX_INT_REGNO_P (N))
1406
1407 #define ANY_QI_REG_P(X) (REG_P (X) && ANY_QI_REGNO_P (REGNO (X)))
1408 #define ANY_QI_REGNO_P(N) \
1409 (TARGET_64BIT ? GENERAL_REGNO_P (N) : QI_REGNO_P (N))
1410
1411 #define REX_INT_REG_P(X) (REG_P (X) && REX_INT_REGNO_P (REGNO (X)))
1412 #define REX_INT_REGNO_P(N) \
1413 IN_RANGE ((N), FIRST_REX_INT_REG, LAST_REX_INT_REG)
1414
1415 #define STACK_REG_P(X) (REG_P (X) && STACK_REGNO_P (REGNO (X)))
1416 #define STACK_REGNO_P(N) IN_RANGE ((N), FIRST_STACK_REG, LAST_STACK_REG)
1417
1418 #define ANY_FP_REG_P(X) (REG_P (X) && ANY_FP_REGNO_P (REGNO (X)))
1419 #define ANY_FP_REGNO_P(N) (STACK_REGNO_P (N) || SSE_REGNO_P (N))
1420
1421 #define X87_FLOAT_MODE_P(MODE) \
1422 (TARGET_80387 && ((MODE) == SFmode || (MODE) == DFmode || (MODE) == XFmode))
1423
1424 #define SSE_REG_P(X) (REG_P (X) && SSE_REGNO_P (REGNO (X)))
1425 #define SSE_REGNO_P(N) \
1426 (IN_RANGE ((N), FIRST_SSE_REG, LAST_SSE_REG) \
1427 || REX_SSE_REGNO_P (N) \
1428 || EXT_REX_SSE_REGNO_P (N))
1429
1430 #define REX_SSE_REGNO_P(N) \
1431 IN_RANGE ((N), FIRST_REX_SSE_REG, LAST_REX_SSE_REG)
1432
1433 #define EXT_REX_SSE_REGNO_P(N) \
1434 IN_RANGE ((N), FIRST_EXT_REX_SSE_REG, LAST_EXT_REX_SSE_REG)
1435
1436 #define SSE_REGNO(N) \
1437 ((N) < 8 ? FIRST_SSE_REG + (N) \
1438 : (N) <= LAST_REX_SSE_REG ? (FIRST_REX_SSE_REG + (N) - 8) \
1439 : (FIRST_EXT_REX_SSE_REG + (N) - 16))
1440
1441 #define MASK_REGNO_P(N) IN_RANGE ((N), FIRST_MASK_REG, LAST_MASK_REG)
1442 #define ANY_MASK_REG_P(X) (REG_P (X) && MASK_REGNO_P (REGNO (X)))
1443
1444 #define SSE_FLOAT_MODE_P(MODE) \
1445 ((TARGET_SSE && (MODE) == SFmode) || (TARGET_SSE2 && (MODE) == DFmode))
1446
1447 #define FMA4_VEC_FLOAT_MODE_P(MODE) \
1448 (TARGET_FMA4 && ((MODE) == V4SFmode || (MODE) == V2DFmode \
1449 || (MODE) == V8SFmode || (MODE) == V4DFmode))
1450
1451 #define MMX_REG_P(X) (REG_P (X) && MMX_REGNO_P (REGNO (X)))
1452 #define MMX_REGNO_P(N) IN_RANGE ((N), FIRST_MMX_REG, LAST_MMX_REG)
1453
1454 #define STACK_TOP_P(X) (REG_P (X) && REGNO (X) == FIRST_STACK_REG)
1455
1456 #define CC_REG_P(X) (REG_P (X) && CC_REGNO_P (REGNO (X)))
1457 #define CC_REGNO_P(X) ((X) == FLAGS_REG || (X) == FPSR_REG)
1458
1459 /* The class value for index registers, and the one for base regs. */
1460
1461 #define INDEX_REG_CLASS INDEX_REGS
1462 #define BASE_REG_CLASS GENERAL_REGS
1463
1464 /* Place additional restrictions on the register class to use when it
1465 is necessary to be able to hold a value of mode MODE in a reload
1466 register for which class CLASS would ordinarily be used.
1467
1468 We avoid classes containing registers from multiple units due to
1469 the limitation in ix86_secondary_memory_needed. We limit these
1470 classes to their "natural mode" single unit register class, depending
1471 on the unit availability.
1472
1473 Please note that reg_class_subset_p is not commutative, so these
1474 conditions mean "... if (CLASS) includes ALL registers from the
1475 register set." */
1476
1477 #define LIMIT_RELOAD_CLASS(MODE, CLASS) \
1478 (((MODE) == QImode && !TARGET_64BIT \
1479 && reg_class_subset_p (Q_REGS, (CLASS))) ? Q_REGS \
1480 : (((MODE) == SImode || (MODE) == DImode) \
1481 && reg_class_subset_p (GENERAL_REGS, (CLASS))) ? GENERAL_REGS \
1482 : (SSE_FLOAT_MODE_P (MODE) && TARGET_SSE_MATH \
1483 && reg_class_subset_p (SSE_REGS, (CLASS))) ? SSE_REGS \
1484 : (X87_FLOAT_MODE_P (MODE) \
1485 && reg_class_subset_p (FLOAT_REGS, (CLASS))) ? FLOAT_REGS \
1486 : (CLASS))
1487
1488 /* If we are copying between general and FP registers, we need a memory
1489 location. The same is true for SSE and MMX registers. */
1490 #define SECONDARY_MEMORY_NEEDED(CLASS1, CLASS2, MODE) \
1491 ix86_secondary_memory_needed ((CLASS1), (CLASS2), (MODE), 1)
1492
1493 /* Get_secondary_mem widens integral modes to BITS_PER_WORD.
1494 There is no need to emit full 64 bit move on 64 bit targets
1495 for integral modes that can be moved using 32 bit move. */
1496 #define SECONDARY_MEMORY_NEEDED_MODE(MODE) \
1497 (GET_MODE_BITSIZE (MODE) < 32 && INTEGRAL_MODE_P (MODE) \
1498 ? mode_for_size (32, GET_MODE_CLASS (MODE), 0) \
1499 : MODE)
1500
1501 /* Return a class of registers that cannot change FROM mode to TO mode. */
1502
1503 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
1504 ix86_cannot_change_mode_class (FROM, TO, CLASS)
1505 \f
1506 /* Stack layout; function entry, exit and calling. */
1507
1508 /* Define this if pushing a word on the stack
1509 makes the stack pointer a smaller address. */
1510 #define STACK_GROWS_DOWNWARD
1511
1512 /* Define this to nonzero if the nominal address of the stack frame
1513 is at the high-address end of the local variables;
1514 that is, each additional local variable allocated
1515 goes at a more negative offset in the frame. */
1516 #define FRAME_GROWS_DOWNWARD 1
1517
1518 /* Offset within stack frame to start allocating local variables at.
1519 If FRAME_GROWS_DOWNWARD, this is the offset to the END of the
1520 first local allocated. Otherwise, it is the offset to the BEGINNING
1521 of the first local allocated. */
1522 #define STARTING_FRAME_OFFSET 0
1523
1524 /* If we generate an insn to push BYTES bytes, this says how many the stack
1525 pointer really advances by. On 386, we have pushw instruction that
1526 decrements by exactly 2 no matter what the position was, there is no pushb.
1527
1528 But as CIE data alignment factor on this arch is -4 for 32bit targets
1529 and -8 for 64bit targets, we need to make sure all stack pointer adjustments
1530 are in multiple of 4 for 32bit targets and 8 for 64bit targets. */
1531
1532 #define PUSH_ROUNDING(BYTES) \
1533 (((BYTES) + UNITS_PER_WORD - 1) & -UNITS_PER_WORD)
1534
1535 /* If defined, the maximum amount of space required for outgoing arguments
1536 will be computed and placed into the variable `crtl->outgoing_args_size'.
1537 No space will be pushed onto the stack for each call; instead, the
1538 function prologue should increase the stack frame size by this amount.
1539
1540 In 32bit mode enabling argument accumulation results in about 5% code size
1541 growth becuase move instructions are less compact than push. In 64bit
1542 mode the difference is less drastic but visible.
1543
1544 FIXME: Unlike earlier implementations, the size of unwind info seems to
1545 actually grow with accumulation. Is that because accumulated args
1546 unwind info became unnecesarily bloated?
1547
1548 With the 64-bit MS ABI, we can generate correct code with or without
1549 accumulated args, but because of OUTGOING_REG_PARM_STACK_SPACE the code
1550 generated without accumulated args is terrible.
1551
1552 If stack probes are required, the space used for large function
1553 arguments on the stack must also be probed, so enable
1554 -maccumulate-outgoing-args so this happens in the prologue. */
1555
1556 #define ACCUMULATE_OUTGOING_ARGS \
1557 ((TARGET_ACCUMULATE_OUTGOING_ARGS && optimize_function_for_speed_p (cfun)) \
1558 || TARGET_STACK_PROBE || TARGET_64BIT_MS_ABI)
1559
1560 /* If defined, a C expression whose value is nonzero when we want to use PUSH
1561 instructions to pass outgoing arguments. */
1562
1563 #define PUSH_ARGS (TARGET_PUSH_ARGS && !ACCUMULATE_OUTGOING_ARGS)
1564
1565 /* We want the stack and args grow in opposite directions, even if
1566 PUSH_ARGS is 0. */
1567 #define PUSH_ARGS_REVERSED 1
1568
1569 /* Offset of first parameter from the argument pointer register value. */
1570 #define FIRST_PARM_OFFSET(FNDECL) 0
1571
1572 /* Define this macro if functions should assume that stack space has been
1573 allocated for arguments even when their values are passed in registers.
1574
1575 The value of this macro is the size, in bytes, of the area reserved for
1576 arguments passed in registers for the function represented by FNDECL.
1577
1578 This space can be allocated by the caller, or be a part of the
1579 machine-dependent stack frame: `OUTGOING_REG_PARM_STACK_SPACE' says
1580 which. */
1581 #define REG_PARM_STACK_SPACE(FNDECL) ix86_reg_parm_stack_space (FNDECL)
1582
1583 #define OUTGOING_REG_PARM_STACK_SPACE(FNTYPE) \
1584 (TARGET_64BIT && ix86_function_type_abi (FNTYPE) == MS_ABI)
1585
1586 /* Define how to find the value returned by a library function
1587 assuming the value has mode MODE. */
1588
1589 #define LIBCALL_VALUE(MODE) ix86_libcall_value (MODE)
1590
1591 /* Define the size of the result block used for communication between
1592 untyped_call and untyped_return. The block contains a DImode value
1593 followed by the block used by fnsave and frstor. */
1594
1595 #define APPLY_RESULT_SIZE (8+108)
1596
1597 /* 1 if N is a possible register number for function argument passing. */
1598 #define FUNCTION_ARG_REGNO_P(N) ix86_function_arg_regno_p (N)
1599
1600 /* Define a data type for recording info about an argument list
1601 during the scan of that argument list. This data type should
1602 hold all necessary information about the function itself
1603 and about the args processed so far, enough to enable macros
1604 such as FUNCTION_ARG to determine where the next arg should go. */
1605
1606 typedef struct ix86_args {
1607 int words; /* # words passed so far */
1608 int nregs; /* # registers available for passing */
1609 int regno; /* next available register number */
1610 int fastcall; /* fastcall or thiscall calling convention
1611 is used */
1612 int sse_words; /* # sse words passed so far */
1613 int sse_nregs; /* # sse registers available for passing */
1614 int warn_avx512f; /* True when we want to warn
1615 about AVX512F ABI. */
1616 int warn_avx; /* True when we want to warn about AVX ABI. */
1617 int warn_sse; /* True when we want to warn about SSE ABI. */
1618 int warn_mmx; /* True when we want to warn about MMX ABI. */
1619 int sse_regno; /* next available sse register number */
1620 int mmx_words; /* # mmx words passed so far */
1621 int mmx_nregs; /* # mmx registers available for passing */
1622 int mmx_regno; /* next available mmx register number */
1623 int maybe_vaarg; /* true for calls to possibly vardic fncts. */
1624 int caller; /* true if it is caller. */
1625 int float_in_sse; /* Set to 1 or 2 for 32bit targets if
1626 SFmode/DFmode arguments should be passed
1627 in SSE registers. Otherwise 0. */
1628 enum calling_abi call_abi; /* Set to SYSV_ABI for sysv abi. Otherwise
1629 MS_ABI for ms abi. */
1630 } CUMULATIVE_ARGS;
1631
1632 /* Initialize a variable CUM of type CUMULATIVE_ARGS
1633 for a call to a function whose data type is FNTYPE.
1634 For a library call, FNTYPE is 0. */
1635
1636 #define INIT_CUMULATIVE_ARGS(CUM, FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS) \
1637 init_cumulative_args (&(CUM), (FNTYPE), (LIBNAME), (FNDECL), \
1638 (N_NAMED_ARGS) != -1)
1639
1640 /* Output assembler code to FILE to increment profiler label # LABELNO
1641 for profiling a function entry. */
1642
1643 #define FUNCTION_PROFILER(FILE, LABELNO) x86_function_profiler (FILE, LABELNO)
1644
1645 #define MCOUNT_NAME "_mcount"
1646
1647 #define MCOUNT_NAME_BEFORE_PROLOGUE "__fentry__"
1648
1649 #define PROFILE_COUNT_REGISTER "edx"
1650
1651 /* EXIT_IGNORE_STACK should be nonzero if, when returning from a function,
1652 the stack pointer does not matter. The value is tested only in
1653 functions that have frame pointers.
1654 No definition is equivalent to always zero. */
1655 /* Note on the 386 it might be more efficient not to define this since
1656 we have to restore it ourselves from the frame pointer, in order to
1657 use pop */
1658
1659 #define EXIT_IGNORE_STACK 1
1660
1661 /* Output assembler code for a block containing the constant parts
1662 of a trampoline, leaving space for the variable parts. */
1663
1664 /* On the 386, the trampoline contains two instructions:
1665 mov #STATIC,ecx
1666 jmp FUNCTION
1667 The trampoline is generated entirely at runtime. The operand of JMP
1668 is the address of FUNCTION relative to the instruction following the
1669 JMP (which is 5 bytes long). */
1670
1671 /* Length in units of the trampoline for entering a nested function. */
1672
1673 #define TRAMPOLINE_SIZE (TARGET_64BIT ? 24 : 10)
1674 \f
1675 /* Definitions for register eliminations.
1676
1677 This is an array of structures. Each structure initializes one pair
1678 of eliminable registers. The "from" register number is given first,
1679 followed by "to". Eliminations of the same "from" register are listed
1680 in order of preference.
1681
1682 There are two registers that can always be eliminated on the i386.
1683 The frame pointer and the arg pointer can be replaced by either the
1684 hard frame pointer or to the stack pointer, depending upon the
1685 circumstances. The hard frame pointer is not used before reload and
1686 so it is not eligible for elimination. */
1687
1688 #define ELIMINABLE_REGS \
1689 {{ ARG_POINTER_REGNUM, STACK_POINTER_REGNUM}, \
1690 { ARG_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM}, \
1691 { FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM}, \
1692 { FRAME_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM}} \
1693
1694 /* Define the offset between two registers, one to be eliminated, and the other
1695 its replacement, at the start of a routine. */
1696
1697 #define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \
1698 ((OFFSET) = ix86_initial_elimination_offset ((FROM), (TO)))
1699 \f
1700 /* Addressing modes, and classification of registers for them. */
1701
1702 /* Macros to check register numbers against specific register classes. */
1703
1704 /* These assume that REGNO is a hard or pseudo reg number.
1705 They give nonzero only if REGNO is a hard reg of the suitable class
1706 or a pseudo reg currently allocated to a suitable hard reg.
1707 Since they use reg_renumber, they are safe only once reg_renumber
1708 has been allocated, which happens in reginfo.c during register
1709 allocation. */
1710
1711 #define REGNO_OK_FOR_INDEX_P(REGNO) \
1712 ((REGNO) < STACK_POINTER_REGNUM \
1713 || REX_INT_REGNO_P (REGNO) \
1714 || (unsigned) reg_renumber[(REGNO)] < STACK_POINTER_REGNUM \
1715 || REX_INT_REGNO_P ((unsigned) reg_renumber[(REGNO)]))
1716
1717 #define REGNO_OK_FOR_BASE_P(REGNO) \
1718 (GENERAL_REGNO_P (REGNO) \
1719 || (REGNO) == ARG_POINTER_REGNUM \
1720 || (REGNO) == FRAME_POINTER_REGNUM \
1721 || GENERAL_REGNO_P ((unsigned) reg_renumber[(REGNO)]))
1722
1723 /* The macros REG_OK_FOR..._P assume that the arg is a REG rtx
1724 and check its validity for a certain class.
1725 We have two alternate definitions for each of them.
1726 The usual definition accepts all pseudo regs; the other rejects
1727 them unless they have been allocated suitable hard regs.
1728 The symbol REG_OK_STRICT causes the latter definition to be used.
1729
1730 Most source files want to accept pseudo regs in the hope that
1731 they will get allocated to the class that the insn wants them to be in.
1732 Source files for reload pass need to be strict.
1733 After reload, it makes no difference, since pseudo regs have
1734 been eliminated by then. */
1735
1736
1737 /* Non strict versions, pseudos are ok. */
1738 #define REG_OK_FOR_INDEX_NONSTRICT_P(X) \
1739 (REGNO (X) < STACK_POINTER_REGNUM \
1740 || REX_INT_REGNO_P (REGNO (X)) \
1741 || REGNO (X) >= FIRST_PSEUDO_REGISTER)
1742
1743 #define REG_OK_FOR_BASE_NONSTRICT_P(X) \
1744 (GENERAL_REGNO_P (REGNO (X)) \
1745 || REGNO (X) == ARG_POINTER_REGNUM \
1746 || REGNO (X) == FRAME_POINTER_REGNUM \
1747 || REGNO (X) >= FIRST_PSEUDO_REGISTER)
1748
1749 /* Strict versions, hard registers only */
1750 #define REG_OK_FOR_INDEX_STRICT_P(X) REGNO_OK_FOR_INDEX_P (REGNO (X))
1751 #define REG_OK_FOR_BASE_STRICT_P(X) REGNO_OK_FOR_BASE_P (REGNO (X))
1752
1753 #ifndef REG_OK_STRICT
1754 #define REG_OK_FOR_INDEX_P(X) REG_OK_FOR_INDEX_NONSTRICT_P (X)
1755 #define REG_OK_FOR_BASE_P(X) REG_OK_FOR_BASE_NONSTRICT_P (X)
1756
1757 #else
1758 #define REG_OK_FOR_INDEX_P(X) REG_OK_FOR_INDEX_STRICT_P (X)
1759 #define REG_OK_FOR_BASE_P(X) REG_OK_FOR_BASE_STRICT_P (X)
1760 #endif
1761
1762 /* TARGET_LEGITIMATE_ADDRESS_P recognizes an RTL expression
1763 that is a valid memory address for an instruction.
1764 The MODE argument is the machine mode for the MEM expression
1765 that wants to use this address.
1766
1767 The other macros defined here are used only in TARGET_LEGITIMATE_ADDRESS_P,
1768 except for CONSTANT_ADDRESS_P which is usually machine-independent.
1769
1770 See legitimize_pic_address in i386.c for details as to what
1771 constitutes a legitimate address when -fpic is used. */
1772
1773 #define MAX_REGS_PER_ADDRESS 2
1774
1775 #define CONSTANT_ADDRESS_P(X) constant_address_p (X)
1776
1777 /* Try a machine-dependent way of reloading an illegitimate address
1778 operand. If we find one, push the reload and jump to WIN. This
1779 macro is used in only one place: `find_reloads_address' in reload.c. */
1780
1781 #define LEGITIMIZE_RELOAD_ADDRESS(X, MODE, OPNUM, TYPE, INDL, WIN) \
1782 do { \
1783 if (ix86_legitimize_reload_address ((X), (MODE), (OPNUM), \
1784 (int)(TYPE), (INDL))) \
1785 goto WIN; \
1786 } while (0)
1787
1788 /* If defined, a C expression to determine the base term of address X.
1789 This macro is used in only one place: `find_base_term' in alias.c.
1790
1791 It is always safe for this macro to not be defined. It exists so
1792 that alias analysis can understand machine-dependent addresses.
1793
1794 The typical use of this macro is to handle addresses containing
1795 a label_ref or symbol_ref within an UNSPEC. */
1796
1797 #define FIND_BASE_TERM(X) ix86_find_base_term (X)
1798
1799 /* Nonzero if the constant value X is a legitimate general operand
1800 when generating PIC code. It is given that flag_pic is on and
1801 that X satisfies CONSTANT_P or is a CONST_DOUBLE. */
1802
1803 #define LEGITIMATE_PIC_OPERAND_P(X) legitimate_pic_operand_p (X)
1804
1805 #define SYMBOLIC_CONST(X) \
1806 (GET_CODE (X) == SYMBOL_REF \
1807 || GET_CODE (X) == LABEL_REF \
1808 || (GET_CODE (X) == CONST && symbolic_reference_mentioned_p (X)))
1809 \f
1810 /* Max number of args passed in registers. If this is more than 3, we will
1811 have problems with ebx (register #4), since it is a caller save register and
1812 is also used as the pic register in ELF. So for now, don't allow more than
1813 3 registers to be passed in registers. */
1814
1815 /* Abi specific values for REGPARM_MAX and SSE_REGPARM_MAX */
1816 #define X86_64_REGPARM_MAX 6
1817 #define X86_64_MS_REGPARM_MAX 4
1818
1819 #define X86_32_REGPARM_MAX 3
1820
1821 #define REGPARM_MAX \
1822 (TARGET_64BIT \
1823 ? (TARGET_64BIT_MS_ABI \
1824 ? X86_64_MS_REGPARM_MAX \
1825 : X86_64_REGPARM_MAX) \
1826 : X86_32_REGPARM_MAX)
1827
1828 #define X86_64_SSE_REGPARM_MAX 8
1829 #define X86_64_MS_SSE_REGPARM_MAX 4
1830
1831 #define X86_32_SSE_REGPARM_MAX (TARGET_SSE ? (TARGET_MACHO ? 4 : 3) : 0)
1832
1833 #define SSE_REGPARM_MAX \
1834 (TARGET_64BIT \
1835 ? (TARGET_64BIT_MS_ABI \
1836 ? X86_64_MS_SSE_REGPARM_MAX \
1837 : X86_64_SSE_REGPARM_MAX) \
1838 : X86_32_SSE_REGPARM_MAX)
1839
1840 #define MMX_REGPARM_MAX (TARGET_64BIT ? 0 : (TARGET_MMX ? 3 : 0))
1841 \f
1842 /* Specify the machine mode that this machine uses
1843 for the index in the tablejump instruction. */
1844 #define CASE_VECTOR_MODE \
1845 (!TARGET_LP64 || (flag_pic && ix86_cmodel != CM_LARGE_PIC) ? SImode : DImode)
1846
1847 /* Define this as 1 if `char' should by default be signed; else as 0. */
1848 #define DEFAULT_SIGNED_CHAR 1
1849
1850 /* Max number of bytes we can move from memory to memory
1851 in one reasonably fast instruction. */
1852 #define MOVE_MAX 16
1853
1854 /* MOVE_MAX_PIECES is the number of bytes at a time which we can
1855 move efficiently, as opposed to MOVE_MAX which is the maximum
1856 number of bytes we can move with a single instruction. */
1857 #define MOVE_MAX_PIECES UNITS_PER_WORD
1858
1859 /* If a memory-to-memory move would take MOVE_RATIO or more simple
1860 move-instruction pairs, we will do a movmem or libcall instead.
1861 Increasing the value will always make code faster, but eventually
1862 incurs high cost in increased code size.
1863
1864 If you don't define this, a reasonable default is used. */
1865
1866 #define MOVE_RATIO(speed) ((speed) ? ix86_cost->move_ratio : 3)
1867
1868 /* If a clear memory operation would take CLEAR_RATIO or more simple
1869 move-instruction sequences, we will do a clrmem or libcall instead. */
1870
1871 #define CLEAR_RATIO(speed) ((speed) ? MIN (6, ix86_cost->move_ratio) : 2)
1872
1873 /* Define if shifts truncate the shift count which implies one can
1874 omit a sign-extension or zero-extension of a shift count.
1875
1876 On i386, shifts do truncate the count. But bit test instructions
1877 take the modulo of the bit offset operand. */
1878
1879 /* #define SHIFT_COUNT_TRUNCATED */
1880
1881 /* Value is 1 if truncating an integer of INPREC bits to OUTPREC bits
1882 is done just by pretending it is already truncated. */
1883 #define TRULY_NOOP_TRUNCATION(OUTPREC, INPREC) 1
1884
1885 /* A macro to update M and UNSIGNEDP when an object whose type is
1886 TYPE and which has the specified mode and signedness is to be
1887 stored in a register. This macro is only called when TYPE is a
1888 scalar type.
1889
1890 On i386 it is sometimes useful to promote HImode and QImode
1891 quantities to SImode. The choice depends on target type. */
1892
1893 #define PROMOTE_MODE(MODE, UNSIGNEDP, TYPE) \
1894 do { \
1895 if (((MODE) == HImode && TARGET_PROMOTE_HI_REGS) \
1896 || ((MODE) == QImode && TARGET_PROMOTE_QI_REGS)) \
1897 (MODE) = SImode; \
1898 } while (0)
1899
1900 /* Specify the machine mode that pointers have.
1901 After generation of rtl, the compiler makes no further distinction
1902 between pointers and any other objects of this machine mode. */
1903 #define Pmode (ix86_pmode == PMODE_DI ? DImode : SImode)
1904
1905 /* A C expression whose value is zero if pointers that need to be extended
1906 from being `POINTER_SIZE' bits wide to `Pmode' are sign-extended and
1907 greater then zero if they are zero-extended and less then zero if the
1908 ptr_extend instruction should be used. */
1909
1910 #define POINTERS_EXTEND_UNSIGNED 1
1911
1912 /* A function address in a call instruction
1913 is a byte address (for indexing purposes)
1914 so give the MEM rtx a byte's mode. */
1915 #define FUNCTION_MODE QImode
1916 \f
1917
1918 /* A C expression for the cost of a branch instruction. A value of 1
1919 is the default; other values are interpreted relative to that. */
1920
1921 #define BRANCH_COST(speed_p, predictable_p) \
1922 (!(speed_p) ? 2 : (predictable_p) ? 0 : ix86_branch_cost)
1923
1924 /* An integer expression for the size in bits of the largest integer machine
1925 mode that should actually be used. We allow pairs of registers. */
1926 #define MAX_FIXED_MODE_SIZE GET_MODE_BITSIZE (TARGET_64BIT ? TImode : DImode)
1927
1928 /* Define this macro as a C expression which is nonzero if accessing
1929 less than a word of memory (i.e. a `char' or a `short') is no
1930 faster than accessing a word of memory, i.e., if such access
1931 require more than one instruction or if there is no difference in
1932 cost between byte and (aligned) word loads.
1933
1934 When this macro is not defined, the compiler will access a field by
1935 finding the smallest containing object; when it is defined, a
1936 fullword load will be used if alignment permits. Unless bytes
1937 accesses are faster than word accesses, using word accesses is
1938 preferable since it may eliminate subsequent memory access if
1939 subsequent accesses occur to other fields in the same word of the
1940 structure, but to different bytes. */
1941
1942 #define SLOW_BYTE_ACCESS 0
1943
1944 /* Nonzero if access to memory by shorts is slow and undesirable. */
1945 #define SLOW_SHORT_ACCESS 0
1946
1947 /* Define this macro to be the value 1 if unaligned accesses have a
1948 cost many times greater than aligned accesses, for example if they
1949 are emulated in a trap handler.
1950
1951 When this macro is nonzero, the compiler will act as if
1952 `STRICT_ALIGNMENT' were nonzero when generating code for block
1953 moves. This can cause significantly more instructions to be
1954 produced. Therefore, do not set this macro nonzero if unaligned
1955 accesses only add a cycle or two to the time for a memory access.
1956
1957 If the value of this macro is always zero, it need not be defined. */
1958
1959 /* #define SLOW_UNALIGNED_ACCESS(MODE, ALIGN) 0 */
1960
1961 /* Define this macro if it is as good or better to call a constant
1962 function address than to call an address kept in a register.
1963
1964 Desirable on the 386 because a CALL with a constant address is
1965 faster than one with a register address. */
1966
1967 #define NO_FUNCTION_CSE
1968 \f
1969 /* Given a comparison code (EQ, NE, etc.) and the first operand of a COMPARE,
1970 return the mode to be used for the comparison.
1971
1972 For floating-point equality comparisons, CCFPEQmode should be used.
1973 VOIDmode should be used in all other cases.
1974
1975 For integer comparisons against zero, reduce to CCNOmode or CCZmode if
1976 possible, to allow for more combinations. */
1977
1978 #define SELECT_CC_MODE(OP, X, Y) ix86_cc_mode ((OP), (X), (Y))
1979
1980 /* Return nonzero if MODE implies a floating point inequality can be
1981 reversed. */
1982
1983 #define REVERSIBLE_CC_MODE(MODE) 1
1984
1985 /* A C expression whose value is reversed condition code of the CODE for
1986 comparison done in CC_MODE mode. */
1987 #define REVERSE_CONDITION(CODE, MODE) ix86_reverse_condition ((CODE), (MODE))
1988
1989 \f
1990 /* Control the assembler format that we output, to the extent
1991 this does not vary between assemblers. */
1992
1993 /* How to refer to registers in assembler output.
1994 This sequence is indexed by compiler's hard-register-number (see above). */
1995
1996 /* In order to refer to the first 8 regs as 32-bit regs, prefix an "e".
1997 For non floating point regs, the following are the HImode names.
1998
1999 For float regs, the stack top is sometimes referred to as "%st(0)"
2000 instead of just "%st". TARGET_PRINT_OPERAND handles this with the
2001 "y" code. */
2002
2003 #define HI_REGISTER_NAMES \
2004 {"ax","dx","cx","bx","si","di","bp","sp", \
2005 "st","st(1)","st(2)","st(3)","st(4)","st(5)","st(6)","st(7)", \
2006 "argp", "flags", "fpsr", "fpcr", "frame", \
2007 "xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7", \
2008 "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7", \
2009 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15", \
2010 "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15", \
2011 "xmm16", "xmm17", "xmm18", "xmm19", \
2012 "xmm20", "xmm21", "xmm22", "xmm23", \
2013 "xmm24", "xmm25", "xmm26", "xmm27", \
2014 "xmm28", "xmm29", "xmm30", "xmm31", \
2015 "k0", "k1", "k2", "k3", "k4", "k5", "k6", "k7" }
2016
2017 #define REGISTER_NAMES HI_REGISTER_NAMES
2018
2019 /* Table of additional register names to use in user input. */
2020
2021 #define ADDITIONAL_REGISTER_NAMES \
2022 { { "eax", 0 }, { "edx", 1 }, { "ecx", 2 }, { "ebx", 3 }, \
2023 { "esi", 4 }, { "edi", 5 }, { "ebp", 6 }, { "esp", 7 }, \
2024 { "rax", 0 }, { "rdx", 1 }, { "rcx", 2 }, { "rbx", 3 }, \
2025 { "rsi", 4 }, { "rdi", 5 }, { "rbp", 6 }, { "rsp", 7 }, \
2026 { "al", 0 }, { "dl", 1 }, { "cl", 2 }, { "bl", 3 }, \
2027 { "ah", 0 }, { "dh", 1 }, { "ch", 2 }, { "bh", 3 }, \
2028 { "ymm0", 21}, { "ymm1", 22}, { "ymm2", 23}, { "ymm3", 24}, \
2029 { "ymm4", 25}, { "ymm5", 26}, { "ymm6", 27}, { "ymm7", 28}, \
2030 { "ymm8", 45}, { "ymm9", 46}, { "ymm10", 47}, { "ymm11", 48}, \
2031 { "ymm12", 49}, { "ymm13", 50}, { "ymm14", 51}, { "ymm15", 52}, \
2032 { "ymm16", 53}, { "ymm17", 54}, { "ymm18", 55}, { "ymm19", 56}, \
2033 { "ymm20", 57}, { "ymm21", 58}, { "ymm22", 59}, { "ymm23", 60}, \
2034 { "ymm24", 61}, { "ymm25", 62}, { "ymm26", 63}, { "ymm27", 64}, \
2035 { "ymm28", 65}, { "ymm29", 66}, { "ymm30", 67}, { "ymm31", 68}, \
2036 { "zmm0", 21}, { "zmm1", 22}, { "zmm2", 23}, { "zmm3", 24}, \
2037 { "zmm4", 25}, { "zmm5", 26}, { "zmm6", 27}, { "zmm7", 28}, \
2038 { "zmm8", 45}, { "zmm9", 46}, { "zmm10", 47}, { "zmm11", 48}, \
2039 { "zmm12", 49}, { "zmm13", 50}, { "zmm14", 51}, { "zmm15", 52}, \
2040 { "zmm16", 53}, { "zmm17", 54}, { "zmm18", 55}, { "zmm19", 56}, \
2041 { "zmm20", 57}, { "zmm21", 58}, { "zmm22", 59}, { "zmm23", 60}, \
2042 { "zmm24", 61}, { "zmm25", 62}, { "zmm26", 63}, { "zmm27", 64}, \
2043 { "zmm28", 65}, { "zmm29", 66}, { "zmm30", 67}, { "zmm31", 68} }
2044
2045 /* Note we are omitting these since currently I don't know how
2046 to get gcc to use these, since they want the same but different
2047 number as al, and ax.
2048 */
2049
2050 #define QI_REGISTER_NAMES \
2051 {"al", "dl", "cl", "bl", "sil", "dil", "bpl", "spl",}
2052
2053 /* These parallel the array above, and can be used to access bits 8:15
2054 of regs 0 through 3. */
2055
2056 #define QI_HIGH_REGISTER_NAMES \
2057 {"ah", "dh", "ch", "bh", }
2058
2059 /* How to renumber registers for dbx and gdb. */
2060
2061 #define DBX_REGISTER_NUMBER(N) \
2062 (TARGET_64BIT ? dbx64_register_map[(N)] : dbx_register_map[(N)])
2063
2064 extern int const dbx_register_map[FIRST_PSEUDO_REGISTER];
2065 extern int const dbx64_register_map[FIRST_PSEUDO_REGISTER];
2066 extern int const svr4_dbx_register_map[FIRST_PSEUDO_REGISTER];
2067
2068 extern int const x86_64_ms_sysv_extra_clobbered_registers[12];
2069
2070 /* Before the prologue, RA is at 0(%esp). */
2071 #define INCOMING_RETURN_ADDR_RTX \
2072 gen_rtx_MEM (VOIDmode, gen_rtx_REG (VOIDmode, STACK_POINTER_REGNUM))
2073
2074 /* After the prologue, RA is at -4(AP) in the current frame. */
2075 #define RETURN_ADDR_RTX(COUNT, FRAME) \
2076 ((COUNT) == 0 \
2077 ? gen_rtx_MEM (Pmode, plus_constant (Pmode, arg_pointer_rtx, \
2078 -UNITS_PER_WORD)) \
2079 : gen_rtx_MEM (Pmode, plus_constant (Pmode, FRAME, UNITS_PER_WORD)))
2080
2081 /* PC is dbx register 8; let's use that column for RA. */
2082 #define DWARF_FRAME_RETURN_COLUMN (TARGET_64BIT ? 16 : 8)
2083
2084 /* Before the prologue, the top of the frame is at 4(%esp). */
2085 #define INCOMING_FRAME_SP_OFFSET UNITS_PER_WORD
2086
2087 /* Describe how we implement __builtin_eh_return. */
2088 #define EH_RETURN_DATA_REGNO(N) ((N) <= DX_REG ? (N) : INVALID_REGNUM)
2089 #define EH_RETURN_STACKADJ_RTX gen_rtx_REG (Pmode, CX_REG)
2090
2091
2092 /* Select a format to encode pointers in exception handling data. CODE
2093 is 0 for data, 1 for code labels, 2 for function pointers. GLOBAL is
2094 true if the symbol may be affected by dynamic relocations.
2095
2096 ??? All x86 object file formats are capable of representing this.
2097 After all, the relocation needed is the same as for the call insn.
2098 Whether or not a particular assembler allows us to enter such, I
2099 guess we'll have to see. */
2100 #define ASM_PREFERRED_EH_DATA_FORMAT(CODE, GLOBAL) \
2101 asm_preferred_eh_data_format ((CODE), (GLOBAL))
2102
2103 /* This is how to output an insn to push a register on the stack.
2104 It need not be very fast code. */
2105
2106 #define ASM_OUTPUT_REG_PUSH(FILE, REGNO) \
2107 do { \
2108 if (TARGET_64BIT) \
2109 asm_fprintf ((FILE), "\tpush{q}\t%%r%s\n", \
2110 reg_names[(REGNO)] + (REX_INT_REGNO_P (REGNO) != 0)); \
2111 else \
2112 asm_fprintf ((FILE), "\tpush{l}\t%%e%s\n", reg_names[(REGNO)]); \
2113 } while (0)
2114
2115 /* This is how to output an insn to pop a register from the stack.
2116 It need not be very fast code. */
2117
2118 #define ASM_OUTPUT_REG_POP(FILE, REGNO) \
2119 do { \
2120 if (TARGET_64BIT) \
2121 asm_fprintf ((FILE), "\tpop{q}\t%%r%s\n", \
2122 reg_names[(REGNO)] + (REX_INT_REGNO_P (REGNO) != 0)); \
2123 else \
2124 asm_fprintf ((FILE), "\tpop{l}\t%%e%s\n", reg_names[(REGNO)]); \
2125 } while (0)
2126
2127 /* This is how to output an element of a case-vector that is absolute. */
2128
2129 #define ASM_OUTPUT_ADDR_VEC_ELT(FILE, VALUE) \
2130 ix86_output_addr_vec_elt ((FILE), (VALUE))
2131
2132 /* This is how to output an element of a case-vector that is relative. */
2133
2134 #define ASM_OUTPUT_ADDR_DIFF_ELT(FILE, BODY, VALUE, REL) \
2135 ix86_output_addr_diff_elt ((FILE), (VALUE), (REL))
2136
2137 /* When we see %v, we will print the 'v' prefix if TARGET_AVX is true. */
2138
2139 #define ASM_OUTPUT_AVX_PREFIX(STREAM, PTR) \
2140 { \
2141 if ((PTR)[0] == '%' && (PTR)[1] == 'v') \
2142 (PTR) += TARGET_AVX ? 1 : 2; \
2143 }
2144
2145 /* A C statement or statements which output an assembler instruction
2146 opcode to the stdio stream STREAM. The macro-operand PTR is a
2147 variable of type `char *' which points to the opcode name in
2148 its "internal" form--the form that is written in the machine
2149 description. */
2150
2151 #define ASM_OUTPUT_OPCODE(STREAM, PTR) \
2152 ASM_OUTPUT_AVX_PREFIX ((STREAM), (PTR))
2153
2154 /* A C statement to output to the stdio stream FILE an assembler
2155 command to pad the location counter to a multiple of 1<<LOG
2156 bytes if it is within MAX_SKIP bytes. */
2157
2158 #ifdef HAVE_GAS_MAX_SKIP_P2ALIGN
2159 #undef ASM_OUTPUT_MAX_SKIP_PAD
2160 #define ASM_OUTPUT_MAX_SKIP_PAD(FILE, LOG, MAX_SKIP) \
2161 if ((LOG) != 0) \
2162 { \
2163 if ((MAX_SKIP) == 0) \
2164 fprintf ((FILE), "\t.p2align %d\n", (LOG)); \
2165 else \
2166 fprintf ((FILE), "\t.p2align %d,,%d\n", (LOG), (MAX_SKIP)); \
2167 }
2168 #endif
2169
2170 /* Write the extra assembler code needed to declare a function
2171 properly. */
2172
2173 #undef ASM_OUTPUT_FUNCTION_LABEL
2174 #define ASM_OUTPUT_FUNCTION_LABEL(FILE, NAME, DECL) \
2175 ix86_asm_output_function_label (FILE, NAME, DECL)
2176
2177 /* Under some conditions we need jump tables in the text section,
2178 because the assembler cannot handle label differences between
2179 sections. This is the case for x86_64 on Mach-O for example. */
2180
2181 #define JUMP_TABLES_IN_TEXT_SECTION \
2182 (flag_pic && ((TARGET_MACHO && TARGET_64BIT) \
2183 || (!TARGET_64BIT && !HAVE_AS_GOTOFF_IN_DATA)))
2184
2185 /* Switch to init or fini section via SECTION_OP, emit a call to FUNC,
2186 and switch back. For x86 we do this only to save a few bytes that
2187 would otherwise be unused in the text section. */
2188 #define CRT_MKSTR2(VAL) #VAL
2189 #define CRT_MKSTR(x) CRT_MKSTR2(x)
2190
2191 #define CRT_CALL_STATIC_FUNCTION(SECTION_OP, FUNC) \
2192 asm (SECTION_OP "\n\t" \
2193 "call " CRT_MKSTR(__USER_LABEL_PREFIX__) #FUNC "\n" \
2194 TEXT_SECTION_ASM_OP);
2195
2196 /* Default threshold for putting data in large sections
2197 with x86-64 medium memory model */
2198 #define DEFAULT_LARGE_SECTION_THRESHOLD 65536
2199 \f
2200 /* Which processor to tune code generation for. These must be in sync
2201 with processor_target_table in i386.c. */
2202
2203 enum processor_type
2204 {
2205 PROCESSOR_GENERIC = 0,
2206 PROCESSOR_I386, /* 80386 */
2207 PROCESSOR_I486, /* 80486DX, 80486SX, 80486DX[24] */
2208 PROCESSOR_PENTIUM,
2209 PROCESSOR_PENTIUMPRO,
2210 PROCESSOR_PENTIUM4,
2211 PROCESSOR_NOCONA,
2212 PROCESSOR_CORE2,
2213 PROCESSOR_NEHALEM,
2214 PROCESSOR_SANDYBRIDGE,
2215 PROCESSOR_HASWELL,
2216 PROCESSOR_BONNELL,
2217 PROCESSOR_SILVERMONT,
2218 PROCESSOR_INTEL,
2219 PROCESSOR_GEODE,
2220 PROCESSOR_K6,
2221 PROCESSOR_ATHLON,
2222 PROCESSOR_K8,
2223 PROCESSOR_AMDFAM10,
2224 PROCESSOR_BDVER1,
2225 PROCESSOR_BDVER2,
2226 PROCESSOR_BDVER3,
2227 PROCESSOR_BDVER4,
2228 PROCESSOR_BTVER1,
2229 PROCESSOR_BTVER2,
2230 PROCESSOR_max
2231 };
2232
2233 extern enum processor_type ix86_tune;
2234 extern enum processor_type ix86_arch;
2235
2236 /* Size of the RED_ZONE area. */
2237 #define RED_ZONE_SIZE 128
2238 /* Reserved area of the red zone for temporaries. */
2239 #define RED_ZONE_RESERVE 8
2240
2241 extern unsigned int ix86_preferred_stack_boundary;
2242 extern unsigned int ix86_incoming_stack_boundary;
2243
2244 /* Smallest class containing REGNO. */
2245 extern enum reg_class const regclass_map[FIRST_PSEUDO_REGISTER];
2246
2247 enum ix86_fpcmp_strategy {
2248 IX86_FPCMP_SAHF,
2249 IX86_FPCMP_COMI,
2250 IX86_FPCMP_ARITH
2251 };
2252 \f
2253 /* To properly truncate FP values into integers, we need to set i387 control
2254 word. We can't emit proper mode switching code before reload, as spills
2255 generated by reload may truncate values incorrectly, but we still can avoid
2256 redundant computation of new control word by the mode switching pass.
2257 The fldcw instructions are still emitted redundantly, but this is probably
2258 not going to be noticeable problem, as most CPUs do have fast path for
2259 the sequence.
2260
2261 The machinery is to emit simple truncation instructions and split them
2262 before reload to instructions having USEs of two memory locations that
2263 are filled by this code to old and new control word.
2264
2265 Post-reload pass may be later used to eliminate the redundant fildcw if
2266 needed. */
2267
2268 enum ix86_entity
2269 {
2270 AVX_U128 = 0,
2271 I387_TRUNC,
2272 I387_FLOOR,
2273 I387_CEIL,
2274 I387_MASK_PM,
2275 MAX_386_ENTITIES
2276 };
2277
2278 enum ix86_stack_slot
2279 {
2280 SLOT_TEMP = 0,
2281 SLOT_CW_STORED,
2282 SLOT_CW_TRUNC,
2283 SLOT_CW_FLOOR,
2284 SLOT_CW_CEIL,
2285 SLOT_CW_MASK_PM,
2286 MAX_386_STACK_LOCALS
2287 };
2288
2289 enum avx_u128_state
2290 {
2291 AVX_U128_CLEAN,
2292 AVX_U128_DIRTY,
2293 AVX_U128_ANY
2294 };
2295
2296 /* Define this macro if the port needs extra instructions inserted
2297 for mode switching in an optimizing compilation. */
2298
2299 #define OPTIMIZE_MODE_SWITCHING(ENTITY) \
2300 ix86_optimize_mode_switching[(ENTITY)]
2301
2302 /* If you define `OPTIMIZE_MODE_SWITCHING', you have to define this as
2303 initializer for an array of integers. Each initializer element N
2304 refers to an entity that needs mode switching, and specifies the
2305 number of different modes that might need to be set for this
2306 entity. The position of the initializer in the initializer -
2307 starting counting at zero - determines the integer that is used to
2308 refer to the mode-switched entity in question. */
2309
2310 #define NUM_MODES_FOR_MODE_SWITCHING \
2311 { AVX_U128_ANY, I387_CW_ANY, I387_CW_ANY, I387_CW_ANY, I387_CW_ANY }
2312
2313 /* ENTITY is an integer specifying a mode-switched entity. If
2314 `OPTIMIZE_MODE_SWITCHING' is defined, you must define this macro to
2315 return an integer value not larger than the corresponding element
2316 in `NUM_MODES_FOR_MODE_SWITCHING', to denote the mode that ENTITY
2317 must be switched into prior to the execution of INSN. */
2318
2319 #define MODE_NEEDED(ENTITY, I) ix86_mode_needed ((ENTITY), (I))
2320
2321 /* If this macro is defined, it is evaluated for every INSN during
2322 mode switching. It determines the mode that an insn results in (if
2323 different from the incoming mode). */
2324
2325 #define MODE_AFTER(ENTITY, MODE, I) ix86_mode_after ((ENTITY), (MODE), (I))
2326
2327 /* If this macro is defined, it is evaluated for every ENTITY that
2328 needs mode switching. It should evaluate to an integer, which is
2329 a mode that ENTITY is assumed to be switched to at function entry. */
2330
2331 #define MODE_ENTRY(ENTITY) ix86_mode_entry (ENTITY)
2332
2333 /* If this macro is defined, it is evaluated for every ENTITY that
2334 needs mode switching. It should evaluate to an integer, which is
2335 a mode that ENTITY is assumed to be switched to at function exit. */
2336
2337 #define MODE_EXIT(ENTITY) ix86_mode_exit (ENTITY)
2338
2339 /* This macro specifies the order in which modes for ENTITY are
2340 processed. 0 is the highest priority. */
2341
2342 #define MODE_PRIORITY_TO_MODE(ENTITY, N) (N)
2343
2344 /* Generate one or more insns to set ENTITY to MODE. HARD_REG_LIVE
2345 is the set of hard registers live at the point where the insn(s)
2346 are to be inserted. */
2347
2348 #define EMIT_MODE_SET(ENTITY, MODE, HARD_REGS_LIVE) \
2349 ix86_emit_mode_set ((ENTITY), (MODE), (HARD_REGS_LIVE))
2350 \f
2351 /* Avoid renaming of stack registers, as doing so in combination with
2352 scheduling just increases amount of live registers at time and in
2353 the turn amount of fxch instructions needed.
2354
2355 ??? Maybe Pentium chips benefits from renaming, someone can try....
2356
2357 Don't rename evex to non-evex sse registers. */
2358
2359 #define HARD_REGNO_RENAME_OK(SRC, TARGET) (!STACK_REGNO_P (SRC) && \
2360 (EXT_REX_SSE_REGNO_P (SRC) == \
2361 EXT_REX_SSE_REGNO_P (TARGET)))
2362
2363 \f
2364 #define FASTCALL_PREFIX '@'
2365 \f
2366 /* Machine specific frame tracking during prologue/epilogue generation. */
2367
2368 #ifndef USED_FOR_TARGET
2369 struct GTY(()) machine_frame_state
2370 {
2371 /* This pair tracks the currently active CFA as reg+offset. When reg
2372 is drap_reg, we don't bother trying to record here the real CFA when
2373 it might really be a DW_CFA_def_cfa_expression. */
2374 rtx cfa_reg;
2375 HOST_WIDE_INT cfa_offset;
2376
2377 /* The current offset (canonically from the CFA) of ESP and EBP.
2378 When stack frame re-alignment is active, these may not be relative
2379 to the CFA. However, in all cases they are relative to the offsets
2380 of the saved registers stored in ix86_frame. */
2381 HOST_WIDE_INT sp_offset;
2382 HOST_WIDE_INT fp_offset;
2383
2384 /* The size of the red-zone that may be assumed for the purposes of
2385 eliding register restore notes in the epilogue. This may be zero
2386 if no red-zone is in effect, or may be reduced from the real
2387 red-zone value by a maximum runtime stack re-alignment value. */
2388 int red_zone_offset;
2389
2390 /* Indicate whether each of ESP, EBP or DRAP currently holds a valid
2391 value within the frame. If false then the offset above should be
2392 ignored. Note that DRAP, if valid, *always* points to the CFA and
2393 thus has an offset of zero. */
2394 BOOL_BITFIELD sp_valid : 1;
2395 BOOL_BITFIELD fp_valid : 1;
2396 BOOL_BITFIELD drap_valid : 1;
2397
2398 /* Indicate whether the local stack frame has been re-aligned. When
2399 set, the SP/FP offsets above are relative to the aligned frame
2400 and not the CFA. */
2401 BOOL_BITFIELD realigned : 1;
2402 };
2403
2404 /* Private to winnt.c. */
2405 struct seh_frame_state;
2406
2407 struct GTY(()) machine_function {
2408 struct stack_local_entry *stack_locals;
2409 const char *some_ld_name;
2410 int varargs_gpr_size;
2411 int varargs_fpr_size;
2412 int optimize_mode_switching[MAX_386_ENTITIES];
2413
2414 /* Number of saved registers USE_FAST_PROLOGUE_EPILOGUE
2415 has been computed for. */
2416 int use_fast_prologue_epilogue_nregs;
2417
2418 /* For -fsplit-stack support: A stack local which holds a pointer to
2419 the stack arguments for a function with a variable number of
2420 arguments. This is set at the start of the function and is used
2421 to initialize the overflow_arg_area field of the va_list
2422 structure. */
2423 rtx split_stack_varargs_pointer;
2424
2425 /* This value is used for amd64 targets and specifies the current abi
2426 to be used. MS_ABI means ms abi. Otherwise SYSV_ABI means sysv abi. */
2427 ENUM_BITFIELD(calling_abi) call_abi : 8;
2428
2429 /* Nonzero if the function accesses a previous frame. */
2430 BOOL_BITFIELD accesses_prev_frame : 1;
2431
2432 /* Nonzero if the function requires a CLD in the prologue. */
2433 BOOL_BITFIELD needs_cld : 1;
2434
2435 /* Set by ix86_compute_frame_layout and used by prologue/epilogue
2436 expander to determine the style used. */
2437 BOOL_BITFIELD use_fast_prologue_epilogue : 1;
2438
2439 /* If true, the current function needs the default PIC register, not
2440 an alternate register (on x86) and must not use the red zone (on
2441 x86_64), even if it's a leaf function. We don't want the
2442 function to be regarded as non-leaf because TLS calls need not
2443 affect register allocation. This flag is set when a TLS call
2444 instruction is expanded within a function, and never reset, even
2445 if all such instructions are optimized away. Use the
2446 ix86_current_function_calls_tls_descriptor macro for a better
2447 approximation. */
2448 BOOL_BITFIELD tls_descriptor_call_expanded_p : 1;
2449
2450 /* If true, the current function has a STATIC_CHAIN is placed on the
2451 stack below the return address. */
2452 BOOL_BITFIELD static_chain_on_stack : 1;
2453
2454 /* If true, it is safe to not save/restore DRAP register. */
2455 BOOL_BITFIELD no_drap_save_restore : 1;
2456
2457 /* During prologue/epilogue generation, the current frame state.
2458 Otherwise, the frame state at the end of the prologue. */
2459 struct machine_frame_state fs;
2460
2461 /* During SEH output, this is non-null. */
2462 struct seh_frame_state * GTY((skip(""))) seh;
2463 };
2464 #endif
2465
2466 #define ix86_stack_locals (cfun->machine->stack_locals)
2467 #define ix86_varargs_gpr_size (cfun->machine->varargs_gpr_size)
2468 #define ix86_varargs_fpr_size (cfun->machine->varargs_fpr_size)
2469 #define ix86_optimize_mode_switching (cfun->machine->optimize_mode_switching)
2470 #define ix86_current_function_needs_cld (cfun->machine->needs_cld)
2471 #define ix86_tls_descriptor_calls_expanded_in_cfun \
2472 (cfun->machine->tls_descriptor_call_expanded_p)
2473 /* Since tls_descriptor_call_expanded is not cleared, even if all TLS
2474 calls are optimized away, we try to detect cases in which it was
2475 optimized away. Since such instructions (use (reg REG_SP)), we can
2476 verify whether there's any such instruction live by testing that
2477 REG_SP is live. */
2478 #define ix86_current_function_calls_tls_descriptor \
2479 (ix86_tls_descriptor_calls_expanded_in_cfun && df_regs_ever_live_p (SP_REG))
2480 #define ix86_static_chain_on_stack (cfun->machine->static_chain_on_stack)
2481
2482 /* Control behavior of x86_file_start. */
2483 #define X86_FILE_START_VERSION_DIRECTIVE false
2484 #define X86_FILE_START_FLTUSED false
2485
2486 /* Flag to mark data that is in the large address area. */
2487 #define SYMBOL_FLAG_FAR_ADDR (SYMBOL_FLAG_MACH_DEP << 0)
2488 #define SYMBOL_REF_FAR_ADDR_P(X) \
2489 ((SYMBOL_REF_FLAGS (X) & SYMBOL_FLAG_FAR_ADDR) != 0)
2490
2491 /* Flags to mark dllimport/dllexport. Used by PE ports, but handy to
2492 have defined always, to avoid ifdefing. */
2493 #define SYMBOL_FLAG_DLLIMPORT (SYMBOL_FLAG_MACH_DEP << 1)
2494 #define SYMBOL_REF_DLLIMPORT_P(X) \
2495 ((SYMBOL_REF_FLAGS (X) & SYMBOL_FLAG_DLLIMPORT) != 0)
2496
2497 #define SYMBOL_FLAG_DLLEXPORT (SYMBOL_FLAG_MACH_DEP << 2)
2498 #define SYMBOL_REF_DLLEXPORT_P(X) \
2499 ((SYMBOL_REF_FLAGS (X) & SYMBOL_FLAG_DLLEXPORT) != 0)
2500
2501 #define SYMBOL_FLAG_STUBVAR (SYMBOL_FLAG_MACH_DEP << 4)
2502 #define SYMBOL_REF_STUBVAR_P(X) \
2503 ((SYMBOL_REF_FLAGS (X) & SYMBOL_FLAG_STUBVAR) != 0)
2504
2505 extern void debug_ready_dispatch (void);
2506 extern void debug_dispatch_window (int);
2507
2508 /* The value at zero is only defined for the BMI instructions
2509 LZCNT and TZCNT, not the BSR/BSF insns in the original isa. */
2510 #define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
2511 ((VALUE) = GET_MODE_BITSIZE (MODE), TARGET_BMI)
2512 #define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
2513 ((VALUE) = GET_MODE_BITSIZE (MODE), TARGET_LZCNT)
2514
2515
2516 /* Flags returned by ix86_get_callcvt (). */
2517 #define IX86_CALLCVT_CDECL 0x1
2518 #define IX86_CALLCVT_STDCALL 0x2
2519 #define IX86_CALLCVT_FASTCALL 0x4
2520 #define IX86_CALLCVT_THISCALL 0x8
2521 #define IX86_CALLCVT_REGPARM 0x10
2522 #define IX86_CALLCVT_SSEREGPARM 0x20
2523
2524 #define IX86_BASE_CALLCVT(FLAGS) \
2525 ((FLAGS) & (IX86_CALLCVT_CDECL | IX86_CALLCVT_STDCALL \
2526 | IX86_CALLCVT_FASTCALL | IX86_CALLCVT_THISCALL))
2527
2528 #define RECIP_MASK_NONE 0x00
2529 #define RECIP_MASK_DIV 0x01
2530 #define RECIP_MASK_SQRT 0x02
2531 #define RECIP_MASK_VEC_DIV 0x04
2532 #define RECIP_MASK_VEC_SQRT 0x08
2533 #define RECIP_MASK_ALL (RECIP_MASK_DIV | RECIP_MASK_SQRT \
2534 | RECIP_MASK_VEC_DIV | RECIP_MASK_VEC_SQRT)
2535 #define RECIP_MASK_DEFAULT (RECIP_MASK_VEC_DIV | RECIP_MASK_VEC_SQRT)
2536
2537 #define TARGET_RECIP_DIV ((recip_mask & RECIP_MASK_DIV) != 0)
2538 #define TARGET_RECIP_SQRT ((recip_mask & RECIP_MASK_SQRT) != 0)
2539 #define TARGET_RECIP_VEC_DIV ((recip_mask & RECIP_MASK_VEC_DIV) != 0)
2540 #define TARGET_RECIP_VEC_SQRT ((recip_mask & RECIP_MASK_VEC_SQRT) != 0)
2541
2542 #define IX86_HLE_ACQUIRE (1 << 16)
2543 #define IX86_HLE_RELEASE (1 << 17)
2544
2545 /* For switching between functions with different target attributes. */
2546 #define SWITCHABLE_TARGET 1
2547
2548 /*
2549 Local variables:
2550 version-control: t
2551 End:
2552 */