1/* Definitions of target machine for GCC for IA-32.
2 Copyright (C) 1988-2025 Free Software Foundation, Inc.
3
4This file is part of GCC.
5
6GCC is free software; you can redistribute it and/or modify
7it under the terms of the GNU General Public License as published by
8the Free Software Foundation; either version 3, or (at your option)
9any later version.
10
11GCC is distributed in the hope that it will be useful,
12but WITHOUT ANY WARRANTY; without even the implied warranty of
13MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14GNU General Public License for more details.
15
16Under Section 7 of GPL version 3, you are granted additional
17permissions described in the GCC Runtime Library Exception, version
183.1, as published by the Free Software Foundation.
19
20You should have received a copy of the GNU General Public License and
21a copy of the GCC Runtime Library Exception along with this program;
22see 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_CMPXCHG16B TARGET_CX16
43#define TARGET_CMPXCHG16B_P(x) TARGET_CX16_P(x)
44
45#define TARGET_LP64 TARGET_ABI_64
46#define TARGET_LP64_P(x) TARGET_ABI_64_P(x)
47#define TARGET_X32 TARGET_ABI_X32
48#define TARGET_X32_P(x) TARGET_ABI_X32_P(x)
49#define TARGET_16BIT TARGET_CODE16
50#define TARGET_16BIT_P(x) TARGET_CODE16_P(x)
51
52#define TARGET_MMX_WITH_SSE (TARGET_64BIT && TARGET_SSE2)
53
54#define TARGET_APX_EGPR (ix86_apx_features & apx_egpr)
55#define TARGET_APX_PUSH2POP2 (ix86_apx_features & apx_push2pop2)
56#define TARGET_APX_NDD (ix86_apx_features & apx_ndd)
57#define TARGET_APX_PPX (ix86_apx_features & apx_ppx)
58#define TARGET_APX_NF (ix86_apx_features & apx_nf)
59#define TARGET_APX_CCMP (ix86_apx_features & apx_ccmp)
60#define TARGET_APX_ZU (ix86_apx_features & apx_zu)
61
62#include "config/vxworks-dummy.h"
63
64#include "config/i386/i386-opts.h"
65
66#define MAX_STRINGOP_ALGS 4
67
68/* Specify what algorithm to use for stringops on known size.
69 When size is unknown, the UNKNOWN_SIZE alg is used. When size is
70 known at compile time or estimated via feedback, the SIZE array
71 is walked in order until MAX is greater then the estimate (or -1
72 means infinity). Corresponding ALG is used then.
73 When NOALIGN is true the code guaranting the alignment of the memory
74 block is skipped.
75
76 For example initializer:
77 {{256, loop}, {-1, rep_prefix_4_byte}}
78 will use loop for blocks smaller or equal to 256 bytes, rep prefix will
79 be used otherwise. */
80struct stringop_algs
81{
82 const enum stringop_alg unknown_size;
83 const struct stringop_strategy {
84 /* Several older compilers delete the default constructor because of the
85 const entries (see PR100246). Manually specifying a CTOR works around
86 this issue. Since this header is used by code compiled with the C
87 compiler we must guard the addition. */
88#ifdef __cplusplus
89 constexpr
90 stringop_strategy (int _max = -1, enum stringop_alg _alg = libcall,
91 int _noalign = false)
92 : max (_max), alg (_alg), noalign (_noalign) {}
93#endif
94 const int max;
95 const enum stringop_alg alg;
96 int noalign;
97 } size [MAX_STRINGOP_ALGS];
98};
99
100/* Analog of COSTS_N_INSNS when optimizing for size. */
101#ifndef COSTS_N_BYTES
102#define COSTS_N_BYTES(N) ((N) * 2)
103#endif
104
105/* Define the specific costs for a given cpu. NB: hard_register is used
106 by TARGET_REGISTER_MOVE_COST and TARGET_MEMORY_MOVE_COST to compute
107 hard register move costs by register allocator. Relative costs of
108 pseudo register load and store versus pseudo register moves in RTL
109 expressions for TARGET_RTX_COSTS can be different from relative
110 costs of hard registers to get the most efficient operations with
111 pseudo registers. */
112
113struct processor_costs {
114 /* Costs used by register allocator. integer->integer register move
115 cost is 2. */
116 struct
117 {
118 const int movzbl_load; /* cost of loading using movzbl */
119 const int int_load[3]; /* cost of loading integer registers
120 in QImode, HImode and SImode relative
121 to reg-reg move (2). */
122 const int int_store[3]; /* cost of storing integer register
123 in QImode, HImode and SImode */
124 const int fp_move; /* cost of reg,reg fld/fst */
125 const int fp_load[3]; /* cost of loading FP register
126 in SFmode, DFmode and XFmode */
127 const int fp_store[3]; /* cost of storing FP register
128 in SFmode, DFmode and XFmode */
129 const int mmx_move; /* cost of moving MMX register. */
130 const int mmx_load[2]; /* cost of loading MMX register
131 in SImode and DImode */
132 const int mmx_store[2]; /* cost of storing MMX register
133 in SImode and DImode */
134 const int xmm_move; /* cost of moving XMM register. */
135 const int ymm_move; /* cost of moving XMM register. */
136 const int zmm_move; /* cost of moving XMM register. */
137 const int sse_load[5]; /* cost of loading SSE register
138 in 32bit, 64bit, 128bit, 256bit and 512bit */
139 const int sse_store[5]; /* cost of storing SSE register
140 in SImode, DImode and TImode. */
141 const int sse_to_integer; /* cost of moving SSE register to integer. */
142 const int integer_to_sse; /* cost of moving integer register to SSE. */
143 const int mask_to_integer; /* cost of moving mask register to integer. */
144 const int integer_to_mask; /* cost of moving integer register to mask. */
145 const int mask_load[3]; /* cost of loading mask registers
146 in QImode, HImode and SImode. */
147 const int mask_store[3]; /* cost of storing mask register
148 in QImode, HImode and SImode. */
149 const int mask_move; /* cost of moving mask register. */
150 } hard_register;
151
152 const int add; /* cost of an add instruction */
153 const int lea; /* cost of a lea instruction */
154 const int shift_var; /* variable shift costs */
155 const int shift_const; /* constant shift costs */
156 const int mult_init[5]; /* cost of starting a multiply
157 in QImode, HImode, SImode, DImode, TImode*/
158 const int mult_bit; /* cost of multiply per each bit set */
159 const int divide[5]; /* cost of a divide/mod
160 in QImode, HImode, SImode, DImode, TImode*/
161 int movsx; /* The cost of movsx operation. */
162 int movzx; /* The cost of movzx operation. */
163 const int large_insn; /* insns larger than this cost more */
164 const int move_ratio; /* The threshold of number of scalar
165 memory-to-memory move insns. */
166 const int clear_ratio; /* The threshold of number of scalar
167 memory clearing insns. */
168 const int int_load[3]; /* cost of loading integer registers
169 in QImode, HImode and SImode relative
170 to reg-reg move (2). */
171 const int int_store[3]; /* cost of storing integer register
172 in QImode, HImode and SImode */
173 const int sse_load[5]; /* cost of loading SSE register
174 in 32bit, 64bit, 128bit, 256bit and 512bit */
175 const int sse_store[5]; /* cost of storing SSE register
176 in 32bit, 64bit, 128bit, 256bit and 512bit */
177 const int sse_unaligned_load[5];/* cost of unaligned load. */
178 const int sse_unaligned_store[5];/* cost of unaligned store. */
179 const int xmm_move, ymm_move, /* cost of moving XMM and YMM register. */
180 zmm_move;
181 const int sse_to_integer; /* cost of moving SSE register to integer. */
182 const int integer_to_sse; /* cost of moving integer register to SSE. */
183 const int gather_static, gather_per_elt; /* Cost of gather load is computed
184 as static + per_item * nelts. */
185 const int scatter_static, scatter_per_elt; /* Cost of gather store is
186 computed as static + per_item * nelts. */
187 const int l1_cache_size; /* size of l1 cache, in kilobytes. */
188 const int l2_cache_size; /* size of l2 cache, in kilobytes. */
189 const int prefetch_block; /* bytes moved to cache for prefetch. */
190 const int simultaneous_prefetches; /* number of parallel prefetch
191 operations. */
192 const int branch_cost; /* Default value for BRANCH_COST. */
193 const int fadd; /* cost of FADD and FSUB instructions. */
194 const int fmul; /* cost of FMUL instruction. */
195 const int fdiv; /* cost of FDIV instruction. */
196 const int fabs; /* cost of FABS instruction. */
197 const int fchs; /* cost of FCHS instruction. */
198 const int fsqrt; /* cost of FSQRT instruction. */
199 /* Specify what algorithm
200 to use for stringops on unknown size. */
201 const int sse_op; /* cost of cheap SSE instruction. */
202 const int addss; /* cost of ADDSS/SD SUBSS/SD instructions. */
203 const int mulss; /* cost of MULSS instructions. */
204 const int mulsd; /* cost of MULSD instructions. */
205 const int fmass; /* cost of FMASS instructions. */
206 const int fmasd; /* cost of FMASD instructions. */
207 const int divss; /* cost of DIVSS instructions. */
208 const int divsd; /* cost of DIVSD instructions. */
209 const int sqrtss; /* cost of SQRTSS instructions. */
210 const int sqrtsd; /* cost of SQRTSD instructions. */
211 const int cvtss2sd; /* cost SSE FP conversions,
212 such as CVTSS2SD. */
213 const int vcvtps2pd256; /* cost 256bit packed FP conversions,
214 such as VCVTPD2PS with larger reg in ymm. */
215 const int vcvtps2pd512; /* cost 512bit packed FP conversions,
216 such as VCVTPD2PS with larger reg in zmm. */
217 const int cvtsi2ss; /* cost of CVTSI2SS instruction. */
218 const int cvtss2si; /* cost of CVT(T)SS2SI instruction. */
219 const int cvtpi2ps; /* cost of CVTPI2PS instruction. */
220 const int cvtps2pi; /* cost of CVT(T)PS2PI instruction. */
221 const int reassoc_int, reassoc_fp, reassoc_vec_int, reassoc_vec_fp;
222 /* Specify reassociation width for integer,
223 fp, vector integer and vector fp
224 operations. Generally should correspond
225 to number of instructions executed in
226 parallel. See also
227 ix86_reassociation_width. */
228 struct stringop_algs *memcpy, *memset;
229 const int cond_taken_branch_cost; /* Cost of taken branch for vectorizer
230 cost model. */
231 const int cond_not_taken_branch_cost;/* Cost of not taken branch for
232 vectorizer cost model. */
233
234 /* The "0:0:8" label alignment specified for some processors generates
235 secondary 8-byte alignment only for those label/jump/loop targets
236 which have primary alignment. */
237 const char *const align_loop; /* Loop alignment. */
238 const char *const align_jump; /* Jump alignment. */
239 const char *const align_label; /* Label alignment. */
240 const char *const align_func; /* Function alignment. */
241
242 const unsigned small_unroll_ninsns; /* Insn count limit for small loop
243 to be unrolled. */
244 const unsigned small_unroll_factor; /* Unroll factor for small loop to
245 be unrolled. */
246 const int br_mispredict_scale; /* Branch mispredict scale for ifcvt
247 threshold. */
248};
249
250extern const struct processor_costs *ix86_cost;
251extern const struct processor_costs ix86_size_cost;
252
253#define ix86_cur_cost() \
254 (optimize_insn_for_size_p () ? &ix86_size_cost : ix86_cost)
255
256/* Macros used in the machine description to test the flags. */
257
258/* configure can arrange to change it. */
259
260#ifndef TARGET_CPU_DEFAULT
261#define TARGET_CPU_DEFAULT PROCESSOR_GENERIC
262#endif
263
264#ifndef TARGET_FPMATH_DEFAULT
265#define TARGET_FPMATH_DEFAULT \
266 (TARGET_64BIT && TARGET_SSE ? FPMATH_SSE : FPMATH_387)
267#endif
268
269#ifndef TARGET_FPMATH_DEFAULT_P
270#define TARGET_FPMATH_DEFAULT_P(x) \
271 (TARGET_64BIT_P(x) && TARGET_SSE_P(x) ? FPMATH_SSE : FPMATH_387)
272#endif
273
274/* If the i387 is disabled or -miamcu is used , then do not return
275 values in it. */
276#define TARGET_FLOAT_RETURNS_IN_80387 \
277 (TARGET_FLOAT_RETURNS && TARGET_80387 && !TARGET_IAMCU)
278#define TARGET_FLOAT_RETURNS_IN_80387_P(x) \
279 (TARGET_FLOAT_RETURNS_P(x) && TARGET_80387_P(x) && !TARGET_IAMCU_P(x))
280
281/* 64bit Sledgehammer mode. For libgcc2 we make sure this is a
282 compile-time constant. */
283#ifdef IN_LIBGCC2
284#undef TARGET_64BIT
285#ifdef __x86_64__
286#define TARGET_64BIT 1
287#else
288#define TARGET_64BIT 0
289#endif
290#else
291#ifndef TARGET_BI_ARCH
292#undef TARGET_64BIT
293#undef TARGET_64BIT_P
294#if TARGET_64BIT_DEFAULT
295#define TARGET_64BIT 1
296#define TARGET_64BIT_P(x) 1
297#else
298#define TARGET_64BIT 0
299#define TARGET_64BIT_P(x) 0
300#endif
301#endif
302#endif
303
304#define HAS_LONG_COND_BRANCH 1
305#define HAS_LONG_UNCOND_BRANCH 1
306
307#define TARGET_CPU_P(CPU) (ix86_tune == PROCESSOR_ ## CPU)
308
309/* Feature tests against the various tunings. */
310enum ix86_tune_indices {
311#undef DEF_TUNE
312#define DEF_TUNE(tune, name, selector) tune,
313#include "x86-tune.def"
314#undef DEF_TUNE
315X86_TUNE_LAST
316};
317
318extern unsigned char ix86_tune_features[X86_TUNE_LAST];
319
320#define TARGET_USE_LEAVE ix86_tune_features[X86_TUNE_USE_LEAVE]
321#define TARGET_PUSH_MEMORY ix86_tune_features[X86_TUNE_PUSH_MEMORY]
322#define TARGET_ZERO_EXTEND_WITH_AND \
323 ix86_tune_features[X86_TUNE_ZERO_EXTEND_WITH_AND]
324#define TARGET_UNROLL_STRLEN ix86_tune_features[X86_TUNE_UNROLL_STRLEN]
325#define TARGET_BRANCH_PREDICTION_HINTS_NOT_TAKEN \
326 ix86_tune_features[X86_TUNE_BRANCH_PREDICTION_HINTS_NOT_TAKEN]
327#define TARGET_BRANCH_PREDICTION_HINTS_TAKEN \
328 ix86_tune_features[X86_TUNE_BRANCH_PREDICTION_HINTS_TAKEN]
329#define TARGET_DOUBLE_WITH_ADD ix86_tune_features[X86_TUNE_DOUBLE_WITH_ADD]
330#define TARGET_USE_SAHF ix86_tune_features[X86_TUNE_USE_SAHF]
331#define TARGET_MOVX ix86_tune_features[X86_TUNE_MOVX]
332#define TARGET_PARTIAL_REG_STALL ix86_tune_features[X86_TUNE_PARTIAL_REG_STALL]
333#define TARGET_PARTIAL_MEMORY_READ_STALL \
334 ix86_tune_features[X86_TUNE_PARTIAL_MEMORY_READ_STALL]
335#define TARGET_PARTIAL_FLAG_REG_STALL \
336 ix86_tune_features[X86_TUNE_PARTIAL_FLAG_REG_STALL]
337#define TARGET_LCP_STALL \
338 ix86_tune_features[X86_TUNE_LCP_STALL]
339#define TARGET_USE_HIMODE_FIOP ix86_tune_features[X86_TUNE_USE_HIMODE_FIOP]
340#define TARGET_USE_SIMODE_FIOP ix86_tune_features[X86_TUNE_USE_SIMODE_FIOP]
341#define TARGET_USE_MOV0 ix86_tune_features[X86_TUNE_USE_MOV0]
342#define TARGET_USE_CLTD ix86_tune_features[X86_TUNE_USE_CLTD]
343#define TARGET_USE_XCHGB ix86_tune_features[X86_TUNE_USE_XCHGB]
344#define TARGET_SPLIT_LONG_MOVES ix86_tune_features[X86_TUNE_SPLIT_LONG_MOVES]
345#define TARGET_READ_MODIFY_WRITE ix86_tune_features[X86_TUNE_READ_MODIFY_WRITE]
346#define TARGET_READ_MODIFY ix86_tune_features[X86_TUNE_READ_MODIFY]
347#define TARGET_PROMOTE_QImode ix86_tune_features[X86_TUNE_PROMOTE_QIMODE]
348#define TARGET_FAST_PREFIX ix86_tune_features[X86_TUNE_FAST_PREFIX]
349#define TARGET_SINGLE_STRINGOP ix86_tune_features[X86_TUNE_SINGLE_STRINGOP]
350#define TARGET_PREFER_KNOWN_REP_MOVSB_STOSB \
351 ix86_tune_features[X86_TUNE_PREFER_KNOWN_REP_MOVSB_STOSB]
352#define TARGET_MISALIGNED_MOVE_STRING_PRO_EPILOGUES \
353 ix86_tune_features[X86_TUNE_MISALIGNED_MOVE_STRING_PRO_EPILOGUES]
354#define TARGET_QIMODE_MATH ix86_tune_features[X86_TUNE_QIMODE_MATH]
355#define TARGET_HIMODE_MATH ix86_tune_features[X86_TUNE_HIMODE_MATH]
356#define TARGET_PROMOTE_QI_REGS ix86_tune_features[X86_TUNE_PROMOTE_QI_REGS]
357#define TARGET_PROMOTE_HI_REGS ix86_tune_features[X86_TUNE_PROMOTE_HI_REGS]
358#define TARGET_SINGLE_POP ix86_tune_features[X86_TUNE_SINGLE_POP]
359#define TARGET_DOUBLE_POP ix86_tune_features[X86_TUNE_DOUBLE_POP]
360#define TARGET_SINGLE_PUSH ix86_tune_features[X86_TUNE_SINGLE_PUSH]
361#define TARGET_DOUBLE_PUSH ix86_tune_features[X86_TUNE_DOUBLE_PUSH]
362#define TARGET_INTEGER_DFMODE_MOVES \
363 ix86_tune_features[X86_TUNE_INTEGER_DFMODE_MOVES]
364#define TARGET_PARTIAL_REG_DEPENDENCY \
365 ix86_tune_features[X86_TUNE_PARTIAL_REG_DEPENDENCY]
366#define TARGET_SSE_PARTIAL_REG_DEPENDENCY \
367 ix86_tune_features[X86_TUNE_SSE_PARTIAL_REG_DEPENDENCY]
368#define TARGET_SSE_PARTIAL_REG_FP_CONVERTS_DEPENDENCY \
369 ix86_tune_features[X86_TUNE_SSE_PARTIAL_REG_FP_CONVERTS_DEPENDENCY]
370#define TARGET_SSE_PARTIAL_REG_CONVERTS_DEPENDENCY \
371 ix86_tune_features[X86_TUNE_SSE_PARTIAL_REG_CONVERTS_DEPENDENCY]
372#define TARGET_SSE_UNALIGNED_LOAD_OPTIMAL \
373 ix86_tune_features[X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL]
374#define TARGET_SSE_UNALIGNED_STORE_OPTIMAL \
375 ix86_tune_features[X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL]
376#define TARGET_SSE_PACKED_SINGLE_INSN_OPTIMAL \
377 ix86_tune_features[X86_TUNE_SSE_PACKED_SINGLE_INSN_OPTIMAL]
378#define TARGET_SSE_SPLIT_REGS ix86_tune_features[X86_TUNE_SSE_SPLIT_REGS]
379#define TARGET_SSE_TYPELESS_STORES \
380 ix86_tune_features[X86_TUNE_SSE_TYPELESS_STORES]
381#define TARGET_SSE_LOAD0_BY_PXOR ix86_tune_features[X86_TUNE_SSE_LOAD0_BY_PXOR]
382#define TARGET_MEMORY_MISMATCH_STALL \
383 ix86_tune_features[X86_TUNE_MEMORY_MISMATCH_STALL]
384#define TARGET_PROLOGUE_USING_MOVE \
385 ix86_tune_features[X86_TUNE_PROLOGUE_USING_MOVE]
386#define TARGET_EPILOGUE_USING_MOVE \
387 ix86_tune_features[X86_TUNE_EPILOGUE_USING_MOVE]
388#define TARGET_SHIFT1 ix86_tune_features[X86_TUNE_SHIFT1]
389#define TARGET_USE_FFREEP ix86_tune_features[X86_TUNE_USE_FFREEP]
390#define TARGET_INTER_UNIT_MOVES_TO_VEC \
391 ix86_tune_features[X86_TUNE_INTER_UNIT_MOVES_TO_VEC]
392#define TARGET_INTER_UNIT_MOVES_FROM_VEC \
393 ix86_tune_features[X86_TUNE_INTER_UNIT_MOVES_FROM_VEC]
394#define TARGET_INTER_UNIT_CONVERSIONS \
395 ix86_tune_features[X86_TUNE_INTER_UNIT_CONVERSIONS]
396#define TARGET_FOUR_JUMP_LIMIT ix86_tune_features[X86_TUNE_FOUR_JUMP_LIMIT]
397#define TARGET_SCHEDULE ix86_tune_features[X86_TUNE_SCHEDULE]
398#define TARGET_USE_BT ix86_tune_features[X86_TUNE_USE_BT]
399#define TARGET_USE_INCDEC ix86_tune_features[X86_TUNE_USE_INCDEC]
400#define TARGET_PAD_RETURNS ix86_tune_features[X86_TUNE_PAD_RETURNS]
401#define TARGET_PAD_SHORT_FUNCTION \
402 ix86_tune_features[X86_TUNE_PAD_SHORT_FUNCTION]
403#define TARGET_EXT_80387_CONSTANTS \
404 ix86_tune_features[X86_TUNE_EXT_80387_CONSTANTS]
405#define TARGET_AVOID_VECTOR_DECODE \
406 ix86_tune_features[X86_TUNE_AVOID_VECTOR_DECODE]
407#define TARGET_TUNE_PROMOTE_HIMODE_IMUL \
408 ix86_tune_features[X86_TUNE_PROMOTE_HIMODE_IMUL]
409#define TARGET_SLOW_IMUL_IMM32_MEM \
410 ix86_tune_features[X86_TUNE_SLOW_IMUL_IMM32_MEM]
411#define TARGET_SLOW_IMUL_IMM8 ix86_tune_features[X86_TUNE_SLOW_IMUL_IMM8]
412#define TARGET_MOVE_M1_VIA_OR ix86_tune_features[X86_TUNE_MOVE_M1_VIA_OR]
413#define TARGET_NOT_UNPAIRABLE ix86_tune_features[X86_TUNE_NOT_UNPAIRABLE]
414#define TARGET_NOT_VECTORMODE ix86_tune_features[X86_TUNE_NOT_VECTORMODE]
415#define TARGET_USE_VECTOR_FP_CONVERTS \
416 ix86_tune_features[X86_TUNE_USE_VECTOR_FP_CONVERTS]
417#define TARGET_USE_VECTOR_CONVERTS \
418 ix86_tune_features[X86_TUNE_USE_VECTOR_CONVERTS]
419#define TARGET_SLOW_PSHUFB \
420 ix86_tune_features[X86_TUNE_SLOW_PSHUFB]
421#define TARGET_AVOID_4BYTE_PREFIXES \
422 ix86_tune_features[X86_TUNE_AVOID_4BYTE_PREFIXES]
423#define TARGET_USE_GATHER_2PARTS \
424 ix86_tune_features[X86_TUNE_USE_GATHER_2PARTS]
425#define TARGET_USE_SCATTER_2PARTS \
426 ix86_tune_features[X86_TUNE_USE_SCATTER_2PARTS]
427#define TARGET_USE_GATHER_4PARTS \
428 ix86_tune_features[X86_TUNE_USE_GATHER_4PARTS]
429#define TARGET_USE_SCATTER_4PARTS \
430 ix86_tune_features[X86_TUNE_USE_SCATTER_4PARTS]
431#define TARGET_USE_GATHER_8PARTS \
432 ix86_tune_features[X86_TUNE_USE_GATHER_8PARTS]
433#define TARGET_USE_SCATTER_8PARTS \
434 ix86_tune_features[X86_TUNE_USE_SCATTER_8PARTS]
435#define TARGET_FUSE_CMP_AND_BRANCH_32 \
436 ix86_tune_features[X86_TUNE_FUSE_CMP_AND_BRANCH_32]
437#define TARGET_FUSE_CMP_AND_BRANCH_64 \
438 ix86_tune_features[X86_TUNE_FUSE_CMP_AND_BRANCH_64]
439#define TARGET_FUSE_CMP_AND_BRANCH \
440 (TARGET_64BIT ? TARGET_FUSE_CMP_AND_BRANCH_64 \
441 : TARGET_FUSE_CMP_AND_BRANCH_32)
442#define TARGET_FUSE_CMP_AND_BRANCH_SOFLAGS \
443 ix86_tune_features[X86_TUNE_FUSE_CMP_AND_BRANCH_SOFLAGS]
444#define TARGET_FUSE_ALU_AND_BRANCH \
445 ix86_tune_features[X86_TUNE_FUSE_ALU_AND_BRANCH]
446#define TARGET_FUSE_ALU_AND_BRANCH_MEM \
447 ix86_tune_features[X86_TUNE_FUSE_ALU_AND_BRANCH_MEM]
448#define TARGET_FUSE_ALU_AND_BRANCH_MEM_IMM \
449 ix86_tune_features[X86_TUNE_FUSE_ALU_AND_BRANCH_MEM_IMM]
450#define TARGET_FUSE_ALU_AND_BRANCH_RIP_RELATIVE\
451 ix86_tune_features[X86_TUNE_FUSE_ALU_AND_BRANCH_RIP_RELATIVE]
452#define TARGET_FUSE_MOV_AND_ALU \
453 ix86_tune_features[X86_TUNE_FUSE_MOV_AND_ALU]
454#define TARGET_OPT_AGU ix86_tune_features[X86_TUNE_OPT_AGU]
455#define TARGET_AVOID_LEA_FOR_ADDR \
456 ix86_tune_features[X86_TUNE_AVOID_LEA_FOR_ADDR]
457#define TARGET_SOFTWARE_PREFETCHING_BENEFICIAL \
458 ix86_tune_features[X86_TUNE_SOFTWARE_PREFETCHING_BENEFICIAL]
459#define TARGET_AVX256_SPLIT_REGS \
460 ix86_tune_features[X86_TUNE_AVX256_SPLIT_REGS]
461#define TARGET_AVX256_AVOID_VEC_PERM \
462 ix86_tune_features[X86_TUNE_AVX256_AVOID_VEC_PERM]
463#define TARGET_AVX512_SPLIT_REGS \
464 ix86_tune_features[X86_TUNE_AVX512_SPLIT_REGS]
465#define TARGET_GENERAL_REGS_SSE_SPILL \
466 ix86_tune_features[X86_TUNE_GENERAL_REGS_SSE_SPILL]
467#define TARGET_AVOID_MEM_OPND_FOR_CMOVE \
468 ix86_tune_features[X86_TUNE_AVOID_MEM_OPND_FOR_CMOVE]
469#define TARGET_SPLIT_MEM_OPND_FOR_FP_CONVERTS \
470 ix86_tune_features[X86_TUNE_SPLIT_MEM_OPND_FOR_FP_CONVERTS]
471#define TARGET_ADJUST_UNROLL \
472 ix86_tune_features[X86_TUNE_ADJUST_UNROLL]
473#define TARGET_AVOID_FALSE_DEP_FOR_BMI \
474 ix86_tune_features[X86_TUNE_AVOID_FALSE_DEP_FOR_BMI]
475#define TARGET_AVOID_FALSE_DEP_FOR_TZCNT \
476 ix86_tune_features[X86_TUNE_AVOID_FALSE_DEP_FOR_TZCNT]
477#define TARGET_AVOID_FALSE_DEP_FOR_BLS \
478 ix86_tune_features[X86_TUNE_AVOID_FALSE_DEP_FOR_BLS]
479#define TARGET_ONE_IF_CONV_INSN \
480 ix86_tune_features[X86_TUNE_ONE_IF_CONV_INSN]
481#define TARGET_AVOID_MFENCE ix86_tune_features[X86_TUNE_AVOID_MFENCE]
482#define TARGET_EXPAND_ABS \
483 ix86_tune_features[X86_TUNE_EXPAND_ABS]
484#define TARGET_V2DF_REDUCTION_PREFER_HADDPD \
485 ix86_tune_features[X86_TUNE_V2DF_REDUCTION_PREFER_HADDPD]
486#define TARGET_DEST_FALSE_DEP_FOR_GLC \
487 ix86_tune_features[X86_TUNE_DEST_FALSE_DEP_FOR_GLC]
488#define TARGET_SLOW_STC ix86_tune_features[X86_TUNE_SLOW_STC]
489#define TARGET_USE_RCR ix86_tune_features[X86_TUNE_USE_RCR]
490#define TARGET_SSE_MOVCC_USE_BLENDV \
491 ix86_tune_features[X86_TUNE_SSE_MOVCC_USE_BLENDV]
492#define TARGET_ALIGN_TIGHT_LOOPS \
493 ix86_tune_features[X86_TUNE_ALIGN_TIGHT_LOOPS]
494#define TARGET_SSE_REDUCTION_PREFER_PSHUF \
495 ix86_tune_features[X86_TUNE_SSE_REDUCTION_PREFER_PSHUF]
496
497
498/* Feature tests against the various architecture variations. */
499enum ix86_arch_indices {
500 X86_ARCH_CMOV,
501 X86_ARCH_CMPXCHG,
502 X86_ARCH_CMPXCHG8B,
503 X86_ARCH_XADD,
504 X86_ARCH_BSWAP,
505
506 X86_ARCH_LAST
507};
508
509extern unsigned char ix86_arch_features[X86_ARCH_LAST];
510
511#define TARGET_CMOV ix86_arch_features[X86_ARCH_CMOV]
512#define TARGET_CMPXCHG ix86_arch_features[X86_ARCH_CMPXCHG]
513#define TARGET_CMPXCHG8B ix86_arch_features[X86_ARCH_CMPXCHG8B]
514#define TARGET_XADD ix86_arch_features[X86_ARCH_XADD]
515#define TARGET_BSWAP ix86_arch_features[X86_ARCH_BSWAP]
516
517/* For sane SSE instruction set generation we need fcomi instruction.
518 It is safe to enable all CMOVE instructions. Also, RDRAND intrinsic
519 expands to a sequence that includes conditional move. */
520#define TARGET_CMOVE (TARGET_CMOV || TARGET_SSE || TARGET_RDRND)
521
522#define TARGET_FISTTP (TARGET_SSE3 && TARGET_80387)
523
524extern unsigned char ix86_prefetch_sse;
525#define TARGET_PREFETCH_SSE ix86_prefetch_sse
526
527#define ASSEMBLER_DIALECT (ix86_asm_dialect)
528
529#define TARGET_SSE_MATH ((ix86_fpmath & FPMATH_SSE) != 0)
530#define TARGET_MIX_SSE_I387 \
531 ((ix86_fpmath & (FPMATH_SSE | FPMATH_387)) == (FPMATH_SSE | FPMATH_387))
532
533#define TARGET_HARD_SF_REGS (TARGET_80387 || TARGET_MMX || TARGET_SSE)
534#define TARGET_HARD_DF_REGS (TARGET_80387 || TARGET_SSE)
535#define TARGET_HARD_XF_REGS (TARGET_80387)
536
537#define TARGET_GNU_TLS (ix86_tls_dialect == TLS_DIALECT_GNU)
538#define TARGET_GNU2_TLS (ix86_tls_dialect == TLS_DIALECT_GNU2)
539#define TARGET_ANY_GNU_TLS (TARGET_GNU_TLS || TARGET_GNU2_TLS)
540#define TARGET_SUN_TLS 0
541#define TARGET_WIN32_TLS 0
542
543#ifndef TARGET_64BIT_DEFAULT
544#define TARGET_64BIT_DEFAULT 0
545#endif
546#ifndef TARGET_TLS_DIRECT_SEG_REFS_DEFAULT
547#define TARGET_TLS_DIRECT_SEG_REFS_DEFAULT 0
548#endif
549
550#define TARGET_SSP_GLOBAL_GUARD (ix86_stack_protector_guard == SSP_GLOBAL)
551#define TARGET_SSP_TLS_GUARD (ix86_stack_protector_guard == SSP_TLS)
552
553/* Fence to use after loop using storent. */
554
555extern GTY(()) tree x86_mfence;
556#define FENCE_FOLLOWING_MOVNT x86_mfence
557
558/* Once GDB has been enhanced to deal with functions without frame
559 pointers, we can change this to allow for elimination of
560 the frame pointer in leaf functions. */
561#define TARGET_DEFAULT 0
562
563/* Extra bits to force. */
564#define TARGET_SUBTARGET_DEFAULT 0
565#define TARGET_SUBTARGET_ISA_DEFAULT 0
566
567/* Extra bits to force on w/ 32-bit mode. */
568#define TARGET_SUBTARGET32_DEFAULT 0
569#define TARGET_SUBTARGET32_ISA_DEFAULT 0
570
571/* Extra bits to force on w/ 64-bit mode. */
572#define TARGET_SUBTARGET64_DEFAULT 0
573/* Enable MMX, SSE and SSE2 by default. */
574#define TARGET_SUBTARGET64_ISA_DEFAULT \
575 (OPTION_MASK_ISA_MMX | OPTION_MASK_ISA_SSE | OPTION_MASK_ISA_SSE2)
576
577/* Replace MACH-O, ifdefs by in-line tests, where possible.
578 (a) Macros defined in config/i386/darwin.h */
579#define TARGET_MACHO 0
580#define TARGET_MACHO_SYMBOL_STUBS 0
581#define MACHOPIC_ATT_STUB 0
582/* (b) Macros defined in config/darwin.h */
583#define MACHO_DYNAMIC_NO_PIC_P 0
584#define MACHOPIC_INDIRECT 0
585#define MACHOPIC_PURE 0
586
587/* For the RDOS */
588#define TARGET_RDOS 0
589
590/* For the Windows 64-bit ABI. */
591#define TARGET_64BIT_MS_ABI (TARGET_64BIT && ix86_cfun_abi () == MS_ABI)
592
593/* For the Windows 32-bit ABI. */
594#define TARGET_32BIT_MS_ABI (!TARGET_64BIT && ix86_cfun_abi () == MS_ABI)
595
596/* This is re-defined by cygming.h. */
597#define TARGET_SEH 0
598
599/* The default abi used by target. */
600#define DEFAULT_ABI SYSV_ABI
601
602/* The default TLS segment register used by target. */
603#define DEFAULT_TLS_SEG_REG \
604 (TARGET_64BIT ? ADDR_SPACE_SEG_FS : ADDR_SPACE_SEG_GS)
605
606/* Subtargets may reset this to 1 in order to enable 96-bit long double
607 with the rounding mode forced to 53 bits. */
608#define TARGET_96_ROUND_53_LONG_DOUBLE 0
609
610#ifndef SUBTARGET_DRIVER_SELF_SPECS
611# define SUBTARGET_DRIVER_SELF_SPECS ""
612#endif
613
614#define DRIVER_SELF_SPECS SUBTARGET_DRIVER_SELF_SPECS
615
616/* -march=native handling only makes sense with compiler running on
617 an x86 or x86_64 chip. If changing this condition, also change
618 the condition in driver-i386.cc. */
619#if defined(__i386__) || defined(__x86_64__)
620/* In driver-i386.cc. */
621extern const char *host_detect_local_cpu (int argc, const char **argv);
622#define EXTRA_SPEC_FUNCTIONS \
623 { "local_cpu_detect", host_detect_local_cpu },
624#define HAVE_LOCAL_CPU_DETECT
625#endif
626
627#if TARGET_64BIT_DEFAULT
628#define OPT_ARCH64 "!m32"
629#define OPT_ARCH32 "m32"
630#else
631#define OPT_ARCH64 "m64|mx32"
632#define OPT_ARCH32 "m64|mx32:;"
633#endif
634
635/* Support for configure-time defaults of some command line options.
636 The order here is important so that -march doesn't squash the
637 tune or cpu values. */
638#define OPTION_DEFAULT_SPECS \
639 {"tune", "%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}" }, \
640 {"tune_32", "%{" OPT_ARCH32 ":%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}}" }, \
641 {"tune_64", "%{" OPT_ARCH64 ":%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}}" }, \
642 {"cpu", "%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}" }, \
643 {"cpu_32", "%{" OPT_ARCH32 ":%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}}" }, \
644 {"cpu_64", "%{" OPT_ARCH64 ":%{!mtune=*:%{!mcpu=*:%{!march=*:-mtune=%(VALUE)}}}}" }, \
645 {"arch", "%{!march=*:-march=%(VALUE)}"}, \
646 {"arch_32", "%{" OPT_ARCH32 ":%{!march=*:-march=%(VALUE)}}"}, \
647 {"arch_64", "%{" OPT_ARCH64 ":%{!march=*:-march=%(VALUE)}}"},
648
649/* Specs for the compiler proper */
650
651#ifndef CC1_CPU_SPEC
652#define CC1_CPU_SPEC_1 ""
653
654#ifndef HAVE_LOCAL_CPU_DETECT
655#define CC1_CPU_SPEC CC1_CPU_SPEC_1
656#else
657#define ARCH_ARG "%{" OPT_ARCH64 ":64;:32}"
658#define CC1_CPU_SPEC CC1_CPU_SPEC_1 \
659"%{march=native:%>march=native %:local_cpu_detect(arch " ARCH_ARG ") \
660 %{!mtune=*:%>mtune=native %:local_cpu_detect(tune " ARCH_ARG ")}} \
661%{mtune=native:%>mtune=native %:local_cpu_detect(tune " ARCH_ARG ")}"
662#endif
663#endif
664
665/* Target CPU builtins. */
666#define TARGET_CPU_CPP_BUILTINS() ix86_target_macros ()
667
668/* Target Pragmas. */
669#define REGISTER_TARGET_PRAGMAS() ix86_register_pragmas ()
670
671#ifndef CC1_SPEC
672#define CC1_SPEC "%(cc1_cpu) "
673#endif
674
675/* This macro defines names of additional specifications to put in the
676 specs that can be used in various specifications like CC1_SPEC. Its
677 definition is an initializer with a subgrouping for each command option.
678
679 Each subgrouping contains a string constant, that defines the
680 specification name, and a string constant that used by the GCC driver
681 program.
682
683 Do not define this macro if it does not need to do anything. */
684
685#ifndef SUBTARGET_EXTRA_SPECS
686#define SUBTARGET_EXTRA_SPECS
687#endif
688
689#define EXTRA_SPECS \
690 { "cc1_cpu", CC1_CPU_SPEC }, \
691 SUBTARGET_EXTRA_SPECS
692
693
694/* Whether to allow x87 floating-point arithmetic on MODE (one of
695 SFmode, DFmode and XFmode) in the current excess precision
696 configuration. */
697#define X87_ENABLE_ARITH(MODE) \
698 (ix86_unsafe_math_optimizations \
699 || ix86_excess_precision == EXCESS_PRECISION_FAST \
700 || (MODE) == XFmode)
701
702/* Likewise, whether to allow direct conversions from integer mode
703 IMODE (HImode, SImode or DImode) to MODE. */
704#define X87_ENABLE_FLOAT(MODE, IMODE) \
705 (ix86_unsafe_math_optimizations \
706 || ix86_excess_precision == EXCESS_PRECISION_FAST \
707 || (MODE) == XFmode \
708 || ((MODE) == DFmode && (IMODE) == SImode) \
709 || (IMODE) == HImode)
710
711/* target machine storage layout */
712
713#define SHORT_TYPE_SIZE 16
714#define INT_TYPE_SIZE 32
715#define LONG_TYPE_SIZE (TARGET_X32 ? 32 : BITS_PER_WORD)
716#define POINTER_SIZE (TARGET_X32 ? 32 : BITS_PER_WORD)
717#define LONG_LONG_TYPE_SIZE 64
718
719#define WIDEST_HARDWARE_FP_SIZE 80
720
721#if defined (TARGET_BI_ARCH) || TARGET_64BIT_DEFAULT
722#define MAX_BITS_PER_WORD 64
723#else
724#define MAX_BITS_PER_WORD 32
725#endif
726
727/* Define this if most significant byte of a word is the lowest numbered. */
728/* That is true on the 80386. */
729
730#define BITS_BIG_ENDIAN 0
731
732/* Define this if most significant byte of a word is the lowest numbered. */
733/* That is not true on the 80386. */
734#define BYTES_BIG_ENDIAN 0
735
736/* Define this if most significant word of a multiword number is the lowest
737 numbered. */
738/* Not true for 80386 */
739#define WORDS_BIG_ENDIAN 0
740
741/* Width of a word, in units (bytes). */
742#define UNITS_PER_WORD (TARGET_64BIT ? 8 : 4)
743
744#ifndef IN_LIBGCC2
745#define MIN_UNITS_PER_WORD 4
746#endif
747
748/* Allocation boundary (in *bits*) for storing arguments in argument list. */
749#define PARM_BOUNDARY BITS_PER_WORD
750
751/* Boundary (in *bits*) on which stack pointer should be aligned. */
752#define STACK_BOUNDARY (TARGET_64BIT_MS_ABI ? 128 : BITS_PER_WORD)
753
754/* Stack boundary of the main function guaranteed by OS. */
755#define MAIN_STACK_BOUNDARY (TARGET_64BIT ? 128 : 32)
756
757/* Minimum stack boundary. */
758#define MIN_STACK_BOUNDARY BITS_PER_WORD
759
760/* Boundary (in *bits*) on which the stack pointer prefers to be
761 aligned; the compiler cannot rely on having this alignment. */
762#define PREFERRED_STACK_BOUNDARY ix86_preferred_stack_boundary
763
764/* It should be MIN_STACK_BOUNDARY. But we set it to 128 bits for
765 both 32bit and 64bit, to support codes that need 128 bit stack
766 alignment for SSE instructions, but can't realign the stack. */
767#define PREFERRED_STACK_BOUNDARY_DEFAULT \
768 (TARGET_IAMCU ? MIN_STACK_BOUNDARY : 128)
769
770/* 1 if -mstackrealign should be turned on by default. It will
771 generate an alternate prologue and epilogue that realigns the
772 runtime stack if nessary. This supports mixing codes that keep a
773 4-byte aligned stack, as specified by i386 psABI, with codes that
774 need a 16-byte aligned stack, as required by SSE instructions. */
775#define STACK_REALIGN_DEFAULT 0
776
777/* Boundary (in *bits*) on which the incoming stack is aligned. */
778#define INCOMING_STACK_BOUNDARY ix86_incoming_stack_boundary
779
780/* According to Windows x64 software convention, the maximum stack allocatable
781 in the prologue is 4G - 8 bytes. Furthermore, there is a limited set of
782 instructions allowed to adjust the stack pointer in the epilog, forcing the
783 use of frame pointer for frames larger than 2 GB. This theorical limit
784 is reduced by 256, an over-estimated upper bound for the stack use by the
785 prologue.
786 We define only one threshold for both the prolog and the epilog. When the
787 frame size is larger than this threshold, we allocate the area to save SSE
788 regs, then save them, and then allocate the remaining. There is no SEH
789 unwind info for this later allocation. */
790#define SEH_MAX_FRAME_SIZE ((2U << 30) - 256)
791
792/* Target OS keeps a vector-aligned (128-bit, 16-byte) stack. This is
793 mandatory for the 64-bit ABI, and may or may not be true for other
794 operating systems. */
795#define TARGET_KEEPS_VECTOR_ALIGNED_STACK TARGET_64BIT
796
797/* Minimum allocation boundary for the code of a function. */
798#define FUNCTION_BOUNDARY 8
799
800/* We will and with this value to test if a custom function descriptor needs
801 a static chain. The function boundary must the adjusted so that the bit
802 this represents is no longer part of the address. 0 Disables the custom
803 function descriptors. */
804#define X86_CUSTOM_FUNCTION_TEST 1
805
806/* C++ stores the virtual bit in the lowest bit of function pointers. */
807#define TARGET_PTRMEMFUNC_VBIT_LOCATION ptrmemfunc_vbit_in_pfn
808
809/* Minimum size in bits of the largest boundary to which any
810 and all fundamental data types supported by the hardware
811 might need to be aligned. No data type wants to be aligned
812 rounder than this.
813
814 Pentium+ prefers DFmode values to be aligned to 64 bit boundary
815 and Pentium Pro XFmode values at 128 bit boundaries.
816
817 When increasing the maximum, also update
818 TARGET_ABSOLUTE_BIGGEST_ALIGNMENT. */
819
820#define BIGGEST_ALIGNMENT \
821 (TARGET_IAMCU ? 32 : (TARGET_AVX512F \
822 ? 512 : (TARGET_AVX ? 256 : 128)))
823
824/* Maximum stack alignment. */
825#define MAX_STACK_ALIGNMENT MAX_OFILE_ALIGNMENT
826
827/* Alignment value for attribute ((aligned)). It is a constant since
828 it is the part of the ABI. We shouldn't change it with -mavx. */
829#define ATTRIBUTE_ALIGNED_VALUE (TARGET_IAMCU ? 32 : 128)
830
831/* Decide whether a variable of mode MODE should be 128 bit aligned. */
832#define ALIGN_MODE_128(MODE) \
833 ((MODE) == XFmode || SSE_REG_MODE_P (MODE))
834
835/* The published ABIs say that doubles should be aligned on word
836 boundaries, so lower the alignment for structure fields unless
837 -malign-double is set. */
838
839/* ??? Blah -- this macro is used directly by libobjc. Since it
840 supports no vector modes, cut out the complexity and fall back
841 on BIGGEST_FIELD_ALIGNMENT. */
842#ifdef IN_TARGET_LIBS
843#ifdef __x86_64__
844#define BIGGEST_FIELD_ALIGNMENT 128
845#else
846#define BIGGEST_FIELD_ALIGNMENT 32
847#endif
848#else
849#define ADJUST_FIELD_ALIGN(FIELD, TYPE, COMPUTED) \
850 x86_field_alignment ((TYPE), (COMPUTED))
851#endif
852
853/* If defined, a C expression to compute the alignment for a static
854 variable. TYPE is the data type, and ALIGN is the alignment that
855 the object would ordinarily have. The value of this macro is used
856 instead of that alignment to align the object.
857
858 If this macro is not defined, then ALIGN is used.
859
860 One use of this macro is to increase alignment of medium-size
861 data to make it all fit in fewer cache lines. Another is to
862 cause character arrays to be word-aligned so that `strcpy' calls
863 that copy constants to character arrays can be done inline. */
864
865#define DATA_ALIGNMENT(TYPE, ALIGN) \
866 ix86_data_alignment ((TYPE), (ALIGN), true)
867
868/* Similar to DATA_ALIGNMENT, but for the cases where the ABI mandates
869 some alignment increase, instead of optimization only purposes. E.g.
870 AMD x86-64 psABI says that variables with array type larger than 15 bytes
871 must be aligned to 16 byte boundaries.
872
873 If this macro is not defined, then ALIGN is used. */
874
875#define DATA_ABI_ALIGNMENT(TYPE, ALIGN) \
876 ix86_data_alignment ((TYPE), (ALIGN), false)
877
878/* If defined, a C expression to compute the alignment for a local
879 variable. TYPE is the data type, and ALIGN is the alignment that
880 the object would ordinarily have. The value of this macro is used
881 instead of that alignment to align the object.
882
883 If this macro is not defined, then ALIGN is 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_ALIGNMENT(TYPE, ALIGN) \
889 ix86_local_alignment ((TYPE), VOIDmode, (ALIGN))
890
891/* If defined, a C expression to compute the alignment for stack slot.
892 TYPE is the data type, MODE is the widest mode available, and ALIGN
893 is the alignment that the slot would ordinarily have. The value of
894 this macro is used instead of that alignment to align the slot.
895
896 If this macro is not defined, then ALIGN is used when TYPE is NULL,
897 Otherwise, LOCAL_ALIGNMENT will be used.
898
899 One use of this macro is to set alignment of stack slot to the
900 maximum alignment of all possible modes which the slot may have. */
901
902#define STACK_SLOT_ALIGNMENT(TYPE, MODE, ALIGN) \
903 ix86_local_alignment ((TYPE), (MODE), (ALIGN))
904
905/* If defined, a C expression to compute the alignment for a local
906 variable DECL.
907
908 If this macro is not defined, then
909 LOCAL_ALIGNMENT (TREE_TYPE (DECL), DECL_ALIGN (DECL)) will be used.
910
911 One use of this macro is to increase alignment of medium-size
912 data to make it all fit in fewer cache lines. */
913
914#define LOCAL_DECL_ALIGNMENT(DECL) \
915 ix86_local_alignment ((DECL), VOIDmode, DECL_ALIGN (DECL))
916
917/* If defined, a C expression to compute the minimum required alignment
918 for dynamic stack realignment purposes for EXP (a TYPE or DECL),
919 MODE, assuming normal alignment ALIGN.
920
921 If this macro is not defined, then (ALIGN) will be used. */
922
923#define MINIMUM_ALIGNMENT(EXP, MODE, ALIGN) \
924 ix86_minimum_alignment ((EXP), (MODE), (ALIGN))
925
926
927/* Set this nonzero if move instructions will actually fail to work
928 when given unaligned data. */
929#define STRICT_ALIGNMENT 0
930
931/* If bit field type is int, don't let it cross an int,
932 and give entire struct the alignment of an int. */
933/* Required on the 386 since it doesn't have bit-field insns. */
934#define PCC_BITFIELD_TYPE_MATTERS 1
935
936#define VECTOR_STORE_FLAG_VALUE(MODE) \
937 (GET_MODE_CLASS (MODE) == MODE_VECTOR_INT ? constm1_rtx : NULL_RTX)
938
939/* Standard register usage. */
940
941/* This processor has special stack-like registers. See reg-stack.cc
942 for details. */
943
944#define STACK_REGS
945
946#define IS_STACK_MODE(MODE) \
947 (X87_FLOAT_MODE_P (MODE) \
948 && (!(SSE_FLOAT_MODE_P (MODE) && TARGET_SSE_MATH) \
949 || TARGET_MIX_SSE_I387))
950
951/* Number of actual hardware registers.
952 The hardware registers are assigned numbers for the compiler
953 from 0 to just below FIRST_PSEUDO_REGISTER.
954 All registers that the compiler knows about must be given numbers,
955 even those that are not normally considered general registers.
956
957 In the 80386 we give the 8 general purpose registers the numbers 0-7.
958 We number the floating point registers 8-15.
959 Note that registers 0-7 can be accessed as a short or int,
960 while only 0-3 may be used with byte `mov' instructions.
961
962 Reg 16 does not correspond to any hardware register, but instead
963 appears in the RTL as an argument pointer prior to reload, and is
964 eliminated during reloading in favor of either the stack or frame
965 pointer. */
966
967#define FIRST_PSEUDO_REGISTER FIRST_PSEUDO_REG
968
969/* Number of hardware registers that go into the DWARF-2 unwind info.
970 If not defined, equals FIRST_PSEUDO_REGISTER. */
971
972#define DWARF_FRAME_REGISTERS 17
973
974/* 1 for registers that have pervasive standard uses
975 and are not available for the register allocator.
976 On the 80386, the stack pointer is such, as is the arg pointer.
977
978 REX registers are disabled for 32bit targets in
979 TARGET_CONDITIONAL_REGISTER_USAGE. */
980
981#define FIXED_REGISTERS \
982/*ax,dx,cx,bx,si,di,bp,sp,st,st1,st2,st3,st4,st5,st6,st7*/ \
983{ 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, \
984/*arg,flags,fpsr,frame*/ \
985 1, 1, 1, 1, \
986/*xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7*/ \
987 0, 0, 0, 0, 0, 0, 0, 0, \
988/* mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7*/ \
989 0, 0, 0, 0, 0, 0, 0, 0, \
990/* r8, r9, r10, r11, r12, r13, r14, r15*/ \
991 0, 0, 0, 0, 0, 0, 0, 0, \
992/*xmm8,xmm9,xmm10,xmm11,xmm12,xmm13,xmm14,xmm15*/ \
993 0, 0, 0, 0, 0, 0, 0, 0, \
994/*xmm16,xmm17,xmm18,xmm19,xmm20,xmm21,xmm22,xmm23*/ \
995 0, 0, 0, 0, 0, 0, 0, 0, \
996/*xmm24,xmm25,xmm26,xmm27,xmm28,xmm29,xmm30,xmm31*/ \
997 0, 0, 0, 0, 0, 0, 0, 0, \
998/* k0, k1, k2, k3, k4, k5, k6, k7*/ \
999 0, 0, 0, 0, 0, 0, 0, 0, \
1000/* r16, r17, r18, r19, r20, r21, r22, r23*/ \
1001 0, 0, 0, 0, 0, 0, 0, 0, \
1002/* r24, r25, r26, r27, r28, r29, r30, r31*/ \
1003 0, 0, 0, 0, 0, 0, 0, 0} \
1004
1005/* 1 for registers not available across function calls.
1006 These must include the FIXED_REGISTERS and also any
1007 registers that can be used without being saved.
1008 The latter must include the registers where values are returned
1009 and the register where structure-value addresses are passed.
1010 Aside from that, you can include as many other registers as you like.
1011
1012 Value is set to 1 if the register is call used unconditionally.
1013 Bit one is set if the register is call used on TARGET_32BIT ABI.
1014 Bit two is set if the register is call used on TARGET_64BIT ABI.
1015 Bit three is set if the register is call used on TARGET_64BIT_MS_ABI.
1016
1017 Proper values are computed in TARGET_CONDITIONAL_REGISTER_USAGE. */
1018
1019#define CALL_USED_REGISTERS_MASK(IS_64BIT_MS_ABI) \
1020 ((IS_64BIT_MS_ABI) ? (1 << 3) : TARGET_64BIT ? (1 << 2) : (1 << 1))
1021
1022#define CALL_USED_REGISTERS \
1023/*ax,dx,cx,bx,si,di,bp,sp,st,st1,st2,st3,st4,st5,st6,st7*/ \
1024{ 1, 1, 1, 0, 4, 4, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, \
1025/*arg,flags,fpsr,frame*/ \
1026 1, 1, 1, 1, \
1027/*xmm0,xmm1,xmm2,xmm3,xmm4,xmm5,xmm6,xmm7*/ \
1028 1, 1, 1, 1, 1, 1, 6, 6, \
1029/* mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7*/ \
1030 1, 1, 1, 1, 1, 1, 1, 1, \
1031/* r8, r9, r10, r11, r12, r13, r14, r15*/ \
1032 1, 1, 1, 1, 2, 2, 2, 2, \
1033/*xmm8,xmm9,xmm10,xmm11,xmm12,xmm13,xmm14,xmm15*/ \
1034 6, 6, 6, 6, 6, 6, 6, 6, \
1035/*xmm16,xmm17,xmm18,xmm19,xmm20,xmm21,xmm22,xmm23*/ \
1036 1, 1, 1, 1, 1, 1, 1, 1, \
1037/*xmm24,xmm25,xmm26,xmm27,xmm28,xmm29,xmm30,xmm31*/ \
1038 1, 1, 1, 1, 1, 1, 1, 1, \
1039 /* k0, k1, k2, k3, k4, k5, k6, k7*/ \
1040 1, 1, 1, 1, 1, 1, 1, 1, \
1041/* r16, r17, r18, r19, r20, r21, r22, r23*/ \
1042 1, 1, 1, 1, 1, 1, 1, 1, \
1043/* r24, r25, r26, r27, r28, r29, r30, r31*/ \
1044 1, 1, 1, 1, 1, 1, 1, 1} \
1045
1046/* Order in which to allocate registers. Each register must be
1047 listed once, even those in FIXED_REGISTERS. List frame pointer
1048 late and fixed registers last. Note that, in general, we prefer
1049 registers listed in CALL_USED_REGISTERS, keeping the others
1050 available for storage of persistent values.
1051
1052 The ADJUST_REG_ALLOC_ORDER actually overwrite the order,
1053 so this is just empty initializer for array. */
1054
1055#define REG_ALLOC_ORDER \
1056{ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, \
1057 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, \
1058 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, \
1059 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, \
1060 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, \
1061 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91}
1062
1063/* ADJUST_REG_ALLOC_ORDER is a macro which permits reg_alloc_order
1064 to be rearranged based on a particular function. When using sse math,
1065 we want to allocate SSE before x87 registers and vice versa. */
1066
1067#define ADJUST_REG_ALLOC_ORDER x86_order_regs_for_local_alloc ()
1068
1069#define INSN_BASE_REG_CLASS(INSN) \
1070 ix86_insn_base_reg_class (INSN)
1071
1072#define REGNO_OK_FOR_INSN_BASE_P(NUM, INSN) \
1073 ix86_regno_ok_for_insn_base_p (NUM, INSN)
1074
1075#define INSN_INDEX_REG_CLASS(INSN) \
1076 ix86_insn_index_reg_class (INSN)
1077
1078#define OVERRIDE_ABI_FORMAT(FNDECL) ix86_call_abi_override (FNDECL)
1079
1080#define HARD_REGNO_NREGS_HAS_PADDING(REGNO, MODE) \
1081 (TARGET_128BIT_LONG_DOUBLE && !TARGET_64BIT \
1082 && GENERAL_REGNO_P (REGNO) \
1083 && ((MODE) == XFmode || (MODE) == XCmode))
1084
1085#define HARD_REGNO_NREGS_WITH_PADDING(REGNO, MODE) ((MODE) == XFmode ? 4 : 8)
1086
1087#define REGMODE_NATURAL_SIZE(MODE) ix86_regmode_natural_size (MODE)
1088
1089#define VALID_AVX256_REG_MODE(MODE) \
1090 ((MODE) == V32QImode || (MODE) == V16HImode || (MODE) == V8SImode \
1091 || (MODE) == V4DImode || (MODE) == V2TImode || (MODE) == V8SFmode \
1092 || (MODE) == V4DFmode || (MODE) == V16HFmode || (MODE) == V16BFmode)
1093
1094#define VALID_AVX256_REG_OR_OI_MODE(MODE) \
1095 (VALID_AVX256_REG_MODE (MODE) || (MODE) == OImode)
1096
1097#define VALID_AVX512F_SCALAR_MODE(MODE) \
1098 ((MODE) == DImode || (MODE) == DFmode \
1099 || (MODE) == SImode || (MODE) == SFmode \
1100 || (MODE) == HImode || (MODE) == HFmode || (MODE) == BFmode)
1101
1102#define VALID_AVX512F_REG_MODE(MODE) \
1103 ((MODE) == V8DImode || (MODE) == V8DFmode || (MODE) == V64QImode \
1104 || (MODE) == V16SImode || (MODE) == V16SFmode || (MODE) == V32HImode \
1105 || (MODE) == V4TImode || (MODE) == V32HFmode || (MODE) == V32BFmode)
1106
1107#define VALID_AVX512F_REG_OR_XI_MODE(MODE) \
1108 (VALID_AVX512F_REG_MODE (MODE) || (MODE) == XImode)
1109
1110#define VALID_AVX512VL_128_REG_MODE(MODE) \
1111 ((MODE) == V2DImode || (MODE) == V2DFmode || (MODE) == V16QImode \
1112 || (MODE) == V4SImode || (MODE) == V4SFmode || (MODE) == V8HImode \
1113 || (MODE) == TFmode || (MODE) == V1TImode || (MODE) == V8HFmode \
1114 || (MODE) == V8BFmode || (MODE) == TImode)
1115
1116#define VALID_AVX512FP16_REG_MODE(MODE) \
1117 ((MODE) == V8HFmode || (MODE) == V16HFmode || (MODE) == V32HFmode)
1118
1119#define VALID_SSE2_TYPE_MODE(MODE) \
1120 ((MODE) == HFmode || (MODE) == BFmode \
1121 || (MODE) == HCmode || (MODE) == BCmode)
1122
1123#define VALID_SSE2_REG_MODE(MODE) \
1124 ((MODE) == V16QImode || (MODE) == V8HImode || (MODE) == V2DFmode \
1125 || (MODE) == V8HFmode || (MODE) == V4HFmode || (MODE) == V2HFmode \
1126 || (MODE) == V8BFmode || (MODE) == V4BFmode || (MODE) == V2BFmode \
1127 || (MODE) == V4QImode || (MODE) == V2HImode || (MODE) == V1SImode \
1128 || (MODE) == V2DImode || (MODE) == V2QImode || (MODE) == HImode \
1129 || (MODE) == DFmode || (MODE) == DImode \
1130 || (MODE) == HFmode || (MODE) == BFmode)
1131
1132#define VALID_SSE_REG_MODE(MODE) \
1133 ((MODE) == V1TImode || (MODE) == TImode \
1134 || (MODE) == V4SFmode || (MODE) == V4SImode \
1135 || (MODE) == SFmode || (MODE) == SImode \
1136 || (MODE) == TFmode || (MODE) == TDmode)
1137
1138#define VALID_MMX_REG_MODE_3DNOW(MODE) \
1139 ((MODE) == V2SFmode || (MODE) == SFmode)
1140
1141/* To match ia32 psABI, V4HFmode should be added here. */
1142#define VALID_MMX_REG_MODE(MODE) \
1143 ((MODE) == V1DImode || (MODE) == DImode \
1144 || (MODE) == V2SImode || (MODE) == SImode \
1145 || (MODE) == V4HImode || (MODE) == V8QImode \
1146 || (MODE) == V4HFmode || (MODE) == V4BFmode)
1147
1148#define VALID_MASK_REG_MODE(MODE) ((MODE) == HImode || (MODE) == QImode)
1149
1150#define VALID_MASK_AVX512BW_MODE(MODE) ((MODE) == SImode || (MODE) == DImode)
1151
1152#define VALID_FP_MODE_P(MODE) \
1153 ((MODE) == SFmode || (MODE) == DFmode || (MODE) == XFmode \
1154 || (MODE) == SCmode || (MODE) == DCmode || (MODE) == XCmode)
1155
1156#define VALID_INT_MODE_P(MODE) \
1157 ((MODE) == QImode || (MODE) == HImode \
1158 || (MODE) == SImode || (MODE) == DImode \
1159 || (MODE) == CQImode || (MODE) == CHImode \
1160 || (MODE) == CSImode || (MODE) == CDImode \
1161 || (MODE) == SDmode || (MODE) == DDmode \
1162 || (MODE) == HFmode || (MODE) == HCmode || (MODE) == BFmode \
1163 || (MODE) == V2HImode || (MODE) == V2HFmode || (MODE) == V2BFmode \
1164 || (MODE) == V1SImode || (MODE) == V4QImode || (MODE) == V2QImode \
1165 || (TARGET_64BIT \
1166 && ((MODE) == TImode || (MODE) == CTImode \
1167 || (MODE) == TFmode || (MODE) == TCmode \
1168 || (MODE) == V8QImode || (MODE) == V4HImode \
1169 || (MODE) == V2SImode || (MODE) == TDmode)))
1170
1171/* Return true for modes passed in SSE registers. */
1172#define SSE_REG_MODE_P(MODE) \
1173 ((MODE) == V1TImode || (MODE) == TImode || (MODE) == V16QImode \
1174 || (MODE) == TFmode || (MODE) == V8HImode || (MODE) == V2DFmode \
1175 || (MODE) == V2DImode || (MODE) == V4SFmode || (MODE) == V4SImode \
1176 || (MODE) == V32QImode || (MODE) == V16HImode || (MODE) == V8SImode \
1177 || (MODE) == V4DImode || (MODE) == V8SFmode || (MODE) == V4DFmode \
1178 || (MODE) == V2TImode || (MODE) == V8DImode || (MODE) == V64QImode \
1179 || (MODE) == V16SImode || (MODE) == V32HImode || (MODE) == V8DFmode \
1180 || (MODE) == V16SFmode \
1181 || (MODE) == V32HFmode || (MODE) == V16HFmode || (MODE) == V8HFmode \
1182 || (MODE) == V32BFmode || (MODE) == V16BFmode || (MODE) == V8BFmode)
1183
1184#define X87_FLOAT_MODE_P(MODE) \
1185 (TARGET_80387 && ((MODE) == SFmode || (MODE) == DFmode || (MODE) == XFmode))
1186
1187#define SSE_FLOAT_MODE_P(MODE) \
1188 ((TARGET_SSE && (MODE) == SFmode) || (TARGET_SSE2 && (MODE) == DFmode))
1189
1190#define SSE_FLOAT_MODE_SSEMATH_OR_HFBF_P(MODE) \
1191 ((SSE_FLOAT_MODE_P (MODE) && TARGET_SSE_MATH) \
1192 || (TARGET_AVX512FP16 && (MODE) == HFmode) \
1193 || (TARGET_AVX10_2 && (MODE) == BFmode))
1194
1195#define FMA4_VEC_FLOAT_MODE_P(MODE) \
1196 (TARGET_FMA4 && ((MODE) == V4SFmode || (MODE) == V2DFmode \
1197 || (MODE) == V8SFmode || (MODE) == V4DFmode))
1198
1199#define VALID_BCST_MODE_P(MODE) \
1200 ((MODE) == SFmode || (MODE) == DFmode \
1201 || (MODE) == SImode || (MODE) == DImode \
1202 || (MODE) == HFmode)
1203
1204/* It is possible to write patterns to move flags; but until someone
1205 does it, */
1206#define AVOID_CCMODE_COPIES
1207
1208/* Specify the modes required to caller save a given hard regno.
1209 We do this on i386 to prevent flags from being saved at all.
1210
1211 Kill any attempts to combine saving of modes. */
1212
1213#define HARD_REGNO_CALLER_SAVE_MODE(REGNO, NREGS, MODE) \
1214 (CC_REGNO_P (REGNO) ? VOIDmode \
1215 : (MODE) == VOIDmode && (NREGS) != 1 ? VOIDmode \
1216 : (MODE) == VOIDmode ? choose_hard_reg_mode ((REGNO), (NREGS), NULL) \
1217 : (MODE) == HImode && !((GENERAL_REGNO_P (REGNO) \
1218 && TARGET_PARTIAL_REG_STALL) \
1219 || MASK_REGNO_P (REGNO)) ? SImode \
1220 : (MODE) == QImode && !(ANY_QI_REGNO_P (REGNO) \
1221 || MASK_REGNO_P (REGNO)) ? SImode \
1222 : (MODE))
1223
1224/* Specify the registers used for certain standard purposes.
1225 The values of these macros are register numbers. */
1226
1227/* on the 386 the pc register is %eip, and is not usable as a general
1228 register. The ordinary mov instructions won't work */
1229/* #define PC_REGNUM */
1230
1231/* Base register for access to arguments of the function. */
1232#define ARG_POINTER_REGNUM ARGP_REG
1233
1234/* Register to use for pushing function arguments. */
1235#define STACK_POINTER_REGNUM SP_REG
1236
1237/* Base register for access to local variables of the function. */
1238#define FRAME_POINTER_REGNUM FRAME_REG
1239#define HARD_FRAME_POINTER_REGNUM BP_REG
1240
1241#define FIRST_INT_REG AX_REG
1242#define LAST_INT_REG SP_REG
1243
1244#define FIRST_INDEX_REG AX_REG
1245#define LAST_INDEX_REG BP_REG
1246
1247#define FIRST_QI_REG AX_REG
1248#define LAST_QI_REG BX_REG
1249
1250/* First & last stack-like regs */
1251#define FIRST_STACK_REG ST0_REG
1252#define LAST_STACK_REG ST7_REG
1253
1254#define FIRST_SSE_REG XMM0_REG
1255#define LAST_SSE_REG XMM7_REG
1256
1257#define FIRST_MMX_REG MM0_REG
1258#define LAST_MMX_REG MM7_REG
1259
1260#define FIRST_REX_INT_REG R8_REG
1261#define LAST_REX_INT_REG R15_REG
1262
1263#define FIRST_REX_SSE_REG XMM8_REG
1264#define LAST_REX_SSE_REG XMM15_REG
1265
1266#define FIRST_EXT_REX_SSE_REG XMM16_REG
1267#define LAST_EXT_REX_SSE_REG XMM31_REG
1268
1269#define FIRST_MASK_REG MASK0_REG
1270#define LAST_MASK_REG MASK7_REG
1271
1272#define FIRST_REX2_INT_REG R16_REG
1273#define LAST_REX2_INT_REG R31_REG
1274
1275/* Override this in other tm.h files to cope with various OS lossage
1276 requiring a frame pointer. */
1277#ifndef SUBTARGET_FRAME_POINTER_REQUIRED
1278#define SUBTARGET_FRAME_POINTER_REQUIRED 0
1279#endif
1280
1281/* Define the shadow offset for asan. Other OS's can override in the
1282 respective tm.h files. */
1283#ifndef SUBTARGET_SHADOW_OFFSET
1284#define SUBTARGET_SHADOW_OFFSET \
1285 (TARGET_LP64 ? HOST_WIDE_INT_C (0x7fff8000) : HOST_WIDE_INT_1 << 29)
1286#endif
1287
1288/* Make sure we can access arbitrary call frames. */
1289#define SETUP_FRAME_ADDRESSES() ix86_setup_frame_addresses ()
1290
1291/* Register to hold the addressing base for position independent
1292 code access to data items. We don't use PIC pointer for 64bit
1293 mode. Define the regnum to dummy value to prevent gcc from
1294 pessimizing code dealing with EBX.
1295
1296 To avoid clobbering a call-saved register unnecessarily, we renumber
1297 the pic register when possible. The change is visible after the
1298 prologue has been emitted. */
1299
1300#define REAL_PIC_OFFSET_TABLE_REGNUM (TARGET_64BIT ? R15_REG : BX_REG)
1301
1302#define PIC_OFFSET_TABLE_REGNUM \
1303 (ix86_use_pseudo_pic_reg () \
1304 ? (pic_offset_table_rtx \
1305 ? INVALID_REGNUM \
1306 : REAL_PIC_OFFSET_TABLE_REGNUM) \
1307 : INVALID_REGNUM)
1308
1309#define GOT_SYMBOL_NAME "_GLOBAL_OFFSET_TABLE_"
1310
1311/* This is overridden by <cygwin.h>. */
1312#define MS_AGGREGATE_RETURN 0
1313
1314#define KEEP_AGGREGATE_RETURN_POINTER 0
1315
1316/* Define the classes of registers for register constraints in the
1317 machine description. Also define ranges of constants.
1318
1319 One of the classes must always be named ALL_REGS and include all hard regs.
1320 If there is more than one class, another class must be named NO_REGS
1321 and contain no registers.
1322
1323 The name GENERAL_REGS must be the name of a class (or an alias for
1324 another name such as ALL_REGS). This is the class of registers
1325 that is allowed by "g" or "r" in a register constraint.
1326 Also, registers outside this class are allocated only when
1327 instructions express preferences for them.
1328
1329 The classes must be numbered in nondecreasing order; that is,
1330 a larger-numbered class must never be contained completely
1331 in a smaller-numbered class. This is why CLOBBERED_REGS class
1332 is listed early, even though in 64-bit mode it contains more
1333 registers than just %eax, %ecx, %edx.
1334
1335 For any two classes, it is very desirable that there be another
1336 class that represents their union.
1337
1338 The flags and fpsr registers are in no class. */
1339
1340enum reg_class
1341{
1342 NO_REGS,
1343 AREG, DREG, CREG, BREG, SIREG, DIREG,
1344 AD_REGS, /* %eax/%edx for DImode */
1345 CLOBBERED_REGS, /* call-clobbered integer registers */
1346 Q_REGS, /* %eax %ebx %ecx %edx */
1347 NON_Q_REGS, /* %esi %edi %ebp %esp */
1348 TLS_GOTBASE_REGS, /* %ebx %ecx %edx %esi %edi %ebp */
1349 LEGACY_GENERAL_REGS, /* %eax %ebx %ecx %edx %esi %edi %ebp %esp */
1350 LEGACY_INDEX_REGS, /* %eax %ebx %ecx %edx %esi %edi %ebp */
1351 GENERAL_REGS, /* %eax %ebx %ecx %edx %esi %edi %ebp %esp
1352 %r8 %r9 %r10 %r11 %r12 %r13 %r14 %r15
1353 %r16 %r17 %r18 %r19 %r20 %r21 %r22 %r23
1354 %r24 %r25 %r26 %r27 %r28 %r29 %r30 %r31 */
1355 INDEX_REGS, /* %eax %ebx %ecx %edx %esi %edi %ebp
1356 %r8 %r9 %r10 %r11 %r12 %r13 %r14 %r15
1357 %r16 %r17 %r18 %r19 %r20 %r21 %r22 %r23
1358 %r24 %r25 %r26 %r27 %r28 %r29 %r30 %r31 */
1359 GENERAL_GPR16, /* %eax %ebx %ecx %edx %esi %edi %ebp %esp
1360 %r8 %r9 %r10 %r11 %r12 %r13 %r14 %r15 */
1361 INDEX_GPR16, /* %eax %ebx %ecx %edx %esi %edi %ebp
1362 %r8 %r9 %r10 %r11 %r12 %r13 %r14 %r15 */
1363 FP_TOP_REG, FP_SECOND_REG, /* %st(0) %st(1) */
1364 FLOAT_REGS,
1365 SSE_FIRST_REG,
1366 NO_REX_SSE_REGS,
1367 SSE_REGS,
1368 ALL_SSE_REGS,
1369 MMX_REGS,
1370 FLOAT_SSE_REGS,
1371 FLOAT_INT_REGS,
1372 INT_SSE_REGS,
1373 FLOAT_INT_SSE_REGS,
1374 MASK_REGS,
1375 ALL_MASK_REGS,
1376 INT_MASK_REGS,
1377 ALL_REGS,
1378 LIM_REG_CLASSES
1379};
1380
1381#define N_REG_CLASSES ((int) LIM_REG_CLASSES)
1382
1383#define INTEGER_CLASS_P(CLASS) \
1384 reg_class_subset_p ((CLASS), GENERAL_REGS)
1385#define FLOAT_CLASS_P(CLASS) \
1386 reg_class_subset_p ((CLASS), FLOAT_REGS)
1387#define SSE_CLASS_P(CLASS) \
1388 reg_class_subset_p ((CLASS), ALL_SSE_REGS)
1389#define INT_SSE_CLASS_P(CLASS) \
1390 reg_class_subset_p ((CLASS), INT_SSE_REGS)
1391#define MMX_CLASS_P(CLASS) \
1392 ((CLASS) == MMX_REGS)
1393#define MASK_CLASS_P(CLASS) \
1394 reg_class_subset_p ((CLASS), ALL_MASK_REGS)
1395#define MAYBE_INTEGER_CLASS_P(CLASS) \
1396 reg_classes_intersect_p ((CLASS), GENERAL_REGS)
1397#define MAYBE_FLOAT_CLASS_P(CLASS) \
1398 reg_classes_intersect_p ((CLASS), FLOAT_REGS)
1399#define MAYBE_SSE_CLASS_P(CLASS) \
1400 reg_classes_intersect_p ((CLASS), ALL_SSE_REGS)
1401#define MAYBE_MMX_CLASS_P(CLASS) \
1402 reg_classes_intersect_p ((CLASS), MMX_REGS)
1403#define MAYBE_MASK_CLASS_P(CLASS) \
1404 reg_classes_intersect_p ((CLASS), ALL_MASK_REGS)
1405
1406#define Q_CLASS_P(CLASS) \
1407 reg_class_subset_p ((CLASS), Q_REGS)
1408
1409#define MAYBE_NON_Q_CLASS_P(CLASS) \
1410 reg_classes_intersect_p ((CLASS), NON_Q_REGS)
1411
1412/* Give names of register classes as strings for dump file. */
1413
1414#define REG_CLASS_NAMES \
1415{ "NO_REGS", \
1416 "AREG", "DREG", "CREG", "BREG", \
1417 "SIREG", "DIREG", \
1418 "AD_REGS", \
1419 "CLOBBERED_REGS", \
1420 "Q_REGS", "NON_Q_REGS", \
1421 "TLS_GOTBASE_REGS", \
1422 "LEGACY_GENERAL_REGS", \
1423 "LEGACY_INDEX_REGS", \
1424 "GENERAL_REGS", \
1425 "INDEX_REGS", \
1426 "GENERAL_GPR16", \
1427 "INDEX_GPR16", \
1428 "FP_TOP_REG", "FP_SECOND_REG", \
1429 "FLOAT_REGS", \
1430 "SSE_FIRST_REG", \
1431 "NO_REX_SSE_REGS", \
1432 "SSE_REGS", \
1433 "ALL_SSE_REGS", \
1434 "MMX_REGS", \
1435 "FLOAT_SSE_REGS", \
1436 "FLOAT_INT_REGS", \
1437 "INT_SSE_REGS", \
1438 "FLOAT_INT_SSE_REGS", \
1439 "MASK_REGS", \
1440 "ALL_MASK_REGS", \
1441 "INT_MASK_REGS", \
1442 "ALL_REGS" }
1443
1444/* Define which registers fit in which classes. This is an initializer
1445 for a vector of HARD_REG_SET of length N_REG_CLASSES.
1446
1447 Note that CLOBBERED_REGS are calculated by
1448 TARGET_CONDITIONAL_REGISTER_USAGE. */
1449
1450#define REG_CLASS_CONTENTS \
1451{ { 0x0, 0x0, 0x0 }, /* NO_REGS */ \
1452 { 0x01, 0x0, 0x0 }, /* AREG */ \
1453 { 0x02, 0x0, 0x0 }, /* DREG */ \
1454 { 0x04, 0x0, 0x0 }, /* CREG */ \
1455 { 0x08, 0x0, 0x0 }, /* BREG */ \
1456 { 0x10, 0x0, 0x0 }, /* SIREG */ \
1457 { 0x20, 0x0, 0x0 }, /* DIREG */ \
1458 { 0x03, 0x0, 0x0 }, /* AD_REGS */ \
1459 { 0x07, 0x0, 0x0 }, /* CLOBBERED_REGS */ \
1460 { 0x0f, 0x0, 0x0 }, /* Q_REGS */ \
1461 { 0x900f0, 0x0, 0x0 }, /* NON_Q_REGS */ \
1462 { 0x7e, 0xff0, 0x0 }, /* TLS_GOTBASE_REGS */ \
1463 { 0x900ff, 0x0, 0x0 }, /* LEGACY_GENERAL_REGS */ \
1464 { 0x7f, 0x0, 0x0 }, /* LEGACY_INDEX_REGS */ \
1465 { 0x900ff, 0xff0, 0xffff000 }, /* GENERAL_REGS */ \
1466 { 0x7f, 0xff0, 0xffff000 }, /* INDEX_REGS */ \
1467 { 0x900ff, 0xff0, 0x0 }, /* GENERAL_GPR16 */ \
1468 { 0x7f, 0xff0, 0x0 }, /* INDEX_GPR16 */ \
1469 { 0x100, 0x0, 0x0 }, /* FP_TOP_REG */ \
1470 { 0x200, 0x0, 0x0 }, /* FP_SECOND_REG */ \
1471 { 0xff00, 0x0, 0x0 }, /* FLOAT_REGS */ \
1472 { 0x100000, 0x0, 0x0 }, /* SSE_FIRST_REG */ \
1473 { 0xff00000, 0x0, 0x0 }, /* NO_REX_SSE_REGS */ \
1474 { 0xff00000, 0xff000, 0x0 }, /* SSE_REGS */ \
1475 { 0xff00000, 0xfffff000, 0xf }, /* ALL_SSE_REGS */ \
1476{ 0xf0000000, 0xf, 0x0 }, /* MMX_REGS */ \
1477 { 0xff0ff00, 0xfffff000, 0xf }, /* FLOAT_SSE_REGS */ \
1478 { 0x9ffff, 0xff0, 0xffff000 }, /* FLOAT_INT_REGS */ \
1479 { 0xff900ff, 0xfffffff0, 0xffff00f }, /* INT_SSE_REGS */ \
1480 { 0xff9ffff, 0xfffffff0, 0xffff00f }, /* FLOAT_INT_SSE_REGS */ \
1481 { 0x0, 0x0, 0xfe0 }, /* MASK_REGS */ \
1482 { 0x0, 0x0, 0xff0 }, /* ALL_MASK_REGS */ \
1483 { 0x900ff, 0xff0, 0xffffff0 }, /* INT_MASK_REGS */ \
1484{ 0xffffffff, 0xffffffff, 0xfffffff } /* ALL_REGS */ \
1485}
1486
1487/* The same information, inverted:
1488 Return the class number of the smallest class containing
1489 reg number REGNO. This could be a conditional expression
1490 or could index an array. */
1491
1492#define REGNO_REG_CLASS(REGNO) (regclass_map[(REGNO)])
1493
1494#define QI_REG_P(X) (REG_P (X) && QI_REGNO_P (REGNO (X)))
1495#define QI_REGNO_P(N) IN_RANGE ((N), FIRST_QI_REG, LAST_QI_REG)
1496
1497#define LEGACY_INT_REG_P(X) (REG_P (X) && LEGACY_INT_REGNO_P (REGNO (X)))
1498#define LEGACY_INT_REGNO_P(N) IN_RANGE ((N), FIRST_INT_REG, LAST_INT_REG)
1499
1500#define LEGACY_INDEX_REG_P(X) (REG_P (X) && LEGACY_INDEX_REGNO_P (REGNO (X)))
1501#define LEGACY_INDEX_REGNO_P(N) \
1502 IN_RANGE ((N), FIRST_INDEX_REG, LAST_INDEX_REG)
1503
1504#define REX_INT_REG_P(X) (REG_P (X) && REX_INT_REGNO_P (REGNO (X)))
1505#define REX_INT_REGNO_P(N) \
1506 IN_RANGE ((N), FIRST_REX_INT_REG, LAST_REX_INT_REG)
1507
1508#define REX2_INT_REG_P(X) (REG_P (X) && REX2_INT_REGNO_P (REGNO (X)))
1509#define REX2_INT_REGNO_P(N) \
1510 IN_RANGE ((N), FIRST_REX2_INT_REG, LAST_REX2_INT_REG)
1511
1512#define GENERAL_REG_P(X) (REG_P (X) && GENERAL_REGNO_P (REGNO (X)))
1513#define GENERAL_REGNO_P(N) \
1514 (LEGACY_INT_REGNO_P (N) || REX_INT_REGNO_P (N) || REX2_INT_REGNO_P (N))
1515
1516#define INDEX_REG_P(X) (REG_P (X) && INDEX_REGNO_P (REGNO (X)))
1517#define INDEX_REGNO_P(N) \
1518 (LEGACY_INDEX_REGNO_P (N) || REX_INT_REGNO_P (N) || REX2_INT_REGNO_P (N))
1519
1520#define GENERAL_GPR16_REGNO_P(N) \
1521 (LEGACY_INT_REGNO_P (N) || REX_INT_REGNO_P (N))
1522
1523#define ANY_QI_REG_P(X) (REG_P (X) && ANY_QI_REGNO_P (REGNO (X)))
1524#define ANY_QI_REGNO_P(N) \
1525 (TARGET_64BIT ? GENERAL_REGNO_P (N) : QI_REGNO_P (N))
1526
1527#define STACK_REG_P(X) (REG_P (X) && STACK_REGNO_P (REGNO (X)))
1528#define STACK_REGNO_P(N) IN_RANGE ((N), FIRST_STACK_REG, LAST_STACK_REG)
1529
1530#define SSE_REG_P(X) (REG_P (X) && SSE_REGNO_P (REGNO (X)))
1531#define SSE_REGNO_P(N) \
1532 (LEGACY_SSE_REGNO_P (N) \
1533 || REX_SSE_REGNO_P (N) \
1534 || EXT_REX_SSE_REGNO_P (N))
1535
1536#define LEGACY_SSE_REGNO_P(N) \
1537 IN_RANGE ((N), FIRST_SSE_REG, LAST_SSE_REG)
1538
1539#define REX_SSE_REGNO_P(N) \
1540 IN_RANGE ((N), FIRST_REX_SSE_REG, LAST_REX_SSE_REG)
1541
1542#define EXT_REX_SSE_REG_P(X) (REG_P (X) && EXT_REX_SSE_REGNO_P (REGNO (X)))
1543
1544#define EXT_REX_SSE_REGNO_P(N) \
1545 IN_RANGE ((N), FIRST_EXT_REX_SSE_REG, LAST_EXT_REX_SSE_REG)
1546
1547#define ANY_FP_REG_P(X) (REG_P (X) && ANY_FP_REGNO_P (REGNO (X)))
1548#define ANY_FP_REGNO_P(N) (STACK_REGNO_P (N) || SSE_REGNO_P (N))
1549
1550#define MASK_REG_P(X) (REG_P (X) && MASK_REGNO_P (REGNO (X)))
1551#define MASK_REGNO_P(N) IN_RANGE ((N), FIRST_MASK_REG, LAST_MASK_REG)
1552#define MASK_PAIR_REGNO_P(N) ((((N) - FIRST_MASK_REG) & 1) == 0)
1553
1554#define MMX_REG_P(X) (REG_P (X) && MMX_REGNO_P (REGNO (X)))
1555#define MMX_REGNO_P(N) IN_RANGE ((N), FIRST_MMX_REG, LAST_MMX_REG)
1556
1557#define CC_REG_P(X) (REG_P (X) && CC_REGNO_P (REGNO (X)))
1558#define CC_REGNO_P(X) ((X) == FLAGS_REG)
1559
1560#define MOD4_SSE_REG_P(X) (REG_P (X) && MOD4_SSE_REGNO_P (REGNO (X)))
1561#define MOD4_SSE_REGNO_P(N) ((N) == XMM0_REG \
1562 || (N) == XMM4_REG \
1563 || (N) == XMM8_REG \
1564 || (N) == XMM12_REG \
1565 || (N) == XMM16_REG \
1566 || (N) == XMM20_REG \
1567 || (N) == XMM24_REG \
1568 || (N) == XMM28_REG)
1569
1570/* First floating point reg */
1571#define FIRST_FLOAT_REG FIRST_STACK_REG
1572#define STACK_TOP_P(X) (REG_P (X) && REGNO (X) == FIRST_FLOAT_REG)
1573
1574#define GET_SSE_REGNO(N) \
1575 ((N) < 8 ? FIRST_SSE_REG + (N) \
1576 : (N) < 16 ? FIRST_REX_SSE_REG + (N) - 8 \
1577 : FIRST_EXT_REX_SSE_REG + (N) - 16)
1578
1579/* The class value for index registers, and the one for base regs. */
1580
1581#define INDEX_REG_CLASS INDEX_REGS
1582#define BASE_REG_CLASS GENERAL_REGS
1583
1584/* Stack layout; function entry, exit and calling. */
1585
1586/* Define this if pushing a word on the stack
1587 makes the stack pointer a smaller address. */
1588#define STACK_GROWS_DOWNWARD 1
1589
1590/* Define this to nonzero if the nominal address of the stack frame
1591 is at the high-address end of the local variables;
1592 that is, each additional local variable allocated
1593 goes at a more negative offset in the frame. */
1594#define FRAME_GROWS_DOWNWARD 1
1595
1596#define PUSH_ROUNDING(BYTES) ix86_push_rounding (BYTES)
1597
1598/* If defined, the maximum amount of space required for outgoing arguments
1599 will be computed and placed into the variable `crtl->outgoing_args_size'.
1600 No space will be pushed onto the stack for each call; instead, the
1601 function prologue should increase the stack frame size by this amount.
1602
1603 In 32bit mode enabling argument accumulation results in about 5% code size
1604 growth because move instructions are less compact than push. In 64bit
1605 mode the difference is less drastic but visible.
1606
1607 FIXME: Unlike earlier implementations, the size of unwind info seems to
1608 actually grow with accumulation. Is that because accumulated args
1609 unwind info became unnecesarily bloated?
1610
1611 With the 64-bit MS ABI, we can generate correct code with or without
1612 accumulated args, but because of OUTGOING_REG_PARM_STACK_SPACE the code
1613 generated without accumulated args is terrible.
1614
1615 If stack probes are required, the space used for large function
1616 arguments on the stack must also be probed, so enable
1617 -maccumulate-outgoing-args so this happens in the prologue.
1618
1619 We must use argument accumulation in interrupt function if stack
1620 may be realigned to avoid DRAP. */
1621
1622#define ACCUMULATE_OUTGOING_ARGS \
1623 ((TARGET_ACCUMULATE_OUTGOING_ARGS \
1624 && optimize_function_for_speed_p (cfun)) \
1625 || (cfun->machine->func_type != TYPE_NORMAL \
1626 && crtl->stack_realign_needed) \
1627 || TARGET_STACK_PROBE \
1628 || TARGET_64BIT_MS_ABI \
1629 || (TARGET_MACHO && crtl->profile))
1630
1631/* We want the stack and args grow in opposite directions, even if
1632 targetm.calls.push_argument returns false. */
1633#define PUSH_ARGS_REVERSED 1
1634
1635/* Offset of first parameter from the argument pointer register value. */
1636#define FIRST_PARM_OFFSET(FNDECL) 0
1637
1638/* Define this macro if functions should assume that stack space has been
1639 allocated for arguments even when their values are passed in registers.
1640
1641 The value of this macro is the size, in bytes, of the area reserved for
1642 arguments passed in registers for the function represented by FNDECL.
1643
1644 This space can be allocated by the caller, or be a part of the
1645 machine-dependent stack frame: `OUTGOING_REG_PARM_STACK_SPACE' says
1646 which. */
1647#define REG_PARM_STACK_SPACE(FNDECL) ix86_reg_parm_stack_space (FNDECL)
1648
1649#define OUTGOING_REG_PARM_STACK_SPACE(FNTYPE) \
1650 (TARGET_64BIT && ix86_function_type_abi (FNTYPE) == MS_ABI)
1651
1652/* Define how to find the value returned by a library function
1653 assuming the value has mode MODE. */
1654
1655#define LIBCALL_VALUE(MODE) ix86_libcall_value (MODE)
1656
1657/* Define the size of the result block used for communication between
1658 untyped_call and untyped_return. The block contains a DImode value
1659 followed by the block used by fnsave and frstor. */
1660
1661#define APPLY_RESULT_SIZE (8+108)
1662
1663/* 1 if N is a possible register number for function argument passing. */
1664#define FUNCTION_ARG_REGNO_P(N) ix86_function_arg_regno_p (N)
1665
1666/* Define a data type for recording info about an argument list
1667 during the scan of that argument list. This data type should
1668 hold all necessary information about the function itself
1669 and about the args processed so far, enough to enable macros
1670 such as FUNCTION_ARG to determine where the next arg should go. */
1671
1672typedef struct ix86_args {
1673 int words; /* # words passed so far */
1674 int nregs; /* # registers available for passing */
1675 int regno; /* next available register number */
1676 int fastcall; /* fastcall or thiscall calling convention
1677 is used */
1678 int sse_words; /* # sse words passed so far */
1679 int sse_nregs; /* # sse registers available for passing */
1680 int warn_avx512f; /* True when we want to warn
1681 about AVX512F ABI. */
1682 int warn_avx; /* True when we want to warn about AVX ABI. */
1683 int warn_sse; /* True when we want to warn about SSE ABI. */
1684 int warn_mmx; /* True when we want to warn about MMX ABI. */
1685 int warn_empty; /* True when we want to warn about empty classes
1686 passing ABI change. */
1687 int sse_regno; /* next available sse register number */
1688 int mmx_words; /* # mmx words passed so far */
1689 int mmx_nregs; /* # mmx registers available for passing */
1690 int mmx_regno; /* next available mmx register number */
1691 int maybe_vaarg; /* true for calls to possibly vardic fncts. */
1692 int caller; /* true if it is caller. */
1693 int float_in_sse; /* Set to 1 or 2 for 32bit targets if
1694 SFmode/DFmode arguments should be passed
1695 in SSE registers. Otherwise 0. */
1696 int stdarg; /* Set to 1 if function is stdarg. */
1697 enum calling_abi call_abi; /* Set to SYSV_ABI for sysv abi. Otherwise
1698 MS_ABI for ms abi. */
1699 tree decl; /* Callee decl. */
1700} CUMULATIVE_ARGS;
1701
1702/* Initialize a variable CUM of type CUMULATIVE_ARGS
1703 for a call to a function whose data type is FNTYPE.
1704 For a library call, FNTYPE is 0. */
1705
1706#define INIT_CUMULATIVE_ARGS(CUM, FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS) \
1707 init_cumulative_args (&(CUM), (FNTYPE), (LIBNAME), (FNDECL), \
1708 (N_NAMED_ARGS) != -1)
1709
1710/* Output assembler code to FILE to increment profiler label # LABELNO
1711 for profiling a function entry. */
1712
1713#define FUNCTION_PROFILER(FILE, LABELNO) \
1714 x86_function_profiler ((FILE), (LABELNO))
1715
1716#define MCOUNT_NAME "_mcount"
1717
1718#define MCOUNT_NAME_BEFORE_PROLOGUE "__fentry__"
1719
1720#define PROFILE_COUNT_REGISTER "edx"
1721
1722/* EXIT_IGNORE_STACK should be nonzero if, when returning from a function,
1723 the stack pointer does not matter. The value is tested only in
1724 functions that have frame pointers.
1725 No definition is equivalent to always zero. */
1726/* Note on the 386 it might be more efficient not to define this since
1727 we have to restore it ourselves from the frame pointer, in order to
1728 use pop */
1729
1730#define EXIT_IGNORE_STACK 1
1731
1732/* Define this macro as a C expression that is nonzero for registers
1733 used by the epilogue or the `return' pattern. */
1734
1735#define EPILOGUE_USES(REGNO) ix86_epilogue_uses (REGNO)
1736
1737/* Output assembler code for a block containing the constant parts
1738 of a trampoline, leaving space for the variable parts. */
1739
1740/* On the 386, the trampoline contains two instructions:
1741 mov #STATIC,ecx
1742 jmp FUNCTION
1743 The trampoline is generated entirely at runtime. The operand of JMP
1744 is the address of FUNCTION relative to the instruction following the
1745 JMP (which is 5 bytes long). */
1746
1747/* Length in units of the trampoline for entering a nested function. */
1748
1749#define TRAMPOLINE_SIZE (TARGET_64BIT ? 28 : 14)
1750
1751/* Definitions for register eliminations.
1752
1753 This is an array of structures. Each structure initializes one pair
1754 of eliminable registers. The "from" register number is given first,
1755 followed by "to". Eliminations of the same "from" register are listed
1756 in order of preference.
1757
1758 There are two registers that can always be eliminated on the i386.
1759 The frame pointer and the arg pointer can be replaced by either the
1760 hard frame pointer or to the stack pointer, depending upon the
1761 circumstances. The hard frame pointer is not used before reload and
1762 so it is not eligible for elimination. */
1763
1764#define ELIMINABLE_REGS \
1765{{ ARG_POINTER_REGNUM, STACK_POINTER_REGNUM}, \
1766 { ARG_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM}, \
1767 { FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM}, \
1768 { FRAME_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM}} \
1769
1770/* Define the offset between two registers, one to be eliminated, and the other
1771 its replacement, at the start of a routine. */
1772
1773#define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \
1774 ((OFFSET) = ix86_initial_elimination_offset ((FROM), (TO)))
1775
1776/* Addressing modes, and classification of registers for them. */
1777
1778/* Macros to check register numbers against specific register classes. */
1779
1780/* These assume that REGNO is a hard or pseudo reg number.
1781 They give nonzero only if REGNO is a hard reg of the suitable class
1782 or a pseudo reg currently allocated to a suitable hard reg.
1783 Since they use reg_renumber, they are safe only once reg_renumber
1784 has been allocated, which happens in reginfo.cc during register
1785 allocation. */
1786
1787#define REGNO_OK_FOR_INDEX_P(REGNO) \
1788 (INDEX_REGNO_P (REGNO) \
1789 || INDEX_REGNO_P (reg_renumber[(REGNO)]))
1790
1791#define REGNO_OK_FOR_BASE_P(REGNO) \
1792 (GENERAL_REGNO_P (REGNO) \
1793 || (REGNO) == ARG_POINTER_REGNUM \
1794 || (REGNO) == FRAME_POINTER_REGNUM \
1795 || GENERAL_REGNO_P (reg_renumber[(REGNO)]))
1796
1797/* Non strict versions, pseudos are ok. */
1798#define REGNO_OK_FOR_INDEX_NONSTRICT_P(REGNO) \
1799 (INDEX_REGNO_P (REGNO) \
1800 || !HARD_REGISTER_NUM_P (REGNO))
1801
1802#define REGNO_OK_FOR_BASE_NONSTRICT_P(REGNO) \
1803 (GENERAL_REGNO_P (REGNO) \
1804 || (REGNO) == ARG_POINTER_REGNUM \
1805 || (REGNO) == FRAME_POINTER_REGNUM \
1806 || !HARD_REGISTER_NUM_P (REGNO))
1807
1808/* TARGET_LEGITIMATE_ADDRESS_P recognizes an RTL expression
1809 that is a valid memory address for an instruction.
1810 The MODE argument is the machine mode for the MEM expression
1811 that wants to use this address.
1812
1813 The other macros defined here are used only in TARGET_LEGITIMATE_ADDRESS_P,
1814 except for CONSTANT_ADDRESS_P which is usually machine-independent.
1815
1816 See legitimize_pic_address in i386.cc for details as to what
1817 constitutes a legitimate address when -fpic is used. */
1818
1819#define MAX_REGS_PER_ADDRESS 2
1820
1821#define CONSTANT_ADDRESS_P(X) constant_address_p (X)
1822
1823/* If defined, a C expression to determine the base term of address X.
1824 This macro is used in only one place: `find_base_term' in alias.cc.
1825
1826 It is always safe for this macro to not be defined. It exists so
1827 that alias analysis can understand machine-dependent addresses.
1828
1829 The typical use of this macro is to handle addresses containing
1830 a label_ref or symbol_ref within an UNSPEC. */
1831
1832#define FIND_BASE_TERM(X) ix86_find_base_term (X)
1833
1834/* Nonzero if the constant value X is a legitimate general operand
1835 when generating PIC code. It is given that flag_pic is on and
1836 that X satisfies CONSTANT_P or is a CONST_DOUBLE. */
1837
1838#define LEGITIMATE_PIC_OPERAND_P(X) legitimate_pic_operand_p (X)
1839
1840#define STRIP_UNARY(X) (UNARY_P (X) ? XEXP (X, 0) : X)
1841
1842#define SYMBOLIC_CONST(X) \
1843 (GET_CODE (X) == SYMBOL_REF \
1844 || GET_CODE (X) == LABEL_REF \
1845 || (GET_CODE (X) == CONST && symbolic_reference_mentioned_p (X)))
1846
1847/* Max number of args passed in registers. If this is more than 3, we will
1848 have problems with ebx (register #4), since it is a caller save register and
1849 is also used as the pic register in ELF. So for now, don't allow more than
1850 3 registers to be passed in registers. */
1851
1852/* Abi specific values for REGPARM_MAX and SSE_REGPARM_MAX */
1853#define X86_64_REGPARM_MAX 6
1854#define X86_64_MS_REGPARM_MAX 4
1855
1856#define X86_32_REGPARM_MAX 3
1857
1858#define REGPARM_MAX \
1859 (TARGET_64BIT \
1860 ? (TARGET_64BIT_MS_ABI \
1861 ? X86_64_MS_REGPARM_MAX \
1862 : X86_64_REGPARM_MAX) \
1863 : X86_32_REGPARM_MAX)
1864
1865#define X86_64_SSE_REGPARM_MAX 8
1866#define X86_64_MS_SSE_REGPARM_MAX 4
1867
1868#define X86_32_SSE_REGPARM_MAX (TARGET_SSE ? (TARGET_MACHO ? 4 : 3) : 0)
1869
1870#define SSE_REGPARM_MAX \
1871 (TARGET_64BIT \
1872 ? (TARGET_64BIT_MS_ABI \
1873 ? X86_64_MS_SSE_REGPARM_MAX \
1874 : X86_64_SSE_REGPARM_MAX) \
1875 : X86_32_SSE_REGPARM_MAX)
1876
1877#define X86_32_MMX_REGPARM_MAX (TARGET_MMX ? (TARGET_MACHO ? 0 : 3) : 0)
1878
1879#define MMX_REGPARM_MAX (TARGET_64BIT ? 0 : X86_32_MMX_REGPARM_MAX)
1880
1881/* Specify the machine mode that this machine uses
1882 for the index in the tablejump instruction. */
1883#define CASE_VECTOR_MODE \
1884 (!TARGET_LP64 || (flag_pic && ix86_cmodel != CM_LARGE_PIC) ? SImode : DImode)
1885
1886/* Define this as 1 if `char' should by default be signed; else as 0. */
1887#define DEFAULT_SIGNED_CHAR 1
1888
1889/* The constant maximum number of bytes that a single instruction can
1890 move quickly between memory and registers or between two memory
1891 locations. */
1892#define MAX_MOVE_MAX 64
1893
1894/* Max number of bytes we can move from memory to memory in one
1895 reasonably fast instruction, as opposed to MOVE_MAX_PIECES which
1896 is the number of bytes at a time which we can move efficiently.
1897 MOVE_MAX_PIECES defaults to MOVE_MAX. */
1898
1899#define MOVE_MAX \
1900 ((TARGET_AVX512F \
1901 && (ix86_move_max == PVW_AVX512 \
1902 || ix86_store_max == PVW_AVX512)) \
1903 ? 64 \
1904 : ((TARGET_AVX \
1905 && (ix86_move_max >= PVW_AVX256 \
1906 || ix86_store_max >= PVW_AVX256)) \
1907 ? 32 \
1908 : ((TARGET_SSE2 \
1909 && TARGET_SSE_UNALIGNED_LOAD_OPTIMAL \
1910 && TARGET_SSE_UNALIGNED_STORE_OPTIMAL) \
1911 ? 16 : UNITS_PER_WORD)))
1912
1913/* STORE_MAX_PIECES is the number of bytes at a time that we can store
1914 efficiently. Allow 16/32/64 bytes only if inter-unit move is enabled
1915 since vec_duplicate enabled by inter-unit move is used to implement
1916 store_by_pieces of 16/32/64 bytes. */
1917#define STORE_MAX_PIECES \
1918 (TARGET_INTER_UNIT_MOVES_TO_VEC \
1919 ? ((TARGET_AVX512F && ix86_store_max == PVW_AVX512) \
1920 ? 64 \
1921 : ((TARGET_AVX \
1922 && ix86_store_max >= PVW_AVX256) \
1923 ? 32 \
1924 : ((TARGET_SSE2 \
1925 && TARGET_SSE_UNALIGNED_STORE_OPTIMAL) \
1926 ? 16 : UNITS_PER_WORD))) \
1927 : UNITS_PER_WORD)
1928
1929/* If a memory-to-memory move would take MOVE_RATIO or more simple
1930 move-instruction pairs, we will do a cpymem or libcall instead.
1931 Increasing the value will always make code faster, but eventually
1932 incurs high cost in increased code size.
1933
1934 If you don't define this, a reasonable default is used. */
1935
1936#define MOVE_RATIO(speed) ((speed) ? ix86_cost->move_ratio : 3)
1937
1938/* If a clear memory operation would take CLEAR_RATIO or more simple
1939 move-instruction sequences, we will do a clrmem or libcall instead. */
1940
1941#define CLEAR_RATIO(speed) ((speed) ? ix86_cost->clear_ratio : 2)
1942
1943/* Define if shifts truncate the shift count which implies one can
1944 omit a sign-extension or zero-extension of a shift count.
1945
1946 On i386, shifts do truncate the count. But bit test instructions
1947 take the modulo of the bit offset operand. */
1948
1949/* #define SHIFT_COUNT_TRUNCATED */
1950
1951/* A macro to update M and UNSIGNEDP when an object whose type is
1952 TYPE and which has the specified mode and signedness is to be
1953 stored in a register. This macro is only called when TYPE is a
1954 scalar type.
1955
1956 On i386 it is sometimes useful to promote HImode and QImode
1957 quantities to SImode. The choice depends on target type. */
1958
1959#define PROMOTE_MODE(MODE, UNSIGNEDP, TYPE) \
1960do { \
1961 if (((MODE) == HImode && TARGET_PROMOTE_HI_REGS) \
1962 || ((MODE) == QImode && TARGET_PROMOTE_QI_REGS)) \
1963 (MODE) = SImode; \
1964} while (0)
1965
1966/* Specify the machine mode that pointers have.
1967 After generation of rtl, the compiler makes no further distinction
1968 between pointers and any other objects of this machine mode. */
1969#define Pmode (ix86_pmode == PMODE_DI ? DImode : SImode)
1970
1971/* Supply a definition of STACK_SAVEAREA_MODE for emit_stack_save.
1972 NONLOCAL needs space to save both shadow stack and stack pointers.
1973
1974 FIXME: We only need to save and restore stack pointer in ptr_mode.
1975 But expand_builtin_setjmp_setup and expand_builtin_longjmp use Pmode
1976 to save and restore stack pointer. See
1977 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84150
1978 */
1979#define STACK_SAVEAREA_MODE(LEVEL) \
1980 ((LEVEL) == SAVE_NONLOCAL ? (TARGET_64BIT ? TImode : DImode) : Pmode)
1981
1982/* Specify the machine_mode of the size increment
1983 operand of an 'allocate_stack' named pattern. */
1984#define STACK_SIZE_MODE Pmode
1985
1986/* A C expression whose value is zero if pointers that need to be extended
1987 from being `POINTER_SIZE' bits wide to `Pmode' are sign-extended and
1988 greater then zero if they are zero-extended and less then zero if the
1989 ptr_extend instruction should be used. */
1990
1991#define POINTERS_EXTEND_UNSIGNED 1
1992
1993/* A function address in a call instruction
1994 is a byte address (for indexing purposes)
1995 so give the MEM rtx a byte's mode. */
1996#define FUNCTION_MODE QImode
1997
1998
1999/* A C expression for the cost of a branch instruction. A value of 1
2000 is the default; other values are interpreted relative to that. */
2001
2002#define BRANCH_COST(speed_p, predictable_p) \
2003 (!(speed_p) ? 2 : (predictable_p) ? 0 : ix86_branch_cost)
2004
2005/* An integer expression for the size in bits of the largest integer machine
2006 mode that should actually be used. We allow pairs of registers. */
2007#define MAX_FIXED_MODE_SIZE GET_MODE_BITSIZE (TARGET_64BIT ? TImode : DImode)
2008
2009/* Define this macro as a C expression which is nonzero if accessing
2010 less than a word of memory (i.e. a `char' or a `short') is no
2011 faster than accessing a word of memory, i.e., if such access
2012 require more than one instruction or if there is no difference in
2013 cost between byte and (aligned) word loads.
2014
2015 When this macro is not defined, the compiler will access a field by
2016 finding the smallest containing object; when it is defined, a
2017 fullword load will be used if alignment permits. Unless bytes
2018 accesses are faster than word accesses, using word accesses is
2019 preferable since it may eliminate subsequent memory access if
2020 subsequent accesses occur to other fields in the same word of the
2021 structure, but to different bytes. */
2022
2023#define SLOW_BYTE_ACCESS 0
2024
2025/* Define this macro if it is as good or better to call a constant
2026 function address than to call an address kept in a register.
2027
2028 Desirable on the 386 because a CALL with a constant address is
2029 faster than one with a register address. */
2030
2031#define NO_FUNCTION_CSE 1
2032
2033/* Given a comparison code (EQ, NE, etc.) and the first operand of a COMPARE,
2034 return the mode to be used for the comparison.
2035
2036 For floating-point equality comparisons, CCFPEQmode should be used.
2037 VOIDmode should be used in all other cases.
2038
2039 For integer comparisons against zero, reduce to CCNOmode or CCZmode if
2040 possible, to allow for more combinations. */
2041
2042#define SELECT_CC_MODE(OP, X, Y) ix86_cc_mode ((OP), (X), (Y))
2043
2044/* Return nonzero if MODE implies a floating point inequality can be
2045 reversed. */
2046
2047#define REVERSIBLE_CC_MODE(MODE) 1
2048
2049/* A C expression whose value is reversed condition code of the CODE for
2050 comparison done in CC_MODE mode. */
2051#define REVERSE_CONDITION(CODE, MODE) ix86_reverse_condition ((CODE), (MODE))
2052
2053
2054/* Control the assembler format that we output, to the extent
2055 this does not vary between assemblers. */
2056
2057/* How to refer to registers in assembler output.
2058 This sequence is indexed by compiler's hard-register-number (see above). */
2059
2060/* In order to refer to the first 8 regs as 32-bit regs, prefix an "e".
2061 For non floating point regs, the following are the HImode names.
2062
2063 For float regs, the stack top is sometimes referred to as "%st(0)"
2064 instead of just "%st". TARGET_PRINT_OPERAND handles this with the
2065 "y" code. */
2066
2067#define HI_REGISTER_NAMES \
2068{"ax","dx","cx","bx","si","di","bp","sp", \
2069 "st","st(1)","st(2)","st(3)","st(4)","st(5)","st(6)","st(7)", \
2070 "argp", "flags", "fpsr", "frame", \
2071 "xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7", \
2072 "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7", \
2073 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15", \
2074 "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15", \
2075 "xmm16", "xmm17", "xmm18", "xmm19", \
2076 "xmm20", "xmm21", "xmm22", "xmm23", \
2077 "xmm24", "xmm25", "xmm26", "xmm27", \
2078 "xmm28", "xmm29", "xmm30", "xmm31", \
2079 "k0", "k1", "k2", "k3", "k4", "k5", "k6", "k7", \
2080 "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23", \
2081 "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31" }
2082
2083#define REGISTER_NAMES HI_REGISTER_NAMES
2084
2085#define QI_REGISTER_NAMES \
2086{"al", "dl", "cl", "bl", "sil", "dil", "bpl", "spl"}
2087
2088#define QI_HIGH_REGISTER_NAMES \
2089{"ah", "dh", "ch", "bh"}
2090
2091/* Table of additional register names to use in user input. */
2092
2093#define ADDITIONAL_REGISTER_NAMES \
2094{ \
2095 { "eax", AX_REG }, { "edx", DX_REG }, { "ecx", CX_REG }, { "ebx", BX_REG }, \
2096 { "esi", SI_REG }, { "edi", DI_REG }, { "ebp", BP_REG }, { "esp", SP_REG }, \
2097 { "rax", AX_REG }, { "rdx", DX_REG }, { "rcx", CX_REG }, { "rbx", BX_REG }, \
2098 { "rsi", SI_REG }, { "rdi", DI_REG }, { "rbp", BP_REG }, { "rsp", SP_REG }, \
2099 { "al", AX_REG }, { "dl", DX_REG }, { "cl", CX_REG }, { "bl", BX_REG }, \
2100 { "sil", SI_REG }, { "dil", DI_REG }, { "bpl", BP_REG }, { "spl", SP_REG }, \
2101 { "ah", AX_REG }, { "dh", DX_REG }, { "ch", CX_REG }, { "bh", BX_REG }, \
2102 { "ymm0", XMM0_REG }, { "ymm1", XMM1_REG }, { "ymm2", XMM2_REG }, { "ymm3", XMM3_REG }, \
2103 { "ymm4", XMM4_REG }, { "ymm5", XMM5_REG }, { "ymm6", XMM6_REG }, { "ymm7", XMM7_REG }, \
2104 { "ymm8", XMM8_REG }, { "ymm9", XMM9_REG }, { "ymm10", XMM10_REG }, { "ymm11", XMM11_REG }, \
2105 { "ymm12", XMM12_REG }, { "ymm13", XMM13_REG }, { "ymm14", XMM14_REG }, { "ymm15", XMM15_REG }, \
2106 { "ymm16", XMM16_REG }, { "ymm17", XMM17_REG }, { "ymm18", XMM18_REG }, { "ymm19", XMM19_REG }, \
2107 { "ymm20", XMM20_REG }, { "ymm21", XMM21_REG }, { "ymm22", XMM22_REG }, { "ymm23", XMM23_REG }, \
2108 { "ymm24", XMM24_REG }, { "ymm25", XMM25_REG }, { "ymm26", XMM26_REG }, { "ymm27", XMM27_REG }, \
2109 { "ymm28", XMM28_REG }, { "ymm29", XMM29_REG }, { "ymm30", XMM30_REG }, { "ymm31", XMM31_REG }, \
2110 { "zmm0", XMM0_REG }, { "zmm1", XMM1_REG }, { "zmm2", XMM2_REG }, { "zmm3", XMM3_REG }, \
2111 { "zmm4", XMM4_REG }, { "zmm5", XMM5_REG }, { "zmm6", XMM6_REG }, { "zmm7", XMM7_REG }, \
2112 { "zmm8", XMM8_REG }, { "zmm9", XMM9_REG }, { "zmm10", XMM10_REG }, { "zmm11", XMM11_REG }, \
2113 { "zmm12", XMM12_REG }, { "zmm13", XMM13_REG }, { "zmm14", XMM14_REG }, { "zmm15", XMM15_REG }, \
2114 { "zmm16", XMM16_REG }, { "zmm17", XMM17_REG }, { "zmm18", XMM18_REG }, { "zmm19", XMM19_REG }, \
2115 { "zmm20", XMM20_REG }, { "zmm21", XMM21_REG }, { "zmm22", XMM22_REG }, { "zmm23", XMM23_REG }, \
2116 { "zmm24", XMM24_REG }, { "zmm25", XMM25_REG }, { "zmm26", XMM26_REG }, { "zmm27", XMM27_REG }, \
2117 { "zmm28", XMM28_REG }, { "zmm29", XMM29_REG }, { "zmm30", XMM30_REG }, { "zmm31", XMM31_REG } \
2118}
2119
2120/* How to renumber registers for gdb. */
2121
2122#define DEBUGGER_REGNO(N) \
2123 (TARGET_64BIT ? debugger64_register_map[(N)] : debugger_register_map[(N)])
2124
2125extern unsigned int const debugger_register_map[FIRST_PSEUDO_REGISTER];
2126extern unsigned int const debugger64_register_map[FIRST_PSEUDO_REGISTER];
2127extern unsigned int const svr4_debugger_register_map[FIRST_PSEUDO_REGISTER];
2128
2129/* Before the prologue, RA is at 0(%esp). */
2130#define INCOMING_RETURN_ADDR_RTX \
2131 gen_rtx_MEM (Pmode, stack_pointer_rtx)
2132
2133/* After the prologue, RA is at -4(AP) in the current frame. */
2134#define RETURN_ADDR_RTX(COUNT, FRAME) \
2135 ((COUNT) == 0 \
2136 ? gen_rtx_MEM (Pmode, plus_constant (Pmode, arg_pointer_rtx, \
2137 -UNITS_PER_WORD)) \
2138 : gen_rtx_MEM (Pmode, plus_constant (Pmode, (FRAME), UNITS_PER_WORD)))
2139
2140/* PC is dbx register 8; let's use that column for RA. */
2141#define DWARF_FRAME_RETURN_COLUMN (TARGET_64BIT ? 16 : 8)
2142
2143/* Before the prologue, there are return address and error code for
2144 exception handler on the top of the frame. */
2145#define INCOMING_FRAME_SP_OFFSET \
2146 (cfun->machine->func_type == TYPE_EXCEPTION \
2147 ? 2 * UNITS_PER_WORD : UNITS_PER_WORD)
2148
2149/* The value of INCOMING_FRAME_SP_OFFSET the assembler assumes in
2150 .cfi_startproc. */
2151#define DEFAULT_INCOMING_FRAME_SP_OFFSET UNITS_PER_WORD
2152
2153/* Describe how we implement __builtin_eh_return. */
2154#define EH_RETURN_DATA_REGNO(N) ((N) <= DX_REG ? (N) : INVALID_REGNUM)
2155#define EH_RETURN_STACKADJ_RTX gen_rtx_REG (Pmode, CX_REG)
2156
2157
2158/* Select a format to encode pointers in exception handling data. CODE
2159 is 0 for data, 1 for code labels, 2 for function pointers. GLOBAL is
2160 true if the symbol may be affected by dynamic relocations.
2161
2162 ??? All x86 object file formats are capable of representing this.
2163 After all, the relocation needed is the same as for the call insn.
2164 Whether or not a particular assembler allows us to enter such, I
2165 guess we'll have to see. */
2166#define ASM_PREFERRED_EH_DATA_FORMAT(CODE, GLOBAL) \
2167 asm_preferred_eh_data_format ((CODE), (GLOBAL))
2168
2169/* These are a couple of extensions to the formats accepted
2170 by asm_fprintf:
2171 %z prints out opcode suffix for word-mode instruction
2172 %r prints out word-mode name for reg_names[arg] */
2173#define ASM_FPRINTF_EXTENSIONS(FILE, ARGS, P) \
2174 case 'z': \
2175 fputc (TARGET_64BIT ? 'q' : 'l', (FILE)); \
2176 break; \
2177 \
2178 case 'r': \
2179 { \
2180 unsigned int regno = va_arg ((ARGS), int); \
2181 if (LEGACY_INT_REGNO_P (regno)) \
2182 fputc (TARGET_64BIT ? 'r' : 'e', (FILE)); \
2183 fputs (reg_names[regno], (FILE)); \
2184 break; \
2185 }
2186
2187/* This is how to output an insn to push a register on the stack. */
2188
2189#define ASM_OUTPUT_REG_PUSH(FILE, REGNO) \
2190 asm_fprintf ((FILE), "\tpush%z\t%%%r\n", (REGNO))
2191
2192/* This is how to output an insn to pop a register from the stack. */
2193
2194#define ASM_OUTPUT_REG_POP(FILE, REGNO) \
2195 asm_fprintf ((FILE), "\tpop%z\t%%%r\n", (REGNO))
2196
2197/* This is how to output an element of a case-vector that is absolute. */
2198
2199#define ASM_OUTPUT_ADDR_VEC_ELT(FILE, VALUE) \
2200 ix86_output_addr_vec_elt ((FILE), (VALUE))
2201
2202/* This is how to output an element of a case-vector that is relative. */
2203
2204#define ASM_OUTPUT_ADDR_DIFF_ELT(FILE, BODY, VALUE, REL) \
2205 ix86_output_addr_diff_elt ((FILE), (VALUE), (REL))
2206
2207/* When we see %v, we will print the 'v' prefix if TARGET_AVX is true. */
2208
2209#define ASM_OUTPUT_AVX_PREFIX(STREAM, PTR) \
2210{ \
2211 if ((PTR)[0] == '%' && (PTR)[1] == 'v') \
2212 (PTR) += TARGET_AVX ? 1 : 2; \
2213}
2214
2215/* A C statement or statements which output an assembler instruction
2216 opcode to the stdio stream STREAM. The macro-operand PTR is a
2217 variable of type `char *' which points to the opcode name in
2218 its "internal" form--the form that is written in the machine
2219 description. */
2220
2221#define ASM_OUTPUT_OPCODE(STREAM, PTR) \
2222 ASM_OUTPUT_AVX_PREFIX ((STREAM), (PTR))
2223
2224/* A C statement to output to the stdio stream FILE an assembler
2225 command to pad the location counter to a multiple of 1<<LOG
2226 bytes if it is within MAX_SKIP bytes. */
2227
2228#ifdef HAVE_GAS_MAX_SKIP_P2ALIGN
2229# define ASM_OUTPUT_MAX_SKIP_ALIGN(FILE,LOG,MAX_SKIP) \
2230 do { \
2231 if ((LOG) != 0) { \
2232 if ((MAX_SKIP) == 0 || (MAX_SKIP) >= (1 << (LOG)) - 1) \
2233 fprintf ((FILE), "\t.p2align %d\n", (LOG)); \
2234 else \
2235 fprintf ((FILE), "\t.p2align %d,,%d\n", (LOG), (MAX_SKIP)); \
2236 } \
2237 } while (0)
2238#endif
2239
2240/* Write the extra assembler code needed to declare a function
2241 properly. */
2242
2243#undef ASM_OUTPUT_FUNCTION_LABEL
2244#define ASM_OUTPUT_FUNCTION_LABEL(FILE, NAME, DECL) \
2245 ix86_asm_output_function_label ((FILE), (NAME), (DECL))
2246
2247/* A C statement (sans semicolon) to output a reference to SYMBOL_REF SYM.
2248 If not defined, assemble_name will be used to output the name of the
2249 symbol. This macro may be used to modify the way a symbol is referenced
2250 depending on information encoded by TARGET_ENCODE_SECTION_INFO. */
2251
2252#ifndef ASM_OUTPUT_SYMBOL_REF
2253#define ASM_OUTPUT_SYMBOL_REF(FILE, SYM) \
2254 do { \
2255 const char *name \
2256 = assemble_name_resolve (XSTR (SYM, 0)); \
2257 /* In -masm=att wrap identifiers that start with $ \
2258 into parens. */ \
2259 if (ASSEMBLER_DIALECT == ASM_ATT \
2260 && name[0] == '$' \
2261 && user_label_prefix[0] == '\0') \
2262 { \
2263 fputc ('(', (FILE)); \
2264 assemble_name_raw ((FILE), name); \
2265 fputc (')', (FILE)); \
2266 } \
2267 else \
2268 assemble_name_raw ((FILE), name); \
2269 } while (0)
2270#endif
2271
2272/* In Intel syntax, we have to quote user-defined labels that would
2273 match (unprefixed) registers or operators. */
2274
2275#undef ASM_OUTPUT_LABELREF
2276#define ASM_OUTPUT_LABELREF(STREAM, NAME) \
2277 ix86_asm_output_labelref ((STREAM), user_label_prefix, (NAME))
2278
2279/* Under some conditions we need jump tables in the text section,
2280 because the assembler cannot handle label differences between
2281 sections. */
2282
2283#define JUMP_TABLES_IN_TEXT_SECTION \
2284 (flag_pic && !(TARGET_64BIT || HAVE_AS_GOTOFF_IN_DATA))
2285
2286/* Switch to init or fini section via SECTION_OP, emit a call to FUNC,
2287 and switch back. For x86 we do this only to save a few bytes that
2288 would otherwise be unused in the text section. */
2289#define CRT_MKSTR2(VAL) #VAL
2290#define CRT_MKSTR(x) CRT_MKSTR2(x)
2291
2292#define CRT_CALL_STATIC_FUNCTION(SECTION_OP, FUNC) \
2293 asm (SECTION_OP "\n\t" \
2294 "call " CRT_MKSTR(__USER_LABEL_PREFIX__) #FUNC "\n" \
2295 TEXT_SECTION_ASM_OP);
2296
2297/* Default threshold for putting data in large sections
2298 with x86-64 medium memory model */
2299#define DEFAULT_LARGE_SECTION_THRESHOLD 65536
2300
2301/* Which processor to tune code generation for. These must be in sync
2302 with processor_cost_table in i386-options.cc. */
2303
2304#define GOT_ALIAS_SET ix86_GOT_alias_set ()
2305
2306enum processor_type
2307{
2308 PROCESSOR_GENERIC = 0,
2309 PROCESSOR_I386, /* 80386 */
2310 PROCESSOR_I486, /* 80486DX, 80486SX, 80486DX[24] */
2311 PROCESSOR_PENTIUM,
2312 PROCESSOR_LAKEMONT,
2313 PROCESSOR_PENTIUMPRO,
2314 PROCESSOR_PENTIUM4,
2315 PROCESSOR_NOCONA,
2316 PROCESSOR_CORE2,
2317 PROCESSOR_NEHALEM,
2318 PROCESSOR_SANDYBRIDGE,
2319 PROCESSOR_HASWELL,
2320 PROCESSOR_BONNELL,
2321 PROCESSOR_SILVERMONT,
2322 PROCESSOR_GOLDMONT,
2323 PROCESSOR_GOLDMONT_PLUS,
2324 PROCESSOR_TREMONT,
2325 PROCESSOR_SIERRAFOREST,
2326 PROCESSOR_GRANDRIDGE,
2327 PROCESSOR_CLEARWATERFOREST,
2328 PROCESSOR_SKYLAKE,
2329 PROCESSOR_SKYLAKE_AVX512,
2330 PROCESSOR_CANNONLAKE,
2331 PROCESSOR_ICELAKE_CLIENT,
2332 PROCESSOR_ICELAKE_SERVER,
2333 PROCESSOR_CASCADELAKE,
2334 PROCESSOR_TIGERLAKE,
2335 PROCESSOR_COOPERLAKE,
2336 PROCESSOR_SAPPHIRERAPIDS,
2337 PROCESSOR_ALDERLAKE,
2338 PROCESSOR_ROCKETLAKE,
2339 PROCESSOR_GRANITERAPIDS,
2340 PROCESSOR_GRANITERAPIDS_D,
2341 PROCESSOR_ARROWLAKE,
2342 PROCESSOR_ARROWLAKE_S,
2343 PROCESSOR_PANTHERLAKE,
2344 PROCESSOR_DIAMONDRAPIDS,
2345 PROCESSOR_INTEL,
2346 PROCESSOR_LUJIAZUI,
2347 PROCESSOR_YONGFENG,
2348 PROCESSOR_SHIJIDADAO,
2349 PROCESSOR_GEODE,
2350 PROCESSOR_K6,
2351 PROCESSOR_ATHLON,
2352 PROCESSOR_K8,
2353 PROCESSOR_AMDFAM10,
2354 PROCESSOR_BDVER1,
2355 PROCESSOR_BDVER2,
2356 PROCESSOR_BDVER3,
2357 PROCESSOR_BDVER4,
2358 PROCESSOR_BTVER1,
2359 PROCESSOR_BTVER2,
2360 PROCESSOR_ZNVER1,
2361 PROCESSOR_ZNVER2,
2362 PROCESSOR_ZNVER3,
2363 PROCESSOR_ZNVER4,
2364 PROCESSOR_ZNVER5,
2365 PROCESSOR_max
2366};
2367
2368#if !defined(IN_LIBGCC2) && !defined(IN_TARGET_LIBS) && !defined(IN_RTS)
2369extern const char *const processor_names[];
2370
2371#include "wide-int-bitmask.h"
2372
2373enum pta_flag
2374{
2375#define DEF_PTA(NAME) _ ## NAME,
2376#include "i386-isa.def"
2377#undef DEF_PTA
2378 END_PTA
2379};
2380
2381/* wide_int_bitmask can handle only 128 flags. */
2382STATIC_ASSERT (END_PTA <= 128);
2383
2384#define WIDE_INT_BITMASK_FROM_NTH(N) (N < 64 ? wide_int_bitmask (0, 1ULL << N) \
2385 : wide_int_bitmask (1ULL << (N - 64), 0))
2386
2387#define DEF_PTA(NAME) constexpr wide_int_bitmask PTA_ ## NAME \
2388 = WIDE_INT_BITMASK_FROM_NTH ((pta_flag) _ ## NAME);
2389#include "i386-isa.def"
2390#undef DEF_PTA
2391
2392constexpr wide_int_bitmask PTA_X86_64_BASELINE = PTA_64BIT | PTA_MMX | PTA_SSE
2393 | PTA_SSE2 | PTA_NO_SAHF | PTA_FXSR;
2394constexpr wide_int_bitmask PTA_X86_64_V2 = (PTA_X86_64_BASELINE
2395 & (~PTA_NO_SAHF))
2396 | PTA_CX16 | PTA_POPCNT | PTA_SSE3 | PTA_SSE4_1 | PTA_SSE4_2 | PTA_SSSE3;
2397constexpr wide_int_bitmask PTA_X86_64_V3 = PTA_X86_64_V2
2398 | PTA_AVX | PTA_AVX2 | PTA_BMI | PTA_BMI2 | PTA_F16C | PTA_FMA | PTA_LZCNT
2399 | PTA_MOVBE | PTA_XSAVE;
2400constexpr wide_int_bitmask PTA_X86_64_V4 = PTA_X86_64_V3
2401 | PTA_AVX512F | PTA_AVX512BW | PTA_AVX512CD | PTA_AVX512DQ | PTA_AVX512VL;
2402
2403constexpr wide_int_bitmask PTA_CORE2 = PTA_64BIT | PTA_MMX | PTA_SSE | PTA_SSE2
2404 | PTA_SSE3 | PTA_SSSE3 | PTA_CX16 | PTA_FXSR;
2405constexpr wide_int_bitmask PTA_NEHALEM = PTA_CORE2 | PTA_SSE4_1 | PTA_SSE4_2
2406 | PTA_POPCNT;
2407constexpr wide_int_bitmask PTA_WESTMERE = PTA_NEHALEM | PTA_PCLMUL;
2408constexpr wide_int_bitmask PTA_SANDYBRIDGE = PTA_WESTMERE | PTA_AVX | PTA_XSAVE
2409 | PTA_XSAVEOPT;
2410constexpr wide_int_bitmask PTA_IVYBRIDGE = PTA_SANDYBRIDGE | PTA_FSGSBASE
2411 | PTA_RDRND | PTA_F16C;
2412constexpr wide_int_bitmask PTA_HASWELL = PTA_IVYBRIDGE | PTA_AVX2 | PTA_BMI
2413 | PTA_BMI2 | PTA_LZCNT | PTA_FMA | PTA_MOVBE | PTA_HLE;
2414constexpr wide_int_bitmask PTA_BROADWELL = PTA_HASWELL | PTA_ADX | PTA_RDSEED
2415 | PTA_PRFCHW;
2416constexpr wide_int_bitmask PTA_SKYLAKE = PTA_BROADWELL | PTA_AES
2417 | PTA_CLFLUSHOPT | PTA_XSAVEC | PTA_XSAVES | PTA_SGX;
2418constexpr wide_int_bitmask PTA_SKYLAKE_AVX512 = PTA_SKYLAKE | PTA_AVX512F
2419 | PTA_AVX512CD | PTA_AVX512VL | PTA_AVX512BW | PTA_AVX512DQ | PTA_PKU
2420 | PTA_CLWB;
2421constexpr wide_int_bitmask PTA_CASCADELAKE = PTA_SKYLAKE_AVX512
2422 | PTA_AVX512VNNI;
2423constexpr wide_int_bitmask PTA_COOPERLAKE = PTA_CASCADELAKE | PTA_AVX512BF16;
2424constexpr wide_int_bitmask PTA_CANNONLAKE = PTA_SKYLAKE | PTA_AVX512F
2425 | PTA_AVX512CD | PTA_AVX512VL | PTA_AVX512BW | PTA_AVX512DQ | PTA_PKU
2426 | PTA_AVX512VBMI | PTA_AVX512IFMA | PTA_SHA;
2427constexpr wide_int_bitmask PTA_ICELAKE_CLIENT = PTA_CANNONLAKE | PTA_AVX512VNNI
2428 | PTA_GFNI | PTA_VAES | PTA_AVX512VBMI2 | PTA_VPCLMULQDQ | PTA_AVX512BITALG
2429 | PTA_RDPID | PTA_AVX512VPOPCNTDQ;
2430constexpr wide_int_bitmask PTA_ROCKETLAKE = PTA_ICELAKE_CLIENT & ~PTA_SGX;
2431constexpr wide_int_bitmask PTA_ICELAKE_SERVER = PTA_ICELAKE_CLIENT
2432 | PTA_PCONFIG | PTA_WBNOINVD | PTA_CLWB;
2433constexpr wide_int_bitmask PTA_TIGERLAKE = PTA_ICELAKE_CLIENT | PTA_MOVDIRI
2434 | PTA_MOVDIR64B | PTA_CLWB | PTA_AVX512VP2INTERSECT | PTA_KL | PTA_WIDEKL;
2435constexpr wide_int_bitmask PTA_SAPPHIRERAPIDS = PTA_ICELAKE_SERVER | PTA_MOVDIRI
2436 | PTA_MOVDIR64B | PTA_ENQCMD | PTA_CLDEMOTE | PTA_PTWRITE | PTA_WAITPKG
2437 | PTA_SERIALIZE | PTA_TSXLDTRK | PTA_AMX_TILE | PTA_AMX_INT8 | PTA_AMX_BF16
2438 | PTA_UINTR | PTA_AVXVNNI | PTA_AVX512FP16 | PTA_AVX512BF16;
2439constexpr wide_int_bitmask PTA_BONNELL = PTA_CORE2 | PTA_MOVBE;
2440constexpr wide_int_bitmask PTA_SILVERMONT = PTA_WESTMERE | PTA_MOVBE
2441 | PTA_RDRND | PTA_PRFCHW;
2442constexpr wide_int_bitmask PTA_GOLDMONT = PTA_SILVERMONT | PTA_AES | PTA_SHA
2443 | PTA_XSAVE | PTA_RDSEED | PTA_XSAVEC | PTA_XSAVES | PTA_CLFLUSHOPT
2444 | PTA_XSAVEOPT | PTA_FSGSBASE;
2445constexpr wide_int_bitmask PTA_GOLDMONT_PLUS = PTA_GOLDMONT | PTA_RDPID
2446 | PTA_SGX | PTA_PTWRITE;
2447constexpr wide_int_bitmask PTA_TREMONT = PTA_GOLDMONT_PLUS | PTA_CLWB
2448 | PTA_GFNI | PTA_MOVDIRI | PTA_MOVDIR64B | PTA_CLDEMOTE | PTA_WAITPKG;
2449constexpr wide_int_bitmask PTA_ALDERLAKE = PTA_TREMONT | PTA_ADX | PTA_AVX
2450 | PTA_AVX2 | PTA_BMI | PTA_BMI2 | PTA_F16C | PTA_FMA | PTA_LZCNT
2451 | PTA_PCONFIG | PTA_PKU | PTA_VAES | PTA_VPCLMULQDQ | PTA_SERIALIZE
2452 | PTA_HRESET | PTA_KL | PTA_WIDEKL | PTA_AVXVNNI;
2453constexpr wide_int_bitmask PTA_SIERRAFOREST = PTA_ALDERLAKE | PTA_AVXIFMA
2454 | PTA_AVXVNNIINT8 | PTA_AVXNECONVERT | PTA_CMPCCXADD | PTA_ENQCMD | PTA_UINTR;
2455constexpr wide_int_bitmask PTA_GRANITERAPIDS = PTA_SAPPHIRERAPIDS | PTA_AMX_FP16
2456 | PTA_PREFETCHI | PTA_AVX10_1;
2457constexpr wide_int_bitmask PTA_GRANITERAPIDS_D = PTA_GRANITERAPIDS
2458 | PTA_AMX_COMPLEX;
2459constexpr wide_int_bitmask PTA_GRANDRIDGE = PTA_SIERRAFOREST;
2460constexpr wide_int_bitmask PTA_ARROWLAKE = PTA_ALDERLAKE | PTA_AVXIFMA
2461 | PTA_AVXVNNIINT8 | PTA_AVXNECONVERT | PTA_CMPCCXADD | PTA_UINTR;
2462constexpr wide_int_bitmask PTA_ARROWLAKE_S = PTA_ARROWLAKE | PTA_AVXVNNIINT16
2463 | PTA_SHA512 | PTA_SM3 | PTA_SM4;
2464constexpr wide_int_bitmask PTA_CLEARWATERFOREST = PTA_SIERRAFOREST
2465 | PTA_AVXVNNIINT16 | PTA_SHA512 | PTA_SM3 | PTA_SM4 | PTA_USER_MSR
2466 | PTA_PREFETCHI;
2467constexpr wide_int_bitmask PTA_PANTHERLAKE = PTA_ARROWLAKE_S | PTA_PREFETCHI;
2468constexpr wide_int_bitmask PTA_DIAMONDRAPIDS = PTA_GRANITERAPIDS_D
2469 | PTA_AVXIFMA | PTA_AVXNECONVERT | PTA_AVXVNNIINT16 | PTA_AVXVNNIINT8
2470 | PTA_CMPCCXADD | PTA_SHA512 | PTA_SM3 | PTA_SM4 | PTA_AVX10_2
2471 | PTA_APX_F | PTA_AMX_AVX512 | PTA_AMX_FP8 | PTA_AMX_TF32 | PTA_AMX_TRANSPOSE
2472 | PTA_MOVRS | PTA_AMX_MOVRS | PTA_USER_MSR;
2473
2474constexpr wide_int_bitmask PTA_BDVER1 = PTA_64BIT | PTA_MMX | PTA_SSE
2475 | PTA_SSE2 | PTA_SSE3 | PTA_SSE4A | PTA_CX16 | PTA_ABM | PTA_SSSE3
2476 | PTA_SSE4_1 | PTA_SSE4_2 | PTA_AES | PTA_PCLMUL | PTA_AVX | PTA_FMA4
2477 | PTA_XOP | PTA_LWP | PTA_PRFCHW | PTA_FXSR | PTA_XSAVE;
2478constexpr wide_int_bitmask PTA_BDVER2 = PTA_BDVER1 | PTA_BMI | PTA_TBM
2479 | PTA_F16C | PTA_FMA;
2480constexpr wide_int_bitmask PTA_BDVER3 = PTA_BDVER2 | PTA_XSAVEOPT
2481 | PTA_FSGSBASE;
2482constexpr wide_int_bitmask PTA_BDVER4 = PTA_BDVER3 | PTA_AVX2 | PTA_BMI2
2483 | PTA_RDRND | PTA_MOVBE | PTA_MWAITX;
2484
2485constexpr wide_int_bitmask PTA_ZNVER1 = PTA_64BIT | PTA_MMX | PTA_SSE | PTA_SSE2
2486 | PTA_SSE3 | PTA_SSE4A | PTA_CX16 | PTA_ABM | PTA_SSSE3 | PTA_SSE4_1
2487 | PTA_SSE4_2 | PTA_AES | PTA_PCLMUL | PTA_AVX | PTA_AVX2 | PTA_BMI | PTA_BMI2
2488 | PTA_F16C | PTA_FMA | PTA_PRFCHW | PTA_FXSR | PTA_XSAVE | PTA_XSAVEOPT
2489 | PTA_FSGSBASE | PTA_RDRND | PTA_MOVBE | PTA_MWAITX | PTA_ADX | PTA_RDSEED
2490 | PTA_CLZERO | PTA_CLFLUSHOPT | PTA_XSAVEC | PTA_XSAVES | PTA_SHA | PTA_LZCNT
2491 | PTA_POPCNT;
2492constexpr wide_int_bitmask PTA_ZNVER2 = PTA_ZNVER1 | PTA_CLWB | PTA_RDPID
2493 | PTA_WBNOINVD;
2494constexpr wide_int_bitmask PTA_ZNVER3 = PTA_ZNVER2 | PTA_VAES | PTA_VPCLMULQDQ
2495 | PTA_PKU;
2496constexpr wide_int_bitmask PTA_ZNVER4 = PTA_ZNVER3 | PTA_AVX512F | PTA_AVX512DQ
2497 | PTA_AVX512IFMA | PTA_AVX512CD | PTA_AVX512BW | PTA_AVX512VL
2498 | PTA_AVX512BF16 | PTA_AVX512VBMI | PTA_AVX512VBMI2 | PTA_GFNI
2499 | PTA_AVX512VNNI | PTA_AVX512BITALG | PTA_AVX512VPOPCNTDQ;
2500constexpr wide_int_bitmask PTA_ZNVER5 = PTA_ZNVER4 | PTA_AVXVNNI
2501 | PTA_MOVDIRI | PTA_MOVDIR64B | PTA_AVX512VP2INTERSECT | PTA_PREFETCHI;
2502
2503constexpr wide_int_bitmask PTA_BTVER1 = PTA_64BIT | PTA_MMX | PTA_SSE
2504 | PTA_SSE2 | PTA_SSE3 | PTA_SSSE3 | PTA_SSE4A | PTA_ABM | PTA_CX16
2505 | PTA_PRFCHW | PTA_FXSR | PTA_XSAVE;
2506constexpr wide_int_bitmask PTA_BTVER2 = PTA_BTVER1 | PTA_SSE4_1 | PTA_SSE4_2
2507 | PTA_AES | PTA_PCLMUL | PTA_AVX | PTA_BMI | PTA_F16C | PTA_MOVBE
2508 | PTA_XSAVEOPT;
2509
2510constexpr wide_int_bitmask PTA_LUJIAZUI = PTA_64BIT | PTA_MMX | PTA_SSE
2511 | PTA_SSE2 | PTA_SSE3 | PTA_CX16 | PTA_ABM | PTA_SSSE3 | PTA_SSE4_1
2512 | PTA_SSE4_2 | PTA_AES | PTA_PCLMUL | PTA_BMI | PTA_BMI2 | PTA_PRFCHW
2513 | PTA_FXSR | PTA_XSAVE | PTA_XSAVEOPT | PTA_FSGSBASE | PTA_RDRND | PTA_MOVBE
2514 | PTA_ADX | PTA_RDSEED | PTA_POPCNT;
2515constexpr wide_int_bitmask PTA_YONGFENG = PTA_LUJIAZUI | PTA_AVX | PTA_AVX2
2516 | PTA_F16C | PTA_FMA | PTA_SHA | PTA_LZCNT;
2517
2518#ifndef GENERATOR_FILE
2519
2520#include "insn-attr-common.h"
2521
2522#include "common/config/i386/i386-cpuinfo.h"
2523
2524class pta
2525{
2526public:
2527 const char *const name; /* processor name or nickname. */
2528 const enum processor_type processor;
2529 const enum attr_cpu schedule;
2530 const wide_int_bitmask flags;
2531 const int model;
2532 const enum feature_priority priority;
2533};
2534
2535extern const pta processor_alias_table[];
2536extern unsigned int const pta_size;
2537extern unsigned int const num_arch_names;
2538#endif
2539
2540#endif
2541
2542extern enum processor_type ix86_tune;
2543extern enum processor_type ix86_arch;
2544
2545/* Size of the RED_ZONE area. */
2546#define RED_ZONE_SIZE 128
2547/* Reserved area of the red zone for temporaries. */
2548#define RED_ZONE_RESERVE 8
2549
2550extern unsigned int ix86_preferred_stack_boundary;
2551extern unsigned int ix86_incoming_stack_boundary;
2552
2553/* Smallest class containing REGNO. */
2554extern enum reg_class const regclass_map[FIRST_PSEUDO_REGISTER];
2555
2556enum ix86_fpcmp_strategy {
2557 IX86_FPCMP_SAHF,
2558 IX86_FPCMP_COMI,
2559 IX86_FPCMP_ARITH
2560};
2561
2562/* To properly truncate FP values into integers, we need to set i387 control
2563 word. We can't emit proper mode switching code before reload, as spills
2564 generated by reload may truncate values incorrectly, but we still can avoid
2565 redundant computation of new control word by the mode switching pass.
2566 The fldcw instructions are still emitted redundantly, but this is probably
2567 not going to be noticeable problem, as most CPUs do have fast path for
2568 the sequence.
2569
2570 The machinery is to emit simple truncation instructions and split them
2571 before reload to instructions having USEs of two memory locations that
2572 are filled by this code to old and new control word.
2573
2574 Post-reload pass may be later used to eliminate the redundant fildcw if
2575 needed. */
2576
2577enum ix86_stack_slot
2578{
2579 SLOT_CW_STORED = 0,
2580 SLOT_CW_ROUNDEVEN,
2581 SLOT_CW_TRUNC,
2582 SLOT_CW_FLOOR,
2583 SLOT_CW_CEIL,
2584 SLOT_STV_TEMP,
2585 SLOT_FLOATxFDI_387,
2586 MAX_386_STACK_LOCALS
2587};
2588
2589enum ix86_entity
2590{
2591 X86_DIRFLAG = 0,
2592 AVX_U128,
2593 I387_ROUNDEVEN,
2594 I387_TRUNC,
2595 I387_FLOOR,
2596 I387_CEIL,
2597 MAX_386_ENTITIES
2598};
2599
2600enum x86_dirflag_state
2601{
2602 X86_DIRFLAG_RESET,
2603 X86_DIRFLAG_ANY
2604};
2605
2606enum avx_u128_state
2607{
2608 AVX_U128_CLEAN,
2609 AVX_U128_DIRTY,
2610 AVX_U128_ANY
2611};
2612
2613/* Define this macro if the port needs extra instructions inserted
2614 for mode switching in an optimizing compilation. */
2615
2616#define OPTIMIZE_MODE_SWITCHING(ENTITY) \
2617 ix86_optimize_mode_switching[(ENTITY)]
2618
2619/* If you define `OPTIMIZE_MODE_SWITCHING', you have to define this as
2620 initializer for an array of integers. Each initializer element N
2621 refers to an entity that needs mode switching, and specifies the
2622 number of different modes that might need to be set for this
2623 entity. The position of the initializer in the initializer -
2624 starting counting at zero - determines the integer that is used to
2625 refer to the mode-switched entity in question. */
2626
2627#define NUM_MODES_FOR_MODE_SWITCHING \
2628 { X86_DIRFLAG_ANY, AVX_U128_ANY, \
2629 I387_CW_ANY, I387_CW_ANY, I387_CW_ANY, I387_CW_ANY }
2630
2631
2632/* Avoid renaming of stack registers, as doing so in combination with
2633 scheduling just increases amount of live registers at time and in
2634 the turn amount of fxch instructions needed.
2635
2636 ??? Maybe Pentium chips benefits from renaming, someone can try....
2637
2638 Don't rename evex to non-evex sse registers. */
2639
2640#define HARD_REGNO_RENAME_OK(SRC, TARGET) \
2641 (!STACK_REGNO_P (SRC) \
2642 && EXT_REX_SSE_REGNO_P (SRC) == EXT_REX_SSE_REGNO_P (TARGET))
2643
2644
2645#define FASTCALL_PREFIX '@'
2646
2647#ifndef USED_FOR_TARGET
2648/* Structure describing stack frame layout.
2649 Stack grows downward:
2650
2651 [arguments]
2652 <- ARG_POINTER
2653 saved pc
2654
2655 saved static chain if ix86_static_chain_on_stack
2656
2657 saved frame pointer if frame_pointer_needed
2658 <- HARD_FRAME_POINTER
2659 [saved regs]
2660 <- reg_save_offset
2661 [padding0]
2662 <- stack_realign_offset
2663 [saved SSE regs]
2664 OR
2665 [stub-saved registers for ms x64 --> sysv clobbers
2666 <- Start of out-of-line, stub-saved/restored regs
2667 (see libgcc/config/i386/(sav|res)ms64*.S)
2668 [XMM6-15]
2669 [RSI]
2670 [RDI]
2671 [?RBX] only if RBX is clobbered
2672 [?RBP] only if RBP and RBX are clobbered
2673 [?R12] only if R12 and all previous regs are clobbered
2674 [?R13] only if R13 and all previous regs are clobbered
2675 [?R14] only if R14 and all previous regs are clobbered
2676 [?R15] only if R15 and all previous regs are clobbered
2677 <- end of stub-saved/restored regs
2678 [padding1]
2679 ]
2680 <- sse_reg_save_offset
2681 [padding2]
2682 | <- FRAME_POINTER
2683 [va_arg registers] |
2684 |
2685 [frame] |
2686 |
2687 [padding2] | = to_allocate
2688 <- STACK_POINTER
2689 */
2690struct GTY(()) ix86_frame
2691{
2692 int nsseregs;
2693 int nregs;
2694 int va_arg_size;
2695 int red_zone_size;
2696 int outgoing_arguments_size;
2697
2698 /* The offsets relative to ARG_POINTER. */
2699 HOST_WIDE_INT frame_pointer_offset;
2700 HOST_WIDE_INT hard_frame_pointer_offset;
2701 HOST_WIDE_INT stack_pointer_offset;
2702 HOST_WIDE_INT hfp_save_offset;
2703 HOST_WIDE_INT reg_save_offset;
2704 HOST_WIDE_INT stack_realign_allocate;
2705 HOST_WIDE_INT stack_realign_offset;
2706 HOST_WIDE_INT sse_reg_save_offset;
2707
2708 /* When save_regs_using_mov is set, emit prologue using
2709 move instead of push instructions. */
2710 bool save_regs_using_mov;
2711
2712 /* Assume without checking that:
2713 EXPENSIVE_P = expensive_function_p (EXPENSIVE_COUNT). */
2714 bool expensive_p;
2715 int expensive_count;
2716};
2717
2718/* Machine specific frame tracking during prologue/epilogue generation. All
2719 values are positive, but since the x86 stack grows downward, are subtratced
2720 from the CFA to produce a valid address. */
2721
2722struct GTY(()) machine_frame_state
2723{
2724 /* This pair tracks the currently active CFA as reg+offset. When reg
2725 is drap_reg, we don't bother trying to record here the real CFA when
2726 it might really be a DW_CFA_def_cfa_expression. */
2727 rtx cfa_reg;
2728 HOST_WIDE_INT cfa_offset;
2729
2730 /* The current offset (canonically from the CFA) of ESP and EBP.
2731 When stack frame re-alignment is active, these may not be relative
2732 to the CFA. However, in all cases they are relative to the offsets
2733 of the saved registers stored in ix86_frame. */
2734 HOST_WIDE_INT sp_offset;
2735 HOST_WIDE_INT fp_offset;
2736
2737 /* The size of the red-zone that may be assumed for the purposes of
2738 eliding register restore notes in the epilogue. This may be zero
2739 if no red-zone is in effect, or may be reduced from the real
2740 red-zone value by a maximum runtime stack re-alignment value. */
2741 int red_zone_offset;
2742
2743 /* Indicate whether each of ESP, EBP or DRAP currently holds a valid
2744 value within the frame. If false then the offset above should be
2745 ignored. Note that DRAP, if valid, *always* points to the CFA and
2746 thus has an offset of zero. */
2747 BOOL_BITFIELD sp_valid : 1;
2748 BOOL_BITFIELD fp_valid : 1;
2749 BOOL_BITFIELD drap_valid : 1;
2750
2751 /* Indicate whether the local stack frame has been re-aligned. When
2752 set, the SP/FP offsets above are relative to the aligned frame
2753 and not the CFA. */
2754 BOOL_BITFIELD realigned : 1;
2755
2756 /* Indicates whether the stack pointer has been re-aligned. When set,
2757 SP/FP continue to be relative to the CFA, but the stack pointer
2758 should only be used for offsets > sp_realigned_offset, while
2759 the frame pointer should be used for offsets <= sp_realigned_fp_last.
2760 The flags realigned and sp_realigned are mutually exclusive. */
2761 BOOL_BITFIELD sp_realigned : 1;
2762
2763 /* When APX_PPX used in prologue, force epilogue to emit
2764 popp instead of move and leave. */
2765 BOOL_BITFIELD apx_ppx_used : 1;
2766
2767 /* If sp_realigned is set, this is the last valid offset from the CFA
2768 that can be used for access with the frame pointer. */
2769 HOST_WIDE_INT sp_realigned_fp_last;
2770
2771 /* If sp_realigned is set, this is the offset from the CFA that the stack
2772 pointer was realigned, and may or may not be equal to sp_realigned_fp_last.
2773 Access via the stack pointer is only valid for offsets that are greater than
2774 this value. */
2775 HOST_WIDE_INT sp_realigned_offset;
2776};
2777
2778/* Private to winnt.cc. */
2779struct seh_frame_state;
2780
2781enum function_type
2782{
2783 TYPE_UNKNOWN = 0,
2784 TYPE_NORMAL,
2785 /* The current function is an interrupt service routine with a
2786 pointer argument as specified by the "interrupt" attribute. */
2787 TYPE_INTERRUPT,
2788 /* The current function is an interrupt service routine with a
2789 pointer argument and an integer argument as specified by the
2790 "interrupt" attribute. */
2791 TYPE_EXCEPTION
2792};
2793
2794enum call_saved_registers_type
2795{
2796 TYPE_DEFAULT_CALL_SAVED_REGISTERS = 0,
2797 /* The current function is a function specified with the "interrupt"
2798 or "no_caller_saved_registers" attribute. */
2799 TYPE_NO_CALLER_SAVED_REGISTERS,
2800 /* The current function is a function specified with the
2801 "no_callee_saved_registers" attribute. */
2802 TYPE_NO_CALLEE_SAVED_REGISTERS,
2803 /* The current function is a function specified with the "noreturn"
2804 attribute. */
2805 TYPE_NO_CALLEE_SAVED_REGISTERS_EXCEPT_BP,
2806};
2807
2808enum queued_insn_type
2809{
2810 TYPE_NONE = 0,
2811 TYPE_ENDBR,
2812 TYPE_PATCHABLE_AREA
2813};
2814
2815struct GTY(()) machine_function {
2816 struct stack_local_entry *stack_locals;
2817 int varargs_gpr_size;
2818 int varargs_fpr_size;
2819 int optimize_mode_switching[MAX_386_ENTITIES];
2820
2821 /* Cached initial frame layout for the current function. */
2822 struct ix86_frame frame;
2823
2824 /* For -fsplit-stack support: A stack local which holds a pointer to
2825 the stack arguments for a function with a variable number of
2826 arguments. This is set at the start of the function and is used
2827 to initialize the overflow_arg_area field of the va_list
2828 structure. */
2829 rtx split_stack_varargs_pointer;
2830
2831 /* This value is used for amd64 targets and specifies the current abi
2832 to be used. MS_ABI means ms abi. Otherwise SYSV_ABI means sysv abi. */
2833 ENUM_BITFIELD(calling_abi) call_abi : 8;
2834
2835 /* Nonzero if the function accesses a previous frame. */
2836 BOOL_BITFIELD accesses_prev_frame : 1;
2837
2838 /* Set by ix86_compute_frame_layout and used by prologue/epilogue
2839 expander to determine the style used. */
2840 BOOL_BITFIELD use_fast_prologue_epilogue : 1;
2841
2842 /* Nonzero if the current function calls pc thunk and
2843 must not use the red zone. */
2844 BOOL_BITFIELD pc_thunk_call_expanded : 1;
2845
2846 /* If true, the current function needs the default PIC register, not
2847 an alternate register (on x86) and must not use the red zone (on
2848 x86_64), even if it's a leaf function. We don't want the
2849 function to be regarded as non-leaf because TLS calls need not
2850 affect register allocation. This flag is set when a TLS call
2851 instruction is expanded within a function, and never reset, even
2852 if all such instructions are optimized away. Use the
2853 ix86_current_function_calls_tls_descriptor macro for a better
2854 approximation. */
2855 BOOL_BITFIELD tls_descriptor_call_expanded_p : 1;
2856
2857 /* If true, the current function has a STATIC_CHAIN is placed on the
2858 stack below the return address. */
2859 BOOL_BITFIELD static_chain_on_stack : 1;
2860
2861 /* If true, it is safe to not save/restore DRAP register. */
2862 BOOL_BITFIELD no_drap_save_restore : 1;
2863
2864 /* Function type. */
2865 ENUM_BITFIELD(function_type) func_type : 2;
2866
2867 /* How to generate indirec branch. */
2868 ENUM_BITFIELD(indirect_branch) indirect_branch_type : 3;
2869
2870 /* If true, the current function has local indirect jumps, like
2871 "indirect_jump" or "tablejump". */
2872 BOOL_BITFIELD has_local_indirect_jump : 1;
2873
2874 /* How to generate function return. */
2875 ENUM_BITFIELD(indirect_branch) function_return_type : 3;
2876
2877 /* Call saved registers type. */
2878 ENUM_BITFIELD(call_saved_registers_type) call_saved_registers : 2;
2879
2880 /* If true, there is register available for argument passing. This
2881 is used only in ix86_function_ok_for_sibcall by 32-bit to determine
2882 if there is scratch register available for indirect sibcall. In
2883 64-bit, rax, r10 and r11 are scratch registers which aren't used to
2884 pass arguments and can be used for indirect sibcall. */
2885 BOOL_BITFIELD arg_reg_available : 1;
2886
2887 /* If true, we're out-of-lining reg save/restore for regs clobbered
2888 by 64-bit ms_abi functions calling a sysv_abi function. */
2889 BOOL_BITFIELD call_ms2sysv : 1;
2890
2891 /* If true, the incoming 16-byte aligned stack has an offset (of 8) and
2892 needs padding prior to out-of-line stub save/restore area. */
2893 BOOL_BITFIELD call_ms2sysv_pad_in : 1;
2894
2895 /* This is the number of extra registers saved by stub (valid range is
2896 0-6). Each additional register is only saved/restored by the stubs
2897 if all successive ones are. (Will always be zero when using a hard
2898 frame pointer.) */
2899 unsigned int call_ms2sysv_extra_regs:3;
2900
2901 /* Nonzero if the function places outgoing arguments on stack. */
2902 BOOL_BITFIELD outgoing_args_on_stack : 1;
2903
2904 /* If true, ENDBR or patchable area is queued at function entrance. */
2905 ENUM_BITFIELD(queued_insn_type) insn_queued_at_entrance : 2;
2906
2907 /* If true, the function label has been emitted. */
2908 BOOL_BITFIELD function_label_emitted : 1;
2909
2910 /* True if the function needs a stack frame. */
2911 BOOL_BITFIELD stack_frame_required : 1;
2912
2913 /* True if we should act silently, rather than raise an error for
2914 invalid calls. */
2915 BOOL_BITFIELD silent_p : 1;
2916
2917 /* True if red zone is used. */
2918 BOOL_BITFIELD red_zone_used : 1;
2919
2920 /* True if inline asm with redzone clobber has been seen. */
2921 BOOL_BITFIELD asm_redzone_clobber_seen : 1;
2922
2923 /* The largest alignment, in bytes, of stack slot actually used. */
2924 unsigned int max_used_stack_alignment;
2925
2926 /* During prologue/epilogue generation, the current frame state.
2927 Otherwise, the frame state at the end of the prologue. */
2928 struct machine_frame_state fs;
2929
2930 /* During SEH output, this is non-null. */
2931 struct seh_frame_state * GTY((skip(""))) seh;
2932};
2933
2934extern GTY(()) tree sysv_va_list_type_node;
2935extern GTY(()) tree ms_va_list_type_node;
2936#endif
2937
2938#define ix86_stack_locals (cfun->machine->stack_locals)
2939#define ix86_varargs_gpr_size (cfun->machine->varargs_gpr_size)
2940#define ix86_varargs_fpr_size (cfun->machine->varargs_fpr_size)
2941#define ix86_optimize_mode_switching (cfun->machine->optimize_mode_switching)
2942#define ix86_pc_thunk_call_expanded (cfun->machine->pc_thunk_call_expanded)
2943#define ix86_tls_descriptor_calls_expanded_in_cfun \
2944 (cfun->machine->tls_descriptor_call_expanded_p)
2945/* Since tls_descriptor_call_expanded is not cleared, even if all TLS
2946 calls are optimized away, we try to detect cases in which it was
2947 optimized away. Since such instructions (use (reg REG_SP)), we can
2948 verify whether there's any such instruction live by testing that
2949 REG_SP is live. */
2950#define ix86_current_function_calls_tls_descriptor \
2951 (ix86_tls_descriptor_calls_expanded_in_cfun && df_regs_ever_live_p (SP_REG))
2952#define ix86_static_chain_on_stack (cfun->machine->static_chain_on_stack)
2953#define ix86_red_zone_used (cfun->machine->red_zone_used)
2954
2955/* Control behavior of x86_file_start. */
2956#define X86_FILE_START_VERSION_DIRECTIVE false
2957#define X86_FILE_START_FLTUSED false
2958
2959/* Flag to mark data that is in the large address area. */
2960#define SYMBOL_FLAG_FAR_ADDR (SYMBOL_FLAG_MACH_DEP << 0)
2961#define SYMBOL_REF_FAR_ADDR_P(X) \
2962 ((SYMBOL_REF_FLAGS (X) & SYMBOL_FLAG_FAR_ADDR) != 0)
2963
2964/* Flags to mark dllimport/dllexport. Used by PE ports, but handy to
2965 have defined always, to avoid ifdefing. */
2966#define SYMBOL_FLAG_DLLIMPORT (SYMBOL_FLAG_MACH_DEP << 1)
2967#define SYMBOL_REF_DLLIMPORT_P(X) \
2968 ((SYMBOL_REF_FLAGS (X) & SYMBOL_FLAG_DLLIMPORT) != 0)
2969
2970#define SYMBOL_FLAG_DLLEXPORT (SYMBOL_FLAG_MACH_DEP << 2)
2971#define SYMBOL_REF_DLLEXPORT_P(X) \
2972 ((SYMBOL_REF_FLAGS (X) & SYMBOL_FLAG_DLLEXPORT) != 0)
2973
2974#define SYMBOL_FLAG_STUBVAR (SYMBOL_FLAG_MACH_DEP << 4)
2975#define SYMBOL_REF_STUBVAR_P(X) \
2976 ((SYMBOL_REF_FLAGS (X) & SYMBOL_FLAG_STUBVAR) != 0)
2977
2978extern void debug_ready_dispatch (void);
2979extern void debug_dispatch_window (int);
2980
2981/* The value at zero is only defined for the BMI instructions
2982 LZCNT and TZCNT, not the BSR/BSF insns in the original isa. */
2983#define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
2984 ((VALUE) = GET_MODE_BITSIZE (MODE), TARGET_BMI ? 2 : 0)
2985#define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
2986 ((VALUE) = GET_MODE_BITSIZE (MODE), TARGET_LZCNT ? 2 : 0)
2987
2988
2989/* Flags returned by ix86_get_callcvt (). */
2990#define IX86_CALLCVT_CDECL 0x1
2991#define IX86_CALLCVT_STDCALL 0x2
2992#define IX86_CALLCVT_FASTCALL 0x4
2993#define IX86_CALLCVT_THISCALL 0x8
2994#define IX86_CALLCVT_REGPARM 0x10
2995#define IX86_CALLCVT_SSEREGPARM 0x20
2996
2997#define IX86_BASE_CALLCVT(FLAGS) \
2998 ((FLAGS) & (IX86_CALLCVT_CDECL | IX86_CALLCVT_STDCALL \
2999 | IX86_CALLCVT_FASTCALL | IX86_CALLCVT_THISCALL))
3000
3001#define RECIP_MASK_NONE 0x00
3002#define RECIP_MASK_DIV 0x01
3003#define RECIP_MASK_SQRT 0x02
3004#define RECIP_MASK_VEC_DIV 0x04
3005#define RECIP_MASK_VEC_SQRT 0x08
3006#define RECIP_MASK_ALL (RECIP_MASK_DIV | RECIP_MASK_SQRT \
3007 | RECIP_MASK_VEC_DIV | RECIP_MASK_VEC_SQRT)
3008#define RECIP_MASK_DEFAULT (RECIP_MASK_VEC_DIV | RECIP_MASK_VEC_SQRT)
3009
3010#define TARGET_RECIP_DIV ((recip_mask & RECIP_MASK_DIV) != 0)
3011#define TARGET_RECIP_SQRT ((recip_mask & RECIP_MASK_SQRT) != 0)
3012#define TARGET_RECIP_VEC_DIV ((recip_mask & RECIP_MASK_VEC_DIV) != 0)
3013#define TARGET_RECIP_VEC_SQRT ((recip_mask & RECIP_MASK_VEC_SQRT) != 0)
3014
3015/* Use 128-bit AVX instructions in the auto-vectorizer. */
3016#define TARGET_PREFER_AVX128 (prefer_vector_width_type == PVW_AVX128)
3017/* Use 256-bit AVX instructions in the auto-vectorizer. */
3018#define TARGET_PREFER_AVX256 (TARGET_PREFER_AVX128 \
3019 || prefer_vector_width_type == PVW_AVX256)
3020
3021#define TARGET_INDIRECT_BRANCH_REGISTER \
3022 (ix86_indirect_branch_register \
3023 || cfun->machine->indirect_branch_type != indirect_branch_keep)
3024
3025#define IX86_HLE_ACQUIRE (1 << 16)
3026#define IX86_HLE_RELEASE (1 << 17)
3027
3028/* For switching between functions with different target attributes. */
3029#define SWITCHABLE_TARGET 1
3030
3031#define TARGET_SUPPORTS_WIDE_INT 1
3032
3033#if !defined(GENERATOR_FILE) && !defined(IN_LIBGCC2)
3034extern enum attr_cpu ix86_schedule;
3035
3036#define NUM_X86_64_MS_CLOBBERED_REGS 12
3037#endif
3038
3039/* __builtin_eh_return can't handle stack realignment, so disable MMX/SSE
3040 in 32-bit libgcc functions that call it. */
3041#ifndef __x86_64__
3042#define LIBGCC2_UNWIND_ATTRIBUTE __attribute__((target ("no-mmx,no-sse")))
3043#endif
3044
3045/*
3046Local variables:
3047version-control: t
3048End:
3049*/
3050

Provided by KDAB

Privacy Policy
Update your C++ knowledge – Modern C++11/14/17 Training
Find out more

source code of gcc/config/i386/i386.h