aboutsummaryrefslogtreecommitdiffstats
path: root/gcc-4.9/gcc/config/i386/x86-tune.def
blob: 215c71f4df44e09ba615caec304632c3093e6d92 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
/* Definitions of x86 tunable features.
   Copyright (C) 2013-2014 Free Software Foundation, Inc.

This file is part of GCC.

GCC is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3, or (at your option)
any later version.

GCC is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License and
a copy of the GCC Runtime Library Exception along with this program;
see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
<http://www.gnu.org/licenses/>.  */

/* Tuning for a given CPU XXXX consists of:
    - adding new CPU into:
	- adding PROCESSOR_XXX to processor_type (in i386.h)
	- possibly adding XXX into CPU attribute in i386.md
	- adding XXX to processor_alias_table (in i386.c)
    - introducing ix86_XXX_cost in i386.c
	- Stringop generation table can be build based on test_stringop
	- script (once rest of tuning is complete)
    - designing a scheduler model in
	- XXXX.md file
	- Updating ix86_issue_rate and ix86_adjust_cost in i386.md
	- possibly updating ia32_multipass_dfa_lookahead, ix86_sched_reorder
	  and ix86_sched_init_global if those tricks are needed.
    - Tunning the flags bellow. Those are split into sections and each
      section is very roughly ordered by importance.  */

/*****************************************************************************/
/* Scheduling flags. 					                     */
/*****************************************************************************/

/* X86_TUNE_SCHEDULE: Enable scheduling.  */
DEF_TUNE (X86_TUNE_SCHEDULE, "schedule",
          m_PENT | m_PPRO | m_CORE_ALL | m_BONNELL | m_SILVERMONT | m_INTEL 
	  | m_K6_GEODE | m_AMD_MULTIPLE | m_GENERIC)

/* X86_TUNE_PARTIAL_REG_DEPENDENCY: Enable more register renaming
   on modern chips.  Preffer stores affecting whole integer register
   over partial stores.  For example preffer MOVZBL or MOVQ to load 8bit
   value over movb.  */
DEF_TUNE (X86_TUNE_PARTIAL_REG_DEPENDENCY, "partial_reg_dependency",
          m_P4_NOCONA | m_CORE_ALL | m_BONNELL | m_SILVERMONT | m_INTEL
	  | m_AMD_MULTIPLE | m_GENERIC)

/* X86_TUNE_SSE_PARTIAL_REG_DEPENDENCY: This knob promotes all store
   destinations to be 128bit to allow register renaming on 128bit SSE units,
   but usually results in one extra microop on 64bit SSE units.
   Experimental results shows that disabling this option on P4 brings over 20%
   SPECfp regression, while enabling it on K8 brings roughly 2.4% regression
   that can be partly masked by careful scheduling of moves.  */
DEF_TUNE (X86_TUNE_SSE_PARTIAL_REG_DEPENDENCY, "sse_partial_reg_dependency",
          m_PPRO | m_P4_NOCONA | m_CORE_ALL | m_BONNELL | m_AMDFAM10
	  | m_BDVER | m_GENERIC)

/* X86_TUNE_SSE_SPLIT_REGS: Set for machines where the type and dependencies
   are resolved on SSE register parts instead of whole registers, so we may
   maintain just lower part of scalar values in proper format leaving the
   upper part undefined.  */
DEF_TUNE (X86_TUNE_SSE_SPLIT_REGS, "sse_split_regs", m_ATHLON_K8)

/* X86_TUNE_PARTIAL_FLAG_REG_STALL: this flag disables use of of flags
   set by instructions affecting just some flags (in particular shifts).
   This is because Core2 resolves dependencies on whole flags register
   and such sequences introduce false dependency on previous instruction
   setting full flags.

   The flags does not affect generation of INC and DEC that is controlled
   by X86_TUNE_USE_INCDEC.

   This flag may be dropped from generic once core2-corei5 machines are
   rare enough.  */
DEF_TUNE (X86_TUNE_PARTIAL_FLAG_REG_STALL, "partial_flag_reg_stall",
          m_CORE2 | m_GENERIC)

/* X86_TUNE_MOVX: Enable to zero extend integer registers to avoid
   partial dependencies.  */
DEF_TUNE (X86_TUNE_MOVX, "movx",
          m_PPRO | m_P4_NOCONA | m_CORE_ALL | m_BONNELL | m_SILVERMONT
	  | m_INTEL | m_GEODE | m_AMD_MULTIPLE  | m_GENERIC)

/* X86_TUNE_MEMORY_MISMATCH_STALL: Avoid partial stores that are followed by
   full sized loads.  */
DEF_TUNE (X86_TUNE_MEMORY_MISMATCH_STALL, "memory_mismatch_stall",
          m_P4_NOCONA | m_CORE_ALL | m_BONNELL | m_SILVERMONT | m_INTEL
	  | m_AMD_MULTIPLE | m_GENERIC)

/* X86_TUNE_FUSE_CMP_AND_BRANCH_32: Fuse compare with a subsequent
   conditional jump instruction for 32 bit TARGET.
   FIXME: revisit for generic.  */
DEF_TUNE (X86_TUNE_FUSE_CMP_AND_BRANCH_32, "fuse_cmp_and_branch_32",
          m_GENERIC | m_CORE_ALL | m_BDVER)

/* X86_TUNE_FUSE_CMP_AND_BRANCH_64: Fuse compare with a subsequent
   conditional jump instruction for TARGET_64BIT.
   FIXME: revisit for generic.  */
DEF_TUNE (X86_TUNE_FUSE_CMP_AND_BRANCH_64, "fuse_cmp_and_branch_64",
          m_GENERIC | m_NEHALEM | m_SANDYBRIDGE | m_HASWELL | m_BDVER)

/* X86_TUNE_FUSE_CMP_AND_BRANCH_SOFLAGS: Fuse compare with a
   subsequent conditional jump instruction when the condition jump
   check sign flag (SF) or overflow flag (OF).  */
DEF_TUNE (X86_TUNE_FUSE_CMP_AND_BRANCH_SOFLAGS, "fuse_cmp_and_branch_soflags",
          m_GENERIC | m_NEHALEM | m_SANDYBRIDGE | m_HASWELL | m_BDVER)

/* X86_TUNE_FUSE_ALU_AND_BRANCH: Fuse alu with a subsequent conditional
   jump instruction when the alu instruction produces the CCFLAG consumed by
   the conditional jump instruction. */
DEF_TUNE (X86_TUNE_FUSE_ALU_AND_BRANCH, "fuse_alu_and_branch",
          m_GENERIC | m_SANDYBRIDGE | m_HASWELL)

/* X86_TUNE_REASSOC_INT_TO_PARALLEL: Try to produce parallel computations
   during reassociation of integer computation.  */
DEF_TUNE (X86_TUNE_REASSOC_INT_TO_PARALLEL, "reassoc_int_to_parallel",
          m_BONNELL)

/* X86_TUNE_REASSOC_FP_TO_PARALLEL: Try to produce parallel computations
   during reassociation of fp computation.  */
DEF_TUNE (X86_TUNE_REASSOC_FP_TO_PARALLEL, "reassoc_fp_to_parallel",
          m_BONNELL | m_SILVERMONT | m_HASWELL | m_INTEL | m_BDVER1
	  | m_BDVER2 | m_GENERIC)

/*****************************************************************************/
/* Function prologue, epilogue and function calling sequences.               */
/*****************************************************************************/

/* X86_TUNE_ACCUMULATE_OUTGOING_ARGS: Allocate stack space for outgoing
   arguments in prologue/epilogue instead of separately for each call
   by push/pop instructions.
   This increase code size by about 5% in 32bit mode, less so in 64bit mode
   because parameters are passed in registers.  It is considerable
   win for targets without stack engine that prevents multple push operations
   to happen in parallel.

   FIXME: the flags is incorrectly enabled for amdfam10, Bulldozer,
   Bobcat and Generic.  This is because disabling it causes large
   regression on mgrid due to IRA limitation leading to unecessary
   use of the frame pointer in 32bit mode.  */
DEF_TUNE (X86_TUNE_ACCUMULATE_OUTGOING_ARGS, "accumulate_outgoing_args",
	  m_PPRO | m_P4_NOCONA | m_BONNELL | m_SILVERMONT | m_INTEL
	  | m_ATHLON_K8)

/* X86_TUNE_PROLOGUE_USING_MOVE: Do not use push/pop in prologues that are
   considered on critical path.  */
DEF_TUNE (X86_TUNE_PROLOGUE_USING_MOVE, "prologue_using_move",
          m_PPRO | m_ATHLON_K8)

/* X86_TUNE_PROLOGUE_USING_MOVE: Do not use push/pop in epilogues that are
   considered on critical path.  */
DEF_TUNE (X86_TUNE_EPILOGUE_USING_MOVE, "epilogue_using_move",
          m_PPRO | m_ATHLON_K8)

/* X86_TUNE_USE_LEAVE: Use "leave" instruction in epilogues where it fits.  */
DEF_TUNE (X86_TUNE_USE_LEAVE, "use_leave",
	  m_386 | m_CORE_ALL | m_K6_GEODE | m_AMD_MULTIPLE | m_GENERIC)

/* X86_TUNE_PUSH_MEMORY: Enable generation of "push mem" instructions.
   Some chips, like 486 and Pentium works faster with separate load
   and push instructions.  */
DEF_TUNE (X86_TUNE_PUSH_MEMORY, "push_memory",
          m_386 | m_P4_NOCONA | m_CORE_ALL | m_K6_GEODE | m_AMD_MULTIPLE
          | m_GENERIC)

/* X86_TUNE_SINGLE_PUSH: Enable if single push insn is preferred
   over esp subtraction.  */
DEF_TUNE (X86_TUNE_SINGLE_PUSH, "single_push", m_386 | m_486 | m_PENT
          | m_K6_GEODE)

/* X86_TUNE_DOUBLE_PUSH. Enable if double push insn is preferred
   over esp subtraction.  */
DEF_TUNE (X86_TUNE_DOUBLE_PUSH, "double_push", m_PENT | m_K6_GEODE)

/* X86_TUNE_SINGLE_POP: Enable if single pop insn is preferred
   over esp addition.  */
DEF_TUNE (X86_TUNE_SINGLE_POP, "single_pop", m_386 | m_486 | m_PENT | m_PPRO)

/* X86_TUNE_DOUBLE_POP: Enable if double pop insn is preferred
   over esp addition.  */
DEF_TUNE (X86_TUNE_DOUBLE_POP, "double_pop", m_PENT)

/*****************************************************************************/
/* Branch predictor tuning  		                                     */
/*****************************************************************************/

/* X86_TUNE_PAD_SHORT_FUNCTION: Make every function to be at least 4
   instructions long.  */
DEF_TUNE (X86_TUNE_PAD_SHORT_FUNCTION, "pad_short_function", m_BONNELL)

/* X86_TUNE_PAD_RETURNS: Place NOP before every RET that is a destination
   of conditional jump or directly preceded by other jump instruction.
   This is important for AND K8-AMDFAM10 because the branch prediction
   architecture expect at most one jump per 2 byte window.  Failing to
   pad returns leads to misaligned return stack.  */
DEF_TUNE (X86_TUNE_PAD_RETURNS, "pad_returns",
          m_ATHLON_K8 | m_AMDFAM10 | m_GENERIC)

/* X86_TUNE_FOUR_JUMP_LIMIT: Some CPU cores are not able to predict more
   than 4 branch instructions in the 16 byte window.  */
DEF_TUNE (X86_TUNE_FOUR_JUMP_LIMIT, "four_jump_limit",
          m_PPRO | m_P4_NOCONA | m_BONNELL | m_SILVERMONT | m_INTEL |
	  m_ATHLON_K8 | m_AMDFAM10)

/*****************************************************************************/
/* Integer instruction selection tuning                                      */
/*****************************************************************************/

/* X86_TUNE_SOFTWARE_PREFETCHING_BENEFICIAL: Enable software prefetching
   at -O3.  For the moment, the prefetching seems badly tuned for Intel
   chips.  */
DEF_TUNE (X86_TUNE_SOFTWARE_PREFETCHING_BENEFICIAL, "software_prefetching_beneficial",
          m_K6_GEODE | m_AMD_MULTIPLE)

/* X86_TUNE_LCP_STALL: Avoid an expensive length-changing prefix stall
   on 16-bit immediate moves into memory on Core2 and Corei7.  */
DEF_TUNE (X86_TUNE_LCP_STALL, "lcp_stall", m_CORE_ALL | m_GENERIC)

/* X86_TUNE_READ_MODIFY: Enable use of read-modify instructions such
   as "add mem, reg".  */
DEF_TUNE (X86_TUNE_READ_MODIFY, "read_modify", ~(m_PENT | m_PPRO))

/* X86_TUNE_USE_INCDEC: Enable use of inc/dec instructions.   */
DEF_TUNE (X86_TUNE_USE_INCDEC, "use_incdec",
          ~(m_P4_NOCONA | m_CORE_ALL | m_BONNELL | m_SILVERMONT | m_INTEL
	    | m_GENERIC))

/* X86_TUNE_INTEGER_DFMODE_MOVES: Enable if integer moves are preferred
   for DFmode copies */
DEF_TUNE (X86_TUNE_INTEGER_DFMODE_MOVES, "integer_dfmode_moves",
          ~(m_PPRO | m_P4_NOCONA | m_CORE_ALL | m_BONNELL | m_SILVERMONT
	    | m_INTEL | m_GEODE | m_AMD_MULTIPLE | m_GENERIC))

/* X86_TUNE_OPT_AGU: Optimize for Address Generation Unit. This flag
   will impact LEA instruction selection. */
DEF_TUNE (X86_TUNE_OPT_AGU, "opt_agu", m_BONNELL | m_SILVERMONT | m_INTEL)

/* X86_TUNE_AVOID_LEA_FOR_ADDR: Avoid lea for address computation.  */
DEF_TUNE (X86_TUNE_AVOID_LEA_FOR_ADDR, "avoid_lea_for_addr",
	  m_BONNELL | m_SILVERMONT)

/* X86_TUNE_SLOW_IMUL_IMM32_MEM: Imul of 32-bit constant and memory is
   vector path on AMD machines.
   FIXME: Do we need to enable this for core? */
DEF_TUNE (X86_TUNE_SLOW_IMUL_IMM32_MEM, "slow_imul_imm32_mem",
          m_K8 | m_AMDFAM10)

/* X86_TUNE_SLOW_IMUL_IMM8: Imul of 8-bit constant is vector path on AMD
   machines.
   FIXME: Do we need to enable this for core? */
DEF_TUNE (X86_TUNE_SLOW_IMUL_IMM8, "slow_imul_imm8",
          m_K8 | m_AMDFAM10)

/* X86_TUNE_AVOID_MEM_OPND_FOR_CMOVE: Try to avoid memory operands for
   a conditional move.  */
DEF_TUNE (X86_TUNE_AVOID_MEM_OPND_FOR_CMOVE, "avoid_mem_opnd_for_cmove",
	  m_BONNELL | m_SILVERMONT | m_INTEL)

/* X86_TUNE_SINGLE_STRINGOP: Enable use of single string operations, such
   as MOVS and STOS (without a REP prefix) to move/set sequences of bytes.  */
DEF_TUNE (X86_TUNE_SINGLE_STRINGOP, "single_stringop", m_386 | m_P4_NOCONA)

/* X86_TUNE_MISALIGNED_MOVE_STRING_PRO_EPILOGUES: Enable generation of
   compact prologues and epilogues by issuing a misaligned moves.  This
   requires target to handle misaligned moves and partial memory stalls
   reasonably well.
   FIXME: This may actualy be a win on more targets than listed here.  */
DEF_TUNE (X86_TUNE_MISALIGNED_MOVE_STRING_PRO_EPILOGUES,
	  "misaligned_move_string_pro_epilogues",
	  m_386 | m_486 | m_CORE_ALL | m_AMD_MULTIPLE | m_GENERIC)

/* X86_TUNE_USE_SAHF: Controls use of SAHF.  */
DEF_TUNE (X86_TUNE_USE_SAHF, "use_sahf",
          m_PPRO | m_P4_NOCONA | m_CORE_ALL | m_BONNELL | m_SILVERMONT
	  | m_INTEL | m_K6_GEODE | m_K8 | m_AMDFAM10 | m_BDVER | m_BTVER
	  | m_GENERIC)

/* X86_TUNE_USE_CLTD: Controls use of CLTD and CTQO instructions.  */
DEF_TUNE (X86_TUNE_USE_CLTD, "use_cltd",
	  ~(m_PENT | m_BONNELL | m_SILVERMONT | m_INTEL  | m_K6))

/* X86_TUNE_USE_BT: Enable use of BT (bit test) instructions.  */
DEF_TUNE (X86_TUNE_USE_BT, "use_bt",
          m_CORE_ALL | m_BONNELL | m_SILVERMONT | m_INTEL | m_AMD_MULTIPLE
	  | m_GENERIC)

/*****************************************************************************/
/* 387 instruction selection tuning                                          */
/*****************************************************************************/

/* X86_TUNE_USE_HIMODE_FIOP: Enables use of x87 instructions with 16bit
   integer operand.
   FIXME: Why this is disabled for modern chips?  */
DEF_TUNE (X86_TUNE_USE_HIMODE_FIOP, "use_himode_fiop",
          m_386 | m_486 | m_K6_GEODE)

/* X86_TUNE_USE_SIMODE_FIOP: Enables use of x87 instructions with 32bit
   integer operand.  */
DEF_TUNE (X86_TUNE_USE_SIMODE_FIOP, "use_simode_fiop",
          ~(m_PENT | m_PPRO | m_CORE_ALL | m_BONNELL | m_SILVERMONT
	    | m_INTEL | m_AMD_MULTIPLE | m_GENERIC))

/* X86_TUNE_USE_FFREEP: Use freep instruction instead of fstp.  */
DEF_TUNE (X86_TUNE_USE_FFREEP, "use_ffreep", m_AMD_MULTIPLE)

/* X86_TUNE_EXT_80387_CONSTANTS: Use fancy 80387 constants, such as PI.  */
DEF_TUNE (X86_TUNE_EXT_80387_CONSTANTS, "ext_80387_constants",
          m_PPRO | m_P4_NOCONA | m_CORE_ALL | m_BONNELL | m_SILVERMONT
	  | m_INTEL | m_K6_GEODE | m_ATHLON_K8 | m_GENERIC)

/*****************************************************************************/
/* SSE instruction selection tuning                                          */
/*****************************************************************************/

/* X86_TUNE_VECTORIZE_DOUBLE: Enable double precision vector
   instructions.  */
DEF_TUNE (X86_TUNE_VECTORIZE_DOUBLE, "vectorize_double", ~m_BONNELL)

/* X86_TUNE_GENERAL_REGS_SSE_SPILL: Try to spill general regs to SSE
   regs instead of memory.  */
DEF_TUNE (X86_TUNE_GENERAL_REGS_SSE_SPILL, "general_regs_sse_spill",
          m_CORE_ALL)

/* X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL: Use movups for misaligned loads instead
   of a sequence loading registers by parts.  */
DEF_TUNE (X86_TUNE_SSE_UNALIGNED_LOAD_OPTIMAL, "sse_unaligned_load_optimal",
          m_NEHALEM | m_SANDYBRIDGE | m_HASWELL | m_AMDFAM10 | m_BDVER
	  | m_BTVER | m_SILVERMONT | m_INTEL | m_GENERIC)

/* X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL: Use movups for misaligned stores instead
   of a sequence loading registers by parts.  */
DEF_TUNE (X86_TUNE_SSE_UNALIGNED_STORE_OPTIMAL, "sse_unaligned_store_optimal",
          m_NEHALEM | m_SANDYBRIDGE | m_HASWELL | m_BDVER | m_SILVERMONT
	  | m_INTEL | m_GENERIC)

/* Use packed single precision instructions where posisble.  I.e. movups instead
   of movupd.  */
DEF_TUNE (X86_TUNE_SSE_PACKED_SINGLE_INSN_OPTIMAL, "sse_packed_single_insn_optimal",
          m_BDVER)

/* X86_TUNE_SSE_TYPELESS_STORES: Always movaps/movups for 128bit stores.   */
DEF_TUNE (X86_TUNE_SSE_TYPELESS_STORES, "sse_typeless_stores",
	  m_AMD_MULTIPLE | m_CORE_ALL | m_GENERIC)

/* X86_TUNE_SSE_LOAD0_BY_PXOR: Always use pxor to load0 as opposed to
   xorps/xorpd and other variants.  */
DEF_TUNE (X86_TUNE_SSE_LOAD0_BY_PXOR, "sse_load0_by_pxor",
	  m_PPRO | m_P4_NOCONA | m_CORE_ALL | m_BDVER | m_BTVER | m_GENERIC)

/* X86_TUNE_INTER_UNIT_MOVES_TO_VEC: Enable moves in from integer
   to SSE registers.  If disabled, the moves will be done by storing
   the value to memory and reloading.  */
DEF_TUNE (X86_TUNE_INTER_UNIT_MOVES_TO_VEC, "inter_unit_moves_to_vec",
          ~(m_AMD_MULTIPLE | m_GENERIC))

/* X86_TUNE_INTER_UNIT_MOVES_TO_VEC: Enable moves in from SSE
   to integer registers.  If disabled, the moves will be done by storing
   the value to memory and reloading.  */
DEF_TUNE (X86_TUNE_INTER_UNIT_MOVES_FROM_VEC, "inter_unit_moves_from_vec",
          ~m_ATHLON_K8)

/* X86_TUNE_INTER_UNIT_CONVERSIONS: Enable float<->integer conversions
   to use both SSE and integer registers at a same time.
   FIXME: revisit importance of this for generic.  */
DEF_TUNE (X86_TUNE_INTER_UNIT_CONVERSIONS, "inter_unit_conversions",
          ~(m_AMDFAM10 | m_BDVER))

/* X86_TUNE_SPLIT_MEM_OPND_FOR_FP_CONVERTS: Try to split memory operand for
   fp converts to destination register.  */
DEF_TUNE (X86_TUNE_SPLIT_MEM_OPND_FOR_FP_CONVERTS, "split_mem_opnd_for_fp_converts",
          m_SILVERMONT | m_INTEL)

/* X86_TUNE_USE_VECTOR_FP_CONVERTS: Prefer vector packed SSE conversion
   from FP to FP.  This form of instructions avoids partial write to the
   destination.  */
DEF_TUNE (X86_TUNE_USE_VECTOR_FP_CONVERTS, "use_vector_fp_converts",
          m_AMDFAM10)

/* X86_TUNE_USE_VECTOR_CONVERTS: Prefer vector packed SSE conversion
   from integer to FP. */
DEF_TUNE (X86_TUNE_USE_VECTOR_CONVERTS, "use_vector_converts", m_AMDFAM10)

/* X86_TUNE_SLOW_SHUFB: Indicates tunings with slow pshufb instruction.  */
DEF_TUNE (X86_TUNE_SLOW_PSHUFB, "slow_pshufb",
          m_BONNELL | m_SILVERMONT | m_INTEL)

/* X86_TUNE_VECTOR_PARALLEL_EXECUTION: Indicates tunings with ability to
   execute 2 or more vector instructions in parallel.  */
DEF_TUNE (X86_TUNE_VECTOR_PARALLEL_EXECUTION, "vec_parallel",
          m_NEHALEM | m_SANDYBRIDGE | m_HASWELL)

/*****************************************************************************/
/* AVX instruction selection tuning (some of SSE flags affects AVX, too)     */
/*****************************************************************************/

/* X86_TUNE_AVX256_UNALIGNED_LOAD_OPTIMAL: if false, unaligned loads are
   split.  */
DEF_TUNE (X86_TUNE_AVX256_UNALIGNED_LOAD_OPTIMAL, "256_unaligned_load_optimal",
          ~(m_NEHALEM | m_SANDYBRIDGE | m_GENERIC))

/* X86_TUNE_AVX256_UNALIGNED_STORE_OPTIMAL: if false, unaligned stores are
   split.  */
DEF_TUNE (X86_TUNE_AVX256_UNALIGNED_STORE_OPTIMAL, "256_unaligned_store_optimal",
          ~(m_NEHALEM | m_SANDYBRIDGE | m_BDVER | m_GENERIC))

/* X86_TUNE_AVX128_OPTIMAL: Enable 128-bit AVX instruction generation for
   the auto-vectorizer.  */
DEF_TUNE (X86_TUNE_AVX128_OPTIMAL, "avx128_optimal", m_BDVER | m_BTVER2)

/*****************************************************************************/
/* Historical relics: tuning flags that helps a specific old CPU designs     */
/*****************************************************************************/

/* X86_TUNE_DOUBLE_WITH_ADD: Use add instead of sal to double value in
   an integer register.  */
DEF_TUNE (X86_TUNE_DOUBLE_WITH_ADD, "double_with_add", ~m_386)

/* X86_TUNE_ALWAYS_FANCY_MATH_387: controls use of fancy 387 operations,
   such as fsqrt, fprem, fsin, fcos, fsincos etc.
   Should be enabled for all targets that always has coprocesor.  */
DEF_TUNE (X86_TUNE_ALWAYS_FANCY_MATH_387, "always_fancy_math_387",
          ~(m_386 | m_486))

/* X86_TUNE_UNROLL_STRLEN: Produce (quite lame) unrolled sequence for
   inline strlen.  This affects only -minline-all-stringops mode. By
   default we always dispatch to a library since our internal strlen
   is bad.  */
DEF_TUNE (X86_TUNE_UNROLL_STRLEN, "unroll_strlen", ~m_386)

/* X86_TUNE_SHIFT1: Enables use of short encoding of "sal reg" instead of
   longer "sal $1, reg".  */
DEF_TUNE (X86_TUNE_SHIFT1, "shift1", ~m_486)

/* X86_TUNE_ZERO_EXTEND_WITH_AND: Use AND instruction instead
   of mozbl/movwl.  */
DEF_TUNE (X86_TUNE_ZERO_EXTEND_WITH_AND, "zero_extend_with_and",  m_486 | m_PENT)

/* X86_TUNE_PROMOTE_HIMODE_IMUL: Modern CPUs have same latency for HImode
   and SImode multiply, but 386 and 486 do HImode multiply faster.  */
DEF_TUNE (X86_TUNE_PROMOTE_HIMODE_IMUL, "promote_himode_imul",
          ~(m_386 | m_486))

/* X86_TUNE_FAST_PREFIX: Enable demoting some 32bit or 64bit arithmetic
   into 16bit/8bit when resulting sequence is shorter.  For example
   for "and $-65536, reg" to 16bit store of 0.  */
DEF_TUNE (X86_TUNE_FAST_PREFIX, "fast_prefix", ~(m_386 | m_486 | m_PENT))

/* X86_TUNE_READ_MODIFY_WRITE: Enable use of read modify write instructions
   such as "add $1, mem".  */
DEF_TUNE (X86_TUNE_READ_MODIFY_WRITE, "read_modify_write", ~m_PENT)

/* X86_TUNE_MOVE_M1_VIA_OR: On pentiums, it is faster to load -1 via OR
   than a MOV.  */
DEF_TUNE (X86_TUNE_MOVE_M1_VIA_OR, "move_m1_via_or", m_PENT)

/* X86_TUNE_NOT_UNPAIRABLE: NOT is not pairable on Pentium, while XOR is,
   but one byte longer.  */
DEF_TUNE (X86_TUNE_NOT_UNPAIRABLE, "not_unpairable", m_PENT)

/* X86_TUNE_PARTIAL_REG_STALL: Pentium pro, unlike later chips, handled
   use of partial registers by renaming.  This improved performance of 16bit
   code where upper halves of registers are not used.  It also leads to
   an penalty whenever a 16bit store is followed by 32bit use.  This flag
   disables production of such sequences in common cases.
   See also X86_TUNE_HIMODE_MATH.

   In current implementation the partial register stalls are not eliminated
   very well - they can be introduced via subregs synthesized by combine
   and can happen in caller/callee saving sequences.  */
DEF_TUNE (X86_TUNE_PARTIAL_REG_STALL, "partial_reg_stall", m_PPRO)

/* X86_TUNE_PROMOTE_QIMODE: When it is cheap, turn 8bit arithmetic to
   corresponding 32bit arithmetic.  */
DEF_TUNE (X86_TUNE_PROMOTE_QIMODE, "promote_qimode",
	  ~m_PPRO)

/* X86_TUNE_PROMOTE_HI_REGS: Same, but for 16bit artihmetic.  Again we avoid
   partial register stalls on PentiumPro targets. */
DEF_TUNE (X86_TUNE_PROMOTE_HI_REGS, "promote_hi_regs", m_PPRO)

/* X86_TUNE_HIMODE_MATH: Enable use of 16bit arithmetic.
   On PPro this flag is meant to avoid partial register stalls.  */
DEF_TUNE (X86_TUNE_HIMODE_MATH, "himode_math", ~m_PPRO)

/* X86_TUNE_SPLIT_LONG_MOVES: Avoid instructions moving immediates
   directly to memory.  */
DEF_TUNE (X86_TUNE_SPLIT_LONG_MOVES, "split_long_moves", m_PPRO)

/* X86_TUNE_USE_XCHGB: Use xchgb %rh,%rl instead of rolw/rorw $8,rx.  */
DEF_TUNE (X86_TUNE_USE_XCHGB, "use_xchgb", m_PENT4)

/* X86_TUNE_USE_MOV0: Use "mov $0, reg" instead of "xor reg, reg" to clear
   integer register.  */
DEF_TUNE (X86_TUNE_USE_MOV0, "use_mov0", m_K6)

/* X86_TUNE_NOT_VECTORMODE: On AMD K6, NOT is vector decoded with memory
   operand that cannot be represented using a modRM byte.  The XOR
   replacement is long decoded, so this split helps here as well.  */
DEF_TUNE (X86_TUNE_NOT_VECTORMODE, "not_vectormode", m_K6)

/* X86_TUNE_AVOID_VECTOR_DECODE: Enable splitters that avoid vector decoded
   forms of instructions on K8 targets.  */
DEF_TUNE (X86_TUNE_AVOID_VECTOR_DECODE, "avoid_vector_decode",
          m_K8)

/* X86_TUNE_AVOID_FALSE_DEP_FOR_BMI: Avoid false dependency
   for bit-manipulation instructions.  */
DEF_TUNE (X86_TUNE_AVOID_FALSE_DEP_FOR_BMI, "avoid_false_dep_for_bmi",
	  m_SANDYBRIDGE | m_HASWELL | m_INTEL | m_GENERIC)

/*****************************************************************************/
/* This never worked well before.                                            */
/*****************************************************************************/

/* X86_TUNE_BRANCH_PREDICTION_HINTS: Branch hints were put in P4 based
   on simulation result. But after P4 was made, no performance benefit
   was observed with branch hints.  It also increases the code size.
   As a result, icc never generates branch hints.  */
DEF_TUNE (X86_TUNE_BRANCH_PREDICTION_HINTS, "branch_prediction_hints", 0)

/* X86_TUNE_QIMODE_MATH: Enable use of 8bit arithmetic.  */
DEF_TUNE (X86_TUNE_QIMODE_MATH, "qimode_math", ~0)

/* X86_TUNE_PROMOTE_QI_REGS: This enables generic code that promotes all 8bit
   arithmetic to 32bit via PROMOTE_MODE macro.  This code generation scheme
   is usually used for RISC targets.  */
DEF_TUNE (X86_TUNE_PROMOTE_QI_REGS, "promote_qi_regs", 0)

/* X86_TUNE_ADJUST_UNROLL: This enables adjusting the unroll factor based
   on hardware capabilities. Bdver3 hardware has a loop buffer which makes
   unrolling small loop less important. For, such architectures we adjust
   the unroll factor so that the unrolled loop fits the loop buffer.  */
DEF_TUNE (X86_TUNE_ADJUST_UNROLL, "adjust_unroll_factor", m_BDVER3 | m_BDVER4)