1 /* Subroutines for the gcc driver.
2 Copyright (C) 2006, 2007 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3, or (at your option)
11 GCC is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
22 #include "coretypes.h"
26 const char *host_detect_local_cpu (int argc
, const char **argv
);
31 /* Returns parameters that describe L1_ASSOC associative cache of size
32 L1_SIZEKB with lines of size L1_LINE. */
35 describe_cache (unsigned l1_sizekb
, unsigned l1_line
,
36 unsigned l1_assoc ATTRIBUTE_UNUSED
)
38 char size
[100], line
[100];
40 /* At the moment, gcc middle-end does not use the information about the
41 associativity of the cache. */
43 sprintf (size
, "--param l1-cache-size=%u", l1_sizekb
);
44 sprintf (line
, "--param l1-cache-line-size=%u", l1_line
);
46 return concat (size
, " ", line
, " ", NULL
);
49 /* Returns the description of caches for an AMD processor. */
52 detect_caches_amd (unsigned max_ext_level
)
54 unsigned eax
, ebx
, ecx
, edx
;
55 unsigned l1_sizekb
, l1_line
, l1_assoc
;
57 if (max_ext_level
< 0x80000005)
60 __cpuid (0x80000005, eax
, ebx
, ecx
, edx
);
63 l1_sizekb
= (ecx
>> 24) & 0xff;
64 l1_assoc
= (ecx
>> 16) & 0xff;
66 return describe_cache (l1_sizekb
, l1_line
, l1_assoc
);
69 /* Stores the size of the L1 cache and cache line, and the associativity
70 of the cache according to REG to L1_SIZEKB, L1_LINE and L1_ASSOC. */
73 decode_caches_intel (unsigned reg
, unsigned *l1_sizekb
, unsigned *l1_line
,
78 if (((reg
>> 31) & 1) != 0)
81 for (i
= 0; i
< 4; i
++)
130 /* Returns the description of caches for an intel processor. */
133 detect_caches_intel (unsigned max_level
)
135 unsigned eax
, ebx
, ecx
, edx
;
136 unsigned l1_sizekb
= 0, l1_line
= 0, assoc
= 0;
141 __cpuid (2, eax
, ebx
, ecx
, edx
);
143 decode_caches_intel (eax
, &l1_sizekb
, &l1_line
, &assoc
);
144 decode_caches_intel (ebx
, &l1_sizekb
, &l1_line
, &assoc
);
145 decode_caches_intel (ecx
, &l1_sizekb
, &l1_line
, &assoc
);
146 decode_caches_intel (edx
, &l1_sizekb
, &l1_line
, &assoc
);
151 return describe_cache (l1_sizekb
, l1_line
, assoc
);
154 /* This will be called by the spec parser in gcc.c when it sees
155 a %:local_cpu_detect(args) construct. Currently it will be called
156 with either "arch" or "tune" as argument depending on if -march=native
157 or -mtune=native is to be substituted.
159 It returns a string containing new command line parameters to be
160 put at the place of the above two options, depending on what CPU
161 this is executed. E.g. "-march=k8" on an AMD64 machine
164 ARGC and ARGV are set depending on the actual arguments given
167 const char *host_detect_local_cpu (int argc
, const char **argv
)
169 enum processor_type processor
= PROCESSOR_I386
;
170 const char *cpu
= "i386";
172 const char *cache
= "";
173 const char *options
= "";
175 unsigned int eax
, ebx
, ecx
, edx
;
177 unsigned int max_level
, ext_level
;
181 unsigned int has_sse3
, has_ssse3
, has_cmpxchg16b
;
182 unsigned int has_cmpxchg8b
, has_cmov
, has_mmx
, has_sse
, has_sse2
;
184 /* Extended features */
185 unsigned int has_lahf_lm
= 0, has_sse4a
= 0;
186 unsigned int has_longmode
= 0, has_3dnowp
= 0, has_3dnow
= 0;
193 arch
= !strcmp (argv
[0], "arch");
195 if (!arch
&& strcmp (argv
[0], "tune"))
198 max_level
= __get_cpuid_max (0, &vendor
);
202 __cpuid (1, eax
, ebx
, ecx
, edx
);
204 /* We don't care for extended family. */
205 family
= (eax
>> 8) & 0x0f;
207 has_sse3
= ecx
& bit_SSE3
;
208 has_ssse3
= ecx
& bit_SSSE3
;
209 has_cmpxchg16b
= ecx
& bit_CMPXCHG16B
;
211 has_cmpxchg8b
= edx
& bit_CMPXCHG8B
;
212 has_cmov
= edx
& bit_CMOV
;
213 has_mmx
= edx
& bit_MMX
;
214 has_sse
= edx
& bit_SSE
;
215 has_sse2
= edx
& bit_SSE2
;
217 /* Check cpuid level of extended features. */
218 __cpuid (0x80000000, ext_level
, ebx
, ecx
, edx
);
220 if (ext_level
> 0x80000000)
222 __cpuid (0x80000001, eax
, ebx
, ecx
, edx
);
224 has_lahf_lm
= ecx
& bit_LAHF_LM
;
225 has_sse4a
= ecx
& bit_SSE4a
;
227 has_longmode
= edx
& bit_LM
;
228 has_3dnowp
= edx
& bit_3DNOWP
;
229 has_3dnow
= edx
& bit_3DNOW
;
234 if (vendor
== *(unsigned int*) "Auth")
235 cache
= detect_caches_amd (ext_level
);
236 else if (vendor
== *(unsigned int*) "Genu")
237 cache
= detect_caches_intel (max_level
);
240 if (vendor
== *(unsigned int*) "Auth")
242 processor
= PROCESSOR_PENTIUM
;
245 processor
= PROCESSOR_K6
;
247 processor
= PROCESSOR_ATHLON
;
248 if (has_sse2
|| has_longmode
)
249 processor
= PROCESSOR_K8
;
251 processor
= PROCESSOR_AMDFAM10
;
253 else if (vendor
== *(unsigned int*) "Geod")
254 processor
= PROCESSOR_GEODE
;
260 processor
= PROCESSOR_I486
;
263 processor
= PROCESSOR_PENTIUM
;
266 processor
= PROCESSOR_PENTIUMPRO
;
269 processor
= PROCESSOR_PENTIUM4
;
272 /* We have no idea. */
273 processor
= PROCESSOR_GENERIC32
;
285 case PROCESSOR_PENTIUM
:
291 case PROCESSOR_PENTIUMPRO
:
293 /* It is Core 2 Duo. */
298 /* It is Core Duo. */
301 /* It is Pentium M. */
304 /* It is Pentium III. */
307 /* It is Pentium II. */
310 /* Default to Pentium Pro. */
314 /* For -mtune, we default to -mtune=generic. */
317 case PROCESSOR_PENTIUM4
:
328 case PROCESSOR_GEODE
:
332 if (arch
&& has_3dnow
)
337 case PROCESSOR_ATHLON
:
344 if (arch
&& has_sse3
)
349 case PROCESSOR_AMDFAM10
:
354 /* Use something reasonable. */
372 else if (has_cmpxchg8b
)
382 options
= concat (options
, "-mcx16 ", NULL
);
384 options
= concat (options
, "-msahf ", NULL
);
388 return concat (cache
, "-m", argv
[0], "=", cpu
, " ", options
, NULL
);
392 /* If we aren't compiling with GCC we just provide a minimal
395 const char *host_detect_local_cpu (int argc
, const char **argv
)
403 arch
= !strcmp (argv
[0], "arch");
405 if (!arch
&& strcmp (argv
[0], "tune"))
410 /* FIXME: i386 is wrong for 64bit compiler. How can we tell if
411 we are generating 64bit or 32bit code? */
417 return concat ("-m", argv
[0], "=", cpu
, NULL
);
419 #endif /* __GNUC__ */