1 | //===-- Host.cpp - Implement OS Host Detection ------------------*- C++ -*-===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | // |
9 | // This file implements the operating system Host detection. |
10 | // |
11 | //===----------------------------------------------------------------------===// |
12 | |
13 | #include "llvm/TargetParser/Host.h" |
14 | #include "llvm/ADT/SmallVector.h" |
15 | #include "llvm/ADT/StringMap.h" |
16 | #include "llvm/ADT/StringRef.h" |
17 | #include "llvm/ADT/StringSwitch.h" |
18 | #include "llvm/Config/llvm-config.h" |
19 | #include "llvm/Support/MemoryBuffer.h" |
20 | #include "llvm/Support/raw_ostream.h" |
21 | #include "llvm/TargetParser/RISCVTargetParser.h" |
22 | #include "llvm/TargetParser/Triple.h" |
23 | #include "llvm/TargetParser/X86TargetParser.h" |
24 | #include <string.h> |
25 | |
26 | // Include the platform-specific parts of this class. |
27 | #ifdef LLVM_ON_UNIX |
28 | #include "Unix/Host.inc" |
29 | #include <sched.h> |
30 | #endif |
31 | #ifdef _WIN32 |
32 | #include "Windows/Host.inc" |
33 | #endif |
34 | #ifdef _MSC_VER |
35 | #include <intrin.h> |
36 | #endif |
37 | #ifdef __MVS__ |
38 | #include "llvm/Support/BCD.h" |
39 | #endif |
40 | #if defined(__APPLE__) |
41 | #include <mach/host_info.h> |
42 | #include <mach/mach.h> |
43 | #include <mach/mach_host.h> |
44 | #include <mach/machine.h> |
45 | #include <sys/param.h> |
46 | #include <sys/sysctl.h> |
47 | #endif |
48 | #ifdef _AIX |
49 | #include <sys/systemcfg.h> |
50 | #endif |
51 | #if defined(__sun__) && defined(__svr4__) |
52 | #include <kstat.h> |
53 | #endif |
54 | #if defined(__GNUC__) || defined(__clang__) |
55 | #if (defined(__i386__) || defined(__x86_64__)) && !defined(_MSC_VER) |
56 | #include <cpuid.h> |
57 | #endif |
58 | #endif |
59 | |
60 | #define DEBUG_TYPE "host-detection" |
61 | |
62 | //===----------------------------------------------------------------------===// |
63 | // |
64 | // Implementations of the CPU detection routines |
65 | // |
66 | //===----------------------------------------------------------------------===// |
67 | |
68 | using namespace llvm; |
69 | |
70 | static std::unique_ptr<llvm::MemoryBuffer> |
71 | LLVM_ATTRIBUTE_UNUSED getProcCpuinfoContent() { |
72 | const char *CPUInfoFile = "/proc/cpuinfo" ; |
73 | if (const char *CpuinfoIntercept = std::getenv(name: "LLVM_CPUINFO" )) |
74 | CPUInfoFile = CpuinfoIntercept; |
75 | llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> Text = |
76 | llvm::MemoryBuffer::getFileAsStream(Filename: CPUInfoFile); |
77 | |
78 | if (std::error_code EC = Text.getError()) { |
79 | llvm::errs() << "Can't read " << CPUInfoFile << ": " << EC.message() |
80 | << "\n" ; |
81 | return nullptr; |
82 | } |
83 | return std::move(*Text); |
84 | } |
85 | |
86 | StringRef sys::detail::getHostCPUNameForPowerPC(StringRef ProcCpuinfoContent) { |
87 | // Access to the Processor Version Register (PVR) on PowerPC is privileged, |
88 | // and so we must use an operating-system interface to determine the current |
89 | // processor type. On Linux, this is exposed through the /proc/cpuinfo file. |
90 | const char *generic = "generic" ; |
91 | |
92 | // The cpu line is second (after the 'processor: 0' line), so if this |
93 | // buffer is too small then something has changed (or is wrong). |
94 | StringRef::const_iterator CPUInfoStart = ProcCpuinfoContent.begin(); |
95 | StringRef::const_iterator CPUInfoEnd = ProcCpuinfoContent.end(); |
96 | |
97 | StringRef::const_iterator CIP = CPUInfoStart; |
98 | |
99 | StringRef::const_iterator CPUStart = nullptr; |
100 | size_t CPULen = 0; |
101 | |
102 | // We need to find the first line which starts with cpu, spaces, and a colon. |
103 | // After the colon, there may be some additional spaces and then the cpu type. |
104 | while (CIP < CPUInfoEnd && CPUStart == nullptr) { |
105 | if (CIP < CPUInfoEnd && *CIP == '\n') |
106 | ++CIP; |
107 | |
108 | if (CIP < CPUInfoEnd && *CIP == 'c') { |
109 | ++CIP; |
110 | if (CIP < CPUInfoEnd && *CIP == 'p') { |
111 | ++CIP; |
112 | if (CIP < CPUInfoEnd && *CIP == 'u') { |
113 | ++CIP; |
114 | while (CIP < CPUInfoEnd && (*CIP == ' ' || *CIP == '\t')) |
115 | ++CIP; |
116 | |
117 | if (CIP < CPUInfoEnd && *CIP == ':') { |
118 | ++CIP; |
119 | while (CIP < CPUInfoEnd && (*CIP == ' ' || *CIP == '\t')) |
120 | ++CIP; |
121 | |
122 | if (CIP < CPUInfoEnd) { |
123 | CPUStart = CIP; |
124 | while (CIP < CPUInfoEnd && (*CIP != ' ' && *CIP != '\t' && |
125 | *CIP != ',' && *CIP != '\n')) |
126 | ++CIP; |
127 | CPULen = CIP - CPUStart; |
128 | } |
129 | } |
130 | } |
131 | } |
132 | } |
133 | |
134 | if (CPUStart == nullptr) |
135 | while (CIP < CPUInfoEnd && *CIP != '\n') |
136 | ++CIP; |
137 | } |
138 | |
139 | if (CPUStart == nullptr) |
140 | return generic; |
141 | |
142 | return StringSwitch<const char *>(StringRef(CPUStart, CPULen)) |
143 | .Case(S: "604e" , Value: "604e" ) |
144 | .Case(S: "604" , Value: "604" ) |
145 | .Case(S: "7400" , Value: "7400" ) |
146 | .Case(S: "7410" , Value: "7400" ) |
147 | .Case(S: "7447" , Value: "7400" ) |
148 | .Case(S: "7455" , Value: "7450" ) |
149 | .Case(S: "G4" , Value: "g4" ) |
150 | .Case(S: "POWER4" , Value: "970" ) |
151 | .Case(S: "PPC970FX" , Value: "970" ) |
152 | .Case(S: "PPC970MP" , Value: "970" ) |
153 | .Case(S: "G5" , Value: "g5" ) |
154 | .Case(S: "POWER5" , Value: "g5" ) |
155 | .Case(S: "A2" , Value: "a2" ) |
156 | .Case(S: "POWER6" , Value: "pwr6" ) |
157 | .Case(S: "POWER7" , Value: "pwr7" ) |
158 | .Case(S: "POWER8" , Value: "pwr8" ) |
159 | .Case(S: "POWER8E" , Value: "pwr8" ) |
160 | .Case(S: "POWER8NVL" , Value: "pwr8" ) |
161 | .Case(S: "POWER9" , Value: "pwr9" ) |
162 | .Case(S: "POWER10" , Value: "pwr10" ) |
163 | .Case(S: "POWER11" , Value: "pwr11" ) |
164 | // FIXME: If we get a simulator or machine with the capabilities of |
165 | // mcpu=future, we should revisit this and add the name reported by the |
166 | // simulator/machine. |
167 | .Default(Value: generic); |
168 | } |
169 | |
170 | StringRef sys::detail::getHostCPUNameForARM(StringRef ProcCpuinfoContent) { |
171 | // The cpuid register on arm is not accessible from user space. On Linux, |
172 | // it is exposed through the /proc/cpuinfo file. |
173 | |
174 | // Read 32 lines from /proc/cpuinfo, which should contain the CPU part line |
175 | // in all cases. |
176 | SmallVector<StringRef, 32> Lines; |
177 | ProcCpuinfoContent.split(A&: Lines, Separator: '\n'); |
178 | |
179 | // Look for the CPU implementer line. |
180 | StringRef Implementer; |
181 | StringRef Hardware; |
182 | StringRef Part; |
183 | for (unsigned I = 0, E = Lines.size(); I != E; ++I) { |
184 | if (Lines[I].starts_with(Prefix: "CPU implementer" )) |
185 | Implementer = Lines[I].substr(Start: 15).ltrim(Chars: "\t :" ); |
186 | if (Lines[I].starts_with(Prefix: "Hardware" )) |
187 | Hardware = Lines[I].substr(Start: 8).ltrim(Chars: "\t :" ); |
188 | if (Lines[I].starts_with(Prefix: "CPU part" )) |
189 | Part = Lines[I].substr(Start: 8).ltrim(Chars: "\t :" ); |
190 | } |
191 | |
192 | if (Implementer == "0x41" ) { // ARM Ltd. |
193 | // MSM8992/8994 may give cpu part for the core that the kernel is running on, |
194 | // which is undeterministic and wrong. Always return cortex-a53 for these SoC. |
195 | if (Hardware.ends_with(Suffix: "MSM8994" ) || Hardware.ends_with(Suffix: "MSM8996" )) |
196 | return "cortex-a53" ; |
197 | |
198 | |
199 | // The CPU part is a 3 digit hexadecimal number with a 0x prefix. The |
200 | // values correspond to the "Part number" in the CP15/c0 register. The |
201 | // contents are specified in the various processor manuals. |
202 | // This corresponds to the Main ID Register in Technical Reference Manuals. |
203 | // and is used in programs like sys-utils |
204 | return StringSwitch<const char *>(Part) |
205 | .Case(S: "0x926" , Value: "arm926ej-s" ) |
206 | .Case(S: "0xb02" , Value: "mpcore" ) |
207 | .Case(S: "0xb36" , Value: "arm1136j-s" ) |
208 | .Case(S: "0xb56" , Value: "arm1156t2-s" ) |
209 | .Case(S: "0xb76" , Value: "arm1176jz-s" ) |
210 | .Case(S: "0xc05" , Value: "cortex-a5" ) |
211 | .Case(S: "0xc07" , Value: "cortex-a7" ) |
212 | .Case(S: "0xc08" , Value: "cortex-a8" ) |
213 | .Case(S: "0xc09" , Value: "cortex-a9" ) |
214 | .Case(S: "0xc0f" , Value: "cortex-a15" ) |
215 | .Case(S: "0xc0e" , Value: "cortex-a17" ) |
216 | .Case(S: "0xc20" , Value: "cortex-m0" ) |
217 | .Case(S: "0xc23" , Value: "cortex-m3" ) |
218 | .Case(S: "0xc24" , Value: "cortex-m4" ) |
219 | .Case(S: "0xc27" , Value: "cortex-m7" ) |
220 | .Case(S: "0xd20" , Value: "cortex-m23" ) |
221 | .Case(S: "0xd21" , Value: "cortex-m33" ) |
222 | .Case(S: "0xd24" , Value: "cortex-m52" ) |
223 | .Case(S: "0xd22" , Value: "cortex-m55" ) |
224 | .Case(S: "0xd23" , Value: "cortex-m85" ) |
225 | .Case(S: "0xc18" , Value: "cortex-r8" ) |
226 | .Case(S: "0xd13" , Value: "cortex-r52" ) |
227 | .Case(S: "0xd16" , Value: "cortex-r52plus" ) |
228 | .Case(S: "0xd15" , Value: "cortex-r82" ) |
229 | .Case(S: "0xd14" , Value: "cortex-r82ae" ) |
230 | .Case(S: "0xd02" , Value: "cortex-a34" ) |
231 | .Case(S: "0xd04" , Value: "cortex-a35" ) |
232 | .Case(S: "0xd8f" , Value: "cortex-a320" ) |
233 | .Case(S: "0xd03" , Value: "cortex-a53" ) |
234 | .Case(S: "0xd05" , Value: "cortex-a55" ) |
235 | .Case(S: "0xd46" , Value: "cortex-a510" ) |
236 | .Case(S: "0xd80" , Value: "cortex-a520" ) |
237 | .Case(S: "0xd88" , Value: "cortex-a520ae" ) |
238 | .Case(S: "0xd07" , Value: "cortex-a57" ) |
239 | .Case(S: "0xd06" , Value: "cortex-a65" ) |
240 | .Case(S: "0xd43" , Value: "cortex-a65ae" ) |
241 | .Case(S: "0xd08" , Value: "cortex-a72" ) |
242 | .Case(S: "0xd09" , Value: "cortex-a73" ) |
243 | .Case(S: "0xd0a" , Value: "cortex-a75" ) |
244 | .Case(S: "0xd0b" , Value: "cortex-a76" ) |
245 | .Case(S: "0xd0e" , Value: "cortex-a76ae" ) |
246 | .Case(S: "0xd0d" , Value: "cortex-a77" ) |
247 | .Case(S: "0xd41" , Value: "cortex-a78" ) |
248 | .Case(S: "0xd42" , Value: "cortex-a78ae" ) |
249 | .Case(S: "0xd4b" , Value: "cortex-a78c" ) |
250 | .Case(S: "0xd47" , Value: "cortex-a710" ) |
251 | .Case(S: "0xd4d" , Value: "cortex-a715" ) |
252 | .Case(S: "0xd81" , Value: "cortex-a720" ) |
253 | .Case(S: "0xd89" , Value: "cortex-a720ae" ) |
254 | .Case(S: "0xd87" , Value: "cortex-a725" ) |
255 | .Case(S: "0xd44" , Value: "cortex-x1" ) |
256 | .Case(S: "0xd4c" , Value: "cortex-x1c" ) |
257 | .Case(S: "0xd48" , Value: "cortex-x2" ) |
258 | .Case(S: "0xd4e" , Value: "cortex-x3" ) |
259 | .Case(S: "0xd82" , Value: "cortex-x4" ) |
260 | .Case(S: "0xd85" , Value: "cortex-x925" ) |
261 | .Case(S: "0xd4a" , Value: "neoverse-e1" ) |
262 | .Case(S: "0xd0c" , Value: "neoverse-n1" ) |
263 | .Case(S: "0xd49" , Value: "neoverse-n2" ) |
264 | .Case(S: "0xd8e" , Value: "neoverse-n3" ) |
265 | .Case(S: "0xd40" , Value: "neoverse-v1" ) |
266 | .Case(S: "0xd4f" , Value: "neoverse-v2" ) |
267 | .Case(S: "0xd84" , Value: "neoverse-v3" ) |
268 | .Case(S: "0xd83" , Value: "neoverse-v3ae" ) |
269 | .Default(Value: "generic" ); |
270 | } |
271 | |
272 | if (Implementer == "0x42" || Implementer == "0x43" ) { // Broadcom | Cavium. |
273 | return StringSwitch<const char *>(Part) |
274 | .Case(S: "0x516" , Value: "thunderx2t99" ) |
275 | .Case(S: "0x0516" , Value: "thunderx2t99" ) |
276 | .Case(S: "0xaf" , Value: "thunderx2t99" ) |
277 | .Case(S: "0x0af" , Value: "thunderx2t99" ) |
278 | .Case(S: "0xa1" , Value: "thunderxt88" ) |
279 | .Case(S: "0x0a1" , Value: "thunderxt88" ) |
280 | .Default(Value: "generic" ); |
281 | } |
282 | |
283 | if (Implementer == "0x46" ) { // Fujitsu Ltd. |
284 | return StringSwitch<const char *>(Part) |
285 | .Case(S: "0x001" , Value: "a64fx" ) |
286 | .Case(S: "0x003" , Value: "fujitsu-monaka" ) |
287 | .Default(Value: "generic" ); |
288 | } |
289 | |
290 | if (Implementer == "0x4e" ) { // NVIDIA Corporation |
291 | return StringSwitch<const char *>(Part) |
292 | .Case(S: "0x004" , Value: "carmel" ) |
293 | .Case(S: "0x10" , Value: "olympus" ) |
294 | .Case(S: "0x010" , Value: "olympus" ) |
295 | .Default(Value: "generic" ); |
296 | } |
297 | |
298 | if (Implementer == "0x48" ) // HiSilicon Technologies, Inc. |
299 | // The CPU part is a 3 digit hexadecimal number with a 0x prefix. The |
300 | // values correspond to the "Part number" in the CP15/c0 register. The |
301 | // contents are specified in the various processor manuals. |
302 | return StringSwitch<const char *>(Part) |
303 | .Case(S: "0xd01" , Value: "tsv110" ) |
304 | .Default(Value: "generic" ); |
305 | |
306 | if (Implementer == "0x51" ) // Qualcomm Technologies, Inc. |
307 | // The CPU part is a 3 digit hexadecimal number with a 0x prefix. The |
308 | // values correspond to the "Part number" in the CP15/c0 register. The |
309 | // contents are specified in the various processor manuals. |
310 | return StringSwitch<const char *>(Part) |
311 | .Case(S: "0x06f" , Value: "krait" ) // APQ8064 |
312 | .Case(S: "0x201" , Value: "kryo" ) |
313 | .Case(S: "0x205" , Value: "kryo" ) |
314 | .Case(S: "0x211" , Value: "kryo" ) |
315 | .Case(S: "0x800" , Value: "cortex-a73" ) // Kryo 2xx Gold |
316 | .Case(S: "0x801" , Value: "cortex-a73" ) // Kryo 2xx Silver |
317 | .Case(S: "0x802" , Value: "cortex-a75" ) // Kryo 3xx Gold |
318 | .Case(S: "0x803" , Value: "cortex-a75" ) // Kryo 3xx Silver |
319 | .Case(S: "0x804" , Value: "cortex-a76" ) // Kryo 4xx Gold |
320 | .Case(S: "0x805" , Value: "cortex-a76" ) // Kryo 4xx/5xx Silver |
321 | .Case(S: "0xc00" , Value: "falkor" ) |
322 | .Case(S: "0xc01" , Value: "saphira" ) |
323 | .Case(S: "0x001" , Value: "oryon-1" ) |
324 | .Default(Value: "generic" ); |
325 | if (Implementer == "0x53" ) { // Samsung Electronics Co., Ltd. |
326 | // The Exynos chips have a convoluted ID scheme that doesn't seem to follow |
327 | // any predictive pattern across variants and parts. |
328 | unsigned Variant = 0, Part = 0; |
329 | |
330 | // Look for the CPU variant line, whose value is a 1 digit hexadecimal |
331 | // number, corresponding to the Variant bits in the CP15/C0 register. |
332 | for (auto I : Lines) |
333 | if (I.consume_front(Prefix: "CPU variant" )) |
334 | I.ltrim(Chars: "\t :" ).getAsInteger(Radix: 0, Result&: Variant); |
335 | |
336 | // Look for the CPU part line, whose value is a 3 digit hexadecimal |
337 | // number, corresponding to the PartNum bits in the CP15/C0 register. |
338 | for (auto I : Lines) |
339 | if (I.consume_front(Prefix: "CPU part" )) |
340 | I.ltrim(Chars: "\t :" ).getAsInteger(Radix: 0, Result&: Part); |
341 | |
342 | unsigned Exynos = (Variant << 12) | Part; |
343 | switch (Exynos) { |
344 | default: |
345 | // Default by falling through to Exynos M3. |
346 | [[fallthrough]]; |
347 | case 0x1002: |
348 | return "exynos-m3" ; |
349 | case 0x1003: |
350 | return "exynos-m4" ; |
351 | } |
352 | } |
353 | |
354 | if (Implementer == "0x61" ) { // Apple |
355 | return StringSwitch<const char *>(Part) |
356 | .Case(S: "0x020" , Value: "apple-m1" ) |
357 | .Case(S: "0x021" , Value: "apple-m1" ) |
358 | .Case(S: "0x022" , Value: "apple-m1" ) |
359 | .Case(S: "0x023" , Value: "apple-m1" ) |
360 | .Case(S: "0x024" , Value: "apple-m1" ) |
361 | .Case(S: "0x025" , Value: "apple-m1" ) |
362 | .Case(S: "0x028" , Value: "apple-m1" ) |
363 | .Case(S: "0x029" , Value: "apple-m1" ) |
364 | .Case(S: "0x030" , Value: "apple-m2" ) |
365 | .Case(S: "0x031" , Value: "apple-m2" ) |
366 | .Case(S: "0x032" , Value: "apple-m2" ) |
367 | .Case(S: "0x033" , Value: "apple-m2" ) |
368 | .Case(S: "0x034" , Value: "apple-m2" ) |
369 | .Case(S: "0x035" , Value: "apple-m2" ) |
370 | .Case(S: "0x038" , Value: "apple-m2" ) |
371 | .Case(S: "0x039" , Value: "apple-m2" ) |
372 | .Case(S: "0x049" , Value: "apple-m3" ) |
373 | .Case(S: "0x048" , Value: "apple-m3" ) |
374 | .Default(Value: "generic" ); |
375 | } |
376 | |
377 | if (Implementer == "0x63" ) { // Arm China. |
378 | return StringSwitch<const char *>(Part) |
379 | .Case(S: "0x132" , Value: "star-mc1" ) |
380 | .Default(Value: "generic" ); |
381 | } |
382 | |
383 | if (Implementer == "0x6d" ) { // Microsoft Corporation. |
384 | // The Microsoft Azure Cobalt 100 CPU is handled as a Neoverse N2. |
385 | return StringSwitch<const char *>(Part) |
386 | .Case(S: "0xd49" , Value: "neoverse-n2" ) |
387 | .Default(Value: "generic" ); |
388 | } |
389 | |
390 | if (Implementer == "0xc0" ) { // Ampere Computing |
391 | return StringSwitch<const char *>(Part) |
392 | .Case(S: "0xac3" , Value: "ampere1" ) |
393 | .Case(S: "0xac4" , Value: "ampere1a" ) |
394 | .Case(S: "0xac5" , Value: "ampere1b" ) |
395 | .Default(Value: "generic" ); |
396 | } |
397 | |
398 | return "generic" ; |
399 | } |
400 | |
401 | namespace { |
402 | StringRef getCPUNameFromS390Model(unsigned int Id, bool HaveVectorSupport) { |
403 | switch (Id) { |
404 | case 2064: // z900 not supported by LLVM |
405 | case 2066: |
406 | case 2084: // z990 not supported by LLVM |
407 | case 2086: |
408 | case 2094: // z9-109 not supported by LLVM |
409 | case 2096: |
410 | return "generic" ; |
411 | case 2097: |
412 | case 2098: |
413 | return "z10" ; |
414 | case 2817: |
415 | case 2818: |
416 | return "z196" ; |
417 | case 2827: |
418 | case 2828: |
419 | return "zEC12" ; |
420 | case 2964: |
421 | case 2965: |
422 | return HaveVectorSupport? "z13" : "zEC12" ; |
423 | case 3906: |
424 | case 3907: |
425 | return HaveVectorSupport? "z14" : "zEC12" ; |
426 | case 8561: |
427 | case 8562: |
428 | return HaveVectorSupport? "z15" : "zEC12" ; |
429 | case 3931: |
430 | case 3932: |
431 | return HaveVectorSupport? "z16" : "zEC12" ; |
432 | case 9175: |
433 | case 9176: |
434 | default: |
435 | return HaveVectorSupport? "z17" : "zEC12" ; |
436 | } |
437 | } |
438 | } // end anonymous namespace |
439 | |
440 | StringRef sys::detail::getHostCPUNameForS390x(StringRef ProcCpuinfoContent) { |
441 | // STIDP is a privileged operation, so use /proc/cpuinfo instead. |
442 | |
443 | // The "processor 0:" line comes after a fair amount of other information, |
444 | // including a cache breakdown, but this should be plenty. |
445 | SmallVector<StringRef, 32> Lines; |
446 | ProcCpuinfoContent.split(A&: Lines, Separator: '\n'); |
447 | |
448 | // Look for the CPU features. |
449 | SmallVector<StringRef, 32> CPUFeatures; |
450 | for (unsigned I = 0, E = Lines.size(); I != E; ++I) |
451 | if (Lines[I].starts_with(Prefix: "features" )) { |
452 | size_t Pos = Lines[I].find(C: ':'); |
453 | if (Pos != StringRef::npos) { |
454 | Lines[I].drop_front(N: Pos + 1).split(A&: CPUFeatures, Separator: ' '); |
455 | break; |
456 | } |
457 | } |
458 | |
459 | // We need to check for the presence of vector support independently of |
460 | // the machine type, since we may only use the vector register set when |
461 | // supported by the kernel (and hypervisor). |
462 | bool HaveVectorSupport = false; |
463 | for (unsigned I = 0, E = CPUFeatures.size(); I != E; ++I) { |
464 | if (CPUFeatures[I] == "vx" ) |
465 | HaveVectorSupport = true; |
466 | } |
467 | |
468 | // Now check the processor machine type. |
469 | for (unsigned I = 0, E = Lines.size(); I != E; ++I) { |
470 | if (Lines[I].starts_with(Prefix: "processor " )) { |
471 | size_t Pos = Lines[I].find(Str: "machine = " ); |
472 | if (Pos != StringRef::npos) { |
473 | Pos += sizeof("machine = " ) - 1; |
474 | unsigned int Id; |
475 | if (!Lines[I].drop_front(N: Pos).getAsInteger(Radix: 10, Result&: Id)) |
476 | return getCPUNameFromS390Model(Id, HaveVectorSupport); |
477 | } |
478 | break; |
479 | } |
480 | } |
481 | |
482 | return "generic" ; |
483 | } |
484 | |
485 | StringRef sys::detail::getHostCPUNameForRISCV(StringRef ProcCpuinfoContent) { |
486 | // There are 24 lines in /proc/cpuinfo |
487 | SmallVector<StringRef> Lines; |
488 | ProcCpuinfoContent.split(A&: Lines, Separator: '\n'); |
489 | |
490 | // Look for uarch line to determine cpu name |
491 | StringRef UArch; |
492 | for (unsigned I = 0, E = Lines.size(); I != E; ++I) { |
493 | if (Lines[I].starts_with(Prefix: "uarch" )) { |
494 | UArch = Lines[I].substr(Start: 5).ltrim(Chars: "\t :" ); |
495 | break; |
496 | } |
497 | } |
498 | |
499 | return StringSwitch<const char *>(UArch) |
500 | .Case(S: "eswin,eic770x" , Value: "sifive-p550" ) |
501 | .Case(S: "sifive,u74-mc" , Value: "sifive-u74" ) |
502 | .Case(S: "sifive,bullet0" , Value: "sifive-u74" ) |
503 | .Default(Value: "" ); |
504 | } |
505 | |
506 | StringRef sys::detail::getHostCPUNameForBPF() { |
507 | #if !defined(__linux__) || !defined(__x86_64__) |
508 | return "generic" ; |
509 | #else |
510 | uint8_t v3_insns[40] __attribute__ ((aligned (8))) = |
511 | /* BPF_MOV64_IMM(BPF_REG_0, 0) */ |
512 | { 0xb7, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, |
513 | /* BPF_MOV64_IMM(BPF_REG_2, 1) */ |
514 | 0xb7, 0x2, 0x0, 0x0, 0x1, 0x0, 0x0, 0x0, |
515 | /* BPF_JMP32_REG(BPF_JLT, BPF_REG_0, BPF_REG_2, 1) */ |
516 | 0xae, 0x20, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0, |
517 | /* BPF_MOV64_IMM(BPF_REG_0, 1) */ |
518 | 0xb7, 0x0, 0x0, 0x0, 0x1, 0x0, 0x0, 0x0, |
519 | /* BPF_EXIT_INSN() */ |
520 | 0x95, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 }; |
521 | |
522 | uint8_t v2_insns[40] __attribute__ ((aligned (8))) = |
523 | /* BPF_MOV64_IMM(BPF_REG_0, 0) */ |
524 | { 0xb7, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, |
525 | /* BPF_MOV64_IMM(BPF_REG_2, 1) */ |
526 | 0xb7, 0x2, 0x0, 0x0, 0x1, 0x0, 0x0, 0x0, |
527 | /* BPF_JMP_REG(BPF_JLT, BPF_REG_0, BPF_REG_2, 1) */ |
528 | 0xad, 0x20, 0x1, 0x0, 0x0, 0x0, 0x0, 0x0, |
529 | /* BPF_MOV64_IMM(BPF_REG_0, 1) */ |
530 | 0xb7, 0x0, 0x0, 0x0, 0x1, 0x0, 0x0, 0x0, |
531 | /* BPF_EXIT_INSN() */ |
532 | 0x95, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 }; |
533 | |
534 | struct bpf_prog_load_attr { |
535 | uint32_t prog_type; |
536 | uint32_t insn_cnt; |
537 | uint64_t insns; |
538 | uint64_t license; |
539 | uint32_t log_level; |
540 | uint32_t log_size; |
541 | uint64_t log_buf; |
542 | uint32_t kern_version; |
543 | uint32_t prog_flags; |
544 | } attr = {}; |
545 | attr.prog_type = 1; /* BPF_PROG_TYPE_SOCKET_FILTER */ |
546 | attr.insn_cnt = 5; |
547 | attr.insns = (uint64_t)v3_insns; |
548 | attr.license = (uint64_t)"DUMMY" ; |
549 | |
550 | int fd = syscall(sysno: 321 /* __NR_bpf */, 5 /* BPF_PROG_LOAD */, &attr, |
551 | sizeof(attr)); |
552 | if (fd >= 0) { |
553 | close(fd: fd); |
554 | return "v3" ; |
555 | } |
556 | |
557 | /* Clear the whole attr in case its content changed by syscall. */ |
558 | memset(s: &attr, c: 0, n: sizeof(attr)); |
559 | attr.prog_type = 1; /* BPF_PROG_TYPE_SOCKET_FILTER */ |
560 | attr.insn_cnt = 5; |
561 | attr.insns = (uint64_t)v2_insns; |
562 | attr.license = (uint64_t)"DUMMY" ; |
563 | fd = syscall(sysno: 321 /* __NR_bpf */, 5 /* BPF_PROG_LOAD */, &attr, sizeof(attr)); |
564 | if (fd >= 0) { |
565 | close(fd: fd); |
566 | return "v2" ; |
567 | } |
568 | return "v1" ; |
569 | #endif |
570 | } |
571 | |
572 | #if defined(__i386__) || defined(_M_IX86) || defined(__x86_64__) || \ |
573 | defined(_M_X64) |
574 | |
575 | /// getX86CpuIDAndInfo - Execute the specified cpuid and return the 4 values in |
576 | /// the specified arguments. If we can't run cpuid on the host, return true. |
577 | static bool getX86CpuIDAndInfo(unsigned value, unsigned *rEAX, unsigned *rEBX, |
578 | unsigned *rECX, unsigned *rEDX) { |
579 | #if (defined(__i386__) || defined(__x86_64__)) && !defined(_MSC_VER) |
580 | return !__get_cpuid(leaf: value, eax: rEAX, ebx: rEBX, ecx: rECX, edx: rEDX); |
581 | #elif defined(_MSC_VER) |
582 | // The MSVC intrinsic is portable across x86 and x64. |
583 | int registers[4]; |
584 | __cpuid(registers, value); |
585 | *rEAX = registers[0]; |
586 | *rEBX = registers[1]; |
587 | *rECX = registers[2]; |
588 | *rEDX = registers[3]; |
589 | return false; |
590 | #else |
591 | return true; |
592 | #endif |
593 | } |
594 | |
595 | namespace llvm { |
596 | namespace sys { |
597 | namespace detail { |
598 | namespace x86 { |
599 | |
600 | VendorSignatures getVendorSignature(unsigned *MaxLeaf) { |
601 | unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0; |
602 | if (MaxLeaf == nullptr) |
603 | MaxLeaf = &EAX; |
604 | else |
605 | *MaxLeaf = 0; |
606 | |
607 | if (getX86CpuIDAndInfo(value: 0, rEAX: MaxLeaf, rEBX: &EBX, rECX: &ECX, rEDX: &EDX) || *MaxLeaf < 1) |
608 | return VendorSignatures::UNKNOWN; |
609 | |
610 | // "Genu ineI ntel" |
611 | if (EBX == 0x756e6547 && EDX == 0x49656e69 && ECX == 0x6c65746e) |
612 | return VendorSignatures::GENUINE_INTEL; |
613 | |
614 | // "Auth enti cAMD" |
615 | if (EBX == 0x68747541 && EDX == 0x69746e65 && ECX == 0x444d4163) |
616 | return VendorSignatures::AUTHENTIC_AMD; |
617 | |
618 | return VendorSignatures::UNKNOWN; |
619 | } |
620 | |
621 | } // namespace x86 |
622 | } // namespace detail |
623 | } // namespace sys |
624 | } // namespace llvm |
625 | |
626 | using namespace llvm::sys::detail::x86; |
627 | |
628 | /// getX86CpuIDAndInfoEx - Execute the specified cpuid with subleaf and return |
629 | /// the 4 values in the specified arguments. If we can't run cpuid on the host, |
630 | /// return true. |
631 | static bool getX86CpuIDAndInfoEx(unsigned value, unsigned subleaf, |
632 | unsigned *rEAX, unsigned *rEBX, unsigned *rECX, |
633 | unsigned *rEDX) { |
634 | // TODO(boomanaiden154): When the minimum toolchain versions for gcc and clang |
635 | // are such that __cpuidex is defined within cpuid.h for both, we can remove |
636 | // the __get_cpuid_count function and share the MSVC implementation between |
637 | // all three. |
638 | #if (defined(__i386__) || defined(__x86_64__)) && !defined(_MSC_VER) |
639 | return !__get_cpuid_count(leaf: value, subleaf: subleaf, eax: rEAX, ebx: rEBX, ecx: rECX, edx: rEDX); |
640 | #elif defined(_MSC_VER) |
641 | int registers[4]; |
642 | __cpuidex(registers, value, subleaf); |
643 | *rEAX = registers[0]; |
644 | *rEBX = registers[1]; |
645 | *rECX = registers[2]; |
646 | *rEDX = registers[3]; |
647 | return false; |
648 | #else |
649 | return true; |
650 | #endif |
651 | } |
652 | |
653 | // Read control register 0 (XCR0). Used to detect features such as AVX. |
654 | static bool getX86XCR0(unsigned *rEAX, unsigned *rEDX) { |
655 | // TODO(boomanaiden154): When the minimum toolchain versions for gcc and clang |
656 | // are such that _xgetbv is supported by both, we can unify the implementation |
657 | // with MSVC and remove all inline assembly. |
658 | #if defined(__GNUC__) || defined(__clang__) |
659 | // Check xgetbv; this uses a .byte sequence instead of the instruction |
660 | // directly because older assemblers do not include support for xgetbv and |
661 | // there is no easy way to conditionally compile based on the assembler used. |
662 | __asm__(".byte 0x0f, 0x01, 0xd0" : "=a" (*rEAX), "=d" (*rEDX) : "c" (0)); |
663 | return false; |
664 | #elif defined(_MSC_FULL_VER) && defined(_XCR_XFEATURE_ENABLED_MASK) |
665 | unsigned long long Result = _xgetbv(_XCR_XFEATURE_ENABLED_MASK); |
666 | *rEAX = Result; |
667 | *rEDX = Result >> 32; |
668 | return false; |
669 | #else |
670 | return true; |
671 | #endif |
672 | } |
673 | |
674 | static void detectX86FamilyModel(unsigned EAX, unsigned *Family, |
675 | unsigned *Model) { |
676 | *Family = (EAX >> 8) & 0xf; // Bits 8 - 11 |
677 | *Model = (EAX >> 4) & 0xf; // Bits 4 - 7 |
678 | if (*Family == 6 || *Family == 0xf) { |
679 | if (*Family == 0xf) |
680 | // Examine extended family ID if family ID is F. |
681 | *Family += (EAX >> 20) & 0xff; // Bits 20 - 27 |
682 | // Examine extended model ID if family ID is 6 or F. |
683 | *Model += ((EAX >> 16) & 0xf) << 4; // Bits 16 - 19 |
684 | } |
685 | } |
686 | |
687 | #define testFeature(F) (Features[F / 32] & (1 << (F % 32))) != 0 |
688 | |
689 | static StringRef getIntelProcessorTypeAndSubtype(unsigned Family, |
690 | unsigned Model, |
691 | const unsigned *Features, |
692 | unsigned *Type, |
693 | unsigned *Subtype) { |
694 | StringRef CPU; |
695 | |
696 | switch (Family) { |
697 | case 3: |
698 | CPU = "i386" ; |
699 | break; |
700 | case 4: |
701 | CPU = "i486" ; |
702 | break; |
703 | case 5: |
704 | if (testFeature(X86::FEATURE_MMX)) { |
705 | CPU = "pentium-mmx" ; |
706 | break; |
707 | } |
708 | CPU = "pentium" ; |
709 | break; |
710 | case 6: |
711 | switch (Model) { |
712 | case 0x0f: // Intel Core 2 Duo processor, Intel Core 2 Duo mobile |
713 | // processor, Intel Core 2 Quad processor, Intel Core 2 Quad |
714 | // mobile processor, Intel Core 2 Extreme processor, Intel |
715 | // Pentium Dual-Core processor, Intel Xeon processor, model |
716 | // 0Fh. All processors are manufactured using the 65 nm process. |
717 | case 0x16: // Intel Celeron processor model 16h. All processors are |
718 | // manufactured using the 65 nm process |
719 | CPU = "core2" ; |
720 | *Type = X86::INTEL_CORE2; |
721 | break; |
722 | case 0x17: // Intel Core 2 Extreme processor, Intel Xeon processor, model |
723 | // 17h. All processors are manufactured using the 45 nm process. |
724 | // |
725 | // 45nm: Penryn , Wolfdale, Yorkfield (XE) |
726 | case 0x1d: // Intel Xeon processor MP. All processors are manufactured using |
727 | // the 45 nm process. |
728 | CPU = "penryn" ; |
729 | *Type = X86::INTEL_CORE2; |
730 | break; |
731 | case 0x1a: // Intel Core i7 processor and Intel Xeon processor. All |
732 | // processors are manufactured using the 45 nm process. |
733 | case 0x1e: // Intel(R) Core(TM) i7 CPU 870 @ 2.93GHz. |
734 | // As found in a Summer 2010 model iMac. |
735 | case 0x1f: |
736 | case 0x2e: // Nehalem EX |
737 | CPU = "nehalem" ; |
738 | *Type = X86::INTEL_COREI7; |
739 | *Subtype = X86::INTEL_COREI7_NEHALEM; |
740 | break; |
741 | case 0x25: // Intel Core i7, laptop version. |
742 | case 0x2c: // Intel Core i7 processor and Intel Xeon processor. All |
743 | // processors are manufactured using the 32 nm process. |
744 | case 0x2f: // Westmere EX |
745 | CPU = "westmere" ; |
746 | *Type = X86::INTEL_COREI7; |
747 | *Subtype = X86::INTEL_COREI7_WESTMERE; |
748 | break; |
749 | case 0x2a: // Intel Core i7 processor. All processors are manufactured |
750 | // using the 32 nm process. |
751 | case 0x2d: |
752 | CPU = "sandybridge" ; |
753 | *Type = X86::INTEL_COREI7; |
754 | *Subtype = X86::INTEL_COREI7_SANDYBRIDGE; |
755 | break; |
756 | case 0x3a: |
757 | case 0x3e: // Ivy Bridge EP |
758 | CPU = "ivybridge" ; |
759 | *Type = X86::INTEL_COREI7; |
760 | *Subtype = X86::INTEL_COREI7_IVYBRIDGE; |
761 | break; |
762 | |
763 | // Haswell: |
764 | case 0x3c: |
765 | case 0x3f: |
766 | case 0x45: |
767 | case 0x46: |
768 | CPU = "haswell" ; |
769 | *Type = X86::INTEL_COREI7; |
770 | *Subtype = X86::INTEL_COREI7_HASWELL; |
771 | break; |
772 | |
773 | // Broadwell: |
774 | case 0x3d: |
775 | case 0x47: |
776 | case 0x4f: |
777 | case 0x56: |
778 | CPU = "broadwell" ; |
779 | *Type = X86::INTEL_COREI7; |
780 | *Subtype = X86::INTEL_COREI7_BROADWELL; |
781 | break; |
782 | |
783 | // Skylake: |
784 | case 0x4e: // Skylake mobile |
785 | case 0x5e: // Skylake desktop |
786 | case 0x8e: // Kaby Lake mobile |
787 | case 0x9e: // Kaby Lake desktop |
788 | case 0xa5: // Comet Lake-H/S |
789 | case 0xa6: // Comet Lake-U |
790 | CPU = "skylake" ; |
791 | *Type = X86::INTEL_COREI7; |
792 | *Subtype = X86::INTEL_COREI7_SKYLAKE; |
793 | break; |
794 | |
795 | // Rocketlake: |
796 | case 0xa7: |
797 | CPU = "rocketlake" ; |
798 | *Type = X86::INTEL_COREI7; |
799 | *Subtype = X86::INTEL_COREI7_ROCKETLAKE; |
800 | break; |
801 | |
802 | // Skylake Xeon: |
803 | case 0x55: |
804 | *Type = X86::INTEL_COREI7; |
805 | if (testFeature(X86::FEATURE_AVX512BF16)) { |
806 | CPU = "cooperlake" ; |
807 | *Subtype = X86::INTEL_COREI7_COOPERLAKE; |
808 | } else if (testFeature(X86::FEATURE_AVX512VNNI)) { |
809 | CPU = "cascadelake" ; |
810 | *Subtype = X86::INTEL_COREI7_CASCADELAKE; |
811 | } else { |
812 | CPU = "skylake-avx512" ; |
813 | *Subtype = X86::INTEL_COREI7_SKYLAKE_AVX512; |
814 | } |
815 | break; |
816 | |
817 | // Cannonlake: |
818 | case 0x66: |
819 | CPU = "cannonlake" ; |
820 | *Type = X86::INTEL_COREI7; |
821 | *Subtype = X86::INTEL_COREI7_CANNONLAKE; |
822 | break; |
823 | |
824 | // Icelake: |
825 | case 0x7d: |
826 | case 0x7e: |
827 | CPU = "icelake-client" ; |
828 | *Type = X86::INTEL_COREI7; |
829 | *Subtype = X86::INTEL_COREI7_ICELAKE_CLIENT; |
830 | break; |
831 | |
832 | // Tigerlake: |
833 | case 0x8c: |
834 | case 0x8d: |
835 | CPU = "tigerlake" ; |
836 | *Type = X86::INTEL_COREI7; |
837 | *Subtype = X86::INTEL_COREI7_TIGERLAKE; |
838 | break; |
839 | |
840 | // Alderlake: |
841 | case 0x97: |
842 | case 0x9a: |
843 | CPU = "alderlake" ; |
844 | *Type = X86::INTEL_COREI7; |
845 | *Subtype = X86::INTEL_COREI7_ALDERLAKE; |
846 | break; |
847 | |
848 | // Gracemont |
849 | case 0xbe: |
850 | CPU = "gracemont" ; |
851 | *Type = X86::INTEL_COREI7; |
852 | *Subtype = X86::INTEL_COREI7_ALDERLAKE; |
853 | break; |
854 | |
855 | // Raptorlake: |
856 | case 0xb7: |
857 | case 0xba: |
858 | case 0xbf: |
859 | CPU = "raptorlake" ; |
860 | *Type = X86::INTEL_COREI7; |
861 | *Subtype = X86::INTEL_COREI7_ALDERLAKE; |
862 | break; |
863 | |
864 | // Meteorlake: |
865 | case 0xaa: |
866 | case 0xac: |
867 | CPU = "meteorlake" ; |
868 | *Type = X86::INTEL_COREI7; |
869 | *Subtype = X86::INTEL_COREI7_ALDERLAKE; |
870 | break; |
871 | |
872 | // Arrowlake: |
873 | case 0xc5: |
874 | // Arrowlake U: |
875 | case 0xb5: |
876 | CPU = "arrowlake" ; |
877 | *Type = X86::INTEL_COREI7; |
878 | *Subtype = X86::INTEL_COREI7_ARROWLAKE; |
879 | break; |
880 | |
881 | // Arrowlake S: |
882 | case 0xc6: |
883 | CPU = "arrowlake-s" ; |
884 | *Type = X86::INTEL_COREI7; |
885 | *Subtype = X86::INTEL_COREI7_ARROWLAKE_S; |
886 | break; |
887 | |
888 | // Lunarlake: |
889 | case 0xbd: |
890 | CPU = "lunarlake" ; |
891 | *Type = X86::INTEL_COREI7; |
892 | *Subtype = X86::INTEL_COREI7_ARROWLAKE_S; |
893 | break; |
894 | |
895 | // Pantherlake: |
896 | case 0xcc: |
897 | CPU = "pantherlake" ; |
898 | *Type = X86::INTEL_COREI7; |
899 | *Subtype = X86::INTEL_COREI7_PANTHERLAKE; |
900 | break; |
901 | |
902 | // Graniterapids: |
903 | case 0xad: |
904 | CPU = "graniterapids" ; |
905 | *Type = X86::INTEL_COREI7; |
906 | *Subtype = X86::INTEL_COREI7_GRANITERAPIDS; |
907 | break; |
908 | |
909 | // Granite Rapids D: |
910 | case 0xae: |
911 | CPU = "graniterapids-d" ; |
912 | *Type = X86::INTEL_COREI7; |
913 | *Subtype = X86::INTEL_COREI7_GRANITERAPIDS_D; |
914 | break; |
915 | |
916 | // Icelake Xeon: |
917 | case 0x6a: |
918 | case 0x6c: |
919 | CPU = "icelake-server" ; |
920 | *Type = X86::INTEL_COREI7; |
921 | *Subtype = X86::INTEL_COREI7_ICELAKE_SERVER; |
922 | break; |
923 | |
924 | // Emerald Rapids: |
925 | case 0xcf: |
926 | CPU = "emeraldrapids" ; |
927 | *Type = X86::INTEL_COREI7; |
928 | *Subtype = X86::INTEL_COREI7_SAPPHIRERAPIDS; |
929 | break; |
930 | |
931 | // Sapphire Rapids: |
932 | case 0x8f: |
933 | CPU = "sapphirerapids" ; |
934 | *Type = X86::INTEL_COREI7; |
935 | *Subtype = X86::INTEL_COREI7_SAPPHIRERAPIDS; |
936 | break; |
937 | |
938 | case 0x1c: // Most 45 nm Intel Atom processors |
939 | case 0x26: // 45 nm Atom Lincroft |
940 | case 0x27: // 32 nm Atom Medfield |
941 | case 0x35: // 32 nm Atom Midview |
942 | case 0x36: // 32 nm Atom Midview |
943 | CPU = "bonnell" ; |
944 | *Type = X86::INTEL_BONNELL; |
945 | break; |
946 | |
947 | // Atom Silvermont codes from the Intel software optimization guide. |
948 | case 0x37: |
949 | case 0x4a: |
950 | case 0x4d: |
951 | case 0x5a: |
952 | case 0x5d: |
953 | case 0x4c: // really airmont |
954 | CPU = "silvermont" ; |
955 | *Type = X86::INTEL_SILVERMONT; |
956 | break; |
957 | // Goldmont: |
958 | case 0x5c: // Apollo Lake |
959 | case 0x5f: // Denverton |
960 | CPU = "goldmont" ; |
961 | *Type = X86::INTEL_GOLDMONT; |
962 | break; |
963 | case 0x7a: |
964 | CPU = "goldmont-plus" ; |
965 | *Type = X86::INTEL_GOLDMONT_PLUS; |
966 | break; |
967 | case 0x86: |
968 | case 0x8a: // Lakefield |
969 | case 0x96: // Elkhart Lake |
970 | case 0x9c: // Jasper Lake |
971 | CPU = "tremont" ; |
972 | *Type = X86::INTEL_TREMONT; |
973 | break; |
974 | |
975 | // Sierraforest: |
976 | case 0xaf: |
977 | CPU = "sierraforest" ; |
978 | *Type = X86::INTEL_SIERRAFOREST; |
979 | break; |
980 | |
981 | // Grandridge: |
982 | case 0xb6: |
983 | CPU = "grandridge" ; |
984 | *Type = X86::INTEL_GRANDRIDGE; |
985 | break; |
986 | |
987 | // Clearwaterforest: |
988 | case 0xdd: |
989 | CPU = "clearwaterforest" ; |
990 | *Type = X86::INTEL_CLEARWATERFOREST; |
991 | break; |
992 | |
993 | // Xeon Phi (Knights Landing + Knights Mill): |
994 | case 0x57: |
995 | CPU = "knl" ; |
996 | *Type = X86::INTEL_KNL; |
997 | break; |
998 | case 0x85: |
999 | CPU = "knm" ; |
1000 | *Type = X86::INTEL_KNM; |
1001 | break; |
1002 | |
1003 | default: // Unknown family 6 CPU, try to guess. |
1004 | // Don't both with Type/Subtype here, they aren't used by the caller. |
1005 | // They're used above to keep the code in sync with compiler-rt. |
1006 | // TODO detect tigerlake host from model |
1007 | if (testFeature(X86::FEATURE_AVX512VP2INTERSECT)) { |
1008 | CPU = "tigerlake" ; |
1009 | } else if (testFeature(X86::FEATURE_AVX512VBMI2)) { |
1010 | CPU = "icelake-client" ; |
1011 | } else if (testFeature(X86::FEATURE_AVX512VBMI)) { |
1012 | CPU = "cannonlake" ; |
1013 | } else if (testFeature(X86::FEATURE_AVX512BF16)) { |
1014 | CPU = "cooperlake" ; |
1015 | } else if (testFeature(X86::FEATURE_AVX512VNNI)) { |
1016 | CPU = "cascadelake" ; |
1017 | } else if (testFeature(X86::FEATURE_AVX512VL)) { |
1018 | CPU = "skylake-avx512" ; |
1019 | } else if (testFeature(X86::FEATURE_CLFLUSHOPT)) { |
1020 | if (testFeature(X86::FEATURE_SHA)) |
1021 | CPU = "goldmont" ; |
1022 | else |
1023 | CPU = "skylake" ; |
1024 | } else if (testFeature(X86::FEATURE_ADX)) { |
1025 | CPU = "broadwell" ; |
1026 | } else if (testFeature(X86::FEATURE_AVX2)) { |
1027 | CPU = "haswell" ; |
1028 | } else if (testFeature(X86::FEATURE_AVX)) { |
1029 | CPU = "sandybridge" ; |
1030 | } else if (testFeature(X86::FEATURE_SSE4_2)) { |
1031 | if (testFeature(X86::FEATURE_MOVBE)) |
1032 | CPU = "silvermont" ; |
1033 | else |
1034 | CPU = "nehalem" ; |
1035 | } else if (testFeature(X86::FEATURE_SSE4_1)) { |
1036 | CPU = "penryn" ; |
1037 | } else if (testFeature(X86::FEATURE_SSSE3)) { |
1038 | if (testFeature(X86::FEATURE_MOVBE)) |
1039 | CPU = "bonnell" ; |
1040 | else |
1041 | CPU = "core2" ; |
1042 | } else if (testFeature(X86::FEATURE_64BIT)) { |
1043 | CPU = "core2" ; |
1044 | } else if (testFeature(X86::FEATURE_SSE3)) { |
1045 | CPU = "yonah" ; |
1046 | } else if (testFeature(X86::FEATURE_SSE2)) { |
1047 | CPU = "pentium-m" ; |
1048 | } else if (testFeature(X86::FEATURE_SSE)) { |
1049 | CPU = "pentium3" ; |
1050 | } else if (testFeature(X86::FEATURE_MMX)) { |
1051 | CPU = "pentium2" ; |
1052 | } else { |
1053 | CPU = "pentiumpro" ; |
1054 | } |
1055 | break; |
1056 | } |
1057 | break; |
1058 | case 15: { |
1059 | if (testFeature(X86::FEATURE_64BIT)) { |
1060 | CPU = "nocona" ; |
1061 | break; |
1062 | } |
1063 | if (testFeature(X86::FEATURE_SSE3)) { |
1064 | CPU = "prescott" ; |
1065 | break; |
1066 | } |
1067 | CPU = "pentium4" ; |
1068 | break; |
1069 | } |
1070 | case 19: |
1071 | switch (Model) { |
1072 | // Diamond Rapids: |
1073 | case 0x01: |
1074 | CPU = "diamondrapids" ; |
1075 | *Type = X86::INTEL_COREI7; |
1076 | *Subtype = X86::INTEL_COREI7_DIAMONDRAPIDS; |
1077 | break; |
1078 | |
1079 | default: // Unknown family 19 CPU. |
1080 | break; |
1081 | } |
1082 | break; |
1083 | default: |
1084 | break; // Unknown. |
1085 | } |
1086 | |
1087 | return CPU; |
1088 | } |
1089 | |
1090 | static const char *getAMDProcessorTypeAndSubtype(unsigned Family, |
1091 | unsigned Model, |
1092 | const unsigned *Features, |
1093 | unsigned *Type, |
1094 | unsigned *Subtype) { |
1095 | const char *CPU = 0; |
1096 | |
1097 | switch (Family) { |
1098 | case 4: |
1099 | CPU = "i486" ; |
1100 | break; |
1101 | case 5: |
1102 | CPU = "pentium" ; |
1103 | switch (Model) { |
1104 | case 6: |
1105 | case 7: |
1106 | CPU = "k6" ; |
1107 | break; |
1108 | case 8: |
1109 | CPU = "k6-2" ; |
1110 | break; |
1111 | case 9: |
1112 | case 13: |
1113 | CPU = "k6-3" ; |
1114 | break; |
1115 | case 10: |
1116 | CPU = "geode" ; |
1117 | break; |
1118 | } |
1119 | break; |
1120 | case 6: |
1121 | if (testFeature(X86::FEATURE_SSE)) { |
1122 | CPU = "athlon-xp" ; |
1123 | break; |
1124 | } |
1125 | CPU = "athlon" ; |
1126 | break; |
1127 | case 15: |
1128 | if (testFeature(X86::FEATURE_SSE3)) { |
1129 | CPU = "k8-sse3" ; |
1130 | break; |
1131 | } |
1132 | CPU = "k8" ; |
1133 | break; |
1134 | case 16: |
1135 | case 18: |
1136 | CPU = "amdfam10" ; |
1137 | *Type = X86::AMDFAM10H; // "amdfam10" |
1138 | switch (Model) { |
1139 | case 2: |
1140 | *Subtype = X86::AMDFAM10H_BARCELONA; |
1141 | break; |
1142 | case 4: |
1143 | *Subtype = X86::AMDFAM10H_SHANGHAI; |
1144 | break; |
1145 | case 8: |
1146 | *Subtype = X86::AMDFAM10H_ISTANBUL; |
1147 | break; |
1148 | } |
1149 | break; |
1150 | case 20: |
1151 | CPU = "btver1" ; |
1152 | *Type = X86::AMD_BTVER1; |
1153 | break; |
1154 | case 21: |
1155 | CPU = "bdver1" ; |
1156 | *Type = X86::AMDFAM15H; |
1157 | if (Model >= 0x60 && Model <= 0x7f) { |
1158 | CPU = "bdver4" ; |
1159 | *Subtype = X86::AMDFAM15H_BDVER4; |
1160 | break; // 60h-7Fh: Excavator |
1161 | } |
1162 | if (Model >= 0x30 && Model <= 0x3f) { |
1163 | CPU = "bdver3" ; |
1164 | *Subtype = X86::AMDFAM15H_BDVER3; |
1165 | break; // 30h-3Fh: Steamroller |
1166 | } |
1167 | if ((Model >= 0x10 && Model <= 0x1f) || Model == 0x02) { |
1168 | CPU = "bdver2" ; |
1169 | *Subtype = X86::AMDFAM15H_BDVER2; |
1170 | break; // 02h, 10h-1Fh: Piledriver |
1171 | } |
1172 | if (Model <= 0x0f) { |
1173 | *Subtype = X86::AMDFAM15H_BDVER1; |
1174 | break; // 00h-0Fh: Bulldozer |
1175 | } |
1176 | break; |
1177 | case 22: |
1178 | CPU = "btver2" ; |
1179 | *Type = X86::AMD_BTVER2; |
1180 | break; |
1181 | case 23: |
1182 | CPU = "znver1" ; |
1183 | *Type = X86::AMDFAM17H; |
1184 | if ((Model >= 0x30 && Model <= 0x3f) || (Model == 0x47) || |
1185 | (Model >= 0x60 && Model <= 0x67) || (Model >= 0x68 && Model <= 0x6f) || |
1186 | (Model >= 0x70 && Model <= 0x7f) || (Model >= 0x84 && Model <= 0x87) || |
1187 | (Model >= 0x90 && Model <= 0x97) || (Model >= 0x98 && Model <= 0x9f) || |
1188 | (Model >= 0xa0 && Model <= 0xaf)) { |
1189 | // Family 17h Models 30h-3Fh (Starship) Zen 2 |
1190 | // Family 17h Models 47h (Cardinal) Zen 2 |
1191 | // Family 17h Models 60h-67h (Renoir) Zen 2 |
1192 | // Family 17h Models 68h-6Fh (Lucienne) Zen 2 |
1193 | // Family 17h Models 70h-7Fh (Matisse) Zen 2 |
1194 | // Family 17h Models 84h-87h (ProjectX) Zen 2 |
1195 | // Family 17h Models 90h-97h (VanGogh) Zen 2 |
1196 | // Family 17h Models 98h-9Fh (Mero) Zen 2 |
1197 | // Family 17h Models A0h-AFh (Mendocino) Zen 2 |
1198 | CPU = "znver2" ; |
1199 | *Subtype = X86::AMDFAM17H_ZNVER2; |
1200 | break; |
1201 | } |
1202 | if ((Model >= 0x10 && Model <= 0x1f) || (Model >= 0x20 && Model <= 0x2f)) { |
1203 | // Family 17h Models 10h-1Fh (Raven1) Zen |
1204 | // Family 17h Models 10h-1Fh (Picasso) Zen+ |
1205 | // Family 17h Models 20h-2Fh (Raven2 x86) Zen |
1206 | *Subtype = X86::AMDFAM17H_ZNVER1; |
1207 | break; |
1208 | } |
1209 | break; |
1210 | case 25: |
1211 | CPU = "znver3" ; |
1212 | *Type = X86::AMDFAM19H; |
1213 | if (Model <= 0x0f || (Model >= 0x20 && Model <= 0x2f) || |
1214 | (Model >= 0x30 && Model <= 0x3f) || (Model >= 0x40 && Model <= 0x4f) || |
1215 | (Model >= 0x50 && Model <= 0x5f)) { |
1216 | // Family 19h Models 00h-0Fh (Genesis, Chagall) Zen 3 |
1217 | // Family 19h Models 20h-2Fh (Vermeer) Zen 3 |
1218 | // Family 19h Models 30h-3Fh (Badami) Zen 3 |
1219 | // Family 19h Models 40h-4Fh (Rembrandt) Zen 3+ |
1220 | // Family 19h Models 50h-5Fh (Cezanne) Zen 3 |
1221 | *Subtype = X86::AMDFAM19H_ZNVER3; |
1222 | break; |
1223 | } |
1224 | if ((Model >= 0x10 && Model <= 0x1f) || (Model >= 0x60 && Model <= 0x6f) || |
1225 | (Model >= 0x70 && Model <= 0x77) || (Model >= 0x78 && Model <= 0x7f) || |
1226 | (Model >= 0xa0 && Model <= 0xaf)) { |
1227 | // Family 19h Models 10h-1Fh (Stones; Storm Peak) Zen 4 |
1228 | // Family 19h Models 60h-6Fh (Raphael) Zen 4 |
1229 | // Family 19h Models 70h-77h (Phoenix, Hawkpoint1) Zen 4 |
1230 | // Family 19h Models 78h-7Fh (Phoenix 2, Hawkpoint2) Zen 4 |
1231 | // Family 19h Models A0h-AFh (Stones-Dense) Zen 4 |
1232 | CPU = "znver4" ; |
1233 | *Subtype = X86::AMDFAM19H_ZNVER4; |
1234 | break; // "znver4" |
1235 | } |
1236 | break; // family 19h |
1237 | case 26: |
1238 | CPU = "znver5" ; |
1239 | *Type = X86::AMDFAM1AH; |
1240 | if (Model <= 0x77) { |
1241 | // Models 00h-0Fh (Breithorn). |
1242 | // Models 10h-1Fh (Breithorn-Dense). |
1243 | // Models 20h-2Fh (Strix 1). |
1244 | // Models 30h-37h (Strix 2). |
1245 | // Models 38h-3Fh (Strix 3). |
1246 | // Models 40h-4Fh (Granite Ridge). |
1247 | // Models 50h-5Fh (Weisshorn). |
1248 | // Models 60h-6Fh (Krackan1). |
1249 | // Models 70h-77h (Sarlak). |
1250 | CPU = "znver5" ; |
1251 | *Subtype = X86::AMDFAM1AH_ZNVER5; |
1252 | break; // "znver5" |
1253 | } |
1254 | break; |
1255 | |
1256 | default: |
1257 | break; // Unknown AMD CPU. |
1258 | } |
1259 | |
1260 | return CPU; |
1261 | } |
1262 | |
1263 | #undef testFeature |
1264 | |
1265 | static void getAvailableFeatures(unsigned ECX, unsigned EDX, unsigned MaxLeaf, |
1266 | unsigned *Features) { |
1267 | unsigned EAX, EBX; |
1268 | |
1269 | auto setFeature = [&](unsigned F) { |
1270 | Features[F / 32] |= 1U << (F % 32); |
1271 | }; |
1272 | |
1273 | if ((EDX >> 15) & 1) |
1274 | setFeature(X86::FEATURE_CMOV); |
1275 | if ((EDX >> 23) & 1) |
1276 | setFeature(X86::FEATURE_MMX); |
1277 | if ((EDX >> 25) & 1) |
1278 | setFeature(X86::FEATURE_SSE); |
1279 | if ((EDX >> 26) & 1) |
1280 | setFeature(X86::FEATURE_SSE2); |
1281 | |
1282 | if ((ECX >> 0) & 1) |
1283 | setFeature(X86::FEATURE_SSE3); |
1284 | if ((ECX >> 1) & 1) |
1285 | setFeature(X86::FEATURE_PCLMUL); |
1286 | if ((ECX >> 9) & 1) |
1287 | setFeature(X86::FEATURE_SSSE3); |
1288 | if ((ECX >> 12) & 1) |
1289 | setFeature(X86::FEATURE_FMA); |
1290 | if ((ECX >> 19) & 1) |
1291 | setFeature(X86::FEATURE_SSE4_1); |
1292 | if ((ECX >> 20) & 1) { |
1293 | setFeature(X86::FEATURE_SSE4_2); |
1294 | setFeature(X86::FEATURE_CRC32); |
1295 | } |
1296 | if ((ECX >> 23) & 1) |
1297 | setFeature(X86::FEATURE_POPCNT); |
1298 | if ((ECX >> 25) & 1) |
1299 | setFeature(X86::FEATURE_AES); |
1300 | |
1301 | if ((ECX >> 22) & 1) |
1302 | setFeature(X86::FEATURE_MOVBE); |
1303 | |
1304 | // If CPUID indicates support for XSAVE, XRESTORE and AVX, and XGETBV |
1305 | // indicates that the AVX registers will be saved and restored on context |
1306 | // switch, then we have full AVX support. |
1307 | const unsigned AVXBits = (1 << 27) | (1 << 28); |
1308 | bool HasAVX = ((ECX & AVXBits) == AVXBits) && !getX86XCR0(rEAX: &EAX, rEDX: &EDX) && |
1309 | ((EAX & 0x6) == 0x6); |
1310 | #if defined(__APPLE__) |
1311 | // Darwin lazily saves the AVX512 context on first use: trust that the OS will |
1312 | // save the AVX512 context if we use AVX512 instructions, even the bit is not |
1313 | // set right now. |
1314 | bool HasAVX512Save = true; |
1315 | #else |
1316 | // AVX512 requires additional context to be saved by the OS. |
1317 | bool HasAVX512Save = HasAVX && ((EAX & 0xe0) == 0xe0); |
1318 | #endif |
1319 | |
1320 | if (HasAVX) |
1321 | setFeature(X86::FEATURE_AVX); |
1322 | |
1323 | bool HasLeaf7 = |
1324 | MaxLeaf >= 0x7 && !getX86CpuIDAndInfoEx(value: 0x7, subleaf: 0x0, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1325 | |
1326 | if (HasLeaf7 && ((EBX >> 3) & 1)) |
1327 | setFeature(X86::FEATURE_BMI); |
1328 | if (HasLeaf7 && ((EBX >> 5) & 1) && HasAVX) |
1329 | setFeature(X86::FEATURE_AVX2); |
1330 | if (HasLeaf7 && ((EBX >> 8) & 1)) |
1331 | setFeature(X86::FEATURE_BMI2); |
1332 | if (HasLeaf7 && ((EBX >> 16) & 1) && HasAVX512Save) { |
1333 | setFeature(X86::FEATURE_AVX512F); |
1334 | setFeature(X86::FEATURE_EVEX512); |
1335 | } |
1336 | if (HasLeaf7 && ((EBX >> 17) & 1) && HasAVX512Save) |
1337 | setFeature(X86::FEATURE_AVX512DQ); |
1338 | if (HasLeaf7 && ((EBX >> 19) & 1)) |
1339 | setFeature(X86::FEATURE_ADX); |
1340 | if (HasLeaf7 && ((EBX >> 21) & 1) && HasAVX512Save) |
1341 | setFeature(X86::FEATURE_AVX512IFMA); |
1342 | if (HasLeaf7 && ((EBX >> 23) & 1)) |
1343 | setFeature(X86::FEATURE_CLFLUSHOPT); |
1344 | if (HasLeaf7 && ((EBX >> 28) & 1) && HasAVX512Save) |
1345 | setFeature(X86::FEATURE_AVX512CD); |
1346 | if (HasLeaf7 && ((EBX >> 29) & 1)) |
1347 | setFeature(X86::FEATURE_SHA); |
1348 | if (HasLeaf7 && ((EBX >> 30) & 1) && HasAVX512Save) |
1349 | setFeature(X86::FEATURE_AVX512BW); |
1350 | if (HasLeaf7 && ((EBX >> 31) & 1) && HasAVX512Save) |
1351 | setFeature(X86::FEATURE_AVX512VL); |
1352 | |
1353 | if (HasLeaf7 && ((ECX >> 1) & 1) && HasAVX512Save) |
1354 | setFeature(X86::FEATURE_AVX512VBMI); |
1355 | if (HasLeaf7 && ((ECX >> 6) & 1) && HasAVX512Save) |
1356 | setFeature(X86::FEATURE_AVX512VBMI2); |
1357 | if (HasLeaf7 && ((ECX >> 8) & 1)) |
1358 | setFeature(X86::FEATURE_GFNI); |
1359 | if (HasLeaf7 && ((ECX >> 10) & 1) && HasAVX) |
1360 | setFeature(X86::FEATURE_VPCLMULQDQ); |
1361 | if (HasLeaf7 && ((ECX >> 11) & 1) && HasAVX512Save) |
1362 | setFeature(X86::FEATURE_AVX512VNNI); |
1363 | if (HasLeaf7 && ((ECX >> 12) & 1) && HasAVX512Save) |
1364 | setFeature(X86::FEATURE_AVX512BITALG); |
1365 | if (HasLeaf7 && ((ECX >> 14) & 1) && HasAVX512Save) |
1366 | setFeature(X86::FEATURE_AVX512VPOPCNTDQ); |
1367 | |
1368 | if (HasLeaf7 && ((EDX >> 2) & 1) && HasAVX512Save) |
1369 | setFeature(X86::FEATURE_AVX5124VNNIW); |
1370 | if (HasLeaf7 && ((EDX >> 3) & 1) && HasAVX512Save) |
1371 | setFeature(X86::FEATURE_AVX5124FMAPS); |
1372 | if (HasLeaf7 && ((EDX >> 8) & 1) && HasAVX512Save) |
1373 | setFeature(X86::FEATURE_AVX512VP2INTERSECT); |
1374 | |
1375 | // EAX from subleaf 0 is the maximum subleaf supported. Some CPUs don't |
1376 | // return all 0s for invalid subleaves so check the limit. |
1377 | bool HasLeaf7Subleaf1 = |
1378 | HasLeaf7 && EAX >= 1 && |
1379 | !getX86CpuIDAndInfoEx(value: 0x7, subleaf: 0x1, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1380 | if (HasLeaf7Subleaf1 && ((EAX >> 5) & 1) && HasAVX512Save) |
1381 | setFeature(X86::FEATURE_AVX512BF16); |
1382 | |
1383 | unsigned MaxExtLevel; |
1384 | getX86CpuIDAndInfo(value: 0x80000000, rEAX: &MaxExtLevel, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1385 | |
1386 | bool HasExtLeaf1 = MaxExtLevel >= 0x80000001 && |
1387 | !getX86CpuIDAndInfo(value: 0x80000001, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1388 | if (HasExtLeaf1 && ((ECX >> 6) & 1)) |
1389 | setFeature(X86::FEATURE_SSE4_A); |
1390 | if (HasExtLeaf1 && ((ECX >> 11) & 1)) |
1391 | setFeature(X86::FEATURE_XOP); |
1392 | if (HasExtLeaf1 && ((ECX >> 16) & 1)) |
1393 | setFeature(X86::FEATURE_FMA4); |
1394 | |
1395 | if (HasExtLeaf1 && ((EDX >> 29) & 1)) |
1396 | setFeature(X86::FEATURE_64BIT); |
1397 | } |
1398 | |
1399 | StringRef sys::getHostCPUName() { |
1400 | unsigned MaxLeaf = 0; |
1401 | const VendorSignatures Vendor = getVendorSignature(MaxLeaf: &MaxLeaf); |
1402 | if (Vendor == VendorSignatures::UNKNOWN) |
1403 | return "generic" ; |
1404 | |
1405 | unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0; |
1406 | getX86CpuIDAndInfo(value: 0x1, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1407 | |
1408 | unsigned Family = 0, Model = 0; |
1409 | unsigned Features[(X86::CPU_FEATURE_MAX + 31) / 32] = {0}; |
1410 | detectX86FamilyModel(EAX, Family: &Family, Model: &Model); |
1411 | getAvailableFeatures(ECX, EDX, MaxLeaf, Features); |
1412 | |
1413 | // These aren't consumed in this file, but we try to keep some source code the |
1414 | // same or similar to compiler-rt. |
1415 | unsigned Type = 0; |
1416 | unsigned Subtype = 0; |
1417 | |
1418 | StringRef CPU; |
1419 | |
1420 | if (Vendor == VendorSignatures::GENUINE_INTEL) { |
1421 | CPU = getIntelProcessorTypeAndSubtype(Family, Model, Features, Type: &Type, |
1422 | Subtype: &Subtype); |
1423 | } else if (Vendor == VendorSignatures::AUTHENTIC_AMD) { |
1424 | CPU = getAMDProcessorTypeAndSubtype(Family, Model, Features, Type: &Type, |
1425 | Subtype: &Subtype); |
1426 | } |
1427 | |
1428 | if (!CPU.empty()) |
1429 | return CPU; |
1430 | |
1431 | return "generic" ; |
1432 | } |
1433 | |
1434 | #elif defined(__APPLE__) && defined(__powerpc__) |
1435 | StringRef sys::getHostCPUName() { |
1436 | host_basic_info_data_t hostInfo; |
1437 | mach_msg_type_number_t infoCount; |
1438 | |
1439 | infoCount = HOST_BASIC_INFO_COUNT; |
1440 | mach_port_t hostPort = mach_host_self(); |
1441 | host_info(hostPort, HOST_BASIC_INFO, (host_info_t)&hostInfo, |
1442 | &infoCount); |
1443 | mach_port_deallocate(mach_task_self(), hostPort); |
1444 | |
1445 | if (hostInfo.cpu_type != CPU_TYPE_POWERPC) |
1446 | return "generic" ; |
1447 | |
1448 | switch (hostInfo.cpu_subtype) { |
1449 | case CPU_SUBTYPE_POWERPC_601: |
1450 | return "601" ; |
1451 | case CPU_SUBTYPE_POWERPC_602: |
1452 | return "602" ; |
1453 | case CPU_SUBTYPE_POWERPC_603: |
1454 | return "603" ; |
1455 | case CPU_SUBTYPE_POWERPC_603e: |
1456 | return "603e" ; |
1457 | case CPU_SUBTYPE_POWERPC_603ev: |
1458 | return "603ev" ; |
1459 | case CPU_SUBTYPE_POWERPC_604: |
1460 | return "604" ; |
1461 | case CPU_SUBTYPE_POWERPC_604e: |
1462 | return "604e" ; |
1463 | case CPU_SUBTYPE_POWERPC_620: |
1464 | return "620" ; |
1465 | case CPU_SUBTYPE_POWERPC_750: |
1466 | return "750" ; |
1467 | case CPU_SUBTYPE_POWERPC_7400: |
1468 | return "7400" ; |
1469 | case CPU_SUBTYPE_POWERPC_7450: |
1470 | return "7450" ; |
1471 | case CPU_SUBTYPE_POWERPC_970: |
1472 | return "970" ; |
1473 | default:; |
1474 | } |
1475 | |
1476 | return "generic" ; |
1477 | } |
1478 | #elif defined(__linux__) && defined(__powerpc__) |
1479 | StringRef sys::getHostCPUName() { |
1480 | std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent(); |
1481 | StringRef Content = P ? P->getBuffer() : "" ; |
1482 | return detail::getHostCPUNameForPowerPC(Content); |
1483 | } |
1484 | #elif defined(__linux__) && (defined(__arm__) || defined(__aarch64__)) |
1485 | StringRef sys::getHostCPUName() { |
1486 | std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent(); |
1487 | StringRef Content = P ? P->getBuffer() : "" ; |
1488 | return detail::getHostCPUNameForARM(Content); |
1489 | } |
1490 | #elif defined(__linux__) && defined(__s390x__) |
1491 | StringRef sys::getHostCPUName() { |
1492 | std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent(); |
1493 | StringRef Content = P ? P->getBuffer() : "" ; |
1494 | return detail::getHostCPUNameForS390x(Content); |
1495 | } |
1496 | #elif defined(__MVS__) |
1497 | StringRef sys::getHostCPUName() { |
1498 | // Get pointer to Communications Vector Table (CVT). |
1499 | // The pointer is located at offset 16 of the Prefixed Save Area (PSA). |
1500 | // It is stored as 31 bit pointer and will be zero-extended to 64 bit. |
1501 | int *StartToCVTOffset = reinterpret_cast<int *>(0x10); |
1502 | // Since its stored as a 31-bit pointer, get the 4 bytes from the start |
1503 | // of address. |
1504 | int ReadValue = *StartToCVTOffset; |
1505 | // Explicitly clear the high order bit. |
1506 | ReadValue = (ReadValue & 0x7FFFFFFF); |
1507 | char *CVT = reinterpret_cast<char *>(ReadValue); |
1508 | // The model number is located in the CVT prefix at offset -6 and stored as |
1509 | // signless packed decimal. |
1510 | uint16_t Id = *(uint16_t *)&CVT[-6]; |
1511 | // Convert number to integer. |
1512 | Id = decodePackedBCD<uint16_t>(Id, false); |
1513 | // Check for vector support. It's stored in field CVTFLAG5 (offset 244), |
1514 | // bit CVTVEF (X'80'). The facilities list is part of the PSA but the vector |
1515 | // extension can only be used if bit CVTVEF is on. |
1516 | bool HaveVectorSupport = CVT[244] & 0x80; |
1517 | return getCPUNameFromS390Model(Id, HaveVectorSupport); |
1518 | } |
1519 | #elif defined(__APPLE__) && (defined(__arm__) || defined(__aarch64__)) |
1520 | // Copied from <mach/machine.h> in the macOS SDK. |
1521 | // |
1522 | // Also available here, though usually not as up-to-date: |
1523 | // https://github.com/apple-oss-distributions/xnu/blob/xnu-11215.41.3/osfmk/mach/machine.h#L403-L452. |
1524 | #define CPUFAMILY_UNKNOWN 0 |
1525 | #define CPUFAMILY_ARM_9 0xe73283ae |
1526 | #define CPUFAMILY_ARM_11 0x8ff620d8 |
1527 | #define CPUFAMILY_ARM_XSCALE 0x53b005f5 |
1528 | #define CPUFAMILY_ARM_12 0xbd1b0ae9 |
1529 | #define CPUFAMILY_ARM_13 0x0cc90e64 |
1530 | #define CPUFAMILY_ARM_14 0x96077ef1 |
1531 | #define CPUFAMILY_ARM_15 0xa8511bca |
1532 | #define CPUFAMILY_ARM_SWIFT 0x1e2d6381 |
1533 | #define CPUFAMILY_ARM_CYCLONE 0x37a09642 |
1534 | #define CPUFAMILY_ARM_TYPHOON 0x2c91a47e |
1535 | #define CPUFAMILY_ARM_TWISTER 0x92fb37c8 |
1536 | #define CPUFAMILY_ARM_HURRICANE 0x67ceee93 |
1537 | #define CPUFAMILY_ARM_MONSOON_MISTRAL 0xe81e7ef6 |
1538 | #define CPUFAMILY_ARM_VORTEX_TEMPEST 0x07d34b9f |
1539 | #define CPUFAMILY_ARM_LIGHTNING_THUNDER 0x462504d2 |
1540 | #define CPUFAMILY_ARM_FIRESTORM_ICESTORM 0x1b588bb3 |
1541 | #define CPUFAMILY_ARM_BLIZZARD_AVALANCHE 0xda33d83d |
1542 | #define CPUFAMILY_ARM_EVEREST_SAWTOOTH 0x8765edea |
1543 | #define CPUFAMILY_ARM_IBIZA 0xfa33415e |
1544 | #define CPUFAMILY_ARM_PALMA 0x72015832 |
1545 | #define CPUFAMILY_ARM_COLL 0x2876f5b5 |
1546 | #define CPUFAMILY_ARM_LOBOS 0x5f4dea93 |
1547 | #define CPUFAMILY_ARM_DONAN 0x6f5129ac |
1548 | #define CPUFAMILY_ARM_BRAVA 0x17d5b93a |
1549 | #define CPUFAMILY_ARM_TAHITI 0x75d4acb9 |
1550 | #define CPUFAMILY_ARM_TUPAI 0x204526d0 |
1551 | |
1552 | StringRef sys::getHostCPUName() { |
1553 | uint32_t Family; |
1554 | size_t Length = sizeof(Family); |
1555 | sysctlbyname("hw.cpufamily" , &Family, &Length, NULL, 0); |
1556 | |
1557 | // This is found by testing on actual hardware, and by looking at: |
1558 | // https://github.com/apple-oss-distributions/xnu/blob/xnu-11215.41.3/osfmk/arm/cpuid.c#L109-L231. |
1559 | // |
1560 | // Another great resource is |
1561 | // https://github.com/AsahiLinux/docs/wiki/Codenames. |
1562 | // |
1563 | // NOTE: We choose to return `apple-mX` instead of `apple-aX`, since the M1, |
1564 | // M2, M3 etc. aliases are more widely known to users than A14, A15, A16 etc. |
1565 | // (and this code is basically only used on host macOS anyways). |
1566 | switch (Family) { |
1567 | case CPUFAMILY_UNKNOWN: |
1568 | return "generic" ; |
1569 | case CPUFAMILY_ARM_9: |
1570 | return "arm920t" ; // or arm926ej-s |
1571 | case CPUFAMILY_ARM_11: |
1572 | return "arm1136jf-s" ; |
1573 | case CPUFAMILY_ARM_XSCALE: |
1574 | return "xscale" ; |
1575 | case CPUFAMILY_ARM_12: // Seems unused by the kernel |
1576 | return "generic" ; |
1577 | case CPUFAMILY_ARM_13: |
1578 | return "cortex-a8" ; |
1579 | case CPUFAMILY_ARM_14: |
1580 | return "cortex-a9" ; |
1581 | case CPUFAMILY_ARM_15: |
1582 | return "cortex-a7" ; |
1583 | case CPUFAMILY_ARM_SWIFT: |
1584 | return "swift" ; |
1585 | case CPUFAMILY_ARM_CYCLONE: |
1586 | return "apple-a7" ; |
1587 | case CPUFAMILY_ARM_TYPHOON: |
1588 | return "apple-a8" ; |
1589 | case CPUFAMILY_ARM_TWISTER: |
1590 | return "apple-a9" ; |
1591 | case CPUFAMILY_ARM_HURRICANE: |
1592 | return "apple-a10" ; |
1593 | case CPUFAMILY_ARM_MONSOON_MISTRAL: |
1594 | return "apple-a11" ; |
1595 | case CPUFAMILY_ARM_VORTEX_TEMPEST: |
1596 | return "apple-a12" ; |
1597 | case CPUFAMILY_ARM_LIGHTNING_THUNDER: |
1598 | return "apple-a13" ; |
1599 | case CPUFAMILY_ARM_FIRESTORM_ICESTORM: // A14 / M1 |
1600 | return "apple-m1" ; |
1601 | case CPUFAMILY_ARM_BLIZZARD_AVALANCHE: // A15 / M2 |
1602 | return "apple-m2" ; |
1603 | case CPUFAMILY_ARM_EVEREST_SAWTOOTH: // A16 |
1604 | case CPUFAMILY_ARM_IBIZA: // M3 |
1605 | case CPUFAMILY_ARM_PALMA: // M3 Max |
1606 | case CPUFAMILY_ARM_LOBOS: // M3 Pro |
1607 | return "apple-m3" ; |
1608 | case CPUFAMILY_ARM_COLL: // A17 Pro |
1609 | return "apple-a17" ; |
1610 | case CPUFAMILY_ARM_DONAN: // M4 |
1611 | case CPUFAMILY_ARM_BRAVA: // M4 Max |
1612 | case CPUFAMILY_ARM_TAHITI: // A18 Pro |
1613 | case CPUFAMILY_ARM_TUPAI: // A18 |
1614 | return "apple-m4" ; |
1615 | default: |
1616 | // Default to the newest CPU we know about. |
1617 | return "apple-m4" ; |
1618 | } |
1619 | } |
1620 | #elif defined(_AIX) |
1621 | StringRef sys::getHostCPUName() { |
1622 | switch (_system_configuration.implementation) { |
1623 | case POWER_4: |
1624 | if (_system_configuration.version == PV_4_3) |
1625 | return "970" ; |
1626 | return "pwr4" ; |
1627 | case POWER_5: |
1628 | if (_system_configuration.version == PV_5) |
1629 | return "pwr5" ; |
1630 | return "pwr5x" ; |
1631 | case POWER_6: |
1632 | if (_system_configuration.version == PV_6_Compat) |
1633 | return "pwr6" ; |
1634 | return "pwr6x" ; |
1635 | case POWER_7: |
1636 | return "pwr7" ; |
1637 | case POWER_8: |
1638 | return "pwr8" ; |
1639 | case POWER_9: |
1640 | return "pwr9" ; |
1641 | // TODO: simplify this once the macro is available in all OS levels. |
1642 | #ifdef POWER_10 |
1643 | case POWER_10: |
1644 | #else |
1645 | case 0x40000: |
1646 | #endif |
1647 | return "pwr10" ; |
1648 | #ifdef POWER_11 |
1649 | case POWER_11: |
1650 | #else |
1651 | case 0x80000: |
1652 | #endif |
1653 | return "pwr11" ; |
1654 | default: |
1655 | return "generic" ; |
1656 | } |
1657 | } |
1658 | #elif defined(__loongarch__) |
1659 | StringRef sys::getHostCPUName() { |
1660 | // Use processor id to detect cpu name. |
1661 | uint32_t processor_id; |
1662 | __asm__("cpucfg %[prid], $zero\n\t" : [prid] "=r" (processor_id)); |
1663 | // Refer PRID_SERIES_MASK in linux kernel: arch/loongarch/include/asm/cpu.h. |
1664 | switch (processor_id & 0xf000) { |
1665 | case 0xc000: // Loongson 64bit, 4-issue |
1666 | return "la464" ; |
1667 | case 0xd000: // Loongson 64bit, 6-issue |
1668 | return "la664" ; |
1669 | // TODO: Others. |
1670 | default: |
1671 | break; |
1672 | } |
1673 | return "generic" ; |
1674 | } |
1675 | #elif defined(__riscv) |
1676 | #if defined(__linux__) |
1677 | // struct riscv_hwprobe |
1678 | struct RISCVHwProbe { |
1679 | int64_t Key; |
1680 | uint64_t Value; |
1681 | }; |
1682 | #endif |
1683 | |
1684 | StringRef sys::getHostCPUName() { |
1685 | #if defined(__linux__) |
1686 | // Try the hwprobe way first. |
1687 | RISCVHwProbe Query[]{{/*RISCV_HWPROBE_KEY_MVENDORID=*/0, 0}, |
1688 | {/*RISCV_HWPROBE_KEY_MARCHID=*/1, 0}, |
1689 | {/*RISCV_HWPROBE_KEY_MIMPID=*/2, 0}}; |
1690 | int Ret = syscall(/*__NR_riscv_hwprobe=*/258, /*pairs=*/Query, |
1691 | /*pair_count=*/std::size(Query), /*cpu_count=*/0, |
1692 | /*cpus=*/0, /*flags=*/0); |
1693 | if (Ret == 0) { |
1694 | RISCV::CPUModel Model{static_cast<uint32_t>(Query[0].Value), Query[1].Value, |
1695 | Query[2].Value}; |
1696 | StringRef Name = RISCV::getCPUNameFromCPUModel(Model); |
1697 | if (!Name.empty()) |
1698 | return Name; |
1699 | } |
1700 | |
1701 | // Then try the cpuinfo way. |
1702 | std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent(); |
1703 | StringRef Content = P ? P->getBuffer() : "" ; |
1704 | StringRef Name = detail::getHostCPUNameForRISCV(Content); |
1705 | if (!Name.empty()) |
1706 | return Name; |
1707 | #endif |
1708 | #if __riscv_xlen == 64 |
1709 | return "generic-rv64" ; |
1710 | #elif __riscv_xlen == 32 |
1711 | return "generic-rv32" ; |
1712 | #else |
1713 | #error "Unhandled value of __riscv_xlen" |
1714 | #endif |
1715 | } |
1716 | #elif defined(__sparc__) |
1717 | #if defined(__linux__) |
1718 | StringRef sys::detail::getHostCPUNameForSPARC(StringRef ProcCpuinfoContent) { |
1719 | SmallVector<StringRef> Lines; |
1720 | ProcCpuinfoContent.split(Lines, '\n'); |
1721 | |
1722 | // Look for cpu line to determine cpu name |
1723 | StringRef Cpu; |
1724 | for (unsigned I = 0, E = Lines.size(); I != E; ++I) { |
1725 | if (Lines[I].starts_with("cpu" )) { |
1726 | Cpu = Lines[I].substr(5).ltrim("\t :" ); |
1727 | break; |
1728 | } |
1729 | } |
1730 | |
1731 | return StringSwitch<const char *>(Cpu) |
1732 | .StartsWith("SuperSparc" , "supersparc" ) |
1733 | .StartsWith("HyperSparc" , "hypersparc" ) |
1734 | .StartsWith("SpitFire" , "ultrasparc" ) |
1735 | .StartsWith("BlackBird" , "ultrasparc" ) |
1736 | .StartsWith("Sabre" , " ultrasparc" ) |
1737 | .StartsWith("Hummingbird" , "ultrasparc" ) |
1738 | .StartsWith("Cheetah" , "ultrasparc3" ) |
1739 | .StartsWith("Jalapeno" , "ultrasparc3" ) |
1740 | .StartsWith("Jaguar" , "ultrasparc3" ) |
1741 | .StartsWith("Panther" , "ultrasparc3" ) |
1742 | .StartsWith("Serrano" , "ultrasparc3" ) |
1743 | .StartsWith("UltraSparc T1" , "niagara" ) |
1744 | .StartsWith("UltraSparc T2" , "niagara2" ) |
1745 | .StartsWith("UltraSparc T3" , "niagara3" ) |
1746 | .StartsWith("UltraSparc T4" , "niagara4" ) |
1747 | .StartsWith("UltraSparc T5" , "niagara4" ) |
1748 | .StartsWith("LEON" , "leon3" ) |
1749 | // niagara7/m8 not supported by LLVM yet. |
1750 | .StartsWith("SPARC-M7" , "niagara4" /* "niagara7" */) |
1751 | .StartsWith("SPARC-S7" , "niagara4" /* "niagara7" */) |
1752 | .StartsWith("SPARC-M8" , "niagara4" /* "m8" */) |
1753 | .Default("generic" ); |
1754 | } |
1755 | #endif |
1756 | |
1757 | StringRef sys::getHostCPUName() { |
1758 | #if defined(__linux__) |
1759 | std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent(); |
1760 | StringRef Content = P ? P->getBuffer() : "" ; |
1761 | return detail::getHostCPUNameForSPARC(Content); |
1762 | #elif defined(__sun__) && defined(__svr4__) |
1763 | char *buf = NULL; |
1764 | kstat_ctl_t *kc; |
1765 | kstat_t *ksp; |
1766 | kstat_named_t *brand = NULL; |
1767 | |
1768 | kc = kstat_open(); |
1769 | if (kc != NULL) { |
1770 | ksp = kstat_lookup(kc, const_cast<char *>("cpu_info" ), -1, NULL); |
1771 | if (ksp != NULL && kstat_read(kc, ksp, NULL) != -1 && |
1772 | ksp->ks_type == KSTAT_TYPE_NAMED) |
1773 | brand = |
1774 | (kstat_named_t *)kstat_data_lookup(ksp, const_cast<char *>("brand" )); |
1775 | if (brand != NULL && brand->data_type == KSTAT_DATA_STRING) |
1776 | buf = KSTAT_NAMED_STR_PTR(brand); |
1777 | } |
1778 | kstat_close(kc); |
1779 | |
1780 | return StringSwitch<const char *>(buf) |
1781 | .Case("TMS390S10" , "supersparc" ) // Texas Instruments microSPARC I |
1782 | .Case("TMS390Z50" , "supersparc" ) // Texas Instruments SuperSPARC I |
1783 | .Case("TMS390Z55" , |
1784 | "supersparc" ) // Texas Instruments SuperSPARC I with SuperCache |
1785 | .Case("MB86904" , "supersparc" ) // Fujitsu microSPARC II |
1786 | .Case("MB86907" , "supersparc" ) // Fujitsu TurboSPARC |
1787 | .Case("RT623" , "hypersparc" ) // Ross hyperSPARC |
1788 | .Case("RT625" , "hypersparc" ) |
1789 | .Case("RT626" , "hypersparc" ) |
1790 | .Case("UltraSPARC-I" , "ultrasparc" ) |
1791 | .Case("UltraSPARC-II" , "ultrasparc" ) |
1792 | .Case("UltraSPARC-IIe" , "ultrasparc" ) |
1793 | .Case("UltraSPARC-IIi" , "ultrasparc" ) |
1794 | .Case("SPARC64-III" , "ultrasparc" ) |
1795 | .Case("SPARC64-IV" , "ultrasparc" ) |
1796 | .Case("UltraSPARC-III" , "ultrasparc3" ) |
1797 | .Case("UltraSPARC-III+" , "ultrasparc3" ) |
1798 | .Case("UltraSPARC-IIIi" , "ultrasparc3" ) |
1799 | .Case("UltraSPARC-IIIi+" , "ultrasparc3" ) |
1800 | .Case("UltraSPARC-IV" , "ultrasparc3" ) |
1801 | .Case("UltraSPARC-IV+" , "ultrasparc3" ) |
1802 | .Case("SPARC64-V" , "ultrasparc3" ) |
1803 | .Case("SPARC64-VI" , "ultrasparc3" ) |
1804 | .Case("SPARC64-VII" , "ultrasparc3" ) |
1805 | .Case("UltraSPARC-T1" , "niagara" ) |
1806 | .Case("UltraSPARC-T2" , "niagara2" ) |
1807 | .Case("UltraSPARC-T2" , "niagara2" ) |
1808 | .Case("UltraSPARC-T2+" , "niagara2" ) |
1809 | .Case("SPARC-T3" , "niagara3" ) |
1810 | .Case("SPARC-T4" , "niagara4" ) |
1811 | .Case("SPARC-T5" , "niagara4" ) |
1812 | // niagara7/m8 not supported by LLVM yet. |
1813 | .Case("SPARC-M7" , "niagara4" /* "niagara7" */) |
1814 | .Case("SPARC-S7" , "niagara4" /* "niagara7" */) |
1815 | .Case("SPARC-M8" , "niagara4" /* "m8" */) |
1816 | .Default("generic" ); |
1817 | #else |
1818 | return "generic" ; |
1819 | #endif |
1820 | } |
1821 | #else |
1822 | StringRef sys::getHostCPUName() { return "generic" ; } |
1823 | namespace llvm { |
1824 | namespace sys { |
1825 | namespace detail { |
1826 | namespace x86 { |
1827 | |
1828 | VendorSignatures getVendorSignature(unsigned *MaxLeaf) { |
1829 | return VendorSignatures::UNKNOWN; |
1830 | } |
1831 | |
1832 | } // namespace x86 |
1833 | } // namespace detail |
1834 | } // namespace sys |
1835 | } // namespace llvm |
1836 | #endif |
1837 | |
1838 | #if defined(__i386__) || defined(_M_IX86) || \ |
1839 | defined(__x86_64__) || defined(_M_X64) |
1840 | const StringMap<bool> sys::getHostCPUFeatures() { |
1841 | unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0; |
1842 | unsigned MaxLevel; |
1843 | StringMap<bool> Features; |
1844 | |
1845 | if (getX86CpuIDAndInfo(value: 0, rEAX: &MaxLevel, rEBX: &EBX, rECX: &ECX, rEDX: &EDX) || MaxLevel < 1) |
1846 | return Features; |
1847 | |
1848 | getX86CpuIDAndInfo(value: 1, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1849 | |
1850 | Features["cx8" ] = (EDX >> 8) & 1; |
1851 | Features["cmov" ] = (EDX >> 15) & 1; |
1852 | Features["mmx" ] = (EDX >> 23) & 1; |
1853 | Features["fxsr" ] = (EDX >> 24) & 1; |
1854 | Features["sse" ] = (EDX >> 25) & 1; |
1855 | Features["sse2" ] = (EDX >> 26) & 1; |
1856 | |
1857 | Features["sse3" ] = (ECX >> 0) & 1; |
1858 | Features["pclmul" ] = (ECX >> 1) & 1; |
1859 | Features["ssse3" ] = (ECX >> 9) & 1; |
1860 | Features["cx16" ] = (ECX >> 13) & 1; |
1861 | Features["sse4.1" ] = (ECX >> 19) & 1; |
1862 | Features["sse4.2" ] = (ECX >> 20) & 1; |
1863 | Features["crc32" ] = Features["sse4.2" ]; |
1864 | Features["movbe" ] = (ECX >> 22) & 1; |
1865 | Features["popcnt" ] = (ECX >> 23) & 1; |
1866 | Features["aes" ] = (ECX >> 25) & 1; |
1867 | Features["rdrnd" ] = (ECX >> 30) & 1; |
1868 | |
1869 | // If CPUID indicates support for XSAVE, XRESTORE and AVX, and XGETBV |
1870 | // indicates that the AVX registers will be saved and restored on context |
1871 | // switch, then we have full AVX support. |
1872 | bool HasXSave = ((ECX >> 27) & 1) && !getX86XCR0(rEAX: &EAX, rEDX: &EDX); |
1873 | bool HasAVXSave = HasXSave && ((ECX >> 28) & 1) && ((EAX & 0x6) == 0x6); |
1874 | #if defined(__APPLE__) |
1875 | // Darwin lazily saves the AVX512 context on first use: trust that the OS will |
1876 | // save the AVX512 context if we use AVX512 instructions, even the bit is not |
1877 | // set right now. |
1878 | bool HasAVX512Save = true; |
1879 | #else |
1880 | // AVX512 requires additional context to be saved by the OS. |
1881 | bool HasAVX512Save = HasAVXSave && ((EAX & 0xe0) == 0xe0); |
1882 | #endif |
1883 | // AMX requires additional context to be saved by the OS. |
1884 | const unsigned AMXBits = (1 << 17) | (1 << 18); |
1885 | bool HasAMXSave = HasXSave && ((EAX & AMXBits) == AMXBits); |
1886 | |
1887 | Features["avx" ] = HasAVXSave; |
1888 | Features["fma" ] = ((ECX >> 12) & 1) && HasAVXSave; |
1889 | // Only enable XSAVE if OS has enabled support for saving YMM state. |
1890 | Features["xsave" ] = ((ECX >> 26) & 1) && HasAVXSave; |
1891 | Features["f16c" ] = ((ECX >> 29) & 1) && HasAVXSave; |
1892 | |
1893 | unsigned MaxExtLevel; |
1894 | getX86CpuIDAndInfo(value: 0x80000000, rEAX: &MaxExtLevel, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1895 | |
1896 | bool HasExtLeaf1 = MaxExtLevel >= 0x80000001 && |
1897 | !getX86CpuIDAndInfo(value: 0x80000001, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1898 | Features["sahf" ] = HasExtLeaf1 && ((ECX >> 0) & 1); |
1899 | Features["lzcnt" ] = HasExtLeaf1 && ((ECX >> 5) & 1); |
1900 | Features["sse4a" ] = HasExtLeaf1 && ((ECX >> 6) & 1); |
1901 | Features["prfchw" ] = HasExtLeaf1 && ((ECX >> 8) & 1); |
1902 | Features["xop" ] = HasExtLeaf1 && ((ECX >> 11) & 1) && HasAVXSave; |
1903 | Features["lwp" ] = HasExtLeaf1 && ((ECX >> 15) & 1); |
1904 | Features["fma4" ] = HasExtLeaf1 && ((ECX >> 16) & 1) && HasAVXSave; |
1905 | Features["tbm" ] = HasExtLeaf1 && ((ECX >> 21) & 1); |
1906 | Features["mwaitx" ] = HasExtLeaf1 && ((ECX >> 29) & 1); |
1907 | |
1908 | Features["64bit" ] = HasExtLeaf1 && ((EDX >> 29) & 1); |
1909 | |
1910 | // Miscellaneous memory related features, detected by |
1911 | // using the 0x80000008 leaf of the CPUID instruction |
1912 | bool HasExtLeaf8 = MaxExtLevel >= 0x80000008 && |
1913 | !getX86CpuIDAndInfo(value: 0x80000008, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1914 | Features["clzero" ] = HasExtLeaf8 && ((EBX >> 0) & 1); |
1915 | Features["rdpru" ] = HasExtLeaf8 && ((EBX >> 4) & 1); |
1916 | Features["wbnoinvd" ] = HasExtLeaf8 && ((EBX >> 9) & 1); |
1917 | |
1918 | bool HasLeaf7 = |
1919 | MaxLevel >= 7 && !getX86CpuIDAndInfoEx(value: 0x7, subleaf: 0x0, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1920 | |
1921 | Features["fsgsbase" ] = HasLeaf7 && ((EBX >> 0) & 1); |
1922 | Features["sgx" ] = HasLeaf7 && ((EBX >> 2) & 1); |
1923 | Features["bmi" ] = HasLeaf7 && ((EBX >> 3) & 1); |
1924 | // AVX2 is only supported if we have the OS save support from AVX. |
1925 | Features["avx2" ] = HasLeaf7 && ((EBX >> 5) & 1) && HasAVXSave; |
1926 | Features["bmi2" ] = HasLeaf7 && ((EBX >> 8) & 1); |
1927 | Features["invpcid" ] = HasLeaf7 && ((EBX >> 10) & 1); |
1928 | Features["rtm" ] = HasLeaf7 && ((EBX >> 11) & 1); |
1929 | // AVX512 is only supported if the OS supports the context save for it. |
1930 | Features["avx512f" ] = HasLeaf7 && ((EBX >> 16) & 1) && HasAVX512Save; |
1931 | if (Features["avx512f" ]) |
1932 | Features["evex512" ] = true; |
1933 | Features["avx512dq" ] = HasLeaf7 && ((EBX >> 17) & 1) && HasAVX512Save; |
1934 | Features["rdseed" ] = HasLeaf7 && ((EBX >> 18) & 1); |
1935 | Features["adx" ] = HasLeaf7 && ((EBX >> 19) & 1); |
1936 | Features["avx512ifma" ] = HasLeaf7 && ((EBX >> 21) & 1) && HasAVX512Save; |
1937 | Features["clflushopt" ] = HasLeaf7 && ((EBX >> 23) & 1); |
1938 | Features["clwb" ] = HasLeaf7 && ((EBX >> 24) & 1); |
1939 | Features["avx512cd" ] = HasLeaf7 && ((EBX >> 28) & 1) && HasAVX512Save; |
1940 | Features["sha" ] = HasLeaf7 && ((EBX >> 29) & 1); |
1941 | Features["avx512bw" ] = HasLeaf7 && ((EBX >> 30) & 1) && HasAVX512Save; |
1942 | Features["avx512vl" ] = HasLeaf7 && ((EBX >> 31) & 1) && HasAVX512Save; |
1943 | |
1944 | Features["avx512vbmi" ] = HasLeaf7 && ((ECX >> 1) & 1) && HasAVX512Save; |
1945 | Features["pku" ] = HasLeaf7 && ((ECX >> 4) & 1); |
1946 | Features["waitpkg" ] = HasLeaf7 && ((ECX >> 5) & 1); |
1947 | Features["avx512vbmi2" ] = HasLeaf7 && ((ECX >> 6) & 1) && HasAVX512Save; |
1948 | Features["shstk" ] = HasLeaf7 && ((ECX >> 7) & 1); |
1949 | Features["gfni" ] = HasLeaf7 && ((ECX >> 8) & 1); |
1950 | Features["vaes" ] = HasLeaf7 && ((ECX >> 9) & 1) && HasAVXSave; |
1951 | Features["vpclmulqdq" ] = HasLeaf7 && ((ECX >> 10) & 1) && HasAVXSave; |
1952 | Features["avx512vnni" ] = HasLeaf7 && ((ECX >> 11) & 1) && HasAVX512Save; |
1953 | Features["avx512bitalg" ] = HasLeaf7 && ((ECX >> 12) & 1) && HasAVX512Save; |
1954 | Features["avx512vpopcntdq" ] = HasLeaf7 && ((ECX >> 14) & 1) && HasAVX512Save; |
1955 | Features["rdpid" ] = HasLeaf7 && ((ECX >> 22) & 1); |
1956 | Features["kl" ] = HasLeaf7 && ((ECX >> 23) & 1); // key locker |
1957 | Features["cldemote" ] = HasLeaf7 && ((ECX >> 25) & 1); |
1958 | Features["movdiri" ] = HasLeaf7 && ((ECX >> 27) & 1); |
1959 | Features["movdir64b" ] = HasLeaf7 && ((ECX >> 28) & 1); |
1960 | Features["enqcmd" ] = HasLeaf7 && ((ECX >> 29) & 1); |
1961 | |
1962 | Features["uintr" ] = HasLeaf7 && ((EDX >> 5) & 1); |
1963 | Features["avx512vp2intersect" ] = |
1964 | HasLeaf7 && ((EDX >> 8) & 1) && HasAVX512Save; |
1965 | Features["serialize" ] = HasLeaf7 && ((EDX >> 14) & 1); |
1966 | Features["tsxldtrk" ] = HasLeaf7 && ((EDX >> 16) & 1); |
1967 | // There are two CPUID leafs which information associated with the pconfig |
1968 | // instruction: |
1969 | // EAX=0x7, ECX=0x0 indicates the availability of the instruction (via the 18th |
1970 | // bit of EDX), while the EAX=0x1b leaf returns information on the |
1971 | // availability of specific pconfig leafs. |
1972 | // The target feature here only refers to the the first of these two. |
1973 | // Users might need to check for the availability of specific pconfig |
1974 | // leaves using cpuid, since that information is ignored while |
1975 | // detecting features using the "-march=native" flag. |
1976 | // For more info, see X86 ISA docs. |
1977 | Features["pconfig" ] = HasLeaf7 && ((EDX >> 18) & 1); |
1978 | Features["amx-bf16" ] = HasLeaf7 && ((EDX >> 22) & 1) && HasAMXSave; |
1979 | Features["avx512fp16" ] = HasLeaf7 && ((EDX >> 23) & 1) && HasAVX512Save; |
1980 | Features["amx-tile" ] = HasLeaf7 && ((EDX >> 24) & 1) && HasAMXSave; |
1981 | Features["amx-int8" ] = HasLeaf7 && ((EDX >> 25) & 1) && HasAMXSave; |
1982 | // EAX from subleaf 0 is the maximum subleaf supported. Some CPUs don't |
1983 | // return all 0s for invalid subleaves so check the limit. |
1984 | bool HasLeaf7Subleaf1 = |
1985 | HasLeaf7 && EAX >= 1 && |
1986 | !getX86CpuIDAndInfoEx(value: 0x7, subleaf: 0x1, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
1987 | Features["sha512" ] = HasLeaf7Subleaf1 && ((EAX >> 0) & 1); |
1988 | Features["sm3" ] = HasLeaf7Subleaf1 && ((EAX >> 1) & 1); |
1989 | Features["sm4" ] = HasLeaf7Subleaf1 && ((EAX >> 2) & 1); |
1990 | Features["raoint" ] = HasLeaf7Subleaf1 && ((EAX >> 3) & 1); |
1991 | Features["avxvnni" ] = HasLeaf7Subleaf1 && ((EAX >> 4) & 1) && HasAVXSave; |
1992 | Features["avx512bf16" ] = HasLeaf7Subleaf1 && ((EAX >> 5) & 1) && HasAVX512Save; |
1993 | Features["amx-fp16" ] = HasLeaf7Subleaf1 && ((EAX >> 21) & 1) && HasAMXSave; |
1994 | Features["cmpccxadd" ] = HasLeaf7Subleaf1 && ((EAX >> 7) & 1); |
1995 | Features["hreset" ] = HasLeaf7Subleaf1 && ((EAX >> 22) & 1); |
1996 | Features["avxifma" ] = HasLeaf7Subleaf1 && ((EAX >> 23) & 1) && HasAVXSave; |
1997 | Features["movrs" ] = HasLeaf7Subleaf1 && ((EAX >> 31) & 1); |
1998 | Features["avxvnniint8" ] = HasLeaf7Subleaf1 && ((EDX >> 4) & 1) && HasAVXSave; |
1999 | Features["avxneconvert" ] = HasLeaf7Subleaf1 && ((EDX >> 5) & 1) && HasAVXSave; |
2000 | Features["amx-complex" ] = HasLeaf7Subleaf1 && ((EDX >> 8) & 1) && HasAMXSave; |
2001 | Features["avxvnniint16" ] = HasLeaf7Subleaf1 && ((EDX >> 10) & 1) && HasAVXSave; |
2002 | Features["prefetchi" ] = HasLeaf7Subleaf1 && ((EDX >> 14) & 1); |
2003 | Features["usermsr" ] = HasLeaf7Subleaf1 && ((EDX >> 15) & 1); |
2004 | bool HasAVX10 = HasLeaf7Subleaf1 && ((EDX >> 19) & 1); |
2005 | bool HasAPXF = HasLeaf7Subleaf1 && ((EDX >> 21) & 1); |
2006 | Features["egpr" ] = HasAPXF; |
2007 | Features["push2pop2" ] = HasAPXF; |
2008 | Features["ppx" ] = HasAPXF; |
2009 | Features["ndd" ] = HasAPXF; |
2010 | Features["ccmp" ] = HasAPXF; |
2011 | Features["nf" ] = HasAPXF; |
2012 | Features["cf" ] = HasAPXF; |
2013 | Features["zu" ] = HasAPXF; |
2014 | |
2015 | bool HasLeafD = MaxLevel >= 0xd && |
2016 | !getX86CpuIDAndInfoEx(value: 0xd, subleaf: 0x1, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
2017 | |
2018 | // Only enable XSAVE if OS has enabled support for saving YMM state. |
2019 | Features["xsaveopt" ] = HasLeafD && ((EAX >> 0) & 1) && HasAVXSave; |
2020 | Features["xsavec" ] = HasLeafD && ((EAX >> 1) & 1) && HasAVXSave; |
2021 | Features["xsaves" ] = HasLeafD && ((EAX >> 3) & 1) && HasAVXSave; |
2022 | |
2023 | bool HasLeaf14 = MaxLevel >= 0x14 && |
2024 | !getX86CpuIDAndInfoEx(value: 0x14, subleaf: 0x0, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
2025 | |
2026 | Features["ptwrite" ] = HasLeaf14 && ((EBX >> 4) & 1); |
2027 | |
2028 | bool HasLeaf19 = |
2029 | MaxLevel >= 0x19 && !getX86CpuIDAndInfo(value: 0x19, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
2030 | Features["widekl" ] = HasLeaf7 && HasLeaf19 && ((EBX >> 2) & 1); |
2031 | |
2032 | bool HasLeaf1E = MaxLevel >= 0x1e && |
2033 | !getX86CpuIDAndInfoEx(value: 0x1e, subleaf: 0x1, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
2034 | Features["amx-fp8" ] = HasLeaf1E && ((EAX >> 4) & 1) && HasAMXSave; |
2035 | Features["amx-transpose" ] = HasLeaf1E && ((EAX >> 5) & 1) && HasAMXSave; |
2036 | Features["amx-tf32" ] = HasLeaf1E && ((EAX >> 6) & 1) && HasAMXSave; |
2037 | Features["amx-avx512" ] = HasLeaf1E && ((EAX >> 7) & 1) && HasAMXSave; |
2038 | Features["amx-movrs" ] = HasLeaf1E && ((EAX >> 8) & 1) && HasAMXSave; |
2039 | |
2040 | bool HasLeaf24 = |
2041 | MaxLevel >= 0x24 && !getX86CpuIDAndInfo(value: 0x24, rEAX: &EAX, rEBX: &EBX, rECX: &ECX, rEDX: &EDX); |
2042 | |
2043 | int AVX10Ver = HasLeaf24 && (EBX & 0xff); |
2044 | int Has512Len = HasLeaf24 && ((EBX >> 18) & 1); |
2045 | Features["avx10.1-256" ] = HasAVX10 && AVX10Ver >= 1; |
2046 | Features["avx10.1-512" ] = HasAVX10 && AVX10Ver >= 1 && Has512Len; |
2047 | Features["avx10.2-256" ] = HasAVX10 && AVX10Ver >= 2; |
2048 | Features["avx10.2-512" ] = HasAVX10 && AVX10Ver >= 2 && Has512Len; |
2049 | |
2050 | return Features; |
2051 | } |
2052 | #elif defined(__linux__) && (defined(__arm__) || defined(__aarch64__)) |
2053 | const StringMap<bool> sys::getHostCPUFeatures() { |
2054 | StringMap<bool> Features; |
2055 | std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent(); |
2056 | if (!P) |
2057 | return Features; |
2058 | |
2059 | SmallVector<StringRef, 32> Lines; |
2060 | P->getBuffer().split(Lines, '\n'); |
2061 | |
2062 | SmallVector<StringRef, 32> CPUFeatures; |
2063 | |
2064 | // Look for the CPU features. |
2065 | for (unsigned I = 0, E = Lines.size(); I != E; ++I) |
2066 | if (Lines[I].starts_with("Features" )) { |
2067 | Lines[I].split(CPUFeatures, ' '); |
2068 | break; |
2069 | } |
2070 | |
2071 | #if defined(__aarch64__) |
2072 | // All of these are "crypto" features, but we must sift out actual features |
2073 | // as the former meaning of "crypto" as a single feature is no more. |
2074 | enum { CAP_AES = 0x1, CAP_PMULL = 0x2, CAP_SHA1 = 0x4, CAP_SHA2 = 0x8 }; |
2075 | uint32_t crypto = 0; |
2076 | #endif |
2077 | |
2078 | for (unsigned I = 0, E = CPUFeatures.size(); I != E; ++I) { |
2079 | StringRef LLVMFeatureStr = StringSwitch<StringRef>(CPUFeatures[I]) |
2080 | #if defined(__aarch64__) |
2081 | .Case("asimd" , "neon" ) |
2082 | .Case("fp" , "fp-armv8" ) |
2083 | .Case("crc32" , "crc" ) |
2084 | .Case("atomics" , "lse" ) |
2085 | .Case("sve" , "sve" ) |
2086 | .Case("sve2" , "sve2" ) |
2087 | #else |
2088 | .Case("half" , "fp16" ) |
2089 | .Case("neon" , "neon" ) |
2090 | .Case("vfpv3" , "vfp3" ) |
2091 | .Case("vfpv3d16" , "vfp3d16" ) |
2092 | .Case("vfpv4" , "vfp4" ) |
2093 | .Case("idiva" , "hwdiv-arm" ) |
2094 | .Case("idivt" , "hwdiv" ) |
2095 | #endif |
2096 | .Default("" ); |
2097 | |
2098 | #if defined(__aarch64__) |
2099 | // We need to check crypto separately since we need all of the crypto |
2100 | // extensions to enable the subtarget feature |
2101 | if (CPUFeatures[I] == "aes" ) |
2102 | crypto |= CAP_AES; |
2103 | else if (CPUFeatures[I] == "pmull" ) |
2104 | crypto |= CAP_PMULL; |
2105 | else if (CPUFeatures[I] == "sha1" ) |
2106 | crypto |= CAP_SHA1; |
2107 | else if (CPUFeatures[I] == "sha2" ) |
2108 | crypto |= CAP_SHA2; |
2109 | #endif |
2110 | |
2111 | if (LLVMFeatureStr != "" ) |
2112 | Features[LLVMFeatureStr] = true; |
2113 | } |
2114 | |
2115 | #if defined(__aarch64__) |
2116 | // LLVM has decided some AArch64 CPUs have all the instructions they _may_ |
2117 | // have, as opposed to all the instructions they _must_ have, so allow runtime |
2118 | // information to correct us on that. |
2119 | uint32_t Aes = CAP_AES | CAP_PMULL; |
2120 | uint32_t Sha2 = CAP_SHA1 | CAP_SHA2; |
2121 | Features["aes" ] = (crypto & Aes) == Aes; |
2122 | Features["sha2" ] = (crypto & Sha2) == Sha2; |
2123 | #endif |
2124 | |
2125 | return Features; |
2126 | } |
2127 | #elif defined(_WIN32) && (defined(__aarch64__) || defined(_M_ARM64)) |
2128 | const StringMap<bool> sys::getHostCPUFeatures() { |
2129 | StringMap<bool> Features; |
2130 | |
2131 | // If we're asking the OS at runtime, believe what the OS says |
2132 | Features["neon" ] = |
2133 | IsProcessorFeaturePresent(PF_ARM_NEON_INSTRUCTIONS_AVAILABLE); |
2134 | Features["crc" ] = |
2135 | IsProcessorFeaturePresent(PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE); |
2136 | |
2137 | // Avoid inferring "crypto" means more than the traditional AES + SHA2 |
2138 | bool TradCrypto = |
2139 | IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE); |
2140 | Features["aes" ] = TradCrypto; |
2141 | Features["sha2" ] = TradCrypto; |
2142 | |
2143 | return Features; |
2144 | } |
2145 | #elif defined(__linux__) && defined(__loongarch__) |
2146 | #include <sys/auxv.h> |
2147 | const StringMap<bool> sys::getHostCPUFeatures() { |
2148 | unsigned long hwcap = getauxval(AT_HWCAP); |
2149 | bool HasFPU = hwcap & (1UL << 3); // HWCAP_LOONGARCH_FPU |
2150 | uint32_t cpucfg2 = 0x2, cpucfg3 = 0x3; |
2151 | __asm__("cpucfg %[cpucfg2], %[cpucfg2]\n\t" : [cpucfg2] "+r" (cpucfg2)); |
2152 | __asm__("cpucfg %[cpucfg3], %[cpucfg3]\n\t" : [cpucfg3] "+r" (cpucfg3)); |
2153 | |
2154 | StringMap<bool> Features; |
2155 | |
2156 | Features["f" ] = HasFPU && (cpucfg2 & (1U << 1)); // CPUCFG.2.FP_SP |
2157 | Features["d" ] = HasFPU && (cpucfg2 & (1U << 2)); // CPUCFG.2.FP_DP |
2158 | |
2159 | Features["lsx" ] = hwcap & (1UL << 4); // HWCAP_LOONGARCH_LSX |
2160 | Features["lasx" ] = hwcap & (1UL << 5); // HWCAP_LOONGARCH_LASX |
2161 | Features["lvz" ] = hwcap & (1UL << 9); // HWCAP_LOONGARCH_LVZ |
2162 | |
2163 | Features["frecipe" ] = cpucfg2 & (1U << 25); // CPUCFG.2.FRECIPE |
2164 | Features["div32" ] = cpucfg2 & (1U << 26); // CPUCFG.2.DIV32 |
2165 | Features["lam-bh" ] = cpucfg2 & (1U << 27); // CPUCFG.2.LAM_BH |
2166 | Features["lamcas" ] = cpucfg2 & (1U << 28); // CPUCFG.2.LAMCAS |
2167 | Features["scq" ] = cpucfg2 & (1U << 30); // CPUCFG.2.SCQ |
2168 | |
2169 | Features["ld-seq-sa" ] = cpucfg3 & (1U << 23); // CPUCFG.3.LD_SEQ_SA |
2170 | |
2171 | // TODO: Need to complete. |
2172 | // Features["llacq-screl"] = cpucfg2 & (1U << 29); // CPUCFG.2.LLACQ_SCREL |
2173 | return Features; |
2174 | } |
2175 | #elif defined(__linux__) && defined(__riscv) |
2176 | const StringMap<bool> sys::getHostCPUFeatures() { |
2177 | RISCVHwProbe Query[]{{/*RISCV_HWPROBE_KEY_BASE_BEHAVIOR=*/3, 0}, |
2178 | {/*RISCV_HWPROBE_KEY_IMA_EXT_0=*/4, 0}, |
2179 | {/*RISCV_HWPROBE_KEY_MISALIGNED_SCALAR_PERF=*/9, 0}}; |
2180 | int Ret = syscall(/*__NR_riscv_hwprobe=*/258, /*pairs=*/Query, |
2181 | /*pair_count=*/std::size(Query), /*cpu_count=*/0, |
2182 | /*cpus=*/0, /*flags=*/0); |
2183 | if (Ret != 0) |
2184 | return {}; |
2185 | |
2186 | StringMap<bool> Features; |
2187 | uint64_t BaseMask = Query[0].Value; |
2188 | // Check whether RISCV_HWPROBE_BASE_BEHAVIOR_IMA is set. |
2189 | if (BaseMask & 1) { |
2190 | Features["i" ] = true; |
2191 | Features["m" ] = true; |
2192 | Features["a" ] = true; |
2193 | } |
2194 | |
2195 | uint64_t ExtMask = Query[1].Value; |
2196 | Features["f" ] = ExtMask & (1 << 0); // RISCV_HWPROBE_IMA_FD |
2197 | Features["d" ] = ExtMask & (1 << 0); // RISCV_HWPROBE_IMA_FD |
2198 | Features["c" ] = ExtMask & (1 << 1); // RISCV_HWPROBE_IMA_C |
2199 | Features["v" ] = ExtMask & (1 << 2); // RISCV_HWPROBE_IMA_V |
2200 | Features["zba" ] = ExtMask & (1 << 3); // RISCV_HWPROBE_EXT_ZBA |
2201 | Features["zbb" ] = ExtMask & (1 << 4); // RISCV_HWPROBE_EXT_ZBB |
2202 | Features["zbs" ] = ExtMask & (1 << 5); // RISCV_HWPROBE_EXT_ZBS |
2203 | Features["zicboz" ] = ExtMask & (1 << 6); // RISCV_HWPROBE_EXT_ZICBOZ |
2204 | Features["zbc" ] = ExtMask & (1 << 7); // RISCV_HWPROBE_EXT_ZBC |
2205 | Features["zbkb" ] = ExtMask & (1 << 8); // RISCV_HWPROBE_EXT_ZBKB |
2206 | Features["zbkc" ] = ExtMask & (1 << 9); // RISCV_HWPROBE_EXT_ZBKC |
2207 | Features["zbkx" ] = ExtMask & (1 << 10); // RISCV_HWPROBE_EXT_ZBKX |
2208 | Features["zknd" ] = ExtMask & (1 << 11); // RISCV_HWPROBE_EXT_ZKND |
2209 | Features["zkne" ] = ExtMask & (1 << 12); // RISCV_HWPROBE_EXT_ZKNE |
2210 | Features["zknh" ] = ExtMask & (1 << 13); // RISCV_HWPROBE_EXT_ZKNH |
2211 | Features["zksed" ] = ExtMask & (1 << 14); // RISCV_HWPROBE_EXT_ZKSED |
2212 | Features["zksh" ] = ExtMask & (1 << 15); // RISCV_HWPROBE_EXT_ZKSH |
2213 | Features["zkt" ] = ExtMask & (1 << 16); // RISCV_HWPROBE_EXT_ZKT |
2214 | Features["zvbb" ] = ExtMask & (1 << 17); // RISCV_HWPROBE_EXT_ZVBB |
2215 | Features["zvbc" ] = ExtMask & (1 << 18); // RISCV_HWPROBE_EXT_ZVBC |
2216 | Features["zvkb" ] = ExtMask & (1 << 19); // RISCV_HWPROBE_EXT_ZVKB |
2217 | Features["zvkg" ] = ExtMask & (1 << 20); // RISCV_HWPROBE_EXT_ZVKG |
2218 | Features["zvkned" ] = ExtMask & (1 << 21); // RISCV_HWPROBE_EXT_ZVKNED |
2219 | Features["zvknha" ] = ExtMask & (1 << 22); // RISCV_HWPROBE_EXT_ZVKNHA |
2220 | Features["zvknhb" ] = ExtMask & (1 << 23); // RISCV_HWPROBE_EXT_ZVKNHB |
2221 | Features["zvksed" ] = ExtMask & (1 << 24); // RISCV_HWPROBE_EXT_ZVKSED |
2222 | Features["zvksh" ] = ExtMask & (1 << 25); // RISCV_HWPROBE_EXT_ZVKSH |
2223 | Features["zvkt" ] = ExtMask & (1 << 26); // RISCV_HWPROBE_EXT_ZVKT |
2224 | Features["zfh" ] = ExtMask & (1 << 27); // RISCV_HWPROBE_EXT_ZFH |
2225 | Features["zfhmin" ] = ExtMask & (1 << 28); // RISCV_HWPROBE_EXT_ZFHMIN |
2226 | Features["zihintntl" ] = ExtMask & (1 << 29); // RISCV_HWPROBE_EXT_ZIHINTNTL |
2227 | Features["zvfh" ] = ExtMask & (1 << 30); // RISCV_HWPROBE_EXT_ZVFH |
2228 | Features["zvfhmin" ] = ExtMask & (1ULL << 31); // RISCV_HWPROBE_EXT_ZVFHMIN |
2229 | Features["zfa" ] = ExtMask & (1ULL << 32); // RISCV_HWPROBE_EXT_ZFA |
2230 | Features["ztso" ] = ExtMask & (1ULL << 33); // RISCV_HWPROBE_EXT_ZTSO |
2231 | Features["zacas" ] = ExtMask & (1ULL << 34); // RISCV_HWPROBE_EXT_ZACAS |
2232 | Features["zicond" ] = ExtMask & (1ULL << 35); // RISCV_HWPROBE_EXT_ZICOND |
2233 | Features["zihintpause" ] = |
2234 | ExtMask & (1ULL << 36); // RISCV_HWPROBE_EXT_ZIHINTPAUSE |
2235 | Features["zve32x" ] = ExtMask & (1ULL << 37); // RISCV_HWPROBE_EXT_ZVE32X |
2236 | Features["zve32f" ] = ExtMask & (1ULL << 38); // RISCV_HWPROBE_EXT_ZVE32F |
2237 | Features["zve64x" ] = ExtMask & (1ULL << 39); // RISCV_HWPROBE_EXT_ZVE64X |
2238 | Features["zve64f" ] = ExtMask & (1ULL << 40); // RISCV_HWPROBE_EXT_ZVE64F |
2239 | Features["zve64d" ] = ExtMask & (1ULL << 41); // RISCV_HWPROBE_EXT_ZVE64D |
2240 | Features["zimop" ] = ExtMask & (1ULL << 42); // RISCV_HWPROBE_EXT_ZIMOP |
2241 | Features["zca" ] = ExtMask & (1ULL << 43); // RISCV_HWPROBE_EXT_ZCA |
2242 | Features["zcb" ] = ExtMask & (1ULL << 44); // RISCV_HWPROBE_EXT_ZCB |
2243 | Features["zcd" ] = ExtMask & (1ULL << 45); // RISCV_HWPROBE_EXT_ZCD |
2244 | Features["zcf" ] = ExtMask & (1ULL << 46); // RISCV_HWPROBE_EXT_ZCF |
2245 | Features["zcmop" ] = ExtMask & (1ULL << 47); // RISCV_HWPROBE_EXT_ZCMOP |
2246 | Features["zawrs" ] = ExtMask & (1ULL << 48); // RISCV_HWPROBE_EXT_ZAWRS |
2247 | |
2248 | // Check whether the processor supports fast misaligned scalar memory access. |
2249 | // NOTE: RISCV_HWPROBE_KEY_MISALIGNED_SCALAR_PERF is only available on |
2250 | // Linux 6.11 or later. If it is not recognized, the key field will be cleared |
2251 | // to -1. |
2252 | if (Query[2].Key != -1 && |
2253 | Query[2].Value == /*RISCV_HWPROBE_MISALIGNED_SCALAR_FAST=*/3) |
2254 | Features["unaligned-scalar-mem" ] = true; |
2255 | |
2256 | return Features; |
2257 | } |
2258 | #else |
2259 | const StringMap<bool> sys::getHostCPUFeatures() { return {}; } |
2260 | #endif |
2261 | |
2262 | #if __APPLE__ |
2263 | /// \returns the \p triple, but with the Host's arch spliced in. |
2264 | static Triple withHostArch(Triple T) { |
2265 | #if defined(__arm__) |
2266 | T.setArch(Triple::arm); |
2267 | T.setArchName("arm" ); |
2268 | #elif defined(__arm64e__) |
2269 | T.setArch(Triple::aarch64, Triple::AArch64SubArch_arm64e); |
2270 | T.setArchName("arm64e" ); |
2271 | #elif defined(__aarch64__) |
2272 | T.setArch(Triple::aarch64); |
2273 | T.setArchName("arm64" ); |
2274 | #elif defined(__x86_64h__) |
2275 | T.setArch(Triple::x86_64); |
2276 | T.setArchName("x86_64h" ); |
2277 | #elif defined(__x86_64__) |
2278 | T.setArch(Triple::x86_64); |
2279 | T.setArchName("x86_64" ); |
2280 | #elif defined(__i386__) |
2281 | T.setArch(Triple::x86); |
2282 | T.setArchName("i386" ); |
2283 | #elif defined(__powerpc__) |
2284 | T.setArch(Triple::ppc); |
2285 | T.setArchName("powerpc" ); |
2286 | #else |
2287 | # error "Unimplemented host arch fixup" |
2288 | #endif |
2289 | return T; |
2290 | } |
2291 | #endif |
2292 | |
2293 | std::string sys::getProcessTriple() { |
2294 | std::string TargetTripleString = updateTripleOSVersion(LLVM_HOST_TRIPLE); |
2295 | Triple PT(Triple::normalize(Str: TargetTripleString)); |
2296 | |
2297 | #if __APPLE__ |
2298 | /// In Universal builds, LLVM_HOST_TRIPLE will have the wrong arch in one of |
2299 | /// the slices. This fixes that up. |
2300 | PT = withHostArch(PT); |
2301 | #endif |
2302 | |
2303 | if (sizeof(void *) == 8 && PT.isArch32Bit()) |
2304 | PT = PT.get64BitArchVariant(); |
2305 | if (sizeof(void *) == 4 && PT.isArch64Bit()) |
2306 | PT = PT.get32BitArchVariant(); |
2307 | |
2308 | return PT.str(); |
2309 | } |
2310 | |
2311 | void sys::printDefaultTargetAndDetectedCPU(raw_ostream &OS) { |
2312 | #if LLVM_VERSION_PRINTER_SHOW_HOST_TARGET_INFO |
2313 | std::string CPU = std::string(sys::getHostCPUName()); |
2314 | if (CPU == "generic" ) |
2315 | CPU = "(unknown)" ; |
2316 | OS << " Default target: " << sys::getDefaultTargetTriple() << '\n' |
2317 | << " Host CPU: " << CPU << '\n'; |
2318 | #endif |
2319 | } |
2320 | |