src/cpu.cpp

   1 /*
   2  * This file is part of OpenTTD.
   3  * OpenTTD is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 2.
   4  * OpenTTD is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
   5  * See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with OpenTTD. If not, see <http://www.gnu.org/licenses/>.
   6  */
   7
   8 /** @file cpu.cpp OS/CPU/compiler dependent CPU specific calls. */
   9
  10 #include "stdafx.h"
  11 #include "core/bitmath_func.hpp"
  12
  13 #include "safeguards.h"
  14
  15 #undef RDTSC_AVAILABLE
  16
  17 /* rdtsc for MSC_VER, uses simple inline assembly, or _rdtsc
  18  * from external win64.asm because VS2005 does not support inline assembly */
  19 #if defined(_MSC_VER) && !defined(RDTSC_AVAILABLE)
  20 #include <intrin.h>
  21 #include <windows.h>
  22 uint64 ottd_rdtsc()
  23 {
  24 #if defined(_M_ARM)
  25         return __rdpmccntr64();
  26 #elif defined(_M_ARM64)
  27         return _ReadStatusReg(ARM64_PMCCNTR_EL0);
  28 #else
  29         return __rdtsc();
  30 #endif
  31 }
  32 #define RDTSC_AVAILABLE
  33 #endif
  34
  35 /* rdtsc for OS/2. Hopefully this works, who knows */
  36 #if defined (__WATCOMC__) && !defined(RDTSC_AVAILABLE)
  37 unsigned __int64 ottd_rdtsc();
  38 # pragma aux ottd_rdtsc = 0x0F 0x31 value [edx eax] parm nomemory modify exact [edx eax] nomemory;
  39 # define RDTSC_AVAILABLE
  40 #endif
  41
  42 /* rdtsc for all other *nix-en (hopefully). Use GCC syntax */
  43 #if (defined(__i386__) || defined(__x86_64__)) && !defined(RDTSC_AVAILABLE)
  44 uint64 ottd_rdtsc()
  45 {
  46         uint32 high, low;
  47         __asm__ __volatile__ ("rdtsc" : "=a" (low), "=d" (high));
  48         return ((uint64)high << 32) | low;
  49 }
  50 # define RDTSC_AVAILABLE
  51 #endif
  52
  53 /* rdtsc for PPC which has this not */
  54 #if (defined(__POWERPC__) || defined(__powerpc__)) && !defined(RDTSC_AVAILABLE)
  55 uint64 ottd_rdtsc()
  56 {
  57         uint32 high = 0, high2 = 0, low;
  58         /* PPC does not have rdtsc, so we cheat by reading the two 32-bit time-counters
  59          * it has, 'Move From Time Base (Upper)'. Since these are two reads, in the
  60          * very unlikely event that the lower part overflows to the upper part while we
  61          * read it; we double-check and reread the registers */
  62         asm volatile (
  63                                   "mftbu %0\n"
  64                                   "mftb %1\n"
  65                                   "mftbu %2\n"
  66                                   "cmpw %3,%4\n"
  67                                   "bne- $-16\n"
  68                                   : "=r" (high), "=r" (low), "=r" (high2)
  69                                   : "0" (high), "2" (high2)
  70                                   );
  71         return ((uint64)high << 32) | low;
  72 }
  73 # define RDTSC_AVAILABLE
  74 #endif
  75
  76 #if defined(__EMSCRIPTEN__) && !defined(RDTSC_AVAILABLE)
  77 /* On emscripten doing TIC/TOC would be ill-advised */
  78 uint64 ottd_rdtsc() {return 0;}
  79 # define RDTSC_AVAILABLE
  80 #endif
  81
  82 /* In all other cases we have no support for rdtsc. No major issue,
  83  * you just won't be able to profile your code with TIC()/TOC() */
  84 #if !defined(RDTSC_AVAILABLE)
  85 #warning "(non-fatal) No support for rdtsc(), you won't be able to profile with TIC/TOC"
  86 uint64 ottd_rdtsc() {return 0;}
  87 #endif
  88
  89
  90 /**
  91  * Definitions for CPU detection:
  92  *
  93  * MSVC offers cpu information while gcc only implements in gcc 4.8
  94  * __builtin_cpu_supports and friends
  95  * http://msdn.microsoft.com/library/vstudio/hskdteyh%28v=vs.100%29.aspx
  96  * http://gcc.gnu.org/onlinedocs/gcc/X86-Built-in-Functions.html
  97  *
  98  * Other platforms/architectures don't have CPUID, so zero the info and then
  99  * most (if not all) of the features are set as if they do not exist.
 100  */
 101 #if defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_X64))
 102 void ottd_cpuid(int info[4], int type)
 103 {
 104         __cpuid(info, type);
 105 }
 106 #elif defined(__x86_64__) || defined(__i386)
 107 void ottd_cpuid(int info[4], int type)
 108 {
 109 #if defined(__i386) && defined(__PIC__)
 110         /* The easy variant would be just cpuid, however... ebx is being used by the GOT (Global Offset Table)
 111          * in case of PIC;
 112          * clobbering ebx is no alternative: some compiler versions don't like this
 113          * and will issue an error message like
 114          *   "can't find a register in class 'BREG' while reloading 'asm'"
 115          */
 116         __asm__ __volatile__ (
 117                         "xchgl %%ebx, %1 \n\t"
 118                         "cpuid           \n\t"
 119                         "xchgl %%ebx, %1 \n\t"
 120                         : "=a" (info[0]), "=r" (info[1]), "=c" (info[2]), "=d" (info[3])
 121                         /* It is safe to write "=r" for (info[1]) as in case that PIC is enabled for i386,
 122                          * the compiler will not choose EBX as target register (but something else).
 123                          */
 124                         : "a" (type)
 125         );
 126 #else
 127         __asm__ __volatile__ (
 128                         "cpuid           \n\t"
 129                         : "=a" (info[0]), "=b" (info[1]), "=c" (info[2]), "=d" (info[3])
 130                         : "a" (type)
 131         );
 132 #endif /* i386 PIC */
 133 }
 134 #else
 135 void ottd_cpuid(int info[4], int type)
 136 {
 137         info[0] = info[1] = info[2] = info[3] = 0;
 138 }
 139 #endif
 140
 141 bool HasCPUIDFlag(uint type, uint index, uint bit)
 142 {
 143         int cpu_info[4] = {-1};
 144         ottd_cpuid(cpu_info, 0);
 145         uint max_info_type = cpu_info[0];
 146         if (max_info_type < type) return false;
 147
 148         ottd_cpuid(cpu_info, type);
 149         return HasBit(cpu_info[index], bit);
 150 }