xpcom/glue/SSE.h

   1 /* vim: set shiftwidth=2 tabstop=8 autoindent cindent expandtab: */
   2 /* ***** BEGIN LICENSE BLOCK *****
   3  * Version: MPL 1.1/GPL 2.0/LGPL 2.1
   4  *
   5  * The contents of this file are subject to the Mozilla Public License Version
   6  * 1.1 (the "License"); you may not use this file except in compliance with
   7  * the License. You may obtain a copy of the License at
   8  * http://www.mozilla.org/MPL/
   9  *
  10  * Software distributed under the License is distributed on an "AS IS" basis,
  11  * WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License
  12  * for the specific language governing rights and limitations under the
  13  * License.
  14  *
  15  * The Original Code is SSE.h
  16  *
  17  * The Initial Developer of the Original Code is the Mozilla Foundation.
  18  * Portions created by the Initial Developer are Copyright (C) 2009
  19  * the Initial Developer. All Rights Reserved.
  20  *
  21  * Contributor(s):
  22  *   L. David Baron <dbaron@dbaron.org>, Mozilla Corporation (original author)
  23  *   Justin Lebar <justin.lebar@gmail.com>, Mozilla Corporation
  24  *
  25  * Alternatively, the contents of this file may be used under the terms of
  26  * either the GNU General Public License Version 2 or later (the "GPL"), or
  27  * the GNU Lesser General Public License Version 2.1 or later (the "LGPL"),
  28  * in which case the provisions of the GPL or the LGPL are applicable instead
  29  * of those above. If you wish to allow use of your version of this file only
  30  * under the terms of either the GPL or the LGPL, and not to allow others to
  31  * use your version of this file under the terms of the MPL, indicate your
  32  * decision by deleting the provisions above and replace them with the notice
  33  * and other provisions required by the GPL or the LGPL. If you do not delete
  34  * the provisions above, a recipient may use your version of this file under
  35  * the terms of any one of the MPL, the GPL or the LGPL.
  36  *
  37  * ***** END LICENSE BLOCK ***** */
  38
  39 /* compile-time and runtime tests for whether to use SSE instructions */
  40
  41 #ifndef mozilla_SSE_h_
  42 #define mozilla_SSE_h_
  43
  44 // for definition of NS_COM_GLUE
  45 #include "nscore.h"
  46
  47 /**
  48  * The public interface of this header consists of a set of macros and
  49  * functions for Intel CPU features.
  50  *
  51  * DETECTING ISA EXTENSIONS
  52  * ========================
  53  *
  54  * This header provides the following functions for determining whether the
  55  * current CPU supports a particular instruction set extension:
  56  *
  57  *    mozilla::supports_mmx
  58  *    mozilla::supports_sse
  59  *    mozilla::supports_sse2
  60  *    mozilla::supports_sse3
  61  *    mozilla::supports_ssse3
  62  *    mozilla::supports_sse4a
  63  *    mozilla::supports_sse4_1
  64  *    mozilla::supports_sse4_2
  65  *
  66  * If you're writing code using inline assembly, you should guard it with a
  67  * call to one of these functions.  For instance:
  68  *
  69  *   if (mozilla::supports_sse2()) {
  70  *     asm(" ... ");
  71  *   }
  72  *   else {
  73  *     ...
  74  *   }
  75  *
  76  * Note that these functions depend on cpuid intrinsics only available in gcc
  77  * 4.3 or later and MSVC 8.0 (Visual C++ 2005) or later, so they return false
  78  * in older compilers.  (This could be fixed by replacing the code with inline
  79  * assembly.)
  80  *
  81  *
  82  * USING INTRINSICS
  83  * ================
  84  *
  85  * This header also provides support for coding using CPU intrinsics.
  86  *
  87  * For each mozilla::supports_abc function, we define a MOZILLA_MAY_SUPPORT_ABC
  88  * macro which indicates that the target/compiler combination we're using is
  89  * compatible with the ABC extension.  For instance, x86_64 with MSVC 2003 is
  90  * compatible with SSE2 but not SSE3, since although there exist x86_64 CPUs
  91  * with SSE3 support, MSVC 2003 only supports through SSE2.
  92  *
  93  * Until gcc fixes #pragma target [1] [2] or our x86 builds require SSE2,
  94  * you'll need to separate code using intrinsics into a file separate from your
  95  * regular code.  Here's the recommended pattern:
  96  *
  97  *  #ifdef MOZILLA_MAY_SUPPORT_ABC
  98  *    namespace mozilla {
  99  *      namespace ABC {
 100  *        void foo();
 101  *      }
 102  *    }
 103  *  #endif
 104  *
 105  *  void foo() {
 106  *    #ifdef MOZILLA_MAY_SUPPORT_ABC
 107  *      if (mozilla::supports_abc()) {
 108  *        mozilla::ABC::foo(); // in a separate file
 109  *        return;
 110  *      }
 111  *    #endif
 112  *
 113  *    foo_unvectorized();
 114  *  }
 115  *
 116  * You'll need to define mozilla::ABC::foo() in a separate file and add the
 117  * -mabc flag when using gcc.
 118  *
 119  * [1] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39787 and
 120  * [2] http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41201 being fixed.
 121  *
 122  */
 123
 124 #if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
 125
 126 #ifdef __MMX__
 127   // It's ok to use MMX instructions based on the -march option (or
 128   // the default for x86_64 or for Intel Mac).
 129   #define MOZILLA_PRESUME_MMX 1
 130 #endif
 131 #ifdef __SSE__
 132   // It's ok to use SSE instructions based on the -march option (or
 133   // the default for x86_64 or for Intel Mac).
 134   #define MOZILLA_PRESUME_SSE 1
 135 #endif
 136 #ifdef __SSE2__
 137   // It's ok to use SSE2 instructions based on the -march option (or
 138   // the default for x86_64 or for Intel Mac).
 139   #define MOZILLA_PRESUME_SSE2 1
 140 #endif
 141 #ifdef __SSE3__
 142   // It's ok to use SSE3 instructions based on the -march option (or the
 143   // default for Intel Mac).
 144   #define MOZILLA_PRESUME_SSE3 1
 145 #endif
 146 #ifdef __SSSE3__
 147   // It's ok to use SSSE3 instructions based on the -march option.
 148   #define MOZILLA_PRESUME_SSSE3 1
 149 #endif
 150 #ifdef __SSE4A__
 151   // It's ok to use SSE4A instructions based on the -march option.
 152   #define MOZILLA_PRESUME_SSE4A 1
 153 #endif
 154 #ifdef __SSE4_1__
 155   // It's ok to use SSE4.1 instructions based on the -march option.
 156   #define MOZILLA_PRESUME_SSE4_1 1
 157 #endif
 158 #ifdef __SSE4_2__
 159   // It's ok to use SSE4.2 instructions based on the -march option.
 160   #define MOZILLA_PRESUME_SSE4_2 1
 161 #endif
 162
 163 #if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)
 164   #define MOZILLA_SSE_HAVE_CPUID_DETECTION
 165 #endif
 166
 167 #elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_AMD64))
 168
 169 #if _MSC_VER >= 1400
 170   // MSVC 2005 or newer on x86 or amd64
 171   #define MOZILLA_SSE_HAVE_CPUID_DETECTION
 172 #endif
 173
 174 #if defined(_M_AMD64)
 175   // MSVC for AMD64 doesn't support MMX, so don't presume it here.
 176
 177   // SSE is always available on AMD64.
 178   #define MOZILLA_PRESUME_SSE
 179   // SSE2 is always available on AMD64.
 180   #define MOZILLA_PRESUME_SSE2
 181 #endif
 182
 183 #elif defined(__SUNPRO_CC) && (defined(__i386) || defined(__x86_64__))
 184 // Sun Studio on x86 or amd64
 185
 186 #define MOZILLA_SSE_HAVE_CPUID_DETECTION
 187
 188 #if defined(__x86_64__)
 189   // MMX is always available on AMD64.
 190   #define MOZILLA_PRESUME_MMX
 191   // SSE is always available on AMD64.
 192   #define MOZILLA_PRESUME_SSE
 193   // SSE2 is always available on AMD64.
 194   #define MOZILLA_PRESUME_SSE2
 195 #endif
 196
 197 #endif
 198
 199 namespace mozilla {
 200
 201   namespace sse_private {
 202 #if defined(MOZILLA_SSE_HAVE_CPUID_DETECTION)
 203 #if !defined(MOZILLA_PRESUME_MMX)
 204     extern bool NS_COM_GLUE mmx_enabled;
 205 #endif
 206 #if !defined(MOZILLA_PRESUME_SSE)
 207     extern bool NS_COM_GLUE sse_enabled;
 208 #endif
 209 #if !defined(MOZILLA_PRESUME_SSE2)
 210     extern bool NS_COM_GLUE sse2_enabled;
 211 #endif
 212 #if !defined(MOZILLA_PRESUME_SSE3)
 213     extern bool NS_COM_GLUE sse3_enabled;
 214 #endif
 215 #if !defined(MOZILLA_PRESUME_SSSE3)
 216     extern bool NS_COM_GLUE ssse3_enabled;
 217 #endif
 218 #if !defined(MOZILLA_PRESUME_SSE4A)
 219     extern bool NS_COM_GLUE sse4a_enabled;
 220 #endif
 221 #if !defined(MOZILLA_PRESUME_SSE4_1)
 222     extern bool NS_COM_GLUE sse4_1_enabled;
 223 #endif
 224 #if !defined(MOZILLA_PRESUME_SSE4_2)
 225     extern bool NS_COM_GLUE sse4_2_enabled;
 226 #endif
 227 #endif
 228   }
 229
 230 #if defined(MOZILLA_PRESUME_MMX)
 231 #define MOZILLA_MAY_SUPPORT_MMX 1
 232   inline bool supports_mmx() { return true; }
 233 #elif defined(MOZILLA_SSE_HAVE_CPUID_DETECTION)
 234 #if !(defined(_MSC_VER) && defined(_M_AMD64))
 235   // Define MOZILLA_MAY_SUPPORT_MMX only if we're not on MSVC for
 236   // AMD64, since that compiler doesn't support MMX.
 237 #define MOZILLA_MAY_SUPPORT_MMX 1
 238 #endif
 239   inline bool supports_mmx() { return sse_private::mmx_enabled; }
 240 #else
 241   inline bool supports_mmx() { return false; }
 242 #endif
 243
 244 #if defined(MOZILLA_PRESUME_SSE)
 245 #define MOZILLA_MAY_SUPPORT_SSE 1
 246   inline bool supports_sse() { return true; }
 247 #elif defined(MOZILLA_SSE_HAVE_CPUID_DETECTION)
 248 #define MOZILLA_MAY_SUPPORT_SSE 1
 249   inline bool supports_sse() { return sse_private::sse_enabled; }
 250 #else
 251   inline bool supports_sse() { return false; }
 252 #endif
 253
 254 #if defined(MOZILLA_PRESUME_SSE2)
 255 #define MOZILLA_MAY_SUPPORT_SSE2 1
 256   inline bool supports_sse2() { return true; }
 257 #elif defined(MOZILLA_SSE_HAVE_CPUID_DETECTION)
 258 #define MOZILLA_MAY_SUPPORT_SSE2 1
 259   inline bool supports_sse2() { return sse_private::sse2_enabled; }
 260 #else
 261   inline bool supports_sse2() { return false; }
 262 #endif
 263
 264 #if defined(MOZILLA_PRESUME_SSE3)
 265 #define MOZILLA_MAY_SUPPORT_SSE3 1
 266   inline bool supports_sse3() { return true; }
 267 #elif defined(MOZILLA_SSE_HAVE_CPUID_DETECTION)
 268 #define MOZILLA_MAY_SUPPORT_SSE3 1
 269   inline bool supports_sse3() { return sse_private::sse3_enabled; }
 270 #else
 271   inline bool supports_sse3() { return false; }
 272 #endif
 273
 274 #if defined(MOZILLA_PRESUME_SSSE3)
 275 #define MOZILLA_MAY_SUPPORT_SSSE3 1
 276   inline bool supports_ssse3() { return true; }
 277 #elif defined(MOZILLA_SSE_HAVE_CPUID_DETECTION)
 278 #define MOZILLA_MAY_SUPPORT_SSSE3 1
 279   inline bool supports_ssse3() { return sse_private::ssse3_enabled; }
 280 #else
 281   inline bool supports_ssse3() { return false; }
 282 #endif
 283
 284 #if defined(MOZILLA_PRESUME_SSE4A)
 285 #define MOZILLA_MAY_SUPPORT_SSE4A 1
 286   inline bool supports_sse4a() { return true; }
 287 #elif defined(MOZILLA_SSE_HAVE_CPUID_DETECTION)
 288 #define MOZILLA_MAY_SUPPORT_SSE4A 1
 289   inline bool supports_sse4a() { return sse_private::sse4a_enabled; }
 290 #else
 291   inline bool supports_sse4a() { return false; }
 292 #endif
 293
 294 #if defined(MOZILLA_PRESUME_SSE4_1)
 295 #define MOZILLA_MAY_SUPPORT_SSE4_1 1
 296   inline bool supports_sse4_1() { return true; }
 297 #elif defined(MOZILLA_SSE_HAVE_CPUID_DETECTION)
 298 #define MOZILLA_MAY_SUPPORT_SSE4_1 1
 299   inline bool supports_sse4_1() { return sse_private::sse4_1_enabled; }
 300 #else
 301   inline bool supports_sse4_1() { return false; }
 302 #endif
 303
 304 #if defined(MOZILLA_PRESUME_SSE4_2)
 305 #define MOZILLA_MAY_SUPPORT_SSE4_2 1
 306   inline bool supports_sse4_2() { return true; }
 307 #elif defined(MOZILLA_SSE_HAVE_CPUID_DETECTION)
 308 #define MOZILLA_MAY_SUPPORT_SSE4_2 1
 309   inline bool supports_sse4_2() { return sse_private::sse4_2_enabled; }
 310 #else
 311   inline bool supports_sse4_2() { return false; }
 312 #endif
 313
 314 }
 315
 316 #endif /* !defined(mozilla_SSE_h_) */