AVX2 core for OGR-NG

common/cliident.cpp

@@ -542,7 +542,7 @@ int CliIsDevelVersion(void)
 
 const char *CliGetFullVersionDescriptor(void)
 {
-  static char buffer[10+32+sizeof("v"CLIENT_VERSIONSTRING"-XXX-99071523-*dev* client for "CLIENT_OS_NAME_EXTENDED)];
+  static char buffer[10+32+sizeof("v" CLIENT_VERSIONSTRING "-XXX-99071523-*dev* client for " CLIENT_OS_NAME_EXTENDED)];
   struct timeval tv; tv.tv_usec = 0;
   tv.tv_sec = CliGetNewestModuleTime();
   sprintf( buffer, "%s v" CLIENT_VERSIONSTRING "-"
@@ -559,7 +559,7 @@ const char *CliGetFullVersionDescriptor(void)
          "%c"  /* limited release or dev branch or public release */
          "-%s" /* date is in bugzilla format yymmddhh */
          "%s"  /* "-*dev*" or "" */
-         " for "CLIENT_OS_NAME_EXTENDED,
+         " for " CLIENT_OS_NAME_EXTENDED,
          utilGetAppName(),
          ((ConIsGUI())?('G'):('C')),
          ((CliIsDevelVersion())?('L'):('R')),

common/confopt.cpp

@@ -405,7 +405,7 @@ struct optionstruct conf_options[CONF_OPTION_COUNT] = {
   "\n"
   "It is possible to have the client rotate through this list, updating its\n"
   "buffers only once for each pass. To do so, 'Dialup-link detection'\n"
-  "and '"ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME"' must be disabled since a buffer\n"
+  "and '" ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME "' must be disabled since a buffer\n"
   "update (new work being made available) would otherwise cause the client\n"
   "to go back to the beginning of the load order.\n"
   /*) */,CONF_MENU_BUFF,CONF_TYPE_ASCIIZ,NULL,NULL,0,0,NULL,NULL
@@ -439,13 +439,13 @@ struct optionstruct conf_options[CONF_OPTION_COUNT] = {
   CONF_FREQUENT_FREQUENCY      , /* CONF_MENU_BUFF */
   CFGTXT("Buffer-level check interval"), "0:00 (on buffer change)",
   /*CFGTXT(*/
-  "This option determines how often '"ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME"'\n"
+  "This option determines how often '" ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME "'\n"
   "should be performed. (More precisely: how much time must elapse between\n"
   "buffer-level checks)\n" 
   "\n"
   "This setting is meaningful only if one of the extensions to normal threshold\n"
   "management is enabled: either implicitly when 'Dialup detection options' are\n"
-  "active or explicitly with '"ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME"'.\n"
+  "active or explicitly with '" ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME "'.\n"
   "\n"
   "The interval specified here is in hours and minutes, and the default denotes\n"
   "that the client should check buffer-levels whenever it detects a change (by\n"
@@ -456,13 +456,13 @@ struct optionstruct conf_options[CONF_OPTION_COUNT] = {
   CONF_FREQUENT_RETRY_FREQUENCY      , /* CONF_MENU_BUFF */
   CFGTXT("Buffer-level check retry interval"), "0:00 (no delay)",
   /*CFGTXT(*/
-  "This option determines how often '"ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME"'\n"
+  "This option determines how often '" ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME "'\n"
   "should be retried after failure. (More precisely: how much time must elapse\n"
   "between buffer-level check retries)\n" 
   "\n"
   "This setting is meaningful only if one of the extensions to normal threshold\n"
   "management is enabled: either implicitly when 'Dialup detection options' are\n"
-  "active or explicitly with '"ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME"'.\n"
+  "active or explicitly with '" ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME "'.\n"
   "\n"
   "The interval specified here is in hours and minutes, and the default denotes\n"
   "that the client should retry the buffer-level checks at most twice per minute\n"
@@ -503,7 +503,7 @@ struct optionstruct conf_options[CONF_OPTION_COUNT] = {
   "should be used instead. If that too is unspecified, then the client will\n"
   "use defaults.\n"
   "\n"
-  "* See also: '"ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME"'\n"
+  "* See also: '" ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME "'\n"
   ,CONF_MENU_BUFF,CONF_TYPE_IARRAY,NULL,NULL,1,0xffff,NULL,NULL
 },
 { 
@@ -523,7 +523,7 @@ struct optionstruct conf_options[CONF_OPTION_COUNT] = {
   "unprocessed packet cannot be predicted.\n"
 #endif
   "\n"
-  "* See also: '"ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME"'\n"
+  "* See also: '" ADDITIONAL_BUFFLEVEL_CHECK_OPTION_NAME "'\n"
   ,CONF_MENU_BUFF,CONF_TYPE_IARRAY,NULL,NULL,0,(14*24),NULL,NULL
 },
 

common/core_ogr_ng.cpp

@@ -84,6 +84,7 @@ return "@(#)$Id: core_ogr_ng.cpp,v 1.47 2015/06/27 21:43:52 zebe Exp $"; }
     CoreDispatchTable *ogrng64_get_dispatch_table_cj1_generic(void);
     CoreDispatchTable *ogrng64_get_dispatch_table_cj1_sse2(void);
     CoreDispatchTable *ogrng64_get_dispatch_table_cj1_sse2_lzcnt(void);
+    CoreDispatchTable *ogrng64_get_dispatch_table_cj1_avx2(void);
 #elif (CLIENT_CPU == CPU_SPARC) && (SIZEOF_LONG == 8)
     CoreDispatchTable *ogrng64_get_dispatch_table(void); 
 #elif (CLIENT_CPU == CPU_S390X) && (SIZEOF_LONG == 8)
@@ -166,6 +167,7 @@ int InitializeCoreTable_ogr_ng(int first_time)
         ogrng64_get_dispatch_table_cj1_generic();
         ogrng64_get_dispatch_table_cj1_sse2();
         ogrng64_get_dispatch_table_cj1_sse2_lzcnt();
+        ogrng64_get_dispatch_table_cj1_avx2();
       #elif (CLIENT_CPU == CPU_S390)
         ogrng_get_dispatch_table();
       #elif (CLIENT_CPU == CPU_S390X)
@@ -225,6 +227,7 @@ const char **corenames_for_contest_ogr_ng()
       "cj-asm-generic",
       "cj-asm-sse2",
       "cj-asm-sse2-lzcnt",
+      "cj-asm-avx2",
   #elif (CLIENT_CPU == CPU_ARM)
       "FLEGE 2.0",
       "FLEGE 2.0 ARMv3",
@@ -356,6 +359,8 @@ int apply_selcore_substitution_rules_ogr_ng(int cindex)
 #  endif
 # elif (CLIENT_CPU == CPU_AMD64)
   unsigned feature = GetProcessorFeatureFlags();
+  if (cindex == 4 && !(feature & CPU_F_AVX2)) /* Core 4 needs AVX2 */
+    cindex = 2; /* If no AVX2, try SSE2 */
   if (cindex == 3 && !(feature & CPU_F_LZCNT)) /* Core 3 needs LZCNT */
     cindex = 2; /* If no LZCNT, try SSE2 */
   if (cindex == 2 && !(feature & CPU_F_SSE2))  /* Core 2 needs SSE2 */
@@ -504,8 +509,11 @@ int selcoreGetPreselectedCoreForProject_ogr_ng()
       }
       if (cindex == -1)
       {
+        /* Assume that if AVX2 is availble it is the best choice */
+        if (detected_flags & CPU_F_AVX2)
+          cindex = 4;
         /* Assume that LZCNT+SSE2 is better then plain SSE2 everywhere */
-        if (detected_flags & CPU_F_LZCNT)
+        else if (detected_flags & CPU_F_LZCNT)
           cindex = 3;
         else if (detected_flags & CPU_F_SSE2)
           cindex = 2;  /* sse2 core */
@@ -638,6 +646,8 @@ int selcoreSelectCore_ogr_ng(Client *client, unsigned int threadindex,
     unit_func.ogr = ogrng64_get_dispatch_table_cj1_sse2();
   else if (coresel == 3)
     unit_func.ogr = ogrng64_get_dispatch_table_cj1_sse2_lzcnt();
+  else if (coresel == 4)
+    unit_func.ogr = ogrng64_get_dispatch_table_cj1_avx2();
   else
   {
     unit_func.ogr = ogrng64_get_dispatch_table();

common/cpucheck.cpp

@@ -1344,7 +1344,8 @@ long __GetRawProcessorID(const char **cpuname, int whattoret = 0 )
          *   0x19 - Sandy Bridge Core iX-2xxx
          *   0x1A - Ivy Bridge Core iX-3xxx
          *   0x1B - Haswell Core iX-4xxx
-         *   0x1C-1F
+         *   0x1C - Kaby Lake Core iX-7xxx
+         *   0x1D-1F
          *   0x20 - AMD Bobcat - Embedded APU
          *   0x21 - AMD Bulldozer - FX
          *   0x22 - AMD Husky - APU
@@ -1488,6 +1489,7 @@ long __GetRawProcessorID(const char **cpuname, int whattoret = 0 )
         { 0x0006450, 0xFFFFFF0, CPU_F_I686, 0x1B, "Core iX-4xxx (Haswell)" },  /* (#4579) */
         { 0x0006460, 0xFFFFFF0, CPU_F_I686, 0x1B, "Core iX-4xxx (Haswell)" },
         { 0x00065E0, 0xFFFFFF0, CPU_F_I686, 0x1B, "Core iX-6xxx (Skylake)" },  /* (#4615) */
+        { 0x00069E0, 0xFFFFFF0, CPU_F_I686, 0x1C, "Core iX-7xxx (Kaby Lake)" },
         { 0x0000000,         0,          0,    0, NULL }
       }; internalxref = &intelxref[0];
     }

common/cpucheck.h

@@ -24,13 +24,13 @@
   #define CPU_F_SSE             (0x00002000L)
   #define CPU_F_SSE2            (0x00004000L)
   #define CPU_F_SSE3            (0x00008000L)
-  #define CPU_F_HYPERTHREAD     (0x00010000L)	/* supported and enabled */
+  #define CPU_F_HYPERTHREAD     (0x00010000L)   /* supported and enabled */
   #define CPU_F_AMD64           (0x00020000L)
   #define CPU_F_EM64T           (0x00040000L)
   #define CPU_F_SSE4_1          (0x00080000L)
   #define CPU_F_SSE4_2          (0x00100000L)
   #define CPU_F_SSSE3           (0x00200000L)
-  #define CPU_F_LZCNT		(0x00400000L)
+  #define CPU_F_LZCNT           (0x00400000L)
   #define CPU_F_AVX_DISABLED    (0x00800000L)   /* supported but disabled (no OS support) */
   #define CPU_F_AVX             (0x01000000L)   /* supported and enabled */
   #define CPU_F_AVX2            (0x02000000L)   /* supported and enabled */

common/mail.cpp

@@ -564,8 +564,8 @@ static int smtp_send_message_header( void * net,
   if (errcode == 0) //send the date
   {
     sprintf( buffer, "\r\nDate: %s"
-        "\r\nX-Mailer: distributed.net v"CLIENT_VERSIONSTRING
-           " client for "CLIENT_OS_NAME_EXTENDED, rfc822Date( buffer + 256 ) );
+        "\r\nX-Mailer: distributed.net v" CLIENT_VERSIONSTRING
+           " client for " CLIENT_OS_NAME_EXTENDED, rfc822Date( buffer + 256 ) );
     if ( put_smtp_line( net, buffer, strlen( buffer ) ) )
       errcode = -1;
   }

common/problem.h

@@ -47,10 +47,10 @@
        #define CORE_MEM_ALIGNMENT 4
      #endif
   #else
-     // For x86, alignment must be 8 for MMX core and 16 for SSE.
-     #if CORE_MEM_ALIGNMENT < 4
+     // For x86, alignment must be 8 for MMX core, 16 for SSE and 32 for AVX2.
+     #if CORE_MEM_ALIGNMENT < 5
        #undef CORE_MEM_ALIGNMENT
-       #define CORE_MEM_ALIGNMENT 4
+       #define CORE_MEM_ALIGNMENT 5
      #endif
   #endif
 #endif

common/util.cpp

@@ -73,7 +73,7 @@ void trace_setsrc( const char *filename )
 void trace_out( int indlevel, const char *format, ... )
 {
   static int indentlevel = -1; /* uninitialized */
-  const char *tracefile = "trace"EXTN_SEP"out";
+  const char *tracefile = "trace" EXTN_SEP "out";
   int old_errno = errno;
   FILE *file;
   va_list arglist;

configure

@@ -659,9 +659,11 @@ add_sources() # $1=os, $2=arch, $3=custom
       TARGET_ADDSRCS="$TARGET_ADDSRCS $OGR/amd64/ogrng64-cj1-generic.cpp"
       TARGET_ADDSRCS="$TARGET_ADDSRCS $OGR/amd64/ogrng64-cj1-sse2.cpp"
       TARGET_ADDSRCS="$TARGET_ADDSRCS $OGR/amd64/ogrng64-cj1-sse2-lzcnt.cpp"
+      TARGET_ADDSRCS="$TARGET_ADDSRCS $OGR/amd64/ogrng64-cj1-avx2.cpp"
       TARGET_ADDNASMS="$TARGET_ADDNASMS $OGR/amd64/ogrng64-cj1-generic-asm.asm"
       TARGET_ADDNASMS="$TARGET_ADDNASMS $OGR/amd64/ogrng64-cj1-sse2-asm.asm"
       TARGET_ADDNASMS="$TARGET_ADDNASMS $OGR/amd64/ogrng64-cj1-sse2-lzcnt-asm.asm"
+      TARGET_ADDNASMS="$TARGET_ADDNASMS $OGR/amd64/ogrng64-cj1-avx2-asm.asm"
     fi
 
     if [ "$HAVE_OGR_P2" = "1" ]; then

makefile.vc

@@ -136,7 +136,7 @@ ZIPEXTRAS = \
 
 OPTS_MSVC = -nologo -D__WIN32__ \
             -W4 -GR- -GA -GF -Gy \
-            -Dsnprintf=_snprintf -DHAVE_SNPRINTF \
+            -DHAVE_SNPRINTF \
             -D_M_$(OPTS_M_PLAT) $(OPTS_CC_CPU) $(OPTS_CC_DEBUG)
             ## *** +++++++++++++++++++++++++++++++++++++++++
 OPTS_LIBS = advapi32.lib user32.lib kernel32.lib gdi32.lib
@@ -157,6 +157,12 @@ OPTS_RC   = -d_Windows -d_M_$(OPTS_M_PLAT)
 #   cl 16.00.xxxx = Visual Studio 2010 (VC10)
 #   cl 17.00.xxxx = Visual Studio 2012 (VC11)
 #   cl 18.00.xxxx = Visual Studio 2013 (VC12)
+#   cl 19.00.xxxx = Visual Studio 2015 (VC14)
+
+# snprintf needs to be defined for Visual Studio 2015 and earlier
+!if ( [plat\win\msvcver.cmd] < 19 )
+OPTS_MSVC = $(OPTS_MSVC) -Dsnprintf=_snprintf
+!endif
 
 !if ( [plat\win\msvcver.cmd] >= 15 )
 OPTS_MSVC = $(OPTS_MSVC) -EHs-c- -GS- -wd4996
@@ -470,6 +476,8 @@ OGRNG_OBJS =                         \
         $(OUTPUTPATH)/ogrng64-cj1-sse2-asm.obj \
         $(OUTPUTPATH)/ogrng64-cj1-sse2-lzcnt.obj    \
         $(OUTPUTPATH)/ogrng64-cj1-sse2-lzcnt-asm.obj \
+        $(OUTPUTPATH)/ogrng64-cj1-avx2.obj    \
+        $(OUTPUTPATH)/ogrng64-cj1-avx2-asm.obj \
         $(OUTPUTPATH)/ogrng_init.obj \
         $(OUTPUTPATH)/ogrng_dat.obj
 !elseif "$(PROCESSOR_ARCHITECTURE)" == "x86"