libatlas/atlas-3.8.4-arm.patch

diff -ru ATLAS/CONFIG/include/atlconf.h ATLAS-3.8.4-arm/CONFIG/include/atlconf.h
--- ATLAS/CONFIG/include/atlconf.h	2011-05-14 19:33:24.000000000 +0200
+++ ATLAS-3.8.4-arm/CONFIG/include/atlconf.h	2011-05-16 13:50:44.000000000 +0200
@@ -14,9 +14,9 @@
              OSWin9x, OSWinNT, OSHPUX, OSFreeBSD, OSOSX};
 #define OSIsWin(OS_) (((OS_) == OSWinNT) || ((OS_) == OSWin9x))
 
-enum ARCHFAM {AFOther=0, AFPPC, AFSPARC, AFALPHA, AFX86, AFIA64, AFMIPS};
+enum ARCHFAM {AFOther=0, AFPPC, AFSPARC, AFALPHA, AFX86, AFIA64, AFMIPS, AFARM};
 
-#define NMACH 37
+#define NMACH 39
 static char *machnam[NMACH] =
    {"UNKNOWN", "POWER3", "POWER4", "POWER5", "PPCG4", "PPCG5",
     "POWER6", "POWER7",
@@ -25,7 +25,7 @@
     "Efficeon", "K7", "HAMMER", "AMD64K10h", "UNKNOWNx86",
     "IA64Itan", "IA64Itan2",
     "USI", "USII", "USIII", "USIV", "UST2", "UnknownUS",
-    "MIPSR1xK", "MIPSICE9"};
+    "MIPSR1xK", "MIPSICE9", "ARMv7VFP", "ARMv7VFPNEON"};
 enum MACHTYPE {MACHOther, IbmPwr3, IbmPwr4, IbmPwr5, PPCG4, PPCG5,
                IbmPwr6, IbmPwr7,
                IntP5, IntP5MMX, IntPPRO, IntPII, IntPIII, IntPM, IntCoreS,
@@ -34,7 +34,8 @@
                IA64Itan, IA64Itan2,
                SunUSI, SunUSII, SunUSIII, SunUSIV, SunUST2, SunUSX,
                MIPSR1xK, /* includes R10K, R12K, R14K, R16K */
-               MIPSICE9   /* SiCortex ICE9 -- like MIPS5K */
+               MIPSICE9,   /* SiCortex ICE9 -- like MIPS5K */
+               ARMv7VFP, ARMv7VFPNEON
                };
 #define MachIsX86(mach_) \
    ( (mach_) >= IntP5 && (mach_) <= x86X )
@@ -51,6 +52,8 @@
 #endif
 #define MachIsPPC(mach_) \
    ( (mach_) >= PPCG4 && (mach_) <= PPCG5 )
+#define MachIsARM(mach_) \
+   ( (mach_) >= ARMv7VFP && (mach_) <= ARMv7VFPNEON )
 
 static char *f2c_namestr[5] = {"UNKNOWN","Add_", "Add__", "NoChange", "UpCase"};
 static char *f2c_intstr[5] =
@@ -63,18 +66,18 @@
 enum F2CINT {f2c_IntErr=0, FintCint, FintClong, FintClonglong, FintCshort};
 enum F2CSTRING {f2c_StrErr=0, fstrSun, fstrCray, fstrStructVal, fstrStructPtr};
 
-#define NISA 6
+#define NISA 7
 static char *ISAXNAM[NISA] =
-   {"", "AltiVec", "SSE3", "SSE2", "SSE1", "3DNow"};
-enum ISAEXT {ISA_None=0, ISA_AV, ISA_SSE3, ISA_SSE2, ISA_SSE1, ISA_3DNow};
+   {"", "AltiVec", "SSE3", "SSE2", "SSE1", "3DNow", "NEON"};
+enum ISAEXT {ISA_None=0, ISA_AV, ISA_SSE3, ISA_SSE2, ISA_SSE1, ISA_3DNow, ISA_NEON};
 
-#define NASMD 7
+#define NASMD 8
 enum ASMDIA
    {ASM_None=0, gas_x86_32, gas_x86_64, gas_sparc, gas_ppc, gas_parisc,
-    gas_mips};
+    gas_mips, gas_arm};
 static char *ASMNAM[NASMD] =
    {"",     "GAS_x8632", "GAS_x8664", "GAS_SPARC", "GAS_PPC", "GAS_PARISC",
-    "GAS_MIPS"};
+    "GAS_MIPS", "GAS_ARM"};
 
 
 /*
diff -ru ATLAS/CONFIG/src/Makefile ATLAS-3.8.4-arm/CONFIG/src/Makefile
--- ATLAS/CONFIG/src/Makefile	2011-05-14 19:33:24.000000000 +0200
+++ ATLAS-3.8.4-arm/CONFIG/src/Makefile	2011-04-20 16:43:26.000000000 +0200
@@ -151,6 +151,11 @@
 	$(MAKE) $(atlrun) atldir=$(mydir) exe=xprobe_sse3 args="$(args)" \
                 redir=config0.out
 	- cat config0.out
+IRun_NEON :
+	$(CC) $(CCFLAGS) -mfpu=neon -o xprobe_neon $(SRCdir)/backend/probe_svec.c $(SRCdir)/backend/probe_NEON.S
+	$(MAKE) $(atlrun) atldir=$(mydir) exe=xprobe_neon args="$(args)" \
+                redir=config0.out
+	- cat config0.out
 
 IRun_GAS_SPARC :
 	$(CC) $(CCFLAGS) -o xprobe_gas_sparc $(SRCdir)/backend/probe_this_asm.c $(SRCdir)/backend/probe_gas_sparc.S
@@ -177,6 +182,11 @@
 	$(MAKE) $(atlrun) atldir=$(mydir) exe=xprobe_gas_x8632 args="$(args)" \
                 redir=config0.out
 	- cat config0.out
+IRun_GAS_ARM :
+	$(CC) $(CCFLAGS) -o xprobe_gas_arm $(SRCdir)/backend/probe_this_asm.c $(SRCdir)/backend/probe_gas_arm.S
+	$(MAKE) $(atlrun) atldir=$(mydir) exe=xprobe_gas_arm args="$(args)" \
+                redir=config0.out
+	- cat config0.out
 
 IRunC2C :
 	- rm -f config0.out xc2c c2cslave.o
diff -ru ATLAS/CONFIG/src/atlcomp.txt ATLAS-3.8.4-arm/CONFIG/src/atlcomp.txt
--- ATLAS/CONFIG/src/atlcomp.txt	2011-05-14 19:33:24.000000000 +0200
+++ ATLAS-3.8.4-arm/CONFIG/src/atlcomp.txt	2012-05-16 19:47:42.000000000 +0200
@@ -164,6 +164,19 @@
 MACH=P4,PM OS=WinNT LVL=0 COMPS=icc,dmc,smc,dkc,skc,xcc
    'icl' '-QxN -O3 -Qprec -fp:extended -fp:except -nologo -Oy'
 #
+# ARM defaults
+#
+MACH=ARMv7VFP,ARMv7VFPNEON OS=ALL LVL=1000 COMPS=icc,xcc
+   'gcc' '-march=armv7-a -O1'
+MACH=ARMv7VFPNEON OS=ALL LVL=1000 COMPS=smc,skc
+   'gcc' '-march=armv7-a -mfpu=neon -ffast-math -O3 -fschedule-insns -fschedule-insns2 -fprefetch-loop-arrays'
+MACH=ARMv7VFP OS=ALL LVL=1000 COMPS=smc,skc
+   'gcc' '-march=armv7-a -mfpu=vfpv3 -O1 -fno-expensive-optimizations'
+MACH=ARMv7VFP,ARMv7VFPNEON OS=ALL LVL=1000 COMPS=dmc,dkc
+   'gcc' '-march=armv7-a -mfpu=vfpv3 -O1 -fno-schedule-insns2'
+MACH=ARMv7VFP,ARMv7VFPNEON OS=ALL LVL=1000 COMPS=f77
+   'gfortran' '-march=armv7-a -O1'
+#
 # Generic defaults
 #
 MACH=ALL OS=ALL LVL=5 COMPS=icc,smc,dmc,skc,dkc,xcc
diff -ru ATLAS/CONFIG/src/atlconf_misc.c ATLAS-3.8.4-arm/CONFIG/src/atlconf_misc.c
--- ATLAS/CONFIG/src/atlconf_misc.c	2011-05-14 19:33:24.000000000 +0200
+++ ATLAS-3.8.4-arm/CONFIG/src/atlconf_misc.c	2011-04-20 16:43:26.000000000 +0200
@@ -480,6 +480,7 @@
       else if (strstr(res, "alpha")) fam = AFALPHA;
       else if (strstr(res, "ia64")) fam = AFIA64;
       else if (strstr(res, "mips")) fam = AFMIPS;
+      else if (strstr(res, "arm")) fam = AFARM;
       else if ( strstr(res, "i686") || strstr(res, "i586") ||
                 strstr(res, "i486") || strstr(res, "i386") ||
                 strstr(res, "x86") || strstr(res, "x86_64") ) fam = AFX86;
@@ -501,6 +502,7 @@
                    strstr(res, "i486") || strstr(res, "i386") ||
                    strstr(res, "x86_64") ) fam = AFX86;
          else if (strstr(res, "mips")) fam = AFMIPS;
+         else if (strstr(res, "arm")) fam = AFARM;
       }
    }
    return(fam);
diff -ru ATLAS/CONFIG/src/backend/archinfo_linux.c ATLAS-3.8.4-arm/CONFIG/src/backend/archinfo_linux.c
--- ATLAS/CONFIG/src/backend/archinfo_linux.c	2011-05-14 19:33:24.000000000 +0200
+++ ATLAS-3.8.4-arm/CONFIG/src/backend/archinfo_linux.c	2011-05-16 14:19:02.000000000 +0200
@@ -166,6 +166,20 @@
          else if (strstr(res, "AMD-K7")) mach = AmdAthlon;
       }
       break;
+   case AFARM:
+      if ( !CmndOneLine(NULL, "fgrep 'Processor' /proc/cpuinfo", res) )
+      {
+         if (strstr(res, "ARMv7") || strstr(res,"v7l")) mach = ARMv7VFP;
+      }
+      else if ( !CmndOneLine(NULL, "fgrep 'cpu' /proc/cpuinfo", res) )
+      {
+         if (strstr(res, "ARMv7") || strstr(res,"v7l")) mach = ARMv7VFP;
+      }
+      if ( !CmndOneLine(NULL, "fgrep 'Features' /proc/cpuinfo", res) )
+      {
+         if (strstr(res, "neon")) mach = ARMv7VFPNEON;
+      }
+      break;
 /*
  *    Add these back if we get machine access and can test
  */
diff -ru ATLAS/README ATLAS-3.8.4-arm/README
--- ATLAS/README	2011-05-14 19:33:23.000000000 +0200
+++ ATLAS-3.8.4-arm/README	2012-06-18 23:19:46.000000000 +0200
@@ -1,3 +1,19 @@
+ATLAS-3.8.4-arm
+released 18 June 2012
+
+Changes since beta 1:
+
+ - assembly headers changed to link properly on Unbuntu 11.10 and 12.04
+ - VFP single-precision GEMM by Clint Whaley backported from ATLAS 3.9.x
+
+Changes since beta 2:
+
+ - VFP assembly GEMM kernels added for hard floating-point ABI (like Ubuntu 12.04)
+
+Changes since beta 3:
+ (none)
+
+
 This is the README index file.  If you want install intructions,
 read ATLAS/INSTALL.txt.  Windows users should read
 ATLAS/doc/Windows.txt as well.  Note that all of this documentation is
diff -ru ATLAS/tune/blas/gemm/CASES/ccases.flg ATLAS-3.8.4-arm/tune/blas/gemm/CASES/ccases.flg
--- ATLAS/tune/blas/gemm/CASES/ccases.flg	2011-05-14 19:34:10.000000000 +0200
+++ ATLAS-3.8.4-arm/tune/blas/gemm/CASES/ccases.flg	2012-06-03 14:25:17.000000000 +0200
@@ -1,5 +1,5 @@
 <ID> <flag> <mb> <nb> <kb> <muladd> <lat> <mu> <nu> <ku> <rout> "<Contributer>"
-20
+25
 304 192 4 3 8 0 4 4 3 8 ATL_mm4x3x8p.c     "R. Clint Whaley" \
 gcc
 -mcpu=ultrasparc -mtune=ultrasparc -fomit-frame-pointer -O
@@ -52,3 +52,18 @@
 gcc
 -x assembler-with-cpp -mips4
 332 192 8 2 4 1 0 8 2 4 ATL_smm8x2x4_av.c  "IBM"
+333 448 4 4 2 1 1 4 4 2 ATL_smm4x4x2pf_arm.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+334 448 4 4 2 1 1 4 4 2 ATL_smm4x4x2dld_arm.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+335 448 4 4 2 1 1 4 4 2 ATL_smm4x4x2pf_armhf.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+336 448 4 4 2 1 1 4 4 2 ATL_smm4x4x2dld_armhf.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+337 480 4 4 4 1 1 4 4 4 ATL_smm4x4x4p_neon.c "Vesperix Corporation" \
+gcc
+-march=armv7-a -mfpu=neon -O2
diff -ru ATLAS/tune/blas/gemm/CASES/dcases.flg ATLAS-3.8.4-arm/tune/blas/gemm/CASES/dcases.flg
--- ATLAS/tune/blas/gemm/CASES/dcases.flg	2011-05-14 19:34:10.000000000 +0200
+++ ATLAS-3.8.4-arm/tune/blas/gemm/CASES/dcases.flg	2012-06-03 14:23:02.000000000 +0200
@@ -1,5 +1,5 @@
 <ID> <flag> <mb> <nb> <kb> <muladd> <lat> <mu> <nu> <ku> <rout> "<Contributer>"
-31
+33
 306 192 4 3 8 0 4 4 3 8 ATL_mm4x3x8p.c     "R. Clint Whaley" \
 gcc
 -mcpu=ultrasparc -mtune=ultrasparc -fomit-frame-pointer -O -fno-schedule-insns -fno-schedule-insns2
@@ -85,6 +85,11 @@
 338 192 8 4 2 1 0 8 4 2  ATL_dmm8x4x2_vsx.c  "IBM" \
 gcc
 -O3 -mvsx
-
+339 448 4 4 2 1 1 4 4 2 ATL_dmm4x4x2pf_arm.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+340 448 4 4 2 1 1 4 4 2 ATL_dmm4x4x2pf_armhf.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
 # NB = 80 and NB = 120 are to force UltraSparc to try NB > L1 Cache,
 # needed for optimal performance on Ultra2 and Ultra4, respectively
diff -ru ATLAS/tune/blas/gemm/CASES/scases.flg ATLAS-3.8.4-arm/tune/blas/gemm/CASES/scases.flg
--- ATLAS/tune/blas/gemm/CASES/scases.flg	2011-05-14 19:34:10.000000000 +0200
+++ ATLAS-3.8.4-arm/tune/blas/gemm/CASES/scases.flg	2012-06-03 14:25:48.000000000 +0200
@@ -1,5 +1,5 @@
 <ID> <flag> <mb> <nb> <kb> <muladd> <lat> <mu> <nu> <ku> <rout> "<Contributer>"
-22
+27
 304 192 4 3 8 0 4 4 3 8 ATL_mm4x3x8p.c     "R. Clint Whaley" \
 gcc
 -mcpu=ultrasparc -mtune=ultrasparc -fomit-frame-pointer -O
@@ -58,3 +58,18 @@
 gcc
 -x assembler-with-cpp -mips4
 332 192 8 2 4 1 0 8 2 4 ATL_smm8x2x4_av.c  "IBM"
+333 448 4 4 2 1 1 4 4 2 ATL_smm4x4x2pf_arm.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+334 448 4 4 2 1 1 4 4 2 ATL_smm4x4x2dld_arm.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+335 448 4 4 2 1 1 4 4 2 ATL_smm4x4x2pf_armhf.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+336 448 4 4 2 1 1 4 4 2 ATL_smm4x4x2dld_armhf.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+337 480 4 4 4 1 1 4 4 4 ATL_smm4x4x4p_neon.c "Vesperix Corporation" \
+gcc
+-march=armv7-a -mfpu=neon -O2
diff -ru ATLAS/tune/blas/gemm/CASES/zcases.flg ATLAS-3.8.4-arm/tune/blas/gemm/CASES/zcases.flg
--- ATLAS/tune/blas/gemm/CASES/zcases.flg	2011-05-14 19:34:10.000000000 +0200
+++ ATLAS-3.8.4-arm/tune/blas/gemm/CASES/zcases.flg	2012-06-03 14:23:36.000000000 +0200
@@ -1,5 +1,5 @@
 <ID> <flag> <mb> <nb> <kb> <muladd> <lat> <mu> <nu> <ku> <rout> "<Contributer>"
-30
+32
 306 192 4 3 8 0 4 4 3 8 ATL_mm4x3x8p.c     "R. Clint Whaley" \
 gcc
 -mcpu=ultrasparc -mtune=ultrasparc -fomit-frame-pointer -O -fno-schedule-insns -fno-schedule-insns2
@@ -82,6 +82,11 @@
 338 192 8 4 2 1 0 8 4 2  ATL_dmm8x4x2_vsx.c  "IBM" \
 gcc
 -O3 -mvsx
-
+339 448 4 4 2 1 1 4 4 2 ATL_dmm4x4x2pf_arm.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
+340 448 4 4 2 1 1 4 4 2 ATL_dmm4x4x2pf_armhf.c "R. Clint Whaley" \
+gcc
+-x assembler-with-cpp -mfpu=vfpv3 -march=armv7-a
 # NB = 80 and NB = 120 are to force UltraSparc to try NB > L1 Cache,
 # needed for optimal performance on Ultra2 and Ultra4, respectively
diff -ru ATLAS/tune/sysinfo/ATL_cputime.c ATLAS-3.8.4-arm/tune/sysinfo/ATL_cputime.c
--- ATLAS/tune/sysinfo/ATL_cputime.c	2011-05-14 19:34:09.000000000 +0200
+++ ATLAS-3.8.4-arm/tune/sysinfo/ATL_cputime.c	2012-05-05 18:16:58.000000000 +0200
@@ -48,6 +48,26 @@
       t0 = clock();
       return(0.0);
    }
+#elif defined(POSIX_HR) /* use the POSIX HR timers */
+   #include <time.h>
+   double ATL_cputime(void)
+   {
+      struct timespec ts;
+      static double t0;
+      double res;
+      static int INIT = 0;
+
+      if (INIT)
+      {
+         clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts);
+         res = ts.tv_sec + 1.0e-9 * ts.tv_nsec;
+         return(res - t0);
+      }
+      clock_gettime(CLOCK_PROCESS_CPUTIME_ID,&ts);
+      t0 = ts.tv_sec + 1.0e-9 * ts.tv_nsec;
+      INIT = 1;
+      return(0.0);
+   }
 #elif defined(UseTimes)
    #include <stdlib.h>
    #include <sys/times.h>
diff -ru ATLAS/tune/sysinfo/ATL_walltime.c ATLAS-3.8.4-arm/tune/sysinfo/ATL_walltime.c
--- ATLAS/tune/sysinfo/ATL_walltime.c	2011-05-14 19:34:09.000000000 +0200
+++ ATLAS-3.8.4-arm/tune/sysinfo/ATL_walltime.c	2012-05-05 18:23:20.000000000 +0200
@@ -79,6 +79,26 @@
       myout = (myout<<32) | msout.LowPart;
       return(myout*freqRecip);
    }
+#elif defined(POSIX_HR) /* use the POSIX HR timers */
+   #include <time.h>
+   double ATL_cputime(void)
+   {
+      struct timespec ts;
+      static double t0;
+      double res;
+      static int INIT = 0;
+
+      if (INIT)
+      {
+         clock_gettime(CLOCK_REALTIME, &ts);
+         res = ts.tv_sec + 1.0e-9 * ts.tv_nsec;
+         return(res - t0);
+      }
+      clock_gettime(CLOCK_REALTIME,&ts);
+      t0 = ts.tv_sec + 1.0e-9 * ts.tv_nsec;
+      INIT = 1;
+      return(0.0);
+   }
 #elif defined(UseTimes)
    #include <stdlib.h>
    #include <sys/times.h>
@@ -97,6 +117,26 @@
    {
       return(gethrtime()*1.0e-9);
    }
+#elif defined(POSIX_HR) /* use the POSIX HR timers */
+   #include <time.h>
+   double ATL_walltime(void)
+   {
+      struct timespec ts;
+      static double t0;
+      double res;
+      static int INIT = 0;
+
+      if (INIT)
+      {
+         clock_gettime(CLOCK_REALTIME, &ts);
+         res = ts.tv_sec + 1.0e-9 * ts.tv_nsec;
+         return(res - t0);
+      }
+      clock_gettime(CLOCK_REALTIME, &ts);
+      t0 = ts.tv_sec + 1.0e-9 * ts.tv_nsec;
+      INIT = 1;
+      return(0.0);
+   }
 /*
  * Without gcc, I know no standard Windows wall-timer, so use cputime
  */