foss/2017b toolchain definition (REVIEW)

easybuild/easyconfigs/f/FFTW/FFTW-3.3.6-gompi-2017b.eb

@@ -0,0 +1,17 @@
+name = 'FFTW'
+version = '3.3.6'
+
+homepage = 'http://www.fftw.org'
+description = """FFTW is a C subroutine library for computing the discrete Fourier transform (DFT)
+ in one or more dimensions, of arbitrary input size, and of both real and complex data."""
+
+toolchain = {'name': 'gompi', 'version': '2017b'}
+toolchainopts = {'pic': True}
+
+source_urls = [homepage]
+sources = ['fftw-%(version)s-pl2.tar.gz']
+checksums = ['a5de35c5c824a78a058ca54278c706cdf3d4abba1c56b63531c2cb05f5d57da2']
+
+runtest = 'check'
+
+moduleclass = 'numlib'

easybuild/easyconfigs/f/foss/foss-2017b.eb

@@ -0,0 +1,33 @@
+easyblock = 'Toolchain'
+
+name = 'foss'
+version = '2017b'
+
+homepage = '(none)'
+description = """GNU Compiler Collection (GCC) based compiler toolchain, including
+ OpenMPI for MPI support, OpenBLAS (BLAS and LAPACK support), FFTW and ScaLAPACK."""
+
+toolchain = {'name': 'dummy', 'version': 'dummy'}
+
+gccver = '6.4.0-2.28'
+
+blaslib = 'OpenBLAS'
+blasver = '0.2.20'
+blas = '%s-%s' % (blaslib, blasver)
+
+# toolchain used to build foss dependencies
+comp_mpi_tc_name = 'gompi'
+comp_mpi_tc = (comp_mpi_tc_name, version)
+
+# we need GCC and OpenMPI as explicit dependencies instead of gompi toolchain
+# because of toolchain preparation functions
+# For binutils, stick to http://wiki.osdev.org/Cross-Compiler_Successful_Builds
+dependencies = [
+    ('GCC', gccver),
+    ('OpenMPI', '2.1.1', '', ('GCC', gccver)),
+    (blaslib, blasver, '', ('GCC', gccver)),
+    ('FFTW', '3.3.6', '', comp_mpi_tc),
+    ('ScaLAPACK', '2.0.2', '-%s' % blas, comp_mpi_tc),
+]
+
+moduleclass = 'toolchain'

easybuild/easyconfigs/g/gompi/gompi-2017b.eb

@@ -0,0 +1,20 @@
+easyblock = "Toolchain"
+
+name = 'gompi'
+version = '2017b'
+
+homepage = '(none)'
+description = """GNU Compiler Collection (GCC) based compiler toolchain,
+ including OpenMPI for MPI support."""
+
+toolchain = {'name': 'dummy', 'version': 'dummy'}
+
+gccver = '6.4.0-2.28'
+
+# compiler toolchain dependencies
+dependencies = [
+    ('GCC', gccver),  # includes both GCC and binutils
+    ('OpenMPI', '2.1.1', '', ('GCC', gccver)),
+]
+
+moduleclass = 'toolchain'

easybuild/easyconfigs/h/HPL/HPL-2.2-foss-2017b.eb

@@ -0,0 +1,19 @@
+name = 'HPL'
+version = '2.2'
+
+homepage = 'http://www.netlib.org/benchmark/hpl/'
+description = """HPL is a software package that solves a (random) dense linear system in double precision (64 bits)
+ arithmetic on distributed-memory computers. It can thus be regarded as a portable as well as freely available
+ implementation of the High Performance Computing Linpack Benchmark."""
+
+toolchain = {'name': 'foss', 'version': '2017b'}
+toolchainopts = {'usempi': True}
+
+sources = [SOURCELOWER_TAR_GZ]
+source_urls = ['http://www.netlib.org/benchmark/%(namelower)s']
+checksums = ['ac7534163a09e21a5fa763e4e16dfc119bc84043f6e6a807aba666518f8df440']
+
+# fix Make dependencies, so parallel build also works
+patches = ['HPL_parallel-make.patch']
+
+moduleclass = 'tools'

easybuild/easyconfigs/o/OpenBLAS/OpenBLAS-0.2.20-GCC-6.4.0-2.28.eb

@@ -0,0 +1,51 @@
+easyblock = 'ConfigureMake'
+
+name = 'OpenBLAS'
+version = '0.2.20'
+
+homepage = 'http://xianyi.github.com/OpenBLAS/'
+description = "OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version."
+
+toolchain = {'name': 'GCC', 'version': '6.4.0-2.28'}
+
+large_src = 'large.tgz'
+timing_src = 'timing.tgz'
+
+source_urls = [
+    # order matters, trying to download the large.tgz/timing.tgz LAPACK tarballs from GitHub causes trouble
+    'http://www.netlib.org/lapack/timing/',
+    'https://github.com/xianyi/OpenBLAS/archive/',
+]
+sources = ['v%(version)s.tar.gz']
+checksums = [
+    '5ef38b15d9c652985774869efd548b8e3e972e1e99475c673b25537ed7bcf394',  # v0.2.20.tar.gz (OpenBLAS)
+    'f328d88b7fa97722f271d7d0cfea1c220e0f8e5ed5ff01d8ef1eb51d6f4243a1',  # large.tgz
+    '999c65f8ea8bd4eac7f1c7f3463d4946917afd20a997807300fe35d70122f3af',  # timing.tgz
+    # OpenBLAS-0.2.20_fix-Intel-L1-cache-size-detection.patch
+    '1d043e4838ec1f90b2b49318b780e3ab13b46133cb72a8d83eb0e3b1b056c4d6',
+    '1e6a046ab658c6e0b351de901d2812db28c2042f9f141416144c2faaf71fbb37',  # OpenBLAS-0.2.20_revert-honor-cpuset.patch
+]
+
+patches = [
+    (large_src, '.'),
+    (timing_src, '.'),
+    'OpenBLAS-%(version)s_fix-Intel-L1-cache-size-detection.patch',
+    'OpenBLAS-%(version)s_revert-honor-cpuset.patch',
+]
+
+skipsteps = ['configure']
+
+buildopts = 'BINARY=64 USE_THREAD=1 USE_OPENMP=1 CC="$CC" FC="$F77"'
+installopts = "USE_THREAD=1 USE_OPENMP=1 PREFIX=%(installdir)s"
+
+# extensive testing can be enabled by uncommenting the line below
+# runtest = 'PATH=.:$PATH lapack-timing'
+
+sanity_check_paths = {
+    'files': ['include/cblas.h', 'include/f77blas.h', 'include/lapacke_config.h', 'include/lapacke.h',
+              'include/lapacke_mangling.h', 'include/lapacke_utils.h', 'include/openblas_config.h',
+              'lib/libopenblas.a', 'lib/libopenblas.%s' % SHLIB_EXT],
+    'dirs': [],
+}
+
+moduleclass = 'numlib'

easybuild/easyconfigs/o/OpenBLAS/OpenBLAS-0.2.20_fix-Intel-L1-cache-size-detection.patch

@@ -0,0 +1,188 @@
+fixes detection of L1 cache size on Intel processors, incl. Intel Haswell)
+see https://github.com/xianyi/OpenBLAS/pull/1236 && https://github.com/xianyi/OpenBLAS/issues/1232
+--- a/cpuid_x86.c
++++ b/cpuid_x86.c
+@@ -71,12 +71,23 @@ void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx)
+   *edx = cpuInfo[3];
+ }
+
++void cpuid_count(int op, int count, int *eax, int *ebx, int *ecx, int *edx)
++{
++  int cpuInfo[4] = {-1};
++  __cpuidex(cpuInfo, op, count);
++  *eax = cpuInfo[0];
++  *ebx = cpuInfo[1];
++  *ecx = cpuInfo[2];
++  *edx = cpuInfo[3];
++}
++
+ #else
+
+ #ifndef CPUIDEMU
+
+ #if defined(__APPLE__) && defined(__i386__)
+ void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx);
++void cpuid_count(int op, int count, int *eax, int *ebx, int *ecx, int *edx);
+ #else
+ static C_INLINE void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
+ #if defined(__i386__) && defined(__PIC__)
+@@ -90,6 +101,19 @@ static C_INLINE void cpuid(int op, int *eax, int *ebx, int *ecx, int *edx){
+     ("cpuid": "=a" (*eax), "=b" (*ebx), "=c" (*ecx), "=d" (*edx) : "a" (op) : "cc");
+ #endif
+ }
++
++static C_INLINE void cpuid_count(int op, int count ,int *eax, int *ebx, int *ecx, int *edx){
++#if defined(__i386__) && defined(__PIC__)
++  __asm__ __volatile__
++    ("mov %%ebx, %%edi;"
++     "cpuid;"
++     "xchgl %%ebx, %%edi;"
++     : "=a" (*eax), "=D" (*ebx), "=c" (*ecx), "=d" (*edx) : "0" (op), "2" (count) : "cc");
++#else
++  __asm__ __volatile__
++    ("cpuid": "=a" (*eax), "=b" (*ebx), "=c" (*ecx), "=d" (*edx) : "0" (op), "2" (count) : "cc");
++#endif
++}
+ #endif
+
+ #else
+@@ -157,6 +157,10 @@ void cpuid(unsigned int op, unsigned int *eax, unsigned int *ebx, unsigned int *
+   *edx = idlist[current].d;
+ }
+
++void cpuid_count (unsigned int op, unsigned int count, unsigned int *eax, unsigned int *ebx, unsigned int *ecx, unsigned int *edx) {
++  return cpuid (op, eax, ebx, ecx, edx);
++}
++
+ #endif
+
+ #endif // _MSC_VER
+@@ -312,9 +336,9 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
+   cpuid(0, &cpuid_level, &ebx, &ecx, &edx);
+
+   if (cpuid_level > 1) {
+-
++    int numcalls =0 ;
+     cpuid(2, &eax, &ebx, &ecx, &edx);
+-
++    numcalls = BITMASK(eax, 0, 0xff); //FIXME some systems may require repeated calls to read all entries
+     info[ 0] = BITMASK(eax,  8, 0xff);
+     info[ 1] = BITMASK(eax, 16, 0xff);
+     info[ 2] = BITMASK(eax, 24, 0xff);
+@@ -335,7 +359,6 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
+     info[14] = BITMASK(edx, 24, 0xff);
+
+     for (i = 0; i < 15; i++){
+-
+       switch (info[i]){
+
+ 	/* This table is from http://www.sandpile.org/ia32/cpuid.htm */
+@@ -637,12 +660,13 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
+ 	LD1.linesize    = 64;
+ 	break;
+       case 0x63 :
+-  DTB.size        = 2048;
+-  DTB.associative = 4;
+-  DTB.linesize    = 32;
+-  LDTB.size       = 4096;
+-  LDTB.associative= 4;
+-  LDTB.linesize   = 32;
++  	DTB.size        = 2048;
++  	DTB.associative = 4;
++  	DTB.linesize    = 32;
++  	LDTB.size       = 4096;
++  	LDTB.associative= 4;
++  	LDTB.linesize   = 32;
++	break;
+       case 0x66 :
+ 	LD1.size        = 8;
+ 	LD1.associative = 4;
+@@ -675,12 +699,13 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
+ 	LC1.associative = 8;
+ 	break;
+       case 0x76 :
+-  ITB.size        = 2048;
+-  ITB.associative = 0;
+-  ITB.linesize    = 8;
+-  LITB.size       = 4096;
+-  LITB.associative= 0;
+-  LITB.linesize   = 8;
++  	ITB.size        = 2048;
++  	ITB.associative = 0;
++  	ITB.linesize    = 8;
++  	LITB.size       = 4096;
++  	LITB.associative= 0;
++  	LITB.linesize   = 8;
++	break;
+       case 0x77 :
+ 	LC1.size        = 16;
+ 	LC1.associative = 4;
+@@ -891,6 +916,68 @@ int get_cacheinfo(int type, cache_info_t *cacheinfo){
+   }
+
+   if (get_vendor() == VENDOR_INTEL) {
++      if(LD1.size<=0 || LC1.size<=0){
++	//If we didn't detect L1 correctly before,
++	int count;
++	for (count=0;count <4;count++) {
++	cpuid_count(4, count, &eax, &ebx, &ecx, &edx);
++        switch (eax &0x1f) {
++        case 0:
++          continue;
++          case 1:
++          case 3:
++          {
++            switch ((eax >>5) &0x07)
++            {
++            case 1:
++            {
++//            fprintf(stderr,"L1 data cache...\n");
++            int sets = ecx+1;
++            int lines = (ebx & 0x0fff) +1;
++            ebx>>=12;
++            int part = (ebx&0x03ff)+1;
++            ebx >>=10;
++            int assoc = (ebx&0x03ff)+1;
++            LD1.size = (assoc*part*lines*sets)/1024;
++            LD1.associative = assoc;
++            LD1.linesize= lines;
++            break;
++            }
++            default: 
++              break;
++           }
++          break;
++          }
++         case 2:
++          {
++            switch ((eax >>5) &0x07)
++            {
++            case 1:
++            {
++//            fprintf(stderr,"L1 instruction cache...\n");
++            int sets = ecx+1;
++            int lines = (ebx & 0x0fff) +1;
++            ebx>>=12;
++            int part = (ebx&0x03ff)+1;
++            ebx >>=10;
++            int assoc = (ebx&0x03ff)+1;
++            LC1.size = (assoc*part*lines*sets)/1024;
++            LC1.associative = assoc;
++            LC1.linesize= lines;
++            break;
++            }
++            default: 
++              break;
++           }
++          break;
++          
++          }
++          default:
++          break;
++        }
++      }
++    }
++
+     cpuid(0x80000000, &cpuid_level, &ebx, &ecx, &edx);
+     if (cpuid_level >= 0x80000006) {
+       if(L2.size<=0){