nmatrix-lapacke 0.2.1 → 0.2.3

data/ext/nmatrix/math/trsm.h

@@ -81,10 +81,14 @@ inline void trsm_nothrow(const enum CBLAS_SIDE side, const enum CBLAS_UPLO uplo,
   // (row-major) trsm: left upper trans nonunit m=3 n=1 1/1 a 3 b 3
 
   if (m == 0 || n == 0) return; /* Quick return if possible. */
+  
+  // Apply necessary offset
+  a -= 1 + lda;
+  b -= 1 + ldb;
 
   if (alpha == 0) { // Handle alpha == 0
-    for (int j = 0; j < n; ++j) {
-      for (int i = 0; i < m; ++i) {
+    for (int j = 1; j <= n; ++j) {
+      for (int i = 1; i <= m; ++i) {
         b[i + j * ldb] = 0;
       }
     }
@@ -96,37 +100,37 @@ inline void trsm_nothrow(const enum CBLAS_SIDE side, const enum CBLAS_UPLO uplo,
 
       /* Form  B := alpha*inv( A )*B. */
       if (uplo == CblasUpper) {
-        for (int j = 0; j < n; ++j) {
+        for (int j = 1; j <= n; ++j) {
           if (alpha != 1) {
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + j * ldb] = alpha * b[i + j * ldb];
             }
           }
-          for (int k = m-1; k >= 0; --k) {
+          for (int k = m; k >= 1; --k) {
             if (b[k + j * ldb] != 0) {
               if (diag == CblasNonUnit) {
                 b[k + j * ldb] /= a[k + k * lda];
               }
 
-              for (int i = 0; i < k-1; ++i) {
+              for (int i = 1; i <= k-1; ++i) {
                 b[i + j * ldb] -= b[k + j * ldb] * a[i + k * lda];
               }
             }
           }
         }
       } else {
-        for (int j = 0; j < n; ++j) {
+        for (int j = 1; j <= n; ++j) {
           if (alpha != 1) {
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + j * ldb] = alpha * b[i + j * ldb];
             }
           }
-          for (int k = 0; k < m; ++k) {
+          for (int k = 1; k <= m; ++k) {
             if (b[k + j * ldb] != 0.) {
               if (diag == CblasNonUnit) {
                 b[k + j * ldb] /= a[k + k * lda];
               }
-              for (int i = k+1; i < m; ++i) {
+              for (int i = k+1; i <= m; ++i) {
                 b[i + j * ldb] -= b[k + j * ldb] * a[i + k * lda];
               }
             }
@@ -137,10 +141,10 @@ inline void trsm_nothrow(const enum CBLAS_SIDE side, const enum CBLAS_UPLO uplo,
 
       /*           Form  B := alpha*inv( A**T )*B. */
       if (uplo == CblasUpper) {
-        for (int j = 0; j < n; ++j) {
-          for (int i = 0; i < m; ++i) {
+        for (int j = 1; j <= n; ++j) {
+          for (int i = 1; i <= m; ++i) {
             DType temp = alpha * b[i + j * ldb];
-            for (int k = 0; k < i; ++k) { // limit was i-1. Lots of similar bugs in this code, probably.
+            for (int k = 1; k <= i-1; ++k) { // limit was i-1. Lots of similar bugs in this code, probably.
               temp -= a[k + i * lda] * b[k + j * ldb];
             }
             if (diag == CblasNonUnit) {
@@ -150,10 +154,10 @@ inline void trsm_nothrow(const enum CBLAS_SIDE side, const enum CBLAS_UPLO uplo,
           }
         }
       } else {
-        for (int j = 0; j < n; ++j) {
-          for (int i = m-1; i >= 0; --i) {
+        for (int j = 1; j <= n; ++j) {
+          for (int i = m; i >= 1; --i) {
             DType temp= alpha * b[i + j * ldb];
-            for (int k = i+1; k < m; ++k) {
+            for (int k = i+1; k <= m; ++k) {
               temp -= a[k + i * lda] * b[k + j * ldb];
             }
             if (diag == CblasNonUnit) {
@@ -171,37 +175,37 @@ inline void trsm_nothrow(const enum CBLAS_SIDE side, const enum CBLAS_UPLO uplo,
       /*           Form  B := alpha*B*inv( A ). */
 
       if (uplo == CblasUpper) {
-        for (int j = 0; j < n; ++j) {
+        for (int j = 1; j <= n; ++j) {
           if (alpha != 1) {
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + j * ldb] = alpha * b[i + j * ldb];
             }
           }
-          for (int k = 0; k < j-1; ++k) {
+          for (int k = 1; k <= j-1; ++k) {
             if (a[k + j * lda] != 0) {
-              for (int i = 0; i < m; ++i) {
+              for (int i = 1; i <= m; ++i) {
                 b[i + j * ldb] -= a[k + j * lda] * b[i + k * ldb];
               }
             }
           }
           if (diag == CblasNonUnit) {
             DType temp = 1 / a[j + j * lda];
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + j * ldb] = temp * b[i + j * ldb];
             }
           }
         }
       } else {
-        for (int j = n-1; j >= 0; --j) {
+        for (int j = n; j >= 1; --j) {
           if (alpha != 1) {
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + j * ldb] = alpha * b[i + j * ldb];
             }
           }
 
-          for (int k = j+1; k < n; ++k) {
+          for (int k = j+1; k <= n; ++k) {
             if (a[k + j * lda] != 0.) {
-              for (int i = 0; i < m; ++i) {
+              for (int i = 1; i <= m; ++i) {
                 b[i + j * ldb] -= a[k + j * lda] * b[i + k * ldb];
               }
             }
@@ -209,7 +213,7 @@ inline void trsm_nothrow(const enum CBLAS_SIDE side, const enum CBLAS_UPLO uplo,
           if (diag == CblasNonUnit) {
             DType temp = 1 / a[j + j * lda];
 
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + j * ldb] = temp * b[i + j * ldb];
             }
           }
@@ -220,45 +224,45 @@ inline void trsm_nothrow(const enum CBLAS_SIDE side, const enum CBLAS_UPLO uplo,
       /*           Form  B := alpha*B*inv( A**T ). */
 
       if (uplo == CblasUpper) {
-        for (int k = n-1; k >= 0; --k) {
+        for (int k = n; k >= 1; --k) {
           if (diag == CblasNonUnit) {
             DType temp= 1 / a[k + k * lda];
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + k * ldb] = temp * b[i + k * ldb];
             }
           }
-          for (int j = 0; j < k-1; ++j) {
+          for (int j = 1; j <= k-1; ++j) {
             if (a[j + k * lda] != 0.) {
               DType temp= a[j + k * lda];
-              for (int i = 0; i < m; ++i) {
+              for (int i = 1; i <= m; ++i) {
                 b[i + j * ldb] -= temp * b[i + k *  ldb];
               }
             }
           }
           if (alpha != 1) {
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + k * ldb] = alpha * b[i + k * ldb];
             }
           }
         }
       } else {
-        for (int k = 0; k < n; ++k) {
+        for (int k = 1; k <= n; ++k) {
           if (diag == CblasNonUnit) {
             DType temp = 1 / a[k + k * lda];
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + k * ldb] = temp * b[i + k * ldb];
             }
           }
-          for (int j = k+1; j < n; ++j) {
+          for (int j = k+1; j <= n; ++j) {
             if (a[j + k * lda] != 0.) {
               DType temp = a[j + k * lda];
-              for (int i = 0; i < m; ++i) {
+              for (int i = 1; i <= m; ++i) {
                 b[i + j * ldb] -= temp * b[i + k * ldb];
               }
             }
           }
           if (alpha != 1) {
-            for (int i = 0; i < m; ++i) {
+            for (int i = 1; i <= m; ++i) {
               b[i + k * ldb] = alpha * b[i + k * ldb];
             }
           }

data/ext/nmatrix/math/util.h

@@ -70,6 +70,20 @@ static inline enum CBLAS_SIDE blas_side_sym(VALUE op) {
   return CblasLeft;
 }
 
+/*
+ * Interprets the LAPACK side argument which could be :left or :right
+ * 
+ * Related to obtaining Q in QR factorization after calling lapack_geqrf
+ */
+
+static inline char lapacke_side_sym(VALUE op) {
+  ID op_id = rb_to_id(op);
+  if (op_id == nm_rb_left)  return 'L';
+  if (op_id == nm_rb_right) return 'R';
+  else rb_raise(rb_eArgError, "Expected :left or :right for side argument");
+  return 'L';
+}
+
 /*
  * Interprets cblas argument which could be :upper or :lower
  *

data/ext/nmatrix/nmatrix.h

@@ -33,6 +33,7 @@
  */
 
 #include <ruby.h>
+#include "ruby_constants.h"
 
 #ifdef __cplusplus
   #include <cmath>
@@ -57,6 +58,28 @@
   #include "nm_memory.h"
 #endif
 
+#ifndef RB_BUILTIN_TYPE
+# define RB_BUILTIN_TYPE(obj) BUILTIN_TYPE(obj)
+#endif
+
+#ifndef RB_FLOAT_TYPE_P
+/* NOTE: assume flonum doesn't exist */
+# define RB_FLOAT_TYPE_P(obj) ( \
+    (!SPECIAL_CONST_P(obj) && BUILTIN_TYPE(obj) == T_FLOAT))
+#endif
+
+#ifndef RB_TYPE_P
+# define RB_TYPE_P(obj, type) ( \
+    ((type) == T_FIXNUM) ? FIXNUM_P(obj) : \
+    ((type) == T_TRUE) ? ((obj) == Qtrue) : \
+    ((type) == T_FALSE) ? ((obj) == Qfalse) : \
+    ((type) == T_NIL) ? ((obj) == Qnil) : \
+    ((type) == T_UNDEF) ? ((obj) == Qundef) : \
+    ((type) == T_SYMBOL) ? SYMBOL_P(obj) : \
+    ((type) == T_FLOAT) ? RB_FLOAT_TYPE_P(obj) : \
+    (!SPECIAL_CONST_P(obj) && BUILTIN_TYPE(obj) == (type)))
+#endif
+
 #ifndef FIX_CONST_VALUE_PTR
 # if defined(__fcc__) || defined(__fcc_version) || \
     defined(__FCC__) || defined(__FCC_VERSION)
@@ -343,11 +366,25 @@ NM_DEF_STRUCT_POST(NM_GC_HOLDER);       // };
 
 #define NM_SRC(val)             (NM_STORAGE(val)->src)
 #define NM_DIM(val)             (NM_STORAGE(val)->dim)
+
+// Returns an int corresponding the data type of the nmatrix. See the dtype_t
+// enum for a list of possible data types.
 #define NM_DTYPE(val)           (NM_STORAGE(val)->dtype)
+
+// Returns a number corresponding the storage type of the nmatrix. See the stype_t
+// enum for a list of possible storage types.
 #define NM_STYPE(val)           (NM_STRUCT(val)->stype)
+
+// Get the shape of the ith dimension (int)
 #define NM_SHAPE(val,i)         (NM_STORAGE(val)->shape[(i)])
+
+// Get the shape of the 0th dimension (int)
 #define NM_SHAPE0(val)          (NM_STORAGE(val)->shape[0])
+
+// Get the shape of the 1st dimenension (int)
 #define NM_SHAPE1(val)          (NM_STORAGE(val)->shape[1])
+
+// Get the default value assigned to the nmatrix.
 #define NM_DEFAULT_VAL(val)     (NM_STORAGE_LIST(val)->default_val)
 
 // Number of elements in a dense nmatrix.
@@ -366,7 +403,8 @@ NM_DEF_STRUCT_POST(NM_GC_HOLDER);       // };
 
 #define RB_FILE_EXISTS(fn)   (rb_funcall(rb_const_get(rb_cObject, rb_intern("File")), rb_intern("exists?"), 1, (fn)) == Qtrue)
 
-#define CheckNMatrixType(v)   if (TYPE(v) != T_DATA || (RDATA(v)->dfree != (RUBY_DATA_FUNC)nm_delete && RDATA(v)->dfree != (RUBY_DATA_FUNC)nm_delete_ref)) rb_raise(rb_eTypeError, "expected NMatrix on left-hand side of operation");
+#define IsNMatrixType(v)  (RB_TYPE_P(v, T_DATA) && (RDATA(v)->dfree == (RUBY_DATA_FUNC)nm_delete || RDATA(v)->dfree == (RUBY_DATA_FUNC)nm_delete_ref))
+#define CheckNMatrixType(v)   if (!IsNMatrixType(v)) rb_raise(rb_eTypeError, "expected NMatrix on left-hand side of operation");
 
 #define NM_IsNMatrix(obj) \
   (rb_obj_is_kind_of(obj, cNMatrix) == Qtrue)

data/ext/nmatrix/storage/common.h

@@ -34,6 +34,7 @@
 
 #include <ruby.h>
 #include <cmath> // pow().
+#include <type_traits>
 
 /*
  * Project Includes
@@ -45,6 +46,11 @@
  * Macros
  */
 
+#define u_int8_t static_assert(false, "Please use uint8_t for cross-platform support and consistency."); uint8_t
+#define u_int16_t static_assert(false, "Please use uint16_t for cross-platform support and consistency."); uint16_t
+#define u_int32_t static_assert(false, "Please use uint32_t for cross-platform support and consistency."); uint32_t
+#define u_int64_t static_assert(false, "Please use uint64_t for cross-platform support and consistency."); uint64_t
+
 extern "C" {
 
 /*
@@ -152,7 +158,7 @@ namespace nm {
   EWOP_INT_INT_DIV(int16_t, int32_t)
   EWOP_INT_INT_DIV(int16_t, int64_t)
   EWOP_INT_INT_DIV(int8_t, int8_t)
-  EWOP_INT_UINT_DIV(int8_t, u_int8_t)
+  EWOP_INT_UINT_DIV(int8_t, uint8_t)
   EWOP_INT_INT_DIV(int8_t, int16_t)
   EWOP_INT_INT_DIV(int8_t, int32_t)
   EWOP_INT_INT_DIV(int8_t, int64_t)
@@ -162,12 +168,12 @@ namespace nm {
   EWOP_UINT_INT_DIV(uint8_t, int32_t)
   EWOP_UINT_INT_DIV(uint8_t, int64_t)
   EWOP_FLOAT_INT_DIV(float, int8_t)
-  EWOP_FLOAT_INT_DIV(float, u_int8_t)
+  EWOP_FLOAT_INT_DIV(float, uint8_t)
   EWOP_FLOAT_INT_DIV(float, int16_t)
   EWOP_FLOAT_INT_DIV(float, int32_t)
   EWOP_FLOAT_INT_DIV(float, int64_t)
   EWOP_FLOAT_INT_DIV(double, int8_t)
-  EWOP_FLOAT_INT_DIV(double, u_int8_t)
+  EWOP_FLOAT_INT_DIV(double, uint8_t)
   EWOP_FLOAT_INT_DIV(double, int16_t)
   EWOP_FLOAT_INT_DIV(double, int32_t)
   EWOP_FLOAT_INT_DIV(double, int64_t)

data/ext/nmatrix/storage/yale/class.h

@@ -376,7 +376,7 @@ public:
       v       = reinterpret_cast<D*>(s->elements);
       v_size  = nm_storage_count_max_elements(s);
 
-    } else if (TYPE(right) == T_ARRAY) {
+    } else if (RB_TYPE_P(right, T_ARRAY)) {
       v_size = RARRAY_LEN(right);
       v      = NM_ALLOC_N(D, v_size);
       if (dtype() == nm::RUBYOBJ) {

data/ext/nmatrix_lapacke/extconf.rb

@@ -25,74 +25,11 @@
 # This file checks for ATLAS and other necessary headers, and
 # generates a Makefile for compiling NMatrix.
 
-require "mkmf"
-
-
-# Function derived from NArray's extconf.rb.
-def have_type(type, header=nil) #:nodoc:
-  printf "checking for %s... ", type
-  STDOUT.flush
-
-  src = <<"SRC"
-#include <ruby.h>
-SRC
-
-
-  src << <<"SRC" unless header.nil?
-#include <#{header}>
-SRC
-
-  r = try_link(src + <<"SRC")
-  int main() { return 0; }
-  int t() { #{type} a; return 0; }
-SRC
-
-  unless r
-    print "no\n"
-    return false
-  end
-
-  $defs.push(format("-DHAVE_%s", type.upcase))
-
-  print "yes\n"
-
-  return true
-end
-
-# Function derived from NArray's extconf.rb.
-def create_conf_h(file) #:nodoc:
-  print "creating #{file}\n"
-  File.open(file, 'w') do |hfile|
-    header_guard = file.upcase.sub(/\s|\./, '_')
-
-    hfile.puts "#ifndef #{header_guard}"
-    hfile.puts "#define #{header_guard}"
-    hfile.puts
-
-    # FIXME: Find a better way to do this:
-    hfile.puts "#define RUBY_2 1" if RUBY_VERSION >= '2.0'
-
-    for line in $defs
-      line =~ /^-D(.*)/
-      hfile.printf "#define %s 1\n", $1
-    end
-
-    hfile.puts
-    hfile.puts "#endif"
-  end
-end
-
-if RUBY_VERSION < '1.9'
-  raise(NotImplementedError, "Sorry, you need at least Ruby 1.9!")
-else
-  #$INSTALLFILES = [['nmatrix.h', '$(archdir)'], ['nmatrix.hpp', '$(archdir)'], ['nmatrix_config.h', '$(archdir)'], ['nm_memory.h', '$(archdir)']]
-  if /cygwin|mingw/ =~ RUBY_PLATFORM
-    #$INSTALLFILES << ['libnmatrix.a', '$(archdir)']
-  end
-end
+require "nmatrix/mkmf"
 
+#$INSTALLFILES = [['nmatrix.h', '$(archdir)'], ['nmatrix.hpp', '$(archdir)'], ['nmatrix_config.h', '$(archdir)'], ['nm_memory.h', '$(archdir)']]
 if /cygwin|mingw/ =~ RUBY_PLATFORM
-  CONFIG["DLDFLAGS"] << " --output-lib libnmatrix.a"
+  #$INSTALLFILES << ['libnmatrix.a', '$(archdir)']
 end
 
 $DEBUG = true
@@ -107,56 +44,6 @@ basenames = %w{nmatrix_lapacke math_lapacke lapacke}
 $objs = basenames.map { |b| "#{b}.o"   }
 $srcs = basenames.map { |b| "#{b}.cpp" }
 
-def find_newer_gplusplus #:nodoc:
-  print "checking for apparent GNU g++ binary with C++0x/C++11 support... "
-  [9,8,7,6,5,4,3].each do |minor|
-    ver = "4.#{minor}"
-    gpp = "g++-#{ver}"
-    result = `which #{gpp}`
-    next if result.empty?
-    CONFIG['CXX'] = gpp
-    puts ver
-    return CONFIG['CXX']
-  end
-  false
-end
-
-def gplusplus_version
-  cxxvar = proc { |n| `#{CONFIG['CXX']} -E -dM - </dev/null | grep #{n}`.chomp.split(' ')[2] }
-  major = cxxvar.call('__GNUC__')
-  minor = cxxvar.call('__GNUC_MINOR__')
-  patch = cxxvar.call('__GNUC_PATCHLEVEL__')
-
-  raise("unable to determine g++ version (match to get version was nil)") if major.nil? || minor.nil? || patch.nil?
-
-  "#{major}.#{minor}.#{patch}"
-end
-
-
-if CONFIG['CXX'] == 'clang++'
-  $CXX_STANDARD = 'c++11'
-
-else
-  version = gplusplus_version
-  if version < '4.3.0' && CONFIG['CXX'] == 'g++'  # see if we can find a newer G++, unless it's been overridden by user
-    if !find_newer_gplusplus
-      raise("You need a version of g++ which supports -std=c++0x or -std=c++11. If you're on a Mac and using Homebrew, we recommend using mac-brew-gcc.sh to install a more recent g++.")
-    end
-    version = gplusplus_version
-  end
-
-  if version < '4.7.0'
-    $CXX_STANDARD = 'c++0x'
-  else
-    $CXX_STANDARD = 'c++11'
-  end
-  puts "using C++ standard... #{$CXX_STANDARD}"
-  puts "g++ reports version... " + `#{CONFIG['CXX']} --version|head -n 1|cut -f 3 -d " "`
-end
-
-# add smmp in to get generic transp; remove smmp2 to eliminate funcptr transp
-
-
 # For some reason, if we try to look for /usr/lib64/atlas on a Mac OS X Mavericks system, and the directory does not
 # exist, it will give a linker error -- even if the lib dir is already correctly included with -L. So we need to check
 # that Dir.exists?(d) for each.
@@ -170,32 +57,12 @@ unless have_library("lapack")
   dir_config("lapack", idefaults[:lapack], ldefaults[:lapack])
 end
 
-
-
-#have_func("rb_scan_args", "ruby.h")
-
 # Order matters here: ATLAS has to go after LAPACK: http://mail.scipy.org/pipermail/scipy-user/2007-January/010717.html
 $libs += " -llapack "
 #To use the Intel MKL, comment out the line above, and also comment out the bit above with have_library and dir_config for lapack.
 #Then add something like the line below (for exactly what linker flags to use see https://software.intel.com/en-us/articles/intel-mkl-link-line-advisor ):
 #$libs += " -L${MKLROOT}/lib/intel64 -lmkl_intel_lp64 -lmkl_core -lmkl_sequential "
 
-
-# For release, these next two should both be changed to -O3.
-$CFLAGS += " -O3" #" -O0 -g "
-#$CFLAGS += " -static -O0 -g "
-$CXXFLAGS += " -O3 -std=#{$CXX_STANDARD}" #" -O0 -g -std=#{$CXX_STANDARD} " #-fmax-errors=10 -save-temps
-#$CPPFLAGS += " -static -O0 -g -std=#{$CXX_STANDARD} "
-
-CONFIG['warnflags'].gsub!('-Wshorten-64-to-32', '') # doesn't work except in Mac-patched gcc (4.2)
-CONFIG['warnflags'].gsub!('-Wdeclaration-after-statement', '')
-CONFIG['warnflags'].gsub!('-Wimplicit-function-declaration', '')
-
-have_func("rb_array_const_ptr", "ruby.h")
-have_macro("FIX_CONST_VALUE_PTR", "ruby.h")
-have_macro("RARRAY_CONST_PTR", "ruby.h")
-have_macro("RARRAY_AREF", "ruby.h")
-
 create_conf_h("nmatrix_lapacke_config.h")
 create_makefile("nmatrix_lapacke")