nmatrix 0.0.9 → 0.1.0.rc1

data/ext/nmatrix/data/meta.h

@@ -9,8 +9,8 @@
 //
 // == Copyright Information
 //
-// SciRuby is Copyright (c) 2010 - 2013, Ruby Science Foundation
-// NMatrix is Copyright (c) 2013, Ruby Science Foundation
+// SciRuby is Copyright (c) 2010 - 2014, Ruby Science Foundation
+// NMatrix is Copyright (c) 2012 - 2014, John Woods and the Ruby Science Foundation
 //
 // Please see LICENSE.txt for additional copyright notices.
 //

data/ext/nmatrix/data/rational.h

@@ -9,8 +9,8 @@
 //
 // == Copyright Information
 //
-// SciRuby is Copyright (c) 2010 - 2013, Ruby Science Foundation
-// NMatrix is Copyright (c) 2013, Ruby Science Foundation
+// SciRuby is Copyright (c) 2010 - 2014, Ruby Science Foundation
+// NMatrix is Copyright (c) 2012 - 2014, John Woods and the Ruby Science Foundation
 //
 // Please see LICENSE.txt for additional copyright notices.
 //

data/ext/nmatrix/data/ruby_object.h

@@ -9,8 +9,8 @@
 //
 // == Copyright Information
 //
-// SciRuby is Copyright (c) 2010 - 2013, Ruby Science Foundation
-// NMatrix is Copyright (c) 2013, Ruby Science Foundation
+// SciRuby is Copyright (c) 2010 - 2014, Ruby Science Foundation
+// NMatrix is Copyright (c) 2012 - 2014, John Woods and the Ruby Science Foundation
 //
 // Please see LICENSE.txt for additional copyright notices.
 //

data/ext/nmatrix/extconf.rb

@@ -8,8 +8,8 @@
 #
 # == Copyright Information
 #
-# SciRuby is Copyright (c) 2010 - 2013, Ruby Science Foundation
-# NMatrix is Copyright (c) 2013, Ruby Science Foundation
+# SciRuby is Copyright (c) 2010 - 2014, Ruby Science Foundation
+# NMatrix is Copyright (c) 2012 - 2014, John Woods and the Ruby Science Foundation
 #
 # Please see LICENSE.txt for additional copyright notices.
 #
@@ -102,22 +102,65 @@ if /cygwin|mingw/ =~ RUBY_PLATFORM
 end
 
 $DEBUG = true
-$CFLAGS = ["-Wall ",$CFLAGS].join(" ")
-
-$srcs = [
-         'nmatrix.cpp',
-         'ruby_constants.cpp',
-
-         'data/data.cpp',
-         'math.cpp',
-         'util/sl_list.cpp',
-         'util/io.cpp',
-         'storage/common.cpp',
-         'storage/storage.cpp',
-         'storage/dense.cpp',
-         'storage/yale/yale.cpp',
-         'storage/list.cpp'
-        ]
+$CFLAGS = ["-Wall -Werror=return-type",$CFLAGS].join(" ")
+$CXXFLAGS = ["-Wall -Werror=return-type",$CXXFLAGS].join(" ")
+$CPPFLAGS = ["-Wall -Werror=return-type",$CPPFLAGS].join(" ")
+
+# When adding objects here, make sure their directories are included in CLEANOBJS down at the bottom of extconf.rb.
+basenames = %w{nmatrix ruby_constants data/data util/io math util/sl_list storage/common storage/storage storage/dense/dense storage/yale/yale storage/list/list}
+$objs = basenames.map { |b| "#{b}.o"   }
+$srcs = basenames.map { |b| "#{b}.cpp" }
+
+#CONFIG['CXX'] = 'clang++'
+CONFIG['CXX'] = 'g++'
+
+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 #:nodoc:
+  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++'
+  $CPP_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'
+    $CPP_STANDARD = 'c++0x'
+  else
+    $CPP_STANDARD = 'c++11'
+  end
+  puts "using C++ standard... #{$CPP_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
 
 # The next line allows the user to supply --with-atlas-dir=/usr/local/atlas,
@@ -138,9 +181,16 @@ idefaults = {lapack: ["/usr/include/atlas"],
              cblas: ["/usr/local/atlas/include", "/usr/include/atlas"],
              atlas: ["/usr/local/atlas/include", "/usr/include/atlas"]}
 
-ldefaults = {lapack: ["/usr/local/lib", "/usr/local/atlas/lib"],
-             cblas: ["/usr/local/lib", "/usr/local/atlas/lib"],
-             atlas: ["/usr/local/atlas/lib", "/usr/local/lib", "/usr/lib"]}
+# 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.
+ldefaults = {lapack: ["/usr/local/lib", "/usr/local/atlas/lib", "/usr/lib64/atlas"].delete_if { |d| !Dir.exists?(d) },
+             cblas: ["/usr/local/lib", "/usr/local/atlas/lib", "/usr/lib64/atlas"].delete_if { |d| !Dir.exists?(d) },
+             atlas: ["/usr/local/lib", "/usr/local/atlas/lib", "/usr/lib", "/usr/lib64/atlas"].delete_if { |d| !Dir.exists?(d) }}
+
+if have_library("clapack") # Usually only applies for Mac OS X
+  $libs += " -lclapack "
+end
 
 unless have_library("lapack")
   dir_config("lapack", idefaults[:lapack], ldefaults[:lapack])
@@ -154,22 +204,19 @@ unless have_library("atlas")
   dir_config("atlas", idefaults[:atlas], ldefaults[:atlas])
 end
 
-# this needs to go before cblas.h checks -- on Ubuntu, the clapack in the
-# include path found for cblas.h doesn't seem to contain all the necessary 
-# functions
-have_header("clapack.h")
-
-# this ensures that we find the header on Ubuntu, where by default the library 
-# can be found but not the header
-unless have_header("cblas.h")
-  find_header("cblas.h", *idefaults[:cblas])
+# If BLAS and LAPACK headers are in an atlas directory, prefer those. Otherwise,
+# we try our luck with the default location.
+if have_header("atlas/cblas.h")
+  have_header("atlas/clapack.h")
+else
+  have_header("cblas.h")
+  have_header("clapack.h")
 end
 
-have_header("cblas.h")
 
 have_func("clapack_dgetrf", ["cblas.h", "clapack.h"])
 have_func("clapack_dgetri", ["cblas.h", "clapack.h"])
-have_func("dgesvd_", "clapack.h")
+have_func("dgesvd_", "clapack.h") # This may not do anything. dgesvd_ seems to be in LAPACK, not CLAPACK.
 
 have_func("cblas_dgemm", "cblas.h")
 
@@ -178,59 +225,15 @@ have_func("cblas_dgemm", "cblas.h")
 #find_library("lapack", "clapack_dgetrf")
 #find_library("cblas", "cblas_dgemm")
 #find_library("atlas", "ATL_dgemmNN")
-
 # Order matters here: ATLAS has to go after LAPACK: http://mail.scipy.org/pipermail/scipy-user/2007-January/010717.html
 $libs += " -llapack -lcblas -latlas "
+#$libs += " -lprofiler "
 
-$objs = %w{nmatrix ruby_constants data/data util/io math util/sl_list storage/common storage/storage storage/dense storage/yale/yale storage/list}.map { |i| i + ".o" }
-
-#CONFIG['CXX'] = 'clang++'
-CONFIG['CXX'] = 'g++'
-
-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 #:nodoc:
-  `LANG="en_US" #{CONFIG['CXX']} -v 2>&1`.lines.to_a.last.match(/gcc\sversion\s(\d\.\d.\d)/).captures.first
-end
-
-
-if CONFIG['CXX'] == 'clang++'
-  $CPP_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'
-    $CPP_STANDARD = 'c++0x'
-  else
-    $CPP_STANDARD = 'c++11'
-  end
-  puts "using C++ standard... #{$CPP_STANDARD}"
-  puts "g++ reports version... " + `#{CONFIG['CXX']} --version|head -n 1|cut -f 3 -d " "`
-end
 
 # For release, these next two should both be changed to -O3.
-$CFLAGS += " -O3 " #" -O0 -g "
+$CFLAGS += " -O3 -g" #" -O0 -g "
 #$CFLAGS += " -static -O0 -g "
-$CPPFLAGS += " -O3 -std=#{$CPP_STANDARD} " #" -O0 -g -std=#{$CPP_STANDARD} " #-fmax-errors=10 -save-temps
+$CPPFLAGS += " -O3 -std=#{$CPP_STANDARD} -g" #" -O0 -g -std=#{$CPP_STANDARD} " #-fmax-errors=10 -save-temps
 #$CPPFLAGS += " -static -O0 -g -std=#{$CPP_STANDARD} "
 
 CONFIG['warnflags'].gsub!('-Wshorten-64-to-32', '') # doesn't work except in Mac-patched gcc (4.2)
@@ -244,15 +247,13 @@ Dir.mkdir("data") unless Dir.exists?("data")
 Dir.mkdir("util") unless Dir.exists?("util")
 Dir.mkdir("storage") unless Dir.exists?("storage")
 Dir.chdir("storage") do
-  Dir.mkdir("yale") unless Dir.exists?("yale")
-  Dir.chdir("yale") do
-    Dir.mkdir("iterators") unless Dir.exists?("iterators")
-  end
+  Dir.mkdir("yale")  unless Dir.exists?("yale")
+  Dir.mkdir("list")  unless Dir.exists?("list")
+  Dir.mkdir("dense") unless Dir.exists?("dense")
 end
 
 # to clean up object files in subdirectories:
 open('Makefile', 'a') do |f|
-  f.write <<EOS
-CLEANOBJS := $(CLEANOBJS) data/*.#{CONFIG["OBJEXT"]} storage/*.#{CONFIG["OBJEXT"]} util/*.#{CONFIG["OBJEXT"]}
-EOS
+  clean_objs_paths = %w{data storage storage/dense storage/yale storage/list util}.map { |d| "#{d}/*.#{CONFIG["OBJEXT"]}" }
+  f.write("CLEANOBJS := $(CLEANOBJS) #{clean_objs_paths.join(' ')}")
 end

data/ext/nmatrix/math.cpp

@@ -9,8 +9,8 @@
 //
 // == Copyright Information
 //
-// SciRuby is Copyright (c) 2010 - 2013, Ruby Science Foundation
-// NMatrix is Copyright (c) 2013, Ruby Science Foundation
+// SciRuby is Copyright (c) 2010 - 2014, Ruby Science Foundation
+// NMatrix is Copyright (c) 2012 - 2014, John Woods and the Ruby Science Foundation
 //
 // Please see LICENSE.txt for additional copyright notices.
 //
@@ -140,7 +140,7 @@
 #include "math/rot.h"
 #include "math/rotg.h"
 #include "math/math.h"
-#include "storage/dense.h"
+#include "storage/dense/dense.h"
 
 #include "nmatrix.h"
 #include "ruby_constants.h"
@@ -150,8 +150,10 @@
  */
 
 extern "C" {
-#ifdef HAVE_CLAPACK_H
+#if defined HAVE_CLAPACK_H
   #include <clapack.h>
+#elif defined HAVE_ATLAS_CLAPACK_H
+  #include <atlas/clapack.h>
 #endif
 
   static VALUE nm_cblas_nrm2(VALUE self, VALUE n, VALUE x, VALUE incx);
@@ -503,8 +505,10 @@ static VALUE nm_cblas_rotg(VALUE self, VALUE ab) {
     return Qnil;
 
   } else {
-    void *pC = ALLOCA_N(char, DTYPE_SIZES[dtype]),
-         *pS = ALLOCA_N(char, DTYPE_SIZES[dtype]);
+    NM_CONSERVATIVE(nm_register_value(self));
+    NM_CONSERVATIVE(nm_register_value(ab));
+    void *pC = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]),
+         *pS = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]);
 
     // extract A and B from the NVector (first two elements)
     void* pA = NM_STORAGE_DENSE(ab)->elements;
@@ -522,7 +526,8 @@ static VALUE nm_cblas_rotg(VALUE self, VALUE ab) {
       rb_ary_store(result, 0, rubyobj_from_cval(pC, dtype).rval);
       rb_ary_store(result, 1, rubyobj_from_cval(pS, dtype).rval);
     }
-
+    NM_CONSERVATIVE(nm_unregister_value(ab));
+    NM_CONSERVATIVE(nm_unregister_value(self));
     return result;
   }
 }
@@ -575,18 +580,18 @@ static VALUE nm_cblas_rot(VALUE self, VALUE n, VALUE x, VALUE incx, VALUE y, VAL
 
     // We need to ensure the cosine and sine arguments are the correct dtype -- which may differ from the actual dtype.
     if (dtype == nm::COMPLEX64) {
-      pC = ALLOCA_N(float,1);
-      pS = ALLOCA_N(float,1);
+      pC = NM_ALLOCA_N(float,1);
+      pS = NM_ALLOCA_N(float,1);
       rubyval_to_cval(c, nm::FLOAT32, pC);
       rubyval_to_cval(s, nm::FLOAT32, pS);
     } else if (dtype == nm::COMPLEX128) {
-      pC = ALLOCA_N(double,1);
-      pS = ALLOCA_N(double,1);
+      pC = NM_ALLOCA_N(double,1);
+      pS = NM_ALLOCA_N(double,1);
       rubyval_to_cval(c, nm::FLOAT64, pC);
       rubyval_to_cval(s, nm::FLOAT64, pS);
     } else {
-      pC = ALLOCA_N(char, DTYPE_SIZES[dtype]);
-      pS = ALLOCA_N(char, DTYPE_SIZES[dtype]);
+      pC = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]);
+      pS = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]);
       rubyval_to_cval(c, dtype, pC);
       rubyval_to_cval(s, dtype, pS);
     }
@@ -646,7 +651,7 @@ static VALUE nm_cblas_nrm2(VALUE self, VALUE n, VALUE x, VALUE incx) {
     if      (dtype == nm::COMPLEX64)  rdtype = nm::FLOAT32;
     else if (dtype == nm::COMPLEX128) rdtype = nm::FLOAT64;
 
-    void *Result = ALLOCA_N(char, DTYPE_SIZES[rdtype]);
+    void *Result = NM_ALLOCA_N(char, DTYPE_SIZES[rdtype]);
 
     ttable[dtype](FIX2INT(n), NM_STORAGE_DENSE(x)->elements, FIX2INT(incx), Result);
 
@@ -698,7 +703,7 @@ static VALUE nm_cblas_asum(VALUE self, VALUE n, VALUE x, VALUE incx) {
   if      (dtype == nm::COMPLEX64)  rdtype = nm::FLOAT32;
   else if (dtype == nm::COMPLEX128) rdtype = nm::FLOAT64;
 
-  void *Result = ALLOCA_N(char, DTYPE_SIZES[rdtype]);
+  void *Result = NM_ALLOCA_N(char, DTYPE_SIZES[rdtype]);
 
   ttable[dtype](FIX2INT(n), NM_STORAGE_DENSE(x)->elements, FIX2INT(incx), Result);
 
@@ -743,8 +748,8 @@ static VALUE nm_cblas_gemm(VALUE self,
 
   nm::dtype_t dtype = NM_DTYPE(a);
 
-  void *pAlpha = ALLOCA_N(char, DTYPE_SIZES[dtype]),
-       *pBeta  = ALLOCA_N(char, DTYPE_SIZES[dtype]);
+  void *pAlpha = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]),
+       *pBeta  = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]);
   rubyval_to_cval(alpha, dtype, pAlpha);
   rubyval_to_cval(beta, dtype, pBeta);
 
@@ -788,8 +793,8 @@ static VALUE nm_cblas_gemv(VALUE self,
 
   nm::dtype_t dtype = NM_DTYPE(a);
 
-  void *pAlpha = ALLOCA_N(char, DTYPE_SIZES[dtype]),
-       *pBeta  = ALLOCA_N(char, DTYPE_SIZES[dtype]);
+  void *pAlpha = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]),
+       *pBeta  = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]);
   rubyval_to_cval(alpha, dtype, pAlpha);
   rubyval_to_cval(beta, dtype, pBeta);
 
@@ -825,7 +830,7 @@ static VALUE nm_cblas_trsm(VALUE self,
   if (!ttable[dtype]) {
     rb_raise(nm_eDataTypeError, "this matrix operation undefined for integer matrices");
   } else {
-    void *pAlpha = ALLOCA_N(char, DTYPE_SIZES[dtype]);
+    void *pAlpha = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]);
     rubyval_to_cval(alpha, dtype, pAlpha);
 
     ttable[dtype](blas_order_sym(order), blas_side_sym(side), blas_uplo_sym(uplo), blas_transpose_sym(trans_a), blas_diag_sym(diag), FIX2INT(m), FIX2INT(n), pAlpha, NM_STORAGE_DENSE(a)->elements, FIX2INT(lda), NM_STORAGE_DENSE(b)->elements, FIX2INT(ldb));
@@ -865,7 +870,7 @@ static VALUE nm_cblas_trmm(VALUE self,
   if (!ttable[dtype]) {
     rb_raise(nm_eDataTypeError, "this matrix operation not yet defined for non-BLAS dtypes");
   } else {
-    void *pAlpha = ALLOCA_N(char, DTYPE_SIZES[dtype]);
+    void *pAlpha = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]);
     rubyval_to_cval(alpha, dtype, pAlpha);
 
     ttable[dtype](blas_order_sym(order), blas_side_sym(side), blas_uplo_sym(uplo), blas_transpose_sym(trans_a), blas_diag_sym(diag), FIX2INT(m), FIX2INT(n), pAlpha, NM_STORAGE_DENSE(a)->elements, FIX2INT(lda), NM_STORAGE_DENSE(b)->elements, FIX2INT(ldb));
@@ -903,8 +908,8 @@ static VALUE nm_cblas_syrk(VALUE self,
   if (!ttable[dtype]) {
     rb_raise(nm_eDataTypeError, "this matrix operation undefined for integer matrices");
   } else {
-    void *pAlpha = ALLOCA_N(char, DTYPE_SIZES[dtype]),
-         *pBeta = ALLOCA_N(char, DTYPE_SIZES[dtype]);
+    void *pAlpha = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]),
+         *pBeta = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]);
     rubyval_to_cval(alpha, dtype, pAlpha);
     rubyval_to_cval(beta, dtype, pBeta);
 
@@ -984,12 +989,12 @@ static VALUE nm_lapack_gesvd(VALUE self, VALUE jobu, VALUE jobvt, VALUE m, VALUE
 
     // only need rwork for complex matrices
     int rwork_size  = (dtype == nm::COMPLEX64 || dtype == nm::COMPLEX128) ? 5 * min_mn : 0;
-    void* rwork     = rwork_size > 0 ? ALLOCA_N(char, DTYPE_SIZES[dtype] * rwork_size) : NULL;
+    void* rwork     = rwork_size > 0 ? NM_ALLOCA_N(char, DTYPE_SIZES[dtype] * rwork_size) : NULL;
     int work_size   = FIX2INT(lwork);
 
     // ignore user argument for lwork if it's too small.
     work_size       = NM_MAX((dtype == nm::COMPLEX64 || dtype == nm::COMPLEX128 ? 2 * min_mn + max_mn : NM_MAX(3*min_mn + max_mn, 5*min_mn)), work_size);
-    void* work      = ALLOCA_N(char, DTYPE_SIZES[dtype] * work_size);
+    void* work      = NM_ALLOCA_N(char, DTYPE_SIZES[dtype] * work_size);
 
     int info = gesvd_table[dtype](JOBU, JOBVT, M, N, NM_STORAGE_DENSE(a)->elements, FIX2INT(lda),
       NM_STORAGE_DENSE(s)->elements, NM_STORAGE_DENSE(u)->elements, FIX2INT(ldu), NM_STORAGE_DENSE(vt)->elements, FIX2INT(ldvt),
@@ -1046,7 +1051,7 @@ static VALUE nm_lapack_gesdd(VALUE self, VALUE jobz, VALUE m, VALUE n, VALUE a,
     int work_size = FIX2INT(lwork); // Make sure we allocate enough work, regardless of the user request.
     if (dtype == nm::COMPLEX64 || dtype == nm::COMPLEX128) {
       int rwork_size = min_mn * (JOBZ == 'N' ? 5 : NM_MAX(5*min_mn + 7, 2*max_mn + 2*min_mn + 1));
-      rwork = ALLOCA_N(char, DTYPE_SIZES[dtype] * rwork_size);
+      rwork = NM_ALLOCA_N(char, DTYPE_SIZES[dtype] * rwork_size);
 
       if (JOBZ == 'N')      work_size = NM_MAX(work_size, 3*min_mn + NM_MAX(max_mn, 6*min_mn));
       else if (JOBZ == 'O') work_size = NM_MAX(work_size, 3*min_mn*min_mn + NM_MAX(max_mn, 5*min_mn*min_mn + 4*min_mn));
@@ -1056,8 +1061,8 @@ static VALUE nm_lapack_gesdd(VALUE self, VALUE jobz, VALUE m, VALUE n, VALUE a,
       else if (JOBZ == 'O') work_size = NM_MAX(work_size, 2*min_mn*min_mn + max_mn + 2*min_mn);
       else                  work_size = NM_MAX(work_size, min_mn*min_mn + max_mn + 2*min_mn);
     }
-    void* work  = ALLOCA_N(char, DTYPE_SIZES[dtype] * work_size);
-    int* iwork  = ALLOCA_N(int, 8*min_mn);
+    void* work  = NM_ALLOCA_N(char, DTYPE_SIZES[dtype] * work_size);
+    int* iwork  = NM_ALLOCA_N(int, 8*min_mn);
 
     int info = gesdd_table[dtype](JOBZ, M, N, NM_STORAGE_DENSE(a)->elements, FIX2INT(lda),
       NM_STORAGE_DENSE(s)->elements, NM_STORAGE_DENSE(u)->elements, FIX2INT(ldu), NM_STORAGE_DENSE(vt)->elements, FIX2INT(ldvt),
@@ -1114,7 +1119,7 @@ static VALUE nm_lapack_geev(VALUE self, VALUE compute_left, VALUE compute_right,
 
     // only need rwork for complex matrices (wi == Qnil for complex)
     int rwork_size  = dtype == nm::COMPLEX64 || dtype == nm::COMPLEX128 ? N * DTYPE_SIZES[dtype] : 0; // 2*N*floattype for complex only, otherwise 0
-    void* rwork     = rwork_size > 0 ? ALLOCA_N(char, rwork_size) : NULL;
+    void* rwork     = rwork_size > 0 ? NM_ALLOCA_N(char, rwork_size) : NULL;
     int work_size   = FIX2INT(lwork);
     void* work;
 
@@ -1123,11 +1128,11 @@ static VALUE nm_lapack_geev(VALUE self, VALUE compute_left, VALUE compute_right,
     // if work size is 0 or -1, query.
     if (work_size <= 0) {
       work_size = -1;
-      work = ALLOC_N(char, DTYPE_SIZES[dtype]); //2*N * DTYPE_SIZES[dtype]);
+      work = NM_ALLOC_N(char, DTYPE_SIZES[dtype]); //2*N * DTYPE_SIZES[dtype]);
       info = geev_table[dtype](JOBVL, JOBVR, N, A, FIX2INT(lda), WR, WI, VL, FIX2INT(ldvl), VR, FIX2INT(ldvr), work, work_size, rwork);
       work_size = (int)(dtype == nm::COMPLEX64 || dtype == nm::FLOAT32 ? reinterpret_cast<float*>(work)[0] : reinterpret_cast<double*>(work)[0]);
       // line above is basically: work_size = (int)(work[0]); // now have new work_size
-      xfree(work);
+      NM_FREE(work);
       if (info == 0)
         rb_warn("geev: calculated optimal lwork of %d; to eliminate this message, use a positive value for lwork (at least 2*shape[i])", work_size);
       else return INT2FIX(info); // error of some kind on query!
@@ -1140,7 +1145,7 @@ static VALUE nm_lapack_geev(VALUE self, VALUE compute_left, VALUE compute_right,
     }
 
     // Allocate work array for actual run
-    work = ALLOCA_N(char, work_size * DTYPE_SIZES[dtype]);
+    work = NM_ALLOCA_N(char, work_size * DTYPE_SIZES[dtype]);
 
     // Perform the actual calculation.
     info = geev_table[dtype](JOBVL, JOBVR, N, A, FIX2INT(lda), WR, WI, VL, FIX2INT(ldvl), VR, FIX2INT(ldvr), work, work_size, rwork);
@@ -1158,7 +1163,7 @@ static VALUE nm_lapack_geev(VALUE self, VALUE compute_left, VALUE compute_right,
 static VALUE nm_clapack_scal(VALUE self, VALUE n, VALUE scale, VALUE vector, VALUE incx) {
   nm::dtype_t dtype = NM_DTYPE(vector);
 
-  void* da      = ALLOCA_N(char, DTYPE_SIZES[dtype]);
+  void* da      = NM_ALLOCA_N(char, DTYPE_SIZES[dtype]);
   rubyval_to_cval(scale, dtype, da);
 
   NAMED_DTYPE_TEMPLATE_TABLE(ttable, nm::math::clapack_scal, void, const int n, const void* da, void* dx, const int incx);
@@ -1251,7 +1256,7 @@ static VALUE nm_clapack_getrf(VALUE self, VALUE order, VALUE m, VALUE n, VALUE a
 
   // Allocate the pivot index array, which is of size MIN(M, N).
   size_t ipiv_size = std::min(M,N);
-  int* ipiv = ALLOCA_N(int, ipiv_size);
+  int* ipiv = NM_ALLOCA_N(int, ipiv_size);
 
   if (!ttable[NM_DTYPE(a)]) {
     rb_raise(nm_eDataTypeError, "this matrix operation undefined for integer matrices");
@@ -1282,7 +1287,7 @@ static VALUE nm_clapack_getrf(VALUE self, VALUE order, VALUE m, VALUE n, VALUE a
  */
 static VALUE nm_clapack_potrf(VALUE self, VALUE order, VALUE uplo, VALUE n, VALUE a, VALUE lda) {
 #ifndef HAVE_CLAPACK_H
-  rb_raise(rb_eNotImpError, "potrf currently requires LAPACK");
+  rb_raise(rb_eNotImpError, "potrf currently requires CLAPACK");
 #endif
 
   static int (*ttable[nm::NUM_DTYPES])(const enum CBLAS_ORDER, const enum CBLAS_UPLO, const int n, void* a, const int lda) = {
@@ -1343,7 +1348,7 @@ static VALUE nm_clapack_getrs(VALUE self, VALUE order, VALUE trans, VALUE n, VAL
   if (TYPE(ipiv) != T_ARRAY) {
     rb_raise(rb_eArgError, "ipiv must be of type Array");
   } else {
-    ipiv_ = ALLOCA_N(int, RARRAY_LEN(ipiv));
+    ipiv_ = NM_ALLOCA_N(int, RARRAY_LEN(ipiv));
     for (int index = 0; index < RARRAY_LEN(ipiv); ++index) {
       ipiv_[index] = FIX2INT( RARRAY_PTR(ipiv)[index] );
     }
@@ -1411,7 +1416,7 @@ static VALUE nm_clapack_potrs(VALUE self, VALUE order, VALUE uplo, VALUE n, VALU
  */
 static VALUE nm_clapack_getri(VALUE self, VALUE order, VALUE n, VALUE a, VALUE lda, VALUE ipiv) {
 #ifndef HAVE_CLAPACK_H
-  rb_raise(rb_eNotImpError, "getri currently requires LAPACK");
+  rb_raise(rb_eNotImpError, "getri currently requires CLAPACK");
 #endif
 
   static int (*ttable[nm::NUM_DTYPES])(const enum CBLAS_ORDER, const int n, void* a, const int lda, const int* ipiv) = {
@@ -1437,7 +1442,7 @@ static VALUE nm_clapack_getri(VALUE self, VALUE order, VALUE n, VALUE a, VALUE l
   if (TYPE(ipiv) != T_ARRAY) {
     rb_raise(rb_eArgError, "ipiv must be of type Array");
   } else {
-    ipiv_ = ALLOCA_N(int, RARRAY_LEN(ipiv));
+    ipiv_ = NM_ALLOCA_N(int, RARRAY_LEN(ipiv));
     for (int index = 0; index < RARRAY_LEN(ipiv); ++index) {
       ipiv_[index] = FIX2INT( RARRAY_PTR(ipiv)[index] );
     }
@@ -1468,7 +1473,7 @@ static VALUE nm_clapack_getri(VALUE self, VALUE order, VALUE n, VALUE a, VALUE l
  */
 static VALUE nm_clapack_potri(VALUE self, VALUE order, VALUE uplo, VALUE n, VALUE a, VALUE lda) {
 #ifndef HAVE_CLAPACK_H
-  rb_raise(rb_eNotImpError, "getri currently requires LAPACK");
+  rb_raise(rb_eNotImpError, "getri currently requires CLAPACK");
 #endif
 
   static int (*ttable[nm::NUM_DTYPES])(const enum CBLAS_ORDER, const enum CBLAS_UPLO, const int n, void* a, const int lda) = {
@@ -1534,7 +1539,7 @@ static VALUE nm_clapack_laswp(VALUE self, VALUE n, VALUE a, VALUE lda, VALUE k1,
   if (TYPE(ipiv) != T_ARRAY) {
     rb_raise(rb_eArgError, "ipiv must be of type Array");
   } else {
-    ipiv_ = ALLOCA_N(int, RARRAY_LEN(ipiv));
+    ipiv_ = NM_ALLOCA_N(int, RARRAY_LEN(ipiv));
     for (int index = 0; index < RARRAY_LEN(ipiv); ++index) {
       ipiv_[index] = FIX2INT( RARRAY_PTR(ipiv)[index] );
     }