lbfgsb 0.3.1 → 0.5.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: b3bd36b7d31d1cbc04ef90be26df3fcf67b73a1dbeba544a1c838f6b35786115
-  data.tar.gz: d09cc5f75361c4ec6c542ed7b663d87c8654e806cfdcd81415754f963074fbd1
+  metadata.gz: d0558e429ff4c590e681165e0af885de04e9b78d4d679afc89f9c0bcfc1b2764
+  data.tar.gz: 5e1a8a157614da1a8541fbe28aad8cb04bf4096a4d3022f1b519c35abf071544
 SHA512:
-  metadata.gz: ce048d758955f403e2d6c176f316e8f7e5db58dc142f91a0ee343fd0115a17b9b61c6a26a145eb1553dab6644f3f6cc5f6fa8ad04898287393247ef34de58fc0
-  data.tar.gz: 6d6e7dd5fbc941b37542de548385a6202322fbfebe5a6a77a9afbb169dcf9d421f7d79435dda881b5c3174948b47eb603a065c6726011f010c66e45b674c41d5
+  metadata.gz: 66997d33a6fd0fb750ae236934fa866a94592161cbba49375bd7d46d94a9c98f0177cd207431d24cd1812bbf10be050106c941de9ac972f7699d0687b97a41c8
+  data.tar.gz: d8ba46b919591239180642f9a45fe5f14337f2aa703ac527ba533449c647be4b64de700a49a88616019241010d37e6118904c992c3886ed7e44c620f771cadc9

data/CHANGELOG.md

@@ -1,3 +1,29 @@
+## 0.5.0
+- Add build option to select FORTRAN integer bit size.
+
+```
+$ gem install lbfgsb -- --with-use-int64
+```
+
+- Add build option to use any blas library.
+
+```
+$ gem install lbfgsb -- --with-blas-dir=/opt/local/openblas/lib --with-blas-lib=openblas
+```
+
+- Change to use 32-bit integer for FORTRAN integer in internal functions by default.
+- Introduce conventional commits.
+
+## 0.4.1
+- Remove dependent gem's type declaration file from installation files.
+
+## 0.4.0
+- Add type declaration file: sig/lbfgsb.rbs
+
+## 0.3.2
+- Refactor native extension codes.
+- Update documentations.
+
 ## 0.3.1
 - Add GC guard to narray given to native extension method.
 

data/LICENSE.txt

@@ -1,4 +1,4 @@
-Copyright (c) 2020-2021 Atsushi Tatsuma
+Copyright (c) 2020-2022 Atsushi Tatsuma
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without

data/README.md

@@ -2,7 +2,7 @@
 
 [![Build Status](https://github.com/yoshoku/lbfgsb.rb/workflows/build/badge.svg)](https://github.com/yoshoku/lbfgsb.rb/actions?query=workflow%3Abuild)
 [![Gem Version](https://badge.fury.io/rb/lbfgsb.svg)](https://badge.fury.io/rb/lbfgsb)
-[![BSD 3-Clause License](https://img.shields.io/badge/License-BSD%203--Clause-orange.svg)](https://github.com/yoshoku/suika/blob/master/LICENSE.txt)
+[![BSD 3-Clause License](https://img.shields.io/badge/License-BSD%203--Clause-orange.svg)](https://github.com/yoshoku/suika/blob/main/LICENSE.txt)
 [![Documentation](http://img.shields.io/badge/api-reference-blue.svg)](https://yoshoku.github.io/lbfgsb.rb/doc/)
 
 Lbfgsb.rb is a Ruby binding for [L-BFGS-B](http://users.iems.northwestern.edu/~nocedal/lbfgsb.html)

data/ext/lbfgsb/extconf.rb

@@ -1,6 +1,8 @@
 require 'mkmf'
 require 'numo/narray'
 
+$defs << '-DUSE_INT64' if with_config('use-int64', false)
+
 $LOAD_PATH.each do |lp|
   if File.exist?(File.join(lp, 'numo/numo/narray.h'))
     $INCFLAGS = "-I#{lp}/numo #{$INCFLAGS}"
@@ -26,10 +28,17 @@ if RUBY_PLATFORM =~ /mswin|cygwin|mingw/
   end
 end
 
-$srcs = Dir.glob("#{$srcdir}/*.c").map { |path| File.basename(path) }
-$srcs.concat(%w[blas.c linpack.c lbfgsb.c])
+$srcs = Dir.glob("#{$srcdir}/**/*.c").map { |path| File.basename(path) }
+
+blas_dir = with_config('blas-dir')
+$LDFLAGS = "-L#{blas_dir} #{$LDFLAGS}" unless blas_dir.nil?
+
+blas_lib = with_config('blas-lib')
+unless blas_lib.nil?
+  abort "#{blas_lib} not found." unless have_library(blas_lib)
+  $srcs.delete('blas.c')
+end
 
-$INCFLAGS << " -I$(srcdir)/src"
 $VPATH << "$(srcdir)/src"
 
 create_makefile('lbfgsb/lbfgsbext')

data/ext/lbfgsb/lbfgsbext.c

@@ -2,40 +2,46 @@
 
 VALUE rb_mLbfgsb;
 
-static
-VALUE lbfgsb_min_l_bfgs_b(VALUE self,
-  VALUE fnc, VALUE x_val, VALUE jcb, VALUE args,
-  VALUE l_val, VALUE u_val, VALUE nbd_val,
-  VALUE maxcor, VALUE ftol, VALUE gtol, VALUE maxiter, VALUE disp)
-{
-  long i;
-  long n_iter;
-  long n_fev;
-  long n_jev;
-  long max_iter = NUM2LONG(maxiter);
+static VALUE lbfgsb_min_l_bfgs_b(VALUE self, VALUE fnc, VALUE x_val, VALUE jcb, VALUE args, VALUE l_val, VALUE u_val,
+                                 VALUE nbd_val, VALUE maxcor, VALUE ftol, VALUE gtol, VALUE maxiter, VALUE disp) {
+  F77_int n_iter;
+  F77_int n_fev;
+  F77_int n_jev;
+#ifdef USE_INT64
+  F77_int max_iter = NUM2LONG(maxiter);
+#else
+  F77_int max_iter = NUM2INT(maxiter);
+#endif
   narray_t* x_nary;
   narray_t* l_nary;
   narray_t* u_nary;
   narray_t* nbd_nary;
-  long n;
-  long m = NUM2LONG(maxcor);
-  double *x_ptr;
-  double *l_ptr;
-  double *u_ptr;
-  long *nbd_ptr;
+  F77_int n;
+#ifdef USE_INT64
+  F77_int m = NUM2LONG(maxcor);
+#else
+  F77_int m = NUM2INT(maxcor);
+#endif
+  double* x_ptr;
+  double* l_ptr;
+  double* u_ptr;
+  F77_int* nbd_ptr;
   double f;
-  double *g;
+  double* g;
   double factr = NUM2DBL(ftol);
   double pgtol = NUM2DBL(gtol);
   double* wa;
-  long* iwa;
+  F77_int* iwa;
   char task[60];
-  long iprint = NIL_P(disp) ? -1 : NUM2LONG(disp);
+#ifdef USE_INT64
+  F77_int iprint = NIL_P(disp) ? -1 : NUM2LONG(disp);
+#else
+  F77_int iprint = NIL_P(disp) ? -1 : NUM2INT(disp);
+#endif
   char csave[60];
-  long lsave[4];
-  long isave[44];
+  F77_int lsave[4];
+  F77_int isave[44];
   double dsave[29];
-  double* g_ptr;
   VALUE g_val;
   VALUE fg_arr;
   VALUE ret;
@@ -45,7 +51,7 @@ VALUE lbfgsb_min_l_bfgs_b(VALUE self,
     rb_raise(rb_eArgError, "x must be a 1-D array.");
     return Qnil;
   }
-  n = (long)NA_SIZE(x_nary);
+  n = (F77_int)NA_SIZE(x_nary);
   if (CLASS_OF(x_val) != numo_cDFloat) {
     x_val = rb_funcall(numo_cDFloat, rb_intern("cast"), 1, x_val);
   }
@@ -58,7 +64,7 @@ VALUE lbfgsb_min_l_bfgs_b(VALUE self,
     rb_raise(rb_eArgError, "l must be a 1-D array.");
     return Qnil;
   }
-  if ((long)NA_SIZE(l_nary) != n) {
+  if ((F77_int)NA_SIZE(l_nary) != n) {
     rb_raise(rb_eArgError, "The size of l must be equal to that of x.");
     return Qnil;
   }
@@ -74,7 +80,7 @@ VALUE lbfgsb_min_l_bfgs_b(VALUE self,
     rb_raise(rb_eArgError, "u must be a 1-D array.");
     return Qnil;
   }
-  if ((long)NA_SIZE(u_nary) != n) {
+  if ((F77_int)NA_SIZE(u_nary) != n) {
     rb_raise(rb_eArgError, "The size of u must be equal to that of x.");
     return Qnil;
   }
@@ -90,13 +96,19 @@ VALUE lbfgsb_min_l_bfgs_b(VALUE self,
     rb_raise(rb_eArgError, "nbd must be a 1-D array.");
     return Qnil;
   }
-  if ((long)NA_SIZE(nbd_nary) != n) {
+  if ((F77_int)NA_SIZE(nbd_nary) != n) {
     rb_raise(rb_eArgError, "The size of nbd must be equal to that of x.");
     return Qnil;
   }
+#ifdef USE_INT64
   if (CLASS_OF(nbd_val) != numo_cInt64) {
     nbd_val = rb_funcall(numo_cInt64, rb_intern("cast"), 1, nbd_val);
   }
+#else
+  if (CLASS_OF(nbd_val) != numo_cInt32) {
+    nbd_val = rb_funcall(numo_cInt32, rb_intern("cast"), 1, nbd_val);
+  }
+#endif
   if (!RTEST(nary_check_contiguous(nbd_val))) {
     nbd_val = nary_dup(nbd_val);
   }
@@ -104,23 +116,20 @@ VALUE lbfgsb_min_l_bfgs_b(VALUE self,
   x_ptr = (double*)na_get_pointer_for_read_write(x_val);
   l_ptr = (double*)na_get_pointer_for_read(l_val);
   u_ptr = (double*)na_get_pointer_for_read(u_val);
-  nbd_ptr = (long*)na_get_pointer_for_read(nbd_val);
+  nbd_ptr = (F77_int*)na_get_pointer_for_read(nbd_val);
   g = ALLOC_N(double, n);
   wa = ALLOC_N(double, (2 * m + 5) * n + 12 * m * m + 12 * m);
-  iwa = ALLOC_N(long, 3 * n);
+  iwa = ALLOC_N(F77_int, 3 * n);
 
   g_val = Qnil;
   f = 0.0;
-  for (i = 0; i < n; g[i++] = 0.0);
+  memset(g, 0, n * sizeof(*g));
   strcpy(task, "START");
   n_fev = 0;
   n_jev = 0;
 
   for (n_iter = 0; n_iter < max_iter;) {
-    setulb_(
-      &n, &m, x_ptr, l_ptr, u_ptr, nbd_ptr, &f, g, &factr, &pgtol, wa, iwa,
-      task, &iprint, csave, lsave, isave, dsave
-    );
+    setulb_(&n, &m, x_ptr, l_ptr, u_ptr, nbd_ptr, &f, g, &factr, &pgtol, wa, iwa, task, &iprint, csave, lsave, isave, dsave);
     if (strncmp(task, "FG", 2) == 0) {
       if (RB_TYPE_P(jcb, T_TRUE)) {
         fg_arr = rb_funcall(self, rb_intern("fnc"), 3, fnc, x_val, args);
@@ -130,15 +139,16 @@ VALUE lbfgsb_min_l_bfgs_b(VALUE self,
         f = NUM2DBL(rb_funcall(self, rb_intern("fnc"), 3, fnc, x_val, args));
         g_val = rb_funcall(self, rb_intern("jcb"), 3, jcb, x_val, args);
       }
-      n_fev += 1;
-      n_jev += 1;
-      if (CLASS_OF(g_val) != numo_cDFloat) g_val = rb_funcall(numo_cDFloat, rb_intern("cast"), 1, g_val);
-      if (!RTEST(nary_check_contiguous(g_val))) g_val = nary_dup(g_val);
-      g_ptr = (double*)na_get_pointer_for_read(g_val);
-      for (i = 0; i < n; i++) g[i] = g_ptr[i];
+      n_fev++;
+      n_jev++;
+      if (CLASS_OF(g_val) != numo_cDFloat)
+        g_val = rb_funcall(numo_cDFloat, rb_intern("cast"), 1, g_val);
+      if (!RTEST(nary_check_contiguous(g_val)))
+        g_val = nary_dup(g_val);
+      memcpy(g, na_get_pointer_for_read(g_val), n * sizeof(*g));
+      RB_GC_GUARD(g_val);
     } else if (strncmp(task, "NEW_X", 5) == 0) {
-      n_iter += 1;
-      continue;
+      n_iter++;
     } else {
       break;
     }
@@ -153,9 +163,15 @@ VALUE lbfgsb_min_l_bfgs_b(VALUE self,
   rb_hash_aset(ret, ID2SYM(rb_intern("x")), x_val);
   rb_hash_aset(ret, ID2SYM(rb_intern("fnc")), DBL2NUM(f));
   rb_hash_aset(ret, ID2SYM(rb_intern("jcb")), g_val);
+#ifdef USE_INT64
   rb_hash_aset(ret, ID2SYM(rb_intern("n_iter")), LONG2NUM(n_iter));
   rb_hash_aset(ret, ID2SYM(rb_intern("n_fev")), LONG2NUM(n_fev));
   rb_hash_aset(ret, ID2SYM(rb_intern("n_jev")), LONG2NUM(n_jev));
+#else
+  rb_hash_aset(ret, ID2SYM(rb_intern("n_iter")), INT2NUM(n_iter));
+  rb_hash_aset(ret, ID2SYM(rb_intern("n_fev")), INT2NUM(n_fev));
+  rb_hash_aset(ret, ID2SYM(rb_intern("n_jev")), INT2NUM(n_jev));
+#endif
   rb_hash_aset(ret, ID2SYM(rb_intern("success")), strncmp(task, "CONV", 4) == 0 ? Qtrue : Qfalse);
 
   RB_GC_GUARD(x_val);
@@ -166,12 +182,17 @@ VALUE lbfgsb_min_l_bfgs_b(VALUE self,
   return ret;
 }
 
-void
-Init_lbfgsbext(void)
-{
+void Init_lbfgsbext(void) {
   rb_mLbfgsb = rb_define_module("Lbfgsb");
   /* The value of double epsilon used in the native extension. */
   rb_define_const(rb_mLbfgsb, "DBL_EPSILON", DBL2NUM(DBL_EPSILON));
+#ifdef USE_INT64
+  /* The bit size of fortran integer. */
+  rb_define_const(rb_mLbfgsb, "SZ_F77_INTEGER", INT2NUM(64));
+#else
+  /* The bit size of fortran integer. */
+  rb_define_const(rb_mLbfgsb, "SZ_F77_INTEGER", INT2NUM(32));
+#endif
   /* @!visibility private */
   rb_define_module_function(rb_mLbfgsb, "min_l_bfgs_b", lbfgsb_min_l_bfgs_b, 12);
 }

data/ext/lbfgsb/lbfgsbext.h

@@ -2,11 +2,13 @@
 #define LBFGSB_RB_H 1
 
 #include <float.h>
+#include <string.h>
 
 #include <ruby.h>
+
 #include <numo/narray.h>
 #include <numo/template.h>
 
-#include <lbfgsb.h>
+#include "src/lbfgsb.h"
 
 #endif /* LBFGSB_RB_H */

data/ext/lbfgsb/src/blas.c

@@ -5,10 +5,9 @@
  */
 #include "blas.h"
 
-int lbfgsb_rb_daxpy_(long *n, double *da, double *dx, long *incx, double *dy, long *incy)
-{
-  long i__1;
-  static long i__, m, ix, iy, mp1;
+void daxpy_(F77_int* n, double* da, double* dx, F77_int* incx, double* dy, F77_int* incy) {
+  F77_int i__1;
+  static F77_int i__, m, ix, iy, mp1;
 
   --dy;
   --dx;
@@ -17,10 +16,10 @@ int lbfgsb_rb_daxpy_(long *n, double *da, double *dx, long *incx, double *dy, lo
   /* uses unrolled loops for increments equal to one. */
   /* jack dongarra, linpack, 3/11/78. */
   if (*n <= 0) {
-    return 0;
+    return;
   }
   if (*da == 0.) {
-    return 0;
+    return;
   }
   if (*incx == 1 && *incy == 1) {
     goto L20;
@@ -42,7 +41,7 @@ int lbfgsb_rb_daxpy_(long *n, double *da, double *dx, long *incx, double *dy, lo
     ix += *incx;
     iy += *incy;
   }
-  return 0;
+  return;
 
   /* code for both increments equal to 1 */
   /* clean-up loop */
@@ -56,7 +55,7 @@ L20:
     dy[i__] += *da * dx[i__];
   }
   if (*n < 4) {
-    return 0;
+    return;
   }
 L40:
   mp1 = m + 1;
@@ -67,13 +66,12 @@ L40:
     dy[i__ + 2] += *da * dx[i__ + 2];
     dy[i__ + 3] += *da * dx[i__ + 3];
   }
-  return 0;
+  return;
 }
 
-int lbfgsb_rb_dcopy_(long *n, double *dx, long *incx, double *dy, long *incy)
-{
-  long i__1;
-  static long i__, m, ix, iy, mp1;
+void dcopy_(F77_int* n, double* dx, F77_int* incx, double* dy, F77_int* incy) {
+  F77_int i__1;
+  static F77_int i__, m, ix, iy, mp1;
 
   --dy;
   --dx;
@@ -82,7 +80,7 @@ int lbfgsb_rb_dcopy_(long *n, double *dx, long *incx, double *dy, long *incy)
   /* uses unrolled loops for increments equal to one. */
   /* jack dongarra, linpack, 3/11/78. */
   if (*n <= 0) {
-    return 0;
+    return;
   }
   if (*incx == 1 && *incy == 1) {
     goto L20;
@@ -104,7 +102,7 @@ int lbfgsb_rb_dcopy_(long *n, double *dx, long *incx, double *dy, long *incy)
     ix += *incx;
     iy += *incy;
   }
-  return 0;
+  return;
 
   /* code for both increments equal to 1 */
   /* clean-up loop */
@@ -118,7 +116,7 @@ L20:
     dy[i__] = dx[i__];
   }
   if (*n < 7) {
-    return 0;
+    return;
   }
 L40:
   mp1 = m + 1;
@@ -132,14 +130,13 @@ L40:
     dy[i__ + 5] = dx[i__ + 5];
     dy[i__ + 6] = dx[i__ + 6];
   }
-  return 0;
+  return;
 }
 
-double lbfgsb_rb_ddot_(long *n, double *dx, long *incx, double *dy, long *incy)
-{
-  long i__1;
+double ddot_(F77_int* n, double* dx, F77_int* incx, double* dy, F77_int* incy) {
+  F77_int i__1;
   double ret_val;
-  static long i__, m, ix, iy, mp1;
+  static F77_int i__, m, ix, iy, mp1;
   static double dtemp;
 
   --dy;
@@ -194,18 +191,17 @@ L40:
   mp1 = m + 1;
   i__1 = *n;
   for (i__ = mp1; i__ <= i__1; i__ += 5) {
-    dtemp = dtemp + dx[i__] * dy[i__] + dx[i__ + 1] * dy[i__ + 1]
-      + dx[i__ + 2] * dy[i__ + 2] + dx[i__ + 3] * dy[i__ + 3] + dx[i__ + 4] * dy[i__ + 4];
+    dtemp = dtemp + dx[i__] * dy[i__] + dx[i__ + 1] * dy[i__ + 1] + dx[i__ + 2] * dy[i__ + 2] + dx[i__ + 3] * dy[i__ + 3] +
+            dx[i__ + 4] * dy[i__ + 4];
   }
 L60:
   ret_val = dtemp;
   return ret_val;
 }
 
-int lbfgsb_rb_dscal_(long *n, double *da, double *dx, long *incx)
-{
-  long i__1, i__2;
-  static long i__, m, mp1, nincx;
+void dscal_(F77_int* n, double* da, double* dx, F77_int* incx) {
+  F77_int i__1, i__2;
+  static F77_int i__, m, mp1, nincx;
 
   --dx;
 
@@ -214,7 +210,7 @@ int lbfgsb_rb_dscal_(long *n, double *da, double *dx, long *incx)
   /* jack dongarra, linpack, 3/11/78. */
   /* modified 3/93 to return if incx .le. 0. */
   if (*n <= 0 || *incx <= 0) {
-    return 0;
+    return;
   }
   if (*incx == 1) {
     goto L20;
@@ -227,7 +223,7 @@ int lbfgsb_rb_dscal_(long *n, double *da, double *dx, long *incx)
   for (i__ = 1; i__2 < 0 ? i__ >= i__1 : i__ <= i__1; i__ += i__2) {
     dx[i__] = *da * dx[i__];
   }
-  return 0;
+  return;
 
   /* code for increment equal to 1 */
   /* clean-up loop */
@@ -241,7 +237,7 @@ L20:
     dx[i__] = *da * dx[i__];
   }
   if (*n < 5) {
-    return 0;
+    return;
   }
 L40:
   mp1 = m + 1;
@@ -253,5 +249,5 @@ L40:
     dx[i__ + 3] = *da * dx[i__ + 3];
     dx[i__ + 4] = *da * dx[i__ + 4];
   }
-  return 0;
+  return;
 }

data/ext/lbfgsb/src/blas.h

@@ -1,11 +1,11 @@
 #ifndef LBFGSB_RB_BLAS_H_
 #define LBFGSB_RB_BLAS_H_
 
-#include <math.h>
+#include "common.h"
 
-extern int lbfgsb_rb_daxpy_(long *n, double *da, double *dx, long *incx, double *dy, long *incy);
-extern int lbfgsb_rb_dcopy_(long *n, double *dx, long *incx, double *dy, long *incy);
-extern double lbfgsb_rb_ddot_(long *n, double *dx, long *incx, double *dy, long *incy);
-extern int lbfgsb_rb_dscal_(long *n, double *da, double *dx, long *incx);
+extern void daxpy_(F77_int* n, double* da, double* dx, F77_int* incx, double* dy, F77_int* incy);
+extern void dcopy_(F77_int* n, double* dx, F77_int* incx, double* dy, F77_int* incy);
+extern double ddot_(F77_int* n, double* dx, F77_int* incx, double* dy, F77_int* incy);
+extern void dscal_(F77_int* n, double* da, double* dx, F77_int* incx);
 
 #endif /* LBFGSB_RB_BLAS_H_ */

data/ext/lbfgsb/src/common.h

@@ -0,0 +1,16 @@
+#ifndef LBFGSB_RB_COMMON_H_
+#define LBFGSB_RB_COMMON_H_
+
+#include <stdint.h>
+#include <inttypes.h>
+#include <math.h>
+
+#ifdef USE_INT64
+typedef int64_t F77_int;
+#define PRIdF77INT PRId64
+#else
+typedef int32_t F77_int;
+#define PRIdF77INT PRId32
+#endif
+
+#endif /* LBFGSB_RB_COMMON_H_ */