chore: update vendored liblbfgs from https://github.com/chokkan/liblbfgs

Author: szhorvat

src/arithmetic_ansi.h

@@ -23,7 +23,7 @@
  * THE SOFTWARE.
  */
 
-/* $Id: arithmetic_ansi.h 65 2010-01-29 12:19:16Z naoaki $ */
+/* $Id$ */
 
 #include <stdlib.h>
 #include <memory.h>
@@ -51,7 +51,7 @@ inline static void vecfree(void *memblock)
 inline static void vecset(lbfgsfloatval_t *x, const lbfgsfloatval_t c, const int n)
 {
     int i;
-
+    
     for (i = 0;i < n;++i) {
         x[i] = c;
     }

src/arithmetic_sse_double.h

@@ -23,14 +23,12 @@
  * THE SOFTWARE.
  */
 
-/* $Id: arithmetic_sse_double.h 65 2010-01-29 12:19:16Z naoaki $ */
+/* $Id$ */
 
 #include <stdlib.h>
-
-#if !defined(__APPLE__)
+#ifndef __APPLE__
 #include <malloc.h>
 #endif
-
 #include <memory.h>
 
 #if     1400 <= _MSC_VER
@@ -43,13 +41,15 @@
 
 inline static void* vecalloc(size_t size)
 {
-#ifdef	_WIN32
+#if     defined(_WIN32)
     void *memblock = _aligned_malloc(size, 16);
-#elif defined(__APPLE__)
-	/* Memory on Mac OS X is already aligned to 16 bytes */
-	void *memblock = malloc(size);
+#elif   defined(__APPLE__)  /* OS X always aligns on 16-byte boundaries */
+    void *memblock = malloc(size);
 #else
-    void *memblock = memalign(16, size);
+    void *memblock = NULL, *p = NULL;
+    if (posix_memalign(&p, 16, size) == 0) {
+        memblock = p;
+    }
 #endif
     if (memblock != NULL) {
         memset(memblock, 0, size);
@@ -59,7 +59,7 @@ inline static void* vecalloc(size_t size)
 
 inline static void vecfree(void *memblock)
 {
-#ifdef	_WIN32
+#ifdef	_MSC_VER
     _aligned_free(memblock);
 #else
     free(memblock);
@@ -199,7 +199,7 @@ inline static void vecfree(void *memblock)
 
 
 
-#if     3 <= __SSE__
+#if     3 <= __SSE__ || defined(__SSE3__)
 /*
     Horizontal add with haddps SSE3 instruction. The work register (rw)
     is unused.

src/arithmetic_sse_float.h

@@ -23,14 +23,12 @@
  * THE SOFTWARE.
  */
 
-/* $Id: arithmetic_sse_float.h 65 2010-01-29 12:19:16Z naoaki $ */
+/* $Id$ */
 
 #include <stdlib.h>
-
-#if !defined(__APPLE__)
+#ifndef __APPLE__
 #include <malloc.h>
 #endif
-
 #include <memory.h>
 
 #if     1400 <= _MSC_VER
@@ -49,7 +47,16 @@
 
 inline static void* vecalloc(size_t size)
 {
+#if     defined(_MSC_VER)
     void *memblock = _aligned_malloc(size, 16);
+#elif   defined(__APPLE__)  /* OS X always aligns on 16-byte boundaries */
+    void *memblock = malloc(size);
+#else
+    void *memblock = NULL, *p = NULL;
+    if (posix_memalign(&p, 16, size) == 0) {
+        memblock = p;
+    }
+#endif
     if (memblock != NULL) {
         memset(memblock, 0, size);
     }
@@ -58,7 +65,11 @@ inline static void* vecalloc(size_t size)
 
 inline static void vecfree(void *memblock)
 {
+#ifdef	_MSC_VER
     _aligned_free(memblock);
+#else
+    free(memblock);
+#endif
 }
 
 #define vecset(x, c, n) \
@@ -189,7 +200,7 @@ inline static void vecfree(void *memblock)
 
 
 
-#if     3 <= __SSE__
+#if     3 <= __SSE__ || defined(__SSE3__)
 /*
     Horizontal add with haddps SSE3 instruction. The work register (rw)
     is unused.