Disable SSE2 optimization with --disable-sse2 configure option.

Now SSE2 optimization for gcc is enabled by default.
rainarch · Aug 9, 2011 · bbd46a6 · bbd46a6
1 parent 62e201c
commit bbd46a6
Show file tree

Hide file tree

Showing 2 changed files with 27 additions and 13 deletions.
diff --git a/configure.in b/configure.in
@@ -92,7 +92,7 @@ dnl Checks for profiling mode
 dnl ------------------------------------------------------------------
 AC_ARG_ENABLE(
   profile,
-  [AS_HELP_STRING([--enable-profile],[Turn on profiling])]
+  [AS_HELP_STRING([--enable-profile],[turn on profiling])]
 )
 
 if test "x$enable_profile" = "xyes"; then
@@ -103,14 +103,13 @@ fi
 dnl ------------------------------------------------------------------
 dnl Checks for SSE2 build
 dnl ------------------------------------------------------------------
-AC_ARG_ENABLE(
-    sse2,
-    [AS_HELP_STRING(
-        [--enable-sse2],
-        [enable SSE2 optimization routines]
-        )],
-    [CFLAGS="-mfpmath=sse -msse2 -DUSE_SSE ${CFLAGS}"]
-)
+AC_ARG_ENABLE([sse2],
+    AS_HELP_STRING([--disable-sse2], [disable SSE2 optimization routines])
+    )
+
+AS_IF([test "x$enable_sse2" != "xno"], [
+    CFLAGS="-mfpmath=sse -msse2 -DUSE_SSE ${CFLAGS}"
+])
 
 
 dnl ------------------------------------------------------------------

diff --git a/lib/crf/src/vecmath.h b/lib/crf/src/vecmath.h
@@ -36,7 +36,9 @@
 #include <math.h>
 #include <memory.h>
 
+#ifdef  USE_SSE
 #include <emmintrin.h>
+#endif/*USE_SSE*/
 
 #ifdef _MSC_VER
 #include <malloc.h>
@@ -169,6 +171,8 @@ inline static floatval_t vecsumlog(floatval_t* x, const int n)
     return s;
 }
 
+#ifdef  USE_SSE
+
 inline static void vecexp(double *values, const int n)
 {
     int i;
@@ -202,8 +206,8 @@ inline static void vecexp(double *values, const int n)
         /* Load four double values. */
         xmm0 = _mm_load_pd(maxlog);
         xmm1 = _mm_load_pd(minlog);
-		x1 = _mm_load_pd(values+i);
-		x2 = _mm_load_pd(values+i+2);
+        x1 = _mm_load_pd(values+i);
+        x2 = _mm_load_pd(values+i+2);
         x1 = _mm_min_pd(x1, xmm0);
         x2 = _mm_min_pd(x2, xmm0);
         x1 = _mm_max_pd(x1, xmm1);
@@ -317,10 +321,21 @@ inline static void vecexp(double *values, const int n)
         a2 = _mm_mul_pd(a2, p2);
 
         /* Store the results. */
-		_mm_store_pd(values+i, a1);
-		_mm_store_pd(values+i+2, a2);
+        _mm_store_pd(values+i, a1);
+        _mm_store_pd(values+i+2, a2);
+    }
+}
+
+#else
+
+inline static void vecexp(double *values, const int n)
+{
+    int i;
+    for (i = 0;i < n;++i) {
+        values[i] = exp(values[i]);
     }
 }
 
+#endif /*USE_SSE*/
 
 #endif/*__VECMATH_H__*/