diff --git a/MagickCore/enhance.c b/MagickCore/enhance.c
index cb0104a..78ee0e4 100644
--- a/MagickCore/enhance.c
+++ b/MagickCore/enhance.c
@@ -344,8 +344,7 @@
clut_view=AcquireVirtualCacheView(clut_image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) \
- if (MaxMap > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(MaxMap,256)
#endif
for (i=0; i <= (ssize_t) MaxMap; i++)
{
@@ -358,8 +357,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -709,8 +707,7 @@
image->filename);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) \
- if (MaxMap > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(MaxMap,256)
#endif
for (i=0; i <= (ssize_t) MaxMap; i++)
{
@@ -731,8 +728,7 @@
*/
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (image->colors > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(image->colors,256)
#endif
for (i=0; i < (ssize_t) image->colors; i++)
{
@@ -760,8 +756,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -917,8 +912,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -1105,8 +1099,7 @@
number_channels=GetPixelChannels(image);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (MaxMap > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(MaxMap,256)
#endif
for (i=0; i < (ssize_t) number_channels; i++)
{
@@ -1144,8 +1137,7 @@
number_channels=GetPixelChannels(image);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (MaxMap > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(MaxMap,256)
#endif
for (i=0; i < (ssize_t) number_channels; i++)
{
@@ -1177,8 +1169,7 @@
*/
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (image->colors > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(image->colors,256)
#endif
for (j=0; j < (ssize_t) image->colors; j++)
{
@@ -1220,8 +1211,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -1374,8 +1364,7 @@
enhance_view=AcquireAuthenticCacheView(enhance_image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -1607,8 +1596,7 @@
number_channels=GetPixelChannels(image);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (MaxMap > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(MaxMap,256)
#endif
for (i=0; i < (ssize_t) number_channels; i++)
{
@@ -1630,8 +1618,7 @@
number_channels=GetPixelChannels(image);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (MaxMap > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(MaxMap,256)
#endif
for (i=0; i < (ssize_t) number_channels; i++)
{
@@ -1661,8 +1648,7 @@
*/
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (image->colors > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(image->colors,256)
#endif
for (j=0; j < (ssize_t) image->colors; j++)
{
@@ -1707,8 +1693,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -1847,8 +1832,7 @@
if (gamma != 0.0)
#if defined(MAGICKCORE_OPENMP_SUPPORT) && (MaxMap > 256)
#pragma omp parallel for \
- if (MaxMap > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(MaxMap,256)
#endif
for (i=0; i <= (ssize_t) MaxMap; i++)
gamma_map[i]=ScaleMapToQuantum((MagickRealType) (MaxMap*pow((double) i/
@@ -1860,8 +1844,7 @@
*/
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(progress,status) \
- if (image->colors > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(image->colors,256)
#endif
for (i=0; i < (ssize_t) image->colors; i++)
{
@@ -1887,8 +1870,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -2058,8 +2040,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -2240,8 +2221,7 @@
if (image->storage_class == PseudoClass)
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (image->colors > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(image->colors,256)
#endif
for (i=0; i < (ssize_t) image->colors; i++)
{
@@ -2269,8 +2249,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -2410,8 +2389,7 @@
if (image->storage_class == PseudoClass)
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (image->colors > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(image->colors,256)
#endif
for (i=0; i < (ssize_t) image->colors; i++)
{
@@ -2439,8 +2417,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -2925,8 +2902,7 @@
*/
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (image->colors > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(image->colors,256)
#endif
for (i=0; i < (ssize_t) image->colors; i++)
switch (colorspace)
@@ -2963,8 +2939,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -3097,8 +3072,7 @@
*/
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(progress,status) \
- if (image->colors > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(image->colors,256)
#endif
for (i=0; i < (ssize_t) image->colors; i++)
{
@@ -3127,8 +3101,7 @@
{
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -3202,8 +3175,7 @@
*/
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
@@ -3383,8 +3355,7 @@
(void) ResetMagickMemory(sigmoidal_map,0,(MaxMap+1)*sizeof(*sigmoidal_map));
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(progress,status) \
- if (MaxMap > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(MaxMap,256)
#endif
for (i=0; i <= (ssize_t) MaxMap; i++)
{
@@ -3448,8 +3419,7 @@
*/
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if (image->colors > 256) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentUno(image->colors,256)
#endif
for (i=0; i < (ssize_t) image->colors; i++)
{
@@ -3475,8 +3445,7 @@
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
- if ((image->rows*image->columns) > 8192) \
- num_threads(GetMagickResourceLimit(ThreadResource))
+ IsConcurrentDos(image->columns,image->rows,64)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{