[Math] Fix Nan from powf() domain error.

This change prevents calling powf(x,y) with negative x.

The SIMD versions using ssePower() already seem to be resistent to
negative pixel values, but the scalar version was not.

This would cause a SIGFPE on apps that have floating point exceptions
enabled.

FIXES: #2066

Signed-off-by: Bram Stolk <[email protected]>

src/OpenColorIO/ops/gamma/GammaOpCPU.cpp

@@ -538,7 +538,10 @@ void GammaMoncurveOpCPUFwd::apply(const void * inImg, void * outImg, long numPix
 
     for(long idx=0; idx<numPixels; ++idx)
     {
-        const float pixel[4] = { in[0], in[1], in[2], in[3] };
+        const float pixel[4] = { std::max(0.0f, in[0]),
+                                 std::max(0.0f, in[1]),
+                                 std::max(0.0f, in[2]),
+                                 std::max(0.0f, in[3]) };
 
         const float data[4] = { std::pow(pixel[0] * red[0] + red[1], red[2]),
                                 std::pow(pixel[1] * grn[0] + grn[1], grn[2]),
@@ -627,7 +630,10 @@ void GammaMoncurveOpCPURev::apply(const void * inImg, void * outImg, long numPix
 
     for(long idx=0; idx<numPixels; ++idx)
     {
-        const float pixel[4] = { in[0], in[1], in[2], in[3] };
+        const float pixel[4] = { std::max(0.0f, in[0]),
+                                 std::max(0.0f, in[1]),
+                                 std::max(0.0f, in[2]),
+                                 std::max(0.0f, in[3]) };
 
         const float data[4] = { std::pow(pixel[0], red[0]) * red[1] - red[2],
                                 std::pow(pixel[1], grn[0]) * grn[1] - grn[2],