Fixing clusterfuzz issue

src/opts/SkXfermode_opts_SSE2.cpp

Index: src/opts/SkXfermode_opts_SSE2.cpp
diff --git a/src/opts/SkXfermode_opts_SSE2.cpp b/src/opts/SkXfermode_opts_SSE2.cpp
index ec76ab3cfbb6533b4295f5a50b0867e5a700bfd5..94f9a4aea3be1591323f603b3d3a8d88fe7ea79a 100644
--- a/src/opts/SkXfermode_opts_SSE2.cpp
+++ b/src/opts/SkXfermode_opts_SSE2.cpp
@@ -644,6 +644,7 @@ extern SkXfermodeProcSIMD gSSE2XfermodeProcs[];
 SkSSE2ProcCoeffXfermode::SkSSE2ProcCoeffXfermode(SkReadBuffer& buffer)
     : INHERITED(buffer) {
     fProcSIMD = reinterpret_cast<void*>(gSSE2XfermodeProcs[this->getMode()]);
+    buffer.validate(fProcSIMD != NULL);

[Stephen White, 2014/06/02 13:13:35]

Not new to this patch, but I wonder why we have such processor-specific
subclasses at all. Shouldn't the SKP format be processor-indepdendent? I.e.,
shouldn't we serialize generic implementations, and have the processing code
initialize callbacks for the platform it's running on?

 }
 
 void SkSSE2ProcCoeffXfermode::xfer32(SkPMColor dst[], const SkPMColor src[],