conversion: truncates input values when printing errors for conversions to floats

[karolherbst]

OpenCL-CTS/test_conformance/conversions/basic_test_conversions.cpp

Line 341 in 5673883

<< s << "." << s << "x vs 0x%" << s << "." << s << "x\n";

This only prints at most 32 bits of the input value.

replacing x with llx fixes it, but not sure if there is a better way?

[bashbaug]

replacing x with llx fixes it, but not sure if there is a better way?

Probably using PRIx64 from inttypes.h?

We've used this in quite a few other places in the CTS. This one is extra tricky because we're building the format string programmatically and the type we're trying to print is templated, but I think it'd be possible to get it working.