A “shmoo plot” refers to a graphical display of test circuit patterns
as two inputs (such as voltage and clock rate) vary. When writing code
to identify the optimal blocking parameters for various kernels, it is
useful to do similar tests by varying inputs such as the threadblock
size and loop unroll factor. Listing A-2 shows the chShmooRange() class,
which encapsulates a parameter range, and the chShmooIterator() class
which enables for loops to easily iterate over a given range.
class chShmooRange {public: chShmooRange( ) { } void Initialize( int value ); bool Initialize( int min, int max, int step ); bool isStatic() const { return m_min==m_max; } friend class chShmooIterator; int min() const { return m_min; } int max() const { return m_max; } private: bool m_initialized; int m_min, m_max, m_step;}; class chShmooIterator{public: chShmooIterator( const chShmooRange& range ); int operator *() const { return m_i; } operator bool() const { return m_i <= m_max; } void operator++(int) { m_i += m_step; };private: int m_i; int m_max; int m_step;};chShmooRange and
chShmooIterator classes. (source on GitHub)The command line parser also includes a specialization that creates a
chShmooRange() based on command-line parameters: prepend “min,” “max,” and
“step” onto the keyword, and the corresponding range will be passed
back. If any of the three are missing, the function returns false. The
concurrencyKernelKernel sample (in the
concurrency/ subdirectory), for example, takes measurements
over ranges of stream count and clock cycle count. The code to extract
these values from the command line is as follows:
chShmooRange streamsRange;
{
const int numStreams = 8;
if ( ! chCommandLineGet( &streamsRange, "Streams", argc, argv ) ) {
streamsRange.Initialize( numStreams );
}
}
chShmooRange cyclesRange;
{
const int minCycles = 8;
const int maxCycles = 512;
const int stepCycles = 8;
cyclesRange.Initialize( minCycles, maxCycles, stepCycles );
chCommandLineGet( &cyclesRange, "Cycles", argc, argv );
}
And users can specify the parameters to the application as follows:
concurrencyKernelKernel --minStreams 2 --maxStreams 16 --stepStreams 2