Today we will continue our discussion on the WRL system to generate and analyze trace data. The trace code itself was written in proprietary assembly language using registers to avoid memory references. 8 of these registers are used for trace generation. The kernel uses a separate register set from user process  Only the trace index and the trace pointer values needed to be copied from user to kernel mode. In the switch from kernel to user they are always calculated. All processes have the shared trace buffer  mapped to the same virtual address space. The linker inserts trace branches as part of the automatic generation of address  references. In addition, it is possible to add custom trace routines by user or kernel. Only the kernel does it for this study. The addresses inserted in the trace buffer by the kernel are physical addresses, whereas those inserted by user addresses are virtual addresses. It is essential for the analysis code to be able to tell one from the other.Similarly, it is important to tell sequences of virtual addresses as belonging to a particular process when more than one is being traced. This is done by making a change mode entry in the trace buffer on every transition from kernel to user or back.The entry indicates the direction of the switch and the process id.
Tracing is done only when a trace flag is set and stored in a trace register.All writes to the trace buffer check for this value. When a write occurs, at the first interrupt following a write to a location in /dev/kmem, tracing is turned on. The set of user programs that are traced are determined by the ones that are specially linked. The kernel is traced only when it is trace-linked. When tracing is off, 4 to 6 additional instructions are executed at every trace point. The user code does not incur any cost but the kernel does. Even when the kernel is not linked for tracing, it is slower because it has to execute instructions to keep the trace buffer consistent between every switch
The trace buffer entries are formatted such that they can be divided into two types based on whether they fill one or two 32 bit words in the buffer. Data reference entries are one word long. Data load/store entries take up a single 32-bit word.  The Load/Store is differentiated by the first two high order bits of the word. The remaining bits are used for address which is guaranteed  to be less than 1GB. Any entry with a non zero high order bits is a data entry. On the other hand, an instruction entry represents the beginning of a basic block and has the first word denoting the type and count. Count is the number of instructions in the basic block starting at the address in the second word.
#codingexercise
Int GetDistinctRangeMode(int [] A)
{
if (A == null) return 0;
Return A.DistinctRangeMode ();
}
Int GetDistinctRangeMedian(int [] A)

{

if (A == null) return 0;

Return A.DistinctRangeMedian ();

}

decimal GetDistinctRangeStdDev(int [] A)


{


if (A == null) return 0;


Return A.DistinctRangeStdDev ();


}