Improving Performance by Pointer Disambiguation

Two pointers are aliased if both point to the same memory location. Storing to memory using a pointer that might be aliased may prevent some optimizations. For example, it may create a dependency between loop iterations that would make vectorization unsafe. Aliasing is not the only source of potential dependencies. In fact, Multiply.c does have other dependencies. In this case however, removal of the dependency created by aliasing allows the compiler to resolve the other loop dependency.

Sometimes, the compiler can generate both a vectorized and a non-vectorized version of a loop and test for aliasing at runtime to select the appropriate code path. If you know that pointers do not alias and inform the compiler, it can avoid the runtime check and generate a single vectorized code path. In Multiply.c, the compiler generates runtime checks to determine whether or not the pointer b in function matvec(FTYPE a[][COLWIDTH], FTYPE b[], FTYPE x[]) is aliased to either a or x . If Multiply.c is compiled with the NOALIAS macro, the restrict qualifier of the argument b informs the compiler that the pointer does not alias with any other pointer, and in particular that the array b does not overlap with a or x.

Remove the -D NOFUNCCALL to restore the call to matvec(), then add the -D NOALIAS option to the command line.

icc -std=c99 -qopt-report=2 -qopt-report-phase=vec -D NOALIAS Multiply.c Driver.c -o MatVector

This conditional compilation replaces the loop in the main program with a function call. Execute MatVector and record the execution time reported in the output. Multiply.optrpt now shows:

LOOP BEGIN at Multiply.c(45,2)
   remark #15542: loop was not vectorized: inner loop was already vectorized

   LOOP BEGIN at Multiply.c(55,3)
   <Peeled>
   LOOP END

   LOOP BEGIN at Multiply.c(55,3)
      remark #15399: vectorization support: unroll factor set to 4
      remark #15300: LOOP WAS VECTORIZED
   LOOP END

   LOOP BEGIN at Multiply.c(55,3)
      remark #25460: No loop optimizations reported
   LOOP END

   LOOP BEGIN at Multiply.c(55,3)
   <Remainder>
      remark #15301: REMAINDER LOOP WAS VECTORIZED
   LOOP END

   LOOP BEGIN at Multiply.c(55,3)
   <Remainder>
   LOOP END
LOOP END

Now that the compiler has been told that the arrays do not overlap, it uses idiom-recognition to resolve the loop dependency and proceeds to vectorize the loop.