-#if 0
- /* This lot is an unrolled loop to copy b->top
- * BN_ULONGs from B to A
- */
-/*
- * I have nothing against unrolling but it's usually done for
- * several reasons, namely:
- * - minimize percentage of decision making code, i.e. branches;
- * - avoid cache trashing;
- * - make it possible to schedule loads earlier;
- * Now let's examine the code below. The cornerstone of C is
- * "programmer is always right" and that's what we love it for:-)
- * For this very reason C compilers have to be paranoid when it
- * comes to data aliasing and assume the worst. Yeah, but what
- * does it mean in real life? This means that loop body below will
- * be compiled to sequence of loads immediately followed by stores
- * as compiler assumes the worst, something in A==B+1 style. As a
- * result CPU pipeline is going to starve for incoming data. Secondly
- * if A and B happen to share same cache line such code is going to
- * cause severe cache trashing. Both factors have severe impact on
- * performance of modern CPUs and this is the reason why this
- * particulare piece of code is #ifdefed away and replaced by more
- * "friendly" version found in #else section below. This comment
- * also applies to BN_copy function.
- *
- * <appro@fy.chalmers.se>
- */
- for (i=b->top&(~7); i>0; i-=8)
- {
- A[0]=B[0]; A[1]=B[1]; A[2]=B[2]; A[3]=B[3];
- A[4]=B[4]; A[5]=B[5]; A[6]=B[6]; A[7]=B[7];
- A+=8;
- B+=8;
- }
- switch (b->top&7)