Re: Poly1305 vs. UMAC vs. new MAC1071

I wrote earlier:
Also, I just noticed "Message Authentication on 64-bit Architectures" by Ted Krovetz (http://eprint.iacr.org/2006/037) which claims 1.1 cycles per byte for UHASH8-128 (120-bit security) on an Athlon64.

I'd like to give an update on this algorithm for those who followed the previous discussion. The algorithm has since been modified a bit and renamed VMAC, and is now described by a draft specification at http://www.fastcrypto.org/vmac/. Performance on 32-bit platforms has also been improved substantially since the ePrint and SAC2006 papers, and I've measured a speed of 3.7 cycles per byte for long messages for VMAC-128 on P4-Northwood with my own implementation (which will be in the next version of Crypto++). (The higher cpb figures in the draft specification are based on the reference implementation.) It should be interesting to do a side-by-side comparison of its performance with Bernstein's upcoming MAC1071.

.

Relevant Pages

  • Re: my assembler is better than your assembler
    ... Therefore we can be sure that "Randy is not the Author" of it. ... don't look at an existing algorithm and then turn ... you're worried about counting cycles when the algorithm is ... not all processors have zero-time nops. ...
    (alt.lang.asm)
  • Re: my assembler is better than your assembler
    ... don't look at an existing algorithm and then turn ... My 'incredible slow' loop does maximal 31 iterations (2..3 cycles) ... Now you might argue that "this routine is so ... Even with that stack frame my code was considerably *faster* ...
    (alt.lang.asm)
  • Re: Array Programming Questions
    ... >>one not) to describe the transformation. ... Let us say the array is: ... > I thought of very similar algorithm but I think there is a flaw. ... give have distinct cycles in them. ...
    (comp.programming)
  • Re: my assembler is better than your assembler
    ... even the loop performs with very few cycles ... waste of time -- first seek a better algorithm. ... My code parts start always aligned (and I know most opcode by heart), ...
    (alt.lang.asm)
  • Re: (graph theory: path problem)how to tell how many path are there from s to t in a graph?
    ... >>Just modify Dijkstra's algorithm. ... >>all the incoming nodes. ... of Dijkstra's algorithm which talks about a "relaxation" step. ... (vertex and edge repeats, and therefore cycles, allowed) ...
    (comp.theory)