utils/gcm-ref (present_gf_pmull): Round width up to a multiple of 64 bits.

Otherwise the later loop, which pulls off 64-bit chunks, gets badly
confused.

Now `gcm-ref' can actually calculate all of the things properly.

	for p in pclmul vmullp64 pmull; do
	  while read u v; do utils/gcm-ref $p $u $v || break 2; done <<EOF
	cde4bef260d7bcda 163547d348b75511
	cde4bef260d7bcda163547d3 48b7551195e77022907dd1df
	cde4bef260d7bcda163547d348b75511 95e77022907dd1dff7dac5c9941d26d0
	cde4bef260d7bcda163547d348b7551195e77022907dd1df f7dac5c9941d26d0c6eb14ad568f86edd1dc9268eeee5332
	cde4bef260d7bcda163547d348b7551195e77022907dd1dff7dac5c9941d26d0 c6eb14ad568f86edd1dc9268eeee533285a6ed810c9b689daaa9060d2d4b6003
	EOF
	done

I wonder what this means about the changes coming up...

diff --git a/utils/gcm-ref b/utils/gcm-ref
index 03672412568658eed60e960767624b2ca0c80ac3..bec00d7532c9abe45921051c78073a35ae4c736b 100755 (executable)
--- a/utils/gcm-ref
+++ b/utils/gcm-ref
@@ -258,6 +258,7 @@ def present_gf_pmull(tag, wd, x, w, n, what):
   if tag == TAG_PRODPIECE or tag == TAG_REDCFULL or tag == TAG_SHIFTED:
     return
   elif tag == TAG_INPUT_V or tag == TAG_KPIECE_V:
   if tag == TAG_PRODPIECE or tag == TAG_REDCFULL or tag == TAG_SHIFTED:
     return
   elif tag == TAG_INPUT_V or tag == TAG_KPIECE_V:

+    w = (w + 63)&~63

     bx = C.ReadBuffer(x.storeb(w/8))
     by = C.WriteBuffer()
     while bx.left: chunk = bx.get(8); by.put(chunk).put(chunk)
     bx = C.ReadBuffer(x.storeb(w/8))
     by = C.WriteBuffer()
     while bx.left: chunk = bx.get(8); by.put(chunk).put(chunk)