math/mpx-mul4-*-sse2.S (mpxmont_redc4): Fix end-of-outer-loop commentary.

  * The carry loop is wrong if the destination is an exact multiple of
    four limbs.  Fortunately, it isn't.

  * The initial pass feeds into the main loop unconditionally, unlike
    `mpxmont_mul4_...' (from which I think the commentary was
    uncritically copied), so being at the end of it doesn't tell you
    anything about whether to start another.  And, indeed, we do indeed
    check the loop-end condition.

diff --git a/math/mpx-mul4-amd64-sse2.S b/math/mpx-mul4-amd64-sse2.S
index a37aba6..17c4f1a 100644 (file)
--- a/math/mpx-mul4-amd64-sse2.S
+++ b/math/mpx-mul4-amd64-sse2.S
@@ -1260,7 +1260,9 @@ FUNC(mpxmont_redc4_amd64_sse2)
        cmp     rdi, DVL4
        jb      0b
 
-       // Deal with the tail end.
+       // Deal with the tail end.  Note that the actual destination length
+       // won't be an exacty number of blocks of four, so it's safe to just
+       // drop through here.
 7:     add     [rdi], C
        mov     C, 0
        add     rdi, 4
@@ -1268,9 +1270,7 @@ FUNC(mpxmont_redc4_amd64_sse2)
        cmp     rdi, DVL
        jb      7b
 
-       // All done for this iteration.  Start the next.  (This must have at
-       // least one follow-on iteration, or we'd not have started this outer
-       // loop.)
+       // All done for this iteration.  Start the next.
 8:     mov     rdi, DV                 // -> Z = dv[i]
        mov     rbx, NV                 // -> X = nv[0]
        cmp     rdi, DVLO               // all done yet?

diff --git a/math/mpx-mul4-x86-sse2.S b/math/mpx-mul4-x86-sse2.S
index 9e1d478..7e7173f 100644 (file)
--- a/math/mpx-mul4-x86-sse2.S
+++ b/math/mpx-mul4-x86-sse2.S
@@ -1062,7 +1062,9 @@ FUNC(mpxmont_redc4_x86_sse2)
        cmp     edi, esi
        jb      0b
 
-       // Deal with the tail end.
+       // Deal with the tail end.  Note that the actual destination length
+       // won't be an exact number of blocks of four, so it's safe to just
+       // drop through here.
 7:     add     [edi], eax
        mov     eax, 0
        add     edi, 4
@@ -1070,9 +1072,7 @@ FUNC(mpxmont_redc4_x86_sse2)
        cmp     edi, edx
        jb      7b
 
-       // All done for this iteration.  Start the next.  (This must have at
-       // least one follow-on iteration, or we'd not have started this outer
-       // loop.)
+       // All done for this iteration.  Start the next.
 8:     mov     edi, [SP + 0]           // -> dv[i - 1]
        mov     ebx, [BP + 28]          // -> X = nv[0]
        lea     edx, [SP + 44]          // -> space for Y