math/mpx-mul4-amd64-sse2.S: Always collect iteration count as 32 bits.

Some ABIs, at least, don't guarantee to zero-extend arguments, and we
use the counter as an address offset.

diff --git a/math/mpx-mul4-amd64-sse2.S b/math/mpx-mul4-amd64-sse2.S
index 098de45..9146a63 100644 (file)
--- a/math/mpx-mul4-amd64-sse2.S
+++ b/math/mpx-mul4-amd64-sse2.S
@@ -1365,7 +1365,7 @@ ENDFUNC
        mov     rbx, r8
        movdqu  xmm8, [r9]
        movdqu  xmm10, [rax]
-       mov     r8, STKARG(1)
+       mov     r8d, STKARG(1)
        mov     r9, STKARG(2)
        mov     r10, rdx
        mov     r11, rcx
@@ -1374,7 +1374,7 @@ ENDFUNC
   .ifeqs "\mode", "mont"
        mov     rbx, rcx
        movdqu  xmm8, [r8]
-       mov     r8, r9
+       mov     r8d, r9d
        mov     r9, STKARG(0)
        mov     r10, rdx
        mov     rcx, rsi
@@ -1402,16 +1402,16 @@ ENDFUNC
        mov     rbx, r9
        movdqu  xmm8, [r10]
        movdqu  xmm10, [r11]
-       mov     r8, STKARG(2)
-       mov     r9, STKARG(3)
        mov     r11, r8
+       mov     r8d, STKARG(2)
+       mov     r9, STKARG(3)
   .endif
   .ifeqs "\mode", "smul"
        mov     rdi, rcx
        mov     rcx, rdx
        mov     rbx, r8
        movdqu  xmm10, [r9]
-       mov     r8, STKARG(0)
+       mov     r8d, STKARG(0)
        mov     r9, STKARG(1)
   .endif
   .ifeqs "\mode", "mmul"
@@ -1422,10 +1422,10 @@ ENDFUNC
        mov     rbx, STKARG(0)
        movdqu  xmm8, [r10]
        movdqu  xmm10, [r11]
-       mov     r8, STKARG(3)
-       mov     r9, STKARG(4)
        mov     r10, r8
        mov     r11, r9
+       mov     r8d, STKARG(3)
+       mov     r9, STKARG(4)
   .endif
   .ifeqs "\mode", "mont"
        mov     r10, STKARG(0)
@@ -1433,9 +1433,9 @@ ENDFUNC
        mov     rcx, rdx
        mov     rbx, r9
        movdqu  xmm8, [r10]
-       mov     r8, STKARG(1)
-       mov     r9, STKARG(2)
        mov     r10, r8
+       mov     r8d, STKARG(1)
+       mov     r9, STKARG(2)
   .endif
 #endif