math/mpx-mul4-amd64-sse2.S: Save a spill by better register allocation.

The Windows code doesn't need to spill r12, because we don't need the
`mi' register after we've loaded and expanded the Montgomery factor.
This doesn't save any stack space because we need 16-byte alignment, but
it does avoid saving and restoring the register.

diff --git a/math/mpx-mul4-amd64-sse2.S b/math/mpx-mul4-amd64-sse2.S
index 17c4f1a..da3e6d6 100644 (file)
--- a/math/mpx-mul4-amd64-sse2.S
+++ b/math/mpx-mul4-amd64-sse2.S
@@ -1155,7 +1155,7 @@ FUNC(mpxmont_redc4_amd64_sse2)
        // outer loop dv        r10     rcx
        // outer loop dv limit  r11     r11
        // nv base              rdx     r8
-       // nv limit             r9      r12*
+       // nv limit             r9      r10*
        // n                    rcx     r9
        // c                    rcx     r9
 
@@ -1183,14 +1183,13 @@ FUNC(mpxmont_redc4_amd64_sse2)
 #  define DV rcx
 #  define DVLO r11
 #  define NV r8
-#  define NVL r12
+#  define NVL r10
 #  define N r9
 #  define C r9d
 
        pushreg rbx
        pushreg rdi
-       pushreg r12
-       stalloc 160
+       stalloc 168
 
        savexmm xmm6,    0
        savexmm xmm7,   16
@@ -1300,8 +1299,7 @@ FUNC(mpxmont_redc4_amd64_sse2)
        rstrxmm xmm14, 128
        rstrxmm xmm15, 144
 
-       stfree  160
-       popreg  r12
+       stfree  168
        popreg  rdi
        popreg  rbx
 #endif