1 files changed, 33 insertions, 11 deletions

diff --git a/lib/mpi/longlong.h b/lib/mpi/longlong.h
index b87487b40a8..29f98624ef9 100644
--- a/lib/mpi/longlong.h
+++ b/lib/mpi/longlong.h
@@ -1200,18 +1200,40 @@ do { \
         "r" ((USItype)(v)) \
         : "%g1", "%g2" __AND_CLOBBER_CC)
 #define UMUL_TIME 39            /* 39 instructions */
-#endif
-#ifndef udiv_qrnnd
-#ifndef LONGLONG_STANDALONE
+/* It's quite necessary to add this much assembler for the sparc.
+   The default udiv_qrnnd (in C) is more than 10 times slower!  */
 #define udiv_qrnnd(q, r, n1, n0, d) \
-do { USItype __r; \
-        (q) = __udiv_qrnnd(&__r, (n1), (n0), (d)); \
-        (r) = __r; \
-} while (0)
-        extern USItype __udiv_qrnnd();
-#define UDIV_TIME 140
-#endif /* LONGLONG_STANDALONE */
-#endif /* udiv_qrnnd */
+  __asm__ ("! Inlined udiv_qrnnd\n\t"                                   \
+           "mov 32,%%g1\n\t"                                            \
+           "subcc       %1,%2,%%g0\n\t"                                 \
+           "1:  bcs     5f\n\t"                                         \
+           "addxcc %0,%0,%0     ! shift n1n0 and a q-bit in lsb\n\t"    \
+           "sub %1,%2,%1        ! this kills msb of n\n\t"              \
+           "addx        %1,%1,%1        ! so this can't give carry\n\t" \
+           "subcc       %%g1,1,%%g1\n\t"                                \
+           "2:  bne     1b\n\t"                                         \
+           "subcc       %1,%2,%%g0\n\t"                                 \
+           "bcs 3f\n\t"                                                 \
+           "addxcc %0,%0,%0     ! shift n1n0 and a q-bit in lsb\n\t"    \
+           "b           3f\n\t"                                         \
+           "sub %1,%2,%1        ! this kills msb of n\n\t"              \
+           "4:  sub     %1,%2,%1\n\t"                                   \
+           "5:  addxcc  %1,%1,%1\n\t"                                   \
+           "bcc 2b\n\t"                                                 \
+           "subcc       %%g1,1,%%g1\n\t"                                \
+           "! Got carry from n.  Subtract next step to cancel this carry.\n\t" \
+           "bne 4b\n\t"                                                 \
+           "addcc       %0,%0,%0        ! shift n1n0 and a 0-bit in lsb\n\t" \
+           "sub %1,%2,%1\n\t"                                           \
+           "3:  xnor    %0,0,%0\n\t"                                    \
+           "! End of inline udiv_qrnnd\n"                               \
+           : "=&r" ((USItype)(q)),                                      \
+             "=&r" ((USItype)(r))                                       \
+           : "r" ((USItype)(d)),                                        \
+             "1" ((USItype)(n1)),                                       \
+             "0" ((USItype)(n0)) : "%g1", "cc")
+#define UDIV_TIME (3+7*32)      /* 7 instructions/iteration. 32 iterations.  */
+#endif
 #endif /* __sparc__ */
 
 /***************************************