[Qemu-devel] [PATCH 1/3] host-utils: Use __int128_t for mul[us]64

[Richard Henderson]

Replace some x86_64 specific inline assembly with something that
all 64-bit hosts ought to optimize well.  At worst this becomes
a call to the gcc __multi3 routine, which is no worse than our
implementation in util/host-utils.c.

With gcc 4.7, we get identical code generation for x86_64.  We
now get native multiplication on ia64 and s390x hosts.  With minor
improvements to gcc we can get it for ppc64 as well.

Signed-off-by: Richard Henderson <address@hidden>
---
 configure                 | 20 ++++++++++++++++++++
 include/qemu/host-utils.h | 17 ++++++++---------
 util/host-utils.c         |  4 ++--
 3 files changed, 30 insertions(+), 11 deletions(-)

diff --git a/configure b/configure
index b7635e4..a1ec120 100755
--- a/configure
+++ b/configure
@@ -3150,6 +3150,22 @@ if compile_prog "" "" ; then
     cpuid_h=yes
 fi
 
+########################################
+# check if __[u]int128_t is usable.
+
+int128=no
+cat > $TMPC << EOF
+__int128_t a;
+__uint128_t b;
+int main (void) {
+  a = a + b;
+  b = a * b;
+  return 0;
+}
+EOF
+if compile_prog "" "" ; then
+    int128=yes
+fi
 
 ##########################################
 # End of CC checks
@@ -3692,6 +3708,10 @@ if test "$cpuid_h" = "yes" ; then
   echo "CONFIG_CPUID_H=y" >> $config_host_mak
 fi
 
+if test "$int128" = "yes" ; then
+  echo "CONFIG_INT128=y" >> $config_host_mak
+fi
+
 if test "$glusterfs" = "yes" ; then
   echo "CONFIG_GLUSTERFS=y" >> $config_host_mak
 fi
diff --git a/include/qemu/host-utils.h b/include/qemu/host-utils.h
index 81c9a75..dbe96ca 100644
--- a/include/qemu/host-utils.h
+++ b/include/qemu/host-utils.h
@@ -27,22 +27,21 @@
 
 #include "qemu/compiler.h"   /* QEMU_GNUC_PREREQ */
 
-#if defined(__x86_64__)
-#define __HAVE_FAST_MULU64__
+#ifdef CONFIG_INT128
 static inline void mulu64(uint64_t *plow, uint64_t *phigh,
                           uint64_t a, uint64_t b)
 {
-    __asm__ ("mul %0\n\t"
-             : "=d" (*phigh), "=a" (*plow)
-             : "a" (a), "0" (b));
+    __uint128_t r = (__uint128_t)a * b;
+    *plow = r;
+    *phigh = r >> 64;
 }
-#define __HAVE_FAST_MULS64__
+
 static inline void muls64(uint64_t *plow, uint64_t *phigh,
                           int64_t a, int64_t b)
 {
-    __asm__ ("imul %0\n\t"
-             : "=d" (*phigh), "=a" (*plow)
-             : "a" (a), "0" (b));
+    __int128_t r = (__int128_t)a * b;
+    *plow = r;
+    *phigh = r >> 64;
 }
 #else
 void muls64(uint64_t *phigh, uint64_t *plow, int64_t a, int64_t b);
diff --git a/util/host-utils.c b/util/host-utils.c
index 5e3915a..2d06a2c 100644
--- a/util/host-utils.c
+++ b/util/host-utils.c
@@ -30,7 +30,7 @@
 //#define DEBUG_MULDIV
 
 /* Long integer helpers */
-#if !defined(__x86_64__)
+#ifndef CONFIG_INT128
 static void add128 (uint64_t *plow, uint64_t *phigh, uint64_t a, uint64_t b)
 {
     *plow += a;
@@ -102,4 +102,4 @@ void muls64 (uint64_t *plow, uint64_t *phigh, int64_t a, 
int64_t b)
            a, b, *phigh, *plow);
 #endif
 }
-#endif /* !defined(__x86_64__) */
+#endif /* !CONFIG_INT128 */
-- 
1.7.11.7