ARM: fix MUL64 inline asm for pre-armv6

Prior to ARMv6, the destination registers of the SMULL instruction
must be distinct from the first source register.  Marking the
output early-clobber ensures it is allocated unique registers.

This restriction is dropped in ARMv6 and later, so allowing overlap
between input and output registers there might give better code.

Signed-off-by: Mans Rullgard <mans@mansr.com>

1 files changed, 11 insertions, 3 deletions

diff --git a/libavcodec/arm/mathops.h b/libavcodec/arm/mathops.h
index dfa941161a..7c2acca2e8 100644
--- a/libavcodec/arm/mathops.h
+++ b/libavcodec/arm/mathops.h
@@ -41,6 +41,8 @@ static inline av_const int MULL(int a, int b, unsigned shift)
 }
 
 #define MULH MULH
+#define MUL64 MUL64
+
 #if HAVE_ARMV6
 static inline av_const int MULH(int a, int b)
 {
@@ -48,6 +50,13 @@ static inline av_const int MULH(int a, int b)
     __asm__ ("smmul %0, %1, %2" : "=r"(r) : "r"(a), "r"(b));
     return r;
 }
+
+static inline av_const int64_t MUL64(int a, int b)
+{
+    int64_t x;
+    __asm__ ("smull %Q0, %R0, %1, %2" : "=r"(x) : "r"(a), "r"(b));
+    return x;
+}
 #else
 static inline av_const int MULH(int a, int b)
 {
@@ -55,15 +64,14 @@ static inline av_const int MULH(int a, int b)
     __asm__ ("smull %0, %1, %2, %3" : "=&r"(lo), "=&r"(hi) : "r"(b), "r"(a));
     return hi;
 }
-#endif
 
 static inline av_const int64_t MUL64(int a, int b)
 {
     int64_t x;
-    __asm__ ("smull %Q0, %R0, %1, %2" : "=r"(x) : "r"(a), "r"(b));
+    __asm__ ("smull %Q0, %R0, %1, %2" : "=&r"(x) : "r"(a), "r"(b));
     return x;
 }
-#define MUL64 MUL64
+#endif
 
 static inline av_const int64_t MAC64(int64_t d, int a, int b)
 {