X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Farm%2Fmathops.h;h=45ac67d43625515678c53d05fdee98f85688267e;hb=82bb3048013201c0095d2853d4623633d912252f;hp=b27b18f871e48893fa865703ce2412951731c71c;hpb=21c65125424ef3dd7e276dea14f8e8ef3292e388;p=ffmpeg

diff --git a/libavcodec/arm/mathops.h b/libavcodec/arm/mathops.h
index b27b18f871e..45ac67d4362 100644
--- a/libavcodec/arm/mathops.h
+++ b/libavcodec/arm/mathops.h
@@ -28,7 +28,7 @@
 
 #if HAVE_INLINE_ASM
 
-#if HAVE_ARMV6
+#if HAVE_ARMV6_INLINE
 #define MULH MULH
 static inline av_const int MULH(int a, int b)
 {
@@ -36,11 +36,35 @@ static inline av_const int MULH(int a, int b)
     __asm__ ("smmul %0, %1, %2" : "=r"(r) : "r"(a), "r"(b));
     return r;
 }
+
+#define FASTDIV FASTDIV
+static av_always_inline av_const int FASTDIV(int a, int b)
+{
+    int r;
+    __asm__ ("cmp     %2, #2               \n\t"
+             "ldr     %0, [%3, %2, lsl #2] \n\t"
+             "ite     le                   \n\t"
+             "lsrle   %0, %1, #1           \n\t"
+             "smmulgt %0, %0, %1           \n\t"
+             : "=&r"(r) : "r"(a), "r"(b), "r"(ff_inverse) : "cc");
+    return r;
+}
+
+#else /* HAVE_ARMV6_INLINE */
+
+#define FASTDIV FASTDIV
+static av_always_inline av_const int FASTDIV(int a, int b)
+{
+    int r, t;
+    __asm__ ("umull %1, %0, %2, %3"
+             : "=&r"(r), "=&r"(t) : "r"(a), "r"(ff_inverse[b]));
+    return r;
+}
 #endif
 
 #define MLS64(d, a, b) MAC64(d, -(a), b)
 
-#if HAVE_ARMV5TE
+#if HAVE_ARMV5TE_INLINE
 
 /* signed 16x16 -> 32 multiply add accumulate */
 #   define MAC16(rt, ra, rb)                                            \
@@ -64,11 +88,14 @@ static inline av_const int mid_pred(int a, int b, int c)
     __asm__ (
         "mov   %0, %2  \n\t"
         "cmp   %1, %2  \n\t"
+        "itt   gt      \n\t"
         "movgt %0, %1  \n\t"
         "movgt %1, %2  \n\t"
         "cmp   %1, %3  \n\t"
+        "it    le      \n\t"
         "movle %1, %3  \n\t"
         "cmp   %0, %1  \n\t"
+        "it    gt      \n\t"
         "movgt %0, %1  \n\t"
         : "=&r"(m), "+r"(a)
         : "r"(b), "r"(c)