[Minios-devel] [UNIKRAFT PATCH v6 7/7] arch/arm64: Reimplement ukarch_ffsl with gcc builtin

[Jia He <justin.he@xxxxxxx>]

Using builtin is better as you can let the GCC folks to maintain the
function.

I've compared the performance between original ffsl and gcc builtin
on a ThunderX2 host(gcc version 7.4.0), default gcc optimization
options.

 for (x=0; x<0xfffffff; x++)

 original:
 real    0m1.723s
 user    0m1.723s
 sys     0m0.000s

 gcc builtin:
 real    0m1.550s
 user    0m1.546s
 sys     0m0.004s

Signed-off-by: Jia He <justin.he@xxxxxxx>
---
V6: change the returned bit postion from [1...32] to [0...31]

 arch/arm/arm64/include/uk/asm/atomic.h | 25 ++-----------------------
 1 file changed, 2 insertions(+), 23 deletions(-)

diff --git a/arch/arm/arm64/include/uk/asm/atomic.h 
b/arch/arm/arm64/include/uk/asm/atomic.h
index 431ac84..7319d66 100644
--- a/arch/arm/arm64/include/uk/asm/atomic.h
+++ b/arch/arm/arm64/include/uk/asm/atomic.h
@@ -68,30 +68,9 @@ static inline unsigned int ukarch_fls(unsigned int x)
  *
  * Undefined if no bit exists, so code should check against 0 first.
  */
-static inline unsigned long ukarch_ffsl(unsigned long word)
+static inline unsigned long ukarch_ffsl(unsigned long x)
 {
-       int clz;
-
-       /* xxxxx10000 = word
-        * xxxxx01111 = word - 1
-        * 0000011111 = word ^ (word - 1)
-        *      4     = 63 - clz(word ^ (word - 1))
-        */
-
-       __asm__("sub x0, %[word], #1\n"
-               "eor x0, x0, %[word]\n"
-               "clz %[clz], x0\n"
-               :
-               /* Outputs: */
-               [clz] "=r"(clz)
-               :
-               /* Inputs: */
-               [word] "r"(word)
-               :
-               /* Clobbers: */
-               "x0");
-
-       return 63 - clz;
+       return __builtin_ffsl(x) - 1;
 }
 
 /**
-- 
2.17.1


_______________________________________________
Minios-devel mailing list
Minios-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/minios-devel