shithub: dav1d

Info  •  Files  •  Log  •  Branches

arm64: cdef: Do saturating subtractions to avoid max operations with 0

Before:                  Cortex A53     A72     A73
cdef_filter_4x4_8bpc_neon:    677.4   433.9   452.9
cdef_filter_4x8_8bpc_neon:   1255.0   815.2   841.8
cdef_filter_8x8_8bpc_neon:   2278.5  1440.0  1505.0
After:
cdef_filter_4x4_8bpc_neon:    645.5   401.9   422.5
cdef_filter_4x8_8bpc_neon:   1193.7   756.6   782.4
cdef_filter_8x8_8bpc_neon:   2162.4  1361.9  1375.6

--- a/src/arm/64/cdef.S

+++ b/src/arm/64/cdef.S

@@ -304,10 +304,8 @@

         uabd            v20.8h, v0.8h,  \s2\().8h   // abs(diff)

         ushl            v17.8h, v16.8h, \shift      // abs(diff) >> shift

         ushl            v21.8h, v20.8h, \shift      // abs(diff) >> shift

-        sub             v17.8h, \thresh_vec, v17.8h // threshold - (abs(diff) >> shift)

-        sub             v21.8h, \thresh_vec, v21.8h // threshold - (abs(diff) >> shift)

-        smax            v17.8h, v29.8h, v17.8h      // imax(0, threshold - ())

-        smax            v21.8h, v29.8h, v21.8h      // imax(0, threshold - ())

+        uqsub           v17.8h, \thresh_vec, v17.8h // imax(0, threshold - (abs(diff) >> shift))

+        uqsub           v21.8h, \thresh_vec, v21.8h // imax(0, threshold - (abs(diff) >> shift))

         cmhi            v18.8h, v0.8h,  \s1\().8h   // px > p0

         cmhi            v22.8h, v0.8h,  \s2\().8h   // px > p1

         smin            v17.8h, v17.8h, v16.8h      // imin(abs(diff), imax())

@@ -334,7 +332,6 @@

         add             x5,  x9,  w5, uxtw #1

         movi            v31.16b,  #255

         movi            v30.8h,   #15

-        movi            v29.8h,   #0

         dup             v28.8h,   w6                // damping

         ushr            v31.8h,   v31.8h, #1        // INT16_MAX

@@ -344,10 +341,8 @@

         clz             v26.8h, v27.8h              // clz(threshold)

         sub             v24.8h, v30.8h, v24.8h      // ulog2(threshold)

         sub             v26.8h, v30.8h, v26.8h      // ulog2(threshold)

-        sub             v24.8h, v28.8h, v24.8h      // damping - ulog2(threshold)

-        sub             v26.8h, v28.8h, v26.8h      // damping - ulog2(threshold)

-        smax            v24.8h, v29.8h, v24.8h      // shift = imax(0, damping - ulog2(threshold))

-        smax            v26.8h, v29.8h, v26.8h      // shift = imax(0, damping - ulog2(threshold))

+        uqsub           v24.8h, v28.8h, v24.8h      // shift = imax(0, damping - ulog2(threshold))

+        uqsub           v26.8h, v28.8h, v26.8h      // shift = imax(0, damping - ulog2(threshold))

         neg             v24.8h, v24.8h              // -shift

         neg             v26.8h, v26.8h              // -shift