[FFmpeg-devel] [PATCH 1/2] aarch64: me_cmp: Avoid redundant loads in ff_pix_abs16_y2_neon

[Martin Storsjö]

This avoids one redundant load per row; pix3 from the previous
iteration can be used as pix2 in the next one.

Before:       Cortex A53    A72    A73
pix_abs_0_2_neon:  138.0   59.7   48.0
After:
pix_abs_0_2_neon:  109.7   50.2   39.5

Signed-off-by: Martin Storsjö <martin at martin.st>
---
 libavcodec/aarch64/me_cmp_neon.S | 24 ++++++++++--------------
 1 file changed, 10 insertions(+), 14 deletions(-)

diff --git a/libavcodec/aarch64/me_cmp_neon.S b/libavcodec/aarch64/me_cmp_neon.S
index 11af4849f9..832a7cb22d 100644
--- a/libavcodec/aarch64/me_cmp_neon.S
+++ b/libavcodec/aarch64/me_cmp_neon.S
@@ -326,9 +326,9 @@ function ff_pix_abs16_y2_neon, export=1
         // w4           int h
 
         // initialize buffers
+        ld1             {v1.16b}, [x2], x3              // Load pix2
         movi            v29.8h, #0                      // clear the accumulator
         movi            v28.8h, #0                      // clear the accumulator
-        add             x5, x2, x3                      // pix2 + stride
         cmp             w4, #4
         b.lt            2f
 
@@ -339,29 +339,25 @@ function ff_pix_abs16_y2_neon, export=1
         // avg2(a, b) = (((a) + (b) + 1) >> 1)
         // abs(x) = (x < 0 ? (-x) : (x))
 
-        ld1             {v1.16b}, [x2], x3              // Load pix2 for first iteration
-        ld1             {v2.16b}, [x5], x3              // Load pix3 for first iteration
+        ld1             {v2.16b}, [x2], x3              // Load pix3 for first iteration
         ld1             {v0.16b}, [x1], x3              // Load pix1 for first iteration
         urhadd          v30.16b, v1.16b, v2.16b         // Rounding halving add, first iteration
-        ld1             {v4.16b}, [x2], x3              // Load pix2 for second iteration
-        ld1             {v5.16b}, [x5], x3              // Load pix3 for second iteartion
+        ld1             {v5.16b}, [x2], x3              // Load pix3 for second iteartion
         uabal           v29.8h, v0.8b, v30.8b           // Absolute difference of lower half, first iteration
         uabal2          v28.8h, v0.16b, v30.16b         // Absolute difference of upper half, first iteration
         ld1             {v3.16b}, [x1], x3              // Load pix1 for second iteration
-        urhadd          v27.16b, v4.16b, v5.16b         // Rounding halving add, second iteration
-        ld1             {v7.16b}, [x2], x3              // Load pix2 for third iteration
-        ld1             {v20.16b}, [x5], x3             // Load pix3 for third iteration
+        urhadd          v27.16b, v2.16b, v5.16b         // Rounding halving add, second iteration
+        ld1             {v20.16b}, [x2], x3             // Load pix3 for third iteration
         uabal           v29.8h, v3.8b, v27.8b           // Absolute difference of lower half for second iteration
         uabal2          v28.8h, v3.16b, v27.16b         // Absolute difference of upper half for second iteration
         ld1             {v6.16b}, [x1], x3              // Load pix1 for third iteration
-        urhadd          v26.16b, v7.16b, v20.16b        // Rounding halving add, third iteration
-        ld1             {v22.16b}, [x2], x3             // Load pix2 for fourth iteration
-        ld1             {v23.16b}, [x5], x3             // Load pix3 for fourth iteration
+        urhadd          v26.16b, v5.16b, v20.16b        // Rounding halving add, third iteration
+        ld1             {v1.16b}, [x2], x3              // Load pix3 for fourth iteration
         uabal           v29.8h, v6.8b, v26.8b           // Absolute difference of lower half for third iteration
         uabal2          v28.8h, v6.16b, v26.16b         // Absolute difference of upper half for third iteration
         ld1             {v21.16b}, [x1], x3             // Load pix1 for fourth iteration
         sub             w4, w4, #4                      // h-= 4
-        urhadd          v25.16b, v22.16b, v23.16b       // Rounding halving add
+        urhadd          v25.16b, v20.16b, v1.16b        // Rounding halving add
         cmp             w4, #4
         uabal           v29.8h, v21.8b, v25.8b          // Absolute difference of lower half for fourth iteration
         uabal2          v28.8h, v21.16b, v25.16b        // Absolute difference of upper half for fourth iteration
@@ -372,11 +368,11 @@ function ff_pix_abs16_y2_neon, export=1
 // iterate by one
 2:
 
-        ld1             {v1.16b}, [x2], x3              // Load pix2
-        ld1             {v2.16b}, [x5], x3              // Load pix3
+        ld1             {v2.16b}, [x2], x3              // Load pix3
         subs            w4, w4, #1
         ld1             {v0.16b}, [x1], x3              // Load pix1
         urhadd          v30.16b, v1.16b, v2.16b         // Rounding halving add
+        mov             v1.16b, v2.16b                  // Shift pix3->pix2
         uabal           v29.8h, v30.8b, v0.8b
         uabal2          v28.8h, v30.16b, v0.16b
 
-- 
2.25.1