[FFmpeg-devel] [PATCH] sws/aarch64/yuv2rgb: use altmacro for the rgb combinations

[Clément Bœsch]

From: Clément Bœsch <clement at stupeflix.com>

---
I just wanted to see if it was possible to do something less dumb than the
current code. I'm not sure I will push this patch (.altmacro is known to cause
issues and it's currently not used for aarch64), but that was just as an
exercise. It might be "inspiring" for the sws yuv2rgb ARM code that currently
needs some rework.

(Thanks Martin!)
---
 libswscale/aarch64/yuv2rgb_neon.S | 45 +++++++++++++++++----------------------
 1 file changed, 20 insertions(+), 25 deletions(-)

diff --git a/libswscale/aarch64/yuv2rgb_neon.S b/libswscale/aarch64/yuv2rgb_neon.S
index cae5384..a9ee8d5 100644
--- a/libswscale/aarch64/yuv2rgb_neon.S
+++ b/libswscale/aarch64/yuv2rgb_neon.S
@@ -123,14 +123,22 @@
     add                 v23.8H, v27.8H, v23.8H                          // Y2 + G2
     add                 v24.8H, v26.8H, v24.8H                          // Y1 + B1
     add                 v25.8H, v27.8H, v25.8H                          // Y2 + B2
-    sqrshrun            \r1, v20.8H, #1                                 // clip_u8((Y1 + R1) >> 1)
-    sqrshrun            \r2, v21.8H, #1                                 // clip_u8((Y2 + R1) >> 1)
-    sqrshrun            \g1, v22.8H, #1                                 // clip_u8((Y1 + G1) >> 1)
-    sqrshrun            \g2, v23.8H, #1                                 // clip_u8((Y2 + G1) >> 1)
-    sqrshrun            \b1, v24.8H, #1                                 // clip_u8((Y1 + B1) >> 1)
-    sqrshrun            \b2, v25.8H, #1                                 // clip_u8((Y2 + B1) >> 1)
-    movi                \a1, #255
-    movi                \a2, #255
+    sqrshrun            v\r1\().8B, v20.8H, #1                          // clip_u8((Y1 + R1) >> 1)
+    sqrshrun            v\r2\().8B, v21.8H, #1                          // clip_u8((Y2 + R1) >> 1)
+    sqrshrun            v\g1\().8B, v22.8H, #1                          // clip_u8((Y1 + G1) >> 1)
+    sqrshrun            v\g2\().8B, v23.8H, #1                          // clip_u8((Y2 + G1) >> 1)
+    sqrshrun            v\b1\().8B, v24.8H, #1                          // clip_u8((Y1 + B1) >> 1)
+    sqrshrun            v\b2\().8B, v25.8H, #1                          // clip_u8((Y2 + B1) >> 1)
+    movi                v\a1\().8B, #255
+    movi                v\a2\().8B, #255
+.endm
+
+.macro compute_rgba_ids ofmt fmt r g b a
+    .ifc \ofmt, \fmt
+        .altmacro
+            compute_rgba %(\r+4),%(\g+4),%(\b+4),%(\a+4), %(\r+16),%(\g+16),%(\b+16),%(\a+16)
+        .noaltmacro
+    .endif
 .endm
 
 .macro declare_func ifmt ofmt
@@ -164,23 +172,10 @@ function ff_\ifmt\()_to_\ofmt\()_neon, export=1
     sub                 v27.8H, v27.8H, v3.8H                           // Y2*(1<<3) - y_offset
     sqdmulh             v26.8H, v26.8H, v0.8H                           // ((Y1*(1<<3) - y_offset) * y_coeff) >> 15
     sqdmulh             v27.8H, v27.8H, v0.8H                           // ((Y2*(1<<3) - y_offset) * y_coeff) >> 15
-
-.ifc \ofmt,argb // 1 2 3 0
-    compute_rgba        v5.8B,v6.8B,v7.8B,v4.8B, v17.8B,v18.8B,v19.8B,v16.8B
-.endif
-
-.ifc \ofmt,rgba // 0 1 2 3
-    compute_rgba        v4.8B,v5.8B,v6.8B,v7.8B, v16.8B,v17.8B,v18.8B,v19.8B
-.endif
-
-.ifc \ofmt,abgr // 3 2 1 0
-    compute_rgba        v7.8B,v6.8B,v5.8B,v4.8B, v19.8B,v18.8B,v17.8B,v16.8B
-.endif
-
-.ifc \ofmt,bgra // 2 1 0 3
-    compute_rgba        v6.8B,v5.8B,v4.8B,v7.8B, v18.8B,v17.8B,v16.8B,v19.8B
-.endif
-
+    compute_rgba_ids    \ofmt, argb, 1, 2, 3, 0
+    compute_rgba_ids    \ofmt, rgba, 0, 1, 2, 3
+    compute_rgba_ids    \ofmt, abgr, 3, 2, 1, 0
+    compute_rgba_ids    \ofmt, bgra, 2, 1, 0, 3
     st4                 { v4.8B, v5.8B, v6.8B, v7.8B}, [x2], #32
     st4                 {v16.8B,v17.8B,v18.8B,v19.8B}, [x2], #32
     subs                w8, w8, #16                                     // width -= 16
-- 
2.7.2