Change #272258

Comments

swscale/tests/sws_ops: split passes when printing ops lists
This affects a large number of conversions across the board, either:

1. Lifting a constant alpha/chroma clear out from the conversion pass:

 rgb24 16x16 -> yuva444p 16x16:
+  [ u8 $XXX] SWS_OP_CLEAR        : {255 _ _ _}
+  [ u8 XXXX] SWS_OP_WRITE        : 1 elem(s) planar >> 0, via {3}
+    ('X' unused, 'z' byteswapped, '=' copied, '$' const, '+' integer, '0' zero)
+ translated micro-ops:
+    u8_clear_x_1
+    u8_write_planar_x
+ Sub-pass #1:
   [ u8 +++X] SWS_OP_READ         : 3 elem(s) packed >> 0
   [ u8 +++X] SWS_OP_CONVERT      : u8 -> f32
   [f32 ...X] SWS_OP_LINEAR       : matrix3+off3 [...]
   [f32 ...X] SWS_OP_DITHER       : 16x16 matrix + {0 3 2 -1}
   [f32 +++X] SWS_OP_CONVERT      : f32 -> u8
-  [ u8 +++$] SWS_OP_CLEAR        : {_ _ _ 255}
-  [ u8 XXXX] SWS_OP_WRITE        : 4 elem(s) planar >> 0
+  [ u8 XXXX] SWS_OP_WRITE        : 3 elem(s) planar >> 0
     ('X' unused, 'z' byteswapped, '=' copied, '$' const, '+' integer, '0' zero)

 gray 16x16 -> yuv444p 16x16:
+  [ u8 $$XX] SWS_OP_CLEAR        : {128 128 _ _}
+  [ u8 XXXX] SWS_OP_WRITE        : 2 elem(s) planar >> 0, via {2, 1}
+    ('X' unused, 'z' byteswapped, '=' copied, '$' const, '+' integer, '0' zero)
+ translated micro-ops:
+    u8_clear_xy_xx
+    u8_write_planar_xy
+ Sub-pass #1:
   [ u8 =XXX] SWS_OP_READ         : 1 elem(s) planar >> 0
   [ u8 =XXX] SWS_OP_CONVERT      : u8 -> f32
   [f32 .XXX] SWS_OP_LINEAR       : luma [...]
   [f32 .XXX] SWS_OP_DITHER       : 16x16 matrix + {0 -1 -1 -1}
   [f32 +XXX] SWS_OP_CONVERT      : f32 -> u8
-  [ u8 +$$X] SWS_OP_CLEAR        : {_ 128 128 _}
-  [ u8 XXXX] SWS_OP_WRITE        : 3 elem(s) planar >> 0
+  [ u8 XXXX] SWS_OP_WRITE        : 1 elem(s) planar >> 0
     ('X' unused, 'z' byteswapped, '=' copied, '$' const, '+' integer, '0' zero)
  translated micro-ops:
     u8_read_planar_x
     u8_to_f32_x
     f32_linear_x_x000x
     f32_dither_x_0_16x16
     f32_to_u8_x
-    u8_clear_yz_xx
-    u8_write_planar_xyz
+    u8_write_planar_x

or

2. Passing through a plane that was previously unmodified by an ops chain:

 gbrap 16x16 -> yuva444p 16x16:
-  [ u8 ====] SWS_OP_READ         : 4 elem(s) planar >> 0, via {2, 0, 1, 3}
-  [ u8 ====] SWS_OP_CONVERT      : u8 -> f32
-  [f32 ...=] SWS_OP_LINEAR       : matrix3+off3 [...]
-  [f32 ...=] SWS_OP_DITHER       : 16x16 matrix + {0 3 2 -1}
-  [f32 +++=] SWS_OP_CONVERT      : f32 -> u8
-  [ u8 XXXX] SWS_OP_WRITE        : 4 elem(s) planar >> 0
+  [ u8 =XXX] SWS_OP_READ         : 1 elem(s) planar >> 0, via {3}
+  [ u8 XXXX] SWS_OP_WRITE        : 1 elem(s) planar >> 0, via {3}
     ('X' unused, 'z' byteswapped, '=' copied, '$' const, '+' integer, '0' zero)
  translated micro-ops:
-    u8_read_planar_xyzw
-    u8_to_f32_xyzw
+    u8_read_planar_x
+    u8_write_planar_x
+ Sub-pass #1:
+  [ u8 ===X] SWS_OP_READ         : 3 elem(s) planar >> 0, via {2, 0, 1}
+  [ u8 ===X] SWS_OP_CONVERT      : u8 -> f32
+  [f32 ...X] SWS_OP_LINEAR       : matrix3+off3 [...]
+  [f32 ...X] SWS_OP_DITHER       : 16x16 matrix + {0 3 2 -1}
+  [f32 +++X] SWS_OP_CONVERT      : f32 -> u8
+  [ u8 XXXX] SWS_OP_WRITE        : 3 elem(s) planar >> 0
+    ('X' unused, 'z' byteswapped, '=' copied, '$' const, '+' integer, '0' zero)
+ translated micro-ops:
+    u8_read_planar_xyz
+    u8_to_f32_xyz
     f32_linear_xyz_xxx0x_xxx0x_xxx0x
     f32_dither_xyz_0_3_2_16x16
-    f32_to_u8_xyzw
-    u8_write_planar_xyzw
+    f32_to_u8_xyz
+    u8_write_planar_xyz

(Op lists are abridged slightly for brevity)

Sponsored-by: Sovereign Tech Fund
Signed-off-by: Niklas Haas <git@haasn.dev>

Changed files

• libswscale/tests/sws_ops.c
• tests/ref/fate/sws-ops-list