3d_gpu/architecture/dynamic_simd/repl.mdwn

   1 # PartitionedSignal nmigen-aware Repl
   2
   3 * <https://bugs.libre-soc.org/show_bug.cgi?id=458>
   4
   5 Partitioned Repl is very similar to [[cat]] and
   6 [[assign]].  The output completely changes depending
   7 on the partition mask.
   8
   9 Take a PartitionedSignal and a repeat quantity of 2:
  10
  11     partition:    p  p  p     (3 bits)
  12     a        :  A3 A2 A1 A0  (32 bits)
  13
  14 The output will be 64 bit in length. However
  15 when the partitions are 32-bit, the output is:
  16
  17     partition:       p     p     p        (3 bits)
  18     out      :  A3 A2 A1 A0 A3 A2 A1 A0  (64 bits)
  19
  20 When 2x16, the top half of A is replicated,
  21 and likewise the lower:
  22
  23     partition:       p     p     p        (3 bits)
  24     out      :  A3 A2 A3 A2 A1 A0 A1 A0  (64 bits)
  25
  26 Finally when 4x8, each byte is concatenated:
  27
  28     partition:       p     p     p        (3 bits)
  29     out      :  A3 A3 A2 A2 A1 A1 A0 A0  (64 bits)
  30