Add comments

diff --git a/techlibs/xilinx/abc_xc7.box b/techlibs/xilinx/abc_xc7.box
index 554cc0cf00caf3a51043373e574ca5ca2aaa7f22..20da3b8a07086b4edf45e77a5183d4a6d9d5d58b 100644 (file)
--- a/techlibs/xilinx/abc_xc7.box
+++ b/techlibs/xilinx/abc_xc7.box
@@ -15,7 +15,10 @@ F7MUX 1 1 3 1
 MUXF8 2 1 3 1
 104 94 273
 
-# Box containing MUXF7.[AB] + MUXF8
+# Box containing MUXF7.[AB] + MUXF8,
+#   Necessary to make these an atomic unit so that
+#   ABC cannot optimise just one of the MUXF7 away
+#   and expect to save on its delay
 # Inputs: I0 I1 I2 I3 S0 S1
 # Outputs: O
 $__MUXF78 3 1 6 1
@@ -40,6 +43,11 @@ CARRY4 4 1 10 8
 
 # SLICEM/A6LUT
 # Box to emulate comb/seq behaviour of RAMD{32,64} and SRL{16,32}
+#   Necessary since RAMD* and SRL* have both combinatorial (i.e.
+#   same-cycle read operation) and sequential (write operation
+#   is only committed on the next clock edge).
+#   To model the combinatorial path, such cells have to be split
+#   into comb and seq parts, with this box modelling only the former.
 # Inputs: A S0 S1 S2 S3 S4 S5
 # Outputs: Y
 $__ABC_LUT6 2000 0 7 1