synthesis: use fast coarse module size estimate to drive flattening policy

Turns out that the improvement in quality of results is roughly
linear or the user can make explicit choices about which modules
to flatten, so no point in anything but a very rough fast module
size estimate.

Signed-off-by: Øyvind Harboe <[email protected]>

Author: oharboe

flow/scripts/synth.tcl

@@ -13,9 +13,15 @@ if {![env_var_equals SYNTH_HIERARCHICAL 1]} {
   # (-flatten part of $synth_args per default)
   synth -run :fine {*}$::env(SYNTH_FULL_ARGS)
 } else {
-  # Perform standard coarse-level synthesis script,
-  # defer flattening until we have decided what hierarchy to keep
-  synth -run :fine
+  # Get a quick estimate of module sizes. Since we're only using
+  # this to decide which modules to flatten and which to keep,
+  # actual quality of results doesn't matter, so no point in running
+  # any optimizations of a coarse synthesis. the user has the
+  # option to tweak MAX_UNGROUP_SIZE or use an explicit list of
+  # modules to flatten if they have a better policy than this
+  # default policy on what to keep and what to flatten.
+  procs
+  memory -nomap
 
   if {[env_var_exists_and_non_empty MAX_UNGROUP_SIZE]} {
     set ungroup_threshold $::env(MAX_UNGROUP_SIZE)