adjust emscrip parallelization heuristics

emscripten.py

@@ -42,7 +42,7 @@ def scan(ll, settings):
   if len(blockaddrs) > 0:
     settings['NECESSARY_BLOCKADDRS'] = blockaddrs
 
-NUM_CHUNKS_PER_CORE = 4
+NUM_CHUNKS_PER_CORE = 5
 MIN_CHUNK_SIZE = 1024*1024
 MAX_CHUNK_SIZE = float(os.environ.get('EMSCRIPT_MAX_CHUNK_SIZE') or 'inf') # configuring this is just for debugging purposes
 
@@ -149,7 +149,7 @@ def emscript(infile, settings, outfile, libraries=[]):
   if cores > 1:
     intended_num_chunks = cores * NUM_CHUNKS_PER_CORE
     chunk_size = max(MIN_CHUNK_SIZE, total_ll_size / intended_num_chunks)
-    chunk_size += len(forwarded_data)/3 + 3*len(meta) # keep ratio of lots of function code to meta (expensive to process) and forwarded (cheap)
+    chunk_size += 3*len(meta) # keep ratio of lots of function code to meta (expensive to process, and done in each parallel task)
     chunk_size = min(MAX_CHUNK_SIZE, chunk_size)
   else:
     chunk_size = MAX_CHUNK_SIZE # if 1 core, just use the max chunk size
@@ -170,7 +170,7 @@ def emscript(infile, settings, outfile, libraries=[]):
     chunks.append(curr)
     curr = ''
   if cores == 1 and total_ll_size < MAX_CHUNK_SIZE: assert len(chunks) == 1, 'no point in splitting up without multiple cores'
-  if DEBUG: print >> sys.stderr, '  emscript: phase 2 working on %d chunks %s (intended chunk size: %.2f MB, meta: %.2f MB, forwarded: %.2f)' % (len(chunks), ('using %d cores' % cores) if len(chunks) > 1 else '', chunk_size/(1024*1024.), len(meta)/(1024*1024.), len(forwarded_data)/(1024*1024.))
+  if DEBUG: print >> sys.stderr, '  emscript: phase 2 working on %d chunks %s (intended chunk size: %.2f MB, meta: %.2f MB, forwarded: %.2f MB, total: %.2f MB)' % (len(chunks), ('using %d cores' % cores) if len(chunks) > 1 else '', chunk_size/(1024*1024.), len(meta)/(1024*1024.), len(forwarded_data)/(1024*1024.), total_ll_size/(1024*1024.))
 
   commands = [(i, chunks[i] + '\n' + meta, settings_file, compiler, forwarded_file, libraries) for i in range(len(chunks))]