Remove min payload size restriction from the compiler and RTSwip/osa1/min_payload_size

Previously we'd require every heap object to have at least one word
space in the payload. This was *incorrectly* documented as follows in
the source code: (now deleted)

    Minimum closure sizes

    This is the minimum number of words in the payload of a
    heap-allocated closure, so that the closure has enough room to be
    overwritten with a forwarding pointer during garbage collection.

This comment was wrong: the GC doesn't require a location in the
payload, forwarding pointers are made using the info table pointer
location.

The real reason why we needed a word in the payload is because
mark-compact collector allocates one bit per word in the heap for
bitmap, but needs two bits per object to work efficiently. Details are
explained in the new note Note [Mark bits in mark-compact collector] in
Compact.h.

We now allocate two bits in the bitmap per word and remove the
restriction. Heap objects can now have size 1, as demonstrated in the
updated tests.

Nofib results:

- 0.0% change in binary sizes and TotalMem
- "contaraints" now allocates -0.1% less. No other changes in
  allocations.
- Runtime/Elapsed results are noisy and inconclusive

1 files changed, 6 insertions, 9 deletions

diff --git a/rts/sm/GC.c b/rts/sm/GC.c
index 92a5e229a1..53c0d53e92 100644
--- a/rts/sm/GC.c
+++ b/rts/sm/GC.c
@@ -1414,17 +1414,14 @@ prepare_collected_gen (generation *gen)
 
     // for a compacted generation, we need to allocate the bitmap
     if (gen->mark) {
-        StgWord bitmap_size; // in bytes
-        bdescr *bitmap_bdescr;
-        StgWord *bitmap;
-
-        bitmap_size = gen->n_old_blocks * BLOCK_SIZE / BITS_IN(W_);
+        // in bytes
+        StgWord bitmap_size = (gen->n_old_blocks * BLOCK_SIZE / BITS_IN(W_)) * 2;
 
         if (bitmap_size > 0) {
-            bitmap_bdescr = allocGroup((StgWord)BLOCK_ROUND_UP(bitmap_size)
-                                       / BLOCK_SIZE);
+            bdescr *bitmap_bdescr =
+                allocGroup((StgWord)BLOCK_ROUND_UP(bitmap_size) / BLOCK_SIZE);
             gen->bitmap = bitmap_bdescr;
-            bitmap = bitmap_bdescr->start;
+            StgWord *bitmap = bitmap_bdescr->start;
 
             debugTrace(DEBUG_gc, "bitmap_size: %d, bitmap: %p",
                        bitmap_size, bitmap);
@@ -1436,7 +1433,7 @@ prepare_collected_gen (generation *gen)
             // block descriptor.
             for (bd=gen->old_blocks; bd != NULL; bd = bd->link) {
                 bd->u.bitmap = bitmap;
-                bitmap += BLOCK_SIZE_W / BITS_IN(W_);
+                bitmap += (BLOCK_SIZE_W / BITS_IN(W_)) * 2;
 
                 // Also at this point we set the BF_MARKED flag
                 // for this block.  The invariant is that