Merge non-moving garbage collector

This introduces a concurrent mark & sweep garbage collector to manage the old
generation. The concurrent nature of this collector typically results in
significantly reduced maximum and mean pause times in applications with large
working sets.

Due to the large and intricate nature of the change I have opted to
preserve the fully-buildable history, including merge commits, which is
described in the "Branch overview" section below.

Collector design
================

The full design of the collector implemented here is described in detail
in a technical note

> B. Gamari. "A Concurrent Garbage Collector For the Glasgow Haskell
> Compiler" (2018)

This document can be requested from @bgamari.
The basic heap structure used in this design is heavily inspired by

> K. Ueno & A. Ohori. "A fully concurrent garbage collector for
> functional programs on multicore processors." /ACM SIGPLAN Notices/
> Vol. 51. No. 9 (presented at ICFP 2016)

This design is intended to allow both marking and sweeping
concurrent to execution of a multi-core mutator. Unlike the Ueno design,
which requires no global synchronization pauses, the collector
introduced here requires a stop-the-world pause at the beginning and end
of the mark phase.

To avoid heap fragmentation, the allocator consists of a number of
fixed-size /sub-allocators/. Each of these sub-allocators allocators into
its own set of /segments/, themselves allocated from the block
allocator. Each segment is broken into a set of fixed-size allocation
blocks (which back allocations) in addition to a bitmap (used to track
the liveness of blocks) and some additional metadata (used also used
to track liveness).

This heap structure enables collection via mark-and-sweep, which can be
performed concurrently via a snapshot-at-the-beginning scheme (although
concurrent collection is not implemented in this patch).

Implementation structure
========================

The majority of the collector is implemented in a handful of files:

 * `rts/Nonmoving.c` is the heart of the beast. It implements the entry-point
   to the nonmoving collector (`nonmoving_collect`), as well as the allocator
   (`nonmoving_allocate`) and a number of utilities for manipulating the heap.

 * `rts/NonmovingMark.c` implements the mark queue functionality, update
   remembered set, and mark loop.

 * `rts/NonmovingSweep.c` implements the sweep loop.

 * `rts/NonmovingScav.c` implements the logic necessary to scavenge the
   nonmoving heap.

Branch overview
===============

```
 * wip/gc/opt-pause:
 |   A variety of small optimisations to further reduce pause times.
 |
 * wip/gc/compact-nfdata:
 |   Introduce support for compact regions into the non-moving
 |\  collector
 | \
 |  \
 | | * wip/gc/segment-header-to-bdescr:
 | | |   Another optimization that we are considering, pushing
 | | |   some segment metadata into the segment descriptor for
 | | |   the sake of locality during mark
 | | |
 | * | wip/gc/shortcutting:
 | | |   Support for indirection shortcutting and the selector optimization
 | | |   in the non-moving heap.
 | | |
 * | | wip/gc/docs:
 | |/    Work on implementation documentation.
 | /
 |/
 * wip/gc/everything:
 |   A roll-up of everything below.
 |\
 | \
 | |\
 | | \
 | | * wip/gc/optimize:
 | | |   A variety of optimizations, primarily to the mark loop.
 | | |   Some of these are microoptimizations but a few are quite
 | | |   significant. In particular, the prefetch patches have
 | | |   produced a nontrivial improvement in mark performance.
 | | |
 | | * wip/gc/aging:
 | | |   Enable support for aging in major collections.
 | | |
 | * | wip/gc/test:
 | | |   Fix up the testsuite to more or less pass.
 | | |
 * | | wip/gc/instrumentation:
 | | |   A variety of runtime instrumentation including statistics
 | | /   support, the nonmoving census, and eventlog support.
 | |/
 | /
 |/
 * wip/gc/nonmoving-concurrent:
 |   The concurrent write barriers.
 |
 * wip/gc/nonmoving-nonconcurrent:
 |   The nonmoving collector without the write barriers necessary
 |   for concurrent collection.
 |
 * wip/gc/preparation:
 |   A merge of the various preparatory patches that aren't directly
 |   implementing the GC.
 |
 |
 * GHC HEAD
 .
 .
 .
```

1 files changed, 86 insertions, 26 deletions

diff --git a/rts/PrimOps.cmm b/rts/PrimOps.cmm
index ec35ee42b4..b66c561dcb 100644
--- a/rts/PrimOps.cmm
+++ b/rts/PrimOps.cmm
@@ -349,8 +349,13 @@ stg_casArrayzh ( gcptr arr, W_ ind, gcptr old, gcptr new )
         // Compare and Swap Succeeded:
         SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, CCCS);
         len = StgMutArrPtrs_ptrs(arr);
+
         // The write barrier.  We must write a byte into the mark table:
         I8[arr + SIZEOF_StgMutArrPtrs + WDS(len) + (ind >> MUT_ARR_PTRS_CARD_BITS )] = 1;
+
+        // Concurrent GC write barrier
+        updateRemembSetPushPtr(old);
+
         return (0,new);
     }
 }
@@ -462,16 +467,45 @@ stg_thawSmallArrayzh ( gcptr src, W_ offset, W_ n )
     cloneSmallArray(stg_SMALL_MUT_ARR_PTRS_DIRTY_info, src, offset, n)
 }
 
+// Concurrent GC write barrier for pointer array copies
+//
+// hdr_size in bytes. dst_off in words, n in words.
+stg_copyArray_barrier ( W_ hdr_size, gcptr dst, W_ dst_off, W_ n)
+{
+    W_ end, p;
+    ASSERT(n > 0);  // Assumes n==0 is handled by caller
+    p = dst + hdr_size + WDS(dst_off);
+    end = p + WDS(n);
+
+again:
+    IF_NONMOVING_WRITE_BARRIER_ENABLED {
+        ccall updateRemembSetPushClosure_(BaseReg "ptr", W_[p] "ptr");
+    }
+    p = p + WDS(1);
+    if (p < end) {
+        goto again;
+    }
+
+    return ();
+}
+
 stg_copySmallArrayzh ( gcptr src, W_ src_off, gcptr dst, W_ dst_off, W_ n)
 {
     W_ dst_p, src_p, bytes;
 
-    SET_INFO(dst, stg_SMALL_MUT_ARR_PTRS_DIRTY_info);
+    if (n > 0) {
+        IF_NONMOVING_WRITE_BARRIER_ENABLED {
+            call stg_copyArray_barrier(SIZEOF_StgSmallMutArrPtrs,
+                                      dst, dst_off, n);
+        }
 
-    dst_p = dst + SIZEOF_StgSmallMutArrPtrs + WDS(dst_off);
-    src_p = src + SIZEOF_StgSmallMutArrPtrs + WDS(src_off);
-    bytes = WDS(n);
-    prim %memcpy(dst_p, src_p, bytes, SIZEOF_W);
+        SET_INFO(dst, stg_SMALL_MUT_ARR_PTRS_DIRTY_info);
+
+        dst_p = dst + SIZEOF_StgSmallMutArrPtrs + WDS(dst_off);
+        src_p = src + SIZEOF_StgSmallMutArrPtrs + WDS(src_off);
+        bytes = WDS(n);
+        prim %memcpy(dst_p, src_p, bytes, SIZEOF_W);
+    }
 
     return ();
 }
@@ -480,15 +514,22 @@ stg_copySmallMutableArrayzh ( gcptr src, W_ src_off, gcptr dst, W_ dst_off, W_ n
 {
     W_ dst_p, src_p, bytes;
 
-    SET_INFO(dst, stg_SMALL_MUT_ARR_PTRS_DIRTY_info);
+    if (n > 0) {
+        IF_NONMOVING_WRITE_BARRIER_ENABLED {
+            call stg_copyArray_barrier(SIZEOF_StgSmallMutArrPtrs,
+                                      dst, dst_off, n);
+        }
 
-    dst_p = dst + SIZEOF_StgSmallMutArrPtrs + WDS(dst_off);
-    src_p = src + SIZEOF_StgSmallMutArrPtrs + WDS(src_off);
-    bytes = WDS(n);
-    if (src == dst) {
-        prim %memmove(dst_p, src_p, bytes, SIZEOF_W);
-    } else {
-        prim %memcpy(dst_p, src_p, bytes, SIZEOF_W);
+        SET_INFO(dst, stg_SMALL_MUT_ARR_PTRS_DIRTY_info);
+
+        dst_p = dst + SIZEOF_StgSmallMutArrPtrs + WDS(dst_off);
+        src_p = src + SIZEOF_StgSmallMutArrPtrs + WDS(src_off);
+        bytes = WDS(n);
+        if (src == dst) {
+            prim %memmove(dst_p, src_p, bytes, SIZEOF_W);
+        } else {
+            prim %memcpy(dst_p, src_p, bytes, SIZEOF_W);
+        }
     }
 
     return ();
@@ -510,6 +551,10 @@ stg_casSmallArrayzh ( gcptr arr, W_ ind, gcptr old, gcptr new )
     } else {
         // Compare and Swap Succeeded:
         SET_HDR(arr, stg_SMALL_MUT_ARR_PTRS_DIRTY_info, CCCS);
+
+        // Concurrent GC write barrier
+        updateRemembSetPushPtr(old);
+
         return (0,new);
     }
 }
@@ -549,7 +594,7 @@ stg_casMutVarzh ( gcptr mv, gcptr old, gcptr new )
         return (1,h);
     } else {
         if (GET_INFO(mv) == stg_MUT_VAR_CLEAN_info) {
-            ccall dirty_MUT_VAR(BaseReg "ptr", mv "ptr");
+            ccall dirty_MUT_VAR(BaseReg "ptr", mv "ptr", old);
         }
         return (0,new);
     }
@@ -562,7 +607,7 @@ stg_casMutVarzh ( gcptr mv, gcptr old, gcptr new )
     } else {
         StgMutVar_var(mv) = new;
         if (GET_INFO(mv) == stg_MUT_VAR_CLEAN_info) {
-            ccall dirty_MUT_VAR(BaseReg "ptr", mv "ptr");
+            ccall dirty_MUT_VAR(BaseReg "ptr", mv "ptr", old);
         }
         return (0,new);
     }
@@ -629,11 +674,12 @@ stg_atomicModifyMutVar2zh ( gcptr mv, gcptr f )
     (h) = prim %cmpxchgW(mv + SIZEOF_StgHeader + OFFSET_StgMutVar_var, x, y);
     if (h != x) { goto retry; }
 #else
+    h = StgMutVar_var(mv);
     StgMutVar_var(mv) = y;
 #endif
 
     if (GET_INFO(mv) == stg_MUT_VAR_CLEAN_info) {
-        ccall dirty_MUT_VAR(BaseReg "ptr", mv "ptr");
+        ccall dirty_MUT_VAR(BaseReg "ptr", mv "ptr", h);
     }
 
     return (x,z);
@@ -755,6 +801,9 @@ stg_addCFinalizzerToWeakzh ( W_ fptr,   // finalizer
         return (0);
     }
 
+    // Write barrier for concurrent non-moving collector
+    updateRemembSetPushPtr(StgWeak_cfinalizers(w))
+
     StgCFinalizerList_link(c) = StgWeak_cfinalizers(w);
     StgWeak_cfinalizers(w) = c;
 
@@ -835,6 +884,8 @@ stg_deRefWeakzh ( gcptr w )
     if (info == stg_WEAK_info) {
         code = 1;
         val = StgWeak_value(w);
+        // See Note [Concurrent read barrier on deRefWeak#] in NonMovingMark.c
+        updateRemembSetPushPtr(val);
     } else {
         code = 0;
         val = w;
@@ -1515,7 +1566,7 @@ stg_takeMVarzh ( P_ mvar /* :: MVar a */ )
      */
     if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
         if (info == stg_MVAR_CLEAN_info) {
-            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr");
+            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr", StgMVar_value(mvar) "ptr");
         }
 
         // We want to put the heap check down here in the slow path,
@@ -1561,6 +1612,9 @@ loop:
         // If the MVar is not already dirty, then we don't need to make
         // it dirty, as it is empty with nothing blocking on it.
         unlockClosure(mvar, info);
+        // However, we do need to ensure that the nonmoving collector
+        // knows about the reference to the value that we just removed...
+        updateRemembSetPushPtr(val);
         return (val);
     }
     qinfo = StgHeader_info(q);
@@ -1574,7 +1628,7 @@ loop:
     // There are putMVar(s) waiting... wake up the first thread on the queue
 
     if (info == stg_MVAR_CLEAN_info) {
-        ccall dirty_MVAR(BaseReg "ptr", mvar "ptr");
+        ccall dirty_MVAR(BaseReg "ptr", mvar "ptr", val "ptr");
     }
 
     tso = StgMVarTSOQueue_tso(q);
@@ -1643,7 +1697,7 @@ loop:
     // There are putMVar(s) waiting... wake up the first thread on the queue
 
     if (info == stg_MVAR_CLEAN_info) {
-        ccall dirty_MVAR(BaseReg "ptr", mvar "ptr");
+        ccall dirty_MVAR(BaseReg "ptr", mvar "ptr", val "ptr");
     }
 
     tso = StgMVarTSOQueue_tso(q);
@@ -1681,7 +1735,7 @@ stg_putMVarzh ( P_ mvar, /* :: MVar a */
     if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) {
 
         if (info == stg_MVAR_CLEAN_info) {
-            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr");
+            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr", StgMVar_value(mvar) "ptr");
         }
 
         // We want to put the heap check down here in the slow path,
@@ -1715,14 +1769,20 @@ stg_putMVarzh ( P_ mvar, /* :: MVar a */
         jump stg_block_putmvar(mvar,val);
     }
 
+    // We are going to mutate the closure, make sure its current pointers
+    // are marked.
+    if (info == stg_MVAR_CLEAN_info) {
+        ccall update_MVAR(BaseReg "ptr", mvar "ptr", StgMVar_value(mvar) "ptr");
+    }
+
     q = StgMVar_head(mvar);
 loop:
     if (q == stg_END_TSO_QUEUE_closure) {
         /* No further takes, the MVar is now full. */
+        StgMVar_value(mvar) = val;
         if (info == stg_MVAR_CLEAN_info) {
-            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr");
+            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr", StgMVar_value(mvar) "ptr");
         }
-        StgMVar_value(mvar) = val;
         unlockClosure(mvar, stg_MVAR_DIRTY_info);
         return ();
     }
@@ -1758,7 +1818,7 @@ loop:
     // indicate that the MVar operation has now completed.
     StgTSO__link(tso) = stg_END_TSO_QUEUE_closure;
 
-    if (TO_W_(StgStack_dirty(stack)) == 0) {
+    if ((TO_W_(StgStack_dirty(stack)) & STACK_DIRTY) == 0) {
         ccall dirty_STACK(MyCapability() "ptr", stack "ptr");
     }
 
@@ -1804,7 +1864,7 @@ loop:
     if (q == stg_END_TSO_QUEUE_closure) {
         /* No further takes, the MVar is now full. */
         if (info == stg_MVAR_CLEAN_info) {
-            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr");
+            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr", StgMVar_value(mvar) "ptr");
         }
 
         StgMVar_value(mvar) = val;
@@ -1843,7 +1903,7 @@ loop:
     // indicate that the MVar operation has now completed.
     StgTSO__link(tso) = stg_END_TSO_QUEUE_closure;
 
-    if (TO_W_(StgStack_dirty(stack)) == 0) {
+    if ((TO_W_(StgStack_dirty(stack)) & STACK_DIRTY) == 0) {
         ccall dirty_STACK(MyCapability() "ptr", stack "ptr");
     }
 
@@ -1875,7 +1935,7 @@ stg_readMVarzh ( P_ mvar, /* :: MVar a */ )
     if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
 
         if (info == stg_MVAR_CLEAN_info) {
-            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr");
+            ccall dirty_MVAR(BaseReg "ptr", mvar "ptr", StgMVar_value(mvar));
         }
 
         ALLOC_PRIM_WITH_CUSTOM_FAILURE