Refactoring and tidy up

This is a port of some of the changes from my private local-GC branch
(which is still in darcs, I haven't converted it to git yet).  There
are a couple of small functional differences in the GC stats: first,
per-thread GC timings should now be more accurate, and secondly we now
report average and maximum pause times. e.g. from minimax +RTS -N8 -s:

                                    Tot time (elapsed)  Avg pause  Max pause
  Gen  0      2755 colls,  2754 par   13.16s    0.93s     0.0003s    0.0150s
  Gen  1       769 colls,   769 par    3.71s    0.26s     0.0003s    0.0059s

1 files changed, 11 insertions, 85 deletions

diff --git a/rts/sm/GCThread.h b/rts/sm/GCThread.h
index 62dd1fb73a..e42a3a1239 100644
--- a/rts/sm/GCThread.h
+++ b/rts/sm/GCThread.h
@@ -15,6 +15,7 @@
 #define SM_GCTHREAD_H
 
 #include "WSDeque.h"
+#include "GetTime.h" // for Ticks
 
 #include "BeginPrivate.h"
 
@@ -115,6 +116,8 @@ typedef struct gen_workspace_ {
    ------------------------------------------------------------------------- */
 
 typedef struct gc_thread_ {
+    Capability *cap;
+
 #ifdef THREADED_RTS
     OSThreadId id;                 // The OS thread that this struct belongs to
     SpinLock   gc_spin;
@@ -162,7 +165,8 @@ typedef struct gc_thread_ {
                                    // instead of the to-space
                                    // corresponding to the object
 
-    lnat thunk_selector_depth;     // ummm.... not used as of now
+    lnat thunk_selector_depth;     // used to avoid unbounded recursion in 
+                                   // evacuate() for THUNK_SELECTOR
 
 #ifdef USE_PAPI
     int papi_events;
@@ -177,10 +181,15 @@ typedef struct gc_thread_ {
     lnat no_work;
     lnat scav_find_work;
 
+    Ticks gc_start_cpu;   // process CPU time
+    Ticks gc_start_elapsed;  // process elapsed time
+    Ticks gc_start_thread_cpu; // thread CPU time
+    lnat gc_start_faults;
+
     // -------------------
     // workspaces
 
-    // array of workspaces, indexed by stp->abs_no.  This is placed
+    // array of workspaces, indexed by gen->abs_no.  This is placed
     // directly at the end of the gc_thread structure so that we can get from
     // the gc_thread pointer to a workspace using only pointer
     // arithmetic, no memory access.  This happens in the inner loop
@@ -191,91 +200,8 @@ typedef struct gc_thread_ {
 
 extern nat n_gc_threads;
 
-/* -----------------------------------------------------------------------------
-   The gct variable is thread-local and points to the current thread's
-   gc_thread structure.  It is heavily accessed, so we try to put gct
-   into a global register variable if possible; if we don't have a
-   register then use gcc's __thread extension to create a thread-local
-   variable.
-
-   Even on x86 where registers are scarce, it is worthwhile using a
-   register variable here: I measured about a 2-5% slowdown with the
-   __thread version.
-   -------------------------------------------------------------------------- */
-
 extern gc_thread **gc_threads;
 
-#if defined(THREADED_RTS)
-
-#define GLOBAL_REG_DECL(type,name,reg) register type name REG(reg);
-
-#define SET_GCT(to) gct = (to)
-
-
-
-#if (defined(i386_HOST_ARCH) && defined(linux_HOST_OS))
-// Using __thread is better than stealing a register on x86/Linux, because
-// we have too few registers available.  In my tests it was worth
-// about 5% in GC performance, but of course that might change as gcc
-// improves. -- SDM 2009/04/03
-//
-// We ought to do the same on MacOS X, but __thread is not
-// supported there yet (gcc 4.0.1).
-
-extern __thread gc_thread* gct;
-#define DECLARE_GCT __thread gc_thread* gct;
-
-
-#elif defined(sparc_HOST_ARCH)
-// On SPARC we can't pin gct to a register. Names like %l1 are just offsets
-//	into the register window, which change on each function call.
-//	
-//	There are eight global (non-window) registers, but they're used for other purposes.
-//	%g0     -- always zero
-//	%g1     -- volatile over function calls, used by the linker
-//	%g2-%g3 -- used as scratch regs by the C compiler (caller saves)
-//	%g4	-- volatile over function calls, used by the linker
-//	%g5-%g7	-- reserved by the OS
-
-extern __thread gc_thread* gct;
-#define DECLARE_GCT __thread gc_thread* gct;
-
-
-#elif defined(REG_Base) && !defined(i386_HOST_ARCH)
-// on i386, REG_Base is %ebx which is also used for PIC, so we don't
-// want to steal it
-
-GLOBAL_REG_DECL(gc_thread*, gct, REG_Base)
-#define DECLARE_GCT /* nothing */
-
-
-#elif defined(REG_R1)
-
-GLOBAL_REG_DECL(gc_thread*, gct, REG_R1)
-#define DECLARE_GCT /* nothing */
-
-
-#elif defined(__GNUC__)
-
-extern __thread gc_thread* gct;
-#define DECLARE_GCT __thread gc_thread* gct;
-
-#else
-
-#error Cannot find a way to declare the thread-local gct
-
-#endif
-
-#else  // not the threaded RTS
-
-extern StgWord8 the_gc_thread[];
-
-#define gct ((gc_thread*)&the_gc_thread)
-#define SET_GCT(to) /*nothing*/
-#define DECLARE_GCT /*nothing*/
-
-#endif
-
 #include "EndPrivate.h"
 
 #endif // SM_GCTHREAD_H