X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=rts%2Fsm%2FGCThread.h;h=ab5aca72b775b8ab6e55aa0277542b4bbc2ca671;hb=91a804680cfc5cb3894d996b9c7fb28b50180cdb;hp=aacef82cbcd404f044ebbfaf62507fbc7c683402;hpb=4e3542263207ae49963811aeb84927027e7bb61d;p=ghc-hetmet.git

diff --git a/rts/sm/GCThread.h b/rts/sm/GCThread.h
index aacef82..ab5aca7 100644
--- a/rts/sm/GCThread.h
+++ b/rts/sm/GCThread.h
@@ -11,12 +11,13 @@
  *
  * ---------------------------------------------------------------------------*/
 
-#ifndef GCTHREAD_H
-#define GCTHREAD_H
+#ifndef SM_GCTHREAD_H
+#define SM_GCTHREAD_H
 
-#include "OSThreads.h"
 #include "WSDeque.h"
 
+BEGIN_RTS_PRIVATE
+
 /* -----------------------------------------------------------------------------
    General scheme
    
@@ -75,7 +76,7 @@
 
 typedef struct step_workspace_ {
     step * step;		// the step for this workspace 
-    struct gc_thread_ * gct;    // the gc_thread that contains this workspace
+    struct gc_thread_ * my_gct; // the gc_thread that contains this workspace
 
     // where objects to be scavenged go
     bdescr *     todo_bd;
@@ -189,8 +190,6 @@ typedef struct gc_thread_ {
 
 extern nat n_gc_threads;
 
-extern gc_thread **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
@@ -203,15 +202,44 @@ extern gc_thread **gc_threads;
    __thread version.
    -------------------------------------------------------------------------- */
 
+extern gc_thread **gc_threads;
+
+#if defined(THREADED_RTS)
+
 #define GLOBAL_REG_DECL(type,name,reg) register type name REG(reg);
 
-#if defined(sparc_HOST_ARCH)
-// Don't use REG_base or R1 for gct on SPARC because they're getting clobbered 
-//	by something else. Not sure what yet. -- BL 2009/01/03
+#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_TARGET_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
@@ -219,11 +247,13 @@ extern __thread gc_thread* gct;
 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;
@@ -235,5 +265,17 @@ extern __thread gc_thread* gct;
 
 #endif
 
-#endif // GCTHREAD_H
+#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
+
+END_RTS_PRIVATE
+
+#endif // SM_GCTHREAD_H