Merge branch 'wip/T14951' into wip/T14068wip/T14068

7 files changed, 35 insertions, 4 deletions

diff --git a/includes/Cmm.h b/includes/Cmm.h
index 57d78ccaa5..18b2aaf324 100644
--- a/includes/Cmm.h
+++ b/includes/Cmm.h
@@ -161,9 +161,11 @@
 
 /* TO_W_(n) converts n to W_ type from a smaller type */
 #if SIZEOF_W == 4
+#define TO_I64(x) %sx64(x)
 #define TO_W_(x) %sx32(x)
 #define HALF_W_(x) %lobits16(x)
 #elif SIZEOF_W == 8
+#define TO_I64(x) (x)
 #define TO_W_(x) %sx64(x)
 #define HALF_W_(x) %lobits32(x)
 #endif
diff --git a/includes/RtsAPI.h b/includes/RtsAPI.h
index 27a5080220..6f011cbf6e 100644
--- a/includes/RtsAPI.h
+++ b/includes/RtsAPI.h
@@ -210,6 +210,30 @@ typedef struct _RTSStats {
 
   GCDetails gc;
 
+  // -----------------------------------
+  // Internal Counters
+
+  // The number of times a GC thread spun on its 'gc_spin' lock.
+  // Will be zero if the rts was not built with PROF_SPIN
+  uint64_t gc_spin_spin;
+  // The number of times a GC thread yielded on its 'gc_spin' lock.
+  // Will be zero if the rts was not built with PROF_SPIN
+  uint64_t gc_spin_yield;
+  // The number of times a GC thread spun on its 'mut_spin' lock.
+  // Will be zero if the rts was not built with PROF_SPIN
+  uint64_t mut_spin_spin;
+  // The number of times a GC thread yielded on its 'mut_spin' lock.
+  // Will be zero if the rts was not built with PROF_SPIN
+  uint64_t mut_spin_yield;
+  // The number of times a GC thread has checked for work across all parallel
+  // GCs
+  uint64_t any_work;
+  // The number of times a GC thread has checked for work and found none across
+  // all parallel GCs
+  uint64_t no_work;
+  // The number of times a GC thread has iterated it's outer loop across all
+  // parallel GCs
+  uint64_t scav_find_work;
 } RTSStats;
 
 void getRTSStats (RTSStats *s);
diff --git a/includes/rts/Flags.h b/includes/rts/Flags.h
index aed4dca384..6487947749 100644
--- a/includes/rts/Flags.h
+++ b/includes/rts/Flags.h
@@ -195,6 +195,7 @@ typedef struct _MISC_FLAGS {
     bool generate_dump_file;
     bool generate_stack_trace;
     bool machineReadable;
+    bool internalCounters;       /* See Note [Internal Counter Stats] */
     StgWord linkerMemBase;       /* address to ask the OS for memory
                                   * for the linker, NULL ==> off */
 } MISC_FLAGS;
diff --git a/includes/rts/SpinLock.h b/includes/rts/SpinLock.h
index 6530a3a2f0..1dca02f795 100644
--- a/includes/rts/SpinLock.h
+++ b/includes/rts/SpinLock.h
@@ -27,7 +27,8 @@
 typedef struct SpinLock_
 {
     StgWord   lock;
-    StgWord64 spin; // DEBUG version counts how much it spins
+    StgWord64 spin;  // incremented every time we spin in ACQUIRE_SPIN_LOCK
+    StgWord64 yield; // incremented every time we yield in ACQUIRE_SPIN_LOCK
 } SpinLock;
 #else
 typedef StgWord SpinLock;
@@ -49,6 +50,7 @@ INLINE_HEADER void ACQUIRE_SPIN_LOCK(SpinLock * p)
             p->spin++;
             busy_wait_nop();
         }
+        p->yield++;
         yieldThread();
     } while (1);
 }
@@ -66,6 +68,7 @@ INLINE_HEADER void initSpinLock(SpinLock * p)
     write_barrier();
     p->lock = 1;
     p->spin = 0;
+    p->yield = 0;
 }
 
 #else
diff --git a/includes/rts/Threads.h b/includes/rts/Threads.h
index fceacdc75d..f72f5ed121 100644
--- a/includes/rts/Threads.h
+++ b/includes/rts/Threads.h
@@ -43,8 +43,6 @@ StgRegTable * resumeThread  (void *);
 //
 int     cmp_thread                       (StgPtr tso1, StgPtr tso2);
 int     rts_getThreadId                  (StgPtr tso);
-HsInt64 rts_getThreadAllocationCounter   (StgPtr tso);
-void    rts_setThreadAllocationCounter   (StgPtr tso, HsInt64 i);
 void    rts_enableThreadAllocationLimit  (StgPtr tso);
 void    rts_disableThreadAllocationLimit (StgPtr tso);
 
diff --git a/includes/rts/storage/GC.h b/includes/rts/storage/GC.h
index 2aed7c57ee..d4182dd7f9 100644
--- a/includes/rts/storage/GC.h
+++ b/includes/rts/storage/GC.h
@@ -120,7 +120,7 @@ typedef struct generation_ {
     // stats information
     uint32_t collections;
     uint32_t par_collections;
-    uint32_t failed_promotions;
+    uint32_t failed_promotions;         // Currently unused
 
     // ------------------------------------
     // Fields below are used during GC only
diff --git a/includes/stg/MiscClosures.h b/includes/stg/MiscClosures.h
index 76cfbd6c8c..1fbfab9fbe 100644
--- a/includes/stg/MiscClosures.h
+++ b/includes/stg/MiscClosures.h
@@ -468,6 +468,9 @@ RTS_FUN_DECL(stg_traceCcszh);
 RTS_FUN_DECL(stg_clearCCSzh);
 RTS_FUN_DECL(stg_traceEventzh);
 RTS_FUN_DECL(stg_traceMarkerzh);
+RTS_FUN_DECL(stg_getThreadAllocationCounterzh);
+RTS_FUN_DECL(stg_setThreadAllocationCounterzh);
+
 
 /* Other misc stuff */
 // See wiki:Commentary/Compiler/Backends/PprC#Prototypes