Dynamic heap count (#86245)

This is an initial implementation for changing the GC heap count dynamically in response to changing load conditions.

Using more heaps will increase memory footprint, but in most cases also improve throughput because more GC work is parallelized, and lock contention on the allocation code path is reduced by spreading the load.

We try to keep the measured overhead from lock contention and GC pauses at 5% or below, and only reduce the heap count if the measured overhead is below 1% and we estimate a significant reduction in memory footprint. There is some more fine tuning for the cases where smaller reductions in either footprint or overhead are possible without impacting the other side of the equation too much.

Because the input data for GC pause etc. are quite noisy, we use a median of 3 filter before the data is used to make decisions. Preliminary data suggests this is effective, but we may need to do more filtering if gains in either direction appear to be small.

Author: PeterSolMS

src/coreclr/gc/gc.cpp

@@ -137,7 +137,8 @@ class GCConfigStringHolder
     INT_CONFIG   (GCConserveMem,             "GCConserveMemory",          "System.GC.ConserveMemory",          0,                  "Specifies how hard GC should try to conserve memory - values 0-9")                       \
     INT_CONFIG   (GCWriteBarrier,            "GCWriteBarrier",            NULL,                                0,                  "Specifies whether GC should use more precise but slower write barrier")                  \
     STRING_CONFIG(GCName,                    "GCName",                    "System.GC.Name",                                        "Specifies the path of the standalone GC implementation.")                                \
-    INT_CONFIG   (GCSpinCountUnit,           "GCSpinCountUnit",           0,                                   0,                  "Specifies the spin count unit used by the GC.")
+    INT_CONFIG   (GCSpinCountUnit,           "GCSpinCountUnit",           0,                                   0,                  "Specifies the spin count unit used by the GC.")                                          \
+    BOOL_CONFIG  (GCDynamicAdaptation,       "GCDynamicAdaptation",       NULL,                                false,              "Enables varying the heap count dynamically in Server GC.")
 // This class is responsible for retreiving configuration information
 // for how the GC should operate.
 class GCConfig

src/coreclr/gc/gcconfig.h

@@ -1068,6 +1068,7 @@ struct ScanContext
 {
     Thread* thread_under_crawl;
     int thread_number;
+    int thread_count;
     uintptr_t stack_limit; // Lowest point on the thread stack that the scanning logic is permitted to read
     bool promotion; //TRUE: Promotion, FALSE: Relocation.
     bool concurrent; //TRUE: concurrent scanning
@@ -1085,6 +1086,7 @@ struct ScanContext
 
         thread_under_crawl = 0;
         thread_number = -1;
+        thread_count = -1;
         stack_limit = 0;
         promotion = false;
         concurrent = false;