runtime: less aggressive per-thread stack segment caching
Introduce global stack segment cache and limit per-thread cache size.
This greatly reduces StackSys memory on workloads that create lots of threads.
benchmark old ns/op new ns/op delta
BenchmarkStackGrowth 665 656 -1.35%
BenchmarkStackGrowth-2 333 328 -1.50%
BenchmarkStackGrowth-4 224 172 -23.21%
BenchmarkStackGrowth-8 124 91 -26.13%
BenchmarkStackGrowth-16 82 47 -41.94%
BenchmarkStackGrowth-32 73 40 -44.79%
BenchmarkStackGrowthDeep 97231 94391 -2.92%
BenchmarkStackGrowthDeep-2 47230 58562 +23.99%
BenchmarkStackGrowthDeep-4 24993 49356 +97.48%
BenchmarkStackGrowthDeep-8 15105 30072 +99.09%
BenchmarkStackGrowthDeep-16 10005 15623 +56.15%
BenchmarkStackGrowthDeep-32 12517 13069 +4.41%
TestStackMem#1,MB 310 12 -96.13%
TestStackMem#2,MB 296 14 -95.27%
TestStackMem#3,MB 479 14 -97.08%
TestStackMem#1,sec 3.22 2.26 -29.81%
TestStackMem#2,sec 2.43 2.15 -11.52%
TestStackMem#3,sec 2.50 2.38 -4.80%
R=sougou, no.smile.face, rsc
CC=golang-dev, msolomon
https://golang.org/cl/7029044
diff --git a/src/pkg/runtime/runtime.h b/src/pkg/runtime/runtime.h
index a228b06..47a7b6e 100644
--- a/src/pkg/runtime/runtime.h
+++ b/src/pkg/runtime/runtime.h
@@ -128,6 +128,13 @@
{
PtrSize = sizeof(void*),
};
+enum
+{
+ // Per-M stack segment cache size.
+ StackCacheSize = 32,
+ // Global <-> per-M stack segment cache transfer batch size.
+ StackCacheBatch = 16,
+};
/*
* structures
@@ -262,7 +269,10 @@
M* schedlink;
uint32 machport; // Return address for Mach IPC (OS X)
MCache *mcache;
- FixAlloc *stackalloc;
+ int32 stackinuse;
+ uint32 stackcachepos;
+ uint32 stackcachecnt;
+ void* stackcache[StackCacheSize];
G* lockedg;
G* idleg;
uintptr createstack[32]; // Stack that created this thread.
