cmd/compile: insert scheduling checks on loop backedges

Loop breaking with a counter.  Benchmarked (see comments),
eyeball checked for sanity on popular loops.  This code
ought to handle loops in general, and properly inserts phi
functions in cases where the earlier version might not have.

Includes test, plus modifications to test/run.go to deal with
timeout and killing looping test.  Tests broken by the addition
of extra code (branch frequency and live vars) for added
checks turn the check insertion off.

If GOEXPERIMENT=preemptibleloops, the compiler inserts reschedule
checks on every backedge of every reducible loop.  Alternately,
specifying GO_GCFLAGS=-d=ssa/insert_resched_checks/on will
enable it for a single compilation, but because the core Go
libraries contain some loops that may run long, this is less
likely to have the desired effect.

This is intended as a tool to help in the study and diagnosis
of GC and other latency problems, now that goal STW GC latency
is on the order of 100 microseconds or less.

Updates #17831.
Updates #10958.

Change-Id: I6206c163a5b0248e3f21eb4fc65f73a179e1f639
Reviewed-on: https://go-review.googlesource.com/33910
Run-TryBot: David Chase <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Keith Randall <[email protected]>

Author: dr2chase

src/cmd/compile/internal/gc/builtin.go

@@ -15,6 +15,7 @@ var runtimeDecls = [...]struct {
 	{"panicwrap", funcTag, 7},
 	{"gopanic", funcTag, 9},
 	{"gorecover", funcTag, 12},
+	{"goschedguarded", funcTag, 5},
 	{"printbool", funcTag, 14},
 	{"printfloat", funcTag, 16},
 	{"printint", funcTag, 18},

src/cmd/compile/internal/gc/builtin/runtime.go

@@ -21,6 +21,7 @@ func panicwrap(string, string, string)
 
 func gopanic(interface{})
 func gorecover(*int32) interface{}
+func goschedguarded()
 
 func printbool(bool)
 func printfloat(float64)

src/cmd/compile/internal/gc/ssa.go

@@ -64,6 +64,9 @@ func buildssa(fn *Node) *ssa.Func {
 	s.config = initssa()
 	s.f = s.config.NewFunc()
 	s.f.Name = name
+	if fn.Func.Pragma&Nosplit != 0 {
+		s.f.NoSplit = true
+	}
 	s.exitCode = fn.Func.Exit
 	s.panics = map[funcLine]*ssa.Block{}
 	s.config.DebugTest = s.config.DebugHashMatch("GOSSAHASH", name)

src/cmd/compile/internal/ssa/compile.go

@@ -5,6 +5,7 @@
 package ssa
 
 import (
+	"cmd/internal/obj"
 	"fmt"
 	"log"
 	"os"
@@ -349,6 +350,8 @@ var passes = [...]pass{
 	{name: "writebarrier", fn: writebarrier, required: true}, // expand write barrier ops
 	{name: "fuse", fn: fuse},
 	{name: "dse", fn: dse},
+	{name: "insert resched checks", fn: insertLoopReschedChecks,
+		disabled: obj.Preemptibleloops_enabled == 0}, // insert resched checks in loops.
 	{name: "tighten", fn: tighten}, // move values closer to their uses
 	{name: "lower", fn: lower, required: true},
 	{name: "lowered cse", fn: cse},
@@ -378,7 +381,13 @@ type constraint struct {
 }
 
 var passOrder = [...]constraint{
-	// prove reliese on common-subexpression elimination for maximum benefits.
+	// "insert resched checks" uses mem, better to clean out stores first.
+	{"dse", "insert resched checks"},
+	// insert resched checks adds new blocks containing generic instructions
+	{"insert resched checks", "lower"},
+	{"insert resched checks", "tighten"},
+
+	// prove relies on common-subexpression elimination for maximum benefits.
 	{"generic cse", "prove"},
 	// deadcode after prove to eliminate all new dead blocks.
 	{"prove", "generic deadcode"},

src/cmd/compile/internal/ssa/func.go

@@ -24,6 +24,7 @@ type Func struct {
 	vid        idAlloc     // value ID allocator
 
 	scheduled bool // Values in Blocks are in final order
+	NoSplit   bool // true if function is marked as nosplit.  Used by schedule check pass.
 
 	// when register allocation is done, maps value ids to locations
 	RegAlloc []Location