cmd/compile: PGO may cause changes to the execution timing of defer in rangefunc

[wdvxdr1123]

Go version

gotip

What did you do?

package main

import (
	"testing"
)

var a int

func save(x int) {
	a = x
}

func foo() {
	for i := range yield1 {
		defer save(i)
	}
}

func yield1(yield func(int) bool) {
	yield(1)
}

func BenchmarkFoo(b *testing.B) {
	for i := 0; i < b.N; i++ {
		foo()
	}
	if a != 1 {
		b.Fatalf("a = %d; want 1", a)
	}
}

collect profile

go test -bench=Foo -cpuprofile=default.pgo

run pgo optimized bench

go test -pgo=default.pgo -bench=Foo

What did you see happen?

--- FAIL: BenchmarkFoo
    example_test.go:28: a = 0; want 1
FAIL
exit status 1
FAIL    tools   0.430s

[gabyhelp]

Related Issues

• cmd/compile: pgo causes worse performance on some benchmarks #68815
• cmd/compile: pgo after CL 449477 cause worse performance on benchmark #68974 (closed)
• cmd/compile: Building a tool chain using PGO does not achieve the 2-7% improvement claimed by blogs #63407 (closed)
• cmd/compile: PGO crashes with '-d pgoinline=2' #58437 (closed)
• runtime: defer is slow #14939 (closed)
• cmd/compile: poor register allocation with PGO #58298

Related Code Changes

• cmd/compile: don't produce a past-the-end pointer in range loops

_{(Emoji vote if this was helpful or unhelpful; more detailed feedback welcome in this discussion.)}

[gopherbot]

Change https://go.dev/cl/732140 mentions this issue: cmd/compile: disable inlining for functions using runtime.deferrangefunc

[zigo101]

same for 1.25.

For 1.24, error on the collect command instead.

[wdvxdr1123]

For 1.24, error on the collect command instead.

@zigo101 I can reproduce this issue too, but this should be because Go default reads default.pgo. You should delete default.pgo before running.