wip: explore accumulated markers

Author: not-matthias

example/helper.go

@@ -0,0 +1,36 @@
+package example
+
+import "time"
+
+//go:noinline
+func recursiveFib(n int) uint64 {
+	if n <= 1 {
+		return uint64(n)
+	}
+	return recursiveFib(n-1) + recursiveFib(n-2)
+}
+
+//go:noinline
+func expensiveOperation() uint64 {
+	// Large memory allocation
+	data := make([]uint64, 1024*1024) // 8 MiB allocation
+	for i := range data {
+		data[i] = 42
+	}
+
+	// Expensive recursive computation that will dominate flamegraph
+	fibResult := recursiveFib(30)
+
+	// More expensive work - sum the data
+	sum := uint64(0)
+	for _, v := range data {
+		sum += v
+	}
+
+	return sum + fibResult
+}
+
+func actualWork() uint64 {
+	time.Sleep(1 * time.Millisecond)
+	return 42
+}

example/timing_test.go

@@ -0,0 +1,27 @@
+package example
+
+import (
+	"runtime"
+	"testing"
+)
+
+func BenchmarkLargeSetup(b *testing.B) {
+	res := expensiveOperation()
+	b.ResetTimer()
+
+	for i := 0; i < b.N; i++ {
+		actualWork()
+	}
+
+	runtime.KeepAlive(res)
+}
+
+func BenchmarkLargeSetupInLoop(b *testing.B) {
+	for i := 0; i < b.N; i++ {
+		b.StopTimer()
+		expensiveOperation()
+		b.StartTimer()
+
+		actualWork()
+	}
+}

testing/testing/benchmark.go

@@ -91,6 +91,10 @@ type codspeed struct {
 	codspeedTimePerRoundNs []time.Duration
 	codspeedItersPerRound  []int64
 	startTimestamp         uint64
+
+	// Accumulated timestamp pairs for batch submission
+	accumulatedStartTimestamps []uint64
+	accumulatedEndTimestamps   []uint64
 }
 
 // B is a type passed to [Benchmark] functions to manage benchmark
@@ -176,14 +180,17 @@ func (b *B) StopTimer() {
 		// b.loop.i |= loopPoisonTimer
 
 		endTimestamp := capi.CurrentTimestamp()
-		b.codspeed.instrument_hooks.AddBenchmarkTimestamps(b.codspeed.startTimestamp, endTimestamp)
+
+		// Accumulate timestamps for batch submission instead of immediate hook call
+		b.codspeed.accumulatedStartTimestamps = append(b.codspeed.accumulatedStartTimestamps, b.codspeed.startTimestamp)
+		b.codspeed.accumulatedEndTimestamps = append(b.codspeed.accumulatedEndTimestamps, endTimestamp)
 
 		// For b.N loops: This will be called in runN which sets b.N to the number of iterations.
 		// For b.Loop() loops: loopSlowPath sets b.N to 0 to prevent b.N loops within b.Loop. However, since
 		// we're starting/stopping the timer for each iteration in the b.Loop() loop, we can use 1 as
 		// the number of iterations for this round.
-		b.codspeedItersPerRound = append(b.codspeedItersPerRound, max(int64(b.N), 1))
-		b.codspeedTimePerRoundNs = append(b.codspeedTimePerRoundNs, highPrecisionTimeSince(b.start))
+		b.codspeed.codspeedItersPerRound = append(b.codspeed.codspeedItersPerRound, max(int64(b.N), 1))
+		b.codspeed.codspeedTimePerRoundNs = append(b.codspeed.codspeedTimePerRoundNs, highPrecisionTimeSince(b.start))
 	}
 }
 
@@ -207,6 +214,21 @@ func (b *B) ResetTimer() {
 	b.duration = 0
 	b.netAllocs = 0
 	b.netBytes = 0
+
+	// Clear accumulated CodSpeed timestamp data
+	b.codspeed.codspeedItersPerRound = b.codspeed.codspeedItersPerRound[:0]
+	b.codspeed.codspeedTimePerRoundNs = b.codspeed.codspeedTimePerRoundNs[:0]
+	b.codspeed.accumulatedStartTimestamps = b.codspeed.accumulatedStartTimestamps[:0]
+	b.codspeed.accumulatedEndTimestamps = b.codspeed.accumulatedEndTimestamps[:0]
+}
+
+func (b *B) submitAccumulatedTimestamps() {
+	for i := 0; i < len(b.codspeed.accumulatedStartTimestamps); i++ {
+		b.codspeed.instrument_hooks.AddBenchmarkTimestamps(
+			b.codspeed.accumulatedStartTimestamps[i],
+			b.codspeed.accumulatedEndTimestamps[i],
+		)
+	}
 }
 
 // SetBytes records the number of bytes processed in a single operation.
@@ -391,8 +413,10 @@ func (b *B) launch() {
 			}
 
 			// Reset the fields from the warmup run
-			b.codspeedItersPerRound = make([]int64, 0)
-			b.codspeedTimePerRoundNs = make([]time.Duration, 0)
+			b.codspeed.codspeedItersPerRound = make([]int64, 0)
+			b.codspeed.codspeedTimePerRoundNs = make([]time.Duration, 0)
+			b.codspeed.accumulatedStartTimestamps = make([]uint64, 0)
+			b.codspeed.accumulatedEndTimestamps = make([]uint64, 0)
 
 			// Final run:
 			benchD := time.Second * b.benchTime.d
@@ -419,7 +443,11 @@ func (b *B) launch() {
 			b.codspeed.instrument_hooks.StopBenchmark()
 		}
 	}
-	b.result = BenchmarkResult{b.N, b.duration, b.bytes, b.netAllocs, b.netBytes, b.codspeedTimePerRoundNs, b.codspeedItersPerRound, b.extra}
+
+	// Submit all accumulated timestamps to CodSpeed hooks before finalizing results
+	b.submitAccumulatedTimestamps()
+
+	b.result = BenchmarkResult{b.N, b.duration, b.bytes, b.netAllocs, b.netBytes, b.codspeed.codspeedTimePerRoundNs, b.codspeed.codspeedItersPerRound, b.extra}
 }
 
 // Elapsed returns the measured elapsed time of the benchmark.
@@ -458,6 +486,10 @@ func (b *B) stopOrScaleBLoop() bool {
 		// Stop the timer so we don't count cleanup time
 		b.StopTimer()
 		b.codspeed.instrument_hooks.StopBenchmark()
+
+		// Submit all accumulated timestamps to CodSpeed hooks before benchmark completion
+		b.submitAccumulatedTimestamps()
+
 		// Commit iteration count
 		b.N = int(b.loop.n)
 		b.loop.done = true
@@ -496,8 +528,10 @@ func (b *B) loopSlowPath() bool {
 		b.N = 0
 		b.loop.i++
 
-		b.codspeedItersPerRound = make([]int64, 0)
-		b.codspeedTimePerRoundNs = make([]time.Duration, 0)
+		b.codspeed.codspeedItersPerRound = make([]int64, 0)
+		b.codspeed.codspeedTimePerRoundNs = make([]time.Duration, 0)
+		b.codspeed.accumulatedStartTimestamps = make([]uint64, 0)
+		b.codspeed.accumulatedEndTimestamps = make([]uint64, 0)
 
 		b.codspeed.instrument_hooks.StartBenchmark()
 		b.ResetTimer()
@@ -515,6 +549,10 @@ func (b *B) loopSlowPath() bool {
 		}
 		b.StopTimer()
 		b.codspeed.instrument_hooks.StopBenchmark()
+
+		// Submit all accumulated timestamps to CodSpeed hooks before benchmark completion
+		b.submitAccumulatedTimestamps()
+
 		// Commit iteration count
 		b.N = int(b.loop.n)
 		b.loop.done = true
@@ -900,8 +938,8 @@ func (s *benchState) processBench(b *B) {
 				Name:                   benchName,
 				Uri:                    benchUri,
 				Pid:                    os.Getpid(),
-				CodspeedTimePerRoundNs: r.CodspeedTimePerRoundNs,
-				CodspeedItersPerRound:  r.CodspeedItersPerRound,
+				CodspeedTimePerRoundNs: r.codspeedTimePerRoundNs,
+				CodspeedItersPerRound:  r.codspeedItersPerRound,
 			}
 
 			goRunnerMetadata, err := findGoRunnerMetadata()