trace-with-dedupe-benchmark-results.txt

tile38 $ benchstat tip+fpcallers.results tip+fpcallers+trace.results
name                                old time/op            new time/op            delta
Tile38WithinCircle100kmRequest                 490µs ± 1%             494µs ± 1%  +0.79%  (p=0.019 n=10+10)
Tile38IntersectsCircle100kmRequest             566µs ± 1%             571µs ± 2%  +0.96%  (p=0.011 n=10+10)
Tile38KNearestLimit100Request                  391µs ± 2%             394µs ± 1%  +0.91%  (p=0.004 n=9+10)

name                                old average-RSS-bytes  new average-RSS-bytes  delta
Tile38WithinCircle100kmRequest                5.38GB ± 1%            5.37GB ± 1%    ~     (p=0.529 n=10+10)
Tile38IntersectsCircle100kmRequest            5.71GB ± 1%            5.70GB ± 1%    ~     (p=0.631 n=10+10)
Tile38KNearestLimit100Request                 7.26GB ± 0%            7.26GB ± 0%    ~     (p=0.529 n=10+10)

name                                old peak-RSS-bytes     new peak-RSS-bytes     delta
Tile38WithinCircle100kmRequest                5.71GB ± 1%            5.70GB ± 1%    ~     (p=0.579 n=10+10)
Tile38IntersectsCircle100kmRequest            5.94GB ± 1%            5.94GB ± 1%    ~     (p=0.684 n=10+10)
Tile38KNearestLimit100Request                 7.49GB ± 0%            7.50GB ± 0%    ~     (p=0.720 n=9+10)

name                                old peak-VM-bytes      new peak-VM-bytes      delta
Tile38WithinCircle100kmRequest                6.42GB ± 0%            6.40GB ± 1%    ~     (p=0.711 n=6+10)
Tile38IntersectsCircle100kmRequest            6.65GB ± 1%            6.64GB ± 1%    ~     (p=0.541 n=10+10)
Tile38KNearestLimit100Request                 8.18GB ± 0%            8.18GB ± 0%    ~     (p=0.245 n=9+8)

name                                old p50-latency-ns     new p50-latency-ns     delta
Tile38WithinCircle100kmRequest                  119k ± 2%              122k ± 1%  +3.20%  (p=0.000 n=10+10)
Tile38IntersectsCircle100kmRequest              173k ± 1%              176k ± 1%  +1.43%  (p=0.004 n=10+9)
Tile38KNearestLimit100Request                   285k ± 2%              289k ± 1%  +1.51%  (p=0.000 n=10+10)

name                                old p90-latency-ns     new p90-latency-ns     delta
Tile38WithinCircle100kmRequest                  722k ± 1%              725k ± 1%    ~     (p=0.315 n=10+9)
Tile38IntersectsCircle100kmRequest              836k ± 2%              838k ± 1%    ~     (p=0.631 n=10+10)
Tile38KNearestLimit100Request                   741k ± 2%              745k ± 1%    ~     (p=0.190 n=10+10)

name                                old p99-latency-ns     new p99-latency-ns     delta
Tile38WithinCircle100kmRequest                 5.75M ± 2%             5.75M ± 2%    ~     (p=0.971 n=10+10)
Tile38IntersectsCircle100kmRequest             5.83M ± 2%             5.91M ± 3%    ~     (p=0.089 n=10+10)
Tile38KNearestLimit100Request                  1.44M ± 3%             1.45M ± 2%    ~     (p=0.684 n=10+10)

name                                old ops/s              new ops/s              delta
Tile38WithinCircle100kmRequest                 12.2k ± 1%             12.1k ± 1%  -0.79%  (p=0.019 n=10+10)
Tile38IntersectsCircle100kmRequest             10.6k ± 1%             10.5k ± 2%  -0.80%  (p=0.022 n=9+10)
Tile38KNearestLimit100Request                  15.4k ± 2%             15.2k ± 1%  -0.91%  (p=0.004 n=9+10)

trace-with-dedupe-diff.txt

diff --git a/src/runtime/runtime2.go b/src/runtime/runtime2.go
index 2e98f895a3..b69da1be9c 100644
--- a/src/runtime/runtime2.go
+++ b/src/runtime/runtime2.go
@@ -472,6 +472,7 @@ type g struct {
 
        raceignore     int8     // ignore race detection events
        sysblocktraced bool     // StartTrace has emitted EvGoInSyscall about this goroutine
+       traceLastEv    byte     // Event type of the last event traced by this goroutine
        tracking       bool     // whether we're tracking this G for sched latency statistics
        trackingSeq    uint8    // used to decide whether to track this G
        trackingStamp  int64    // timestamp of when the G last started being tracked
@@ -495,6 +496,7 @@ type g struct {
        labels         unsafe.Pointer // profiler labels
        timer          *timer         // cached timer for time.Sleep
        selectDone     atomic.Uint32  // are we participating in a select and did someone win the race?
+       traceLastStack uint32         // stack ID for the last traced event on this G
 
        // goroutineProfiled indicates the status of this goroutine's stack for the
        // current in-progress goroutine profile
diff --git a/src/runtime/trace.go b/src/runtime/trace.go
index 7484c45f71..122b0aca2f 100644
--- a/src/runtime/trace.go
+++ b/src/runtime/trace.go
@@ -239,7 +239,7 @@ func StartTrace() error {
 
        // Obtain current stack ID to use in all traceEvGoCreate events below.
        stkBuf := make([]uintptr, traceStackSize)
-       stackID := traceStackID(mp, stkBuf, 2)
+       stackID := uint64(traceStackID(mp, stkBuf, 2))
 
        profBuf := newProfBuf(2, profBufWordCount, profBufTagCount) // after the timestamp, header is [pp.id, gp.goid]
        trace.cpuLogRead = profBuf
@@ -753,7 +753,34 @@ func traceEventLocked(extraBytes int, mp *m, pid int32, bufp *traceBufPtr, ev by
        } else if skip == 0 {
                buf.varint(0)
        } else if skip > 0 {
-               buf.varint(traceStackID(mp, buf.stk[:], skip))
+               if mp != nil && mp.curg != nil {
+                       lastEv := mp.curg.traceLastEv
+                       switch ev {
+                       case traceEvGoStart, traceEvGoStartLocal:
+                               switch lastEv {
+                               case traceEvGoStop,
+                                       traceEvGoSched,
+                                       traceEvGoPreempt,
+                                       traceEvGoSleep,
+                                       traceEvGoBlock,
+                                       traceEvGoBlockSend,
+                                       traceEvGoBlockRecv,
+                                       traceEvGoBlockSelect,
+                                       traceEvGoBlockSync,
+                                       traceEvGoBlockCond,
+                                       traceEvGoBlockNet,
+                                       traceEvGoSysCall:
+                                       stackID = mp.curg.traceLastStack
+                               default:
+                                       stackID = traceStackID(mp, buf.stk[:], skip)
+                               }
+                       default:
+                               stackID = traceStackID(mp, buf.stk[:], skip)
+                       }
+               } else {
+                       stackID = traceStackID(mp, buf.stk[:], skip)
+               }
+               buf.varint(uint64(stackID))
        }
        evSize := buf.pos - startPos
        if evSize > maxSize {
@@ -763,6 +790,10 @@ func traceEventLocked(extraBytes int, mp *m, pid int32, bufp *traceBufPtr, ev by
                // Fill in actual length.
                *lenp = byte(evSize - 2)
        }
+       if mp != nil && mp.curg != nil {
+               mp.curg.traceLastEv = ev
+               mp.curg.traceLastStack = stackID
+       }
 }
 
 // traceCPUSample writes a CPU profile sample stack to the execution tracer's
@@ -877,7 +908,7 @@ const logicalStackSentinel = ^uintptr(0)
 // stack table, and returns its unique ID. pcBuf should have a length equal to
 // traceStackSize. skip controls the number of leaf frames to omit in order to
 // hide tracer internals from stack traces, see CL 5523.
-func traceStackID(mp *m, pcBuf []uintptr, skip int) uint64 {
+func traceStackID(mp *m, pcBuf []uintptr, skip int) uint32 {
        gp := getg()
        curgp := mp.curg
        nstk := 1
@@ -916,7 +947,7 @@ func traceStackID(mp *m, pcBuf []uintptr, skip int) uint64 {
                nstk-- // skip runtime.main
        }
        id := trace.stackTab.put(pcBuf[:nstk])
-       return uint64(id)
+       return id
 }
 
 // tracefpunwindoff returns false if frame pointer unwinding for the tracer is