Files
Jesse Gross 2e036e7cdf mlx, mlxrunner: move the MLX engine out of x/
The MLX runner is the only Go inference runner left and is no longer
experimental, so its packages leave x/. The bindings become a top-level
mlx package beside the carried patches in mlx/compat, mirroring how
llama/ holds the llama.cpp integration, and the runner becomes mlxrunner
with the architectures nested under the package they implement.
Subpackages move with their parent unless listed.

  x/mlxrunner/mlx            mlx
  x/internal/mlxthread       mlx/mlxthread
  x/internal/mlxthreadtest   mlx/mlxthread/mlxthreadtest
  x/internal/mlxtest         mlx/mlxtest
  x/quant                    mlx/quant
  mlx/compat/*.patch         mlx/compat/mlx-c   (MLX patches go in mlx/compat/mlx)
  x/mlxrunner                mlxrunner
  x/models/nn                mlxrunner/nn
  x/models/<arch>            mlxrunner/model/<arch>
  x/mlxrunner/imports.go     mlxrunner/model/architectures   (new package)
  x/create                   create
  x/safetensors              fs/safetensors
  x/tokenizer                mlxrunner/tokenizer

Every package keeps its name, so the Go changes are the import path
rewrites the moves force, and the CMake, Dockerfile, CI cache keys, drift
check and Darwin payload script follow the new paths. Four edits are not
paths: the runner's blank architecture imports become the package
mlxrunner/model/architectures, so the list to extend for a new model sits
beside the architecture directories; a depguard rule keeps the two test
harnesses out of non-test code, as the x/internal placement used to; the
CI change filter's two entries for the long-deleted x/imagegen/mlx now
name the bindings' CMake project and the carried patches, so a change to
either builds the payload; and the tokenizer parity test reads its
fixtures from its own testdata instead of walking out of x/.

x/server and x/imagegen/manifest stay for the next two commits.
2026-09-16 14:06:08 -07:00

361 lines
7.4 KiB
Go

package mlxthread
import (
"context"
"errors"
"reflect"
"runtime"
"strings"
"sync"
"sync/atomic"
"testing"
"time"
)
func TestDoRunsInOrder(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
defer thread.Stop(context.Background(), nil)
var got []int
for i := 0; i < 5; i++ {
i := i
if err := thread.Do(context.Background(), func() error {
got = append(got, i)
return nil
}); err != nil {
t.Fatal(err)
}
}
if want := []int{0, 1, 2, 3, 4}; !reflect.DeepEqual(got, want) {
t.Fatalf("got %v, want %v", got, want)
}
}
func TestDoPropagatesPanicToCaller(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
defer thread.Stop(context.Background(), nil)
defer func() {
got := recover()
pe, ok := got.(*panicError)
if !ok {
t.Fatalf("got panic %T (%v), want *panicError", got, got)
}
if pe.value != "boom" {
t.Fatalf("got panic value %v, want boom", pe.value)
}
if !strings.Contains(string(pe.stack), "TestDoPropagatesPanicToCaller") {
t.Fatalf("captured stack missing original frame:\n%s", pe.stack)
}
}()
_ = thread.Do(context.Background(), func() error {
panic("boom")
})
}
func TestDoCancelsBeforeJobStarts(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
defer thread.Stop(context.Background(), nil)
running := make(chan struct{})
release := make(chan struct{})
errCh := make(chan error, 1)
go func() {
errCh <- thread.Do(context.Background(), func() error {
close(running)
<-release
return nil
})
}()
<-running
ctx, cancel := context.WithCancel(context.Background())
cancel()
err = thread.Do(ctx, func() error {
t.Fatal("canceled job should not run")
return nil
})
if !errors.Is(err, context.Canceled) {
t.Fatalf("got %v, want %v", err, context.Canceled)
}
close(release)
if err := <-errCh; err != nil {
t.Fatal(err)
}
}
func TestAlreadyCanceledContextDoesNotEnqueue(t *testing.T) {
t.Run("Do", func(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
defer thread.Stop(context.Background(), nil)
ctx, cancel := context.WithCancel(context.Background())
cancel()
ran := false
err = thread.Do(ctx, func() error {
ran = true
return nil
})
if !errors.Is(err, context.Canceled) {
t.Fatalf("got %v, want %v", err, context.Canceled)
}
if ran {
t.Fatal("canceled job ran")
}
})
t.Run("Stop", func(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
defer thread.Stop(context.Background(), nil)
ctx, cancel := context.WithCancel(context.Background())
cancel()
cleaned := false
err = thread.Stop(ctx, func() {
cleaned = true
})
if !errors.Is(err, context.Canceled) {
t.Fatalf("got %v, want %v", err, context.Canceled)
}
if cleaned {
t.Fatal("cleanup ran for canceled stop")
}
if err := thread.Do(context.Background(), func() error { return nil }); err != nil {
t.Fatalf("thread did not accept work after canceled Stop: %v", err)
}
})
}
func TestCallReturnsValue(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
defer thread.Stop(context.Background(), nil)
got, err := Call(context.Background(), thread, func() (int, error) {
return 42, nil
})
if err != nil {
t.Fatal(err)
}
if got != 42 {
t.Fatalf("got %d, want 42", got)
}
}
func TestDoRunsConcurrentlySubmittedWorkSerially(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
defer thread.Stop(context.Background(), nil)
oldProcs := runtime.GOMAXPROCS(8)
defer runtime.GOMAXPROCS(oldProcs)
const goroutines = 16
const iterations = 64
var active atomic.Int32
var count atomic.Int64
var wg sync.WaitGroup
errCh := make(chan error, goroutines)
for range goroutines {
wg.Add(1)
go func() {
defer wg.Done()
for range iterations {
if err := thread.Do(context.Background(), func() error {
if got := active.Add(1); got != 1 {
return errors.New("thread executed jobs concurrently")
}
runtime.Gosched()
count.Add(1)
if got := active.Add(-1); got != 0 {
return errors.New("thread active count did not return to zero")
}
return nil
}); err != nil {
errCh <- err
return
}
}
}()
}
wg.Wait()
close(errCh)
for err := range errCh {
t.Fatal(err)
}
if got, want := count.Load(), int64(goroutines*iterations); got != want {
t.Fatalf("got %d jobs, want %d", got, want)
}
}
func TestStopRunsCleanupAndRejectsWork(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
cleaned := 0
if err := thread.Stop(context.Background(), func() {
cleaned++
}); err != nil {
t.Fatal(err)
}
if cleaned != 1 {
t.Fatalf("cleanup ran %d times, want 1", cleaned)
}
if err := thread.Stop(context.Background(), func() {
cleaned++
}); err != nil {
t.Fatal(err)
}
if cleaned != 1 {
t.Fatalf("cleanup ran %d times after second Stop, want 1", cleaned)
}
err = thread.Do(context.Background(), func() error {
t.Fatal("job should not run after stop")
return nil
})
if !errors.Is(err, ErrStopped) {
t.Fatalf("got %v, want %v", err, ErrStopped)
}
}
func TestStopCanceledBeforeEnqueueCanBeRetried(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
defer thread.Stop(context.Background(), nil)
running := make(chan struct{})
release := make(chan struct{})
errCh := make(chan error, 1)
go func() {
errCh <- thread.Do(context.Background(), func() error {
close(running)
<-release
return nil
})
}()
<-running
ctx, cancel := context.WithTimeout(context.Background(), 10*time.Millisecond)
defer cancel()
cleanupRan := false
err = thread.Stop(ctx, func() {
cleanupRan = true
})
if !errors.Is(err, context.DeadlineExceeded) {
t.Fatalf("got %v, want %v", err, context.DeadlineExceeded)
}
if cleanupRan {
t.Fatal("cleanup ran even though stop was not enqueued")
}
close(release)
if err := <-errCh; err != nil {
t.Fatal(err)
}
if err := thread.Do(context.Background(), func() error { return nil }); err != nil {
t.Fatalf("thread did not accept work after canceled Stop: %v", err)
}
cleanupRan = false
if err := thread.Stop(context.Background(), func() {
cleanupRan = true
}); err != nil {
t.Fatal(err)
}
if !cleanupRan {
t.Fatal("cleanup did not run on retried Stop")
}
}
func TestStopWaitsForActiveWorkBeforeCleanup(t *testing.T) {
thread, err := Start("test", nil)
if err != nil {
t.Fatal(err)
}
running := make(chan struct{})
release := make(chan struct{})
jobErr := make(chan error, 1)
go func() {
jobErr <- thread.Do(context.Background(), func() error {
close(running)
<-release
return nil
})
}()
<-running
cleaned := make(chan struct{})
stopErr := make(chan error, 1)
go func() {
stopErr <- thread.Stop(context.Background(), func() {
close(cleaned)
})
}()
select {
case <-cleaned:
t.Fatal("cleanup ran before active job completed")
case <-time.After(10 * time.Millisecond):
}
err = thread.Do(context.Background(), func() error {
return errors.New("work should be rejected once Stop starts")
})
if !errors.Is(err, ErrStopped) {
t.Fatalf("got %v, want %v", err, ErrStopped)
}
close(release)
if err := <-jobErr; err != nil {
t.Fatal(err)
}
if err := <-stopErr; err != nil {
t.Fatal(err)
}
select {
case <-cleaned:
default:
t.Fatal("cleanup did not run")
}
}