update cuda tests for -sync flag, comment out unified slice

cuda/buffer_test.go

@@ -6,7 +6,7 @@ import (
 )
 
 func init() {
-	Init(0, "auto")
+	Init(0, "auto", false)
 }
 
 func TestBuffer(t *testing.T) {

cuda/reduce_test.go

@@ -13,7 +13,7 @@ func initTest() {
 		return
 	}
 	LockThread()
-	Init(0, "auto")
+	Init(0, "auto", false)
 	{
 		inh1 := make([]float32, 1000)
 		for i := range inh1 {

cuda/slice.go

@@ -14,9 +14,9 @@ func NewSlice(nComp int, m *data.Mesh) *data.Slice {
 }
 
 // Make a GPU Slice with nComp components each of size length.
-func NewUnifiedSlice(nComp int, m *data.Mesh) *data.Slice {
-	return newSlice(nComp, m, cu.MemAllocHost, data.UnifiedMemory)
-}
+//func NewUnifiedSlice(nComp int, m *data.Mesh) *data.Slice {
+//	return newSlice(nComp, m, cu.MemAllocHost, data.UnifiedMemory)
+//}
 
 func newSlice(nComp int, m *data.Mesh, alloc func(int64) unsafe.Pointer, memType int8) *data.Slice {
 	data.EnableGPU(memFree, cu.MemFreeHost, memCpy, memCpyDtoH, memCpyHtoD)
@@ -47,16 +47,13 @@ func memCpyHtoD(dst, src unsafe.Pointer, bytes int64) {
 }
 
 func memCpy(dst, src unsafe.Pointer, bytes int64) {
-	if synchronous { // debug
-		Sync()
-	}
+	Sync()
 	cu.MemcpyAsync(cu.DevicePtr(uintptr(dst)), cu.DevicePtr(uintptr(src)), bytes, stream0)
-	if synchronous { // debug
-		Sync()
-	}
+	Sync()
 }
 
 // Memset sets the Slice's components to the specified values.
+// To be carefully used on unified slice (need sync)
 func Memset(s *data.Slice, val ...float32) {
 	if synchronous { // debug
 		Sync()

cuda/slice_test.go

@@ -6,7 +6,7 @@ import (
 )
 
 func init() {
-	Init(0, "auto")
+	Init(0, "auto", false)
 }
 
 func TestSlice(t *testing.T) {
@@ -93,18 +93,18 @@ func TestSliceHost(t *testing.T) {
 	N0, N1, N2 := 1, 10, 10
 	c := 1e-6
 	m := data.NewMesh(N0, N1, N2, c, c, c)
-	a := NewUnifiedSlice(3, m)
+	a := NewSlice(3, m)
 	defer a.Free()
 
-	b := a.Host()
+	b := a.HostCopy().Host()
 	if b[0][0] != 0 || b[1][42] != 0 || b[2][99] != 0 {
-		t.Fail()
+		t.Error("slice not inited to zero")
 	}
 
 	Memset(a, 1, 2, 3)
-	b = a.Host()
+	b = a.HostCopy().Host()
 	if b[0][0] != 1 || b[1][42] != 2 || b[2][99] != 3 {
-		t.Fail()
+		t.Error("slice memset")
 	}
 }
 

data/slice.go

@@ -11,7 +11,6 @@ import (
 )
 
 // Slice is like a [][]float32, but may be stored in GPU or host memory.
-// TODO: unified memory is not used anymore, can be removed. Then we can split cuda.Slice and data.Slice?
 type Slice struct {
 	ptr_    [MAX_COMP]unsafe.Pointer // keeps data local // TODO: rm (premature optimization)
 	ptrs    []unsafe.Pointer         // points into ptr_