[SE] Don't pack raw device mem args

Summary:
Step 4 of getting GlobalDeviceMemory to own its handle.

Take out code to pack untyped device memory types as kernel arguments.
When GlobalDeviceMemory owns its handle, users will never touch untyped
device memory types, so they will never pass them as kernel args.

Reviewers: jlebar

Subscribers: jprice, parallel_libs-commits

Differential Revision: https://reviews.llvm.org/D24177

llvm-svn: 280496

diff --git a/parallel-libs/streamexecutor/include/streamexecutor/PackedKernelArgumentArray.h b/parallel-libs/streamexecutor/include/streamexecutor/PackedKernelArgumentArray.h
index 4d5ca17..ba53ea4 100644 (file)
--- a/parallel-libs/streamexecutor/include/streamexecutor/PackedKernelArgumentArray.h
+++ b/parallel-libs/streamexecutor/include/streamexecutor/PackedKernelArgumentArray.h
@@ -209,36 +209,12 @@ private:
     Types[Index] = KernelArgumentType::GLOBAL_DEVICE_MEMORY;
   }
 
-  // Pack a SharedDeviceMemoryBase argument.
-  void PackOneArgument(size_t Index, const SharedDeviceMemoryBase &Argument) {
-    ++SharedCount;
-    Addresses[Index] = nullptr;
-    Sizes[Index] = Argument.getByteCount();
-    Types[Index] = KernelArgumentType::SHARED_DEVICE_MEMORY;
-  }
-
-  // Pack a SharedDeviceMemoryBase pointer argument.
-  void PackOneArgument(size_t Index, SharedDeviceMemoryBase *Argument) {
-    ++SharedCount;
-    Addresses[Index] = nullptr;
-    Sizes[Index] = Argument->getByteCount();
-    Types[Index] = KernelArgumentType::SHARED_DEVICE_MEMORY;
-  }
-
-  // Pack a const SharedDeviceMemoryBase pointer argument.
-  void PackOneArgument(size_t Index, const SharedDeviceMemoryBase *Argument) {
-    ++SharedCount;
-    Addresses[Index] = nullptr;
-    Sizes[Index] = Argument->getByteCount();
-    Types[Index] = KernelArgumentType::SHARED_DEVICE_MEMORY;
-  }
-
   // Pack a SharedDeviceMemory argument.
   template <typename T>
   void PackOneArgument(size_t Index, const SharedDeviceMemory<T> &Argument) {
     ++SharedCount;
     Addresses[Index] = nullptr;
-    Sizes[Index] = Argument.getByteCount();
+    Sizes[Index] = Argument.getElementCount() * sizeof(T);
     Types[Index] = KernelArgumentType::SHARED_DEVICE_MEMORY;
   }
 
@@ -247,7 +223,7 @@ private:
   void PackOneArgument(size_t Index, SharedDeviceMemory<T> *Argument) {
     ++SharedCount;
     Addresses[Index] = nullptr;
-    Sizes[Index] = Argument->getByteCount();
+    Sizes[Index] = Argument->getElementCount() * sizeof(T);
     Types[Index] = KernelArgumentType::SHARED_DEVICE_MEMORY;
   }
 
@@ -256,7 +232,7 @@ private:
   void PackOneArgument(size_t Index, const SharedDeviceMemory<T> *Argument) {
     ++SharedCount;
     Addresses[Index] = nullptr;
-    Sizes[Index] = Argument->getByteCount();
+    Sizes[Index] = Argument->getElementCount() * sizeof(T);
     Types[Index] = KernelArgumentType::SHARED_DEVICE_MEMORY;
   }
 

diff --git a/parallel-libs/streamexecutor/lib/unittests/PackedKernelArgumentArrayTest.cpp b/parallel-libs/streamexecutor/lib/unittests/PackedKernelArgumentArrayTest.cpp
index 7e87b59..929af07 100644 (file)
--- a/parallel-libs/streamexecutor/lib/unittests/PackedKernelArgumentArrayTest.cpp
+++ b/parallel-libs/streamexecutor/lib/unittests/PackedKernelArgumentArrayTest.cpp
@@ -12,8 +12,11 @@
 ///
 //===----------------------------------------------------------------------===//
 
+#include "SimpleHostPlatformDevice.h"
+#include "streamexecutor/Device.h"
 #include "streamexecutor/DeviceMemory.h"
 #include "streamexecutor/PackedKernelArgumentArray.h"
+#include "streamexecutor/PlatformInterfaces.h"
 
 #include "llvm/ADT/Twine.h"
 
@@ -32,21 +35,19 @@ using Type = se::KernelArgumentType;
 class DeviceMemoryPackingTest : public ::testing::Test {
 public:
   DeviceMemoryPackingTest()
-      : Value(42), Handle(&Value), ByteCount(15), ElementCount(5),
-        UntypedGlobal(Handle, ByteCount),
-        TypedGlobal(se::GlobalDeviceMemory<int>::makeFromElementCount(
-            Handle, ElementCount)),
-        UntypedShared(ByteCount),
+      : Device(&PDevice), Value(42), Handle(&Value), ByteCount(15),
+        ElementCount(5),
+        TypedGlobal(getOrDie(Device.allocateDeviceMemory<int>(ElementCount))),
         TypedShared(
             se::SharedDeviceMemory<int>::makeFromElementCount(ElementCount)) {}
 
+  se::test::SimpleHostPlatformDevice PDevice;
+  se::Device Device;
   int Value;
   void *Handle;
   size_t ByteCount;
   size_t ElementCount;
-  se::GlobalDeviceMemoryBase UntypedGlobal;
   se::GlobalDeviceMemory<int> TypedGlobal;
-  se::SharedDeviceMemoryBase UntypedShared;
   se::SharedDeviceMemory<int> TypedShared;
 };
 
@@ -73,38 +74,18 @@ TEST_F(DeviceMemoryPackingTest, SingleValue) {
   EXPECT_EQ(0u, Array.getSharedCount());
 }
 
-TEST_F(DeviceMemoryPackingTest, SingleUntypedGlobal) {
-  auto Array = se::make_kernel_argument_pack(UntypedGlobal);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 0);
-  EXPECT_EQ(1u, Array.getArgumentCount());
-  EXPECT_EQ(0u, Array.getSharedCount());
-}
-
-TEST_F(DeviceMemoryPackingTest, SingleUntypedGlobalPointer) {
-  auto Array = se::make_kernel_argument_pack(&UntypedGlobal);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 0);
-  EXPECT_EQ(1u, Array.getArgumentCount());
-  EXPECT_EQ(0u, Array.getSharedCount());
-}
-
-TEST_F(DeviceMemoryPackingTest, SingleConstUntypedGlobalPointer) {
-  const se::GlobalDeviceMemoryBase *ConstPointer = &UntypedGlobal;
-  auto Array = se::make_kernel_argument_pack(ConstPointer);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 0);
-  EXPECT_EQ(1u, Array.getArgumentCount());
-  EXPECT_EQ(0u, Array.getSharedCount());
-}
-
 TEST_F(DeviceMemoryPackingTest, SingleTypedGlobal) {
   auto Array = se::make_kernel_argument_pack(TypedGlobal);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 0);
+  ExpectEqual(TypedGlobal.getHandle(), sizeof(void *),
+              Type::GLOBAL_DEVICE_MEMORY, Array, 0);
   EXPECT_EQ(1u, Array.getArgumentCount());
   EXPECT_EQ(0u, Array.getSharedCount());
 }
 
 TEST_F(DeviceMemoryPackingTest, SingleTypedGlobalPointer) {
   auto Array = se::make_kernel_argument_pack(&TypedGlobal);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 0);
+  ExpectEqual(TypedGlobal.getHandle(), sizeof(void *),
+              Type::GLOBAL_DEVICE_MEMORY, Array, 0);
   EXPECT_EQ(1u, Array.getArgumentCount());
   EXPECT_EQ(0u, Array.getSharedCount());
 }
@@ -112,48 +93,24 @@ TEST_F(DeviceMemoryPackingTest, SingleTypedGlobalPointer) {
 TEST_F(DeviceMemoryPackingTest, SingleConstTypedGlobalPointer) {
   const se::GlobalDeviceMemory<int> *ArgumentPointer = &TypedGlobal;
   auto Array = se::make_kernel_argument_pack(ArgumentPointer);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 0);
+  ExpectEqual(TypedGlobal.getHandle(), sizeof(void *),
+              Type::GLOBAL_DEVICE_MEMORY, Array, 0);
   EXPECT_EQ(1u, Array.getArgumentCount());
   EXPECT_EQ(0u, Array.getSharedCount());
 }
 
-TEST_F(DeviceMemoryPackingTest, SingleUntypedShared) {
-  auto Array = se::make_kernel_argument_pack(UntypedShared);
-  ExpectEqual(nullptr, UntypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 0);
-  EXPECT_EQ(1u, Array.getArgumentCount());
-  EXPECT_EQ(1u, Array.getSharedCount());
-}
-
-TEST_F(DeviceMemoryPackingTest, SingleUntypedSharedPointer) {
-  auto Array = se::make_kernel_argument_pack(&UntypedShared);
-  ExpectEqual(nullptr, UntypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 0);
-  EXPECT_EQ(1u, Array.getArgumentCount());
-  EXPECT_EQ(1u, Array.getSharedCount());
-}
-
-TEST_F(DeviceMemoryPackingTest, SingleConstUntypedSharedPointer) {
-  const se::SharedDeviceMemoryBase *ArgumentPointer = &UntypedShared;
-  auto Array = se::make_kernel_argument_pack(ArgumentPointer);
-  ExpectEqual(nullptr, UntypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 0);
-  EXPECT_EQ(1u, Array.getArgumentCount());
-  EXPECT_EQ(1u, Array.getSharedCount());
-}
-
 TEST_F(DeviceMemoryPackingTest, SingleTypedShared) {
   auto Array = se::make_kernel_argument_pack(TypedShared);
-  ExpectEqual(nullptr, TypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 0);
+  ExpectEqual(nullptr, TypedShared.getElementCount() * sizeof(int),
+              Type::SHARED_DEVICE_MEMORY, Array, 0);
   EXPECT_EQ(1u, Array.getArgumentCount());
   EXPECT_EQ(1u, Array.getSharedCount());
 }
 
 TEST_F(DeviceMemoryPackingTest, SingleTypedSharedPointer) {
   auto Array = se::make_kernel_argument_pack(&TypedShared);
-  ExpectEqual(nullptr, TypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 0);
+  ExpectEqual(nullptr, TypedShared.getElementCount() * sizeof(int),
+              Type::SHARED_DEVICE_MEMORY, Array, 0);
   EXPECT_EQ(1u, Array.getArgumentCount());
   EXPECT_EQ(1u, Array.getSharedCount());
 }
@@ -161,42 +118,33 @@ TEST_F(DeviceMemoryPackingTest, SingleTypedSharedPointer) {
 TEST_F(DeviceMemoryPackingTest, SingleConstTypedSharedPointer) {
   const se::SharedDeviceMemory<int> *ArgumentPointer = &TypedShared;
   auto Array = se::make_kernel_argument_pack(ArgumentPointer);
-  ExpectEqual(nullptr, TypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 0);
+  ExpectEqual(nullptr, TypedShared.getElementCount() * sizeof(int),
+              Type::SHARED_DEVICE_MEMORY, Array, 0);
   EXPECT_EQ(1u, Array.getArgumentCount());
   EXPECT_EQ(1u, Array.getSharedCount());
 }
 
 TEST_F(DeviceMemoryPackingTest, PackSeveralArguments) {
-  const se::GlobalDeviceMemoryBase *UntypedGlobalPointer = &UntypedGlobal;
   const se::GlobalDeviceMemory<int> *TypedGlobalPointer = &TypedGlobal;
-  const se::SharedDeviceMemoryBase *UntypedSharedPointer = &UntypedShared;
   const se::SharedDeviceMemory<int> *TypedSharedPointer = &TypedShared;
-  auto Array = se::make_kernel_argument_pack(
-      Value, UntypedGlobal, &UntypedGlobal, UntypedGlobalPointer, TypedGlobal,
-      &TypedGlobal, TypedGlobalPointer, UntypedShared, &UntypedShared,
-      UntypedSharedPointer, TypedShared, &TypedShared, TypedSharedPointer);
+  auto Array = se::make_kernel_argument_pack(Value, TypedGlobal, &TypedGlobal,
+                                             TypedGlobalPointer, TypedShared,
+                                             &TypedShared, TypedSharedPointer);
   ExpectEqual(&Value, sizeof(Value), Type::VALUE, Array, 0);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 1);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 2);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 3);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 4);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 5);
-  ExpectEqual(Handle, sizeof(void *), Type::GLOBAL_DEVICE_MEMORY, Array, 6);
-  ExpectEqual(nullptr, UntypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 7);
-  ExpectEqual(nullptr, UntypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 8);
-  ExpectEqual(nullptr, UntypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 9);
-  ExpectEqual(nullptr, TypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 10);
-  ExpectEqual(nullptr, TypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 11);
-  ExpectEqual(nullptr, TypedShared.getByteCount(), Type::SHARED_DEVICE_MEMORY,
-              Array, 12);
-  EXPECT_EQ(13u, Array.getArgumentCount());
-  EXPECT_EQ(6u, Array.getSharedCount());
+  ExpectEqual(TypedGlobal.getHandle(), sizeof(void *),
+              Type::GLOBAL_DEVICE_MEMORY, Array, 1);
+  ExpectEqual(TypedGlobal.getHandle(), sizeof(void *),
+              Type::GLOBAL_DEVICE_MEMORY, Array, 2);
+  ExpectEqual(TypedGlobal.getHandle(), sizeof(void *),
+              Type::GLOBAL_DEVICE_MEMORY, Array, 3);
+  ExpectEqual(nullptr, TypedShared.getElementCount() * sizeof(int),
+              Type::SHARED_DEVICE_MEMORY, Array, 4);
+  ExpectEqual(nullptr, TypedShared.getElementCount() * sizeof(int),
+              Type::SHARED_DEVICE_MEMORY, Array, 5);
+  ExpectEqual(nullptr, TypedShared.getElementCount() * sizeof(int),
+              Type::SHARED_DEVICE_MEMORY, Array, 6);
+  EXPECT_EQ(7u, Array.getArgumentCount());
+  EXPECT_EQ(3u, Array.getSharedCount());
 }
 
 } // namespace