[SYCL] Add queue shortcut functions with placeholder accessors (#7266)

Andrew Lamzed-Short · web-flow · commit 5ee066e4b638 · 2022-11-14T09:46:11.000+01:00
Implementation of the explicit copy-based queue shortcut functions defined in Table 29 of [Section 4.6.5.2 of the specification](https://registry.khronos.org/SYCL/specs/sycl-2020/html/sycl-2020.html#sec:queue-shortcuts) (the last 7 rows of the table to be precise).
diff --git a/sycl/include/sycl/queue.hpp b/sycl/include/sycl/queue.hpp
@@ -1133,6 +1133,124 @@ class __SYCL_EXPORT queue {
         CodeLoc);
   }
 
+  /// Copies data from a memory region pointed to by a placeholder accessor to
+  /// another memory region pointed to by a shared_ptr.
+  ///
+  /// \param Src is a placeholder accessor to the source memory.
+  /// \param Dest is a shared_ptr to the destination memory.
+  /// \return an event representing copy operation.
+  template <typename SrcT, int SrcDims, access_mode SrcMode, target SrcTgt,
+            access::placeholder IsPlaceholder, typename DestT>
+  event copy(accessor<SrcT, SrcDims, SrcMode, SrcTgt, IsPlaceholder> Src,
+             std::shared_ptr<DestT> Dest _CODELOCPARAM(&CodeLoc)) {
+    return submit([&](handler &CGH) {
+      CGH.require(Src);
+      CGH.copy(Src, Dest);
+    } _CODELOCFW(CodeLoc));
+  }
+
+  /// Copies data from a memory region pointed to by a shared_ptr to another
+  /// memory region pointed to by a placeholder accessor.
+  ///
+  /// \param Src is a shared_ptr to the source memory.
+  /// \param Dest is a placeholder accessor to the destination memory.
+  /// \return an event representing copy operation.
+  template <typename SrcT, typename DestT, int DestDims, access_mode DestMode,
+            target DestTgt, access::placeholder IsPlaceholder>
+  event copy(std::shared_ptr<SrcT> Src,
+             accessor<DestT, DestDims, DestMode, DestTgt, IsPlaceholder> Dest
+                 _CODELOCPARAM(&CodeLoc)) {
+    return submit([&](handler &CGH) {
+      CGH.require(Dest);
+      CGH.copy(Src, Dest);
+    } _CODELOCFW(CodeLoc));
+  }
+
+  /// Copies data from a memory region pointed to by a placeholder accessor to
+  /// another memory region pointed to by a raw pointer.
+  ///
+  /// \param Src is a placeholder accessor to the source memory.
+  /// \param Dest is a raw pointer to the destination memory.
+  /// \return an event representing copy operation.
+  template <typename SrcT, int SrcDims, access_mode SrcMode, target SrcTgt,
+            access::placeholder IsPlaceholder, typename DestT>
+  event copy(accessor<SrcT, SrcDims, SrcMode, SrcTgt, IsPlaceholder> Src,
+             DestT *Dest _CODELOCPARAM(&CodeLoc)) {
+    return submit([&](handler &CGH) {
+      CGH.require(Src);
+      CGH.copy(Src, Dest);
+    } _CODELOCFW(CodeLoc));
+  }
+
+  /// Copies data from a memory region pointed to by a raw pointer to another
+  /// memory region pointed to by a placeholder accessor.
+  ///
+  /// \param Src is a raw pointer to the source memory.
+  /// \param Dest is a placeholder accessor to the destination memory.
+  /// \return an event representing copy operation.
+  template <typename SrcT, typename DestT, int DestDims, access_mode DestMode,
+            target DestTgt, access::placeholder IsPlaceholder>
+  event copy(const SrcT *Src,
+             accessor<DestT, DestDims, DestMode, DestTgt, IsPlaceholder> Dest
+                 _CODELOCPARAM(&CodeLoc)) {
+    return submit([&](handler &CGH) {
+      CGH.require(Dest);
+      CGH.copy(Src, Dest);
+    } _CODELOCFW(CodeLoc));
+  }
+
+  /// Copies data from one memory region to another, both pointed by placeholder
+  /// accessors.
+  ///
+  /// \param Src is a placeholder accessor to the source memory.
+  /// \param Dest is a placeholder accessor to the destination memory.
+  /// \return an event representing copy operation.
+  template <typename SrcT, int SrcDims, access_mode SrcMode, target SrcTgt,
+            access::placeholder IsSrcPlaceholder, typename DestT, int DestDims,
+            access_mode DestMode, target DestTgt,
+            access::placeholder IsDestPlaceholder>
+  event
+  copy(accessor<SrcT, SrcDims, SrcMode, SrcTgt, IsSrcPlaceholder> Src,
+       accessor<DestT, DestDims, DestMode, DestTgt, IsDestPlaceholder> Dest
+           _CODELOCPARAM(&CodeLoc)) {
+    return submit([&](handler &CGH) {
+      CGH.require(Src);
+      CGH.require(Dest);
+      CGH.copy(Src, Dest);
+    } _CODELOCFW(CodeLoc));
+  }
+
+  /// Provides guarantees that the memory object accessed via Acc is updated
+  /// on the host after operation is complete.
+  ///
+  /// \param Acc is a SYCL accessor that needs to be updated on host.
+  /// \return an event representing update_host operation.
+  template <typename T, int Dims, access_mode Mode, target Tgt,
+            access::placeholder IsPlaceholder>
+  event update_host(
+      accessor<T, Dims, Mode, Tgt, IsPlaceholder> Acc _CODELOCPARAM(&CodeLoc)) {
+    return submit([&](handler &CGH) {
+      CGH.require(Acc);
+      CGH.update_host(Acc);
+    } _CODELOCFW(CodeLoc));
+  }
+
+  /// Fills the specified memory with the specified data.
+  ///
+  /// \param Dest is the placeholder accessor to the memory to fill.
+  /// \param Src is the data to fill the memory with. T should be
+  /// trivially copyable.
+  /// \return an event representing fill operation.
+  template <typename T, int Dims, access_mode Mode, target Tgt,
+            access::placeholder IsPlaceholder>
+  event fill(accessor<T, Dims, Mode, Tgt, IsPlaceholder> Dest,
+             const T &Src _CODELOCPARAM(&CodeLoc)) {
+    return submit([&](handler &CGH) {
+      CGH.require(Dest);
+      CGH.fill<T>(Dest, Src);
+    } _CODELOCFW(CodeLoc));
+  }
+
 // Clean KERNELFUNC macros.
 #undef _KERNELFUNCPARAM
 
diff --git a/sycl/unittests/queue/CMakeLists.txt b/sycl/unittests/queue/CMakeLists.txt
@@ -4,4 +4,5 @@ add_sycl_unittest(QueueTests OBJECT
   USM.cpp
   Wait.cpp
   GetProfilingInfo.cpp
+  ShortcutFunctions.cpp
 )
diff --git a/sycl/unittests/queue/ShortcutFunctions.cpp b/sycl/unittests/queue/ShortcutFunctions.cpp
@@ -0,0 +1,235 @@
+//==-------------- ShortcutFunctions.cpp --- queue unit tests --------------==//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <detail/context_impl.hpp>
+#include <gtest/gtest.h>
+#include <helpers/PiMock.hpp>
+#include <sycl/ext/oneapi/accessor_property_list.hpp>
+#include <sycl/handler.hpp>
+#include <sycl/queue.hpp>
+#include <sycl/sycl.hpp>
+
+#include <memory>
+
+using namespace sycl;
+
+namespace {
+struct TestCtx {
+  bool BufferFillCalled = false;
+  bool BufferReadCalled = false;
+  bool BufferWriteCalled = false;
+  bool BufferCopyCalled = false;
+};
+} // namespace
+
+static std::unique_ptr<TestCtx> TestContext;
+
+pi_result redefinedEnqueueMemBufferWrite(pi_queue command_queue, pi_mem buffer,
+                                         pi_bool blocking_write, size_t offset,
+                                         size_t size, const void *ptr,
+                                         pi_uint32 num_events_in_wait_list,
+                                         const pi_event *event_wait_list,
+                                         pi_event *event) {
+  TestContext->BufferWriteCalled = true;
+  return PI_SUCCESS;
+}
+
+pi_result redefinedEnqueueMemBufferRead(pi_queue queue, pi_mem buffer,
+                                        pi_bool blocking_read, size_t offset,
+                                        size_t size, void *ptr,
+                                        pi_uint32 num_events_in_wait_list,
+                                        const pi_event *event_wait_list,
+                                        pi_event *event) {
+  TestContext->BufferReadCalled = true;
+  return PI_SUCCESS;
+}
+
+pi_result redefinedEnqueueMemBufferCopy(pi_queue command_queue,
+                                        pi_mem src_buffer, pi_mem dst_buffer,
+                                        size_t src_offset, size_t dst_offset,
+                                        size_t size,
+                                        pi_uint32 num_events_in_wait_list,
+                                        const pi_event *event_wait_list,
+                                        pi_event *event) {
+  TestContext->BufferCopyCalled = true;
+  return PI_SUCCESS;
+}
+
+pi_result redefinedEnqueueMemBufferFill(pi_queue command_queue, pi_mem buffer,
+                                        const void *pattern,
+                                        size_t pattern_size, size_t offset,
+                                        size_t size,
+                                        pi_uint32 num_events_in_wait_list,
+                                        const pi_event *event_wait_list,
+                                        pi_event *event) {
+  TestContext->BufferFillCalled = true;
+  return PI_SUCCESS;
+}
+
+TEST(ShortcutFunctions, ShortcutsCallCorrectPIFunctions) {
+  unittest::PiMock Mock;
+  platform Plt = Mock.getPlatform();
+
+  Mock.redefine<detail::PiApiKind::piEnqueueMemBufferWrite>(
+      redefinedEnqueueMemBufferWrite);
+  Mock.redefine<detail::PiApiKind::piEnqueueMemBufferRead>(
+      redefinedEnqueueMemBufferRead);
+  Mock.redefine<detail::PiApiKind::piEnqueueMemBufferCopy>(
+      redefinedEnqueueMemBufferCopy);
+
+  Mock.redefine<detail::PiApiKind::piEnqueueMemBufferFill>(
+      redefinedEnqueueMemBufferFill);
+
+  context Ctx(Plt);
+  queue Q{Ctx, default_selector()};
+
+  constexpr std::size_t Size = 1;
+
+  // Queue.copy(accessor src, shared_ptr dest);
+  {
+    TestContext.reset(new TestCtx());
+
+    int Data[Size];
+    buffer<int> Buf(Data, Size);
+
+    accessor<int, 1, access::mode::read, access::target::device,
+             access::placeholder::true_t, ext::oneapi::accessor_property_list<>>
+        Src(Buf);
+    ASSERT_TRUE(Src.is_placeholder());
+
+    std::shared_ptr<int> Dest = std::make_shared<int>(0);
+
+    Q.copy(Src, Dest);
+    Q.wait();
+
+    EXPECT_TRUE(TestContext->BufferReadCalled);
+  }
+
+  // Queue.copy(shared_ptr src, accessor dest);
+  {
+    TestContext.reset(new TestCtx());
+
+    int Data[Size];
+    buffer<int> Buf(Data, Size);
+
+    std::shared_ptr<int> Src = std::make_shared<int>(42);
+
+    accessor<int, 1, access::mode::write, access::target::device,
+             access::placeholder::true_t, ext::oneapi::accessor_property_list<>>
+        Dest(Buf);
+    ASSERT_TRUE(Dest.is_placeholder());
+
+    Q.copy(Src, Dest);
+    Q.wait();
+
+    EXPECT_TRUE(TestContext->BufferWriteCalled);
+  }
+
+  // Queue.copy(accessor src, ptr* dest);
+  {
+    TestContext.reset(new TestCtx());
+
+    int Data[Size];
+    buffer<int> Buf(Data, Size);
+
+    accessor<int, 1, access::mode::read, access::target::device,
+             access::placeholder::true_t, ext::oneapi::accessor_property_list<>>
+        Src(Buf);
+    ASSERT_TRUE(Src.is_placeholder());
+
+    std::unique_ptr<int> Dest = std::make_unique<int>(0);
+
+    Q.copy(Src, Dest.get());
+    Q.wait();
+
+    EXPECT_TRUE(TestContext->BufferReadCalled);
+  }
+
+  // Queue.copy(ptr* src, accessor dest);
+  {
+    TestContext.reset(new TestCtx());
+
+    int Data[Size];
+    buffer<int> Buf(Data, Size);
+
+    std::unique_ptr<int> Src = std::make_unique<int>(42);
+
+    accessor<int, 1, access::mode::write, access::target::device,
+             access::placeholder::true_t, ext::oneapi::accessor_property_list<>>
+        Dest(Buf);
+    ASSERT_TRUE(Dest.is_placeholder());
+
+    Q.copy(Src.get(), Dest);
+    Q.wait();
+
+    EXPECT_TRUE(TestContext->BufferWriteCalled);
+  }
+
+  // Queue.copy(accessor src, accessor dest);
+  {
+    TestContext.reset(new TestCtx());
+
+    int SrcData[Size];
+    buffer<int> SrcBuf(SrcData, Size);
+
+    int DestData[Size];
+    buffer<int> DestBuf(DestData, Size);
+
+    accessor<int, 1, access::mode::read, access::target::device,
+             access::placeholder::true_t, ext::oneapi::accessor_property_list<>>
+        Src(SrcBuf);
+    accessor<int, 1, access::mode::write, access::target::device,
+             access::placeholder::true_t, ext::oneapi::accessor_property_list<>>
+        Dest(DestBuf);
+
+    ASSERT_TRUE(Src.is_placeholder());
+    ASSERT_TRUE(Dest.is_placeholder());
+
+    Q.copy(Src, Dest);
+    Q.wait();
+
+    EXPECT_TRUE(TestContext->BufferCopyCalled);
+  }
+
+  // Queue.update_host(accessor acc);
+  {
+    TestContext.reset(new TestCtx());
+
+    int Data[Size];
+    buffer<int> Buf(Data, Size);
+
+    accessor<int, 1, access::mode::read_write, access::target::device,
+             access::placeholder::true_t, ext::oneapi::accessor_property_list<>>
+        Acc(Buf);
+
+    ASSERT_TRUE(Acc.is_placeholder());
+
+    Q.update_host(Acc);
+    Q.wait();
+
+    // No PI functions expected.
+  }
+
+  // Queue.fill<T>(accessor Dest, T src)
+  {
+    TestContext.reset(new TestCtx());
+
+    int Data[Size];
+    buffer<int> Buf(Data, Size);
+
+    accessor<int, 1, access::mode::read_write, access::target::device,
+             access::placeholder::true_t, ext::oneapi::accessor_property_list<>>
+        Acc(Buf);
+    ASSERT_TRUE(Acc.is_placeholder());
+
+    Q.fill(Acc, 42);
+    Q.wait();
+
+    EXPECT_TRUE(TestContext->BufferFillCalled);
+  }
+}

Original file line number	Diff line number	Diff line change
`@@ -4,4 +4,5 @@ add_sycl_unittest(QueueTests OBJECT`
`4`	`4`	`USM.cpp`
`5`	`5`	`Wait.cpp`
`6`	`6`	`GetProfilingInfo.cpp`
	`7`	`+ ShortcutFunctions.cpp`
`7`	`8`	`)`