https://github.com/RossBrunton updated
https://github.com/llvm/llvm-project/pull/150036
>From d28c1a4958738e3ada876a831d9179cf134a4e66 Mon Sep 17 00:00:00 2001
From: Ross Brunton <r...@codeplay.com>
Date: Tue, 22 Jul 2025 15:50:18 +0100
Subject: [PATCH 1/3] [Offload] Add (a new version of) olWaitQueue
Not to be confused with olSyncQueue, which used to be called olWaitQueue
until #150023. This function causes a queue to wait until all the
provided events have completed before running any future scheduled work.
---
offload/liboffload/API/Queue.td | 17 ++
offload/liboffload/src/OffloadImpl.cpp | 22 +++
offload/unittests/OffloadAPI/CMakeLists.txt | 3 +-
.../OffloadAPI/device_code/CMakeLists.txt | 2 +
.../OffloadAPI/device_code/sequence.c | 11 ++
.../OffloadAPI/queue/olWaitQueue.cpp | 148 ++++++++++++++++++
6 files changed, 202 insertions(+), 1 deletion(-)
create mode 100644 offload/unittests/OffloadAPI/device_code/sequence.c
create mode 100644 offload/unittests/OffloadAPI/queue/olWaitQueue.cpp
diff --git a/offload/liboffload/API/Queue.td b/offload/liboffload/API/Queue.td
index 19327cdab4254..43c723de54510 100644
--- a/offload/liboffload/API/Queue.td
+++ b/offload/liboffload/API/Queue.td
@@ -41,6 +41,23 @@ def : Function {
let returns = [];
}
+def : Function {
+ let name = "olWaitQueue";
+ let desc = "Make any future work submitted to this queue wait until the
provided events are complete.";
+ let details = [
+ "All events in `Events` must complete beforet he queue is unblocked.",
+ "The input events can be from any queue on any device provided by the
same platform as `Queue`.",
+ ];
+ let params = [
+ Param<"ol_queue_handle_t", "Queue", "handle of the queue", PARAM_IN>,
+ Param<"ol_event_handle_t *", "Events", "list of `NumEvents` events to
wait for", PARAM_IN>,
+ Param<"size_t", "NumEvents", "size of `Events`", PARAM_IN>,
+ ];
+ let returns = [
+ Return<"OL_ERRC_INVALID_NULL_HANDLE", ["Any event handle in the list is
NULL"]>,
+ ];
+}
+
def : Enum {
let name = "ol_queue_info_t";
let desc = "Supported queue info.";
diff --git a/offload/liboffload/src/OffloadImpl.cpp
b/offload/liboffload/src/OffloadImpl.cpp
index d93e4f1db58a7..c155a6b85387c 100644
--- a/offload/liboffload/src/OffloadImpl.cpp
+++ b/offload/liboffload/src/OffloadImpl.cpp
@@ -500,6 +500,28 @@ Error olSyncQueue_impl(ol_queue_handle_t Queue) {
return Error::success();
}
+Error olWaitQueue_impl(ol_queue_handle_t Queue, ol_event_handle_t *Events,
+ size_t NumEvents) {
+ auto *Device = Queue->Device->Device;
+
+ for (size_t I = 0; I < NumEvents; I++) {
+ auto *Event = Events[I];
+
+ if (!Event)
+ return Plugin::error(ErrorCode::INVALID_NULL_HANDLE,
+ "olWaitQueue asked to wait on a NULL event");
+
+ // Do nothing if the event is for this queue
+ if (Event->Queue == Queue)
+ continue;
+
+ if (auto Err = Device->waitEvent(Event->EventInfo, Queue->AsyncInfo))
+ return Err;
+ }
+
+ return Error::success();
+}
+
Error olGetQueueInfoImplDetail(ol_queue_handle_t Queue,
ol_queue_info_t PropName, size_t PropSize,
void *PropValue, size_t *PropSizeRet) {
diff --git a/offload/unittests/OffloadAPI/CMakeLists.txt
b/offload/unittests/OffloadAPI/CMakeLists.txt
index f09cfc6bb0876..2621eaeb64e82 100644
--- a/offload/unittests/OffloadAPI/CMakeLists.txt
+++ b/offload/unittests/OffloadAPI/CMakeLists.txt
@@ -39,7 +39,8 @@ add_offload_unittest("queue"
queue/olSyncQueue.cpp
queue/olDestroyQueue.cpp
queue/olGetQueueInfo.cpp
- queue/olGetQueueInfoSize.cpp)
+ queue/olGetQueueInfoSize.cpp
+ queue/olWaitQueue.cpp)
add_offload_unittest("symbol"
symbol/olGetSymbol.cpp
diff --git a/offload/unittests/OffloadAPI/device_code/CMakeLists.txt
b/offload/unittests/OffloadAPI/device_code/CMakeLists.txt
index 11c8ccbd6c7c5..0e4695ee9969f 100644
--- a/offload/unittests/OffloadAPI/device_code/CMakeLists.txt
+++ b/offload/unittests/OffloadAPI/device_code/CMakeLists.txt
@@ -8,6 +8,7 @@ add_offload_test_device_code(localmem_static.c localmem_static)
add_offload_test_device_code(global.c global)
add_offload_test_device_code(global_ctor.c global_ctor)
add_offload_test_device_code(global_dtor.c global_dtor)
+add_offload_test_device_code(sequence.c sequence)
add_custom_target(offload_device_binaries DEPENDS
foo.bin
@@ -19,5 +20,6 @@ add_custom_target(offload_device_binaries DEPENDS
global.bin
global_ctor.bin
global_dtor.bin
+ sequence.bin
)
set(OFFLOAD_TEST_DEVICE_CODE_PATH ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE)
diff --git a/offload/unittests/OffloadAPI/device_code/sequence.c
b/offload/unittests/OffloadAPI/device_code/sequence.c
new file mode 100644
index 0000000000000..22504086ffa38
--- /dev/null
+++ b/offload/unittests/OffloadAPI/device_code/sequence.c
@@ -0,0 +1,11 @@
+#include <gpuintrin.h>
+#include <stdint.h>
+
+__gpu_kernel void sequence(uint32_t idx, uint32_t *inout) {
+ if (idx == 0)
+ inout[idx] = 0;
+ else if (idx == 1)
+ inout[idx] = 1;
+ else
+ inout[idx] = inout[idx-1] + inout[idx-2];
+}
diff --git a/offload/unittests/OffloadAPI/queue/olWaitQueue.cpp
b/offload/unittests/OffloadAPI/queue/olWaitQueue.cpp
new file mode 100644
index 0000000000000..fdf272dafa911
--- /dev/null
+++ b/offload/unittests/OffloadAPI/queue/olWaitQueue.cpp
@@ -0,0 +1,148 @@
+//===------- Offload API tests - olWaitQueue
------------------------------===//
+//
+// 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 "../common/Fixtures.hpp"
+#include <OffloadAPI.h>
+#include <gtest/gtest.h>
+
+struct olWaitQueueTest : OffloadProgramTest {
+ void SetUp() override {
+ RETURN_ON_FATAL_FAILURE(OffloadProgramTest::SetUpWith("sequence"));
+ ASSERT_SUCCESS(
+ olGetSymbol(Program, "sequence", OL_SYMBOL_KIND_KERNEL, &Kernel));
+ LaunchArgs.Dimensions = 1;
+ LaunchArgs.GroupSize = {1, 1, 1};
+ LaunchArgs.NumGroups = {1, 1, 1};
+ LaunchArgs.DynSharedMemory = 0;
+ }
+
+ void TearDown() override {
+ RETURN_ON_FATAL_FAILURE(OffloadProgramTest::TearDown());
+ }
+
+ ol_symbol_handle_t Kernel = nullptr;
+ ol_kernel_launch_size_args_t LaunchArgs{};
+};
+OFFLOAD_TESTS_INSTANTIATE_DEVICE_FIXTURE(olWaitQueueTest);
+
+TEST_P(olWaitQueueTest, Success) {
+ constexpr size_t NUM_KERNELS = 16;
+ ol_queue_handle_t Queues[NUM_KERNELS];
+ ol_event_handle_t Events[NUM_KERNELS];
+
+ void *Mem;
+ ASSERT_SUCCESS(olMemAlloc(Device, OL_ALLOC_TYPE_MANAGED,
+ NUM_KERNELS * sizeof(uint32_t), &Mem));
+ struct {
+ uint32_t Idx;
+ void *Mem;
+ } Args{0, Mem};
+
+ for (size_t I = 0; I < NUM_KERNELS; I++) {
+ Args.Idx = I;
+
+ ASSERT_SUCCESS(olCreateQueue(Device, &Queues[I]));
+
+ if (I > 0)
+ ASSERT_SUCCESS(olWaitQueue(Queues[I], &Events[I - 1], 1));
+
+ ASSERT_SUCCESS(olLaunchKernel(Queues[I], Device, Kernel, &Args,
+ sizeof(Args), &LaunchArgs, &Events[I]));
+ }
+
+ ASSERT_SUCCESS(olSyncEvent(Events[NUM_KERNELS - 1]));
+
+ uint32_t *Data = (uint32_t *)Mem;
+ for (uint32_t i = 2; i < NUM_KERNELS; i++) {
+ ASSERT_EQ(Data[i], Data[i - 1] + Data[i - 2]);
+ }
+}
+
+TEST_P(olWaitQueueTest, SuccessSingleQueue) {
+ constexpr size_t NUM_KERNELS = 16;
+ ol_queue_handle_t Queue;
+ ol_event_handle_t Events[NUM_KERNELS];
+
+ ASSERT_SUCCESS(olCreateQueue(Device, &Queue));
+
+ void *Mem;
+ ASSERT_SUCCESS(olMemAlloc(Device, OL_ALLOC_TYPE_MANAGED,
+ NUM_KERNELS * sizeof(uint32_t), &Mem));
+ struct {
+ uint32_t Idx;
+ void *Mem;
+ } Args{0, Mem};
+
+ for (size_t I = 0; I < NUM_KERNELS; I++) {
+ Args.Idx = I;
+
+ if (I > 0)
+ ASSERT_SUCCESS(olWaitQueue(Queue, &Events[I - 1], 1));
+
+ ASSERT_SUCCESS(olLaunchKernel(Queue, Device, Kernel, &Args, sizeof(Args),
+ &LaunchArgs, &Events[I]));
+ }
+
+ ASSERT_SUCCESS(olSyncEvent(Events[NUM_KERNELS - 1]));
+
+ uint32_t *Data = (uint32_t *)Mem;
+ for (uint32_t i = 2; i < NUM_KERNELS; i++) {
+ ASSERT_EQ(Data[i], Data[i - 1] + Data[i - 2]);
+ }
+}
+
+TEST_P(olWaitQueueTest, SuccessMultipleEvents) {
+ constexpr size_t NUM_KERNELS = 16;
+ ol_queue_handle_t Queues[NUM_KERNELS];
+ ol_event_handle_t Events[NUM_KERNELS];
+
+ void *Mem;
+ ASSERT_SUCCESS(olMemAlloc(Device, OL_ALLOC_TYPE_MANAGED,
+ NUM_KERNELS * sizeof(uint32_t), &Mem));
+ struct {
+ uint32_t Idx;
+ void *Mem;
+ } Args{0, Mem};
+
+ for (size_t I = 0; I < NUM_KERNELS; I++) {
+ Args.Idx = I;
+
+ ASSERT_SUCCESS(olCreateQueue(Device, &Queues[I]));
+
+ if (I > 0)
+ ASSERT_SUCCESS(olWaitQueue(Queues[I], Events, I));
+
+ ASSERT_SUCCESS(olLaunchKernel(Queues[I], Device, Kernel, &Args,
+ sizeof(Args), &LaunchArgs, &Events[I]));
+ }
+
+ ASSERT_SUCCESS(olSyncEvent(Events[NUM_KERNELS - 1]));
+
+ uint32_t *Data = (uint32_t *)Mem;
+ for (uint32_t i = 2; i < NUM_KERNELS; i++) {
+ ASSERT_EQ(Data[i], Data[i - 1] + Data[i - 2]);
+ }
+}
+
+TEST_P(olWaitQueueTest, InvalidNullQueue) {
+ ol_event_handle_t Event;
+ ASSERT_ERROR(OL_ERRC_INVALID_NULL_HANDLE, olWaitQueue(nullptr, &Event, 1));
+}
+
+TEST_P(olWaitQueueTest, InvalidNullEvent) {
+ ol_queue_handle_t Queue;
+ ASSERT_SUCCESS(olCreateQueue(Device, &Queue));
+ ASSERT_ERROR(OL_ERRC_INVALID_NULL_POINTER, olWaitQueue(Queue, nullptr, 1));
+}
+
+TEST_P(olWaitQueueTest, InvalidNullInnerEvent) {
+ ol_queue_handle_t Queue;
+ ASSERT_SUCCESS(olCreateQueue(Device, &Queue));
+ ol_event_handle_t Event = nullptr;
+ ASSERT_ERROR(OL_ERRC_INVALID_NULL_HANDLE, olWaitQueue(Queue, &Event, 1));
+}
>From ceacd7f70f6dcfd30e879c4c2e4a7326eeaef247 Mon Sep 17 00:00:00 2001
From: Ross Brunton <r...@codeplay.com>
Date: Tue, 22 Jul 2025 15:59:15 +0100
Subject: [PATCH 2/3] Clang-format
---
offload/unittests/OffloadAPI/device_code/sequence.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/offload/unittests/OffloadAPI/device_code/sequence.c
b/offload/unittests/OffloadAPI/device_code/sequence.c
index 22504086ffa38..7662f2d817496 100644
--- a/offload/unittests/OffloadAPI/device_code/sequence.c
+++ b/offload/unittests/OffloadAPI/device_code/sequence.c
@@ -7,5 +7,5 @@ __gpu_kernel void sequence(uint32_t idx, uint32_t *inout) {
else if (idx == 1)
inout[idx] = 1;
else
- inout[idx] = inout[idx-1] + inout[idx-2];
+ inout[idx] = inout[idx - 1] + inout[idx - 2];
}
>From ef58e6a7cee20c9263a1fdaa4589c7c08831933c Mon Sep 17 00:00:00 2001
From: Ross Brunton <r...@codeplay.com>
Date: Tue, 22 Jul 2025 16:01:05 +0100
Subject: [PATCH 3/3] Typo
---
offload/liboffload/API/Queue.td | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/offload/liboffload/API/Queue.td b/offload/liboffload/API/Queue.td
index 43c723de54510..1193f40fc96ab 100644
--- a/offload/liboffload/API/Queue.td
+++ b/offload/liboffload/API/Queue.td
@@ -45,7 +45,7 @@ def : Function {
let name = "olWaitQueue";
let desc = "Make any future work submitted to this queue wait until the
provided events are complete.";
let details = [
- "All events in `Events` must complete beforet he queue is unblocked.",
+ "All events in `Events` must complete before the queue is unblocked.",
"The input events can be from any queue on any device provided by the
same platform as `Queue`.",
];
let params = [
_______________________________________________
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
