Skip to content

[Offload] Add liboffload unit tests for shared/local memory #147040

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Jul 7, 2025
Merged

[Offload] Add liboffload unit tests for shared/local memory #147040

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c878ebf
[Offload] Add liboffload unit tests for shared/local memory
RossBrunton Jul 4, 2025
ea2183e
Add static test
RossBrunton Jul 7, 2025
82e76b9
Actually add the test
RossBrunton Jul 7, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 11 additions & 1 deletion offload/unittests/OffloadAPI/device_code/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,16 @@ add_offload_test_device_code(foo.c foo)
add_offload_test_device_code(bar.c bar)
# Compile with optimizations to eliminate AMDGPU implicit arguments.
add_offload_test_device_code(noargs.c noargs -O3)
add_offload_test_device_code(localmem.c localmem)
add_offload_test_device_code(localmem_reduction.c localmem_reduction)
add_offload_test_device_code(localmem_static.c localmem_static)

add_custom_target(offload_device_binaries DEPENDS foo.bin bar.bin noargs.bin)
add_custom_target(offload_device_binaries DEPENDS
foo.bin
bar.bin
noargs.bin
localmem.bin
localmem_reduction.bin
localmem_static.bin
)
set(OFFLOAD_TEST_DEVICE_CODE_PATH ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE)
11 changes: 11 additions & 0 deletions offload/unittests/OffloadAPI/device_code/localmem.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,11 @@
#include <gpuintrin.h>
#include <stdint.h>

extern __gpu_local uint32_t shared_mem[];

__gpu_kernel void localmem(uint32_t *out) {
shared_mem[__gpu_thread_id(0)] = __gpu_thread_id(0);
shared_mem[__gpu_thread_id(0)] *= 2;
out[__gpu_thread_id(0) + (__gpu_num_threads(0) * __gpu_block_id(0))] =
shared_mem[__gpu_thread_id(0)];
}
16 changes: 16 additions & 0 deletions offload/unittests/OffloadAPI/device_code/localmem_reduction.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,16 @@
#include <gpuintrin.h>
#include <stdint.h>

extern __gpu_local uint32_t shared_mem[];
Copy link
Contributor

@jhuber6 jhuber6 Jul 4, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No test for static shared memory? That would require [[clang::loader_uninitialized]] to work.

Copy link
Contributor Author

@RossBrunton RossBrunton Jul 7, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've added one. Since the number of tests is getting a bit large, I've also added a macro to the olLaunchKernel test.


__gpu_kernel void localmem_reduction(uint32_t *out) {
shared_mem[__gpu_thread_id(0)] = 2;

__gpu_sync_threads();

if (__gpu_thread_id(0) == 0) {
out[__gpu_block_id(0)] = 0;
for (uint32_t i = 0; i < __gpu_num_threads(0); i++)
out[__gpu_block_id(0)] += shared_mem[i];
}
}
17 changes: 17 additions & 0 deletions offload/unittests/OffloadAPI/device_code/localmem_static.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,17 @@
#include <gpuintrin.h>
#include <stdint.h>

[[clang::loader_uninitialized]]
__gpu_local uint32_t shared_mem[64];

__gpu_kernel void localmem_static(uint32_t *out) {
shared_mem[__gpu_thread_id(0)] = 2;

__gpu_sync_threads();

if (__gpu_thread_id(0) == 0) {
out[__gpu_block_id(0)] = 0;
for (uint32_t i = 0; i < __gpu_num_threads(0); i++)
out[__gpu_block_id(0)] += shared_mem[i];
}
}
101 changes: 85 additions & 16 deletions offload/unittests/OffloadAPI/kernel/olLaunchKernel.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,21 +38,19 @@ struct LaunchKernelTestBase : OffloadQueueTest {
ol_kernel_launch_size_args_t LaunchArgs{};
};

struct olLaunchKernelTest : LaunchKernelTestBase {
void SetUp() override {
RETURN_ON_FATAL_FAILURE(LaunchKernelTestBase::SetUpKernel("foo"));
}
};
OFFLOAD_TESTS_INSTANTIATE_DEVICE_FIXTURE(olLaunchKernelTest);

struct olLaunchKernelNoArgsTest : LaunchKernelTestBase {
void SetUp() override {
RETURN_ON_FATAL_FAILURE(LaunchKernelTestBase::SetUpKernel("noargs"));
}
};
OFFLOAD_TESTS_INSTANTIATE_DEVICE_FIXTURE(olLaunchKernelNoArgsTest);

TEST_P(olLaunchKernelTest, Success) {
#define KERNEL_TEST(NAME, KERNEL) \
struct olLaunchKernel##NAME##Test : LaunchKernelTestBase { \
void SetUp() override { LaunchKernelTestBase::SetUpKernel(#KERNEL); } \
}; \
OFFLOAD_TESTS_INSTANTIATE_DEVICE_FIXTURE(olLaunchKernel##NAME##Test);

KERNEL_TEST(Foo, foo)
KERNEL_TEST(NoArgs, noargs)
KERNEL_TEST(LocalMem, localmem)
KERNEL_TEST(LocalMemReduction, localmem_reduction)
KERNEL_TEST(LocalMemStatic, localmem_static)

TEST_P(olLaunchKernelFooTest, Success) {
void *Mem;
ASSERT_SUCCESS(olMemAlloc(Device, OL_ALLOC_TYPE_MANAGED,
LaunchArgs.GroupSize.x * sizeof(uint32_t), &Mem));
Expand Down Expand Up @@ -80,7 +78,7 @@ TEST_P(olLaunchKernelNoArgsTest, Success) {
ASSERT_SUCCESS(olWaitQueue(Queue));
}

TEST_P(olLaunchKernelTest, SuccessSynchronous) {
TEST_P(olLaunchKernelFooTest, SuccessSynchronous) {
void *Mem;
ASSERT_SUCCESS(olMemAlloc(Device, OL_ALLOC_TYPE_MANAGED,
LaunchArgs.GroupSize.x * sizeof(uint32_t), &Mem));
Expand All @@ -99,3 +97,74 @@ TEST_P(olLaunchKernelTest, SuccessSynchronous) {

ASSERT_SUCCESS(olMemFree(Mem));
}

TEST_P(olLaunchKernelLocalMemTest, Success) {
LaunchArgs.NumGroups.x = 4;
LaunchArgs.DynSharedMemory = 64 * sizeof(uint32_t);

void *Mem;
ASSERT_SUCCESS(olMemAlloc(Device, OL_ALLOC_TYPE_MANAGED,
LaunchArgs.GroupSize.x * LaunchArgs.NumGroups.x *
sizeof(uint32_t),
&Mem));
struct {
void *Mem;
} Args{Mem};

ASSERT_SUCCESS(olLaunchKernel(Queue, Device, Kernel, &Args, sizeof(Args),
&LaunchArgs, nullptr));

ASSERT_SUCCESS(olWaitQueue(Queue));

uint32_t *Data = (uint32_t *)Mem;
for (uint32_t i = 0; i < LaunchArgs.GroupSize.x * LaunchArgs.NumGroups.x; i++)
ASSERT_EQ(Data[i], (i % 64) * 2);

ASSERT_SUCCESS(olMemFree(Mem));
}

TEST_P(olLaunchKernelLocalMemReductionTest, Success) {
LaunchArgs.NumGroups.x = 4;
LaunchArgs.DynSharedMemory = 64 * sizeof(uint32_t);

void *Mem;
ASSERT_SUCCESS(olMemAlloc(Device, OL_ALLOC_TYPE_MANAGED,
LaunchArgs.NumGroups.x * sizeof(uint32_t), &Mem));
struct {
void *Mem;
} Args{Mem};

ASSERT_SUCCESS(olLaunchKernel(Queue, Device, Kernel, &Args, sizeof(Args),
&LaunchArgs, nullptr));

ASSERT_SUCCESS(olWaitQueue(Queue));

uint32_t *Data = (uint32_t *)Mem;
for (uint32_t i = 0; i < LaunchArgs.NumGroups.x; i++)
ASSERT_EQ(Data[i], 2 * LaunchArgs.GroupSize.x);

ASSERT_SUCCESS(olMemFree(Mem));
}

TEST_P(olLaunchKernelLocalMemStaticTest, Success) {
LaunchArgs.NumGroups.x = 4;
LaunchArgs.DynSharedMemory = 0;

void *Mem;
ASSERT_SUCCESS(olMemAlloc(Device, OL_ALLOC_TYPE_MANAGED,
LaunchArgs.NumGroups.x * sizeof(uint32_t), &Mem));
struct {
void *Mem;
} Args{Mem};

ASSERT_SUCCESS(olLaunchKernel(Queue, Device, Kernel, &Args, sizeof(Args),
&LaunchArgs, nullptr));

ASSERT_SUCCESS(olWaitQueue(Queue));

uint32_t *Data = (uint32_t *)Mem;
for (uint32_t i = 0; i < LaunchArgs.NumGroups.x; i++)
ASSERT_EQ(Data[i], 2 * LaunchArgs.GroupSize.x);

ASSERT_SUCCESS(olMemFree(Mem));
}
Loading