Skip to content

Add E2E test for bar.sync and bar.arrive #936

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

Open
wants to merge 1 commit into
base: SYCLomatic
Choose a base branch
from
Open

Add E2E test for bar.sync and bar.arrive #936

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
82 changes: 73 additions & 9 deletions features/feature_case/asm/asm_bar.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@

#include <cuda_runtime.h>

__device__ void bar(int *arr, int *brr) {
__global__ void bar_warp_sync(int *arr, int *brr) {
arr[threadIdx.x] = threadIdx.x + 10;
if (threadIdx.x % 2 == 0) {
for (int i = 0; i < 1000; ++i)
Expand All @@ -18,32 +18,96 @@ __device__ void bar(int *arr, int *brr) {
arr[threadIdx.x] = 0;
}

asm volatile ("bar.warp.sync %0;" :: "r"(0b1010101010));
asm volatile("bar.warp.sync %0;" ::"r"(0b1010101010));
if (threadIdx.x == 1) {
for (int i = 0; i < 10; ++i) {
brr[i] = arr[i];
}
}
}

__global__ void kernel(int *arr, int *brr) {
bar(arr, brr);
}

int main() {
int test_bar_warp_sync() {
int *arr, *brr;
cudaMallocManaged(&arr, sizeof(int) * 10);
cudaMemset(arr, 0, sizeof(int) * 10);
cudaMallocManaged(&brr, sizeof(int) * 10);
cudaMemset(brr, 0, sizeof(int) * 10);

kernel<<<1, 10>>>(arr, brr);
bar_warp_sync<<<1, 10>>>(arr, brr);
cudaDeviceSynchronize();
cudaFree(arr);
int res = 0;
for (int i = 1; i < 10; i+= 2)
for (int i = 1; i < 10; i += 2)
if (brr[i] != i + 10 || brr[i - 1] != 0)
res = 1;
cudaFree(brr);
return res;
}

constexpr unsigned int WIDTH = 10;
constexpr unsigned int HEIGHT = 32 * 4;

__global__ void bar_sync_bar_arrive(int *arr, int *results) {
__shared__ int buffer[HEIGHT / 2];
int value;

// Initialize provider threads
if (threadIdx.x < HEIGHT / 2)
value = arr[threadIdx.x * WIDTH] + arr[(threadIdx.x + HEIGHT / 2) * WIDTH];
else
value = 0;

for (unsigned int i = 1; i < WIDTH + 1; ++i) {
if (threadIdx.x < HEIGHT / 2) {
// provider
buffer[threadIdx.x] = value;
asm volatile("bar.arrive 1, %0;" ::"r"(HEIGHT));
value = arr[threadIdx.x * WIDTH + i] +
arr[(threadIdx.x + HEIGHT / 2) * WIDTH + i];
asm volatile("bar.sync 2, %0;" ::"r"(HEIGHT));
} else {
// consumer
asm volatile("bar.sync 1, %0;" ::"r"(HEIGHT));
value = buffer[threadIdx.x - HEIGHT / 2];
asm volatile("bar.arrive 2, %0;" ::"r"(HEIGHT));
results[threadIdx.x - HEIGHT / 2] += value;
}
}
}

int test_bar_sync_bar_arrive() {
int *arr, *results;

cudaMallocManaged(&arr, sizeof(int) * WIDTH * HEIGHT);
for (unsigned int i = 0; i < WIDTH * HEIGHT; ++i) {
arr[i] = i;
}

cudaMallocManaged(&results, sizeof(int) * HEIGHT / 2);
cudaMemset(results, 0, sizeof(int) * HEIGHT / 2);

bar_sync_bar_arrive<<<1, HEIGHT>>>(arr, results);
cudaDeviceSynchronize();

int expected = 0;
for (unsigned int i = 0; i < WIDTH * HEIGHT; ++i) {
expected += i;
}
int result = 0;
for (unsigned int i = 0; i < HEIGHT / 2; ++i) {
result += results[i];
}

int res = result != expected;

cudaFree(arr);
cudaFree(results);
return res;
}

int main() {
int res = 0;
res |= test_bar_warp_sync();
res |= test_bar_sync_bar_arrive();
return res;
}