Skip to content

Refactor stridingMemcpyKernel to be unroll friendly #3

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
Artem-B wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Refactor stridingMemcpyKernel to be unroll friendly #3

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e8662a1
Parameterize unroll factor in stridingMemcpyKernel and add uint4 spec…
Artem-B Apr 14, 2026
110247f
Convert main loop to use indices. Fix tail processing.
Artem-B Apr 15, 2026
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
59 changes: 23 additions & 36 deletions library/kernels.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,11 @@ __device__ void write_to_regular_memory(T *dst, T val) {
*dst = val;
}

template<>
__device__ void write_to_regular_memory<uint4>(uint4 *dst, uint4 val) {
*dst = {val.x, val.y, val.z, val.w};
}

template<typename T>
__device__ void write_to_multicast_memory(T *dst, T val) {
#if __CUDA_ARCH__ >= 900
Expand All @@ -59,54 +64,36 @@ __device__ void reduce_from_multicast_red(T *dst, T *val) {
#endif
}

template<typename T, write_to_memory<T> write>
template<typename T, write_to_memory<T> write, int UNROLL_FACTOR = 12>
__global__ void stridingMemcpyKernel(unsigned int totalThreadCount, unsigned long long loopCount, T* dst, T* src, size_t sizeInElement) {
T *dstEnd = dst + sizeInElement;
size_t chunkSizeInElement = sizeInElement / totalThreadCount;

size_t globalThreadId = blockDim.x * blockIdx.x + threadIdx.x;
dst += globalThreadId;
src += globalThreadId;
size_t maxElementIndex = sizeInElement - globalThreadId;

// Calculate where to end the big pipelined copy
size_t bigChunkSizeInElement = chunkSizeInElement / 12;
T *dstBigEnd = dst + (bigChunkSizeInElement * 12) * totalThreadCount;

size_t bigChunkSizeInElement = chunkSizeInElement / UNROLL_FACTOR;
for (unsigned int i = 0; i < loopCount; i++) {
T* cdst = dst;
T* csrc = src;
for (unsigned long long j = 0; j < bigChunkSizeInElement; j++) {
unsigned long long offset = j * UNROLL_FACTOR * totalThreadCount;

while (cdst < dstBigEnd) {
T pipe_0 = *csrc; csrc += totalThreadCount;
T pipe_1 = *csrc; csrc += totalThreadCount;
T pipe_2 = *csrc; csrc += totalThreadCount;
T pipe_3 = *csrc; csrc += totalThreadCount;
T pipe_4 = *csrc; csrc += totalThreadCount;
T pipe_5 = *csrc; csrc += totalThreadCount;
T pipe_6 = *csrc; csrc += totalThreadCount;
T pipe_7 = *csrc; csrc += totalThreadCount;
T pipe_8 = *csrc; csrc += totalThreadCount;
T pipe_9 = *csrc; csrc += totalThreadCount;
T pipe_10 = *csrc; csrc += totalThreadCount;
T pipe_11 = *csrc; csrc += totalThreadCount;

write(cdst, pipe_0); cdst += totalThreadCount;
write(cdst, pipe_1); cdst += totalThreadCount;
write(cdst, pipe_2); cdst += totalThreadCount;
write(cdst, pipe_3); cdst += totalThreadCount;
write(cdst, pipe_4); cdst += totalThreadCount;
write(cdst, pipe_5); cdst += totalThreadCount;
write(cdst, pipe_6); cdst += totalThreadCount;
write(cdst, pipe_7); cdst += totalThreadCount;
write(cdst, pipe_8); cdst += totalThreadCount;
write(cdst, pipe_9); cdst += totalThreadCount;
write(cdst, pipe_10); cdst += totalThreadCount;
write(cdst, pipe_11); cdst += totalThreadCount;
T pipe[UNROLL_FACTOR];
#pragma unroll
for (int k = 0; k < UNROLL_FACTOR; ++k) {
pipe[k] = src[offset + k * totalThreadCount];
}
#pragma unroll
for (int k = 0; k < UNROLL_FACTOR; ++k) {
write(dst + offset + k * totalThreadCount, pipe[k]);
}
}

// Take care of copies that didn't get aligned properly
while (cdst < dstEnd) {
write(cdst, *csrc); cdst += totalThreadCount; csrc += totalThreadCount;
// Take care of copies that didn't get aligned properly and remainder elements
size_t start_offset = bigChunkSizeInElement * UNROLL_FACTOR * totalThreadCount;
for (size_t offset = start_offset; offset < maxElementIndex; offset += totalThreadCount) {
write(dst + offset, src[offset]);
}
}
}
Expand Down