Skip to content

fix(runtime): restore strict CUDA feature builds #31

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
d-tietjen merged 1 commit into from
Oct 7, 2026
Merged

fix(runtime): restore strict CUDA feature builds #31

Show file tree
Hide file tree
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
14 changes: 9 additions & 5 deletions crates/shardcache-runtime/src/connector.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,8 @@ pub struct ConnectorTransferReport {
}

#[derive(Debug)]
// Preserve the public handle shape and keep submission free of a new allocation.
#[allow(clippy::large_enum_variant)]
pub enum ConnectorTransferHandle {
Completed(ConnectorTransferReport),
#[cfg(all(feature = "cuda", target_os = "linux"))]
Expand Down Expand Up @@ -690,7 +692,7 @@ mod tests {
device_ordinal: 0,
stream_ordinal: 0,
allocation_id: 9001,
dst_device_ptr: device_buffer.as_device_ptr().as_raw() as u64,
dst_device_ptr: device_buffer.as_device_ptr().as_raw(),
dst_base_offset_bytes: 0,
}),
&mut engine,
Expand Down Expand Up @@ -748,7 +750,7 @@ mod tests {
device_ordinal: 0,
stream_ordinal: 0,
allocation_id: 9002,
dst_device_ptr: device_buffer.as_device_ptr().as_raw() as u64,
dst_device_ptr: device_buffer.as_device_ptr().as_raw(),
dst_base_offset_bytes: 0,
},
);
Expand Down Expand Up @@ -820,7 +822,7 @@ mod tests {
device_ordinal: 0,
stream_ordinal: 0,
allocation_id: 9003,
dst_device_ptr: device_buffer.as_device_ptr().as_raw() as u64,
dst_device_ptr: device_buffer.as_device_ptr().as_raw(),
dst_base_offset_bytes: 0,
},
);
Expand Down Expand Up @@ -890,7 +892,7 @@ mod tests {
let mut sorted = values.to_vec();
sorted.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
let mid = sorted.len() / 2;
if sorted.len() % 2 == 0 {
if sorted.len().is_multiple_of(2) {
(sorted[mid - 1] + sorted[mid]) / 2.0
} else {
sorted[mid]
Expand All @@ -907,6 +909,8 @@ mod tests {
}

#[cfg(all(feature = "cuda", target_os = "linux"))]
// Keep the independent CUDA benchmark dimensions visible at each call site.
#[allow(clippy::too_many_arguments)]
fn run_real_cuda_benchmark_case(
case_name: &str,
cuda: shardmap::cuda::CudaConfig,
Expand Down Expand Up @@ -959,7 +963,7 @@ mod tests {
device_ordinal: 0,
stream_ordinal: 0,
allocation_id: 9100,
dst_device_ptr: device_buffer.as_device_ptr().as_raw() as u64,
dst_device_ptr: device_buffer.as_device_ptr().as_raw(),
dst_base_offset_bytes: 0,
},
);
Expand Down
73 changes: 39 additions & 34 deletions crates/shardcache-runtime/src/cuda_engine.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -480,7 +480,10 @@ impl CudaTransferEngine {
let aligned = aligned_host_range(ptr, len, HOST_PAGE_BYTES)?;
let existing = self.registered_host_ranges();
for uncovered in uncovered_host_ranges(aligned, &existing) {
crate::cuda_ffi::register_host_region(uncovered.start as *mut u8, uncovered.len())?;
crate::cuda_ffi::register_host_region(
uncovered.start as *mut u8,
uncovered.end - uncovered.start,
)?;
self.temporary_registered_regions
.push(RegisteredHostRegion { range: uncovered });
}
Expand All @@ -499,7 +502,10 @@ impl CudaTransferEngine {
let mut existing = Vec::with_capacity(self.cached_session_views.len());
for cached in &self.cached_session_views {
for uncovered in uncovered_host_ranges(cached.range, &existing) {
crate::cuda_ffi::register_host_region(uncovered.start as *mut u8, uncovered.len())?;
crate::cuda_ffi::register_host_region(
uncovered.start as *mut u8,
uncovered.end - uncovered.start,
)?;
self.persistent_registered_regions
.push(RegisteredHostRegion { range: uncovered });
existing.push(uncovered);
Expand Down Expand Up @@ -576,26 +582,26 @@ impl CudaTransferEngine {
continue;
}

if matches!(transfer.target(), RuntimeTransferTarget::Gpu(_)) {
if let Some(plan) = detect_strided_host_copy_plan(
if matches!(transfer.target(), RuntimeTransferTarget::Gpu(_))
&& let Some(plan) = detect_strided_host_copy_plan(
&packed.offsets,
&packed.lengths,
&dst_offsets,
index,
) {
let src = unsafe { src_base.add(plan.src_offset) };
self.submit_host_copy_2d(
transfer.target(),
&descriptors[index],
src,
plan.row_bytes,
plan.row_count,
plan.dst_pitch,
)?;
index += plan.row_count;
self.advance_stream();
continue;
}
)
{
let src = unsafe { src_base.add(plan.src_offset) };
self.submit_host_copy_2d(
transfer.target(),
&descriptors[index],
src,
plan.row_bytes,
plan.row_count,
plan.dst_pitch,
)?;
index += plan.row_count;
self.advance_stream();
continue;
}

let start_index = index;
Expand Down Expand Up @@ -669,23 +675,22 @@ impl CudaTransferEngine {
continue;
}

if matches!(transfer.target(), RuntimeTransferTarget::Gpu(_)) {
if let Some(plan) =
if matches!(transfer.target(), RuntimeTransferTarget::Gpu(_))
&& let Some(plan) =
detect_strided_host_copy_plan(offsets, lengths, &dst_offsets, index)
{
let src = unsafe { src_base.add(plan.src_offset) };
self.submit_host_copy_2d(
transfer.target(),
&descriptors[index],
src,
plan.row_bytes,
plan.row_count,
plan.dst_pitch,
)?;
index += plan.row_count;
self.advance_stream();
continue;
}
{
let src = unsafe { src_base.add(plan.src_offset) };
self.submit_host_copy_2d(
transfer.target(),
&descriptors[index],
src,
plan.row_bytes,
plan.row_count,
plan.dst_pitch,
)?;
index += plan.row_count;
self.advance_stream();
continue;
}

let start_index = index;
Expand Down
Loading