Skip to content

Fix pjrt stream executor tsan issue #363

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 2 commits into
base: rocm-jaxlib-v0.6.0
Choose a base branch
from
+6 −5
Open

Fix pjrt stream executor tsan issue #363

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3736b7a
Sync Literal::Storage data_
alekstheod Sep 19, 2025
1df25c4
Fix prj_stream_executor for test xla/tests/fuzz/rand_000041.hlo_test_…
alekstheod Sep 19, 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
5 changes: 0 additions & 5 deletions build_tools/rocm/tsan_ignore_list.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,17 +15,12 @@ race:xla::gpu::AllocateDestinationBuffer
race:xla::LocalDeviceState::ThenRelease

# To be fixed
race:xla::LiteralBase::Piece::DeallocateBuffers
race:xla::PjRtStreamExecutorLoadedExecutable::ExecuteHelper
race:xla::PjRtStreamExecutorClient::BufferFromHostBufferInternal
race:xla::HloRunnerPjRt::TransferLiteralsFromDevice
race:xla::MutableLiteralBase::~MutableLiteralBase
race:xla::MutableLiteralBase::PopulateR1<int>
race:xla::xla::gpu::GpuCompiler::CompileSingleModule
race:xla::LiteralBase::Piece::Storage::Storage
race:xla::LocalClient::TransferFromOutfeedLocal
race:llvm::cl::opt_storage<bool, false, false>::setValue<int>
race:xla::gpu::(anonymous namespace)::RecoverExp2Pattern::initStaticsIfNeeded<std::tuple<mlir::Float32Type, mlir::FloatTF32Type, mlir::BFloat16Type>>
race:lld::lldMain
race:llvm::*
race:xla::gpu::GpuExecutable::ExecuteAsyncOnStream
6 changes: 6 additions & 0 deletions xla/literal.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,7 @@ limitations under the License.
#include <initializer_list>
#include <limits>
#include <memory>
#include <mutex>
#include <optional>
#include <ostream>
#include <string>
Expand Down Expand Up @@ -1107,6 +1108,7 @@ class LiteralBase {
Storage(Storage&& other) { *this = std::move(other); }
Storage& operator=(Storage&& other) {
rep_ = std::move(other.rep_);
std::lock_guard<std::mutex> lock(mutex_);
data_ = other.data_;

if (auto* inline_rep = GetDenseInlinedRep()) {
Expand All @@ -1127,6 +1129,7 @@ class LiteralBase {
template <typename Rep, typename... Args>
Rep& Emplace(Args... args) {
Rep& emplaced = rep_.emplace<Rep>(std::forward<Args>(args)...);
std::lock_guard<std::mutex> lock(mutex_);
if constexpr (std::is_same_v<Rep, DenseRep> ||
std::is_same_v<Rep, DenseInlinedRep>) {
data_ = emplaced.data;
Expand Down Expand Up @@ -1157,11 +1160,13 @@ class LiteralBase {
TupleRep* GetTupleRep() { return std::get_if<TupleRep>(&rep_); }

const char* data() const {
std::lock_guard<std::mutex> lock(mutex_);
DCHECK_EQ(dense_data(), data_) << "cached data pointer is stale";
return data_;
}

char* data() {
std::lock_guard<std::mutex> lock(mutex_);
DCHECK_EQ(dense_data(), data_) << "cached data pointer is stale";
return data_;
}
Expand All @@ -1174,6 +1179,7 @@ class LiteralBase {
}

std::variant<Uninitialized, TupleRep, DenseRep, DenseInlinedRep> rep_;
mutable std::mutex mutex_;
char* data_ = nullptr; // cached `rep_.data` value for dense reps
};

Expand Down