Completes migration of iree/base/threading and adds comprehensive tests.

Finishes the threading module migration by removing backward-compatibility
shim headers and updating all consumers to use direct includes. This makes
dependencies explicit and removes the indirection that was masking the
true module structure.

Adds a full test and benchmark suite:
- Mutex: correctness under contention, try_lock semantics, performance
  characteristics of slim vs fat mutex across thread counts
- Notification: producer/consumer patterns, timeout behavior, spurious
  wakeup handling, throughput under varying contention
- Thread: creation/join lifecycle, thread-local storage, affinity,
  priority inheritance behavior
- Call-once: concurrent initialization races, exception safety

Co-Authored-By: Claude <noreply@anthropic.com>

Author: benvanik

runtime/bindings/python/loop.cc

@@ -10,7 +10,7 @@
 #include <vector>
 
 #include "./hal.h"
-#include "iree/base/internal/synchronization.h"
+#include "iree/base/threading/mutex.h"
 
 namespace iree::python {
 

runtime/bindings/tflite/BUILD.bazel

@@ -41,7 +41,6 @@ iree_runtime_cc_library(
     deps = [
         "//runtime/src/iree/base",
         "//runtime/src/iree/base/internal",
-        "//runtime/src/iree/base/internal:synchronization",
         "//runtime/src/iree/base/threading",
         "//runtime/src/iree/hal",
         "//runtime/src/iree/hal/drivers",

runtime/bindings/tflite/CMakeLists.txt

@@ -31,7 +31,7 @@ iree_cc_library(
   DEPS
     iree::base
     iree::base::internal
-    iree::base::internal::synchronization
+    iree::base::threading
     iree::hal
     iree::hal::drivers
     iree::modules::hal

runtime/src/iree/base/internal/BUILD.bazel

@@ -75,7 +75,6 @@ iree_runtime_cc_library(
     deps = [
         ":atomic_slist",
         ":internal",
-        ":synchronization",
         "//runtime/src/iree/base",
     ],
 )
@@ -86,8 +85,8 @@ iree_runtime_cc_library(
     hdrs = ["atomic_slist.h"],
     deps = [
         ":internal",
-        ":synchronization",
         "//runtime/src/iree/base:core_headers",
+        "//runtime/src/iree/base/threading",
     ],
 )
 
@@ -133,7 +132,6 @@ cc_binary_benchmark(
     ],
 )
 
-
 iree_runtime_cc_library(
     name = "cpu",
     srcs = ["cpu.c"],
@@ -155,7 +153,6 @@ iree_runtime_cc_library(
     deps = [
         ":internal",
         ":path",
-        ":synchronization",
         "//build_tools:dl",
         "//build_tools:pthreads",
         "//runtime/src/iree/base",
@@ -305,33 +302,6 @@ iree_runtime_cc_library(
     hdrs = ["span.h"],
 )
 
-iree_runtime_cc_library(
-    name = "synchronization",
-    hdrs = ["synchronization.h"],
-    deps = ["//runtime/src/iree/base/threading"],
-)
-
-cc_binary_benchmark(
-    name = "synchronization_benchmark",
-    testonly = True,
-    srcs = ["synchronization_benchmark.cc"],
-    deps = [
-        ":synchronization",
-        "//runtime/src/iree/testing:benchmark_main",
-        "@com_google_benchmark//:benchmark",
-    ],
-)
-
-iree_runtime_cc_test(
-    name = "synchronization_test",
-    srcs = ["synchronization_test.cc"],
-    deps = [
-        ":synchronization",
-        "//runtime/src/iree/testing:gtest",
-        "//runtime/src/iree/testing:gtest_main",
-    ],
-)
-
 iree_runtime_cc_library(
     name = "time",
     srcs = [
@@ -392,7 +362,6 @@ iree_runtime_cc_library(
     ],
     hdrs = ["wait_handle.h"],
     deps = [
-        ":synchronization",
         "//runtime/src/iree/base",
         "//runtime/src/iree/base:core_headers",
         "//runtime/src/iree/base/threading",
@@ -428,30 +397,9 @@ iree_runtime_cc_library(
     hdrs = ["event_pool.h"],
     deps = [
         ":internal",
-        ":synchronization",
         ":wait_handle",
         "//runtime/src/iree/base",
         "//runtime/src/iree/base:core_headers",
-    ],
-)
-
-iree_runtime_cc_library(
-    name = "threading",
-    hdrs = [
-        "threading.h",
-        "threading_impl.h",
-    ],
-    deps = ["//runtime/src/iree/base/threading:thread"],
-)
-
-iree_runtime_cc_test(
-    name = "threading_test",
-    srcs = ["threading_test.cc"],
-    deps = [
-        ":internal",
-        ":synchronization",
-        ":threading",
-        "//runtime/src/iree/testing:gtest",
-        "//runtime/src/iree/testing:gtest_main",
+        "//runtime/src/iree/base/threading",
     ],
 )

runtime/src/iree/base/internal/CMakeLists.txt

@@ -61,7 +61,6 @@ iree_cc_library(
   DEPS
     ::atomic_slist
     ::internal
-    ::synchronization
     iree::base
   PUBLIC
 )
@@ -75,8 +74,8 @@ iree_cc_library(
     "atomic_slist.c"
   DEPS
     ::internal
-    ::synchronization
     iree::base::core_headers
+    iree::base::threading
   PUBLIC
 )
 
@@ -154,7 +153,6 @@ iree_cc_library(
     ${CMAKE_DL_LIBS}
     ::internal
     ::path
-    ::synchronization
     iree::base
     iree::base::threading
   PUBLIC
@@ -331,39 +329,6 @@ iree_cc_library(
   PUBLIC
 )
 
-iree_cc_library(
-  NAME
-    synchronization
-  HDRS
-    "synchronization.h"
-  DEPS
-    iree::base::threading
-  PUBLIC
-)
-
-iree_cc_binary_benchmark(
-  NAME
-    synchronization_benchmark
-  SRCS
-    "synchronization_benchmark.cc"
-  DEPS
-    ::synchronization
-    benchmark
-    iree::testing::benchmark_main
-  TESTONLY
-)
-
-iree_cc_test(
-  NAME
-    synchronization_test
-  SRCS
-    "synchronization_test.cc"
-  DEPS
-    ::synchronization
-    iree::testing::gtest
-    iree::testing::gtest_main
-)
-
 iree_cc_library(
   NAME
     time
@@ -427,7 +392,6 @@ iree_cc_library(
     "wait_handle_posix.h"
     "wait_handle_win32.c"
   DEPS
-    ::synchronization
     iree::base
     iree::base::core_headers
     iree::base::threading
@@ -458,37 +422,13 @@ iree_cc_library(
     "event_pool.c"
   DEPS
     ::internal
-    ::synchronization
     ::wait_handle
     iree::base
     iree::base::core_headers
+    iree::base::threading
   PUBLIC
 )
 
-iree_cc_library(
-  NAME
-    threading
-  HDRS
-    "threading.h"
-    "threading_impl.h"
-  DEPS
-    iree::base::threading::thread
-  PUBLIC
-)
-
-iree_cc_test(
-  NAME
-    threading_test
-  SRCS
-    "threading_test.cc"
-  DEPS
-    ::internal
-    ::synchronization
-    ::threading
-    iree::testing::gtest
-    iree::testing::gtest_main
-)
-
 ### BAZEL_TO_CMAKE_PRESERVES_ALL_CONTENT_BELOW_THIS_LINE ###
 
 if(EMSCRIPTEN)

runtime/src/iree/base/internal/atomic_slist.h

@@ -17,7 +17,7 @@
 
 #include "iree/base/alignment.h"
 #include "iree/base/internal/atomics.h"
-#include "iree/base/internal/synchronization.h"
+#include "iree/base/threading/mutex.h"
 
 #ifdef __cplusplus
 extern "C" {

runtime/src/iree/base/internal/event_pool.c

@@ -10,7 +10,7 @@
 #include <stddef.h>
 #include <string.h>
 
-#include "iree/base/internal/synchronization.h"
+#include "iree/base/threading/mutex.h"
 
 struct iree_event_pool_t {
   // Allocator used to create the event pool.

runtime/src/iree/base/internal/memory.h

@@ -72,6 +72,31 @@ void iree_memory_flush_icache(void* base_address, iree_host_size_t length);
 // C11 aligned_alloc shim
 //===----------------------------------------------------------------------===//
 
+// !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
+// WARNING: DO NOT USE THIS FUNCTION FOR NORMAL ALLOCATIONS.
+// !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
+//
+// This function BYPASSES the iree_allocator_t infrastructure entirely. Using it
+// means:
+//   - No integration with custom allocators (arena, pool, tracking, etc)
+//   - No ability to use iree_allocator_free() - must use iree_aligned_free()
+//   - Memory accounting and debugging tools cannot track these allocations
+//   - Breaks the allocator abstraction that enables embedded/custom deployments
+//
+// VALID USES (require explicit approval):
+//   - NUMA allocation fallback (numa_*.c) - special physical placement needs
+//   - Status storage (status.c internal) - bootstrapping before allocators
+//     exist and that allocators use for handling when they fail
+//   - Test/benchmark code where allocator integration is irrelevant
+//
+// FOR ALIGNED ALLOCATIONS IN NORMAL CODE:
+//   - Embed aligned data within a larger struct that is normally allocated
+//   - Use iree_allocator_malloc with extra padding and manual alignment
+//   - Reconsider whether 64-byte alignment is actually necessary
+//
+// If you think you need this function, you are almost certainly wrong.
+// Ask for guidance before using it.
+
 // Allocates |size| bytes of uninitialized storage whose alignment is specified
 // by |alignment|. The |size| parameter must be an integral multiple of
 // |alignment|. The returned pointer must be freed using iree_aligned_free.