Fix GitHub Actions for threading.

PiperOrigin-RevId: 922926969
Change-Id: I968e6b130b6125dbc0ec65ba913d04292270b563

src/engine/engine_crossplatform.h

@@ -69,6 +69,16 @@
   #endif
 #endif
 
+// Atomics helper for size_t.
+#if defined(_MSC_VER) && !defined(__clang__)
+  #include <intrin.h>
+  #define mj_atomic_add_size_t(ptr, val) \
+      (size_t)_InterlockedExchangeAdd64((__int64 volatile*)(ptr), (__int64)(val))
+#else
+  #define mj_atomic_add_size_t(ptr, val) \
+      __atomic_fetch_add(ptr, val, __ATOMIC_RELAXED)
+#endif
+
 #ifdef __cplusplus
 extern "C" {
 #endif

src/engine/engine_memory.c

@@ -16,7 +16,6 @@
 
 #include <inttypes.h>  // IWYU pragma: keep
 #include <limits.h>
-#include <stdatomic.h>
 #include <stddef.h>
 #include <stdint.h>
 #include <stdio.h>
@@ -216,8 +215,7 @@ static inline void* stackalloc(mjData* d, size_t size, size_t alignment,
   // call in mju_dispatch: atomically reserve space on the stack
   if (d->threadlock) {
     size_t alloc_size = size + alignment - 1 + 2 * mjREDZONE;
-    size_t old_pstack = atomic_fetch_add_explicit(
-        (_Atomic size_t*)&d->pstack, alloc_size, memory_order_relaxed);
+    size_t old_pstack = mj_atomic_add_size_t(&d->pstack, alloc_size);
 
     // check for stack overflow
     size_t stack_available_bytes = (size_t)d->narena - d->parena;

src/engine/engine_thread.cc

@@ -24,6 +24,10 @@
 #include <mujoco/mjmodel.h>
 #include "engine/engine_memory.h"
 
+#if defined(__cpp_lib_atomic_wait)
+
+// ----------------------------- multithreaded implementation --------------------------------------
+
 // context for thread pool stored on mjData
 class ThreadPoolContext {
  public:
@@ -129,7 +133,6 @@ class ThreadPoolContext {
 };
 
 
-
 // create a thread pool with nthread threads
 void mju_threadpool(mjData* d, int nthread) {
   if (d->threadpool) {
@@ -188,3 +191,23 @@ int mju_numThread(const mjData* d) {
   ThreadPoolContext* ctx = reinterpret_cast<ThreadPoolContext*>(d->threadpool);
   return ctx ? ctx->ThreadCount() + 1 : 1;
 }
+
+#else
+
+// -------------------------------------- fallback  ------------------------------------------------
+
+void mju_threadpool(mjData* d, int nthread) {
+  // No-op
+}
+
+void mju_dispatch(const mjModel* m, mjData* d, mjTaskFunc func, void* arg, int ntask) {
+  for (int i = 0; i < ntask; i++) {
+    func(m, d, arg, 0, i);
+  }
+}
+
+int mju_numThread(const mjData* d) {
+  return 1;
+}
+
+#endif  // __cpp_lib_atomic_wait