threadpool: improve abort handling

Do not use threadpool->ec (exit code) to decide whether to exit the compute loop.
threadpool->ec is not atomic which makes thread-sanitizer rightfully unhappy about it.

We already have an atomic threadpool->stop flag for this which is consitent with the
rest of the code.

While at it add an explicit atomic_load for n_threads_cur for consistency.

Author: max-krasnyansky

ggml/src/ggml.c

@@ -19928,34 +19928,33 @@ struct ggml_cplan ggml_graph_plan(
 
 static thread_ret_t ggml_graph_compute_thread(void * data) {
     struct ggml_compute_state * state = (struct ggml_compute_state *) data;
+    struct ggml_threadpool    * tp    = state->threadpool;
 
-    const struct ggml_cgraph * cgraph = state->threadpool->cgraph;
-    const struct ggml_cplan  * cplan  = state->threadpool->cplan;
+    const struct ggml_cgraph * cgraph = tp->cgraph;
+    const struct ggml_cplan  * cplan  = tp->cplan;
 
     set_numa_thread_affinity(state->ith);
 
     struct ggml_compute_params params = {
         /*.ith       =*/ state->ith,
-        /*.nth       =*/ state->threadpool->n_threads_cur,
+        /*.nth       =*/ atomic_load_explicit(&tp->n_threads_cur, memory_order_relaxed),
         /*.wsize     =*/ cplan->work_size,
         /*.wdata     =*/ cplan->work_data,
-        /*.threadpool=*/ state->threadpool,
+        /*.threadpool=*/ tp,
     };
 
-    for (int node_n = 0; node_n < cgraph->n_nodes; node_n++) {
+    for (int node_n = 0; node_n < cgraph->n_nodes && !tp->stop; node_n++) {
         struct ggml_tensor * node = cgraph->nodes[node_n];
 
         ggml_compute_forward(&params, node);
 
-        if (state->ith == 0 && cplan->abort_callback && cplan->abort_callback(cplan->abort_callback_data)) {
-            state->threadpool->ec = GGML_STATUS_ABORTED;
+        if (state->ith == 0 && cplan->abort_callback &&
+                cplan->abort_callback(cplan->abort_callback_data)) {
+            tp->stop = true;
+            tp->ec   = GGML_STATUS_ABORTED;
         }
 
         ggml_barrier(state->threadpool);
-
-        if (state->threadpool->ec != GGML_STATUS_SUCCESS) {
-            break;
-        }
     }
 
     return 0;
@@ -20220,6 +20219,7 @@ enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cpl
         threadpool->cgraph           = cgraph;
         threadpool->cplan            = cplan;
         threadpool->current_chunk    = 0;
+        threadpool->stop             = false;
         threadpool->ec               = GGML_STATUS_SUCCESS;
     }