Include allocator module in LLVM_passes timer

compiler/rustc_codegen_ssa/src/back/write.rs

@@ -110,7 +110,6 @@ pub struct ModuleConfig {
     pub lint_llvm_ir: bool,
     pub no_prepopulate_passes: bool,
     pub no_builtins: bool,
-    pub time_module: bool,
     pub vectorize_loop: bool,
     pub vectorize_slp: bool,
     pub merge_functions: bool,
@@ -228,10 +227,6 @@ impl ModuleConfig {
             no_prepopulate_passes: sess.opts.cg.no_prepopulate_passes,
             no_builtins: no_builtins || sess.target.no_builtins,
 
-            // Exclude metadata and allocator modules from time_passes output,
-            // since they throw off the "LLVM passes" measurement.
-            time_module: if_regular!(true, false),
-
             // Copy what clang does by turning on loop vectorization at O2 and
             // slp vectorization at O3.
             vectorize_loop: !sess.opts.cg.no_vectorize_loops
@@ -1740,7 +1735,7 @@ fn spawn_work<'a, B: ExtraBackendMethods>(
     llvm_start_time: &mut Option<VerboseTimingGuard<'a>>,
     work: WorkItem<B>,
 ) {
-    if cgcx.config(work.module_kind()).time_module && llvm_start_time.is_none() {
+    if llvm_start_time.is_none() {
         *llvm_start_time = Some(cgcx.prof.verbose_generic_activity("LLVM_passes"));
     }
 

compiler/rustc_codegen_ssa/src/base.rs

@@ -681,33 +681,6 @@ pub fn codegen_crate<B: ExtraBackendMethods>(
 
     let ongoing_codegen = start_async_codegen(backend.clone(), tcx, target_cpu, autodiff_items);
 
-    // Codegen an allocator shim, if necessary.
-    if let Some(kind) = allocator_kind_for_codegen(tcx) {
-        let llmod_id =
-            cgu_name_builder.build_cgu_name(LOCAL_CRATE, &["crate"], Some("allocator")).to_string();
-        let module_llvm = tcx.sess.time("write_allocator_module", || {
-            backend.codegen_allocator(
-                tcx,
-                &llmod_id,
-                kind,
-                // If allocator_kind is Some then alloc_error_handler_kind must
-                // also be Some.
-                tcx.alloc_error_handler_kind(()).unwrap(),
-            )
-        });
-
-        ongoing_codegen.wait_for_signal_to_codegen_item();
-        ongoing_codegen.check_for_errors(tcx.sess);
-
-        // These modules are generally cheap and won't throw off scheduling.
-        let cost = 0;
-        submit_codegened_module_to_llvm(
-            &ongoing_codegen.coordinator,
-            ModuleCodegen::new_allocator(llmod_id, module_llvm),
-            cost,
-        );
-    }
-
     // For better throughput during parallel processing by LLVM, we used to sort
     // CGUs largest to smallest. This would lead to better thread utilization
     // by, for example, preventing a large CGU from being processed last and
@@ -823,6 +796,35 @@ pub fn codegen_crate<B: ExtraBackendMethods>(
         }
     }
 
+    // Codegen an allocator shim, if necessary.
+    // Do this last to ensure the LLVM_passes timer doesn't start while no CGUs have been codegened
+    // yet for the backend to optimize.
+    if let Some(kind) = allocator_kind_for_codegen(tcx) {
+        let llmod_id =
+            cgu_name_builder.build_cgu_name(LOCAL_CRATE, &["crate"], Some("allocator")).to_string();
+        let module_llvm = tcx.sess.time("write_allocator_module", || {
+            backend.codegen_allocator(
+                tcx,
+                &llmod_id,
+                kind,
+                // If allocator_kind is Some then alloc_error_handler_kind must
+                // also be Some.
+                tcx.alloc_error_handler_kind(()).unwrap(),
+            )
+        });
+
+        ongoing_codegen.wait_for_signal_to_codegen_item();
+        ongoing_codegen.check_for_errors(tcx.sess);
+
+        // These modules are generally cheap and won't throw off scheduling.
+        let cost = 0;
+        submit_codegened_module_to_llvm(
+            &ongoing_codegen.coordinator,
+            ModuleCodegen::new_allocator(llmod_id, module_llvm),
+            cost,
+        );
+    }
+
     ongoing_codegen.codegen_finished(tcx);
 
     // Since the main thread is sometimes blocked during codegen, we keep track