Break prof_accum into multiple compilation units.

Break prof_accum into multiple compilation units, in order to thwart
compiler optimizations such as inlining and tail call optimization that
would alter backtraces.

Makefile.in

@@ -128,6 +128,8 @@ TESTS_UNIT := $(srcroot)test/unit/bitmap.c \
 	$(srcroot)test/unit/tsd.c \
 	$(srcroot)test/unit/util.c \
 	$(srcroot)test/unit/zero.c
+TESTS_UNIT_AUX := $(srcroot)test/unit/prof_accum_a.c \
+	$(srcroot)test/unit/prof_accum_b.c
 TESTS_INTEGRATION := $(srcroot)test/integration/aligned_alloc.c \
 	$(srcroot)test/integration/allocated.c \
 	$(srcroot)test/integration/mallocx.c \
@@ -155,9 +157,10 @@ C_TESTLIB_STRESS_OBJS := $(C_TESTLIB_SRCS:$(srcroot)%.c=$(objroot)%.stress.$(O))
 C_TESTLIB_OBJS := $(C_TESTLIB_UNIT_OBJS) $(C_TESTLIB_INTEGRATION_OBJS) $(C_UTIL_INTEGRATION_OBJS) $(C_TESTLIB_STRESS_OBJS)
 
 TESTS_UNIT_OBJS := $(TESTS_UNIT:$(srcroot)%.c=$(objroot)%.$(O))
+TESTS_UNIT_AUX_OBJS := $(TESTS_UNIT_AUX:$(srcroot)%.c=$(objroot)%.$(O))
 TESTS_INTEGRATION_OBJS := $(TESTS_INTEGRATION:$(srcroot)%.c=$(objroot)%.$(O))
 TESTS_STRESS_OBJS := $(TESTS_STRESS:$(srcroot)%.c=$(objroot)%.$(O))
-TESTS_OBJS := $(TESTS_UNIT_OBJS) $(TESTS_INTEGRATION_OBJS) $(TESTS_STRESS_OBJS)
+TESTS_OBJS := $(TESTS_UNIT_OBJS) $(TESTS_UNIT_AUX_OBJS) $(TESTS_INTEGRATION_OBJS) $(TESTS_STRESS_OBJS)
 
 .PHONY: all dist build_doc_html build_doc_man build_doc
 .PHONY: install_bin install_include install_lib
@@ -206,6 +209,12 @@ $(C_TESTLIB_STRESS_OBJS): $(objroot)test/src/%.stress.$(O): $(srcroot)test/src/%
 $(C_TESTLIB_STRESS_OBJS): CPPFLAGS += -DJEMALLOC_STRESS_TEST -DJEMALLOC_STRESS_TESTLIB
 $(C_TESTLIB_OBJS): CPPFLAGS += -I$(srcroot)test/include -I$(objroot)test/include
 $(TESTS_UNIT_OBJS): CPPFLAGS += -DJEMALLOC_UNIT_TEST
+$(TESTS_UNIT_AUX_OBJS): CPPFLAGS += -DJEMALLOC_UNIT_TEST
+define make-unit-link-dep
+$(1): TESTS_UNIT_LINK_OBJS += $(2)
+$(1): $(2)
+endef
+$(foreach test, $(TESTS_UNIT:$(srcroot)test/unit/%.c=$(objroot)test/unit/%$(EXE)), $(eval $(call make-unit-link-dep,$(test),$(filter $(test:%=%_a.$(O)) $(test:%=%_b.$(O)),$(TESTS_UNIT_AUX_OBJS)))))
 $(TESTS_INTEGRATION_OBJS): CPPFLAGS += -DJEMALLOC_INTEGRATION_TEST
 $(TESTS_STRESS_OBJS): CPPFLAGS += -DJEMALLOC_STRESS_TEST
 $(TESTS_OBJS): $(objroot)test/%.$(O): $(srcroot)test/%.c
@@ -248,7 +257,7 @@ $(STATIC_LIBS):
 	@mkdir -p $(@D)
 	$(AR) $(ARFLAGS)@AROUT@ $+
 
-$(objroot)test/unit/%$(EXE): $(objroot)test/unit/%.$(O) $(C_JET_OBJS) $(C_TESTLIB_UNIT_OBJS)
+$(objroot)test/unit/%$(EXE): $(objroot)test/unit/%.$(O) $(TESTS_UNIT_LINK_OBJS) $(C_JET_OBJS) $(C_TESTLIB_UNIT_OBJS)
 	@mkdir -p $(@D)
 	$(CC) $(LDTARGET) $(filter %.$(O),$^) $(call RPATH,$(objroot)lib) $(LDFLAGS) $(filter-out -lm,$(LIBS)) -lm $(EXTRA_LDFLAGS)
 
@@ -359,7 +368,7 @@ coverage: check
 	$(SHELL) $(srcroot)coverage.sh $(srcroot)test/src unit $(C_TESTLIB_UNIT_OBJS)
 	$(SHELL) $(srcroot)coverage.sh $(srcroot)test/src integration $(C_TESTLIB_INTEGRATION_OBJS)
 	$(SHELL) $(srcroot)coverage.sh $(srcroot)test/src stress $(C_TESTLIB_STRESS_OBJS)
-	$(SHELL) $(srcroot)coverage.sh $(srcroot)test/unit unit $(TESTS_UNIT_OBJS)
+	$(SHELL) $(srcroot)coverage.sh $(srcroot)test/unit unit $(TESTS_UNIT_OBJS) $(TESTS_UNIT_AUX_OBJS)
 	$(SHELL) $(srcroot)coverage.sh $(srcroot)test/integration integration $(TESTS_INTEGRATION_OBJS)
 	$(SHELL) $(srcroot)coverage.sh $(srcroot)test/stress integration $(TESTS_STRESS_OBJS)
 endif

test/unit/prof_accum.c

@@ -1,9 +1,4 @@
-#include "test/jemalloc_test.h"
+#include "prof_accum.h"
-
-#define	NTHREADS		4
-#define	NALLOCS_PER_THREAD	50
-#define	DUMP_INTERVAL		1
-#define	BT_COUNT_CHECK_INTERVAL	5
 
 #ifdef JEMALLOC_PROF
 const char *malloc_conf =
@@ -21,37 +16,6 @@ prof_dump_open_intercept(bool propagate_err, const char *filename)
 	return (fd);
 }
 
-#define	alloc_n_proto(n)						\
-JEMALLOC_NOINLINE static void	*alloc_##n(unsigned bits);
-
-#define	alloc_n_gen(n)							\
-JEMALLOC_NOINLINE static void *						\
-alloc_##n(unsigned bits)						\
-{									\
-	void *p;							\
-									\
-	if (bits == 0)							\
-		p = mallocx(1, 0);					\
-	else {								\
-		switch (bits & 0x1U) {					\
-		case 0:							\
-			p = alloc_0(bits >> 1);				\
-			break;						\
-		case 1:							\
-			p = alloc_1(bits >> 1);				\
-			break;						\
-		default: not_reached();					\
-		}							\
-	}								\
-	/* Intentionally sabotage tail call optimization. */		\
-	assert_ptr_not_null(p, "Unexpected mallocx() failure");		\
-	return (p);							\
-}
-alloc_n_proto(0)
-alloc_n_proto(1)
-alloc_n_gen(0)
-alloc_n_gen(1)
-
 static void *
 alloc_from_permuted_backtrace(unsigned thd_ind, unsigned iteration)
 {

test/unit/prof_accum.h

@@ -0,0 +1,29 @@
+#include "test/jemalloc_test.h"
+
+#define	NTHREADS		4
+#define	NALLOCS_PER_THREAD	50
+#define	DUMP_INTERVAL		1
+#define	BT_COUNT_CHECK_INTERVAL	5
+
+#define	alloc_n_proto(n)						\
+void	*alloc_##n(unsigned bits);
+alloc_n_proto(0)
+alloc_n_proto(1)
+
+#define	alloc_n_gen(n)							\
+void *									\
+alloc_##n(unsigned bits)						\
+{									\
+									\
+	if (bits == 0) {						\
+		void *p = mallocx(1, 0);				\
+		assert_ptr_not_null(p, "Unexpected mallocx() failure");	\
+		return (p);						\
+	} else {							\
+		switch (bits & 0x1U) {					\
+		case 0: return (alloc_0(bits >> 1));			\
+		case 1: return (alloc_1(bits >> 1));			\
+		default: not_reached();					\
+		}							\
+	}								\
+}

test/unit/prof_accum_a.c

@@ -0,0 +1,3 @@
+#include "prof_accum.h"
+
+alloc_n_gen(0)

test/unit/prof_accum_b.c

@@ -0,0 +1,3 @@
+#include "prof_accum.h"
+
+alloc_n_gen(1)