Add a benchmark of large allocations.

Makefile.in

@@ -289,6 +289,7 @@ CPP_SRCS :=
 TESTS_INTEGRATION_CPP :=
 endif
 TESTS_STRESS := $(srcroot)test/stress/microbench.c \
+	$(srcroot)test/stress/large_microbench.c \
 	$(srcroot)test/stress/hookbench.c
 
 

test/include/test/bench.h

@@ -0,0 +1,39 @@
+static inline void
+time_func(timedelta_t *timer, uint64_t nwarmup, uint64_t niter,
+    void (*func)(void)) {
+	uint64_t i;
+
+	for (i = 0; i < nwarmup; i++) {
+		func();
+	}
+	timer_start(timer);
+	for (i = 0; i < niter; i++) {
+		func();
+	}
+	timer_stop(timer);
+}
+
+static inline void
+compare_funcs(uint64_t nwarmup, uint64_t niter, const char *name_a,
+    void (*func_a), const char *name_b, void (*func_b)) {
+	timedelta_t timer_a, timer_b;
+	char ratio_buf[6];
+	void *p;
+
+	p = mallocx(1, 0);
+	if (p == NULL) {
+		test_fail("Unexpected mallocx() failure");
+		return;
+	}
+
+	time_func(&timer_a, nwarmup, niter, func_a);
+	time_func(&timer_b, nwarmup, niter, func_b);
+
+	timer_ratio(&timer_a, &timer_b, ratio_buf, sizeof(ratio_buf));
+	malloc_printf("%"FMTu64" iterations, %s=%"FMTu64"us, "
+	    "%s=%"FMTu64"us, ratio=1:%s\n",
+	    niter, name_a, timer_usec(&timer_a), name_b, timer_usec(&timer_b),
+	    ratio_buf);
+
+	dallocx(p, 0);
+}

test/stress/large_microbench.c

@@ -0,0 +1,33 @@
+#include "test/jemalloc_test.h"
+#include "test/bench.h"
+
+static void
+large_mallocx_free(void) {
+	/*
+	 * We go a bit larger than the large minclass on its own to better
+	 * expose costs from things like zeroing.
+	 */
+	void *p = mallocx(SC_LARGE_MINCLASS, MALLOCX_TCACHE_NONE);
+	assert_ptr_not_null(p, "mallocx shouldn't fail");
+	free(p);
+}
+
+static void
+small_mallocx_free(void) {
+	void *p = mallocx(16, 0);
+	assert_ptr_not_null(p, "mallocx shouldn't fail");
+	free(p);
+}
+
+TEST_BEGIN(test_large_vs_small) {
+	compare_funcs(100*1000, 1*1000*1000, "large mallocx",
+	    large_mallocx_free, "small mallocx", small_mallocx_free);
+}
+TEST_END
+
+int
+main(void) {
+	return test_no_reentrancy(
+	    test_large_vs_small);
+}
+

test/stress/microbench.c

@@ -1,44 +1,5 @@
 #include "test/jemalloc_test.h"
-
-static inline void
-time_func(timedelta_t *timer, uint64_t nwarmup, uint64_t niter,
-    void (*func)(void)) {
-	uint64_t i;
-
-	for (i = 0; i < nwarmup; i++) {
-		func();
-	}
-	timer_start(timer);
-	for (i = 0; i < niter; i++) {
-		func();
-	}
-	timer_stop(timer);
-}
-
-void
-compare_funcs(uint64_t nwarmup, uint64_t niter, const char *name_a,
-    void (*func_a), const char *name_b, void (*func_b)) {
-	timedelta_t timer_a, timer_b;
-	char ratio_buf[6];
-	void *p;
-
-	p = mallocx(1, 0);
-	if (p == NULL) {
-		test_fail("Unexpected mallocx() failure");
-		return;
-	}
-
-	time_func(&timer_a, nwarmup, niter, func_a);
-	time_func(&timer_b, nwarmup, niter, func_b);
-
-	timer_ratio(&timer_a, &timer_b, ratio_buf, sizeof(ratio_buf));
-	malloc_printf("%"FMTu64" iterations, %s=%"FMTu64"us, "
-	    "%s=%"FMTu64"us, ratio=1:%s\n",
-	    niter, name_a, timer_usec(&timer_a), name_b, timer_usec(&timer_b),
-	    ratio_buf);
-
-	dallocx(p, 0);
-}
+#include "test/bench.h"
 
 static void
 malloc_free(void) {