Merge branch 'feature/upstream' into develop

1 files changed, 27 insertions, 11 deletions

diff --git a/src/tc-time.c b/src/tc-time.c
index e842e77..e86ed67 100644
--- a/src/tc-time.c
+++ b/src/tc-time.c
@@ -5,6 +5,7 @@
  *	modify it under the terms of the WTF Public License version 2 or
  *	(at your option) any later version.
  */
+#include <assert.h>
 #include <math.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -36,10 +37,12 @@ static const struct timespec pairs[] = {
 };
 
 /*
- * Variant that uses full 64 bit division and is thus slower on
- * a handful of hardware.
+ * The playing fields it not level at all!
+ * HX_timespec_add_DIVQ benefits from inlining and near jumps,
+ * while HX_timespec_add has to go via PLT, and also has the PIC tax.
+ * It is actually worse by 7% on i7-8250U/-m64.
  */
-static struct timespec *HX_timespec_add_FDIV(struct timespec *r,
+static struct timespec *HX_timespec_add_DIVQ(struct timespec *r,
     const struct timespec *a, const struct timespec *b)
 {
 	long long p, q;
@@ -105,6 +108,20 @@ HX_timespec_mulf_S(struct timespec *r, const struct timespec *a, double f)
 	return r;
 }
 
+static void test_basic(void)
+{
+	struct timeval a = {1, 769298}, b = {2, 430520}, c;
+	struct timespec p = {1, 769298000}, q = {2, 430520000}, r;
+	HX_timeval_sub(&c, &b, &a);
+	HX_timespec_sub(&r, &q, &p);
+	printf(HX_TIMEVAL_FMT "\n", HX_TIMEVAL_EXP(&c));
+	printf(HX_TIMESPEC_FMT "\n", HX_TIMESPEC_EXP(&r));
+	assert(c.tv_sec == 0);
+	assert(c.tv_usec == 661222);
+	assert(r.tv_sec == 0);
+	assert(r.tv_nsec == 661222000);
+}
+
 static void test_same(void)
 {
 	static const struct timespec zero = {0, 0};
@@ -205,9 +222,9 @@ static void test_add(void)
 	for (a = pairs; a < pairs + ARRAY_SIZE(pairs); ++a) {
 		for (b = pairs; b < pairs + ARRAY_SIZE(pairs); ++b) {
 			HX_timespec_add(&r, a, b);
-			print_op2(&r, a, "+N", b);
-			HX_timespec_add_FDIV(&s, a, b);
-			print_op2(&r, a, "+F", b);
+			print_op2(&r, a, "+L", b);
+			HX_timespec_add_DIVQ(&s, a, b);
+			print_op2(&r, a, "+Q", b);
 			if (r.tv_sec != s.tv_sec || r.tv_nsec != s.tv_nsec)
 				abort();
 			HX_timespec_sub(&r, a, b);
@@ -288,7 +305,7 @@ static void test_adds(void)
 {
 	printf("# Test addition speed\n");
 	test_adds_1("normal:  ", HX_timespec_add);
-	test_adds_1("fulldiv: ", HX_timespec_add_FDIV);
+	test_adds_1("div64:   ", HX_timespec_add_DIVQ);
 	printf("\n");
 }
 
@@ -335,11 +352,10 @@ static void test_mul(void)
 static void test_muls_1i(const char *text, mul_func_t fn)
 {
 	struct timespec r, s, start, delta;
-	unsigned int i;
 
 	printf("%s", text);
 	clock_gettime(clock_id, &start);
-	for (i = 0; i < step_mul; ++i) {
+	for (time_t i = 0; i < step_mul; ++i) {
 		r.tv_sec  = -i;
 		r.tv_nsec = -i / 4;
 		(*fn)(&s, &r, 7);
@@ -352,11 +368,10 @@ static void test_muls_1i(const char *text, mul_func_t fn)
 static void test_muls_1f(const char *text, mulf_func_t fn)
 {
 	struct timespec r, s, start, delta;
-	unsigned int i;
 
 	printf("%s", text);
 	clock_gettime(clock_id, &start);
-	for (i = 0; i < step_mul; ++i) {
+	for (time_t i = 0; i < step_mul; ++i) {
 		r.tv_sec  = -i;
 		r.tv_nsec = -i / 4;
 		(*fn)(&s, &r, 7);
@@ -382,6 +397,7 @@ int main(void)
 	if (HX_init() <= 0)
 		abort();
 
+	test_basic();
 	test_same();
 	test_neg();
 	test_add();