patina_bench: Misc tweaks

Signed-off-by: msdx321 <[email protected]>

src/components/implementation/tests/bench_sched_yield/bench_sched_yield.c

@@ -17,36 +17,47 @@
 
 /* lo and hi is actually running at the same prio */
 #define ITERATION 10000
-/* #define PRINT_ALL */
+#define PRINT_ALL
 
 thdid_t yield_hi = 0, yield_lo = 0;
+
 volatile cycles_t start;
-volatile cycles_t end;
+volatile int count;
 
 struct perfdata perf;
 cycles_t result[ITERATION] = {0, };
 
 /***
- * We're measuring 2-way context switch time. 
+ * We're measuring one-way context switch time. 
  */
 void
 yield_hi_thd(void *d)
 {
-	/* Never stops running; low priority controls how many iters to run. */
-	while (1) {
+	cycles_t end;
+
+	while (count < ITERATION) {
 		debug("h1,");
+
+		start = time_now();
 		sched_thd_yield_to(yield_lo);
+		end = time_now();
+
 		debug("h2,");
+
+		perfdata_add(&perf, end - start);
+
+		count++;
 	}
+
+	while (1) ;
 }
 
 void
 yield_lo_thd(void *d)
 {
-	int i;
-	int first = 0;
+	cycles_t end;
 
-	for (i = 0; i < ITERATION + 1; i++) {
+	while (count < ITERATION) {
 		debug("l1,");
 
 		start = time_now();
@@ -55,16 +66,16 @@ yield_lo_thd(void *d)
 
 		debug("l2,");
 
-		if (first == 0) first = 1;
-		else perfdata_add(&perf, end - start);
+		perfdata_add(&perf, end - start);
+
+		count++;
 	}
 
-	perfdata_calc(&perf);
 #ifdef PRINT_ALL
-	perfdata_all(&perf);
-#else
-	perfdata_print(&perf);
+	perfdata_raw(&perf);
 #endif
+	perfdata_calc(&perf);
+	perfdata_print(&perf);
 
 	while (1) ;
 }
@@ -77,6 +88,8 @@ test_yield(void)
 		SCHED_PARAM_CONS(SCHEDP_PRIO, 6)
 	};
 
+	count = 0;
+
 	perfdata_init(&perf, "Context switch time", result, ITERATION);
 
 	printc("Create threads:\n");

src/components/implementation/tests/bench_sl_yield/bench_sl_yield.c

@@ -19,7 +19,7 @@
 
 /* lo and hi is actually running at the same prio */
 #define ITERATION 10000
-/* #define PRINT_ALL */
+#define PRINT_ALL
 
 /* Ensure this is the same as what is in sl_mod_fprr.c */
 #define SL_FPRR_NPRIOS 32
@@ -32,29 +32,40 @@ struct sl_thd *testing_thread;
 thdid_t thdid1, thdid2;
 
 volatile cycles_t start;
-volatile cycles_t end;
+volatile int count;
 
 struct perfdata perf;
 cycles_t result[ITERATION] = {0, };
 
 static void
 thd1_fn()
 {
-	/* Never stops running; low priority controls how many iters to run. */
-	while (1) {
+	cycles_t end;
+
+	while (count < ITERATION) {
 		debug("h1,");
+
+		start = time_now();
 		sl_thd_yield(thdid2);
+		end = time_now();
+
 		debug("h2,");
+
+		perfdata_add(&perf, end - start);
+
+		count++;
 	}
+
+	while (1); 
 }
 
 static void
 thd2_fn()
 {
 	int i;
-	int first = 0;
+	cycles_t end;
 
-	for (i = 0; i < ITERATION + 1; i++) {
+	while (count < ITERATION) {
 		debug("l1,");
 
 		start = time_now();
@@ -63,16 +74,16 @@ thd2_fn()
 
 		debug("l2,");
 
-		if (first == 0) first = 1;
-		else perfdata_add(&perf, end - start);
+		perfdata_add(&perf, end - start);
+
+		count++;
 	}
 
-	perfdata_calc(&perf);
 #ifdef PRINT_ALL
-	perfdata_all(&perf);
-#else
-	perfdata_print(&perf);
+	perfdata_raw(&perf);
 #endif
+	perfdata_calc(&perf);
+	perfdata_print(&perf);
 
 	while (1) ;
 }
@@ -104,6 +115,8 @@ cos_init(void)
 	struct cos_defcompinfo *defci = cos_defcompinfo_curr_get();
 	struct cos_compinfo    *ci    = cos_compinfo_get(defci);
 
+	count = 0;
+
 	PRINTC("Thread switch benchmark for the scheduling library (sl)\n");
 	cos_meminfo_init(&(ci->mi), BOOT_MEM_KM_BASE, COS_MEM_KERN_PA_SZ, BOOT_CAPTBL_SELF_UNTYPED_PT);
 	cos_defcompinfo_init();

src/components/implementation/tests/patina_chan_bench/patina_chan_bench.c

@@ -10,6 +10,20 @@
 #include <patina.h>
 #include <perfdata.h>
 
+#define COLD_CACHE
+#ifdef COLD_CACHE
+#define cache_flush() __cache_flush()
+#define COLD_OFFSET 1
+#define COLD_INDEX 0
+#else
+#define cache_flush()
+#define COLD_OFFSET 0
+#define COLD_INDEX -1
+#endif
+
+#define CACHE_SIZE 512 * 1024
+#define CACHE_LINE_SIZE 32
+
 #undef PATINA_CHAN_TRACE_DEBUG
 #ifdef PATINA_CHAN_TRACE_DEBUG
 #define debug(format, ...) printc(format, ##__VA_ARGS__)
@@ -18,11 +32,14 @@
 #endif
 
 /* One low-priority thread and one high-priority thread contends on the lock */
+#ifdef COLD_CACHE
+#define ITERATION 10 * 10
+#else
 #define ITERATION 10 * 1000
-#define PRINT_ALL
+#endif
+#undef PRINT_ALL
 
 /* Two options are available: Sender at low/high prio, data words 4 */
-#undef READER_HIGH
 #define DATA_WORDS 2
 
 thdid_t chan_reader = 0, chan_writer = 0;
@@ -60,6 +77,20 @@ patina_chan_r_t rid2;
 patina_chan_s_t sid;
 patina_chan_s_t sid2;
 
+volatile char pool[CACHE_SIZE * 4] = {
+  0,
+};
+
+void
+__cache_flush()
+{
+	int agg = 1;
+	for (int i = 0; i < CACHE_SIZE * 4; i += CACHE_LINE_SIZE) {
+		pool[i] += agg;
+		agg = pool[i];
+	}
+}
+
 /***
  * The two threads reciprocally sends and receives.
  */
@@ -84,10 +115,10 @@ void
 chan_writer_thd(void *d)
 {
 	int i;
-	int first = 0;
 
-	for (int i = 0; i < ITERATION + 1; i++) {
+	for (int i = 0; i < ITERATION + COLD_OFFSET; i++) {
 		debug("w1,");
+		cache_flush();
 		ts1[0] = time_now();
 		debug("ts1: %d,", ts1[0]);
 		debug("w2,");
@@ -99,35 +130,25 @@ chan_writer_thd(void *d)
 		ts3[0] = time_now();
 		debug("w5,");
 
-		if (first == 0)
-			first = 1;
-		else {
-			if (ts2[0] > ts1[0] && ts3[0] > ts2[0]) {
-				perfdata_add(&perf1, ts2[0] - ts1[0]);
-				perfdata_add(&perf2, ts3[0] - ts2[0]);
-				perfdata_add(&perf3, ts3[0] - ts1[0]);
-			}
+		if (ts2[0] > ts1[0] && ts3[0] > ts2[0] && i != COLD_INDEX) {
+			perfdata_add(&perf1, ts2[0] - ts1[0]);
+			perfdata_add(&perf2, ts3[0] - ts2[0]);
+			perfdata_add(&perf3, ts3[0] - ts1[0]);
 		}
 	}
 
+#ifdef PRINT_ALL
+	perfdata_raw(&perf1);
+	perfdata_raw(&perf2);
+	perfdata_raw(&perf3);
+#endif
 	perfdata_calc(&perf1);
 	perfdata_calc(&perf2);
 	perfdata_calc(&perf3);
-#ifdef PRINT_ALL
-#ifdef READER_HIGH
-	perfdata_all(&perf1);
-#else
-	perfdata_all(&perf2);
-#endif
-	perfdata_all(&perf3);
-#else
-#ifdef READER_HIGH
+
 	perfdata_print(&perf1);
-#else
 	perfdata_print(&perf2);
-#endif
 	perfdata_print(&perf3);
-#endif
 
 	while (1)
 		;
@@ -137,18 +158,13 @@ void
 test_chan(void)
 {
 	int      i;
-	int      first = 0;
 	cycles_t begin, end;
 
-#ifdef READER_HIGH
 	sched_param_t sps[] = {SCHED_PARAM_CONS(SCHEDP_PRIO, 4), SCHED_PARAM_CONS(SCHEDP_PRIO, 6)};
-#else
-	sched_param_t sps[] = {SCHED_PARAM_CONS(SCHEDP_PRIO, 6), SCHED_PARAM_CONS(SCHEDP_PRIO, 4)};
-#endif
 
 	/* Uncontended lock taking/releasing */
 	perfdata_init(&perf1, "Uncontended channel - selfloop", result1, ITERATION);
-	for (i = 0; i < ITERATION + 1; i++) {
+	for (i = 0; i < ITERATION; i++) {
 		begin = time_now();
 
 		debug("send\n");
@@ -157,17 +173,13 @@ test_chan(void)
 		patina_channel_recv(rid, tmp, 1, 0);
 
 		end = time_now();
-		if (first == 0)
-			first = 1;
-		else
-			perfdata_add(&perf1, end - begin);
+		perfdata_add(&perf1, end - begin);
 	}
-	perfdata_calc(&perf1);
 #ifdef PRINT_ALL
-	perfdata_all(&perf1);
-#else
-	perfdata_print(&perf1);
+	perfdata_raw(&perf1);
 #endif
+	perfdata_calc(&perf1);
+	perfdata_print(&perf1);
 
 	perfdata_init(&perf1, "Contended channel - reader high use this", result1, ITERATION);
 	perfdata_init(&perf2, "Contended channel - writer high use this", result2, ITERATION);
@@ -176,11 +188,11 @@ test_chan(void)
 	printc("Create threads:\n");
 
 	chan_reader = sched_thd_create(chan_reader_thd, NULL);
-	printc("\tcreating reader thread %d at prio %d\n", chan_reader, sps[0]);
+	printc("\tcreating reader thread %d at prio %d\n", chan_reader, sps[1]);
 	sched_thd_param_set(chan_reader, sps[0]);
 
 	chan_writer = sched_thd_create(chan_writer_thd, NULL);
-	printc("\tcreating writer thread %d at prio %d\n", chan_writer, sps[1]);
+	printc("\tcreating writer thread %d at prio %d\n", chan_writer, sps[0]);
 	sched_thd_param_set(chan_writer, sps[1]);
 }
 

src/components/implementation/tests/patina_chan_bench_inter_recv/patina_chan_bench_inter_recv.c

@@ -22,19 +22,13 @@ patina_chan_r_t rid;
 patina_event_t  evt;
 
 /* Keep these settings below consistent with the sender side */
-#undef READER_HIGH
-#define USE_EVTMGR
+#undef USE_EVTMGR
 
 #define TEST_CHAN_ITEM_SZ sizeof(u32_t)
 #define TEST_CHAN_NSLOTS 2
 #define TEST_CHAN_SEND_ID 3
 #define TEST_CHAN_RECV_ID 4
-/* We are the receiver, and we don't care about data gathering */
-#ifdef READER_HIGH
 #define TEST_CHAN_PRIO_SELF 4
-#else
-#define TEST_CHAN_PRIO_SELF 5
-#endif
 
 typedef unsigned int cycles_32_t;