Adding a flag to control steps granularity in hope of better performance

CMakeLists.txt

@@ -20,6 +20,8 @@ add_library(ink lib.c include/ink.h)
 # Removes several checks to improve performance in cases where ink is used as a bytecode
 # add_definitions(-DNOEXTRACHECKS)
 
+add_definitions(-DINK_STEP_BATCH_COUNT=5)
+
 add_executable(ink_exe main.c)
 target_link_libraries(ink_exe PUBLIC ink)
 target_include_directories(ink PUBLIC include)

lib.c

@@ -26,6 +26,14 @@
 #define min(x, y) ((x) > (y) ? (y) : (x))
 #define max(x, y) ((x) < (y) ? (y) : (x))
 
+#ifdef __GNUC__
+#define likely(x)       __builtin_expect(!!(x), 1)
+#define unlikely(x)     __builtin_expect(!!(x), 0)
+#else
+#define likely(x) (!!(x))
+#define unlikely(x) (!!(x))
+#endif
+
 struct label {
     int active;
     int dest;
@@ -1071,35 +1079,41 @@ int ink_can_run(struct context* pContext) {
 }
 
 int ink_step_everyone(struct context* pContext) {
-    int out;
-    pContext->routine_current = -1;
-    for(;;) {
-        /* Increment to next runnable routine */
-        do{
-            ++(pContext->routine_current);
-        } while(
-                pContext->routine_current < pContext->routines_top
-                && pContext->routines[pContext->routine_current].panic != 0
-                && pContext->routines[pContext->routine_current].parse_error.is_set
-                && pContext->routines[pContext->routine_current].runtime_error.is_set
-                );
-        /* Exit condition */
-        if(pContext->routine_current >= pContext->routines_top) break;
-
-        /* Kill? */
-        if(pContext->routines[pContext->routine_current].panic == INK_ROUTINE_SUCCESS) {
-            ink_kill_routine(pContext, pContext->routine_current);
-        }
-
-        /* Step! */
-        out = ink_step(pContext);
-        if(out == 0) {
-            pContext->routines[pContext->routine_current].panic = INK_ROUTINE_SUCCESS;
-        } else if(out < 0) {
-            pContext->routines[pContext->routine_current].panic = out;
-        }
-    }
-    return 0;
+	int idx;
+	int out;
+	pContext->routine_current = -1;
+	for(;;) {
+		top_label:
+		/* Increment to next runnable routine */
+		do{
+			++(pContext->routine_current);
+		} while(
+			pContext->routine_current < pContext->routines_top
+			&& pContext->routines[pContext->routine_current].panic != 0
+			&& pContext->routines[pContext->routine_current].parse_error.is_set
+			&& pContext->routines[pContext->routine_current].runtime_error.is_set
+			);
+		/* Exit condition */
+		if(pContext->routine_current >= pContext->routines_top) break;
+		
+		/* Kill? */
+		if(pContext->routines[pContext->routine_current].panic == INK_ROUTINE_SUCCESS) {
+			ink_kill_routine(pContext, pContext->routine_current);
+		}
+		
+		/* Step! */
+		for(idx = 0; idx < INK_STEP_BATCH_COUNT; ++idx) {
+			out = ink_step(pContext);
+			if (unlikely(out == 0)) {
+				pContext->routines[pContext->routine_current].panic = INK_ROUTINE_SUCCESS;
+				goto top_label;
+			} else if (unlikely(out < 0)) {
+				pContext->routines[pContext->routine_current].panic = out;
+				goto top_label;
+			}
+		}
+	}
+	return 0;
 }
 
 int ink_new_type(

main.c

@@ -4,8 +4,10 @@
 
 int main(int argc, char** argv) {
 	char read_buffer[2048];
+	struct timespec start_time, end_time;
 	clock_t begin, end;
 	double time_spent;
+	double s_total = 0;
 	struct context* ctx;
 	char** end_argv;
 	ctx = ink_make_default_context();
@@ -29,12 +31,16 @@ int main(int argc, char** argv) {
 	}
 	
 	begin = clock();
+	clock_gettime(CLOCK_MONOTONIC, &start_time);
 	while(ink_can_run(ctx)) {
 		ink_step_everyone(ctx);
 	}
+	clock_gettime(CLOCK_MONOTONIC, &end_time);
     ink_gc(ctx);
 	end = clock();
-	time_spent = ctx->steps/((double)(end - begin) / CLOCKS_PER_SEC);
-	printf("\nExecuted in %u steps\nCollected %u times\nExecution freq: %uHz\n", ctx->steps, ctx->collections, (unsigned int)time_spent);
+	time_spent = ctx->steps/(double)((end - begin) / CLOCKS_PER_SEC);
+	s_total += (end_time.tv_sec - start_time.tv_sec) + (end_time.tv_nsec - start_time.tv_nsec) / 1.0e9;
+	s_total = ctx->steps / s_total;
+	printf("\nExecuted in %u steps\nCollected %u times\nExecution freq: %lfHz\n", ctx->steps, ctx->collections, s_total);
 	return ctx->panic;
 }

test/test03.nk

@@ -0,0 +1,19 @@
+
+fn work do
+    12 13 + 25 + 13 + 12 + 25 + 100 ==
+    12 13 + 25 + 13 + 12 + 25 + 100 == -
+    12 13 + 25 + 13 + 12 + 25 + 100 ==
+    12 13 + 25 + 13 + 12 + 25 + 100 == - +
+    12 13 + 25 + 13 + 12 + 25 + 100 ==
+    12 13 + 25 + 13 + 12 + 25 + 100 == -
+    12 13 + 25 + 13 + 12 + 25 + 100 ==
+    12 13 + 25 + 13 + 12 + 25 + 100 == - + +
+    12 13 + 25 + 13 + 12 + 25 + 100 ==
+    12 13 + 25 + 13 + 12 + 25 + 100 == -
+    12 13 + 25 + 13 + 12 + 25 + 100 ==
+    12 13 + 25 + 13 + 12 + 25 + 100 == - +
+    12 13 + 25 + 13 + 12 + 25 + 100 ==
+    12 13 + 25 + 13 + 12 + 25 + 100 == -
+    12 13 + 25 + 13 + 12 + 25 + 100 ==
+    12 13 + 25 + 13 + 12 + 25 + 100 == - + + + 0 ==
+end

test/test06.nk

@@ -3,3 +3,22 @@
 [ 72 101 108 108 111 32 87 111 114 108 100 10 ]
 stack.dump
 # This prints the array above
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+work work work work work
+