diff --git a/m1/perf/Makefile b/m1/perf/Makefile
index 5a36cdc..8c85070 100644
--- a/m1/perf/Makefile
+++ b/m1/perf/Makefile
@@ -36,6 +36,7 @@ $(LIBFPVM_A):
 clean:
 		$(MAKE) -C $(M1SWLIB_DIR) clean
 		$(MAKE) -C $(COMPILER_DIR) clean
+		rm -f $(M1SWLIB_DIR)/sched.o
 		rm -f $(OBJS)
 
 path:
diff --git a/m1/perf/TODO b/m1/perf/TODO
index 5935d5e..aeff81a 100644
--- a/m1/perf/TODO
+++ b/m1/perf/TODO
@@ -1,8 +1,9 @@
 Done:
 - dynamically allocate scheduler context
+- see if preferring critical path can improve code efficiency (YES !)
 
 Pending:
-- see if preferring critical path can improve code efficiency
+- see if dynamically adjusting the critical path leads to further improvements
 - test IF
 - run result comparison against full set of patches
 - check if result comparison actually compares meaningful data
diff --git a/m1/perf/sched.c b/m1/perf/sched.c
index 49c6ed6..ec1ebda 100644
--- a/m1/perf/sched.c
+++ b/m1/perf/sched.c
@@ -31,7 +31,8 @@
 #include <hw/pfpu.h>
 
 
-#define	REG_STATS
+//#define	REG_STATS
+//#define	LCPF	/* longest critical path first */
 
 #ifdef DEBUG
 #define	Dprintf printf
@@ -64,6 +65,9 @@ struct insn {
 	int earliest;	/* earliest cycle dependencies seen so far are met */
 	struct list dependants;	/* list of dependencies (constant) */
 	int num_dependants;	/* number of unresolved dependencies */
+#ifdef LCPF
+	int distance;		/* minimum cycles on this path until the end */
+#endif
 };
 
 
@@ -376,6 +380,18 @@ static void init_scheduler(struct fpvm_fragment *frag)
 		else
 			list_add_tail(&sc->ready[0], &insn->more);
 	}
+
+#ifdef LCPF
+	struct data_ref *dep;
+
+	for (i = frag->ninstructions-1; i >= 0; i--) {
+		insn = sc->insns+i;
+		foreach (dep, &insn->dependants)
+			if (dep->insn->distance > insn->distance)
+				insn->distance = dep->insn->distance;
+		insn->distance += insn->latency;
+	}
+#endif
 }
 
 
@@ -441,6 +457,9 @@ static int schedule(unsigned int *code)
 	int remaining;
 	int i, last, end;
 	struct insn *insn;
+#ifdef LCPF
+	struct insn *best;
+#endif
 
 	remaining = sc->frag->ninstructions;
 	for (i = 0; remaining; i++) {
@@ -451,17 +470,32 @@ static int schedule(unsigned int *code)
 		    i, remaining, count(&sc->waiting), count(&sc->ready[i]));
 
 		list_concat(&sc->waiting, &sc->ready[i]);
+#ifdef LCPF
+		best = NULL;
+#endif
 		foreach (insn, &sc->waiting) {
 			end = i+insn->latency;
 			if (end >= PFPU_PROGSIZE)
 				return -1;
 			if (!FIELD(code[end]).dest) {
+#ifdef LCPF
+				if (!best || best->distance < insn->distance)
+					best = insn;
+#else
 				issue(insn, i, code);
 				list_del(&insn->more);
 				remaining--;
 				break;
+#endif
 			}
 		}
+#ifdef LCPF
+		if (best) {
+			issue(best, i, code);
+			list_del(&best->more);
+			remaining--;
+		}
+#endif
 		if (FIELD(code[i]).dest)
 			put_reg(sc->pfpu_regs[FIELD(code[i]).dest].vm_reg);
 	}