From c009a13f5cdd1cc70ef4affa49a7a4e2d1604987 Mon Sep 17 00:00:00 2001
From: Werner Almesberger <werner@almesberger.net>
Date: Sun, 18 Sep 2011 01:28:02 -0300
Subject: [PATCH] m1/perf/: cleaned up sched.c a little; added TODO

---
 m1/perf/TODO    |  8 ++++++++
 m1/perf/sched.c | 37 +++++++++++++++++++++----------------
 2 files changed, 29 insertions(+), 16 deletions(-)
 create mode 100644 m1/perf/TODO

diff --git a/m1/perf/TODO b/m1/perf/TODO
new file mode 100644
index 0000000..f4f212e
--- /dev/null
+++ b/m1/perf/TODO
@@ -0,0 +1,8 @@
+- see if preferring critical path can improve code efficiency
+- test IF
+- run result comparison against full set of patches
+- check if result comparison actually compares meaningful data
+- compare run time and code size for all patches
+- see what optimization changes (may interfere with profiling)
+- build into Flickernoise (some things may need adapting, e.g., abort())
+- review code, see if things can be simplified
diff --git a/m1/perf/sched.c b/m1/perf/sched.c
index 82eaeca..0732b8f 100644
--- a/m1/perf/sched.c
+++ b/m1/perf/sched.c
@@ -31,7 +31,11 @@
 #include <hw/pfpu.h>
 
 
+#ifdef DEBUG
+#define	Dprintf printf
+#else
 #define	Dprintf(...)
+#endif
 
 
 #define	MAX_LATENCY	8	/* maximum latency; okay to make this bigger */
@@ -199,7 +203,9 @@ static int alloc_reg(struct insn *setter)
 		abort();
 	reg->vm_reg = vm_reg;
 	pfpu_reg = reg-pfpu_regs;
-Dprintf("  alloc reg %d -> %d\n", vm_reg, pfpu_reg);
+
+	Dprintf("  alloc reg %d -> %d\n", vm_reg, pfpu_reg);
+
 	vm_idx = reg2idx(vm_reg);
 	regs[vm_idx].setter = setter;
 	regs[vm_idx].pfpu_reg = pfpu_reg;
@@ -299,8 +305,9 @@ static struct vm_reg *add_data_ref(struct insn *insn, struct data_ref *ref,
 		list_add_tail(&ref->dep->dependants, &ref->more);
 		ref->dep->num_dependants++;
 		insn->unresolved++;
-Dprintf("insn %lu: reg %d setter %lu unresolved %d\n",
-   insn-insns, reg_num, reg->setter-insns, insn->unresolved);
+
+		Dprintf("insn %lu: reg %d setter %lu unresolved %d\n",
+		    insn-insns, reg_num, reg->setter-insns, insn->unresolved);
 	} else {
 		list_init(&ref->more);
 	}
@@ -308,7 +315,6 @@ Dprintf("insn %lu: reg %d setter %lu unresolved %d\n",
 }
 
 
-int catch = 0;
 static void init_scheduler(struct fpvm_fragment *frag)
 {
 	int i;
@@ -319,12 +325,6 @@ static void init_scheduler(struct fpvm_fragment *frag)
 	for (i = 0; i != PFPU_PROGSIZE; i++)
 		list_init(&ready[i]);
 
-#if 0
-if (frag->ninstructions > 10) {
-  frag->ninstructions = 10;
-catch = 1;
-}
-#endif
 	for (i = 0; i != frag->ninstructions; i++) {
 		insn = insns+i;
 		memset(insn, 0, sizeof(struct insn));
@@ -363,8 +363,9 @@ static void issue(struct insn *insn, int cycle, unsigned *code)
 	int end;
 	end = cycle+insn->latency;
 
-Dprintf("cycle %d: insn %lu L %d (A %d B %d)\n",
-  cycle, insn-insns, insn->latency, insn->vm_insn->opa, insn->vm_insn->opb);
+	Dprintf("cycle %d: insn %lu L %d (A %d B %d)\n", cycle,
+	    insn-insns, insn->latency, insn->vm_insn->opa, insn->vm_insn->opb);
+
 	switch (insn->arity) {
 	case 3:
 		/* fall through */
@@ -389,7 +390,8 @@ Dprintf("cycle %d: insn %lu L %d (A %d B %d)\n",
 		if (ref->insn->earliest <= end)
 			ref->insn->earliest = end+1;
 		if (!--ref->insn->unresolved) {
-Dprintf("  unlocked %lu -> %u\n", ref->insn-insns, ref->insn->earliest);
+			Dprintf("  unlocked %lu -> %u\n", ref->insn-insns,
+			    ref->insn->earliest);
 			list_del(&ref->insn->more);
 			list_add_tail(ready+ref->insn->earliest,
 			    &ref->insn->more);
@@ -398,6 +400,7 @@ Dprintf("  unlocked %lu -> %u\n", ref->insn-insns, ref->insn->earliest);
 }
 
 
+#ifdef DEBUG
 static int count(const struct list *list)
 {
 	int n = 0;
@@ -407,6 +410,7 @@ static int count(const struct list *list)
 		n++;
 	return n;
 }
+#endif
 
 
 static int schedule(struct fpvm_fragment *frag, unsigned int *code)
@@ -419,10 +423,11 @@ static int schedule(struct fpvm_fragment *frag, unsigned int *code)
 	for (i = 0; remaining; i++) {
 		if (i == PFPU_PROGSIZE)
 			return -1;
-Dprintf("@%d --- remaining %d, waiting %d + ready %d = ", i, remaining,
-  count(&waiting), count(&ready[i]));
+
+		Dprintf("@%d --- remaining %d, waiting %d + ready %d\n",
+		    i, remaining, count(&waiting), count(&ready[i]));
+
 		list_concat(&waiting, &ready[i]);
-Dprintf("%d\n", count(&waiting));
 		foreach (insn, &waiting) {
 			end = i+insn->latency;
 			if (end >= PFPU_PROGSIZE)