3 files changed, 34 insertions, 2 deletions

diff --git a/src/freedreno/ir3/ir3.c b/src/freedreno/ir3/ir3.c
index 9678389e8b5..d4fcd995ed0 100644
--- a/src/freedreno/ir3/ir3.c
+++ b/src/freedreno/ir3/ir3.c
@@ -1155,11 +1155,33 @@ ir3_count_instructions(struct ir3 *ir)
 	unsigned cnt = 1;
 	foreach_block (block, &ir->block_list) {
 		block->start_ip = cnt;
+		foreach_instr (instr, &block->instr_list) {
+			instr->ip = cnt++;
+		}
 		block->end_ip = cnt;
+	}
+	return cnt;
+}
+
+/* When counting instructions for RA, we insert extra fake instructions at the
+ * beginning of each block, where values become live, and at the end where
+ * values die. This prevents problems where values live-in at the beginning or
+ * live-out at the end of a block from being treated as if they were
+ * live-in/live-out at the first/last instruction, which would be incorrect.
+ * In ir3_legalize these ip's are assumed to be actual ip's of the final
+ * program, so it would be incorrect to use this everywhere.
+ */
+
+unsigned
+ir3_count_instructions_ra(struct ir3 *ir)
+{
+	unsigned cnt = 1;
+	foreach_block (block, &ir->block_list) {
+		block->start_ip = cnt++;
 		foreach_instr (instr, &block->instr_list) {
 			instr->ip = cnt++;
-			block->end_ip = instr->ip;
 		}
+		block->end_ip = cnt++;
 	}
 	return cnt;
 }
diff --git a/src/freedreno/ir3/ir3.h b/src/freedreno/ir3/ir3.h
index 351490aecf7..6cb882002f0 100644
--- a/src/freedreno/ir3/ir3.h
+++ b/src/freedreno/ir3/ir3.h
@@ -594,6 +594,7 @@ void ir3_block_clear_mark(struct ir3_block *block);
 void ir3_clear_mark(struct ir3 *shader);
 
 unsigned ir3_count_instructions(struct ir3 *ir);
+unsigned ir3_count_instructions_ra(struct ir3 *ir);
 
 void ir3_find_ssa_uses(struct ir3 *ir, void *mem_ctx, bool falsedeps);
 
diff --git a/src/freedreno/ir3/ir3_ra.c b/src/freedreno/ir3/ir3_ra.c
index 72682dc6988..f74174498d0 100644
--- a/src/freedreno/ir3/ir3_ra.c
+++ b/src/freedreno/ir3/ir3_ra.c
@@ -541,7 +541,7 @@ ra_init(struct ir3_ra_ctx *ctx)
 	unsigned n, base;
 
 	ir3_clear_mark(ctx->ir);
-	n = ir3_count_instructions(ctx->ir);
+	n = ir3_count_instructions_ra(ctx->ir);
 
 	ctx->instrd = rzalloc_array(NULL, struct ir3_ra_instr_data, n);
 
@@ -950,6 +950,15 @@ ra_calc_block_live_values(struct ir3_ra_ctx *ctx, struct ir3_block *block)
 
 	/* the remaining live should match liveout (for extra sanity testing): */
 	if (RA_DEBUG) {
+		unsigned new_dead = 0;
+		BITSET_FOREACH_SET (name, live, ctx->alloc_count) {
+			/* Is this the last use? */
+			if (ctx->use[name] != block->end_ip)
+				continue;
+			new_dead += name_size(ctx, name);
+			d("NEW_DEAD: %u (new_dead=%u)", name, new_dead);
+			BITSET_CLEAR(live, name);
+		}
 		unsigned liveout = 0;
 		BITSET_FOREACH_SET (name, bd->liveout, ctx->alloc_count) {
 			liveout += name_size(ctx, name);