freedreno/ir3: better cleanup when removing unused instructions

Do a better job of pruning when removing unused instructions, including
cleaning up dangling false-deps.

This introduces a new ssa src pointer iterator, which makes it easy to
clear links without having to think about whether it is a normal ssa
src or a false-dep.

Signed-off-by: Rob Clark <[email protected]>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4440>

4 files changed, 36 insertions, 10 deletions

diff --git a/src/freedreno/ir3/ir3.h b/src/freedreno/ir3/ir3.h
index 65c16791e62..7fb0d823d25 100644
--- a/src/freedreno/ir3/ir3.h
+++ b/src/freedreno/ir3/ir3.h
@@ -1118,13 +1118,16 @@ static inline unsigned __ssa_src_cnt(struct ir3_instruction *instr)
 	return cnt;
 }
 
-static inline struct ir3_instruction * __ssa_src_n(struct ir3_instruction *instr, unsigned n)
+static inline struct ir3_instruction **
+__ssa_srcp_n(struct ir3_instruction *instr, unsigned n)
 {
 	if (n == (instr->regs_count + instr->deps_count))
-		return instr->address;
+		return &instr->address;
 	if (n >= instr->regs_count)
-		return instr->deps[n - instr->regs_count];
-	return ssa(instr->regs[n]);
+		return &instr->deps[n - instr->regs_count];
+	if (ssa(instr->regs[n]))
+		return &instr->regs[n]->instr;
+	return NULL;
 }
 
 static inline bool __is_false_dep(struct ir3_instruction *instr, unsigned n)
@@ -1136,12 +1139,18 @@ static inline bool __is_false_dep(struct ir3_instruction *instr, unsigned n)
 	return false;
 }
 
-#define __src_cnt(__instr) ((__instr)->address ? (__instr)->regs_count : (__instr)->regs_count - 1)
+#define foreach_ssa_srcp_n(__srcp, __n, __instr) \
+	for (struct ir3_instruction **__srcp = (void *)~0; __srcp; __srcp = NULL) \
+		for (unsigned __cnt = __ssa_src_cnt(__instr), __n = 0; __n < __cnt; __n++) \
+			if ((__srcp = __ssa_srcp_n(__instr, __n)))
+
+#define foreach_ssa_srcp(__srcp, __instr) \
+	foreach_ssa_srcp_n(__srcp, __i, __instr)
 
 /* iterator for an instruction's SSA sources (instr), also returns src #: */
 #define foreach_ssa_src_n(__srcinst, __n, __instr) \
-	for (unsigned __cnt = __ssa_src_cnt(__instr), __n = 0; __n < __cnt; __n++) \
-		if ((__srcinst = __ssa_src_n(__instr, __n)))
+	foreach_ssa_srcp_n(__srcp, __n, __instr) \
+		if ((__srcinst = *__srcp))
 
 /* iterator for an instruction's SSA sources (instr): */
 #define foreach_ssa_src(__srcinst, __instr) \
diff --git a/src/freedreno/ir3/ir3_depth.c b/src/freedreno/ir3/ir3_depth.c
index 00d0c54fac5..90a0ddef51f 100644
--- a/src/freedreno/ir3/ir3_depth.c
+++ b/src/freedreno/ir3/ir3_depth.c
@@ -95,7 +95,7 @@ remove_unused_by_block(struct ir3_block *block)
 				/* tex (cat5) instructions have a writemask, so we can
 				 * mask off unused components.  Other instructions do not.
 				 */
-				if (is_tex_or_prefetch(src) && (src->regs[0]->wrmask > 1)) {
+				if (src && is_tex_or_prefetch(src) && (src->regs[0]->wrmask > 1)) {
 					src->regs[0]->wrmask &= ~(1 << instr->split.off);
 
 					/* prune no-longer needed right-neighbors.  We could
@@ -114,6 +114,13 @@ remove_unused_by_block(struct ir3_block *block)
 					}
 				}
 			}
+
+			/* prune false-deps, etc: */
+			foreach_ssa_use (use, instr)
+				foreach_ssa_srcp_n (srcp, n, use)
+					if (*srcp == instr)
+						*srcp = NULL;
+
 			list_delinit(&instr->node);
 			progress = true;
 		}
@@ -206,8 +213,14 @@ compute_depth_and_remove_unused(struct ir3 *ir, struct ir3_shader_variant *so)
 void
 ir3_depth(struct ir3 *ir, struct ir3_shader_variant *so)
 {
+	void *mem_ctx = ralloc_context(NULL);
 	bool progress;
+
+	ir3_find_ssa_uses(ir, mem_ctx, true);
+
 	do {
 		progress = compute_depth_and_remove_unused(ir, so);
 	} while (progress);
+
+	ralloc_free(mem_ctx);
 }
diff --git a/src/freedreno/ir3/ir3_postsched.c b/src/freedreno/ir3/ir3_postsched.c
index 09e58750c04..521078a04c9 100644
--- a/src/freedreno/ir3/ir3_postsched.c
+++ b/src/freedreno/ir3/ir3_postsched.c
@@ -685,7 +685,7 @@ cleanup_self_movs(struct ir3 *ir)
 			}
 
 			for (unsigned i = 0; i < instr->deps_count; i++) {
-				if (is_self_mov(instr->deps[i])) {
+				if (instr->deps[i] && is_self_mov(instr->deps[i])) {
 					list_delinit(&instr->deps[i]->node);
 					instr->deps[i] = instr->deps[i]->regs[1]->instr;
 				}
diff --git a/src/freedreno/ir3/ir3_print.c b/src/freedreno/ir3/ir3_print.c
index 44c8278276a..b18a3d27e34 100644
--- a/src/freedreno/ir3/ir3_print.c
+++ b/src/freedreno/ir3/ir3_print.c
@@ -73,7 +73,11 @@ static void print_instr_name(struct ir3_instruction *instr, bool flags)
 	printf("%04u:", instr->name);
 	printf("%04u:", instr->ip);
 	printf("%03d:", instr->depth);
-	printf("%03u: ", instr->use_count);
+	if (instr->flags & IR3_INSTR_UNUSED) {
+		printf("XXX: ");
+	} else {
+		printf("%03u: ", instr->use_count);
+	}
 
 	if (flags) {
 		printf("\t");