freedreno/ir3: don't DCE ij_pix if used for pre-fs-texture-fetch

When we enable pre-dispatch texture fetch, we could have a scenario
where the barycentric i/j coord sysval is not used in the shader, but
only used for the varying fetch for the pre-dispatch texture fetch.
In this case we need to take care not to DCE this sysval.

Signed-off-by: Rob Clark <[email protected]>
Reviewed-by: Kristian H. Kristensen <[email protected]>

3 files changed, 14 insertions, 6 deletions

diff --git a/src/freedreno/ir3/ir3.h b/src/freedreno/ir3/ir3.h
index f7a87ee076a..8b8788a8a97 100644
--- a/src/freedreno/ir3/ir3.h
+++ b/src/freedreno/ir3/ir3.h
@@ -1062,13 +1062,13 @@ void ir3_print(struct ir3 *ir);
 void ir3_print_instr(struct ir3_instruction *instr);
 
 /* depth calculation: */
+struct ir3_shader_variant;
 int ir3_delayslots(struct ir3_instruction *assigner,
 		struct ir3_instruction *consumer, unsigned n);
 void ir3_insert_by_depth(struct ir3_instruction *instr, struct list_head *list);
-void ir3_depth(struct ir3 *ir);
+void ir3_depth(struct ir3 *ir, struct ir3_shader_variant *so);
 
 /* copy-propagate: */
-struct ir3_shader_variant;
 void ir3_cp(struct ir3 *ir, struct ir3_shader_variant *so);
 
 /* group neighbors and insert mov's to resolve conflicts: */
diff --git a/src/freedreno/ir3/ir3_compiler_nir.c b/src/freedreno/ir3/ir3_compiler_nir.c
index 081d4cf19ec..39bef63a780 100644
--- a/src/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/freedreno/ir3/ir3_compiler_nir.c
@@ -3142,7 +3142,7 @@ ir3_compile_shader_nir(struct ir3_compiler *compiler,
 		ir3_print(ir);
 	}
 
-	ir3_depth(ir);
+	ir3_depth(ir, so);
 
 	if (ir3_shader_debug & IR3_DBG_OPTMSGS) {
 		printf("AFTER DEPTH:\n");
diff --git a/src/freedreno/ir3/ir3_depth.c b/src/freedreno/ir3/ir3_depth.c
index bc42bfdeb3b..49feda7b195 100644
--- a/src/freedreno/ir3/ir3_depth.c
+++ b/src/freedreno/ir3/ir3_depth.c
@@ -27,6 +27,7 @@
 #include "util/u_math.h"
 
 #include "ir3.h"
+#include "ir3_shader.h"
 
 /*
  * Instruction Depth:
@@ -209,7 +210,7 @@ remove_unused_by_block(struct ir3_block *block)
 }
 
 static bool
-compute_depth_and_remove_unused(struct ir3 *ir)
+compute_depth_and_remove_unused(struct ir3 *ir, struct ir3_shader_variant *so)
 {
 	unsigned i;
 	bool progress = false;
@@ -221,6 +222,13 @@ compute_depth_and_remove_unused(struct ir3 *ir)
 	 */
 	list_for_each_entry (struct ir3_block, block, &ir->block_list, node) {
 		list_for_each_entry (struct ir3_instruction, instr, &block->instr_list, node) {
+			/* special case, if pre-fs texture fetch used, we cannot
+			 * eliminate the barycentric i/j input
+			 */
+			if (so->num_sampler_prefetch &&
+					(instr->opc == OPC_META_INPUT) &&
+					(instr->input.sysval == SYSTEM_VALUE_BARYCENTRIC_PIXEL))
+				continue;
 			instr->flags |= IR3_INSTR_UNUSED;
 		}
 	}
@@ -263,10 +271,10 @@ compute_depth_and_remove_unused(struct ir3 *ir)
 }
 
 void
-ir3_depth(struct ir3 *ir)
+ir3_depth(struct ir3 *ir, struct ir3_shader_variant *so)
 {
 	bool progress;
 	do {
-		progress = compute_depth_and_remove_unused(ir);
+		progress = compute_depth_and_remove_unused(ir, so);
 	} while (progress);
 }