freedreno/ir3: fix fallout of unused false-depth elimination

Since we were MARK flag for both preventing loops, and tracking whether
instructions were used, we could end up in an infinite loop due to
bd2ca2bcdd.  Instead invert the logic.. mark all instructions UNUSED
up front and clear the flag as we visit them.

Fixes: bd2ca2bcdd freedreno/ir3: eliminate unused false-deps
Signed-off-by: Rob Clark <robdclark@gmail.com>

diff --git a/src/gallium/drivers/freedreno/ir3/ir3.h b/src/gallium/drivers/freedreno/ir3/ir3.h
index 500faa85377e3a0a9386036ab6b96b4aae614a5b..1152ea300b1ebb044074d8b7de65998bad9fa3e6 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3.h
@@ -762,7 +762,6 @@ static inline bool writes_pred(struct ir3_instruction *instr)
 static inline struct ir3_instruction *ssa(struct ir3_register *reg)
 {
        if (reg->flags & (IR3_REG_SSA | IR3_REG_ARRAY)) {
-               debug_assert(!(reg->instr && (reg->instr->flags & IR3_INSTR_UNUSED)));
                return reg->instr;
        }
        return NULL;

diff --git a/src/gallium/drivers/freedreno/ir3/ir3_depth.c b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
index 3ece13928f424130cbcd8de2ad033e883118980b..18795f77eb9735057d948ba1540edf25ef56f18f 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3_depth.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
@@ -137,14 +137,12 @@ ir3_instr_depth(struct ir3_instruction *instr, unsigned boost, bool falsedep)
 {
        struct ir3_instruction *src;
 
-       /* if we've already visited this instruction, bail now: */
-       if (falsedep) {
-               /* don't mark falsedep's as used, but process them normally: */
-               if (instr->flags & IR3_INSTR_MARK)
-                       return;
-       } else if (ir3_instr_check_mark(instr)) {
+       /* don't mark falsedep's as used, but otherwise process them normally: */
+       if (!falsedep)
+               instr->flags &= ~IR3_INSTR_UNUSED;
+
+       if (ir3_instr_check_mark(instr))
                return;
-       }
 
        instr->depth = 0;
 
@@ -175,14 +173,9 @@ remove_unused_by_block(struct ir3_block *block)
 {
        bool progress = false;
        list_for_each_entry_safe (struct ir3_instruction, instr, &block->instr_list, node) {
-               if (!ir3_instr_check_mark(instr)) {
-                       if (instr->opc == OPC_END)
-                               continue;
-                       /* mark it, in case it is input, so we can
-                        * remove unused inputs:
-                        */
-                       instr->flags |= IR3_INSTR_UNUSED;
-                       /* and remove from instruction list: */
+               if (instr->opc == OPC_END)
+                       continue;
+               if (instr->flags & IR3_INSTR_UNUSED) {
                        list_delinit(&instr->node);
                        progress = true;
                }
@@ -197,6 +190,16 @@ compute_depth_and_remove_unused(struct ir3 *ir)
        bool progress = false;
 
        ir3_clear_mark(ir);
+
+       /* initially mark everything as unused, we'll clear the flag as we
+        * visit the instructions:
+        */
+       list_for_each_entry (struct ir3_block, block, &ir->block_list, node) {
+               list_for_each_entry (struct ir3_instruction, instr, &block->instr_list, node) {
+                       instr->flags |= IR3_INSTR_UNUSED;
+               }
+       }
+
        for (i = 0; i < ir->noutputs; i++)
                if (ir->outputs[i])
                        ir3_instr_depth(ir->outputs[i], 0, false);
@@ -220,7 +223,7 @@ compute_depth_and_remove_unused(struct ir3 *ir)
         */
        for (i = 0; i < ir->indirects_count; i++) {
                struct ir3_instruction *instr = ir->indirects[i];
-               if (instr->flags & IR3_INSTR_UNUSED)
+               if (instr && (instr->flags & IR3_INSTR_UNUSED))
                        ir->indirects[i] = NULL;
        }