aco: run opt_algebraic in a loop

Totals from affected shaders:
SGPRS: 13920 -> 13656 (-1.90 %)
VGPRS: 12972 -> 12960 (-0.09 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 1005680 -> 1000648 (-0.50 %) bytes
LDS: 91 -> 91 (0.00 %) blocks
Max Waves: 688 -> 688 (0.00 %)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>

diff --git a/src/amd/compiler/aco_instruction_selection_setup.cpp b/src/amd/compiler/aco_instruction_selection_setup.cpp
index 9727020a8a929e0c8e0b411c071de1f0579645d5..b100ace3c8c0704c89d689fa63a5d5a27fb11f29 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection_setup.cpp
+++ b/src/amd/compiler/aco_instruction_selection_setup.cpp
@@ -1331,9 +1331,14 @@ setup_isel_context(Program* program,
       nir_lower_idiv(nir, nir_lower_idiv_precise);
 
       /* optimize the lowered ALU operations */
-      nir_copy_prop(nir);
-      nir_opt_constant_folding(nir);
-      nir_opt_algebraic(nir);
+      bool more_algebraic = true;
+      while (more_algebraic) {
+         more_algebraic = false;
+         NIR_PASS_V(nir, nir_copy_prop);
+         NIR_PASS_V(nir, nir_opt_dce);
+         NIR_PASS_V(nir, nir_opt_constant_folding);
+         NIR_PASS(more_algebraic, nir, nir_opt_algebraic);
+      }
 
       /* Do late algebraic optimization to turn add(a, neg(b)) back into
       * subs, then the mandatory cleanup after algebraic.  Note that it may