nir: Handle vec8/16 in bool_to_bitsize

[mesa.git] / src / compiler / nir / nir_lower_phis_to_scalar.c

diff --git a/src/compiler/nir/nir_lower_phis_to_scalar.c b/src/compiler/nir/nir_lower_phis_to_scalar.c
index 41ae19e8391759316832d029cf7ba2d7b7bfa445..45894dee352f4c28e0176cd4dc2735d8707a83c8 100644 (file)
--- a/src/compiler/nir/nir_lower_phis_to_scalar.c
+++ b/src/compiler/nir/nir_lower_phis_to_scalar.c
@@ -75,10 +75,15 @@ is_phi_src_scalarizable(nir_phi_src *src,
       return should_lower_phi(nir_instr_as_phi(src_instr), state);
 
    case nir_instr_type_load_const:
-   case nir_instr_type_ssa_undef:
       /* These are trivially scalarizable */
       return true;
 
+   case nir_instr_type_ssa_undef:
+      /* The caller of this function is going to OR the results and we don't
+       * want undefs to count so we return false.
+       */
+      return false;
+
    case nir_instr_type_intrinsic: {
       nir_intrinsic_instr *src_intrin = nir_instr_as_intrinsic(src_instr);
 
@@ -88,15 +93,18 @@ is_phi_src_scalarizable(nir_phi_src *src,
          return deref->mode == nir_var_shader_in ||
                 deref->mode == nir_var_uniform ||
                 deref->mode == nir_var_mem_ubo ||
-                deref->mode == nir_var_mem_ssbo;
+                deref->mode == nir_var_mem_ssbo ||
+                deref->mode == nir_var_mem_global;
       }
 
       case nir_intrinsic_interp_deref_at_centroid:
       case nir_intrinsic_interp_deref_at_sample:
       case nir_intrinsic_interp_deref_at_offset:
+      case nir_intrinsic_interp_deref_at_vertex:
       case nir_intrinsic_load_uniform:
       case nir_intrinsic_load_ubo:
       case nir_intrinsic_load_ssbo:
+      case nir_intrinsic_load_global:
       case nir_intrinsic_load_input:
          return true;
       default:
@@ -148,11 +156,16 @@ should_lower_phi(nir_phi_instr *phi, struct lower_phis_to_scalar_state *state)
     */
    entry = _mesa_hash_table_insert(state->phi_table, phi, (void *)(intptr_t)1);
 
-   bool scalarizable = true;
+   bool scalarizable = false;
 
    nir_foreach_phi_src(src, phi) {
+      /* This loop ignores srcs that are not scalarizable because its likely
+       * still worth copying to temps if another phi source is scalarizable.
+       * This reduces register spilling by a huge amount in the i965 driver for
+       * Deus Ex: MD.
+       */
       scalarizable = is_phi_src_scalarizable(src, state);
-      if (!scalarizable)
+      if (scalarizable)
          break;
    }
 
@@ -223,7 +236,7 @@ lower_phis_to_scalar_block(nir_block *block,
          nir_foreach_phi_src(src, phi) {
             /* We need to insert a mov to grab the i'th component of src */
             nir_alu_instr *mov = nir_alu_instr_create(state->mem_ctx,
-                                                      nir_op_imov);
+                                                      nir_op_mov);
             nir_ssa_dest_init(&mov->instr, &mov->dest.dest, 1, bit_size, NULL);
             mov->dest.write_mask = 1;
             nir_src_copy(&mov->src[0].src, &src->src, state->mem_ctx);