nir: add atomic lowering support for AoA
authorTimothy Arceri <t_arceri@yahoo.com.au>
Wed, 22 Jul 2015 22:32:00 +0000 (08:32 +1000)
committerTimothy Arceri <t_arceri@yahoo.com.au>
Fri, 16 Oct 2015 21:43:21 +0000 (08:43 +1100)
Cc: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
src/glsl/nir/nir_lower_atomics.c

index 6f9ecc019ec69b560763e8872af3cc72896e092c..46e137652a19acd358c07b4429bd895e4d836841 100644 (file)
@@ -72,20 +72,22 @@ lower_instr(nir_intrinsic_instr *instr, nir_function_impl *impl)
 
    nir_ssa_def *offset_def = &offset_const->def;
 
-   if (instr->variables[0]->deref.child != NULL) {
-      assert(instr->variables[0]->deref.child->deref_type ==
-             nir_deref_type_array);
-      nir_deref_array *deref_array =
-         nir_deref_as_array(instr->variables[0]->deref.child);
-      assert(deref_array->deref.child == NULL);
+   nir_deref *tail = &instr->variables[0]->deref;
+   while (tail->child != NULL) {
+      assert(tail->child->deref_type == nir_deref_type_array);
+      nir_deref_array *deref_array = nir_deref_as_array(tail->child);
+      tail = tail->child;
 
-      offset_const->value.u[0] +=
-         deref_array->base_offset * ATOMIC_COUNTER_SIZE;
+      unsigned child_array_elements = tail->child != NULL ?
+         glsl_get_aoa_size(tail->type) : 1;
+
+      offset_const->value.u[0] += deref_array->base_offset *
+         child_array_elements * ATOMIC_COUNTER_SIZE;
 
       if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
          nir_load_const_instr *atomic_counter_size =
                nir_load_const_instr_create(mem_ctx, 1);
-         atomic_counter_size->value.u[0] = ATOMIC_COUNTER_SIZE;
+         atomic_counter_size->value.u[0] = child_array_elements * ATOMIC_COUNTER_SIZE;
          nir_instr_insert_before(&instr->instr, &atomic_counter_size->instr);
 
          nir_alu_instr *mul = nir_alu_instr_create(mem_ctx, nir_op_imul);
@@ -102,7 +104,7 @@ lower_instr(nir_intrinsic_instr *instr, nir_function_impl *impl)
          add->src[0].src.is_ssa = true;
          add->src[0].src.ssa = &mul->dest.dest.ssa;
          add->src[1].src.is_ssa = true;
-         add->src[1].src.ssa = &offset_const->def;
+         add->src[1].src.ssa = offset_def;
          nir_instr_insert_before(&instr->instr, &add->instr);
 
          offset_def = &add->dest.dest.ssa;