nir/algebraic: Make algebraic_parser_test.sh executable.

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

diff --git a/src/compiler/nir/nir_lower_io.c b/src/compiler/nir/nir_lower_io.c
index 3bdacee66bfd312ba4538afa8acfa31a7e3f3842..644243078129eaa3ba13dec183ecefb7eecdfa9c 100644 (file)
--- a/src/compiler/nir/nir_lower_io.c
+++ b/src/compiler/nir/nir_lower_io.c
@@ -112,10 +112,8 @@ get_io_offset(nir_builder *b, nir_deref_instr *deref,
       assert(glsl_type_is_scalar((*p)->type));
 
       /* We always lower indirect dereferences for "compact" array vars. */
-      nir_const_value *const_index = nir_src_as_const_value((*p)->arr.index);
-      assert(const_index);
-
-      const unsigned total_offset = *component + const_index->u32[0];
+      const unsigned index = nir_src_as_uint((*p)->arr.index);
+      const unsigned total_offset = *component + index;
       const unsigned slot_offset = total_offset / 4;
       *component = total_offset % 4;
       return nir_imm_int(b, type_size(glsl_vec4_type()) * slot_offset);
@@ -129,8 +127,7 @@ get_io_offset(nir_builder *b, nir_deref_instr *deref,
          unsigned size = type_size((*p)->type);
 
          nir_ssa_def *mul =
-            nir_imul(b, nir_imm_int(b, size),
-                     nir_ssa_for_src(b, (*p)->arr.index, 1));
+            nir_imul_imm(b, nir_ssa_for_src(b, (*p)->arr.index, 1), size);
 
          offset = nir_iadd(b, offset, mul);
       } else if ((*p)->deref_type == nir_deref_type_struct) {
@@ -141,7 +138,7 @@ get_io_offset(nir_builder *b, nir_deref_instr *deref,
          for (unsigned i = 0; i < (*p)->strct.index; i++) {
             field_offset += type_size(glsl_get_struct_field(parent->type, i));
          }
-         offset = nir_iadd(b, offset, nir_imm_int(b, field_offset));
+         offset = nir_iadd_imm(b, offset, field_offset);
       } else {
          unreachable("Unsupported deref type");
       }
@@ -280,6 +277,10 @@ lower_atomic(nir_intrinsic_instr *intrin, struct lower_io_state *state,
    OP(atomic_and)
    OP(atomic_or)
    OP(atomic_xor)
+   OP(atomic_fadd)
+   OP(atomic_fmin)
+   OP(atomic_fmax)
+   OP(atomic_fcomp_swap)
 #undef OP
    default:
       unreachable("Invalid atomic");
@@ -380,6 +381,10 @@ nir_lower_io_block(nir_block *block,
       case nir_intrinsic_deref_atomic_xor:
       case nir_intrinsic_deref_atomic_exchange:
       case nir_intrinsic_deref_atomic_comp_swap:
+      case nir_intrinsic_deref_atomic_fadd:
+      case nir_intrinsic_deref_atomic_fmin:
+      case nir_intrinsic_deref_atomic_fmax:
+      case nir_intrinsic_deref_atomic_fcomp_swap:
          /* We can lower the io for this nir instrinsic */
          break;
       case nir_intrinsic_interp_deref_at_centroid:
@@ -441,6 +446,10 @@ nir_lower_io_block(nir_block *block,
       case nir_intrinsic_deref_atomic_xor:
       case nir_intrinsic_deref_atomic_exchange:
       case nir_intrinsic_deref_atomic_comp_swap:
+      case nir_intrinsic_deref_atomic_fadd:
+      case nir_intrinsic_deref_atomic_fmin:
+      case nir_intrinsic_deref_atomic_fmax:
+      case nir_intrinsic_deref_atomic_fcomp_swap:
          assert(vertex_index == NULL);
          replacement = lower_atomic(intrin, state, var, offset);
          break;
@@ -510,8 +519,6 @@ nir_lower_io(nir_shader *shader, nir_variable_mode modes,
 {
    bool progress = false;
 
-   nir_assert_unlowered_derefs(shader, nir_lower_load_store_derefs | nir_lower_interp_derefs | nir_lower_atomic_derefs);
-
    nir_foreach_function(function, shader) {
       if (function->impl) {
          progress |= nir_lower_io_impl(function->impl, modes,
@@ -531,6 +538,7 @@ nir_get_io_offset_src(nir_intrinsic_instr *instr)
    switch (instr->intrinsic) {
    case nir_intrinsic_load_input:
    case nir_intrinsic_load_output:
+   case nir_intrinsic_load_shared:
    case nir_intrinsic_load_uniform:
       return &instr->src[0];
    case nir_intrinsic_load_ubo:
@@ -539,6 +547,7 @@ nir_get_io_offset_src(nir_intrinsic_instr *instr)
    case nir_intrinsic_load_per_vertex_output:
    case nir_intrinsic_load_interpolated_input:
    case nir_intrinsic_store_output:
+   case nir_intrinsic_store_shared:
       return &instr->src[1];
    case nir_intrinsic_store_ssbo:
    case nir_intrinsic_store_per_vertex_output: