static void
brw_nir_lower_uniforms(nir_shader *nir, bool is_scalar)
{
if (is_scalar) {
nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
type_size_scalar_bytes);
nir_lower_io(nir, nir_var_uniform, type_size_scalar_bytes, 0);
} else {
nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
type_size_vec4_bytes);
nir_lower_io(nir, nir_var_uniform, type_size_vec4_bytes, 0);
}
}
nir_shader *
brw_create_nir(struct brw_context *brw,
const struct gl_shader_program *shader_prog,
const struct gl_program *prog,
gl_shader_stage stage,
bool is_scalar)
{
struct gl_context *ctx = &brw->ctx;
const nir_shader_compiler_options *options =
ctx->Const.ShaderCompilerOptions[stage].NirOptions;
struct gl_shader *shader = shader_prog ? shader_prog->_LinkedShaders[stage] : NULL;
bool debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
nir_shader *nir;
/* First, lower the GLSL IR or Mesa IR to NIR */
if (shader_prog) {
nir = glsl_to_nir(shader, options);
} else {
nir = prog_to_nir(prog, options);
nir_convert_to_ssa(nir); /* turn registers into SSA */
}
nir_validate_shader(nir);
nir_lower_global_vars_to_local(nir);
nir_validate_shader(nir);
nir_lower_tex_projector(nir);
nir_validate_shader(nir);
nir_normalize_cubemap_coords(nir);
nir_validate_shader(nir);
nir_split_var_copies(nir);
nir_validate_shader(nir);
nir_optimize(nir, is_scalar);
/* Lower a bunch of stuff */
nir_lower_var_copies(nir);
nir_validate_shader(nir);
/* Get rid of split copies */
nir_optimize(nir, is_scalar);
if (is_scalar) {
nir_assign_var_locations(&nir->uniforms,
&nir->num_uniforms,
type_size_scalar);
nir_assign_var_locations(&nir->inputs, &nir->num_inputs, type_size_scalar);
nir_assign_var_locations(&nir->outputs, &nir->num_outputs, type_size_scalar);
nir_lower_io(nir, type_size_scalar);
} else {
nir_assign_var_locations(&nir->uniforms,
&nir->num_uniforms,
type_size_vec4);
nir_assign_var_locations(&nir->inputs, &nir->num_inputs, type_size_vec4);
foreach_list_typed(nir_variable, var, node, &nir->outputs)
var->data.driver_location = var->data.location;
nir_lower_io(nir, type_size_vec4);
}
nir_validate_shader(nir);
nir_remove_dead_variables(nir);
nir_validate_shader(nir);
if (shader_prog) {
nir_lower_samplers(nir, shader_prog);
nir_validate_shader(nir);
}
nir_lower_system_values(nir);
nir_validate_shader(nir);
nir_lower_atomics(nir);
nir_validate_shader(nir);
nir_optimize(nir, is_scalar);
if (brw->gen >= 6) {
/* Try and fuse multiply-adds */
nir_opt_peephole_ffma(nir);
nir_validate_shader(nir);
}
nir_opt_algebraic_late(nir);
nir_validate_shader(nir);
nir_lower_locals_to_regs(nir);
nir_validate_shader(nir);
nir_lower_to_source_mods(nir);
nir_validate_shader(nir);
nir_copy_prop(nir);
nir_validate_shader(nir);
nir_opt_dce(nir);
nir_validate_shader(nir);
if (unlikely(debug_enabled)) {
/* Re-index SSA defs so we print more sensible numbers. */
nir_foreach_overload(nir, overload) {
if (overload->impl)
nir_index_ssa_defs(overload->impl);
}
fprintf(stderr, "NIR (SSA form) for %s shader:\n",
_mesa_shader_stage_to_string(stage));
nir_print_shader(nir, stderr);
}
nir_convert_from_ssa(nir, is_scalar);
nir_validate_shader(nir);
if (!is_scalar) {
nir_lower_vec_to_movs(nir);
nir_validate_shader(nir);
}
/* This is the last pass we run before we start emitting stuff. It
* determines when we need to insert boolean resolves on Gen <= 5. We
* run it last because it stashes data in instr->pass_flags and we don't
* want that to be squashed by other NIR passes.
*/
if (brw->gen <= 5)
brw_nir_analyze_boolean_resolves(nir);
nir_sweep(nir);
if (unlikely(debug_enabled)) {
fprintf(stderr, "NIR (final form) for %s shader:\n",
_mesa_shader_stage_to_string(stage));
nir_print_shader(nir, stderr);
}
return nir;
}