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;
bool progress;
nir_shader *nir;
/* First, lower the GLSL IR or Mesa IR to NIR */
if (shader_prog) {
nir = glsl_to_nir(shader_prog, stage, options);
nir_remove_dead_variables(nir, nir_var_shader_in | nir_var_shader_out);
NIR_PASS_V(nir, nir_lower_io_to_temporaries,
nir_shader_get_entrypoint(nir), true, false);
} else {
nir = prog_to_nir(prog, options);
NIR_PASS_V(nir, nir_convert_to_ssa); /* turn registers into SSA */
}
nir_validate_shader(nir);
(void)progress;
nir = brw_preprocess_nir(brw->screen->compiler, nir);
if (stage == MESA_SHADER_FRAGMENT) {
static const struct nir_lower_wpos_ytransform_options wpos_options = {
.state_tokens = {STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM, 0, 0, 0},
.fs_coord_pixel_center_integer = 1,
.fs_coord_origin_upper_left = 1,
};
_mesa_add_state_reference(prog->Parameters,
(gl_state_index *) wpos_options.state_tokens);
NIR_PASS(progress, nir, nir_lower_wpos_ytransform, &wpos_options);
}
NIR_PASS(progress, nir, nir_lower_system_values);
NIR_PASS_V(nir, brw_nir_lower_uniforms, is_scalar);
nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));
if (shader_prog) {
NIR_PASS_V(nir, nir_lower_samplers, shader_prog);
NIR_PASS_V(nir, nir_lower_atomics, shader_prog);
}
return nir;
}
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;
}
static void
nir_optimize(nir_shader *nir, bool is_scalar)
{
bool progress;
do {
progress = false;
nir_lower_vars_to_ssa(nir);
nir_validate_shader(nir);
if (is_scalar) {
nir_lower_alu_to_scalar(nir);
nir_validate_shader(nir);
}
progress |= nir_copy_prop(nir);
nir_validate_shader(nir);
if (is_scalar) {
nir_lower_phis_to_scalar(nir);
nir_validate_shader(nir);
}
progress |= nir_copy_prop(nir);
nir_validate_shader(nir);
progress |= nir_opt_dce(nir);
nir_validate_shader(nir);
progress |= nir_opt_cse(nir);
nir_validate_shader(nir);
progress |= nir_opt_peephole_select(nir);
nir_validate_shader(nir);
progress |= nir_opt_algebraic(nir);
nir_validate_shader(nir);
progress |= nir_opt_constant_folding(nir);
nir_validate_shader(nir);
progress |= nir_opt_remove_phis(nir);
nir_validate_shader(nir);
} while (progress);
}
static nir_shader *
create_passthrough_tcs(void *mem_ctx, const struct brw_compiler *compiler,
const nir_shader_compiler_options *options,
const struct brw_tcs_prog_key *key)
{
nir_builder b;
nir_builder_init_simple_shader(&b, mem_ctx, MESA_SHADER_TESS_CTRL,
options);
nir_shader *nir = b.shader;
nir_variable *var;
nir_intrinsic_instr *load;
nir_intrinsic_instr *store;
nir_ssa_def *zero = nir_imm_int(&b, 0);
nir_ssa_def *invoc_id =
nir_load_system_value(&b, nir_intrinsic_load_invocation_id, 0);
nir->info->inputs_read = key->outputs_written;
nir->info->outputs_written = key->outputs_written;
nir->info->tcs.vertices_out = key->input_vertices;
nir->info->name = ralloc_strdup(nir, "passthrough");
nir->num_uniforms = 8 * sizeof(uint32_t);
var = nir_variable_create(nir, nir_var_uniform, glsl_vec4_type(), "hdr_0");
var->data.location = 0;
var = nir_variable_create(nir, nir_var_uniform, glsl_vec4_type(), "hdr_1");
var->data.location = 1;
/* Write the patch URB header. */
for (int i = 0; i <= 1; i++) {
load = nir_intrinsic_instr_create(nir, nir_intrinsic_load_uniform);
load->num_components = 4;
load->src[0] = nir_src_for_ssa(zero);
nir_ssa_dest_init(&load->instr, &load->dest, 4, 32, NULL);
nir_intrinsic_set_base(load, i * 4 * sizeof(uint32_t));
nir_builder_instr_insert(&b, &load->instr);
store = nir_intrinsic_instr_create(nir, nir_intrinsic_store_output);
store->num_components = 4;
store->src[0] = nir_src_for_ssa(&load->dest.ssa);
store->src[1] = nir_src_for_ssa(zero);
nir_intrinsic_set_base(store, VARYING_SLOT_TESS_LEVEL_INNER - i);
nir_intrinsic_set_write_mask(store, WRITEMASK_XYZW);
nir_builder_instr_insert(&b, &store->instr);
}
/* Copy inputs to outputs. */
uint64_t varyings = key->outputs_written;
while (varyings != 0) {
const int varying = ffsll(varyings) - 1;
load = nir_intrinsic_instr_create(nir,
nir_intrinsic_load_per_vertex_input);
load->num_components = 4;
load->src[0] = nir_src_for_ssa(invoc_id);
load->src[1] = nir_src_for_ssa(zero);
nir_ssa_dest_init(&load->instr, &load->dest, 4, 32, NULL);
nir_intrinsic_set_base(load, varying);
nir_builder_instr_insert(&b, &load->instr);
store = nir_intrinsic_instr_create(nir,
nir_intrinsic_store_per_vertex_output);
store->num_components = 4;
store->src[0] = nir_src_for_ssa(&load->dest.ssa);
store->src[1] = nir_src_for_ssa(invoc_id);
store->src[2] = nir_src_for_ssa(zero);
nir_intrinsic_set_base(store, varying);
nir_intrinsic_set_write_mask(store, WRITEMASK_XYZW);
nir_builder_instr_insert(&b, &store->instr);
varyings &= ~BITFIELD64_BIT(varying);
}
nir_validate_shader(nir);
nir = brw_preprocess_nir(compiler, nir);
return nir;
}
