From dfbabc6bad775e1575ff4a97a3c871341cd57f77 Mon Sep 17 00:00:00 2001 From: Rob Clark Date: Fri, 25 Mar 2016 15:10:50 -0400 Subject: nir/lower-io: add support for lowering inputs Signed-off-by: Rob Clark Reviewed-by: Jason Ekstrand --- src/compiler/nir/nir.h | 3 +- src/compiler/nir/nir_lower_io_to_temporaries.c | 51 ++++++++++++++++++++++---- src/mesa/drivers/dri/i965/brw_nir.c | 4 +- 3 files changed, 48 insertions(+), 10 deletions(-) diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h index 5410f0b8020..c96eaf9c3fc 100644 --- a/src/compiler/nir/nir.h +++ b/src/compiler/nir/nir.h @@ -2279,7 +2279,8 @@ bool nir_lower_indirect_derefs(nir_shader *shader, nir_variable_mode modes); bool nir_lower_locals_to_regs(nir_shader *shader); -void nir_lower_io_to_temporaries(nir_shader *shader, nir_function *entrypoint); +void nir_lower_io_to_temporaries(nir_shader *shader, nir_function *entrypoint, + bool outputs, bool inputs); void nir_shader_gather_info(nir_shader *shader, nir_function_impl *entrypoint); diff --git a/src/compiler/nir/nir_lower_io_to_temporaries.c b/src/compiler/nir/nir_lower_io_to_temporaries.c index 4b4363b8cde..3153a49b7be 100644 --- a/src/compiler/nir/nir_lower_io_to_temporaries.c +++ b/src/compiler/nir/nir_lower_io_to_temporaries.c @@ -22,9 +22,12 @@ */ /* - * Implements a pass that lowers output variables to a temporary plus an - * output variable with a single copy at each exit point of the shader. - * This way the output variable is only ever written. + * Implements a pass that lowers output and/or input variables to a + * temporary plus an output variable with a single copy at each exit + * point of the shader and/or an input variable with a single copy + * at the entrance point of the shader. This way the output variable + * is only ever written once and/or input is only read once, and there + * are no indirect outut/input accesses. */ #include "nir.h" @@ -33,6 +36,7 @@ struct lower_io_state { nir_shader *shader; nir_function *entrypoint; struct exec_list old_outputs; + struct exec_list old_inputs; }; static void @@ -88,6 +92,16 @@ emit_output_copies_impl(struct lower_io_state *state, nir_function_impl *impl) } } +static void +emit_input_copies_impl(struct lower_io_state *state, nir_function_impl *impl) +{ + if (impl->function == state->entrypoint) { + nir_cursor cursor = nir_before_block(nir_start_block(impl)); + emit_copies(cursor, state->shader, &state->old_inputs, + &state->shader->inputs); + } +} + static nir_variable * create_shadow_temp(struct lower_io_state *state, nir_variable *var) { @@ -103,8 +117,8 @@ create_shadow_temp(struct lower_io_state *state, nir_variable *var) /* Reparent the constant initializer (if any) */ ralloc_steal(nvar, nvar->constant_initializer); - /* Give the output a new name with @out-temp appended */ - const char *mode = "out"; + /* Give the original a new name with @-temp appended */ + const char *mode = (temp->data.mode == nir_var_shader_in) ? "in" : "out"; temp->name = ralloc_asprintf(var, "%s@%s-temp", mode, nvar->name); temp->data.mode = nir_var_global; temp->constant_initializer = NULL; @@ -113,7 +127,8 @@ create_shadow_temp(struct lower_io_state *state, nir_variable *var) } void -nir_lower_io_to_temporaries(nir_shader *shader, nir_function *entrypoint) +nir_lower_io_to_temporaries(nir_shader *shader, nir_function *entrypoint, + bool outputs, bool inputs) { struct lower_io_state state; @@ -122,7 +137,16 @@ nir_lower_io_to_temporaries(nir_shader *shader, nir_function *entrypoint) state.shader = shader; state.entrypoint = entrypoint; - exec_list_move_nodes_to(&shader->outputs, &state.old_outputs); + + if (inputs) + exec_list_move_nodes_to(&shader->inputs, &state.old_inputs); + else + exec_list_make_empty(&state.old_inputs); + + if (outputs) + exec_list_move_nodes_to(&shader->outputs, &state.old_outputs); + else + exec_list_make_empty(&state.old_outputs); /* Walk over all of the outputs turn each output into a temporary and * make a new variable for the actual output. @@ -132,15 +156,26 @@ nir_lower_io_to_temporaries(nir_shader *shader, nir_function *entrypoint) exec_list_push_tail(&shader->outputs, &output->node); } + /* and same for inputs: */ + nir_foreach_variable(var, &state.old_inputs) { + nir_variable *input = create_shadow_temp(&state, var); + exec_list_push_tail(&shader->inputs, &input->node); + } + nir_foreach_function(function, shader) { if (function->impl == NULL) continue; - emit_output_copies_impl(&state, function->impl); + if (inputs) + emit_input_copies_impl(&state, function->impl); + + if (outputs) + emit_output_copies_impl(&state, function->impl); nir_metadata_preserve(function->impl, nir_metadata_block_index | nir_metadata_dominance); } + exec_list_append(&shader->globals, &state.old_inputs); exec_list_append(&shader->globals, &state.old_outputs); } diff --git a/src/mesa/drivers/dri/i965/brw_nir.c b/src/mesa/drivers/dri/i965/brw_nir.c index 07d24b27442..c501bc157ef 100644 --- a/src/mesa/drivers/dri/i965/brw_nir.c +++ b/src/mesa/drivers/dri/i965/brw_nir.c @@ -563,7 +563,9 @@ brw_create_nir(struct brw_context *brw, /* First, lower the GLSL IR or Mesa IR to NIR */ if (shader_prog) { nir = glsl_to_nir(shader_prog, stage, options); - OPT_V(nir_lower_io_to_temporaries, nir_shader_get_entrypoint(nir)); + OPT_V(nir_lower_io_to_temporaries, + nir_shader_get_entrypoint(nir), + true, false); } else { nir = prog_to_nir(prog, options); OPT_V(nir_convert_to_ssa); /* turn registers into SSA */ -- cgit v1.2.3