*size = max;
}
-extern "C" {
+static void
+st_nir_opts(nir_shader *nir)
+{
+ bool progress;
+ do {
+ progress = false;
+
+ NIR_PASS(progress, nir, nir_copy_prop);
+ NIR_PASS(progress, nir, nir_opt_remove_phis);
+ NIR_PASS(progress, nir, nir_opt_dce);
+ if (nir_opt_trivial_continues(nir)) {
+ progress = true;
+ NIR_PASS(progress, nir, nir_copy_prop);
+ NIR_PASS(progress, nir, nir_opt_dce);
+ }
+ NIR_PASS(progress, nir, nir_opt_if);
+ NIR_PASS(progress, nir, nir_opt_dead_cf);
+ NIR_PASS(progress, nir, nir_opt_cse);
+ NIR_PASS(progress, nir, nir_opt_peephole_select, 8);
+
+ NIR_PASS(progress, nir, nir_opt_algebraic);
+ NIR_PASS(progress, nir, nir_opt_constant_folding);
+
+ NIR_PASS(progress, nir, nir_opt_undef);
+ NIR_PASS(progress, nir, nir_opt_conditional_discard);
+ if (nir->options->max_unroll_iterations) {
+ NIR_PASS(progress, nir, nir_opt_loop_unroll, (nir_variable_mode)0);
+ }
+ } while (progress);
+}
/* First third of converting glsl_to_nir.. this leaves things in a pre-
* nir_lower_io state, so that shader variants can more easily insert/
* replace variables, etc.
*/
-nir_shader *
+static nir_shader *
st_glsl_to_nir(struct st_context *st, struct gl_program *prog,
struct gl_shader_program *shader_program,
gl_shader_stage stage)
if (prog->nir)
return prog->nir;
- return glsl_to_nir(shader_program, stage, options);
+ nir_shader *nir = glsl_to_nir(shader_program, stage, options);
+
+ st_nir_opts(nir);
+
+ return nir;
}
/* Second third of converting glsl_to_nir. This creates uniforms, gathers
*/
_mesa_associate_uniform_storage(st->ctx, shader_program, prog, true);
- NIR_PASS_V(nir, nir_lower_io_to_temporaries,
- nir_shader_get_entrypoint(nir),
- true, true);
- NIR_PASS_V(nir, nir_lower_global_vars_to_local);
- NIR_PASS_V(nir, nir_split_var_copies);
- NIR_PASS_V(nir, nir_lower_var_copies);
-
- /* fragment shaders may need : */
- if (prog->info.stage == MESA_SHADER_FRAGMENT) {
- static const gl_state_index wposTransformState[STATE_LENGTH] = {
- STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM
- };
- nir_lower_wpos_ytransform_options wpos_options = { { 0 } };
- struct pipe_screen *pscreen = st->pipe->screen;
-
- memcpy(wpos_options.state_tokens, wposTransformState,
- sizeof(wpos_options.state_tokens));
- wpos_options.fs_coord_origin_upper_left =
- pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT);
- wpos_options.fs_coord_origin_lower_left =
- pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
- wpos_options.fs_coord_pixel_center_integer =
- pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
- wpos_options.fs_coord_pixel_center_half_integer =
- pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER);
-
- if (nir_lower_wpos_ytransform(nir, &wpos_options)) {
- nir_validate_shader(nir);
- _mesa_add_state_reference(prog->Parameters, wposTransformState);
- }
- }
-
- NIR_PASS_V(nir, nir_lower_system_values);
-
- nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));
- prog->info = nir->info;
-
st_set_prog_affected_state_flags(prog);
NIR_PASS_V(nir, st_nir_lower_builtin);
exec_list_move_nodes_to(&new_list, var_list);
}
-/* Last third of preparing nir from glsl, which happens after shader
- * variant lowering.
- */
-void
-st_finalize_nir(struct st_context *st, struct gl_program *prog,
- struct gl_shader_program *shader_program, nir_shader *nir)
-{
- struct pipe_screen *screen = st->pipe->screen;
-
- NIR_PASS_V(nir, nir_split_var_copies);
- NIR_PASS_V(nir, nir_lower_var_copies);
- NIR_PASS_V(nir, nir_lower_io_types);
-
- if (nir->info.stage == MESA_SHADER_VERTEX) {
- /* Needs special handling so drvloc matches the vbo state: */
- st_nir_assign_vs_in_locations(prog, nir);
- /* Re-lower global vars, to deal with any dead VS inputs. */
- NIR_PASS_V(nir, nir_lower_global_vars_to_local);
-
- sort_varyings(&nir->outputs);
- st_nir_assign_var_locations(&nir->outputs,
- &nir->num_outputs);
- st_nir_fixup_varying_slots(st, &nir->outputs);
- } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
- sort_varyings(&nir->inputs);
- st_nir_assign_var_locations(&nir->inputs,
- &nir->num_inputs);
- st_nir_fixup_varying_slots(st, &nir->inputs);
- st_nir_assign_var_locations(&nir->outputs,
- &nir->num_outputs);
- } else if (nir->info.stage == MESA_SHADER_COMPUTE) {
- /* TODO? */
- } else {
- unreachable("invalid shader type for tgsi bypass\n");
- }
-
- NIR_PASS_V(nir, nir_lower_atomics_to_ssbo,
- st->ctx->Const.Program[nir->info.stage].MaxAtomicBuffers);
-
- st_nir_assign_uniform_locations(prog, shader_program,
- &nir->uniforms, &nir->num_uniforms);
-
- if (screen->get_param(screen, PIPE_CAP_NIR_SAMPLERS_AS_DEREF))
- NIR_PASS_V(nir, nir_lower_samplers_as_deref, shader_program);
- else
- NIR_PASS_V(nir, nir_lower_samplers, shader_program);
-}
-
static void
set_st_program(struct gl_program *prog,
struct gl_shader_program *shader_program,
set_st_program(prog, shader_program, nir);
prog->nir = nir;
+
+ NIR_PASS_V(nir, nir_lower_io_to_temporaries,
+ nir_shader_get_entrypoint(nir),
+ true, true);
+ NIR_PASS_V(nir, nir_lower_global_vars_to_local);
+ NIR_PASS_V(nir, nir_split_var_copies);
+ NIR_PASS_V(nir, nir_lower_var_copies);
+}
+
+static void
+st_nir_link_shaders(nir_shader **producer, nir_shader **consumer)
+{
+ nir_lower_io_arrays_to_elements(*producer, *consumer);
+
+ NIR_PASS_V(*producer, nir_remove_dead_variables, nir_var_shader_out);
+ NIR_PASS_V(*consumer, nir_remove_dead_variables, nir_var_shader_in);
+
+ if (nir_remove_unused_varyings(*producer, *consumer)) {
+ NIR_PASS_V(*producer, nir_lower_global_vars_to_local);
+ NIR_PASS_V(*consumer, nir_lower_global_vars_to_local);
+
+ /* The backend might not be able to handle indirects on
+ * temporaries so we need to lower indirects on any of the
+ * varyings we have demoted here.
+ *
+ * TODO: radeonsi shouldn't need to do this, however LLVM isn't
+ * currently smart enough to handle indirects without causing excess
+ * spilling causing the gpu to hang.
+ *
+ * See the following thread for more details of the problem:
+ * https://lists.freedesktop.org/archives/mesa-dev/2017-July/162106.html
+ */
+ nir_variable_mode indirect_mask = nir_var_local;
+
+ NIR_PASS_V(*producer, nir_lower_indirect_derefs, indirect_mask);
+ NIR_PASS_V(*consumer, nir_lower_indirect_derefs, indirect_mask);
+
+ st_nir_opts(*producer);
+ st_nir_opts(*consumer);
+ }
}
+extern "C" {
+
bool
st_link_nir(struct gl_context *ctx,
struct gl_shader_program *shader_program)
{
struct st_context *st = st_context(ctx);
+ /* Determine first and last stage. */
+ unsigned first = MESA_SHADER_STAGES;
+ unsigned last = 0;
+ for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
+ if (!shader_program->_LinkedShaders[i])
+ continue;
+ if (first == MESA_SHADER_STAGES)
+ first = i;
+ last = i;
+ }
+
for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
struct gl_linked_shader *shader = shader_program->_LinkedShaders[i];
if (shader == NULL)
continue;
st_nir_get_mesa_program(ctx, shader_program, shader);
+
+ nir_variable_mode mask = (nir_variable_mode) 0;
+ if (i != first)
+ mask = (nir_variable_mode)(mask | nir_var_shader_in);
+
+ if (i != last)
+ mask = (nir_variable_mode)(mask | nir_var_shader_out);
+
+ nir_shader *nir = shader->Program->nir;
+ nir_lower_io_to_scalar_early(nir, mask);
+ st_nir_opts(nir);
+ }
+
+ /* Linking the stages in the opposite order (from fragment to vertex)
+ * ensures that inter-shader outputs written to in an earlier stage
+ * are eliminated if they are (transitively) not used in a later
+ * stage.
+ */
+ int next = last;
+ for (int i = next - 1; i >= 0; i--) {
+ struct gl_linked_shader *shader = shader_program->_LinkedShaders[i];
+ if (shader == NULL)
+ continue;
+
+ st_nir_link_shaders(&shader->Program->nir,
+ &shader_program->_LinkedShaders[next]->Program->nir);
+ next = i;
}
+ int prev = -1;
for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
struct gl_linked_shader *shader = shader_program->_LinkedShaders[i];
if (shader == NULL)
continue;
+ nir_shader *nir = shader->Program->nir;
+
+ /* fragment shaders may need : */
+ if (nir->info.stage == MESA_SHADER_FRAGMENT) {
+ static const gl_state_index wposTransformState[STATE_LENGTH] = {
+ STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM
+ };
+ nir_lower_wpos_ytransform_options wpos_options = { { 0 } };
+ struct pipe_screen *pscreen = st->pipe->screen;
+
+ memcpy(wpos_options.state_tokens, wposTransformState,
+ sizeof(wpos_options.state_tokens));
+ wpos_options.fs_coord_origin_upper_left =
+ pscreen->get_param(pscreen,
+ PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT);
+ wpos_options.fs_coord_origin_lower_left =
+ pscreen->get_param(pscreen,
+ PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
+ wpos_options.fs_coord_pixel_center_integer =
+ pscreen->get_param(pscreen,
+ PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
+ wpos_options.fs_coord_pixel_center_half_integer =
+ pscreen->get_param(pscreen,
+ PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER);
+
+ if (nir_lower_wpos_ytransform(nir, &wpos_options)) {
+ nir_validate_shader(nir);
+ _mesa_add_state_reference(shader->Program->Parameters,
+ wposTransformState);
+ }
+ }
+
+ NIR_PASS_V(nir, nir_lower_system_values);
+
+ nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));
+ shader->Program->info = nir->info;
+
+ if (prev != -1) {
+ nir_compact_varyings(shader_program->_LinkedShaders[prev]->Program->nir,
+ nir, ctx->API != API_OPENGL_COMPAT);
+ }
+ prev = i;
+
st_glsl_to_nir_post_opts(st, shader->Program, shader_program);
assert(shader->Program);
return true;
}
+/* Last third of preparing nir from glsl, which happens after shader
+ * variant lowering.
+ */
+void
+st_finalize_nir(struct st_context *st, struct gl_program *prog,
+ struct gl_shader_program *shader_program, nir_shader *nir)
+{
+ struct pipe_screen *screen = st->pipe->screen;
+
+ NIR_PASS_V(nir, nir_split_var_copies);
+ NIR_PASS_V(nir, nir_lower_var_copies);
+ NIR_PASS_V(nir, nir_lower_io_types);
+
+ if (nir->info.stage == MESA_SHADER_VERTEX) {
+ /* Needs special handling so drvloc matches the vbo state: */
+ st_nir_assign_vs_in_locations(prog, nir);
+ /* Re-lower global vars, to deal with any dead VS inputs. */
+ NIR_PASS_V(nir, nir_lower_global_vars_to_local);
+
+ sort_varyings(&nir->outputs);
+ st_nir_assign_var_locations(&nir->outputs,
+ &nir->num_outputs);
+ st_nir_fixup_varying_slots(st, &nir->outputs);
+ } else if (nir->info.stage == MESA_SHADER_FRAGMENT) {
+ sort_varyings(&nir->inputs);
+ st_nir_assign_var_locations(&nir->inputs,
+ &nir->num_inputs);
+ st_nir_fixup_varying_slots(st, &nir->inputs);
+ st_nir_assign_var_locations(&nir->outputs,
+ &nir->num_outputs);
+ } else if (nir->info.stage == MESA_SHADER_COMPUTE) {
+ /* TODO? */
+ } else {
+ unreachable("invalid shader type for tgsi bypass\n");
+ }
+
+ NIR_PASS_V(nir, nir_lower_atomics_to_ssbo,
+ st->ctx->Const.Program[nir->info.stage].MaxAtomicBuffers);
+
+ st_nir_assign_uniform_locations(prog, shader_program,
+ &nir->uniforms, &nir->num_uniforms);
+
+ if (screen->get_param(screen, PIPE_CAP_NIR_SAMPLERS_AS_DEREF))
+ NIR_PASS_V(nir, nir_lower_samplers_as_deref, shader_program);
+ else
+ NIR_PASS_V(nir, nir_lower_samplers, shader_program);
+}
+
} /* extern "C" */
projects / mesa.git / blobdiff