/* * Copyright © 2015 Intel Corporation * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), * to deal in the Software without restriction, including without limitation * the rights to use, copy, modify, merge, publish, distribute, sublicense, * and/or sell copies of the Software, and to permit persons to whom the * Software is furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice (including the next * paragraph) shall be included in all copies or substantial portions of the * Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS * IN THE SOFTWARE. */ #include "nir.h" #include "nir_builder.h" #include "nir_control_flow.h" #include "nir_vla.h" /* * TODO: write a proper inliner for GPUs. * This heuristic just inlines small functions, * and tail calls get inlined as well. */ static bool nir_function_can_inline(nir_function *function) { bool can_inline = true; if (!function->should_inline) { if (function->impl) { if (function->impl->num_blocks > 2) can_inline = false; if (function->impl->ssa_alloc > 45) can_inline = false; } } return can_inline; } static bool function_ends_in_jump(nir_function_impl *impl) { nir_block *last_block = nir_impl_last_block(impl); return nir_block_ends_in_jump(last_block); } /* A cast is used to deref function in/out params. However the bindless * textures spec allows both uniforms and functions temps to be passed to a * function param defined the same way. To deal with this we need to update * this when we inline and know what variable mode we are dealing with. */ static void fixup_cast_deref_mode(nir_deref_instr *deref) { nir_deref_instr *parent = nir_src_as_deref(deref->parent); if (parent && parent->modes & nir_var_uniform && deref->modes & nir_var_function_temp) { deref->modes |= nir_var_uniform; deref->modes ^= nir_var_function_temp; nir_foreach_use(use, &deref->def) { if (nir_src_parent_instr(use)->type != nir_instr_type_deref) continue; /* Recurse into children */ fixup_cast_deref_mode(nir_instr_as_deref(nir_src_parent_instr(use))); } } } void nir_inline_function_impl(struct nir_builder *b, const nir_function_impl *impl, nir_def **params, struct hash_table *shader_var_remap) { nir_function_impl *copy = nir_function_impl_clone(b->shader, impl); exec_list_append(&b->impl->locals, ©->locals); nir_foreach_block(block, copy) { nir_foreach_instr_safe(instr, block) { switch (instr->type) { case nir_instr_type_deref: { nir_deref_instr *deref = nir_instr_as_deref(instr); /* Note: This shouldn't change the mode of anything but the * replaced nir_intrinsic_load_param intrinsics handled later in * this switch table. Any incorrect modes should have already been * detected by previous nir_vaidate calls. */ if (deref->deref_type == nir_deref_type_cast) { fixup_cast_deref_mode(deref); break; } if (deref->deref_type != nir_deref_type_var) break; /* We don't need to remap function variables. We already cloned * them as part of nir_function_impl_clone and appended them to * b->impl->locals. */ if (deref->var->data.mode == nir_var_function_temp) break; /* If no map is provided, we assume that there are either no * shader variables or they already live b->shader (this is the * case for function inlining within a single shader. */ if (shader_var_remap == NULL) break; struct hash_entry *entry = _mesa_hash_table_search(shader_var_remap, deref->var); if (entry == NULL) { nir_variable *nvar = nir_variable_clone(deref->var, b->shader); nir_shader_add_variable(b->shader, nvar); entry = _mesa_hash_table_insert(shader_var_remap, deref->var, nvar); } deref->var = entry->data; break; } case nir_instr_type_intrinsic: { nir_intrinsic_instr *load = nir_instr_as_intrinsic(instr); if (load->intrinsic != nir_intrinsic_load_param) break; unsigned param_idx = nir_intrinsic_param_idx(load); assert(param_idx < impl->function->num_params); nir_def_replace(&load->def, params[param_idx]); break; } case nir_instr_type_jump: /* Returns have to be lowered for this to work */ assert(nir_instr_as_jump(instr)->type != nir_jump_return); break; default: break; } } } bool nest_if = function_ends_in_jump(copy); /* Pluck the body out of the function and place it here */ nir_cf_list body; nir_cf_list_extract(&body, ©->body); if (nest_if) { nir_if *cf = nir_push_if(b, nir_imm_true(b)); nir_cf_reinsert(&body, nir_after_cf_list(&cf->then_list)); nir_pop_if(b, cf); } else { /* Insert a nop at the cursor so we can keep track of where things are as * we add/remove stuff from the CFG. */ nir_intrinsic_instr *nop = nir_nop(b); nir_cf_reinsert(&body, nir_before_instr(&nop->instr)); b->cursor = nir_instr_remove(&nop->instr); } } static bool inline_function_impl(nir_function_impl *impl, struct set *inlined); static bool inline_functions_pass(nir_builder *b, nir_instr *instr, void *cb_data) { struct set *inlined = cb_data; if (instr->type != nir_instr_type_call) return false; nir_call_instr *call = nir_instr_as_call(instr); assert(call->callee->impl); if (b->shader->options->driver_functions && b->shader->info.stage == MESA_SHADER_KERNEL) { bool last_instr = (instr == nir_block_last_instr(instr->block)); if (!nir_function_can_inline(call->callee) && !last_instr) { return false; } } /* Make sure that the function we're calling is already inlined */ inline_function_impl(call->callee->impl, inlined); b->cursor = nir_instr_remove(&call->instr); /* Rewrite all of the uses of the callee's parameters to use the call * instructions sources. In order to ensure that the "load" happens * here and not later (for register sources), we make sure to convert it * to an SSA value first. */ const unsigned num_params = call->num_params; NIR_VLA(nir_def *, params, num_params); for (unsigned i = 0; i < num_params; i++) { params[i] = call->params[i].ssa; } nir_inline_function_impl(b, call->callee->impl, params, NULL); return true; } static bool inline_function_impl(nir_function_impl *impl, struct set *inlined) { if (_mesa_set_search(inlined, impl)) return false; /* Already inlined */ bool progress; progress = nir_function_instructions_pass(impl, inline_functions_pass, nir_metadata_none, inlined); if (progress) { /* Indices are completely messed up now */ nir_index_ssa_defs(impl); } _mesa_set_add(inlined, impl); return progress; } /** A pass to inline all functions in a shader into their callers * * For most use-cases, function inlining is a multi-step process. The general * pattern employed by SPIR-V consumers and others is as follows: * * 1. nir_lower_variable_initializers(shader, nir_var_function_temp) * * This is needed because local variables from the callee are simply added * to the locals list for the caller and the information about where the * constant initializer logically happens is lost. If the callee is * called in a loop, this can cause the variable to go from being * initialized once per loop iteration to being initialized once at the * top of the caller and values to persist from one invocation of the * callee to the next. The simple solution to this problem is to get rid * of constant initializers before function inlining. * * 2. nir_lower_returns(shader) * * nir_inline_functions assumes that all functions end "naturally" by * execution reaching the end of the function without any return * instructions causing instant jumps to the end. Thanks to NIR being * structured, we can't represent arbitrary jumps to various points in the * program which is what an early return in the callee would have to turn * into when we inline it into the caller. Instead, we require returns to * be lowered which lets us just copy+paste the callee directly into the * caller. * * 3. nir_inline_functions(shader) * * This does the actual function inlining and the resulting shader will * contain no call instructions. * * 4. nir_opt_deref(shader) * * Most functions contain pointer parameters where the result of a deref * instruction is passed in as a parameter, loaded via a load_param * intrinsic, and then turned back into a deref via a cast. Function * inlining will get rid of the load_param but we are still left with a * cast. Running nir_opt_deref gets rid of the intermediate cast and * results in a whole deref chain again. This is currently required by a * number of optimizations and lowering passes at least for certain * variable modes. * * 5. Loop over the functions and delete all but the main entrypoint. * * In the Intel Vulkan driver this looks like this: * * nir_remove_non_entrypoints(nir); * * While nir_inline_functions does get rid of all call instructions, it * doesn't get rid of any functions because it doesn't know what the "root * function" is. Instead, it's up to the individual driver to know how to * decide on a root function and delete the rest. With SPIR-V, * spirv_to_nir returns the root function and so we can just use == whereas * with GL, you may have to look for a function named "main". * * 6. nir_lower_variable_initializers(shader, ~nir_var_function_temp) * * Lowering constant initializers on inputs, outputs, global variables, * etc. requires that we know the main entrypoint so that we know where to * initialize them. Otherwise, we would have to assume that anything * could be a main entrypoint and initialize them at the start of every * function but that would clearly be wrong if any of those functions were * ever called within another function. Simply requiring a single- * entrypoint function shader is the best way to make it well-defined. */ bool nir_inline_functions(nir_shader *shader) { struct set *inlined = _mesa_pointer_set_create(NULL); bool progress = false; nir_foreach_function_impl(impl, shader) { progress = inline_function_impl(impl, inlined) || progress; } _mesa_set_destroy(inlined, NULL); return progress; } struct lower_link_state { struct hash_table *shader_var_remap; const nir_shader *link_shader; unsigned printf_index_offset; }; static bool lower_calls_vars_instr(struct nir_builder *b, nir_instr *instr, void *cb_data) { struct lower_link_state *state = cb_data; switch (instr->type) { case nir_instr_type_deref: { nir_deref_instr *deref = nir_instr_as_deref(instr); if (deref->deref_type != nir_deref_type_var) return false; if (deref->var->data.mode == nir_var_function_temp) return false; assert(state->shader_var_remap); struct hash_entry *entry = _mesa_hash_table_search(state->shader_var_remap, deref->var); if (entry == NULL) { nir_variable *nvar = nir_variable_clone(deref->var, b->shader); nir_shader_add_variable(b->shader, nvar); entry = _mesa_hash_table_insert(state->shader_var_remap, deref->var, nvar); } deref->var = entry->data; break; } case nir_instr_type_call: { nir_call_instr *ncall = nir_instr_as_call(instr); if (!ncall->callee->name) return false; nir_function *func = nir_shader_get_function_for_name(b->shader, ncall->callee->name); if (func) { ncall->callee = func; break; } nir_function *new_func; new_func = nir_shader_get_function_for_name(state->link_shader, ncall->callee->name); if (new_func) ncall->callee = nir_function_clone(b->shader, new_func); break; } case nir_instr_type_intrinsic: { /* Reindex the offset of the printf intrinsic by the number of already * present printfs in the shader where functions are linked into. */ if (state->printf_index_offset == 0) return false; nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr); if (intrin->intrinsic != nir_intrinsic_printf) return false; b->cursor = nir_before_instr(instr); nir_src_rewrite(&intrin->src[0], nir_iadd_imm(b, intrin->src[0].ssa, state->printf_index_offset)); break; } default: break; } return true; } static bool lower_call_function_impl(struct nir_builder *b, nir_function *callee, const nir_function_impl *impl, struct lower_link_state *state) { nir_function_impl *copy = nir_function_impl_clone(b->shader, impl); copy->function = callee; callee->impl = copy; return nir_function_instructions_pass(copy, lower_calls_vars_instr, nir_metadata_none, state); } static bool function_link_pass(struct nir_builder *b, nir_instr *instr, void *cb_data) { struct lower_link_state *state = cb_data; if (instr->type != nir_instr_type_call) return false; nir_call_instr *call = nir_instr_as_call(instr); nir_function *func = NULL; if (!call->callee->name) return false; if (call->callee->impl) return false; func = nir_shader_get_function_for_name(state->link_shader, call->callee->name); if (!func || !func->impl) { return false; } return lower_call_function_impl(b, call->callee, func->impl, state); } bool nir_link_shader_functions(nir_shader *shader, const nir_shader *link_shader) { void *ra_ctx = ralloc_context(NULL); struct hash_table *copy_vars = _mesa_pointer_hash_table_create(ra_ctx); bool progress = false, overall_progress = false; struct lower_link_state state = { .shader_var_remap = copy_vars, .link_shader = link_shader, .printf_index_offset = shader->printf_info_count, }; /* do progress passes inside the pass */ do { progress = false; nir_foreach_function_impl(impl, shader) { bool this_progress = nir_function_instructions_pass(impl, function_link_pass, nir_metadata_none, &state); if (this_progress) nir_index_ssa_defs(impl); progress |= this_progress; } overall_progress |= progress; } while (progress); if (overall_progress && link_shader->printf_info_count > 0) { shader->printf_info = reralloc(shader, shader->printf_info, u_printf_info, shader->printf_info_count + link_shader->printf_info_count); for (unsigned i = 0; i < link_shader->printf_info_count; i++){ const u_printf_info *src_info = &link_shader->printf_info[i]; u_printf_info *dst_info = &shader->printf_info[shader->printf_info_count++]; dst_info->num_args = src_info->num_args; dst_info->arg_sizes = ralloc_array(shader, unsigned, dst_info->num_args); memcpy(dst_info->arg_sizes, src_info->arg_sizes, sizeof(dst_info->arg_sizes[0]) * dst_info->num_args); dst_info->string_size = src_info->string_size; dst_info->strings = ralloc_memdup(shader, src_info->strings, dst_info->string_size); } } ralloc_free(ra_ctx); return overall_progress; } static void nir_mark_used_functions(struct nir_function *func, struct set *used_funcs); static bool mark_used_pass_cb(struct nir_builder *b, nir_instr *instr, void *data) { struct set *used_funcs = data; if (instr->type != nir_instr_type_call) return false; nir_call_instr *call = nir_instr_as_call(instr); _mesa_set_add(used_funcs, call->callee); nir_mark_used_functions(call->callee, used_funcs); return true; } static void nir_mark_used_functions(struct nir_function *func, struct set *used_funcs) { if (func->impl) { nir_function_instructions_pass(func->impl, mark_used_pass_cb, nir_metadata_none, used_funcs); } } void nir_cleanup_functions(nir_shader *nir) { if (!nir->options->driver_functions) { nir_remove_non_entrypoints(nir); return; } struct set *used_funcs = _mesa_set_create(NULL, _mesa_hash_pointer, _mesa_key_pointer_equal); foreach_list_typed_safe(nir_function, func, node, &nir->functions) { if (func->is_entrypoint) { _mesa_set_add(used_funcs, func); nir_mark_used_functions(func, used_funcs); } } foreach_list_typed_safe(nir_function, func, node, &nir->functions) { if (!_mesa_set_search(used_funcs, func)) exec_node_remove(&func->node); } _mesa_set_destroy(used_funcs, NULL); }