/* * Copyright © 2015 Thomas Helland * * 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_loop_analyze.h" #include "util/bitset.h" #include "nir.h" #include "nir_constant_expressions.h" typedef enum { undefined, basic_induction } nir_loop_variable_type; typedef struct { /* A link for the work list */ struct list_head process_link; bool in_loop; /* The ssa_def associated with this info */ nir_def *def; /* The type of this ssa_def */ nir_loop_variable_type type; /* True if variable is in an if branch */ bool in_if_branch; /* True if variable is in a nested loop */ bool in_nested_loop; /* Could be a basic_induction if following uniforms are inlined */ nir_src *init_src; nir_alu_src *update_src; /** * SSA def of the phi-node associated with this induction variable. * * Every loop induction variable has an associated phi node in the loop * header. This may point to the same SSA def as \c def. If, however, \c def * is the increment of the induction variable, this will point to the SSA * def being incremented. */ nir_def *basis; } nir_loop_variable; typedef struct { /* The loop we store information for */ nir_loop *loop; /* Loop_variable for all ssa_defs in function */ nir_loop_variable *loop_vars; BITSET_WORD *loop_vars_init; /* A list of the loop_vars to analyze */ struct list_head process_list; nir_variable_mode indirect_mask; bool force_unroll_sampler_indirect; } loop_info_state; static nir_loop_variable * get_loop_var(nir_def *value, loop_info_state *state) { nir_loop_variable *var = &(state->loop_vars[value->index]); if (!BITSET_TEST(state->loop_vars_init, value->index)) { var->in_loop = false; var->def = value; var->in_if_branch = false; var->in_nested_loop = false; var->init_src = NULL; var->update_src = NULL; var->type = undefined; BITSET_SET(state->loop_vars_init, value->index); } return var; } typedef struct { loop_info_state *state; bool in_if_branch; bool in_nested_loop; } init_loop_state; static bool init_loop_def(nir_def *def, void *void_init_loop_state) { init_loop_state *loop_init_state = void_init_loop_state; nir_loop_variable *var = get_loop_var(def, loop_init_state->state); if (loop_init_state->in_nested_loop) { var->in_nested_loop = true; } else if (loop_init_state->in_if_branch) { var->in_if_branch = true; } else { /* Add to the tail of the list. That way we start at the beginning of * the defs in the loop instead of the end when walking the list. This * means less recursive calls. Only add defs that are not in nested * loops or conditional blocks. */ list_addtail(&var->process_link, &loop_init_state->state->process_list); } var->in_loop = true; return true; } /** Calculate an estimated cost in number of instructions * * We do this so that we don't unroll loops which will later get massively * inflated due to int64 or fp64 lowering. The estimates provided here don't * have to be massively accurate; they just have to be good enough that loop * unrolling doesn't cause things to blow up too much. */ static unsigned instr_cost(loop_info_state *state, nir_instr *instr, const nir_shader_compiler_options *options) { if (instr->type == nir_instr_type_intrinsic || instr->type == nir_instr_type_tex) return 1; if (instr->type != nir_instr_type_alu) return 0; nir_alu_instr *alu = nir_instr_as_alu(instr); const nir_op_info *info = &nir_op_infos[alu->op]; unsigned cost = 1; if (nir_op_is_selection(alu->op)) { nir_scalar cond_scalar = { alu->src[0].src.ssa, 0 }; if (nir_is_terminator_condition_with_two_inputs(cond_scalar)) { nir_instr *sel_cond = alu->src[0].src.ssa->parent_instr; nir_alu_instr *sel_alu = nir_instr_as_alu(sel_cond); nir_scalar rhs, lhs; lhs = nir_scalar_chase_alu_src(cond_scalar, 0); rhs = nir_scalar_chase_alu_src(cond_scalar, 1); /* If the selects condition is a comparision between a constant and * a basic induction variable we know that it will be eliminated once * the loop is unrolled so here we assign it a cost of 0. */ if ((nir_src_is_const(sel_alu->src[0].src) && get_loop_var(rhs.def, state)->type == basic_induction) || (nir_src_is_const(sel_alu->src[1].src) && get_loop_var(lhs.def, state)->type == basic_induction)) { /* Also if the selects condition is only used by the select then * remove that alu instructons cost from the cost total also. */ if (!list_is_singular(&sel_alu->def.uses) || nir_def_used_by_if(&sel_alu->def)) return 0; else return -1; } } } if (alu->op == nir_op_flrp) { if ((options->lower_flrp16 && alu->def.bit_size == 16) || (options->lower_flrp32 && alu->def.bit_size == 32) || (options->lower_flrp64 && alu->def.bit_size == 64)) cost *= 3; } /* Assume everything 16 or 32-bit is cheap. * * There are no 64-bit ops that don't have a 64-bit thing as their * destination or first source. */ if (alu->def.bit_size < 64 && nir_src_bit_size(alu->src[0].src) < 64) return cost; bool is_fp64 = alu->def.bit_size == 64 && nir_alu_type_get_base_type(info->output_type) == nir_type_float; for (unsigned i = 0; i < info->num_inputs; i++) { if (nir_src_bit_size(alu->src[i].src) == 64 && nir_alu_type_get_base_type(info->input_types[i]) == nir_type_float) is_fp64 = true; } if (is_fp64) { /* If it's something lowered normally, it's expensive. */ if (options->lower_doubles_options & nir_lower_doubles_op_to_options_mask(alu->op)) cost *= 20; /* If it's full software, it's even more expensive */ if (options->lower_doubles_options & nir_lower_fp64_full_software) { cost *= 100; state->loop->info->has_soft_fp64 = true; } return cost; } else { if (options->lower_int64_options & nir_lower_int64_op_to_options_mask(alu->op)) { /* These require a doing the division algorithm. */ if (alu->op == nir_op_idiv || alu->op == nir_op_udiv || alu->op == nir_op_imod || alu->op == nir_op_umod || alu->op == nir_op_irem) return cost * 100; /* Other int64 lowering isn't usually all that expensive */ return cost * 5; } return cost; } } static bool init_loop_block(nir_block *block, loop_info_state *state, bool in_if_branch, bool in_nested_loop) { init_loop_state init_state = { .in_if_branch = in_if_branch, .in_nested_loop = in_nested_loop, .state = state }; nir_foreach_instr(instr, block) { nir_foreach_def(instr, init_loop_def, &init_state); } return true; } static inline bool is_var_alu(nir_loop_variable *var) { return var->def->parent_instr->type == nir_instr_type_alu; } static inline bool is_var_phi(nir_loop_variable *var) { return var->def->parent_instr->type == nir_instr_type_phi; } /* If all of the instruction sources point to identical ALU instructions (as * per nir_instrs_equal), return one of the ALU instructions. Otherwise, * return NULL. */ static nir_alu_instr * phi_instr_as_alu(nir_phi_instr *phi) { nir_alu_instr *first = NULL; nir_foreach_phi_src(src, phi) { if (src->src.ssa->parent_instr->type != nir_instr_type_alu) return NULL; nir_alu_instr *alu = nir_instr_as_alu(src->src.ssa->parent_instr); if (first == NULL) { first = alu; } else { if (!nir_instrs_equal(&first->instr, &alu->instr)) return NULL; } } return first; } static bool alu_src_has_identity_swizzle(nir_alu_instr *alu, unsigned src_idx) { assert(nir_op_infos[alu->op].input_sizes[src_idx] == 0); for (unsigned i = 0; i < alu->def.num_components; i++) { if (alu->src[src_idx].swizzle[i] != i) return false; } return true; } static bool is_only_uniform_src(nir_src *src) { nir_instr *instr = src->ssa->parent_instr; switch (instr->type) { case nir_instr_type_alu: { /* Return true if all sources return true. */ nir_alu_instr *alu = nir_instr_as_alu(instr); for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++) { if (!is_only_uniform_src(&alu->src[i].src)) return false; } return true; } case nir_instr_type_intrinsic: { nir_intrinsic_instr *inst = nir_instr_as_intrinsic(instr); /* current uniform inline only support load ubo */ return inst->intrinsic == nir_intrinsic_load_ubo; } case nir_instr_type_load_const: /* Always return true for constants. */ return true; default: return false; } } static bool compute_induction_information(loop_info_state *state) { unsigned num_induction_vars = 0; list_for_each_entry_safe(nir_loop_variable, var, &state->process_list, process_link) { /* Things in nested loops or conditionals should not have been added into * the procss_list. */ assert(!var->in_if_branch && !var->in_nested_loop); /* We are only interested in checking phis for the basic induction * variable case as its simple to detect. All basic induction variables * have a phi node */ if (!is_var_phi(var)) continue; nir_phi_instr *phi = nir_instr_as_phi(var->def->parent_instr); nir_loop_variable *alu_src_var = NULL; nir_foreach_phi_src(src, phi) { nir_loop_variable *src_var = get_loop_var(src->src.ssa, state); /* If one of the sources is in an if branch or nested loop then don't * attempt to go any further. */ if (src_var->in_if_branch || src_var->in_nested_loop) break; /* Detect inductions variables that are incremented in both branches * of an unnested if rather than in a loop block. */ if (is_var_phi(src_var)) { nir_phi_instr *src_phi = nir_instr_as_phi(src_var->def->parent_instr); nir_alu_instr *src_phi_alu = phi_instr_as_alu(src_phi); if (src_phi_alu) { src_var = get_loop_var(&src_phi_alu->def, state); if (!src_var->in_if_branch) break; } } if (!src_var->in_loop && !var->init_src) { var->init_src = &src->src; } else if (is_var_alu(src_var) && !var->update_src) { alu_src_var = src_var; nir_alu_instr *alu = nir_instr_as_alu(src_var->def->parent_instr); /* Check for unsupported alu operations */ if (alu->op != nir_op_iadd && alu->op != nir_op_fadd && alu->op != nir_op_imul && alu->op != nir_op_fmul && alu->op != nir_op_ishl && alu->op != nir_op_ishr && alu->op != nir_op_ushr) break; if (nir_op_infos[alu->op].num_inputs == 2) { for (unsigned i = 0; i < 2; i++) { /* Is one of the operands const or uniform, and the other the phi. * The phi source can't be swizzled in any way. */ if (alu->src[1 - i].src.ssa == &phi->def && alu_src_has_identity_swizzle(alu, 1 - i)) { if (is_only_uniform_src(&alu->src[i].src)) var->update_src = alu->src + i; } } } if (!var->update_src) break; } else { var->update_src = NULL; break; } } if (var->update_src && var->init_src && is_only_uniform_src(var->init_src)) { alu_src_var->init_src = var->init_src; alu_src_var->update_src = var->update_src; alu_src_var->basis = var->def; alu_src_var->type = basic_induction; var->basis = var->def; var->type = basic_induction; num_induction_vars += 2; } else { var->init_src = NULL; var->update_src = NULL; var->basis = NULL; } } nir_loop_info *info = state->loop->info; ralloc_free(info->induction_vars); info->num_induction_vars = 0; /* record induction variables into nir_loop_info */ if (num_induction_vars) { info->induction_vars = ralloc_array(info, nir_loop_induction_variable, num_induction_vars); list_for_each_entry(nir_loop_variable, var, &state->process_list, process_link) { if (var->type == basic_induction) { nir_loop_induction_variable *ivar = &info->induction_vars[info->num_induction_vars++]; ivar->def = var->def; ivar->init_src = var->init_src; ivar->update_src = var->update_src; } } /* don't overflow */ assert(info->num_induction_vars <= num_induction_vars); } return num_induction_vars != 0; } static bool find_loop_terminators(loop_info_state *state) { bool success = false; foreach_list_typed_safe(nir_cf_node, node, node, &state->loop->body) { if (node->type == nir_cf_node_if) { nir_if *nif = nir_cf_node_as_if(node); nir_block *break_blk = NULL; nir_block *continue_from_blk = NULL; bool continue_from_then = true; nir_block *last_then = nir_if_last_then_block(nif); nir_block *last_else = nir_if_last_else_block(nif); if (nir_block_ends_in_break(last_then)) { break_blk = last_then; continue_from_blk = last_else; continue_from_then = false; } else if (nir_block_ends_in_break(last_else)) { break_blk = last_else; continue_from_blk = last_then; } /* If there is a break then we should find a terminator. If we can * not find a loop terminator, but there is a break-statement then * we should return false so that we do not try to find trip-count */ if (!nir_is_trivial_loop_if(nif, break_blk)) { state->loop->info->complex_loop = true; return false; } /* Continue if the if contained no jumps at all */ if (!break_blk) continue; if (nif->condition.ssa->parent_instr->type == nir_instr_type_phi) { state->loop->info->complex_loop = true; return false; } nir_loop_terminator *terminator = rzalloc(state->loop->info, nir_loop_terminator); list_addtail(&terminator->loop_terminator_link, &state->loop->info->loop_terminator_list); terminator->nif = nif; terminator->break_block = break_blk; terminator->continue_from_block = continue_from_blk; terminator->continue_from_then = continue_from_then; terminator->conditional_instr = nif->condition.ssa->parent_instr; success = true; } } return success; } /* This function looks for an array access within a loop that uses an * induction variable for the array index. If found it returns the size of the * array, otherwise 0 is returned. If we find an induction var we pass it back * to the caller via array_index_out. */ static unsigned find_array_access_via_induction(loop_info_state *state, nir_deref_instr *deref, nir_loop_variable **array_index_out) { for (nir_deref_instr *d = deref; d; d = nir_deref_instr_parent(d)) { if (d->deref_type != nir_deref_type_array) continue; nir_loop_variable *array_index = get_loop_var(d->arr.index.ssa, state); if (array_index->type != basic_induction) continue; if (array_index_out) *array_index_out = array_index; nir_deref_instr *parent = nir_deref_instr_parent(d); if (glsl_type_is_array_or_matrix(parent->type)) { return glsl_get_length(parent->type); } else { assert(glsl_type_is_vector(parent->type)); return glsl_get_vector_elements(parent->type); } } return 0; } static bool guess_loop_limit(loop_info_state *state, nir_const_value *limit_val, nir_scalar basic_ind) { unsigned min_array_size = 0; nir_foreach_block_in_cf_node(block, &state->loop->cf_node) { nir_foreach_instr(instr, block) { if (instr->type != nir_instr_type_intrinsic) continue; nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr); /* Check for arrays variably-indexed by a loop induction variable. */ if (intrin->intrinsic == nir_intrinsic_load_deref || intrin->intrinsic == nir_intrinsic_store_deref || intrin->intrinsic == nir_intrinsic_copy_deref) { nir_loop_variable *array_idx = NULL; unsigned array_size = find_array_access_via_induction(state, nir_src_as_deref(intrin->src[0]), &array_idx); if (array_idx && basic_ind.def == array_idx->def && (min_array_size == 0 || min_array_size > array_size)) { /* Array indices are scalars */ assert(basic_ind.def->num_components == 1); min_array_size = array_size; } if (intrin->intrinsic != nir_intrinsic_copy_deref) continue; array_size = find_array_access_via_induction(state, nir_src_as_deref(intrin->src[1]), &array_idx); if (array_idx && basic_ind.def == array_idx->def && (min_array_size == 0 || min_array_size > array_size)) { /* Array indices are scalars */ assert(basic_ind.def->num_components == 1); min_array_size = array_size; } } } } if (min_array_size) { *limit_val = nir_const_value_for_uint(min_array_size, basic_ind.def->bit_size); return true; } return false; } static nir_op invert_comparison_if_needed(nir_op alu_op, bool invert); /* Returns whether "limit_op(a, b) alu_op c" is equivalent to "(a alu_op c) || (b alu_op c)". */ static bool is_minmax_compatible(nir_op limit_op, nir_op alu_op, bool limit_rhs, bool invert_cond) { bool is_max; switch (limit_op) { case nir_op_imin: case nir_op_fmin: case nir_op_umin: is_max = false; break; case nir_op_imax: case nir_op_fmax: case nir_op_umax: is_max = true; break; default: return false; } if (nir_op_infos[limit_op].input_types[0] != nir_op_infos[alu_op].input_types[0]) return false; /* Comparisons we can split are: * - min(a, b) < c * - c < max(a, b) * - max(a, b) >= c * - c >= min(a, b) */ switch (invert_comparison_if_needed(alu_op, invert_cond)) { case nir_op_ilt: case nir_op_flt: case nir_op_ult: return (!limit_rhs && !is_max) || (limit_rhs && is_max); case nir_op_ige: case nir_op_fge: case nir_op_uge: return (!limit_rhs && is_max) || (limit_rhs && !is_max); default: return false; } } static bool try_find_limit_of_alu(nir_scalar limit, nir_const_value *limit_val, nir_op alu_op, bool invert_cond, nir_loop_terminator *terminator, loop_info_state *state) { if (!nir_scalar_is_alu(limit)) return false; nir_op limit_op = nir_scalar_alu_op(limit); if (is_minmax_compatible(limit_op, alu_op, !terminator->induction_rhs, invert_cond)) { for (unsigned i = 0; i < 2; i++) { nir_scalar src = nir_scalar_chase_alu_src(limit, i); if (nir_scalar_is_const(src)) { *limit_val = nir_scalar_as_const_value(src); terminator->exact_trip_count_unknown = true; return true; } } } return false; } static nir_const_value eval_const_unop(nir_op op, unsigned bit_size, nir_const_value src0, unsigned execution_mode) { assert(nir_op_infos[op].num_inputs == 1); nir_const_value dest; nir_const_value *src[1] = { &src0 }; nir_eval_const_opcode(op, &dest, 1, bit_size, src, execution_mode); return dest; } static nir_const_value eval_const_binop(nir_op op, unsigned bit_size, nir_const_value src0, nir_const_value src1, unsigned execution_mode) { assert(nir_op_infos[op].num_inputs == 2); nir_const_value dest; nir_const_value *src[2] = { &src0, &src1 }; nir_eval_const_opcode(op, &dest, 1, bit_size, src, execution_mode); return dest; } static int find_replacement(const nir_scalar *originals, nir_scalar key, unsigned num_replacements) { for (int i = 0; i < num_replacements; i++) { if (nir_scalar_equal(originals[i], key)) return i; } return -1; } /** * Try to evaluate an ALU instruction as a constant with a replacement * * Much like \c nir_opt_constant_folding.c:try_fold_alu, this method attempts * to evaluate an ALU instruction as a constant. There are two significant * differences. * * First, this method performs the evaluation recursively. If any source of * the ALU instruction is not itself a constant, it is first evaluated. * * Second, if the SSA value \c original is encountered as a source of the ALU * instruction, the value \c replacement is substituted. * * The intended purpose of this function is to evaluate an arbitrary * expression involving a loop induction variable. In this case, \c original * would be the phi node associated with the induction variable, and * \c replacement is the initial value of the induction variable. * * \returns true if the ALU instruction can be evaluated as constant (after * applying the previously described substitution) or false otherwise. */ static bool try_eval_const_alu(nir_const_value *dest, nir_scalar alu_s, const nir_scalar *originals, const nir_const_value *replacements, unsigned num_replacements, unsigned execution_mode) { nir_alu_instr *alu = nir_instr_as_alu(alu_s.def->parent_instr); if (nir_op_infos[alu->op].output_size) return false; /* In the case that any outputs/inputs have unsized types, then we need to * guess the bit-size. In this case, the validator ensures that all * bit-sizes match so we can just take the bit-size from first * output/input with an unsized type. If all the outputs/inputs are sized * then we don't need to guess the bit-size at all because the code we * generate for constant opcodes in this case already knows the sizes of * the types involved and does not need the provided bit-size for anything * (although it still requires to receive a valid bit-size). */ unsigned bit_size = 0; if (!nir_alu_type_get_type_size(nir_op_infos[alu->op].output_type)) { bit_size = alu->def.bit_size; } else { for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++) { if (!nir_alu_type_get_type_size(nir_op_infos[alu->op].input_types[i])) bit_size = alu->src[i].src.ssa->bit_size; } if (bit_size == 0) bit_size = 32; } nir_const_value src[NIR_MAX_VEC_COMPONENTS]; nir_const_value *src_ptrs[NIR_MAX_VEC_COMPONENTS]; for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++) { nir_scalar src_s = nir_scalar_chase_alu_src(alu_s, i); src_ptrs[i] = &src[i]; if (nir_scalar_is_const(src_s)) { src[i] = nir_scalar_as_const_value(src_s); continue; } int r = find_replacement(originals, src_s, num_replacements); if (r >= 0) { src[i] = replacements[r]; } else if (!nir_scalar_is_alu(src_s) || !try_eval_const_alu(&src[i], src_s, originals, replacements, num_replacements, execution_mode)) { return false; } } nir_eval_const_opcode(alu->op, dest, 1, bit_size, src_ptrs, execution_mode); return true; } static nir_op invert_comparison_if_needed(nir_op alu_op, bool invert) { if (!invert) return alu_op; switch (alu_op) { case nir_op_fge: return nir_op_flt; case nir_op_ige: return nir_op_ilt; case nir_op_uge: return nir_op_ult; case nir_op_flt: return nir_op_fge; case nir_op_ilt: return nir_op_ige; case nir_op_ult: return nir_op_uge; case nir_op_feq: return nir_op_fneu; case nir_op_ieq: return nir_op_ine; case nir_op_fneu: return nir_op_feq; case nir_op_ine: return nir_op_ieq; default: unreachable("Unsuported comparison!"); } } static int32_t get_iteration(nir_op cond_op, nir_const_value initial, nir_const_value step, nir_const_value limit, bool invert_cond, unsigned bit_size, unsigned execution_mode) { nir_const_value span, iter; unsigned iter_bit_size = bit_size; switch (invert_comparison_if_needed(cond_op, invert_cond)) { case nir_op_ine: /* In order for execution to be here, limit must be the same as initial. * Otherwise will_break_on_first_iteration would have returned false. * If step is zero, the loop is infinite. Otherwise the loop will * execute once. */ return step.u64 == 0 ? -1 : 1; case nir_op_ige: case nir_op_ilt: case nir_op_ieq: span = eval_const_binop(nir_op_isub, bit_size, limit, initial, execution_mode); iter = eval_const_binop(nir_op_idiv, bit_size, span, step, execution_mode); break; case nir_op_uge: case nir_op_ult: span = eval_const_binop(nir_op_isub, bit_size, limit, initial, execution_mode); iter = eval_const_binop(nir_op_udiv, bit_size, span, step, execution_mode); break; case nir_op_fneu: /* In order for execution to be here, limit must be the same as initial. * Otherwise will_break_on_first_iteration would have returned false. * If step is zero, the loop is infinite. Otherwise the loop will * execute once. * * This is a little more tricky for floating point since X-Y might still * be X even if Y is not zero. Instead check that (initial + step) != * initial. */ span = eval_const_binop(nir_op_fadd, bit_size, initial, step, execution_mode); iter = eval_const_binop(nir_op_feq, bit_size, initial, span, execution_mode); /* return (initial + step) == initial ? -1 : 1 */ return iter.b ? -1 : 1; case nir_op_fge: case nir_op_flt: case nir_op_feq: span = eval_const_binop(nir_op_fsub, bit_size, limit, initial, execution_mode); iter = eval_const_binop(nir_op_fdiv, bit_size, span, step, execution_mode); iter = eval_const_unop(nir_op_f2i64, bit_size, iter, execution_mode); iter_bit_size = 64; break; default: return -1; } uint64_t iter_u64 = nir_const_value_as_uint(iter, iter_bit_size); return iter_u64 > u_intN_max(iter_bit_size) ? -1 : (int)iter_u64; } static int32_t get_iteration_empirical(nir_scalar cond, nir_alu_instr *incr_alu, nir_scalar basis, nir_const_value initial, nir_scalar limit_basis, nir_const_value limit, bool invert_cond, unsigned execution_mode, unsigned max_unroll_iterations) { int iter_count = 0; nir_const_value result; const nir_scalar incr = nir_get_scalar(&incr_alu->def, basis.comp); const nir_scalar original[] = {basis, limit_basis}; nir_const_value replacement[] = {initial, limit}; while (iter_count <= max_unroll_iterations) { bool success; success = try_eval_const_alu(&result, cond, original, replacement, 2, execution_mode); if (!success) return -1; const bool cond_succ = invert_cond ? !result.b : result.b; if (cond_succ) return iter_count; iter_count++; success = try_eval_const_alu(&result, incr, original, replacement, 2, execution_mode); assert(success); replacement[0] = result; } return -1; } static bool will_break_on_first_iteration(nir_scalar cond, nir_scalar basis, nir_scalar limit_basis, nir_const_value initial, nir_const_value limit, bool invert_cond, unsigned execution_mode) { nir_const_value result; const nir_scalar originals[2] = { basis, limit_basis }; const nir_const_value replacements[2] = { initial, limit }; ASSERTED bool success = try_eval_const_alu(&result, cond, originals, replacements, 2, execution_mode); assert(success); return invert_cond ? !result.b : result.b; } static bool test_iterations(int32_t iter_int, nir_const_value step, nir_const_value limit, nir_op cond_op, unsigned bit_size, nir_alu_type induction_base_type, nir_const_value initial, bool limit_rhs, bool invert_cond, unsigned execution_mode) { assert(nir_op_infos[cond_op].num_inputs == 2); nir_const_value iter_src; nir_op mul_op; nir_op add_op; switch (induction_base_type) { case nir_type_float: iter_src = nir_const_value_for_float(iter_int, bit_size); mul_op = nir_op_fmul; add_op = nir_op_fadd; break; case nir_type_int: case nir_type_uint: iter_src = nir_const_value_for_int(iter_int, bit_size); mul_op = nir_op_imul; add_op = nir_op_iadd; break; default: unreachable("Unhandled induction variable base type!"); } /* Multiple the iteration count we are testing by the number of times we * step the induction variable each iteration. */ nir_const_value mul_result = eval_const_binop(mul_op, bit_size, iter_src, step, execution_mode); /* Add the initial value to the accumulated induction variable total */ nir_const_value add_result = eval_const_binop(add_op, bit_size, mul_result, initial, execution_mode); nir_const_value *src[2]; src[limit_rhs ? 0 : 1] = &add_result; src[limit_rhs ? 1 : 0] = &limit; /* Evaluate the loop exit condition */ nir_const_value result; nir_eval_const_opcode(cond_op, &result, 1, bit_size, src, execution_mode); return invert_cond ? !result.b : result.b; } static int calculate_iterations(nir_scalar basis, nir_scalar limit_basis, nir_const_value initial, nir_const_value step, nir_const_value limit, nir_alu_instr *alu, nir_scalar cond, nir_op alu_op, bool limit_rhs, bool invert_cond, unsigned execution_mode, unsigned max_unroll_iterations) { /* nir_op_isub should have been lowered away by this point */ assert(alu->op != nir_op_isub); /* Make sure the alu type for our induction variable is compatible with the * conditional alus input type. If its not something has gone really wrong. */ nir_alu_type induction_base_type = nir_alu_type_get_base_type(nir_op_infos[alu->op].output_type); if (induction_base_type == nir_type_int || induction_base_type == nir_type_uint) { assert(nir_alu_type_get_base_type(nir_op_infos[alu_op].input_types[1]) == nir_type_int || nir_alu_type_get_base_type(nir_op_infos[alu_op].input_types[1]) == nir_type_uint); } else { assert(nir_alu_type_get_base_type(nir_op_infos[alu_op].input_types[0]) == induction_base_type); } /* do-while loops can increment the starting value before the condition is * checked. e.g. * * do { * ndx++; * } while (ndx < 3); * * Here we check if the induction variable is used directly by the loop * condition and if so we assume we need to step the initial value. */ unsigned trip_offset = 0; nir_alu_instr *cond_alu = nir_instr_as_alu(cond.def->parent_instr); if (cond_alu->src[0].src.ssa == &alu->def || cond_alu->src[1].src.ssa == &alu->def) { trip_offset = 1; } unsigned bit_size = nir_src_bit_size(alu->src[0].src); /* get_iteration works under assumption that iterator will be * incremented or decremented until it hits the limit, * however if the loop condition is false on the first iteration * get_iteration's assumption is broken. Handle such loops first. */ if (will_break_on_first_iteration(cond, basis, limit_basis, initial, limit, invert_cond, execution_mode)) { return 0; } /* For loops incremented with addition operation, it's easy to * calculate the number of iterations theoretically. Even though it * is possible for other operations as well, it is much more error * prone, and doesn't cover all possible cases. So, we try to * emulate the loop. */ int iter_int; switch (alu->op) { case nir_op_iadd: case nir_op_fadd: assert(nir_src_bit_size(alu->src[0].src) == nir_src_bit_size(alu->src[1].src)); iter_int = get_iteration(alu_op, initial, step, limit, invert_cond, bit_size, execution_mode); break; case nir_op_fmul: /* Detecting non-zero loop counts when the loop increment is floating * point multiplication triggers a preexisting problem in * glsl-fs-loop-unroll-mul-fp64.shader_test. See * https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3445#note_1779438. */ return -1; case nir_op_imul: case nir_op_ishl: case nir_op_ishr: case nir_op_ushr: return get_iteration_empirical(cond, alu, basis, initial, limit_basis, limit, invert_cond, execution_mode, max_unroll_iterations); default: unreachable("Invalid induction variable increment operation."); } /* If iter_int is negative the loop is ill-formed or is the conditional is * unsigned with a huge iteration count so don't bother going any further. */ if (iter_int < 0) return -1; nir_op actual_alu_op = invert_comparison_if_needed(alu_op, invert_cond); if (actual_alu_op == nir_op_ine || actual_alu_op == nir_op_fneu) return iter_int; /* An explanation from the GLSL unrolling pass: * * Make sure that the calculated number of iterations satisfies the exit * condition. This is needed to catch off-by-one errors and some types of * ill-formed loops. For example, we need to detect that the following * loop does not have a maximum iteration count. * * for (float x = 0.0; x != 0.9; x += 0.2); */ for (int bias = -1; bias <= 1; bias++) { const int iter_bias = iter_int + bias; if (iter_bias < 1) continue; if (test_iterations(iter_bias, step, limit, alu_op, bit_size, induction_base_type, initial, limit_rhs, invert_cond, execution_mode)) { return iter_bias - trip_offset; } } return -1; } static bool get_induction_and_limit_vars(nir_scalar cond, nir_scalar *ind, nir_scalar *limit, bool *limit_rhs, loop_info_state *state) { nir_scalar rhs, lhs; lhs = nir_scalar_chase_alu_src(cond, 0); rhs = nir_scalar_chase_alu_src(cond, 1); nir_loop_variable *src0_lv = get_loop_var(lhs.def, state); nir_loop_variable *src1_lv = get_loop_var(rhs.def, state); if (src0_lv->type == basic_induction) { if (!nir_src_is_const(*src0_lv->init_src)) return false; *ind = lhs; *limit = rhs; *limit_rhs = true; return true; } else if (src1_lv->type == basic_induction) { if (!nir_src_is_const(*src1_lv->init_src)) return false; *ind = rhs; *limit = lhs; *limit_rhs = false; return true; } else { return false; } } static bool try_find_trip_count_vars_in_logical_op(nir_scalar *cond, nir_scalar *ind, nir_scalar *limit, bool *limit_rhs, loop_info_state *state) { const nir_op alu_op = nir_scalar_alu_op(*cond); bool exit_loop_on_false = alu_op == nir_op_ieq || alu_op == nir_op_inot; nir_scalar logical_op = exit_loop_on_false ? nir_scalar_chase_alu_src(*cond, 0) : *cond; if (alu_op == nir_op_ieq) { nir_scalar zero = nir_scalar_chase_alu_src(*cond, 1); if (!nir_scalar_is_alu(logical_op) || !nir_scalar_is_const(zero)) { /* Maybe we had it the wrong way, flip things around */ nir_scalar tmp = zero; zero = logical_op; logical_op = tmp; /* If we still didn't find what we need then return */ if (!nir_scalar_is_const(zero)) return false; } /* If the loop is not breaking on (x && y) == 0 then return */ if (nir_scalar_as_uint(zero) != 0) return false; } if (!nir_scalar_is_alu(logical_op)) return false; if ((exit_loop_on_false && (nir_scalar_alu_op(logical_op) != nir_op_iand)) || (!exit_loop_on_false && (nir_scalar_alu_op(logical_op) != nir_op_ior))) return false; /* Check if iand src is a terminator condition and try get induction var * and trip limit var. */ bool found_induction_var = false; for (unsigned i = 0; i < 2; i++) { nir_scalar src = nir_scalar_chase_alu_src(logical_op, i); if (nir_is_terminator_condition_with_two_inputs(src) && get_induction_and_limit_vars(src, ind, limit, limit_rhs, state)) { *cond = src; found_induction_var = true; /* If we've found one with a constant limit, stop. */ if (nir_scalar_is_const(*limit)) return true; } } return found_induction_var; } /* Run through each of the terminators of the loop and try to infer a possible * trip-count. We need to check them all, and set the lowest trip-count as the * trip-count of our loop. If one of the terminators has an undecidable * trip-count we can not safely assume anything about the duration of the * loop. */ static void find_trip_count(loop_info_state *state, unsigned execution_mode, unsigned max_unroll_iterations) { bool trip_count_known = true; bool guessed_trip_count = false; nir_loop_terminator *limiting_terminator = NULL; int max_trip_count = -1; list_for_each_entry(nir_loop_terminator, terminator, &state->loop->info->loop_terminator_list, loop_terminator_link) { nir_scalar cond = { terminator->nif->condition.ssa, 0 }; if (!nir_scalar_is_alu(cond)) { /* If we get here the loop is dead and will get cleaned up by the * nir_opt_dead_cf pass. */ trip_count_known = false; terminator->exact_trip_count_unknown = true; continue; } nir_op alu_op = nir_scalar_alu_op(cond); bool invert_cond = terminator->continue_from_then; bool limit_rhs; nir_scalar basic_ind = { NULL, 0 }; nir_scalar limit; if ((alu_op == nir_op_inot || alu_op == nir_op_ieq || alu_op == nir_op_ior) && try_find_trip_count_vars_in_logical_op(&cond, &basic_ind, &limit, &limit_rhs, state)) { /* The loop is exiting on (x && y) == 0 so we need to get the * inverse of x or y (i.e. which ever contained the induction var) in * order to compute the trip count. */ if (alu_op == nir_op_inot || alu_op == nir_op_ieq) invert_cond = !invert_cond; alu_op = nir_scalar_alu_op(cond); trip_count_known = false; terminator->conditional_instr = cond.def->parent_instr; terminator->exact_trip_count_unknown = true; } if (!basic_ind.def) { if (nir_is_supported_terminator_condition(cond)) { /* Extract and inverse the comparision if it is wrapped in an inot */ if (alu_op == nir_op_inot) { cond = nir_scalar_chase_alu_src(cond, 0); alu_op = nir_scalar_alu_op(cond); invert_cond = !invert_cond; } get_induction_and_limit_vars(cond, &basic_ind, &limit, &limit_rhs, state); } } /* The comparison has to have a basic induction variable for us to be * able to find trip counts. */ if (!basic_ind.def) { trip_count_known = false; terminator->exact_trip_count_unknown = true; continue; } terminator->induction_rhs = !limit_rhs; /* Attempt to find a constant limit for the loop */ nir_const_value limit_val; if (nir_scalar_is_const(limit)) { limit_val = nir_scalar_as_const_value(limit); } else { trip_count_known = false; if (!try_find_limit_of_alu(limit, &limit_val, alu_op, invert_cond, terminator, state)) { /* Guess loop limit based on array access */ if (!guess_loop_limit(state, &limit_val, basic_ind)) { terminator->exact_trip_count_unknown = true; continue; } guessed_trip_count = true; } } /* We have determined that we have the following constants: * (With the typical int i = 0; i < x; i++; as an example) * - Upper limit. * - Starting value * - Step / iteration size * Thats all thats needed to calculate the trip-count */ nir_loop_variable *lv = get_loop_var(basic_ind.def, state); /* The basic induction var might be a vector but, because we guarantee * earlier that the phi source has a scalar swizzle, we can take the * component from basic_ind. */ nir_scalar initial_s = { lv->init_src->ssa, basic_ind.comp }; nir_scalar alu_s = { lv->update_src->src.ssa, lv->update_src->swizzle[basic_ind.comp] }; /* We are not guaranteed by that at one of these sources is a constant. * Try to find one. */ if (!nir_scalar_is_const(initial_s) || !nir_scalar_is_const(alu_s)) continue; nir_const_value initial_val = nir_scalar_as_const_value(initial_s); nir_const_value step_val = nir_scalar_as_const_value(alu_s); int iterations = calculate_iterations(nir_get_scalar(lv->basis, basic_ind.comp), limit, initial_val, step_val, limit_val, nir_instr_as_alu(nir_src_parent_instr(&lv->update_src->src)), cond, alu_op, limit_rhs, invert_cond, execution_mode, max_unroll_iterations); /* Where we not able to calculate the iteration count */ if (iterations == -1) { trip_count_known = false; guessed_trip_count = false; terminator->exact_trip_count_unknown = true; continue; } if (guessed_trip_count) { guessed_trip_count = false; terminator->exact_trip_count_unknown = true; if (state->loop->info->guessed_trip_count == 0 || state->loop->info->guessed_trip_count > iterations) state->loop->info->guessed_trip_count = iterations; continue; } /* If this is the first run or we have found a smaller amount of * iterations than previously (we have identified a more limiting * terminator) set the trip count and limiting terminator. */ if (max_trip_count == -1 || iterations < max_trip_count) { max_trip_count = iterations; limiting_terminator = terminator; } } state->loop->info->exact_trip_count_known = trip_count_known; if (max_trip_count > -1) state->loop->info->max_trip_count = max_trip_count; state->loop->info->limiting_terminator = limiting_terminator; } static bool force_unroll_array_access(loop_info_state *state, nir_deref_instr *deref, bool contains_sampler) { unsigned array_size = find_array_access_via_induction(state, deref, NULL); if (array_size) { if ((array_size == state->loop->info->max_trip_count) && nir_deref_mode_must_be(deref, nir_var_shader_in | nir_var_shader_out | nir_var_shader_temp | nir_var_function_temp)) return true; if (nir_deref_mode_must_be(deref, state->indirect_mask)) return true; if (contains_sampler && state->force_unroll_sampler_indirect) return true; } return false; } static bool force_unroll_heuristics(loop_info_state *state, nir_block *block) { nir_foreach_instr(instr, block) { if (instr->type == nir_instr_type_tex) { nir_tex_instr *tex_instr = nir_instr_as_tex(instr); int sampler_idx = nir_tex_instr_src_index(tex_instr, nir_tex_src_sampler_deref); if (sampler_idx >= 0) { nir_deref_instr *deref = nir_instr_as_deref(tex_instr->src[sampler_idx].src.ssa->parent_instr); if (force_unroll_array_access(state, deref, true)) return true; } } if (instr->type != nir_instr_type_intrinsic) continue; nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr); /* Check for arrays variably-indexed by a loop induction variable. * Unrolling the loop may convert that access into constant-indexing. */ if (intrin->intrinsic == nir_intrinsic_load_deref || intrin->intrinsic == nir_intrinsic_store_deref || intrin->intrinsic == nir_intrinsic_copy_deref) { if (force_unroll_array_access(state, nir_src_as_deref(intrin->src[0]), false)) return true; if (intrin->intrinsic == nir_intrinsic_copy_deref && force_unroll_array_access(state, nir_src_as_deref(intrin->src[1]), false)) return true; } } return false; } static void get_loop_info(loop_info_state *state, nir_function_impl *impl) { nir_shader *shader = impl->function->shader; const nir_shader_compiler_options *options = shader->options; /* Add all entries in the outermost part of the loop to the processing list * Mark the entries in conditionals or in nested loops accordingly */ foreach_list_typed_safe(nir_cf_node, node, node, &state->loop->body) { switch (node->type) { case nir_cf_node_block: init_loop_block(nir_cf_node_as_block(node), state, false, false); break; case nir_cf_node_if: nir_foreach_block_in_cf_node(block, node) init_loop_block(block, state, true, false); break; case nir_cf_node_loop: nir_foreach_block_in_cf_node(block, node) { init_loop_block(block, state, false, true); } break; case nir_cf_node_function: break; } } /* Try to find all simple terminators of the loop. If we can't find any, * or we find possible terminators that have side effects then bail. */ if (!find_loop_terminators(state)) { list_for_each_entry_safe(nir_loop_terminator, terminator, &state->loop->info->loop_terminator_list, loop_terminator_link) { list_del(&terminator->loop_terminator_link); ralloc_free(terminator); } return; } if (!compute_induction_information(state)) return; /* Run through each of the terminators and try to compute a trip-count */ find_trip_count(state, impl->function->shader->info.float_controls_execution_mode, impl->function->shader->options->max_unroll_iterations); nir_foreach_block_in_cf_node(block, &state->loop->cf_node) { nir_foreach_instr(instr, block) { state->loop->info->instr_cost += instr_cost(state, instr, options); } if (state->loop->info->force_unroll) continue; if (force_unroll_heuristics(state, block)) { state->loop->info->force_unroll = true; } } } static loop_info_state * initialize_loop_info_state(nir_loop *loop, void *mem_ctx, nir_function_impl *impl) { loop_info_state *state = rzalloc(mem_ctx, loop_info_state); state->loop_vars = ralloc_array(mem_ctx, nir_loop_variable, impl->ssa_alloc); state->loop_vars_init = rzalloc_array(mem_ctx, BITSET_WORD, BITSET_WORDS(impl->ssa_alloc)); state->loop = loop; list_inithead(&state->process_list); if (loop->info) ralloc_free(loop->info); loop->info = rzalloc(loop, nir_loop_info); list_inithead(&loop->info->loop_terminator_list); return state; } static void process_loops(nir_cf_node *cf_node, nir_variable_mode indirect_mask, bool force_unroll_sampler_indirect) { switch (cf_node->type) { case nir_cf_node_block: return; case nir_cf_node_if: { nir_if *if_stmt = nir_cf_node_as_if(cf_node); foreach_list_typed(nir_cf_node, nested_node, node, &if_stmt->then_list) process_loops(nested_node, indirect_mask, force_unroll_sampler_indirect); foreach_list_typed(nir_cf_node, nested_node, node, &if_stmt->else_list) process_loops(nested_node, indirect_mask, force_unroll_sampler_indirect); return; } case nir_cf_node_loop: { nir_loop *loop = nir_cf_node_as_loop(cf_node); assert(!nir_loop_has_continue_construct(loop)); foreach_list_typed(nir_cf_node, nested_node, node, &loop->body) process_loops(nested_node, indirect_mask, force_unroll_sampler_indirect); break; } default: unreachable("unknown cf node type"); } nir_loop *loop = nir_cf_node_as_loop(cf_node); nir_function_impl *impl = nir_cf_node_get_function(cf_node); void *mem_ctx = ralloc_context(NULL); loop_info_state *state = initialize_loop_info_state(loop, mem_ctx, impl); state->indirect_mask = indirect_mask; state->force_unroll_sampler_indirect = force_unroll_sampler_indirect; get_loop_info(state, impl); ralloc_free(mem_ctx); } void nir_loop_analyze_impl(nir_function_impl *impl, nir_variable_mode indirect_mask, bool force_unroll_sampler_indirect) { nir_index_ssa_defs(impl); foreach_list_typed(nir_cf_node, node, node, &impl->body) process_loops(node, indirect_mask, force_unroll_sampler_indirect); }