/* Transformation Utilities for Loop Vectorization. Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009 Free Software Foundation, Inc. Contributed by Dorit Naishlos This file is part of GCC. GCC is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3, or (at your option) any later version. GCC is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with GCC; see the file COPYING3. If not see . */ #include "config.h" #include "system.h" #include "coretypes.h" #include "tm.h" #include "ggc.h" #include "tree.h" #include "target.h" #include "rtl.h" #include "basic-block.h" #include "diagnostic.h" #include "tree-flow.h" #include "tree-dump.h" #include "timevar.h" #include "cfgloop.h" #include "expr.h" #include "optabs.h" #include "params.h" #include "recog.h" #include "tree-data-ref.h" #include "tree-chrec.h" #include "tree-scalar-evolution.h" #include "tree-vectorizer.h" #include "langhooks.h" #include "tree-pass.h" #include "toplev.h" #include "real.h" /* Utility functions for the code transformation. */ static bool vect_transform_stmt (gimple, gimple_stmt_iterator *, bool *, slp_tree, slp_instance); static tree vect_create_destination_var (tree, tree); static tree vect_create_data_ref_ptr (gimple, struct loop*, tree, tree *, gimple *, bool, bool *, tree); static tree vect_create_addr_base_for_vector_ref (gimple, gimple_seq *, tree, struct loop *); static tree vect_get_new_vect_var (tree, enum vect_var_kind, const char *); static tree vect_get_vec_def_for_operand (tree, gimple, tree *); static tree vect_init_vector (gimple, tree, tree, gimple_stmt_iterator *); static void vect_finish_stmt_generation (gimple stmt, gimple vec_stmt, gimple_stmt_iterator *); static bool vect_is_simple_cond (tree, loop_vec_info); static void vect_create_epilog_for_reduction (tree, gimple, int, enum tree_code, gimple); static tree get_initial_def_for_reduction (gimple, tree, tree *); /* Utility function dealing with loop peeling (not peeling itself). */ static void vect_generate_tmps_on_preheader (loop_vec_info, tree *, tree *, tree *); static tree vect_build_loop_niters (loop_vec_info); static void vect_update_ivs_after_vectorizer (loop_vec_info, tree, edge); static tree vect_gen_niters_for_prolog_loop (loop_vec_info, tree); static void vect_update_init_of_dr (struct data_reference *, tree niters); static void vect_update_inits_of_drs (loop_vec_info, tree); static int vect_min_worthwhile_factor (enum tree_code); static int cost_for_stmt (gimple stmt) { stmt_vec_info stmt_info = vinfo_for_stmt (stmt); switch (STMT_VINFO_TYPE (stmt_info)) { case load_vec_info_type: return TARG_SCALAR_LOAD_COST; case store_vec_info_type: return TARG_SCALAR_STORE_COST; case op_vec_info_type: case condition_vec_info_type: case assignment_vec_info_type: case reduc_vec_info_type: case induc_vec_info_type: case type_promotion_vec_info_type: case type_demotion_vec_info_type: case type_conversion_vec_info_type: case call_vec_info_type: return TARG_SCALAR_STMT_COST; case undef_vec_info_type: default: gcc_unreachable (); } } /* Function vect_estimate_min_profitable_iters Return the number of iterations required for the vector version of the loop to be profitable relative to the cost of the scalar version of the loop. TODO: Take profile info into account before making vectorization decisions, if available. */ int vect_estimate_min_profitable_iters (loop_vec_info loop_vinfo) { int i; int min_profitable_iters; int peel_iters_prologue; int peel_iters_epilogue; int vec_inside_cost = 0; int vec_outside_cost = 0; int scalar_single_iter_cost = 0; int scalar_outside_cost = 0; int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo); int nbbs = loop->num_nodes; int byte_misalign = LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo); int peel_guard_costs = 0; int innerloop_iters = 0, factor; VEC (slp_instance, heap) *slp_instances; slp_instance instance; /* Cost model disabled. */ if (!flag_vect_cost_model) { if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "cost model disabled."); return 0; } /* Requires loop versioning tests to handle misalignment. */ if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))) { /* FIXME: Make cost depend on complexity of individual check. */ vec_outside_cost += VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)); if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "cost model: Adding cost of checks for loop " "versioning to treat misalignment.\n"); } if (VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo))) { /* FIXME: Make cost depend on complexity of individual check. */ vec_outside_cost += VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo)); if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "cost model: Adding cost of checks for loop " "versioning aliasing.\n"); } if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)) || VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo))) { vec_outside_cost += TARG_COND_TAKEN_BRANCH_COST; } /* Count statements in scalar loop. Using this as scalar cost for a single iteration for now. TODO: Add outer loop support. TODO: Consider assigning different costs to different scalar statements. */ /* FORNOW. */ if (loop->inner) innerloop_iters = 50; /* FIXME */ for (i = 0; i < nbbs; i++) { gimple_stmt_iterator si; basic_block bb = bbs[i]; if (bb->loop_father == loop->inner) factor = innerloop_iters; else factor = 1; for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si)) { gimple stmt = gsi_stmt (si); stmt_vec_info stmt_info = vinfo_for_stmt (stmt); /* Skip stmts that are not vectorized inside the loop. */ if (!STMT_VINFO_RELEVANT_P (stmt_info) && (!STMT_VINFO_LIVE_P (stmt_info) || STMT_VINFO_DEF_TYPE (stmt_info) != vect_reduction_def)) continue; scalar_single_iter_cost += cost_for_stmt (stmt) * factor; vec_inside_cost += STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) * factor; /* FIXME: for stmts in the inner-loop in outer-loop vectorization, some of the "outside" costs are generated inside the outer-loop. */ vec_outside_cost += STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info); } } /* Add additional cost for the peeled instructions in prologue and epilogue loop. FORNOW: If we don't know the value of peel_iters for prologue or epilogue at compile-time - we assume it's vf/2 (the worst would be vf-1). TODO: Build an expression that represents peel_iters for prologue and epilogue to be used in a run-time test. */ if (byte_misalign < 0) { peel_iters_prologue = vf/2; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "cost model: " "prologue peel iters set to vf/2."); /* If peeling for alignment is unknown, loop bound of main loop becomes unknown. */ peel_iters_epilogue = vf/2; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "cost model: " "epilogue peel iters set to vf/2 because " "peeling for alignment is unknown ."); /* If peeled iterations are unknown, count a taken branch and a not taken branch per peeled loop. Even if scalar loop iterations are known, vector iterations are not known since peeled prologue iterations are not known. Hence guards remain the same. */ peel_guard_costs += 2 * (TARG_COND_TAKEN_BRANCH_COST + TARG_COND_NOT_TAKEN_BRANCH_COST); } else { if (byte_misalign) { struct data_reference *dr = LOOP_VINFO_UNALIGNED_DR (loop_vinfo); int element_size = GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (dr)))); tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (DR_STMT (dr))); int nelements = TYPE_VECTOR_SUBPARTS (vectype); peel_iters_prologue = nelements - (byte_misalign / element_size); } else peel_iters_prologue = 0; if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo)) { peel_iters_epilogue = vf/2; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "cost model: " "epilogue peel iters set to vf/2 because " "loop iterations are unknown ."); /* If peeled iterations are known but number of scalar loop iterations are unknown, count a taken branch per peeled loop. */ peel_guard_costs += 2 * TARG_COND_TAKEN_BRANCH_COST; } else { int niters = LOOP_VINFO_INT_NITERS (loop_vinfo); peel_iters_prologue = niters < peel_iters_prologue ? niters : peel_iters_prologue; peel_iters_epilogue = (niters - peel_iters_prologue) % vf; } } vec_outside_cost += (peel_iters_prologue * scalar_single_iter_cost) + (peel_iters_epilogue * scalar_single_iter_cost) + peel_guard_costs; /* FORNOW: The scalar outside cost is incremented in one of the following ways: 1. The vectorizer checks for alignment and aliasing and generates a condition that allows dynamic vectorization. A cost model check is ANDED with the versioning condition. Hence scalar code path now has the added cost of the versioning check. if (cost > th & versioning_check) jmp to vector code Hence run-time scalar is incremented by not-taken branch cost. 2. The vectorizer then checks if a prologue is required. If the cost model check was not done before during versioning, it has to be done before the prologue check. if (cost <= th) prologue = scalar_iters if (prologue == 0) jmp to vector code else execute prologue if (prologue == num_iters) go to exit Hence the run-time scalar cost is incremented by a taken branch, plus a not-taken branch, plus a taken branch cost. 3. The vectorizer then checks if an epilogue is required. If the cost model check was not done before during prologue check, it has to be done with the epilogue check. if (prologue == 0) jmp to vector code else execute prologue if (prologue == num_iters) go to exit vector code: if ((cost <= th) | (scalar_iters-prologue-epilogue == 0)) jmp to epilogue Hence the run-time scalar cost should be incremented by 2 taken branches. TODO: The back end may reorder the BBS's differently and reverse conditions/branch directions. Change the estimates below to something more reasonable. */ /* If the number of iterations is known and we do not do versioning, we can decide whether to vectorize at compile time. Hence the scalar version do not carry cost model guard costs. */ if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) || VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)) || VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo))) { /* Cost model check occurs at versioning. */ if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)) || VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo))) scalar_outside_cost += TARG_COND_NOT_TAKEN_BRANCH_COST; else { /* Cost model check occurs at prologue generation. */ if (LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo) < 0) scalar_outside_cost += 2 * TARG_COND_TAKEN_BRANCH_COST + TARG_COND_NOT_TAKEN_BRANCH_COST; /* Cost model check occurs at epilogue generation. */ else scalar_outside_cost += 2 * TARG_COND_TAKEN_BRANCH_COST; } } /* Add SLP costs. */ slp_instances = LOOP_VINFO_SLP_INSTANCES (loop_vinfo); for (i = 0; VEC_iterate (slp_instance, slp_instances, i, instance); i++) { vec_outside_cost += SLP_INSTANCE_OUTSIDE_OF_LOOP_COST (instance); vec_inside_cost += SLP_INSTANCE_INSIDE_OF_LOOP_COST (instance); } /* Calculate number of iterations required to make the vector version profitable, relative to the loop bodies only. The following condition must hold true: SIC * niters + SOC > VIC * ((niters-PL_ITERS-EP_ITERS)/VF) + VOC where SIC = scalar iteration cost, VIC = vector iteration cost, VOC = vector outside cost, VF = vectorization factor, PL_ITERS = prologue iterations, EP_ITERS= epilogue iterations SOC = scalar outside cost for run time cost model check. */ if ((scalar_single_iter_cost * vf) > vec_inside_cost) { if (vec_outside_cost <= 0) min_profitable_iters = 1; else { min_profitable_iters = ((vec_outside_cost - scalar_outside_cost) * vf - vec_inside_cost * peel_iters_prologue - vec_inside_cost * peel_iters_epilogue) / ((scalar_single_iter_cost * vf) - vec_inside_cost); if ((scalar_single_iter_cost * vf * min_profitable_iters) <= ((vec_inside_cost * min_profitable_iters) + ((vec_outside_cost - scalar_outside_cost) * vf))) min_profitable_iters++; } } /* vector version will never be profitable. */ else { if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "cost model: vector iteration cost = %d " "is divisible by scalar iteration cost = %d by a factor " "greater than or equal to the vectorization factor = %d .", vec_inside_cost, scalar_single_iter_cost, vf); return -1; } if (vect_print_dump_info (REPORT_COST)) { fprintf (vect_dump, "Cost model analysis: \n"); fprintf (vect_dump, " Vector inside of loop cost: %d\n", vec_inside_cost); fprintf (vect_dump, " Vector outside of loop cost: %d\n", vec_outside_cost); fprintf (vect_dump, " Scalar iteration cost: %d\n", scalar_single_iter_cost); fprintf (vect_dump, " Scalar outside cost: %d\n", scalar_outside_cost); fprintf (vect_dump, " prologue iterations: %d\n", peel_iters_prologue); fprintf (vect_dump, " epilogue iterations: %d\n", peel_iters_epilogue); fprintf (vect_dump, " Calculated minimum iters for profitability: %d\n", min_profitable_iters); } min_profitable_iters = min_profitable_iters < vf ? vf : min_profitable_iters; /* Because the condition we create is: if (niters <= min_profitable_iters) then skip the vectorized loop. */ min_profitable_iters--; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, " Profitability threshold = %d\n", min_profitable_iters); return min_profitable_iters; } /* TODO: Close dependency between vect_model_*_cost and vectorizable_* functions. Design better to avoid maintenance issues. */ /* Function vect_model_reduction_cost. Models cost for a reduction operation, including the vector ops generated within the strip-mine loop, the initial definition before the loop, and the epilogue code that must be generated. */ static bool vect_model_reduction_cost (stmt_vec_info stmt_info, enum tree_code reduc_code, int ncopies) { int outer_cost = 0; enum tree_code code; optab optab; tree vectype; gimple stmt, orig_stmt; tree reduction_op; enum machine_mode mode; loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); /* Cost of reduction op inside loop. */ STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) += ncopies * TARG_VEC_STMT_COST; stmt = STMT_VINFO_STMT (stmt_info); switch (get_gimple_rhs_class (gimple_assign_rhs_code (stmt))) { case GIMPLE_SINGLE_RHS: gcc_assert (TREE_OPERAND_LENGTH (gimple_assign_rhs1 (stmt)) == ternary_op); reduction_op = TREE_OPERAND (gimple_assign_rhs1 (stmt), 2); break; case GIMPLE_UNARY_RHS: reduction_op = gimple_assign_rhs1 (stmt); break; case GIMPLE_BINARY_RHS: reduction_op = gimple_assign_rhs2 (stmt); break; default: gcc_unreachable (); } vectype = get_vectype_for_scalar_type (TREE_TYPE (reduction_op)); if (!vectype) { if (vect_print_dump_info (REPORT_COST)) { fprintf (vect_dump, "unsupported data-type "); print_generic_expr (vect_dump, TREE_TYPE (reduction_op), TDF_SLIM); } return false; } mode = TYPE_MODE (vectype); orig_stmt = STMT_VINFO_RELATED_STMT (stmt_info); if (!orig_stmt) orig_stmt = STMT_VINFO_STMT (stmt_info); code = gimple_assign_rhs_code (orig_stmt); /* Add in cost for initial definition. */ outer_cost += TARG_SCALAR_TO_VEC_COST; /* Determine cost of epilogue code. We have a reduction operator that will reduce the vector in one statement. Also requires scalar extract. */ if (!nested_in_vect_loop_p (loop, orig_stmt)) { if (reduc_code < NUM_TREE_CODES) outer_cost += TARG_VEC_STMT_COST + TARG_VEC_TO_SCALAR_COST; else { int vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1); tree bitsize = TYPE_SIZE (TREE_TYPE (gimple_assign_lhs (orig_stmt))); int element_bitsize = tree_low_cst (bitsize, 1); int nelements = vec_size_in_bits / element_bitsize; optab = optab_for_tree_code (code, vectype, optab_default); /* We have a whole vector shift available. */ if (VECTOR_MODE_P (mode) && optab_handler (optab, mode)->insn_code != CODE_FOR_nothing && optab_handler (vec_shr_optab, mode)->insn_code != CODE_FOR_nothing) /* Final reduction via vector shifts and the reduction operator. Also requires scalar extract. */ outer_cost += ((exact_log2(nelements) * 2) * TARG_VEC_STMT_COST + TARG_VEC_TO_SCALAR_COST); else /* Use extracts and reduction op for final reduction. For N elements, we have N extracts and N-1 reduction ops. */ outer_cost += ((nelements + nelements - 1) * TARG_VEC_STMT_COST); } } STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info) = outer_cost; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_reduction_cost: inside_cost = %d, " "outside_cost = %d .", STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info), STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info)); return true; } /* Function vect_model_induction_cost. Models cost for induction operations. */ static void vect_model_induction_cost (stmt_vec_info stmt_info, int ncopies) { /* loop cost for vec_loop. */ STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) = ncopies * TARG_VEC_STMT_COST; /* prologue cost for vec_init and vec_step. */ STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info) = 2 * TARG_SCALAR_TO_VEC_COST; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_induction_cost: inside_cost = %d, " "outside_cost = %d .", STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info), STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info)); } /* Function vect_model_simple_cost. Models cost for simple operations, i.e. those that only emit ncopies of a single op. Right now, this does not account for multiple insns that could be generated for the single vector op. We will handle that shortly. */ void vect_model_simple_cost (stmt_vec_info stmt_info, int ncopies, enum vect_def_type *dt, slp_tree slp_node) { int i; int inside_cost = 0, outside_cost = 0; /* The SLP costs were already calculated during SLP tree build. */ if (PURE_SLP_STMT (stmt_info)) return; inside_cost = ncopies * TARG_VEC_STMT_COST; /* FORNOW: Assuming maximum 2 args per stmts. */ for (i = 0; i < 2; i++) { if (dt[i] == vect_constant_def || dt[i] == vect_invariant_def) outside_cost += TARG_SCALAR_TO_VEC_COST; } if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_simple_cost: inside_cost = %d, " "outside_cost = %d .", inside_cost, outside_cost); /* Set the costs either in STMT_INFO or SLP_NODE (if exists). */ stmt_vinfo_set_inside_of_loop_cost (stmt_info, slp_node, inside_cost); stmt_vinfo_set_outside_of_loop_cost (stmt_info, slp_node, outside_cost); } /* Function vect_cost_strided_group_size For strided load or store, return the group_size only if it is the first load or store of a group, else return 1. This ensures that group size is only returned once per group. */ static int vect_cost_strided_group_size (stmt_vec_info stmt_info) { gimple first_stmt = DR_GROUP_FIRST_DR (stmt_info); if (first_stmt == STMT_VINFO_STMT (stmt_info)) return DR_GROUP_SIZE (stmt_info); return 1; } /* Function vect_model_store_cost Models cost for stores. In the case of strided accesses, one access has the overhead of the strided access attributed to it. */ void vect_model_store_cost (stmt_vec_info stmt_info, int ncopies, enum vect_def_type dt, slp_tree slp_node) { int group_size; int inside_cost = 0, outside_cost = 0; /* The SLP costs were already calculated during SLP tree build. */ if (PURE_SLP_STMT (stmt_info)) return; if (dt == vect_constant_def || dt == vect_invariant_def) outside_cost = TARG_SCALAR_TO_VEC_COST; /* Strided access? */ if (DR_GROUP_FIRST_DR (stmt_info) && !slp_node) group_size = vect_cost_strided_group_size (stmt_info); /* Not a strided access. */ else group_size = 1; /* Is this an access in a group of stores, which provide strided access? If so, add in the cost of the permutes. */ if (group_size > 1) { /* Uses a high and low interleave operation for each needed permute. */ inside_cost = ncopies * exact_log2(group_size) * group_size * TARG_VEC_STMT_COST; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_store_cost: strided group_size = %d .", group_size); } /* Costs of the stores. */ inside_cost += ncopies * TARG_VEC_STORE_COST; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_store_cost: inside_cost = %d, " "outside_cost = %d .", inside_cost, outside_cost); /* Set the costs either in STMT_INFO or SLP_NODE (if exists). */ stmt_vinfo_set_inside_of_loop_cost (stmt_info, slp_node, inside_cost); stmt_vinfo_set_outside_of_loop_cost (stmt_info, slp_node, outside_cost); } /* Function vect_model_load_cost Models cost for loads. In the case of strided accesses, the last access has the overhead of the strided access attributed to it. Since unaligned accesses are supported for loads, we also account for the costs of the access scheme chosen. */ void vect_model_load_cost (stmt_vec_info stmt_info, int ncopies, slp_tree slp_node) { int group_size; int alignment_support_cheme; gimple first_stmt; struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info), *first_dr; int inside_cost = 0, outside_cost = 0; /* The SLP costs were already calculated during SLP tree build. */ if (PURE_SLP_STMT (stmt_info)) return; /* Strided accesses? */ first_stmt = DR_GROUP_FIRST_DR (stmt_info); if (first_stmt && !slp_node) { group_size = vect_cost_strided_group_size (stmt_info); first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt)); } /* Not a strided access. */ else { group_size = 1; first_dr = dr; } alignment_support_cheme = vect_supportable_dr_alignment (first_dr); /* Is this an access in a group of loads providing strided access? If so, add in the cost of the permutes. */ if (group_size > 1) { /* Uses an even and odd extract operations for each needed permute. */ inside_cost = ncopies * exact_log2(group_size) * group_size * TARG_VEC_STMT_COST; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_load_cost: strided group_size = %d .", group_size); } /* The loads themselves. */ switch (alignment_support_cheme) { case dr_aligned: { inside_cost += ncopies * TARG_VEC_LOAD_COST; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_load_cost: aligned."); break; } case dr_unaligned_supported: { /* Here, we assign an additional cost for the unaligned load. */ inside_cost += ncopies * TARG_VEC_UNALIGNED_LOAD_COST; if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_load_cost: unaligned supported by " "hardware."); break; } case dr_explicit_realign: { inside_cost += ncopies * (2*TARG_VEC_LOAD_COST + TARG_VEC_STMT_COST); /* FIXME: If the misalignment remains fixed across the iterations of the containing loop, the following cost should be added to the outside costs. */ if (targetm.vectorize.builtin_mask_for_load) inside_cost += TARG_VEC_STMT_COST; break; } case dr_explicit_realign_optimized: { if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_load_cost: unaligned software " "pipelined."); /* Unaligned software pipeline has a load of an address, an initial load, and possibly a mask operation to "prime" the loop. However, if this is an access in a group of loads, which provide strided access, then the above cost should only be considered for one access in the group. Inside the loop, there is a load op and a realignment op. */ if ((!DR_GROUP_FIRST_DR (stmt_info)) || group_size > 1 || slp_node) { outside_cost = 2*TARG_VEC_STMT_COST; if (targetm.vectorize.builtin_mask_for_load) outside_cost += TARG_VEC_STMT_COST; } inside_cost += ncopies * (TARG_VEC_LOAD_COST + TARG_VEC_STMT_COST); break; } default: gcc_unreachable (); } if (vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "vect_model_load_cost: inside_cost = %d, " "outside_cost = %d .", inside_cost, outside_cost); /* Set the costs either in STMT_INFO or SLP_NODE (if exists). */ stmt_vinfo_set_inside_of_loop_cost (stmt_info, slp_node, inside_cost); stmt_vinfo_set_outside_of_loop_cost (stmt_info, slp_node, outside_cost); } /* Function vect_get_new_vect_var. Returns a name for a new variable. The current naming scheme appends the prefix "vect_" or "vect_p" (depending on the value of VAR_KIND) to the name of vectorizer generated variables, and appends that to NAME if provided. */ static tree vect_get_new_vect_var (tree type, enum vect_var_kind var_kind, const char *name) { const char *prefix; tree new_vect_var; switch (var_kind) { case vect_simple_var: prefix = "vect_"; break; case vect_scalar_var: prefix = "stmp_"; break; case vect_pointer_var: prefix = "vect_p"; break; default: gcc_unreachable (); } if (name) { char* tmp = concat (prefix, name, NULL); new_vect_var = create_tmp_var (type, tmp); free (tmp); } else new_vect_var = create_tmp_var (type, prefix); /* Mark vector typed variable as a gimple register variable. */ if (TREE_CODE (type) == VECTOR_TYPE) DECL_GIMPLE_REG_P (new_vect_var) = true; return new_vect_var; } /* Function vect_create_addr_base_for_vector_ref. Create an expression that computes the address of the first memory location that will be accessed for a data reference. Input: STMT: The statement containing the data reference. NEW_STMT_LIST: Must be initialized to NULL_TREE or a statement list. OFFSET: Optional. If supplied, it is be added to the initial address. LOOP: Specify relative to which loop-nest should the address be computed. For example, when the dataref is in an inner-loop nested in an outer-loop that is now being vectorized, LOOP can be either the outer-loop, or the inner-loop. The first memory location accessed by the following dataref ('in' points to short): for (i=0; iloop_father; tree data_ref_base = unshare_expr (DR_BASE_ADDRESS (dr)); tree base_name; tree data_ref_base_var; tree vec_stmt; tree addr_base, addr_expr; tree dest; gimple_seq seq = NULL; tree base_offset = unshare_expr (DR_OFFSET (dr)); tree init = unshare_expr (DR_INIT (dr)); tree vect_ptr_type, addr_expr2; tree step = TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dr))); gcc_assert (loop); if (loop != containing_loop) { loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); gcc_assert (nested_in_vect_loop_p (loop, stmt)); data_ref_base = unshare_expr (STMT_VINFO_DR_BASE_ADDRESS (stmt_info)); base_offset = unshare_expr (STMT_VINFO_DR_OFFSET (stmt_info)); init = unshare_expr (STMT_VINFO_DR_INIT (stmt_info)); } /* Create data_ref_base */ base_name = build_fold_indirect_ref (data_ref_base); data_ref_base_var = create_tmp_var (TREE_TYPE (data_ref_base), "batmp"); add_referenced_var (data_ref_base_var); data_ref_base = force_gimple_operand (data_ref_base, &seq, true, data_ref_base_var); gimple_seq_add_seq (new_stmt_list, seq); /* Create base_offset */ base_offset = size_binop (PLUS_EXPR, fold_convert (sizetype, base_offset), fold_convert (sizetype, init)); dest = create_tmp_var (sizetype, "base_off"); add_referenced_var (dest); base_offset = force_gimple_operand (base_offset, &seq, true, dest); gimple_seq_add_seq (new_stmt_list, seq); if (offset) { tree tmp = create_tmp_var (sizetype, "offset"); add_referenced_var (tmp); offset = fold_build2 (MULT_EXPR, sizetype, fold_convert (sizetype, offset), step); base_offset = fold_build2 (PLUS_EXPR, sizetype, base_offset, offset); base_offset = force_gimple_operand (base_offset, &seq, false, tmp); gimple_seq_add_seq (new_stmt_list, seq); } /* base + base_offset */ addr_base = fold_build2 (POINTER_PLUS_EXPR, TREE_TYPE (data_ref_base), data_ref_base, base_offset); vect_ptr_type = build_pointer_type (STMT_VINFO_VECTYPE (stmt_info)); /* addr_expr = addr_base */ addr_expr = vect_get_new_vect_var (vect_ptr_type, vect_pointer_var, get_name (base_name)); add_referenced_var (addr_expr); vec_stmt = fold_convert (vect_ptr_type, addr_base); addr_expr2 = vect_get_new_vect_var (vect_ptr_type, vect_pointer_var, get_name (base_name)); add_referenced_var (addr_expr2); vec_stmt = force_gimple_operand (vec_stmt, &seq, false, addr_expr2); gimple_seq_add_seq (new_stmt_list, seq); if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "created "); print_generic_expr (vect_dump, vec_stmt, TDF_SLIM); } return vec_stmt; } /* Function vect_create_data_ref_ptr. Create a new pointer to vector type (vp), that points to the first location accessed in the loop by STMT, along with the def-use update chain to appropriately advance the pointer through the loop iterations. Also set aliasing information for the pointer. This vector pointer is used by the callers to this function to create a memory reference expression for vector load/store access. Input: 1. STMT: a stmt that references memory. Expected to be of the form GIMPLE_ASSIGN or GIMPLE_ASSIGN . 2. AT_LOOP: the loop where the vector memref is to be created. 3. OFFSET (optional): an offset to be added to the initial address accessed by the data-ref in STMT. 4. ONLY_INIT: indicate if vp is to be updated in the loop, or remain pointing to the initial address. 5. TYPE: if not NULL indicates the required type of the data-ref. Output: 1. Declare a new ptr to vector_type, and have it point to the base of the data reference (initial addressed accessed by the data reference). For example, for vector of type V8HI, the following code is generated: v8hi *vp; vp = (v8hi *)initial_address; if OFFSET is not supplied: initial_address = &a[init]; if OFFSET is supplied: initial_address = &a[init + OFFSET]; Return the initial_address in INITIAL_ADDRESS. 2. If ONLY_INIT is true, just return the initial pointer. Otherwise, also update the pointer in each iteration of the loop. Return the increment stmt that updates the pointer in PTR_INCR. 3. Set INV_P to true if the access pattern of the data reference in the vectorized loop is invariant. Set it to false otherwise. 4. Return the pointer. */ static tree vect_create_data_ref_ptr (gimple stmt, struct loop *at_loop, tree offset, tree *initial_address, gimple *ptr_incr, bool only_init, bool *inv_p, tree type) { tree base_name; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); bool nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt); struct loop *containing_loop = (gimple_bb (stmt))->loop_father; tree vectype = STMT_VINFO_VECTYPE (stmt_info); tree vect_ptr_type; tree vect_ptr; tree tag; tree new_temp; gimple vec_stmt; gimple_seq new_stmt_list = NULL; edge pe; basic_block new_bb; tree vect_ptr_init; struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info); tree vptr; gimple_stmt_iterator incr_gsi; bool insert_after; tree indx_before_incr, indx_after_incr; gimple incr; tree step; /* Check the step (evolution) of the load in LOOP, and record whether it's invariant. */ if (nested_in_vect_loop) step = STMT_VINFO_DR_STEP (stmt_info); else step = DR_STEP (STMT_VINFO_DATA_REF (stmt_info)); if (tree_int_cst_compare (step, size_zero_node) == 0) *inv_p = true; else *inv_p = false; /* Create an expression for the first address accessed by this load in LOOP. */ base_name = build_fold_indirect_ref (unshare_expr (DR_BASE_ADDRESS (dr))); if (vect_print_dump_info (REPORT_DETAILS)) { tree data_ref_base = base_name; fprintf (vect_dump, "create vector-pointer variable to type: "); print_generic_expr (vect_dump, vectype, TDF_SLIM); if (TREE_CODE (data_ref_base) == VAR_DECL) fprintf (vect_dump, " vectorizing a one dimensional array ref: "); else if (TREE_CODE (data_ref_base) == ARRAY_REF) fprintf (vect_dump, " vectorizing a multidimensional array ref: "); else if (TREE_CODE (data_ref_base) == COMPONENT_REF) fprintf (vect_dump, " vectorizing a record based array ref: "); else if (TREE_CODE (data_ref_base) == SSA_NAME) fprintf (vect_dump, " vectorizing a pointer ref: "); print_generic_expr (vect_dump, base_name, TDF_SLIM); } /** (1) Create the new vector-pointer variable: **/ if (type) vect_ptr_type = build_pointer_type (type); else vect_ptr_type = build_pointer_type (vectype); if (TREE_CODE (DR_BASE_ADDRESS (dr)) == SSA_NAME && TYPE_RESTRICT (TREE_TYPE (DR_BASE_ADDRESS (dr)))) vect_ptr_type = build_qualified_type (vect_ptr_type, TYPE_QUAL_RESTRICT); vect_ptr = vect_get_new_vect_var (vect_ptr_type, vect_pointer_var, get_name (base_name)); if (TREE_CODE (DR_BASE_ADDRESS (dr)) == SSA_NAME && TYPE_RESTRICT (TREE_TYPE (DR_BASE_ADDRESS (dr)))) { get_alias_set (base_name); DECL_POINTER_ALIAS_SET (vect_ptr) = DECL_POINTER_ALIAS_SET (SSA_NAME_VAR (DR_BASE_ADDRESS (dr))); } add_referenced_var (vect_ptr); /** (2) Add aliasing information to the new vector-pointer: (The points-to info (DR_PTR_INFO) may be defined later.) **/ tag = DR_SYMBOL_TAG (dr); gcc_assert (tag); /* If tag is a variable (and NOT_A_TAG) than a new symbol memory tag must be created with tag added to its may alias list. */ if (!MTAG_P (tag)) new_type_alias (vect_ptr, tag, DR_REF (dr)); else { set_symbol_mem_tag (vect_ptr, tag); mark_sym_for_renaming (tag); } /** Note: If the dataref is in an inner-loop nested in LOOP, and we are vectorizing LOOP (i.e. outer-loop vectorization), we need to create two def-use update cycles for the pointer: One relative to the outer-loop (LOOP), which is what steps (3) and (4) below do. The other is relative to the inner-loop (which is the inner-most loop containing the dataref), and this is done be step (5) below. When vectorizing inner-most loops, the vectorized loop (LOOP) is also the inner-most loop, and so steps (3),(4) work the same, and step (5) is redundant. Steps (3),(4) create the following: vp0 = &base_addr; LOOP: vp1 = phi(vp0,vp2) ... ... vp2 = vp1 + step goto LOOP If there is an inner-loop nested in loop, then step (5) will also be applied, and an additional update in the inner-loop will be created: vp0 = &base_addr; LOOP: vp1 = phi(vp0,vp2) ... inner: vp3 = phi(vp1,vp4) vp4 = vp3 + inner_step if () goto inner ... vp2 = vp1 + step if () goto LOOP */ /** (3) Calculate the initial address the vector-pointer, and set the vector-pointer to point to it before the loop: **/ /* Create: (&(base[init_val+offset]) in the loop preheader. */ new_temp = vect_create_addr_base_for_vector_ref (stmt, &new_stmt_list, offset, loop); pe = loop_preheader_edge (loop); if (new_stmt_list) { new_bb = gsi_insert_seq_on_edge_immediate (pe, new_stmt_list); gcc_assert (!new_bb); } *initial_address = new_temp; /* Create: p = (vectype *) initial_base */ vec_stmt = gimple_build_assign (vect_ptr, fold_convert (vect_ptr_type, new_temp)); vect_ptr_init = make_ssa_name (vect_ptr, vec_stmt); gimple_assign_set_lhs (vec_stmt, vect_ptr_init); new_bb = gsi_insert_on_edge_immediate (pe, vec_stmt); gcc_assert (!new_bb); /** (4) Handle the updating of the vector-pointer inside the loop. This is needed when ONLY_INIT is false, and also when AT_LOOP is the inner-loop nested in LOOP (during outer-loop vectorization). **/ if (only_init && at_loop == loop) /* No update in loop is required. */ { /* Copy the points-to information if it exists. */ if (DR_PTR_INFO (dr)) duplicate_ssa_name_ptr_info (vect_ptr_init, DR_PTR_INFO (dr)); vptr = vect_ptr_init; } else { /* The step of the vector pointer is the Vector Size. */ tree step = TYPE_SIZE_UNIT (vectype); /* One exception to the above is when the scalar step of the load in LOOP is zero. In this case the step here is also zero. */ if (*inv_p) step = size_zero_node; standard_iv_increment_position (loop, &incr_gsi, &insert_after); create_iv (vect_ptr_init, fold_convert (vect_ptr_type, step), vect_ptr, loop, &incr_gsi, insert_after, &indx_before_incr, &indx_after_incr); incr = gsi_stmt (incr_gsi); set_vinfo_for_stmt (incr, new_stmt_vec_info (incr, loop_vinfo)); /* Copy the points-to information if it exists. */ if (DR_PTR_INFO (dr)) { duplicate_ssa_name_ptr_info (indx_before_incr, DR_PTR_INFO (dr)); duplicate_ssa_name_ptr_info (indx_after_incr, DR_PTR_INFO (dr)); } merge_alias_info (vect_ptr_init, indx_before_incr); merge_alias_info (vect_ptr_init, indx_after_incr); if (ptr_incr) *ptr_incr = incr; vptr = indx_before_incr; } if (!nested_in_vect_loop || only_init) return vptr; /** (5) Handle the updating of the vector-pointer inside the inner-loop nested in LOOP, if exists: **/ gcc_assert (nested_in_vect_loop); if (!only_init) { standard_iv_increment_position (containing_loop, &incr_gsi, &insert_after); create_iv (vptr, fold_convert (vect_ptr_type, DR_STEP (dr)), vect_ptr, containing_loop, &incr_gsi, insert_after, &indx_before_incr, &indx_after_incr); incr = gsi_stmt (incr_gsi); set_vinfo_for_stmt (incr, new_stmt_vec_info (incr, loop_vinfo)); /* Copy the points-to information if it exists. */ if (DR_PTR_INFO (dr)) { duplicate_ssa_name_ptr_info (indx_before_incr, DR_PTR_INFO (dr)); duplicate_ssa_name_ptr_info (indx_after_incr, DR_PTR_INFO (dr)); } merge_alias_info (vect_ptr_init, indx_before_incr); merge_alias_info (vect_ptr_init, indx_after_incr); if (ptr_incr) *ptr_incr = incr; return indx_before_incr; } else gcc_unreachable (); } /* Function bump_vector_ptr Increment a pointer (to a vector type) by vector-size. If requested, i.e. if PTR-INCR is given, then also connect the new increment stmt to the existing def-use update-chain of the pointer, by modifying the PTR_INCR as illustrated below: The pointer def-use update-chain before this function: DATAREF_PTR = phi (p_0, p_2) .... PTR_INCR: p_2 = DATAREF_PTR + step The pointer def-use update-chain after this function: DATAREF_PTR = phi (p_0, p_2) .... NEW_DATAREF_PTR = DATAREF_PTR + BUMP .... PTR_INCR: p_2 = NEW_DATAREF_PTR + step Input: DATAREF_PTR - ssa_name of a pointer (to vector type) that is being updated in the loop. PTR_INCR - optional. The stmt that updates the pointer in each iteration of the loop. The increment amount across iterations is expected to be vector_size. BSI - location where the new update stmt is to be placed. STMT - the original scalar memory-access stmt that is being vectorized. BUMP - optional. The offset by which to bump the pointer. If not given, the offset is assumed to be vector_size. Output: Return NEW_DATAREF_PTR as illustrated above. */ static tree bump_vector_ptr (tree dataref_ptr, gimple ptr_incr, gimple_stmt_iterator *gsi, gimple stmt, tree bump) { stmt_vec_info stmt_info = vinfo_for_stmt (stmt); struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info); tree vectype = STMT_VINFO_VECTYPE (stmt_info); tree ptr_var = SSA_NAME_VAR (dataref_ptr); tree update = TYPE_SIZE_UNIT (vectype); gimple incr_stmt; ssa_op_iter iter; use_operand_p use_p; tree new_dataref_ptr; if (bump) update = bump; incr_stmt = gimple_build_assign_with_ops (POINTER_PLUS_EXPR, ptr_var, dataref_ptr, update); new_dataref_ptr = make_ssa_name (ptr_var, incr_stmt); gimple_assign_set_lhs (incr_stmt, new_dataref_ptr); vect_finish_stmt_generation (stmt, incr_stmt, gsi); /* Copy the points-to information if it exists. */ if (DR_PTR_INFO (dr)) duplicate_ssa_name_ptr_info (new_dataref_ptr, DR_PTR_INFO (dr)); merge_alias_info (new_dataref_ptr, dataref_ptr); if (!ptr_incr) return new_dataref_ptr; /* Update the vector-pointer's cross-iteration increment. */ FOR_EACH_SSA_USE_OPERAND (use_p, ptr_incr, iter, SSA_OP_USE) { tree use = USE_FROM_PTR (use_p); if (use == dataref_ptr) SET_USE (use_p, new_dataref_ptr); else gcc_assert (tree_int_cst_compare (use, update) == 0); } return new_dataref_ptr; } /* Function vect_create_destination_var. Create a new temporary of type VECTYPE. */ static tree vect_create_destination_var (tree scalar_dest, tree vectype) { tree vec_dest; const char *new_name; tree type; enum vect_var_kind kind; kind = vectype ? vect_simple_var : vect_scalar_var; type = vectype ? vectype : TREE_TYPE (scalar_dest); gcc_assert (TREE_CODE (scalar_dest) == SSA_NAME); new_name = get_name (scalar_dest); if (!new_name) new_name = "var_"; vec_dest = vect_get_new_vect_var (type, kind, new_name); add_referenced_var (vec_dest); return vec_dest; } /* Function vect_init_vector. Insert a new stmt (INIT_STMT) that initializes a new vector variable with the vector elements of VECTOR_VAR. Place the initialization at BSI if it is not NULL. Otherwise, place the initialization at the loop preheader. Return the DEF of INIT_STMT. It will be used in the vectorization of STMT. */ static tree vect_init_vector (gimple stmt, tree vector_var, tree vector_type, gimple_stmt_iterator *gsi) { stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt); tree new_var; gimple init_stmt; tree vec_oprnd; edge pe; tree new_temp; basic_block new_bb; new_var = vect_get_new_vect_var (vector_type, vect_simple_var, "cst_"); add_referenced_var (new_var); init_stmt = gimple_build_assign (new_var, vector_var); new_temp = make_ssa_name (new_var, init_stmt); gimple_assign_set_lhs (init_stmt, new_temp); if (gsi) vect_finish_stmt_generation (stmt, init_stmt, gsi); else { loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); if (nested_in_vect_loop_p (loop, stmt)) loop = loop->inner; pe = loop_preheader_edge (loop); new_bb = gsi_insert_on_edge_immediate (pe, init_stmt); gcc_assert (!new_bb); } if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "created new init_stmt: "); print_gimple_stmt (vect_dump, init_stmt, 0, TDF_SLIM); } vec_oprnd = gimple_assign_lhs (init_stmt); return vec_oprnd; } /* For constant and loop invariant defs of SLP_NODE this function returns (vector) defs (VEC_OPRNDS) that will be used in the vectorized stmts. OP_NUM determines if we gather defs for operand 0 or operand 1 of the scalar stmts. NUMBER_OF_VECTORS is the number of vector defs to create. */ static void vect_get_constant_vectors (slp_tree slp_node, VEC(tree,heap) **vec_oprnds, unsigned int op_num, unsigned int number_of_vectors) { VEC (gimple, heap) *stmts = SLP_TREE_SCALAR_STMTS (slp_node); gimple stmt = VEC_index (gimple, stmts, 0); stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt); tree vectype = STMT_VINFO_VECTYPE (stmt_vinfo); int nunits; tree vec_cst; tree t = NULL_TREE; int j, number_of_places_left_in_vector; tree vector_type; tree op, vop; int group_size = VEC_length (gimple, stmts); unsigned int vec_num, i; int number_of_copies = 1; VEC (tree, heap) *voprnds = VEC_alloc (tree, heap, number_of_vectors); bool constant_p, is_store; if (STMT_VINFO_DATA_REF (stmt_vinfo)) { is_store = true; op = gimple_assign_rhs1 (stmt); } else { is_store = false; op = gimple_op (stmt, op_num + 1); } if (CONSTANT_CLASS_P (op)) { vector_type = vectype; constant_p = true; } else { vector_type = get_vectype_for_scalar_type (TREE_TYPE (op)); gcc_assert (vector_type); constant_p = false; } nunits = TYPE_VECTOR_SUBPARTS (vector_type); /* NUMBER_OF_COPIES is the number of times we need to use the same values in created vectors. It is greater than 1 if unrolling is performed. For example, we have two scalar operands, s1 and s2 (e.g., group of strided accesses of size two), while NUNITS is four (i.e., four scalars of this type can be packed in a vector). The output vector will contain two copies of each scalar operand: {s1, s2, s1, s2}. (NUMBER_OF_COPIES will be 2). If GROUP_SIZE > NUNITS, the scalars will be split into several vectors containing the operands. For example, NUNITS is four as before, and the group size is 8 (s1, s2, ..., s8). We will create two vectors {s1, s2, s3, s4} and {s5, s6, s7, s8}. */ number_of_copies = least_common_multiple (nunits, group_size) / group_size; number_of_places_left_in_vector = nunits; for (j = 0; j < number_of_copies; j++) { for (i = group_size - 1; VEC_iterate (gimple, stmts, i, stmt); i--) { if (is_store) op = gimple_assign_rhs1 (stmt); else op = gimple_op (stmt, op_num + 1); /* Create 'vect_ = {op0,op1,...,opn}'. */ t = tree_cons (NULL_TREE, op, t); number_of_places_left_in_vector--; if (number_of_places_left_in_vector == 0) { number_of_places_left_in_vector = nunits; if (constant_p) vec_cst = build_vector (vector_type, t); else vec_cst = build_constructor_from_list (vector_type, t); VEC_quick_push (tree, voprnds, vect_init_vector (stmt, vec_cst, vector_type, NULL)); t = NULL_TREE; } } } /* Since the vectors are created in the reverse order, we should invert them. */ vec_num = VEC_length (tree, voprnds); for (j = vec_num - 1; j >= 0; j--) { vop = VEC_index (tree, voprnds, j); VEC_quick_push (tree, *vec_oprnds, vop); } VEC_free (tree, heap, voprnds); /* In case that VF is greater than the unrolling factor needed for the SLP group of stmts, NUMBER_OF_VECTORS to be created is greater than NUMBER_OF_SCALARS/NUNITS or NUNITS/NUMBER_OF_SCALARS, and hence we have to replicate the vectors. */ while (number_of_vectors > VEC_length (tree, *vec_oprnds)) { for (i = 0; VEC_iterate (tree, *vec_oprnds, i, vop) && i < vec_num; i++) VEC_quick_push (tree, *vec_oprnds, vop); } } /* Get vectorized definitions from SLP_NODE that contains corresponding vectorized def-stmts. */ static void vect_get_slp_vect_defs (slp_tree slp_node, VEC (tree,heap) **vec_oprnds) { tree vec_oprnd; gimple vec_def_stmt; unsigned int i; gcc_assert (SLP_TREE_VEC_STMTS (slp_node)); for (i = 0; VEC_iterate (gimple, SLP_TREE_VEC_STMTS (slp_node), i, vec_def_stmt); i++) { gcc_assert (vec_def_stmt); vec_oprnd = gimple_get_lhs (vec_def_stmt); VEC_quick_push (tree, *vec_oprnds, vec_oprnd); } } /* Get vectorized definitions for SLP_NODE. If the scalar definitions are loop invariants or constants, collect them and call vect_get_constant_vectors() to create vector stmts. Otherwise, the def-stmts must be already vectorized and the vectorized stmts must be stored in the LEFT/RIGHT node of SLP_NODE, and we call vect_get_slp_vect_defs() to retrieve them. If VEC_OPRNDS1 is NULL, don't get vector defs for the second operand (from the right node. This is used when the second operand must remain scalar. */ static void vect_get_slp_defs (slp_tree slp_node, VEC (tree,heap) **vec_oprnds0, VEC (tree,heap) **vec_oprnds1) { gimple first_stmt; enum tree_code code; int number_of_vects; HOST_WIDE_INT lhs_size_unit, rhs_size_unit; first_stmt = VEC_index (gimple, SLP_TREE_SCALAR_STMTS (slp_node), 0); /* The number of vector defs is determined by the number of vector statements in the node from which we get those statements. */ if (SLP_TREE_LEFT (slp_node)) number_of_vects = SLP_TREE_NUMBER_OF_VEC_STMTS (SLP_TREE_LEFT (slp_node)); else { number_of_vects = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node); /* Number of vector stmts was calculated according to LHS in vect_schedule_slp_instance(), fix it by replacing LHS with RHS, if necessary. See vect_get_smallest_scalar_type() for details. */ vect_get_smallest_scalar_type (first_stmt, &lhs_size_unit, &rhs_size_unit); if (rhs_size_unit != lhs_size_unit) { number_of_vects *= rhs_size_unit; number_of_vects /= lhs_size_unit; } } /* Allocate memory for vectorized defs. */ *vec_oprnds0 = VEC_alloc (tree, heap, number_of_vects); /* SLP_NODE corresponds either to a group of stores or to a group of unary/binary operations. We don't call this function for loads. */ if (SLP_TREE_LEFT (slp_node)) /* The defs are already vectorized. */ vect_get_slp_vect_defs (SLP_TREE_LEFT (slp_node), vec_oprnds0); else /* Build vectors from scalar defs. */ vect_get_constant_vectors (slp_node, vec_oprnds0, 0, number_of_vects); if (STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt))) /* Since we don't call this function with loads, this is a group of stores. */ return; code = gimple_assign_rhs_code (first_stmt); if (get_gimple_rhs_class (code) != GIMPLE_BINARY_RHS || !vec_oprnds1) return; /* The number of vector defs is determined by the number of vector statements in the node from which we get those statements. */ if (SLP_TREE_RIGHT (slp_node)) number_of_vects = SLP_TREE_NUMBER_OF_VEC_STMTS (SLP_TREE_RIGHT (slp_node)); else number_of_vects = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node); *vec_oprnds1 = VEC_alloc (tree, heap, number_of_vects); if (SLP_TREE_RIGHT (slp_node)) /* The defs are already vectorized. */ vect_get_slp_vect_defs (SLP_TREE_RIGHT (slp_node), vec_oprnds1); else /* Build vectors from scalar defs. */ vect_get_constant_vectors (slp_node, vec_oprnds1, 1, number_of_vects); } /* Function get_initial_def_for_induction Input: STMT - a stmt that performs an induction operation in the loop. IV_PHI - the initial value of the induction variable Output: Return a vector variable, initialized with the first VF values of the induction variable. E.g., for an iv with IV_PHI='X' and evolution S, for a vector of 4 units, we want to return: [X, X + S, X + 2*S, X + 3*S]. */ static tree get_initial_def_for_induction (gimple iv_phi) { stmt_vec_info stmt_vinfo = vinfo_for_stmt (iv_phi); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); tree scalar_type = TREE_TYPE (gimple_phi_result (iv_phi)); tree vectype; int nunits; edge pe = loop_preheader_edge (loop); struct loop *iv_loop; basic_block new_bb; tree vec, vec_init, vec_step, t; tree access_fn; tree new_var; tree new_name; gimple init_stmt, induction_phi, new_stmt; tree induc_def, vec_def, vec_dest; tree init_expr, step_expr; int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); int i; bool ok; int ncopies; tree expr; stmt_vec_info phi_info = vinfo_for_stmt (iv_phi); bool nested_in_vect_loop = false; gimple_seq stmts = NULL; imm_use_iterator imm_iter; use_operand_p use_p; gimple exit_phi; edge latch_e; tree loop_arg; gimple_stmt_iterator si; basic_block bb = gimple_bb (iv_phi); vectype = get_vectype_for_scalar_type (scalar_type); gcc_assert (vectype); nunits = TYPE_VECTOR_SUBPARTS (vectype); ncopies = vf / nunits; gcc_assert (phi_info); gcc_assert (ncopies >= 1); /* Find the first insertion point in the BB. */ si = gsi_after_labels (bb); if (INTEGRAL_TYPE_P (scalar_type) || POINTER_TYPE_P (scalar_type)) step_expr = build_int_cst (scalar_type, 0); else step_expr = build_real (scalar_type, dconst0); /* Is phi in an inner-loop, while vectorizing an enclosing outer-loop? */ if (nested_in_vect_loop_p (loop, iv_phi)) { nested_in_vect_loop = true; iv_loop = loop->inner; } else iv_loop = loop; gcc_assert (iv_loop == (gimple_bb (iv_phi))->loop_father); latch_e = loop_latch_edge (iv_loop); loop_arg = PHI_ARG_DEF_FROM_EDGE (iv_phi, latch_e); access_fn = analyze_scalar_evolution (iv_loop, PHI_RESULT (iv_phi)); gcc_assert (access_fn); ok = vect_is_simple_iv_evolution (iv_loop->num, access_fn, &init_expr, &step_expr); gcc_assert (ok); pe = loop_preheader_edge (iv_loop); /* Create the vector that holds the initial_value of the induction. */ if (nested_in_vect_loop) { /* iv_loop is nested in the loop to be vectorized. init_expr had already been created during vectorization of previous stmts; We obtain it from the STMT_VINFO_VEC_STMT of the defining stmt. */ tree iv_def = PHI_ARG_DEF_FROM_EDGE (iv_phi, loop_preheader_edge (iv_loop)); vec_init = vect_get_vec_def_for_operand (iv_def, iv_phi, NULL); } else { /* iv_loop is the loop to be vectorized. Create: vec_init = [X, X+S, X+2*S, X+3*S] (S = step_expr, X = init_expr) */ new_var = vect_get_new_vect_var (scalar_type, vect_scalar_var, "var_"); add_referenced_var (new_var); new_name = force_gimple_operand (init_expr, &stmts, false, new_var); if (stmts) { new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); gcc_assert (!new_bb); } t = NULL_TREE; t = tree_cons (NULL_TREE, init_expr, t); for (i = 1; i < nunits; i++) { /* Create: new_name_i = new_name + step_expr */ enum tree_code code = POINTER_TYPE_P (scalar_type) ? POINTER_PLUS_EXPR : PLUS_EXPR; init_stmt = gimple_build_assign_with_ops (code, new_var, new_name, step_expr); new_name = make_ssa_name (new_var, init_stmt); gimple_assign_set_lhs (init_stmt, new_name); new_bb = gsi_insert_on_edge_immediate (pe, init_stmt); gcc_assert (!new_bb); if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "created new init_stmt: "); print_gimple_stmt (vect_dump, init_stmt, 0, TDF_SLIM); } t = tree_cons (NULL_TREE, new_name, t); } /* Create a vector from [new_name_0, new_name_1, ..., new_name_nunits-1] */ vec = build_constructor_from_list (vectype, nreverse (t)); vec_init = vect_init_vector (iv_phi, vec, vectype, NULL); } /* Create the vector that holds the step of the induction. */ if (nested_in_vect_loop) /* iv_loop is nested in the loop to be vectorized. Generate: vec_step = [S, S, S, S] */ new_name = step_expr; else { /* iv_loop is the loop to be vectorized. Generate: vec_step = [VF*S, VF*S, VF*S, VF*S] */ expr = build_int_cst (scalar_type, vf); new_name = fold_build2 (MULT_EXPR, scalar_type, expr, step_expr); } t = NULL_TREE; for (i = 0; i < nunits; i++) t = tree_cons (NULL_TREE, unshare_expr (new_name), t); gcc_assert (CONSTANT_CLASS_P (new_name)); vec = build_vector (vectype, t); vec_step = vect_init_vector (iv_phi, vec, vectype, NULL); /* Create the following def-use cycle: loop prolog: vec_init = ... vec_step = ... loop: vec_iv = PHI ... STMT ... vec_loop = vec_iv + vec_step; */ /* Create the induction-phi that defines the induction-operand. */ vec_dest = vect_get_new_vect_var (vectype, vect_simple_var, "vec_iv_"); add_referenced_var (vec_dest); induction_phi = create_phi_node (vec_dest, iv_loop->header); set_vinfo_for_stmt (induction_phi, new_stmt_vec_info (induction_phi, loop_vinfo)); induc_def = PHI_RESULT (induction_phi); /* Create the iv update inside the loop */ new_stmt = gimple_build_assign_with_ops (PLUS_EXPR, vec_dest, induc_def, vec_step); vec_def = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, vec_def); gsi_insert_before (&si, new_stmt, GSI_SAME_STMT); set_vinfo_for_stmt (new_stmt, new_stmt_vec_info (new_stmt, loop_vinfo)); /* Set the arguments of the phi node: */ add_phi_arg (induction_phi, vec_init, pe, UNKNOWN_LOCATION); add_phi_arg (induction_phi, vec_def, loop_latch_edge (iv_loop), UNKNOWN_LOCATION); /* In case that vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits), we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. For more details see documentation in vectorizable_operation. */ if (ncopies > 1) { stmt_vec_info prev_stmt_vinfo; /* FORNOW. This restriction should be relaxed. */ gcc_assert (!nested_in_vect_loop); /* Create the vector that holds the step of the induction. */ expr = build_int_cst (scalar_type, nunits); new_name = fold_build2 (MULT_EXPR, scalar_type, expr, step_expr); t = NULL_TREE; for (i = 0; i < nunits; i++) t = tree_cons (NULL_TREE, unshare_expr (new_name), t); gcc_assert (CONSTANT_CLASS_P (new_name)); vec = build_vector (vectype, t); vec_step = vect_init_vector (iv_phi, vec, vectype, NULL); vec_def = induc_def; prev_stmt_vinfo = vinfo_for_stmt (induction_phi); for (i = 1; i < ncopies; i++) { /* vec_i = vec_prev + vec_step */ new_stmt = gimple_build_assign_with_ops (PLUS_EXPR, vec_dest, vec_def, vec_step); vec_def = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, vec_def); gsi_insert_before (&si, new_stmt, GSI_SAME_STMT); set_vinfo_for_stmt (new_stmt, new_stmt_vec_info (new_stmt, loop_vinfo)); STMT_VINFO_RELATED_STMT (prev_stmt_vinfo) = new_stmt; prev_stmt_vinfo = vinfo_for_stmt (new_stmt); } } if (nested_in_vect_loop) { /* Find the loop-closed exit-phi of the induction, and record the final vector of induction results: */ exit_phi = NULL; FOR_EACH_IMM_USE_FAST (use_p, imm_iter, loop_arg) { if (!flow_bb_inside_loop_p (iv_loop, gimple_bb (USE_STMT (use_p)))) { exit_phi = USE_STMT (use_p); break; } } if (exit_phi) { stmt_vec_info stmt_vinfo = vinfo_for_stmt (exit_phi); /* FORNOW. Currently not supporting the case that an inner-loop induction is not used in the outer-loop (i.e. only outside the outer-loop). */ gcc_assert (STMT_VINFO_RELEVANT_P (stmt_vinfo) && !STMT_VINFO_LIVE_P (stmt_vinfo)); STMT_VINFO_VEC_STMT (stmt_vinfo) = new_stmt; if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "vector of inductions after inner-loop:"); print_gimple_stmt (vect_dump, new_stmt, 0, TDF_SLIM); } } } if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "transform induction: created def-use cycle: "); print_gimple_stmt (vect_dump, induction_phi, 0, TDF_SLIM); fprintf (vect_dump, "\n"); print_gimple_stmt (vect_dump, SSA_NAME_DEF_STMT (vec_def), 0, TDF_SLIM); } STMT_VINFO_VEC_STMT (phi_info) = induction_phi; return induc_def; } /* Function vect_get_vec_def_for_operand. OP is an operand in STMT. This function returns a (vector) def that will be used in the vectorized stmt for STMT. In the case that OP is an SSA_NAME which is defined in the loop, then STMT_VINFO_VEC_STMT of the defining stmt holds the relevant def. In case OP is an invariant or constant, a new stmt that creates a vector def needs to be introduced. */ static tree vect_get_vec_def_for_operand (tree op, gimple stmt, tree *scalar_def) { tree vec_oprnd; gimple vec_stmt; gimple def_stmt; stmt_vec_info def_stmt_info = NULL; stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt); tree vectype = STMT_VINFO_VECTYPE (stmt_vinfo); unsigned int nunits = TYPE_VECTOR_SUBPARTS (vectype); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); tree vec_inv; tree vec_cst; tree t = NULL_TREE; tree def; int i; enum vect_def_type dt; bool is_simple_use; tree vector_type; if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "vect_get_vec_def_for_operand: "); print_generic_expr (vect_dump, op, TDF_SLIM); } is_simple_use = vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt); gcc_assert (is_simple_use); if (vect_print_dump_info (REPORT_DETAILS)) { if (def) { fprintf (vect_dump, "def = "); print_generic_expr (vect_dump, def, TDF_SLIM); } if (def_stmt) { fprintf (vect_dump, " def_stmt = "); print_gimple_stmt (vect_dump, def_stmt, 0, TDF_SLIM); } } switch (dt) { /* Case 1: operand is a constant. */ case vect_constant_def: { if (scalar_def) *scalar_def = op; /* Create 'vect_cst_ = {cst,cst,...,cst}' */ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "Create vector_cst. nunits = %d", nunits); for (i = nunits - 1; i >= 0; --i) { t = tree_cons (NULL_TREE, op, t); } vec_cst = build_vector (vectype, t); return vect_init_vector (stmt, vec_cst, vectype, NULL); } /* Case 2: operand is defined outside the loop - loop invariant. */ case vect_invariant_def: { vector_type = get_vectype_for_scalar_type (TREE_TYPE (def)); gcc_assert (vector_type); nunits = TYPE_VECTOR_SUBPARTS (vector_type); if (scalar_def) *scalar_def = def; /* Create 'vec_inv = {inv,inv,..,inv}' */ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "Create vector_inv."); for (i = nunits - 1; i >= 0; --i) { t = tree_cons (NULL_TREE, def, t); } /* FIXME: use build_constructor directly. */ vec_inv = build_constructor_from_list (vector_type, t); return vect_init_vector (stmt, vec_inv, vector_type, NULL); } /* Case 3: operand is defined inside the loop. */ case vect_loop_def: { if (scalar_def) *scalar_def = NULL/* FIXME tuples: def_stmt*/; /* Get the def from the vectorized stmt. */ def_stmt_info = vinfo_for_stmt (def_stmt); vec_stmt = STMT_VINFO_VEC_STMT (def_stmt_info); gcc_assert (vec_stmt); if (gimple_code (vec_stmt) == GIMPLE_PHI) vec_oprnd = PHI_RESULT (vec_stmt); else if (is_gimple_call (vec_stmt)) vec_oprnd = gimple_call_lhs (vec_stmt); else vec_oprnd = gimple_assign_lhs (vec_stmt); return vec_oprnd; } /* Case 4: operand is defined by a loop header phi - reduction */ case vect_reduction_def: { struct loop *loop; gcc_assert (gimple_code (def_stmt) == GIMPLE_PHI); loop = (gimple_bb (def_stmt))->loop_father; /* Get the def before the loop */ op = PHI_ARG_DEF_FROM_EDGE (def_stmt, loop_preheader_edge (loop)); return get_initial_def_for_reduction (stmt, op, scalar_def); } /* Case 5: operand is defined by loop-header phi - induction. */ case vect_induction_def: { gcc_assert (gimple_code (def_stmt) == GIMPLE_PHI); /* Get the def from the vectorized stmt. */ def_stmt_info = vinfo_for_stmt (def_stmt); vec_stmt = STMT_VINFO_VEC_STMT (def_stmt_info); gcc_assert (vec_stmt && gimple_code (vec_stmt) == GIMPLE_PHI); vec_oprnd = PHI_RESULT (vec_stmt); return vec_oprnd; } default: gcc_unreachable (); } } /* Function vect_get_vec_def_for_stmt_copy Return a vector-def for an operand. This function is used when the vectorized stmt to be created (by the caller to this function) is a "copy" created in case the vectorized result cannot fit in one vector, and several copies of the vector-stmt are required. In this case the vector-def is retrieved from the vector stmt recorded in the STMT_VINFO_RELATED_STMT field of the stmt that defines VEC_OPRND. DT is the type of the vector def VEC_OPRND. Context: In case the vectorization factor (VF) is bigger than the number of elements that can fit in a vectype (nunits), we have to generate more than one vector stmt to vectorize the scalar stmt. This situation arises when there are multiple data-types operated upon in the loop; the smallest data-type determines the VF, and as a result, when vectorizing stmts operating on wider types we need to create 'VF/nunits' "copies" of the vector stmt (each computing a vector of 'nunits' results, and together computing 'VF' results in each iteration). This function is called when vectorizing such a stmt (e.g. vectorizing S2 in the illustration below, in which VF=16 and nunits=4, so the number of copies required is 4): scalar stmt: vectorized into: STMT_VINFO_RELATED_STMT S1: x = load VS1.0: vx.0 = memref0 VS1.1 VS1.1: vx.1 = memref1 VS1.2 VS1.2: vx.2 = memref2 VS1.3 VS1.3: vx.3 = memref3 S2: z = x + ... VSnew.0: vz0 = vx.0 + ... VSnew.1 VSnew.1: vz1 = vx.1 + ... VSnew.2 VSnew.2: vz2 = vx.2 + ... VSnew.3 VSnew.3: vz3 = vx.3 + ... The vectorization of S1 is explained in vectorizable_load. The vectorization of S2: To create the first vector-stmt out of the 4 copies - VSnew.0 - the function 'vect_get_vec_def_for_operand' is called to get the relevant vector-def for each operand of S2. For operand x it returns the vector-def 'vx.0'. To create the remaining copies of the vector-stmt (VSnew.j), this function is called to get the relevant vector-def for each operand. It is obtained from the respective VS1.j stmt, which is recorded in the STMT_VINFO_RELATED_STMT field of the stmt that defines VEC_OPRND. For example, to obtain the vector-def 'vx.1' in order to create the vector stmt 'VSnew.1', this function is called with VEC_OPRND='vx.0'. Given 'vx0' we obtain the stmt that defines it ('VS1.0'); from the STMT_VINFO_RELATED_STMT field of 'VS1.0' we obtain the next copy - 'VS1.1', and return its def ('vx.1'). Overall, to create the above sequence this function will be called 3 times: vx.1 = vect_get_vec_def_for_stmt_copy (dt, vx.0); vx.2 = vect_get_vec_def_for_stmt_copy (dt, vx.1); vx.3 = vect_get_vec_def_for_stmt_copy (dt, vx.2); */ static tree vect_get_vec_def_for_stmt_copy (enum vect_def_type dt, tree vec_oprnd) { gimple vec_stmt_for_operand; stmt_vec_info def_stmt_info; /* Do nothing; can reuse same def. */ if (dt == vect_invariant_def || dt == vect_constant_def ) return vec_oprnd; vec_stmt_for_operand = SSA_NAME_DEF_STMT (vec_oprnd); def_stmt_info = vinfo_for_stmt (vec_stmt_for_operand); gcc_assert (def_stmt_info); vec_stmt_for_operand = STMT_VINFO_RELATED_STMT (def_stmt_info); gcc_assert (vec_stmt_for_operand); vec_oprnd = gimple_get_lhs (vec_stmt_for_operand); if (gimple_code (vec_stmt_for_operand) == GIMPLE_PHI) vec_oprnd = PHI_RESULT (vec_stmt_for_operand); else vec_oprnd = gimple_get_lhs (vec_stmt_for_operand); return vec_oprnd; } /* Get vectorized definitions for the operands to create a copy of an original stmt. See vect_get_vec_def_for_stmt_copy() for details. */ static void vect_get_vec_defs_for_stmt_copy (enum vect_def_type *dt, VEC(tree,heap) **vec_oprnds0, VEC(tree,heap) **vec_oprnds1) { tree vec_oprnd = VEC_pop (tree, *vec_oprnds0); vec_oprnd = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd); VEC_quick_push (tree, *vec_oprnds0, vec_oprnd); if (vec_oprnds1 && *vec_oprnds1) { vec_oprnd = VEC_pop (tree, *vec_oprnds1); vec_oprnd = vect_get_vec_def_for_stmt_copy (dt[1], vec_oprnd); VEC_quick_push (tree, *vec_oprnds1, vec_oprnd); } } /* Get vectorized definitions for OP0 and OP1, or SLP_NODE if it is not NULL. */ static void vect_get_vec_defs (tree op0, tree op1, gimple stmt, VEC(tree,heap) **vec_oprnds0, VEC(tree,heap) **vec_oprnds1, slp_tree slp_node) { if (slp_node) vect_get_slp_defs (slp_node, vec_oprnds0, vec_oprnds1); else { tree vec_oprnd; *vec_oprnds0 = VEC_alloc (tree, heap, 1); vec_oprnd = vect_get_vec_def_for_operand (op0, stmt, NULL); VEC_quick_push (tree, *vec_oprnds0, vec_oprnd); if (op1) { *vec_oprnds1 = VEC_alloc (tree, heap, 1); vec_oprnd = vect_get_vec_def_for_operand (op1, stmt, NULL); VEC_quick_push (tree, *vec_oprnds1, vec_oprnd); } } } /* Function vect_finish_stmt_generation. Insert a new stmt. */ static void vect_finish_stmt_generation (gimple stmt, gimple vec_stmt, gimple_stmt_iterator *gsi) { stmt_vec_info stmt_info = vinfo_for_stmt (stmt); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); gcc_assert (gimple_code (stmt) != GIMPLE_LABEL); gsi_insert_before (gsi, vec_stmt, GSI_SAME_STMT); set_vinfo_for_stmt (vec_stmt, new_stmt_vec_info (vec_stmt, loop_vinfo)); if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "add new stmt: "); print_gimple_stmt (vect_dump, vec_stmt, 0, TDF_SLIM); } gimple_set_location (vec_stmt, gimple_location (gsi_stmt (*gsi))); } /* Function get_initial_def_for_reduction Input: STMT - a stmt that performs a reduction operation in the loop. INIT_VAL - the initial value of the reduction variable Output: ADJUSTMENT_DEF - a tree that holds a value to be added to the final result of the reduction (used for adjusting the epilog - see below). Return a vector variable, initialized according to the operation that STMT performs. This vector will be used as the initial value of the vector of partial results. Option1 (adjust in epilog): Initialize the vector as follows: add: [0,0,...,0,0] mult: [1,1,...,1,1] min/max: [init_val,init_val,..,init_val,init_val] bit and/or: [init_val,init_val,..,init_val,init_val] and when necessary (e.g. add/mult case) let the caller know that it needs to adjust the result by init_val. Option2: Initialize the vector as follows: add: [0,0,...,0,init_val] mult: [1,1,...,1,init_val] min/max: [init_val,init_val,...,init_val] bit and/or: [init_val,init_val,...,init_val] and no adjustments are needed. For example, for the following code: s = init_val; for (i=0;iloop_father); vecdef = vect_get_vec_def_for_operand (init_val, stmt, NULL); switch (code) { case WIDEN_SUM_EXPR: case DOT_PROD_EXPR: case PLUS_EXPR: if (nested_in_vect_loop) *adjustment_def = vecdef; else *adjustment_def = init_val; /* Create a vector of zeros for init_def. */ if (SCALAR_FLOAT_TYPE_P (scalar_type)) def_for_init = build_real (scalar_type, dconst0); else def_for_init = build_int_cst (scalar_type, 0); for (i = nunits - 1; i >= 0; --i) t = tree_cons (NULL_TREE, def_for_init, t); init_def = build_vector (vectype, t); break; case MIN_EXPR: case MAX_EXPR: *adjustment_def = NULL_TREE; init_def = vecdef; break; default: gcc_unreachable (); } return init_def; } /* Function vect_create_epilog_for_reduction Create code at the loop-epilog to finalize the result of a reduction computation. VECT_DEF is a vector of partial results. REDUC_CODE is the tree-code for the epilog reduction. NCOPIES is > 1 in case the vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits). In this case we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. For more details see documentation in vectorizable_operation. STMT is the scalar reduction stmt that is being vectorized. REDUCTION_PHI is the phi-node that carries the reduction computation. This function: 1. Creates the reduction def-use cycle: sets the arguments for REDUCTION_PHI: The loop-entry argument is the vectorized initial-value of the reduction. The loop-latch argument is VECT_DEF - the vector of partial sums. 2. "Reduces" the vector of partial results VECT_DEF into a single result, by applying the operation specified by REDUC_CODE if available, or by other means (whole-vector shifts or a scalar loop). The function also creates a new phi node at the loop exit to preserve loop-closed form, as illustrated below. The flow at the entry to this function: loop: vec_def = phi # REDUCTION_PHI VECT_DEF = vector_stmt # vectorized form of STMT s_loop = scalar_stmt # (scalar) STMT loop_exit: s_out0 = phi # (scalar) EXIT_PHI use use The above is transformed by this function into: loop: vec_def = phi # REDUCTION_PHI VECT_DEF = vector_stmt # vectorized form of STMT s_loop = scalar_stmt # (scalar) STMT loop_exit: s_out0 = phi # (scalar) EXIT_PHI v_out1 = phi # NEW_EXIT_PHI v_out2 = reduce s_out3 = extract_field s_out4 = adjust_result use use */ static void vect_create_epilog_for_reduction (tree vect_def, gimple stmt, int ncopies, enum tree_code reduc_code, gimple reduction_phi) { stmt_vec_info stmt_info = vinfo_for_stmt (stmt); stmt_vec_info prev_phi_info; tree vectype; enum machine_mode mode; loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); basic_block exit_bb; tree scalar_dest; tree scalar_type; gimple new_phi = NULL, phi; gimple_stmt_iterator exit_gsi; tree vec_dest; tree new_temp = NULL_TREE; tree new_name; gimple epilog_stmt = NULL; tree new_scalar_dest, new_dest; gimple exit_phi; tree bitsize, bitpos, bytesize; enum tree_code code = gimple_assign_rhs_code (stmt); tree adjustment_def; tree vec_initial_def, def; tree orig_name; imm_use_iterator imm_iter; use_operand_p use_p; bool extract_scalar_result = false; tree reduction_op, expr; gimple orig_stmt; gimple use_stmt; bool nested_in_vect_loop = false; VEC(gimple,heap) *phis = NULL; enum vect_def_type dt = vect_unknown_def_type; int j, i; if (nested_in_vect_loop_p (loop, stmt)) { loop = loop->inner; nested_in_vect_loop = true; } switch (get_gimple_rhs_class (gimple_assign_rhs_code (stmt))) { case GIMPLE_SINGLE_RHS: gcc_assert (TREE_OPERAND_LENGTH (gimple_assign_rhs1 (stmt)) == ternary_op); reduction_op = TREE_OPERAND (gimple_assign_rhs1 (stmt), 2); break; case GIMPLE_UNARY_RHS: reduction_op = gimple_assign_rhs1 (stmt); break; case GIMPLE_BINARY_RHS: reduction_op = gimple_assign_rhs2 (stmt); break; default: gcc_unreachable (); } vectype = get_vectype_for_scalar_type (TREE_TYPE (reduction_op)); gcc_assert (vectype); mode = TYPE_MODE (vectype); /*** 1. Create the reduction def-use cycle ***/ /* For the case of reduction, vect_get_vec_def_for_operand returns the scalar def before the loop, that defines the initial value of the reduction variable. */ vec_initial_def = vect_get_vec_def_for_operand (reduction_op, stmt, &adjustment_def); phi = reduction_phi; def = vect_def; for (j = 0; j < ncopies; j++) { /* 1.1 set the loop-entry arg of the reduction-phi: */ add_phi_arg (phi, vec_initial_def, loop_preheader_edge (loop), UNKNOWN_LOCATION); /* 1.2 set the loop-latch arg for the reduction-phi: */ if (j > 0) def = vect_get_vec_def_for_stmt_copy (dt, def); add_phi_arg (phi, def, loop_latch_edge (loop), UNKNOWN_LOCATION); if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "transform reduction: created def-use cycle: "); print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); fprintf (vect_dump, "\n"); print_gimple_stmt (vect_dump, SSA_NAME_DEF_STMT (def), 0, TDF_SLIM); } phi = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (phi)); } /*** 2. Create epilog code The reduction epilog code operates across the elements of the vector of partial results computed by the vectorized loop. The reduction epilog code consists of: step 1: compute the scalar result in a vector (v_out2) step 2: extract the scalar result (s_out3) from the vector (v_out2) step 3: adjust the scalar result (s_out3) if needed. Step 1 can be accomplished using one the following three schemes: (scheme 1) using reduc_code, if available. (scheme 2) using whole-vector shifts, if available. (scheme 3) using a scalar loop. In this case steps 1+2 above are combined. The overall epilog code looks like this: s_out0 = phi # original EXIT_PHI v_out1 = phi # NEW_EXIT_PHI v_out2 = reduce # step 1 s_out3 = extract_field # step 2 s_out4 = adjust_result # step 3 (step 3 is optional, and steps 1 and 2 may be combined). Lastly, the uses of s_out0 are replaced by s_out4. ***/ /* 2.1 Create new loop-exit-phi to preserve loop-closed form: v_out1 = phi */ exit_bb = single_exit (loop)->dest; def = vect_def; prev_phi_info = NULL; for (j = 0; j < ncopies; j++) { phi = create_phi_node (SSA_NAME_VAR (vect_def), exit_bb); set_vinfo_for_stmt (phi, new_stmt_vec_info (phi, loop_vinfo)); if (j == 0) new_phi = phi; else { def = vect_get_vec_def_for_stmt_copy (dt, def); STMT_VINFO_RELATED_STMT (prev_phi_info) = phi; } SET_PHI_ARG_DEF (phi, single_exit (loop)->dest_idx, def); prev_phi_info = vinfo_for_stmt (phi); } exit_gsi = gsi_after_labels (exit_bb); /* 2.2 Get the relevant tree-code to use in the epilog for schemes 2,3 (i.e. when reduc_code is not available) and in the final adjustment code (if needed). Also get the original scalar reduction variable as defined in the loop. In case STMT is a "pattern-stmt" (i.e. - it represents a reduction pattern), the tree-code and scalar-def are taken from the original stmt that the pattern-stmt (STMT) replaces. Otherwise (it is a regular reduction) - the tree-code and scalar-def are taken from STMT. */ orig_stmt = STMT_VINFO_RELATED_STMT (stmt_info); if (!orig_stmt) { /* Regular reduction */ orig_stmt = stmt; } else { /* Reduction pattern */ stmt_vec_info stmt_vinfo = vinfo_for_stmt (orig_stmt); gcc_assert (STMT_VINFO_IN_PATTERN_P (stmt_vinfo)); gcc_assert (STMT_VINFO_RELATED_STMT (stmt_vinfo) == stmt); } code = gimple_assign_rhs_code (orig_stmt); scalar_dest = gimple_assign_lhs (orig_stmt); scalar_type = TREE_TYPE (scalar_dest); new_scalar_dest = vect_create_destination_var (scalar_dest, NULL); bitsize = TYPE_SIZE (scalar_type); bytesize = TYPE_SIZE_UNIT (scalar_type); /* In case this is a reduction in an inner-loop while vectorizing an outer loop - we don't need to extract a single scalar result at the end of the inner-loop. The final vector of partial results will be used in the vectorized outer-loop, or reduced to a scalar result at the end of the outer-loop. */ if (nested_in_vect_loop) goto vect_finalize_reduction; /* FORNOW */ gcc_assert (ncopies == 1); /* 2.3 Create the reduction code, using one of the three schemes described above. */ if (reduc_code < NUM_TREE_CODES) { tree tmp; /*** Case 1: Create: v_out2 = reduc_expr */ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "Reduce using direct vector reduction."); vec_dest = vect_create_destination_var (scalar_dest, vectype); tmp = build1 (reduc_code, vectype, PHI_RESULT (new_phi)); epilog_stmt = gimple_build_assign (vec_dest, tmp); new_temp = make_ssa_name (vec_dest, epilog_stmt); gimple_assign_set_lhs (epilog_stmt, new_temp); gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); extract_scalar_result = true; } else { enum tree_code shift_code = 0; bool have_whole_vector_shift = true; int bit_offset; int element_bitsize = tree_low_cst (bitsize, 1); int vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1); tree vec_temp; if (optab_handler (vec_shr_optab, mode)->insn_code != CODE_FOR_nothing) shift_code = VEC_RSHIFT_EXPR; else have_whole_vector_shift = false; /* Regardless of whether we have a whole vector shift, if we're emulating the operation via tree-vect-generic, we don't want to use it. Only the first round of the reduction is likely to still be profitable via emulation. */ /* ??? It might be better to emit a reduction tree code here, so that tree-vect-generic can expand the first round via bit tricks. */ if (!VECTOR_MODE_P (mode)) have_whole_vector_shift = false; else { optab optab = optab_for_tree_code (code, vectype, optab_default); if (optab_handler (optab, mode)->insn_code == CODE_FOR_nothing) have_whole_vector_shift = false; } if (have_whole_vector_shift) { /*** Case 2: Create: for (offset = VS/2; offset >= element_size; offset/=2) { Create: va' = vec_shift Create: va = vop } */ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "Reduce using vector shifts"); vec_dest = vect_create_destination_var (scalar_dest, vectype); new_temp = PHI_RESULT (new_phi); for (bit_offset = vec_size_in_bits/2; bit_offset >= element_bitsize; bit_offset /= 2) { tree bitpos = size_int (bit_offset); epilog_stmt = gimple_build_assign_with_ops (shift_code, vec_dest, new_temp, bitpos); new_name = make_ssa_name (vec_dest, epilog_stmt); gimple_assign_set_lhs (epilog_stmt, new_name); gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); epilog_stmt = gimple_build_assign_with_ops (code, vec_dest, new_name, new_temp); new_temp = make_ssa_name (vec_dest, epilog_stmt); gimple_assign_set_lhs (epilog_stmt, new_temp); gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); } extract_scalar_result = true; } else { tree rhs; /*** Case 3: Create: s = extract_field for (offset = element_size; offset < vector_size; offset += element_size;) { Create: s' = extract_field Create: s = op } */ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "Reduce using scalar code. "); vec_temp = PHI_RESULT (new_phi); vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1); rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, bitsize, bitsize_zero_node); epilog_stmt = gimple_build_assign (new_scalar_dest, rhs); new_temp = make_ssa_name (new_scalar_dest, epilog_stmt); gimple_assign_set_lhs (epilog_stmt, new_temp); gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); for (bit_offset = element_bitsize; bit_offset < vec_size_in_bits; bit_offset += element_bitsize) { tree bitpos = bitsize_int (bit_offset); tree rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, bitsize, bitpos); epilog_stmt = gimple_build_assign (new_scalar_dest, rhs); new_name = make_ssa_name (new_scalar_dest, epilog_stmt); gimple_assign_set_lhs (epilog_stmt, new_name); gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); epilog_stmt = gimple_build_assign_with_ops (code, new_scalar_dest, new_name, new_temp); new_temp = make_ssa_name (new_scalar_dest, epilog_stmt); gimple_assign_set_lhs (epilog_stmt, new_temp); gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); } extract_scalar_result = false; } } /* 2.4 Extract the final scalar result. Create: s_out3 = extract_field */ if (extract_scalar_result) { tree rhs; gcc_assert (!nested_in_vect_loop); if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "extract scalar result"); if (BYTES_BIG_ENDIAN) bitpos = size_binop (MULT_EXPR, bitsize_int (TYPE_VECTOR_SUBPARTS (vectype) - 1), TYPE_SIZE (scalar_type)); else bitpos = bitsize_zero_node; rhs = build3 (BIT_FIELD_REF, scalar_type, new_temp, bitsize, bitpos); epilog_stmt = gimple_build_assign (new_scalar_dest, rhs); new_temp = make_ssa_name (new_scalar_dest, epilog_stmt); gimple_assign_set_lhs (epilog_stmt, new_temp); gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); } vect_finalize_reduction: /* 2.5 Adjust the final result by the initial value of the reduction variable. (When such adjustment is not needed, then 'adjustment_def' is zero). For example, if code is PLUS we create: new_temp = loop_exit_def + adjustment_def */ if (adjustment_def) { if (nested_in_vect_loop) { gcc_assert (TREE_CODE (TREE_TYPE (adjustment_def)) == VECTOR_TYPE); expr = build2 (code, vectype, PHI_RESULT (new_phi), adjustment_def); new_dest = vect_create_destination_var (scalar_dest, vectype); } else { gcc_assert (TREE_CODE (TREE_TYPE (adjustment_def)) != VECTOR_TYPE); expr = build2 (code, scalar_type, new_temp, adjustment_def); new_dest = vect_create_destination_var (scalar_dest, scalar_type); } epilog_stmt = gimple_build_assign (new_dest, expr); new_temp = make_ssa_name (new_dest, epilog_stmt); gimple_assign_set_lhs (epilog_stmt, new_temp); SSA_NAME_DEF_STMT (new_temp) = epilog_stmt; gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); } /* 2.6 Handle the loop-exit phi */ /* Replace uses of s_out0 with uses of s_out3: Find the loop-closed-use at the loop exit of the original scalar result. (The reduction result is expected to have two immediate uses - one at the latch block, and one at the loop exit). */ phis = VEC_alloc (gimple, heap, 10); FOR_EACH_IMM_USE_FAST (use_p, imm_iter, scalar_dest) { if (!flow_bb_inside_loop_p (loop, gimple_bb (USE_STMT (use_p)))) { exit_phi = USE_STMT (use_p); VEC_quick_push (gimple, phis, exit_phi); } } /* We expect to have found an exit_phi because of loop-closed-ssa form. */ gcc_assert (!VEC_empty (gimple, phis)); for (i = 0; VEC_iterate (gimple, phis, i, exit_phi); i++) { if (nested_in_vect_loop) { stmt_vec_info stmt_vinfo = vinfo_for_stmt (exit_phi); /* FORNOW. Currently not supporting the case that an inner-loop reduction is not used in the outer-loop (but only outside the outer-loop). */ gcc_assert (STMT_VINFO_RELEVANT_P (stmt_vinfo) && !STMT_VINFO_LIVE_P (stmt_vinfo)); epilog_stmt = adjustment_def ? epilog_stmt : new_phi; STMT_VINFO_VEC_STMT (stmt_vinfo) = epilog_stmt; set_vinfo_for_stmt (epilog_stmt, new_stmt_vec_info (epilog_stmt, loop_vinfo)); if (adjustment_def) STMT_VINFO_RELATED_STMT (vinfo_for_stmt (epilog_stmt)) = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (new_phi)); continue; } /* Replace the uses: */ orig_name = PHI_RESULT (exit_phi); FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, orig_name) FOR_EACH_IMM_USE_ON_STMT (use_p, imm_iter) SET_USE (use_p, new_temp); } VEC_free (gimple, heap, phis); } /* Function vectorizable_reduction. Check if STMT performs a reduction operation that can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. This function also handles reduction idioms (patterns) that have been recognized in advance during vect_pattern_recog. In this case, STMT may be of this form: X = pattern_expr (arg0, arg1, ..., X) and it's STMT_VINFO_RELATED_STMT points to the last stmt in the original sequence that had been detected and replaced by the pattern-stmt (STMT). In some cases of reduction patterns, the type of the reduction variable X is different than the type of the other arguments of STMT. In such cases, the vectype that is used when transforming STMT into a vector stmt is different than the vectype that is used to determine the vectorization factor, because it consists of a different number of elements than the actual number of elements that are being operated upon in parallel. For example, consider an accumulation of shorts into an int accumulator. On some targets it's possible to vectorize this pattern operating on 8 shorts at a time (hence, the vectype for purposes of determining the vectorization factor should be V8HI); on the other hand, the vectype that is used to create the vector form is actually V4SI (the type of the result). Upon entry to this function, STMT_VINFO_VECTYPE records the vectype that indicates what is the actual level of parallelism (V8HI in the example), so that the right vectorization factor would be derived. This vectype corresponds to the type of arguments to the reduction stmt, and should *NOT* be used to create the vectorized stmt. The right vectype for the vectorized stmt is obtained from the type of the result X: get_vectype_for_scalar_type (TREE_TYPE (X)) This means that, contrary to "regular" reductions (or "regular" stmts in general), the following equation: STMT_VINFO_VECTYPE == get_vectype_for_scalar_type (TREE_TYPE (X)) does *NOT* necessarily hold for reduction patterns. */ bool vectorizable_reduction (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt) { tree vec_dest; tree scalar_dest; tree loop_vec_def0 = NULL_TREE, loop_vec_def1 = NULL_TREE; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); tree vectype = STMT_VINFO_VECTYPE (stmt_info); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); enum tree_code code, orig_code, epilog_reduc_code = 0; enum machine_mode vec_mode; int op_type; optab optab, reduc_optab; tree new_temp = NULL_TREE; tree def; gimple def_stmt; enum vect_def_type dt; gimple new_phi = NULL; tree scalar_type; bool is_simple_use; gimple orig_stmt; stmt_vec_info orig_stmt_info; tree expr = NULL_TREE; int i; int nunits = TYPE_VECTOR_SUBPARTS (vectype); int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits; int epilog_copies; stmt_vec_info prev_stmt_info, prev_phi_info; gimple first_phi = NULL; bool single_defuse_cycle = false; tree reduc_def; gimple new_stmt = NULL; int j; tree ops[3]; if (nested_in_vect_loop_p (loop, stmt)) loop = loop->inner; gcc_assert (ncopies >= 1); /* FORNOW: SLP not supported. */ if (STMT_SLP_TYPE (stmt_info)) return false; /* 1. Is vectorizable reduction? */ /* Not supportable if the reduction variable is used in the loop. */ if (STMT_VINFO_RELEVANT (stmt_info) > vect_used_in_outer) return false; /* Reductions that are not used even in an enclosing outer-loop, are expected to be "live" (used out of the loop). */ if (STMT_VINFO_RELEVANT (stmt_info) == vect_unused_in_loop && !STMT_VINFO_LIVE_P (stmt_info)) return false; /* Make sure it was already recognized as a reduction computation. */ if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_reduction_def) return false; /* 2. Has this been recognized as a reduction pattern? Check if STMT represents a pattern that has been recognized in earlier analysis stages. For stmts that represent a pattern, the STMT_VINFO_RELATED_STMT field records the last stmt in the original sequence that constitutes the pattern. */ orig_stmt = STMT_VINFO_RELATED_STMT (stmt_info); if (orig_stmt) { orig_stmt_info = vinfo_for_stmt (orig_stmt); gcc_assert (STMT_VINFO_RELATED_STMT (orig_stmt_info) == stmt); gcc_assert (STMT_VINFO_IN_PATTERN_P (orig_stmt_info)); gcc_assert (!STMT_VINFO_IN_PATTERN_P (stmt_info)); } /* 3. Check the operands of the operation. The first operands are defined inside the loop body. The last operand is the reduction variable, which is defined by the loop-header-phi. */ gcc_assert (is_gimple_assign (stmt)); /* Flatten RHS */ switch (get_gimple_rhs_class (gimple_assign_rhs_code (stmt))) { case GIMPLE_SINGLE_RHS: op_type = TREE_OPERAND_LENGTH (gimple_assign_rhs1 (stmt)); if (op_type == ternary_op) { tree rhs = gimple_assign_rhs1 (stmt); ops[0] = TREE_OPERAND (rhs, 0); ops[1] = TREE_OPERAND (rhs, 1); ops[2] = TREE_OPERAND (rhs, 2); code = TREE_CODE (rhs); } else return false; break; case GIMPLE_BINARY_RHS: code = gimple_assign_rhs_code (stmt); op_type = TREE_CODE_LENGTH (code); gcc_assert (op_type == binary_op); ops[0] = gimple_assign_rhs1 (stmt); ops[1] = gimple_assign_rhs2 (stmt); break; case GIMPLE_UNARY_RHS: return false; default: gcc_unreachable (); } scalar_dest = gimple_assign_lhs (stmt); scalar_type = TREE_TYPE (scalar_dest); if (!POINTER_TYPE_P (scalar_type) && !INTEGRAL_TYPE_P (scalar_type) && !SCALAR_FLOAT_TYPE_P (scalar_type)) return false; /* All uses but the last are expected to be defined in the loop. The last use is the reduction variable. */ for (i = 0; i < op_type-1; i++) { is_simple_use = vect_is_simple_use (ops[i], loop_vinfo, &def_stmt, &def, &dt); gcc_assert (is_simple_use); if (dt != vect_loop_def && dt != vect_invariant_def && dt != vect_constant_def && dt != vect_induction_def) return false; } is_simple_use = vect_is_simple_use (ops[i], loop_vinfo, &def_stmt, &def, &dt); gcc_assert (is_simple_use); gcc_assert (dt == vect_reduction_def); gcc_assert (gimple_code (def_stmt) == GIMPLE_PHI); if (orig_stmt) gcc_assert (orig_stmt == vect_is_simple_reduction (loop_vinfo, def_stmt)); else gcc_assert (stmt == vect_is_simple_reduction (loop_vinfo, def_stmt)); if (STMT_VINFO_LIVE_P (vinfo_for_stmt (def_stmt))) return false; /* 4. Supportable by target? */ /* 4.1. check support for the operation in the loop */ optab = optab_for_tree_code (code, vectype, optab_default); if (!optab) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "no optab."); return false; } vec_mode = TYPE_MODE (vectype); if (optab_handler (optab, vec_mode)->insn_code == CODE_FOR_nothing) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "op not supported by target."); if (GET_MODE_SIZE (vec_mode) != UNITS_PER_WORD || LOOP_VINFO_VECT_FACTOR (loop_vinfo) < vect_min_worthwhile_factor (code)) return false; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "proceeding using word mode."); } /* Worthwhile without SIMD support? */ if (!VECTOR_MODE_P (TYPE_MODE (vectype)) && LOOP_VINFO_VECT_FACTOR (loop_vinfo) < vect_min_worthwhile_factor (code)) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "not worthwhile without SIMD support."); return false; } /* 4.2. Check support for the epilog operation. If STMT represents a reduction pattern, then the type of the reduction variable may be different than the type of the rest of the arguments. For example, consider the case of accumulation of shorts into an int accumulator; The original code: S1: int_a = (int) short_a; orig_stmt-> S2: int_acc = plus ; was replaced with: STMT: int_acc = widen_sum This means that: 1. The tree-code that is used to create the vector operation in the epilog code (that reduces the partial results) is not the tree-code of STMT, but is rather the tree-code of the original stmt from the pattern that STMT is replacing. I.e, in the example above we want to use 'widen_sum' in the loop, but 'plus' in the epilog. 2. The type (mode) we use to check available target support for the vector operation to be created in the *epilog*, is determined by the type of the reduction variable (in the example above we'd check this: plus_optab[vect_int_mode]). However the type (mode) we use to check available target support for the vector operation to be created *inside the loop*, is determined by the type of the other arguments to STMT (in the example we'd check this: widen_sum_optab[vect_short_mode]). This is contrary to "regular" reductions, in which the types of all the arguments are the same as the type of the reduction variable. For "regular" reductions we can therefore use the same vector type (and also the same tree-code) when generating the epilog code and when generating the code inside the loop. */ if (orig_stmt) { /* This is a reduction pattern: get the vectype from the type of the reduction variable, and get the tree-code from orig_stmt. */ orig_code = gimple_assign_rhs_code (orig_stmt); vectype = get_vectype_for_scalar_type (TREE_TYPE (def)); if (!vectype) { if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "unsupported data-type "); print_generic_expr (vect_dump, TREE_TYPE (def), TDF_SLIM); } return false; } vec_mode = TYPE_MODE (vectype); } else { /* Regular reduction: use the same vectype and tree-code as used for the vector code inside the loop can be used for the epilog code. */ orig_code = code; } if (!reduction_code_for_scalar_code (orig_code, &epilog_reduc_code)) return false; reduc_optab = optab_for_tree_code (epilog_reduc_code, vectype, optab_default); if (!reduc_optab) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "no optab for reduction."); epilog_reduc_code = NUM_TREE_CODES; } if (optab_handler (reduc_optab, vec_mode)->insn_code == CODE_FOR_nothing) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "reduc op not supported by target."); epilog_reduc_code = NUM_TREE_CODES; } if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = reduc_vec_info_type; if (!vect_model_reduction_cost (stmt_info, epilog_reduc_code, ncopies)) return false; return true; } /** Transform. **/ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform reduction."); /* Create the destination vector */ vec_dest = vect_create_destination_var (scalar_dest, vectype); /* In case the vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits), we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. For more details see documentation in vectorizable_operation. */ /* If the reduction is used in an outer loop we need to generate VF intermediate results, like so (e.g. for ncopies=2): r0 = phi (init, r0) r1 = phi (init, r1) r0 = x0 + r0; r1 = x1 + r1; (i.e. we generate VF results in 2 registers). In this case we have a separate def-use cycle for each copy, and therefore for each copy we get the vector def for the reduction variable from the respective phi node created for this copy. Otherwise (the reduction is unused in the loop nest), we can combine together intermediate results, like so (e.g. for ncopies=2): r = phi (init, r) r = x0 + r; r = x1 + r; (i.e. we generate VF/2 results in a single register). In this case for each copy we get the vector def for the reduction variable from the vectorized reduction operation generated in the previous iteration. */ if (STMT_VINFO_RELEVANT (stmt_info) == vect_unused_in_loop) { single_defuse_cycle = true; epilog_copies = 1; } else epilog_copies = ncopies; prev_stmt_info = NULL; prev_phi_info = NULL; for (j = 0; j < ncopies; j++) { if (j == 0 || !single_defuse_cycle) { /* Create the reduction-phi that defines the reduction-operand. */ new_phi = create_phi_node (vec_dest, loop->header); set_vinfo_for_stmt (new_phi, new_stmt_vec_info (new_phi, loop_vinfo)); } /* Handle uses. */ if (j == 0) { loop_vec_def0 = vect_get_vec_def_for_operand (ops[0], stmt, NULL); if (op_type == ternary_op) { loop_vec_def1 = vect_get_vec_def_for_operand (ops[1], stmt, NULL); } /* Get the vector def for the reduction variable from the phi node */ reduc_def = PHI_RESULT (new_phi); first_phi = new_phi; } else { enum vect_def_type dt = vect_unknown_def_type; /* Dummy */ loop_vec_def0 = vect_get_vec_def_for_stmt_copy (dt, loop_vec_def0); if (op_type == ternary_op) loop_vec_def1 = vect_get_vec_def_for_stmt_copy (dt, loop_vec_def1); if (single_defuse_cycle) reduc_def = gimple_assign_lhs (new_stmt); else reduc_def = PHI_RESULT (new_phi); STMT_VINFO_RELATED_STMT (prev_phi_info) = new_phi; } /* Arguments are ready. create the new vector stmt. */ if (op_type == binary_op) expr = build2 (code, vectype, loop_vec_def0, reduc_def); else expr = build3 (code, vectype, loop_vec_def0, loop_vec_def1, reduc_def); new_stmt = gimple_build_assign (vec_dest, expr); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); if (j == 0) STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt; else STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); prev_phi_info = vinfo_for_stmt (new_phi); } /* Finalize the reduction-phi (set its arguments) and create the epilog reduction code. */ if (!single_defuse_cycle) new_temp = gimple_assign_lhs (*vec_stmt); vect_create_epilog_for_reduction (new_temp, stmt, epilog_copies, epilog_reduc_code, first_phi); return true; } /* Checks if CALL can be vectorized in type VECTYPE. Returns a function declaration if the target has a vectorized version of the function, or NULL_TREE if the function cannot be vectorized. */ tree vectorizable_function (gimple call, tree vectype_out, tree vectype_in) { tree fndecl = gimple_call_fndecl (call); enum built_in_function code; /* We only handle functions that do not read or clobber memory -- i.e. const or novops ones. */ if (!(gimple_call_flags (call) & (ECF_CONST | ECF_NOVOPS))) return NULL_TREE; if (!fndecl || TREE_CODE (fndecl) != FUNCTION_DECL || !DECL_BUILT_IN (fndecl)) return NULL_TREE; code = DECL_FUNCTION_CODE (fndecl); return targetm.vectorize.builtin_vectorized_function (code, vectype_out, vectype_in); } /* Function vectorizable_call. Check if STMT performs a function call that can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_call (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt) { tree vec_dest; tree scalar_dest; tree op, type; tree vec_oprnd0 = NULL_TREE, vec_oprnd1 = NULL_TREE; stmt_vec_info stmt_info = vinfo_for_stmt (stmt), prev_stmt_info; tree vectype_out, vectype_in; int nunits_in; int nunits_out; loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); tree fndecl, new_temp, def, rhs_type, lhs_type; gimple def_stmt; enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type}; gimple new_stmt; int ncopies, j; VEC(tree, heap) *vargs = NULL; enum { NARROW, NONE, WIDEN } modifier; size_t i, nargs; if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def) return false; /* FORNOW: SLP not supported. */ if (STMT_SLP_TYPE (stmt_info)) return false; /* Is STMT a vectorizable call? */ if (!is_gimple_call (stmt)) return false; if (TREE_CODE (gimple_call_lhs (stmt)) != SSA_NAME) return false; /* Process function arguments. */ rhs_type = NULL_TREE; nargs = gimple_call_num_args (stmt); /* Bail out if the function has more than two arguments, we do not have interesting builtin functions to vectorize with more than two arguments. No arguments is also not good. */ if (nargs == 0 || nargs > 2) return false; for (i = 0; i < nargs; i++) { op = gimple_call_arg (stmt, i); /* We can only handle calls with arguments of the same type. */ if (rhs_type && rhs_type != TREE_TYPE (op)) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "argument types differ."); return false; } rhs_type = TREE_TYPE (op); if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt[i])) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } } vectype_in = get_vectype_for_scalar_type (rhs_type); if (!vectype_in) return false; nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in); lhs_type = TREE_TYPE (gimple_call_lhs (stmt)); vectype_out = get_vectype_for_scalar_type (lhs_type); if (!vectype_out) return false; nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out); /* FORNOW */ if (nunits_in == nunits_out / 2) modifier = NARROW; else if (nunits_out == nunits_in) modifier = NONE; else if (nunits_out == nunits_in / 2) modifier = WIDEN; else return false; /* For now, we only vectorize functions if a target specific builtin is available. TODO -- in some cases, it might be profitable to insert the calls for pieces of the vector, in order to be able to vectorize other operations in the loop. */ fndecl = vectorizable_function (stmt, vectype_out, vectype_in); if (fndecl == NULL_TREE) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "function is not vectorizable."); return false; } gcc_assert (ZERO_SSA_OPERANDS (stmt, SSA_OP_ALL_VIRTUALS)); if (modifier == NARROW) ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_out; else ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in; /* Sanity check: make sure that at least one copy of the vectorized stmt needs to be generated. */ gcc_assert (ncopies >= 1); if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = call_vec_info_type; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vectorizable_call ==="); vect_model_simple_cost (stmt_info, ncopies, dt, NULL); return true; } /** Transform. **/ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform operation."); /* Handle def. */ scalar_dest = gimple_call_lhs (stmt); vec_dest = vect_create_destination_var (scalar_dest, vectype_out); prev_stmt_info = NULL; switch (modifier) { case NONE: for (j = 0; j < ncopies; ++j) { /* Build argument list for the vectorized call. */ if (j == 0) vargs = VEC_alloc (tree, heap, nargs); else VEC_truncate (tree, vargs, 0); for (i = 0; i < nargs; i++) { op = gimple_call_arg (stmt, i); if (j == 0) vec_oprnd0 = vect_get_vec_def_for_operand (op, stmt, NULL); else vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd0); VEC_quick_push (tree, vargs, vec_oprnd0); } new_stmt = gimple_build_call_vec (fndecl, vargs); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_call_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); mark_symbols_for_renaming (new_stmt); if (j == 0) STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt; else STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); } break; case NARROW: for (j = 0; j < ncopies; ++j) { /* Build argument list for the vectorized call. */ if (j == 0) vargs = VEC_alloc (tree, heap, nargs * 2); else VEC_truncate (tree, vargs, 0); for (i = 0; i < nargs; i++) { op = gimple_call_arg (stmt, i); if (j == 0) { vec_oprnd0 = vect_get_vec_def_for_operand (op, stmt, NULL); vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd0); } else { vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd1); vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt[nargs], vec_oprnd0); } VEC_quick_push (tree, vargs, vec_oprnd0); VEC_quick_push (tree, vargs, vec_oprnd1); } new_stmt = gimple_build_call_vec (fndecl, vargs); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_call_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); mark_symbols_for_renaming (new_stmt); if (j == 0) STMT_VINFO_VEC_STMT (stmt_info) = new_stmt; else STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); } *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info); break; case WIDEN: /* No current target implements this case. */ return false; } VEC_free (tree, heap, vargs); /* Update the exception handling table with the vector stmt if necessary. */ if (maybe_clean_or_replace_eh_stmt (stmt, *vec_stmt)) gimple_purge_dead_eh_edges (gimple_bb (stmt)); /* The call in STMT might prevent it from being removed in dce. We however cannot remove it here, due to the way the ssa name it defines is mapped to the new definition. So just replace rhs of the statement with something harmless. */ type = TREE_TYPE (scalar_dest); new_stmt = gimple_build_assign (gimple_call_lhs (stmt), fold_convert (type, integer_zero_node)); set_vinfo_for_stmt (new_stmt, stmt_info); set_vinfo_for_stmt (stmt, NULL); STMT_VINFO_STMT (stmt_info) = new_stmt; gsi_replace (gsi, new_stmt, false); SSA_NAME_DEF_STMT (gimple_assign_lhs (new_stmt)) = new_stmt; return true; } /* Function vect_gen_widened_results_half Create a vector stmt whose code, type, number of arguments, and result variable are CODE, OP_TYPE, and VEC_DEST, and its arguments are VEC_OPRND0 and VEC_OPRND1. The new vector stmt is to be inserted at BSI. In the case that CODE is a CALL_EXPR, this means that a call to DECL needs to be created (DECL is a function-decl of a target-builtin). STMT is the original scalar stmt that we are vectorizing. */ static gimple vect_gen_widened_results_half (enum tree_code code, tree decl, tree vec_oprnd0, tree vec_oprnd1, int op_type, tree vec_dest, gimple_stmt_iterator *gsi, gimple stmt) { gimple new_stmt; tree new_temp; tree sym; ssa_op_iter iter; /* Generate half of the widened result: */ if (code == CALL_EXPR) { /* Target specific support */ if (op_type == binary_op) new_stmt = gimple_build_call (decl, 2, vec_oprnd0, vec_oprnd1); else new_stmt = gimple_build_call (decl, 1, vec_oprnd0); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_call_set_lhs (new_stmt, new_temp); } else { /* Generic support */ gcc_assert (op_type == TREE_CODE_LENGTH (code)); if (op_type != binary_op) vec_oprnd1 = NULL; new_stmt = gimple_build_assign_with_ops (code, vec_dest, vec_oprnd0, vec_oprnd1); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_temp); } vect_finish_stmt_generation (stmt, new_stmt, gsi); if (code == CALL_EXPR) { FOR_EACH_SSA_TREE_OPERAND (sym, new_stmt, iter, SSA_OP_ALL_VIRTUALS) { if (TREE_CODE (sym) == SSA_NAME) sym = SSA_NAME_VAR (sym); mark_sym_for_renaming (sym); } } return new_stmt; } /* Check if STMT performs a conversion operation, that can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_conversion (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt, slp_tree slp_node) { tree vec_dest; tree scalar_dest; tree op0; tree vec_oprnd0 = NULL_TREE, vec_oprnd1 = NULL_TREE; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); enum tree_code code, code1 = ERROR_MARK, code2 = ERROR_MARK; tree decl1 = NULL_TREE, decl2 = NULL_TREE; tree new_temp; tree def; gimple def_stmt; enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type}; gimple new_stmt = NULL; stmt_vec_info prev_stmt_info; int nunits_in; int nunits_out; tree vectype_out, vectype_in; int ncopies, j; tree expr; tree rhs_type, lhs_type; tree builtin_decl; enum { NARROW, NONE, WIDEN } modifier; int i; VEC(tree,heap) *vec_oprnds0 = NULL; tree vop0; tree integral_type; VEC(tree,heap) *dummy = NULL; int dummy_int; /* Is STMT a vectorizable conversion? */ if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def) return false; if (!is_gimple_assign (stmt)) return false; if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME) return false; code = gimple_assign_rhs_code (stmt); if (code != FIX_TRUNC_EXPR && code != FLOAT_EXPR) return false; /* Check types of lhs and rhs. */ op0 = gimple_assign_rhs1 (stmt); rhs_type = TREE_TYPE (op0); vectype_in = get_vectype_for_scalar_type (rhs_type); if (!vectype_in) return false; nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in); scalar_dest = gimple_assign_lhs (stmt); lhs_type = TREE_TYPE (scalar_dest); vectype_out = get_vectype_for_scalar_type (lhs_type); if (!vectype_out) return false; nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out); /* FORNOW */ if (nunits_in == nunits_out / 2) modifier = NARROW; else if (nunits_out == nunits_in) modifier = NONE; else if (nunits_out == nunits_in / 2) modifier = WIDEN; else return false; if (modifier == NONE) gcc_assert (STMT_VINFO_VECTYPE (stmt_info) == vectype_out); /* Bail out if the types are both integral or non-integral. */ if ((INTEGRAL_TYPE_P (rhs_type) && INTEGRAL_TYPE_P (lhs_type)) || (!INTEGRAL_TYPE_P (rhs_type) && !INTEGRAL_TYPE_P (lhs_type))) return false; integral_type = INTEGRAL_TYPE_P (rhs_type) ? vectype_in : vectype_out; if (modifier == NARROW) ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_out; else ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in; /* FORNOW: SLP with multiple types is not supported. The SLP analysis verifies this, so we can safely override NCOPIES with 1 here. */ if (slp_node) ncopies = 1; /* Sanity check: make sure that at least one copy of the vectorized stmt needs to be generated. */ gcc_assert (ncopies >= 1); /* Check the operands of the operation. */ if (!vect_is_simple_use (op0, loop_vinfo, &def_stmt, &def, &dt[0])) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } /* Supportable by target? */ if ((modifier == NONE && !targetm.vectorize.builtin_conversion (code, integral_type)) || (modifier == WIDEN && !supportable_widening_operation (code, stmt, vectype_in, &decl1, &decl2, &code1, &code2, &dummy_int, &dummy)) || (modifier == NARROW && !supportable_narrowing_operation (code, stmt, vectype_in, &code1, &dummy_int, &dummy))) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "conversion not supported by target."); return false; } if (modifier != NONE) { STMT_VINFO_VECTYPE (stmt_info) = vectype_in; /* FORNOW: SLP not supported. */ if (STMT_SLP_TYPE (stmt_info)) return false; } if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = type_conversion_vec_info_type; return true; } /** Transform. **/ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform conversion."); /* Handle def. */ vec_dest = vect_create_destination_var (scalar_dest, vectype_out); if (modifier == NONE && !slp_node) vec_oprnds0 = VEC_alloc (tree, heap, 1); prev_stmt_info = NULL; switch (modifier) { case NONE: for (j = 0; j < ncopies; j++) { tree sym; ssa_op_iter iter; if (j == 0) vect_get_vec_defs (op0, NULL, stmt, &vec_oprnds0, NULL, slp_node); else vect_get_vec_defs_for_stmt_copy (dt, &vec_oprnds0, NULL); builtin_decl = targetm.vectorize.builtin_conversion (code, integral_type); for (i = 0; VEC_iterate (tree, vec_oprnds0, i, vop0); i++) { /* Arguments are ready. create the new vector stmt. */ new_stmt = gimple_build_call (builtin_decl, 1, vop0); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_call_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); FOR_EACH_SSA_TREE_OPERAND (sym, new_stmt, iter, SSA_OP_ALL_VIRTUALS) { if (TREE_CODE (sym) == SSA_NAME) sym = SSA_NAME_VAR (sym); mark_sym_for_renaming (sym); } if (slp_node) VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt); } if (j == 0) STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt; else STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); } break; case WIDEN: /* In case the vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits), we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. */ for (j = 0; j < ncopies; j++) { if (j == 0) vec_oprnd0 = vect_get_vec_def_for_operand (op0, stmt, NULL); else vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd0); STMT_VINFO_VECTYPE (stmt_info) = vectype_in; /* Generate first half of the widened result: */ new_stmt = vect_gen_widened_results_half (code1, decl1, vec_oprnd0, vec_oprnd1, unary_op, vec_dest, gsi, stmt); if (j == 0) STMT_VINFO_VEC_STMT (stmt_info) = new_stmt; else STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); /* Generate second half of the widened result: */ new_stmt = vect_gen_widened_results_half (code2, decl2, vec_oprnd0, vec_oprnd1, unary_op, vec_dest, gsi, stmt); STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); } break; case NARROW: /* In case the vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits), we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. */ for (j = 0; j < ncopies; j++) { /* Handle uses. */ if (j == 0) { vec_oprnd0 = vect_get_vec_def_for_operand (op0, stmt, NULL); vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd0); } else { vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd1); vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd0); } /* Arguments are ready. Create the new vector stmt. */ expr = build2 (code1, vectype_out, vec_oprnd0, vec_oprnd1); new_stmt = gimple_build_assign_with_ops (code1, vec_dest, vec_oprnd0, vec_oprnd1); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); if (j == 0) STMT_VINFO_VEC_STMT (stmt_info) = new_stmt; else STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); } *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info); } if (vec_oprnds0) VEC_free (tree, heap, vec_oprnds0); return true; } /* Function vectorizable_assignment. Check if STMT performs an assignment (copy) that can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_assignment (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt, slp_tree slp_node) { tree vec_dest; tree scalar_dest; tree op; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); tree vectype = STMT_VINFO_VECTYPE (stmt_info); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); tree new_temp; tree def; gimple def_stmt; enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type}; int nunits = TYPE_VECTOR_SUBPARTS (vectype); int ncopies; int i; VEC(tree,heap) *vec_oprnds = NULL; tree vop; /* Multiple types in SLP are handled by creating the appropriate number of vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in case of SLP. */ if (slp_node) ncopies = 1; else ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits; gcc_assert (ncopies >= 1); if (ncopies > 1) return false; /* FORNOW */ if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def) return false; /* Is vectorizable assignment? */ if (!is_gimple_assign (stmt)) return false; scalar_dest = gimple_assign_lhs (stmt); if (TREE_CODE (scalar_dest) != SSA_NAME) return false; if (gimple_assign_single_p (stmt) || gimple_assign_rhs_code (stmt) == PAREN_EXPR) op = gimple_assign_rhs1 (stmt); else return false; if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt[0])) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = assignment_vec_info_type; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vectorizable_assignment ==="); vect_model_simple_cost (stmt_info, ncopies, dt, NULL); return true; } /** Transform. **/ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform assignment."); /* Handle def. */ vec_dest = vect_create_destination_var (scalar_dest, vectype); /* Handle use. */ vect_get_vec_defs (op, NULL, stmt, &vec_oprnds, NULL, slp_node); /* Arguments are ready. create the new vector stmt. */ for (i = 0; VEC_iterate (tree, vec_oprnds, i, vop); i++) { *vec_stmt = gimple_build_assign (vec_dest, vop); new_temp = make_ssa_name (vec_dest, *vec_stmt); gimple_assign_set_lhs (*vec_stmt, new_temp); vect_finish_stmt_generation (stmt, *vec_stmt, gsi); STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt; if (slp_node) VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), *vec_stmt); } VEC_free (tree, heap, vec_oprnds); return true; } /* Function vect_min_worthwhile_factor. For a loop where we could vectorize the operation indicated by CODE, return the minimum vectorization factor that makes it worthwhile to use generic vectors. */ static int vect_min_worthwhile_factor (enum tree_code code) { switch (code) { case PLUS_EXPR: case MINUS_EXPR: case NEGATE_EXPR: return 4; case BIT_AND_EXPR: case BIT_IOR_EXPR: case BIT_XOR_EXPR: case BIT_NOT_EXPR: return 2; default: return INT_MAX; } } /* Function vectorizable_induction Check if PHI performs an induction computation that can be vectorized. If VEC_STMT is also passed, vectorize the induction PHI: create a vectorized phi to replace it, put it in VEC_STMT, and add it to the same basic block. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_induction (gimple phi, gimple_stmt_iterator *gsi ATTRIBUTE_UNUSED, gimple *vec_stmt) { stmt_vec_info stmt_info = vinfo_for_stmt (phi); tree vectype = STMT_VINFO_VECTYPE (stmt_info); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); int nunits = TYPE_VECTOR_SUBPARTS (vectype); int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits; tree vec_def; gcc_assert (ncopies >= 1); /* FORNOW. This restriction should be relaxed. */ if (nested_in_vect_loop_p (loop, phi) && ncopies > 1) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "multiple types in nested loop."); return false; } if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; /* FORNOW: SLP not supported. */ if (STMT_SLP_TYPE (stmt_info)) return false; gcc_assert (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def); if (gimple_code (phi) != GIMPLE_PHI) return false; if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = induc_vec_info_type; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vectorizable_induction ==="); vect_model_induction_cost (stmt_info, ncopies); return true; } /** Transform. **/ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform induction phi."); vec_def = get_initial_def_for_induction (phi); *vec_stmt = SSA_NAME_DEF_STMT (vec_def); return true; } /* Function vectorizable_operation. Check if STMT performs a binary or unary operation that can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_operation (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt, slp_tree slp_node) { tree vec_dest; tree scalar_dest; tree op0, op1 = NULL; tree vec_oprnd1 = NULL_TREE; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); tree vectype = STMT_VINFO_VECTYPE (stmt_info); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); enum tree_code code; enum machine_mode vec_mode; tree new_temp; int op_type; optab optab; int icode; enum machine_mode optab_op2_mode; tree def; gimple def_stmt; enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type}; gimple new_stmt = NULL; stmt_vec_info prev_stmt_info; int nunits_in = TYPE_VECTOR_SUBPARTS (vectype); int nunits_out; tree vectype_out; int ncopies; int j, i; VEC(tree,heap) *vec_oprnds0 = NULL, *vec_oprnds1 = NULL; tree vop0, vop1; unsigned int k; bool shift_p = false; bool scalar_shift_arg = false; /* Multiple types in SLP are handled by creating the appropriate number of vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in case of SLP. */ if (slp_node) ncopies = 1; else ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in; gcc_assert (ncopies >= 1); if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def) return false; /* Is STMT a vectorizable binary/unary operation? */ if (!is_gimple_assign (stmt)) return false; if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME) return false; scalar_dest = gimple_assign_lhs (stmt); vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest)); if (!vectype_out) return false; nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out); if (nunits_out != nunits_in) return false; code = gimple_assign_rhs_code (stmt); /* For pointer addition, we should use the normal plus for the vector addition. */ if (code == POINTER_PLUS_EXPR) code = PLUS_EXPR; /* Support only unary or binary operations. */ op_type = TREE_CODE_LENGTH (code); if (op_type != unary_op && op_type != binary_op) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "num. args = %d (not unary/binary op).", op_type); return false; } op0 = gimple_assign_rhs1 (stmt); if (!vect_is_simple_use (op0, loop_vinfo, &def_stmt, &def, &dt[0])) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } if (op_type == binary_op) { op1 = gimple_assign_rhs2 (stmt); if (!vect_is_simple_use (op1, loop_vinfo, &def_stmt, &def, &dt[1])) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } } /* If this is a shift/rotate, determine whether the shift amount is a vector, or scalar. If the shift/rotate amount is a vector, use the vector/vector shift optabs. */ if (code == LSHIFT_EXPR || code == RSHIFT_EXPR || code == LROTATE_EXPR || code == RROTATE_EXPR) { shift_p = true; /* vector shifted by vector */ if (dt[1] == vect_loop_def) { optab = optab_for_tree_code (code, vectype, optab_vector); if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "vector/vector shift/rotate found."); } /* See if the machine has a vector shifted by scalar insn and if not then see if it has a vector shifted by vector insn */ else if (dt[1] == vect_constant_def || dt[1] == vect_invariant_def) { optab = optab_for_tree_code (code, vectype, optab_scalar); if (optab && (optab_handler (optab, TYPE_MODE (vectype))->insn_code != CODE_FOR_nothing)) { scalar_shift_arg = true; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "vector/scalar shift/rotate found."); } else { optab = optab_for_tree_code (code, vectype, optab_vector); if (vect_print_dump_info (REPORT_DETAILS) && optab && (optab_handler (optab, TYPE_MODE (vectype))->insn_code != CODE_FOR_nothing)) fprintf (vect_dump, "vector/vector shift/rotate found."); } } else { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "operand mode requires invariant argument."); return false; } } else optab = optab_for_tree_code (code, vectype, optab_default); /* Supportable by target? */ if (!optab) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "no optab."); return false; } vec_mode = TYPE_MODE (vectype); icode = (int) optab_handler (optab, vec_mode)->insn_code; if (icode == CODE_FOR_nothing) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "op not supported by target."); /* Check only during analysis. */ if (GET_MODE_SIZE (vec_mode) != UNITS_PER_WORD || (LOOP_VINFO_VECT_FACTOR (loop_vinfo) < vect_min_worthwhile_factor (code) && !vec_stmt)) return false; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "proceeding using word mode."); } /* Worthwhile without SIMD support? Check only during analysis. */ if (!VECTOR_MODE_P (TYPE_MODE (vectype)) && LOOP_VINFO_VECT_FACTOR (loop_vinfo) < vect_min_worthwhile_factor (code) && !vec_stmt) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "not worthwhile without SIMD support."); return false; } if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = op_vec_info_type; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vectorizable_operation ==="); vect_model_simple_cost (stmt_info, ncopies, dt, NULL); return true; } /** Transform. **/ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform binary/unary operation."); /* Handle def. */ vec_dest = vect_create_destination_var (scalar_dest, vectype); /* Allocate VECs for vector operands. In case of SLP, vector operands are created in the previous stages of the recursion, so no allocation is needed, except for the case of shift with scalar shift argument. In that case we store the scalar operand in VEC_OPRNDS1 for every vector stmt to be created to vectorize the SLP group, i.e., SLP_NODE->VEC_STMTS_SIZE. In case of loop-based vectorization we allocate VECs of size 1. We allocate VEC_OPRNDS1 only in case of binary operation. */ if (!slp_node) { vec_oprnds0 = VEC_alloc (tree, heap, 1); if (op_type == binary_op) vec_oprnds1 = VEC_alloc (tree, heap, 1); } else if (scalar_shift_arg) vec_oprnds1 = VEC_alloc (tree, heap, slp_node->vec_stmts_size); /* In case the vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits), we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. In doing so, we record a pointer from one copy of the vector stmt to the next, in the field STMT_VINFO_RELATED_STMT. This is necessary in order to allow following stages to find the correct vector defs to be used when vectorizing stmts that use the defs of the current stmt. The example below illustrates the vectorization process when VF=16 and nunits=4 (i.e - we need to create 4 vectorized stmts): before vectorization: RELATED_STMT VEC_STMT S1: x = memref - - S2: z = x + 1 - - step 1: vectorize stmt S1 (done in vectorizable_load. See more details there): RELATED_STMT VEC_STMT VS1_0: vx0 = memref0 VS1_1 - VS1_1: vx1 = memref1 VS1_2 - VS1_2: vx2 = memref2 VS1_3 - VS1_3: vx3 = memref3 - - S1: x = load - VS1_0 S2: z = x + 1 - - step2: vectorize stmt S2 (done here): To vectorize stmt S2 we first need to find the relevant vector def for the first operand 'x'. This is, as usual, obtained from the vector stmt recorded in the STMT_VINFO_VEC_STMT of the stmt that defines 'x' (S1). This way we find the stmt VS1_0, and the relevant vector def 'vx0'. Having found 'vx0' we can generate the vector stmt VS2_0, and as usual, record it in the STMT_VINFO_VEC_STMT of stmt S2. When creating the second copy (VS2_1), we obtain the relevant vector def from the vector stmt recorded in the STMT_VINFO_RELATED_STMT of stmt VS1_0. This way we find the stmt VS1_1 and the relevant vector def 'vx1'. Using 'vx1' we create stmt VS2_1 and record a pointer to it in the STMT_VINFO_RELATED_STMT of the vector stmt VS2_0. Similarly when creating stmts VS2_2 and VS2_3. This is the resulting chain of stmts and pointers: RELATED_STMT VEC_STMT VS1_0: vx0 = memref0 VS1_1 - VS1_1: vx1 = memref1 VS1_2 - VS1_2: vx2 = memref2 VS1_3 - VS1_3: vx3 = memref3 - - S1: x = load - VS1_0 VS2_0: vz0 = vx0 + v1 VS2_1 - VS2_1: vz1 = vx1 + v1 VS2_2 - VS2_2: vz2 = vx2 + v1 VS2_3 - VS2_3: vz3 = vx3 + v1 - - S2: z = x + 1 - VS2_0 */ prev_stmt_info = NULL; for (j = 0; j < ncopies; j++) { /* Handle uses. */ if (j == 0) { if (op_type == binary_op && scalar_shift_arg) { /* Vector shl and shr insn patterns can be defined with scalar operand 2 (shift operand). In this case, use constant or loop invariant op1 directly, without extending it to vector mode first. */ optab_op2_mode = insn_data[icode].operand[2].mode; if (!VECTOR_MODE_P (optab_op2_mode)) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "operand 1 using scalar mode."); vec_oprnd1 = op1; VEC_quick_push (tree, vec_oprnds1, vec_oprnd1); if (slp_node) { /* Store vec_oprnd1 for every vector stmt to be created for SLP_NODE. We check during the analysis that all the shift arguments are the same. TODO: Allow different constants for different vector stmts generated for an SLP instance. */ for (k = 0; k < slp_node->vec_stmts_size - 1; k++) VEC_quick_push (tree, vec_oprnds1, vec_oprnd1); } } } /* vec_oprnd1 is available if operand 1 should be of a scalar-type (a special case for certain kind of vector shifts); otherwise, operand 1 should be of a vector type (the usual case). */ if (op_type == binary_op && !vec_oprnd1) vect_get_vec_defs (op0, op1, stmt, &vec_oprnds0, &vec_oprnds1, slp_node); else vect_get_vec_defs (op0, NULL_TREE, stmt, &vec_oprnds0, NULL, slp_node); } else vect_get_vec_defs_for_stmt_copy (dt, &vec_oprnds0, &vec_oprnds1); /* Arguments are ready. Create the new vector stmt. */ for (i = 0; VEC_iterate (tree, vec_oprnds0, i, vop0); i++) { vop1 = ((op_type == binary_op) ? VEC_index (tree, vec_oprnds1, i) : NULL); new_stmt = gimple_build_assign_with_ops (code, vec_dest, vop0, vop1); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); if (slp_node) VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt); } if (slp_node) continue; if (j == 0) STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt; else STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); } VEC_free (tree, heap, vec_oprnds0); if (vec_oprnds1) VEC_free (tree, heap, vec_oprnds1); return true; } /* Get vectorized definitions for loop-based vectorization. For the first operand we call vect_get_vec_def_for_operand() (with OPRND containing scalar operand), and for the rest we get a copy with vect_get_vec_def_for_stmt_copy() using the previous vector definition (stored in OPRND). See vect_get_vec_def_for_stmt_copy() for details. The vectors are collected into VEC_OPRNDS. */ static void vect_get_loop_based_defs (tree *oprnd, gimple stmt, enum vect_def_type dt, VEC (tree, heap) **vec_oprnds, int multi_step_cvt) { tree vec_oprnd; /* Get first vector operand. */ /* All the vector operands except the very first one (that is scalar oprnd) are stmt copies. */ if (TREE_CODE (TREE_TYPE (*oprnd)) != VECTOR_TYPE) vec_oprnd = vect_get_vec_def_for_operand (*oprnd, stmt, NULL); else vec_oprnd = vect_get_vec_def_for_stmt_copy (dt, *oprnd); VEC_quick_push (tree, *vec_oprnds, vec_oprnd); /* Get second vector operand. */ vec_oprnd = vect_get_vec_def_for_stmt_copy (dt, vec_oprnd); VEC_quick_push (tree, *vec_oprnds, vec_oprnd); *oprnd = vec_oprnd; /* For conversion in multiple steps, continue to get operands recursively. */ if (multi_step_cvt) vect_get_loop_based_defs (oprnd, stmt, dt, vec_oprnds, multi_step_cvt - 1); } /* Create vectorized demotion statements for vector operands from VEC_OPRNDS. For multi-step conversions store the resulting vectors and call the function recursively. */ static void vect_create_vectorized_demotion_stmts (VEC (tree, heap) **vec_oprnds, int multi_step_cvt, gimple stmt, VEC (tree, heap) *vec_dsts, gimple_stmt_iterator *gsi, slp_tree slp_node, enum tree_code code, stmt_vec_info *prev_stmt_info) { unsigned int i; tree vop0, vop1, new_tmp, vec_dest; gimple new_stmt; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); vec_dest = VEC_pop (tree, vec_dsts); for (i = 0; i < VEC_length (tree, *vec_oprnds); i += 2) { /* Create demotion operation. */ vop0 = VEC_index (tree, *vec_oprnds, i); vop1 = VEC_index (tree, *vec_oprnds, i + 1); new_stmt = gimple_build_assign_with_ops (code, vec_dest, vop0, vop1); new_tmp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_tmp); vect_finish_stmt_generation (stmt, new_stmt, gsi); if (multi_step_cvt) /* Store the resulting vector for next recursive call. */ VEC_replace (tree, *vec_oprnds, i/2, new_tmp); else { /* This is the last step of the conversion sequence. Store the vectors in SLP_NODE or in vector info of the scalar statement (or in STMT_VINFO_RELATED_STMT chain). */ if (slp_node) VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt); else { if (!*prev_stmt_info) STMT_VINFO_VEC_STMT (stmt_info) = new_stmt; else STMT_VINFO_RELATED_STMT (*prev_stmt_info) = new_stmt; *prev_stmt_info = vinfo_for_stmt (new_stmt); } } } /* For multi-step demotion operations we first generate demotion operations from the source type to the intermediate types, and then combine the results (stored in VEC_OPRNDS) in demotion operation to the destination type. */ if (multi_step_cvt) { /* At each level of recursion we have have of the operands we had at the previous level. */ VEC_truncate (tree, *vec_oprnds, (i+1)/2); vect_create_vectorized_demotion_stmts (vec_oprnds, multi_step_cvt - 1, stmt, vec_dsts, gsi, slp_node, code, prev_stmt_info); } } /* Function vectorizable_type_demotion Check if STMT performs a binary or unary operation that involves type demotion, and if it can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_type_demotion (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt, slp_tree slp_node) { tree vec_dest; tree scalar_dest; tree op0; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); enum tree_code code, code1 = ERROR_MARK; tree def; gimple def_stmt; enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type}; stmt_vec_info prev_stmt_info; int nunits_in; int nunits_out; tree vectype_out; int ncopies; int j, i; tree vectype_in; int multi_step_cvt = 0; VEC (tree, heap) *vec_oprnds0 = NULL; VEC (tree, heap) *vec_dsts = NULL, *interm_types = NULL, *tmp_vec_dsts = NULL; tree last_oprnd, intermediate_type; if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def) return false; /* Is STMT a vectorizable type-demotion operation? */ if (!is_gimple_assign (stmt)) return false; if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME) return false; code = gimple_assign_rhs_code (stmt); if (!CONVERT_EXPR_CODE_P (code)) return false; op0 = gimple_assign_rhs1 (stmt); vectype_in = get_vectype_for_scalar_type (TREE_TYPE (op0)); if (!vectype_in) return false; nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in); scalar_dest = gimple_assign_lhs (stmt); vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest)); if (!vectype_out) return false; nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out); if (nunits_in >= nunits_out) return false; /* Multiple types in SLP are handled by creating the appropriate number of vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in case of SLP. */ if (slp_node) ncopies = 1; else ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_out; gcc_assert (ncopies >= 1); if (! ((INTEGRAL_TYPE_P (TREE_TYPE (scalar_dest)) && INTEGRAL_TYPE_P (TREE_TYPE (op0))) || (SCALAR_FLOAT_TYPE_P (TREE_TYPE (scalar_dest)) && SCALAR_FLOAT_TYPE_P (TREE_TYPE (op0)) && CONVERT_EXPR_CODE_P (code)))) return false; /* Check the operands of the operation. */ if (!vect_is_simple_use (op0, loop_vinfo, &def_stmt, &def, &dt[0])) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } /* Supportable by target? */ if (!supportable_narrowing_operation (code, stmt, vectype_in, &code1, &multi_step_cvt, &interm_types)) return false; STMT_VINFO_VECTYPE (stmt_info) = vectype_in; if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = type_demotion_vec_info_type; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vectorizable_demotion ==="); vect_model_simple_cost (stmt_info, ncopies, dt, NULL); return true; } /** Transform. **/ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform type demotion operation. ncopies = %d.", ncopies); /* In case of multi-step demotion, we first generate demotion operations to the intermediate types, and then from that types to the final one. We create vector destinations for the intermediate type (TYPES) received from supportable_narrowing_operation, and store them in the correct order for future use in vect_create_vectorized_demotion_stmts(). */ if (multi_step_cvt) vec_dsts = VEC_alloc (tree, heap, multi_step_cvt + 1); else vec_dsts = VEC_alloc (tree, heap, 1); vec_dest = vect_create_destination_var (scalar_dest, vectype_out); VEC_quick_push (tree, vec_dsts, vec_dest); if (multi_step_cvt) { for (i = VEC_length (tree, interm_types) - 1; VEC_iterate (tree, interm_types, i, intermediate_type); i--) { vec_dest = vect_create_destination_var (scalar_dest, intermediate_type); VEC_quick_push (tree, vec_dsts, vec_dest); } } /* In case the vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits), we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. */ last_oprnd = op0; prev_stmt_info = NULL; for (j = 0; j < ncopies; j++) { /* Handle uses. */ if (slp_node) vect_get_slp_defs (slp_node, &vec_oprnds0, NULL); else { VEC_free (tree, heap, vec_oprnds0); vec_oprnds0 = VEC_alloc (tree, heap, (multi_step_cvt ? vect_pow2 (multi_step_cvt) * 2 : 2)); vect_get_loop_based_defs (&last_oprnd, stmt, dt[0], &vec_oprnds0, vect_pow2 (multi_step_cvt) - 1); } /* Arguments are ready. Create the new vector stmts. */ tmp_vec_dsts = VEC_copy (tree, heap, vec_dsts); vect_create_vectorized_demotion_stmts (&vec_oprnds0, multi_step_cvt, stmt, tmp_vec_dsts, gsi, slp_node, code1, &prev_stmt_info); } VEC_free (tree, heap, vec_oprnds0); VEC_free (tree, heap, vec_dsts); VEC_free (tree, heap, tmp_vec_dsts); VEC_free (tree, heap, interm_types); *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info); return true; } /* Create vectorized promotion statements for vector operands from VEC_OPRNDS0 and VEC_OPRNDS1 (for binary operations). For multi-step conversions store the resulting vectors and call the function recursively. */ static void vect_create_vectorized_promotion_stmts (VEC (tree, heap) **vec_oprnds0, VEC (tree, heap) **vec_oprnds1, int multi_step_cvt, gimple stmt, VEC (tree, heap) *vec_dsts, gimple_stmt_iterator *gsi, slp_tree slp_node, enum tree_code code1, enum tree_code code2, tree decl1, tree decl2, int op_type, stmt_vec_info *prev_stmt_info) { int i; tree vop0, vop1, new_tmp1, new_tmp2, vec_dest; gimple new_stmt1, new_stmt2; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); VEC (tree, heap) *vec_tmp; vec_dest = VEC_pop (tree, vec_dsts); vec_tmp = VEC_alloc (tree, heap, VEC_length (tree, *vec_oprnds0) * 2); for (i = 0; VEC_iterate (tree, *vec_oprnds0, i, vop0); i++) { if (op_type == binary_op) vop1 = VEC_index (tree, *vec_oprnds1, i); else vop1 = NULL_TREE; /* Generate the two halves of promotion operation. */ new_stmt1 = vect_gen_widened_results_half (code1, decl1, vop0, vop1, op_type, vec_dest, gsi, stmt); new_stmt2 = vect_gen_widened_results_half (code2, decl2, vop0, vop1, op_type, vec_dest, gsi, stmt); if (is_gimple_call (new_stmt1)) { new_tmp1 = gimple_call_lhs (new_stmt1); new_tmp2 = gimple_call_lhs (new_stmt2); } else { new_tmp1 = gimple_assign_lhs (new_stmt1); new_tmp2 = gimple_assign_lhs (new_stmt2); } if (multi_step_cvt) { /* Store the results for the recursive call. */ VEC_quick_push (tree, vec_tmp, new_tmp1); VEC_quick_push (tree, vec_tmp, new_tmp2); } else { /* Last step of promotion sequience - store the results. */ if (slp_node) { VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt1); VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt2); } else { if (!*prev_stmt_info) STMT_VINFO_VEC_STMT (stmt_info) = new_stmt1; else STMT_VINFO_RELATED_STMT (*prev_stmt_info) = new_stmt1; *prev_stmt_info = vinfo_for_stmt (new_stmt1); STMT_VINFO_RELATED_STMT (*prev_stmt_info) = new_stmt2; *prev_stmt_info = vinfo_for_stmt (new_stmt2); } } } if (multi_step_cvt) { /* For multi-step promotion operation we first generate we call the function recurcively for every stage. We start from the input type, create promotion operations to the intermediate types, and then create promotions to the output type. */ *vec_oprnds0 = VEC_copy (tree, heap, vec_tmp); VEC_free (tree, heap, vec_tmp); vect_create_vectorized_promotion_stmts (vec_oprnds0, vec_oprnds1, multi_step_cvt - 1, stmt, vec_dsts, gsi, slp_node, code1, code2, decl2, decl2, op_type, prev_stmt_info); } } /* Function vectorizable_type_promotion Check if STMT performs a binary or unary operation that involves type promotion, and if it can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_type_promotion (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt, slp_tree slp_node) { tree vec_dest; tree scalar_dest; tree op0, op1 = NULL; tree vec_oprnd0=NULL, vec_oprnd1=NULL; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); enum tree_code code, code1 = ERROR_MARK, code2 = ERROR_MARK; tree decl1 = NULL_TREE, decl2 = NULL_TREE; int op_type; tree def; gimple def_stmt; enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type}; stmt_vec_info prev_stmt_info; int nunits_in; int nunits_out; tree vectype_out; int ncopies; int j, i; tree vectype_in; tree intermediate_type = NULL_TREE; int multi_step_cvt = 0; VEC (tree, heap) *vec_oprnds0 = NULL, *vec_oprnds1 = NULL; VEC (tree, heap) *vec_dsts = NULL, *interm_types = NULL, *tmp_vec_dsts = NULL; if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def) return false; /* Is STMT a vectorizable type-promotion operation? */ if (!is_gimple_assign (stmt)) return false; if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME) return false; code = gimple_assign_rhs_code (stmt); if (!CONVERT_EXPR_CODE_P (code) && code != WIDEN_MULT_EXPR) return false; op0 = gimple_assign_rhs1 (stmt); vectype_in = get_vectype_for_scalar_type (TREE_TYPE (op0)); if (!vectype_in) return false; nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in); scalar_dest = gimple_assign_lhs (stmt); vectype_out = get_vectype_for_scalar_type (TREE_TYPE (scalar_dest)); if (!vectype_out) return false; nunits_out = TYPE_VECTOR_SUBPARTS (vectype_out); if (nunits_in <= nunits_out) return false; /* Multiple types in SLP are handled by creating the appropriate number of vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in case of SLP. */ if (slp_node) ncopies = 1; else ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits_in; gcc_assert (ncopies >= 1); if (! ((INTEGRAL_TYPE_P (TREE_TYPE (scalar_dest)) && INTEGRAL_TYPE_P (TREE_TYPE (op0))) || (SCALAR_FLOAT_TYPE_P (TREE_TYPE (scalar_dest)) && SCALAR_FLOAT_TYPE_P (TREE_TYPE (op0)) && CONVERT_EXPR_CODE_P (code)))) return false; /* Check the operands of the operation. */ if (!vect_is_simple_use (op0, loop_vinfo, &def_stmt, &def, &dt[0])) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } op_type = TREE_CODE_LENGTH (code); if (op_type == binary_op) { op1 = gimple_assign_rhs2 (stmt); if (!vect_is_simple_use (op1, loop_vinfo, &def_stmt, &def, &dt[1])) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } } /* Supportable by target? */ if (!supportable_widening_operation (code, stmt, vectype_in, &decl1, &decl2, &code1, &code2, &multi_step_cvt, &interm_types)) return false; /* Binary widening operation can only be supported directly by the architecture. */ gcc_assert (!(multi_step_cvt && op_type == binary_op)); STMT_VINFO_VECTYPE (stmt_info) = vectype_in; if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = type_promotion_vec_info_type; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vectorizable_promotion ==="); vect_model_simple_cost (stmt_info, 2*ncopies, dt, NULL); return true; } /** Transform. **/ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform type promotion operation. ncopies = %d.", ncopies); /* Handle def. */ /* In case of multi-step promotion, we first generate promotion operations to the intermediate types, and then from that types to the final one. We store vector destination in VEC_DSTS in the correct order for recursive creation of promotion operations in vect_create_vectorized_promotion_stmts(). Vector destinations are created according to TYPES recieved from supportable_widening_operation(). */ if (multi_step_cvt) vec_dsts = VEC_alloc (tree, heap, multi_step_cvt + 1); else vec_dsts = VEC_alloc (tree, heap, 1); vec_dest = vect_create_destination_var (scalar_dest, vectype_out); VEC_quick_push (tree, vec_dsts, vec_dest); if (multi_step_cvt) { for (i = VEC_length (tree, interm_types) - 1; VEC_iterate (tree, interm_types, i, intermediate_type); i--) { vec_dest = vect_create_destination_var (scalar_dest, intermediate_type); VEC_quick_push (tree, vec_dsts, vec_dest); } } if (!slp_node) { vec_oprnds0 = VEC_alloc (tree, heap, (multi_step_cvt ? vect_pow2 (multi_step_cvt) : 1)); if (op_type == binary_op) vec_oprnds1 = VEC_alloc (tree, heap, 1); } /* In case the vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits), we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. */ prev_stmt_info = NULL; for (j = 0; j < ncopies; j++) { /* Handle uses. */ if (j == 0) { if (slp_node) vect_get_slp_defs (slp_node, &vec_oprnds0, &vec_oprnds1); else { vec_oprnd0 = vect_get_vec_def_for_operand (op0, stmt, NULL); VEC_quick_push (tree, vec_oprnds0, vec_oprnd0); if (op_type == binary_op) { vec_oprnd1 = vect_get_vec_def_for_operand (op1, stmt, NULL); VEC_quick_push (tree, vec_oprnds1, vec_oprnd1); } } } else { vec_oprnd0 = vect_get_vec_def_for_stmt_copy (dt[0], vec_oprnd0); VEC_replace (tree, vec_oprnds0, 0, vec_oprnd0); if (op_type == binary_op) { vec_oprnd1 = vect_get_vec_def_for_stmt_copy (dt[1], vec_oprnd1); VEC_replace (tree, vec_oprnds1, 0, vec_oprnd1); } } /* Arguments are ready. Create the new vector stmts. */ tmp_vec_dsts = VEC_copy (tree, heap, vec_dsts); vect_create_vectorized_promotion_stmts (&vec_oprnds0, &vec_oprnds1, multi_step_cvt, stmt, tmp_vec_dsts, gsi, slp_node, code1, code2, decl1, decl2, op_type, &prev_stmt_info); } VEC_free (tree, heap, vec_dsts); VEC_free (tree, heap, tmp_vec_dsts); VEC_free (tree, heap, interm_types); VEC_free (tree, heap, vec_oprnds0); VEC_free (tree, heap, vec_oprnds1); *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info); return true; } /* Function vect_strided_store_supported. Returns TRUE is INTERLEAVE_HIGH and INTERLEAVE_LOW operations are supported, and FALSE otherwise. */ static bool vect_strided_store_supported (tree vectype) { optab interleave_high_optab, interleave_low_optab; int mode; mode = (int) TYPE_MODE (vectype); /* Check that the operation is supported. */ interleave_high_optab = optab_for_tree_code (VEC_INTERLEAVE_HIGH_EXPR, vectype, optab_default); interleave_low_optab = optab_for_tree_code (VEC_INTERLEAVE_LOW_EXPR, vectype, optab_default); if (!interleave_high_optab || !interleave_low_optab) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "no optab for interleave."); return false; } if (optab_handler (interleave_high_optab, mode)->insn_code == CODE_FOR_nothing || optab_handler (interleave_low_optab, mode)->insn_code == CODE_FOR_nothing) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "interleave op not supported by target."); return false; } return true; } /* Function vect_permute_store_chain. Given a chain of interleaved stores in DR_CHAIN of LENGTH that must be a power of 2, generate interleave_high/low stmts to reorder the data correctly for the stores. Return the final references for stores in RESULT_CHAIN. E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8. The input is 4 vectors each containing 8 elements. We assign a number to each element, the input sequence is: 1st vec: 0 1 2 3 4 5 6 7 2nd vec: 8 9 10 11 12 13 14 15 3rd vec: 16 17 18 19 20 21 22 23 4th vec: 24 25 26 27 28 29 30 31 The output sequence should be: 1st vec: 0 8 16 24 1 9 17 25 2nd vec: 2 10 18 26 3 11 19 27 3rd vec: 4 12 20 28 5 13 21 30 4th vec: 6 14 22 30 7 15 23 31 i.e., we interleave the contents of the four vectors in their order. We use interleave_high/low instructions to create such output. The input of each interleave_high/low operation is two vectors: 1st vec 2nd vec 0 1 2 3 4 5 6 7 the even elements of the result vector are obtained left-to-right from the high/low elements of the first vector. The odd elements of the result are obtained left-to-right from the high/low elements of the second vector. The output of interleave_high will be: 0 4 1 5 and of interleave_low: 2 6 3 7 The permutation is done in log LENGTH stages. In each stage interleave_high and interleave_low stmts are created for each pair of vectors in DR_CHAIN, where the first argument is taken from the first half of DR_CHAIN and the second argument from it's second half. In our example, I1: interleave_high (1st vec, 3rd vec) I2: interleave_low (1st vec, 3rd vec) I3: interleave_high (2nd vec, 4th vec) I4: interleave_low (2nd vec, 4th vec) The output for the first stage is: I1: 0 16 1 17 2 18 3 19 I2: 4 20 5 21 6 22 7 23 I3: 8 24 9 25 10 26 11 27 I4: 12 28 13 29 14 30 15 31 The output of the second stage, i.e. the final result is: I1: 0 8 16 24 1 9 17 25 I2: 2 10 18 26 3 11 19 27 I3: 4 12 20 28 5 13 21 30 I4: 6 14 22 30 7 15 23 31. */ static bool vect_permute_store_chain (VEC(tree,heap) *dr_chain, unsigned int length, gimple stmt, gimple_stmt_iterator *gsi, VEC(tree,heap) **result_chain) { tree perm_dest, vect1, vect2, high, low; gimple perm_stmt; tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt)); tree scalar_dest; int i; unsigned int j; enum tree_code high_code, low_code; scalar_dest = gimple_assign_lhs (stmt); /* Check that the operation is supported. */ if (!vect_strided_store_supported (vectype)) return false; *result_chain = VEC_copy (tree, heap, dr_chain); for (i = 0; i < exact_log2 (length); i++) { for (j = 0; j < length/2; j++) { vect1 = VEC_index (tree, dr_chain, j); vect2 = VEC_index (tree, dr_chain, j+length/2); /* Create interleaving stmt: in the case of big endian: high = interleave_high (vect1, vect2) and in the case of little endian: high = interleave_low (vect1, vect2). */ perm_dest = create_tmp_var (vectype, "vect_inter_high"); DECL_GIMPLE_REG_P (perm_dest) = 1; add_referenced_var (perm_dest); if (BYTES_BIG_ENDIAN) { high_code = VEC_INTERLEAVE_HIGH_EXPR; low_code = VEC_INTERLEAVE_LOW_EXPR; } else { low_code = VEC_INTERLEAVE_HIGH_EXPR; high_code = VEC_INTERLEAVE_LOW_EXPR; } perm_stmt = gimple_build_assign_with_ops (high_code, perm_dest, vect1, vect2); high = make_ssa_name (perm_dest, perm_stmt); gimple_assign_set_lhs (perm_stmt, high); vect_finish_stmt_generation (stmt, perm_stmt, gsi); VEC_replace (tree, *result_chain, 2*j, high); /* Create interleaving stmt: in the case of big endian: low = interleave_low (vect1, vect2) and in the case of little endian: low = interleave_high (vect1, vect2). */ perm_dest = create_tmp_var (vectype, "vect_inter_low"); DECL_GIMPLE_REG_P (perm_dest) = 1; add_referenced_var (perm_dest); perm_stmt = gimple_build_assign_with_ops (low_code, perm_dest, vect1, vect2); low = make_ssa_name (perm_dest, perm_stmt); gimple_assign_set_lhs (perm_stmt, low); vect_finish_stmt_generation (stmt, perm_stmt, gsi); VEC_replace (tree, *result_chain, 2*j+1, low); } dr_chain = VEC_copy (tree, heap, *result_chain); } return true; } /* Function vectorizable_store. Check if STMT defines a non scalar data-ref (array/pointer/structure) that can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_store (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt, slp_tree slp_node) { tree scalar_dest; tree data_ref; tree op; tree vec_oprnd = NULL_TREE; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info), *first_dr = NULL; tree vectype = STMT_VINFO_VECTYPE (stmt_info); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); enum machine_mode vec_mode; tree dummy; enum dr_alignment_support alignment_support_scheme; tree def; gimple def_stmt; enum vect_def_type dt; stmt_vec_info prev_stmt_info = NULL; tree dataref_ptr = NULL_TREE; int nunits = TYPE_VECTOR_SUBPARTS (vectype); int ncopies; int j; gimple next_stmt, first_stmt = NULL; bool strided_store = false; unsigned int group_size, i; VEC(tree,heap) *dr_chain = NULL, *oprnds = NULL, *result_chain = NULL; bool inv_p; VEC(tree,heap) *vec_oprnds = NULL; bool slp = (slp_node != NULL); stmt_vec_info first_stmt_vinfo; unsigned int vec_num; /* Multiple types in SLP are handled by creating the appropriate number of vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in case of SLP. */ if (slp) ncopies = 1; else ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits; gcc_assert (ncopies >= 1); /* FORNOW. This restriction should be relaxed. */ if (nested_in_vect_loop_p (loop, stmt) && ncopies > 1) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "multiple types in nested loop."); return false; } if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def) return false; /* Is vectorizable store? */ if (!is_gimple_assign (stmt)) return false; scalar_dest = gimple_assign_lhs (stmt); if (TREE_CODE (scalar_dest) != ARRAY_REF && TREE_CODE (scalar_dest) != INDIRECT_REF && !STMT_VINFO_STRIDED_ACCESS (stmt_info)) return false; gcc_assert (gimple_assign_single_p (stmt)); op = gimple_assign_rhs1 (stmt); if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt)) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } /* The scalar rhs type needs to be trivially convertible to the vector component type. This should always be the case. */ if (!useless_type_conversion_p (TREE_TYPE (vectype), TREE_TYPE (op))) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "??? operands of different types"); return false; } vec_mode = TYPE_MODE (vectype); /* FORNOW. In some cases can vectorize even if data-type not supported (e.g. - array initialization with 0). */ if (optab_handler (mov_optab, (int)vec_mode)->insn_code == CODE_FOR_nothing) return false; if (!STMT_VINFO_DATA_REF (stmt_info)) return false; if (STMT_VINFO_STRIDED_ACCESS (stmt_info)) { strided_store = true; first_stmt = DR_GROUP_FIRST_DR (stmt_info); if (!vect_strided_store_supported (vectype) && !PURE_SLP_STMT (stmt_info) && !slp) return false; if (first_stmt == stmt) { /* STMT is the leader of the group. Check the operands of all the stmts of the group. */ next_stmt = DR_GROUP_NEXT_DR (stmt_info); while (next_stmt) { gcc_assert (gimple_assign_single_p (next_stmt)); op = gimple_assign_rhs1 (next_stmt); if (!vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt)) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt)); } } } if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = store_vec_info_type; vect_model_store_cost (stmt_info, ncopies, dt, NULL); return true; } /** Transform. **/ if (strided_store) { first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt)); group_size = DR_GROUP_SIZE (vinfo_for_stmt (first_stmt)); DR_GROUP_STORE_COUNT (vinfo_for_stmt (first_stmt))++; /* FORNOW */ gcc_assert (!nested_in_vect_loop_p (loop, stmt)); /* We vectorize all the stmts of the interleaving group when we reach the last stmt in the group. */ if (DR_GROUP_STORE_COUNT (vinfo_for_stmt (first_stmt)) < DR_GROUP_SIZE (vinfo_for_stmt (first_stmt)) && !slp) { *vec_stmt = NULL; return true; } if (slp) strided_store = false; /* VEC_NUM is the number of vect stmts to be created for this group. */ if (slp) vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node); else vec_num = group_size; } else { first_stmt = stmt; first_dr = dr; group_size = vec_num = 1; first_stmt_vinfo = stmt_info; } if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform store. ncopies = %d",ncopies); dr_chain = VEC_alloc (tree, heap, group_size); oprnds = VEC_alloc (tree, heap, group_size); alignment_support_scheme = vect_supportable_dr_alignment (first_dr); gcc_assert (alignment_support_scheme); gcc_assert (alignment_support_scheme == dr_aligned); /* FORNOW */ /* In case the vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits), we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. For more details see documentation in vect_get_vec_def_for_copy_stmt. */ /* In case of interleaving (non-unit strided access): S1: &base + 2 = x2 S2: &base = x0 S3: &base + 1 = x1 S4: &base + 3 = x3 We create vectorized stores starting from base address (the access of the first stmt in the chain (S2 in the above example), when the last store stmt of the chain (S4) is reached: VS1: &base = vx2 VS2: &base + vec_size*1 = vx0 VS3: &base + vec_size*2 = vx1 VS4: &base + vec_size*3 = vx3 Then permutation statements are generated: VS5: vx5 = VEC_INTERLEAVE_HIGH_EXPR < vx0, vx3 > VS6: vx6 = VEC_INTERLEAVE_LOW_EXPR < vx0, vx3 > ... And they are put in STMT_VINFO_VEC_STMT of the corresponding scalar stmts (the order of the data-refs in the output of vect_permute_store_chain corresponds to the order of scalar stmts in the interleaving chain - see the documentation of vect_permute_store_chain()). In case of both multiple types and interleaving, above vector stores and permutation stmts are created for every copy. The result vector stmts are put in STMT_VINFO_VEC_STMT for the first copy and in the corresponding STMT_VINFO_RELATED_STMT for the next copies. */ prev_stmt_info = NULL; for (j = 0; j < ncopies; j++) { gimple new_stmt; gimple ptr_incr; if (j == 0) { if (slp) { /* Get vectorized arguments for SLP_NODE. */ vect_get_slp_defs (slp_node, &vec_oprnds, NULL); vec_oprnd = VEC_index (tree, vec_oprnds, 0); } else { /* For interleaved stores we collect vectorized defs for all the stores in the group in DR_CHAIN and OPRNDS. DR_CHAIN is then used as an input to vect_permute_store_chain(), and OPRNDS as an input to vect_get_vec_def_for_stmt_copy() for the next copy. If the store is not strided, GROUP_SIZE is 1, and DR_CHAIN and OPRNDS are of size 1. */ next_stmt = first_stmt; for (i = 0; i < group_size; i++) { /* Since gaps are not supported for interleaved stores, GROUP_SIZE is the exact number of stmts in the chain. Therefore, NEXT_STMT can't be NULL_TREE. In case that there is no interleaving, GROUP_SIZE is 1, and only one iteration of the loop will be executed. */ gcc_assert (next_stmt && gimple_assign_single_p (next_stmt)); op = gimple_assign_rhs1 (next_stmt); vec_oprnd = vect_get_vec_def_for_operand (op, next_stmt, NULL); VEC_quick_push(tree, dr_chain, vec_oprnd); VEC_quick_push(tree, oprnds, vec_oprnd); next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt)); } } /* We should have catched mismatched types earlier. */ gcc_assert (useless_type_conversion_p (vectype, TREE_TYPE (vec_oprnd))); dataref_ptr = vect_create_data_ref_ptr (first_stmt, NULL, NULL_TREE, &dummy, &ptr_incr, false, &inv_p, NULL); gcc_assert (!inv_p); } else { /* For interleaved stores we created vectorized defs for all the defs stored in OPRNDS in the previous iteration (previous copy). DR_CHAIN is then used as an input to vect_permute_store_chain(), and OPRNDS as an input to vect_get_vec_def_for_stmt_copy() for the next copy. If the store is not strided, GROUP_SIZE is 1, and DR_CHAIN and OPRNDS are of size 1. */ for (i = 0; i < group_size; i++) { op = VEC_index (tree, oprnds, i); vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt); vec_oprnd = vect_get_vec_def_for_stmt_copy (dt, op); VEC_replace(tree, dr_chain, i, vec_oprnd); VEC_replace(tree, oprnds, i, vec_oprnd); } dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt, NULL_TREE); } if (strided_store) { result_chain = VEC_alloc (tree, heap, group_size); /* Permute. */ if (!vect_permute_store_chain (dr_chain, group_size, stmt, gsi, &result_chain)) return false; } next_stmt = first_stmt; for (i = 0; i < vec_num; i++) { if (i > 0) /* Bump the vector pointer. */ dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt, NULL_TREE); if (slp) vec_oprnd = VEC_index (tree, vec_oprnds, i); else if (strided_store) /* For strided stores vectorized defs are interleaved in vect_permute_store_chain(). */ vec_oprnd = VEC_index (tree, result_chain, i); data_ref = build_fold_indirect_ref (dataref_ptr); /* Arguments are ready. Create the new vector stmt. */ new_stmt = gimple_build_assign (data_ref, vec_oprnd); vect_finish_stmt_generation (stmt, new_stmt, gsi); mark_symbols_for_renaming (new_stmt); if (slp) continue; if (j == 0) STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt; else STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt)); if (!next_stmt) break; } } VEC_free (tree, heap, dr_chain); VEC_free (tree, heap, oprnds); if (result_chain) VEC_free (tree, heap, result_chain); return true; } /* Function vect_setup_realignment This function is called when vectorizing an unaligned load using the dr_explicit_realign[_optimized] scheme. This function generates the following code at the loop prolog: p = initial_addr; x msq_init = *(floor(p)); # prolog load realignment_token = call target_builtin; loop: x msq = phi (msq_init, ---) The stmts marked with x are generated only for the case of dr_explicit_realign_optimized. The code above sets up a new (vector) pointer, pointing to the first location accessed by STMT, and a "floor-aligned" load using that pointer. It also generates code to compute the "realignment-token" (if the relevant target hook was defined), and creates a phi-node at the loop-header bb whose arguments are the result of the prolog-load (created by this function) and the result of a load that takes place in the loop (to be created by the caller to this function). For the case of dr_explicit_realign_optimized: The caller to this function uses the phi-result (msq) to create the realignment code inside the loop, and sets up the missing phi argument, as follows: loop: msq = phi (msq_init, lsq) lsq = *(floor(p')); # load in loop result = realign_load (msq, lsq, realignment_token); For the case of dr_explicit_realign: loop: msq = *(floor(p)); # load in loop p' = p + (VS-1); lsq = *(floor(p')); # load in loop result = realign_load (msq, lsq, realignment_token); Input: STMT - (scalar) load stmt to be vectorized. This load accesses a memory location that may be unaligned. BSI - place where new code is to be inserted. ALIGNMENT_SUPPORT_SCHEME - which of the two misalignment handling schemes is used. Output: REALIGNMENT_TOKEN - the result of a call to the builtin_mask_for_load target hook, if defined. Return value - the result of the loop-header phi node. */ static tree vect_setup_realignment (gimple stmt, gimple_stmt_iterator *gsi, tree *realignment_token, enum dr_alignment_support alignment_support_scheme, tree init_addr, struct loop **at_loop) { stmt_vec_info stmt_info = vinfo_for_stmt (stmt); tree vectype = STMT_VINFO_VECTYPE (stmt_info); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); edge pe; tree scalar_dest = gimple_assign_lhs (stmt); tree vec_dest; gimple inc; tree ptr; tree data_ref; gimple new_stmt; basic_block new_bb; tree msq_init = NULL_TREE; tree new_temp; gimple phi_stmt; tree msq = NULL_TREE; gimple_seq stmts = NULL; bool inv_p; bool compute_in_loop = false; bool nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt); struct loop *containing_loop = (gimple_bb (stmt))->loop_father; struct loop *loop_for_initial_load; gcc_assert (alignment_support_scheme == dr_explicit_realign || alignment_support_scheme == dr_explicit_realign_optimized); /* We need to generate three things: 1. the misalignment computation 2. the extra vector load (for the optimized realignment scheme). 3. the phi node for the two vectors from which the realignment is done (for the optimized realignment scheme). */ /* 1. Determine where to generate the misalignment computation. If INIT_ADDR is NULL_TREE, this indicates that the misalignment calculation will be generated by this function, outside the loop (in the preheader). Otherwise, INIT_ADDR had already been computed for us by the caller, inside the loop. Background: If the misalignment remains fixed throughout the iterations of the loop, then both realignment schemes are applicable, and also the misalignment computation can be done outside LOOP. This is because we are vectorizing LOOP, and so the memory accesses in LOOP advance in steps that are a multiple of VS (the Vector Size), and therefore the misalignment in different vectorized LOOP iterations is always the same. The problem arises only if the memory access is in an inner-loop nested inside LOOP, which is now being vectorized using outer-loop vectorization. This is the only case when the misalignment of the memory access may not remain fixed throughout the iterations of the inner-loop (as explained in detail in vect_supportable_dr_alignment). In this case, not only is the optimized realignment scheme not applicable, but also the misalignment computation (and generation of the realignment token that is passed to REALIGN_LOAD) have to be done inside the loop. In short, INIT_ADDR indicates whether we are in a COMPUTE_IN_LOOP mode or not, which in turn determines if the misalignment is computed inside the inner-loop, or outside LOOP. */ if (init_addr != NULL_TREE) { compute_in_loop = true; gcc_assert (alignment_support_scheme == dr_explicit_realign); } /* 2. Determine where to generate the extra vector load. For the optimized realignment scheme, instead of generating two vector loads in each iteration, we generate a single extra vector load in the preheader of the loop, and in each iteration reuse the result of the vector load from the previous iteration. In case the memory access is in an inner-loop nested inside LOOP, which is now being vectorized using outer-loop vectorization, we need to determine whether this initial vector load should be generated at the preheader of the inner-loop, or can be generated at the preheader of LOOP. If the memory access has no evolution in LOOP, it can be generated in the preheader of LOOP. Otherwise, it has to be generated inside LOOP (in the preheader of the inner-loop). */ if (nested_in_vect_loop) { tree outerloop_step = STMT_VINFO_DR_STEP (stmt_info); bool invariant_in_outerloop = (tree_int_cst_compare (outerloop_step, size_zero_node) == 0); loop_for_initial_load = (invariant_in_outerloop ? loop : loop->inner); } else loop_for_initial_load = loop; if (at_loop) *at_loop = loop_for_initial_load; /* 3. For the case of the optimized realignment, create the first vector load at the loop preheader. */ if (alignment_support_scheme == dr_explicit_realign_optimized) { /* Create msq_init = *(floor(p1)) in the loop preheader */ gcc_assert (!compute_in_loop); pe = loop_preheader_edge (loop_for_initial_load); vec_dest = vect_create_destination_var (scalar_dest, vectype); ptr = vect_create_data_ref_ptr (stmt, loop_for_initial_load, NULL_TREE, &init_addr, &inc, true, &inv_p, NULL_TREE); data_ref = build1 (ALIGN_INDIRECT_REF, vectype, ptr); new_stmt = gimple_build_assign (vec_dest, data_ref); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_temp); mark_symbols_for_renaming (new_stmt); new_bb = gsi_insert_on_edge_immediate (pe, new_stmt); gcc_assert (!new_bb); msq_init = gimple_assign_lhs (new_stmt); } /* 4. Create realignment token using a target builtin, if available. It is done either inside the containing loop, or before LOOP (as determined above). */ if (targetm.vectorize.builtin_mask_for_load) { tree builtin_decl; /* Compute INIT_ADDR - the initial addressed accessed by this memref. */ if (compute_in_loop) gcc_assert (init_addr); /* already computed by the caller. */ else { /* Generate the INIT_ADDR computation outside LOOP. */ init_addr = vect_create_addr_base_for_vector_ref (stmt, &stmts, NULL_TREE, loop); pe = loop_preheader_edge (loop); new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); gcc_assert (!new_bb); } builtin_decl = targetm.vectorize.builtin_mask_for_load (); new_stmt = gimple_build_call (builtin_decl, 1, init_addr); vec_dest = vect_create_destination_var (scalar_dest, gimple_call_return_type (new_stmt)); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_call_set_lhs (new_stmt, new_temp); if (compute_in_loop) gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT); else { /* Generate the misalignment computation outside LOOP. */ pe = loop_preheader_edge (loop); new_bb = gsi_insert_on_edge_immediate (pe, new_stmt); gcc_assert (!new_bb); } *realignment_token = gimple_call_lhs (new_stmt); /* The result of the CALL_EXPR to this builtin is determined from the value of the parameter and no global variables are touched which makes the builtin a "const" function. Requiring the builtin to have the "const" attribute makes it unnecessary to call mark_call_clobbered. */ gcc_assert (TREE_READONLY (builtin_decl)); } if (alignment_support_scheme == dr_explicit_realign) return msq; gcc_assert (!compute_in_loop); gcc_assert (alignment_support_scheme == dr_explicit_realign_optimized); /* 5. Create msq = phi in loop */ pe = loop_preheader_edge (containing_loop); vec_dest = vect_create_destination_var (scalar_dest, vectype); msq = make_ssa_name (vec_dest, NULL); phi_stmt = create_phi_node (msq, containing_loop->header); SSA_NAME_DEF_STMT (msq) = phi_stmt; add_phi_arg (phi_stmt, msq_init, pe, UNKNOWN_LOCATION); return msq; } /* Function vect_strided_load_supported. Returns TRUE is EXTRACT_EVEN and EXTRACT_ODD operations are supported, and FALSE otherwise. */ static bool vect_strided_load_supported (tree vectype) { optab perm_even_optab, perm_odd_optab; int mode; mode = (int) TYPE_MODE (vectype); perm_even_optab = optab_for_tree_code (VEC_EXTRACT_EVEN_EXPR, vectype, optab_default); if (!perm_even_optab) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "no optab for perm_even."); return false; } if (optab_handler (perm_even_optab, mode)->insn_code == CODE_FOR_nothing) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "perm_even op not supported by target."); return false; } perm_odd_optab = optab_for_tree_code (VEC_EXTRACT_ODD_EXPR, vectype, optab_default); if (!perm_odd_optab) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "no optab for perm_odd."); return false; } if (optab_handler (perm_odd_optab, mode)->insn_code == CODE_FOR_nothing) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "perm_odd op not supported by target."); return false; } return true; } /* Function vect_permute_load_chain. Given a chain of interleaved loads in DR_CHAIN of LENGTH that must be a power of 2, generate extract_even/odd stmts to reorder the input data correctly. Return the final references for loads in RESULT_CHAIN. E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8. The input is 4 vectors each containing 8 elements. We assign a number to each element, the input sequence is: 1st vec: 0 1 2 3 4 5 6 7 2nd vec: 8 9 10 11 12 13 14 15 3rd vec: 16 17 18 19 20 21 22 23 4th vec: 24 25 26 27 28 29 30 31 The output sequence should be: 1st vec: 0 4 8 12 16 20 24 28 2nd vec: 1 5 9 13 17 21 25 29 3rd vec: 2 6 10 14 18 22 26 30 4th vec: 3 7 11 15 19 23 27 31 i.e., the first output vector should contain the first elements of each interleaving group, etc. We use extract_even/odd instructions to create such output. The input of each extract_even/odd operation is two vectors 1st vec 2nd vec 0 1 2 3 4 5 6 7 and the output is the vector of extracted even/odd elements. The output of extract_even will be: 0 2 4 6 and of extract_odd: 1 3 5 7 The permutation is done in log LENGTH stages. In each stage extract_even and extract_odd stmts are created for each pair of vectors in DR_CHAIN in their order. In our example, E1: extract_even (1st vec, 2nd vec) E2: extract_odd (1st vec, 2nd vec) E3: extract_even (3rd vec, 4th vec) E4: extract_odd (3rd vec, 4th vec) The output for the first stage will be: E1: 0 2 4 6 8 10 12 14 E2: 1 3 5 7 9 11 13 15 E3: 16 18 20 22 24 26 28 30 E4: 17 19 21 23 25 27 29 31 In order to proceed and create the correct sequence for the next stage (or for the correct output, if the second stage is the last one, as in our example), we first put the output of extract_even operation and then the output of extract_odd in RESULT_CHAIN (which is then copied to DR_CHAIN). The input for the second stage is: 1st vec (E1): 0 2 4 6 8 10 12 14 2nd vec (E3): 16 18 20 22 24 26 28 30 3rd vec (E2): 1 3 5 7 9 11 13 15 4th vec (E4): 17 19 21 23 25 27 29 31 The output of the second stage: E1: 0 4 8 12 16 20 24 28 E2: 2 6 10 14 18 22 26 30 E3: 1 5 9 13 17 21 25 29 E4: 3 7 11 15 19 23 27 31 And RESULT_CHAIN after reordering: 1st vec (E1): 0 4 8 12 16 20 24 28 2nd vec (E3): 1 5 9 13 17 21 25 29 3rd vec (E2): 2 6 10 14 18 22 26 30 4th vec (E4): 3 7 11 15 19 23 27 31. */ static bool vect_permute_load_chain (VEC(tree,heap) *dr_chain, unsigned int length, gimple stmt, gimple_stmt_iterator *gsi, VEC(tree,heap) **result_chain) { tree perm_dest, data_ref, first_vect, second_vect; gimple perm_stmt; tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt)); int i; unsigned int j; /* Check that the operation is supported. */ if (!vect_strided_load_supported (vectype)) return false; *result_chain = VEC_copy (tree, heap, dr_chain); for (i = 0; i < exact_log2 (length); i++) { for (j = 0; j < length; j +=2) { first_vect = VEC_index (tree, dr_chain, j); second_vect = VEC_index (tree, dr_chain, j+1); /* data_ref = permute_even (first_data_ref, second_data_ref); */ perm_dest = create_tmp_var (vectype, "vect_perm_even"); DECL_GIMPLE_REG_P (perm_dest) = 1; add_referenced_var (perm_dest); perm_stmt = gimple_build_assign_with_ops (VEC_EXTRACT_EVEN_EXPR, perm_dest, first_vect, second_vect); data_ref = make_ssa_name (perm_dest, perm_stmt); gimple_assign_set_lhs (perm_stmt, data_ref); vect_finish_stmt_generation (stmt, perm_stmt, gsi); mark_symbols_for_renaming (perm_stmt); VEC_replace (tree, *result_chain, j/2, data_ref); /* data_ref = permute_odd (first_data_ref, second_data_ref); */ perm_dest = create_tmp_var (vectype, "vect_perm_odd"); DECL_GIMPLE_REG_P (perm_dest) = 1; add_referenced_var (perm_dest); perm_stmt = gimple_build_assign_with_ops (VEC_EXTRACT_ODD_EXPR, perm_dest, first_vect, second_vect); data_ref = make_ssa_name (perm_dest, perm_stmt); gimple_assign_set_lhs (perm_stmt, data_ref); vect_finish_stmt_generation (stmt, perm_stmt, gsi); mark_symbols_for_renaming (perm_stmt); VEC_replace (tree, *result_chain, j/2+length/2, data_ref); } dr_chain = VEC_copy (tree, heap, *result_chain); } return true; } /* Function vect_transform_strided_load. Given a chain of input interleaved data-refs (in DR_CHAIN), build statements to perform their permutation and ascribe the result vectorized statements to the scalar statements. */ static bool vect_transform_strided_load (gimple stmt, VEC(tree,heap) *dr_chain, int size, gimple_stmt_iterator *gsi) { stmt_vec_info stmt_info = vinfo_for_stmt (stmt); gimple first_stmt = DR_GROUP_FIRST_DR (stmt_info); gimple next_stmt, new_stmt; VEC(tree,heap) *result_chain = NULL; unsigned int i, gap_count; tree tmp_data_ref; /* DR_CHAIN contains input data-refs that are a part of the interleaving. RESULT_CHAIN is the output of vect_permute_load_chain, it contains permuted vectors, that are ready for vector computation. */ result_chain = VEC_alloc (tree, heap, size); /* Permute. */ if (!vect_permute_load_chain (dr_chain, size, stmt, gsi, &result_chain)) return false; /* Put a permuted data-ref in the VECTORIZED_STMT field. Since we scan the chain starting from it's first node, their order corresponds the order of data-refs in RESULT_CHAIN. */ next_stmt = first_stmt; gap_count = 1; for (i = 0; VEC_iterate (tree, result_chain, i, tmp_data_ref); i++) { if (!next_stmt) break; /* Skip the gaps. Loads created for the gaps will be removed by dead code elimination pass later. No need to check for the first stmt in the group, since it always exists. DR_GROUP_GAP is the number of steps in elements from the previous access (if there is no gap DR_GROUP_GAP is 1). We skip loads that correspond to the gaps. */ if (next_stmt != first_stmt && gap_count < DR_GROUP_GAP (vinfo_for_stmt (next_stmt))) { gap_count++; continue; } while (next_stmt) { new_stmt = SSA_NAME_DEF_STMT (tmp_data_ref); /* We assume that if VEC_STMT is not NULL, this is a case of multiple copies, and we put the new vector statement in the first available RELATED_STMT. */ if (!STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt))) STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)) = new_stmt; else { if (!DR_GROUP_SAME_DR_STMT (vinfo_for_stmt (next_stmt))) { gimple prev_stmt = STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)); gimple rel_stmt = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (prev_stmt)); while (rel_stmt) { prev_stmt = rel_stmt; rel_stmt = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (rel_stmt)); } STMT_VINFO_RELATED_STMT (vinfo_for_stmt (prev_stmt)) = new_stmt; } } next_stmt = DR_GROUP_NEXT_DR (vinfo_for_stmt (next_stmt)); gap_count = 1; /* If NEXT_STMT accesses the same DR as the previous statement, put the same TMP_DATA_REF as its vectorized statement; otherwise get the next data-ref from RESULT_CHAIN. */ if (!next_stmt || !DR_GROUP_SAME_DR_STMT (vinfo_for_stmt (next_stmt))) break; } } VEC_free (tree, heap, result_chain); return true; } /* Create NCOPIES permutation statements using the mask MASK_BYTES (by building a vector of type MASK_TYPE from it) and two input vectors placed in DR_CHAIN at FIRST_VEC_INDX and SECOND_VEC_INDX for the first copy and shifting by STRIDE elements of DR_CHAIN for every copy. (STRIDE is the number of vectorized stmts for NODE divided by the number of copies). VECT_STMTS_COUNTER specifies the index in the vectorized stmts of NODE, where the created stmts must be inserted. */ static inline void vect_create_mask_and_perm (gimple stmt, gimple next_scalar_stmt, int *mask_array, int mask_nunits, tree mask_element_type, tree mask_type, int first_vec_indx, int second_vec_indx, gimple_stmt_iterator *gsi, slp_tree node, tree builtin_decl, tree vectype, VEC(tree,heap) *dr_chain, int ncopies, int vect_stmts_counter) { tree t = NULL_TREE, mask_vec, mask, perm_dest; gimple perm_stmt = NULL; stmt_vec_info next_stmt_info; int i, group_size, stride, dr_chain_size; tree first_vec, second_vec, data_ref; tree sym; ssa_op_iter iter; VEC (tree, heap) *params = NULL; /* Create a vector mask. */ for (i = mask_nunits - 1; i >= 0; --i) t = tree_cons (NULL_TREE, build_int_cst (mask_element_type, mask_array[i]), t); mask_vec = build_vector (mask_type, t); mask = vect_init_vector (stmt, mask_vec, mask_type, NULL); group_size = VEC_length (gimple, SLP_TREE_SCALAR_STMTS (node)); stride = SLP_TREE_NUMBER_OF_VEC_STMTS (node) / ncopies; dr_chain_size = VEC_length (tree, dr_chain); /* Initialize the vect stmts of NODE to properly insert the generated stmts later. */ for (i = VEC_length (gimple, SLP_TREE_VEC_STMTS (node)); i < (int) SLP_TREE_NUMBER_OF_VEC_STMTS (node); i++) VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (node), NULL); perm_dest = vect_create_destination_var (gimple_assign_lhs (stmt), vectype); for (i = 0; i < ncopies; i++) { first_vec = VEC_index (tree, dr_chain, first_vec_indx); second_vec = VEC_index (tree, dr_chain, second_vec_indx); /* Build argument list for the vectorized call. */ VEC_free (tree, heap, params); params = VEC_alloc (tree, heap, 3); VEC_quick_push (tree, params, first_vec); VEC_quick_push (tree, params, second_vec); VEC_quick_push (tree, params, mask); /* Generate the permute statement. */ perm_stmt = gimple_build_call_vec (builtin_decl, params); data_ref = make_ssa_name (perm_dest, perm_stmt); gimple_call_set_lhs (perm_stmt, data_ref); vect_finish_stmt_generation (stmt, perm_stmt, gsi); FOR_EACH_SSA_TREE_OPERAND (sym, perm_stmt, iter, SSA_OP_ALL_VIRTUALS) { if (TREE_CODE (sym) == SSA_NAME) sym = SSA_NAME_VAR (sym); mark_sym_for_renaming (sym); } /* Store the vector statement in NODE. */ VEC_replace (gimple, SLP_TREE_VEC_STMTS (node), stride * i + vect_stmts_counter, perm_stmt); first_vec_indx += stride; second_vec_indx += stride; } /* Mark the scalar stmt as vectorized. */ next_stmt_info = vinfo_for_stmt (next_scalar_stmt); STMT_VINFO_VEC_STMT (next_stmt_info) = perm_stmt; } /* Given FIRST_MASK_ELEMENT - the mask element in element representation, return in CURRENT_MASK_ELEMENT its equivalent in target specific representation. Check that the mask is valid and return FALSE if not. Return TRUE in NEED_NEXT_VECTOR if the permutation requires to move to the next vector, i.e., the current first vector is not needed. */ static bool vect_get_mask_element (gimple stmt, int first_mask_element, int m, int mask_nunits, bool only_one_vec, int index, int *mask, int *current_mask_element, bool *need_next_vector) { int i; static int number_of_mask_fixes = 1; static bool mask_fixed = false; static bool needs_first_vector = false; /* Convert to target specific representation. */ *current_mask_element = first_mask_element + m; /* Adjust the value in case it's a mask for second and third vectors. */ *current_mask_element -= mask_nunits * (number_of_mask_fixes - 1); if (*current_mask_element < mask_nunits) needs_first_vector = true; /* We have only one input vector to permute but the mask accesses values in the next vector as well. */ if (only_one_vec && *current_mask_element >= mask_nunits) { if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "permutation requires at least two vectors "); print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); } return false; } /* The mask requires the next vector. */ if (*current_mask_element >= mask_nunits * 2) { if (needs_first_vector || mask_fixed) { /* We either need the first vector too or have already moved to the next vector. In both cases, this permutation needs three vectors. */ if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "permutation requires at " "least three vectors "); print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); } return false; } /* We move to the next vector, dropping the first one and working with the second and the third - we need to adjust the values of the mask accordingly. */ *current_mask_element -= mask_nunits * number_of_mask_fixes; for (i = 0; i < index; i++) mask[i] -= mask_nunits * number_of_mask_fixes; (number_of_mask_fixes)++; mask_fixed = true; } *need_next_vector = mask_fixed; /* This was the last element of this mask. Start a new one. */ if (index == mask_nunits - 1) { number_of_mask_fixes = 1; mask_fixed = false; needs_first_vector = false; } return true; } /* Generate vector permute statements from a list of loads in DR_CHAIN. If ANALYZE_ONLY is TRUE, only check that it is possible to create valid permute statements for SLP_NODE_INSTANCE. */ bool vect_transform_slp_perm_load (gimple stmt, VEC (tree, heap) *dr_chain, gimple_stmt_iterator *gsi, int vf, slp_instance slp_node_instance, bool analyze_only) { stmt_vec_info stmt_info = vinfo_for_stmt (stmt); tree mask_element_type = NULL_TREE, mask_type; int i, j, k, m, scale, mask_nunits, nunits, vec_index = 0, scalar_index; slp_tree node; tree vectype = STMT_VINFO_VECTYPE (stmt_info), builtin_decl; gimple next_scalar_stmt; int group_size = SLP_INSTANCE_GROUP_SIZE (slp_node_instance); int first_mask_element; int index, unroll_factor, *mask, current_mask_element, ncopies; bool only_one_vec = false, need_next_vector = false; int first_vec_index, second_vec_index, orig_vec_stmts_num, vect_stmts_counter; if (!targetm.vectorize.builtin_vec_perm) { if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "no builtin for vect permute for "); print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); } return false; } builtin_decl = targetm.vectorize.builtin_vec_perm (vectype, &mask_element_type); if (!builtin_decl || !mask_element_type) { if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "no builtin for vect permute for "); print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); } return false; } mask_type = get_vectype_for_scalar_type (mask_element_type); mask_nunits = TYPE_VECTOR_SUBPARTS (mask_type); mask = (int *) xmalloc (sizeof (int) * mask_nunits); nunits = TYPE_VECTOR_SUBPARTS (vectype); scale = mask_nunits / nunits; unroll_factor = SLP_INSTANCE_UNROLLING_FACTOR (slp_node_instance); /* The number of vector stmts to generate based only on SLP_NODE_INSTANCE unrolling factor. */ orig_vec_stmts_num = group_size * SLP_INSTANCE_UNROLLING_FACTOR (slp_node_instance) / nunits; if (orig_vec_stmts_num == 1) only_one_vec = true; /* Number of copies is determined by the final vectorization factor relatively to SLP_NODE_INSTANCE unrolling factor. */ ncopies = vf / SLP_INSTANCE_UNROLLING_FACTOR (slp_node_instance); /* Generate permutation masks for every NODE. Number of masks for each NODE is equal to GROUP_SIZE. E.g., we have a group of three nodes with three loads from the same location in each node, and the vector size is 4. I.e., we have a a0b0c0a1b1c1... sequence and we need to create the following vectors: for a's: a0a0a0a1 a1a1a2a2 a2a3a3a3 for b's: b0b0b0b1 b1b1b2b2 b2b3b3b3 ... The masks for a's should be: {0,0,0,3} {3,3,6,6} {6,9,9,9} (in target scpecific type, e.g., in bytes for Altivec. The last mask is illegal since we assume two operands for permute operation, and the mask element values can't be outside that range. Hence, the last mask must be converted into {2,5,5,5}. For the first two permutations we need the first and the second input vectors: {a0,b0,c0,a1} and {b1,c1,a2,b2}, and for the last permutation we need the second and the third vectors: {b1,c1,a2,b2} and {c2,a3,b3,c3}. */ for (i = 0; VEC_iterate (slp_tree, SLP_INSTANCE_LOADS (slp_node_instance), i, node); i++) { scalar_index = 0; index = 0; vect_stmts_counter = 0; vec_index = 0; first_vec_index = vec_index++; if (only_one_vec) second_vec_index = first_vec_index; else second_vec_index = vec_index++; for (j = 0; j < unroll_factor; j++) { for (k = 0; k < group_size; k++) { first_mask_element = (i + j * group_size) * scale; for (m = 0; m < scale; m++) { if (!vect_get_mask_element (stmt, first_mask_element, m, mask_nunits, only_one_vec, index, mask, ¤t_mask_element, &need_next_vector)) return false; mask[index++] = current_mask_element; } if (index == mask_nunits) { index = 0; if (!analyze_only) { if (need_next_vector) { first_vec_index = second_vec_index; second_vec_index = vec_index; } next_scalar_stmt = VEC_index (gimple, SLP_TREE_SCALAR_STMTS (node), scalar_index++); vect_create_mask_and_perm (stmt, next_scalar_stmt, mask, mask_nunits, mask_element_type, mask_type, first_vec_index, second_vec_index, gsi, node, builtin_decl, vectype, dr_chain, ncopies, vect_stmts_counter++); } } } } } free (mask); return true; } /* vectorizable_load. Check if STMT reads a non scalar data-ref (array/pointer/structure) that can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_load (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt, slp_tree slp_node, slp_instance slp_node_instance) { tree scalar_dest; tree vec_dest = NULL; tree data_ref = NULL; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); stmt_vec_info prev_stmt_info; loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); struct loop *containing_loop = (gimple_bb (stmt))->loop_father; bool nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt); struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info), *first_dr; tree vectype = STMT_VINFO_VECTYPE (stmt_info); tree new_temp; int mode; gimple new_stmt = NULL; tree dummy; enum dr_alignment_support alignment_support_scheme; tree dataref_ptr = NULL_TREE; gimple ptr_incr; int nunits = TYPE_VECTOR_SUBPARTS (vectype); int ncopies; int i, j, group_size; tree msq = NULL_TREE, lsq; tree offset = NULL_TREE; tree realignment_token = NULL_TREE; gimple phi = NULL; VEC(tree,heap) *dr_chain = NULL; bool strided_load = false; gimple first_stmt; tree scalar_type; bool inv_p; bool compute_in_loop = false; struct loop *at_loop; int vec_num; bool slp = (slp_node != NULL); bool slp_perm = false; enum tree_code code; /* Multiple types in SLP are handled by creating the appropriate number of vectorized stmts for each SLP node. Hence, NCOPIES is always 1 in case of SLP. */ if (slp) ncopies = 1; else ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits; gcc_assert (ncopies >= 1); /* FORNOW. This restriction should be relaxed. */ if (nested_in_vect_loop && ncopies > 1) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "multiple types in nested loop."); return false; } if (slp && SLP_INSTANCE_LOAD_PERMUTATION (slp_node_instance)) slp_perm = true; if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def) return false; /* Is vectorizable load? */ if (!is_gimple_assign (stmt)) return false; scalar_dest = gimple_assign_lhs (stmt); if (TREE_CODE (scalar_dest) != SSA_NAME) return false; code = gimple_assign_rhs_code (stmt); if (code != ARRAY_REF && code != INDIRECT_REF && !STMT_VINFO_STRIDED_ACCESS (stmt_info)) return false; if (!STMT_VINFO_DATA_REF (stmt_info)) return false; scalar_type = TREE_TYPE (DR_REF (dr)); mode = (int) TYPE_MODE (vectype); /* FORNOW. In some cases can vectorize even if data-type not supported (e.g. - data copies). */ if (optab_handler (mov_optab, mode)->insn_code == CODE_FOR_nothing) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "Aligned load, but unsupported type."); return false; } /* The vector component type needs to be trivially convertible to the scalar lhs. This should always be the case. */ if (!useless_type_conversion_p (TREE_TYPE (scalar_dest), TREE_TYPE (vectype))) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "??? operands of different types"); return false; } /* Check if the load is a part of an interleaving chain. */ if (STMT_VINFO_STRIDED_ACCESS (stmt_info)) { strided_load = true; /* FORNOW */ gcc_assert (! nested_in_vect_loop); /* Check if interleaving is supported. */ if (!vect_strided_load_supported (vectype) && !PURE_SLP_STMT (stmt_info) && !slp) return false; } if (!vec_stmt) /* transformation not required. */ { STMT_VINFO_TYPE (stmt_info) = load_vec_info_type; vect_model_load_cost (stmt_info, ncopies, NULL); return true; } if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform load."); /** Transform. **/ if (strided_load) { first_stmt = DR_GROUP_FIRST_DR (stmt_info); /* Check if the chain of loads is already vectorized. */ if (STMT_VINFO_VEC_STMT (vinfo_for_stmt (first_stmt))) { *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info); return true; } first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt)); group_size = DR_GROUP_SIZE (vinfo_for_stmt (first_stmt)); /* VEC_NUM is the number of vect stmts to be created for this group. */ if (slp) { strided_load = false; vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node); } else vec_num = group_size; dr_chain = VEC_alloc (tree, heap, vec_num); } else { first_stmt = stmt; first_dr = dr; group_size = vec_num = 1; } alignment_support_scheme = vect_supportable_dr_alignment (first_dr); gcc_assert (alignment_support_scheme); /* In case the vectorization factor (VF) is bigger than the number of elements that we can fit in a vectype (nunits), we have to generate more than one vector stmt - i.e - we need to "unroll" the vector stmt by a factor VF/nunits. In doing so, we record a pointer from one copy of the vector stmt to the next, in the field STMT_VINFO_RELATED_STMT. This is necessary in order to allow following stages to find the correct vector defs to be used when vectorizing stmts that use the defs of the current stmt. The example below illustrates the vectorization process when VF=16 and nunits=4 (i.e - we need to create 4 vectorized stmts): before vectorization: RELATED_STMT VEC_STMT S1: x = memref - - S2: z = x + 1 - - step 1: vectorize stmt S1: We first create the vector stmt VS1_0, and, as usual, record a pointer to it in the STMT_VINFO_VEC_STMT of the scalar stmt S1. Next, we create the vector stmt VS1_1, and record a pointer to it in the STMT_VINFO_RELATED_STMT of the vector stmt VS1_0. Similarly, for VS1_2 and VS1_3. This is the resulting chain of stmts and pointers: RELATED_STMT VEC_STMT VS1_0: vx0 = memref0 VS1_1 - VS1_1: vx1 = memref1 VS1_2 - VS1_2: vx2 = memref2 VS1_3 - VS1_3: vx3 = memref3 - - S1: x = load - VS1_0 S2: z = x + 1 - - See in documentation in vect_get_vec_def_for_stmt_copy for how the information we recorded in RELATED_STMT field is used to vectorize stmt S2. */ /* In case of interleaving (non-unit strided access): S1: x2 = &base + 2 S2: x0 = &base S3: x1 = &base + 1 S4: x3 = &base + 3 Vectorized loads are created in the order of memory accesses starting from the access of the first stmt of the chain: VS1: vx0 = &base VS2: vx1 = &base + vec_size*1 VS3: vx3 = &base + vec_size*2 VS4: vx4 = &base + vec_size*3 Then permutation statements are generated: VS5: vx5 = VEC_EXTRACT_EVEN_EXPR < vx0, vx1 > VS6: vx6 = VEC_EXTRACT_ODD_EXPR < vx0, vx1 > ... And they are put in STMT_VINFO_VEC_STMT of the corresponding scalar stmts (the order of the data-refs in the output of vect_permute_load_chain corresponds to the order of scalar stmts in the interleaving chain - see the documentation of vect_permute_load_chain()). The generation of permutation stmts and recording them in STMT_VINFO_VEC_STMT is done in vect_transform_strided_load(). In case of both multiple types and interleaving, the vector loads and permutation stmts above are created for every copy. The result vector stmts are put in STMT_VINFO_VEC_STMT for the first copy and in the corresponding STMT_VINFO_RELATED_STMT for the next copies. */ /* If the data reference is aligned (dr_aligned) or potentially unaligned on a target that supports unaligned accesses (dr_unaligned_supported) we generate the following code: p = initial_addr; indx = 0; loop { p = p + indx * vectype_size; vec_dest = *(p); indx = indx + 1; } Otherwise, the data reference is potentially unaligned on a target that does not support unaligned accesses (dr_explicit_realign_optimized) - then generate the following code, in which the data in each iteration is obtained by two vector loads, one from the previous iteration, and one from the current iteration: p1 = initial_addr; msq_init = *(floor(p1)) p2 = initial_addr + VS - 1; realignment_token = call target_builtin; indx = 0; loop { p2 = p2 + indx * vectype_size lsq = *(floor(p2)) vec_dest = realign_load (msq, lsq, realignment_token) indx = indx + 1; msq = lsq; } */ /* If the misalignment remains the same throughout the execution of the loop, we can create the init_addr and permutation mask at the loop preheader. Otherwise, it needs to be created inside the loop. This can only occur when vectorizing memory accesses in the inner-loop nested within an outer-loop that is being vectorized. */ if (nested_in_vect_loop_p (loop, stmt) && (TREE_INT_CST_LOW (DR_STEP (dr)) % GET_MODE_SIZE (TYPE_MODE (vectype)) != 0)) { gcc_assert (alignment_support_scheme != dr_explicit_realign_optimized); compute_in_loop = true; } if ((alignment_support_scheme == dr_explicit_realign_optimized || alignment_support_scheme == dr_explicit_realign) && !compute_in_loop) { msq = vect_setup_realignment (first_stmt, gsi, &realignment_token, alignment_support_scheme, NULL_TREE, &at_loop); if (alignment_support_scheme == dr_explicit_realign_optimized) { phi = SSA_NAME_DEF_STMT (msq); offset = size_int (TYPE_VECTOR_SUBPARTS (vectype) - 1); } } else at_loop = loop; prev_stmt_info = NULL; for (j = 0; j < ncopies; j++) { /* 1. Create the vector pointer update chain. */ if (j == 0) dataref_ptr = vect_create_data_ref_ptr (first_stmt, at_loop, offset, &dummy, &ptr_incr, false, &inv_p, NULL_TREE); else dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt, NULL_TREE); for (i = 0; i < vec_num; i++) { if (i > 0) dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi, stmt, NULL_TREE); /* 2. Create the vector-load in the loop. */ switch (alignment_support_scheme) { case dr_aligned: gcc_assert (aligned_access_p (first_dr)); data_ref = build_fold_indirect_ref (dataref_ptr); break; case dr_unaligned_supported: { int mis = DR_MISALIGNMENT (first_dr); tree tmis = (mis == -1 ? size_zero_node : size_int (mis)); tmis = size_binop (MULT_EXPR, tmis, size_int(BITS_PER_UNIT)); data_ref = build2 (MISALIGNED_INDIRECT_REF, vectype, dataref_ptr, tmis); break; } case dr_explicit_realign: { tree ptr, bump; tree vs_minus_1 = size_int (TYPE_VECTOR_SUBPARTS (vectype) - 1); if (compute_in_loop) msq = vect_setup_realignment (first_stmt, gsi, &realignment_token, dr_explicit_realign, dataref_ptr, NULL); data_ref = build1 (ALIGN_INDIRECT_REF, vectype, dataref_ptr); vec_dest = vect_create_destination_var (scalar_dest, vectype); new_stmt = gimple_build_assign (vec_dest, data_ref); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); copy_virtual_operands (new_stmt, stmt); mark_symbols_for_renaming (new_stmt); msq = new_temp; bump = size_binop (MULT_EXPR, vs_minus_1, TYPE_SIZE_UNIT (scalar_type)); ptr = bump_vector_ptr (dataref_ptr, NULL, gsi, stmt, bump); data_ref = build1 (ALIGN_INDIRECT_REF, vectype, ptr); break; } case dr_explicit_realign_optimized: data_ref = build1 (ALIGN_INDIRECT_REF, vectype, dataref_ptr); break; default: gcc_unreachable (); } vec_dest = vect_create_destination_var (scalar_dest, vectype); new_stmt = gimple_build_assign (vec_dest, data_ref); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); mark_symbols_for_renaming (new_stmt); /* 3. Handle explicit realignment if necessary/supported. Create in loop: vec_dest = realign_load (msq, lsq, realignment_token) */ if (alignment_support_scheme == dr_explicit_realign_optimized || alignment_support_scheme == dr_explicit_realign) { tree tmp; lsq = gimple_assign_lhs (new_stmt); if (!realignment_token) realignment_token = dataref_ptr; vec_dest = vect_create_destination_var (scalar_dest, vectype); tmp = build3 (REALIGN_LOAD_EXPR, vectype, msq, lsq, realignment_token); new_stmt = gimple_build_assign (vec_dest, tmp); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); if (alignment_support_scheme == dr_explicit_realign_optimized) { gcc_assert (phi); if (i == vec_num - 1 && j == ncopies - 1) add_phi_arg (phi, lsq, loop_latch_edge (containing_loop), UNKNOWN_LOCATION); msq = lsq; } } /* 4. Handle invariant-load. */ if (inv_p) { gcc_assert (!strided_load); gcc_assert (nested_in_vect_loop_p (loop, stmt)); if (j == 0) { int k; tree t = NULL_TREE; tree vec_inv, bitpos, bitsize = TYPE_SIZE (scalar_type); /* CHECKME: bitpos depends on endianess? */ bitpos = bitsize_zero_node; vec_inv = build3 (BIT_FIELD_REF, scalar_type, new_temp, bitsize, bitpos); vec_dest = vect_create_destination_var (scalar_dest, NULL_TREE); new_stmt = gimple_build_assign (vec_dest, vec_inv); new_temp = make_ssa_name (vec_dest, new_stmt); gimple_assign_set_lhs (new_stmt, new_temp); vect_finish_stmt_generation (stmt, new_stmt, gsi); for (k = nunits - 1; k >= 0; --k) t = tree_cons (NULL_TREE, new_temp, t); /* FIXME: use build_constructor directly. */ vec_inv = build_constructor_from_list (vectype, t); new_temp = vect_init_vector (stmt, vec_inv, vectype, gsi); new_stmt = SSA_NAME_DEF_STMT (new_temp); } else gcc_unreachable (); /* FORNOW. */ } /* Collect vector loads and later create their permutation in vect_transform_strided_load (). */ if (strided_load || slp_perm) VEC_quick_push (tree, dr_chain, new_temp); /* Store vector loads in the corresponding SLP_NODE. */ if (slp && !slp_perm) VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt); } if (slp && !slp_perm) continue; if (slp_perm) { if (!vect_transform_slp_perm_load (stmt, dr_chain, gsi, LOOP_VINFO_VECT_FACTOR (loop_vinfo), slp_node_instance, false)) { VEC_free (tree, heap, dr_chain); return false; } } else { if (strided_load) { if (!vect_transform_strided_load (stmt, dr_chain, group_size, gsi)) return false; *vec_stmt = STMT_VINFO_VEC_STMT (stmt_info); VEC_free (tree, heap, dr_chain); dr_chain = VEC_alloc (tree, heap, group_size); } else { if (j == 0) STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt; else STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; prev_stmt_info = vinfo_for_stmt (new_stmt); } } } if (dr_chain) VEC_free (tree, heap, dr_chain); return true; } /* Function vectorizable_live_operation. STMT computes a value that is used outside the loop. Check if it can be supported. */ bool vectorizable_live_operation (gimple stmt, gimple_stmt_iterator *gsi ATTRIBUTE_UNUSED, gimple *vec_stmt ATTRIBUTE_UNUSED) { stmt_vec_info stmt_info = vinfo_for_stmt (stmt); loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); int i; int op_type; tree op; tree def; gimple def_stmt; enum vect_def_type dt; enum tree_code code; enum gimple_rhs_class rhs_class; gcc_assert (STMT_VINFO_LIVE_P (stmt_info)); if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_reduction_def) return false; if (!is_gimple_assign (stmt)) return false; if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME) return false; /* FORNOW. CHECKME. */ if (nested_in_vect_loop_p (loop, stmt)) return false; code = gimple_assign_rhs_code (stmt); op_type = TREE_CODE_LENGTH (code); rhs_class = get_gimple_rhs_class (code); gcc_assert (rhs_class != GIMPLE_UNARY_RHS || op_type == unary_op); gcc_assert (rhs_class != GIMPLE_BINARY_RHS || op_type == binary_op); /* FORNOW: support only if all uses are invariant. This means that the scalar operations can remain in place, unvectorized. The original last scalar value that they compute will be used. */ for (i = 0; i < op_type; i++) { if (rhs_class == GIMPLE_SINGLE_RHS) op = TREE_OPERAND (gimple_op (stmt, 1), i); else op = gimple_op (stmt, i + 1); if (op && !vect_is_simple_use (op, loop_vinfo, &def_stmt, &def, &dt)) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "use not simple."); return false; } if (dt != vect_invariant_def && dt != vect_constant_def) return false; } /* No transformation is required for the cases we currently support. */ return true; } /* Function vect_is_simple_cond. Input: LOOP - the loop that is being vectorized. COND - Condition that is checked for simple use. Returns whether a COND can be vectorized. Checks whether condition operands are supportable using vec_is_simple_use. */ static bool vect_is_simple_cond (tree cond, loop_vec_info loop_vinfo) { tree lhs, rhs; tree def; enum vect_def_type dt; if (!COMPARISON_CLASS_P (cond)) return false; lhs = TREE_OPERAND (cond, 0); rhs = TREE_OPERAND (cond, 1); if (TREE_CODE (lhs) == SSA_NAME) { gimple lhs_def_stmt = SSA_NAME_DEF_STMT (lhs); if (!vect_is_simple_use (lhs, loop_vinfo, &lhs_def_stmt, &def, &dt)) return false; } else if (TREE_CODE (lhs) != INTEGER_CST && TREE_CODE (lhs) != REAL_CST && TREE_CODE (lhs) != FIXED_CST) return false; if (TREE_CODE (rhs) == SSA_NAME) { gimple rhs_def_stmt = SSA_NAME_DEF_STMT (rhs); if (!vect_is_simple_use (rhs, loop_vinfo, &rhs_def_stmt, &def, &dt)) return false; } else if (TREE_CODE (rhs) != INTEGER_CST && TREE_CODE (rhs) != REAL_CST && TREE_CODE (rhs) != FIXED_CST) return false; return true; } /* vectorizable_condition. Check if STMT is conditional modify expression that can be vectorized. If VEC_STMT is also passed, vectorize the STMT: create a vectorized stmt using VEC_COND_EXPR to replace it, put it in VEC_STMT, and insert it at BSI. Return FALSE if not a vectorizable STMT, TRUE otherwise. */ bool vectorizable_condition (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt) { tree scalar_dest = NULL_TREE; tree vec_dest = NULL_TREE; tree op = NULL_TREE; tree cond_expr, then_clause, else_clause; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); tree vectype = STMT_VINFO_VECTYPE (stmt_info); tree vec_cond_lhs, vec_cond_rhs, vec_then_clause, vec_else_clause; tree vec_compare, vec_cond_expr; tree new_temp; loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); enum machine_mode vec_mode; tree def; enum vect_def_type dt; int nunits = TYPE_VECTOR_SUBPARTS (vectype); int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits; enum tree_code code; gcc_assert (ncopies >= 1); if (ncopies > 1) return false; /* FORNOW */ if (!STMT_VINFO_RELEVANT_P (stmt_info)) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_loop_def) return false; /* FORNOW: SLP not supported. */ if (STMT_SLP_TYPE (stmt_info)) return false; /* FORNOW: not yet supported. */ if (STMT_VINFO_LIVE_P (stmt_info)) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "value used after loop."); return false; } /* Is vectorizable conditional operation? */ if (!is_gimple_assign (stmt)) return false; code = gimple_assign_rhs_code (stmt); if (code != COND_EXPR) return false; gcc_assert (gimple_assign_single_p (stmt)); op = gimple_assign_rhs1 (stmt); cond_expr = TREE_OPERAND (op, 0); then_clause = TREE_OPERAND (op, 1); else_clause = TREE_OPERAND (op, 2); if (!vect_is_simple_cond (cond_expr, loop_vinfo)) return false; /* We do not handle two different vector types for the condition and the values. */ if (TREE_TYPE (TREE_OPERAND (cond_expr, 0)) != TREE_TYPE (vectype)) return false; if (TREE_CODE (then_clause) == SSA_NAME) { gimple then_def_stmt = SSA_NAME_DEF_STMT (then_clause); if (!vect_is_simple_use (then_clause, loop_vinfo, &then_def_stmt, &def, &dt)) return false; } else if (TREE_CODE (then_clause) != INTEGER_CST && TREE_CODE (then_clause) != REAL_CST && TREE_CODE (then_clause) != FIXED_CST) return false; if (TREE_CODE (else_clause) == SSA_NAME) { gimple else_def_stmt = SSA_NAME_DEF_STMT (else_clause); if (!vect_is_simple_use (else_clause, loop_vinfo, &else_def_stmt, &def, &dt)) return false; } else if (TREE_CODE (else_clause) != INTEGER_CST && TREE_CODE (else_clause) != REAL_CST && TREE_CODE (else_clause) != FIXED_CST) return false; vec_mode = TYPE_MODE (vectype); if (!vec_stmt) { STMT_VINFO_TYPE (stmt_info) = condition_vec_info_type; return expand_vec_cond_expr_p (op, vec_mode); } /* Transform */ /* Handle def. */ scalar_dest = gimple_assign_lhs (stmt); vec_dest = vect_create_destination_var (scalar_dest, vectype); /* Handle cond expr. */ vec_cond_lhs = vect_get_vec_def_for_operand (TREE_OPERAND (cond_expr, 0), stmt, NULL); vec_cond_rhs = vect_get_vec_def_for_operand (TREE_OPERAND (cond_expr, 1), stmt, NULL); vec_then_clause = vect_get_vec_def_for_operand (then_clause, stmt, NULL); vec_else_clause = vect_get_vec_def_for_operand (else_clause, stmt, NULL); /* Arguments are ready. Create the new vector stmt. */ vec_compare = build2 (TREE_CODE (cond_expr), vectype, vec_cond_lhs, vec_cond_rhs); vec_cond_expr = build3 (VEC_COND_EXPR, vectype, vec_compare, vec_then_clause, vec_else_clause); *vec_stmt = gimple_build_assign (vec_dest, vec_cond_expr); new_temp = make_ssa_name (vec_dest, *vec_stmt); gimple_assign_set_lhs (*vec_stmt, new_temp); vect_finish_stmt_generation (stmt, *vec_stmt, gsi); return true; } /* Function vect_transform_stmt. Create a vectorized stmt to replace STMT, and insert it at BSI. */ static bool vect_transform_stmt (gimple stmt, gimple_stmt_iterator *gsi, bool *strided_store, slp_tree slp_node, slp_instance slp_node_instance) { bool is_store = false; gimple vec_stmt = NULL; stmt_vec_info stmt_info = vinfo_for_stmt (stmt); gimple orig_stmt_in_pattern; bool done; loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); switch (STMT_VINFO_TYPE (stmt_info)) { case type_demotion_vec_info_type: done = vectorizable_type_demotion (stmt, gsi, &vec_stmt, slp_node); gcc_assert (done); break; case type_promotion_vec_info_type: done = vectorizable_type_promotion (stmt, gsi, &vec_stmt, slp_node); gcc_assert (done); break; case type_conversion_vec_info_type: done = vectorizable_conversion (stmt, gsi, &vec_stmt, slp_node); gcc_assert (done); break; case induc_vec_info_type: gcc_assert (!slp_node); done = vectorizable_induction (stmt, gsi, &vec_stmt); gcc_assert (done); break; case op_vec_info_type: done = vectorizable_operation (stmt, gsi, &vec_stmt, slp_node); gcc_assert (done); break; case assignment_vec_info_type: done = vectorizable_assignment (stmt, gsi, &vec_stmt, slp_node); gcc_assert (done); break; case load_vec_info_type: done = vectorizable_load (stmt, gsi, &vec_stmt, slp_node, slp_node_instance); gcc_assert (done); break; case store_vec_info_type: done = vectorizable_store (stmt, gsi, &vec_stmt, slp_node); gcc_assert (done); if (STMT_VINFO_STRIDED_ACCESS (stmt_info) && !slp_node) { /* In case of interleaving, the whole chain is vectorized when the last store in the chain is reached. Store stmts before the last one are skipped, and there vec_stmt_info shouldn't be freed meanwhile. */ *strided_store = true; if (STMT_VINFO_VEC_STMT (stmt_info)) is_store = true; } else is_store = true; break; case condition_vec_info_type: gcc_assert (!slp_node); done = vectorizable_condition (stmt, gsi, &vec_stmt); gcc_assert (done); break; case call_vec_info_type: gcc_assert (!slp_node); done = vectorizable_call (stmt, gsi, &vec_stmt); break; case reduc_vec_info_type: gcc_assert (!slp_node); done = vectorizable_reduction (stmt, gsi, &vec_stmt); gcc_assert (done); break; default: if (!STMT_VINFO_LIVE_P (stmt_info)) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "stmt not supported."); gcc_unreachable (); } } /* Handle inner-loop stmts whose DEF is used in the loop-nest that is being vectorized, but outside the immediately enclosing loop. */ if (vec_stmt && nested_in_vect_loop_p (loop, stmt) && STMT_VINFO_TYPE (stmt_info) != reduc_vec_info_type && (STMT_VINFO_RELEVANT (stmt_info) == vect_used_in_outer || STMT_VINFO_RELEVANT (stmt_info) == vect_used_in_outer_by_reduction)) { struct loop *innerloop = loop->inner; imm_use_iterator imm_iter; use_operand_p use_p; tree scalar_dest; gimple exit_phi; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "Record the vdef for outer-loop vectorization."); /* Find the relevant loop-exit phi-node, and reord the vec_stmt there (to be used when vectorizing outer-loop stmts that use the DEF of STMT). */ if (gimple_code (stmt) == GIMPLE_PHI) scalar_dest = PHI_RESULT (stmt); else scalar_dest = gimple_assign_lhs (stmt); FOR_EACH_IMM_USE_FAST (use_p, imm_iter, scalar_dest) { if (!flow_bb_inside_loop_p (innerloop, gimple_bb (USE_STMT (use_p)))) { exit_phi = USE_STMT (use_p); STMT_VINFO_VEC_STMT (vinfo_for_stmt (exit_phi)) = vec_stmt; } } } /* Handle stmts whose DEF is used outside the loop-nest that is being vectorized. */ if (STMT_VINFO_LIVE_P (stmt_info) && STMT_VINFO_TYPE (stmt_info) != reduc_vec_info_type) { done = vectorizable_live_operation (stmt, gsi, &vec_stmt); gcc_assert (done); } if (vec_stmt) { STMT_VINFO_VEC_STMT (stmt_info) = vec_stmt; orig_stmt_in_pattern = STMT_VINFO_RELATED_STMT (stmt_info); if (orig_stmt_in_pattern) { stmt_vec_info stmt_vinfo = vinfo_for_stmt (orig_stmt_in_pattern); /* STMT was inserted by the vectorizer to replace a computation idiom. ORIG_STMT_IN_PATTERN is a stmt in the original sequence that computed this idiom. We need to record a pointer to VEC_STMT in the stmt_info of ORIG_STMT_IN_PATTERN. See more details in the documentation of vect_pattern_recog. */ if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo)) { gcc_assert (STMT_VINFO_RELATED_STMT (stmt_vinfo) == stmt); STMT_VINFO_VEC_STMT (stmt_vinfo) = vec_stmt; } } } return is_store; } /* This function builds ni_name = number of iterations loop executes on the loop preheader. */ static tree vect_build_loop_niters (loop_vec_info loop_vinfo) { tree ni_name, var; gimple_seq stmts = NULL; edge pe; struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); tree ni = unshare_expr (LOOP_VINFO_NITERS (loop_vinfo)); var = create_tmp_var (TREE_TYPE (ni), "niters"); add_referenced_var (var); ni_name = force_gimple_operand (ni, &stmts, false, var); pe = loop_preheader_edge (loop); if (stmts) { basic_block new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); gcc_assert (!new_bb); } return ni_name; } /* This function generates the following statements: ni_name = number of iterations loop executes ratio = ni_name / vf ratio_mult_vf_name = ratio * vf and places them at the loop preheader edge. */ static void vect_generate_tmps_on_preheader (loop_vec_info loop_vinfo, tree *ni_name_ptr, tree *ratio_mult_vf_name_ptr, tree *ratio_name_ptr) { edge pe; basic_block new_bb; gimple_seq stmts; tree ni_name; tree var; tree ratio_name; tree ratio_mult_vf_name; struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); tree ni = LOOP_VINFO_NITERS (loop_vinfo); int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); tree log_vf; pe = loop_preheader_edge (loop); /* Generate temporary variable that contains number of iterations loop executes. */ ni_name = vect_build_loop_niters (loop_vinfo); log_vf = build_int_cst (TREE_TYPE (ni), exact_log2 (vf)); /* Create: ratio = ni >> log2(vf) */ ratio_name = fold_build2 (RSHIFT_EXPR, TREE_TYPE (ni_name), ni_name, log_vf); if (!is_gimple_val (ratio_name)) { var = create_tmp_var (TREE_TYPE (ni), "bnd"); add_referenced_var (var); stmts = NULL; ratio_name = force_gimple_operand (ratio_name, &stmts, true, var); pe = loop_preheader_edge (loop); new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); gcc_assert (!new_bb); } /* Create: ratio_mult_vf = ratio << log2 (vf). */ ratio_mult_vf_name = fold_build2 (LSHIFT_EXPR, TREE_TYPE (ratio_name), ratio_name, log_vf); if (!is_gimple_val (ratio_mult_vf_name)) { var = create_tmp_var (TREE_TYPE (ni), "ratio_mult_vf"); add_referenced_var (var); stmts = NULL; ratio_mult_vf_name = force_gimple_operand (ratio_mult_vf_name, &stmts, true, var); pe = loop_preheader_edge (loop); new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); gcc_assert (!new_bb); } *ni_name_ptr = ni_name; *ratio_mult_vf_name_ptr = ratio_mult_vf_name; *ratio_name_ptr = ratio_name; return; } /* Function vect_update_ivs_after_vectorizer. "Advance" the induction variables of LOOP to the value they should take after the execution of LOOP. This is currently necessary because the vectorizer does not handle induction variables that are used after the loop. Such a situation occurs when the last iterations of LOOP are peeled, because: 1. We introduced new uses after LOOP for IVs that were not originally used after LOOP: the IVs of LOOP are now used by an epilog loop. 2. LOOP is going to be vectorized; this means that it will iterate N/VF times, whereas the loop IVs should be bumped N times. Input: - LOOP - a loop that is going to be vectorized. The last few iterations of LOOP were peeled. - NITERS - the number of iterations that LOOP executes (before it is vectorized). i.e, the number of times the ivs should be bumped. - UPDATE_E - a successor edge of LOOP->exit that is on the (only) path coming out from LOOP on which there are uses of the LOOP ivs (this is the path from LOOP->exit to epilog_loop->preheader). The new definitions of the ivs are placed in LOOP->exit. The phi args associated with the edge UPDATE_E in the bb UPDATE_E->dest are updated accordingly. Assumption 1: Like the rest of the vectorizer, this function assumes a single loop exit that has a single predecessor. Assumption 2: The phi nodes in the LOOP header and in update_bb are organized in the same order. Assumption 3: The access function of the ivs is simple enough (see vect_can_advance_ivs_p). This assumption will be relaxed in the future. Assumption 4: Exactly one of the successors of LOOP exit-bb is on a path coming out of LOOP on which the ivs of LOOP are used (this is the path that leads to the epilog loop; other paths skip the epilog loop). This path starts with the edge UPDATE_E, and its destination (denoted update_bb) needs to have its phis updated. */ static void vect_update_ivs_after_vectorizer (loop_vec_info loop_vinfo, tree niters, edge update_e) { struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); basic_block exit_bb = single_exit (loop)->dest; gimple phi, phi1; gimple_stmt_iterator gsi, gsi1; basic_block update_bb = update_e->dest; /* gcc_assert (vect_can_advance_ivs_p (loop_vinfo)); */ /* Make sure there exists a single-predecessor exit bb: */ gcc_assert (single_pred_p (exit_bb)); for (gsi = gsi_start_phis (loop->header), gsi1 = gsi_start_phis (update_bb); !gsi_end_p (gsi) && !gsi_end_p (gsi1); gsi_next (&gsi), gsi_next (&gsi1)) { tree access_fn = NULL; tree evolution_part; tree init_expr; tree step_expr; tree var, ni, ni_name; gimple_stmt_iterator last_gsi; phi = gsi_stmt (gsi); phi1 = gsi_stmt (gsi1); if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "vect_update_ivs_after_vectorizer: phi: "); print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); } /* Skip virtual phi's. */ if (!is_gimple_reg (SSA_NAME_VAR (PHI_RESULT (phi)))) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "virtual phi. skip."); continue; } /* Skip reduction phis. */ if (STMT_VINFO_DEF_TYPE (vinfo_for_stmt (phi)) == vect_reduction_def) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "reduc phi. skip."); continue; } access_fn = analyze_scalar_evolution (loop, PHI_RESULT (phi)); gcc_assert (access_fn); STRIP_NOPS (access_fn); evolution_part = unshare_expr (evolution_part_in_loop_num (access_fn, loop->num)); gcc_assert (evolution_part != NULL_TREE); /* FORNOW: We do not support IVs whose evolution function is a polynomial of degree >= 2 or exponential. */ gcc_assert (!tree_is_chrec (evolution_part)); step_expr = evolution_part; init_expr = unshare_expr (initial_condition_in_loop_num (access_fn, loop->num)); if (POINTER_TYPE_P (TREE_TYPE (init_expr))) ni = fold_build2 (POINTER_PLUS_EXPR, TREE_TYPE (init_expr), init_expr, fold_convert (sizetype, fold_build2 (MULT_EXPR, TREE_TYPE (niters), niters, step_expr))); else ni = fold_build2 (PLUS_EXPR, TREE_TYPE (init_expr), fold_build2 (MULT_EXPR, TREE_TYPE (init_expr), fold_convert (TREE_TYPE (init_expr), niters), step_expr), init_expr); var = create_tmp_var (TREE_TYPE (init_expr), "tmp"); add_referenced_var (var); last_gsi = gsi_last_bb (exit_bb); ni_name = force_gimple_operand_gsi (&last_gsi, ni, false, var, true, GSI_SAME_STMT); /* Fix phi expressions in the successor bb. */ SET_PHI_ARG_DEF (phi1, update_e->dest_idx, ni_name); } } /* Return the more conservative threshold between the min_profitable_iters returned by the cost model and the user specified threshold, if provided. */ static unsigned int conservative_cost_threshold (loop_vec_info loop_vinfo, int min_profitable_iters) { unsigned int th; int min_scalar_loop_bound; min_scalar_loop_bound = ((PARAM_VALUE (PARAM_MIN_VECT_LOOP_BOUND) * LOOP_VINFO_VECT_FACTOR (loop_vinfo)) - 1); /* Use the cost model only if it is more conservative than user specified threshold. */ th = (unsigned) min_scalar_loop_bound; if (min_profitable_iters && (!min_scalar_loop_bound || min_profitable_iters > min_scalar_loop_bound)) th = (unsigned) min_profitable_iters; if (th && vect_print_dump_info (REPORT_COST)) fprintf (vect_dump, "Vectorization may not be profitable."); return th; } /* Function vect_do_peeling_for_loop_bound Peel the last iterations of the loop represented by LOOP_VINFO. The peeled iterations form a new epilog loop. Given that the loop now iterates NITERS times, the new epilog loop iterates NITERS % VECTORIZATION_FACTOR times. The original loop will later be made to iterate NITERS / VECTORIZATION_FACTOR times (this value is placed into RATIO). */ static void vect_do_peeling_for_loop_bound (loop_vec_info loop_vinfo, tree *ratio) { tree ni_name, ratio_mult_vf_name; struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); struct loop *new_loop; edge update_e; basic_block preheader; int loop_num; bool check_profitability = false; unsigned int th = 0; int min_profitable_iters; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vect_do_peeling_for_loop_bound ==="); initialize_original_copy_tables (); /* Generate the following variables on the preheader of original loop: ni_name = number of iteration the original loop executes ratio = ni_name / vf ratio_mult_vf_name = ratio * vf */ vect_generate_tmps_on_preheader (loop_vinfo, &ni_name, &ratio_mult_vf_name, ratio); loop_num = loop->num; /* If cost model check not done during versioning and peeling for alignment. */ if (!VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)) && !VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo)) && !LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo)) { check_profitability = true; /* Get profitability threshold for vectorized loop. */ min_profitable_iters = LOOP_VINFO_COST_MODEL_MIN_ITERS (loop_vinfo); th = conservative_cost_threshold (loop_vinfo, min_profitable_iters); } new_loop = slpeel_tree_peel_loop_to_edge (loop, single_exit (loop), ratio_mult_vf_name, ni_name, false, th, check_profitability); gcc_assert (new_loop); gcc_assert (loop_num == loop->num); #ifdef ENABLE_CHECKING slpeel_verify_cfg_after_peeling (loop, new_loop); #endif /* A guard that controls whether the new_loop is to be executed or skipped is placed in LOOP->exit. LOOP->exit therefore has two successors - one is the preheader of NEW_LOOP, where the IVs from LOOP are used. The other is a bb after NEW_LOOP, where these IVs are not used. Find the edge that is on the path where the LOOP IVs are used and need to be updated. */ preheader = loop_preheader_edge (new_loop)->src; if (EDGE_PRED (preheader, 0)->src == single_exit (loop)->dest) update_e = EDGE_PRED (preheader, 0); else update_e = EDGE_PRED (preheader, 1); /* Update IVs of original loop as if they were advanced by ratio_mult_vf_name steps. */ vect_update_ivs_after_vectorizer (loop_vinfo, ratio_mult_vf_name, update_e); /* After peeling we have to reset scalar evolution analyzer. */ scev_reset (); free_original_copy_tables (); } /* Function vect_gen_niters_for_prolog_loop Set the number of iterations for the loop represented by LOOP_VINFO to the minimum between LOOP_NITERS (the original iteration count of the loop) and the misalignment of DR - the data reference recorded in LOOP_VINFO_UNALIGNED_DR (LOOP_VINFO). As a result, after the execution of this loop, the data reference DR will refer to an aligned location. The following computation is generated: If the misalignment of DR is known at compile time: addr_mis = int mis = DR_MISALIGNMENT (dr); Else, compute address misalignment in bytes: addr_mis = addr & (vectype_size - 1) prolog_niters = min (LOOP_NITERS, ((VF - addr_mis/elem_size)&(VF-1))/step) (elem_size = element type size; an element is the scalar element whose type is the inner type of the vectype) When the step of the data-ref in the loop is not 1 (as in interleaved data and SLP), the number of iterations of the prolog must be divided by the step (which is equal to the size of interleaved group). The above formulas assume that VF == number of elements in the vector. This may not hold when there are multiple-types in the loop. In this case, for some data-references in the loop the VF does not represent the number of elements that fit in the vector. Therefore, instead of VF we use TYPE_VECTOR_SUBPARTS. */ static tree vect_gen_niters_for_prolog_loop (loop_vec_info loop_vinfo, tree loop_niters) { struct data_reference *dr = LOOP_VINFO_UNALIGNED_DR (loop_vinfo); struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); tree var; gimple_seq stmts; tree iters, iters_name; edge pe; basic_block new_bb; gimple dr_stmt = DR_STMT (dr); stmt_vec_info stmt_info = vinfo_for_stmt (dr_stmt); tree vectype = STMT_VINFO_VECTYPE (stmt_info); int vectype_align = TYPE_ALIGN (vectype) / BITS_PER_UNIT; tree niters_type = TREE_TYPE (loop_niters); int step = 1; int element_size = GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (dr)))); int nelements = TYPE_VECTOR_SUBPARTS (vectype); if (STMT_VINFO_STRIDED_ACCESS (stmt_info)) step = DR_GROUP_SIZE (vinfo_for_stmt (DR_GROUP_FIRST_DR (stmt_info))); pe = loop_preheader_edge (loop); if (LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo) > 0) { int byte_misalign = LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo); int elem_misalign = byte_misalign / element_size; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "known alignment = %d.", byte_misalign); iters = build_int_cst (niters_type, (((nelements - elem_misalign) & (nelements - 1)) / step)); } else { gimple_seq new_stmts = NULL; tree start_addr = vect_create_addr_base_for_vector_ref (dr_stmt, &new_stmts, NULL_TREE, loop); tree ptr_type = TREE_TYPE (start_addr); tree size = TYPE_SIZE (ptr_type); tree type = lang_hooks.types.type_for_size (tree_low_cst (size, 1), 1); tree vectype_size_minus_1 = build_int_cst (type, vectype_align - 1); tree elem_size_log = build_int_cst (type, exact_log2 (vectype_align/nelements)); tree nelements_minus_1 = build_int_cst (type, nelements - 1); tree nelements_tree = build_int_cst (type, nelements); tree byte_misalign; tree elem_misalign; new_bb = gsi_insert_seq_on_edge_immediate (pe, new_stmts); gcc_assert (!new_bb); /* Create: byte_misalign = addr & (vectype_size - 1) */ byte_misalign = fold_build2 (BIT_AND_EXPR, type, fold_convert (type, start_addr), vectype_size_minus_1); /* Create: elem_misalign = byte_misalign / element_size */ elem_misalign = fold_build2 (RSHIFT_EXPR, type, byte_misalign, elem_size_log); /* Create: (niters_type) (nelements - elem_misalign)&(nelements - 1) */ iters = fold_build2 (MINUS_EXPR, type, nelements_tree, elem_misalign); iters = fold_build2 (BIT_AND_EXPR, type, iters, nelements_minus_1); iters = fold_convert (niters_type, iters); } /* Create: prolog_loop_niters = min (iters, loop_niters) */ /* If the loop bound is known at compile time we already verified that it is greater than vf; since the misalignment ('iters') is at most vf, there's no need to generate the MIN_EXPR in this case. */ if (TREE_CODE (loop_niters) != INTEGER_CST) iters = fold_build2 (MIN_EXPR, niters_type, iters, loop_niters); if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "niters for prolog loop: "); print_generic_expr (vect_dump, iters, TDF_SLIM); } var = create_tmp_var (niters_type, "prolog_loop_niters"); add_referenced_var (var); stmts = NULL; iters_name = force_gimple_operand (iters, &stmts, false, var); /* Insert stmt on loop preheader edge. */ if (stmts) { basic_block new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); gcc_assert (!new_bb); } return iters_name; } /* Function vect_update_init_of_dr NITERS iterations were peeled from LOOP. DR represents a data reference in LOOP. This function updates the information recorded in DR to account for the fact that the first NITERS iterations had already been executed. Specifically, it updates the OFFSET field of DR. */ static void vect_update_init_of_dr (struct data_reference *dr, tree niters) { tree offset = DR_OFFSET (dr); niters = fold_build2 (MULT_EXPR, sizetype, fold_convert (sizetype, niters), fold_convert (sizetype, DR_STEP (dr))); offset = fold_build2 (PLUS_EXPR, sizetype, offset, niters); DR_OFFSET (dr) = offset; } /* Function vect_update_inits_of_drs NITERS iterations were peeled from the loop represented by LOOP_VINFO. This function updates the information recorded for the data references in the loop to account for the fact that the first NITERS iterations had already been executed. Specifically, it updates the initial_condition of the access_function of all the data_references in the loop. */ static void vect_update_inits_of_drs (loop_vec_info loop_vinfo, tree niters) { unsigned int i; VEC (data_reference_p, heap) *datarefs = LOOP_VINFO_DATAREFS (loop_vinfo); struct data_reference *dr; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vect_update_inits_of_dr ==="); for (i = 0; VEC_iterate (data_reference_p, datarefs, i, dr); i++) vect_update_init_of_dr (dr, niters); } /* Function vect_do_peeling_for_alignment Peel the first 'niters' iterations of the loop represented by LOOP_VINFO. 'niters' is set to the misalignment of one of the data references in the loop, thereby forcing it to refer to an aligned location at the beginning of the execution of this loop. The data reference for which we are peeling is recorded in LOOP_VINFO_UNALIGNED_DR. */ static void vect_do_peeling_for_alignment (loop_vec_info loop_vinfo) { struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); tree niters_of_prolog_loop, ni_name; tree n_iters; struct loop *new_loop; bool check_profitability = false; unsigned int th = 0; int min_profitable_iters; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vect_do_peeling_for_alignment ==="); initialize_original_copy_tables (); ni_name = vect_build_loop_niters (loop_vinfo); niters_of_prolog_loop = vect_gen_niters_for_prolog_loop (loop_vinfo, ni_name); /* If cost model check not done during versioning. */ if (!VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)) && !VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo))) { check_profitability = true; /* Get profitability threshold for vectorized loop. */ min_profitable_iters = LOOP_VINFO_COST_MODEL_MIN_ITERS (loop_vinfo); th = conservative_cost_threshold (loop_vinfo, min_profitable_iters); } /* Peel the prolog loop and iterate it niters_of_prolog_loop. */ new_loop = slpeel_tree_peel_loop_to_edge (loop, loop_preheader_edge (loop), niters_of_prolog_loop, ni_name, true, th, check_profitability); gcc_assert (new_loop); #ifdef ENABLE_CHECKING slpeel_verify_cfg_after_peeling (new_loop, loop); #endif /* Update number of times loop executes. */ n_iters = LOOP_VINFO_NITERS (loop_vinfo); LOOP_VINFO_NITERS (loop_vinfo) = fold_build2 (MINUS_EXPR, TREE_TYPE (n_iters), n_iters, niters_of_prolog_loop); /* Update the init conditions of the access functions of all data refs. */ vect_update_inits_of_drs (loop_vinfo, niters_of_prolog_loop); /* After peeling we have to reset scalar evolution analyzer. */ scev_reset (); free_original_copy_tables (); } /* Function vect_create_cond_for_align_checks. Create a conditional expression that represents the alignment checks for all of data references (array element references) whose alignment must be checked at runtime. Input: COND_EXPR - input conditional expression. New conditions will be chained with logical AND operation. LOOP_VINFO - two fields of the loop information are used. LOOP_VINFO_PTR_MASK is the mask used to check the alignment. LOOP_VINFO_MAY_MISALIGN_STMTS contains the refs to be checked. Output: COND_EXPR_STMT_LIST - statements needed to construct the conditional expression. The returned value is the conditional expression to be used in the if statement that controls which version of the loop gets executed at runtime. The algorithm makes two assumptions: 1) The number of bytes "n" in a vector is a power of 2. 2) An address "a" is aligned if a%n is zero and that this test can be done as a&(n-1) == 0. For example, for 16 byte vectors the test is a&0xf == 0. */ static void vect_create_cond_for_align_checks (loop_vec_info loop_vinfo, tree *cond_expr, gimple_seq *cond_expr_stmt_list) { struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); VEC(gimple,heap) *may_misalign_stmts = LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo); gimple ref_stmt; int mask = LOOP_VINFO_PTR_MASK (loop_vinfo); tree mask_cst; unsigned int i; tree psize; tree int_ptrsize_type; char tmp_name[20]; tree or_tmp_name = NULL_TREE; tree and_tmp, and_tmp_name; gimple and_stmt; tree ptrsize_zero; tree part_cond_expr; /* Check that mask is one less than a power of 2, i.e., mask is all zeros followed by all ones. */ gcc_assert ((mask != 0) && ((mask & (mask+1)) == 0)); /* CHECKME: what is the best integer or unsigned type to use to hold a cast from a pointer value? */ psize = TYPE_SIZE (ptr_type_node); int_ptrsize_type = lang_hooks.types.type_for_size (tree_low_cst (psize, 1), 0); /* Create expression (mask & (dr_1 || ... || dr_n)) where dr_i is the address of the first vector of the i'th data reference. */ for (i = 0; VEC_iterate (gimple, may_misalign_stmts, i, ref_stmt); i++) { gimple_seq new_stmt_list = NULL; tree addr_base; tree addr_tmp, addr_tmp_name; tree or_tmp, new_or_tmp_name; gimple addr_stmt, or_stmt; /* create: addr_tmp = (int)(address_of_first_vector) */ addr_base = vect_create_addr_base_for_vector_ref (ref_stmt, &new_stmt_list, NULL_TREE, loop); if (new_stmt_list != NULL) gimple_seq_add_seq (cond_expr_stmt_list, new_stmt_list); sprintf (tmp_name, "%s%d", "addr2int", i); addr_tmp = create_tmp_var (int_ptrsize_type, tmp_name); add_referenced_var (addr_tmp); addr_tmp_name = make_ssa_name (addr_tmp, NULL); addr_stmt = gimple_build_assign_with_ops (NOP_EXPR, addr_tmp_name, addr_base, NULL_TREE); SSA_NAME_DEF_STMT (addr_tmp_name) = addr_stmt; gimple_seq_add_stmt (cond_expr_stmt_list, addr_stmt); /* The addresses are OR together. */ if (or_tmp_name != NULL_TREE) { /* create: or_tmp = or_tmp | addr_tmp */ sprintf (tmp_name, "%s%d", "orptrs", i); or_tmp = create_tmp_var (int_ptrsize_type, tmp_name); add_referenced_var (or_tmp); new_or_tmp_name = make_ssa_name (or_tmp, NULL); or_stmt = gimple_build_assign_with_ops (BIT_IOR_EXPR, new_or_tmp_name, or_tmp_name, addr_tmp_name); SSA_NAME_DEF_STMT (new_or_tmp_name) = or_stmt; gimple_seq_add_stmt (cond_expr_stmt_list, or_stmt); or_tmp_name = new_or_tmp_name; } else or_tmp_name = addr_tmp_name; } /* end for i */ mask_cst = build_int_cst (int_ptrsize_type, mask); /* create: and_tmp = or_tmp & mask */ and_tmp = create_tmp_var (int_ptrsize_type, "andmask" ); add_referenced_var (and_tmp); and_tmp_name = make_ssa_name (and_tmp, NULL); and_stmt = gimple_build_assign_with_ops (BIT_AND_EXPR, and_tmp_name, or_tmp_name, mask_cst); SSA_NAME_DEF_STMT (and_tmp_name) = and_stmt; gimple_seq_add_stmt (cond_expr_stmt_list, and_stmt); /* Make and_tmp the left operand of the conditional test against zero. if and_tmp has a nonzero bit then some address is unaligned. */ ptrsize_zero = build_int_cst (int_ptrsize_type, 0); part_cond_expr = fold_build2 (EQ_EXPR, boolean_type_node, and_tmp_name, ptrsize_zero); if (*cond_expr) *cond_expr = fold_build2 (TRUTH_AND_EXPR, boolean_type_node, *cond_expr, part_cond_expr); else *cond_expr = part_cond_expr; } /* Function vect_vfa_segment_size. Create an expression that computes the size of segment that will be accessed for a data reference. The functions takes into account that realignment loads may access one more vector. Input: DR: The data reference. VECT_FACTOR: vectorization factor. Return an expression whose value is the size of segment which will be accessed by DR. */ static tree vect_vfa_segment_size (struct data_reference *dr, tree vect_factor) { tree segment_length = fold_build2 (MULT_EXPR, integer_type_node, DR_STEP (dr), vect_factor); if (vect_supportable_dr_alignment (dr) == dr_explicit_realign_optimized) { tree vector_size = TYPE_SIZE_UNIT (STMT_VINFO_VECTYPE (vinfo_for_stmt (DR_STMT (dr)))); segment_length = fold_build2 (PLUS_EXPR, integer_type_node, segment_length, vector_size); } return fold_convert (sizetype, segment_length); } /* Function vect_create_cond_for_alias_checks. Create a conditional expression that represents the run-time checks for overlapping of address ranges represented by a list of data references relations passed as input. Input: COND_EXPR - input conditional expression. New conditions will be chained with logical AND operation. LOOP_VINFO - field LOOP_VINFO_MAY_ALIAS_STMTS contains the list of ddrs to be checked. Output: COND_EXPR - conditional expression. COND_EXPR_STMT_LIST - statements needed to construct the conditional expression. The returned value is the conditional expression to be used in the if statement that controls which version of the loop gets executed at runtime. */ static void vect_create_cond_for_alias_checks (loop_vec_info loop_vinfo, tree * cond_expr, gimple_seq * cond_expr_stmt_list) { struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); VEC (ddr_p, heap) * may_alias_ddrs = LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo); tree vect_factor = build_int_cst (integer_type_node, LOOP_VINFO_VECT_FACTOR (loop_vinfo)); ddr_p ddr; unsigned int i; tree part_cond_expr; /* Create expression ((store_ptr_0 + store_segment_length_0) < load_ptr_0) || (load_ptr_0 + load_segment_length_0) < store_ptr_0)) && ... && ((store_ptr_n + store_segment_length_n) < load_ptr_n) || (load_ptr_n + load_segment_length_n) < store_ptr_n)) */ if (VEC_empty (ddr_p, may_alias_ddrs)) return; for (i = 0; VEC_iterate (ddr_p, may_alias_ddrs, i, ddr); i++) { struct data_reference *dr_a, *dr_b; gimple dr_group_first_a, dr_group_first_b; tree addr_base_a, addr_base_b; tree segment_length_a, segment_length_b; gimple stmt_a, stmt_b; dr_a = DDR_A (ddr); stmt_a = DR_STMT (DDR_A (ddr)); dr_group_first_a = DR_GROUP_FIRST_DR (vinfo_for_stmt (stmt_a)); if (dr_group_first_a) { stmt_a = dr_group_first_a; dr_a = STMT_VINFO_DATA_REF (vinfo_for_stmt (stmt_a)); } dr_b = DDR_B (ddr); stmt_b = DR_STMT (DDR_B (ddr)); dr_group_first_b = DR_GROUP_FIRST_DR (vinfo_for_stmt (stmt_b)); if (dr_group_first_b) { stmt_b = dr_group_first_b; dr_b = STMT_VINFO_DATA_REF (vinfo_for_stmt (stmt_b)); } addr_base_a = vect_create_addr_base_for_vector_ref (stmt_a, cond_expr_stmt_list, NULL_TREE, loop); addr_base_b = vect_create_addr_base_for_vector_ref (stmt_b, cond_expr_stmt_list, NULL_TREE, loop); segment_length_a = vect_vfa_segment_size (dr_a, vect_factor); segment_length_b = vect_vfa_segment_size (dr_b, vect_factor); if (vect_print_dump_info (REPORT_DR_DETAILS)) { fprintf (vect_dump, "create runtime check for data references "); print_generic_expr (vect_dump, DR_REF (dr_a), TDF_SLIM); fprintf (vect_dump, " and "); print_generic_expr (vect_dump, DR_REF (dr_b), TDF_SLIM); } part_cond_expr = fold_build2 (TRUTH_OR_EXPR, boolean_type_node, fold_build2 (LT_EXPR, boolean_type_node, fold_build2 (POINTER_PLUS_EXPR, TREE_TYPE (addr_base_a), addr_base_a, segment_length_a), addr_base_b), fold_build2 (LT_EXPR, boolean_type_node, fold_build2 (POINTER_PLUS_EXPR, TREE_TYPE (addr_base_b), addr_base_b, segment_length_b), addr_base_a)); if (*cond_expr) *cond_expr = fold_build2 (TRUTH_AND_EXPR, boolean_type_node, *cond_expr, part_cond_expr); else *cond_expr = part_cond_expr; } if (vect_print_dump_info (REPORT_VECTORIZED_LOOPS)) fprintf (vect_dump, "created %u versioning for alias checks.\n", VEC_length (ddr_p, may_alias_ddrs)); } /* Function vect_loop_versioning. If the loop has data references that may or may not be aligned or/and has data reference relations whose independence was not proven then two versions of the loop need to be generated, one which is vectorized and one which isn't. A test is then generated to control which of the loops is executed. The test checks for the alignment of all of the data references that may or may not be aligned. An additional sequence of runtime tests is generated for each pairs of DDRs whose independence was not proven. The vectorized version of loop is executed only if both alias and alignment tests are passed. The test generated to check which version of loop is executed is modified to also check for profitability as indicated by the cost model initially. */ static void vect_loop_versioning (loop_vec_info loop_vinfo) { struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); struct loop *nloop; tree cond_expr = NULL_TREE; gimple_seq cond_expr_stmt_list = NULL; basic_block condition_bb; gimple_stmt_iterator gsi, cond_exp_gsi; basic_block merge_bb; basic_block new_exit_bb; edge new_exit_e, e; gimple orig_phi, new_phi; tree arg; unsigned prob = 4 * REG_BR_PROB_BASE / 5; gimple_seq gimplify_stmt_list = NULL; tree scalar_loop_iters = LOOP_VINFO_NITERS (loop_vinfo); int min_profitable_iters = 0; unsigned int th; /* Get profitability threshold for vectorized loop. */ min_profitable_iters = LOOP_VINFO_COST_MODEL_MIN_ITERS (loop_vinfo); th = conservative_cost_threshold (loop_vinfo, min_profitable_iters); cond_expr = fold_build2 (GT_EXPR, boolean_type_node, scalar_loop_iters, build_int_cst (TREE_TYPE (scalar_loop_iters), th)); cond_expr = force_gimple_operand (cond_expr, &cond_expr_stmt_list, false, NULL_TREE); if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo))) vect_create_cond_for_align_checks (loop_vinfo, &cond_expr, &cond_expr_stmt_list); if (VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo))) vect_create_cond_for_alias_checks (loop_vinfo, &cond_expr, &cond_expr_stmt_list); cond_expr = fold_build2 (NE_EXPR, boolean_type_node, cond_expr, integer_zero_node); cond_expr = force_gimple_operand (cond_expr, &gimplify_stmt_list, true, NULL_TREE); gimple_seq_add_seq (&cond_expr_stmt_list, gimplify_stmt_list); initialize_original_copy_tables (); nloop = loop_version (loop, cond_expr, &condition_bb, prob, prob, REG_BR_PROB_BASE - prob, true); free_original_copy_tables(); /* Loop versioning violates an assumption we try to maintain during vectorization - that the loop exit block has a single predecessor. After versioning, the exit block of both loop versions is the same basic block (i.e. it has two predecessors). Just in order to simplify following transformations in the vectorizer, we fix this situation here by adding a new (empty) block on the exit-edge of the loop, with the proper loop-exit phis to maintain loop-closed-form. */ merge_bb = single_exit (loop)->dest; gcc_assert (EDGE_COUNT (merge_bb->preds) == 2); new_exit_bb = split_edge (single_exit (loop)); new_exit_e = single_exit (loop); e = EDGE_SUCC (new_exit_bb, 0); for (gsi = gsi_start_phis (merge_bb); !gsi_end_p (gsi); gsi_next (&gsi)) { orig_phi = gsi_stmt (gsi); new_phi = create_phi_node (SSA_NAME_VAR (PHI_RESULT (orig_phi)), new_exit_bb); arg = PHI_ARG_DEF_FROM_EDGE (orig_phi, e); add_phi_arg (new_phi, arg, new_exit_e, gimple_phi_arg_location_from_edge (orig_phi, e)); SET_PHI_ARG_DEF (orig_phi, e->dest_idx, PHI_RESULT (new_phi)); } /* End loop-exit-fixes after versioning. */ update_ssa (TODO_update_ssa); if (cond_expr_stmt_list) { cond_exp_gsi = gsi_last_bb (condition_bb); gsi_insert_seq_before (&cond_exp_gsi, cond_expr_stmt_list, GSI_SAME_STMT); } } /* Remove a group of stores (for SLP or interleaving), free their stmt_vec_info. */ static void vect_remove_stores (gimple first_stmt) { gimple next = first_stmt; gimple tmp; gimple_stmt_iterator next_si; while (next) { /* Free the attached stmt_vec_info and remove the stmt. */ next_si = gsi_for_stmt (next); gsi_remove (&next_si, true); tmp = DR_GROUP_NEXT_DR (vinfo_for_stmt (next)); free_stmt_vec_info (next); next = tmp; } } /* Vectorize SLP instance tree in postorder. */ static bool vect_schedule_slp_instance (slp_tree node, slp_instance instance, unsigned int vectorization_factor) { gimple stmt; bool strided_store, is_store; gimple_stmt_iterator si; stmt_vec_info stmt_info; unsigned int vec_stmts_size, nunits, group_size; tree vectype; int i; slp_tree loads_node; if (!node) return false; vect_schedule_slp_instance (SLP_TREE_LEFT (node), instance, vectorization_factor); vect_schedule_slp_instance (SLP_TREE_RIGHT (node), instance, vectorization_factor); stmt = VEC_index (gimple, SLP_TREE_SCALAR_STMTS (node), 0); stmt_info = vinfo_for_stmt (stmt); /* VECTYPE is the type of the destination. */ vectype = get_vectype_for_scalar_type (TREE_TYPE (gimple_assign_lhs (stmt))); nunits = (unsigned int) TYPE_VECTOR_SUBPARTS (vectype); group_size = SLP_INSTANCE_GROUP_SIZE (instance); /* For each SLP instance calculate number of vector stmts to be created for the scalar stmts in each node of the SLP tree. Number of vector elements in one vector iteration is the number of scalar elements in one scalar iteration (GROUP_SIZE) multiplied by VF divided by vector size. */ vec_stmts_size = (vectorization_factor * group_size) / nunits; /* In case of load permutation we have to allocate vectorized statements for all the nodes that participate in that permutation. */ if (SLP_INSTANCE_LOAD_PERMUTATION (instance)) { for (i = 0; VEC_iterate (slp_tree, SLP_INSTANCE_LOADS (instance), i, loads_node); i++) { if (!SLP_TREE_VEC_STMTS (loads_node)) { SLP_TREE_VEC_STMTS (loads_node) = VEC_alloc (gimple, heap, vec_stmts_size); SLP_TREE_NUMBER_OF_VEC_STMTS (loads_node) = vec_stmts_size; } } } if (!SLP_TREE_VEC_STMTS (node)) { SLP_TREE_VEC_STMTS (node) = VEC_alloc (gimple, heap, vec_stmts_size); SLP_TREE_NUMBER_OF_VEC_STMTS (node) = vec_stmts_size; } if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "------>vectorizing SLP node starting from: "); print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); } /* Loads should be inserted before the first load. */ if (SLP_INSTANCE_FIRST_LOAD_STMT (instance) && STMT_VINFO_STRIDED_ACCESS (stmt_info) && !REFERENCE_CLASS_P (gimple_get_lhs (stmt))) si = gsi_for_stmt (SLP_INSTANCE_FIRST_LOAD_STMT (instance)); else si = gsi_for_stmt (stmt); is_store = vect_transform_stmt (stmt, &si, &strided_store, node, instance); if (is_store) { if (DR_GROUP_FIRST_DR (stmt_info)) /* If IS_STORE is TRUE, the vectorization of the interleaving chain was completed - free all the stores in the chain. */ vect_remove_stores (DR_GROUP_FIRST_DR (stmt_info)); else /* FORNOW: SLP originates only from strided stores. */ gcc_unreachable (); return true; } /* FORNOW: SLP originates only from strided stores. */ return false; } static bool vect_schedule_slp (loop_vec_info loop_vinfo) { VEC (slp_instance, heap) *slp_instances = LOOP_VINFO_SLP_INSTANCES (loop_vinfo); slp_instance instance; unsigned int i; bool is_store = false; for (i = 0; VEC_iterate (slp_instance, slp_instances, i, instance); i++) { /* Schedule the tree of INSTANCE. */ is_store = vect_schedule_slp_instance (SLP_INSTANCE_TREE (instance), instance, LOOP_VINFO_VECT_FACTOR (loop_vinfo)); if (vect_print_dump_info (REPORT_VECTORIZED_LOOPS) || vect_print_dump_info (REPORT_UNVECTORIZED_LOOPS)) fprintf (vect_dump, "vectorizing stmts using SLP."); } return is_store; } /* Function vect_transform_loop. The analysis phase has determined that the loop is vectorizable. Vectorize the loop - created vectorized stmts to replace the scalar stmts in the loop, and update the loop exit condition. */ void vect_transform_loop (loop_vec_info loop_vinfo) { struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo); int nbbs = loop->num_nodes; gimple_stmt_iterator si; int i; tree ratio = NULL; int vectorization_factor = LOOP_VINFO_VECT_FACTOR (loop_vinfo); bool strided_store; bool slp_scheduled = false; unsigned int nunits; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vec_transform_loop ==="); if (VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)) || VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo))) vect_loop_versioning (loop_vinfo); /* CHECKME: we wouldn't need this if we called update_ssa once for all loops. */ bitmap_zero (vect_memsyms_to_rename); /* Peel the loop if there are data refs with unknown alignment. Only one data ref with unknown store is allowed. */ if (LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo)) vect_do_peeling_for_alignment (loop_vinfo); /* If the loop has a symbolic number of iterations 'n' (i.e. it's not a compile time constant), or it is a constant that doesn't divide by the vectorization factor, then an epilog loop needs to be created. We therefore duplicate the loop: the original loop will be vectorized, and will compute the first (n/VF) iterations. The second copy of the loop will remain scalar and will compute the remaining (n%VF) iterations. (VF is the vectorization factor). */ if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) || (LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) && LOOP_VINFO_INT_NITERS (loop_vinfo) % vectorization_factor != 0)) vect_do_peeling_for_loop_bound (loop_vinfo, &ratio); else ratio = build_int_cst (TREE_TYPE (LOOP_VINFO_NITERS (loop_vinfo)), LOOP_VINFO_INT_NITERS (loop_vinfo) / vectorization_factor); /* 1) Make sure the loop header has exactly two entries 2) Make sure we have a preheader basic block. */ gcc_assert (EDGE_COUNT (loop->header->preds) == 2); split_edge (loop_preheader_edge (loop)); /* FORNOW: the vectorizer supports only loops which body consist of one basic block (header + empty latch). When the vectorizer will support more involved loop forms, the order by which the BBs are traversed need to be reconsidered. */ for (i = 0; i < nbbs; i++) { basic_block bb = bbs[i]; stmt_vec_info stmt_info; gimple phi; for (si = gsi_start_phis (bb); !gsi_end_p (si); gsi_next (&si)) { phi = gsi_stmt (si); if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "------>vectorizing phi: "); print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); } stmt_info = vinfo_for_stmt (phi); if (!stmt_info) continue; if (!STMT_VINFO_RELEVANT_P (stmt_info) && !STMT_VINFO_LIVE_P (stmt_info)) continue; if ((TYPE_VECTOR_SUBPARTS (STMT_VINFO_VECTYPE (stmt_info)) != (unsigned HOST_WIDE_INT) vectorization_factor) && vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "multiple-types."); if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def) { if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform phi."); vect_transform_stmt (phi, NULL, NULL, NULL, NULL); } } for (si = gsi_start_bb (bb); !gsi_end_p (si);) { gimple stmt = gsi_stmt (si); bool is_store; if (vect_print_dump_info (REPORT_DETAILS)) { fprintf (vect_dump, "------>vectorizing statement: "); print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); } stmt_info = vinfo_for_stmt (stmt); /* vector stmts created in the outer-loop during vectorization of stmts in an inner-loop may not have a stmt_info, and do not need to be vectorized. */ if (!stmt_info) { gsi_next (&si); continue; } if (!STMT_VINFO_RELEVANT_P (stmt_info) && !STMT_VINFO_LIVE_P (stmt_info)) { gsi_next (&si); continue; } gcc_assert (STMT_VINFO_VECTYPE (stmt_info)); nunits = (unsigned int) TYPE_VECTOR_SUBPARTS (STMT_VINFO_VECTYPE (stmt_info)); if (!STMT_SLP_TYPE (stmt_info) && nunits != (unsigned int) vectorization_factor && vect_print_dump_info (REPORT_DETAILS)) /* For SLP VF is set according to unrolling factor, and not to vector size, hence for SLP this print is not valid. */ fprintf (vect_dump, "multiple-types."); /* SLP. Schedule all the SLP instances when the first SLP stmt is reached. */ if (STMT_SLP_TYPE (stmt_info)) { if (!slp_scheduled) { slp_scheduled = true; if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== scheduling SLP instances ==="); vect_schedule_slp (loop_vinfo); } /* Hybrid SLP stmts must be vectorized in addition to SLP. */ if (!vinfo_for_stmt (stmt) || PURE_SLP_STMT (stmt_info)) { gsi_next (&si); continue; } } /* -------- vectorize statement ------------ */ if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "transform statement."); strided_store = false; is_store = vect_transform_stmt (stmt, &si, &strided_store, NULL, NULL); if (is_store) { if (STMT_VINFO_STRIDED_ACCESS (stmt_info)) { /* Interleaving. If IS_STORE is TRUE, the vectorization of the interleaving chain was completed - free all the stores in the chain. */ vect_remove_stores (DR_GROUP_FIRST_DR (stmt_info)); gsi_remove (&si, true); continue; } else { /* Free the attached stmt_vec_info and remove the stmt. */ free_stmt_vec_info (stmt); gsi_remove (&si, true); continue; } } gsi_next (&si); } /* stmts in BB */ } /* BBs in loop */ slpeel_make_loop_iterate_ntimes (loop, ratio); mark_set_for_renaming (vect_memsyms_to_rename); /* The memory tags and pointers in vectorized statements need to have their SSA forms updated. FIXME, why can't this be delayed until all the loops have been transformed? */ update_ssa (TODO_update_ssa); if (vect_print_dump_info (REPORT_VECTORIZED_LOOPS)) fprintf (vect_dump, "LOOP VECTORIZED."); if (loop->inner && vect_print_dump_info (REPORT_VECTORIZED_LOOPS)) fprintf (vect_dump, "OUTER LOOP VECTORIZED."); }