/* Vectorizer Specific Loop Manipulations Copyright (C) 2003-2023 Free Software Foundation, Inc. Contributed by Dorit Naishlos and Ira Rosen 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 "backend.h" #include "tree.h" #include "gimple.h" #include "cfghooks.h" #include "tree-pass.h" #include "ssa.h" #include "fold-const.h" #include "cfganal.h" #include "gimplify.h" #include "gimple-iterator.h" #include "gimplify-me.h" #include "tree-cfg.h" #include "tree-ssa-loop-manip.h" #include "tree-into-ssa.h" #include "tree-ssa.h" #include "cfgloop.h" #include "tree-scalar-evolution.h" #include "tree-vectorizer.h" #include "tree-ssa-loop-ivopts.h" #include "gimple-fold.h" #include "tree-ssa-loop-niter.h" #include "internal-fn.h" #include "stor-layout.h" #include "optabs-query.h" #include "vec-perm-indices.h" #include "insn-config.h" #include "rtl.h" #include "recog.h" #include "langhooks.h" #include "tree-vector-builder.h" #include "optabs-tree.h" /************************************************************************* Simple Loop Peeling Utilities Utilities to support loop peeling for vectorization purposes. *************************************************************************/ /* Renames the use *OP_P. */ static void rename_use_op (use_operand_p op_p) { tree new_name; if (TREE_CODE (USE_FROM_PTR (op_p)) != SSA_NAME) return; new_name = get_current_def (USE_FROM_PTR (op_p)); /* Something defined outside of the loop. */ if (!new_name) return; /* An ordinary ssa name defined in the loop. */ SET_USE (op_p, new_name); } /* Renames the variables in basic block BB. Allow renaming of PHI arguments on edges incoming from outer-block header if RENAME_FROM_OUTER_LOOP is true. */ static void rename_variables_in_bb (basic_block bb, bool rename_from_outer_loop) { gimple *stmt; use_operand_p use_p; ssa_op_iter iter; edge e; edge_iterator ei; class loop *loop = bb->loop_father; class loop *outer_loop = NULL; if (rename_from_outer_loop) { gcc_assert (loop); outer_loop = loop_outer (loop); } for (gimple_stmt_iterator gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi)) { stmt = gsi_stmt (gsi); FOR_EACH_SSA_USE_OPERAND (use_p, stmt, iter, SSA_OP_ALL_USES) rename_use_op (use_p); } FOR_EACH_EDGE (e, ei, bb->preds) { if (!flow_bb_inside_loop_p (loop, e->src)) { if (!rename_from_outer_loop) continue; if (e->src != outer_loop->header) { if (outer_loop->inner->next) { /* If outer_loop has 2 inner loops, allow there to be an extra basic block which decides which of the two loops to use using LOOP_VECTORIZED. */ if (!single_pred_p (e->src) || single_pred (e->src) != outer_loop->header) continue; } } } for (gphi_iterator gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi)) rename_use_op (PHI_ARG_DEF_PTR_FROM_EDGE (gsi.phi (), e)); } } struct adjust_info { tree from, to; basic_block bb; }; /* A stack of values to be adjusted in debug stmts. We have to process them LIFO, so that the closest substitution applies. If we processed them FIFO, without the stack, we might substitute uses with a PHI DEF that would soon become non-dominant, and when we got to the suitable one, it wouldn't have anything to substitute any more. */ static vec adjust_vec; /* Adjust any debug stmts that referenced AI->from values to use the loop-closed AI->to, if the references are dominated by AI->bb and not by the definition of AI->from. */ static void adjust_debug_stmts_now (adjust_info *ai) { basic_block bbphi = ai->bb; tree orig_def = ai->from; tree new_def = ai->to; imm_use_iterator imm_iter; gimple *stmt; basic_block bbdef = gimple_bb (SSA_NAME_DEF_STMT (orig_def)); gcc_assert (dom_info_available_p (CDI_DOMINATORS)); /* Adjust any debug stmts that held onto non-loop-closed references. */ FOR_EACH_IMM_USE_STMT (stmt, imm_iter, orig_def) { use_operand_p use_p; basic_block bbuse; if (!is_gimple_debug (stmt)) continue; gcc_assert (gimple_debug_bind_p (stmt)); bbuse = gimple_bb (stmt); if ((bbuse == bbphi || dominated_by_p (CDI_DOMINATORS, bbuse, bbphi)) && !(bbuse == bbdef || dominated_by_p (CDI_DOMINATORS, bbuse, bbdef))) { if (new_def) FOR_EACH_IMM_USE_ON_STMT (use_p, imm_iter) SET_USE (use_p, new_def); else { gimple_debug_bind_reset_value (stmt); update_stmt (stmt); } } } } /* Adjust debug stmts as scheduled before. */ static void adjust_vec_debug_stmts (void) { if (!MAY_HAVE_DEBUG_BIND_STMTS) return; gcc_assert (adjust_vec.exists ()); while (!adjust_vec.is_empty ()) { adjust_debug_stmts_now (&adjust_vec.last ()); adjust_vec.pop (); } } /* Adjust any debug stmts that referenced FROM values to use the loop-closed TO, if the references are dominated by BB and not by the definition of FROM. If adjust_vec is non-NULL, adjustments will be postponed until adjust_vec_debug_stmts is called. */ static void adjust_debug_stmts (tree from, tree to, basic_block bb) { adjust_info ai; if (MAY_HAVE_DEBUG_BIND_STMTS && TREE_CODE (from) == SSA_NAME && ! SSA_NAME_IS_DEFAULT_DEF (from) && ! virtual_operand_p (from)) { ai.from = from; ai.to = to; ai.bb = bb; if (adjust_vec.exists ()) adjust_vec.safe_push (ai); else adjust_debug_stmts_now (&ai); } } /* Change E's phi arg in UPDATE_PHI to NEW_DEF, and record information to adjust any debug stmts that referenced the old phi arg, presumably non-loop-closed references left over from other transformations. */ static void adjust_phi_and_debug_stmts (gimple *update_phi, edge e, tree new_def) { tree orig_def = PHI_ARG_DEF_FROM_EDGE (update_phi, e); gcc_assert (TREE_CODE (orig_def) != SSA_NAME || orig_def != new_def); SET_PHI_ARG_DEF (update_phi, e->dest_idx, new_def); if (MAY_HAVE_DEBUG_BIND_STMTS) adjust_debug_stmts (orig_def, PHI_RESULT (update_phi), gimple_bb (update_phi)); } /* Define one loop rgroup control CTRL from loop LOOP. INIT_CTRL is the value that the control should have during the first iteration and NEXT_CTRL is the value that it should have on subsequent iterations. */ static void vect_set_loop_control (class loop *loop, tree ctrl, tree init_ctrl, tree next_ctrl) { gphi *phi = create_phi_node (ctrl, loop->header); add_phi_arg (phi, init_ctrl, loop_preheader_edge (loop), UNKNOWN_LOCATION); add_phi_arg (phi, next_ctrl, loop_latch_edge (loop), UNKNOWN_LOCATION); } /* Add SEQ to the end of LOOP's preheader block. */ static void add_preheader_seq (class loop *loop, gimple_seq seq) { if (seq) { edge pe = loop_preheader_edge (loop); basic_block new_bb = gsi_insert_seq_on_edge_immediate (pe, seq); gcc_assert (!new_bb); } } /* Add SEQ to the beginning of LOOP's header block. */ static void add_header_seq (class loop *loop, gimple_seq seq) { if (seq) { gimple_stmt_iterator gsi = gsi_after_labels (loop->header); gsi_insert_seq_before (&gsi, seq, GSI_SAME_STMT); } } /* Return true if the target can interleave elements of two vectors. OFFSET is 0 if the first half of the vectors should be interleaved or 1 if the second half should. When returning true, store the associated permutation in INDICES. */ static bool interleave_supported_p (vec_perm_indices *indices, tree vectype, unsigned int offset) { poly_uint64 nelts = TYPE_VECTOR_SUBPARTS (vectype); poly_uint64 base = exact_div (nelts, 2) * offset; vec_perm_builder sel (nelts, 2, 3); for (unsigned int i = 0; i < 3; ++i) { sel.quick_push (base + i); sel.quick_push (base + i + nelts); } indices->new_vector (sel, 2, nelts); return can_vec_perm_const_p (TYPE_MODE (vectype), TYPE_MODE (vectype), *indices); } /* Try to use permutes to define the masks in DEST_RGM using the masks in SRC_RGM, given that the former has twice as many masks as the latter. Return true on success, adding any new statements to SEQ. */ static bool vect_maybe_permute_loop_masks (gimple_seq *seq, rgroup_controls *dest_rgm, rgroup_controls *src_rgm) { tree src_masktype = src_rgm->type; tree dest_masktype = dest_rgm->type; machine_mode src_mode = TYPE_MODE (src_masktype); insn_code icode1, icode2; if (dest_rgm->max_nscalars_per_iter <= src_rgm->max_nscalars_per_iter && (icode1 = optab_handler (vec_unpacku_hi_optab, src_mode)) != CODE_FOR_nothing && (icode2 = optab_handler (vec_unpacku_lo_optab, src_mode)) != CODE_FOR_nothing) { /* Unpacking the source masks gives at least as many mask bits as we need. We can then VIEW_CONVERT any excess bits away. */ machine_mode dest_mode = insn_data[icode1].operand[0].mode; gcc_assert (dest_mode == insn_data[icode2].operand[0].mode); tree unpack_masktype = vect_halve_mask_nunits (src_masktype, dest_mode); for (unsigned int i = 0; i < dest_rgm->controls.length (); ++i) { tree src = src_rgm->controls[i / 2]; tree dest = dest_rgm->controls[i]; tree_code code = ((i & 1) == (BYTES_BIG_ENDIAN ? 0 : 1) ? VEC_UNPACK_HI_EXPR : VEC_UNPACK_LO_EXPR); gassign *stmt; if (dest_masktype == unpack_masktype) stmt = gimple_build_assign (dest, code, src); else { tree temp = make_ssa_name (unpack_masktype); stmt = gimple_build_assign (temp, code, src); gimple_seq_add_stmt (seq, stmt); stmt = gimple_build_assign (dest, VIEW_CONVERT_EXPR, build1 (VIEW_CONVERT_EXPR, dest_masktype, temp)); } gimple_seq_add_stmt (seq, stmt); } return true; } vec_perm_indices indices[2]; if (dest_masktype == src_masktype && interleave_supported_p (&indices[0], src_masktype, 0) && interleave_supported_p (&indices[1], src_masktype, 1)) { /* The destination requires twice as many mask bits as the source, so we can use interleaving permutes to double up the number of bits. */ tree masks[2]; for (unsigned int i = 0; i < 2; ++i) masks[i] = vect_gen_perm_mask_checked (src_masktype, indices[i]); for (unsigned int i = 0; i < dest_rgm->controls.length (); ++i) { tree src = src_rgm->controls[i / 2]; tree dest = dest_rgm->controls[i]; gimple *stmt = gimple_build_assign (dest, VEC_PERM_EXPR, src, src, masks[i & 1]); gimple_seq_add_stmt (seq, stmt); } return true; } return false; } /* Populate DEST_RGM->controls, given that they should add up to STEP. STEP = MIN_EXPR ; First length (MIN (X, VF/N)): loop_len_15 = MIN_EXPR ; Second length: tmp = STEP - loop_len_15; loop_len_16 = MIN (tmp, VF/N); Third length: tmp2 = tmp - loop_len_16; loop_len_17 = MIN (tmp2, VF/N); Last length: loop_len_18 = tmp2 - loop_len_17; */ static void vect_adjust_loop_lens_control (tree iv_type, gimple_seq *seq, rgroup_controls *dest_rgm, tree step) { tree ctrl_type = dest_rgm->type; poly_uint64 nitems_per_ctrl = TYPE_VECTOR_SUBPARTS (ctrl_type) * dest_rgm->factor; tree length_limit = build_int_cst (iv_type, nitems_per_ctrl); for (unsigned int i = 0; i < dest_rgm->controls.length (); ++i) { tree ctrl = dest_rgm->controls[i]; if (i == 0) { /* First iteration: MIN (X, VF/N) capped to the range [0, VF/N]. */ gassign *assign = gimple_build_assign (ctrl, MIN_EXPR, step, length_limit); gimple_seq_add_stmt (seq, assign); } else if (i == dest_rgm->controls.length () - 1) { /* Last iteration: Remain capped to the range [0, VF/N]. */ gassign *assign = gimple_build_assign (ctrl, MINUS_EXPR, step, dest_rgm->controls[i - 1]); gimple_seq_add_stmt (seq, assign); } else { /* (MIN (remain, VF*I/N)) capped to the range [0, VF/N]. */ step = gimple_build (seq, MINUS_EXPR, iv_type, step, dest_rgm->controls[i - 1]); gassign *assign = gimple_build_assign (ctrl, MIN_EXPR, step, length_limit); gimple_seq_add_stmt (seq, assign); } } } /* Helper for vect_set_loop_condition_partial_vectors. Generate definitions for all the rgroup controls in RGC and return a control that is nonzero when the loop needs to iterate. Add any new preheader statements to PREHEADER_SEQ. Use LOOP_COND_GSI to insert code before the exit gcond. RGC belongs to loop LOOP. The loop originally iterated NITERS times and has been vectorized according to LOOP_VINFO. If NITERS_SKIP is nonnull, the first iteration of the vectorized loop starts with NITERS_SKIP dummy iterations of the scalar loop before the real work starts. The mask elements for these dummy iterations must be 0, to ensure that the extra iterations do not have an effect. It is known that: NITERS * RGC->max_nscalars_per_iter * RGC->factor does not overflow. However, MIGHT_WRAP_P says whether an induction variable that starts at 0 and has step: VF * RGC->max_nscalars_per_iter * RGC->factor might overflow before hitting a value above: (NITERS + NITERS_SKIP) * RGC->max_nscalars_per_iter * RGC->factor This means that we cannot guarantee that such an induction variable would ever hit a value that produces a set of all-false masks or zero lengths for RGC. Note: the cost of the code generated by this function is modeled by vect_estimate_min_profitable_iters, so changes here may need corresponding changes there. */ static tree vect_set_loop_controls_directly (class loop *loop, loop_vec_info loop_vinfo, gimple_seq *preheader_seq, gimple_seq *header_seq, gimple_stmt_iterator loop_cond_gsi, rgroup_controls *rgc, tree niters, tree niters_skip, bool might_wrap_p, tree *iv_step, tree *compare_step) { tree compare_type = LOOP_VINFO_RGROUP_COMPARE_TYPE (loop_vinfo); tree iv_type = LOOP_VINFO_RGROUP_IV_TYPE (loop_vinfo); bool use_masks_p = LOOP_VINFO_FULLY_MASKED_P (loop_vinfo); tree ctrl_type = rgc->type; unsigned int nitems_per_iter = rgc->max_nscalars_per_iter * rgc->factor; poly_uint64 nitems_per_ctrl = TYPE_VECTOR_SUBPARTS (ctrl_type) * rgc->factor; poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); tree length_limit = NULL_TREE; /* For length, we need length_limit to ensure length in range. */ if (!use_masks_p) length_limit = build_int_cst (compare_type, nitems_per_ctrl); /* Calculate the maximum number of item values that the rgroup handles in total, the number that it handles for each iteration of the vector loop, and the number that it should skip during the first iteration of the vector loop. */ tree nitems_total = niters; tree nitems_step = build_int_cst (iv_type, vf); tree nitems_skip = niters_skip; if (nitems_per_iter != 1) { /* We checked before setting LOOP_VINFO_USING_PARTIAL_VECTORS_P that these multiplications don't overflow. */ tree compare_factor = build_int_cst (compare_type, nitems_per_iter); tree iv_factor = build_int_cst (iv_type, nitems_per_iter); nitems_total = gimple_build (preheader_seq, MULT_EXPR, compare_type, nitems_total, compare_factor); nitems_step = gimple_build (preheader_seq, MULT_EXPR, iv_type, nitems_step, iv_factor); if (nitems_skip) nitems_skip = gimple_build (preheader_seq, MULT_EXPR, compare_type, nitems_skip, compare_factor); } /* Create an induction variable that counts the number of items processed. */ tree index_before_incr, index_after_incr; gimple_stmt_iterator incr_gsi; bool insert_after; standard_iv_increment_position (loop, &incr_gsi, &insert_after); if (LOOP_VINFO_USING_DECREMENTING_IV_P (loop_vinfo)) { /* Create an IV that counts down from niters_total and whose step is the (variable) amount processed in the current iteration: ... _10 = (unsigned long) count_12(D); ... # ivtmp_9 = PHI _36 = (MIN_EXPR | SELECT_VL) ; ... vect__4.8_28 = .LEN_LOAD (_17, 32B, _36, 0); ... ivtmp_35 = ivtmp_9 - POLY_INT_CST [4, 4]; ... if (ivtmp_9 > POLY_INT_CST [4, 4]) goto ; [83.33%] else goto ; [16.67%] */ nitems_total = gimple_convert (preheader_seq, iv_type, nitems_total); tree step = rgc->controls.length () == 1 ? rgc->controls[0] : make_ssa_name (iv_type); /* Create decrement IV. */ if (LOOP_VINFO_USING_SELECT_VL_P (loop_vinfo)) { create_iv (nitems_total, MINUS_EXPR, step, NULL_TREE, loop, &incr_gsi, insert_after, &index_before_incr, &index_after_incr); tree len = gimple_build (header_seq, IFN_SELECT_VL, iv_type, index_before_incr, nitems_step); gimple_seq_add_stmt (header_seq, gimple_build_assign (step, len)); } else { create_iv (nitems_total, MINUS_EXPR, nitems_step, NULL_TREE, loop, &incr_gsi, insert_after, &index_before_incr, &index_after_incr); gimple_seq_add_stmt (header_seq, gimple_build_assign (step, MIN_EXPR, index_before_incr, nitems_step)); } *iv_step = step; *compare_step = nitems_step; return LOOP_VINFO_USING_SELECT_VL_P (loop_vinfo) ? index_after_incr : index_before_incr; } /* Create increment IV. */ create_iv (build_int_cst (iv_type, 0), PLUS_EXPR, nitems_step, NULL_TREE, loop, &incr_gsi, insert_after, &index_before_incr, &index_after_incr); tree zero_index = build_int_cst (compare_type, 0); tree test_index, test_limit, first_limit; gimple_stmt_iterator *test_gsi; if (might_wrap_p) { /* In principle the loop should stop iterating once the incremented IV reaches a value greater than or equal to: NITEMS_TOTAL +[infinite-prec] NITEMS_SKIP However, there's no guarantee that this addition doesn't overflow the comparison type, or that the IV hits a value above it before wrapping around. We therefore adjust the limit down by one IV step: (NITEMS_TOTAL +[infinite-prec] NITEMS_SKIP) -[infinite-prec] NITEMS_STEP and compare the IV against this limit _before_ incrementing it. Since the comparison type is unsigned, we actually want the subtraction to saturate at zero: (NITEMS_TOTAL +[infinite-prec] NITEMS_SKIP) -[sat] NITEMS_STEP And since NITEMS_SKIP < NITEMS_STEP, we can reassociate this as: NITEMS_TOTAL -[sat] (NITEMS_STEP - NITEMS_SKIP) where the rightmost subtraction can be done directly in COMPARE_TYPE. */ test_index = index_before_incr; tree adjust = gimple_convert (preheader_seq, compare_type, nitems_step); if (nitems_skip) adjust = gimple_build (preheader_seq, MINUS_EXPR, compare_type, adjust, nitems_skip); test_limit = gimple_build (preheader_seq, MAX_EXPR, compare_type, nitems_total, adjust); test_limit = gimple_build (preheader_seq, MINUS_EXPR, compare_type, test_limit, adjust); test_gsi = &incr_gsi; /* Get a safe limit for the first iteration. */ if (nitems_skip) { /* The first vector iteration can handle at most NITEMS_STEP items. NITEMS_STEP <= CONST_LIMIT, and adding NITEMS_SKIP to that cannot overflow. */ tree const_limit = build_int_cst (compare_type, LOOP_VINFO_VECT_FACTOR (loop_vinfo) * nitems_per_iter); first_limit = gimple_build (preheader_seq, MIN_EXPR, compare_type, nitems_total, const_limit); first_limit = gimple_build (preheader_seq, PLUS_EXPR, compare_type, first_limit, nitems_skip); } else /* For the first iteration it doesn't matter whether the IV hits a value above NITEMS_TOTAL. That only matters for the latch condition. */ first_limit = nitems_total; } else { /* Test the incremented IV, which will always hit a value above the bound before wrapping. */ test_index = index_after_incr; test_limit = nitems_total; if (nitems_skip) test_limit = gimple_build (preheader_seq, PLUS_EXPR, compare_type, test_limit, nitems_skip); test_gsi = &loop_cond_gsi; first_limit = test_limit; } /* Convert the IV value to the comparison type (either a no-op or a demotion). */ gimple_seq test_seq = NULL; test_index = gimple_convert (&test_seq, compare_type, test_index); gsi_insert_seq_before (test_gsi, test_seq, GSI_SAME_STMT); /* Provide a definition of each control in the group. */ tree next_ctrl = NULL_TREE; tree ctrl; unsigned int i; FOR_EACH_VEC_ELT_REVERSE (rgc->controls, i, ctrl) { /* Previous controls will cover BIAS items. This control covers the next batch. */ poly_uint64 bias = nitems_per_ctrl * i; tree bias_tree = build_int_cst (compare_type, bias); /* See whether the first iteration of the vector loop is known to have a full control. */ poly_uint64 const_limit; bool first_iteration_full = (poly_int_tree_p (first_limit, &const_limit) && known_ge (const_limit, (i + 1) * nitems_per_ctrl)); /* Rather than have a new IV that starts at BIAS and goes up to TEST_LIMIT, prefer to use the same 0-based IV for each control and adjust the bound down by BIAS. */ tree this_test_limit = test_limit; if (i != 0) { this_test_limit = gimple_build (preheader_seq, MAX_EXPR, compare_type, this_test_limit, bias_tree); this_test_limit = gimple_build (preheader_seq, MINUS_EXPR, compare_type, this_test_limit, bias_tree); } /* Create the initial control. First include all items that are within the loop limit. */ tree init_ctrl = NULL_TREE; if (!first_iteration_full) { tree start, end; if (first_limit == test_limit) { /* Use a natural test between zero (the initial IV value) and the loop limit. The "else" block would be valid too, but this choice can avoid the need to load BIAS_TREE into a register. */ start = zero_index; end = this_test_limit; } else { /* FIRST_LIMIT is the maximum number of items handled by the first iteration of the vector loop. Test the portion associated with this control. */ start = bias_tree; end = first_limit; } if (use_masks_p) init_ctrl = vect_gen_while (preheader_seq, ctrl_type, start, end, "max_mask"); else { init_ctrl = make_temp_ssa_name (compare_type, NULL, "max_len"); gimple_seq seq = vect_gen_len (init_ctrl, start, end, length_limit); gimple_seq_add_seq (preheader_seq, seq); } } /* Now AND out the bits that are within the number of skipped items. */ poly_uint64 const_skip; if (nitems_skip && !(poly_int_tree_p (nitems_skip, &const_skip) && known_le (const_skip, bias))) { gcc_assert (use_masks_p); tree unskipped_mask = vect_gen_while_not (preheader_seq, ctrl_type, bias_tree, nitems_skip); if (init_ctrl) init_ctrl = gimple_build (preheader_seq, BIT_AND_EXPR, ctrl_type, init_ctrl, unskipped_mask); else init_ctrl = unskipped_mask; } if (!init_ctrl) { /* First iteration is full. */ if (use_masks_p) init_ctrl = build_minus_one_cst (ctrl_type); else init_ctrl = length_limit; } /* Get the control value for the next iteration of the loop. */ if (use_masks_p) { gimple_seq stmts = NULL; next_ctrl = vect_gen_while (&stmts, ctrl_type, test_index, this_test_limit, "next_mask"); gsi_insert_seq_before (test_gsi, stmts, GSI_SAME_STMT); } else { next_ctrl = make_temp_ssa_name (compare_type, NULL, "next_len"); gimple_seq seq = vect_gen_len (next_ctrl, test_index, this_test_limit, length_limit); gsi_insert_seq_before (test_gsi, seq, GSI_SAME_STMT); } vect_set_loop_control (loop, ctrl, init_ctrl, next_ctrl); } int partial_load_bias = LOOP_VINFO_PARTIAL_LOAD_STORE_BIAS (loop_vinfo); if (partial_load_bias != 0) { tree adjusted_len = rgc->bias_adjusted_ctrl; gassign *minus = gimple_build_assign (adjusted_len, PLUS_EXPR, rgc->controls[0], build_int_cst (TREE_TYPE (rgc->controls[0]), partial_load_bias)); gimple_seq_add_stmt (header_seq, minus); } return next_ctrl; } /* Set up the iteration condition and rgroup controls for LOOP, given that LOOP_VINFO_USING_PARTIAL_VECTORS_P is true for the vectorized loop. LOOP_VINFO describes the vectorization of LOOP. NITERS is the number of iterations of the original scalar loop that should be handled by the vector loop. NITERS_MAYBE_ZERO and FINAL_IV are as for vect_set_loop_condition. Insert the branch-back condition before LOOP_COND_GSI and return the final gcond. */ static gcond * vect_set_loop_condition_partial_vectors (class loop *loop, edge exit_edge, loop_vec_info loop_vinfo, tree niters, tree final_iv, bool niters_maybe_zero, gimple_stmt_iterator loop_cond_gsi) { gimple_seq preheader_seq = NULL; gimple_seq header_seq = NULL; bool use_masks_p = LOOP_VINFO_FULLY_MASKED_P (loop_vinfo); tree compare_type = LOOP_VINFO_RGROUP_COMPARE_TYPE (loop_vinfo); unsigned int compare_precision = TYPE_PRECISION (compare_type); tree orig_niters = niters; /* Type of the initial value of NITERS. */ tree ni_actual_type = TREE_TYPE (niters); unsigned int ni_actual_precision = TYPE_PRECISION (ni_actual_type); tree niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); if (niters_skip) niters_skip = gimple_convert (&preheader_seq, compare_type, niters_skip); /* Convert NITERS to the same size as the compare. */ if (compare_precision > ni_actual_precision && niters_maybe_zero) { /* We know that there is always at least one iteration, so if the count is zero then it must have wrapped. Cope with this by subtracting 1 before the conversion and adding 1 to the result. */ gcc_assert (TYPE_UNSIGNED (ni_actual_type)); niters = gimple_build (&preheader_seq, PLUS_EXPR, ni_actual_type, niters, build_minus_one_cst (ni_actual_type)); niters = gimple_convert (&preheader_seq, compare_type, niters); niters = gimple_build (&preheader_seq, PLUS_EXPR, compare_type, niters, build_one_cst (compare_type)); } else niters = gimple_convert (&preheader_seq, compare_type, niters); /* Iterate over all the rgroups and fill in their controls. We could use the first control from any rgroup for the loop condition; here we arbitrarily pick the last. */ tree test_ctrl = NULL_TREE; tree iv_step = NULL_TREE; tree compare_step = NULL_TREE; rgroup_controls *rgc; rgroup_controls *iv_rgc = nullptr; unsigned int i; auto_vec *controls = use_masks_p ? &LOOP_VINFO_MASKS (loop_vinfo).rgc_vec : &LOOP_VINFO_LENS (loop_vinfo); FOR_EACH_VEC_ELT (*controls, i, rgc) if (!rgc->controls.is_empty ()) { /* First try using permutes. This adds a single vector instruction to the loop for each mask, but needs no extra loop invariants or IVs. */ unsigned int nmasks = i + 1; if (use_masks_p && (nmasks & 1) == 0) { rgroup_controls *half_rgc = &(*controls)[nmasks / 2 - 1]; if (!half_rgc->controls.is_empty () && vect_maybe_permute_loop_masks (&header_seq, rgc, half_rgc)) continue; } if (!LOOP_VINFO_USING_DECREMENTING_IV_P (loop_vinfo) || !iv_rgc || (iv_rgc->max_nscalars_per_iter * iv_rgc->factor != rgc->max_nscalars_per_iter * rgc->factor)) { /* See whether zero-based IV would ever generate all-false masks or zero length before wrapping around. */ bool might_wrap_p = vect_rgroup_iv_might_wrap_p (loop_vinfo, rgc); /* Set up all controls for this group. */ test_ctrl = vect_set_loop_controls_directly (loop, loop_vinfo, &preheader_seq, &header_seq, loop_cond_gsi, rgc, niters, niters_skip, might_wrap_p, &iv_step, &compare_step); iv_rgc = rgc; } if (LOOP_VINFO_USING_DECREMENTING_IV_P (loop_vinfo) && rgc->controls.length () > 1) { /* vect_set_loop_controls_directly creates an IV whose step is equal to the expected sum of RGC->controls. Use that information to populate RGC->controls. */ tree iv_type = LOOP_VINFO_RGROUP_IV_TYPE (loop_vinfo); gcc_assert (iv_step); vect_adjust_loop_lens_control (iv_type, &header_seq, rgc, iv_step); } } /* Emit all accumulated statements. */ add_preheader_seq (loop, preheader_seq); add_header_seq (loop, header_seq); /* Get a boolean result that tells us whether to iterate. */ gcond *cond_stmt; if (LOOP_VINFO_USING_DECREMENTING_IV_P (loop_vinfo) && !LOOP_VINFO_USING_SELECT_VL_P (loop_vinfo)) { gcc_assert (compare_step); tree_code code = (exit_edge->flags & EDGE_TRUE_VALUE) ? LE_EXPR : GT_EXPR; cond_stmt = gimple_build_cond (code, test_ctrl, compare_step, NULL_TREE, NULL_TREE); } else { tree_code code = (exit_edge->flags & EDGE_TRUE_VALUE) ? EQ_EXPR : NE_EXPR; tree zero_ctrl = build_zero_cst (TREE_TYPE (test_ctrl)); cond_stmt = gimple_build_cond (code, test_ctrl, zero_ctrl, NULL_TREE, NULL_TREE); } gsi_insert_before (&loop_cond_gsi, cond_stmt, GSI_SAME_STMT); /* The loop iterates (NITERS - 1) / VF + 1 times. Subtract one from this to get the latch count. */ tree step = build_int_cst (compare_type, LOOP_VINFO_VECT_FACTOR (loop_vinfo)); tree niters_minus_one = fold_build2 (PLUS_EXPR, compare_type, niters, build_minus_one_cst (compare_type)); loop->nb_iterations = fold_build2 (TRUNC_DIV_EXPR, compare_type, niters_minus_one, step); if (final_iv) { gassign *assign = gimple_build_assign (final_iv, orig_niters); gsi_insert_on_edge_immediate (exit_edge, assign); } return cond_stmt; } /* Set up the iteration condition and rgroup controls for LOOP in AVX512 style, given that LOOP_VINFO_USING_PARTIAL_VECTORS_P is true for the vectorized loop. LOOP_VINFO describes the vectorization of LOOP. NITERS is the number of iterations of the original scalar loop that should be handled by the vector loop. NITERS_MAYBE_ZERO and FINAL_IV are as for vect_set_loop_condition. Insert the branch-back condition before LOOP_COND_GSI and return the final gcond. */ static gcond * vect_set_loop_condition_partial_vectors_avx512 (class loop *loop, edge exit_edge, loop_vec_info loop_vinfo, tree niters, tree final_iv, bool niters_maybe_zero, gimple_stmt_iterator loop_cond_gsi) { tree niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); tree iv_type = LOOP_VINFO_RGROUP_IV_TYPE (loop_vinfo); poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); tree orig_niters = niters; gimple_seq preheader_seq = NULL; /* Create an IV that counts down from niters and whose step is the number of iterations processed in the current iteration. Produce the controls with compares like the following. # iv_2 = PHI rem_4 = MIN ; remv_6 = { rem_4, rem_4, rem_4, ... } mask_5 = { 0, 0, 1, 1, 2, 2, ... } < remv6; iv_3 = iv_2 - VF; if (iv_2 > VF) continue; Where the constant is built with elements at most VF - 1 and repetitions according to max_nscalars_per_iter which is guarnateed to be the same within a group. */ /* Convert NITERS to the determined IV type. */ if (TYPE_PRECISION (iv_type) > TYPE_PRECISION (TREE_TYPE (niters)) && niters_maybe_zero) { /* We know that there is always at least one iteration, so if the count is zero then it must have wrapped. Cope with this by subtracting 1 before the conversion and adding 1 to the result. */ gcc_assert (TYPE_UNSIGNED (TREE_TYPE (niters))); niters = gimple_build (&preheader_seq, PLUS_EXPR, TREE_TYPE (niters), niters, build_minus_one_cst (TREE_TYPE (niters))); niters = gimple_convert (&preheader_seq, iv_type, niters); niters = gimple_build (&preheader_seq, PLUS_EXPR, iv_type, niters, build_one_cst (iv_type)); } else niters = gimple_convert (&preheader_seq, iv_type, niters); /* Bias the initial value of the IV in case we need to skip iterations at the beginning. */ tree niters_adj = niters; if (niters_skip) { tree skip = gimple_convert (&preheader_seq, iv_type, niters_skip); niters_adj = gimple_build (&preheader_seq, PLUS_EXPR, iv_type, niters, skip); } /* The iteration step is the vectorization factor. */ tree iv_step = build_int_cst (iv_type, vf); /* Create the decrement IV. */ tree index_before_incr, index_after_incr; gimple_stmt_iterator incr_gsi; bool insert_after; standard_iv_increment_position (loop, &incr_gsi, &insert_after); create_iv (niters_adj, MINUS_EXPR, iv_step, NULL_TREE, loop, &incr_gsi, insert_after, &index_before_incr, &index_after_incr); /* Iterate over all the rgroups and fill in their controls. */ for (auto &rgc : LOOP_VINFO_MASKS (loop_vinfo).rgc_vec) { if (rgc.controls.is_empty ()) continue; tree ctrl_type = rgc.type; poly_uint64 nitems_per_ctrl = TYPE_VECTOR_SUBPARTS (ctrl_type); tree vectype = rgc.compare_type; /* index_after_incr is the IV specifying the remaining iterations in the next iteration. */ tree rem = index_after_incr; /* When the data type for the compare to produce the mask is smaller than the IV type we need to saturate. Saturate to the smallest possible value (IV_TYPE) so we only have to saturate once (CSE will catch redundant ones we add). */ if (TYPE_PRECISION (TREE_TYPE (vectype)) < TYPE_PRECISION (iv_type)) rem = gimple_build (&incr_gsi, false, GSI_CONTINUE_LINKING, UNKNOWN_LOCATION, MIN_EXPR, TREE_TYPE (rem), rem, iv_step); rem = gimple_convert (&incr_gsi, false, GSI_CONTINUE_LINKING, UNKNOWN_LOCATION, TREE_TYPE (vectype), rem); /* Build a data vector composed of the remaining iterations. */ rem = gimple_build_vector_from_val (&incr_gsi, false, GSI_CONTINUE_LINKING, UNKNOWN_LOCATION, vectype, rem); /* Provide a definition of each vector in the control group. */ tree next_ctrl = NULL_TREE; tree first_rem = NULL_TREE; tree ctrl; unsigned int i; FOR_EACH_VEC_ELT_REVERSE (rgc.controls, i, ctrl) { /* Previous controls will cover BIAS items. This control covers the next batch. */ poly_uint64 bias = nitems_per_ctrl * i; /* Build the constant to compare the remaining iters against, this is sth like { 0, 0, 1, 1, 2, 2, 3, 3, ... } appropriately split into pieces. */ unsigned n = TYPE_VECTOR_SUBPARTS (ctrl_type).to_constant (); tree_vector_builder builder (vectype, n, 1); for (unsigned i = 0; i < n; ++i) { unsigned HOST_WIDE_INT val = (i + bias.to_constant ()) / rgc.max_nscalars_per_iter; gcc_assert (val < vf.to_constant ()); builder.quick_push (build_int_cst (TREE_TYPE (vectype), val)); } tree cmp_series = builder.build (); /* Create the initial control. First include all items that are within the loop limit. */ tree init_ctrl = NULL_TREE; poly_uint64 const_limit; /* See whether the first iteration of the vector loop is known to have a full control. */ if (poly_int_tree_p (niters, &const_limit) && known_ge (const_limit, (i + 1) * nitems_per_ctrl)) init_ctrl = build_minus_one_cst (ctrl_type); else { /* The remaining work items initially are niters. Saturate, splat and compare. */ if (!first_rem) { first_rem = niters; if (TYPE_PRECISION (TREE_TYPE (vectype)) < TYPE_PRECISION (iv_type)) first_rem = gimple_build (&preheader_seq, MIN_EXPR, TREE_TYPE (first_rem), first_rem, iv_step); first_rem = gimple_convert (&preheader_seq, TREE_TYPE (vectype), first_rem); first_rem = gimple_build_vector_from_val (&preheader_seq, vectype, first_rem); } init_ctrl = gimple_build (&preheader_seq, LT_EXPR, ctrl_type, cmp_series, first_rem); } /* Now AND out the bits that are within the number of skipped items. */ poly_uint64 const_skip; if (niters_skip && !(poly_int_tree_p (niters_skip, &const_skip) && known_le (const_skip, bias))) { /* For integer mode masks it's cheaper to shift out the bits since that avoids loading a constant. */ gcc_assert (GET_MODE_CLASS (TYPE_MODE (ctrl_type)) == MODE_INT); init_ctrl = gimple_build (&preheader_seq, VIEW_CONVERT_EXPR, lang_hooks.types.type_for_mode (TYPE_MODE (ctrl_type), 1), init_ctrl); /* ??? But when the shift amount isn't constant this requires a round-trip to GRPs. We could apply the bias to either side of the compare instead. */ tree shift = gimple_build (&preheader_seq, MULT_EXPR, TREE_TYPE (niters_skip), niters_skip, build_int_cst (TREE_TYPE (niters_skip), rgc.max_nscalars_per_iter)); init_ctrl = gimple_build (&preheader_seq, LSHIFT_EXPR, TREE_TYPE (init_ctrl), init_ctrl, shift); init_ctrl = gimple_build (&preheader_seq, VIEW_CONVERT_EXPR, ctrl_type, init_ctrl); } /* Get the control value for the next iteration of the loop. */ next_ctrl = gimple_build (&incr_gsi, false, GSI_CONTINUE_LINKING, UNKNOWN_LOCATION, LT_EXPR, ctrl_type, cmp_series, rem); vect_set_loop_control (loop, ctrl, init_ctrl, next_ctrl); } } /* Emit all accumulated statements. */ add_preheader_seq (loop, preheader_seq); /* Adjust the exit test using the decrementing IV. */ tree_code code = (exit_edge->flags & EDGE_TRUE_VALUE) ? LE_EXPR : GT_EXPR; /* When we peel for alignment with niter_skip != 0 this can cause niter + niter_skip to wrap and since we are comparing the value before the decrement here we get a false early exit. We can't compare the value after decrement either because that decrement could wrap as well as we're not doing a saturating decrement. To avoid this situation we force a larger iv_type. */ gcond *cond_stmt = gimple_build_cond (code, index_before_incr, iv_step, NULL_TREE, NULL_TREE); gsi_insert_before (&loop_cond_gsi, cond_stmt, GSI_SAME_STMT); /* The loop iterates (NITERS - 1 + NITERS_SKIP) / VF + 1 times. Subtract one from this to get the latch count. */ tree niters_minus_one = fold_build2 (PLUS_EXPR, TREE_TYPE (orig_niters), orig_niters, build_minus_one_cst (TREE_TYPE (orig_niters))); tree niters_adj2 = fold_convert (iv_type, niters_minus_one); if (niters_skip) niters_adj2 = fold_build2 (PLUS_EXPR, iv_type, niters_minus_one, fold_convert (iv_type, niters_skip)); loop->nb_iterations = fold_build2 (TRUNC_DIV_EXPR, iv_type, niters_adj2, iv_step); if (final_iv) { gassign *assign = gimple_build_assign (final_iv, orig_niters); gsi_insert_on_edge_immediate (single_exit (loop), assign); } return cond_stmt; } /* Like vect_set_loop_condition, but handle the case in which the vector loop handles exactly VF scalars per iteration. */ static gcond * vect_set_loop_condition_normal (loop_vec_info /* loop_vinfo */, edge exit_edge, class loop *loop, tree niters, tree step, tree final_iv, bool niters_maybe_zero, gimple_stmt_iterator loop_cond_gsi) { tree indx_before_incr, indx_after_incr; gcond *cond_stmt; gcond *orig_cond; edge pe = loop_preheader_edge (loop); gimple_stmt_iterator incr_gsi; bool insert_after; enum tree_code code; tree niters_type = TREE_TYPE (niters); orig_cond = get_loop_exit_condition (exit_edge); gcc_assert (orig_cond); loop_cond_gsi = gsi_for_stmt (orig_cond); tree init, limit; if (!niters_maybe_zero && integer_onep (step)) { /* In this case we can use a simple 0-based IV: A: x = 0; do { ... x += 1; } while (x < NITERS); */ code = (exit_edge->flags & EDGE_TRUE_VALUE) ? GE_EXPR : LT_EXPR; init = build_zero_cst (niters_type); limit = niters; } else { /* The following works for all values of NITERS except 0: B: x = 0; do { ... x += STEP; } while (x <= NITERS - STEP); so that the loop continues to iterate if x + STEP - 1 < NITERS but stops if x + STEP - 1 >= NITERS. However, if NITERS is zero, x never hits a value above NITERS - STEP before wrapping around. There are two obvious ways of dealing with this: - start at STEP - 1 and compare x before incrementing it - start at -1 and compare x after incrementing it The latter is simpler and is what we use. The loop in this case looks like: C: x = -1; do { ... x += STEP; } while (x < NITERS - STEP); In both cases the loop limit is NITERS - STEP. */ gimple_seq seq = NULL; limit = force_gimple_operand (niters, &seq, true, NULL_TREE); limit = gimple_build (&seq, MINUS_EXPR, TREE_TYPE (limit), limit, step); if (seq) { basic_block new_bb = gsi_insert_seq_on_edge_immediate (pe, seq); gcc_assert (!new_bb); } if (niters_maybe_zero) { /* Case C. */ code = (exit_edge->flags & EDGE_TRUE_VALUE) ? GE_EXPR : LT_EXPR; init = build_all_ones_cst (niters_type); } else { /* Case B. */ code = (exit_edge->flags & EDGE_TRUE_VALUE) ? GT_EXPR : LE_EXPR; init = build_zero_cst (niters_type); } } standard_iv_increment_position (loop, &incr_gsi, &insert_after); create_iv (init, PLUS_EXPR, step, NULL_TREE, loop, &incr_gsi, insert_after, &indx_before_incr, &indx_after_incr); indx_after_incr = force_gimple_operand_gsi (&loop_cond_gsi, indx_after_incr, true, NULL_TREE, true, GSI_SAME_STMT); limit = force_gimple_operand_gsi (&loop_cond_gsi, limit, true, NULL_TREE, true, GSI_SAME_STMT); cond_stmt = gimple_build_cond (code, indx_after_incr, limit, NULL_TREE, NULL_TREE); gsi_insert_before (&loop_cond_gsi, cond_stmt, GSI_SAME_STMT); /* Record the number of latch iterations. */ if (limit == niters) /* Case A: the loop iterates NITERS times. Subtract one to get the latch count. */ loop->nb_iterations = fold_build2 (MINUS_EXPR, niters_type, niters, build_int_cst (niters_type, 1)); else /* Case B or C: the loop iterates (NITERS - STEP) / STEP + 1 times. Subtract one from this to get the latch count. */ loop->nb_iterations = fold_build2 (TRUNC_DIV_EXPR, niters_type, limit, step); if (final_iv) { gassign *assign; gcc_assert (single_pred_p (exit_edge->dest)); tree phi_dest = integer_zerop (init) ? final_iv : copy_ssa_name (indx_after_incr); /* Make sure to maintain LC SSA form here and elide the subtraction if the value is zero. */ gphi *phi = create_phi_node (phi_dest, exit_edge->dest); add_phi_arg (phi, indx_after_incr, exit_edge, UNKNOWN_LOCATION); if (!integer_zerop (init)) { assign = gimple_build_assign (final_iv, MINUS_EXPR, phi_dest, init); gimple_stmt_iterator gsi = gsi_after_labels (exit_edge->dest); gsi_insert_before (&gsi, assign, GSI_SAME_STMT); } } return cond_stmt; } /* If we're using fully-masked loops, make LOOP iterate: N == (NITERS - 1) / STEP + 1 times. When NITERS is zero, this is equivalent to making the loop execute (1 << M) / STEP times, where M is the precision of NITERS. NITERS_MAYBE_ZERO is true if this last case might occur. If we're not using fully-masked loops, make LOOP iterate: N == (NITERS - STEP) / STEP + 1 times, where NITERS is known to be outside the range [1, STEP - 1]. This is equivalent to making the loop execute NITERS / STEP times when NITERS is nonzero and (1 << M) / STEP times otherwise. NITERS_MAYBE_ZERO again indicates whether this last case might occur. If FINAL_IV is nonnull, it is an SSA name that should be set to N * STEP on exit from the loop. Assumption: the exit-condition of LOOP is the last stmt in the loop. */ void vect_set_loop_condition (class loop *loop, edge loop_e, loop_vec_info loop_vinfo, tree niters, tree step, tree final_iv, bool niters_maybe_zero) { gcond *cond_stmt; gcond *orig_cond = get_loop_exit_condition (loop_e); gimple_stmt_iterator loop_cond_gsi = gsi_for_stmt (orig_cond); if (loop_vinfo && LOOP_VINFO_USING_PARTIAL_VECTORS_P (loop_vinfo)) { if (LOOP_VINFO_PARTIAL_VECTORS_STYLE (loop_vinfo) == vect_partial_vectors_avx512) cond_stmt = vect_set_loop_condition_partial_vectors_avx512 (loop, loop_e, loop_vinfo, niters, final_iv, niters_maybe_zero, loop_cond_gsi); else cond_stmt = vect_set_loop_condition_partial_vectors (loop, loop_e, loop_vinfo, niters, final_iv, niters_maybe_zero, loop_cond_gsi); } else cond_stmt = vect_set_loop_condition_normal (loop_vinfo, loop_e, loop, niters, step, final_iv, niters_maybe_zero, loop_cond_gsi); /* Remove old loop exit test. */ stmt_vec_info orig_cond_info; if (loop_vinfo && (orig_cond_info = loop_vinfo->lookup_stmt (orig_cond))) loop_vinfo->remove_stmt (orig_cond_info); else gsi_remove (&loop_cond_gsi, true); if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "New loop exit condition: %G", (gimple *) cond_stmt); } /* Given LOOP this function generates a new copy of it and puts it on E which is either the entry or exit of LOOP. If SCALAR_LOOP is non-NULL, assume LOOP and SCALAR_LOOP are equivalent and copy the basic blocks from SCALAR_LOOP instead of LOOP, but to either the entry or exit of LOOP. If FLOW_LOOPS then connect LOOP to SCALAR_LOOP as a continuation. This is correct for cases where one loop continues from the other like in the vectorizer, but not true for uses in e.g. loop distribution where the contents of the loop body are split but the iteration space of both copies remains the same. If UPDATED_DOMS is not NULL it is update with the list of basic blocks whoms dominators were updated during the peeling. */ class loop * slpeel_tree_duplicate_loop_to_edge_cfg (class loop *loop, edge loop_exit, class loop *scalar_loop, edge scalar_exit, edge e, edge *new_e, bool flow_loops) { class loop *new_loop; basic_block *new_bbs, *bbs, *pbbs; bool at_exit; bool was_imm_dom; basic_block exit_dest; edge exit, new_exit; bool duplicate_outer_loop = false; exit = loop_exit; at_exit = (e == exit); if (!at_exit && e != loop_preheader_edge (loop)) return NULL; if (scalar_loop == NULL) { scalar_loop = loop; scalar_exit = loop_exit; } else if (scalar_loop == loop) scalar_exit = loop_exit; else { /* Loop has been version, match exits up using the aux index. */ for (edge exit : get_loop_exit_edges (scalar_loop)) if (exit->aux == loop_exit->aux) { scalar_exit = exit; break; } gcc_assert (scalar_exit); } bbs = XNEWVEC (basic_block, scalar_loop->num_nodes + 1); pbbs = bbs + 1; get_loop_body_with_size (scalar_loop, pbbs, scalar_loop->num_nodes); /* Allow duplication of outer loops. */ if (scalar_loop->inner) duplicate_outer_loop = true; /* Generate new loop structure. */ new_loop = duplicate_loop (scalar_loop, loop_outer (scalar_loop)); duplicate_subloops (scalar_loop, new_loop); exit_dest = exit->dest; was_imm_dom = (get_immediate_dominator (CDI_DOMINATORS, exit_dest) == loop->header ? true : false); /* Also copy the pre-header, this avoids jumping through hoops to duplicate the loop entry PHI arguments. Create an empty pre-header unconditionally for this. */ basic_block preheader = split_edge (loop_preheader_edge (scalar_loop)); edge entry_e = single_pred_edge (preheader); bbs[0] = preheader; new_bbs = XNEWVEC (basic_block, scalar_loop->num_nodes + 1); copy_bbs (bbs, scalar_loop->num_nodes + 1, new_bbs, &scalar_exit, 1, &new_exit, NULL, at_exit ? loop->latch : e->src, true); exit = loop_exit; basic_block new_preheader = new_bbs[0]; gcc_assert (new_exit); /* Record the new loop exit information. new_loop doesn't have SCEV data and so we must initialize the exit information. */ if (new_e) *new_e = new_exit; /* Before installing PHI arguments make sure that the edges into them match that of the scalar loop we analyzed. This makes sure the SLP tree matches up between the main vectorized loop and the epilogue vectorized copies. */ if (single_succ_edge (preheader)->dest_idx != single_succ_edge (new_bbs[0])->dest_idx) { basic_block swap_bb = new_bbs[1]; gcc_assert (EDGE_COUNT (swap_bb->preds) == 2); std::swap (EDGE_PRED (swap_bb, 0), EDGE_PRED (swap_bb, 1)); EDGE_PRED (swap_bb, 0)->dest_idx = 0; EDGE_PRED (swap_bb, 1)->dest_idx = 1; } if (duplicate_outer_loop) { class loop *new_inner_loop = get_loop_copy (scalar_loop->inner); if (loop_preheader_edge (scalar_loop)->dest_idx != loop_preheader_edge (new_inner_loop)->dest_idx) { basic_block swap_bb = new_inner_loop->header; gcc_assert (EDGE_COUNT (swap_bb->preds) == 2); std::swap (EDGE_PRED (swap_bb, 0), EDGE_PRED (swap_bb, 1)); EDGE_PRED (swap_bb, 0)->dest_idx = 0; EDGE_PRED (swap_bb, 1)->dest_idx = 1; } } add_phi_args_after_copy (new_bbs, scalar_loop->num_nodes + 1, NULL); /* Skip new preheader since it's deleted if copy loop is added at entry. */ for (unsigned i = (at_exit ? 0 : 1); i < scalar_loop->num_nodes + 1; i++) rename_variables_in_bb (new_bbs[i], duplicate_outer_loop); /* Rename the exit uses. */ for (edge exit : get_loop_exit_edges (new_loop)) for (auto gsi = gsi_start_phis (exit->dest); !gsi_end_p (gsi); gsi_next (&gsi)) { tree orig_def = PHI_ARG_DEF_FROM_EDGE (gsi.phi (), exit); rename_use_op (PHI_ARG_DEF_PTR_FROM_EDGE (gsi.phi (), exit)); if (MAY_HAVE_DEBUG_BIND_STMTS) adjust_debug_stmts (orig_def, PHI_RESULT (gsi.phi ()), exit->dest); } auto loop_exits = get_loop_exit_edges (loop); auto_vec doms; if (at_exit) /* Add the loop copy at exit. */ { if (scalar_loop != loop && new_exit->dest != exit_dest) { new_exit = redirect_edge_and_branch (new_exit, exit_dest); flush_pending_stmts (new_exit); } auto_vec new_phis; hash_map new_phi_args; /* First create the empty phi nodes so that when we flush the statements they can be filled in. However because there is no order between the PHI nodes in the exits and the loop headers we need to order them base on the order of the two headers. First record the new phi nodes. */ for (auto gsi_from = gsi_start_phis (scalar_exit->dest); !gsi_end_p (gsi_from); gsi_next (&gsi_from)) { gimple *from_phi = gsi_stmt (gsi_from); tree new_res = copy_ssa_name (gimple_phi_result (from_phi)); gphi *res = create_phi_node (new_res, new_preheader); new_phis.safe_push (res); } /* Then redirect the edges and flush the changes. This writes out the new SSA names. */ for (edge exit : loop_exits) { edge temp_e = redirect_edge_and_branch (exit, new_preheader); flush_pending_stmts (temp_e); } /* Record the new SSA names in the cache so that we can skip materializing them again when we fill in the rest of the LCSSA variables. */ for (auto phi : new_phis) { tree new_arg = gimple_phi_arg (phi, 0)->def; if (!SSA_VAR_P (new_arg)) continue; /* If the PHI MEM node dominates the loop then we shouldn't create a new LC-SSSA PHI for it in the intermediate block. */ /* A MEM phi that consitutes a new DEF for the vUSE chain can either be a .VDEF or a PHI that operates on MEM. And said definition must not be inside the main loop. Or we must be a parameter. In the last two cases we may remove a non-MEM PHI node, but since they dominate both loops the removal is unlikely to cause trouble as the exits must already be using them. */ if (virtual_operand_p (new_arg) && (SSA_NAME_IS_DEFAULT_DEF (new_arg) || !flow_bb_inside_loop_p (loop, gimple_bb (SSA_NAME_DEF_STMT (new_arg))))) { auto gsi = gsi_for_stmt (phi); remove_phi_node (&gsi, true); continue; } new_phi_args.put (new_arg, gimple_phi_result (phi)); if (TREE_CODE (new_arg) != SSA_NAME) continue; } /* Copy the current loop LC PHI nodes between the original loop exit block and the new loop header. This allows us to later split the preheader block and still find the right LC nodes. */ edge loop_entry = single_succ_edge (new_preheader); if (flow_loops) for (auto gsi_from = gsi_start_phis (loop->header), gsi_to = gsi_start_phis (new_loop->header); !gsi_end_p (gsi_from) && !gsi_end_p (gsi_to); gsi_next (&gsi_from), gsi_next (&gsi_to)) { gimple *from_phi = gsi_stmt (gsi_from); gimple *to_phi = gsi_stmt (gsi_to); tree new_arg = PHI_ARG_DEF_FROM_EDGE (from_phi, loop_latch_edge (loop)); /* Check if we've already created a new phi node during edge redirection. If we have, only propagate the value downwards. */ if (tree *res = new_phi_args.get (new_arg)) { adjust_phi_and_debug_stmts (to_phi, loop_entry, *res); continue; } tree new_res = copy_ssa_name (gimple_phi_result (from_phi)); gphi *lcssa_phi = create_phi_node (new_res, new_preheader); /* Main loop exit should use the final iter value. */ add_phi_arg (lcssa_phi, new_arg, loop_exit, UNKNOWN_LOCATION); adjust_phi_and_debug_stmts (to_phi, loop_entry, new_res); } set_immediate_dominator (CDI_DOMINATORS, new_preheader, e->src); if (was_imm_dom || duplicate_outer_loop) set_immediate_dominator (CDI_DOMINATORS, exit_dest, new_exit->src); /* And remove the non-necessary forwarder again. Keep the other one so we have a proper pre-header for the loop at the exit edge. */ redirect_edge_pred (single_succ_edge (preheader), single_pred (preheader)); delete_basic_block (preheader); set_immediate_dominator (CDI_DOMINATORS, scalar_loop->header, loop_preheader_edge (scalar_loop)->src); } else /* Add the copy at entry. */ { /* Copy the current loop LC PHI nodes between the original loop exit block and the new loop header. This allows us to later split the preheader block and still find the right LC nodes. */ if (flow_loops) for (auto gsi_from = gsi_start_phis (new_loop->header), gsi_to = gsi_start_phis (loop->header); !gsi_end_p (gsi_from) && !gsi_end_p (gsi_to); gsi_next (&gsi_from), gsi_next (&gsi_to)) { gimple *from_phi = gsi_stmt (gsi_from); gimple *to_phi = gsi_stmt (gsi_to); tree new_arg = PHI_ARG_DEF_FROM_EDGE (from_phi, loop_latch_edge (new_loop)); adjust_phi_and_debug_stmts (to_phi, loop_preheader_edge (loop), new_arg); } if (scalar_loop != loop) { /* Remove the non-necessary forwarder of scalar_loop again. */ redirect_edge_pred (single_succ_edge (preheader), single_pred (preheader)); delete_basic_block (preheader); set_immediate_dominator (CDI_DOMINATORS, scalar_loop->header, loop_preheader_edge (scalar_loop)->src); preheader = split_edge (loop_preheader_edge (loop)); entry_e = single_pred_edge (preheader); } redirect_edge_and_branch_force (entry_e, new_preheader); flush_pending_stmts (entry_e); set_immediate_dominator (CDI_DOMINATORS, new_preheader, entry_e->src); redirect_edge_and_branch_force (new_exit, preheader); flush_pending_stmts (new_exit); set_immediate_dominator (CDI_DOMINATORS, preheader, new_exit->src); /* And remove the non-necessary forwarder again. Keep the other one so we have a proper pre-header for the loop at the exit edge. */ redirect_edge_pred (single_succ_edge (new_preheader), single_pred (new_preheader)); delete_basic_block (new_preheader); set_immediate_dominator (CDI_DOMINATORS, new_loop->header, loop_preheader_edge (new_loop)->src); } free (new_bbs); free (bbs); checking_verify_dominators (CDI_DOMINATORS); return new_loop; } /* Given the condition expression COND, put it as the last statement of GUARD_BB; set both edges' probability; set dominator of GUARD_TO to DOM_BB; return the skip edge. GUARD_TO is the target basic block to skip the loop. PROBABILITY is the skip edge's probability. Mark the new edge as irreducible if IRREDUCIBLE_P is true. */ static edge slpeel_add_loop_guard (basic_block guard_bb, tree cond, basic_block guard_to, basic_block dom_bb, profile_probability probability, bool irreducible_p) { gimple_stmt_iterator gsi; edge new_e, enter_e; gcond *cond_stmt; gimple_seq gimplify_stmt_list = NULL; enter_e = EDGE_SUCC (guard_bb, 0); enter_e->flags &= ~EDGE_FALLTHRU; enter_e->flags |= EDGE_FALSE_VALUE; gsi = gsi_last_bb (guard_bb); cond = force_gimple_operand_1 (cond, &gimplify_stmt_list, is_gimple_condexpr_for_cond, NULL_TREE); if (gimplify_stmt_list) gsi_insert_seq_after (&gsi, gimplify_stmt_list, GSI_NEW_STMT); cond_stmt = gimple_build_cond_from_tree (cond, NULL_TREE, NULL_TREE); gsi = gsi_last_bb (guard_bb); gsi_insert_after (&gsi, cond_stmt, GSI_NEW_STMT); /* Add new edge to connect guard block to the merge/loop-exit block. */ new_e = make_edge (guard_bb, guard_to, EDGE_TRUE_VALUE); new_e->probability = probability; if (irreducible_p) new_e->flags |= EDGE_IRREDUCIBLE_LOOP; enter_e->probability = probability.invert (); set_immediate_dominator (CDI_DOMINATORS, guard_to, dom_bb); /* Split enter_e to preserve LOOPS_HAVE_PREHEADERS. */ if (enter_e->dest->loop_father->header == enter_e->dest) split_edge (enter_e); return new_e; } /* This function verifies that the following restrictions apply to LOOP: (1) it consists of exactly 2 basic blocks - header, and an empty latch for innermost loop and 5 basic blocks for outer-loop. (2) it is single entry, single exit (3) its exit condition is the last stmt in the header (4) E is the entry/exit edge of LOOP. */ bool slpeel_can_duplicate_loop_p (const class loop *loop, const_edge exit_e, const_edge e) { edge entry_e = loop_preheader_edge (loop); gcond *orig_cond = get_loop_exit_condition (exit_e); gimple_stmt_iterator loop_exit_gsi = gsi_last_bb (exit_e->src); unsigned int num_bb = loop->inner? 5 : 2; /* All loops have an outer scope; the only case loop->outer is NULL is for the function itself. */ if (!loop_outer (loop) || loop->num_nodes != num_bb || !empty_block_p (loop->latch) || !exit_e /* Verify that new loop exit condition can be trivially modified. */ || (!orig_cond || orig_cond != gsi_stmt (loop_exit_gsi)) || (e != exit_e && e != entry_e)) return false; basic_block *bbs = XNEWVEC (basic_block, loop->num_nodes); get_loop_body_with_size (loop, bbs, loop->num_nodes); bool ret = can_copy_bbs_p (bbs, loop->num_nodes); free (bbs); return ret; } /* Function find_loop_location. Extract the location of the loop in the source code. If the loop is not well formed for vectorization, an estimated location is calculated. Return the loop location if succeed and NULL if not. */ dump_user_location_t find_loop_location (class loop *loop) { gimple *stmt = NULL; basic_block bb; gimple_stmt_iterator si; if (!loop) return dump_user_location_t (); /* For the root of the loop tree return the function location. */ if (!loop_outer (loop)) return dump_user_location_t::from_function_decl (cfun->decl); if (loops_state_satisfies_p (LOOPS_HAVE_RECORDED_EXITS)) { /* We only care about the loop location, so use any exit with location information. */ for (edge e : get_loop_exit_edges (loop)) { stmt = get_loop_exit_condition (e); if (stmt && LOCATION_LOCUS (gimple_location (stmt)) > BUILTINS_LOCATION) return stmt; } } /* If we got here the loop is probably not "well formed", try to estimate the loop location */ if (!loop->header) return dump_user_location_t (); bb = loop->header; for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si)) { stmt = gsi_stmt (si); if (LOCATION_LOCUS (gimple_location (stmt)) > BUILTINS_LOCATION) return stmt; } return dump_user_location_t (); } /* Return true if the phi described by STMT_INFO defines an IV of the loop to be vectorized. */ static bool iv_phi_p (stmt_vec_info stmt_info) { gphi *phi = as_a (stmt_info->stmt); if (virtual_operand_p (PHI_RESULT (phi))) return false; if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_reduction_def || STMT_VINFO_DEF_TYPE (stmt_info) == vect_double_reduction_def) return false; return true; } /* Return true if vectorizer can peel for nonlinear iv. */ static bool vect_can_peel_nonlinear_iv_p (loop_vec_info loop_vinfo, stmt_vec_info stmt_info) { enum vect_induction_op_type induction_type = STMT_VINFO_LOOP_PHI_EVOLUTION_TYPE (stmt_info); tree niters_skip; /* Init_expr will be update by vect_update_ivs_after_vectorizer, if niters or vf is unkown: For shift, when shift mount >= precision, there would be UD. For mult, don't known how to generate init_expr * pow (step, niters) for variable niters. For neg, it should be ok, since niters of vectorized main loop will always be multiple of 2. */ if ((!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) || !LOOP_VINFO_VECT_FACTOR (loop_vinfo).is_constant ()) && induction_type != vect_step_op_neg) { if (dump_enabled_p ()) dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, "Peeling for epilogue is not supported" " for nonlinear induction except neg" " when iteration count is unknown.\n"); return false; } /* Avoid compile time hog on vect_peel_nonlinear_iv_init. */ if (induction_type == vect_step_op_mul) { tree step_expr = STMT_VINFO_LOOP_PHI_EVOLUTION_PART (stmt_info); tree type = TREE_TYPE (step_expr); if (wi::exact_log2 (wi::to_wide (step_expr)) == -1 && LOOP_VINFO_INT_NITERS(loop_vinfo) >= TYPE_PRECISION (type)) { if (dump_enabled_p ()) dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, "Avoid compile time hog on" " vect_peel_nonlinear_iv_init" " for nonlinear induction vec_step_op_mul" " when iteration count is too big.\n"); return false; } } /* Also doens't support peel for neg when niter is variable. ??? generate something like niter_expr & 1 ? init_expr : -init_expr? */ niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); if ((niters_skip != NULL_TREE && (TREE_CODE (niters_skip) != INTEGER_CST || (HOST_WIDE_INT) TREE_INT_CST_LOW (niters_skip) < 0)) || (!vect_use_loop_mask_for_alignment_p (loop_vinfo) && LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) < 0)) { if (dump_enabled_p ()) dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, "Peeling for alignement is not supported" " for nonlinear induction when niters_skip" " is not constant.\n"); return false; } return true; } /* Function vect_can_advance_ivs_p In case the number of iterations that LOOP iterates is unknown at compile time, an epilog loop will be generated, and the loop induction variables (IVs) will be "advanced" to the value they are supposed to take just before the epilog loop. Here we check that the access function of the loop IVs and the expression that represents the loop bound are simple enough. These restrictions will be relaxed in the future. */ bool vect_can_advance_ivs_p (loop_vec_info loop_vinfo) { class loop *loop = LOOP_VINFO_LOOP (loop_vinfo); basic_block bb = loop->header; gphi_iterator gsi; /* Analyze phi functions of the loop header. */ if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "vect_can_advance_ivs_p:\n"); for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi)) { tree evolution_part; enum vect_induction_op_type induction_type; gphi *phi = gsi.phi (); stmt_vec_info phi_info = loop_vinfo->lookup_stmt (phi); if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "Analyze phi: %G", phi_info->stmt); /* Skip virtual phi's. The data dependences that are associated with virtual defs/uses (i.e., memory accesses) are analyzed elsewhere. Skip reduction phis. */ if (!iv_phi_p (phi_info)) { if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "reduc or virtual phi. skip.\n"); continue; } induction_type = STMT_VINFO_LOOP_PHI_EVOLUTION_TYPE (phi_info); if (induction_type != vect_step_op_add) { if (!vect_can_peel_nonlinear_iv_p (loop_vinfo, phi_info)) return false; continue; } /* Analyze the evolution function. */ evolution_part = STMT_VINFO_LOOP_PHI_EVOLUTION_PART (phi_info); if (evolution_part == NULL_TREE) { if (dump_enabled_p ()) dump_printf (MSG_MISSED_OPTIMIZATION, "No access function or evolution.\n"); return false; } /* FORNOW: We do not transform initial conditions of IVs which evolution functions are not invariants in the loop. */ if (!expr_invariant_in_loop_p (loop, evolution_part)) { if (dump_enabled_p ()) dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, "evolution not invariant in loop.\n"); return false; } /* FORNOW: We do not transform initial conditions of IVs which evolution functions are a polynomial of degree >= 2. */ if (tree_is_chrec (evolution_part)) { if (dump_enabled_p ()) dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, "evolution is chrec.\n"); return false; } } return true; } /* 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) { gphi_iterator gsi, gsi1; class loop *loop = LOOP_VINFO_LOOP (loop_vinfo); basic_block update_bb = update_e->dest; basic_block exit_bb = LOOP_VINFO_IV_EXIT (loop_vinfo)->dest; /* Make sure there exists a single-predecessor exit bb: */ gcc_assert (single_pred_p (exit_bb)); gcc_assert (single_succ_edge (exit_bb) == update_e); 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 init_expr; tree step_expr, off; tree type; tree var, ni, ni_name; gimple_stmt_iterator last_gsi; gphi *phi = gsi.phi (); gphi *phi1 = gsi1.phi (); stmt_vec_info phi_info = loop_vinfo->lookup_stmt (phi); if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "vect_update_ivs_after_vectorizer: phi: %G", (gimple *) phi); /* Skip reduction and virtual phis. */ if (!iv_phi_p (phi_info)) { if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "reduc or virtual phi. skip.\n"); continue; } type = TREE_TYPE (gimple_phi_result (phi)); step_expr = STMT_VINFO_LOOP_PHI_EVOLUTION_PART (phi_info); step_expr = unshare_expr (step_expr); /* FORNOW: We do not support IVs whose evolution function is a polynomial of degree >= 2 or exponential. */ gcc_assert (!tree_is_chrec (step_expr)); init_expr = PHI_ARG_DEF_FROM_EDGE (phi, loop_preheader_edge (loop)); gimple_seq stmts = NULL; enum vect_induction_op_type induction_type = STMT_VINFO_LOOP_PHI_EVOLUTION_TYPE (phi_info); if (induction_type == vect_step_op_add) { tree stype = TREE_TYPE (step_expr); off = fold_build2 (MULT_EXPR, stype, fold_convert (stype, niters), step_expr); if (POINTER_TYPE_P (type)) ni = fold_build_pointer_plus (init_expr, off); else ni = fold_convert (type, fold_build2 (PLUS_EXPR, stype, fold_convert (stype, init_expr), off)); } /* Don't bother call vect_peel_nonlinear_iv_init. */ else if (induction_type == vect_step_op_neg) ni = init_expr; else ni = vect_peel_nonlinear_iv_init (&stmts, init_expr, niters, step_expr, induction_type); var = create_tmp_var (type, "tmp"); last_gsi = gsi_last_bb (exit_bb); gimple_seq new_stmts = NULL; ni_name = force_gimple_operand (ni, &new_stmts, false, var); /* Exit_bb shouldn't be empty. */ if (!gsi_end_p (last_gsi)) { gsi_insert_seq_after (&last_gsi, stmts, GSI_SAME_STMT); gsi_insert_seq_after (&last_gsi, new_stmts, GSI_SAME_STMT); } else { gsi_insert_seq_before (&last_gsi, stmts, GSI_SAME_STMT); gsi_insert_seq_before (&last_gsi, new_stmts, GSI_SAME_STMT); } /* Fix phi expressions in the successor bb. */ adjust_phi_and_debug_stmts (phi1, update_e, ni_name); } } /* Return a gimple value containing the misalignment (measured in vector elements) for the loop described by LOOP_VINFO, i.e. how many elements it is away from a perfectly aligned address. Add any new statements to SEQ. */ static tree get_misalign_in_elems (gimple **seq, loop_vec_info loop_vinfo) { dr_vec_info *dr_info = LOOP_VINFO_UNALIGNED_DR (loop_vinfo); stmt_vec_info stmt_info = dr_info->stmt; tree vectype = STMT_VINFO_VECTYPE (stmt_info); poly_uint64 target_align = DR_TARGET_ALIGNMENT (dr_info); unsigned HOST_WIDE_INT target_align_c; tree target_align_minus_1; bool negative = tree_int_cst_compare (DR_STEP (dr_info->dr), size_zero_node) < 0; tree offset = (negative ? size_int ((-TYPE_VECTOR_SUBPARTS (vectype) + 1) * TREE_INT_CST_LOW (TYPE_SIZE_UNIT (TREE_TYPE (vectype)))) : size_zero_node); tree start_addr = vect_create_addr_base_for_vector_ref (loop_vinfo, stmt_info, seq, offset); tree type = unsigned_type_for (TREE_TYPE (start_addr)); if (target_align.is_constant (&target_align_c)) target_align_minus_1 = build_int_cst (type, target_align_c - 1); else { tree vla = build_int_cst (type, target_align); tree vla_align = fold_build2 (BIT_AND_EXPR, type, vla, fold_build2 (MINUS_EXPR, type, build_int_cst (type, 0), vla)); target_align_minus_1 = fold_build2 (MINUS_EXPR, type, vla_align, build_int_cst (type, 1)); } HOST_WIDE_INT elem_size = int_cst_value (TYPE_SIZE_UNIT (TREE_TYPE (vectype))); tree elem_size_log = build_int_cst (type, exact_log2 (elem_size)); /* Create: misalign_in_bytes = addr & (target_align - 1). */ tree int_start_addr = fold_convert (type, start_addr); tree misalign_in_bytes = fold_build2 (BIT_AND_EXPR, type, int_start_addr, target_align_minus_1); /* Create: misalign_in_elems = misalign_in_bytes / element_size. */ tree misalign_in_elems = fold_build2 (RSHIFT_EXPR, type, misalign_in_bytes, elem_size_log); return misalign_in_elems; } /* Function vect_gen_prolog_loop_niters Generate the number of iterations which should be peeled as prolog for the loop represented by LOOP_VINFO. It is calculated as 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 & (target_align - 1) prolog_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) The computations will be emitted at the end of BB. We also compute and store upper bound (included) of the result in BOUND. 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_prolog_loop_niters (loop_vec_info loop_vinfo, basic_block bb, int *bound) { dr_vec_info *dr_info = LOOP_VINFO_UNALIGNED_DR (loop_vinfo); tree var; tree niters_type = TREE_TYPE (LOOP_VINFO_NITERS (loop_vinfo)); gimple_seq stmts = NULL, new_stmts = NULL; tree iters, iters_name; stmt_vec_info stmt_info = dr_info->stmt; tree vectype = STMT_VINFO_VECTYPE (stmt_info); poly_uint64 target_align = DR_TARGET_ALIGNMENT (dr_info); if (LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) > 0) { int npeel = LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo); if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "known peeling = %d.\n", npeel); iters = build_int_cst (niters_type, npeel); *bound = LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo); } else { tree misalign_in_elems = get_misalign_in_elems (&stmts, loop_vinfo); tree type = TREE_TYPE (misalign_in_elems); HOST_WIDE_INT elem_size = int_cst_value (TYPE_SIZE_UNIT (TREE_TYPE (vectype))); /* We only do prolog peeling if the target alignment is known at compile time. */ poly_uint64 align_in_elems = exact_div (target_align, elem_size); tree align_in_elems_minus_1 = build_int_cst (type, align_in_elems - 1); tree align_in_elems_tree = build_int_cst (type, align_in_elems); /* Create: (niters_type) ((align_in_elems - misalign_in_elems) & (align_in_elems - 1)). */ bool negative = tree_int_cst_compare (DR_STEP (dr_info->dr), size_zero_node) < 0; if (negative) iters = fold_build2 (MINUS_EXPR, type, misalign_in_elems, align_in_elems_tree); else iters = fold_build2 (MINUS_EXPR, type, align_in_elems_tree, misalign_in_elems); iters = fold_build2 (BIT_AND_EXPR, type, iters, align_in_elems_minus_1); iters = fold_convert (niters_type, iters); unsigned HOST_WIDE_INT align_in_elems_c; if (align_in_elems.is_constant (&align_in_elems_c)) *bound = align_in_elems_c - 1; else *bound = -1; } if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "niters for prolog loop: %T\n", iters); var = create_tmp_var (niters_type, "prolog_loop_niters"); iters_name = force_gimple_operand (iters, &new_stmts, false, var); if (new_stmts) gimple_seq_add_seq (&stmts, new_stmts); if (stmts) { gcc_assert (single_succ_p (bb)); gimple_stmt_iterator gsi = gsi_last_bb (bb); if (gsi_end_p (gsi)) gsi_insert_seq_before (&gsi, stmts, GSI_SAME_STMT); else gsi_insert_seq_after (&gsi, stmts, GSI_SAME_STMT); } return iters_name; } /* Function vect_update_init_of_dr If CODE is PLUS, the vector loop starts NITERS iterations after the scalar one, otherwise CODE is MINUS and the vector loop starts NITERS iterations before the scalar one (using masking to skip inactive elements). This function updates the information recorded in DR to account for the difference. Specifically, it updates the OFFSET field of DR_INFO. */ static void vect_update_init_of_dr (dr_vec_info *dr_info, tree niters, tree_code code) { struct data_reference *dr = dr_info->dr; tree offset = dr_info->offset; if (!offset) offset = build_zero_cst (sizetype); niters = fold_build2 (MULT_EXPR, sizetype, fold_convert (sizetype, niters), fold_convert (sizetype, DR_STEP (dr))); offset = fold_build2 (code, sizetype, fold_convert (sizetype, offset), niters); dr_info->offset = offset; } /* Function vect_update_inits_of_drs Apply vect_update_inits_of_dr to all accesses in LOOP_VINFO. CODE and NITERS are as for vect_update_inits_of_dr. */ void vect_update_inits_of_drs (loop_vec_info loop_vinfo, tree niters, tree_code code) { unsigned int i; vec datarefs = LOOP_VINFO_DATAREFS (loop_vinfo); struct data_reference *dr; DUMP_VECT_SCOPE ("vect_update_inits_of_dr"); /* Adjust niters to sizetype. We used to insert the stmts on loop preheader here, but since we might use these niters to update the epilogues niters and data references we can't insert them here as this definition might not always dominate its uses. */ if (!types_compatible_p (sizetype, TREE_TYPE (niters))) niters = fold_convert (sizetype, niters); FOR_EACH_VEC_ELT (datarefs, i, dr) { dr_vec_info *dr_info = loop_vinfo->lookup_dr (dr); if (!STMT_VINFO_GATHER_SCATTER_P (dr_info->stmt) && !STMT_VINFO_SIMD_LANE_ACCESS_P (dr_info->stmt)) vect_update_init_of_dr (dr_info, niters, code); } } /* For the information recorded in LOOP_VINFO prepare the loop for peeling by masking. This involves calculating the number of iterations to be peeled and then aligning all memory references appropriately. */ void vect_prepare_for_masked_peels (loop_vec_info loop_vinfo) { tree misalign_in_elems; tree type = TREE_TYPE (LOOP_VINFO_NITERS (loop_vinfo)); gcc_assert (vect_use_loop_mask_for_alignment_p (loop_vinfo)); /* From the information recorded in LOOP_VINFO get the number of iterations that need to be skipped via masking. */ if (LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) > 0) { poly_int64 misalign = (LOOP_VINFO_VECT_FACTOR (loop_vinfo) - LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo)); misalign_in_elems = build_int_cst (type, misalign); } else { gimple_seq seq1 = NULL, seq2 = NULL; misalign_in_elems = get_misalign_in_elems (&seq1, loop_vinfo); misalign_in_elems = fold_convert (type, misalign_in_elems); misalign_in_elems = force_gimple_operand (misalign_in_elems, &seq2, true, NULL_TREE); gimple_seq_add_seq (&seq1, seq2); if (seq1) { edge pe = loop_preheader_edge (LOOP_VINFO_LOOP (loop_vinfo)); basic_block new_bb = gsi_insert_seq_on_edge_immediate (pe, seq1); gcc_assert (!new_bb); } } if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "misalignment for fully-masked loop: %T\n", misalign_in_elems); LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo) = misalign_in_elems; vect_update_inits_of_drs (loop_vinfo, misalign_in_elems, MINUS_EXPR); } /* This function builds ni_name = number of iterations. Statements are emitted on the loop preheader edge. If NEW_VAR_P is not NULL, set it to TRUE if new ssa_var is generated. */ tree vect_build_loop_niters (loop_vec_info loop_vinfo, bool *new_var_p) { tree ni = unshare_expr (LOOP_VINFO_NITERS (loop_vinfo)); if (TREE_CODE (ni) == INTEGER_CST) return ni; else { tree ni_name, var; gimple_seq stmts = NULL; edge pe = loop_preheader_edge (LOOP_VINFO_LOOP (loop_vinfo)); var = create_tmp_var (TREE_TYPE (ni), "niters"); ni_name = force_gimple_operand (ni, &stmts, false, var); if (stmts) { gsi_insert_seq_on_edge_immediate (pe, stmts); if (new_var_p != NULL) *new_var_p = true; } return ni_name; } } /* Calculate the number of iterations above which vectorized loop will be preferred than scalar loop. NITERS_PROLOG is the number of iterations of prolog loop. If it's integer const, the integer number is also passed in INT_NITERS_PROLOG. BOUND_PROLOG is the upper bound (inclusive) of the number of iterations of the prolog loop. BOUND_EPILOG is the corresponding value for the epilog loop. If CHECK_PROFITABILITY is true, TH is the threshold below which the scalar (rather than vectorized) loop will be executed. This function stores the upper bound (inclusive) of the result in BOUND_SCALAR. */ static tree vect_gen_scalar_loop_niters (tree niters_prolog, int int_niters_prolog, int bound_prolog, poly_int64 bound_epilog, int th, poly_uint64 *bound_scalar, bool check_profitability) { tree type = TREE_TYPE (niters_prolog); tree niters = fold_build2 (PLUS_EXPR, type, niters_prolog, build_int_cst (type, bound_epilog)); *bound_scalar = bound_prolog + bound_epilog; if (check_profitability) { /* TH indicates the minimum niters of vectorized loop, while we compute the maximum niters of scalar loop. */ th--; /* Peeling for constant times. */ if (int_niters_prolog >= 0) { *bound_scalar = upper_bound (int_niters_prolog + bound_epilog, th); return build_int_cst (type, *bound_scalar); } /* Peeling an unknown number of times. Note that both BOUND_PROLOG and BOUND_EPILOG are inclusive upper bounds. */ if (known_ge (th, bound_prolog + bound_epilog)) { *bound_scalar = th; return build_int_cst (type, th); } /* Need to do runtime comparison. */ else if (maybe_gt (th, bound_epilog)) { *bound_scalar = upper_bound (*bound_scalar, th); return fold_build2 (MAX_EXPR, type, build_int_cst (type, th), niters); } } return niters; } /* NITERS is the number of times that the original scalar loop executes after peeling. Work out the maximum number of iterations N that can be handled by the vectorized form of the loop and then either: a) set *STEP_VECTOR_PTR to the vectorization factor and generate: niters_vector = N b) set *STEP_VECTOR_PTR to one and generate: niters_vector = N / vf In both cases, store niters_vector in *NITERS_VECTOR_PTR and add any new statements on the loop preheader edge. NITERS_NO_OVERFLOW is true if NITERS doesn't overflow (i.e. if NITERS is always nonzero). */ void vect_gen_vector_loop_niters (loop_vec_info loop_vinfo, tree niters, tree *niters_vector_ptr, tree *step_vector_ptr, bool niters_no_overflow) { tree ni_minus_gap, var; tree niters_vector, step_vector, type = TREE_TYPE (niters); poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); edge pe = loop_preheader_edge (LOOP_VINFO_LOOP (loop_vinfo)); tree log_vf = NULL_TREE; /* If epilogue loop is required because of data accesses with gaps, we subtract one iteration from the total number of iterations here for correct calculation of RATIO. */ if (LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo)) { ni_minus_gap = fold_build2 (MINUS_EXPR, type, niters, build_one_cst (type)); if (!is_gimple_val (ni_minus_gap)) { var = create_tmp_var (type, "ni_gap"); gimple *stmts = NULL; ni_minus_gap = force_gimple_operand (ni_minus_gap, &stmts, true, var); gsi_insert_seq_on_edge_immediate (pe, stmts); } } else ni_minus_gap = niters; /* To silence some unexpected warnings, simply initialize to 0. */ unsigned HOST_WIDE_INT const_vf = 0; if (vf.is_constant (&const_vf) && !LOOP_VINFO_USING_PARTIAL_VECTORS_P (loop_vinfo)) { /* Create: niters >> log2(vf) */ /* If it's known that niters == number of latch executions + 1 doesn't overflow, we can generate niters >> log2(vf); otherwise we generate (niters - vf) >> log2(vf) + 1 by using the fact that we know ratio will be at least one. */ log_vf = build_int_cst (type, exact_log2 (const_vf)); if (niters_no_overflow) niters_vector = fold_build2 (RSHIFT_EXPR, type, ni_minus_gap, log_vf); else niters_vector = fold_build2 (PLUS_EXPR, type, fold_build2 (RSHIFT_EXPR, type, fold_build2 (MINUS_EXPR, type, ni_minus_gap, build_int_cst (type, vf)), log_vf), build_int_cst (type, 1)); step_vector = build_one_cst (type); } else { niters_vector = ni_minus_gap; step_vector = build_int_cst (type, vf); } if (!is_gimple_val (niters_vector)) { var = create_tmp_var (type, "bnd"); gimple_seq stmts = NULL; niters_vector = force_gimple_operand (niters_vector, &stmts, true, var); gsi_insert_seq_on_edge_immediate (pe, stmts); /* Peeling algorithm guarantees that vector loop bound is at least ONE, we set range information to make niters analyzer's life easier. Note the number of latch iteration value can be TYPE_MAX_VALUE so we have to represent the vector niter TYPE_MAX_VALUE + 1 >> log_vf. */ if (stmts != NULL && log_vf) { if (niters_no_overflow) { value_range vr (type, wi::one (TYPE_PRECISION (type)), wi::rshift (wi::max_value (TYPE_PRECISION (type), TYPE_SIGN (type)), exact_log2 (const_vf), TYPE_SIGN (type))); set_range_info (niters_vector, vr); } /* For VF == 1 the vector IV might also overflow so we cannot assert a minimum value of 1. */ else if (const_vf > 1) { value_range vr (type, wi::one (TYPE_PRECISION (type)), wi::rshift (wi::max_value (TYPE_PRECISION (type), TYPE_SIGN (type)) - (const_vf - 1), exact_log2 (const_vf), TYPE_SIGN (type)) + 1); set_range_info (niters_vector, vr); } } } *niters_vector_ptr = niters_vector; *step_vector_ptr = step_vector; return; } /* Given NITERS_VECTOR which is the number of iterations for vectorized loop specified by LOOP_VINFO after vectorization, compute the number of iterations before vectorization (niters_vector * vf) and store it to NITERS_VECTOR_MULT_VF_PTR. */ static void vect_gen_vector_loop_niters_mult_vf (loop_vec_info loop_vinfo, tree niters_vector, tree *niters_vector_mult_vf_ptr) { /* We should be using a step_vector of VF if VF is variable. */ int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo).to_constant (); tree type = TREE_TYPE (niters_vector); tree log_vf = build_int_cst (type, exact_log2 (vf)); basic_block exit_bb = LOOP_VINFO_IV_EXIT (loop_vinfo)->dest; gcc_assert (niters_vector_mult_vf_ptr != NULL); tree niters_vector_mult_vf = fold_build2 (LSHIFT_EXPR, type, niters_vector, log_vf); if (!is_gimple_val (niters_vector_mult_vf)) { tree var = create_tmp_var (type, "niters_vector_mult_vf"); gimple_seq stmts = NULL; niters_vector_mult_vf = force_gimple_operand (niters_vector_mult_vf, &stmts, true, var); gimple_stmt_iterator gsi = gsi_start_bb (exit_bb); gsi_insert_seq_before (&gsi, stmts, GSI_SAME_STMT); } *niters_vector_mult_vf_ptr = niters_vector_mult_vf; } /* Function slpeel_add_loop_guard adds guard skipping from the beginning of SKIP_LOOP to the beginning of UPDATE_LOOP. GUARD_EDGE and MERGE_EDGE are two pred edges of the merge point before UPDATE_LOOP. The two loops appear like below: guard_bb: if (cond) goto merge_bb; else goto skip_loop; skip_loop: header_a: i_1 = PHI; ... i_2 = i_1 + 1; if (cond_a) goto latch_a; else goto exit_a; latch_a: goto header_a; exit_a: i_5 = PHI; merge_bb: ;; PHI (i_x = PHI) to be created at merge point. update_loop: header_b: i_3 = PHI; ;; Use of i_5 to be replaced with i_x. ... i_4 = i_3 + 1; if (cond_b) goto latch_b; else goto exit_bb; latch_b: goto header_b; exit_bb: This function creates PHI nodes at merge_bb and replaces the use of i_5 in the update_loop's PHI node with the result of new PHI result. */ static void slpeel_update_phi_nodes_for_guard1 (class loop *skip_loop, class loop *update_loop, edge guard_edge, edge merge_edge) { location_t merge_loc, guard_loc; edge orig_e = loop_preheader_edge (skip_loop); edge update_e = loop_preheader_edge (update_loop); gphi_iterator gsi_orig, gsi_update; for ((gsi_orig = gsi_start_phis (skip_loop->header), gsi_update = gsi_start_phis (update_loop->header)); !gsi_end_p (gsi_orig) && !gsi_end_p (gsi_update); gsi_next (&gsi_orig), gsi_next (&gsi_update)) { gphi *orig_phi = gsi_orig.phi (); gphi *update_phi = gsi_update.phi (); /* Generate new phi node at merge bb of the guard. */ tree new_res = copy_ssa_name (PHI_RESULT (orig_phi)); gphi *new_phi = create_phi_node (new_res, guard_edge->dest); /* Merge bb has two incoming edges: GUARD_EDGE and MERGE_EDGE. Set the args in NEW_PHI for these edges. */ tree merge_arg = PHI_ARG_DEF_FROM_EDGE (update_phi, update_e); tree guard_arg = PHI_ARG_DEF_FROM_EDGE (orig_phi, orig_e); merge_loc = gimple_phi_arg_location_from_edge (update_phi, update_e); guard_loc = gimple_phi_arg_location_from_edge (orig_phi, orig_e); add_phi_arg (new_phi, merge_arg, merge_edge, merge_loc); add_phi_arg (new_phi, guard_arg, guard_edge, guard_loc); /* Update phi in UPDATE_PHI. */ adjust_phi_and_debug_stmts (update_phi, update_e, new_res); } } /* LOOP_VINFO is an epilogue loop whose corresponding main loop can be skipped. Return a value that equals: - MAIN_LOOP_VALUE when LOOP_VINFO is entered from the main loop and - SKIP_VALUE when the main loop is skipped. */ tree vect_get_main_loop_result (loop_vec_info loop_vinfo, tree main_loop_value, tree skip_value) { gcc_assert (loop_vinfo->main_loop_edge); tree phi_result = make_ssa_name (TREE_TYPE (main_loop_value)); basic_block bb = loop_vinfo->main_loop_edge->dest; gphi *new_phi = create_phi_node (phi_result, bb); add_phi_arg (new_phi, main_loop_value, loop_vinfo->main_loop_edge, UNKNOWN_LOCATION); add_phi_arg (new_phi, skip_value, loop_vinfo->skip_main_loop_edge, UNKNOWN_LOCATION); return phi_result; } /* Function vect_do_peeling. Input: - LOOP_VINFO: Represent a loop to be vectorized, which looks like: preheader: LOOP: header_bb: loop_body if (exit_loop_cond) goto exit_bb else goto header_bb exit_bb: - NITERS: The number of iterations of the loop. - NITERSM1: The number of iterations of the loop's latch. - NITERS_NO_OVERFLOW: No overflow in computing NITERS. - TH, CHECK_PROFITABILITY: Threshold of niters to vectorize loop if CHECK_PROFITABILITY is true. Output: - *NITERS_VECTOR and *STEP_VECTOR describe how the main loop should iterate after vectorization; see vect_set_loop_condition for details. - *NITERS_VECTOR_MULT_VF_VAR is either null or an SSA name that should be set to the number of scalar iterations handled by the vector loop. The SSA name is only used on exit from the loop. This function peels prolog and epilog from the loop, adds guards skipping PROLOG and EPILOG for various conditions. As a result, the changed CFG would look like: guard_bb_1: if (prefer_scalar_loop) goto merge_bb_1 else goto guard_bb_2 guard_bb_2: if (skip_prolog) goto merge_bb_2 else goto prolog_preheader prolog_preheader: PROLOG: prolog_header_bb: prolog_body if (exit_prolog_cond) goto prolog_exit_bb else goto prolog_header_bb prolog_exit_bb: merge_bb_2: vector_preheader: VECTOR LOOP: vector_header_bb: vector_body if (exit_vector_cond) goto vector_exit_bb else goto vector_header_bb vector_exit_bb: guard_bb_3: if (skip_epilog) goto merge_bb_3 else goto epilog_preheader merge_bb_1: epilog_preheader: EPILOG: epilog_header_bb: epilog_body if (exit_epilog_cond) goto merge_bb_3 else goto epilog_header_bb merge_bb_3: Note this function peels prolog and epilog only if it's necessary, as well as guards. This function returns the epilogue loop if a decision was made to vectorize it, otherwise NULL. The analysis resulting in this epilogue loop's loop_vec_info was performed in the same vect_analyze_loop call as the main loop's. At that time vect_analyze_loop constructs a list of accepted loop_vec_info's for lower vectorization factors than the main loop. This list is stored in the main loop's loop_vec_info in the 'epilogue_vinfos' member. Everytime we decide to vectorize the epilogue loop for a lower vectorization factor, the loop_vec_info sitting at the top of the epilogue_vinfos list is removed, updated and linked to the epilogue loop. This is later used to vectorize the epilogue. The reason the loop_vec_info needs updating is that it was constructed based on the original main loop, and the epilogue loop is a copy of this loop, so all links pointing to statements in the original loop need updating. Furthermore, these loop_vec_infos share the data_reference's records, which will also need to be updated. TODO: Guard for prefer_scalar_loop should be emitted along with versioning conditions if loop versioning is needed. */ class loop * vect_do_peeling (loop_vec_info loop_vinfo, tree niters, tree nitersm1, tree *niters_vector, tree *step_vector, tree *niters_vector_mult_vf_var, int th, bool check_profitability, bool niters_no_overflow, tree *advance) { edge e, guard_e; tree type = TREE_TYPE (niters), guard_cond; basic_block guard_bb, guard_to; profile_probability prob_prolog, prob_vector, prob_epilog; int estimated_vf; int prolog_peeling = 0; bool vect_epilogues = loop_vinfo->epilogue_vinfos.length () > 0; /* We currently do not support prolog peeling if the target alignment is not known at compile time. 'vect_gen_prolog_loop_niters' depends on the target alignment being constant. */ dr_vec_info *dr_info = LOOP_VINFO_UNALIGNED_DR (loop_vinfo); if (dr_info && !DR_TARGET_ALIGNMENT (dr_info).is_constant ()) return NULL; if (!vect_use_loop_mask_for_alignment_p (loop_vinfo)) prolog_peeling = LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo); poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); poly_uint64 bound_epilog = 0; if (!LOOP_VINFO_USING_PARTIAL_VECTORS_P (loop_vinfo) && LOOP_VINFO_PEELING_FOR_NITER (loop_vinfo)) bound_epilog += vf - 1; if (LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo)) bound_epilog += 1; bool epilog_peeling = maybe_ne (bound_epilog, 0U); poly_uint64 bound_scalar = bound_epilog; if (!prolog_peeling && !epilog_peeling) return NULL; /* Before doing any peeling make sure to reset debug binds outside of the loop refering to defs not in LC SSA. */ class loop *loop = LOOP_VINFO_LOOP (loop_vinfo); for (unsigned i = 0; i < loop->num_nodes; ++i) { basic_block bb = LOOP_VINFO_BBS (loop_vinfo)[i]; imm_use_iterator ui; gimple *use_stmt; for (gphi_iterator gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi)) { FOR_EACH_IMM_USE_STMT (use_stmt, ui, gimple_phi_result (gsi.phi ())) if (gimple_debug_bind_p (use_stmt) && loop != gimple_bb (use_stmt)->loop_father && !flow_loop_nested_p (loop, gimple_bb (use_stmt)->loop_father)) { gimple_debug_bind_reset_value (use_stmt); update_stmt (use_stmt); } } for (gimple_stmt_iterator gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi)) { ssa_op_iter op_iter; def_operand_p def_p; FOR_EACH_SSA_DEF_OPERAND (def_p, gsi_stmt (gsi), op_iter, SSA_OP_DEF) FOR_EACH_IMM_USE_STMT (use_stmt, ui, DEF_FROM_PTR (def_p)) if (gimple_debug_bind_p (use_stmt) && loop != gimple_bb (use_stmt)->loop_father && !flow_loop_nested_p (loop, gimple_bb (use_stmt)->loop_father)) { gimple_debug_bind_reset_value (use_stmt); update_stmt (use_stmt); } } } prob_vector = profile_probability::guessed_always ().apply_scale (9, 10); estimated_vf = vect_vf_for_cost (loop_vinfo); if (estimated_vf == 2) estimated_vf = 3; prob_prolog = prob_epilog = profile_probability::guessed_always () .apply_scale (estimated_vf - 1, estimated_vf); class loop *prolog, *epilog = NULL; class loop *first_loop = loop; bool irred_flag = loop_preheader_edge (loop)->flags & EDGE_IRREDUCIBLE_LOOP; /* SSA form needs to be up-to-date since we are going to manually update SSA form in slpeel_tree_duplicate_loop_to_edge_cfg and delete all update SSA state after that, so we have to make sure to not lose any pending update needs. */ gcc_assert (!need_ssa_update_p (cfun)); /* If we're vectorizing an epilogue loop, we have ensured that the virtual operand is in SSA form throughout the vectorized main loop. Normally it is possible to trace the updated vector-stmt vdefs back to scalar-stmt vdefs and vector-stmt vuses back to scalar-stmt vuses, meaning that the effect of the SSA update remains local to the main loop. However, there are rare cases in which the vectorized loop should have vdefs even when the original scalar loop didn't. For example, vectorizing a load with IFN_LOAD_LANES introduces clobbers of the temporary vector array, which in turn needs new vdefs. If the scalar loop doesn't write to memory, these new vdefs will be the only ones in the vector loop. We are currently defering updating virtual SSA form and creating of a virtual PHI for this case so we do not have to make sure the newly introduced virtual def is in LCSSA form. */ if (MAY_HAVE_DEBUG_BIND_STMTS) { gcc_assert (!adjust_vec.exists ()); adjust_vec.create (32); } initialize_original_copy_tables (); /* Record the anchor bb at which the guard should be placed if the scalar loop might be preferred. */ basic_block anchor = loop_preheader_edge (loop)->src; /* Generate the number of iterations for the prolog loop. We do this here so that we can also get the upper bound on the number of iterations. */ tree niters_prolog; int bound_prolog = 0; if (prolog_peeling) { niters_prolog = vect_gen_prolog_loop_niters (loop_vinfo, anchor, &bound_prolog); /* If algonment peeling is known, we will always execute prolog. */ if (TREE_CODE (niters_prolog) == INTEGER_CST) prob_prolog = profile_probability::always (); } else niters_prolog = build_int_cst (type, 0); loop_vec_info epilogue_vinfo = NULL; if (vect_epilogues) { epilogue_vinfo = loop_vinfo->epilogue_vinfos[0]; loop_vinfo->epilogue_vinfos.ordered_remove (0); } tree niters_vector_mult_vf = NULL_TREE; /* Saving NITERs before the loop, as this may be changed by prologue. */ tree before_loop_niters = LOOP_VINFO_NITERS (loop_vinfo); edge update_e = NULL, skip_e = NULL; unsigned int lowest_vf = constant_lower_bound (vf); /* Prolog loop may be skipped. */ bool skip_prolog = (prolog_peeling != 0); /* Skip this loop to epilog when there are not enough iterations to enter this vectorized loop. If true we should perform runtime checks on the NITERS to check whether we should skip the current vectorized loop. If we know the number of scalar iterations we may choose to add a runtime check if this number "maybe" smaller than the number of iterations required when we know the number of scalar iterations may potentially be smaller than the number of iterations required to enter this loop, for this we use the upper bounds on the prolog and epilog peeling. When we don't know the number of iterations and don't require versioning it is because we have asserted that there are enough scalar iterations to enter the main loop, so this skip is not necessary. When we are versioning then we only add such a skip if we have chosen to vectorize the epilogue. */ bool skip_vector = (LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) ? maybe_lt (LOOP_VINFO_INT_NITERS (loop_vinfo), bound_prolog + bound_epilog) : (!LOOP_REQUIRES_VERSIONING (loop_vinfo) || vect_epilogues)); /* Epilog loop must be executed if the number of iterations for epilog loop is known at compile time, otherwise we need to add a check at the end of vector loop and skip to the end of epilog loop. */ bool skip_epilog = (prolog_peeling < 0 || !LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) || !vf.is_constant ()); /* PEELING_FOR_GAPS is special because epilog loop must be executed. */ if (LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo)) skip_epilog = false; class loop *scalar_loop = LOOP_VINFO_SCALAR_LOOP (loop_vinfo); auto_vec original_counts; basic_block *original_bbs = NULL; if (skip_vector) { split_edge (loop_preheader_edge (loop)); if (epilog_peeling && (vect_epilogues || scalar_loop == NULL)) { original_bbs = get_loop_body (loop); for (unsigned int i = 0; i < loop->num_nodes; i++) original_counts.safe_push(original_bbs[i]->count); } /* Due to the order in which we peel prolog and epilog, we first propagate probability to the whole loop. The purpose is to avoid adjusting probabilities of both prolog and vector loops separately. Note in this case, the probability of epilog loop needs to be scaled back later. */ basic_block bb_before_loop = loop_preheader_edge (loop)->src; if (prob_vector.initialized_p ()) { scale_bbs_frequencies (&bb_before_loop, 1, prob_vector); scale_loop_profile (loop, prob_vector, -1); } } if (vect_epilogues) { /* Make sure to set the epilogue's epilogue scalar loop, such that we can use the original scalar loop as remaining epilogue if necessary. */ LOOP_VINFO_SCALAR_LOOP (epilogue_vinfo) = LOOP_VINFO_SCALAR_LOOP (loop_vinfo); LOOP_VINFO_SCALAR_IV_EXIT (epilogue_vinfo) = LOOP_VINFO_SCALAR_IV_EXIT (loop_vinfo); } if (prolog_peeling) { e = loop_preheader_edge (loop); edge exit_e = LOOP_VINFO_IV_EXIT (loop_vinfo); gcc_checking_assert (slpeel_can_duplicate_loop_p (loop, exit_e, e)); /* Peel prolog and put it on preheader edge of loop. */ edge scalar_e = LOOP_VINFO_SCALAR_IV_EXIT (loop_vinfo); edge prolog_e = NULL; prolog = slpeel_tree_duplicate_loop_to_edge_cfg (loop, exit_e, scalar_loop, scalar_e, e, &prolog_e); gcc_assert (prolog); prolog->force_vectorize = false; first_loop = prolog; reset_original_copy_tables (); /* Update the number of iterations for prolog loop. */ tree step_prolog = build_one_cst (TREE_TYPE (niters_prolog)); vect_set_loop_condition (prolog, prolog_e, NULL, niters_prolog, step_prolog, NULL_TREE, false); /* Skip the prolog loop. */ if (skip_prolog) { guard_cond = fold_build2 (EQ_EXPR, boolean_type_node, niters_prolog, build_int_cst (type, 0)); guard_bb = loop_preheader_edge (prolog)->src; basic_block bb_after_prolog = loop_preheader_edge (loop)->src; guard_to = split_edge (loop_preheader_edge (loop)); guard_e = slpeel_add_loop_guard (guard_bb, guard_cond, guard_to, guard_bb, prob_prolog.invert (), irred_flag); e = EDGE_PRED (guard_to, 0); e = (e != guard_e ? e : EDGE_PRED (guard_to, 1)); slpeel_update_phi_nodes_for_guard1 (prolog, loop, guard_e, e); scale_bbs_frequencies (&bb_after_prolog, 1, prob_prolog); scale_loop_profile (prolog, prob_prolog, bound_prolog - 1); } /* Update init address of DRs. */ vect_update_inits_of_drs (loop_vinfo, niters_prolog, PLUS_EXPR); /* Update niters for vector loop. */ LOOP_VINFO_NITERS (loop_vinfo) = fold_build2 (MINUS_EXPR, type, niters, niters_prolog); LOOP_VINFO_NITERSM1 (loop_vinfo) = fold_build2 (MINUS_EXPR, type, LOOP_VINFO_NITERSM1 (loop_vinfo), niters_prolog); bool new_var_p = false; niters = vect_build_loop_niters (loop_vinfo, &new_var_p); /* It's guaranteed that vector loop bound before vectorization is at least VF, so set range information for newly generated var. */ if (new_var_p) { value_range vr (type, wi::to_wide (build_int_cst (type, lowest_vf)), wi::to_wide (TYPE_MAX_VALUE (type))); set_range_info (niters, vr); } /* Prolog iterates at most bound_prolog times, latch iterates at most bound_prolog - 1 times. */ record_niter_bound (prolog, bound_prolog - 1, false, true); delete_update_ssa (); adjust_vec_debug_stmts (); scev_reset (); } basic_block bb_before_epilog = NULL; if (epilog_peeling) { e = LOOP_VINFO_IV_EXIT (loop_vinfo); gcc_checking_assert (slpeel_can_duplicate_loop_p (loop, e, e)); /* Peel epilog and put it on exit edge of loop. If we are vectorizing said epilog then we should use a copy of the main loop as a starting point. This loop may have already had some preliminary transformations to allow for more optimal vectorization, for example if-conversion. If we are not vectorizing the epilog then we should use the scalar loop as the transformations mentioned above make less or no sense when not vectorizing. */ edge scalar_e = LOOP_VINFO_SCALAR_IV_EXIT (loop_vinfo); epilog = vect_epilogues ? get_loop_copy (loop) : scalar_loop; edge epilog_e = vect_epilogues ? e : scalar_e; edge new_epilog_e = NULL; epilog = slpeel_tree_duplicate_loop_to_edge_cfg (loop, e, epilog, epilog_e, e, &new_epilog_e); LOOP_VINFO_EPILOGUE_IV_EXIT (loop_vinfo) = new_epilog_e; gcc_assert (epilog); epilog->force_vectorize = false; bb_before_epilog = loop_preheader_edge (epilog)->src; /* Scalar version loop may be preferred. In this case, add guard and skip to epilog. Note this only happens when the number of iterations of loop is unknown at compile time, otherwise this won't be vectorized. */ if (skip_vector) { /* Additional epilogue iteration is peeled if gap exists. */ tree t = vect_gen_scalar_loop_niters (niters_prolog, prolog_peeling, bound_prolog, bound_epilog, th, &bound_scalar, check_profitability); /* Build guard against NITERSM1 since NITERS may overflow. */ guard_cond = fold_build2 (LT_EXPR, boolean_type_node, nitersm1, t); guard_bb = anchor; guard_to = split_edge (loop_preheader_edge (epilog)); guard_e = slpeel_add_loop_guard (guard_bb, guard_cond, guard_to, guard_bb, prob_vector.invert (), irred_flag); skip_e = guard_e; e = EDGE_PRED (guard_to, 0); e = (e != guard_e ? e : EDGE_PRED (guard_to, 1)); slpeel_update_phi_nodes_for_guard1 (first_loop, epilog, guard_e, e); /* Simply propagate profile info from guard_bb to guard_to which is a merge point of control flow. */ profile_count old_count = guard_to->count; guard_to->count = guard_bb->count; /* Restore the counts of the epilog loop if we didn't use the scalar loop. */ if (vect_epilogues || scalar_loop == NULL) { gcc_assert(epilog->num_nodes == loop->num_nodes); basic_block *bbs = get_loop_body (epilog); for (unsigned int i = 0; i < epilog->num_nodes; i++) { gcc_assert(get_bb_original (bbs[i]) == original_bbs[i]); bbs[i]->count = original_counts[i]; } free (bbs); free (original_bbs); } else if (old_count.nonzero_p ()) scale_loop_profile (epilog, guard_to->count.probability_in (old_count), -1); /* Only need to handle basic block before epilog loop if it's not the guard_bb, which is the case when skip_vector is true. */ if (guard_bb != bb_before_epilog) bb_before_epilog->count = single_pred_edge (bb_before_epilog)->count (); bb_before_epilog = loop_preheader_edge (epilog)->src; } /* If loop is peeled for non-zero constant times, now niters refers to orig_niters - prolog_peeling, it won't overflow even the orig_niters overflows. */ niters_no_overflow |= (prolog_peeling > 0); vect_gen_vector_loop_niters (loop_vinfo, niters, niters_vector, step_vector, niters_no_overflow); if (!integer_onep (*step_vector)) { /* On exit from the loop we will have an easy way of calcalating NITERS_VECTOR / STEP * STEP. Install a dummy definition until then. */ niters_vector_mult_vf = make_ssa_name (TREE_TYPE (*niters_vector)); SSA_NAME_DEF_STMT (niters_vector_mult_vf) = gimple_build_nop (); *niters_vector_mult_vf_var = niters_vector_mult_vf; } else vect_gen_vector_loop_niters_mult_vf (loop_vinfo, *niters_vector, &niters_vector_mult_vf); /* Update IVs of original loop as if they were advanced by niters_vector_mult_vf steps. */ gcc_checking_assert (vect_can_advance_ivs_p (loop_vinfo)); update_e = skip_vector ? e : loop_preheader_edge (epilog); vect_update_ivs_after_vectorizer (loop_vinfo, niters_vector_mult_vf, update_e); if (skip_epilog) { guard_cond = fold_build2 (EQ_EXPR, boolean_type_node, niters, niters_vector_mult_vf); guard_bb = LOOP_VINFO_IV_EXIT (loop_vinfo)->dest; edge epilog_e = LOOP_VINFO_EPILOGUE_IV_EXIT (loop_vinfo); guard_to = epilog_e->dest; guard_e = slpeel_add_loop_guard (guard_bb, guard_cond, guard_to, skip_vector ? anchor : guard_bb, prob_epilog.invert (), irred_flag); if (vect_epilogues) epilogue_vinfo->skip_this_loop_edge = guard_e; edge main_iv = LOOP_VINFO_IV_EXIT (loop_vinfo); gphi_iterator gsi2 = gsi_start_phis (main_iv->dest); for (gphi_iterator gsi = gsi_start_phis (guard_to); !gsi_end_p (gsi); gsi_next (&gsi)) { /* We are expecting all of the PHIs we have on epilog_e to be also on the main loop exit. But sometimes a stray virtual definition can appear at epilog_e which we can then take as the same on all exits, we've removed the LC SSA PHI on the main exit before so we wouldn't need to create a loop PHI for it. */ if (virtual_operand_p (gimple_phi_result (*gsi)) && (gsi_end_p (gsi2) || !virtual_operand_p (gimple_phi_result (*gsi2)))) add_phi_arg (*gsi, gimple_phi_arg_def_from_edge (*gsi, epilog_e), guard_e, UNKNOWN_LOCATION); else { add_phi_arg (*gsi, gimple_phi_result (*gsi2), guard_e, UNKNOWN_LOCATION); gsi_next (&gsi2); } } /* Only need to handle basic block before epilog loop if it's not the guard_bb, which is the case when skip_vector is true. */ if (guard_bb != bb_before_epilog) { prob_epilog = prob_vector * prob_epilog + prob_vector.invert (); scale_bbs_frequencies (&bb_before_epilog, 1, prob_epilog); } scale_loop_profile (epilog, prob_epilog, -1); } unsigned HOST_WIDE_INT bound; if (bound_scalar.is_constant (&bound)) { gcc_assert (bound != 0); /* -1 to convert loop iterations to latch iterations. */ record_niter_bound (epilog, bound - 1, false, true); scale_loop_profile (epilog, profile_probability::always (), bound - 1); } delete_update_ssa (); adjust_vec_debug_stmts (); scev_reset (); } if (vect_epilogues) { epilog->aux = epilogue_vinfo; LOOP_VINFO_LOOP (epilogue_vinfo) = epilog; LOOP_VINFO_IV_EXIT (epilogue_vinfo) = LOOP_VINFO_EPILOGUE_IV_EXIT (loop_vinfo); loop_constraint_clear (epilog, LOOP_C_INFINITE); /* We now must calculate the number of NITERS performed by the previous loop and EPILOGUE_NITERS to be performed by the epilogue. */ tree niters = fold_build2 (PLUS_EXPR, TREE_TYPE (niters_vector_mult_vf), niters_prolog, niters_vector_mult_vf); /* If skip_vector we may skip the previous loop, we insert a phi-node to determine whether we are coming from the previous vectorized loop using the update_e edge or the skip_vector basic block using the skip_e edge. */ if (skip_vector) { gcc_assert (update_e != NULL && skip_e != NULL); gphi *new_phi = create_phi_node (make_ssa_name (TREE_TYPE (niters)), update_e->dest); tree new_ssa = make_ssa_name (TREE_TYPE (niters)); gimple *stmt = gimple_build_assign (new_ssa, niters); gimple_stmt_iterator gsi; if (TREE_CODE (niters_vector_mult_vf) == SSA_NAME && SSA_NAME_DEF_STMT (niters_vector_mult_vf)->bb != NULL) { gsi = gsi_for_stmt (SSA_NAME_DEF_STMT (niters_vector_mult_vf)); gsi_insert_after (&gsi, stmt, GSI_NEW_STMT); } else { gsi = gsi_last_bb (update_e->src); gsi_insert_before (&gsi, stmt, GSI_NEW_STMT); } niters = new_ssa; add_phi_arg (new_phi, niters, update_e, UNKNOWN_LOCATION); add_phi_arg (new_phi, build_zero_cst (TREE_TYPE (niters)), skip_e, UNKNOWN_LOCATION); niters = PHI_RESULT (new_phi); epilogue_vinfo->main_loop_edge = update_e; epilogue_vinfo->skip_main_loop_edge = skip_e; } /* Set ADVANCE to the number of iterations performed by the previous loop and its prologue. */ *advance = niters; /* Subtract the number of iterations performed by the vectorized loop from the number of total iterations. */ tree epilogue_niters = fold_build2 (MINUS_EXPR, TREE_TYPE (niters), before_loop_niters, niters); LOOP_VINFO_NITERS (epilogue_vinfo) = epilogue_niters; LOOP_VINFO_NITERSM1 (epilogue_vinfo) = fold_build2 (MINUS_EXPR, TREE_TYPE (epilogue_niters), epilogue_niters, build_one_cst (TREE_TYPE (epilogue_niters))); /* Decide what to do if the number of epilogue iterations is not a multiple of the epilogue loop's vectorization factor. We should have rejected the loop during the analysis phase if this fails. */ bool res = vect_determine_partial_vectors_and_peeling (epilogue_vinfo); gcc_assert (res); } adjust_vec.release (); free_original_copy_tables (); return vect_epilogues ? epilog : NULL; } /* Function vect_create_cond_for_niters_checks. Create a conditional expression that represents the run-time checks for loop's niter. The loop is guaranteed to terminate if the run-time checks hold. Input: COND_EXPR - input conditional expression. New conditions will be chained with logical AND operation. If it is NULL, then the function is used to return the number of alias checks. LOOP_VINFO - field LOOP_VINFO_MAY_ALIAS_STMTS contains the list of ddrs to be checked. Output: COND_EXPR - conditional expression. The returned COND_EXPR 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_niters_checks (loop_vec_info loop_vinfo, tree *cond_expr) { tree part_cond_expr = LOOP_VINFO_NITERS_ASSUMPTIONS (loop_vinfo); if (*cond_expr) *cond_expr = fold_build2 (TRUTH_AND_EXPR, boolean_type_node, *cond_expr, part_cond_expr); else *cond_expr = part_cond_expr; } /* Set *COND_EXPR to a tree that is true when both the original *COND_EXPR and PART_COND_EXPR are true. Treat a null *COND_EXPR as "true". */ static void chain_cond_expr (tree *cond_expr, tree part_cond_expr) { 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_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) { const vec &may_misalign_stmts = LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo); stmt_vec_info stmt_info; int mask = LOOP_VINFO_PTR_MASK (loop_vinfo); tree mask_cst; unsigned int i; tree int_ptrsize_type; char tmp_name[20]; tree or_tmp_name = NULL_TREE; tree 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)); int_ptrsize_type = signed_type_for (ptr_type_node); /* Create expression (mask & (dr_1 || ... || dr_n)) where dr_i is the address of the first vector of the i'th data reference. */ FOR_EACH_VEC_ELT (may_misalign_stmts, i, stmt_info) { gimple_seq new_stmt_list = NULL; tree addr_base; tree addr_tmp_name; tree new_or_tmp_name; gimple *addr_stmt, *or_stmt; tree vectype = STMT_VINFO_VECTYPE (stmt_info); bool negative = tree_int_cst_compare (DR_STEP (STMT_VINFO_DATA_REF (stmt_info)), size_zero_node) < 0; tree offset = negative ? size_int ((-TYPE_VECTOR_SUBPARTS (vectype) + 1) * TREE_INT_CST_LOW (TYPE_SIZE_UNIT (TREE_TYPE (vectype)))) : size_zero_node; /* create: addr_tmp = (int)(address_of_first_vector) */ addr_base = vect_create_addr_base_for_vector_ref (loop_vinfo, stmt_info, &new_stmt_list, offset); if (new_stmt_list != NULL) gimple_seq_add_seq (cond_expr_stmt_list, new_stmt_list); sprintf (tmp_name, "addr2int%d", i); addr_tmp_name = make_temp_ssa_name (int_ptrsize_type, NULL, tmp_name); addr_stmt = gimple_build_assign (addr_tmp_name, NOP_EXPR, addr_base); 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, "orptrs%d", i); new_or_tmp_name = make_temp_ssa_name (int_ptrsize_type, NULL, tmp_name); or_stmt = gimple_build_assign (new_or_tmp_name, BIT_IOR_EXPR, or_tmp_name, addr_tmp_name); 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_name = make_temp_ssa_name (int_ptrsize_type, NULL, "andmask"); and_stmt = gimple_build_assign (and_tmp_name, BIT_AND_EXPR, or_tmp_name, mask_cst); 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); chain_cond_expr (cond_expr, part_cond_expr); } /* If LOOP_VINFO_CHECK_UNEQUAL_ADDRS contains , ..., , create a tree representation of: (&A1 != &B1) && ... && (&An != &Bn). Set *COND_EXPR to a tree that is true when both the original *COND_EXPR and this new condition are true. Treat a null *COND_EXPR as "true". */ static void vect_create_cond_for_unequal_addrs (loop_vec_info loop_vinfo, tree *cond_expr) { const vec &pairs = LOOP_VINFO_CHECK_UNEQUAL_ADDRS (loop_vinfo); unsigned int i; vec_object_pair *pair; FOR_EACH_VEC_ELT (pairs, i, pair) { tree addr1 = build_fold_addr_expr (pair->first); tree addr2 = build_fold_addr_expr (pair->second); tree part_cond_expr = fold_build2 (NE_EXPR, boolean_type_node, addr1, addr2); chain_cond_expr (cond_expr, part_cond_expr); } } /* Create an expression that is true when all lower-bound conditions for the vectorized loop are met. Chain this condition with *COND_EXPR. */ static void vect_create_cond_for_lower_bounds (loop_vec_info loop_vinfo, tree *cond_expr) { const vec &lower_bounds = LOOP_VINFO_LOWER_BOUNDS (loop_vinfo); for (unsigned int i = 0; i < lower_bounds.length (); ++i) { tree expr = lower_bounds[i].expr; tree type = unsigned_type_for (TREE_TYPE (expr)); expr = fold_convert (type, expr); poly_uint64 bound = lower_bounds[i].min_value; if (!lower_bounds[i].unsigned_p) { expr = fold_build2 (PLUS_EXPR, type, expr, build_int_cstu (type, bound - 1)); bound += bound - 1; } tree part_cond_expr = fold_build2 (GE_EXPR, boolean_type_node, expr, build_int_cstu (type, bound)); chain_cond_expr (cond_expr, part_cond_expr); } } /* 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. If it is NULL, then the function is used to return the number of alias checks. LOOP_VINFO - field LOOP_VINFO_MAY_ALIAS_STMTS contains the list of ddrs to be checked. Output: COND_EXPR - conditional expression. The returned COND_EXPR is the conditional expression to be used in the if statement that controls which version of the loop gets executed at runtime. */ void vect_create_cond_for_alias_checks (loop_vec_info loop_vinfo, tree * cond_expr) { const vec &comp_alias_ddrs = LOOP_VINFO_COMP_ALIAS_DDRS (loop_vinfo); if (comp_alias_ddrs.is_empty ()) return; create_runtime_alias_checks (LOOP_VINFO_LOOP (loop_vinfo), &comp_alias_ddrs, cond_expr); if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "created %u versioning for alias checks.\n", comp_alias_ddrs.length ()); } /* 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 threshold TH. The versioning precondition(s) are placed in *COND_EXPR and *COND_EXPR_STMT_LIST. */ class loop * vect_loop_versioning (loop_vec_info loop_vinfo, gimple *loop_vectorized_call) { class loop *loop = LOOP_VINFO_LOOP (loop_vinfo), *nloop; class loop *scalar_loop = LOOP_VINFO_SCALAR_LOOP (loop_vinfo); basic_block condition_bb; gphi_iterator gsi; gimple_stmt_iterator cond_exp_gsi; basic_block merge_bb; basic_block new_exit_bb; edge new_exit_e, e; gphi *orig_phi, *new_phi; tree cond_expr = NULL_TREE; gimple_seq cond_expr_stmt_list = NULL; tree arg; profile_probability prob = profile_probability::likely (); gimple_seq gimplify_stmt_list = NULL; tree scalar_loop_iters = LOOP_VINFO_NITERSM1 (loop_vinfo); bool version_align = LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT (loop_vinfo); bool version_alias = LOOP_REQUIRES_VERSIONING_FOR_ALIAS (loop_vinfo); bool version_niter = LOOP_REQUIRES_VERSIONING_FOR_NITERS (loop_vinfo); poly_uint64 versioning_threshold = LOOP_VINFO_VERSIONING_THRESHOLD (loop_vinfo); tree version_simd_if_cond = LOOP_REQUIRES_VERSIONING_FOR_SIMD_IF_COND (loop_vinfo); unsigned th = LOOP_VINFO_COST_MODEL_THRESHOLD (loop_vinfo); if (vect_apply_runtime_profitability_check_p (loop_vinfo) && !ordered_p (th, versioning_threshold)) cond_expr = fold_build2 (GE_EXPR, boolean_type_node, scalar_loop_iters, build_int_cst (TREE_TYPE (scalar_loop_iters), th - 1)); if (maybe_ne (versioning_threshold, 0U)) { tree expr = fold_build2 (GE_EXPR, boolean_type_node, scalar_loop_iters, build_int_cst (TREE_TYPE (scalar_loop_iters), versioning_threshold - 1)); if (cond_expr) cond_expr = fold_build2 (BIT_AND_EXPR, boolean_type_node, expr, cond_expr); else cond_expr = expr; } tree cost_name = NULL_TREE; profile_probability prob2 = profile_probability::always (); if (cond_expr && EXPR_P (cond_expr) && (version_niter || version_align || version_alias || version_simd_if_cond)) { cost_name = cond_expr = force_gimple_operand_1 (unshare_expr (cond_expr), &cond_expr_stmt_list, is_gimple_val, NULL_TREE); /* Split prob () into two so that the overall probability of passing both the cost-model and versioning checks is the orig prob. */ prob2 = prob = prob.sqrt (); } if (version_niter) vect_create_cond_for_niters_checks (loop_vinfo, &cond_expr); if (cond_expr) { gimple_seq tem = NULL; cond_expr = force_gimple_operand_1 (unshare_expr (cond_expr), &tem, is_gimple_condexpr_for_cond, NULL_TREE); gimple_seq_add_seq (&cond_expr_stmt_list, tem); } if (version_align) vect_create_cond_for_align_checks (loop_vinfo, &cond_expr, &cond_expr_stmt_list); if (version_alias) { vect_create_cond_for_unequal_addrs (loop_vinfo, &cond_expr); vect_create_cond_for_lower_bounds (loop_vinfo, &cond_expr); vect_create_cond_for_alias_checks (loop_vinfo, &cond_expr); } if (version_simd_if_cond) { gcc_assert (dom_info_available_p (CDI_DOMINATORS)); if (flag_checking) if (basic_block bb = gimple_bb (SSA_NAME_DEF_STMT (version_simd_if_cond))) gcc_assert (bb != loop->header && dominated_by_p (CDI_DOMINATORS, loop->header, bb) && (scalar_loop == NULL || (bb != scalar_loop->header && dominated_by_p (CDI_DOMINATORS, scalar_loop->header, bb)))); tree zero = build_zero_cst (TREE_TYPE (version_simd_if_cond)); tree c = fold_build2 (NE_EXPR, boolean_type_node, version_simd_if_cond, zero); if (cond_expr) cond_expr = fold_build2 (TRUTH_AND_EXPR, boolean_type_node, c, cond_expr); else cond_expr = c; if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "created versioning for simd if condition check.\n"); } cond_expr = force_gimple_operand_1 (unshare_expr (cond_expr), &gimplify_stmt_list, is_gimple_condexpr_for_cond, NULL_TREE); gimple_seq_add_seq (&cond_expr_stmt_list, gimplify_stmt_list); /* Compute the outermost loop cond_expr and cond_expr_stmt_list are invariant in. */ class loop *outermost = outermost_invariant_loop_for_expr (loop, cond_expr); for (gimple_stmt_iterator gsi = gsi_start (cond_expr_stmt_list); !gsi_end_p (gsi); gsi_next (&gsi)) { gimple *stmt = gsi_stmt (gsi); update_stmt (stmt); ssa_op_iter iter; use_operand_p use_p; basic_block def_bb; FOR_EACH_SSA_USE_OPERAND (use_p, stmt, iter, SSA_OP_USE) if ((def_bb = gimple_bb (SSA_NAME_DEF_STMT (USE_FROM_PTR (use_p)))) && flow_bb_inside_loop_p (outermost, def_bb)) outermost = superloop_at_depth (loop, bb_loop_depth (def_bb) + 1); } /* Search for the outermost loop we can version. Avoid versioning of non-perfect nests but allow if-conversion versioned loops inside. */ class loop *loop_to_version = loop; if (flow_loop_nested_p (outermost, loop)) { if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "trying to apply versioning to outer loop %d\n", outermost->num); if (outermost->num == 0) outermost = superloop_at_depth (loop, 1); /* And avoid applying versioning on non-perfect nests. */ while (loop_to_version != outermost && (e = single_exit (loop_outer (loop_to_version))) && !(e->flags & EDGE_COMPLEX) && (!loop_outer (loop_to_version)->inner->next || vect_loop_vectorized_call (loop_to_version)) && (!loop_outer (loop_to_version)->inner->next || !loop_outer (loop_to_version)->inner->next->next)) loop_to_version = loop_outer (loop_to_version); } /* Apply versioning. If there is already a scalar version created by if-conversion re-use that. Note we cannot re-use the copy of an if-converted outer-loop when vectorizing the inner loop only. */ gcond *cond; if ((!loop_to_version->inner || loop == loop_to_version) && loop_vectorized_call) { gcc_assert (scalar_loop); condition_bb = gimple_bb (loop_vectorized_call); cond = as_a (*gsi_last_bb (condition_bb)); gimple_cond_set_condition_from_tree (cond, cond_expr); update_stmt (cond); if (cond_expr_stmt_list) { cond_exp_gsi = gsi_for_stmt (loop_vectorized_call); gsi_insert_seq_before (&cond_exp_gsi, cond_expr_stmt_list, GSI_SAME_STMT); } /* if-conversion uses profile_probability::always () for both paths, reset the paths probabilities appropriately. */ edge te, fe; extract_true_false_edges_from_block (condition_bb, &te, &fe); te->probability = prob; fe->probability = prob.invert (); /* We can scale loops counts immediately but have to postpone scaling the scalar loop because we re-use it during peeling. Ifcvt duplicates loop preheader, loop body and produces an basic block after loop exit. We need to scale all that. */ basic_block preheader = loop_preheader_edge (loop_to_version)->src; preheader->count = preheader->count.apply_probability (prob * prob2); scale_loop_frequencies (loop_to_version, prob * prob2); single_exit (loop_to_version)->dest->count = preheader->count; LOOP_VINFO_SCALAR_LOOP_SCALING (loop_vinfo) = (prob * prob2).invert (); nloop = scalar_loop; if (dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "reusing %sloop version created by if conversion\n", loop_to_version != loop ? "outer " : ""); } else { if (loop_to_version != loop && dump_enabled_p ()) dump_printf_loc (MSG_NOTE, vect_location, "applying loop versioning to outer loop %d\n", loop_to_version->num); unsigned orig_pe_idx = loop_preheader_edge (loop)->dest_idx; initialize_original_copy_tables (); nloop = loop_version (loop_to_version, cond_expr, &condition_bb, prob * prob2, (prob * prob2).invert (), prob * prob2, (prob * prob2).invert (), true); /* We will later insert second conditional so overall outcome of both is prob * prob2. */ edge true_e, false_e; extract_true_false_edges_from_block (condition_bb, &true_e, &false_e); true_e->probability = prob; false_e->probability = prob.invert (); gcc_assert (nloop); nloop = get_loop_copy (loop); /* For cycle vectorization with SLP we rely on the PHI arguments appearing in the same order as the SLP node operands which for the loop PHI nodes means the preheader edge dest index needs to remain the same for the analyzed loop which also becomes the vectorized one. Make it so in case the state after versioning differs by redirecting the first edge into the header to the same destination which moves it last. */ if (loop_preheader_edge (loop)->dest_idx != orig_pe_idx) { edge e = EDGE_PRED (loop->header, 0); ssa_redirect_edge (e, e->dest); flush_pending_stmts (e); } gcc_assert (loop_preheader_edge (loop)->dest_idx == orig_pe_idx); /* Kill off IFN_LOOP_VECTORIZED_CALL in the copy, nobody will reap those otherwise; they also refer to the original loops. */ class loop *l = loop; while (gimple *call = vect_loop_vectorized_call (l)) { call = SSA_NAME_DEF_STMT (get_current_def (gimple_call_lhs (call))); fold_loop_internal_call (call, boolean_false_node); l = loop_outer (l); } free_original_copy_tables (); 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); } /* 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. If loop versioning wasn't done from loop, but scalar_loop instead, merge_bb will have already just a single successor. */ merge_bb = single_exit (loop_to_version)->dest; if (EDGE_COUNT (merge_bb->preds) >= 2) { gcc_assert (EDGE_COUNT (merge_bb->preds) >= 2); new_exit_bb = split_edge (single_exit (loop_to_version)); new_exit_e = single_exit (loop_to_version); e = EDGE_SUCC (new_exit_bb, 0); for (gsi = gsi_start_phis (merge_bb); !gsi_end_p (gsi); gsi_next (&gsi)) { tree new_res; orig_phi = gsi.phi (); new_res = copy_ssa_name (PHI_RESULT (orig_phi)); new_phi = create_phi_node (new_res, 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)); adjust_phi_and_debug_stmts (orig_phi, e, PHI_RESULT (new_phi)); } } update_ssa (TODO_update_ssa_no_phi); } /* Split the cost model check off to a separate BB. Costing assumes this is the only thing we perform when we enter the scalar loop from a failed cost decision. */ if (cost_name && TREE_CODE (cost_name) == SSA_NAME) { gimple *def = SSA_NAME_DEF_STMT (cost_name); gcc_assert (gimple_bb (def) == condition_bb); /* All uses of the cost check are 'true' after the check we are going to insert. */ replace_uses_by (cost_name, boolean_true_node); /* And we're going to build the new single use of it. */ gcond *cond = gimple_build_cond (NE_EXPR, cost_name, boolean_false_node, NULL_TREE, NULL_TREE); edge e = split_block (gimple_bb (def), def); gimple_stmt_iterator gsi = gsi_for_stmt (def); gsi_insert_after (&gsi, cond, GSI_NEW_STMT); edge true_e, false_e; extract_true_false_edges_from_block (e->dest, &true_e, &false_e); e->flags &= ~EDGE_FALLTHRU; e->flags |= EDGE_TRUE_VALUE; edge e2 = make_edge (e->src, false_e->dest, EDGE_FALSE_VALUE); e->probability = prob2; e2->probability = prob2.invert (); e->dest->count = e->count (); set_immediate_dominator (CDI_DOMINATORS, false_e->dest, e->src); auto_vec adj; for (basic_block son = first_dom_son (CDI_DOMINATORS, e->dest); son; son = next_dom_son (CDI_DOMINATORS, son)) if (EDGE_COUNT (son->preds) > 1) adj.safe_push (son); for (auto son : adj) set_immediate_dominator (CDI_DOMINATORS, son, e->src); //debug_bb (condition_bb); //debug_bb (e->src); } if (version_niter) { /* The versioned loop could be infinite, we need to clear existing niter information which is copied from the original loop. */ gcc_assert (loop_constraint_set_p (loop, LOOP_C_FINITE)); vect_free_loop_info_assumptions (nloop); } if (LOCATION_LOCUS (vect_location.get_location_t ()) != UNKNOWN_LOCATION && dump_enabled_p ()) { if (version_alias) dump_printf_loc (MSG_OPTIMIZED_LOCATIONS | MSG_PRIORITY_USER_FACING, vect_location, "loop versioned for vectorization because of " "possible aliasing\n"); if (version_align) dump_printf_loc (MSG_OPTIMIZED_LOCATIONS | MSG_PRIORITY_USER_FACING, vect_location, "loop versioned for vectorization to enhance " "alignment\n"); } return nloop; }