X-Git-Url: http://git.sourceforge.jp/view?p=pf3gnuchains%2Fgcc-fork.git;a=blobdiff_plain;f=gcc%2Ftree-vect-patterns.c;h=c3f24a186f027852adf3f052e0fe47f82a08e43b;hp=0d78882751338fb5abc992e338285382e78e1395;hb=760ff94e3788e67aab56bcc7b3dd3507ddda6aca;hpb=97594f7be7ee4321d26f44ef2a28849721947df6 diff --git a/gcc/tree-vect-patterns.c b/gcc/tree-vect-patterns.c index 0d788827513..c3f24a186f0 100644 --- a/gcc/tree-vect-patterns.c +++ b/gcc/tree-vect-patterns.c @@ -1,5 +1,5 @@ /* Analysis Utilities for Loop Vectorization. - Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 + Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011, 2012 Free Software Foundation, Inc. Contributed by Dorit Nuzman @@ -49,13 +49,66 @@ static gimple vect_recog_dot_prod_pattern (VEC (gimple, heap) **, tree *, static gimple vect_recog_pow_pattern (VEC (gimple, heap) **, tree *, tree *); static gimple vect_recog_over_widening_pattern (VEC (gimple, heap) **, tree *, tree *); +static gimple vect_recog_widen_shift_pattern (VEC (gimple, heap) **, + tree *, tree *); +static gimple vect_recog_vector_vector_shift_pattern (VEC (gimple, heap) **, + tree *, tree *); +static gimple vect_recog_sdivmod_pow2_pattern (VEC (gimple, heap) **, + tree *, tree *); +static gimple vect_recog_mixed_size_cond_pattern (VEC (gimple, heap) **, + tree *, tree *); +static gimple vect_recog_bool_pattern (VEC (gimple, heap) **, tree *, tree *); static vect_recog_func_ptr vect_vect_recog_func_ptrs[NUM_PATTERNS] = { vect_recog_widen_mult_pattern, vect_recog_widen_sum_pattern, vect_recog_dot_prod_pattern, vect_recog_pow_pattern, - vect_recog_over_widening_pattern}; + vect_recog_widen_shift_pattern, + vect_recog_over_widening_pattern, + vect_recog_vector_vector_shift_pattern, + vect_recog_sdivmod_pow2_pattern, + vect_recog_mixed_size_cond_pattern, + vect_recog_bool_pattern}; +static inline void +append_pattern_def_seq (stmt_vec_info stmt_info, gimple stmt) +{ + gimple_seq_add_stmt_without_update (&STMT_VINFO_PATTERN_DEF_SEQ (stmt_info), + stmt); +} + +static inline void +new_pattern_def_seq (stmt_vec_info stmt_info, gimple stmt) +{ + STMT_VINFO_PATTERN_DEF_SEQ (stmt_info) = NULL; + append_pattern_def_seq (stmt_info, stmt); +} + +/* If the LHS of DEF_STMT has a single use, and that statement is + in the same loop, return it. */ + +static gimple +vect_single_imm_use (gimple def_stmt) +{ + stmt_vec_info stmt_vinfo = vinfo_for_stmt (def_stmt); + loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); + struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); + tree lhs = gimple_assign_lhs (def_stmt); + use_operand_p use_p; + gimple use_stmt; + + if (!single_imm_use (lhs, &use_p, &use_stmt)) + return NULL; + + if (!gimple_bb (use_stmt)) + return NULL; + + if (!flow_bb_inside_loop_p (loop, gimple_bb (use_stmt))) + return NULL; + + gcc_assert (vinfo_for_stmt (use_stmt)); + return use_stmt; +} /* Function widened_name_p @@ -82,7 +135,8 @@ widened_name_p (tree name, gimple use_stmt, tree *half_type, gimple *def_stmt, stmt_vinfo = vinfo_for_stmt (use_stmt); loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); - if (!vect_is_simple_use (name, loop_vinfo, NULL, def_stmt, &def, &dt)) + if (!vect_is_simple_use (name, use_stmt, loop_vinfo, NULL, def_stmt, &def, + &dt)) return false; if (dt != vect_internal_def @@ -106,8 +160,8 @@ widened_name_p (tree name, gimple use_stmt, tree *half_type, gimple *def_stmt, || (TYPE_PRECISION (type) < (TYPE_PRECISION (*half_type) * 2))) return false; - if (!vect_is_simple_use (oprnd0, loop_vinfo, NULL, &dummy_gimple, &dummy, - &dt)) + if (!vect_is_simple_use (oprnd0, *def_stmt, loop_vinfo, + NULL, &dummy_gimple, &dummy, &dt)) return false; return true; @@ -256,6 +310,8 @@ vect_recog_dot_prod_pattern (VEC (gimple, heap) **stmts, tree *type_in, we know that oprnd1 is the reduction variable (defined by a loop-header phi), and oprnd0 is an ssa-name defined by a stmt in the loop body. Left to check that oprnd0 is defined by a (widen_)mult_expr */ + if (TREE_CODE (oprnd0) != SSA_NAME) + return NULL; prod_type = half_type; stmt = SSA_NAME_DEF_STMT (oprnd0); @@ -333,27 +389,37 @@ vect_recog_dot_prod_pattern (VEC (gimple, heap) **stmts, tree *type_in, } -/* Handle two cases of multiplication by a constant. The first one is when - the constant, CONST_OPRND, fits the type (HALF_TYPE) of the second - operand (OPRND). In that case, we can peform widen-mult from HALF_TYPE to - TYPE. +/* Handle widening operation by a constant. At the moment we support MULT_EXPR + and LSHIFT_EXPR. + + For MULT_EXPR we check that CONST_OPRND fits HALF_TYPE, and for LSHIFT_EXPR + we check that CONST_OPRND is less or equal to the size of HALF_TYPE. Otherwise, if the type of the result (TYPE) is at least 4 times bigger than - HALF_TYPE, and CONST_OPRND fits an intermediate type (2 times smaller than - TYPE), we can perform widen-mult from the intermediate type to TYPE and - replace a_T = (TYPE) a_t; with a_it - (interm_type) a_t; */ + HALF_TYPE, and there is an intermediate type (2 times smaller than TYPE) + that satisfies the above restrictions, we can perform a widening opeartion + from the intermediate type to TYPE and replace a_T = (TYPE) a_t; + with a_it = (interm_type) a_t; */ static bool -vect_handle_widen_mult_by_const (gimple stmt, tree const_oprnd, tree *oprnd, - VEC (gimple, heap) **stmts, tree type, - tree *half_type, gimple def_stmt) +vect_handle_widen_op_by_const (gimple stmt, enum tree_code code, + tree const_oprnd, tree *oprnd, + VEC (gimple, heap) **stmts, tree type, + tree *half_type, gimple def_stmt) { tree new_type, new_oprnd, tmp; gimple new_stmt; loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (vinfo_for_stmt (stmt)); struct loop *loop = LOOP_VINFO_LOOP (loop_info); - if (int_fits_type_p (const_oprnd, *half_type)) + if (code != MULT_EXPR && code != LSHIFT_EXPR) + return false; + + if (((code == MULT_EXPR && int_fits_type_p (const_oprnd, *half_type)) + || (code == LSHIFT_EXPR + && compare_tree_int (const_oprnd, TYPE_PRECISION (*half_type)) + != 1)) + && TYPE_PRECISION (type) == (TYPE_PRECISION (*half_type) * 2)) { /* CONST_OPRND is a constant of HALF_TYPE. */ *oprnd = gimple_assign_rhs1 (def_stmt); @@ -366,14 +432,16 @@ vect_handle_widen_mult_by_const (gimple stmt, tree const_oprnd, tree *oprnd, || !vinfo_for_stmt (def_stmt)) return false; - /* TYPE is 4 times bigger than HALF_TYPE, try widen-mult for + /* TYPE is 4 times bigger than HALF_TYPE, try widening operation for a type 2 times bigger than HALF_TYPE. */ new_type = build_nonstandard_integer_type (TYPE_PRECISION (type) / 2, TYPE_UNSIGNED (type)); - if (!int_fits_type_p (const_oprnd, new_type)) + if ((code == MULT_EXPR && !int_fits_type_p (const_oprnd, new_type)) + || (code == LSHIFT_EXPR + && compare_tree_int (const_oprnd, TYPE_PRECISION (new_type)) == 1)) return false; - /* Use NEW_TYPE for widen_mult. */ + /* Use NEW_TYPE for widening operation. */ if (STMT_VINFO_RELATED_STMT (vinfo_for_stmt (def_stmt))) { new_stmt = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (def_stmt)); @@ -383,6 +451,7 @@ vect_handle_widen_mult_by_const (gimple stmt, tree const_oprnd, tree *oprnd, || TREE_TYPE (gimple_assign_lhs (new_stmt)) != new_type) return false; + VEC_safe_push (gimple, heap, *stmts, def_stmt); *oprnd = gimple_assign_lhs (new_stmt); } else @@ -394,7 +463,6 @@ vect_handle_widen_mult_by_const (gimple stmt, tree const_oprnd, tree *oprnd, new_oprnd = make_ssa_name (tmp, NULL); new_stmt = gimple_build_assign_with_ops (NOP_EXPR, new_oprnd, *oprnd, NULL_TREE); - SSA_NAME_DEF_STMT (new_oprnd) = new_stmt; STMT_VINFO_RELATED_STMT (vinfo_for_stmt (def_stmt)) = new_stmt; VEC_safe_push (gimple, heap, *stmts, def_stmt); *oprnd = new_oprnd; @@ -493,7 +561,7 @@ vect_recog_widen_mult_pattern (VEC (gimple, heap) **stmts, enum tree_code dummy_code; int dummy_int; VEC (tree, heap) *dummy_vec; - bool op0_ok, op1_ok; + bool op1_ok; if (!is_gimple_assign (last_stmt)) return NULL; @@ -513,38 +581,23 @@ vect_recog_widen_mult_pattern (VEC (gimple, heap) **stmts, return NULL; /* Check argument 0. */ - op0_ok = widened_name_p (oprnd0, last_stmt, &half_type0, &def_stmt0, false); + if (!widened_name_p (oprnd0, last_stmt, &half_type0, &def_stmt0, false)) + return NULL; /* Check argument 1. */ op1_ok = widened_name_p (oprnd1, last_stmt, &half_type1, &def_stmt1, false); - /* In case of multiplication by a constant one of the operands may not match - the pattern, but not both. */ - if (!op0_ok && !op1_ok) - return NULL; - - if (op0_ok && op1_ok) + if (op1_ok) { oprnd0 = gimple_assign_rhs1 (def_stmt0); oprnd1 = gimple_assign_rhs1 (def_stmt1); } - else if (!op0_ok) - { - if (TREE_CODE (oprnd0) == INTEGER_CST - && TREE_CODE (half_type1) == INTEGER_TYPE - && vect_handle_widen_mult_by_const (last_stmt, oprnd0, &oprnd1, - stmts, type, - &half_type1, def_stmt1)) - half_type0 = half_type1; - else - return NULL; - } - else if (!op1_ok) + else { if (TREE_CODE (oprnd1) == INTEGER_CST && TREE_CODE (half_type0) == INTEGER_TYPE - && vect_handle_widen_mult_by_const (last_stmt, oprnd1, &oprnd0, - stmts, type, - &half_type0, def_stmt0)) + && vect_handle_widen_op_by_const (last_stmt, MULT_EXPR, oprnd1, + &oprnd0, stmts, type, + &half_type0, def_stmt0)) half_type1 = half_type0; else return NULL; @@ -555,25 +608,15 @@ vect_recog_widen_mult_pattern (VEC (gimple, heap) **stmts, Use unsigned TYPE as the type for WIDEN_MULT_EXPR. */ if (TYPE_UNSIGNED (type) != TYPE_UNSIGNED (half_type0)) { - tree lhs = gimple_assign_lhs (last_stmt), use_lhs; - imm_use_iterator imm_iter; - use_operand_p use_p; - int nuses = 0; - gimple use_stmt = NULL; + gimple use_stmt; + tree use_lhs; tree use_type; if (TYPE_UNSIGNED (type) == TYPE_UNSIGNED (half_type1)) return NULL; - FOR_EACH_IMM_USE_FAST (use_p, imm_iter, lhs) - { - if (is_gimple_debug (USE_STMT (use_p))) - continue; - use_stmt = USE_STMT (use_p); - nuses++; - } - - if (nuses != 1 || !is_gimple_assign (use_stmt) + use_stmt = vect_single_imm_use (last_stmt); + if (!use_stmt || !is_gimple_assign (use_stmt) || gimple_assign_rhs_code (use_stmt) != NOP_EXPR) return NULL; @@ -613,7 +656,6 @@ vect_recog_widen_mult_pattern (VEC (gimple, heap) **stmts, var = vect_recog_temp_ssa_var (type, NULL); pattern_stmt = gimple_build_assign_with_ops (WIDEN_MULT_EXPR, var, oprnd0, oprnd1); - SSA_NAME_DEF_STMT (var) = pattern_stmt; if (vect_print_dump_info (REPORT_DETAILS)) print_gimple_stmt (vect_dump, pattern_stmt, 0, TDF_SLIM); @@ -697,7 +739,6 @@ vect_recog_pow_pattern (VEC (gimple, heap) **stmts, tree *type_in, var = vect_recog_temp_ssa_var (TREE_TYPE (base), NULL); stmt = gimple_build_assign_with_ops (MULT_EXPR, var, base, base); - SSA_NAME_DEF_STMT (var) = stmt; return stmt; } @@ -820,7 +861,6 @@ vect_recog_widen_sum_pattern (VEC (gimple, heap) **stmts, tree *type_in, var = vect_recog_temp_ssa_var (type, NULL); pattern_stmt = gimple_build_assign_with_ops (WIDEN_SUM_EXPR, var, oprnd0, oprnd1); - SSA_NAME_DEF_STMT (var) = pattern_stmt; if (vect_print_dump_info (REPORT_DETAILS)) { @@ -861,7 +901,7 @@ vect_recog_widen_sum_pattern (VEC (gimple, heap) **stmts, tree *type_in, NEW_DEF_STMT - in case DEF has to be promoted, we create two pattern statements for STMT: the first one is a type promotion and the second one is the operation itself. We return the type promotion statement - in NEW_DEF_STMT and further store it in STMT_VINFO_PATTERN_DEF_STMT of + in NEW_DEF_STMT and further store it in STMT_VINFO_PATTERN_DEF_SEQ of the second pattern statement. */ static bool @@ -877,6 +917,8 @@ vect_operation_fits_smaller_type (gimple stmt, tree def, tree *new_type, loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (vinfo_for_stmt (stmt)); struct loop *loop = LOOP_VINFO_LOOP (loop_info); + *op0 = NULL_TREE; + *op1 = NULL_TREE; *new_def_stmt = NULL; if (!is_gimple_assign (stmt)) @@ -895,6 +937,11 @@ vect_operation_fits_smaller_type (gimple stmt, tree def, tree *new_type, || TREE_CODE (const_oprnd) != INTEGER_CST) return false; + /* If oprnd has other uses besides that in stmt we cannot mark it + as being part of a pattern only. */ + if (!has_single_use (oprnd)) + return false; + /* If we are in the middle of a sequence, we use DEF from a previous statement. Otherwise, OPRND has to be a result of type promotion. */ if (*new_type) @@ -982,7 +1029,7 @@ vect_operation_fits_smaller_type (gimple stmt, tree def, tree *new_type, a. Its type is not sufficient for the operation, we create a new stmt: a type conversion for OPRND from HALF_TYPE to INTERM_TYPE. We store this statement in NEW_DEF_STMT, and it is later put in - STMT_VINFO_PATTERN_DEF_STMT of the pattern statement for STMT. + STMT_VINFO_PATTERN_DEF_SEQ of the pattern statement for STMT. b. OPRND is good to use in the new statement. */ if (first) { @@ -999,6 +1046,7 @@ vect_operation_fits_smaller_type (gimple stmt, tree def, tree *new_type, || TREE_TYPE (gimple_assign_lhs (new_stmt)) != interm_type) return false; + VEC_safe_push (gimple, heap, *stmts, def_stmt); oprnd = gimple_assign_lhs (new_stmt); } else @@ -1010,7 +1058,6 @@ vect_operation_fits_smaller_type (gimple stmt, tree def, tree *new_type, new_oprnd = make_ssa_name (tmp, NULL); new_stmt = gimple_build_assign_with_ops (NOP_EXPR, new_oprnd, oprnd, NULL_TREE); - SSA_NAME_DEF_STMT (new_oprnd) = new_stmt; STMT_VINFO_RELATED_STMT (vinfo_for_stmt (def_stmt)) = new_stmt; VEC_safe_push (gimple, heap, *stmts, def_stmt); oprnd = new_oprnd; @@ -1032,7 +1079,6 @@ vect_operation_fits_smaller_type (gimple stmt, tree def, tree *new_type, new_oprnd = make_ssa_name (tmp, NULL); new_stmt = gimple_build_assign_with_ops (NOP_EXPR, new_oprnd, oprnd, NULL_TREE); - SSA_NAME_DEF_STMT (new_oprnd) = new_stmt; oprnd = new_oprnd; *new_def_stmt = new_stmt; } @@ -1068,23 +1114,18 @@ vect_operation_fits_smaller_type (gimple stmt, tree def, tree *new_type, constants. Check if S3 and S4 can be done on a smaller type than 'TYPE', it can either be 'type' or some intermediate type. For now, we expect S5 to be a type - demotion operation. We also check that S3 and S4 have only one use. -. + demotion operation. We also check that S3 and S4 have only one use. */ -*/ static gimple vect_recog_over_widening_pattern (VEC (gimple, heap) **stmts, tree *type_in, tree *type_out) { gimple stmt = VEC_pop (gimple, *stmts); gimple pattern_stmt = NULL, new_def_stmt, prev_stmt = NULL, use_stmt = NULL; - tree op0, op1, vectype = NULL_TREE, lhs, use_lhs, use_type; - imm_use_iterator imm_iter; - use_operand_p use_p; - int nuses = 0; + tree op0, op1, vectype = NULL_TREE, use_lhs, use_type; tree var = NULL_TREE, new_type = NULL_TREE, tmp, new_oprnd; bool first; - struct loop *loop = (gimple_bb (stmt))->loop_father; + tree type = NULL; first = true; while (1) @@ -1105,19 +1146,8 @@ vect_recog_over_widening_pattern (VEC (gimple, heap) **stmts, } /* STMT can be performed on a smaller type. Check its uses. */ - lhs = gimple_assign_lhs (stmt); - nuses = 0; - FOR_EACH_IMM_USE_FAST (use_p, imm_iter, lhs) - { - if (is_gimple_debug (USE_STMT (use_p))) - continue; - use_stmt = USE_STMT (use_p); - nuses++; - } - - if (nuses != 1 || !is_gimple_assign (use_stmt) - || !gimple_bb (use_stmt) - || !flow_bb_inside_loop_p (loop, gimple_bb (use_stmt))) + use_stmt = vect_single_imm_use (stmt); + if (!use_stmt || !is_gimple_assign (use_stmt)) return NULL; /* Create pattern statement for STMT. */ @@ -1129,17 +1159,17 @@ vect_recog_over_widening_pattern (VEC (gimple, heap) **stmts, statetments, except for the case when the last statement in the sequence doesn't have a corresponding pattern statement. In such case we associate the last pattern statement with the last statement - in the sequence. Therefore, we only add an original statetement to + in the sequence. Therefore, we only add the original statement to the list if we know that it is not the last. */ if (prev_stmt) VEC_safe_push (gimple, heap, *stmts, prev_stmt); var = vect_recog_temp_ssa_var (new_type, NULL); - pattern_stmt = gimple_build_assign_with_ops ( - gimple_assign_rhs_code (stmt), var, op0, op1); - SSA_NAME_DEF_STMT (var) = pattern_stmt; + pattern_stmt + = gimple_build_assign_with_ops (gimple_assign_rhs_code (stmt), var, + op0, op1); STMT_VINFO_RELATED_STMT (vinfo_for_stmt (stmt)) = pattern_stmt; - STMT_VINFO_PATTERN_DEF_STMT (vinfo_for_stmt (stmt)) = new_def_stmt; + new_pattern_def_seq (vinfo_for_stmt (stmt), new_def_stmt); if (vect_print_dump_info (REPORT_DETAILS)) { @@ -1147,6 +1177,7 @@ vect_recog_over_widening_pattern (VEC (gimple, heap) **stmts, print_gimple_stmt (vect_dump, pattern_stmt, 0, TDF_SLIM); } + type = gimple_expr_type (stmt); prev_stmt = stmt; stmt = use_stmt; @@ -1162,11 +1193,15 @@ vect_recog_over_widening_pattern (VEC (gimple, heap) **stmts, { use_lhs = gimple_assign_lhs (use_stmt); use_type = TREE_TYPE (use_lhs); - /* Support only type promotion or signedess change. */ + /* Support only type demotion or signedess change. */ if (!INTEGRAL_TYPE_P (use_type) - || TYPE_PRECISION (new_type) > TYPE_PRECISION (use_type)) + || TYPE_PRECISION (type) <= TYPE_PRECISION (use_type)) return NULL; + /* Check that NEW_TYPE is not bigger than the conversion result. */ + if (TYPE_PRECISION (new_type) > TYPE_PRECISION (use_type)) + return NULL; + if (TYPE_UNSIGNED (new_type) != TYPE_UNSIGNED (use_type) || TYPE_PRECISION (new_type) != TYPE_PRECISION (use_type)) { @@ -1176,7 +1211,6 @@ vect_recog_over_widening_pattern (VEC (gimple, heap) **stmts, new_oprnd = make_ssa_name (tmp, NULL); pattern_stmt = gimple_build_assign_with_ops (NOP_EXPR, new_oprnd, var, NULL_TREE); - SSA_NAME_DEF_STMT (new_oprnd) = pattern_stmt; STMT_VINFO_RELATED_STMT (vinfo_for_stmt (use_stmt)) = pattern_stmt; *type_in = get_vectype_for_scalar_type (new_type); @@ -1192,8 +1226,8 @@ vect_recog_over_widening_pattern (VEC (gimple, heap) **stmts, else { if (prev_stmt) - STMT_VINFO_PATTERN_DEF_STMT (vinfo_for_stmt (use_stmt)) - = STMT_VINFO_PATTERN_DEF_STMT (vinfo_for_stmt (prev_stmt)); + STMT_VINFO_PATTERN_DEF_SEQ (vinfo_for_stmt (use_stmt)) + = STMT_VINFO_PATTERN_DEF_SEQ (vinfo_for_stmt (prev_stmt)); *type_in = vectype; *type_out = NULL_TREE; @@ -1215,6 +1249,1039 @@ vect_recog_over_widening_pattern (VEC (gimple, heap) **stmts, return pattern_stmt; } +/* Detect widening shift pattern: + + type a_t; + TYPE a_T, res_T; + + S1 a_t = ; + S2 a_T = (TYPE) a_t; + S3 res_T = a_T << CONST; + + where type 'TYPE' is at least double the size of type 'type'. + + Also detect cases where the shift result is immediately converted + to another type 'result_type' that is no larger in size than 'TYPE'. + In those cases we perform a widen-shift that directly results in + 'result_type', to avoid a possible over-widening situation: + + type a_t; + TYPE a_T, res_T; + result_type res_result; + + S1 a_t = ; + S2 a_T = (TYPE) a_t; + S3 res_T = a_T << CONST; + S4 res_result = (result_type) res_T; + '--> res_result' = a_t w<< CONST; + + And a case when 'TYPE' is 4 times bigger than 'type'. In that case we + create an additional pattern stmt for S2 to create a variable of an + intermediate type, and perform widen-shift on the intermediate type: + + type a_t; + interm_type a_it; + TYPE a_T, res_T, res_T'; + + S1 a_t = ; + S2 a_T = (TYPE) a_t; + '--> a_it = (interm_type) a_t; + S3 res_T = a_T << CONST; + '--> res_T' = a_it <<* CONST; + + Input/Output: + + * STMTS: Contains a stmt from which the pattern search begins. + In case of unsigned widen-shift, the original stmt (S3) is replaced with S4 + in STMTS. When an intermediate type is used and a pattern statement is + created for S2, we also put S2 here (before S3). + + Output: + + * TYPE_IN: The type of the input arguments to the pattern. + + * TYPE_OUT: The type of the output of this pattern. + + * Return value: A new stmt that will be used to replace the sequence of + stmts that constitute the pattern. In this case it will be: + WIDEN_LSHIFT_EXPR . */ + +static gimple +vect_recog_widen_shift_pattern (VEC (gimple, heap) **stmts, + tree *type_in, tree *type_out) +{ + gimple last_stmt = VEC_pop (gimple, *stmts); + gimple def_stmt0; + tree oprnd0, oprnd1; + tree type, half_type0; + gimple pattern_stmt; + tree vectype, vectype_out = NULL_TREE; + tree dummy; + tree var; + enum tree_code dummy_code; + int dummy_int; + VEC (tree, heap) * dummy_vec; + gimple use_stmt; + + if (!is_gimple_assign (last_stmt) || !vinfo_for_stmt (last_stmt)) + return NULL; + + if (STMT_VINFO_IN_PATTERN_P (vinfo_for_stmt (last_stmt))) + return NULL; + + if (gimple_assign_rhs_code (last_stmt) != LSHIFT_EXPR) + return NULL; + + oprnd0 = gimple_assign_rhs1 (last_stmt); + oprnd1 = gimple_assign_rhs2 (last_stmt); + if (TREE_CODE (oprnd0) != SSA_NAME || TREE_CODE (oprnd1) != INTEGER_CST) + return NULL; + + /* Check operand 0: it has to be defined by a type promotion. */ + if (!widened_name_p (oprnd0, last_stmt, &half_type0, &def_stmt0, false)) + return NULL; + + /* Check operand 1: has to be positive. We check that it fits the type + in vect_handle_widen_op_by_const (). */ + if (tree_int_cst_compare (oprnd1, size_zero_node) <= 0) + return NULL; + + oprnd0 = gimple_assign_rhs1 (def_stmt0); + type = gimple_expr_type (last_stmt); + + /* Check for subsequent conversion to another type. */ + use_stmt = vect_single_imm_use (last_stmt); + if (use_stmt && is_gimple_assign (use_stmt) + && CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (use_stmt)) + && !STMT_VINFO_IN_PATTERN_P (vinfo_for_stmt (use_stmt))) + { + tree use_lhs = gimple_assign_lhs (use_stmt); + tree use_type = TREE_TYPE (use_lhs); + + if (INTEGRAL_TYPE_P (use_type) + && TYPE_PRECISION (use_type) <= TYPE_PRECISION (type)) + { + last_stmt = use_stmt; + type = use_type; + } + } + + /* Check if this a widening operation. */ + if (!vect_handle_widen_op_by_const (last_stmt, LSHIFT_EXPR, oprnd1, + &oprnd0, stmts, + type, &half_type0, def_stmt0)) + return NULL; + + /* Pattern detected. */ + if (vect_print_dump_info (REPORT_DETAILS)) + fprintf (vect_dump, "vect_recog_widen_shift_pattern: detected: "); + + /* Check target support. */ + vectype = get_vectype_for_scalar_type (half_type0); + vectype_out = get_vectype_for_scalar_type (type); + + if (!vectype + || !vectype_out + || !supportable_widening_operation (WIDEN_LSHIFT_EXPR, last_stmt, + vectype_out, vectype, + &dummy, &dummy, &dummy_code, + &dummy_code, &dummy_int, + &dummy_vec)) + return NULL; + + *type_in = vectype; + *type_out = vectype_out; + + /* Pattern supported. Create a stmt to be used to replace the pattern. */ + var = vect_recog_temp_ssa_var (type, NULL); + pattern_stmt = + gimple_build_assign_with_ops (WIDEN_LSHIFT_EXPR, var, oprnd0, oprnd1); + + if (vect_print_dump_info (REPORT_DETAILS)) + print_gimple_stmt (vect_dump, pattern_stmt, 0, TDF_SLIM); + + VEC_safe_push (gimple, heap, *stmts, last_stmt); + return pattern_stmt; +} + +/* Detect a vector by vector shift pattern that wouldn't be otherwise + vectorized: + + type a_t; + TYPE b_T, res_T; + + S1 a_t = ; + S2 b_T = ; + S3 res_T = b_T op a_t; + + where type 'TYPE' is a type with different size than 'type', + and op is <<, >> or rotate. + + Also detect cases: + + type a_t; + TYPE b_T, c_T, res_T; + + S0 c_T = ; + S1 a_t = (type) c_T; + S2 b_T = ; + S3 res_T = b_T op a_t; + + Input/Output: + + * STMTS: Contains a stmt from which the pattern search begins, + i.e. the shift/rotate stmt. The original stmt (S3) is replaced + with a shift/rotate which has same type on both operands, in the + second case just b_T op c_T, in the first case with added cast + from a_t to c_T in STMT_VINFO_PATTERN_DEF_SEQ. + + Output: + + * TYPE_IN: The type of the input arguments to the pattern. + + * TYPE_OUT: The type of the output of this pattern. + + * Return value: A new stmt that will be used to replace the shift/rotate + S3 stmt. */ + +static gimple +vect_recog_vector_vector_shift_pattern (VEC (gimple, heap) **stmts, + tree *type_in, tree *type_out) +{ + gimple last_stmt = VEC_pop (gimple, *stmts); + tree oprnd0, oprnd1, lhs, var; + gimple pattern_stmt, def_stmt; + enum tree_code rhs_code; + stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt); + loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); + enum vect_def_type dt; + tree def; + + if (!is_gimple_assign (last_stmt)) + return NULL; + + rhs_code = gimple_assign_rhs_code (last_stmt); + switch (rhs_code) + { + case LSHIFT_EXPR: + case RSHIFT_EXPR: + case LROTATE_EXPR: + case RROTATE_EXPR: + break; + default: + return NULL; + } + + if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo)) + return NULL; + + lhs = gimple_assign_lhs (last_stmt); + oprnd0 = gimple_assign_rhs1 (last_stmt); + oprnd1 = gimple_assign_rhs2 (last_stmt); + if (TREE_CODE (oprnd0) != SSA_NAME + || TREE_CODE (oprnd1) != SSA_NAME + || TYPE_MODE (TREE_TYPE (oprnd0)) == TYPE_MODE (TREE_TYPE (oprnd1)) + || TYPE_PRECISION (TREE_TYPE (oprnd1)) + != GET_MODE_PRECISION (TYPE_MODE (TREE_TYPE (oprnd1))) + || TYPE_PRECISION (TREE_TYPE (lhs)) + != TYPE_PRECISION (TREE_TYPE (oprnd0))) + return NULL; + + if (!vect_is_simple_use (oprnd1, last_stmt, loop_vinfo, NULL, &def_stmt, + &def, &dt)) + return NULL; + + if (dt != vect_internal_def) + return NULL; + + *type_in = get_vectype_for_scalar_type (TREE_TYPE (oprnd0)); + *type_out = *type_in; + if (*type_in == NULL_TREE) + return NULL; + + def = NULL_TREE; + if (gimple_assign_cast_p (def_stmt)) + { + tree rhs1 = gimple_assign_rhs1 (def_stmt); + if (TYPE_MODE (TREE_TYPE (rhs1)) == TYPE_MODE (TREE_TYPE (oprnd0)) + && TYPE_PRECISION (TREE_TYPE (rhs1)) + == TYPE_PRECISION (TREE_TYPE (oprnd0))) + def = rhs1; + } + + if (def == NULL_TREE) + { + def = vect_recog_temp_ssa_var (TREE_TYPE (oprnd0), NULL); + def_stmt = gimple_build_assign_with_ops (NOP_EXPR, def, oprnd1, + NULL_TREE); + new_pattern_def_seq (stmt_vinfo, def_stmt); + } + + /* Pattern detected. */ + if (vect_print_dump_info (REPORT_DETAILS)) + fprintf (vect_dump, "vect_recog_vector_vector_shift_pattern: detected: "); + + /* Pattern supported. Create a stmt to be used to replace the pattern. */ + var = vect_recog_temp_ssa_var (TREE_TYPE (oprnd0), NULL); + pattern_stmt = gimple_build_assign_with_ops (rhs_code, var, oprnd0, def); + + if (vect_print_dump_info (REPORT_DETAILS)) + print_gimple_stmt (vect_dump, pattern_stmt, 0, TDF_SLIM); + + VEC_safe_push (gimple, heap, *stmts, last_stmt); + return pattern_stmt; +} + +/* Detect a signed division by power of two constant that wouldn't be + otherwise vectorized: + + type a_t, b_t; + + S1 a_t = b_t / N; + + where type 'type' is a signed integral type and N is a constant positive + power of two. + + Similarly handle signed modulo by power of two constant: + + S4 a_t = b_t % N; + + Input/Output: + + * STMTS: Contains a stmt from which the pattern search begins, + i.e. the division stmt. S1 is replaced by: + S3 y_t = b_t < 0 ? N - 1 : 0; + S2 x_t = b_t + y_t; + S1' a_t = x_t >> log2 (N); + + S4 is replaced by (where *_T temporaries have unsigned type): + S9 y_T = b_t < 0 ? -1U : 0U; + S8 z_T = y_T >> (sizeof (type_t) * CHAR_BIT - log2 (N)); + S7 z_t = (type) z_T; + S6 w_t = b_t + z_t; + S5 x_t = w_t & (N - 1); + S4' a_t = x_t - z_t; + + Output: + + * TYPE_IN: The type of the input arguments to the pattern. + + * TYPE_OUT: The type of the output of this pattern. + + * Return value: A new stmt that will be used to replace the division + S1 or modulo S4 stmt. */ + +static gimple +vect_recog_sdivmod_pow2_pattern (VEC (gimple, heap) **stmts, + tree *type_in, tree *type_out) +{ + gimple last_stmt = VEC_pop (gimple, *stmts); + tree oprnd0, oprnd1, vectype, itype, cond; + gimple pattern_stmt, def_stmt; + enum tree_code rhs_code; + stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt); + loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); + optab optab; + + if (!is_gimple_assign (last_stmt)) + return NULL; + + rhs_code = gimple_assign_rhs_code (last_stmt); + switch (rhs_code) + { + case TRUNC_DIV_EXPR: + case TRUNC_MOD_EXPR: + break; + default: + return NULL; + } + + if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo)) + return NULL; + + oprnd0 = gimple_assign_rhs1 (last_stmt); + oprnd1 = gimple_assign_rhs2 (last_stmt); + itype = TREE_TYPE (oprnd0); + if (TREE_CODE (oprnd0) != SSA_NAME + || TREE_CODE (oprnd1) != INTEGER_CST + || TREE_CODE (itype) != INTEGER_TYPE + || TYPE_UNSIGNED (itype) + || TYPE_PRECISION (itype) != GET_MODE_PRECISION (TYPE_MODE (itype)) + || !integer_pow2p (oprnd1) + || tree_int_cst_sgn (oprnd1) != 1) + return NULL; + + vectype = get_vectype_for_scalar_type (itype); + if (vectype == NULL_TREE) + return NULL; + + /* If the target can handle vectorized division or modulo natively, + don't attempt to optimize this. */ + optab = optab_for_tree_code (rhs_code, vectype, optab_default); + if (optab != NULL) + { + enum machine_mode vec_mode = TYPE_MODE (vectype); + int icode = (int) optab_handler (optab, vec_mode); + if (icode != CODE_FOR_nothing + || GET_MODE_SIZE (vec_mode) == UNITS_PER_WORD) + return NULL; + } + + /* Pattern detected. */ + if (vect_print_dump_info (REPORT_DETAILS)) + fprintf (vect_dump, "vect_recog_sdivmod_pow2_pattern: detected: "); + + cond = build2 (LT_EXPR, boolean_type_node, oprnd0, build_int_cst (itype, 0)); + if (rhs_code == TRUNC_DIV_EXPR) + { + tree var = vect_recog_temp_ssa_var (itype, NULL); + def_stmt + = gimple_build_assign_with_ops3 (COND_EXPR, var, cond, + fold_build2 (MINUS_EXPR, itype, + oprnd1, + build_int_cst (itype, + 1)), + build_int_cst (itype, 0)); + new_pattern_def_seq (stmt_vinfo, def_stmt); + var = vect_recog_temp_ssa_var (itype, NULL); + def_stmt + = gimple_build_assign_with_ops (PLUS_EXPR, var, oprnd0, + gimple_assign_lhs (def_stmt)); + append_pattern_def_seq (stmt_vinfo, def_stmt); + + pattern_stmt + = gimple_build_assign_with_ops (RSHIFT_EXPR, + vect_recog_temp_ssa_var (itype, NULL), + var, + build_int_cst (itype, + tree_log2 (oprnd1))); + } + else + { + tree signmask; + STMT_VINFO_PATTERN_DEF_SEQ (stmt_vinfo) = NULL; + if (compare_tree_int (oprnd1, 2) == 0) + { + signmask = vect_recog_temp_ssa_var (itype, NULL); + def_stmt + = gimple_build_assign_with_ops3 (COND_EXPR, signmask, cond, + build_int_cst (itype, 1), + build_int_cst (itype, 0)); + append_pattern_def_seq (stmt_vinfo, def_stmt); + } + else + { + tree utype + = build_nonstandard_integer_type (TYPE_PRECISION (itype), 1); + tree vecutype = get_vectype_for_scalar_type (utype); + tree shift + = build_int_cst (utype, GET_MODE_BITSIZE (TYPE_MODE (itype)) + - tree_log2 (oprnd1)); + tree var = vect_recog_temp_ssa_var (utype, NULL); + stmt_vec_info def_stmt_vinfo; + + def_stmt + = gimple_build_assign_with_ops3 (COND_EXPR, var, cond, + build_int_cst (utype, -1), + build_int_cst (utype, 0)); + def_stmt_vinfo = new_stmt_vec_info (def_stmt, loop_vinfo, NULL); + set_vinfo_for_stmt (def_stmt, def_stmt_vinfo); + STMT_VINFO_VECTYPE (def_stmt_vinfo) = vecutype; + append_pattern_def_seq (stmt_vinfo, def_stmt); + var = vect_recog_temp_ssa_var (utype, NULL); + def_stmt + = gimple_build_assign_with_ops (RSHIFT_EXPR, var, + gimple_assign_lhs (def_stmt), + shift); + def_stmt_vinfo = new_stmt_vec_info (def_stmt, loop_vinfo, NULL); + set_vinfo_for_stmt (def_stmt, def_stmt_vinfo); + STMT_VINFO_VECTYPE (def_stmt_vinfo) = vecutype; + append_pattern_def_seq (stmt_vinfo, def_stmt); + signmask = vect_recog_temp_ssa_var (itype, NULL); + def_stmt + = gimple_build_assign_with_ops (NOP_EXPR, signmask, var, + NULL_TREE); + append_pattern_def_seq (stmt_vinfo, def_stmt); + } + def_stmt + = gimple_build_assign_with_ops (PLUS_EXPR, + vect_recog_temp_ssa_var (itype, NULL), + oprnd0, signmask); + append_pattern_def_seq (stmt_vinfo, def_stmt); + def_stmt + = gimple_build_assign_with_ops (BIT_AND_EXPR, + vect_recog_temp_ssa_var (itype, NULL), + gimple_assign_lhs (def_stmt), + fold_build2 (MINUS_EXPR, itype, + oprnd1, + build_int_cst (itype, + 1))); + append_pattern_def_seq (stmt_vinfo, def_stmt); + + pattern_stmt + = gimple_build_assign_with_ops (MINUS_EXPR, + vect_recog_temp_ssa_var (itype, NULL), + gimple_assign_lhs (def_stmt), + signmask); + } + + if (vect_print_dump_info (REPORT_DETAILS)) + print_gimple_stmt (vect_dump, pattern_stmt, 0, TDF_SLIM); + + VEC_safe_push (gimple, heap, *stmts, last_stmt); + + *type_in = vectype; + *type_out = vectype; + return pattern_stmt; +} + +/* Function vect_recog_mixed_size_cond_pattern + + Try to find the following pattern: + + type x_t, y_t; + TYPE a_T, b_T, c_T; + loop: + S1 a_T = x_t CMP y_t ? b_T : c_T; + + where type 'TYPE' is an integral type which has different size + from 'type'. b_T and c_T are constants and if 'TYPE' is wider + than 'type', the constants need to fit into an integer type + with the same width as 'type'. + + Input: + + * LAST_STMT: A stmt from which the pattern search begins. + + Output: + + * TYPE_IN: The type of the input arguments to the pattern. + + * TYPE_OUT: The type of the output of this pattern. + + * Return value: A new stmt that will be used to replace the pattern. + Additionally a def_stmt is added. + + a_it = x_t CMP y_t ? b_it : c_it; + a_T = (TYPE) a_it; */ + +static gimple +vect_recog_mixed_size_cond_pattern (VEC (gimple, heap) **stmts, tree *type_in, + tree *type_out) +{ + gimple last_stmt = VEC_index (gimple, *stmts, 0); + tree cond_expr, then_clause, else_clause; + stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt), def_stmt_info; + tree type, vectype, comp_vectype, itype, vecitype; + enum machine_mode cmpmode; + gimple pattern_stmt, def_stmt; + loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); + + if (!is_gimple_assign (last_stmt) + || gimple_assign_rhs_code (last_stmt) != COND_EXPR + || STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_internal_def) + return NULL; + + cond_expr = gimple_assign_rhs1 (last_stmt); + then_clause = gimple_assign_rhs2 (last_stmt); + else_clause = gimple_assign_rhs3 (last_stmt); + + if (TREE_CODE (then_clause) != INTEGER_CST + || TREE_CODE (else_clause) != INTEGER_CST) + return NULL; + + if (!COMPARISON_CLASS_P (cond_expr)) + return NULL; + + comp_vectype + = get_vectype_for_scalar_type (TREE_TYPE (TREE_OPERAND (cond_expr, 0))); + if (comp_vectype == NULL_TREE) + return NULL; + + type = gimple_expr_type (last_stmt); + cmpmode = GET_MODE_INNER (TYPE_MODE (comp_vectype)); + + if (GET_MODE_BITSIZE (TYPE_MODE (type)) == GET_MODE_BITSIZE (cmpmode)) + return NULL; + + vectype = get_vectype_for_scalar_type (type); + if (vectype == NULL_TREE) + return NULL; + + if (expand_vec_cond_expr_p (vectype, comp_vectype)) + return NULL; + + itype = build_nonstandard_integer_type (GET_MODE_BITSIZE (cmpmode), + TYPE_UNSIGNED (type)); + if (itype == NULL_TREE + || GET_MODE_BITSIZE (TYPE_MODE (itype)) != GET_MODE_BITSIZE (cmpmode)) + return NULL; + + vecitype = get_vectype_for_scalar_type (itype); + if (vecitype == NULL_TREE) + return NULL; + + if (!expand_vec_cond_expr_p (vecitype, comp_vectype)) + return NULL; + + if (GET_MODE_BITSIZE (TYPE_MODE (type)) > GET_MODE_BITSIZE (cmpmode)) + { + if (!int_fits_type_p (then_clause, itype) + || !int_fits_type_p (else_clause, itype)) + return NULL; + } + + def_stmt + = gimple_build_assign_with_ops3 (COND_EXPR, + vect_recog_temp_ssa_var (itype, NULL), + unshare_expr (cond_expr), + fold_convert (itype, then_clause), + fold_convert (itype, else_clause)); + pattern_stmt + = gimple_build_assign_with_ops (NOP_EXPR, + vect_recog_temp_ssa_var (type, NULL), + gimple_assign_lhs (def_stmt), NULL_TREE); + + new_pattern_def_seq (stmt_vinfo, def_stmt); + def_stmt_info = new_stmt_vec_info (def_stmt, loop_vinfo, NULL); + set_vinfo_for_stmt (def_stmt, def_stmt_info); + STMT_VINFO_VECTYPE (def_stmt_info) = vecitype; + *type_in = vecitype; + *type_out = vectype; + + return pattern_stmt; +} + + +/* Helper function of vect_recog_bool_pattern. Called recursively, return + true if bool VAR can be optimized that way. */ + +static bool +check_bool_pattern (tree var, loop_vec_info loop_vinfo) +{ + gimple def_stmt; + enum vect_def_type dt; + tree def, rhs1; + enum tree_code rhs_code; + + if (!vect_is_simple_use (var, NULL, loop_vinfo, NULL, &def_stmt, &def, &dt)) + return false; + + if (dt != vect_internal_def) + return false; + + if (!is_gimple_assign (def_stmt)) + return false; + + if (!has_single_use (def)) + return false; + + rhs1 = gimple_assign_rhs1 (def_stmt); + rhs_code = gimple_assign_rhs_code (def_stmt); + switch (rhs_code) + { + case SSA_NAME: + return check_bool_pattern (rhs1, loop_vinfo); + + CASE_CONVERT: + if ((TYPE_PRECISION (TREE_TYPE (rhs1)) != 1 + || !TYPE_UNSIGNED (TREE_TYPE (rhs1))) + && TREE_CODE (TREE_TYPE (rhs1)) != BOOLEAN_TYPE) + return false; + return check_bool_pattern (rhs1, loop_vinfo); + + case BIT_NOT_EXPR: + return check_bool_pattern (rhs1, loop_vinfo); + + case BIT_AND_EXPR: + case BIT_IOR_EXPR: + case BIT_XOR_EXPR: + if (!check_bool_pattern (rhs1, loop_vinfo)) + return false; + return check_bool_pattern (gimple_assign_rhs2 (def_stmt), loop_vinfo); + + default: + if (TREE_CODE_CLASS (rhs_code) == tcc_comparison) + { + tree vecitype, comp_vectype; + + /* If the comparison can throw, then is_gimple_condexpr will be + false and we can't make a COND_EXPR/VEC_COND_EXPR out of it. */ + if (stmt_could_throw_p (def_stmt)) + return false; + + comp_vectype = get_vectype_for_scalar_type (TREE_TYPE (rhs1)); + if (comp_vectype == NULL_TREE) + return false; + + if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE) + { + enum machine_mode mode = TYPE_MODE (TREE_TYPE (rhs1)); + tree itype + = build_nonstandard_integer_type (GET_MODE_BITSIZE (mode), 1); + vecitype = get_vectype_for_scalar_type (itype); + if (vecitype == NULL_TREE) + return false; + } + else + vecitype = comp_vectype; + return expand_vec_cond_expr_p (vecitype, comp_vectype); + } + return false; + } +} + + +/* Helper function of adjust_bool_pattern. Add a cast to TYPE to a previous + stmt (SSA_NAME_DEF_STMT of VAR) by moving the COND_EXPR from RELATED_STMT + to PATTERN_DEF_SEQ and adding a cast as RELATED_STMT. */ + +static tree +adjust_bool_pattern_cast (tree type, tree var) +{ + stmt_vec_info stmt_vinfo = vinfo_for_stmt (SSA_NAME_DEF_STMT (var)); + gimple cast_stmt, pattern_stmt; + + gcc_assert (!STMT_VINFO_PATTERN_DEF_SEQ (stmt_vinfo)); + pattern_stmt = STMT_VINFO_RELATED_STMT (stmt_vinfo); + new_pattern_def_seq (stmt_vinfo, pattern_stmt); + cast_stmt + = gimple_build_assign_with_ops (NOP_EXPR, + vect_recog_temp_ssa_var (type, NULL), + gimple_assign_lhs (pattern_stmt), + NULL_TREE); + STMT_VINFO_RELATED_STMT (stmt_vinfo) = cast_stmt; + return gimple_assign_lhs (cast_stmt); +} + + +/* Helper function of vect_recog_bool_pattern. Do the actual transformations, + recursively. VAR is an SSA_NAME that should be transformed from bool + to a wider integer type, OUT_TYPE is the desired final integer type of + the whole pattern, TRUEVAL should be NULL unless optimizing + BIT_AND_EXPR into a COND_EXPR with one integer from one of the operands + in the then_clause, STMTS is where statements with added pattern stmts + should be pushed to. */ + +static tree +adjust_bool_pattern (tree var, tree out_type, tree trueval, + VEC (gimple, heap) **stmts) +{ + gimple stmt = SSA_NAME_DEF_STMT (var); + enum tree_code rhs_code, def_rhs_code; + tree itype, cond_expr, rhs1, rhs2, irhs1, irhs2; + location_t loc; + gimple pattern_stmt, def_stmt; + + rhs1 = gimple_assign_rhs1 (stmt); + rhs2 = gimple_assign_rhs2 (stmt); + rhs_code = gimple_assign_rhs_code (stmt); + loc = gimple_location (stmt); + switch (rhs_code) + { + case SSA_NAME: + CASE_CONVERT: + irhs1 = adjust_bool_pattern (rhs1, out_type, NULL_TREE, stmts); + itype = TREE_TYPE (irhs1); + pattern_stmt + = gimple_build_assign_with_ops (SSA_NAME, + vect_recog_temp_ssa_var (itype, NULL), + irhs1, NULL_TREE); + break; + + case BIT_NOT_EXPR: + irhs1 = adjust_bool_pattern (rhs1, out_type, NULL_TREE, stmts); + itype = TREE_TYPE (irhs1); + pattern_stmt + = gimple_build_assign_with_ops (BIT_XOR_EXPR, + vect_recog_temp_ssa_var (itype, NULL), + irhs1, build_int_cst (itype, 1)); + break; + + case BIT_AND_EXPR: + /* Try to optimize x = y & (a < b ? 1 : 0); into + x = (a < b ? y : 0); + + E.g. for: + bool a_b, b_b, c_b; + TYPE d_T; + + S1 a_b = x1 CMP1 y1; + S2 b_b = x2 CMP2 y2; + S3 c_b = a_b & b_b; + S4 d_T = (TYPE) c_b; + + we would normally emit: + + S1' a_T = x1 CMP1 y1 ? 1 : 0; + S2' b_T = x2 CMP2 y2 ? 1 : 0; + S3' c_T = a_T & b_T; + S4' d_T = c_T; + + but we can save one stmt by using the + result of one of the COND_EXPRs in the other COND_EXPR and leave + BIT_AND_EXPR stmt out: + + S1' a_T = x1 CMP1 y1 ? 1 : 0; + S3' c_T = x2 CMP2 y2 ? a_T : 0; + S4' f_T = c_T; + + At least when VEC_COND_EXPR is implemented using masks + cond ? 1 : 0 is as expensive as cond ? var : 0, in both cases it + computes the comparison masks and ands it, in one case with + all ones vector, in the other case with a vector register. + Don't do this for BIT_IOR_EXPR, because cond ? 1 : var; is + often more expensive. */ + def_stmt = SSA_NAME_DEF_STMT (rhs2); + def_rhs_code = gimple_assign_rhs_code (def_stmt); + if (TREE_CODE_CLASS (def_rhs_code) == tcc_comparison) + { + tree def_rhs1 = gimple_assign_rhs1 (def_stmt); + irhs1 = adjust_bool_pattern (rhs1, out_type, NULL_TREE, stmts); + if (TYPE_PRECISION (TREE_TYPE (irhs1)) + == GET_MODE_BITSIZE (TYPE_MODE (TREE_TYPE (def_rhs1)))) + { + gimple tstmt; + stmt_vec_info stmt_def_vinfo = vinfo_for_stmt (def_stmt); + irhs2 = adjust_bool_pattern (rhs2, out_type, irhs1, stmts); + tstmt = VEC_pop (gimple, *stmts); + gcc_assert (tstmt == def_stmt); + VEC_quick_push (gimple, *stmts, stmt); + STMT_VINFO_RELATED_STMT (vinfo_for_stmt (stmt)) + = STMT_VINFO_RELATED_STMT (stmt_def_vinfo); + gcc_assert (!STMT_VINFO_PATTERN_DEF_SEQ (stmt_def_vinfo)); + STMT_VINFO_RELATED_STMT (stmt_def_vinfo) = NULL; + return irhs2; + } + else + irhs2 = adjust_bool_pattern (rhs2, out_type, NULL_TREE, stmts); + goto and_ior_xor; + } + def_stmt = SSA_NAME_DEF_STMT (rhs1); + def_rhs_code = gimple_assign_rhs_code (def_stmt); + if (TREE_CODE_CLASS (def_rhs_code) == tcc_comparison) + { + tree def_rhs1 = gimple_assign_rhs1 (def_stmt); + irhs2 = adjust_bool_pattern (rhs2, out_type, NULL_TREE, stmts); + if (TYPE_PRECISION (TREE_TYPE (irhs2)) + == GET_MODE_BITSIZE (TYPE_MODE (TREE_TYPE (def_rhs1)))) + { + gimple tstmt; + stmt_vec_info stmt_def_vinfo = vinfo_for_stmt (def_stmt); + irhs1 = adjust_bool_pattern (rhs1, out_type, irhs2, stmts); + tstmt = VEC_pop (gimple, *stmts); + gcc_assert (tstmt == def_stmt); + VEC_quick_push (gimple, *stmts, stmt); + STMT_VINFO_RELATED_STMT (vinfo_for_stmt (stmt)) + = STMT_VINFO_RELATED_STMT (stmt_def_vinfo); + gcc_assert (!STMT_VINFO_PATTERN_DEF_SEQ (stmt_def_vinfo)); + STMT_VINFO_RELATED_STMT (stmt_def_vinfo) = NULL; + return irhs1; + } + else + irhs1 = adjust_bool_pattern (rhs1, out_type, NULL_TREE, stmts); + goto and_ior_xor; + } + /* FALLTHRU */ + case BIT_IOR_EXPR: + case BIT_XOR_EXPR: + irhs1 = adjust_bool_pattern (rhs1, out_type, NULL_TREE, stmts); + irhs2 = adjust_bool_pattern (rhs2, out_type, NULL_TREE, stmts); + and_ior_xor: + if (TYPE_PRECISION (TREE_TYPE (irhs1)) + != TYPE_PRECISION (TREE_TYPE (irhs2))) + { + int prec1 = TYPE_PRECISION (TREE_TYPE (irhs1)); + int prec2 = TYPE_PRECISION (TREE_TYPE (irhs2)); + int out_prec = TYPE_PRECISION (out_type); + if (absu_hwi (out_prec - prec1) < absu_hwi (out_prec - prec2)) + irhs2 = adjust_bool_pattern_cast (TREE_TYPE (irhs1), rhs2); + else if (absu_hwi (out_prec - prec1) > absu_hwi (out_prec - prec2)) + irhs1 = adjust_bool_pattern_cast (TREE_TYPE (irhs2), rhs1); + else + { + irhs1 = adjust_bool_pattern_cast (out_type, rhs1); + irhs2 = adjust_bool_pattern_cast (out_type, rhs2); + } + } + itype = TREE_TYPE (irhs1); + pattern_stmt + = gimple_build_assign_with_ops (rhs_code, + vect_recog_temp_ssa_var (itype, NULL), + irhs1, irhs2); + break; + + default: + gcc_assert (TREE_CODE_CLASS (rhs_code) == tcc_comparison); + if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE + || !TYPE_UNSIGNED (TREE_TYPE (rhs1)) + || (TYPE_PRECISION (TREE_TYPE (rhs1)) + != GET_MODE_BITSIZE (TYPE_MODE (TREE_TYPE (rhs1))))) + { + enum machine_mode mode = TYPE_MODE (TREE_TYPE (rhs1)); + itype + = build_nonstandard_integer_type (GET_MODE_BITSIZE (mode), 1); + } + else + itype = TREE_TYPE (rhs1); + cond_expr = build2_loc (loc, rhs_code, itype, rhs1, rhs2); + if (trueval == NULL_TREE) + trueval = build_int_cst (itype, 1); + else + gcc_checking_assert (useless_type_conversion_p (itype, + TREE_TYPE (trueval))); + pattern_stmt + = gimple_build_assign_with_ops3 (COND_EXPR, + vect_recog_temp_ssa_var (itype, NULL), + cond_expr, trueval, + build_int_cst (itype, 0)); + break; + } + + VEC_safe_push (gimple, heap, *stmts, stmt); + gimple_set_location (pattern_stmt, loc); + STMT_VINFO_RELATED_STMT (vinfo_for_stmt (stmt)) = pattern_stmt; + return gimple_assign_lhs (pattern_stmt); +} + + +/* Function vect_recog_bool_pattern + + Try to find pattern like following: + + bool a_b, b_b, c_b, d_b, e_b; + TYPE f_T; + loop: + S1 a_b = x1 CMP1 y1; + S2 b_b = x2 CMP2 y2; + S3 c_b = a_b & b_b; + S4 d_b = x3 CMP3 y3; + S5 e_b = c_b | d_b; + S6 f_T = (TYPE) e_b; + + where type 'TYPE' is an integral type. + + Input: + + * LAST_STMT: A stmt at the end from which the pattern + search begins, i.e. cast of a bool to + an integer type. + + Output: + + * TYPE_IN: The type of the input arguments to the pattern. + + * TYPE_OUT: The type of the output of this pattern. + + * Return value: A new stmt that will be used to replace the pattern. + + Assuming size of TYPE is the same as size of all comparisons + (otherwise some casts would be added where needed), the above + sequence we create related pattern stmts: + S1' a_T = x1 CMP1 y1 ? 1 : 0; + S3' c_T = x2 CMP2 y2 ? a_T : 0; + S4' d_T = x3 CMP3 y3 ? 1 : 0; + S5' e_T = c_T | d_T; + S6' f_T = e_T; + + Instead of the above S3' we could emit: + S2' b_T = x2 CMP2 y2 ? 1 : 0; + S3' c_T = a_T | b_T; + but the above is more efficient. */ + +static gimple +vect_recog_bool_pattern (VEC (gimple, heap) **stmts, tree *type_in, + tree *type_out) +{ + gimple last_stmt = VEC_pop (gimple, *stmts); + enum tree_code rhs_code; + tree var, lhs, rhs, vectype; + stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt); + loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); + gimple pattern_stmt; + + if (!is_gimple_assign (last_stmt)) + return NULL; + + var = gimple_assign_rhs1 (last_stmt); + lhs = gimple_assign_lhs (last_stmt); + + if ((TYPE_PRECISION (TREE_TYPE (var)) != 1 + || !TYPE_UNSIGNED (TREE_TYPE (var))) + && TREE_CODE (TREE_TYPE (var)) != BOOLEAN_TYPE) + return NULL; + + rhs_code = gimple_assign_rhs_code (last_stmt); + if (CONVERT_EXPR_CODE_P (rhs_code)) + { + if (TREE_CODE (TREE_TYPE (lhs)) != INTEGER_TYPE + || TYPE_PRECISION (TREE_TYPE (lhs)) == 1) + return NULL; + vectype = get_vectype_for_scalar_type (TREE_TYPE (lhs)); + if (vectype == NULL_TREE) + return NULL; + + if (!check_bool_pattern (var, loop_vinfo)) + return NULL; + + rhs = adjust_bool_pattern (var, TREE_TYPE (lhs), NULL_TREE, stmts); + lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL); + if (useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs))) + pattern_stmt + = gimple_build_assign_with_ops (SSA_NAME, lhs, rhs, NULL_TREE); + else + pattern_stmt + = gimple_build_assign_with_ops (NOP_EXPR, lhs, rhs, NULL_TREE); + *type_out = vectype; + *type_in = vectype; + VEC_safe_push (gimple, heap, *stmts, last_stmt); + return pattern_stmt; + } + else if (rhs_code == SSA_NAME + && STMT_VINFO_DATA_REF (stmt_vinfo)) + { + stmt_vec_info pattern_stmt_info; + vectype = STMT_VINFO_VECTYPE (stmt_vinfo); + gcc_assert (vectype != NULL_TREE); + if (!VECTOR_MODE_P (TYPE_MODE (vectype))) + return NULL; + if (!check_bool_pattern (var, loop_vinfo)) + return NULL; + + rhs = adjust_bool_pattern (var, TREE_TYPE (vectype), NULL_TREE, stmts); + lhs = build1 (VIEW_CONVERT_EXPR, TREE_TYPE (vectype), lhs); + if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs))) + { + tree rhs2 = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL); + gimple cast_stmt + = gimple_build_assign_with_ops (NOP_EXPR, rhs2, rhs, NULL_TREE); + new_pattern_def_seq (stmt_vinfo, cast_stmt); + rhs = rhs2; + } + pattern_stmt + = gimple_build_assign_with_ops (SSA_NAME, lhs, rhs, NULL_TREE); + pattern_stmt_info = new_stmt_vec_info (pattern_stmt, loop_vinfo, NULL); + set_vinfo_for_stmt (pattern_stmt, pattern_stmt_info); + STMT_VINFO_DATA_REF (pattern_stmt_info) + = STMT_VINFO_DATA_REF (stmt_vinfo); + STMT_VINFO_DR_BASE_ADDRESS (pattern_stmt_info) + = STMT_VINFO_DR_BASE_ADDRESS (stmt_vinfo); + STMT_VINFO_DR_INIT (pattern_stmt_info) = STMT_VINFO_DR_INIT (stmt_vinfo); + STMT_VINFO_DR_OFFSET (pattern_stmt_info) + = STMT_VINFO_DR_OFFSET (stmt_vinfo); + STMT_VINFO_DR_STEP (pattern_stmt_info) = STMT_VINFO_DR_STEP (stmt_vinfo); + STMT_VINFO_DR_ALIGNED_TO (pattern_stmt_info) + = STMT_VINFO_DR_ALIGNED_TO (stmt_vinfo); + DR_STMT (STMT_VINFO_DATA_REF (stmt_vinfo)) = pattern_stmt; + *type_out = vectype; + *type_in = vectype; + VEC_safe_push (gimple, heap, *stmts, last_stmt); + return pattern_stmt; + } + else + return NULL; +} + /* Mark statements that are involved in a pattern. */ @@ -1227,30 +2294,42 @@ vect_mark_pattern_stmts (gimple orig_stmt, gimple pattern_stmt, loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (orig_stmt_info); gimple def_stmt; - set_vinfo_for_stmt (pattern_stmt, - new_stmt_vec_info (pattern_stmt, loop_vinfo, NULL)); - gimple_set_bb (pattern_stmt, gimple_bb (orig_stmt)); pattern_stmt_info = vinfo_for_stmt (pattern_stmt); + if (pattern_stmt_info == NULL) + { + pattern_stmt_info = new_stmt_vec_info (pattern_stmt, loop_vinfo, NULL); + set_vinfo_for_stmt (pattern_stmt, pattern_stmt_info); + } + gimple_set_bb (pattern_stmt, gimple_bb (orig_stmt)); STMT_VINFO_RELATED_STMT (pattern_stmt_info) = orig_stmt; STMT_VINFO_DEF_TYPE (pattern_stmt_info) - = STMT_VINFO_DEF_TYPE (orig_stmt_info); + = STMT_VINFO_DEF_TYPE (orig_stmt_info); STMT_VINFO_VECTYPE (pattern_stmt_info) = pattern_vectype; STMT_VINFO_IN_PATTERN_P (orig_stmt_info) = true; STMT_VINFO_RELATED_STMT (orig_stmt_info) = pattern_stmt; - STMT_VINFO_PATTERN_DEF_STMT (pattern_stmt_info) - = STMT_VINFO_PATTERN_DEF_STMT (orig_stmt_info); - if (STMT_VINFO_PATTERN_DEF_STMT (pattern_stmt_info)) + STMT_VINFO_PATTERN_DEF_SEQ (pattern_stmt_info) + = STMT_VINFO_PATTERN_DEF_SEQ (orig_stmt_info); + if (STMT_VINFO_PATTERN_DEF_SEQ (pattern_stmt_info)) { - def_stmt = STMT_VINFO_PATTERN_DEF_STMT (pattern_stmt_info); - set_vinfo_for_stmt (def_stmt, - new_stmt_vec_info (def_stmt, loop_vinfo, NULL)); - gimple_set_bb (def_stmt, gimple_bb (orig_stmt)); - def_stmt_info = vinfo_for_stmt (def_stmt); - STMT_VINFO_RELATED_STMT (def_stmt_info) = orig_stmt; - STMT_VINFO_DEF_TYPE (def_stmt_info) - = STMT_VINFO_DEF_TYPE (orig_stmt_info); - STMT_VINFO_VECTYPE (def_stmt_info) = pattern_vectype; + gimple_stmt_iterator si; + for (si = gsi_start (STMT_VINFO_PATTERN_DEF_SEQ (pattern_stmt_info)); + !gsi_end_p (si); gsi_next (&si)) + { + def_stmt = gsi_stmt (si); + def_stmt_info = vinfo_for_stmt (def_stmt); + if (def_stmt_info == NULL) + { + def_stmt_info = new_stmt_vec_info (def_stmt, loop_vinfo, NULL); + set_vinfo_for_stmt (def_stmt, def_stmt_info); + } + gimple_set_bb (def_stmt, gimple_bb (orig_stmt)); + STMT_VINFO_RELATED_STMT (def_stmt_info) = orig_stmt; + STMT_VINFO_DEF_TYPE (def_stmt_info) + = STMT_VINFO_DEF_TYPE (orig_stmt_info); + if (STMT_VINFO_VECTYPE (def_stmt_info) == NULL_TREE) + STMT_VINFO_VECTYPE (def_stmt_info) = pattern_vectype; + } } } @@ -1277,9 +2356,9 @@ vect_mark_pattern_stmts (gimple orig_stmt, gimple pattern_stmt, for vect_recog_pattern. */ static void -vect_pattern_recog_1 ( - gimple (* vect_recog_func) (VEC (gimple, heap) **, tree *, tree *), - gimple_stmt_iterator si) +vect_pattern_recog_1 (vect_recog_func_ptr vect_recog_func, + gimple_stmt_iterator si, + VEC (gimple, heap) **stmts_to_replace) { gimple stmt = gsi_stmt (si), pattern_stmt; stmt_vec_info stmt_info; @@ -1289,14 +2368,14 @@ vect_pattern_recog_1 ( enum tree_code code; int i; gimple next; - VEC (gimple, heap) *stmts_to_replace = VEC_alloc (gimple, heap, 1); - VEC_quick_push (gimple, stmts_to_replace, stmt); - pattern_stmt = (* vect_recog_func) (&stmts_to_replace, &type_in, &type_out); + VEC_truncate (gimple, *stmts_to_replace, 0); + VEC_quick_push (gimple, *stmts_to_replace, stmt); + pattern_stmt = (* vect_recog_func) (stmts_to_replace, &type_in, &type_out); if (!pattern_stmt) return; - stmt = VEC_last (gimple, stmts_to_replace); + stmt = VEC_last (gimple, *stmts_to_replace); stmt_info = vinfo_for_stmt (stmt); loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); @@ -1304,8 +2383,6 @@ vect_pattern_recog_1 ( { /* No need to check target support (already checked by the pattern recognition function). */ - if (type_out) - gcc_assert (VECTOR_MODE_P (TYPE_MODE (type_out))); pattern_vectype = type_out ? type_out : type_in; } else @@ -1361,8 +2438,8 @@ vect_pattern_recog_1 ( /* It is possible that additional pattern stmts are created and inserted in STMTS_TO_REPLACE. We create a stmt_info for each of them, and mark the relevant statements. */ - for (i = 0; VEC_iterate (gimple, stmts_to_replace, i, stmt) - && (unsigned) i < (VEC_length (gimple, stmts_to_replace) - 1); + for (i = 0; VEC_iterate (gimple, *stmts_to_replace, i, stmt) + && (unsigned) i < (VEC_length (gimple, *stmts_to_replace) - 1); i++) { stmt_info = vinfo_for_stmt (stmt); @@ -1375,8 +2452,6 @@ vect_pattern_recog_1 ( vect_mark_pattern_stmts (stmt, pattern_stmt, NULL_TREE); } - - VEC_free (gimple, heap, stmts_to_replace); } @@ -1465,7 +2540,8 @@ vect_pattern_recog (loop_vec_info loop_vinfo) unsigned int nbbs = loop->num_nodes; gimple_stmt_iterator si; unsigned int i, j; - gimple (* vect_recog_func_ptr) (VEC (gimple, heap) **, tree *, tree *); + vect_recog_func_ptr vect_recog_func; + VEC (gimple, heap) *stmts_to_replace = VEC_alloc (gimple, heap, 1); if (vect_print_dump_info (REPORT_DETAILS)) fprintf (vect_dump, "=== vect_pattern_recog ==="); @@ -1480,9 +2556,12 @@ vect_pattern_recog (loop_vec_info loop_vinfo) /* Scan over all generic vect_recog_xxx_pattern functions. */ for (j = 0; j < NUM_PATTERNS; j++) { - vect_recog_func_ptr = vect_vect_recog_func_ptrs[j]; - vect_pattern_recog_1 (vect_recog_func_ptr, si); + vect_recog_func = vect_vect_recog_func_ptrs[j]; + vect_pattern_recog_1 (vect_recog_func, si, + &stmts_to_replace); } } } + + VEC_free (gimple, heap, stmts_to_replace); }