gcc/config/aarch64/tuning_models/generic_armv8_a.h

   1 /* Tuning model description for AArch64 architecture.
   2    Copyright (C) 2023-2025 Free Software Foundation, Inc.
   3
   4    This file is part of GCC.
   5
   6    GCC is free software; you can redistribute it and/or modify it
   7    under the terms of the GNU General Public License as published by
   8    the Free Software Foundation; either version 3, or (at your option)
   9    any later version.
  10
  11    GCC is distributed in the hope that it will be useful, but
  12    WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14    General Public License for more details.
  15
  16    You should have received a copy of the GNU General Public License
  17    along with GCC; see the file COPYING3.  If not see
  18    <http://www.gnu.org/licenses/>.  */
  19
  20 #ifndef GCC_AARCH64_H_GENERIC_ARMV8_A
  21 #define GCC_AARCH64_H_GENERIC_ARMV8_A
  22
  23 #include "generic.h"
  24
  25 static const struct cpu_addrcost_table generic_armv8_a_addrcost_table =
  26 {
  27     {
  28       1, /* hi  */
  29       0, /* si  */
  30       0, /* di  */
  31       1, /* ti  */
  32     },
  33   0, /* pre_modify  */
  34   0, /* post_modify  */
  35   0, /* post_modify_ld3_st3  */
  36   0, /* post_modify_ld4_st4  */
  37   0, /* register_offset  */
  38   0, /* register_sextend  */
  39   0, /* register_zextend  */
  40   0 /* imm_offset  */
  41 };
  42
  43 static const struct cpu_regmove_cost generic_armv8_a_regmove_cost =
  44 {
  45   1, /* GP2GP  */
  46   /* Avoid the use of slow int<->fp moves for spilling by setting
  47      their cost higher than memmov_cost.  */
  48   5, /* GP2FP  */
  49   5, /* FP2GP  */
  50   2 /* FP2FP  */
  51 };
  52
  53 /* Generic costs for Advanced SIMD vector operations.   */
  54 static const advsimd_vec_cost generic_armv8_a_advsimd_vector_cost =
  55 {
  56   1, /* int_stmt_cost  */
  57   1, /* fp_stmt_cost  */
  58   0, /* ld2_st2_permute_cost  */
  59   0, /* ld3_st3_permute_cost  */
  60   0, /* ld4_st4_permute_cost  */
  61   2, /* permute_cost  */
  62   2, /* reduc_i8_cost  */
  63   2, /* reduc_i16_cost  */
  64   2, /* reduc_i32_cost  */
  65   2, /* reduc_i64_cost  */
  66   2, /* reduc_f16_cost  */
  67   2, /* reduc_f32_cost  */
  68   2, /* reduc_f64_cost  */
  69   2, /* store_elt_extra_cost  */
  70   2, /* vec_to_scalar_cost  */
  71   1, /* scalar_to_vec_cost  */
  72   1, /* align_load_cost  */
  73   1, /* unalign_load_cost  */
  74   1, /* unalign_store_cost  */
  75   1  /* store_cost  */
  76 };
  77
  78 /* Generic costs for SVE vector operations.  */
  79 static const sve_vec_cost generic_armv8_a_sve_vector_cost =
  80 {
  81   {
  82     1, /* int_stmt_cost  */
  83     1, /* fp_stmt_cost  */
  84     0, /* ld2_st2_permute_cost  */
  85     0, /* ld3_st3_permute_cost  */
  86     0, /* ld4_st4_permute_cost  */
  87     2, /* permute_cost  */
  88     2, /* reduc_i8_cost  */
  89     2, /* reduc_i16_cost  */
  90     2, /* reduc_i32_cost  */
  91     2, /* reduc_i64_cost  */
  92     2, /* reduc_f16_cost  */
  93     2, /* reduc_f32_cost  */
  94     2, /* reduc_f64_cost  */
  95     2, /* store_elt_extra_cost  */
  96     2, /* vec_to_scalar_cost  */
  97     1, /* scalar_to_vec_cost  */
  98     1, /* align_load_cost  */
  99     1, /* unalign_load_cost  */
 100     1, /* unalign_store_cost  */
 101     1  /* store_cost  */
 102   },
 103   2, /* clast_cost  */
 104   2, /* fadda_f16_cost  */
 105   2, /* fadda_f32_cost  */
 106   2, /* fadda_f64_cost  */
 107   4, /* gather_load_x32_cost  */
 108   2, /* gather_load_x64_cost  */
 109   12, /* gather_load_x32_init_cost  */
 110   4, /* gather_load_x64_init_cost  */
 111   1 /* scatter_store_elt_cost  */
 112 };
 113
 114 /* Generic costs for vector insn classes.  */
 115 static const struct cpu_vector_cost generic_armv8_a_vector_cost =
 116 {
 117   1, /* scalar_int_stmt_cost  */
 118   1, /* scalar_fp_stmt_cost  */
 119   1, /* scalar_load_cost  */
 120   1, /* scalar_store_cost  */
 121   3, /* cond_taken_branch_cost  */
 122   1, /* cond_not_taken_branch_cost  */
 123   &generic_armv8_a_advsimd_vector_cost, /* advsimd  */
 124   &generic_armv8_a_sve_vector_cost, /* sve */
 125   nullptr /* issue_info  */
 126 };
 127
 128 /* Generic costs for branch instructions.  */
 129 static const struct cpu_branch_cost generic_armv8_a_branch_cost =
 130 {
 131   1,  /* Predictable.  */
 132   3   /* Unpredictable.  */
 133 };
 134
 135 /* Generic approximation modes.  */
 136 static const cpu_approx_modes generic_armv8_a_approx_modes =
 137 {
 138   AARCH64_APPROX_NONE,  /* division  */
 139   AARCH64_APPROX_NONE,  /* sqrt  */
 140   AARCH64_APPROX_NONE   /* recip_sqrt  */
 141 };
 142
 143 /* Generic prefetch settings (which disable prefetch).  */
 144 static const cpu_prefetch_tune generic_armv8_a_prefetch_tune =
 145 {
 146   0,                    /* num_slots  */
 147   -1,                   /* l1_cache_size  */
 148   -1,                   /* l1_cache_line_size  */
 149   -1,                   /* l2_cache_size  */
 150   true,                 /* prefetch_dynamic_strides */
 151   -1,                   /* minimum_stride */
 152   -1                    /* default_opt_level  */
 153 };
 154
 155 static const struct tune_params generic_armv8_a_tunings =
 156 {
 157   &cortexa76_extra_costs,
 158   &generic_armv8_a_addrcost_table,
 159   &generic_armv8_a_regmove_cost,
 160   &generic_armv8_a_vector_cost,
 161   &generic_armv8_a_branch_cost,
 162   &generic_armv8_a_approx_modes,
 163   SVE_NOT_IMPLEMENTED, /* sve_width  */
 164   { 4, /* load_int.  */
 165     2, /* store_int.  */
 166     5, /* load_fp.  */
 167     2, /* store_fp.  */
 168     4, /* load_pred.  */
 169     4 /* store_pred.  */
 170   }, /* memmov_cost.  */
 171   3, /* issue_rate  */
 172   AARCH64_FUSE_BASE, /* fusible_ops  */
 173   "32:16",      /* function_align.  */
 174   "4",          /* jump_align.  */
 175   "32:16",      /* loop_align.  */
 176   2,    /* int_reassoc_width.  */
 177   4,    /* fp_reassoc_width.  */
 178   1,    /* fma_reassoc_width.  */
 179   2,    /* vec_reassoc_width.  */
 180   2,    /* min_div_recip_mul_sf.  */
 181   2,    /* min_div_recip_mul_df.  */
 182   0,    /* max_case_values.  */
 183   tune_params::AUTOPREFETCHER_WEAK,     /* autoprefetcher_model.  */
 184   (AARCH64_EXTRA_TUNE_BASE
 185    | AARCH64_EXTRA_TUNE_CSE_SVE_VL_CONSTANTS
 186    | AARCH64_EXTRA_TUNE_MATCHED_VECTOR_THROUGHPUT),     /* tune_flags.  */
 187   &generic_prefetch_tune,
 188   AARCH64_LDP_STP_POLICY_ALWAYS,   /* ldp_policy_model.  */
 189   AARCH64_LDP_STP_POLICY_ALWAYS    /* stp_policy_model.  */
 190 };
 191
 192 #endif /* GCC_AARCH64_H_GENERIC_ARMV8_A.  */