1 /* Profiling definitions for the FRV simulator 2 Copyright (C) 1998-2024 Free Software Foundation, Inc. 3 Contributed by Red Hat. 4 5 This file is part of the GNU Simulators. 6 7 This program is free software; you can redistribute it and/or modify 8 it under the terms of the GNU General Public License as published by 9 the Free Software Foundation; either version 3 of the License, or 10 (at your option) any later version. 11 12 This program is distributed in the hope that it will be useful, 13 but WITHOUT ANY WARRANTY; without even the implied warranty of 14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 15 GNU General Public License for more details. 16 17 You should have received a copy of the GNU General Public License 18 along with this program. If not, see <http://www.gnu.org/licenses/>. */ 19 20 #ifndef PROFILE_H 21 #define PROFILE_H 22 23 #include "opcodes/frv-desc.h" 24 25 /* This struct defines the state of profiling. All fields are of general 26 use to all machines. */ 27 typedef struct 28 { 29 long vliw_insns; /* total number of VLIW insns. */ 30 long vliw_wait; /* number of cycles that the current VLIW insn must wait. */ 31 long post_wait; /* number of cycles that post processing in the current 32 VLIW insn must wait. */ 33 long vliw_cycles;/* number of cycles used by current VLIW insn. */ 34 35 int past_first_p; /* Not the first insns in the VLIW */ 36 37 /* Register latencies. Must be signed since they can be temporarily 38 negative. */ 39 int gr_busy[64]; /* Cycles until GR is available. */ 40 int fr_busy[64]; /* Cycles until FR is available. */ 41 int acc_busy[64]; /* Cycles until FR is available. */ 42 int ccr_busy[8]; /* Cycles until ICC/FCC is available. */ 43 int spr_busy[4096]; /* Cycles until spr is available. */ 44 int idiv_busy[2]; /* Cycles until integer division unit is available. */ 45 int fdiv_busy[2]; /* Cycles until float division unit is available. */ 46 int fsqrt_busy[2]; /* Cycles until square root unit is available. */ 47 int float_busy[4]; /* Cycles until floating point unit is available. */ 48 int media_busy[4]; /* Cycles until media unit is available. */ 49 int branch_penalty; /* Cycles until branch is complete. */ 50 51 int gr_latency[64]; /* Cycles until target GR is available. */ 52 int fr_latency[64]; /* Cycles until target FR is available. */ 53 int acc_latency[64]; /* Cycles until target FR is available. */ 54 int ccr_latency[8]; /* Cycles until target ICC/FCC is available. */ 55 int spr_latency[4096]; /* Cycles until target spr is available. */ 56 57 /* Some registers are busy for a shorter number of cycles than normal 58 depending on how they are used next. the xxx_busy_adjust arrays keep track 59 of how many cycles to adjust down. 60 */ 61 int fr_busy_adjust[64]; 62 int acc_busy_adjust[64]; 63 64 /* Register flags. Each bit represents one register. */ 65 DI cur_gr_complex; 66 DI prev_gr_complex; 67 68 /* Keep track of the total queued post-processing time required before a 69 resource is available. This is applied to the resource's latency once all 70 pending loads for the resource are completed. */ 71 int fr_ptime[64]; 72 73 int branch_hint; /* hint field from branch insn. */ 74 USI branch_address; /* Address of predicted branch. */ 75 USI insn_fetch_address;/* Address of sequential insns fetched. */ 76 int mclracc_acc; /* ACC number of register cleared by mclracc. */ 77 int mclracc_A; /* A field of mclracc. */ 78 79 /* We need to know when the first branch of a vliw insn is taken, so that 80 we don't consider the remaining branches in the vliw insn. */ 81 int vliw_branch_taken; 82 83 /* Keep track of the maximum load stall for each VLIW insn. */ 84 int vliw_load_stall; 85 86 /* Need to know if all cache entries are affected by various cache 87 operations. */ 88 int all_cache_entries; 89 } FRV_PROFILE_STATE; 90 91 #define DUAL_REG(reg) ((reg) >= 0 && (reg) < 63 ? (reg) + 1 : -1) 92 #define DUAL_DOUBLE(reg) ((reg) >= 0 && (reg) < 61 ? (reg) + 2 : -1) 93 94 /* Return the GNER register associated with the given GR register. 95 There is no GNER associated with gr0. */ 96 #define GNER_FOR_GR(gr) ((gr) > 63 ? -1 : \ 97 (gr) > 31 ? H_SPR_GNER0 : \ 98 (gr) > 0 ? H_SPR_GNER1 : \ 99 -1) 100 /* Return the GNER register associated with the given GR register. 101 There is no GNER associated with gr0. */ 102 #define FNER_FOR_FR(fr) ((fr) > 63 ? -1 : \ 103 (fr) > 31 ? H_SPR_FNER0 : \ 104 (fr) > 0 ? H_SPR_FNER1 : \ 105 -1) 106 107 /* Top up the latency of the given GR by the given number of cycles. */ 108 void update_GR_latency (SIM_CPU *, INT, int); 109 void update_GRdouble_latency (SIM_CPU *, INT, int); 110 void update_GR_latency_for_load (SIM_CPU *, INT, int); 111 void update_GRdouble_latency_for_load (SIM_CPU *, INT, int); 112 void update_GR_latency_for_swap (SIM_CPU *, INT, int); 113 void update_FR_latency (SIM_CPU *, INT, int); 114 void update_FRdouble_latency (SIM_CPU *, INT, int); 115 void update_FR_latency_for_load (SIM_CPU *, INT, int); 116 void update_FRdouble_latency_for_load (SIM_CPU *, INT, int); 117 void update_FR_ptime (SIM_CPU *, INT, int); 118 void update_FRdouble_ptime (SIM_CPU *, INT, int); 119 void decrease_ACC_busy (SIM_CPU *, INT, int); 120 void decrease_FR_busy (SIM_CPU *, INT, int); 121 void decrease_GR_busy (SIM_CPU *, INT, int); 122 void increase_FR_busy (SIM_CPU *, INT, int); 123 void increase_ACC_busy (SIM_CPU *, INT, int); 124 void update_ACC_latency (SIM_CPU *, INT, int); 125 void update_CCR_latency (SIM_CPU *, INT, int); 126 void update_SPR_latency (SIM_CPU *, INT, int); 127 void update_idiv_resource_latency (SIM_CPU *, INT, int); 128 void update_fdiv_resource_latency (SIM_CPU *, INT, int); 129 void update_fsqrt_resource_latency (SIM_CPU *, INT, int); 130 void update_float_resource_latency (SIM_CPU *, INT, int); 131 void update_media_resource_latency (SIM_CPU *, INT, int); 132 void update_branch_penalty (SIM_CPU *, int); 133 void update_ACC_ptime (SIM_CPU *, INT, int); 134 void update_SPR_ptime (SIM_CPU *, INT, int); 135 void vliw_wait_for_GR (SIM_CPU *, INT); 136 void vliw_wait_for_GRdouble (SIM_CPU *, INT); 137 void vliw_wait_for_FR (SIM_CPU *, INT); 138 void vliw_wait_for_FRdouble (SIM_CPU *, INT); 139 void vliw_wait_for_CCR (SIM_CPU *, INT); 140 void vliw_wait_for_ACC (SIM_CPU *, INT); 141 void vliw_wait_for_SPR (SIM_CPU *, INT); 142 void vliw_wait_for_idiv_resource (SIM_CPU *, INT); 143 void vliw_wait_for_fdiv_resource (SIM_CPU *, INT); 144 void vliw_wait_for_fsqrt_resource (SIM_CPU *, INT); 145 void vliw_wait_for_float_resource (SIM_CPU *, INT); 146 void vliw_wait_for_media_resource (SIM_CPU *, INT); 147 void load_wait_for_GR (SIM_CPU *, INT); 148 void load_wait_for_FR (SIM_CPU *, INT); 149 void load_wait_for_GRdouble (SIM_CPU *, INT); 150 void load_wait_for_FRdouble (SIM_CPU *, INT); 151 void enforce_full_fr_latency (SIM_CPU *, INT); 152 void enforce_full_acc_latency (SIM_CPU *, INT); 153 void post_wait_for_FR (SIM_CPU *, INT); 154 void post_wait_for_FRdouble (SIM_CPU *, INT); 155 void post_wait_for_ACC (SIM_CPU *, INT); 156 void post_wait_for_CCR (SIM_CPU *, INT); 157 void post_wait_for_SPR (SIM_CPU *, INT); 158 void post_wait_for_fdiv (SIM_CPU *, INT); 159 void post_wait_for_fsqrt (SIM_CPU *, INT); 160 void post_wait_for_float (SIM_CPU *, INT); 161 void post_wait_for_media (SIM_CPU *, INT); 162 163 void trace_vliw_wait_cycles (SIM_CPU *); 164 void handle_resource_wait (SIM_CPU *); 165 166 void request_cache_load (SIM_CPU *, INT, int, int); 167 void request_cache_flush (SIM_CPU *, FRV_CACHE *, int); 168 void request_cache_invalidate (SIM_CPU *, FRV_CACHE *, int); 169 void request_cache_preload (SIM_CPU *, FRV_CACHE *, int); 170 void request_cache_unlock (SIM_CPU *, FRV_CACHE *, int); 171 int load_pending_for_register (SIM_CPU *, int, int, int); 172 173 void set_use_is_gr_complex (SIM_CPU *, INT); 174 void set_use_not_gr_complex (SIM_CPU *, INT); 175 int use_is_gr_complex (SIM_CPU *, INT); 176 177 typedef struct 178 { 179 SI address; 180 unsigned reqno; 181 } FRV_INSN_FETCH_BUFFER; 182 183 extern FRV_INSN_FETCH_BUFFER frv_insn_fetch_buffer[]; 184 185 PROFILE_INFO_CPU_CALLBACK_FN frv_profile_info; 186 187 enum { 188 /* Simulator specific profile bits begin here. */ 189 /* Profile caches. */ 190 PROFILE_CACHE_IDX = PROFILE_NEXT_IDX, 191 /* Profile parallelization. */ 192 PROFILE_PARALLEL_IDX 193 }; 194 195 /* Masks so WITH_PROFILE can have symbolic values. 196 The case choice here is on purpose. The lowercase parts are args to 197 --with-profile. */ 198 #define PROFILE_cache (1 << PROFILE_INSN_IDX) 199 #define PROFILE_parallel (1 << PROFILE_INSN_IDX) 200 201 /* Preprocessor macros to simplify tests of WITH_PROFILE. */ 202 #define WITH_PROFILE_CACHE_P (WITH_PROFILE & PROFILE_insn) 203 #define WITH_PROFILE_PARALLEL_P (WITH_PROFILE & PROFILE_insn) 204 205 #define FRV_COUNT_CYCLES(cpu, condition) \ 206 ((PROFILE_MODEL_P (cpu) && (condition)) || frv_interrupt_state.timer.enabled) 207 208 /* Modelling support. */ 209 extern int frv_save_profile_model_p; 210 211 extern enum FRV_INSN_MODELING { 212 FRV_INSN_NO_MODELING = 0, 213 FRV_INSN_MODEL_PASS_1, 214 FRV_INSN_MODEL_PASS_2, 215 FRV_INSN_MODEL_WRITEBACK 216 } model_insn; 217 218 void 219 frv_model_advance_cycles (SIM_CPU *, int); 220 void 221 frv_model_trace_wait_cycles (SIM_CPU *, int, const char *); 222 223 /* Register types for queued load requests. */ 224 #define REGTYPE_NONE 0 225 #define REGTYPE_FR 1 226 #define REGTYPE_ACC 2 227 228 SI frv_ref_SI (SI); 229 230 #endif /* PROFILE_H */ 231