| 1 |
38 |
julius |
;; Scheduling description for IBM POWER5 processor.
|
| 2 |
|
|
;; Copyright (C) 2003, 2004, 2007 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
|
| 8 |
|
|
;; by the Free Software Foundation; either version 3, or (at your
|
| 9 |
|
|
;; option) any later version.
|
| 10 |
|
|
;;
|
| 11 |
|
|
;; GCC is distributed in the hope that it will be useful, but WITHOUT
|
| 12 |
|
|
;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
| 13 |
|
|
;; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
| 14 |
|
|
;; 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 |
|
|
;; .
|
| 19 |
|
|
|
| 20 |
|
|
;; Sources: IBM Red Book and White Paper on POWER5
|
| 21 |
|
|
|
| 22 |
|
|
;; The POWER5 has 2 iu, 2 fpu, 2 lsu per engine (2 engines per chip).
|
| 23 |
|
|
;; Instructions that update more than one register get broken into two
|
| 24 |
|
|
;; (split) or more internal ops. The chip can issue up to 5
|
| 25 |
|
|
;; internal ops per cycle.
|
| 26 |
|
|
|
| 27 |
|
|
(define_automaton "power5iu,power5fpu,power5misc")
|
| 28 |
|
|
|
| 29 |
|
|
(define_cpu_unit "iu1_power5,iu2_power5" "power5iu")
|
| 30 |
|
|
(define_cpu_unit "lsu1_power5,lsu2_power5" "power5misc")
|
| 31 |
|
|
(define_cpu_unit "fpu1_power5,fpu2_power5" "power5fpu")
|
| 32 |
|
|
(define_cpu_unit "bpu_power5,cru_power5" "power5misc")
|
| 33 |
|
|
(define_cpu_unit "du1_power5,du2_power5,du3_power5,du4_power5,du5_power5"
|
| 34 |
|
|
"power5misc")
|
| 35 |
|
|
|
| 36 |
|
|
(define_reservation "lsq_power5"
|
| 37 |
|
|
"(du1_power5,lsu1_power5)\
|
| 38 |
|
|
|(du2_power5,lsu2_power5)\
|
| 39 |
|
|
|(du3_power5,lsu2_power5)\
|
| 40 |
|
|
|(du4_power5,lsu1_power5)")
|
| 41 |
|
|
|
| 42 |
|
|
(define_reservation "iq_power5"
|
| 43 |
|
|
"(du1_power5,iu1_power5)\
|
| 44 |
|
|
|(du2_power5,iu2_power5)\
|
| 45 |
|
|
|(du3_power5,iu2_power5)\
|
| 46 |
|
|
|(du4_power5,iu1_power5)")
|
| 47 |
|
|
|
| 48 |
|
|
(define_reservation "fpq_power5"
|
| 49 |
|
|
"(du1_power5,fpu1_power5)\
|
| 50 |
|
|
|(du2_power5,fpu2_power5)\
|
| 51 |
|
|
|(du3_power5,fpu2_power5)\
|
| 52 |
|
|
|(du4_power5,fpu1_power5)")
|
| 53 |
|
|
|
| 54 |
|
|
; Dispatch slots are allocated in order conforming to program order.
|
| 55 |
|
|
(absence_set "du1_power5" "du2_power5,du3_power5,du4_power5,du5_power5")
|
| 56 |
|
|
(absence_set "du2_power5" "du3_power5,du4_power5,du5_power5")
|
| 57 |
|
|
(absence_set "du3_power5" "du4_power5,du5_power5")
|
| 58 |
|
|
(absence_set "du4_power5" "du5_power5")
|
| 59 |
|
|
|
| 60 |
|
|
|
| 61 |
|
|
; Load/store
|
| 62 |
|
|
(define_insn_reservation "power5-load" 4 ; 3
|
| 63 |
|
|
(and (eq_attr "type" "load")
|
| 64 |
|
|
(eq_attr "cpu" "power5"))
|
| 65 |
|
|
"lsq_power5")
|
| 66 |
|
|
|
| 67 |
|
|
(define_insn_reservation "power5-load-ext" 5
|
| 68 |
|
|
(and (eq_attr "type" "load_ext")
|
| 69 |
|
|
(eq_attr "cpu" "power5"))
|
| 70 |
|
|
"du1_power5+du2_power5,lsu1_power5,nothing,nothing,iu2_power5")
|
| 71 |
|
|
|
| 72 |
|
|
(define_insn_reservation "power5-load-ext-update" 5
|
| 73 |
|
|
(and (eq_attr "type" "load_ext_u")
|
| 74 |
|
|
(eq_attr "cpu" "power5"))
|
| 75 |
|
|
"du1_power5+du2_power5+du3_power5+du4_power5,\
|
| 76 |
|
|
lsu1_power5+iu2_power5,nothing,nothing,iu2_power5")
|
| 77 |
|
|
|
| 78 |
|
|
(define_insn_reservation "power5-load-ext-update-indexed" 5
|
| 79 |
|
|
(and (eq_attr "type" "load_ext_ux")
|
| 80 |
|
|
(eq_attr "cpu" "power5"))
|
| 81 |
|
|
"du1_power5+du2_power5+du3_power5+du4_power5,\
|
| 82 |
|
|
iu1_power5,lsu2_power5+iu1_power5,nothing,nothing,iu2_power5")
|
| 83 |
|
|
|
| 84 |
|
|
(define_insn_reservation "power5-load-update-indexed" 3
|
| 85 |
|
|
(and (eq_attr "type" "load_ux")
|
| 86 |
|
|
(eq_attr "cpu" "power5"))
|
| 87 |
|
|
"du1_power5+du2_power5+du3_power5+du4_power5,\
|
| 88 |
|
|
iu1_power5,lsu2_power5+iu2_power5")
|
| 89 |
|
|
|
| 90 |
|
|
(define_insn_reservation "power5-load-update" 4 ; 3
|
| 91 |
|
|
(and (eq_attr "type" "load_u")
|
| 92 |
|
|
(eq_attr "cpu" "power5"))
|
| 93 |
|
|
"du1_power5+du2_power5,lsu1_power5+iu2_power5")
|
| 94 |
|
|
|
| 95 |
|
|
(define_insn_reservation "power5-fpload" 6 ; 5
|
| 96 |
|
|
(and (eq_attr "type" "fpload")
|
| 97 |
|
|
(eq_attr "cpu" "power5"))
|
| 98 |
|
|
"lsq_power5")
|
| 99 |
|
|
|
| 100 |
|
|
(define_insn_reservation "power5-fpload-update" 6 ; 5
|
| 101 |
|
|
(and (eq_attr "type" "fpload_u,fpload_ux")
|
| 102 |
|
|
(eq_attr "cpu" "power5"))
|
| 103 |
|
|
"du1_power5+du2_power5,lsu1_power5+iu2_power5")
|
| 104 |
|
|
|
| 105 |
|
|
(define_insn_reservation "power5-store" 12
|
| 106 |
|
|
(and (eq_attr "type" "store")
|
| 107 |
|
|
(eq_attr "cpu" "power5"))
|
| 108 |
|
|
"(du1_power5,lsu1_power5,iu1_power5)\
|
| 109 |
|
|
|(du2_power5,lsu2_power5,iu2_power5)\
|
| 110 |
|
|
|(du3_power5,lsu2_power5,iu2_power5)\
|
| 111 |
|
|
|(du4_power5,lsu1_power5,iu1_power5)")
|
| 112 |
|
|
|
| 113 |
|
|
(define_insn_reservation "power5-store-update" 12
|
| 114 |
|
|
(and (eq_attr "type" "store_u")
|
| 115 |
|
|
(eq_attr "cpu" "power5"))
|
| 116 |
|
|
"du1_power5+du2_power5,lsu1_power5+iu2_power5,iu1_power5")
|
| 117 |
|
|
|
| 118 |
|
|
(define_insn_reservation "power5-store-update-indexed" 12
|
| 119 |
|
|
(and (eq_attr "type" "store_ux")
|
| 120 |
|
|
(eq_attr "cpu" "power5"))
|
| 121 |
|
|
"du1_power5+du2_power5+du3_power5+du4_power5,\
|
| 122 |
|
|
iu1_power5,lsu2_power5+iu2_power5,iu2_power5")
|
| 123 |
|
|
|
| 124 |
|
|
(define_insn_reservation "power5-fpstore" 12
|
| 125 |
|
|
(and (eq_attr "type" "fpstore")
|
| 126 |
|
|
(eq_attr "cpu" "power5"))
|
| 127 |
|
|
"(du1_power5,lsu1_power5,fpu1_power5)\
|
| 128 |
|
|
|(du2_power5,lsu2_power5,fpu2_power5)\
|
| 129 |
|
|
|(du3_power5,lsu2_power5,fpu2_power5)\
|
| 130 |
|
|
|(du4_power5,lsu1_power5,fpu1_power5)")
|
| 131 |
|
|
|
| 132 |
|
|
(define_insn_reservation "power5-fpstore-update" 12
|
| 133 |
|
|
(and (eq_attr "type" "fpstore_u,fpstore_ux")
|
| 134 |
|
|
(eq_attr "cpu" "power5"))
|
| 135 |
|
|
"du1_power5+du2_power5,lsu1_power5+iu2_power5,fpu1_power5")
|
| 136 |
|
|
|
| 137 |
|
|
(define_insn_reservation "power5-llsc" 11
|
| 138 |
|
|
(and (eq_attr "type" "load_l,store_c,sync")
|
| 139 |
|
|
(eq_attr "cpu" "power5"))
|
| 140 |
|
|
"du1_power5+du2_power5+du3_power5+du4_power5,\
|
| 141 |
|
|
lsu1_power5")
|
| 142 |
|
|
|
| 143 |
|
|
|
| 144 |
|
|
; Integer latency is 2 cycles
|
| 145 |
|
|
(define_insn_reservation "power5-integer" 2
|
| 146 |
|
|
(and (eq_attr "type" "integer")
|
| 147 |
|
|
(eq_attr "cpu" "power5"))
|
| 148 |
|
|
"iq_power5")
|
| 149 |
|
|
|
| 150 |
|
|
(define_insn_reservation "power5-two" 2
|
| 151 |
|
|
(and (eq_attr "type" "two")
|
| 152 |
|
|
(eq_attr "cpu" "power5"))
|
| 153 |
|
|
"(du1_power5+du2_power5,iu1_power5,nothing,iu2_power5)\
|
| 154 |
|
|
|(du2_power5+du3_power5,iu2_power5,nothing,iu2_power5)\
|
| 155 |
|
|
|(du3_power5+du4_power5,iu2_power5,nothing,iu1_power5)\
|
| 156 |
|
|
|(du4_power5+du1_power5,iu1_power5,nothing,iu1_power5)")
|
| 157 |
|
|
|
| 158 |
|
|
(define_insn_reservation "power5-three" 2
|
| 159 |
|
|
(and (eq_attr "type" "three")
|
| 160 |
|
|
(eq_attr "cpu" "power5"))
|
| 161 |
|
|
"(du1_power5+du2_power5+du3_power5,\
|
| 162 |
|
|
iu1_power5,nothing,iu2_power5,nothing,iu2_power5)\
|
| 163 |
|
|
|(du2_power5+du3_power5+du4_power5,\
|
| 164 |
|
|
iu2_power5,nothing,iu2_power5,nothing,iu1_power5)\
|
| 165 |
|
|
|(du3_power5+du4_power5+du1_power5,\
|
| 166 |
|
|
iu2_power5,nothing,iu1_power5,nothing,iu1_power5)\
|
| 167 |
|
|
|(du4_power5+du1_power5+du2_power5,\
|
| 168 |
|
|
iu1_power5,nothing,iu2_power5,nothing,iu2_power5)")
|
| 169 |
|
|
|
| 170 |
|
|
(define_insn_reservation "power5-insert" 4
|
| 171 |
|
|
(and (eq_attr "type" "insert_word")
|
| 172 |
|
|
(eq_attr "cpu" "power5"))
|
| 173 |
|
|
"du1_power5+du2_power5,iu1_power5,nothing,iu2_power5")
|
| 174 |
|
|
|
| 175 |
|
|
(define_insn_reservation "power5-cmp" 3
|
| 176 |
|
|
(and (eq_attr "type" "cmp,fast_compare")
|
| 177 |
|
|
(eq_attr "cpu" "power5"))
|
| 178 |
|
|
"iq_power5")
|
| 179 |
|
|
|
| 180 |
|
|
(define_insn_reservation "power5-compare" 2
|
| 181 |
|
|
(and (eq_attr "type" "compare,delayed_compare")
|
| 182 |
|
|
(eq_attr "cpu" "power5"))
|
| 183 |
|
|
"du1_power5+du2_power5,iu1_power5,iu2_power5")
|
| 184 |
|
|
|
| 185 |
|
|
(define_bypass 4 "power5-compare" "power5-branch,power5-crlogical,power5-delayedcr,power5-mfcr,power5-mfcrf")
|
| 186 |
|
|
|
| 187 |
|
|
(define_insn_reservation "power5-lmul-cmp" 7
|
| 188 |
|
|
(and (eq_attr "type" "lmul_compare")
|
| 189 |
|
|
(eq_attr "cpu" "power5"))
|
| 190 |
|
|
"du1_power5+du2_power5,iu1_power5*6,iu2_power5")
|
| 191 |
|
|
|
| 192 |
|
|
(define_bypass 10 "power5-lmul-cmp" "power5-branch,power5-crlogical,power5-delayedcr,power5-mfcr,power5-mfcrf")
|
| 193 |
|
|
|
| 194 |
|
|
(define_insn_reservation "power5-imul-cmp" 5
|
| 195 |
|
|
(and (eq_attr "type" "imul_compare")
|
| 196 |
|
|
(eq_attr "cpu" "power5"))
|
| 197 |
|
|
"du1_power5+du2_power5,iu1_power5*4,iu2_power5")
|
| 198 |
|
|
|
| 199 |
|
|
(define_bypass 8 "power5-imul-cmp" "power5-branch,power5-crlogical,power5-delayedcr,power5-mfcr,power5-mfcrf")
|
| 200 |
|
|
|
| 201 |
|
|
(define_insn_reservation "power5-lmul" 7
|
| 202 |
|
|
(and (eq_attr "type" "lmul")
|
| 203 |
|
|
(eq_attr "cpu" "power5"))
|
| 204 |
|
|
"(du1_power5,iu1_power5*6)\
|
| 205 |
|
|
|(du2_power5,iu2_power5*6)\
|
| 206 |
|
|
|(du3_power5,iu2_power5*6)\
|
| 207 |
|
|
|(du4_power5,iu1_power5*6)")
|
| 208 |
|
|
|
| 209 |
|
|
(define_insn_reservation "power5-imul" 5
|
| 210 |
|
|
(and (eq_attr "type" "imul")
|
| 211 |
|
|
(eq_attr "cpu" "power5"))
|
| 212 |
|
|
"(du1_power5,iu1_power5*4)\
|
| 213 |
|
|
|(du2_power5,iu2_power5*4)\
|
| 214 |
|
|
|(du3_power5,iu2_power5*4)\
|
| 215 |
|
|
|(du4_power5,iu1_power5*4)")
|
| 216 |
|
|
|
| 217 |
|
|
(define_insn_reservation "power5-imul3" 4
|
| 218 |
|
|
(and (eq_attr "type" "imul2,imul3")
|
| 219 |
|
|
(eq_attr "cpu" "power5"))
|
| 220 |
|
|
"(du1_power5,iu1_power5*3)\
|
| 221 |
|
|
|(du2_power5,iu2_power5*3)\
|
| 222 |
|
|
|(du3_power5,iu2_power5*3)\
|
| 223 |
|
|
|(du4_power5,iu1_power5*3)")
|
| 224 |
|
|
|
| 225 |
|
|
|
| 226 |
|
|
; SPR move only executes in first IU.
|
| 227 |
|
|
; Integer division only executes in second IU.
|
| 228 |
|
|
(define_insn_reservation "power5-idiv" 36
|
| 229 |
|
|
(and (eq_attr "type" "idiv")
|
| 230 |
|
|
(eq_attr "cpu" "power5"))
|
| 231 |
|
|
"du1_power5+du2_power5,iu2_power5*35")
|
| 232 |
|
|
|
| 233 |
|
|
(define_insn_reservation "power5-ldiv" 68
|
| 234 |
|
|
(and (eq_attr "type" "ldiv")
|
| 235 |
|
|
(eq_attr "cpu" "power5"))
|
| 236 |
|
|
"du1_power5+du2_power5,iu2_power5*67")
|
| 237 |
|
|
|
| 238 |
|
|
|
| 239 |
|
|
(define_insn_reservation "power5-mtjmpr" 3
|
| 240 |
|
|
(and (eq_attr "type" "mtjmpr,mfjmpr")
|
| 241 |
|
|
(eq_attr "cpu" "power5"))
|
| 242 |
|
|
"du1_power5,bpu_power5")
|
| 243 |
|
|
|
| 244 |
|
|
|
| 245 |
|
|
; Branches take dispatch Slot 4. The presence_sets prevent other insn from
|
| 246 |
|
|
; grabbing previous dispatch slots once this is assigned.
|
| 247 |
|
|
(define_insn_reservation "power5-branch" 2
|
| 248 |
|
|
(and (eq_attr "type" "jmpreg,branch")
|
| 249 |
|
|
(eq_attr "cpu" "power5"))
|
| 250 |
|
|
"(du5_power5\
|
| 251 |
|
|
|du4_power5+du5_power5\
|
| 252 |
|
|
|du3_power5+du4_power5+du5_power5\
|
| 253 |
|
|
|du2_power5+du3_power5+du4_power5+du5_power5\
|
| 254 |
|
|
|du1_power5+du2_power5+du3_power5+du4_power5+du5_power5),bpu_power5")
|
| 255 |
|
|
|
| 256 |
|
|
|
| 257 |
|
|
; Condition Register logical ops are split if non-destructive (RT != RB)
|
| 258 |
|
|
(define_insn_reservation "power5-crlogical" 2
|
| 259 |
|
|
(and (eq_attr "type" "cr_logical")
|
| 260 |
|
|
(eq_attr "cpu" "power5"))
|
| 261 |
|
|
"du1_power5,cru_power5")
|
| 262 |
|
|
|
| 263 |
|
|
(define_insn_reservation "power5-delayedcr" 4
|
| 264 |
|
|
(and (eq_attr "type" "delayed_cr")
|
| 265 |
|
|
(eq_attr "cpu" "power5"))
|
| 266 |
|
|
"du1_power5+du2_power5,cru_power5,cru_power5")
|
| 267 |
|
|
|
| 268 |
|
|
; 4 mfcrf (each 3 cyc, 1/cyc) + 3 fxu
|
| 269 |
|
|
(define_insn_reservation "power5-mfcr" 6
|
| 270 |
|
|
(and (eq_attr "type" "mfcr")
|
| 271 |
|
|
(eq_attr "cpu" "power5"))
|
| 272 |
|
|
"du1_power5+du2_power5+du3_power5+du4_power5,\
|
| 273 |
|
|
du1_power5+du2_power5+du3_power5+du4_power5+cru_power5,\
|
| 274 |
|
|
cru_power5,cru_power5,cru_power5")
|
| 275 |
|
|
|
| 276 |
|
|
; mfcrf (1 field)
|
| 277 |
|
|
(define_insn_reservation "power5-mfcrf" 3
|
| 278 |
|
|
(and (eq_attr "type" "mfcrf")
|
| 279 |
|
|
(eq_attr "cpu" "power5"))
|
| 280 |
|
|
"du1_power5,cru_power5")
|
| 281 |
|
|
|
| 282 |
|
|
; mtcrf (1 field)
|
| 283 |
|
|
(define_insn_reservation "power5-mtcr" 4
|
| 284 |
|
|
(and (eq_attr "type" "mtcr")
|
| 285 |
|
|
(eq_attr "cpu" "power5"))
|
| 286 |
|
|
"du1_power5,iu1_power5")
|
| 287 |
|
|
|
| 288 |
|
|
; Basic FP latency is 6 cycles
|
| 289 |
|
|
(define_insn_reservation "power5-fp" 6
|
| 290 |
|
|
(and (eq_attr "type" "fp,dmul")
|
| 291 |
|
|
(eq_attr "cpu" "power5"))
|
| 292 |
|
|
"fpq_power5")
|
| 293 |
|
|
|
| 294 |
|
|
(define_insn_reservation "power5-fpcompare" 5
|
| 295 |
|
|
(and (eq_attr "type" "fpcompare")
|
| 296 |
|
|
(eq_attr "cpu" "power5"))
|
| 297 |
|
|
"fpq_power5")
|
| 298 |
|
|
|
| 299 |
|
|
(define_insn_reservation "power5-sdiv" 33
|
| 300 |
|
|
(and (eq_attr "type" "sdiv,ddiv")
|
| 301 |
|
|
(eq_attr "cpu" "power5"))
|
| 302 |
|
|
"(du1_power5,fpu1_power5*28)\
|
| 303 |
|
|
|(du2_power5,fpu2_power5*28)\
|
| 304 |
|
|
|(du3_power5,fpu2_power5*28)\
|
| 305 |
|
|
|(du4_power5,fpu1_power5*28)")
|
| 306 |
|
|
|
| 307 |
|
|
(define_insn_reservation "power5-sqrt" 40
|
| 308 |
|
|
(and (eq_attr "type" "ssqrt,dsqrt")
|
| 309 |
|
|
(eq_attr "cpu" "power5"))
|
| 310 |
|
|
"(du1_power5,fpu1_power5*35)\
|
| 311 |
|
|
|(du2_power5,fpu2_power5*35)\
|
| 312 |
|
|
|(du3_power5,fpu2_power5*35)\
|
| 313 |
|
|
|(du4_power5,fpu2_power5*35)")
|
| 314 |
|
|
|
| 315 |
|
|
(define_insn_reservation "power5-isync" 2
|
| 316 |
|
|
(and (eq_attr "type" "isync")
|
| 317 |
|
|
(eq_attr "cpu" "power5"))
|
| 318 |
|
|
"du1_power5+du2_power5+du3_power5+du4_power5,\
|
| 319 |
|
|
lsu1_power5")
|
| 320 |
|
|
|
