rt_gccstream/gcc/config/rs6000/e300c2c3.md

190 lines
8.1 KiB
Markdown

;; Pipeline description for Motorola PowerPC e300c3 core.
;; Copyright (C) 2008, 2009 Free Software Foundation, Inc.
;; Contributed by Edmar Wienskoski (edmar@freescale.com)
;;
;; 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
;; <http://www.gnu.org/licenses/>.
(define_automaton "ppce300c3_most,ppce300c3_long,ppce300c3_retire")
(define_cpu_unit "ppce300c3_decode_0,ppce300c3_decode_1" "ppce300c3_most")
;; We don't simulate general issue queue (GIC). If we have SU insn
;; and then SU1 insn, they can not be issued on the same cycle
;; (although SU1 insn and then SU insn can be issued) because the SU
;; insn will go to SU1 from GIC0 entry. Fortunately, the first cycle
;; multipass insn scheduling will find the situation and issue the SU1
;; insn and then the SU insn.
(define_cpu_unit "ppce300c3_issue_0,ppce300c3_issue_1" "ppce300c3_most")
;; We could describe completion buffers slots in combination with the
;; retirement units and the order of completion but the result
;; automaton would behave in the same way because we can not describe
;; real latency time with taking in order completion into account.
;; Actually we could define the real latency time by querying reserved
;; automaton units but the current scheduler uses latency time before
;; issuing insns and making any reservations.
;;
;; So our description is aimed to achieve a insn schedule in which the
;; insns would not wait in the completion buffer.
(define_cpu_unit "ppce300c3_retire_0,ppce300c3_retire_1" "ppce300c3_retire")
;; Branch unit:
(define_cpu_unit "ppce300c3_bu" "ppce300c3_most")
;; IU:
(define_cpu_unit "ppce300c3_iu0_stage0,ppce300c3_iu1_stage0" "ppce300c3_most")
;; IU: This used to describe non-pipelined division.
(define_cpu_unit "ppce300c3_mu_div" "ppce300c3_long")
;; SRU:
(define_cpu_unit "ppce300c3_sru_stage0" "ppce300c3_most")
;; Here we simplified LSU unit description not describing the stages.
(define_cpu_unit "ppce300c3_lsu" "ppce300c3_most")
;; FPU:
(define_cpu_unit "ppce300c3_fpu" "ppce300c3_most")
;; The following units are used to make automata deterministic
(define_cpu_unit "present_ppce300c3_decode_0" "ppce300c3_most")
(define_cpu_unit "present_ppce300c3_issue_0" "ppce300c3_most")
(define_cpu_unit "present_ppce300c3_retire_0" "ppce300c3_retire")
(define_cpu_unit "present_ppce300c3_iu0_stage0" "ppce300c3_most")
;; The following sets to make automata deterministic when option ndfa is used.
(presence_set "present_ppce300c3_decode_0" "ppce300c3_decode_0")
(presence_set "present_ppce300c3_issue_0" "ppce300c3_issue_0")
(presence_set "present_ppce300c3_retire_0" "ppce300c3_retire_0")
(presence_set "present_ppce300c3_iu0_stage0" "ppce300c3_iu0_stage0")
;; Some useful abbreviations.
(define_reservation "ppce300c3_decode"
"ppce300c3_decode_0|ppce300c3_decode_1+present_ppce300c3_decode_0")
(define_reservation "ppce300c3_issue"
"ppce300c3_issue_0|ppce300c3_issue_1+present_ppce300c3_issue_0")
(define_reservation "ppce300c3_retire"
"ppce300c3_retire_0|ppce300c3_retire_1+present_ppce300c3_retire_0")
(define_reservation "ppce300c3_iu_stage0"
"ppce300c3_iu0_stage0|ppce300c3_iu1_stage0+present_ppce300c3_iu0_stage0")
;; Compares can be executed either one of the IU or SRU
(define_insn_reservation "ppce300c3_cmp" 1
(and (eq_attr "type" "cmp,compare,delayed_compare,fast_compare")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+(ppce300c3_iu_stage0|ppce300c3_sru_stage0) \
+ppce300c3_retire")
;; Other one cycle IU insns
(define_insn_reservation "ppce300c3_iu" 1
(and (eq_attr "type" "integer,insert_word,isel")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_iu_stage0+ppce300c3_retire")
;; Branch. Actually this latency time is not used by the scheduler.
(define_insn_reservation "ppce300c3_branch" 1
(and (eq_attr "type" "jmpreg,branch")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_bu,ppce300c3_retire")
;; Multiply is non-pipelined but can be executed in any IU
(define_insn_reservation "ppce300c3_multiply" 2
(and (eq_attr "type" "imul,imul2,imul3,imul_compare")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_iu_stage0, \
ppce300c3_iu_stage0+ppce300c3_retire")
;; Divide. We use the average latency time here. We omit reserving a
;; retire unit because of the result automata will be huge.
(define_insn_reservation "ppce300c3_divide" 20
(and (eq_attr "type" "idiv")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_iu_stage0+ppce300c3_mu_div,\
ppce300c3_mu_div*19")
;; CR logical
(define_insn_reservation "ppce300c3_cr_logical" 1
(and (eq_attr "type" "cr_logical,delayed_cr")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_sru_stage0+ppce300c3_retire")
;; Mfcr
(define_insn_reservation "ppce300c3_mfcr" 1
(and (eq_attr "type" "mfcr")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_sru_stage0+ppce300c3_retire")
;; Mtcrf
(define_insn_reservation "ppce300c3_mtcrf" 1
(and (eq_attr "type" "mtcr")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_sru_stage0+ppce300c3_retire")
;; Mtjmpr
(define_insn_reservation "ppce300c3_mtjmpr" 1
(and (eq_attr "type" "mtjmpr,mfjmpr")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_sru_stage0+ppce300c3_retire")
;; Float point instructions
(define_insn_reservation "ppce300c3_fpcompare" 3
(and (eq_attr "type" "fpcompare")
(eq_attr "cpu" "ppce300c3"))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,nothing,ppce300c3_retire")
(define_insn_reservation "ppce300c3_fp" 3
(and (eq_attr "type" "fp")
(eq_attr "cpu" "ppce300c3"))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,nothing,ppce300c3_retire")
(define_insn_reservation "ppce300c3_dmul" 4
(and (eq_attr "type" "dmul")
(eq_attr "cpu" "ppce300c3"))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,ppce300c3_fpu,nothing,ppce300c3_retire")
; Divides are not pipelined
(define_insn_reservation "ppce300c3_sdiv" 18
(and (eq_attr "type" "sdiv")
(eq_attr "cpu" "ppce300c3"))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,ppce300c3_fpu*17")
(define_insn_reservation "ppce300c3_ddiv" 33
(and (eq_attr "type" "ddiv")
(eq_attr "cpu" "ppce300c3"))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,ppce300c3_fpu*32")
;; Loads
(define_insn_reservation "ppce300c3_load" 2
(and (eq_attr "type" "load,load_ext,load_ext_u,load_ext_ux,load_ux,load_u")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_lsu,ppce300c3_retire")
(define_insn_reservation "ppce300c3_fpload" 2
(and (eq_attr "type" "fpload,fpload_ux,fpload_u")
(eq_attr "cpu" "ppce300c3"))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_lsu,ppce300c3_retire")
;; Stores.
(define_insn_reservation "ppce300c3_store" 2
(and (eq_attr "type" "store,store_ux,store_u")
(ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_lsu,ppce300c3_retire")
(define_insn_reservation "ppce300c3_fpstore" 2
(and (eq_attr "type" "fpstore,fpstore_ux,fpstore_u")
(eq_attr "cpu" "ppce300c3"))
"ppce300c3_decode,ppce300c3_issue+ppce300c3_lsu,ppce300c3_retire")