gcc/config/mips/74k.md

;; DFA-based pipeline description for MIPS32 model 74k.
;; Contributed by MIPS Technologies and CodeSourcery.
;;
;; Reference:
;;   "MIPS32 74K Microarchitecure Specification Rev. 01.02 Jun 15, 2006"
;;   "MIPS32 74Kf Processor Core Datasheet Jun 2, 2006"
;;
;; Copyright (C) 2007-2021 Free Software Foundation, Inc.
;;
;; 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 "r74k_mdu_pipe, r74k_alu_pipe, r74k_agen_pipe, r74k_fpu")
(define_cpu_unit "r74k_mul" "r74k_mdu_pipe")
(define_cpu_unit "r74k_alu" "r74k_alu_pipe")
(define_cpu_unit "r74k_agen" "r74k_agen_pipe")
(define_cpu_unit "r74k_fpu_arith" "r74k_fpu")
(define_cpu_unit "r74k_fpu_ldst" "r74k_fpu")

;; --------------------------------------------------------------
;; Producers
;; --------------------------------------------------------------

;; ALU: Logicals/Arithmetics
;; - Logicals, move (addu/addiu with rt = 0), Set less than, 
;;   sign extend - 1 cycle
(define_insn_reservation "r74k_int_logical" 1
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "logical,move,signext,slt"))
  "r74k_alu")

;; - Arithmetics - 2 cycles
(define_insn_reservation "r74k_int_arith" 2
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "arith,const,shift,clz"))
  "r74k_alu")

(define_insn_reservation "r74k_int_nop" 0
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "nop"))
  "nothing")

(define_insn_reservation "r74k_int_cmove" 4
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "condmove"))
  "r74k_agen*2")

;; MDU: fully pipelined multiplier
;; mult - delivers result to hi/lo in 4 cycle (pipelined)
(define_insn_reservation "r74k_int_mult" 4
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "imul"))
  "r74k_alu+r74k_mul")

;; madd, msub - delivers result to hi/lo in 4 cycle (pipelined)
(define_insn_reservation "r74k_int_madd" 4
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "imadd"))
  "r74k_alu+r74k_mul")

;; mul - delivers result to general register in 7 cycles
(define_insn_reservation "r74k_int_mul3" 7
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "imul3"))
  "r74k_alu+r74k_mul")

;; mfhi, mflo, mflhxu - deliver result to gpr in 7 cycles
(define_insn_reservation "r74k_int_mfhilo" 7
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "mfhi,mflo"))
  "r74k_alu+r74k_mul")

;; mthi, mtlo, mtlhx - deliver result to hi/lo, thence madd, handled as bypass
(define_insn_reservation "r74k_int_mthilo" 7
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "mthi,mtlo"))
  "r74k_alu+r74k_mul")

;; div - default to 50 cycles for 32bit operands.  Faster for 8 bit,
;; but is tricky to identify.
(define_insn_reservation "r74k_int_div" 50
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "idiv"))
  "r74k_alu+r74k_mul*50")

;; call
(define_insn_reservation "r74k_int_call" 1
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "call"))
  "r74k_agen")

;; branch/jump
(define_insn_reservation "r74k_int_jump" 1
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "branch,jump"))
  "r74k_agen")

;; loads: lb, lbu, lh, lhu, ll, lw, lwl, lwr, lwpc, lwxs
;; prefetch: prefetch, prefetchx
(define_insn_reservation "r74k_int_load" 3 
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "load,prefetch,prefetchx"))
  "r74k_agen")

;; stores
(define_insn_reservation "r74k_int_store" 1
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "store"))
  "r74k_agen")


;; Unknowns - Currently these include blockage, consttable and alignment
;;            rtls.  They do not really affect scheduling latency, (blockage
;;            affects scheduling via log links, but not used here).
;;
(define_insn_reservation "r74k_unknown" 1 
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "unknown,atomic,syncloop"))
  "r74k_alu")

(define_insn_reservation "r74k_multi" 10
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "multi"))
  "(r74k_alu+r74k_agen)*10")

;; --------------------------------------------------------------
;; Bypass to Consumer
;; --------------------------------------------------------------

;; load->next use :  3 cycles (Default)
;; load->load base:  4 cycles
;; load->store base: 4 cycles
(define_bypass 4 "r74k_int_load" "r74k_int_load")
(define_bypass 4 "r74k_int_load" "r74k_int_store" "!mips_store_data_bypass_p")

;; logical/move/slt/signext->next use :  1 cycles (Default)
;; logical/move/slt/signext->load base:  2 cycles
;; logical/move/slt/signext->store base: 2 cycles
(define_bypass 2 "r74k_int_logical" "r74k_int_load")
(define_bypass 2 "r74k_int_logical" "r74k_int_store"
  "!mips_store_data_bypass_p")

;; arith->next use :  2 cycles (Default)
;; arith->load base:  3 cycles
;; arith->store base: 3 cycles
(define_bypass 3 "r74k_int_arith" "r74k_int_load")
(define_bypass 3 "r74k_int_arith" "r74k_int_store" "!mips_store_data_bypass_p")

;; cmove->next use :  4 cycles (Default)
;; cmove->load base:  5 cycles
;; cmove->store base: 5 cycles
(define_bypass 5 "r74k_int_cmove"  "r74k_int_load")
(define_bypass 5 "r74k_int_cmove"  "r74k_int_store"
  "!mips_store_data_bypass_p")

;; mult/madd/msub->int_mfhilo  : 4 cycles (default)
;; mult->madd/msub             : 1 cycles
;; madd/msub->madd/msub        : 1 cycles
(define_bypass 1 "r74k_int_mult,r74k_int_mul3" "r74k_int_madd"
  "mips_linked_madd_p")
(define_bypass 1 "r74k_int_madd" "r74k_int_madd"
  "mips_linked_madd_p")


;; --------------------------------------------------------------
;; DSP instructions
;; --------------------------------------------------------------

;; Non-saturating insn have the same latency as normal ALU operations,
(define_insn_reservation "r74k_dsp_alu" 2
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "dspalu"))
  "r74k_alu")

;; Saturating insn takes an extra cycle.
(define_insn_reservation "r74k_dsp_alu_sat" 3
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "dspalusat"))
  "r74k_alu")

;; dpaq_s, dpau, dpsq_s, dpsu, maq_s, mulsaq
;; - delivers result to hi/lo in 6 cycle (bypass at M4)
(define_insn_reservation "r74k_dsp_mac" 6
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "dspmac"))
  "r74k_alu+r74k_mul")

;; dpaq_sa, dpsq_sa, maq_sa
;; - delivers result to hi/lo in 7 cycle (bypass at WB)
(define_insn_reservation "r74k_dsp_mac_sat" 7
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "dspmacsat"))
  "r74k_alu+r74k_mul")

;; extp, extpdp, extpdpv, extpv, extr, extrv
;; - same latency as "mul"
(define_insn_reservation "r74k_dsp_acc_ext" 7
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "accext"))
  "r74k_alu+r74k_mul")

;; mthlip, shilo, shilov
;; - same latency as "mul"
(define_insn_reservation "r74k_dsp_acc_mod" 7
  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
       (eq_attr "type" "accmod"))
  "r74k_alu+r74k_mul")

;; dspalu   ->load/store base
;; dspalusat->load/store base
;; - we should never see these in real life.

;; dsp_mac->dsp_mac     : 1 cycles (repeat rate of 1)
;; dsp_mac->dsp_mac_sat : 1 cycles (repeat rate of 1)
(define_bypass 1 "r74k_dsp_mac" "r74k_dsp_mac")
(define_bypass 1 "r74k_dsp_mac" "r74k_dsp_mac_sat")

;; dsp_mac_sat->dsp_mac_sat : 2 cycles (repeat rate of 2)
;; dsp_mac_sat->dsp_mac     : 2 cycles (repeat rate of 2)
(define_bypass 2 "r74k_dsp_mac_sat" "r74k_dsp_mac_sat")
(define_bypass 2 "r74k_dsp_mac_sat" "r74k_dsp_mac")

(define_bypass 1 "r74k_int_mult" "r74k_dsp_mac")
(define_bypass 1 "r74k_int_mult" "r74k_dsp_mac_sat")

(define_bypass 1 "r74k_int_mul3" "r74k_dsp_mac" "mips_linked_madd_p")
(define_bypass 1 "r74k_int_mul3" "r74k_dsp_mac_sat" "mips_linked_madd_p")

;; Assuming the following is true (bypass at M4)
;;  AP AF AM MB M1 M2 M3 M4 WB GR GC
;;              AP AF AM MB M1 M2 M3 M4 WB GR GC
;; dsp_mac->dsp_acc_ext : 4 cycles
;; dsp_mac->dsp_acc_mod : 4 cycles
(define_bypass 4 "r74k_dsp_mac" "r74k_dsp_acc_ext")
(define_bypass 4 "r74k_dsp_mac" "r74k_dsp_acc_mod")

;; Assuming the following is true (bypass at WB)
;;  AP AF AM MB M1 M2 M3 M4 WB GR GC
;;                 AP AF AM MB M1 M2 M3 M4 WB GR GC
;; dsp_mac_sat->dsp_acc_ext : 5 cycles
;; dsp_mac_sat->dsp_acc_mod : 5 cycles
(define_bypass 5 "r74k_dsp_mac_sat" "r74k_dsp_acc_ext")
(define_bypass 5 "r74k_dsp_mac_sat" "r74k_dsp_acc_mod")


;; --------------------------------------------------------------
;; Floating Point Instructions
;; --------------------------------------------------------------

;; 74Kf FPU runs at 1:1 or 2:1 core/FPU clock ratio.

;; fadd, fabs, fneg, 
(define_insn_reservation "r74kf1_1_fadd" 4
  (and (eq_attr "cpu" "74kf1_1")
       (eq_attr "type" "fadd,fabs,fneg"))
  "r74k_fpu_arith")

(define_insn_reservation "r74kf2_1_fadd" 8
  (and (eq_attr "cpu" "74kf2_1")
       (eq_attr "type" "fadd,fabs,fneg"))
  "r74k_fpu_arith*2")

(define_insn_reservation "r74kf3_2_fadd" 6
  (and (eq_attr "cpu" "74kf3_2")
       (eq_attr "type" "fadd,fabs,fneg"))
  "r74k_fpu_arith")

;; fmove, fcmove
(define_insn_reservation "r74kf1_1_fmove" 4
  (and (eq_attr "cpu" "74kf1_1")
       (eq_attr "type" "fmove"))
  "r74k_fpu_arith")

(define_insn_reservation "r74kf2_1_fmove" 8
  (and (eq_attr "cpu" "74kf2_1")
       (eq_attr "type" "fmove"))
  "r74k_fpu_arith*2")

(define_insn_reservation "r74kf3_2_fmove" 6
  (and (eq_attr "cpu" "74kf3_2")
       (eq_attr "type" "fmove"))
  "r74k_fpu_arith")

;; fload
(define_insn_reservation "r74kf1_1_fload" 4
  (and (eq_attr "cpu" "74kf1_1")
       (eq_attr "type" "fpload,fpidxload"))
  "r74k_agen+r74k_fpu_ldst")

(define_insn_reservation "r74kf2_1_fload" 8
  (and (eq_attr "cpu" "74kf2_1")
       (eq_attr "type" "fpload,fpidxload"))
  "r74k_agen+(r74k_fpu_ldst*2)")

(define_insn_reservation "r74kf3_2_fload" 6
  (and (eq_attr "cpu" "74kf3_2")
       (eq_attr "type" "fpload,fpidxload"))
  "r74k_agen+r74k_fpu_ldst")

;; fstore
(define_insn_reservation "r74kf1_1_fstore" 1
  (and (eq_attr "cpu" "74kf1_1")
       (eq_attr "type" "fpstore,fpidxstore"))
  "r74k_agen+r74k_fpu_ldst")

(define_insn_reservation "r74kf2_1_fstore" 2
  (and (eq_attr "cpu" "74kf2_1")
       (eq_attr "type" "fpstore,fpidxstore"))
  "r74k_agen+(r74k_fpu_ldst*2)")

(define_insn_reservation "r74kf3_2_fstore" 1
  (and (eq_attr "cpu" "74kf3_2")
       (eq_attr "type" "fpstore,fpidxstore"))
  "r74k_agen+r74k_fpu_ldst")

;; fmul, fmadd
(define_insn_reservation "r74kf1_1_fmul_sf" 4
  (and (eq_attr "cpu" "74kf1_1")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "SF")))
  "r74k_fpu_arith")

(define_insn_reservation "r74kf2_1_fmul_sf" 8
  (and (eq_attr "cpu" "74kf2_1")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "SF")))
  "r74k_fpu_arith*2")

(define_insn_reservation "r74kf3_2_fmul_sf" 6
  (and (eq_attr "cpu" "74kf3_2")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "SF")))
  "r74k_fpu_arith")

(define_insn_reservation "r74kf1_1_fmul_df" 5
  (and (eq_attr "cpu" "74kf1_1")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "DF")))
  "r74k_fpu_arith*2")

(define_insn_reservation "r74kf2_1_fmul_df" 10
  (and (eq_attr "cpu" "74kf2_1")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "DF")))
  "r74k_fpu_arith*4")

(define_insn_reservation "r74kf3_2_fmul_df" 7
  (and (eq_attr "cpu" "74kf3_2")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "DF")))
  "r74k_fpu_arith*2")

;; fdiv, fsqrt
(define_insn_reservation "r74kf1_1_fdiv_sf" 17
  (and (eq_attr "cpu" "74kf1_1")
       (and (eq_attr "type" "fdiv,fsqrt")
	    (eq_attr "mode" "SF")))
  "r74k_fpu_arith*14")

(define_insn_reservation "r74kf2_1_fdiv_sf" 34
  (and (eq_attr "cpu" "74kf2_1")
       (and (eq_attr "type" "fdiv,fsqrt")
	    (eq_attr "mode" "SF")))
  "r74k_fpu_arith*28")

(define_insn_reservation "r74kf3_2_fdiv_sf" 25
  (and (eq_attr "cpu" "74kf3_2")
       (and (eq_attr "type" "fdiv,fsqrt")
	    (eq_attr "mode" "SF")))
  "r74k_fpu_arith*14")

(define_insn_reservation "r74kf1_1_fdiv_df" 32
  (and (eq_attr "cpu" "74kf1_1")
       (and (eq_attr "type" "fdiv,fsqrt")
	    (eq_attr "mode" "DF")))
  "r74k_fpu_arith*29")

(define_insn_reservation "r74kf2_1_fdiv_df" 64
  (and (eq_attr "cpu" "74kf2_1")
       (and (eq_attr "type" "fdiv,fsqrt")
	    (eq_attr "mode" "DF")))
  "r74k_fpu_arith*58")

(define_insn_reservation "r74kf3_2_fdiv_df" 48
  (and (eq_attr "cpu" "74kf3_2")
       (and (eq_attr "type" "fdiv,fsqrt")
	    (eq_attr "mode" "DF")))
  "r74k_fpu_arith*29")

;; frsqrt
(define_insn_reservation "r74kf1_1_frsqrt_sf" 17
  (and (eq_attr "cpu" "74kf1_1")
       (and (eq_attr "type" "frsqrt")
	    (eq_attr "mode" "SF")))
  "r74k_fpu_arith*14")

(define_insn_reservation "r74kf2_1_frsqrt_sf" 34
  (and (eq_attr "cpu" "74kf2_1")
       (and (eq_attr "type" "frsqrt")
	    (eq_attr "mode" "SF")))
  "r74k_fpu_arith*28")

(define_insn_reservation "r74kf3_2_frsqrt_sf" 25
  (and (eq_attr "cpu" "74kf3_2")
       (and (eq_attr "type" "frsqrt")
	    (eq_attr "mode" "SF")))
  "r74k_fpu_arith*14")

(define_insn_reservation "r74kf1_1_frsqrt_df" 36
  (and (eq_attr "cpu" "74kf1_1")
       (and (eq_attr "type" "frsqrt")
	    (eq_attr "mode" "DF")))
  "r74k_fpu_arith*31")

(define_insn_reservation "r74kf2_1_frsqrt_df" 72
  (and (eq_attr "cpu" "74kf2_1")
       (and (eq_attr "type" "frsqrt")
	    (eq_attr "mode" "DF")))
  "r74k_fpu_arith*62")

(define_insn_reservation "r74kf3_2_frsqrt_df" 54
  (and (eq_attr "cpu" "74kf3_2")
       (and (eq_attr "type" "frsqrt")
	    (eq_attr "mode" "DF")))
  "r74k_fpu_arith*31")

;; fcmp
(define_insn_reservation "r74kf1_1_fcmp" 4
  (and (eq_attr "cpu" "74kf1_1")
       (eq_attr "type" "fcmp"))
  "r74k_fpu_arith")

(define_insn_reservation "r74kf2_1_fcmp" 8
  (and (eq_attr "cpu" "74kf2_1")
       (eq_attr "type" "fcmp"))
  "r74k_fpu_arith*2")

(define_insn_reservation "r74kf3_2_fcmp" 6
  (and (eq_attr "cpu" "74kf3_2")
       (eq_attr "type" "fcmp"))
  "r74k_fpu_arith")

;; fcvt
(define_insn_reservation "r74kf1_1_fcvt" 4
  (and (eq_attr "cpu" "74kf1_1")
       (eq_attr "type" "fcvt"))
  "r74k_fpu_arith")

(define_insn_reservation "r74kf2_1_fcvt" 8
  (and (eq_attr "cpu" "74kf2_1")
       (eq_attr "type" "fcvt"))
  "r74k_fpu_arith*2")

(define_insn_reservation "r74kf3_2_fcvt" 6
  (and (eq_attr "cpu" "74kf3_2")
       (eq_attr "type" "fcvt"))
  "r74k_fpu_arith")

;; fxfer (MTC1, DMTC1: latency is 4) (MFC1, DMFC1: latency is 1)
(define_insn_reservation "r74kf1_1_fxfer_to_c1" 4
  (and (eq_attr "cpu" "74kf1_1")
       (eq_attr "type" "mtc"))
  "r74k_fpu_arith")

(define_insn_reservation "r74kf2_1_fxfer_to_c1" 8
  (and (eq_attr "cpu" "74kf2_1")
       (eq_attr "type" "mtc"))
  "r74k_fpu_arith*2")

(define_insn_reservation "r74kf3_2_fxfer_to_c1" 6
  (and (eq_attr "cpu" "74kf3_2")
       (eq_attr "type" "mtc"))
  "r74k_fpu_arith")

(define_insn_reservation "r74kf1_1_fxfer_from_c1" 1
  (and (eq_attr "cpu" "74kf1_1")
       (eq_attr "type" "mfc"))
  "r74k_fpu_arith")

(define_insn_reservation "r74kf2_1_fxfer_from_c1" 2
  (and (eq_attr "cpu" "74kf2_1")
       (eq_attr "type" "mfc"))
  "r74k_fpu_arith*2")

(define_insn_reservation "r74kf3_2_fxfer_from_c1" 1
  (and (eq_attr "cpu" "74kf3_2")
       (eq_attr "type" "mfc"))
  "r74k_fpu_arith")