;; DFA-based pipeline descriptions for MIPS Technologies 24K core.
;; Contributed by Chao-ying Fu (fu@mips.com), Nigel Stephens (nigel@mips.com)
;;   and David Ung (davidu@mips.com)
;;
;; The 24kf2_1 is a single-issue processor with a half-clocked fpu.
;; The 24kf1_1 is 24k with 1:1 clocked fpu.
;;
;; References:
;;   "MIPS32 24K Processor Core Family Software User's Manual, Rev 3.04."
;;
;; Copyright (C) 2005, 2007 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 "r24k_cpu, r24k_mdu, r24k_fpu")

;; Integer execution unit.
(define_cpu_unit "r24k_iss"		"r24k_cpu")
(define_cpu_unit "r24k_ixu_arith"	"r24k_cpu")
(define_cpu_unit "r24k_mul3a"	        "r24k_mdu")
(define_cpu_unit "r24k_mul3b"	        "r24k_mdu")
(define_cpu_unit "r24k_mul3c"	        "r24k_mdu")

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

;; 1. Loads: lb, lbu, lh, lhu, ll, lw, lwl, lwr, lwpc, lwxs
(define_insn_reservation "r24k_int_load" 2
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "load"))
  "r24k_iss+r24k_ixu_arith")


;; 2. Arithmetic: add, addi, addiu, addiupc, addu, and, andi, clo, clz,
;;    ext, ins, lui, movn, movz, nor, or, ori, rotr, rotrv, seb, seh, sll,
;;    sllv, slt, slti, sltiu, sltu, sra, srav, srl, srlv, sub, subu, wsbh,
;;    xor, xori
;; (movn/movz is not matched, we'll need to split condmov to
;;  differentiate between integer/float moves)
(define_insn_reservation "r24k_int_arith" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "arith,const,logical,move,nop,shift,signext,slt"))
  "r24k_iss+r24k_ixu_arith")


;; 3. Links: bgezal, bgezall, bltzal, bltzall, jal, jalr, jalx
;; 3a. jr/jalr consumer
(define_insn_reservation "r24k_int_jump" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "call,jump"))
  "r24k_iss+r24k_ixu_arith")

;; 3b. branch consumer
(define_insn_reservation "r24k_int_branch" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "branch"))
  "r24k_iss+r24k_ixu_arith")


;; 4. MDU: fully pipelined multiplier
;; mult - delivers result to hi/lo in 1 cycle (pipelined)
(define_insn_reservation "r24k_int_mult" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "imul"))
  "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")

;; madd, msub - delivers result to hi/lo in 1 cycle (pipelined)
(define_insn_reservation "r24k_int_madd" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "imadd"))
  "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")

;; mul - delivers result to gpr in 5 cycles
(define_insn_reservation "r24k_int_mul3" 5
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "imul3"))
  "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)*5")

;; mfhi, mflo, mflhxu - deliver result to gpr in 5 cycles
(define_insn_reservation "r24k_int_mfhilo" 5
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "mfhilo"))
  "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")

;; mthi, mtlo, mtlhx - deliver result to hi/lo, thence madd, handled as bypass
(define_insn_reservation "r24k_int_mthilo" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "mthilo"))
  "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")

;; div - default to 36 cycles for 32bit operands.  Faster for 24bit, 16bit and
;; 8bit, but is tricky to identify.
(define_insn_reservation "r24k_int_div" 36
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "idiv"))
  "r24k_iss+(r24k_mul3a+r24k_mul3b+r24k_mul3c)*36")


;; 5. Cop: cfc1, di, ei, mfc0, mtc0
;; (Disabled until we add proper cop0 support)
;;(define_insn_reservation "r24k_int_cop" 3
;;  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
;;       (eq_attr "type" "cop0"))
;;  "r24k_iss+r24k_ixu_arith")


;; 6. Store
(define_insn_reservation "r24k_int_store" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (and (eq_attr "type" "store")
	    (eq_attr "mode" "!unknown")))
  "r24k_iss+r24k_ixu_arith")

;; 6.1 Special case - matches the cprestore pattern which don't set the mode
;;     attrib. This avoids being set as r24k_int_store and have it checked
;;     against store_data_bypass_p, which would then fail because cprestore
;;     does not have a normal SET pattern.
(define_insn_reservation "r24k_unknown_store" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (and (eq_attr "type" "store")
	    (eq_attr "mode" "unknown")))
  "r24k_iss+r24k_ixu_arith")


;; 7. Multiple instructions
(define_insn_reservation "r24k_int_multi" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "multi"))
  "r24k_iss+r24k_ixu_arith+r24k_fpu_arith+(r24k_mul3a+r24k_mul3b+r24k_mul3c)")


;; 8. 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 "r24k_int_unknown" 0
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "unknown"))
  "r24k_iss")


;; 9. Prefetch
(define_insn_reservation "r24k_int_prefetch" 1
  (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
       (eq_attr "type" "prefetch,prefetchx"))
  "r24k_iss+r24k_ixu_arith")


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

;; load->next use :  2 cycles (Default)
;; load->load base:  3 cycles
;; load->store base: 3 cycles
;; load->prefetch:   3 cycles
(define_bypass 3 "r24k_int_load" "r24k_int_load")
(define_bypass 3 "r24k_int_load" "r24k_int_store" "!store_data_bypass_p")
(define_bypass 3 "r24k_int_load" "r24k_int_prefetch")

;; arith->next use :  1 cycles (Default)
;; arith->load base:  2 cycles
;; arith->store base: 2 cycles
;; arith->prefetch:   2 cycles
(define_bypass 2 "r24k_int_arith" "r24k_int_load")
(define_bypass 2 "r24k_int_arith" "r24k_int_store" "!store_data_bypass_p")
(define_bypass 2 "r24k_int_arith" "r24k_int_prefetch")

;; mul3->next use : 5 cycles (default)
;; mul3->l/s base : 6 cycles
;; mul3->prefetch : 6 cycles
(define_bypass 6 "r24k_int_mul3" "r24k_int_load")
(define_bypass 6 "r24k_int_mul3" "r24k_int_store" "!store_data_bypass_p")
(define_bypass 6 "r24k_int_mul3" "r24k_int_prefetch")

;; mul3->madd/msub : 1 cycle
(define_bypass 1 "r24k_int_mul3" "r24k_int_madd" "mips_linked_madd_p")

;; mfhilo->next use  : 5 cycles (default)
;; mfhilo->l/s base  : 6 cycles
;; mfhilo->prefetch  : 6 cycles
;; mthilo->madd/msub : 2 cycle (only for mthi/lo not mfhi/lo)
(define_bypass 6 "r24k_int_mfhilo" "r24k_int_load")
(define_bypass 6 "r24k_int_mfhilo" "r24k_int_store" "!store_data_bypass_p")
(define_bypass 6 "r24k_int_mfhilo" "r24k_int_prefetch")
(define_bypass 2 "r24k_int_mthilo" "r24k_int_madd")

;; cop->next use : 3 cycles (Default)
;; cop->l/s base : 4 cycles
;; (define_bypass 4 "r24k_int_cop" "r24k_int_load")
;; (define_bypass 4 "r24k_int_cop" "r24k_int_store" "!store_data_bypass_p")

;; multi->next use : 1 cycles (Default)
;; multi->l/s base : 2 cycles
;; multi->prefetch : 2 cycles
(define_bypass 2 "r24k_int_multi" "r24k_int_load")
(define_bypass 2 "r24k_int_multi" "r24k_int_store" "!store_data_bypass_p")
(define_bypass 2 "r24k_int_multi" "r24k_int_prefetch")


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

(define_cpu_unit "r24k_fpu_arith" "r24k_fpu")

;; The 24k is a single issue cpu, and the fpu runs at half clock speed,
;; so each fpu instruction ties up the shared instruction scheduler for
;; 1 cycle, and the fpu scheduler for 2 cycles.
;;
;; These timings are therefore twice the values in the 24K manual,
;; which are quoted in fpu clocks.
;;
;; The 24kf1_1 is a 24k configured with 1:1 cpu and fpu, so use
;; the unscaled timings

(define_reservation "r24kf2_1_fpu_iss"	"r24k_iss+(r24k_fpu_arith*2)")

;; fadd, fabs, fneg
(define_insn_reservation "r24kf2_1_fadd" 8
  (and (eq_attr "cpu" "24kf2_1")
       (eq_attr "type" "fadd,fabs,fneg"))
  "r24kf2_1_fpu_iss")

;; fmove, fcmove
(define_insn_reservation "r24kf2_1_fmove" 8
  (and (eq_attr "cpu" "24kf2_1")
       (eq_attr "type" "fmove,condmove"))
  "r24kf2_1_fpu_iss")

;; fload
(define_insn_reservation "r24kf2_1_fload" 6
  (and (eq_attr "cpu" "24kf2_1")
       (eq_attr "type" "fpload,fpidxload"))
  "r24kf2_1_fpu_iss")

;; fstore
(define_insn_reservation "r24kf2_1_fstore" 2
  (and (eq_attr "cpu" "24kf2_1")
       (eq_attr "type" "fpstore"))
  "r24kf2_1_fpu_iss")

;; fmul, fmadd
(define_insn_reservation "r24kf2_1_fmul_sf" 8
  (and (eq_attr "cpu" "24kf2_1")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "SF")))
  "r24kf2_1_fpu_iss")

(define_insn_reservation "r24kf2_1_fmul_df" 10
  (and (eq_attr "cpu" "24kf2_1")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "DF")))
  "r24kf2_1_fpu_iss,(r24k_fpu_arith*2)")


;; fdiv, fsqrt, frsqrt
(define_insn_reservation "r24kf2_1_fdiv_sf" 34
  (and (eq_attr "cpu" "24kf2_1")
       (and (eq_attr "type" "fdiv,fsqrt,frsqrt")
	    (eq_attr "mode" "SF")))
  "r24kf2_1_fpu_iss,(r24k_fpu_arith*26)")

(define_insn_reservation "r24kf2_1_fdiv_df" 64
  (and (eq_attr "cpu" "24kf2_1")
       (and (eq_attr "type" "fdiv,fsqrt")
	    (eq_attr "mode" "DF")))
  "r24kf2_1_fpu_iss,(r24k_fpu_arith*56)")

;; frsqrt
(define_insn_reservation "r24kf2_1_frsqrt_df" 70
  (and (eq_attr "cpu" "24kf2_1")
       (and (eq_attr "type" "frsqrt")
	    (eq_attr "mode" "DF")))
  "r24kf2_1_fpu_iss,(r24k_fpu_arith*60)")

;; fcmp
(define_insn_reservation "r24kf2_1_fcmp" 4
  (and (eq_attr "cpu" "24kf2_1")
       (eq_attr "type" "fcmp"))
  "r24kf2_1_fpu_iss")

;; fcmp -> movf.fmt & movt.fmt bypass (dependency must be on the condition)
(define_bypass 2 "r24kf2_1_fcmp" "r24kf2_1_fmove")

;; fcvt (cvt.d.s, cvt.[sd].[wl])
(define_insn_reservation "r24kf2_1_fcvt_i2f_s2d" 8
  (and (eq_attr "cpu" "24kf2_1")
       (and (eq_attr "type" "fcvt")
	    (eq_attr "cnv_mode" "I2S,I2D,S2D")))
  "r24kf2_1_fpu_iss")

;; fcvt (cvt.s.d)
(define_insn_reservation "r24kf2_1_fcvt_s2d" 12
  (and (eq_attr "cpu" "24kf2_1")
       (and (eq_attr "type" "fcvt")
	    (eq_attr "cnv_mode" "D2S")))
  "r24kf2_1_fpu_iss")

;; fcvt (cvt.[wl].[sd], etc)
(define_insn_reservation "r24kf2_1_fcvt_f2i" 10
  (and (eq_attr "cpu" "24kf2_1")
       (and (eq_attr "type" "fcvt")
	    (eq_attr "cnv_mode" "S2I,D2I")))
  "r24kf2_1_fpu_iss")

;; fxfer (mfc1, mfhc1, mtc1, mthc1)
(define_insn_reservation "r24kf2_1_fxfer" 4
  (and (eq_attr "cpu" "24kf2_1")
       (eq_attr "type" "mfc,mtc"))
  "r24kf2_1_fpu_iss")

;; --------------------------------------------------------------
;; Bypass to Consumer
;; --------------------------------------------------------------
;; r24kf2_1_fcvt_f2i->l/s base : 11 cycles
;; r24kf2_1_fcvt_f2i->prefetch : 11 cycles
(define_bypass 11 "r24kf2_1_fcvt_f2i" "r24k_int_load")
(define_bypass 11 "r24kf2_1_fcvt_f2i" "r24k_int_store" "!store_data_bypass_p")
(define_bypass 11 "r24kf2_1_fcvt_f2i" "r24k_int_prefetch")

;; r24kf2_1_fxfer->l/s base : 5 cycles
;; r24kf2_1_fxfer->prefetch : 5 cycles
(define_bypass 5 "r24kf2_1_fxfer" "r24k_int_load")
(define_bypass 5 "r24kf2_1_fxfer" "r24k_int_store" "!store_data_bypass_p")
(define_bypass 5 "r24kf2_1_fxfer" "r24k_int_prefetch")

;; --------------------------------------------------------------
;; The 24kf1_1 is a 24k configured with 1:1 cpu and fpu, so use
;; the unscaled timings
;; --------------------------------------------------------------

(define_reservation "r24kf1_1_fpu_iss"	"r24k_iss+r24k_fpu_arith")

;; fadd, fabs, fneg
(define_insn_reservation "r24kf1_1_fadd" 4
  (and (eq_attr "cpu" "24kf1_1")
       (eq_attr "type" "fadd,fabs,fneg"))
  "r24kf1_1_fpu_iss")

;; fmove, fcmove
(define_insn_reservation "r24kf1_1_fmove" 4
  (and (eq_attr "cpu" "24kf1_1")
       (eq_attr "type" "fmove,condmove"))
  "r24kf1_1_fpu_iss")

;; fload
(define_insn_reservation "r24kf1_1_fload" 3
  (and (eq_attr "cpu" "24kf1_1")
       (eq_attr "type" "fpload,fpidxload"))
  "r24kf1_1_fpu_iss")

;; fstore
(define_insn_reservation "r24kf1_1_fstore" 1
  (and (eq_attr "cpu" "24kf1_1")
       (eq_attr "type" "fpstore"))
  "r24kf1_1_fpu_iss")

;; fmul, fmadd
(define_insn_reservation "r24kf1_1_fmul_sf" 4
  (and (eq_attr "cpu" "24kf1_1")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "SF")))
  "r24kf1_1_fpu_iss")

(define_insn_reservation "r24kf1_1_fmul_df" 5
  (and (eq_attr "cpu" "24kf1_1")
       (and (eq_attr "type" "fmul,fmadd")
	    (eq_attr "mode" "DF")))
  "r24kf1_1_fpu_iss,r24k_fpu_arith")


;; fdiv, fsqrt, frsqrt
(define_insn_reservation "r24kf1_1_fdiv_sf" 17
  (and (eq_attr "cpu" "24kf1_1")
       (and (eq_attr "type" "fdiv,fsqrt,frsqrt")
	    (eq_attr "mode" "SF")))
  "r24kf1_1_fpu_iss,(r24k_fpu_arith*13)")

(define_insn_reservation "r24kf1_1_fdiv_df" 32
  (and (eq_attr "cpu" "24kf1_1")
       (and (eq_attr "type" "fdiv,fsqrt")
	    (eq_attr "mode" "DF")))
  "r24kf1_1_fpu_iss,(r24k_fpu_arith*28)")

;; frsqrt
(define_insn_reservation "r24kf1_1_frsqrt_df" 35
  (and (eq_attr "cpu" "24kf1_1")
       (and (eq_attr "type" "frsqrt")
	    (eq_attr "mode" "DF")))
  "r24kf1_1_fpu_iss,(r24k_fpu_arith*30)")

;; fcmp
(define_insn_reservation "r24kf1_1_fcmp" 2
  (and (eq_attr "cpu" "24kf1_1")
       (eq_attr "type" "fcmp"))
  "r24kf1_1_fpu_iss")

;; fcmp -> movf.fmt & movt.fmt bypass (dependency must be on the condition)
(define_bypass 1 "r24kf1_1_fcmp" "r24kf1_1_fmove")

;; fcvt (cvt.d.s, cvt.[sd].[wl])
(define_insn_reservation "r24kf1_1_fcvt_i2f_s2d" 4
  (and (eq_attr "cpu" "24kf1_1")
       (and (eq_attr "type" "fcvt")
	    (eq_attr "cnv_mode" "I2S,I2D,S2D")))
  "r24kf1_1_fpu_iss")

;; fcvt (cvt.s.d)
(define_insn_reservation "r24kf1_1_fcvt_s2d" 6
  (and (eq_attr "cpu" "24kf1_1")
       (and (eq_attr "type" "fcvt")
	    (eq_attr "cnv_mode" "D2S")))
  "r24kf1_1_fpu_iss")

;; fcvt (cvt.[wl].[sd], etc)
(define_insn_reservation "r24kf1_1_fcvt_f2i" 5
  (and (eq_attr "cpu" "24kf1_1")
       (and (eq_attr "type" "fcvt")
	    (eq_attr "cnv_mode" "S2I,D2I")))
  "r24kf1_1_fpu_iss")

;; fxfer (mfc1, mfhc1, mtc1, mthc1)
(define_insn_reservation "r24kf1_1_fxfer" 2
  (and (eq_attr "cpu" "24kf1_1")
       (eq_attr "type" "mfc,mtc"))
  "r24kf1_1_fpu_iss")

;; --------------------------------------------------------------
;; Bypass to Consumer
;; --------------------------------------------------------------
;; r24kf1_1_fcvt_f2i->l/s base : 6 cycles
;; r24kf1_1_fcvt_f2i->prefetch : 6 cycles
(define_bypass 6 "r24kf1_1_fcvt_f2i" "r24k_int_load")
(define_bypass 6 "r24kf1_1_fcvt_f2i" "r24k_int_store" "!store_data_bypass_p")
(define_bypass 6 "r24kf1_1_fcvt_f2i" "r24k_int_prefetch")

;; r24kf1_1_fxfer->l/s base : 3 cycles
;; r24kf1_1_fxfer->prefetch : 3 cycles
(define_bypass 3 "r24kf1_1_fxfer" "r24k_int_load")
(define_bypass 3 "r24kf1_1_fxfer" "r24k_int_store" "!store_data_bypass_p")
(define_bypass 3 "r24kf1_1_fxfer" "r24k_int_prefetch")