;; Faraday FA626TE Pipeline Description
;; Copyright (C) 2010-2020 Free Software Foundation, Inc.
;; Written by Mingfeng Wu, based on ARM926EJ-S Pipeline Description.
;;
;; 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/>. */
;; These descriptions are based on the information contained in the
;; FMP626 Core Design Note, Copyright (c) 2010 Faraday Technology Corp.
;; Pipeline architecture
;; S E M W(Q1) Q2
;; ___________________________________________
;; shifter alu
;; mul1 mul2 mul3
;; ld/st1 ld/st2 ld/st3 ld/st4 ld/st5
;; This automaton provides a pipeline description for the Faraday
;; FMP626 core.
;;
;; The model given here assumes that the condition for all conditional
;; instructions is "true", i.e., that all of the instructions are
;; actually executed.
(define_automaton "fmp626")
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Pipelines
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; There is a single pipeline
;;
;; The ALU pipeline has fetch, decode, execute, memory, and
;; write stages. We only need to model the execute, memory and write
;; stages.
(define_cpu_unit "fmp626_core" "fmp626")
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; ALU Instructions
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; ALU instructions require two cycles to execute, and use the ALU
;; pipeline in each of the three stages. The results are available
;; after the execute stage has finished.
;;
;; If the destination register is the PC, the pipelines are stalled
;; for several cycles. That case is not modeled here.
;; ALU operations
(define_insn_reservation "mp626_alu_op" 1
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "alu_imm,alus_imm,alu_sreg,alus_sreg,\
logic_imm,logics_imm,logic_reg,logics_reg,\
adc_imm,adcs_imm,adc_reg,adcs_reg,\
adr,bfm,rev,\
shift_imm,shift_reg,\
mov_imm,mov_reg,mvn_imm,mvn_reg"))
"fmp626_core")
(define_insn_reservation "mp626_alu_shift_op" 2
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "alu_shift_imm,logic_shift_imm,alus_shift_imm,logics_shift_imm,\
alu_shift_reg,logic_shift_reg,alus_shift_reg,logics_shift_reg,\
extend,\
mov_shift,mov_shift_reg,\
mvn_shift,mvn_shift_reg"))
"fmp626_core")
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Multiplication Instructions
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
(define_insn_reservation "mp626_mult1" 2
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "smulwy,smlawy,smulxy,smlaxy"))
"fmp626_core")
(define_insn_reservation "mp626_mult2" 2
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "mul,mla"))
"fmp626_core")
(define_insn_reservation "mp626_mult3" 3
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "muls,mlas,smull,smlal,umull,umlal,smlalxy,smlawx"))
"fmp626_core*2")
(define_insn_reservation "mp626_mult4" 4
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "smulls,smlals,umulls,umlals"))
"fmp626_core*3")
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Load/Store Instructions
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; The models for load/store instructions do not accurately describe
;; the difference between operations with a base register writeback
;; (such as "ldm!"). These models assume that all memory references
;; hit in dcache.
(define_insn_reservation "mp626_load1_op" 5
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "load_4,load_byte"))
"fmp626_core")
(define_insn_reservation "mp626_load2_op" 6
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "load_8,load_12"))
"fmp626_core*2")
(define_insn_reservation "mp626_load3_op" 7
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "load_16"))
"fmp626_core*3")
(define_insn_reservation "mp626_store1_op" 0
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "store_4"))
"fmp626_core")
(define_insn_reservation "mp626_store2_op" 1
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "store_8,store_12"))
"fmp626_core*2")
(define_insn_reservation "mp626_store3_op" 2
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "store_16"))
"fmp626_core*3")
(define_bypass 1 "mp626_load1_op,mp626_load2_op,mp626_load3_op"
"mp626_store1_op,mp626_store2_op,mp626_store3_op"
"arm_no_early_store_addr_dep")
(define_bypass 1 "mp626_alu_op,mp626_alu_shift_op,mp626_mult1,mp626_mult2,\
mp626_mult3,mp626_mult4" "mp626_store1_op"
"arm_no_early_store_addr_dep")
(define_bypass 1 "mp626_alu_shift_op" "mp626_alu_op")
(define_bypass 1 "mp626_alu_shift_op" "mp626_alu_shift_op"
"arm_no_early_alu_shift_dep")
(define_bypass 1 "mp626_mult1,mp626_mult2" "mp626_alu_shift_op"
"arm_no_early_alu_shift_dep")
(define_bypass 2 "mp626_mult3" "mp626_alu_shift_op"
"arm_no_early_alu_shift_dep")
(define_bypass 3 "mp626_mult4" "mp626_alu_shift_op"
"arm_no_early_alu_shift_dep")
(define_bypass 1 "mp626_mult1,mp626_mult2" "mp626_alu_op")
(define_bypass 2 "mp626_mult3" "mp626_alu_op")
(define_bypass 3 "mp626_mult4" "mp626_alu_op")
(define_bypass 4 "mp626_load1_op" "mp626_alu_op")
(define_bypass 5 "mp626_load2_op" "mp626_alu_op")
(define_bypass 6 "mp626_load3_op" "mp626_alu_op")
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Branch and Call Instructions
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Branch instructions are difficult to model accurately. The FMP626
;; core can predict most branches. If the branch is predicted
;; correctly, and predicted early enough, the branch can be completely
;; eliminated from the instruction stream. Some branches can
;; therefore appear to require zero cycle to execute. We assume that
;; all branches are predicted correctly, and that the latency is
;; therefore the minimum value.
(define_insn_reservation "mp626_branch_op" 0
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "branch"))
"fmp626_core")
;; The latency for a call is actually the latency when the result is available.
;; i.e. R0 ready for int return value.
(define_insn_reservation "mp626_call_op" 1
(and (eq_attr "tune" "fmp626")
(eq_attr "type" "call"))
"fmp626_core")