;; DFA scheduling description for SH4. ;; Copyright (C) 2004-2020 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 ;; . ;; Load and store instructions save a cycle if they are aligned on a ;; four byte boundary. Using a function unit for stores encourages ;; gcc to separate load and store instructions by one instruction, ;; which makes it more likely that the linker will be able to word ;; align them when relaxing. ;; The following description models the SH4 pipeline using the DFA based ;; scheduler. The DFA based description is better way to model a ;; superscalar pipeline as compared to function unit reservation model. ;; 1. The function unit based model is oriented to describe at most one ;; unit reservation by each insn. It is difficult to model unit reservations ;; in multiple pipeline units by same insn. This can be done using DFA ;; based description. ;; 2. The execution performance of DFA based scheduler does not depend on ;; processor complexity. ;; 3. Writing all unit reservations for an instruction class is a more natural ;; description of the pipeline and makes the interface to the hazard ;; recognizer simpler than the old function unit based model. ;; 4. The DFA model is richer and is a part of greater overall framework ;; of RCSP. ;; Two automata are defined to reduce number of states ;; which a single large automaton will have. (Factoring) (define_automaton "inst_pipeline,fpu_pipe") ;; This unit is basically the decode unit of the processor. ;; Since SH4 is a dual issue machine,it is as if there are two ;; units so that any insn can be processed by either one ;; of the decoding unit. (define_cpu_unit "pipe_01,pipe_02" "inst_pipeline") ;; The fixed point arithmetic calculator(?? EX Unit). (define_cpu_unit "int" "inst_pipeline") ;; f1_1 and f1_2 are floating point units.Actually there is ;; a f1 unit which can overlap with other f1 unit but ;; not another F1 unit.It is as though there were two ;; f1 units. (define_cpu_unit "f1_1,f1_2" "fpu_pipe") ;; The floating point units (except FS - F2 always precedes it.) (define_cpu_unit "F0,F1,F2,F3" "fpu_pipe") ;; This is basically the MA unit of SH4 ;; used in LOAD/STORE pipeline. (define_cpu_unit "memory" "inst_pipeline") ;; However, there are LS group insns that don't use it, even ones that ;; complete in 0 cycles. So we use an extra unit for the issue of LS insns. (define_cpu_unit "load_store" "inst_pipeline") ;; The address calculator used for branch instructions. ;; This will be reserved after "issue" of branch instructions ;; and this is to make sure that no two branch instructions ;; can be issued in parallel. (define_cpu_unit "pcr_addrcalc" "inst_pipeline") ;; ---------------------------------------------------- ;; This reservation is to simplify the dual issue description. (define_reservation "issue" "pipe_01|pipe_02") ;; This is to express the locking of D stage. ;; Note that the issue of a CO group insn also effectively locks the D stage. (define_reservation "d_lock" "pipe_01+pipe_02") ;; Every FE instruction but fipr / ftrv starts with issue and this. (define_reservation "F01" "F0+F1") ;; This is to simplify description where F1,F2,FS ;; are used simultaneously. (define_reservation "fpu" "F1+F2") ;; This is to highlight the fact that f1 ;; cannot overlap with F1. (exclusion_set "f1_1,f1_2" "F1") (define_insn_reservation "nil" 0 (eq_attr "type" "nil") "nothing") ;; Although reg moves have a latency of zero ;; we need to highlight that they use D stage ;; for one cycle. ;; Group: MT (define_insn_reservation "reg_mov" 0 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "move")) "issue") ;; Group: LS (define_insn_reservation "freg_mov" 0 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "fmove")) "issue+load_store") ;; We don't model all pipeline stages; we model the issue ('D') stage ;; inasmuch as we allow only two instructions to issue simultaneously, ;; and CO instructions prevent any simultaneous issue of another instruction. ;; (This uses pipe_01 and pipe_02). ;; Double issue of EX insns is prevented by using the int unit in the EX stage. ;; Double issue of EX / BR insns is prevented by using the int unit / ;; pcr_addrcalc unit in the EX stage. ;; Double issue of BR / LS instructions is prevented by using the ;; pcr_addrcalc / load_store unit in the issue cycle. ;; Double issue of FE instructions is prevented by using F0 in the first ;; pipeline stage after the first D stage. ;; There is no need to describe the [ES]X / [MN]A / S stages after a D stage ;; (except in the cases outlined above), nor to describe the FS stage after ;; the F2 stage. ;; Other MT group instructions(1 step operations) ;; Group: MT ;; Latency: 1 ;; Issue Rate: 1 (define_insn_reservation "mt" 1 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "mt_group")) "issue") ;; Fixed Point Arithmetic Instructions(1 step operations) ;; Group: EX ;; Latency: 1 ;; Issue Rate: 1 (define_insn_reservation "sh4_simple_arith" 1 (and (eq_attr "pipe_model" "sh4") (eq_attr "insn_class" "ex_group")) "issue,int") ;; Load and store instructions have no alignment peculiarities for the SH4, ;; but they use the load-store unit, which they share with the fmove type ;; insns (fldi[01]; fmov frn,frm; flds; fsts; fabs; fneg) . ;; Loads have a latency of two. ;; However, call insns can only paired with a preceding insn, and have ;; a delay slot, so that we want two more insns to be scheduled between the ;; load of the function address and the call. This is equivalent to a ;; latency of three. ;; ADJUST_COST can only properly handle reductions of the cost, so we ;; use a latency of three here, which gets multiplied by 10 to yield 30. ;; We only do this for SImode loads of general registers, to make the work ;; for ADJUST_COST easier. ;; Load Store instructions. (MOV.[BWL]@(d,GBR) ;; Group: LS ;; Latency: 2 ;; Issue Rate: 1 (define_insn_reservation "sh4_load" 2 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "load,pcload")) "issue+load_store,nothing,memory") ;; calls / sfuncs need an extra instruction for their delay slot. ;; Moreover, estimating the latency for SImode loads as 3 will also allow ;; adjust_cost to meaningfully bump it back up to 3 if they load the shift ;; count of a dynamic shift. (define_insn_reservation "sh4_load_si" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "load_si,pcload_si")) "issue+load_store,nothing,memory") ;; (define_bypass 2 "sh4_load_si" "!sh4_call") ;; The load latency is upped to three higher if the dependent insn does ;; double precision computation. We want the 'default' latency to reflect ;; that increased latency because otherwise the insn priorities won't ;; allow proper scheduling. (define_insn_reservation "sh4_fload" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "fload,pcfload")) "issue+load_store,nothing,memory") ;; (define_bypass 2 "sh4_fload" "!") (define_insn_reservation "sh4_store" 1 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "store,fstore")) "issue+load_store,nothing,memory") (define_insn_reservation "mac_mem" 1 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "mac_mem")) "d_lock,nothing,memory") ;; Load Store instructions. ;; Group: LS ;; Latency: 1 ;; Issue Rate: 1 (define_insn_reservation "sh4_gp_fpul" 1 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "gp_fpul")) "issue+load_store") ;; Load Store instructions. ;; Group: LS ;; Latency: 3 ;; Issue Rate: 1 (define_insn_reservation "sh4_fpul_gp" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "fpul_gp")) "issue+load_store") ;; Branch (BF,BF/S,BT,BT/S,BRA) ;; Group: BR ;; Latency when taken: 2 (or 1) ;; Issue Rate: 1 ;; The latency is 1 when displacement is 0. ;; We can't really do much with the latency, even if we could express it, ;; but the pairing restrictions are useful to take into account. ;; ??? If the branch is likely, we might want to fill the delay slot; ;; if the branch is likely, but not very likely, should we pretend to use ;; a resource that CO instructions use, to get a pairable delay slot insn? (define_insn_reservation "sh4_branch" 1 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "cbranch,jump")) "issue+pcr_addrcalc") ;; Branch Far (JMP,RTS,BRAF) ;; Group: CO ;; Latency: 3 ;; Issue Rate: 2 ;; ??? Scheduling happens before branch shortening, and hence jmp and braf ;; can't be distinguished from bra for the "jump" pattern. (define_insn_reservation "sh4_return" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "return,jump_ind")) "d_lock*2") ;; RTE ;; Group: CO ;; Latency: 5 ;; Issue Rate: 5 ;; this instruction can be executed in any of the pipelines ;; and blocks the pipeline for next 4 stages. (define_insn_reservation "sh4_return_from_exp" 5 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "rte")) "d_lock*5") ;; OCBP, OCBWB ;; Group: CO ;; Latency: 1-5 ;; Issue Rate: 1 ;; cwb is used for the sequence ;; ocbwb @%0 ;; extu.w %0,%2 ;; or %1,%2 ;; mov.l %0,@%2 ;; ocbwb on its own would be "d_lock,nothing,memory*5" (define_insn_reservation "ocbwb" 6 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "cwb")) "d_lock*2,(d_lock+memory)*3,issue+load_store+memory,memory*2") ;; LDS to PR,JSR ;; Group: CO ;; Latency: 3 ;; Issue Rate: 2 ;; The SX stage is blocked for last 2 cycles. ;; OTOH, the only time that has an effect for insns generated by the compiler ;; is when lds to PR is followed by sts from PR - and that is highly unlikely - ;; or when we are doing a function call - and we don't do inter-function ;; scheduling. For the function call case, it's really best that we end with ;; something that models an rts. (define_insn_reservation "sh4_lds_to_pr" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "prset") ) "d_lock*2") ;; calls introduce a longisch delay that is likely to flush the pipelines ;; of the caller's instructions. Ordinary functions tend to end with a ;; load to restore a register (in the delay slot of rts), while sfuncs ;; tend to end with an EX or MT insn. But that is not actually relevant, ;; since there are no instructions that contend for memory access early. ;; We could, of course, provide exact scheduling information for specific ;; sfuncs, if that should prove useful. (define_insn_reservation "sh4_call" 16 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "call,sfunc")) "d_lock*16") ;; LDS.L to PR ;; Group: CO ;; Latency: 3 ;; Issue Rate: 2 ;; The SX unit is blocked for last 2 cycles. (define_insn_reservation "ldsmem_to_pr" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "pload")) "d_lock*2") ;; STS from PR ;; Group: CO ;; Latency: 2 ;; Issue Rate: 2 ;; The SX unit in second and third cycles. (define_insn_reservation "sts_from_pr" 2 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "prget")) "d_lock*2") ;; STS.L from PR ;; Group: CO ;; Latency: 2 ;; Issue Rate: 2 (define_insn_reservation "sh4_prstore_mem" 2 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "pstore")) "d_lock*2,nothing,memory") ;; LDS to FPSCR ;; Group: CO ;; Latency: 4 ;; Issue Rate: 1 ;; F1 is blocked for last three cycles. (define_insn_reservation "fpscr_load" 4 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "gp_fpscr")) "d_lock,nothing,F1*3") ;; LDS.L to FPSCR ;; Group: CO ;; Latency: 1 / 4 ;; Latency to update Rn is 1 and latency to update FPSCR is 4 ;; Issue Rate: 1 ;; F1 is blocked for last three cycles. (define_insn_reservation "fpscr_load_mem" 4 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "mem_fpscr")) "d_lock,nothing,(F1+memory),F1*2") ;; Fixed point multiplication (DMULS.L DMULU.L MUL.L MULS.W,MULU.W) ;; Group: CO ;; Latency: 4 / 4 ;; Issue Rate: 2 (define_insn_reservation "multi" 4 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "smpy,dmpy")) "d_lock,(d_lock+f1_1),(f1_1|f1_2)*3,F2") ;; Fixed STS from, and LDS to MACL / MACH ;; Group: CO ;; Latency: 3 ;; Issue Rate: 1 (define_insn_reservation "sh4_mac_gp" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "mac_gp,gp_mac,mem_mac")) "d_lock") ;; Single precision floating point computation FCMP/EQ, ;; FCMP/GT, FADD, FLOAT, FMAC, FMUL, FSUB, FTRC, FRCHG, FSCHG ;; Group: FE ;; Latency: 3/4 ;; Issue Rate: 1 (define_insn_reservation "fp_arith" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "fp,fp_cmp")) "issue,F01,F2") ;; We don't model the resource usage of this exactly because that would ;; introduce a bogus latency. (define_insn_reservation "sh4_fpscr_toggle" 1 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "fpscr_toggle")) "issue") (define_insn_reservation "fp_arith_ftrc" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "ftrc_s")) "issue,F01,F2") (define_bypass 1 "fp_arith_ftrc" "sh4_fpul_gp") ;; Single Precision FDIV/SQRT ;; Group: FE ;; Latency: 12/13 (FDIV); 11/12 (FSQRT) ;; Issue Rate: 1 ;; We describe fdiv here; fsqrt is actually one cycle faster. (define_insn_reservation "fp_div" 12 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "fdiv")) "issue,F01+F3,F2+F3,F3*7,F1+F3,F2") ;; Double Precision floating point computation ;; (FCNVDS, FCNVSD, FLOAT, FTRC) ;; Group: FE ;; Latency: (3,4)/5 ;; Issue Rate: 1 (define_insn_reservation "dp_float" 4 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "dfp_conv")) "issue,F01,F1+F2,F2") ;; Double-precision floating-point (FADD,FMUL,FSUB) ;; Group: FE ;; Latency: (7,8)/9 ;; Issue Rate: 1 (define_insn_reservation "fp_double_arith" 8 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "dfp_arith,dfp_mul")) "issue,F01,F1+F2,fpu*4,F2") ;; Double-precision FCMP (FCMP/EQ,FCMP/GT) ;; Group: CO ;; Latency: 3/5 ;; Issue Rate: 2 (define_insn_reservation "fp_double_cmp" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "dfp_cmp")) "d_lock,(d_lock+F01),F1+F2,F2") ;; Double precision FDIV/SQRT ;; Group: FE ;; Latency: (24,25)/26 ;; Issue Rate: 1 (define_insn_reservation "dp_div" 25 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "dfdiv")) "issue,F01+F3,F1+F2+F3,F2+F3,F3*16,F1+F3,(fpu+F3)*2,F2") ;; Use the branch-not-taken case to model arith3 insns. For the branch taken ;; case, we'd get a d_lock instead of issue at the end. (define_insn_reservation "arith3" 3 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "arith3")) "issue,d_lock+pcr_addrcalc,issue") ;; arith3b insns schedule the same no matter if the branch is taken or not. (define_insn_reservation "arith3b" 2 (and (eq_attr "pipe_model" "sh4") (eq_attr "type" "arith3")) "issue,d_lock+pcr_addrcalc")