* [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
@ 2015-01-15 10:48 James Greenhalgh
2015-01-16 10:39 ` Marcus Shawcroft
0 siblings, 1 reply; 14+ messages in thread
From: James Greenhalgh @ 2015-01-15 10:48 UTC (permalink / raw)
To: gcc-patches; +Cc: marcus.shawcroft, richard.earnshaw
[-- Attachment #1: Type: text/plain, Size: 883 bytes --]
Hi,
This patch implements a new scheduler model for the ARM Cortex-A57
processor.
This model results in better code generation for the Cortex-A57 with a
more mixed blend of instruction types, particularly when scheduling for
the new instructions introduced in ARMv8-A.
I haven't yet wired it up for the ARM back end, though it is likely
to also be appropriate there. As is convention for these shared
models I've added the .md file to config/arm/cortex-a57.md
I've bootstrapped and tested the patch on an aarch64-none-linux-gnu
target with no regressions.
Is this patch OK for trunk?
Thanks,
James
---
2015-01-15 James Greenhalgh <james.greenhalgh@arm.com>
* config/arm/cortex-a57.md: New.
* config/aarch64/aarch64.md: Include it.
* config/aarch64/aarch64-cores.def (cortex-a57): Tune for it.
* config/aarch64/aarch64-tune.md: Regenerate.
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-AArch64-Add-a-new-scheduling-description-for-the-ARM.patch --]
[-- Type: text/x-patch; name=0001-AArch64-Add-a-new-scheduling-description-for-the-ARM.patch, Size: 31834 bytes --]
diff --git a/gcc/config/aarch64/aarch64-cores.def b/gcc/config/aarch64/aarch64-cores.def
index 18f5c48..e30beb0 100644
--- a/gcc/config/aarch64/aarch64-cores.def
+++ b/gcc/config/aarch64/aarch64-cores.def
@@ -35,7 +35,7 @@
/* V8 Architecture Processors. */
AARCH64_CORE("cortex-a53", cortexa53, cortexa53, 8, AARCH64_FL_FOR_ARCH8 | AARCH64_FL_CRC, cortexa53)
-AARCH64_CORE("cortex-a57", cortexa15, cortexa15, 8, AARCH64_FL_FOR_ARCH8 | AARCH64_FL_CRC, cortexa57)
+AARCH64_CORE("cortex-a57", cortexa57, cortexa57, 8, AARCH64_FL_FOR_ARCH8 | AARCH64_FL_CRC, cortexa57)
AARCH64_CORE("thunderx", thunderx, thunderx, 8, AARCH64_FL_FOR_ARCH8 | AARCH64_FL_CRC | AARCH64_FL_CRYPTO, thunderx)
/* V8 big.LITTLE implementations. */
diff --git a/gcc/config/aarch64/aarch64-tune.md b/gcc/config/aarch64/aarch64-tune.md
index c717ea848bcd4a4d9bce8b338b3643438ec7474e..f8455006e194c6b3261fc86a71ca8a88bd177f2a 100644
--- a/gcc/config/aarch64/aarch64-tune.md
+++ b/gcc/config/aarch64/aarch64-tune.md
@@ -1,5 +1,5 @@
;; -*- buffer-read-only: t -*-
;; Generated automatically by gentune.sh from aarch64-cores.def
(define_attr "tune"
- "cortexa53,cortexa15,thunderx,cortexa57cortexa53"
+ "cortexa53,cortexa57,thunderx,cortexa57cortexa53"
(const (symbol_ref "((enum attr_tune) aarch64_tune)")))
diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
index fc729515910718a6e4792de92b543d4e2524d568..6c9d76760dfa86e8649789776a21b771848d727c 100644
--- a/gcc/config/aarch64/aarch64.md
+++ b/gcc/config/aarch64/aarch64.md
@@ -188,7 +188,7 @@ (define_attr "enabled" "no,yes"
;; Scheduling
(include "../arm/cortex-a53.md")
-(include "../arm/cortex-a15.md")
+(include "../arm/cortex-a57.md")
(include "thunderx.md")
;; -------------------------------------------------------------------
diff --git a/gcc/config/arm/cortex-a57.md b/gcc/config/arm/cortex-a57.md
index ...c9782f20e9534f97bdb5e9f180f7f61010f930f0 100644
--- a/gcc/config/arm/cortex-a57.md
+++ b/gcc/config/arm/cortex-a57.md
@@ -0,0 +1,797 @@
+;; ARM Cortex-A57 pipeline description
+;; Copyright (C) 2014-2015 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 "cortex_a57")
+
+(define_attr "cortex_a57_neon_type"
+ "neon_abd, neon_abd_q, neon_arith_acc, neon_arith_acc_q,
+ neon_arith_basic, neon_arith_complex,
+ neon_reduc_add_acc, neon_multiply, neon_multiply_q,
+ neon_multiply_long, neon_mla, neon_mla_q, neon_mla_long,
+ neon_sat_mla_long, neon_shift_acc, neon_shift_imm_basic,
+ neon_shift_imm_complex,
+ neon_shift_reg_basic, neon_shift_reg_basic_q, neon_shift_reg_complex,
+ neon_shift_reg_complex_q, neon_fp_negabs, neon_fp_arith,
+ neon_fp_arith_q, neon_fp_reductions_q, neon_fp_cvt_int,
+ neon_fp_cvt_int_q, neon_fp_cvt16, neon_fp_minmax, neon_fp_mul,
+ neon_fp_mul_q, neon_fp_mla, neon_fp_mla_q, neon_fp_recpe_rsqrte,
+ neon_fp_recpe_rsqrte_q, neon_fp_recps_rsqrts, neon_fp_recps_rsqrts_q,
+ neon_bitops, neon_bitops_q, neon_from_gp,
+ neon_from_gp_q, neon_move, neon_tbl3_tbl4, neon_zip_q, neon_to_gp,
+ neon_load_a, neon_load_b, neon_load_c, neon_load_d, neon_load_e,
+ neon_load_f, neon_store_a, neon_store_b, neon_store_complex,
+ unknown"
+ (cond [
+ (eq_attr "type" "neon_abd, neon_abd_long")
+ (const_string "neon_abd")
+ (eq_attr "type" "neon_abd_q")
+ (const_string "neon_abd_q")
+ (eq_attr "type" "neon_arith_acc, neon_reduc_add_acc,\
+ neon_reduc_add_acc_q")
+ (const_string "neon_arith_acc")
+ (eq_attr "type" "neon_arith_acc_q")
+ (const_string "neon_arith_acc_q")
+ (eq_attr "type" "neon_add, neon_add_q, neon_add_long,\
+ neon_add_widen, neon_neg, neon_neg_q,\
+ neon_reduc_add, neon_reduc_add_q,\
+ neon_reduc_add_long, neon_sub, neon_sub_q,\
+ neon_sub_long, neon_sub_widen, neon_logic,\
+ neon_logic_q, neon_tst, neon_tst_q")
+ (const_string "neon_arith_basic")
+ (eq_attr "type" "neon_abs, neon_abs_q, neon_add_halve_narrow_q,\
+ neon_add_halve, neon_add_halve_q,\
+ neon_sub_halve, neon_sub_halve_q, neon_qabs,\
+ neon_qabs_q, neon_qadd, neon_qadd_q, neon_qneg,\
+ neon_qneg_q, neon_qsub, neon_qsub_q,\
+ neon_sub_halve_narrow_q,\
+ neon_compare, neon_compare_q,\
+ neon_compare_zero, neon_compare_zero_q,\
+ neon_minmax, neon_minmax_q, neon_reduc_minmax,\
+ neon_reduc_minmax_q")
+ (const_string "neon_arith_complex")
+
+ (eq_attr "type" "neon_mul_b, neon_mul_h, neon_mul_s,\
+ neon_mul_h_scalar, neon_mul_s_scalar,\
+ neon_sat_mul_b, neon_sat_mul_h,\
+ neon_sat_mul_s, neon_sat_mul_h_scalar,\
+ neon_sat_mul_s_scalar,\
+ neon_mul_b_long, neon_mul_h_long,\
+ neon_mul_s_long, neon_mul_d_long,\
+ neon_mul_h_scalar_long, neon_mul_s_scalar_long,\
+ neon_sat_mul_b_long, neon_sat_mul_h_long,\
+ neon_sat_mul_s_long, neon_sat_mul_h_scalar_long,\
+ neon_sat_mul_s_scalar_long")
+ (const_string "neon_multiply")
+ (eq_attr "type" "neon_mul_b_q, neon_mul_h_q, neon_mul_s_q,\
+ neon_mul_h_scalar_q, neon_mul_s_scalar_q,\
+ neon_sat_mul_b_q, neon_sat_mul_h_q,\
+ neon_sat_mul_s_q, neon_sat_mul_h_scalar_q,\
+ neon_sat_mul_s_scalar_q")
+ (const_string "neon_multiply_q")
+ (eq_attr "type" "neon_mla_b, neon_mla_h, neon_mla_s,\
+ neon_mla_h_scalar, neon_mla_s_scalar,\
+ neon_mla_b_long, neon_mla_h_long,\
+ neon_mla_s_long,\
+ neon_mla_h_scalar_long, neon_mla_s_scalar_long")
+ (const_string "neon_mla")
+ (eq_attr "type" "neon_mla_b_q, neon_mla_h_q, neon_mla_s_q,\
+ neon_mla_h_scalar_q, neon_mla_s_scalar_q")
+ (const_string "neon_mla_q")
+ (eq_attr "type" "neon_sat_mla_b_long, neon_sat_mla_h_long,\
+ neon_sat_mla_s_long, neon_sat_mla_h_scalar_long,\
+ neon_sat_mla_s_scalar_long")
+ (const_string "neon_sat_mla_long")
+
+ (eq_attr "type" "neon_shift_acc, neon_shift_acc_q")
+ (const_string "neon_shift_acc")
+ (eq_attr "type" "neon_shift_imm, neon_shift_imm_q,\
+ neon_shift_imm_narrow_q, neon_shift_imm_long")
+ (const_string "neon_shift_imm_basic")
+ (eq_attr "type" "neon_sat_shift_imm, neon_sat_shift_imm_q,\
+ neon_sat_shift_imm_narrow_q")
+ (const_string "neon_shift_imm_complex")
+ (eq_attr "type" "neon_shift_reg")
+ (const_string "neon_shift_reg_basic")
+ (eq_attr "type" "neon_shift_reg_q")
+ (const_string "neon_shift_reg_basic_q")
+ (eq_attr "type" "neon_sat_shift_reg")
+ (const_string "neon_shift_reg_complex")
+ (eq_attr "type" "neon_sat_shift_reg_q")
+ (const_string "neon_shift_reg_complex_q")
+
+ (eq_attr "type" "neon_fp_neg_s, neon_fp_neg_s_q,\
+ neon_fp_abs_s, neon_fp_abs_s_q,\
+ neon_fp_neg_d, neon_fp_neg_d_q,\
+ neon_fp_abs_d, neon_fp_abs_d_q")
+ (const_string "neon_fp_negabs")
+ (eq_attr "type" "neon_fp_addsub_s, neon_fp_abd_s,\
+ neon_fp_reduc_add_s, neon_fp_compare_s,\
+ neon_fp_minmax_s, neon_fp_round_s,\
+ neon_fp_addsub_d, neon_fp_abd_d,\
+ neon_fp_reduc_add_d, neon_fp_compare_d,\
+ neon_fp_minmax_d, neon_fp_round_d,\
+ neon_fp_reduc_minmax_s, neon_fp_reduc_minmax_d")
+ (const_string "neon_fp_arith")
+ (eq_attr "type" "neon_fp_addsub_s_q, neon_fp_abd_s_q,\
+ neon_fp_reduc_add_s_q, neon_fp_compare_s_q,\
+ neon_fp_minmax_s_q, neon_fp_round_s_q,\
+ neon_fp_addsub_d_q, neon_fp_abd_d_q,\
+ neon_fp_reduc_add_d_q, neon_fp_compare_d_q,\
+ neon_fp_minmax_d_q, neon_fp_round_d_q")
+ (const_string "neon_fp_arith_q")
+ (eq_attr "type" "neon_fp_reduc_minmax_s_q,\
+ neon_fp_reduc_minmax_d_q,\
+ neon_fp_reduc_add_s_q, neon_fp_reduc_add_d_q")
+ (const_string "neon_fp_reductions_q")
+ (eq_attr "type" "neon_fp_to_int_s, neon_int_to_fp_s,\
+ neon_fp_to_int_d, neon_int_to_fp_d")
+ (const_string "neon_fp_cvt_int")
+ (eq_attr "type" "neon_fp_to_int_s_q, neon_int_to_fp_s_q,\
+ neon_fp_to_int_d_q, neon_int_to_fp_d_q")
+ (const_string "neon_fp_cvt_int_q")
+ (eq_attr "type" "neon_fp_cvt_narrow_s_q, neon_fp_cvt_widen_h")
+ (const_string "neon_fp_cvt16")
+ (eq_attr "type" "neon_fp_mul_s, neon_fp_mul_s_scalar,\
+ neon_fp_mul_d")
+ (const_string "neon_fp_mul")
+ (eq_attr "type" "neon_fp_mul_s_q, neon_fp_mul_s_scalar_q,\
+ neon_fp_mul_d_q, neon_fp_mul_d_scalar_q")
+ (const_string "neon_fp_mul_q")
+ (eq_attr "type" "neon_fp_mla_s, neon_fp_mla_s_scalar,\
+ neon_fp_mla_d")
+ (const_string "neon_fp_mla")
+ (eq_attr "type" "neon_fp_mla_s_q, neon_fp_mla_s_scalar_q,
+ neon_fp_mla_d_q, neon_fp_mla_d_scalar_q")
+ (const_string "neon_fp_mla_q")
+ (eq_attr "type" "neon_fp_recpe_s, neon_fp_rsqrte_s,\
+ neon_fp_recpx_s,\
+ neon_fp_recpe_d, neon_fp_rsqrte_d,\
+ neon_fp_recpx_d")
+ (const_string "neon_fp_recpe_rsqrte")
+ (eq_attr "type" "neon_fp_recpe_s_q, neon_fp_rsqrte_s_q,\
+ neon_fp_recpx_s_q,\
+ neon_fp_recpe_d_q, neon_fp_rsqrte_d_q,\
+ neon_fp_recpx_d_q")
+ (const_string "neon_fp_recpe_rsqrte_q")
+ (eq_attr "type" "neon_fp_recps_s, neon_fp_rsqrts_s,\
+ neon_fp_recps_d, neon_fp_rsqrts_d")
+ (const_string "neon_fp_recps_rsqrts")
+ (eq_attr "type" "neon_fp_recps_s_q, neon_fp_rsqrts_s_q,\
+ neon_fp_recps_d_q, neon_fp_rsqrts_d_q")
+ (const_string "neon_fp_recps_rsqrts_q")
+ (eq_attr "type" "neon_bsl, neon_cls, neon_cnt,\
+ neon_rev, neon_permute, neon_rbit,\
+ neon_tbl1, neon_tbl2, neon_zip,\
+ neon_dup, neon_dup_q, neon_ext, neon_ext_q,\
+ neon_move, neon_move_q, neon_move_narrow_q")
+ (const_string "neon_bitops")
+ (eq_attr "type" "neon_bsl_q, neon_cls_q, neon_cnt_q,\
+ neon_rev_q, neon_permute_q, neon_rbit_q")
+ (const_string "neon_bitops_q")
+ (eq_attr "type" "neon_from_gp,f_mcr,f_mcrr")
+ (const_string "neon_from_gp")
+ (eq_attr "type" "neon_from_gp_q")
+ (const_string "neon_from_gp_q")
+ (eq_attr "type" "neon_tbl3, neon_tbl4")
+ (const_string "neon_tbl3_tbl4")
+ (eq_attr "type" "neon_zip_q")
+ (const_string "neon_zip_q")
+ (eq_attr "type" "neon_to_gp, neon_to_gp_q,f_mrc,f_mrrc")
+ (const_string "neon_to_gp")
+
+ (eq_attr "type" "f_loads, f_loadd,\
+ neon_load1_1reg, neon_load1_1reg_q,\
+ neon_load1_2reg, neon_load1_2reg_q")
+ (const_string "neon_load_a")
+ (eq_attr "type" "neon_load1_3reg, neon_load1_3reg_q,\
+ neon_load1_4reg, neon_load1_4reg_q")
+ (const_string "neon_load_b")
+ (eq_attr "type" "neon_load1_one_lane, neon_load1_one_lane_q,\
+ neon_load1_all_lanes, neon_load1_all_lanes_q,\
+ neon_load2_2reg, neon_load2_2reg_q,\
+ neon_load2_all_lanes, neon_load2_all_lanes_q")
+ (const_string "neon_load_c")
+ (eq_attr "type" "neon_load2_4reg, neon_load2_4reg_q,\
+ neon_load3_3reg, neon_load3_3reg_q,\
+ neon_load3_one_lane, neon_load3_one_lane_q,\
+ neon_load4_4reg, neon_load4_4reg_q")
+ (const_string "neon_load_d")
+ (eq_attr "type" "neon_load2_one_lane, neon_load2_one_lane_q,\
+ neon_load3_all_lanes, neon_load3_all_lanes_q,\
+ neon_load4_all_lanes, neon_load4_all_lanes_q")
+ (const_string "neon_load_e")
+ (eq_attr "type" "neon_load4_one_lane, neon_load4_one_lane_q")
+ (const_string "neon_load_f")
+
+ (eq_attr "type" "f_stores, f_stored,\
+ neon_store1_1reg")
+ (const_string "neon_store_a")
+ (eq_attr "type" "neon_store1_2reg, neon_store1_1reg_q")
+ (const_string "neon_store_b")
+ (eq_attr "type" "neon_store1_3reg, neon_store1_3reg_q,\
+ neon_store3_3reg, neon_store3_3reg_q,\
+ neon_store2_4reg, neon_store2_4reg_q,\
+ neon_store4_4reg, neon_store4_4reg_q,\
+ neon_store2_2reg, neon_store2_2reg_q,\
+ neon_store3_one_lane, neon_store3_one_lane_q,\
+ neon_store4_one_lane, neon_store4_one_lane_q,\
+ neon_store1_4reg, neon_store1_4reg_q,\
+ neon_store1_one_lane, neon_store1_one_lane_q,\
+ neon_store2_one_lane, neon_store2_one_lane_q")
+ (const_string "neon_store_complex")]
+ (const_string "unknown")))
+
+;; The Cortex-A57 core is modelled as a triple issue pipeline that has
+;; the following functional units.
+;; 1. Two pipelines for integer operations: SX1, SX2
+
+(define_cpu_unit "ca57_sx1_issue" "cortex_a57")
+(define_reservation "ca57_sx1" "ca57_sx1_issue")
+
+(define_cpu_unit "ca57_sx2_issue" "cortex_a57")
+(define_reservation "ca57_sx2" "ca57_sx2_issue")
+
+;; 2. One pipeline for complex integer operations: MX
+
+(define_cpu_unit "ca57_mx_issue"
+ "cortex_a57")
+(define_reservation "ca57_mx" "ca57_mx_issue")
+(define_reservation "ca57_mx_block" "ca57_mx_issue")
+
+;; 3. Two asymmetric pipelines for Neon and FP operations: CX1, CX2
+(define_automaton "cortex_a57_cx")
+
+(define_cpu_unit "ca57_cx1_issue"
+ "cortex_a57_cx")
+(define_cpu_unit "ca57_cx2_issue"
+ "cortex_a57_cx")
+
+(define_reservation "ca57_cx1" "ca57_cx1_issue")
+
+(define_reservation "ca57_cx2" "ca57_cx2_issue")
+(define_reservation "ca57_cx2_block" "ca57_cx2_issue*2")
+
+;; 4. One pipeline for branch operations: BX
+
+(define_cpu_unit "ca57_bx_issue" "cortex_a57")
+(define_reservation "ca57_bx" "ca57_bx_issue")
+
+;; 5. Two pipelines for load and store operations: LS1, LS2. The most
+;; valuable thing we can do is force a structural hazard to split
+;; up loads/stores.
+
+(define_cpu_unit "ca57_ls_issue" "cortex_a57")
+(define_cpu_unit "ca57_ldr, ca57_str" "cortex_a57")
+(define_reservation "ca57_load_model" "ca57_ls_issue,ca57_ldr*2")
+(define_reservation "ca57_store_model" "ca57_ls_issue,ca57_str")
+
+;; Block all issue queues.
+
+(define_reservation "ca57_block" "ca57_cx1_issue + ca57_cx2_issue
+ + ca57_mx_issue + ca57_sx1_issue
+ + ca57_sx2_issue + ca57_ls_issue")
+
+;; Simple Execution Unit:
+;;
+;; Simple ALU without shift
+(define_insn_reservation "cortex_a57_alu" 2
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "alu_imm,alus_imm,logic_imm,logics_imm,\
+ alu_sreg,alus_sreg,logic_reg,logics_reg,\
+ adc_imm,adcs_imm,adc_reg,adcs_reg,\
+ adr,bfm,clz,rbit,rev,alu_dsp_reg,\
+ shift_imm,shift_reg,\
+ mov_imm,mov_reg,\
+ mvn_imm,mvn_reg,\
+ mrs,multiple,no_insn"))
+ "ca57_sx1|ca57_sx2")
+
+;; ALU ops with immediate shift
+(define_insn_reservation "cortex_a57_alu_shift" 3
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "extend,\
+ alu_shift_imm,alus_shift_imm,\
+ crc,logic_shift_imm,logics_shift_imm,\
+ mov_shift,mvn_shift"))
+ "ca57_mx")
+
+;; Multi-Cycle Execution Unit:
+;;
+;; ALU ops with register controlled shift
+(define_insn_reservation "cortex_a57_alu_shift_reg" 3
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "alu_shift_reg,alus_shift_reg,\
+ logic_shift_reg,logics_shift_reg,\
+ mov_shift_reg,mvn_shift_reg"))
+ "ca57_mx")
+
+;; All multiplies
+;; TODO: AArch32 and AArch64 have different behaviour
+(define_insn_reservation "cortex_a57_mult32" 3
+ (and (eq_attr "tune" "cortexa57")
+ (ior (eq_attr "mul32" "yes")
+ (eq_attr "mul64" "yes")))
+ "ca57_mx")
+
+;; Integer divide
+(define_insn_reservation "cortex_a57_div" 10
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "udiv,sdiv"))
+ "ca57_mx_issue,ca57_mx_block*3")
+
+;; Block all issue pipes for a cycle
+(define_insn_reservation "cortex_a57_block" 1
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "block"))
+ "ca57_block")
+
+;; Branch execution Unit
+;;
+;; Branches take one issue slot.
+;; No latency as there is no result
+(define_insn_reservation "cortex_a57_branch" 0
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "branch"))
+ "ca57_bx")
+
+;; Load-store execution Unit
+;;
+;; Loads of up to two words.
+(define_insn_reservation "cortex_a57_load1" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "load_byte,load1,load2"))
+ "ca57_load_model")
+
+;; Loads of three or four words.
+(define_insn_reservation "cortex_a57_load3" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "load3,load4"))
+ "ca57_ls_issue*2,ca57_load_model")
+
+;; Stores of up to two words.
+(define_insn_reservation "cortex_a57_store1" 0
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "store1,store2"))
+ "ca57_store_model")
+
+;; Stores of three or four words.
+(define_insn_reservation "cortex_a57_store3" 0
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "store3,store4"))
+ "ca57_ls_issue*2,ca57_store_model")
+
+;; Advanced SIMD Unit - Integer Arithmetic Instructions.
+
+(define_insn_reservation "cortex_a57_neon_abd" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_abd"))
+ "ca57_cx1|ca57_cx2")
+
+(define_insn_reservation "cortex_a57_neon_abd_q" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_abd_q"))
+ "ca57_cx1+ca57_cx2")
+
+(define_insn_reservation "cortex_a57_neon_aba" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_arith_acc"))
+ "ca57_cx2")
+
+(define_insn_reservation "cortex_a57_neon_aba_q" 8
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_arith_acc_q"))
+ "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
+
+(define_insn_reservation "cortex_a57_neon_arith_basic" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_arith_basic"))
+ "ca57_cx1|ca57_cx2")
+
+(define_insn_reservation "cortex_a57_neon_arith_complex" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_arith_complex"))
+ "ca57_cx1|ca57_cx2")
+
+;; Integer Multiply Instructions.
+
+(define_insn_reservation "cortex_a57_neon_multiply" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_multiply"))
+ "ca57_cx1")
+
+(define_insn_reservation "cortex_a57_neon_multiply_q" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_multiply_q"))
+ "ca57_cx1+(ca57_cx1_issue,ca57_cx1)")
+
+(define_insn_reservation "cortex_a57_neon_mla" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_mla"))
+ "ca57_cx1")
+
+(define_insn_reservation "cortex_a57_neon_mla_q" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_mla_q"))
+ "ca57_cx1+(ca57_cx1_issue,ca57_cx1)")
+
+(define_insn_reservation "cortex_a57_neon_sat_mla_long" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_sat_mla_long"))
+ "ca57_cx1")
+
+;; Integer Shift Instructions.
+
+(define_insn_reservation
+ "cortex_a57_neon_shift_acc" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_shift_acc"))
+ "ca57_cx2")
+
+(define_insn_reservation
+ "cortex_a57_neon_shift_imm_basic" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_shift_imm_basic"))
+ "ca57_cx2")
+
+(define_insn_reservation
+ "cortex_a57_neon_shift_imm_complex" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_shift_imm_complex"))
+ "ca57_cx2")
+
+(define_insn_reservation
+ "cortex_a57_neon_shift_reg_basic" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_shift_reg_basic"))
+ "ca57_cx2")
+
+(define_insn_reservation
+ "cortex_a57_neon_shift_reg_basic_q" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_shift_reg_basic_q"))
+ "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_shift_reg_complex" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_shift_reg_complex"))
+ "ca57_cx2")
+
+(define_insn_reservation
+ "cortex_a57_neon_shift_reg_complex_q" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_shift_reg_complex_q"))
+ "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
+
+;; Floating Point Instructions.
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_negabs" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_negabs"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_arith" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_arith"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_arith_q" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_arith_q"))
+ "(ca57_cx1+ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_reductions_q" 10
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_reductions_q"))
+ "(ca57_cx1+ca57_cx2),(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_cvt_int" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_cvt_int"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_cvt_int_q" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_cvt_int_q"))
+ "(ca57_cx1+ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_cvt16" 10
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_cvt16"))
+ "(ca57_cx1_issue+ca57_cx2_issue),(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_mul" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_mul"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_mul_q" 5
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_mul_q"))
+ "(ca57_cx1+ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_mla" 9
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_mla"))
+ "(ca57_cx1,ca57_cx1)|(ca57_cx2,ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_mla_q" 9
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_mla_q"))
+ "(ca57_cx1+ca57_cx2),(ca57_cx1,ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_recpe_rsqrte" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_recpe_rsqrte"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_recpe_rsqrte_q" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_recpe_rsqrte_q"))
+ "(ca57_cx1+ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_recps_rsqrts" 10
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_recps_rsqrts"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_fp_recps_rsqrts_q" 10
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_fp_recps_rsqrts_q"))
+ "(ca57_cx1+ca57_cx2)")
+
+;; Miscellaneous Instructions.
+
+(define_insn_reservation
+ "cortex_a57_neon_bitops" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_bitops"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_bitops_q" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_bitops_q"))
+ "(ca57_cx1+ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_from_gp" 9
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_from_gp"))
+ "(ca57_ls_issue+ca57_cx1_issue,ca57_cx1)
+ |(ca57_ls_issue+ca57_cx2_issue,ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_from_gp_q" 9
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_from_gp_q"))
+ "(ca57_ls_issue+ca57_cx1_issue,ca57_cx1)
+ +(ca57_ls_issue+ca57_cx2_issue,ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_tbl3_tbl4" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_tbl3_tbl4"))
+ "(ca57_cx1_issue,ca57_cx1)
+ +(ca57_cx2_issue,ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_zip_q" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_zip_q"))
+ "(ca57_cx1_issue,ca57_cx1)
+ +(ca57_cx2_issue,ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_to_gp" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_to_gp"))
+ "((ca57_ls_issue+ca57_sx1_issue),ca57_sx1)
+ |((ca57_ls_issue+ca57_sx2_issue),ca57_sx2)")
+
+;; Load Instructions.
+
+(define_insn_reservation
+ "cortex_a57_neon_load_a" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_load_a"))
+ "ca57_load_model")
+
+(define_insn_reservation
+ "cortex_a57_neon_load_b" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_load_b"))
+ "ca57_ls_issue,ca57_ls_issue+ca57_ldr,ca57_ldr*2")
+
+(define_insn_reservation
+ "cortex_a57_neon_load_c" 9
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_load_c"))
+ "ca57_load_model+(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_load_d" 11
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_load_d"))
+ "ca57_cx1_issue+ca57_cx2_issue,
+ ca57_ls_issue+ca57_ls_issue,ca57_ldr*2")
+
+(define_insn_reservation
+ "cortex_a57_neon_load_e" 9
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_load_e"))
+ "ca57_load_model+(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation
+ "cortex_a57_neon_load_f" 11
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_load_f"))
+ "ca57_cx1_issue+ca57_cx2_issue,
+ ca57_ls_issue+ca57_ls_issue,ca57_ldr*2")
+
+;; Store Instructions.
+
+(define_insn_reservation
+ "cortex_a57_neon_store_a" 0
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_store_a"))
+ "ca57_store_model")
+
+(define_insn_reservation
+ "cortex_a57_neon_store_b" 0
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_store_b"))
+ "ca57_store_model")
+
+;; These block issue for a number of cycles proportional to the number
+;; of 64-bit chunks they will store, we don't attempt to model that
+;; precisely, treat them as blocking execution for two cycles when
+;; issued.
+(define_insn_reservation
+ "cortex_a57_neon_store_complex" 0
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "cortex_a57_neon_type" "neon_store_complex"))
+ "ca57_block*2")
+
+;; Floating-Point Operations.
+
+(define_insn_reservation "cortex_a57_fp_const" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "fconsts,fconstd"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation "cortex_a57_fp_add_sub" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "fadds,faddd"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation "cortex_a57_fp_mul" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "fmuls,fmuld"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation "cortex_a57_fp_mac" 10
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "fmacs,ffmas,fmacd,ffmad"))
+ "(ca57_cx1,nothing,nothing,ca57_cx1) \
+ |(ca57_cx2,nothing,nothing,ca57_cx2)")
+
+(define_insn_reservation "cortex_a57_fp_cvt" 6
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "f_cvt,f_cvtf2i,f_cvti2f"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation "cortex_a57_fp_cmp" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "fcmps,fcmpd"))
+ "ca57_cx2")
+
+(define_insn_reservation "cortex_a57_fp_arith" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "ffariths,ffarithd"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation "cortex_a57_fp_cpys" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "fmov"))
+ "(ca57_cx1|ca57_cx2)")
+
+(define_insn_reservation "cortex_a57_fp_divs" 12
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "fdivs, fsqrts,\
+ neon_fp_div_s, neon_fp_sqrt_s"))
+ "ca57_cx2_block*5")
+
+(define_insn_reservation "cortex_a57_fp_divd" 16
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "fdivd, fsqrtd, neon_fp_div_d, neon_fp_sqrt_d"))
+ "ca57_cx2_block*3")
+
+(define_insn_reservation "cortex_a57_neon_fp_div_q" 20
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "fdivd, fsqrtd,\
+ neon_fp_div_s_q, neon_fp_div_d_q,\
+ neon_fp_sqrt_s_q, neon_fp_sqrt_d_q"))
+ "ca57_cx2_block*3")
+
+(define_insn_reservation "cortex_a57_crypto_simple" 4
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "crypto_aese,crypto_aesmc,crypto_sha1_fast"))
+ "ca57_cx2")
+
+(define_insn_reservation "cortex_a57_crypto_complex" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "crypto_sha1_slow"))
+ "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
+
+(define_insn_reservation "cortex_a57_crypto_xor" 7
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "crypto_sha1_xor"))
+ "(ca57_cx1+ca57_cx2)")
+
+;; We lie with calls. They take up all issue slots, but are otherwise
+;; not harmful.
+(define_insn_reservation "cortex_a57_call" 1
+ (and (eq_attr "tune" "cortexa57")
+ (eq_attr "type" "call"))
+ "ca57_sx1_issue+ca57_sx2_issue+ca57_cx1_issue+ca57_cx2_issue\
+ +ca57_mx_issue+ca57_bx_issue+ca57_ls_issue"
+)
+
+;; Simple execution unit bypasses
+(define_bypass 1 "cortex_a57_alu"
+ "cortex_a57_alu,cortex_a57_alu_shift,cortex_a57_alu_shift_reg")
+(define_bypass 2 "cortex_a57_alu_shift"
+ "cortex_a57_alu,cortex_a57_alu_shift,cortex_a57_alu_shift_reg")
+(define_bypass 2 "cortex_a57_alu_shift_reg"
+ "cortex_a57_alu,cortex_a57_alu_shift,cortex_a57_alu_shift_reg")
+(define_bypass 1 "cortex_a57_alu" "cortex_a57_load1,cortex_a57_load3")
+(define_bypass 2 "cortex_a57_alu_shift" "cortex_a57_load1,cortex_a57_load3")
+(define_bypass 2 "cortex_a57_alu_shift_reg"
+ "cortex_a57_load1,cortex_a57_load3")
+
+;; An MLA or a MUL can feed a dependent MLA.
+(define_bypass 5 "cortex_a57_neon_*mla*,cortex_a57_neon_*mul*"
+ "cortex_a57_neon_*mla*")
+
+(define_bypass 5 "cortex_a57_fp_mul,cortex_a57_fp_mac"
+ "cortex_a57_fp_mac")
+
+;; We don't need to care about control hazards, either the branch is
+;; predicted in which case we pay no penalty, or the branch is
+;; mispredicted in which case instruction scheduling will be unlikely to
+;; help.
+(define_bypass 1 "cortex_a57_*"
+ "cortex_a57_call,cortex_a57_branch")
+
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-15 10:48 [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor James Greenhalgh
@ 2015-01-16 10:39 ` Marcus Shawcroft
2015-01-16 11:18 ` Ramana Radhakrishnan
2015-01-16 15:16 ` [AArch64] Add a new scheduling " James Greenhalgh
0 siblings, 2 replies; 14+ messages in thread
From: Marcus Shawcroft @ 2015-01-16 10:39 UTC (permalink / raw)
To: James Greenhalgh, ramana.radhakrishnan
Cc: gcc-patches, Marcus Shawcroft, Richard Earnshaw
On 15 January 2015 at 09:50, James Greenhalgh <james.greenhalgh@arm.com> wrote:
> 2015-01-15 James Greenhalgh <james.greenhalgh@arm.com>
>
> * config/arm/cortex-a57.md: New.
> * config/aarch64/aarch64.md: Include it.
> * config/aarch64/aarch64-cores.def (cortex-a57): Tune for it.
> * config/aarch64/aarch64-tune.md: Regenerate.
This is low risk and only affects aarch64 backend. I think we should
take this for gcc 5, so OK.
Ramana do you want this wired up for aarch32 in a followup?
/Marcus
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-16 10:39 ` Marcus Shawcroft
@ 2015-01-16 11:18 ` Ramana Radhakrishnan
2015-01-19 18:06 ` [ARM] Wire up the new scheduler " James Greenhalgh
2015-01-16 15:16 ` [AArch64] Add a new scheduling " James Greenhalgh
1 sibling, 1 reply; 14+ messages in thread
From: Ramana Radhakrishnan @ 2015-01-16 11:18 UTC (permalink / raw)
To: Marcus Shawcroft, James Greenhalgh
Cc: gcc-patches, Marcus Shawcroft, Richard Earnshaw
On 16/01/15 10:20, Marcus Shawcroft wrote:
> On 15 January 2015 at 09:50, James Greenhalgh <james.greenhalgh@arm.com> wrote:
>
>> 2015-01-15 James Greenhalgh <james.greenhalgh@arm.com>
>>
>> * config/arm/cortex-a57.md: New.
>> * config/aarch64/aarch64.md: Include it.
>> * config/aarch64/aarch64-cores.def (cortex-a57): Tune for it.
>> * config/aarch64/aarch64-tune.md: Regenerate.
>
> This is low risk and only affects aarch64 backend. I think we should
> take this for gcc 5, so OK.
>
> Ramana do you want this wired up for aarch32 in a followup?
Yes please. I think as long as it works, let's just take it for 5.0. I'd
rather not have divergence between the backends as this just creates
problems for us elsewhere.
Ramana
>
> /Marcus
>
^ permalink raw reply [flat|nested] 14+ messages in thread
* [ARM] Wire up the new scheduler description for the ARM Cortex-A57 processor
2015-01-16 11:18 ` Ramana Radhakrishnan
@ 2015-01-19 18:06 ` James Greenhalgh
2015-01-23 15:24 ` James Greenhalgh
2015-01-26 9:14 ` Ramana Radhakrishnan
0 siblings, 2 replies; 14+ messages in thread
From: James Greenhalgh @ 2015-01-19 18:06 UTC (permalink / raw)
To: gcc-patches; +Cc: richard.earnshaw, ramana.radhakrishnan
[-- Attachment #1: Type: text/plain, Size: 1503 bytes --]
On Fri, Jan 16, 2015 at 11:14:42AM +0000, Ramana Radhakrishnan wrote:
>
>
> On 16/01/15 10:20, Marcus Shawcroft wrote:
> > On 15 January 2015 at 09:50, James Greenhalgh <james.greenhalgh@arm.com> wrote:
> >
> >> 2015-01-15 James Greenhalgh <james.greenhalgh@arm.com>
> >>
> >> * config/arm/cortex-a57.md: New.
> >> * config/aarch64/aarch64.md: Include it.
> >> * config/aarch64/aarch64-cores.def (cortex-a57): Tune for it.
> >> * config/aarch64/aarch64-tune.md: Regenerate.
> >
> > This is low risk and only affects aarch64 backend. I think we should
> > take this for gcc 5, so OK.
> >
> > Ramana do you want this wired up for aarch32 in a followup?
>
> Yes please. I think as long as it works, let's just take it for 5.0. I'd
> rather not have divergence between the backends as this just creates
> problems for us elsewhere.
Hi,
OK, that patch looks like this. I know I've missed the stage 3 deadline,
but I agree that the two back ends should not diverge too much - and
this patch is small and gives a nice little boost to performance on
some popular benchmarks.
I've regression tested it over the weekend with no issues.
OK?
Cheers,
James
---
2014-01-19 James Greenhalgh <james.greenhalgh@arm.com>
* config/arm/arm-cores.def (cortex-a57): Use the new Cortex-A57
pipeline model.
config/arm/arm.md: Include the new Cortex-A57 model.
(generic_sched): Don't use generic_sched when tuning for
Cortex-A57.
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-ARM-Wire-up-the-new-scheduler-description-for-the-AR.patch --]
[-- Type: text/x-patch; name=0001-ARM-Wire-up-the-new-scheduler-description-for-the-AR.patch, Size: 1751 bytes --]
diff --git a/gcc/config/arm/arm-cores.def b/gcc/config/arm/arm-cores.def
index fa13eb9..f24fefd 100644
--- a/gcc/config/arm/arm-cores.def
+++ b/gcc/config/arm/arm-cores.def
@@ -166,7 +166,7 @@ ARM_CORE("cortex-a17.cortex-a7", cortexa17cortexa7, cortexa7, 7A, FL_LDSCHED |
/* V8 Architecture Processors */
ARM_CORE("cortex-a53", cortexa53, cortexa53, 8A, FL_LDSCHED | FL_CRC32, cortex_a53)
-ARM_CORE("cortex-a57", cortexa57, cortexa15, 8A, FL_LDSCHED | FL_CRC32, cortex_a57)
+ARM_CORE("cortex-a57", cortexa57, cortexa57, 8A, FL_LDSCHED | FL_CRC32, cortex_a57)
ARM_CORE("xgene1", xgene1, xgene1, 8A, FL_LDSCHED, xgene1)
/* V8 big.LITTLE implementations */
diff --git a/gcc/config/arm/arm.md b/gcc/config/arm/arm.md
index 5e6649a5dd39c58fac66bb26e0ffae32d2b4a6b8..149f5bc90d4087c6b1425f8e1fb70503cec18970 100644
--- a/gcc/config/arm/arm.md
+++ b/gcc/config/arm/arm.md
@@ -391,8 +391,8 @@ (define_attr "generic_sched" "yes,no"
arm926ejs,arm1020e,arm1026ejs,arm1136js,\
arm1136jfs,cortexa5,cortexa7,cortexa8,\
cortexa9,cortexa12,cortexa15,cortexa17,\
- cortexa53,cortexm4,cortexm7,marvell_pj4,\
- xgene1")
+ cortexa53,cortexa57,cortexm4,cortexm7,\
+ marvell_pj4,xgene1")
(eq_attr "tune_cortexr4" "yes"))
(const_string "no")
(const_string "yes"))))
@@ -425,6 +425,7 @@ (define_attr "generic_vfp" "yes,no"
(include "cortex-a15.md")
(include "cortex-a17.md")
(include "cortex-a53.md")
+(include "cortex-a57.md")
(include "cortex-r4.md")
(include "cortex-r4f.md")
(include "cortex-m7.md")
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [ARM] Wire up the new scheduler description for the ARM Cortex-A57 processor
2015-01-19 18:06 ` [ARM] Wire up the new scheduler " James Greenhalgh
@ 2015-01-23 15:24 ` James Greenhalgh
2015-01-26 9:14 ` Ramana Radhakrishnan
1 sibling, 0 replies; 14+ messages in thread
From: James Greenhalgh @ 2015-01-23 15:24 UTC (permalink / raw)
To: gcc-patches; +Cc: Richard Earnshaw, Ramana Radhakrishnan
*Ping*
Cheers,
James
On Mon, Jan 19, 2015 at 05:44:27PM +0000, James Greenhalgh wrote:
>
> On Fri, Jan 16, 2015 at 11:14:42AM +0000, Ramana Radhakrishnan wrote:
> >
> >
> > On 16/01/15 10:20, Marcus Shawcroft wrote:
> > > On 15 January 2015 at 09:50, James Greenhalgh <james.greenhalgh@arm.com> wrote:
> > >
> > >> 2015-01-15 James Greenhalgh <james.greenhalgh@arm.com>
> > >>
> > >> * config/arm/cortex-a57.md: New.
> > >> * config/aarch64/aarch64.md: Include it.
> > >> * config/aarch64/aarch64-cores.def (cortex-a57): Tune for it.
> > >> * config/aarch64/aarch64-tune.md: Regenerate.
> > >
> > > This is low risk and only affects aarch64 backend. I think we should
> > > take this for gcc 5, so OK.
> > >
> > > Ramana do you want this wired up for aarch32 in a followup?
> >
> > Yes please. I think as long as it works, let's just take it for 5.0. I'd
> > rather not have divergence between the backends as this just creates
> > problems for us elsewhere.
>
> Hi,
>
> OK, that patch looks like this. I know I've missed the stage 3 deadline,
> but I agree that the two back ends should not diverge too much - and
> this patch is small and gives a nice little boost to performance on
> some popular benchmarks.
>
> I've regression tested it over the weekend with no issues.
>
> OK?
>
> Cheers,
> James
>
> ---
> 2014-01-19 James Greenhalgh <james.greenhalgh@arm.com>
>
> * config/arm/arm-cores.def (cortex-a57): Use the new Cortex-A57
> pipeline model.
> config/arm/arm.md: Include the new Cortex-A57 model.
> (generic_sched): Don't use generic_sched when tuning for
> Cortex-A57.
>
> diff --git a/gcc/config/arm/arm-cores.def b/gcc/config/arm/arm-cores.def
> index fa13eb9..f24fefd 100644
> --- a/gcc/config/arm/arm-cores.def
> +++ b/gcc/config/arm/arm-cores.def
> @@ -166,7 +166,7 @@ ARM_CORE("cortex-a17.cortex-a7", cortexa17cortexa7, cortexa7, 7A, FL_LDSCHED |
>
> /* V8 Architecture Processors */
> ARM_CORE("cortex-a53", cortexa53, cortexa53, 8A, FL_LDSCHED | FL_CRC32, cortex_a53)
> -ARM_CORE("cortex-a57", cortexa57, cortexa15, 8A, FL_LDSCHED | FL_CRC32, cortex_a57)
> +ARM_CORE("cortex-a57", cortexa57, cortexa57, 8A, FL_LDSCHED | FL_CRC32, cortex_a57)
> ARM_CORE("xgene1", xgene1, xgene1, 8A, FL_LDSCHED, xgene1)
>
> /* V8 big.LITTLE implementations */
> diff --git a/gcc/config/arm/arm.md b/gcc/config/arm/arm.md
> index 5e6649a5dd39c58fac66bb26e0ffae32d2b4a6b8..149f5bc90d4087c6b1425f8e1fb70503cec18970 100644
> --- a/gcc/config/arm/arm.md
> +++ b/gcc/config/arm/arm.md
> @@ -391,8 +391,8 @@ (define_attr "generic_sched" "yes,no"
> arm926ejs,arm1020e,arm1026ejs,arm1136js,\
> arm1136jfs,cortexa5,cortexa7,cortexa8,\
> cortexa9,cortexa12,cortexa15,cortexa17,\
> - cortexa53,cortexm4,cortexm7,marvell_pj4,\
> - xgene1")
> + cortexa53,cortexa57,cortexm4,cortexm7,\
> + marvell_pj4,xgene1")
> (eq_attr "tune_cortexr4" "yes"))
> (const_string "no")
> (const_string "yes"))))
> @@ -425,6 +425,7 @@ (define_attr "generic_vfp" "yes,no"
> (include "cortex-a15.md")
> (include "cortex-a17.md")
> (include "cortex-a53.md")
> +(include "cortex-a57.md")
> (include "cortex-r4.md")
> (include "cortex-r4f.md")
> (include "cortex-m7.md")
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [ARM] Wire up the new scheduler description for the ARM Cortex-A57 processor
2015-01-19 18:06 ` [ARM] Wire up the new scheduler " James Greenhalgh
2015-01-23 15:24 ` James Greenhalgh
@ 2015-01-26 9:14 ` Ramana Radhakrishnan
1 sibling, 0 replies; 14+ messages in thread
From: Ramana Radhakrishnan @ 2015-01-26 9:14 UTC (permalink / raw)
To: James Greenhalgh; +Cc: gcc-patches, Richard Earnshaw, Ramana Radhakrishnan
On Mon, Jan 19, 2015 at 5:44 PM, James Greenhalgh
<james.greenhalgh@arm.com> wrote:
>
> On Fri, Jan 16, 2015 at 11:14:42AM +0000, Ramana Radhakrishnan wrote:
>>
>>
>> On 16/01/15 10:20, Marcus Shawcroft wrote:
>> > On 15 January 2015 at 09:50, James Greenhalgh <james.greenhalgh@arm.com> wrote:
>> >
>> >> 2015-01-15 James Greenhalgh <james.greenhalgh@arm.com>
>> >>
>> >> * config/arm/cortex-a57.md: New.
>> >> * config/aarch64/aarch64.md: Include it.
>> >> * config/aarch64/aarch64-cores.def (cortex-a57): Tune for it.
>> >> * config/aarch64/aarch64-tune.md: Regenerate.
>> >
>> > This is low risk and only affects aarch64 backend. I think we should
>> > take this for gcc 5, so OK.
>> >
>> > Ramana do you want this wired up for aarch32 in a followup?
>>
>> Yes please. I think as long as it works, let's just take it for 5.0. I'd
>> rather not have divergence between the backends as this just creates
>> problems for us elsewhere.
>
> Hi,
>
> OK, that patch looks like this. I know I've missed the stage 3 deadline,
> but I agree that the two back ends should not diverge too much - and
> this patch is small and gives a nice little boost to performance on
> some popular benchmarks.
>
> I've regression tested it over the weekend with no issues.
>
> OK?
This is OK as this is a tuning patch that was posted before the stage3
deadline and only wires in a pipeline description.
Ramana
>
> Cheers,
> James
>
> ---
> 2014-01-19 James Greenhalgh <james.greenhalgh@arm.com>
>
> * config/arm/arm-cores.def (cortex-a57): Use the new Cortex-A57
> pipeline model.
> config/arm/arm.md: Include the new Cortex-A57 model.
> (generic_sched): Don't use generic_sched when tuning for
> Cortex-A57.
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-16 10:39 ` Marcus Shawcroft
2015-01-16 11:18 ` Ramana Radhakrishnan
@ 2015-01-16 15:16 ` James Greenhalgh
2015-01-16 15:38 ` Ramana Radhakrishnan
1 sibling, 1 reply; 14+ messages in thread
From: James Greenhalgh @ 2015-01-16 15:16 UTC (permalink / raw)
To: Marcus Shawcroft
Cc: Ramana Radhakrishnan, gcc-patches, Marcus Shawcroft, Richard Earnshaw
[-- Attachment #1: Type: text/plain, Size: 1024 bytes --]
On Fri, Jan 16, 2015 at 10:20:40AM +0000, Marcus Shawcroft wrote:
> On 15 January 2015 at 09:50, James Greenhalgh <james.greenhalgh@arm.com> wrote:
>
> > 2015-01-15 James Greenhalgh <james.greenhalgh@arm.com>
> >
> > * config/arm/cortex-a57.md: New.
> > * config/aarch64/aarch64.md: Include it.
> > * config/aarch64/aarch64-cores.def (cortex-a57): Tune for it.
> > * config/aarch64/aarch64-tune.md: Regenerate.
>
> This is low risk and only affects aarch64 backend. I think we should
> take this for gcc 5, so OK.
Thanks Marcus,
I committed this as revision 219724, but I made a mistake in applying
the patch and we ended up with two copies of the model in the same
file.
I've committed the attached as revision 219746 as the obvious
fixup to my mistake.
This should resolve the build failure that doko was seeing.
Thanks,
James
---
2015-01-16 James Greenhalgh <james.greenhalgh@arm.com>
* config/arm/cortex-a57.md: Remove duplicate of file accidentally
introduced in 219724.
[-- Attachment #2: fixup.diff --]
[-- Type: text/x-diff, Size: 29665 bytes --]
Index: gcc/ChangeLog
===================================================================
--- gcc/ChangeLog (revision 219744)
+++ gcc/ChangeLog (working copy)
@@ -1,3 +1,8 @@
+2015-01-16 James Greenhalgh <james.greenhalgh@arm.com>
+
+ * config/arm/cortex-a57.md: Remove duplicate of file accidentally
+ introduced in 219724.
+
2015-01-16 David Edelsohn <dje.gcc@gmail.com>
PR target/64623
Index: gcc/config/arm/cortex-a57.md
===================================================================
--- gcc/config/arm/cortex-a57.md (revision 219744)
+++ gcc/config/arm/cortex-a57.md (working copy)
@@ -795,800 +795,3 @@
(define_bypass 1 "cortex_a57_*"
"cortex_a57_call,cortex_a57_branch")
-;; ARM Cortex-A57 pipeline description
-;; Copyright (C) 2014-2015 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 "cortex_a57")
-
-(define_attr "cortex_a57_neon_type"
- "neon_abd, neon_abd_q, neon_arith_acc, neon_arith_acc_q,
- neon_arith_basic, neon_arith_complex,
- neon_reduc_add_acc, neon_multiply, neon_multiply_q,
- neon_multiply_long, neon_mla, neon_mla_q, neon_mla_long,
- neon_sat_mla_long, neon_shift_acc, neon_shift_imm_basic,
- neon_shift_imm_complex,
- neon_shift_reg_basic, neon_shift_reg_basic_q, neon_shift_reg_complex,
- neon_shift_reg_complex_q, neon_fp_negabs, neon_fp_arith,
- neon_fp_arith_q, neon_fp_reductions_q, neon_fp_cvt_int,
- neon_fp_cvt_int_q, neon_fp_cvt16, neon_fp_minmax, neon_fp_mul,
- neon_fp_mul_q, neon_fp_mla, neon_fp_mla_q, neon_fp_recpe_rsqrte,
- neon_fp_recpe_rsqrte_q, neon_fp_recps_rsqrts, neon_fp_recps_rsqrts_q,
- neon_bitops, neon_bitops_q, neon_from_gp,
- neon_from_gp_q, neon_move, neon_tbl3_tbl4, neon_zip_q, neon_to_gp,
- neon_load_a, neon_load_b, neon_load_c, neon_load_d, neon_load_e,
- neon_load_f, neon_store_a, neon_store_b, neon_store_complex,
- unknown"
- (cond [
- (eq_attr "type" "neon_abd, neon_abd_long")
- (const_string "neon_abd")
- (eq_attr "type" "neon_abd_q")
- (const_string "neon_abd_q")
- (eq_attr "type" "neon_arith_acc, neon_reduc_add_acc,\
- neon_reduc_add_acc_q")
- (const_string "neon_arith_acc")
- (eq_attr "type" "neon_arith_acc_q")
- (const_string "neon_arith_acc_q")
- (eq_attr "type" "neon_add, neon_add_q, neon_add_long,\
- neon_add_widen, neon_neg, neon_neg_q,\
- neon_reduc_add, neon_reduc_add_q,\
- neon_reduc_add_long, neon_sub, neon_sub_q,\
- neon_sub_long, neon_sub_widen, neon_logic,\
- neon_logic_q, neon_tst, neon_tst_q")
- (const_string "neon_arith_basic")
- (eq_attr "type" "neon_abs, neon_abs_q, neon_add_halve_narrow_q,\
- neon_add_halve, neon_add_halve_q,\
- neon_sub_halve, neon_sub_halve_q, neon_qabs,\
- neon_qabs_q, neon_qadd, neon_qadd_q, neon_qneg,\
- neon_qneg_q, neon_qsub, neon_qsub_q,\
- neon_sub_halve_narrow_q,\
- neon_compare, neon_compare_q,\
- neon_compare_zero, neon_compare_zero_q,\
- neon_minmax, neon_minmax_q, neon_reduc_minmax,\
- neon_reduc_minmax_q")
- (const_string "neon_arith_complex")
-
- (eq_attr "type" "neon_mul_b, neon_mul_h, neon_mul_s,\
- neon_mul_h_scalar, neon_mul_s_scalar,\
- neon_sat_mul_b, neon_sat_mul_h,\
- neon_sat_mul_s, neon_sat_mul_h_scalar,\
- neon_sat_mul_s_scalar,\
- neon_mul_b_long, neon_mul_h_long,\
- neon_mul_s_long, neon_mul_d_long,\
- neon_mul_h_scalar_long, neon_mul_s_scalar_long,\
- neon_sat_mul_b_long, neon_sat_mul_h_long,\
- neon_sat_mul_s_long, neon_sat_mul_h_scalar_long,\
- neon_sat_mul_s_scalar_long")
- (const_string "neon_multiply")
- (eq_attr "type" "neon_mul_b_q, neon_mul_h_q, neon_mul_s_q,\
- neon_mul_h_scalar_q, neon_mul_s_scalar_q,\
- neon_sat_mul_b_q, neon_sat_mul_h_q,\
- neon_sat_mul_s_q, neon_sat_mul_h_scalar_q,\
- neon_sat_mul_s_scalar_q")
- (const_string "neon_multiply_q")
- (eq_attr "type" "neon_mla_b, neon_mla_h, neon_mla_s,\
- neon_mla_h_scalar, neon_mla_s_scalar,\
- neon_mla_b_long, neon_mla_h_long,\
- neon_mla_s_long,\
- neon_mla_h_scalar_long, neon_mla_s_scalar_long")
- (const_string "neon_mla")
- (eq_attr "type" "neon_mla_b_q, neon_mla_h_q, neon_mla_s_q,\
- neon_mla_h_scalar_q, neon_mla_s_scalar_q")
- (const_string "neon_mla_q")
- (eq_attr "type" "neon_sat_mla_b_long, neon_sat_mla_h_long,\
- neon_sat_mla_s_long, neon_sat_mla_h_scalar_long,\
- neon_sat_mla_s_scalar_long")
- (const_string "neon_sat_mla_long")
-
- (eq_attr "type" "neon_shift_acc, neon_shift_acc_q")
- (const_string "neon_shift_acc")
- (eq_attr "type" "neon_shift_imm, neon_shift_imm_q,\
- neon_shift_imm_narrow_q, neon_shift_imm_long")
- (const_string "neon_shift_imm_basic")
- (eq_attr "type" "neon_sat_shift_imm, neon_sat_shift_imm_q,\
- neon_sat_shift_imm_narrow_q")
- (const_string "neon_shift_imm_complex")
- (eq_attr "type" "neon_shift_reg")
- (const_string "neon_shift_reg_basic")
- (eq_attr "type" "neon_shift_reg_q")
- (const_string "neon_shift_reg_basic_q")
- (eq_attr "type" "neon_sat_shift_reg")
- (const_string "neon_shift_reg_complex")
- (eq_attr "type" "neon_sat_shift_reg_q")
- (const_string "neon_shift_reg_complex_q")
-
- (eq_attr "type" "neon_fp_neg_s, neon_fp_neg_s_q,\
- neon_fp_abs_s, neon_fp_abs_s_q,\
- neon_fp_neg_d, neon_fp_neg_d_q,\
- neon_fp_abs_d, neon_fp_abs_d_q")
- (const_string "neon_fp_negabs")
- (eq_attr "type" "neon_fp_addsub_s, neon_fp_abd_s,\
- neon_fp_reduc_add_s, neon_fp_compare_s,\
- neon_fp_minmax_s, neon_fp_round_s,\
- neon_fp_addsub_d, neon_fp_abd_d,\
- neon_fp_reduc_add_d, neon_fp_compare_d,\
- neon_fp_minmax_d, neon_fp_round_d,\
- neon_fp_reduc_minmax_s, neon_fp_reduc_minmax_d")
- (const_string "neon_fp_arith")
- (eq_attr "type" "neon_fp_addsub_s_q, neon_fp_abd_s_q,\
- neon_fp_reduc_add_s_q, neon_fp_compare_s_q,\
- neon_fp_minmax_s_q, neon_fp_round_s_q,\
- neon_fp_addsub_d_q, neon_fp_abd_d_q,\
- neon_fp_reduc_add_d_q, neon_fp_compare_d_q,\
- neon_fp_minmax_d_q, neon_fp_round_d_q")
- (const_string "neon_fp_arith_q")
- (eq_attr "type" "neon_fp_reduc_minmax_s_q,\
- neon_fp_reduc_minmax_d_q,\
- neon_fp_reduc_add_s_q, neon_fp_reduc_add_d_q")
- (const_string "neon_fp_reductions_q")
- (eq_attr "type" "neon_fp_to_int_s, neon_int_to_fp_s,\
- neon_fp_to_int_d, neon_int_to_fp_d")
- (const_string "neon_fp_cvt_int")
- (eq_attr "type" "neon_fp_to_int_s_q, neon_int_to_fp_s_q,\
- neon_fp_to_int_d_q, neon_int_to_fp_d_q")
- (const_string "neon_fp_cvt_int_q")
- (eq_attr "type" "neon_fp_cvt_narrow_s_q, neon_fp_cvt_widen_h")
- (const_string "neon_fp_cvt16")
- (eq_attr "type" "neon_fp_mul_s, neon_fp_mul_s_scalar,\
- neon_fp_mul_d")
- (const_string "neon_fp_mul")
- (eq_attr "type" "neon_fp_mul_s_q, neon_fp_mul_s_scalar_q,\
- neon_fp_mul_d_q, neon_fp_mul_d_scalar_q")
- (const_string "neon_fp_mul_q")
- (eq_attr "type" "neon_fp_mla_s, neon_fp_mla_s_scalar,\
- neon_fp_mla_d")
- (const_string "neon_fp_mla")
- (eq_attr "type" "neon_fp_mla_s_q, neon_fp_mla_s_scalar_q,
- neon_fp_mla_d_q, neon_fp_mla_d_scalar_q")
- (const_string "neon_fp_mla_q")
- (eq_attr "type" "neon_fp_recpe_s, neon_fp_rsqrte_s,\
- neon_fp_recpx_s,\
- neon_fp_recpe_d, neon_fp_rsqrte_d,\
- neon_fp_recpx_d")
- (const_string "neon_fp_recpe_rsqrte")
- (eq_attr "type" "neon_fp_recpe_s_q, neon_fp_rsqrte_s_q,\
- neon_fp_recpx_s_q,\
- neon_fp_recpe_d_q, neon_fp_rsqrte_d_q,\
- neon_fp_recpx_d_q")
- (const_string "neon_fp_recpe_rsqrte_q")
- (eq_attr "type" "neon_fp_recps_s, neon_fp_rsqrts_s,\
- neon_fp_recps_d, neon_fp_rsqrts_d")
- (const_string "neon_fp_recps_rsqrts")
- (eq_attr "type" "neon_fp_recps_s_q, neon_fp_rsqrts_s_q,\
- neon_fp_recps_d_q, neon_fp_rsqrts_d_q")
- (const_string "neon_fp_recps_rsqrts_q")
- (eq_attr "type" "neon_bsl, neon_cls, neon_cnt,\
- neon_rev, neon_permute, neon_rbit,\
- neon_tbl1, neon_tbl2, neon_zip,\
- neon_dup, neon_dup_q, neon_ext, neon_ext_q,\
- neon_move, neon_move_q, neon_move_narrow_q")
- (const_string "neon_bitops")
- (eq_attr "type" "neon_bsl_q, neon_cls_q, neon_cnt_q,\
- neon_rev_q, neon_permute_q, neon_rbit_q")
- (const_string "neon_bitops_q")
- (eq_attr "type" "neon_from_gp,f_mcr,f_mcrr")
- (const_string "neon_from_gp")
- (eq_attr "type" "neon_from_gp_q")
- (const_string "neon_from_gp_q")
- (eq_attr "type" "neon_tbl3, neon_tbl4")
- (const_string "neon_tbl3_tbl4")
- (eq_attr "type" "neon_zip_q")
- (const_string "neon_zip_q")
- (eq_attr "type" "neon_to_gp, neon_to_gp_q,f_mrc,f_mrrc")
- (const_string "neon_to_gp")
-
- (eq_attr "type" "f_loads, f_loadd,\
- neon_load1_1reg, neon_load1_1reg_q,\
- neon_load1_2reg, neon_load1_2reg_q")
- (const_string "neon_load_a")
- (eq_attr "type" "neon_load1_3reg, neon_load1_3reg_q,\
- neon_load1_4reg, neon_load1_4reg_q")
- (const_string "neon_load_b")
- (eq_attr "type" "neon_load1_one_lane, neon_load1_one_lane_q,\
- neon_load1_all_lanes, neon_load1_all_lanes_q,\
- neon_load2_2reg, neon_load2_2reg_q,\
- neon_load2_all_lanes, neon_load2_all_lanes_q")
- (const_string "neon_load_c")
- (eq_attr "type" "neon_load2_4reg, neon_load2_4reg_q,\
- neon_load3_3reg, neon_load3_3reg_q,\
- neon_load3_one_lane, neon_load3_one_lane_q,\
- neon_load4_4reg, neon_load4_4reg_q")
- (const_string "neon_load_d")
- (eq_attr "type" "neon_load2_one_lane, neon_load2_one_lane_q,\
- neon_load3_all_lanes, neon_load3_all_lanes_q,\
- neon_load4_all_lanes, neon_load4_all_lanes_q")
- (const_string "neon_load_e")
- (eq_attr "type" "neon_load4_one_lane, neon_load4_one_lane_q")
- (const_string "neon_load_f")
-
- (eq_attr "type" "f_stores, f_stored,\
- neon_store1_1reg")
- (const_string "neon_store_a")
- (eq_attr "type" "neon_store1_2reg, neon_store1_1reg_q")
- (const_string "neon_store_b")
- (eq_attr "type" "neon_store1_3reg, neon_store1_3reg_q,\
- neon_store3_3reg, neon_store3_3reg_q,\
- neon_store2_4reg, neon_store2_4reg_q,\
- neon_store4_4reg, neon_store4_4reg_q,\
- neon_store2_2reg, neon_store2_2reg_q,\
- neon_store3_one_lane, neon_store3_one_lane_q,\
- neon_store4_one_lane, neon_store4_one_lane_q,\
- neon_store1_4reg, neon_store1_4reg_q,\
- neon_store1_one_lane, neon_store1_one_lane_q,\
- neon_store2_one_lane, neon_store2_one_lane_q")
- (const_string "neon_store_complex")]
- (const_string "unknown")))
-
-;; The Cortex-A57 core is modelled as a triple issue pipeline that has
-;; the following functional units.
-;; 1. Two pipelines for integer operations: SX1, SX2
-
-(define_cpu_unit "ca57_sx1_issue" "cortex_a57")
-(define_reservation "ca57_sx1" "ca57_sx1_issue")
-
-(define_cpu_unit "ca57_sx2_issue" "cortex_a57")
-(define_reservation "ca57_sx2" "ca57_sx2_issue")
-
-;; 2. One pipeline for complex integer operations: MX
-
-(define_cpu_unit "ca57_mx_issue"
- "cortex_a57")
-(define_reservation "ca57_mx" "ca57_mx_issue")
-(define_reservation "ca57_mx_block" "ca57_mx_issue")
-
-;; 3. Two asymmetric pipelines for Neon and FP operations: CX1, CX2
-(define_automaton "cortex_a57_cx")
-
-(define_cpu_unit "ca57_cx1_issue"
- "cortex_a57_cx")
-(define_cpu_unit "ca57_cx2_issue"
- "cortex_a57_cx")
-
-(define_reservation "ca57_cx1" "ca57_cx1_issue")
-
-(define_reservation "ca57_cx2" "ca57_cx2_issue")
-(define_reservation "ca57_cx2_block" "ca57_cx2_issue*2")
-
-;; 4. One pipeline for branch operations: BX
-
-(define_cpu_unit "ca57_bx_issue" "cortex_a57")
-(define_reservation "ca57_bx" "ca57_bx_issue")
-
-;; 5. Two pipelines for load and store operations: LS1, LS2. The most
-;; valuable thing we can do is force a structural hazard to split
-;; up loads/stores.
-
-(define_cpu_unit "ca57_ls_issue" "cortex_a57")
-(define_cpu_unit "ca57_ldr, ca57_str" "cortex_a57")
-(define_reservation "ca57_load_model" "ca57_ls_issue,ca57_ldr*2")
-(define_reservation "ca57_store_model" "ca57_ls_issue,ca57_str")
-
-;; Block all issue queues.
-
-(define_reservation "ca57_block" "ca57_cx1_issue + ca57_cx2_issue
- + ca57_mx_issue + ca57_sx1_issue
- + ca57_sx2_issue + ca57_ls_issue")
-
-;; Simple Execution Unit:
-;;
-;; Simple ALU without shift
-(define_insn_reservation "cortex_a57_alu" 2
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "alu_imm,alus_imm,logic_imm,logics_imm,\
- alu_sreg,alus_sreg,logic_reg,logics_reg,\
- adc_imm,adcs_imm,adc_reg,adcs_reg,\
- adr,bfm,clz,rbit,rev,alu_dsp_reg,\
- shift_imm,shift_reg,\
- mov_imm,mov_reg,\
- mvn_imm,mvn_reg,\
- mrs,multiple,no_insn"))
- "ca57_sx1|ca57_sx2")
-
-;; ALU ops with immediate shift
-(define_insn_reservation "cortex_a57_alu_shift" 3
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "extend,\
- alu_shift_imm,alus_shift_imm,\
- crc,logic_shift_imm,logics_shift_imm,\
- mov_shift,mvn_shift"))
- "ca57_mx")
-
-;; Multi-Cycle Execution Unit:
-;;
-;; ALU ops with register controlled shift
-(define_insn_reservation "cortex_a57_alu_shift_reg" 3
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "alu_shift_reg,alus_shift_reg,\
- logic_shift_reg,logics_shift_reg,\
- mov_shift_reg,mvn_shift_reg"))
- "ca57_mx")
-
-;; All multiplies
-;; TODO: AArch32 and AArch64 have different behaviour
-(define_insn_reservation "cortex_a57_mult32" 3
- (and (eq_attr "tune" "cortexa57")
- (ior (eq_attr "mul32" "yes")
- (eq_attr "mul64" "yes")))
- "ca57_mx")
-
-;; Integer divide
-(define_insn_reservation "cortex_a57_div" 10
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "udiv,sdiv"))
- "ca57_mx_issue,ca57_mx_block*3")
-
-;; Block all issue pipes for a cycle
-(define_insn_reservation "cortex_a57_block" 1
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "block"))
- "ca57_block")
-
-;; Branch execution Unit
-;;
-;; Branches take one issue slot.
-;; No latency as there is no result
-(define_insn_reservation "cortex_a57_branch" 0
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "branch"))
- "ca57_bx")
-
-;; Load-store execution Unit
-;;
-;; Loads of up to two words.
-(define_insn_reservation "cortex_a57_load1" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "load_byte,load1,load2"))
- "ca57_load_model")
-
-;; Loads of three or four words.
-(define_insn_reservation "cortex_a57_load3" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "load3,load4"))
- "ca57_ls_issue*2,ca57_load_model")
-
-;; Stores of up to two words.
-(define_insn_reservation "cortex_a57_store1" 0
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "store1,store2"))
- "ca57_store_model")
-
-;; Stores of three or four words.
-(define_insn_reservation "cortex_a57_store3" 0
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "store3,store4"))
- "ca57_ls_issue*2,ca57_store_model")
-
-;; Advanced SIMD Unit - Integer Arithmetic Instructions.
-
-(define_insn_reservation "cortex_a57_neon_abd" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_abd"))
- "ca57_cx1|ca57_cx2")
-
-(define_insn_reservation "cortex_a57_neon_abd_q" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_abd_q"))
- "ca57_cx1+ca57_cx2")
-
-(define_insn_reservation "cortex_a57_neon_aba" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_arith_acc"))
- "ca57_cx2")
-
-(define_insn_reservation "cortex_a57_neon_aba_q" 8
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_arith_acc_q"))
- "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
-
-(define_insn_reservation "cortex_a57_neon_arith_basic" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_arith_basic"))
- "ca57_cx1|ca57_cx2")
-
-(define_insn_reservation "cortex_a57_neon_arith_complex" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_arith_complex"))
- "ca57_cx1|ca57_cx2")
-
-;; Integer Multiply Instructions.
-
-(define_insn_reservation "cortex_a57_neon_multiply" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_multiply"))
- "ca57_cx1")
-
-(define_insn_reservation "cortex_a57_neon_multiply_q" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_multiply_q"))
- "ca57_cx1+(ca57_cx1_issue,ca57_cx1)")
-
-(define_insn_reservation "cortex_a57_neon_mla" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_mla"))
- "ca57_cx1")
-
-(define_insn_reservation "cortex_a57_neon_mla_q" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_mla_q"))
- "ca57_cx1+(ca57_cx1_issue,ca57_cx1)")
-
-(define_insn_reservation "cortex_a57_neon_sat_mla_long" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_sat_mla_long"))
- "ca57_cx1")
-
-;; Integer Shift Instructions.
-
-(define_insn_reservation
- "cortex_a57_neon_shift_acc" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_shift_acc"))
- "ca57_cx2")
-
-(define_insn_reservation
- "cortex_a57_neon_shift_imm_basic" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_shift_imm_basic"))
- "ca57_cx2")
-
-(define_insn_reservation
- "cortex_a57_neon_shift_imm_complex" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_shift_imm_complex"))
- "ca57_cx2")
-
-(define_insn_reservation
- "cortex_a57_neon_shift_reg_basic" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_shift_reg_basic"))
- "ca57_cx2")
-
-(define_insn_reservation
- "cortex_a57_neon_shift_reg_basic_q" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_shift_reg_basic_q"))
- "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_shift_reg_complex" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_shift_reg_complex"))
- "ca57_cx2")
-
-(define_insn_reservation
- "cortex_a57_neon_shift_reg_complex_q" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_shift_reg_complex_q"))
- "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
-
-;; Floating Point Instructions.
-
-(define_insn_reservation
- "cortex_a57_neon_fp_negabs" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_negabs"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_arith" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_arith"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_arith_q" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_arith_q"))
- "(ca57_cx1+ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_reductions_q" 10
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_reductions_q"))
- "(ca57_cx1+ca57_cx2),(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_cvt_int" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_cvt_int"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_cvt_int_q" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_cvt_int_q"))
- "(ca57_cx1+ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_cvt16" 10
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_cvt16"))
- "(ca57_cx1_issue+ca57_cx2_issue),(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_mul" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_mul"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_mul_q" 5
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_mul_q"))
- "(ca57_cx1+ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_mla" 9
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_mla"))
- "(ca57_cx1,ca57_cx1)|(ca57_cx2,ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_mla_q" 9
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_mla_q"))
- "(ca57_cx1+ca57_cx2),(ca57_cx1,ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_recpe_rsqrte" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_recpe_rsqrte"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_recpe_rsqrte_q" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_recpe_rsqrte_q"))
- "(ca57_cx1+ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_recps_rsqrts" 10
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_recps_rsqrts"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_fp_recps_rsqrts_q" 10
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_fp_recps_rsqrts_q"))
- "(ca57_cx1+ca57_cx2)")
-
-;; Miscellaneous Instructions.
-
-(define_insn_reservation
- "cortex_a57_neon_bitops" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_bitops"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_bitops_q" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_bitops_q"))
- "(ca57_cx1+ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_from_gp" 9
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_from_gp"))
- "(ca57_ls_issue+ca57_cx1_issue,ca57_cx1)
- |(ca57_ls_issue+ca57_cx2_issue,ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_from_gp_q" 9
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_from_gp_q"))
- "(ca57_ls_issue+ca57_cx1_issue,ca57_cx1)
- +(ca57_ls_issue+ca57_cx2_issue,ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_tbl3_tbl4" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_tbl3_tbl4"))
- "(ca57_cx1_issue,ca57_cx1)
- +(ca57_cx2_issue,ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_zip_q" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_zip_q"))
- "(ca57_cx1_issue,ca57_cx1)
- +(ca57_cx2_issue,ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_to_gp" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_to_gp"))
- "((ca57_ls_issue+ca57_sx1_issue),ca57_sx1)
- |((ca57_ls_issue+ca57_sx2_issue),ca57_sx2)")
-
-;; Load Instructions.
-
-(define_insn_reservation
- "cortex_a57_neon_load_a" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_load_a"))
- "ca57_load_model")
-
-(define_insn_reservation
- "cortex_a57_neon_load_b" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_load_b"))
- "ca57_ls_issue,ca57_ls_issue+ca57_ldr,ca57_ldr*2")
-
-(define_insn_reservation
- "cortex_a57_neon_load_c" 9
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_load_c"))
- "ca57_load_model+(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_load_d" 11
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_load_d"))
- "ca57_cx1_issue+ca57_cx2_issue,
- ca57_ls_issue+ca57_ls_issue,ca57_ldr*2")
-
-(define_insn_reservation
- "cortex_a57_neon_load_e" 9
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_load_e"))
- "ca57_load_model+(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation
- "cortex_a57_neon_load_f" 11
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_load_f"))
- "ca57_cx1_issue+ca57_cx2_issue,
- ca57_ls_issue+ca57_ls_issue,ca57_ldr*2")
-
-;; Store Instructions.
-
-(define_insn_reservation
- "cortex_a57_neon_store_a" 0
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_store_a"))
- "ca57_store_model")
-
-(define_insn_reservation
- "cortex_a57_neon_store_b" 0
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_store_b"))
- "ca57_store_model")
-
-;; These block issue for a number of cycles proportional to the number
-;; of 64-bit chunks they will store, we don't attempt to model that
-;; precisely, treat them as blocking execution for two cycles when
-;; issued.
-(define_insn_reservation
- "cortex_a57_neon_store_complex" 0
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "cortex_a57_neon_type" "neon_store_complex"))
- "ca57_block*2")
-
-;; Floating-Point Operations.
-
-(define_insn_reservation "cortex_a57_fp_const" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "fconsts,fconstd"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation "cortex_a57_fp_add_sub" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "fadds,faddd"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation "cortex_a57_fp_mul" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "fmuls,fmuld"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation "cortex_a57_fp_mac" 10
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "fmacs,ffmas,fmacd,ffmad"))
- "(ca57_cx1,nothing,nothing,ca57_cx1) \
- |(ca57_cx2,nothing,nothing,ca57_cx2)")
-
-(define_insn_reservation "cortex_a57_fp_cvt" 6
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "f_cvt,f_cvtf2i,f_cvti2f"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation "cortex_a57_fp_cmp" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "fcmps,fcmpd"))
- "ca57_cx2")
-
-(define_insn_reservation "cortex_a57_fp_arith" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "ffariths,ffarithd"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation "cortex_a57_fp_cpys" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "fmov"))
- "(ca57_cx1|ca57_cx2)")
-
-(define_insn_reservation "cortex_a57_fp_divs" 12
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "fdivs, fsqrts,\
- neon_fp_div_s, neon_fp_sqrt_s"))
- "ca57_cx2_block*5")
-
-(define_insn_reservation "cortex_a57_fp_divd" 16
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "fdivd, fsqrtd, neon_fp_div_d, neon_fp_sqrt_d"))
- "ca57_cx2_block*3")
-
-(define_insn_reservation "cortex_a57_neon_fp_div_q" 20
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "fdivd, fsqrtd,\
- neon_fp_div_s_q, neon_fp_div_d_q,\
- neon_fp_sqrt_s_q, neon_fp_sqrt_d_q"))
- "ca57_cx2_block*3")
-
-(define_insn_reservation "cortex_a57_crypto_simple" 4
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "crypto_aese,crypto_aesmc,crypto_sha1_fast"))
- "ca57_cx2")
-
-(define_insn_reservation "cortex_a57_crypto_complex" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "crypto_sha1_slow"))
- "ca57_cx2+(ca57_cx2_issue,ca57_cx2)")
-
-(define_insn_reservation "cortex_a57_crypto_xor" 7
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "crypto_sha1_xor"))
- "(ca57_cx1+ca57_cx2)")
-
-;; We lie with calls. They take up all issue slots, but are otherwise
-;; not harmful.
-(define_insn_reservation "cortex_a57_call" 1
- (and (eq_attr "tune" "cortexa57")
- (eq_attr "type" "call"))
- "ca57_sx1_issue+ca57_sx2_issue+ca57_cx1_issue+ca57_cx2_issue\
- +ca57_mx_issue+ca57_bx_issue+ca57_ls_issue"
-)
-
-;; Simple execution unit bypasses
-(define_bypass 1 "cortex_a57_alu"
- "cortex_a57_alu,cortex_a57_alu_shift,cortex_a57_alu_shift_reg")
-(define_bypass 2 "cortex_a57_alu_shift"
- "cortex_a57_alu,cortex_a57_alu_shift,cortex_a57_alu_shift_reg")
-(define_bypass 2 "cortex_a57_alu_shift_reg"
- "cortex_a57_alu,cortex_a57_alu_shift,cortex_a57_alu_shift_reg")
-(define_bypass 1 "cortex_a57_alu" "cortex_a57_load1,cortex_a57_load3")
-(define_bypass 2 "cortex_a57_alu_shift" "cortex_a57_load1,cortex_a57_load3")
-(define_bypass 2 "cortex_a57_alu_shift_reg"
- "cortex_a57_load1,cortex_a57_load3")
-
-;; An MLA or a MUL can feed a dependent MLA.
-(define_bypass 5 "cortex_a57_neon_*mla*,cortex_a57_neon_*mul*"
- "cortex_a57_neon_*mla*")
-
-(define_bypass 5 "cortex_a57_fp_mul,cortex_a57_fp_mac"
- "cortex_a57_fp_mac")
-
-;; We don't need to care about control hazards, either the branch is
-;; predicted in which case we pay no penalty, or the branch is
-;; mispredicted in which case instruction scheduling will be unlikely to
-;; help.
-(define_bypass 1 "cortex_a57_*"
- "cortex_a57_call,cortex_a57_branch")
-
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-16 15:16 ` [AArch64] Add a new scheduling " James Greenhalgh
@ 2015-01-16 15:38 ` Ramana Radhakrishnan
2015-01-19 18:09 ` James Greenhalgh
0 siblings, 1 reply; 14+ messages in thread
From: Ramana Radhakrishnan @ 2015-01-16 15:38 UTC (permalink / raw)
To: James Greenhalgh
Cc: Marcus Shawcroft, Ramana Radhakrishnan, gcc-patches,
Marcus Shawcroft, Richard Earnshaw
On Fri, Jan 16, 2015 at 3:06 PM, James Greenhalgh
<james.greenhalgh@arm.com> wrote:
> On Fri, Jan 16, 2015 at 10:20:40AM +0000, Marcus Shawcroft wrote:
>> On 15 January 2015 at 09:50, James Greenhalgh <james.greenhalgh@arm.com> wrote:
>>
>> > 2015-01-15 James Greenhalgh <james.greenhalgh@arm.com>
>> >
>> > * config/arm/cortex-a57.md: New.
>> > * config/aarch64/aarch64.md: Include it.
>> > * config/aarch64/aarch64-cores.def (cortex-a57): Tune for it.
>> > * config/aarch64/aarch64-tune.md: Regenerate.
>>
>> This is low risk and only affects aarch64 backend. I think we should
>> take this for gcc 5, so OK.
>
> Thanks Marcus,
>
> I committed this as revision 219724, but I made a mistake in applying
> the patch and we ended up with two copies of the model in the same
> file.
>
> I've committed the attached as revision 219746 as the obvious
> fixup to my mistake.
>
> This should resolve the build failure that doko was seeing.
Can you also spin up a changes.html patch for the ARM / AArch64 ports
while you are at it ?
Ramana
>
> Thanks,
> James
>
> ---
> 2015-01-16 James Greenhalgh <james.greenhalgh@arm.com>
>
> * config/arm/cortex-a57.md: Remove duplicate of file accidentally
> introduced in 219724.
>
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-16 15:38 ` Ramana Radhakrishnan
@ 2015-01-19 18:09 ` James Greenhalgh
2015-01-19 21:29 ` Gerald Pfeifer
0 siblings, 1 reply; 14+ messages in thread
From: James Greenhalgh @ 2015-01-19 18:09 UTC (permalink / raw)
To: Ramana Radhakrishnan
Cc: Marcus Shawcroft, gcc-patches, Marcus Shawcroft,
Richard Earnshaw, gerald
On Fri, Jan 16, 2015 at 03:34:30PM +0000, Ramana Radhakrishnan wrote:
> On Fri, Jan 16, 2015 at 3:06 PM, James Greenhalgh
> <james.greenhalgh@arm.com> wrote:
> > On Fri, Jan 16, 2015 at 10:20:40AM +0000, Marcus Shawcroft wrote:
> >> On 15 January 2015 at 09:50, James Greenhalgh <james.greenhalgh@arm.com> wrote:
> >>
> >> > 2015-01-15 James Greenhalgh <james.greenhalgh@arm.com>
> >> >
> >> > * config/arm/cortex-a57.md: New.
> >> > * config/aarch64/aarch64.md: Include it.
> >> > * config/aarch64/aarch64-cores.def (cortex-a57): Tune for it.
> >> > * config/aarch64/aarch64-tune.md: Regenerate.
> >>
> >> This is low risk and only affects aarch64 backend. I think we should
> >> take this for gcc 5, so OK.
> >
> > Thanks Marcus,
> >
> > I committed this as revision 219724, but I made a mistake in applying
> > the patch and we ended up with two copies of the model in the same
> > file.
> >
> > I've committed the attached as revision 219746 as the obvious
> > fixup to my mistake.
> >
> > This should resolve the build failure that doko was seeing.
>
> Can you also spin up a changes.html patch for the ARM / AArch64 ports
> while you are at it ?
Of course, how does the below look to you?
OK after the Cortex-A57 scheduling description goes in to the ARM port?
Thanks,
James
---
Index: htdocs/gcc-5/changes.html
===================================================================
RCS file: /cvs/gcc/wwwdocs/htdocs/gcc-5/changes.html,v
retrieving revision 1.65
diff -u -r1.65 changes.html
--- htdocs/gcc-5/changes.html 14 Jan 2015 11:59:44 -0000 1.65
+++ htdocs/gcc-5/changes.html 16 Jan 2015 16:10:10 -0000
@@ -439,6 +439,12 @@
<h3 id="aarch64">AArch64</h3>
<ul>
+ <li>Code generation for the ARM Cortex-A57 processor has been improved.
+ A more accurate instruction scheduling model for the processor is
+ now used, and a number of compiler tuning parameters have been set
+ to offer increased performance when compiling with
+ <code>-mcpu=cortex-a57</code> or <code>-mtune=cortex-a57</code>.
+ </li>
<li> A workaround for the ARM Cortex-A53 erratum 835769 has been
added and can be enabled by giving the
<code>-mfix-cortex-a53-835769</code> option.
@@ -471,6 +477,12 @@
configure option with either of <code>--with-tune</code> or
<code>--with-arch</code>.
</li>
+ <li>Code generation for the ARM Cortex-A57 processor has been improved.
+ A more accurate instruction scheduling model for the processor is
+ now used, and a number of compiler tuning parameters have been set
+ to offer increased performance when compiling with
+ <code>-mcpu=cortex-a57</code> or <code>-mtune=cortex-a57</code>.
+ </li>
<li> Support for the Cortex-A17 processor has been added through the
<code>-mcpu=cortex-a17</code> and <code>-mtune=cortex-a17</code> options.
</li>
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-19 18:09 ` James Greenhalgh
@ 2015-01-19 21:29 ` Gerald Pfeifer
2015-01-19 21:58 ` James Greenhalgh
0 siblings, 1 reply; 14+ messages in thread
From: Gerald Pfeifer @ 2015-01-19 21:29 UTC (permalink / raw)
To: James Greenhalgh
Cc: Ramana Radhakrishnan, Marcus Shawcroft, gcc-patches,
Marcus Shawcroft, Richard Earnshaw
On Monday 2015-01-19 17:52, James Greenhalgh wrote:
> OK after the Cortex-A57 scheduling description goes in to the ARM port?
Yes, thanks, except that once will be sufficient. ;-) (The current
patch features the same hunk twice?)
Gerald
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-19 21:29 ` Gerald Pfeifer
@ 2015-01-19 21:58 ` James Greenhalgh
2015-01-20 8:58 ` Ramana Radhakrishnan
2015-01-20 9:42 ` Marcus Shawcroft
0 siblings, 2 replies; 14+ messages in thread
From: James Greenhalgh @ 2015-01-19 21:58 UTC (permalink / raw)
To: Gerald Pfeifer
Cc: Ramana Radhakrishnan, Marcus Shawcroft, gcc-patches,
Marcus Shawcroft, Richard Earnshaw
On Mon, Jan 19, 2015 at 08:57:31PM +0000, Gerald Pfeifer wrote:
> On Monday 2015-01-19 17:52, James Greenhalgh wrote:
> > OK after the Cortex-A57 scheduling description goes in to the ARM port?
>
> Yes, thanks, except that once will be sufficient. ;-) (The current
> patch features the same hunk twice?)
Once under AArch64 and once under ARM. I'm happy to drop one or the
other hunk. Neither is incorrect, but I agree it is odd to say the same
thing twice.
Ramana, Marcus, Richard, any opinions on how you would like this resolved?
Perhaps an ARM/AArch64 common changes section? Though I'm not sure which
other changes would go in to it.
Cheers,
James
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-19 21:58 ` James Greenhalgh
@ 2015-01-20 8:58 ` Ramana Radhakrishnan
2015-01-20 9:42 ` Marcus Shawcroft
1 sibling, 0 replies; 14+ messages in thread
From: Ramana Radhakrishnan @ 2015-01-20 8:58 UTC (permalink / raw)
To: James Greenhalgh, Gerald Pfeifer
Cc: Marcus Shawcroft, gcc-patches, Marcus Shawcroft, Richard Earnshaw
On 19/01/15 21:05, James Greenhalgh wrote:
> On Mon, Jan 19, 2015 at 08:57:31PM +0000, Gerald Pfeifer wrote:
>> On Monday 2015-01-19 17:52, James Greenhalgh wrote:
>>> OK after the Cortex-A57 scheduling description goes in to the ARM port?
>>
>> Yes, thanks, except that once will be sufficient. ;-) (The current
>> patch features the same hunk twice?)
>
> Once under AArch64 and once under ARM. I'm happy to drop one or the
> other hunk. Neither is incorrect, but I agree it is odd to say the same
> thing twice.
>
> Ramana, Marcus, Richard, any opinions on how you would like this resolved?
> Perhaps an ARM/AArch64 common changes section? Though I'm not sure which
> other changes would go in to it.
>
I'm not sure if a "common" section improves readability. I'd rather this
remained as it is today.
My 10 paise.
Ramana
> Cheers,
> James
>
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-19 21:58 ` James Greenhalgh
2015-01-20 8:58 ` Ramana Radhakrishnan
@ 2015-01-20 9:42 ` Marcus Shawcroft
2015-01-20 16:11 ` Gerald Pfeifer
1 sibling, 1 reply; 14+ messages in thread
From: Marcus Shawcroft @ 2015-01-20 9:42 UTC (permalink / raw)
To: James Greenhalgh, Gerald Pfeifer
Cc: Ramana Radhakrishnan, Marcus Shawcroft, gcc-patches, Richard Earnshaw
On 19/01/15 21:05, James Greenhalgh wrote:
> On Mon, Jan 19, 2015 at 08:57:31PM +0000, Gerald Pfeifer wrote:
>> On Monday 2015-01-19 17:52, James Greenhalgh wrote:
>>> OK after the Cortex-A57 scheduling description goes in to the ARM port?
>>
>> Yes, thanks, except that once will be sufficient. ;-) (The current
>> patch features the same hunk twice?)
>
> Once under AArch64 and once under ARM. I'm happy to drop one or the
> other hunk. Neither is incorrect, but I agree it is odd to say the same
> thing twice.
>
> Ramana, Marcus, Richard, any opinions on how you would like this resolved?
> Perhaps an ARM/AArch64 common changes section? Though I'm not sure which
> other changes would go in to it.
I'd prefer separate sections, IMHO that is more useful. /Marcus
^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor
2015-01-20 9:42 ` Marcus Shawcroft
@ 2015-01-20 16:11 ` Gerald Pfeifer
0 siblings, 0 replies; 14+ messages in thread
From: Gerald Pfeifer @ 2015-01-20 16:11 UTC (permalink / raw)
To: Ramana Radhakrishnan, Marcus Shawcroft
Cc: James Greenhalgh, Marcus Shawcroft, gcc-patches, Richard Earnshaw
On Tuesday 2015-01-20 08:15, Ramana Radhakrishnan wrote:
> I'm not sure if a "common" section improves readability. I'd rather
> this remained as it is today.
On Tuesday 2015-01-20 09:27, Marcus Shawcroft wrote:
> I'd prefer separate sections, IMHO that is more useful. /Marcus
Okay, then let's go that way.
Gerald
^ permalink raw reply [flat|nested] 14+ messages in thread
end of thread, other threads:[~2015-01-26 8:25 UTC | newest]
Thread overview: 14+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-01-15 10:48 [AArch64] Add a new scheduling description for the ARM Cortex-A57 processor James Greenhalgh
2015-01-16 10:39 ` Marcus Shawcroft
2015-01-16 11:18 ` Ramana Radhakrishnan
2015-01-19 18:06 ` [ARM] Wire up the new scheduler " James Greenhalgh
2015-01-23 15:24 ` James Greenhalgh
2015-01-26 9:14 ` Ramana Radhakrishnan
2015-01-16 15:16 ` [AArch64] Add a new scheduling " James Greenhalgh
2015-01-16 15:38 ` Ramana Radhakrishnan
2015-01-19 18:09 ` James Greenhalgh
2015-01-19 21:29 ` Gerald Pfeifer
2015-01-19 21:58 ` James Greenhalgh
2015-01-20 8:58 ` Ramana Radhakrishnan
2015-01-20 9:42 ` Marcus Shawcroft
2015-01-20 16:11 ` Gerald Pfeifer
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).