package/gcc/5.4.0/831-ARM-PR-target-70473-Reduce-size-of-Cortex-A8-automat.patch

   1 From 89bd94250827f887b38211ac048b4006229dc9c0 Mon Sep 17 00:00:00 2001
   2 From: ktkachov <ktkachov@138bc75d-0d04-0410-961f-82ee72b054a4>
   3 Date: Fri, 26 Aug 2016 10:23:20 +0000
   4 Subject: [PATCH] [ARM] PR target/70473: Reduce size of Cortex-A8 automaton
   5
   6         PR target/70473
   7         * config/arm/cortex-a8-neon.md (cortex_a8_vfp_muld): Reduce
   8         reservation duration to 15 cycles.
   9         (cortex_a8_vfp_macs): Likewise.
  10         (cortex_a8_vfp_macd): Likewise.
  11         (cortex_a8_vfp_divs): Likewise.
  12         (cortex_a8_vfp_divd): Likewise.
  13
  14 git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@239772 138bc75d-0d04-0410-961f-82ee72b054a4
  15
  16 [Backport from gcc upstream commit 79f5d5e3cd5a668d5ecb2b6b0cce80ef5706ac53.]
  17 Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
  18 ---
  19  gcc/config/arm/cortex-a8-neon.md | 14 +++++++++-----
  20  1 file changed, 9 insertions(+), 5 deletions(-)
  21
  22 diff --git a/gcc/config/arm/cortex-a8-neon.md b/gcc/config/arm/cortex-a8-neon.md
  23 index 04fa01e..4e6d05d 100644
  24 --- a/gcc/config/arm/cortex-a8-neon.md
  25 +++ b/gcc/config/arm/cortex-a8-neon.md
  26 @@ -357,30 +357,34 @@
  27         (eq_attr "type" "fmuls"))
  28    "cortex_a8_vfp,cortex_a8_vfplite*11")
  29
  30 +;; Don't model a reservation for more than 15 cycles as this explodes the
  31 +;; state space of the automaton for little gain.  It is unlikely that the
  32 +;; scheduler will find enough instructions to hide the full latency of the
  33 +;; instructions.
  34  (define_insn_reservation "cortex_a8_vfp_muld" 17
  35    (and (eq_attr "tune" "cortexa8")
  36         (eq_attr "type" "fmuld"))
  37 -  "cortex_a8_vfp,cortex_a8_vfplite*16")
  38 +  "cortex_a8_vfp,cortex_a8_vfplite*15")
  39
  40  (define_insn_reservation "cortex_a8_vfp_macs" 21
  41    (and (eq_attr "tune" "cortexa8")
  42         (eq_attr "type" "fmacs,ffmas"))
  43 -  "cortex_a8_vfp,cortex_a8_vfplite*20")
  44 +  "cortex_a8_vfp,cortex_a8_vfplite*15")
  45
  46  (define_insn_reservation "cortex_a8_vfp_macd" 26
  47    (and (eq_attr "tune" "cortexa8")
  48         (eq_attr "type" "fmacd,ffmad"))
  49 -  "cortex_a8_vfp,cortex_a8_vfplite*25")
  50 +  "cortex_a8_vfp,cortex_a8_vfplite*15")
  51
  52  (define_insn_reservation "cortex_a8_vfp_divs" 37
  53    (and (eq_attr "tune" "cortexa8")
  54         (eq_attr "type" "fdivs, fsqrts"))
  55 -  "cortex_a8_vfp,cortex_a8_vfplite*36")
  56 +  "cortex_a8_vfp,cortex_a8_vfplite*15")
  57
  58  (define_insn_reservation "cortex_a8_vfp_divd" 65
  59    (and (eq_attr "tune" "cortexa8")
  60         (eq_attr "type" "fdivd, fsqrtd"))
  61 -  "cortex_a8_vfp,cortex_a8_vfplite*64")
  62 +  "cortex_a8_vfp,cortex_a8_vfplite*15")
  63
  64  ;; Comparisons can actually take 7 cycles sometimes instead of four,
  65  ;; but given all the other instructions lumped into type=ffarith that
  66 --
  67 2.7.4
  68