--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/gcc/config/sparc/ultra3.md	Fri Jul 17 14:47:48 2009 +0900
@@ -0,0 +1,189 @@
+;; Scheduling description for UltraSPARC-III.
+;;   Copyright (C) 2002, 2004, 2007 Free Software Foundation, Inc.
+;;
+;; This file is part of GCC.
+;;
+;; GCC is free software; you can redistribute it and/or modify
+;; it under the terms of the GNU General Public License as published by
+;; the Free Software Foundation; either version 3, or (at your option)
+;; any later version.
+;;
+;; GCC is distributed in the hope that it will be useful,
+;; but WITHOUT ANY WARRANTY; without even the implied warranty of
+;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+;; GNU General Public License for more details.
+;;
+;; You should have received a copy of the GNU General Public License
+;; along with GCC; see the file COPYING3.  If not see
+;; <http://www.gnu.org/licenses/>.
+
+;; UltraSPARC-III is a quad-issue processor.
+;;
+;; It is also a much simpler beast than Ultra-I/II, no silly
+;; slotting rules and both integer units are fully symmetric.
+;; It does still have single-issue instructions though.
+
+(define_automaton "ultrasparc3_0,ultrasparc3_1")
+
+(define_cpu_unit "us3_ms,us3_br,us3_fpm" "ultrasparc3_0")
+(define_cpu_unit "us3_a0,us3_a1,us3_slot0,\
+                  us3_slot1,us3_slot2,us3_slot3,us3_fpa" "ultrasparc3_1")
+(define_cpu_unit "us3_load_writeback" "ultrasparc3_1")
+
+(define_reservation "us3_slotany" "(us3_slot0 | us3_slot1 | us3_slot2 | us3_slot3)")
+(define_reservation "us3_single_issue" "us3_slot0 + us3_slot1 + us3_slot2 + us3_slot3")
+(define_reservation "us3_ax" "(us3_a0 | us3_a1)")
+
+(define_insn_reservation "us3_single" 1
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "multi,savew,flushw,iflush,trap"))
+  "us3_single_issue")
+
+(define_insn_reservation "us3_integer" 1
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "ialu,shift,compare"))
+  "us3_ax + us3_slotany")
+
+(define_insn_reservation "us3_ialuX" 5
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "ialu,shift,compare"))
+  "us3_single_issue*4, nothing")
+
+(define_insn_reservation "us3_cmove" 2
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "cmove"))
+  "us3_ms + us3_br + us3_slotany, nothing")
+
+;; ??? Not entirely accurate.
+;; ??? It can run from 6 to 9 cycles.  The first cycle the MS pipe
+;; ??? is needed, and the instruction group is broken right after
+;; ??? the imul.  Then 'helper' instructions are generated to perform
+;; ??? each further stage of the multiplication, each such 'helper' is
+;; ??? single group.  So, the reservation aspect is represented accurately
+;; ??? here, but the variable cycles are not.
+;; ??? Currently I have no idea how to determine the variability, but once
+;; ??? known we can simply add a define_bypass or similar to model it.
+(define_insn_reservation "us3_imul" 7
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "imul"))
+  "us3_ms + us3_slotany, us3_single_issue*4, nothing*2")
+
+(define_insn_reservation "us3_idiv" 72
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "idiv"))
+  "us3_ms + us3_slotany, us3_single_issue*69, nothing*2")
+
+;; UltraSPARC-III has a similar load delay as UltraSPARC-I/II except
+;; that all loads except 32-bit/64-bit unsigned loads take the extra
+;; delay for sign/zero extension.
+(define_insn_reservation "us3_2cycle_load" 2
+  (and (eq_attr "cpu" "ultrasparc3")
+    (and (eq_attr "type" "load,fpload")
+      (eq_attr "us3load_type" "2cycle")))
+  "us3_ms + us3_slotany, us3_load_writeback")
+
+(define_insn_reservation "us3_load_delayed" 3
+  (and (eq_attr "cpu" "ultrasparc3")
+    (and (eq_attr "type" "load,sload")
+      (eq_attr "us3load_type" "3cycle")))
+  "us3_ms + us3_slotany, nothing, us3_load_writeback")
+
+(define_insn_reservation "us3_store" 1
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "store,fpstore"))
+  "us3_ms + us3_slotany")
+
+(define_insn_reservation "us3_branch" 1
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "branch"))
+  "us3_br + us3_slotany")
+
+(define_insn_reservation "us3_call_jmpl" 1
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "call,sibcall,call_no_delay_slot,uncond_branch"))
+  "us3_br + us3_ms + us3_slotany")
+
+(define_insn_reservation "us3_fmov" 3
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fpmove"))
+  "us3_fpa + us3_slotany, nothing*2")
+
+(define_insn_reservation "us3_fcmov" 3
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fpcmove"))
+  "us3_fpa + us3_br + us3_slotany, nothing*2")
+
+(define_insn_reservation "us3_fcrmov" 3
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fpcrmove"))
+  "us3_fpa + us3_ms + us3_slotany, nothing*2")
+
+(define_insn_reservation "us3_faddsub" 4
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fp"))
+  "us3_fpa + us3_slotany, nothing*3")
+
+(define_insn_reservation "us3_fpcmp" 5
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fpcmp"))
+  "us3_fpa + us3_slotany, nothing*4")
+
+(define_insn_reservation "us3_fmult" 4
+ (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fpmul"))
+  "us3_fpm + us3_slotany, nothing*3")
+
+(define_insn_reservation "us3_fdivs" 17
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fpdivs"))
+  "(us3_fpm + us3_slotany), us3_fpm*14, nothing*2")
+
+(define_insn_reservation "us3_fsqrts" 20
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fpsqrts"))
+  "(us3_fpm + us3_slotany), us3_fpm*17, nothing*2")
+
+(define_insn_reservation "us3_fdivd" 20
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fpdivd"))
+  "(us3_fpm + us3_slotany), us3_fpm*17, nothing*2")
+
+(define_insn_reservation "us3_fsqrtd" 29
+  (and (eq_attr "cpu" "ultrasparc3")
+    (eq_attr "type" "fpsqrtd"))
+  "(us3_fpm + us3_slotany), us3_fpm*26, nothing*2")
+
+;; Any store may multi issue with the insn creating the source
+;; data as long as that creating insn is not an FPU div/sqrt.
+;; We need a special guard function because this bypass does
+;; not apply to the address inputs of the store.
+(define_bypass 0 "us3_integer,us3_faddsub,us3_fmov,us3_fcmov,us3_fmult" "us3_store"
+   "store_data_bypass_p")
+
+;; An integer branch may execute in the same cycle as the compare
+;; creating the condition codes.
+(define_bypass 0 "us3_integer" "us3_branch")
+
+;; If FMOVfcc is user of FPCMP, latency is only 1 cycle.
+(define_bypass 1 "us3_fpcmp" "us3_fcmov")
+
+;; VIS scheduling
+(define_insn_reservation "us3_fga"
+  3
+  (and (eq_attr "cpu" "ultrasparc3")
+       (eq_attr "type" "fga"))
+  "us3_fpa + us3_slotany, nothing*2")
+
+(define_insn_reservation "us3_fgm"
+  4
+  (and (eq_attr "cpu" "ultrasparc3")
+       (eq_attr "type" "fgm_pack,fgm_mul,fgm_cmp"))
+  "us3_fpm + us3_slotany, nothing*3")
+
+(define_insn_reservation "us3_pdist"
+  4
+  (and (eq_attr "cpu" "ultrasparc3")
+       (eq_attr "type" "fgm_pdist"))
+  "us3_fpm + us3_slotany, nothing*3")
+
+(define_bypass 1 "us3_pdist" "us3_pdist")