aboutsummaryrefslogtreecommitdiffstats
path: root/gcc-4.4.3/gcc/config/rs6000/power5.md
diff options
context:
space:
mode:
authorJing Yu <jingyu@google.com>2010-07-22 14:03:48 -0700
committerJing Yu <jingyu@google.com>2010-07-22 14:03:48 -0700
commitb094d6c4bf572654a031ecc4afe675154c886dc5 (patch)
tree89394c56b05e13a5413ee60237d65b0214fd98e2 /gcc-4.4.3/gcc/config/rs6000/power5.md
parentdc34721ac3bf7e3c406fba8cfe9d139393345ec5 (diff)
downloadtoolchain_gcc-b094d6c4bf572654a031ecc4afe675154c886dc5.tar.gz
toolchain_gcc-b094d6c4bf572654a031ecc4afe675154c886dc5.tar.bz2
toolchain_gcc-b094d6c4bf572654a031ecc4afe675154c886dc5.zip
commit gcc-4.4.3 which is used to build gcc-4.4.3 Android toolchain in master.
The source is based on fsf gcc-4.4.3 and contains local patches which are recorded in gcc-4.4.3/README.google. Change-Id: Id8c6d6927df274ae9749196a1cc24dbd9abc9887
Diffstat (limited to 'gcc-4.4.3/gcc/config/rs6000/power5.md')
-rw-r--r--gcc-4.4.3/gcc/config/rs6000/power5.md321
1 files changed, 321 insertions, 0 deletions
diff --git a/gcc-4.4.3/gcc/config/rs6000/power5.md b/gcc-4.4.3/gcc/config/rs6000/power5.md
new file mode 100644
index 000000000..1b73e093e
--- /dev/null
+++ b/gcc-4.4.3/gcc/config/rs6000/power5.md
@@ -0,0 +1,321 @@
+;; Scheduling description for IBM POWER5 processor.
+;; Copyright (C) 2003, 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/>.
+
+;; Sources: IBM Red Book and White Paper on POWER5
+
+;; The POWER5 has 2 iu, 2 fpu, 2 lsu per engine (2 engines per chip).
+;; Instructions that update more than one register get broken into two
+;; (split) or more internal ops. The chip can issue up to 5
+;; internal ops per cycle.
+
+(define_automaton "power5iu,power5fpu,power5misc")
+
+(define_cpu_unit "iu1_power5,iu2_power5" "power5iu")
+(define_cpu_unit "lsu1_power5,lsu2_power5" "power5misc")
+(define_cpu_unit "fpu1_power5,fpu2_power5" "power5fpu")
+(define_cpu_unit "bpu_power5,cru_power5" "power5misc")
+(define_cpu_unit "du1_power5,du2_power5,du3_power5,du4_power5,du5_power5"
+ "power5misc")
+
+(define_reservation "lsq_power5"
+ "(du1_power5,lsu1_power5)\
+ |(du2_power5,lsu2_power5)\
+ |(du3_power5,lsu2_power5)\
+ |(du4_power5,lsu1_power5)")
+
+(define_reservation "iq_power5"
+ "(du1_power5,iu1_power5)\
+ |(du2_power5,iu2_power5)\
+ |(du3_power5,iu2_power5)\
+ |(du4_power5,iu1_power5)")
+
+(define_reservation "fpq_power5"
+ "(du1_power5,fpu1_power5)\
+ |(du2_power5,fpu2_power5)\
+ |(du3_power5,fpu2_power5)\
+ |(du4_power5,fpu1_power5)")
+
+; Dispatch slots are allocated in order conforming to program order.
+(absence_set "du1_power5" "du2_power5,du3_power5,du4_power5,du5_power5")
+(absence_set "du2_power5" "du3_power5,du4_power5,du5_power5")
+(absence_set "du3_power5" "du4_power5,du5_power5")
+(absence_set "du4_power5" "du5_power5")
+
+
+; Load/store
+(define_insn_reservation "power5-load" 4 ; 3
+ (and (eq_attr "type" "load")
+ (eq_attr "cpu" "power5"))
+ "lsq_power5")
+
+(define_insn_reservation "power5-load-ext" 5
+ (and (eq_attr "type" "load_ext")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,lsu1_power5,nothing,nothing,iu2_power5")
+
+(define_insn_reservation "power5-load-ext-update" 5
+ (and (eq_attr "type" "load_ext_u")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5+du3_power5+du4_power5,\
+ lsu1_power5+iu2_power5,nothing,nothing,iu2_power5")
+
+(define_insn_reservation "power5-load-ext-update-indexed" 5
+ (and (eq_attr "type" "load_ext_ux")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5+du3_power5+du4_power5,\
+ iu1_power5,lsu2_power5+iu1_power5,nothing,nothing,iu2_power5")
+
+(define_insn_reservation "power5-load-update-indexed" 3
+ (and (eq_attr "type" "load_ux")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5+du3_power5+du4_power5,\
+ iu1_power5,lsu2_power5+iu2_power5")
+
+(define_insn_reservation "power5-load-update" 4 ; 3
+ (and (eq_attr "type" "load_u")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,lsu1_power5+iu2_power5")
+
+(define_insn_reservation "power5-fpload" 6 ; 5
+ (and (eq_attr "type" "fpload")
+ (eq_attr "cpu" "power5"))
+ "lsq_power5")
+
+(define_insn_reservation "power5-fpload-update" 6 ; 5
+ (and (eq_attr "type" "fpload_u,fpload_ux")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,lsu1_power5+iu2_power5")
+
+(define_insn_reservation "power5-store" 12
+ (and (eq_attr "type" "store")
+ (eq_attr "cpu" "power5"))
+ "(du1_power5,lsu1_power5,iu1_power5)\
+ |(du2_power5,lsu2_power5,iu2_power5)\
+ |(du3_power5,lsu2_power5,iu2_power5)\
+ |(du4_power5,lsu1_power5,iu1_power5)")
+
+(define_insn_reservation "power5-store-update" 12
+ (and (eq_attr "type" "store_u")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,lsu1_power5+iu2_power5,iu1_power5")
+
+(define_insn_reservation "power5-store-update-indexed" 12
+ (and (eq_attr "type" "store_ux")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5+du3_power5+du4_power5,\
+ iu1_power5,lsu2_power5+iu2_power5,iu2_power5")
+
+(define_insn_reservation "power5-fpstore" 12
+ (and (eq_attr "type" "fpstore")
+ (eq_attr "cpu" "power5"))
+ "(du1_power5,lsu1_power5,fpu1_power5)\
+ |(du2_power5,lsu2_power5,fpu2_power5)\
+ |(du3_power5,lsu2_power5,fpu2_power5)\
+ |(du4_power5,lsu1_power5,fpu1_power5)")
+
+(define_insn_reservation "power5-fpstore-update" 12
+ (and (eq_attr "type" "fpstore_u,fpstore_ux")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,lsu1_power5+iu2_power5,fpu1_power5")
+
+(define_insn_reservation "power5-llsc" 11
+ (and (eq_attr "type" "load_l,store_c,sync")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5+du3_power5+du4_power5,\
+ lsu1_power5")
+
+
+; Integer latency is 2 cycles
+(define_insn_reservation "power5-integer" 2
+ (and (eq_attr "type" "integer,insert_dword,shift,trap,\
+ var_shift_rotate,cntlz,exts")
+ (eq_attr "cpu" "power5"))
+ "iq_power5")
+
+(define_insn_reservation "power5-two" 2
+ (and (eq_attr "type" "two")
+ (eq_attr "cpu" "power5"))
+ "(du1_power5+du2_power5,iu1_power5,nothing,iu2_power5)\
+ |(du2_power5+du3_power5,iu2_power5,nothing,iu2_power5)\
+ |(du3_power5+du4_power5,iu2_power5,nothing,iu1_power5)\
+ |(du4_power5+du1_power5,iu1_power5,nothing,iu1_power5)")
+
+(define_insn_reservation "power5-three" 2
+ (and (eq_attr "type" "three")
+ (eq_attr "cpu" "power5"))
+ "(du1_power5+du2_power5+du3_power5,\
+ iu1_power5,nothing,iu2_power5,nothing,iu2_power5)\
+ |(du2_power5+du3_power5+du4_power5,\
+ iu2_power5,nothing,iu2_power5,nothing,iu1_power5)\
+ |(du3_power5+du4_power5+du1_power5,\
+ iu2_power5,nothing,iu1_power5,nothing,iu1_power5)\
+ |(du4_power5+du1_power5+du2_power5,\
+ iu1_power5,nothing,iu2_power5,nothing,iu2_power5)")
+
+(define_insn_reservation "power5-insert" 4
+ (and (eq_attr "type" "insert_word")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,iu1_power5,nothing,iu2_power5")
+
+(define_insn_reservation "power5-cmp" 3
+ (and (eq_attr "type" "cmp,fast_compare")
+ (eq_attr "cpu" "power5"))
+ "iq_power5")
+
+(define_insn_reservation "power5-compare" 2
+ (and (eq_attr "type" "compare,delayed_compare,var_delayed_compare")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,iu1_power5,iu2_power5")
+
+(define_bypass 4 "power5-compare" "power5-branch,power5-crlogical,power5-delayedcr,power5-mfcr,power5-mfcrf")
+
+(define_insn_reservation "power5-lmul-cmp" 7
+ (and (eq_attr "type" "lmul_compare")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,iu1_power5*6,iu2_power5")
+
+(define_bypass 10 "power5-lmul-cmp" "power5-branch,power5-crlogical,power5-delayedcr,power5-mfcr,power5-mfcrf")
+
+(define_insn_reservation "power5-imul-cmp" 5
+ (and (eq_attr "type" "imul_compare")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,iu1_power5*4,iu2_power5")
+
+(define_bypass 8 "power5-imul-cmp" "power5-branch,power5-crlogical,power5-delayedcr,power5-mfcr,power5-mfcrf")
+
+(define_insn_reservation "power5-lmul" 7
+ (and (eq_attr "type" "lmul")
+ (eq_attr "cpu" "power5"))
+ "(du1_power5,iu1_power5*6)\
+ |(du2_power5,iu2_power5*6)\
+ |(du3_power5,iu2_power5*6)\
+ |(du4_power5,iu1_power5*6)")
+
+(define_insn_reservation "power5-imul" 5
+ (and (eq_attr "type" "imul")
+ (eq_attr "cpu" "power5"))
+ "(du1_power5,iu1_power5*4)\
+ |(du2_power5,iu2_power5*4)\
+ |(du3_power5,iu2_power5*4)\
+ |(du4_power5,iu1_power5*4)")
+
+(define_insn_reservation "power5-imul3" 4
+ (and (eq_attr "type" "imul2,imul3")
+ (eq_attr "cpu" "power5"))
+ "(du1_power5,iu1_power5*3)\
+ |(du2_power5,iu2_power5*3)\
+ |(du3_power5,iu2_power5*3)\
+ |(du4_power5,iu1_power5*3)")
+
+
+; SPR move only executes in first IU.
+; Integer division only executes in second IU.
+(define_insn_reservation "power5-idiv" 36
+ (and (eq_attr "type" "idiv")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,iu2_power5*35")
+
+(define_insn_reservation "power5-ldiv" 68
+ (and (eq_attr "type" "ldiv")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,iu2_power5*67")
+
+
+(define_insn_reservation "power5-mtjmpr" 3
+ (and (eq_attr "type" "mtjmpr,mfjmpr")
+ (eq_attr "cpu" "power5"))
+ "du1_power5,bpu_power5")
+
+
+; Branches take dispatch Slot 4. The presence_sets prevent other insn from
+; grabbing previous dispatch slots once this is assigned.
+(define_insn_reservation "power5-branch" 2
+ (and (eq_attr "type" "jmpreg,branch")
+ (eq_attr "cpu" "power5"))
+ "(du5_power5\
+ |du4_power5+du5_power5\
+ |du3_power5+du4_power5+du5_power5\
+ |du2_power5+du3_power5+du4_power5+du5_power5\
+ |du1_power5+du2_power5+du3_power5+du4_power5+du5_power5),bpu_power5")
+
+
+; Condition Register logical ops are split if non-destructive (RT != RB)
+(define_insn_reservation "power5-crlogical" 2
+ (and (eq_attr "type" "cr_logical")
+ (eq_attr "cpu" "power5"))
+ "du1_power5,cru_power5")
+
+(define_insn_reservation "power5-delayedcr" 4
+ (and (eq_attr "type" "delayed_cr")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5,cru_power5,cru_power5")
+
+; 4 mfcrf (each 3 cyc, 1/cyc) + 3 fxu
+(define_insn_reservation "power5-mfcr" 6
+ (and (eq_attr "type" "mfcr")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5+du3_power5+du4_power5,\
+ du1_power5+du2_power5+du3_power5+du4_power5+cru_power5,\
+ cru_power5,cru_power5,cru_power5")
+
+; mfcrf (1 field)
+(define_insn_reservation "power5-mfcrf" 3
+ (and (eq_attr "type" "mfcrf")
+ (eq_attr "cpu" "power5"))
+ "du1_power5,cru_power5")
+
+; mtcrf (1 field)
+(define_insn_reservation "power5-mtcr" 4
+ (and (eq_attr "type" "mtcr")
+ (eq_attr "cpu" "power5"))
+ "du1_power5,iu1_power5")
+
+; Basic FP latency is 6 cycles
+(define_insn_reservation "power5-fp" 6
+ (and (eq_attr "type" "fp,dmul")
+ (eq_attr "cpu" "power5"))
+ "fpq_power5")
+
+(define_insn_reservation "power5-fpcompare" 5
+ (and (eq_attr "type" "fpcompare")
+ (eq_attr "cpu" "power5"))
+ "fpq_power5")
+
+(define_insn_reservation "power5-sdiv" 33
+ (and (eq_attr "type" "sdiv,ddiv")
+ (eq_attr "cpu" "power5"))
+ "(du1_power5,fpu1_power5*28)\
+ |(du2_power5,fpu2_power5*28)\
+ |(du3_power5,fpu2_power5*28)\
+ |(du4_power5,fpu1_power5*28)")
+
+(define_insn_reservation "power5-sqrt" 40
+ (and (eq_attr "type" "ssqrt,dsqrt")
+ (eq_attr "cpu" "power5"))
+ "(du1_power5,fpu1_power5*35)\
+ |(du2_power5,fpu2_power5*35)\
+ |(du3_power5,fpu2_power5*35)\
+ |(du4_power5,fpu2_power5*35)")
+
+(define_insn_reservation "power5-isync" 2
+ (and (eq_attr "type" "isync")
+ (eq_attr "cpu" "power5"))
+ "du1_power5+du2_power5+du3_power5+du4_power5,\
+ lsu1_power5")
+