ARM Cortex-R4F scheduler.
Paul Brook
paul@codesourcery.com
Mon Sep 1 13:39:00 GMT 2008
The patch below adds a scheduler for the ARM Cortex-R4F VFP unit.
This unit has very limited dual-issue, which means splitting up the
VFP "farith" and "ffarith" types. Other than that it's fairly
straightforward.
Tested on arm-none-eabi.
Applied to svn trunk.
Paul
2008-09-01 Paul Brook <paul@codesourcery.com>
gcc/
* config/arm/arm.md: Include cortex-r4f.md.
(attr fpu): Update type list.
(attr type): Add fcpys, ffariths, ffarithd, fadds, faddd, fconsts,
fconstd, fcmps and fcmpd.
(attr tune_cortexr4): Define.
(attr generic_sched, attr generic_vfp): Use tune_cortexr4.
* config/arm/vfp.md: Document fcpys, ffariths, ffarithd, fadds, faddd,
fconsts, fconstd, fcmps and fcmpd. Use them in insn patterns.
* config/arm/arm.c (arm_issue_rate): Add cortexr4f.
* config/arm/arm1020e.md (v10_ffarith, v10_farith): Use new insn
types.
* config/arm/cortex-a8-neon.md (cortex_a8_vfp_add_sub,
cortex_a8_vfp_farith: Ditto.
* config/arm/vfp11.md (vfp_ffarith, vfp_farith): Ditto.
* config/arm/cortex-r4.md: Use tune_cortexr4.
* config/arm/cortex-r4f.md: New file.
Index: gcc/config/arm/arm1020e.md
===================================================================
--- gcc/config/arm/arm1020e.md (revision 139860)
+++ gcc/config/arm/arm1020e.md (working copy)
@@ -269,12 +269,12 @@ (define_attr "vfp10" "yes,no"
;; first execute state. We model this by using 1020a_e in the first cycle.
(define_insn_reservation "v10_ffarith" 5
(and (eq_attr "vfp10" "yes")
- (eq_attr "type" "ffarith"))
+ (eq_attr "type" "fcpys,ffariths,ffarithd,fcmps,fcmpd"))
"1020a_e+v10_fmac")
(define_insn_reservation "v10_farith" 5
(and (eq_attr "vfp10" "yes")
- (eq_attr "type" "farith"))
+ (eq_attr "type" "faddd,fadds"))
"1020a_e+v10_fmac")
(define_insn_reservation "v10_cvt" 5
Index: gcc/config/arm/arm.c
===================================================================
--- gcc/config/arm/arm.c (revision 139860)
+++ gcc/config/arm/arm.c (working copy)
@@ -19024,6 +19024,7 @@ arm_issue_rate (void)
switch (arm_tune)
{
case cortexr4:
+ case cortexr4f:
case cortexa8:
return 2;
Index: gcc/config/arm/cortex-a8-neon.md
===================================================================
--- gcc/config/arm/cortex-a8-neon.md (revision 139860)
+++ gcc/config/arm/cortex-a8-neon.md (working copy)
@@ -134,7 +134,7 @@ (define_reservation "cortex_a8_vfp"
(define_insn_reservation "cortex_a8_vfp_add_sub" 10
(and (eq_attr "tune" "cortexa8")
- (eq_attr "type" "farith"))
+ (eq_attr "type" "fconsts,fconstd,fadds,faddd"))
"cortex_a8_vfp,cortex_a8_vfplite*9")
(define_insn_reservation "cortex_a8_vfp_muls" 12
@@ -172,7 +172,7 @@ (define_insn_reservation "cortex_a8_vfp_
;; take four cycles, we pick that latency.
(define_insn_reservation "cortex_a8_vfp_farith" 4
(and (eq_attr "tune" "cortexa8")
- (eq_attr "type" "ffarith"))
+
(eq_attr "type" "fcpys,ffariths,ffarithd,fconsts,fconstd,fcmps,fcmpd"))
"cortex_a8_vfp,cortex_a8_vfplite*3")
(define_insn_reservation "cortex_a8_vfp_cvt" 7
Index: gcc/config/arm/cortex-r4.md
===================================================================
--- gcc/config/arm/cortex-r4.md (revision 139860)
+++ gcc/config/arm/cortex-r4.md (working copy)
@@ -77,24 +77,24 @@ (define_reservation "cortex_r4_branch" "
;; Data processing instructions. Moves without shifts are kept separate
;; for the purposes of the dual-issue constraints above.
(define_insn_reservation "cortex_r4_alu" 2
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(and (eq_attr "type" "alu")
(not (eq_attr "insn" "mov"))))
"cortex_r4_alu")
(define_insn_reservation "cortex_r4_mov" 2
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(and (eq_attr "type" "alu")
(eq_attr "insn" "mov")))
"cortex_r4_mov")
(define_insn_reservation "cortex_r4_alu_shift" 2
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "alu_shift"))
"cortex_r4_alu")
(define_insn_reservation "cortex_r4_alu_shift_reg" 2
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "alu_shift_reg"))
"cortex_r4_alu_shift_reg")
@@ -127,32 +127,32 @@ (define_bypass 1 "cortex_r4_alu,cortex_r
;; Multiplication instructions.
(define_insn_reservation "cortex_r4_mul_4" 4
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "mul,smmul"))
"cortex_r4_mul_2")
(define_insn_reservation "cortex_r4_mul_3" 3
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "smulxy,smulwy,smuad,smusd"))
"cortex_r4_mul")
(define_insn_reservation "cortex_r4_mla_4" 4
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "mla,smmla"))
"cortex_r4_mul_2")
(define_insn_reservation "cortex_r4_mla_3" 3
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "smlaxy,smlawy,smlad,smlsd"))
"cortex_r4_mul")
(define_insn_reservation "cortex_r4_smlald" 3
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "smlald,smlsld"))
"cortex_r4_mul")
(define_insn_reservation "cortex_r4_mull" 4
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "smull,umull,umlal,umaal"))
"cortex_r4_mul_2")
@@ -195,19 +195,19 @@ (define_bypass 2 "cortex_r4_mul_4,cortex
;; is performed with B having ten more leading zeros than A.
;; This gives a latency of nine for udiv and ten for sdiv.
(define_insn_reservation "cortex_r4_udiv" 9
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "udiv"))
"cortex_r4_div_9")
(define_insn_reservation "cortex_r4_sdiv" 10
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "sdiv"))
"cortex_r4_div_10")
;; Branches. We assume correct prediction.
(define_insn_reservation "cortex_r4_branch" 0
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "branch"))
"cortex_r4_branch")
@@ -215,7 +215,7 @@ (define_insn_reservation "cortex_r4_bran
;; number is used as "positive infinity" so that everything should be
;; finished by the time of return.
(define_insn_reservation "cortex_r4_call" 32
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "call"))
"nothing")
@@ -226,12 +226,12 @@ (define_insn_reservation "cortex_r4_call
;; accesses following are correctly aligned.
(define_insn_reservation "cortex_r4_load_1_2" 3
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "load1,load2"))
"cortex_r4_load_store")
(define_insn_reservation "cortex_r4_load_3_4" 4
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "load3,load4"))
"cortex_r4_load_store_2")
@@ -281,12 +281,12 @@ (define_bypass 5 "cortex_r4_mul_4,cortex
;; Store instructions.
(define_insn_reservation "cortex_r4_store_1_2" 0
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "store1,store2"))
"cortex_r4_load_store")
(define_insn_reservation "cortex_r4_store_3_4" 0
- (and (eq_attr "tune" "cortexr4")
+ (and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "store3,store4"))
"cortex_r4_load_store_2")
Index: gcc/config/arm/vfp.md
===================================================================
--- gcc/config/arm/vfp.md (revision 139860)
+++ gcc/config/arm/vfp.md (working copy)
@@ -24,8 +24,15 @@ (define_constants
)
;; The VFP "type" attributes differ from those used in the FPA model.
-;; ffarith Fast floating point insns, e.g. abs, neg, cpy, cmp.
-;; farith Most arithmetic insns.
+;; fcpys Single precision cpy.
+;; ffariths Single precision abs, neg.
+;; ffarithd Double precision abs, neg, cpy.
+;; fadds Single precision add/sub.
+;; faddd Double precision add/sub.
+;; fconsts Single precision load immediate.
+;; fconstd Double precision load immediate.
+;; fcmps Single precision comparison.
+;; fcmpd Double precision comparison.
;; fmuls Single precision multiply.
;; fmuld Double precision multiply.
;; fmacs Single precision multiply-accumulate.
@@ -74,7 +81,7 @@ (define_insn "*arm_movsi_vfp"
}
"
[(set_attr "predicable" "yes")
-
(set_attr "type" "*,*,*,*,load1,store1,r_2_f,f_2_r,ffarith,f_loads,f_stores")
+
(set_attr "type" "*,*,*,*,load1,store1,r_2_f,f_2_r,fcpys,f_loads,f_stores")
(set_attr "pool_range" "*,*,*,*,4096,*,*,*,*,1020,*")
(set_attr "neg_pool_range" "*,*,*,*,4084,*,*,*,*,1008,*")]
)
@@ -111,7 +118,7 @@ (define_insn "*thumb2_movsi_vfp"
}
"
[(set_attr "predicable" "yes")
-
(set_attr "type" "*,*,*,*,load1,store1,r_2_f,f_2_r,ffarith,f_load,f_store")
+ (set_attr "type" "*,*,*,*,load1,store1,r_2_f,f_2_r,fcpys,f_load,f_store")
(set_attr "pool_range" "*,*,*,*,4096,*,*,*,*,1020,*")
(set_attr "neg_pool_range" "*,*,*,*, 0,*,*,*,*,1008,*")]
)
@@ -145,7 +152,7 @@ (define_insn "*arm_movdi_vfp"
gcc_unreachable ();
}
"
- [(set_attr "type" "*,load2,store2,r_2_f,f_2_r,ffarith,f_loadd,f_stored")
+ [(set_attr "type" "*,load2,store2,r_2_f,f_2_r,ffarithd,f_loadd,f_stored")
(set_attr "length" "8,8,8,4,4,4,4,4")
(set_attr "pool_range" "*,1020,*,*,*,*,1020,*")
(set_attr "neg_pool_range" "*,1008,*,*,*,*,1008,*")]
@@ -172,7 +179,7 @@ (define_insn "*thumb2_movdi_vfp"
abort ();
}
"
- [(set_attr "type" "*,load2,store2,r_2_f,f_2_r,ffarith,f_load,f_store")
+ [(set_attr "type" "*,load2,store2,r_2_f,f_2_r,ffarithd,f_load,f_store")
(set_attr "length" "8,8,8,4,4,4,4,4")
(set_attr "pool_range" "*,4096,*,*,*,*,1020,*")
(set_attr "neg_pool_range" "*, 0,*,*,*,*,1008,*")]
@@ -214,7 +221,7 @@ (define_insn "*movsf_vfp"
"
[(set_attr "predicable" "yes")
(set_attr "type"
- "r_2_f,f_2_r,farith,f_loads,f_stores,load1,store1,ffarith,*")
+ "r_2_f,f_2_r,fconsts,f_loads,f_stores,load1,store1,fcpys,*")
(set_attr "pool_range" "*,*,*,1020,*,4096,*,*,*")
(set_attr "neg_pool_range" "*,*,*,1008,*,4080,*,*,*")]
)
@@ -250,7 +257,7 @@ (define_insn "*thumb2_movsf_vfp"
"
[(set_attr "predicable" "yes")
(set_attr "type"
- "r_2_f,f_2_r,farith,f_load,f_store,load1,store1,ffarith,*")
+ "r_2_f,f_2_r,fconsts,f_load,f_store,load1,store1,fcpys,*")
(set_attr "pool_range" "*,*,*,1020,*,4092,*,*,*")
(set_attr "neg_pool_range" "*,*,*,1008,*,0,*,*,*")]
)
@@ -288,7 +295,7 @@ (define_insn "*movdf_vfp"
}
"
[(set_attr "type"
- "r_2_f,f_2_r,farith,f_loadd,f_stored,load2,store2,ffarith,*")
+ "r_2_f,f_2_r,fconstd,f_loadd,f_stored,load2,store2,ffarithd,*")
(set_attr "length" "4,4,4,8,8,4,4,4,8")
(set_attr "pool_range" "*,*,*,1020,*,1020,*,*,*")
(set_attr "neg_pool_range" "*,*,*,1008,*,1008,*,*,*")]
@@ -320,7 +327,7 @@ (define_insn "*thumb2_movdf_vfp"
}
"
[(set_attr "type"
- "r_2_f,f_2_r,farith,load2,store2,f_load,f_store,ffarith,*")
+ "r_2_f,f_2_r,fconstd,load2,store2,f_load,f_store,ffarithd,*")
(set_attr "length" "4,4,4,8,8,4,4,4,8")
(set_attr "pool_range" "*,*,*,4096,*,1020,*,*,*")
(set_attr "neg_pool_range" "*,*,*,0,*,1008,*,*,*")]
@@ -349,7 +356,7 @@ (define_insn "*movsfcc_vfp"
fmrs%D3\\t%0, %2\;fmrs%d3\\t%0, %1"
[(set_attr "conds" "use")
(set_attr "length" "4,4,8,4,4,8,4,4,8")
-
(set_attr "type" "ffarith,ffarith,ffarith,r_2_f,r_2_f,r_2_f,f_2_r,f_2_r,f_2_r")]
+
(set_attr "type" "fcpys,fcpys,fcpys,r_2_f,r_2_f,r_2_f,f_2_r,f_2_r,f_2_r")]
)
(define_insn "*thumb2_movsfcc_vfp"
@@ -372,7 +379,7 @@ (define_insn "*thumb2_movsfcc_vfp"
ite\\t%D3\;fmrs%D3\\t%0, %2\;fmrs%d3\\t%0, %1"
[(set_attr "conds" "use")
(set_attr "length" "6,6,10,6,6,10,6,6,10")
-
(set_attr "type" "ffarith,ffarith,ffarith,r_2_f,r_2_f,r_2_f,f_2_r,f_2_r,f_2_r")]
+
(set_attr "type" "fcpys,fcpys,fcpys,r_2_f,r_2_f,r_2_f,f_2_r,f_2_r,f_2_r")]
)
(define_insn "*movdfcc_vfp"
@@ -395,7 +402,7 @@ (define_insn "*movdfcc_vfp"
fmrrd%D3\\t%Q0, %R0, %P2\;fmrrd%d3\\t%Q0, %R0, %P1"
[(set_attr "conds" "use")
(set_attr "length" "4,4,8,4,4,8,4,4,8")
-
(set_attr "type" "ffarith,ffarith,ffarith,r_2_f,r_2_f,r_2_f,f_2_r,f_2_r,f_2_r")]
+
(set_attr "type" "ffarithd,ffarithd,ffarithd,r_2_f,r_2_f,r_2_f,f_2_r,f_2_r,f_2_r")]
)
(define_insn "*thumb2_movdfcc_vfp"
@@ -418,7 +425,7 @@ (define_insn "*thumb2_movdfcc_vfp"
ite\\t%D3\;fmrrd%D3\\t%Q0, %R0, %P2\;fmrrd%d3\\t%Q0, %R0, %P1"
[(set_attr "conds" "use")
(set_attr "length" "6,6,10,6,6,10,6,6,10")
-
(set_attr "type" "ffarith,ffarith,ffarith,r_2_f,r_2_f,r_2_f,f_2_r,f_2_r,f_2_r")]
+
(set_attr "type" "ffarithd,ffarithd,ffarithd,r_2_f,r_2_f,r_2_f,f_2_r,f_2_r,f_2_r")]
)
@@ -430,7 +437,7 @@ (define_insn "*abssf2_vfp"
"TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_VFP"
"fabss%?\\t%0, %1"
[(set_attr "predicable" "yes")
- (set_attr "type" "ffarith")]
+ (set_attr "type" "ffariths")]
)
(define_insn "*absdf2_vfp"
@@ -439,7 +446,7 @@ (define_insn "*absdf2_vfp"
"TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_VFP"
"fabsd%?\\t%P0, %P1"
[(set_attr "predicable" "yes")
- (set_attr "type" "ffarith")]
+ (set_attr "type" "ffarithd")]
)
(define_insn "*negsf2_vfp"
@@ -450,7 +457,7 @@ (define_insn "*negsf2_vfp"
fnegs%?\\t%0, %1
eor%?\\t%0, %1, #-2147483648"
[(set_attr "predicable" "yes")
- (set_attr "type" "ffarith")]
+ (set_attr "type" "ffariths")]
)
(define_insn_and_split "*negdf2_vfp"
@@ -496,7 +503,7 @@ (define_insn_and_split "*negdf2_vfp"
"
[(set_attr "predicable" "yes")
(set_attr "length" "4,4,8")
- (set_attr "type" "ffarith")]
+ (set_attr "type" "ffarithd")]
)
@@ -509,7 +516,7 @@ (define_insn "*addsf3_vfp"
"TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_VFP"
"fadds%?\\t%0, %1, %2"
[(set_attr "predicable" "yes")
- (set_attr "type" "farith")]
+ (set_attr "type" "fadds")]
)
(define_insn "*adddf3_vfp"
@@ -519,7 +526,7 @@ (define_insn "*adddf3_vfp"
"TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_VFP"
"faddd%?\\t%P0, %P1, %P2"
[(set_attr "predicable" "yes")
- (set_attr "type" "farith")]
+ (set_attr "type" "faddd")]
)
@@ -530,7 +537,7 @@ (define_insn "*subsf3_vfp"
"TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_VFP"
"fsubs%?\\t%0, %1, %2"
[(set_attr "predicable" "yes")
- (set_attr "type" "farith")]
+ (set_attr "type" "fadds")]
)
(define_insn "*subdf3_vfp"
@@ -540,7 +547,7 @@ (define_insn "*subdf3_vfp"
"TARGET_32BIT && TARGET_HARD_FLOAT && TARGET_VFP"
"fsubd%?\\t%P0, %P1, %P2"
[(set_attr "predicable" "yes")
- (set_attr "type" "farith")]
+ (set_attr "type" "faddd")]
)
@@ -909,7 +916,7 @@ (define_insn "*cmpsf_vfp"
fcmps%?\\t%0, %1
fcmpzs%?\\t%0"
[(set_attr "predicable" "yes")
- (set_attr "type" "ffarith")]
+ (set_attr "type" "fcmps")]
)
(define_insn "*cmpsf_trap_vfp"
@@ -921,7 +928,7 @@ (define_insn "*cmpsf_trap_vfp"
fcmpes%?\\t%0, %1
fcmpezs%?\\t%0"
[(set_attr "predicable" "yes")
- (set_attr "type" "ffarith")]
+ (set_attr "type" "fcmpd")]
)
(define_insn "*cmpdf_vfp"
@@ -933,7 +940,7 @@ (define_insn "*cmpdf_vfp"
fcmpd%?\\t%P0, %P1
fcmpzd%?\\t%P0"
[(set_attr "predicable" "yes")
- (set_attr "type" "ffarith")]
+ (set_attr "type" "fcmpd")]
)
(define_insn "*cmpdf_trap_vfp"
@@ -945,7 +952,7 @@ (define_insn "*cmpdf_trap_vfp"
fcmped%?\\t%P0, %P1
fcmpezd%?\\t%P0"
[(set_attr "predicable" "yes")
- (set_attr "type" "ffarith")]
+ (set_attr "type" "fcmpd")]
)
Index: gcc/config/arm/cortex-r4f.md
===================================================================
--- gcc/config/arm/cortex-r4f.md (revision 0)
+++ gcc/config/arm/cortex-r4f.md (revision 0)
@@ -0,0 +1,161 @@
+;; ARM Crotex-R4F VFP pipeline description
+;; Copyright (C) 2007 Free Software Foundation, Inc.
+;; Written by CodeSourcery.
+;;
+;; 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/>.
+
+;; With the exception of simple VMOV <freg>, <freg> instructions and
+;; the accululate operand of a multiply-accumulate instruction, all
+;; registers are early registers. Thus base latencies are 1 more than
+;; those listed in the TRM.
+
+;; We use the A, B abd C units from the integer core, plus two additional
+;; units to enforce VFP dual issue constraints.
+
+;; A B C V1 VMLA
+;; fcpy 1 2
+;; farith 1 2 1
+;; fmrc 1 2
+;; fconst 1 2 * *
+;; ffarith 1 2 * *
+;; fmac 1 2 1 2
+;; fdiv 1 2 *
+;; f_loads * * *
+;; f_stores * * *
+
+(define_cpu_unit "cortex_r4_v1" "cortex_r4")
+
+(define_cpu_unit "cortex_r4_vmla" "cortex_r4")
+
+(define_reservation "cortex_r4_issue_ab"
+ "(cortex_r4_issue_a|cortex_r4_issue_b)")
+(define_reservation "cortex_r4_single_issue"
+ "cortex_r4_issue_a+cortex_r4_issue_b")
+
+(define_insn_reservation "cortex_r4_fcpys" 2
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "fcpys"))
+ "cortex_r4_issue_ab")
+
+(define_insn_reservation "cortex_r4_ffariths" 2
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "ffariths,fconsts,fcmps"))
+ "cortex_r4_issue_ab+cortex_r4_issue_c+cortex_r4_v1")
+
+(define_insn_reservation "cortex_r4_fariths" 3
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "fadds,fmuls"))
+ "(cortex_r4_issue_a+cortex_r4_v1)|cortex_r4_issue_b")
+
+(define_insn_reservation "cortex_r4_fmacs" 6
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "fmacs"))
+ "(cortex_r4_issue_a+cortex_r4_v1)|(cortex_r4_issue_b+cortex_r4_vmla)")
+
+(define_insn_reservation "cortex_r4_fdivs" 17
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "fdivs"))
+ "cortex_r4_issue_ab+cortex_r4_v1,cortex_r4_issue_a+cortex_r4_v1")
+
+(define_insn_reservation "cortex_r4_floads" 2
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "f_loads"))
+ "cortex_r4_issue_a+cortex_r4_issue_c+cortex_r4_v1")
+
+(define_insn_reservation "cortex_r4_fstores" 1
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "f_stores"))
+ "cortex_r4_issue_a+cortex_r4_issue_c+cortex_r4_vmla")
+
+(define_insn_reservation "cortex_r4_mcr" 2
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "r_2_f"))
+ "cortex_r4_issue_ab")
+
+(define_insn_reservation "cortex_r4_mrc" 3
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "f_2_r"))
+ "cortex_r4_issue_ab")
+
+;; Bypasses for normal (not early) regs.
+(define_bypass 1 "cortex_r4_ffariths,cortex_r4_fcpys,cortex_r4_mcr"
+ "cortex_r4_fcpys")
+(define_bypass 2 "cortex_r4_fariths"
+ "cortex_r4_fcpys")
+(define_bypass 5 "cortex_r4_fmacs"
+ "cortex_r4_fcpys")
+(define_bypass 16 "cortex_r4_fdivs"
+ "cortex_r4_fcpys")
+
+(define_bypass 1 "cortex_r4_ffariths,cortex_r4_fcpys,cortex_r4_mcr"
+ "cortex_r4_fmacs"
+ "arm_no_early_mul_dep")
+(define_bypass 2 "cortex_r4_fariths"
+ "cortex_r4_fmacs"
+ "arm_no_early_mul_dep")
+;; mac->mac has an extra forwarding path.
+(define_bypass 3 "cortex_r4_fmacs"
+ "cortex_r4_fmacs"
+ "arm_no_early_mul_dep")
+(define_bypass 16 "cortex_r4_fdivs"
+ "cortex_r4_fmacs"
+ "arm_no_early_mul_dep")
+
+;; Double precision operations. These can not dual issue.
+
+(define_insn_reservation "cortex_r4_fmacd" 20
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "fmacd"))
+ "cortex_r4_single_issue*13")
+
+(define_insn_reservation "cortex_r4_farith" 10
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "faddd,fmuld"))
+ "cortex_r4_single_issue*3")
+
+;; FIXME: The short cycle count suggests these instructions complete
+;; out of order. Chances are this is not a pipelined operation.
+(define_insn_reservation "cortex_r4_fdivd" 97
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "fdivd"))
+ "cortex_r4_single_issue*3")
+
+(define_insn_reservation "cortex_r4_ffarithd" 2
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "ffarithd,fconstd"))
+ "cortex_r4_single_issue")
+
+(define_insn_reservation "cortex_r4_fcmpd" 2
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "fcmpd"))
+ "cortex_r4_single_issue*2")
+
+(define_insn_reservation "cortex_r4_f_cvt" 8
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "f_cvt"))
+ "cortex_r4_single_issue*3")
+
+(define_insn_reservation "cortex_r4_f_memd" 8
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "f_loadd,f_stored"))
+ "cortex_r4_single_issue")
+
+(define_insn_reservation "cortex_r4_f_flag" 1
+ (and (eq_attr "tune_cortexr4" "yes")
+ (eq_attr "type" "f_stores"))
+ "cortex_r4_single_issue")
+
Index: gcc/config/arm/arm.md
===================================================================
--- gcc/config/arm/arm.md (revision 139860)
+++ gcc/config/arm/arm.md (working copy)
@@ -157,7 +157,7 @@ (define_attr "shift" "" (const_int 0))
; Floating Point Unit. If we only have floating point emulation, then there
; is no point in scheduling the floating point insns. (Well, for best
; performance we should try and group them together).
-(define_attr "fpu" "none,fpa,fpe2,fpe3,maverick,vfp"
+(define_attr "fpu" "none,fpa,fpe2,fpe3,maverick,vfp,vfpv3d16,vfpv3,neon"
(const (symbol_ref "arm_fpu_attr")))
; LENGTH of an instruction (in bytes)
@@ -239,7 +239,7 @@ (define_attr "insn"
;
(define_attr "type"
- "alu,alu_shift,alu_shift_reg,mult,block,float,fdivx,fdivd,fdivs,fmul,fmuls,fmuld,fmacs,fmacd,ffmul,farith,ffarith,f_flag,float_em,f_load,f_store,f_loads,f_loadd,f_stores,f_stored,f_mem_r,r_mem_f,f_2_r,r_2_f,f_cvt,branch,call,load_byte,load1,load2,load3,load4,store1,store2,store3,store4,mav_farith,mav_dmult"
+ "alu,alu_shift,alu_shift_reg,mult,block,float,fdivx,fdivd,fdivs,fmul,fmuls,fmuld,fmacs,fmacd,ffmul,farith,ffarith,f_flag,float_em,f_load,f_store,f_loads,f_loadd,f_stores,f_stored,f_mem_r,r_mem_f,f_2_r,r_2_f,f_cvt,branch,call,load_byte,load1,load2,load3,load4,store1,store2,store3,store4,mav_farith,mav_dmult,fconsts,fconstd,fadds,faddd,ffariths,ffarithd,fcmps,fcmpd,fcpys"
(if_then_else
(eq_attr "insn" "smulxy,smlaxy,smlalxy,smulwy,smlawx,mul,muls,mla,mlas,umull,umulls,umlal,umlals,smull,smulls,smlal,smlals")
(const_string "mult")
@@ -331,18 +331,26 @@ (define_mode_iterator ANY64 [DI DF V8QI
;; Processor type. This is created automatically from arm-cores.def.
(include "arm-tune.md")
+(define_attr "tune_cortexr4" "yes,no"
+ (const (if_then_else
+ (eq_attr "tune" "cortexr4,cortexr4f")
+ (const_string "yes")
+ (const_string "no"))))
+
;; True if the generic scheduling description should be used.
(define_attr "generic_sched" "yes,no"
(const (if_then_else
-
(eq_attr "tune" "arm926ejs,arm1020e,arm1026ejs,arm1136js,arm1136jfs,cortexa8,cortexr4")
+ (ior
(eq_attr "tune" "arm926ejs,arm1020e,arm1026ejs,arm1136js,arm1136jfs,cortexa8")
+ (eq_attr "tune_cortexr4" "yes"))
(const_string "no")
(const_string "yes"))))
(define_attr "generic_vfp" "yes,no"
(const (if_then_else
(and (eq_attr "fpu" "vfp")
- (eq_attr "tune" "!arm1020e,arm1022e,cortexa8"))
+ (eq_attr "tune" "!arm1020e,arm1022e,cortexa8")
+ (eq_attr "tune_cortexr4" "no"))
(const_string "yes")
(const_string "no"))))
@@ -353,6 +361,7 @@ (define_attr "generic_vfp" "yes,no"
(include "arm1136jfs.md")
(include "cortex-a8.md")
(include "cortex-r4.md")
+(include "cortex-r4f.md")
(include "vfp11.md")
Index: gcc/config/arm/vfp11.md
===================================================================
--- gcc/config/arm/vfp11.md (revision 139860)
+++ gcc/config/arm/vfp11.md (working copy)
@@ -51,12 +51,12 @@ (define_cpu_unit "fmstat" "vfp11")
(define_insn_reservation "vfp_ffarith" 4
(and (eq_attr "generic_vfp" "yes")
- (eq_attr "type" "ffarith"))
+ (eq_attr "type" "fcpys,ffariths,ffarithd,fcmps,fcmpd"))
"fmac")
(define_insn_reservation "vfp_farith" 8
(and (eq_attr "generic_vfp" "yes")
- (eq_attr "type" "farith,f_cvt,fmuls,fmacs"))
+ (eq_attr "type" "fadds,faddd,fconsts,fconstd,f_cvt,fmuls,fmacs"))
"fmac")
(define_insn_reservation "vfp_fmul" 9
More information about the Gcc-patches
mailing list