Atom pipeline model, tuning and insn selection.

2009-04-06  Joey Ye  <joey.ye@intel.com>
	    Xuepeng Guo <xuepeng.guo@intel.com>
	    H.J. Lu  <hongjiu.lu@intel.com>

	Atom pipeline model, tuning and insn selection.
	* config.gcc (atom): Add atom config options and target.

	* config/i386/atom.md: New.

	* config/i386/i386.c (atom_cost): New cost.
	(m_ATOM): New macro flag.
	(initial_ix86_tune_features): Set m_ATOM.
	(x86_accumulate_outgoing_args): Likewise.
	(x86_arch_always_fancy_math_387): Likewise.
	(processor_target): Add Atom cost.
	(cpu_names): Add Atom cpu name.
	(override_options): Set Atom ISA.
	(ix86_issue_rate): New case PROCESSOR_ATOM.
	(ix86_adjust_cost): Likewise.

	* config/i386/i386.h (TARGET_ATOM): New target macro.
	(ix86_tune_indices): Add X86_TUNE_OPT_AGU.
	(TARGET_OPT_AGU): New target option.
	(target_cpu_default): Add TARGET_CPU_DEFAULT_atom.
	(processor_type): Add PROCESSOR_ATOM.

	* config/i386/i386.md (cpu): Add new value "atom".
	(use_carry, movu): New attr.
	(atom.md): Include atom.md.
	(adddi3_carry_rex64): Set attr "use_carry".
	(addqi3_carry): Likewise.
	(addhi3_carry): Likewise.
	(addsi3_carry): Likewise.
	(*addsi3_carry_zext): Likewise.
	(subdi3_carry_rex64): Likewise.
	(subqi3_carry): Likewise.
	(subhi3_carry): Likewise.
	(subsi3_carry): Likewise.
	(x86_movdicc_0_m1_rex64): Likewise.
	(*x86_movdicc_0_m1_se): Likewise.
	(x86_movsicc_0_m1): Likewise.
	(*x86_movsicc_0_m1_se): Likewise.
	(*adddi_1_rex64): Emit add insn as much as possible.
	(*addsi_1): Likewise.
	(return_internal): Set atom_unit.
	(return_internal_long): Likewise.
	(return_pop_internal): Likewise.
	(*rcpsf2_sse): Set atom_sse_attr attr.
	(*qrt<mode>2_sse): Likewise.
	(*prefetch_sse): Likewise.

	* config/i386/i386-c.c (ix86_target_macros_internal): New case
	PROCESSOR_ATOM.
	(ix86_target_macros_internal): Likewise.

	* config/i386/sse.md (cpu): Set attr "atom_sse_attr".
	(*prefetch_sse_rex): Likewise.
	(sse_rcpv4sf2): Likewise.
	(sse_vmrcpv4sf2): Likewise.
	(sse_sqrtv4sf2): Likewise.
	(<sse>_vmsqrt<mode>2): Likewise.
	(sse_ldmxcsr): Likewise.
	(sse_stmxcsr): Likewise.
	(*sse_sfence): Likewise.
	(sse2_clflush): Likewise.
	(*sse2_mfence): Likewise.
	(*sse2_lfence): Likewise.
	(avx_movup<avxmodesuffixf2c><avxmodesuffix>): Set attr "movu".
	(<sse>_movup<ssemodesuffixf2c>): Likewise.
	(avx_movdqu<avxmodesuffix>): Likewise.
	(avx_lddqu<avxmodesuffix>): Likewise.
	(sse2_movntv2di): Change attr "type" to "ssemov".
	(sse2_movntsi): Likewise.
	(rsqrtv8sf2): Change attr "type" to "sseadd".
	(sse3_addsubv2df3): Set attr "atom_unit".
	(sse3_h<plusminus_insn>v4sf3): Likewise.
	(*sse2_pmaddwd): Likewise.
	(*vec_extractv2di_1_rex64): Likewise.
	(*vec_extractv2di_1_avx): Likewise.
	(sse2_psadbw): Likewise.
	(ssse3_phaddwv8hi3): Likewise.
	(ssse3_phaddwv4hi3): Likewise.
	(ssse3_phadddv4si3): Likewise.
	(ssse3_phadddv2si3): Likewise.
	(ssse3_phaddswv8hi3): Likewise.
	(ssse3_phaddswv4hi3): Likewise.
	(ssse3_phsubwv8hi3): Likewise.
	(ssse3_phsubwv4hi3): Likewise.
	(ssse3_phsubdv4si3): Likewise.
	(ssse3_phsubdv2si3): Likewise.
	(ssse3_phsubswv8hi3): Likewise.
	(ssse3_phsubswv4hi3): Likewise.
	(ssse3_pmaddubsw128): Likewise.
	(sse3_pmaddubsw: Likewise.
	(ssse3_palignrti): Likewise.
	(ssse3_palignrdi): Likewise.

Co-Authored-By: H.J. Lu <hongjiu.lu@intel.com>
Co-Authored-By: Xuepeng Guo <xuepeng.guo@intel.com>

From-SVN: r145624

gcc/ChangeLog

+2009-04-06  Joey Ye  <joey.ye@intel.com>
+	    Xuepeng Guo <xuepeng.guo@intel.com>
+	    H.J. Lu  <hongjiu.lu@intel.com>
+
+	Atom pipeline model, tuning and insn selection.
+	* config.gcc (atom): Add atom config options and target.
+
+	* config/i386/atom.md: New.
+
+	* config/i386/i386.c (atom_cost): New cost.
+	(m_ATOM): New macro flag.
+	(initial_ix86_tune_features): Set m_ATOM.
+	(x86_accumulate_outgoing_args): Likewise.
+	(x86_arch_always_fancy_math_387): Likewise.
+	(processor_target): Add Atom cost.
+	(cpu_names): Add Atom cpu name.
+	(override_options): Set Atom ISA.
+	(ix86_issue_rate): New case PROCESSOR_ATOM.
+	(ix86_adjust_cost): Likewise.
+
+	* config/i386/i386.h (TARGET_ATOM): New target macro.
+	(ix86_tune_indices): Add X86_TUNE_OPT_AGU.
+	(TARGET_OPT_AGU): New target option.
+	(target_cpu_default): Add TARGET_CPU_DEFAULT_atom.
+	(processor_type): Add PROCESSOR_ATOM.
+
+	* config/i386/i386.md (cpu): Add new value "atom".
+	(use_carry, movu): New attr.
+	(atom.md): Include atom.md.
+	(adddi3_carry_rex64): Set attr "use_carry".
+	(addqi3_carry): Likewise.
+	(addhi3_carry): Likewise.
+	(addsi3_carry): Likewise.
+	(*addsi3_carry_zext): Likewise.
+	(subdi3_carry_rex64): Likewise.
+	(subqi3_carry): Likewise.
+	(subhi3_carry): Likewise.
+	(subsi3_carry): Likewise.
+	(x86_movdicc_0_m1_rex64): Likewise.
+	(*x86_movdicc_0_m1_se): Likewise.
+	(x86_movsicc_0_m1): Likewise.
+	(*x86_movsicc_0_m1_se): Likewise.
+	(*adddi_1_rex64): Emit add insn as much as possible.
+	(*addsi_1): Likewise.
+	(return_internal): Set atom_unit.
+	(return_internal_long): Likewise.
+	(return_pop_internal): Likewise.
+	(*rcpsf2_sse): Set atom_sse_attr attr.
+	(*qrt<mode>2_sse): Likewise.
+	(*prefetch_sse): Likewise.
+
+	* config/i386/i386-c.c (ix86_target_macros_internal): New case
+	PROCESSOR_ATOM.
+	(ix86_target_macros_internal): Likewise.
+
+	* config/i386/sse.md (cpu): Set attr "atom_sse_attr".
+	(*prefetch_sse_rex): Likewise.
+	(sse_rcpv4sf2): Likewise.
+	(sse_vmrcpv4sf2): Likewise.
+	(sse_sqrtv4sf2): Likewise.
+	(<sse>_vmsqrt<mode>2): Likewise.
+	(sse_ldmxcsr): Likewise.
+	(sse_stmxcsr): Likewise.
+	(*sse_sfence): Likewise.
+	(sse2_clflush): Likewise.
+	(*sse2_mfence): Likewise.
+	(*sse2_lfence): Likewise.
+	(avx_movup<avxmodesuffixf2c><avxmodesuffix>): Set attr "movu".
+	(<sse>_movup<ssemodesuffixf2c>): Likewise.
+	(avx_movdqu<avxmodesuffix>): Likewise.
+	(avx_lddqu<avxmodesuffix>): Likewise.
+	(sse2_movntv2di): Change attr "type" to "ssemov".
+	(sse2_movntsi): Likewise.
+	(rsqrtv8sf2): Change attr "type" to "sseadd".
+	(sse3_addsubv2df3): Set attr "atom_unit".
+	(sse3_h<plusminus_insn>v4sf3): Likewise.
+	(*sse2_pmaddwd): Likewise.
+	(*vec_extractv2di_1_rex64): Likewise.
+	(*vec_extractv2di_1_avx): Likewise.
+	(sse2_psadbw): Likewise.
+	(ssse3_phaddwv8hi3): Likewise.
+	(ssse3_phaddwv4hi3): Likewise.
+	(ssse3_phadddv4si3): Likewise.
+	(ssse3_phadddv2si3): Likewise.
+	(ssse3_phaddswv8hi3): Likewise.
+	(ssse3_phaddswv4hi3): Likewise.
+	(ssse3_phsubwv8hi3): Likewise.
+	(ssse3_phsubwv4hi3): Likewise.
+	(ssse3_phsubdv4si3): Likewise.
+	(ssse3_phsubdv2si3): Likewise.
+	(ssse3_phsubswv8hi3): Likewise.
+	(ssse3_phsubswv4hi3): Likewise.
+	(ssse3_pmaddubsw128): Likewise.
+	(sse3_pmaddubsw: Likewise.
+	(ssse3_palignrti): Likewise.
+	(ssse3_palignrdi): Likewise.
+
 2009-04-06  Gerald Pfeifer  <gerald@pfeifer.com>
 
 	* doc/install.texi (Specific): Fix two cross-references to MinGW.

gcc/config.gcc

@@ -1074,7 +1074,7 @@ i[34567]86-*-linux* | i[34567]86-*-kfreebsd*-gnu | i[34567]86-*-knetbsd*-gnu | i
 			tmake_file="${tmake_file} i386/t-linux64"
 			need_64bit_hwint=yes
 			case X"${with_cpu}" in
-			Xgeneric|Xcore2|Xnocona|Xx86-64|Xamdfam10|Xbarcelona|Xk8|Xopteron|Xathlon64|Xathlon-fx)
+			Xgeneric|Xatom|Xcore2|Xnocona|Xx86-64|Xamdfam10|Xbarcelona|Xk8|Xopteron|Xathlon64|Xathlon-fx)
 				;;
 			X)
 				if test x$with_cpu_64 = x; then
@@ -1083,7 +1083,7 @@ i[34567]86-*-linux* | i[34567]86-*-kfreebsd*-gnu | i[34567]86-*-knetbsd*-gnu | i
 				;;
 			*)
 				echo "Unsupported CPU used in --with-cpu=$with_cpu, supported values:" 1>&2
-				echo "generic core2 nocona x86-64 amdfam10 barcelona k8 opteron athlon64 athlon-fx" 1>&2
+				echo "generic atom core2 nocona x86-64 amdfam10 barcelona k8 opteron athlon64 athlon-fx" 1>&2
 				exit 1
 				;;
 			esac
@@ -1189,7 +1189,7 @@ i[34567]86-*-solaris2*)
 		need_64bit_hwint=yes
 		use_gcc_stdint=wrap
 		case X"${with_cpu}" in
-		Xgeneric|Xcore2|Xnocona|Xx86-64|Xamdfam10|Xbarcelona|Xk8|Xopteron|Xathlon64|Xathlon-fx)
+		Xgeneric|Xatom|Xcore2|Xnocona|Xx86-64|Xamdfam10|Xbarcelona|Xk8|Xopteron|Xathlon64|Xathlon-fx)
 			;;
 		X)
 			if test x$with_cpu_64 = x; then
@@ -1198,7 +1198,7 @@ i[34567]86-*-solaris2*)
 			;;
 		*)
 			echo "Unsupported CPU used in --with-cpu=$with_cpu, supported values:" 1>&2
-			echo "generic core2 nocona x86-64 amdfam10 barcelona k8 opteron athlon64 athlon-fx" 1>&2
+			echo "generic atom core2 nocona x86-64 amdfam10 barcelona k8 opteron athlon64 athlon-fx" 1>&2
 			exit 1
 			;;
 		esac
@@ -2801,7 +2801,7 @@ case "${target}" in
 				esac
 				# OK
 				;;
-			"" | amdfam10 | barcelona | k8 | opteron | athlon64 | athlon-fx | nocona | core2 | generic)
+			"" | amdfam10 | barcelona | k8 | opteron | athlon64 | athlon-fx | nocona | core2 | atom | generic)
 				# OK
 				;;
 			*)

gcc/config/i386/atom.md

+;; Atom Scheduling
+;; Copyright (C) 2009 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/>.
+;;
+;; Atom is an in-order core with two integer pipelines.
+
+
+(define_attr "atom_unit" "sishuf,simul,jeu,complex,other" 
+  (const_string "other"))
+
+(define_attr "atom_sse_attr" "rcp,movdup,lfence,fence,prefetch,sqrt,mxcsr,other"
+  (const_string "other"))
+
+(define_automaton "atom")
+
+;;  Atom has two ports: port 0 and port 1 connecting to all execution units
+(define_cpu_unit "atom-port-0,atom-port-1" "atom")
+
+;;  EU: Execution Unit
+;;  Atom EUs are connected by port 0 or port 1. 
+
+(define_cpu_unit "atom-eu-0, atom-eu-1,
+                  atom-imul-1, atom-imul-2, atom-imul-3, atom-imul-4"
+                  "atom")
+
+;; Some EUs have duplicated copied and can be accessed via either
+;; port 0 or port 1
+;; (define_reservation "atom-port-either" "(atom-port-0 | atom-port-1)")
+
+;;; Some instructions is dual-pipe execution, need both ports
+;;; Complex multi-op macro-instructoins need both ports and all EUs
+(define_reservation "atom-port-dual" "(atom-port-0 + atom-port-1)")
+(define_reservation "atom-all-eu" "(atom-eu-0 + atom-eu-1 + 
+                                    atom-imul-1 + atom-imul-2 + atom-imul-3 +
+                                    atom-imul-4)")
+
+;;; Most of simple instructions have 1 cycle latency. Some of them
+;;; issue in port 0, some in port 0 and some in either port.
+(define_reservation "atom-simple-0" "(atom-port-0 + atom-eu-0)")
+(define_reservation "atom-simple-1" "(atom-port-1 + atom-eu-1)")
+(define_reservation "atom-simple-either" "(atom-simple-0 | atom-simple-1)")
+
+;;; Some insn issues in port 0 with 3 cycle latency and 1 cycle tput
+(define_reservation "atom-eu-0-3-1" "(atom-port-0 + atom-eu-0, nothing*2)")
+
+;;; fmul insn can have 4 or 5 cycles latency
+(define_reservation "atom-fmul-5c" "(atom-port-0 + atom-eu-0), nothing*4")
+(define_reservation "atom-fmul-4c" "(atom-port-0 + atom-eu-0), nothing*3")
+
+;;; fadd can has 5 cycles latency depends on instruction forms
+(define_reservation "atom-fadd-5c" "(atom-port-1 + atom-eu-1), nothing*5")
+
+;;; imul insn has 5 cycles latency
+(define_reservation "atom-imul-32" 
+                    "atom-imul-1, atom-imul-2, atom-imul-3, atom-imul-4, 
+                     atom-port-0")
+;;; imul instruction excludes other non-FP instructions.
+(exclusion_set "atom-eu-0, atom-eu-1" 
+               "atom-imul-1, atom-imul-2, atom-imul-3, atom-imul-4")
+
+;;; dual-execution instructions can have 1,2,4,5 cycles latency depends on 
+;;; instruction forms
+(define_reservation "atom-dual-1c" "(atom-port-dual + atom-eu-0 + atom-eu-1)")
+(define_reservation "atom-dual-2c"
+                    "(atom-port-dual + atom-eu-0 + atom-eu-1, nothing)")
+(define_reservation "atom-dual-5c"
+                    "(atom-port-dual + atom-eu-0 + atom-eu-1, nothing*4)")
+
+;;; Complex macro-instruction has variants of latency, and uses both ports.
+(define_reservation "atom-complex" "(atom-port-dual + atom-all-eu)")
+
+(define_insn_reservation  "atom_other" 9
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "other")
+            (eq_attr "atom_unit" "!jeu")))
+  "atom-complex, atom-all-eu*8")
+
+;; return has type "other" with atom_unit "jeu"
+(define_insn_reservation  "atom_other_2" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "other")
+            (eq_attr "atom_unit" "jeu")))
+  "atom-dual-1c")
+
+(define_insn_reservation  "atom_multi" 9
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "multi"))
+  "atom-complex, atom-all-eu*8")
+
+;; Normal alu insns without carry
+(define_insn_reservation  "atom_alu" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "alu")
+            (and (eq_attr "memory" "none")
+                 (eq_attr "use_carry" "0"))))
+  "atom-simple-either")
+
+;; Normal alu insns without carry
+(define_insn_reservation  "atom_alu_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "alu")
+            (and (eq_attr "memory" "!none")
+                 (eq_attr "use_carry" "0"))))
+  "atom-simple-either")
+
+;; Alu insn consuming CF, such as add/sbb
+(define_insn_reservation  "atom_alu_carry" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "alu")
+            (and (eq_attr "memory" "none")
+                 (eq_attr "use_carry" "1"))))
+  "atom-simple-either")
+
+;; Alu insn consuming CF, such as add/sbb
+(define_insn_reservation  "atom_alu_carry_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "alu")
+            (and (eq_attr "memory" "!none")
+                (eq_attr "use_carry" "1"))))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_alu1" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "alu1")
+            (eq_attr "memory" "none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_alu1_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "alu1")
+            (eq_attr "memory" "!none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_negnot" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "negnot")
+            (eq_attr "memory" "none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_negnot_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "negnot")
+            (eq_attr "memory" "!none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_imov" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imov")
+            (eq_attr "memory" "none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_imov_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imov")
+            (eq_attr "memory" "!none")))
+  "atom-simple-either")
+
+;; 16<-16, 32<-32
+(define_insn_reservation  "atom_imovx" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imovx")
+            (and (eq_attr "memory" "none")
+                 (ior (and (match_operand:HI 0 "register_operand")
+                           (match_operand:HI 1 "general_operand"))
+                      (and (match_operand:SI 0 "register_operand")
+                           (match_operand:SI 1 "general_operand"))))))
+  "atom-simple-either")
+
+;; 16<-16, 32<-32, mem
+(define_insn_reservation  "atom_imovx_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imovx")
+            (and (eq_attr "memory" "!none")
+                 (ior (and (match_operand:HI 0 "register_operand")
+                           (match_operand:HI 1 "general_operand"))
+                      (and (match_operand:SI 0 "register_operand")
+                           (match_operand:SI 1 "general_operand"))))))
+  "atom-simple-either")
+
+;; 32<-16, 32<-8, 64<-16, 64<-8, 64<-32, 8<-8
+(define_insn_reservation  "atom_imovx_2" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imovx")
+            (and (eq_attr "memory" "none")
+                 (ior (match_operand:QI 0 "register_operand")
+                      (ior (and (match_operand:SI 0 "register_operand")
+                                (not (match_operand:SI 1 "general_operand")))
+                           (match_operand:DI 0 "register_operand"))))))
+  "atom-simple-0")
+
+;; 32<-16, 32<-8, 64<-16, 64<-8, 64<-32, 8<-8, mem
+(define_insn_reservation  "atom_imovx_2_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imovx")
+            (and (eq_attr "memory" "!none")
+                 (ior (match_operand:QI 0 "register_operand")
+                      (ior (and (match_operand:SI 0 "register_operand")
+                                (not (match_operand:SI 1 "general_operand")))
+                           (match_operand:DI 0 "register_operand"))))))
+  "atom-simple-0")
+
+;; 16<-8
+(define_insn_reservation  "atom_imovx_3" 3
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imovx")
+            (and (match_operand:HI 0 "register_operand")
+                 (match_operand:QI 1 "general_operand"))))
+  "atom-complex, atom-all-eu*2")
+
+(define_insn_reservation  "atom_lea" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "lea")
+            (eq_attr "mode" "!HI")))
+  "atom-simple-either")
+
+;; lea 16bit address is complex insn
+(define_insn_reservation  "atom_lea_2" 2
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "lea")
+            (eq_attr "mode" "HI")))
+  "atom-complex, atom-all-eu")
+
+(define_insn_reservation  "atom_incdec" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "incdec")
+            (eq_attr "memory" "none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_incdec_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "incdec")
+            (eq_attr "memory" "!none")))
+  "atom-simple-either")
+
+;; simple shift instruction use SHIFT eu, none memory
+(define_insn_reservation  "atom_ishift" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ishift")
+            (and (eq_attr "memory" "none") (eq_attr "prefix_0f" "0"))))
+  "atom-simple-0")
+
+;; simple shift instruction use SHIFT eu, memory
+(define_insn_reservation  "atom_ishift_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ishift")
+            (and (eq_attr "memory" "!none") (eq_attr "prefix_0f" "0"))))
+  "atom-simple-0")
+
+;; DF shift (prefixed with 0f) is complex insn with latency of 7 cycles
+(define_insn_reservation  "atom_ishift_3" 7
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ishift")
+            (eq_attr "prefix_0f" "1")))
+  "atom-complex, atom-all-eu*6")
+
+(define_insn_reservation  "atom_ishift1" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ishift1")
+            (eq_attr "memory" "none")))
+  "atom-simple-0")
+
+(define_insn_reservation  "atom_ishift1_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ishift1")
+            (eq_attr "memory" "!none")))
+  "atom-simple-0")
+
+(define_insn_reservation  "atom_rotate" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "rotate")
+            (eq_attr "memory" "none")))
+  "atom-simple-0")
+
+(define_insn_reservation  "atom_rotate_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "rotate")
+            (eq_attr "memory" "!none")))
+  "atom-simple-0")
+
+(define_insn_reservation  "atom_rotate1" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "rotate1")
+            (eq_attr "memory" "none")))
+  "atom-simple-0")
+
+(define_insn_reservation  "atom_rotate1_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "rotate1")
+            (eq_attr "memory" "!none")))
+  "atom-simple-0")
+
+(define_insn_reservation  "atom_imul" 5
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imul")
+            (and (eq_attr "memory" "none") (eq_attr "mode" "SI"))))
+  "atom-imul-32")
+
+(define_insn_reservation  "atom_imul_mem" 5
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imul")
+            (and (eq_attr "memory" "!none") (eq_attr "mode" "SI"))))
+  "atom-imul-32")
+
+;; latency set to 10 as common 64x64 imul
+(define_insn_reservation  "atom_imul_3" 10
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "imul")
+            (eq_attr "mode" "!SI")))
+  "atom-complex, atom-all-eu*9")
+
+(define_insn_reservation  "atom_idiv" 65
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "idiv"))
+  "atom-complex, atom-all-eu*32, nothing*32")
+
+(define_insn_reservation  "atom_icmp" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "icmp")
+            (eq_attr "memory" "none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_icmp_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "icmp")
+            (eq_attr "memory" "!none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_test" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "test")
+            (eq_attr "memory" "none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_test_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "test")
+            (eq_attr "memory" "!none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_ibr" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ibr")
+            (eq_attr "memory" "!load")))
+  "atom-simple-1")
+
+;; complex if jump target is from address
+(define_insn_reservation  "atom_ibr_2" 2
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ibr")
+            (eq_attr "memory" "load")))
+  "atom-complex, atom-all-eu")
+
+(define_insn_reservation  "atom_setcc" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "setcc")
+            (eq_attr "memory" "!store")))
+  "atom-simple-either")
+
+;; 2 cycles complex if target is in memory
+(define_insn_reservation  "atom_setcc_2" 2
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "setcc")
+            (eq_attr "memory" "store")))
+  "atom-complex, atom-all-eu")
+
+(define_insn_reservation  "atom_icmov" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "icmov")
+            (eq_attr "memory" "none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_icmov_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "icmov")
+            (eq_attr "memory" "!none")))
+  "atom-simple-either")
+
+;; UCODE if segreg, ignored
+(define_insn_reservation  "atom_push" 2
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "push"))
+  "atom-dual-2c")
+
+;; pop r64 is 1 cycle. UCODE if segreg, ignored
+(define_insn_reservation  "atom_pop" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "pop")
+            (eq_attr "mode" "DI")))
+  "atom-dual-1c")
+
+;; pop non-r64 is 2 cycles. UCODE if segreg, ignored
+(define_insn_reservation  "atom_pop_2" 2
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "pop")
+            (eq_attr "mode" "!DI")))
+  "atom-dual-2c")
+
+;; UCODE if segreg, ignored
+(define_insn_reservation  "atom_call" 1
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "call"))
+  "atom-dual-1c")
+
+(define_insn_reservation  "atom_callv" 1
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "callv"))
+  "atom-dual-1c")
+
+(define_insn_reservation  "atom_leave" 3
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "leave"))
+  "atom-complex, atom-all-eu*2")
+
+(define_insn_reservation  "atom_str" 3
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "str"))
+  "atom-complex, atom-all-eu*2")
+
+(define_insn_reservation  "atom_sselog" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sselog")
+            (eq_attr "memory" "none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_sselog_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sselog")
+            (eq_attr "memory" "!none")))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_sselog1" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sselog1")
+            (eq_attr "memory" "none")))
+  "atom-simple-0")
+
+(define_insn_reservation  "atom_sselog1_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sselog1")
+            (eq_attr "memory" "!none")))
+  "atom-simple-0")
+
+;; not pmad, not psad
+(define_insn_reservation  "atom_sseiadd" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseiadd")
+            (and (not (match_operand:V2DI 0 "register_operand"))
+                 (and (eq_attr "atom_unit" "!simul")
+                      (eq_attr "atom_unit" "!complex")))))
+  "atom-simple-either")
+
+;; pmad, psad and 64
+(define_insn_reservation  "atom_sseiadd_2" 4
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseiadd")
+            (and (not (match_operand:V2DI 0 "register_operand"))
+                 (and (eq_attr "atom_unit" "simul" )
+                      (eq_attr "mode" "DI")))))
+  "atom-fmul-4c")
+
+;; pmad, psad and 128
+(define_insn_reservation  "atom_sseiadd_3" 5
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseiadd")
+            (and (not (match_operand:V2DI 0 "register_operand"))
+                 (and (eq_attr "atom_unit" "simul" )
+                      (eq_attr "mode" "TI")))))
+  "atom-fmul-5c")
+
+;; if paddq(64 bit op), phadd/phsub
+(define_insn_reservation  "atom_sseiadd_4" 6
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseiadd")
+            (ior (match_operand:V2DI 0 "register_operand")
+                 (eq_attr "atom_unit" "complex"))))
+  "atom-complex, atom-all-eu*5")
+
+;; if immediate op. 
+(define_insn_reservation  "atom_sseishft" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseishft")
+            (and (eq_attr "atom_unit" "!sishuf")
+                 (match_operand 2 "immediate_operand"))))
+  "atom-simple-either")
+
+;; if palignr or psrldq
+(define_insn_reservation  "atom_sseishft_2" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseishft")
+            (and (eq_attr "atom_unit" "sishuf")
+                 (match_operand 2 "immediate_operand"))))
+  "atom-simple-0")
+
+;; if reg/mem op
+(define_insn_reservation  "atom_sseishft_3" 2
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseishft")
+            (not (match_operand 2 "immediate_operand"))))
+  "atom-complex, atom-all-eu")
+
+(define_insn_reservation  "atom_sseimul" 1
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "sseimul"))
+  "atom-simple-0")
+
+;; rcpss or rsqrtss
+(define_insn_reservation  "atom_sse" 4
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sse")
+            (and (eq_attr "atom_sse_attr" "rcp") (eq_attr "mode" "SF"))))
+  "atom-fmul-4c")
+
+;; movshdup, movsldup. Suggest to type sseishft
+(define_insn_reservation  "atom_sse_2" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sse")
+            (eq_attr "atom_sse_attr" "movdup")))
+  "atom-simple-0")
+
+;; lfence
+(define_insn_reservation  "atom_sse_3" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sse")
+            (eq_attr "atom_sse_attr" "lfence")))
+  "atom-simple-either")
+
+;; sfence,clflush,mfence, prefetch
+(define_insn_reservation  "atom_sse_4" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sse")
+            (ior (eq_attr "atom_sse_attr" "fence")
+                 (eq_attr "atom_sse_attr" "prefetch"))))
+  "atom-simple-0")
+
+;; rcpps, rsqrtss, sqrt, ldmxcsr
+(define_insn_reservation  "atom_sse_5" 7
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sse")
+            (ior (ior (eq_attr "atom_sse_attr" "sqrt")
+                      (eq_attr "atom_sse_attr" "mxcsr"))
+                 (and (eq_attr "atom_sse_attr" "rcp")
+                      (eq_attr "mode" "V4SF")))))
+  "atom-complex, atom-all-eu*6")
+
+;; xmm->xmm
+(define_insn_reservation  "atom_ssemov" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssemov")
+            (and (match_operand 0 "register_operand" "xy") (match_operand 1 "register_operand" "xy"))))
+  "atom-simple-either")
+
+;; reg->xmm
+(define_insn_reservation  "atom_ssemov_2" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssemov")
+            (and (match_operand 0 "register_operand" "xy") (match_operand 1 "register_operand" "r"))))
+  "atom-simple-0")
+
+;; xmm->reg
+(define_insn_reservation  "atom_ssemov_3" 3
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssemov")
+            (and (match_operand 0 "register_operand" "r") (match_operand 1 "register_operand" "xy"))))
+  "atom-eu-0-3-1")
+
+;; mov mem
+(define_insn_reservation  "atom_ssemov_4" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssemov")
+            (and (eq_attr "movu" "0") (eq_attr "memory" "!none"))))
+  "atom-simple-0")
+
+;; movu mem
+(define_insn_reservation  "atom_ssemov_5" 2
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssemov")
+            (ior (eq_attr "movu" "1") (eq_attr "memory" "!none"))))
+  "atom-complex, atom-all-eu")
+
+;; no memory simple
+(define_insn_reservation  "atom_sseadd" 5
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseadd")
+            (and (eq_attr "memory" "none")
+                 (and (eq_attr "mode" "!V2DF")
+                      (eq_attr "atom_unit" "!complex")))))
+  "atom-fadd-5c")
+
+;; memory simple
+(define_insn_reservation  "atom_sseadd_mem" 5
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseadd")
+            (and (eq_attr "memory" "!none")
+                 (and (eq_attr "mode" "!V2DF")
+                      (eq_attr "atom_unit" "!complex")))))
+  "atom-dual-5c")
+
+;; maxps, minps, *pd, hadd, hsub
+(define_insn_reservation  "atom_sseadd_3" 8
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseadd")
+            (ior (eq_attr "mode" "V2DF") (eq_attr "atom_unit" "complex"))))
+  "atom-complex, atom-all-eu*7")
+
+;; Except dppd/dpps
+(define_insn_reservation  "atom_ssemul" 5
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssemul")
+            (eq_attr "mode" "!SF")))
+  "atom-fmul-5c")
+
+;; Except dppd/dpps, 4 cycle if mulss
+(define_insn_reservation  "atom_ssemul_2" 4
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssemul")
+            (eq_attr "mode" "SF")))
+  "atom-fmul-4c")
+
+(define_insn_reservation  "atom_ssecmp" 1
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "ssecmp"))
+  "atom-simple-either")
+
+(define_insn_reservation  "atom_ssecomi" 10
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "ssecomi"))
+  "atom-complex, atom-all-eu*9")
+
+;; no memory and cvtpi2ps, cvtps2pi, cvttps2pi
+(define_insn_reservation  "atom_ssecvt" 5
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssecvt")
+            (ior (and (match_operand:V2SI 0 "register_operand")
+                      (match_operand:V4SF 1 "register_operand"))
+                 (and (match_operand:V4SF 0 "register_operand")
+                      (match_operand:V2SI 1 "register_operand")))))
+  "atom-fadd-5c")
+
+;; memory and cvtpi2ps, cvtps2pi, cvttps2pi
+(define_insn_reservation  "atom_ssecvt_2" 5
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssecvt")
+            (ior (and (match_operand:V2SI 0 "register_operand")
+                      (match_operand:V4SF 1 "memory_operand"))
+                 (and (match_operand:V4SF 0 "register_operand")
+                      (match_operand:V2SI 1 "memory_operand")))))
+  "atom-dual-5c")
+
+;; otherwise. 7 cycles average for cvtss2sd
+(define_insn_reservation  "atom_ssecvt_3" 7
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "ssecvt")
+            (not (ior (and (match_operand:V2SI 0 "register_operand")
+                           (match_operand:V4SF 1 "nonimmediate_operand"))
+                      (and (match_operand:V4SF 0 "register_operand")
+                           (match_operand:V2SI 1 "nonimmediate_operand"))))))
+  "atom-complex, atom-all-eu*6")
+
+;; memory and cvtsi2sd
+(define_insn_reservation  "atom_sseicvt" 5
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseicvt")
+            (and (match_operand:V2DF 0 "register_operand")
+                 (match_operand:SI 1 "memory_operand"))))
+  "atom-dual-5c")
+
+;; otherwise. 8 cycles average for cvtsd2si
+(define_insn_reservation  "atom_sseicvt_2" 8
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "sseicvt")
+            (not (and (match_operand:V2DF 0 "register_operand")
+                      (match_operand:SI 1 "memory_operand")))))
+  "atom-complex, atom-all-eu*7")
+
+(define_insn_reservation  "atom_ssediv" 62
+  (and (eq_attr "cpu" "atom")
+       (eq_attr "type" "ssediv"))
+  "atom-complex, atom-all-eu*12, nothing*49")
+
+;; simple for fmov
+(define_insn_reservation  "atom_fmov" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "fmov")
+            (eq_attr "memory" "none")))
+  "atom-simple-either")
+
+;; simple for fmov
+(define_insn_reservation  "atom_fmov_mem" 1
+  (and (eq_attr "cpu" "atom")
+       (and (eq_attr "type" "fmov")
+            (eq_attr "memory" "!none")))
+  "atom-simple-either")
+
+;; Define bypass here
+
+;; There will be no stall from lea to non-mem EX insns
+(define_bypass 0 "atom_lea"
+                 "atom_alu_carry,
+                  atom_alu,atom_alu1,atom_negnot,atom_imov,atom_imovx,
+                  atom_incdec, atom_setcc, atom_icmov, atom_pop")
+
+(define_bypass 0 "atom_lea"
+                 "atom_alu_mem, atom_alu_carry_mem, atom_alu1_mem,
+                  atom_imovx_mem, atom_imovx_2_mem,
+                  atom_imov_mem, atom_icmov_mem, atom_fmov_mem"
+                 "!ix86_agi_dependent")
+
+;; There will be 3 cycles stall from EX insns to AGAN insns LEA
+(define_bypass 4 "atom_alu_carry,
+                  atom_alu,atom_alu1,atom_negnot,atom_imov,atom_imovx,
+                  atom_incdec,atom_ishift,atom_ishift1,atom_rotate,
+                  atom_rotate1, atom_setcc, atom_icmov, atom_pop,
+                  atom_alu_mem, atom_alu_carry_mem, atom_alu1_mem,
+                  atom_imovx_mem, atom_imovx_2_mem,
+                  atom_imov_mem, atom_icmov_mem, atom_fmov_mem"
+                 "atom_lea")
+
+;; There will be 3 cycles stall from EX insns to insns need addr calculation
+(define_bypass 4 "atom_alu_carry,
+                  atom_alu,atom_alu1,atom_negnot,atom_imov,atom_imovx,
+                  atom_incdec,atom_ishift,atom_ishift1,atom_rotate,
+                  atom_rotate1, atom_setcc, atom_icmov, atom_pop,
+                  atom_imovx_mem, atom_imovx_2_mem,
+                  atom_alu_mem, atom_alu_carry_mem, atom_alu1_mem,
+                  atom_imov_mem, atom_icmov_mem, atom_fmov_mem"
+                 "atom_alu_mem, atom_alu_carry_mem, atom_alu1_mem,
+                  atom_negnot_mem, atom_imov_mem, atom_incdec_mem,
+                  atom_imovx_mem, atom_imovx_2_mem,
+                  atom_imul_mem, atom_icmp_mem,
+                  atom_test_mem, atom_icmov_mem, atom_sselog_mem,
+                  atom_sselog1_mem, atom_fmov_mem, atom_sseadd_mem,
+                  atom_ishift_mem, atom_ishift1_mem, 
+                  atom_rotate_mem, atom_rotate1_mem"
+                  "ix86_agi_dependent")
+
+;; Stall from imul to lea is 8 cycles.
+(define_bypass 9 "atom_imul, atom_imul_mem" "atom_lea")
+
+;; Stall from imul to memory address is 8 cycles.
+(define_bypass 9 "atom_imul, atom_imul_mem" 
+                 "atom_alu_mem, atom_alu_carry_mem, atom_alu1_mem,
+                  atom_negnot_mem, atom_imov_mem, atom_incdec_mem,
+                  atom_ishift_mem, atom_ishift1_mem, atom_rotate_mem,
+                  atom_rotate1_mem, atom_imul_mem, atom_icmp_mem,
+                  atom_test_mem, atom_icmov_mem, atom_sselog_mem,
+                  atom_sselog1_mem, atom_fmov_mem, atom_sseadd_mem"
+                  "ix86_agi_dependent")
+
+;; There will be 0 cycle stall from cmp/test to jcc
+
+;; There will be 1 cycle stall from flag producer to cmov and adc/sbb
+(define_bypass 2 "atom_icmp, atom_test, atom_alu, atom_alu_carry,
+                  atom_alu1, atom_negnot, atom_incdec, atom_ishift,
+                  atom_ishift1, atom_rotate, atom_rotate1"
+                 "atom_icmov, atom_alu_carry")

gcc/config/i386/i386-c.c

@@ -119,6 +119,10 @@ ix86_target_macros_internal (int isa_flag,
       def_or_undef (parse_in, "__core2");
       def_or_undef (parse_in, "__core2__");
       break;
+    case PROCESSOR_ATOM:
+      def_or_undef (parse_in, "__atom");
+      def_or_undef (parse_in, "__atom__");
+      break;
     /* use PROCESSOR_max to not set/unset the arch macro.  */
     case PROCESSOR_max:
       break;
@@ -187,6 +191,9 @@ ix86_target_macros_internal (int isa_flag,
     case PROCESSOR_CORE2:
       def_or_undef (parse_in, "__tune_core2__");
       break;
+    case PROCESSOR_ATOM:
+      def_or_undef (parse_in, "__tune_atom__");
+      break;
     case PROCESSOR_GENERIC32:
     case PROCESSOR_GENERIC64:
       break;

gcc/config/i386/i386.c

@@ -1036,6 +1036,79 @@ struct processor_costs core2_cost = {
   1,                                    /* cond_not_taken_branch_cost.  */
 };
 
+static const
+struct processor_costs atom_cost = {
+  COSTS_N_INSNS (1),			/* cost of an add instruction */
+  COSTS_N_INSNS (1) + 1,		/* cost of a lea instruction */
+  COSTS_N_INSNS (1),			/* variable shift costs */
+  COSTS_N_INSNS (1),			/* constant shift costs */
+  {COSTS_N_INSNS (3),			/* cost of starting multiply for QI */
+   COSTS_N_INSNS (4),			/*                               HI */
+   COSTS_N_INSNS (3),			/*                               SI */
+   COSTS_N_INSNS (4),			/*                               DI */
+   COSTS_N_INSNS (2)},			/*                               other */
+  0,					/* cost of multiply per each bit set */
+  {COSTS_N_INSNS (18),			/* cost of a divide/mod for QI */
+   COSTS_N_INSNS (26),			/*                          HI */
+   COSTS_N_INSNS (42),			/*                          SI */
+   COSTS_N_INSNS (74),			/*                          DI */
+   COSTS_N_INSNS (74)},			/*                          other */
+  COSTS_N_INSNS (1),			/* cost of movsx */
+  COSTS_N_INSNS (1),			/* cost of movzx */
+  8,					/* "large" insn */
+  17,					/* MOVE_RATIO */
+  2,					/* cost for loading QImode using movzbl */
+  {4, 4, 4},				/* cost of loading integer registers
+					   in QImode, HImode and SImode.
+					   Relative to reg-reg move (2).  */
+  {4, 4, 4},				/* cost of storing integer registers */
+  4,					/* cost of reg,reg fld/fst */
+  {12, 12, 12},				/* cost of loading fp registers
+					   in SFmode, DFmode and XFmode */
+  {6, 6, 8},				/* cost of storing fp registers
+					   in SFmode, DFmode and XFmode */
+  2,					/* cost of moving MMX register */
+  {8, 8},				/* cost of loading MMX registers
+					   in SImode and DImode */
+  {8, 8},				/* cost of storing MMX registers
+					   in SImode and DImode */
+  2,					/* cost of moving SSE register */
+  {8, 8, 8},				/* cost of loading SSE registers
+					   in SImode, DImode and TImode */
+  {8, 8, 8},				/* cost of storing SSE registers
+					   in SImode, DImode and TImode */
+  5,					/* MMX or SSE register to integer */
+  32,					/* size of l1 cache.  */
+  256,					/* size of l2 cache.  */
+  64,					/* size of prefetch block */
+  6,					/* number of parallel prefetches */
+  3,					/* Branch cost */
+  COSTS_N_INSNS (8),			/* cost of FADD and FSUB insns.  */
+  COSTS_N_INSNS (8),			/* cost of FMUL instruction.  */
+  COSTS_N_INSNS (20),			/* cost of FDIV instruction.  */
+  COSTS_N_INSNS (8),			/* cost of FABS instruction.  */
+  COSTS_N_INSNS (8),			/* cost of FCHS instruction.  */
+  COSTS_N_INSNS (40),			/* cost of FSQRT instruction.  */
+  {{libcall, {{11, loop}, {-1, rep_prefix_4_byte}}},
+   {libcall, {{32, loop}, {64, rep_prefix_4_byte},
+          {8192, rep_prefix_8_byte}, {-1, libcall}}}},
+  {{libcall, {{8, loop}, {15, unrolled_loop},
+          {2048, rep_prefix_4_byte}, {-1, libcall}}},
+   {libcall, {{24, loop}, {32, unrolled_loop},
+          {8192, rep_prefix_8_byte}, {-1, libcall}}}},
+  1,                                    /* scalar_stmt_cost.  */
+  1,                                    /* scalar load_cost.  */
+  1,                                    /* scalar_store_cost.  */
+  1,                                    /* vec_stmt_cost.  */
+  1,                                    /* vec_to_scalar_cost.  */
+  1,                                    /* scalar_to_vec_cost.  */
+  1,                                    /* vec_align_load_cost.  */
+  2,                                    /* vec_unalign_load_cost.  */
+  1,                                    /* vec_store_cost.  */
+  3,                                    /* cond_taken_branch_cost.  */
+  1,                                    /* cond_not_taken_branch_cost.  */
+};
+
 /* Generic64 should produce code tuned for Nocona and K8.  */
 static const
 struct processor_costs generic64_cost = {
@@ -1194,6 +1267,7 @@ const struct processor_costs *ix86_cost = &pentium_cost;
 #define m_PENT4  (1<<PROCESSOR_PENTIUM4)
 #define m_NOCONA  (1<<PROCESSOR_NOCONA)
 #define m_CORE2  (1<<PROCESSOR_CORE2)
+#define m_ATOM  (1<<PROCESSOR_ATOM)
 
 #define m_GEODE  (1<<PROCESSOR_GEODE)
 #define m_K6  (1<<PROCESSOR_K6)
@@ -1231,10 +1305,11 @@ static unsigned int initial_ix86_tune_features[X86_TUNE_LAST] = {
   m_486 | m_PENT,
 
   /* X86_TUNE_UNROLL_STRLEN */
-  m_486 | m_PENT | m_PPRO | m_AMD_MULTIPLE | m_K6 | m_CORE2 | m_GENERIC,
+  m_486 | m_PENT | m_ATOM | m_PPRO | m_AMD_MULTIPLE | m_K6
+  | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_DEEP_BRANCH_PREDICTION */
-  m_PPRO | m_K6_GEODE | m_AMD_MULTIPLE | m_PENT4 | m_GENERIC,
+  m_ATOM | m_PPRO | m_K6_GEODE | m_AMD_MULTIPLE | m_PENT4 | m_GENERIC,
 
   /* X86_TUNE_BRANCH_PREDICTION_HINTS: Branch hints were put in P4 based
      on simulation result. But after P4 was made, no performance benefit
@@ -1246,12 +1321,12 @@ static unsigned int initial_ix86_tune_features[X86_TUNE_LAST] = {
   ~m_386,
 
   /* X86_TUNE_USE_SAHF */
-  m_PPRO | m_K6_GEODE | m_K8 | m_AMDFAM10 | m_PENT4
+  m_ATOM | m_PPRO | m_K6_GEODE | m_K8 | m_AMDFAM10 | m_PENT4
   | m_NOCONA | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_MOVX: Enable to zero extend integer registers to avoid
      partial dependencies.  */
-  m_AMD_MULTIPLE | m_PPRO | m_PENT4 | m_NOCONA
+  m_AMD_MULTIPLE | m_ATOM | m_PPRO | m_PENT4 | m_NOCONA
   | m_CORE2 | m_GENERIC | m_GEODE /* m_386 | m_K6 */,
 
   /* X86_TUNE_PARTIAL_REG_STALL: We probably ought to watch for partial
@@ -1271,13 +1346,13 @@ static unsigned int initial_ix86_tune_features[X86_TUNE_LAST] = {
   m_386 | m_486 | m_K6_GEODE,
 
   /* X86_TUNE_USE_SIMODE_FIOP */
-  ~(m_PPRO | m_AMD_MULTIPLE | m_PENT | m_CORE2 | m_GENERIC),
+  ~(m_PPRO | m_AMD_MULTIPLE | m_PENT | m_ATOM | m_CORE2 | m_GENERIC),
 
   /* X86_TUNE_USE_MOV0 */
   m_K6,
 
   /* X86_TUNE_USE_CLTD */
-  ~(m_PENT | m_K6 | m_CORE2 | m_GENERIC),
+  ~(m_PENT | m_ATOM | m_K6 | m_CORE2 | m_GENERIC),
 
   /* X86_TUNE_USE_XCHGB: Use xchgb %rh,%rl instead of rolw/rorw $8,rx.  */
   m_PENT4,
@@ -1292,8 +1367,8 @@ static unsigned int initial_ix86_tune_features[X86_TUNE_LAST] = {
   ~(m_PENT | m_PPRO),
 
   /* X86_TUNE_PROMOTE_QIMODE */
-  m_K6_GEODE | m_PENT | m_386 | m_486 | m_AMD_MULTIPLE | m_CORE2
-  | m_GENERIC /* | m_PENT4 ? */,
+  m_K6_GEODE | m_PENT | m_ATOM | m_386 | m_486 | m_AMD_MULTIPLE
+  | m_CORE2 | m_GENERIC /* | m_PENT4 ? */,
 
   /* X86_TUNE_FAST_PREFIX */
   ~(m_PENT | m_486 | m_386),
@@ -1317,26 +1392,28 @@ static unsigned int initial_ix86_tune_features[X86_TUNE_LAST] = {
   m_PPRO,
 
   /* X86_TUNE_ADD_ESP_4: Enable if add/sub is preferred over 1/2 push/pop.  */
-  m_AMD_MULTIPLE | m_K6_GEODE | m_PENT4 | m_NOCONA | m_CORE2 | m_GENERIC,
+  m_ATOM | m_AMD_MULTIPLE | m_K6_GEODE | m_PENT4 | m_NOCONA
+  | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_ADD_ESP_8 */
-  m_AMD_MULTIPLE | m_PPRO | m_K6_GEODE | m_386
+  m_AMD_MULTIPLE | m_ATOM | m_PPRO | m_K6_GEODE | m_386
   | m_486 | m_PENT4 | m_NOCONA | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_SUB_ESP_4 */
-  m_AMD_MULTIPLE | m_PPRO | m_PENT4 | m_NOCONA | m_CORE2 | m_GENERIC,
+  m_AMD_MULTIPLE | m_ATOM | m_PPRO | m_PENT4 | m_NOCONA | m_CORE2
+  | m_GENERIC,
 
   /* X86_TUNE_SUB_ESP_8 */
-  m_AMD_MULTIPLE | m_PPRO | m_386 | m_486
+  m_AMD_MULTIPLE | m_ATOM | m_PPRO | m_386 | m_486
   | m_PENT4 | m_NOCONA | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_INTEGER_DFMODE_MOVES: Enable if integer moves are preferred
      for DFmode copies */
-  ~(m_AMD_MULTIPLE | m_PENT4 | m_NOCONA | m_PPRO | m_CORE2
+  ~(m_AMD_MULTIPLE | m_ATOM | m_PENT4 | m_NOCONA | m_PPRO | m_CORE2
     | m_GENERIC | m_GEODE),
 
   /* X86_TUNE_PARTIAL_REG_DEPENDENCY */
-  m_AMD_MULTIPLE | m_PENT4 | m_NOCONA | m_CORE2 | m_GENERIC,
+  m_AMD_MULTIPLE | m_ATOM | m_PENT4 | m_NOCONA | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_SSE_PARTIAL_REG_DEPENDENCY: In the Generic model we have a
      conflict here in between PPro/Pentium4 based chips that thread 128bit
@@ -1347,7 +1424,8 @@ static unsigned int initial_ix86_tune_features[X86_TUNE_LAST] = {
      shows that disabling this option on P4 brings over 20% SPECfp regression,
      while enabling it on K8 brings roughly 2.4% regression that can be partly
      masked by careful scheduling of moves.  */
-  m_PENT4 | m_NOCONA | m_PPRO | m_CORE2 | m_GENERIC | m_AMDFAM10,
+  m_ATOM | m_PENT4 | m_NOCONA | m_PPRO | m_CORE2 | m_GENERIC
+  | m_AMDFAM10,
 
   /* X86_TUNE_SSE_UNALIGNED_MOVE_OPTIMAL */
   m_AMDFAM10,
@@ -1365,13 +1443,13 @@ static unsigned int initial_ix86_tune_features[X86_TUNE_LAST] = {
   m_PPRO | m_PENT4 | m_NOCONA,
 
   /* X86_TUNE_MEMORY_MISMATCH_STALL */
-  m_AMD_MULTIPLE | m_PENT4 | m_NOCONA | m_CORE2 | m_GENERIC,
+  m_AMD_MULTIPLE | m_ATOM | m_PENT4 | m_NOCONA | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_PROLOGUE_USING_MOVE */
-  m_ATHLON_K8 | m_PPRO | m_CORE2 | m_GENERIC,
+  m_ATHLON_K8 | m_ATOM | m_PPRO | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_EPILOGUE_USING_MOVE */
-  m_ATHLON_K8 | m_PPRO | m_CORE2 | m_GENERIC,
+  m_ATHLON_K8 | m_ATOM | m_PPRO | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_SHIFT1 */
   ~m_486,
@@ -1380,29 +1458,32 @@ static unsigned int initial_ix86_tune_features[X86_TUNE_LAST] = {
   m_AMD_MULTIPLE,
 
   /* X86_TUNE_INTER_UNIT_MOVES */
-  ~(m_AMD_MULTIPLE | m_GENERIC),
+  ~(m_AMD_MULTIPLE | m_ATOM | m_GENERIC),
 
   /* X86_TUNE_INTER_UNIT_CONVERSIONS */
   ~(m_AMDFAM10),
 
   /* X86_TUNE_FOUR_JUMP_LIMIT: Some CPU cores are not able to predict more
      than 4 branch instructions in the 16 byte window.  */
-  m_PPRO | m_AMD_MULTIPLE | m_PENT4 | m_NOCONA | m_CORE2 | m_GENERIC,
+  m_ATOM | m_PPRO | m_AMD_MULTIPLE | m_PENT4 | m_NOCONA | m_CORE2
+  | m_GENERIC,
 
   /* X86_TUNE_SCHEDULE */
-  m_PPRO | m_AMD_MULTIPLE | m_K6_GEODE | m_PENT | m_CORE2 | m_GENERIC,
+  m_PPRO | m_AMD_MULTIPLE | m_K6_GEODE | m_PENT | m_ATOM | m_CORE2
+  | m_GENERIC,
 
   /* X86_TUNE_USE_BT */
-  m_AMD_MULTIPLE | m_CORE2 | m_GENERIC,
+  m_AMD_MULTIPLE | m_ATOM | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_USE_INCDEC */
-  ~(m_PENT4 | m_NOCONA | m_GENERIC),
+  ~(m_PENT4 | m_NOCONA | m_GENERIC | m_ATOM),
 
   /* X86_TUNE_PAD_RETURNS */
   m_AMD_MULTIPLE | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_EXT_80387_CONSTANTS */
-  m_K6_GEODE | m_ATHLON_K8 | m_PENT4 | m_NOCONA | m_PPRO | m_CORE2 | m_GENERIC,
+  m_K6_GEODE | m_ATHLON_K8 | m_ATOM | m_PENT4 | m_NOCONA | m_PPRO
+  | m_CORE2 | m_GENERIC,
 
   /* X86_TUNE_SHORTEN_X87_SSE */
   ~m_K8,
@@ -1447,6 +1528,10 @@ static unsigned int initial_ix86_tune_features[X86_TUNE_LAST] = {
      with a subsequent conditional jump instruction into a single
      compare-and-branch uop.  */
   m_CORE2,
+
+  /* X86_TUNE_OPT_AGU: Optimize for Address Generation Unit. This flag
+     will impact LEA instruction selection. */
+  m_ATOM,
 };
 
 /* Feature tests against the various architecture variations.  */
@@ -1472,10 +1557,11 @@ static unsigned int initial_ix86_arch_features[X86_ARCH_LAST] = {
 };
 
 static const unsigned int x86_accumulate_outgoing_args
-  = m_AMD_MULTIPLE | m_PENT4 | m_NOCONA | m_PPRO | m_CORE2 | m_GENERIC;
+  = m_AMD_MULTIPLE | m_ATOM | m_PENT4 | m_NOCONA | m_PPRO | m_CORE2
+    | m_GENERIC;
 
 static const unsigned int x86_arch_always_fancy_math_387
-  = m_PENT | m_PPRO | m_AMD_MULTIPLE | m_PENT4
+  = m_PENT | m_ATOM | m_PPRO | m_AMD_MULTIPLE | m_PENT4
     | m_NOCONA | m_CORE2 | m_GENERIC;
 
 static enum stringop_alg stringop_alg = no_stringop;
@@ -1958,7 +2044,8 @@ static const struct ptt processor_target_table[PROCESSOR_max] =
   {&core2_cost, 16, 10, 16, 10, 16},
   {&generic32_cost, 16, 7, 16, 7, 16},
   {&generic64_cost, 16, 10, 16, 10, 16},
-  {&amdfam10_cost, 32, 24, 32, 7, 32}
+  {&amdfam10_cost, 32, 24, 32, 7, 32},
+  {&atom_cost, 16, 7, 16, 7, 16}
 };
 
 static const char *const cpu_names[TARGET_CPU_DEFAULT_max] =
@@ -1976,6 +2063,7 @@ static const char *const cpu_names[TARGET_CPU_DEFAULT_max] =
   "prescott",
   "nocona",
   "core2",
+  "atom",
   "geode",
   "k6",
   "k6-2",
@@ -2534,6 +2622,9 @@ override_options (bool main_args_p)
       {"core2", PROCESSOR_CORE2, CPU_CORE2,
 	PTA_64BIT | PTA_MMX | PTA_SSE | PTA_SSE2 | PTA_SSE3
 	| PTA_SSSE3 | PTA_CX16},
+      {"atom", PROCESSOR_ATOM, CPU_ATOM,
+	PTA_64BIT | PTA_MMX | PTA_SSE | PTA_SSE2 | PTA_SSE3
+	| PTA_SSSE3 | PTA_CX16},
       {"geode", PROCESSOR_GEODE, CPU_GEODE,
 	PTA_MMX | PTA_3DNOW | PTA_3DNOW_A |PTA_PREFETCH_SSE},
       {"k6", PROCESSOR_K6, CPU_K6, PTA_MMX},
@@ -19026,6 +19117,7 @@ ix86_issue_rate (void)
   switch (ix86_tune)
     {
     case PROCESSOR_PENTIUM:
+    case PROCESSOR_ATOM:
     case PROCESSOR_K6:
       return 2;
 
@@ -19226,6 +19318,7 @@ ix86_adjust_cost (rtx insn, rtx link, rtx dep_insn, int cost)
     case PROCESSOR_ATHLON:
     case PROCESSOR_K8:
     case PROCESSOR_AMDFAM10:
+    case PROCESSOR_ATOM:
     case PROCESSOR_GENERIC32:
     case PROCESSOR_GENERIC64:
       memory = get_attr_memory (insn);

gcc/config/i386/i386.h

@@ -231,6 +231,7 @@ extern const struct processor_costs ix86_size_cost;
 #define TARGET_GENERIC64 (ix86_tune == PROCESSOR_GENERIC64)
 #define TARGET_GENERIC (TARGET_GENERIC32 || TARGET_GENERIC64)
 #define TARGET_AMDFAM10 (ix86_tune == PROCESSOR_AMDFAM10)
+#define TARGET_ATOM (ix86_tune == PROCESSOR_ATOM)
 
 /* Feature tests against the various tunings.  */
 enum ix86_tune_indices {
@@ -295,6 +296,7 @@ enum ix86_tune_indices {
   X86_TUNE_USE_VECTOR_FP_CONVERTS,
   X86_TUNE_USE_VECTOR_CONVERTS,
   X86_TUNE_FUSE_CMP_AND_BRANCH,
+  X86_TUNE_OPT_AGU,
 
   X86_TUNE_LAST
 };
@@ -382,6 +384,7 @@ extern unsigned char ix86_tune_features[X86_TUNE_LAST];
 	ix86_tune_features[X86_TUNE_USE_VECTOR_CONVERTS]
 #define TARGET_FUSE_CMP_AND_BRANCH \
 	ix86_tune_features[X86_TUNE_FUSE_CMP_AND_BRANCH]
+#define TARGET_OPT_AGU ix86_tune_features[X86_TUNE_OPT_AGU]
 
 /* Feature tests against the various architecture variations.  */
 enum ix86_arch_indices {
@@ -567,6 +570,7 @@ enum target_cpu_default
   TARGET_CPU_DEFAULT_prescott,
   TARGET_CPU_DEFAULT_nocona,
   TARGET_CPU_DEFAULT_core2,
+  TARGET_CPU_DEFAULT_atom,
 
   TARGET_CPU_DEFAULT_geode,
   TARGET_CPU_DEFAULT_k6,
@@ -2272,6 +2276,7 @@ enum processor_type
   PROCESSOR_GENERIC32,
   PROCESSOR_GENERIC64,
   PROCESSOR_AMDFAM10,
+  PROCESSOR_ATOM,
   PROCESSOR_max
 };
 

gcc/config/i386/i386.md

@@ -316,7 +316,7 @@
 
 
 ;; Processor type.
-(define_attr "cpu" "none,pentium,pentiumpro,geode,k6,athlon,k8,core2,
+(define_attr "cpu" "none,pentium,pentiumpro,geode,k6,athlon,k8,core2,atom,
 		    generic64,amdfam10"
   (const (symbol_ref "ix86_schedule")))
 
@@ -612,6 +612,12 @@
 (define_attr "i387_cw" "trunc,floor,ceil,mask_pm,uninitialized,any"
   (const_string "any"))
 
+;; Define attribute to classify add/sub insns that consumes carry flag (CF)
+(define_attr "use_carry" "0,1" (const_string "0"))
+
+;; Define attribute to indicate unaligned ssemov insns
+(define_attr "movu" "0,1" (const_string "0"))
+
 ;; Describe a user's asm statement.
 (define_asm_attributes
   [(set_attr "length" "128")
@@ -727,6 +733,7 @@
 (include "k6.md")
 (include "athlon.md")
 (include "geode.md")
+(include "atom.md")
 
 
 ;; Operand and operator predicates and constraints
@@ -5816,6 +5823,7 @@
   "TARGET_64BIT && ix86_binary_operator_ok (PLUS, DImode, operands)"
   "adc{q}\t{%2, %0|%0, %2}"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "mode" "DI")])
 
@@ -5890,6 +5898,7 @@
   "ix86_binary_operator_ok (PLUS, QImode, operands)"
   "adc{b}\t{%2, %0|%0, %2}"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "mode" "QI")])
 
@@ -5902,6 +5911,7 @@
   "ix86_binary_operator_ok (PLUS, HImode, operands)"
   "adc{w}\t{%2, %0|%0, %2}"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "mode" "HI")])
 
@@ -5914,6 +5924,7 @@
   "ix86_binary_operator_ok (PLUS, SImode, operands)"
   "adc{l}\t{%2, %0|%0, %2}"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "mode" "SI")])
 
@@ -5927,6 +5938,7 @@
   "TARGET_64BIT && ix86_binary_operator_ok (PLUS, SImode, operands)"
   "adc{l}\t{%2, %k0|%k0, %2}"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "mode" "SI")])
 
@@ -6156,9 +6168,9 @@
    (set_attr "mode" "SI")])
 
 (define_insn "*adddi_1_rex64"
-  [(set (match_operand:DI 0 "nonimmediate_operand" "=r,rm,r")
-	(plus:DI (match_operand:DI 1 "nonimmediate_operand" "%0,0,r")
-		 (match_operand:DI 2 "x86_64_general_operand" "rme,re,le")))
+  [(set (match_operand:DI 0 "nonimmediate_operand" "=r,rm,r,r")
+	(plus:DI (match_operand:DI 1 "nonimmediate_operand" "%0,0,r,r")
+		 (match_operand:DI 2 "x86_64_general_operand" "rme,re,0,le")))
    (clobber (reg:CC FLAGS_REG))]
   "TARGET_64BIT && ix86_binary_operator_ok (PLUS, DImode, operands)"
 {
@@ -6179,6 +6191,10 @@
 	}
 
     default:
+      /* Use add as much as possible to replace lea for AGU optimization. */
+      if (which_alternative == 2 && TARGET_OPT_AGU)
+        return "add{q}\t{%1, %0|%0, %1}";
+        
       gcc_assert (rtx_equal_p (operands[0], operands[1]));
 
       /* Make things pretty and `subl $4,%eax' rather than `addl $-4, %eax'.
@@ -6197,8 +6213,11 @@
     }
 }
   [(set (attr "type")
-     (cond [(eq_attr "alternative" "2")
+     (cond [(and (eq_attr "alternative" "2") 
+                 (eq (symbol_ref "TARGET_OPT_AGU") (const_int 0)))
 	      (const_string "lea")
+            (eq_attr "alternative" "3")
+              (const_string "lea")
 	    ; Current assemblers are broken and do not allow @GOTOFF in
 	    ; ought but a memory context.
 	    (match_operand:DI 2 "pic_symbolic_operand" "")
@@ -6215,8 +6234,7 @@
 	(plus:DI (match_operand:DI 1 "register_operand" "")
 		 (match_operand:DI 2 "x86_64_nonmemory_operand" "")))
    (clobber (reg:CC FLAGS_REG))]
-  "TARGET_64BIT && reload_completed
-   && true_regnum (operands[0]) != true_regnum (operands[1])"
+  "TARGET_64BIT && reload_completed"
   [(set (match_dup 0)
 	(plus:DI (match_dup 1)
 		 (match_dup 2)))]
@@ -6420,9 +6438,9 @@
 
 
 (define_insn "*addsi_1"
-  [(set (match_operand:SI 0 "nonimmediate_operand" "=r,rm,r")
-	(plus:SI (match_operand:SI 1 "nonimmediate_operand" "%0,0,r")
-		 (match_operand:SI 2 "general_operand" "g,ri,li")))
+  [(set (match_operand:SI 0 "nonimmediate_operand" "=r,rm,r,r")
+	(plus:SI (match_operand:SI 1 "nonimmediate_operand" "%0,0,r,r")
+		 (match_operand:SI 2 "general_operand" "g,ri,0,li")))
    (clobber (reg:CC FLAGS_REG))]
   "ix86_binary_operator_ok (PLUS, SImode, operands)"
 {
@@ -6443,6 +6461,10 @@
 	}
 
     default:
+      /* Use add as much as possible to replace lea for AGU optimization. */
+      if (which_alternative == 2 && TARGET_OPT_AGU)
+        return "add{l}\t{%1, %0|%0, %1}";
+
       gcc_assert (rtx_equal_p (operands[0], operands[1]));
 
       /* Make things pretty and `subl $4,%eax' rather than `addl $-4, %eax'.
@@ -6459,7 +6481,10 @@
     }
 }
   [(set (attr "type")
-     (cond [(eq_attr "alternative" "2")
+     (cond [(and (eq_attr "alternative" "2") 
+                 (eq (symbol_ref "TARGET_OPT_AGU") (const_int 0)))
+               (const_string "lea")
+            (eq_attr "alternative" "3")
 	      (const_string "lea")
 	    ; Current assemblers are broken and do not allow @GOTOFF in
 	    ; ought but a memory context.
@@ -6477,8 +6502,7 @@
 	(plus (match_operand 1 "register_operand" "")
               (match_operand 2 "nonmemory_operand" "")))
    (clobber (reg:CC FLAGS_REG))]
-  "reload_completed
-   && true_regnum (operands[0]) != true_regnum (operands[1])"
+  "reload_completed"
   [(const_int 0)]
 {
   rtx pat;
@@ -7580,6 +7604,7 @@
   "TARGET_64BIT && ix86_binary_operator_ok (MINUS, DImode, operands)"
   "sbb{q}\t{%2, %0|%0, %2}"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "mode" "DI")])
 
@@ -7628,6 +7653,7 @@
   "ix86_binary_operator_ok (MINUS, QImode, operands)"
   "sbb{b}\t{%2, %0|%0, %2}"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "mode" "QI")])
 
@@ -7640,6 +7666,7 @@
   "ix86_binary_operator_ok (MINUS, HImode, operands)"
   "sbb{w}\t{%2, %0|%0, %2}"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "mode" "HI")])
 
@@ -7652,6 +7679,7 @@
   "ix86_binary_operator_ok (MINUS, SImode, operands)"
   "sbb{l}\t{%2, %0|%0, %2}"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "mode" "SI")])
 
@@ -15275,6 +15303,7 @@
   "reload_completed"
   "ret"
   [(set_attr "length" "1")
+   (set_attr "atom_unit" "jeu")
    (set_attr "length_immediate" "0")
    (set_attr "modrm" "0")])
 
@@ -15287,6 +15316,7 @@
   "reload_completed"
   "rep\;ret"
   [(set_attr "length" "1")
+   (set_attr "atom_unit" "jeu")
    (set_attr "length_immediate" "0")
    (set_attr "prefix_rep" "1")
    (set_attr "modrm" "0")])
@@ -15297,6 +15327,7 @@
   "reload_completed"
   "ret\t%0"
   [(set_attr "length" "3")
+   (set_attr "atom_unit" "jeu")
    (set_attr "length_immediate" "2")
    (set_attr "modrm" "0")])
 
@@ -16418,6 +16449,7 @@
   "TARGET_SSE_MATH"
   "%vrcpss\t{%1, %d0|%d0, %1}"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "rcp")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "SF")])
 
@@ -16777,6 +16809,7 @@
   "TARGET_SSE_MATH"
   "%vrsqrtss\t{%1, %d0|%d0, %1}"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "rcp")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "SF")])
 
@@ -16797,6 +16830,7 @@
   "SSE_FLOAT_MODE_P (<MODE>mode) && TARGET_SSE_MATH"
   "%vsqrts<ssemodefsuffix>\t{%1, %d0|%d0, %1}"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "sqrt")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "<MODE>")
    (set_attr "athlon_decode" "*")
@@ -19850,6 +19884,7 @@
   ; Since we don't have the proper number of operands for an alu insn,
   ; fill in all the blanks.
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "memory" "none")
    (set_attr "imm_disp" "false")
@@ -19865,6 +19900,7 @@
   ""
   "sbb{q}\t%0, %0"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "memory" "none")
    (set_attr "imm_disp" "false")
@@ -19908,6 +19944,7 @@
   ; Since we don't have the proper number of operands for an alu insn,
   ; fill in all the blanks.
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "memory" "none")
    (set_attr "imm_disp" "false")
@@ -19923,6 +19960,7 @@
   ""
   "sbb{l}\t%0, %0"
   [(set_attr "type" "alu")
+   (set_attr "use_carry" "1")
    (set_attr "pent_pair" "pu")
    (set_attr "memory" "none")
    (set_attr "imm_disp" "false")
@@ -20255,7 +20293,8 @@
     }
 }
   [(set (attr "type")
-	(cond [(eq_attr "alternative" "0")
+	(cond [(and (eq_attr "alternative" "0") 
+	            (eq (symbol_ref "TARGET_OPT_AGU") (const_int 0)))
 		 (const_string "alu")
 	       (match_operand:SI 2 "const0_operand" "")
 		 (const_string "imov")
@@ -20298,7 +20337,8 @@
     }
 }
   [(set (attr "type")
-	(cond [(eq_attr "alternative" "0")
+	(cond [(and (eq_attr "alternative" "0")
+	            (eq (symbol_ref "TARGET_OPT_AGU") (const_int 0)))
 		 (const_string "alu")
 	       (match_operand:DI 2 "const0_operand" "")
 		 (const_string "imov")
@@ -21790,6 +21830,7 @@
   return patterns[locality];
 }
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "prefetch")
    (set_attr "memory" "none")])
 
 (define_insn "*prefetch_sse_rex"
@@ -21808,6 +21849,7 @@
   return patterns[locality];
 }
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "prefetch")
    (set_attr "memory" "none")])
 
 (define_insn "*prefetch_3dnow"

gcc/config/i386/sse.md

@@ -338,6 +338,7 @@
    && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
   "vmovup<avxmodesuffixf2c>\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssemov")
+   (set_attr "movu" "1")
    (set_attr "prefix" "vex")
    (set_attr "mode" "<MODE>")])
 
@@ -363,6 +364,7 @@
    && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
   "movup<ssemodesuffixf2c>\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssemov")
+   (set_attr "movu" "1")
    (set_attr "mode" "<MODE>")])
 
 (define_insn "avx_movdqu<avxmodesuffix>"
@@ -373,6 +375,7 @@
   "TARGET_AVX && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
   "vmovdqu\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssemov")
+   (set_attr "movu" "1")
    (set_attr "prefix" "vex")
    (set_attr "mode" "<avxvecmode>")])
 
@@ -383,6 +386,7 @@
   "TARGET_SSE2 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
   "movdqu\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssemov")
+   (set_attr "movu" "1")
    (set_attr "prefix_data16" "1")
    (set_attr "mode" "TI")])
 
@@ -424,7 +428,7 @@
 		     UNSPEC_MOVNT))]
   "TARGET_SSE2"
   "movntdq\t{%1, %0|%0, %1}"
-  [(set_attr "type" "ssecvt")
+  [(set_attr "type" "ssemov")
    (set_attr "prefix_data16" "1")
    (set_attr "mode" "TI")])
 
@@ -434,7 +438,7 @@
 		   UNSPEC_MOVNT))]
   "TARGET_SSE2"
   "movnti\t{%1, %0|%0, %1}"
-  [(set_attr "type" "ssecvt")
+  [(set_attr "type" "ssemov")
    (set_attr "mode" "V2DF")])
 
 (define_insn "avx_lddqu<avxmodesuffix>"
@@ -445,6 +449,7 @@
   "TARGET_AVX"
   "vlddqu\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssecvt")
+   (set_attr "movu" "1")
    (set_attr "prefix" "vex")
    (set_attr "mode" "<avxvecmode>")])
 
@@ -454,7 +459,8 @@
 		      UNSPEC_LDDQU))]
   "TARGET_SSE3"
   "lddqu\t{%1, %0|%0, %1}"
-  [(set_attr "type" "ssecvt")
+  [(set_attr "type" "ssemov")
+   (set_attr "movu" "1")
    (set_attr "prefix_rep" "1")
    (set_attr "mode" "TI")])
 
@@ -761,6 +767,7 @@
   "TARGET_SSE"
   "%vrcpps\t{%1, %0|%0, %1}"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "rcp")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "V4SF")])
 
@@ -787,6 +794,7 @@
   "TARGET_SSE"
   "rcpss\t{%1, %0|%0, %1}"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "rcp")
    (set_attr "mode" "SF")])
 
 (define_expand "sqrtv8sf2"
@@ -832,6 +840,7 @@
   "TARGET_SSE"
   "%vsqrtps\t{%1, %0|%0, %1}"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "sqrt")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "V4SF")])
 
@@ -876,6 +885,7 @@
   "SSE_VEC_FLOAT_MODE_P (<MODE>mode)"
   "sqrts<ssemodesuffixf2c>\t{%1, %0|%0, %1}"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "sqrt")
    (set_attr "mode" "<ssescalarmode>")])
 
 (define_expand "rsqrtv8sf2"
@@ -1039,7 +1049,7 @@
 	 (const_int 1)))]
   "SSE_VEC_FLOAT_MODE_P (<MODE>mode)"
   "<maxminfprefix>s<ssemodesuffixf2c>\t{%2, %0|%0, %2}"
-  [(set_attr "type" "sse")
+  [(set_attr "type" "sseadd")
    (set_attr "mode" "<ssescalarmode>")])
 
 ;; These versions of the min/max patterns implement exactly the operations
@@ -1175,6 +1185,7 @@
   "TARGET_SSE3"
   "addsubpd\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "mode" "V2DF")])
 
 (define_insn "avx_h<plusminus_insn>v4df3"
@@ -1298,6 +1309,7 @@
   "TARGET_SSE3"
   "h<plusminus_mnemonic>ps\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_rep" "1")
    (set_attr "mode" "V4SF")])
 
@@ -5066,6 +5078,7 @@
   "TARGET_SSE2 && ix86_binary_operator_ok (MULT, V8HImode, operands)"
   "pmaddwd\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "simul")
    (set_attr "prefix_data16" "1")
    (set_attr "mode" "TI")])
 
@@ -7025,6 +7038,7 @@
    movq\t{%H1, %0|%0, %H1}
    mov{q}\t{%H1, %0|%0, %H1}"
   [(set_attr "type" "ssemov,sseishft,ssemov,imov")
+   (set_attr "atom_unit" "*,sishuf,*,*")
    (set_attr "memory" "*,none,*,*")
    (set_attr "mode" "V2SF,TI,TI,DI")])
 
@@ -7057,6 +7071,7 @@
    psrldq\t{$8, %0|%0, 8}
    movq\t{%H1, %0|%0, %H1}"
   [(set_attr "type" "ssemov,sseishft,ssemov")
+   (set_attr "atom_unit" "*,sishuf,*")
    (set_attr "memory" "*,none,*")
    (set_attr "mode" "V2SF,TI,TI")])
 
@@ -7614,6 +7629,7 @@
   "TARGET_SSE2"
   "psadbw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "simul")
    (set_attr "prefix_data16" "1")
    (set_attr "mode" "TI")])
 
@@ -7635,7 +7651,7 @@
 	  UNSPEC_MOVMSK))]
   "SSE_VEC_FLOAT_MODE_P (<MODE>mode)"
   "%vmovmskp<ssemodesuffixf2c>\t{%1, %0|%0, %1}"
-  [(set_attr "type" "ssecvt")
+  [(set_attr "type" "ssemov")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "<MODE>")])
 
@@ -7645,7 +7661,7 @@
 		   UNSPEC_MOVMSK))]
   "TARGET_SSE2"
   "%vpmovmskb\t{%1, %0|%0, %1}"
-  [(set_attr "type" "ssecvt")
+  [(set_attr "type" "ssemov")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "SI")])
@@ -7668,7 +7684,7 @@
   "TARGET_SSE2 && !TARGET_64BIT"
   ;; @@@ check ordering of operands in intel/nonintel syntax
   "%vmaskmovdqu\t{%2, %1|%1, %2}"
-  [(set_attr "type" "ssecvt")
+  [(set_attr "type" "ssemov")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "TI")])
@@ -7682,7 +7698,7 @@
   "TARGET_SSE2 && TARGET_64BIT"
   ;; @@@ check ordering of operands in intel/nonintel syntax
   "%vmaskmovdqu\t{%2, %1|%1, %2}"
-  [(set_attr "type" "ssecvt")
+  [(set_attr "type" "ssemov")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "TI")])
@@ -7693,6 +7709,7 @@
   "TARGET_SSE"
   "%vldmxcsr\t%0"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "mxcsr")
    (set_attr "prefix" "maybe_vex")
    (set_attr "memory" "load")])
 
@@ -7702,6 +7719,7 @@
   "TARGET_SSE"
   "%vstmxcsr\t%0"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "mxcsr")
    (set_attr "prefix" "maybe_vex")
    (set_attr "memory" "store")])
 
@@ -7720,6 +7738,7 @@
   "TARGET_SSE || TARGET_3DNOW_A"
   "sfence"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "fence")
    (set_attr "memory" "unknown")])
 
 (define_insn "sse2_clflush"
@@ -7728,6 +7747,7 @@
   "TARGET_SSE2"
   "clflush\t%a0"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "fence")
    (set_attr "memory" "unknown")])
 
 (define_expand "sse2_mfence"
@@ -7745,6 +7765,7 @@
   "TARGET_64BIT || TARGET_SSE2"
   "mfence"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "fence")
    (set_attr "memory" "unknown")])
 
 (define_expand "sse2_lfence"
@@ -7762,6 +7783,7 @@
   "TARGET_SSE2"
   "lfence"
   [(set_attr "type" "sse")
+   (set_attr "atom_sse_attr" "lfence")
    (set_attr "memory" "unknown")])
 
 (define_insn "sse3_mwait"
@@ -7885,6 +7907,7 @@
   "TARGET_SSSE3"
   "phaddw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "TI")])
@@ -7913,6 +7936,7 @@
   "TARGET_SSSE3"
   "phaddw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "DI")])
 
@@ -7967,6 +7991,7 @@
   "TARGET_SSSE3"
   "phaddd\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "TI")])
@@ -7987,6 +8012,7 @@
   "TARGET_SSSE3"
   "phaddd\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "DI")])
 
@@ -8073,6 +8099,7 @@
   "TARGET_SSSE3"
   "phaddsw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "TI")])
@@ -8101,6 +8128,7 @@
   "TARGET_SSSE3"
   "phaddsw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "DI")])
 
@@ -8187,6 +8215,7 @@
   "TARGET_SSSE3"
   "phsubw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "TI")])
@@ -8215,6 +8244,7 @@
   "TARGET_SSSE3"
   "phsubw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "DI")])
 
@@ -8269,6 +8299,7 @@
   "TARGET_SSSE3"
   "phsubd\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "TI")])
@@ -8289,6 +8320,7 @@
   "TARGET_SSSE3"
   "phsubd\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "DI")])
 
@@ -8375,6 +8407,7 @@
   "TARGET_SSSE3"
   "phsubsw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "TI")])
@@ -8403,6 +8436,7 @@
   "TARGET_SSSE3"
   "phsubsw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "complex")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "DI")])
 
@@ -8509,6 +8543,7 @@
   "TARGET_SSSE3"
   "pmaddubsw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "simul")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "TI")])
@@ -8547,6 +8582,7 @@
   "TARGET_SSSE3"
   "pmaddubsw\t{%2, %0|%0, %2}"
   [(set_attr "type" "sseiadd")
+   (set_attr "atom_unit" "simul")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "DI")])
 
@@ -8754,6 +8790,7 @@
   return "palignr\t{%3, %2, %0|%0, %2, %3}";
 }
   [(set_attr "type" "sseishft")
+   (set_attr "atom_unit" "sishuf")
    (set_attr "prefix_data16" "1")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "TI")])
@@ -8770,6 +8807,7 @@
   return "palignr\t{%3, %2, %0|%0, %2, %3}";
 }
   [(set_attr "type" "sseishft")
+   (set_attr "atom_unit" "sishuf")
    (set_attr "prefix_extra" "1")
    (set_attr "mode" "DI")])
 
@@ -8956,7 +8994,7 @@
 		     UNSPEC_MOVNTDQA))]
   "TARGET_SSE4_1"
   "%vmovntdqa\t{%1, %0|%0, %1}"
-  [(set_attr "type" "ssecvt")
+  [(set_attr "type" "ssemov")
    (set_attr "prefix_extra" "1")
    (set_attr "prefix" "maybe_vex")
    (set_attr "mode" "TI")])