[RFA] PowerPC e300c2 and e300c3 cores support

[Edmar Wienskoski <edmar at freescale dot com>]

Freescale would like to contribute this patch to gcc. It creates a scheduler
and proper default flags for the e300c2 and e300c3 cores.

Could some PowerPC maintainer review and commit the patch ?

Regards,
Edmar

2008-02-21  Edmar Wienskoski  <edmar@freescale.com>

	* config.gcc (powerpc*-*-*): Add new cores e300c2 and e300c3.
	* config/rs6000/e300c2c3.md: New file.
	* config/rs6000/rs6000.c (processor_costs): Add new costs for
	e300c2 and e300c3.
	(rs6000_override_options): Add e300c2 and e300c3 cases to
	processor_target_table. Do not allow usage of Altivec or Spe
	with e300 cores. Initialize rs6000_cost for e300c2 and e300c3.
	(rs6000_issue_rate): Set issue rate for e300c2 and e300c3.
	* config/rs6000/rs6000.h (processor_type): Add
	PROCESSOR_PPCE300C2 and PROCESSOR_PPCE300C3.
	(ASM_CPU_SPEC): Add e300c2 and e300c3.
	* config/rs6000/rs6000.md (define_attr "cpu"): Add ppce300c2
	and ppce300c3. Include e300c2c3.md.
	* doc/invoke.texi: Add 8540 to list of cpus in rs6000 cpu
	section.


Index: gcc/gcc/config.gcc
===================================================================
--- gcc/gcc/config.gcc	(revision 132526)
+++ gcc/gcc/config.gcc	(working copy)
@@ -3130,7 +3130,7 @@
 			| rios | rios1 | rios2 | rsc | rsc1 | rs64a \
 			| 401 | 403 | 405 | 405fp | 440 | 440fp | 505 \
 			| 601 | 602 | 603 | 603e | ec603e | 604 \
-			| 604e | 620 | 630 | 740 | 750 | 7400 | 7450 \
+			| 604e | 620 | 630 | 740 | 750 | 7400 | 7450 | e300c[23] \
 			| 854[08] | 801 | 821 | 823 | 860 | 970 | G3 | G4 | G5 | cell)
 				# OK
 				;;
Index: gcc/gcc/config/rs6000/e300c2c3.md
===================================================================
--- gcc/gcc/config/rs6000/e300c2c3.md	(revision 0)
+++ gcc/gcc/config/rs6000/e300c2c3.md	(revision 0)
@@ -0,0 +1,189 @@
+;; Pipeline description for Motorola PowerPC e300c3 core.
+;;   Copyright (C) 2008 Free Software Foundation, Inc.
+;;   Contributed by Peter Steinmetz (steinmtz@us.ibm.com)
+;;
+;; 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/>.
+
+(define_automaton "ppce300c3_most,ppce300c3_long,ppce300c3_retire")
+(define_cpu_unit "ppce300c3_decode_0,ppce300c3_decode_1" "ppce300c3_most")
+
+;; We don't simulate general issue queue (GIC).  If we have SU insn
+;; and then SU1 insn, they can not be issued on the same cycle
+;; (although SU1 insn and then SU insn can be issued) because the SU
+;; insn will go to SU1 from GIC0 entry.  Fortunately, the first cycle
+;; multipass insn scheduling will find the situation and issue the SU1
+;; insn and then the SU insn.
+(define_cpu_unit "ppce300c3_issue_0,ppce300c3_issue_1"   "ppce300c3_most")
+
+;; We could describe completion buffers slots in combination with the
+;; retirement units and the order of completion but the result
+;; automaton would behave in the same way because we can not describe
+;; real latency time with taking in order completion into account.
+;; Actually we could define the real latency time by querying reserved
+;; automaton units but the current scheduler uses latency time before
+;; issuing insns and making any reservations.
+;;
+;; So our description is aimed to achieve a insn schedule in which the
+;; insns would not wait in the completion buffer.
+(define_cpu_unit "ppce300c3_retire_0,ppce300c3_retire_1" "ppce300c3_retire")
+
+;; Branch unit:
+(define_cpu_unit "ppce300c3_bu" "ppce300c3_most")
+
+;; IU:
+(define_cpu_unit "ppce300c3_iu0_stage0,ppce300c3_iu1_stage0" "ppce300c3_most")
+
+;; IU: This used to describe non-pipelined division.
+(define_cpu_unit "ppce300c3_mu_div" "ppce300c3_long")
+
+;; SRU:
+(define_cpu_unit "ppce300c3_sru_stage0" "ppce300c3_most")
+
+;; Here we simplified LSU unit description not describing the stages.
+(define_cpu_unit "ppce300c3_lsu" "ppce300c3_most")
+
+;; FPU:
+(define_cpu_unit "ppce300c3_fpu" "ppce300c3_most")
+
+;; The following units are used to make automata deterministic
+(define_cpu_unit "present_ppce300c3_decode_0" "ppce300c3_most")
+(define_cpu_unit "present_ppce300c3_issue_0" "ppce300c3_most")
+(define_cpu_unit "present_ppce300c3_retire_0" "ppce300c3_retire")
+(define_cpu_unit "present_ppce300c3_iu0_stage0" "ppce300c3_most")
+
+;; The following sets to make automata deterministic when option ndfa is used.
+(presence_set "present_ppce300c3_decode_0" "ppce300c3_decode_0")
+(presence_set "present_ppce300c3_issue_0" "ppce300c3_issue_0")
+(presence_set "present_ppce300c3_retire_0" "ppce300c3_retire_0")
+(presence_set "present_ppce300c3_iu0_stage0" "ppce300c3_iu0_stage0")
+
+;; Some useful abbreviations.
+(define_reservation "ppce300c3_decode"
+    "ppce300c3_decode_0|ppce300c3_decode_1+present_ppce300c3_decode_0")
+(define_reservation "ppce300c3_issue"
+    "ppce300c3_issue_0|ppce300c3_issue_1+present_ppce300c3_issue_0")
+(define_reservation "ppce300c3_retire"
+   "ppce300c3_retire_0|ppce300c3_retire_1+present_ppce300c3_retire_0")
+(define_reservation "ppce300c3_iu_stage0"
+   "ppce300c3_iu0_stage0|ppce300c3_iu1_stage0+present_ppce300c3_iu0_stage0")
+
+;; Compares can be executed either one of the IU or SRU
+(define_insn_reservation "ppce300c3_cmp" 1
+  (and (eq_attr "type" "cmp,compare,delayed_compare,fast_compare")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+(ppce300c3_iu_stage0|ppce300c3_sru_stage0) \
+        +ppce300c3_retire")
+
+;; Other one cycle IU insns
+(define_insn_reservation "ppce300c3_iu" 1
+  (and (eq_attr "type" "integer,insert_word")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_iu_stage0+ppce300c3_retire")
+
+;; Branch.  Actually this latency time is not used by the scheduler.
+(define_insn_reservation "ppce300c3_branch" 1
+  (and (eq_attr "type" "jmpreg,branch")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_bu,ppce300c3_retire")
+
+;; Multiply is non-pipelined but can be executed in any IU
+(define_insn_reservation "ppce300c3_multiply" 2
+  (and (eq_attr "type" "imul,imul2,imul3,imul_compare")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_iu_stage0, \
+   ppce300c3_iu_stage0+ppce300c3_retire")
+
+;; Divide.  We use the average latency time here.  We omit reserving a
+;; retire unit because of the result automata will be huge.
+(define_insn_reservation "ppce300c3_divide" 20
+  (and (eq_attr "type" "idiv")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_iu_stage0+ppce300c3_mu_div,\
+   ppce300c3_mu_div*19")
+
+;; CR logical
+(define_insn_reservation "ppce300c3_cr_logical" 1
+  (and (eq_attr "type" "cr_logical,delayed_cr")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_sru_stage0+ppce300c3_retire")
+
+;; Mfcr
+(define_insn_reservation "ppce300c3_mfcr" 1
+  (and (eq_attr "type" "mfcr")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_sru_stage0+ppce300c3_retire")
+
+;; Mtcrf
+(define_insn_reservation "ppce300c3_mtcrf" 1
+  (and (eq_attr "type" "mtcr")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_sru_stage0+ppce300c3_retire")
+
+;; Mtjmpr
+(define_insn_reservation "ppce300c3_mtjmpr" 1
+  (and (eq_attr "type" "mtjmpr,mfjmpr")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_sru_stage0+ppce300c3_retire")
+
+;; Float point instructions
+(define_insn_reservation "ppce300c3_fpcompare" 3
+  (and (eq_attr "type" "fpcompare")
+       (eq_attr "cpu" "ppce300c3"))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,nothing,ppce300c3_retire")
+
+(define_insn_reservation "ppce300c3_fp" 3
+  (and (eq_attr "type" "fp")
+       (eq_attr "cpu" "ppce300c3"))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,nothing,ppce300c3_retire")
+
+(define_insn_reservation "ppce300c3_dmul" 4
+  (and (eq_attr "type" "dmul")
+       (eq_attr "cpu" "ppce300c3"))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,ppce300c3_fpu,nothing,ppce300c3_retire")
+
+; Divides are not pipelined
+(define_insn_reservation "ppce300c3_sdiv" 18
+  (and (eq_attr "type" "sdiv")
+       (eq_attr "cpu" "ppce300c3"))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,ppce300c3_fpu*17")
+
+(define_insn_reservation "ppce300c3_ddiv" 33
+  (and (eq_attr "type" "ddiv")
+       (eq_attr "cpu" "ppce300c3"))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_fpu,ppce300c3_fpu*32")
+
+;; Loads
+(define_insn_reservation "ppce300c3_load" 2
+  (and (eq_attr "type" "load,load_ext,load_ext_u,load_ext_ux,load_ux,load_u")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_lsu,ppce300c3_retire")
+
+(define_insn_reservation "ppce300c3_fpload" 2
+  (and (eq_attr "type" "fpload,fpload_ux,fpload_u")
+       (eq_attr "cpu" "ppce300c3"))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_lsu,ppce300c3_retire")
+
+;; Stores.
+(define_insn_reservation "ppce300c3_store" 2
+  (and (eq_attr "type" "store,store_ux,store_u")
+       (ior (eq_attr "cpu" "ppce300c2") (eq_attr "cpu" "ppce300c3")))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_lsu,ppce300c3_retire")
+
+(define_insn_reservation "ppce300c3_fpstore" 2
+  (and (eq_attr "type" "fpstore,fpstore_ux,fpstore_u")
+       (eq_attr "cpu" "ppce300c3"))
+  "ppce300c3_decode,ppce300c3_issue+ppce300c3_lsu,ppce300c3_retire")
Index: gcc/gcc/config/rs6000/rs6000.c
===================================================================
--- gcc/gcc/config/rs6000/rs6000.c	(revision 132526)
+++ gcc/gcc/config/rs6000/rs6000.c	(working copy)
@@ -667,6 +667,21 @@
   1,			/* prefetch streams /*/
 };
 
+/* Instruction costs on E300C2 and E300C3 cores.  */
+static const
+struct processor_costs ppce300c2c3_cost = {
+  COSTS_N_INSNS (4),    /* mulsi */
+  COSTS_N_INSNS (4),    /* mulsi_const */
+  COSTS_N_INSNS (4),    /* mulsi_const9 */
+  COSTS_N_INSNS (4),    /* muldi */
+  COSTS_N_INSNS (19),   /* divsi */
+  COSTS_N_INSNS (19),   /* divdi */
+  COSTS_N_INSNS (3),    /* fp */
+  COSTS_N_INSNS (4),    /* dmul */
+  COSTS_N_INSNS (18),   /* sdiv */
+  COSTS_N_INSNS (33),   /* ddiv */
+};
+
 /* Instruction costs on POWER4 and POWER5 processors.  */
 static const
 struct processor_costs power4_cost = {
@@ -1418,6 +1433,8 @@
 	 {"8540", PROCESSOR_PPC8540, POWERPC_BASE_MASK | MASK_STRICT_ALIGN},
 	 /* 8548 has a dummy entry for now.  */
 	 {"8548", PROCESSOR_PPC8540, POWERPC_BASE_MASK | MASK_STRICT_ALIGN},
+	 {"e300c2", PROCESSOR_PPCE300C2, POWERPC_BASE_MASK | MASK_SOFT_FLOAT},
+	 {"e300c3", PROCESSOR_PPCE300C3, POWERPC_BASE_MASK},
 	 {"860", PROCESSOR_MPCCORE, POWERPC_BASE_MASK | MASK_SOFT_FLOAT},
 	 {"970", PROCESSOR_POWER4,
 	  POWERPC_7400_MASK | MASK_PPC_GPOPT | MASK_MFCRF | MASK_POWERPC64},
@@ -1524,6 +1541,14 @@
   if (TARGET_E500)
     rs6000_isel = 1;
 
+  if (rs6000_cpu == PROCESSOR_PPCE300C2 || rs6000_cpu == PROCESSOR_PPCE300C3)
+    {
+      if (TARGET_ALTIVEC)
+	error ("AltiVec not supported in this target");
+      if (TARGET_SPE)
+	error ("Spe not supported in this target");
+    }
+
   /* If we are optimizing big endian systems for space, use the load/store
      multiple and string instructions.  */
   if (BYTES_BIG_ENDIAN && optimize_size)
@@ -1836,6 +1861,11 @@
 	rs6000_cost = &ppc8540_cost;
 	break;
 
+      case PROCESSOR_PPCE300C2:
+      case PROCESSOR_PPCE300C3:
+	rs6000_cost = &ppce300c2c3_cost;
+	break;
+
       case PROCESSOR_POWER4:
       case PROCESSOR_POWER5:
 	rs6000_cost = &power4_cost;
@@ -18502,6 +18532,8 @@
   case CPU_PPC7400:
   case CPU_PPC8540:
   case CPU_CELL:
+  case CPU_PPCE300C2:
+  case CPU_PPCE300C3:
     return 2;
   case CPU_RIOS2:
   case CPU_PPC604:
Index: gcc/gcc/config/rs6000/rs6000.h
===================================================================
--- gcc/gcc/config/rs6000/rs6000.h	(revision 132526)
+++ gcc/gcc/config/rs6000/rs6000.h	(working copy)
@@ -117,6 +117,8 @@
 %{mcpu=G5: -mpower4 -maltivec} \
 %{mcpu=8540: -me500} \
 %{mcpu=8548: -me500} \
+%{mcpu=e300c2: -me300} \
+%{mcpu=e300c3: -me300} \
 %{maltivec: -maltivec} \
 -many"
 
@@ -262,6 +264,8 @@
    PROCESSOR_PPC7400,
    PROCESSOR_PPC7450,
    PROCESSOR_PPC8540,
+   PROCESSOR_PPCE300C2,
+   PROCESSOR_PPCE300C3,
    PROCESSOR_POWER4,
    PROCESSOR_POWER5,
    PROCESSOR_POWER6,
Index: gcc/gcc/config/rs6000/rs6000.md
===================================================================
--- gcc/gcc/config/rs6000/rs6000.md	(revision 132526)
+++ gcc/gcc/config/rs6000/rs6000.md	(working copy)
@@ -133,7 +133,7 @@
 ;; Processor type -- this attribute must exactly match the processor_type
 ;; enumeration in rs6000.h.
 
-(define_attr "cpu" "rios1,rios2,rs64a,mpccore,ppc403,ppc405,ppc440,ppc601,ppc603,ppc604,ppc604e,ppc620,ppc630,ppc750,ppc7400,ppc7450,ppc8540,power4,power5,power6,cell"
+(define_attr "cpu" "rios1,rios2,rs64a,mpccore,ppc403,ppc405,ppc440,ppc601,ppc603,ppc604,ppc604e,ppc620,ppc630,ppc750,ppc7400,ppc7450,ppc8540,ppce300c2,ppce300c3,power4,power5,power6,cell"
   (const (symbol_ref "rs6000_cpu_attr")))
 
 
@@ -166,6 +166,7 @@
 (include "7xx.md")
 (include "7450.md")
 (include "8540.md")
+(include "e300c2c3.md")
 (include "power4.md")
 (include "power5.md")
 (include "power6.md")