[PATCH] Improve Niagara-3 tuning.

[David Miller]

This adjusts the costs of various operations on Niagara-3, in
particular FP operation latency increases to 9 cycles compared to
Niagara-2's 6 cycles.  Integer multiplies increase to 9 cycles as
well, which likely indicates that they go through the FPU unit too.

I also fixed up the integer divide costs for Niagara-2, as it's
pretty clear I have no clue how to compute averages :-)

Commited to trunk.

gcc/

        * gcc/config/sparc/sparc.c (niagara2_costs): Adjust integer
        divide costs.
        (niagara3_costs): New.
        (sparc_option_override): Use it.
        * gcc/config/sparc/niagara2.md: Adjust with more accurate
        Niagara-3 reservations.

diff --git a/gcc/config/sparc/niagara2.md b/gcc/config/sparc/niagara2.md
index 999e13b..d53be84 100644
--- a/gcc/config/sparc/niagara2.md
+++ b/gcc/config/sparc/niagara2.md
@@ -45,12 +45,22 @@
  "niag2_pipe")
 
 (define_insn_reservation "niag2_imul" 5
- (and (eq_attr "cpu" "niagara2,niagara3")
+ (and (eq_attr "cpu" "niagara2")
    (eq_attr "type" "imul"))
  "niag2_pipe*5")
 
-(define_insn_reservation "niag2_idiv" 31
- (and (eq_attr "cpu" "niagara2,niagara3")
+(define_insn_reservation "niag3_imul" 9
+ (and (eq_attr "cpu" "niagara3")
+   (eq_attr "type" "imul"))
+ "niag2_pipe*9")
+
+(define_insn_reservation "niag2_idiv" 26
+ (and (eq_attr "cpu" "niagara2")
+   (eq_attr "type" "idiv"))
+ "niag2_pipe*26")
+
+(define_insn_reservation "niag3_idiv" 31
+ (and (eq_attr "cpu" "niagara3")
    (eq_attr "type" "idiv"))
  "niag2_pipe*31")
 
@@ -69,22 +79,42 @@
     (eq_attr "type" "store,fpstore"))
   "niag2_pipe")
 
-(define_insn_reservation "niag2_fp" 3
-  (and (eq_attr "cpu" "niagara2,niagara3")
+(define_insn_reservation "niag2_fp" 6
+  (and (eq_attr "cpu" "niagara2")
     (eq_attr "type" "fpmove,fpcmove,fpcrmove,fpcmp,fpmul"))
-  "niag2_pipe*3")
+  "niag2_pipe*6")
+
+(define_insn_reservation "niag3_fp" 9
+  (and (eq_attr "cpu" "niagara3")
+    (eq_attr "type" "fpmove,fpcmove,fpcrmove,fpcmp,fpmul"))
+  "niag2_pipe*9")
 
 (define_insn_reservation "niag2_fdivs" 19
-  (and (eq_attr "cpu" "niagara2,niagara3")
+  (and (eq_attr "cpu" "niagara2")
     (eq_attr "type" "fpdivs"))
   "niag2_pipe*19")
 
+(define_insn_reservation "niag3_fdivs" 23
+  (and (eq_attr "cpu" "niagara3")
+    (eq_attr "type" "fpdivs"))
+  "niag2_pipe*23")
+
 (define_insn_reservation "niag2_fdivd" 33
-  (and (eq_attr "cpu" "niagara2,niagara3")
+  (and (eq_attr "cpu" "niagara2")
     (eq_attr "type" "fpdivd"))
   "niag2_pipe*33")
 
+(define_insn_reservation "niag3_fdivd" 37
+  (and (eq_attr "cpu" "niagara3")
+    (eq_attr "type" "fpdivd"))
+  "niag2_pipe*37")
+
 (define_insn_reservation "niag2_vis" 6
-  (and (eq_attr "cpu" "niagara2,niagara3")
+  (and (eq_attr "cpu" "niagara2")
     (eq_attr "type" "fga,fgm_pack,fgm_mul,fgm_cmp,fgm_pdist,edge"))
   "niag2_pipe*6")
+
+(define_insn_reservation "niag3_vis" 9
+  (and (eq_attr "cpu" "niagara3")
+    (eq_attr "type" "fga,fgm_pack,fgm_mul,fgm_cmp,fgm_pdist,edge"))
+  "niag2_pipe*9")
diff --git a/gcc/config/sparc/sparc.c b/gcc/config/sparc/sparc.c
index 7e124a0..6ccc97b 100644
--- a/gcc/config/sparc/sparc.c
+++ b/gcc/config/sparc/sparc.c
@@ -269,8 +269,32 @@ struct processor_costs niagara2_costs = {
   COSTS_N_INSNS (5), /* imul */
   COSTS_N_INSNS (5), /* imulX */
   0, /* imul bit factor */
-  COSTS_N_INSNS (31), /* idiv, average of 12 - 41 cycle range */
-  COSTS_N_INSNS (31), /* idivX, average of 12 - 41 cycle range */
+  COSTS_N_INSNS (26), /* idiv, average of 12 - 41 cycle range */
+  COSTS_N_INSNS (26), /* idivX, average of 12 - 41 cycle range */
+  COSTS_N_INSNS (1), /* movcc/movr */
+  0, /* shift penalty */
+};
+
+static const
+struct processor_costs niagara3_costs = {
+  COSTS_N_INSNS (3), /* int load */
+  COSTS_N_INSNS (3), /* int signed load */
+  COSTS_N_INSNS (3), /* int zeroed load */
+  COSTS_N_INSNS (3), /* float load */
+  COSTS_N_INSNS (9), /* fmov, fneg, fabs */
+  COSTS_N_INSNS (9), /* fadd, fsub */
+  COSTS_N_INSNS (9), /* fcmp */
+  COSTS_N_INSNS (9), /* fmov, fmovr */
+  COSTS_N_INSNS (9), /* fmul */
+  COSTS_N_INSNS (23), /* fdivs */
+  COSTS_N_INSNS (37), /* fdivd */
+  COSTS_N_INSNS (23), /* fsqrts */
+  COSTS_N_INSNS (37), /* fsqrtd */
+  COSTS_N_INSNS (9), /* imul */
+  COSTS_N_INSNS (9), /* imulX */
+  0, /* imul bit factor */
+  COSTS_N_INSNS (31), /* idiv, average of 17 - 45 cycle range */
+  COSTS_N_INSNS (30), /* idivX, average of 16 - 44 cycle range */
   COSTS_N_INSNS (1), /* movcc/movr */
   0, /* shift penalty */
 };
@@ -917,9 +941,11 @@ sparc_option_override (void)
       sparc_costs = &niagara_costs;
       break;
     case PROCESSOR_NIAGARA2:
+      sparc_costs = &niagara2_costs;
+      break;
     case PROCESSOR_NIAGARA3:
     case PROCESSOR_NIAGARA4:
-      sparc_costs = &niagara2_costs;
+      sparc_costs = &niagara3_costs;
       break;
     case PROCESSOR_NATIVE:
       gcc_unreachable ();