+ cs_userbp optimizer switch, so on x86-64 the compiler can make use of rbp if it is not needed as frame pointer

git-svn-id: trunk@26633 -

compiler/globtype.pas

@@ -263,7 +263,7 @@ interface
          cs_opt_level1,cs_opt_level2,cs_opt_level3,
          cs_opt_regvar,cs_opt_uncertain,cs_opt_size,cs_opt_stackframe,
          cs_opt_peephole,cs_opt_asmcse,cs_opt_loopunroll,cs_opt_tailrecursion,cs_opt_nodecse,
-         cs_opt_nodedfa,cs_opt_loopstrength,cs_opt_scheduler,cs_opt_autoinline,cs_useebp,
+         cs_opt_nodedfa,cs_opt_loopstrength,cs_opt_scheduler,cs_opt_autoinline,cs_useebp,cs_userbp,
          cs_opt_reorder_fields,cs_opt_fastmath,
          { Allow removing expressions whose result is not used, even when this
            can change program behaviour (range check errors disappear,
@@ -308,7 +308,7 @@ interface
          'LEVEL1','LEVEL2','LEVEL3',
          'REGVAR','UNCERTAIN','SIZE','STACKFRAME',
          'PEEPHOLE','ASMCSE','LOOPUNROLL','TAILREC','CSE',
-         'DFA','STRENGTH','SCHEDULE','AUTOINLINE','USEEBP',
+         'DFA','STRENGTH','SCHEDULE','AUTOINLINE','USEEBP','USERBP',
          'ORDERFIELDS','FASTMATH','DEADVALUES','REMOVEEMPTYPROCS',
          'CONSTPROP',
          'DEADSTORE'

compiler/x86_64/cgcpu.pas

@@ -62,7 +62,7 @@ unit cgcpu;
 
     procedure Tcgx86_64.init_register_allocators;
       const
-        win64_saved_std_regs : array[0..6] of tsuperregister = (RS_RBX,RS_RDI,RS_RSI,RS_R12,RS_R13,RS_R14,RS_R15);
+        win64_saved_std_regs : array[0..7] of tsuperregister = (RS_RBX,RS_RDI,RS_RSI,RS_R12,RS_R13,RS_R14,RS_R15,RS_RBP);
         others_saved_std_regs : array[0..4] of tsuperregister = (RS_RBX,RS_R12,RS_R13,RS_R14,RS_R15);
         saved_regs_length : array[boolean] of longint = (5,7);
 
@@ -96,8 +96,16 @@ unit cgcpu;
               end;
           end;
         if target_info.system=system_x86_64_win64 then
-          rg[R_INTREGISTER]:=trgcpu.create(R_INTREGISTER,R_SUBWHOLE,[RS_RAX,RS_RDX,RS_RCX,RS_R8,RS_R9,RS_R10,
-            RS_R11,RS_RBX,RS_RSI,RS_RDI,RS_R12,RS_R13,RS_R14,RS_R15],first_int_imreg,[])
+          begin
+            if (cs_userbp in current_settings.optimizerswitches) and assigned(current_procinfo) and (current_procinfo.framepointer=NR_STACK_POINTER_REG) then
+              begin
+                rg[R_INTREGISTER]:=trgcpu.create(R_INTREGISTER,R_SUBWHOLE,[RS_RAX,RS_RDX,RS_RCX,RS_R8,RS_R9,RS_R10,
+                  RS_R11,RS_RBX,RS_RSI,RS_RDI,RS_R12,RS_R13,RS_R14,RS_R15,RS_RBP],first_int_imreg,[]);
+              end
+            else
+              rg[R_INTREGISTER]:=trgcpu.create(R_INTREGISTER,R_SUBWHOLE,[RS_RAX,RS_RDX,RS_RCX,RS_R8,RS_R9,RS_R10,
+                RS_R11,RS_RBX,RS_RSI,RS_RDI,RS_R12,RS_R13,RS_R14,RS_R15],first_int_imreg,[])
+          end
         else
           rg[R_INTREGISTER]:=trgcpu.create(R_INTREGISTER,R_SUBWHOLE,[RS_RAX,RS_RDX,RS_RCX,RS_RSI,RS_RDI,RS_R8,
             RS_R9,RS_R10,RS_R11,RS_RBX,RS_R12,RS_R13,RS_R14,RS_R15],first_int_imreg,[]);

compiler/x86_64/cpuinfo.pas

@@ -106,14 +106,14 @@ Const
                                  genericlevel3optimizerswitches-
                                  { no need to write info about those }
                                  [cs_opt_level1,cs_opt_level2,cs_opt_level3]+
-                                 [cs_opt_regvar,cs_opt_loopunroll,cs_opt_stackframe,
+                                 [cs_opt_regvar,cs_opt_loopunroll,cs_opt_stackframe,cs_userbp,
 				  cs_opt_tailrecursion,cs_opt_nodecse,cs_opt_reorder_fields,cs_opt_fastmath];
 
    level1optimizerswitches = genericlevel1optimizerswitches;
    level2optimizerswitches = genericlevel2optimizerswitches + level1optimizerswitches + 
      [cs_opt_regvar,cs_opt_stackframe,cs_opt_tailrecursion,cs_opt_nodecse];
    level3optimizerswitches = genericlevel3optimizerswitches + level2optimizerswitches + [{,cs_opt_loopunroll}];
-   level4optimizerswitches = genericlevel4optimizerswitches + level3optimizerswitches + [];
+   level4optimizerswitches = genericlevel4optimizerswitches + level3optimizerswitches + [cs_userbp];
 
 type
    tcpuflags =