+ MIPS: implemented parameter location reusing, eliminating second copy of (potentially large) records passed by value. When parameter is passed both in registers and stack, let it have a single LOC_REFERENCE location on callee side, and store relevant registers on stack (into 16-byte area reserved by ABI) early in prologue.

git-svn-id: trunk@24970 -

compiler/mips/cgcpu.pas

@@ -1157,6 +1157,20 @@ end;
 
 { *********** entry/exit code and address loading ************ }
 
+procedure FixupOffsets(p:TObject;arg:pointer);
+var
+  sym: tabstractnormalvarsym absolute p;
+begin
+  if (tsym(p).typ=paravarsym) and
+    (sym.localloc.loc=LOC_REFERENCE) and
+    (sym.localloc.reference.base=NR_FRAME_POINTER_REG) then
+    begin
+      sym.localloc.reference.base:=NR_STACK_POINTER_REG;
+      Inc(sym.localloc.reference.offset,PLongint(arg)^);
+    end;
+end;
+
+
 procedure TCGMIPS.g_proc_entry(list: tasmlist; localsize: longint; nostackframe: boolean);
 var
   lastintoffset,lastfpuoffset,
@@ -1283,28 +1297,20 @@ begin
       list.concat(Taicpu.op_none(A_P_SET_NOMACRO));
     end;
 
-  with TMIPSProcInfo(current_procinfo) do
-    begin
-      href.offset:=0;
-      //if current_procinfo.framepointer<>NR_STACK_POINTER_REG then
-        href.base:=NR_FRAME_POINTER_REG;
+  href.base:=NR_STACK_POINTER_REG;
+
+  for i:=0 to MIPS_MAX_REGISTERS_USED_IN_CALL-1 do
+    if TMIPSProcInfo(current_procinfo).register_used[i] then
+      begin
+        reg:=parasupregs[i];
+        href.offset:=i*sizeof(aint)+LocalSize;
+        list.concat(taicpu.op_reg_ref(A_SW, newreg(R_INTREGISTER,reg,R_SUBWHOLE), href));
+      end;
 
-      for i:=0 to MIPS_MAX_REGISTERS_USED_IN_CALL-1 do
-        if (register_used[i]) then
-          begin
-            reg:=parasupregs[i];
-            if register_offset[i]=-1 then
-              comment(V_warning,'Register parameter has offset -1 in TCGMIPS.g_proc_entry');
-
-            //if current_procinfo.framepointer=NR_STACK_POINTER_REG then
-            //  href.offset:=register_offset[i]+Localsize
-            //else
-            href.offset:=register_offset[i];
-            list.concat(taicpu.op_reg_ref(A_SW, newreg(R_INTREGISTER,reg,R_SUBWHOLE), href));
-        end;
-    end;
   list.concatList(helplist);
   helplist.Free;
+  if current_procinfo.has_nestedprocs then
+    current_procinfo.procdef.parast.SymList.ForEachCall(@FixupOffsets,@LocalSize);
 end;
 
 

compiler/mips/cpupara.pas

@@ -31,8 +31,6 @@ interface
       symconst,symbase,symsym,symtype,symdef,paramgr,parabase,cgbase,cgutils;
 
     const
-      MIPS_MAX_OFFSET = 20;
-
       { The value below is OK for O32 and N32 calling conventions }
       MIPS_MAX_REGISTERS_USED_IN_CALL = 6;
 
@@ -63,9 +61,6 @@ interface
     type
       tparasupregs = array[0..MIPS_MAX_REGISTERS_USED_IN_CALL-1] of tsuperregister;
       tparasupregsused = array[0..MIPS_MAX_REGISTERS_USED_IN_CALL-1] of boolean;
-      tparasupregsize = array[0..MIPS_MAX_REGISTERS_USED_IN_CALL-1] of tcgsize;
-      tparasuprename = array[0..MIPS_MAX_REGISTERS_USED_IN_CALL-1] of shortstring;
-      tparasupregsoffset = array[0..MIPS_MAX_REGISTERS_USED_IN_CALL-1] of longint;
 
     const
 
@@ -79,6 +74,7 @@ interface
         function  create_paraloc_info(p : TAbstractProcDef; side: tcallercallee):longint;override;
         function  create_varargs_paraloc_info(p : TAbstractProcDef; varargspara:tvarargsparalist):longint;override;
         function  get_funcretloc(p : tabstractprocdef; side: tcallercallee; forcetempdef: tdef): tcgpara;override;
+        function  param_use_paraloc(const cgpara: tcgpara): boolean; override;
       private
         intparareg,
         intparasize : longint;
@@ -119,6 +115,16 @@ implementation
       end;
 
 
+    function TMIPSParaManager.param_use_paraloc(const cgpara: tcgpara): boolean;
+      var
+        paraloc: pcgparalocation;
+      begin
+        paraloc:=cgpara.location;
+        if not assigned(paraloc) then
+          internalerror(200410102);
+        result:=(paraloc^.loc=LOC_REFERENCE) and (paraloc^.next=nil);
+      end;
+
 
     { true if a parameter is too large to copy and only the address is pushed }
     function TMIPSParaManager.push_addr_param(varspez:tvarspez;def : tdef;calloption : tproccalloption) : boolean;
@@ -231,7 +237,7 @@ implementation
     procedure TMIPSParaManager.create_paraloc_info_intern(p : tabstractprocdef; side: tcallercallee;paras:tparalist);
       var
         paraloc      : pcgparalocation;
-        i            : integer;
+        i,j          : integer;
         hp           : tparavarsym;
         paracgsize   : tcgsize;
         paralen      : longint;
@@ -242,6 +248,7 @@ implementation
         alignment    : longint;
         tmp          : longint;
         firstparaloc : boolean;
+        reg_and_stack: boolean;
       begin
         fpparareg := 0;
         for i:=0 to paras.count-1 do
@@ -340,6 +347,9 @@ implementation
               can_use_float := false;
 
             firstparaloc:=true;
+            { Is parameter split between stack and registers? }
+            reg_and_stack:=(side=calleeside) and
+              (paralen+intparasize>16) and (intparasize<16);
             while paralen>0 do
               begin
                 paraloc:=hp.paraloc[side].add_location;
@@ -368,11 +378,14 @@ implementation
                     paraloc^.register:=newreg(R_INTREGISTER,parasupregs[0],R_SUBWHOLE);
                     inc(intparasize,align(tcgsize2size[paraloc^.size],sizeof(aint)));
                   end
-                { In case of po_delphi_nested_cc, the parent frame pointer
-                  is always passed on the stack. }
+                { "In case of po_delphi_nested_cc, the parent frame pointer
+                  is always passed on the stack". On other targets it is
+                  used to provide caller-side stack cleanup and prevent stackframe
+                  optimization. For MIPS this does not matter. }
                 else if (intparareg<mips_nb_used_registers) and
+                   (not reg_and_stack) {and
                    (not(vo_is_parentfp in hp.varoptions) or
-                    not(po_delphi_nested_cc in p.procoptions)) then
+                    not(po_delphi_nested_cc in p.procoptions))} then
                   begin
                     if (can_use_float) then
                       begin
@@ -418,6 +431,13 @@ implementation
                   end
                 else
                   begin
+                    if reg_and_stack then
+                      begin
+                        for j:=intparareg to mips_nb_used_registers-1 do
+                          tmipsprocinfo(current_procinfo).register_used[j]:=true;
+                        { all registers used now }
+                        intparareg:=mips_nb_used_registers;
+                      end;
                     paraloc^.loc:=LOC_REFERENCE;
                     paraloc^.size:=int_cgsize(paralen);
                     paraloc^.def:=get_paraloc_def(locdef,paralen,firstparaloc);

compiler/mips/cpupi.pas

@@ -41,9 +41,6 @@ interface
       intregssave,
       floatregssave : byte;
       register_used : tparasupregsused;
-      register_size : tparasupregsize;
-      register_name : tparasuprename;
-      register_offset : tparasupregsoffset;
       computed_local_size : longint;
       save_gp_ref: treference;
       //intparareg,
@@ -66,20 +63,12 @@ implementation
       tgobj,paramgr,symconst;
 
     constructor TMIPSProcInfo.create(aparent: tprocinfo);
-      var
-        i : longint;
       begin
         inherited create(aparent);
         { if (cs_generate_stackframes in current_settings.localswitches) or
            not (cs_opt_stackframe in current_settings.optimizerswitches) then }
           include(flags,pi_needs_stackframe);
-        for i:=low(tparasupregs)  to high(tparasupregs) do
-          begin
-            register_used[i]:=false;
-            register_size[i]:=OS_NO;
-            register_name[i]:='invalid';
-            register_offset[i]:=-1;
-          end;
+
         floatregssave:=12; { f20-f31 }
         intregssave:=10;   { r16-r23,r30,r31 }
         computed_local_size:=-1;