* removed extra deallocation and allocation of function result

    registers, allows for most optimal code possible to be generated
    for trivial functions (but in some cases worsens generated code
    because the register allocator is not 100% optimal)

git-svn-id: trunk@2208 -

compiler/ncgutil.pas

@@ -1183,9 +1183,6 @@ implementation
                       if getsupreg(resloc.register64.reglo)<first_int_imreg then
                         begin
                           cg.getcpuregister(list,resloc.register64.reglo);
-                          cg.ungetcpuregister(list,resloc.register64.reglo);
-                          { for the optimizer }
-                          cg.a_reg_alloc(list,resloc.register64.reglo);
                         end;
                       case restmploc.loc of
                         LOC_REFERENCE :
@@ -1203,9 +1200,6 @@ implementation
                       if getsupreg(resloc.register64.reghi)<first_int_imreg then
                         begin
                           cg.getcpuregister(list,resloc.register64.reghi);
-                          cg.ungetcpuregister(list,resloc.register64.reghi);
-                          { for the optimizer }
-                          cg.a_reg_alloc(list,resloc.register64.reghi);
                         end;
                       case restmploc.loc of
                         LOC_REFERENCE :
@@ -1228,9 +1222,6 @@ implementation
                       if getsupreg(funcretloc.register)<first_int_imreg then
                         begin
                           cg.getcpuregister(list,funcretloc.register);
-                          cg.ungetcpuregister(list,hreg);
-                          { for the optimizer }
-                          cg.a_reg_alloc(list,funcretloc.register);
                         end;
                       { it could be that a structure is passed in memory but the function is expected to
                         return a pointer to this memory }
@@ -1245,7 +1236,6 @@ implementation
                   if getsupreg(funcretloc.register)<first_fpu_imreg then
                     begin
                       cg.getcpuregister(list,funcretloc.register);
-                      cg.ungetcpuregister(list,funcretloc.register);
                     end;
                   { we can't do direct moves between fpu and mm registers }
                   if restmploc.loc in [LOC_MMREGISTER,LOC_CMMREGISTER] then
@@ -1257,7 +1247,6 @@ implementation
                   if getsupreg(funcretloc.register)<first_mm_imreg then
                     begin
                       cg.getcpuregister(list,funcretloc.register);
-                      cg.ungetcpuregister(list,funcretloc.register);
                     end;
                   cg.a_loadmm_loc_reg(list,restmploc.size,restmploc,funcretloc.register,mms_movescalar);
                 end;