![](/file/12652/www.mactech.com.tar/www.mactech.com/sites/all/themes/custom_front/images/you_are_here_red.gif)
![](/file/12652/www.mactech.com.tar/www.mactech.com/sites/default/files/beta-site.gif)
|
Volume Number: | 4 | |
Issue Number: | 9 | |
Column Tag: | Pascal Procedures |
Fun With Regions, Part II
By Stephen Dubin, V.M.D., Ph.D.,, Thomas W. Moore, PH.D., Drexel University, Sheel Kishore, MS
In our previous paper (Fun with Regions Part I: High Level Language Implementation), we showed how it was possible to estimate the area of an arbitrarily drawn region from high level languages such as Pascal and C using repeated application of the ROM subroutine, PtInRgn. Although this approach is simple and intuitive, execution time is excessive for large or complex regions. Whenever part of a high level language routine consumes an inconveniently long execution time, the possibility of using assembly language to achieve better efficiency should be considered. Two fundamental approaches may be applied. A study of the code generated by the compiler may reveal unnecessary looping, inefficient use of registers or other complexities which can be streamlined. If savings can be made within loops executed many times, the resulting speedup can be significant. The second approach applies when the task at hand is relatively simple and straightforward. In this case, determination of a specific efficient algorithm is the key, with translation to assembly code following directly. In dealing with the area computation part of our regions manipulation, we have explored both of these approaches and found the latter approach to be clearly superior.
In the first article, we provided the C Language code for the area calculation program at the end of the article with major subroutines interpolated into the text as implemented in Pascal. In this installment, the “tables are turned”; the interpolated routines are implemented for the Megamax C development system and the Consulair (MDS) assembly language system [This is included in the source code due to space considerations-ED]. At the end of the article, a program using the most useful of the assembly language optimizations is shown for Turbo Pascal with explanation of minor changes needed for the TML Pascal Development system. We recognize that our program may not be the most elegant or efficient approach to the problems; but even where an attempt at optimization yielded poor or marginal results, an interesting and - hopefully - useful technique is explored.
Figure 1.
Several authors ( Morton, M.: Reduce Your Time in the Traps! MacTutor October 1986 pp 21-24. and Knaster, S.: “How to Write Macintosh Software.” Hayden, Hasbrouck Heights NJ, 1986, p 368 ) have advocated bypassing the trap dispatcher as a means of speeding routines in which ROM calls are made repeatedly. Certainly our CountPix routine, since it calls PtInRgn for every point within the region bounding box, is a candidate for this type of optimization. Mike Morton presents the underlying mechanism for this strategy, points out some cautions and pitfalls and shows how to do this task in Pascal using INLINE calls. Briefly, any call to the ROM must go through an intermediate step of finding the “true” address of the call in the particular version of the ROM in your machine before it can be invoked. When a ROM call is to be used many times, this address may be determined one time by means of the GetTrapAddress function early in the program; then you may employ some means of jumping to this address directly whenever the particular ROM call is to be used. In C language, the address might be acquired as follows:
trap = gettrapaddress(0xa8e8); /* trap is a global long integer */ /* A8E8 is the trap # for ptinrgn */
In order to see how the jump might be made, consider the following “glue” routine which is used for ptinrgn by the Megamax system:
boolean ptinrgn(pt, rgn) /* as copied from qd13.c */ point *pt; rgnhandle rgn; { asm { subq #2,A7 /* make room on the stack for the result */ move.l pt(A6),A0/* address of point into A0 */ move.l (A0),-(A7) /* dereference and put onto stack*/ move.l rgn(A6),-(A7)/* region handle onto stack */ dc.w 0xa8e8/* call the ROM for ptinrgn */ move.b (A7)+,D0 /* result into D0 where C expects to find the answer */ ext.w D0 /* sign extend the result */ } }
Note that with Megamax inline assembly, the compiler takes care of setting up (and tearing down) the stack frame. Automatic (local) variables are accessed using the name of the variable as a displacement from A6. Global variables are treated similarly as offsets from A4. Thus if we have safely installed the true ROM address of ptinrgn in “trap”, we can write a “new improved” version of the glue routine as follows:
boolean zptinrgn(pt, rgn) /* same as ptinrgn except */ point *pt;/* bypasses the trap dispatcher */ rgnhandle rgn; { asm { subq #2,A7 /* make room on the stack for the result */ move.l pt(A6),A0/* address of point into A0*/ move.l (A0),-(A7) /* dereference and put onto stack */ move.l rgn(A6),-(A7)/* region handle onto stack */ move.l trap(A4),A2/* address of “true address” of ptinrgn into A2 */ jsr (A2) /* dereference once and jump there */ move.b (A7)+,D0 /* result into D0 where C expects to find the answer */ ext.w D0 /* sign extend the result */ } }
When this version was used in place of ptinrgn, the time needed to estimate the area of a region was decreased by 15% for small simple regions and about 9% for larger and more complex ones. Although this would ordinarily be considered a significant improvement, it is little comfort to know that a five minute computation can now be completed in only four and a quarter minutes.
Upon examination of the disassembled code for counbtpix(), we noticed that the most often used variables were the points delimiting the region bounding box as well as the “exploring” point on which we called ptinrgn. Following classical optimization strategy, the next step was to set up these data structures on registers rather than to fetch them every time the coordinates of the exploring point were incremented. The code for this implementation of the countpix function is shown below:
{1} bcountpix(theregion) /* sets up test point on registers */ rgnhandle theregion; { asm{ move.l trap(A4),A2; address of ptinregn move.w #0xA8E8,D0 ; trap number for ptinrgn dc.w 0xA146 ; call the trap, address is in A0 move.l A0,A2 ; put it into A2 move.l theregion(A6),A3 ; regionhandle move.l (A3),A1 ; dereference once move.l 2(A1),D4 ; topleft of rgnbox move.l 6(A1),D5 ; botright of rgnbox move.l D4,D6 ; copy of TL as VH current point hortest: cmp.w D5,D6 ; compare horizontal blt.s vertest ; go on swap D4 ; D4 is now HV addq.w #1,D4 ; down 1 row swap D4; now D4 is back to HV move.l D4,D6 ; make this the current test point vertest: swap D6; now is HV swap D5; now is RB cmp.w D5,D6 ; compare vertical blt.s pointest ; go on bra.s done; pointest: swap D5; back to BR swap D6; back to VH subq #2,A7 ; make room for result move.l D6,-(A7) ; point onto stack move.l theregion(A6),-(A7) ; rgnhandle onto stack jsr (A2); go to ROM move.b (A7)+,D0 ; result onto stack tst.b D0; was it true? beq skip; not this time addq.l #1,numpix(A4); yes, increment the counter skip: addq.w #1,D6 ; over 1 column bra.s hortest ; back for another point done: } }
As with the previous attempt at optimization, the speed increase with this approach was marginal at best. Our final attempt in this direction was to examine the code for ptinregn in the ROM in order to transpose (plagiarize?) it directly into the above routine. The result was surprising as well as disappointing. Although there was a measurable but tiny improvement for small simple regions, ones for which optimization was not needed anyway, the time needed to calculate the area of large, complex or disjoint regions increased significantly!! Our theory as to why this happens is based on the way in which the 68000 accesses ROM and RAM. Accesses to RAM (where the program resides) are shared with the video display, sound generator and disk speed controller. This leads to a RAM access rate of approximately six megahertz. The ROM has a “direct line” and is accessed at 7.83 MHz (Inside Macintosh. III-18, Addison-Wesley, Reading MA, 1985).
All of this preoccupation with ptinregn led to an understanding of why the area computation takes so long for large or complex regions. A flow chart of how ptinregn works is shown in figure 1. Unless the region under examination is rectangular, it may be necessary to examine all of the region data, one word at a time, before deciding whether the point is indeed in the region. This is particularly true as the exploring point moves toward increasing values in the vertical (y) component. Clearly, our original countpix procedure, which calls ptinregn on every point in the bounding box, covers the same ground many times. Based on a conviction that the region information should be adequate to permit estimation of the area with one pass, we resolved to implement a specific algorithm “from the ground up.”
As mentioned in the first installment of this article, region information is stored in memory in a way designed to require minimal space. A clear understanding of this method of encoding region boundaries is necessary in order to design our area calculating algorithm. To illustrate this process, consider the simple region plotted in Figure 2. The numbers on the plot are the coordinates of the “corners” of the outlined region. A memory dump of the data representing this region is shown below the graph. In order to design an algorithm for area calculation, we must understand the method of encoding the region in memory. As explained in our previous paper, the first five words of this data list are the data size in bytes (44) followed by the “upper left” and “lower right” coordinates of the rectangular boundary of the region - the regnbbox (100,100,220,200). Following these five words we find the information needed to compute area. Only horizontal boundary information is stored. The region being defined consists of the (rectangular) area under a given boundary line, extending down to the next horizontal boundary line encountered. Horizontal boundary lines are indicated by a y coordinate word, followed by start and stop x values (more than one pair if the line has multiple segments). The flag word, #7FFF, marks the end of the boundary segments at a particular y value. Therefore, the sixth word of data (100) is the y coordinate of the top boundary line of the region. It is followed by x values 100 (start of line) and 200 (end of line). Then we get a flag indicating that no more boundary segments exist at this level. The next horizontal boundary is the line from (125,150) to (180,150). Therefore we find 150 (y value) and 125,180 (x start and stop values) to be the next three words of the data. In this manner, the remaining data can be seen to define the region of figure 2. The double flag indicates the end of the data table.
The question now is how to use this table to calculate the area of the region thus defined. The arrangement of the data suggests dividing the area into rectangular pieces and adding their areas. We might start by subtracting the first x value (100) from the second (200) to get the width of the top of the first rectangle (marked “A”). The y value of this line could be put aside to be subtracted from the next y value (170) yielding the height of the rectangle. The product of these dimensions is one component of the final area
The x values following the 150 are endpoints of a new horizontal boundary line. Since this line falls under the previous boundary, it represents the bottom of a rectangular piece rather than the top of a new one. From here (y = 150 ) our region will now grow downward in two rectangular pieces, B and C. To calculate the areas of these pieces, we must make use of the two new x values found in the data table (125,180). If we arrange all x values found so far in order of magnitude (100, 125, 180, 200), the appropriate widths can be found by pairing the values and subtracting the first from the second in each pair. This is a rule we can use in our algorithm: maintain an ordered list of all encountered x values, pair them and subtract the first of each pair from the second. The sum of these differences will be the total width at the top of each of the rectangular regions. The y coordinate at the top of B and C is subtracted from the next y coordinate found (170-150) to determine the height of these rectangles. Height times width is then added to the accumulating total area.
One problem remains: how to end the process? Following y value 170, we find x values of 100 and 125 in the data. One leg of our descending area ends here so we would like our x table to list just 180, 200 (the top of D) from here on. Therefore, the final rule we need for our area algorithm is to remove entries from the x value ordered list whenever they are matched by a newly encountered x value. At the final y value (220), we subtract the remaining x pair (200-180), and multiply by the last y difference (220-170), giving the area of the last piece, D.
A flowchart of this process is given in figure 3. The routine as implemented to provide a linkable object file using the MDS (now Consulair) 68000 assembly system is shown below. Using the TTAA (Tom Terrific Area Algorithm), even the largest and most complex region that could be drawn on the Macintosh screen could have its area estimated in less than 20 seconds. Such regions take as long as ten minutes using the old CountPix.
This is the code for use with the MDS assembler to produce the file ACountPix.REL. This can then be linked to a Pascal or other “main” program.
{2} ; ;ACountPix.asm ;Pascal Usage: Function ACountPix( theRegion:RgnHandle) : LongInt; ;This function emulates CountPix ; Written by Thomas W. Moore, Ph.D. and Stephen Dubin, V.M.D., Ph.D. ;Copyright © 1987 XDEF ACountPix XREF myBUF ; The buffer is allocated in the calling program even though it might be ; more elegant to allocate it here with a DS statement; however Turbo ; Pascal V1.0 seems intolerant of this. ( p 335 of the Turbo Pascal manual) ;-------------------------------- INCLUDES ------------------------ Include Traps.D ; Use System and ToolBox traps Include ToolEqu.D; Use ToolBox equates ACountPix: link A6,#0 ; set up frame pointer movem.lA0-A3/D0-D7,-(A7) ; save the world clr.l -(A7) ; make room on stack for result movea.l8(A6),A0 ; region handle into A0 movea.l(A0),A0 ; dereference => pointer in A0 clr.l D7; set area to zero lea myBUF(A5),A1 ; lowest address of x list rectcheck:; see whether it is a rect and if so - do the job here cmpi.w #10,(A0) ; is this a single rectangle bne.s morework ; if not do the big job move.w 4(A0),D1 ; left move.w 8(A0),D2 ; right move.w 2(A0),D3 ; top move.w 6(A0),D4 ; bottom sub.w D1,D2 ; width sub.w D3,D4 ; height mulu.w D2,D4 ; area in D4 move.w D4,D7 ; lower word into D7 bra done morework: ; get ready for some serious work lea 10(A0),A0 ; beginning of region info clr.l D4; clr.l D2; move.l #512,D3 ; size of buffer to hold ordered list of x values adda.l D3,A1 ; highest address in buffer movea.lA1,A2 ; copy in A2 movea.lA1,A3 ; another in A3 move.w #-1,(A1) ; -1 in highest x address so that 1st x entry will be greater gety: ; read in y coordinate of next horizontalboundary move.w (A0)+, D3; latest y value jsr calc getx: move.w (A0)+,D1 ; new x value cmpi.w #$7fff,D1; flag indicates no more x values at this y bne storex; if no flag, it is a new x move.w (A0),D1 ; next word of region info cmpi.w #$7fff,D1; all done? beq done; yes go home bra gety; no, get next y storex: ; place new x value in proper place in ordered list movea.lA3,A1 ; A3 points to highest x value in ordered list cmp.w (A1),D1 ; compare new x value to largest entry bne s1; if not equal, it must be added to list addq #2,A3 ; if match, remove from list bra getx; next x s1: lea -2(A3),A3 ; add a space at high end of list for new x bgt.s insert ; if new x value is greatest, put it on top mkroom: ; new x is not greatest so we must move list values up to make room move.w (A1)+,-4(A1) ; move data up (1 word net distance) cmp.w (A1),D1 ; compare next list entry beq.s remove ; if it matches, remove it bcc.s insert ; it is greater, so put it above cmpa.l A1,A2 ; are we at bottom? bne mkroom; no, move another one up insert: ; insert new x value in ordered place in list move.w D1,-(A1) ; insert above present location bra getx; remove: ; erases an entry from the list subq #2,A1 ; point to next higher r1: cmpa.l A1,A3 ; is it the top? beq shrink; yes so exit move.w -(A1),4(A1); move greater x values down to replace bra r1; value removed shrink: addq #4,A3 ; if a match occurred, list shrinks by 2 words bra getx; one that we didn’t insert and one that we erased calc: ; determine new Height sub.w D3,D4 ; Y old - Y new neg.w D4; Height of the rectangle(s) newW: ; prepare for Width calculation clr.l D2; Will receive width clr.l D1; work reg movea.lA2,A1 ; reset A1 to point to least x value in list dx:; check to see if all x pairs have been used. ; multiply H x W and add to area cmpa.l A1,A3 ; A3 points to greatest x value in list bne morex ; if not equal, not all x’s have been used mulu D4,D2 ; H x W add.l D2,D7 ; add to accumulating area move.w D3,D4 ; for next time rts morex: ; subtracts x values in pairs adding differences to accumulating W move.w -(A1),D1 ; Xi (lower x value of a pair) sub.w -(A1),D1 ; Xi - Xi+1 (length of a horizontal boundary segment) neg.w D1; Xi+1 - Xi (correct sign) add.w D1,D2 ; W (add to accumulating width) bra dx; done: move.l D7,12(A6); store result “under” the last parameter movem.l(A7)+,A0-A3/D0-D7 ; restore registers unlk A6; restore original stack move.l (A7)+,A0 ; get return address addq.l #4,A7 ; remove parameters jmp (A0); return this way end
The same algorithm can be used with the very convenient inline assembly facility of the Megamax and other C development systems. Because these compilers take care of “tending the stack” for you, the entry and exit procedures are significantly simplified. For the Megamax systems, they are as follows:
{3} acountpix(theregion) rgnhandle theregion; { intbuf[1000]; asm{ move.l theregion(A6),A0 ; regionhandle note: local variables are referred off A6 move.l (A0),A0 ; dereference once => region pointer clr.l D7; set area to zero lea buf(A6),A1 ; lowest address of x list rectcheck: /* Everything in between is the same as in ACountPix.Asm above */ done: move.l D7,numpix(A4);report the answer note: global variables are referred offA4 } }
This is the code for our main calling program as implemented in Turbo Pascal:
{4} {PasArea.Pas } {Copyright 1987 by Stephen Dubin, V.M.D.and Thomas W. Moore,Ph.D. } {Prepared with Turbo Pascal V1.0 } { Users of other Pascal systems should particularly check the “preamble”} { portion of their program (Linking directives, “uses”, “includes”, etc.} { also check usage of type “point” - TML doesn’t like use of pt.h and} { pt.v as control elements in a for statement. } program PasArea; {$R-} { Turn off range checking } {$I-} { Turn off I/O error checking } {$R PasArea.rsrc} { Identify resource file } {$U-} { Turn off auto link to runtime units } {$L ACountPix.Rel } { Link in Assembly Language Segment} {$D+} { Embed Procedure Labels } uses Memtypes,QuickDraw,OSIntf,ToolIntf,PackIntf; const FileMenuID = 1;{ the File menu} OptionMenuID = 2;{ the option menu} WindResID = 1; { the resource id of my window} type BUF = array[1..512] of Integer; { Make it bigger if you are really paranoid} var myMenus : Array[FileMenuId..OptionMenuID] of MenuHandle; Done : Boolean; MyWindow : WindowPtr; TotalRegion : RgnHandle; Numpix : Longint; myBUF : BUF; function ACountPix( theRegion:RgnHandle) : LongInt; external; function CountPix(theRegion : RgnHandle): LongInt; var pt : Point; rgn : Region; temp : LongInt; x : Integer; y : Integer; begin temp := 0; rgn := theRegion^^; for x := rgn.rgnBBox.left to rgn.rgnBBox.right do begin pt.h := x; for y := rgn.rgnBBox.top to rgn.rgnBBox.bottom do begin pt.v := y; if PtInRgn( pt, TheRegion) then temp := temp + 1; end; end; CountPix := temp; end; { Turbo seems to accept pt.h and pt.v as control elements but TML does} { not. Some format checkers agree with TML} procedure Wipe; var r : Rect; begin SetRect(r,0,0,504,300); EraseRect(r); end; procedure Data; var rgn : Region; rgnpntr : Ptr; size : Integer; thebuf : BUF; bfpntr : Ptr; myString : Str255; i : Integer; x : Integer; y : Integer; begin Wipe; TextSize(9); TextFont(Monaco); rgn := totalRegion^^; rgnpntr := ptr(totalRegion^); size := rgn.rgnSize; if size > 800 then size:= 800; bfpntr := ptr(@thebuf); BlockMove(rgnpntr,bfpntr,size); MoveTo(10,10); DrawString(‘Here are the first 400 words of the region data. (FLAG = 32767)’); x := 10; y := 20; for i := 1 to (size div 2) do begin MoveTo(x,y); NumToString(theBuf[i],myString); if theBuf[i] < 32766 then begin if theBuf[i] <10 then DrawString(‘ ‘); if theBuf[i] <100 then DrawString(‘ ‘); if theBuf[i] < 1000 then DrawString(‘ ‘); if theBuf[i] < 10000 then DrawString(‘ ‘); DrawString(MyString); end; if theBuf[i] > 32766 then DrawString(‘ FLAG’); x := x + 30; if (i mod 16) = 0 then begin x := 10; y := y+10; end; end; end; procedure OvalRegion; var RectA : Rect; begin Wipe; TotalRegion := NewRgn; SetRect(RectA, 170,175,195,200); OpenRgn; ShowPen; FrameOval(RectA); HidePen; CloseRgn(TotalRegion); end; procedure Contour; var p1 : Point; p2 : Point; OldTick : Longint; begin Wipe; TotalRegion := NewRgn; OldTick := TickCount; Repeat GetMouse(p1); MoveTo(p1.h,p1.v); p2 := p1; Until Button = True; OpenRgn; ShowPen; PenMode(patXor); Repeat GetMouse(p2); Repeat Until (OldTick <> TickCount); LineTo(p2.h,p2.v); Until Button <> True; Repeat Until (OldTick <> TickCount); LineTo(p1.h,p1.v); PenNormal; HidePen; CloseRgn(TotalRegion); InvertRgn(TotalRegion); end; procedure Example; begin Wipe; OpenRgn; TotalRegion := NewRgn; ShowPen; MoveTo(100,100); LineTo(200,100); LineTo(200,220); LineTo(180,220); LineTo(180,150); LineTo(125,150); LineTo(125,170); LineTo(125,170); LineTo(100,170); LineTo(100,100); HidePen; CloseRgn(TotalRegion); end; procedure FreeBox; var p1 : Point; p2 : Point; p3 : Point; OldTick : Longint; MyRect : Rect; begin Wipe; TotalRegion := NewRgn; OldTick := TickCount; PenPat(gray); PenMode(patXor); Repeat GetMouse(p1); p2 := p1; Until Button = True; OpenRgn; ShowPen; PenMode(patXor); Repeat Pt2Rect(p1,p2,MyRect); Repeat Until (OldTick <> TickCount); FrameRect(MyRect); Repeat GetMouse(p3); Until EqualPt(p2,p3) <> True; Repeat Until (OldTick <> TickCount); FrameRect(MyRect); p2 := p3; Until Button <> True; Pennormal; HidePen; PenPat(black); FrameRect(MyRect); CloseRgn(TotalRegion); InvertRgn(TotalRegion); end; procedure Area; var NumTix : LongInt; MoreTix : LongInt; TicString : Str255; PixString : Str255; begin TextFont(Monaco); TextSize(9); TextMode(0); MoveTo(10,20); DrawString(‘ Using Pascal ‘); NumTix := TickCount; NumPix := CountPix( TotalRegion ); MoreTix := TickCount - NumTix; NumToString(MoreTix,TicString); NumToString(NumPix,PixString); MoveTo(10,30); DrawString(‘ Tickcount = ‘); MoveTo(120,30); DrawString(TicString); MoveTo(10,40); DrawString(‘ Pixel Number = ‘); MoveTo(120,40); DrawString(PixString); MoveTo(10,50); DrawString(‘ Using Tom Terrific ‘); NumTix := TickCount; NumPix := ACountPix( TotalRegion ); MoreTix := TickCount - NumTix; NumToString(MoreTix,TicString); NumToString(NumPix,PixString); MoveTo(10,60); DrawString(‘ Tickcount = ‘); MoveTo(120,60); DrawString(TicString); MoveTo(10,70); DrawString(‘ Pixel Number = ‘); MoveTo(120,70); DrawString(PixString); end; procedure ProcessMenu(codeWord : Longint); var menuNum : Integer; itemNum : Integer; begin if codeWord <> 0 then begin menuNum := HiWord(codeWord); itemNum := LoWord(codeWord); case menuNum of FileMenuID :Done := true; OptionMenuID : begin case ItemNum of 1:Contour; {Contour} 2:FreeBox; {Freebox} 3:OvalRegion; {Oval} 4:Example; {Example} 5: Area; {Area} 6:Data; {Region Data} end; { of ItemNum case} end;{ of MenuNum case} end; HiliteMenu(0); end; end; procedure DealWithMouseDowns(theEvent: EventRecord); var location : Integer; windowPointedTo : WindowPtr; mouseLoc : point; windowLoc : integer; VandH : Longint; Height : Integer; Width : Integer; begin mouseLoc := theEvent.where; windowLoc := FindWindow(mouseLoc,windowPointedTo); case windowLoc of inMenuBar : begin ProcessMenu(MenuSelect(mouseLoc)); end; end; end; procedure MainEventLoop; var Event : EventRecord; theItem : integer; begin repeat SystemTask; if GetNextEvent(everyEvent, Event) then begin case Event.what of mouseDown : DealWithMouseDowns(Event); end; end; until Done; end; procedure MakeMenus; var index : Integer; begin for index := FileMenuId to OptionMenuID do begin myMenus[index] := GetMenu(index); InsertMenu(myMenus[index],0); end; DrawMenuBar; end; { Main Program } begin Done := false; FlushEvents(everyEvent,0); InitGraf(@thePort); InitFonts; InitWindows; InitMenus; InitDialogs(nil); InitCursor; MoreMasters; MoreMasters; MakeMenus; MyWindow := GetNewWindow(WindResID,nil,Pointer(-1)); SetPort(MyWindow); TotalRegion := NewRgn; {Lazy way to avoid bomb if your select “Area” first} MainEventLoop; end. Here is the resource file for use with the above program (Turbo Pascal): * * Resource listing from file: “PasArea.R”. * PasArea.rsrc Type AREA = STR ,0 PasArea, by Stephen Dubin and Thomas W. Moore Copyright © 1987 Type WIND ,1 Fun with Regions II 40 5 330 505 Visible NoGoAway 0 0 Type MENU ,1 File Quit ,2 Option Contour Freebox Oval Example Compute Area Region Data
In order to compile the same program with TML Pascal V2.0, a few minor adjustments were needed. The preamble was changed to:
{5} program TMPasArea; {$T APPL AREA} { set the type and creator} {$B+} { set the bundle bit} {$L TMPasAreaRes}{ link the resource file too...} uses MacIntf; { Constant, Type and Variable declarations as above are the same as in PasArea.Pas above} {Declare the Assembly Language routine as external } function ACountPix( theRegion:RgnHandle) : LongInt; external; {$U ACountPix } { This directive will not appear in the .link file unless it follows the declaration of the } { relocatable object file as external}
The only change needed in the body of the program was in the high level CountPix function. A form that compiled with TML is:
{6} function CountPix(theRegion : RgnHandle): LongInt; var pt : Point; rgn : Region; temp : LongInt; x : Integer; y : Integer; begin temp := 0; rgn := theRegion^^; for x := rgn.rgnBBox.left to rgn.rgnBBox.right do begin pt.h := x; for y := rgn.rgnBBox.top to rgn.rgnBBox.bottom do begin pt.v := y; if PtInRgn( pt, TheRegion) then temp := temp + 1; end; end; CountPix := temp; end;
TML does not seem to like having pt.h and pt.v as control elements. PasMat, a Pascal formatting and syntax checking program, agrees with TML on this point. In keeping with our local traditions, the first non-comment line of our TML resource file was “TMPasAreaRes”. Although it probably is of little interest in these days of monstrous memories, the TML version of the program requires 3,305 bytes of memory; whereas the Turbo program weighs in at a hefty 10,855 bytes.
Some final zingers for the reader - Although it was certainly necessary for us to use assembler to plumb the depths of the ROM and to work out the algorithm for making our area measurement lightning fast; one might consider whether the same algorithm might now be implemented entirely from C, Pascal or possibly Basic. Would the speed be degraded to any appreciable extent? Will a new call AreaRgn be found in the 512K Roms on the Mack III’s?
![](/file/12652/www.mactech.com.tar/www.mactech.com/sites/all/themes/custom_front/img/search_text.gif)
- SPREAD THE WORD:
- Slashdot
- Digg
- Del.icio.us
- Newsvine