Post by thread:[PIC]: How to make a simple and efficient computed

Hello, I'm porting a software that is running ok on a PIC16F877 to a PIC18F452 and now I'm having problems with the computed goto's. On the PIC16: MOVLW 0x01 MOVWF PCLATH MOVF state,W_ ADDWF PCL,F_ GOTO case0 GOTO case1 GOTO case2 etc On the PIC18 strange things happen: MOVLW 0x01 MOVWF PCLATH RLCF dkstate, W_ ADDWF PCL GOTO case0 GOTO case1 GOTO case2 etc Someone have an example code about computed goto's on a PIC18? I've seen the "Unlimited computed goto (or table lookup)" page by Bob Ammerman on PICLIST, but it seens overkill. I need very fast code. Thank you very much, Brusque ----------------------------------------------------------------- Edson Brusque C.I.Tronics Lighting Designers Ltda Research and Development Blumenau - SC - Brazil Say NO to HTML mail http://www.citronics.com.br ----------------------------------------------------------------- -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email spam_OUTlistservTakeThisOuTmitvma.mit.edu with SET PICList DIGEST in the body

> I'm porting a software that is running ok on a PIC16F877 to a PIC18F452 {Quote hidden}> and now I'm having problems with the computed goto's. > > On the PIC16: > > MOVLW 0x01 > MOVWF PCLATH > MOVF state,W_ > ADDWF PCL,F_ > GOTO case0 > GOTO case1 > GOTO case2 > etc > > On the PIC18 strange things happen: > MOVLW 0x01 > MOVWF PCLATH > RLCF dkstate, W_ > ADDWF PCL > GOTO case0 > GOTO case1 > GOTO case2 > etc First, DOCUMENT YOUR CODE!!! If you had forced yourself to explain what each instruction was doing, you might have found the bug yourself. Second, why are you hard setting PCLATH to 1? This is very bad programming practise and a bug waiting to happen (maybe it just did). Your code will only work if all the GOTO instructions are located from 100h - 1FFh. This should be derived symbolically, or at least the same symbolic constant used to force the address of the table as derive the value to stick into PCLATH. Note that the PIC16 addresses program memory as whole words, with each instruction taking up one word. The PIC18 addresses program memory in bytes, with instructions taking up either 2 or 4 bytes. The PIC18 GOTO instruction takes up 4 bytes, so the DKSTATE needs to be multiplied be 4, not 2, before used to index into the table. Also, this type of GOTO table can only support up to 64 entries because that are all that fit into a 256 byte region. Larger tables require computation of the PCLATH value, since a single PCLATH value addresses a 256 byte region. Note also that the PIC18 has a PCLATU register in addition to PCLATH. You can get away with ignoring PCLATU on the 18F452 because it only has 32Kbytes of program memory, but this is another bug waiting to happen. I would use assembly conditionals to insert the proper code if this is ever ported to a machine with more than 64Kbytes. Or, at least generate an assembly error in that case so that someone is forced to inspect the code. It's easy to do now, a real pain 4 years from now when you've forgotten all about this little section of code. While you're at it, you can generate an assembly error automatically is someone tries to grow the table beyond 64 entries. ***************************************************************** Embed Inc, embedded system specialists in Littleton Massachusetts (978) 742-9014, http://www.embedinc.com -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email .....listservKILLspam@spam@mitvma.mit.edu with SET PICList DIGEST in the body

Below is some working code. Note that the branch instruction is two bytes long, but the goto is four, thus some of the problem. Also, you need to use the rlncf so you don't accidentally end up on an odd address (though that might be thrown out when it gets loaded into the pc). Harold movlw high(RXJA) movwf pclath ; Update pc latch for table calls rlncf RxState,0 ; Get receive state*2 in W since bra instructions are 2 bytes long andlw 0x07 ; Get rid of invalid states addlw low(RXJA) ; w now holds low half of address we want to jump to skpnc incf pclath,1 ; Add carry into pclath if needed movwf pcl ; Jump into table RXJA bra RxState0 ; Go handle wait for FE (state 0) bra RxState1 ; Go handle wait for start code (state 1) bra RxState2 ; Go handle receive data (state 2) return ; should never be executed (invalid state 3) On Tue, 13 Aug 2002 11:21:49 -0300 "Edson Brusque (Listas)" <brusque.listasKILLspamCITRONICS.COM.BR> writes: {Quote hidden}> Hello, > > I'm porting a software that is running ok on a PIC16F877 to a > PIC18F452 > and now I'm having problems with the computed goto's. > > On the PIC16: > > MOVLW 0x01 > MOVWF PCLATH > MOVF state,W_ > ADDWF PCL,F_ > GOTO case0 > GOTO case1 > GOTO case2 > etc > > On the PIC18 strange things happen: > MOVLW 0x01 > MOVWF PCLATH > RLCF dkstate, W_ > ADDWF PCL > GOTO case0 > GOTO case1 > GOTO case2 > etc > > Someone have an example code about computed goto's on a PIC18? > > I've seen the "Unlimited computed goto (or table lookup)" page > by Bob > Ammerman on PICLIST, but it seens overkill. I need very fast code. > > Thank you very much, > > Brusque > > ----------------------------------------------------------------- > Edson Brusque C.I.Tronics Lighting Designers Ltda > Research and Development Blumenau - SC - Brazil > Say NO to HTML mail http://www.citronics.com.br > ----------------------------------------------------------------- > > -- > http://www.piclist.com#nomail Going offline? Don't AutoReply us! > email .....listservKILLspam.....mitvma.mit.edu with SET PICList DIGEST in the body > > FCC Rules Online at http://hallikainen.com/FccRules Lighting control for theatre and television at http://www.dovesystems.com Reach broadcasters, engineers, manufacturers, compliance labs, and attorneys. Advertise at http://www.hallikainen.com/FccRules/ . ________________________________________________________________ GET INTERNET ACCESS FROM JUNO! Juno offers FREE or PREMIUM Internet access for less! Join Juno today! For your FREE software, visit: dl.http://www.juno.com/get/web/. -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email EraseMElistservspam_OUTTakeThisOuTmitvma.mit.edu with SET PICList DIGEST in the body

Hello Olin, > First, DOCUMENT YOUR CODE!!! If you had forced yourself to explain what > each instruction was doing, you might have found the bug yourself. :) I agree with you. But this is documented on the starting of the file. > Second, why are you hard setting PCLATH to 1? This is very bad programming > practise and a bug waiting to happen (maybe it just did). Your code will > only work if all the GOTO instructions are located from 100h - 1FFh. This > should be derived symbolically, or at least the same symbolic constant used > to force the address of the table as derive the value to stick into PCLATH. This is the reason why this code is documented on the starting of the file with a big WARNING! I'm using PCH (by CCS) and haven't found a way to derive the PCLATH value simbolically. CCS have a label_address() function but I can't find a way to make something like "PCLATH = hi(label_address(dkstart));" to work. > Note that the PIC16 addresses program memory as whole words, with each > instruction taking up one word. The PIC18 addresses program memory in > bytes, with instructions taking up either 2 or 4 bytes. The PIC18 GOTO > instruction takes up 4 bytes, so the DKSTATE needs to be multiplied be 4, > not 2, before used to index into the table. Also, this type of GOTO table > can only support up to 64 entries because that are all that fit into a 256 > byte region. Larger tables require computation of the PCLATH value, since a > single PCLATH value addresses a 256 byte region. Yes, this is the problem. I've changed the code to: MOVLW 0x01 MOVWF PCLATH MOVF dkstate,W_ MULLW 4 MOVF PRODL,W_ ADDWF PCL GOTO case0 GOTO case1 GOTO case2 and now it seens to be working fine. The 64 "cases" limitation isn't a issue in this application. I'm taking the precautions to not having PCL "pages" crossing. > Note also that the PIC18 has a PCLATU register in addition to PCLATH. You > can get away with ignoring PCLATU on the 18F452 because it only has 32Kbytes > of program memory, but this is another bug waiting to happen. I would use > assembly conditionals to insert the proper code if this is ever ported to a > machine with more than 64Kbytes. Or, at least generate an assembly error in > that case so that someone is forced to inspect the code. It's easy to do > now, a real pain 4 years from now when you've forgotten all about this > little section of code. While you're at it, you can generate an assembly > error automatically is someone tries to grow the table beyond 64 entries. This is a good idea. I don't think this code will ever be implemented on more powerfull machines, but better be safe than sorry. :) Thank you very much, Brusque ----------------------------------------------------------------- Edson Brusque C.I.Tronics Lighting Designers Ltda Research and Development Blumenau - SC - Brazil Say NO to HTML mail http://www.citronics.com.br ----------------------------------------------------------------- -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email listservspam_OUTmitvma.mit.edu with SET PICList DIGEST in the body

Hello Harold, > Below is some working code. Note that the branch instruction is two > bytes long, but the goto is four, thus some of the problem. yes, this was the problem. >Also, you > need to use the rlncf so you don't accidentally end up on an odd address > (though that might be thrown out when it gets loaded into the pc). The rlncf is a bit strange for me, as it makes "0b10110000" into "0b01100001" (the eighth bit turns first), so you have to be sure about the state of this bit. Not a problem in this code anyway. Maybe I'll use it. Your code is smart. Just don't know how to port it to CCS... oh boy! Thank you, Brusque ----------------------------------------------------------------- Edson Brusque C.I.Tronics Lighting Designers Ltda Research and Development Blumenau - SC - Brazil Say NO to HTML mail http://www.citronics.com.br ----------------------------------------------------------------- -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email @spam@listservKILLspammitvma.mit.edu with SET PICList DIGEST in the body

> > First, DOCUMENT YOUR CODE!!! If you had forced yourself to explain what > > each instruction was doing, you might have found the bug yourself. > > :) I agree with you. But this is documented on the starting of the file. That's no excuse. Yes, header comments are good, but individual details must be documented too. > This is the reason why this code is documented on the starting of the > file with a big WARNING! I'm using PCH (by CCS) and haven't found a way to > derive the PCLATH value simbolically. I thought CCS was a C compiler, but the code you showed was in assembler. > This is a good idea. I don't think this code will ever be implemented on > more powerfull machines, Yeah, right. That's probably what the original author who wrote the 16F877 code thought too. > but better be safe than sorry. :) Yes. ***************************************************************** Embed Inc, embedded system specialists in Littleton Massachusetts (978) 742-9014, http://www.embedinc.com -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email KILLspamlistservKILLspammitvma.mit.edu with SET PICList DIGEST in the body

Hello Olin, > I thought CCS was a C compiler, but the code you showed was in assembler. that's because you can use inline assembly on a C compiler when you want maximum optimization. CCS assembly support is very good, but it always lacks something when you really need... > > This is a good idea. I don't think this code will ever be implemented > > on more powerfull machines, > Yeah, right. That's probably what the original author who wrote the 16F877 > code thought too. Hahaha, good point but I'm the original author and I only wrote the code for the 16F877 because the 18F452s haven't arrived at that time and I really was needing something working to send to the client. Best regards, Brusque ----------------------------------------------------------------- Edson Brusque C.I.Tronics Lighting Designers Ltda Research and Development Blumenau - SC - Brazil Say NO to HTML mail http://www.citronics.com.br ----------------------------------------------------------------- -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email RemoveMElistservTakeThisOuTmitvma.mit.edu with SET PICList DIGEST in the body

Hello Olin, Harold et all, I've been studing (mmm, is this conjugation correct?) the Bob's code at http://www.piclist.com/techref/microchip/18c/ongoto-ba.htm and adapted it to my situation: // COMPUTED GOTO #asm rlncf state, W_ // put state*2 movwf gotoarg // in gotoarg rcall comp_goto // This is the trick, comp_goto deals with TOSL and TOSH goto case00 // then returns to the correct "goto caseXX" goto case01 goto case02 goto case03 comp_goto: rlncf gotoarg, W_ // add gotoarg*2 addwf TOSL,F // on TOSL movlw 0 addwfc TOSH,F // if carry, increments TOSH return // returns directly to the "goto caseXX" #endasm It's not very efficient. My "brute-force" method uses 8 instruction and Bob's method uses 14, but it works. Bob's code have 16-bit limit, my implementation is 7-bit (128 gotos possible). Now I'll try to optimize it. Best regards, Brusque ----------------------------------------------------------------- Edson Brusque C.I.Tronics Lighting Designers Ltda Research and Development Blumenau - SC - Brazil Say NO to HTML mail http://www.citronics.com.br ----------------------------------------------------------------- -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email spamBeGonelistservspamBeGonemitvma.mit.edu with SET PICList DIGEST in the body

On Tue, 13 Aug 2002 18:29:43 -0300 "Edson Brusque (Listas)" <TakeThisOuTbrusque.listasEraseMEspam_OUTCITRONICS.COM.BR> writes: > The rlncf is a bit strange for me, as it makes "0b10110000" into > "0b01100001" (the eighth bit turns first), so you have to be sure > about the > state of this bit. Not a problem in this code anyway. Maybe I'll use > it. > Oops! I forgot that the "no carry" means to just leave it out of the loop, but there is, nonetheless, a loop of the msb to the lsb. Should just do a ANDLW 0xfe afterwards to clear that bit, or do an rlcf, but do a clrc first... Harold FCC Rules Online at http://hallikainen.com/FccRules Lighting control for theatre and television at http://www.dovesystems.com Reach broadcasters, engineers, manufacturers, compliance labs, and attorneys. Advertise at http://www.hallikainen.com/FccRules/ . ________________________________________________________________ GET INTERNET ACCESS FROM JUNO! Juno offers FREE or PREMIUM Internet access for less! Join Juno today! For your FREE software, visit: dl.http://www.juno.com/get/web/. -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email RemoveMElistservTakeThisOuTmitvma.mit.edu with SET PICList DIGEST in the body

As I recall, the application is in C, so you could always use a switch-case statement and let the compiler deal with it! Harold FCC Rules Online at http://hallikainen.com/FccRules Lighting control for theatre and television at http://www.dovesystems.com Reach broadcasters, engineers, manufacturers, compliance labs, and attorneys. Advertise at http://www.hallikainen.com/FccRules/ . ________________________________________________________________ GET INTERNET ACCESS FROM JUNO! Juno offers FREE or PREMIUM Internet access for less! Join Juno today! For your FREE software, visit: dl.http://www.juno.com/get/web/. -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email listservEraseME.....mitvma.mit.edu with SET PICList DIGEST in the body

Harold M Hallikainen <EraseMEPICLISTmitvma.mit.edu> wrote: > > The rlncf is a bit strange for me, as it makes "0b10110000" into > > "0b01100001" (the eighth bit turns first), so you have to be sure > > about the state of this bit. > > Oops! I forgot that the "no carry" means to just leave it out of > the loop, but there is, nonetheless, a loop of the msb to the lsb. > Should just do a ANDLW 0xfe afterwards to clear that bit, or do > an rlcf, but do a clrc first... I missed the start of this thread, so I haven't seen the code, but can't you just add WREG to itself to do this? -Andrew === Andrew Warren -- RemoveMEaiwEraseMEEraseMEcypress.com === Principal Design Engineer === Cypress Semiconductor Corporation === === Opinions expressed above do not === necessarily represent those of === Cypress Semiconductor Corporation -- http://www.piclist.com#nomail Going offline? Don't AutoReply us! email RemoveMElistservspam_OUTKILLspammitvma.mit.edu with SET PICList DIGEST in the body

Hello Harold et all, > As I recall, the application is in C, so you could always use a > switch-case statement and let the compiler deal with it! yes, but the Assembly code generated by the C compiler isn't that tight. I managed to make my computed goto code a little more smart. It have a limitation of 63 possible GOTOs, but executes on just 11 instruction cycles: // COMPUTED GOTO #asm rlncf state, W_ // put state*4 addwf WREG, W_ // in WREG addlw 6 // align future PUSH position push // put PCL+2 on the TOS addwf TOSL,F // add WREG to TOSL clrf WREG addwfc TOSH,F // if carry, increments TOSH return // returns directly to the "goto caseXX" goto case00 goto case01 goto case02 goto case03 etc... #endasm Althought this code works perfectly while I'm simulating it in MPLAB, but doesn't work on the device. This is very strange. I've made some tests and it seens to be a bug on the PIC18F452 part(!). It's like the STKPTR isn't incremented when PUSH is executed. Someone want to try it? Best regards, Brusque ----------------------------------------------------------------- Edson Brusque C.I.Tronics Lighting Designers Ltda Research and Development Blumenau - SC - Brazil Say NO to HTML mail http://www.citronics.com.br ----------------------------------------------------------------- -- http://www.piclist.com hint: To leave the PICList EraseMEpiclist-unsubscribe-requestspamspamBeGonemitvma.mit.edu

On Tue, 13 Aug 2002 18:42:15 -0700 Andrew Warren <RemoveMEaiwKILLspamCYPRESS.COM> writes: > I missed the start of this thread, so I haven't seen the code, > but can't you just add WREG to itself to do this? > Yes you can! And why didn't I do it? Too clever, I guess. Just didn't think of it! That saves us one instruction, though putting an andlw after the add and prior to moving it to pcl would allow us to force the lsb low and force some higher bits low to prevent us jumping to invalid states. Adding wreg to itself is a very clever idea, though! Harold FCC Rules Online at http://hallikainen.com/FccRules Lighting control for theatre and television at http://www.dovesystems.com Reach broadcasters, engineers, manufacturers, compliance labs, and attorneys. Advertise at http://www.hallikainen.com/FccRules/ . ________________________________________________________________ GET INTERNET ACCESS FROM JUNO! Juno offers FREE or PREMIUM Internet access for less! Join Juno today! For your FREE software, visit: dl.http://www.juno.com/get/web/. -- http://www.piclist.com hint: To leave the PICList piclist-unsubscribe-requestSTOPspamspam_OUTmitvma.mit.edu