On 2012-08-16, David Brown <david.brown@removethis.hesbynett.no> wrote:> On 16/08/12 20:38, Jon Kirwan wrote: >> On Thu, 16 Aug 2012 20:09:56 +0200, David Brown >> <david.brown@removethis.hesbynett.no> wrote: >> >>> <snip> >>> Anyway, Andr? - if you have never used an 8051, then don't go there now. >>> <snip> >> >> I think this is good advice, broadly speaking. The 8051 was >> incredible in its day. Seriously - a nice chip. But it is >> seriously hampered in accessing external memory as data. Just >> take a look at what is involved in something as basic as a >> memcpy() for example. > > Yes, it was powerful in its day - but its day was some 30 years ago! > For the last 15 years at least, it has only existed because it is > popular, and it is only popular because it is popular, not because it is > a technically or economically good choice (somewhat like the x86 > architecture, or most MS software).Well said. If you're starting out right now, I'd recommend: * Atmel AVR for cheap 8-bit. * TI MSP430 for cheap 16-bit. There are dozens and dozens of cheap eval/demo boards for both of the above. [Don't bother with the MSP430 'X' parts which try to be 20-bit processors -- if you want more than 64K of address space, jump to an ARM part.] Do not, under any conditions, touch a PIC part the CPU architecture is _amazingly_ awful. 8051 is slightly less painful, but it's still bad compared to an AVR or MSP430. Personally, I think the MSP430 is easier to use than the AVR (especially if you end up doing any assembly language stuff), but the AVR is still good. * ARM Cortex-M for cheap 32-bit. There are about a half-dozen uC vendors to choose from (NXP, TI/Stellaris, Motorola, Samsung, Toshiba, and so on.). The above are all available as single-chip solutions (on-chip SRAM and flash). If you want to use external flash and SDRAM (e.g. to run Linux), then I'd go with: * ARM9 or Cortex-A (Atmel, TI, Motorla, etc.) GCC is avaialable for all the above. Once you've learned how to use gcc and gnu binutils, those skills apply to everything from an AVR/MSP with 8K of flash and 256 bytes of RAM to an IBM system Z "mainframe". -- Grant Edwards grant.b.edwards Yow! My haircut is totally at traditional! gmail.com
WAS:anybody know a good C compiler....
Started by ●August 16, 2012
Reply by ●August 16, 20122012-08-16
Reply by ●August 16, 20122012-08-16
In article <k0jm3k$mmc$1@reader1.panix.com>, invalid@invalid.invalid says...> > On 2012-08-16, David Brown <david.brown@removethis.hesbynett.no> wrote: > > On 16/08/12 20:38, Jon Kirwan wrote: > >> On Thu, 16 Aug 2012 20:09:56 +0200, David Brown > >> <david.brown@removethis.hesbynett.no> wrote: > >> > >>> <snip> > >>> Anyway, Andr? - if you have never used an 8051, then don't go there now. > >>> <snip> > >> > >> I think this is good advice, broadly speaking. The 8051 was > >> incredible in its day. Seriously - a nice chip. But it is > >> seriously hampered in accessing external memory as data. Just > >> take a look at what is involved in something as basic as a > >> memcpy() for example. > > > > Yes, it was powerful in its day - but its day was some 30 years ago! > > For the last 15 years at least, it has only existed because it is > > popular, and it is only popular because it is popular, not because it is > > a technically or economically good choice (somewhat like the x86 > > architecture, or most MS software). > > Well said. > > > If you're starting out right now, I'd recommend: > > * Atmel AVR for cheap 8-bit. > > * TI MSP430 for cheap 16-bit. > > There are dozens and dozens of cheap eval/demo boards for both of the > above. [Don't bother with the MSP430 'X' parts which try to be 20-bit > processors -- if you want more than 64K of address space, jump to an > ARM part.] > > Do not, under any conditions, touch a PIC part the CPU architecture is > _amazingly_ awful. 8051 is slightly less painful, but it's still bad > compared to an AVR or MSP430. Personally, I think the MSP430 is > easier to use than the AVR (especially if you end up doing any > assembly language stuff), but the AVR is still goodI've only looked briefly at the AVR, but I've done a dozen designs with various MSP430 variants. I've used some of the larger flash versions, but never had to use more than about 36K of that program memory. I used the larger chips mostly because the later versions have faster clock rates. The one limitation that I think about most is the limited RAM--- about 10K in the versions I've used. If you need a big buffer to accept interrupt-driven input while you handle the vagaries of SD card write timing, that 10K can seem a bit limited. If you need big buffers (such as for a camera), it's time to move up to a Cortex M3 or M4. You can start at 64K RAM and go up to 160K without much searching.> > * ARM Cortex-M for cheap 32-bit. > > There are about a half-dozen uC vendors to choose from (NXP, > TI/Stellaris, Motorola, Samsung, Toshiba, and so on.).And ST. I've been pretty happy with the peripherals and supporting libraries for the STM32 chips.> > The above are all available as single-chip solutions (on-chip SRAM and > flash). If you want to use external flash and SDRAM (e.g. to run > Linux), then I'd go with: > > * ARM9 or Cortex-A (Atmel, TI, Motorla, etc.)Yup. 160K RAM hardly gets you started on an OS which wants lots of buffers for ethernet stacks, file system, video and USB drivers. Beagleboards and their kin are good place to start here. I worked in that area for half a year----but I'm really happier down in the Cortex and MSP430 environment.> > GCC is avaialable for all the above. Once you've learned how to use > gcc and gnu binutils, those skills apply to everything from an AVR/MSP > with 8K of flash and 256 bytes of RAM to an IBM system Z "mainframe".Mark Borgerson
Reply by ●August 17, 20122012-08-17
On 2012-08-17, Mark Borgerson <mborgerson@comcast.net> wrote:> I've only looked briefly at the AVR, but I've done a dozen designs with > various MSP430 variants.The main annoyance in the AVR is that it's strictly an 8-bit CPU. It can't really do any 16-bit operations. That would be fine if it had an 8 bit address space, but it's got two different 16-bit address spaces. So, if you need to do anything at all with pointers (and C is designed to use pointers quite a bit) the AVR falls down badly compared to the MSP430 -- which has sixteen 16-bit registers, a single 16-bit address space, and a very elegent, orthogonal instructions set (rather reminiscent of the PDP-11) where any register can be used as a pointer, accumulator, index, whatever. The old "8-bit" 8080/Z80 and 6800/6502 processors had a 1 or 2 16-bit registers and a couple 16-bit operations specifically to deal with the pointer issue. Unfortuantely, the AVR failed to learn from them and is missing those features -- and it hurts.>> The above are all available as single-chip solutions (on-chip SRAM and >> flash). If you want to use external flash and SDRAM (e.g. to run >> Linux), then I'd go with: >> >> * ARM9 or Cortex-A (Atmel, TI, Motorla, etc.) > > Yup. 160K RAM hardly gets you started on an OS which wants lots of > buffers for ethernet stacks, file system, video and USB drivers.I should add that there are a _few_ (very few) Cortex-M parts that have external busses and include SDRAM controllers (NXP has one family), so it is possible to do large-memory stuff with Cortex-M. But, it's far more common to move up to the Cortex-A or the slightly older ARM9 families if you need memory measured in MB rather than KB.> Beagleboards and their kin are good place to start here. I worked in > that area for half a year----but I'm really happier down in the > Cortex and MSP430 environment.-- Grant Edwards grant.b.edwards Yow! I joined scientology at at a garage sale!! gmail.com
Reply by ●August 17, 20122012-08-17
On 17/08/2012 15:17, Grant Edwards wrote:> On 2012-08-17, Mark Borgerson <mborgerson@comcast.net> wrote: > >> I've only looked briefly at the AVR, but I've done a dozen designs with >> various MSP430 variants. > > The main annoyance in the AVR is that it's strictly an 8-bit CPU. It > can't really do any 16-bit operations. That would be fine if it had > an 8 bit address space, but it's got two different 16-bit address > spaces. So, if you need to do anything at all with pointers (and C is > designed to use pointers quite a bit) the AVR falls down badly > compared to the MSP430 -- which has sixteen 16-bit registers, a single > 16-bit address space, and a very elegent, orthogonal instructions set > (rather reminiscent of the PDP-11) where any register can be used as a > pointer, accumulator, index, whatever. > > The old "8-bit" 8080/Z80 and 6800/6502 processors had a 1 or 2 16-bit > registers and a couple 16-bit operations specifically to deal with the > pointer issue. Unfortuantely, the AVR failed to learn from them and > is missing those features -- and it hurts.Strictly speaking, the AVR also has 16-bit registers (X, Y and Z) made from pairs of 8-bit registers, and it has a few 16-bit operations (such as addiw, subiw and movw). But it would have benefited enormously from a forth pointer register, better addressing modes with these registers, and a few extra instructions (such as 16-bit atomic moves between the SP and a pointer register).> >>> The above are all available as single-chip solutions (on-chip SRAM and >>> flash). If you want to use external flash and SDRAM (e.g. to run >>> Linux), then I'd go with: >>> >>> * ARM9 or Cortex-A (Atmel, TI, Motorla, etc.) >> >> Yup. 160K RAM hardly gets you started on an OS which wants lots of >> buffers for ethernet stacks, file system, video and USB drivers. > > I should add that there are a _few_ (very few) Cortex-M parts that > have external busses and include SDRAM controllers (NXP has one > family), so it is possible to do large-memory stuff with Cortex-M.Some Freescale Kinetis (Cortex-M4) parts have external memory buses, IIRC.> But, it's far more common to move up to the Cortex-A or the slightly > older ARM9 families if you need memory measured in MB rather than KB. > >> Beagleboards and their kin are good place to start here. I worked in >> that area for half a year----but I'm really happier down in the >> Cortex and MSP430 environment. >
Reply by ●August 17, 20122012-08-17
David Brown wrote:> "But it would have benefited enormously from > a forth pointer register,"Not you, too! :)> On 17/08/2012 15:17, Grant Edwards wrote:[ ... ]>> The old "8-bit" 8080/Z80 and 6800/6502 processors had a 1 or 2 16-bit >> registers and a couple 16-bit operations specifically to deal with the >> pointer issue. Unfortuantely, the AVR failed to learn from them and >> is missing those features -- and it hurts. > > Strictly speaking, the AVR also has 16-bit registers (X, Y and Z) made > from pairs of 8-bit registers, and it has a few 16-bit operations (such > as addiw, subiw and movw). But it would have benefited enormously from > a forth pointer register, better addressing modes with these registers, > and a few extra instructions (such as 16-bit atomic moves between the SP > and a pointer register).Grant Edwards was more dramatic than I would have been, but the 8-bit roots show in all the doubled-up instructions below. A lot of potential pointer problems go away because gcc has inlined the heck out of the code, and much of the parameter passing has vanished (less obvious here than in other snippets, I guess): unsigned char led, i, w; int mask, val, pwm; // randomize time to next interrupt .. for (i=0, w=0; i < (127/32); i++) // randomly pick a number [31/32 .. 1+1/32) scaled into 8 bits w = w * 2 + random_bit(); a2: 33 0f add r19, r19 unsigned short lfsr = 0xACE1u; static int random_bit () { // taps: 16 14 13 11; characteristic polynomial: x^16 + x^14 + x^13 + x^11 + 1 lfsr = (lfsr >> 1) ^ (-(lfsr & 1u) & 0xB400u); a4: ca 01 movw r24, r20 a6: 96 95 lsr r25 a8: 87 95 ror r24 aa: 44 27 eor r20, r20 ac: 55 27 eor r21, r21 ae: 46 1b sub r20, r22 b0: 57 0b sbc r21, r23 b2: 40 70 andi r20, 0x00 ; 0 b4: 54 7b andi r21, 0xB4 ; 180 b6: 48 27 eor r20, r24 b8: 59 27 eor r21, r25 unsigned char led, i, w; int mask, val, pwm; // randomize time to next interrupt .. for (i=0, w=0; i < (127/32); i++) // randomly pick a number [31/32 .. 1+1/32) scaled into 8 bits w = w * 2 + random_bit(); ba: ba 01 movw r22, r20 bc: 61 70 andi r22, 0x01 ; 1 be: 70 70 andi r23, 0x00 ; 0 c0: 36 0f add r19, r22
Reply by ●August 17, 20122012-08-17
On Fri, 17 Aug 2012 13:17:16 +0000 (UTC), Grant Edwards <invalid@invalid.invalid> wrote:>On 2012-08-17, Mark Borgerson <mborgerson@comcast.net> wrote: > >> I've only looked briefly at the AVR, but I've done a dozen designs with >> various MSP430 variants. > >The main annoyance in the AVR is that it's strictly an 8-bit CPU. It >can't really do any 16-bit operations. That would be fine if it had >an 8 bit address space, but it's got two different 16-bit address >spaces. So, if you need to do anything at all with pointers (and C is >designed to use pointers quite a bit) the AVR falls down badly >compared to the MSP430 -- which has sixteen 16-bit registers, a single >16-bit address space, and a very elegent, orthogonal instructions set >(rather reminiscent of the PDP-11) where any register can be used as a >pointer, accumulator, index, whatever. > >The old "8-bit" 8080/Z80 and 6800/6502 processors had a 1 or 2 16-bit >registers and a couple 16-bit operations specifically to deal with the >pointer issue. Unfortuantely, the AVR failed to learn from them and >is missing those features -- and it hurts.The 6800 had a couple of 16-bit registers (the stack pointer and index register, plus, of course the PC), and you do some limited operations on X (load, store, compare, increment, decrement). On the 6502, OTOH, neither the stack pointer*, or either of the index registers, X and Y, were 16 bit. But 6502 had a few very useful zero-page indirect addressing modes, which allowed using most of the first 256 bytes of memory as a collection of 16-bit index registers. *The 8-bit stack pointer was possibly the single biggest annoyance on the 6502. Or maybe the lack of a zero-page indirect addressing mode that *didn't* also use X or Y�
Reply by ●August 17, 20122012-08-17
On Thursday, August 16, 2012 11:57:44 PM UTC+12, Andre wrote:> Do you know a good and CHEAP programmer for the ATMEL series of MCU ?? > > one that support the whole series but more certainly the mid-range > > including the AT89C55WD ??Do you have a dusty drawer full of AT89C55WD ? If you want low cost, easy to use, the new Atmel AT89LP series are supported by Atmel web programmers. Or, look at SiLabs C8051F5xx series tool sticks, programming AND Debug for around $20-$30. Or the Zilog Z51F series, you can get going for ~$29, again programming AND debug. The parts above are all wide Supply, which is rather harder to find in 32 bit variants. -jg
Reply by ●August 17, 20122012-08-17
On Aug 17, 9:13�pm, Robert Wessel <robertwess...@yahoo.com> wrote:> ... > The 6800 had a couple of 16-bit registers (the stack pointer and index > register, plus, of course the PC), and you do some limited operations > on X (load, store, compare, increment, decrement).And TSX/TXS... whoever has written lots of SWI handlers will remember those well (including the +1/-1 ...). They lived on the HC11, probably on the 12 (whatever they call that today) to this day. The 6809 was a huge step forward but did not make it into the future, it came too late (although I learned the trade mostly on it), the 68k came out and changed the game entirely. Dimiter ------------------------------------------------------ Dimiter Popoff Transgalactic Instruments http://www.tgi-sci.com ------------------------------------------------------ http://www.flickr.com/photos/didi_tgi/sets/72157600228621276/
Reply by ●August 20, 20122012-08-20
On Fri, 17 Aug 2012 13:13:35 -0500, Robert Wessel <robertwessel2@yahoo.com> wrote:>On the 6502, OTOH, neither the stack pointer*, or either of the index >registers, X and Y, were 16 bit. But 6502 had a few very useful >zero-page indirect addressing modes, which allowed using most of the >first 256 bytes of memory as a collection of 16-bit index registers. > >*The 8-bit stack pointer was possibly the single biggest annoyance on >the 6502. Or maybe the lack of a zero-page indirect addressing mode >that *didn't* also use X or Y…Yes. But the 65802 and 65816 did have a full complement of 16-bit registers and operations. The 802 was pin compatible with the 6502. The 802 is no longer available, but the 816 comes in 2 flavors: MPU or SoC. IMO, the MPU version is overpriced, but the SoC version is pretty reasonable for what it does. George
Reply by ●August 20, 20122012-08-20
On Mon, 20 Aug 2012 16:14:38 -0400, George Neuner <gneuner2@comcast.net> wrote:>On Fri, 17 Aug 2012 13:13:35 -0500, Robert Wessel ><robertwessel2@yahoo.com> wrote: > >>On the 6502, OTOH, neither the stack pointer*, or either of the index >>registers, X and Y, were 16 bit. But 6502 had a few very useful >>zero-page indirect addressing modes, which allowed using most of the >>first 256 bytes of memory as a collection of 16-bit index registers. >> >>*The 8-bit stack pointer was possibly the single biggest annoyance on >>the 6502. Or maybe the lack of a zero-page indirect addressing mode >>that *didn't* also use X or Y� > >Yes. But the 65802 and 65816 did have a full complement of 16-bit >registers and operations. The 802 was pin compatible with the 6502. > >The 802 is no longer available, but the 816 comes in 2 flavors: MPU or >SoC. IMO, the MPU version is overpriced, but the SoC version is >pretty reasonable for what it does.While true, it also postdated the 8080/Z80/6800/6502 (which were variously introduced in 1974-1976) by a decade (the 65816 first shipped in 1984). Nor was it particularly successful, certainly not by comparison to the above four devices, or the more contemporary 68K and 8086 families.
