when would you use static and const?

[Pommie]

In pics all constants are stored in rom as the only sensible way to use them is with a movlw # or addlw # etc. Using define or const will still produce the same code. Strings stored as constants are compiled to retlw tables or if you are using asm then the DT "Hello World!" will be stored as a series of retlw instructions.

Mike.

 

[Pommie]

Hi guys,

I'm trying to understand the PIC18F block diagram (specifically, i have PIC18F4520).

Since the PIC18F is 8-bit, i.e. the address bus and data bus are 8-bit, wouldn't you expect the Program Coutner register to be 8-bit as well?

in the Block Diagram, it appears to be 20-bit.

View attachment 83272

An 8 bit computer has a 8 bit data bus only, the address bus can be any width. BTW, in the above diagram the address bus is 12 bits wide.

Mike.

 

[3v0]

Think about it. Would you be happy with an 8bit PC on an 18F?

 

[Pommie]

Think about it. Would you be happy with an 8bit PC on an 18F?

Having only 256 instructions could be a little restrictive.

Mike.

 

[NorthGuy]

In pics all constants are stored in rom as the only sensible way to use them is with a movlw # or addlw # etc. Using define or const will still produce the same code. Strings stored as constants are compiled to retlw tables or if you are using asm then the DT "Hello World!" will be stored as a series of retlw instructions.

On 16-bit PICs, the linker has a provision for a RAM data section which is then getting initalized from ROM at startup, so the same thing is stored twice. I suspect that C compiler could be using this mechanism.

They have a direct mechanism which allows parts of the program memory to be mapped into the data space, but since the program memory is usually bigger it cannot be mapped all at once. I guess that is why direct access may not be implemented in C.

I use Assembler and I create a constant section at the beginning of the program memory which I use for constants that don't change. I then map it into the data memory space and I never change this mapping. As a result, it is accessible from anywhere as if it was in the data memory, but read-only. The only drawback is that you need to use an ugly #psvoffset prefix for all such variables, but this could be avoided with better Assembler compiler.

 

[3v0]

I think that most C compilers including C18 has a provision in the c0 startup code to initialize ram variables from rom where needed. Just saying.

 

[Pommie]

On 16-bit PICs,

Sorry, I should have made it clear above that I was referring to 16 series (8 bit) pics. I would assume that 16 bit compilers keep constants in ROM and use the same mechanism as yourself for efficiency. However, would it be more efficient to use an immediate addressing mode (like movlw) to move a constant into a register?

Mike.

 

[NorthGuy]

Sorry, I should have made it clear above that I was referring to 16 series (8 bit) pics. I would assume that 16 bit compilers keep constants in ROM and use the same mechanism as yourself for efficiency. However, would it be more efficient to use an immediate addressing mode (like movlw) to move a constant into a register?

On 16-bit PICs, instructions reading from the mapped memory take about 3 times longer than instructions dealing with regular memory. However, you do not really read constants from the memory very often, so I don't think it's a big deal.

The organization of their program memory is very funny. Even though every instruction uses 3 bytes, these 3 bytes are spaced by 2 in the program memory space. That is first 3-byte instruction is at address 0x200, second 3-byte instruction is at address 0x202, next one is at address 0x204 etc. When you buy a chip with 64K of memory you get 64K/3 = 22016 instructions with the last address being 0x5800.

When the program memory is mapped into the data space, only 2 of the 3 instruction bytes are visible, while the 3-rd byte is hidden. When you put your constants into program memory, the data go into the two visible bytes and the third byte is set to 0. When the hidden byte is zero, the instruction becomes "NOP", so you can safely execute through the data located in the program memory without doing much harm.

Mapping is very useful. For example, you may have a routine which prints text on LCD. You can pass a pointer to a regular data memory location (e.g. decimal number) or you can pass a pointer to a piece of program memory mapped to data memory (e.g. constatnt text). It won't make any difference except that the mapped memory cannot be written.

 

[3v0]

Thanks northguy I have not done much with the 16 bit pics but they are very attractive. I was thinking there should be a like button.

 

[heydonms]

In pics all constants are stored in rom as the only sensible way to use them is with a movlw # or addlw # etc. Using define or const will still produce the same code.

That makes sense, except for handling pointers. You can have a pointer to a const but you can't use FSR and INDF to reference program memory. If the compiler was smart enough and/or the code simple enough it might be possible to replace any pointer based code with literals but I'm not sure that is always going to work.

 

[electroRF]

Hi dear friends.

I thank you very very much!

I learned a lot of this thread and i'm going to do (keep doing) a lot of reading.

Especially on Memory mapping which is new to me and you discussed this term.

Thanks a lot