If you do this: Y_in_RAM = *&Y_in_ROM; The compiler correctly interprets that dereferencing the reference to Y_in_ROM is unnecessary, so it creates the machine code as if the instruction were "Y_in_RAM = Y_in_ROM;". Such as: mov.w 0xf03e,w0 mov.w w0,0x093e Although, in my case, Y_in_ROM is not a named variable. The value I want is somewhere in a 1K swath of ROM that is copied out to an external EEPROM chip on power up if a certain signature is not found in the EEPROM chip. This typically only happens once the first time the board is powered up at the CM's factory. Subsequently the user may alter some EEPROM values to customize operation of the machine. Such as changing the operation of the clock from a 12-hour day to a 24-hour day. The entire EEPROM contents are also mirrored into RAM so all reads of EEPROM data may be read from RAM to avoid the I2C time delay that would be involved if reads of EEPROM data had to always come from the external EEPROM chip. All writes to EEPROM are handled by a function that immediately mirrors the written data back into RAM so the RAM image is always a full and complete copy of what is in EEPROM. All variables in this EE_RAM area are named so they may be referenced by the code. The only reference to the EE_ROM data is the name EE_ROM_Base at the beginning of the 1K of ROM at address 0xF000. I also have a reference to the base of the EEPROM shadow RAM, named EE_RAM_Base at address 0x900. And of course I have a reference to the named variable Y_in_RAM at lets say address 0x93E. Now, if the user hits the button to return the machine back to the factory fresh state, some but not all of the values in ROM must be copied back out to the EEPROM chip. So, what I need to do is copy lets say the clock selection data from ROM at address 0xF03E to the variable Y_in_RAM. So I try this: Y-in_RAM = *(&EE_ROM_Base - &EE_RAM_Base + &Y-in_RAM); But, instead of the compiler understanding that it knows all of the relevant information at compile time which should essentially produce the same machine code as above it creates machine code that includes doing all of the math in registers to create a pointer and then sets Y-in_RAM to what is pointed to, all at runtime. So here is the question, is there a syntax that will clue the compiler to do the pointer math at compile time to simply create machine language similar to: mov.w 0xf03e,w0 mov.w w0,0x093e Can the compiler be induced to understand that the addresses &EE_ROM_Base (0xF000), &EE_RAM_Base (0x900) and &Y_in_RAM (0x93E) are all constant so that 0xF000 - 0x900 + 0x93E = the constant 0xF03E. This should reduce to "Y_in_RAM = *(&0xF03E)"; Which should reduce to "Y_in_RAM = *&0xF03E"; Which should produce the machine code: mov.w 0xf03e,w0 mov.w w0,0x093e But for some reason, any syntax I try, always produces the machine code: mov.w #0xf000,w0 mov.w #0x93e,w1 add.w w0,w1,w0 mov.w #0x900,w1 sub.w w0,w1,w0 mov.w [w0],w0 mov.w w0,0x093e It is doing all of the pointer math at run time even though all values are constant and known at compile time. -- View this message in context: http://old.nabble.com/Cause-GCC-to-do-address-math-at-compile-time-tp33115582p33115582.html Sent from the gcc - Help mailing list archive at Nabble.com.
