> comment #2
> I've changed the LPM_R0_ZPLUS_* macros in common/macros.inc to look for AVR_HAVE_ELPM instead of BIG_CODE.
>My guess is that there are still problems because of finding the definition of AVR_RAMPZ_ADDR.

RAMPZ register address is defined based on RAMPZ macro in device header file. Building library for specific arch will result in error as AVR_RAMPZ_ADDR is not defined.

(include/avr/common.h)
284 #if defined(RAMPZ)
285 #  define AVR_RAMPZ_REG   RAMPZ
286 #  if _AVR_ARCH_ >= 100
287 #    define AVR_RAMPZ_ADDR  _SFR_MEM_ADDR(RAMPZ)
288 #  else
289 #    define AVR_RAMPZ_ADDR  _SFR_IO_ADDR(RAMPZ)
290 #  endif
291 #endif

It is possible to hard-code RAMPZ register address irrespective of device header file (gcc also hard-codes RAMPZ address) as follows:

+#  define AVR_RAMPZ_REG   (_SFR_IO8(0x3b))
+#  define AVR_RAMPZ_ADDR  (_SFR_IO8(0x3b))

This will enable building library for a architecture than a device.
--
This fix shall be included until we have per device library build system - Joerg Wunsch (http://lists.nongnu.org/archive/html/avr-libc-dev/2013-10/msg00002.html)

I've changed the LPM_R0_ZPLUS_* macros in common/macros.inc to look for _AVR_HAVE_ELPM_ instead of BIG_CODE.

My guess is that there are still problems because of finding the definition of AVR_RAMPZ_ADDR.

To really fix this, avr-libc needs to build a library-per-device.

It looks like that the pmstring code uses the LPM_R0_ZPLUS_* macros von common/macros.inc, which will only use the 'elpm' if _AVR_ENHANCED_ is set. Unfortunately this is no longer used??

Therefore memcmp_PF etc. don't work with atmega1284p or probably other 128k devices.

As I'm not familiar with the setting of flags for the family avr51 I have no idea for a path