Back in 2004 and 2005, two features were added to GNU binutils for msp430: -mP : Enables polymorphs -mQ : Enables relaxation
As best I can reconstruct it: Polymorphs are extended jump instructions like "beq" which have a constant size regardless of whether the target is within a 1KB code window allowing use of the PC-relative short jump instructions (je #target), or is farther away and must be reached with a jump-around sequence (jne 1f; br #target; 1:). If the jump is found to be in range when relocations are resolved at link time, the short form is used and the remainder of the space filled with nops. Relaxation is a separate but supportive operation that detects situations where the shorter jump is selected at link time and attempts to move objects to eliminate the nops. Neither flag is enabled by default. There are comments in the binutils code indicating that these flags are disabled because they are dangerous and don't work. Contemporaneous email on the binutils mailing list suggests certain problems were never resolved. I have created an assembly-language test case that confirms that polymorphs will generate incorrect jumps. gcc does not currently emit any polymorph jump instructions: it attempts to determine the jump distance and select the appropriate instruction. All this leaves me relatively convinced that both features are candidates for removal. I'm guessing that the only use of either of these features is in historical projects which have hand-written assembly code that uses the polymorph jump instructions between functions. If there is anybody out there who actively uses either the -mP or -mQ flags to any msp430 binutils programs, or gets polymorphs or relaxation to work in some other way, please respond and explain what you're doing. At this time, it is highly likely that these features will be removed during a binutils validation phase in preparation for object file enhancements that will improve/support 20-bit code and data operations. The fewer non-functional/untested "features" I have to work around the better the functional ones will be. Response to this email will determine whether I keep and work around them, or remove them (and if so, whether I ever put them back). So if you care about polymorphs, speak up. (It is more likely that relaxation will return in a future optimization that supports link-time assignment of objects to near (16-bit) or far (20-bit) addresses, but AFAICT what's there won't help with that, and we need to get the basic functionality right first.) Peter ------------------------------------------------------------------------------ Write once. Port to many. Get the SDK and tools to simplify cross-platform app development. Create new or port existing apps to sell to consumers worldwide. Explore the Intel AppUpSM program developer opportunity. appdeveloper.intel.com/join http://p.sf.net/sfu/intel-appdev _______________________________________________ Mspgcc-users mailing list Mspgcc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mspgcc-users
