New in 20020728: There are quite a few new features in this release — probably because it's been over three years since the last release:

• Split I&D executables (JMAGIC) are now supported, allowing up to twice as much code and data in executable files as was possible previously. This executable format keeps code and data in two separate 16-bit address spaces. The linker 'ld' generates split I&D executables when given the '-j' option.

• The assembler 'as' now supports multi-word "aliased" instructions which automatically assemble into the proper sequence of single-word instructions:

  • The data movement and long jump/call instructions now appear to accept immediate address arguments. Internally, the assembler generates code to use the reserved %t6 register to load the immediate address, and then generates the indirect form of the original instruction using that register. This means you can replace three lines of assembly code with one, and improve readability dramatically.

    Opcodes with aliased forms come from the load/store word family (mov/ldi, mov/ld and mov/st), the load/store byte family (movb/ldb and movb/stb), and the long jump and procedure call/return family (jal, j and call). This means you can now write simply call _printf and the assembler will worry about generating the three machine instructions necessary to load the two 8-bit halves of the value _printf into a register and then make an indirect linked jump using that register.

  • There's a new family of stack pseudo-instructions — push and pop — for saving and restoring register values.

  • Now only two CPP macros are used for assembling the included SMOKE-16 programs: FUNCTION(name) and RET. They implement the prologue and epilogue for procedures that need the ability to call and be called by other procedures obeying the SMOKE-16 calling convention. Note: The prologue macro creates a global symbol for the procedure name.

• The assembler also supports a .rdata pseudo-section for read-only data, and allows mapping it to either the data section or the text section. Placing it in the data section is a good idea for split I&D executables; placing it in the text section using the '-R' option means it's shareable for other types of executables.

• The linker 'ld' now supports object archives in a modified a.out file format (LMAGIC). There's a new command 'nm' to list the contents of SMOKE-16 object files and object archives. Support for portable 'ar' archives is still to come, and there's no archive utility 'ar' yet. Instead, object archives are created using a special '-Z' argument to the linker. Naturally the linker now supports an archive search path using the standard '-L' argument, and linking of libraries called libx.a using the standard '-lx' argument. The SMOKE-16 standard library 'libc' is now an object archive, not a monolithic object file as it was in previous releases, so executables linked with it grow only enough to accomodate functionality they actually use.

• The emulator 'emu' has been improved greatly, especially the disassembler. It now displays correct immediate word values for sethi/movb sequences, and in many cases is able to display a useful symbolic name as well. A syntax-coloring filter script 'emucolor' is also included for those wishing to display the disassembly in technicolor. Usage in GNU bash: bin/emu -disassemble image 2>&1 | ./emucolor

Portability: Symbols and archive member filenames are now portable across platforms and character sets (they are now restricted to printable ASCII and a few control characters.) Although the emulator 'emu' is still restricted to the host machine's native character encoding for the system-call interface, the assembler 'as' can be used to produce ASCII executables even on non-ASCII systems using the '-portable' argument. The two remaining portability problems problems are the strings passed through the emulator system-call interface and the set of errno values.