dasm-a64 is the ARM64 port of DynASM, the awesome dynamic assembler of luajit. All the instructions for AArch64 execution state of ARMv8-A architecture profile are supported.

Comparing with the other ports of DynASM, A relative stupid encoding schema is used: most encoding information is specified in the instruction templates, the parse and encoding engine is relatively simple. Only very base functions are provided in the encoding engine, the templates use these base functions to specify the encoding for each instruction. So there may be many different templates for one instructions(e.g. more than 100 templates for ld1 instruction).

DynASM does not provide any library or executable, it would be used as the preprocessor of any JIT compiler or dynamic assembler.

I do not have access to a real arm64 device now, so dasm-a64 is tested on the virtual machine provided by Foundation Model from ARM. The test environment is provided by linaro

The lamp image is used for the simulation environment(there's arm64 gcc and gdb in this image):

Foundation_v8pkg/models/Linux64_GCC-4.1/Foundation_v8 \
--cores=4 \
--no-secure-memory \
--visualization \
--gicv3 \
--data=fvp_bl1.bin@0x0 \
--data=fvp_fip.bin@0x8000000 \
--network nat \
--block-device=vexpress64-openembedded_lamp-armv8-gcc-4.9_20140923-688.img

A simple brainfxck virtual machine is provided in the test directory. This is an ARM64 port from Peter Cawley's tutorial of DynASM

$ cd test/bf $ make test

The result looks like this:

root@genericarmv8:~/dasm-a64/test/bf# make
normal bf interpreter running...
    AAAAAAAAAAAAAAAABBBBBBB....
    ....
real    22m 13.93s
user    22m 13.69s
sys     0m 0.00s

jit running...
    AAAAAAAAAAAAAAAABBBBBB.....
    ....
real    1m 51.86s
user    1m 51.84s
sys     0m 0.00s
root@genericarmv8:~/dasm-a64/test/bf#

Peter Cawley provides an unoffical but great DynASM tutorial here: http://corsix.github.io/dynasm-doc/tutorial.html