The decoder API is now suitable to be made public. decoder.h is the sole
entry point for the decoder and it has been moved to `include/`
Signed-off-by: Ronald Caesar <github43132@proton.me>
Assign indices in generate_a64_table.py at the very end, right before
generation. This is much safer than maintaining a running counter during
parsing.
Signed-off-by: Ronald Caesar <github43132@proton.me>
Switch to hashing bits [32:21] instead of bits [27:20] and [7:4]. This
massively reduces the average amount of instructions in a bucket.
Signed-off-by: Ronald Caesar <github43132@proton.me>
512 instructions are being declared in every bucket.
Size per bucket:
512 x 8 bytes (ptr) = 4096 bytes
Total Table Size:
4096 buckets x 4096 bytes ≈ 16.7 MB
This will destroy the CPU cache performance.
To fix this we generate small seperate arrays for each bucket that
actually has content and point to them.
Signed-off-by: Ronald Caesar <github43132@proton.me>
generate_a64_table.py only generated instruction variants. For
example, the instruction ADD has forms. The default form with bit 31
set to 0, and its variant with bit 31 set to 1. The script ignored the
default form and added the varient.
Signed-off-by: Ronald Caesar <github43132@proton.me>
- Added default values for input/output directories
- Improve error handling for missing cli arguments
Signed-off-by: Ronald Caesar <github43132@proton.me>
Adds a python script, tools/generate_a64_table.py, to parse ARM's
machine readable XML. The script generates a static C lookup table
containing instruction mnemonics, masks, and values.
Signed-off-by: Ronald Caesar <github43132@proton.me>
