Run commands from the repository root after installation.
Use this bounded interpreter for small integer tools. For functions, containers or large integers, use MicroPython.
from gpu_rv32 import compile_program, run_program
image = compile_program("print(2 * 8)\n")
answer = run_program(image)
assert answer["status"] == answer["error"] == 0
print(answer["stdout"], end="") # 16The guest supports signed 32-bit integers, variables, assignment, expressions,
print(integer), if/else, while, pass, comments and spaces for indentation.
Expressions support + - * // %, comparisons/chains, & | ^ ~ << >>,
and/or/not, and True/False. // and % use Python's floor rule; arithmetic
wraps at 32 bits. Negative shifts fail; shifts of 32 or more produce zero for
left shifts, and zero or -1 for right shifts according to the sign of the input.
This is a bounded language subset: no imports, function definitions, strings,
containers, floats, for/range, break/continue, or augmented assignment.
The limits are 4095 UTF-8 source bytes, 256 nonblank statements, 32 variables,
31-byte names, 16 nested blocks/expression levels, and 4095 stdout bytes.
Guest RAM is 64 KiB, with at least 32 KiB reserved for stack. Every source uses
the same compiled interpreter; only initial source/budget RAM changes.
compile_program(source, statement_budget=10000) returns the fixed ELF,
machine code, initialized RAM, symbols and build hashes. It works without CUDA.
run_program(image, instruction_budget=20000000) returns value, stdout,
guest error/error_name, VM status/status_name, executed steps, GPU kernel
time and host request time. Always check both VM status and guest error.
Guest errors 0..6 mean success, syntax, undefined name, division by zero, capacity/nesting, statement budget and negative shift.
Check VM status separately from guest errors. See quick start for artifacts and timing.