a native SASS backend, and ggml-cuda lowers (#174) #103

	name: Numeric Regression (SLATEC)

	on:
	push:
	branches: [master]
	pull_request:
	branches: [master]

	jobs:
	numeric:
	runs-on: ubuntu-latest
	defaults:
	run:
	shell: bash -el {0} # login shell so the conda env is active
	steps:
	- uses: actions/checkout@v4

	- name: Build Booth
	run: make

	- name: Setup tinygrad emulator
	run: \|
	# The correct gfx1200 emulation lives on our tinygrad fork; upstream's
	# amd mockgpu gets rdna4 wrong (rdna3 is fine on both). booth-emu is a
	# frozen tag at the verified commit; the fork's branch tip has since
	# drifted and gets gfx1200 wrong too, hence a tag not a branch.
	git clone --branch booth-emu --depth 1 https://github.com/Zaneham/tinygrad.git /tmp/tinygrad
	pip install -e /tmp/tinygrad

	- name: Install LFortran
	uses: mamba-org/setup-micromamba@v1
	with:
	environment-name: lf
	create-args: >-
	python=3.11
	lfortran=0.65.0

	- name: Numeric regression vs baseline
	env:
	KATH: ./kath
	LFORTRAN: lfortran
	TINYGRAD_PATH: /tmp/tinygrad
	run: python tests/numeric/run_numeric.py --backends cpu,rdna3,rdna4

Navigation Menu