[maleadt]

On devices without Float64 support, method_tables adds float64_overrides, changing the method table type GPUCompiler's interpreter specializes on. The precompile workload covered only supports_fp64 = true. On devices such as Arc A-series GPUs, the first kernel compilation therefore inferred and compiled the interpreter from scratch: typeinf_local alone took 7.4 s. Precompile both settings to cover these devices.

Arc A770, Julia 1.12.7, median of 3 fresh processes:

	main	PR
first @oneapi launch in a session	8.61 s	2.41 s
first KernelAbstractions launch in a session	8.84 s	2.63 s
precompiling oneAPI.jl	34.6 s	41.4 s

Devices with Float64 support remain covered. The workload still compiles only on Julia 1.12+.

The full test suite passes on an Arc A770 (Julia 1.12.7).

[codecov]

Codecov Report

❌ Patch coverage is 87.50000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 86.63%. Comparing base (446f8ac) to head (80c4df5).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
src/compiler/precompile.jl	87.50%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #746      +/-   ##
==========================================
- Coverage   86.89%   86.63%   -0.27%     
==========================================
  Files          55       55              
  Lines        5090     4997      -93     
==========================================
- Hits         4423     4329      -94     
- Misses        667      668       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.