Example models where asserts are actually defined/used?
Closed this issue · 6 comments
I'm having trouble understanding how this test suite is actually used in practice. All the examples I can find (including from spike and sail) define the assert macros (like RVMODEL_IO_ASSERT_GPR_EQ) as empty, meaning the tests will never fail! If that is the case, what are you actually testing?
I also don't understand how the F/D tests are supposed to work. For example, in the fadd tests, it looks like correctval is always being set to 0, meaning asserts wouldn't work even if defined! Where is the expected answer supposed to come from?
Yes, I'm trying to debug problems with my model -- isn't that what this test suite is for?
And how can I ignore the correctval for FP tests? What is it actually testing if there's no expected result?
Hi @stkaplan,
I'm currently trying to do this for my processor. No idea if this is an efficient or correct approach, but might help you get on the right track: https://github.com/black-parrot-sdk/bp-riscv-arch-test/blob/master/blackparrot/env/model_test.h
Assertions: use the scratch register to form a small stack. Implement a minimal comparison and print out a != b or a == b.
Correctness: after test completion, copy the entire signature out to a file (I use a testbench hook to print). Offline, compare the example signature from Sail model with the DUT signature.
I see. I assumed these were supposed to work like the tests from https://github.com/riscv-software-src/riscv-tests -- exiting with error on failures. I didn't understand the whole test signature part until digging in further after you brought it up. So this is mainly designed to run in the RISCOF framework, comparing results of my model against Sail or Spike?
As a side note, what is the relationship between this repo and https://github.com/riscv-software-src/riscv-tests? Is this going to be replacing that one? Both seem to be actively updated, but the tests in this repo seem more comprehensive, supported more extensions than riscv-tests.