[PATCH vkd3d 1/2] tests: Fix shader tests crashes on WARP driver.

[Nikolay Sivov]

On 12/15/21 20:57, Henri Verbeet wrote:
> On Sat, 11 Dec 2021 at 20:44, Nikolay Sivov <nsivov at codeweavers.com> wrote:
>> @@ -419,9 +418,11 @@ static void parse_test_directive(struct shader_context *context, const char *lin
>>           hr = create_root_signature(context->c.device, &root_signature_desc, &context->c.root_signature);
>>           ok(hr == S_OK, "Failed to create root signature, hr %#x.\n", hr);
>>
>> -        pso = create_pipeline_state(context->c.device, context->c.root_signature,
>> +        if (context->c.pso)
>> +            ID3D12PipelineState_Release(context->c.pso);
>> +        context->c.pso = create_pipeline_state(context->c.device, context->c.root_signature,
>>                   context->c.render_target_desc.Format, NULL, &ps, NULL);
>> -        if (!pso)
>> +        if (!context->c.pso)
>>               return;
>>
> I suppose that's an improvement in the sense that it should work in
> practice for most of our existing tests, but it doesn't strike me as
> entirely correct either. In particular, the restriction is that we
> can't destroy the state object until all command lists referencing it
> have finished executing on the GPU. That mostly works out in practice
> because the typical pattern is to do a "draw", followed by a "probe",
> in which case the "probe" will wait for the command list to finish
> executing, but I could certainly imagine writing tests where we do
> multiple draws before probing.
Would it be correct in terms of test results to flush current command 
list, on repeated "draw"s? I see there is ::Reset() that according to 
docs will reinitialize recording, if last Close() succeeded. Test could 
use "pso != NULL" as a marker that "probe" wasn't used since last draw, 
and close-exec-wait-reset right before starting new recording.

Another brute force option is to simply allow for arbitrary number of 
state objects in test context, accumulated indefinitely until next "probe".