| Summary: | radv_shader_compile_to_nir doesn't check for bad entry point | ||
|---|---|---|---|
| Product: | Mesa | Reporter: | Dave Gilbert <freedesktop> |
| Component: | Drivers/Vulkan/radeon | Assignee: | mesa-dev |
| Status: | RESOLVED WONTFIX | QA Contact: | mesa-dev |
| Severity: | normal | ||
| Priority: | medium | ||
| Version: | unspecified | ||
| Hardware: | Other | ||
| OS: | All | ||
| Whiteboard: | |||
| i915 platform: | i915 features: | ||
Should be fixed with https://patchwork.freedesktop.org/patch/249829/ That's not the job of the driver. You should be running with validation layers enabled while developing applications. Actually, I do agree with Jason's point of view. We are not going to fix anything inside the driver. |
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.
I screwed up my vulkan calling and ended up getting radv_shader_compile_to_nir to seg because it has: 251 entry_point = spirv_to_nir(spirv, module->size / 4, spec_entries, num_spec_entries, stage, entrypoint_name, &spirv_options, &nir_options); 252 nir = entry_point->shader; assert(nir->info.stage == stage); it's dereferencing 'entry_point' on 252, but if the caller screws up entry_point can be NULL from spirv_to_nir; Could this failure be made cleaner? (Nothing else printed anything first before the seg by default; spirv_to_nir took a bit of an odd route; it checks b->entry_point = NULL then call vrn_fail that jumps back to it's setjmp handler and cleans up - so it's not obvious to me why it also has a ralloc_free(b), return NULL if the vtn_fail is going to take the setjmp). My screwup incidentally was a cut-and-paste where I'd copied the code for my vertex shader for my fragment shader and so I was trying to find the fragment shader entrypoint in my vertex shader.