Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix userdata binding corner case #1673

Merged
merged 1 commit into from
Apr 24, 2023
Merged

Fix userdata binding corner case #1673

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions src/liboslexec/oslexec_pvt.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -191,8 +191,10 @@ struct UserDataNeeded {
{
if (a.name != b.name)
return a.name < b.name;
if (a.layer_num != b.layer_num)
return a.layer_num < b.layer_num;
// Checking for layer_num means that if derivs differ find_userdata_index
// may find the wrong layer symbol with the wrong derivs setting.
//if (a.layer_num != b.layer_num)
// return a.layer_num < b.layer_num;
if (a.type.basetype != b.type.basetype)
return a.type.basetype < b.type.basetype;
if (a.type.aggregate != b.type.aggregate)
Expand Down
7 changes: 6 additions & 1 deletion src/liboslexec/shadingsys.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4498,8 +4498,13 @@ osl_bind_interpolated_param(void* sg_, ustring_pod name, long long type,
sg->context->incr_get_userdata_calls();
}
if (status == 2) {
int udata_size = (userdata_has_derivs ? 3 : 1) * TYPEDESC(type).size();
// If userdata was present, copy it to the shader variable
memcpy(symbol_data, userdata_data, symbol_data_size);
memcpy(symbol_data, userdata_data,
std::min(symbol_data_size, udata_size));
if (symbol_data_size > udata_size)
memset((char*)symbol_data + udata_size, 0,
symbol_data_size - udata_size);
return 1;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

wide_shadingsys.cpp __OSL_OP(bind_interpolated_param) could use this same type of fix, although because of the SOA data layout it might be more complicated. In particular can't just copy less bytes. Are there any unit tests to exercise this exact issue?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

With the fixed version of the set<> index, the safeguard in bind_interpolated_param shouldn't be necessary. I added it first to confirm we were getting garbage, then I left it in case the policy changed. No unit test, but with the other fix it won't even fail.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Are you making the statement
OSL_ASSERT(symbol_data_size == udata_size);
?

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Aside:

I would love to eventually work toward as much as possible merging the separate scalar and batch versions of the ShadingSystem and BackendLLVM classes, or making them both inherit from a common base class with most of the functionality that is shared, or making batch a derived class of scalar, or something. Making "batch-ness" be more of a mode or a subclass that has only a few critical methods overloaded/replaced, will help to reduce these recurring issues where the scalar and batch systems have large numbers of identical or very similar methods and patchers (myself included) often only remember to change one of them.

In retrospect, I think it's less problematic how we did the OptiX back end, in which there are a variety of #if USE_OPTIX (for compile time) and if (optixmode) (for runtime) clauses inside the methods that really need them, rather than completely replicating the class hierarchy. That seems to suffer from fewer cases where we make a change to scalar or optix but forget to make the corresponding fix to the other... because for most of it, there are not separate code paths.

Granted, the changes necessary for batch shading are more intrusive and extensive -- the whole Cuda shtick is that you can largely write the code as if it's running on one point at a time, so the Cuda/OptiX code is closer in spirt to the CPU scalar path than the CPU simd path is to the CPU scalar path.

}
return 0; // no such user data
Expand Down