Skip to content

r300/ci: uninitialized reads in fragment shaders lead to flakes

Some R400 failures have showed up without any apparent reason, seems like uninitalized read problem (in FS).

https://gitlab.freedesktop.org/mesa/mesa/-/jobs/65354454

The spec@glsl-1.10@execution@fs-frontfacing-ternary-vec4-neg-1.0-1.0,Fail reads uninitialized gl_FrontFacing variable (we don't support it for <R500 currently, the current code implies the older HW just can't do it) and thus depends on the specific content of the uninitialized variable. We should probably do similar thing like we did recently for gl_ClipVertex in cc14d402 and replace the read with undef early so we don't have flakes.

Similarly for the rest

  spec@glsl-1.10@execution@variable-indexing@vs-varying-array-mat3-index-rd,UnexpectedImprovement(Pass)
  spec@glsl-1.10@execution@variable-indexing@vs-varying-array-mat3-index-row-rd,Fail
  spec@glsl-1.20@execution@variable-indexing@vs-varying-array-mat3-col-rd,Fail
  spec@glsl-1.20@execution@variable-indexing@vs-varying-array-mat3-col-row-rd,Fail
  spec@glsl-1.20@execution@variable-indexing@vs-varying-array-mat3-index-rd,Fail

here we run out of the varying slots, thus some reads are uninitialized as well. Same solution as above, although here it might be more tricky to figure out early which slots will be actually unused.

Edited by Pavel Ondračka
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information