> + /* > + * Note: the first call is here to lower RECT and TXP, the second one > + * to lower the TXS(lod) instructions generated by the RECT lowering > + * done in the first pass. > + * > + * FIXME: we should probably have a > + * > + * do nir_lower_tex() while (progress) > + * > + * loop here, but it looks like nir_lower_tex() is not ready for that > + * (it keeps returning true and lowering the same instructions over > and > + * over again). > + */ > + NIR_PASS(progress, nir, nir_lower_tex, &lower_tex_options); > + lower_tex_options.lower_rect = false; > + lower_tex_options.lower_txp = 0; > NIR_PASS(progress, nir, nir_lower_tex, &lower_tex_options);
Hmm, maybe we could just have a second indepedent lower_tex_options to do just the lower_txs_lod lowering? No progress loop needed, no complex explanation needed. It's a little strange, but it's okay :) > + case nir_texop_txs: > + emit_sysval_read(ctx, &instr->instr); > + return; See op switch comment. --- Everything later in the patch looks great!
signature.asc
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
