The struct padding got broken by
c789b981b244333cfc903bcd1e2fefc010500013.
This caused serious performance regression because part of the key was
uninitialized and hence the shader always recompiled (at least on release
builds...).
While here also fix key size calculation when the number of samplers
and the number of sampler views are different.
v2: add comment
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
key->clip_halfz = !llvm->draw->rasterizer->gl_rasterization_rules;
key->need_edgeflags = (llvm->draw->vs.edgeflag_output ? TRUE : FALSE);
key->ucp_enable = llvm->draw->rasterizer->clip_plane_enable;
- key->pad = 0;
+ key->pad1 = 0;
+ key->pad2 = 0;
/* All variants of this shader will have the same value for
* nr_samplers. Not yet trying to compact away holes in the
unsigned clip_halfz:1;
unsigned bypass_viewport:1;
unsigned need_edgeflags:1;
+ /*
+ * it is important there are no holes in this struct
+ * (and all padding gets zeroed).
+ */
+ unsigned pad1:1;
unsigned ucp_enable:PIPE_MAX_CLIP_PLANES;
- unsigned pad:33-PIPE_MAX_CLIP_PLANES;
+ unsigned pad2:32-PIPE_MAX_CLIP_PLANES;
/* Variable number of vertex elements:
*/
vs->variant_key_size =
draw_llvm_variant_key_size(
- vs->base.info.file_max[TGSI_FILE_INPUT]+1,
- vs->base.info.file_max[TGSI_FILE_SAMPLER]+1);
+ vs->base.info.file_max[TGSI_FILE_INPUT]+1,
+ MAX2(vs->base.info.file_max[TGSI_FILE_SAMPLER]+1,
+ vs->base.info.file_max[TGSI_FILE_SAMPLER_VIEW]+1));
vs->base.state.stream_output = state->stream_output;
vs->base.draw = draw;