During debugging gpgpu_fill test on various platforms, I found out few things that can affect newer gens: Seting number of threads in TS in gen8_fill_interface_descriptor to 1. This field was omitted in earlier platforms (apparently without any side effects). Not setting it for newer platforms results in GPU hang. Adding pipeline selection mask to gen9_gpgpu_fillfunc, which is necessary from SKL. Changes gen7_emit_interface_descriptor_load to gen8 version. Signed-off-by: Katarzyna Dec <katarzyna.dec@xxxxxxxxx> Cc: Lukasz Kalamarz <lukasz.kalamarz@xxxxxxxxx> Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@xxxxxxxxx> --- lib/gpgpu_fill.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/lib/gpgpu_fill.c b/lib/gpgpu_fill.c index 4d98643d..e7a1edaa 100644 --- a/lib/gpgpu_fill.c +++ b/lib/gpgpu_fill.c @@ -336,6 +336,8 @@ gen8_fill_interface_descriptor(struct intel_batchbuffer *batch, struct igt_buf * idd->desc5.constant_urb_entry_read_offset = 0; idd->desc5.constant_urb_entry_read_length = 1; /* grf 1 */ + idd->desc6.num_threads_in_tg = 1; + return offset; } @@ -790,12 +792,13 @@ gen9_gpgpu_fillfunc(struct intel_batchbuffer *batch, batch->ptr = batch->buffer; /* GPGPU pipeline */ - OUT_BATCH(GEN7_PIPELINE_SELECT | PIPELINE_SELECT_GPGPU); + OUT_BATCH(GEN7_PIPELINE_SELECT | GEN9_PIPELINE_SELECTION_MASK | + PIPELINE_SELECT_GPGPU); gen9_emit_state_base_address(batch); gen8_emit_vfe_state_gpgpu(batch); gen7_emit_curbe_load(batch, curbe_buffer); - gen7_emit_interface_descriptor_load(batch, interface_descriptor); + gen8_emit_interface_descriptor_load(batch, interface_descriptor); gen8_emit_gpgpu_walk(batch, x, y, width, height); OUT_BATCH(MI_BATCH_BUFFER_END); -- 2.14.3 _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
