<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Thu, Apr 6, 2017 at 1:58 AM, Samuel Iglesias Gonsálvez <<a href="mailto:siglesias@igalia.com" target="_blank">siglesias@igalia.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">On Broadwell we still need to do a resolve between the subpass that writes and the subpass that reads when there is a self-dependency because HW could not see fast-clears and works on the render cache as if there was regular non-fast-clear surface. Fixes 16 tests on BDW: dEQP-VK.renderpass.formats.*.input.clear.store.self_dep* Signed-off-by: Samuel Iglesias Gonsálvez <<a href="mailto:siglesias@igalia.com">siglesias@igalia.com</a>> --- src/intel/vulkan/anv_blorp.c | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-) diff --git a/src/intel/vulkan/anv_blorp.c b/src/intel/vulkan/anv_blorp.c index 72a468a744a..daf13d360c4 100644 --- a/src/intel/vulkan/anv_blorp.c +++ b/src/intel/vulkan/anv_blorp.c @@ -1457,6 +1457,7 @@ ccs_resolve_attachment(struct anv_cmd_buffer *cmd_buffer, * still hot in the cache. */ bool found_draw = false; + bool self_dep = false; enum anv_subpass_usage usage = 0; for (uint32_t s = subpass_idx + 1; s < pass->subpass_count; s++) { usage |= pass->attachments[att].subpass_usage[s]; @@ -1466,6 +1467,8 @@ ccs_resolve_attachment(struct anv_cmd_buffer *cmd_buffer, * wait to resolve until then. */ found_draw = true; + if (pass->attachments[att].subpass_usage[s] & ANV_SUBPASS_USAGE_INPUT) + self_dep = true; break; } } @@ -1509,10 +1512,19 @@ ccs_resolve_attachment(struct anv_cmd_buffer *cmd_buffer, */ } } else if (usage & ANV_SUBPASS_USAGE_INPUT) { - /* Input attachments are clear-color aware so, at least on Sky Lake, we + /* Input attachments are clear-color aware, at least on Sky Lake, so we * can frequently sample from them with no resolves at all. + * + * However on Broadwell we still need to do resolves when there is a + * self-dependency because HW could not see fast-clears and works + * on the render cache as if there was regular non-fast-clear surface. + * To avoid any inconsistency, we force the resolve. */ - if (att_state->aux_usage != att_state->input_aux_usage) { + if (cmd_buffer->device->info.gen == 8 && self_dep && + att_state->aux_usage == ISL_AUX_USAGE_CCS_D && + att_state->input_aux_usage == ISL_AUX_USAGE_CCS_D) { + resolve_op = BLORP_FAST_CLEAR_OP_RESOLVE_FULL; </blockquote><div> Would it make a tad bit more sense to move this down in the if-ladder a bit? It would let us simplify the condition because we would already know that aux_usage == input_aux_usage but I think it's also the most special of our special cases. </div><div>Either way, </div><div>Reviewed-by: Jason Ekstrand <<a href="mailto:jason@jlekstrand.net">jason@jlekstrand.net</a>> </div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> + } else if (att_state->aux_usage != att_state->input_aux_usage) { assert(att_state->input_aux_usage == ISL_AUX_USAGE_NONE); resolve_op = BLORP_FAST_CLEAR_OP_RESOLVE_FULL; } else if (!att_state->clear_color_is_zero_one) { -- 2.11.0 </blockquote></div> </div></div>