<div dir="ltr"> <div class="gmail_extra"> <div class="gmail_quote">On Wed, Aug 31, 2016 at 8:29 PM, Nanley Chery <<a href="mailto:nanleychery@gmail.com" target="_blank">nanleychery@gmail.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Provides an FPS increase of ~30% on the Sascha triangle and multisampling demos. Clears that happen within a render pass via vkCmdClearAttachments are safe even if the clear color changes. This is because the meta implementation does not use LOAD_OP_CLEAR which avoids any conflicts with 3DSTATE_CLEAR_PARAMS. Signed-off-by: Nanley Chery <<a href="mailto:nanley.g.chery@intel.com">nanley.g.chery@intel.com</a>> --- src/intel/vulkan/anv_pass.c | 11 +++++++++++ src/intel/vulkan/gen8_cmd_buffer.c | 6 ++++++ src/intel/vulkan/genX_cmd_buffer.c | 4 +--- 3 files changed, 18 insertions(+), 3 deletions(-) diff --git a/src/intel/vulkan/anv_pass.c b/src/intel/vulkan/anv_pass.c index 69c3c7e..823f9cf 100644 --- a/src/intel/vulkan/anv_pass.c +++ b/src/intel/vulkan/anv_pass.c @@ -155,5 +155,16 @@ void anv_GetRenderAreaGranularity( VkRenderPass renderPass, VkExtent2D* pGranularity) { + ANV_FROM_HANDLE(anv_render_pass, pass, renderPass); + + /* This granularity is needed for HiZ fast clears */ + for (unsigned i = 0; i < pass->subpass_count; ++i) { + if (pass->subpasses[i].depth_stencil_attachment != + VK_ATTACHMENT_UNUSED) { + *pGranularity = (VkExtent2D) { 8, 4 }; + return; + } + } </blockquote><div> </div>Thanks for remembering this! As mentioned in an earlier e-mail, I'm not sure this is needed for Sky Lake or Broadwell with a format other than D16_UNORM but setting it all the time doesn't hurt. 8x4 isn't onerous and once we get fast clears, it'll get much bigger anyway. <div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> + *pGranularity = (VkExtent2D) { 1, 1 }; } diff --git a/src/intel/vulkan/gen8_cmd_buffer.c b/src/intel/vulkan/gen8_cmd_buffer.c index 7f65fe2..ec91ecd 100644 --- a/src/intel/vulkan/gen8_cmd_buffer.c +++ b/src/intel/vulkan/gen8_cmd_buffer.c @@ -451,6 +451,12 @@ genX(cmd_buffer_do_hz_op)(struct anv_cmd_buffer *cmd_buffer, enum anv_hz_op op) cmd_state->render_area.extent.height % align_h) return; } + + anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CLEAR_PARAMS), cp) { + cp.DepthClearValueValid = true; + cp.DepthClearValue = + cmd_buffer->state.attachments[ds].clear_value.depthStencil.depth; + } </blockquote><div> </div><div>Hrm... I'm not sure where the best place to set CLEAR_PARAMS is. It might almost be better in BeginSubpass... In any case, I think this works. We can move it later if we want. </div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> break; case ANV_HZ_OP_DEPTH_RESOLVE: if (cmd_buffer->state.pass->attachments[ds].store_op != diff --git a/src/intel/vulkan/genX_cmd_buffer.c b/src/intel/vulkan/genX_cmd_buffer.c index 349d2a4..7d2a6bd 100644 --- a/src/intel/vulkan/genX_cmd_buffer.c +++ b/src/intel/vulkan/genX_cmd_buffer.c @@ -1159,9 +1159,6 @@ cmd_buffer_emit_depth_stencil(struct anv_cmd_buffer *cmd_buffer) } else { anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_STENCIL_BUFFER), sb); } - - /* Clear the clear params. */ - anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CLEAR_PARAMS), cp); } /** @@ -1196,6 +1193,7 @@ void genX(CmdBeginRenderPass)( genX(cmd_buffer_set_subpass)(cmd_buffer, pass->subpasses); genX(cmd_buffer_do_hz_op)(cmd_buffer, ANV_HZ_OP_HIZ_RESOLVE); + genX(cmd_buffer_do_hz_op)(cmd_buffer, ANV_HZ_OP_CLEAR); anv_cmd_buffer_clear_subpass(cmd_buffer); } -- 2.9.3 </blockquote></div> </div></div>