<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Fri, Feb 10, 2017 at 3:18 PM, Nanley Chery <span dir="ltr"><<a href="mailto:nanleychery@gmail.com" target="_blank">nanleychery@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div class="gmail-HOEnZb"><div class="gmail-h5">On Fri, Feb 10, 2017 at 02:55:42PM -0800, Nanley Chery wrote:<br>
> On Fri, Feb 10, 2017 at 11:02:19AM -0800, Jason Ekstrand wrote:<br>
> > It's a bit hard to measure because it almost gets lost in the noise,<br>
> > but this seemed to help Dota 2 by a percent or two on my Broadwell<br>
> > GT3e desktop.<br>
> ><br>
> > Reviewed-by: Lionel Landwerlin <<a href="mailto:lionel.g.landwerlin@intel.com">lionel.g.landwerlin@intel.com</a><wbr>><br>
> > ---<br>
> > src/intel/vulkan/genX_<wbr>pipeline.c | 171 ++++++++++++++++++++++++++++++<wbr>+--------<br>
> > 1 file changed, 137 insertions(+), 34 deletions(-)<br>
> ><br>
> > diff --git a/src/intel/vulkan/genX_<wbr>pipeline.c b/src/intel/vulkan/genX_<wbr>pipeline.c<br>
> > index 71b9e30..d2af8b9 100644<br>
> > --- a/src/intel/vulkan/genX_<wbr>pipeline.c<br>
> > +++ b/src/intel/vulkan/genX_<wbr>pipeline.c<br>
> > @@ -638,6 +638,133 @@ static const uint32_t vk_to_gen_stencil_op[] = {<br>
> > [VK_STENCIL_OP_DECREMENT_AND_<wbr>WRAP] = STENCILOP_DECR,<br>
> > };<br>
> ><br>
> > +static bool<br>
> > +may_write_stencil_face(const VkStencilOpState *face,<br>
> > + VkCompareOp depthCompareOp)<br>
> > +{<br>
><br>
> This function triggers some false negatives. One example is<br>
><br>
> face->compareOp == VK_COMPARE_OP_GREATER &&<br>
> face->depthFailOp == VK_STENCIL_OP_KEEP &&<br>
> face->passOp == VK_STENCIL_OP_KEEP &&<br>
> face->failOp == VK_STENCIL_OP_INVERT &&<br>
> depthCompareOp == don't care;<br>
><br>
> This function returns false even though a stencil write occurs if the<br>
> stencil test results in a fail.<br>
<br>
</div></div>Thanks for pointing out in the office that this actually isn't a false<br>
negative. This function is correct, but I still think it could be<br>
simplified. I really needed a visual to evaluate it as it currently is<br>
written.<span class="gmail-HOEnZb"><font color="#888888"><br></font></span></blockquote><div><br></div><div>There is a way I could write it that may be clearer. How about:<br><br>may_write_stencil_face(const VkStencilOpState *face,<br>
VkCompareOp depthCompareOp)<br>
{<br></div><div> /* If compareOp is ALWAYS, then failOp doesn't matter */<br></div><div> if (face->compareOp == VK_COMPARE_OP_ALWAYS)<br></div><div> face->failOp = VK_STENCIL_OP_KEEP;<br><br></div><div> if (face->compareOp == VK_COMPARE_OP_NEVER ||<br></div><div> face->depthCompareOp == VK_COMPARE_OP_NEVER)<br></div><div> face->passOp = VK_STENCIL_OP_KEEP;<br><br><div> if (face->compareOp == VK_COMPARE_OP_NEVER ||<br></div><div> face->depthCompareOp == VK_COMPARE_OP_ALWAYS)<br></div> face->depthFailOp = VK_STENCIL_OP_KEEP;<br><br></div><div> return face->failOp != VK_STENCIL_OP_KEEP ||<br></div><div> face->depthFailOp != VK_STENCIL_OP_KEEP ||<br></div><div> face->passOp != VK_STENCIL_OP_KEEP;<br>}<br><br></div><div>I thought about writing it that way at one point but I decided sanitizing the stencil ops wasn't needed. I can switch if you'd rather.<br></div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><span class="gmail-HOEnZb"><font color="#888888">
-Nanley<br>
</font></span><div class="gmail-HOEnZb"><div class="gmail-h5"><br>
><br>
> The possible input combinations that affect writing to stencil can be<br>
> divided up into 10 sets.<br>
><br>
> s = stencil compare op<br>
> d = depth compare op<br>
> df = depth fail op<br>
> sf = stencil fail op<br>
> sp = stencil pass op<br>
><br>
> s.Always | !s.Always<br>
> d.Never | !d.Never | s.Never |<br>
> | d.Always | | |<br>
> !keep | | | |<br>
> | | | |<br>
> ------------|----------|------<wbr>|--------------|--------------<br>
> | | | |<br>
> keep df.Keep | sp.Keep | | sf.Keep |<br>
> | | | |<br>
> | | | |<br>
><br>
> We know that stencil won't be written if the function inputs land us in<br>
> 3 of these sets (denoted by the *.Keep's above). I think this function<br>
> would be easier to understand if it determined if we fell in one of the<br>
> 3 instead of the 7. How about something like this:<br>
><br>
> static bool<br>
> wont_write_stencil_face(const VkStencilOpState *face,<br>
> VkCompareOp depthCompareOp)<br>
> {<br>
> if (face->compareOp == VK_COMPARE_OP_NEVER &&<br>
> face->failOp == VK_STENCIL_OP_KEEP)<br>
> return true;<br>
><br>
> if (face->compareOp == VK_COMPARE_OP_ALWAYS) {<br>
> if (depthCompareOp == VK_COMPARE_OP_NEVER &&<br>
> face->depthFailOp == VK_STENCIL_OP_KEEP)<br>
> return true;<br>
> if (depthCompareOp == VK_COMPARE_OP_ALWAYS &&<br>
> face->passOp == VK_STENCIL_OP_KEEP)<br>
> return true;<br>
> }<br>
> return false;<br>
> }<br>
><br>
> -Nanley<br>
><br>
> > + /* If compareOp is ALWAYS then the stencil test will never fail and we can<br>
> > + * ignore failOp. If it's not ALWAYS and failOp is not KEEP, we may write<br>
> > + * stencil.<br>
> > + */<br>
> > + if (face->compareOp != VK_COMPARE_OP_ALWAYS &&<br>
> > + face->failOp != VK_STENCIL_OP_KEEP)<br>
> > + return true;<br>
> > +<br>
> > +<br>
> > + /* If the compareOp is NEVER then the stencil test will never pass and the<br>
> > + * passOp and depthFailOp don't matter. If compareOp isn't NEVER, then we<br>
> > + * need to take them into account.<br>
> > + */<br>
> > + if (face->compareOp != VK_COMPARE_OP_NEVER) {<br>
> > + /* If depthOp is NEVER then passOp doesn't matter. */<br>
> > + if (depthCompareOp != VK_COMPARE_OP_NEVER &&<br>
> > + face->passOp != VK_STENCIL_OP_KEEP)<br>
> > + return true;<br>
> > +<br>
> > + /* If depthOp is ALWAYS then depthFailOp doesn't matter. */<br>
> > + if (depthCompareOp != VK_COMPARE_OP_ALWAYS &&<br>
> > + face->depthFailOp != VK_STENCIL_OP_KEEP)<br>
> > + return true;<br>
> > + }<br>
> > +<br>
> > + return false;<br>
> > +}<br>
> > +<br>
> > +/* Intel hardware is fairly sensitive to whether or not depth/stencil writes<br>
> > + * are enabled. In the presence of discards, it's fairly easy to get into the<br>
> > + * non-promoted case which means a fairly big performance hit. From the Iron<br>
> > + * Lake PRM, Vol 2, pt. 1, section 8.4.3.2, "Early Depth Test Cases":<br>
> > + *<br>
> > + * "Non-promoted depth (N) is active whenever the depth test can be done<br>
> > + * early but it cannot determine whether or not to write source depth to<br>
> > + * the depth buffer, therefore the depth write must be performed post pixel<br>
> > + * shader. This includes cases where the pixel shader can kill pixels,<br>
> > + * including via sampler chroma key, as well as cases where the alpha test<br>
> > + * function is enabled, which kills pixels based on a programmable alpha<br>
> > + * test. In this case, even if the depth test fails, the pixel cannot be<br>
> > + * killed if a stencil write is indicated. Whether or not the stencil write<br>
> > + * happens depends on whether or not the pixel is killed later. In these<br>
> > + * cases if stencil test fails and stencil writes are off, the pixels can<br>
> > + * also be killed early. If stencil writes are enabled, the pixels must be<br>
> > + * treated as Computed depth (described above)."<br>
> > + *<br>
> > + * The same thing as mentioned in the stencil case can happen in the depth<br>
> > + * case as well if it thinks it writes depth but, thanks to the depth test<br>
> > + * being GL_EQUAL, the write doesn't actually matter. A little extra work<br>
> > + * up-front to try and disable depth and stencil writes can make a big<br>
> > + * difference.<br>
> > + *<br>
> > + * Unfortunately, the way depth and stencil testing is specified, there are<br>
> > + * many case where, regardless of depth/stencil writes being enabled, nothing<br>
> > + * actually gets written due to some other bit of state being set. This<br>
> > + * function attempts to "sanitize" the depth stencil state and disable writes<br>
> > + * and sometimes even testing whenever possible.<br>
> > + */<br>
> > +static void<br>
> > +sanitize_ds_state(<wbr>VkPipelineDepthStencilStateCre<wbr>ateInfo *state,<br>
> > + bool *stencilWriteEnable,<br>
> > + VkImageAspectFlags ds_aspects)<br>
> > +{<br>
> > + *stencilWriteEnable = state->stencilTestEnable;<br>
> > +<br>
> > + /* If the depth test is disabled, we won't be writing anything. */<br>
> > + if (!state->depthTestEnable)<br>
> > + state->depthWriteEnable = false;<br>
> > +<br>
> > + /* The Vulkan spec requires that if either depth or stencil is not present,<br>
> > + * the pipeline is to act as if the test silently passes.<br>
> > + */<br>
> > + if (!(ds_aspects & VK_IMAGE_ASPECT_DEPTH_BIT)) {<br>
> > + state->depthWriteEnable = false;<br>
> > + state->depthCompareOp = VK_COMPARE_OP_ALWAYS;<br>
> > + }<br>
> > +<br>
> > + if (!(ds_aspects & VK_IMAGE_ASPECT_STENCIL_BIT)) {<br>
> > + *stencilWriteEnable = false;<br>
> > + state->front.compareOp = VK_COMPARE_OP_ALWAYS;<br>
> > + state->back.compareOp = VK_COMPARE_OP_ALWAYS;<br>
> > + }<br>
> > +<br>
> > + /* If the stencil test is enabled and always fails, then we will never get<br>
> > + * to the depth test so we can just disable the depth test entirely.<br>
> > + */<br>
> > + if (state->stencilTestEnable &&<br>
> > + state->front.compareOp == VK_COMPARE_OP_NEVER &&<br>
> > + state->back.compareOp == VK_COMPARE_OP_NEVER) {<br>
> > + state->depthTestEnable = false;<br>
> > + state->depthWriteEnable = false;<br>
> > + }<br>
> > +<br>
> > + /* If depthCompareOp is EQUAL then the value we would be writing to the<br>
> > + * depth buffer is the same as the value that's already there so there's no<br>
> > + * point in writing it.<br>
> > + */<br>
> > + if (state->depthCompareOp == VK_COMPARE_OP_EQUAL)<br>
> > + state->depthWriteEnable = false;<br>
> > +<br>
> > + /* If the stencil ops are such that we don't actually ever modify the<br>
> > + * stencil buffer, we should disable writes.<br>
> > + */<br>
> > + if (!may_write_stencil_face(&<wbr>state->front, state->depthCompareOp) &&<br>
> > + !may_write_stencil_face(&<wbr>state->back, state->depthCompareOp))<br>
> > + *stencilWriteEnable = false;<br>
> > +<br>
> > + /* If the depth test always passes and we never write out depth, that's the<br>
> > + * same as if the depth test is disabled entirely.<br>
> > + */<br>
> > + if (state->depthCompareOp == VK_COMPARE_OP_ALWAYS &&<br>
> > + !state->depthWriteEnable)<br>
> > + state->depthTestEnable = false;<br>
> > +<br>
> > + /* If the stencil test always passes and we never write out stencil, that's<br>
> > + * the same as if the stencil test is disabled entirely.<br>
> > + */<br>
> > + if (state->front.compareOp == VK_COMPARE_OP_ALWAYS &&<br>
> > + state->back.compareOp == VK_COMPARE_OP_ALWAYS &&<br>
> > + !*stencilWriteEnable)<br>
> > + state->stencilTestEnable = false;<br>
> > +}<br>
> > +<br>
> > static void<br>
> > emit_ds_state(struct anv_pipeline *pipeline,<br>
> > const VkPipelineDepthStencilStateCre<wbr>ateInfo *pCreateInfo,<br>
> > @@ -663,12 +790,20 @@ emit_ds_state(struct anv_pipeline *pipeline,<br>
> > return;<br>
> > }<br>
> ><br>
> > + VkImageAspectFlags ds_aspects = 0;<br>
> > + if (subpass->depth_stencil_<wbr>attachment != VK_ATTACHMENT_UNUSED) {<br>
> > + VkFormat depth_stencil_format =<br>
> > + pass->attachments[subpass-><wbr>depth_stencil_attachment].<wbr>format;<br>
> > + ds_aspects = vk_format_aspects(depth_<wbr>stencil_format);<br>
> > + }<br>
> > +<br>
> > VkPipelineDepthStencilStateCre<wbr>ateInfo info = *pCreateInfo;<br>
> > + sanitize_ds_state(&info, &pipeline->writes_stencil, ds_aspects);<br>
> > + pipeline->writes_depth = info.depthWriteEnable;<br>
> > + pipeline->depth_test_enable = info.depthTestEnable;<br>
> ><br>
> > /* VkBool32 depthBoundsTestEnable; // optional (depth_bounds_test) */<br>
> ><br>
> > - pipeline->writes_stencil = info.stencilTestEnable;<br>
> > -<br>
> > #if GEN_GEN <= 7<br>
> > struct GENX(DEPTH_STENCIL_STATE) depth_stencil = {<br>
> > #else<br>
> > @@ -690,38 +825,6 @@ emit_ds_state(struct anv_pipeline *pipeline,<br>
> > .BackfaceStencilTestFunction = vk_to_gen_compare_op[info.<wbr>back.compareOp],<br>
> > };<br>
> ><br>
> > - VkImageAspectFlags aspects = 0;<br>
> > - if (subpass->depth_stencil_<wbr>attachment != VK_ATTACHMENT_UNUSED) {<br>
> > - VkFormat depth_stencil_format =<br>
> > - pass->attachments[subpass-><wbr>depth_stencil_attachment].<wbr>format;<br>
> > - aspects = vk_format_aspects(depth_<wbr>stencil_format);<br>
> > - }<br>
> > -<br>
> > - /* The Vulkan spec requires that if either depth or stencil is not present,<br>
> > - * the pipeline is to act as if the test silently passes.<br>
> > - */<br>
> > - if (!(aspects & VK_IMAGE_ASPECT_DEPTH_BIT)) {<br>
> > - depth_stencil.<wbr>DepthBufferWriteEnable = false;<br>
> > - depth_stencil.<wbr>DepthTestFunction = PREFILTEROPALWAYS;<br>
> > - }<br>
> > -<br>
> > - if (!(aspects & VK_IMAGE_ASPECT_STENCIL_BIT)) {<br>
> > - pipeline->writes_stencil = false;<br>
> > - depth_stencil.<wbr>StencilTestFunction = PREFILTEROPALWAYS;<br>
> > - depth_stencil.<wbr>BackfaceStencilTestFunction = PREFILTEROPALWAYS;<br>
> > - }<br>
> > -<br>
> > - /* From the Broadwell PRM:<br>
> > - *<br>
> > - * "If Depth_Test_Enable = 1 AND Depth_Test_func = EQUAL, the<br>
> > - * Depth_Write_Enable must be set to 0."<br>
> > - */<br>
> > - if (info.depthTestEnable && info.depthCompareOp == VK_COMPARE_OP_EQUAL)<br>
> > - depth_stencil.<wbr>DepthBufferWriteEnable = false;<br>
> > -<br>
> > - pipeline->writes_depth = depth_stencil.<wbr>DepthBufferWriteEnable;<br>
> > - pipeline->depth_test_enable = depth_stencil.DepthTestEnable;<br>
> > -<br>
> > #if GEN_GEN <= 7<br>
> > GENX(DEPTH_STENCIL_STATE_pack)<wbr>(NULL, depth_stencil_dw, &depth_stencil);<br>
> > #else<br>
> > --<br>
> > 2.5.0.400.gff86faf<br>
> ><br>
> > ______________________________<wbr>_________________<br>
> > mesa-dev mailing list<br>
> > <a href="mailto:mesa-dev@lists.freedesktop.org">mesa-dev@lists.freedesktop.org</a><br>
> > <a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev" rel="noreferrer" target="_blank">https://lists.freedesktop.org/<wbr>mailman/listinfo/mesa-dev</a><br>
</div></div></blockquote></div><br></div></div>