<div dir="auto">R-b, assuming you disable the ext and add todos for the htile stuff.</div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Fri, May 24, 2019, 9:28 AM Samuel Pitoiset <<a href="mailto:samuel.pitoiset@gmail.com">samuel.pitoiset@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
  
    
  
  <div text="#000000" bgcolor="#FFFFFF">
    <p>It's a bit difficult and invasive to support variable sample
      locations during layout transitions actually.</p>
    <p>So for now, I disabled HTILE for depth/stencil images that might
      require sample locations.<br>
    </p>
    <div class="m_-1581030183729916673moz-cite-prefix">On 5/22/19 10:20 PM, Marek Olšák wrote:<br>
    </div>
    <blockquote type="cite">
      
      <div dir="ltr">
        <div class="gmail_quote">
          <div class="gmail_attr">The depth decompress pass needs to
            know the sample locations.<br>
          </div>
          <div class="gmail_attr"><br>
          </div>
          <div class="gmail_attr">If shader loads read from compressed
            depth, the texture hardware will always use the standard
            locations for decompression.</div>
          <div class="gmail_attr"><br>
          </div>
          <div class="gmail_attr">Marek<br>
          </div>
          <div dir="ltr" class="gmail_attr"><br>
          </div>
          <div dir="ltr" class="gmail_attr">On Tue, May 21, 2019 at 8:17
            PM Bas Nieuwenhuizen <<a href="mailto:bas@basnieuwenhuizen.nl" target="_blank" rel="noreferrer">bas@basnieuwenhuizen.nl</a>>
            wrote:<br>
          </div>
          <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">So this does not seem to
            use the sample locations during layout transitions?<br>
            <br>
            AFAIK those are needed for e.g. HTILE decompression as it is
            based on<br>
            equations somehow.<br>
            <br>
            On Thu, May 16, 2019 at 11:51 AM Samuel Pitoiset<br>
            <<a href="mailto:samuel.pitoiset@gmail.com" target="_blank" rel="noreferrer">samuel.pitoiset@gmail.com</a>>
            wrote:<br>
            ><br>
            > Basically, this extension allows applications to use
            custom<br>
            > sample locations. It doesn't support variable sample
            locations<br>
            > during subpass. Note that we don't have to upload the
            user<br>
            > sample locations because the spec doesn't allow this.<br>
            ><br>
            > Only enabled on VI+ because it's untested on older
            chips.<br>
            ><br>
            > v2: - change sampleLocationCoordinateRange[1] to 0.9375<br>
            >     - compute and emit PA_SC_CENTROID_PRIORITY_{0,1}<br>
            >     - rebased on top of master<br>
            >     - some cleanups<br>
            ><br>
            > Signed-off-by: Samuel Pitoiset <<a href="mailto:samuel.pitoiset@gmail.com" target="_blank" rel="noreferrer">samuel.pitoiset@gmail.com</a>><br>
            > ---<br>
            >  src/amd/vulkan/radv_cmd_buffer.c  | 223
            ++++++++++++++++++++++++++++++<br>
            >  src/amd/vulkan/radv_device.c      |  27 ++++<br>
            >  src/amd/vulkan/radv_extensions.py |   1 +<br>
            >  src/amd/vulkan/radv_pipeline.c    |  30 ++++<br>
            >  src/amd/vulkan/radv_private.h     |  26 +++-<br>
            >  5 files changed, 300 insertions(+), 7 deletions(-)<br>
            ><br>
            > diff --git a/src/amd/vulkan/radv_cmd_buffer.c
            b/src/amd/vulkan/radv_cmd_buffer.c<br>
            > index 4f592bc7f68..fb79c1c6713 100644<br>
            > --- a/src/amd/vulkan/radv_cmd_buffer.c<br>
            > +++ b/src/amd/vulkan/radv_cmd_buffer.c<br>
            > @@ -105,6 +105,7 @@ radv_bind_dynamic_state(struct
            radv_cmd_buffer *cmd_buffer,<br>
            >         dest->viewport.count =
            src->viewport.count;<br>
            >         dest->scissor.count = src->scissor.count;<br>
            >         dest->discard_rectangle.count =
            src->discard_rectangle.count;<br>
            > +       dest->sample_location.count =
            src->sample_location.count;<br>
            ><br>
            >         if (copy_mask & RADV_DYNAMIC_VIEWPORT) {<br>
            >                 if
            (memcmp(&dest->viewport.viewports,
            &src->viewport.viewports,<br>
            > @@ -192,6 +193,22 @@ radv_bind_dynamic_state(struct
            radv_cmd_buffer *cmd_buffer,<br>
            >                 }<br>
            >         }<br>
            ><br>
            > +       if (copy_mask &
            RADV_DYNAMIC_SAMPLE_LOCATIONS) {<br>
            > +               if (dest->sample_location.per_pixel
            != src->sample_location.per_pixel ||<br>
            > +                 
             dest->sample_location.grid_size.width !=
            src->sample_location.grid_size.width ||<br>
            > +                 
             dest->sample_location.grid_size.height !=
            src->sample_location.grid_size.height ||<br>
            > +                 
             memcmp(&dest->sample_location.locations,<br>
            > +                         
            &src->sample_location.locations,<br>
            > +                         
            src->sample_location.count *
            sizeof(VkSampleLocationEXT))) {<br>
            > +                     
             dest->sample_location.per_pixel =
            src->sample_location.per_pixel;<br>
            > +                     
             dest->sample_location.grid_size =
            src->sample_location.grid_size;<br>
            > +                     
             typed_memcpy(dest->sample_location.locations,<br>
            > +                                   
            src->sample_location.locations,<br>
            > +                                   
            src->sample_location.count);<br>
            > +                       dest_mask |=
            RADV_DYNAMIC_SAMPLE_LOCATIONS;<br>
            > +               }<br>
            > +       }<br>
            > +<br>
            >         cmd_buffer->state.dirty |= dest_mask;<br>
            >  }<br>
            ><br>
            > @@ -632,6 +649,190 @@
            radv_emit_descriptor_pointers(struct radv_cmd_buffer
            *cmd_buffer,<br>
            >         }<br>
            >  }<br>
            ><br>
            > +/**<br>
            > + * Convert the user sample locations to hardware
            sample locations (the values<br>
            > + * that will be emitted by
            PA_SC_AA_SAMPLE_LOCS_PIXEL_*).<br>
            > + */<br>
            > +static void<br>
            > +radv_convert_user_sample_locs(struct
            radv_sample_locations_state *state,<br>
            > +                             uint32_t x, uint32_t y,
            VkOffset2D *sample_locs)<br>
            > +{<br>
            > +       uint32_t x_offset = x %
            state->grid_size.width;<br>
            > +       uint32_t y_offset = y %
            state->grid_size.height;<br>
            > +       uint32_t num_samples =
            (uint32_t)state->per_pixel;<br>
            > +       VkSampleLocationEXT *user_locs;<br>
            > +       uint32_t pixel_offset;<br>
            > +<br>
            > +       pixel_offset = (x_offset + y_offset *
            state->grid_size.width) * num_samples;<br>
            > +<br>
            > +       assert(pixel_offset <=
            MAX_SAMPLE_LOCATIONS);<br>
            > +       user_locs =
            &state->locations[pixel_offset];<br>
            > +<br>
            > +       for (uint32_t i = 0; i < num_samples; i++) {<br>
            > +               float shifted_pos_x = user_locs[i].x -
            0.5;<br>
            > +               float shifted_pos_y = user_locs[i].y -
            0.5;<br>
            > +<br>
            > +               int32_t scaled_pos_x =
            floor(shifted_pos_x * 16);<br>
            > +               int32_t scaled_pos_y =
            floor(shifted_pos_y * 16);<br>
            > +<br>
            > +               sample_locs[i].x = CLAMP(scaled_pos_x,
            -8, 7);<br>
            > +               sample_locs[i].y = CLAMP(scaled_pos_y,
            -8, 7);<br>
            > +       }<br>
            > +}<br>
            > +<br>
            > +/**<br>
            > + * Compute the PA_SC_AA_SAMPLE_LOCS_PIXEL_* mask based
            on hardware sample<br>
            > + * locations.<br>
            > + */<br>
            > +static void<br>
            > +radv_compute_sample_locs_pixel(uint32_t num_samples,
            VkOffset2D *sample_locs,<br>
            > +                              uint32_t
            *sample_locs_pixel)<br>
            > +{<br>
            > +       for (uint32_t i = 0; i < num_samples; i++) {<br>
            > +               uint32_t sample_reg_idx = i / 4;<br>
            > +               uint32_t sample_loc_idx = i % 4;<br>
            > +               int32_t pos_x = sample_locs[i].x;<br>
            > +               int32_t pos_y = sample_locs[i].y;<br>
            > +<br>
            > +               uint32_t shift_x = 8 * sample_loc_idx;<br>
            > +               uint32_t shift_y = shift_x + 4;<br>
            > +<br>
            > +               sample_locs_pixel[sample_reg_idx] |=
            (pos_x & 0xf) << shift_x;<br>
            > +               sample_locs_pixel[sample_reg_idx] |=
            (pos_y & 0xf) << shift_y;<br>
            > +       }<br>
            > +}<br>
            > +<br>
            > +/**<br>
            > + * Compute the PA_SC_CENTROID_PRIORITY_* mask based on
            the top left hardware<br>
            > + * sample locations.<br>
            > + */<br>
            > +static uint64_t<br>
            > +radv_compute_centroid_priority(struct radv_cmd_buffer
            *cmd_buffer,<br>
            > +                              VkOffset2D *sample_locs,<br>
            > +                              uint32_t num_samples)<br>
            > +{<br>
            > +       uint32_t centroid_priorities[num_samples];<br>
            > +       uint32_t sample_mask = num_samples - 1;<br>
            > +       uint32_t distances[num_samples];<br>
            > +       uint64_t centroid_priority = 0;<br>
            > +<br>
            > +       /* Compute the distances from center for each
            sample. */<br>
            > +       for (int i = 0; i < num_samples; i++) {<br>
            > +               distances[i] = (sample_locs[i].x *
            sample_locs[i].x) +<br>
            > +                              (sample_locs[i].y *
            sample_locs[i].y);<br>
            > +       }<br>
            > +<br>
            > +       /* Compute the centroid priorities by looking
            at the distances array. */<br>
            > +       for (int i = 0; i < num_samples; i++) {<br>
            > +               uint32_t min_idx = 0;<br>
            > +<br>
            > +               for (int j = 1; j < num_samples;
            j++) {<br>
            > +                       if (distances[j] <
            distances[min_idx])<br>
            > +                               min_idx = j;<br>
            > +               }<br>
            > +<br>
            > +               centroid_priorities[i] = min_idx;<br>
            > +               distances[min_idx] = 0xffffffff;<br>
            > +       }<br>
            > +<br>
            > +       /* Compute the final centroid priority. */<br>
            > +       for (int i = 0; i < 8; i++) {<br>
            > +               centroid_priority |=<br>
            > +                       centroid_priorities[i &
            sample_mask] << (i * 4);<br>
            > +       }<br>
            > +<br>
            > +       return centroid_priority << 32 |
            centroid_priority;<br>
            > +}<br>
            > +<br>
            > +/**<br>
            > + * Emit the sample locations that are specified with
            VK_EXT_sample_locations.<br>
            > + */<br>
            > +static void<br>
            > +radv_emit_sample_locations(struct radv_cmd_buffer
            *cmd_buffer)<br>
            > +{<br>
            > +       struct radv_pipeline *pipeline =
            cmd_buffer->state.pipeline;<br>
            > +       struct radv_multisample_state *ms =
            &pipeline-><a href="http://graphics.ms" rel="noreferrer noreferrer" target="_blank">graphics.ms</a>;<br>
            > +       struct radv_sample_locations_state
            *sample_location =<br>
            > +             
             &cmd_buffer->state.dynamic.sample_location;<br>
            > +       uint32_t num_samples =
            (uint32_t)sample_location->per_pixel;<br>
            > +       struct radeon_cmdbuf *cs = cmd_buffer->cs;<br>
            > +       uint32_t sample_locs_pixel[4][2] = {};<br>
            > +       VkOffset2D sample_locs[4][8]; /* 8 is the max.
            sample count supported */<br>
            > +       uint32_t max_sample_dist = 0;<br>
            > +       uint64_t centroid_priority;<br>
            > +<br>
            > +       if
            (!cmd_buffer->state.dynamic.sample_location.count)<br>
            > +               return;<br>
            > +<br>
            > +       /* Convert the user sample locations to
            hardware sample locations. */<br>
            > +       radv_convert_user_sample_locs(sample_location,
            0, 0, sample_locs[0]);<br>
            > +       radv_convert_user_sample_locs(sample_location,
            1, 0, sample_locs[1]);<br>
            > +       radv_convert_user_sample_locs(sample_location,
            0, 1, sample_locs[2]);<br>
            > +       radv_convert_user_sample_locs(sample_location,
            1, 1, sample_locs[3]);<br>
            > +<br>
            > +       /* Compute the PA_SC_AA_SAMPLE_LOCS_PIXEL_*
            mask. */<br>
            > +       for (uint32_t i = 0; i < 4; i++) {<br>
            > +             
             radv_compute_sample_locs_pixel(num_samples, sample_locs[i],<br>
            > +                                             
            sample_locs_pixel[i]);<br>
            > +       }<br>
            > +<br>
            > +       /* Compute the PA_SC_CENTROID_PRIORITY_* mask.
            */<br>
            > +       centroid_priority =<br>
            > +             
             radv_compute_centroid_priority(cmd_buffer, sample_locs[0],<br>
            > +                                             
            num_samples);<br>
            > +<br>
            > +       /* Compute the maximum sample distance from the
            specified locations. */<br>
            > +       for (uint32_t i = 0; i < num_samples; i++) {<br>
            > +               VkOffset2D offset = sample_locs[0][i];<br>
            > +               max_sample_dist = MAX2(max_sample_dist,<br>
            > +                                     
            MAX2(abs(offset.x), abs(offset.y)));<br>
            > +       }<br>
            > +<br>
            > +       /* Emit the specified user sample locations. */<br>
            > +       switch (num_samples) {<br>
            > +       case 2:<br>
            > +       case 4:<br>
            > +               radeon_set_context_reg(cs,
            R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0,
            sample_locs_pixel[0][0]);<br>
            > +               radeon_set_context_reg(cs,
            R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0,
            sample_locs_pixel[1][0]);<br>
            > +               radeon_set_context_reg(cs,
            R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0,
            sample_locs_pixel[2][0]);<br>
            > +               radeon_set_context_reg(cs,
            R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0,
            sample_locs_pixel[3][0]);<br>
            > +               break;<br>
            > +       case 8:<br>
            > +               radeon_set_context_reg(cs,
            R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0,
            sample_locs_pixel[0][0]);<br>
            > +               radeon_set_context_reg(cs,
            R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0,
            sample_locs_pixel[1][0]);<br>
            > +               radeon_set_context_reg(cs,
            R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0,
            sample_locs_pixel[2][0]);<br>
            > +               radeon_set_context_reg(cs,
            R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0,
            sample_locs_pixel[3][0]);<br>
            > +               radeon_set_context_reg(cs,
            R_028BFC_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_1,
            sample_locs_pixel[0][1]);<br>
            > +               radeon_set_context_reg(cs,
            R_028C0C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_1,
            sample_locs_pixel[1][1]);<br>
            > +               radeon_set_context_reg(cs,
            R_028C1C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_1,
            sample_locs_pixel[2][1]);<br>
            > +               radeon_set_context_reg(cs,
            R_028C2C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_1,
            sample_locs_pixel[3][1]);<br>
            > +               break;<br>
            > +       default:<br>
            > +               unreachable("invalid number of
            samples");<br>
            > +       }<br>
            > +<br>
            > +       /* Emit the maximum sample distance and the
            centroid priority. */<br>
            > +       uint32_t pa_sc_aa_config =
            ms->pa_sc_aa_config;<br>
            > +<br>
            > +       pa_sc_aa_config &=
            C_028BE0_MAX_SAMPLE_DIST;<br>
            > +       pa_sc_aa_config |=
            S_028BE0_MAX_SAMPLE_DIST(max_sample_dist);<br>
            > +<br>
            > +       radeon_set_context_reg_seq(cs,
            R_028BE0_PA_SC_AA_CONFIG, 1);<br>
            > +       radeon_emit(cs, pa_sc_aa_config);<br>
            > +<br>
            > +       radeon_set_context_reg_seq(cs,
            R_028BD4_PA_SC_CENTROID_PRIORITY_0, 2);<br>
            > +       radeon_emit(cs, centroid_priority);<br>
            > +       radeon_emit(cs, centroid_priority >> 32);<br>
            > +<br>
            > +       /* GFX9: Flush DFSM when the AA mode changes.
            */<br>
            > +       if (cmd_buffer->device->dfsm_allowed) {<br>
            > +               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE,
            0, 0));<br>
            > +               radeon_emit(cs,
            EVENT_TYPE(V_028A90_FLUSH_DFSM) | EVENT_INDEX(0));<br>
            > +       }<br>
            > +<br>
            > +     
             cmd_buffer->state.context_roll_without_scissor_emitted =
            true;<br>
            > +}<br>
            > +<br>
            >  static void<br>
            >  radv_emit_inline_push_consts(struct radv_cmd_buffer
            *cmd_buffer,<br>
            >                              struct radv_pipeline
            *pipeline,<br>
            > @@ -1775,6 +1976,9 @@
            radv_cmd_buffer_flush_dynamic_state(struct radv_cmd_buffer
            *cmd_buffer)<br>
            >         if (states &
            RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE)<br>
            >               
             radv_emit_discard_rectangle(cmd_buffer);<br>
            ><br>
            > +       if (states &
            RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS)<br>
            > +               radv_emit_sample_locations(cmd_buffer);<br>
            > +<br>
            >         cmd_buffer->state.dirty &= ~states;<br>
            >  }<br>
            ><br>
            > @@ -3266,6 +3470,25 @@ void
            radv_CmdSetDiscardRectangleEXT(<br>
            >         state->dirty |=
            RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE;<br>
            >  }<br>
            ><br>
            > +void radv_CmdSetSampleLocationsEXT(<br>
            > +       VkCommandBuffer                           
             commandBuffer,<br>
            > +       const VkSampleLocationsInfoEXT*           
             pSampleLocationsInfo)<br>
            > +{<br>
            > +       RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer,
            commandBuffer);<br>
            > +       struct radv_cmd_state *state =
            &cmd_buffer->state;<br>
            > +<br>
            > +     
             assert(pSampleLocationsInfo->sampleLocationsCount <=
            MAX_SAMPLE_LOCATIONS);<br>
            > +<br>
            > +       state->dynamic.sample_location.per_pixel =
            pSampleLocationsInfo->sampleLocationsPerPixel;<br>
            > +       state->dynamic.sample_location.grid_size =
            pSampleLocationsInfo->sampleLocationGridSize;<br>
            > +       state->dynamic.sample_location.count =
            pSampleLocationsInfo->sampleLocationsCount;<br>
            > +     
             typed_memcpy(&state->dynamic.sample_location.locations[0],<br>
            > +                   
            pSampleLocationsInfo->pSampleLocations,<br>
            > +                   
            pSampleLocationsInfo->sampleLocationsCount);<br>
            > +<br>
            > +       state->dirty |=
            RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS;<br>
            > +}<br>
            > +<br>
            >  void radv_CmdExecuteCommands(<br>
            >         VkCommandBuffer                           
             commandBuffer,<br>
            >         uint32_t                                   
            commandBufferCount,<br>
            > diff --git a/src/amd/vulkan/radv_device.c
            b/src/amd/vulkan/radv_device.c<br>
            > index c0e317a97e5..e523bba1223 100644<br>
            > --- a/src/amd/vulkan/radv_device.c<br>
            > +++ b/src/amd/vulkan/radv_device.c<br>
            > @@ -1360,6 +1360,19 @@ void
            radv_GetPhysicalDeviceProperties2(<br>
            >                       
             props->maxDescriptorSetUpdateAfterBindInlineUniformBlocks
            = MAX_INLINE_UNIFORM_BLOCK_COUNT;<br>
            >                         break;<br>
            >                 }<br>
            > +               case
            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLE_LOCATIONS_PROPERTIES_EXT:
            {<br>
            > +                     
             VkPhysicalDeviceSampleLocationsPropertiesEXT *properties =<br>
            > +                             
             (VkPhysicalDeviceSampleLocationsPropertiesEXT *)ext;<br>
            > +                     
             properties->sampleLocationSampleCounts =
            VK_SAMPLE_COUNT_2_BIT |<br>
            > +                                                     
                      VK_SAMPLE_COUNT_4_BIT |<br>
            > +                                                     
                      VK_SAMPLE_COUNT_8_BIT;<br>
            > +                     
             properties->maxSampleLocationGridSize = (VkExtent2D){ 2
            , 2 };<br>
            > +                     
             properties->sampleLocationCoordinateRange[0] = 0.0f;<br>
            > +                     
             properties->sampleLocationCoordinateRange[1] = 0.9375f;<br>
            > +                     
             properties->sampleLocationSubPixelBits = 4;<br>
            > +                     
             properties->variableSampleLocations = VK_FALSE;<br>
            > +                       break;<br>
            > +               }<br>
            >                 default:<br>
            >                         break;<br>
            >                 }<br>
            > @@ -5361,3 +5374,17 @@ VkResult
            radv_GetCalibratedTimestampsEXT(<br>
            ><br>
            >         return VK_SUCCESS;<br>
            >  }<br>
            > +<br>
            > +void radv_GetPhysicalDeviceMultisamplePropertiesEXT(<br>
            > +    VkPhysicalDevice                           
            physicalDevice,<br>
            > +    VkSampleCountFlagBits                     
             samples,<br>
            > +    VkMultisamplePropertiesEXT*               
             pMultisampleProperties)<br>
            > +{<br>
            > +       if (samples & (VK_SAMPLE_COUNT_2_BIT |<br>
            > +                      VK_SAMPLE_COUNT_4_BIT |<br>
            > +                      VK_SAMPLE_COUNT_8_BIT)) {<br>
            > +             
             pMultisampleProperties->maxSampleLocationGridSize =
            (VkExtent2D){ 2, 2 };<br>
            > +       } else {<br>
            > +             
             pMultisampleProperties->maxSampleLocationGridSize =
            (VkExtent2D){ 0, 0 };<br>
            > +       }<br>
            > +}<br>
            > diff --git a/src/amd/vulkan/radv_extensions.py
            b/src/amd/vulkan/radv_extensions.py<br>
            > index 0b5af56a435..2c2c4dc74d2 100644<br>
            > --- a/src/amd/vulkan/radv_extensions.py<br>
            > +++ b/src/amd/vulkan/radv_extensions.py<br>
            > @@ -119,6 +119,7 @@ EXTENSIONS = [<br>
            >      Extension('VK_EXT_memory_priority',               
               1, True),<br>
            >      Extension('VK_EXT_pci_bus_info',                 
                2, True),<br>
            >      Extension('VK_EXT_pipeline_creation_feedback',   
                1, True),<br>
            > +    Extension('VK_EXT_sample_locations',             
                1, 'device->rad_info.chip_class >= GFX8'),<br>
            >      Extension('VK_EXT_sampler_filter_minmax',         
               1, 'device->rad_info.chip_class >= GFX7'),<br>
            >      Extension('VK_EXT_scalar_block_layout',           
               1, 'device->rad_info.chip_class >= GFX7'),<br>
            >      Extension('VK_EXT_shader_viewport_index_layer',   
               1, True),<br>
            > diff --git a/src/amd/vulkan/radv_pipeline.c
            b/src/amd/vulkan/radv_pipeline.c<br>
            > index 56fd65bec29..7f8e00b5c3c 100644<br>
            > --- a/src/amd/vulkan/radv_pipeline.c<br>
            > +++ b/src/amd/vulkan/radv_pipeline.c<br>
            > @@ -1267,6 +1267,8 @@ static unsigned
            radv_dynamic_state_mask(VkDynamicState state)<br>
            >                 return RADV_DYNAMIC_STENCIL_REFERENCE;<br>
            >         case VK_DYNAMIC_STATE_DISCARD_RECTANGLE_EXT:<br>
            >                 return RADV_DYNAMIC_DISCARD_RECTANGLE;<br>
            > +       case VK_DYNAMIC_STATE_SAMPLE_LOCATIONS_EXT:<br>
            > +               return RADV_DYNAMIC_SAMPLE_LOCATIONS;<br>
            >         default:<br>
            >                 unreachable("Unhandled dynamic state");<br>
            >         }<br>
            > @@ -1297,6 +1299,11 @@ static uint32_t
            radv_pipeline_needed_dynamic_state(const
            VkGraphicsPipelineCreat<br>
            >         if
            (!vk_find_struct_const(pCreateInfo->pNext,
            PIPELINE_DISCARD_RECTANGLE_STATE_CREATE_INFO_EXT))<br>
            >                 states &=
            ~RADV_DYNAMIC_DISCARD_RECTANGLE;<br>
            ><br>
            > +       if (!pCreateInfo->pMultisampleState ||<br>
            > +         
             !vk_find_struct_const(pCreateInfo->pMultisampleState->pNext,<br>
            > +                               
             PIPELINE_SAMPLE_LOCATIONS_STATE_CREATE_INFO_EXT))<br>
            > +               states &=
            ~RADV_DYNAMIC_SAMPLE_LOCATIONS;<br>
            > +<br>
            >         /* TODO: blend constants & line width. */<br>
            ><br>
            >         return states;<br>
            > @@ -1426,6 +1433,29 @@
            radv_pipeline_init_dynamic_state(struct radv_pipeline
            *pipeline,<br>
            >                 }<br>
            >         }<br>
            ><br>
            > +       if (needed_states &
            RADV_DYNAMIC_SAMPLE_LOCATIONS) {<br>
            > +               const
            VkPipelineSampleLocationsStateCreateInfoEXT
            *sample_location_info =<br>
            > +                     
             vk_find_struct_const(pCreateInfo->pMultisampleState->pNext,<br>
            > +                                           
            PIPELINE_SAMPLE_LOCATIONS_STATE_CREATE_INFO_EXT);<br>
            > +               /* If sampleLocationsEnable is
            VK_FALSE, the default sample<br>
            > +                * locations are used and the values
            specified in<br>
            > +                * sampleLocationsInfo are ignored.<br>
            > +                */<br>
            > +               if
            (sample_location_info->sampleLocationsEnable) {<br>
            > +                       const VkSampleLocationsInfoEXT
            *pSampleLocationsInfo =<br>
            > +                             
             &sample_location_info->sampleLocationsInfo;<br>
            > +<br>
            > +                     
             assert(pSampleLocationsInfo->sampleLocationsCount <=
            MAX_SAMPLE_LOCATIONS);<br>
            > +<br>
            > +                     
             dynamic->sample_location.per_pixel =
            pSampleLocationsInfo->sampleLocationsPerPixel;<br>
            > +                     
             dynamic->sample_location.grid_size =
            pSampleLocationsInfo->sampleLocationGridSize;<br>
            > +                     
             dynamic->sample_location.count =
            pSampleLocationsInfo->sampleLocationsCount;<br>
            > +                     
             typed_memcpy(&dynamic->sample_location.locations[0],<br>
            > +                                   
            pSampleLocationsInfo->pSampleLocations,<br>
            > +                                   
            pSampleLocationsInfo->sampleLocationsCount);<br>
            > +               }<br>
            > +       }<br>
            > +<br>
            >         pipeline->dynamic_state.mask = states;<br>
            >  }<br>
            ><br>
            > diff --git a/src/amd/vulkan/radv_private.h
            b/src/amd/vulkan/radv_private.h<br>
            > index 7834a505562..5c84bf74352 100644<br>
            > --- a/src/amd/vulkan/radv_private.h<br>
            > +++ b/src/amd/vulkan/radv_private.h<br>
            > @@ -91,6 +91,7 @@ typedef uint32_t xcb_window_t;<br>
            >  #define MAX_VIEWPORTS   16<br>
            >  #define MAX_SCISSORS    16<br>
            >  #define MAX_DISCARD_RECTANGLES 4<br>
            > +#define MAX_SAMPLE_LOCATIONS 32<br>
            >  #define MAX_PUSH_CONSTANTS_SIZE 128<br>
            >  #define MAX_PUSH_DESCRIPTORS 32<br>
            >  #define MAX_DYNAMIC_UNIFORM_BUFFERS 16<br>
            > @@ -852,7 +853,8 @@ enum radv_dynamic_state_bits {<br>
            >         RADV_DYNAMIC_STENCIL_WRITE_MASK   = 1 <<
            7,<br>
            >         RADV_DYNAMIC_STENCIL_REFERENCE    = 1 <<
            8,<br>
            >         RADV_DYNAMIC_DISCARD_RECTANGLE    = 1 <<
            9,<br>
            > -       RADV_DYNAMIC_ALL                  = (1 <<
            10) - 1,<br>
            > +       RADV_DYNAMIC_SAMPLE_LOCATIONS     = 1 <<
            10,<br>
            > +       RADV_DYNAMIC_ALL                  = (1 <<
            11) - 1,<br>
            >  };<br>
            ><br>
            >  enum radv_cmd_dirty_bits {<br>
            > @@ -868,12 +870,13 @@ enum radv_cmd_dirty_bits {<br>
            >         RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK     
              = 1 << 7,<br>
            >         RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE       
             = 1 << 8,<br>
            >         RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE       
             = 1 << 9,<br>
            > -       RADV_CMD_DIRTY_DYNAMIC_ALL                     
             = (1 << 10) - 1,<br>
            > -       RADV_CMD_DIRTY_PIPELINE                       
              = 1 << 10,<br>
            > -       RADV_CMD_DIRTY_INDEX_BUFFER                   
              = 1 << 11,<br>
            > -       RADV_CMD_DIRTY_FRAMEBUFFER                     
             = 1 << 12,<br>
            > -       RADV_CMD_DIRTY_VERTEX_BUFFER                   
             = 1 << 13,<br>
            > -       RADV_CMD_DIRTY_STREAMOUT_BUFFER               
              = 1 << 14,<br>
            > +       RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS       
              = 1 << 10,<br>
            > +       RADV_CMD_DIRTY_DYNAMIC_ALL                     
             = (1 << 11) - 1,<br>
            > +       RADV_CMD_DIRTY_PIPELINE                       
              = 1 << 11,<br>
            > +       RADV_CMD_DIRTY_INDEX_BUFFER                   
              = 1 << 12,<br>
            > +       RADV_CMD_DIRTY_FRAMEBUFFER                     
             = 1 << 13,<br>
            > +       RADV_CMD_DIRTY_VERTEX_BUFFER                   
             = 1 << 14,<br>
            > +       RADV_CMD_DIRTY_STREAMOUT_BUFFER               
              = 1 << 15,<br>
            >  };<br>
            ><br>
            >  enum radv_cmd_flush_bits {<br>
            > @@ -950,6 +953,13 @@ struct
            radv_discard_rectangle_state {<br>
            >         VkRect2D                                       
              rectangles[MAX_DISCARD_RECTANGLES];<br>
            >  };<br>
            ><br>
            > +struct radv_sample_locations_state {<br>
            > +       VkSampleCountFlagBits per_pixel;<br>
            > +       VkExtent2D grid_size;<br>
            > +       uint32_t count;<br>
            > +       VkSampleLocationEXT
            locations[MAX_SAMPLE_LOCATIONS];<br>
            > +};<br>
            > +<br>
            >  struct radv_dynamic_state {<br>
            >         /**<br>
            >          * Bitmask of (1 << VK_DYNAMIC_STATE_*).<br>
            > @@ -992,6 +1002,8 @@ struct radv_dynamic_state {<br>
            >         } stencil_reference;<br>
            ><br>
            >         struct radv_discard_rectangle_state           
               discard_rectangle;<br>
            > +<br>
            > +       struct radv_sample_locations_state             
              sample_location;<br>
            >  };<br>
            ><br>
            >  extern const struct radv_dynamic_state
            default_dynamic_state;<br>
            > --<br>
            > 2.21.0<br>
            ><br>
            > _______________________________________________<br>
            > mesa-dev mailing list<br>
            > <a href="mailto:mesa-dev@lists.freedesktop.org" target="_blank" rel="noreferrer">mesa-dev@lists.freedesktop.org</a><br>
            > <a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev" rel="noreferrer noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/mesa-dev</a><br>
            _______________________________________________<br>
            mesa-dev mailing list<br>
            <a href="mailto:mesa-dev@lists.freedesktop.org" target="_blank" rel="noreferrer">mesa-dev@lists.freedesktop.org</a><br>
            <a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev" rel="noreferrer noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/mesa-dev</a></blockquote>
        </div>
      </div>
    </blockquote>
  </div>

</blockquote></div>