radv: implement dynamic sample locations enable

VK_EXT_sample_locations is only supported on < GFX10 due to some weird
issues on recent GPUs. extendedDynamicState3SampleLocationsEnable is
only enabled on GFX6-GFX9 for the same reason.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22947>

docs/relnotes/new_features.txt

@@ -1 +1,2 @@
 VK_EXT_attachment_feedback_loop_dynamic_state on RADV
+extendedDynamicState3SampleLocationsEnable on RADV

src/amd/vulkan/radv_cmd_buffer.c

@@ -203,6 +203,7 @@ radv_bind_dynamic_state(struct radv_cmd_buffer *cmd_buffer, const struct radv_dy
    RADV_CMP_COPY(vk.ms.alpha_to_coverage_enable, RADV_DYNAMIC_ALPHA_TO_COVERAGE_ENABLE);
    RADV_CMP_COPY(vk.ms.sample_mask, RADV_DYNAMIC_SAMPLE_MASK);
    RADV_CMP_COPY(vk.ms.rasterization_samples, RADV_DYNAMIC_RASTERIZATION_SAMPLES);
+   RADV_CMP_COPY(vk.ms.sample_locations_enable, RADV_DYNAMIC_SAMPLE_LOCATIONS_ENABLE);
 
    RADV_CMP_COPY(vk.ds.depth.bounds_test.min, RADV_DYNAMIC_DEPTH_BOUNDS);
    RADV_CMP_COPY(vk.ds.depth.bounds_test.max, RADV_DYNAMIC_DEPTH_BOUNDS);
@@ -1088,7 +1089,7 @@ radv_emit_sample_locations(struct radv_cmd_buffer *cmd_buffer)
    VkOffset2D sample_locs[4][8]; /* 8 is the max. sample count supported */
    uint64_t centroid_priority;
 
-   if (!d->sample_location.count)
+   if (!d->sample_location.count || !d->vk.ms.sample_locations_enable)
       return;
 
    /* Convert the user sample locations to hardware sample locations. */
@@ -6485,8 +6486,6 @@ radv_bind_multisample_state(struct radv_cmd_buffer *cmd_buffer,
       cmd_buffer->state.ms.sample_shading_enable = true;
       cmd_buffer->state.ms.min_sample_shading = ms->min_sample_shading;
    }
-
-   cmd_buffer->state.ms.uses_user_sample_locations = ms->uses_user_sample_locations;
 }
 
 static void
@@ -7564,6 +7563,17 @@ radv_CmdSetColorBlendEquationEXT(VkCommandBuffer commandBuffer, uint32_t firstAt
    state->dirty |= RADV_CMD_DIRTY_DYNAMIC_COLOR_BLEND_EQUATION;
 }
 
+VKAPI_ATTR void VKAPI_CALL
+radv_CmdSetSampleLocationsEnableEXT(VkCommandBuffer commandBuffer, VkBool32 sampleLocationsEnable)
+{
+   RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+   struct radv_cmd_state *state = &cmd_buffer->state;
+
+   state->dynamic.vk.ms.sample_locations_enable = sampleLocationsEnable;
+
+   state->dirty |= RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS_ENABLE;
+}
+
 VKAPI_ATTR void VKAPI_CALL
 radv_CmdSetDiscardRectangleEnableEXT(VkCommandBuffer commandBuffer, VkBool32 discardRectangleEnable)
 {
@@ -8953,7 +8963,7 @@ radv_get_ngg_culling_settings(struct radv_cmd_buffer *cmd_buffer, bool vp_y_inve
    /* Small primitive culling assumes a sample position at (0.5, 0.5)
     * so don't enable it with user sample locations.
     */
-   if (!cmd_buffer->state.ms.uses_user_sample_locations) {
+   if (!d->vk.ms.sample_locations_enable) {
       nggc_settings |= radv_nggc_small_primitives;
 
       /* small_prim_precision = num_samples / 2^subpixel_bits
@@ -9061,7 +9071,8 @@ radv_emit_all_graphics_states(struct radv_cmd_buffer *cmd_buffer, const struct r
         (RADV_CMD_DIRTY_PIPELINE | RADV_CMD_DIRTY_DYNAMIC_CULL_MODE |
          RADV_CMD_DIRTY_DYNAMIC_FRONT_FACE | RADV_CMD_DIRTY_DYNAMIC_RASTERIZER_DISCARD_ENABLE |
          RADV_CMD_DIRTY_DYNAMIC_VIEWPORT | RADV_CMD_DIRTY_DYNAMIC_CONSERVATIVE_RAST_MODE |
-         RADV_CMD_DIRTY_DYNAMIC_RASTERIZATION_SAMPLES | RADV_CMD_DIRTY_DYNAMIC_PRIMITIVE_TOPOLOGY)) &&
+         RADV_CMD_DIRTY_DYNAMIC_RASTERIZATION_SAMPLES | RADV_CMD_DIRTY_DYNAMIC_PRIMITIVE_TOPOLOGY |
+         RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS_ENABLE)) &&
        cmd_buffer->state.has_nggc)
       radv_emit_ngg_culling_state(cmd_buffer);
 

src/amd/vulkan/radv_physical_device.c

@@ -1244,7 +1244,7 @@ radv_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice,
          features->extendedDynamicState3ColorWriteMask = true;
          features->extendedDynamicState3RasterizationSamples = true;
          features->extendedDynamicState3ColorBlendEquation = true;
-         features->extendedDynamicState3SampleLocationsEnable = false; /* TODO */
+         features->extendedDynamicState3SampleLocationsEnable = pdevice->rad_info.gfx_level < GFX10;
          features->extendedDynamicState3LineRasterizationMode = true;
          features->extendedDynamicState3ExtraPrimitiveOverestimationSize = false;
          features->extendedDynamicState3AlphaToOneEnable = false;

src/amd/vulkan/radv_pipeline_graphics.c

@@ -370,8 +370,6 @@ radv_pipeline_init_multisample_state(const struct radv_device *device,
       ms->sample_shading_enable = true;
       ms->min_sample_shading = state->ms->min_sample_shading;
    }
-
-   ms->uses_user_sample_locations = state->ms && state->ms->sample_locations_enable;
 }
 
 static uint32_t
@@ -515,6 +513,8 @@ radv_dynamic_state_mask(VkDynamicState state)
       return RADV_DYNAMIC_DISCARD_RECTANGLE_MODE;
    case VK_DYNAMIC_STATE_ATTACHMENT_FEEDBACK_LOOP_ENABLE_EXT:
       return RADV_DYNAMIC_ATTACHMENT_FEEDBACK_LOOP_ENABLE;
+   case VK_DYNAMIC_STATE_SAMPLE_LOCATIONS_ENABLE_EXT:
+      return RADV_DYNAMIC_SAMPLE_LOCATIONS_ENABLE;
    default:
       unreachable("Unhandled dynamic state");
    }
@@ -590,7 +590,8 @@ radv_pipeline_needed_dynamic_state(const struct radv_graphics_pipeline *pipeline
        !state->dr->rectangle_count)
       states &= ~RADV_DYNAMIC_DISCARD_RECTANGLE;
 
-   if (!state->ms || !state->ms->sample_locations_enable)
+   if (!(pipeline->dynamic_states & RADV_DYNAMIC_SAMPLE_LOCATIONS_ENABLE) &&
+       (!state->ms || !state->ms->sample_locations_enable))
       states &= ~RADV_DYNAMIC_SAMPLE_LOCATIONS;
 
    if (!(pipeline->dynamic_states & RADV_DYNAMIC_LINE_STIPPLE_ENABLE) &&
@@ -1041,6 +1042,10 @@ radv_pipeline_init_dynamic_state(struct radv_graphics_pipeline *pipeline,
       dynamic->vk.ms.rasterization_samples = state->ms->rasterization_samples;
    }
 
+   if (states & RADV_DYNAMIC_SAMPLE_LOCATIONS_ENABLE) {
+      dynamic->vk.ms.sample_locations_enable = state->ms->sample_locations_enable;
+   }
+
    if (states & RADV_DYNAMIC_SAMPLE_LOCATIONS) {
       unsigned count = state->ms->sample_locations->per_pixel *
                        state->ms->sample_locations->grid_size.width *

src/amd/vulkan/radv_private.h

@@ -1279,7 +1279,8 @@ enum radv_dynamic_state_bits {
    RADV_DYNAMIC_DISCARD_RECTANGLE_ENABLE = 1ull << 46,
    RADV_DYNAMIC_DISCARD_RECTANGLE_MODE = 1ull << 47,
    RADV_DYNAMIC_ATTACHMENT_FEEDBACK_LOOP_ENABLE = 1ull << 48,
-   RADV_DYNAMIC_ALL = (1ull << 49) - 1,
+   RADV_DYNAMIC_SAMPLE_LOCATIONS_ENABLE = 1ull << 49,
+   RADV_DYNAMIC_ALL = (1ull << 50) - 1,
 };
 
 enum radv_cmd_dirty_bits {
@@ -1334,16 +1335,17 @@ enum radv_cmd_dirty_bits {
    RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE_ENABLE = 1ull << 46,
    RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE_MODE = 1ull << 47,
    RADV_CMD_DIRTY_DYNAMIC_ATTACHMENT_FEEDBACK_LOOP_ENABLE = 1ull << 48,
-   RADV_CMD_DIRTY_DYNAMIC_ALL = (1ull << 49) - 1,
-   RADV_CMD_DIRTY_PIPELINE = 1ull << 49,
-   RADV_CMD_DIRTY_INDEX_BUFFER = 1ull << 50,
-   RADV_CMD_DIRTY_FRAMEBUFFER = 1ull << 51,
-   RADV_CMD_DIRTY_VERTEX_BUFFER = 1ull << 52,
-   RADV_CMD_DIRTY_STREAMOUT_BUFFER = 1ull << 53,
-   RADV_CMD_DIRTY_GUARDBAND = 1ull << 54,
-   RADV_CMD_DIRTY_RBPLUS = 1ull << 55,
-   RADV_CMD_DIRTY_NGG_QUERY = 1ull << 56,
-   RADV_CMD_DIRTY_OCCLUSION_QUERY = 1ull << 57,
+   RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS_ENABLE = 1ull << 49,
+   RADV_CMD_DIRTY_DYNAMIC_ALL = (1ull << 50) - 1,
+   RADV_CMD_DIRTY_PIPELINE = 1ull << 50,
+   RADV_CMD_DIRTY_INDEX_BUFFER = 1ull << 51,
+   RADV_CMD_DIRTY_FRAMEBUFFER = 1ull << 52,
+   RADV_CMD_DIRTY_VERTEX_BUFFER = 1ull << 53,
+   RADV_CMD_DIRTY_STREAMOUT_BUFFER = 1ull << 54,
+   RADV_CMD_DIRTY_GUARDBAND = 1ull << 55,
+   RADV_CMD_DIRTY_RBPLUS = 1ull << 56,
+   RADV_CMD_DIRTY_NGG_QUERY = 1ull << 57,
+   RADV_CMD_DIRTY_OCCLUSION_QUERY = 1ull << 58,
 };
 
 enum radv_cmd_flush_bits {
@@ -1581,7 +1583,6 @@ enum rgp_flush_bits {
 
 struct radv_multisample_state {
    bool sample_shading_enable;
-   bool uses_user_sample_locations;
    float min_sample_shading;
 };