panvk: enable KHR_line_rasterization support

This allows users to toggle between rectangular and bresenham style
rasterization.

The bresenham style rasterization is performed by disabling
multisampling and changing the end-points to be axis-aligned. This is
similar to what we already do in Gallium.

Acked-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: Benjamin Lee <benjamin.lee@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33844>

docs/features.txt

@@ -513,7 +513,7 @@ Vulkan 1.4 -- all DONE: anv, lvp, nvk, radv/gfx8+, tu/a7xx+, vn
   VK_KHR_dynamic_rendering_local_read                   DONE (anv, lvp, nvk, radv, tu, vn)
   VK_KHR_global_priority                                DONE (anv, lvp, nvk, panvk, radv, tu, vn)
   VK_KHR_index_type_uint8                               DONE (anv, lvp, nvk, panvk, pvr, radv, tu, v3dv, vn)
-  VK_KHR_line_rasterization                             DONE (anv, lvp, nvk, radv, tu, v3dv, vn)
+  VK_KHR_line_rasterization                             DONE (anv, lvp, nvk, panvk, radv, tu, v3dv, vn)
   VK_KHR_load_store_op_none                             DONE (anv, lvp, nvk, radv, tu, v3dv, vn)
   VK_KHR_maintenance5                                   DONE (anv, lvp, nvk, radv, tu, v3dv, vn)
   VK_KHR_maintenance6                                   DONE (anv, lvp, nvk, radv, tu, vn)

docs/relnotes/new_features.txt

@@ -25,3 +25,4 @@ EXT_shader_framebuffer_image_fetch_coherent on v3d
 KHR_blend_equation_advanced on v3d
 KHR_blend_equation_advanced_coherent on v3d
 KHR_partial_update on etnaviv
+VK_KHR_line_rasterization on panvk

src/panfrost/vulkan/csf/panvk_vX_cmd_draw.c

@@ -1464,6 +1464,7 @@ prepare_dcd(struct panvk_cmd_buffer *cmdbuf)
    bool dcd0_dirty =
       dyn_gfx_state_dirty(cmdbuf, RS_RASTERIZER_DISCARD_ENABLE) ||
       dyn_gfx_state_dirty(cmdbuf, RS_CULL_MODE) ||
+      dyn_gfx_state_dirty(cmdbuf, RS_LINE_MODE) ||
       dyn_gfx_state_dirty(cmdbuf, RS_FRONT_FACE) ||
       dyn_gfx_state_dirty(cmdbuf, MS_RASTERIZATION_SAMPLES) ||
       dyn_gfx_state_dirty(cmdbuf, MS_SAMPLE_MASK) ||
@@ -1477,10 +1478,15 @@ prepare_dcd(struct panvk_cmd_buffer *cmdbuf)
       dyn_gfx_state_dirty(cmdbuf, DS_STENCIL_TEST_ENABLE) ||
       dyn_gfx_state_dirty(cmdbuf, DS_STENCIL_OP) ||
       dyn_gfx_state_dirty(cmdbuf, DS_STENCIL_WRITE_MASK) ||
+      /* line mode needs primitive topology */
+      dyn_gfx_state_dirty(cmdbuf, IA_PRIMITIVE_TOPOLOGY) ||
       fs_user_dirty(cmdbuf) || gfx_state_dirty(cmdbuf, RENDER_STATE) ||
       gfx_state_dirty(cmdbuf, OQ);
    bool dcd1_dirty = dyn_gfx_state_dirty(cmdbuf, MS_RASTERIZATION_SAMPLES) ||
                      dyn_gfx_state_dirty(cmdbuf, MS_SAMPLE_MASK) ||
+                     /* line mode needs primitive topology */
+                     dyn_gfx_state_dirty(cmdbuf, RS_LINE_MODE) ||
+                     dyn_gfx_state_dirty(cmdbuf, IA_PRIMITIVE_TOPOLOGY) ||
                      fs_user_dirty(cmdbuf) ||
                      gfx_state_dirty(cmdbuf, RENDER_STATE);
 
@@ -1488,10 +1494,27 @@ prepare_dcd(struct panvk_cmd_buffer *cmdbuf)
       &cmdbuf->vk.dynamic_graphics_state;
    const struct vk_rasterization_state *rs =
       &cmdbuf->vk.dynamic_graphics_state.rs;
+   const struct vk_input_assembly_state *ia =
+      &cmdbuf->vk.dynamic_graphics_state.ia;
+
    bool alpha_to_coverage = dyns->ms.alpha_to_coverage_enable;
    bool writes_z = writes_depth(cmdbuf);
    bool writes_s = writes_stencil(cmdbuf);
 
+   bool msaa = dyns->ms.rasterization_samples > 1;
+   if ((ia->primitive_topology == VK_PRIMITIVE_TOPOLOGY_LINE_LIST ||
+        ia->primitive_topology == VK_PRIMITIVE_TOPOLOGY_LINE_STRIP) &&
+       rs->line.mode == VK_LINE_RASTERIZATION_MODE_BRESENHAM) {
+      /* we need to disable MSAA when rendering bresenham lines.
+       *
+       * From the Vulkan spec:
+       *   "When Bresenham lines are being rasterized, sample locations may
+       *    all be treated as being at the pixel center (this may affect
+       *    attribute and depth interpolation).""
+       */
+       msaa = false;
+   }
+
    if (dcd0_dirty) {
       struct mali_dcd_flags_0_packed dcd0;
       pan_pack(&dcd0, DCD_FLAGS_0, cfg) {
@@ -1530,11 +1553,14 @@ prepare_dcd(struct panvk_cmd_buffer *cmdbuf)
             cfg.overdraw_alpha1 = true;
          }
 
+         if (rs->line.mode == VK_LINE_RASTERIZATION_MODE_BRESENHAM)
+            cfg.aligned_line_ends = true;
+
          cfg.front_face_ccw = rs->front_face == VK_FRONT_FACE_COUNTER_CLOCKWISE;
          cfg.cull_front_face = (rs->cull_mode & VK_CULL_MODE_FRONT_BIT) != 0;
          cfg.cull_back_face = (rs->cull_mode & VK_CULL_MODE_BACK_BIT) != 0;
 
-         cfg.multisample_enable = dyns->ms.rasterization_samples > 1;
+         cfg.multisample_enable = msaa;
          cfg.occlusion_query = cmdbuf->state.gfx.occlusion_query.mode;
          cfg.alpha_to_coverage = alpha_to_coverage;
       }
@@ -1546,9 +1572,7 @@ prepare_dcd(struct panvk_cmd_buffer *cmdbuf)
    if (dcd1_dirty) {
       struct mali_dcd_flags_1_packed dcd1;
       pan_pack(&dcd1, DCD_FLAGS_1, cfg) {
-         cfg.sample_mask = dyns->ms.rasterization_samples > 1
+         cfg.sample_mask = msaa ? dyns->ms.sample_mask : UINT16_MAX;
-                              ? dyns->ms.sample_mask
-                              : UINT16_MAX;
 
          if (fs) {
             cfg.render_target_mask =

src/panfrost/vulkan/jm/panvk_vX_cmd_draw.c

@@ -185,6 +185,9 @@ panvk_draw_prepare_fs_rsd(struct panvk_cmd_buffer *cmdbuf,
                 dyn_gfx_state_dirty(cmdbuf, RS_DEPTH_CLIP_ENABLE) ||
                 dyn_gfx_state_dirty(cmdbuf, RS_DEPTH_BIAS_ENABLE) ||
                 dyn_gfx_state_dirty(cmdbuf, RS_DEPTH_BIAS_FACTORS) ||
+                dyn_gfx_state_dirty(cmdbuf, RS_LINE_MODE) ||
+                /* line mode needs primitive topology */
+                dyn_gfx_state_dirty(cmdbuf, IA_PRIMITIVE_TOPOLOGY) ||
                 dyn_gfx_state_dirty(cmdbuf, CB_LOGIC_OP_ENABLE) ||
                 dyn_gfx_state_dirty(cmdbuf, CB_LOGIC_OP) ||
                 dyn_gfx_state_dirty(cmdbuf, CB_ATTACHMENT_COUNT) ||
@@ -219,6 +222,7 @@ panvk_draw_prepare_fs_rsd(struct panvk_cmd_buffer *cmdbuf,
    const struct vk_rasterization_state *rs = &dyns->rs;
    const struct vk_color_blend_state *cb = &dyns->cb;
    const struct vk_depth_stencil_state *ds = &dyns->ds;
+   const struct vk_input_assembly_state *ia = &dyns->ia;
    const struct panvk_shader *fs = get_fs(cmdbuf);
    const struct pan_shader_info *fs_info = fs ? &fs->info : NULL;
    unsigned bd_count = MAX2(cb->attachment_count, 1);
@@ -227,6 +231,20 @@ panvk_draw_prepare_fs_rsd(struct panvk_cmd_buffer *cmdbuf,
    bool writes_z = writes_depth(cmdbuf);
    bool writes_s = writes_stencil(cmdbuf);
 
+   bool msaa = dyns->ms.rasterization_samples > 1;
+   if ((ia->primitive_topology == VK_PRIMITIVE_TOPOLOGY_LINE_LIST ||
+        ia->primitive_topology == VK_PRIMITIVE_TOPOLOGY_LINE_STRIP) &&
+       rs->line.mode == VK_LINE_RASTERIZATION_MODE_BRESENHAM) {
+      /* we need to disable MSAA when rendering bresenham lines.
+       *
+       * From the Vulkan spec:
+       *   "When Bresenham lines are being rasterized, sample locations may
+       *    all be treated as being at the pixel center (this may affect
+       *    attribute and depth interpolation).""
+       */
+      msaa = false;
+   }
+
    struct panfrost_ptr ptr = panvk_cmd_alloc_desc_aggregate(
       cmdbuf, PAN_DESC(RENDERER_STATE), PAN_DESC_ARRAY(bd_count, BLEND));
    if (!ptr.gpu)
@@ -251,7 +269,6 @@ panvk_draw_prepare_fs_rsd(struct panvk_cmd_buffer *cmdbuf,
 
    pan_pack(rsd, RENDERER_STATE, cfg) {
       bool alpha_to_coverage = dyns->ms.alpha_to_coverage_enable;
-      bool msaa = dyns->ms.rasterization_samples > 1;
 
       if (fs) {
          pan_shader_prepare_rsd(fs_info, fs_code, &cfg);
@@ -284,7 +301,7 @@ panvk_draw_prepare_fs_rsd(struct panvk_cmd_buffer *cmdbuf,
          cfg.properties.pixel_kill_operation = earlyzs.kill;
          cfg.properties.zs_update_operation = earlyzs.update;
          cfg.multisample_misc.evaluate_per_sample =
-            (fs->info.fs.sample_shading && msaa);
+            (fs->info.fs.sample_shading && dyns->ms.rasterization_samples > 1);
       } else {
          cfg.properties.depth_source = MALI_DEPTH_SOURCE_FIXED_FUNCTION;
          cfg.properties.allow_forward_pixel_to_kill = true;
@@ -314,6 +331,9 @@ panvk_draw_prepare_fs_rsd(struct panvk_cmd_buffer *cmdbuf,
       cfg.stencil_mask_misc.front_facing_depth_bias = rs->depth_bias.enable;
       cfg.stencil_mask_misc.back_facing_depth_bias = rs->depth_bias.enable;
 
+      if (rs->line.mode == VK_LINE_RASTERIZATION_MODE_BRESENHAM)
+         cfg.stencil_mask_misc.aligned_line_ends = true;
+
       cfg.depth_units = rs->depth_bias.constant_factor;
       cfg.depth_factor = rs->depth_bias.slope_factor;
       cfg.depth_bias_clamp = rs->depth_bias.clamp;

src/panfrost/vulkan/panvk_physical_device.c

@@ -202,6 +202,7 @@ get_device_extensions(const struct panvk_physical_device *device,
       .KHR_image_format_list = true,
       .KHR_imageless_framebuffer = true,
       .KHR_index_type_uint8 = true,
+      .KHR_line_rasterization = true,
       .KHR_maintenance1 = true,
       .KHR_maintenance2 = true,
       .KHR_maintenance3 = true,
@@ -251,6 +252,7 @@ get_device_extensions(const struct panvk_physical_device *device,
       .EXT_image_drm_format_modifier = true,
       .EXT_image_robustness = true,
       .EXT_index_type_uint8 = true,
+      .EXT_line_rasterization = true,
       .EXT_physical_device_drm = true,
       .EXT_pipeline_creation_cache_control = true,
       .EXT_pipeline_creation_feedback = true,
@@ -390,6 +392,10 @@ get_features(const struct panvk_physical_device *device,
       .shaderSubgroupRotate = true,
       .shaderSubgroupRotateClustered = true,
 
+      /* VK_KHR_line_rasterization */
+      .rectangularLines = true,
+      .bresenhamLines = true,
+
       /* VK_EXT_graphics_pipeline_library */
       .graphicsPipelineLibrary = true,
 
@@ -854,6 +860,9 @@ get_device_properties(const struct panvk_instance *instance,
       /* XXX: VK_KHR_maintenance4 */
       .maxBufferSize = 1 << 30,
 
+      /* VK_KHR_line_rasterization */
+      .lineSubPixelPrecisionBits = 8,
+
       /* VK_EXT_custom_border_color */
       .maxCustomBorderColorSamplers = 32768,