Merge branch 'kk-post-depth-coverage' into 'main'

kk: Support VK_EXT_post_depth_coverage

See merge request mesa/mesa!41283

docs/features.txt

@@ -671,7 +671,7 @@ Khronos extensions that are not part of any Vulkan version:
   VK_EXT_pci_bus_info                                   DONE (anv, hasvk, nvk, radv, vn)
   VK_EXT_physical_device_drm                            DONE (anv, hasvk, hk, nvk, panvk, pvr, radv, tu, v3dv, vn)
   VK_EXT_pipeline_library_group_handles                 DONE (anv, lvp, radv, vn)
-  VK_EXT_post_depth_coverage                            DONE (anv/gfx11+, lvp, nvk, radv/gfx10+, tu, vn)
+  VK_EXT_post_depth_coverage                            DONE (anv/gfx11+, kk, lvp, nvk, radv/gfx10+, tu, vn)
   VK_EXT_present_timing                                 DONE (anv, hk, nvk, radv, tu, panvk)
   VK_EXT_primitive_restart_index                        DONE (lvp, radv)
   VK_EXT_primitive_topology_list_restart                DONE (anv, hasvk, lvp, nvk, panvk, radv, tu, v3dv, vn, nvk)

src/kosmickrisp/compiler/msl_nir_lower_common.c

@@ -369,6 +369,15 @@ lower_sample_shading(nir_builder *b, nir_intrinsic_instr *intr, void *data)
       return true;
    }
 
+   if (intr->intrinsic == nir_intrinsic_load_sample_mask_in) {
+      b->cursor = nir_after_instr(&intr->instr);
+      nir_def *sample_id = nir_load_sample_id(b);
+      nir_def *sample_bit = nir_ishl(b, nir_imm_int(b, 1), sample_id);
+      nir_def *sample_mask_bit = nir_iand(b, &intr->def, sample_bit);
+      nir_def_rewrite_uses_after(&intr->def, sample_mask_bit);
+      return true;
+   }
+
    if (intr->intrinsic != nir_intrinsic_load_interpolated_input)
       return false;
 

src/kosmickrisp/compiler/nir_to_msl.c

@@ -62,14 +62,24 @@ static const char *sysval_table[SYSTEM_VALUE_MAX] = {
       "uint mtl_AmplificationID [[amplification_id]]",
    [SYSTEM_VALUE_FIRST_VERTEX] = "uint gl_FirstVertex [[base_vertex]]",
 };
+static const char *sysval_sample_mask_in_post_depth_coverage =
+   "uint gl_SampleMask [[sample_mask, post_depth_coverage]]";
 
 static void
 emit_sysvals(struct nir_to_msl_ctx *ctx, nir_shader *shader)
 {
    unsigned i;
    BITSET_FOREACH_SET(i, shader->info.system_values_read, SYSTEM_VALUE_MAX) {
-      assert(sysval_table[i]);
-      P_IND(ctx, "%s,\n", sysval_table[i]);
+      const char *sysval;
+      if (i == SYSTEM_VALUE_SAMPLE_MASK_IN &&
+          ctx->shader->info.stage == MESA_SHADER_FRAGMENT &&
+          ctx->shader->info.fs.post_depth_coverage)
+         sysval = sysval_sample_mask_in_post_depth_coverage;
+      else
+         sysval = sysval_table[i];
+
+      assert(sysval);
+      P_IND(ctx, "%s,\n", sysval);
    }
 }
 
@@ -1005,7 +1015,7 @@ intrinsic_to_msl(struct nir_to_msl_ctx *ctx, nir_intrinsic_instr *instr)
       P(ctx, "gl_SampleID;\n");
       break;
    case nir_intrinsic_load_sample_mask_in:
-      P(ctx, "gl_SampleMask & (1u << gl_SampleID);\n");
+      P(ctx, "gl_SampleMask;\n");
       break;
    case nir_intrinsic_load_sample_pos:
       P(ctx, "get_sample_position(gl_SampleID);\n");

src/kosmickrisp/vulkan/kk_physical_device.c

@@ -143,6 +143,7 @@ kk_get_device_extensions(const struct kk_instance *instance,
       .EXT_image_2d_view_of_3d = true,
       .EXT_load_store_op_none = true,
       .EXT_mutable_descriptor_type = true,
+      .EXT_post_depth_coverage = true,
       .EXT_shader_atomic_float = true,
       .EXT_shader_replicated_composites = true,
 

src/kosmickrisp/vulkan/kk_shader.c

@@ -127,6 +127,7 @@ struct kk_fs_key {
    struct vk_depth_stencil_state ds;
    uint32_t rasterization_samples;
    uint16_t static_sample_mask;
+   bool sample_shading_enable;
    bool has_depth;
 };
 
@@ -151,6 +152,7 @@ kk_populate_fs_key(struct kk_fs_key *key,
    if (state->ms) {
       key->rasterization_samples = state->ms->rasterization_samples;
       key->static_sample_mask = state->ms->sample_mask;
+      key->sample_shading_enable = state->ms->sample_shading_enable;
    }
 
    /* Depth writes are removed unless there's an actual attachment */
@@ -400,6 +402,9 @@ static void
 kk_lower_fs(struct kk_device *dev, nir_shader *nir,
             const struct vk_graphics_pipeline_state *state)
 {
+   nir->info.fs.uses_sample_shading |= state->ms &&
+                                       state->ms->sample_shading_enable;
+
    /* msl_nir_lower_sample_shading needs to go before blending since
     * nir_lower_blend will always set uses_sample_shading to true if there's any
     * output read. I believe we do not need to lower it always, that is why it
@@ -440,8 +445,12 @@ kk_lower_fs(struct kk_device *dev, nir_shader *nir,
          if (!nir->info.fs.early_fragment_tests) {
             nir_function_impl *entrypoint = nir_shader_get_entrypoint(nir);
             nir_builder b = nir_builder_at(nir_after_impl(entrypoint));
-            nir_discard_if(&b,
-                           nir_ieq_imm(&b, nir_load_sample_mask_in(&b), 0u));
+
+            nir_def *sample_id = nir_load_sample_id(&b);
+            nir_def *sample_bit = nir_ishl(&b, nir_imm_int(&b, 1), sample_id);
+            nir_def *sample_mask_bit = nir_iand(&b, nir_load_sample_mask_in(&b),
+                                                sample_bit);
+            nir_discard_if(&b, nir_ieq_imm(&b, sample_mask_bit, 0u));
          }
       }
       NIR_PASS(_, nir, msl_lower_static_sample_mask, state->ms->sample_mask);