Merge branch 'master' into arcpatch-D16436

# Conflicts:
#	source/blender/draw/engines/overlay/shaders/infos/overlay_outline_info.hh
#	source/blender/draw/engines/overlay/shaders/infos/overlay_wireframe_info.hh
#	source/blender/gpu/intern/gpu_shader_create_info.cc
#	source/blender/gpu/intern/gpu_shader_create_info.hh
#	source/blender/gpu/metal/kernels/depth_2d_update_info.hh
#	source/blender/gpu/metal/mtl_shader_generator.mm

source/blender/draw/CMakeLists.txt

@@ -477,9 +477,11 @@ set(GLSL_SRC
   engines/workbench/shaders/workbench_prepass_pointcloud_vert.glsl
   engines/workbench/shaders/workbench_prepass_vert.glsl
   engines/workbench/shaders/workbench_shadow_caps_geom.glsl
+  engines/workbench/shaders/workbench_shadow_caps_vert_no_geom.glsl
   engines/workbench/shaders/workbench_shadow_debug_frag.glsl
   engines/workbench/shaders/workbench_shadow_geom.glsl
   engines/workbench/shaders/workbench_shadow_vert.glsl
+  engines/workbench/shaders/workbench_shadow_vert_no_geom.glsl
   engines/workbench/shaders/workbench_transparent_accum_frag.glsl
   engines/workbench/shaders/workbench_transparent_resolve_frag.glsl
   engines/workbench/shaders/workbench_volume_frag.glsl

source/blender/draw/engines/workbench/shaders/infos/workbench_shadow_info.hh

@@ -13,12 +13,17 @@ GPU_SHADER_INTERFACE_INFO(workbench_shadow_iface, "vData")
 
 GPU_SHADER_CREATE_INFO(workbench_shadow_common)
     .vertex_in(0, Type::VEC3, "pos")
-    .vertex_out(workbench_shadow_iface)
     .push_constant(Type::FLOAT, "lightDistance")
     .push_constant(Type::VEC3, "lightDirection")
-    .vertex_source("workbench_shadow_vert.glsl")
     .additional_info("draw_mesh");
 
+/* `workbench_shadow_vert.glsl` only used by geometry shader path.
+ * Vertex output iface not needed by non-geometry shader variants,
+ * as only gl_Position is returned. */
+GPU_SHADER_CREATE_INFO(workbench_shadow_common_geom)
+    .vertex_out(workbench_shadow_iface)
+    .vertex_source("workbench_shadow_vert.glsl");
+
 /** \} */
 
 /* -------------------------------------------------------------------- */
@@ -26,13 +31,25 @@ GPU_SHADER_CREATE_INFO(workbench_shadow_common)
  * \{ */
 
 GPU_SHADER_CREATE_INFO(workbench_shadow_manifold)
+    .additional_info("workbench_shadow_common_geom")
     .geometry_layout(PrimitiveIn::LINES_ADJACENCY, PrimitiveOut::TRIANGLE_STRIP, 4, 1)
     .geometry_source("workbench_shadow_geom.glsl");
 
 GPU_SHADER_CREATE_INFO(workbench_shadow_no_manifold)
+    .additional_info("workbench_shadow_common_geom")
     .geometry_layout(PrimitiveIn::LINES_ADJACENCY, PrimitiveOut::TRIANGLE_STRIP, 4, 2)
     .geometry_source("workbench_shadow_geom.glsl");
 
+GPU_SHADER_CREATE_INFO(workbench_shadow_manifold_no_geom)
+    .vertex_source("workbench_shadow_vert_no_geom.glsl")
+    /* Inject SSBO vertex fetch declaration using 2 output triangles. */
+    .define("VAR_MANIFOLD", "\n#pragma USE_SSBO_VERTEX_FETCH(TriangleList, 6)");
+
+GPU_SHADER_CREATE_INFO(workbench_shadow_no_manifold_no_geom)
+    .vertex_source("workbench_shadow_vert_no_geom.glsl")
+    /* Inject SSBO vertex fetch declaration using 4 output triangles. */
+    .define("VAR_NO_MANIFOLD", "\n#pragma USE_SSBO_VERTEX_FETCH(TriangleList, 12)");
+
 /** \} */
 
 /* -------------------------------------------------------------------- */
@@ -40,9 +57,13 @@ GPU_SHADER_CREATE_INFO(workbench_shadow_no_manifold)
  * \{ */
 
 GPU_SHADER_CREATE_INFO(workbench_shadow_caps)
+    .additional_info("workbench_shadow_common_geom")
     .geometry_layout(PrimitiveIn::TRIANGLES, PrimitiveOut::TRIANGLE_STRIP, 3, 2)
     .geometry_source("workbench_shadow_caps_geom.glsl");
 
+GPU_SHADER_CREATE_INFO(workbench_shadow_caps_no_geom)
+    .vertex_source("workbench_shadow_caps_vert_no_geom.glsl");
+
 /** \} */
 
 /* -------------------------------------------------------------------- */

source/blender/draw/engines/workbench/shaders/workbench_shadow_caps_vert_no_geom.glsl

@@ -0,0 +1,105 @@
+#pragma BLENDER_REQUIRE(common_view_lib.glsl)
+#pragma USE_SSBO_VERTEX_FETCH(TriangleList, 6)
+
+#ifdef DOUBLE_MANIFOLD
+#  define vert_len 6 /* Triangle Strip with 6 verts = 4 triangles = 12 verts*/
+#  define emit_triangle_count 2
+#else
+#  define vert_len 6 /* Triangle Strip with 6 verts = 4 triangles = 12 verts*/
+#  define emit_triangle_count 2
+#endif
+
+struct VertexData {
+  vec3 pos;           /* local position */
+  vec4 frontPosition; /* final ndc position */
+  vec4 backPosition;
+};
+
+/* Input geometry triangle list. */
+VertexData vData[3] = {};
+
+#define DISCARD_VERTEX \
+  gl_Position = vec4(0.0); \
+  return;
+
+vec4 get_pos(int v, bool backface)
+{
+  return (backface) ? vData[v].backPosition : vData[v].frontPosition;
+}
+
+void emit_cap(const bool front, bool reversed, int triangle_vertex_id)
+{
+  /* Inverse. */
+  ivec2 idx = (reversed) ? ivec2(2, 1) : ivec2(1, 2);
+
+  /* Output position depending on vertex ID. */
+  switch (triangle_vertex_id) {
+    case 0: {
+      gl_Position = (front) ? vData[0].frontPosition : vData[0].backPosition;
+    } break;
+
+    case 1: {
+      gl_Position = (front) ? vData[idx.x].frontPosition : vData[idx.y].backPosition;
+    } break;
+
+    case 2: {
+      gl_Position = (front) ? vData[idx.y].frontPosition : vData[idx.x].backPosition;
+    } break;
+  }
+
+  /* Apply depth bias. Prevents Z-fighting artifacts when fast-math is enabled. */
+  gl_Position.z += 0.00005;
+}
+
+void main()
+{
+  /* Output Data indexing. */
+  int input_prim_index = int(gl_VertexID / 6);
+  int output_vertex_id = gl_VertexID % 6;
+  int output_triangle_id = output_vertex_id / 3;
+
+  /* Source primitive data location derived from output primitive. */
+  int input_base_vertex_id = input_prim_index * 3;
+
+  /* In data is triangles - Should be guaranteed. */
+  /* Read input position data. */
+  vData[0].pos = vertex_fetch_attribute(input_base_vertex_id + 0, pos, vec3);
+  vData[1].pos = vertex_fetch_attribute(input_base_vertex_id + 1, pos, vec3);
+  vData[2].pos = vertex_fetch_attribute(input_base_vertex_id + 2, pos, vec3);
+
+  /* Calculate front/back Positions. */
+  vData[0].frontPosition = point_object_to_ndc(vData[0].pos);
+  vData[0].backPosition = point_object_to_ndc(vData[0].pos + lightDirection * lightDistance);
+
+  vData[1].frontPosition = point_object_to_ndc(vData[1].pos);
+  vData[1].backPosition = point_object_to_ndc(vData[1].pos + lightDirection * lightDistance);
+
+  vData[2].frontPosition = point_object_to_ndc(vData[2].pos);
+  vData[2].backPosition = point_object_to_ndc(vData[2].pos + lightDirection * lightDistance);
+
+  /* Geometry shader equivalent calc. */
+  vec3 v10 = vData[0].pos - vData[1].pos;
+  vec3 v12 = vData[2].pos - vData[1].pos;
+
+  vec3 n = cross(v12, v10);
+  float facing = dot(n, lightDirection);
+
+  bool backface = facing > 0.0;
+
+#ifdef DOUBLE_MANIFOLD
+  /* In case of non manifold geom, we only increase/decrease
+   * the stencil buffer by one but do every faces as they were facing the light. */
+  bool invert = backface;
+  const bool is_manifold = false;
+#else
+  const bool invert = false;
+  const bool is_manifold = true;
+#endif
+
+  if (!is_manifold || !backface) {
+    bool do_front = (output_triangle_id==0)?true:false;
+    emit_cap(do_front, invert, output_vertex_id%3);
+  } else {
+    DISCARD_VERTEX
+  }
+}

source/blender/draw/engines/workbench/shaders/workbench_shadow_vert_no_geom.glsl

@@ -0,0 +1,160 @@
+#pragma BLENDER_REQUIRE(common_view_lib.glsl)
+
+/* Two variants, split pass, generates either 2 triangles or 6 triangles depending on input
+ * geometry manifold type */
+
+#ifdef DOUBLE_MANIFOLD
+#  define vert_len 12 /* Triangle Strip with 6 verts = 4 triangles = 12 verts*/
+#  define emit_triangle_count 4
+#else
+#  define vert_len 6 /* Triang;e Strip with 4 verts = 2 triangles = 6 verts*/
+#  define emit_triangle_count 2
+#endif
+
+#define DEGENERATE_TRIS_WORKAROUND
+#define DEGENERATE_TRIS_AREA_THRESHOLD 4e-15
+
+#define len_sqr(a) dot(a, a)
+
+struct VertexData {
+  vec3 pos;           /* local position */
+  vec4 frontPosition; /* final ndc position */
+  vec4 backPosition;
+};
+
+#define DISCARD_VERTEX \
+  gl_Position = vec4(0.0); \
+  return;
+
+/* Extra positions for primary triangle */
+VertexData vData[4];
+
+void extrude_edge(bool invert, int output_vertex_id)
+{
+  /* Reverse order if backfacing the light. */
+  ivec2 idx = (invert) ? ivec2(1, 2) : ivec2(2, 1);
+
+  /* Either outputs first or second quad, depending on double manifold status. */
+  int triangle_vertex_id = output_vertex_id % 6;
+
+  switch (triangle_vertex_id) {
+    case 0:
+      gl_Position = vData[idx.x].frontPosition;
+      break;
+    case 1:
+    case 4:
+      gl_Position = vData[idx.y].frontPosition;
+      break;
+    case 2:
+    case 3:
+      gl_Position = vData[idx.x].backPosition;
+      break;
+    case 5:
+      gl_Position = vData[idx.y].backPosition;
+      break;
+  }
+
+  /* Apply depth bias. Prevents Z-fighting artifacts when fast-math is enabled. */
+  gl_Position.z += 0.00005;
+}
+
+void main()
+{
+  /* Output Data indexing. */
+#ifdef DOUBLE_MANIFOLD
+  int input_prim_index = int(gl_VertexID / 12);
+  int output_vertex_id = gl_VertexID % 12;
+  int output_quad_id = (output_vertex_id >= 6) ? 1 : 0;
+#else
+  int input_prim_index = int(gl_VertexID / 6);
+  int output_vertex_id = gl_VertexID % 6;
+  int output_quad_id = 0;
+#endif
+
+  /* Source primitive data location derived from output primitive. */
+  int input_base_vertex_id = (input_prim_index * 4);
+
+  /* IN DATA is lines_adjacency - Should be guaranteed. */
+  /* Read input position data. */
+  vData[0].pos = vertex_fetch_attribute(input_base_vertex_id + 0, pos, vec3);
+  vData[1].pos = vertex_fetch_attribute(input_base_vertex_id + 1, pos, vec3);
+  vData[2].pos = vertex_fetch_attribute(input_base_vertex_id + 2, pos, vec3);
+  vData[3].pos = vertex_fetch_attribute(input_base_vertex_id + 3, pos, vec3);
+
+  /* Calculate front/back Positions. */
+  vData[0].frontPosition = point_object_to_ndc(vData[0].pos);
+  vData[0].backPosition = point_object_to_ndc(vData[0].pos + lightDirection * lightDistance);
+
+  vData[1].frontPosition = point_object_to_ndc(vData[1].pos);
+  vData[1].backPosition = point_object_to_ndc(vData[1].pos + lightDirection * lightDistance);
+
+  vData[2].frontPosition = point_object_to_ndc(vData[2].pos);
+  vData[2].backPosition = point_object_to_ndc(vData[2].pos + lightDirection * lightDistance);
+
+  vData[3].frontPosition = point_object_to_ndc(vData[3].pos);
+  vData[3].backPosition = point_object_to_ndc(vData[3].pos + lightDirection * lightDistance);
+
+  /* Geometry shader equivalent path. */
+  vec3 v10 = vData[0].pos - vData[1].pos;
+  vec3 v12 = vData[2].pos - vData[1].pos;
+  vec3 v13 = vData[3].pos - vData[1].pos;
+  vec3 n1 = cross(v12, v10);
+  vec3 n2 = cross(v13, v12);
+
+#ifdef DEGENERATE_TRIS_WORKAROUND
+  /* Check if area is null. */
+  vec2 faces_area = vec2(len_sqr(n1), len_sqr(n2));
+  bvec2 degen_faces = lessThan(abs(faces_area), vec2(DEGENERATE_TRIS_AREA_THRESHOLD));
+
+  /* Both triangles are degenerate, abort. */
+  if (all(degen_faces)) {
+    DISCARD_VERTEX
+  }
+#endif
+
+  vec2 facing = vec2(dot(n1, lightDirection), dot(n2, lightDirection));
+
+  /* WATCH: maybe unpredictable in some cases. */
+  bool is_manifold = any(notEqual(vData[0].pos, vData[3].pos));
+
+  bvec2 backface = greaterThan(facing, vec2(0.0));
+
+#ifdef DEGENERATE_TRIS_WORKAROUND
+#  ifndef DOUBLE_MANIFOLD
+  /* If the mesh is known to be manifold and we don't use double count,
+   * only create an quad if the we encounter a facing geom. */
+  if ((degen_faces.x && backface.y) || (degen_faces.y && backface.x)) {
+    DISCARD_VERTEX
+  }
+#  endif
+
+  /* If one of the 2 triangles is degenerate, replace edge by a non-manifold one. */
+  backface.x = (degen_faces.x) ? !backface.y : backface.x;
+  backface.y = (degen_faces.y) ? !backface.x : backface.y;
+  is_manifold = (any(degen_faces)) ? false : is_manifold;
+#endif
+
+  /* If both faces face the same direction it's not an outline edge. */
+  if (backface.x == backface.y) {
+    DISCARD_VERTEX
+  }
+
+  /* Output correct output vertex depending on */
+  if (output_vertex_id < 6) {
+    /* Always output if first 6 vertices. */
+    extrude_edge(backface.x, output_vertex_id);
+  }
+  else {
+#ifdef DOUBLE_MANIFOLD
+    /* If part of double manifold, only output this vertex if correctly contributing */
+    if (is_manifold) {
+      extrude_edge(backface.x, output_vertex_id);
+    }
+    else {
+      DISCARD_VERTEX
+    }
+#else
+    DISCARD_VERTEX
+#endif
+  }
+}

source/blender/gpu/intern/gpu_shader_create_info.cc

@@ -75,9 +75,9 @@ void ShaderCreateInfo::finalize()
     if (info.early_fragment_test_) {
       early_fragment_test_ = true;
     }
-    /* Override depth-write with additional info if this specifies a writing mode
+    /* Modify depth write if has been changed from default.
-     * other than the default. */
+     * `UNCHANGED` implies gl_FragDepth is not used at all. */
-    if (info.depth_write_ != DepthWrite::NONE) {
+    if (info.depth_write_ != DepthWrite::UNCHANGED) {
       depth_write_ = info.depth_write_;
     }
 
@@ -343,6 +343,13 @@ void gpu_shader_create_info_init()
     overlay_motion_path_line = overlay_motion_path_line_no_geom;
     overlay_motion_path_line_clipped = overlay_motion_path_line_clipped_no_geom;
 
+    /* Workbench shadows.
+     * Note: Updates additional-info used by workbench shadow permutations.
+     * Must be prepared prior to permutation preparation. */
+    workbench_shadow_manifold = workbench_shadow_manifold_no_geom;
+    workbench_shadow_no_manifold = workbench_shadow_no_manifold_no_geom;
+    workbench_shadow_caps = workbench_shadow_caps_no_geom;
+
     /* Conservative rasterization. */
     basic_depth_mesh_conservative = basic_depth_mesh_conservative_no_geom;
     basic_depth_mesh_conservative_clipped = basic_depth_mesh_conservative_no_geom_clipped;

source/blender/gpu/intern/gpu_shader_create_info.hh

@@ -189,8 +189,8 @@ ENUM_OPERATORS(BuiltinBits, BuiltinBits::USE_DEBUG_PRINT);
  * https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_conservative_depth.txt
  */
 enum class DepthWrite {
-  /* NONE specified as default to indicate gl_FragDepth is not used. */
+  /* UNCHANGED specified as default to indicate gl_FragDepth is not used. */
-  NONE = 0,
+  UNCHANGED = 0,
   ANY,
   GREATER,
   LESS,
@@ -344,7 +344,7 @@ struct ShaderCreateInfo {
   /** If true, force the use of the GL shader introspection for resource location. */
   bool legacy_resource_location_ = false;
   /** Allow optimization when fragment shader writes to `gl_FragDepth`. */
-  DepthWrite depth_write_ = DepthWrite::NONE;
+  DepthWrite depth_write_ = DepthWrite::UNCHANGED;
   /** GPU Backend compatibility flag. Temporary requirement until Metal enablement is fully
    * complete. */
   bool metal_backend_only_ = false;

source/blender/gpu/metal/kernels/depth_2d_update_info.hh

@@ -24,19 +24,19 @@ GPU_SHADER_CREATE_INFO(depth_2d_update_float)
     .metal_backend_only(true)
     .fragment_source("depth_2d_update_float_frag.glsl")
     .additional_info("depth_2d_update_info_base")
-    // .do_static_compilation(true)
+    .do_static_compilation(true)
     .depth_write(DepthWrite::ANY);
 
 GPU_SHADER_CREATE_INFO(depth_2d_update_int24)
     .metal_backend_only(true)
     .fragment_source("depth_2d_update_int24_frag.glsl")
     .additional_info("depth_2d_update_info_base")
-    // .do_static_compilation(true)
+    .do_static_compilation(true)
     .depth_write(DepthWrite::ANY);
 
 GPU_SHADER_CREATE_INFO(depth_2d_update_int32)
     .metal_backend_only(true)
     .fragment_source("depth_2d_update_int32_frag.glsl")
     .additional_info("depth_2d_update_info_base")
-    // .do_static_compilation(true)
+    .do_static_compilation(true)
     .depth_write(DepthWrite::ANY);

source/blender/gpu/metal/mtl_batch.mm

@@ -472,21 +472,6 @@ id<MTLRenderCommandEncoder> MTLBatch::bind(uint v_first, uint v_count, uint i_fi
                                                   this->elem ? @"(indexed)" : @"",
                                                   active_shader_->get_interface()->get_name()]];
   }
-
-  /* Ensure Context Render Pipeline State is fully setup and ready to execute the draw. */
-  MTLPrimitiveType mtl_prim_type = gpu_prim_type_to_metal(this->prim_type);
-  if (!ctx->ensure_render_pipeline_state(mtl_prim_type)) {
-    printf("FAILED TO ENSURE RENDER PIPELINE STATE");
-    BLI_assert(false);
-
-    if (G.debug & G_DEBUG_GPU) {
-      [rec popDebugGroup];
-    }
-    return nil;
-  }
-
-  /*** Bind Vertex Buffers and Index Buffers **/
-
   /* SSBO Vertex Fetch Buffer bindings. */
   if (uses_ssbo_fetch) {
 
@@ -514,7 +499,7 @@ id<MTLRenderCommandEncoder> MTLBatch::bind(uint v_first, uint v_count, uint i_fi
     }
     rps.bind_vertex_buffer(idx_buffer, 0, MTL_SSBO_VERTEX_FETCH_IBO_INDEX);
 
-    /* Ensure all attributes are set */
+    /* Ensure all attributes are set. */
     active_shader_->ssbo_vertex_fetch_bind_attributes_end(rec);
 
     /* Bind NULL Buffers for unused vertex data slots. */
@@ -529,7 +514,7 @@ id<MTLRenderCommandEncoder> MTLBatch::bind(uint v_first, uint v_count, uint i_fi
     /* Flag whether Indexed rendering is used or not. */
     int &uniform_ssbo_use_indexed = active_shader_->uni_ssbo_uses_indexed_rendering;
     BLI_assert(uniform_ssbo_use_indexed != -1);
-    int uses_indexed_rendering = (mtl_elem != NULL) ? 1 : 0;
+    int uses_indexed_rendering = (mtl_elem != nullptr) ? 1 : 0;
     active_shader_->uniform_int(uniform_ssbo_use_indexed, 1, 1, &uses_indexed_rendering);
 
     /* Set SSBO-fetch-mode status uniforms. */
@@ -547,6 +532,19 @@ id<MTLRenderCommandEncoder> MTLBatch::bind(uint v_first, uint v_count, uint i_fi
                                   (const int *)(&v_count));
   }
 
+  /* Ensure Context Render Pipeline State is fully setup and ready to execute the draw.
+   * This should happen after all other final rendeirng setup is complete. */
+  MTLPrimitiveType mtl_prim_type = gpu_prim_type_to_metal(this->prim_type);
+  if (!ctx->ensure_render_pipeline_state(mtl_prim_type)) {
+    printf("FAILED TO ENSURE RENDER PIPELINE STATE");
+    BLI_assert(false);
+
+    if (G.debug & G_DEBUG_GPU) {
+      [rec popDebugGroup];
+    }
+    return nil;
+  }
+
   /* Bind Vertex Buffers. */
   for (int i = 0; i < num_buffers; i++) {
     MTLVertBuf *buf_at_index = buffers[i];
@@ -755,6 +753,10 @@ void MTLBatch::draw_advanced(int v_first, int v_count, int i_first, int i_count)
         mtl_vertex_count_fits_primitive_type(
             output_num_verts, active_shader_->get_ssbo_vertex_fetch_output_prim_type()),
         "Output Vertex count is not compatible with the requested output vertex primitive type");
+
+    /* Set depth stencil state (requires knowledge of primitive type). */
+    ctx->ensure_depth_stencil_state(active_shader_->get_ssbo_vertex_fetch_output_prim_type());
+
     [rec drawPrimitives:active_shader_->get_ssbo_vertex_fetch_output_prim_type()
             vertexStart:0
             vertexCount:output_num_verts

source/blender/gpu/metal/mtl_shader_generator.mm

@@ -675,7 +675,7 @@ bool MTLShader::generate_msl_from_glsl(const shader::ShaderCreateInfo *info)
 
     /* NOTE(Metal): FragDepth output mode specified in create-info 'DepthWrite depth_write_'.
      * If parsing without create-info, manual extraction will be required. */
-    msl_iface.uses_gl_FragDepth = (info->depth_write_ != DepthWrite::NONE) &&
+    msl_iface.uses_gl_FragDepth = (info->depth_write_ != DepthWrite::UNCHANGED) &&
                                   shd_builder_->glsl_fragment_source_.find("gl_FragDepth") !=
                                       std::string::npos;
     msl_iface.depth_write = info->depth_write_;

source/blender/gpu/metal/mtl_state.mm

@@ -282,27 +282,25 @@ void MTLStateManager::set_stencil_test(const eGPUStencilTest test, const eGPUSte
           context_, MTLStencilOperationKeep, MTLStencilOperationKeep, MTLStencilOperationReplace);
       break;
     case GPU_STENCIL_OP_COUNT_DEPTH_PASS:
-      /* Winding inversed due to flipped Y coordinate system in Metal. */
       mtl_stencil_set_op_separate(context_,
-                                  GPU_CULL_FRONT,
+                                  GPU_CULL_BACK,
                                   MTLStencilOperationKeep,
                                   MTLStencilOperationKeep,
                                   MTLStencilOperationIncrementWrap);
       mtl_stencil_set_op_separate(context_,
-                                  GPU_CULL_BACK,
+                                  GPU_CULL_FRONT,
                                   MTLStencilOperationKeep,
                                   MTLStencilOperationKeep,
                                   MTLStencilOperationDecrementWrap);
       break;
     case GPU_STENCIL_OP_COUNT_DEPTH_FAIL:
-      /* Winding inversed due to flipped Y coordinate system in Metal. */
       mtl_stencil_set_op_separate(context_,
-                                  GPU_CULL_FRONT,
+                                  GPU_CULL_BACK,
                                   MTLStencilOperationKeep,
                                   MTLStencilOperationDecrementWrap,
                                   MTLStencilOperationKeep);
       mtl_stencil_set_op_separate(context_,
-                                  GPU_CULL_BACK,
+                                  GPU_CULL_FRONT,
                                   MTLStencilOperationKeep,
                                   MTLStencilOperationIncrementWrap,
                                   MTLStencilOperationKeep);