v3dv: StencilOp and StencilTestEnable are now dynamic

This commit introduces a significant change when we emit STENCIL_CFG,
with any dynamic state: we stop to use cl_emit_with_prepacked, and use
directly cl_emit. The reason is that now most of the STENCIL_CFG
parameters are dynamic, any improvement of using
cl_emit_with_prepacked is minimized. Also gets the code simpler, and
avoid the need to be extra careful with the fact that
cl_emit_with_prepaked doesn't override values.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27609>

src/broadcom/vulkan/v3dv_cmd_buffer.c

@@ -3006,6 +3006,7 @@ v3dv_cmd_buffer_emit_pre_draw(struct v3dv_cmd_buffer *cmd_buffer,
    if (*dirty & (V3DV_CMD_DIRTY_PIPELINE) ||
        BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_RS_CULL_MODE) ||
        BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_RS_FRONT_FACE) ||
+       BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_TEST_ENABLE) ||
        BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_DEPTH_BOUNDS_TEST_ENABLE)) {
       v3dv_X(device, cmd_buffer_emit_configuration_bits)(cmd_buffer);
    }
@@ -3028,7 +3029,8 @@ v3dv_cmd_buffer_emit_pre_draw(struct v3dv_cmd_buffer *cmd_buffer,
    bool any_dynamic_stencil_dirty =
       BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_COMPARE_MASK) ||
       BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_WRITE_MASK) ||
-      BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_REFERENCE);
+      BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_REFERENCE) ||
+      BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_OP);
 
    if (*dirty & V3DV_CMD_DIRTY_PIPELINE || any_dynamic_stencil_dirty)
       v3dv_X(device, cmd_buffer_emit_stencil)(cmd_buffer);

src/broadcom/vulkan/v3dvx_cmd_buffer.c

@@ -1442,29 +1442,40 @@ v3dX(cmd_buffer_emit_stencil)(struct v3dv_cmd_buffer *cmd_buffer)
    bool any_dynamic_stencil_state =
       BITSET_TEST(dyn->set, MESA_VK_DYNAMIC_DS_STENCIL_COMPARE_MASK) ||
       BITSET_TEST(dyn->set, MESA_VK_DYNAMIC_DS_STENCIL_WRITE_MASK) ||
-      BITSET_TEST(dyn->set, MESA_VK_DYNAMIC_DS_STENCIL_REFERENCE);
+      BITSET_TEST(dyn->set, MESA_VK_DYNAMIC_DS_STENCIL_REFERENCE) ||
+      BITSET_TEST(dyn->set, MESA_VK_DYNAMIC_DS_STENCIL_OP);
 
    bool emitted_stencil = false;
    const struct vk_stencil_test_face_state *front = &dyn->ds.stencil.front;
    const struct vk_stencil_test_face_state *back = &dyn->ds.stencil.back;
 
-   for (uint32_t i = 0; i < 2; i++) {
+   const bool needs_front_and_back = any_dynamic_stencil_state ?
+      memcmp(front, back, sizeof(*front)) != 0 :
+      pipeline->emit_stencil_cfg[1] == true;
+   const unsigned stencil_packets = needs_front_and_back ? 2 : 1;
+
+   for (uint32_t i = 0; i < stencil_packets; i++) {
       if (pipeline->emit_stencil_cfg[i]) {
          if (any_dynamic_stencil_state) {
-            cl_emit_with_prepacked(&job->bcl, STENCIL_CFG,
+            const struct vk_stencil_test_face_state *stencil_state =
-                                   pipeline->stencil_cfg[i], config) {
+               i == 0 ? front : back;
-               if (BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_COMPARE_MASK)) {
+
-                  config.stencil_test_mask =
+            /* If we have any dynamic stencil state we just emit the entire
-                     i == 0 ? front->compare_mask : back->compare_mask;
+             * packet since for simplicity
-               }
+             */
-               if (BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_WRITE_MASK)) {
+            cl_emit(&job->bcl, STENCIL_CFG, config) {
-                  config.stencil_write_mask =
+               config.front_config = !needs_front_and_back || i == 0;
-                     i == 0 ? front->write_mask : back->write_mask;
+               config.back_config = !needs_front_and_back || i == 1;
-               }
+               config.stencil_test_mask = stencil_state->compare_mask & 0xff;
-               if (BITSET_TEST(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_REFERENCE)) {
+               config.stencil_write_mask = stencil_state->write_mask & 0xff;
-                  config.stencil_ref_value =
+               config.stencil_ref_value = stencil_state->reference & 0xff;
-                     i == 0 ? front->reference : back->reference;
+               config.stencil_test_function = stencil_state->op.compare;
-               }
+               config.stencil_pass_op =
+                  v3dX(translate_stencil_op)(stencil_state->op.pass);
+               config.depth_test_fail_op =
+                  v3dX(translate_stencil_op)(stencil_state->op.depth_fail);
+               config.stencil_test_fail_op =
+                  v3dX(translate_stencil_op)(stencil_state->op.fail);
             }
          } else {
             cl_emit_prepacked(&job->bcl, &pipeline->stencil_cfg[i]);
@@ -1476,6 +1487,7 @@ v3dX(cmd_buffer_emit_stencil)(struct v3dv_cmd_buffer *cmd_buffer)
       BITSET_CLEAR(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_COMPARE_MASK);
       BITSET_CLEAR(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_REFERENCE);
       BITSET_CLEAR(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_WRITE_MASK);
+      BITSET_CLEAR(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_OP);
    }
 }
 
@@ -1970,8 +1982,11 @@ v3dX(cmd_buffer_emit_configuration_bits)(struct v3dv_cmd_buffer *cmd_buffer)
    struct vk_dynamic_graphics_state *dyn =
       &cmd_buffer->vk.dynamic_graphics_state;
 
+   /* Disable depth/stencil if we don't have a D/S attachment */
    bool has_depth =
       pipeline->rendering_info.depth_attachment_format != VK_FORMAT_UNDEFINED;
+   bool has_stencil =
+      pipeline->rendering_info.stencil_attachment_format != VK_FORMAT_UNDEFINED;
 
    cl_emit_with_prepacked(&job->bcl, CFG_BITS, pipeline->cfg_bits, config) {
       if (dyn->ds.depth.test_enable && has_depth) {
@@ -1981,6 +1996,8 @@ v3dX(cmd_buffer_emit_configuration_bits)(struct v3dv_cmd_buffer *cmd_buffer)
          config.depth_test_function = VK_COMPARE_OP_ALWAYS;
       }
 
+      config.stencil_enable = dyn->ds.stencil.test_enable && has_stencil;
+
       cmd_buffer->state.z_updates_enable = config.z_updates_enable;
 #if V3D_VERSION == 42
       bool enable_ez = cmd_buffer_update_ez_state(cmd_buffer, pipeline);
@@ -2012,6 +2029,7 @@ v3dX(cmd_buffer_emit_configuration_bits)(struct v3dv_cmd_buffer *cmd_buffer)
    BITSET_CLEAR(dyn->dirty, MESA_VK_DYNAMIC_RS_CULL_MODE);
    BITSET_CLEAR(dyn->dirty, MESA_VK_DYNAMIC_RS_FRONT_FACE);
    BITSET_CLEAR(dyn->dirty, MESA_VK_DYNAMIC_DS_DEPTH_BOUNDS_TEST_ENABLE);
+   BITSET_CLEAR(dyn->dirty, MESA_VK_DYNAMIC_DS_STENCIL_TEST_ENABLE);
 }
 
 void

src/broadcom/vulkan/v3dvx_pipeline.c

@@ -199,13 +199,6 @@ pack_cfg_bits(struct v3dv_pipeline *pipeline,
 
       config.blend_enable = pipeline->blend.enables != 0;
 
-      /* Disable depth/stencil if we don't have a D/S attachment */
-      const struct vk_render_pass_state *ri = &pipeline->rendering_info;
-      bool has_stencil = ri->stencil_attachment_format != VK_FORMAT_UNDEFINED;
-
-      config.stencil_enable =
-         ds_info ? ds_info->stencilTestEnable && has_stencil: false;
-
 #if V3D_VERSION >= 71
       /* From the Vulkan spec:
        *
@@ -240,8 +233,8 @@ pack_cfg_bits(struct v3dv_pipeline *pipeline,
    };
 }
 
-static uint32_t
+uint32_t
-translate_stencil_op(VkStencilOp op)
+v3dX(translate_stencil_op)(VkStencilOp op)
 {
    switch (op) {
    case VK_STENCIL_OP_KEEP:
@@ -283,33 +276,20 @@ pack_single_stencil_cfg(struct v3dv_pipeline *pipeline,
     *
     * In our case, 's' is always 8, so we clamp to that to prevent our packing
     * functions to assert in debug mode if they see larger values.
-    *
-    * If we have dynamic state we need to make sure we set the corresponding
-    * state bits to 0, since cl_emit_with_prepacked ORs the new value with
-    * the old.
     */
-   const uint8_t write_mask =
-      BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_WRITE_MASK) ?
-      0 : stencil_state->writeMask & 0xff;
-
-   const uint8_t compare_mask =
-      BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_COMPARE_MASK) ?
-      0 : stencil_state->compareMask & 0xff;
-
-   const uint8_t reference =
-      BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_REFERENCE) ?
-      0 : stencil_state->reference & 0xff;
-
    v3dvx_pack(stencil_cfg, STENCIL_CFG, config) {
       config.front_config = is_front;
       config.back_config = is_back;
-      config.stencil_write_mask = write_mask;
+      config.stencil_write_mask = stencil_state->writeMask & 0xff;
-      config.stencil_test_mask = compare_mask;
+      config.stencil_test_mask = stencil_state->compareMask & 0xff;
       config.stencil_test_function = stencil_state->compareOp;
-      config.stencil_pass_op = translate_stencil_op(stencil_state->passOp);
+      config.stencil_pass_op =
-      config.depth_test_fail_op = translate_stencil_op(stencil_state->depthFailOp);
+         v3dX(translate_stencil_op)(stencil_state->passOp);
-      config.stencil_test_fail_op = translate_stencil_op(stencil_state->failOp);
+      config.depth_test_fail_op =
-      config.stencil_ref_value = reference;
+         v3dX(translate_stencil_op)(stencil_state->depthFailOp);
+      config.stencil_test_fail_op =
+         v3dX(translate_stencil_op)(stencil_state->failOp);
+      config.stencil_ref_value = stencil_state->reference & 0xff;
    }
 }
 
@@ -320,8 +300,10 @@ pack_stencil_cfg(struct v3dv_pipeline *pipeline,
 {
    assert(sizeof(pipeline->stencil_cfg) == 2 * cl_packet_length(STENCIL_CFG));
 
-   if (!ds_info || !ds_info->stencilTestEnable)
+   if ((!ds_info || !ds_info->stencilTestEnable) &&
+       (!BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_TEST_ENABLE))) {
       return;
+   }
 
    const struct vk_render_pass_state *ri = &pipeline->rendering_info;
    if (ri->stencil_attachment_format == VK_FORMAT_UNDEFINED)
@@ -330,7 +312,9 @@ pack_stencil_cfg(struct v3dv_pipeline *pipeline,
    const bool any_dynamic_stencil_states =
       BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_WRITE_MASK) ||
       BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_COMPARE_MASK) ||
-      BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_REFERENCE);
+      BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_REFERENCE) ||
+      BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_TEST_ENABLE) ||
+      BITSET_TEST(state->dynamic, MESA_VK_DYNAMIC_DS_STENCIL_OP);
 
    /* If front != back or we have dynamic stencil state we can't emit a single
     * packet for both faces.

src/broadcom/vulkan/v3dvx_private.h

@@ -368,3 +368,6 @@ void
 v3dX(viewport_compute_xform)(const VkViewport *viewport,
                              float scale[3],
                              float translate[3]);
+
+uint32_t
+v3dX(translate_stencil_op)(VkStencilOp op);