anv: add dirty tracking for push constant data

This allows us to skip allocating state if it exists already. There are
different scenarios where this can help: when updating only descriptors
(not push constant data) and after blorp or simple shader run.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/10898
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28689>

src/intel/vulkan/anv_cmd_buffer.c

@@ -497,6 +497,7 @@ anv_cmd_buffer_set_ray_query_buffer(struct anv_cmd_buffer *cmd_buffer,
    pipeline_state->push_constants.ray_query_globals =
       anv_address_physical(ray_query_globals_addr);
    cmd_buffer->state.push_constants_dirty |= stages;
+   pipeline_state->push_constants_data_dirty = true;
 }
 
 /**
@@ -696,14 +697,17 @@ void anv_CmdBindPipeline(
                modified = true;
             }
          }
-         if (modified)
+         if (modified) {
             cmd_buffer->state.push_constants_dirty |= stages;
+            state->push_constants_data_dirty = true;
+         }
       }
 
       if ((new_pipeline->fs_msaa_flags & INTEL_MSAA_FLAG_ENABLE_DYNAMIC) &&
           push->gfx.fs_msaa_flags != new_pipeline->fs_msaa_flags) {
          push->gfx.fs_msaa_flags = new_pipeline->fs_msaa_flags;
          cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_FRAGMENT_BIT;
+         state->push_constants_data_dirty = true;
       }
 
       anv_cmd_buffer_flush_pipeline_state(cmd_buffer, old_pipeline, new_pipeline);
@@ -922,6 +926,7 @@ anv_cmd_buffer_bind_descriptor_set(struct anv_cmd_buffer *cmd_buffer,
    else
       cmd_buffer->state.descriptors_dirty |= dirty_stages;
    cmd_buffer->state.push_constants_dirty |= dirty_stages;
+   pipe_state->push_constants_data_dirty = true;
 }
 
 #define ANV_GRAPHICS_STAGE_BITS \
@@ -1306,6 +1311,7 @@ void anv_CmdPushConstants2KHR(
 
       memcpy(pipe_state->push_constants.client_data + pInfo->offset,
              pInfo->pValues, pInfo->size);
+      pipe_state->push_constants_data_dirty = true;
    }
    if (pInfo->stageFlags & VK_SHADER_STAGE_COMPUTE_BIT) {
       struct anv_cmd_pipeline_state *pipe_state =
@@ -1313,6 +1319,7 @@ void anv_CmdPushConstants2KHR(
 
       memcpy(pipe_state->push_constants.client_data + pInfo->offset,
              pInfo->pValues, pInfo->size);
+      pipe_state->push_constants_data_dirty = true;
    }
    if (pInfo->stageFlags & ANV_RT_STAGE_BITS) {
       struct anv_cmd_pipeline_state *pipe_state =
@@ -1320,6 +1327,7 @@ void anv_CmdPushConstants2KHR(
 
       memcpy(pipe_state->push_constants.client_data + pInfo->offset,
              pInfo->pValues, pInfo->size);
+      pipe_state->push_constants_data_dirty = true;
    }
 
    cmd_buffer->state.push_constants_dirty |= pInfo->stageFlags;

src/intel/vulkan/anv_private.h

@@ -3556,6 +3556,9 @@ struct anv_cmd_pipeline_state {
 
    struct anv_push_constants push_constants;
 
+   /** Tracks whether the push constant data has changed and need to be reemitted */
+   bool                                         push_constants_data_dirty;
+
    /* Push constant state allocated when flushing push constants. */
    struct anv_state          push_constants_state;
 
@@ -3743,6 +3746,7 @@ struct anv_cmd_state {
 
    VkShaderStageFlags                           descriptors_dirty;
    VkShaderStageFlags                           push_descriptors_dirty;
+   /** Tracks the 3DSTATE_CONSTANT_* instruction that needs to be reemitted */
    VkShaderStageFlags                           push_constants_dirty;
 
    struct {

src/intel/vulkan/genX_cmd_buffer.c

@@ -394,6 +394,7 @@ genX(cmd_buffer_emit_state_base_address)(struct anv_cmd_buffer *cmd_buffer)
        * state heap. If we change it, we need to reemit the push constants.
        */
       cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_COMPUTE_BIT;
+      cmd_buffer->state.compute.base.push_constants_data_dirty = true;
 #endif
    }
 }
@@ -2751,6 +2752,7 @@ genX(flush_descriptor_buffers)(struct anv_cmd_buffer *cmd_buffer,
       cmd_buffer->state.push_constants_dirty |=
          (cmd_buffer->state.descriptor_buffers.offsets_dirty &
           pipe_state->pipeline->active_stages);
+      pipe_state->push_constants_data_dirty = true;
       cmd_buffer->state.descriptor_buffers.offsets_dirty &=
          ~pipe_state->pipeline->active_stages;
    }
@@ -2943,6 +2945,7 @@ genX(BeginCommandBuffer)(
     * flag them dirty here to make sure they get emitted.
     */
    cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_ALL_GRAPHICS;
+   cmd_buffer->state.gfx.base.push_constants_data_dirty = true;
 
    if (cmd_buffer->usage_flags &
        VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT) {

src/intel/vulkan/genX_cmd_compute.c

@@ -138,6 +138,7 @@ genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer)
        * so flag push constants as dirty if we change the pipeline.
        */
       cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_COMPUTE_BIT;
+      comp_state->base.push_constants_data_dirty = true;
    }
 
    cmd_buffer->state.descriptors_dirty |=
@@ -179,8 +180,13 @@ genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer)
    }
 
    if (cmd_buffer->state.push_constants_dirty & VK_SHADER_STAGE_COMPUTE_BIT) {
-      comp_state->push_data =
-         anv_cmd_buffer_cs_push_constants(cmd_buffer);
+
+      if (comp_state->push_data.alloc_size == 0 ||
+          comp_state->base.push_constants_data_dirty) {
+         comp_state->push_data =
+            anv_cmd_buffer_cs_push_constants(cmd_buffer);
+         comp_state->base.push_constants_data_dirty = false;
+      }
 
 #if GFX_VERx10 < 125
       if (comp_state->push_data.alloc_size) {
@@ -218,6 +224,7 @@ anv_cmd_buffer_push_base_group_id(struct anv_cmd_buffer *cmd_buffer,
       push->cs.base_work_group_id[2] = baseGroupZ;
 
       cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_COMPUTE_BIT;
+      cmd_buffer->state.compute.base.push_constants_data_dirty = true;
    }
 }
 

src/intel/vulkan/genX_cmd_draw.c

@@ -486,10 +486,12 @@ cmd_buffer_flush_gfx_push_constants(struct anv_cmd_buffer *cmd_buffer,
       }
    }
 
-   /* Resets the push constant state so that we allocate a new one if
-    * needed.
+    /* Setting NULL resets the push constant state so that we allocate a new one
+    * if needed. If push constant data not dirty, get_push_range_address can
+    * re-use existing allocation.
     */
-   gfx_state->base.push_constants_state = ANV_STATE_NULL;
+   if (gfx_state->base.push_constants_data_dirty)
+      gfx_state->base.push_constants_state = ANV_STATE_NULL;
 
    anv_foreach_stage(stage, dirty_stages) {
       unsigned buffer_count = 0;
@@ -560,6 +562,7 @@ cmd_buffer_flush_gfx_push_constants(struct anv_cmd_buffer *cmd_buffer,
 #endif
 
    cmd_buffer->state.push_constants_dirty &= ~flushed;
+   gfx_state->base.push_constants_data_dirty = false;
 }
 
 #if GFX_VERx10 >= 125

src/intel/vulkan/genX_gfx_state.c

@@ -1343,6 +1343,7 @@ genX(cmd_buffer_flush_gfx_runtime_state)(struct anv_cmd_buffer *cmd_buffer)
        push->gfx.tcs_input_vertices != dyn->ts.patch_control_points) {
       push->gfx.tcs_input_vertices = dyn->ts.patch_control_points;
       cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT;
+      gfx->base.push_constants_data_dirty = true;
    }
 
 #undef GET