vk/pipeline: Track VB's that are actually used by the pipeline

Previously, we just blasted out whatever VB's we had marked as "dirty"
regardless of which ones were used by the pipeline.  Given that the stride
of the VB is embedded in the pipeline this can cause problems.  One problem
is if the pipeline doesn't use the given VB binding we emit a bogus stride.
Another problem is that we weren't properly resetting the dirty bits when
the pipeline changed.

src/vulkan/device.c

@@ -2384,8 +2384,10 @@ void anv_CmdBindPipeline(
     VkPipeline                                  _pipeline)
 {
    struct anv_cmd_buffer *cmd_buffer = (struct anv_cmd_buffer *) cmdBuffer;
+   struct anv_pipeline *pipeline = (struct anv_pipeline *) _pipeline;
 
-   cmd_buffer->pipeline = (struct anv_pipeline *) _pipeline;
+   cmd_buffer->pipeline = pipeline;
+   cmd_buffer->vb_dirty |= pipeline->vb_used;
    cmd_buffer->dirty |= ANV_CMD_BUFFER_PIPELINE_DIRTY;
 }
 
@@ -2617,18 +2619,20 @@ anv_cmd_buffer_flush_state(struct anv_cmd_buffer *cmd_buffer)
 {
    struct anv_pipeline *pipeline = cmd_buffer->pipeline;
    struct anv_bindings *bindings = cmd_buffer->bindings;
-   const uint32_t num_buffers = __builtin_popcount(cmd_buffer->vb_dirty);
-   const uint32_t num_dwords = 1 + num_buffers * 4;
    uint32_t *p;
 
-   if (cmd_buffer->vb_dirty) {
+   uint32_t vb_emit = cmd_buffer->vb_dirty & pipeline->vb_used;
+   const uint32_t num_buffers = __builtin_popcount(vb_emit);
+   const uint32_t num_dwords = 1 + num_buffers * 4;
+
+   if (vb_emit) {
       p = anv_batch_emitn(&cmd_buffer->batch, num_dwords,
                           GEN8_3DSTATE_VERTEX_BUFFERS);
       uint32_t vb, i = 0;
-      for_each_bit(vb, cmd_buffer->vb_dirty) {
+      for_each_bit(vb, vb_emit) {
          struct anv_buffer *buffer = bindings->vb[vb].buffer;
          uint32_t offset = bindings->vb[vb].offset;
-      
+
          struct GEN8_VERTEX_BUFFER_STATE state = {
             .VertexBufferIndex = vb,
             .MemoryObjectControlState = 0,
@@ -2653,7 +2657,7 @@ anv_cmd_buffer_flush_state(struct anv_cmd_buffer *cmd_buffer)
       anv_batch_emit_merge(&cmd_buffer->batch,
                            cmd_buffer->rs_state->state_sf, pipeline->state_sf);
 
-   cmd_buffer->vb_dirty = 0;
+   cmd_buffer->vb_dirty &= ~vb_emit;
    cmd_buffer->dirty = 0;
 }
 

src/vulkan/pipeline.c

@@ -62,11 +62,13 @@ emit_vertex_input(struct anv_pipeline *pipeline, VkPipelineVertexInputCreateInfo
    const uint32_t num_dwords = 1 + info->attributeCount * 2;
    uint32_t *p;
    bool instancing_enable[32];
-   
+
+   pipeline->vb_used = 0;
    for (uint32_t i = 0; i < info->bindingCount; i++) {
       const VkVertexInputBindingDescription *desc =
          &info->pVertexBindingDescriptions[i];
-      
+
+      pipeline->vb_used |= 1 << desc->binding;
       pipeline->binding_stride[desc->binding] = desc->strideInBytes;
 
       /* Step rate is programmed per vertex element (attribute), not

src/vulkan/private.h

@@ -610,6 +610,7 @@ struct anv_pipeline {
    uint32_t                                     gs_vec4;
    uint32_t                                     gs_vertex_count;
 
+   uint32_t                                     vb_used;
    uint32_t                                     binding_stride[MAX_VBS];
 
    uint32_t                                     state_sf[GEN8_3DSTATE_SF_length];