anv: Make a batch decoder for each queue family

The decoder context needs to know what engine it's associated with.
Nowadays, we have render, compute, blitter, even video engines being
used from the same driver.  Rather than trying to have a single decoder
and thwacking the engine field back and forth between calls, we make
one per queue family, and stash a pointer in anv_queue for easy access.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21149>

src/intel/vulkan/anv_batch_chain.c

@@ -1177,7 +1177,7 @@ anv_cmd_buffer_exec_batch_debug(struct anv_queue *queue,
          uint64_t pass_batch_offset =
             khr_perf_query_preamble_offset(perf_query_pool, perf_query_pass);
 
-         intel_print_batch(&device->decoder_ctx,
+         intel_print_batch(queue->decoder,
                            pass_batch_bo->map + pass_batch_offset, 64,
                            pass_batch_bo->offset + pass_batch_offset, false);
       }
@@ -1185,12 +1185,12 @@ anv_cmd_buffer_exec_batch_debug(struct anv_queue *queue,
       for (uint32_t i = 0; i < cmd_buffer_count; i++) {
          struct anv_batch_bo **bo = u_vector_tail(&cmd_buffers[i]->seen_bbos);
          device->cmd_buffer_being_decoded = cmd_buffers[i];
-         intel_print_batch(&device->decoder_ctx, (*bo)->bo->map,
+         intel_print_batch(queue->decoder, (*bo)->bo->map,
                            (*bo)->bo->size, (*bo)->bo->offset, false);
          device->cmd_buffer_being_decoded = NULL;
       }
    } else {
-      intel_print_batch(&device->decoder_ctx, device->trivial_batch_bo->map,
+      intel_print_batch(queue->decoder, device->trivial_batch_bo->map,
                         device->trivial_batch_bo->size,
                         device->trivial_batch_bo->offset, false);
    }
@@ -1383,7 +1383,7 @@ anv_queue_submit_simple_batch(struct anv_queue *queue,
 #endif
 
    if (INTEL_DEBUG(DEBUG_BATCH)) {
-      intel_print_batch(&device->decoder_ctx,
+      intel_print_batch(queue->decoder,
                         batch_bo->map,
                         batch_bo->size,
                         batch_bo->offset, false);

src/intel/vulkan/anv_device.c

@@ -3200,22 +3200,26 @@ VkResult anv_CreateDevice(
       goto fail_alloc;
 
    if (INTEL_DEBUG(DEBUG_BATCH)) {
-      const unsigned decode_flags =
-         INTEL_BATCH_DECODE_FULL |
-         (INTEL_DEBUG(DEBUG_COLOR) ? INTEL_BATCH_DECODE_IN_COLOR : 0) |
-         INTEL_BATCH_DECODE_OFFSETS |
-         INTEL_BATCH_DECODE_FLOATS;
+      for (unsigned i = 0; i < physical_device->queue.family_count; i++) {
+         struct intel_batch_decode_ctx *decoder = &device->decoder[i];
 
-      intel_batch_decode_ctx_init(&device->decoder_ctx,
-                                  &physical_device->compiler->isa,
-                                  &physical_device->info,
-                                  stderr, decode_flags, NULL,
-                                  decode_get_bo, NULL, device);
+         const unsigned decode_flags =
+            INTEL_BATCH_DECODE_FULL |
+            (INTEL_DEBUG(DEBUG_COLOR) ? INTEL_BATCH_DECODE_IN_COLOR : 0) |
+            INTEL_BATCH_DECODE_OFFSETS |
+            INTEL_BATCH_DECODE_FLOATS;
 
-      device->decoder_ctx.dynamic_base = DYNAMIC_STATE_POOL_MIN_ADDRESS;
-      device->decoder_ctx.surface_base = INTERNAL_SURFACE_STATE_POOL_MIN_ADDRESS;
-      device->decoder_ctx.instruction_base =
-         INSTRUCTION_STATE_POOL_MIN_ADDRESS;
+         intel_batch_decode_ctx_init(decoder,
+                                     &physical_device->compiler->isa,
+                                     &physical_device->info,
+                                     stderr, decode_flags, NULL,
+                                     decode_get_bo, NULL, device);
+
+         decoder->engine = physical_device->queue.families[i].engine_class;
+         decoder->dynamic_base = DYNAMIC_STATE_POOL_MIN_ADDRESS;
+         decoder->surface_base = INTERNAL_SURFACE_STATE_POOL_MIN_ADDRESS;
+         decoder->instruction_base = INSTRUCTION_STATE_POOL_MIN_ADDRESS;
+      }
    }
 
    anv_device_set_physical(device, physical_device);
@@ -3635,6 +3639,8 @@ void anv_DestroyDevice(
    if (!device)
       return;
 
+   struct anv_physical_device *pdevice = device->physical;
+
    anv_device_utrace_finish(device);
 
    anv_device_finish_blorp(device);
@@ -3707,8 +3713,10 @@ void anv_DestroyDevice(
 
    intel_gem_destroy_context(device->fd, device->context_id);
 
-   if (INTEL_DEBUG(DEBUG_BATCH))
-      intel_batch_decode_ctx_finish(&device->decoder_ctx);
+   if (INTEL_DEBUG(DEBUG_BATCH)) {
+      for (unsigned i = 0; i < pdevice->queue.family_count; i++)
+         intel_batch_decode_ctx_finish(&device->decoder[i]);
+   }
 
    close(device->fd);
 

src/intel/vulkan/anv_private.h

@@ -1050,6 +1050,8 @@ struct anv_queue {
 
    const struct anv_queue_family *           family;
 
+   struct intel_batch_decode_ctx *           decoder;
+
    uint32_t                                  exec_flags;
 
    /** Synchronization object for debug purposes (DEBUG_SYNC) */
@@ -1221,7 +1223,7 @@ struct anv_device {
     pthread_mutex_t                             mutex;
     pthread_cond_t                              queue_submit;
 
-    struct intel_batch_decode_ctx               decoder_ctx;
+    struct intel_batch_decode_ctx               decoder[ANV_MAX_QUEUE_FAMILIES];
     /*
      * When decoding a anv_cmd_buffer, we might need to search for BOs through
      * the cmd_buffer's list.

src/intel/vulkan/anv_queue.c

@@ -59,6 +59,8 @@ anv_queue_init(struct anv_device *device, struct anv_queue *queue,
    queue->family = &pdevice->queue.families[queue->vk.queue_family_index];
    queue->exec_flags = exec_flags;
 
+   queue->decoder = &device->decoder[queue->vk.queue_family_index];
+
    return VK_SUCCESS;
 }