anv: add support for INTEL_DEBUG=bat

As requested by Ken ;)

v2: Also decode simple batches (Caio)
    Fix u_vector usage issues (Lionel)

v3: Make binding/instruction/state/surface available (Lionel)

v4: Going through device pools for simple batches (Lionel)
    Centralize search BO callbacks into anv_device.c (Lionel)

v5: Clear decoded batch buffer var after use (Caio)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

src/intel/vulkan/anv_batch_chain.c

@@ -1729,10 +1729,20 @@ anv_cmd_buffer_execbuf(struct anv_device *device,
       }
    }
 
-   if (cmd_buffer)
+   if (cmd_buffer) {
+      if (unlikely(INTEL_DEBUG & DEBUG_BATCH)) {
+         struct anv_batch_bo **bo = u_vector_head(&cmd_buffer->seen_bbos);
+
+         device->cmd_buffer_being_decoded = cmd_buffer;
+         gen_print_batch(&device->decoder_ctx, (*bo)->bo.map,
+                         (*bo)->bo.size, (*bo)->bo.offset);
+         device->cmd_buffer_being_decoded = NULL;
+      }
+
       result = setup_execbuf_for_cmd_buffer(&execbuf, cmd_buffer);
-   else
+   } else {
       result = setup_empty_execbuf(&execbuf, device);
+   }
 
    if (result != VK_SUCCESS)
       return result;

src/intel/vulkan/anv_device.c

@@ -1735,6 +1735,63 @@ anv_device_init_hiz_clear_value_bo(struct anv_device *device)
    anv_gem_munmap(map, device->hiz_clear_bo.size);
 }
 
+static bool
+get_bo_from_pool(struct gen_batch_decode_bo *ret,
+                 struct anv_block_pool *pool,
+                 uint64_t address)
+{
+   for (uint32_t i = 0; i < pool->nbos; i++) {
+      uint64_t bo_address = pool->bos[i].offset & (~0ull >> 16);
+      uint32_t bo_size = pool->bos[i].size;
+      if (address >= bo_address && address < (bo_address + bo_size)) {
+         *ret = (struct gen_batch_decode_bo) {
+            .addr = bo_address,
+            .size = bo_size,
+            .map = pool->bos[i].map,
+         };
+         return true;
+      }
+   }
+   return false;
+}
+
+/* Finding a buffer for batch decoding */
+static struct gen_batch_decode_bo
+decode_get_bo(void *v_batch, uint64_t address)
+{
+   struct anv_device *device = v_batch;
+   struct gen_batch_decode_bo ret_bo = {};
+
+   if (get_bo_from_pool(&ret_bo, &device->dynamic_state_pool.block_pool, address))
+      return ret_bo;
+   if (get_bo_from_pool(&ret_bo, &device->instruction_state_pool.block_pool, address))
+      return ret_bo;
+   if (get_bo_from_pool(&ret_bo, &device->binding_table_pool.block_pool, address))
+      return ret_bo;
+   if (get_bo_from_pool(&ret_bo, &device->surface_state_pool.block_pool, address))
+      return ret_bo;
+
+   if (!device->cmd_buffer_being_decoded)
+      return (struct gen_batch_decode_bo) { };
+
+   struct anv_batch_bo **bo;
+
+   u_vector_foreach(bo, &device->cmd_buffer_being_decoded->seen_bbos) {
+      /* The decoder zeroes out the top 16 bits, so we need to as well */
+      uint64_t bo_address = (*bo)->bo.offset & (~0ull >> 16);
+
+      if (address >= bo_address && address < bo_address + (*bo)->bo.size) {
+         return (struct gen_batch_decode_bo) {
+            .addr = bo_address,
+            .size = (*bo)->bo.size,
+            .map = (*bo)->bo.map,
+         };
+      }
+   }
+
+   return (struct gen_batch_decode_bo) { };
+}
+
 VkResult anv_CreateDevice(
     VkPhysicalDevice                            physicalDevice,
     const VkDeviceCreateInfo*                   pCreateInfo,
@@ -1802,6 +1859,17 @@ VkResult anv_CreateDevice(
    if (!device)
       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 
+   const unsigned decode_flags =
+      GEN_BATCH_DECODE_FULL |
+      ((INTEL_DEBUG & DEBUG_COLOR) ? GEN_BATCH_DECODE_IN_COLOR : 0) |
+      GEN_BATCH_DECODE_OFFSETS |
+      GEN_BATCH_DECODE_FLOATS;
+
+   gen_batch_decode_ctx_init(&device->decoder_ctx,
+                             &physical_device->info,
+                             stderr, decode_flags, NULL,
+                             decode_get_bo, NULL, device);
+
    device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
    device->instance = physical_device->instance;
    device->chipset_id = physical_device->chipset_id;
@@ -2089,6 +2157,8 @@ void anv_DestroyDevice(
 
    anv_gem_destroy_context(device, device->context_id);
 
+   gen_batch_decode_ctx_finish(&device->decoder_ctx);
+
    close(device->fd);
 
    vk_free(&device->alloc, device);

src/intel/vulkan/anv_private.h

@@ -44,6 +44,7 @@
 #endif
 
 #include "common/gen_clflush.h"
+#include "common/gen_decoder.h"
 #include "common/gen_gem.h"
 #include "dev/gen_device_info.h"
 #include "blorp/blorp.h"
@@ -1120,6 +1121,13 @@ struct anv_device {
     pthread_mutex_t                             mutex;
     pthread_cond_t                              queue_submit;
     bool                                        _lost;
+
+    struct gen_batch_decode_ctx                 decoder_ctx;
+    /*
+     * When decoding a anv_cmd_buffer, we might need to search for BOs through
+     * the cmd_buffer's list.
+     */
+    struct anv_cmd_buffer                      *cmd_buffer_being_decoded;
 };
 
 static inline struct anv_state_pool *

src/intel/vulkan/anv_queue.c

@@ -100,6 +100,9 @@ anv_device_submit_simple_batch(struct anv_device *device,
    execbuf.rsvd1 = device->context_id;
    execbuf.rsvd2 = 0;
 
+   if (unlikely(INTEL_DEBUG & DEBUG_BATCH))
+      gen_print_batch(&device->decoder_ctx, bo.map, bo.size, bo.offset);
+
    result = anv_device_execbuf(device, &execbuf, exec_bos);
    if (result != VK_SUCCESS)
       goto fail;