anv_device.c

/*
 * Copyright © 2015 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include <assert.h>
#include <stdbool.h>
#include <string.h>
#include <sys/mman.h>
#include <sys/sysinfo.h>
#include <unistd.h>
#include <fcntl.h>
#include <xf86drm.h>
#include <drm_fourcc.h>

#include "anv_private.h"
#include "util/strtod.h"
#include "util/debug.h"
#include "util/build_id.h"
#include "util/disk_cache.h"
#include "util/mesa-sha1.h"
#include "vk_util.h"
#include "common/gen_defines.h"

#include "genxml/gen7_pack.h"

static void
compiler_debug_log(void *data, const char *fmt, ...)
{ }

static void
compiler_perf_log(void *data, const char *fmt, ...)
{
   va_list args;
   va_start(args, fmt);

   if (unlikely(INTEL_DEBUG & DEBUG_PERF))
      intel_logd_v(fmt, args);

   va_end(args);
}

static VkResult
anv_compute_heap_size(int fd, uint64_t gtt_size, uint64_t *heap_size)
{
   /* Query the total ram from the system */
   struct sysinfo info;
   sysinfo(&info);

   uint64_t total_ram = (uint64_t)info.totalram * (uint64_t)info.mem_unit;

   /* We don't want to burn too much ram with the GPU.  If the user has 4GiB
    * or less, we use at most half.  If they have more than 4GiB, we use 3/4.
    */
   uint64_t available_ram;
   if (total_ram <= 4ull * 1024ull * 1024ull * 1024ull)
      available_ram = total_ram / 2;
   else
      available_ram = total_ram * 3 / 4;

   /* We also want to leave some padding for things we allocate in the driver,
    * so don't go over 3/4 of the GTT either.
    */
   uint64_t available_gtt = gtt_size * 3 / 4;

   *heap_size = MIN2(available_ram, available_gtt);

   return VK_SUCCESS;
}

static VkResult
anv_physical_device_init_heaps(struct anv_physical_device *device, int fd)
{
   uint64_t gtt_size;
   if (anv_gem_get_context_param(fd, 0, I915_CONTEXT_PARAM_GTT_SIZE,
                                 &gtt_size) == -1) {
      /* If, for whatever reason, we can't actually get the GTT size from the
       * kernel (too old?) fall back to the aperture size.
       */
      anv_perf_warn(NULL, NULL,
                    "Failed to get I915_CONTEXT_PARAM_GTT_SIZE: %m");

      if (anv_gem_get_aperture(fd, &gtt_size) == -1) {
         return vk_errorf(NULL, NULL, VK_ERROR_INITIALIZATION_FAILED,
                          "failed to get aperture size: %m");
      }
   }

   device->supports_48bit_addresses = (device->info.gen >= 8) &&
      gtt_size > (4ULL << 30 /* GiB */);

   uint64_t heap_size = 0;
   VkResult result = anv_compute_heap_size(fd, gtt_size, &heap_size);
   if (result != VK_SUCCESS)
      return result;

   if (heap_size > (2ull << 30) && !device->supports_48bit_addresses) {
      /* When running with an overridden PCI ID, we may get a GTT size from
       * the kernel that is greater than 2 GiB but the execbuf check for 48bit
       * address support can still fail.  Just clamp the address space size to
       * 2 GiB if we don't have 48-bit support.
       */
      intel_logw("%s:%d: The kernel reported a GTT size larger than 2 GiB but "
                        "not support for 48-bit addresses",
                        __FILE__, __LINE__);
      heap_size = 2ull << 30;
   }

   if (heap_size <= 3ull * (1ull << 30)) {
      /* In this case, everything fits nicely into the 32-bit address space,
       * so there's no need for supporting 48bit addresses on client-allocated
       * memory objects.
       */
      device->memory.heap_count = 1;
      device->memory.heaps[0] = (struct anv_memory_heap) {
         .size = heap_size,
         .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
         .supports_48bit_addresses = false,
      };
   } else {
      /* Not everything will fit nicely into a 32-bit address space.  In this
       * case we need a 64-bit heap.  Advertise a small 32-bit heap and a
       * larger 48-bit heap.  If we're in this case, then we have a total heap
       * size larger than 3GiB which most likely means they have 8 GiB of
       * video memory and so carving off 1 GiB for the 32-bit heap should be
       * reasonable.
       */
      const uint64_t heap_size_32bit = 1ull << 30;
      const uint64_t heap_size_48bit = heap_size - heap_size_32bit;

      assert(device->supports_48bit_addresses);

      device->memory.heap_count = 2;
      device->memory.heaps[0] = (struct anv_memory_heap) {
         .size = heap_size_48bit,
         .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
         .supports_48bit_addresses = true,
      };
      device->memory.heaps[1] = (struct anv_memory_heap) {
         .size = heap_size_32bit,
         .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
         .supports_48bit_addresses = false,
      };
   }

   uint32_t type_count = 0;
   for (uint32_t heap = 0; heap < device->memory.heap_count; heap++) {
      uint32_t valid_buffer_usage = ~0;

      /* There appears to be a hardware issue in the VF cache where it only
       * considers the bottom 32 bits of memory addresses.  If you happen to
       * have two vertex buffers which get placed exactly 4 GiB apart and use
       * them in back-to-back draw calls, you can get collisions.  In order to
       * solve this problem, we require vertex and index buffers be bound to
       * memory allocated out of the 32-bit heap.
       */
      if (device->memory.heaps[heap].supports_48bit_addresses) {
         valid_buffer_usage &= ~(VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
                                 VK_BUFFER_USAGE_VERTEX_BUFFER_BIT);
      }

      if (device->info.has_llc) {
         /* Big core GPUs share LLC with the CPU and thus one memory type can be
          * both cached and coherent at the same time.
          */
         device->memory.types[type_count++] = (struct anv_memory_type) {
            .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
                             VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
                             VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
                             VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
            .heapIndex = heap,
            .valid_buffer_usage = valid_buffer_usage,
         };
      } else {
         /* The spec requires that we expose a host-visible, coherent memory
          * type, but Atom GPUs don't share LLC. Thus we offer two memory types
          * to give the application a choice between cached, but not coherent and
          * coherent but uncached (WC though).
          */
         device->memory.types[type_count++] = (struct anv_memory_type) {
            .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
                             VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
                             VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
            .heapIndex = heap,
            .valid_buffer_usage = valid_buffer_usage,
         };
         device->memory.types[type_count++] = (struct anv_memory_type) {
            .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
                             VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
                             VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
            .heapIndex = heap,
            .valid_buffer_usage = valid_buffer_usage,
         };
      }
   }
   device->memory.type_count = type_count;

   return VK_SUCCESS;
}

static VkResult
anv_physical_device_init_uuids(struct anv_physical_device *device)
{
   const struct build_id_note *note =
      build_id_find_nhdr_for_addr(anv_physical_device_init_uuids);
   if (!note) {
      return vk_errorf(device->instance, device,
                       VK_ERROR_INITIALIZATION_FAILED,
                       "Failed to find build-id");
   }

   unsigned build_id_len = build_id_length(note);
   if (build_id_len < 20) {
      return vk_errorf(device->instance, device,
                       VK_ERROR_INITIALIZATION_FAILED,
                       "build-id too short.  It needs to be a SHA");
   }

   memcpy(device->driver_build_sha1, build_id_data(note), 20);

   struct mesa_sha1 sha1_ctx;
   uint8_t sha1[20];
   STATIC_ASSERT(VK_UUID_SIZE <= sizeof(sha1));

   /* The pipeline cache UUID is used for determining when a pipeline cache is
    * invalid.  It needs both a driver build and the PCI ID of the device.
    */
   _mesa_sha1_init(&sha1_ctx);
   _mesa_sha1_update(&sha1_ctx, build_id_data(note), build_id_len);
   _mesa_sha1_update(&sha1_ctx, &device->chipset_id,
                     sizeof(device->chipset_id));
   _mesa_sha1_final(&sha1_ctx, sha1);
   memcpy(device->pipeline_cache_uuid, sha1, VK_UUID_SIZE);

   /* The driver UUID is used for determining sharability of images and memory
    * between two Vulkan instances in separate processes.  People who want to
    * share memory need to also check the device UUID (below) so all this
    * needs to be is the build-id.
    */
   memcpy(device->driver_uuid, build_id_data(note), VK_UUID_SIZE);

   /* The device UUID uniquely identifies the given device within the machine.
    * Since we never have more than one device, this doesn't need to be a real
    * UUID.  However, on the off-chance that someone tries to use this to
    * cache pre-tiled images or something of the like, we use the PCI ID and
    * some bits of ISL info to ensure that this is safe.
    */
   _mesa_sha1_init(&sha1_ctx);
   _mesa_sha1_update(&sha1_ctx, &device->chipset_id,
                     sizeof(device->chipset_id));
   _mesa_sha1_update(&sha1_ctx, &device->isl_dev.has_bit6_swizzling,
                     sizeof(device->isl_dev.has_bit6_swizzling));
   _mesa_sha1_final(&sha1_ctx, sha1);
   memcpy(device->device_uuid, sha1, VK_UUID_SIZE);

   return VK_SUCCESS;
}

static void
anv_physical_device_init_disk_cache(struct anv_physical_device *device)
{
#ifdef ENABLE_SHADER_CACHE
   char renderer[9];
   MAYBE_UNUSED int len = snprintf(renderer, sizeof(renderer), "anv_%04x",
                                   device->chipset_id);
   assert(len == sizeof(renderer) - 1);

   char timestamp[41];
   _mesa_sha1_format(timestamp, device->driver_build_sha1);

   device->disk_cache = disk_cache_create(renderer, timestamp, 0);
#else
   device->disk_cache = NULL;
#endif
}

static void
anv_physical_device_free_disk_cache(struct anv_physical_device *device)
{
#ifdef ENABLE_SHADER_CACHE
   if (device->disk_cache)
      disk_cache_destroy(device->disk_cache);
#else
   assert(device->disk_cache == NULL);
#endif
}

static VkResult
anv_physical_device_init(struct anv_physical_device *device,
                         struct anv_instance *instance,
                         const char *primary_path,
                         const char *path)
{
   VkResult result;
   int fd;
   int master_fd = -1;

   brw_process_intel_debug_variable();

   fd = open(path, O_RDWR | O_CLOEXEC);
   if (fd < 0)
      return vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);

   device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
   device->instance = instance;

   assert(strlen(path) < ARRAY_SIZE(device->path));
   strncpy(device->path, path, ARRAY_SIZE(device->path));

   device->no_hw = getenv("INTEL_NO_HW") != NULL;

   const int pci_id_override = gen_get_pci_device_id_override();
   if (pci_id_override < 0) {
      device->chipset_id = anv_gem_get_param(fd, I915_PARAM_CHIPSET_ID);
      if (!device->chipset_id) {
         result = vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);
         goto fail;
      }
   } else {
      device->chipset_id = pci_id_override;
      device->no_hw = true;
   }

   device->name = gen_get_device_name(device->chipset_id);
   if (!gen_get_device_info(device->chipset_id, &device->info)) {
      result = vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);
      goto fail;
   }

   if (device->info.is_haswell) {
      intel_logw("Haswell Vulkan support is incomplete");
   } else if (device->info.gen == 7 && !device->info.is_baytrail) {
      intel_logw("Ivy Bridge Vulkan support is incomplete");
   } else if (device->info.gen == 7 && device->info.is_baytrail) {
      intel_logw("Bay Trail Vulkan support is incomplete");
   } else if (device->info.gen >= 8 && device->info.gen <= 10) {
      /* Gen8-10 fully supported */
   } else if (device->info.gen == 11) {
      intel_logw("Vulkan is not yet fully supported on gen11.");
   } else {
      result = vk_errorf(device->instance, device,
                         VK_ERROR_INCOMPATIBLE_DRIVER,
                         "Vulkan not yet supported on %s", device->name);
      goto fail;
   }

   device->cmd_parser_version = -1;
   if (device->info.gen == 7) {
      device->cmd_parser_version =
         anv_gem_get_param(fd, I915_PARAM_CMD_PARSER_VERSION);
      if (device->cmd_parser_version == -1) {
         result = vk_errorf(device->instance, device,
                            VK_ERROR_INITIALIZATION_FAILED,
                            "failed to get command parser version");
         goto fail;
      }
   }

   if (!anv_gem_get_param(fd, I915_PARAM_HAS_WAIT_TIMEOUT)) {
      result = vk_errorf(device->instance, device,
                         VK_ERROR_INITIALIZATION_FAILED,
                         "kernel missing gem wait");
      goto fail;
   }

   if (!anv_gem_get_param(fd, I915_PARAM_HAS_EXECBUF2)) {
      result = vk_errorf(device->instance, device,
                         VK_ERROR_INITIALIZATION_FAILED,
                         "kernel missing execbuf2");
      goto fail;
   }

   if (!device->info.has_llc &&
       anv_gem_get_param(fd, I915_PARAM_MMAP_VERSION) < 1) {
      result = vk_errorf(device->instance, device,
                         VK_ERROR_INITIALIZATION_FAILED,
                         "kernel missing wc mmap");
      goto fail;
   }

   result = anv_physical_device_init_heaps(device, fd);
   if (result != VK_SUCCESS)
      goto fail;

   device->has_exec_async = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_ASYNC);
   device->has_exec_capture = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_CAPTURE);
   device->has_exec_fence = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_FENCE);
   device->has_syncobj = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_FENCE_ARRAY);
   device->has_syncobj_wait = device->has_syncobj &&
                              anv_gem_supports_syncobj_wait(fd);
   device->has_context_priority = anv_gem_has_context_priority(fd);

   device->use_softpin = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_SOFTPIN)
      && device->supports_48bit_addresses;

   device->has_context_isolation =
      anv_gem_get_param(fd, I915_PARAM_HAS_CONTEXT_ISOLATION);

   bool swizzled = anv_gem_get_bit6_swizzle(fd, I915_TILING_X);

   /* Starting with Gen10, the timestamp frequency of the command streamer may
    * vary from one part to another. We can query the value from the kernel.
    */
   if (device->info.gen >= 10) {
      int timestamp_frequency =
         anv_gem_get_param(fd, I915_PARAM_CS_TIMESTAMP_FREQUENCY);

      if (timestamp_frequency < 0)
         intel_logw("Kernel 4.16-rc1+ required to properly query CS timestamp frequency");
      else
         device->info.timestamp_frequency = timestamp_frequency;
   }

   /* GENs prior to 8 do not support EU/Subslice info */
   if (device->info.gen >= 8) {
      device->subslice_total = anv_gem_get_param(fd, I915_PARAM_SUBSLICE_TOTAL);
      device->eu_total = anv_gem_get_param(fd, I915_PARAM_EU_TOTAL);

      /* Without this information, we cannot get the right Braswell
       * brandstrings, and we have to use conservative numbers for GPGPU on
       * many platforms, but otherwise, things will just work.
       */
      if (device->subslice_total < 1 || device->eu_total < 1) {
         intel_logw("Kernel 4.1 required to properly query GPU properties");
      }
   } else if (device->info.gen == 7) {
      device->subslice_total = 1 << (device->info.gt - 1);
   }

   if (device->info.is_cherryview &&
       device->subslice_total > 0 && device->eu_total > 0) {
      /* Logical CS threads = EUs per subslice * num threads per EU */
      uint32_t max_cs_threads =
         device->eu_total / device->subslice_total * device->info.num_thread_per_eu;

      /* Fuse configurations may give more threads than expected, never less. */
      if (max_cs_threads > device->info.max_cs_threads)
         device->info.max_cs_threads = max_cs_threads;
   }

   device->compiler = brw_compiler_create(NULL, &device->info);
   if (device->compiler == NULL) {
      result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
      goto fail;
   }
   device->compiler->shader_debug_log = compiler_debug_log;
   device->compiler->shader_perf_log = compiler_perf_log;
   device->compiler->supports_pull_constants = false;
   device->compiler->constant_buffer_0_is_relative =
      device->info.gen < 8 || !device->has_context_isolation;
   device->compiler->supports_shader_constants = true;

   isl_device_init(&device->isl_dev, &device->info, swizzled);

   result = anv_physical_device_init_uuids(device);
   if (result != VK_SUCCESS)
      goto fail;

   anv_physical_device_init_disk_cache(device);

   if (instance->enabled_extensions.KHR_display) {
      master_fd = open(primary_path, O_RDWR | O_CLOEXEC);
      if (master_fd >= 0) {
         /* prod the device with a GETPARAM call which will fail if
          * we don't have permission to even render on this device
          */
         if (anv_gem_get_param(master_fd, I915_PARAM_CHIPSET_ID) == 0) {
            close(master_fd);
            master_fd = -1;
         }
      }
   }
   device->master_fd = master_fd;

   result = anv_init_wsi(device);
   if (result != VK_SUCCESS) {
      ralloc_free(device->compiler);
      anv_physical_device_free_disk_cache(device);
      goto fail;
   }

   anv_physical_device_get_supported_extensions(device,
                                                &device->supported_extensions);


   device->local_fd = fd;

   return VK_SUCCESS;

fail:
   close(fd);
   if (master_fd != -1)
      close(master_fd);
   return result;
}

static void
anv_physical_device_finish(struct anv_physical_device *device)
{
   anv_finish_wsi(device);
   anv_physical_device_free_disk_cache(device);
   ralloc_free(device->compiler);
   close(device->local_fd);
   if (device->master_fd >= 0)
      close(device->master_fd);
}

static void *
default_alloc_func(void *pUserData, size_t size, size_t align,
                   VkSystemAllocationScope allocationScope)
{
   return malloc(size);
}

static void *
default_realloc_func(void *pUserData, void *pOriginal, size_t size,
                     size_t align, VkSystemAllocationScope allocationScope)
{
   return realloc(pOriginal, size);
}

static void
default_free_func(void *pUserData, void *pMemory)
{
   free(pMemory);
}

static const VkAllocationCallbacks default_alloc = {
   .pUserData = NULL,
   .pfnAllocation = default_alloc_func,
   .pfnReallocation = default_realloc_func,
   .pfnFree = default_free_func,
};

VkResult anv_EnumerateInstanceExtensionProperties(
    const char*                                 pLayerName,
    uint32_t*                                   pPropertyCount,
    VkExtensionProperties*                      pProperties)
{
   VK_OUTARRAY_MAKE(out, pProperties, pPropertyCount);

   for (int i = 0; i < ANV_INSTANCE_EXTENSION_COUNT; i++) {
      if (anv_instance_extensions_supported.extensions[i]) {
         vk_outarray_append(&out, prop) {
            *prop = anv_instance_extensions[i];
         }
      }
   }

   return vk_outarray_status(&out);
}

VkResult anv_CreateInstance(
    const VkInstanceCreateInfo*                 pCreateInfo,
    const VkAllocationCallbacks*                pAllocator,
    VkInstance*                                 pInstance)
{
   struct anv_instance *instance;
   VkResult result;

   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO);

   struct anv_instance_extension_table enabled_extensions = {};
   for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
      int idx;
      for (idx = 0; idx < ANV_INSTANCE_EXTENSION_COUNT; idx++) {
         if (strcmp(pCreateInfo->ppEnabledExtensionNames[i],
                    anv_instance_extensions[idx].extensionName) == 0)
            break;
      }

      if (idx >= ANV_INSTANCE_EXTENSION_COUNT)
         return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);

      if (!anv_instance_extensions_supported.extensions[idx])
         return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);

      enabled_extensions.extensions[idx] = true;
   }

   instance = vk_alloc2(&default_alloc, pAllocator, sizeof(*instance), 8,
                         VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
   if (!instance)
      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);

   instance->_loader_data.loaderMagic = ICD_LOADER_MAGIC;

   if (pAllocator)
      instance->alloc = *pAllocator;
   else
      instance->alloc = default_alloc;

   if (pCreateInfo->pApplicationInfo &&
       pCreateInfo->pApplicationInfo->apiVersion != 0) {
      instance->apiVersion = pCreateInfo->pApplicationInfo->apiVersion;
   } else {
      anv_EnumerateInstanceVersion(&instance->apiVersion);
   }

   instance->enabled_extensions = enabled_extensions;

   for (unsigned i = 0; i < ARRAY_SIZE(instance->dispatch.entrypoints); i++) {
      /* Vulkan requires that entrypoints for extensions which have not been
       * enabled must not be advertised.
       */
      if (!anv_entrypoint_is_enabled(i, instance->apiVersion,
                                     &instance->enabled_extensions, NULL)) {
         instance->dispatch.entrypoints[i] = NULL;
      } else if (anv_dispatch_table.entrypoints[i] != NULL) {
         instance->dispatch.entrypoints[i] = anv_dispatch_table.entrypoints[i];
      } else {
         instance->dispatch.entrypoints[i] =
            anv_tramp_dispatch_table.entrypoints[i];
      }
   }

   instance->physicalDeviceCount = -1;

   result = vk_debug_report_instance_init(&instance->debug_report_callbacks);
   if (result != VK_SUCCESS) {
      vk_free2(&default_alloc, pAllocator, instance);
      return vk_error(result);
   }

   instance->pipeline_cache_enabled =
      env_var_as_boolean("ANV_ENABLE_PIPELINE_CACHE", true);

   _mesa_locale_init();

   VG(VALGRIND_CREATE_MEMPOOL(instance, 0, false));

   *pInstance = anv_instance_to_handle(instance);

   return VK_SUCCESS;
}

void anv_DestroyInstance(
    VkInstance                                  _instance,
    const VkAllocationCallbacks*                pAllocator)
{
   ANV_FROM_HANDLE(anv_instance, instance, _instance);

   if (!instance)
      return;

   if (instance->physicalDeviceCount > 0) {
      /* We support at most one physical device. */
      assert(instance->physicalDeviceCount == 1);
      anv_physical_device_finish(&instance->physicalDevice);
   }

   VG(VALGRIND_DESTROY_MEMPOOL(instance));

   vk_debug_report_instance_destroy(&instance->debug_report_callbacks);

   _mesa_locale_fini();

   vk_free(&instance->alloc, instance);
}

static VkResult
anv_enumerate_devices(struct anv_instance *instance)
{
   /* TODO: Check for more devices ? */
   drmDevicePtr devices[8];
   VkResult result = VK_ERROR_INCOMPATIBLE_DRIVER;
   int max_devices;

   instance->physicalDeviceCount = 0;

   max_devices = drmGetDevices2(0, devices, ARRAY_SIZE(devices));
   if (max_devices < 1)
      return VK_ERROR_INCOMPATIBLE_DRIVER;

   for (unsigned i = 0; i < (unsigned)max_devices; i++) {
      if (devices[i]->available_nodes & 1 << DRM_NODE_RENDER &&
          devices[i]->bustype == DRM_BUS_PCI &&
          devices[i]->deviceinfo.pci->vendor_id == 0x8086) {

         result = anv_physical_device_init(&instance->physicalDevice,
                        instance,
                        devices[i]->nodes[DRM_NODE_PRIMARY],
                        devices[i]->nodes[DRM_NODE_RENDER]);
         if (result != VK_ERROR_INCOMPATIBLE_DRIVER)
            break;
      }
   }
   drmFreeDevices(devices, max_devices);

   if (result == VK_SUCCESS)
      instance->physicalDeviceCount = 1;

   return result;
}

static VkResult
anv_instance_ensure_physical_device(struct anv_instance *instance)
{
   if (instance->physicalDeviceCount < 0) {
      VkResult result = anv_enumerate_devices(instance);
      if (result != VK_SUCCESS &&
          result != VK_ERROR_INCOMPATIBLE_DRIVER)
         return result;
   }

   return VK_SUCCESS;
}

VkResult anv_EnumeratePhysicalDevices(
    VkInstance                                  _instance,
    uint32_t*                                   pPhysicalDeviceCount,
    VkPhysicalDevice*                           pPhysicalDevices)
{
   ANV_FROM_HANDLE(anv_instance, instance, _instance);
   VK_OUTARRAY_MAKE(out, pPhysicalDevices, pPhysicalDeviceCount);

   VkResult result = anv_instance_ensure_physical_device(instance);
   if (result != VK_SUCCESS)
      return result;

   if (instance->physicalDeviceCount == 0)
      return VK_SUCCESS;

   assert(instance->physicalDeviceCount == 1);
   vk_outarray_append(&out, i) {
      *i = anv_physical_device_to_handle(&instance->physicalDevice);
   }

   return vk_outarray_status(&out);
}

VkResult anv_EnumeratePhysicalDeviceGroups(
    VkInstance                                  _instance,
    uint32_t*                                   pPhysicalDeviceGroupCount,
    VkPhysicalDeviceGroupProperties*            pPhysicalDeviceGroupProperties)
{
   ANV_FROM_HANDLE(anv_instance, instance, _instance);
   VK_OUTARRAY_MAKE(out, pPhysicalDeviceGroupProperties,
                         pPhysicalDeviceGroupCount);

   VkResult result = anv_instance_ensure_physical_device(instance);
   if (result != VK_SUCCESS)
      return result;

   if (instance->physicalDeviceCount == 0)
      return VK_SUCCESS;

   assert(instance->physicalDeviceCount == 1);

   vk_outarray_append(&out, p) {
      p->physicalDeviceCount = 1;
      memset(p->physicalDevices, 0, sizeof(p->physicalDevices));
      p->physicalDevices[0] =
         anv_physical_device_to_handle(&instance->physicalDevice);
      p->subsetAllocation = VK_FALSE;

      vk_foreach_struct(ext, p->pNext)
         anv_debug_ignored_stype(ext->sType);
   }

   return vk_outarray_status(&out);
}

void anv_GetPhysicalDeviceFeatures(
    VkPhysicalDevice                            physicalDevice,
    VkPhysicalDeviceFeatures*                   pFeatures)
{
   ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);

   *pFeatures = (VkPhysicalDeviceFeatures) {
      .robustBufferAccess                       = true,
      .fullDrawIndexUint32                      = true,
      .imageCubeArray                           = true,
      .independentBlend                         = true,
      .geometryShader                           = true,
      .tessellationShader                       = true,
      .sampleRateShading                        = true,
      .dualSrcBlend                             = true,
      .logicOp                                  = true,
      .multiDrawIndirect                        = true,
      .drawIndirectFirstInstance                = true,
      .depthClamp                               = true,
      .depthBiasClamp                           = true,
      .fillModeNonSolid                         = true,
      .depthBounds                              = false,
      .wideLines                                = true,
      .largePoints                              = true,
      .alphaToOne                               = true,
      .multiViewport                            = true,
      .samplerAnisotropy                        = true,
      .textureCompressionETC2                   = pdevice->info.gen >= 8 ||
                                                  pdevice->info.is_baytrail,
      .textureCompressionASTC_LDR               = pdevice->info.gen >= 9, /* FINISHME CHV */
      .textureCompressionBC                     = true,
      .occlusionQueryPrecise                    = true,
      .pipelineStatisticsQuery                  = true,
      .fragmentStoresAndAtomics                 = true,
      .shaderTessellationAndGeometryPointSize   = true,
      .shaderImageGatherExtended                = true,
      .shaderStorageImageExtendedFormats        = true,
      .shaderStorageImageMultisample            = false,
      .shaderStorageImageReadWithoutFormat      = false,
      .shaderStorageImageWriteWithoutFormat     = true,
      .shaderUniformBufferArrayDynamicIndexing  = true,
      .shaderSampledImageArrayDynamicIndexing   = true,
      .shaderStorageBufferArrayDynamicIndexing  = true,
      .shaderStorageImageArrayDynamicIndexing   = true,
      .shaderClipDistance                       = true,
      .shaderCullDistance                       = true,
      .shaderFloat64                            = pdevice->info.gen >= 8 &&
                                                  pdevice->info.has_64bit_types,
      .shaderInt64                              = pdevice->info.gen >= 8 &&
                                                  pdevice->info.has_64bit_types,
      .shaderInt16                              = pdevice->info.gen >= 8,
      .shaderResourceMinLod                     = false,
      .variableMultisampleRate                  = true,
      .inheritedQueries                         = true,
   };

   /* We can't do image stores in vec4 shaders */
   pFeatures->vertexPipelineStoresAndAtomics =
      pdevice->compiler->scalar_stage[MESA_SHADER_VERTEX] &&
      pdevice->compiler->scalar_stage[MESA_SHADER_GEOMETRY];
}

void anv_GetPhysicalDeviceFeatures2(
    VkPhysicalDevice                            physicalDevice,
    VkPhysicalDeviceFeatures2*                  pFeatures)
{
   anv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);

   vk_foreach_struct(ext, pFeatures->pNext) {
      switch (ext->sType) {
      case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: {
         VkPhysicalDeviceProtectedMemoryFeatures *features = (void *)ext;
         features->protectedMemory = VK_FALSE;
         break;
      }

      case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES: {
         VkPhysicalDeviceMultiviewFeatures *features =
            (VkPhysicalDeviceMultiviewFeatures *)ext;
         features->multiview = true;
         features->multiviewGeometryShader = true;
         features->multiviewTessellationShader = true;
         break;
      }

      case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTER_FEATURES: {
         VkPhysicalDeviceVariablePointerFeatures *features = (void *)ext;
         features->variablePointersStorageBuffer = true;
         features->variablePointers = true;
         break;
      }

      case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES: {
         VkPhysicalDeviceSamplerYcbcrConversionFeatures *features =
            (VkPhysicalDeviceSamplerYcbcrConversionFeatures *) ext;
         features->samplerYcbcrConversion = true;
         break;
      }

      case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETER_FEATURES: {
         VkPhysicalDeviceShaderDrawParameterFeatures *features = (void *)ext;
         features->shaderDrawParameters = true;
         break;
      }

      case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES_KHR: {
         VkPhysicalDevice16BitStorageFeaturesKHR *features =
            (VkPhysicalDevice16BitStorageFeaturesKHR *)ext;
         ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);

         features->storageBuffer16BitAccess = pdevice->info.gen >= 8;
         features->uniformAndStorageBuffer16BitAccess = pdevice->info.gen >= 8;
         features->storagePushConstant16 = pdevice->info.gen >= 8;
         features->storageInputOutput16 = false;
         break;
      }

      case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR: {
         VkPhysicalDevice8BitStorageFeaturesKHR *features =
            (VkPhysicalDevice8BitStorageFeaturesKHR *)ext;
         ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);

         features->storageBuffer8BitAccess = pdevice->info.gen >= 8;
         features->uniformAndStorageBuffer8BitAccess = pdevice->info.gen >= 8;
         features->storagePushConstant8 = pdevice->info.gen >= 8;
         break;
      }

      default:
         anv_debug_ignored_stype(ext->sType);
         break;
      }
   }
}

void anv_GetPhysicalDeviceProperties(
    VkPhysicalDevice                            physicalDevice,
    VkPhysicalDeviceProperties*                 pProperties)
{
   ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
   const struct gen_device_info *devinfo = &pdevice->info;

   /* See assertions made when programming the buffer surface state. */
   const uint32_t max_raw_buffer_sz = devinfo->gen >= 7 ?
                                      (1ul << 30) : (1ul << 27);

   const uint32_t max_samplers = (devinfo->gen >= 8 || devinfo->is_haswell) ?
                                 128 : 16;

   VkSampleCountFlags sample_counts =
      isl_device_get_sample_counts(&pdevice->isl_dev);

   VkPhysicalDeviceLimits limits = {
      .maxImageDimension1D                      = (1 << 14),
      .maxImageDimension2D                      = (1 << 14),
      .maxImageDimension3D                      = (1 << 11),
      .maxImageDimensionCube                    = (1 << 14),
      .maxImageArrayLayers                      = (1 << 11),
      .maxTexelBufferElements                   = 128 * 1024 * 1024,
      .maxUniformBufferRange                    = (1ul << 27),
      .maxStorageBufferRange                    = max_raw_buffer_sz,
      .maxPushConstantsSize                     = MAX_PUSH_CONSTANTS_SIZE,
      .maxMemoryAllocationCount                 = UINT32_MAX,
      .maxSamplerAllocationCount                = 64 * 1024,
      .bufferImageGranularity                   = 64, /* A cache line */
      .sparseAddressSpaceSize                   = 0,
      .maxBoundDescriptorSets                   = MAX_SETS,
      .maxPerStageDescriptorSamplers            = max_samplers,
      .maxPerStageDescriptorUniformBuffers      = 64,
      .maxPerStageDescriptorStorageBuffers      = 64,
      .maxPerStageDescriptorSampledImages       = max_samplers,
      .maxPerStageDescriptorStorageImages       = 64,
      .maxPerStageDescriptorInputAttachments    = 64,
      .maxPerStageResources                     = 250,
      .maxDescriptorSetSamplers                 = 6 * max_samplers, /* number of stages * maxPerStageDescriptorSamplers */
      .maxDescriptorSetUniformBuffers           = 6 * 64,           /* number of stages * maxPerStageDescriptorUniformBuffers */
      .maxDescriptorSetUniformBuffersDynamic    = MAX_DYNAMIC_BUFFERS / 2,
      .maxDescriptorSetStorageBuffers           = 6 * 64,           /* number of stages * maxPerStageDescriptorStorageBuffers */
      .maxDescriptorSetStorageBuffersDynamic    = MAX_DYNAMIC_BUFFERS / 2,
      .maxDescriptorSetSampledImages            = 6 * max_samplers, /* number of stages * maxPerStageDescriptorSampledImages */
      .maxDescriptorSetStorageImages            = 6 * 64,           /* number of stages * maxPerStageDescriptorStorageImages */
      .maxDescriptorSetInputAttachments         = 256,
      .maxVertexInputAttributes                 = MAX_VBS,
      .maxVertexInputBindings                   = MAX_VBS,
      .maxVertexInputAttributeOffset            = 2047,
      .maxVertexInputBindingStride              = 2048,
      .maxVertexOutputComponents                = 128,
      .maxTessellationGenerationLevel           = 64,
      .maxTessellationPatchSize                 = 32,
      .maxTessellationControlPerVertexInputComponents = 128,
      .maxTessellationControlPerVertexOutputComponents = 128,
      .maxTessellationControlPerPatchOutputComponents = 128,
      .maxTessellationControlTotalOutputComponents = 2048,
      .maxTessellationEvaluationInputComponents = 128,
      .maxTessellationEvaluationOutputComponents = 128,
      .maxGeometryShaderInvocations             = 32,
      .maxGeometryInputComponents               = 64,
      .maxGeometryOutputComponents              = 128,
      .maxGeometryOutputVertices                = 256,
      .maxGeometryTotalOutputComponents         = 1024,
      .maxFragmentInputComponents               = 112, /* 128 components - (POS, PSIZ, CLIP_DIST0, CLIP_DIST1) */
      .maxFragmentOutputAttachments             = 8,
      .maxFragmentDualSrcAttachments            = 1,
      .maxFragmentCombinedOutputResources       = 8,
      .maxComputeSharedMemorySize               = 32768,
      .maxComputeWorkGroupCount                 = { 65535, 65535, 65535 },
      .maxComputeWorkGroupInvocations           = 16 * devinfo->max_cs_threads,
      .maxComputeWorkGroupSize = {
         16 * devinfo->max_cs_threads,
         16 * devinfo->max_cs_threads,
         16 * devinfo->max_cs_threads,
      },
      .subPixelPrecisionBits                    = 4 /* FIXME */,
      .subTexelPrecisionBits                    = 4 /* FIXME */,
      .mipmapPrecisionBits                      = 4 /* FIXME */,
      .maxDrawIndexedIndexValue                 = UINT32_MAX,
      .maxDrawIndirectCount                     = UINT32_MAX,
      .maxSamplerLodBias                        = 16,
      .maxSamplerAnisotropy                     = 16,
      .maxViewports                             = MAX_VIEWPORTS,
      .maxViewportDimensions                    = { (1 << 14), (1 << 14) },
      .viewportBoundsRange                      = { INT16_MIN, INT16_MAX },
      .viewportSubPixelBits                     = 13, /* We take a float? */
      .minMemoryMapAlignment                    = 4096, /* A page */
      .minTexelBufferOffsetAlignment            = 1,
      /* We need 16 for UBO block reads to work and 32 for push UBOs */
      .minUniformBufferOffsetAlignment          = 32,
      .minStorageBufferOffsetAlignment          = 4,
      .minTexelOffset                           = -8,
      .maxTexelOffset                           = 7,
      .minTexelGatherOffset                     = -32,
      .maxTexelGatherOffset                     = 31,
      .minInterpolationOffset                   = -0.5,
      .maxInterpolationOffset                   = 0.4375,
      .subPixelInterpolationOffsetBits          = 4,
      .maxFramebufferWidth                      = (1 << 14),
      .maxFramebufferHeight                     = (1 << 14),
      .maxFramebufferLayers                     = (1 << 11),
      .framebufferColorSampleCounts             = sample_counts,
      .framebufferDepthSampleCounts             = sample_counts,
      .framebufferStencilSampleCounts           = sample_counts,
      .framebufferNoAttachmentsSampleCounts     = sample_counts,
      .maxColorAttachments                      = MAX_RTS,
      .sampledImageColorSampleCounts            = sample_counts,
      .sampledImageIntegerSampleCounts          = VK_SAMPLE_COUNT_1_BIT,
      .sampledImageDepthSampleCounts            = sample_counts,
      .sampledImageStencilSampleCounts          = sample_counts,
      .storageImageSampleCounts                 = VK_SAMPLE_COUNT_1_BIT,
      .maxSampleMaskWords                       = 1,
      .timestampComputeAndGraphics              = false,
      .timestampPeriod                          = 1000000000.0 / devinfo->timestamp_frequency,
      .maxClipDistances                         = 8,
      .maxCullDistances                         = 8,
      .maxCombinedClipAndCullDistances          = 8,
      .discreteQueuePriorities                  = 1,
      .pointSizeRange                           = { 0.125, 255.875 },
      .lineWidthRange                           = { 0.0, 7.9921875 },
      .pointSizeGranularity                     = (1.0 / 8.0),
      .lineWidthGranularity                     = (1.0 / 128.0),
      .strictLines                              = false, /* FINISHME */
      .standardSampleLocations                  = true,
      .optimalBufferCopyOffsetAlignment         = 128,
      .optimalBufferCopyRowPitchAlignment       = 128,
      .nonCoherentAtomSize                      = 64,
   };

   *pProperties = (VkPhysicalDeviceProperties) {
      .apiVersion = anv_physical_device_api_version(pdevice),
      .driverVersion = vk_get_driver_version(),
      .vendorID = 0x8086,
      .deviceID = pdevice->chipset_id,
      .deviceType = VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU,
      .limits = limits,
      .sparseProperties = {0}, /* Broadwell doesn't do sparse. */
   };

   snprintf(pProperties->deviceName, sizeof(pProperties->deviceName),
            "%s", pdevice->name);
   memcpy(pProperties->pipelineCacheUUID,
          pdevice->pipeline_cache_uuid, VK_UUID_SIZE);
}

void anv_GetPhysicalDeviceProperties2(
    VkPhysicalDevice                            physicalDevice,
    VkPhysicalDeviceProperties2*                pProperties)
{
   ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);

   anv_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);

   vk_foreach_struct(ext, pProperties->pNext) {
      switch (ext->sType) {
      case VK_STRU