mesa/src/vulkan/anv_cmd_buffer.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 <unistd.h>
#include <fcntl.h>

#include "anv_private.h"

/** \file anv_cmd_buffer.c
 *
 * This file contains all of the stuff for emitting commands into a command
 * buffer.  This includes implementations of most of the vkCmd*
 * entrypoints.  This file is concerned entirely with state emission and
 * not with the command buffer data structure itself.  As far as this file
 * is concerned, most of anv_cmd_buffer is magic.
 */

static void
anv_cmd_state_init(struct anv_cmd_state *state)
{
   state->rs_state = NULL;
   state->vp_state = NULL;
   state->cb_state = NULL;
   state->ds_state = NULL;
   memset(&state->state_vf, 0, sizeof(state->state_vf));
   memset(&state->descriptors, 0, sizeof(state->descriptors));

   state->dirty = 0;
   state->vb_dirty = 0;
   state->descriptors_dirty = 0;
   state->pipeline = NULL;
   state->vp_state = NULL;
   state->rs_state = NULL;
   state->ds_state = NULL;
}

VkResult anv_CreateCommandBuffer(
    VkDevice                                    _device,
    const VkCmdBufferCreateInfo*                pCreateInfo,
    VkCmdBuffer*                                pCmdBuffer)
{
   ANV_FROM_HANDLE(anv_device, device, _device);
   ANV_FROM_HANDLE(anv_cmd_pool, pool, pCreateInfo->cmdPool);
   struct anv_cmd_buffer *cmd_buffer;
   VkResult result;

   cmd_buffer = anv_device_alloc(device, sizeof(*cmd_buffer), 8,
                                 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
   if (cmd_buffer == NULL)
      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);

   cmd_buffer->device = device;

   result = anv_cmd_buffer_init_batch_bo_chain(cmd_buffer);
   if (result != VK_SUCCESS)
      goto fail;

   anv_state_stream_init(&cmd_buffer->surface_state_stream,
                         &device->surface_state_block_pool);
   anv_state_stream_init(&cmd_buffer->dynamic_state_stream,
                         &device->dynamic_state_block_pool);

   cmd_buffer->level = pCreateInfo->level;
   cmd_buffer->opt_flags = 0;

   anv_cmd_state_init(&cmd_buffer->state);

   list_addtail(&cmd_buffer->pool_link, &pool->cmd_buffers);

   *pCmdBuffer = anv_cmd_buffer_to_handle(cmd_buffer);

   return VK_SUCCESS;

 fail: anv_device_free(device, cmd_buffer);

   return result;
}

VkResult anv_DestroyCommandBuffer(
    VkDevice                                    _device,
    VkCmdBuffer                                 _cmd_buffer)
{
   ANV_FROM_HANDLE(anv_device, device, _device);
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, _cmd_buffer);

   list_del(&cmd_buffer->pool_link);

   anv_cmd_buffer_fini_batch_bo_chain(cmd_buffer);

   anv_state_stream_finish(&cmd_buffer->surface_state_stream);
   anv_state_stream_finish(&cmd_buffer->dynamic_state_stream);
   anv_device_free(device, cmd_buffer);

   return VK_SUCCESS;
}

VkResult anv_ResetCommandBuffer(
    VkCmdBuffer                                 cmdBuffer,
    VkCmdBufferResetFlags                       flags)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);

   anv_cmd_buffer_reset_batch_bo_chain(cmd_buffer);

   anv_cmd_state_init(&cmd_buffer->state);

   return VK_SUCCESS;
}

void
anv_cmd_buffer_emit_state_base_address(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_device *device = cmd_buffer->device;
   struct anv_bo *scratch_bo = NULL;

   cmd_buffer->state.scratch_size =
      anv_block_pool_size(&device->scratch_block_pool);
   if (cmd_buffer->state.scratch_size > 0)
      scratch_bo = &device->scratch_block_pool.bo;

   anv_batch_emit(&cmd_buffer->batch, GEN8_STATE_BASE_ADDRESS,
                  .GeneralStateBaseAddress = { scratch_bo, 0 },
                  .GeneralStateMemoryObjectControlState = GEN8_MOCS,
                  .GeneralStateBaseAddressModifyEnable = true,
                  .GeneralStateBufferSize = 0xfffff,
                  .GeneralStateBufferSizeModifyEnable = true,

                  .SurfaceStateBaseAddress = { anv_cmd_buffer_current_surface_bo(cmd_buffer), 0 },
                  .SurfaceStateMemoryObjectControlState = GEN8_MOCS,
                  .SurfaceStateBaseAddressModifyEnable = true,

                  .DynamicStateBaseAddress = { &device->dynamic_state_block_pool.bo, 0 },
                  .DynamicStateMemoryObjectControlState = GEN8_MOCS,
                  .DynamicStateBaseAddressModifyEnable = true,
                  .DynamicStateBufferSize = 0xfffff,
                  .DynamicStateBufferSizeModifyEnable = true,

                  .IndirectObjectBaseAddress = { NULL, 0 },
                  .IndirectObjectMemoryObjectControlState = GEN8_MOCS,
                  .IndirectObjectBaseAddressModifyEnable = true,
                  .IndirectObjectBufferSize = 0xfffff,
                  .IndirectObjectBufferSizeModifyEnable = true,

                  .InstructionBaseAddress = { &device->instruction_block_pool.bo, 0 },
                  .InstructionMemoryObjectControlState = GEN8_MOCS,
                  .InstructionBaseAddressModifyEnable = true,
                  .InstructionBufferSize = 0xfffff,
                  .InstructionBuffersizeModifyEnable = true);

   /* After re-setting the surface state base address, we have to do some
    * cache flusing so that the sampler engine will pick up the new
    * SURFACE_STATE objects and binding tables. From the Broadwell PRM,
    * Shared Function > 3D Sampler > State > State Caching (page 96):
    *
    *    Coherency with system memory in the state cache, like the texture
    *    cache is handled partially by software. It is expected that the
    *    command stream or shader will issue Cache Flush operation or
    *    Cache_Flush sampler message to ensure that the L1 cache remains
    *    coherent with system memory.
    *
    *    [...]
    *
    *    Whenever the value of the Dynamic_State_Base_Addr,
    *    Surface_State_Base_Addr are altered, the L1 state cache must be
    *    invalidated to ensure the new surface or sampler state is fetched
    *    from system memory.
    *
    * The PIPE_CONTROL command has a "State Cache Invalidation Enable" bit
    * which, according the PIPE_CONTROL instruction documentation in the
    * Broadwell PRM:
    *
    *    Setting this bit is independent of any other bit in this packet.
    *    This bit controls the invalidation of the L1 and L2 state caches
    *    at the top of the pipe i.e. at the parsing time.
    *
    * Unfortunately, experimentation seems to indicate that state cache
    * invalidation through a PIPE_CONTROL does nothing whatsoever in
    * regards to surface state and binding tables.  In stead, it seems that
    * invalidating the texture cache is what is actually needed.
    *
    * XXX:  As far as we have been able to determine through
    * experimentation, shows that flush the texture cache appears to be
    * sufficient.  The theory here is that all of the sampling/rendering
    * units cache the binding table in the texture cache.  However, we have
    * yet to be able to actually confirm this.
    */
   anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
                  .TextureCacheInvalidationEnable = true);
}

VkResult anv_BeginCommandBuffer(
    VkCmdBuffer                                 cmdBuffer,
    const VkCmdBufferBeginInfo*                 pBeginInfo)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);

   cmd_buffer->opt_flags = pBeginInfo->flags;

   if (cmd_buffer->level == VK_CMD_BUFFER_LEVEL_SECONDARY) {
      cmd_buffer->state.framebuffer =
         anv_framebuffer_from_handle(pBeginInfo->framebuffer);
      cmd_buffer->state.pass =
         anv_render_pass_from_handle(pBeginInfo->renderPass);

      /* FIXME: We shouldn't be starting on the first subpass */
      anv_cmd_buffer_begin_subpass(cmd_buffer,
                                   &cmd_buffer->state.pass->subpasses[0]);
   }

   anv_cmd_buffer_emit_state_base_address(cmd_buffer);
   cmd_buffer->state.current_pipeline = UINT32_MAX;

   return VK_SUCCESS;
}

VkResult anv_EndCommandBuffer(
    VkCmdBuffer                                 cmdBuffer)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   struct anv_device *device = cmd_buffer->device;

   anv_cmd_buffer_end_batch_buffer(cmd_buffer);

   if (cmd_buffer->level == VK_CMD_BUFFER_LEVEL_PRIMARY) {
      /* The algorithm used to compute the validate list is not threadsafe as
       * it uses the bo->index field.  We have to lock the device around it.
       * Fortunately, the chances for contention here are probably very low.
       */
      pthread_mutex_lock(&device->mutex);
      anv_cmd_buffer_prepare_execbuf(cmd_buffer);
      pthread_mutex_unlock(&device->mutex);
   }

   return VK_SUCCESS;
}

void anv_CmdBindPipeline(
    VkCmdBuffer                                 cmdBuffer,
    VkPipelineBindPoint                         pipelineBindPoint,
    VkPipeline                                  _pipeline)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_pipeline, pipeline, _pipeline);

   switch (pipelineBindPoint) {
   case VK_PIPELINE_BIND_POINT_COMPUTE:
      cmd_buffer->state.compute_pipeline = pipeline;
      cmd_buffer->state.compute_dirty |= ANV_CMD_BUFFER_PIPELINE_DIRTY;
      break;

   case VK_PIPELINE_BIND_POINT_GRAPHICS:
      cmd_buffer->state.pipeline = pipeline;
      cmd_buffer->state.vb_dirty |= pipeline->vb_used;
      cmd_buffer->state.dirty |= ANV_CMD_BUFFER_PIPELINE_DIRTY;
      break;

   default:
      assert(!"invalid bind point");
      break;
   }
}

void anv_CmdBindDynamicViewportState(
    VkCmdBuffer                                 cmdBuffer,
    VkDynamicViewportState                      dynamicViewportState)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_dynamic_vp_state, vp_state, dynamicViewportState);

   cmd_buffer->state.vp_state = vp_state;
   cmd_buffer->state.dirty |= ANV_CMD_BUFFER_VP_DIRTY;
}

void anv_CmdBindDynamicRasterState(
    VkCmdBuffer                                 cmdBuffer,
    VkDynamicRasterState                        dynamicRasterState)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_dynamic_rs_state, rs_state, dynamicRasterState);

   cmd_buffer->state.rs_state = rs_state;
   cmd_buffer->state.dirty |= ANV_CMD_BUFFER_RS_DIRTY;
}

void anv_CmdBindDynamicColorBlendState(
    VkCmdBuffer                                 cmdBuffer,
    VkDynamicColorBlendState                    dynamicColorBlendState)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_dynamic_cb_state, cb_state, dynamicColorBlendState);

   cmd_buffer->state.cb_state = cb_state;
   cmd_buffer->state.dirty |= ANV_CMD_BUFFER_CB_DIRTY;
}

void anv_CmdBindDynamicDepthStencilState(
    VkCmdBuffer                                 cmdBuffer,
    VkDynamicDepthStencilState                  dynamicDepthStencilState)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_dynamic_ds_state, ds_state, dynamicDepthStencilState);

   cmd_buffer->state.ds_state = ds_state;
   cmd_buffer->state.dirty |= ANV_CMD_BUFFER_DS_DIRTY;
}

void anv_CmdBindDescriptorSets(
    VkCmdBuffer                                 cmdBuffer,
    VkPipelineBindPoint                         pipelineBindPoint,
    VkPipelineLayout                            _layout,
    uint32_t                                    firstSet,
    uint32_t                                    setCount,
    const VkDescriptorSet*                      pDescriptorSets,
    uint32_t                                    dynamicOffsetCount,
    const uint32_t*                             pDynamicOffsets)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_pipeline_layout, layout, _layout);
   struct anv_descriptor_set_layout *set_layout;

   assert(firstSet + setCount < MAX_SETS);

   uint32_t dynamic_slot = 0;
   for (uint32_t i = 0; i < setCount; i++) {
      ANV_FROM_HANDLE(anv_descriptor_set, set, pDescriptorSets[i]);
      set_layout = layout->set[firstSet + i].layout;

      cmd_buffer->state.descriptors[firstSet + i].set = set;

      assert(set_layout->num_dynamic_buffers <
             ARRAY_SIZE(cmd_buffer->state.descriptors[0].dynamic_offsets));
      memcpy(cmd_buffer->state.descriptors[firstSet + i].dynamic_offsets,
             pDynamicOffsets + dynamic_slot,
             set_layout->num_dynamic_buffers * sizeof(*pDynamicOffsets));

      cmd_buffer->state.descriptors_dirty |= set_layout->shader_stages;

      dynamic_slot += set_layout->num_dynamic_buffers;
   }
}

void anv_CmdBindIndexBuffer(
    VkCmdBuffer                                 cmdBuffer,
    VkBuffer                                    _buffer,
    VkDeviceSize                                offset,
    VkIndexType                                 indexType)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);

   static const uint32_t vk_to_gen_index_type[] = {
      [VK_INDEX_TYPE_UINT16]                    = INDEX_WORD,
      [VK_INDEX_TYPE_UINT32]                    = INDEX_DWORD,
   };

   struct GEN8_3DSTATE_VF vf = {
      GEN8_3DSTATE_VF_header,
      .CutIndex = (indexType == VK_INDEX_TYPE_UINT16) ? UINT16_MAX : UINT32_MAX,
   };
   GEN8_3DSTATE_VF_pack(NULL, cmd_buffer->state.state_vf, &vf);

   cmd_buffer->state.dirty |= ANV_CMD_BUFFER_INDEX_BUFFER_DIRTY;

   anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_INDEX_BUFFER,
                  .IndexFormat = vk_to_gen_index_type[indexType],
                  .MemoryObjectControlState = GEN8_MOCS,
                  .BufferStartingAddress = { buffer->bo, buffer->offset + offset },
                  .BufferSize = buffer->size - offset);
}

void anv_CmdBindVertexBuffers(
    VkCmdBuffer                                 cmdBuffer,
    uint32_t                                    startBinding,
    uint32_t                                    bindingCount,
    const VkBuffer*                             pBuffers,
    const VkDeviceSize*                         pOffsets)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   struct anv_vertex_binding *vb = cmd_buffer->state.vertex_bindings;

   /* We have to defer setting up vertex buffer since we need the buffer
    * stride from the pipeline. */

   assert(startBinding + bindingCount < MAX_VBS);
   for (uint32_t i = 0; i < bindingCount; i++) {
      vb[startBinding + i].buffer = anv_buffer_from_handle(pBuffers[i]);
      vb[startBinding + i].offset = pOffsets[i];
      cmd_buffer->state.vb_dirty |= 1 << (startBinding + i);
   }
}

static VkResult
cmd_buffer_emit_binding_table(struct anv_cmd_buffer *cmd_buffer,
                              unsigned stage, struct anv_state *bt_state)
{
   struct anv_framebuffer *fb = cmd_buffer->state.framebuffer;
   struct anv_subpass *subpass = cmd_buffer->state.subpass;
   struct anv_pipeline_layout *layout;
   uint32_t attachments, bias, size;

   if (stage == VK_SHADER_STAGE_COMPUTE)
      layout = cmd_buffer->state.compute_pipeline->layout;
   else
      layout = cmd_buffer->state.pipeline->layout;

   if (stage == VK_SHADER_STAGE_FRAGMENT) {
      bias = MAX_RTS;
      attachments = subpass->color_count;
   } else {
      bias = 0;
      attachments = 0;
   }

   /* This is a little awkward: layout can be NULL but we still have to
    * allocate and set a binding table for the PS stage for render
    * targets. */
   uint32_t surface_count = layout ? layout->stage[stage].surface_count : 0;

   if (attachments + surface_count == 0)
      return VK_SUCCESS;

   size = (bias + surface_count) * sizeof(uint32_t);
   *bt_state = anv_cmd_buffer_alloc_surface_state(cmd_buffer, size, 32);
   uint32_t *bt_map = bt_state->map;

   if (bt_state->map == NULL)
      return VK_ERROR_OUT_OF_DEVICE_MEMORY;

   /* This is highly annoying.  The Vulkan spec puts the depth-stencil
    * attachments in with the color attachments.  Unfortunately, thanks to
    * other aspects of the API, we cana't really saparate them before this
    * point.  Therefore, we have to walk all of the attachments but only
    * put the color attachments into the binding table.
    */
   for (uint32_t a = 0; a < attachments; a++) {
      const struct anv_attachment_view *attachment =
         fb->attachments[subpass->color_attachments[a]];

      assert(attachment->attachment_type == ANV_ATTACHMENT_VIEW_TYPE_COLOR);
      const struct anv_color_attachment_view *view =
         (const struct anv_color_attachment_view *)attachment;

      struct anv_state state =
         anv_cmd_buffer_alloc_surface_state(cmd_buffer, 64, 64);

      if (state.map == NULL)
         return VK_ERROR_OUT_OF_DEVICE_MEMORY;

      memcpy(state.map, view->view.surface_state.map, 64);

      /* The address goes in dwords 8 and 9 of the SURFACE_STATE */
      *(uint64_t *)(state.map + 8 * 4) =
         anv_reloc_list_add(anv_cmd_buffer_current_surface_relocs(cmd_buffer),
                            cmd_buffer->device,
                            state.offset + 8 * 4,
                            view->view.bo, view->view.offset);

      bt_map[a] = state.offset;
   }

   if (layout == NULL)
      return VK_SUCCESS;

   for (uint32_t set = 0; set < layout->num_sets; set++) {
      struct anv_descriptor_set_binding *d = &cmd_buffer->state.descriptors[set];
      struct anv_descriptor_set_layout *set_layout = layout->set[set].layout;
      struct anv_descriptor_slot *surface_slots =
         set_layout->stage[stage].surface_start;

      uint32_t start = bias + layout->set[set].surface_start[stage];

      for (uint32_t b = 0; b < set_layout->stage[stage].surface_count; b++) {
         struct anv_surface_view *view =
            d->set->descriptors[surface_slots[b].index].view;

         if (!view)
            continue;

         struct anv_state state =
            anv_cmd_buffer_alloc_surface_state(cmd_buffer, 64, 64);

         if (state.map == NULL)
            return VK_ERROR_OUT_OF_DEVICE_MEMORY;

         uint32_t offset;
         if (surface_slots[b].dynamic_slot >= 0) {
            uint32_t dynamic_offset =
               d->dynamic_offsets[surface_slots[b].dynamic_slot];

            offset = view->offset + dynamic_offset;
            anv_fill_buffer_surface_state(state.map, view->format, offset,
                                          view->range - dynamic_offset);
         } else {
            offset = view->offset;
            memcpy(state.map, view->surface_state.map, 64);
         }

         /* The address goes in dwords 8 and 9 of the SURFACE_STATE */
         *(uint64_t *)(state.map + 8 * 4) =
            anv_reloc_list_add(anv_cmd_buffer_current_surface_relocs(cmd_buffer),
                               cmd_buffer->device,
                               state.offset + 8 * 4,
                               view->bo, offset);

         bt_map[start + b] = state.offset;
      }
   }

   return VK_SUCCESS;
}

static VkResult
cmd_buffer_emit_samplers(struct anv_cmd_buffer *cmd_buffer,
                         unsigned stage, struct anv_state *state)
{
   struct anv_pipeline_layout *layout;
   uint32_t sampler_count;

   if (stage == VK_SHADER_STAGE_COMPUTE)
      layout = cmd_buffer->state.compute_pipeline->layout;
   else
      layout = cmd_buffer->state.pipeline->layout;

   sampler_count = layout ? layout->stage[stage].sampler_count : 0;
   if (sampler_count == 0)
      return VK_SUCCESS;

   uint32_t size = sampler_count * 16;
   *state = anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, size, 32);

   if (state->map == NULL)
      return VK_ERROR_OUT_OF_DEVICE_MEMORY;

   for (uint32_t set = 0; set < layout->num_sets; set++) {
      struct anv_descriptor_set_binding *d = &cmd_buffer->state.descriptors[set];
      struct anv_descriptor_set_layout *set_layout = layout->set[set].layout;
      struct anv_descriptor_slot *sampler_slots =
         set_layout->stage[stage].sampler_start;

      uint32_t start = layout->set[set].sampler_start[stage];

      for (uint32_t b = 0; b < set_layout->stage[stage].sampler_count; b++) {
         struct anv_sampler *sampler =
            d->set->descriptors[sampler_slots[b].index].sampler;

         if (!sampler)
            continue;

         memcpy(state->map + (start + b) * 16,
                sampler->state, sizeof(sampler->state));
      }
   }

   return VK_SUCCESS;
}

static VkResult
flush_descriptor_set(struct anv_cmd_buffer *cmd_buffer, uint32_t stage)
{
   struct anv_state surfaces = { 0, }, samplers = { 0, };
   VkResult result;

   result = cmd_buffer_emit_samplers(cmd_buffer, stage, &samplers);
   if (result != VK_SUCCESS)
      return result;
   result = cmd_buffer_emit_binding_table(cmd_buffer, stage, &surfaces);
   if (result != VK_SUCCESS)
      return result;

   static const uint32_t sampler_state_opcodes[] = {
      [VK_SHADER_STAGE_VERTEX]                  = 43,
      [VK_SHADER_STAGE_TESS_CONTROL]            = 44, /* HS */
      [VK_SHADER_STAGE_TESS_EVALUATION]         = 45, /* DS */
      [VK_SHADER_STAGE_GEOMETRY]                = 46,
      [VK_SHADER_STAGE_FRAGMENT]                = 47,
      [VK_SHADER_STAGE_COMPUTE]                 = 0,
   };

   static const uint32_t binding_table_opcodes[] = {
      [VK_SHADER_STAGE_VERTEX]                  = 38,
      [VK_SHADER_STAGE_TESS_CONTROL]            = 39,
      [VK_SHADER_STAGE_TESS_EVALUATION]         = 40,
      [VK_SHADER_STAGE_GEOMETRY]                = 41,
      [VK_SHADER_STAGE_FRAGMENT]                = 42,
      [VK_SHADER_STAGE_COMPUTE]                 = 0,
   };

   if (samplers.alloc_size > 0) {
      anv_batch_emit(&cmd_buffer->batch,
                     GEN8_3DSTATE_SAMPLER_STATE_POINTERS_VS,
                     ._3DCommandSubOpcode  = sampler_state_opcodes[stage],
                     .PointertoVSSamplerState = samplers.offset);
   }

   if (surfaces.alloc_size > 0) {
      anv_batch_emit(&cmd_buffer->batch,
                     GEN8_3DSTATE_BINDING_TABLE_POINTERS_VS,
                     ._3DCommandSubOpcode  = binding_table_opcodes[stage],
                     .PointertoVSBindingTable = surfaces.offset);
   }

   return VK_SUCCESS;
}

static void
flush_descriptor_sets(struct anv_cmd_buffer *cmd_buffer)
{
   uint32_t s, dirty = cmd_buffer->state.descriptors_dirty &
                       cmd_buffer->state.pipeline->active_stages;

   VkResult result = VK_SUCCESS;
   for_each_bit(s, dirty) {
      result = flush_descriptor_set(cmd_buffer, s);
      if (result != VK_SUCCESS)
         break;
   }

   if (result != VK_SUCCESS) {
      assert(result == VK_ERROR_OUT_OF_DEVICE_MEMORY);

      result = anv_cmd_buffer_new_surface_state_bo(cmd_buffer);
      assert(result == VK_SUCCESS);

      /* Re-emit state base addresses so we get the new surface state base
       * address before we start emitting binding tables etc.
       */
      anv_cmd_buffer_emit_state_base_address(cmd_buffer);

      /* Re-emit all active binding tables */
      for_each_bit(s, cmd_buffer->state.pipeline->active_stages) {
         result = flush_descriptor_set(cmd_buffer, s);

         /* It had better succeed this time */
         assert(result == VK_SUCCESS);
      }
   }

   cmd_buffer->state.descriptors_dirty &= ~cmd_buffer->state.pipeline->active_stages;
}

static struct anv_state
anv_cmd_buffer_emit_dynamic(struct anv_cmd_buffer *cmd_buffer,
                             uint32_t *a, uint32_t dwords, uint32_t alignment)
{
   struct anv_state state;

   state = anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
                                              dwords * 4, alignment);
   memcpy(state.map, a, dwords * 4);

   VG(VALGRIND_CHECK_MEM_IS_DEFINED(state.map, dwords * 4));

   return state;
}

static struct anv_state
anv_cmd_buffer_merge_dynamic(struct anv_cmd_buffer *cmd_buffer,
                             uint32_t *a, uint32_t *b,
                             uint32_t dwords, uint32_t alignment)
{
   struct anv_state state;
   uint32_t *p;

   state = anv_cmd_buffer_alloc_dynamic_state(cmd_buffer,
                                              dwords * 4, alignment);
   p = state.map;
   for (uint32_t i = 0; i < dwords; i++)
      p[i] = a[i] | b[i];

   VG(VALGRIND_CHECK_MEM_IS_DEFINED(p, dwords * 4));

   return state;
}

static VkResult
flush_compute_descriptor_set(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_device *device = cmd_buffer->device;
   struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
   struct anv_state surfaces = { 0, }, samplers = { 0, };
   VkResult result;

   result = cmd_buffer_emit_samplers(cmd_buffer,
                                     VK_SHADER_STAGE_COMPUTE, &samplers);
   if (result != VK_SUCCESS)
      return result;
   result = cmd_buffer_emit_binding_table(cmd_buffer,
                                          VK_SHADER_STAGE_COMPUTE, &surfaces);
   if (result != VK_SUCCESS)
      return result;

   struct GEN8_INTERFACE_DESCRIPTOR_DATA desc = {
      .KernelStartPointer = pipeline->cs_simd,
      .KernelStartPointerHigh = 0,
      .BindingTablePointer = surfaces.offset,
      .BindingTableEntryCount = 0,
      .SamplerStatePointer = samplers.offset,
      .SamplerCount = 0,
      .NumberofThreadsinGPGPUThreadGroup = 0 /* FIXME: Really? */
   };

   uint32_t size = GEN8_INTERFACE_DESCRIPTOR_DATA_length * sizeof(uint32_t);
   struct anv_state state =
      anv_state_pool_alloc(&device->dynamic_state_pool, size, 64);

   GEN8_INTERFACE_DESCRIPTOR_DATA_pack(NULL, state.map, &desc);

   anv_batch_emit(&cmd_buffer->batch, GEN8_MEDIA_INTERFACE_DESCRIPTOR_LOAD,
                  .InterfaceDescriptorTotalLength = size,
                  .InterfaceDescriptorDataStartAddress = state.offset);

   return VK_SUCCESS;
}

static void
anv_cmd_buffer_flush_compute_state(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
   VkResult result;

   assert(pipeline->active_stages == VK_SHADER_STAGE_COMPUTE_BIT);

   if (cmd_buffer->state.current_pipeline != GPGPU) {
      anv_batch_emit(&cmd_buffer->batch, GEN8_PIPELINE_SELECT,
                     .PipelineSelection = GPGPU);
      cmd_buffer->state.current_pipeline = GPGPU;
   }

   if (cmd_buffer->state.compute_dirty & ANV_CMD_BUFFER_PIPELINE_DIRTY)
      anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch);

   if ((cmd_buffer->state.descriptors_dirty & VK_SHADER_STAGE_COMPUTE_BIT) ||
       (cmd_buffer->state.compute_dirty & ANV_CMD_BUFFER_PIPELINE_DIRTY)) {
      result = flush_compute_descriptor_set(cmd_buffer);
      assert(result == VK_SUCCESS);
      cmd_buffer->state.descriptors_dirty &= ~VK_SHADER_STAGE_COMPUTE;
   }

   cmd_buffer->state.compute_dirty = 0;
}

static void
anv_cmd_buffer_flush_state(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_pipeline *pipeline = cmd_buffer->state.pipeline;
   uint32_t *p;

   uint32_t vb_emit = cmd_buffer->state.vb_dirty & pipeline->vb_used;

   assert((pipeline->active_stages & VK_SHADER_STAGE_COMPUTE_BIT) == 0);

   if (cmd_buffer->state.current_pipeline != _3D) {
      anv_batch_emit(&cmd_buffer->batch, GEN8_PIPELINE_SELECT,
                     .PipelineSelection = _3D);
      cmd_buffer->state.current_pipeline = _3D;
   }

   if (vb_emit) {
      const uint32_t num_buffers = __builtin_popcount(vb_emit);
      const uint32_t num_dwords = 1 + num_buffers * 4;

      p = anv_batch_emitn(&cmd_buffer->batch, num_dwords,
                          GEN8_3DSTATE_VERTEX_BUFFERS);
      uint32_t vb, i = 0;
      for_each_bit(vb, vb_emit) {
         struct anv_buffer *buffer = cmd_buffer->state.vertex_bindings[vb].buffer;
         uint32_t offset = cmd_buffer->state.vertex_bindings[vb].offset;

         struct GEN8_VERTEX_BUFFER_STATE state = {
            .VertexBufferIndex = vb,
            .MemoryObjectControlState = GEN8_MOCS,
            .AddressModifyEnable = true,
            .BufferPitch = pipeline->binding_stride[vb],
            .BufferStartingAddress = { buffer->bo, buffer->offset + offset },
            .BufferSize = buffer->size - offset
         };

         GEN8_VERTEX_BUFFER_STATE_pack(&cmd_buffer->batch, &p[1 + i * 4], &state);
         i++;
      }
   }

   if (cmd_buffer->state.dirty & ANV_CMD_BUFFER_PIPELINE_DIRTY) {
      /* If somebody compiled a pipeline after starting a command buffer the
       * scratch bo may have grown since we started this cmd buffer (and
       * emitted STATE_BASE_ADDRESS).  If we're binding that pipeline now,
       * reemit STATE_BASE_ADDRESS so that we use the bigger scratch bo. */
      if (cmd_buffer->state.scratch_size < pipeline->total_scratch)
         anv_cmd_buffer_emit_state_base_address(cmd_buffer);

      anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch);
   }

   if (cmd_buffer->state.descriptors_dirty)
      flush_descriptor_sets(cmd_buffer);

   if (cmd_buffer->state.dirty & ANV_CMD_BUFFER_VP_DIRTY) {
      struct anv_dynamic_vp_state *vp_state = cmd_buffer->state.vp_state;
      anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_SCISSOR_STATE_POINTERS,
                     .ScissorRectPointer = vp_state->scissor.offset);
      anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_VIEWPORT_STATE_POINTERS_CC,
                     .CCViewportPointer = vp_state->cc_vp.offset);
      anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP,
                     .SFClipViewportPointer = vp_state->sf_clip_vp.offset);
   }

   if (cmd_buffer->state.dirty & (ANV_CMD_BUFFER_PIPELINE_DIRTY |
                                  ANV_CMD_BUFFER_RS_DIRTY)) {
      anv_batch_emit_merge(&cmd_buffer->batch,
                           cmd_buffer->state.rs_state->state_sf,
                           pipeline->state_sf);
      anv_batch_emit_merge(&cmd_buffer->batch,
                           cmd_buffer->state.rs_state->state_raster,
                           pipeline->state_raster);
   }

   if (cmd_buffer->state.ds_state &&
       (cmd_buffer->state.dirty & (ANV_CMD_BUFFER_PIPELINE_DIRTY |
                                   ANV_CMD_BUFFER_DS_DIRTY))) {
      anv_batch_emit_merge(&cmd_buffer->batch,
                           cmd_buffer->state.ds_state->state_wm_depth_stencil,
                           pipeline->state_wm_depth_stencil);
   }

   if (cmd_buffer->state.dirty & (ANV_CMD_BUFFER_CB_DIRTY |
                                  ANV_CMD_BUFFER_DS_DIRTY)) {
      struct anv_state state;
      if (cmd_buffer->state.ds_state == NULL)
         state = anv_cmd_buffer_emit_dynamic(cmd_buffer,
                                             cmd_buffer->state.cb_state->state_color_calc,
                                             GEN8_COLOR_CALC_STATE_length, 64);
      else if (cmd_buffer->state.cb_state == NULL)
         state = anv_cmd_buffer_emit_dynamic(cmd_buffer,
                                             cmd_buffer->state.ds_state->state_color_calc,
                                             GEN8_COLOR_CALC_STATE_length, 64);
      else
         state = anv_cmd_buffer_merge_dynamic(cmd_buffer,
                                              cmd_buffer->state.ds_state->state_color_calc,
                                              cmd_buffer->state.cb_state->state_color_calc,
                                              GEN8_COLOR_CALC_STATE_length, 64);

      anv_batch_emit(&cmd_buffer->batch,
                     GEN8_3DSTATE_CC_STATE_POINTERS,
                     .ColorCalcStatePointer = state.offset,
                     .ColorCalcStatePointerValid = true);
   }

   if (cmd_buffer->state.dirty & (ANV_CMD_BUFFER_PIPELINE_DIRTY |
                                  ANV_CMD_BUFFER_INDEX_BUFFER_DIRTY)) {
      anv_batch_emit_merge(&cmd_buffer->batch,
                           cmd_buffer->state.state_vf, pipeline->state_vf);
   }

   cmd_buffer->state.vb_dirty &= ~vb_emit;
   cmd_buffer->state.dirty = 0;
}

void anv_CmdDraw(
    VkCmdBuffer                                 cmdBuffer,
    uint32_t                                    firstVertex,
    uint32_t                                    vertexCount,
    uint32_t                                    firstInstance,
    uint32_t                                    instanceCount)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);

   anv_cmd_buffer_flush_state(cmd_buffer);

   anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
                  .VertexAccessType = SEQUENTIAL,
                  .VertexCountPerInstance = vertexCount,
                  .StartVertexLocation = firstVertex,
                  .InstanceCount = instanceCount,
                  .StartInstanceLocation = firstInstance,
                  .BaseVertexLocation = 0);
}

void anv_CmdDrawIndexed(
    VkCmdBuffer                                 cmdBuffer,
    uint32_t                                    firstIndex,
    uint32_t                                    indexCount,
    int32_t                                     vertexOffset,
    uint32_t                                    firstInstance,
    uint32_t                                    instanceCount)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);

   anv_cmd_buffer_flush_state(cmd_buffer);

   anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
                  .VertexAccessType = RANDOM,
                  .VertexCountPerInstance = indexCount,
                  .StartVertexLocation = firstIndex,
                  .InstanceCount = instanceCount,
                  .StartInstanceLocation = firstInstance,
                  .BaseVertexLocation = vertexOffset);
}

static void
anv_batch_lrm(struct anv_batch *batch,
              uint32_t reg, struct anv_bo *bo, uint32_t offset)
{
   anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_MEM,
                  .RegisterAddress = reg,
                  .MemoryAddress = { bo, offset });
}

static void
anv_batch_lri(struct anv_batch *batch, uint32_t reg, uint32_t imm)
{
   anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_IMM,
                  .RegisterOffset = reg,
                  .DataDWord = imm);
}

/* Auto-Draw / Indirect Registers */
#define GEN7_3DPRIM_END_OFFSET          0x2420
#define GEN7_3DPRIM_START_VERTEX        0x2430
#define GEN7_3DPRIM_VERTEX_COUNT        0x2434
#define GEN7_3DPRIM_INSTANCE_COUNT      0x2438
#define GEN7_3DPRIM_START_INSTANCE      0x243C
#define GEN7_3DPRIM_BASE_VERTEX         0x2440

void anv_CmdDrawIndirect(
    VkCmdBuffer                                 cmdBuffer,
    VkBuffer                                    _buffer,
    VkDeviceSize                                offset,
    uint32_t                                    count,
    uint32_t                                    stride)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
   struct anv_bo *bo = buffer->bo;
   uint32_t bo_offset = buffer->offset + offset;

   anv_cmd_buffer_flush_state(cmd_buffer);

   anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_VERTEX_COUNT, bo, bo_offset);
   anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_INSTANCE_COUNT, bo, bo_offset + 4);
   anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_VERTEX, bo, bo_offset + 8);
   anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_INSTANCE, bo, bo_offset + 12);
   anv_batch_lri(&cmd_buffer->batch, GEN7_3DPRIM_BASE_VERTEX, 0);

   anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
                  .IndirectParameterEnable = true,
                  .VertexAccessType = SEQUENTIAL);
}

void anv_CmdDrawIndexedIndirect(
    VkCmdBuffer                                 cmdBuffer,
    VkBuffer                                    _buffer,
    VkDeviceSize                                offset,
    uint32_t                                    count,
    uint32_t                                    stride)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
   struct anv_bo *bo = buffer->bo;
   uint32_t bo_offset = buffer->offset + offset;

   anv_cmd_buffer_flush_state(cmd_buffer);

   anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_VERTEX_COUNT, bo, bo_offset);
   anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_INSTANCE_COUNT, bo, bo_offset + 4);
   anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_VERTEX, bo, bo_offset + 8);
   anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_BASE_VERTEX, bo, bo_offset + 12);
   anv_batch_lrm(&cmd_buffer->batch, GEN7_3DPRIM_START_INSTANCE, bo, bo_offset + 16);

   anv_batch_emit(&cmd_buffer->batch, GEN8_3DPRIMITIVE,
                  .IndirectParameterEnable = true,
                  .VertexAccessType = RANDOM);
}

void anv_CmdDispatch(
    VkCmdBuffer                                 cmdBuffer,
    uint32_t                                    x,
    uint32_t                                    y,
    uint32_t                                    z)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
   struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;

   anv_cmd_buffer_flush_compute_state(cmd_buffer);

   anv_batch_emit(&cmd_buffer->batch, GEN8_GPGPU_WALKER,
                  .SIMDSize = prog_data->simd_size / 16,
                  .ThreadDepthCounterMaximum = 0,
                  .ThreadHeightCounterMaximum = 0,
                  .ThreadWidthCounterMaximum = pipeline->cs_thread_width_max,
                  .ThreadGroupIDXDimension = x,
                  .ThreadGroupIDYDimension = y,
                  .ThreadGroupIDZDimension = z,
                  .RightExecutionMask = pipeline->cs_right_mask,
                  .BottomExecutionMask = 0xffffffff);

   anv_batch_emit(&cmd_buffer->batch, GEN8_MEDIA_STATE_FLUSH);
}

#define GPGPU_DISPATCHDIMX 0x2500
#define GPGPU_DISPATCHDIMY 0x2504
#define GPGPU_DISPATCHDIMZ 0x2508

void anv_CmdDispatchIndirect(
    VkCmdBuffer                                 cmdBuffer,
    VkBuffer                                    _buffer,
    VkDeviceSize                                offset)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
   struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
   struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
   struct anv_bo *bo = buffer->bo;
   uint32_t bo_offset = buffer->offset + offset;

   anv_cmd_buffer_flush_compute_state(cmd_buffer);

   anv_batch_lrm(&cmd_buffer->batch, GPGPU_DISPATCHDIMX, bo, bo_offset);
   anv_batch_lrm(&cmd_buffer->batch, GPGPU_DISPATCHDIMY, bo, bo_offset + 4);
   anv_batch_lrm(&cmd_buffer->batch, GPGPU_DISPATCHDIMZ, bo, bo_offset + 8);

   anv_batch_emit(&cmd_buffer->batch, GEN8_GPGPU_WALKER,
                  .IndirectParameterEnable = true,
                  .SIMDSize = prog_data->simd_size / 16,
                  .ThreadDepthCounterMaximum = 0,
                  .ThreadHeightCounterMaximum = 0,
                  .ThreadWidthCounterMaximum = pipeline->cs_thread_width_max,
                  .RightExecutionMask = pipeline->cs_right_mask,
                  .BottomExecutionMask = 0xffffffff);

   anv_batch_emit(&cmd_buffer->batch, GEN8_MEDIA_STATE_FLUSH);
}

void anv_CmdSetEvent(
    VkCmdBuffer                                 cmdBuffer,
    VkEvent                                     event,
    VkPipelineStageFlags                        stageMask)
{
   stub();
}

void anv_CmdResetEvent(
    VkCmdBuffer                                 cmdBuffer,
    VkEvent                                     event,
    VkPipelineStageFlags                        stageMask)
{
   stub();
}

void anv_CmdWaitEvents(
    VkCmdBuffer                                 cmdBuffer,
    uint32_t                                    eventCount,
    const VkEvent*                              pEvents,
    VkPipelineStageFlags                        srcStageMask,
    VkPipelineStageFlags                        destStageMask,
    uint32_t                                    memBarrierCount,
    const void* const*                          ppMemBarriers)
{
   stub();
}

void anv_CmdPipelineBarrier(
    VkCmdBuffer                                 cmdBuffer,
    VkPipelineStageFlags                        srcStageMask,
    VkPipelineStageFlags                        destStageMask,
    VkBool32                                    byRegion,
    uint32_t                                    memBarrierCount,
    const void* const*                          ppMemBarriers)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   uint32_t b, *dw;

   struct GEN8_PIPE_CONTROL cmd = {
      GEN8_PIPE_CONTROL_header,
      .PostSyncOperation = NoWrite,
   };

   /* XXX: I think waitEvent is a no-op on our HW.  We should verify that. */

   if (anv_clear_mask(&srcStageMask, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT)) {
      /* This is just what PIPE_CONTROL does */
   }

   if (anv_clear_mask(&srcStageMask,
                      VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT |
                      VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
                      VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
                      VK_PIPELINE_STAGE_TESS_CONTROL_SHADER_BIT |
                      VK_PIPELINE_STAGE_TESS_EVALUATION_SHADER_BIT |
                      VK_PIPELINE_STAGE_GEOMETRY_SHADER_BIT |
                      VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
                      VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
                      VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
                      VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT)) {
      cmd.StallAtPixelScoreboard = true;
   }


   if (anv_clear_mask(&srcStageMask,
                      VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
                      VK_PIPELINE_STAGE_TRANSFER_BIT |
                      VK_PIPELINE_STAGE_TRANSITION_BIT)) {
      cmd.CommandStreamerStallEnable = true;
   }

   if (anv_clear_mask(&srcStageMask, VK_PIPELINE_STAGE_HOST_BIT)) {
      anv_finishme("VK_PIPE_EVENT_CPU_SIGNAL_BIT");
   }

   /* On our hardware, all stages will wait for execution as needed. */
   (void)destStageMask;

   /* We checked all known VkPipeEventFlags. */
   anv_assert(srcStageMask == 0);

   /* XXX: Right now, we're really dumb and just flush whatever categories
    * the app asks for.  One of these days we may make this a bit better
    * but right now that's all the hardware allows for in most areas.
    */
   VkMemoryOutputFlags out_flags = 0;
   VkMemoryInputFlags in_flags = 0;

   for (uint32_t i = 0; i < memBarrierCount; i++) {
      const struct anv_common *common = ppMemBarriers[i];
      switch (common->sType) {
      case VK_STRUCTURE_TYPE_MEMORY_BARRIER: {
         ANV_COMMON_TO_STRUCT(VkMemoryBarrier, barrier, common);
         out_flags |= barrier->outputMask;
         in_flags |= barrier->inputMask;
         break;
      }
      case VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER: {
         ANV_COMMON_TO_STRUCT(VkBufferMemoryBarrier, barrier, common);
         out_flags |= barrier->outputMask;
         in_flags |= barrier->inputMask;
         break;
      }
      case VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER: {
         ANV_COMMON_TO_STRUCT(VkImageMemoryBarrier, barrier, common);
         out_flags |= barrier->outputMask;
         in_flags |= barrier->inputMask;
         break;
      }
      default:
         unreachable("Invalid memory barrier type");
      }
   }

   for_each_bit(b, out_flags) {
      switch ((VkMemoryOutputFlags)(1 << b)) {
      case VK_MEMORY_OUTPUT_HOST_WRITE_BIT:
         break; /* FIXME: Little-core systems */
      case VK_MEMORY_OUTPUT_SHADER_WRITE_BIT:
         cmd.DCFlushEnable = true;
         break;
      case VK_MEMORY_OUTPUT_COLOR_ATTACHMENT_BIT:
         cmd.RenderTargetCacheFlushEnable = true;
         break;
      case VK_MEMORY_OUTPUT_DEPTH_STENCIL_ATTACHMENT_BIT:
         cmd.DepthCacheFlushEnable = true;
         break;
      case VK_MEMORY_OUTPUT_TRANSFER_BIT:
         cmd.RenderTargetCacheFlushEnable = true;
         cmd.DepthCacheFlushEnable = true;
         break;
      default:
         unreachable("Invalid memory output flag");
      }
   }

   for_each_bit(b, out_flags) {
      switch ((VkMemoryInputFlags)(1 << b)) {
      case VK_MEMORY_INPUT_HOST_READ_BIT:
         break; /* FIXME: Little-core systems */
      case VK_MEMORY_INPUT_INDIRECT_COMMAND_BIT:
      case VK_MEMORY_INPUT_INDEX_FETCH_BIT:
      case VK_MEMORY_INPUT_VERTEX_ATTRIBUTE_FETCH_BIT:
         cmd.VFCacheInvalidationEnable = true;
         break;
      case VK_MEMORY_INPUT_UNIFORM_READ_BIT:
         cmd.ConstantCacheInvalidationEnable = true;
         /* fallthrough */
      case VK_MEMORY_INPUT_SHADER_READ_BIT:
         cmd.DCFlushEnable = true;
         cmd.TextureCacheInvalidationEnable = true;
         break;
      case VK_MEMORY_INPUT_COLOR_ATTACHMENT_BIT:
      case VK_MEMORY_INPUT_DEPTH_STENCIL_ATTACHMENT_BIT:
         break; /* XXX: Hunh? */
      case VK_MEMORY_INPUT_TRANSFER_BIT:
         cmd.TextureCacheInvalidationEnable = true;
         break;
      }
   }

   dw = anv_batch_emit_dwords(&cmd_buffer->batch, GEN8_PIPE_CONTROL_length);
   GEN8_PIPE_CONTROL_pack(&cmd_buffer->batch, dw, &cmd);
}

void anv_CmdPushConstants(
    VkCmdBuffer                                 cmdBuffer,
    VkPipelineLayout                            layout,
    VkShaderStageFlags                          stageFlags,
    uint32_t                                    start,
    uint32_t                                    length,
    const void*                                 values)
{
   stub();
}

static void
anv_cmd_buffer_emit_depth_stencil(struct anv_cmd_buffer *cmd_buffer)
{
   struct anv_subpass *subpass = cmd_buffer->state.subpass;
   struct anv_framebuffer *fb = cmd_buffer->state.framebuffer;
   const struct anv_depth_stencil_view *view;

   static const struct anv_depth_stencil_view null_view =
      { .depth_format = D16_UNORM, .depth_stride = 0, .stencil_stride = 0 };

   if (subpass->depth_stencil_attachment != VK_ATTACHMENT_UNUSED) {
      const struct anv_attachment_view *aview =
         fb->attachments[subpass->depth_stencil_attachment];
      assert(aview->attachment_type == ANV_ATTACHMENT_VIEW_TYPE_DEPTH_STENCIL);
      view = (const struct anv_depth_stencil_view *)aview;
   } else {
      view = &null_view;
   }

   /* FIXME: Implement the PMA stall W/A */
   /* FIXME: Width and Height are wrong */

   anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_DEPTH_BUFFER,
                  .SurfaceType = SURFTYPE_2D,
                  .DepthWriteEnable = view->depth_stride > 0,
                  .StencilWriteEnable = view->stencil_stride > 0,
                  .HierarchicalDepthBufferEnable = false,
                  .SurfaceFormat = view->depth_format,
                  .SurfacePitch = view->depth_stride > 0 ? view->depth_stride - 1 : 0,
                  .SurfaceBaseAddress = { view->bo,  view->depth_offset },
                  .Height = cmd_buffer->state.framebuffer->height - 1,
                  .Width = cmd_buffer->state.framebuffer->width - 1,
                  .LOD = 0,
                  .Depth = 1 - 1,
                  .MinimumArrayElement = 0,
                  .DepthBufferObjectControlState = GEN8_MOCS,
                  .RenderTargetViewExtent = 1 - 1,
                  .SurfaceQPitch = view->depth_qpitch >> 2);

   /* Disable hierarchial depth buffers. */
   anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_HIER_DEPTH_BUFFER);

   anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_STENCIL_BUFFER,
                  .StencilBufferEnable = view->stencil_stride > 0,
                  .StencilBufferObjectControlState = GEN8_MOCS,
                  .SurfacePitch = view->stencil_stride > 0 ? view->stencil_stride - 1 : 0,
                  .SurfaceBaseAddress = { view->bo, view->stencil_offset },
                  .SurfaceQPitch = view->stencil_qpitch >> 2);

   /* Clear the clear params. */
   anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_CLEAR_PARAMS);
}

void
anv_cmd_buffer_begin_subpass(struct anv_cmd_buffer *cmd_buffer,
                             struct anv_subpass *subpass)
{
   cmd_buffer->state.subpass = subpass;

   cmd_buffer->state.descriptors_dirty |= VK_SHADER_STAGE_FRAGMENT_BIT;

   anv_cmd_buffer_emit_depth_stencil(cmd_buffer);
}

void anv_CmdBeginRenderPass(
    VkCmdBuffer                                 cmdBuffer,
    const VkRenderPassBeginInfo*                pRenderPassBegin,
    VkRenderPassContents                        contents)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
   ANV_FROM_HANDLE(anv_render_pass, pass, pRenderPassBegin->renderPass);
   ANV_FROM_HANDLE(anv_framebuffer, framebuffer, pRenderPassBegin->framebuffer);

   cmd_buffer->state.framebuffer = framebuffer;
   cmd_buffer->state.pass = pass;

   const VkRect2D *render_area = &pRenderPassBegin->renderArea;

   anv_batch_emit(&cmd_buffer->batch, GEN8_3DSTATE_DRAWING_RECTANGLE,
                  .ClippedDrawingRectangleYMin = render_area->offset.y,
                  .ClippedDrawingRectangleXMin = render_area->offset.x,
                  .ClippedDrawingRectangleYMax =
                     render_area->offset.y + render_area->extent.height - 1,
                  .ClippedDrawingRectangleXMax =
                     render_area->offset.x + render_area->extent.width - 1,
                  .DrawingRectangleOriginY = 0,
                  .DrawingRectangleOriginX = 0);

   anv_cmd_buffer_clear_attachments(cmd_buffer, pass,
                                    pRenderPassBegin->pAttachmentClearValues);

   anv_cmd_buffer_begin_subpass(cmd_buffer, pass->subpasses);
}

void anv_CmdNextSubpass(
    VkCmdBuffer                                 cmdBuffer,
    VkRenderPassContents                        contents)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);

   assert(cmd_buffer->level == VK_CMD_BUFFER_LEVEL_PRIMARY);

   anv_cmd_buffer_begin_subpass(cmd_buffer, cmd_buffer->state.subpass + 1);
}

void anv_CmdEndRenderPass(
    VkCmdBuffer                                 cmdBuffer)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);

   /* Emit a flushing pipe control at the end of a pass.  This is kind of a
    * hack but it ensures that render targets always actually get written.
    * Eventually, we should do flushing based on image format transitions
    * or something of that nature.
    */
   anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
                  .PostSyncOperation = NoWrite,
                  .RenderTargetCacheFlushEnable = true,
                  .InstructionCacheInvalidateEnable = true,
                  .DepthCacheFlushEnable = true,
                  .VFCacheInvalidationEnable = true,
                  .TextureCacheInvalidationEnable = true,
                  .CommandStreamerStallEnable = true);
}

void anv_CmdExecuteCommands(
    VkCmdBuffer                                 cmdBuffer,
    uint32_t                                    cmdBuffersCount,
    const VkCmdBuffer*                          pCmdBuffers)
{
   ANV_FROM_HANDLE(anv_cmd_buffer, primary, cmdBuffer);

   assert(primary->level == VK_CMD_BUFFER_LEVEL_PRIMARY);

   anv_assert(primary->state.subpass == &primary->state.pass->subpasses[0]);

   for (uint32_t i = 0; i < cmdBuffersCount; i++) {
      ANV_FROM_HANDLE(anv_cmd_buffer, secondary, pCmdBuffers[i]);

      assert(secondary->level == VK_CMD_BUFFER_LEVEL_SECONDARY);

      anv_cmd_buffer_add_secondary(primary, secondary);
   }
}

VkResult anv_CreateCommandPool(
    VkDevice                                    _device,
    const VkCmdPoolCreateInfo*                  pCreateInfo,
    VkCmdPool*                                  pCmdPool)
{
   ANV_FROM_HANDLE(anv_device, device, _device);
   struct anv_cmd_pool *pool;

   pool = anv_device_alloc(device, sizeof(*pool), 8,
                           VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
   if (pool == NULL)
      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);

   list_inithead(&pool->cmd_buffers);

   *pCmdPool = anv_cmd_pool_to_handle(pool);

   return VK_SUCCESS;
}

VkResult anv_DestroyCommandPool(
    VkDevice                                    _device,
    VkCmdPool                                   cmdPool)
{
   ANV_FROM_HANDLE(anv_device, device, _device);
   ANV_FROM_HANDLE(anv_cmd_pool, pool, cmdPool);

   anv_ResetCommandPool(_device, cmdPool, 0);

   anv_device_free(device, pool);

   return VK_SUCCESS;
}

VkResult anv_ResetCommandPool(
    VkDevice                                    device,
    VkCmdPool                                   cmdPool,
    VkCmdPoolResetFlags                         flags)
{
   ANV_FROM_HANDLE(anv_cmd_pool, pool, cmdPool);

   list_for_each_entry_safe(struct anv_cmd_buffer, cmd_buffer,
                            &pool->cmd_buffers, pool_link) {
      anv_DestroyCommandBuffer(device, anv_cmd_buffer_to_handle(cmd_buffer));
   }

   return VK_SUCCESS;
}