/**
 ******************************************************************************
 * Xenia : Xbox 360 Emulator Research Project                                 *
 ******************************************************************************
 * Copyright 2022 Ben Vanik. All rights reserved.                             *
 * Released under the BSD license - see LICENSE in the root for more details. *
 ******************************************************************************
 */

#include "xenia/gpu/d3d12/d3d12_render_target_cache.h"

#include <algorithm>
#include <cstdint>
#include <cstring>
#include <iterator>
#include <memory>
#include <string>
#include <tuple>
#include <utility>

#include "third_party/dxbc/DXBCChecksum.h"

#include "xenia/base/assert.h"
#include "xenia/base/cvar.h"
#include "xenia/base/logging.h"
#include "xenia/base/math.h"
#include "xenia/base/string.h"
#include "xenia/gpu/d3d12/d3d12_command_processor.h"
#include "xenia/gpu/d3d12/d3d12_texture_cache.h"
#include "xenia/gpu/d3d12/deferred_command_list.h"
#include "xenia/gpu/draw_util.h"
#include "xenia/gpu/dxbc.h"
#include "xenia/gpu/dxbc_shader_translator.h"
#include "xenia/gpu/gpu_flags.h"
#include "xenia/gpu/trace_writer.h"
#include "xenia/gpu/xenos.h"
#include "xenia/ui/d3d12/d3d12_provider.h"
#include "xenia/ui/d3d12/d3d12_util.h"

DEFINE_bool(
    native_stencil_value_output_d3d12_intel, false,
    "Allow stencil reference output usage on Direct3D 12 on Intel GPUs - not "
    "working on UHD Graphics 630 as of March 2021 (driver 27.20.0100.8336).",
    "GPU");
// TODO(Triang3l): Make ROV the default when it's optimized better (for
// instance, using static shader modifications to pass render target
// parameters).
DEFINE_string(
    render_target_path_d3d12, "",
    "Render target emulation path to use on Direct3D 12.\n"
    "Use: [any, rtv, rov]\n"
    " rtv:\n"
    "  Host render targets and fixed-function blending and depth / stencil "
    "testing, copying between render targets when needed.\n"
    "  Lower accuracy (limited pixel format support).\n"
    "  Performance limited primarily by render target layout changes requiring "
    "copying, but generally higher.\n"
    " rov:\n"
    "  Manual pixel packing, blending and depth / stencil testing, with free "
    "render target layout changes.\n"
    "  Requires a GPU supporting rasterizer-ordered views.\n"
    "  Highest accuracy (all pixel formats handled in software).\n"
    "  Performance limited primarily by overdraw.\n"
    "  On AMD drivers, currently causes shader compiler crashes in many "
    "cases.\n"
    " Any other value:\n"
    "  Choose what is considered the most optimal for the system (currently "
    "always RTV because the ROV path is much slower now, except for Intel "
    "GPUs, which have a bug in stencil testing that causes Xbox 360 Direct3D 9 "
    "clears not to work).",
    "GPU");

namespace xe {
namespace gpu {
namespace d3d12 {

// Generated with `xb buildshaders`.
namespace shaders {
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/clear_uint2_ps.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/fullscreen_vs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/host_depth_store_1xmsaa_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/host_depth_store_2xmsaa_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/host_depth_store_4xmsaa_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/passthrough_position_xy_vs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_clear_32bpp_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_clear_32bpp_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_clear_64bpp_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_clear_64bpp_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_fast_32bpp_1x2xmsaa_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_fast_32bpp_1x2xmsaa_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_fast_32bpp_4xmsaa_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_fast_32bpp_4xmsaa_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_fast_64bpp_1x2xmsaa_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_fast_64bpp_1x2xmsaa_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_fast_64bpp_4xmsaa_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_fast_64bpp_4xmsaa_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_128bpp_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_128bpp_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_16bpp_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_16bpp_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_32bpp_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_32bpp_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_64bpp_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_64bpp_scaled_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_8bpp_cs.h"
#include "xenia/gpu/shaders/bytecode/d3d12_5_1/resolve_full_8bpp_scaled_cs.h"
}  // namespace shaders

const D3D12RenderTargetCache::ResolveCopyShaderCode
    D3D12RenderTargetCache::kResolveCopyShaders[size_t(
        draw_util::ResolveCopyShaderIndex::kCount)] = {
        {shaders::resolve_fast_32bpp_1x2xmsaa_cs,
         sizeof(shaders::resolve_fast_32bpp_1x2xmsaa_cs),
         shaders::resolve_fast_32bpp_1x2xmsaa_scaled_cs,
         sizeof(shaders::resolve_fast_32bpp_1x2xmsaa_scaled_cs)},
        {shaders::resolve_fast_32bpp_4xmsaa_cs,
         sizeof(shaders::resolve_fast_32bpp_4xmsaa_cs),
         shaders::resolve_fast_32bpp_4xmsaa_scaled_cs,
         sizeof(shaders::resolve_fast_32bpp_4xmsaa_scaled_cs)},
        {shaders::resolve_fast_64bpp_1x2xmsaa_cs,
         sizeof(shaders::resolve_fast_64bpp_1x2xmsaa_cs),
         shaders::resolve_fast_64bpp_1x2xmsaa_scaled_cs,
         sizeof(shaders::resolve_fast_64bpp_1x2xmsaa_scaled_cs)},
        {shaders::resolve_fast_64bpp_4xmsaa_cs,
         sizeof(shaders::resolve_fast_64bpp_4xmsaa_cs),
         shaders::resolve_fast_64bpp_4xmsaa_scaled_cs,
         sizeof(shaders::resolve_fast_64bpp_4xmsaa_scaled_cs)},
        {shaders::resolve_full_8bpp_cs, sizeof(shaders::resolve_full_8bpp_cs),
         shaders::resolve_full_8bpp_scaled_cs,
         sizeof(shaders::resolve_full_8bpp_scaled_cs)},
        {shaders::resolve_full_16bpp_cs, sizeof(shaders::resolve_full_16bpp_cs),
         shaders::resolve_full_16bpp_scaled_cs,
         sizeof(shaders::resolve_full_16bpp_scaled_cs)},
        {shaders::resolve_full_32bpp_cs, sizeof(shaders::resolve_full_32bpp_cs),
         shaders::resolve_full_32bpp_scaled_cs,
         sizeof(shaders::resolve_full_32bpp_scaled_cs)},
        {shaders::resolve_full_64bpp_cs, sizeof(shaders::resolve_full_64bpp_cs),
         shaders::resolve_full_64bpp_scaled_cs,
         sizeof(shaders::resolve_full_64bpp_scaled_cs)},
        {shaders::resolve_full_128bpp_cs,
         sizeof(shaders::resolve_full_128bpp_cs),
         shaders::resolve_full_128bpp_scaled_cs,
         sizeof(shaders::resolve_full_128bpp_scaled_cs)},
};

const uint32_t D3D12RenderTargetCache::kTransferUsedRootParameters[size_t(
    TransferRootSignatureIndex::kCount)] = {
    // kColor
    kTransferUsedRootParameterColorSRVBit |
        kTransferUsedRootParameterAddressConstantBit,
    // kDepth
    kTransferUsedRootParameterDepthSRVBit |
        kTransferUsedRootParameterAddressConstantBit,
    // kDepthStencil
    kTransferUsedRootParameterDepthSRVBit |
        kTransferUsedRootParameterStencilSRVBit |
        kTransferUsedRootParameterAddressConstantBit,
    // kColorToStencilBit
    kTransferUsedRootParameterStencilMaskConstantBit |
        kTransferUsedRootParameterColorSRVBit |
        kTransferUsedRootParameterAddressConstantBit,
    // kStencilToStencilBit
    kTransferUsedRootParameterStencilMaskConstantBit |
        kTransferUsedRootParameterStencilSRVBit |
        kTransferUsedRootParameterAddressConstantBit,
    // kColorAndHostDepth
    kTransferUsedRootParameterColorSRVBit |
        kTransferUsedRootParameterAddressConstantBit |
        kTransferUsedRootParameterHostDepthSRVBit |
        kTransferUsedRootParameterHostDepthAddressConstantBit,
    // kDepthAndHostDepth
    kTransferUsedRootParameterDepthSRVBit |
        kTransferUsedRootParameterAddressConstantBit |
        kTransferUsedRootParameterHostDepthSRVBit |
        kTransferUsedRootParameterHostDepthAddressConstantBit,
    // kDepthStencilAndHostDepth
    kTransferUsedRootParameterDepthSRVBit |
        kTransferUsedRootParameterStencilSRVBit |
        kTransferUsedRootParameterAddressConstantBit |
        kTransferUsedRootParameterHostDepthSRVBit |
        kTransferUsedRootParameterHostDepthAddressConstantBit,
};

const D3D12RenderTargetCache::TransferModeInfo
    D3D12RenderTargetCache::kTransferModes[size_t(TransferMode::kCount)] = {
        // kColorToDepth
        {TransferOutput::kDepth, TransferRootSignatureIndex::kColor,
         TransferRootSignatureIndex::kColor},
        // kColorToColor
        {TransferOutput::kColor, TransferRootSignatureIndex::kColor,
         TransferRootSignatureIndex::kColor},
        // kDepthToDepth
        {TransferOutput::kDepth, TransferRootSignatureIndex::kDepth,
         TransferRootSignatureIndex::kDepthStencil},
        // kDepthToColor
        {TransferOutput::kColor, TransferRootSignatureIndex::kDepthStencil,
         TransferRootSignatureIndex::kDepthStencil},
        // kColorToStencilBit
        {TransferOutput::kStencilBit,
         TransferRootSignatureIndex::kColorToStencilBit,
         TransferRootSignatureIndex::kColorToStencilBit},
        // kDepthToStencilBit
        {TransferOutput::kStencilBit,
         TransferRootSignatureIndex::kStencilToStencilBit,
         TransferRootSignatureIndex::kStencilToStencilBit},
        // kColorAndHostDepthToDepth
        {TransferOutput::kDepth, TransferRootSignatureIndex::kColorAndHostDepth,
         TransferRootSignatureIndex::kColorAndHostDepth},
        // kDepthAndHostDepthToDepth
        {TransferOutput::kDepth, TransferRootSignatureIndex::kDepthAndHostDepth,
         TransferRootSignatureIndex::kDepthStencilAndHostDepth},
};

D3D12RenderTargetCache::~D3D12RenderTargetCache() { Shutdown(true); }

bool D3D12RenderTargetCache::Initialize() {
  const ui::d3d12::D3D12Provider& provider =
      command_processor_.GetD3D12Provider();
  ID3D12Device* device = provider.GetDevice();

  if (cvars::render_target_path_d3d12 == "rtv") {
    path_ = Path::kHostRenderTargets;
  } else if (cvars::render_target_path_d3d12 == "rov") {
    path_ = Path::kPixelShaderInterlock;
  } else {
    // As of April 2021 (driver version 27.20.0100.9316), on Intel (tested on
    // UHD Graphics 630), the "always" stencil comparison function isn't working
    // properly, so clears in the Xbox 360's Direct3D 9 don't work. Forcing ROV
    // there.
#if 1
    // The ROV path is currently much slower generally.
    // TODO(Triang3l): Make ROV the default when it's optimized better (for
    // instance, using static shader modifications to pass render target
    // parameters).
    path_ = provider.GetAdapterVendorID() ==
                    ui::GraphicsProvider::GpuVendorID::kIntel
                ? Path::kPixelShaderInterlock
                : Path::kHostRenderTargets;
#else
    // The AMD shader compiler crashes very often with Xenia's custom
    // output-merger code as of March 2021.
    path_ =
        provider.GetAdapterVendorID() == ui::GraphicsProvider::GpuVendorID::kAMD
            ? Path::kHostRenderTargets
            : Path::kPixelShaderInterlock;
#endif
  }
  if (path_ == Path::kPixelShaderInterlock &&
      !provider.AreRasterizerOrderedViewsSupported()) {
    path_ = Path::kHostRenderTargets;
  }

  // Create the buffer for reinterpreting EDRAM contents.
  uint32_t edram_buffer_size =
      xenos::kEdramSizeBytes *
      (draw_resolution_scale_x() * draw_resolution_scale_y());
  D3D12_RESOURCE_DESC edram_buffer_desc;
  ui::d3d12::util::FillBufferResourceDesc(
      edram_buffer_desc, edram_buffer_size,
      D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS);
  // The first operation will likely be depth self-comparison with host render
  // targets or drawing with ROV.
  edram_buffer_state_ = D3D12_RESOURCE_STATE_UNORDERED_ACCESS;
  // Creating zeroed for stable initial value with ROV (though on a real
  // console it has to be cleared anyway probably) and not to leak irrelevant
  // data to trace dumps when not covered by host render targets entirely.
  if (FAILED(device->CreateCommittedResource(
          &ui::d3d12::util::kHeapPropertiesDefault, D3D12_HEAP_FLAG_NONE,
          &edram_buffer_desc, edram_buffer_state_, nullptr,
          IID_PPV_ARGS(&edram_buffer_)))) {
    XELOGE("D3D12RenderTargetCache: Failed to create the EDRAM buffer");
    Shutdown();
    return false;
  }
  edram_buffer_->SetName(L"EDRAM Buffer");
  edram_buffer_modification_status_ =
      EdramBufferModificationStatus::kUnmodified;

  // Create non-shader-visible descriptors of the EDRAM buffer for copying.
  D3D12_DESCRIPTOR_HEAP_DESC edram_buffer_descriptor_heap_desc;
  edram_buffer_descriptor_heap_desc.Type =
      D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV;
  edram_buffer_descriptor_heap_desc.NumDescriptors =
      uint32_t(EdramBufferDescriptorIndex::kCount);
  edram_buffer_descriptor_heap_desc.Flags = D3D12_DESCRIPTOR_HEAP_FLAG_NONE;
  edram_buffer_descriptor_heap_desc.NodeMask = 0;
  if (FAILED(device->CreateDescriptorHeap(
          &edram_buffer_descriptor_heap_desc,
          IID_PPV_ARGS(&edram_buffer_descriptor_heap_)))) {
    XELOGE(
        "D3D12RenderTargetCache: Failed to create the descriptor heap for "
        "EDRAM buffer views");
    Shutdown();
    return false;
  }
  edram_buffer_descriptor_heap_start_ =
      edram_buffer_descriptor_heap_->GetCPUDescriptorHandleForHeapStart();
  ui::d3d12::util::CreateBufferRawSRV(
      device,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kRawSRV)),
      edram_buffer_, edram_buffer_size);
  ui::d3d12::util::CreateBufferTypedSRV(
      device,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kR32UintSRV)),
      edram_buffer_, DXGI_FORMAT_R32_UINT, edram_buffer_size >> 2);
  ui::d3d12::util::CreateBufferTypedSRV(
      device,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kR32G32UintSRV)),
      edram_buffer_, DXGI_FORMAT_R32G32_UINT, edram_buffer_size >> 3);
  ui::d3d12::util::CreateBufferTypedSRV(
      device,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kR32G32B32A32UintSRV)),
      edram_buffer_, DXGI_FORMAT_R32G32B32A32_UINT, edram_buffer_size >> 4);
  ui::d3d12::util::CreateBufferRawUAV(
      device,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kRawUAV)),
      edram_buffer_, edram_buffer_size);
  ui::d3d12::util::CreateBufferTypedUAV(
      device,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kR32UintUAV)),
      edram_buffer_, DXGI_FORMAT_R32_UINT, edram_buffer_size >> 2);
  ui::d3d12::util::CreateBufferTypedUAV(
      device,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kR32G32UintUAV)),
      edram_buffer_, DXGI_FORMAT_R32G32_UINT, edram_buffer_size >> 3);
  ui::d3d12::util::CreateBufferTypedUAV(
      device,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kR32G32B32A32UintUAV)),
      edram_buffer_, DXGI_FORMAT_R32G32B32A32_UINT, edram_buffer_size >> 4);

  bool draw_resolution_scaled = IsDrawResolutionScaled();

  // Create the resolve copying root signature.
  D3D12_ROOT_PARAMETER resolve_copy_root_parameters[4];
  // Parameter 0 is constants.
  resolve_copy_root_parameters[0].ParameterType =
      D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
  resolve_copy_root_parameters[0].Constants.ShaderRegister = 0;
  resolve_copy_root_parameters[0].Constants.RegisterSpace = 0;
  // Binding all of the shared memory at 1x resolution, portions with scaled
  // resolution.
  resolve_copy_root_parameters[0].Constants.Num32BitValues =
      (draw_resolution_scaled
           ? sizeof(draw_util::ResolveCopyShaderConstants::DestRelative)
           : sizeof(draw_util::ResolveCopyShaderConstants)) /
      sizeof(uint32_t);
  resolve_copy_root_parameters[0].ShaderVisibility =
      D3D12_SHADER_VISIBILITY_ALL;
  // Parameter 1 is the destination (shared memory).
  D3D12_DESCRIPTOR_RANGE resolve_copy_dest_range;
  resolve_copy_dest_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
  resolve_copy_dest_range.NumDescriptors = 1;
  resolve_copy_dest_range.BaseShaderRegister = 0;
  resolve_copy_dest_range.RegisterSpace = 0;
  resolve_copy_dest_range.OffsetInDescriptorsFromTableStart = 0;
  resolve_copy_root_parameters[1].ParameterType =
      D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
  resolve_copy_root_parameters[1].DescriptorTable.NumDescriptorRanges = 1;
  resolve_copy_root_parameters[1].DescriptorTable.pDescriptorRanges =
      &resolve_copy_dest_range;
  resolve_copy_root_parameters[1].ShaderVisibility =
      D3D12_SHADER_VISIBILITY_ALL;
  // Parameter 2 is the source (EDRAM).
  D3D12_DESCRIPTOR_RANGE resolve_copy_source_range;
  resolve_copy_source_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
  resolve_copy_source_range.NumDescriptors = 1;
  resolve_copy_source_range.BaseShaderRegister = 0;
  resolve_copy_source_range.RegisterSpace = 0;
  resolve_copy_source_range.OffsetInDescriptorsFromTableStart = 0;
  resolve_copy_root_parameters[2].ParameterType =
      D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
  resolve_copy_root_parameters[2].DescriptorTable.NumDescriptorRanges = 1;
  resolve_copy_root_parameters[2].DescriptorTable.pDescriptorRanges =
      &resolve_copy_source_range;
  resolve_copy_root_parameters[2].ShaderVisibility =
      D3D12_SHADER_VISIBILITY_ALL;
  // Parameter 3 is the resolution scale.
  if (draw_resolution_scaled) {
    resolve_copy_root_parameters[3].ParameterType =
        D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
    resolve_copy_root_parameters[3].Constants.ShaderRegister = 1;
    resolve_copy_root_parameters[3].Constants.RegisterSpace = 0;
    // Binding all of the shared memory at 1x resolution, portions with scaled
    // resolution.
    resolve_copy_root_parameters[3].Constants.Num32BitValues =
        sizeof(draw_util::ResolveResolutionScaleConstant) / sizeof(uint32_t);
    resolve_copy_root_parameters[3].ShaderVisibility =
        D3D12_SHADER_VISIBILITY_ALL;
  }
  D3D12_ROOT_SIGNATURE_DESC resolve_copy_root_signature_desc;
  resolve_copy_root_signature_desc.NumParameters =
      draw_resolution_scaled ? 4 : 3;
  resolve_copy_root_signature_desc.pParameters = resolve_copy_root_parameters;
  resolve_copy_root_signature_desc.NumStaticSamplers = 0;
  resolve_copy_root_signature_desc.pStaticSamplers = nullptr;
  resolve_copy_root_signature_desc.Flags = D3D12_ROOT_SIGNATURE_FLAG_NONE;
  resolve_copy_root_signature_ = ui::d3d12::util::CreateRootSignature(
      provider, resolve_copy_root_signature_desc);
  if (resolve_copy_root_signature_ == nullptr) {
    XELOGE(
        "D3D12RenderTargetCache: Failed to create the resolve copy root "
        "signature");
    Shutdown();
    return false;
  }

  // Create the resolve copying pipelines.
  for (size_t i = 0; i < size_t(draw_util::ResolveCopyShaderIndex::kCount);
       ++i) {
    const draw_util::ResolveCopyShaderInfo& resolve_copy_shader_info =
        draw_util::resolve_copy_shader_info[i];
    const ResolveCopyShaderCode& resolve_copy_shader_code =
        kResolveCopyShaders[i];
    // Somewhat verification whether resolve_copy_shaders_ is up to date.
    assert_true(resolve_copy_shader_code.unscaled &&
                resolve_copy_shader_code.unscaled_size &&
                resolve_copy_shader_code.scaled &&
                resolve_copy_shader_code.scaled_size);
    ID3D12PipelineState* resolve_copy_pipeline =
        ui::d3d12::util::CreateComputePipeline(
            device,
            draw_resolution_scaled ? resolve_copy_shader_code.scaled
                                   : resolve_copy_shader_code.unscaled,
            draw_resolution_scaled ? resolve_copy_shader_code.scaled_size
                                   : resolve_copy_shader_code.unscaled_size,
            resolve_copy_root_signature_);
    if (resolve_copy_pipeline == nullptr) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create {} resolve copy pipeline",
          resolve_copy_shader_info.debug_name);
      Shutdown();
      return false;
    }
    std::u16string resolve_copy_pipeline_name =
        xe::to_utf16(resolve_copy_shader_info.debug_name);
    resolve_copy_pipeline->SetName(
        reinterpret_cast<LPCWSTR>(resolve_copy_pipeline_name.c_str()));
    resolve_copy_pipelines_[i] = resolve_copy_pipeline;
  }

  // Using the cvar on emulator initialization so used pipelines are consistent
  // across different titles launched in one emulator instance.
  use_stencil_reference_output_ =
      cvars::native_stencil_value_output &&
      provider.IsPSSpecifiedStencilReferenceSupported() &&
      (cvars::native_stencil_value_output_d3d12_intel ||
       provider.GetAdapterVendorID() !=
           ui::GraphicsProvider::GpuVendorID::kIntel);

  if (path_ == Path::kHostRenderTargets) {
    // Host render targets.

    gamma_render_target_as_srgb_ = cvars::gamma_render_target_as_srgb;

    depth_float24_conversion_ = GetConfigDepthFloat24Conversion();

    // Check if 2x MSAA is supported or needs to be emulated with 4x MSAA
    // instead.
    if (cvars::native_2x_msaa) {
      msaa_2x_supported_ = true;
      static const DXGI_FORMAT kRenderTargetDXGIFormats[] = {
          DXGI_FORMAT_R16G16B16A16_FLOAT,
          DXGI_FORMAT_R16G16B16A16_SNORM,
          DXGI_FORMAT_R32G32_FLOAT,
          DXGI_FORMAT_D32_FLOAT_S8X24_UINT,
          DXGI_FORMAT_R10G10B10A2_UNORM,
          DXGI_FORMAT_R8G8B8A8_UNORM,
          DXGI_FORMAT_R16G16_FLOAT,
          DXGI_FORMAT_R16G16_SNORM,
          DXGI_FORMAT_R32_FLOAT,
          DXGI_FORMAT_D24_UNORM_S8_UINT,
          // For ownership transfer.
          DXGI_FORMAT_R16G16B16A16_UINT,
          DXGI_FORMAT_R32G32_UINT,
          DXGI_FORMAT_R16G16_UINT,
          DXGI_FORMAT_R32_UINT,
      };
      D3D12_FEATURE_DATA_MULTISAMPLE_QUALITY_LEVELS multisample_quality_levels;
      multisample_quality_levels.SampleCount = 2;
      multisample_quality_levels.Flags =
          D3D12_MULTISAMPLE_QUALITY_LEVELS_FLAG_NONE;
      for (size_t i = 0; i < xe::countof(kRenderTargetDXGIFormats); ++i) {
        multisample_quality_levels.Format = kRenderTargetDXGIFormats[i];
        multisample_quality_levels.NumQualityLevels = 0;
        if (FAILED(device->CheckFeatureSupport(
                D3D12_FEATURE_MULTISAMPLE_QUALITY_LEVELS,
                &multisample_quality_levels,
                sizeof(multisample_quality_levels))) ||
            !multisample_quality_levels.NumQualityLevels) {
          msaa_2x_supported_ = false;
          break;
        }
      }
    } else {
      msaa_2x_supported_ = false;
    }
    if (!msaa_2x_supported_) {
      XELOGW(
          "2x MSAA is not supported, emulated via top-left and bottom-right "
          "samples of 4x MSAA");
    }

    descriptor_pool_color_ =
        std::make_unique<ui::d3d12::D3D12CpuDescriptorPool>(
            provider, D3D12_DESCRIPTOR_HEAP_TYPE_RTV, 11);
    descriptor_pool_depth_ =
        std::make_unique<ui::d3d12::D3D12CpuDescriptorPool>(
            provider, D3D12_DESCRIPTOR_HEAP_TYPE_DSV, 11);
    descriptor_pool_srv_ = std::make_unique<ui::d3d12::D3D12CpuDescriptorPool>(
        provider, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV, 11);

    // Create null render target descriptors for gaps, must be fully typed
    // (though in pipeline states, DXGI_FORMAT_UNKNOWN must be used instead -
    // this would also cause a mismatching format error in the debug layer, but
    // it's a bug in the debug layer itself - needs to be suppressed, and
    // already fixed in some version of Windows).
    null_rtv_descriptor_ss_ = descriptor_pool_color_->AllocateDescriptor();
    null_rtv_descriptor_ms_ = descriptor_pool_color_->AllocateDescriptor();
    if (!null_rtv_descriptor_ss_ || !null_rtv_descriptor_ms_) {
      Shutdown();
      return false;
    }
    D3D12_RENDER_TARGET_VIEW_DESC null_rtv_desc;
    // The format doesn't matter, but it must be bindable as a render target,
    // not DXGI_FORMAT_UNKNOWN.
    null_rtv_desc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
    null_rtv_desc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2D;
    null_rtv_desc.Texture2D.MipSlice = 0;
    null_rtv_desc.Texture2D.PlaneSlice = 0;
    device->CreateRenderTargetView(nullptr, &null_rtv_desc,
                                   null_rtv_descriptor_ss_.GetHandle());
    null_rtv_desc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2DMS;
    device->CreateRenderTargetView(nullptr, &null_rtv_desc,
                                   null_rtv_descriptor_ms_.GetHandle());

    // For host depth -> same depth transfers, host depth storing root signature
    // and pipelines.
    D3D12_ROOT_PARAMETER
    host_depth_store_root_parameters[kHostDepthStoreRootParameterCount];
    // Constants.
    D3D12_ROOT_PARAMETER& host_depth_store_root_constants =
        host_depth_store_root_parameters[kHostDepthStoreRootParameterConstants];
    host_depth_store_root_constants.ParameterType =
        D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
    host_depth_store_root_constants.Constants.ShaderRegister = 0;
    host_depth_store_root_constants.Constants.RegisterSpace = 0;
    host_depth_store_root_constants.Constants.Num32BitValues =
        sizeof(HostDepthStoreConstants) / sizeof(uint32_t);
    host_depth_store_root_constants.ShaderVisibility =
        D3D12_SHADER_VISIBILITY_ALL;
    // Source.
    D3D12_DESCRIPTOR_RANGE host_depth_store_root_source_range;
    host_depth_store_root_source_range.RangeType =
        D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
    host_depth_store_root_source_range.NumDescriptors = 1;
    host_depth_store_root_source_range.BaseShaderRegister = 0;
    host_depth_store_root_source_range.RegisterSpace = 0;
    host_depth_store_root_source_range.OffsetInDescriptorsFromTableStart = 0;
    D3D12_ROOT_PARAMETER& host_depth_store_root_source =
        host_depth_store_root_parameters[kHostDepthStoreRootParameterSource];
    host_depth_store_root_source.ParameterType =
        D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
    host_depth_store_root_source.DescriptorTable.NumDescriptorRanges = 1;
    host_depth_store_root_source.DescriptorTable.pDescriptorRanges =
        &host_depth_store_root_source_range;
    host_depth_store_root_source.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
    // Destination.
    D3D12_DESCRIPTOR_RANGE host_depth_store_root_dest_range;
    host_depth_store_root_dest_range.RangeType =
        D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
    host_depth_store_root_dest_range.NumDescriptors = 1;
    host_depth_store_root_dest_range.BaseShaderRegister = 0;
    host_depth_store_root_dest_range.RegisterSpace = 0;
    host_depth_store_root_dest_range.OffsetInDescriptorsFromTableStart = 0;
    D3D12_ROOT_PARAMETER& host_depth_store_root_dest =
        host_depth_store_root_parameters[kHostDepthStoreRootParameterDest];
    host_depth_store_root_dest.ParameterType =
        D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
    host_depth_store_root_dest.DescriptorTable.NumDescriptorRanges = 1;
    host_depth_store_root_dest.DescriptorTable.pDescriptorRanges =
        &host_depth_store_root_dest_range;
    host_depth_store_root_dest.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
    // Root signature.
    D3D12_ROOT_SIGNATURE_DESC host_depth_store_root_desc;
    host_depth_store_root_desc.NumParameters =
        UINT(xe::countof(host_depth_store_root_parameters));
    host_depth_store_root_desc.pParameters = host_depth_store_root_parameters;
    host_depth_store_root_desc.NumStaticSamplers = 0;
    host_depth_store_root_desc.pStaticSamplers = nullptr;
    host_depth_store_root_desc.Flags = D3D12_ROOT_SIGNATURE_FLAG_NONE;
    host_depth_store_root_signature_ = ui::d3d12::util::CreateRootSignature(
        provider, host_depth_store_root_desc);
    if (!host_depth_store_root_signature_) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the host depth storing "
          "root signature");
      Shutdown();
      return false;
    }
    // Pipelines.
    // 1 sample.
    host_depth_store_pipelines_[size_t(xenos::MsaaSamples::k1X)] =
        ui::d3d12::util::CreateComputePipeline(
            device, shaders::host_depth_store_1xmsaa_cs,
            sizeof(shaders::host_depth_store_1xmsaa_cs),
            host_depth_store_root_signature_);
    if (!host_depth_store_pipelines_[size_t(xenos::MsaaSamples::k1X)]) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the 1-sample host depth "
          "storing pipeline");
      Shutdown();
      return false;
    }
    host_depth_store_pipelines_[size_t(xenos::MsaaSamples::k1X)]->SetName(
        L"Host Depth Store 1xMSAA");
    // 2 samples.
    host_depth_store_pipelines_[size_t(xenos::MsaaSamples::k2X)] =
        ui::d3d12::util::CreateComputePipeline(
            device, shaders::host_depth_store_2xmsaa_cs,
            sizeof(shaders::host_depth_store_2xmsaa_cs),
            host_depth_store_root_signature_);
    if (!host_depth_store_pipelines_[size_t(xenos::MsaaSamples::k2X)]) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the 2-sample host depth "
          "storing pipeline");
      Shutdown();
      return false;
    }
    host_depth_store_pipelines_[size_t(xenos::MsaaSamples::k2X)]->SetName(
        L"Host Depth Store 2xMSAA");
    // 4 samples.
    host_depth_store_pipelines_[size_t(xenos::MsaaSamples::k4X)] =
        ui::d3d12::util::CreateComputePipeline(
            device, shaders::host_depth_store_4xmsaa_cs,
            sizeof(shaders::host_depth_store_4xmsaa_cs),
            host_depth_store_root_signature_);
    if (!host_depth_store_pipelines_[size_t(xenos::MsaaSamples::k4X)]) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the 4-sample host depth "
          "storing pipeline");
      Shutdown();
      return false;
    }
    host_depth_store_pipelines_[size_t(xenos::MsaaSamples::k4X)]->SetName(
        L"Host Depth Store 4xMSAA");

    // Transfer and clear vertex buffer, for quads of up to tile granularity.
    transfer_vertex_buffer_pool_ =
        std::make_unique<ui::d3d12::D3D12UploadBufferPool>(
            provider,
            std::max(ui::d3d12::D3D12UploadBufferPool::kDefaultPageSize,
                     sizeof(float) * 2 * 6 *
                         Transfer::kMaxCutoutBorderRectangles *
                         xenos::kEdramTileCount));

    // Transfer root signatures.
    D3D12_DESCRIPTOR_RANGE transfer_root_color_srv_range;
    transfer_root_color_srv_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
    transfer_root_color_srv_range.NumDescriptors = 1;
    transfer_root_color_srv_range.BaseShaderRegister =
        kTransferSRVRegisterColor;
    transfer_root_color_srv_range.RegisterSpace = 0;
    transfer_root_color_srv_range.OffsetInDescriptorsFromTableStart = 0;
    D3D12_DESCRIPTOR_RANGE transfer_root_depth_srv_range;
    transfer_root_depth_srv_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
    transfer_root_depth_srv_range.NumDescriptors = 1;
    transfer_root_depth_srv_range.BaseShaderRegister =
        kTransferSRVRegisterDepth;
    transfer_root_depth_srv_range.RegisterSpace = 0;
    transfer_root_depth_srv_range.OffsetInDescriptorsFromTableStart = 0;
    D3D12_DESCRIPTOR_RANGE transfer_root_stencil_srv_range;
    transfer_root_stencil_srv_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
    transfer_root_stencil_srv_range.NumDescriptors = 1;
    transfer_root_stencil_srv_range.BaseShaderRegister =
        kTransferSRVRegisterStencil;
    transfer_root_stencil_srv_range.RegisterSpace = 0;
    transfer_root_stencil_srv_range.OffsetInDescriptorsFromTableStart = 0;
    D3D12_DESCRIPTOR_RANGE transfer_root_host_depth_srv_range;
    transfer_root_host_depth_srv_range.RangeType =
        D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
    transfer_root_host_depth_srv_range.NumDescriptors = 1;
    transfer_root_host_depth_srv_range.BaseShaderRegister =
        kTransferSRVRegisterHostDepth;
    transfer_root_host_depth_srv_range.RegisterSpace = 0;
    transfer_root_host_depth_srv_range.OffsetInDescriptorsFromTableStart = 0;
    D3D12_ROOT_PARAMETER
    transfer_root_parameters[kTransferUsedRootParameterCount];
    D3D12_ROOT_SIGNATURE_DESC transfer_root_desc;
    transfer_root_desc.pParameters = transfer_root_parameters;
    transfer_root_desc.NumStaticSamplers = 0;
    transfer_root_desc.pStaticSamplers = nullptr;
    transfer_root_desc.Flags =
        D3D12_ROOT_SIGNATURE_FLAG_ALLOW_INPUT_ASSEMBLER_INPUT_LAYOUT;
    for (size_t i = 0; i < size_t(TransferRootSignatureIndex::kCount); ++i) {
      uint32_t transfer_root_mask = kTransferUsedRootParameters[i];
      // Stencil mask constant.
      if (transfer_root_mask &
          kTransferUsedRootParameterStencilMaskConstantBit) {
        D3D12_ROOT_PARAMETER& transfer_root_stencil_mask_constant =
            transfer_root_parameters[xe::bit_count(
                transfer_root_mask &
                (kTransferUsedRootParameterStencilMaskConstantBit - 1))];
        transfer_root_stencil_mask_constant.ParameterType =
            D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
        transfer_root_stencil_mask_constant.Constants.ShaderRegister =
            kTransferCBVRegisterStencilMask;
        transfer_root_stencil_mask_constant.Constants.RegisterSpace = 0;
        transfer_root_stencil_mask_constant.Constants.Num32BitValues = 1;
        transfer_root_stencil_mask_constant.ShaderVisibility =
            D3D12_SHADER_VISIBILITY_PIXEL;
      }
      // Color SRV.
      if (transfer_root_mask & kTransferUsedRootParameterColorSRVBit) {
        D3D12_ROOT_PARAMETER& transfer_root_color_srv =
            transfer_root_parameters[xe::bit_count(
                transfer_root_mask &
                (kTransferUsedRootParameterColorSRVBit - 1))];
        transfer_root_color_srv.ParameterType =
            D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
        transfer_root_color_srv.DescriptorTable.NumDescriptorRanges = 1;
        transfer_root_color_srv.DescriptorTable.pDescriptorRanges =
            &transfer_root_color_srv_range;
        transfer_root_color_srv.ShaderVisibility =
            D3D12_SHADER_VISIBILITY_PIXEL;
      }
      // Depth SRV.
      if (transfer_root_mask & kTransferUsedRootParameterDepthSRVBit) {
        D3D12_ROOT_PARAMETER& transfer_root_depth_srv =
            transfer_root_parameters[xe::bit_count(
                transfer_root_mask &
                (kTransferUsedRootParameterDepthSRVBit - 1))];
        transfer_root_depth_srv.ParameterType =
            D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
        transfer_root_depth_srv.DescriptorTable.NumDescriptorRanges = 1;
        transfer_root_depth_srv.DescriptorTable.pDescriptorRanges =
            &transfer_root_depth_srv_range;
        transfer_root_depth_srv.ShaderVisibility =
            D3D12_SHADER_VISIBILITY_PIXEL;
      }
      // Stencil SRV.
      if (transfer_root_mask & kTransferUsedRootParameterStencilSRVBit) {
        D3D12_ROOT_PARAMETER& transfer_root_stencil_srv =
            transfer_root_parameters[xe::bit_count(
                transfer_root_mask &
                (kTransferUsedRootParameterStencilSRVBit - 1))];
        transfer_root_stencil_srv.ParameterType =
            D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
        transfer_root_stencil_srv.DescriptorTable.NumDescriptorRanges = 1;
        transfer_root_stencil_srv.DescriptorTable.pDescriptorRanges =
            &transfer_root_stencil_srv_range;
        transfer_root_stencil_srv.ShaderVisibility =
            D3D12_SHADER_VISIBILITY_PIXEL;
      }
      // Address constant.
      if (transfer_root_mask & kTransferUsedRootParameterAddressConstantBit) {
        D3D12_ROOT_PARAMETER& transfer_root_address_constant =
            transfer_root_parameters[xe::bit_count(
                transfer_root_mask &
                (kTransferUsedRootParameterAddressConstantBit - 1))];
        transfer_root_address_constant.ParameterType =
            D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
        transfer_root_address_constant.Constants.ShaderRegister =
            kTransferCBVRegisterAddress;
        transfer_root_address_constant.Constants.RegisterSpace = 0;
        transfer_root_address_constant.Constants.Num32BitValues =
            sizeof(TransferAddressConstant) / sizeof(uint32_t);
        transfer_root_address_constant.ShaderVisibility =
            D3D12_SHADER_VISIBILITY_PIXEL;
      }
      // Host depth SRV.
      if (transfer_root_mask & kTransferUsedRootParameterHostDepthSRVBit) {
        D3D12_ROOT_PARAMETER& transfer_root_host_depth_srv =
            transfer_root_parameters[xe::bit_count(
                transfer_root_mask &
                (kTransferUsedRootParameterHostDepthSRVBit - 1))];
        transfer_root_host_depth_srv.ParameterType =
            D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
        transfer_root_host_depth_srv.DescriptorTable.NumDescriptorRanges = 1;
        transfer_root_host_depth_srv.DescriptorTable.pDescriptorRanges =
            &transfer_root_host_depth_srv_range;
        transfer_root_host_depth_srv.ShaderVisibility =
            D3D12_SHADER_VISIBILITY_PIXEL;
      }
      // Host depth address constant.
      if (transfer_root_mask &
          kTransferUsedRootParameterHostDepthAddressConstantBit) {
        D3D12_ROOT_PARAMETER& transfer_root_host_address_constant =
            transfer_root_parameters[xe::bit_count(
                transfer_root_mask &
                (kTransferUsedRootParameterHostDepthAddressConstantBit - 1))];
        transfer_root_host_address_constant.ParameterType =
            D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
        transfer_root_host_address_constant.Constants.ShaderRegister =
            kTransferCBVRegisterHostDepthAddress;
        transfer_root_host_address_constant.Constants.RegisterSpace = 0;
        transfer_root_host_address_constant.Constants.Num32BitValues =
            sizeof(TransferAddressConstant) / sizeof(uint32_t);
        transfer_root_host_address_constant.ShaderVisibility =
            D3D12_SHADER_VISIBILITY_PIXEL;
      }
      transfer_root_desc.NumParameters = xe::bit_count(transfer_root_mask);
      assert_true(transfer_root_desc.NumParameters <=
                  kTransferUsedRootParameterCount);
      transfer_root_signatures_[i] =
          ui::d3d12::util::CreateRootSignature(provider, transfer_root_desc);
      if (!transfer_root_signatures_[i]) {
        XELOGE(
            "D3D12RenderTargetCache: Failed to create the render target "
            "ownership transfer root signature {:X}",
            transfer_root_mask);
        Shutdown();
        return false;
      }
    }

    // Dumping root signatures.
    D3D12_DESCRIPTOR_RANGE dump_root_source_range;
    dump_root_source_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
    dump_root_source_range.NumDescriptors = 1;
    dump_root_source_range.BaseShaderRegister = 0;
    dump_root_source_range.RegisterSpace = 0;
    dump_root_source_range.OffsetInDescriptorsFromTableStart = 0;
    D3D12_DESCRIPTOR_RANGE dump_root_stencil_range;
    dump_root_stencil_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
    dump_root_stencil_range.NumDescriptors = 1;
    dump_root_stencil_range.BaseShaderRegister = 1;
    dump_root_stencil_range.RegisterSpace = 0;
    dump_root_stencil_range.OffsetInDescriptorsFromTableStart = 0;
    D3D12_DESCRIPTOR_RANGE dump_root_edram_range;
    dump_root_edram_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
    dump_root_edram_range.NumDescriptors = 1;
    dump_root_edram_range.BaseShaderRegister = 0;
    dump_root_edram_range.RegisterSpace = 0;
    dump_root_edram_range.OffsetInDescriptorsFromTableStart = 0;
    D3D12_ROOT_PARAMETER
    dump_root_color_parameters[kDumpRootParameterColorCount];
    D3D12_ROOT_PARAMETER
    dump_root_depth_parameters[kDumpRootParameterDepthCount];
    for (uint32_t i = 0; i < 2; ++i) {
      // Offsets.
      D3D12_ROOT_PARAMETER& dump_root_offsets =
          i ? dump_root_depth_parameters[kDumpRootParameterOffsets]
            : dump_root_color_parameters[kDumpRootParameterOffsets];
      dump_root_offsets.ParameterType =
          D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
      dump_root_offsets.Constants.ShaderRegister = kDumpCbufferOffsets;
      dump_root_offsets.Constants.RegisterSpace = 0;
      dump_root_offsets.Constants.Num32BitValues =
          sizeof(DumpOffsets) / sizeof(uint32_t);
      dump_root_offsets.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
      // Source.
      D3D12_ROOT_PARAMETER& dump_root_source =
          i ? dump_root_depth_parameters[kDumpRootParameterSource]
            : dump_root_color_parameters[kDumpRootParameterSource];
      dump_root_source.ParameterType =
          D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
      dump_root_source.DescriptorTable.NumDescriptorRanges = 1;
      dump_root_source.DescriptorTable.pDescriptorRanges =
          &dump_root_source_range;
      dump_root_source.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
      // Stencil.
      if (i) {
        D3D12_ROOT_PARAMETER& dump_root_stencil =
            dump_root_depth_parameters[kDumpRootParameterDepthStencil];
        dump_root_stencil.ParameterType =
            D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
        dump_root_stencil.DescriptorTable.NumDescriptorRanges = 1;
        dump_root_stencil.DescriptorTable.pDescriptorRanges =
            &dump_root_stencil_range;
        dump_root_stencil.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
      }
      // Pitches.
      D3D12_ROOT_PARAMETER& dump_root_pitches =
          i ? dump_root_depth_parameters[kDumpRootParameterDepthPitches]
            : dump_root_color_parameters[kDumpRootParameterColorPitches];
      dump_root_pitches.ParameterType =
          D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
      dump_root_pitches.Constants.ShaderRegister = kDumpCbufferPitches;
      dump_root_pitches.Constants.RegisterSpace = 0;
      dump_root_pitches.Constants.Num32BitValues =
          sizeof(DumpPitches) / sizeof(uint32_t);
      dump_root_pitches.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
      // EDRAM.
      D3D12_ROOT_PARAMETER& dump_root_edram =
          i ? dump_root_depth_parameters[kDumpRootParameterDepthEdram]
            : dump_root_color_parameters[kDumpRootParameterColorEdram];
      dump_root_edram.ParameterType =
          D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
      dump_root_edram.DescriptorTable.NumDescriptorRanges = 1;
      dump_root_edram.DescriptorTable.pDescriptorRanges =
          &dump_root_edram_range;
      dump_root_edram.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
    }
    D3D12_ROOT_SIGNATURE_DESC dump_root_desc;
    dump_root_desc.NumParameters =
        UINT(xe::countof(dump_root_color_parameters));
    dump_root_desc.pParameters = dump_root_color_parameters;
    dump_root_desc.NumStaticSamplers = 0;
    dump_root_desc.pStaticSamplers = nullptr;
    dump_root_desc.Flags = D3D12_ROOT_SIGNATURE_FLAG_NONE;
    dump_root_signature_color_ =
        ui::d3d12::util::CreateRootSignature(provider, dump_root_desc);
    if (!dump_root_signature_color_) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the color render target "
          "dumping root signature");
      Shutdown();
      return false;
    }
    dump_root_desc.NumParameters =
        UINT(xe::countof(dump_root_depth_parameters));
    dump_root_desc.pParameters = dump_root_depth_parameters;
    dump_root_signature_depth_ =
        ui::d3d12::util::CreateRootSignature(provider, dump_root_desc);
    if (!dump_root_signature_depth_) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the depth render target "
          "dumping root signature");
      Shutdown();
      return false;
    }

    // k_32_FLOAT and k_32_32_FLOAT clear root signature and pipelines.
    D3D12_ROOT_PARAMETER uint32_rtv_clear_root_constants;
    uint32_rtv_clear_root_constants.ParameterType =
        D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
    uint32_rtv_clear_root_constants.Constants.ShaderRegister = 0;
    uint32_rtv_clear_root_constants.Constants.RegisterSpace = 0;
    uint32_rtv_clear_root_constants.Constants.Num32BitValues = 2;
    uint32_rtv_clear_root_constants.ShaderVisibility =
        D3D12_SHADER_VISIBILITY_PIXEL;
    D3D12_ROOT_SIGNATURE_DESC uint32_rtv_clear_root_desc;
    uint32_rtv_clear_root_desc.NumParameters = 1;
    uint32_rtv_clear_root_desc.pParameters = &uint32_rtv_clear_root_constants;
    uint32_rtv_clear_root_desc.NumStaticSamplers = 0;
    uint32_rtv_clear_root_desc.pStaticSamplers = nullptr;
    uint32_rtv_clear_root_desc.Flags = D3D12_ROOT_SIGNATURE_FLAG_NONE;
    uint32_rtv_clear_root_signature_ = ui::d3d12::util::CreateRootSignature(
        provider, uint32_rtv_clear_root_desc);
    if (!uint32_rtv_clear_root_signature_) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the k_32_FLOAT / "
          "k_32_32_FLOAT render target clearing root signature");
      Shutdown();
      return false;
    }
    D3D12_GRAPHICS_PIPELINE_STATE_DESC uint32_rtv_clear_pipeline_desc = {};
    uint32_rtv_clear_pipeline_desc.pRootSignature =
        uint32_rtv_clear_root_signature_;
    uint32_rtv_clear_pipeline_desc.VS.pShaderBytecode = shaders::fullscreen_vs;
    uint32_rtv_clear_pipeline_desc.VS.BytecodeLength =
        sizeof(shaders::fullscreen_vs);
    uint32_rtv_clear_pipeline_desc.PS.pShaderBytecode = shaders::clear_uint2_ps;
    uint32_rtv_clear_pipeline_desc.PS.BytecodeLength =
        sizeof(shaders::clear_uint2_ps);
    uint32_rtv_clear_pipeline_desc.BlendState.RenderTarget[0]
        .RenderTargetWriteMask = D3D12_COLOR_WRITE_ENABLE_ALL;
    uint32_rtv_clear_pipeline_desc.RasterizerState.FillMode =
        D3D12_FILL_MODE_SOLID;
    uint32_rtv_clear_pipeline_desc.RasterizerState.CullMode =
        D3D12_CULL_MODE_NONE;
    uint32_rtv_clear_pipeline_desc.RasterizerState.DepthClipEnable = TRUE;
    uint32_rtv_clear_pipeline_desc.PrimitiveTopologyType =
        D3D12_PRIMITIVE_TOPOLOGY_TYPE_TRIANGLE;
    uint32_rtv_clear_pipeline_desc.NumRenderTargets = 1;
    for (size_t i = 0; i < 2; ++i) {
      uint32_rtv_clear_pipeline_desc.RTVFormats[0] =
          GetColorOwnershipTransferDXGIFormat(
              i ? xenos::ColorRenderTargetFormat::k_32_32_FLOAT
                : xenos::ColorRenderTargetFormat::k_32_FLOAT);
      for (size_t j = size_t(xenos::MsaaSamples::k1X);
           j <= size_t(xenos::MsaaSamples::k4X); ++j) {
        if (xenos::MsaaSamples(j) == xenos::MsaaSamples::k2X &&
            !msaa_2x_supported_) {
          // Using sample 0 as 0 and 3 as 1 for 2x instead.
          uint32_rtv_clear_pipeline_desc.SampleMask = 0b1001;
          uint32_rtv_clear_pipeline_desc.SampleDesc.Count = 4;
        } else {
          uint32_rtv_clear_pipeline_desc.SampleMask = UINT_MAX;
          uint32_rtv_clear_pipeline_desc.SampleDesc.Count = 1 << j;
        }
        ID3D12PipelineState* uint32_rtv_clear_pipeline;
        if (FAILED(device->CreateGraphicsPipelineState(
                &uint32_rtv_clear_pipeline_desc,
                IID_PPV_ARGS(&uint32_rtv_clear_pipeline)))) {
          XELOGE(
              "D3D12RenderTargetCache: Failed to create the {} {}-sample "
              "render target clearing pipeline",
              i ? "k_32_32_FLOAT" : "k_32_FLOAT", uint32_t(1) << j);
          Shutdown();
          return false;
        }
        uint32_rtv_clear_pipelines_[i][j] = uint32_rtv_clear_pipeline;
        std::wstring uint32_rtv_clear_pipeline_name =
            fmt::format(L"Resolve Clear {} {}xMSAA",
                        i ? L"k_32_32_FLOAT" : L"k_32_FLOAT", uint32_t(1) << j);
        uint32_rtv_clear_pipeline->SetName(
            reinterpret_cast<LPCWSTR>(uint32_rtv_clear_pipeline_name.c_str()));
      }
    }

    // FXC-compiled depth / stencil dumping shader is ~2 KB, reserve 4 KB for
    // some additional space.
    built_shader_.reserve(1024);
  } else if (path_ == Path::kPixelShaderInterlock) {
    // Pixel shader interlock (rasterizer-ordered view).

    // Blending is done in linear space directly in shaders.
    gamma_render_target_as_srgb_ = false;

    // Always true float24 depth.
    depth_float24_conversion_ = DepthFloat24Conversion::kOnOutputRounding;

    // Only ForcedSampleCount, which doesn't support 2x.
    msaa_2x_supported_ = false;

    // Create the resolve EDRAM buffer clearing root signature.
    D3D12_ROOT_PARAMETER resolve_rov_clear_root_parameters[3];
    // Parameter 0 is constants.
    resolve_rov_clear_root_parameters[0].ParameterType =
        D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
    resolve_rov_clear_root_parameters[0].Constants.ShaderRegister = 0;
    resolve_rov_clear_root_parameters[0].Constants.RegisterSpace = 0;
    // Binding all of the shared memory at 1x resolution, portions with scaled
    // resolution.
    resolve_rov_clear_root_parameters[0].Constants.Num32BitValues =
        sizeof(draw_util::ResolveClearShaderConstants) / sizeof(uint32_t);
    resolve_rov_clear_root_parameters[0].ShaderVisibility =
        D3D12_SHADER_VISIBILITY_ALL;
    // Parameter 1 is the destination (EDRAM).
    D3D12_DESCRIPTOR_RANGE resolve_rov_clear_dest_range;
    resolve_rov_clear_dest_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
    resolve_rov_clear_dest_range.NumDescriptors = 1;
    resolve_rov_clear_dest_range.BaseShaderRegister = 0;
    resolve_rov_clear_dest_range.RegisterSpace = 0;
    resolve_rov_clear_dest_range.OffsetInDescriptorsFromTableStart = 0;
    resolve_rov_clear_root_parameters[1].ParameterType =
        D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
    resolve_rov_clear_root_parameters[1].DescriptorTable.NumDescriptorRanges =
        1;
    resolve_rov_clear_root_parameters[1].DescriptorTable.pDescriptorRanges =
        &resolve_rov_clear_dest_range;
    resolve_rov_clear_root_parameters[1].ShaderVisibility =
        D3D12_SHADER_VISIBILITY_ALL;
    // Parameter 2 is the resolution scale.
    if (draw_resolution_scaled) {
      resolve_rov_clear_root_parameters[2].ParameterType =
          D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
      resolve_rov_clear_root_parameters[2].Constants.ShaderRegister = 1;
      resolve_rov_clear_root_parameters[2].Constants.RegisterSpace = 0;
      // Binding all of the shared memory at 1x resolution, portions with scaled
      // resolution.
      resolve_rov_clear_root_parameters[2].Constants.Num32BitValues =
          sizeof(draw_util::ResolveResolutionScaleConstant) / sizeof(uint32_t);
      resolve_rov_clear_root_parameters[2].ShaderVisibility =
          D3D12_SHADER_VISIBILITY_ALL;
    }
    D3D12_ROOT_SIGNATURE_DESC resolve_rov_clear_root_signature_desc;
    resolve_rov_clear_root_signature_desc.NumParameters =
        draw_resolution_scaled ? 3 : 2;
    resolve_rov_clear_root_signature_desc.pParameters =
        resolve_rov_clear_root_parameters;
    resolve_rov_clear_root_signature_desc.NumStaticSamplers = 0;
    resolve_rov_clear_root_signature_desc.pStaticSamplers = nullptr;
    resolve_rov_clear_root_signature_desc.Flags =
        D3D12_ROOT_SIGNATURE_FLAG_NONE;
    resolve_rov_clear_root_signature_ = ui::d3d12::util::CreateRootSignature(
        provider, resolve_rov_clear_root_signature_desc);
    if (resolve_rov_clear_root_signature_ == nullptr) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the resolve EDRAM buffer "
          "clear root signature");
      Shutdown();
      return false;
    }

    // Create the resolve EDRAM buffer clearing pipelines.
    resolve_rov_clear_32bpp_pipeline_ = ui::d3d12::util::CreateComputePipeline(
        device,
        draw_resolution_scaled ? shaders::resolve_clear_32bpp_scaled_cs
                               : shaders::resolve_clear_32bpp_cs,
        draw_resolution_scaled ? sizeof(shaders::resolve_clear_32bpp_scaled_cs)
                               : sizeof(shaders::resolve_clear_32bpp_cs),
        resolve_rov_clear_root_signature_);
    if (resolve_rov_clear_32bpp_pipeline_ == nullptr) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the 32bpp resolve EDRAM "
          "buffer clear pipeline");
      Shutdown();
      return false;
    }
    resolve_rov_clear_32bpp_pipeline_->SetName(L"Resolve Clear 32bpp");
    resolve_rov_clear_64bpp_pipeline_ = ui::d3d12::util::CreateComputePipeline(
        device,
        draw_resolution_scaled ? shaders::resolve_clear_64bpp_scaled_cs
                               : shaders::resolve_clear_64bpp_cs,
        draw_resolution_scaled ? sizeof(shaders::resolve_clear_64bpp_scaled_cs)
                               : sizeof(shaders::resolve_clear_64bpp_cs),
        resolve_rov_clear_root_signature_);
    if (resolve_rov_clear_64bpp_pipeline_ == nullptr) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create the 64bpp resolve EDRAM "
          "buffer clear pipeline");
      Shutdown();
      return false;
    }
    resolve_rov_clear_64bpp_pipeline_->SetName(L"Resolve Clear 64bpp");
  } else {
    assert_unhandled_case(path_);
    Shutdown();
    return false;
  }

  InitializeCommon();

  return true;
}

void D3D12RenderTargetCache::Shutdown(bool from_destructor) {
  ui::d3d12::util::ReleaseAndNull(resolve_rov_clear_64bpp_pipeline_);
  ui::d3d12::util::ReleaseAndNull(resolve_rov_clear_32bpp_pipeline_);
  ui::d3d12::util::ReleaseAndNull(resolve_rov_clear_root_signature_);

  for (size_t i = 0; i < 2; ++i) {
    for (size_t j = size_t(xenos::MsaaSamples::k1X);
         j <= size_t(xenos::MsaaSamples::k4X); ++j) {
      ui::d3d12::util::ReleaseAndNull(uint32_rtv_clear_pipelines_[i][j]);
    }
  }
  ui::d3d12::util::ReleaseAndNull(uint32_rtv_clear_root_signature_);

  for (const auto& dump_pipeline_pair : dump_pipelines_) {
    if (dump_pipeline_pair.second) {
      dump_pipeline_pair.second->Release();
    }
  }
  dump_pipelines_.clear();
  ui::d3d12::util::ReleaseAndNull(dump_root_signature_depth_);
  ui::d3d12::util::ReleaseAndNull(dump_root_signature_color_);

  for (const auto& transfer_pipeline_array_pair :
       transfer_stencil_bit_pipelines_) {
    for (ID3D12PipelineState* transfer_pipeline :
         transfer_pipeline_array_pair.second) {
      if (transfer_pipeline) {
        transfer_pipeline->Release();
      }
    }
  }
  transfer_stencil_bit_pipelines_.clear();
  for (const auto& transfer_pipeline_pair : transfer_pipelines_) {
    if (transfer_pipeline_pair.second) {
      transfer_pipeline_pair.second->Release();
    }
  }
  transfer_pipelines_.clear();
  for (size_t i = 0; i < xe::countof(transfer_root_signatures_); ++i) {
    ui::d3d12::util::ReleaseAndNull(transfer_root_signatures_[i]);
  }

  transfer_vertex_buffer_pool_.reset();

  for (size_t i = 0; i < xe::countof(host_depth_store_pipelines_); ++i) {
    ui::d3d12::util::ReleaseAndNull(host_depth_store_pipelines_[i]);
  }
  ui::d3d12::util::ReleaseAndNull(host_depth_store_root_signature_);

  null_rtv_descriptor_ms_.Free();
  null_rtv_descriptor_ss_.Free();
  descriptor_pool_srv_.reset();
  descriptor_pool_depth_.reset();
  descriptor_pool_color_.reset();

  for (size_t i = 0; i < xe::countof(resolve_copy_pipelines_); ++i) {
    ui::d3d12::util::ReleaseAndNull(resolve_copy_pipelines_[i]);
  }
  ui::d3d12::util::ReleaseAndNull(resolve_copy_root_signature_);

  edram_snapshot_restore_pool_.reset();
  ui::d3d12::util::ReleaseAndNull(edram_snapshot_download_buffer_);

  ui::d3d12::util::ReleaseAndNull(edram_buffer_descriptor_heap_);
  ui::d3d12::util::ReleaseAndNull(edram_buffer_);

  if (!from_destructor) {
    ShutdownCommon();
  }
}

void D3D12RenderTargetCache::CompletedSubmissionUpdated() {
  if (edram_snapshot_restore_pool_) {
    edram_snapshot_restore_pool_->Reclaim(
        command_processor_.GetCompletedSubmission());
  }
  if (transfer_vertex_buffer_pool_) {
    transfer_vertex_buffer_pool_->Reclaim(
        command_processor_.GetCompletedSubmission());
  }
}

void D3D12RenderTargetCache::BeginSubmission() {
  // New command list - render targets not bound.
  InvalidateCommandListRenderTargets();
  // ExecuteCommandLists is a full UAV barrier.
  if (edram_buffer_modification_status_ !=
      EdramBufferModificationStatus::kUnmodified) {
    assert_true(edram_buffer_state_ == D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
    edram_buffer_modification_status_ =
        EdramBufferModificationStatus::kUnmodified;
    PixelShaderInterlockFullEdramBarrierPlaced();
  }
}

bool D3D12RenderTargetCache::Update(
    bool is_rasterization_done, reg::RB_DEPTHCONTROL normalized_depth_control,
    uint32_t normalized_color_mask, const Shader& vertex_shader) {
  if (!RenderTargetCache::Update(is_rasterization_done,
                                 normalized_depth_control,
                                 normalized_color_mask, vertex_shader)) {
    return false;
  }
  switch (GetPath()) {
    case Path::kHostRenderTargets: {
      RenderTarget* const* depth_and_color_render_targets =
          last_update_accumulated_render_targets();
      PerformTransfersAndResolveClears(1 + xenos::kMaxColorRenderTargets,
                                       depth_and_color_render_targets,
                                       last_update_transfers());
      SetCommandListRenderTargets(depth_and_color_render_targets);
    } break;
    case Path::kPixelShaderInterlock: {
      // For ROV, only the barrier is needed - already scheduled if required.
      // But the buffer will be used for ROV drawing now.
      TransitionEdramBuffer(D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
      // Commit preceding UAV (but not ROV) writes like clears as they aren't
      // synchronized with ROV accesses.
      CommitEdramBufferUAVWrites(EdramBufferModificationStatus::kAsUAV);
      // TODO(Triang3l): Check if this draw call modifies color or depth /
      // stencil, at least coarsely, to prevent useless barriers.
      MarkEdramBufferModified(EdramBufferModificationStatus::kAsROV);
    } break;
    default:
      assert_unhandled_case(GetPath());
      return false;
  }
  return true;
}

void D3D12RenderTargetCache::WriteEdramRawSRVDescriptor(
    D3D12_CPU_DESCRIPTOR_HANDLE handle) {
  const ui::d3d12::D3D12Provider& provider =
      command_processor_.GetD3D12Provider();
  ID3D12Device* device = provider.GetDevice();
  device->CopyDescriptorsSimple(
      1, handle,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kRawSRV)),
      D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
}

void D3D12RenderTargetCache::WriteEdramRawUAVDescriptor(
    D3D12_CPU_DESCRIPTOR_HANDLE handle) {
  const ui::d3d12::D3D12Provider& provider =
      command_processor_.GetD3D12Provider();
  ID3D12Device* device = provider.GetDevice();
  device->CopyDescriptorsSimple(
      1, handle,
      provider.OffsetViewDescriptor(
          edram_buffer_descriptor_heap_start_,
          uint32_t(EdramBufferDescriptorIndex::kRawUAV)),
      D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
}

void D3D12RenderTargetCache::WriteEdramUintPow2SRVDescriptor(
    D3D12_CPU_DESCRIPTOR_HANDLE handle, uint32_t element_size_bytes_pow2) {
  EdramBufferDescriptorIndex descriptor_index;
  switch (element_size_bytes_pow2) {
    case 2:
      descriptor_index = EdramBufferDescriptorIndex::kR32UintSRV;
      break;
    case 3:
      descriptor_index = EdramBufferDescriptorIndex::kR32G32UintSRV;
      break;
    case 4:
      descriptor_index = EdramBufferDescriptorIndex::kR32G32B32A32UintSRV;
      break;
    default:
      assert_unhandled_case(element_size_bytes_pow2);
      return;
  }
  const ui::d3d12::D3D12Provider& provider =
      command_processor_.GetD3D12Provider();
  ID3D12Device* device = provider.GetDevice();
  device->CopyDescriptorsSimple(
      1, handle,
      provider.OffsetViewDescriptor(edram_buffer_descriptor_heap_start_,
                                    uint32_t(descriptor_index)),
      D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
}

void D3D12RenderTargetCache::WriteEdramUintPow2UAVDescriptor(
    D3D12_CPU_DESCRIPTOR_HANDLE handle, uint32_t element_size_bytes_pow2) {
  EdramBufferDescriptorIndex descriptor_index;
  switch (element_size_bytes_pow2) {
    case 2:
      descriptor_index = EdramBufferDescriptorIndex::kR32UintUAV;
      break;
    case 3:
      descriptor_index = EdramBufferDescriptorIndex::kR32G32UintUAV;
      break;
    case 4:
      descriptor_index = EdramBufferDescriptorIndex::kR32G32B32A32UintUAV;
      break;
    default:
      assert_unhandled_case(element_size_bytes_pow2);
      return;
  }
  const ui::d3d12::D3D12Provider& provider =
      command_processor_.GetD3D12Provider();
  ID3D12Device* device = provider.GetDevice();
  device->CopyDescriptorsSimple(
      1, handle,
      provider.OffsetViewDescriptor(edram_buffer_descriptor_heap_start_,
                                    uint32_t(descriptor_index)),
      D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
}

bool D3D12RenderTargetCache::Resolve(const Memory& memory,
                                     D3D12SharedMemory& shared_memory,
                                     D3D12TextureCache& texture_cache,
                                     uint32_t& written_address_out,
                                     uint32_t& written_length_out) {
  written_address_out = 0;
  written_length_out = 0;

  bool draw_resolution_scaled = IsDrawResolutionScaled();

  draw_util::ResolveInfo resolve_info;
  if (!draw_util::GetResolveInfo(
          register_file(), memory, trace_writer_, draw_resolution_scaled,
          IsFixed16TruncatedToMinus1To1(), resolve_info)) {
    return false;
  }

  // Nothing to copy/clear.
  if (!resolve_info.address.width_div_8 || !resolve_info.address.height_div_8) {
    return true;
  }

  draw_util::ResolveResolutionScaleConstant resolution_scale_constant;
  resolution_scale_constant.resolution_scale_x = draw_resolution_scale_x();
  resolution_scale_constant.resolution_scale_y = draw_resolution_scale_y();

  DeferredCommandList& command_list =
      command_processor_.GetDeferredCommandList();

  // Copying.
  bool copied = false;
  if (resolve_info.copy_dest_length) {
    if (GetPath() == Path::kHostRenderTargets) {
      // Dump the current contents of the render targets owning the affected
      // range to edram_buffer_.
      // TODO(Triang3l): Direct host render target -> shared memory resolve
      // shaders for non-converting cases.
      uint32_t dump_base;
      uint32_t dump_row_length_used;
      uint32_t dump_rows;
      uint32_t dump_pitch;
      resolve_info.GetCopyEdramTileSpan(dump_base, dump_row_length_used,
                                        dump_rows, dump_pitch);
      DumpRenderTargets(dump_base, dump_row_length_used, dump_rows, dump_pitch);
    }

    draw_util::ResolveCopyShaderConstants copy_shader_constants;
    uint32_t copy_group_count_x, copy_group_count_y;
    draw_util::ResolveCopyShaderIndex copy_shader = resolve_info.GetCopyShader(
        draw_resolution_scale_x(), draw_resolution_scale_y(),
        copy_shader_constants, copy_group_count_x, copy_group_count_y);
    assert_true(copy_group_count_x && copy_group_count_y);
    if (copy_shader != draw_util::ResolveCopyShaderIndex::kUnknown) {
      const draw_util::ResolveCopyShaderInfo& copy_shader_info =
          draw_util::resolve_copy_shader_info[size_t(copy_shader)];

      // Make sure there is memory to write to.
      bool copy_dest_committed;
      if (draw_resolution_scaled) {
        copy_dest_committed =
            texture_cache.EnsureScaledResolveMemoryCommitted(
                resolve_info.copy_dest_base, resolve_info.copy_dest_length) &&
            texture_cache.MakeScaledResolveRangeCurrent(
                resolve_info.copy_dest_base, resolve_info.copy_dest_length);
      } else {
        copy_dest_committed = shared_memory.RequestRange(
            resolve_info.copy_dest_base, resolve_info.copy_dest_length);
      }
      if (copy_dest_committed) {
        // Write the descriptors and transition the resources.
        // Full shared memory without resolution scaling, range of the scaled
        // resolve buffer with scaling because only at least 128 * 2^20 R32
        // elements must be addressable
        // (D3D12_REQ_BUFFER_RESOURCE_TEXEL_COUNT_2_TO_EXP).
        ui::d3d12::util::DescriptorCpuGpuHandlePair descriptor_dest;
        ui::d3d12::util::DescriptorCpuGpuHandlePair descriptor_source;
        ui::d3d12::util::DescriptorCpuGpuHandlePair descriptors[2];
        if (command_processor_.RequestOneUseSingleViewDescriptors(
                bindless_resources_used_ ? uint32_t(draw_resolution_scaled) : 2,
                descriptors)) {
          if (bindless_resources_used_) {
            if (draw_resolution_scaled) {
              descriptor_dest = descriptors[0];
            } else {
              descriptor_dest =
                  command_processor_
                      .GetSharedMemoryUintPow2BindlessUAVHandlePair(
                          copy_shader_info.dest_bpe_log2);
            }
            if (copy_shader_info.source_is_raw) {
              descriptor_source =
                  command_processor_.GetSystemBindlessViewHandlePair(
                      D3D12CommandProcessor::SystemBindlessView::kEdramRawSRV);
            } else {
              descriptor_source =
                  command_processor_.GetEdramUintPow2BindlessSRVHandlePair(
                      copy_shader_info.source_bpe_log2);
            }
          } else {
            descriptor_dest = descriptors[0];
            if (!draw_resolution_scaled) {
              shared_memory.WriteUintPow2UAVDescriptor(
                  descriptor_dest.first, copy_shader_info.dest_bpe_log2);
            }
            descriptor_source = descriptors[1];
            if (copy_shader_info.source_is_raw) {
              WriteEdramRawSRVDescriptor(descriptor_source.first);
            } else {
              WriteEdramUintPow2SRVDescriptor(descriptor_source.first,
                                              copy_shader_info.source_bpe_log2);
            }
          }
          if (draw_resolution_scaled) {
            texture_cache.CreateCurrentScaledResolveRangeUintPow2UAV(
                descriptor_dest.first, copy_shader_info.dest_bpe_log2);
            texture_cache.TransitionCurrentScaledResolveRange(
                D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
          } else {
            shared_memory.UseForWriting();
          }
          TransitionEdramBuffer(D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE);

          // Submit the resolve.
          command_list.D3DSetComputeRootSignature(resolve_copy_root_signature_);
          if (draw_resolution_scaled) {
            command_list.D3DSetComputeRoot32BitConstants(
                3, sizeof(resolution_scale_constant) / sizeof(uint32_t),
                &resolution_scale_constant, 0);
          }
          command_list.D3DSetComputeRootDescriptorTable(
              2, descriptor_source.second);
          command_list.D3DSetComputeRootDescriptorTable(1,
                                                        descriptor_dest.second);
          if (draw_resolution_scaled) {
            command_list.D3DSetComputeRoot32BitConstants(
                0,
                sizeof(copy_shader_constants.dest_relative) / sizeof(uint32_t),
                &copy_shader_constants.dest_relative, 0);
          } else {
            command_list.D3DSetComputeRoot32BitConstants(
                0, sizeof(copy_shader_constants) / sizeof(uint32_t),
                &copy_shader_constants, 0);
          }
          command_processor_.SetExternalPipeline(
              resolve_copy_pipelines_[size_t(copy_shader)]);
          command_processor_.SubmitBarriers();
          command_list.D3DDispatch(copy_group_count_x, copy_group_count_y, 1);

          // Order the resolve with other work using the destination as a UAV.
          if (draw_resolution_scaled) {
            texture_cache.MarkCurrentScaledResolveRangeUAVWritesCommitNeeded();
          } else {
            shared_memory.MarkUAVWritesCommitNeeded();
          }

          // Invalidate textures and mark the range as scaled if needed.
          texture_cache.MarkRangeAsResolved(resolve_info.copy_dest_base,
                                            resolve_info.copy_dest_length);
          written_address_out = resolve_info.copy_dest_base;
          written_length_out = resolve_info.copy_dest_length;
          copied = true;
        }
      } else {
        XELOGE(
            "D3D12RenderTargetCache: Failed to obtain the resolve destination "
            "memory region");
      }
    }
  } else {
    copied = true;
  }

  // Clearing.
  bool cleared = false;
  bool clear_depth = resolve_info.IsClearingDepth();
  bool clear_color = resolve_info.IsClearingColor();
  if (clear_depth || clear_color) {
    switch (GetPath()) {
      case Path::kHostRenderTargets: {
        Transfer::Rectangle clear_rectangle;
        RenderTarget* clear_render_targets[2];
        // If PrepareHostRenderTargetsResolveClear returns false, may be just an
        // empty region (success) or an error - don't care.
        if (PrepareHostRenderTargetsResolveClear(
                resolve_info, clear_rectangle, clear_render_targets[0],
                clear_transfers_[0], clear_render_targets[1],
                clear_transfers_[1])) {
          uint64_t clear_values[2];
          clear_values[0] = resolve_info.rb_depth_clear;
          clear_values[1] = resolve_info.rb_color_clear |
                            (uint64_t(resolve_info.rb_color_clear_lo) << 32);
          PerformTransfersAndResolveClears(2, clear_render_targets,
                                           clear_transfers_, clear_values,
                                           &clear_rectangle);
        }
        cleared = true;
      } break;
      case Path::kPixelShaderInterlock: {
        ui::d3d12::util::DescriptorCpuGpuHandlePair descriptor_edram;
        bool descriptor_edram_obtained;
        if (bindless_resources_used_) {
          descriptor_edram = command_processor_.GetSystemBindlessViewHandlePair(
              D3D12CommandProcessor::SystemBindlessView ::
                  kEdramR32G32B32A32UintUAV);
          descriptor_edram_obtained = true;
        } else {
          descriptor_edram_obtained =
              command_processor_.RequestOneUseSingleViewDescriptors(
                  1, &descriptor_edram);
          if (descriptor_edram_obtained) {
            WriteEdramUintPow2UAVDescriptor(descriptor_edram.first, 4);
          }
        }
        if (descriptor_edram_obtained) {
          TransitionEdramBuffer(D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
          // Should be safe to only commit once (if was UAV / ROV previously -
          // if there was nothing to copy, only to clear, for some reason, for
          // instance), overlap of the depth and the color ranges is highly
          // unlikely.
          CommitEdramBufferUAVWrites();
          command_list.D3DSetComputeRootSignature(
              resolve_rov_clear_root_signature_);
          if (draw_resolution_scaled) {
            command_list.D3DSetComputeRoot32BitConstants(
                2, sizeof(resolution_scale_constant) / sizeof(uint32_t),
                &resolution_scale_constant, 0);
          }
          command_list.D3DSetComputeRootDescriptorTable(
              1, descriptor_edram.second);
          std::pair<uint32_t, uint32_t> clear_group_count =
              resolve_info.GetClearShaderGroupCount(draw_resolution_scale_x(),
                                                    draw_resolution_scale_y());
          assert_true(clear_group_count.first && clear_group_count.second);
          if (clear_depth) {
            draw_util::ResolveClearShaderConstants depth_clear_constants;
            resolve_info.GetDepthClearShaderConstants(depth_clear_constants);
            command_list.D3DSetComputeRoot32BitConstants(
                0, sizeof(depth_clear_constants) / sizeof(uint32_t),
                &depth_clear_constants, 0);
            command_processor_.SetExternalPipeline(
                resolve_rov_clear_32bpp_pipeline_);
            command_processor_.SubmitBarriers();
            command_list.D3DDispatch(clear_group_count.first,
                                     clear_group_count.second, 1);
          }
          if (clear_color) {
            draw_util::ResolveClearShaderConstants color_clear_constants;
            resolve_info.GetColorClearShaderConstants(color_clear_constants);
            if (clear_depth) {
              // Non-RT-specific constants have already been set.
              command_list.D3DSetComputeRoot32BitConstants(
                  0,
                  sizeof(color_clear_constants.rt_specific) / sizeof(uint32_t),
                  &color_clear_constants.rt_specific,
                  offsetof(draw_util::ResolveClearShaderConstants,
                           rt_specific) /
                      sizeof(uint32_t));
            } else {
              command_list.D3DSetComputeRoot32BitConstants(
                  0, sizeof(color_clear_constants) / sizeof(uint32_t),
                  &color_clear_constants, 0);
            }
            command_processor_.SetExternalPipeline(
                resolve_info.color_edram_info.format_is_64bpp
                    ? resolve_rov_clear_64bpp_pipeline_
                    : resolve_rov_clear_32bpp_pipeline_);
            command_processor_.SubmitBarriers();
            command_list.D3DDispatch(clear_group_count.first,
                                     clear_group_count.second, 1);
          }
          MarkEdramBufferModified();
          cleared = true;
        }
      } break;
      default:
        assert_unhandled_case(GetPath());
    }
  } else {
    cleared = true;
  }

  return copied && cleared;
}

bool D3D12RenderTargetCache::InitializeTraceSubmitDownloads() {
  if (IsDrawResolutionScaled()) {
    // No 1:1 mapping.
    return false;
  }
  if (!edram_snapshot_download_buffer_) {
    D3D12_RESOURCE_DESC edram_snapshot_download_buffer_desc;
    ui::d3d12::util::FillBufferResourceDesc(edram_snapshot_download_buffer_desc,
                                            xenos::kEdramSizeBytes,
                                            D3D12_RESOURCE_FLAG_NONE);
    const ui::d3d12::D3D12Provider& provider =
        command_processor_.GetD3D12Provider();
    ID3D12Device* device = provider.GetDevice();
    if (FAILED(device->CreateCommittedResource(
            &ui::d3d12::util::kHeapPropertiesReadback,
            provider.GetHeapFlagCreateNotZeroed(),
            &edram_snapshot_download_buffer_desc,
            D3D12_RESOURCE_STATE_COPY_DEST, nullptr,
            IID_PPV_ARGS(&edram_snapshot_download_buffer_)))) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create a EDRAM snapshot download "
          "buffer");
      return false;
    }
  }
  if (GetPath() == Path::kHostRenderTargets) {
    // Dump all host render targets to edram_buffer_.
    DumpRenderTargets(0, xenos::kEdramTileCount, 1, xenos::kEdramTileCount);
  }
  TransitionEdramBuffer(D3D12_RESOURCE_STATE_COPY_SOURCE);
  command_processor_.SubmitBarriers();
  command_processor_.GetDeferredCommandList().D3DCopyBufferRegion(
      edram_snapshot_download_buffer_, 0, edram_buffer_, 0,
      xenos::kEdramSizeBytes);
  return true;
}

void D3D12RenderTargetCache::InitializeTraceCompleteDownloads() {
  if (!edram_snapshot_download_buffer_) {
    return;
  }
  void* download_mapping;
  if (SUCCEEDED(edram_snapshot_download_buffer_->Map(0, nullptr,
                                                     &download_mapping))) {
    trace_writer_.WriteEdramSnapshot(download_mapping);
    D3D12_RANGE download_write_range = {};
    edram_snapshot_download_buffer_->Unmap(0, &download_write_range);
  } else {
    XELOGE(
        "D3D12RenderTargetCache: Failed to map the EDRAM snapshot download "
        "buffer");
  }
  edram_snapshot_download_buffer_->Release();
  edram_snapshot_download_buffer_ = nullptr;
}

void D3D12RenderTargetCache::RestoreEdramSnapshot(const void* snapshot) {
  if (IsDrawResolutionScaled()) {
    // No 1:1 mapping.
    return;
  }

  // Create the buffer - will be used for copying to either a 32-bit 1280x2048
  // render target or the EDRAM buffer.
  const ui::d3d12::D3D12Provider& provider =
      command_processor_.GetD3D12Provider();
  if (!edram_snapshot_restore_pool_) {
    edram_snapshot_restore_pool_ =
        std::make_unique<ui::d3d12::D3D12UploadBufferPool>(
            provider, xenos::kEdramSizeBytes);
  }
  ID3D12Resource* upload_buffer;
  size_t upload_buffer_offset;
  void* upload_buffer_mapping = edram_snapshot_restore_pool_->Request(
      command_processor_.GetCurrentSubmission(), xenos::kEdramSizeBytes, 1,
      &upload_buffer, &upload_buffer_offset, nullptr);
  if (!upload_buffer_mapping) {
    XELOGE(
        "D3D12RenderTargetCache: Failed to get a buffer for restoring a EDRAM "
        "snapshot");
    return;
  }

  DeferredCommandList& command_list =
      command_processor_.GetDeferredCommandList();

  switch (GetPath()) {
    case Path::kHostRenderTargets: {
      // k_32_FLOAT because it's unambiguous (not effected by something like
      // DXGI_FORMAT_R8G8B8A8 vs. DXGI_FORMAT_B8G8R8A8).
      D3D12RenderTarget* full_edram_render_target =
          static_cast<D3D12RenderTarget*>(
              PrepareFullEdram1280xRenderTargetForSnapshotRestoration(
                  xenos::ColorRenderTargetFormat::k_32_FLOAT));
      if (!full_edram_render_target) {
        return;
      }
      D3D12_TEXTURE_COPY_LOCATION copy_source_location;
      copy_source_location.pResource = upload_buffer;
      copy_source_location.Type = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT;
      UINT64 copy_total_bytes;
      D3D12_RESOURCE_DESC full_edram_render_target_desc =
          full_edram_render_target->resource()->GetDesc();
      provider.GetDevice()->GetCopyableFootprints(
          &full_edram_render_target_desc, 0, 1, 0,
          &copy_source_location.PlacedFootprint, nullptr, nullptr,
          &copy_total_bytes);
      // 1280 width * sizeof(uint32_t) is aligned to
      // D3D12_TEXTURE_DATA_PITCH_ALIGNMENT (256).
      assert_true(copy_total_bytes <= xenos::kEdramSizeBytes);
      assert_false(full_edram_render_target->key().Is64bpp());
      uint32_t pitch_tiles =
          full_edram_render_target->key().pitch_tiles_at_32bpp;
      uint32_t tile_rows = xenos::kEdramTileCount / pitch_tiles;
      assert_true(pitch_tiles * tile_rows == xenos::kEdramTileCount);
      const uint8_t* snapshot_sample_row =
          reinterpret_cast<const uint8_t*>(snapshot);
      for (uint32_t y_tile = 0; y_tile < tile_rows; ++y_tile) {
        uint8_t* upload_buffer_tile_row_origin =
            reinterpret_cast<uint8_t*>(upload_buffer_mapping) +
            copy_source_location.PlacedFootprint.Offset +
            xenos::kEdramTileHeightSamples * y_tile *
                copy_source_location.PlacedFootprint.Footprint.RowPitch;
        for (uint32_t x_tile = 0; x_tile < pitch_tiles; ++x_tile) {
          uint8_t* upload_buffer_sample_row =
              upload_buffer_tile_row_origin +
              sizeof(uint32_t) * xenos::kEdramTileWidthSamples * x_tile;
          for (uint32_t sample_row = 0;
               sample_row < xenos::kEdramTileHeightSamples; ++sample_row) {
            std::memcpy(upload_buffer_sample_row, snapshot_sample_row,
                        sizeof(uint32_t) * xenos::kEdramTileWidthSamples);
            snapshot_sample_row +=
                sizeof(uint32_t) * xenos::kEdramTileWidthSamples;
            upload_buffer_sample_row +=
                copy_source_location.PlacedFootprint.Footprint.RowPitch;
          }
        }
      }
      command_processor_.PushTransitionBarrier(
          full_edram_render_target->resource(),
          full_edram_render_target->SetResourceState(
              D3D12_RESOURCE_STATE_COPY_DEST),
          D3D12_RESOURCE_STATE_COPY_DEST);
      command_processor_.SubmitBarriers();
      D3D12_TEXTURE_COPY_LOCATION copy_dest_location;
      copy_dest_location.pResource = full_edram_render_target->resource();
      copy_dest_location.Type = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX;
      copy_dest_location.SubresourceIndex = 0;
      command_list.D3DCopyTextureRegion(&copy_dest_location, 0, 0, 0,
                                        &copy_source_location, nullptr);
    } break;

    case Path::kPixelShaderInterlock: {
      std::memcpy(upload_buffer_mapping, snapshot, xenos::kEdramSizeBytes);
      TransitionEdramBuffer(D3D12_RESOURCE_STATE_COPY_DEST);
      command_processor_.SubmitBarriers();
      command_list.D3DCopyBufferRegion(edram_buffer_, 0, upload_buffer,
                                       UINT64(upload_buffer_offset),
                                       xenos::kEdramSizeBytes);
    } break;

    default:
      assert_unhandled_case(GetPath());
  }
}

DXGI_FORMAT D3D12RenderTargetCache::GetColorResourceDXGIFormat(
    xenos::ColorRenderTargetFormat format) const {
  // Typed should be preferred over typeless so there are more opportunities for
  // compression.
  switch (format) {
    case xenos::ColorRenderTargetFormat::k_8_8_8_8:
    case xenos::ColorRenderTargetFormat::k_8_8_8_8_GAMMA:
      if (gamma_render_target_as_srgb_) {
        // Can toggle between UNORM and UNORM_SRGB for the same data.
        return DXGI_FORMAT_R8G8B8A8_TYPELESS;
      }
      return DXGI_FORMAT_R8G8B8A8_UNORM;
    case xenos::ColorRenderTargetFormat::k_2_10_10_10:
    case xenos::ColorRenderTargetFormat::k_2_10_10_10_AS_10_10_10_10:
      return DXGI_FORMAT_R10G10B10A2_UNORM;
    case xenos::ColorRenderTargetFormat::k_2_10_10_10_FLOAT:
    case xenos::ColorRenderTargetFormat::k_2_10_10_10_FLOAT_AS_16_16_16_16:
      return DXGI_FORMAT_R16G16B16A16_FLOAT;
    // SNORM has two representations of -1.
    case xenos::ColorRenderTargetFormat::k_16_16:
      return DXGI_FORMAT_R16G16_TYPELESS;
    case xenos::ColorRenderTargetFormat::k_16_16_16_16:
      return DXGI_FORMAT_R16G16B16A16_TYPELESS;
    // Floating-point - ensure NaN propagation during ownership transfer for
    // unmodified data.
    case xenos::ColorRenderTargetFormat::k_16_16_FLOAT:
      return DXGI_FORMAT_R16G16_TYPELESS;
    case xenos::ColorRenderTargetFormat::k_16_16_16_16_FLOAT:
      return DXGI_FORMAT_R16G16B16A16_TYPELESS;
    // TODO(Triang3l): Check if NaN propagation defined in the D3D11.3
    // specification can be relied on for 32-bit float render targets.
    case xenos::ColorRenderTargetFormat::k_32_FLOAT:
      return DXGI_FORMAT_R32_TYPELESS;
    case xenos::ColorRenderTargetFormat::k_32_32_FLOAT:
      return DXGI_FORMAT_R32G32_TYPELESS;
    default:
      assert_unhandled_case(format);
      return DXGI_FORMAT_UNKNOWN;
  }
}

DXGI_FORMAT D3D12RenderTargetCache::GetColorDrawDXGIFormat(
    xenos::ColorRenderTargetFormat format) const {
  switch (format) {
    case xenos::ColorRenderTargetFormat::k_8_8_8_8:
      return DXGI_FORMAT_R8G8B8A8_UNORM;
    case xenos::ColorRenderTargetFormat::k_8_8_8_8_GAMMA:
      return gamma_render_target_as_srgb_ ? DXGI_FORMAT_R8G8B8A8_UNORM_SRGB
                                          : DXGI_FORMAT_R8G8B8A8_UNORM;
    case xenos::ColorRenderTargetFormat::k_16_16:
      return DXGI_FORMAT_R16G16_SNORM;
    case xenos::ColorRenderTargetFormat::k_16_16_16_16:
      return DXGI_FORMAT_R16G16B16A16_SNORM;
    case xenos::ColorRenderTargetFormat::k_16_16_FLOAT:
      return DXGI_FORMAT_R16G16_FLOAT;
    case xenos::ColorRenderTargetFormat::k_16_16_16_16_FLOAT:
      return DXGI_FORMAT_R16G16B16A16_FLOAT;
    case xenos::ColorRenderTargetFormat::k_32_FLOAT:
      return DXGI_FORMAT_R32_FLOAT;
    case xenos::ColorRenderTargetFormat::k_32_32_FLOAT:
      return DXGI_FORMAT_R32G32_FLOAT;
    default:
      return GetColorResourceDXGIFormat(format);
  }
}

DXGI_FORMAT D3D12RenderTargetCache::GetColorOwnershipTransferDXGIFormat(
    xenos::ColorRenderTargetFormat format, bool* is_integer_out) const {
  if (is_integer_out) {
    *is_integer_out = true;
  }
  switch (format) {
    case xenos::ColorRenderTargetFormat::k_16_16:
    case xenos::ColorRenderTargetFormat::k_16_16_FLOAT:
      return DXGI_FORMAT_R16G16_UINT;
    case xenos::ColorRenderTargetFormat::k_16_16_16_16:
    case xenos::ColorRenderTargetFormat::k_16_16_16_16_FLOAT:
      return DXGI_FORMAT_R16G16B16A16_UINT;
    case xenos::ColorRenderTargetFormat::k_32_FLOAT:
      return DXGI_FORMAT_R32_UINT;
    case xenos::ColorRenderTargetFormat::k_32_32_FLOAT:
      return DXGI_FORMAT_R32G32_UINT;
    default:
      if (is_integer_out) {
        *is_integer_out = false;
      }
      return GetColorDrawDXGIFormat(format);
  }
}

DXGI_FORMAT D3D12RenderTargetCache::GetDepthResourceDXGIFormat(
    xenos::DepthRenderTargetFormat format) {
  switch (format) {
    case xenos::DepthRenderTargetFormat::kD24S8:
      return DXGI_FORMAT_R24G8_TYPELESS;
    case xenos::DepthRenderTargetFormat::kD24FS8:
      return DXGI_FORMAT_R32G8X24_TYPELESS;
    default:
      assert_unhandled_case(format);
      return DXGI_FORMAT_UNKNOWN;
  }
}

DXGI_FORMAT D3D12RenderTargetCache::GetDepthDSVDXGIFormat(
    xenos::DepthRenderTargetFormat format) {
  switch (format) {
    case xenos::DepthRenderTargetFormat::kD24S8:
      return DXGI_FORMAT_D24_UNORM_S8_UINT;
    case xenos::DepthRenderTargetFormat::kD24FS8:
      return DXGI_FORMAT_D32_FLOAT_S8X24_UINT;
    default:
      assert_unhandled_case(format);
      return DXGI_FORMAT_UNKNOWN;
  }
}

DXGI_FORMAT D3D12RenderTargetCache::GetDepthSRVDepthDXGIFormat(
    xenos::DepthRenderTargetFormat format) {
  switch (format) {
    case xenos::DepthRenderTargetFormat::kD24S8:
      return DXGI_FORMAT_R24_UNORM_X8_TYPELESS;
    case xenos::DepthRenderTargetFormat::kD24FS8:
      return DXGI_FORMAT_R32_FLOAT_X8X24_TYPELESS;
    default:
      assert_unhandled_case(format);
      return DXGI_FORMAT_UNKNOWN;
  }
}

DXGI_FORMAT D3D12RenderTargetCache::GetDepthSRVStencilDXGIFormat(
    xenos::DepthRenderTargetFormat format) {
  switch (format) {
    case xenos::DepthRenderTargetFormat::kD24S8:
      return DXGI_FORMAT_X24_TYPELESS_G8_UINT;
    case xenos::DepthRenderTargetFormat::kD24FS8:
      return DXGI_FORMAT_X32_TYPELESS_G8X24_UINT;
    default:
      assert_unhandled_case(format);
      return DXGI_FORMAT_UNKNOWN;
  }
}

RenderTargetCache::RenderTarget* D3D12RenderTargetCache::CreateRenderTarget(
    RenderTargetKey key) {
  ID3D12Device* device = command_processor_.GetD3D12Provider().GetDevice();

  D3D12_RESOURCE_DESC resource_desc;
  resource_desc.Dimension = D3D12_RESOURCE_DIMENSION_TEXTURE2D;
  resource_desc.Alignment = 0;
  resource_desc.Width = key.GetWidth() * draw_resolution_scale_x();
  resource_desc.Height =
      GetRenderTargetHeight(key.pitch_tiles_at_32bpp, key.msaa_samples) *
      draw_resolution_scale_y();
  resource_desc.DepthOrArraySize = 1;
  resource_desc.MipLevels = 1;
  if (key.is_depth) {
    resource_desc.Format = GetDepthResourceDXGIFormat(key.GetDepthFormat());
  } else {
    resource_desc.Format = GetColorResourceDXGIFormat(key.GetColorFormat());
  }
  assert_true(resource_desc.Format != DXGI_FORMAT_UNKNOWN);
  if (resource_desc.Format == DXGI_FORMAT_UNKNOWN) {
    XELOGE("D3D12RenderTargetCache: Unknown {} render target format {}",
           key.is_depth ? "depth" : "color", key.resource_format);
    return nullptr;
  }
  if (key.msaa_samples == xenos::MsaaSamples::k2X && !msaa_2x_supported()) {
    resource_desc.SampleDesc.Count = 4;
  } else {
    resource_desc.SampleDesc.Count = UINT(1) << UINT(key.msaa_samples);
  }
  resource_desc.SampleDesc.Quality = 0;
  resource_desc.Layout = D3D12_TEXTURE_LAYOUT_UNKNOWN;
  resource_desc.Flags = key.is_depth ? D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL
                                     : D3D12_RESOURCE_FLAG_ALLOW_RENDER_TARGET;
  // The first access will be ownership transfer into this render target or
  // starting to draw directly.
  D3D12_RESOURCE_STATES resource_state =
      key.is_depth ? D3D12_RESOURCE_STATE_DEPTH_WRITE
                   : D3D12_RESOURCE_STATE_RENDER_TARGET;
  D3D12_CLEAR_VALUE optimized_clear_value;
  if (key.is_depth) {
    optimized_clear_value.Format = GetDepthDSVDXGIFormat(key.GetDepthFormat());
    // Fixed-point depth is generally direct (1 being the farthest),
    // floating-point is used for more uniform precision across the range (0
    // being the farthest).
    optimized_clear_value.DepthStencil.Depth =
        key.GetDepthFormat() == xenos::DepthRenderTargetFormat::kD24S8 ? 1.0f
                                                                       : 0.0f;
    optimized_clear_value.DepthStencil.Stencil = 0;
  } else {
    optimized_clear_value.Format = GetColorDrawDXGIFormat(key.GetColorFormat());
    optimized_clear_value.Color[0] = 0.0f;
    optimized_clear_value.Color[1] = 0.0f;
    optimized_clear_value.Color[2] = 0.0f;
    optimized_clear_value.Color[3] = 0.0f;
  }
  // Create zeroed for more determinism, primarily with respect to compression
  // and depth float24 / float32 mirroring.
  Microsoft::WRL::ComPtr<ID3D12Resource> resource;
  if (FAILED(device->CreateCommittedResource(
          &ui::d3d12::util::kHeapPropertiesDefault, D3D12_HEAP_FLAG_NONE,
          &resource_desc, resource_state, &optimized_clear_value,
          IID_PPV_ARGS(&resource)))) {
    return nullptr;
  }
  {
    std::u16string resource_name = xe::to_utf16(key.GetDebugName());
    resource->SetName(reinterpret_cast<LPCWSTR>(resource_name.c_str()));
  }

  ui::d3d12::D3D12CpuDescriptorPool& descriptor_pool =
      key.is_depth ? *descriptor_pool_depth_ : *descriptor_pool_color_;
  ui::d3d12::D3D12CpuDescriptorPool::Descriptor descriptor_draw =
      descriptor_pool.AllocateDescriptor();
  ui::d3d12::D3D12CpuDescriptorPool::Descriptor descriptor_srv =
      descriptor_pool_srv_->AllocateDescriptor();
  if (!descriptor_draw.IsValid() || !descriptor_srv.IsValid()) {
    return nullptr;
  }
  D3D12_CPU_DESCRIPTOR_HANDLE descriptor_draw_handle =
      descriptor_draw.GetHandle();
  ui::d3d12::D3D12CpuDescriptorPool::Descriptor descriptor_draw_srgb;
  ui::d3d12::D3D12CpuDescriptorPool::Descriptor descriptor_load_separate;
  ui::d3d12::D3D12CpuDescriptorPool::Descriptor descriptor_srv_stencil;
  D3D12_SHADER_RESOURCE_VIEW_DESC srv_desc;
  srv_desc.Shader4ComponentMapping = D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING;
  if (resource_desc.SampleDesc.Count > 1) {
    srv_desc.ViewDimension = D3D12_SRV_DIMENSION_TEXTURE2DMS;
  } else {
    srv_desc.ViewDimension = D3D12_SRV_DIMENSION_TEXTURE2D;
    srv_desc.Texture2D.MostDetailedMip = 0;
    srv_desc.Texture2D.MipLevels = 1;
    srv_desc.Texture2D.PlaneSlice = 0;
    srv_desc.Texture2D.ResourceMinLODClamp = 0.0f;
  }
  if (key.is_depth) {
    // DSV and stencil SRV.
    descriptor_srv_stencil = descriptor_pool_srv_->AllocateDescriptor();
    if (!descriptor_srv_stencil.IsValid()) {
      return nullptr;
    }
    D3D12_DEPTH_STENCIL_VIEW_DESC dsv_desc;
    dsv_desc.Format = optimized_clear_value.Format;
    dsv_desc.Flags = D3D12_DSV_FLAG_NONE;
    D3D12_SHADER_RESOURCE_VIEW_DESC stencil_srv_desc;
    stencil_srv_desc.Format =
        GetDepthSRVStencilDXGIFormat(key.GetDepthFormat());
    stencil_srv_desc.Shader4ComponentMapping =
        D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING;
    if (resource_desc.SampleDesc.Count > 1) {
      dsv_desc.ViewDimension = D3D12_DSV_DIMENSION_TEXTURE2DMS;
      stencil_srv_desc.ViewDimension = D3D12_SRV_DIMENSION_TEXTURE2DMS;
    } else {
      dsv_desc.ViewDimension = D3D12_DSV_DIMENSION_TEXTURE2D;
      dsv_desc.Texture2D.MipSlice = 0;
      stencil_srv_desc.ViewDimension = D3D12_SRV_DIMENSION_TEXTURE2D;
      stencil_srv_desc.Texture2D.MostDetailedMip = 0;
      stencil_srv_desc.Texture2D.MipLevels = 1;
      stencil_srv_desc.Texture2D.PlaneSlice = 1;
      stencil_srv_desc.Texture2D.ResourceMinLODClamp = 0.0f;
    }
    device->CreateDepthStencilView(resource.Get(), &dsv_desc,
                                   descriptor_draw_handle);
    device->CreateShaderResourceView(resource.Get(), &stencil_srv_desc,
                                     descriptor_srv_stencil.GetHandle());
    // Depth SRV.
    srv_desc.Format = GetDepthSRVDepthDXGIFormat(key.GetDepthFormat());
  } else {
    // Drawing RTV.
    D3D12_RENDER_TARGET_VIEW_DESC rtv_desc;
    rtv_desc.Format = optimized_clear_value.Format;
    if (resource_desc.SampleDesc.Count > 1) {
      rtv_desc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2DMS;
    } else {
      rtv_desc.ViewDimension = D3D12_RTV_DIMENSION_TEXTURE2D;
      rtv_desc.Texture2D.MipSlice = 0;
      rtv_desc.Texture2D.PlaneSlice = 0;
    }
    device->CreateRenderTargetView(resource.Get(), &rtv_desc,
                                   descriptor_draw_handle);
    // sRGB drawing RTV.
    switch (key.GetColorFormat()) {
      case xenos::ColorRenderTargetFormat::k_8_8_8_8:
      case xenos::ColorRenderTargetFormat::k_8_8_8_8_GAMMA:
        if (gamma_render_target_as_srgb_) {
          descriptor_draw_srgb = descriptor_pool.AllocateDescriptor();
          if (!descriptor_draw_srgb.IsValid()) {
            return nullptr;
          }
          rtv_desc.Format = DXGI_FORMAT_R8G8B8A8_UNORM_SRGB;
          device->CreateRenderTargetView(resource.Get(), &rtv_desc,
                                         descriptor_draw_srgb.GetHandle());
        }
        break;
      default:
        break;
    }
    // Ownership transfer RTV.
    DXGI_FORMAT load_format =
        GetColorOwnershipTransferDXGIFormat(key.GetColorFormat());
    if (rtv_desc.Format != load_format) {
      descriptor_load_separate = descriptor_pool.AllocateDescriptor();
      if (!descriptor_load_separate.IsValid()) {
        return nullptr;
      }
      rtv_desc.Format = load_format;
      device->CreateRenderTargetView(resource.Get(), &rtv_desc,
                                     descriptor_load_separate.GetHandle());
    }
    // SRV for ownership transfer and dumping.
    srv_desc.Format = load_format;
  }
  device->CreateShaderResourceView(resource.Get(), &srv_desc,
                                   descriptor_srv.GetHandle());

  return new D3D12RenderTarget(
      key, resource.Get(), std::move(descriptor_draw),
      std::move(descriptor_draw_srgb), std::move(descriptor_load_separate),
      std::move(descriptor_srv), std::move(descriptor_srv_stencil),
      resource_state);
}

bool D3D12RenderTargetCache::IsHostDepthEncodingDifferent(
    xenos::DepthRenderTargetFormat format) const {
  if (format == xenos::DepthRenderTargetFormat::kD24FS8) {
    return depth_float24_conversion_ == DepthFloat24Conversion::kOnCopy;
  }
  return false;
}

void D3D12RenderTargetCache::RequestPixelShaderInterlockBarrier() {
  CommitEdramBufferUAVWrites();
}

void D3D12RenderTargetCache::TransitionEdramBuffer(
    D3D12_RESOURCE_STATES new_state) {
  if (command_processor_.PushTransitionBarrier(
          edram_buffer_, edram_buffer_state_, new_state)) {
    // Resetting edram_buffer_modification_status_ only if the barrier has been
    // truly inserted - in particular, not resetting it for UAV > UAV as
    // barriers are dropped if the state hasn't been changed.
    edram_buffer_modification_status_ =
        EdramBufferModificationStatus::kUnmodified;
  }
  edram_buffer_state_ = new_state;
}

void D3D12RenderTargetCache::MarkEdramBufferModified(
    EdramBufferModificationStatus modification_status) {
  assert_true(modification_status !=
              EdramBufferModificationStatus::kUnmodified);
  assert_true(edram_buffer_state_ == D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
  if (edram_buffer_state_ != D3D12_RESOURCE_STATE_UNORDERED_ACCESS) {
    return;
  }
  // max because being modified as a UAV requires stricter synchronization than
  // as ROV.
  edram_buffer_modification_status_ =
      std::max(edram_buffer_modification_status_, modification_status);
}

void D3D12RenderTargetCache::CommitEdramBufferUAVWrites(
    EdramBufferModificationStatus commit_status) {
  assert_true(commit_status != EdramBufferModificationStatus::kUnmodified);
  if (edram_buffer_modification_status_ < commit_status) {
    return;
  }
  assert_true(edram_buffer_state_ == D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
  if (edram_buffer_state_ == D3D12_RESOURCE_STATE_UNORDERED_ACCESS) {
    command_processor_.PushUAVBarrier(edram_buffer_);
  }
  edram_buffer_modification_status_ =
      EdramBufferModificationStatus::kUnmodified;
  PixelShaderInterlockFullEdramBarrierPlaced();
}

ID3D12PipelineState* const*
D3D12RenderTargetCache::GetOrCreateTransferPipelines(TransferShaderKey key) {
  const TransferModeInfo& mode = kTransferModes[size_t(key.mode)];
  bool dest_is_stencil_bit = (mode.output == TransferOutput::kStencilBit);

  if (dest_is_stencil_bit) {
    auto pipelines_it = transfer_stencil_bit_pipelines_.find(key);
    if (pipelines_it != transfer_stencil_bit_pipelines_.end()) {
      return pipelines_it->second[0] ? pipelines_it->second.data() : nullptr;
    }
  } else {
    auto pipeline_it = transfer_pipelines_.find(key);
    if (pipeline_it != transfer_pipelines_.end()) {
      return pipeline_it->second ? &pipeline_it->second : nullptr;
    }
  }

  uint32_t rs = kTransferUsedRootParameters[size_t(
      use_stencil_reference_output_ ? mode.root_signature_with_stencil_ref
                                    : mode.root_signature_no_stencil_ref)];

  // If not dest_is_color, it's depth, or stencil bit - 40-sample columns are
  // swapped as opposed to color source.
  bool dest_is_color = (mode.output == TransferOutput::kColor);

  xenos::ColorRenderTargetFormat dest_color_format =
      xenos::ColorRenderTargetFormat(key.dest_resource_format);
  xenos::DepthRenderTargetFormat dest_depth_format =
      xenos::DepthRenderTargetFormat(key.dest_resource_format);
  bool dest_is_64bpp =
      dest_is_color && xenos::IsColorRenderTargetFormat64bpp(dest_color_format);

  xenos::ColorRenderTargetFormat source_color_format =
      xenos::ColorRenderTargetFormat(key.source_resource_format);
  xenos::DepthRenderTargetFormat source_depth_format =
      xenos::DepthRenderTargetFormat(key.source_resource_format);
  // If not source_is_color, it's depth / stencil - 40-sample columns are
  // swapped as opposed to color destination.
  bool source_is_color = (rs & kTransferUsedRootParameterColorSRVBit) != 0;
  bool source_is_64bpp;
  uint32_t source_color_format_component_count;
  uint32_t source_color_srv_component_mask;
  bool source_color_is_uint;
  if (source_is_color) {
    assert_zero(rs & kTransferUsedRootParameterDepthSRVBit);
    assert_zero(rs & kTransferUsedRootParameterStencilSRVBit);
    source_is_64bpp =
        xenos::IsColorRenderTargetFormat64bpp(source_color_format);
    source_color_format_component_count =
        xenos::GetColorRenderTargetFormatComponentCount(source_color_format);
    if (dest_is_stencil_bit) {
      if (source_is_64bpp && !dest_is_64bpp) {
        // Need one component, but choosing from the two 32bpp halves of the
        // 64bpp sample.
        source_color_srv_component_mask =
            0b1 | (0b1 << (source_color_format_component_count >> 1));
      } else {
        // Red is at least 8 bits per component in all formats.
        source_color_srv_component_mask = 0b1;
      }
    } else {
      source_color_srv_component_mask =
          (uint32_t(1) << source_color_format_component_count) - 1;
    }
    GetColorOwnershipTransferDXGIFormat(source_color_format,
                                        &source_color_is_uint);
  } else {
    source_is_64bpp = false;
    source_color_format_component_count = 0;
    source_color_srv_component_mask = 0;
    source_color_is_uint = false;
  }

  bool shader_uses_stencil_reference_output =
      mode.output == TransferOutput::kDepth && use_stencil_reference_output_;

  // Because of built_shader_.resize(), pointers can't be kept persistently
  // here! Resizing also zeroes the memory.

  built_shader_.clear();

  // RDEF, ISGN, OSGN, SHEX, optionally SFI0, STAT.
  uint32_t blob_count = 5 + uint32_t(shader_uses_stencil_reference_output);

  // Allocate space for the container header and the blob offsets.
  built_shader_.resize(sizeof(dxbc::ContainerHeader) / sizeof(uint32_t) +
                       blob_count);
  uint32_t blob_offset_position_dwords =
      sizeof(dxbc::ContainerHeader) / sizeof(uint32_t);
  uint32_t blob_position_dwords = uint32_t(built_shader_.size());
  constexpr uint32_t kBlobHeaderSizeDwords =
      sizeof(dxbc::BlobHeader) / sizeof(uint32_t);

  uint32_t name_ptr;

  // ***************************************************************************
  // Resource definition
  // ***************************************************************************

  built_shader_[blob_offset_position_dwords] =
      uint32_t(blob_position_dwords * sizeof(uint32_t));
  uint32_t rdef_position_dwords = blob_position_dwords + kBlobHeaderSizeDwords;
  // Not needed, as the next operation done is resize, to allocate the space for
  // both the blob header and the resource definition header.
  // built_shader_.resize(rdef_position_dwords);

  // Allocate space for the RDEF header.
  built_shader_.resize(rdef_position_dwords +
                       sizeof(dxbc::RdefHeader) / sizeof(uint32_t));
  // Generator name.
  dxbc::AppendAlignedString(built_shader_, "Xenia");

  // Constant types - uint (aka "dword" when it's scalar) only.
  // Names.
  name_ptr = uint32_t((built_shader_.size() - rdef_position_dwords) *
                      sizeof(uint32_t));
  uint32_t rdef_dword_name_ptr = name_ptr;
  name_ptr += dxbc::AppendAlignedString(built_shader_, "dword");
  // Types.
  uint32_t rdef_type_uint_position_dwords = uint32_t(built_shader_.size());
  uint32_t rdef_type_uint_ptr =
      uint32_t((rdef_type_uint_position_dwords - rdef_position_dwords) *
               sizeof(uint32_t));
  built_shader_.resize(rdef_type_uint_position_dwords +
                       sizeof(dxbc::RdefType) / sizeof(uint32_t));
  {
    auto& rdef_type_uint = *reinterpret_cast<dxbc::RdefType*>(
        built_shader_.data() + rdef_type_uint_position_dwords);
    rdef_type_uint.variable_class = dxbc::RdefVariableClass::kScalar;
    rdef_type_uint.variable_type = dxbc::RdefVariableType::kUInt;
    rdef_type_uint.row_count = 1;
    rdef_type_uint.column_count = 1;
    rdef_type_uint.name_ptr = rdef_dword_name_ptr;
  }

  // Constants, if needed:
  // - uint xe_transfer_stencil_mask
  // - uint xe_transfer_address
  // - uint xe_transfer_host_depth_address
  uint32_t rdef_constant_count = 0;
  uint32_t rdef_constant_index_stencil_mask =
      (rs & kTransferUsedRootParameterStencilMaskConstantBit)
          ? rdef_constant_count++
          : UINT32_MAX;
  assert_false(dest_is_stencil_bit &&
               rdef_constant_index_stencil_mask == UINT32_MAX);
  uint32_t rdef_constant_index_address =
      (rs & kTransferUsedRootParameterAddressConstantBit)
          ? rdef_constant_count++
          : UINT32_MAX;
  assert_true(rdef_constant_index_address != UINT32_MAX);
  uint32_t rdef_constant_index_host_depth_address =
      (rs & kTransferUsedRootParameterHostDepthAddressConstantBit)
          ? rdef_constant_count++
          : UINT32_MAX;
  // Names.
  name_ptr = uint32_t((built_shader_.size() - rdef_position_dwords) *
                      sizeof(uint32_t));
  uint32_t rdef_xe_transfer_stencil_mask_name_ptr = name_ptr;
  if (rdef_constant_index_stencil_mask != UINT32_MAX) {
    name_ptr +=
        dxbc::AppendAlignedString(built_shader_, "xe_transfer_stencil_mask");
  }
  uint32_t rdef_xe_transfer_address_name_ptr = name_ptr;
  if (rdef_constant_index_address != UINT32_MAX) {
    name_ptr += dxbc::AppendAlignedString(built_shader_, "xe_transfer_address");
  }
  uint32_t rdef_xe_transfer_host_depth_address_name_ptr = name_ptr;
  if (rdef_constant_index_host_depth_address != UINT32_MAX) {
    name_ptr += dxbc::AppendAlignedString(built_shader_,
                                          "xe_transfer_host_depth_address");
  }
  // Constants.
  uint32_t rdef_constants_position_dwords = uint32_t(built_shader_.size());
  uint32_t rdef_constants_ptr =
      uint32_t((rdef_constants_position_dwords - rdef_position_dwords) *
               sizeof(uint32_t));
  built_shader_.resize(rdef_constants_position_dwords +
                       sizeof(dxbc::RdefVariable) / sizeof(uint32_t) *
                           rdef_constant_count);
  {
    auto rdef_constants = reinterpret_cast<dxbc::RdefVariable*>(
        built_shader_.data() + rdef_constants_position_dwords);
    // uint xe_transfer_stencil_mask
    if (rdef_constant_index_stencil_mask != UINT32_MAX) {
      dxbc::RdefVariable& rdef_constant_stencil_mask =
          rdef_constants[rdef_constant_index_stencil_mask];
      rdef_constant_stencil_mask.name_ptr =
          rdef_xe_transfer_stencil_mask_name_ptr;
      rdef_constant_stencil_mask.size_bytes = sizeof(uint32_t);
      rdef_constant_stencil_mask.flags = dxbc::kRdefVariableFlagUsed;
      rdef_constant_stencil_mask.type_ptr = rdef_type_uint_ptr;
      rdef_constant_stencil_mask.start_texture = UINT32_MAX;
      rdef_constant_stencil_mask.start_sampler = UINT32_MAX;
    }
    // uint xe_transfer_address
    if (rdef_constant_index_address != UINT32_MAX) {
      dxbc::RdefVariable& rdef_constant_address =
          rdef_constants[rdef_constant_index_address];
      rdef_constant_address.name_ptr = rdef_xe_transfer_address_name_ptr;
      rdef_constant_address.size_bytes = sizeof(uint32_t);
      rdef_constant_address.flags = dxbc::kRdefVariableFlagUsed;
      rdef_constant_address.type_ptr = rdef_type_uint_ptr;
      rdef_constant_address.start_texture = UINT32_MAX;
      rdef_constant_address.start_sampler = UINT32_MAX;
    }
    // uint xe_transfer_host_depth_address
    if (rdef_constant_index_host_depth_address != UINT32_MAX) {
      dxbc::RdefVariable& rdef_constant_host_depth_address =
          rdef_constants[rdef_constant_index_host_depth_address];
      rdef_constant_host_depth_address.name_ptr =
          rdef_xe_transfer_host_depth_address_name_ptr;
      rdef_constant_host_depth_address.size_bytes = sizeof(uint32_t);
      rdef_constant_host_depth_address.flags = dxbc::kRdefVariableFlagUsed;
      rdef_constant_host_depth_address.type_ptr = rdef_type_uint_ptr;
      rdef_constant_host_depth_address.start_texture = UINT32_MAX;
      rdef_constant_host_depth_address.start_sampler = UINT32_MAX;
    }
  }

  // Constant buffers, if needed:
  // - xe_transfer_stencil_mask { uint xe_transfer_stencil_mask; }
  // - xe_transfer_address { uint xe_transfer_address; }
  // - xe_transfer_host_depth_address { uint xe_transfer_host_depth_address; }
  // Reusing the constant names for constant buffers.
  uint32_t rdef_cbuffer_count = 0;
  uint32_t cbuffer_index_stencil_mask =
      rdef_constant_index_stencil_mask != UINT32_MAX ? rdef_cbuffer_count++
                                                     : UINT32_MAX;
  uint32_t cbuffer_index_address = rdef_constant_index_address != UINT32_MAX
                                       ? rdef_cbuffer_count++
                                       : UINT32_MAX;
  uint32_t cbuffer_index_host_depth_address =
      rdef_constant_index_host_depth_address != UINT32_MAX
          ? rdef_cbuffer_count++
          : UINT32_MAX;
  uint32_t rdef_cbuffer_position_dwords = uint32_t(built_shader_.size());
  built_shader_.resize(rdef_cbuffer_position_dwords +
                       sizeof(dxbc::RdefCbuffer) / sizeof(uint32_t) *
                           rdef_cbuffer_count);
  {
    auto rdef_cbuffers = reinterpret_cast<dxbc::RdefCbuffer*>(
        built_shader_.data() + rdef_cbuffer_position_dwords);
    // xe_transfer_stencil_mask
    if (cbuffer_index_stencil_mask != UINT32_MAX) {
      dxbc::RdefCbuffer& rdef_cbuffer_stencil_mask =
          rdef_cbuffers[cbuffer_index_stencil_mask];
      rdef_cbuffer_stencil_mask.name_ptr =
          rdef_xe_transfer_stencil_mask_name_ptr;
      rdef_cbuffer_stencil_mask.variable_count = 1;
      rdef_cbuffer_stencil_mask.variables_ptr =
          uint32_t(rdef_constants_ptr + sizeof(dxbc::RdefVariable) *
                                            rdef_constant_index_stencil_mask);
      rdef_cbuffer_stencil_mask.size_vector_aligned_bytes =
          sizeof(uint32_t) * 4;
    }
    // xe_transfer_address
    if (cbuffer_index_address != UINT32_MAX) {
      dxbc::RdefCbuffer& rdef_cbuffer_address =
          rdef_cbuffers[cbuffer_index_address];
      rdef_cbuffer_address.name_ptr = rdef_xe_transfer_address_name_ptr;
      rdef_cbuffer_address.variable_count = 1;
      rdef_cbuffer_address.variables_ptr =
          uint32_t(rdef_constants_ptr +
                   sizeof(dxbc::RdefVariable) * rdef_constant_index_address);
      rdef_cbuffer_address.size_vector_aligned_bytes = sizeof(uint32_t) * 4;
    }
    // xe_transfer_host_depth_address
    if (cbuffer_index_host_depth_address != UINT32_MAX) {
      dxbc::RdefCbuffer& rdef_cbuffer_host_depth_address =
          rdef_cbuffers[cbuffer_index_host_depth_address];
      rdef_cbuffer_host_depth_address.name_ptr =
          rdef_xe_transfer_host_depth_address_name_ptr;
      rdef_cbuffer_host_depth_address.variable_count = 1;
      rdef_cbuffer_host_depth_address.variables_ptr = uint32_t(
          rdef_constants_ptr +
          sizeof(dxbc::RdefVariable) * rdef_constant_index_host_depth_address);
      rdef_cbuffer_host_depth_address.size_vector_aligned_bytes =
          sizeof(uint32_t) * 4;
    }
  }

  // Bindings.
  // - Texture2D/Texture2DMS<floatN/uintN> xe_transfer_color
  // - Texture2D/Texture2DMS<float> xe_transfer_depth
  // - Texture2D/Texture2DMS<uint2> xe_transfer_stencil
  // - Texture2D<float>/Texture2DMS<float>/Buffer<uint> xe_transfer_host_depth
  // - Constant buffers
  uint32_t rdef_srv_count = 0;
  uint32_t srv_index_color = (rs & kTransferUsedRootParameterColorSRVBit)
                                 ? rdef_srv_count++
                                 : UINT32_MAX;
  uint32_t srv_index_depth = (rs & kTransferUsedRootParameterDepthSRVBit)
                                 ? rdef_srv_count++
                                 : UINT32_MAX;
  uint32_t srv_index_stencil = (rs & kTransferUsedRootParameterStencilSRVBit)
                                   ? rdef_srv_count++
                                   : UINT32_MAX;
  uint32_t srv_index_host_depth =
      (rs & kTransferUsedRootParameterHostDepthSRVBit) ? rdef_srv_count++
                                                       : UINT32_MAX;
  uint32_t rdef_binding_count = rdef_srv_count + rdef_cbuffer_count;
  // Names.
  name_ptr = uint32_t((built_shader_.size() - rdef_position_dwords) *
                      sizeof(uint32_t));
  uint32_t rdef_xe_transfer_color_name_ptr = name_ptr;
  if (srv_index_color != UINT32_MAX) {
    name_ptr += dxbc::AppendAlignedString(built_shader_, "xe_transfer_color");
  }
  uint32_t rdef_xe_transfer_depth_name_ptr = name_ptr;
  if (srv_index_depth != UINT32_MAX) {
    name_ptr += dxbc::AppendAlignedString(built_shader_, "xe_transfer_depth");
  }
  uint32_t rdef_xe_transfer_stencil_name_ptr = name_ptr;
  if (srv_index_stencil != UINT32_MAX) {
    name_ptr += dxbc::AppendAlignedString(built_shader_, "xe_transfer_stencil");
  }
  uint32_t rdef_xe_transfer_host_depth_name_ptr = name_ptr;
  if (srv_index_host_depth != UINT32_MAX) {
    name_ptr +=
        dxbc::AppendAlignedString(built_shader_, "xe_transfer_host_depth");
  }
  // Bindings.
  uint32_t rdef_binding_position_dwords = uint32_t(built_shader_.size());
  built_shader_.resize(rdef_binding_position_dwords +
                       sizeof(dxbc::RdefInputBind) / sizeof(uint32_t) *
                           rdef_binding_count);
  {
    auto rdef_bindings = reinterpret_cast<dxbc::RdefInputBind*>(
        built_shader_.data() + rdef_binding_position_dwords);
    uint32_t rdef_binding_index = 0;
    // xe_transfer_color
    if (srv_index_color != UINT32_MAX) {
      dxbc::RdefInputBind& rdef_binding_color =
          rdef_bindings[rdef_binding_index++];
      rdef_binding_color.name_ptr = rdef_xe_transfer_color_name_ptr;
      rdef_binding_color.type = dxbc::RdefInputType::kTexture;
      rdef_binding_color.return_type = source_color_is_uint
                                           ? dxbc::ResourceReturnType::kUInt
                                           : dxbc::ResourceReturnType::kFloat;
      if (key.source_msaa_samples != xenos::MsaaSamples::k1X) {
        rdef_binding_color.dimension = dxbc::RdefDimension::kSRVTexture2DMS;
      } else {
        rdef_binding_color.dimension = dxbc::RdefDimension::kSRVTexture2D;
        rdef_binding_color.sample_count = UINT32_MAX;
      }
      rdef_binding_color.bind_point = kTransferSRVRegisterColor;
      rdef_binding_color.bind_count = 1;
      assert_not_zero(source_color_srv_component_mask);
      rdef_binding_color.flags =
          (32 - xe::lzcnt(source_color_srv_component_mask) - 1)
          << dxbc::kRdefInputFlagsComponentsShift;
      rdef_binding_color.id = srv_index_color;
    }
    // xe_transfer_depth
    if (srv_index_depth != UINT32_MAX) {
      dxbc::RdefInputBind& rdef_binding_depth =
          rdef_bindings[rdef_binding_index++];
      rdef_binding_depth.name_ptr = rdef_xe_transfer_depth_name_ptr;
      rdef_binding_depth.type = dxbc::RdefInputType::kTexture;
      rdef_binding_depth.return_type = dxbc::ResourceReturnType::kFloat;
      if (key.source_msaa_samples != xenos::MsaaSamples::k1X) {
        rdef_binding_depth.dimension = dxbc::RdefDimension::kSRVTexture2DMS;
      } else {
        rdef_binding_depth.dimension = dxbc::RdefDimension::kSRVTexture2D;
        rdef_binding_depth.sample_count = UINT32_MAX;
      }
      rdef_binding_depth.bind_point = kTransferSRVRegisterDepth;
      rdef_binding_depth.bind_count = 1;
      rdef_binding_depth.id = srv_index_depth;
    }
    // xe_transfer_stencil
    if (srv_index_stencil != UINT32_MAX) {
      dxbc::RdefInputBind& rdef_binding_stencil =
          rdef_bindings[rdef_binding_index++];
      rdef_binding_stencil.name_ptr = rdef_xe_transfer_stencil_name_ptr;
      rdef_binding_stencil.type = dxbc::RdefInputType::kTexture;
      rdef_binding_stencil.return_type = dxbc::ResourceReturnType::kUInt;
      if (key.source_msaa_samples != xenos::MsaaSamples::k1X) {
        rdef_binding_stencil.dimension = dxbc::RdefDimension::kSRVTexture2DMS;
      } else {
        rdef_binding_stencil.dimension = dxbc::RdefDimension::kSRVTexture2D;
        rdef_binding_stencil.sample_count = UINT32_MAX;
      }
      rdef_binding_stencil.bind_point = kTransferSRVRegisterStencil;
      rdef_binding_stencil.bind_count = 1;
      rdef_binding_stencil.flags = dxbc::kRdefInputFlags2Component;
      rdef_binding_stencil.id = srv_index_stencil;
    }
    // xe_transfer_host_depth
    if (srv_index_host_depth != UINT32_MAX) {
      dxbc::RdefInputBind& rdef_binding_host_depth =
          rdef_bindings[rdef_binding_index++];
      rdef_binding_host_depth.name_ptr = rdef_xe_transfer_host_depth_name_ptr;
      rdef_binding_host_depth.type = dxbc::RdefInputType::kTexture;
      if (key.host_depth_source_is_copy) {
        // Float as uint.
        rdef_binding_host_depth.return_type = dxbc::ResourceReturnType::kUInt;
        rdef_binding_host_depth.dimension = dxbc::RdefDimension::kSRVBuffer;
        rdef_binding_host_depth.sample_count = UINT32_MAX;
      } else {
        rdef_binding_host_depth.return_type = dxbc::ResourceReturnType::kFloat;
        if (key.host_depth_source_msaa_samples != xenos::MsaaSamples::k1X) {
          rdef_binding_host_depth.dimension =
              dxbc::RdefDimension::kSRVTexture2DMS;
        } else {
          rdef_binding_host_depth.dimension =
              dxbc::RdefDimension::kSRVTexture2D;
          rdef_binding_host_depth.sample_count = UINT32_MAX;
        }
      }
      rdef_binding_host_depth.bind_point = kTransferSRVRegisterHostDepth;
      rdef_binding_host_depth.bind_count = 1;
      rdef_binding_host_depth.id = srv_index_host_depth;
    }
    // xe_transfer_stencil_mask
    if (cbuffer_index_stencil_mask != UINT32_MAX) {
      dxbc::RdefInputBind& rdef_binding_stencil_mask =
          rdef_bindings[rdef_binding_index++];
      rdef_binding_stencil_mask.name_ptr =
          rdef_xe_transfer_stencil_mask_name_ptr;
      rdef_binding_stencil_mask.type = dxbc::RdefInputType::kCbuffer;
      rdef_binding_stencil_mask.bind_point = kTransferCBVRegisterStencilMask;
      rdef_binding_stencil_mask.bind_count = 1;
      rdef_binding_stencil_mask.flags = dxbc::kRdefInputFlagUserPacked;
      rdef_binding_stencil_mask.id = cbuffer_index_stencil_mask;
    }
    // xe_transfer_address
    if (cbuffer_index_address != UINT32_MAX) {
      dxbc::RdefInputBind& rdef_binding_address =
          rdef_bindings[rdef_binding_index++];
      rdef_binding_address.name_ptr = rdef_xe_transfer_address_name_ptr;
      rdef_binding_address.type = dxbc::RdefInputType::kCbuffer;
      rdef_binding_address.bind_point = kTransferCBVRegisterAddress;
      rdef_binding_address.bind_count = 1;
      rdef_binding_address.flags = dxbc::kRdefInputFlagUserPacked;
      rdef_binding_address.id = cbuffer_index_address;
    }
    // xe_transfer_host_depth_address
    if (cbuffer_index_host_depth_address != UINT32_MAX) {
      dxbc::RdefInputBind& rdef_binding_host_depth_address =
          rdef_bindings[rdef_binding_index++];
      rdef_binding_host_depth_address.name_ptr =
          rdef_xe_transfer_host_depth_address_name_ptr;
      rdef_binding_host_depth_address.type = dxbc::RdefInputType::kCbuffer;
      rdef_binding_host_depth_address.bind_point =
          kTransferCBVRegisterHostDepthAddress;
      rdef_binding_host_depth_address.bind_count = 1;
      rdef_binding_host_depth_address.flags = dxbc::kRdefInputFlagUserPacked;
      rdef_binding_host_depth_address.id = cbuffer_index_host_depth_address;
    }
  }

  // Header.
  {
    auto& rdef_header = *reinterpret_cast<dxbc::RdefHeader*>(
        built_shader_.data() + rdef_position_dwords);
    rdef_header.cbuffer_count = rdef_cbuffer_count;
    rdef_header.cbuffers_ptr =
        uint32_t((rdef_cbuffer_position_dwords - rdef_position_dwords) *
                 sizeof(uint32_t));
    rdef_header.input_bind_count = rdef_binding_count;
    rdef_header.input_binds_ptr =
        uint32_t((rdef_binding_position_dwords - rdef_position_dwords) *
                 sizeof(uint32_t));
    rdef_header.shader_model = dxbc::RdefShaderModel::kPixelShader5_1;
    rdef_header.compile_flags =
        dxbc::kCompileFlagNoPreshader | dxbc::kCompileFlagPreferFlowControl |
        dxbc::kCompileFlagIeeeStrictness | dxbc::kCompileFlagAllResourcesBound;
    // Generator name is right after the header.
    rdef_header.generator_name_ptr = sizeof(dxbc::RdefHeader);
    rdef_header.fourcc = dxbc::RdefHeader::FourCC::k5_1;
    rdef_header.InitializeSizes();
  }

  {
    auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
        built_shader_.data() + blob_position_dwords);
    blob_header.fourcc = dxbc::BlobHeader::FourCC::kResourceDefinition;
    blob_position_dwords = uint32_t(built_shader_.size());
    blob_header.size_bytes =
        (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
        built_shader_[blob_offset_position_dwords++];
  }

  // ***************************************************************************
  // Input signature
  // ***************************************************************************

  // Registers for accessing in the shader code - multiple inputs may be packed
  // into the same register.
  enum InputRegister : uint32_t {
    kInputRegisterPosition,
    kInputRegisterSampleIndex,
    kInputRegisterCount,
  };

  // Position, and for multisampled, sample index.
  uint32_t isgn_parameter_count =
      1 + uint32_t(key.dest_msaa_samples != xenos::MsaaSamples::k1X);

  // Reserve space for the header and the parameters.
  built_shader_[blob_offset_position_dwords] =
      uint32_t(blob_position_dwords * sizeof(uint32_t));
  uint32_t isgn_position_dwords = blob_position_dwords + kBlobHeaderSizeDwords;
  built_shader_.resize(isgn_position_dwords +
                       sizeof(dxbc::Signature) / sizeof(uint32_t) +
                       sizeof(dxbc::SignatureParameter) / sizeof(uint32_t) *
                           isgn_parameter_count);

  // Names (after the parameters).
  name_ptr = uint32_t((built_shader_.size() - isgn_position_dwords) *
                      sizeof(uint32_t));
  uint32_t isgn_sv_position_name_ptr = name_ptr;
  name_ptr += dxbc::AppendAlignedString(built_shader_, "SV_Position");
  uint32_t isgn_sv_sample_index_name_ptr = name_ptr;
  if (key.dest_msaa_samples != xenos::MsaaSamples::k1X) {
    name_ptr += dxbc::AppendAlignedString(built_shader_, "SV_SampleIndex");
  }

  // Header and parameters.
  {
    // Header.
    auto& isgn_header = *reinterpret_cast<dxbc::Signature*>(
        built_shader_.data() + isgn_position_dwords);
    isgn_header.parameter_count = isgn_parameter_count;
    isgn_header.parameter_info_ptr = sizeof(dxbc::Signature);
    // Parameters.
    auto isgn_parameters = reinterpret_cast<dxbc::SignatureParameter*>(
        built_shader_.data() + isgn_position_dwords +
        sizeof(dxbc::Signature) / sizeof(uint32_t));
    uint32_t isgn_parameter_index = 0;
    // SV_Position.xy
    dxbc::SignatureParameter& isgn_sv_position =
        isgn_parameters[isgn_parameter_index++];
    isgn_sv_position.semantic_name_ptr = isgn_sv_position_name_ptr;
    isgn_sv_position.system_value = dxbc::Name::kPosition;
    isgn_sv_position.component_type =
        dxbc::SignatureRegisterComponentType::kFloat32;
    isgn_sv_position.register_index = kInputRegisterPosition;
    isgn_sv_position.mask = 0b1111;
    isgn_sv_position.always_reads_mask = 0b0011;
    // SV_SampleIndex
    if (key.dest_msaa_samples != xenos::MsaaSamples::k1X) {
      dxbc::SignatureParameter& isgn_sv_sample_index =
          isgn_parameters[isgn_parameter_index++];
      isgn_sv_sample_index.semantic_name_ptr = isgn_sv_sample_index_name_ptr;
      isgn_sv_sample_index.system_value = dxbc::Name::kSampleIndex;
      isgn_sv_sample_index.component_type =
          dxbc::SignatureRegisterComponentType::kUInt32;
      isgn_sv_sample_index.register_index = kInputRegisterSampleIndex;
      isgn_sv_sample_index.mask = 0b0001;
      isgn_sv_sample_index.always_reads_mask = 0b0001;
    }
  }

  {
    auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
        built_shader_.data() + blob_position_dwords);
    blob_header.fourcc = dxbc::BlobHeader::FourCC::kInputSignature;
    blob_position_dwords = uint32_t(built_shader_.size());
    blob_header.size_bytes =
        (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
        built_shader_[blob_offset_position_dwords++];
  }

  // ***************************************************************************
  // Output signature
  // ***************************************************************************

  // Color or depth.
  uint32_t osgn_parameter_count = 0;
  uint32_t osgn_parameter_index_sv_target =
      mode.output == TransferOutput::kColor ? osgn_parameter_count++
                                            : UINT32_MAX;
  uint32_t osgn_parameter_index_sv_depth = mode.output == TransferOutput::kDepth
                                               ? osgn_parameter_count++
                                               : UINT32_MAX;
  uint32_t osgn_parameter_index_sv_stencil_ref =
      shader_uses_stencil_reference_output ? osgn_parameter_count++
                                           : UINT32_MAX;

  // Reserve space for the header and the parameters.
  built_shader_[blob_offset_position_dwords] =
      uint32_t(blob_position_dwords * sizeof(uint32_t));
  uint32_t osgn_position_dwords = blob_position_dwords + kBlobHeaderSizeDwords;
  built_shader_.resize(osgn_position_dwords +
                       sizeof(dxbc::Signature) / sizeof(uint32_t) +
                       sizeof(dxbc::SignatureParameter) / sizeof(uint32_t) *
                           osgn_parameter_count);

  // Names (after the parameters).
  name_ptr = uint32_t((built_shader_.size() - osgn_position_dwords) *
                      sizeof(uint32_t));
  uint32_t osgn_sv_target_name_ptr = name_ptr;
  if (osgn_parameter_index_sv_target != UINT32_MAX) {
    name_ptr += dxbc::AppendAlignedString(built_shader_, "SV_Target");
  }
  uint32_t osgn_sv_depth_name_ptr = name_ptr;
  if (osgn_parameter_index_sv_depth != UINT32_MAX) {
    name_ptr += dxbc::AppendAlignedString(built_shader_, "SV_Depth");
  }
  uint32_t osgn_sv_stencil_ref_name_ptr = name_ptr;
  if (osgn_parameter_index_sv_stencil_ref != UINT32_MAX) {
    name_ptr += dxbc::AppendAlignedString(built_shader_, "SV_StencilRef");
  }

  bool dest_color_is_uint;
  if (mode.output == TransferOutput::kColor) {
    GetColorOwnershipTransferDXGIFormat(dest_color_format, &dest_color_is_uint);
  } else {
    dest_color_is_uint = false;
  }

  // Header and parameters.
  {
    // Header.
    auto& osgn_header = *reinterpret_cast<dxbc::Signature*>(
        built_shader_.data() + osgn_position_dwords);
    osgn_header.parameter_count = osgn_parameter_count;
    osgn_header.parameter_info_ptr = sizeof(dxbc::Signature);
    // Parameters.
    auto osgn_parameters = reinterpret_cast<dxbc::SignatureParameter*>(
        built_shader_.data() + osgn_position_dwords +
        sizeof(dxbc::Signature) / sizeof(uint32_t));
    // SV_Target
    if (osgn_parameter_index_sv_target != UINT32_MAX) {
      dxbc::SignatureParameter& osgn_sv_target =
          osgn_parameters[osgn_parameter_index_sv_target];
      osgn_sv_target.semantic_name_ptr = osgn_sv_target_name_ptr;
      osgn_sv_target.component_type =
          dest_color_is_uint ? dxbc::SignatureRegisterComponentType::kUInt32
                             : dxbc::SignatureRegisterComponentType::kFloat32;
      osgn_sv_target.register_index = 0;
      osgn_sv_target.mask = 0b1111;
    }
    // SV_Depth
    if (osgn_parameter_index_sv_depth != UINT32_MAX) {
      dxbc::SignatureParameter& osgn_sv_depth =
          osgn_parameters[osgn_parameter_index_sv_depth];
      osgn_sv_depth.semantic_name_ptr = osgn_sv_depth_name_ptr;
      osgn_sv_depth.component_type =
          dxbc::SignatureRegisterComponentType::kFloat32;
      osgn_sv_depth.register_index = UINT32_MAX;
      osgn_sv_depth.mask = 0b0001;
      osgn_sv_depth.never_writes_mask = 0b1110;
    }
    // SV_StencilRef
    if (osgn_parameter_index_sv_stencil_ref != UINT32_MAX) {
      dxbc::SignatureParameter& osgn_sv_stencil_ref =
          osgn_parameters[osgn_parameter_index_sv_stencil_ref];
      osgn_sv_stencil_ref.semantic_name_ptr = osgn_sv_stencil_ref_name_ptr;
      // Older versions of FXC incorrectly expect SV_StencilRef to be float,
      // it's always uint in DXC and also in the latest versions of FXC.
      osgn_sv_stencil_ref.component_type =
          dxbc::SignatureRegisterComponentType::kUInt32;
      osgn_sv_stencil_ref.register_index = UINT32_MAX;
      osgn_sv_stencil_ref.mask = 0b0001;
      osgn_sv_stencil_ref.never_writes_mask = 0b1110;
    }
  }

  {
    auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
        built_shader_.data() + blob_position_dwords);
    blob_header.fourcc = dxbc::BlobHeader::FourCC::kOutputSignature;
    blob_position_dwords = uint32_t(built_shader_.size());
    blob_header.size_bytes =
        (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
        built_shader_[blob_offset_position_dwords++];
  }

  // ***************************************************************************
  // Shader program
  // ***************************************************************************

  built_shader_[blob_offset_position_dwords] =
      uint32_t(blob_position_dwords * sizeof(uint32_t));
  uint32_t shex_position_dwords = blob_position_dwords + kBlobHeaderSizeDwords;
  built_shader_.resize(shex_position_dwords);

  built_shader_.push_back(
      dxbc::VersionToken(dxbc::ProgramType::kPixelShader, 5, 1));
  // Reserve space for the length token.
  built_shader_.push_back(0);

  dxbc::Statistics stat;
  std::memset(&stat, 0, sizeof(dxbc::Statistics));
  dxbc::Assembler a(built_shader_, stat);

  a.OpDclGlobalFlags(dxbc::kGlobalFlagAllResourcesBound);
  if (cbuffer_index_stencil_mask != UINT32_MAX) {
    a.OpDclConstantBuffer(
        dxbc::Src::CB(dxbc::Src::Dcl, cbuffer_index_stencil_mask,
                      kTransferCBVRegisterStencilMask,
                      kTransferCBVRegisterStencilMask),
        1);
  }
  if (cbuffer_index_address != UINT32_MAX) {
    a.OpDclConstantBuffer(
        dxbc::Src::CB(dxbc::Src::Dcl, cbuffer_index_address,
                      kTransferCBVRegisterAddress, kTransferCBVRegisterAddress),
        1);
  }
  if (cbuffer_index_host_depth_address != UINT32_MAX) {
    a.OpDclConstantBuffer(
        dxbc::Src::CB(dxbc::Src::Dcl, cbuffer_index_host_depth_address,
                      kTransferCBVRegisterHostDepthAddress,
                      kTransferCBVRegisterHostDepthAddress),
        1);
  }
  if (srv_index_color != UINT32_MAX) {
    a.OpDclResource(
        key.source_msaa_samples != xenos::MsaaSamples::k1X
            ? dxbc::ResourceDimension::kTexture2DMS
            : dxbc::ResourceDimension::kTexture2D,
        dxbc::ResourceReturnTypeX4Token(source_color_is_uint
                                            ? dxbc::ResourceReturnType::kUInt
                                            : dxbc::ResourceReturnType::kFloat),
        dxbc::Src::T(dxbc::Src::Dcl, srv_index_color, kTransferSRVRegisterColor,
                     kTransferSRVRegisterColor));
  }
  if (srv_index_depth != UINT32_MAX) {
    a.OpDclResource(
        key.source_msaa_samples != xenos::MsaaSamples::k1X
            ? dxbc::ResourceDimension::kTexture2DMS
            : dxbc::ResourceDimension::kTexture2D,
        dxbc::ResourceReturnTypeX4Token(dxbc::ResourceReturnType::kFloat),
        dxbc::Src::T(dxbc::Src::Dcl, srv_index_depth, kTransferSRVRegisterDepth,
                     kTransferSRVRegisterDepth));
  }
  if (srv_index_stencil != UINT32_MAX) {
    a.OpDclResource(
        key.source_msaa_samples != xenos::MsaaSamples::k1X
            ? dxbc::ResourceDimension::kTexture2DMS
            : dxbc::ResourceDimension::kTexture2D,
        dxbc::ResourceReturnTypeX4Token(dxbc::ResourceReturnType::kUInt),
        dxbc::Src::T(dxbc::Src::Dcl, srv_index_stencil,
                     kTransferSRVRegisterStencil, kTransferSRVRegisterStencil));
  }
  if (srv_index_host_depth != UINT32_MAX) {
    a.OpDclResource(
        key.host_depth_source_is_copy
            ? dxbc::ResourceDimension::kBuffer
            : (key.host_depth_source_msaa_samples != xenos::MsaaSamples::k1X
                   ? dxbc::ResourceDimension::kTexture2DMS
                   : dxbc::ResourceDimension::kTexture2D),
        dxbc::ResourceReturnTypeX4Token(dxbc::ResourceReturnType::kFloat),
        dxbc::Src::T(dxbc::Src::Dcl, srv_index_host_depth,
                     kTransferSRVRegisterHostDepth,
                     kTransferSRVRegisterHostDepth));
  }
  a.OpDclInputPSSIV(dxbc::InterpolationMode::kLinearNoPerspective,
                    dxbc::Dest::V1D(kInputRegisterPosition, 0b0011),
                    dxbc::Name::kPosition);
  if (key.dest_msaa_samples != xenos::MsaaSamples::k1X) {
    a.OpDclInputPSSGV(dxbc::Dest::V1D(kInputRegisterSampleIndex, 0b0001),
                      dxbc::Name::kSampleIndex);
  }
  if (osgn_parameter_index_sv_target != UINT32_MAX) {
    a.OpDclOutput(dxbc::Dest::O(0));
  }
  if (osgn_parameter_index_sv_depth != UINT32_MAX) {
    a.OpDclOutput(dxbc::Dest::ODepth());
  }
  if (osgn_parameter_index_sv_stencil_ref != UINT32_MAX) {
    a.OpDclOutput(dxbc::Dest::OStencilRef());
  }
  // r0:r2 are involved at least in common addressing code. Texture loads
  // usually can overwrite some of the addressing temps as they are only needed
  // for the coordinates for that load. Currently 3 temps are enough.
  a.OpDclTemps(3);

  uint32_t draw_resolution_scale_x = this->draw_resolution_scale_x();
  uint32_t draw_resolution_scale_y = this->draw_resolution_scale_y();

  uint32_t tile_width_samples_scaled =
      xenos::kEdramTileWidthSamples * draw_resolution_scale_x;
  uint32_t tile_height_samples_scaled =
      xenos::kEdramTileHeightSamples * draw_resolution_scale_y;

  // Split the destination pixel index into 32bpp tile in r0.z and
  // 32bpp-tile-relative pixel index in r0.xy.
  // r0.xy = pixel XY as uint
  a.OpFToU(dxbc::Dest::R(0, 0b0011), dxbc::Src::V1D(kInputRegisterPosition));
  uint32_t dest_sample_width_log2 =
      uint32_t(dest_is_64bpp) +
      uint32_t(key.dest_msaa_samples >= xenos::MsaaSamples::k4X);
  uint32_t dest_sample_height_log2 =
      uint32_t(key.dest_msaa_samples >= xenos::MsaaSamples::k2X);
  uint32_t dest_tile_width_divide_scale, dest_tile_width_divide_upper_shift;
  draw_util::GetEdramTileWidthDivideScaleAndUpperShift(
      draw_resolution_scale_x, dest_tile_width_divide_scale,
      dest_tile_width_divide_upper_shift);
  assert_true(dest_tile_width_divide_upper_shift >= dest_sample_width_log2);
  // Need the host tile size in pixels, not samples.
  dest_tile_width_divide_upper_shift -= dest_sample_width_log2;
  static_assert(
      TextureCache::kMaxDrawResolutionScaleAlongAxis <= 3,
      "D3D12RenderTargetCache EDRAM range ownership transfer shader generation "
      "supports Y draw resolution scaling factors of only up to 3");
  if (draw_resolution_scale_y == 3) {
    // r0.zw = upper 32 bits in the division process of pixel XY by pixel count
    // in a 32bpp tile
    a.OpUMul(dxbc::Dest::R(0, 0b1100), dxbc::Dest::Null(),
             dxbc::Src::R(0, 0b0100 << 4),
             dxbc::Src::LU(0, 0, dest_tile_width_divide_scale,
                           draw_util::kDivideScale3));
    // r0.zw = 32bpp tile XY index
    a.OpUShR(dxbc::Dest::R(0, 0b1100), dxbc::Src::R(0),
             dxbc::Src::LU(
                 0, 0, dest_tile_width_divide_upper_shift,
                 draw_util::kDivideUpperShift3 + 4 - dest_sample_height_log2));
    // r0.xy = destination pixel XY index within the 32bpp tile
    a.OpIMAd(
        dxbc::Dest::R(0, 0b0011), dxbc::Src::R(0, 0b1110),
        dxbc::Src::LI(
            -int32_t((80 * draw_resolution_scale_x) >> dest_sample_width_log2),
            -int32_t((16 * draw_resolution_scale_y) >> dest_sample_height_log2),
            0, 0),
        dxbc::Src::R(0, 0b0100));
  } else {
    assert_true(draw_resolution_scale_y <= 2);
    uint32_t dest_tile_height_pixels_log2 =
        (draw_resolution_scale_y == 2 ? 5 : 4) - dest_sample_height_log2;
    // r0.z = upper 32 bits in the division process of pixel X by pixel count in
    // a 32bpp tile
    a.OpUMul(dxbc::Dest::R(0, 0b0100), dxbc::Dest::Null(),
             dxbc::Src::R(0, dxbc::Src::kXXXX),
             dxbc::Src::LU(dest_tile_width_divide_scale));
    // r0.zw = 32bpp tile XY index
    a.OpUShR(dxbc::Dest::R(0, 0b1100), dxbc::Src::R(0, 0b0110 << 4),
             dxbc::Src::LU(0, 0, dest_tile_width_divide_upper_shift,
                           dest_tile_height_pixels_log2));
    // r0.x = destination pixel X index within the 32bpp tile
    a.OpIMAd(dxbc::Dest::R(0, 0b0001), dxbc::Src::R(0, dxbc::Src::kZZZZ),
             dxbc::Src::LI(-int32_t((80 * draw_resolution_scale_x) >>
                                    dest_sample_width_log2)),
             dxbc::Src::R(0, dxbc::Src::kXXXX));
    // r0.y = destination pixel Y index within the 32bpp tile
    a.OpAnd(dxbc::Dest::R(0, 0b0010), dxbc::Src::R(0, dxbc::Src::kYYYY),
            dxbc::Src::LU((1 << dest_tile_height_pixels_log2) - 1));
  }
  // r1.x = destination pitch in 32bpp tiles
  a.OpUBFE(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(xenos::kEdramPitchTilesBits),
           dxbc::Src::LU(0),
           dxbc::Src::CB(cbuffer_index_address, kTransferCBVRegisterAddress, 0,
                         dxbc::Src::kXXXX));
  // r0.z = 32bpp tile index relative to the destination base
  // r0.w = free
  // r1.x = free
  a.OpUMAd(dxbc::Dest::R(0, 0b0100), dxbc::Src::R(1, dxbc::Src::kXXXX),
           dxbc::Src::R(0, dxbc::Src::kWWWW),
           dxbc::Src::R(0, dxbc::Src::kZZZZ));

  // Now the tile index doesn't have any dependencies on the destination. The
  // dword index within the source tile, however, is calculated from both the
  // source and the destination pixel size, sample count and color vs. depth.

  // Source can be 64bpp or 32bpp - depth if only depth is available, color in
  // all other cases.

  // Load the source to r1 (or low to r0, high to r1 if need 64bpp color as the
  // result, as the address is loaded to r1).

  // Source pixel and sample index within the 32bpp tile.
  // X to r1.x (or keep r0.x if not modifying).
  // Y to r1.y (or keep r0.y if not modifying).
  // Sample index to r1.z (or use v# if not modifying); r1.z will also be set
  // to 0 before sampling for the LOD of the single-sampled source (needs to
  // be in the register).
  // If 64bpp -> 32bpp, also the needed half in r0.w.

  dxbc::Src dest_sample(
      dxbc::Src::V1D(kInputRegisterSampleIndex, dxbc::Src::kXXXX));
  dxbc::Src source_sample(dest_sample);
  uint32_t source_tile_pixel_x_reg = 0;
  uint32_t source_tile_pixel_y_reg = 0;

  // First sample bit at 4x in Direct3D 10.1+ - horizontal sample.
  // Second sample bit at 4x in Direct3D 10.1+ - vertical sample.
  // At 2x:
  // - Native 2x: top is 1 in Direct3D 10.1+, bottom is 0.
  // - 2x as 4x: top is 0, bottom is 3.

  if (!source_is_64bpp && dest_is_64bpp) {
    // 32bpp -> 64bpp, need two samples of the source.
    if (key.source_msaa_samples >= xenos::MsaaSamples::k4X) {
      // 32bpp -> 64bpp, 4x ->.
      // Source has 32bpp halves in two adjacent samples.
      if (key.dest_msaa_samples >= xenos::MsaaSamples::k4X) {
        // 32bpp -> 64bpp, 4x -> 4x.
        // 1 destination horizontal sample = 2 source horizontal samples.
        // D p0,0 s0,0 = S p0,0 s0,0 | S p0,0 s1,0
        // D p0,0 s1,0 = S p1,0 s0,0 | S p1,0 s1,0
        // D p0,0 s0,1 = S p0,0 s0,1 | S p0,0 s1,1
        // D p0,0 s1,1 = S p1,0 s0,1 | S p1,0 s1,1
        // Thus destination horizontal sample -> source horizontal pixel,
        // vertical samples are 1:1.
        a.OpAnd(dxbc::Dest::R(1, 0b0100), dest_sample, dxbc::Src::LU(0b10));
        source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
        a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(31), dxbc::Src::LU(1),
                dxbc::Src::R(0, dxbc::Src::kXXXX),
                dxbc::Src::V1D(kInputRegisterSampleIndex, dxbc::Src::kXXXX));
        source_tile_pixel_x_reg = 1;
      } else if (key.dest_msaa_samples == xenos::MsaaSamples::k2X) {
        // 32bpp -> 64bpp, 4x -> 2x.
        // 1 destination horizontal pixel = 2 source horizontal samples.
        // D p0,0 s0 = S p0,0 s0,0 | S p0,0 s1,0
        // D p0,0 s1 = S p0,0 s0,1 | S p0,0 s1,1
        // D p1,0 s0 = S p1,0 s0,0 | S p1,0 s1,0
        // D p1,0 s1 = S p1,0 s0,1 | S p1,0 s1,1
        // Pixel index can be reused. Sample 1 (for native 2x) or 0 (for 2x as
        // 4x) should become samples 01, sample 0 or 3 should become samples 23.
        source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
        if (msaa_2x_supported_) {
          a.OpXOr(dxbc::Dest::R(1, 0b0100), dest_sample, dxbc::Src::LU(1));
          a.OpIShL(dxbc::Dest::R(1, 0b0100), source_sample, dxbc::Src::LU(1));
        } else {
          a.OpAnd(dxbc::Dest::R(1, 0b0100), dest_sample, dxbc::Src::LU(0b10));
        }
      } else {
        // 32bpp -> 64bpp, 4x -> 1x.
        // 1 destination horizontal pixel = 2 source horizontal samples.
        // D p0,0 = S p0,0 s0,0 | S p0,0 s1,0
        // D p0,1 = S p0,0 s0,1 | S p0,0 s1,1
        // Horizontal pixel index can be reused. Vertical pixel 1 should
        // become sample 2.
        a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1), dxbc::Src::LU(1),
                dxbc::Src::R(0, dxbc::Src::kYYYY), dxbc::Src::LU(0));
        source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
        a.OpUShR(dxbc::Dest::R(1, 0b0010), dxbc::Src::R(0, dxbc::Src::kYYYY),
                 dxbc::Src::LU(1));
        source_tile_pixel_y_reg = 1;
      }
    } else {
      // 32bpp -> 64bpp, 1x/2x ->.
      // Source has 32bpp halves in two adjacent pixels.
      if (key.dest_msaa_samples >= xenos::MsaaSamples::k4X) {
        // 32bpp -> 64bpp, 1x/2x -> 4x.
        // The X part.
        // 1 destination horizontal sample = 2 source horizontal pixels.
        a.OpIShL(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(0, dxbc::Src::kXXXX),
                 dxbc::Src::LU(2));
        a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(1), dxbc::Src::LU(1),
                dxbc::Src::V1D(kInputRegisterSampleIndex, dxbc::Src::kXXXX),
                dxbc::Src::R(1, dxbc::Src::kXXXX));
        source_tile_pixel_x_reg = 1;
        // Y is handled by common code.
      } else {
        // 32bpp -> 64bpp, 1x/2x -> 1x/2x.
        // The X part.
        // 1 destination horizontal pixel = 2 source horizontal pixels.
        a.OpIShL(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(0, dxbc::Src::kXXXX),
                 dxbc::Src::LU(1));
        source_tile_pixel_x_reg = 1;
        // Y is handled by common code.
      }
    }
  } else if (source_is_64bpp && !dest_is_64bpp) {
    // 64bpp -> 32bpp, also the half to r0.w.
    if (key.dest_msaa_samples >= xenos::MsaaSamples::k4X) {
      // 64bpp -> 32bpp, -> 4x.
      // The needed half is in the destination horizontal sample index.
      if (key.source_msaa_samples >= xenos::MsaaSamples::k4X) {
        // 64bpp -> 32bpp, 4x -> 4x.
        // D p0,0 s0,0 = S s0,0 low
        // D p0,0 s1,0 = S s0,0 high
        // D p1,0 s0,0 = S s1,0 low
        // D p1,0 s1,0 = S s1,0 high
        // Vertical pixel and sample (second bit) addressing is the same.
        // However, 1 horizontal destination pixel = 1 horizontal source sample.
        a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1), dxbc::Src::LU(0),
                dxbc::Src::R(0, dxbc::Src::kXXXX), dest_sample);
        source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
        // 2 destination horizontal samples = 1 source horizontal sample, thus
        // 2 destination horizontal pixels = 1 source horizontal pixel.
        a.OpUShR(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(0, dxbc::Src::kXXXX),
                 dxbc::Src::LU(1));
        source_tile_pixel_x_reg = 1;
      } else {
        // 64bpp -> 32bpp, 1x/2x -> 4x.
        // 2 destination horizontal samples = 1 source horizontal pixel, thus
        // 1 destination horizontal pixel = 1 source horizontal pixel. Can reuse
        // horizontal pixel index.
        // Y is handled by common code.
      }
      // Half in r0.w from the destination horizontal sample index.
      a.OpAnd(dxbc::Dest::R(0, 0b1000), dest_sample, dxbc::Src::LU(1));
    } else {
      // 64bpp -> 32bpp, -> 1x/2x.
      // The needed half is in the destination horizontal pixel index.
      if (key.source_msaa_samples >= xenos::MsaaSamples::k4X) {
        // 64bpp -> 32bpp, 4x -> 1x/2x.
        // (Destination horizontal pixel >> 1) & 1 = source horizontal sample
        // (first bit).
        a.OpUBFE(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1), dxbc::Src::LU(1),
                 dxbc::Src::R(0, dxbc::Src::kXXXX));
        source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
        if (key.dest_msaa_samples == xenos::MsaaSamples::k2X) {
          // 64bpp -> 32bpp, 4x -> 2x.
          // Destination vertical samples (1/0 in the first bit for native 2x or
          // 0/1 in the second bit for 2x as 4x) = source vertical samples
          // (second bit).
          if (msaa_2x_supported_) {
            a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1),
                    dxbc::Src::LU(1), dest_sample, source_sample);
            a.OpXOr(dxbc::Dest::R(1, 0b0100), source_sample,
                    dxbc::Src::LU(1 << 1));
          } else {
            a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1),
                    dxbc::Src::LU(0), source_sample, dest_sample);
          }
        } else {
          // 64bpp -> 32bpp, 4x -> 1x.
          // 1 destination vertical pixel = 1 source vertical sample.
          a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1), dxbc::Src::LU(1),
                  dxbc::Src::R(0, dxbc::Src::kYYYY), source_sample);
          a.OpUShR(dxbc::Dest::R(1, 0b0010), dxbc::Src::R(0, dxbc::Src::kYYYY),
                   dxbc::Src::LU(1));
          source_tile_pixel_y_reg = 1;
        }
        // 2 destination horizontal pixels = 1 source horizontal sample.
        // 4 destination horizontal pixels = 1 source horizontal pixel.
        a.OpUShR(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(0, dxbc::Src::kXXXX),
                 dxbc::Src::LU(2));
        source_tile_pixel_x_reg = 1;
      } else {
        // 64bpp -> 32bpp, 1x/2x -> 1x/2x.
        // The X part.
        // 2 destination horizontal pixels = 1 destination source pixel.
        a.OpUShR(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(0, dxbc::Src::kXXXX),
                 dxbc::Src::LU(1));
        source_tile_pixel_x_reg = 1;
        // Y is handled by common code.
      }
      // Half in r0.w from the destination horizontal pixel index.
      a.OpAnd(dxbc::Dest::R(0, 0b1000), dxbc::Src::R(0, dxbc::Src::kXXXX),
              dxbc::Src::LU(1));
    }
  } else {
    // Same bit count.
    if (key.source_msaa_samples != key.dest_msaa_samples) {
      if (key.source_msaa_samples >= xenos::MsaaSamples::k4X) {
        // Same BPP, 4x -> 1x/2x.
        if (key.dest_msaa_samples == xenos::MsaaSamples::k2X) {
          // Same BPP, 4x -> 2x.
          // Horizontal pixels to samples. Vertical sample (1/0 in the first bit
          // for native 2x or 0/1 in the second bit for 2x as 4x) to second
          // sample bit.
          source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
          if (msaa_2x_supported_) {
            a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(31),
                    dxbc::Src::LU(1), dest_sample,
                    dxbc::Src::R(0, dxbc::Src::kXXXX));
            a.OpXOr(dxbc::Dest::R(1, 0b0100), source_sample,
                    dxbc::Src::LU(1 << 1));
          } else {
            a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1),
                    dxbc::Src::LU(0), dxbc::Src::R(0, dxbc::Src::kXXXX),
                    dest_sample);
          }
          a.OpUShR(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(0, dxbc::Src::kXXXX),
                   dxbc::Src::LU(1));
          source_tile_pixel_x_reg = 1;
        } else {
          // Same BPP, 4x -> 1x.
          // Pixels to samples.
          a.OpAnd(dxbc::Dest::R(1, 0b0100), dxbc::Src::R(0, dxbc::Src::kXXXX),
                  dxbc::Src::LU(1));
          source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
          a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1), dxbc::Src::LU(1),
                  dxbc::Src::R(0, dxbc::Src::kYYYY), source_sample);
          a.OpUShR(dxbc::Dest::R(1, 0b0011), dxbc::Src::R(0), dxbc::Src::LU(1));
          source_tile_pixel_x_reg = 1;
          source_tile_pixel_y_reg = 1;
        }
      } else {
        // Same BPP, 1x/2x -> 1x/2x/4x (as long as they're different).
        // Only the X part - Y is handled by common code.
        if (key.dest_msaa_samples >= xenos::MsaaSamples::k4X) {
          // Horizontal samples to pixels.
          a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(31), dxbc::Src::LU(1),
                  dxbc::Src::R(0, dxbc::Src::kXXXX), dest_sample);
          source_tile_pixel_x_reg = 1;
        }
      }
    }
  }
  // Common source Y and sample index for 1x/2x AA sources, independent of bits
  // per sample.
  if (key.source_msaa_samples < xenos::MsaaSamples::k4X &&
      key.source_msaa_samples != key.dest_msaa_samples) {
    if (key.dest_msaa_samples >= xenos::MsaaSamples::k4X) {
      // 1x/2x -> 4x.
      if (key.source_msaa_samples == xenos::MsaaSamples::k2X) {
        // 2x -> 4x.
        // Vertical samples (second bit) of 4x destination to vertical sample
        // (1, 0 for native 2x, or 0, 3 for 2x as 4x) of 2x source.
        a.OpUShR(dxbc::Dest::R(1, 0b0100), dest_sample, dxbc::Src::LU(1));
        source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
        if (msaa_2x_supported_) {
          a.OpXOr(dxbc::Dest::R(1, 0b0100), source_sample, dxbc::Src::LU(1));
        } else {
          a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1), dxbc::Src::LU(1),
                  source_sample, source_sample);
        }
      } else {
        // 1x -> 4x.
        // Vertical samples (second bit) to Y pixels.
        a.OpUShR(dxbc::Dest::R(1, 0b0010), dest_sample, dxbc::Src::LU(1));
        a.OpBFI(dxbc::Dest::R(1, 0b0010), dxbc::Src::LU(31), dxbc::Src::LU(1),
                dxbc::Src::R(0, dxbc::Src::kYYYY),
                dxbc::Src::R(1, dxbc::Src::kYYYY));
        source_tile_pixel_y_reg = 1;
      }
    } else {
      // 1x/2x -> different 1x/2x.
      if (key.source_msaa_samples == xenos::MsaaSamples::k2X) {
        // 2x -> 1x.
        // Vertical pixels of 2x destination to vertical samples (1, 0 for
        // native 2x, or 0, 3 for 2x as 4x) of 1x source.
        a.OpAnd(dxbc::Dest::R(1, 0b0100), dxbc::Src::R(0, dxbc::Src::kYYYY),
                dxbc::Src::LU(1));
        source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
        if (msaa_2x_supported_) {
          a.OpXOr(dxbc::Dest::R(1, 0b0100), source_sample, dxbc::Src::LU(1));
        } else {
          a.OpBFI(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(1), dxbc::Src::LU(1),
                  source_sample, source_sample);
        }
        a.OpUShR(dxbc::Dest::R(1, 0b0010), dxbc::Src::R(0, dxbc::Src::kYYYY),
                 dxbc::Src::LU(1));
        source_tile_pixel_y_reg = 1;
      } else {
        // 1x -> 2x.
        // Vertical samples (1/0 in the first bit for native 2x or 0/1 in the
        // second bit for 2x as 4x) of 2x destination to vertical pixels of 1x
        // source.
        if (msaa_2x_supported_) {
          a.OpBFI(dxbc::Dest::R(1, 0b0010), dxbc::Src::LU(31), dxbc::Src::LU(1),
                  dxbc::Src::R(0, dxbc::Src::kYYYY), dest_sample);
          a.OpXOr(dxbc::Dest::R(1, 0b0010), dxbc::Src::R(1, dxbc::Src::kYYYY),
                  dxbc::Src::LU(1));
        } else {
          a.OpUShR(dxbc::Dest::R(1, 0b0010), dest_sample, dxbc::Src::LU(1));
          a.OpBFI(dxbc::Dest::R(1, 0b0010), dxbc::Src::LU(31), dxbc::Src::LU(1),
                  dxbc::Src::R(0, dxbc::Src::kYYYY),
                  dxbc::Src::R(1, dxbc::Src::kYYYY));
        }
        source_tile_pixel_y_reg = 1;
      }
    }
  }

  uint32_t source_pixel_width_dwords_log2 =
      uint32_t(key.source_msaa_samples >= xenos::MsaaSamples::k4X) +
      uint32_t(source_is_64bpp);

  if (source_is_color != dest_is_color) {
    // Copying between color and depth / stencil - swap 40-32bpp-sample columns
    // in the pixel index within the source 32bpp tile using r1.w as temporary.
    uint32_t source_32bpp_tile_half_pixels =
        tile_width_samples_scaled >> (1 + source_pixel_width_dwords_log2);
    a.OpULT(dxbc::Dest::R(1, 0b1000),
            dxbc::Src::R(source_tile_pixel_x_reg, dxbc::Src::kXXXX),
            dxbc::Src::LU(source_32bpp_tile_half_pixels));
    a.OpMovC(dxbc::Dest::R(1, 0b1000), dxbc::Src::R(1, dxbc::Src::kWWWW),
             dxbc::Src::LI(int32_t(source_32bpp_tile_half_pixels)),
             dxbc::Src::LI(-int32_t(source_32bpp_tile_half_pixels)));
    a.OpIAdd(dxbc::Dest::R(1, 0b0001),
             dxbc::Src::R(source_tile_pixel_x_reg, dxbc::Src::kXXXX),
             dxbc::Src::R(1, dxbc::Src::kWWWW));
    source_tile_pixel_x_reg = 1;
    // r1.w = free
  }

  // Current register allocation:
  // r0.xy = pixel index within the destination 32bpp tile
  // r0.z = 32bpp tile index relative to the destination base
  // r0.w for 64bpp -> 32bpp - needed 32bpp half index of 64bpp data
  // r1.xy = pixel index within the source 32bpp tile
  // r1.z for 2x/4x -> = sample index within the source pixel

  // Apply the source 32bpp tile index.
  // r1.w = destination to source EDRAM tile adjustment
  a.OpIBFE(dxbc::Dest::R(1, 0b1000), dxbc::Src::LU(xenos::kEdramBaseTilesBits),
           dxbc::Src::LU(xenos::kEdramPitchTilesBits * 2),
           dxbc::Src::CB(cbuffer_index_address, kTransferCBVRegisterAddress, 0,
                         dxbc::Src::kXXXX));
  // r1.w = 32bpp tile index within the source
  a.OpIAdd(dxbc::Dest::R(1, 0b1000), dxbc::Src::R(0, dxbc::Src::kZZZZ),
           dxbc::Src::R(1, dxbc::Src::kWWWW));
  // r2.x = source pitch in 32bpp tiles
  a.OpUBFE(dxbc::Dest::R(2, 0b0001), dxbc::Src::LU(xenos::kEdramPitchTilesBits),
           dxbc::Src::LU(xenos::kEdramPitchTilesBits),
           dxbc::Src::CB(cbuffer_index_address, kTransferCBVRegisterAddress, 0,
                         dxbc::Src::kXXXX));
  // r1.w = source tile row
  // r2.x = source 32bpp tile within the row
  a.OpUDiv(dxbc::Dest::R(1, 0b1000), dxbc::Dest::R(2, 0b0001),
           dxbc::Src::R(1, dxbc::Src::kWWWW),
           dxbc::Src::R(2, dxbc::Src::kXXXX));
  // r1.x = pixel X within the source texture
  // r2.x = free
  a.OpUMAd(dxbc::Dest::R(1, 0b0001),
           dxbc::Src::LU(tile_width_samples_scaled >>
                         source_pixel_width_dwords_log2),
           dxbc::Src::R(2, dxbc::Src::kXXXX),
           dxbc::Src::R(source_tile_pixel_x_reg, dxbc::Src::kXXXX));
  // r1.y = pixel Y within the source texture
  // r1.w = free
  a.OpUMAd(dxbc::Dest::R(1, 0b0010),
           dxbc::Src::LU(
               tile_height_samples_scaled >>
               uint32_t(key.source_msaa_samples >= xenos::MsaaSamples::k2X)),
           dxbc::Src::R(1, dxbc::Src::kWWWW),
           dxbc::Src::R(source_tile_pixel_y_reg, dxbc::Src::kYYYY));

  // Load the source to r1, or, for 32bpp | 32bpp -> 64bpp, the first dword to
  // r0 since addressing will not be needed anymore for color, and the second
  // dword to r1.
  // Depth will be loaded to w before loading stencil (so it doesn't overwrite
  // the coordinates needed for stencil loading).
  // Stencil will be loaded to x.
  // Color will be loaded to x...w.
  bool source_load_is_two_dwords = !source_is_64bpp && dest_is_64bpp;
  if (key.source_msaa_samples != xenos::MsaaSamples::k1X) {
    for (uint32_t i = 0; i <= uint32_t(source_load_is_two_dwords); ++i) {
      uint32_t source_load_register = source_load_is_two_dwords ? i : 1;
      if (srv_index_depth != UINT32_MAX) {
        a.OpLdMS(dxbc::Dest::R(source_load_register, 0b1000), dxbc::Src::R(1),
                 0b0011,
                 dxbc::Src::T(srv_index_depth, kTransferSRVRegisterDepth,
                              dxbc::Src::kXXXX),
                 source_sample);
      }
      if (srv_index_stencil != UINT32_MAX) {
        a.OpLdMS(dxbc::Dest::R(source_load_register, 0b0001), dxbc::Src::R(1),
                 0b0011,
                 dxbc::Src::T(srv_index_stencil, kTransferSRVRegisterStencil,
                              dxbc::Src::kYYYY),
                 source_sample);
      } else if (srv_index_color != UINT32_MAX) {
        a.OpLdMS(dxbc::Dest::R(source_load_register,
                               source_color_srv_component_mask),
                 dxbc::Src::R(1), 0b0011,
                 dxbc::Src::T(srv_index_color, kTransferSRVRegisterColor),
                 source_sample);
      }
      if (source_load_is_two_dwords && !i) {
        // Go to the next sample or pixel along X if need to load two dwords.
        if (key.source_msaa_samples >= xenos::MsaaSamples::k4X) {
          a.OpOr(dxbc::Dest::R(1, 0b0100), source_sample, dxbc::Src::LU(1));
          source_sample = dxbc::Src::R(1, dxbc::Src::kZZZZ);
        } else {
          a.OpOr(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX),
                 dxbc::Src::LU(1));
        }
      }
    }
  } else {
    // Write zero to the LOD index in r1.z.
    a.OpMov(dxbc::Dest::R(1, 0b0100), dxbc::Src::LU(0));
    dxbc::Src source_coordinates(dxbc::Src::R(1, 0b10000100));
    for (uint32_t i = 0; i <= uint32_t(source_load_is_two_dwords); ++i) {
      uint32_t source_load_register = source_load_is_two_dwords ? i : 1;
      if (srv_index_depth != UINT32_MAX) {
        a.OpLd(dxbc::Dest::R(source_load_register, 0b1000), source_coordinates,
               0b1011,
               dxbc::Src::T(srv_index_depth, kTransferSRVRegisterDepth,
                            dxbc::Src::kXXXX));
      }
      if (srv_index_stencil != UINT32_MAX) {
        a.OpLd(dxbc::Dest::R(source_load_register, 0b0001), source_coordinates,
               0b1011,
               dxbc::Src::T(srv_index_stencil, kTransferSRVRegisterStencil,
                            dxbc::Src::kYYYY));
      } else if (srv_index_color != UINT32_MAX) {
        a.OpLd(dxbc::Dest::R(source_load_register,
                             source_color_srv_component_mask),
               source_coordinates, 0b1011,
               dxbc::Src::T(srv_index_color, kTransferSRVRegisterColor));
      }
      if (source_load_is_two_dwords && !i) {
        // Go to the next pixel along X if need to load two dwords.
        a.OpOr(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX),
               dxbc::Src::LU(1));
      }
    }
  }
  // Pick the needed 32bpp half of the 64bpp color based on r0.w.
  if (source_is_64bpp && !dest_is_64bpp) {
    uint32_t source_color_half_component_count =
        source_color_format_component_count >> 1;
    if (dest_is_stencil_bit) {
      a.OpMovC(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(0, dxbc::Src::kWWWW),
               dxbc::Src::R(1).Select(source_color_half_component_count),
               dxbc::Src::R(1, dxbc::Src::kXXXX));
    } else {
      uint32_t color_high_dword_swizzle =
          (source_color_half_component_count * 0b01010101) &
          ~((uint32_t(1) << (source_color_half_component_count * 2)) - 1);
      for (uint32_t i = 0; i < source_color_half_component_count; ++i) {
        color_high_dword_swizzle |= (source_color_half_component_count + i)
                                    << (i * 2);
      }
      a.OpMovC(dxbc::Dest::R(1, (1 << source_color_half_component_count) - 1),
               dxbc::Src::R(0, dxbc::Src::kWWWW),
               dxbc::Src::R(1, color_high_dword_swizzle), dxbc::Src::R(1));
    }
  }

  if (osgn_parameter_index_sv_stencil_ref != UINT32_MAX &&
      srv_index_stencil != UINT32_MAX) {
    // For the depth -> depth case, write the stencil loaded to r1.x directly to
    // the output.
    assert_true(mode.output == TransferOutput::kDepth);
    a.OpMov(dxbc::Dest::OStencilRef(), dxbc::Src::R(1, dxbc::Src::kXXXX));
  }

  if (dest_is_64bpp) {
    // Handle construction of 64bpp color, either from two 32-bit samples in r0
    // and r1, or from one 64bpp sample in r1. Using r2.x as temporary when
    // needed.
    // If color_packed_in_r0x_and_r1x, use the generic path for combining two
    // 32-bit samples - as raw in r0.x and r1.x - into the destination.
    bool color_packed_in_r0x_and_r1x = false;
    if (source_is_color) {
      switch (source_color_format) {
        case xenos::ColorRenderTargetFormat::k_8_8_8_8:
        case xenos::ColorRenderTargetFormat::k_8_8_8_8_GAMMA: {
          color_packed_in_r0x_and_r1x = true;
          for (uint32_t i = 0; i < 2; ++i) {
            a.OpMAd(dxbc::Dest::R(i), dxbc::Src::R(i), dxbc::Src::LF(255.0f),
                    dxbc::Src::LF(0.5f));
            a.OpFToU(dxbc::Dest::R(i), dxbc::Src::R(i));
            for (uint32_t j = 1; j < 4; ++j) {
              a.OpBFI(dxbc::Dest::R(i, 0b0001), dxbc::Src::LU(8),
                      dxbc::Src::LU(j * 8), dxbc::Src::R(i).Select(j),
                      dxbc::Src::R(i, dxbc::Src::kXXXX));
            }
          }
        } break;
        case xenos::ColorRenderTargetFormat::k_2_10_10_10:
        case xenos::ColorRenderTargetFormat::k_2_10_10_10_AS_10_10_10_10: {
          color_packed_in_r0x_and_r1x = true;
          for (uint32_t i = 0; i < 2; ++i) {
            a.OpMAd(dxbc::Dest::R(i), dxbc::Src::R(i),
                    dxbc::Src::LF(1023.0f, 1023.0f, 1023.0f, 3.0f),
                    dxbc::Src::LF(0.5f));
            a.OpFToU(dxbc::Dest::R(i), dxbc::Src::R(i));
            for (uint32_t j = 1; j < 4; ++j) {
              a.OpBFI(dxbc::Dest::R(i, 0b0001), dxbc::Src::LU(j == 3 ? 2 : 10),
                      dxbc::Src::LU(j * 10), dxbc::Src::R(i).Select(j),
                      dxbc::Src::R(i, dxbc::Src::kXXXX));
            }
          }
        } break;
        case xenos::ColorRenderTargetFormat::k_2_10_10_10_FLOAT:
        case xenos::ColorRenderTargetFormat::
            k_2_10_10_10_FLOAT_AS_16_16_16_16: {
          color_packed_in_r0x_and_r1x = true;
          for (uint32_t i = 0; i < 2; ++i) {
            // Float16 has a wider range for both color and alpha, also NaNs -
            // clamp and convert.
            for (uint32_t j = 0; j < 3; ++j) {
              DxbcShaderTranslator::UnclampedFloat32To7e3(a, i, j, i, j, 2, 0);
              if (j) {
                a.OpBFI(dxbc::Dest::R(i, 0b0001), dxbc::Src::LU(10),
                        dxbc::Src::LU(j * 10), dxbc::Src::R(i).Select(j),
                        dxbc::Src::R(i, dxbc::Src::kXXXX));
              }
            }
            // Saturate and convert the alpha.
            a.OpMov(dxbc::Dest::R(i, 0b1000), dxbc::Src::R(i, dxbc::Src::kWWWW),
                    true);
            a.OpMAd(dxbc::Dest::R(i, 0b1000), dxbc::Src::R(i, dxbc::Src::kWWWW),
                    dxbc::Src::LF(3.0f), dxbc::Src::LF(0.5f));
            a.OpFToU(dxbc::Dest::R(i, 0b1000),
                     dxbc::Src::R(i, dxbc::Src::kWWWW));
            a.OpBFI(dxbc::Dest::R(i, 0b0001), dxbc::Src::LU(2),
                    dxbc::Src::LU(30), dxbc::Src::R(i, dxbc::Src::kWWWW),
                    dxbc::Src::R(i, dxbc::Src::kXXXX));
          }
        } break;
        // All 64bpp formats, and all 16 bits per component formats, are
        // represented as integers in ownership transfer for safe handling of
        // NaNs and -32768 / -32767.
        case xenos::ColorRenderTargetFormat::k_16_16:
        case xenos::ColorRenderTargetFormat::k_16_16_FLOAT: {
          if (dest_color_format ==
              xenos::ColorRenderTargetFormat::k_32_32_FLOAT) {
            for (uint32_t i = 0; i < 2; ++i) {
              a.OpBFI(dxbc::Dest::O(0, 1 << i), dxbc::Src::LU(16),
                      dxbc::Src::LU(16), dxbc::Src::R(i, dxbc::Src::kYYYY),
                      dxbc::Src::R(i, dxbc::Src::kXXXX));
            }
          } else {
            a.OpMov(dxbc::Dest::O(0, 0b0011), dxbc::Src::R(0));
            a.OpMov(dxbc::Dest::O(0, 0b1100), dxbc::Src::R(1, 0b0100 << 4));
          }
        } break;
        case xenos::ColorRenderTargetFormat::k_16_16_16_16:
        case xenos::ColorRenderTargetFormat::k_16_16_16_16_FLOAT: {
          if (dest_color_format ==
              xenos::ColorRenderTargetFormat::k_32_32_FLOAT) {
            a.OpBFI(dxbc::Dest::O(0, 0b0011), dxbc::Src::LU(16),
                    dxbc::Src::LU(16), dxbc::Src::R(1, 0b1101),
                    dxbc::Src::R(1, 0b1000));
          } else {
            a.OpMov(dxbc::Dest::O(0), dxbc::Src::R(1));
          }
        } break;
        case xenos::ColorRenderTargetFormat::k_32_FLOAT: {
          color_packed_in_r0x_and_r1x = true;
        } break;
        case xenos::ColorRenderTargetFormat::k_32_32_FLOAT: {
          if (dest_color_format ==
              xenos::ColorRenderTargetFormat::k_32_32_FLOAT) {
            a.OpMov(dxbc::Dest::O(0, 0b0011), dxbc::Src::R(1));
          } else {
            a.OpUBFE(dxbc::Dest::O(0), dxbc::Src::LU(16),
                     dxbc::Src::LU(0, 16, 0, 16), dxbc::Src::R(1, 0b01010000));
          }
        } break;
      }
    } else {
      assert_not_zero(rs & kTransferUsedRootParameterDepthSRVBit);
      color_packed_in_r0x_and_r1x = true;
      for (uint32_t i = 0; i < 2; ++i) {
        switch (source_depth_format) {
          case xenos::DepthRenderTargetFormat::kD24S8: {
            // Round to the nearest even integer. This seems to be the correct,
            // adding +0.5 and rounding towards zero results in red instead of
            // black in the 4D5307E6 clear shader.
            a.OpMul(dxbc::Dest::R(i, 0b1000), dxbc::Src::R(i, dxbc::Src::kWWWW),
                    dxbc::Src::LF(float(0xFFFFFF)));
            a.OpRoundNE(dxbc::Dest::R(i, 0b1000),
                        dxbc::Src::R(i, dxbc::Src::kWWWW));
            a.OpFToU(dxbc::Dest::R(i, 0b1000),
                     dxbc::Src::R(i, dxbc::Src::kWWWW));
          } break;
          case xenos::DepthRenderTargetFormat::kD24FS8: {
            // Convert using r1.y as temporary.
            DxbcShaderTranslator::PreClampedDepthTo20e4(a, i, 3, i, 3, 1, 1,
                                                        true);
          } break;
        }
        // Merge depth and stencil into r0/r1.x.
        a.OpBFI(dxbc::Dest::R(i, 0b0001), dxbc::Src::LU(24), dxbc::Src::LU(8),
                dxbc::Src::R(i, dxbc::Src::kWWWW),
                dxbc::Src::R(i, dxbc::Src::kXXXX));
      }
    }
    if (color_packed_in_r0x_and_r1x) {
      if (dest_color_format == xenos::ColorRenderTargetFormat::k_32_32_FLOAT) {
        a.OpMov(dxbc::Dest::O(0, 0b0001), dxbc::Src::R(0, dxbc::Src::kXXXX));
        a.OpMov(dxbc::Dest::O(0, 0b0010), dxbc::Src::R(1, dxbc::Src::kXXXX));
      } else {
        for (uint32_t i = 0; i < 2; ++i) {
          a.OpUBFE(dxbc::Dest::O(0, 0b11 << (i * 2)), dxbc::Src::LU(16),
                   dxbc::Src::LU(0, 16, 0, 16),
                   dxbc::Src::R(i, dxbc::Src::kXXXX));
        }
      }
    }
  } else {
    // Handle a 32bpp destination (32bpp color, or depth / stencil). If
    // color_packed_in_r1x is true, a raw 32bpp color value was written, and
    // common handling will be done.
    bool color_packed_in_r1x = false;
    bool depth_loaded_in_guest_format = false;
    if (source_is_color) {
      switch (source_color_format) {
        case xenos::ColorRenderTargetFormat::k_8_8_8_8:
        case xenos::ColorRenderTargetFormat::k_8_8_8_8_GAMMA: {
          if (dest_is_stencil_bit) {
            a.OpMAd(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX),
                    dxbc::Src::LF(255.0f), dxbc::Src::LF(0.5f));
            a.OpFToU(dxbc::Dest::R(1, 0b0001),
                     dxbc::Src::R(1, dxbc::Src::kXXXX));
          } else if (dest_is_color &&
                     (dest_color_format ==
                          xenos::ColorRenderTargetFormat::k_8_8_8_8 ||
                      dest_color_format ==
                          xenos::ColorRenderTargetFormat::k_8_8_8_8_GAMMA)) {
            // Same format - passthrough.
            a.OpMov(dxbc::Dest::O(0), dxbc::Src::R(1));
          } else if (mode.output == TransferOutput::kDepth) {
            // When need only depth, not stencil, skip the red component.
            a.OpMAd(dxbc::Dest::R(
                        1, osgn_parameter_index_sv_stencil_ref != UINT32_MAX
                               ? 0b1111
                               : 0b1110),
                    dxbc::Src::R(1), dxbc::Src::LF(255.0f),
                    dxbc::Src::LF(0.5f));
            a.OpFToU(dxbc::Dest::R(1, 0b1110), dxbc::Src::R(1));
            if (osgn_parameter_index_sv_stencil_ref != UINT32_MAX) {
              // Write the red component to the stencil reference.
              a.OpFToU(dxbc::Dest::OStencilRef(),
                       dxbc::Src::R(1, dxbc::Src::kXXXX));
            }
            // Put depth in 0:23 of r1.w.
            // r1.y = 0xGGBB0000.
            a.OpBFI(dxbc::Dest::R(1, 0b0010), dxbc::Src::LU(8),
                    dxbc::Src::LU(8), dxbc::Src::R(1, dxbc::Src::kZZZZ),
                    dxbc::Src::R(1, dxbc::Src::kYYYY));
            // r1.w = 0xGGBBAA00.
            a.OpBFI(dxbc::Dest::R(1, 0b1000), dxbc::Src::LU(8),
                    dxbc::Src::LU(16), dxbc::Src::R(1, dxbc::Src::kWWWW),
                    dxbc::Src::R(1, dxbc::Src::kYYYY));
          } else {
            color_packed_in_r1x = true;
            a.OpMAd(dxbc::Dest::R(1), dxbc::Src::R(1), dxbc::Src::LF(255.0f),
                    dxbc::Src::LF(0.5f));
            a.OpFToU(dxbc::Dest::R(1), dxbc::Src::R(1));
            for (uint32_t i = 1; i < 4; ++i) {
              a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(8),
                      dxbc::Src::LU(i * 8), dxbc::Src::R(1).Select(i),
                      dxbc::Src::R(1, dxbc::Src::kXXXX));
            }
          }
        } break;
        case xenos::ColorRenderTargetFormat::k_2_10_10_10:
        case xenos::ColorRenderTargetFormat::k_2_10_10_10_AS_10_10_10_10: {
          if (dest_is_stencil_bit) {
            a.OpMAd(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX),
                    dxbc::Src::LF(1023.0f), dxbc::Src::LF(0.5f));
            a.OpFToU(dxbc::Dest::R(1, 0b0001),
                     dxbc::Src::R(1, dxbc::Src::kXXXX));
          } else if (dest_is_color &&
                     (dest_color_format ==
                          xenos::ColorRenderTargetFormat::k_2_10_10_10 ||
                      dest_color_format == xenos::ColorRenderTargetFormat::
                                               k_2_10_10_10_AS_10_10_10_10)) {
            a.OpMov(dxbc::Dest::O(0), dxbc::Src::R(1));
          } else {
            color_packed_in_r1x = true;
            a.OpMAd(dxbc::Dest::R(1), dxbc::Src::R(1),
                    dxbc::Src::LF(1023.0f, 1023.0f, 1023.0f, 3.0f),
                    dxbc::Src::LF(0.5f));
            a.OpFToU(dxbc::Dest::R(1), dxbc::Src::R(1));
            for (uint32_t i = 1; i < 4; ++i) {
              a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(i == 3 ? 2 : 10),
                      dxbc::Src::LU(i * 10), dxbc::Src::R(1).Select(i),
                      dxbc::Src::R(1, dxbc::Src::kXXXX));
            }
          }
        } break;
        case xenos::ColorRenderTargetFormat::k_2_10_10_10_FLOAT:
        case xenos::ColorRenderTargetFormat::
            k_2_10_10_10_FLOAT_AS_16_16_16_16: {
          if (dest_is_stencil_bit) {
            DxbcShaderTranslator::UnclampedFloat32To7e3(a, 1, 0, 1, 0, 2, 0);
          } else if (dest_is_color &&
                     (dest_color_format ==
                          xenos::ColorRenderTargetFormat::k_2_10_10_10_FLOAT ||
                      dest_color_format ==
                          xenos::ColorRenderTargetFormat::
                              k_2_10_10_10_FLOAT_AS_16_16_16_16)) {
            a.OpMov(dxbc::Dest::O(0), dxbc::Src::R(1));
          } else {
            color_packed_in_r1x = true;
            // Float16 has a wider range for both color and alpha, also NaNs -
            // clamp and convert.
            for (uint32_t i = 0; i < 3; ++i) {
              DxbcShaderTranslator::UnclampedFloat32To7e3(a, 1, i, 1, i, 2, 0);
              if (i) {
                a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(10),
                        dxbc::Src::LU(i * 10), dxbc::Src::R(1).Select(i),
                        dxbc::Src::R(1, dxbc::Src::kXXXX));
              }
            }
            // Saturate and convert the alpha.
            a.OpMov(dxbc::Dest::R(1, 0b1000), dxbc::Src::R(1, dxbc::Src::kWWWW),
                    true);
            a.OpMAd(dxbc::Dest::R(1, 0b1000), dxbc::Src::R(1, dxbc::Src::kWWWW),
                    dxbc::Src::LF(3.0f), dxbc::Src::LF(0.5f));
            a.OpFToU(dxbc::Dest::R(1, 0b1000),
                     dxbc::Src::R(1, dxbc::Src::kWWWW));
            a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(2),
                    dxbc::Src::LU(30), dxbc::Src::R(1, dxbc::Src::kWWWW),
                    dxbc::Src::R(1, dxbc::Src::kXXXX));
          }
        } break;
        case xenos::ColorRenderTargetFormat::k_16_16:
        case xenos::ColorRenderTargetFormat::k_16_16_16_16:
        case xenos::ColorRenderTargetFormat::k_16_16_FLOAT:
        case xenos::ColorRenderTargetFormat::k_16_16_16_16_FLOAT: {
          // All 16 bits per component formats are represented as integers in
          // ownership transfer for safe handling of NaNs and -32768 / -32767.
          if (dest_is_stencil_bit) {
            // High bits are not important for discarding, as only one bit is
            // checked - already loaded to red.
          } else if (dest_is_color &&
                     (dest_color_format ==
                          xenos::ColorRenderTargetFormat::k_16_16 ||
                      dest_color_format ==
                          xenos::ColorRenderTargetFormat::k_16_16_FLOAT)) {
            a.OpMov(dxbc::Dest::O(0, 0b0011), dxbc::Src::R(1));
          } else {
            color_packed_in_r1x = true;
            a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(16),
                    dxbc::Src::LU(16), dxbc::Src::R(1, dxbc::Src::kYYYY),
                    dxbc::Src::R(1, dxbc::Src::kXXXX));
          }
        } break;
        case xenos::ColorRenderTargetFormat::k_32_FLOAT:
        case xenos::ColorRenderTargetFormat::k_32_32_FLOAT: {
          color_packed_in_r1x = true;
        } break;
      }
    } else if (rs & kTransferUsedRootParameterDepthSRVBit) {
      if (dest_is_color || dest_depth_format != source_depth_format) {
        // Need to reinterpret the depth value as color or as a different depth
        // format. Convert the depth within r1.w.
        depth_loaded_in_guest_format = true;
        switch (source_depth_format) {
          case xenos::DepthRenderTargetFormat::kD24S8: {
            // Round to the nearest even integer. This seems to be the correct,
            // adding +0.5 and rounding towards zero results in red instead of
            // black in the 4D5307E6 clear shader.
            a.OpMul(dxbc::Dest::R(1, 0b1000), dxbc::Src::R(1, dxbc::Src::kWWWW),
                    dxbc::Src::LF(float(0xFFFFFF)));
            a.OpRoundNE(dxbc::Dest::R(1, 0b1000),
                        dxbc::Src::R(1, dxbc::Src::kWWWW));
            a.OpFToU(dxbc::Dest::R(1, 0b1000),
                     dxbc::Src::R(1, dxbc::Src::kWWWW));
          } break;
          case xenos::DepthRenderTargetFormat::kD24FS8: {
            // Convert using r1.y as temporary.
            DxbcShaderTranslator::PreClampedDepthTo20e4(a, 1, 3, 1, 3, 1, 1,
                                                        true);
          } break;
        }
        if (dest_is_color) {
          // Merge depth and stencil into r1.x for reinterpretation as color.
          color_packed_in_r1x = true;
          a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(24), dxbc::Src::LU(8),
                  dxbc::Src::R(1, dxbc::Src::kWWWW),
                  dxbc::Src::R(1, dxbc::Src::kXXXX));
        }
      }
    }
    switch (mode.output) {
      case TransferOutput::kColor:
        // Unless a special path was taken, unpack the raw 32bpp value into the
        // 32bpp color output. Any register can be used as temporary if needed -
        // this is the end of the shader.
        if (color_packed_in_r1x) {
          switch (dest_color_format) {
            case xenos::ColorRenderTargetFormat::k_8_8_8_8:
            case xenos::ColorRenderTargetFormat::k_8_8_8_8_GAMMA: {
              a.OpUBFE(dxbc::Dest::R(1), dxbc::Src::LU(8),
                       dxbc::Src::LU(0, 8, 16, 24),
                       dxbc::Src::R(1, dxbc::Src::kXXXX));
              a.OpUToF(dxbc::Dest::R(1), dxbc::Src::R(1));
              a.OpMul(dxbc::Dest::O(0), dxbc::Src::R(1),
                      dxbc::Src::LF(1.0f / 255.0f));
            } break;
            case xenos::ColorRenderTargetFormat::k_2_10_10_10:
            case xenos::ColorRenderTargetFormat::k_2_10_10_10_AS_10_10_10_10: {
              a.OpUBFE(dxbc::Dest::R(1), dxbc::Src::LU(10, 10, 10, 2),
                       dxbc::Src::LU(0, 10, 20, 30),
                       dxbc::Src::R(1, dxbc::Src::kXXXX));
              a.OpUToF(dxbc::Dest::R(1), dxbc::Src::R(1));
              a.OpMul(dxbc::Dest::O(0), dxbc::Src::R(1),
                      dxbc::Src::LF(1.0f / 1023.0f, 1.0f / 1023.0f,
                                    1.0f / 1023.0f, 1.0f / 3.0f));
            } break;
            case xenos::ColorRenderTargetFormat::k_2_10_10_10_FLOAT:
            case xenos::ColorRenderTargetFormat::
                k_2_10_10_10_FLOAT_AS_16_16_16_16: {
              // Color using r1.yz as temporary.
              for (uint32_t i = 0; i < 3; ++i) {
                DxbcShaderTranslator::Float7e3To32(a, dxbc::Dest::O(0, 1 << i),
                                                   1, 0, i * 10, 1, 1, 1, 2);
              }
              // Alpha.
              a.OpUBFE(dxbc::Dest::R(1, 0b1000), dxbc::Src::LU(2),
                       dxbc::Src::LU(30), dxbc::Src::R(1, dxbc::Src::kXXXX));
              a.OpUToF(dxbc::Dest::R(1, 0b1000),
                       dxbc::Src::R(1, dxbc::Src::kWWWW));
              a.OpMul(dxbc::Dest::O(0, 0b1000),
                      dxbc::Src::R(1, dxbc::Src::kWWWW),
                      dxbc::Src::LF(1.0f / 3.0f));
            } break;
            case xenos::ColorRenderTargetFormat::k_16_16:
            case xenos::ColorRenderTargetFormat::k_16_16_FLOAT: {
              // All 16 bits per component formats are represented as integers
              // in ownership transfer for safe handling of NaNs and
              // -32768 / -32767.
              a.OpUBFE(dxbc::Dest::O(0, 0b0011), dxbc::Src::LU(16),
                       dxbc::Src::LU(0, 16, 0, 0),
                       dxbc::Src::R(1, dxbc::Src::kXXXX));
            } break;
            case xenos::ColorRenderTargetFormat::k_32_FLOAT: {
              // Already as a 32-bit value.
              a.OpMov(dxbc::Dest::O(0, 0b0001),
                      dxbc::Src::R(1, dxbc::Src::kXXXX));
            } break;
            default:
              // A 64bpp format (handled separately) or an invalid one.
              assert_unhandled_case(dest_color_format);
          }
        }
        break;
      case TransferOutput::kDepth:
        if (source_is_color || depth_loaded_in_guest_format) {
          if (color_packed_in_r1x) {
            // Extract the depth bits to r1.w.
            a.OpUBFE(dxbc::Dest::R(1, 0b1000), dxbc::Src::LU(24),
                     dxbc::Src::LU(8), dxbc::Src::R(1, dxbc::Src::kXXXX));
            if (osgn_parameter_index_sv_stencil_ref != UINT32_MAX) {
              // Extract the stencil bits to the stencil reference.
              // The depth -> depth case is handled earlier, not long after
              // loading the stencil, for simplicity.
              a.OpUBFE(dxbc::Dest::OStencilRef(), dxbc::Src::LU(8),
                       dxbc::Src::LU(0), dxbc::Src::R(1, dxbc::Src::kXXXX));
            }
          }
          // r1.w contains the depth in the guest format. If a host depth source
          // is available, need to check if it's up to date - if it is, the host
          // precision value needs to be written. Otherwise, the new guest value
          // needs to be converted to the host format. Using `if` here because
          // it's likely that the values will either be the same - if not
          // modified - or different - if cleared or totally overwritten - in
          // large amounts of samples, usually whole waves, at once.
          if (rs & kTransferUsedRootParameterHostDepthSRVBit) {
            // Load the host float32 depth to r0.x, check if, when converted to
            // the guest format, it's the same as the guest source, thus up to
            // date, and if it is, write host float32 depth to r1.w, otherwise
            // do the guest -> host conversion on the `else` path.

            // Current register allocation:
            // r0.xy = pixel index within the destination 32bpp tile
            // r0.z = 32bpp tile index relative to the destination base
            // r1.w = depth in guest format

            if (key.host_depth_source_is_copy) {
              // Get the address in the EDRAM scratch buffer and load from
              // there.
              // The beginning of the buffer is (0, 0) of the destination.
              // 40-sample columns are not swapped for addressing simplicity
              // (because this is used for depth -> depth transfers, where
              // swapping isn't needed).
              // Convert samples to pixels.
              assert_true(key.host_depth_source_msaa_samples ==
                          xenos::MsaaSamples::k1X);
              if (key.dest_msaa_samples >= xenos::MsaaSamples::k2X) {
                if (key.dest_msaa_samples >= xenos::MsaaSamples::k4X) {
                  // Horizontal sample index in bit 0.
                  a.OpBFI(dxbc::Dest::R(0, 0b0001), dxbc::Src::LU(31),
                          dxbc::Src::LU(1), dxbc::Src::R(0, dxbc::Src::kXXXX),
                          dest_sample);
                }
                // Vertical sample index as 1 or 0 in bit 0 for true 2x or as 0
                // or 1 in bit 1 for 4x or for 2x emulated as 4x.
                if (key.dest_msaa_samples == xenos::MsaaSamples::k2X &&
                    msaa_2x_supported_) {
                  a.OpBFI(dxbc::Dest::R(0, 0b0010), dxbc::Src::LU(31),
                          dxbc::Src::LU(1), dxbc::Src::R(0, dxbc::Src::kYYYY),
                          dest_sample);
                  a.OpXOr(dxbc::Dest::R(0, 0b0010),
                          dxbc::Src::R(0, dxbc::Src::kYYYY), dxbc::Src::LU(1));
                } else {
                  // Using r0.w as a temporary.
                  a.OpUShR(dxbc::Dest::R(0, 0b1000), dest_sample,
                           dxbc::Src::LU(1));
                  a.OpBFI(dxbc::Dest::R(0, 0b0010), dxbc::Src::LU(31),
                          dxbc::Src::LU(1), dxbc::Src::R(0, dxbc::Src::kYYYY),
                          dxbc::Src::R(0, dxbc::Src::kWWWW));
                }
              }
              // Combine the tile sample index and the tile index into buffer
              // address to r0.x.
              a.OpUMAd(dxbc::Dest::R(0, 0b0001),
                       dxbc::Src::LU(tile_width_samples_scaled),
                       dxbc::Src::R(0, dxbc::Src::kYYYY),
                       dxbc::Src::R(0, dxbc::Src::kXXXX));
              a.OpUMAd(dxbc::Dest::R(0, 0b0001),
                       dxbc::Src::LU(tile_width_samples_scaled *
                                     tile_height_samples_scaled),
                       dxbc::Src::R(0, dxbc::Src::kZZZZ),
                       dxbc::Src::R(0, dxbc::Src::kXXXX));
              // Load from the buffer.
              a.OpLd(dxbc::Dest::R(0, 0b0001),
                     dxbc::Src::R(0, dxbc::Src::kXXXX), 0b0001,
                     dxbc::Src::T(srv_index_host_depth,
                                  kTransferSRVRegisterHostDepth,
                                  dxbc::Src::kXXXX));
            } else {
              // Adjust the tile index from the destination to the host depth
              // source.
              // r0.w = destination to host depth source EDRAM tile adjustment
              a.OpIBFE(dxbc::Dest::R(0, 0b1000),
                       dxbc::Src::LU(xenos::kEdramBaseTilesBits),
                       dxbc::Src::LU(xenos::kEdramPitchTilesBits * 2),
                       dxbc::Src::CB(cbuffer_index_host_depth_address,
                                     kTransferCBVRegisterHostDepthAddress, 0,
                                     dxbc::Src::kXXXX));
              // r0.z = tile index relative to the host depth source base
              // r0.w = free
              a.OpIAdd(dxbc::Dest::R(0, 0b0100),
                       dxbc::Src::R(0, dxbc::Src::kZZZZ),
                       dxbc::Src::R(0, dxbc::Src::kWWWW));
              // Convert position and sample index from within the destination
              // tile to within the host depth source tile, like for the guest
              // render target, but for 32bpp -> 32bpp only.
              dxbc::Src host_depth_source_sample(dest_sample);
              if (key.host_depth_source_msaa_samples != key.dest_msaa_samples) {
                if (key.host_depth_source_msaa_samples >=
                    xenos::MsaaSamples::k4X) {
                  // 4x -> 1x/2x.
                  if (key.dest_msaa_samples == xenos::MsaaSamples::k2X) {
                    // 4x -> 2x.
                    // Horizontal pixels to samples. Vertical sample (1, 0 in
                    // the first bit for native 2x or 0, 1 in the second bit for
                    // 2x as 4x) to second sample bit.
                    host_depth_source_sample =
                        dxbc::Src::R(0, dxbc::Src::kWWWW);
                    if (msaa_2x_supported_) {
                      a.OpBFI(dxbc::Dest::R(0, 0b1000), dxbc::Src::LU(31),
                              dxbc::Src::LU(1), dest_sample,
                              dxbc::Src::R(0, dxbc::Src::kXXXX));
                      a.OpXOr(dxbc::Dest::R(0, 0b1000),
                              host_depth_source_sample, dxbc::Src::LU(1 << 1));
                    } else {
                      a.OpBFI(dxbc::Dest::R(0, 0b1000), dxbc::Src::LU(1),
                              dxbc::Src::LU(0),
                              dxbc::Src::R(0, dxbc::Src::kXXXX), dest_sample);
                    }
                    a.OpUShR(dxbc::Dest::R(0, 0b0001),
                             dxbc::Src::R(0, dxbc::Src::kXXXX),
                             dxbc::Src::LU(1));
                  } else {
                    // 4x -> 1x.
                    // Pixels to samples.
                    a.OpAnd(dxbc::Dest::R(0, 0b1000),
                            dxbc::Src::R(0, dxbc::Src::kXXXX),
                            dxbc::Src::LU(1));
                    host_depth_source_sample =
                        dxbc::Src::R(0, dxbc::Src::kWWWW);
                    a.OpBFI(dxbc::Dest::R(0, 0b1000), dxbc::Src::LU(1),
                            dxbc::Src::LU(1), dxbc::Src::R(0, dxbc::Src::kYYYY),
                            host_depth_source_sample);
                    a.OpUShR(dxbc::Dest::R(0, 0b0011), dxbc::Src::R(0),
                             dxbc::Src::LU(1));
                  }
                } else {
                  // 1x/2x -> 1x/2x/4x (as long as they're different).
                  // Only the X part - Y is handled by common code.
                  if (key.dest_msaa_samples >= xenos::MsaaSamples::k4X) {
                    // Horizontal samples to pixels.
                    a.OpBFI(dxbc::Dest::R(0, 0b0001), dxbc::Src::LU(31),
                            dxbc::Src::LU(1), dxbc::Src::R(0, dxbc::Src::kXXXX),
                            dest_sample);
                  }
                }
                // Host depth source Y and sample index for 1x/2x AA sources.
                if (key.host_depth_source_msaa_samples <
                    xenos::MsaaSamples::k4X) {
                  if (key.dest_msaa_samples >= xenos::MsaaSamples::k4X) {
                    // 1x/2x -> 4x.
                    if (key.host_depth_source_msaa_samples ==
                        xenos::MsaaSamples::k2X) {
                      // 2x -> 4x.
                      // Vertical samples (second bit) of 4x destination to
                      // vertical sample (1, 0 for native 2x, or 0, 3 for 2x as
                      // 4x) of 2x source.
                      a.OpUShR(dxbc::Dest::R(0, 0b1000), dest_sample,
                               dxbc::Src::LU(1));
                      host_depth_source_sample =
                          dxbc::Src::R(0, dxbc::Src::kWWWW);
                      if (msaa_2x_supported_) {
                        a.OpXOr(dxbc::Dest::R(0, 0b1000),
                                host_depth_source_sample, dxbc::Src::LU(1));
                      } else {
                        a.OpBFI(dxbc::Dest::R(0, 0b1000), dxbc::Src::LU(1),
                                dxbc::Src::LU(1), host_depth_source_sample,
                                host_depth_source_sample);
                      }
                    } else {
                      // 1x -> 4x.
                      // Vertical samples (second bit) to Y pixels, using r0.w
                      // (not needed without source MSAA) as a temporary.
                      a.OpUShR(dxbc::Dest::R(0, 0b1000), dest_sample,
                               dxbc::Src::LU(1));
                      a.OpBFI(dxbc::Dest::R(0, 0b0010), dxbc::Src::LU(31),
                              dxbc::Src::LU(1),
                              dxbc::Src::R(0, dxbc::Src::kYYYY),
                              dxbc::Src::R(0, dxbc::Src::kWWWW));
                    }
                  } else {
                    // 1x/2x -> different 1x/2x.
                    if (key.host_depth_source_msaa_samples ==
                        xenos::MsaaSamples::k2X) {
                      // 2x -> 1x.
                      // Vertical pixels of 2x destination to vertical samples
                      // (1, 0 for native 2x, or 0, 3 for 2x as 4x) of 1x
                      // source.
                      a.OpAnd(dxbc::Dest::R(0, 0b1000),
                              dxbc::Src::R(0, dxbc::Src::kYYYY),
                              dxbc::Src::LU(1));
                      host_depth_source_sample =
                          dxbc::Src::R(0, dxbc::Src::kWWWW);
                      if (msaa_2x_supported_) {
                        a.OpXOr(dxbc::Dest::R(0, 0b1000),
                                host_depth_source_sample, dxbc::Src::LU(1));
                      } else {
                        a.OpBFI(dxbc::Dest::R(0, 0b1000), dxbc::Src::LU(1),
                                dxbc::Src::LU(1), host_depth_source_sample,
                                host_depth_source_sample);
                      }
                      a.OpUShR(dxbc::Dest::R(0, 0b0010),
                               dxbc::Src::R(0, dxbc::Src::kYYYY),
                               dxbc::Src::LU(1));
                    } else {
                      // 1x -> 2x.
                      // Vertical samples (1, 0 in the first bit for native 2x
                      // or 0, 1 in the second bit for 2x as 4x) of 2x
                      // destination to vertical pixels of 1x source.
                      // Using r0.w (not needed without source MSAA) as a
                      // temporary.
                      if (msaa_2x_supported_) {
                        a.OpBFI(dxbc::Dest::R(0, 0b0010), dxbc::Src::LU(31),
                                dxbc::Src::LU(1),
                                dxbc::Src::R(0, dxbc::Src::kYYYY), dest_sample);
                        a.OpXOr(dxbc::Dest::R(0, 0b0010),
                                dxbc::Src::R(0, dxbc::Src::kYYYY),
                                dxbc::Src::LU(1));
                      } else {
                        a.OpUShR(dxbc::Dest::R(0, 0b1000), dest_sample,
                                 dxbc::Src::LU(1));
                        a.OpBFI(dxbc::Dest::R(0, 0b0010), dxbc::Src::LU(31),
                                dxbc::Src::LU(1),
                                dxbc::Src::R(0, dxbc::Src::kYYYY),
                                dxbc::Src::R(0, dxbc::Src::kWWWW));
                      }
                    }
                  }
                }
              }
              // r1.x = host depth source pitch in tiles
              a.OpUBFE(dxbc::Dest::R(1, 0b0001),
                       dxbc::Src::LU(xenos::kEdramPitchTilesBits),
                       dxbc::Src::LU(xenos::kEdramPitchTilesBits),
                       dxbc::Src::CB(cbuffer_index_host_depth_address,
                                     kTransferCBVRegisterHostDepthAddress, 0,
                                     dxbc::Src::kXXXX));
              // r0.z = host depth source tile row
              // r1.x = host depth source tile within the row
              a.OpUDiv(dxbc::Dest::R(0, 0b0100), dxbc::Dest::R(1, 0b0001),
                       dxbc::Src::R(0, dxbc::Src::kZZZZ),
                       dxbc::Src::R(1, dxbc::Src::kXXXX));
              // r0.x = pixel X within the host depth source texture
              // r1.x = free
              a.OpUMAd(
                  dxbc::Dest::R(0, 0b0001),
                  dxbc::Src::LU(tile_width_samples_scaled >>
                                uint32_t(key.host_depth_source_msaa_samples >=
                                         xenos::MsaaSamples::k4X)),
                  dxbc::Src::R(1, dxbc::Src::kXXXX),
                  dxbc::Src::R(0, dxbc::Src::kXXXX));
              // r0.y = pixel Y within the host depth source texture
              // r0.z = free
              a.OpUMAd(
                  dxbc::Dest::R(0, 0b0010),
                  dxbc::Src::LU(tile_height_samples_scaled >>
                                uint32_t(key.host_depth_source_msaa_samples >=
                                         xenos::MsaaSamples::k2X)),
                  dxbc::Src::R(0, dxbc::Src::kZZZZ),
                  dxbc::Src::R(0, dxbc::Src::kYYYY));
              // Load from the host depth texture.
              if (key.host_depth_source_msaa_samples !=
                  xenos::MsaaSamples::k1X) {
                a.OpLdMS(dxbc::Dest::R(0, 0b0001), dxbc::Src::R(0), 0b0011,
                         dxbc::Src::T(srv_index_host_depth,
                                      kTransferSRVRegisterHostDepth,
                                      dxbc::Src::kXXXX),
                         host_depth_source_sample);
              } else {
                // Write zero to the LOD index in r0.z.
                a.OpMov(dxbc::Dest::R(0, 0b0100), dxbc::Src::LU(0));
                a.OpLd(dxbc::Dest::R(0, 0b0001), dxbc::Src::R(0, 0b10000100),
                       0b1011,
                       dxbc::Src::T(srv_index_host_depth,
                                    kTransferSRVRegisterHostDepth,
                                    dxbc::Src::kXXXX));
              }
            }
            // Convert the host depth value in r0.x to the guest format in r0.y
            // using r0.z as a temporary and check if it matches the value in
            // the currently owning guest render target.
            switch (dest_depth_format) {
              case xenos::DepthRenderTargetFormat::kD24S8: {
                // Round to the nearest even integer. This seems to be the
                // correct, adding +0.5 and rounding towards zero results in red
                // instead of black in the 4D5307E6 clear shader.
                a.OpMul(dxbc::Dest::R(0, 0b0010),
                        dxbc::Src::R(0, dxbc::Src::kXXXX),
                        dxbc::Src::LF(float(0xFFFFFF)));
                a.OpRoundNE(dxbc::Dest::R(0, 0b0010),
                            dxbc::Src::R(0, dxbc::Src::kYYYY));
                a.OpFToU(dxbc::Dest::R(0, 0b0010),
                         dxbc::Src::R(0, dxbc::Src::kYYYY));
              } break;
              case xenos::DepthRenderTargetFormat::kD24FS8: {
                DxbcShaderTranslator::PreClampedDepthTo20e4(a, 0, 1, 0, 0, 0, 2,
                                                            true);
              } break;
            }
            a.OpIEq(dxbc::Dest::R(0, 0b0010), dxbc::Src::R(0, dxbc::Src::kYYYY),
                    dxbc::Src::R(1, dxbc::Src::kWWWW));
            a.OpIf(true, dxbc::Src::R(0, dxbc::Src::kYYYY));
            // If the host depth is up to date, write it to oDepth at the host
            // precision instead of converting the guest depth.
            a.OpMov(dxbc::Dest::R(1, 0b1000),
                    dxbc::Src::R(0, dxbc::Src::kXXXX));
            a.OpElse();
          }
          // Convert using r0.x as a temporary.
          switch (dest_depth_format) {
            case xenos::DepthRenderTargetFormat::kD24S8: {
              // Multiplying by 1.0 / 0xFFFFFF produces an incorrect result (for
              // 0xC00000, for instance - which is 2_10_10_10 clear to 0001) -
              // rescale from 0...0xFFFFFF to 0...0x1000000 doing what true
              // float division followed by multiplication does (on x86-64 MSVC
              // with default SSE rounding) - values starting from 0x800000
              // become bigger by 1; then accurately bias the result's exponent.
              a.OpUShR(dxbc::Dest::R(0, 0b0001),
                       dxbc::Src::R(1, dxbc::Src::kWWWW), dxbc::Src::LU(23));
              a.OpIAdd(dxbc::Dest::R(1, 0b1000),
                       dxbc::Src::R(1, dxbc::Src::kWWWW),
                       dxbc::Src::R(0, dxbc::Src::kXXXX));
              a.OpUToF(dxbc::Dest::R(1, 0b1000),
                       dxbc::Src::R(1, dxbc::Src::kWWWW));
              a.OpMul(dxbc::Dest::R(1, 0b1000),
                      dxbc::Src::R(1, dxbc::Src::kWWWW),
                      dxbc::Src::LF(1.0f / float(1 << 24)));
            } break;
            case xenos::DepthRenderTargetFormat::kD24FS8: {
              DxbcShaderTranslator::Depth20e4To32(a, dxbc::Dest::R(1, 0b1000),
                                                  1, 3, 0, 1, 3, 0, 0, true);
            } break;
          }
          // Host depth is different, or not available - convert the guest depth
          // to the destination format.
          if (rs & kTransferUsedRootParameterHostDepthSRVBit) {
            // Close the conditional for the host / guest depth.
            a.OpEndIf();
          }
        }
        a.OpMov(dxbc::Dest::ODepth(), dxbc::Src::R(1, dxbc::Src::kWWWW));
        break;
      case TransferOutput::kStencilBit:
        // Discard the sample if the needed stencil bit is not set.
        assert_true(cbuffer_index_stencil_mask != UINT32_MAX);
        a.OpAnd(dxbc::Dest::R(0, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX),
                dxbc::Src::CB(cbuffer_index_stencil_mask,
                              kTransferCBVRegisterStencilMask, 0,
                              dxbc::Src::kXXXX));
        a.OpDiscard(false, dxbc::Src::R(0, dxbc::Src::kXXXX));
        break;
    }
  }

  if (dest_is_color) {
    // Fill the unused components of the color result.
    uint32_t dest_color_component_count =
        xenos::GetColorRenderTargetFormatComponentCount(dest_color_format);
    uint32_t dest_color_unwritten_mask =
        0b1111 & ~uint32_t((1 << dest_color_component_count) - 1);
    if (dest_color_component_count < 4) {
      a.OpMov(dxbc::Dest::O(0, dest_color_unwritten_mask), dxbc::Src::LU(0));
    }
  }

  a.OpRet();

  // Write the shader program length in dwords.
  built_shader_[shex_position_dwords + 1] =
      uint32_t(built_shader_.size()) - shex_position_dwords;

  {
    auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
        built_shader_.data() + blob_position_dwords);
    blob_header.fourcc = dxbc::BlobHeader::FourCC::kShaderEx;
    blob_position_dwords = uint32_t(built_shader_.size());
    blob_header.size_bytes =
        (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
        built_shader_[blob_offset_position_dwords++];
  }

  // ***************************************************************************
  // Shader feature info
  // ***************************************************************************

  if (shader_uses_stencil_reference_output) {
    built_shader_[blob_offset_position_dwords] =
        uint32_t(blob_position_dwords * sizeof(uint32_t));
    uint32_t sfi0_position_dwords =
        blob_position_dwords + kBlobHeaderSizeDwords;
    built_shader_.resize(sfi0_position_dwords +
                         sizeof(dxbc::ShaderFeatureInfo) / sizeof(uint32_t));
    auto& shader_feature_info = *reinterpret_cast<dxbc::ShaderFeatureInfo*>(
        built_shader_.data() + sfi0_position_dwords);
    shader_feature_info.feature_flags[0] |= dxbc::kShaderFeature0_StencilRef;
    {
      auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
          built_shader_.data() + blob_position_dwords);
      blob_header.fourcc = dxbc::BlobHeader::FourCC::kShaderFeatureInfo;
      blob_position_dwords = uint32_t(built_shader_.size());
      blob_header.size_bytes =
          (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
          built_shader_[blob_offset_position_dwords++];
    }
  }

  // ***************************************************************************
  // Statistics
  // ***************************************************************************

  built_shader_[blob_offset_position_dwords] =
      uint32_t(blob_position_dwords * sizeof(uint32_t));
  uint32_t stat_position_dwords = blob_position_dwords + kBlobHeaderSizeDwords;
  built_shader_.resize(stat_position_dwords +
                       sizeof(dxbc::Statistics) / sizeof(uint32_t));
  std::memcpy(built_shader_.data() + stat_position_dwords, &stat,
              sizeof(dxbc::Statistics));
  {
    auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
        built_shader_.data() + blob_position_dwords);
    blob_header.fourcc = dxbc::BlobHeader::FourCC::kStatistics;
    blob_position_dwords = uint32_t(built_shader_.size());
    blob_header.size_bytes =
        (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
        built_shader_[blob_offset_position_dwords++];
  }

  // ***************************************************************************
  // Container header
  // ***************************************************************************

  uint32_t built_shader_size_bytes =
      uint32_t(built_shader_.size() * sizeof(uint32_t));
  {
    auto& container_header =
        *reinterpret_cast<dxbc::ContainerHeader*>(built_shader_.data());
    container_header.InitializeIdentification();
    container_header.size_bytes = built_shader_size_bytes;
    container_header.blob_count = blob_count;
    CalculateDXBCChecksum(
        reinterpret_cast<unsigned char*>(built_shader_.data()),
        static_cast<unsigned int>(built_shader_size_bytes),
        reinterpret_cast<unsigned int*>(&container_header.hash));
  }

  // ***************************************************************************
  // Pipeline
  // ***************************************************************************

  ID3D12PipelineState* const* pipelines;
  ID3D12Device* device = command_processor_.GetD3D12Provider().GetDevice();
  D3D12_INPUT_ELEMENT_DESC pipeline_input_element_desc;
  pipeline_input_element_desc.SemanticName = "POSITION";
  pipeline_input_element_desc.SemanticIndex = 0;
  pipeline_input_element_desc.Format = DXGI_FORMAT_R32G32_FLOAT;
  pipeline_input_element_desc.InputSlot = 0;
  pipeline_input_element_desc.AlignedByteOffset = 0;
  pipeline_input_element_desc.InputSlotClass =
      D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA;
  pipeline_input_element_desc.InstanceDataStepRate = 0;
  D3D12_GRAPHICS_PIPELINE_STATE_DESC pipeline_desc = {};
  pipeline_desc.pRootSignature = transfer_root_signatures_[size_t(
      use_stencil_reference_output_ ? mode.root_signature_with_stencil_ref
                                    : mode.root_signature_no_stencil_ref)];
  pipeline_desc.VS.pShaderBytecode = shaders::passthrough_position_xy_vs;
  pipeline_desc.VS.BytecodeLength = sizeof(shaders::passthrough_position_xy_vs);
  pipeline_desc.PS.pShaderBytecode = built_shader_.data();
  pipeline_desc.PS.BytecodeLength = built_shader_size_bytes;
  if (key.dest_msaa_samples == xenos::MsaaSamples::k2X && !msaa_2x_supported_) {
    // Using sample 0 as 0 and 3 as 1 for 2x instead.
    pipeline_desc.SampleMask = 0b1001;
    pipeline_desc.SampleDesc.Count = 4;
  } else {
    pipeline_desc.SampleMask = UINT_MAX;
    pipeline_desc.SampleDesc.Count = UINT(1) << UINT(key.dest_msaa_samples);
  }
  pipeline_desc.RasterizerState.FillMode = D3D12_FILL_MODE_SOLID;
  pipeline_desc.RasterizerState.CullMode = D3D12_CULL_MODE_NONE;
  pipeline_desc.RasterizerState.DepthClipEnable = TRUE;
  pipeline_desc.InputLayout.pInputElementDescs = &pipeline_input_element_desc;
  pipeline_desc.InputLayout.NumElements = 1;
  pipeline_desc.PrimitiveTopologyType = D3D12_PRIMITIVE_TOPOLOGY_TYPE_TRIANGLE;
  if (dest_is_stencil_bit) {
    pipeline_desc.DepthStencilState.StencilEnable = TRUE;
    pipeline_desc.DepthStencilState.FrontFace.StencilFailOp =
        D3D12_STENCIL_OP_KEEP;
    pipeline_desc.DepthStencilState.FrontFace.StencilDepthFailOp =
        D3D12_STENCIL_OP_KEEP;
    pipeline_desc.DepthStencilState.FrontFace.StencilPassOp =
        D3D12_STENCIL_OP_REPLACE;
    pipeline_desc.DepthStencilState.FrontFace.StencilFunc =
        D3D12_COMPARISON_FUNC_ALWAYS;
    pipeline_desc.DepthStencilState.BackFace =
        pipeline_desc.DepthStencilState.FrontFace;
    pipeline_desc.DSVFormat = GetDepthDSVDXGIFormat(dest_depth_format);
    // Even if creation fails, still store the null pointers not to try to
    // create again.
    std::array<ID3D12PipelineState*, 8>& stencil_bit_pipelines =
        transfer_stencil_bit_pipelines_
            .emplace(std::piecewise_construct, std::make_tuple(key),
                     std::make_tuple())
            .first->second;
    bool stencil_pipelines_created = true;
    for (uint32_t i = 0; i < 8; ++i) {
      pipeline_desc.DepthStencilState.StencilWriteMask = UINT8(1) << i;
      if (SUCCEEDED(device->CreateGraphicsPipelineState(
              &pipeline_desc, IID_PPV_ARGS(&stencil_bit_pipelines[i])))) {
        continue;
      }
      stencil_pipelines_created = false;
      for (uint32_t j = 0; j < i; ++j) {
        stencil_bit_pipelines[j]->Release();
        stencil_bit_pipelines[j] = nullptr;
      }
      break;
    }
    pipelines =
        stencil_pipelines_created ? stencil_bit_pipelines.data() : nullptr;
  } else {
    if (dest_is_color) {
      pipeline_desc.BlendState.RenderTarget[0].RenderTargetWriteMask =
          D3D12_COLOR_WRITE_ENABLE_ALL;
      pipeline_desc.NumRenderTargets = 1;
      pipeline_desc.RTVFormats[0] =
          GetColorOwnershipTransferDXGIFormat(dest_color_format);
    } else {
      pipeline_desc.DepthStencilState.DepthEnable = TRUE;
      pipeline_desc.DepthStencilState.DepthWriteMask =
          D3D12_DEPTH_WRITE_MASK_ALL;
      pipeline_desc.DepthStencilState.DepthFunc =
          cvars::depth_transfer_not_equal_test ? D3D12_COMPARISON_FUNC_NOT_EQUAL
                                               : D3D12_COMPARISON_FUNC_ALWAYS;
      if (use_stencil_reference_output_) {
        pipeline_desc.DepthStencilState.StencilEnable = TRUE;
        pipeline_desc.DepthStencilState.StencilWriteMask = UINT8_MAX;
        pipeline_desc.DepthStencilState.FrontFace.StencilFailOp =
            D3D12_STENCIL_OP_KEEP;
        pipeline_desc.DepthStencilState.FrontFace.StencilDepthFailOp =
            cvars::depth_transfer_not_equal_test ? D3D12_STENCIL_OP_REPLACE
                                                 : D3D12_STENCIL_OP_KEEP;
        pipeline_desc.DepthStencilState.FrontFace.StencilPassOp =
            D3D12_STENCIL_OP_REPLACE;
        // Using ALWAYS, not NOT_EQUAL, so depth writing is unaffected by
        // stencil being different.
        pipeline_desc.DepthStencilState.FrontFace.StencilFunc =
            D3D12_COMPARISON_FUNC_ALWAYS;
        pipeline_desc.DepthStencilState.BackFace =
            pipeline_desc.DepthStencilState.FrontFace;
      }
      pipeline_desc.DSVFormat = GetDepthDSVDXGIFormat(dest_depth_format);
    }
    ID3D12PipelineState* pipeline;
    if (FAILED(device->CreateGraphicsPipelineState(&pipeline_desc,
                                                   IID_PPV_ARGS(&pipeline)))) {
      pipeline = nullptr;
    }
    // Even if creation fails, still store the null pointer not to try to create
    // again.
    // Return a pointer to the persistent location.
    ID3D12PipelineState*& inserted_pipeline =
        transfer_pipelines_.emplace(key, pipeline).first->second;
    pipelines = inserted_pipeline ? &inserted_pipeline : nullptr;
  }
  // TODO(Triang3l): Pipeline state name debug names (lots of variables - but
  // not very important since everything can be derived from the bindings and
  // outputs in a debugger).

  if (!pipelines) {
    // Stencil bit copying uses only the stencil SRV for depth / stencil source,
    // can't use srv_index_depth for checking.
    const char* source_format_name =
        (rs & kTransferUsedRootParameterColorSRVBit)
            ? xenos::GetColorRenderTargetFormatName(source_color_format)
            : xenos::GetDepthRenderTargetFormatName(source_depth_format);
    const char* dest_format_name =
        mode.output == TransferOutput::kColor
            ? xenos::GetColorRenderTargetFormatName(dest_color_format)
            : xenos::GetDepthRenderTargetFormatName(dest_depth_format);
    if (srv_index_host_depth != UINT32_MAX) {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create a render target ownership "
          "transfer pipeline for {}-sample {} + {}-sample host depth{} -> "
          "{}-sample {} for mode {}",
          uint32_t(1) << uint32_t(key.source_msaa_samples), source_format_name,
          uint32_t(1) << uint32_t(key.host_depth_source_msaa_samples),
          key.host_depth_source_is_copy ? " copy" : "",
          uint32_t(1) << uint32_t(key.dest_msaa_samples), dest_format_name,
          uint32_t(key.mode));
    } else {
      XELOGE(
          "D3D12RenderTargetCache: Failed to create a render target ownership "
          "transfer pipeline for {}-sample {} -> {}-sample {} for mode {}",
          uint32_t(1) << uint32_t(key.source_msaa_samples), source_format_name,
          uint32_t(1) << uint32_t(key.dest_msaa_samples), dest_format_name,
          uint32_t(key.mode));
    }
  }
  return pipelines;
}

void D3D12RenderTargetCache::PerformTransfersAndResolveClears(
    uint32_t render_target_count, RenderTarget* const* render_targets,
    const std::vector<Transfer>* render_target_transfers,
    const uint64_t* render_target_resolve_clear_values,
    const Transfer::Rectangle* resolve_clear_rectangle) {
  assert_true(GetPath() == Path::kHostRenderTargets);

  const ui::d3d12::D3D12Provider& provider =
      command_processor_.GetD3D12Provider();
  ID3D12Device* device = provider.GetDevice();
  uint64_t current_submission = command_processor_.GetCurrentSubmission();
  DeferredCommandList& command_list =
      command_processor_.GetDeferredCommandList();

  bool resolve_clear_needed =
      render_target_resolve_clear_values && resolve_clear_rectangle;
  D3D12_RECT clear_rect;
  if (resolve_clear_needed) {
    // Assuming the rectangle is already clamped by the setup function from the
    // common render target cache.
    clear_rect.left =
        LONG(resolve_clear_rectangle->x_pixels * draw_resolution_scale_x());
    clear_rect.top =
        LONG(resolve_clear_rectangle->y_pixels * draw_resolution_scale_y());
    clear_rect.right = LONG((resolve_clear_rectangle->x_pixels +
                             resolve_clear_rectangle->width_pixels) *
                            draw_resolution_scale_x());
    clear_rect.bottom = LONG((resolve_clear_rectangle->y_pixels +
                              resolve_clear_rectangle->height_pixels) *
                             draw_resolution_scale_y());
  }

  // Do host depth storing for the depth destination (assuming there can be only
  // one depth destination) where depth destination == host depth source.
  bool host_depth_store_set_up = false;
  for (uint32_t i = 0; i < render_target_count; ++i) {
    RenderTarget* dest_rt = render_targets[i];
    if (!dest_rt) {
      continue;
    }
    auto& dest_d3d12_rt = *static_cast<D3D12RenderTarget*>(dest_rt);
    RenderTargetKey dest_rt_key = dest_d3d12_rt.key();
    if (!dest_rt_key.is_depth) {
      continue;
    }
    const std::vector<Transfer>& depth_transfers = render_target_transfers[i];
    for (const Transfer& transfer : depth_transfers) {
      if (transfer.host_depth_source != dest_rt) {
        continue;
      }
      if (!host_depth_store_set_up) {
        // Bindings.
        // 0 - source.
        // 1 - EDRAM if bindful.
        ui::d3d12::util::DescriptorCpuGpuHandlePair
            host_depth_store_descriptors[2];
        if (!command_processor_.RequestOneUseSingleViewDescriptors(
                1 + uint32_t(!bindless_resources_used_),
                host_depth_store_descriptors)) {
          continue;
        }
        command_list.D3DSetComputeRootSignature(
            host_depth_store_root_signature_);
        // Destination (EDRAM uint4 buffer).
        if (bindless_resources_used_) {
          command_list.D3DSetComputeRootDescriptorTable(
              kHostDepthStoreRootParameterDest,
              command_processor_.GetEdramUintPow2BindlessUAVHandlePair(4)
                  .second);
        } else {
          const ui::d3d12::util::DescriptorCpuGpuHandlePair&
              host_depth_store_descriptor_dest =
                  host_depth_store_descriptors[1];
          WriteEdramUintPow2UAVDescriptor(
              host_depth_store_descriptor_dest.first, 4);
          command_list.D3DSetComputeRootDescriptorTable(
              kHostDepthStoreRootParameterDest,
              host_depth_store_descriptor_dest.second);
        }
        // Depth source texture.
        const ui::d3d12::util::DescriptorCpuGpuHandlePair&
            host_depth_store_descriptor_source =
                host_depth_store_descriptors[0];
        device->CopyDescriptorsSimple(
            1, host_depth_store_descriptor_source.first,
            dest_d3d12_rt.descriptor_srv().GetHandle(),
            D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
        command_list.D3DSetComputeRootDescriptorTable(
            kHostDepthStoreRootParameterSource,
            host_depth_store_descriptor_source.second);
        // Render target constant.
        HostDepthStoreRenderTargetConstant
            host_depth_store_render_target_constant =
                GetHostDepthStoreRenderTargetConstant(
                    dest_rt_key.pitch_tiles_at_32bpp, msaa_2x_supported_);
        command_list.D3DSetComputeRoot32BitConstants(
            kHostDepthStoreRootParameterConstants,
            sizeof(host_depth_store_render_target_constant) / sizeof(uint32_t),
            &host_depth_store_render_target_constant,
            offsetof(HostDepthStoreConstants, render_target) /
                sizeof(uint32_t));
        // Barriers - don't need to try to combine them with the rest of
        // render target transfer barriers now - if this happens, after host
        // depth storing, NON_PIXEL_SHADER_RESOURCE -> DEPTH_WRITE will be done
        // anyway even in the best case, so it's not possible to have all the
        // barriers in one place here.
        TransitionEdramBuffer(D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
        command_processor_.PushTransitionBarrier(
            dest_d3d12_rt.resource(),
            dest_d3d12_rt.SetResourceState(
                D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE),
            D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE);
        // Pipeline.
        command_processor_.SetExternalPipeline(
            host_depth_store_pipelines_[size_t(dest_rt_key.msaa_samples)]);
        host_depth_store_set_up = true;
      }
      Transfer::Rectangle
          transfer_rectangles[Transfer::kMaxRectanglesWithCutout];
      uint32_t transfer_rectangle_count = transfer.GetRectangles(
          dest_rt_key.base_tiles, dest_rt_key.pitch_tiles_at_32bpp,
          dest_rt_key.msaa_samples, false, transfer_rectangles,
          resolve_clear_rectangle);
      assert_not_zero(transfer_rectangle_count);
      HostDepthStoreRectangleConstant host_depth_store_rectangle_constant;
      for (uint32_t j = 0; j < transfer_rectangle_count; ++j) {
        uint32_t group_count_x, group_count_y;
        GetHostDepthStoreRectangleInfo(
            transfer_rectangles[j], dest_rt_key.msaa_samples,
            host_depth_store_rectangle_constant, group_count_x, group_count_y);
        command_list.D3DSetComputeRoot32BitConstants(
            kHostDepthStoreRootParameterConstants,
            sizeof(host_depth_store_rectangle_constant) / sizeof(uint32_t),
            &host_depth_store_rectangle_constant,
            offsetof(HostDepthStoreConstants, rectangle) / sizeof(uint32_t));
        command_processor_.SubmitBarriers();
        command_list.D3DDispatch(group_count_x, group_count_y, 1);
        MarkEdramBufferModified();
      }
    }
    break;
  }

  // Try to insert as many barriers as possible in one place, hoping that in the
  // best case (no cross-copying between current render targets), barriers will
  // need to be only inserted here, not between transfers. In case of
  // cross-copying, if the destination use is going to happen before the source
  // use, choose the destination state, otherwise the source state - to match
  // the order in which transfers will actually happen (otherwise there will be
  // just a useless switch back and forth).
  for (uint32_t i = 0; i < render_target_count; ++i) {
    RenderTarget* dest_rt = render_targets[i];
    if (!dest_rt) {
      continue;
    }
    auto& dest_d3d12_rt = *static_cast<D3D12RenderTarget*>(dest_rt);
    const std::vector<Transfer>& dest_transfers = render_target_transfers[i];
    if (!resolve_clear_needed && dest_transfers.empty()) {
      continue;
    }
    // Transition the sources, only if not going to be used as destinations
    // earlier.
    for (const Transfer& transfer : render_target_transfers[i]) {
      bool source_previously_used_as_dest = false;
      bool host_depth_source_previously_used_as_dest = false;
      for (uint32_t j = 0; j < i; ++j) {
        if (render_target_transfers[j].empty()) {
          continue;
        }
        const RenderTarget* previous_rt = render_targets[j];
        if (transfer.source == previous_rt) {
          source_previously_used_as_dest = true;
        }
        if (transfer.host_depth_source == previous_rt) {
          host_depth_source_previously_used_as_dest = true;
        }
      }
      if (!source_previously_used_as_dest) {
        auto& source_d3d12_rt =
            *static_cast<D3D12RenderTarget*>(transfer.source);
        command_processor_.PushTransitionBarrier(
            source_d3d12_rt.resource(),
            source_d3d12_rt.SetResourceState(
                D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE),
            D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE);
      }
      // transfer.host_depth_source == dest_rt means the EDRAM buffer will be
      // used instead, no need to transition.
      if (transfer.host_depth_source && transfer.host_depth_source != dest_rt &&
          !host_depth_source_previously_used_as_dest) {
        auto& host_depth_source_d3d12_rt =
            *static_cast<D3D12RenderTarget*>(transfer.host_depth_source);
        command_processor_.PushTransitionBarrier(
            host_depth_source_d3d12_rt.resource(),
            host_depth_source_d3d12_rt.SetResourceState(
                D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE),
            D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE);
      }
    }
    // Transition the destination, only if not going to be used as a source
    // earlier.
    bool dest_used_previously_as_source = false;
    for (uint32_t j = 0; j < i; ++j) {
      for (const Transfer& previous_transfer : render_target_transfers[j]) {
        if (previous_transfer.source == dest_rt ||
            previous_transfer.host_depth_source == dest_rt) {
          dest_used_previously_as_source = true;
          break;
        }
      }
    }
    if (!dest_used_previously_as_source) {
      D3D12_RESOURCE_STATES dest_state =
          dest_d3d12_rt.key().is_depth ? D3D12_RESOURCE_STATE_DEPTH_WRITE
                                       : D3D12_RESOURCE_STATE_RENDER_TARGET;
      command_processor_.PushTransitionBarrier(
          dest_d3d12_rt.resource(), dest_d3d12_rt.SetResourceState(dest_state),
          dest_state);
    }
  }
  if (host_depth_store_set_up) {
    // Will be reading copied host depth from the EDRAM buffer.
    TransitionEdramBuffer(D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE);
  }

  // Copy source descriptors to the shader-visible heap.
  // Clear previously set shader-visible descriptor indices.
  for (uint32_t i = 0; i < render_target_count; ++i) {
    RenderTarget* dest_rt = render_targets[i];
    if (!dest_rt) {
      continue;
    }
    for (const Transfer& transfer : render_target_transfers[i]) {
      assert_not_null(transfer.source);
      auto& source_d3d12_rt = *static_cast<D3D12RenderTarget*>(transfer.source);
      source_d3d12_rt.SetTemporarySRVDescriptorIndex(UINT32_MAX);
      source_d3d12_rt.SetTemporarySRVDescriptorIndexStencil(UINT32_MAX);
      auto* host_depth_source_d3d12_rt =
          static_cast<D3D12RenderTarget*>(transfer.host_depth_source);
      if (host_depth_source_d3d12_rt) {
        host_depth_source_d3d12_rt->SetTemporarySRVDescriptorIndex(UINT32_MAX);
        host_depth_source_d3d12_rt->SetTemporarySRVDescriptorIndexStencil(
            UINT32_MAX);
      }
    }
  }
  current_temporary_descriptors_cpu_.clear();
  uint32_t host_depth_copy_srv_index;
  if (host_depth_store_set_up && !bindless_resources_used_) {
    host_depth_copy_srv_index =
        uint32_t(current_temporary_descriptors_cpu_.size());
    current_temporary_descriptors_cpu_.push_back(provider.OffsetViewDescriptor(
        edram_buffer_descriptor_heap_start_,
        uint32_t(EdramBufferDescriptorIndex::kR32UintSRV)));
  } else {
    host_depth_copy_srv_index = UINT32_MAX;
  }
  for (uint32_t i = 0; i < render_target_count; ++i) {
    RenderTarget* dest_rt = render_targets[i];
    if (!dest_rt) {
      continue;
    }
    bool dest_is_depth = dest_rt->key().is_depth;
    for (const Transfer& transfer : render_target_transfers[i]) {
      assert_not_null(transfer.source);
      auto& source_d3d12_rt = *static_cast<D3D12RenderTarget*>(transfer.source);
      if (source_d3d12_rt.temporary_srv_descriptor_index() == UINT32_MAX) {
        source_d3d12_rt.SetTemporarySRVDescriptorIndex(
            uint32_t(current_temporary_descriptors_cpu_.size()));
        current_temporary_descriptors_cpu_.push_back(
            source_d3d12_rt.descriptor_srv().GetHandle());
      }
      if (source_d3d12_rt.key().is_depth &&
          source_d3d12_rt.temporary_srv_descriptor_index_stencil() ==
              UINT32_MAX) {
        source_d3d12_rt.SetTemporarySRVDescriptorIndexStencil(
            uint32_t(current_temporary_descriptors_cpu_.size()));
        current_temporary_descriptors_cpu_.push_back(
            source_d3d12_rt.descriptor_srv_stencil().GetHandle());
      }
      bool source_is_depth = source_d3d12_rt.key().is_depth;
      auto* host_depth_source_d3d12_rt =
          static_cast<D3D12RenderTarget*>(transfer.host_depth_source);
      // The host_depth_source_d3d12_rt == dest_rt case would use the EDRAM
      // buffer instead.
      if (host_depth_source_d3d12_rt && host_depth_source_d3d12_rt != dest_rt &&
          host_depth_source_d3d12_rt->temporary_srv_descriptor_index() ==
              UINT32_MAX) {
        host_depth_source_d3d12_rt->SetTemporarySRVDescriptorIndex(
            uint32_t(current_temporary_descriptors_cpu_.size()));
        current_temporary_descriptors_cpu_.push_back(
            host_depth_source_d3d12_rt->descriptor_srv().GetHandle());
      }
    }
  }
  uint32_t descriptor_count =
      uint32_t(current_temporary_descriptors_cpu_.size());
  current_temporary_descriptors_gpu_.resize(descriptor_count);
  if (!command_processor_.RequestOneUseSingleViewDescriptors(
          descriptor_count, current_temporary_descriptors_gpu_.data())) {
    return;
  }
  for (uint32_t i = 0; i < descriptor_count; ++i) {
    device->CopyDescriptorsSimple(1,
                                  current_temporary_descriptors_gpu_[i].first,
                                  current_temporary_descriptors_cpu_[i],
                                  D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
  }

  // Perform the transfers and clears.

  bool transfer_viewport_set = false;
  float pixels_to_ndc_unscaled =
      2.0f / float(D3D12_REQ_TEXTURE2D_U_OR_V_DIMENSION);
  float pixels_to_ndc_x = pixels_to_ndc_unscaled * draw_resolution_scale_x();
  float pixels_to_ndc_y = pixels_to_ndc_unscaled * draw_resolution_scale_y();

  TransferRootSignatureIndex last_transfer_root_signature_index =
      TransferRootSignatureIndex::kCount;
  uint32_t transfer_root_parameters_set = 0;
  uint32_t last_descriptor_index_color = UINT32_MAX;
  uint32_t last_descriptor_index_depth = UINT32_MAX;
  uint32_t last_descriptor_index_stencil = UINT32_MAX;
  uint32_t last_descriptor_index_host_depth_non_copy = UINT32_MAX;
  bool last_descriptor_host_depth_is_copy = false;
  TransferAddressConstant last_address_constant;
  TransferAddressConstant last_host_depth_address_constant;

  for (uint32_t i = 0; i < render_target_count; ++i) {
    RenderTarget* dest_rt = render_targets[i];
    if (!dest_rt) {
      continue;
    }

    const std::vector<Transfer>& current_transfers = render_target_transfers[i];
    if (current_transfers.empty() && !resolve_clear_needed) {
      continue;
    }

    auto& dest_d3d12_rt = *static_cast<D3D12RenderTarget*>(dest_rt);
    RenderTargetKey dest_rt_key = dest_d3d12_rt.key();

    // Late barrier in case there was cross-copying that prevented merging of
    // barriers.
    D3D12_RESOURCE_STATES dest_state = dest_rt_key.is_depth
                                           ? D3D12_RESOURCE_STATE_DEPTH_WRITE
                                           : D3D12_RESOURCE_STATE_RENDER_TARGET;
    command_processor_.PushTransitionBarrier(
        dest_d3d12_rt.resource(), dest_d3d12_rt.SetResourceState(dest_state),
        dest_state);

    if (!current_transfers.empty()) {
      are_current_command_list_render_targets_valid_ = false;
      if (dest_rt_key.is_depth) {
        command_list.D3DOMSetRenderTargets(
            0, nullptr, FALSE, &dest_d3d12_rt.descriptor_draw().GetHandle());
        if (!use_stencil_reference_output_) {
          command_processor_.SetStencilReference(UINT8_MAX);
        }
      } else {
        command_list.D3DOMSetRenderTargets(
            1,
            &(dest_d3d12_rt.descriptor_load_separate().IsValid()
                  ? dest_d3d12_rt.descriptor_load_separate().GetHandle()
                  : dest_d3d12_rt.descriptor_draw().GetHandle()),
            FALSE, nullptr);
      }

      uint32_t dest_pitch_tiles = dest_rt_key.GetPitchTiles();
      bool dest_is_64bpp = dest_rt_key.Is64bpp();

      // Gather shader keys and sort to reduce pipeline state and binding
      // switches. Also gather stencil rectangles to clear if needed.
      bool need_stencil_bit_draws =
          dest_rt_key.is_depth && !use_stencil_reference_output_;
      current_transfer_invocations_.clear();
      current_transfer_invocations_.reserve(
          current_transfers.size() << uint32_t(need_stencil_bit_draws));
      uint32_t rt_sort_index = 0;
      TransferShaderKey new_transfer_shader_key;
      new_transfer_shader_key.dest_msaa_samples = dest_rt_key.msaa_samples;
      new_transfer_shader_key.dest_resource_format =
          dest_rt_key.resource_format;
      uint32_t stencil_clear_rectangle_count = 0;
      for (uint32_t j = 0; j <= uint32_t(need_stencil_bit_draws); ++j) {
        // j == 0 - color or depth.
        // j == 1 - stencil bits.
        // Stencil bit writing always requires a different root signature,
        // handle these separately. Stencil never has a host depth source.
        // Clear previously set sort indices.
        for (const Transfer& transfer : current_transfers) {
          auto* host_depth_source_d3d12_rt =
              static_cast<D3D12RenderTarget*>(transfer.host_depth_source);
          if (host_depth_source_d3d12_rt) {
            host_depth_source_d3d12_rt->SetTemporarySortIndex(UINT32_MAX);
          }
          assert_not_null(transfer.source);
          auto& source_d3d12_rt =
              *static_cast<D3D12RenderTarget*>(transfer.source);
          source_d3d12_rt.SetTemporarySortIndex(UINT32_MAX);
        }
        for (const Transfer& transfer : current_transfers) {
          assert_not_null(transfer.source);
          auto& source_d3d12_rt =
              *static_cast<D3D12RenderTarget*>(transfer.source);
          D3D12RenderTarget* host_depth_source_d3d12_rt =
              j ? nullptr
                : static_cast<D3D12RenderTarget*>(transfer.host_depth_source);
          if (host_depth_source_d3d12_rt &&
              host_depth_source_d3d12_rt->temporary_sort_index() ==
                  UINT32_MAX) {
            host_depth_source_d3d12_rt->SetTemporarySortIndex(rt_sort_index++);
          }
          if (source_d3d12_rt.temporary_sort_index() == UINT32_MAX) {
            source_d3d12_rt.SetTemporarySortIndex(rt_sort_index++);
          }
          RenderTargetKey source_rt_key = source_d3d12_rt.key();
          new_transfer_shader_key.source_msaa_samples =
              source_rt_key.msaa_samples;
          new_transfer_shader_key.source_resource_format =
              source_rt_key.resource_format;
          bool host_depth_source_is_copy =
              host_depth_source_d3d12_rt == &dest_d3d12_rt;
          new_transfer_shader_key.host_depth_source_is_copy =
              host_depth_source_is_copy;
          // The host depth copy buffer has only raw samples.
          new_transfer_shader_key.host_depth_source_msaa_samples =
              (host_depth_source_d3d12_rt && !host_depth_source_is_copy)
                  ? host_depth_source_d3d12_rt->key().msaa_samples
                  : xenos::MsaaSamples::k1X;
          if (j) {
            new_transfer_shader_key.mode =
                source_rt_key.is_depth ? TransferMode::kDepthToStencilBit
                                       : TransferMode::kColorToStencilBit;
            stencil_clear_rectangle_count +=
                transfer.GetRectangles(dest_rt_key.base_tiles, dest_pitch_tiles,
                                       dest_rt_key.msaa_samples, dest_is_64bpp,
                                       nullptr, resolve_clear_rectangle);
          } else {
            if (dest_rt_key.is_depth) {
              if (host_depth_source_d3d12_rt) {
                new_transfer_shader_key.mode =
                    source_rt_key.is_depth
                        ? TransferMode::kDepthAndHostDepthToDepth
                        : TransferMode::kColorAndHostDepthToDepth;
              } else {
                new_transfer_shader_key.mode =
                    source_rt_key.is_depth ? TransferMode::kDepthToDepth
                                           : TransferMode::kColorToDepth;
              }
            } else {
              new_transfer_shader_key.mode = source_rt_key.is_depth
                                                 ? TransferMode::kDepthToColor
                                                 : TransferMode::kColorToColor;
            }
          }
          current_transfer_invocations_.emplace_back(transfer,
                                                     new_transfer_shader_key);
          if (j) {
            current_transfer_invocations_.back().transfer.host_depth_source =
                nullptr;
          }
        }
      }
      std::sort(current_transfer_invocations_.begin(),
                current_transfer_invocations_.end());

      // Clear the stencil to 0 where it will be loaded - will be setting the
      // bits that need to be 1 by discarding samples. Clearing everything here
      // to reduce context switches internally in the driver if clear causes
      // them.
      if (stencil_clear_rectangle_count) {
        command_processor_.SubmitBarriers();
        D3D12_RECT* stencil_clear_rect_write_ptr =
            command_list.ClearDepthStencilViewAllocatedRects(
                dest_d3d12_rt.descriptor_draw().GetHandle(),
                D3D12_CLEAR_FLAG_STENCIL, 0.0f, 0,
                stencil_clear_rectangle_count);
        assert_not_null(stencil_clear_rect_write_ptr);
        for (const Transfer& transfer : current_transfers) {
          Transfer::Rectangle transfer_stencil_clear_rectangles
              [Transfer::kMaxRectanglesWithCutout];
          uint32_t transfer_stencil_clear_rectangle_count =
              transfer.GetRectangles(dest_rt_key.base_tiles, dest_pitch_tiles,
                                     dest_rt_key.msaa_samples, dest_is_64bpp,
                                     transfer_stencil_clear_rectangles,
                                     resolve_clear_rectangle);
          for (uint32_t j = 0; j < transfer_stencil_clear_rectangle_count;
               ++j) {
            const Transfer::Rectangle& stencil_clear_rectangle =
                transfer_stencil_clear_rectangles[j];
            stencil_clear_rect_write_ptr->left = LONG(
                stencil_clear_rectangle.x_pixels * draw_resolution_scale_x());
            stencil_clear_rect_write_ptr->top = LONG(
                stencil_clear_rectangle.y_pixels * draw_resolution_scale_y());
            stencil_clear_rect_write_ptr->right =
                LONG((stencil_clear_rectangle.x_pixels +
                      stencil_clear_rectangle.width_pixels) *
                     draw_resolution_scale_x());
            stencil_clear_rect_write_ptr->bottom =
                LONG((stencil_clear_rectangle.y_pixels +
                      stencil_clear_rectangle.height_pixels) *
                     draw_resolution_scale_y());
            ++stencil_clear_rect_write_ptr;
          }
        }
      }

      // Perform the transfers for the render target.

      if (!transfer_viewport_set) {
        transfer_viewport_set = true;
        // Will be passing NDC directly, set the viewport to the maximum host
        // render target size for simplicity. Using a power-of-two scale for
        // exact pixel coordinates.
        D3D12_VIEWPORT transfer_viewport;
        transfer_viewport.TopLeftX = 0.0f;
        transfer_viewport.TopLeftY = 0.0f;
        transfer_viewport.Width = float(D3D12_REQ_TEXTURE2D_U_OR_V_DIMENSION);
        transfer_viewport.Height = float(D3D12_REQ_TEXTURE2D_U_OR_V_DIMENSION);
        transfer_viewport.MinDepth = 0.0f;
        transfer_viewport.MaxDepth = 1.0f;
        command_processor_.SetViewport(transfer_viewport);
        // TODO(Triang3l): Reduce scissor to the smallest transfer region for
        // more tiling friendliness.
        D3D12_RECT transfer_scissor;
        transfer_scissor.left = 0;
        transfer_scissor.top = 0;
        transfer_scissor.right = LONG(D3D12_REQ_TEXTURE2D_U_OR_V_DIMENSION);
        transfer_scissor.bottom = LONG(D3D12_REQ_TEXTURE2D_U_OR_V_DIMENSION);
        command_processor_.SetScissorRect(transfer_scissor);
      }

      for (auto it = current_transfer_invocations_.cbegin();
           it != current_transfer_invocations_.cend(); ++it) {
        const TransferInvocation& transfer_invocation_first = *it;
        // Will be merging transfers from the same source into one mesh.
        auto it_merged_first = it, it_merged_last = it;
        uint32_t transfer_rectangle_count =
            transfer_invocation_first.transfer.GetRectangles(
                dest_rt_key.base_tiles, dest_pitch_tiles,
                dest_rt_key.msaa_samples, dest_is_64bpp, nullptr,
                resolve_clear_rectangle);
        for (auto it_merge = std::next(it_merged_first);
             it_merge != current_transfer_invocations_.cend(); ++it_merge) {
          if (!transfer_invocation_first.CanBeMergedIntoOneDraw(*it_merge)) {
            break;
          }
          transfer_rectangle_count += it_merge->transfer.GetRectangles(
              dest_rt_key.base_tiles, dest_pitch_tiles,
              dest_rt_key.msaa_samples, dest_is_64bpp, nullptr,
              resolve_clear_rectangle);
          it_merged_last = it_merge;
        }
        assert_not_zero(transfer_rectangle_count);
        // Skip the merged transfers in the subsequent iterations.
        it = it_merged_last;

        assert_not_null(it->transfer.source);
        auto& source_d3d12_rt =
            *static_cast<D3D12RenderTarget*>(it->transfer.source);
        auto* host_depth_source_d3d12_rt =
            static_cast<D3D12RenderTarget*>(it->transfer.host_depth_source);
        TransferShaderKey transfer_shader_key = it->shader_key;
        const TransferModeInfo& transfer_mode_info =
            kTransferModes[size_t(transfer_shader_key.mode)];
        TransferRootSignatureIndex transfer_root_signature_index =
            use_stencil_reference_output_
                ? transfer_mode_info.root_signature_with_stencil_ref
                : transfer_mode_info.root_signature_no_stencil_ref;
        uint32_t transfer_root_parameters_used =
            kTransferUsedRootParameters[size_t(transfer_root_signature_index)];
        bool is_stencil_bit =
            (transfer_root_parameters_used &
             kTransferUsedRootParameterStencilMaskConstantBit) != 0;

        // Late barriers in case there was cross-copying that prevented merging
        // of barriers.
        command_processor_.PushTransitionBarrier(
            source_d3d12_rt.resource(),
            source_d3d12_rt.SetResourceState(
                D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE),
            D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE);
        if (host_depth_source_d3d12_rt) {
          if (transfer_shader_key.host_depth_source_is_copy) {
            // Reading copied host depth from the EDRAM buffer.
            TransitionEdramBuffer(D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE);
          } else {
            // Reading host depth from the texture.
            command_processor_.PushTransitionBarrier(
                host_depth_source_d3d12_rt->resource(),
                host_depth_source_d3d12_rt->SetResourceState(
                    D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE),
                D3D12_RESOURCE_STATE_PIXEL_SHADER_RESOURCE);
          }
        }

        uint32_t transfer_vertex_count = 6 * transfer_rectangle_count;
        D3D12_VERTEX_BUFFER_VIEW transfer_rectangle_buffer_view;
        transfer_rectangle_buffer_view.StrideInBytes = sizeof(float) * 2;
        transfer_rectangle_buffer_view.SizeInBytes =
            transfer_rectangle_buffer_view.StrideInBytes *
            transfer_vertex_count;
        float* transfer_rectangle_write_ptr =
            reinterpret_cast<float*>(transfer_vertex_buffer_pool_->Request(
                current_submission, transfer_rectangle_buffer_view.SizeInBytes,
                sizeof(float), nullptr, nullptr,
                &transfer_rectangle_buffer_view.BufferLocation));
        if (!transfer_rectangle_write_ptr) {
          continue;
        }
        for (auto it_merged = it_merged_first; it_merged <= it_merged_last;
             ++it_merged) {
          Transfer::Rectangle transfer_invocation_rectangles
              [Transfer::kMaxRectanglesWithCutout];
          uint32_t transfer_invocation_rectangle_count =
              it_merged->transfer.GetRectangles(
                  dest_rt_key.base_tiles, dest_pitch_tiles,
                  dest_rt_key.msaa_samples, dest_is_64bpp,
                  transfer_invocation_rectangles, resolve_clear_rectangle);
          assert_not_zero(transfer_invocation_rectangle_count);
          for (uint32_t j = 0; j < transfer_invocation_rectangle_count; ++j) {
            const Transfer::Rectangle& transfer_rectangle =
                transfer_invocation_rectangles[j];
            float transfer_rectangle_x0 =
                -1.0f + transfer_rectangle.x_pixels * pixels_to_ndc_x;
            float transfer_rectangle_y0 =
                1.0f - transfer_rectangle.y_pixels * pixels_to_ndc_y;
            float transfer_rectangle_x1 =
                transfer_rectangle_x0 +
                transfer_rectangle.width_pixels * pixels_to_ndc_x;
            float transfer_rectangle_y1 =
                transfer_rectangle_y0 -
                transfer_rectangle.height_pixels * pixels_to_ndc_y;
            // O-*
            // |/
            // *
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_x0;
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_y0;
            // *-O
            // |/
            // *
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_x1;
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_y0;
            // *-*
            // |/
            // O
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_x0;
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_y1;
            //   *
            //  /|
            // O-*
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_x0;
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_y1;
            //   O
            //  /|
            // *-*
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_x1;
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_y0;
            //   *
            //  /|
            // *-O
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_x1;
            *(transfer_rectangle_write_ptr++) = transfer_rectangle_y1;
          }
        }
        command_processor_.SetPrimitiveTopology(
            D3D_PRIMITIVE_TOPOLOGY_TRIANGLELIST);
        command_list.D3DIASetVertexBuffers(0, 1,
                                           &transfer_rectangle_buffer_view);

        ID3D12PipelineState* const* transfer_pipelines =
            GetOrCreateTransferPipelines(transfer_shader_key);
        if (!transfer_pipelines) {
          continue;
        }
        if (last_transfer_root_signature_index !=
            transfer_root_signature_index) {
          last_transfer_root_signature_index = transfer_root_signature_index;
          command_processor_.SetExternalGraphicsRootSignature(
              transfer_root_signatures_[size_t(transfer_root_signature_index)]);
          transfer_root_parameters_set = 0;
        }

        // Invalidate outdated bindings.
        if (transfer_root_parameters_used &
            kTransferUsedRootParameterColorSRVBit) {
          uint32_t descriptor_index_color =
              source_d3d12_rt.temporary_srv_descriptor_index();
          assert_true(descriptor_index_color != UINT32_MAX);
          if (last_descriptor_index_color != descriptor_index_color) {
            last_descriptor_index_color = descriptor_index_color;
            transfer_root_parameters_set &=
                ~kTransferUsedRootParameterColorSRVBit;
          }
        }
        if (transfer_root_parameters_used &
            kTransferUsedRootParameterDepthSRVBit) {
          uint32_t descriptor_index_depth =
              source_d3d12_rt.temporary_srv_descriptor_index();
          assert_true(descriptor_index_depth != UINT32_MAX);
          if (last_descriptor_index_depth != descriptor_index_depth) {
            last_descriptor_index_depth = descriptor_index_depth;
            transfer_root_parameters_set &=
                ~kTransferUsedRootParameterDepthSRVBit;
          }
        }
        if (transfer_root_parameters_used &
            kTransferUsedRootParameterStencilSRVBit) {
          uint32_t descriptor_index_stencil =
              source_d3d12_rt.temporary_srv_descriptor_index_stencil();
          assert_true(descriptor_index_stencil != UINT32_MAX);
          if (last_descriptor_index_stencil != descriptor_index_stencil) {
            last_descriptor_index_stencil = descriptor_index_stencil;
            transfer_root_parameters_set &=
                ~kTransferUsedRootParameterStencilSRVBit;
          }
        }
        if (transfer_root_parameters_used &
            kTransferUsedRootParameterHostDepthSRVBit) {
          if (transfer_shader_key.host_depth_source_is_copy) {
            if (!last_descriptor_host_depth_is_copy) {
              last_descriptor_host_depth_is_copy = true;
              transfer_root_parameters_set &=
                  ~kTransferUsedRootParameterHostDepthSRVBit;
            }
          } else {
            assert_not_null(host_depth_source_d3d12_rt);
            uint32_t descriptor_index_host_depth =
                host_depth_source_d3d12_rt->temporary_srv_descriptor_index();
            assert_true(descriptor_index_host_depth != UINT32_MAX);
            if (last_descriptor_host_depth_is_copy ||
                last_descriptor_index_host_depth_non_copy !=
                    descriptor_index_host_depth) {
              transfer_root_parameters_set &=
                  ~kTransferUsedRootParameterHostDepthSRVBit;
            }
            last_descriptor_host_depth_is_copy = false;
            last_descriptor_index_host_depth_non_copy =
                descriptor_index_host_depth;
          }
        }
        if (transfer_root_parameters_used &
            kTransferUsedRootParameterAddressConstantBit) {
          RenderTargetKey source_rt_key = source_d3d12_rt.key();
          TransferAddressConstant address_constant;
          address_constant.dest_pitch = dest_pitch_tiles;
          address_constant.source_pitch = source_rt_key.GetPitchTiles();
          address_constant.source_to_dest = int32_t(dest_rt_key.base_tiles) -
                                            int32_t(source_rt_key.base_tiles);
          if (last_address_constant != address_constant) {
            last_address_constant = address_constant;
            transfer_root_parameters_set &=
                ~kTransferUsedRootParameterAddressConstantBit;
          }
        }
        if (transfer_root_parameters_used &
            kTransferUsedRootParameterHostDepthAddressConstantBit) {
          assert_not_null(host_depth_source_d3d12_rt);
          RenderTargetKey host_depth_source_rt_key =
              host_depth_source_d3d12_rt->key();
          TransferAddressConstant host_depth_address_constant;
          host_depth_address_constant.dest_pitch = dest_pitch_tiles;
          host_depth_address_constant.source_pitch =
              host_depth_source_rt_key.GetPitchTiles();
          host_depth_address_constant.source_to_dest =
              int32_t(dest_rt_key.base_tiles) -
              int32_t(host_depth_source_rt_key.base_tiles);
          if (last_host_depth_address_constant != host_depth_address_constant) {
            last_host_depth_address_constant = host_depth_address_constant;
            transfer_root_parameters_set &=
                ~kTransferUsedRootParameterHostDepthAddressConstantBit;
          }
        }

        // Apply the new bindings.
        uint32_t transfer_root_parameters_unset =
            transfer_root_parameters_used & ~transfer_root_parameters_set;
        if (transfer_root_parameters_unset &
            kTransferUsedRootParameterHostDepthAddressConstantBit) {
          command_list.D3DSetGraphicsRoot32BitConstants(
              xe::bit_count(
                  transfer_root_parameters_used &
                  (kTransferUsedRootParameterHostDepthAddressConstantBit - 1)),
              sizeof(last_host_depth_address_constant) / sizeof(uint32_t),
              &last_host_depth_address_constant, 0);
          transfer_root_parameters_set |=
              kTransferUsedRootParameterHostDepthAddressConstantBit;
        }
        if (transfer_root_parameters_unset &
            kTransferUsedRootParameterHostDepthSRVBit) {
          D3D12_GPU_DESCRIPTOR_HANDLE descriptor_gpu_handle;
          if (last_descriptor_host_depth_is_copy) {
            if (bindless_resources_used_) {
              descriptor_gpu_handle =
                  command_processor_
                      .GetSystemBindlessViewHandlePair(
                          D3D12CommandProcessor::SystemBindlessView ::
                              kEdramR32UintSRV)
                      .second;
            } else {
              assert_true(host_depth_copy_srv_index != UINT32_MAX);
              descriptor_gpu_handle =
                  current_temporary_descriptors_gpu_[host_depth_copy_srv_index]
                      .second;
            }
          } else {
            assert_true(last_descriptor_index_host_depth_non_copy !=
                        UINT32_MAX);
            descriptor_gpu_handle =
                current_temporary_descriptors_gpu_
                    [last_descriptor_index_host_depth_non_copy]
                        .second;
          }
          command_list.D3DSetGraphicsRootDescriptorTable(
              xe::bit_count(transfer_root_parameters_used &
                            (kTransferUsedRootParameterHostDepthSRVBit - 1)),
              descriptor_gpu_handle);
          transfer_root_parameters_set |=
              kTransferUsedRootParameterHostDepthSRVBit;
        }
        if (transfer_root_parameters_unset &
            kTransferUsedRootParameterAddressConstantBit) {
          command_list.D3DSetGraphicsRoot32BitConstants(
              xe::bit_count(transfer_root_parameters_used &
                            (kTransferUsedRootParameterAddressConstantBit - 1)),
              sizeof(last_address_constant) / sizeof(uint32_t),
              &last_address_constant, 0);
          transfer_root_parameters_set |=
              kTransferUsedRootParameterAddressConstantBit;
        }
        if (transfer_root_parameters_unset &
            kTransferUsedRootParameterStencilSRVBit) {
          assert_true(last_descriptor_index_stencil != UINT32_MAX);
          command_list.D3DSetGraphicsRootDescriptorTable(
              xe::bit_count(transfer_root_parameters_used &
                            (kTransferUsedRootParameterStencilSRVBit - 1)),
              current_temporary_descriptors_gpu_[last_descriptor_index_stencil]
                  .second);
          transfer_root_parameters_set |=
              kTransferUsedRootParameterStencilSRVBit;
        }
        if (transfer_root_parameters_unset &
            kTransferUsedRootParameterDepthSRVBit) {
          assert_true(last_descriptor_index_depth != UINT32_MAX);
          command_list.D3DSetGraphicsRootDescriptorTable(
              xe::bit_count(transfer_root_parameters_used &
                            (kTransferUsedRootParameterDepthSRVBit - 1)),
              current_temporary_descriptors_gpu_[last_descriptor_index_depth]
                  .second);
          transfer_root_parameters_set |= kTransferUsedRootParameterDepthSRVBit;
        }
        if (transfer_root_parameters_unset &
            kTransferUsedRootParameterColorSRVBit) {
          assert_true(last_descriptor_index_color != UINT32_MAX);
          command_list.D3DSetGraphicsRootDescriptorTable(
              xe::bit_count(transfer_root_parameters_used &
                            (kTransferUsedRootParameterColorSRVBit - 1)),
              current_temporary_descriptors_gpu_[last_descriptor_index_color]
                  .second);
          transfer_root_parameters_set |= kTransferUsedRootParameterColorSRVBit;
        }

        // Draw the transfer rectangles.
        command_processor_.SubmitBarriers();
        for (uint32_t j = 0; j <= uint32_t(is_stencil_bit) * 7; ++j) {
          if (is_stencil_bit) {
            uint32_t transfer_stencil_bit = uint32_t(1) << j;
            command_list.D3DSetGraphicsRoot32BitConstants(
                xe::bit_count(
                    transfer_root_parameters_used &
                    (kTransferUsedRootParameterStencilMaskConstantBit - 1)),
                sizeof(transfer_stencil_bit) / sizeof(uint32_t),
                &transfer_stencil_bit, 0);
          }
          command_processor_.SetExternalPipeline(transfer_pipelines[j]);
          command_list.D3DDrawInstanced(transfer_vertex_count, 1, 0, 0);
        }
      }
    }

    // Perform the clear.
    if (resolve_clear_needed) {
      uint64_t clear_value = render_target_resolve_clear_values[i];
      if (dest_rt_key.is_depth) {
        uint32_t depth_guest_clear_value =
            (uint32_t(clear_value) >> 8) & 0xFFFFFF;
        float depth_host_clear_value = 0.0f;
        switch (dest_rt_key.GetDepthFormat()) {
          case xenos::DepthRenderTargetFormat::kD24S8:
            depth_host_clear_value =
                xenos::UNorm24To32(depth_guest_clear_value);
            break;
          case xenos::DepthRenderTargetFormat::kD24FS8:
            // Taking [0, 2) -> [0, 1) remapping into account.
            depth_host_clear_value =
                xenos::Float20e4To32(depth_guest_clear_value) * 0.5f;
            break;
        }
        command_processor_.PushTransitionBarrier(
            dest_d3d12_rt.resource(),
            dest_d3d12_rt.SetResourceState(D3D12_RESOURCE_STATE_DEPTH_WRITE),
            D3D12_RESOURCE_STATE_DEPTH_WRITE);
        command_processor_.SubmitBarriers();
        command_list.D3DClearDepthStencilView(
            dest_d3d12_rt.descriptor_draw().GetHandle(),
            D3D12_CLEAR_FLAG_DEPTH | D3D12_CLEAR_FLAG_STENCIL,
            depth_host_clear_value, UINT(clear_value) & 0xFF, 1, &clear_rect);
      } else {
        float color_clear_value[4] = {};
        bool clear_via_drawing = false;
        switch (dest_rt_key.GetColorFormat()) {
          case xenos::ColorRenderTargetFormat::k_8_8_8_8:
          case xenos::ColorRenderTargetFormat::k_8_8_8_8_GAMMA: {
            for (uint32_t j = 0; j < 4; ++j) {
              color_clear_value[j] =
                  ((clear_value >> (j * 8)) & 0xFF) * (1.0f / 0xFF);
            }
          } break;
          case xenos::ColorRenderTargetFormat::k_2_10_10_10:
          case xenos::ColorRenderTargetFormat::k_2_10_10_10_AS_10_10_10_10: {
            for (uint32_t j = 0; j < 3; ++j) {
              color_clear_value[j] =
                  ((clear_value >> (j * 10)) & 0x3FF) * (1.0f / 0x3FF);
            }
            color_clear_value[3] = ((clear_value >> 30) & 0x3) * (1.0f / 0x3);
          } break;
          case xenos::ColorRenderTargetFormat::k_2_10_10_10_FLOAT:
          case xenos::ColorRenderTargetFormat::
              k_2_10_10_10_FLOAT_AS_16_16_16_16: {
            for (uint32_t j = 0; j < 3; ++j) {
              color_clear_value[j] =
                  xenos::Float7e3To32((clear_value >> (j * 10)) & 0x3FF);
            }
            color_clear_value[3] = ((clear_value >> 30) & 0x3) * (1.0f / 0x3);
          } break;
          case xenos::ColorRenderTargetFormat::k_16_16:
          case xenos::ColorRenderTargetFormat::k_16_16_FLOAT: {
            // Using uint for loading both. Disregarding the current -32...32
            // vs. -1...1 settings for consistency with color clear via depth
            // aliasing.
            for (uint32_t j = 0; j < 2; ++j) {
              color_clear_value[j] = float((clear_value >> (j * 16)) & 0xFFFF);
            }
          } break;
          case xenos::ColorRenderTargetFormat::k_16_16_16_16:
          case xenos::ColorRenderTargetFormat::k_16_16_16_16_FLOAT: {
            // Using uint for loading both. Disregarding the current -32...32
            // vs. -1...1 settings for consistency with color clear via depth
            // aliasing.
            for (uint32_t j = 0; j < 4; ++j) {
              color_clear_value[j] = float((clear_value >> (j * 16)) & 0xFFFF);
            }
          } break;
          case xenos::ColorRenderTargetFormat::k_32_FLOAT: {
            // Using uint for proper denormal and NaN handling.
            color_clear_value[0] = float(uint32_t(clear_value));
            // Numbers > 2^24 can't be represented with a step of 1 as floats,
            // need to clear by drawing a uint rectangle.
            if (uint64_t(color_clear_value[0]) != uint32_t(clear_value)) {
              clear_via_drawing = true;
            }
          } break;
          case xenos::ColorRenderTargetFormat::k_32_32_FLOAT: {
            // Using uint for proper denormal and NaN handling.
            color_clear_value[0] = float(uint32_t(clear_value));
            color_clear_value[1] = float(uint32_t(clear_value >> 32));
            // Numbers > 2^24 can't be represented with a step of 1 as floats,
            // need to clear by drawing a uint rectangle.
            if (uint64_t(color_clear_value[0]) != uint32_t(clear_value) ||
                uint64_t(color_clear_value[1]) != uint32_t(clear_value >> 32)) {
              clear_via_drawing = true;
            }
          } break;
        }
        command_processor_.PushTransitionBarrier(
            dest_d3d12_rt.resource(),
            dest_d3d12_rt.SetResourceState(D3D12_RESOURCE_STATE_RENDER_TARGET),
            D3D12_RESOURCE_STATE_RENDER_TARGET);
        if (clear_via_drawing) {
          command_list.D3DOMSetRenderTargets(
              1,
              &(dest_d3d12_rt.descriptor_load_separate().IsValid()
                    ? dest_d3d12_rt.descriptor_load_separate().GetHandle()
                    : dest_d3d12_rt.descriptor_draw().GetHandle()),
              FALSE, nullptr);
          are_current_command_list_render_targets_valid_ = true;
          D3D12_VIEWPORT clear_viewport;
          clear_viewport.TopLeftX = float(clear_rect.left);
          clear_viewport.TopLeftY = float(clear_rect.top);
          clear_viewport.Width = float(clear_rect.right - clear_rect.left);
          clear_viewport.Height = float(clear_rect.bottom - clear_rect.top);
          clear_viewport.MinDepth = 0.0f;
          clear_viewport.MaxDepth = 1.0f;
          command_processor_.SetViewport(clear_viewport);
          command_processor_.SetScissorRect(clear_rect);
          command_processor_.SetExternalGraphicsRootSignature(
              uint32_rtv_clear_root_signature_);
          uint32_t clear_via_drawing_value[2] = {uint32_t(clear_value),
                                                 uint32_t(clear_value >> 32)};
          command_list.D3DSetGraphicsRoot32BitConstants(
              0, 2, clear_via_drawing_value, 0);
          command_processor_.SetExternalPipeline(
              uint32_rtv_clear_pipelines_[size_t(
                  dest_rt_key.GetColorFormat() ==
                  xenos::ColorRenderTargetFormat::k_32_32_FLOAT)]
                                         [size_t(dest_rt_key.msaa_samples)]);
          command_processor_.SetPrimitiveTopology(
              D3D_PRIMITIVE_TOPOLOGY_TRIANGLELIST);
          command_list.D3DDrawInstanced(3, 1, 0, 0);
        } else {
          command_processor_.SubmitBarriers();
          command_list.D3DClearRenderTargetView(
              dest_d3d12_rt.descriptor_load_separate().IsValid()
                  ? dest_d3d12_rt.descriptor_load_separate().GetHandle()
                  : dest_d3d12_rt.descriptor_draw().GetHandle(),
              color_clear_value, 1, &clear_rect);
        }
      }
    }
  }
}

void D3D12RenderTargetCache::SetCommandListRenderTargets(
    RenderTarget* const* depth_and_color_render_targets) {
  assert_true(GetPath() == Path::kHostRenderTargets);

  // Ensure the render targets are in the needed resource state.
  if (depth_and_color_render_targets[0]) {
    auto& d3d12_rt =
        *static_cast<D3D12RenderTarget*>(depth_and_color_render_targets[0]);
    command_processor_.PushTransitionBarrier(
        d3d12_rt.resource(),
        d3d12_rt.SetResourceState(D3D12_RESOURCE_STATE_DEPTH_WRITE),
        D3D12_RESOURCE_STATE_DEPTH_WRITE);
  }
  for (uint32_t i = 0; i < xenos::kMaxColorRenderTargets; ++i) {
    RenderTarget* render_target = depth_and_color_render_targets[1 + i];
    if (!render_target) {
      continue;
    }
    auto& d3d12_rt = *static_cast<D3D12RenderTarget*>(render_target);
    command_processor_.PushTransitionBarrier(
        d3d12_rt.resource(),
        d3d12_rt.SetResourceState(D3D12_RESOURCE_STATE_RENDER_TARGET),
        D3D12_RESOURCE_STATE_RENDER_TARGET);
  }

  // Bind the render targets.
  if (are_current_command_list_render_targets_valid_ &&
      std::memcmp(current_command_list_render_targets_,
                  depth_and_color_render_targets,
                  sizeof(current_command_list_render_targets_))) {
    are_current_command_list_render_targets_valid_ = false;
  }
  uint32_t render_targets_are_srgb;
  if (gamma_render_target_as_srgb_) {
    render_targets_are_srgb = last_update_accumulated_color_targets_are_gamma();
    if (are_current_command_list_render_targets_srgb_ !=
        render_targets_are_srgb) {
      are_current_command_list_render_targets_srgb_ = render_targets_are_srgb;
      are_current_command_list_render_targets_valid_ = false;
    }
  } else {
    render_targets_are_srgb = 0;
  }
  if (!are_current_command_list_render_targets_valid_) {
    std::memcpy(current_command_list_render_targets_,
                depth_and_color_render_targets,
                sizeof(current_command_list_render_targets_));
    D3D12_CPU_DESCRIPTOR_HANDLE dsv_handle;
    if (depth_and_color_render_targets[0]) {
      dsv_handle = static_cast<const D3D12RenderTarget*>(
                       depth_and_color_render_targets[0])
                       ->descriptor_draw()
                       .GetHandle();
    }
    D3D12_CPU_DESCRIPTOR_HANDLE rtv_handles[xenos::kMaxColorRenderTargets];
    uint32_t rtv_count = 0;
    for (uint32_t i = 0; i < xenos::kMaxColorRenderTargets; ++i) {
      const RenderTarget* render_target = depth_and_color_render_targets[1 + i];
      if (!render_target) {
        continue;
      }
      // Fill the gaps with a null descriptor.
      while (rtv_count < i) {
        rtv_handles[rtv_count++] =
            render_target->key().msaa_samples != xenos::MsaaSamples::k1X
                ? null_rtv_descriptor_ms_.GetHandle()
                : null_rtv_descriptor_ss_.GetHandle();
      }
      auto& d3d12_rt = *static_cast<const D3D12RenderTarget*>(render_target);
      rtv_handles[rtv_count++] =
          (render_targets_are_srgb & (uint32_t(1) << i))
              ? d3d12_rt.descriptor_draw_srgb().GetHandle()
              : d3d12_rt.descriptor_draw().GetHandle();
    }
    command_processor_.GetDeferredCommandList().D3DOMSetRenderTargets(
        rtv_count, rtv_handles, FALSE,
        depth_and_color_render_targets[0] ? &dsv_handle : nullptr);
    are_current_command_list_render_targets_valid_ = true;
  }
}

ID3D12PipelineState* D3D12RenderTargetCache::GetOrCreateDumpPipeline(
    DumpPipelineKey key) {
  auto pipeline_it = dump_pipelines_.find(key);
  if (pipeline_it != dump_pipelines_.end()) {
    return pipeline_it->second;
  }

  // Because of built_shader_.resize(), pointers can't be kept persistently
  // here! Resizing also zeroes the memory.

  built_shader_.clear();

  // RDEF, ISGN, OSGN, SHEX, STAT.
  constexpr uint32_t kBlobCount = 5;

  // Allocate space for the container header and the blob offsets.
  built_shader_.resize(sizeof(dxbc::ContainerHeader) / sizeof(uint32_t) +
                       kBlobCount);
  uint32_t blob_offset_position_dwords =
      sizeof(dxbc::ContainerHeader) / sizeof(uint32_t);
  uint32_t blob_position_dwords = uint32_t(built_shader_.size());
  constexpr uint32_t kBlobHeaderSizeDwords =
      sizeof(dxbc::BlobHeader) / sizeof(uint32_t);

  uint32_t name_ptr;

  // ***************************************************************************
  // Resource definition
  // ***************************************************************************

  built_shader_[blob_offset_position_dwords] =
      uint32_t(blob_position_dwords * sizeof(uint32_t));
  uint32_t rdef_position_dwords = blob_position_dwords + kBlobHeaderSizeDwords;
  // Not needed, as the next operation done is resize, to allocate the space for
  // both the blob header and the resource definition header.
  // built_shader_.resize(rdef_position_dwords);

  // Allocate space for the RDEF header.
  built_shader_.resize(rdef_position_dwords +
                       sizeof(dxbc::RdefHeader) / sizeof(uint32_t));
  // Generator name.
  dxbc::AppendAlignedString(built_shader_, "Xenia");

  // Constant types - uint (aka "dword" when it's scalar) only.
  // Names.
  name_ptr = uint32_t((built_shader_.size() - rdef_position_dwords) *
                      sizeof(uint32_t));
  uint32_t rdef_dword_name_ptr = name_ptr;
  name_ptr += dxbc::AppendAlignedString(built_shader_, "dword");
  // Types.
  uint32_t rdef_type_uint_position_dwords = uint32_t(built_shader_.size());
  uint32_t rdef_type_uint_ptr =
      uint32_t((rdef_type_uint_position_dwords - rdef_position_dwords) *
               sizeof(uint32_t));
  built_shader_.resize(rdef_type_uint_position_dwords +
                       sizeof(dxbc::RdefType) / sizeof(uint32_t));
  {
    auto& rdef_type_uint = *reinterpret_cast<dxbc::RdefType*>(
        built_shader_.data() + rdef_type_uint_position_dwords);
    rdef_type_uint.variable_class = dxbc::RdefVariableClass::kScalar;
    rdef_type_uint.variable_type = dxbc::RdefVariableType::kUInt;
    rdef_type_uint.row_count = 1;
    rdef_type_uint.column_count = 1;
    rdef_type_uint.name_ptr = rdef_dword_name_ptr;
  }

  // Constants:
  // - uint xe_edram_dump_offsets
  // - uint xe_edram_dump_pitches
  enum Constant : uint32_t {
    kConstantOffsets,
    kConstantPitches,
    kConstantCount,
  };
  // Names.
  name_ptr = uint32_t((built_shader_.size() - rdef_position_dwords) *
                      sizeof(uint32_t));
  uint32_t rdef_xe_edram_dump_offsets_name_ptr = name_ptr;
  name_ptr += dxbc::AppendAlignedString(built_shader_, "xe_edram_dump_offsets");
  uint32_t rdef_xe_edram_dump_pitches_name_ptr = name_ptr;
  name_ptr += dxbc::AppendAlignedString(built_shader_, "xe_edram_dump_pitches");
  // Constants.
  uint32_t rdef_constants_position_dwords = uint32_t(built_shader_.size());
  uint32_t rdef_constants_ptr =
      uint32_t((rdef_constants_position_dwords - rdef_position_dwords) *
               sizeof(uint32_t));
  built_shader_.resize(rdef_constants_position_dwords +
                       sizeof(dxbc::RdefVariable) / sizeof(uint32_t) *
                           kConstantCount);
  {
    auto rdef_constants = reinterpret_cast<dxbc::RdefVariable*>(
        built_shader_.data() + rdef_constants_position_dwords);
    // uint xe_edram_dump_offsets
    dxbc::RdefVariable& rdef_constant_offsets =
        rdef_constants[kConstantOffsets];
    rdef_constant_offsets.name_ptr = rdef_xe_edram_dump_offsets_name_ptr;
    rdef_constant_offsets.size_bytes = sizeof(uint32_t);
    rdef_constant_offsets.flags = dxbc::kRdefVariableFlagUsed;
    rdef_constant_offsets.type_ptr = rdef_type_uint_ptr;
    rdef_constant_offsets.start_texture = UINT32_MAX;
    rdef_constant_offsets.start_sampler = UINT32_MAX;
    // uint xe_edram_dump_pitches
    dxbc::RdefVariable& rdef_constant_pitches =
        rdef_constants[kConstantPitches];
    rdef_constant_pitches.name_ptr = rdef_xe_edram_dump_pitches_name_ptr;
    rdef_constant_pitches.size_bytes = sizeof(uint32_t);
    rdef_constant_pitches.flags = dxbc::kRdefVariableFlagUsed;
    rdef_constant_pitches.type_ptr = rdef_type_uint_ptr;
    rdef_constant_pitches.start_texture = UINT32_MAX;
    rdef_constant_pitches.start_sampler = UINT32_MAX;
  }

  // Constant buffers:
  // - xe_edram_dump_offsets : b0 { uint xe_edram_dump_offsets; }
  // - xe_edram_dump_pitches : b1 { uint xe_edram_dump_pitches; }
  // Reusing the constant names for constant buffers.
  uint32_t rdef_cbuffer_position_dwords = uint32_t(built_shader_.size());
  built_shader_.resize(rdef_cbuffer_position_dwords +
                       sizeof(dxbc::RdefCbuffer) / sizeof(uint32_t) *
                           kDumpCbufferCount);
  {
    auto rdef_cbuffers = reinterpret_cast<dxbc::RdefCbuffer*>(
        built_shader_.data() + rdef_cbuffer_position_dwords);
    // xe_edram_dump_offsets
    dxbc::RdefCbuffer& rdef_cbuffer_offsets =
        rdef_cbuffers[kDumpCbufferOffsets];
    rdef_cbuffer_offsets.name_ptr = rdef_xe_edram_dump_offsets_name_ptr;
    rdef_cbuffer_offsets.variable_count = 1;
    rdef_cbuffer_offsets.variables_ptr = uint32_t(
        rdef_constants_ptr + sizeof(dxbc::RdefVariable) * kConstantOffsets);
    rdef_cbuffer_offsets.size_vector_aligned_bytes = sizeof(uint32_t) * 4;
    // xe_edram_dump_pitches
    dxbc::RdefCbuffer& rdef_cbuffer_pitches =
        rdef_cbuffers[kDumpCbufferPitches];
    rdef_cbuffer_pitches.name_ptr = rdef_xe_edram_dump_pitches_name_ptr;
    rdef_cbuffer_pitches.variable_count = 1;
    rdef_cbuffer_pitches.variables_ptr = uint32_t(
        rdef_constants_ptr + sizeof(dxbc::RdefVariable) * kConstantPitches);
    rdef_cbuffer_pitches.size_vector_aligned_bytes = sizeof(uint32_t) * 4;
  }

  // Bindings.
  // - Texture2D/Texture2DMS<float4/uint4> xe_edram_dump_source : t0
  // - Optionally, Texture2D/Texture2DMS<uint2> xe_edram_dump_stencil : t1
  // - RWBuffer<uint/uint2> xe_edram : u0
  // - Constant buffers
  uint32_t rdef_binding_count = 1 + key.is_depth + 1 + kDumpCbufferCount;
  // Names.
  name_ptr = uint32_t((built_shader_.size() - rdef_position_dwords) *
                      sizeof(uint32_t));
  uint32_t rdef_xe_edram_dump_source_name_ptr = name_ptr;
  name_ptr += dxbc::AppendAlignedString(built_shader_, "xe_edram_dump_source");
  uint32_t rdef_xe_edram_dump_stencil_name_ptr = name_ptr;
  if (key.is_depth) {
    name_ptr +=
        dxbc::AppendAlignedString(built_shader_, "xe_edram_dump_stencil");
  }
  uint32_t rdef_xe_edram_name_ptr = name_ptr;
  name_ptr += dxbc::AppendAlignedString(built_shader_, "xe_edram");
  // Bindings.
  uint32_t rdef_binding_position_dwords = uint32_t(built_shader_.size());
  built_shader_.resize(rdef_binding_position_dwords +
                       sizeof(dxbc::RdefInputBind) / sizeof(uint32_t) *
                           rdef_binding_count);
  bool source_is_uint;
  if (key.is_depth) {
    source_is_uint = false;
  } else {
    GetColorOwnershipTransferDXGIFormat(key.GetColorFormat(), &source_is_uint);
  }
  dxbc::ResourceReturnType source_return_type =
      source_is_uint ? dxbc::ResourceReturnType::kUInt
                     : dxbc::ResourceReturnType::kFloat;
  uint32_t source_component_count =
      key.is_depth ? 1
                   : xenos::GetColorRenderTargetFormatComponentCount(
                         key.GetColorFormat());
  bool format_is_64bpp = !key.is_depth && xenos::IsColorRenderTargetFormat64bpp(
                                              key.GetColorFormat());
  {
    auto rdef_bindings = reinterpret_cast<dxbc::RdefInputBind*>(
        built_shader_.data() + rdef_binding_position_dwords);
    uint32_t rdef_binding_index = 0;
    // xe_edram_dump_source
    dxbc::RdefInputBind& rdef_binding_source =
        rdef_bindings[rdef_binding_index++];
    rdef_binding_source.name_ptr = rdef_xe_edram_dump_source_name_ptr;
    rdef_binding_source.type = dxbc::RdefInputType::kTexture;
    rdef_binding_source.return_type = source_return_type;
    if (key.msaa_samples != xenos::MsaaSamples::k1X) {
      rdef_binding_source.dimension = dxbc::RdefDimension::kSRVTexture2DMS;
      // Sample count is dynamic on Shader Model 5.
    } else {
      rdef_binding_source.dimension = dxbc::RdefDimension::kSRVTexture2D;
      rdef_binding_source.sample_count = UINT32_MAX;
    }
    rdef_binding_source.bind_count = 1;
    rdef_binding_source.flags = (source_component_count - 1)
                                << dxbc::kRdefInputFlagsComponentsShift;
    // xe_edram_dump_stencil
    if (key.is_depth) {
      dxbc::RdefInputBind& rdef_binding_stencil =
          rdef_bindings[rdef_binding_index++];
      rdef_binding_stencil.name_ptr = rdef_xe_edram_dump_stencil_name_ptr;
      rdef_binding_stencil.type = dxbc::RdefInputType::kTexture;
      rdef_binding_stencil.return_type = dxbc::ResourceReturnType::kUInt;
      rdef_binding_stencil.dimension = rdef_binding_source.dimension;
      rdef_binding_stencil.sample_count = rdef_binding_source.sample_count;
      rdef_binding_stencil.bind_point = 1;
      rdef_binding_stencil.bind_count = 1;
      rdef_binding_stencil.flags = dxbc::kRdefInputFlags2Component;
      rdef_binding_stencil.id = 1;
    }
    // xe_edram
    dxbc::RdefInputBind& rdef_binding_edram =
        rdef_bindings[rdef_binding_index++];
    rdef_binding_edram.name_ptr = rdef_xe_edram_name_ptr;
    rdef_binding_edram.type = dxbc::RdefInputType::kUAVRWTyped;
    rdef_binding_edram.return_type = dxbc::ResourceReturnType::kUInt;
    rdef_binding_edram.dimension = dxbc::RdefDimension::kUAVBuffer;
    rdef_binding_edram.sample_count = UINT32_MAX;
    rdef_binding_edram.bind_count = 1;
    rdef_binding_edram.flags =
        format_is_64bpp ? dxbc::kRdefInputFlags2Component : 0;
    // xe_edram_dump_offsets
    dxbc::RdefInputBind& rdef_binding_offsets =
        rdef_bindings[rdef_binding_index++];
    rdef_binding_offsets.name_ptr = rdef_xe_edram_dump_offsets_name_ptr;
    rdef_binding_offsets.type = dxbc::RdefInputType::kCbuffer;
    rdef_binding_offsets.bind_point = kDumpCbufferOffsets;
    rdef_binding_offsets.bind_count = 1;
    rdef_binding_offsets.flags = dxbc::kRdefInputFlagUserPacked;
    rdef_binding_offsets.id = kDumpCbufferOffsets;
    // xe_edram_dump_pitches
    dxbc::RdefInputBind& rdef_binding_pitches =
        rdef_bindings[rdef_binding_index++];
    rdef_binding_pitches.name_ptr = rdef_xe_edram_dump_pitches_name_ptr;
    rdef_binding_pitches.type = dxbc::RdefInputType::kCbuffer;
    rdef_binding_pitches.bind_point = kDumpCbufferPitches;
    rdef_binding_pitches.bind_count = 1;
    rdef_binding_pitches.flags = dxbc::kRdefInputFlagUserPacked;
    rdef_binding_pitches.id = kDumpCbufferPitches;
  }

  // Header.
  {
    auto& rdef_header = *reinterpret_cast<dxbc::RdefHeader*>(
        built_shader_.data() + rdef_position_dwords);
    rdef_header.cbuffer_count = kDumpCbufferCount;
    rdef_header.cbuffers_ptr =
        uint32_t((rdef_cbuffer_position_dwords - rdef_position_dwords) *
                 sizeof(uint32_t));
    rdef_header.input_bind_count = rdef_binding_count;
    rdef_header.input_binds_ptr =
        uint32_t((rdef_binding_position_dwords - rdef_position_dwords) *
                 sizeof(uint32_t));
    rdef_header.shader_model = dxbc::RdefShaderModel::kComputeShader5_1;
    rdef_header.compile_flags =
        dxbc::kCompileFlagNoPreshader | dxbc::kCompileFlagPreferFlowControl |
        dxbc::kCompileFlagIeeeStrictness | dxbc::kCompileFlagAllResourcesBound;
    // Generator name is right after the header.
    rdef_header.generator_name_ptr = sizeof(dxbc::RdefHeader);
    rdef_header.fourcc = dxbc::RdefHeader::FourCC::k5_1;
    rdef_header.InitializeSizes();
  }

  {
    auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
        built_shader_.data() + blob_position_dwords);
    blob_header.fourcc = dxbc::BlobHeader::FourCC::kResourceDefinition;
    blob_position_dwords = uint32_t(built_shader_.size());
    blob_header.size_bytes =
        (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
        built_shader_[blob_offset_position_dwords++];
  }

  // ***************************************************************************
  // Input and output signatures (empty)
  // ***************************************************************************

  for (uint32_t i = 0; i < 2; ++i) {
    built_shader_[blob_offset_position_dwords] =
        uint32_t(blob_position_dwords * sizeof(uint32_t));
    uint32_t signature_position_dwords =
        blob_position_dwords + kBlobHeaderSizeDwords;
    built_shader_.resize(signature_position_dwords +
                         sizeof(dxbc::Signature) / sizeof(uint32_t));
    {
      auto& signature = *reinterpret_cast<dxbc::Signature*>(
          built_shader_.data() + signature_position_dwords);
      // Empty - just set parameter pointer to the end.
      signature.parameter_info_ptr = sizeof(dxbc::Signature);
    }
    {
      auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
          built_shader_.data() + blob_position_dwords);
      blob_header.fourcc = i ? dxbc::BlobHeader::FourCC::kOutputSignature
                             : dxbc::BlobHeader::FourCC::kInputSignature;
      blob_position_dwords = uint32_t(built_shader_.size());
      blob_header.size_bytes =
          (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
          built_shader_[blob_offset_position_dwords++];
    }
  }

  // ***************************************************************************
  // Shader program
  // ***************************************************************************

  built_shader_[blob_offset_position_dwords] =
      uint32_t(blob_position_dwords * sizeof(uint32_t));
  uint32_t shex_position_dwords = blob_position_dwords + kBlobHeaderSizeDwords;
  built_shader_.resize(shex_position_dwords);

  built_shader_.push_back(
      dxbc::VersionToken(dxbc::ProgramType::kComputeShader, 5, 1));
  // Reserve space for the length token.
  built_shader_.push_back(0);

  dxbc::Statistics stat;
  std::memset(&stat, 0, sizeof(dxbc::Statistics));
  dxbc::Assembler a(built_shader_, stat);

  a.OpDclGlobalFlags(dxbc::kGlobalFlagAllResourcesBound);
  a.OpDclConstantBuffer(dxbc::Src::CB(dxbc::Src::Dcl, kDumpCbufferOffsets,
                                      kDumpCbufferOffsets, kDumpCbufferOffsets),
                        1);
  a.OpDclConstantBuffer(dxbc::Src::CB(dxbc::Src::Dcl, kDumpCbufferPitches,
                                      kDumpCbufferPitches, kDumpCbufferPitches),
                        1);
  // Source texture.
  dxbc::ResourceDimension source_dimension =
      key.msaa_samples != xenos::MsaaSamples::k1X
          ? dxbc::ResourceDimension::kTexture2DMS
          : dxbc::ResourceDimension::kTexture2D;
  a.OpDclResource(source_dimension,
                  dxbc::ResourceReturnTypeX4Token(
                      source_is_uint ? dxbc::ResourceReturnType::kUInt
                                     : dxbc::ResourceReturnType::kFloat),
                  dxbc::Src::T(dxbc::Src::Dcl, 0, 0, 0));
  // Source stencil texture.
  if (key.is_depth) {
    a.OpDclResource(
        source_dimension,
        dxbc::ResourceReturnTypeX4Token(dxbc::ResourceReturnType::kUInt),
        dxbc::Src::T(dxbc::Src::Dcl, 1, 1, 1));
  }
  // EDRAM buffer.
  a.OpDclUnorderedAccessViewTyped(
      dxbc::ResourceDimension::kBuffer, 0,
      dxbc::ResourceReturnTypeX4Token(dxbc::ResourceReturnType::kUInt),
      dxbc::Src::U(dxbc::Src::Dcl, 0, 0, 0));
  a.OpDclInput(dxbc::Dest::VThreadID(0b0011));
  // r0 - addressing before the load, then addressing and conversion scratch
  // r1 - addressing scratch before the load, then data
  stat.temp_register_count = 2;
  a.OpDclTemps(stat.temp_register_count);
  // There's no strict dependency on the group size here, for simplicity of
  // calculations especially with resolution scaling, dividing manually (as the
  // group size is not unlimited). The only restriction is that an integer
  // multiple of it must be 80x16 samples (and no larger than that) for 32bpp,
  // or 40x16 samples for 64bpp (because only a half of the pair of tiles may
  // need to be dumped). The group size limit in Direct3D 11 is 1024, and 40x16
  // fits in it, while 80x16 doesn't.
  a.OpDclThreadGroup(40, 16, 1);

  uint32_t draw_resolution_scale_x = this->draw_resolution_scale_x();
  uint32_t draw_resolution_scale_y = this->draw_resolution_scale_y();

  // For now, as the exact addressing in 64bpp render targets relatively to
  // 32bpp is unknown, treating 64bpp tiles as storing 40x16 samples rather than
  // 80x16 for simplicity of addressing into the texture.

  // Get the parts of the address along Y - tile row index within the dispatch
  // to r0.w, sample Y within the tile to r0.y.
  static_assert(
      TextureCache::kMaxDrawResolutionScaleAlongAxis <= 3,
      "D3D12RenderTargetCache render target dump shader generation supports Y "
      "draw resolution scaling factors of only up to 3");
  if (draw_resolution_scale_y == 3) {
    // Multiplication part of the division by the (16 * scale) tile height,
    // specifically 48 here, or 16 * 3.
    // r0.w = (Y * kDivideScale3) >> 32
    a.OpUMul(dxbc::Dest::R(0, 0b1000), dxbc::Dest::Null(),
             dxbc::Src::VThreadID(dxbc::Src::kYYYY),
             dxbc::Src::LU(draw_util::kDivideScale3));
    // Shift part of the division by 16 * scale.
    // r0.w = Y tile position
    a.OpUShR(dxbc::Dest::R(0, 0b1000), dxbc::Src::R(0, dxbc::Src::kWWWW),
             dxbc::Src::LU(draw_util::kDivideUpperShift3 + 4));
    // Take the remainder of the performed division to r0.y.
    // r0.y = Y sample position within the tile
    // r0.w = Y tile position
    a.OpIMAd(dxbc::Dest::R(0, 0b0010), dxbc::Src::R(0, dxbc::Src::kWWWW),
             dxbc::Src::LI(-16 * draw_resolution_scale_y),
             dxbc::Src::VThreadID(dxbc::Src::kYYYY));
  } else {
    assert_true(draw_resolution_scale_y <= 2);
    // Tile height is a power of two, can use bit operations.
    // Get the tile row index into r0.w.
    // r0.w = Y tile position.
    a.OpUShR(dxbc::Dest::R(0, 0b1000), dxbc::Src::VThreadID(dxbc::Src::kYYYY),
             dxbc::Src::LU(draw_resolution_scale_y == 2 ? 5 : 4));
    // Get the Y sample position within the tile into r0.y.
    // r0.y = Y sample position within the tile
    // r0.w = Y tile position
    a.OpAnd(dxbc::Dest::R(0, 0b0010), dxbc::Src::VThreadID(dxbc::Src::kYYYY),
            dxbc::Src::LU((16 * draw_resolution_scale_y) - 1));
  }

  // Get the X tile offset within the dispatch to r0.z.
  uint32_t tile_width = xenos::kEdramTileWidthSamples * draw_resolution_scale_x;
  uint32_t tile_width_divide_scale;
  uint32_t tile_width_divide_upper_shift;
  draw_util::GetEdramTileWidthDivideScaleAndUpperShift(
      draw_resolution_scale_x, tile_width_divide_scale,
      tile_width_divide_upper_shift);
  if (format_is_64bpp) {
    tile_width >>= 1;
    assert_not_zero(tile_width_divide_upper_shift);
    --tile_width_divide_upper_shift;
  }
  // Multiplication part of the division by 80|40 * scale.
  // r0.y = Y sample position within the tile
  // r0.z = (X * tile_width_divide_scale) >> 32
  // r0.w = Y tile position
  a.OpUMul(dxbc::Dest::R(0, 0b0100), dxbc::Dest::Null(),
           dxbc::Src::VThreadID(dxbc::Src::kXXXX),
           dxbc::Src::LU(tile_width_divide_scale));
  // Shift part of the division by 80|40 * scale.
  // r0.y = Y sample position within the tile
  // r0.z = X tile position
  // r0.w = Y tile position
  a.OpUShR(dxbc::Dest::R(0, 0b0100), dxbc::Src::R(0, dxbc::Src::kZZZZ),
           dxbc::Src::LU(tile_width_divide_upper_shift));

  // Extract the dump rectangle tile row pitch to r0.x.
  // r0.x = dump rectangle pitch in tiles
  // r0.y = Y sample position within the tile
  // r0.z = X tile position
  // r0.w = Y tile position
  a.OpUBFE(dxbc::Dest::R(0, 0b0001), dxbc::Src::LU(xenos::kEdramPitchTilesBits),
           dxbc::Src::LU(0),
           dxbc::Src::CB(kDumpCbufferPitches, kDumpCbufferPitches, 0,
                         dxbc::Src::kXXXX));
  // Get the tile index in the EDRAM relative to the dump rectangle base tile to
  // r0.w.
  // r0.x = free
  // r0.y = Y sample position within the tile
  // r0.z = X tile position
  // r0.w = tile index relative to the dump rectangle base
  a.OpUMAd(dxbc::Dest::R(0, 0b1000), dxbc::Src::R(0, dxbc::Src::kWWWW),
           dxbc::Src::R(0, dxbc::Src::kXXXX),
           dxbc::Src::R(0, dxbc::Src::kZZZZ));

  // Take the X sample index within the tile as the remainder of the division of
  // the thread index by tile width to r0.x.
  // r0.x = X sample position within the tile
  // r0.y = Y sample position within the tile
  // r0.z = free
  // r0.w = tile index relative to the dump rectangle base
  a.OpIMAd(dxbc::Dest::R(0, 0b0001), dxbc::Src::R(0, dxbc::Src::kZZZZ),
           dxbc::Src::LI(-int32_t(tile_width)),
           dxbc::Src::VThreadID(dxbc::Src::kXXXX));

  // Extract the index of the first tile of the dispatch in the EDRAM to r0.z.
  // r0.x = X sample position within the tile
  // r0.y = Y sample position within the tile
  // r0.z = first EDRAM tile index in the dispatch
  // r0.w = tile index relative to the dump rectangle base
  a.OpUBFE(dxbc::Dest::R(0, 0b0100), dxbc::Src::LU(xenos::kEdramBaseTilesBits),
           dxbc::Src::LU(0),
           dxbc::Src::CB(kDumpCbufferOffsets, kDumpCbufferOffsets, 0,
                         dxbc::Src::kXXXX));
  // Add the base tile in the dispatch to the dispatch-local tile index to r0.w.
  // r0.x = X sample position within the tile
  // r0.y = Y sample position within the tile
  // r0.z = free
  // r0.w = tile index in the EDRAM
  a.OpIAdd(dxbc::Dest::R(0, 0b1000), dxbc::Src::R(0, dxbc::Src::kWWWW),
           dxbc::Src::R(0, dxbc::Src::kZZZZ));
  // Convert the tile index to samples and add the X sample index to it to r0.z.
  // r0.x = X sample position within the tile
  // r0.y = Y sample position within the tile
  // r0.z = tile sample offset in the EDRAM plus X sample offset
  // r0.w = tile index in the EDRAM
  a.OpUMAd(dxbc::Dest::R(0, 0b0100), dxbc::Src::R(0, dxbc::Src::kWWWW),
           dxbc::Src::LU(
               draw_resolution_scale_x * draw_resolution_scale_y *
               (xenos::kEdramTileWidthSamples >> uint32_t(format_is_64bpp)) *
               xenos::kEdramTileHeightSamples),
           dxbc::Src::R(0, dxbc::Src::kXXXX));
  // Add the contribution of the Y sample position within the tile to the sample
  // address in the EDRAM to r0.w.
  // r0.x = X sample position within the tile
  // r0.y = Y sample position within the tile
  // r0.z = sample offset in the EDRAM without the depth column swapping
  // r0.w = tile index in the EDRAM
  a.OpUMAd(dxbc::Dest::R(0, 0b0100), dxbc::Src::R(0, dxbc::Src::kYYYY),
           dxbc::Src::LU(tile_width), dxbc::Src::R(0, dxbc::Src::kZZZZ));
  if (key.is_depth) {
    uint32_t tile_width_half = tile_width >> 1;
    // Get which 40-sample half within the tile is being processed to r1.x.
    // r0.x = X sample position within the tile
    // r0.y = Y sample position within the tile
    // r0.z = sample offset in the EDRAM without the depth column swapping
    // r0.w = tile index in the EDRAM
    // r1.x = 0xFFFFFFFF if in the right 40-sample half, 0 otherwise
    a.OpUGE(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(0, dxbc::Src::kXXXX),
            dxbc::Src::LU(tile_width_half));
    // Get the offset needed to swap 40-sample halves for depth.
    // r0.x = X sample position within the tile
    // r0.y = Y sample position within the tile
    // r0.z = sample offset in the EDRAM without the depth column swapping
    // r0.w = tile index in the EDRAM
    // r1.x = depth half-tile flipping offset
    a.OpMovC(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX),
             dxbc::Src::LI(-int32_t(tile_width_half)),
             dxbc::Src::LI(int32_t(tile_width_half)));
    // Swap 40-sample columns in the depth buffer in the destination address in
    // r0.w to get the final address of the sample in EDRAM.
    // r0.x = X sample position within the tile
    // r0.y = Y sample position within the tile
    // r0.z = sample offset in the EDRAM
    // r0.w = tile index in the EDRAM
    // r1.x = free
    a.OpIAdd(dxbc::Dest::R(0, 0b0100), dxbc::Src::R(0, dxbc::Src::kZZZZ),
             dxbc::Src::R(1, dxbc::Src::kXXXX));
  }

  // Extract the source texture base tile index to r1.x.
  // r0.x = X sample position within the tile
  // r0.y = Y sample position within the tile
  // r0.z = sample offset in the EDRAM
  // r0.w = tile index in the EDRAM
  // r1.x = source texture base tile index
  a.OpUBFE(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(xenos::kEdramBaseTilesBits),
           dxbc::Src::LU(xenos::kEdramBaseTilesBits),
           dxbc::Src::CB(kDumpCbufferOffsets, kDumpCbufferOffsets, 0,
                         dxbc::Src::kXXXX));
  // Get the linear tile index within the source texture to r0.w.
  // r0.x = X sample position within the tile
  // r0.y = Y sample position within the tile
  // r0.z = sample offset in the EDRAM
  // r0.w = linear tile index in the source texture
  // r1.x = free
  a.OpIAdd(dxbc::Dest::R(0, 0b1000), dxbc::Src::R(0, dxbc::Src::kWWWW),
           -dxbc::Src::R(1, dxbc::Src::kXXXX));
  // Get the source texture pitch in tiles to r1.x.
  // r0.x = X sample position within the tile
  // r0.y = Y sample position within the tile
  // r0.z = sample offset in the EDRAM
  // r0.w = linear tile index in the source texture
  // r1.x = source texture pitch in tiles
  a.OpUBFE(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(xenos::kEdramPitchTilesBits),
           dxbc::Src::LU(xenos::kEdramPitchTilesBits),
           dxbc::Src::CB(kDumpCbufferPitches, kDumpCbufferPitches, 0,
                         dxbc::Src::kXXXX));
  // Split the linear tile index in the source texture into X and Y in tiles.
  // Get the source texture pitch in tiles to r1.x.
  // r0.x = X sample position within the tile
  // r0.y = Y sample position within the tile
  // r0.z = sample offset in the EDRAM
  // r0.w = X tile index within the tile row in the source texture
  // r1.x = Y tile row index within the source texture
  a.OpUDiv(dxbc::Dest::R(1, 0b0001), dxbc::Dest::R(0, 0b1000),
           dxbc::Src::R(0, dxbc::Src::kWWWW),
           dxbc::Src::R(1, dxbc::Src::kXXXX));
  // Add the source texture tile X offset to the source texture sample X
  // coordinate.
  // r0.x = X sample position within the source texture
  // r0.y = Y sample position within the tile
  // r0.z = sample offset in the EDRAM
  // r0.w = free
  // r1.x = Y tile row index within the source texture
  a.OpUMAd(dxbc::Dest::R(0, 0b0001), dxbc::Src::R(0, dxbc::Src::kWWWW),
           dxbc::Src::LU(tile_width), dxbc::Src::R(0, dxbc::Src::kXXXX));
  // Add the source texture tile Y offset to the source texture sample Y
  // coordinate.
  // r0.x = X sample position within the source texture
  // r0.y = Y sample position within the source texture
  // r0.z = sample offset in the EDRAM
  // r1.x = free
  a.OpUMAd(
      dxbc::Dest::R(0, 0b0010), dxbc::Src::R(1, dxbc::Src::kXXXX),
      dxbc::Src::LU(xenos::kEdramTileHeightSamples * draw_resolution_scale_y),
      dxbc::Src::R(0, dxbc::Src::kYYYY));
  // Will be using the source texture coordinates from r0.xy, and for
  // single-sampled source, LOD from r0.w.
  dxbc::Src source_address_src(dxbc::Src::R(0, 0b11000100));
  if (key.msaa_samples >= xenos::MsaaSamples::k2X) {
    if (key.msaa_samples >= xenos::MsaaSamples::k4X) {
      // 4x MSAA source texture sample index - bit 0 for horizontal, bit 1 for
      // vertical.
      // Extract the horizontal sample index to r0.w.
      // r0.x = X sample position within the source texture
      // r0.y = Y sample position within the source texture
      // r0.z = sample offset in the EDRAM
      // r0.w = horizontal sample index within the source pixel
      a.OpAnd(dxbc::Dest::R(0, 0b1000), dxbc::Src::R(0, dxbc::Src::kXXXX),
              dxbc::Src::LU(1));
      // Insert the vertical sample index to r0.w.
      // r0.x = X sample position within the source texture
      // r0.y = Y sample position within the source texture
      // r0.z = sample offset in the EDRAM
      // r0.w = sample index within the source pixel
      a.OpBFI(dxbc::Dest::R(0, 0b1000), dxbc::Src::LU(1), dxbc::Src::LU(1),
              dxbc::Src::R(0, dxbc::Src::kYYYY),
              dxbc::Src::R(0, dxbc::Src::kWWWW));
      // Convert sample to pixel coordinates in the source texture to r0.xy.
      // r0.x = X pixel position within the source texture
      // r0.y = Y pixel position within the source texture
      // r0.z = sample offset in the EDRAM
      // r0.w = sample index within the source pixel
      a.OpUShR(dxbc::Dest::R(0, 0b0011), dxbc::Src::R(0), dxbc::Src::LU(1));
    } else {
      // 2x MSAA source texture sample index.
      // Extract the vertical sample index to r0.w.
      // r0.x = X pixel position within the source texture
      // r0.y = Y sample position within the source texture
      // r0.z = sample offset in the EDRAM
      // r0.w = vertical sample index within the destination pixel
      a.OpAnd(dxbc::Dest::R(0, 0b1000), dxbc::Src::R(0, dxbc::Src::kYYYY),
              dxbc::Src::LU(1));
      // Convert the 2x MSAA sample index from the guest to Direct3D 10.1+.
      // r0.x = X pixel position within the source texture
      // r0.y = Y sample position within the source texture
      // r0.z = sample offset in the EDRAM
      // r0.w = sample index within the source pixel
      a.OpMovC(dxbc::Dest::R(0, 0b1000), dxbc::Src::R(0, dxbc::Src::kWWWW),
               dxbc::Src::LU(draw_util::GetD3D10SampleIndexForGuest2xMSAA(
                   1, msaa_2x_supported_)),
               dxbc::Src::LU(draw_util::GetD3D10SampleIndexForGuest2xMSAA(
                   0, msaa_2x_supported_)));
      // Convert sample Y to pixel Y in the source texture to r0.y.
      // r0.x = X pixel position within the source texture
      // r0.y = Y pixel position within the source texture
      // r0.z = sample offset in the EDRAM
      // r0.w = sample index within the source pixel
      a.OpUShR(dxbc::Dest::R(0, 0b0010), dxbc::Src::R(0, dxbc::Src::kYYYY),
               dxbc::Src::LU(1));
    }
    // Load the source to r1.
    // r0.x = X pixel position within the source texture if stencil is needed
    // r0.y = Y pixel position within the source texture if stencil is needed
    // r0.z = sample offset in the EDRAM
    // r0.w = sample index within the source pixel if stencil is needed
    // r1 = source texel value
    a.OpLdMS(dxbc::Dest::R(1, (1 << source_component_count) - 1),
             source_address_src, 0b0011, dxbc::Src::T(0, 0),
             dxbc::Src::R(0, dxbc::Src::kWWWW));
    if (key.is_depth) {
      // Load the source stencil to r1.y.
      // r0.x = free
      // r0.y = free
      // r0.z = sample offset in the EDRAM
      // r0.w = free
      // r1.x = source depth value
      // r1.y = source stencil value
      a.OpLdMS(dxbc::Dest::R(1, 0b0010), source_address_src, 0b0011,
               dxbc::Src::T(1, 1), dxbc::Src::R(0, dxbc::Src::kWWWW));
    }
  } else {
    // Write the LOD index (0) to the register with texture coordinates for
    // loading from the single-sampled source texture.
    // r0.x = X pixel position within the source texture
    // r0.y = Y pixel position within the source texture
    // r0.z = sample offset in the EDRAM
    // r0.w = LOD for the texture load (zero)
    a.OpMov(dxbc::Dest::R(0, 0b1000), dxbc::Src::LF(0.0f));
    // Load the source to r1.
    // r0.x = X pixel position within the source texture if stencil is needed
    // r0.y = Y pixel position within the source texture if stencil is needed
    // r0.z = sample offset in the EDRAM
    // r0.w = LOD for the texture load (zero)
    // r1 = source texel value
    a.OpLd(dxbc::Dest::R(1, (1 << source_component_count) - 1),
           source_address_src, 0b1011, dxbc::Src::T(0, 0));
    if (key.is_depth) {
      // Load the source stencil to r1.y.
      // r0.x = free
      // r0.y = free
      // r0.z = sample offset in the EDRAM
      // r0.w = free
      // r1.x = source depth value
      // r1.y = source stencil value
      a.OpLd(dxbc::Dest::R(1, 0b0010), source_address_src, 0b1011,
             dxbc::Src::T(1, 1));
    }
  }

  // Pack in the needed format, writing the result to r1.x for 32bpp or r1.xy
  // for 64bpp.
  // r0.xyw are usable as temporary storage.
  if (key.is_depth) {
    switch (key.GetDepthFormat()) {
      case xenos::DepthRenderTargetFormat::kD24S8:
        // Round to the nearest even integer. This seems to be the correct,
        // adding +0.5 and rounding towards zero results in red instead of
        // black in the 4D5307E6 clear shader.
        a.OpMul(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX),
                dxbc::Src::LF(float(0xFFFFFF)));
        a.OpRoundNE(dxbc::Dest::R(1, 0b0001),
                    dxbc::Src::R(1, dxbc::Src::kXXXX));
        a.OpFToU(dxbc::Dest::R(1, 0b0001), dxbc::Src::R(1, dxbc::Src::kXXXX));
        break;
      case xenos::DepthRenderTargetFormat::kD24FS8:
        // Convert to [0, 2) float24 from [0, 1) float32, using r0.x as
        // temporary.
        DxbcShaderTranslator::PreClampedDepthTo20e4(a, 1, 0, 1, 0, 0, 0, true);
        break;
    }
    // Combine 24-bit depth and stencil into r1.x.
    a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(24), dxbc::Src::LU(8),
            dxbc::Src::R(1, dxbc::Src::kXXXX),
            dxbc::Src::R(1, dxbc::Src::kYYYY));
  } else {
    switch (key.GetColorFormat()) {
      case xenos::ColorRenderTargetFormat::k_8_8_8_8:
      case xenos::ColorRenderTargetFormat::k_8_8_8_8_GAMMA:
        if (!source_is_uint) {
          a.OpMAd(dxbc::Dest::R(1), dxbc::Src::R(1), dxbc::Src::LF(255.0f),
                  dxbc::Src::LF(0.5f));
          a.OpFToU(dxbc::Dest::R(1), dxbc::Src::R(1));
        }
        for (uint32_t i = 1; i < 4; ++i) {
          a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(8),
                  dxbc::Src::LU(i * 8), dxbc::Src::R(1).Select(i),
                  dxbc::Src::R(1, dxbc::Src::kXXXX));
        }
        break;
      case xenos::ColorRenderTargetFormat::k_2_10_10_10:
      case xenos::ColorRenderTargetFormat::k_2_10_10_10_AS_10_10_10_10:
        if (!source_is_uint) {
          a.OpMAd(dxbc::Dest::R(1), dxbc::Src::R(1),
                  dxbc::Src::LF(1023.0f, 1023.0f, 1023.0f, 3.0f),
                  dxbc::Src::LF(0.5f));
          a.OpFToU(dxbc::Dest::R(1), dxbc::Src::R(1));
        }
        for (uint32_t i = 1; i < 4; ++i) {
          a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(i == 3 ? 2 : 10),
                  dxbc::Src::LU(i * 10), dxbc::Src::R(1).Select(i),
                  dxbc::Src::R(1, dxbc::Src::kXXXX));
        }
        break;
      case xenos::ColorRenderTargetFormat::k_2_10_10_10_FLOAT:
      case xenos::ColorRenderTargetFormat::k_2_10_10_10_FLOAT_AS_16_16_16_16:
        // Float16 has a wider range for both color and alpha, also NaNs.
        // Color - clamp and convert.
        // Convert red in r1.x to the result register r1.x - the same, but
        // UnclampedFloat32To7e3 allows that - using r0.x as a temporary.
        DxbcShaderTranslator::UnclampedFloat32To7e3(a, 1, 0, 1, 0, 0, 0);
        for (uint32_t i = 1; i < 3; ++i) {
          // Convert green and blue to a temporary register r0.x using r0.y
          // as an internal temporary, then insert them into the result in
          // r1.x.
          DxbcShaderTranslator::UnclampedFloat32To7e3(a, 0, 0, 1, i, 0, 1);
          a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(10),
                  dxbc::Src::LU(i * 10), dxbc::Src::R(0, dxbc::Src::kXXXX),
                  dxbc::Src::R(1, dxbc::Src::kXXXX));
        }
        // Alpha - saturate and convert.
        a.OpMov(dxbc::Dest::R(1, 0b1000), dxbc::Src::R(1, dxbc::Src::kWWWW),
                true);
        a.OpMAd(dxbc::Dest::R(1, 0b1000), dxbc::Src::R(1, dxbc::Src::kWWWW),
                dxbc::Src::LF(3.0f), dxbc::Src::LF(0.5f));
        a.OpFToU(dxbc::Dest::R(1, 0b1000), dxbc::Src::R(1, dxbc::Src::kWWWW));
        a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(2), dxbc::Src::LU(30),
                dxbc::Src::R(1, dxbc::Src::kWWWW),
                dxbc::Src::R(1, dxbc::Src::kXXXX));
        break;
      case xenos::ColorRenderTargetFormat::k_16_16:
      case xenos::ColorRenderTargetFormat::k_16_16_FLOAT:
        assert_true(source_is_uint);
        a.OpBFI(dxbc::Dest::R(1, 0b0001), dxbc::Src::LU(16), dxbc::Src::LU(16),
                dxbc::Src::R(1, dxbc::Src::kYYYY),
                dxbc::Src::R(1, dxbc::Src::kXXXX));
        break;
      case xenos::ColorRenderTargetFormat::k_16_16_16_16:
      case xenos::ColorRenderTargetFormat::k_16_16_16_16_FLOAT:
        assert_true(source_is_uint);
        a.OpBFI(dxbc::Dest::R(1, 0b0011), dxbc::Src::LU(16), dxbc::Src::LU(16),
                dxbc::Src::R(1, 0b1101), dxbc::Src::R(1, 0b1000));
        break;
      case xenos::ColorRenderTargetFormat::k_32_FLOAT:
      case xenos::ColorRenderTargetFormat::k_32_32_FLOAT:
        assert_true(source_is_uint);
        // Already has the needed representation.
        break;
    }
  }

  // Write the sample to the destination address stored in r0.z.
  a.OpStoreUAVTyped(
      dxbc::Dest::U(0, 0), dxbc::Src::R(0, dxbc::Src::kZZZZ), 1,
      dxbc::Src::R(1, format_is_64bpp ? 0b0100 : dxbc::Src::kXXXX));

  a.OpRet();

  // Write the shader program length in dwords.
  built_shader_[shex_position_dwords + 1] =
      uint32_t(built_shader_.size()) - shex_position_dwords;

  {
    auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
        built_shader_.data() + blob_position_dwords);
    blob_header.fourcc = dxbc::BlobHeader::FourCC::kShaderEx;
    blob_position_dwords = uint32_t(built_shader_.size());
    blob_header.size_bytes =
        (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
        built_shader_[blob_offset_position_dwords++];
  }

  // ***************************************************************************
  // Statistics
  // ***************************************************************************

  built_shader_[blob_offset_position_dwords] =
      uint32_t(blob_position_dwords * sizeof(uint32_t));
  uint32_t stat_position_dwords = blob_position_dwords + kBlobHeaderSizeDwords;
  built_shader_.resize(stat_position_dwords +
                       sizeof(dxbc::Statistics) / sizeof(uint32_t));
  std::memcpy(built_shader_.data() + stat_position_dwords, &stat,
              sizeof(dxbc::Statistics));
  {
    auto& blob_header = *reinterpret_cast<dxbc::BlobHeader*>(
        built_shader_.data() + blob_position_dwords);
    blob_header.fourcc = dxbc::BlobHeader::FourCC::kStatistics;
    blob_position_dwords = uint32_t(built_shader_.size());
    blob_header.size_bytes =
        (blob_position_dwords - kBlobHeaderSizeDwords) * sizeof(uint32_t) -
        built_shader_[blob_offset_position_dwords++];
  }

  // ***************************************************************************
  // Container header
  // ***************************************************************************

  uint32_t built_shader_size_bytes =
      uint32_t(built_shader_.size() * sizeof(uint32_t));
  {
    auto& container_header =
        *reinterpret_cast<dxbc::ContainerHeader*>(built_shader_.data());
    container_header.InitializeIdentification();
    container_header.size_bytes = built_shader_size_bytes;
    container_header.blob_count = kBlobCount;
    CalculateDXBCChecksum(
        reinterpret_cast<unsigned char*>(built_shader_.data()),
        static_cast<unsigned int>(built_shader_size_bytes),
        reinterpret_cast<unsigned int*>(&container_header.hash));
  }

  // ***************************************************************************
  // Pipeline
  // ***************************************************************************
  ID3D12PipelineState* pipeline = ui::d3d12::util::CreateComputePipeline(
      command_processor_.GetD3D12Provider().GetDevice(), built_shader_.data(),
      built_shader_size_bytes,
      key.is_depth ? dump_root_signature_depth_ : dump_root_signature_color_);
  const char* format_name =
      key.is_depth
          ? xenos::GetDepthRenderTargetFormatName(key.GetDepthFormat())
          : xenos::GetColorRenderTargetFormatName(key.GetColorFormat());
  if (pipeline) {
    std::u16string pipeline_name =
        xe::to_utf16(fmt::format("RT Dump {} {}xMSAA", format_name,
                                 uint32_t(1) << uint32_t(key.msaa_samples)));
    pipeline->SetName(reinterpret_cast<LPCWSTR>(pipeline_name.c_str()));
  } else {
    XELOGE(
        "D3D12RenderTargetCache: Failed to create a render target dumping "
        "pipeline for {}-sample render targets with format {}",
        uint32_t(1) << uint32_t(key.msaa_samples), format_name);
  }
  // Even if creation fails, still store the null pointer not to try to create
  // again.
  dump_pipelines_.emplace(key, pipeline);
  return pipeline;
}

void D3D12RenderTargetCache::DumpRenderTargets(uint32_t dump_base,
                                               uint32_t dump_row_length_used,
                                               uint32_t dump_rows,
                                               uint32_t dump_pitch) {
  assert_true(GetPath() == Path::kHostRenderTargets);

  GetResolveCopyRectanglesToDump(dump_base, dump_row_length_used, dump_rows,
                                 dump_pitch, dump_rectangles_);
  if (dump_rectangles_.empty()) {
    return;
  }

  // Clear previously set temporary indices.
  for (const ResolveCopyDumpRectangle& rectangle : dump_rectangles_) {
    auto& d3d12_rt = *static_cast<D3D12RenderTarget*>(rectangle.render_target);
    d3d12_rt.SetTemporarySortIndex(UINT32_MAX);
    d3d12_rt.SetTemporarySRVDescriptorIndex(UINT32_MAX);
    d3d12_rt.SetTemporarySRVDescriptorIndexStencil(UINT32_MAX);
  }
  // Gather all needed barriers and info needed to create descriptors and to
  // sort the invocations.
  TransitionEdramBuffer(D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
  dump_invocations_.clear();
  dump_invocations_.reserve(dump_rectangles_.size());
  current_temporary_descriptors_cpu_.clear();
  bool any_sources_32bpp_64bpp[2] = {};
  uint32_t rt_sort_index = 0;
  for (const ResolveCopyDumpRectangle& rectangle : dump_rectangles_) {
    auto& d3d12_rt = *static_cast<D3D12RenderTarget*>(rectangle.render_target);
    command_processor_.PushTransitionBarrier(
        d3d12_rt.resource(),
        d3d12_rt.SetResourceState(
            D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE),
        D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE);
    if (d3d12_rt.temporary_sort_index() == UINT32_MAX) {
      d3d12_rt.SetTemporarySortIndex(rt_sort_index++);
    }
    if (d3d12_rt.temporary_srv_descriptor_index() == UINT32_MAX) {
      d3d12_rt.SetTemporarySRVDescriptorIndex(
          uint32_t(current_temporary_descriptors_cpu_.size()));
      current_temporary_descriptors_cpu_.push_back(
          d3d12_rt.descriptor_srv().GetHandle());
    }
    RenderTargetKey rt_key = d3d12_rt.key();
    if (rt_key.is_depth &&
        d3d12_rt.temporary_srv_descriptor_index_stencil() == UINT32_MAX) {
      d3d12_rt.SetTemporarySRVDescriptorIndexStencil(
          uint32_t(current_temporary_descriptors_cpu_.size()));
      current_temporary_descriptors_cpu_.push_back(
          d3d12_rt.descriptor_srv_stencil().GetHandle());
    }
    any_sources_32bpp_64bpp[size_t(rt_key.Is64bpp())] = true;
    DumpPipelineKey pipeline_key;
    pipeline_key.msaa_samples = rt_key.msaa_samples;
    pipeline_key.resource_format = rt_key.resource_format;
    pipeline_key.is_depth = rt_key.is_depth;
    dump_invocations_.emplace_back(rectangle, pipeline_key);
  }
  // 32bpp and 64bpp.
  size_t edram_uav_indices[2] = {SIZE_MAX, SIZE_MAX};
  const ui::d3d12::D3D12Provider& provider =
      command_processor_.GetD3D12Provider();
  if (!bindless_resources_used_) {
    if (any_sources_32bpp_64bpp[0]) {
      edram_uav_indices[0] = current_temporary_descriptors_cpu_.size();
      current_temporary_descriptors_cpu_.push_back(
          provider.OffsetViewDescriptor(
              edram_buffer_descriptor_heap_start_,
              uint32_t(EdramBufferDescriptorIndex::kR32UintUAV)));
    }
    if (any_sources_32bpp_64bpp[1]) {
      edram_uav_indices[1] = current_temporary_descriptors_cpu_.size();
      current_temporary_descriptors_cpu_.push_back(
          provider.OffsetViewDescriptor(
              edram_buffer_descriptor_heap_start_,
              uint32_t(EdramBufferDescriptorIndex::kR32G32UintUAV)));
    }
  }

  // Copy source descriptors to a shader-visible heap.
  ID3D12Device* device = provider.GetDevice();
  uint32_t descriptor_count =
      uint32_t(current_temporary_descriptors_cpu_.size());
  current_temporary_descriptors_gpu_.resize(descriptor_count);
  if (!command_processor_.RequestOneUseSingleViewDescriptors(
          descriptor_count, current_temporary_descriptors_gpu_.data())) {
    return;
  }
  for (uint32_t i = 0; i < descriptor_count; ++i) {
    device->CopyDescriptorsSimple(1,
                                  current_temporary_descriptors_gpu_[i].first,
                                  current_temporary_descriptors_cpu_[i],
                                  D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
  }

  // Sort the invocations to reduce context and binding switches.
  std::sort(dump_invocations_.begin(), dump_invocations_.end());

  // Dump the render targets.
  DeferredCommandList& command_list =
      command_processor_.GetDeferredCommandList();
  ID3D12RootSignature* last_root_signature = nullptr;
  uint32_t root_parameters_set = 0;
  uint32_t last_descriptor_index_source = UINT32_MAX;
  uint32_t last_descriptor_index_stencil = UINT32_MAX;
  bool last_edram_uav_is_64bpp = false;
  DumpOffsets last_offsets;
  DumpPitches last_pitches;
  for (const DumpInvocation& invocation : dump_invocations_) {
    const ResolveCopyDumpRectangle& rectangle = invocation.rectangle;
    auto& d3d12_rt = *static_cast<D3D12RenderTarget*>(rectangle.render_target);
    RenderTargetKey rt_key = d3d12_rt.key();
    DumpPipelineKey pipeline_key = invocation.pipeline_key;
    ID3D12PipelineState* pipeline = GetOrCreateDumpPipeline(pipeline_key);
    if (!pipeline) {
      continue;
    }
    command_processor_.SetExternalPipeline(pipeline);

    ID3D12RootSignature* root_signature = pipeline_key.is_depth
                                              ? dump_root_signature_depth_
                                              : dump_root_signature_color_;
    if (last_root_signature != root_signature) {
      last_root_signature = root_signature;
      command_list.D3DSetComputeRootSignature(root_signature);
      root_parameters_set = 0;
    }

    DumpRootParameter root_parameter_edram = pipeline_key.is_depth
                                                 ? kDumpRootParameterDepthEdram
                                                 : kDumpRootParameterColorEdram;
    uint32_t root_parameter_edram_bit = uint32_t(1) << root_parameter_edram;
    bool format_is_64bpp = rt_key.Is64bpp();
    if (last_edram_uav_is_64bpp != format_is_64bpp) {
      last_edram_uav_is_64bpp = format_is_64bpp;
      root_parameters_set &= ~root_parameter_edram_bit;
    }
    if (!(root_parameters_set & root_parameter_edram_bit)) {
      D3D12_GPU_DESCRIPTOR_HANDLE descriptor_handle_edram;
      if (bindless_resources_used_) {
        descriptor_handle_edram = command_processor_
                                      .GetEdramUintPow2BindlessUAVHandlePair(
                                          2 + uint32_t(last_edram_uav_is_64bpp))
                                      .second;
      } else {
        assert_true(edram_uav_indices[size_t(last_edram_uav_is_64bpp)] !=
                    SIZE_MAX);
        descriptor_handle_edram =
            current_temporary_descriptors_gpu_[edram_uav_indices[size_t(
                                                   last_edram_uav_is_64bpp)]]
                .second;
      }
      command_list.D3DSetComputeRootDescriptorTable(root_parameter_edram,
                                                    descriptor_handle_edram);
      root_parameters_set |= root_parameter_edram_bit;
    }

    DumpRootParameter root_parameter_pitches =
        pipeline_key.is_depth ? kDumpRootParameterDepthPitches
                              : kDumpRootParameterColorPitches;
    uint32_t root_parameter_pitches_bit = uint32_t(1) << root_parameter_pitches;
    DumpPitches pitches;
    pitches.dest_pitch = dump_pitch;
    pitches.source_pitch = rt_key.GetPitchTiles();
    if (last_pitches != pitches) {
      last_pitches = pitches;
      root_parameters_set &= ~root_parameter_pitches_bit;
    }
    if (!(root_parameters_set & root_parameter_pitches_bit)) {
      command_list.D3DSetComputeRoot32BitConstants(
          root_parameter_pitches, sizeof(last_pitches) / sizeof(uint32_t),
          &last_pitches, 0);
      root_parameters_set |= root_parameter_pitches_bit;
    }

    if (pipeline_key.is_depth) {
      constexpr uint32_t kDumpRootParameterDepthStencilBit =
          uint32_t(1) << kDumpRootParameterDepthStencil;
      uint32_t descriptor_index_stencil =
          d3d12_rt.temporary_srv_descriptor_index_stencil();
      assert_true(descriptor_index_stencil != UINT32_MAX);
      if (last_descriptor_index_stencil != descriptor_index_stencil) {
        last_descriptor_index_stencil = descriptor_index_stencil;
        root_parameters_set &= ~kDumpRootParameterDepthStencilBit;
      }
      if (!(root_parameters_set & kDumpRootParameterDepthStencilBit)) {
        command_list.D3DSetComputeRootDescriptorTable(
            kDumpRootParameterDepthStencil,
            current_temporary_descriptors_gpu_[last_descriptor_index_stencil]
                .second);
        root_parameters_set |= kDumpRootParameterDepthStencilBit;
      }
    }

    constexpr uint32_t kDumpRootParameterSourceBit =
        uint32_t(1) << kDumpRootParameterSource;
    uint32_t descriptor_index_source =
        d3d12_rt.temporary_srv_descriptor_index();
    assert_true(descriptor_index_source != UINT32_MAX);
    if (last_descriptor_index_source != descriptor_index_source) {
      last_descriptor_index_source = descriptor_index_source;
      root_parameters_set &= ~kDumpRootParameterSourceBit;
    }
    if (!(root_parameters_set & kDumpRootParameterSourceBit)) {
      command_list.D3DSetComputeRootDescriptorTable(
          kDumpRootParameterSource,
          current_temporary_descriptors_gpu_[last_descriptor_index_source]
              .second);
      root_parameters_set |= kDumpRootParameterSourceBit;
    }

    constexpr uint32_t kDumpRootParameterOffsetsBit =
        uint32_t(1) << kDumpRootParameterOffsets;
    DumpOffsets offsets;
    offsets.source_base_tiles = rt_key.base_tiles;
    ResolveCopyDumpRectangle::Dispatch
        dispatches[ResolveCopyDumpRectangle::kMaxDispatches];
    uint32_t dispatch_count =
        rectangle.GetDispatches(dump_pitch, dump_row_length_used, dispatches);
    for (uint32_t i = 0; i < dispatch_count; ++i) {
      const ResolveCopyDumpRectangle::Dispatch& dispatch = dispatches[i];
      offsets.dispatch_first_tile = dump_base + dispatch.offset;
      if (last_offsets != offsets) {
        last_offsets = offsets;
        root_parameters_set &= ~kDumpRootParameterOffsetsBit;
      }
      if (!(root_parameters_set & kDumpRootParameterOffsetsBit)) {
        command_list.D3DSetComputeRoot32BitConstants(
            kDumpRootParameterOffsets, sizeof(last_offsets) / sizeof(uint32_t),
            &last_offsets, 0);
        root_parameters_set |= kDumpRootParameterOffsetsBit;
      }
      command_processor_.SubmitBarriers();
      // Processing 40 x 16 x scale samples per dispatch (a 32bpp tile in two
      // dispatches at 1x1 scale, 64bpp in one dispatch).
      command_list.D3DDispatch(
          (dispatch.width_tiles * draw_resolution_scale_x())
              << uint32_t(!format_is_64bpp),
          dispatch.height_tiles * draw_resolution_scale_y(), 1);
    }
    MarkEdramBufferModified();
  }
}

}  // namespace d3d12
}  // namespace gpu
}  // namespace xe