// Copyright 2018 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include "gpu/command_buffer/client/webgpu_implementation.h"

#include <dawn/wire/client/webgpu.h>

#include <algorithm>
#include <vector>

#include "base/compiler_specific.h"
#include "base/notimplemented.h"
#include "base/numerics/checked_math.h"
#include "base/run_loop.h"
#include "base/trace_event/trace_event.h"
#include "base/types/optional_util.h"
#include "gpu/command_buffer/client/dawn_client_memory_transfer_service.h"
#include "gpu/command_buffer/client/dawn_client_serializer.h"
#include "gpu/command_buffer/client/gpu_control.h"
#include "gpu/command_buffer/client/shared_memory_limits.h"
#include "third_party/perfetto/include/perfetto/tracing/track_event_args.h"

#define GPU_CLIENT_SINGLE_THREAD_CHECK()

namespace gpu {
namespace webgpu {

#if BUILDFLAG(USE_DAWN)
DawnWireServices::~DawnWireServices() {
  wgpuDawnWireClientInstanceRelease(wgpu_instance_);
}

DawnWireServices::DawnWireServices(
    WebGPUImplementation* webgpu_implementation,
    WebGPUCmdHelper* helper,
    MappedMemoryManager* mapped_memory,
    std::unique_ptr<TransferBuffer> transfer_buffer,
    bool support_locking)
    : lock_(support_locking ? std::make_optional<base::Lock>() : std::nullopt),
      memory_transfer_service_(mapped_memory),
      serializer_(webgpu_implementation,
                  helper,
                  &memory_transfer_service_,
                  std::move(transfer_buffer)),
      wire_client_(dawn::wire::WireClientDescriptor{
          &serializer_,
          &memory_transfer_service_,
      }) {
  std::vector<WGPUInstanceFeatureName> instance_features = {
      WGPUInstanceFeatureName_TimedWaitAny};
  WGPUInstanceDescriptor instance_desc = WGPU_INSTANCE_DESCRIPTOR_INIT;
  instance_desc.requiredFeatureCount = instance_features.size();
  instance_desc.requiredFeatures = instance_features.data();
  wgpu_instance_ = wire_client_.ReserveInstance(&instance_desc).instance;
  DCHECK(wgpu_instance_);
}

base::WeakPtr<DawnWireServices> DawnWireServices::AsWeakPtr() {
  return weak_ptr_factory_.GetWeakPtr();
}

WGPUInstance DawnWireServices::GetWGPUInstance() const {
  return wgpu_instance_;
}

void DawnWireServices::Disconnect() {
  {
    base::AutoLockMaybe lock(OptionalToPtr(lock_));
    disconnected_ = true;
  }
  wire_client_.Disconnect();
  serializer_.Disconnect();
  memory_transfer_service_.Disconnect();
}

void DawnWireServices::HandleCommands(const cmds::DawnReturnCommandsInfo& info,
                                      size_t size) {
  TRACE_EVENT(TRACE_DISABLED_BY_DEFAULT("gpu.dawn"), "DawnReturnCommands",
              perfetto::TerminatingFlow::Global(
                  info.header.return_data_header.trace_id));

  base::AutoLockMaybe lock(OptionalToPtr(lock_));
  if (disconnected_) {
    return;
  }

  // Commands from the GPU process are expected to be well-formed.
  CHECK(wire_client_.HandleCommands(info.deserialized_buffer, size));
}

void DawnWireServices::ProcessEvents() {
  wgpuDawnWireClientInstanceProcessEvents(wgpu_instance_);
}

dawn::wire::ReservedBuffer DawnWireServices::ReserveBuffer(
    WGPUDevice device,
    const WGPUBufferDescriptor* desc) {
  base::AutoLockMaybe lock(OptionalToPtr(lock_));
  return wire_client_.ReserveBuffer(device, desc);
}

dawn::wire::ReservedTexture DawnWireServices::ReserveTexture(
    WGPUDevice device,
    const WGPUTextureDescriptor* desc) {
  base::AutoLockMaybe lock(OptionalToPtr(lock_));
  return wire_client_.ReserveTexture(device, desc);
}

void DawnWireServices::Commit() {
  base::AutoLockMaybe lock(OptionalToPtr(lock_));
  serializer_.Commit();
}

void DawnWireServices::SetAwaitingFlush(bool awaiting_flush) {
  base::AutoLockMaybe lock(OptionalToPtr(lock_));
  serializer_.SetAwaitingFlush(awaiting_flush);
}

bool DawnWireServices::EnsureAwaitingFlush() {
  base::AutoLockMaybe lock(OptionalToPtr(lock_));
  // If there is already a flush waiting, we don't need to flush.
  // We only want to ask for a flush on state transition from
  // false -> true.
  if (serializer_.AwaitingFlush()) {
    return false;
  }

  // Set the state to waiting for flush.
  serializer_.SetAwaitingFlush(true);
  return true;
}

void DawnWireServices::FreeMappedResources(WebGPUCmdHelper* helper) {
  base::AutoLockMaybe lock(OptionalToPtr(lock_));
  memory_transfer_service_.FreeHandles(helper);
}
#endif

// Include the auto-generated part of this file. We split this because it means
// we can easily edit the non-auto generated parts right here in this file
// instead of having to edit some template or the code generator.
#include "gpu/command_buffer/client/webgpu_implementation_impl_autogen.h"

WebGPUImplementation::WebGPUImplementation(
    WebGPUCmdHelper* helper,
    TransferBufferInterface* transfer_buffer,
    GpuControl* gpu_control,
    bool support_locking)
    : ImplementationBase(helper, transfer_buffer, gpu_control),
      helper_(helper),
      main_task_runner_(base::SequencedTaskRunner::GetCurrentDefault()) {}

WebGPUImplementation::~WebGPUImplementation() {
  LoseContext();

  // Before destroying WebGPUImplementation, all mappable buffers
  // must be destroyed first. This means that all shared memory mappings are
  // detached. If they are not destroyed, MappedMemoryManager (member of
  // base class ImplementationBase) will assert on destruction that some
  // memory blocks are in use. Calling |FreeMappedResources| marks all
  // blocks that are no longer in use as free.
#if BUILDFLAG(USE_DAWN)
  if (dawn_wire_) {
    dawn_wire_->FreeMappedResources(helper_);
  }
#endif

  // Wait for commands to finish before we continue destruction.
  // WebGPUImplementation no longer owns the WebGPU transfer buffer, but still
  // owns the GPU command buffer. We should not free shared memory that the
  // GPU process is using.
  helper_->Finish();
}

void WebGPUImplementation::LoseContext() {
  lost_ = true;
#if BUILDFLAG(USE_DAWN)
  if (dawn_wire_) {
    dawn_wire_->Disconnect();
  }
#endif
}

gpu::ContextResult WebGPUImplementation::Initialize(
    const SharedMemoryLimits& limits) {
  TRACE_EVENT0("gpu", "WebGPUImplementation::Initialize");
  auto result = ImplementationBase::Initialize(limits);
  if (result != gpu::ContextResult::kSuccess) {
    return result;
  }

  std::unique_ptr<TransferBuffer> transfer_buffer =
      std::make_unique<TransferBuffer>(helper_);
  if (!transfer_buffer->Initialize(
          limits.start_transfer_buffer_size,
          /* start offset */ 0, limits.min_transfer_buffer_size,
          limits.max_transfer_buffer_size, kAlignment)) {
    return gpu::ContextResult::kFatalFailure;
  }

#if BUILDFLAG(USE_DAWN)
  dawn_wire_ = base::MakeRefCounted<DawnWireServices>(
      this, helper_, mapped_memory_.get(), std::move(transfer_buffer),
      main_task_runner_ != nullptr);
#endif

  return gpu::ContextResult::kSuccess;
}

// ContextSupport implementation.
void WebGPUImplementation::SetAggressivelyFreeResources(
    bool aggressively_free_resources) {
  NOTIMPLEMENTED();
}
void WebGPUImplementation::SetErrorMessageCallback(
    base::RepeatingCallback<void(const char*, int32_t)> callback) {
  NOTIMPLEMENTED();
}
base::span<uint8_t> WebGPUImplementation::MapTransferCacheEntry(
    uint32_t serialized_size) {
  NOTREACHED();
}
void WebGPUImplementation::UnmapAndCreateTransferCacheEntry(uint32_t type,
                                                            uint32_t id) {
  NOTREACHED();
}
bool WebGPUImplementation::ThreadsafeLockTransferCacheEntry(uint32_t type,
                                                            uint32_t id) {
  NOTREACHED();
}
void WebGPUImplementation::UnlockTransferCacheEntries(
    const std::vector<std::pair<uint32_t, uint32_t>>& entries) {
  NOTREACHED();
}
void WebGPUImplementation::DeleteTransferCacheEntry(uint32_t type,
                                                    uint32_t id) {
  NOTREACHED();
}
unsigned int WebGPUImplementation::GetTransferBufferFreeSize() const {
  NOTREACHED();
}

// InterfaceBase implementation.
void WebGPUImplementation::GenSyncTokenCHROMIUM(GLbyte* sync_token) {
  // Need to commit the commands to the GPU command buffer first for SyncToken
  // to work.
#if BUILDFLAG(USE_DAWN)
  dawn_wire_->Commit();
#endif
  ImplementationBase::GenSyncToken(sync_token);
}
void WebGPUImplementation::GenUnverifiedSyncTokenCHROMIUM(GLbyte* sync_token) {
  // Need to commit the commands to the GPU command buffer first for SyncToken
  // to work.
#if BUILDFLAG(USE_DAWN)
  dawn_wire_->Commit();
#endif
  ImplementationBase::GenUnverifiedSyncToken(sync_token);
}
void WebGPUImplementation::VerifySyncTokensCHROMIUM(GLbyte** sync_tokens,
                                                    GLsizei count) {
  ImplementationBase::VerifySyncTokens(sync_tokens, count);
}
void WebGPUImplementation::WaitSyncTokenCHROMIUM(const GLbyte* sync_token) {
  // Need to commit the commands to the GPU command buffer first for SyncToken
  // to work.
#if BUILDFLAG(USE_DAWN)
  dawn_wire_->Commit();
#endif
  ImplementationBase::WaitSyncToken(sync_token);
}
void WebGPUImplementation::ShallowFlushCHROMIUM() {
  FlushCommands();
}

// ImplementationBase implementation.
void WebGPUImplementation::IssueShallowFlush() {
  NOTIMPLEMENTED();
}

void WebGPUImplementation::SetGLError(GLenum error,
                                      const char* function_name,
                                      const char* msg) {
  GPU_CLIENT_LOG("[" << GetLogPrefix() << "] Client Synthesized Error: "
                     << gles2::GLES2Util::GetStringError(error) << ": "
                     << function_name << ": " << msg);
  NOTIMPLEMENTED();
}

// GpuControlClient implementation.
void WebGPUImplementation::OnGpuControlLostContext() {
  LoseContext();

  // This should never occur more than once.
  DCHECK(!lost_context_callback_run_);
  lost_context_callback_run_ = true;
  if (!lost_context_callback_.is_null()) {
    std::move(lost_context_callback_).Run();
  }
}
void WebGPUImplementation::OnGpuControlLostContextMaybeReentrant() {
  // If this function is called, we are guaranteed to also get a call
  // to |OnGpuControlLostContext| when the callstack unwinds. Thus, this
  // function only handles immediately setting state so that other operations
  // which occur while the callstack is unwinding are aware that the context
  // is lost.
  lost_ = true;
}
void WebGPUImplementation::OnGpuControlErrorMessage(const char* message,
                                                    int32_t id) {
  NOTIMPLEMENTED();
}
void WebGPUImplementation::OnGpuControlReturnData(
    base::span<const uint8_t> data) {
  if (lost_) {
    return;
  }

#if BUILDFLAG(USE_DAWN)
  TRACE_EVENT1(TRACE_DISABLED_BY_DEFAULT("gpu.dawn"),
               "WebGPUImplementation::OnGpuControlReturnData", "bytes",
               data.size());

  CHECK_GT(data.size(), sizeof(cmds::DawnReturnDataHeader));

  const cmds::DawnReturnDataHeader& dawnReturnDataHeader =
      *reinterpret_cast<const cmds::DawnReturnDataHeader*>(data.data());

  switch (dawnReturnDataHeader.return_data_type) {
    case DawnReturnDataType::kDawnCommands: {
      CHECK_GE(data.size(), sizeof(cmds::DawnReturnCommandsInfo));

      const cmds::DawnReturnCommandsInfo* dawn_return_commands_info =
          reinterpret_cast<const cmds::DawnReturnCommandsInfo*>(data.data());
      dawn_wire_->HandleCommands(
          *dawn_return_commands_info,
          data.size() -
              offsetof(cmds::DawnReturnCommandsInfo, deserialized_buffer));

      // Call ProcessEvents now, always posting the task to do so to the
      // runner to handle the callbacks in a fresh call stack to avoid
      // re-entrancy.
      CHECK(main_task_runner_);
      main_task_runner_->PostTask(
          FROM_HERE, base::BindOnce(&DawnWireServices::ProcessEvents,
                                    dawn_wire_->AsWeakPtr()));
    } break;

    default:
      NOTREACHED();
  }
#endif
}

void WebGPUImplementation::FlushCommands() {
#if BUILDFLAG(USE_DAWN)
  dawn_wire_->Commit();
  helper_->Flush();
#endif
}

bool WebGPUImplementation::EnsureAwaitingFlush() {
#if BUILDFLAG(USE_DAWN)
  return dawn_wire_->EnsureAwaitingFlush();
#else
  return false;
#endif
}

void WebGPUImplementation::FlushAwaitingCommands() {
#if BUILDFLAG(USE_DAWN)
  dawn_wire_->Commit();
  helper_->FlushLazy();
  dawn_wire_->SetAwaitingFlush(false);
#endif
}

scoped_refptr<APIChannel> WebGPUImplementation::GetAPIChannel() const {
#if BUILDFLAG(USE_DAWN)
  return dawn_wire_.get();
#else
  return nullptr;
#endif
}

ReservedBuffer WebGPUImplementation::ReserveBuffer(
    WGPUDevice device,
    const WGPUBufferDescriptor* optionalDesc) {
#if BUILDFLAG(USE_DAWN)
  // Commit because we need to make sure messages that free a previously used
  // buffer is seen first. ReserveBuffer may reuse an existing ID.
  dawn_wire_->Commit();

  WGPUBufferDescriptor placeholderDesc;
  if (optionalDesc == nullptr) {
    placeholderDesc = {};  // Zero initialize.
    optionalDesc = &placeholderDesc;
  }

  auto reserved = dawn_wire_->ReserveBuffer(device, optionalDesc);
  ReservedBuffer result;
  result.buffer = reserved.buffer;
  result.id = reserved.handle.id;
  result.generation = reserved.handle.generation;
  result.deviceId = reserved.deviceHandle.id;
  result.deviceGeneration = reserved.deviceHandle.generation;
  return result;
#else
  return {};
#endif
}

ReservedTexture WebGPUImplementation::ReserveTexture(
    WGPUDevice device,
    const WGPUTextureDescriptor* optionalDesc) {
#if BUILDFLAG(USE_DAWN)
  // Commit because we need to make sure messages that free a previously used
  // texture are seen first. ReserveTexture may reuse an existing ID.
  dawn_wire_->Commit();

  WGPUTextureDescriptor placeholderDesc;
  if (optionalDesc == nullptr) {
    placeholderDesc = {};  // Zero initialize.
    optionalDesc = &placeholderDesc;
  }

  auto reserved = dawn_wire_->ReserveTexture(device, optionalDesc);
  ReservedTexture result;
  result.texture = reserved.texture;
  result.id = reserved.handle.id;
  result.generation = reserved.handle.generation;
  result.deviceId = reserved.deviceHandle.id;
  result.deviceGeneration = reserved.deviceHandle.generation;
  return result;
#else
  NOTREACHED();
#endif
}

WGPUDevice WebGPUImplementation::DeprecatedEnsureDefaultDeviceSync() {
  NOTIMPLEMENTED();
  return nullptr;
}

void WebGPUImplementation::AssociateMailbox(
    GLuint device_id,
    GLuint device_generation,
    GLuint texture_id,
    GLuint texture_generation,
    uint64_t usage,
    uint64_t internal_usage,
    const WGPUTextureFormat* view_formats,
    GLuint view_format_count,
    MailboxFlags flags,
    const Mailbox& mailbox) {
#if BUILDFLAG(USE_DAWN)
  // Commit previous Dawn commands as they may manipulate texture object IDs
  // and need to be resolved prior to the AssociateMailbox command. Otherwise
  // the service side might not know, for example that the previous texture
  // using that ID has been released.
  dawn_wire_->Commit();

  // The command buffer transfer data in 4-byte "entries". So the array of data
  // we pass must have a byte-length that's a multiple of 4.
  constexpr size_t kEntrySize = 4u;
  static_assert(sizeof(mailbox.name) % kEntrySize == 0u);
  static_assert(sizeof(WGPUTextureFormat) % kEntrySize == 0u);

  size_t num_bytes =
      sizeof(mailbox.name) + sizeof(WGPUTextureFormat) * view_format_count;
  std::vector<char> immediate_data(num_bytes);

  uint32_t num_entries = ComputeNumEntries(immediate_data.size());

  UNSAFE_TODO(
      memcpy(immediate_data.data(), mailbox.name, sizeof(mailbox.name)));
  UNSAFE_TODO(memcpy(immediate_data.data() + sizeof(mailbox.name), view_formats,
                     sizeof(WGPUTextureFormat) * view_format_count));

  helper_->AssociateMailboxImmediate(
      device_id, device_generation, texture_id, texture_generation, usage,
      internal_usage, flags, view_format_count, num_entries,
      UNSAFE_TODO(reinterpret_cast<GLuint*>(immediate_data.data())));
#endif
}

void WebGPUImplementation::AssociateMailboxForBuffer(GLuint device_id,
                                                     GLuint device_generation,
                                                     GLuint buffer_id,
                                                     GLuint buffer_generation,
                                                     uint64_t usage,
                                                     const Mailbox& mailbox) {
#if BUILDFLAG(USE_DAWN)
  // Commit previous Dawn commands as they may manipulate buffer object IDs
  // and need to be resolved prior to the AssociateMailboxForBuffer command.
  // Otherwise the service side might not know, for example that the previous
  // buffer using that ID has been released.
  dawn_wire_->Commit();

  // The command buffer transfer data in 4-byte "entries". So the array of data
  // we pass must have a byte-length that's a multiple of 4.
  constexpr size_t kEntrySize = 4u;
  static_assert(sizeof(mailbox.name) % kEntrySize == 0u);

  helper_->AssociateMailboxForBufferImmediate(
      device_id, device_generation, buffer_id, buffer_generation, usage,
      reinterpret_cast<const GLuint*>(mailbox.name));
#endif
}

void WebGPUImplementation::DissociateMailbox(GLuint texture_id,
                                             GLuint texture_generation) {
#if BUILDFLAG(USE_DAWN)
  // Commit previous Dawn commands that might be rendering to the texture, prior
  // to Dissociating the shared image from that texture.
  dawn_wire_->Commit();
  helper_->DissociateMailbox(texture_id, texture_generation);
#endif
}

void WebGPUImplementation::DissociateMailboxForBuffer(
    GLuint buffer_id,
    GLuint buffer_generation) {
#if BUILDFLAG(USE_DAWN)
  // Commit previous Dawn commands that might be rendering to the buffer, prior
  // to Dissociating the shared image from that buffer.
  dawn_wire_->Commit();
  helper_->DissociateMailboxForBuffer(buffer_id, buffer_generation);
#endif
}

void WebGPUImplementation::DissociateMailboxForPresent(
    GLuint device_id,
    GLuint device_generation,
    GLuint texture_id,
    GLuint texture_generation) {
#if BUILDFLAG(USE_DAWN)
  // Commit previous Dawn commands that might be rendering to the texture, prior
  // to Dissociating the shared image from that texture.
  dawn_wire_->Commit();
  helper_->DissociateMailboxForPresent(device_id, device_generation, texture_id,
                                       texture_generation);
#endif
}

void WebGPUImplementation::SetWebGPUExecutionContextToken(uint32_t type,
                                                          uint32_t high_high,
                                                          uint32_t high_low,
                                                          uint32_t low_high,
                                                          uint32_t low_low) {
#if BUILDFLAG(USE_DAWN)
  helper_->SetWebGPUExecutionContextToken(type, high_high, high_low, low_high,
                                          low_low);
#endif
}

}  // namespace webgpu
}  // namespace gpu