D3D11触发Fence后Map调用仍缓慢的问题排查

现状概述 #

运行计算着色器，通过D3D11 Query统计得着色器本身耗时约0.22ms，但实际总耗时平均约0.6ms，Map调用占用了部分时间。着色器负责计算自上一帧以来所有变更像素的dirty RECT并返回CPU，功能运行正常。

核心问题：正常场景下总耗时约0.6ms，性能理想；但程序后台运行且存在GPU密集型应用（如Blender、3A游戏）时，总耗时从0.6ms飙升至9ms，Map调用速度异常缓慢，需要定位原因。

当前实现代码细节 #

结果缓冲区创建 #

buffer_desc.ByteWidth = sizeof(RECT);
buffer_desc.StructureByteStride = sizeof(RECT);
buffer_desc.Usage = D3D11_USAGE_DEFAULT;
buffer_desc.BindFlags = D3D11_BIND_UNORDERED_ACCESS;
buffer_desc.MiscFlags = D3D11_RESOURCE_MISC_BUFFER_STRUCTURED;
buffer_desc.CPUAccessFlags = 0;
hr = ID3D11Device5_CreateBuffer(ctx->device, &buffer_desc, NULL, &ctx->reduction_final);

UAV设置 #

uav_desc.Format = DXGI_FORMAT_UNKNOWN;
uav_desc.ViewDimension = D3D11_UAV_DIMENSION_BUFFER;
uav_desc.Buffer.FirstElement = 0;
uav_desc.Buffer.NumElements = 1;
uav_desc.Buffer.Flags = 0;
hr = ID3D11Device5_CreateUnorderedAccessView(ctx->device, (ID3D11Resource*) ctx->reduction_final, &uav_desc, &ctx->reduction_final_UAV);

暂存缓冲区创建 #

buffer_desc.ByteWidth = sizeof(RECT);
buffer_desc.StructureByteStride = sizeof(RECT); 
buffer_desc.Usage = D3D11_USAGE_STAGING;
buffer_desc.BindFlags = 0;
buffer_desc.MiscFlags = D3D11_RESOURCE_MISC_BUFFER_STRUCTURED;
buffer_desc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
hr = ID3D11Device5_CreateBuffer(ctx->device, &buffer_desc, NULL, &ctx->cpu_staging);

Map调用实现 #

hr = ID3D11DeviceContext4_Map(ctx->device_ctx, (ID3D11Resource*) ctx->cpu_staging, 0, D3D11_MAP_READ, 0, &mapped_res);

Fence与事件创建 #

hr = ID3D11Device5_CreateFence(ctx->device, 1, D3D11_FENCE_FLAG_NONE, &IID_ID3D11Fence, (void**) &ctx->fence);
ctx->fence_handle = CreateEventA(NULL, FALSE, FALSE, NULL);

Dispatch后的复制与Fence信号 #

ID3D11DeviceContext4_CopyResource(ctx->device_ctx, (ID3D11Resource*) ctx->cpu_staging, (ID3D11Resource*) ctx->reduction_final);
ID3D11DeviceContext4_Signal(ctx->device_ctx, ctx->fence, ++ctx->fence_counter);
hr = ID3D11Fence_SetEventOnCompletion(ctx->fence, ctx->fence_counter, ctx->fence_handle);

等待与Map读取逻辑 #

DWORD waitResult = WaitForSingleObject(ctx->fence_handle, 1);
if (waitResult != WAIT_OBJECT_0)
{
    LOG("Timeout Count: %u", ++timeout_count);
    return false;
}
else
{
    LOG("Timeout Count: %u", timeout_count);

    ID3D11DeviceContext4_Map(ctx->device_ctx, (ID3D11Resource*) ctx->cpu_staging, 0, D3D11_MAP_READ, 0, &mapped_res);
    memcpy(rect, mapped_res.pData, sizeof(RECT));
    ID3D11DeviceContext4_Unmap(ctx->device_ctx, (ID3D11Resource*) ctx->cpu_staging, 0);
    return true;
}

核心疑问 #

1. 对Map调用的工作原理存在哪些误解？
2. 有没有办法缩短Map调用的耗时？
3. Fence是否可能在复制操作完成前就触发信号？

已尝试的优化方案 #

• 设置SetGPUThreadPriority为7
• 设置AvSetMmThreadCharacteristicsW为"DisplayPostProcessing"、"Games"和"Capture"
• 设置AvSetMmThreadPriority为AVRT_PRIORITY_CRITICAL

内容的提问来源于stack exchange，提问作者vulkur