copy region scatter bench

copy-region-scatter.cpp

#pragma comment(lib, "d3d11.lib")
#include <d3d11.h>
#include <stdio.h>
#include <windows.h>
#include <assert.h>
#define WIDTH 8192
#define HEIGHT 8192

int buf[WIDTH * HEIGHT];
int buf2[WIDTH * HEIGHT];

void delupms(void *b, size_t length)
{
  int sum = 0;
  char *buf = (char *)b;
  char *end = buf + length;
  while (buf < end)
  {
    sum += *buf;
    buf += 0x1000;
  }
  printf("sum: %d\n", sum);
}

double ticksToMS;
int main(int argc, char **argv)
{
  int size;
  ID3D11Device *device;
  ID3D11DeviceContext *context;
  D3D_FEATURE_LEVEL level;
  int creation_flags = 0; // D3D11_CREATE_DEVICE_DEBUG;
  D3D11CreateDevice(NULL, D3D_DRIVER_TYPE_HARDWARE, NULL,
                    creation_flags,
                    NULL, 0, D3D11_SDK_VERSION, &device,
                    &level, &context);
  printf("%p\n", device);
  if (argc > 1)
    size = strtol(argv[1], 0, NULL);
  else
    size = D3D11_REQ_TEXTURE2D_U_OR_V_DIMENSION;
  int size_low = 0;
  int trial_size;
  HRESULT result;
  size_t byte_size = sizeof(buf);

  LARGE_INTEGER freq;
  QueryPerformanceFrequency(&freq);
  printf("frequency %lld\n", freq.QuadPart);
  ticksToMS = 1000. / freq.QuadPart;
  trial_size = size;
  result = 0x800000;

  D3D11_TEXTURE2D_DESC desc;
  desc.Width = WIDTH;
  desc.Height = HEIGHT;
  desc.MipLevels = 1;
  desc.ArraySize = 1;
  desc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
  desc.SampleDesc.Count = 1;
  desc.SampleDesc.Quality = 0;
  desc.Usage = D3D11_USAGE_DEFAULT;
  desc.CPUAccessFlags = 0;
  desc.MiscFlags = 0;
  desc.BindFlags = D3D11_BIND_SHADER_RESOURCE;

  {
    LARGE_INTEGER before;
    LARGE_INTEGER after;
    LONGLONG time;

    ID3D11Texture2D *src_texture;
    D3D11_TEXTURE2D_DESC small_desc = desc;
    //small_desc.Height = 1;
    result = device->CreateTexture2D(&small_desc, NULL, &src_texture);
    assert(result == S_OK);
    QueryPerformanceCounter(&before);
    D3D11_BOX box;
    box.left = 0;
    box.right = WIDTH;
    box.top = 0;
    box.bottom = 1;
    box.front = 0;
    box.back = 1;
    printf("before updatesubresource time\n");
    context->UpdateSubresource(src_texture, 0, &box, buf, WIDTH * 4, sizeof(buf));
    QueryPerformanceCounter(&after);
    time = after.QuadPart - before.QuadPart;
    printf("%x updatesubresource time: %.1fms\n", result, time * ticksToMS);

    ID3D11Texture2D *dst_texture;
    result = device->CreateTexture2D(&desc, NULL, &dst_texture);

    while (true) {
    D3D11_QUERY_DESC query_desc = {};
    query_desc.Query = D3D11_QUERY_TIMESTAMP_DISJOINT;
    ID3D11Query *query;
    device->CreateQuery(&query_desc, &query);

    D3D11_QUERY_DESC time_query_desc = {};
    time_query_desc.Query = D3D11_QUERY_TIMESTAMP;
    ID3D11Query *time_before_query;
    device->CreateQuery(&time_query_desc, &time_before_query);
    ID3D11Query *time_after_query;
    device->CreateQuery(&time_query_desc, &time_after_query);

    context->Begin(query);
    context->Begin(time_before_query);
    context->End(time_before_query);

    for (int y = 0; y < HEIGHT; y++) {
      for (int x = 0; x < WIDTH; x++) {
        D3D11_BOX box;
        box.left = 0;
        box.right = 1;
        box.top = 0;
        box.bottom = 1;
        box.bottom = 1;
        box.front = 0;
        box.back = 1;
        context->CopySubresourceRegion(dst_texture, x, y, 0, 0, src_texture, 0, &box);
      }
    }
    context->Begin(time_after_query);
    context->End(time_after_query);
    context->End(query);

    D3D11_QUERY_DATA_TIMESTAMP_DISJOINT data;
    do
    {
      result = context->GetData(query, &data, sizeof(data), 0);
    } while (result != 0);
    UINT64 time_before;
    UINT64 time_after;
    do
    {
      result = context->GetData(time_after_query, &time_after, sizeof(time_after), 0);
    } while (result != 0);
    QueryPerformanceCounter(&after);
    do
    {
      result = context->GetData(time_before_query, &time_before, sizeof(time_before), 0);
    } while (result != 0);
    assert(!data.Disjoint);
    printf("copy time: %fms\n", (time_after - time_before) * 1000. / data.Frequency);
      time_before_query->Release();
  time_after_query->Release();
  query->Release();
  }

  }
}