Inaccurate/Wrong calculations

Inaccurate/Wrong calculations

• seu

  thanks for the response

  it's a weird one though - changing the way doubles behave without any form of huge warning is sort of dangerous! I've changed the flag and it all seems to work.

  benw

  
  

• Ashok Karmani

  Okay, I agree that over the course of seconds since 1970, a few 100ths of milliseconds may be acceptable... however, here's another one -

  
  http://www.benwalkermusic.co.uk/pics/sums2.JPG

  
  Apparently 1153221506 + 33 = 1153221504. It looks like it's wrapping around something, yet if I try with ints and then recast back to doubles, I still get 1153221504...
  

  
  

• Garrett Serack - MSFT

  It certainly looks like something is wrong there.
  
  However, can you post an entire self-contained method that exhibits the problem (i.e. one that someone else can drop into a project and try)
  
  

• gg1

  There is nothing "wrong" with the answer you are getting. Both differences are exact -- to the precision of the operands. Do the subtraction yourself on paper if you like.

  1153221555.7980 - 1153221506.7907 is exactly 49.0073. Both value you see are that value. Both include some meaningless "insignificant digits" which is an artifact to floating point math on computers, particularly when producing a small number by subtracting two large close numbers.

  Now, what exactly do these number represent If the numbers are standard DateTime value (representing days & fractions of days), you are arguing over a tenth of a second in 300,000 years!

  
  
  

• kevin delafield

  Hi,
  
  I think it's partly to do with the scale of the numbers I'm using.
  
  If I assign the difference to a new variable, I get the wrong number, however the difference in the 'Watch' window is the correct number.
  
  I'm going to try breaking it up into ints and doubles to keep the precision in both integer and decimal parts - not at the desk at the moment so will try again tomorrow...
  
  
  benw
  
  
  

• stephenl

  What do you mean by "see" Is it the output of the difference that is giving you the wrong #
  
  

• ruckycharms

  Again, what do you mean by "see". If you use something like Console.Write(someDouble.ToString()), it will be formatted to the user's # of decimal points preference (e.g. in the US that two decimal places) rounding the value.
  
  

• Tyler M

  Screenshot posted at

  http://www.benwalkermusic.co.uk/pics/sums.JPG

  FWIW I've tried converting my UTC Times back to DateTime structures and then calculating the difference, however the returned TimeSpan structure claims that the difference is zero...

  
  

• Naveeeen

  Nailed the source of the problem, but no solution: It's DirectX... Add references as neccessary

  using System;
  using System.Drawing;
  using System.Windows.Forms;
  using Microsoft.DirectX;
  using Microsoft.DirectX.Direct3D;
  using Direct3D=Microsoft.DirectX.Direct3D;

  namespace SumTest
  {
  public class Test : Form
  {
  Device device = null; // Our rendering device
  PresentParameters presentParams = new PresentParameters();
  private System.Windows.Forms.Panel panel1;
  private System.Windows.Forms.Label label1;
  private IntPtr handle;

  bool pause = false;

  
  public Test()
  {
  InitializeComponent();
  this.Text = "How not to add up";
  }

  public bool InitializeGraphics()
  {
  handle = this.panel1.Handle;

  try
  {
  presentParams.Windowed=true;
  presentParams.SwapEffect = SwapEffect.Discard;
  device = new Device(0, DeviceType.Hardware, handle, CreateFlags.SoftwareVertexProcessing, presentParams);
  pause = false;
  return true;
  }
  catch (DirectXException)
  {
  return false;
  }
  }
  
  private void Render()
  {
  label1.Text = CalcNumber().ToString("0.000000000");
  
  if (device == null)
  return;

  device.Clear(ClearFlags.Target, System.Drawing.Color.White, 1.0f, 0);
  device.BeginScene();
  device.EndScene();
  device.Present(handle);

  
  }

  private double CalcNumber()
  {
  double d1 = 1153221506.123456;
  double d2 = 32.9565461;
  return d1+d2;
  }

  protected override void OnPaint(System.Windows.Forms.PaintEventArgs e)
  {
  this.Render(); // Render on painting
  }
  protected override void OnKeyPress(System.Windows.Forms.KeyPressEventArgs e)
  {
  if ((int)(byte)e.KeyChar == (int)System.Windows.Forms.Keys.Escape)
  this.Close(); // Esc was pressed
  }
  protected override void OnResize(System.EventArgs e)
  {
  pause = ((this.WindowState == FormWindowState.Minimized) || !this.Visible);
  }

  private void InitializeComponent()
  {
  this.panel1 = new System.Windows.Forms.Panel();
  this.label1 = new System.Windows.Forms.Label();
  this.SuspendLayout();
  //
  // panel1
  //
  this.panel1.Location = new System.Drawing.Point(8, 8);
  this.panel1.Name = "panel1";
  this.panel1.Size = new System.Drawing.Size(416, 392);
  this.panel1.TabIndex = 0;
  //
  // label1
  //
  this.label1.Location = new System.Drawing.Point(40, 424);
  this.label1.Name = "label1";
  this.label1.Size = new System.Drawing.Size(368, 23);
  this.label1.TabIndex = 1;
  this.label1.Text = "label1";
  //
  // Matrices
  //
  this.AutoScaleBaseSize = new System.Drawing.Size(5, 13);
  this.ClientSize = new System.Drawing.Size(432, 469);
  this.Controls.Add(this.label1);
  this.Controls.Add(this.panel1);
  this.Name = "Matrices";
  this.ResumeLayout(false);

  }

  /// <summary>
  /// The main entry point for the application.
  /// </summary>
  static void Main()
  {
  using (Test frm = new Test())
  {
  frm.Show();
  
  if (!frm.InitializeGraphics()) // Initialize Direct3D
  {
  MessageBox.Show("Could not initialize Direct3D");
  return;
  }
  
  // While the form is still valid, render and process messages
  while(frm.Created)
  {
  frm.Render();
  Application.DoEvents();
  }
  }
  }
  }
  }
  

  If you comment out the Graphics Initialisation code, the answer is correct. Surely there must be other people using DirectX and doubles that have had this problem I'm reposting this on the DX forum too, so hopefully someone will have some ideas...

  
  

• William Lowers

  See:
  I'm running in from within the debugger - I copy the variables/expressions into the watch window below to check what's going on. As the values are assigned, the value appears at the side of the variable/expression name.
  
  There is no Console.ToString or similar that I call directly - it's whatever the watch window uses for it's formatting (afaik 16 decimal places). I'm simply trying to debug some code that does a quick linear interpolation between two UTC timestamps in double format. The problem arises when I try to calculate the difference in the two timestamps - wherby the answer never seems to be 'correct'. I'll try and post a screenshot tomorrow from the desk.
  
  
  
  

• Suganya Mahadevan

  Check out this article for more detail.
  
  Essentially, when creating a DirectX object, you should use the CreateFlags.FpuPreserve flag to keep the CLR's double precision.
  
  
  
• Inaccurate/Wrong calculations