c++ - strange results with /fp:fast -

-

We have some code that looks like this: 

  Inline ant calms_group ( Double X) {If (x> 0.0) {// Some Return 1; } Else {// return something else 0; Unfortunately, when using Flag  / fp: fast , we get the  calc_something (0) == 1  so we are clearly taking the wrong code path This is only when we use the method on many points in our code with different standards, so I think there is some inflammatory customization here from the compiler (Microsoft Visual Studio 2008, SP1). 
 
 In addition, the above problem is solved when we use the interface 
 
  inline inc calc_something (const double & x) {  P>) but I do not know whether this fix is ​​strange behavior can anyone explain this behavior? If I do not understand what is happening then we have to delete the  / fp: fast  switch, but it will slow down our application very much.

y = x + 20; sit in that comparison. Y = y - 20; y is already on the FP stack, so instead of load x the compiler compares against just y . But due to rounding errors, y is not enough 0.0 as it is supposed to be, and you get strange results. A better explanation for

: C ++ FAQ from Light. This is part of what I am trying to achieve, I just do not remember, as far as I've read it yet.

Changing it in a reference context is fine because the compiler is concerned about aliasing. It forces a load from x because it can not assume that its value has not changed at any point after making it y , and since X is exactly the same as 0.0 [which is worthy of representation in every floating point format], I'm familiar with) spherical errors disappear.

I'm pretty sure that MS provides you a program that lets you set up FP flags based on a per-function. Or you can move this routine to a different file and give custom flags of that file. Either way, this can prevent your entire program from being victimized just to keep a regular happy.


Comments

Post a Comment

Popular posts from this blog

windows - Heroku throws SQLITE3 Read only exception -

-
After I deploy the app to Heroku, I have to run migration scripts and get this error message ... ITES \ padrino \ prophetmargin & gt; Heroku Rack AR: Rack mutated! SQLite3 :: ReadOnlyException: only one try to write database to read: CREATE TABLE "schema_migrations" ( "version" varchar (255) NOT NULL) /disk1/home/slugs/215264_925fd2c_65a3/mnt/.bundle/gems/gems/ padrino-core- is 0.9.11 / lib / padrino-core / cli / rake.rb: 9: in `init ' how can this be? I also tried to run Heroku dbpush SQLite: //db/my-db.db and that did not work Horoku sqlite3 but using postgres does not do. I'm not sure why you are receiving this error, however, as I have used sqlite3 in development and when pushing it into its travel, they do some magic which migrates to postgrads. I'm absolutely sure how Heroku does 'backwash' this database but it seems that this is happening to you because it's the SQLite db file which is obviously due to Heroku's
Read more

lex - Building a lexical Analyzer in Java -

-
I am currently learning lexical analysis in compiler design. To know how a lexical analyzer actually works, I am trying to build myself. I am planning to make it in Java The input to the lexical analyzer is a .tex file that is of the following format. \ section {script} {intro} \ section {Scope} arbitrary text \ section {relevance} uncontrolled text \ subdivision (profit) arbitrary text \ subsubsection {in real life} \ subdivision {Ingredient} \ end {document} The output of the lecture may be a table of contents with a page number in another file. 1 Introduction 1 1.1 Scope 1 1.2 Relevance 2 1.2.1 Benefits 2 1.2.1.1 Real life 2 1.2.2 loss 3 I hope that This problem is within the scope of Language Analysis Read the .exe file and check '\' and continue to check whether it is actually on search In the sectioning command is set to indicate whether or not a flag variable is the type of sectioning. I hope the above approach will work for the construction of Lexa
Read more

python - rename keys in a dictionary -

-
I want to change the names of keys in a dictionary which are ints, and I need them so that the inputs with key zero They can sort correctly. For example, my keys are like this: '1', '101', '11' and I may need them: '001', '101', '011' What is doing, but I know that there is a better way tmpDict = {} in the ADC for the old: t MpDict ['% 04d'% int (olden)] = addict [oldkey] New controlled = tmpDict You are going wrong about this if you want to draw entries on a line in a specific way, then you have to sort on the extraction. for sorted (D, key = int): print '% s:% R'% (k, d [k]) < / Div>
Read more