Sender: Leavy (☆ 星 ☆ ☆), letter area: Fortran Title: [Collection] Help About the large number of letter station: Nanjing University Small Lily Station (Tue Sep 7 21:38:48 2004) Vapour2004 (Vapour) in THU SEP 2 13:04:04 2004) Next: How to achieve more than 2GB of PGF90 in Linux86-64? It is easy to use under the PGCC, as long as INCLUDE STDLIB.H is used in Malloc. How to do it? I am in a hurry ... I greeted Leavy (☆ ☆ ☆) in THU Sep 2 13:20:48 2004) mentioned: I have encountered such a difficulty, one array 1G more It can't be opened, and then split in the program, I can't study the PGI manual, there may be TSY (TSY) in Thu Sep 2 16:32:29 2004) mentioned: I have never encountered such a big thing. I can't I don't have Hoho, why don't you use C, then use Fortran calls? TSY (TSY) is mentioned in THU Sep 2 16:52:01 2004) mentioned: I remember that I have seen this saying: machine configuration is 512M memory, virtual memory 1G (Windows), exchange zone 512M (Linux) other What is forgotten, it is like P4 1.7 ~~ 1, Windows uses F90, array is allocatable, integer and real * 4 can only reach 256m (61 * 1024 * 1024), the double precision can only reach 31 * 1024 * 1024, if it is big, although the compilation is not wrong, it will display the error related to virtual memory, even if your virtual memory is reserved, the static array can be opened to 1.6G, but compiling will give a warning, However, the result is correct, as if Windows can only use 256M memory 2, Linux can be opened to 708M and then make mistakes. For static arrays, only real memory space can be used, and the switching area is not used. But then I don't know, I haven't used it so much, so I haven't tried vapour2004 (vapour) in THU Sep 2 17:52:36 2004) mentioned: First of all I use the machine is 2 x Opteron246 The memory is 16GB, so don't discuss the problem of switching space. Secondly, I used to write a program to use C. Fortran is not good. Now I must use Fortran, not, nothing, just want to skilled Fortran. Of course, it is good in C, of course, if not include stdlib.h, use dynamic memory Malloc to compile, there is no problem when running in the decoder, but it is not greater than 2GB (note unit). I would like to have a similar situation in Fortran, I am trying to use Fortran's dynamic memory allocation Malloc. I want to have one of the headers or libraries to break through 2GB restrictions, just PGI's bookscript book. Not clear, but the PGI does not need to directly apply for 2GB or more array without dynamic memory allocation, but I don't satisfy such a program. If Fortran is so important and simple, the problem can not be solved, I will see it, I'm just dead. . Because Fortran has always been called in numerical calculation, the quantum mechanics calculation medium wave function is often very large, and the most recent mission has such a calculation, with a total memory amount of about 12GB, one of which is about 8GB, this time I just wanted to use Fortran, no one persuade me to use any other law. Also I know that opneron is 40 or 48 addressing, this problem does not have to be discussed.
Leavy (☆ 星 ☆ ☆) in THU Sep 2 18:05:43 2004) mentioned: Opteron should not limit the memory? Is there a unlimit? Compilation can pass smoothly? I used 2G memory P4 microcomputer, some array didn't seem to exceed 900M (possibly deviation, unknown) Leavy (☆ 星 ☆ ☆) in Thu Sep 2 18:07:08 2004) mention: I am in Linux The next single static array must not exceed 900m, total memory 2gleavy (☆ 星 河 ☆) in THU Sep 2 18:10:08 2004) mentioned: Are you compiled with those options? Time We have Opteron test, but unfortunately I didn't try the big memory recommendation to high performance calculate the HPC version to ask, there are many big cows: Pyogicy (I step on the moon) in THU Sep 2 18: 19:51 2004) He mentioned that 64-bit should not have 2G limitations, even static allocation, it should be sufficient. Leavy (☆ ☆) in Thu Sep 2 18:35:14 2004) mentioned: I remember to unlimit on SGI, I have to add -64yogicy (I step on the moon color) at THU SEP 2 18: 53:48 2004) He mentioned: Yes, compiled with 64. However, it seems that the static large array is compiled with PGF90 64 on Opteron, but I don't know why. But if you use dynamic assignments. Vapour2004 (Vapour) is mentioned in Fri Sep 3 08:05:40 2004). Thank you for your discussion. Regarding the arrays in C, I will mention it, each returned pointer is 40, and more than 40 is all NIL. Therefore, AMD64 is not a real 64-bit machine (mainly referred to), even if the official announcement technology details refer to the address of 48, and then there is no longer there. There is a real 64-bit machine. Here is a suggestion here, you should work hard to solve the problem, so the discussion of the cloud is also the discussion, such as compilation parameters, etc., the manual is very clear, and there is no practical experience, and The bottom is not very solid, should not say that 64 bits should have no 2GB restrictions, because the operation of the program is not only closely related to the hardware and the operating system programming environment. Regarding the issue of Limit, each programmer should be viewed directly with the limit command before programming in the Linux environment, so that the necessary modifications. I said that these words are absolutely good, just hope this edition is more readable. Thank you again for your enthusiastic heroes. Yogicy (I am coming to the moon) in Fri Sep 3 10:33:19 2004) mentioned: Anyway, I am on the opnerocate on a 3.4G array of array Leavy (☆ 星 ☆ ☆) in Fri Sep 3 11:44: 31 2004) He mentioned: I didn't really use Opteron. I didn't expect to test the big memory vapour2004 (vapour) on Sat Sepou 4 06:19:16 2004) mentioned: I really want to know how you applaid you. What is the large group? Is PGI5.x? Can you tell us, which header files, specific instances.
