Specifying the Transformation¶

You must supply exactly one of the following options:

pamflip's predecessor (before Netpbm 10.7 - August 2002) pnmflip did not have the -xform option and instead allowed you to specify any number of the other options, including zero. It applied all the indicated transformations, in the order given, just like with pamflip's -xform option. (Reason for the change: this kind of interpretation of options is inconsistent with the rest of Netpbm and most of the Unix world, and thus hard to understand and to implement).

-leftright

-lr

Flip left for right.

-topbottom

-tb

Flip top for bottom.

-transpose

-xy

Transpose horizontal for vertical. I.e. make the pixel at (x,y) be at (y,x).

-rotate90

-r90

-ccw

Rotate counterclockwise 90 degrees.

-rotate180

-r180

Rotate 180 degrees.

-rotate270

-r270

-cw

Rotate counterclockwise 270 degrees (clockwise 90 degrees)

-null

No change. (The purpose of this option is the convenience of programs that invoke pamflip after computing the kind of transformation desired, including none at all).

This option was new in Netpbm 10.13 (December 2002).

-xform=xform1,xform2...

Apply all the transforms listed, in order. The valid values for the transforms are as follows and have the same meanings as the identically named options above.

• leftright
• topbottom
• transpose

  This option was new in Netpbm 10.13 (December 2002).

Memory Management¶

The following options help pamflip use memory efficiently. Some flipping operations on very large images can cause pamflip to have a very large working set, which means if you don't have enough real memory, the program can page thrash, which means it takes a ridiculous amount time to run. If your entire image fits in real memory, you don't have a problem. If you're just flipping top for bottom or left for right, you don't have a problem. Otherwise, pay attention. If you're interested in the details of the thrashing problem and how pamflip approaches it, you're invited to read a complete explanation in comments in the source code.

-memsize=mebibytes

mebibytes is the size in mebibytes (aka megabytes) of memory available for pamflip. It is the lesser of the amount of real or virtual memory available.

pamflip does nothing special to allocate real memory or control it's allocation -- it gets whatever it gets just by referencing virtual memory normally. The real memory figure in question is the maximum amount that pamflip can be expected to end up with by doing that. This is just about impossible for you to know, of course, but you can estimate. The total real memory in your system should be a major factor in your estimate.

If pamflip cannot fit the entire image in the amount of memory you specify, it does the transformation in chunks, using temporary files for intermediate results.

In strict horizontal transformations (left for right) or the null transformation, pamflip never keeps more than one row in memory, so the memory size is irrelevant and pamflip doesn't use temporary files.

In strict vertical transformations (top for bottom), even when you allow pamflip enough memory to keep the entire image in virtual memory, it accesses it in a single pass, which does not cause any thrashing; the chunks in temporary files thus don't help.

The real memory is important when you do a column for row type of transformation (e.g. -rotate90). In that case, even if pamflip can fit the entire image in virtual memory at once, if it does not also fit in real memory, the program will thrash like crazy because of the order in which pamflip accesses the pixels, and that means it will take a ridiculously long time to run. A proper -memsize drastically reduces the paging.

If you specify -memsize too large, pamflip may attempt to get more virtual memory than the system allows it and fail. If it can get the virtual memory, but -memsize is larger than the amount of real memory the system allows it and the transformation is row for column, it will page thrash and run very slowly. A value even slightly too high is the same as infinity.

If you specify -memsize too small, the program will run slightly more slowly because of extra overhead in manipulating temporary files. Also, if your environment isn't set up to make temporary files possible, pamflip will fail.

Doing the entire transformation "in memory" doesn't speed things up as much as you might think, because even with the temporary files, the data is just as likely to be in memory. Virtual memory gets paged to disk and disk files get cached in memory. In fact, the pixels fit much more compactly into memory when stored in a temporary file than when stored "in memory" because pamflip uses a more efficient format. So you're likely to have less disk I/O when you allow pamflip less memory.

If you do not specify -memsize, pamflip assumes infinity.

This option did not exist before Netpbm 10.7 (August 2002).

Before Netpbm 10.42 (March 2008), this option applied only to real memory. pamflip would always keep the entire image in virtual memory and if it could not get enough virtual memory, it failed. pamflip accessed the pixels in an order designed to keep real memory use within the specified amount.

-pagesize=bytes

bytes is the size in bytes of a paging unit -- the amount of memory that gets paged in or out as an indivisible unit -- in your system. The default is 4KiB.

This option has no effect.

Before Netpbm 10.42 (March 2008), pamflip used it to control its use of real memory.

This option did not exist before Netpbm 10.7 (August 2002).

Miscellaneous¶

-verbose

This option causes pamflip to issue messages to Standard Error about its progress.