Fixes issue where Ki,Mi,Gi,Ti,Pi behave like K,KB etc instead of KiB.

[RobertTheBrucey]

Fixes issue where Ki,Mi,Gi,Ti,Pi behave like K,KB etc instead of KiB.

Fixes #1618
As described in issue 1618, the Ki,Mi,etc units do not behave as expected. The code in parse.c explicitly handles K,KB and KiB, but not Ki, this results in Ki being implicitly handled by the "K" check. This commit adds explicit handling of Ki,Mi, etc to match the behavior of KiB, MiB, etc.
This change also brings the behaviour in line with the documentation, though the examples have incorrect numbers for TB and have duplicated entries between kb_base=1000 and kb_base=1024.

Sean Ashton [email protected]

[RobertTheBrucey]

Tested with this jobfile:

[global]
size=2M
ioengine=posixaio

[job-M]
bs=1M

[job-MB]
bs=1MB

[job-Mi]
bs=1Mi

[job-MiB]
bs=1MiB

[job-M-1000]
bs=1M
kb_base=1000

[job-MB-1000]
bs=1MB
kb_base=1000

[job-Mi-1000]
bs=1Mi
kb_base=1000

[job-MiB-1000]
bs=1MiB
kb_base=1000

With results:

job-M: (g=0): rw=read, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=posixaio, iodepth=1
job-MB: (g=0): rw=read, bs=(R) 1024KiB-1024KiB, (W) 1024KiB-1024KiB, (T) 1024KiB-1024KiB, ioengine=posixaio, iodepth=1
job-Mi: (g=0): rw=read, bs=(R) 977KiB-977KiB, (W) 977KiB-977KiB, (T) 977KiB-977KiB, ioengine=posixaio, iodepth=1
job-MiB: (g=0): rw=read, bs=(R) 977KiB-977KiB, (W) 977KiB-977KiB, (T) 977KiB-977KiB, ioengine=posixaio, iodepth=1
job-M-1000: (g=0): rw=read, bs=(R) 1000kB-1000kB, (W) 1000kB-1000kB, (T) 1000kB-1000kB, ioengine=posixaio, iodepth=1
job-MB-1000: (g=0): rw=read, bs=(R) 1000kB-1000kB, (W) 1000kB-1000kB, (T) 1000kB-1000kB, ioengine=posixaio, iodepth=1
job-Mi-1000: (g=0): rw=read, bs=(R) 1049kB-1049kB, (W) 1049kB-1049kB, (T) 1049kB-1049kB, ioengine=posixaio, iodepth=1
job-MiB-1000: (g=0): rw=read, bs=(R) 1049kB-1049kB, (W) 1049kB-1049kB, (T) 1049kB-1049kB, ioengine=posixaio, iodepth=1