Re: Power PC and EPICS

["Noboru Yamamoto" <noboru.yamamoto@kek.jp>]

> 
> Dear All,
> 
> We are considering upgrading to a Power PC based CPU board.
> I would be interested to hear of users experiences with
> EPICS on any Power PC based board.
> 
We uses Force PowerCore 603 for development of KEKB control system.
In a few weeks, we will get Force PowerCore6750, which uses  PPC-750
as a CPU.
EPICS R3.13beta11 runs on PCore603. 
So far ,  We have tested  limited number  of dirvers:
	drvGpib.c
	Camac support and  camacLib with Hytec 2992 Serial Driver
	drvSerial/devSerial
	Modbus+ suppport
	Arcnet driver

Mr. Jun-ich Odagiri is the person who ported EPICS to Froce PPC board
in KEK. He wrote a memo on the port of EPICS to PPC in Japanse ( sorry).
Here I briefly summarize his memo. There are two categories in the 
modifications. One related to the difference between compilers 
for 68K and PPC. Another is CPU board specific.

1) Compiler differences.
 1-1) signed or unsigned integral variables.
 According to our study, the compiler for MC68K and the compiler for PPC
in VxWorks have different default types for integral variables.

		char		|	short	|	int		|	long
68k	|	signed	|	signed	|	unsigned	|	unsigned
PPC	|	unsigned |	unsigned |	signed	|	signed

In some cases, this difference causes different behaviour between
two platforms. You must use "signed" or "unsigned"  keywords explicitly.

For "char" type, compiler option "-fsigned-char" can be used.

  1-2) "_" in the global name.
 EPICS uses global symbol table(i.e. it uses symFindByName() function)
 to find device/record support routines.
68K compiler adds "_" in a front of the global symbol name, however,
PPC compiler DOES NOT.
Some EPICS routines assume that the global name have "_" at the front of 
symbol name. You need to fix this in the following files.

 bdt/bdtServ.c
 db/devLib.c
 db/drvTS.c
 db/iocInit.c
 dbStatic/dbStaticRun.c
 dbTools/PVSvx.c
 dev/devAiSymb.c
 dev/devAoSymb.c
 dev/devLiSymb.c
 dev/devLoSymb.c
 dev/devSiSymb.c
 dev/old/drvGpib.c ( for Hideos support)
 rec/gsubRecord.c
 rec/subRecord.c
 drv/ansi/drvEpvxi.c
 libvxWorks/ldpp.c

 1-3) -mnobitfield option
 A compiler option "-mnobitfield" does not exists for PPC compiler.

 1-4) va_list in errPrintfVX.c.

 In 68K compiler, va_list is "typedef"-ed as pointer type, while PPC 
compiler "typedef" va_list as an array type. There is a code segment
in src/libCoom/errPrintfVX.c which assgined "va_list" value 
to a "va_list" variable.
 In 68K compiler, it is a simple pointer assignment. In PPC compiler,
it turns to be an array assginment which is not allowed in the 
current compiler.

2) Board specific problems .
　2-1) vxMemProbe() does not work on Force PPC board.  Because of
this, some device/driver support routine　fail to find VME modules.
 We currently uses  workaround to fix this problem. That is to supply
numbers of VME modules in a startup script, like:

 nigpib_num_links=2
 ht2992_num_cards=1

  2-2)Slave A24 access to a main memory.
  NI-1014, GP-IB interface board for VME from Natinal Instruments, uses 
DMA to exchange data between the board and a cpu board. It uses slave A24
access to the main memory on the CPU board. Default settiing of BSP for
Force CPU disables slave A24 access. 

 2-3) Word bundary.
 GP-IB driver for NI-1014 uses data structure cc_ary shown below:

 struct cc_ary{
	void *cc_ccb;
 	short cc_ONE;
	void *cc_n_1addr;
	short cc_TWO;
}

This structure relflects structure of registers on NI-1014 board and 
should not be changed. However, compiler for Power PC add pad bytes between
cc_ccb and cc_ONE. 

We have modified the structure as follows:

typedef struct DmaStuffStruct {
	char cc_array[12];
	char cc_byte;
}
 
>
> In particular, did anyone make use of the following drivers and
> did they have to make any code changes to get them working:
> 
> "drvBc635"   - Bancomm driver - not likely to have been used by anyone
> outside of this
>                project.
> "drvXy240"   - TTL Card.
> "drvXy566"   - Analogue Input.
> "drvOms"     - Motor controller.
> "drvTip810"  - CanBus.
> "drvAb"      - Allen-Bradley.
> "drvPmac"    - PMAC Motion Controller.
Sorry, we use NONE of these driver in our system and don't have chance to
test them.

> 
> Many thanks for your time,
> 
> Andy
> 
> ===================================================================
> Andy Foster
> Gemini 8-m Telescopes Project	Tel:       1 (808) 933 7106 (direct)
> Suite 207			Fax:       1 (808) 933 7123
> Hilo Plaza			E-mail:    afoster@gemini.edu or
> 180 Kinoole Street                         ajf@noao.edu
> Hilo
> Hawaii 96720
> USA
> 

Best regards,
Noboru Yamamoto