Power 6/32

3D rendering by a Power 6/32 in Young Sherlock Holmes

The Power 6/32, code named Tahoe, from Computer Consoles Incorporated, was a supermini; it used the VERSAbus as its bus. It is most noted from the CSRG porting 4.3 BSD to it, in the 4.3 BSD Tahoe release.

Although it was not very successful commercially (in part because CCI exited the computer business shortly after it was released), it was very influential as the first non-VAX machine that BSD UNIX was ported to. The Tahoe did help separate out a lot of VAX-specific code from the base, allowing BSD to become more portable, but the Power 6/32 machine quickly disappeared off the market, and not much is known about them.

The CPU was re-sold (under other names) in systems from several vendors (viewed by CCI as OEMs): Unisys sold the 6/32 under the name 'Unisys 7000/40', using their own bastardized BSD+SYSV Unix. (Kuwait Petroleum in Denmark had one of them.) ICL had the 'Clan 7'. Harris had the HCX-5, -7, and -9 models (for which a small amount of information survives - below), running the SysV derivative HCX/UX.

Pixar used a Power 6/32 to render the stained-glass knight in the 1985 movie Young Sherlock Holmes.

Details

The Power 6/32 CPU was implemented in TTL; it had a clock rate of 100nS.

The performance is given in this table, from dhrystone.c (below):

 *----------------DHRYSTONE VERSION 1.0 RESULTS BEGIN--------------------------
 *
 * MACHINE      MICROPROCESSOR  OPERATING       COMPILER        DHRYSTONES/SEC.
 * TYPE                         SYSTEM                          NO REG  REGS
 * --------------------------   ------------    -----------     ---------------
 * CCI POWER 6/32               COS(SV+4.2)     cc              7500    7800
 * CCI POWER 6/32               POWER 6 UNIX/V  cc              8236    8498
 * CCI POWER 6/32               4.2 Rel. 1.2b   cc              8963    9544
 * VAX 11/780   -               UNIX 5.2        cc              1515    1562
 * VAX 11/780   -               UNIX 4.3bsd     cc              1646    1662

Which may give some indication on the initial reasons why the Power 6/32 was chosen as the successor to the VAX by CSRG.

Instruction Set

Although apparently no documentation survives and is accessible, we can infer a lot from the instruction decoding/printing routines in the debuggers from 4.3 BSD Tahoe, and their input file 'instrs' (below).

• Instructions start with a 1-byte opcode, and, depending on the opcode and 'addressing mode' (below), are followed by a basic set of 0-3 operands. (Two instructions, 'ediv' and 'emul', take 4 operands; if any operand uses the 'indexed' addressing mode, an extra operand is required for each such use.)
• Operands start with 1 byte, containing 4 bits of 'addressing mode' and 4 bits of 'register':
  • The CPU has sixteen 32-bit registers r0 ... r13 (fp), r14 (sp), r15 (pc).
  • The addressing modes (reminiscent of those of the PDP-11), and their encodings, are:

Encoding	Mode	notation
0x0	short literals 0-15
0x1	short literals 16-31
0x2	short literals 32-47
0x3	short literals 48-63
0x4	indexed	[r]
0x5	register immediate	r
0x6	register deferred	(r)
0x7	auto-decrement	-(r)
0x8	auto-increment	(r)+
0x9	auto-increment deferred	*(r)+
0xA	byte displacement	BD(r)
0xB	byte displacement deferred	*BD(r)
0xC	short-word displacement	WD(r)
0xD	short-word displacement deferred	*WD(r)
0xE	long-word displacement	LD(r)
0xF	long-word displacement deferred	*LD(r)

The "indexed" mode apparently adds the contents of the named register to the address generated by the following operand. (It likely cannot be used with the "register immediate" addressing mode.)

For operands of the 6 'displacement' modes, the operand is followed by a value of the indicated length.

Opcode table

The opcode map and mnemonics for the assembler are as follows:

	0	1	2	3	4	5	6	7	8	9	a	b	c	d	e	f
0x00	halt	-	-	-	-	sinf	ldf	ldd	addb2	movb	addw2	movw	addl2	movl	bbs	-
0x10	nop	brb	-	brw	-	cosf	lnf	lnd	addb3	cmpb	addw3	cmpw	addl3	cmpl	bbc	-
0x20	rei	bneq/bnequ	-	cvtwl	-	atanf	stf	std	subb2	mcomb	subw2	mcomw	subl2	mcoml	emul	aoblss
0x30	bpt	beql/beqlu	-	cvtwb	-	logf	cmpf	cmpd	subb3	bitb	subw3	bitw	subl3	bitl	ediv	aobleq
0x40	ret	bgtr	-	-	-	sqrtf	cmpf2	cmpd2	shll	clrb	shlq	clrw	mull2	clrl	shal	-
0x50	-	bleq	-	-	-	expf	tstf	tstd	shrl	tstb	shrq	tstw	mull3	tstl	shar	bbssi
0x60	ldpctx	-	-	-	-	-	-	pushd	-	incb	-	incw	divl2	incl	-	cvtlb
0x70	svpctx	jmp	-	-	-	-	cvlf	cvld	-	decb	-	decw	divl3	decl	-	cvtlw
0x80	-	bgeq	movs2	-	-	-	cvfl	cvdl	orb2	cvtbl	orw2	bispsw	orl2	adwc	adda	-
0x90	-	blss	cmps2	-	-	-	-	ldfd	orb3	cvtbw	orw3	bicpsw	orl3	sbwc	suba	-
0xa0	-	bgtru	-	-	-	-	cvdf	-	andb2	movzbl	andw2	loadr	andl2	mtpr	ffs	-
0xb0	-	blequ	-	-	-	-	negf	negd	andb3	movzbw	andw3	storer	andl3	mfpr	ffc	calls
0xc0	prober	bvc	movs3	movzwl	-	-	addf	addd	xorb2	movob	xorw2	movow	xorl2	movpsl	btcs	kcall
0xd0	probew	bvs	cmps3	-	-	-	subf	subd	xorb3	pushb	xorw3	pushw	xorl3	pushl	-	-
0xe0	insque	bgequ/bcs	-	-	-	-	mulf	muld	mnegb	movab	mnegw	movaw	mnegl	moval	-	-
0xf0	remque	bcc/blssu	-	-	-	-	divf	divd	movblk	pushab	-	pushaw	casel	pushal	callf	-

A number of the more complex arithmetic instructions, many of them the floating point and the trigonometric ones (CVLF, CVLD, CVFL, CVDL, LDFD, CVDF, ADDF, ADDD, SUBF, SUBD, MULF, MULD, DIVF, DIVD, SINF, COSF, ATANF, LOGF, SQRTF and EXPF), do not seem to be implemented in the Power 6/32; instead, they were emulated in software, after an attempt to execute them resulted in a trap.

See also

• Harris CX series

External links

• Harris HCX-5 - ad
• Harris HCX-9 - ad
• The Harris HCX-9 aka TAHOE platform
• Installing and Operating 4.3BSD UNIX on the Tahoe
• adb - debugger source
• dbx
• instrs - basic instruction table
  • instrs - alphabetical list
• dhrystone.c
• Cook, Carpenter, Catmull 1987 "The Reyes Image Rendering Architecture" in ACM Computer Graphics Vol. 21, No. 4 PDF indicating this machine was used at Pixar for a number of films