Received: (from major@localhost) by minnie.cs.adfa.oz.au (8.8.5/8.8.5) id NAA02467 for pups-liszt; Wed, 15 Apr 1998 13:09:51 +1000 (EST) X-Authentication-Warning: minnie.cs.adfa.oz.au: major set sender to owner-pups@minnie.cs.adfa.oz.au using -f Received: from renoir.op.net (root@renoir.op.net [209.152.193.4]) by minnie.cs.adfa.oz.au (8.8.5/8.8.5) with ESMTP id NAA02451 for ; Wed, 15 Apr 1998 13:09:41 +1000 (EST) Received: from goppelt.op.net (d-phlarc1-00.ppp.op.net [209.152.199.64]) by renoir.op.net (o1/$Revision: 1.16 $) with SMTP id XAA00270; Tue, 14 Apr 1998 23:09:28 -0400 (EDT) Message-Id: <199804150309.XAA00270@renoir.op.net> Comments: Authenticated sender is From: "Ed G." To: pete@dunnington.u-net.com (Pete Turnbull) Date: Tue, 14 Apr 1998 23:09:26 -0400 MIME-Version: 1.0 Content-type: text/plain; charset=US-ASCII Content-transfer-encoding: 7BIT Subject: Re: Floating Point-How Important Reply-to: edgee@cyberpass.net CC: pups@minnie.cs.adfa.oz.au In-reply-to: <9804111346.ZM7828@indy.dunnington.york.ac.uk> References: "Ed G." "Re: Floating Point-How Important to Unix?" (Apr 10, 22:40) X-mailer: Pegasus Mail for Win32 (v2.54) Sender: owner-pups@minnie.cs.adfa.oz.au Precedence: bulk > What about position-independent code? Your query got me thinking about the various addressing modes of the PDP-11 and how they might affect my brute force approach to estimating floating point ops for C programs. Is this what you meant when you asked about position independent code? And yes, these addressing modes could mean the death knell for my approach. Index mode is definitely a problem as C programs seem to use r5 as a frame pointer with both positive and *negative* 16 bit offsets (see assembly language listing of my square root program below). I don't think PC relative mode (e.g., clr addr) is a problem (if the data segment follows the text, then the offsets would all be positive and all less than the size of the program). Is there such a thing as PC relative mode for the jmp op code? In other words, can you make long + or -32K relative jumps on the PDP-11? If so, this too could potentially confound my estimates. .globl _absv .text _absv: ~~absv: jsr r5,csv ~n=4 jbr L1 L2:clrf r0 cmpf 4(r5),r0 cfcc jge L4 movf 4(r5),r0 negf r0 jbr L3 jbr L5 L4:movf 4(r5),r0 jbr L3 L5:L3:jmp cret L1:jbr L2 .globl _mysqrt .text _mysqrt: ~~mysqrt: jsr r5,csv ~n=4 jbr L6 L7:~g=177762 ~err=177752 movf 4(r5),r0 divf $40400,r0 movf r0,-16(r5) .data L10000:77777;177776;177777;177777 .text movf 4(r5),r0 divf L10000,r0 movf r0,-26(r5) movf -16(r5),r0 movf r0,-(sp) mov $L9,-(sp) jsr pc,_printf add $12,sp L10:movf -16(r5),r0 mulf -16(r5),r0 subf 4(r5),r0 movf r0,-(sp) jsr pc,_absv add $10,sp cmpf -26(r5),r0 cfcc jgt L11 movf -16(r5),r0 mulf -16(r5),r0 addf 4(r5),r0 movf $40400,r1 mulf -16(r5),r1 divf r1,r0 movf r0,-16(r5) movf -16(r5),r0 movf r0,-(sp) mov $L12,-(sp) jsr pc,_printf add $12,sp jbr L10 L11:movf -16(r5),r0 jbr L8 L8:jmp cret L6:sub $20,sp jbr L7 .globl _main .text _main: ~~main: jsr r5,csv jbr L13 L14:.data L10001:77777;177776;177777;177777 .text movf L10001,r0 movf r0,-16(r5) ~n=177762 movf -16(r5),r0 movf r0,-(sp) jsr pc,_mysqrt add $10,sp movf r0,-(sp) mov $L16,-(sp) jsr pc,_printf add $12,sp L15:jmp cret L13:sub $10,sp jbr L14 .globl fltused .globl .data L9:.byte 111,156,151,164,151,141,154,40,147,165,145,163,163,72 .byte 40,45,56,61,66,146,12,12,0 L12:.byte 147,165,145,163,163,72,40,45,56,61,66,146,12,0 L16:.byte 12,115,171,40,163,161,165,141,162,145,40,162,157,157 .byte 164,40,151,163,72,40,45,56,61,66,146,12,0