amull@Morgan.COM (Andrew P. Mullhaupt) (12/07/89)
We are having a good time with the new Cyrix FasMath 83d87 plug-in replacement for the 80387. It's pretty fast. It would be even faster if we could generate code which would take full advantage of it, as opposed to just 80387 code. Note that this code would be expected to run on 80387 chips also, but perhaps not optimally. 1. Is this really actually possible (optimal code for 83D87 which doesn't break the 80387). It would seem so. 2. Is there any plan for a compiler to generate such code? Language is not the main issue, but code generation is. (There is interest, but somewhat less excited interest, in the code generation for the II2 80287/80387 replacements.) Thanks in advance. Andrew Mullhaupt
BHB3@PSUVM.BITNET (12/08/89)
In article <582@s5.Morgan.COM>, amull@Morgan.COM (Andrew P. Mullhaupt) says: > >We are having a good time with the new Cyrix FasMath 83d87 plug-in >replacement for the 80387. It's pretty fast. It would be even faster >if we could generate code which would take full advantage of it, as >opposed to just 80387 code. Note that this code would be expected to >run on 80387 chips also, but perhaps not optimally. > >1. Is this really actually possible (optimal code for 83D87 which >doesn't break the 80387). It would seem so. > >2. Is there any plan for a compiler to generate such code? Language >is not the main issue, but code generation is. > >(There is interest, but somewhat less excited interest, in the code >generation for the II2 80287/80387 replacements.) > >Thanks in advance. >Andrew Mullhaupt As far as I understand the 80387 instruction set and the Cyrix one are excactly the same, except that the Cyrix executes many of them much quicker than the 80387. You could use the Borland or Microsoft Assembler. Both contai n the full 80387 instruction set. I suppose certain sequences of instruction would make optimal use of the Cyrix chip. I doubt any compiler manufacturers are going to put out special modules for it. The CYrix is competing against the Weitek 3167 series, which are faster than it. The selling point of the Cyr ix chip is that is is compatable with all 80x87 software, whereas the Weitek chips aren't. Brent H. Besler Ford Motor Scientific Research
amull@Morgan.COM (Andrew P. Mullhaupt) (12/11/89)
In article <89341.162025BHB3@PSUVM.BITNET>, BHB3@PSUVM.BITNET writes: > > As far as I understand the 80387 instruction set and the Cyrix one are > excactly the same, except that the Cyrix executes many of them much quicker > than the 80387. You could use the Borland or Microsoft Assembler. Both contai > n the full 80387 instruction set. I suppose certain sequences of instruction > would make optimal use of the Cyrix chip. I doubt any compiler manufacturers > are going to put out special modules for it. The CYrix is competing against > the Weitek 3167 series, which are faster than it. The selling point of the Cyr > ix chip is that is is compatable with all 80x87 software, whereas the Weitek > chips aren't. Our information is that the Cyrix is about the same speed as the Weitek 3167 when it gets a good instruction stream. It turns out that NDP FORTRAN and C both claim to generate Cyrix code as separate from their ability to generate 80387 code. Does anyone actually have both of these chips (Cyrix, Weitek) and know which is faster from first hand experience? The fact that Cyrix is as fast, more widely useful, and somewhat less expensive will prove decisive unless we get information to the contrary. (We are consideing a purchase, and we are debating whether to go through the trouble of testing the Weitek, since at least one of our critical applications (APL2/32) doesn't support the Weitek. Later, Andrew Mullhaupt