By: Brett (ggtgp.delete@this.yahoo.com), August 8, 2022 12:31 am
Room: Moderated Discussions
Marcus (m.delete@this.bitsnbites.eu) on August 7, 2022 11:30 pm wrote:
> Megol (golem960.delete@this.gmail.com) on August 7, 2022 2:01 pm wrote:
> > Marcus (m.delete@this.bitsnbites.eu) on August 6, 2022 4:36 am wrote:
> > > Have you seen the 68080? It's a 64-bit 4-wide OoO 68k CPU with more registers and instructions
> > > than the 32-bit 68k line (68000-68060). It's implemented in an FPGA. Pretty impressive IMO.
> >
> > AFAIK it's an in-order 2 wide superscalar with load-execute-store pipelines and powerful instruction fusion?
> > Personally the extensions are too hacky but yes, pretty impressive.
> >
>
> I have trouble finding any definitive information and thorough documentation,
> but check out http://www.apollo-core.com/index.htm?page=features
>
> It states:
>
> * Out of Order CPU
> * Out of Order FPU
> * Executes up to 4 instructions per clock cycle
>
> OTOH it only has two integer execution units, according to the same
> page, so four instructions may be when mixing in FPU & AMMX?
A typical 4 way RISC arch will have around two integer, a load pipe, a load/store pipe, FPU mul, FPU add.
So up to six issue, but the decode and complete are four wide, determining width.
> Megol (golem960.delete@this.gmail.com) on August 7, 2022 2:01 pm wrote:
> > Marcus (m.delete@this.bitsnbites.eu) on August 6, 2022 4:36 am wrote:
> > > Have you seen the 68080? It's a 64-bit 4-wide OoO 68k CPU with more registers and instructions
> > > than the 32-bit 68k line (68000-68060). It's implemented in an FPGA. Pretty impressive IMO.
> >
> > AFAIK it's an in-order 2 wide superscalar with load-execute-store pipelines and powerful instruction fusion?
> > Personally the extensions are too hacky but yes, pretty impressive.
> >
>
> I have trouble finding any definitive information and thorough documentation,
> but check out http://www.apollo-core.com/index.htm?page=features
>
> It states:
>
> * Out of Order CPU
> * Out of Order FPU
> * Executes up to 4 instructions per clock cycle
>
> OTOH it only has two integer execution units, according to the same
> page, so four instructions may be when mixing in FPU & AMMX?
A typical 4 way RISC arch will have around two integer, a load pipe, a load/store pipe, FPU mul, FPU add.
So up to six issue, but the decode and complete are four wide, determining width.