@NovaCoder

Awesome! I've left a comment on the YouTube video.

Not sure my 1260 @ 50mhz could run this. Just shows how capable AGA is given plenty of CPU/RAM though.

_________________
A1200, 68060/64MB/1.2GB/WiFi/AGAtoCRT/OS3.9 Pegasos I, G3 600Mhz/512/9200SE/80GB WinUAE, Ryzen 5 2400G/Vega11, 8GB DDR4, 256GB SSD,Win 10 Pro x64 Amiga Forever !

That is VERY impressing!

I didn' think an AGA 68K Machine could do that!

_________________
God created Paula so that Allister Brimble and Dave Whittaker could do music

Check my Amiga gameplays (ITA)!

@matthey
Quote:

The pipelined Pentium FPU is up to 2x as fast on paper but the unpipelined 68060 FPU can operate in parallel with the dual integer and branch units of the 68060.

As can the Pentium FPU. Advantage: Pentium.

Quote:

The Pentium stack based FPU is not as good of a design either.

The x86 (legacy) FPU isn't a pure stack based thingy, it's partially stack based and partially register based - one can freely select one operand to be any of the 8 stack slots. This in combination with that Pentium can execute one FXCHG instruction in parallel with another FPU instruction makes your point moot.

Quote:

The 68060 never received the software and optimization support that the Pentium did (especially from Motorola who were pushing PPC) or it would have made the higher clocked Pentium owners put their pencils away. With equivalent memory speed and caches, the 68060 is competitive with x86 and PPC processors running at 2x the clock speed.

Bull. Pentium and the 68060 are in most cases comparable and if one should be considered superior than the other it would be the Pentium. It have a pipelined FPU and a 64 bit data bus while the 68060 have early execution of some instructions.

Quote:

There isn't a compiler yet that generated good quality code for the 68060 (needs instruction scheduling as any superscalar processor). Nova's Q2 AGA should be running at 15+ fps on AGA IMO. The biggest problem with the c2p is the slow chip memory speed. C= never saw a need to improve it. Nova is using Kalm's c2p code for his ports so it still only takes a few percent (less than 10%) of the CPU power of a 68060.

Quote:

megol wrote: @matthey Quote: The pipelined Pentium FPU is up to 2x as fast on paper but the unpipelined 68060 FPU can operate in parallel with the dual integer and branch units of the 68060. As can the Pentium FPU. Advantage: Pentium.

Parallelism with more than 1 cache/memory access per cycle? The x86 integer units needed to use the stack (cache/memory) more because of fewer registers too. The 68060 FPU was a better design with more parallelism possible and was easier to program. The Pentium had pipelining, higher clock speeds and better compilers. The maximum potential of the x86 is certainly higher but more difficult to achieve. I think the 68060 FPU would be easier for a compiler to generate good code but no one cared as most implementations (including GCC) were a bare minimum of support with no instructions scheduling for FPU instructions. The x86 FPU would have the advantage in pure FPU code but I think the 68060 could level the playing field with the more common mixed FPU/integer code and possibly have an advantage with mostly integer instructions with a few FPU instructions.

Quote:

Quote: The 68060 never received the software and optimization support that the Pentium did (especially from Motorola who were pushing PPC) or it would have made the higher clocked Pentium owners put their pencils away. With equivalent memory speed and caches, the 68060 is competitive with x86 and PPC processors running at 2x the clock speed. Bull. Pentium and the 68060 are in most cases comparable and if one should be considered superior than the other it would be the Pentium. It have a pipelined FPU and a 64 bit data bus while the 68060 have early execution of some instructions.

My point was that back in the day, Pentium owners had an advantage they thought was due to a superior processor but it had more to do with support and software. We can see that now that less than optimized fps games using FPU instructions run at comparable (if not better) speeds clock for clock compared to the Pentium. That's with the Pentium having bragging rights of a pipelined FPU and a 64 bit data bus giving up to 2x the performance each in theory. A Motorola engineer reported that tests showed very little advantage in most code (less than 5% on average as I recall) for the wider data bus.

Last edited by matthey on 14-May-2013 at 04:44 PM.

@matthey
Quote:

Quote: matthey wrote: [quote] megol wrote: @matthey [quote] The pipelined Pentium FPU is up to 2x as fast on paper but the unpipelined 68060 FPU can operate in parallel with the dual integer and branch units of the 68060. As can the Pentium FPU. Advantage: Pentium.

Parallelism with more than 1 cache/memory access per cycle? The x86 integer units needed to use the stack (cache/memory) more because of fewer registers too. The 68060 FPU was a better design with more parallelism possible and was easier to program. The Pentium had pipelining, higher clock speeds and better compilers. The maximum potential of the x86 is certainly higher but more difficult to achieve. I think the 68060 FPU would be easier for a compiler to generate good code but no one cared as most implementations (including GCC) were a bare minimum of support with no instructions scheduling for FPU instructions. The x86 FPU would have the advantage in pure FPU code but I think the 68060 could level the playing field with the more common mixed FPU/integer code and possibly have an advantage with mostly integer instructions with a few FPU instructions.
[/quote]

Most inner loops (the critical ones) need no more than the 7 registers available, some requires 8 registers which isn't a problem in assembly (the stack pointer is a general register too). If more values are needed the Pentium allows two 32 bit memory operations per clock.
I have coded graphics code for the Pentium and only had problems with registers twice - one of the solutions was using self modifying code for two pseduo-constants (self modifying code was still inexpensive back then).

Generating good FPU code for the Pentium is easy. The 68060 have no advantage there - BY USING THE FREE FXCHG INSTRUCTION THE FPU CAN BE TREATED AS A PURE REGISTER MACHINE. And that was what most compilers did back then.

The cases where the 68060 had advantages are very uncommon, the cases where the Pentium have advantages are common for FPU intensive code.

Quote:

Quote: [quote] The 68060 never received the software and optimization support that the Pentium did (especially from Motorola who were pushing PPC) or it would have made the higher clocked Pentium owners put their pencils away. With equivalent memory speed and caches, the 68060 is competitive with x86 and PPC processors running at 2x the clock speed. Bull. Pentium and the 68060 are in most cases comparable and if one should be considered superior than the other it would be the Pentium. It have a pipelined FPU and a 64 bit data bus while the 68060 have early execution of some instructions.

My point was that back in the day, Pentium owners had an advantage they thought was due to a superior processor but it had more to do with support and software. We can see that now that less than optimized fps games using FPU instructions run at comparable (if not better) speeds clock for clock compared to the Pentium. That's with the Pentium having bragging rights of a pipelined FPU and a 64 bit data bus giving up to 2x the performance each in theory. A Motorola engineer reported that tests showed very little advantage in most code (less than 5% on average as I recall) for the wider data bus.
[/quote]

Do I need to remind you that it was _you_ that claimed the 68060 was 2x as fast as comparable clocked x86 and PPC processors?!?

As my first Pentium used a P24T overdrive processor in a 80486 motherboard I do know what the difference a 32 bit data bus made compared to a 64 bit one...