Go to advanced search

by phiren
Tue Jul 09, 2019 3:19 pm
Forum: General discussion
Topic: Pi 4 - full specification of VideoCore 6
Replies: 97
Views: 33881

Re: Pi 4 - full specification of VideoCore 6

Each ALU typically have 2 floating point operators, and as you pointed out in a earlier post videocore 6 is no exception, with both a multiply and additive floating point operator. Ah. No. The QPUs can do two floating point operations per cycle; One in the add ALU and one in the multiply ALU. That'...
by phiren
Tue Jul 09, 2019 4:05 am
Forum: General discussion
Topic: Pi 4 - full specification of VideoCore 6
Replies: 97
Views: 33881

Re: Pi 4 - full specification of VideoCore 6

Each QPU has 2 ALUs, so Raspberry Pi 3's 12 QPU graphics processor actually has 24 ALU x 2 x 0.4GHz = 19.2GFLOPs, with a clock of 0.5GHz, you get the advertised 24GFLOPs. Where is your extra x2 coming from? I had 12 QPUs x 2 ALUs x 0.4ghz = 9.6 GFLOPs which already accounts for the fact that each Q...
by phiren
Fri Jul 05, 2019 5:43 pm
Forum: General discussion
Topic: Pi 4 - full specification of VideoCore 6
Replies: 97
Views: 33881

Re: Pi 4 - full specification of VideoCore 6

(Not knowing which ones were there in the first place, but being mildly curious due to ML) vc4 had v8adds, v8subs, v8muld, v8min and v8max which operated on four 8bit uint values packed into a 32bit register. Multiplication was in the range 0.0 to 1.0 and addition/subtraction saturated. There were ...
by phiren
Fri Jul 05, 2019 4:54 am
Forum: General discussion
Topic: Pi 4 - full specification of VideoCore 6
Replies: 97
Views: 33881

Re: Pi 4 - full specification of VideoCore 6

- Technical information on VC. Is the VC6 an uprev of VC4 and so by-and-large compatible with VC4 software (e.g. open source compilers) or is the architecture significantly different? Are there more QPUs? More VPUs? Is the VPU instruction set similar? I notice OpenGL ES 3.0 is supported (up from 2....
by phiren
Tue Apr 01, 2014 11:21 pm
Forum: Other
Topic: Hard Float Setting for Building Linux Kernel
Replies: 1
Views: 4779

Re: Hard Float Setting for Building Linux Kernel

I ran across this problem, myself. You probably don't care by now, but I'm replying so that anyone else who stumbles across this doesn't have to research it themselves. The linux kernel doesn't use any floating point code internally, so removing the -mfloat-soft flags from the makefiles will have no...

Go to advanced search