Technical discussions about the TI C55x DSPs (including the c5501, c5502, c5503, c5507, c5509, c5510 and OMAP5910).
Gary- > I believe your assumptions are incorrect. The whole point of "soft- > parallelism" is to do it at runtime, not compile-time. This type of > behavior is often seen in pipelined processors, especially those with > multiple execution pipelines. Does TI documentation support your comments? Is there a specific SPRU-number or other doc that indicates run-time decisions? I'm dubious about "intelligent run-time decisions" on 55x devices as those are relatively old (designed in 2000/2001) and TI hasn't introduced anything dramatically new for the CPU/core of the 55x line in the last few years, as they've done for C64x (e.g. C64x+ core). -Jeff > --- In c...@yahoogroups.com, bla bla <perro_verde_lunar@...> wrote: > > > > Hi, > > > > I've read the Instruction Set Reference Guide but my > > problem it is not how to obtain paralleism on my DSP > > code, my problem how to handle parallelism in the > > simulator I am programming. I could program the rules > > and let my simulator handle parallelism but losing too > > many MIPS in the process. > > > > Parallelism is determined at compile time in C55x, and > > reading the CPU Reference Guide (SPRU371F), page 1-7, > > it's clearly written that the instruction decoder > > "determines whether the CPU has been INSTRUCTED to > > execute two instructions in parallel", so I assume > > that somewhere in the binary there are some hints for > > the soft dual parallelism and other exceptions cases. > > Maybe the DSP's decoder has a big hardware LUT that > > automatically tells the DSP to execute the following > > two instructions in parallel. > > > > Thanks for your answer and time. > > > > Carl. Check Out Industry's First Single-Chip, Multi-Format, Real-Time HD Video Transcoding Solution for Commercial & Consumer End Equipment: www.ti.com/dm6467
Hi Jeff, In the CPU reference guide book (SPRU371F), at page 1-7 it's written: "The instruction decoder: - identifies instruction boundaries so that it can decode 8-, 16-, 24-, 32-, 40- and 48-bit instruction. - DETERMINES WHETHER THE CPU HAS BEEN INSTRUCTED TO EXECUTE TWO INSTRUCTIONS IN PARALLEL. - Sends decoded execution commands..." So, I think that support our point of view. Thanks for your answer and time. Carl. --- Jeff Brower <j...@signalogic.com> escribió: > Gary- > > > I believe your assumptions are incorrect. The > whole point of "soft- > > parallelism" is to do it at runtime, not > compile-time. This type of > > behavior is often seen in pipelined processors, > especially those with > > multiple execution pipelines. > > Does TI documentation support your comments? Is > there a specific SPRU-number or > other doc that indicates run-time decisions? > > I'm dubious about "intelligent run-time decisions" > on 55x devices as those are > relatively old (designed in 2000/2001) and TI hasn't > introduced anything dramatically > new for the CPU/core of the 55x line in the last few > years, as they've done for C64x > (e.g. C64x+ core). > > -Jeff > > > --- In c...@yahoogroups.com, bla bla > <perro_verde_lunar@...> wrote: > > > > > > Hi, > > > > > > I've read the Instruction Set Reference Guide > but my > > > problem it is not how to obtain paralleism on my > DSP > > > code, my problem how to handle parallelism in > the > > > simulator I am programming. I could program the > rules > > > and let my simulator handle parallelism but > losing too > > > many MIPS in the process. > > > > > > Parallelism is determined at compile time in > C55x, and > > > reading the CPU Reference Guide (SPRU371F), page > 1-7, > > > it's clearly written that the instruction > decoder > > > "determines whether the CPU has been INSTRUCTED > to > > > execute two instructions in parallel", so I > assume > > > that somewhere in the binary there are some > hints for > > > the soft dual parallelism and other exceptions > cases. > > > Maybe the DSP's decoder has a big hardware LUT > that > > > automatically tells the DSP to execute the > following > > > two instructions in parallel. > > > > > > Thanks for your answer and time. > > > > > > Carl. > ------------------------------------ Check Out Industry's First Single-Chip, Multi-Format, Real-Time HD Video Transcoding Solution for Commercial & Consumer End Equipment: www.ti.com/dm6467