DSP

[SimonLothar]

TeamFX proposed to use the DSP-features of the fx-MPUs.

Indeed this seems to be possible on SH-7305-based calculators.
Bit 12 of sr must be set. Otherwise DSP-instructions give an illegal instruction error.
Though after a hardware-RESET the instruction SETRC gives an illegal instruction error irrespective of sr bit 12. After a controlled off/ON, SETRC behaves itself. The reason of which has yet to be unravelled.
With the renesas SH compiler comes a library sh3dspnb.lib (together with ensigdsp.h and _dspc_type.h ) which works with the old SDK. The library functions are documented in the SHC-manual chapter 10.3.5.
Most information is covered in [Renesas REJ09B0317-0400 Rev. 4.00 Revision Date: May 15, 2006].
The Instructions PWSB and PWAD (see [Renesas REJ09B0317-0400 Rev. 4.00 Revision Date: May 15, 2006] page 93) are not supported by the old SDK.

[TeamFX Profile]

Some related discussion can be found here:
https://www.cemetech.net/forum/viewtopic.php?t=11918

[Jsec42 Profile]

What would some of the advantages of abusing the DSP hardware be?

[SimonLothar]

What would some of the advantages of abusing the DSP hardware be?

It is speed. The DSP is able to process up to four operations in one cycle.
Consider the sh3dsp-library function "mean", which calculates the mean of 16-bit values (max. 2048 measurements), f. i. to reduce measurement noise.
It runs about four to five times faster compared to a conventional programmed C function on a SH-4A-fx-9860.

Though, to be honest: On fx-calculators I think the use of the DSP-functions has some academical value to gain experience, only. Which is enough for me, but not necessarily for anybody. At this juncture the crash after RESET prevents the DSP to be used beyond an academical approach.

Additionally it looks as if they lost interest in DSP with the regular SH-4A MPUs. At least I did not find any proper documentation, yet. Perhaps the SH-4A-MPUs are too fast as such and a DSP is not worth the trouble any more.

If you abuse it or use it, is up to you.

BTW (off-topic): I dropped fx_calculators_SuperH_based.chm (20) with a few news. The most (possibly only) interesting thing seems to be the (I hope unabridged) lists of SH7337/7355-, SH7291- and SH7305-registers, which I gathered based on a suggestion of TeamFX.

[brickviking Profile]

Could you explain what you mean by "Dropped"? I don't know this term, except in "I dropped a plate".

Thanks.

(Post 16)

[SimonLothar]

Could you explain what you mean by "Dropped"? I don't know this term, except in "I dropped a plate".

Oh, sry. I do so much Drag&Drop all day long. Seems that I subconsciously got accustomed to the term "drop a file". The percentage of dropping files is above 99.9 % compared to dropping plates, as far as I am concerned. To be honest I cannot remember the last instance me dropping a plate or any other dishes.
But to answer your question: I meant "uploaded".

[SimonLothar]

What would some of the advantages of abusing the DSP hardware be?

The library function "Variance" seems to be three to four times faster compared to a conventional approach.

[Jsec42 Profile]

It is very possible that they lost interest in the DSP instructions in the SH4a. After all, the Casio 9860gii/2, Prism, and CP-400 are tied for the fastest graphing calculator when set to their top speed.