Compiler/TMS320F280049: Proper Optimization Settings for fast control

Manish,

Thanks for the help.   I removed the volatile.

I put Vars in Ram M1 and Coefs in Ram GS0  (also tried this the other way too),  Function Called from RAM LS4...

No change.  Disassembly is below.

Any other ideas?

This is being called from an interrupt...  would that cause a problem or would it be better to call a function from the interrupt? (there is usually a minor penalty for adding a function, so I would prefer not to do it).

Thanks,

Kyle

2018    		CNTL_2P2Z_F_C(ontrol_Coef, Control_Var);
        C$L52:
00a8b3:   761F0019    MOVW         DP, #0x19
00a8b5:   E2AF0002    MOV32        R0H, @0x2, UNCF
00a8b7:   E2AF0100    MOV32        R1H, @0x0, UNCF
00a8b9:   E7200008    SUBF32       R0H, R1H, R0H
00a8bb:   7700        NOP          
00a8bc:   E2030004    MOV32        @0x4, R0H
00a8be:   761F0300    MOVW         DP, #0x300
00a8c0:   E2AF0006    MOV32        R0H, @0x6, UNCF
00a8c2:   761F0019    MOVW         DP, #0x19
00a8c4:   E2AF040E    MOV32        R4H, @0xe, UNCF
00a8c6:   761F0300    MOVW         DP, #0x300
00a8c8:   E2AF0108    MOV32        R1H, @0x8, UNCF
00a8ca:   761F0019    MOVW         DP, #0x19
00a8cc:   E2AF030C    MOV32        R3H, @0xc, UNCF
00a8ce:   761F0300    MOVW         DP, #0x300
00a8d0:   E2AF0200    MOV32        R2H, @0x0, UNCF
00a8d2:   761F0019    MOVW         DP, #0x19
00a8d4:   E3010508    MPYF32       R0H, R4H, R0H
                   || MOV32        R5H, @0x8
00a8d6:   761F0300    MOVW         DP, #0x300
00a8d8:   E302CB02    MPYF32       R1H, R3H, R1H
                   || MOV32        R3H, @0x2
00a8da:   761F0019    MOVW         DP, #0x19
00a8dc:   E3055406    MPYF32       R2H, R5H, R2H
                   || MOV32        R4H, @0x6
00a8de:   761F0300    MOVW         DP, #0x300
00a8e0:   E3120504    ADDF32       R0H, R0H, R1H
                   || MOV32        R5H, @0x4
00a8e2:   761F0019    MOVW         DP, #0x19
00a8e4:   E74100E1    MPYF32       R1H, R4H, R3H
                   || ADDF32       R0H, R0H, R2H
00a8e6:   E2AF0604    MOV32        R6H, @0x4, UNCF
00a8e8:   E7408172    MPYF32       R2H, R6H, R5H
                   || ADDF32       R0H, R0H, R1H
00a8ea:   7700        NOP          
00a8eb:   E7100080    ADDF32       R0H, R0H, R2H
00a8ed:   7700        NOP          
00a8ee:   E2030010    MOV32        @0x10, R0H
00a8f0:   0606        MOVL         ACC, @0x6
00a8f1:   1E08        MOVL         @0x8, ACC
00a8f2:   0604        MOVL         ACC, @0x4
00a8f3:   1E06        MOVL         @0x6, ACC
00a8f4:   0610        MOVL         ACC, @0x10
00a8f5:   761F0300    MOVW         DP, #0x300
00a8f7:   E2AF000A    MOV32        R0H, @0xa, UNCF
00a8f9:   761F0019    MOVW         DP, #0x19
00a8fb:   1E0A        MOVL         @0xa, ACC
00a8fc:   E2AF010A    MOV32        R1H, @0xa, UNCF
00a8fe:   E6940001    CMPF32       R1H, R0H
00a900:   AD14        MOVST0       NF,ZF
00a901:   6404        SB           C$L53, LT
00a902:   761F0300    MOVW         DP, #0x300
00a904:   060A        MOVL         ACC, @0xa
        C$L53:
00a905:   761F0019    MOVW         DP, #0x19
00a907:   1E0A        MOVL         @0xa, ACC
00a908:   761F0300    MOVW         DP, #0x300
00a90a:   E2AF000C    MOV32        R0H, @0xc, UNCF
00a90c:   761F0019    MOVW         DP, #0x19
00a90e:   E2AF010A    MOV32        R1H, @0xa, UNCF
00a910:   E6940001    CMPF32       R1H, R0H
00a912:   AD14        MOVST0       NF,ZF
00a913:   6204        SB           C$L54, GT
00a914:   761F0300    MOVW         DP, #0x300
00a916:   060C        MOVL         ACC, @0xc
        C$L54:
00a917:   761F0019    MOVW         DP, #0x19
00a919:   1E0A        MOVL         @0xa, ACC
00a91a:   060C        MOVL         ACC, @0xc
00a91b:   1E0E        MOVL         @0xe, ACC
00a91c:   060A        MOVL         ACC, @0xa
00a91d:   761F0300    MOVW         DP, #0x300
00a91f:   E2AF000E    MOV32        R0H, @0xe, UNCF
00a921:   761F0019    MOVW         DP, #0x19
00a923:   1E0C        MOVL         @0xc, ACC
00a924:   E2AF010A    MOV32        R1H, @0xa, UNCF
00a926:   E6940001    CMPF32       R1H, R0H
00a928:   AD14        MOVST0       NF,ZF
00a929:   6503        SB           C$L55, LEQ
00a92a:   060A        MOVL         ACC, @0xa
00a92b:   6F04        SB           C$L56, UNC
        C$L55:
00a92c:   761F0300    MOVW         DP, #0x300
00a92e:   060E        MOVL         ACC, @0xe
        C$L56:
00a92f:   761F0019    MOVW         DP, #0x19
00a931:   1E0A        MOVL         @0xa, ACC

Manish,
Trying a few other things.

Setting the optimization settings to "2-Global Optimizations" (or above) and setting the size vs speed to at least 2... seemed to help a lot.

Those options didn't seem to help without your previous recommendations.

It's still not down to the 42 cycles advertised, but it's much closer.


2023 CNTL_2P2Z_F_C(Control_Coef, Control_Var);
00a765: E2AF0516 MOV32 R5H, @0x16, UNCF
00a767: E30A9418 MPYF32 R2H, R2H, R5H
|| MOV32 R4H, @0x18
00a769: E700011C MPYF32 R4H, R3H, R4H
00a76b: BDA60F2A MOV32 R6H, @XAR6
00a76d: E7100112 ADDF32 R2H, R2H, R4H
00a76f: BDA70F22 MOV32 R4H, @XAR7
00a771: 7700 NOP
00a772: E2AF0110 MOV32 R1H, @0x10, UNCF
00a774: E7000071 MPYF32 R1H, R6H, R1H
00a776: E2AF0512 MOV32 R5H, @0x12, UNCF
00a778: E7411365 MPYF32 R5H, R4H, R5H
|| ADDF32 R1H, R1H, R2H
00a77a: E2AF0614 MOV32 R6H, @0x14, UNCF
00a77c: E7429382 MPYF32 R2H, R0H, R6H
|| ADDF32 R1H, R1H, R5H
00a77e: 7700 NOP
00a77f: E312AE1A ADDF32 R5H, R2H, R1H
|| MOV32 R6H, @0x1a
00a781: E2AF021C MOV32 R2H, @0x1c, UNCF
00a783: E6CF0029 MOV32 R1H, R5H, UNCF
00a785: E6970031 MINF32 R1H, R6H
00a787: E2AF061E MOV32 R6H, @0x1e, UNCF
00a789: E6960011 MAXF32 R1H, R2H
00a78b: 761F0019 MOVW DP, #0x19
00a78d: E6CF000A MOV32 R2H, R1H, UNCF
00a78f: E2030004 MOV32 @0x4, R0H
00a791: C308 MOVL @0x8, XAR7
00a792: E2030006 MOV32 @0x6, R0H
00a794: E203030E MOV32 @0xe, R3H
00a796: E203010C MOV32 @0xc, R1H
00a798: E2030510 MOV32 @0x10, R5H
00a79a: E6960032 MAXF32 R2H, R6H
00a79c: 6F47 SB C$L178, UNC
C$L177:
00a79d: 761F0300 MOVW DP, #0x300

With macros, profiling is not exact or repeatable because of surrounding code affecting the code compiled. 

I re-profiles and am getting 43 cycles 

;***************************************************************
;* FNAME: _CNTL_2P2Z_F_Routine          FR SIZE:   4           *
;*                                                             *
;* FUNCTION ENVIRONMENT                                        *
;*                                                             *
;* FUNCTION PROPERTIES                                         *
;*                            0 Parameter,  0 Auto,  4 SOE     *
;***************************************************************

_CNTL_2P2Z_F_Routine:
;* R3H   assigned to $O$C1
;* R1H   assigned to $O$v4
;* R0H   assigned to $O$v3
;* R2H   assigned to $O$v1
	.dwcfi	cfa_offset, -2
	.dwcfi	save_reg_to_mem, 26, 0
        MOVW      DP,#_cntl_2p2z_vars1+12 ; [CPU_ARAU] 
	.dwpsn	file "../Float_Profiling-Routines.c",line 215,column 3,is_stmt,isa 0
        MOV32     R0H,@_cntl_2p2z_vars1+12 ; [CPU_FPU] |215| 
        MOV32     R1H,@_cntl_2p2z_vars1+10 ; [CPU_FPU] |215| 

        SUBF32    R2H,R1H,R0H           ; [CPU_FPU] |215| 
||      MOV32     R3H,@_cntl_2p2z_vars1+6 ; [CPU_FPU] |215| 

        MOV32     *SP++,R4H             ; [CPU_FPU] 
	.dwcfi	save_reg_to_mem, 60, 2
	.dwcfi	cfa_offset, -4
        MOV32     @_cntl_2p2z_vars1+4,R2H ; [CPU_FPU] |215| 
        MOVW      DP,#_cntl_2p2z_coeffs1+6 ; [CPU_ARAU] 
        MOV32     R0H,@_cntl_2p2z_coeffs1+6 ; [CPU_FPU] |215| 
        MOVW      DP,#_cntl_2p2z_vars1+2 ; [CPU_ARAU] 
        MOV32     R4H,@_cntl_2p2z_vars1+2 ; [CPU_FPU] |215| 
        MOVD32    R1H,@_cntl_2p2z_vars1 ; [CPU_FPU] |215| 
        MOVW      DP,#_cntl_2p2z_coeffs1+8 ; [CPU_ARAU] 

        MOV32     R4H,@_cntl_2p2z_coeffs1+8 ; [CPU_FPU] |215| 
||      MPYF32    R0H,R4H,R0H           ; [CPU_FPU] |215| 

        MPYF32    R4H,R1H,R4H           ; [CPU_FPU] |215| 
        MOV32     *SP++,R5H             ; [CPU_FPU] 
	.dwcfi	save_reg_to_mem, 64, 4
	.dwcfi	cfa_offset, -6

        MOV32     R4H,@_cntl_2p2z_coeffs1 ; [CPU_FPU] |215| 
||      ADDF32    R0H,R0H,R4H           ; [CPU_FPU] |215| 

        MOVW      DP,#_cntl_2p2z_vars1+8 ; [CPU_ARAU] 
        MOV32     R5H,@_cntl_2p2z_vars1+8 ; [CPU_FPU] |215| 
        MPYF32    R4H,R5H,R4H           ; [CPU_FPU] |215| 
        MOVW      DP,#_cntl_2p2z_coeffs1+2 ; [CPU_ARAU] 

        MOV32     R4H,@_cntl_2p2z_coeffs1+2 ; [CPU_FPU] |215| 
||      ADDF32    R0H,R0H,R4H           ; [CPU_FPU] |215| 

        MPYF32    R4H,R3H,R4H           ; [CPU_FPU] |215| 
        NOP       ; [CPU_ALU] 

        MOV32     R4H,@_cntl_2p2z_coeffs1+4 ; [CPU_FPU] |215| 
||      ADDF32    R0H,R0H,R4H           ; [CPU_FPU] |215| 

        MOVW      DP,#_cntl_2p2z_vars1+8 ; [CPU_ARAU] 
        MOV32     @_cntl_2p2z_vars1+8,R3H ; [CPU_FPU] |215| 

        MOV32     @_cntl_2p2z_vars1+6,R2H ; [CPU_FPU] |215| 
||      MPYF32    R4H,R2H,R4H           ; [CPU_FPU] |215| 

        MOVW      DP,#_cntl_2p2z_coeffs1+10 ; [CPU_ARAU] 
        ADDF32    R0H,R0H,R4H           ; [CPU_FPU] |215| 
        MOV32     R2H,@_cntl_2p2z_coeffs1+10 ; [CPU_FPU] |215| 
        MINF32    R0H,R2H               ; [CPU_FPU] |215| 
        MOV32     R2H,@_cntl_2p2z_coeffs1+12 ; [CPU_FPU] |215| 
        MOVW      DP,#_cntl_2p2z_vars1  ; [CPU_ARAU] 
        MAXF32    R0H,R2H               ; [CPU_FPU] |215| 
        MOV32     @_cntl_2p2z_vars1,R0H ; [CPU_FPU] |215| 
        MOVW      DP,#_cntl_2p2z_coeffs1+14 ; [CPU_ARAU] 
        MOV32     R1H,@_cntl_2p2z_coeffs1+14 ; [CPU_FPU] |215| 
        MOVW      DP,#_cntl_2p2z_vars1+14 ; [CPU_ARAU] 
        MAXF32    R0H,R1H               ; [CPU_FPU] |215| 
        MOV32     @_cntl_2p2z_vars1+14,R0H ; [CPU_FPU] |215| 
        MOV32     R5H,*--SP             ; [CPU_FPU] 
	.dwcfi	cfa_offset, -4
	.dwcfi	restore_reg, 64
        MOV32     R4H,*--SP             ; [CPU_FPU] 
	.dwcfi	cfa_offset, -2
	.dwcfi	restore_reg, 60
$C$DW$88	.dwtag  DW_TAG_TI_branch
	.dwattr $C$DW$88, DW_AT_low_pc(0x00)
	.dwattr $C$DW$88, DW_AT_TI_return

        LRETR     ; [CPU_ALU] 

Also please note macros are not the best way to program , we do have some legacy modules such as the one you are referring to. However they are in non maintained state. 

We have now moved to DCL or the Digital Controller Library which is much more feature rich and well tested that these modules. I will highly recommend using that as that is being actively maintained and supported.  

Manish,

Thanks.  I'll move toward the DCL library.

One quick questions on it:

I'm assuming that the same coefficients (a1, 12, b0, b1, b3, etc...) match the previous 2P2Z.  Looking at the equations,  that appears to be the case.  I want to double check.

Also,  Anything I need to know about SFRA and if plan on changing that around.  It works well, but if you plan one make changes, let me know.

Thank you for the help,

Kyle

Manish,
I think you have an issue with your CompDesigner GUI.

Your Frequency is off. Even with your example file (SFRAData.csv) from the folder the plant has has a crossover of about 1.2kHz. The Comp designer shows ~0.25 kHz. (Using C2000Ware_DigitalPower_SDK_1_02_00_00)

I checked it versus an old version and the data looks fine.

The Magnitdue shows up correctly, just the frequency looks to have a scaling factor applied to it and is for some reason connected to the Control ISR frequency... I can't see how this would be correct...

Kyle, 

I just opened the compensation designer on my end and it looks ok to me. 

The crossover that the GUI reports is of the open loop gain , which is affected by the control frequency and hence that is expected. 

Can you post some snapshot of what you are seeing to enable me to debug what you are observing?

-Manish

The crossover of the open loop certainly.

But what I'm seeing is that the plant data is being changed as well (red line).  That should be static. 

Below is a screenshot of the "SFRAData.csv" that gets installed with version 1.02. 

On the left: is the data in the old compensation designer with the knee in the plant about 1.2kHz... Crossover at 4kHz

On the right: is the new compensation designer  with the same data and same poles and zeros..  notice the red line has moved and the knee is now at 200Hz...  and this obviously affects the Open Loop crossover as well.  This is incorrect.  The plant should be fixed.  

The data clearly shows that the knee is about 1.2kHz .  The plant doesn't move with control frequency (you might lose phase, or gain near the control frequency, but not like this).

This doesn't match your documentation which shows this data correctly (plant knee at ~1.2kHz.  So at some point it worked well...  It does not on my machine after a fresh install.

I also added the same controller a 2P2Z with the same poles, zeros, and magnitude and get totally different graphs and the B coefficients are different (accounting for the sign change you mentioned in the A coefficients).

The second image is just the compensator using the same values...  that doesn't even match...

If it helps the File Version from windows details says 2.1.0.0, and Product version is 2.01.0.0  (old version is 1.10.0.0 and 1.10.0.0 respectively).

Kyle thanks for the details. I will look into this and reply back. But considering holiday schedule it may only in the next year.

Kyle,

I checked back at this and am unable to repeat the issue.

Can you confirm both the GUI's are pointing to the same SFRA data.csv

For example I pointed them both to the SFRA data.csv and with all the settings correct I am able to get the same results.

I believe the issue you might be seeing is that in the new compensation designer Kdc is no londer in dB ??

Confirmed, same file (and files... I tried this with multiple files).  I wouldn't go through the trouble of saying there is an issue if I did do this.  

I can see your plant is identical between the two.  That is not the case with mine... this is where there is something off.  Please post a zip file with your compensation designer and the associated DLL to make it run.  I'm attaching a zip of my non-working one with the example plant that data shows 1.2kHz knee, but shows up as 202Hz in the comp designer.

In no case should the plant ever change.  Those are measure values and will not shift.  The fact that mine are moving is incorrect.

NonWorkingCompDesigner.zip

This would be the case as to why the compensation is different... I get that. But still, the plant should never move... that is the bigger issue.

Manish,
Good, I'm not crazy.

The save comp didn't fix the issue, but manually changing the path in the XML file to the SFRA directory did. Not sure why the save comp button didn't work, but either way it appears to work now.

And now time to migrate and try these newer control functions.

Thanks,
Kyle