EVMK2H: MATHLIB

George,

George Rodionov said:

Is it possible to attach a header file and apply functions from the MATHLIB library without changing the algorithm (replace math.h to mathlib.h)? How do I do this?

I use the mathlib_c66x_3_1_2_4 library an EVMK2H device.

Yes, MATHLIB can be used to override functions that are defined in math.h. In order to do this, you will need to rebuild the MATHLIB with the OVERRIDE_RTS enabled in the compiler settings as this is not provided as a default functionality for this library. 

Please refer to this document that I had created some time back that provides information to rebuild MATHLIB with OVERRIDE_RTS=1 and linking to your DSP application.

Using_MATHLIB_RTS_functions_in_TI_RTOS_DSP_application.pdf

I am also providing a prebuilt version of MATHLIB and a reference project for you to check the linker settings.

C66x_Benhmark_using_MATHLIB.zip

Hope this helps.

Regards,

Rahul

Thanks for the answer, but it's not clear yet. And it doesn't work.

Q1: Do I need to exclude math.h from the project now? Or I can have him stay.
Q2: Do I need to go up all the c-files of the project in the "Link Order", according to the pdf instructions?
Q3: After completing the steps to connecting mathlib_rts.ae66, when going through the sin function and other mathematical functions, should they switch to MATHLIB files? This is not so.

The MATLIB download page has the mathlib_rts_c66x_3_1_2_1_Win32.exe library. The default option OVERRIDE_RTS = 1 is set in it. I used it and use provided file (of course separately).

After all in map-file I got:

C:/ti/mathlib_rts_c66x_3_1_2_1/packages/ti/mathlib/lib/mathlib_rts.ae66
asindp.oe66 1376 0 0
powdp.oe66 1088 0 0
atan2dp.oe66 992 0 0
powsp.oe66 704 0 0
log10dp.oe66 640 0 0
expdp.oe66 448 0 0
cosdp.oe66 416 0 0
sindp.oe66 384 0 0
divdp.oe66 160 0 0
divsp.oe66 160 0 0
sqrtdp.oe66 160 0 0
sqrtsp.oe66 160 0 0
tables.oe66 0 0 160
+--+--------------------------------+--------+---------+-----------+
Total: 6688 0 160

Q4: The map file still contains sin and other functions. I'm not confidence that it is work properly.

Q5: The provided file mathlib_rts.ae66 = 269 kB, the file with the library mathlib_rts_c66x_3_1_2_1 mathlib_rts.ae66 = 241 kB. This is suspicious.
With the provided file, the calculation in the float ceases to be performed.

Q6: How to understand that sin and other function from MATHLIB? The way with map file not convincing. My map file has both functions.

George

George Rodionov said:

Q1: Do I need to exclude math.h from the project now? Or I can have him stay.

No, you don`t need to remove math.h file.

George Rodionov said:

Q2: Do I need to go up all the c-files of the project in the "Link Order", according to the pdf instructions?

you are misinterpreting the build process. Link order is not done at file level, it is a project setting as linking step happens once after all files have been compiled so I don`t understand the need to set link order for all c files that you are referring to in the question.

George Rodionov said:

Q3: After completing the steps to connecting mathlib_rts.ae66, when going through the sin function and other mathematical functions, should they switch to MATHLIB files? This is not so.

Did you try to build the example project that I shared. This has been used by multiple users and confirmed to be working.

George Rodionov said:

The MATLIB download page has the mathlib_rts_c66x_3_1_2_1_Win32.exe library. The default option OVERRIDE_RTS = 1 is set in it. I used it and use provided file (of course separately).

I don`t agree with your observation. I did a fresh install of the version of MATHLIB that you referred and searched to confirm that the override_rts flag is set to 0 as you can see from screenshot below:

George Rodionov said:

After all in map-file I got:

C:/ti/mathlib_rts_c66x_3_1_2_1/packages/ti/mathlib/lib/mathlib_rts.ae66
asindp.oe66 1376 0 0
powdp.oe66 1088 0 0
atan2dp.oe66 992 0 0
powsp.oe66 704 0 0
log10dp.oe66 640 0 0
expdp.oe66 448 0 0
cosdp.oe66 416 0 0
sindp.oe66 384 0 0
divdp.oe66 160 0 0
divsp.oe66 160 0 0
sqrtdp.oe66 160 0 0
sqrtsp.oe66 160 0 0
tables.oe66 0 0 160
+--+--------------------------------+--------+---------+-----------+
Total: 6688 0 160

Q4: The map file still contains sin and other functions. I'm not confidence that it is work properly

If you are not confident that linking the library is working with your setup, then remove mathlib_rts.ae66 from the link step. You can directly add the required MATHLIB source files to your project. This approach takes advantage of the compiler link behavior where if a function is defined in the project then the API calls will not link to the RTS library. This is the easiest way to test the override functionality. 

George Rodionov said:

Q5: The provided file mathlib_rts.ae66 = 269 kB, the file with the library mathlib_rts_c66x_3_1_2_1 mathlib_rts.ae66 = 241 kB. This is suspicious.
With the provided file, the calculation in the float ceases to be performed.

Size of resulting binary will depend on the compiler version and the level of inlining and optimized code size it will generate.

George Rodionov said:

Q6: How to understand that sin and other function from MATHLIB? The way with map file not convincing. My map file has both functions.

Map file is the only feedback the compiler provides to all of the linked symbols in your application binary. MAthlib only contains a subset of the RTS library functions so you are likely to see object files from both libraries being used but for functions that do exist in mathlib the mathlib object files will override the functions in the RTS library.

Hope the responses help.

Rahul Prabhu said:

Did you try to build the example project that I shared. This has been used by multiple users and confirmed to be working.

You do not share example, only library

I will deal with the example, but it is not yet available

Q2: Do I need to add all the c-files of the project in the "Link Order", or not?

Rahul Prabhu said:

I don`t agree with your observation. I did a fresh install of the version of MATHLIB that you referred and searched to confirm that the override_rts flag is set to 0 as you can see from screenshot below:

I told about library  mathlib_rts_c66x_3_1_2_1_Win32.exe . I check it as you, get the next results:

Rahul Prabhu said:

If you are not confident that linking the library is working with your setup, then remove mathlib_rts.ae66 from the link step. You can directly add the required MATHLIB source files to your project. This approach takes advantage of the compiler link behavior where if a function is defined in the project then the API calls will not link to the RTS library. This is the easiest way to test the override functionality. 

How to do that? In your instruction has no description, and in TI instruction I can't find.

Could you share the example?

George

I still wait example.

George.

George,

Sorry, I wasn`t able to get to this last week. I have posted the example for your reference now. The example was originally created for AM572x device and if you want to use DDR and MSMC memory, just update the linker command file and update the cache setting if required (MAR bits to enable regions.)

AM57xx_C66x_Benhmark_using_MATHLIB.zip

Hope this helps.

Regards,

Rahul

How to understand wich function replace sin, cos and so on? In MATHLIB two type functions sp and dp - how to replace standart function with sp or dp - how can I switch between sp and dp?

George

In this document  Texas Instruments Test Results MATHLIB 3.1.2.1 c66x what does it mean RTC, C, Inline, Vector? Where is descriptions of test?

George

The test report is generated by running the code provided in test folder here:

mathlib_c66x_3_x_x_x\packages\ti\mathlib\src

In each folder, there is a _d file and function source provided with test input and output vectors that are used to check for correctness. Please review and let us know if you have further questions.

Regards

Rahul

I tried to execute cosdp_66_LE_ELF:
1. Add in workspace project C:\ti\mathlib_c66x_3_1_1_0\packages\ti\mathlib\src\cosdp\c66\cosdp_66_LE_ELF.

2. Select device 66AK2H14 and XDS2xx.

3. After try of compiling arise error

"C:/ti/mathlib_c66x_3_1_1_0/packages/ti/mathlib/src/cosdp/cosdp_d.c", line 149: error #20: identifier "DBL_MIN" is undefined"

Where defined DBL_MIN? It is double min value?

I set this value to 0 and get the next result:

--------------------------------------------------------------------------------
Verification Results: cosDP
--------------------------------------------------------------------------------
Pre-defined Data: Passed
Special Case Data: Passed
Extended Range Data: Passed
Random Data (seed = 7878): Passed
--------------------------------------------------------------------------------

--------------------------------------------------------------------------------
Cycle Profile: cosDP
--------------------------------------------------------------------------------
RTS: 254 cycles
ASM: 128 cycles
C: 128 cycles
Inline: 292 cycles
Vector: 30 cycles
--------------------------------------------------------------------------------

--------------------------------------------------------------------------------
Memory Profile: cosDP
--------------------------------------------------------------------------------
ASM: 32 bytes
C: 768 bytes
Vector: 448 bytes
--------------------------------------------------------------------------------

Is it correct?

George