Why a opcode can't using in CCS 3.3

 hello ,

 thank for  your reply!

Another interesting question is how do I use this addu instruction?

Whether it is no longer in use.

PS:

1.ADD —      Add  Two Signed Integers Without Saturation

2.ADDU —       Add  Two Unsigned Integers Without Saturation  

Good!

 I understand, thank you!

another question, can you help me ?

i want to finish a multiply:  -32768 x 32768.

so the code is follwing: 

mpy in1,in2,result
stnw in1,*out1++
stnw in2,*out1++   
stnw result,*out1++

I printf them :  printf("%x\n",out1[i]);

FFFF8000
8000
40000000

you know that is worng! why ?  

is that mpy can finish this task :Multiply Signed by Signed, 16 LSB × 16 LSB

Try

mpysu in1, in2, result

I found that I will not use the instructions: shr

WHEN  the code :

             shr A1,15,A1                       (A1 is a 32bit register)

The instructions will be added automatically according to the sign bit 0 or 1.

   BUT:when i codeing following :

       shr A2:A1,15,A2:A1

for example:

A2:A1:

oxffffffff   :0xfffffff =>0xff : 0x fffe1fff

0x2 :0x1 ====>0x0: 0x40000

The instructions are all added 0, regardless of the highest level of 64

The upper 24 bits of the high register are ignored when doing a 40-bit shift.  Only bit 39 matters when determining the sign of the result.

I get 64 - bit register set, the data obtained by MPY32 instructions, the following:

mpy32 in2,in1,result4:result3

Then, I need to shift the 64 - bit data.result4:result3 .

I don't know why the SHR only 48-bit effectively.

Can you give me some help,SIHFT 64 set of registers.

Thank you for your reply,

         You are so clever to solve the problem.

but If i consider the 64th sign-bit. i think the sifhter code:

           SHL     A5,0x1f,B4
           SHRU    A4,0x1,A3

           OR      B4,A3,A4
||        SHR    A5,0x1,A5

I have already tested it,Thank you again for your help!

Can I ask you another question?

1.how to implement subtraction with 64-signed bit.

Such as a model

         _SUB  A5:A4,A3:A2,A3:A2

2. in TMS320C6000 Integer Division  SPRA707 - October 2000 .

int sdiv(int num, int den) ,this function can complete 32-signed bit division.

but i want a function that a 64-signed bit are divided by a 32-signed bit. or  a 64-signed bit are divided by a 64-signed bit.

 I am doing assem.bly work In my company, at the same time I need your guidance(TI documents or some program code).

best  wishes to you and your family ！

I didn't think up that solution for 64-bit shift, I let the compiler do it for me.  I wrote:

unsigned long long func(unsigned long long x) { return x >> 1; }

and looked at what the compiler generated.  Similarly, we can do the same thing for subtraction:

long long func(long long x, long long y) { return x - y; }

The compiler generates:

           SUBU    A4,B4,A7:A6

           SUB     A5,B5,A3
||         EXT     A7,24,24,A5

           ADD     A3,A5,A5 
||         MV      A6,A4  

longya lang said:

 a 64-signed bit are divided by a 64-signed

Call __c6xabi_divlli (__divlli in COFF mode) instead of writing code to do so, it's what the compiler will do.

I think you taught me a method of learning rather than merely a knowledge. I will study hard and make use of C6x compiler. Thank you for your guidance!

Another question: imath64.obj (.text:__c6xabi_divlli) .Is this funtion  defined in imath64.c file?And in which lib? 

This function may be in the rts64plus.lib.i think.

 Here is the thing.

Earlier, I used the function _divlli() in the mian. C file ，and call it sucessfully.I can get the right result.

Then I tried to call the _divlli() in the assembly file. but failed！(The figure below)
In addition, in that assemble file，the function sdiv（）which is my-coding function can using successfuly！

i don't know what happened? "undefined   symbol" Means that the lack of libraries, but when I called _divllli() in C file,we didn't meet with that problem .i think this function is in rts64plus.lib. so,

 i copy the function code from imath64.c to main.c.and call it in assmble file.the compiler print: __divlli is defined multiple times. i know all the thing happen in the linking section .

So,in the assembly code,How can i call a  function which is in a .lib file . For example, _devlli() in rts64plus.lib.

PS:

this is my function for 64-signed bit DIV.

long long divlli64(long long a, long long b)
 {
  /*-----------------------------------------------------------------------*/
   /* CHECK SIGNS, TAKE ABSOLUTE VALUE, AND USED UNSIGNED DIVIDE.           */
   /*-----------------------------------------------------------------------*/
   long long sign  = (a^b)>>63;    
  unsigned long long ua=(a==LLONG_MIN?a:llabs(a)); 
  unsigned long long ub=(b==LLONG_MIN?b:llabs(b));
    unsigned long long q =_divull(ua,ub);

   if (b == 0)
       return a?(((unsigned long long)-1) >> 1)^sign : 0;
                         /* saturation value or 0 */
   return sign?-q:q;
 }

but i can't using it ,when i pass a 64-bit para to  it. the following is the code:

 .call in6=_divlli64(in3:in4,in1:in2)

the compiler print: >> internal error: Missing symbol table entry for in3:in4

(The figure of the project is at the bottom)

Hello Archaeologist!

In coding some assembly  file ,i have encountered some trouble: 64-bit div.for example,using the _divlli() which you told me yesterday in my  assembly code.

These problemsI have already submitted in the following URL.

http://e2e.ti.com/support/development_tools/compiler/f/343/t/287024.aspx

You need the information contained in the TMS320C6000 Optimizing Compiler v7.4 User's Guide (SPRU187U) http://www.ti.com/lit/ug/spru187u/spru187u.pdf section 7.5 "Interfacing C and C++ With Assembly Language", section 6.12 "Object File Symbol Naming Conventions (Linknames)," and chapter 5 "Linking C/C++ Code"

Dear my friend!

                 Thank you for all your help these days!

                I have already completed  the task , transforming the fixed-point floating-point program  to a fixed-point  program.Under the guidance of you, I was able to complete. Thank you very much.

 PS: the last probelom,The last question does the problem of naming convention.