Linear Assembly For Floating-Point Dot Product Inner Loop With Lddw - Texas Instruments TMS320C6000 Programmer's Manual

6.4.2.2 Floating-Point Dot Product
Example 6–16. Linear Assembly for Floating-Point Dot Product Inner Loop with LDDW
LDDW
LDDW
MPYSP
MPYSP
ADDSP
ADDSP
[cntr] SUB
[cntr] B
Using Word Access for Short Data and Doubleword Access for Floating-Point Data
Two MPY instructions are now necessary to multiply the second set of array
elements:
The first MPY instruction multiplies the 16 least significant bits (LSBs) in
each source register: a[i]
The MPYH instruction multiplies the 16 most significant bits (MSBs) of
each source register: a[i+1]
The two ADD instructions accumulate the sums of the even and odd elements:
sum0 and sum1.
Note:
This is true only when the 'C6x is in little-endian mode. In big-endian mode,
MPY operates on a[i+1] and b[i+1] and MPYH operates on a[i] and b[i]. See
the TMS320C6000 Peripherals Reference Guide for more information.
Example 6–16 shows the list of 'C6x instructions that execute the unrolled
floating-point dot product loop. Symbolic variable names are used instead of
actual registers. Using symbolic names for data and pointers makes code eas-
ier to write and allows the optimizer to allocate registers. However, you must
use the .reg assembly optimizer directive. See the TMS320C6000 Optimizing
C/C++ Compiler User's Guide for more information on writing linear assembly
code.
*a++,ai1:ai0 ; load a[i+0] & a[i+1] from memory
*b++,bi1:bi0 ; load b[i+0] & b[i+1] from memory
ai0,bi0,pi0 ; a[i+0] * b[i+0]
ai1,bi1,pi1 ; a[i+1] * b[i+1]
pi0,sum0,sum0 ; sum0 += (a[i+0] * b[i+0])
pi1,sum1,sum1 ; sum1 += (a[i+1] * b[i+1])
cntr,1,cntr ; decrement loop counter
LOOP ; branch to loop
The two load doubleword (LDDW) instructions load a[i], a[i+1], b[i], and b[i+1]
on each iteration.
Two MPYSP instructions are now necessary to multiply the second set of array
elements.
The two ADDSP instructions accumulate the sums of the even and odd
elements: sum0 and sum1.
b[i].
b [i+1].
Optimizing Assembly Code via Linear Assembly 6-21