Champs,
One of my customers enabled the Local (-o1) opt level in the CCS3.3 to save flash space but encountered with an error result issue with the same C code.
The error comes from below statement:
iq10Temp.lData = _IQ10mpy(iq10Temp.lData,g_iq10PfcVolt.lData);
With P = iq10Temp.IData, the assembly code after optimization is:
1074 000001e0 761F- MOVW DP,#_g_iq10PfcVolt
000001e1 0000
1075 000001e2 A9AC MOVL XT,P
1076 000001e3 5605- IMPYL P,XT,@_g_iq10PfcVolt ; |435|
000001e4 0004
1077 000001e5 A9AC MOVL XT,P ; |435|
1078 000001e6 5663- QMPYL ACC,XT,@_g_iq10PfcVolt ; |435|
000001e7 0004
1079 000001e8 5689 ASR64 ACC:P,#10 ; |435|
Obviously, there insert an additional instruction markered with red MOVL XT,P , which leads to the incorrect multiply result because the desired non-variable XT has been modified after the red unexpected instruction.
Here comes the question, why this instruction added in the assembly code? How does the compiler do to this _IQ10mpy statement when local opt level enabled?
If the auxiliary register P is replaced by other registers, such as XAR4, this issue will also not exist, then what is the principle on allocating auxiliary register?
Thanks for your suggestion.
Best regards,
Ricky Zhang
