Compiling without Make about coremark-pro

[petertorelli]

It looks like you've succeeded in porting it. I'm pretty sure you are encountering errors regarding floating point accuracy since the loops test is a very accuracy sensitive test, and some compiler options can impact this. Try enabling BMDEBUG and DEBUG_ACCURATE_BITS. This will print out the values that failed the bit-accuracy test. You may need to experiment with the clang floating-point options.

[muke101]

I've tried to look into this and it seems like you're right, but I can't find how to fix the problem. Although for my purposes I don't actually need the results to be correct, I just need to run the program - are you able to comment at all on how if at all the floating point accuracy might effect the behavior of the program?

[petertorelli]

You can alter MIN_ACC_BITS_FP32 and MIN_ACC_BITS_FP64 to reduce the number of bits compared. Here is an example of how it is used in th_bignum.c:

void acc_summary_info(intparts *sig, intparts *ref, e_u32 accbits) {
	svals(&acc_summary.sig_exp,sig->exp);
	svals(&acc_summary.ref_exp,ref->exp);
	svals(&acc_summary.bits,accbits);
	if (accbits<MIN_ACC_BITS_FP)
		acc_summary.counts[accbits]++;
	else
		acc_summary.counts[MAX_ACC_COUNTS]++;
}