There is a problem when llvm IR generate an unnamed temporary value inside another llvm IR instruction
es:
result:
%tmp2.flt = load double, double* getelementptr inbounds ([100000 x double], [100000 x double]* @array, i64 0, i64 0), align 16, !taffo.initweight !14, !taffo.info !1
expected:
%tmp2.flt = load double, double* getelementptr inbounds ([100000 x double], [100000 x double]* @array.fixp, i64 0, i64 0), align 16, !taffo.initweight !14, !taffo.info !1

So when the instruction is converted FloatToFixed::fallback (here) is called
then a call to fallbackMatchValue (here) is performed which dosen't change the operand because TypeConversion is not set for the GEP
array.zip