error_demo1.c 1.0e10f*((1.0f+1.0e-7f)-1.0f) 
only about 7 digits accuracy, power of 10 not exact 
ans= 1.192093e+03 
force store, break optimization, no difference here 
ans= 1.192093e+03 

expression with float  ans=1192.09 
expression with double ans=1000 

error_demo1.c (1.0f+0.5e-7f)+1.0f  too small 
ans= 0.000000e+00 
force store, break optimization, no difference here 
ans= 0.000000e+00