FPU single precision add inaccuracy
Created by: jrrk
I am testing the improved, pipelined FPU under ariane-dev branch. I am trying to run TestFloat and in the main it works well, but I am having problems with certain particular operands, as per the attached testcase:
TEST_FP_OP2_S_( 2, fadd.s, 33354290, 33357822, -3531 .952392578125 ); # actually returns 33357822 TEST_FP_OP2_S_( 3, fadd.s, -1.701411733192644E+38, -1.701411733192644E+38, -2. 350970765024232E-38 ); # actually returns NAN
I can add more test vectors later if needed