Test case: ---------------------- void main() { auto p = -0.0; assert(p == 0); } ---------------------- Running the program with -m64 caused an Assertion failure. (This works in 32-bit.) An equivalent test which also fails in 64-bit: ---------------------- void main() { double p = -0.0; assert(p == -0.0); } ----------------------
The problem is that, when comparing with 0.0 or -0.0, the backend will generate an integer 'cmp' instruction, but -0.0 has a different bit pattern than 0.0, so the equality will fail. cmp QWORD PTR [rbp-0x8],0x0 je ... The code works if the type is 'float' because it uses an 'add eax, eax' trick make both 0.0f and -0.0f set the ZF flag. This code also work if the type if 'real' because it uses x87.
I think 0.0 == -0.0, but 0.0 !is -0.0.
Commit pushed to master at https://github.com/D-Programming-Language/dmd https://github.com/D-Programming-Language/dmd/commit/a8ffc8ab60aa3e1ae965e5764f0900e14a0881c1 Fix issue 7546 64-bit floating-point issue with negative zero: -0.0 == 0.0 is false Just duplicate the code for float == float.