5556 – [64-bit] Wrong Implicit Conversion to Double

The following weird result happens only with -m64 enabled. To reproduce the bug from this test case, -inline must be disabled to prevent the two functions from being inlined, though in the more complex cases that I reduced it from inlining could be enabled.
import std.stdio;
float fun(float[] data, sizediff_t k) {
 stderr.writeln("Returning ", data[k]); // Prints "Returning 2"
 return data[k];
}
double gun(float[] data) {
 return fun(data, data.length / 2);
}
void main() {
 float[] test = [1,4,2,3,5];
 stderr.writeln(gun(test)); // Prints 5.30499e-215
}

https://github.com/D-Programming-Language/dmd/commit/e5b63251d2c0fb4d2aa7929d3560e0f4d42be458
https://github.com/D-Programming-Language/dmd/commit/bc29f3797e695fe7ff3ad5da5f3c5bc6e43a84da