d3dx9: Implement D3DXFloat16to32Array and D3DXFloat32to16Array.

[Stefan Dösinger]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


Am 07.02.2011 um 03:37 schrieb Misha Koshelev:
> +    FLOAT full[10] = { 1.0f, -2.0f, 6.55e4, 6.10352e-5, 5.96046e-8, 0.0f, -0.0f, INFINITY, -INFINITY, NAN },
> +        full_exp[10] = { 1.0f, -2.0f, 6.55e4, 6.10352e-5, 5.96046e-8, 0.0f, -0.0f, 65536.0f, -131008.0f, 131008.0f },
> +        full_res[10];
I'd rather call them "single" instead of full because 32 bit floats are usually referred to as single precision(doubles are double precision, and 16 bit floats half precision)

> math.c:2238: Test failed: Got 7fff, expected 7c00 for index 7.
Looks like Windows doesn't do INF, and returns NaN instead. Your -INF value test already expects NaN as return value, but with the sign bit set. INF or -INF should have all mantissa bits set to 0. NaN looks like +/- INF, just with mantissa != 0(so there are many possible encodings of NaN).

However, it is also possible that Windows doesn't support those special values at all. This is suggested by the fact that +/- NaN is happily converted into a proper single precision value.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG/MacGPG2 v2.0.17 (Darwin)
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=EhgD
-----END PGP SIGNATURE-----