Let me preface this by saying I know very little about quadratic forms and most of what I know is about quadratic forms in two variables, whilst this question is about a quadratic form in three variables.
Let $n\ge 1$ (Actually, $n> 1$, to make it interesting, see the edit) be a positive integer, $n\equiv 1\pmod 8$. Then there exist $x,y,z\in\mathbb{Z}$ $$ n = x^2 + 4y^2+4z^2. $$ Question: Can we always find such $(x,y,z)$ with $x\equiv \pm 3\pmod 8$?
My attempt: Suppose $n\equiv 9\pmod{16}$ and $n$ is represented over the integers by $u^2+16v^2+16w^2$. Then we can take $(x,y,z)=(u,2v,2w)$. Now, according to this, the form $u^2+16v^2+16w^2$ is regular, which I believe means such a representation of $n$ does indeed exist.
For $n\equiv 1\pmod{16}$, the existence of a representation with $x\equiv \pm 3\pmod{8}$ is equivalent to $n$ being represented over the integers by $$ u^2 + (4v+2u)^2 + (4w+2u)^2 = 9u^2+16v^2+16w^2+16uv+16uw. $$ However, I know basically nothing about this quadratic form. Its discriminant is $256$ and does not show up in the aforementioned list of regular forms.
Any pointers in the right direction are appreciated.
Edit (added after ThibautOphelia's answer): There is no such solution when $n=1$, but I'm mainly interested in the case $n>1$.
I have most of it. All that remains are certain squares.
The other form in the genus of $f=x^2 + 16 y^2 + 16 z^2$ has reduced representative $g = 4 x^2 + 9 y^2 + 9 z^2 + 2 y z + 4zx + 4xy.$ In Jones and Pall (1939), this is written as (4,9,9,1,2,2) on the last page, as they halve the last three coefficients.
The form $f$ is regular, and does take care of $9 \pmod{16}.$ ADDED: meanwhile, if a square is $9 \pmod{16}$ then its square root is an integer and is $\pm 3 \pmod{16}$
The form $g$ is what we call spinor regular. It represents the same things as $f$ with the exception of certain squares; it does not represent any $m^2$ when no prime $q \equiv 3 \pmod 4$ divides $m.$ But it does represent all other numbers $n \equiv 1 \pmod{16},$ including any such number that is not a square, or a square that is divisible by some $q \equiv 3 \pmod 4.$
We take such $n$ as $$ n = 4 x^2 + 9 y^2 + 9 z^2 + 2 y z + 4zx + 4xy. $$
We may write this using what Kap called a homothety, $$ n = (2x+y+z)^2 + 4 (y-z)^2 + 4 (y+z)^2 $$ Now: since $n$ is odd, we know $2x+y+z$ is odd, and so are $y \pm z$ But then $(y \pm z)^2 \equiv 1 \pmod 8,$ next $$4(y \pm z)^2 \equiv 4 \pmod {32},$$ then $4(y - z)^2 + 4(y + z)^2 \equiv 8 \pmod {32},$ so $$4(y - z)^2 + 4(y + z)^2 \equiv 8 \pmod {16}.$$
It now follows from $$ n = (2x+y+z)^2 + 4 (y-z)^2 + 4 (y+z)^2 \equiv 1 \pmod{16}$$ that $(2x+y+z)^2 \equiv 9 \pmod{16}$ and $$ 2x+y+z \equiv \pm 3 \pmod 8 $$
To repeat, your desired representation is always possible when $n \equiv 1 \pmod 8$ is not a spinor exception for the genus of $x^2 + 16 y^2 + 16 z^2$
Jones and Pall(1939)
Earnest and Haensch, preprint at arxiv