Should you hardcode texture coordinates array in vertex shader when possible? - performance

Question

So i have made a voxel engine where instead of passing texture coordinates array to VAO, i hardcode that array in vertex shader:
const vec2 texcoords[4] = vec2[4](
const vec2(1.0, 1.0),
const vec2(1.0, 0.0),
const vec2(0.0, 0.0),
const vec2(0.0, 1.0)
);
And instead i pass vertexIndex to VAO (so vertexIndex determines which index of this array to use).
so it saves me up a LOT of memory, i could fit every info about each voxel in single uint. but i want to know what downsides does this method have, should i use this when memory profit is not as much?
EDIT:
here is code for chunk.vert:
#version 460 core
in uint vertex;
out vec2 pass_texcoord;
out float pass_faceLight;
uniform mat4 projection;
uniform mat4 view;
uniform uint texAtlasRowsCount;
uniform vec3 chunkPos;
#define vertexIndex (vertex & 0x600000) >> 21
#define xc (vertex & 0x3F)
#define yc ((vertex & 0xFC0) >> 6)
#define zc ((vertex & 0x3F000) >> 12)
#define light ((vertex & 0x1C0000) >> 18) * 0.2F
#define pos vec4(vec3(xc, yc, zc) + chunkPos, 1)
#define position projection * view * pos
const vec2 texcoords[4] = vec2[4](
const vec2(1.0, 1.0),
const vec2(1.0, 0.0),
const vec2(0.0, 0.0),
const vec2(0.0, 1.0)
);
void main() {
uint row = ((vertex & 0x3F800000) >> 23) - 1;
const uint col = row / texAtlasRowsCount;
row %= texAtlasRowsCount;
pass_texcoord = texcoords[vertexIndex];
pass_texcoord.x += row;
pass_texcoord.y += col;
pass_texcoord /= texAtlasRowsCount;
//float AmbientOclussion = float((vertex & 0xC0000000) >> 30);
pass_faceLight = light;
gl_Position = position;
}