blob: a9909fbe8060909dab6505257909fc44bb5974e7 [file] [log] [blame]
struct vert_main_outputs {
float4 tint_symbol : SV_Position;
};
struct vert_main_inputs {
float2 a_particlePos : TEXCOORD0;
float2 a_particleVel : TEXCOORD1;
float2 a_pos : TEXCOORD2;
};
struct frag_main_outputs {
float4 tint_symbol_1 : SV_Target0;
};
struct comp_main_inputs {
uint3 gl_GlobalInvocationID : SV_DispatchThreadID;
};
cbuffer cbuffer_params : register(b0) {
uint4 params[2];
};
RWByteAddressBuffer particlesA : register(u1);
RWByteAddressBuffer particlesB : register(u2);
float4 vert_main_inner(float2 a_particlePos, float2 a_particleVel, float2 a_pos) {
float angle = -(atan2(a_particleVel.x, a_particleVel.y));
float2 pos = float2(((a_pos.x * cos(angle)) - (a_pos.y * sin(angle))), ((a_pos.x * sin(angle)) + (a_pos.y * cos(angle))));
return float4((pos + a_particlePos), 0.0f, 1.0f);
}
float4 frag_main_inner() {
return (1.0f).xxxx;
}
void comp_main_inner(uint3 gl_GlobalInvocationID) {
uint index = gl_GlobalInvocationID.x;
if ((index >= 5u)) {
return;
}
float2 vPos = asfloat(particlesA.Load2((0u + (uint(index) * 16u))));
float2 vVel = asfloat(particlesA.Load2((8u + (uint(index) * 16u))));
float2 cMass = (0.0f).xx;
float2 cVel = (0.0f).xx;
float2 colVel = (0.0f).xx;
int cMassCount = int(0);
int cVelCount = int(0);
float2 pos = (0.0f).xx;
float2 vel = (0.0f).xx;
{
uint i = 0u;
while(true) {
if ((i < 5u)) {
} else {
break;
}
if ((i == index)) {
{
i = (i + 1u);
}
continue;
}
pos = asfloat(particlesA.Load2((0u + (uint(i) * 16u)))).xy;
vel = asfloat(particlesA.Load2((8u + (uint(i) * 16u)))).xy;
if ((distance(pos, vPos) < asfloat(params[0u].y))) {
cMass = (cMass + pos);
cMassCount = (cMassCount + int(1));
}
if ((distance(pos, vPos) < asfloat(params[0u].z))) {
colVel = (colVel - (pos - vPos));
}
if ((distance(pos, vPos) < asfloat(params[0u].w))) {
cVel = (cVel + vel);
cVelCount = (cVelCount + int(1));
}
{
i = (i + 1u);
}
continue;
}
}
if ((cMassCount > int(0))) {
float2 v = cMass;
float v_1 = float(cMassCount);
float2 v_2 = (v / float2(v_1, float(cMassCount)));
cMass = (v_2 - vPos);
}
if ((cVelCount > int(0))) {
float2 v_3 = cVel;
float v_4 = float(cVelCount);
cVel = (v_3 / float2(v_4, float(cVelCount)));
}
vVel = (((vVel + (cMass * asfloat(params[1u].x))) + (colVel * asfloat(params[1u].y))) + (cVel * asfloat(params[1u].z)));
vVel = (normalize(vVel) * clamp(length(vVel), 0.0f, 0.10000000149011611938f));
vPos = (vPos + (vVel * asfloat(params[0u].x)));
if ((vPos.x < -1.0f)) {
vPos.x = 1.0f;
}
if ((vPos.x > 1.0f)) {
vPos.x = -1.0f;
}
if ((vPos.y < -1.0f)) {
vPos.y = 1.0f;
}
if ((vPos.y > 1.0f)) {
vPos.y = -1.0f;
}
uint v_5 = (uint(index) * 16u);
particlesB.Store2((0u + v_5), asuint(vPos));
uint v_6 = (uint(index) * 16u);
particlesB.Store2((8u + v_6), asuint(vVel));
}
vert_main_outputs vert_main(vert_main_inputs inputs) {
vert_main_outputs v_7 = {vert_main_inner(inputs.a_particlePos, inputs.a_particleVel, inputs.a_pos)};
return v_7;
}
frag_main_outputs frag_main() {
frag_main_outputs v_8 = {frag_main_inner()};
return v_8;
}
[numthreads(1, 1, 1)]
void comp_main(comp_main_inputs inputs) {
comp_main_inner(inputs.gl_GlobalInvocationID);
}