blob: 4550e4c4116b5a956dc2a266a9ac5c721e7c4659 [file] [log] [blame]
struct VertexOutput {
float4 position;
float4 color;
float2 quad_pos;
};
struct VertexInput {
float3 position;
float4 color;
float2 quad_pos;
};
struct Particle {
float3 position;
float lifetime;
float4 color;
float2 velocity;
};
struct vertex_main_outputs {
float4 tint_symbol_4 : SV_Position;
};
struct vs_main_outputs {
float4 VertexOutput_color : TEXCOORD0;
float2 VertexOutput_quad_pos : TEXCOORD1;
float4 VertexOutput_position : SV_Position;
};
struct vs_main_inputs {
float3 VertexInput_position : TEXCOORD0;
float4 VertexInput_color : TEXCOORD1;
float2 VertexInput_quad_pos : TEXCOORD2;
};
struct simulate_inputs {
uint3 GlobalInvocationID : SV_DispatchThreadID;
};
struct export_level_inputs {
uint3 coord : SV_DispatchThreadID;
};
static float2 rand_seed = (0.0f).xx;
cbuffer cbuffer_render_params : register(b5) {
uint4 render_params[6];
};
cbuffer cbuffer_sim_params : register(b0) {
uint4 sim_params[2];
};
RWByteAddressBuffer data : register(u1);
Texture1D<float4> tint_symbol_2 : register(t2);
cbuffer cbuffer_ubo : register(b3) {
uint4 ubo[1];
};
ByteAddressBuffer buf_in : register(t4);
RWByteAddressBuffer buf_out : register(u5);
Texture2D<float4> tex_in : register(t6);
RWTexture2D<float4> tex_out : register(u7);
void asinh_468a48() {
float16_t arg_0 = float16_t(0.0h);
float16_t v = arg_0;
float16_t res = log((v + sqrt(((v * v) + float16_t(1.0h)))));
}
float4 vertex_main_inner() {
asinh_468a48();
return (0.0f).xxxx;
}
void fragment_main() {
asinh_468a48();
}
[numthreads(1, 1, 1)]
void rgba32uintin() {
asinh_468a48();
}
float4x4 v_1(uint start_byte_offset) {
float4 v_2 = asfloat(render_params[(start_byte_offset / 16u)]);
float4 v_3 = asfloat(render_params[((16u + start_byte_offset) / 16u)]);
float4 v_4 = asfloat(render_params[((32u + start_byte_offset) / 16u)]);
return float4x4(v_2, v_3, v_4, asfloat(render_params[((48u + start_byte_offset) / 16u)]));
}
VertexOutput vs_main_inner(VertexInput tint_symbol) {
float3 v_5 = asfloat(render_params[4u].xyz);
float3 quad_pos = mul(tint_symbol.quad_pos, float2x3(v_5, asfloat(render_params[5u].xyz)));
float3 position = (tint_symbol.position - (quad_pos + 0.00999999977648258209f));
VertexOutput tint_symbol_1 = (VertexOutput)0;
float4x4 v_6 = v_1(0u);
tint_symbol_1.position = mul(float4(position, 1.0f), v_6);
tint_symbol_1.color = tint_symbol.color;
tint_symbol_1.quad_pos = tint_symbol.quad_pos;
VertexOutput v_7 = tint_symbol_1;
return v_7;
}
void v_8(uint offset, Particle obj) {
data.Store3((offset + 0u), asuint(obj.position));
data.Store((offset + 12u), asuint(obj.lifetime));
data.Store4((offset + 16u), asuint(obj.color));
data.Store2((offset + 32u), asuint(obj.velocity));
}
Particle v_9(uint offset) {
float3 v_10 = asfloat(data.Load3((offset + 0u)));
float v_11 = asfloat(data.Load((offset + 12u)));
float4 v_12 = asfloat(data.Load4((offset + 16u)));
Particle v_13 = {v_10, v_11, v_12, asfloat(data.Load2((offset + 32u)))};
return v_13;
}
void simulate_inner(uint3 GlobalInvocationID) {
float2 v_14 = asfloat(sim_params[1u]).xy;
float2 v_15 = (v_14 * float2(GlobalInvocationID.xy));
rand_seed = (v_15 * asfloat(sim_params[1u]).zw);
uint idx = GlobalInvocationID[0u];
Particle particle = v_9((0u + (uint(idx) * 48u)));
uint v_16 = (uint(idx) * 48u);
Particle v_17 = particle;
v_8((0u + v_16), v_17);
}
void export_level_inner(uint3 coord) {
uint2 v_18 = (0u).xx;
tex_out.GetDimensions(v_18[0u], v_18[1u]);
if (all((coord.xy < uint2(v_18)))) {
uint dst_offset = (coord[0u] << ((coord[1u] * ubo[0u].x) & 31u));
uint src_offset = ((coord[0u] - 2u) + ((coord[1u] >> (2u & 31u)) * ubo[0u].x));
float a = asfloat(buf_in.Load((0u + (uint((src_offset << (0u & 31u))) * 4u))));
float b = asfloat(buf_in.Load((0u + (uint((src_offset + 1u)) * 4u))));
float c = asfloat(buf_in.Load((0u + (uint(((src_offset + 1u) + ubo[0u].x)) * 4u))));
float d = asfloat(buf_in.Load((0u + (uint(((src_offset + 1u) + ubo[0u].x)) * 4u))));
float sum = dot(float4(a, b, c, d), (1.0f).xxxx);
uint v_19 = (uint(dst_offset) * 4u);
float v_20 = (sum / 4.0f);
float v_21 = floor(v_20);
buf_out.Store((0u + v_19), asuint((sum - ((((v_20 < 0.0f)) ? (ceil(v_20)) : (v_21)) * 4.0f))));
float4 v_22 = float4(a, (a * b), ((a / b) + c), sum);
float4 probabilities = (v_22 + max(sum, 0.0f));
tex_out[int2(coord.xy)] = probabilities;
}
}
vertex_main_outputs vertex_main() {
vertex_main_outputs v_23 = {vertex_main_inner()};
return v_23;
}
vs_main_outputs vs_main(vs_main_inputs inputs) {
VertexInput v_24 = {inputs.VertexInput_position, inputs.VertexInput_color, inputs.VertexInput_quad_pos};
VertexOutput v_25 = vs_main_inner(v_24);
vs_main_outputs v_26 = {v_25.color, v_25.quad_pos, v_25.position};
return v_26;
}
[numthreads(64, 1, 1)]
void simulate(simulate_inputs inputs) {
simulate_inner(inputs.GlobalInvocationID);
}
[numthreads(64, 1, 1)]
void export_level(export_level_inputs inputs) {
export_level_inner(inputs.coord);
}