#version 450

const int MAX_LIGHTS = 4;

struct Light {
    vec4 direction;
    vec4 color;
};

layout(location = 0) in vec3 inPosition;
layout(location = 1) in vec3 inNormal;
layout(location = 2) in vec2 inUV;

layout(location = 0) out vec2 fragUV;
layout(location = 1) out vec3 fragNormal;
layout(location = 2) out vec3 fragWorldPos;
layout(location = 3) out vec3 fragLightSpacePos;
layout(location = 4) out vec2 fragVelocity;   // TAA：clip 空间运动向量

// GPUDriven：模型矩阵从实例缓冲按实例索引取。
// 同一 mesh 的 N 个实体 = 1 次 drawIndexed(instanceCount=N)，
// draw 侧用 firstInstance 指向该 batch 在实例数组中的起始下标。
layout(std430, set = 0, binding = 1) readonly buffer ObjectBuffer {
    mat4 models[];   // 连续 mat4 数组（16B 对齐，std430/std140 布局一致）
} objects;

// 上一帧实例矩阵（与 ObjectBuffer 同下标 = 同一实例的上一帧姿态）
layout(std430, set = 0, binding = 2) readonly buffer PrevObjectBuffer {
    mat4 prevModels[];
} prevObjects;

layout(set = 0, binding = 0) uniform UniformBufferObject {
    mat4 view;
    mat4 proj;              // 抖动后（光栅化）
    mat4 lightSpace;
    vec4 ambientColor;
    Light lights[MAX_LIGHTS];
    mat4 projNoJitter;      // 未抖动（velocity 当前帧）
    mat4 prevView;          // velocity 上一帧
    mat4 prevProj;
} ubo;

void main() {
    mat4 model = objects.models[gl_InstanceIndex];

    vec4 worldPos = model * vec4(inPosition, 1.0);
    gl_Position = ubo.proj * ubo.view * worldPos;

    // ---- TAA velocity：当前帧与上一帧的 clip 空间位置差（UV 空间）----
    // 抖动只进光栅化 gl_Position，不进 velocity（否则抖动本身被当成运动）
    vec4 currClip = ubo.projNoJitter * ubo.view * worldPos;
    vec4 prevWorldPos = prevObjects.prevModels[gl_InstanceIndex] * vec4(inPosition, 1.0);
    vec4 prevClip = ubo.prevProj * ubo.prevView * prevWorldPos;
    fragVelocity = (currClip.xy / currClip.w - prevClip.xy / prevClip.w) * 0.5;

    fragWorldPos = worldPos.xyz;
    // 光空间坐标（含 Vulkan [0,1] 深度重映射，与 shadow pass 写入端共用同一矩阵）
    fragLightSpacePos = (ubo.lightSpace * worldPos).xyz;

    mat3 normalMatrix = mat3(transpose(inverse(model)));
    fragNormal = normalize(normalMatrix * inNormal);

    fragUV = inUV;
}