#include "rendering/sdl3gpu/sdl3gpu_shader.hpp"

#include <SDL3/SDL_log.h>

#include <algorithm>  // std::min, std::max, std::floor
#include <cmath>      // std::floor
#include <cstring>    // memcpy, strlen

#ifndef __APPLE__
#include "rendering/sdl3gpu/postfx_frag_spv.h"
#include "rendering/sdl3gpu/postfx_vert_spv.h"
#endif

#ifdef __APPLE__
// ============================================================================
// MSL shaders (Metal Shading Language) — macOS
// ============================================================================

// NOLINTBEGIN(readability-identifier-naming)
static const char* POSTFX_VERT_MSL = R"(
#include <metal_stdlib>
using namespace metal;

struct PostVOut {
    float4 pos [[position]];
    float2 uv;
};

vertex PostVOut postfx_vs(uint vid [[vertex_id]]) {
    const float2 positions[3] = { {-1.0, -1.0}, {3.0, -1.0}, {-1.0, 3.0} };
    const float2 uvs[3]       = { { 0.0,  1.0}, {2.0,  1.0}, { 0.0,-1.0} };
    PostVOut out;
    out.pos = float4(positions[vid], 0.0, 1.0);
    out.uv  = uvs[vid];
    return out;
}
)";

static const char* POSTFX_FRAG_MSL = R"(
#include <metal_stdlib>
using namespace metal;

struct PostVOut {
    float4 pos [[position]];
    float2 uv;
};

struct PostFXUniforms {
    float vignette_strength;
    float chroma_strength;
    float scanline_strength;
    float screen_height;
    float mask_strength;
    float gamma_strength;
    float curvature;
    float bleeding;
    float pixel_scale;
    float time;
    float oversample;  // 1.0 = sin SS, 3.0 = 3× supersampling
    float flicker;     // 0 = off, 1 = phosphor flicker ~50 Hz
};

// YCbCr helpers for NTSC bleeding
static float3 rgb_to_ycc(float3 rgb) {
    return float3(
         0.299f*rgb.r + 0.587f*rgb.g + 0.114f*rgb.b,
        -0.169f*rgb.r - 0.331f*rgb.g + 0.500f*rgb.b + 0.5f,
         0.500f*rgb.r - 0.419f*rgb.g - 0.081f*rgb.b + 0.5f
    );
}
static float3 ycc_to_rgb(float3 ycc) {
    float y  = ycc.x;
    float cb = ycc.y - 0.5f;
    float cr = ycc.z - 0.5f;
    return clamp(float3(
        y + 1.402f*cr,
        y - 0.344f*cb - 0.714f*cr,
        y + 1.772f*cb
    ), 0.0f, 1.0f);
}

fragment float4 postfx_fs(PostVOut                 in    [[stage_in]],
                           texture2d<float>         scene [[texture(0)]],
                           sampler                  samp  [[sampler(0)]],
                           constant PostFXUniforms& u     [[buffer(0)]]) {
    float2 uv = in.uv;

    // Curvatura barrel CRT
    if (u.curvature > 0.0f) {
        float2 c = uv - 0.5f;
        float rsq = dot(c, c);
        float2 dist = float2(0.05f, 0.1f) * u.curvature;
        float2 barrelScale = 1.0f - 0.23f * dist;
        c += c * (dist * rsq);
        c *= barrelScale;
        if (abs(c.x) >= 0.5f || abs(c.y) >= 0.5f) {
            return float4(0.0f, 0.0f, 0.0f, 1.0f);
        }
        uv = c + 0.5f;
    }

    // Muestra base
    float3 base = scene.sample(samp, uv).rgb;

    // Sangrado NTSC — difuminado horizontal de crominancia.
    // step = 1 pixel de juego en espacio UV (corrige SS: scene.get_width() = game_w * oversample).
    float3 colour;
    if (u.bleeding > 0.0f) {
        float tw    = float(scene.get_width());
        float step  = u.oversample / tw;  // 1 pixel lógico en UV
        float3 ycc    = rgb_to_ycc(base);
        float3 ycc_l2 = rgb_to_ycc(scene.sample(samp, uv - float2(2.0f*step, 0.0f)).rgb);
        float3 ycc_l1 = rgb_to_ycc(scene.sample(samp, uv - float2(1.0f*step, 0.0f)).rgb);
        float3 ycc_r1 = rgb_to_ycc(scene.sample(samp, uv + float2(1.0f*step, 0.0f)).rgb);
        float3 ycc_r2 = rgb_to_ycc(scene.sample(samp, uv + float2(2.0f*step, 0.0f)).rgb);
        ycc.yz = (ycc_l2.yz + ycc_l1.yz*2.0f + ycc.yz*2.0f + ycc_r1.yz*2.0f + ycc_r2.yz) / 8.0f;
        colour = mix(base, ycc_to_rgb(ycc), u.bleeding);
    } else {
        colour = base;
    }

    // Aberración cromática (drift animado con time para efecto NTSC real)
    float ca = u.chroma_strength * 0.005f * (1.0f + 0.15f * sin(u.time * 7.3f));
    colour.r = scene.sample(samp, uv + float2(ca, 0.0f)).r;
    colour.b = scene.sample(samp, uv - float2(ca, 0.0f)).b;

    // Corrección gamma (linealizar antes de scanlines, codificar después)
    if (u.gamma_strength > 0.0f) {
        float3 lin = pow(colour, float3(2.4f));
        colour = mix(colour, lin, u.gamma_strength);
    }

    // Scanlines — 1 pixel físico oscuro por fila lógica.
    // Usa uv.y (independiente del offset de letterbox) con pixel_scale para
    // calcular la posición dentro de la fila en coordenadas físicas.
    // 3x: 1 dark + 2 bright. 4x: 1 dark + 3 bright.
    // bright=3.5×, dark floor=0.42 (mantiene aspecto CRT original).
    if (u.scanline_strength > 0.0f) {
        float ps          = max(1.0f, round(u.pixel_scale));
        float frac_in_row = fract(uv.y * u.screen_height);
        float row_pos     = floor(frac_in_row * ps);
        float is_dark     = step(ps - 1.0f, row_pos);
        float scan        = mix(3.5f, 0.42f, is_dark);
        colour           *= mix(1.0f, scan, u.scanline_strength);
    }

    if (u.gamma_strength > 0.0f) {
        float3 enc = pow(colour, float3(1.0f/2.2f));
        colour = mix(colour, enc, u.gamma_strength);
    }

    // Viñeta
    float2 d = uv - 0.5f;
    float vignette = 1.0f - dot(d, d) * u.vignette_strength;
    colour *= clamp(vignette, 0.0f, 1.0f);

    // Máscara de fósforo RGB — después de scanlines (orden original):
    // filas brillantes saturadas → máscara invisible, filas oscuras → RGB visible.
    if (u.mask_strength > 0.0f) {
        float whichMask = fract(in.pos.x * 0.3333333f);
        float3 mask = float3(0.80f);
        if (whichMask < 0.3333333f)       mask.x = 1.0f;
        else if (whichMask < 0.6666667f)  mask.y = 1.0f;
        else                              mask.z = 1.0f;
        colour = mix(colour, colour * mask, u.mask_strength);
    }

    // Parpadeo de fósforo CRT (~50 Hz)
    if (u.flicker > 0.0f) {
        float flicker_wave = sin(u.time * 100.0f) * 0.5f + 0.5f;
        colour *= 1.0f - u.flicker * 0.04f * flicker_wave;
    }

    return float4(colour, 1.0f);
}
)";
// NOLINTEND(readability-identifier-naming)

#endif  // __APPLE__

namespace Rendering {

    // ---------------------------------------------------------------------------
    // Destructor
    // ---------------------------------------------------------------------------
    SDL3GPUShader::~SDL3GPUShader() {
        destroy();
    }

    // ---------------------------------------------------------------------------
    // init
    // ---------------------------------------------------------------------------
    auto SDL3GPUShader::init(SDL_Window* window,
        SDL_Texture* texture,
        const std::string& /*vertex_source*/,
        const std::string& /*fragment_source*/) -> bool {
        // Si ya estaba inicializado (p.ej. al cambiar borde), liberar recursos
        // de textura/pipeline pero mantener el device vivo para evitar conflictos
        // con SDL_Renderer en Windows/Vulkan.
        if (is_initialized_) {
            cleanup();
        }

        window_ = window;

        // Dimensions from the SDL_Texture placeholder
        float fw = 0.0F;
        float fh = 0.0F;
        SDL_GetTextureSize(texture, &fw, &fh);
        game_width_ = static_cast<int>(fw);
        game_height_ = static_cast<int>(fh);
        tex_width_ = game_width_ * oversample_;
        tex_height_ = game_height_ * oversample_;
        uniforms_.screen_height = static_cast<float>(tex_height_);  // Altura de la textura GPU
        uniforms_.oversample = static_cast<float>(oversample_);

        // ----------------------------------------------------------------
        // 1. Create GPU device (solo si no existe ya)
        // ----------------------------------------------------------------
        if (device_ == nullptr) {
#ifdef __APPLE__
            const SDL_GPUShaderFormat PREFERRED = SDL_GPU_SHADERFORMAT_MSL | SDL_GPU_SHADERFORMAT_METALLIB;
#else
            const SDL_GPUShaderFormat PREFERRED = SDL_GPU_SHADERFORMAT_SPIRV;
#endif
            device_ = SDL_CreateGPUDevice(PREFERRED, false, nullptr);
            if (device_ == nullptr) {
                SDL_Log("SDL3GPUShader: SDL_CreateGPUDevice failed: %s", SDL_GetError());
                return false;
            }
            SDL_Log("SDL3GPUShader: driver = %s", SDL_GetGPUDeviceDriver(device_));

            // ----------------------------------------------------------------
            // 2. Claim window (una sola vez — no liberar hasta destroy())
            // ----------------------------------------------------------------
            if (!SDL_ClaimWindowForGPUDevice(device_, window_)) {
                SDL_Log("SDL3GPUShader: SDL_ClaimWindowForGPUDevice failed: %s", SDL_GetError());
                SDL_DestroyGPUDevice(device_);
                device_ = nullptr;
                return false;
            }
            SDL_SetGPUSwapchainParameters(device_, window_, SDL_GPU_SWAPCHAINCOMPOSITION_SDR, vsync_ ? SDL_GPU_PRESENTMODE_VSYNC : SDL_GPU_PRESENTMODE_IMMEDIATE);
        }

        // ----------------------------------------------------------------
        // 3. Create scene texture (upload target + sampler source)
        //    Format: B8G8R8A8_UNORM matches SDL ARGB8888 byte layout on LE
        // ----------------------------------------------------------------
        SDL_GPUTextureCreateInfo tex_info = {};
        tex_info.type = SDL_GPU_TEXTURETYPE_2D;
        tex_info.format = SDL_GPU_TEXTUREFORMAT_B8G8R8A8_UNORM;
        tex_info.usage = SDL_GPU_TEXTUREUSAGE_SAMPLER;
        tex_info.width = static_cast<Uint32>(tex_width_);
        tex_info.height = static_cast<Uint32>(tex_height_);
        tex_info.layer_count_or_depth = 1;
        tex_info.num_levels = 1;
        scene_texture_ = SDL_CreateGPUTexture(device_, &tex_info);
        if (scene_texture_ == nullptr) {
            SDL_Log("SDL3GPUShader: failed to create scene texture: %s", SDL_GetError());
            cleanup();
            return false;
        }

        // ----------------------------------------------------------------
        // 4. Create upload transfer buffer (CPU → GPU, size = w*h*4 bytes)
        // ----------------------------------------------------------------
        SDL_GPUTransferBufferCreateInfo tb_info = {};
        tb_info.usage = SDL_GPU_TRANSFERBUFFERUSAGE_UPLOAD;
        tb_info.size = static_cast<Uint32>(tex_width_ * tex_height_ * 4);
        upload_buffer_ = SDL_CreateGPUTransferBuffer(device_, &tb_info);
        if (upload_buffer_ == nullptr) {
            SDL_Log("SDL3GPUShader: failed to create upload buffer: %s", SDL_GetError());
            cleanup();
            return false;
        }

        // ----------------------------------------------------------------
        // 5. Create samplers: NEAREST (pixel art) + LINEAR (supersampling)
        // ----------------------------------------------------------------
        SDL_GPUSamplerCreateInfo samp_info = {};
        samp_info.min_filter = SDL_GPU_FILTER_NEAREST;
        samp_info.mag_filter = SDL_GPU_FILTER_NEAREST;
        samp_info.mipmap_mode = SDL_GPU_SAMPLERMIPMAPMODE_NEAREST;
        samp_info.address_mode_u = SDL_GPU_SAMPLERADDRESSMODE_CLAMP_TO_EDGE;
        samp_info.address_mode_v = SDL_GPU_SAMPLERADDRESSMODE_CLAMP_TO_EDGE;
        samp_info.address_mode_w = SDL_GPU_SAMPLERADDRESSMODE_CLAMP_TO_EDGE;
        sampler_ = SDL_CreateGPUSampler(device_, &samp_info);
        if (sampler_ == nullptr) {
            SDL_Log("SDL3GPUShader: failed to create sampler: %s", SDL_GetError());
            cleanup();
            return false;
        }

        SDL_GPUSamplerCreateInfo lsamp_info = {};
        lsamp_info.min_filter = SDL_GPU_FILTER_LINEAR;
        lsamp_info.mag_filter = SDL_GPU_FILTER_LINEAR;
        lsamp_info.mipmap_mode = SDL_GPU_SAMPLERMIPMAPMODE_NEAREST;
        lsamp_info.address_mode_u = SDL_GPU_SAMPLERADDRESSMODE_CLAMP_TO_EDGE;
        lsamp_info.address_mode_v = SDL_GPU_SAMPLERADDRESSMODE_CLAMP_TO_EDGE;
        lsamp_info.address_mode_w = SDL_GPU_SAMPLERADDRESSMODE_CLAMP_TO_EDGE;
        linear_sampler_ = SDL_CreateGPUSampler(device_, &lsamp_info);
        if (linear_sampler_ == nullptr) {
            SDL_Log("SDL3GPUShader: failed to create linear sampler: %s", SDL_GetError());
            cleanup();
            return false;
        }

        // ----------------------------------------------------------------
        // 6. Create PostFX graphics pipeline
        // ----------------------------------------------------------------
        if (!createPipeline()) {
            cleanup();
            return false;
        }

        is_initialized_ = true;
        SDL_Log("SDL3GPUShader: initialized OK (%dx%d)", tex_width_, tex_height_);
        return true;
    }

    // ---------------------------------------------------------------------------
    // createPipeline
    // ---------------------------------------------------------------------------
    auto SDL3GPUShader::createPipeline() -> bool {
        const SDL_GPUTextureFormat SWAPCHAIN_FMT = SDL_GetGPUSwapchainTextureFormat(device_, window_);

#ifdef __APPLE__
        SDL_GPUShader* vert = createShaderMSL(device_, POSTFX_VERT_MSL, "postfx_vs", SDL_GPU_SHADERSTAGE_VERTEX, 0, 0);
        SDL_GPUShader* frag = createShaderMSL(device_, POSTFX_FRAG_MSL, "postfx_fs", SDL_GPU_SHADERSTAGE_FRAGMENT, 1, 1);
#else
        SDL_GPUShader* vert = createShaderSPIRV(device_, kpostfx_vert_spv, kpostfx_vert_spv_size, "main", SDL_GPU_SHADERSTAGE_VERTEX, 0, 0);
        SDL_GPUShader* frag = createShaderSPIRV(device_, kpostfx_frag_spv, kpostfx_frag_spv_size, "main", SDL_GPU_SHADERSTAGE_FRAGMENT, 1, 1);
#endif

        if ((vert == nullptr) || (frag == nullptr)) {
            SDL_Log("SDL3GPUShader: failed to compile PostFX shaders");
            if (vert != nullptr) { SDL_ReleaseGPUShader(device_, vert); }
            if (frag != nullptr) { SDL_ReleaseGPUShader(device_, frag); }
            return false;
        }

        SDL_GPUColorTargetBlendState no_blend = {};
        no_blend.enable_blend = false;
        no_blend.enable_color_write_mask = false;

        SDL_GPUColorTargetDescription color_target = {};
        color_target.format = SWAPCHAIN_FMT;
        color_target.blend_state = no_blend;

        SDL_GPUVertexInputState no_input = {};

        SDL_GPUGraphicsPipelineCreateInfo pipe_info = {};
        pipe_info.vertex_shader = vert;
        pipe_info.fragment_shader = frag;
        pipe_info.vertex_input_state = no_input;
        pipe_info.primitive_type = SDL_GPU_PRIMITIVETYPE_TRIANGLELIST;
        pipe_info.target_info.num_color_targets = 1;
        pipe_info.target_info.color_target_descriptions = &color_target;

        pipeline_ = SDL_CreateGPUGraphicsPipeline(device_, &pipe_info);

        SDL_ReleaseGPUShader(device_, vert);
        SDL_ReleaseGPUShader(device_, frag);

        if (pipeline_ == nullptr) {
            SDL_Log("SDL3GPUShader: pipeline creation failed: %s", SDL_GetError());
            return false;
        }
        return true;
    }

    // ---------------------------------------------------------------------------
    // uploadPixels — copies ARGB8888 CPU pixels into the GPU transfer buffer.
    // Con supersampling (oversample_ > 1) expande cada pixel del juego a un bloque
    // oversample × oversample y hornea la scanline oscura en la última fila del bloque.
    // ---------------------------------------------------------------------------
    void SDL3GPUShader::uploadPixels(const Uint32* pixels, int width, int height) {
        if (!is_initialized_ || (upload_buffer_ == nullptr)) { return; }

        void* mapped = SDL_MapGPUTransferBuffer(device_, upload_buffer_, false);
        if (mapped == nullptr) {
            SDL_Log("SDL3GPUShader: SDL_MapGPUTransferBuffer failed: %s", SDL_GetError());
            return;
        }

        if (oversample_ <= 1) {
            // Path sin supersampling: copia directa
            std::memcpy(mapped, pixels, static_cast<size_t>(width * height * 4));
        } else {
            // Path con supersampling: expande cada pixel a OS×OS, oscurece última fila.
            // Replica la fórmula del shader: mix(3.5, 0.42, scanline_strength).
            auto* out = static_cast<Uint32*>(mapped);
            const int OS = oversample_;
            const float BRIGHT_MUL = 1.0F + (baked_scanline_strength_ * 2.5F);  // rows 0..OS-2
            const float DARK_MUL = 1.0F - (baked_scanline_strength_ * 0.58F);   // row OS-1

            for (int y = 0; y < height; ++y) {
                for (int x = 0; x < width; ++x) {
                    const Uint32 SRC = pixels[(y * width) + x];
                    const Uint32 ALPHA = (SRC >> 24) & 0xFFU;
                    const auto FR = static_cast<float>((SRC >> 16) & 0xFFU);
                    const auto FG = static_cast<float>((SRC >> 8) & 0xFFU);
                    const auto FB = static_cast<float>(SRC & 0xFFU);

                    auto make_px = [ALPHA](float rv, float gv, float bv) -> Uint32 {
                        auto cl = [](float v) -> Uint32 { return static_cast<Uint32>(std::min(255.0F, v)); };
                        return (ALPHA << 24) | (cl(rv) << 16) | (cl(gv) << 8) | cl(bv);
                    };

                    const Uint32 BRIGHT = make_px(FR * BRIGHT_MUL, FG * BRIGHT_MUL, FB * BRIGHT_MUL);
                    const Uint32 DARK = make_px(FR * DARK_MUL, FG * DARK_MUL, FB * DARK_MUL);

                    for (int dy = 0; dy < OS; ++dy) {
                        const Uint32 OUT_PX = (dy == OS - 1) ? DARK : BRIGHT;
                        const int DST_Y = (y * OS) + dy;
                        for (int dx = 0; dx < OS; ++dx) {
                            out[(DST_Y * (width * OS)) + ((x * OS) + dx)] = OUT_PX;
                        }
                    }
                }
            }
        }

        SDL_UnmapGPUTransferBuffer(device_, upload_buffer_);
    }

    // ---------------------------------------------------------------------------
    // render — upload scene texture + PostFX pass → swapchain
    // ---------------------------------------------------------------------------
    void SDL3GPUShader::render() {
        if (!is_initialized_) { return; }

        SDL_GPUCommandBuffer* cmd = SDL_AcquireGPUCommandBuffer(device_);
        if (cmd == nullptr) {
            SDL_Log("SDL3GPUShader: SDL_AcquireGPUCommandBuffer failed: %s", SDL_GetError());
            return;
        }

        // ---- Copy pass: transfer buffer → scene texture ----
        SDL_GPUCopyPass* copy = SDL_BeginGPUCopyPass(cmd);
        if (copy != nullptr) {
            SDL_GPUTextureTransferInfo src = {};
            src.transfer_buffer = upload_buffer_;
            src.offset = 0;
            src.pixels_per_row = static_cast<Uint32>(tex_width_);
            src.rows_per_layer = static_cast<Uint32>(tex_height_);

            SDL_GPUTextureRegion dst = {};
            dst.texture = scene_texture_;
            dst.w = static_cast<Uint32>(tex_width_);
            dst.h = static_cast<Uint32>(tex_height_);
            dst.d = 1;

            SDL_UploadToGPUTexture(copy, &src, &dst, false);
            SDL_EndGPUCopyPass(copy);
        }

        // ---- Acquire swapchain texture ----
        SDL_GPUTexture* swapchain = nullptr;
        Uint32 sw = 0;
        Uint32 sh = 0;
        if (!SDL_AcquireGPUSwapchainTexture(cmd, window_, &swapchain, &sw, &sh)) {
            SDL_Log("SDL3GPUShader: SDL_AcquireGPUSwapchainTexture failed: %s", SDL_GetError());
            SDL_SubmitGPUCommandBuffer(cmd);
            return;
        }
        if (swapchain == nullptr) {
            // Window minimized — skip frame
            SDL_SubmitGPUCommandBuffer(cmd);
            return;
        }

        // ---- Render pass: PostFX → swapchain ----
        SDL_GPUColorTargetInfo color_target = {};
        color_target.texture = swapchain;
        color_target.load_op = SDL_GPU_LOADOP_CLEAR;
        color_target.store_op = SDL_GPU_STOREOP_STORE;
        color_target.clear_color = {.r = 0.0F, .g = 0.0F, .b = 0.0F, .a = 1.0F};

        SDL_GPURenderPass* pass = SDL_BeginGPURenderPass(cmd, &color_target, 1, nullptr);
        if (pass != nullptr) {
            SDL_BindGPUGraphicsPipeline(pass, pipeline_);

            // Calcular viewport usando las dimensiones lógicas del canvas (game_width_/height_),
            // no las de la textura GPU (que pueden ser game×3 con supersampling).
            // El GPU escala la textura para cubrir el viewport independientemente de su resolución.
            float vx = 0.0F;
            float vy = 0.0F;
            float vw = 0.0F;
            float vh = 0.0F;
            if (integer_scale_) {
                const int SCALE = std::max(1, std::min(static_cast<int>(sw) / game_width_, static_cast<int>(sh) / game_height_));
                vw = static_cast<float>(game_width_ * SCALE);
                vh = static_cast<float>(game_height_ * SCALE);
            } else {
                const float SCALE = std::min(
                    static_cast<float>(sw) / static_cast<float>(game_width_),
                    static_cast<float>(sh) / static_cast<float>(game_height_));
                vw = static_cast<float>(game_width_) * SCALE;
                vh = static_cast<float>(game_height_) * SCALE;
            }
            vx = std::floor((static_cast<float>(sw) - vw) * 0.5F);
            vy = std::floor((static_cast<float>(sh) - vh) * 0.5F);
            SDL_GPUViewport vp = {.x = vx, .y = vy, .w = vw, .h = vh, .min_depth = 0.0F, .max_depth = 1.0F};
            SDL_SetGPUViewport(pass, &vp);

            // pixel_scale: pixels físicos por pixel lógico de juego (para scanlines sin SS).
            // Con SS las scanlines están horneadas en CPU → scanline_strength=0 → no se usa.
            uniforms_.pixel_scale = (game_height_ > 0)
                ? (vh / static_cast<float>(game_height_))
                : 1.0F;
            uniforms_.time = static_cast<float>(SDL_GetTicks()) / 1000.0F;
            uniforms_.oversample = static_cast<float>(oversample_);

            // Con supersampling usamos LINEAR para que el escalado a zooms no-múltiplo-de-3
            // promedia correctamente las filas de scanline horneadas en CPU.
            SDL_GPUSampler* active_sampler = (oversample_ > 1 && linear_sampler_ != nullptr)
                ? linear_sampler_
                : sampler_;

            SDL_GPUTextureSamplerBinding binding = {};
            binding.texture = scene_texture_;
            binding.sampler = active_sampler;
            SDL_BindGPUFragmentSamplers(pass, 0, &binding, 1);

            SDL_PushGPUFragmentUniformData(cmd, 0, &uniforms_, sizeof(PostFXUniforms));

            SDL_DrawGPUPrimitives(pass, 3, 1, 0, 0);
            SDL_EndGPURenderPass(pass);
        }

        SDL_SubmitGPUCommandBuffer(cmd);
    }

    // ---------------------------------------------------------------------------
    // cleanup — libera pipeline/texturas/buffer pero mantiene device + swapchain
    // ---------------------------------------------------------------------------
    void SDL3GPUShader::cleanup() {
        is_initialized_ = false;

        if (device_ != nullptr) {
            SDL_WaitForGPUIdle(device_);

            if (pipeline_ != nullptr) {
                SDL_ReleaseGPUGraphicsPipeline(device_, pipeline_);
                pipeline_ = nullptr;
            }
            if (scene_texture_ != nullptr) {
                SDL_ReleaseGPUTexture(device_, scene_texture_);
                scene_texture_ = nullptr;
            }
            if (upload_buffer_ != nullptr) {
                SDL_ReleaseGPUTransferBuffer(device_, upload_buffer_);
                upload_buffer_ = nullptr;
            }
            if (sampler_ != nullptr) {
                SDL_ReleaseGPUSampler(device_, sampler_);
                sampler_ = nullptr;
            }
            if (linear_sampler_ != nullptr) {
                SDL_ReleaseGPUSampler(device_, linear_sampler_);
                linear_sampler_ = nullptr;
            }
            // device_ y el claim de la ventana se mantienen vivos
        }
    }

    // ---------------------------------------------------------------------------
    // destroy — limpieza completa incluyendo device y swapchain (solo al cerrar)
    // ---------------------------------------------------------------------------
    void SDL3GPUShader::destroy() {
        cleanup();

        if (device_ != nullptr) {
            if (window_ != nullptr) {
                SDL_ReleaseWindowFromGPUDevice(device_, window_);
            }
            SDL_DestroyGPUDevice(device_);
            device_ = nullptr;
        }
        window_ = nullptr;
    }

    // ---------------------------------------------------------------------------
    // Shader creation helpers
    // ---------------------------------------------------------------------------
    auto SDL3GPUShader::createShaderMSL(SDL_GPUDevice* device,
        const char* msl_source,
        const char* entrypoint,
        SDL_GPUShaderStage stage,
        Uint32 num_samplers,
        Uint32 num_uniform_buffers) -> SDL_GPUShader* {
        SDL_GPUShaderCreateInfo info = {};
        info.code = reinterpret_cast<const Uint8*>(msl_source);
        info.code_size = std::strlen(msl_source) + 1;
        info.entrypoint = entrypoint;
        info.format = SDL_GPU_SHADERFORMAT_MSL;
        info.stage = stage;
        info.num_samplers = num_samplers;
        info.num_uniform_buffers = num_uniform_buffers;
        SDL_GPUShader* shader = SDL_CreateGPUShader(device, &info);
        if (shader == nullptr) {
            SDL_Log("SDL3GPUShader: MSL shader '%s' failed: %s", entrypoint, SDL_GetError());
        }
        return shader;
    }

    auto SDL3GPUShader::createShaderSPIRV(SDL_GPUDevice* device,  // NOLINT(readability-convert-member-functions-to-static)
        const uint8_t* spv_code,
        size_t spv_size,
        const char* entrypoint,
        SDL_GPUShaderStage stage,
        Uint32 num_samplers,
        Uint32 num_uniform_buffers) -> SDL_GPUShader* {
        SDL_GPUShaderCreateInfo info = {};
        info.code = spv_code;
        info.code_size = spv_size;
        info.entrypoint = entrypoint;
        info.format = SDL_GPU_SHADERFORMAT_SPIRV;
        info.stage = stage;
        info.num_samplers = num_samplers;
        info.num_uniform_buffers = num_uniform_buffers;
        SDL_GPUShader* shader = SDL_CreateGPUShader(device, &info);
        if (shader == nullptr) {
            SDL_Log("SDL3GPUShader: SPIRV shader '%s' failed: %s", entrypoint, SDL_GetError());
        }
        return shader;
    }

    void SDL3GPUShader::setPostFXParams(const PostFXParams& p) {
        uniforms_.vignette_strength = p.vignette;
        uniforms_.chroma_strength = p.chroma;
        uniforms_.mask_strength = p.mask;
        uniforms_.gamma_strength = p.gamma;
        uniforms_.curvature = p.curvature;
        uniforms_.bleeding = p.bleeding;
        uniforms_.flicker = p.flicker;

        // Con supersampling las scanlines se hornean en CPU (uploadPixels).
        // El shader recibe strength=0 para no aplicarlas de nuevo en GPU.
        baked_scanline_strength_ = p.scanlines;
        uniforms_.scanline_strength = (oversample_ > 1) ? 0.0F : p.scanlines;
    }

    void SDL3GPUShader::setVSync(bool vsync) {
        vsync_ = vsync;
        if (device_ != nullptr && window_ != nullptr) {
            SDL_SetGPUSwapchainParameters(device_, window_, SDL_GPU_SWAPCHAINCOMPOSITION_SDR, vsync_ ? SDL_GPU_PRESENTMODE_VSYNC : SDL_GPU_PRESENTMODE_IMMEDIATE);
        }
    }

    void SDL3GPUShader::setScaleMode(bool integer_scale) {
        integer_scale_ = integer_scale;
    }

    // ---------------------------------------------------------------------------
    // setOversample — cambia el factor SS; recrea texturas si ya está inicializado
    // ---------------------------------------------------------------------------
    void SDL3GPUShader::setOversample(int factor) {
        const int NEW_FACTOR = std::max(1, factor);
        if (NEW_FACTOR == oversample_) { return; }
        oversample_ = NEW_FACTOR;
        if (is_initialized_) {
            reinitTexturesAndBuffer();
            // scanline_strength se actualizará en el próximo setPostFXParams
        }
    }

    // ---------------------------------------------------------------------------
    // reinitTexturesAndBuffer — recrea scene_texture_ y upload_buffer_ con el
    // tamaño actual (game × oversample_). No toca pipeline ni samplers.
    // ---------------------------------------------------------------------------
    auto SDL3GPUShader::reinitTexturesAndBuffer() -> bool {
        if (device_ == nullptr) { return false; }
        SDL_WaitForGPUIdle(device_);

        if (scene_texture_ != nullptr) {
            SDL_ReleaseGPUTexture(device_, scene_texture_);
            scene_texture_ = nullptr;
        }
        if (upload_buffer_ != nullptr) {
            SDL_ReleaseGPUTransferBuffer(device_, upload_buffer_);
            upload_buffer_ = nullptr;
        }

        tex_width_ = game_width_ * oversample_;
        tex_height_ = game_height_ * oversample_;
        uniforms_.screen_height = static_cast<float>(tex_height_);
        uniforms_.oversample = static_cast<float>(oversample_);

        SDL_GPUTextureCreateInfo tex_info = {};
        tex_info.type = SDL_GPU_TEXTURETYPE_2D;
        tex_info.format = SDL_GPU_TEXTUREFORMAT_B8G8R8A8_UNORM;
        tex_info.usage = SDL_GPU_TEXTUREUSAGE_SAMPLER;
        tex_info.width = static_cast<Uint32>(tex_width_);
        tex_info.height = static_cast<Uint32>(tex_height_);
        tex_info.layer_count_or_depth = 1;
        tex_info.num_levels = 1;
        scene_texture_ = SDL_CreateGPUTexture(device_, &tex_info);
        if (scene_texture_ == nullptr) {
            SDL_Log("SDL3GPUShader: reinit — failed to create scene texture: %s", SDL_GetError());
            return false;
        }

        SDL_GPUTransferBufferCreateInfo tb_info = {};
        tb_info.usage = SDL_GPU_TRANSFERBUFFERUSAGE_UPLOAD;
        tb_info.size = static_cast<Uint32>(tex_width_ * tex_height_ * 4);
        upload_buffer_ = SDL_CreateGPUTransferBuffer(device_, &tb_info);
        if (upload_buffer_ == nullptr) {
            SDL_Log("SDL3GPUShader: reinit — failed to create upload buffer: %s", SDL_GetError());
            SDL_ReleaseGPUTexture(device_, scene_texture_);
            scene_texture_ = nullptr;
            return false;
        }

        SDL_Log("SDL3GPUShader: oversample %d → texture %dx%d", oversample_, tex_width_, tex_height_);
        return true;
    }

}  // namespace Rendering