path: root/source/blender/draw/engines/eevee/eevee_lights.c

/*
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version 2
 * of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
 *
 * Copyright 2016, Blender Foundation.
 */

/** \file
 * \ingroup DNA
 */

#include "DRW_render.h"

#include "BLI_dynstr.h"
#include "BLI_rand.h"
#include "BLI_rect.h"

#include "BKE_object.h"

#include "DEG_depsgraph_query.h"

#include "eevee_private.h"

#define SHADOW_CASTER_ALLOC_CHUNK 16

// #define DEBUG_CSM
// #define DEBUG_SHADOW_DISTRIBUTION

static struct {
  struct GPUShader *shadow_sh;
  struct GPUShader *shadow_store_cube_sh[SHADOW_METHOD_MAX];
  struct GPUShader *shadow_store_cube_high_sh[SHADOW_METHOD_MAX];
  struct GPUShader *shadow_store_cascade_sh[SHADOW_METHOD_MAX];
  struct GPUShader *shadow_store_cascade_high_sh[SHADOW_METHOD_MAX];
  struct GPUShader *shadow_copy_cube_sh[SHADOW_METHOD_MAX];
  struct GPUShader *shadow_copy_cascade_sh[SHADOW_METHOD_MAX];
} e_data = {NULL}; /* Engine data */

extern char datatoc_shadow_vert_glsl[];
extern char datatoc_shadow_frag_glsl[];
extern char datatoc_shadow_process_vert_glsl[];
extern char datatoc_shadow_process_geom_glsl[];
extern char datatoc_shadow_store_frag_glsl[];
extern char datatoc_shadow_copy_frag_glsl[];
extern char datatoc_concentric_samples_lib_glsl[];

extern char datatoc_common_view_lib_glsl[];

/* Prototypes */
static void eevee_light_setup(Object *ob, EEVEE_Light *evli);
static float light_attenuation_radius_get(Light *la, float light_threshold);

/* *********** LIGHT BITS *********** */
static void lightbits_set_single(EEVEE_LightBits *bitf, uint idx, bool val)
{
  if (val) {
    bitf->fields[idx / 8] |= (1 << (idx % 8));
  }
  else {
    bitf->fields[idx / 8] &= ~(1 << (idx % 8));
  }
}

static void lightbits_set_all(EEVEE_LightBits *bitf, bool val)
{
  memset(bitf, (val) ? 0xFF : 0x00, sizeof(EEVEE_LightBits));
}

static void lightbits_or(EEVEE_LightBits *r, const EEVEE_LightBits *v)
{
  for (int i = 0; i < MAX_LIGHTBITS_FIELDS; ++i) {
    r->fields[i] |= v->fields[i];
  }
}

static bool lightbits_get(const EEVEE_LightBits *r, uint idx)
{
  return r->fields[idx / 8] & (1 << (idx % 8));
}

static void lightbits_convert(EEVEE_LightBits *r,
                              const EEVEE_LightBits *bitf,
                              const int *light_bit_conv_table,
                              uint table_length)
{
  for (int i = 0; i < table_length; ++i) {
    if (lightbits_get(bitf, i) != 0) {
      if (light_bit_conv_table[i] >= 0) {
        r->fields[i / 8] |= (1 << (i % 8));
      }
    }
  }
}

/* *********** FUNCTIONS *********** */

void EEVEE_lights_init(EEVEE_ViewLayerData *sldata)
{
  const uint shadow_ubo_size = sizeof(EEVEE_Shadow) * MAX_SHADOW +
                               sizeof(EEVEE_ShadowCube) * MAX_SHADOW_CUBE +
                               sizeof(EEVEE_ShadowCascade) * MAX_SHADOW_CASCADE;

  const DRWContextState *draw_ctx = DRW_context_state_get();
  const Scene *scene_eval = DEG_get_evaluated_scene(draw_ctx->depsgraph);

  if (!e_data.shadow_sh) {
    e_data.shadow_sh = DRW_shader_create_with_lib(datatoc_shadow_vert_glsl,
                                                  NULL,
                                                  datatoc_shadow_frag_glsl,
                                                  datatoc_common_view_lib_glsl,
                                                  NULL);
  }

  if (!sldata->lights) {
    sldata->lights = MEM_callocN(sizeof(EEVEE_LightsInfo), "EEVEE_LightsInfo");
    sldata->light_ubo = DRW_uniformbuffer_create(sizeof(EEVEE_Light) * MAX_LIGHT, NULL);
    sldata->shadow_ubo = DRW_uniformbuffer_create(shadow_ubo_size, NULL);
    sldata->shadow_render_ubo = DRW_uniformbuffer_create(sizeof(EEVEE_ShadowRender), NULL);

    for (int i = 0; i < 2; ++i) {
      sldata->shcasters_buffers[i].shadow_casters = MEM_callocN(
          sizeof(EEVEE_ShadowCaster) * SHADOW_CASTER_ALLOC_CHUNK, "EEVEE_ShadowCaster buf");
      sldata->shcasters_buffers[i].flags = MEM_callocN(sizeof(sldata->shcasters_buffers[0].flags) *
                                                           SHADOW_CASTER_ALLOC_CHUNK,
                                                       "EEVEE_shcast_buffer flags buf");
      sldata->shcasters_buffers[i].alloc_count = SHADOW_CASTER_ALLOC_CHUNK;
      sldata->shcasters_buffers[i].count = 0;
    }

    sldata->lights->shcaster_frontbuffer = &sldata->shcasters_buffers[0];
    sldata->lights->shcaster_backbuffer = &sldata->shcasters_buffers[1];
  }

  /* Flip buffers */
  SWAP(EEVEE_ShadowCasterBuffer *,
       sldata->lights->shcaster_frontbuffer,
       sldata->lights->shcaster_backbuffer);

  const int sh_method = scene_eval->eevee.shadow_method;
  int sh_cube_size = scene_eval->eevee.shadow_cube_size;
  int sh_cascade_size = scene_eval->eevee.shadow_cascade_size;
  const bool sh_high_bitdepth = (scene_eval->eevee.flag & SCE_EEVEE_SHADOW_HIGH_BITDEPTH) != 0;
  sldata->lights->soft_shadows = (scene_eval->eevee.flag & SCE_EEVEE_SHADOW_SOFT) != 0;

  EEVEE_LightsInfo *linfo = sldata->lights;
  if ((linfo->shadow_cube_size != sh_cube_size) || (linfo->shadow_method != sh_method) ||
      (linfo->shadow_high_bitdepth != sh_high_bitdepth)) {
    BLI_assert((sh_cube_size > 0) && (sh_cube_size <= 4096));
    DRW_TEXTURE_FREE_SAFE(sldata->shadow_cube_pool);
    DRW_TEXTURE_FREE_SAFE(sldata->shadow_cube_target);
    DRW_TEXTURE_FREE_SAFE(sldata->shadow_cube_blur);

    /* Compute adequate size for the octahedral map. */
    linfo->shadow_cube_store_size = OCTAHEDRAL_SIZE_FROM_CUBESIZE(sh_cube_size);

    CLAMP(linfo->shadow_cube_store_size, 1, 4096);
    CLAMP(sh_cube_size, 1, 4096);

    linfo->shadow_render_data.cube_texel_size = 1.0f / sh_cube_size;
  }

  if ((linfo->shadow_cascade_size != sh_cascade_size) || (linfo->shadow_method != sh_method) ||
      (linfo->shadow_high_bitdepth != sh_high_bitdepth)) {
    BLI_assert((sh_cascade_size > 0) && (sh_cascade_size <= 4096));
    DRW_TEXTURE_FREE_SAFE(sldata->shadow_cascade_pool);
    DRW_TEXTURE_FREE_SAFE(sldata->shadow_cascade_target);
    DRW_TEXTURE_FREE_SAFE(sldata->shadow_cascade_blur);

    CLAMP(sh_cascade_size, 1, 4096);
  }

  linfo->shadow_high_bitdepth = sh_high_bitdepth;
  linfo->shadow_method = sh_method;
  linfo->shadow_cube_size = sh_cube_size;
  linfo->shadow_cascade_size = sh_cascade_size;

  /* only compile the ones needed. reduce startup time. */
  if ((sh_method == SHADOW_ESM) && !e_data.shadow_copy_cube_sh[SHADOW_ESM]) {
    e_data.shadow_copy_cube_sh[SHADOW_ESM] = DRW_shader_create(datatoc_shadow_process_vert_glsl,
                                                               datatoc_shadow_process_geom_glsl,
                                                               datatoc_shadow_copy_frag_glsl,
                                                               "#define ESM\n"
                                                               "#define COPY\n");
    e_data.shadow_copy_cascade_sh[SHADOW_ESM] = DRW_shader_create(datatoc_shadow_process_vert_glsl,
                                                                  datatoc_shadow_process_geom_glsl,
                                                                  datatoc_shadow_copy_frag_glsl,
                                                                  "#define ESM\n"
                                                                  "#define COPY\n"
                                                                  "#define CSM\n");
  }
  else if ((sh_method == SHADOW_VSM) && !e_data.shadow_copy_cube_sh[SHADOW_VSM]) {
    e_data.shadow_copy_cube_sh[SHADOW_VSM] = DRW_shader_create(datatoc_shadow_process_vert_glsl,
                                                               datatoc_shadow_process_geom_glsl,
                                                               datatoc_shadow_copy_frag_glsl,
                                                               "#define VSM\n"
                                                               "#define COPY\n");
    e_data.shadow_copy_cascade_sh[SHADOW_VSM] = DRW_shader_create(datatoc_shadow_process_vert_glsl,
                                                                  datatoc_shadow_process_geom_glsl,
                                                                  datatoc_shadow_copy_frag_glsl,
                                                                  "#define VSM\n"
                                                                  "#define COPY\n"
                                                                  "#define CSM\n");
  }
}

static GPUShader *eevee_lights_get_store_sh(int shadow_method, bool high_blur, bool cascade)
{
  GPUShader **shader;

  if (cascade) {
    shader = (high_blur) ? &e_data.shadow_store_cascade_high_sh[shadow_method] :
                           &e_data.shadow_store_cascade_sh[shadow_method];
  }
  else {
    shader = (high_blur) ? &e_data.shadow_store_cube_high_sh[shadow_method] :
                           &e_data.shadow_store_cube_sh[shadow_method];
  }

  if (*shader == NULL) {
    DynStr *ds_frag = BLI_dynstr_new();
    BLI_dynstr_append(ds_frag, datatoc_concentric_samples_lib_glsl);
    BLI_dynstr_append(ds_frag, datatoc_shadow_store_frag_glsl);
    char *store_shadow_shader_str = BLI_dynstr_get_cstring(ds_frag);
    BLI_dynstr_free(ds_frag);

    ds_frag = BLI_dynstr_new();
    BLI_dynstr_append(ds_frag, (shadow_method == SHADOW_VSM) ? "#define VSM\n" : "#define ESM\n");
    if (high_blur) {
      BLI_dynstr_append(ds_frag, "#define HIGH_BLUR\n");
    }
    if (cascade) {
      BLI_dynstr_append(ds_frag, "#define CSM\n");
    }
    char *define_str = BLI_dynstr_get_cstring(ds_frag);
    BLI_dynstr_free(ds_frag);

    *shader = DRW_shader_create(datatoc_shadow_process_vert_glsl,
                                datatoc_shadow_process_geom_glsl,
                                store_shadow_shader_str,
                                define_str);

    MEM_freeN(store_shadow_shader_str);
    MEM_freeN(define_str);
  }

  return *shader;
}

static DRWPass *eevee_lights_cube_store_pass_get(EEVEE_PassList *psl,
                                                 EEVEE_ViewLayerData *sldata,
                                                 int shadow_method,
                                                 int shadow_samples_len)
{
  bool high_blur = shadow_samples_len > 16;
  DRWPass **pass = (high_blur) ? &psl->shadow_cube_store_pass : &psl->shadow_cube_store_high_pass;
  if (*pass == NULL) {
    *pass = DRW_pass_create("Shadow Cube Storage Pass", DRW_STATE_WRITE_COLOR);
    GPUShader *shader = eevee_lights_get_store_sh(shadow_method, high_blur, false);
    DRWShadingGroup *grp = DRW_shgroup_create(shader, *pass);
    DRW_shgroup_uniform_texture_ref(grp, "shadowTexture", &sldata->shadow_cube_blur);
    DRW_shgroup_uniform_block(grp, "shadow_render_block", sldata->shadow_render_ubo);

    DRW_shgroup_call_procedural_triangles(grp, NULL, 6);
  }
  return *pass;
}

static DRWPass *eevee_lights_cascade_store_pass_get(EEVEE_PassList *psl,
                                                    EEVEE_ViewLayerData *sldata,
                                                    int shadow_method,
                                                    int shadow_samples_len)
{
  bool high_blur = shadow_samples_len > 16;
  DRWPass **pass = (high_blur) ? &psl->shadow_cascade_store_pass :
                                 &psl->shadow_cascade_store_high_pass;
  if (*pass == NULL) {
    *pass = DRW_pass_create("Shadow Cascade Storage Pass", DRW_STATE_WRITE_COLOR);
    GPUShader *shader = eevee_lights_get_store_sh(shadow_method, high_blur, true);
    DRWShadingGroup *grp = DRW_shgroup_create(shader, *pass);
    DRW_shgroup_uniform_texture_ref(grp, "shadowTexture", &sldata->shadow_cascade_blur);
    DRW_shgroup_uniform_block(grp, "shadow_render_block", sldata->shadow_render_ubo);

    DRW_shgroup_call_procedural_triangles(grp, NULL, MAX_CASCADE_NUM);
  }
  return *pass;
}

void EEVEE_lights_cache_init(EEVEE_ViewLayerData *sldata, EEVEE_Data *vedata)
{
  EEVEE_LightsInfo *linfo = sldata->lights;
  EEVEE_StorageList *stl = vedata->stl;
  EEVEE_PassList *psl = vedata->psl;

  linfo->shcaster_frontbuffer->count = 0;
  linfo->num_light = 0;
  linfo->num_cube_layer = 0;
  linfo->num_cascade_layer = 0;
  linfo->gpu_cube_len = linfo->gpu_cascade_len = linfo->gpu_shadow_len = 0;
  linfo->cpu_cube_len = linfo->cpu_cascade_len = 0;
  memset(linfo->light_ref, 0, sizeof(linfo->light_ref));
  memset(linfo->shadow_cube_ref, 0, sizeof(linfo->shadow_cube_ref));
  memset(linfo->shadow_cascade_ref, 0, sizeof(linfo->shadow_cascade_ref));
  memset(linfo->new_shadow_id, -1, sizeof(linfo->new_shadow_id));

  /* Shadow Casters: Reset flags. */
  memset(linfo->shcaster_backbuffer->flags,
         (char)SHADOW_CASTER_PRUNED,
         linfo->shcaster_backbuffer->alloc_count);
  memset(linfo->shcaster_frontbuffer->flags, 0x00, linfo->shcaster_frontbuffer->alloc_count);

  psl->shadow_cube_store_pass = NULL;
  psl->shadow_cube_store_high_pass = NULL;
  psl->shadow_cascade_store_pass = NULL;
  psl->shadow_cascade_store_high_pass = NULL;

  {
    DRW_PASS_CREATE(psl->shadow_cube_copy_pass, DRW_STATE_WRITE_COLOR);

    DRWShadingGroup *grp = DRW_shgroup_create(e_data.shadow_copy_cube_sh[linfo->shadow_method],
                                              psl->shadow_cube_copy_pass);
    DRW_shgroup_uniform_texture_ref(grp, "shadowTexture", &sldata->shadow_cube_target);
    DRW_shgroup_uniform_block(grp, "shadow_render_block", sldata->shadow_render_ubo);

    DRW_shgroup_call_procedural_triangles(grp, NULL, 6);
  }

  {
    DRW_PASS_CREATE(psl->shadow_cascade_copy_pass, DRW_STATE_WRITE_COLOR);

    DRWShadingGroup *grp = DRW_shgroup_create(e_data.shadow_copy_cascade_sh[linfo->shadow_method],
                                              psl->shadow_cascade_copy_pass);
    DRW_shgroup_uniform_texture_ref(grp, "shadowTexture", &sldata->shadow_cascade_target);
    DRW_shgroup_uniform_block(grp, "shadow_render_block", sldata->shadow_render_ubo);

    DRW_shgroup_call_procedural_triangles(grp, NULL, MAX_CASCADE_NUM);
  }

  {
    DRWState state = DRW_STATE_WRITE_COLOR | DRW_STATE_WRITE_DEPTH | DRW_STATE_DEPTH_LESS_EQUAL;
    DRW_PASS_CREATE(psl->shadow_pass, state);

    stl->g_data->shadow_shgrp = DRW_shgroup_create(e_data.shadow_sh, psl->shadow_pass);
  }
}

void EEVEE_lights_cache_add(EEVEE_ViewLayerData *sldata, Object *ob)
{
  EEVEE_LightsInfo *linfo = sldata->lights;

  const DRWContextState *draw_ctx = DRW_context_state_get();
  const float threshold = draw_ctx->scene->eevee.light_threshold;
  /* Step 1 find all lights in the scene and setup them */
  if (linfo->num_light >= MAX_LIGHT) {
    printf("Too many lights in the scene !!!\n");
  }
  else {
    Light *la = (Light *)ob->data;

    /* Early out if light has no power. */
    if (la->energy == 0.0f || is_zero_v3(&la->r)) {
      return;
    }

    EEVEE_Light *evli = linfo->light_data + linfo->num_light;
    eevee_light_setup(ob, evli);

    /* We do not support shadowmaps for dupli lights. */
    if ((ob->base_flag & BASE_FROM_DUPLI) != 0) {
      linfo->num_light++;
      return;
    }

    EEVEE_LightEngineData *led = EEVEE_light_data_ensure(ob);

    /* Save previous shadow id. */
    int prev_cube_sh_id = led->prev_cube_shadow_id;

    /* Default light without shadows */
    led->data.ld.shadow_id = -1;
    led->prev_cube_shadow_id = -1;

    if (la->mode & LA_SHADOW) {
      if (la->type == LA_SUN) {
        int cascade_nbr = la->cascade_count;

        if ((linfo->gpu_cascade_len + 1) <= MAX_SHADOW_CASCADE) {
          /* Save Light object. */
          linfo->shadow_cascade_ref[linfo->cpu_cascade_len] = ob;

          /* Store indices. */
          EEVEE_ShadowCascadeData *data = &led->data.scad;
          data->shadow_id = linfo->gpu_shadow_len;
          data->cascade_id = linfo->gpu_cascade_len;
          data->layer_id = linfo->num_cascade_layer;

          /* Increment indices. */
          linfo->gpu_shadow_len += 1;
          linfo->gpu_cascade_len += 1;
          linfo->num_cascade_layer += cascade_nbr;

          linfo->cpu_cascade_len += 1;
        }
      }
      else if (la->type == LA_SPOT || la->type == LA_LOCAL || la->type == LA_AREA) {
        if ((linfo->gpu_cube_len + 1) <= MAX_SHADOW_CUBE) {
          /* Save Light object. */
          linfo->shadow_cube_ref[linfo->cpu_cube_len] = ob;

          /* For light update tracking. */
          if ((prev_cube_sh_id >= 0) && (prev_cube_sh_id < linfo->shcaster_backbuffer->count)) {
            linfo->new_shadow_id[prev_cube_sh_id] = linfo->cpu_cube_len;
          }
          led->prev_cube_shadow_id = linfo->cpu_cube_len;

          /* Saving light bounds for later. */
          BLI_assert(linfo->cpu_cube_len >= 0 && linfo->cpu_cube_len < MAX_LIGHT);
          copy_v3_v3(linfo->shadow_bounds[linfo->cpu_cube_len].center, ob->obmat[3]);
          linfo->shadow_bounds[linfo->cpu_cube_len].radius = light_attenuation_radius_get(
              la, threshold);

          EEVEE_ShadowCubeData *data = &led->data.scd;
          /* Store indices. */
          data->shadow_id = linfo->gpu_shadow_len;
          data->cube_id = linfo->gpu_cube_len;
          data->layer_id = linfo->num_cube_layer;

          /* Increment indices. */
          linfo->gpu_shadow_len += 1;
          linfo->gpu_cube_len += 1;
          linfo->num_cube_layer += 1;

          linfo->cpu_cube_len += 1;
        }
      }
    }

    led->data.ld.light_id = linfo->num_light;
    linfo->light_ref[linfo->num_light] = ob;
    linfo->num_light++;
  }
}

/* Add a shadow caster to the shadowpasses */
void EEVEE_lights_cache_shcaster_add(EEVEE_ViewLayerData *UNUSED(sldata),
                                     EEVEE_StorageList *stl,
                                     struct GPUBatch *geom,
                                     Object *ob)
{
  DRW_shgroup_call(stl->g_data->shadow_shgrp, geom, ob);
}

void EEVEE_lights_cache_shcaster_material_add(EEVEE_ViewLayerData *sldata,
                                              EEVEE_PassList *psl,
                                              struct GPUMaterial *gpumat,
                                              struct GPUBatch *geom,
                                              struct Object *ob,
                                              const float *alpha_threshold)
{
  /* TODO / PERF : reuse the same shading group for objects with the same material */
  DRWShadingGroup *grp = DRW_shgroup_material_create(gpumat, psl->shadow_pass);

  if (grp == NULL) {
    return;
  }

  /* Grrr needed for correctness but not 99% of the time not needed.
   * TODO detect when needed? */
  DRW_shgroup_uniform_block(grp, "probe_block", sldata->probe_ubo);
  DRW_shgroup_uniform_block(grp, "grid_block", sldata->grid_ubo);
  DRW_shgroup_uniform_block(grp, "planar_block", sldata->planar_ubo);
  DRW_shgroup_uniform_block(grp, "light_block", sldata->light_ubo);
  DRW_shgroup_uniform_block(grp, "shadow_block", sldata->shadow_ubo);
  DRW_shgroup_uniform_block(grp, "common_block", sldata->common_ubo);

  if (alpha_threshold != NULL) {
    DRW_shgroup_uniform_float(grp, "alphaThreshold", alpha_threshold, 1);
  }

  DRW_shgroup_call(grp, geom, ob);
}

/* Make that object update shadow casting lights inside its influence bounding box. */
void EEVEE_lights_cache_shcaster_object_add(EEVEE_ViewLayerData *sldata, Object *ob)
{
  if ((ob->base_flag & BASE_FROM_DUPLI) != 0) {
    /* TODO: Special case for dupli objects because we cannot save the object pointer. */
    return;
  }

  EEVEE_ObjectEngineData *oedata = EEVEE_object_data_ensure(ob);
  EEVEE_LightsInfo *linfo = sldata->lights;
  EEVEE_ShadowCasterBuffer *backbuffer = linfo->shcaster_backbuffer;
  EEVEE_ShadowCasterBuffer *frontbuffer = linfo->shcaster_frontbuffer;
  int past_id = oedata->shadow_caster_id;

  /* Update flags in backbuffer. */
  if (past_id > -1 && past_id < backbuffer->count) {
    backbuffer->flags[past_id] &= ~SHADOW_CASTER_PRUNED;

    if (oedata->need_update) {
      backbuffer->flags[past_id] |= SHADOW_CASTER_UPDATED;
    }
  }

  /* Update id. */
  oedata->shadow_caster_id = frontbuffer->count++;

  /* Make sure shadow_casters is big enough. */
  if (oedata->shadow_caster_id >= frontbuffer->alloc_count) {
    frontbuffer->alloc_count += SHADOW_CASTER_ALLOC_CHUNK;
    frontbuffer->shadow_casters = MEM_reallocN(
        frontbuffer->shadow_casters, sizeof(EEVEE_ShadowCaster) * frontbuffer->alloc_count);
    frontbuffer->flags = MEM_reallocN(frontbuffer->flags,
                                      sizeof(EEVEE_ShadowCaster) * frontbuffer->alloc_count);
  }

  EEVEE_ShadowCaster *shcaster = frontbuffer->shadow_casters + oedata->shadow_caster_id;

  if (oedata->need_update) {
    frontbuffer->flags[oedata->shadow_caster_id] = SHADOW_CASTER_UPDATED;
  }

  /* Update World AABB in frontbuffer. */
  BoundBox *bb = BKE_object_boundbox_get(ob);
  float min[3], max[3];
  INIT_MINMAX(min, max);
  for (int i = 0; i < 8; ++i) {
    float vec[3];
    copy_v3_v3(vec, bb->vec[i]);
    mul_m4_v3(ob->obmat, vec);
    minmax_v3v3_v3(min, max, vec);
  }

  EEVEE_BoundBox *aabb = &shcaster->bbox;
  add_v3_v3v3(aabb->center, min, max);
  mul_v3_fl(aabb->center, 0.5f);
  sub_v3_v3v3(aabb->halfdim, aabb->center, max);

  aabb->halfdim[0] = fabsf(aabb->halfdim[0]);
  aabb->halfdim[1] = fabsf(aabb->halfdim[1]);
  aabb->halfdim[2] = fabsf(aabb->halfdim[2]);

  oedata->need_update = false;
}

void EEVEE_lights_cache_finish(EEVEE_ViewLayerData *sldata, EEVEE_Data *vedata)
{
  EEVEE_LightsInfo *linfo = sldata->lights;
  eGPUTextureFormat shadow_pool_format = GPU_R32F;

  sldata->common_data.la_num_light = linfo->num_light;

  /* Setup enough layers. */
  /* Free textures if number mismatch. */
  if (linfo->num_cube_layer != linfo->cache_num_cube_layer) {
    DRW_TEXTURE_FREE_SAFE(sldata->shadow_cube_pool);
    linfo->cache_num_cube_layer = linfo->num_cube_layer;
    linfo->update_flag |= LIGHT_UPDATE_SHADOW_CUBE;
  }

  if (linfo->num_cascade_layer != linfo->cache_num_cascade_layer) {
    DRW_TEXTURE_FREE_SAFE(sldata->shadow_cascade_pool);
    linfo->cache_num_cascade_layer = linfo->num_cascade_layer;
  }

  switch (linfo->shadow_method) {
    case SHADOW_ESM:
      shadow_pool_format = ((linfo->shadow_high_bitdepth) ? GPU_R32F : GPU_R16F);
      break;
    case SHADOW_VSM:
      shadow_pool_format = ((linfo->shadow_high_bitdepth) ? GPU_RG32F : GPU_RG16F);
      break;
    default:
      BLI_assert(!"Incorrect Shadow Method");
      break;
  }

  /* Cubemaps */
  if (!sldata->shadow_cube_target) {
    sldata->shadow_cube_target = DRW_texture_create_cube(
        linfo->shadow_cube_size, GPU_DEPTH_COMPONENT24, 0, NULL);
    sldata->shadow_cube_blur = DRW_texture_create_cube(
        linfo->shadow_cube_size, shadow_pool_format, DRW_TEX_FILTER, NULL);
  }
  GPU_framebuffer_ensure_config(
      &sldata->shadow_cube_copy_fb,
      {GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(sldata->shadow_cube_blur)});

  if (!sldata->shadow_cube_pool) {
    sldata->shadow_cube_pool = DRW_texture_create_2d_array(linfo->shadow_cube_store_size,
                                                           linfo->shadow_cube_store_size,
                                                           max_ii(1, linfo->num_cube_layer),
                                                           shadow_pool_format,
                                                           DRW_TEX_FILTER,
                                                           NULL);
  }
  GPU_framebuffer_ensure_config(&sldata->shadow_cube_target_fb,
                                {GPU_ATTACHMENT_TEXTURE(sldata->shadow_cube_target)});
  GPU_framebuffer_ensure_config(
      &sldata->shadow_cube_store_fb,
      {GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(sldata->shadow_cube_pool)});

  /* CSM */
  if (!sldata->shadow_cascade_target) {
    sldata->shadow_cascade_target = DRW_texture_create_2d_array(linfo->shadow_cascade_size,
                                                                linfo->shadow_cascade_size,
                                                                MAX_CASCADE_NUM,
                                                                GPU_DEPTH_COMPONENT24,
                                                                0,
                                                                NULL);
    sldata->shadow_cascade_blur = DRW_texture_create_2d_array(linfo->shadow_cascade_size,
                                                              linfo->shadow_cascade_size,
                                                              MAX_CASCADE_NUM,
                                                              shadow_pool_format,
                                                              DRW_TEX_FILTER,
                                                              NULL);
  }
  GPU_framebuffer_ensure_config(
      &sldata->shadow_cascade_copy_fb,
      {GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(sldata->shadow_cascade_blur)});

  if (!sldata->shadow_cascade_pool) {
    sldata->shadow_cascade_pool = DRW_texture_create_2d_array(linfo->shadow_cascade_size,
                                                              linfo->shadow_cascade_size,
                                                              max_ii(1, linfo->num_cascade_layer),
                                                              shadow_pool_format,
                                                              DRW_TEX_FILTER,
                                                              NULL);
  }
  GPU_framebuffer_ensure_config(&sldata->shadow_cascade_target_fb,
                                {GPU_ATTACHMENT_TEXTURE(sldata->shadow_cascade_target)});
  GPU_framebuffer_ensure_config(
      &sldata->shadow_cascade_store_fb,
      {GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(sldata->shadow_cascade_pool)});

  /* Update Lights UBOs. */
  EEVEE_lights_update(sldata, vedata);
}

float light_attenuation_radius_get(Light *la, float light_threshold)
{
  if (la->mode & LA_CUSTOM_ATTENUATION) {
    return la->att_dist;
  }

  /* Compute max light power. */
  float power = max_fff(la->r, la->g, la->b);
  power *= fabsf(la->energy / 100.0f);
  power *= max_ff(1.0f, la->spec_fac);
  /* Compute the distance (using the inverse square law)
   * at which the light power reaches the light_threshold. */
  float distance = sqrtf(max_ff(1e-16, power / max_ff(1e-16, light_threshold)));
  return distance;
}

static void light_shape_parameters_set(EEVEE_Light *evli, const Light *la, float scale[3])
{
  if (la->type == LA_SPOT) {
    /* Spot size & blend */
    evli->sizex = scale[0] / scale[2];
    evli->sizey = scale[1] / scale[2];
    evli->spotsize = cosf(la->spotsize * 0.5f);
    evli->spotblend = (1.0f - evli->spotsize) * la->spotblend;
    evli->radius = max_ff(0.001f, la->area_size);
  }
  else if (la->type == LA_AREA) {
    evli->sizex = max_ff(0.003f, la->area_size * scale[0] * 0.5f);
    if (ELEM(la->area_shape, LA_AREA_RECT, LA_AREA_ELLIPSE)) {
      evli->sizey = max_ff(0.003f, la->area_sizey * scale[1] * 0.5f);
    }
    else {
      evli->sizey = max_ff(0.003f, la->area_size * scale[1] * 0.5f);
    }
  }
  else if (la->type == LA_SUN) {
    evli->radius = max_ff(0.001f, tanf(la->sun_angle / 2.0f));
  }
  else {
    evli->radius = max_ff(0.001f, la->area_size);
  }
}

static float light_shape_power_get(const Light *la, const EEVEE_Light *evli)
{
  float power;
  /* Make illumination power constant */
  if (la->type == LA_AREA) {
    power = 1.0f / (evli->sizex * evli->sizey * 4.0f * M_PI) * /* 1/(w*h*Pi) */
            0.8f; /* XXX : Empirical, Fit cycles power */
    if (ELEM(la->area_shape, LA_AREA_DISK, LA_AREA_ELLIPSE)) {
      /* Scale power to account for the lower area of the ellipse compared to the surrounding
       * rectangle. */
      power *= 4.0f / M_PI;
    }
  }
  else if (la->type == LA_SPOT || la->type == LA_LOCAL) {
    power = 1.0f / (4.0f * evli->radius * evli->radius * M_PI * M_PI); /* 1/(4*r²*Pi²) */

    /* for point lights (a.k.a radius == 0.0) */
    // power = M_PI * M_PI * 0.78; /* XXX : Empirical, Fit cycles power */
  }
  else {
    power = 1.0f / (evli->radius * evli->radius * M_PI); /* 1/(r²*Pi) */
    /* Make illumination power closer to cycles for bigger radii. Cycles uses a cos^3 term that we
     * cannot reproduce so we account for that by scaling the light power. This function is the
     * result of a rough manual fitting. */
    power += 1.0f / (2.0f * M_PI); /* power *= 1 + r²/2 */
  }
  return power;
}

/* Update buffer with light data */
static void eevee_light_setup(Object *ob, EEVEE_Light *evli)
{
  Light *la = (Light *)ob->data;
  float mat[4][4], scale[3], power, att_radius;

  const DRWContextState *draw_ctx = DRW_context_state_get();
  const float light_threshold = draw_ctx->scene->eevee.light_threshold;

  /* Position */
  copy_v3_v3(evli->position, ob->obmat[3]);

  /* Color */
  copy_v3_v3(evli->color, &la->r);

  evli->spec = la->spec_fac;

  /* Influence Radius */
  att_radius = light_attenuation_radius_get(la, light_threshold);
  /* Take the inverse square of this distance. */
  evli->invsqrdist = 1.0 / max_ff(1e-4f, att_radius * att_radius);

  /* Vectors */
  normalize_m4_m4_ex(mat, ob->obmat, scale);
  copy_v3_v3(evli->forwardvec, mat[2]);
  normalize_v3(evli->forwardvec);
  negate_v3(evli->forwardvec);

  copy_v3_v3(evli->rightvec, mat[0]);
  normalize_v3(evli->rightvec);

  copy_v3_v3(evli->upvec, mat[1]);
  normalize_v3(evli->upvec);

  /* Make sure we have a consistent Right Hand coord frame.
   * (in case of negatively scaled Z axis) */
  float cross[3];
  cross_v3_v3v3(cross, evli->rightvec, evli->forwardvec);
  if (dot_v3v3(cross, evli->upvec) < 0.0) {
    negate_v3(evli->upvec);
  }

  light_shape_parameters_set(evli, la, scale);

  /* Light Type */
  evli->light_type = (float)la->type;
  if ((la->type == LA_AREA) && ELEM(la->area_shape, LA_AREA_DISK, LA_AREA_ELLIPSE)) {
    evli->light_type = LAMPTYPE_AREA_ELLIPSE;
  }

  power = light_shape_power_get(la, evli);
  mul_v3_fl(evli->color, power * la->energy);

  /* No shadow by default */
  evli->shadowid = -1.0f;
}

/**
 * Special ball distribution:
 * Point are distributed in a way that when they are orthogonaly
 * projected into any plane, the resulting distribution is (close to)
 * a uniform disc distribution.
 */
static void sample_ball(int sample_ofs, float radius, float rsample[3])
{
  double ht_point[3];
  double ht_offset[3] = {0.0, 0.0, 0.0};
  uint ht_primes[3] = {2, 3, 7};

  BLI_halton_3d(ht_primes, ht_offset, sample_ofs, ht_point);

  float omega = ht_point[1] * 2.0f * M_PI;

  rsample[2] = ht_point[0] * 2.0f - 1.0f; /* cos theta */

  float r = sqrtf(fmaxf(0.0f, 1.0f - rsample[2] * rsample[2])); /* sin theta */

  rsample[0] = r * cosf(omega);
  rsample[1] = r * sinf(omega);

  radius *= sqrt(sqrt(ht_point[2]));
  mul_v3_fl(rsample, radius);
}

static void sample_rectangle(
    int sample_ofs, float x_axis[3], float y_axis[3], float size_x, float size_y, float rsample[3])
{
  double ht_point[2];
  double ht_offset[2] = {0.0, 0.0};
  uint ht_primes[2] = {2, 3};

  BLI_halton_2d(ht_primes, ht_offset, sample_ofs, ht_point);

  /* Change ditribution center to be 0,0 */
  ht_point[0] = (ht_point[0] > 0.5f) ? ht_point[0] - 1.0f : ht_point[0];
  ht_point[1] = (ht_point[1] > 0.5f) ? ht_point[1] - 1.0f : ht_point[1];

  zero_v3(rsample);
  madd_v3_v3fl(rsample, x_axis, (ht_point[0] * 2.0f) * size_x);
  madd_v3_v3fl(rsample, y_axis, (ht_point[1] * 2.0f) * size_y);
}

static void sample_ellipse(
    int sample_ofs, float x_axis[3], float y_axis[3], float size_x, float size_y, float rsample[3])
{
  double ht_point[2];
  double ht_offset[2] = {0.0, 0.0};
  uint ht_primes[2] = {2, 3};

  BLI_halton_2d(ht_primes, ht_offset, sample_ofs, ht_point);

  /* Uniform disc sampling. */
  float omega = ht_point[1] * 2.0f * M_PI;
  float r = sqrtf(ht_point[0]);
  ht_point[0] = r * cosf(omega) * size_x;
  ht_point[1] = r * sinf(omega) * size_y;

  zero_v3(rsample);
  madd_v3_v3fl(rsample, x_axis, ht_point[0]);
  madd_v3_v3fl(rsample, y_axis, ht_point[1]);
}

static void shadow_cube_random_position_set(EEVEE_Light *evli,
                                            Light *la,
                                            int sample_ofs,
                                            float ws_sample_pos[3])
{
  float jitter[3];

#ifndef DEBUG_SHADOW_DISTRIBUTION
  int i = sample_ofs;
#else
  for (int i = 0; i <= sample_ofs; ++i) {
#endif
  switch (la->type) {
    case LA_AREA:
      if (ELEM(la->area_shape, LA_AREA_RECT, LA_AREA_SQUARE)) {
        sample_rectangle(i, evli->rightvec, evli->upvec, evli->sizex, evli->sizey, jitter);
      }
      else {
        sample_ellipse(i, evli->rightvec, evli->upvec, evli->sizex, evli->sizey, jitter);
      }
      break;
    default:
      sample_ball(i, evli->radius, jitter);
  }
#ifdef DEBUG_SHADOW_DISTRIBUTION
  float p[3];
  add_v3_v3v3(p, jitter, ws_sample_pos);
  DRW_debug_sphere(p, 0.01f, (float[4]){1.0f, (sample_ofs == i) ? 1.0f : 0.0f, 0.0f, 1.0f});
}
#endif
add_v3_v3(ws_sample_pos, jitter);
}

static void eevee_shadow_cube_setup(Object *ob,
                                    EEVEE_LightsInfo *linfo,
                                    EEVEE_LightEngineData *led,
                                    int sample_ofs)
{
  EEVEE_ShadowCubeData *sh_data = &led->data.scd;
  EEVEE_Light *evli = linfo->light_data + sh_data->light_id;
  EEVEE_Shadow *ubo_data = linfo->shadow_data + sh_data->shadow_id;
  EEVEE_ShadowCube *cube_data = linfo->shadow_cube_data + sh_data->cube_id;
  Light *la = (Light *)ob->data;

  copy_v3_v3(cube_data->position, ob->obmat[3]);

  if (linfo->soft_shadows) {
    shadow_cube_random_position_set(evli, la, sample_ofs, cube_data->position);
  }

  ubo_data->bias = 0.05f * la->bias;
  ubo_data->near = la->clipsta;
  ubo_data->far = 1.0f / (evli->invsqrdist * evli->invsqrdist);
  ubo_data->exp = (linfo->shadow_method == SHADOW_VSM) ? la->bleedbias : la->bleedexp;

  evli->shadowid = (float)(sh_data->shadow_id);
  ubo_data->shadow_start = (float)(sh_data->layer_id);
  ubo_data->data_start = (float)(sh_data->cube_id);
  ubo_data->shadow_blur = la->soft * 0.02f; /* Used by translucence shadowmap blur */

  ubo_data->contact_dist = (la->mode & LA_SHAD_CONTACT) ? la->contact_dist : 0.0f;
  ubo_data->contact_bias = 0.05f * la->contact_bias;
  ubo_data->contact_spread = la->contact_spread;
  ubo_data->contact_thickness = la->contact_thickness;
}

static void shadow_cascade_random_matrix_set(float mat[4][4], float radius, int sample_ofs)
{
  float jitter[3];

#ifndef DEBUG_SHADOW_DISTRIBUTION
  int i = sample_ofs;
#else
    for (int i = 0; i <= sample_ofs; ++i) {
#endif
  sample_ellipse(i, mat[0], mat[1], radius, radius, jitter);
#ifdef DEBUG_SHADOW_DISTRIBUTION
  float p[3];
  add_v3_v3v3(p, jitter, mat[2]);
  DRW_debug_sphere(p, 0.01f, (float[4]){1.0f, (sample_ofs == i) ? 1.0f : 0.0f, 0.0f, 1.0f});
}
#endif

add_v3_v3(mat[2], jitter);
orthogonalize_m4(mat, 2);
}

#define LERP(t, a, b) ((a) + (t) * ((b) - (a)))

static double round_to_digits(double value, int digits)
{
  double factor = pow(10.0, digits - ceil(log10(fabs(value))));
  return round(value * factor) / factor;
}

static void frustum_min_bounding_sphere(const float corners[8][3],
                                        float r_center[3],
                                        float *r_radius)
{
#if 0 /* Simple solution but waste too much space. */
  float minvec[3], maxvec[3];

  /* compute the bounding box */
  INIT_MINMAX(minvec, maxvec);
  for (int i = 0; i < 8; ++i) {
    minmax_v3v3_v3(minvec, maxvec, corners[i]);
  }

  /* compute the bounding sphere of this box */
  r_radius = len_v3v3(minvec, maxvec) * 0.5f;
  add_v3_v3v3(r_center, minvec, maxvec);
  mul_v3_fl(r_center, 0.5f);
#else
      /* Find averaged center. */
      zero_v3(r_center);
      for (int i = 0; i < 8; ++i) {
        add_v3_v3(r_center, corners[i]);
      }
      mul_v3_fl(r_center, 1.0f / 8.0f);

      /* Search the largest distance from the sphere center. */
      *r_radius = 0.0f;
      for (int i = 0; i < 8; ++i) {
        float rad = len_squared_v3v3(corners[i], r_center);
        if (rad > *r_radius) {
          *r_radius = rad;
        }
      }

      /* TODO try to reduce the radius further by moving the center.
       * Remember we need a __stable__ solution! */

      /* Try to reduce float imprecision leading to shimmering. */
      *r_radius = (float)round_to_digits(sqrtf(*r_radius), 3);
#endif
}

static void eevee_shadow_cascade_setup(Object *ob,
                                       EEVEE_LightsInfo *linfo,
                                       EEVEE_LightEngineData *led,
                                       DRWView *view,
                                       float view_near,
                                       float view_far,
                                       int sample_ofs)
{
  Light *la = (Light *)ob->data;

  /* Camera Matrices */
  float persinv[4][4], vp_projmat[4][4];
  DRW_view_persmat_get(view, persinv, true);
  DRW_view_winmat_get(view, vp_projmat, false);
  bool is_persp = DRW_view_is_persp_get(view);

  /* Lights Matrices */
  int cascade_nbr = la->cascade_count;

  EEVEE_ShadowCascadeData *sh_data = &led->data.scad;
  EEVEE_Light *evli = linfo->light_data + sh_data->light_id;
  EEVEE_Shadow *ubo_data = linfo->shadow_data + sh_data->shadow_id;
  EEVEE_ShadowCascade *cascade_data = linfo->shadow_cascade_data + sh_data->cascade_id;

  /* obmat = Object Space > World Space */
  /* viewmat = World Space > View Space */
  float(*viewmat)[4] = sh_data->viewmat;
#if 0 /* done at culling time */
  normalize_m4_m4(viewmat, ob->obmat);
#endif

  if (linfo->soft_shadows) {
    shadow_cascade_random_matrix_set(viewmat, evli->radius, sample_ofs);
  }

  copy_m4_m4(sh_data->viewinv, viewmat);
  invert_m4(viewmat);

  /* The technique consists into splitting
   * the view frustum into several sub-frustum
   * that are individually receiving one shadow map */

  float csm_start, csm_end;

  if (is_persp) {
    csm_start = view_near;
    csm_end = max_ff(view_far, -la->cascade_max_dist);
    /* Avoid artifacts */
    csm_end = min_ff(view_near, csm_end);
  }
  else {
    csm_start = -view_far;
    csm_end = view_far;
  }

  /* init near/far */
  for (int c = 0; c < MAX_CASCADE_NUM; ++c) {
    cascade_data->split_start[c] = csm_end;
    cascade_data->split_end[c] = csm_end;
  }

  /* Compute split planes */
  float splits_start_ndc[MAX_CASCADE_NUM];
  float splits_end_ndc[MAX_CASCADE_NUM];

  {
    /* Nearest plane */
    float p[4] = {1.0f, 1.0f, csm_start, 1.0f};
    /* TODO: we don't need full m4 multiply here */
    mul_m4_v4(vp_projmat, p);
    splits_start_ndc[0] = p[2];
    if (is_persp) {
      splits_start_ndc[0] /= p[3];
    }
  }

  {
    /* Farthest plane */
    float p[4] = {1.0f, 1.0f, csm_end, 1.0f};
    /* TODO: we don't need full m4 multiply here */
    mul_m4_v4(vp_projmat, p);
    splits_end_ndc[cascade_nbr - 1] = p[2];
    if (is_persp) {
      splits_end_ndc[cascade_nbr - 1] /= p[3];
    }
  }

  cascade_data->split_start[0] = csm_start;
  cascade_data->split_end[cascade_nbr - 1] = csm_end;

  for (int c = 1; c < cascade_nbr; ++c) {
    /* View Space */
    float linear_split = LERP(((float)(c) / (float)cascade_nbr), csm_start, csm_end);
    float exp_split = csm_start * powf(csm_end / csm_start, (float)(c) / (float)cascade_nbr);

    if (is_persp) {
      cascade_data->split_start[c] = LERP(la->cascade_exponent, linear_split, exp_split);
    }
    else {
      cascade_data->split_start[c] = linear_split;
    }
    cascade_data->split_end[c - 1] = cascade_data->split_start[c];

    /* Add some overlap for smooth transition */
    cascade_data->split_start[c] = LERP(la->cascade_fade,
                                        cascade_data->split_end[c - 1],
                                        (c > 1) ? cascade_data->split_end[c - 2] :
                                                  cascade_data->split_start[0]);

    /* NDC Space */
    {
      float p[4] = {1.0f, 1.0f, cascade_data->split_start[c], 1.0f};
      /* TODO: we don't need full m4 multiply here */
      mul_m4_v4(vp_projmat, p);
      splits_start_ndc[c] = p[2];

      if (is_persp) {
        splits_start_ndc[c] /= p[3];
      }
    }

    {
      float p[4] = {1.0f, 1.0f, cascade_data->split_end[c - 1], 1.0f};
      /* TODO: we don't need full m4 multiply here */
      mul_m4_v4(vp_projmat, p);
      splits_end_ndc[c - 1] = p[2];

      if (is_persp) {
        splits_end_ndc[c - 1] /= p[3];
      }
    }
  }

  /* Set last cascade split fade distance into the first split_start. */
  float prev_split = (cascade_nbr > 1) ? cascade_data->split_end[cascade_nbr - 2] :
                                         cascade_data->split_start[0];
  cascade_data->split_start[0] = LERP(
      la->cascade_fade, cascade_data->split_end[cascade_nbr - 1], prev_split);

  /* For each cascade */
  for (int c = 0; c < cascade_nbr; ++c) {
    float(*projmat)[4] = sh_data->projmat[c];
    /* Given 8 frustum corners */
    float corners[8][3] = {
        /* Near Cap */
        {1.0f, -1.0f, splits_start_ndc[c]},
        {-1.0f, -1.0f, splits_start_ndc[c]},
        {-1.0f, 1.0f, splits_start_ndc[c]},
        {1.0f, 1.0f, splits_start_ndc[c]},
        /* Far Cap */
        {1.0f, -1.0f, splits_end_ndc[c]},
        {-1.0f, -1.0f, splits_end_ndc[c]},
        {-1.0f, 1.0f, splits_end_ndc[c]},
        {1.0f, 1.0f, splits_end_ndc[c]},
    };

    /* Transform them into world space */
    for (int i = 0; i < 8; ++i) {
      mul_project_m4_v3(persinv, corners[i]);
    }

    float center[3];
    frustum_min_bounding_sphere(corners, center, &(sh_data->radius[c]));

#ifdef DEBUG_CSM
    float dbg_col[4] = {0.0f, 0.0f, 0.0f, 1.0f};
    if (c < 3) {
      dbg_col[c] = 1.0f;
    }
    DRW_debug_bbox((BoundBox *)&corners, dbg_col);
    DRW_debug_sphere(center, sh_data->radius[c], dbg_col);
#endif

    /* Project into lightspace */
    mul_m4_v3(viewmat, center);

    /* Snap projection center to nearest texel to cancel shimmering. */
    float shadow_origin[2], shadow_texco[2];
    /* Light to texture space. */
    mul_v2_v2fl(shadow_origin, center, linfo->shadow_cascade_size / (2.0f * sh_data->radius[c]));

    /* Find the nearest texel. */
    shadow_texco[0] = roundf(shadow_origin[0]);
    shadow_texco[1] = roundf(shadow_origin[1]);

    /* Compute offset. */
    sub_v2_v2(shadow_texco, shadow_origin);
    /* Texture to light space. */
    mul_v2_fl(shadow_texco, (2.0f * sh_data->radius[c]) / linfo->shadow_cascade_size);

    /* Apply offset. */
    add_v2_v2(center, shadow_texco);

    /* Expand the projection to cover frustum range */
    rctf rect_cascade;
    BLI_rctf_init_pt_radius(&rect_cascade, center, sh_data->radius[c]);
    orthographic_m4(projmat,
                    rect_cascade.xmin,
                    rect_cascade.xmax,
                    rect_cascade.ymin,
                    rect_cascade.ymax,
                    la->clipsta,
                    la->clipend);

    mul_m4_m4m4(sh_data->viewprojmat[c], projmat, viewmat);
    mul_m4_m4m4(cascade_data->shadowmat[c], texcomat, sh_data->viewprojmat[c]);

#ifdef DEBUG_CSM
    DRW_debug_m4_as_bbox(sh_data->viewprojmat[c], dbg_col, true);
#endif
  }

  ubo_data->bias = 0.05f * la->bias;
  ubo_data->near = la->clipsta;
  ubo_data->far = la->clipend;
  ubo_data->exp = (linfo->shadow_method == SHADOW_VSM) ? la->bleedbias : la->bleedexp;

  evli->shadowid = (float)(sh_data->shadow_id);
  ubo_data->shadow_start = (float)(sh_data->layer_id);
  ubo_data->data_start = (float)(sh_data->cascade_id);
  ubo_data->shadow_blur = la->soft * 0.02f; /* Used by translucence shadowmap blur */

  ubo_data->contact_dist = (la->mode & LA_SHAD_CONTACT) ? la->contact_dist : 0.0f;
  ubo_data->contact_bias = 0.05f * la->contact_bias;
  ubo_data->contact_spread = la->contact_spread;
  ubo_data->contact_thickness = la->contact_thickness;
}

/* Used for checking if object is inside the shadow volume. */
static bool sphere_bbox_intersect(const EEVEE_BoundSphere *bs, const EEVEE_BoundBox *bb)
{
  /* We are testing using a rougher AABB vs AABB test instead of full AABB vs Sphere. */
  /* TODO test speed with AABB vs Sphere. */
  bool x = fabsf(bb->center[0] - bs->center[0]) <= (bb->halfdim[0] + bs->radius);
  bool y = fabsf(bb->center[1] - bs->center[1]) <= (bb->halfdim[1] + bs->radius);
  bool z = fabsf(bb->center[2] - bs->center[2]) <= (bb->halfdim[2] + bs->radius);

  return x && y && z;
}

void EEVEE_lights_update(EEVEE_ViewLayerData *sldata, EEVEE_Data *vedata)
{
  EEVEE_StorageList *stl = vedata->stl;
  EEVEE_EffectsInfo *effects = stl->effects;
  EEVEE_LightsInfo *linfo = sldata->lights;
  Object *ob;
  int i;
  char *flag;
  EEVEE_ShadowCaster *shcaster;
  EEVEE_BoundSphere *bsphere;
  EEVEE_ShadowCasterBuffer *frontbuffer = linfo->shcaster_frontbuffer;
  EEVEE_ShadowCasterBuffer *backbuffer = linfo->shcaster_backbuffer;

  EEVEE_LightBits update_bits = {{0}};
  if ((linfo->update_flag & LIGHT_UPDATE_SHADOW_CUBE) != 0) {
    /* Update all lights. */
    lightbits_set_all(&update_bits, true);
  }
  else {
    /* Search for deleted shadow casters and if shcaster WAS in shadow radius. */
    /* No need to run this if we already update all lights. */
    EEVEE_LightBits past_bits = {{0}};
    EEVEE_LightBits curr_bits = {{0}};
    shcaster = backbuffer->shadow_casters;
    flag = backbuffer->flags;
    for (i = 0; i < backbuffer->count; ++i, ++flag, ++shcaster) {
      /* If the shadowcaster has been deleted or updated. */
      if (*flag != 0) {
        /* Add the lights that were intersecting with its BBox. */
        lightbits_or(&past_bits, &shcaster->bits);
      }
    }
    /* Convert old bits to new bits and add result to final update bits. */
    /* NOTE: This might be overkill since all lights are tagged to refresh if
     * the light count changes. */
    lightbits_convert(&curr_bits, &past_bits, linfo->new_shadow_id, MAX_LIGHT);
    lightbits_or(&update_bits, &curr_bits);
  }

  /* Search for updates in current shadow casters. */
  shcaster = frontbuffer->shadow_casters;
  flag = frontbuffer->flags;
  for (i = 0; i < frontbuffer->count; i++, flag++, shcaster++) {
    /* Run intersection checks to fill the bitfields. */
    bsphere = linfo->shadow_bounds;
    for (int j = 0; j < linfo->cpu_cube_len; j++, bsphere++) {
      bool iter = sphere_bbox_intersect(bsphere, &shcaster->bbox);
      lightbits_set_single(&shcaster->bits, j, iter);
    }
    /* Only add to final bits if objects has been updated. */
    if (*flag != 0) {
      lightbits_or(&update_bits, &shcaster->bits);
    }
  }

  /* Setup shadow cube in UBO and tag for update if necessary. */
  for (i = 0; (i < MAX_SHADOW_CUBE) && (ob = linfo->shadow_cube_ref[i]); i++) {
    EEVEE_LightEngineData *led = EEVEE_light_data_ensure(ob);

    eevee_shadow_cube_setup(ob, linfo, led, effects->taa_current_sample - 1);
    if (lightbits_get(&update_bits, i) != 0 || linfo->soft_shadows) {
      led->need_update = true;
    }
  }

  /* Resize shcasters buffers if too big. */
  if (frontbuffer->alloc_count - frontbuffer->count > SHADOW_CASTER_ALLOC_CHUNK) {
    frontbuffer->alloc_count = (frontbuffer->count / SHADOW_CASTER_ALLOC_CHUNK) *
                               SHADOW_CASTER_ALLOC_CHUNK;
    frontbuffer->alloc_count += (frontbuffer->count % SHADOW_CASTER_ALLOC_CHUNK != 0) ?
                                    SHADOW_CASTER_ALLOC_CHUNK :
                                    0;
    frontbuffer->shadow_casters = MEM_reallocN(
        frontbuffer->shadow_casters, sizeof(EEVEE_ShadowCaster) * frontbuffer->alloc_count);
    frontbuffer->flags = MEM_reallocN(frontbuffer->flags,
                                      sizeof(EEVEE_ShadowCaster) * frontbuffer->alloc_count);
  }
}

static void eevee_ensure_cube_views(float near, float far, const float pos[3], DRWView *view[6])
{
  float winmat[4][4], viewmat[4][4];
  perspective_m4(winmat, -near, near, -near, near, near, far);

  for (int i = 0; i < 6; i++) {
    unit_m4(viewmat);
    negate_v3_v3(viewmat[3], pos);
    mul_m4_m4m4(viewmat, cubefacemat[i], viewmat);

    if (view[i] == NULL) {
      view[i] = DRW_view_create(viewmat, winmat, NULL, NULL, NULL);
    }
    else {
      DRW_view_update(view[i], viewmat, winmat, NULL, NULL);
    }
  }
}

static void eevee_ensure_cascade_views(EEVEE_ShadowCascadeData *cascade_data,
                                       int cascade_count,
                                       DRWView *view[4])
{
  for (int i = 0; i < cascade_count; i++) {
    if (view[i] == NULL) {
      view[i] = DRW_view_create(cascade_data->viewmat, cascade_data->projmat[i], NULL, NULL, NULL);
    }
    else {
      DRW_view_update(view[i], cascade_data->viewmat, cascade_data->projmat[i], NULL, NULL);
    }
  }
}

/* this refresh lights shadow buffers */
void EEVEE_draw_shadows(EEVEE_ViewLayerData *sldata, EEVEE_Data *vedata, DRWView *view)
{
  EEVEE_PassList *psl = vedata->psl;
  EEVEE_StorageList *stl = vedata->stl;
  EEVEE_EffectsInfo *effects = stl->effects;
  EEVEE_PrivateData *g_data = stl->g_data;
  EEVEE_LightsInfo *linfo = sldata->lights;
  const DRWContextState *draw_ctx = DRW_context_state_get();
  const float light_threshold = draw_ctx->scene->eevee.light_threshold;
  Object *ob;
  int i;

  int saved_ray_type = sldata->common_data.ray_type;

  /* TODO: make it optional if we don't draw shadows. */
  sldata->common_data.ray_type = EEVEE_RAY_SHADOW;
  DRW_uniformbuffer_update(sldata->common_ubo, &sldata->common_data);

  /* Precompute all shadow/view test before rendering and trashing the culling cache. */
  bool cube_visible[MAX_SHADOW_CUBE];
  for (i = 0; (ob = linfo->shadow_cube_ref[i]) && (i < MAX_SHADOW_CUBE); i++) {
    Light *la = (Light *)ob->data;
    BoundSphere bsphere = {
        .center = {ob->obmat[3][0], ob->obmat[3][1], ob->obmat[3][2]},
        .radius = light_attenuation_radius_get(la, light_threshold),
    };
    cube_visible[i] = DRW_culling_sphere_test(view, &bsphere);
  }

  bool cascade_visible[MAX_SHADOW_CASCADE];
  for (i = 0; (ob = linfo->shadow_cascade_ref[i]) && (i < MAX_SHADOW_CASCADE); i++) {
    EEVEE_LightEngineData *led = EEVEE_light_data_get(ob);
    EEVEE_ShadowCascadeData *sh_data = &led->data.scad;
    float plane[4];
    normalize_m4_m4(sh_data->viewmat, ob->obmat);

    plane_from_point_normal_v3(plane, sh_data->viewmat[3], sh_data->viewmat[2]);
    /* TODO: check against near/far instead of "local Z = 0" plane.
     * Or even the cascades AABB. */
    cascade_visible[i] = DRW_culling_plane_test(view, plane);
  }

  /* Cube Shadow Maps */
  DRW_stats_group_start("Cube Shadow Maps");
  /* Render each shadow to one layer of the array */
  for (i = 0; (ob = linfo->shadow_cube_ref[i]) && (i < MAX_SHADOW_CUBE); i++) {
    EEVEE_LightEngineData *led = EEVEE_light_data_ensure(ob);
    Light *la = (Light *)ob->data;

    if (!led->need_update || !cube_visible[i]) {
      continue;
    }

    EEVEE_ShadowRender *srd = &linfo->shadow_render_data;
    EEVEE_ShadowCubeData *evscd = &led->data.scd;
    EEVEE_ShadowCube *cube_data = linfo->shadow_cube_data + evscd->cube_id;

    srd->clip_near = la->clipsta;
    srd->clip_far = light_attenuation_radius_get(la, light_threshold);
    srd->stored_texel_size = 1.0 / (float)linfo->shadow_cube_store_size;
    srd->exponent = la->bleedexp;

    eevee_ensure_cube_views(
        srd->clip_near, srd->clip_far, cube_data->position, g_data->cube_views);

    /* Render shadow cube */
    /* Render 6 faces separately: seems to be faster for the general case.
     * The only time it's more beneficial is when the CPU culling overhead
     * outweigh the instancing overhead. which is rarely the case. */
    for (int j = 0; j < 6; j++) {
      DRW_view_set_active(g_data->cube_views[j]);

      GPU_framebuffer_texture_cubeface_attach(
          sldata->shadow_cube_target_fb, sldata->shadow_cube_target, 0, j, 0);
      GPU_framebuffer_bind(sldata->shadow_cube_target_fb);
      GPU_framebuffer_clear_depth(sldata->shadow_cube_target_fb, 1.0f);
      DRW_draw_pass(psl->shadow_pass);
    }

    /* 0.001f is arbitrary, but it should be relatively small so that filter size is not too big.
     */
    float filter_texture_size = la->soft * 0.001f;
    float filter_pixel_size = ceil(filter_texture_size / srd->cube_texel_size);

    /* TODO: OPTI: Don't do this intermediate step if no filter is needed. */
    {
      srd->filter_size[0] = srd->cube_texel_size * ((filter_pixel_size > 1.0f) ? 1.5f : 0.0f);
      srd->view_count = 6;
      srd->base_id = 0;
      DRW_uniformbuffer_update(sldata->shadow_render_ubo, srd);

      /* Copy using a small 3x3 box filter */
      GPU_framebuffer_bind(sldata->shadow_cube_copy_fb);
      DRW_draw_pass(psl->shadow_cube_copy_pass);
    }
    /* Push it to shadowmap array */
    {
      /* Adjust constants if concentric samples change. */
      const float max_filter_size = 7.5f;
      const float magic = 4.5f; /* Dunno why but that works. */
      const int max_sample = 256;

      if (filter_pixel_size > 2.0f) {
        srd->filter_size[0] = srd->cube_texel_size * max_filter_size * magic;
        filter_pixel_size = max_ff(0.0f, filter_pixel_size - 3.0f);
        /* Compute number of concentric samples. Depends directly on filter size. */
        float pix_size_sqr = filter_pixel_size * filter_pixel_size;
        srd->shadow_samples_len[0] = min_ii(
            max_sample, 4 + 8 * (int)filter_pixel_size + 4 * (int)(pix_size_sqr));
      }
      else {
        srd->filter_size[0] = 0.0f;
        srd->shadow_samples_len[0] = 4;
      }
      srd->view_count = 1;
      srd->base_id = evscd->layer_id;
      srd->shadow_samples_len_inv[0] = 1.0f / (float)srd->shadow_samples_len[0];
      DRW_uniformbuffer_update(sldata->shadow_render_ubo, srd);

      DRWPass *store_pass = eevee_lights_cube_store_pass_get(
          psl, sldata, linfo->shadow_method, srd->shadow_samples_len[0]);

      GPU_framebuffer_bind(sldata->shadow_cube_store_fb);
      DRW_draw_pass(store_pass);
    }

    if (linfo->soft_shadows == false) {
      led->need_update = false;
    }
  }
  linfo->update_flag &= ~LIGHT_UPDATE_SHADOW_CUBE;
  DRW_stats_group_end();

  float near = DRW_view_near_distance_get(view);
  float far = DRW_view_far_distance_get(view);

  /* Cascaded Shadow Maps */
  DRW_stats_group_start("Cascaded Shadow Maps");
  for (i = 0; (ob = linfo->shadow_cascade_ref[i]) && (i < MAX_SHADOW_CASCADE); i++) {
    if (!cascade_visible[i]) {
      continue;
    }

    EEVEE_LightEngineData *led = EEVEE_light_data_ensure(ob);
    Light *la = (Light *)ob->data;

    EEVEE_ShadowCascadeData *evscd = &led->data.scad;
    EEVEE_ShadowRender *srd = &linfo->shadow_render_data;

    srd->clip_near = la->clipsta;
    srd->clip_far = la->clipend;
    srd->view_count = la->cascade_count;
    srd->stored_texel_size = 1.0 / (float)linfo->shadow_cascade_size;

    DRW_uniformbuffer_update(sldata->shadow_render_ubo, &linfo->shadow_render_data);

    eevee_shadow_cascade_setup(ob, linfo, led, view, near, far, effects->taa_current_sample - 1);

    /* Meh, Reusing the cube views. */
    BLI_assert(MAX_CASCADE_NUM <= 6);
    eevee_ensure_cascade_views(evscd, la->cascade_count, g_data->cube_views);

    /* Render shadow cascades */
    /* Render cascade separately: seems to be faster for the general case.
     * The only time it's more beneficial is when the CPU culling overhead
     * outweigh the instancing overhead. which is rarely the case. */
    for (int j = 0; j < la->cascade_count; j++) {
      DRW_view_set_active(g_data->cube_views[j]);

      GPU_framebuffer_texture_layer_attach(
          sldata->shadow_cascade_target_fb, sldata->shadow_cascade_target, 0, j, 0);
      GPU_framebuffer_bind(sldata->shadow_cascade_target_fb);
      GPU_framebuffer_clear_depth(sldata->shadow_cascade_target_fb, 1.0f);
      DRW_draw_pass(psl->shadow_pass);
    }
    /* Copy using a small 3x3 box filter */
    {
      /* NOTE: We always do it in the case of CSM because of artifacts in the farthest cascade. */
      copy_v4_fl(srd->filter_size, srd->stored_texel_size);
      srd->base_id = 0;
      DRW_uniformbuffer_update(sldata->shadow_render_ubo, srd);

      GPU_framebuffer_bind(sldata->shadow_cascade_copy_fb);
      DRW_draw_pass(psl->shadow_cascade_copy_pass);
    }
    /* Push it to shadowmap array and blur more */
    {
      int max_pass_sample = 0;

      for (int j = 0; j < la->cascade_count; j++) {
        /* 0.01f factor to convert to percentage */
        float filter_texture_size = la->soft * 0.01f / evscd->radius[j];
        float filter_pixel_size = ceil(linfo->shadow_cascade_size * filter_texture_size);
        /* Adjust constants if concentric samples change. */
        const float max_filter_size = 7.5f;
        const float magic = 3.2f; /* Arbitrary: less banding */
        const int max_sample = 256;

        if (filter_pixel_size > 2.0f) {
          srd->filter_size[j] = srd->stored_texel_size * max_filter_size * magic;
          filter_pixel_size = max_ff(0.0f, filter_pixel_size - 3.0f);
          /* Compute number of concentric samples. Depends directly on filter size. */
          float pix_size_sqr = filter_pixel_size * filter_pixel_size;
          srd->shadow_samples_len[j] = min_ii(
              max_sample, 4 + 8 * (int)filter_pixel_size + 4 * (int)(pix_size_sqr));
        }
        else {
          srd->filter_size[j] = 0.0f;
          srd->shadow_samples_len[j] = 4;
        }
        srd->shadow_samples_len_inv[j] = 1.0f / (float)srd->shadow_samples_len[j];
        max_pass_sample = max_ii(max_pass_sample, srd->shadow_samples_len[j]);
      }
      srd->base_id = evscd->layer_id;
      DRW_uniformbuffer_update(sldata->shadow_render_ubo, srd);

      /* XXX(fclem) this create drawcalls outside of cache generation. */
      DRWPass *store_pass = eevee_lights_cascade_store_pass_get(
          psl, sldata, linfo->shadow_method, max_pass_sample);

      GPU_framebuffer_bind(sldata->shadow_cascade_store_fb);
      DRW_draw_pass(store_pass);
    }
  }

  DRW_stats_group_end();

  DRW_view_set_active(view);

  DRW_uniformbuffer_update(sldata->light_ubo, &linfo->light_data);
  DRW_uniformbuffer_update(sldata->shadow_ubo, &linfo->shadow_data); /* Update all data at once */

  sldata->common_data.ray_type = saved_ray_type;
  DRW_uniformbuffer_update(sldata->common_ubo, &sldata->common_data);
}

void EEVEE_lights_free(void)
{
  DRW_SHADER_FREE_SAFE(e_data.shadow_sh);
  for (int i = 0; i < SHADOW_METHOD_MAX; ++i) {
    DRW_SHADER_FREE_SAFE(e_data.shadow_store_cube_sh[i]);
    DRW_SHADER_FREE_SAFE(e_data.shadow_store_cube_high_sh[i]);
    DRW_SHADER_FREE_SAFE(e_data.shadow_store_cascade_sh[i]);
    DRW_SHADER_FREE_SAFE(e_data.shadow_store_cascade_high_sh[i]);
    DRW_SHADER_FREE_SAFE(e_data.shadow_copy_cube_sh[i]);
    DRW_SHADER_FREE_SAFE(e_data.shadow_copy_cascade_sh[i]);
  }
}