GPencil: Use indexed rendering instead of instances

This allows using instancing in other ways, like resources indexing.

17 files changed, 228 insertions, 838 deletions

diff --git a/source/blender/draw/engines/eevee_next/eevee_sync.cc b/source/blender/draw/engines/eevee_next/eevee_sync.cc
index 08cda6f47cf..0915dace1e5 100644
--- a/source/blender/draw/engines/eevee_next/eevee_sync.cc
+++ b/source/blender/draw/engines/eevee_next/eevee_sync.cc
@@ -248,15 +248,15 @@ static void gpencil_stroke_sync(bGPDlayer * /*gpl*/,
     return;
   }
 
+  GPUBatch *geom = DRW_cache_gpencil_get(iter.ob, iter.cfra);
+
   if (show_fill) {
-    GPUBatch *geom = DRW_cache_gpencil_fills_get(iter.ob, iter.cfra);
     int vfirst = gps->runtime.fill_start * 3;
     int vcount = gps->tot_triangles * 3;
     gpencil_drawcall_add(iter, geom, material, vfirst, vcount, false);
   }
 
   if (show_stroke) {
-    GPUBatch *geom = DRW_cache_gpencil_strokes_get(iter.ob, iter.cfra);
     /* Start one vert before to have gl_InstanceID > 0 (see shader). */
     int vfirst = gps->runtime.stroke_start - 1;
     /* Include "potential" cyclic vertex and start adj vertex (see shader). */
diff --git a/source/blender/draw/engines/eevee_next/shaders/eevee_geom_gpencil_vert.glsl b/source/blender/draw/engines/eevee_next/shaders/eevee_geom_gpencil_vert.glsl
index 38debf14eda..87a5bf71c45 100644
--- a/source/blender/draw/engines/eevee_next/shaders/eevee_geom_gpencil_vert.glsl
+++ b/source/blender/draw/engines/eevee_next/shaders/eevee_geom_gpencil_vert.glsl
@@ -16,30 +16,18 @@ void main()
   float hardness;
   vec2 thickness;
 
-  gl_Position = gpencil_vertex(ma,
-                               ma1,
-                               ma2,
-                               ma3,
-                               pos,
-                               pos1,
-                               pos2,
-                               pos3,
-                               uv1,
-                               uv2,
-                               col1,
-                               col2,
-                               fcol1,
-                               /* TODO */
-                               vec4(1024.0, 1024.0, 1.0 / 1024.0, 1.0 / 1024.0),
-                               interp.P,
-                               interp.N,
-                               g_color,
-                               strength,
-                               g_uvs,
-                               sspos,
-                               aspect,
-                               thickness,
-                               hardness);
+  gl_Position = gpencil_vertex(
+      /* TODO */
+      vec4(1024.0, 1024.0, 1.0 / 1024.0, 1.0 / 1024.0),
+      interp.P,
+      interp.N,
+      g_color,
+      strength,
+      g_uvs,
+      sspos,
+      aspect,
+      thickness,
+      hardness);
 #ifdef MAT_VELOCITY
   /* GPencil do not support deformation motion blur. */
   vec3 lP_curr = transform_point(ModelMatrixInverse, interp.P);
diff --git a/source/blender/draw/engines/gpencil/gpencil_engine.c b/source/blender/draw/engines/gpencil/gpencil_engine.c
index b24e4c605e4..956830e6b4c 100644
--- a/source/blender/draw/engines/gpencil/gpencil_engine.c
+++ b/source/blender/draw/engines/gpencil/gpencil_engine.c
@@ -346,7 +346,7 @@ typedef struct gpIterPopulateData {
   int vfirst, vcount;
 } gpIterPopulateData;
 
-#define DISABLE_BATCHING 0
+#define DISABLE_BATCHING 1
 
 static void gpencil_drawcall_flush(gpIterPopulateData *iter)
 {
@@ -377,6 +377,7 @@ static void gpencil_drawcall_add(
   else {
     DRW_shgroup_call_range(iter->grp, iter->ob, geom, v_first, v_count);
   }
+  return;
 #endif
 
   int last = iter->vfirst + iter->vcount;
@@ -516,22 +517,32 @@ static void gpencil_stroke_cache_populate(bGPDlayer *gpl,
 
   bool do_sbuffer = (iter->do_sbuffer_call == DRAW_NOW);
 
+  GPUBatch *geom = do_sbuffer ? DRW_cache_gpencil_sbuffer_get(iter->ob) :
+                                DRW_cache_gpencil_get(iter->ob, iter->pd->cfra);
+  if (geom != iter->geom) {
+    gpencil_drawcall_flush(iter);
+
+    GPUVertBuf *position_tx = do_sbuffer ?
+                                  DRW_cache_gpencil_sbuffer_position_buffer_get(iter->ob) :
+                                  DRW_cache_gpencil_position_buffer_get(iter->ob, iter->pd->cfra);
+    GPUVertBuf *color_tx = do_sbuffer ?
+                               DRW_cache_gpencil_sbuffer_color_buffer_get(iter->ob) :
+                               DRW_cache_gpencil_color_buffer_get(iter->ob, iter->pd->cfra);
+    DRW_shgroup_buffer_texture(iter->grp, "gp_pos_tx", position_tx);
+    DRW_shgroup_buffer_texture(iter->grp, "gp_col_tx", color_tx);
+  }
+
   if (show_fill) {
-    GPUBatch *geom = do_sbuffer ? DRW_cache_gpencil_sbuffer_fill_get(iter->ob) :
-                                  DRW_cache_gpencil_fills_get(iter->ob, iter->pd->cfra);
     int vfirst = gps->runtime.fill_start * 3;
     int vcount = gps->tot_triangles * 3;
     gpencil_drawcall_add(iter, geom, false, vfirst, vcount);
   }
 
   if (show_stroke) {
-    GPUBatch *geom = do_sbuffer ? DRW_cache_gpencil_sbuffer_stroke_get(iter->ob) :
-                                  DRW_cache_gpencil_strokes_get(iter->ob, iter->pd->cfra);
-    /* Start one vert before to have gl_InstanceID > 0 (see shader). */
-    int vfirst = gps->runtime.stroke_start - 1;
-    /* Include "potential" cyclic vertex and start adj vertex (see shader). */
-    int vcount = gps->totpoints + 1 + 1;
-    gpencil_drawcall_add(iter, geom, true, vfirst, vcount);
+    int vfirst = gps->runtime.stroke_start * 3;
+    /* Include "potential" cyclic vertex (see shader). */
+    int vcount = (gps->totpoints + 1) * 2 * 3;
+    gpencil_drawcall_add(iter, geom, false, vfirst, vcount);
   }
 
   iter->stroke_index_last = gps->runtime.stroke_start + gps->totpoints + 1;
diff --git a/source/blender/draw/engines/gpencil/gpencil_shader_shared.h b/source/blender/draw/engines/gpencil/gpencil_shader_shared.h
index 4c621e955b9..3f0f73e7c13 100644
--- a/source/blender/draw/engines/gpencil/gpencil_shader_shared.h
+++ b/source/blender/draw/engines/gpencil/gpencil_shader_shared.h
@@ -41,6 +41,9 @@ enum gpLightType {
   GP_LIGHT_TYPE_AMBIENT = 3u,
 };
 
+#define GP_IS_STROKE_VERTEX_BIT (1 << 30)
+#define GP_VERTEX_ID_SHIFT 2
+
 /* Avoid compiler funkiness with enum types not being strongly typed in C. */
 #ifndef GPU_SHADER
 #  define gpMaterialFlag uint
diff --git a/source/blender/draw/engines/gpencil/shaders/gpencil_common_lib.glsl b/source/blender/draw/engines/gpencil/shaders/gpencil_common_lib.glsl
index 7ddfdc5f65c..642939136c8 100644
--- a/source/blender/draw/engines/gpencil/shaders/gpencil_common_lib.glsl
+++ b/source/blender/draw/engines/gpencil/shaders/gpencil_common_lib.glsl
@@ -1,92 +1,4 @@
 
-/* Must match C declaration. */
-struct gpMaterial {
-  vec4 stroke_color;
-  vec4 fill_color;
-  vec4 fill_mix_color;
-  vec4 fill_uv_rot_scale;
-  vec4 fill_uv_offset;
-  /* Put float/int at the end to avoid padding error */
-  /* Some drivers are completely messing the alignment or the fetches here.
-   * We are forced to pack these into vec4 otherwise we only get 0.0 as value. */
-  vec4 gp_mat_packed_1;
-  // float stroke_texture_mix;
-  // float stroke_u_scale;
-  // float fill_texture_mix;
-  // int gp_flag;
-  /* Please ensure 16 byte alignment (multiple of vec4). */
-};
-
-#define MATERIAL(m) materials[m + gpMaterialOffset]
-
-#define stroke_texture_mix gp_mat_packed_1.x
-#define stroke_u_scale gp_mat_packed_1.y
-#define fill_texture_mix gp_mat_packed_1.z
-#define GP_FLAG(m) floatBitsToInt(MATERIAL(m).gp_mat_packed_1.w)
-
-/* flag */
-#define GP_STROKE_ALIGNMENT_STROKE 1
-#define GP_STROKE_ALIGNMENT_OBJECT 2
-#define GP_STROKE_ALIGNMENT_FIXED 3
-#define GP_STROKE_ALIGNMENT 0x3
-#define GP_STROKE_OVERLAP (1 << 2)
-#define GP_STROKE_TEXTURE_USE (1 << 3)
-#define GP_STROKE_TEXTURE_STENCIL (1 << 4)
-#define GP_STROKE_TEXTURE_PREMUL (1 << 5)
-#define GP_STROKE_DOTS (1 << 6)
-#define GP_STROKE_HOLDOUT (1 << 7)
-#define GP_FILL_HOLDOUT (1 << 8)
-#define GP_FILL_TEXTURE_USE (1 << 10)
-#define GP_FILL_TEXTURE_PREMUL (1 << 11)
-#define GP_FILL_TEXTURE_CLIP (1 << 12)
-#define GP_FILL_GRADIENT_USE (1 << 13)
-#define GP_FILL_GRADIENT_RADIAL (1 << 14)
-/* High bits are used to pass material ID to fragment shader. */
-#define GP_MATID_SHIFT 16
-
-/* Multiline defines can crash blender with certain GPU drivers. */
-/* clang-format off */
-#define GP_FILL_FLAGS (GP_FILL_TEXTURE_USE | GP_FILL_TEXTURE_PREMUL | GP_FILL_TEXTURE_CLIP | GP_FILL_GRADIENT_USE | GP_FILL_GRADIENT_RADIAL | GP_FILL_HOLDOUT)
-/* clang-format on */
-
-#define GP_FLAG_TEST(flag, val) (((flag) & (val)) != 0)
-
-/* Must match C declaration. */
-struct gpLight {
-  vec4 color_type;
-  vec4 right;
-  vec4 up;
-  vec4 forward;
-  vec4 position;
-  /* Please ensure 16 byte alignment (multiple of vec4). */
-};
-
-#define spot_size right.w
-#define spot_blend up.w
-
-#define GP_LIGHT_TYPE_POINT 0.0
-#define GP_LIGHT_TYPE_SPOT 1.0
-#define GP_LIGHT_TYPE_SUN 2.0
-#define GP_LIGHT_TYPE_AMBIENT 3.0
-
-#ifdef GP_MATERIAL_BUFFER_LEN
-
-layout(std140) uniform gpMaterialBlock
-{
-  gpMaterial materials[GP_MATERIAL_BUFFER_LEN];
-};
-
-#endif
-
-#ifdef GPENCIL_LIGHT_BUFFER_LEN
-
-layout(std140) uniform gpLightBlock
-{
-  gpLight lights[GPENCIL_LIGHT_BUFFER_LEN];
-};
-
-#endif
-
 /* Must match eGPLayerBlendModes */
 #define MODE_REGULAR 0
 #define MODE_HARDLIGHT 1
@@ -149,510 +61,3 @@ void blend_mode_output(
       break;
   }
 }
-
-#ifndef USE_GPU_SHADER_CREATE_INFO
-
-IN_OUT ShaderStageInterface
-{
-  vec4 finalColorMul;
-  vec4 finalColorAdd;
-  vec3 finalPos;
-  vec2 finalUvs;
-  noperspective float strokeThickness;
-  noperspective float unclampedThickness;
-  noperspective float strokeHardeness;
-  flat vec2 strokeAspect;
-  flat vec2 strokePt1;
-  flat vec2 strokePt2;
-  flat int matFlag;
-  flat float depth;
-};
-
-#endif
-
-#ifdef GPU_FRAGMENT_SHADER
-
-#  define linearstep(p0, p1, v) (clamp(((v) - (p0)) / abs((p1) - (p0)), 0.0, 1.0))
-
-float stroke_round_cap_mask(vec2 p1, vec2 p2, vec2 aspect, float thickness, float hardfac)
-{
-  /* We create our own uv space to avoid issues with triangulation and linear
-   * interpolation artifacts. */
-  vec2 line = p2.xy - p1.xy;
-  vec2 pos = gl_FragCoord.xy - p1.xy;
-  float line_len = length(line);
-  float half_line_len = line_len * 0.5;
-  /* Normalize */
-  line = (line_len > 0.0) ? (line / line_len) : vec2(1.0, 0.0);
-  /* Create a uv space that englobe the whole segment into a capsule. */
-  vec2 uv_end;
-  uv_end.x = max(abs(dot(line, pos) - half_line_len) - half_line_len, 0.0);
-  uv_end.y = dot(vec2(-line.y, line.x), pos);
-  /* Divide by stroke radius. */
-  uv_end /= thickness;
-  uv_end *= aspect;
-
-  float dist = clamp(1.0 - length(uv_end) * 2.0, 0.0, 1.0);
-  if (hardfac > 0.999) {
-    return step(1e-8, dist);
-  }
-  else {
-    /* Modulate the falloff profile */
-    float hardness = 1.0 - hardfac;
-    dist = pow(dist, mix(0.01, 10.0, hardness));
-    return smoothstep(0.0, 1.0, dist);
-  }
-}
-
-#endif
-
-uniform vec2 sizeViewport;
-uniform vec2 sizeViewportInv;
-
-/* Per Object */
-uniform bool strokeOrder3d;
-uniform int gpMaterialOffset;
-uniform float thicknessScale;
-uniform float thicknessWorldScale;
-#define thicknessIsScreenSpace (thicknessWorldScale < 0.0)
-
-#ifdef GPU_VERTEX_SHADER
-
-/* Per Layer */
-uniform float thicknessOffset;
-uniform float vertexColorOpacity;
-uniform vec4 layerTint;
-uniform float layerOpacity; /* Used for onion skin. */
-uniform float strokeIndexOffset = 0.0;
-
-/* All of these attributes are quad loaded the same way
- * as GL_LINES_ADJACENCY would feed a geometry shader:
- * - ma reference the previous adjacency point.
- * - ma1 reference the current line first point.
- * - ma2 reference the current line second point.
- * - ma3 reference the next adjacency point.
- * Note that we are rendering quad instances and not using any index buffer (except for fills).
- */
-/* x is material index, y is stroke_id, z is point_id, w is aspect & rotation & hardness packed. */
-in ivec4 ma;
-in ivec4 ma1;
-in ivec4 ma2;
-in ivec4 ma3;
-/* Position contains thickness in 4th component. */
-in vec4 pos;  /* Prev adj vert */
-in vec4 pos1; /* Current edge */
-in vec4 pos2; /* Current edge */
-in vec4 pos3; /* Next adj vert */
-/* xy is UV for fills, z is U of stroke, w is strength. */
-in vec4 uv1;
-in vec4 uv2;
-in vec4 col1;
-in vec4 col2;
-in vec4 fcol1;
-/* WARNING: Max attribute count is actually 14 because OSX OpenGL implementation
- * considers gl_VertexID and gl_InstanceID as vertex attribute. (see T74536) */
-#  define stroke_id1 ma1.y
-#  define point_id1 ma1.z
-#  define thickness1 pos1.w
-#  define thickness2 pos2.w
-#  define strength1 uv1.w
-#  define strength2 uv2.w
-/* Packed! need to be decoded. */
-#  define hardness1 ma1.w
-#  define hardness2 ma2.w
-#  define uvrot1 ma1.w
-#  define aspect1 ma1.w
-
-vec2 decode_aspect(int packed_data)
-{
-  float asp = float(uint(packed_data) & 0x1FFu) * (1.0 / 255.0);
-  return (asp > 1.0) ? vec2(1.0, (asp - 1.0)) : vec2(asp, 1.0);
-}
-
-float decode_uvrot(int packed_data)
-{
-  uint udata = uint(packed_data);
-  float uvrot = 1e-8 + float((udata & 0x1FE00u) >> 9u) * (1.0 / 255.0);
-  return ((udata & 0x20000u) != 0u) ? -uvrot : uvrot;
-}
-
-float decode_hardness(int packed_data)
-{
-  return float((uint(packed_data) & 0x3FC0000u) >> 18u) * (1.0 / 255.0);
-}
-
-void discard_vert()
-{
-  /* We set the vertex at the camera origin to generate 0 fragments. */
-  gl_Position = vec4(0.0, 0.0, -3e36, 0.0);
-}
-
-vec2 project_to_screenspace(vec4 v)
-{
-  return ((v.xy / v.w) * 0.5 + 0.5) * sizeViewport;
-}
-
-vec2 rotate_90deg(vec2 v)
-{
-  /* Counter Clock-Wise. */
-  return vec2(-v.y, v.x);
-}
-
-mat4 model_matrix_get()
-{
-  return ModelMatrix;
-}
-
-vec3 transform_point(mat4 m, vec3 v)
-{
-  return (m * vec4(v, 1.0)).xyz;
-}
-
-vec2 safe_normalize(vec2 v)
-{
-  float len_sqr = dot(v, v);
-  if (len_sqr > 0.0) {
-    return v / sqrt(len_sqr);
-  }
-  else {
-    return vec2(1.0, 0.0);
-  }
-}
-
-vec2 safe_normalize_len(vec2 v, out float len)
-{
-  len = sqrt(dot(v, v));
-  if (len > 0.0) {
-    return v / len;
-  }
-  else {
-    return vec2(1.0, 0.0);
-  }
-}
-
-float stroke_thickness_modulate(float thickness)
-{
-  /* Modify stroke thickness by object and layer factors. */
-  thickness *= thicknessScale;
-  thickness += thicknessOffset;
-  thickness = max(1.0, thickness);
-
-  if (thicknessIsScreenSpace) {
-    /* Multiply offset by view Z so that offset is constant in screenspace.
-     * (e.i: does not change with the distance to camera) */
-    thickness *= gl_Position.w;
-  }
-  else {
-    /* World space point size. */
-    thickness *= thicknessWorldScale * drw_view.winmat[1][1] * sizeViewport.y;
-  }
-  return thickness;
-}
-
-float clamp_small_stroke_thickness(float thickness)
-{
-  /* To avoid aliasing artifacts, we clamp the line thickness and
-   * reduce its opacity in the fragment shader. */
-  float min_thickness = gl_Position.w * 1.3;
-  thickness = max(min_thickness, thickness);
-
-  return thickness;
-}
-
-#  ifdef GP_MATERIAL_BUFFER_LEN
-void color_output(vec4 stroke_col, vec4 vert_col, float vert_strength, float mix_tex)
-{
-  /* Mix stroke with other colors. */
-  vec4 mixed_col = stroke_col;
-  mixed_col.rgb = mix(mixed_col.rgb, vert_col.rgb, vert_col.a * vertexColorOpacity);
-  mixed_col.rgb = mix(mixed_col.rgb, layerTint.rgb, layerTint.a);
-  mixed_col.a *= vert_strength * layerOpacity;
-  /**
-   * This is what the fragment shader looks like.
-   * out = col * finalColorMul + col.a * finalColorAdd.
-   * finalColorMul is how much of the texture color to keep.
-   * finalColorAdd is how much of the mixed color to add.
-   * Note that we never add alpha. This is to keep the texture act as a stencil.
-   * We do however, modulate the alpha (reduce it).
-   */
-  /* We add the mixed color. This is 100% mix (no texture visible). */
-  finalColorMul = vec4(mixed_col.aaa, mixed_col.a);
-  finalColorAdd = vec4(mixed_col.rgb * mixed_col.a, 0.0);
-  /* Then we blend according to the texture mix factor.
-   * Note that we keep the alpha modulation. */
-  finalColorMul.rgb *= mix_tex;
-  finalColorAdd.rgb *= 1.0 - mix_tex;
-}
-#  endif
-
-void stroke_vertex()
-{
-  int m = ma1.x;
-  bool is_dot = false;
-  bool is_squares = false;
-
-#  ifdef GP_MATERIAL_BUFFER_LEN
-  if (m != -1) {
-    is_dot = GP_FLAG_TEST(GP_FLAG(m), GP_STROKE_ALIGNMENT);
-    is_squares = !GP_FLAG_TEST(GP_FLAG(m), GP_STROKE_DOTS);
-  }
-#  endif
-
-  /* Special Case. Stroke with single vert are rendered as dots. Do not discard them. */
-  if (!is_dot && ma.x == -1 && ma2.x == -1) {
-    is_dot = true;
-    is_squares = false;
-  }
-
-  /* Endpoints, we discard the vertices. */
-  if (ma1.x == -1 || (!is_dot && ma2.x == -1)) {
-    discard_vert();
-    return;
-  }
-
-  mat4 model_mat = model_matrix_get();
-
-  /* Avoid using a vertex attribute for quad positioning. */
-  float x = float(gl_VertexID & 1) * 2.0 - 1.0; /* [-1..1] */
-  float y = float(gl_VertexID & 2) - 1.0;       /* [-1..1] */
-
-  bool use_curr = is_dot || (x == -1.0);
-
-  vec3 wpos_adj = transform_point(model_mat, (use_curr) ? pos.xyz : pos3.xyz);
-  vec3 wpos1 = transform_point(model_mat, pos1.xyz);
-  vec3 wpos2 = transform_point(model_mat, pos2.xyz);
-
-  vec4 ndc_adj = point_world_to_ndc(wpos_adj);
-  vec4 ndc1 = point_world_to_ndc(wpos1);
-  vec4 ndc2 = point_world_to_ndc(wpos2);
-
-  gl_Position = (use_curr) ? ndc1 : ndc2;
-  finalPos = (use_curr) ? wpos1 : wpos2;
-
-  vec2 ss_adj = project_to_screenspace(ndc_adj);
-  vec2 ss1 = project_to_screenspace(ndc1);
-  vec2 ss2 = project_to_screenspace(ndc2);
-  /* Screenspace Lines tangents. */
-  float line_len;
-  vec2 line = safe_normalize_len(ss2 - ss1, line_len);
-  vec2 line_adj = safe_normalize((use_curr) ? (ss1 - ss_adj) : (ss_adj - ss2));
-
-  float thickness = abs((use_curr) ? thickness1 : thickness2);
-  thickness = stroke_thickness_modulate(thickness);
-  float clampedThickness = clamp_small_stroke_thickness(thickness);
-
-  finalUvs = vec2(x, y) * 0.5 + 0.5;
-  strokeHardeness = decode_hardness(use_curr ? hardness1 : hardness2);
-
-  if (is_dot) {
-#  ifdef GP_MATERIAL_BUFFER_LEN
-    int alignement = GP_FLAG(m) & GP_STROKE_ALIGNMENT;
-    /* For one point strokes use object alignment. */
-    if (ma.x == -1 && ma2.x == -1 && alignement == GP_STROKE_ALIGNMENT_STROKE) {
-      alignement = GP_STROKE_ALIGNMENT_OBJECT;
-    }
-#  endif
-
-    vec2 x_axis;
-#  ifdef GP_MATERIAL_BUFFER_LEN
-    if (alignement == GP_STROKE_ALIGNMENT_STROKE) {
-      x_axis = (ma2.x == -1) ? line_adj : line;
-    }
-    else if (alignement == GP_STROKE_ALIGNMENT_FIXED) {
-      /* Default for no-material drawing. */
-      x_axis = vec2(1.0, 0.0);
-    }
-    else
-#  endif
-    { /* GP_STROKE_ALIGNMENT_OBJECT */
-      vec4 ndc_x = point_world_to_ndc(wpos1 + model_mat[0].xyz);
-      vec2 ss_x = project_to_screenspace(ndc_x);
-      x_axis = safe_normalize(ss_x - ss1);
-    }
-
-    /* Rotation: Encoded as Cos + Sin sign. */
-    float uv_rot = decode_uvrot(uvrot1);
-    float rot_sin = sqrt(max(0.0, 1.0 - uv_rot * uv_rot)) * sign(uv_rot);
-    float rot_cos = abs(uv_rot);
-    x_axis = mat2(rot_cos, -rot_sin, rot_sin, rot_cos) * x_axis;
-
-#  ifdef GP_MATERIAL_BUFFER_LEN
-    if (is_dot) {
-      float alignment_cos = MATERIAL(m).fill_uv_offset.z;
-      float alignment_sin = MATERIAL(m).fill_uv_offset.w;
-      x_axis = mat2(alignment_cos, -alignment_sin, alignment_sin, alignment_cos) * x_axis;
-    }
-#  endif
-
-    vec2 y_axis = rotate_90deg(x_axis);
-
-    strokeAspect = decode_aspect(aspect1);
-
-    x *= strokeAspect.x;
-    y *= strokeAspect.y;
-
-    /* Invert for vertex shader. */
-    strokeAspect = 1.0 / strokeAspect;
-
-    gl_Position.xy += (x * x_axis + y * y_axis) * sizeViewportInv.xy * clampedThickness;
-
-    strokePt1 = ss1;
-    strokePt2 = ss1 + x_axis * 0.5;
-    strokeThickness = (is_squares) ? 1e18 : (clampedThickness / gl_Position.w);
-    unclampedThickness = (is_squares) ? 1e18 : (thickness / gl_Position.w);
-  }
-  else {
-    bool is_stroke_start = (ma.x == -1 && x == -1);
-    bool is_stroke_end = (ma3.x == -1 && x == 1);
-
-    /* Mitter tangent vector. */
-    vec2 miter_tan = safe_normalize(line_adj + line);
-    float miter_dot = dot(miter_tan, line_adj);
-    /* Break corners after a certain angle to avoid really thick corners. */
-    const float miter_limit = 0.5; /* cos(60°) */
-    bool miter_break = (miter_dot < miter_limit);
-    miter_tan = (miter_break || is_stroke_start || is_stroke_end) ? line : (miter_tan / miter_dot);
-
-    vec2 miter = rotate_90deg(miter_tan);
-
-    strokePt1.xy = ss1;
-    strokePt2.xy = ss2;
-    strokeThickness = clampedThickness / gl_Position.w;
-    unclampedThickness = thickness / gl_Position.w;
-    strokeAspect = vec2(1.0);
-
-    vec2 screen_ofs = miter * y;
-
-    /* Reminder: we packed the cap flag into the sign of strength and thickness sign. */
-    if ((is_stroke_start && strength1 > 0.0) || (is_stroke_end && thickness1 > 0.0) ||
-        (miter_break && !is_stroke_start && !is_stroke_end)) {
-      screen_ofs += line * x;
-    }
-
-    gl_Position.xy += screen_ofs * sizeViewportInv.xy * clampedThickness;
-
-    finalUvs.x = (use_curr) ? uv1.z : uv2.z;
-#  ifdef GP_MATERIAL_BUFFER_LEN
-    finalUvs.x *= MATERIAL(m).stroke_u_scale;
-#  endif
-  }
-
-#  ifdef GP_MATERIAL_BUFFER_LEN
-  vec4 vert_col = (use_curr) ? col1 : col2;
-  float vert_strength = abs((use_curr) ? strength1 : strength2);
-  vec4 stroke_col = MATERIAL(m).stroke_color;
-  float mix_tex = MATERIAL(m).stroke_texture_mix;
-
-  /* Special case: We don't use vertex color if material Holdout. */
-  if (GP_FLAG_TEST(GP_FLAG(m), GP_STROKE_HOLDOUT)) {
-    vert_col = vec4(0.0);
-  }
-
-  color_output(stroke_col, vert_col, vert_strength, mix_tex);
-
-  matFlag = GP_FLAG(m) & ~GP_FILL_FLAGS;
-#  endif
-
-  if (strokeOrder3d) {
-    /* Use the fragment depth (see fragment shader). */
-    depth = -1.0;
-  }
-#  ifdef GP_MATERIAL_BUFFER_LEN
-  else if (GP_FLAG_TEST(GP_FLAG(m), GP_STROKE_OVERLAP)) {
-    /* Use the index of the point as depth.
-     * This means the stroke can overlap itself. */
-    depth = (point_id1 + strokeIndexOffset + 1.0) * 0.0000002;
-  }
-#  endif
-  else {
-    /* Use the index of first point of the stroke as depth.
-     * We render using a greater depth test this means the stroke
-     * cannot overlap itself.
-     * We offset by one so that the fill can be overlapped by its stroke.
-     * The offset is ok since we pad the strokes data because of adjacency infos. */
-    depth = (stroke_id1 + strokeIndexOffset + 1.0) * 0.0000002;
-  }
-}
-
-void fill_vertex()
-{
-  mat4 model_mat = model_matrix_get();
-
-  vec3 wpos = transform_point(model_mat, pos1.xyz);
-  gl_Position = point_world_to_ndc(wpos);
-  finalPos = wpos;
-
-#  ifdef GP_MATERIAL_BUFFER_LEN
-  int m = ma1.x;
-
-  vec4 fill_col = MATERIAL(m).fill_color;
-  float mix_tex = MATERIAL(m).fill_texture_mix;
-
-  /* Special case: We don't modulate alpha in gradient mode. */
-  if (GP_FLAG_TEST(GP_FLAG(m), GP_FILL_GRADIENT_USE)) {
-    fill_col.a = 1.0;
-  }
-
-  /* Decode fill opacity. */
-  vec4 fcol_decode = vec4(fcol1.rgb, floor(fcol1.a / 10.0));
-  float fill_opacity = fcol1.a - (fcol_decode.a * 10);
-  fcol_decode.a /= 10000.0;
-
-  /* Special case: We don't use vertex color if material Holdout. */
-  if (GP_FLAG_TEST(GP_FLAG(m), GP_FILL_HOLDOUT)) {
-    fcol_decode = vec4(0.0);
-  }
-
-  /* Apply opacity. */
-  fill_col.a *= fill_opacity;
-  /* If factor is > 1 force opacity. */
-  if (fill_opacity > 1.0) {
-    fill_col.a += fill_opacity - 1.0;
-  }
-
-  fill_col.a = clamp(fill_col.a, 0.0, 1.0);
-
-  color_output(fill_col, fcol_decode, 1.0, mix_tex);
-
-  matFlag = GP_FLAG(m) & GP_FILL_FLAGS;
-  matFlag |= m << GP_MATID_SHIFT;
-
-  vec2 loc = MATERIAL(m).fill_uv_offset.xy;
-  mat2x2 rot_scale = mat2x2(MATERIAL(m).fill_uv_rot_scale.xy, MATERIAL(m).fill_uv_rot_scale.zw);
-  finalUvs = rot_scale * uv1.xy + loc;
-#  endif
-
-  strokeHardeness = 1.0;
-  strokeThickness = 1e18;
-  unclampedThickness = 1e20;
-  strokeAspect = vec2(1.0);
-  strokePt1 = strokePt2 = vec2(0.0);
-
-  if (strokeOrder3d) {
-    /* Use the fragment depth (see fragment shader). */
-    depth = -1.0;
-    /* We still offset the fills a little to avoid overlaps */
-    gl_Position.z += 0.000002;
-  }
-  else {
-    /* Use the index of first point of the stroke as depth. */
-    depth = (stroke_id1 + strokeIndexOffset) * 0.0000002;
-  }
-}
-
-void gpencil_vertex()
-{
-  /* Trick to detect if a drawcall is stroke or fill.
-   * This does mean that we need to draw an empty stroke segment before starting
-   * to draw the real stroke segments. */
-  bool is_fill = (gl_InstanceID == 0);
-
-  if (!is_fill) {
-    stroke_vertex();
-  }
-  else {
-    fill_vertex();
-  }
-}
-
-#endif
diff --git a/source/blender/draw/engines/gpencil/shaders/gpencil_vert.glsl b/source/blender/draw/engines/gpencil/shaders/gpencil_vert.glsl
index 8ed03b23809..9b1db09ab3c 100644
--- a/source/blender/draw/engines/gpencil/shaders/gpencil_vert.glsl
+++ b/source/blender/draw/engines/gpencil/shaders/gpencil_vert.glsl
@@ -31,23 +31,11 @@ void main()
   vec4 vert_color;
   vec3 vert_N;
 
+  ivec4 ma1 = floatBitsToInt(texelFetch(gp_pos_tx, gpencil_stroke_point_id() * 3 + 1));
   gpMaterial gp_mat = materials[ma1.x + gpMaterialOffset];
   gpMaterialFlag gp_flag = floatBitsToUint(gp_mat._flag);
 
-  gl_Position = gpencil_vertex(ma,
-                               ma1,
-                               ma2,
-                               ma3,
-                               pos,
-                               pos1,
-                               pos2,
-                               pos3,
-                               uv1,
-                               uv2,
-                               col1,
-                               col2,
-                               fcol1,
-                               vec4(viewportSize, 1.0 / viewportSize),
+  gl_Position = gpencil_vertex(vec4(viewportSize, 1.0 / viewportSize),
                                gp_flag,
                                gp_mat._alignment_rot,
                                gp_interp.pos,
@@ -60,7 +48,7 @@ void main()
                                gp_interp.thickness,
                                gp_interp.hardness);
 
-  if (GPENCIL_IS_STROKE_VERTEX) {
+  if (gpencil_is_stroke_vertex()) {
     if (!flag_test(gp_flag, GP_STROKE_ALIGNMENT)) {
       gp_interp.uv.x *= gp_mat._stroke_u_scale;
     }
@@ -96,6 +84,9 @@ void main()
     }
   }
   else {
+    int stroke_point_id = gpencil_stroke_point_id();
+    vec4 uv1 = texelFetch(gp_col_tx, stroke_point_id * 2 + 2);
+    vec4 fcol1 = texelFetch(gp_col_tx, stroke_point_id * 2 + 1);
     vec4 fill_col = gp_mat.fill_color;
 
     /* Special case: We don't modulate alpha in gradient mode. */
diff --git a/source/blender/draw/engines/gpencil/shaders/infos/gpencil_info.hh b/source/blender/draw/engines/gpencil/shaders/infos/gpencil_info.hh
index edd51e71242..da2776254e6 100644
--- a/source/blender/draw/engines/gpencil/shaders/infos/gpencil_info.hh
+++ b/source/blender/draw/engines/gpencil/shaders/infos/gpencil_info.hh
@@ -22,10 +22,10 @@ GPU_SHADER_CREATE_INFO(gpencil_geometry)
     .do_static_compilation(true)
     .define("GP_LIGHT")
     .typedef_source("gpencil_defines.h")
-    .sampler(0, ImageType::FLOAT_2D, "gpFillTexture")
-    .sampler(1, ImageType::FLOAT_2D, "gpStrokeTexture")
-    .sampler(2, ImageType::DEPTH_2D, "gpSceneDepthTexture")
-    .sampler(3, ImageType::FLOAT_2D, "gpMaskTexture")
+    .sampler(2, ImageType::FLOAT_2D, "gpFillTexture")
+    .sampler(3, ImageType::FLOAT_2D, "gpStrokeTexture")
+    .sampler(4, ImageType::DEPTH_2D, "gpSceneDepthTexture")
+    .sampler(5, ImageType::FLOAT_2D, "gpMaskTexture")
     .uniform_buf(4, "gpMaterial", "materials[GPENCIL_MATERIAL_BUFFER_LEN]", Frequency::BATCH)
     .uniform_buf(3, "gpLight", "lights[GPENCIL_LIGHT_BUFFER_LEN]", Frequency::BATCH)
     .push_constant(Type::VEC2, "viewportSize")
diff --git a/source/blender/draw/engines/overlay/overlay_outline.cc b/source/blender/draw/engines/overlay/overlay_outline.cc
index 5ea02376b67..50d42effe00 100644
--- a/source/blender/draw/engines/overlay/overlay_outline.cc
+++ b/source/blender/draw/engines/overlay/overlay_outline.cc
@@ -172,7 +172,6 @@ void OVERLAY_outline_cache_init(OVERLAY_Data *vedata)
 typedef struct iterData {
   Object *ob;
   DRWShadingGroup *stroke_grp;
-  DRWShadingGroup *fill_grp;
   int cfra;
   float plane[4];
 } iterData;
@@ -193,12 +192,17 @@ static void gpencil_layer_cache_populate(bGPDlayer *gpl,
    * Convert to world units (by default, 1 meter = 2000 pixels). */
   float thickness_scale = (is_screenspace) ? -1.0f : (gpd->pixfactor / 2000.0f);
 
+  GPUVertBuf *position_tx = DRW_cache_gpencil_position_buffer_get(iter->ob, iter->cfra);
+  GPUVertBuf *color_tx = DRW_cache_gpencil_color_buffer_get(iter->ob, iter->cfra);
+
   DRWShadingGroup *grp = iter->stroke_grp = DRW_shgroup_create_sub(iter->stroke_grp);
   DRW_shgroup_uniform_bool_copy(grp, "gpStrokeOrder3d", is_stroke_order_3d);
   DRW_shgroup_uniform_float_copy(grp, "gpThicknessScale", object_scale);
   DRW_shgroup_uniform_float_copy(grp, "gpThicknessOffset", float(gpl->line_change));
   DRW_shgroup_uniform_float_copy(grp, "gpThicknessWorldScale", thickness_scale);
   DRW_shgroup_uniform_vec4_copy(grp, "gpDepthPlane", iter->plane);
+  DRW_shgroup_buffer_texture(grp, "gp_pos_tx", position_tx);
+  DRW_shgroup_buffer_texture(grp, "gp_col_tx", color_tx);
 }
 
 static void gpencil_stroke_cache_populate(bGPDlayer * /*gpl*/,
@@ -219,20 +223,19 @@ static void gpencil_stroke_cache_populate(bGPDlayer * /*gpl*/,
     return;
   }
 
+  struct GPUBatch *geom = DRW_cache_gpencil_get(iter->ob, iter->cfra);
+
   if (show_fill) {
-    struct GPUBatch *geom = DRW_cache_gpencil_fills_get(iter->ob, iter->cfra);
     int vfirst = gps->runtime.fill_start * 3;
     int vcount = gps->tot_triangles * 3;
-    DRW_shgroup_call_range(iter->fill_grp, iter->ob, geom, vfirst, vcount);
+    DRW_shgroup_call_range(iter->stroke_grp, iter->ob, geom, vfirst, vcount);
   }
 
   if (show_stroke) {
-    struct GPUBatch *geom = DRW_cache_gpencil_strokes_get(iter->ob, iter->cfra);
-    /* Start one vert before to have gl_InstanceID > 0 (see shader). */
-    int vfirst = gps->runtime.stroke_start - 1;
-    /* Include "potential" cyclic vertex and start adj vertex (see shader). */
-    int vcount = gps->totpoints + 1 + 1;
-    DRW_shgroup_call_instance_range(iter->stroke_grp, iter->ob, geom, vfirst, vcount);
+    int vfirst = gps->runtime.stroke_start * 3;
+    /* Include "potential" cyclic vertex (see shader). */
+    int vcount = (gps->totpoints + 1) * 2 * 3;
+    DRW_shgroup_call_range(iter->stroke_grp, iter->ob, geom, vfirst, vcount);
   }
 }
 
@@ -247,7 +250,6 @@ static void OVERLAY_outline_gpencil(OVERLAY_PrivateData *pd, Object *ob)
   iterData iter{};
   iter.ob = ob;
   iter.stroke_grp = pd->outlines_gpencil_grp;
-  iter.fill_grp = DRW_shgroup_create_sub(pd->outlines_gpencil_grp);
   iter.cfra = pd->cfra;
 
   if (gpd->draw_mode == GP_DRAWMODE_2D) {
diff --git a/source/blender/draw/engines/overlay/shaders/overlay_outline_prepass_gpencil_vert.glsl b/source/blender/draw/engines/overlay/shaders/overlay_outline_prepass_gpencil_vert.glsl
index 7d8fb0c2cb8..851e0884354 100644
--- a/source/blender/draw/engines/overlay/shaders/overlay_outline_prepass_gpencil_vert.glsl
+++ b/source/blender/draw/engines/overlay/shaders/overlay_outline_prepass_gpencil_vert.glsl
@@ -34,20 +34,7 @@ void main()
   float unused_strength;
   vec2 unused_uv;
 
-  gl_Position = gpencil_vertex(ma,
-                               ma1,
-                               ma2,
-                               ma3,
-                               pos,
-                               pos1,
-                               pos2,
-                               pos3,
-                               uv1,
-                               uv2,
-                               col1,
-                               col2,
-                               fcol1,
-                               vec4(sizeViewport, sizeViewportInv),
+  gl_Position = gpencil_vertex(vec4(sizeViewport, sizeViewportInv),
                                world_pos,
                                unused_N,
                                unused_color,
diff --git a/source/blender/draw/intern/draw_cache.h b/source/blender/draw/intern/draw_cache.h
index 058f28f094d..44abf3c7064 100644
--- a/source/blender/draw/intern/draw_cache.h
+++ b/source/blender/draw/intern/draw_cache.h
@@ -13,6 +13,7 @@ extern "C" {
 
 struct GPUBatch;
 struct GPUMaterial;
+struct GPUVertBuf;
 struct ModifierData;
 struct Object;
 struct PTCacheEdit;
@@ -257,14 +258,16 @@ struct GPUBatch *DRW_cache_volume_selection_surface_get(struct Object *ob);
 
 /* GPencil */
 
-struct GPUBatch *DRW_cache_gpencil_strokes_get(struct Object *ob, int cfra);
-struct GPUBatch *DRW_cache_gpencil_fills_get(struct Object *ob, int cfra);
+struct GPUBatch *DRW_cache_gpencil_get(struct Object *ob, int cfra);
+struct GPUVertBuf *DRW_cache_gpencil_position_buffer_get(struct Object *ob, int cfra);
+struct GPUVertBuf *DRW_cache_gpencil_color_buffer_get(struct Object *ob, int cfra);
 struct GPUBatch *DRW_cache_gpencil_edit_lines_get(struct Object *ob, int cfra);
 struct GPUBatch *DRW_cache_gpencil_edit_points_get(struct Object *ob, int cfra);
 struct GPUBatch *DRW_cache_gpencil_edit_curve_handles_get(struct Object *ob, int cfra);
 struct GPUBatch *DRW_cache_gpencil_edit_curve_points_get(struct Object *ob, int cfra);
-struct GPUBatch *DRW_cache_gpencil_sbuffer_stroke_get(struct Object *ob);
-struct GPUBatch *DRW_cache_gpencil_sbuffer_fill_get(struct Object *ob);
+struct GPUBatch *DRW_cache_gpencil_sbuffer_get(struct Object *ob);
+struct GPUVertBuf *DRW_cache_gpencil_sbuffer_position_buffer_get(struct Object *ob);
+struct GPUVertBuf *DRW_cache_gpencil_sbuffer_color_buffer_get(struct Object *ob);
 int DRW_gpencil_material_count_get(struct bGPdata *gpd);
 
 struct GPUBatch *DRW_cache_gpencil_face_wireframe_get(struct Object *ob);
diff --git a/source/blender/draw/intern/draw_cache_impl_gpencil.cc b/source/blender/draw/intern/draw_cache_impl_gpencil.cc
index 94925bfae71..eadb7772f4c 100644
--- a/source/blender/draw/intern/draw_cache_impl_gpencil.cc
+++ b/source/blender/draw/intern/draw_cache_impl_gpencil.cc
@@ -23,12 +23,14 @@
 #include "DEG_depsgraph_query.h"
 
 #include "BLI_hash.h"
+#include "BLI_math_vec_types.hh"
 #include "BLI_polyfill_2d.h"
 
 #include "draw_cache.h"
 #include "draw_cache_impl.h"
 
 #include "../engines/gpencil/gpencil_defines.h"
+#include "../engines/gpencil/gpencil_shader_shared.h"
 
 #define BEZIER_HANDLE (1 << 3)
 #define COLOR_SHIFT 5
@@ -41,11 +43,13 @@ typedef struct GpencilBatchCache {
   /** Instancing Data */
   GPUVertBuf *vbo;
   GPUVertBuf *vbo_col;
-  /** Fill Topology */
+  /** Indices in material order, then stroke order with fill first.
+   * Strokes can be individually rendered using `gps->runtime.stroke_start` and
+   * `gps->runtime.stroke_start`. */
   GPUIndexBuf *ibo;
-  /** Instancing Batches */
-  GPUBatch *stroke_batch;
-  GPUBatch *fill_batch;
+  /** Batches */
+  GPUBatch *geom_batch;
+  /** Stroke lines only */
   GPUBatch *lines_batch;
 
   /** Edit Mode */
@@ -117,8 +121,7 @@ static void gpencil_batch_cache_clear(GpencilBatchCache *cache)
   }
 
   GPU_BATCH_DISCARD_SAFE(cache->lines_batch);
-  GPU_BATCH_DISCARD_SAFE(cache->fill_batch);
-  GPU_BATCH_DISCARD_SAFE(cache->stroke_batch);
+  GPU_BATCH_DISCARD_SAFE(cache->geom_batch);
   GPU_VERTBUF_DISCARD_SAFE(cache->vbo);
   GPU_VERTBUF_DISCARD_SAFE(cache->vbo_col);
   GPU_INDEXBUF_DISCARD_SAFE(cache->ibo);
@@ -173,9 +176,10 @@ void DRW_gpencil_batch_cache_free(bGPdata *gpd)
 
 /* MUST match the format below. */
 typedef struct gpStrokeVert {
-  int32_t mat, stroke_id, point_id, packed_asp_hard_rot;
   /** Position and thickness packed in the same attribute. */
   float pos[3], thickness;
+  /** Material Index, Stroke Index, Point Index, Packed aspect + hardness + rotation. */
+  int32_t mat, stroke_id, point_id, packed_asp_hard_rot;
   /** UV and strength packed in the same attribute. */
   float uv_fill[2], u_stroke, strength;
 } gpStrokeVert;
@@ -184,12 +188,9 @@ static GPUVertFormat *gpencil_stroke_format(void)
 {
   static GPUVertFormat format = {0};
   if (format.attr_len == 0) {
-    GPU_vertformat_attr_add(&format, "ma", GPU_COMP_I32, 4, GPU_FETCH_INT);
     GPU_vertformat_attr_add(&format, "pos", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
+    GPU_vertformat_attr_add(&format, "ma", GPU_COMP_I32, 4, GPU_FETCH_INT);
     GPU_vertformat_attr_add(&format, "uv", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
-    /* IMPORTANT: This means having only 4 attributes
-     * to fit into GPU module limit of 16 attributes. */
-    GPU_vertformat_multiload_enable(&format, 4);
   }
   return &format;
 }
@@ -239,9 +240,6 @@ static GPUVertFormat *gpencil_color_format(void)
   if (format.attr_len == 0) {
     GPU_vertformat_attr_add(&format, "col", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
     GPU_vertformat_attr_add(&format, "fcol", GPU_COMP_F32, 4, GPU_FETCH_FLOAT);
-    /* IMPORTANT: This means having only 4 attributes
-     * to fit into GPU module limit of 16 attributes. */
-    GPU_vertformat_multiload_enable(&format, 4);
   }
   return &format;
 }
@@ -296,7 +294,8 @@ BLI_INLINE int32_t pack_rotation_aspect_hardness(float rot, float asp, float har
   return packed;
 }
 
-static void gpencil_buffer_add_point(gpStrokeVert *verts,
+static void gpencil_buffer_add_point(GPUIndexBufBuilder *ibo,
+                                     gpStrokeVert *verts,
                                      gpColorVert *cols,
                                      const bGPDstroke *gps,
                                      const bGPDspoint *pt,
@@ -330,27 +329,36 @@ static void gpencil_buffer_add_point(gpStrokeVert *verts,
 
   vert->packed_asp_hard_rot = pack_rotation_aspect_hardness(
       pt->uv_rot, aspect_ratio, gps->hardeness);
+
+  if (!is_endpoint) {
+    /* Issue a Quad per point. */
+    /* The attribute loading uses a different shader and will undo this bit packing. */
+    int v_mat = (v << GP_VERTEX_ID_SHIFT) | GP_IS_STROKE_VERTEX_BIT;
+    GPU_indexbuf_add_tri_verts(ibo, v_mat + 0, v_mat + 1, v_mat + 2);
+    GPU_indexbuf_add_tri_verts(ibo, v_mat + 2, v_mat + 1, v_mat + 3);
+  }
 }
 
-static void gpencil_buffer_add_stroke(gpStrokeVert *verts,
+static void gpencil_buffer_add_stroke(GPUIndexBufBuilder *ibo,
+                                      gpStrokeVert *verts,
                                       gpColorVert *cols,
                                       const bGPDstroke *gps)
 {
   const bGPDspoint *pts = gps->points;
   int pts_len = gps->totpoints;
   bool is_cyclic = gpencil_stroke_is_cyclic(gps);
-  int v = gps->runtime.stroke_start;
+  int v = gps->runtime.vertex_start;
 
   /* First point for adjacency (not drawn). */
   int adj_idx = (is_cyclic) ? (pts_len - 1) : min_ii(pts_len - 1, 1);
-  gpencil_buffer_add_point(verts, cols, gps, &pts[adj_idx], v++, true);
+  gpencil_buffer_add_point(ibo, verts, cols, gps, &pts[adj_idx], v++, true);
 
   for (int i = 0; i < pts_len; i++) {
-    gpencil_buffer_add_point(verts, cols, gps, &pts[i], v++, false);
+    gpencil_buffer_add_point(ibo, verts, cols, gps, &pts[i], v++, false);
   }
   /* Draw line to first point to complete the loop for cyclic strokes. */
   if (is_cyclic) {
-    gpencil_buffer_add_point(verts, cols, gps, &pts[0], v, false);
+    gpencil_buffer_add_point(ibo, verts, cols, gps, &pts[0], v, false);
     /* UV factor needs to be adjusted for the last point to not be equal to the UV factor of the
      * first point. It should be the factor of the last point plus the distance from the last point
      * to the first.
@@ -361,16 +369,20 @@ static void gpencil_buffer_add_stroke(gpStrokeVert *verts,
   }
   /* Last adjacency point (not drawn). */
   adj_idx = (is_cyclic) ? 1 : max_ii(0, pts_len - 2);
-  gpencil_buffer_add_point(verts, cols, gps, &pts[adj_idx], v++, true);
+  gpencil_buffer_add_point(ibo, verts, cols, gps, &pts[adj_idx], v++, true);
 }
 
 static void gpencil_buffer_add_fill(GPUIndexBufBuilder *ibo, const bGPDstroke *gps)
 {
   int tri_len = gps->tot_triangles;
-  int v = gps->runtime.stroke_start;
+  int v = gps->runtime.vertex_start + 1;
   for (int i = 0; i < tri_len; i++) {
     uint *tri = gps->triangles[i].verts;
-    GPU_indexbuf_add_tri_verts(ibo, v + tri[0], v + tri[1], v + tri[2]);
+    /* The attribute loading uses a different shader and will undo this bit packing. */
+    GPU_indexbuf_add_tri_verts(ibo,
+                               (v + tri[0]) << GP_VERTEX_ID_SHIFT,
+                               (v + tri[1]) << GP_VERTEX_ID_SHIFT,
+                               (v + tri[2]) << GP_VERTEX_ID_SHIFT);
   }
 }
 
@@ -380,10 +392,10 @@ static void gpencil_stroke_iter_cb(bGPDlayer *UNUSED(gpl),
                                    void *thunk)
 {
   gpIterData *iter = (gpIterData *)thunk;
-  gpencil_buffer_add_stroke(iter->verts, iter->cols, gps);
   if (gps->tot_triangles > 0) {
     gpencil_buffer_add_fill(&iter->ibo, gps);
   }
+  gpencil_buffer_add_stroke(&iter->ibo, iter->verts, iter->cols, gps);
 }
 
 static void gpencil_object_verts_count_cb(bGPDlayer *UNUSED(gpl),
@@ -392,12 +404,15 @@ static void gpencil_object_verts_count_cb(bGPDlayer *UNUSED(gpl),
                                           void *thunk)
 {
   gpIterData *iter = (gpIterData *)thunk;
-
-  /* Store first index offset */
-  gps->runtime.stroke_start = iter->vert_len;
+  int stroke_vert_len = gps->totpoints + gpencil_stroke_is_cyclic(gps);
+  gps->runtime.vertex_start = iter->vert_len;
+  /* Add additional padding at the start and end. */
+  iter->vert_len += 1 + stroke_vert_len + 1;
+  /* Store first index offset. */
   gps->runtime.fill_start = iter->tri_len;
-  iter->vert_len += gps->totpoints + 2 + gpencil_stroke_is_cyclic(gps);
   iter->tri_len += gps->tot_triangles;
+  gps->runtime.stroke_start = iter->tri_len;
+  iter->tri_len += stroke_vert_len * 2;
 }
 
 static void gpencil_batches_ensure(Object *ob, GpencilBatchCache *cache, int cfra)
@@ -407,7 +422,7 @@ static void gpencil_batches_ensure(Object *ob, GpencilBatchCache *cache, int cfr
   if (cache->vbo == NULL) {
     /* Should be discarded together. */
     BLI_assert(cache->vbo == NULL && cache->ibo == NULL);
-    BLI_assert(cache->fill_batch == NULL && cache->stroke_batch == NULL);
+    BLI_assert(cache->geom_batch == NULL);
     /* TODO/PERF: Could be changed to only do it if needed.
      * For now it's simpler to assume we always need it
      * since multiple viewport could or could not need it.
@@ -420,25 +435,26 @@ static void gpencil_batches_ensure(Object *ob, GpencilBatchCache *cache, int cfr
         .gpd = gpd,
         .verts = NULL,
         .ibo = {0},
-        .vert_len = 1, /* Start at 1 for the gl_InstanceID trick to work (see vert shader). */
+        .vert_len = 0,
         .tri_len = 0,
         .curve_len = 0,
     };
     BKE_gpencil_visible_stroke_advanced_iter(
         NULL, ob, NULL, gpencil_object_verts_count_cb, &iter, do_onion, cfra);
 
+    GPUUsageType vbo_flag = GPU_USAGE_STATIC | GPU_USAGE_FLAG_BUFFER_TEXTURE_ONLY;
     /* Create VBOs. */
     GPUVertFormat *format = gpencil_stroke_format();
     GPUVertFormat *format_col = gpencil_color_format();
-    cache->vbo = GPU_vertbuf_create_with_format(format);
-    cache->vbo_col = GPU_vertbuf_create_with_format(format_col);
+    cache->vbo = GPU_vertbuf_create_with_format_ex(format, vbo_flag);
+    cache->vbo_col = GPU_vertbuf_create_with_format_ex(format_col, vbo_flag);
     /* Add extra space at the end of the buffer because of quad load. */
     GPU_vertbuf_data_alloc(cache->vbo, iter.vert_len + 2);
     GPU_vertbuf_data_alloc(cache->vbo_col, iter.vert_len + 2);
     iter.verts = (gpStrokeVert *)GPU_vertbuf_get_data(cache->vbo);
     iter.cols = (gpColorVert *)GPU_vertbuf_get_data(cache->vbo_col);
     /* Create IBO. */
-    GPU_indexbuf_init(&iter.ibo, GPU_PRIM_TRIS, iter.tri_len, iter.vert_len);
+    GPU_indexbuf_init(&iter.ibo, GPU_PRIM_TRIS, iter.tri_len, 0xFFFFFFFFu);
 
     /* Fill buffers with data. */
     BKE_gpencil_visible_stroke_advanced_iter(
@@ -453,33 +469,39 @@ static void gpencil_batches_ensure(Object *ob, GpencilBatchCache *cache, int cfr
 
     /* Finish the IBO. */
     cache->ibo = GPU_indexbuf_build(&iter.ibo);
-
     /* Create the batches */
-    cache->fill_batch = GPU_batch_create(GPU_PRIM_TRIS, cache->vbo, cache->ibo);
-    GPU_batch_vertbuf_add(cache->fill_batch, cache->vbo_col);
-    cache->stroke_batch = GPU_batch_create(GPU_PRIM_TRI_STRIP, gpencil_dummy_buffer_get(), NULL);
-    GPU_batch_instbuf_add_ex(cache->stroke_batch, cache->vbo, 0);
-    GPU_batch_instbuf_add_ex(cache->stroke_batch, cache->vbo_col, 0);
+    cache->geom_batch = GPU_batch_create(GPU_PRIM_TRIS, cache->vbo, cache->ibo);
+    /* Allow creation of buffer texture. */
+    GPU_vertbuf_use(cache->vbo);
+    GPU_vertbuf_use(cache->vbo_col);
 
     gpd->flag &= ~GP_DATA_CACHE_IS_DIRTY;
     cache->is_dirty = false;
   }
 }
 
-GPUBatch *DRW_cache_gpencil_strokes_get(Object *ob, int cfra)
+GPUBatch *DRW_cache_gpencil_get(Object *ob, int cfra)
 {
   GpencilBatchCache *cache = gpencil_batch_cache_get(ob, cfra);
   gpencil_batches_ensure(ob, cache, cfra);
 
-  return cache->stroke_batch;
+  return cache->geom_batch;
 }
 
-GPUBatch *DRW_cache_gpencil_fills_get(Object *ob, int cfra)
+GPUVertBuf *DRW_cache_gpencil_position_buffer_get(Object *ob, int cfra)
 {
   GpencilBatchCache *cache = gpencil_batch_cache_get(ob, cfra);
   gpencil_batches_ensure(ob, cache, cfra);
 
-  return cache->fill_batch;
+  return cache->vbo;
+}
+
+GPUVertBuf *DRW_cache_gpencil_color_buffer_get(Object *ob, int cfra)
+{
+  GpencilBatchCache *cache = gpencil_batch_cache_get(ob, cfra);
+  gpencil_batches_ensure(ob, cache, cfra);
+
+  return cache->vbo_col;
 }
 
 static void gpencil_lines_indices_cb(bGPDlayer *UNUSED(gpl),
@@ -490,7 +512,7 @@ static void gpencil_lines_indices_cb(bGPDlayer *UNUSED(gpl),
   gpIterData *iter = (gpIterData *)thunk;
   int pts_len = gps->totpoints + gpencil_stroke_is_cyclic(gps);
 
-  int start = gps->runtime.stroke_start + 1;
+  int start = gps->runtime.vertex_start + 1;
   int end = start + pts_len;
   for (int i = start; i < end; i++) {
     GPU_indexbuf_add_generic_vert(&iter->ibo, i);
@@ -573,7 +595,7 @@ static void gpencil_sbuffer_stroke_ensure(bGPdata *gpd, bool do_stroke, bool do_
   /* DRW_cache_gpencil_sbuffer_stroke_data_get need to have been called previously. */
   BLI_assert(gps != NULL);
 
-  if (do_stroke && (gpd->runtime.sbuffer_stroke_batch == NULL)) {
+  if (do_stroke && (gpd->runtime.sbuffer_batch == NULL)) {
     gps->points = (bGPDspoint *)MEM_mallocN(vert_len * sizeof(*gps->points), __func__);
 
     const DRWContextState *draw_ctx = DRW_context_state_get();
@@ -597,35 +619,24 @@ static void gpencil_sbuffer_stroke_ensure(bGPdata *gpd, bool do_stroke, bool do_
     /* Calc uv data along the stroke. */
     BKE_gpencil_stroke_uv_update(gps);
 
+    int vert_len = gps->tot_triangles + (gps->totpoints + gpencil_stroke_is_cyclic(gps)) * 2;
+    /* Create IBO. */
+    GPUIndexBufBuilder ibo_builder;
+    GPU_indexbuf_init(&ibo_builder, GPU_PRIM_TRIS, vert_len, 0xFFFFFFFFu);
     /* Create VBO. */
+    GPUUsageType vbo_flag = GPU_USAGE_STATIC | GPU_USAGE_FLAG_BUFFER_TEXTURE_ONLY;
     GPUVertFormat *format = gpencil_stroke_format();
     GPUVertFormat *format_color = gpencil_color_format();
-    GPUVertBuf *vbo = GPU_vertbuf_create_with_format(format);
-    GPUVertBuf *vbo_col = GPU_vertbuf_create_with_format(format_color);
-    /* Add extra space at the end (and start) of the buffer because of quad load and cyclic. */
-    GPU_vertbuf_data_alloc(vbo, 1 + vert_len + 1 + 2);
-    GPU_vertbuf_data_alloc(vbo_col, 1 + vert_len + 1 + 2);
+    GPUVertBuf *vbo = GPU_vertbuf_create_with_format_ex(format, vbo_flag);
+    GPUVertBuf *vbo_col = GPU_vertbuf_create_with_format_ex(format_color, vbo_flag);
+    /* Add extra space at the end the buffer because of quad load and cyclic. */
+    GPU_vertbuf_data_alloc(vbo, vert_len + 2);
+    GPU_vertbuf_data_alloc(vbo_col, vert_len + 2);
     gpStrokeVert *verts = (gpStrokeVert *)GPU_vertbuf_get_data(vbo);
     gpColorVert *cols = (gpColorVert *)GPU_vertbuf_get_data(vbo_col);
 
-    /* Fill buffers with data. */
-    gpencil_buffer_add_stroke(verts, cols, gps);
-
-    GPUBatch *batch = GPU_batch_create(GPU_PRIM_TRI_STRIP, gpencil_dummy_buffer_get(), NULL);
-    GPU_batch_instbuf_add_ex(batch, vbo, true);
-    GPU_batch_instbuf_add_ex(batch, vbo_col, true);
-
-    gpd->runtime.sbuffer_stroke_batch = batch;
-
-    MEM_freeN(gps->points);
-  }
-
-  if (do_fill && (gpd->runtime.sbuffer_fill_batch == NULL)) {
-    /* Create IBO. */
-    GPUIndexBufBuilder ibo_builder;
-    GPU_indexbuf_init(&ibo_builder, GPU_PRIM_TRIS, gps->tot_triangles, vert_len);
-
-    if (gps->tot_triangles > 0) {
+    /* Create fill indices. */
+    if (do_fill && gps->tot_triangles > 0) {
       float(*tpoints2d)[2] = (float(*)[2])MEM_mallocN(sizeof(*tpoints2d) * vert_len, __func__);
       /* Triangulate in 2D. */
       for (int i = 0; i < vert_len; i++) {
@@ -645,40 +656,56 @@ static void gpencil_sbuffer_stroke_ensure(bGPdata *gpd, bool do_stroke, bool do_
       MEM_freeN(tpoints2d);
     }
 
-    GPUIndexBuf *ibo = GPU_indexbuf_build(&ibo_builder);
-    GPUVertBuf *vbo = gpd->runtime.sbuffer_stroke_batch->inst[0];
-    GPUVertBuf *vbo_col = gpd->runtime.sbuffer_stroke_batch->inst[1];
+    /* Fill buffers with data. */
+    gpencil_buffer_add_stroke(&ibo_builder, verts, cols, gps);
+
+    GPUBatch *batch = GPU_batch_create_ex(GPU_PRIM_TRIS,
+                                          gpencil_dummy_buffer_get(),
+                                          GPU_indexbuf_build(&ibo_builder),
+                                          GPU_BATCH_OWNS_INDEX);
 
-    GPUBatch *batch = GPU_batch_create_ex(GPU_PRIM_TRIS, vbo, ibo, GPU_BATCH_OWNS_INDEX);
-    GPU_batch_vertbuf_add(batch, vbo_col);
+    gpd->runtime.sbuffer_position_buf = vbo;
+    gpd->runtime.sbuffer_color_buf = vbo_col;
+    gpd->runtime.sbuffer_batch = batch;
 
-    gpd->runtime.sbuffer_fill_batch = batch;
+    MEM_freeN(gps->points);
   }
 }
 
-GPUBatch *DRW_cache_gpencil_sbuffer_stroke_get(Object *ob)
+GPUBatch *DRW_cache_gpencil_sbuffer_get(Object *ob)
+{
+  bGPdata *gpd = (bGPdata *)ob->data;
+  /* Fill batch also need stroke batch to be created (vbo is shared). */
+  gpencil_sbuffer_stroke_ensure(gpd, true, true);
+
+  return gpd->runtime.sbuffer_batch;
+}
+
+GPUVertBuf *DRW_cache_gpencil_sbuffer_position_buffer_get(Object *ob)
 {
   bGPdata *gpd = (bGPdata *)ob->data;
-  gpencil_sbuffer_stroke_ensure(gpd, true, false);
+  /* Fill batch also need stroke batch to be created (vbo is shared). */
+  gpencil_sbuffer_stroke_ensure(gpd, true, true);
 
-  return gpd->runtime.sbuffer_stroke_batch;
+  return gpd->runtime.sbuffer_position_buf;
 }
 
-GPUBatch *DRW_cache_gpencil_sbuffer_fill_get(Object *ob)
+GPUVertBuf *DRW_cache_gpencil_sbuffer_color_buffer_get(Object *ob)
 {
   bGPdata *gpd = (bGPdata *)ob->data;
   /* Fill batch also need stroke batch to be created (vbo is shared). */
   gpencil_sbuffer_stroke_ensure(gpd, true, true);
 
-  return gpd->runtime.sbuffer_fill_batch;
+  return gpd->runtime.sbuffer_color_buf;
 }
 
 void DRW_cache_gpencil_sbuffer_clear(Object *ob)
 {
   bGPdata *gpd = (bGPdata *)ob->data;
   MEM_SAFE_FREE(gpd->runtime.sbuffer_gps);
-  GPU_BATCH_DISCARD_SAFE(gpd->runtime.sbuffer_fill_batch);
-  GPU_BATCH_DISCARD_SAFE(gpd->runtime.sbuffer_stroke_batch);
+  GPU_BATCH_DISCARD_SAFE(gpd->runtime.sbuffer_batch);
+  GPU_VERTBUF_DISCARD_SAFE(gpd->runtime.sbuffer_position_buf);
+  GPU_VERTBUF_DISCARD_SAFE(gpd->runtime.sbuffer_color_buf);
 }
 
 /** \} */
diff --git a/source/blender/draw/intern/draw_manager_data.cc b/source/blender/draw/intern/draw_manager_data.cc
index 981206e56fe..b9e0db71122 100644
--- a/source/blender/draw/intern/draw_manager_data.cc
+++ b/source/blender/draw/intern/draw_manager_data.cc
@@ -589,7 +589,7 @@ void DRW_shgroup_buffer_texture(DRWShadingGroup *shgroup,
                                 const char *name,
                                 GPUVertBuf *vertex_buffer)
 {
-  int location = GPU_shader_get_ssbo(shgroup->shader, name);
+  int location = GPU_shader_get_texture_binding(shgroup->shader, name);
   if (location == -1) {
     return;
   }
@@ -606,7 +606,7 @@ void DRW_shgroup_buffer_texture_ref(DRWShadingGroup *shgroup,
                                     const char *name,
                                     GPUVertBuf **vertex_buffer)
 {
-  int location = GPU_shader_get_ssbo(shgroup->shader, name);
+  int location = GPU_shader_get_texture_binding(shgroup->shader, name);
   if (location == -1) {
     return;
   }
@@ -695,7 +695,7 @@ static void drw_call_obinfos_init(DRWObjectInfos *ob_infos, Object *ob)
   drw_call_calc_orco(ob, ob_infos->orcotexfac);
   /* Random float value. */
   uint random = (DST.dupli_source) ?
-                     DST.dupli_source->random_id :
+                    DST.dupli_source->random_id :
                      /* TODO(fclem): this is rather costly to do at runtime. Maybe we can
                       * put it in ob->runtime and make depsgraph ensure it is up to date. */
                      BLI_hash_int_2d(BLI_hash_string(ob->id.name + 2), 0);
diff --git a/source/blender/draw/intern/shaders/common_gpencil_lib.glsl b/source/blender/draw/intern/shaders/common_gpencil_lib.glsl
index 123c493b572..def841b07aa 100644
--- a/source/blender/draw/intern/shaders/common_gpencil_lib.glsl
+++ b/source/blender/draw/intern/shaders/common_gpencil_lib.glsl
@@ -90,10 +90,15 @@ float gpencil_clamp_small_stroke_thickness(float thickness, vec4 ndc_pos)
 
 #ifdef GPU_VERTEX_SHADER
 
-/* Trick to detect if a drawcall is stroke or fill.
- * This does mean that we need to draw an empty stroke segment before starting
- * to draw the real stroke segments. */
-#  define GPENCIL_IS_STROKE_VERTEX (gl_InstanceID != 0)
+int gpencil_stroke_point_id()
+{
+  return (gl_VertexID & ~GP_IS_STROKE_VERTEX_BIT) >> GP_VERTEX_ID_SHIFT;
+}
+
+bool gpencil_is_stroke_vertex()
+{
+  return flag_test(gl_VertexID, GP_IS_STROKE_VERTEX_BIT);
+}
 
 /**
  * Returns value of gl_Position.
@@ -120,20 +125,7 @@ float gpencil_clamp_small_stroke_thickness(float thickness, vec4 ndc_pos)
  * WARNING: Max attribute count is actually 14 because OSX OpenGL implementation
  * considers gl_VertexID and gl_InstanceID as vertex attribute. (see T74536)
  */
-vec4 gpencil_vertex(ivec4 ma,
-                    ivec4 ma1,
-                    ivec4 ma2,
-                    ivec4 ma3,
-                    vec4 pos,
-                    vec4 pos1,
-                    vec4 pos2,
-                    vec4 pos3,
-                    vec4 uv1,
-                    vec4 uv2,
-                    vec4 col1,
-                    vec4 col2,
-                    vec4 fcol1,
-                    vec4 viewport_size,
+vec4 gpencil_vertex(vec4 viewport_size,
                     gpMaterialFlag material_flags,
                     vec2 alignment_rot,
                     /* World Position. */
@@ -155,6 +147,24 @@ vec4 gpencil_vertex(ivec4 ma,
                     /* Stroke hardness. */
                     out float out_hardness)
 {
+  int stroke_point_id = (gl_VertexID & ~GP_IS_STROKE_VERTEX_BIT) >> GP_VERTEX_ID_SHIFT;
+
+  /* Attribute Loading. */
+  vec4 pos = texelFetch(gp_pos_tx, (stroke_point_id - 1) * 3 + 0);
+  vec4 pos1 = texelFetch(gp_pos_tx, (stroke_point_id + 0) * 3 + 0);
+  vec4 pos2 = texelFetch(gp_pos_tx, (stroke_point_id + 1) * 3 + 0);
+  vec4 pos3 = texelFetch(gp_pos_tx, (stroke_point_id + 2) * 3 + 0);
+  ivec4 ma = floatBitsToInt(texelFetch(gp_pos_tx, (stroke_point_id - 1) * 3 + 1));
+  ivec4 ma1 = floatBitsToInt(texelFetch(gp_pos_tx, (stroke_point_id + 0) * 3 + 1));
+  ivec4 ma2 = floatBitsToInt(texelFetch(gp_pos_tx, (stroke_point_id + 1) * 3 + 1));
+  ivec4 ma3 = floatBitsToInt(texelFetch(gp_pos_tx, (stroke_point_id + 2) * 3 + 1));
+  vec4 uv1 = texelFetch(gp_pos_tx, (stroke_point_id + 0) * 3 + 2);
+  vec4 uv2 = texelFetch(gp_pos_tx, (stroke_point_id + 1) * 3 + 2);
+
+  vec4 col1 = texelFetch(gp_col_tx, (stroke_point_id + 0) * 2 + 0);
+  vec4 col2 = texelFetch(gp_col_tx, (stroke_point_id + 1) * 2 + 0);
+  vec4 fcol1 = texelFetch(gp_col_tx, (stroke_point_id + 0) * 2 + 1);
+
 #  define thickness1 pos1.w
 #  define thickness2 pos2.w
 #  define strength1 uv1.w
@@ -167,7 +177,7 @@ vec4 gpencil_vertex(ivec4 ma,
 
   vec4 out_ndc;
 
-  if (GPENCIL_IS_STROKE_VERTEX) {
+  if (gpencil_is_stroke_vertex()) {
     bool is_dot = flag_test(material_flags, GP_STROKE_ALIGNMENT);
     bool is_squares = !flag_test(material_flags, GP_STROKE_DOTS);
 
@@ -177,13 +187,6 @@ vec4 gpencil_vertex(ivec4 ma,
       is_squares = false;
     }
 
-    /* Endpoints, we discard the vertices. */
-    if (ma1.x == -1 || (!is_dot && ma2.x == -1)) {
-      /* We set the vertex at the camera origin to generate 0 fragments. */
-      out_ndc = vec4(0.0, 0.0, -3e36, 0.0);
-      return out_ndc;
-    }
-
     /* Avoid using a vertex attribute for quad positioning. */
     float x = float(gl_VertexID & 1) * 2.0 - 1.0; /* [-1..1] */
     float y = float(gl_VertexID & 2) - 1.0;       /* [-1..1] */
@@ -336,8 +339,7 @@ vec4 gpencil_vertex(ivec4 ma,
     out_N = safe_normalize(N);
 
     /* Decode fill opacity. */
-    out_color = vec4(fcol1.rgb, floor(fcol1.a / 10.0));
-    out_color.a /= 10000.0;
+    out_color = vec4(fcol1.rgb, floor(fcol1.a / 10.0) / 10000.0);
 
     /* We still offset the fills a little to avoid overlaps */
     out_ndc.z += 0.000002;
@@ -355,20 +357,7 @@ vec4 gpencil_vertex(ivec4 ma,
   return out_ndc;
 }
 
-vec4 gpencil_vertex(ivec4 ma,
-                    ivec4 ma1,
-                    ivec4 ma2,
-                    ivec4 ma3,
-                    vec4 pos,
-                    vec4 pos1,
-                    vec4 pos2,
-                    vec4 pos3,
-                    vec4 uv1,
-                    vec4 uv2,
-                    vec4 col1,
-                    vec4 col2,
-                    vec4 fcol1,
-                    vec4 viewport_size,
+vec4 gpencil_vertex(vec4 viewport_size,
                     out vec3 out_P,
                     out vec3 out_N,
                     out vec4 out_color,
@@ -379,20 +368,7 @@ vec4 gpencil_vertex(ivec4 ma,
                     out vec2 out_thickness,
                     out float out_hardness)
 {
-  return gpencil_vertex(ma,
-                        ma1,
-                        ma2,
-                        ma3,
-                        pos,
-                        pos1,
-                        pos2,
-                        pos3,
-                        uv1,
-                        uv2,
-                        col1,
-                        col2,
-                        fcol1,
-                        viewport_size,
+  return gpencil_vertex(viewport_size,
                         0u,
                         vec2(1.0, 0.0),
                         out_P,
diff --git a/source/blender/draw/intern/shaders/draw_view_info.hh b/source/blender/draw/intern/shaders/draw_view_info.hh
index 7b500f66a68..114dbab799f 100644
--- a/source/blender/draw/intern/shaders/draw_view_info.hh
+++ b/source/blender/draw/intern/shaders/draw_view_info.hh
@@ -122,26 +122,15 @@ GPU_SHADER_CREATE_INFO(draw_volume).additional_info("draw_modelmat", "draw_resou
 GPU_SHADER_CREATE_INFO(draw_gpencil)
     .typedef_source("gpencil_shader_shared.h")
     .define("DRW_GPENCIL_INFO")
-    .vertex_in(0, Type::IVEC4, "ma")
-    .vertex_in(1, Type::IVEC4, "ma1")
-    .vertex_in(2, Type::IVEC4, "ma2")
-    .vertex_in(3, Type::IVEC4, "ma3")
-    .vertex_in(4, Type::VEC4, "pos")
-    .vertex_in(5, Type::VEC4, "pos1")
-    .vertex_in(6, Type::VEC4, "pos2")
-    .vertex_in(7, Type::VEC4, "pos3")
-    .vertex_in(8, Type::VEC4, "uv1")
-    .vertex_in(9, Type::VEC4, "uv2")
-    .vertex_in(10, Type::VEC4, "col1")
-    .vertex_in(11, Type::VEC4, "col2")
-    .vertex_in(12, Type::VEC4, "fcol1")
+    .sampler(0, ImageType::FLOAT_BUFFER, "gp_pos_tx")
+    .sampler(1, ImageType::FLOAT_BUFFER, "gp_col_tx")
     /* Per Object */
     .push_constant(Type::FLOAT, "gpThicknessScale") /* TODO(fclem): Replace with object info. */
     .push_constant(Type::FLOAT, "gpThicknessWorldScale") /* TODO(fclem): Same as above. */
     .define("gpThicknessIsScreenSpace", "(gpThicknessWorldScale < 0.0)")
     /* Per Layer */
     .push_constant(Type::FLOAT, "gpThicknessOffset")
-    .additional_info("draw_modelmat", "draw_resource_id_uniform", "draw_object_infos");
+    .additional_info("draw_modelmat", "draw_object_infos");
 
 /** \} */
 
diff --git a/source/blender/gpu/intern/gpu_texture_private.hh b/source/blender/gpu/intern/gpu_texture_private.hh
index 2ad31183206..124b1751b96 100644
--- a/source/blender/gpu/intern/gpu_texture_private.hh
+++ b/source/blender/gpu/intern/gpu_texture_private.hh
@@ -588,7 +588,7 @@ inline eGPUFrameBufferBits to_framebuffer_bits(eGPUTextureFormat tex_format)
 
 static inline eGPUTextureFormat to_texture_format(const GPUVertFormat *format)
 {
-  if (format->attr_len > 1 || format->attr_len == 0) {
+  if (format->attr_len == 0) {
     BLI_assert_msg(0, "Incorrect vertex format for buffer texture");
     return GPU_DEPTH_COMPONENT24;
   }
diff --git a/source/blender/gpu/intern/gpu_vertex_format.cc b/source/blender/gpu/intern/gpu_vertex_format.cc
index b30e3c358c8..76d95ac1b55 100644
--- a/source/blender/gpu/intern/gpu_vertex_format.cc
+++ b/source/blender/gpu/intern/gpu_vertex_format.cc
@@ -361,8 +361,12 @@ void VertexFormat_texture_buffer_pack(GPUVertFormat *format)
    * minimum per-vertex stride, which mandates 4-byte alignment in Metal.
    * This additional alignment padding caused smaller data types, e.g. U16,
    * to mis-align. */
-  BLI_assert_msg(format->attr_len == 1,
-                 "Texture buffer mode should only use a single vertex attribute.");
+  for (int i = 0; i < format->attr_len; i++) {
+    /* The buffer texture setup uses the first attribute for type and size.
+     * Make sure all attributes use the same size. */
+    BLI_assert_msg(format->attrs[i].size == format->attrs[0].size,
+                   "Texture buffer mode should only use a attributes with the same size.");
+  }
 
   /* Pack vertex format without minimum stride, as this is not required by texture buffers. */
   VertexFormat_pack_impl(format, 1);
diff --git a/source/blender/makesdna/DNA_gpencil_types.h b/source/blender/makesdna/DNA_gpencil_types.h
index 26dbb544f52..8b3f4956cfe 100644
--- a/source/blender/makesdna/DNA_gpencil_types.h
+++ b/source/blender/makesdna/DNA_gpencil_types.h
@@ -243,12 +243,15 @@ typedef struct bGPDstroke_Runtime {
   /** Runtime falloff factor (only for transform). */
   float multi_frame_falloff;
 
-  /** Vertex offset in the VBO where this stroke starts. */
+  /** Triangle offset in the IBO where this stroke starts. */
   int stroke_start;
   /** Triangle offset in the IBO where this fill starts. */
   int fill_start;
+  /** Vertex offset in the VBO where this stroke starts. */
+  int vertex_start;
   /** Curve Handles offset in the IBO where this handle starts. */
   int curve_start;
+  int _pad0;
 
   /** Original stroke (used to dereference evaluated data) */
   struct bGPDstroke *gps_orig;
@@ -613,8 +616,9 @@ typedef struct bGPdata_Runtime {
   /** Stroke buffer. */
   void *sbuffer;
   /** Temp batches cleared after drawing. */
-  struct GPUBatch *sbuffer_stroke_batch;
-  struct GPUBatch *sbuffer_fill_batch;
+  struct GPUVertBuf *sbuffer_position_buf;
+  struct GPUVertBuf *sbuffer_color_buf;
+  struct GPUBatch *sbuffer_batch;
   /** Temp stroke used for drawing. */
   struct bGPDstroke *sbuffer_gps;