Bug 693330: Change shadings to decompose to meshes at render time.

Currently, the mupdf code loads shadings at parse time, and
instantly decomposes them into a mesh of triangles. This mesh
of triangles is the transformed and rendered as required.

Unfortunately the storage space for the mesh is typically much
greater than the original representation.

In this commit, we move the shading stream parsing/decomposition
code into a general 'fz_process_mesh' function within res_shade.
We then grab a copy of the buffer at load time, and 'process'
(decompose/paint) at render time.

For the test file on the bug, memory falls from the reported 660Mb
to 30Mb. For another test file (txt9780547775815_ingested.pdf
page 271) it reduces memory use from 750Meg to 33Meg. These figures
could be further reduced by storing the compressed streams from the
pdf file rather than the uncompressed ones.

Incorporating typo fix and unused function removal from Sebras. Thanks.

Remove unused function in shading code

1 files changed, 76 insertions, 746 deletions

diff --git a/pdf/pdf_shade.c b/pdf/pdf_shade.c
index 39e71626..6fbfd6b7 100644
--- a/pdf/pdf_shade.c
+++ b/pdf/pdf_shade.c
@@ -1,326 +1,8 @@
 #include "fitz-internal.h"
 #include "mupdf-internal.h"
 
-#define HUGENUM 32000 /* how far to extend axial/radial shadings */
+/* FIXME: Remove this somehow */
 #define FUNSEGS 32 /* size of sampled mesh for function-based shadings */
-#define RADSEGS 32 /* how many segments to generate for radial meshes */
-#define SUBDIV 3 /* how many levels to subdivide patches */
-
-struct vertex
-{
-	float x, y;
-	float c[FZ_MAX_COLORS];
-};
-
-static void
-pdf_grow_mesh(fz_context *ctx, fz_shade *shade, int amount)
-{
-	int cap = shade->mesh_cap;
-
-	if (shade->mesh_len + amount < cap)
-		return;
-
-	if (cap == 0)
-		cap = 1024;
-
-	while (shade->mesh_len + amount > cap)
-		cap = (cap * 3) / 2;
-
-	shade->mesh = fz_resize_array(ctx, shade->mesh, cap, sizeof(float));
-	shade->mesh_cap = cap;
-}
-
-static void
-pdf_add_vertex(fz_context *ctx, fz_shade *shade, struct vertex *v)
-{
-	int ncomp = shade->use_function ? 1 : shade->colorspace->n;
-	int i;
-	pdf_grow_mesh(ctx, shade, 2 + ncomp);
-	shade->mesh[shade->mesh_len++] = v->x;
-	shade->mesh[shade->mesh_len++] = v->y;
-	for (i = 0; i < ncomp; i++)
-		shade->mesh[shade->mesh_len++] = v->c[i];
-}
-
-static void
-pdf_add_triangle(fz_context *ctx, fz_shade *shade,
-	struct vertex *v0,
-	struct vertex *v1,
-	struct vertex *v2)
-{
-	pdf_add_vertex(ctx, shade, v0);
-	pdf_add_vertex(ctx, shade, v1);
-	pdf_add_vertex(ctx, shade, v2);
-}
-
-static void
-pdf_add_quad(fz_context *ctx, fz_shade *shade,
-	struct vertex *v0,
-	struct vertex *v1,
-	struct vertex *v2,
-	struct vertex *v3)
-{
-	pdf_add_triangle(ctx, shade, v0, v1, v3);
-	pdf_add_triangle(ctx, shade, v1, v3, v2);
-}
-
-/* Subdivide and tessellate tensor-patches */
-
-typedef struct pdf_tensor_patch_s pdf_tensor_patch;
-
-struct pdf_tensor_patch_s
-{
-	fz_point pole[4][4];
-	float color[4][FZ_MAX_COLORS];
-};
-
-static void
-triangulate_patch(fz_context *ctx, pdf_tensor_patch p, fz_shade *shade)
-{
-	struct vertex v0, v1, v2, v3;
-
-	v0.x = p.pole[0][0].x;
-	v0.y = p.pole[0][0].y;
-	memcpy(v0.c, p.color[0], sizeof(v0.c));
-
-	v1.x = p.pole[0][3].x;
-	v1.y = p.pole[0][3].y;
-	memcpy(v1.c, p.color[1], sizeof(v1.c));
-
-	v2.x = p.pole[3][3].x;
-	v2.y = p.pole[3][3].y;
-	memcpy(v2.c, p.color[2], sizeof(v2.c));
-
-	v3.x = p.pole[3][0].x;
-	v3.y = p.pole[3][0].y;
-	memcpy(v3.c, p.color[3], sizeof(v3.c));
-
-	pdf_add_quad(ctx, shade, &v0, &v1, &v2, &v3);
-}
-
-static inline void midcolor(float *c, float *c1, float *c2)
-{
-	int i;
-	for (i = 0; i < FZ_MAX_COLORS; i++)
-		c[i] = (c1[i] + c2[i]) * 0.5f;
-}
-
-static void
-split_curve(fz_point *pole, fz_point *q0, fz_point *q1, int polestep)
-{
-	/*
-	split bezier curve given by control points pole[0]..pole[3]
-	using de casteljau algo at midpoint and build two new
-	bezier curves q0[0]..q0[3] and q1[0]..q1[3]. all indices
-	should be multiplies by polestep == 1 for vertical bezier
-	curves in patch and == 4 for horizontal bezier curves due
-	to C's multi-dimensional matrix memory layout.
-	*/
-
-	float x12 = (pole[1 * polestep].x + pole[2 * polestep].x) * 0.5f;
-	float y12 = (pole[1 * polestep].y + pole[2 * polestep].y) * 0.5f;
-
-	q0[1 * polestep].x = (pole[0 * polestep].x + pole[1 * polestep].x) * 0.5f;
-	q0[1 * polestep].y = (pole[0 * polestep].y + pole[1 * polestep].y) * 0.5f;
-	q1[2 * polestep].x = (pole[2 * polestep].x + pole[3 * polestep].x) * 0.5f;
-	q1[2 * polestep].y = (pole[2 * polestep].y + pole[3 * polestep].y) * 0.5f;
-
-	q0[2 * polestep].x = (q0[1 * polestep].x + x12) * 0.5f;
-	q0[2 * polestep].y = (q0[1 * polestep].y + y12) * 0.5f;
-	q1[1 * polestep].x = (x12 + q1[2 * polestep].x) * 0.5f;
-	q1[1 * polestep].y = (y12 + q1[2 * polestep].y) * 0.5f;
-
-	q0[3 * polestep].x = (q0[2 * polestep].x + q1[1 * polestep].x) * 0.5f;
-	q0[3 * polestep].y = (q0[2 * polestep].y + q1[1 * polestep].y) * 0.5f;
-	q1[0 * polestep].x = (q0[2 * polestep].x + q1[1 * polestep].x) * 0.5f;
-	q1[0 * polestep].y = (q0[2 * polestep].y + q1[1 * polestep].y) * 0.5f;
-
-	q0[0 * polestep].x = pole[0 * polestep].x;
-	q0[0 * polestep].y = pole[0 * polestep].y;
-	q1[3 * polestep].x = pole[3 * polestep].x;
-	q1[3 * polestep].y = pole[3 * polestep].y;
-}
-
-static void
-split_stripe(pdf_tensor_patch *p, pdf_tensor_patch *s0, pdf_tensor_patch *s1)
-{
-	/*
-	split all horizontal bezier curves in patch,
-	creating two new patches with half the width.
-	*/
-	split_curve(&p->pole[0][0], &s0->pole[0][0], &s1->pole[0][0], 4);
-	split_curve(&p->pole[0][1], &s0->pole[0][1], &s1->pole[0][1], 4);
-	split_curve(&p->pole[0][2], &s0->pole[0][2], &s1->pole[0][2], 4);
-	split_curve(&p->pole[0][3], &s0->pole[0][3], &s1->pole[0][3], 4);
-
-	/* interpolate the colors for the two new patches. */
-	memcpy(s0->color[0], p->color[0], sizeof(s0->color[0]));
-	memcpy(s0->color[1], p->color[1], sizeof(s0->color[1]));
-	midcolor(s0->color[2], p->color[1], p->color[2]);
-	midcolor(s0->color[3], p->color[0], p->color[3]);
-
-	memcpy(s1->color[0], s0->color[3], sizeof(s1->color[0]));
-	memcpy(s1->color[1], s0->color[2], sizeof(s1->color[1]));
-	memcpy(s1->color[2], p->color[2], sizeof(s1->color[2]));
-	memcpy(s1->color[3], p->color[3], sizeof(s1->color[3]));
-}
-
-static void
-draw_stripe(fz_context *ctx, pdf_tensor_patch *p, fz_shade *shade, int depth)
-{
-	pdf_tensor_patch s0, s1;
-
-	/* split patch into two half-height patches */
-	split_stripe(p, &s0, &s1);
-
-	depth--;
-	if (depth == 0)
-	{
-		/* if no more subdividing, draw two new patches... */
-		triangulate_patch(ctx, s1, shade);
-		triangulate_patch(ctx, s0, shade);
-	}
-	else
-	{
-		/* ...otherwise, continue subdividing. */
-		draw_stripe(ctx, &s1, shade, depth);
-		draw_stripe(ctx, &s0, shade, depth);
-	}
-}
-
-static void
-split_patch(pdf_tensor_patch *p, pdf_tensor_patch *s0, pdf_tensor_patch *s1)
-{
-	/*
-	split all vertical bezier curves in patch,
-	creating two new patches with half the height.
-	*/
-	split_curve(p->pole[0], s0->pole[0], s1->pole[0], 1);
-	split_curve(p->pole[1], s0->pole[1], s1->pole[1], 1);
-	split_curve(p->pole[2], s0->pole[2], s1->pole[2], 1);
-	split_curve(p->pole[3], s0->pole[3], s1->pole[3], 1);
-
-	/* interpolate the colors for the two new patches. */
-	memcpy(s0->color[0], p->color[0], sizeof(s0->color[0]));
-	midcolor(s0->color[1], p->color[0], p->color[1]);
-	midcolor(s0->color[2], p->color[2], p->color[3]);
-	memcpy(s0->color[3], p->color[3], sizeof(s0->color[3]));
-
-	memcpy(s1->color[0], s0->color[1], sizeof(s1->color[0]));
-	memcpy(s1->color[1], p->color[1], sizeof(s1->color[1]));
-	memcpy(s1->color[2], p->color[2], sizeof(s1->color[2]));
-	memcpy(s1->color[3], s0->color[2], sizeof(s1->color[3]));
-}
-
-static void
-draw_patch(fz_context *ctx, fz_shade *shade, pdf_tensor_patch *p, int depth, int origdepth)
-{
-	pdf_tensor_patch s0, s1;
-
-	/* split patch into two half-width patches */
-	split_patch(p, &s0, &s1);
-
-	depth--;
-	if (depth == 0)
-	{
-		/* if no more subdividing, draw two new patches... */
-		draw_stripe(ctx, &s0, shade, origdepth);
-		draw_stripe(ctx, &s1, shade, origdepth);
-	}
-	else
-	{
-		/* ...otherwise, continue subdividing. */
-		draw_patch(ctx, shade, &s0, depth, origdepth);
-		draw_patch(ctx, shade, &s1, depth, origdepth);
-	}
-}
-
-static fz_point
-pdf_compute_tensor_interior(
-	fz_point a, fz_point b, fz_point c, fz_point d,
-	fz_point e, fz_point f, fz_point g, fz_point h)
-{
-	fz_point pt;
-
-	/* see equations at page 330 in pdf 1.7 */
-
-	pt.x = -4 * a.x;
-	pt.x += 6 * (b.x + c.x);
-	pt.x += -2 * (d.x + e.x);
-	pt.x += 3 * (f.x + g.x);
-	pt.x += -1 * h.x;
-	pt.x /= 9;
-
-	pt.y = -4 * a.y;
-	pt.y += 6 * (b.y + c.y);
-	pt.y += -2 * (d.y + e.y);
-	pt.y += 3 * (f.y + g.y);
-	pt.y += -1 * h.y;
-	pt.y /= 9;
-
-	return pt;
-}
-
-static void
-pdf_make_tensor_patch(pdf_tensor_patch *p, int type, fz_point *pt)
-{
-	if (type == 6)
-	{
-		/* see control point stream order at page 325 in pdf 1.7 */
-
-		p->pole[0][0] = pt[0];
-		p->pole[0][1] = pt[1];
-		p->pole[0][2] = pt[2];
-		p->pole[0][3] = pt[3];
-		p->pole[1][3] = pt[4];
-		p->pole[2][3] = pt[5];
-		p->pole[3][3] = pt[6];
-		p->pole[3][2] = pt[7];
-		p->pole[3][1] = pt[8];
-		p->pole[3][0] = pt[9];
-		p->pole[2][0] = pt[10];
-		p->pole[1][0] = pt[11];
-
-		/* see equations at page 330 in pdf 1.7 */
-
-		p->pole[1][1] = pdf_compute_tensor_interior(
-			p->pole[0][0], p->pole[0][1], p->pole[1][0], p->pole[0][3],
-			p->pole[3][0], p->pole[3][1], p->pole[1][3], p->pole[3][3]);
-
-		p->pole[1][2] = pdf_compute_tensor_interior(
-			p->pole[0][3], p->pole[0][2], p->pole[1][3], p->pole[0][0],
-			p->pole[3][3], p->pole[3][2], p->pole[1][0], p->pole[3][0]);
-
-		p->pole[2][1] = pdf_compute_tensor_interior(
-			p->pole[3][0], p->pole[3][1], p->pole[2][0], p->pole[3][3],
-			p->pole[0][0], p->pole[0][1], p->pole[2][3], p->pole[0][3]);
-
-		p->pole[2][2] = pdf_compute_tensor_interior(
-			p->pole[3][3], p->pole[3][2], p->pole[2][3], p->pole[3][0],
-			p->pole[0][3], p->pole[0][2], p->pole[2][0], p->pole[0][0]);
-	}
-	else if (type == 7)
-	{
-		/* see control point stream order at page 330 in pdf 1.7 */
-
-		p->pole[0][0] = pt[0];
-		p->pole[0][1] = pt[1];
-		p->pole[0][2] = pt[2];
-		p->pole[0][3] = pt[3];
-		p->pole[1][3] = pt[4];
-		p->pole[2][3] = pt[5];
-		p->pole[3][3] = pt[6];
-		p->pole[3][2] = pt[7];
-		p->pole[3][1] = pt[8];
-		p->pole[3][0] = pt[9];
-		p->pole[2][0] = pt[10];
-		p->pole[1][0] = pt[11];
-		p->pole[1][1] = pt[12];
-		p->pole[1][2] = pt[13];
-		p->pole[2][2] = pt[14];
-		p->pole[2][1] = pt[15];
-	}
-}
 
 /* Sample various functions into lookup tables */
 
@@ -370,13 +52,11 @@ pdf_load_function_based_shading(fz_shade *shade, pdf_document *xref, pdf_obj *di
 {
 	pdf_obj *obj;
 	float x0, y0, x1, y1;
+	float fv[2];
 	fz_matrix matrix;
-	struct vertex v[4];
 	int xx, yy;
-	float x, y;
-	float xn, yn;
-	int i;
 	fz_context *ctx = xref->ctx;
+	float *p;
 
 	x0 = y0 = 0;
 	x1 = y1 = 1;
@@ -393,39 +73,26 @@ pdf_load_function_based_shading(fz_shade *shade, pdf_document *xref, pdf_obj *di
 	obj = pdf_dict_gets(dict, "Matrix");
 	if (pdf_array_len(obj) == 6)
 		matrix = pdf_to_matrix(ctx, obj);
-
-	for (yy = 0; yy < FUNSEGS; yy++)
+	shade->u.f.matrix = matrix;
+	shade->u.f.xdivs = FUNSEGS;
+	shade->u.f.ydivs = FUNSEGS;
+	shade->u.f.fn_vals = fz_malloc(ctx, (FUNSEGS+1)*(FUNSEGS+1)*shade->colorspace->n*sizeof(float));
+	shade->u.f.domain[0][0] = x0;
+	shade->u.f.domain[0][1] = y0;
+	shade->u.f.domain[1][0] = x1;
+	shade->u.f.domain[1][1] = y1;
+
+	p = shade->u.f.fn_vals;
+	for (yy = 0; yy <= FUNSEGS; yy++)
 	{
-		y = y0 + (y1 - y0) * yy / FUNSEGS;
-		yn = y0 + (y1 - y0) * (yy + 1) / FUNSEGS;
+		fv[1] = y0 + (y1 - y0) * yy / FUNSEGS;
 
-		for (xx = 0; xx < FUNSEGS; xx++)
+		for (xx = 0; xx <= FUNSEGS; xx++)
 		{
-			x = x0 + (x1 - x0) * xx / FUNSEGS;
-			xn = x0 + (x1 - x0) * (xx + 1) / FUNSEGS;
-
-			v[0].x = x; v[0].y = y;
-			v[1].x = xn; v[1].y = y;
-			v[2].x = xn; v[2].y = yn;
-			v[3].x = x; v[3].y = yn;
+			fv[0] = x0 + (x1 - x0) * xx / FUNSEGS;
 
-			for (i = 0; i < 4; i++)
-			{
-				fz_point pt;
-				float fv[2];
-
-				fv[0] = v[i].x;
-				fv[1] = v[i].y;
-				pdf_eval_function(ctx, func, fv, 2, v[i].c, shade->colorspace->n);
-
-				pt.x = v[i].x;
-				pt.y = v[i].y;
-				pt = fz_transform_point(matrix, pt);
-				v[i].x = pt.x;
-				v[i].y = pt.y;
-			}
-
-			pdf_add_quad(ctx, shade, &v[0], &v[1], &v[2], &v[3]);
+			pdf_eval_function(ctx, func, fv, 2, p, shade->colorspace->n);
+			p += shade->colorspace->n;
 		}
 	}
 }
@@ -436,15 +103,13 @@ pdf_load_axial_shading(fz_shade *shade, pdf_document *xref, pdf_obj *dict, int f
 	pdf_obj *obj;
 	float d0, d1;
 	int e0, e1;
-	float x0, y0, x1, y1;
-	struct vertex p1, p2;
 	fz_context *ctx = xref->ctx;
 
 	obj = pdf_dict_gets(dict, "Coords");
-	x0 = pdf_to_real(pdf_array_get(obj, 0));
-	y0 = pdf_to_real(pdf_array_get(obj, 1));
-	x1 = pdf_to_real(pdf_array_get(obj, 2));
-	y1 = pdf_to_real(pdf_array_get(obj, 3));
+	shade->u.a_or_r.coords[0][0] = pdf_to_real(pdf_array_get(obj, 0));
+	shade->u.a_or_r.coords[0][1] = pdf_to_real(pdf_array_get(obj, 1));
+	shade->u.a_or_r.coords[1][0] = pdf_to_real(pdf_array_get(obj, 2));
+	shade->u.a_or_r.coords[1][1] = pdf_to_real(pdf_array_get(obj, 3));
 
 	d0 = 0;
 	d1 = 1;
@@ -465,20 +130,8 @@ pdf_load_axial_shading(fz_shade *shade, pdf_document *xref, pdf_obj *dict, int f
 
 	pdf_sample_shade_function(ctx, shade, funcs, func, d0, d1);
 
-	shade->type = FZ_LINEAR;
-
-	shade->extend[0] = e0;
-	shade->extend[1] = e1;
-
-	p1.x = x0;
-	p1.y = y0;
-	p1.c[0] = 0;
-	pdf_add_vertex(ctx, shade, &p1);
-
-	p2.x = x1;
-	p2.y = y1;
-	p2.c[0] = 0;
-	pdf_add_vertex(ctx, shade, &p2);
+	shade->u.a_or_r.extend[0] = e0;
+	shade->u.a_or_r.extend[1] = e1;
 }
 
 static void
@@ -487,17 +140,15 @@ pdf_load_radial_shading(fz_shade *shade, pdf_document *xref, pdf_obj *dict, int
 	pdf_obj *obj;
 	float d0, d1;
 	int e0, e1;
-	float x0, y0, r0, x1, y1, r1;
-	struct vertex p1, p2;
 	fz_context *ctx = xref->ctx;
 
 	obj = pdf_dict_gets(dict, "Coords");
-	x0 = pdf_to_real(pdf_array_get(obj, 0));
-	y0 = pdf_to_real(pdf_array_get(obj, 1));
-	r0 = pdf_to_real(pdf_array_get(obj, 2));
-	x1 = pdf_to_real(pdf_array_get(obj, 3));
-	y1 = pdf_to_real(pdf_array_get(obj, 4));
-	r1 = pdf_to_real(pdf_array_get(obj, 5));
+	shade->u.a_or_r.coords[0][0] = pdf_to_real(pdf_array_get(obj, 0));
+	shade->u.a_or_r.coords[0][1] = pdf_to_real(pdf_array_get(obj, 1));
+	shade->u.a_or_r.coords[0][2] = pdf_to_real(pdf_array_get(obj, 2));
+	shade->u.a_or_r.coords[1][0] = pdf_to_real(pdf_array_get(obj, 3));
+	shade->u.a_or_r.coords[1][1] = pdf_to_real(pdf_array_get(obj, 4));
+	shade->u.a_or_r.coords[1][2] = pdf_to_real(pdf_array_get(obj, 5));
 
 	d0 = 0;
 	d1 = 1;
@@ -518,31 +169,12 @@ pdf_load_radial_shading(fz_shade *shade, pdf_document *xref, pdf_obj *dict, int
 
 	pdf_sample_shade_function(ctx, shade, funcs, func, d0, d1);
 
-	shade->type = FZ_RADIAL;
-
-	shade->extend[0] = e0;
-	shade->extend[1] = e1;
-
-	p1.x = x0;
-	p1.y = y0;
-	p1.c[0] = r0;
-	pdf_add_vertex(ctx, shade, &p1);
-
-	p2.x = x1;
-	p2.y = y1;
-	p2.c[0] = r1;
-	pdf_add_vertex(ctx, shade, &p2);
+	shade->u.a_or_r.extend[0] = e0;
+	shade->u.a_or_r.extend[1] = e1;
 }
 
 /* Type 4-7 -- Triangle and patch mesh shadings */
 
-static inline float read_sample(fz_stream *stream, int bits, float min, float max)
-{
-	/* we use pow(2,x) because (1<<x) would overflow the math on 32-bit samples */
-	float bitscale = 1 / (powf(2, bits) - 1);
-	return min + fz_read_bits(stream, bits) * (max - min) * bitscale;
-}
-
 struct mesh_params
 {
 	int vprow;
@@ -556,53 +188,53 @@ struct mesh_params
 };
 
 static void
-pdf_load_mesh_params(pdf_document *xref, pdf_obj *dict, struct mesh_params *p)
+pdf_load_mesh_params(fz_shade *shade, pdf_document *xref, pdf_obj *dict)
 {
 	pdf_obj *obj;
 	int i, n;
 
-	p->x0 = p->y0 = 0;
-	p->x1 = p->y1 = 1;
+	shade->u.m.x0 = shade->u.m.y0 = 0;
+	shade->u.m.x1 = shade->u.m.y1 = 1;
 	for (i = 0; i < FZ_MAX_COLORS; i++)
 	{
-		p->c0[i] = 0;
-		p->c1[i] = 1;
+		shade->u.m.c0[i] = 0;
+		shade->u.m.c1[i] = 1;
 	}
 
-	p->vprow = pdf_to_int(pdf_dict_gets(dict, "VerticesPerRow"));
-	p->bpflag = pdf_to_int(pdf_dict_gets(dict, "BitsPerFlag"));
-	p->bpcoord = pdf_to_int(pdf_dict_gets(dict, "BitsPerCoordinate"));
-	p->bpcomp = pdf_to_int(pdf_dict_gets(dict, "BitsPerComponent"));
+	shade->u.m.vprow = pdf_to_int(pdf_dict_gets(dict, "VerticesPerRow"));
+	shade->u.m.bpflag = pdf_to_int(pdf_dict_gets(dict, "BitsPerFlag"));
+	shade->u.m.bpcoord = pdf_to_int(pdf_dict_gets(dict, "BitsPerCoordinate"));
+	shade->u.m.bpcomp = pdf_to_int(pdf_dict_gets(dict, "BitsPerComponent"));
 
 	obj = pdf_dict_gets(dict, "Decode");
 	if (pdf_array_len(obj) >= 6)
 	{
 		n = (pdf_array_len(obj) - 4) / 2;
-		p->x0 = pdf_to_real(pdf_array_get(obj, 0));
-		p->x1 = pdf_to_real(pdf_array_get(obj, 1));
-		p->y0 = pdf_to_real(pdf_array_get(obj, 2));
-		p->y1 = pdf_to_real(pdf_array_get(obj, 3));
+		shade->u.m.x0 = pdf_to_real(pdf_array_get(obj, 0));
+		shade->u.m.x1 = pdf_to_real(pdf_array_get(obj, 1));
+		shade->u.m.y0 = pdf_to_real(pdf_array_get(obj, 2));
+		shade->u.m.y1 = pdf_to_real(pdf_array_get(obj, 3));
 		for (i = 0; i < n; i++)
 		{
-			p->c0[i] = pdf_to_real(pdf_array_get(obj, 4 + i * 2));
-			p->c1[i] = pdf_to_real(pdf_array_get(obj, 5 + i * 2));
+			shade->u.m.c0[i] = pdf_to_real(pdf_array_get(obj, 4 + i * 2));
+			shade->u.m.c1[i] = pdf_to_real(pdf_array_get(obj, 5 + i * 2));
 		}
 	}
 
-	if (p->vprow < 2)
-		p->vprow = 2;
+	if (shade->u.m.vprow < 2)
+		shade->u.m.vprow = 2;
 
-	if (p->bpflag != 2 && p->bpflag != 4 && p->bpflag != 8)
-		p->bpflag = 8;
+	if (shade->u.m.bpflag != 2 && shade->u.m.bpflag != 4 && shade->u.m.bpflag != 8)
+		shade->u.m.bpflag = 8;
 
-	if (p->bpcoord != 1 && p->bpcoord != 2 && p->bpcoord != 4 &&
-		p->bpcoord != 8 && p->bpcoord != 12 && p->bpcoord != 16 &&
-		p->bpcoord != 24 && p->bpcoord != 32)
-		p->bpcoord = 8;
+	if (shade->u.m.bpcoord != 1 && shade->u.m.bpcoord != 2 && shade->u.m.bpcoord != 4 &&
+		shade->u.m.bpcoord != 8 && shade->u.m.bpcoord != 12 && shade->u.m.bpcoord != 16 &&
+		shade->u.m.bpcoord != 24 && shade->u.m.bpcoord != 32)
+		shade->u.m.bpcoord = 8;
 
-	if (p->bpcomp != 1 && p->bpcomp != 2 && p->bpcomp != 4 &&
-		p->bpcomp != 8 && p->bpcomp != 12 && p->bpcomp != 16)
-		p->bpcomp = 8;
+	if (shade->u.m.bpcomp != 1 && shade->u.m.bpcomp != 2 && shade->u.m.bpcomp != 4 &&
+		shade->u.m.bpcomp != 8 && shade->u.m.bpcomp != 12 && shade->u.m.bpcomp != 16)
+		shade->u.m.bpcomp = 8;
 }
 
 static void
@@ -610,67 +242,15 @@ pdf_load_type4_shade(fz_shade *shade, pdf_document *xref, pdf_obj *dict,
 	int funcs, pdf_function **func)
 {
 	fz_context *ctx = xref->ctx;
-	struct mesh_params p;
-	struct vertex va, vb, vc, vd;
-	int ncomp;
-	int flag;
-	int i;
 	fz_stream *stream;
 
-	pdf_load_mesh_params(xref, dict, &p);
+	pdf_load_mesh_params(shade, xref, dict);
 
 	if (funcs > 0)
-	{
-		ncomp = 1;
-		pdf_sample_shade_function(ctx, shade, funcs, func, p.c0[0], p.c1[0]);
-	}
-	else
-		ncomp = shade->colorspace->n;
+		pdf_sample_shade_function(ctx, shade, funcs, func, shade->u.m.c0[0], shade->u.m.c1[0]);
 
 	stream = pdf_open_stream(xref, pdf_to_num(dict), pdf_to_gen(dict));
-
-	while (!fz_is_eof_bits(stream))
-	{
-		flag = fz_read_bits(stream, p.bpflag);
-		vd.x = read_sample(stream, p.bpcoord, p.x0, p.x1);
-		vd.y = read_sample(stream, p.bpcoord, p.y0, p.y1);
-		for (i = 0; i < ncomp; i++)
-			vd.c[i] = read_sample(stream, p.bpcomp, p.c0[i], p.c1[i]);
-
-		switch (flag)
-		{
-		case 0: /* start new triangle */
-			va = vd;
-
-			fz_read_bits(stream, p.bpflag);
-			vb.x = read_sample(stream, p.bpcoord, p.x0, p.x1);
-			vb.y = read_sample(stream, p.bpcoord, p.y0, p.y1);
-			for (i = 0; i < ncomp; i++)
-				vb.c[i] = read_sample(stream, p.bpcomp, p.c0[i], p.c1[i]);
-
-			fz_read_bits(stream, p.bpflag);
-			vc.x = read_sample(stream, p.bpcoord, p.x0, p.x1);
-			vc.y = read_sample(stream, p.bpcoord, p.y0, p.y1);
-			for (i = 0; i < ncomp; i++)
-				vc.c[i] = read_sample(stream, p.bpcomp, p.c0[i], p.c1[i]);
-
-			pdf_add_triangle(ctx, shade, &va, &vb, &vc);
-			break;
-
-		case 1: /* Vb, Vc, Vd */
-			va = vb;
-			vb = vc;
-			vc = vd;
-			pdf_add_triangle(ctx, shade, &va, &vb, &vc);
-			break;
-
-		case 2: /* Va, Vc, Vd */
-			vb = vc;
-			vc = vd;
-			pdf_add_triangle(ctx, shade, &va, &vb, &vc);
-			break;
-		}
-	}
+	shade->buffer = fz_read_all(stream, 1024 /* FIXME */);
 	fz_close(stream);
 }
 
@@ -679,50 +259,15 @@ pdf_load_type5_shade(fz_shade *shade, pdf_document *xref, pdf_obj *dict,
 	int funcs, pdf_function **func)
 {
 	fz_context *ctx = xref->ctx;
-	struct mesh_params p;
-	struct vertex *buf, *ref;
-	int first;
-	int ncomp;
-	int i, k;
 	fz_stream *stream;
 
-	pdf_load_mesh_params(xref, dict, &p);
+	pdf_load_mesh_params(shade, xref, dict);
 
 	if (funcs > 0)
-	{
-		ncomp = 1;
-		pdf_sample_shade_function(ctx, shade, funcs, func, p.c0[0], p.c1[0]);
-	}
-	else
-		ncomp = shade->colorspace->n;
-
-	ref = fz_malloc_array(ctx, p.vprow, sizeof(struct vertex));
-	buf = fz_malloc_array(ctx, p.vprow, sizeof(struct vertex));
-	first = 1;
+		pdf_sample_shade_function(ctx, shade, funcs, func, shade->u.m.c0[0], shade->u.m.c1[0]);
 
 	stream = pdf_open_stream(xref, pdf_to_num(dict), pdf_to_gen(dict));
-
-	while (!fz_is_eof_bits(stream))
-	{
-		for (i = 0; i < p.vprow; i++)
-		{
-			buf[i].x = read_sample(stream, p.bpcoord, p.x0, p.x1);
-			buf[i].y = read_sample(stream, p.bpcoord, p.y0, p.y1);
-			for (k = 0; k < ncomp; k++)
-				buf[i].c[k] = read_sample(stream, p.bpcomp, p.c0[k], p.c1[k]);
-		}
-
-		if (!first)
-			for (i = 0; i < p.vprow - 1; i++)
-				pdf_add_quad(ctx, shade,
-					&ref[i], &ref[i+1], &buf[i+1], &buf[i]);
-
-		memcpy(ref, buf, p.vprow * sizeof(struct vertex));
-		first = 0;
-	}
-
-	fz_free(ctx, ref);
-	fz_free(ctx, buf);
+	shade->buffer = fz_read_all(stream, 1024 /* FIXME */);
 	fz_close(stream);
 }
 
@@ -733,121 +278,15 @@ pdf_load_type6_shade(fz_shade *shade, pdf_document *xref, pdf_obj *dict,
 	int funcs, pdf_function **func)
 {
 	fz_context *ctx = xref->ctx;
-	struct mesh_params p;
-	int haspatch, hasprevpatch;
-	float prevc[4][FZ_MAX_COLORS];
-	fz_point prevp[12];
-	int ncomp;
-	int i, k;
 	fz_stream *stream;
 
-	pdf_load_mesh_params(xref, dict, &p);
+	pdf_load_mesh_params(shade, xref, dict);
 
 	if (funcs > 0)
-	{
-		ncomp = 1;
-		pdf_sample_shade_function(ctx, shade, funcs, func, p.c0[0], p.c1[0]);
-	}
-	else
-		ncomp = shade->colorspace->n;
-
-	hasprevpatch = 0;
+		pdf_sample_shade_function(ctx, shade, funcs, func, shade->u.m.c0[0], shade->u.m.c1[0]);
 
 	stream = pdf_open_stream(xref, pdf_to_num(dict), pdf_to_gen(dict));
-
-	while (!fz_is_eof_bits(stream))
-	{
-		float c[4][FZ_MAX_COLORS];
-		fz_point v[12];
-		int startcolor;
-		int startpt;
-		int flag;
-
-		flag = fz_read_bits(stream, p.bpflag);
-
-		if (flag == 0)
-		{
-			startpt = 0;
-			startcolor = 0;
-		}
-		else
-		{
-			startpt = 4;
-			startcolor = 2;
-		}
-
-		for (i = startpt; i < 12; i++)
-		{
-			v[i].x = read_sample(stream, p.bpcoord, p.x0, p.x1);
-			v[i].y = read_sample(stream, p.bpcoord, p.y0, p.y1);
-		}
-
-		for (i = startcolor; i < 4; i++)
-		{
-			for (k = 0; k < ncomp; k++)
-				c[i][k] = read_sample(stream, p.bpcomp, p.c0[k], p.c1[k]);
-		}
-
-		haspatch = 0;
-
-		if (flag == 0)
-		{
-			haspatch = 1;
-		}
-		else if (flag == 1 && hasprevpatch)
-		{
-			v[0] = prevp[3];
-			v[1] = prevp[4];
-			v[2] = prevp[5];
-			v[3] = prevp[6];
-			memcpy(c[0], prevc[1], ncomp * sizeof(float));
-			memcpy(c[1], prevc[2], ncomp * sizeof(float));
-
-			haspatch = 1;
-		}
-		else if (flag == 2 && hasprevpatch)
-		{
-			v[0] = prevp[6];
-			v[1] = prevp[7];
-			v[2] = prevp[8];
-			v[3] = prevp[9];
-			memcpy(c[0], prevc[2], ncomp * sizeof(float));
-			memcpy(c[1], prevc[3], ncomp * sizeof(float));
-
-			haspatch = 1;
-		}
-		else if (flag == 3 && hasprevpatch)
-		{
-			v[0] = prevp[ 9];
-			v[1] = prevp[10];
-			v[2] = prevp[11];
-			v[3] = prevp[ 0];
-			memcpy(c[0], prevc[3], ncomp * sizeof(float));
-			memcpy(c[1], prevc[0], ncomp * sizeof(float));
-
-			haspatch = 1;
-		}
-
-		if (haspatch)
-		{
-			pdf_tensor_patch patch;
-
-			pdf_make_tensor_patch(&patch, 6, v);
-
-			for (i = 0; i < 4; i++)
-				memcpy(patch.color[i], c[i], ncomp * sizeof(float));
-
-			draw_patch(ctx, shade, &patch, SUBDIV, SUBDIV);
-
-			for (i = 0; i < 12; i++)
-				prevp[i] = v[i];
-
-			for (i = 0; i < 4; i++)
-				memcpy(prevc[i], c[i], ncomp * sizeof(float));
-
-			hasprevpatch = 1;
-		}
-	}
+	shade->buffer = fz_read_all(stream, 1024 /* FIXME */);
 	fz_close(stream);
 }
 
@@ -856,121 +295,15 @@ pdf_load_type7_shade(fz_shade *shade, pdf_document *xref, pdf_obj *dict,
 	int funcs, pdf_function **func)
 {
 	fz_context *ctx = xref->ctx;
-	struct mesh_params p;
-	int haspatch, hasprevpatch;
-	float prevc[4][FZ_MAX_COLORS];
-	fz_point prevp[16];
-	int ncomp;
-	int i, k;
 	fz_stream *stream;
 
-	pdf_load_mesh_params(xref, dict, &p);
+	pdf_load_mesh_params(shade, xref, dict);
 
 	if (funcs > 0)
-	{
-		ncomp = 1;
-		pdf_sample_shade_function(ctx, shade, funcs, func, p.c0[0], p.c1[0]);
-	}
-	else
-		ncomp = shade->colorspace->n;
-
-	hasprevpatch = 0;
+		pdf_sample_shade_function(ctx, shade, funcs, func, shade->u.m.c0[0], shade->u.m.c1[0]);
 
 	stream = pdf_open_stream(xref, pdf_to_num(dict), pdf_to_gen(dict));
-
-	while (!fz_is_eof_bits(stream))
-	{
-		float c[4][FZ_MAX_COLORS];
-		fz_point v[16];
-		int startcolor;
-		int startpt;
-		int flag;
-
-		flag = fz_read_bits(stream, p.bpflag);
-
-		if (flag == 0)
-		{
-			startpt = 0;
-			startcolor = 0;
-		}
-		else
-		{
-			startpt = 4;
-			startcolor = 2;
-		}
-
-		for (i = startpt; i < 16; i++)
-		{
-			v[i].x = read_sample(stream, p.bpcoord, p.x0, p.x1);
-			v[i].y = read_sample(stream, p.bpcoord, p.y0, p.y1);
-		}
-
-		for (i = startcolor; i < 4; i++)
-		{
-			for (k = 0; k < ncomp; k++)
-				c[i][k] = read_sample(stream, p.bpcomp, p.c0[k], p.c1[k]);
-		}
-
-		haspatch = 0;
-
-		if (flag == 0)
-		{
-			haspatch = 1;
-		}
-		else if (flag == 1 && hasprevpatch)
-		{
-			v[0] = prevp[3];
-			v[1] = prevp[4];
-			v[2] = prevp[5];
-			v[3] = prevp[6];
-			memcpy(c[0], prevc[1], ncomp * sizeof(float));
-			memcpy(c[1], prevc[2], ncomp * sizeof(float));
-
-			haspatch = 1;
-		}
-		else if (flag == 2 && hasprevpatch)
-		{
-			v[0] = prevp[6];
-			v[1] = prevp[7];
-			v[2] = prevp[8];
-			v[3] = prevp[9];
-			memcpy(c[0], prevc[2], ncomp * sizeof(float));
-			memcpy(c[1], prevc[3], ncomp * sizeof(float));
-
-			haspatch = 1;
-		}
-		else if (flag == 3 && hasprevpatch)
-		{
-			v[0] = prevp[ 9];
-			v[1] = prevp[10];
-			v[2] = prevp[11];
-			v[3] = prevp[ 0];
-			memcpy(c[0], prevc[3], ncomp * sizeof(float));
-			memcpy(c[1], prevc[0], ncomp * sizeof(float));
-
-			haspatch = 1;
-		}
-
-		if (haspatch)
-		{
-			pdf_tensor_patch patch;
-
-			pdf_make_tensor_patch(&patch, 7, v);
-
-			for (i = 0; i < 4; i++)
-				memcpy(patch.color[i], c[i], ncomp * sizeof(float));
-
-			draw_patch(ctx, shade, &patch, SUBDIV, SUBDIV);
-
-			for (i = 0; i < 16; i++)
-				prevp[i] = v[i];
-
-			for (i = 0; i < 4; i++)
-				memcpy(prevc[i], c[i], FZ_MAX_COLORS * sizeof(float));
-
-			hasprevpatch = 1;
-		}
-	}
+	shade->buffer = fz_read_all(stream, 1024 /* FIXME */);
 	fz_close(stream);
 }
 
@@ -996,17 +329,11 @@ pdf_load_shading_dict(pdf_document *xref, pdf_obj *dict, fz_matrix transform)
 	{
 		shade = fz_malloc_struct(ctx, fz_shade);
 		FZ_INIT_STORABLE(shade, 1, fz_free_shade_imp);
-		shade->type = FZ_MESH;
+		shade->type = FZ_MESH_TYPE4;
 		shade->use_background = 0;
 		shade->use_function = 0;
 		shade->matrix = transform;
 		shade->bbox = fz_infinite_rect;
-		shade->extend[0] = 0;
-		shade->extend[1] = 0;
-
-		shade->mesh_len = 0;
-		shade->mesh_cap = 0;
-		shade->mesh = NULL;
 
 		shade->colorspace = NULL;
 
@@ -1069,6 +396,7 @@ pdf_load_shading_dict(pdf_document *xref, pdf_obj *dict, fz_matrix transform)
 			}
 		}
 
+		shade->type = type;
 		switch (type)
 		{
 		case 1: pdf_load_function_based_shading(shade, xref, dict, func[0]); break;
@@ -1103,7 +431,9 @@ fz_shade_size(fz_shade *s)
 {
 	if (s == NULL)
 		return 0;
-	return sizeof(*s) + s->mesh_cap * sizeof(*s->mesh) + s->colorspace->size;
+	if (s->type != 1)
+		return sizeof(*s);
+	return sizeof(*s) + sizeof(float) * s->u.f.xdivs * s->u.f.ydivs * s->colorspace->n;
 }
 
 fz_shade *