fixes scale on derivative maps

8 files changed, 212 insertions, 30 deletions

diff --git a/source/blender/blenkernel/BKE_DerivedMesh.h b/source/blender/blenkernel/BKE_DerivedMesh.h
index 429d5e84785..5f451ce7821 100644
--- a/source/blender/blenkernel/BKE_DerivedMesh.h
+++ b/source/blender/blenkernel/BKE_DerivedMesh.h
@@ -94,6 +94,7 @@ struct DerivedMesh {
 	BVHCache bvhCache;
 	struct GPUDrawObject *drawObject;
 	DerivedMeshType type;
+	float auto_bump_scale;
 
 	/* Misc. Queries */
 
@@ -578,6 +579,7 @@ void DM_vertex_attributes_from_gpu(DerivedMesh *dm,
 	struct GPUVertexAttribs *gattribs, DMVertexAttribs *attribs);
 
 void DM_add_tangent_layer(DerivedMesh *dm);
+void DM_calc_auto_bump_scale(DerivedMesh *dm);
 
 /* Set object's bounding box based on DerivedMesh min/max data */
 void DM_set_object_boundbox(struct Object *ob, DerivedMesh *dm);
diff --git a/source/blender/blenkernel/intern/DerivedMesh.c b/source/blender/blenkernel/intern/DerivedMesh.c
index 4b91911a6ab..c829d840ace 100644
--- a/source/blender/blenkernel/intern/DerivedMesh.c
+++ b/source/blender/blenkernel/intern/DerivedMesh.c
@@ -184,6 +184,7 @@ void DM_init(DerivedMesh *dm, DerivedMeshType type,
 	DM_init_funcs(dm);
 	
 	dm->needsFree = 1;
+	dm->auto_bump_scale = -1.0f;
 }
 
 void DM_from_template(DerivedMesh *dm, DerivedMesh *source, DerivedMeshType type,
@@ -1831,6 +1832,159 @@ void DM_add_tangent_layer(DerivedMesh *dm)
 	MEM_freeN(vtangents);
 }
 
+void DM_calc_auto_bump_scale(DerivedMesh *dm)
+{
+	int totvert= dm->getNumVerts(dm);
+	int totface= dm->getNumFaces(dm);
+
+	MVert * mvert = dm->getVertArray(dm);
+	MFace * mface = dm->getFaceArray(dm);
+	MTFace * mtface = dm->getFaceDataArray(dm, CD_MTFACE);
+
+	if(mtface)
+	{
+		double dsum = 0.0;
+		int nr_accumulated = 0;
+		int f;
+
+		for ( f=0; f<totface; f++ )
+		{
+			{
+				float * verts[4], * tex_coords[4];
+				const int nr_verts = mface[f].v4!=0 ? 4 : 3;
+				int i, is_degenerate;
+
+				verts[0]=mvert[mface[f].v1].co; verts[1]=mvert[mface[f].v2].co; verts[2]=mvert[mface[f].v3].co;
+				tex_coords[0]=mtface[f].uv[0]; tex_coords[1]=mtface[f].uv[1]; tex_coords[2]=mtface[f].uv[2];
+				if(nr_verts==4)
+				{
+					verts[3]=mvert[mface[f].v4].co;
+					tex_coords[3]=mtface[f].uv[3];
+				}
+
+				// discard degenerate faces
+				is_degenerate = 0;
+				if(	equals_v3v3(verts[0], verts[1]) || equals_v3v3(verts[0], verts[2]) || equals_v3v3(verts[1], verts[2]) ||
+					equals_v2v2(tex_coords[0], tex_coords[1]) || equals_v2v2(tex_coords[0], tex_coords[2]) || equals_v2v2(tex_coords[1], tex_coords[2]) )
+				{
+					is_degenerate = 1;
+				}
+
+				// verify last vertex as well if this is a quad
+				if ( is_degenerate==0 && nr_verts==4 )
+				{
+					if(	equals_v3v3(verts[3], verts[0]) || equals_v3v3(verts[3], verts[1]) || equals_v3v3(verts[3], verts[2]) ||
+						equals_v2v2(tex_coords[3], tex_coords[0]) || equals_v2v2(tex_coords[3], tex_coords[1]) || equals_v2v2(tex_coords[3], tex_coords[2]) )
+					{
+						is_degenerate = 1;
+					}
+
+					// verify the winding is consistent
+					if ( is_degenerate==0 )
+					{
+						float prev_edge[2];
+						int is_signed = 0;
+						sub_v2_v2v2(prev_edge, tex_coords[0], tex_coords[3]);
+
+						i = 0;
+						while ( is_degenerate==0 && i<4 )
+						{
+							float cur_edge[2], signed_area;
+							sub_v2_v2v2(cur_edge, tex_coords[(i+1)&0x3], tex_coords[i]);
+							signed_area = prev_edge[0]*cur_edge[1] - prev_edge[1]*cur_edge[0];
+							if ( i==0 ) is_signed = signed_area<0.0f ? 1 : 0;
+							else if((is_signed!=0)!=(signed_area<0.0f)) is_degenerate=1;
+
+							if ( is_degenerate==0 )
+							{
+								copy_v2_v2(prev_edge, cur_edge);
+								++i;
+							}
+						}
+					}
+				}
+
+				// proceed if not a degenerate face
+				if ( is_degenerate==0 )
+				{
+					int nr_tris_to_pile=0;
+					// quads split at shortest diagonal
+					int offs = 0;		// initial triangulation is 0,1,2 and 0, 2, 3
+					if ( nr_verts==4 )
+					{
+						float pos_len_diag0, pos_len_diag1;
+						float vtmp[3];
+						sub_v3_v3v3(vtmp, verts[2], verts[0]);
+						pos_len_diag0 = dot_v3v3(vtmp, vtmp);
+						sub_v3_v3v3(vtmp, verts[3], verts[1]);
+						pos_len_diag1 = dot_v3v3(vtmp, vtmp);
+
+						if(pos_len_diag1<pos_len_diag0)
+							offs=1;		// alter split
+						else if(pos_len_diag0==pos_len_diag1)		// do UV check instead
+						{
+							float tex_len_diag0, tex_len_diag1;
+
+							sub_v2_v2v2(vtmp, tex_coords[2], tex_coords[0]);
+							tex_len_diag0 = dot_v2v2(vtmp, vtmp);
+							sub_v2_v2v2(vtmp, tex_coords[3], tex_coords[1]);
+							tex_len_diag1 = dot_v2v2(vtmp, vtmp);
+
+							if(tex_len_diag1<tex_len_diag0)
+							{
+								offs=1;		// alter split
+							}
+						}
+					}
+					nr_tris_to_pile = nr_verts-2 ;
+					if ( nr_tris_to_pile==1 || nr_tris_to_pile==2 )
+					{
+						const int indices[] = {offs+0, offs+1, offs+2, offs+0, offs+2, (offs+3)&0x3 };
+						int t;
+						for ( t=0; t<nr_tris_to_pile; t++ )
+						{
+							float f2x_area_uv;
+							float * p0 = verts[indices[t*3+0]];
+							float * p1 = verts[indices[t*3+1]];
+							float * p2 = verts[indices[t*3+2]];
+
+							float edge_t0[2], edge_t1[2];
+							sub_v2_v2v2(edge_t0, tex_coords[indices[t*3+1]], tex_coords[indices[t*3+0]]);
+							sub_v2_v2v2(edge_t1, tex_coords[indices[t*3+2]], tex_coords[indices[t*3+0]]);
+
+							f2x_area_uv = fabsf(edge_t0[0]*edge_t1[1] - edge_t0[1]*edge_t1[0]);
+							if ( f2x_area_uv>FLT_EPSILON )
+							{
+								float norm[3], v0[3], v1[3], f2x_surf_area, fsurf_ratio;
+								sub_v3_v3v3(v0, p1, p0);
+								sub_v3_v3v3(v1, p2, p0);
+								cross_v3_v3v3(norm, v0, v1);
+
+								f2x_surf_area = len_v3(norm);
+								fsurf_ratio = f2x_surf_area/f2x_area_uv;	// tri area divided by texture area
+
+								++nr_accumulated;
+								dsum += (double)(fsurf_ratio);
+							}
+						}
+					}
+				}
+			}
+		}
+
+		// finalize
+		{
+			const float avg_area_ratio = (nr_accumulated>0) ? ((float)(dsum / nr_accumulated)) : 1.0f;
+			const float use_as_render_bump_scale = sqrtf(avg_area_ratio);		// use width of average surface ratio as your bump scale
+			dm->auto_bump_scale = use_as_render_bump_scale;
+		}
+	}
+	else
+	{
+		dm->auto_bump_scale = 1.0f;
+	}
+}
+
 void DM_vertex_attributes_from_gpu(DerivedMesh *dm, GPUVertexAttribs *gattribs, DMVertexAttribs *attribs)
 {
 	CustomData *vdata, *fdata, *tfdata = NULL;
@@ -1851,6 +2005,10 @@ void DM_vertex_attributes_from_gpu(DerivedMesh *dm, GPUVertexAttribs *gattribs,
 	else
 		tfdata = fdata;
 
+	/* calc auto bump scale if necessary */
+	if(dm->auto_bump_scale<=0.0f)
+		DM_calc_auto_bump_scale(dm);
+
 	/* add a tangent layer if necessary */
 	for(b = 0; b < gattribs->totlayer; b++)
 		if(gattribs->layer[b].type == CD_TANGENT)
diff --git a/source/blender/gpu/GPU_material.h b/source/blender/gpu/GPU_material.h
index 418bea5fe9c..5689f02d2d7 100644
--- a/source/blender/gpu/GPU_material.h
+++ b/source/blender/gpu/GPU_material.h
@@ -83,7 +83,8 @@ typedef enum GPUBuiltin {
 	GPU_INVERSE_OBJECT_MATRIX = 8,
 	GPU_VIEW_POSITION = 16,
 	GPU_VIEW_NORMAL = 32,
-	GPU_OBCOLOR = 64
+	GPU_OBCOLOR = 64,
+	GPU_AUTO_BUMPSCALE = 128
 } GPUBuiltin;
 
 typedef enum GPUBlendMode {
@@ -129,7 +130,7 @@ void GPU_material_free(struct Material *ma);
 void GPU_materials_free(void);
 
 void GPU_material_bind(GPUMaterial *material, int oblay, int viewlay, double time, int mipmap);
-void GPU_material_bind_uniforms(GPUMaterial *material, float obmat[][4], float viewmat[][4], float viewinv[][4], float obcol[4]);
+void GPU_material_bind_uniforms(GPUMaterial *material, float obmat[][4], float viewmat[][4], float viewinv[][4], float obcol[4], float autobumpscale);
 void GPU_material_unbind(GPUMaterial *material);
 int GPU_material_bound(GPUMaterial *material);
 
@@ -162,6 +163,7 @@ typedef enum GPUDynamicType {
 	GPU_DYNAMIC_OBJECT_VIEWIMAT = 3,
 	GPU_DYNAMIC_OBJECT_IMAT = 4,
 	GPU_DYNAMIC_OBJECT_COLOR = 5,
+	GPU_DYNAMIC_OBJECT_AUTOBUMPSCALE = 15,
 	GPU_DYNAMIC_LAMP_FIRST = 6,
 	GPU_DYNAMIC_LAMP_DYNVEC = 6,
 	GPU_DYNAMIC_LAMP_DYNCO = 7,
diff --git a/source/blender/gpu/intern/gpu_codegen.c b/source/blender/gpu/intern/gpu_codegen.c
index 185f9eb185e..05c459b703d 100644
--- a/source/blender/gpu/intern/gpu_codegen.c
+++ b/source/blender/gpu/intern/gpu_codegen.c
@@ -344,6 +344,8 @@ const char *GPU_builtin_name(GPUBuiltin builtin)
 		return "varnormal";
 	else if(builtin == GPU_OBCOLOR)
 		return "unfobcolor";
+	else if(builtin == GPU_AUTO_BUMPSCALE)
+		return "unfobautobumpscale";
 	else
 		return "";
 }
diff --git a/source/blender/gpu/intern/gpu_draw.c b/source/blender/gpu/intern/gpu_draw.c
index 3cd3cde8aad..6a11010002f 100644
--- a/source/blender/gpu/intern/gpu_draw.c
+++ b/source/blender/gpu/intern/gpu_draw.c
@@ -60,6 +60,7 @@
 #include "BKE_node.h"
 #include "BKE_object.h"
 #include "BKE_scene.h"
+#include "BKE_DerivedMesh.h"
 
 #include "BLI_threads.h"
 #include "BLI_blenlib.h"
@@ -1159,7 +1160,7 @@ int GPU_enable_material(int nr, void *attribs)
 			gpumat = GPU_material_from_blender(GMS.gscene, mat);
 			GPU_material_vertex_attributes(gpumat, gattribs);
 			GPU_material_bind(gpumat, GMS.gob->lay, GMS.glay, 1.0, !(GMS.gob->mode & OB_MODE_TEXTURE_PAINT));
-			GPU_material_bind_uniforms(gpumat, GMS.gob->obmat, GMS.gviewmat, GMS.gviewinv, GMS.gob->col);
+			GPU_material_bind_uniforms(gpumat, GMS.gob->obmat, GMS.gviewmat, GMS.gviewinv, GMS.gob->col, GMS.gob->derivedFinal->auto_bump_scale);
 			GMS.gboundmat= mat;
 
 			/* for glsl use alpha blend mode, unless it's set to solid and
diff --git a/source/blender/gpu/intern/gpu_material.c b/source/blender/gpu/intern/gpu_material.c
index 397c0e32c69..2d8487deb71 100644
--- a/source/blender/gpu/intern/gpu_material.c
+++ b/source/blender/gpu/intern/gpu_material.c
@@ -94,7 +94,7 @@ struct GPUMaterial {
 	/* for passing uniforms */
 	int viewmatloc, invviewmatloc;
 	int obmatloc, invobmatloc;
-	int obcolloc;
+	int obcolloc, obautobumpscaleloc;
 
 	ListBase lamps;
 };
@@ -212,7 +212,8 @@ static int GPU_material_construct_end(GPUMaterial *material)
 			material->invobmatloc = GPU_shader_get_uniform(shader, GPU_builtin_name(GPU_INVERSE_OBJECT_MATRIX));
 		if(material->builtins & GPU_OBCOLOR)
 			material->obcolloc = GPU_shader_get_uniform(shader, GPU_builtin_name(GPU_OBCOLOR));
-
+		if(material->builtins & GPU_AUTO_BUMPSCALE)
+			material->obautobumpscaleloc = GPU_shader_get_uniform(shader, GPU_builtin_name(GPU_AUTO_BUMPSCALE));
 		return 1;
 	}
 
@@ -273,7 +274,7 @@ void GPU_material_bind(GPUMaterial *material, int oblay, int viewlay, double tim
 	}
 }
 
-void GPU_material_bind_uniforms(GPUMaterial *material, float obmat[][4], float viewmat[][4], float viewinv[][4], float obcol[4])
+void GPU_material_bind_uniforms(GPUMaterial *material, float obmat[][4], float viewmat[][4], float viewinv[][4], float obcol[4], float autobumpscale)
 {
 	if(material->pass) {
 		GPUShader *shader = GPU_pass_shader(material->pass);
@@ -300,7 +301,9 @@ void GPU_material_bind_uniforms(GPUMaterial *material, float obmat[][4], float v
 			CLAMP(col[3], 0.0f, 1.0f);
 			GPU_shader_uniform_vector(shader, material->obcolloc, 4, 1, col);
 		}
-
+		if(material->builtins & GPU_AUTO_BUMPSCALE) {
+			GPU_shader_uniform_vector(shader, material->obautobumpscaleloc, 1, 1, &autobumpscale);
+		}
 		/* update lamps */
 		for(nlink=material->lamps.first; nlink; nlink=nlink->next) {
 			lamp= nlink->data;
@@ -1087,8 +1090,7 @@ static void do_material_tex(GPUShadeInput *shi)
 						/* ntap bumpmap image */
 						int iBumpSpace;
 						float ima_x, ima_y;
-						float hScale = 0.1f; // compatibility adjustment factor for all bumpspace types
-						float hScaleTex = 13.0f; // factor for scaling texspace bumps
+						float hScale; 
 
 						float imag_tspace_dimension_x = 1024.0f;		// only used for texture space variant
 						float aspect = 1.0f;
@@ -1096,16 +1098,35 @@ static void do_material_tex(GPUShadeInput *shi)
 						GPUNodeLink *surf_pos = GPU_builtin(GPU_VIEW_POSITION);
 						GPUNodeLink *vR1, *vR2;
 						GPUNodeLink *dBs, *dBt, *fDet;
-						
+
+						hScale = 0.1;		// compatibility adjustment factor for all bumpspace types
 						if( mtex->texflag & MTEX_BUMP_TEXTURESPACE )
-							hScale = hScaleTex;
+							hScale = 13.0f;		// factor for scaling texspace bumps
+						else if(found_deriv_map!=0)
+							hScale = 1.0f;
+
+						// resolve texture resolution
+						if( (mtex->texflag & MTEX_BUMP_TEXTURESPACE) || found_deriv_map ) {
+							ImBuf *ibuf= BKE_image_get_ibuf(tex->ima, &tex->iuser);
+							ima_x= 512.0f; ima_y= 512.f;		// prevent calling textureSize, glsl 1.3 only
+							if(ibuf) {
+								ima_x= ibuf->x;
+								ima_y= ibuf->y;
+								aspect = ((float) ima_y) / ima_x;
+							}
+						}
 
 						// The negate on norfac is done because the
 						// normal in the renderer points inward which corresponds
 						// to inverting the bump map. Should this ever change
 						// this negate must be removed.
 						norfac = -hScale * mtex->norfac;
+						if(found_deriv_map) norfac /= sqrtf(ima_x*ima_y);
+
 						tnorfac = GPU_uniform(&norfac);
+
+						if(found_deriv_map)
+							GPU_link(mat, "math_multiply", tnorfac, GPU_builtin(GPU_AUTO_BUMPSCALE), &tnorfac);
 						
 						if(GPU_link_changed(stencil))
 							GPU_link(mat, "math_multiply", tnorfac, stencil, &tnorfac);
@@ -1152,17 +1173,6 @@ static void do_material_tex(GPUShadeInput *shi)
 							
 							iBumpSpacePrev = iBumpSpace;
 						}
-
-						// resolve texture resolution
-						if( (mtex->texflag & MTEX_BUMP_TEXTURESPACE) || found_deriv_map ) {
-							ImBuf *ibuf= BKE_image_get_ibuf(tex->ima, &tex->iuser);
-							ima_x= 512.0f; ima_y= 512.f;		// prevent calling textureSize, glsl 1.3 only
-							if(ibuf) {
-								ima_x= ibuf->x;
-								ima_y= ibuf->y;
-								aspect = ((float) ima_y) / ima_x;
-							}
-						}
 						
 						
 						if(found_deriv_map) {
@@ -1703,6 +1713,7 @@ GPUShaderExport *GPU_shader_export(struct Scene *scene, struct Material *ma)
 		{ GPU_OBJECT_MATRIX, GPU_DYNAMIC_OBJECT_MAT, GPU_DATA_16F },
 		{ GPU_INVERSE_OBJECT_MATRIX, GPU_DYNAMIC_OBJECT_IMAT, GPU_DATA_16F },
 		{ GPU_OBCOLOR, GPU_DYNAMIC_OBJECT_COLOR, GPU_DATA_4F },
+		{ GPU_AUTO_BUMPSCALE, GPU_DYNAMIC_OBJECT_AUTOBUMPSCALE, GPU_DATA_1F },
 		{ 0 }
 	};
 
diff --git a/source/blender/python/intern/gpu.c b/source/blender/python/intern/gpu.c
index 57ce7b59067..0da63297077 100644
--- a/source/blender/python/intern/gpu.c
+++ b/source/blender/python/intern/gpu.c
@@ -87,6 +87,7 @@ PyInit_gpu(void)
 	PY_MODULE_ADD_CONSTANT(m, GPU_DYNAMIC_OBJECT_VIEWIMAT);
 	PY_MODULE_ADD_CONSTANT(m, GPU_DYNAMIC_OBJECT_IMAT);
 	PY_MODULE_ADD_CONSTANT(m, GPU_DYNAMIC_OBJECT_COLOR);
+	PY_MODULE_ADD_CONSTANT(m, GPU_DYNAMIC_OBJECT_AUTOBUMPSCALE);
 	PY_MODULE_ADD_CONSTANT(m, GPU_DYNAMIC_LAMP_DYNVEC);
 	PY_MODULE_ADD_CONSTANT(m, GPU_DYNAMIC_LAMP_DYNCO);
 	PY_MODULE_ADD_CONSTANT(m, GPU_DYNAMIC_LAMP_DYNIMAT);
diff --git a/source/blender/render/intern/source/render_texture.c b/source/blender/render/intern/source/render_texture.c
index ae48b0f777f..e55d2676b17 100644
--- a/source/blender/render/intern/source/render_texture.c
+++ b/source/blender/render/intern/source/render_texture.c
@@ -1932,11 +1932,13 @@ static int ntap_bump_compute(NTapBump *ntap_bump, ShadeInput *shi, MTex *mtex, T
 	float *nvec = texres->nor;
 	texres->nor = NULL;
 
-	if( mtex->texflag & MTEX_BUMP_TEXTURESPACE ) {
-		if(tex->ima)
-			Hscale *= 13.0f; // appears to be a sensible default value
-	} else
-		Hscale *= 0.1f; // factor 0.1 proved to look like the previous bump code
+	if(found_deriv_map==0) {
+		if( mtex->texflag & MTEX_BUMP_TEXTURESPACE ) {
+			if(tex->ima)
+				Hscale *= 13.0f; // appears to be a sensible default value
+		} else
+			Hscale *= 0.1f; // factor 0.1 proved to look like the previous bump code
+	}
 
 	if( !ntap_bump->init_done ) {
 		copy_v3_v3(ntap_bump->vNacc, shi->vn);
@@ -1958,15 +1960,18 @@ static int ntap_bump_compute(NTapBump *ntap_bump, ShadeInput *shi, MTex *mtex, T
 	}
 	
 	if(found_deriv_map) {
-		float dBdu, dBdv;
+		float dBdu, dBdv, auto_bump;
 		float s = 1;		// negate this if flipped texture coordinate
 		texco_mapping(shi, tex, mtex, co, dx, dy, texvec, dxt, dyt);
 		rgbnor = multitex_mtex(shi, mtex, texvec, dxt, dyt, texres);
+
+		auto_bump = shi->obr->ob->derivedFinal->auto_bump_scale;
+		auto_bump /= sqrtf((float) (dimx*dimy));
 		
 		// this variant using a derivative map is described here
 		// http://mmikkelsen3d.blogspot.com/2011/07/derivative-maps.html
-		dBdu = Hscale*dimx*(2*texres->tr-1);
-		dBdv = Hscale*dimy*(2*texres->tg-1);
+		dBdu = auto_bump*Hscale*dimx*(2*texres->tr-1);
+		dBdv = auto_bump*Hscale*dimy*(2*texres->tg-1);
 
 		dHdx = dBdu*dxt[0] + s * dBdv*dxt[1];
 		dHdy = dBdu*dyt[0] + s * dBdv*dyt[1];