new bumpmapping options for the renderer

oldbump -> original
newbump -> compatible
 *new*	-> default (3tap)
 *new*	-> best quality (5tap)

the latter two have an option to apply bumpmapping in 
 viewspace - much like displacement mapping
 objectspace - default (scales with the object)
 texturespace - much like normal mapping (scales)

7 files changed, 463 insertions, 197 deletions

diff --git a/source/blender/blenkernel/intern/material.c b/source/blender/blenkernel/intern/material.c
index c22384a6bdc..c49d8310789 100644
--- a/source/blender/blenkernel/intern/material.c
+++ b/source/blender/blenkernel/intern/material.c
@@ -820,7 +820,7 @@ static void do_init_render_material(Material *ma, int r_mode, float *amb)
 
 			/* always get derivatives for these textures */
 			if ELEM3(mtex->tex->type, TEX_IMAGE, TEX_PLUGIN, TEX_ENVMAP) ma->texco |= TEXCO_OSA;
-			else if(mtex->texflag & MTEX_NEW_BUMP) ma->texco |= TEXCO_OSA;
+			else if(mtex->texflag & (MTEX_COMPAT_BUMP|MTEX_3TAP_BUMP|MTEX_5TAP_BUMP)) ma->texco |= TEXCO_OSA;
 			
 			if(ma->texco & (TEXCO_ORCO|TEXCO_REFL|TEXCO_NORM|TEXCO_STRAND|TEXCO_STRESS)) needuv= 1;
 			else if(ma->texco & (TEXCO_GLOB|TEXCO_UV|TEXCO_OBJECT|TEXCO_SPEED)) needuv= 1;
diff --git a/source/blender/blenkernel/intern/texture.c b/source/blender/blenkernel/intern/texture.c
index 11c0ce74b4a..48903fa5f5f 100644
--- a/source/blender/blenkernel/intern/texture.c
+++ b/source/blender/blenkernel/intern/texture.c
@@ -630,7 +630,8 @@ void default_mtex(MTex *mtex)
 	mtex->size[1]= 1.0;
 	mtex->size[2]= 1.0;
 	mtex->tex= 0;
-	mtex->texflag= MTEX_NEW_BUMP;
+	mtex->texflag= MTEX_3TAP_BUMP;
+	mtex->texflag= MTEX_BUMP_OBJECTSPACE;
 	mtex->colormodel= 0;
 	mtex->r= 1.0;
 	mtex->g= 0.0;
diff --git a/source/blender/blenloader/intern/readfile.c b/source/blender/blenloader/intern/readfile.c
index e34a84139aa..77230637342 100644
--- a/source/blender/blenloader/intern/readfile.c
+++ b/source/blender/blenloader/intern/readfile.c
@@ -9991,12 +9991,15 @@ static void do_versions(FileData *fd, Library *lib, Main *main)
 			for(a=0; a<MAX_MTEX; a++) {
 				if(ma->mtex[a]) {
 					tex= ma->mtex[a]->tex;
-					if(!tex)
-						ma->mtex[a]->texflag |= MTEX_NEW_BUMP;
-					else {
+					if(!tex) {
+						ma->mtex[a]->texflag |= MTEX_3TAP_BUMP;
+						ma->mtex[a]->texflag |= MTEX_BUMP_OBJECTSPACE;
+					} else {
 						tex= (Tex*)newlibadr(fd, ma->id.lib, tex);
-						if(tex && tex->type == 0) /* invalid type */
-							ma->mtex[a]->texflag |= MTEX_NEW_BUMP;
+						if(tex && tex->type == 0) { /* invalid type */
+							ma->mtex[a]->texflag |= MTEX_3TAP_BUMP;
+							ma->mtex[a]->texflag |= MTEX_BUMP_OBJECTSPACE;
+						}
 					}
 				}
 			}
diff --git a/source/blender/makesdna/DNA_texture_types.h b/source/blender/makesdna/DNA_texture_types.h
index 4d11890f5a5..41eebca821c 100644
--- a/source/blender/makesdna/DNA_texture_types.h
+++ b/source/blender/makesdna/DNA_texture_types.h
@@ -454,7 +454,11 @@ typedef struct TexMapping {
 #define MTEX_VIEWSPACE		16
 #define MTEX_DUPLI_MAPTO	32
 #define MTEX_OB_DUPLI_ORIG	64
-#define MTEX_NEW_BUMP		128
+#define MTEX_COMPAT_BUMP	128
+#define MTEX_3TAP_BUMP		256
+#define MTEX_5TAP_BUMP		512
+#define MTEX_BUMP_OBJECTSPACE	1024
+#define MTEX_BUMP_TEXTURESPACE	2048
 
 /* blendtype */
 #define MTEX_BLEND		0
diff --git a/source/blender/makesrna/intern/rna_material.c b/source/blender/makesrna/intern/rna_material.c
index aecb89a0fee..4c8ed475a8d 100644
--- a/source/blender/makesrna/intern/rna_material.c
+++ b/source/blender/makesrna/intern/rna_material.c
@@ -378,9 +378,17 @@ static void rna_def_material_mtex(BlenderRNA *brna)
 
 	static EnumPropertyItem prop_bump_method_items[] = {
 		{0, "BUMP_ORIGINAL", 0, "Original", ""},
-		{MTEX_NEW_BUMP, "BUMP_IMPROVED", 0, "Improved", ""},
+		{MTEX_COMPAT_BUMP, "BUMP_COMPATIBLE", 0, "Compatible", ""},
+		{MTEX_3TAP_BUMP, "BUMP_DEFAULT", 0, "Default", ""},
+		{MTEX_5TAP_BUMP, "BUMP_BEST_QUALITY", 0, "Best Quality", ""},
 		{0, NULL, 0, NULL, NULL}};
 
+	static EnumPropertyItem prop_bump_space_items[] = {
+		{0, "BUMP_VIEWSPACE", 0, "ViewSpace", ""},
+		{MTEX_BUMP_OBJECTSPACE, "BUMP_OBJECTSPACE", 0, "ObjectSpace", ""},
+		{MTEX_BUMP_TEXTURESPACE, "BUMP_TEXTURESPACE", 0, "TextureSpace", ""},
+		{0, NULL, 0, NULL, NULL}};
+	
 	srna= RNA_def_struct(brna, "MaterialTextureSlot", "TextureSlot");
 	RNA_def_struct_sdna(srna, "MTex");
 	RNA_def_struct_ui_text(srna, "Material Texture Slot", "Texture slot for textures in a Material datablock");
@@ -687,6 +695,12 @@ static void rna_def_material_mtex(BlenderRNA *brna)
 	RNA_def_property_enum_items(prop, prop_bump_method_items);
 	RNA_def_property_ui_text(prop, "Bump Method", "Method to use for bump mapping");
 	RNA_def_property_update(prop, 0, "rna_Material_update");
+	
+	prop= RNA_def_property(srna, "bump_objectspace", PROP_ENUM, PROP_NONE);
+	RNA_def_property_enum_bitflag_sdna(prop, NULL, "texflag");
+	RNA_def_property_enum_items(prop, prop_bump_space_items);
+	RNA_def_property_ui_text(prop, "Bump Space", "Space to apply bump mapping in");
+	RNA_def_property_update(prop, 0, "rna_Material_update");
 }
 
 static void rna_def_material_colors(StructRNA *srna)
diff --git a/source/blender/render/intern/include/texture.h b/source/blender/render/intern/include/texture.h
index fb941d1b7f3..8eb91f3299f 100644
--- a/source/blender/render/intern/include/texture.h
+++ b/source/blender/render/intern/include/texture.h
@@ -48,6 +48,7 @@ if(tex->saturation != 1.0f) { \
 	hsv_to_rgb(_hsv[0], _hsv[1], _hsv[2], &texres->tr, &texres->tg, &texres->tb); \
 } \
 
+#define RGBTOBW(r,g,b)	( r*0.35 + g*0.45 + b*0.2 )		/* keep this in sync with gpu_shader_material.glsl:rgbtobw */
 
 struct HaloRen;
 struct ShadeInput;
diff --git a/source/blender/render/intern/source/texture.c b/source/blender/render/intern/source/texture.c
index 9563e802945..a243149254e 100644
--- a/source/blender/render/intern/source/texture.c
+++ b/source/blender/render/intern/source/texture.c
@@ -1607,6 +1607,7 @@ static void texco_mapping(ShadeInput* shi, Tex* tex, MTex* mtex, float* co, floa
 		do_2d_mapping(mtex, texvec, shi->vlr, shi->facenor, dxt, dyt);
 
 		// translate and scale
+		/*
 		texvec[0] = mtex->size[0]*(texvec[0] - 0.5f) + mtex->ofs[0] + 0.5f;
 		texvec[1] = mtex->size[1]*(texvec[1] - 0.5f) + mtex->ofs[1] + 0.5f;
 		if (shi->osatex) {
@@ -1615,6 +1616,7 @@ static void texco_mapping(ShadeInput* shi, Tex* tex, MTex* mtex, float* co, floa
 			dyt[0] = mtex->size[0]*dyt[0];
 			dyt[1] = mtex->size[1]*dyt[1];
 		}
+		*/
 		
 		/* problem: repeat-mirror is not a 'repeat' but 'extend' in imagetexture.c */
 		// TXF: bug was here, only modify texvec when repeat mode set, old code affected other modes too.
@@ -1667,6 +1669,410 @@ static void texco_mapping(ShadeInput* shi, Tex* tex, MTex* mtex, float* co, floa
 	}
 }
 
+/* Bump code from 2.5 development cycle, has a number of bugs, but here for compatibility */
+
+typedef struct CompatibleBump {
+	float nu[3], nv[3], nn[3];
+	float dudnu, dudnv, dvdnu, dvdnv;
+	int nunvdone;
+} CompatibleBump;
+
+static void compatible_bump_init(CompatibleBump *compat_bump)
+{
+	memset(compat_bump, 0, sizeof(*compat_bump));
+
+	compat_bump->dudnu = 1.0f;
+	compat_bump->dvdnv = 1.0f;
+}
+
+static void compatible_bump_uv_derivs(CompatibleBump *compat_bump, ShadeInput *shi, MTex *mtex, int i)
+{
+	// uvmapping only, calculation of normal tangent u/v partial derivatives
+	// (should not be here, dudnu, dudnv, dvdnu & dvdnv should probably be part of ShadeInputUV struct,
+	//  nu/nv in ShadeInput and this calculation should then move to shadeinput.c, shade_input_set_shade_texco() func.)
+	// NOTE: test for shi->obr->ob here, since vlr/obr/obi can be 'fake' when called from fastshade(), another reason to move it..
+	// NOTE: shi->v1 is NULL when called from displace_render_vert, assigning verts in this case is not trivial because the shi quad face side is not know.
+	if ((mtex->texflag & MTEX_COMPAT_BUMP) && shi->obr && shi->obr->ob && shi->v1) {
+		if(mtex->mapto & (MAP_NORM|MAP_WARP) && !((mtex->tex->type==TEX_IMAGE) && (mtex->tex->imaflag & TEX_NORMALMAP))) {
+			MTFace* tf = RE_vlakren_get_tface(shi->obr, shi->vlr, i, NULL, 0);
+			int j1 = shi->i1, j2 = shi->i2, j3 = shi->i3;
+
+			vlr_set_uv_indices(shi->vlr, &j1, &j2, &j3);
+
+			// compute ortho basis around normal
+			if(!compat_bump->nunvdone) {
+				// render normal is negated
+				compat_bump->nn[0] = -shi->vn[0];
+				compat_bump->nn[1] = -shi->vn[1];
+				compat_bump->nn[2] = -shi->vn[2];
+				ortho_basis_v3v3_v3(compat_bump->nu, compat_bump->nv, compat_bump->nn);
+				compat_bump->nunvdone= 1;
+			}
+
+			if (tf) {
+				float *uv1 = tf->uv[j1], *uv2 = tf->uv[j2], *uv3 = tf->uv[j3];
+				const float an[3] = {fabsf(compat_bump->nn[0]), fabsf(compat_bump->nn[1]), fabsf(compat_bump->nn[2])};
+				const int a1 = (an[0] > an[1] && an[0] > an[2]) ? 1 : 0;
+				const int a2 = (an[2] > an[0] && an[2] > an[1]) ? 1 : 2;
+				const float dp1_a1 = shi->v1->co[a1] - shi->v3->co[a1];
+				const float dp1_a2 = shi->v1->co[a2] - shi->v3->co[a2];
+				const float dp2_a1 = shi->v2->co[a1] - shi->v3->co[a1];
+				const float dp2_a2 = shi->v2->co[a2] - shi->v3->co[a2];
+				const float du1 = uv1[0] - uv3[0], du2 = uv2[0] - uv3[0];
+				const float dv1 = uv1[1] - uv3[1], dv2 = uv2[1] - uv3[1];
+				const float dpdu_a1 = dv2*dp1_a1 - dv1*dp2_a1;
+				const float dpdu_a2 = dv2*dp1_a2 - dv1*dp2_a2;
+				const float dpdv_a1 = du1*dp2_a1 - du2*dp1_a1;
+				const float dpdv_a2 = du1*dp2_a2 - du2*dp1_a2;
+				float d = dpdu_a1*dpdv_a2 - dpdv_a1*dpdu_a2;
+				float uvd = du1*dv2 - dv1*du2;
+
+				if (uvd == 0.f) uvd = 1e-5f;
+				if (d == 0.f) d = 1e-5f;
+				d = uvd / d;
+
+				compat_bump->dudnu = (dpdv_a2*compat_bump->nu[a1] - dpdv_a1*compat_bump->nu[a2])*d;
+				compat_bump->dvdnu = (dpdu_a1*compat_bump->nu[a2] - dpdu_a2*compat_bump->nu[a1])*d;
+				compat_bump->dudnv = (dpdv_a2*compat_bump->nv[a1] - dpdv_a1*compat_bump->nv[a2])*d;
+				compat_bump->dvdnv = (dpdu_a1*compat_bump->nv[a2] - dpdu_a2*compat_bump->nv[a1])*d;
+			}
+		}
+	}
+}
+
+static int compatible_bump_compute(CompatibleBump *compat_bump, ShadeInput *shi, MTex *mtex, Tex *tex, TexResult *texres, float Tnor, float *co, float *dx, float *dy, float *texvec, float *dxt, float *dyt)
+{
+	TexResult ttexr = {0, 0, 0, 0, 0, texres->talpha, NULL};	// temp TexResult
+	float tco[3], texv[3], cd, ud, vd, du, dv, idu, idv;
+	const int fromrgb = ((tex->type == TEX_IMAGE) || ((tex->flag & TEX_COLORBAND)!=0));
+	const float bf = 0.04f*Tnor*mtex->norfac;
+	int rgbnor;
+	// disable internal bump eval
+	float* nvec = texres->nor;
+	texres->nor = NULL;
+	// du & dv estimates, constant value defaults
+	du = dv = 0.01f;
+
+	// compute ortho basis around normal
+	if(!compat_bump->nunvdone) {
+		// render normal is negated
+		negate_v3_v3(compat_bump->nn, shi->vn);
+		ortho_basis_v3v3_v3(compat_bump->nu, compat_bump->nv, compat_bump->nn);
+		compat_bump->nunvdone= 1;
+	}
+
+	// two methods, either constant based on main image resolution,
+	// (which also works without osa, though of course not always good (or even very bad) results),
+	// or based on tex derivative max values (osa only). Not sure which is best...
+
+	if (!shi->osatex && (tex->type == TEX_IMAGE) && tex->ima) {
+		// in case we have no proper derivatives, fall back to
+		// computing du/dv it based on image size
+		ImBuf* ibuf = BKE_image_get_ibuf(tex->ima, &tex->iuser);
+		if (ibuf) {
+			du = 1.f/(float)ibuf->x;
+			dv = 1.f/(float)ibuf->y;
+		}
+	}
+	else if (shi->osatex) {
+		// we have derivatives, can compute proper du/dv
+		if (tex->type == TEX_IMAGE) {	// 2d image, use u & v max. of dx/dy 2d vecs
+			const float adx[2] = {fabsf(dx[0]), fabsf(dx[1])};
+			const float ady[2] = {fabsf(dy[0]), fabsf(dy[1])};
+			du = MAX2(adx[0], ady[0]);
+			dv = MAX2(adx[1], ady[1]);
+		}
+		else {	// 3d procedural, estimate from all dx/dy elems
+			const float adx[3] = {fabsf(dx[0]), fabsf(dx[1]), fabsf(dx[2])};
+			const float ady[3] = {fabsf(dy[0]), fabsf(dy[1]), fabsf(dy[2])};
+			du = MAX3(adx[0], adx[1], adx[2]);
+			dv = MAX3(ady[1], ady[1], ady[2]);
+		}
+	}
+
+	// center, main return value
+	texco_mapping(shi, tex, mtex, co, dx, dy, texvec, dxt, dyt);
+	rgbnor = multitex_mtex(shi, mtex, texvec, dxt, dyt, texres);
+	cd = fromrgb ? (texres->tr + texres->tg + texres->tb)*0.33333333f : texres->tin;
+
+	if (mtex->texco == TEXCO_UV) {
+		// for the uv case, use the same value for both du/dv,
+		// since individually scaling the normal derivatives makes them useless...
+		du = MIN2(du, dv);
+		idu = (du < 1e-5f) ? bf : (bf/du);
+
+		// +u val
+		tco[0] = co[0] + compat_bump->dudnu*du;
+		tco[1] = co[1] + compat_bump->dvdnu*du;
+		tco[2] = 0.f;
+		texco_mapping(shi, tex, mtex, tco, dx, dy, texv, dxt, dyt);
+		multitex_mtex(shi, mtex, texv, dxt, dyt, &ttexr);
+		ud = idu*(cd - (fromrgb ? (ttexr.tr + ttexr.tg + ttexr.tb)*0.33333333f : ttexr.tin));
+
+		// +v val
+		tco[0] = co[0] + compat_bump->dudnv*du;
+		tco[1] = co[1] + compat_bump->dvdnv*du;
+		tco[2] = 0.f;
+		texco_mapping(shi, tex, mtex, tco, dx, dy, texv, dxt, dyt);
+		multitex_mtex(shi, mtex, texv, dxt, dyt, &ttexr);
+		vd = idu*(cd - (fromrgb ? (ttexr.tr + ttexr.tg + ttexr.tb)*0.33333333f : ttexr.tin));
+	}
+	else {
+		float tu[3], tv[3];
+
+		copy_v3_v3(tu, compat_bump->nu);
+		copy_v3_v3(tv, compat_bump->nv);
+
+		idu = (du < 1e-5f) ? bf : (bf/du);
+		idv = (dv < 1e-5f) ? bf : (bf/dv);
+
+		if ((mtex->texco == TEXCO_ORCO) && shi->obr && shi->obr->ob) {
+			mul_mat3_m4_v3(shi->obr->ob->imat, tu);
+			mul_mat3_m4_v3(shi->obr->ob->imat, tv);
+			normalize_v3(tu);
+			normalize_v3(tv);
+		}
+		else if (mtex->texco == TEXCO_GLOB) {
+			mul_mat3_m4_v3(R.viewinv, tu);
+			mul_mat3_m4_v3(R.viewinv, tv);
+		}
+		else if (mtex->texco == TEXCO_OBJECT && mtex->object) {
+			mul_mat3_m4_v3(mtex->object->imat, tu);
+			mul_mat3_m4_v3(mtex->object->imat, tv);
+			normalize_v3(tu);
+			normalize_v3(tv);
+		}
+
+		// +u val
+		tco[0] = co[0] + tu[0]*du;
+		tco[1] = co[1] + tu[1]*du;
+		tco[2] = co[2] + tu[2]*du;
+		texco_mapping(shi, tex, mtex, tco, dx, dy, texv, dxt, dyt);
+		multitex_mtex(shi, mtex, texv, dxt, dyt, &ttexr);
+		ud = idu*(cd - (fromrgb ? (ttexr.tr + ttexr.tg + ttexr.tb)*0.33333333f : ttexr.tin));
+
+		// +v val
+		tco[0] = co[0] + tv[0]*dv;
+		tco[1] = co[1] + tv[1]*dv;
+		tco[2] = co[2] + tv[2]*dv;
+		texco_mapping(shi, tex, mtex, tco, dx, dy, texv, dxt, dyt);
+		multitex_mtex(shi, mtex, texv, dxt, dyt, &ttexr);
+		vd = idv*(cd - (fromrgb ? (ttexr.tr + ttexr.tg + ttexr.tb)*0.33333333f : ttexr.tin));
+	}
+
+	// bumped normal
+	compat_bump->nu[0] += ud*compat_bump->nn[0];
+	compat_bump->nu[1] += ud*compat_bump->nn[1];
+	compat_bump->nu[2] += ud*compat_bump->nn[2];
+	compat_bump->nv[0] += vd*compat_bump->nn[0];
+	compat_bump->nv[1] += vd*compat_bump->nn[1];
+	compat_bump->nv[2] += vd*compat_bump->nn[2];
+	cross_v3_v3v3(nvec, compat_bump->nu, compat_bump->nv);
+
+	nvec[0] = -nvec[0];
+	nvec[1] = -nvec[1];
+	nvec[2] = -nvec[2];
+	texres->nor = nvec;
+
+	rgbnor |= TEX_NOR;
+	return rgbnor;
+}
+
+/* Improved bump code from later in 2.5 development cycle */
+
+typedef struct NTapBump {
+	int nunvdone;
+
+	// bumpmapping
+	float vNacc[3]; // original surface normal minus the surface gradient of every bump map which is encountered
+	float vR1[3], vR2[3]; // cross products (sigma_y, original_normal), (original_normal, sigma_x)
+	float sgn_det; // sign of the determinant of the matrix {sigma_x, sigma_y, original_normal}
+} NTapBump;
+
+static void ntap_bump_init(NTapBump *ntap_bump)
+{
+	memset(ntap_bump, 0, sizeof(*ntap_bump));
+}
+
+static int ntap_bump_compute(NTapBump *ntap_bump, ShadeInput *shi, MTex *mtex, Tex *tex, TexResult *texres, float Tnor, float *co, float *dx, float *dy, float *texvec, float *dxt, float *dyt)
+{
+	TexResult ttexr = {0, 0, 0, 0, 0, texres->talpha, NULL};	// temp TexResult
+
+	const int fromrgb = ((tex->type == TEX_IMAGE) || ((tex->flag & TEX_COLORBAND)!=0));
+	// TODO: solve this Hscale issue more elegantly.
+	float Hscale = 0.1f * Tnor*mtex->norfac; // factor 0.1 proved to look like the previous bump code
+	if( mtex->texflag & MTEX_BUMP_TEXTURESPACE )
+		Hscale *= 130.0f;
+
+	// 2 channels for 2D texture and 3 for 3D textures.
+	const int nr_channels = (mtex->texco == TEXCO_UV)? 2 : 3;
+	int c, rgbnor;
+	float dHdx, dHdy;
+
+	// disable internal bump eval in sampler, save pointer
+	float *nvec = texres->nor;
+	texres->nor = NULL;
+
+	if(!(mtex->texflag & MTEX_5TAP_BUMP)) {
+		// compute height derivatives with respect to output image pixel coordinates x and y
+		float STll[3], STlr[3], STul[3];
+		float Hll, Hlr, Hul;
+
+		texco_mapping(shi, tex, mtex, co, dx, dy, texvec, dxt, dyt);
+
+		for(c=0; c<nr_channels; c++) {
+			// dx contains the derivatives (du/dx, dv/dx)
+			// dy contains the derivatives (du/dy, dv/dy)
+			STll[c] = texvec[c];
+			STlr[c] = texvec[c]+dxt[c];
+			STul[c] = texvec[c]+dyt[c];
+		}
+
+		// clear unused derivatives
+		for(c=nr_channels; c<3; c++) {
+			STll[c] = 0.0f;
+			STlr[c] = 0.0f;
+			STul[c] = 0.0f;
+		}
+
+		// use texres for the center sample, set rgbnor
+		rgbnor = multitex_mtex(shi, mtex, STll, dxt, dyt, texres);
+		Hll = (fromrgb)? RGBTOBW(texres->tr, texres->tg, texres->tb) : texres->tin;
+
+		// use ttexr for the other 2 taps
+		multitex_mtex(shi, mtex, STlr, dxt, dyt, &ttexr);
+		Hlr = (fromrgb)? RGBTOBW(ttexr.tr, ttexr.tg, ttexr.tb) : ttexr.tin;
+
+		multitex_mtex(shi, mtex, STul, dxt, dyt, &ttexr);
+		Hul = (fromrgb)? RGBTOBW(ttexr.tr, ttexr.tg, ttexr.tb) : ttexr.tin;
+
+		dHdx = Hscale*(Hlr - Hll);
+		dHdy = Hscale*(Hul - Hll);
+	}
+	else {
+		/* same as above, but doing 5 taps, increasing quality at cost of speed */
+		float STc[3], STl[3], STr[3], STd[3], STu[3];
+		float Hc, Hl, Hr, Hd, Hu;
+
+		texco_mapping(shi, tex, mtex, co, dx, dy, texvec, dxt, dyt);
+
+		for(c=0; c<nr_channels; c++) {
+			STc[c] = texvec[c];
+			STl[c] = texvec[c] - 0.5f*dxt[c];
+			STr[c] = texvec[c] + 0.5f*dxt[c];
+			STd[c] = texvec[c] - 0.5f*dyt[c];
+			STu[c] = texvec[c] + 0.5f*dyt[c];
+		}
+
+		// clear unused derivatives
+		for(c=nr_channels; c<3; c++) {
+			STc[c] = 0.0f;
+			STl[c] = 0.0f;
+			STr[c] = 0.0f;
+			STd[c] = 0.0f;
+			STu[c] = 0.0f;
+		}
+
+		// use texres for the center sample, set rgbnor
+		rgbnor = multitex_mtex(shi, mtex, STc, dxt, dyt, texres);
+		Hc = (fromrgb)? RGBTOBW(texres->tr, texres->tg, texres->tb) : texres->tin;
+
+		// use ttexr for the other taps
+		multitex_mtex(shi, mtex, STl, dxt, dyt, &ttexr);
+		Hl = (fromrgb)? RGBTOBW(ttexr.tr, ttexr.tg, ttexr.tb) : ttexr.tin;
+		multitex_mtex(shi, mtex, STr, dxt, dyt, &ttexr);
+		Hr = (fromrgb)? RGBTOBW(ttexr.tr, ttexr.tg, ttexr.tb) : ttexr.tin;
+		multitex_mtex(shi, mtex, STd, dxt, dyt, &ttexr);
+		Hd = (fromrgb)? RGBTOBW(ttexr.tr, ttexr.tg, ttexr.tb) : ttexr.tin;
+		multitex_mtex(shi, mtex, STu, dxt, dyt, &ttexr);
+		Hu = (fromrgb)? RGBTOBW(ttexr.tr, ttexr.tg, ttexr.tb) : ttexr.tin;
+
+		dHdx = Hscale*(Hr - Hl);
+		dHdy = Hscale*(Hu - Hd);
+	}
+
+	// restore pointer
+	texres->nor = nvec;
+
+	/* replaced newbump with code based on listing 1 and 2 of
+		[Mik10] Mikkelsen M. S.: Bump Mapping Unparametrized Surfaces on the GPU.
+		-> http://jbit.net/~sparky/sfgrad_bump/mm_sfgrad_bump.pdf */
+
+	if(!ntap_bump->nunvdone) {
+		// initialize normal perturbation vectors
+		int xyz;
+		float fDet, abs_fDet;
+		// object2view and inverted matrix
+		float obj2view[3][3], view2obj[3][3], tmp[4][4];
+		// local copies of derivatives and normal
+		float dPdx[3], dPdy[3], vN[3];
+		VECCOPY(dPdx, shi->dxco);
+		VECCOPY(dPdy, shi->dyco);
+		VECCOPY(vN, shi->vn);
+		
+		if( mtex->texflag & MTEX_BUMP_OBJECTSPACE ) {
+			// TODO: these calculations happen for every pixel!
+			//	-> move to shi->obi
+			mul_m4_m4m4(tmp, shi->obr->ob->obmat, R.viewmat);
+			copy_m3_m4(obj2view, tmp); // use only upper left 3x3 matrix
+			invert_m3_m3(view2obj, obj2view);
+		
+			// generate the surface derivatives in object space
+			mul_m3_v3(view2obj, dPdx);
+			mul_m3_v3( view2obj, dPdy );
+			// generate the unit normal in object space
+			mul_transposed_m3_v3( obj2view, vN );
+			normalize_v3(vN);
+		}
+		
+		cross_v3_v3v3(ntap_bump->vR1, dPdy, vN);
+		cross_v3_v3v3(ntap_bump->vR2, vN, dPdx);
+		fDet = dot_v3v3(dPdx, ntap_bump->vR1);
+		ntap_bump->sgn_det = (fDet < 0)? -1.0f: 1.0f;
+		abs_fDet = ntap_bump->sgn_det * fDet;
+
+		if( mtex->texflag & MTEX_BUMP_TEXTURESPACE ) {
+			// crazy hack solution that gives results similar to normal mapping - part 1
+			normalize_v3(ntap_bump->vR1);
+			normalize_v3(ntap_bump->vR2);
+			abs_fDet = 1.0f;
+		}
+		
+		for(xyz=0; xyz<3; xyz++)
+				ntap_bump->vNacc[xyz] = abs_fDet * vN[xyz];
+	
+		if( mtex->texflag & MTEX_BUMP_OBJECTSPACE ) {
+			// pre do transform of texres->nor by the inverse transposed of obj2view
+			mul_transposed_m3_v3( view2obj, ntap_bump->vNacc );
+			mul_transposed_m3_v3( view2obj, ntap_bump->vR1 );
+			mul_transposed_m3_v3( view2obj, ntap_bump->vR2 );
+		}
+
+		ntap_bump->nunvdone= 1;
+	}
+
+	if( mtex->texflag & MTEX_BUMP_TEXTURESPACE ) {
+		// crazy hack solution that gives results similar to normal mapping - part 2
+		float vec[2];
+		vec[0] = tex->ima->gen_x*dxt[0];
+		vec[1] = tex->ima->gen_y*dxt[1];
+		dHdx *= 1.0f/len_v2(vec);
+		vec[0] = tex->ima->gen_x*dyt[0];
+		vec[1] = tex->ima->gen_y*dyt[1];
+		dHdy *= 1.0f/len_v2(vec);
+	}
+	
+	// subtract the surface gradient from vNacc
+	for(c=0; c<3; c++) {
+		float vSurfGrad_compi = ntap_bump->sgn_det * (dHdx * ntap_bump->vR1[c] + dHdy * ntap_bump->vR2[c]);
+		ntap_bump->vNacc[c] -= vSurfGrad_compi;
+		texres->nor[c] = ntap_bump->vNacc[c]; // copy
+	}
+
+	rgbnor |= TEX_NOR;
+	return rgbnor;
+}
+
 void do_material_tex(ShadeInput *shi)
 {
 	MTex *mtex;
@@ -1676,8 +2082,13 @@ void do_material_tex(ShadeInput *shi)
 	float fact, facm, factt, facmm, stencilTin=1.0;
 	float texvec[3], dxt[3], dyt[3], tempvec[3], norvec[3], warpvec[3]={0.0f, 0.0f, 0.0f}, Tnor=1.0;
 	int tex_nr, rgbnor= 0, warpdone=0;
-	float nu[3] = {0,0,0}, nv[3] = {0,0,0}, nn[3] = {0,0,0}, dudnu = 1.f, dudnv = 0.f, dvdnu = 0.f, dvdnv = 1.f; // bump mapping
-	int nunvdone= 0, newbump;
+
+	CompatibleBump compat_bump;
+	NTapBump ntap_bump;
+	int use_compat_bump, use_ntap_bump;
+
+	compatible_bump_init(&compat_bump);
+	ntap_bump_init(&ntap_bump);
 
 	if (R.r.scemode & R_NO_TEX) return;
 	/* here: test flag if there's a tex (todo) */
@@ -1693,9 +2104,15 @@ void do_material_tex(ShadeInput *shi)
 			tex= mtex->tex;
 			if(tex==0) continue;
 
+			use_compat_bump= (mtex->texflag & MTEX_COMPAT_BUMP);
+			use_ntap_bump= (mtex->texflag & (MTEX_3TAP_BUMP|MTEX_5TAP_BUMP));
+
 			/* XXX texture node trees don't work for this yet */
-			newbump= (mtex->texflag & MTEX_NEW_BUMP) && !(tex->nodetree && tex->use_nodes);
-			
+			if(tex->nodetree && tex->use_nodes) {
+				use_compat_bump = 0;
+				use_ntap_bump = 0;
+			}
+
 			/* which coords */
 			if(mtex->texco==TEXCO_ORCO) {
 				if(mtex->texflag & MTEX_DUPLI_MAPTO) {
@@ -1769,57 +2186,7 @@ void do_material_tex(ShadeInput *shi)
 					dx= suv->dxuv;
 					dy= suv->dyuv; 
 
-					// uvmapping only, calculation of normal tangent u/v partial derivatives
-					// (should not be here, dudnu, dudnv, dvdnu & dvdnv should probably be part of ShadeInputUV struct,
-					//  nu/nv in ShadeInput and this calculation should then move to shadeinput.c, shade_input_set_shade_texco() func.)
-					// NOTE: test for shi->obr->ob here, since vlr/obr/obi can be 'fake' when called from fastshade(), another reason to move it..
-					// NOTE: shi->v1 is NULL when called from displace_render_vert, assigning verts in this case is not trivial because the shi quad face side is not know.
-					if ((mtex->texflag & MTEX_NEW_BUMP) && shi->obr && shi->obr->ob && shi->v1) {
-						if(mtex->mapto & (MAP_NORM|MAP_WARP) && !((tex->type==TEX_IMAGE) && (tex->imaflag & TEX_NORMALMAP))) {
-							MTFace* tf = RE_vlakren_get_tface(shi->obr, shi->vlr, i, NULL, 0);
-							int j1 = shi->i1, j2 = shi->i2, j3 = shi->i3;
-
-							vlr_set_uv_indices(shi->vlr, &j1, &j2, &j3);
-
-							// compute ortho basis around normal
-							if(!nunvdone) {
-								// render normal is negated
-								nn[0] = -shi->vn[0];
-								nn[1] = -shi->vn[1];
-								nn[2] = -shi->vn[2];
-								ortho_basis_v3v3_v3( nu, nv,nn);
-								nunvdone= 1;
-							}
-
-							if (tf) {
-								float *uv1 = tf->uv[j1], *uv2 = tf->uv[j2], *uv3 = tf->uv[j3];
-								const float an[3] = {fabsf(nn[0]), fabsf(nn[1]), fabsf(nn[2])};
-								const int a1 = (an[0] > an[1] && an[0] > an[2]) ? 1 : 0;
-								const int a2 = (an[2] > an[0] && an[2] > an[1]) ? 1 : 2;
-								const float dp1_a1 = shi->v1->co[a1] - shi->v3->co[a1];
-								const float dp1_a2 = shi->v1->co[a2] - shi->v3->co[a2];
-								const float dp2_a1 = shi->v2->co[a1] - shi->v3->co[a1];
-								const float dp2_a2 = shi->v2->co[a2] - shi->v3->co[a2];
-								const float du1 = uv1[0] - uv3[0], du2 = uv2[0] - uv3[0];
-								const float dv1 = uv1[1] - uv3[1], dv2 = uv2[1] - uv3[1];
-								const float dpdu_a1 = dv2*dp1_a1 - dv1*dp2_a1;
-								const float dpdu_a2 = dv2*dp1_a2 - dv1*dp2_a2;
-								const float dpdv_a1 = du1*dp2_a1 - du2*dp1_a1;
-								const float dpdv_a2 = du1*dp2_a2 - du2*dp1_a2;
-								float d = dpdu_a1*dpdv_a2 - dpdv_a1*dpdu_a2;
-								float uvd = du1*dv2 - dv1*du2;
-
-								if (uvd == 0.f) uvd = 1e-5f;
-								if (d == 0.f) d = 1e-5f;
-								d = uvd / d;
-
-								dudnu = (dpdv_a2*nu[a1] - dpdv_a1*nu[a2])*d;
-								dvdnu = (dpdu_a1*nu[a2] - dpdu_a2*nu[a1])*d;
-								dudnv = (dpdv_a2*nv[a1] - dpdv_a1*nv[a2])*d;
-								dvdnv = (dpdu_a1*nv[a2] - dpdu_a2*nv[a1])*d;
-							}
-						}
-					}
+					compatible_bump_uv_derivs(&compat_bump, shi, mtex, i);
 				}
 			}
 			else if(mtex->texco==TEXCO_WINDOW) {
@@ -1858,138 +2225,14 @@ void do_material_tex(ShadeInput *shi)
 			}
 
 			/* XXX texture node trees don't work for this yet */
-			if(newbump) {
-				// compute ortho basis around normal
-				if(!nunvdone) {
-					// render normal is negated
-					nn[0] = -shi->vn[0];
-					nn[1] = -shi->vn[1];
-					nn[2] = -shi->vn[2];
-					ortho_basis_v3v3_v3( nu, nv,nn);
-					nunvdone= 1;
-				}
-
-				if(texres.nor && !((tex->type==TEX_IMAGE) && (tex->imaflag & TEX_NORMALMAP))) {
-					TexResult ttexr = {0, 0, 0, 0, 0, texres.talpha, NULL};	// temp TexResult
-					float tco[3], texv[3], cd, ud, vd, du, dv, idu, idv;
-					const int fromrgb = ((tex->type == TEX_IMAGE) || ((tex->flag & TEX_COLORBAND)!=0));
-					const float bf = 0.04f*Tnor*stencilTin*mtex->norfac;
-					// disable internal bump eval
-					float* nvec = texres.nor;
-					texres.nor = NULL;
-					// du & dv estimates, constant value defaults
-					du = dv = 0.01f;
-
-					// two methods, either constant based on main image resolution,
-					// (which also works without osa, though of course not always good (or even very bad) results),
-					// or based on tex derivative max values (osa only). Not sure which is best...
-
-					if (!shi->osatex && (tex->type == TEX_IMAGE) && tex->ima) {
-						// in case we have no proper derivatives, fall back to
-						// computing du/dv it based on image size
-						ImBuf* ibuf = BKE_image_get_ibuf(tex->ima, &tex->iuser);
-						if (ibuf) {
-							du = 1.f/(float)ibuf->x;
-							dv = 1.f/(float)ibuf->y;
-						}
-					}
-					else if (shi->osatex) {
-						// we have derivatives, can compute proper du/dv
-						if (tex->type == TEX_IMAGE) {	// 2d image, use u & v max. of dx/dy 2d vecs
-							const float adx[2] = {fabsf(dx[0]), fabsf(dx[1])};
-							const float ady[2] = {fabsf(dy[0]), fabsf(dy[1])};
-							du = MAX2(adx[0], ady[0]);
-							dv = MAX2(adx[1], ady[1]);
-						}
-						else {	// 3d procedural, estimate from all dx/dy elems
-							const float adx[3] = {fabsf(dx[0]), fabsf(dx[1]), fabsf(dx[2])};
-							const float ady[3] = {fabsf(dy[0]), fabsf(dy[1]), fabsf(dy[2])};
-							du = MAX3(adx[0], adx[1], adx[2]);
-							dv = MAX3(ady[1], ady[1], ady[2]);
-						}
-					}
-
-					// center, main return value
-					texco_mapping(shi, tex, mtex, co, dx, dy, texvec, dxt, dyt);
-					rgbnor = multitex_mtex(shi, mtex, texvec, dxt, dyt, &texres);
-					cd = fromrgb ? (texres.tr + texres.tg + texres.tb)*0.33333333f : texres.tin;
-
-					if (mtex->texco == TEXCO_UV) {
-						// for the uv case, use the same value for both du/dv,
-						// since individually scaling the normal derivatives makes them useless...
-						du = MIN2(du, dv);
-						idu = (du < 1e-5f) ? bf : (bf/du);
-
-						// +u val
-						tco[0] = co[0] + dudnu*du;
-						tco[1] = co[1] + dvdnu*du;
-						tco[2] = 0.f;
-						texco_mapping(shi, tex, mtex, tco, dx, dy, texv, dxt, dyt);
-						multitex_mtex(shi, mtex, texv, dxt, dyt, &ttexr);
-						ud = idu*(cd - (fromrgb ? (ttexr.tr + ttexr.tg + ttexr.tb)*0.33333333f : ttexr.tin));
-
-						// +v val
-						tco[0] = co[0] + dudnv*du;
-						tco[1] = co[1] + dvdnv*du;
-						tco[2] = 0.f;
-						texco_mapping(shi, tex, mtex, tco, dx, dy, texv, dxt, dyt);
-						multitex_mtex(shi, mtex, texv, dxt, dyt, &ttexr);
-						vd = idu*(cd - (fromrgb ? (ttexr.tr + ttexr.tg + ttexr.tb)*0.33333333f : ttexr.tin));
-					}
-					else {
-						float tu[3] = {nu[0], nu[1], nu[2]}, tv[3] = {nv[0], nv[1], nv[2]};
-
-						idu = (du < 1e-5f) ? bf : (bf/du);
-						idv = (dv < 1e-5f) ? bf : (bf/dv);
-
-						if ((mtex->texco == TEXCO_ORCO) && shi->obr && shi->obr->ob) {
-							mul_mat3_m4_v3(shi->obr->ob->imat, tu);
-							mul_mat3_m4_v3(shi->obr->ob->imat, tv);
-							normalize_v3(tu);
-							normalize_v3(tv);
-						}
-						else if (mtex->texco == TEXCO_GLOB) {
-							mul_mat3_m4_v3(R.viewinv, tu);
-							mul_mat3_m4_v3(R.viewinv, tv);
-						}
-						else if (mtex->texco == TEXCO_OBJECT && mtex->object) {
-							mul_mat3_m4_v3(mtex->object->imat, tu);
-							mul_mat3_m4_v3(mtex->object->imat, tv);
-							normalize_v3(tu);
-							normalize_v3(tv);
-						}
-
-						// +u val
-						tco[0] = co[0] + tu[0]*du;
-						tco[1] = co[1] + tu[1]*du;
-						tco[2] = co[2] + tu[2]*du;
-						texco_mapping(shi, tex, mtex, tco, dx, dy, texv, dxt, dyt);
-						multitex_mtex(shi, mtex, texv, dxt, dyt, &ttexr);
-						ud = idu*(cd - (fromrgb ? (ttexr.tr + ttexr.tg + ttexr.tb)*0.33333333f : ttexr.tin));
-
-						// +v val
-						tco[0] = co[0] + tv[0]*dv;
-						tco[1] = co[1] + tv[1]*dv;
-						tco[2] = co[2] + tv[2]*dv;
-						texco_mapping(shi, tex, mtex, tco, dx, dy, texv, dxt, dyt);
-						multitex_mtex(shi, mtex, texv, dxt, dyt, &ttexr);
-						vd = idv*(cd - (fromrgb ? (ttexr.tr + ttexr.tg + ttexr.tb)*0.33333333f : ttexr.tin));
-					}
-
-					// bumped normal
-					nu[0] += ud*nn[0];
-					nu[1] += ud*nn[1];
-					nu[2] += ud*nn[2];
-					nv[0] += vd*nn[0];
-					nv[1] += vd*nn[1];
-					nv[2] += vd*nn[2];
-					cross_v3_v3v3(nvec, nu, nv);
-
-					nvec[0] = -nvec[0];
-					nvec[1] = -nvec[1];
-					nvec[2] = -nvec[2];
-					texres.nor = nvec;
-					rgbnor |= TEX_NOR;
+			if(texres.nor && !((tex->type==TEX_IMAGE) && (tex->imaflag & TEX_NORMALMAP))) {
+				if(use_compat_bump) {
+					rgbnor = compatible_bump_compute(&compat_bump, shi, mtex, tex,
+						&texres, Tnor*stencilTin, co, dx, dy, texvec, dxt, dyt);
+ 				}
+				else if(use_ntap_bump) {
+					rgbnor = ntap_bump_compute(&ntap_bump, shi, mtex, tex,
+						&texres, Tnor*stencilTin, co, dx, dy, texvec, dxt, dyt);
 				}
 				else {
 					texco_mapping(shi, tex, mtex, co, dx, dy, texvec, dxt, dyt);
@@ -2184,7 +2427,7 @@ void do_material_tex(ShadeInput *shi)
 					}
 					else {
 						/* XXX texture node trees don't work for this yet */
-						if (newbump) {
+						if (use_compat_bump || use_ntap_bump) {
 							shi->vn[0] = texres.nor[0];
 							shi->vn[1] = texres.nor[1];
 							shi->vn[2] = texres.nor[2];