Fix cycles CUDA sm 1.3 build with 32 bit compiler, tweaked voronoi

and brick code so that it can be uninlined.

build_files/buildbot/config/user-config-glibc211-i686.py

@@ -110,8 +110,7 @@ BF_JACK_LIB_STATIC = '${BF_ZLIB}/lib/libjack.a'
 # Cycles
 WITH_BF_CYCLES = True
 WITH_BF_CYCLES_CUDA_BINARIES = True
-#BF_CYCLES_CUDA_BINARIES_ARCH = ['sm_13', 'sm_20', 'sm_21', 'sm_30']
+BF_CYCLES_CUDA_BINARIES_ARCH = ['sm_13', 'sm_20', 'sm_21', 'sm_30']
-BF_CYCLES_CUDA_BINARIES_ARCH = ['sm_20', 'sm_21', 'sm_30']
 
 WITH_BF_OIIO = True
 WITH_BF_STATICOIIO = True

build_files/buildbot/config/user-config-glibc27-i686.py

@@ -97,8 +97,7 @@ WITH_BF_JACK = True
 # Cycles
 WITH_BF_CYCLES = True
 WITH_BF_CYCLES_CUDA_BINARIES = True
-#BF_CYCLES_CUDA_BINARIES_ARCH = ['sm_13', 'sm_20', 'sm_21', 'sm_30']
+BF_CYCLES_CUDA_BINARIES_ARCH = ['sm_13', 'sm_20', 'sm_21', 'sm_30']
-BF_CYCLES_CUDA_BINARIES_ARCH = ['sm_20', 'sm_21', 'sm_30']
 
 WITH_BF_OIIO = True
 WITH_BF_STATICOIIO = True

intern/cycles/kernel/kernel_types.h

@@ -102,6 +102,7 @@ CCL_NAMESPACE_BEGIN
 #define __IMAGE_TEXTURES__
 #define __EXTRA_NODES__
 #define __HOLDOUT__
+#define __NORMAL_MAP__
 #endif
 
 #ifdef __KERNEL_ADV_SHADING__

intern/cycles/kernel/svm/svm.h

@@ -401,9 +401,13 @@ __device_noinline void svm_eval_nodes(KernelGlobals *kg, ShaderData *sd, ShaderT
 			case NODE_LIGHT_FALLOFF:
 				svm_node_light_falloff(sd, stack, node);
 				break;
+#endif			
+#ifdef __ANISOTROPIC__
 			case NODE_TANGENT:
 				svm_node_tangent(kg, sd, stack, node);
 				break;
+#endif			
+#ifdef __NORMAL_MAP__
 			case NODE_NORMAL_MAP:
 				svm_node_normal_map(kg, sd, stack, node);
 				break;

intern/cycles/kernel/svm/svm_brick.h

@@ -28,9 +28,9 @@ __device_noinline float brick_noise(int n) /* fast integer noise */
 	return 0.5f * ((float)nn / 1073741824.0f);
 }
 
-__device_noinline float svm_brick(float3 p, float scale, float mortar_size, float bias,
+__device_noinline float2 svm_brick(float3 p, float scale, float mortar_size, float bias,
 	float brick_width, float row_height, float offset_amount, int offset_frequency,
-	float squash_amount, int squash_frequency, float *tint)
+	float squash_amount, int squash_frequency)
 {	
 	p *= scale;
 
@@ -50,11 +50,12 @@ __device_noinline float svm_brick(float3 p, float scale, float mortar_size, floa
 	x = (p.x+offset) - brick_width*bricknum;
 	y = p.y - row_height*rownum;
 
-	*tint = clamp((brick_noise((rownum << 16) + (bricknum & 0xFFFF)) + bias), 0.0f, 1.0f);
+	return make_float2(
+		clamp((brick_noise((rownum << 16) + (bricknum & 0xFFFF)) + bias), 0.0f, 1.0f),
 
-	return (x < mortar_size || y < mortar_size ||
+		(x < mortar_size || y < mortar_size ||
 		x > (brick_width - mortar_size) ||
-		y > (row_height - mortar_size)) ? 1.0f : 0.0f;
+		y > (row_height - mortar_size)) ? 1.0f : 0.0f);
 }
 
 __device void svm_node_tex_brick(KernelGlobals *kg, ShaderData *sd, float *stack, uint4 node, int *offset)
@@ -70,8 +71,6 @@ __device void svm_node_tex_brick(KernelGlobals *kg, ShaderData *sd, float *stack
 	/* RNA properties */
 	uint offset_frequency, squash_frequency;
 	
-	float tint = 0.0f;
-
 	decode_node_uchar4(node.y, &co_offset, &color1_offset, &color2_offset, &mortar_offset);
 	decode_node_uchar4(node.z, &scale_offset, &mortar_size_offset, &bias_offset, &brick_width_offset);
 	decode_node_uchar4(node.w, &row_height_offset, &color_offset, &fac_offset, NULL);
@@ -92,9 +91,11 @@ __device void svm_node_tex_brick(KernelGlobals *kg, ShaderData *sd, float *stack
 	float offset_amount = __int_as_float(node3.z);
 	float squash_amount = __int_as_float(node3.w);
 	
-	float f = svm_brick(co, scale, mortar_size, bias, brick_width, row_height,
+	float2 f2 = svm_brick(co, scale, mortar_size, bias, brick_width, row_height,
-		offset_amount, offset_frequency, squash_amount, squash_frequency,
+		offset_amount, offset_frequency, squash_amount, squash_frequency);
-		&tint);
+
+	float tint = f2.x;
+	float f = f2.y;
 	
 	if(f != 1.0f) {
 		float facm = 1.0f - tint;

intern/cycles/kernel/svm/svm_texture.h

@@ -42,8 +42,12 @@ __device float voronoi_distance(NodeDistanceMetric distance_metric, float3 d, fl
 
 /* Voronoi / Worley like */
 
-__device_noinline void voronoi(float3 p, NodeDistanceMetric distance_metric, float e, float da[4], float3 pa[4])
+__device_noinline float4 voronoi_Fn(float3 p, float e, int n1, int n2)
 {
+	float da[4];
+	float3 pa[4];
+	NodeDistanceMetric distance_metric = NODE_VORONOI_DISTANCE_SQUARED;
+
 	/* returns distances in da and point coords in pa */
 	int xx, yy, zz, xi, yi, zi;
 
@@ -105,33 +109,20 @@ __device_noinline void voronoi(float3 p, NodeDistanceMetric distance_metric, flo
 			}
 		}
 	}
+
+	float4 result = make_float4(pa[n1].x, pa[n1].y, pa[n1].z, da[n1]);
+
+	if(n2 != -1)
+		result = make_float4(pa[n2].x, pa[n2].y, pa[n2].z, da[n2]) - result;
+
+	return result;
 }
 
-__device float voronoi_Fn(float3 p, int n)
+__device float voronoi_F1(float3 p) { return voronoi_Fn(p, 0.0f, 0, -1).w; }
-{
+__device float voronoi_F2(float3 p) { return voronoi_Fn(p, 0.0f, 1, -1).w; }
-	float da[4];
+__device float voronoi_F3(float3 p) { return voronoi_Fn(p, 0.0f, 2, -1).w; }
-	float3 pa[4];
+__device float voronoi_F4(float3 p) { return voronoi_Fn(p, 0.0f, 3, -1).w; }
-
+__device float voronoi_F1F2(float3 p) { return voronoi_Fn(p, 0.0f, 0, 1).w; }
-	voronoi(p, NODE_VORONOI_DISTANCE_SQUARED, 0, da, pa);
-
-	return da[n];
-}
-
-__device float voronoi_FnFn(float3 p, int n1, int n2)
-{
-	float da[4];
-	float3 pa[4];
-
-	voronoi(p, NODE_VORONOI_DISTANCE_SQUARED, 0, da, pa);
-
-	return da[n2] - da[n1];
-}
-
-__device float voronoi_F1(float3 p) { return voronoi_Fn(p, 0); }
-__device float voronoi_F2(float3 p) { return voronoi_Fn(p, 1); }
-__device float voronoi_F3(float3 p) { return voronoi_Fn(p, 2); }
-__device float voronoi_F4(float3 p) { return voronoi_Fn(p, 3); }
-__device float voronoi_F1F2(float3 p) { return voronoi_FnFn(p, 0, 1); }
 
 __device float voronoi_Cr(float3 p)
 {

intern/cycles/kernel/svm/svm_voronoi.h

@@ -23,21 +23,18 @@ CCL_NAMESPACE_BEGIN
 __device_noinline float4 svm_voronoi(NodeVoronoiColoring coloring, float scale, float3 p)
 {
 	/* compute distance and point coordinate of 4 nearest neighbours */
-	float da[4];
+	float4 dpa0 = voronoi_Fn(p*scale, 1.0f, 0, -1);
-	float3 pa[4];
-
-	voronoi(p*scale, NODE_VORONOI_DISTANCE_SQUARED, 1.0f, da, pa);
 
 	/* output */
 	float fac;
 	float3 color;
 
 	if(coloring == NODE_VORONOI_INTENSITY) {
-		fac = fabsf(da[0]);
+		fac = fabsf(dpa0.w);
 		color = make_float3(fac, fac, fac);
 	}
 	else {
-		color = cellnoise_color(pa[0]);
+		color = cellnoise_color(float4_to_float3(dpa0));
 		fac = average(color);
 	}