Cycles: OpenCL kernel split

This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.

Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.

Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.

This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.

More feature will be enabled once they're ported to the split kernel and
tested.

Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.

Based on the research paper:

  https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf

Here's the documentation:

  https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit

Design discussion of the patch:

  https://developer.blender.org/T44197

Differential Revision: https://developer.blender.org/D1200

1 files changed, 21 insertions, 21 deletions

diff --git a/intern/cycles/kernel/svm/svm_geometry.h b/intern/cycles/kernel/svm/svm_geometry.h
index efbefa77d28..bb06254c3a9 100644
--- a/intern/cycles/kernel/svm/svm_geometry.h
+++ b/intern/cycles/kernel/svm/svm_geometry.h
@@ -23,15 +23,15 @@ ccl_device void svm_node_geometry(KernelGlobals *kg, ShaderData *sd, float *stac
 	float3 data;
 
 	switch(type) {
-		case NODE_GEOM_P: data = sd->P; break;
-		case NODE_GEOM_N: data = sd->N; break;
+		case NODE_GEOM_P: data = ccl_fetch(sd, P); break;
+		case NODE_GEOM_N: data = ccl_fetch(sd, N); break;
 #ifdef __DPDU__
 		case NODE_GEOM_T: data = primitive_tangent(kg, sd); break;
 #endif
-		case NODE_GEOM_I: data = sd->I; break;
-		case NODE_GEOM_Ng: data = sd->Ng; break;
+		case NODE_GEOM_I: data = ccl_fetch(sd, I); break;
+		case NODE_GEOM_Ng: data = ccl_fetch(sd, Ng); break;
 #ifdef __UV__
-		case NODE_GEOM_uv: data = make_float3(sd->u, sd->v, 0.0f); break;
+		case NODE_GEOM_uv: data = make_float3(ccl_fetch(sd, u), ccl_fetch(sd, v), 0.0f); break;
 #endif
 	}
 
@@ -44,8 +44,8 @@ ccl_device void svm_node_geometry_bump_dx(KernelGlobals *kg, ShaderData *sd, flo
 	float3 data;
 
 	switch(type) {
-		case NODE_GEOM_P: data = sd->P + sd->dP.dx; break;
-		case NODE_GEOM_uv: data = make_float3(sd->u + sd->du.dx, sd->v + sd->dv.dx, 0.0f); break;
+		case NODE_GEOM_P: data = ccl_fetch(sd, P) + ccl_fetch(sd, dP).dx; break;
+		case NODE_GEOM_uv: data = make_float3(ccl_fetch(sd, u) + ccl_fetch(sd, du).dx, ccl_fetch(sd, v) + ccl_fetch(sd, dv).dx, 0.0f); break;
 		default: svm_node_geometry(kg, sd, stack, type, out_offset); return;
 	}
 
@@ -61,8 +61,8 @@ ccl_device void svm_node_geometry_bump_dy(KernelGlobals *kg, ShaderData *sd, flo
 	float3 data;
 
 	switch(type) {
-		case NODE_GEOM_P: data = sd->P + sd->dP.dy; break;
-		case NODE_GEOM_uv: data = make_float3(sd->u + sd->du.dy, sd->v + sd->dv.dy, 0.0f); break;
+		case NODE_GEOM_P: data = ccl_fetch(sd, P) + ccl_fetch(sd, dP).dy; break;
+		case NODE_GEOM_uv: data = make_float3(ccl_fetch(sd, u) + ccl_fetch(sd, du).dy, ccl_fetch(sd, v) + ccl_fetch(sd, dv).dy, 0.0f); break;
 		default: svm_node_geometry(kg, sd, stack, type, out_offset); return;
 	}
 
@@ -83,9 +83,9 @@ ccl_device void svm_node_object_info(KernelGlobals *kg, ShaderData *sd, float *s
 			stack_store_float3(stack, out_offset, object_location(kg, sd));
 			return;
 		}
-		case NODE_INFO_OB_INDEX: data = object_pass_id(kg, sd->object); break;
+		case NODE_INFO_OB_INDEX: data = object_pass_id(kg, ccl_fetch(sd, object)); break;
 		case NODE_INFO_MAT_INDEX: data = shader_pass_id(kg, sd); break;
-		case NODE_INFO_OB_RANDOM: data = object_random_number(kg, sd->object); break;
+		case NODE_INFO_OB_RANDOM: data = object_random_number(kg, ccl_fetch(sd, object)); break;
 		default: data = 0.0f; break;
 	}
 
@@ -98,44 +98,44 @@ ccl_device void svm_node_particle_info(KernelGlobals *kg, ShaderData *sd, float
 {
 	switch(type) {
 		case NODE_INFO_PAR_INDEX: {
-			int particle_id = object_particle_id(kg, sd->object);
+			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
 			stack_store_float(stack, out_offset, particle_index(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_AGE: {
-			int particle_id = object_particle_id(kg, sd->object);
+			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
 			stack_store_float(stack, out_offset, particle_age(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_LIFETIME: {
-			int particle_id = object_particle_id(kg, sd->object);
+			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
 			stack_store_float(stack, out_offset, particle_lifetime(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_LOCATION: {
-			int particle_id = object_particle_id(kg, sd->object);
+			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
 			stack_store_float3(stack, out_offset, particle_location(kg, particle_id));
 			break;
 		}
 #if 0	/* XXX float4 currently not supported in SVM stack */
 		case NODE_INFO_PAR_ROTATION: {
-			int particle_id = object_particle_id(kg, sd->object);
+			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
 			stack_store_float4(stack, out_offset, particle_rotation(kg, particle_id));
 			break;
 		}
 #endif
 		case NODE_INFO_PAR_SIZE: {
-			int particle_id = object_particle_id(kg, sd->object);
+			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
 			stack_store_float(stack, out_offset, particle_size(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_VELOCITY: {
-			int particle_id = object_particle_id(kg, sd->object);
+			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
 			stack_store_float3(stack, out_offset, particle_velocity(kg, particle_id));
 			break;
 		}
 		case NODE_INFO_PAR_ANGULAR_VELOCITY: {
-			int particle_id = object_particle_id(kg, sd->object);
+			int particle_id = object_particle_id(kg, ccl_fetch(sd, object));
 			stack_store_float3(stack, out_offset, particle_angular_velocity(kg, particle_id));
 			break;
 		}
@@ -153,7 +153,7 @@ ccl_device void svm_node_hair_info(KernelGlobals *kg, ShaderData *sd, float *sta
 
 	switch(type) {
 		case NODE_INFO_CURVE_IS_STRAND: {
-			data = (sd->type & PRIMITIVE_ALL_CURVE) != 0;
+			data = (ccl_fetch(sd, type) & PRIMITIVE_ALL_CURVE) != 0;
 			stack_store_float(stack, out_offset, data);
 			break;
 		}
@@ -165,7 +165,7 @@ ccl_device void svm_node_hair_info(KernelGlobals *kg, ShaderData *sd, float *sta
 			break;
 		}
 		/*case NODE_INFO_CURVE_FADE: {
-			data = sd->curve_transparency;
+			data = ccl_fetch(sd, curve_transparency);
 			stack_store_float(stack, out_offset, data);
 			break;
 		}*/