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, 63 insertions, 22 deletions

diff --git a/intern/cycles/kernel/kernel_emission.h b/intern/cycles/kernel/kernel_emission.h
index 6c5a5fac8c5..1fee5835360 100644
--- a/intern/cycles/kernel/kernel_emission.h
+++ b/intern/cycles/kernel/kernel_emission.h
@@ -17,12 +17,20 @@
 CCL_NAMESPACE_BEGIN
 
 /* Direction Emission */
-
 ccl_device_noinline float3 direct_emissive_eval(KernelGlobals *kg,
-	LightSample *ls, float3 I, differential3 dI, float t, float time, int bounce, int transparent_bounce)
+	LightSample *ls, float3 I, differential3 dI, float t, float time, int bounce, int transparent_bounce
+#ifdef __SPLIT_KERNEL__
+	,ShaderData *sd_input
+#endif
+)
 {
 	/* setup shading at emitter */
-	ShaderData sd;
+#ifdef __SPLIT_KERNEL__
+	ShaderData *sd = sd_input;
+#else
+	ShaderData sd_object;
+	ShaderData *sd = &sd_object;
+#endif
 	float3 eval;
 
 #ifdef __BACKGROUND_MIS__
@@ -37,23 +45,23 @@ ccl_device_noinline float3 direct_emissive_eval(KernelGlobals *kg,
 		ray.dP = differential3_zero();
 		ray.dD = dI;
 
-		shader_setup_from_background(kg, &sd, &ray, bounce+1, transparent_bounce);
-		eval = shader_eval_background(kg, &sd, 0, SHADER_CONTEXT_EMISSION);
+		shader_setup_from_background(kg, sd, &ray, bounce+1, transparent_bounce);
+		eval = shader_eval_background(kg, sd, 0, SHADER_CONTEXT_EMISSION);
 	}
 	else
 #endif
 	{
-		shader_setup_from_sample(kg, &sd, ls->P, ls->Ng, I, ls->shader, ls->object, ls->prim, ls->u, ls->v, t, time, bounce+1, transparent_bounce);
+		shader_setup_from_sample(kg, sd, ls->P, ls->Ng, I, ls->shader, ls->object, ls->prim, ls->u, ls->v, t, time, bounce+1, transparent_bounce);
 
-		ls->Ng = sd.Ng;
+		ls->Ng = ccl_fetch(sd, Ng);
 
 		/* no path flag, we're evaluating this for all closures. that's weak but
 		 * we'd have to do multiple evaluations otherwise */
-		shader_eval_surface(kg, &sd, 0.0f, 0, SHADER_CONTEXT_EMISSION);
+		shader_eval_surface(kg, sd, 0.0f, 0, SHADER_CONTEXT_EMISSION);
 
 		/* evaluate emissive closure */
-		if(sd.flag & SD_EMISSION)
-			eval = shader_emissive_eval(kg, &sd);
+		if(ccl_fetch(sd, flag) & SD_EMISSION)
+			eval = shader_emissive_eval(kg, sd);
 		else
 			eval = make_float3(0.0f, 0.0f, 0.0f);
 	}
@@ -63,9 +71,14 @@ ccl_device_noinline float3 direct_emissive_eval(KernelGlobals *kg,
 	return eval;
 }
 
+/* The argument sd_DL is meaningful only for split kernel. Other uses can just pass NULL */
 ccl_device_noinline bool direct_emission(KernelGlobals *kg, ShaderData *sd,
 	LightSample *ls, Ray *ray, BsdfEval *eval, bool *is_lamp,
-	int bounce, int transparent_bounce)
+	int bounce, int transparent_bounce
+#ifdef __SPLIT_KERNEL__
+	, ShaderData *sd_DL
+#endif
+	)
 {
 	if(ls->pdf == 0.0f)
 		return false;
@@ -74,7 +87,14 @@ ccl_device_noinline bool direct_emission(KernelGlobals *kg, ShaderData *sd,
 	differential3 dD = differential3_zero();
 
 	/* evaluate closure */
-	float3 light_eval = direct_emissive_eval(kg, ls, -ls->D, dD, ls->t, sd->time, bounce, transparent_bounce);
+
+	float3 light_eval = direct_emissive_eval(kg, ls, -ls->D, dD, ls->t, ccl_fetch(sd, time),
+	                                         bounce,
+	                                         transparent_bounce
+#ifdef __SPLIT_KERNEL__
+	                                         ,sd_DL
+#endif
+	                                         );
 
 	if(is_zero(light_eval))
 		return false;
@@ -83,7 +103,7 @@ ccl_device_noinline bool direct_emission(KernelGlobals *kg, ShaderData *sd,
 	float bsdf_pdf;
 
 #ifdef __VOLUME__
-	if(sd->prim != PRIM_NONE)
+	if(ccl_fetch(sd, prim) != PRIM_NONE)
 		shader_bsdf_eval(kg, sd, ls->D, eval, &bsdf_pdf);
 	else
 		shader_volume_phase_eval(kg, sd, ls->D, eval, &bsdf_pdf);
@@ -118,8 +138,8 @@ ccl_device_noinline bool direct_emission(KernelGlobals *kg, ShaderData *sd,
 
 	if(ls->shader & SHADER_CAST_SHADOW) {
 		/* setup ray */
-		bool transmit = (dot(sd->Ng, ls->D) < 0.0f);
-		ray->P = ray_offset(sd->P, (transmit)? -sd->Ng: sd->Ng);
+		bool transmit = (dot(ccl_fetch(sd, Ng), ls->D) < 0.0f);
+		ray->P = ray_offset(ccl_fetch(sd, P), (transmit)? -ccl_fetch(sd, Ng): ccl_fetch(sd, Ng));
 
 		if(ls->t == FLT_MAX) {
 			/* distant light */
@@ -132,7 +152,7 @@ ccl_device_noinline bool direct_emission(KernelGlobals *kg, ShaderData *sd,
 			ray->D = normalize_len(ray->D, &ray->t);
 		}
 
-		ray->dP = sd->dP;
+		ray->dP = ccl_fetch(sd, dP);
 		ray->dD = differential3_zero();
 	}
 	else {
@@ -154,14 +174,14 @@ ccl_device_noinline float3 indirect_primitive_emission(KernelGlobals *kg, Shader
 	float3 L = shader_emissive_eval(kg, sd);
 
 #ifdef __HAIR__
-	if(!(path_flag & PATH_RAY_MIS_SKIP) && (sd->flag & SD_USE_MIS) && (sd->type & PRIMITIVE_ALL_TRIANGLE))
+	if(!(path_flag & PATH_RAY_MIS_SKIP) && (ccl_fetch(sd, flag) & SD_USE_MIS) && (ccl_fetch(sd, type) & PRIMITIVE_ALL_TRIANGLE))
 #else
-	if(!(path_flag & PATH_RAY_MIS_SKIP) && (sd->flag & SD_USE_MIS))
+	if(!(path_flag & PATH_RAY_MIS_SKIP) && (ccl_fetch(sd, flag) & SD_USE_MIS))
 #endif
 	{
 		/* multiple importance sampling, get triangle light pdf,
 		 * and compute weight with respect to BSDF pdf */
-		float pdf = triangle_light_pdf(kg, sd->Ng, sd->I, t);
+		float pdf = triangle_light_pdf(kg, ccl_fetch(sd, Ng), ccl_fetch(sd, I), t);
 		float mis_weight = power_heuristic(bsdf_pdf, pdf);
 
 		return L*mis_weight;
@@ -172,7 +192,12 @@ ccl_device_noinline float3 indirect_primitive_emission(KernelGlobals *kg, Shader
 
 /* Indirect Lamp Emission */
 
-ccl_device_noinline bool indirect_lamp_emission(KernelGlobals *kg, PathState *state, Ray *ray, float3 *emission)
+/* The argument sd is meaningful only for split kernel. Other uses can just pass NULL */
+ccl_device_noinline bool indirect_lamp_emission(KernelGlobals *kg, PathState *state, Ray *ray, float3 *emission
+#ifdef __SPLIT_KERNEL__
+                                                ,ShaderData *sd
+#endif
+                                                )
 {
 	bool hit_lamp = false;
 
@@ -196,7 +221,13 @@ ccl_device_noinline bool indirect_lamp_emission(KernelGlobals *kg, PathState *st
 		}
 #endif
 
-		float3 L = direct_emissive_eval(kg, &ls, -ray->D, ray->dD, ls.t, ray->time, state->bounce, state->transparent_bounce);
+		float3 L = direct_emissive_eval(kg, &ls, -ray->D, ray->dD, ls.t, ray->time,
+		                                state->bounce,
+		                                state->transparent_bounce
+#ifdef __SPLIT_KERNEL__
+		                                ,sd
+#endif
+		                                );
 
 #ifdef __VOLUME__
 		if(state->volume_stack[0].shader != SHADER_NONE) {
@@ -225,7 +256,11 @@ ccl_device_noinline bool indirect_lamp_emission(KernelGlobals *kg, PathState *st
 
 /* Indirect Background */
 
-ccl_device_noinline float3 indirect_background(KernelGlobals *kg, PathState *state, Ray *ray)
+ccl_device_noinline float3 indirect_background(KernelGlobals *kg, ccl_addr_space PathState *state, ccl_addr_space Ray *ray
+#ifdef __SPLIT_KERNEL__
+                                               ,ShaderData *sd_global
+#endif
+                                               )
 {
 #ifdef __BACKGROUND__
 	int shader = kernel_data.background.surface_shader;
@@ -241,11 +276,17 @@ ccl_device_noinline float3 indirect_background(KernelGlobals *kg, PathState *sta
 			return make_float3(0.0f, 0.0f, 0.0f);
 	}
 
+#ifdef __SPLIT_KERNEL__
 	/* evaluate background closure */
+	Ray priv_ray = *ray;
+	shader_setup_from_background(kg, sd_global, &priv_ray, state->bounce+1, state->transparent_bounce);
+	float3 L = shader_eval_background(kg, sd_global, state->flag, SHADER_CONTEXT_EMISSION);
+#else
 	ShaderData sd;
 	shader_setup_from_background(kg, &sd, ray, state->bounce+1, state->transparent_bounce);
 
 	float3 L = shader_eval_background(kg, &sd, state->flag, SHADER_CONTEXT_EMISSION);
+#endif
 
 #ifdef __BACKGROUND_MIS__
 	/* check if background light exists or if we should skip pdf */