[Bf-blender-cvs] [e8b5a5bf5b6] master: Cycles: Speedup transparent shadows in split kernel

[Hristo Gueorguiev]

Commit: e8b5a5bf5b63ef1c8980f8da95be32cad4d2cf0e
Author: Hristo Gueorguiev
Date:   Wed Mar 8 16:26:39 2017 +0100
Branches: master
https://developer.blender.org/rBe8b5a5bf5b63ef1c8980f8da95be32cad4d2cf0e

Cycles: Speedup transparent shadows in split kernel

This commit enables record-all transparent shadows rays.

Perfromance results:

               R9 290 render time (without synchronization), seconds
                        Before    After   Change
BMW                      261.5    262.5   +0.4 %
Classroom                869.6    867.3   -0.3 %
Fishy Cat                657.4    639.8   -2.7 %
Koro                    1909.8    692.8  -63.7 %
Pabellon Barcelona      1633.3   1238.0  -24.2 %
Pabellon Barcelona(*)   1158.1    903.8  -22.0 %

(*) without glossy connected to volume

===================================================================

M	intern/cycles/kernel/kernel_shadow.h
M	intern/cycles/kernel/kernel_types.h

===================================================================

diff --git a/intern/cycles/kernel/kernel_shadow.h b/intern/cycles/kernel/kernel_shadow.h
index 68a7ccfd903..4efc6c8118d 100644
--- a/intern/cycles/kernel/kernel_shadow.h
+++ b/intern/cycles/kernel/kernel_shadow.h
@@ -152,7 +152,13 @@ ccl_device bool shadow_blocked_transparent_all_loop(KernelGlobals *kg,
 		int bounce = state->transparent_bounce;
 		Intersection *isect = hits;
 #    ifdef __VOLUME__
-		PathState ps = *state;
+#      ifdef __SPLIT_KERNEL__
+		ccl_addr_space PathState *ps = &kernel_split_state.state_shadow[ccl_global_id(1) * ccl_global_size(0) + ccl_global_id(0)];
+#      else
+		PathState ps_object;
+		PathState *ps = &ps_object;
+#      endif
+		*ps = *state;
 #    endif
 		sort_intersections(hits, num_hits);
 		for(int hit = 0; hit < num_hits; hit++, isect++) {
@@ -171,7 +177,7 @@ ccl_device bool shadow_blocked_transparent_all_loop(KernelGlobals *kg,
 			                                   shadow_sd,
 			                                   state,
 #ifdef __VOLUME__
-			                                   &ps,
+			                                   ps,
 #endif
 			                                   isect,
 			                                   ray,
@@ -188,8 +194,8 @@ ccl_device bool shadow_blocked_transparent_all_loop(KernelGlobals *kg,
 		}
 #    ifdef __VOLUME__
 		/* Attenuation for last line segment towards light. */
-		if(ps.volume_stack[0].shader != SHADER_NONE) {
-			kernel_volume_shadow(kg, shadow_sd, &ps, ray, &throughput);
+		if(ps->volume_stack[0].shader != SHADER_NONE) {
+			kernel_volume_shadow(kg, shadow_sd, ps, ray, &throughput);
 		}
 #    endif
 		*shadow = throughput;
@@ -214,7 +220,10 @@ ccl_device bool shadow_blocked_transparent_all(KernelGlobals *kg,
                                                uint max_hits,
                                                float3 *shadow)
 {
-#    ifdef __KERNEL_CUDA__
+#    ifdef __SPLIT_KERNEL__
+	Intersection hits_[SHADOW_STACK_MAX_HITS];
+	Intersection *hits = &hits_[0];
+#    elif defined(__KERNEL_CUDA__)
 	Intersection *hits = kg->hits_stack;
 #    else
 	Intersection hits_stack[SHADOW_STACK_MAX_HITS];
diff --git a/intern/cycles/kernel/kernel_types.h b/intern/cycles/kernel/kernel_types.h
index 6c18cab6406..cac710c5ff3 100644
--- a/intern/cycles/kernel/kernel_types.h
+++ b/intern/cycles/kernel/kernel_types.h
@@ -80,9 +80,9 @@ CCL_NAMESPACE_BEGIN
 #  define __CMJ__
 #  define __VOLUME__
 #  define __VOLUME_SCATTER__
+#  define __SHADOW_RECORD_ALL__
 #  ifndef __SPLIT_KERNEL__
 #    define __VOLUME_DECOUPLED__
-#    define __SHADOW_RECORD_ALL__
 #    define __VOLUME_RECORD_ALL__
 #  endif
 #endif  /* __KERNEL_CPU__ */
@@ -131,6 +131,7 @@ CCL_NAMESPACE_BEGIN
 #    define __SUBSURFACE__
 #    define __VOLUME__
 #    define __VOLUME_SCATTER__
+#    define __SHADOW_RECORD_ALL__
 #  endif  /* __KERNEL_OPENCL_AMD__ */
 
 #  ifdef __KERNEL_OPENCL_INTEL_CPU__