Cycles: Initialize the RNG state from the kernel instead of the host

This allows to save a memory copy, which will be particularly useful for network rendering.

Reviewers: sergey, brecht, dingto, juicyfruit, maiself

Differential Revision: https://developer.blender.org/D2323

4 files changed, 11 insertions, 10 deletions

diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index 694f19a808a..56bcafbce38 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -177,6 +177,7 @@ set(SRC_UTIL_HEADERS
 	../util/util_atomic.h
 	../util/util_color.h
 	../util/util_half.h
+	../util/util_hash.h
 	../util/util_math.h
 	../util/util_math_fast.h
 	../util/util_static_assert.h
diff --git a/intern/cycles/kernel/kernel_path_common.h b/intern/cycles/kernel/kernel_path_common.h
index 1912dfa16ed..13597eab287 100644
--- a/intern/cycles/kernel/kernel_path_common.h
+++ b/intern/cycles/kernel/kernel_path_common.h
@@ -14,6 +14,8 @@
  * limitations under the License.
  */
 
+#include "util_hash.h"
+
 CCL_NAMESPACE_BEGIN
 
 ccl_device_inline void kernel_path_trace_setup(KernelGlobals *kg,
@@ -28,6 +30,10 @@ ccl_device_inline void kernel_path_trace_setup(KernelGlobals *kg,
 
 	int num_samples = kernel_data.integrator.aa_samples;
 
+	if(sample == 0) {
+		*rng_state = hash_int_2d(x, y);
+	}
+
 	path_rng_init(kg, rng_state, sample, num_samples, rng, x, y, &filter_u, &filter_v);
 
 	/* sample camera ray */
diff --git a/intern/cycles/render/buffers.cpp b/intern/cycles/render/buffers.cpp
index 1e170d3a96e..cb20e811708 100644
--- a/intern/cycles/render/buffers.cpp
+++ b/intern/cycles/render/buffers.cpp
@@ -135,15 +135,7 @@ void RenderBuffers::reset(Device *device, BufferParams& params_)
 	/* allocate rng state */
 	rng_state.resize(params.width, params.height);
 
-	uint *init_state = rng_state.resize(params.width, params.height);
-	int x, y, width = params.width, height = params.height;
-	
-	for(y = 0; y < height; y++)
-		for(x = 0; x < width; x++)
-			init_state[y*width + x] = hash_int_2d(params.full_x+x, params.full_y+y);
-
 	device->mem_alloc(rng_state, MEM_READ_WRITE);
-	device->mem_copy_to(rng_state);
 }
 
 bool RenderBuffers::copy_from_device()
diff --git a/intern/cycles/util/util_hash.h b/intern/cycles/util/util_hash.h
index 3ff2802b46d..98c3a681ff2 100644
--- a/intern/cycles/util/util_hash.h
+++ b/intern/cycles/util/util_hash.h
@@ -21,7 +21,7 @@
 
 CCL_NAMESPACE_BEGIN
 
-static inline uint hash_int_2d(uint kx, uint ky)
+ccl_device_inline uint hash_int_2d(uint kx, uint ky)
 {
 #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
 
@@ -44,11 +44,12 @@ static inline uint hash_int_2d(uint kx, uint ky)
 #undef rot
 }
 
-static inline uint hash_int(uint k)
+ccl_device_inline uint hash_int(uint k)
 {
 	return hash_int_2d(k, 0);
 }
 
+#ifndef __KERNEL_GPU__
 static inline uint hash_string(const char *str)
 {
 	uint i = 0, c;
@@ -58,6 +59,7 @@ static inline uint hash_string(const char *str)
 
 	return i;
 }
+#endif
 
 CCL_NAMESPACE_END