Code refactor: remove rng_state buffer and compute hash on the fly.

A little faster on some benchmark scenes, a little slower on others, seems
about performance neutral on average and saves a little memory.

7 files changed, 8 insertions, 23 deletions

diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index 6a1106328fb..72330b02a28 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -171,7 +171,7 @@ public:
 
 	DeviceRequestedFeatures requested_features;
 
-	KernelFunctions<void(*)(KernelGlobals *, float *, unsigned int *, int, int, int, int, int)>   path_trace_kernel;
+	KernelFunctions<void(*)(KernelGlobals *, float *, int, int, int, int, int)>   path_trace_kernel;
 	KernelFunctions<void(*)(KernelGlobals *, uchar4 *, float *, float, int, int, int, int)>       convert_to_half_float_kernel;
 	KernelFunctions<void(*)(KernelGlobals *, uchar4 *, float *, float, int, int, int, int)>       convert_to_byte_kernel;
 	KernelFunctions<void(*)(KernelGlobals *, uint4 *, float4 *, float*, int, int, int, int, int)> shader_kernel;
@@ -192,7 +192,7 @@ public:
 	KernelFunctions<void(*)(int, int, int, int, int, float*, int*, float*, float3*, int*, int)>                       filter_finalize_kernel;
 
 	KernelFunctions<void(*)(KernelGlobals *, ccl_constant KernelData*, ccl_global void*, int, ccl_global char*,
-	                       ccl_global uint*, int, int, int, int, int, int, int, int, ccl_global int*, int,
+	                       int, int, int, int, int, int, int, int, ccl_global int*, int,
 	                       ccl_global char*, ccl_global unsigned int*, unsigned int, ccl_global float*)>        data_init_kernel;
 	unordered_map<string, KernelFunctions<void(*)(KernelGlobals*, KernelData*)> > split_kernels;
 
@@ -617,7 +617,6 @@ public:
 	void path_trace(DeviceTask &task, RenderTile &tile, KernelGlobals *kg)
 	{
 		float *render_buffer = (float*)tile.buffer;
-		uint *rng_state = (uint*)tile.rng_state;
 		int start_sample = tile.start_sample;
 		int end_sample = tile.start_sample + tile.num_samples;
 
@@ -629,7 +628,7 @@ public:
 
 			for(int y = tile.y; y < tile.y + tile.h; y++) {
 				for(int x = tile.x; x < tile.x + tile.w; x++) {
-					path_trace_kernel()(kg, render_buffer, rng_state,
+					path_trace_kernel()(kg, render_buffer,
 					                    sample, x, y, tile.offset, tile.stride);
 				}
 			}
@@ -913,7 +912,6 @@ bool CPUSplitKernel::enqueue_split_kernel_data_init(const KernelDimensions& dim,
 			                           (void*)split_data.device_pointer,
 			                           num_global_elements,
 			                           (char*)ray_state.device_pointer,
-			                           (uint*)rtile.rng_state,
 			                           rtile.start_sample,
 			                           rtile.start_sample + rtile.num_samples,
 			                           rtile.x,
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 7ee74e9a512..e5464dcf34e 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -1322,7 +1322,6 @@ public:
 		wtile->start_sample = sample;
 		wtile->num_samples = 1;
 		wtile->buffer = (float*)cuda_device_ptr(rtile.buffer);
-		wtile->rng_state = (uint*)cuda_device_ptr(rtile.rng_state);
 
 		mem_alloc("work_tiles", work_tiles, MEM_READ_ONLY);
 		mem_copy_to(work_tiles);
@@ -1945,7 +1944,6 @@ bool CUDASplitKernel::enqueue_split_kernel_data_init(const KernelDimensions& dim
 	CUdeviceptr d_use_queues_flag = device->cuda_device_ptr(use_queues_flag.device_pointer);
 	CUdeviceptr d_work_pool_wgs = device->cuda_device_ptr(work_pool_wgs.device_pointer);
 
-	CUdeviceptr d_rng_state = device->cuda_device_ptr(rtile.rng_state);
 	CUdeviceptr d_buffer = device->cuda_device_ptr(rtile.buffer);
 
 	int end_sample = rtile.start_sample + rtile.num_samples;
@@ -1955,7 +1953,6 @@ bool CUDASplitKernel::enqueue_split_kernel_data_init(const KernelDimensions& dim
 		CUdeviceptr* split_data_buffer;
 		int* num_elements;
 		CUdeviceptr* ray_state;
-		CUdeviceptr* rng_state;
 		int* start_sample;
 		int* end_sample;
 		int* sx;
@@ -1976,7 +1973,6 @@ bool CUDASplitKernel::enqueue_split_kernel_data_init(const KernelDimensions& dim
 		&d_split_data,
 		&num_global_elements,
 		&d_ray_state,
-		&d_rng_state,
 		&rtile.start_sample,
 		&end_sample,
 		&rtile.x,
diff --git a/intern/cycles/device/device_multi.cpp b/intern/cycles/device/device_multi.cpp
index bc505b676fc..164ed50bdf6 100644
--- a/intern/cycles/device/device_multi.cpp
+++ b/intern/cycles/device/device_multi.cpp
@@ -281,7 +281,6 @@ public:
 		foreach(SubDevice& sub, devices) {
 			if(sub.device == sub_device) {
 				if(tile.buffer) tile.buffer = sub.ptr_map[tile.buffer];
-				if(tile.rng_state) tile.rng_state = sub.ptr_map[tile.rng_state];
 			}
 		}
 	}
diff --git a/intern/cycles/device/device_network.cpp b/intern/cycles/device/device_network.cpp
index 571ba9465ca..4ff8647f66b 100644
--- a/intern/cycles/device/device_network.cpp
+++ b/intern/cycles/device/device_network.cpp
@@ -737,7 +737,6 @@ protected:
 					tile = entry.tile;
 
 					if(tile.buffer) tile.buffer = ptr_map[tile.buffer];
-					if(tile.rng_state) tile.rng_state = ptr_map[tile.rng_state];
 
 					result = true;
 					break;
@@ -769,7 +768,6 @@ protected:
 		thread_scoped_lock acquire_lock(acquire_mutex);
 
 		if(tile.buffer) tile.buffer = ptr_imap[tile.buffer];
-		if(tile.rng_state) tile.rng_state = ptr_imap[tile.rng_state];
 
 		{
 			thread_scoped_lock lock(rpc_lock);
diff --git a/intern/cycles/device/device_network.h b/intern/cycles/device/device_network.h
index a5d24c66018..7bfebaf5aec 100644
--- a/intern/cycles/device/device_network.h
+++ b/intern/cycles/device/device_network.h
@@ -142,7 +142,7 @@ public:
 		archive & tile.x & tile.y & tile.w & tile.h;
 		archive & tile.start_sample & tile.num_samples & tile.sample;
 		archive & tile.resolution & tile.offset & tile.stride;
-		archive & tile.buffer & tile.rng_state;
+		archive & tile.buffer;
 	}
 
 	void write()
@@ -303,7 +303,7 @@ public:
 		*archive & tile.x & tile.y & tile.w & tile.h;
 		*archive & tile.start_sample & tile.num_samples & tile.sample;
 		*archive & tile.resolution & tile.offset & tile.stride;
-		*archive & tile.buffer & tile.rng_state;
+		*archive & tile.buffer;
 
 		tile.buffers = NULL;
 	}
diff --git a/intern/cycles/device/opencl/opencl_mega.cpp b/intern/cycles/device/opencl/opencl_mega.cpp
index ec47fdafa3d..f4555eaba4f 100644
--- a/intern/cycles/device/opencl/opencl_mega.cpp
+++ b/intern/cycles/device/opencl/opencl_mega.cpp
@@ -62,7 +62,6 @@ public:
 		/* Cast arguments to cl types. */
 		cl_mem d_data = CL_MEM_PTR(const_mem_map["__data"]->device_pointer);
 		cl_mem d_buffer = CL_MEM_PTR(rtile.buffer);
-		cl_mem d_rng_state = CL_MEM_PTR(rtile.rng_state);
 		cl_int d_x = rtile.x;
 		cl_int d_y = rtile.y;
 		cl_int d_w = rtile.w;
@@ -79,8 +78,7 @@ public:
 			kernel_set_args(ckPathTraceKernel,
 			                0,
 			                d_data,
-			                d_buffer,
-			                d_rng_state);
+			                d_buffer);
 
 		set_kernel_arg_buffers(ckPathTraceKernel, &start_arg_index);
 
diff --git a/intern/cycles/device/opencl/opencl_split.cpp b/intern/cycles/device/opencl/opencl_split.cpp
index 16a96213100..976cc9df46d 100644
--- a/intern/cycles/device/opencl/opencl_split.cpp
+++ b/intern/cycles/device/opencl/opencl_split.cpp
@@ -192,7 +192,6 @@ struct CachedSplitMemory {
 	int id;
 	device_memory *split_data;
 	device_memory *ray_state;
-	device_ptr *rng_state;
 	device_memory *queue_index;
 	device_memory *use_queues_flag;
 	device_memory *work_pools;
@@ -225,8 +224,7 @@ public:
 					            kg,
 					            data,
 					            *cached_memory.split_data,
-					            *cached_memory.ray_state,
-					            *cached_memory.rng_state);
+					            *cached_memory.ray_state);
 
 				device->set_kernel_arg_buffers(program(), &start_arg_index);
 
@@ -356,8 +354,7 @@ public:
 			                kernel_data,
 			                split_data,
 			                num_global_elements,
-			                ray_state,
-			                rtile.rng_state);
+			                ray_state);
 
 			device->set_kernel_arg_buffers(device->program_data_init(), &start_arg_index);
 
@@ -401,7 +398,6 @@ public:
 
 		cached_memory.split_data = &split_data;
 		cached_memory.ray_state = &ray_state;
-		cached_memory.rng_state = &rtile.rng_state;
 		cached_memory.queue_index = &queue_index;
 		cached_memory.use_queues_flag = &use_queues_flag;
 		cached_memory.work_pools = &work_pool_wgs;