Cycles: Optionally output luminance from the shader evaluation kernel

This makes it possible to move some parts of evaluation from host to the device
and hopefully reduce memory usage by avoid having full RGBA buffer on the host.

Reviewers: juicyfruit, lukasstockner97, brecht

Reviewed By: lukasstockner97, brecht

Differential Revision: https://developer.blender.org/D1702

intern/cycles/device/device_cpu.cpp

@@ -343,7 +343,7 @@ public:
 #ifdef WITH_OSL
 		OSLShader::thread_init(&kg, &kernel_globals, &osl_globals);
 #endif
-		void(*shader_kernel)(KernelGlobals*, uint4*, float4*, int, int, int, int);
+		void(*shader_kernel)(KernelGlobals*, uint4*, float4*, float*, int, int, int, int);
 
 #ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX2
 		if(system_cpu_support_avx2())
@@ -374,8 +374,14 @@ public:
 
 		for(int sample = 0; sample < task.num_samples; sample++) {
 			for(int x = task.shader_x; x < task.shader_x + task.shader_w; x++)
-				shader_kernel(&kg, (uint4*)task.shader_input, (float4*)task.shader_output,
+				shader_kernel(&kg,
-					task.shader_eval_type, x, task.offset, sample);
+				              (uint4*)task.shader_input,
+				              (float4*)task.shader_output,
+				              (float*)task.shader_output_luma,
+				              task.shader_eval_type,
+				              x,
+				              task.offset,
+				              sample);
 
 			if(task.get_cancel() || task_pool.canceled())
 				break;

intern/cycles/device/device_cuda.cpp

@@ -726,6 +726,7 @@ public:
 		CUfunction cuShader;
 		CUdeviceptr d_input = cuda_device_ptr(task.shader_input);
 		CUdeviceptr d_output = cuda_device_ptr(task.shader_output);
+		CUdeviceptr d_output_luma = cuda_device_ptr(task.shader_output_luma);
 
 		/* get kernel function */
 		if(task.shader_eval_type >= SHADER_EVAL_BAKE) {
@@ -747,13 +748,18 @@ public:
 				int shader_w = min(shader_chunk_size, end - shader_x);
 
 				/* pass in parameters */
-				void *args[] = {&d_input,
+				void *args[8];
-								 &d_output,
+				int arg = 0;
-								 &task.shader_eval_type,
+				args[arg++] = &d_input;
-								 &shader_x,
+				args[arg++] = &d_output;
-								 &shader_w,
+				if(task.shader_eval_type < SHADER_EVAL_BAKE) {
-								 &offset,
+					args[arg++] = &d_output_luma;
-								 &sample};
+				}
+				args[arg++] = &task.shader_eval_type;
+				args[arg++] = &shader_x;
+				args[arg++] = &shader_w;
+				args[arg++] = &offset;
+				args[arg++] = &sample;
 
 				/* launch kernel */
 				int threads_per_block;

intern/cycles/device/device_multi.cpp

@@ -316,6 +316,7 @@ public:
 				if(task.rgba_half) subtask.rgba_half = sub.ptr_map[task.rgba_half];
 				if(task.shader_input) subtask.shader_input = sub.ptr_map[task.shader_input];
 				if(task.shader_output) subtask.shader_output = sub.ptr_map[task.shader_output];
+				if(task.shader_output_luma) subtask.shader_output_luma = sub.ptr_map[task.shader_output_luma];
 
 				sub.device->task_add(subtask);
 			}

intern/cycles/device/device_network.cpp

@@ -648,6 +648,9 @@ protected:
 			if(task.shader_output)
 				task.shader_output = device_ptr_from_client_pointer(task.shader_output);
 
+			if(task.shader_output)luma)
+				task.shader_output_luma = device_ptr_from_client_pointer(task.shader_output_luma);
+
 
 			task.acquire_tile = function_bind(&DeviceServer::task_acquire_tile, this, _1, _2);
 			task.release_tile = function_bind(&DeviceServer::task_release_tile, this, _1);

intern/cycles/device/device_network.h

@@ -132,7 +132,7 @@ public:
 		archive & type & task.x & task.y & task.w & task.h;
 		archive & task.rgba_byte & task.rgba_half & task.buffer & task.sample & task.num_samples;
 		archive & task.offset & task.stride;
-		archive & task.shader_input & task.shader_output & task.shader_eval_type;
+		archive & task.shader_input & task.shader_output & task.shader_output_luma & task.shader_eval_type;
 		archive & task.shader_x & task.shader_w;
 		archive & task.need_finish_queue;
 	}
@@ -291,7 +291,7 @@ public:
 		*archive & type & task.x & task.y & task.w & task.h;
 		*archive & task.rgba_byte & task.rgba_half & task.buffer & task.sample & task.num_samples;
 		*archive & task.offset & task.stride;
-		*archive & task.shader_input & task.shader_output & task.shader_eval_type;
+		*archive & task.shader_input & task.shader_output & task.shader_output_luma & task.shader_eval_type;
 		*archive & task.shader_x & task.shader_w;
 		*archive & task.need_finish_queue;
 

intern/cycles/device/device_opencl.cpp

@@ -1304,6 +1304,7 @@ public:
 		cl_mem d_data = CL_MEM_PTR(const_mem_map["__data"]->device_pointer);
 		cl_mem d_input = CL_MEM_PTR(task.shader_input);
 		cl_mem d_output = CL_MEM_PTR(task.shader_output);
+		cl_mem d_output_luma = CL_MEM_PTR(task.shader_output_luma);
 		cl_int d_shader_eval_type = task.shader_eval_type;
 		cl_int d_shader_x = task.shader_x;
 		cl_int d_shader_w = task.shader_w;
@@ -1330,6 +1331,12 @@ public:
 				                d_input,
 				                d_output);
 
+		if(task.shader_eval_type < SHADER_EVAL_BAKE) {
+			start_arg_index += kernel_set_args(kernel,
+			                                   start_arg_index,
+			                                   d_output_luma);
+		}
+
 #define KERNEL_TEX(type, ttype, name) \
 		set_kernel_arg_mem(kernel, &start_arg_index, #name);
 #include "kernel_textures.h"

intern/cycles/device/device_task.cpp

@@ -29,7 +29,7 @@ CCL_NAMESPACE_BEGIN
 DeviceTask::DeviceTask(Type type_)
 : type(type_), x(0), y(0), w(0), h(0), rgba_byte(0), rgba_half(0), buffer(0),
   sample(0), num_samples(1),
-  shader_input(0), shader_output(0),
+  shader_input(0), shader_output(0), shader_output_luma(0),
   shader_eval_type(0), shader_x(0), shader_w(0)
 {
 	last_update_time = time_dt();

intern/cycles/device/device_task.h

@@ -46,7 +46,7 @@ public:
 	int offset, stride;
 
 	device_ptr shader_input;
-	device_ptr shader_output;
+	device_ptr shader_output, shader_output_luma;
 	int shader_eval_type;
 	int shader_x, shader_w;
 

intern/cycles/kernel/kernel_bake.h

@@ -453,7 +453,13 @@ ccl_device void kernel_bake_evaluate(KernelGlobals *kg, ccl_global uint4 *input,
 		output[i] += make_float4(out.x, out.y, out.z, 1.0f) * output_fac;
 }
 
-ccl_device void kernel_shader_evaluate(KernelGlobals *kg, ccl_global uint4 *input, ccl_global float4 *output, ShaderEvalType type, int i, int sample)
+ccl_device void kernel_shader_evaluate(KernelGlobals *kg,
+                                       ccl_global uint4 *input,
+                                       ccl_global float4 *output,
+                                       ccl_global float *output_luma,
+                                       ShaderEvalType type,
+                                       int i,
+                                       int sample)
 {
 	ShaderData sd;
 	uint4 in = input[i];
@@ -500,10 +506,22 @@ ccl_device void kernel_shader_evaluate(KernelGlobals *kg, ccl_global uint4 *inpu
 	}
 	
 	/* write output */
-	if(sample == 0)
+	if(sample == 0) {
-		output[i] = make_float4(out.x, out.y, out.z, 0.0f);
+		if(output != NULL) {
-	else
+			output[i] = make_float4(out.x, out.y, out.z, 0.0f);
-		output[i] += make_float4(out.x, out.y, out.z, 0.0f);
+		}
+		if(output_luma != NULL) {
+			output_luma[i] = average(out);
+		}
+	}
+	else {
+		if(output != NULL) {
+			output[i] += make_float4(out.x, out.y, out.z, 0.0f);
+		}
+		if(output_luma != NULL) {
+			output_luma[i] += average(out);
+		}
+	}
 }
 
 CCL_NAMESPACE_END

intern/cycles/kernel/kernels/cpu/kernel_cpu.h

@@ -42,6 +42,7 @@ void KERNEL_FUNCTION_FULL_NAME(convert_to_half_float)(KernelGlobals *kg,
 void KERNEL_FUNCTION_FULL_NAME(shader)(KernelGlobals *kg,
                                        uint4 *input,
                                        float4 *output,
+                                       float *output_luma,
                                        int type,
                                        int i,
                                        int offset,

intern/cycles/kernel/kernels/cpu/kernel_cpu_impl.h

@@ -99,12 +99,14 @@ void KERNEL_FUNCTION_FULL_NAME(convert_to_half_float)(KernelGlobals *kg,
 void KERNEL_FUNCTION_FULL_NAME(shader)(KernelGlobals *kg,
                                        uint4 *input,
                                        float4 *output,
+                                       float *output_luma,
                                        int type,
                                        int i,
                                        int offset,
                                        int sample)
 {
 	if(type >= SHADER_EVAL_BAKE) {
+		kernel_assert(output_luma == NULL);
 		kernel_bake_evaluate(kg,
 		                     input,
 		                     output,
@@ -117,6 +119,7 @@ void KERNEL_FUNCTION_FULL_NAME(shader)(KernelGlobals *kg,
 		kernel_shader_evaluate(kg,
 		                       input,
 		                       output,
+		                       output_luma,
 		                       (ShaderEvalType)type,
 		                       i,
 		                       sample);

intern/cycles/kernel/kernels/cuda/kernel.cu

@@ -159,12 +159,26 @@ kernel_cuda_convert_to_half_float(uchar4 *rgba, float *buffer, float sample_scal
 
 extern "C" __global__ void
 CUDA_LAUNCH_BOUNDS(CUDA_THREADS_BLOCK_WIDTH, CUDA_KERNEL_MAX_REGISTERS)
-kernel_cuda_shader(uint4 *input, float4 *output, int type, int sx, int sw, int offset, int sample)
+kernel_cuda_shader(uint4 *input,
+                   float4 *output,
+                   float *output_luma,
+                   int type,
+                   int sx,
+                   int sw,
+                   int offset,
+                   int sample)
 {
 	int x = sx + blockDim.x*blockIdx.x + threadIdx.x;
 
-	if(x < sx + sw)
+	if(x < sx + sw) {
-		kernel_shader_evaluate(NULL, input, output, (ShaderEvalType)type, x, sample);
+		kernel_shader_evaluate(NULL,
+		                       input,
+		                       output,
+		                       output_luma,
+		                       (ShaderEvalType)type, 
+		                       x,
+		                       sample);
+	}
 }
 
 extern "C" __global__ void

intern/cycles/kernel/kernels/opencl/kernel.cl

@@ -61,6 +61,7 @@ __kernel void kernel_ocl_shader(
 	ccl_constant KernelData *data,
 	ccl_global uint4 *input,
 	ccl_global float4 *output,
+	ccl_global float *output_luma,
 
 #define KERNEL_TEX(type, ttype, name) \
 	ccl_global type *name,
@@ -78,8 +79,15 @@ __kernel void kernel_ocl_shader(
 
 	int x = sx + get_global_id(0);
 
-	if(x < sx + sw)
+	if(x < sx + sw) {
-		kernel_shader_evaluate(kg, input, output, (ShaderEvalType)type, x, sample);
+		kernel_shader_evaluate(kg,
+		                       input,
+		                       output,
+		                       output_luma,
+		                       (ShaderEvalType)type,
+		                       x,
+		                       sample);
+	}
 }
 
 __kernel void kernel_ocl_bake(