Fix T40370: cycles CUDA baking timeout with high number of AA samples.

Now baking does one AA sample at a time, just like final render. There is
also some code for shader antialiasing that solves T40369 but it is disabled
for now because there may be unpredictable side effects.

3 files changed, 51 insertions, 36 deletions

diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index b0739dd20b4..71bf2d23d6e 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -393,7 +393,8 @@ public:
 #ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX
 		if(system_cpu_support_avx()) {
 			for(int x = task.shader_x; x < task.shader_x + task.shader_w; x++) {
-				kernel_cpu_avx_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x);
+				for(int sample = 0; sample < task.num_samples; sample++)
+					kernel_cpu_avx_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x, sample);
 
 				if(task.get_cancel() || task_pool.canceled())
 					break;
@@ -404,7 +405,8 @@ public:
 #ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE41			
 		if(system_cpu_support_sse41()) {
 			for(int x = task.shader_x; x < task.shader_x + task.shader_w; x++) {
-				kernel_cpu_sse41_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x);
+				for(int sample = 0; sample < task.num_samples; sample++)
+					kernel_cpu_sse41_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x, sample);
 
 				if(task.get_cancel() || task_pool.canceled())
 					break;
@@ -415,7 +417,8 @@ public:
 #ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE3
 		if(system_cpu_support_sse3()) {
 			for(int x = task.shader_x; x < task.shader_x + task.shader_w; x++) {
-				kernel_cpu_sse3_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x);
+				for(int sample = 0; sample < task.num_samples; sample++)
+					kernel_cpu_sse3_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x, sample);
 
 				if(task.get_cancel() || task_pool.canceled())
 					break;
@@ -426,7 +429,8 @@ public:
 #ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE2
 		if(system_cpu_support_sse2()) {
 			for(int x = task.shader_x; x < task.shader_x + task.shader_w; x++) {
-				kernel_cpu_sse2_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x);
+				for(int sample = 0; sample < task.num_samples; sample++)
+					kernel_cpu_sse2_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x, sample);
 
 				if(task.get_cancel() || task_pool.canceled())
 					break;
@@ -436,7 +440,8 @@ public:
 #endif
 		{
 			for(int x = task.shader_x; x < task.shader_x + task.shader_w; x++) {
-				kernel_cpu_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x);
+				for(int sample = 0; sample < task.num_samples; sample++)
+					kernel_cpu_shader(&kg, (uint4*)task.shader_input, (float4*)task.shader_output, task.shader_eval_type, x, sample);
 
 				if(task.get_cancel() || task_pool.canceled())
 					break;
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 0429bfc6e97..0aa09ac5383 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -764,40 +764,45 @@ public:
 
 			int shader_w = min(shader_chunk_size, end - shader_x);
 
-			/* pass in parameters */
-			int offset = 0;
+			for(int sample = 0; sample < task.num_samples; sample++) {
+				/* pass in parameters */
+				int offset = 0;
 
-			cuda_assert(cuParamSetv(cuShader, offset, &d_input, sizeof(d_input)));
-			offset += sizeof(d_input);
+				cuda_assert(cuParamSetv(cuShader, offset, &d_input, sizeof(d_input)));
+				offset += sizeof(d_input);
 
-			cuda_assert(cuParamSetv(cuShader, offset, &d_output, sizeof(d_output)));
-			offset += sizeof(d_output);
+				cuda_assert(cuParamSetv(cuShader, offset, &d_output, sizeof(d_output)));
+				offset += sizeof(d_output);
 
-			int shader_eval_type = task.shader_eval_type;
-			offset = align_up(offset, __alignof(shader_eval_type));
+				int shader_eval_type = task.shader_eval_type;
+				offset = align_up(offset, __alignof(shader_eval_type));
 
-			cuda_assert(cuParamSeti(cuShader, offset, task.shader_eval_type));
-			offset += sizeof(task.shader_eval_type);
+				cuda_assert(cuParamSeti(cuShader, offset, task.shader_eval_type));
+				offset += sizeof(task.shader_eval_type);
 
-			cuda_assert(cuParamSeti(cuShader, offset, shader_x));
-			offset += sizeof(shader_x);
+				cuda_assert(cuParamSeti(cuShader, offset, shader_x));
+				offset += sizeof(shader_x);
 
-			cuda_assert(cuParamSeti(cuShader, offset, shader_w));
-			offset += sizeof(shader_w);
+				cuda_assert(cuParamSeti(cuShader, offset, shader_w));
+				offset += sizeof(shader_w);
 
-			cuda_assert(cuParamSetSize(cuShader, offset));
+				cuda_assert(cuParamSeti(cuShader, offset, sample));
+				offset += sizeof(sample);
 
-			/* launch kernel */
-			int threads_per_block;
-			cuda_assert(cuFuncGetAttribute(&threads_per_block, CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK, cuShader));
+				cuda_assert(cuParamSetSize(cuShader, offset));
 
-			int xblocks = (shader_w + threads_per_block - 1)/threads_per_block;
+				/* launch kernel */
+				int threads_per_block;
+				cuda_assert(cuFuncGetAttribute(&threads_per_block, CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK, cuShader));
 
-			cuda_assert(cuFuncSetCacheConfig(cuShader, CU_FUNC_CACHE_PREFER_L1));
-			cuda_assert(cuFuncSetBlockShape(cuShader, threads_per_block, 1, 1));
-			cuda_assert(cuLaunchGrid(cuShader, xblocks, 1));
+				int xblocks = (shader_w + threads_per_block - 1)/threads_per_block;
 
-			cuda_assert(cuCtxSynchronize());
+				cuda_assert(cuFuncSetCacheConfig(cuShader, CU_FUNC_CACHE_PREFER_L1));
+				cuda_assert(cuFuncSetBlockShape(cuShader, threads_per_block, 1, 1));
+				cuda_assert(cuLaunchGrid(cuShader, xblocks, 1));
+
+				cuda_assert(cuCtxSynchronize());
+			}
 		}
 
 		cuda_pop_context();
diff --git a/intern/cycles/device/device_opencl.cpp b/intern/cycles/device/device_opencl.cpp
index f841daba124..abfe445414a 100644
--- a/intern/cycles/device/device_opencl.cpp
+++ b/intern/cycles/device/device_opencl.cpp
@@ -1067,19 +1067,24 @@ public:
 		else
 			kernel = ckShaderKernel;
 
-		opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_data), (void*)&d_data));
-		opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_input), (void*)&d_input));
-		opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_output), (void*)&d_output));
+		for(int sample = 0; sample < task.num_samples; sample++) {
+			cl_int d_sample = task.sample;
+
+			opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_data), (void*)&d_data));
+			opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_input), (void*)&d_input));
+			opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_output), (void*)&d_output));
 
 #define KERNEL_TEX(type, ttype, name) \
-	set_kernel_arg_mem(kernel, &narg, #name);
+		set_kernel_arg_mem(kernel, &narg, #name);
 #include "kernel_textures.h"
 
-		opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_shader_eval_type), (void*)&d_shader_eval_type));
-		opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_shader_x), (void*)&d_shader_x));
-		opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_shader_w), (void*)&d_shader_w));
+			opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_shader_eval_type), (void*)&d_shader_eval_type));
+			opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_shader_x), (void*)&d_shader_x));
+			opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_shader_w), (void*)&d_shader_w));
+			opencl_assert(clSetKernelArg(kernel, narg++, sizeof(d_sample), (void*)&d_sample));
 
-		enqueue_kernel(kernel, task.shader_w, 1);
+			enqueue_kernel(kernel, task.shader_w, 1);
+		}
 	}
 
 	void thread_run(DeviceTask *task)