Cycles: OpenCL kernel split

This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.

Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.

Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.

This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.

More feature will be enabled once they're ported to the split kernel and
tested.

Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.

Based on the research paper:

  https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf

Here's the documentation:

  https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit

Design discussion of the patch:

  https://developer.blender.org/T44197

Differential Revision: https://developer.blender.org/D1200

1 files changed, 132 insertions, 0 deletions

diff --git a/intern/cycles/kernel/kernel_queues.h b/intern/cycles/kernel/kernel_queues.h
new file mode 100644
index 00000000000..9e65e2b0768
--- /dev/null
+++ b/intern/cycles/kernel/kernel_queues.h
@@ -0,0 +1,132 @@
+/*
+ * Copyright 2011-2015 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef __KERNEL_QUEUE_H__
+#define __KERNEL_QUEUE_H__
+
+/*
+ * Queue utility functions for split kernel
+ */
+
+#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable
+#pragma OPENCL EXTENSION cl_khr_local_int32_base_atomics : enable
+
+/*
+ * Enqueue ray index into the queue
+ */
+ccl_device void enqueue_ray_index (
+              int ray_index,               /* Ray index to be enqueued */
+              int queue_number,            /* Queue in which the ray index should be enqueued*/
+              ccl_global int *queues,      /* Buffer of all queues */
+              int queue_size,              /* Size of each queue */
+              ccl_global int *queue_index /* Array of size num_queues; Used for atomic increment */
+              )
+{
+	/* This thread's queue index */
+	int my_queue_index = atomic_inc(&queue_index[queue_number]) + (queue_number * queue_size);
+	queues[my_queue_index] = ray_index;
+}
+
+/*
+ * Get the ray index for this thread
+ * Returns a positive ray_index for threads that have to do some work;
+ * Returns 'QUEUE_EMPTY_SLOT' for threads that don't have any work
+ * i.e All ray's in the queue has been successfully allocated and there
+ * is no more ray to allocate to other threads.
+ */
+ccl_device int get_ray_index (
+              int thread_index,       /* Global thread index */
+              int queue_number,       /* Queue to operate on */
+              ccl_global int *queues, /* Buffer of all queues */
+              int queuesize,          /* Size of a queue */
+              int empty_queue         /* Empty the queue slot as soon as we fetch the ray index */
+              )
+{
+	int ray_index = queues[queue_number * queuesize + thread_index];
+
+	if(empty_queue && ray_index != QUEUE_EMPTY_SLOT) {
+		queues[queue_number * queuesize + thread_index] = QUEUE_EMPTY_SLOT;
+	}
+
+	return ray_index;
+}
+
+/* The following functions are to realize Local memory variant of enqueue ray index function */
+
+/* All threads should call this function */
+ccl_device void enqueue_ray_index_local(
+              int ray_index,                               /* Ray index to enqueue*/
+              int queue_number,                            /* Queue in which to enqueue ray index */
+              char enqueue_flag,                           /* True for threads whose ray index has to be enqueued */
+              int queuesize,                               /* queue size */
+              ccl_local unsigned int *local_queue_atomics,   /* To to local queue atomics */
+              ccl_global int *Queue_data,                  /* Queues */
+              ccl_global int *Queue_index                  /* To do global queue atomics */
+              )
+{
+	int lidx = get_local_id(1) * get_local_size(0) + get_local_id(0);
+
+	/* Get local queue id */
+	unsigned int lqidx;
+	if(enqueue_flag) {
+		lqidx = atomic_inc(local_queue_atomics);
+	}
+	barrier(CLK_LOCAL_MEM_FENCE);
+
+	/* Get global queue offset */
+	if(lidx == 0) {
+		*local_queue_atomics = atomic_add(&Queue_index[queue_number], *local_queue_atomics);
+	}
+	barrier(CLK_LOCAL_MEM_FENCE);
+
+	/* Get global queue index and enqueue ray */
+	if(enqueue_flag) {
+		unsigned int my_gqidx = queue_number * queuesize + (*local_queue_atomics) + lqidx;
+		Queue_data[my_gqidx] = ray_index;
+	}
+}
+
+ccl_device unsigned int get_local_queue_index(
+               int queue_number, /* Queue in which to enqueue the ray; -1 if no queue */
+               ccl_local unsigned int *local_queue_atomics
+               )
+{
+	int my_lqidx = atomic_inc(&local_queue_atomics[queue_number]);
+	return my_lqidx;
+}
+
+ccl_device unsigned int get_global_per_queue_offset(
+               int queue_number,
+               ccl_local unsigned int *local_queue_atomics,
+               ccl_global int* global_queue_atomics
+               )
+{
+	unsigned int queue_offset = atomic_add((&global_queue_atomics[queue_number]), local_queue_atomics[queue_number]);
+	return queue_offset;
+}
+
+ccl_device unsigned int get_global_queue_index(
+               int queue_number,
+               int queuesize,
+               unsigned int lqidx,
+               ccl_local unsigned int * global_per_queue_offset
+               )
+{
+	int my_gqidx = queuesize * queue_number + lqidx + global_per_queue_offset[queue_number];
+	return my_gqidx;
+}
+
+#endif // __KERNEL_QUEUE_H__