gallium: Initial port of Thomas slab suballocator to pipebuffer.

Not tested yet -- just compiles.

This includes only the slab algorithm. Fencing is already
implemented in pb_bufmgr_fence and time-based caching will be commited in a
separate module shortly.

4 files changed, 440 insertions, 0 deletions

diff --git a/src/gallium/auxiliary/pipebuffer/Makefile b/src/gallium/auxiliary/pipebuffer/Makefile
index a9fa518c674..0770cad45a1 100644
--- a/src/gallium/auxiliary/pipebuffer/Makefile
+++ b/src/gallium/auxiliary/pipebuffer/Makefile
@@ -9,6 +9,7 @@ C_SOURCES = \
 	pb_bufmgr_fenced.c \
 	pb_bufmgr_mm.c \
 	pb_bufmgr_pool.c \
+	pb_bufmgr_slab.c \
 	pb_winsys.c
 
 include ../../Makefile.template
diff --git a/src/gallium/auxiliary/pipebuffer/SConscript b/src/gallium/auxiliary/pipebuffer/SConscript
index 3d41fd84a6c..14c3c77e37d 100644
--- a/src/gallium/auxiliary/pipebuffer/SConscript
+++ b/src/gallium/auxiliary/pipebuffer/SConscript
@@ -8,6 +8,7 @@ pipebuffer = env.ConvenienceLibrary(
 		'pb_bufmgr_fenced.c',
 		'pb_bufmgr_mm.c',
 		'pb_bufmgr_pool.c',
+		'pb_bufmgr_slab.c',
 		'pb_winsys.c',
 	])
 
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
index 0cf8e92e377..a2377f70e2f 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
@@ -114,6 +114,15 @@ mm_bufmgr_create_from_buffer(struct pb_buffer *buffer,
                              size_t size, size_t align2);
 
 
+struct pb_manager *
+pb_slab_manager_create(struct pb_manager *provider, 
+                       const struct pb_desc *desc,
+                       size_t smallestSize,
+                       size_t numSizes,
+                       size_t desiredNumBuffers,
+                       size_t maxSlabSize,
+                       size_t pageAlignment);
+                       
 /** 
  * Fenced buffer manager.
  *
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
new file mode 100644
index 00000000000..676e8e29b9c
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
@@ -0,0 +1,429 @@
+/**************************************************************************
+ *
+ * Copyright 2006-2008 Tungsten Graphics, Inc., Cedar Park, TX., USA
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, FREE of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * S-lab pool implementation.
+ * 
+ * @author Thomas Hellstrom <thomas-at-tungstengraphics-dot-com>
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_error.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "util/u_double_list.h"
+#include "util/u_time.h"
+
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+#define DRI_SLABPOOL_ALLOC_RETRIES 100
+
+
+struct pb_slab;
+
+struct pb_slab_buffer
+{
+   struct pb_buffer base;
+   
+   struct pb_slab *slab;
+   struct list_head head;
+   unsigned mapCount;
+   size_t start;
+   _glthread_Cond event;
+};
+
+struct pb_slab
+{
+   struct list_head head;
+   struct list_head freeBuffers;
+   size_t numBuffers;
+   size_t numFree;
+   struct pb_slab_buffer *buffers;
+   struct pb_slab_size_header *header;
+   
+   struct pb_buffer *bo;
+   size_t pageAlignment;
+   void *virtual;   
+};
+
+struct pb_slab_size_header 
+{
+   struct list_head slabs;
+   struct list_head freeSlabs;
+   struct pb_slab_manager *pool;
+   size_t bufSize;
+   _glthread_Mutex mutex;
+};
+
+/**
+ * The data of this structure remains constant after
+ * initialization and thus needs no mutex protection.
+ */
+struct pb_slab_manager 
+{
+   struct pb_manager base;
+
+   struct pb_desc desc;
+   size_t *bucketSizes;
+   size_t numBuckets;
+   size_t pageSize;
+   struct pb_manager *provider;
+   unsigned pageAlignment;
+   unsigned maxSlabSize;
+   unsigned desiredNumBuffers;
+   struct pb_slab_size_header *headers;
+};
+
+
+static INLINE struct pb_slab_buffer *
+pb_slab_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   return (struct pb_slab_buffer *)buf;
+}
+
+
+static INLINE struct pb_slab_manager *
+pb_slab_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct pb_slab_manager *)mgr;
+}
+
+
+/**
+ * Delete a buffer from the slab header delayed list and put
+ * it on the slab FREE list.
+ */
+static void
+pb_slab_buffer_destroy(struct pb_buffer *_buf)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+   struct pb_slab *slab = buf->slab;
+   struct pb_slab_size_header *header = slab->header;
+   struct list_head *list = &buf->head;
+
+   _glthread_LOCK_MUTEX(header->mutex);
+   
+   assert(buf->base.base.refcount == 0);
+   
+   buf->mapCount = 0;
+
+   LIST_DEL(list);
+   LIST_ADDTAIL(list, &slab->freeBuffers);
+   slab->numFree++;
+
+   if (slab->head.next == &slab->head)
+      LIST_ADDTAIL(&slab->head, &header->slabs);
+
+   if (slab->numFree == slab->numBuffers) {
+      list = &slab->head;
+      LIST_DEL(list);
+      LIST_ADDTAIL(list, &header->freeSlabs);
+   }
+
+   if (header->slabs.next == &header->slabs || slab->numFree
+	 != slab->numBuffers) {
+
+      struct list_head *next;
+
+      for (list = header->freeSlabs.next, next = list->next; list
+	    != &header->freeSlabs; list = next, next = list->next) {
+
+	 slab = LIST_ENTRY(struct pb_slab, list, head);
+
+	 LIST_DELINIT(list);
+	 pb_reference(&slab->bo, NULL);
+	 FREE(slab->buffers);
+	 FREE(slab);
+      }
+   }
+   
+   _glthread_UNLOCK_MUTEX(header->mutex);
+}
+
+
+static void *
+pb_slab_buffer_map(struct pb_buffer *_buf, 
+                   unsigned flags)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+
+   ++buf->mapCount;
+   return (void *) ((uint8_t *) buf->slab->virtual + buf->start);
+}
+
+
+static void
+pb_slab_buffer_unmap(struct pb_buffer *_buf)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+
+   --buf->mapCount;
+   if (buf->mapCount == 0) 
+       _glthread_COND_BROADCAST(buf->event);
+}
+
+
+static void
+pb_slab_buffer_get_base_buffer(struct pb_buffer *_buf,
+                               struct pb_buffer **base_buf,
+                               unsigned *offset)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+   pb_get_base_buffer(buf->slab->bo, base_buf, offset);
+   *offset += buf->start;
+}
+
+
+static const struct pb_vtbl 
+pb_slab_buffer_vtbl = {
+      pb_slab_buffer_destroy,
+      pb_slab_buffer_map,
+      pb_slab_buffer_unmap,
+      pb_slab_buffer_get_base_buffer
+};
+
+
+static enum pipe_error
+pb_slab_create(struct pb_slab_size_header *header)
+{
+   struct pb_slab_manager *pool = header->pool;
+   size_t size = header->bufSize * pool->desiredNumBuffers;
+   struct pb_slab *slab;
+   struct pb_slab_buffer *buf;
+   size_t numBuffers;
+   int ret;
+   unsigned i;
+
+   slab = CALLOC_STRUCT(pb_slab);
+   if (!slab)
+      return PIPE_ERROR_OUT_OF_MEMORY;
+
+   /*
+    * FIXME: We should perhaps allow some variation in slabsize in order
+    * to efficiently reuse slabs.
+    */
+
+   size = (size <= pool->maxSlabSize) ? size : pool->maxSlabSize;
+   size = (size + pool->pageSize - 1) & ~(pool->pageSize - 1);
+
+   slab->bo = pool->provider->create_buffer(pool->provider, size, &pool->desc);
+   if(!slab->bo)
+      goto out_err0;
+
+   slab->virtual = pb_map(slab->bo, 
+			 PIPE_BUFFER_USAGE_CPU_READ |
+			 PIPE_BUFFER_USAGE_CPU_WRITE);
+   if(!slab->virtual)
+      goto out_err1;
+
+   pb_unmap(slab->bo);
+
+   numBuffers = slab->bo->base.size / header->bufSize;
+
+   slab->buffers = CALLOC(numBuffers, sizeof(*slab->buffers));
+   if (!slab->buffers) {
+      ret = PIPE_ERROR_OUT_OF_MEMORY;
+      goto out_err1;
+   }
+
+   LIST_INITHEAD(&slab->head);
+   LIST_INITHEAD(&slab->freeBuffers);
+   slab->numBuffers = numBuffers;
+   slab->numFree = 0;
+   slab->header = header;
+
+   buf = slab->buffers;
+   for (i=0; i < numBuffers; ++i) {
+      buf->base.base.refcount = 0;
+      buf->base.base.size = header->bufSize;
+      buf->base.base.alignment = 0;
+      buf->base.base.usage = 0;
+      buf->base.vtbl = &pb_slab_buffer_vtbl;
+      buf->slab = slab;
+      buf->start = i* header->bufSize;
+      buf->mapCount = 0;
+      _glthread_INIT_COND(buf->event);
+      LIST_ADDTAIL(&buf->head, &slab->freeBuffers);
+      slab->numFree++;
+      buf++;
+   }
+
+   LIST_ADDTAIL(&slab->head, &header->slabs);
+
+   return PIPE_OK;
+
+out_err1: 
+   pb_reference(&slab->bo, NULL);
+out_err0: 
+   FREE(slab);
+   return ret;
+}
+
+
+static struct pb_buffer *
+pb_slab_manager_create_buffer(struct pb_manager *_pool,
+                              size_t size,
+                              const struct pb_desc *desc)
+{
+   struct pb_slab_manager *pool = pb_slab_manager(_pool);
+   struct pb_slab_size_header *header;
+   unsigned i;
+   static struct pb_slab_buffer *buf;
+   struct pb_slab *slab;
+   struct list_head *list;
+   int count = DRI_SLABPOOL_ALLOC_RETRIES;
+
+   /*
+    * FIXME: Check for compatibility.
+    */
+
+   header = pool->headers;
+   for (i=0; i<pool->numBuckets; ++i) {
+      if (header->bufSize >= size)
+	 break;
+      header++;
+   }
+
+   if (i >= pool->numBuckets)
+      /* Fall back to allocate a buffer object directly from the provider. */
+      return pool->provider->create_buffer(pool->provider, size, desc);
+
+
+   _glthread_LOCK_MUTEX(header->mutex);
+   while (header->slabs.next == &header->slabs && count > 0) {
+      if (header->slabs.next != &header->slabs)
+	 break;
+
+      _glthread_UNLOCK_MUTEX(header->mutex);
+      if (count != DRI_SLABPOOL_ALLOC_RETRIES)
+	 util_time_sleep(1);
+      _glthread_LOCK_MUTEX(header->mutex);
+      (void) pb_slab_create(header);
+      count--;
+   }
+
+   list = header->slabs.next;
+   if (list == &header->slabs) {
+      _glthread_UNLOCK_MUTEX(header->mutex);
+      return NULL;
+   }
+   slab = LIST_ENTRY(struct pb_slab, list, head);
+   if (--slab->numFree == 0)
+      LIST_DELINIT(list);
+
+   list = slab->freeBuffers.next;
+   LIST_DELINIT(list);
+
+   _glthread_UNLOCK_MUTEX(header->mutex);
+   buf = LIST_ENTRY(struct pb_slab_buffer, list, head);
+   ++buf->base.base.refcount;
+   return &buf->base;
+}
+
+
+static void
+pb_slab_manager_destroy(struct pb_manager *_pool)
+{
+   struct pb_slab_manager *pool = pb_slab_manager(_pool);
+
+   FREE(pool->headers);
+   FREE(pool->bucketSizes);
+   FREE(pool);
+}
+
+
+struct pb_manager *
+pb_slab_manager_create(struct pb_manager *provider, 
+                       const struct pb_desc *desc,
+                       size_t smallestSize,
+                       size_t numSizes,
+                       size_t desiredNumBuffers,
+                       size_t maxSlabSize,
+                       size_t pageAlignment)
+{
+   struct pb_slab_manager *pool;
+   size_t i;
+
+   pool = CALLOC_STRUCT(pb_slab_manager);
+   if (!pool)
+      goto out_err0;
+
+   pool->bucketSizes = CALLOC(numSizes, sizeof(*pool->bucketSizes));
+   if (!pool->bucketSizes)
+      goto out_err1;
+
+   pool->headers = CALLOC(numSizes, sizeof(*pool->headers));
+   if (!pool->headers)
+      goto out_err2;
+
+   pool->desc = *desc;
+   pool->numBuckets = numSizes;
+#ifdef WIN32
+   pool->pageSize = 4096;
+#else
+   pool->pageSize = getpagesize();
+#endif
+   pool->provider = provider;
+   pool->pageAlignment = pageAlignment;
+   pool->maxSlabSize = maxSlabSize;
+   pool->desiredNumBuffers = desiredNumBuffers;
+
+   for (i=0; i<pool->numBuckets; ++i) {
+      struct pb_slab_size_header *header = &pool->headers[i];
+      
+      pool->bucketSizes[i] = (smallestSize << i);
+      
+      _glthread_INIT_MUTEX(header->mutex);
+
+      LIST_INITHEAD(&header->slabs);
+      LIST_INITHEAD(&header->freeSlabs);
+
+      header->pool = pool;
+      header->bufSize = (smallestSize << i);
+   }
+
+   pool->base.destroy = pb_slab_manager_destroy;
+   pool->base.create_buffer = pb_slab_manager_create_buffer;
+
+   return &pool->base;
+
+out_err2: 
+   FREE(pool->bucketSizes);
+out_err1: 
+   FREE(pool);
+out_err0:
+   return NULL;
+}