P4 to Git Change 1576247 by gandryey@gera-w8 on 2018/07/04 16:55:54

SWDEV-158017 - CL_DEVICE_GLOBAL_FREE_MEMORY_AMD doesn't work correctly on PAL backend
	- Change the free memory detection to account possible memory allocation with oversubscription.
	- Make sure resource cache isn't counted as allocated memory

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#593 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuresource.cpp#244 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuresource.hpp#88 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#94 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.hpp#31 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.cpp#68 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.hpp#25 edit

rocclr/runtime/device/gpu/gpudevice.cpp

@@ -1805,7 +1805,8 @@ bool Device::globalFreeMemory(size_t* freeMemory) const {
 
   // Fill free memory info
   freeMemory[TotalFreeMemory] =
-      (memInfo.cardMemAvailableBytes + memInfo.cardExtMemAvailableBytes) / Ki;
+      (memInfo.cardMemAvailableBytes + memInfo.cardExtMemAvailableBytes +   
+       resourceCache().lclCacheSize()) / Ki;
   freeMemory[LargestFreeBlock] =
       std::max(memInfo.cardLargestFreeBlockBytes, memInfo.cardExtLargestFreeBlockBytes) / Ki;
   if (settings().apuSystem_) {

rocclr/runtime/device/gpu/gpuresource.cpp

@@ -1748,6 +1748,9 @@ bool ResourceCache::addCalResource(Resource::CalResourceDesc* desc, GslResourceR
       // Add the current resource to the cache
       resCache_.push_front({descCached, ref});
       cacheSize_ += size;
+      if (desc->type_ == Resource::Local) {
+          lclCacheSize_ += size;
+      }
       result = true;
     }
   }
@@ -1777,10 +1780,13 @@ GslResourceReference* ResourceCache::findCalResource(Resource::CalResourceDesc*
         (entry->mipLevels_ == desc->mipLevels_) && (entry->isAllocSVM_ == desc->isAllocSVM_) &&
         (entry->isAllocExecute_ == desc->isAllocExecute_)) {
       ref = it.second;
+      cacheSize_ -= size;
+      if (entry->type_ == Resource::Local) {
+          lclCacheSize_ -= size;
+      }
       delete it.first;
       // Remove the found etry from the cache
       resCache_.remove(it);
-      cacheSize_ -= size;
       break;
     }
   }
@@ -1823,12 +1829,16 @@ void ResourceCache::removeLast() {
 
   size_t size = getResourceSize(entry.first);
 
+  cacheSize_ -= size;
+  if (entry.first->type_ == Resource::Local) {
+      lclCacheSize_ -= size;
+  }
+
   // Delete CalResourceDesc
   delete entry.first;
 
   // Destroy GSL resource
   entry.second->release();
-  cacheSize_ -= size;
 }
 
 }  // namespace gpu

rocclr/runtime/device/gpu/gpuresource.hpp

@@ -428,7 +428,10 @@ class ResourceCache : public amd::HeapObject {
  public:
   //! Default constructor
   ResourceCache(size_t cacheSizeLimit)
-      : lockCacheOps_("CAL resource cache", true), cacheSize_(0), cacheSizeLimit_(cacheSizeLimit) {}
+      : lockCacheOps_("CAL resource cache", true)
+      , cacheSize_(0)
+      , lclCacheSize_(0)
+      , cacheSizeLimit_(cacheSizeLimit) {}
 
   //! Default destructor
   ~ResourceCache();
@@ -446,6 +449,9 @@ class ResourceCache : public amd::HeapObject {
   //! Destroys cache
   bool free(size_t minCacheEntries = 0);
 
+  //! Returns the size of local memory, stored in the cache
+  size_t lclCacheSize() const { return lclCacheSize_; }
+
  private:
   //! Disable copy constructor
   ResourceCache(const ResourceCache&);
@@ -462,6 +468,7 @@ class ResourceCache : public amd::HeapObject {
   amd::Monitor lockCacheOps_;  //!< Lock to serialise cache access
 
   size_t cacheSize_;       //!< Current cache size in bytes
+  size_t lclCacheSize_;    //!< Local memory stored in the cache
   size_t cacheSizeLimit_;  //!< Cache size limit in bytes
 
   //! CAL resource cache