P4 to Git Change 1084962 by gandryey@gera-dev-w7 on 2014/10/07 14:25:46

ECR #304775 - Remove GPU_PREALLOC_ADDR_SPACE key
	- Previously the key was used for 64bit ISA testing on boards with <4GB frame buffer

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudefs.hpp#114 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#467 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.cpp#289 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.hpp#88 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#215 edit

rocclr/runtime/device/gpu/gpudefs.hpp

@@ -57,9 +57,6 @@ const static uint CalOfflineImpl = 0xffffffff;
 //! Alignment restriciton for the pinned memory
 const static size_t PinnedMemoryAlignment = 4 * Ki;
 
-//! Reserved address space
-const static cl_ulong ReservedAdressSpaceSize = static_cast<cl_ulong>(4) * Gi;
-
 //! Defines all supported ASIC families
 enum AsicFamilies {
     Family7xx,

rocclr/runtime/device/gpu/gpudevice.cpp

@@ -520,14 +520,6 @@ void Device::fillDeviceInfo(
             info_.globalMemSize_   +=
                 (static_cast<cl_ulong>(calAttr.uncachedRemoteRAM) * Mi) / 2;
         }
-        // Check if runtime has to reserve address space for testing
-        if (settings().use64BitPtr_ && settings().preallocAddrSpace_ &&
-            (info_.globalMemSize_ > ReservedAdressSpaceSize)) {
-            info_.globalMemSize_ -= ReservedAdressSpaceSize;
-        }
-        else {
-            reinterpret_cast<gpu::Settings*>(settings_)->preallocAddrSpace_ = false;
-        }
 
         // We try to calculate the largest available memory size from
         // the largest available block in either heap.  In theory this
@@ -954,9 +946,6 @@ Device::initializeHeapResources()
         }
 
         size_t dummySize = amd::Os::pageSize();
-        if (heap()->isVirtual() && settings().preallocAddrSpace_) {
-            dummySize = static_cast<size_t>(ReservedAdressSpaceSize - Mi);
-        }
 
         // Allocate a dummy page for NULL pointer processing
         dummyPage_ = new(*context_) amd::Buffer(*context_, 0, dummySize);

rocclr/runtime/device/gpu/gpusettings.cpp

@@ -110,9 +110,6 @@ Settings::Settings()
     minWorkloadTime_ = 1;       // 0.1 ms
     maxWorkloadTime_ = 5000;    // 500 ms
 
-    // Preallocates address space
-    preallocAddrSpace_ = GPU_PREALLOC_ADDR_SPACE;
-
     // Controls tiled images in persistent
     //!@note IOL for Linux doesn't setup tiling aperture in CMM/QS
     linearPersistentImage_ = false;

rocclr/runtime/device/gpu/gpusettings.hpp

@@ -63,7 +63,6 @@ public:
             uint    viPlus_: 1;         //!< VI and post VI features
             uint    rectLinearDMA_: 1;  //!< Rectangular linear DRMDMA support
             uint    threadTraceEnable_: 1;  //!< Thread trace enable
-            uint    preallocAddrSpace_: 1;  //!< Preallocates address space
             uint    linearPersistentImage_: 1;  //!< Allocates linear images in persistent
             uint    useSingleScratch_: 1;   //!< Allocates single scratch per device
             uint    sdmaProfiling_: 1;  //!< Enables SDMA profiling
@@ -74,7 +73,7 @@ public:
             uint    asyncMemCopy_: 1;   //!< Use async memory transfers
             uint    hsailDirectSRD_: 1; //!< Controls direct SRD for HSAIL
             uint    useDeviceQueue_: 1; //!< Submit to separate device queue
-            uint    reserved_: 1;
+            uint    reserved_: 2;
         };
         uint    value_;
     };

rocclr/runtime/utils/flags.hpp

@@ -102,8 +102,6 @@ release(bool, GPU_FORCE_64BIT_PTR, 0,                                         \
         "Forces 64 bit pointers on GPU")                                      \
 release(bool, GPU_FORCE_OCL20_32BIT, 0,                                       \
         "Forces 32 bit apps to take CLANG\HSAIL path")                        \
-release(bool, GPU_PREALLOC_ADDR_SPACE, 0,                                     \
-        "Preallocates 4GB address space. Valid for boards > 4GB")             \
 release(bool, GPU_RAW_TIMESTAMP, 0,                                           \
         "Reports GPU raw timestamps in GPU timeline")                         \
 release(bool, CPU_IMAGE_SUPPORT, true,                                        \