P4 to Git Change 1062136 by jatang@jatang-opencl-hsa-stg2 on 2014/08/01 14:19:58

EPR #402950 - Properly align the scratch buffer to 64K.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#452 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuresource.cpp#189 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/include/cal/calcl.h#29 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp#95 edit


[ROCm/clr commit: 268ff9830e]

projects/clr/rocclr/runtime/device/gpu/gpudevice.cpp

@@ -2289,9 +2289,8 @@ Device::allocScratch(uint regNum, const VirtualGPU* vgpu)
             ScopedLockVgpus lock(*this);
             std::vector<Memory*>& mems = scratch_[s]->memObjs_;
 
-            // Calculate the size of the new buffer +
-            // (64 Ki) for alignment with generic address space
-            size_t size = calcScratchBufferSize(regNum) + 64 * Ki;
+            // Calculate the size of the new buffer
+            size_t size = calcScratchBufferSize(regNum);
 
             scratch_[s]->destroyMemory();
 

projects/clr/rocclr/runtime/device/gpu/gpuresource.cpp

@@ -346,6 +346,7 @@ Resource::create(MemoryType memType, CreateParams* params, bool heap)
     bool useRowPitch = false;
 
     desc.vaBase = 0;
+    desc.minAlignment = 0;
     desc.section = GSL_SECTION_REGULAR;
     if (NULL != params && NULL != params->owner_) {   //make sure params not NULL
         mcaddr svmPtr = reinterpret_cast<mcaddr>(params->owner_->getSvmPtr());
@@ -453,7 +454,12 @@ Resource::create(MemoryType memType, CreateParams* params, bool heap)
             // Check resource cache first for an appropriate resource
             gslRef_ = dev().resourceCache().findCalResource(&cal_);
             if (memType == Scratch) {
-                desc.vaBase = static_cast<mcaddr>(0x100000000ULL);
+                if ((dev().settings().hsail_) || (dev().settings().oclVersion_ >= OpenCL20)) {
+                    desc.minAlignment = 64 * Ki;
+                }
+                else {
+                    desc.vaBase = static_cast<mcaddr>(0x100000000ULL);
+                }
             }
             else if ((gslRef_ != NULL) && (!dev().settings().use64BitPtr_)) {
                 // Make sure runtime didn't pick a resource with > 4GB address

projects/clr/rocclr/runtime/device/gpu/gslbe/src/include/cal/calcl.h

@@ -145,6 +145,7 @@ typedef struct CALresourceDescRec {
     CALuint64       busAddress[2];
     mcaddr          vaBase;
     gslMemObjectAttribSection section;
+    CALuint         minAlignment;
 } CALresourceDesc;
 
 typedef enum CALresallocsliceviewflagsRec {

projects/clr/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp

@@ -741,6 +741,7 @@ CALGSLDevice::resAlloc(const CALresourceDesc* desc) const
     attribs.location = desc->type;
     attribs.vaBase   = desc->vaBase;
     attribs.section = desc->section;
+    attribs.minAlignment = desc->minAlignment;
 
     //!@note GSL asserts with tiled 1D images of any type.
     if ((desc->dimension == GSL_MOA_BUFFER) ||