[L0] Only Override max allocation limits given env

- Change the defaults from always allowing > 4GB allocations to making the user have to request > 4GB
  allocation support given the max allocation allowed on that system is less than 4GB.
- This ensures performance is maintained on systems that dont handle > 4GB allocations natively and
  avoids breaking Ahead of Time (AOT) binaries that were built without > 4GB resource support.
- By setting UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS=1 , the L0 Adapter
  will force the modules to be built with stateless or > 4GB support and
  will allow for the allocations to exceed the max single allocation size limit
  for that device.

Signed-off-by: Spruit, Neil R <neil.r.spruit@intel.com>

source/adapters/level_zero/device.cpp

@@ -268,9 +268,10 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(
  return ReturnValue(uint32_t{64});
  }
  case UR_DEVICE_INFO_MAX_MEM_ALLOC_SIZE:
- // if not optimized for 32-bit access, return total memory size.
- // otherwise, return only maximum allocatable size.
- if (Device->useOptimized32bitAccess() == 0) {
+ // if the user wishes to allocate large allocations on a system that usually
+ // does not allow that allocation size, then we return the max global mem
+ // size as the limit.
+ if (Device->useRelaxedAllocationLimits() == 1) {
  return ReturnValue(uint64_t{calculateGlobalMemSize(Device)});
  } else {
  return ReturnValue(uint64_t{Device->ZeDeviceProperties->maxMemAllocSize});
@@ -923,20 +924,15 @@ ur_device_handle_t_::useImmediateCommandLists() {
  }
 }
 
-int32_t ur_device_handle_t_::useOptimized32bitAccess() {
- static const int32_t Optimize32bitAccessMode = [this] {
- // If device is Intel(R) Data Center GPU Max,
- // use default provided by L0 driver.
- // TODO: Use IP versioning to select based on range of devices
- if (this->isPVC())
- return -1;
- const char *UrRet = std::getenv("UR_L0_USE_OPTIMIZED_32BIT_ACCESS");
+int32_t ur_device_handle_t_::useRelaxedAllocationLimits() {
+ static const int32_t EnableRelaxedAllocationLimits = [this] {
+ const char *UrRet = std::getenv("UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS");
  if (!UrRet)
  return 0;
  return std::atoi(UrRet);
  }();
 
- return Optimize32bitAccessMode;
+ return EnableRelaxedAllocationLimits;
 }
 
 ur_result_t ur_device_handle_t_::initialize(int SubSubDeviceOrdinal,

source/adapters/level_zero/device.hpp

@@ -160,7 +160,7 @@ struct ur_device_handle_t_ : _ur_object {
  // provide support for only one, like for Intel(R)
  // Data Center GPU Max, for which L0 driver only
  // supports stateless.
- int32_t useOptimized32bitAccess();
+ int32_t useRelaxedAllocationLimits();
 
  bool isSubDevice() { return RootDevice != nullptr; }
 

source/adapters/level_zero/program.cpp

@@ -161,7 +161,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urProgramBuildExp(
  ZeBuildOptions += pOptions;
  }
 
- if (phDevices[0]->useOptimized32bitAccess() == 0) {
+ if (phDevices[0]->useRelaxedAllocationLimits() == 1) {
  ZeBuildOptions += " -ze-opt-greater-than-4GB-buffer-required";
  }
 
@@ -256,7 +256,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urProgramCompile(
  // ze-opt-greater-than-4GB-buffer-required to disable
  // stateful optimizations and be able to use larger than
  // 4GB allocations on these kernels.
- if (Context->Devices[0]->useOptimized32bitAccess() == 0) {
+ if (Context->Devices[0]->useRelaxedAllocationLimits() == 1) {
  Program->BuildFlags += " -ze-opt-greater-than-4GB-buffer-required";
  }
  }

source/adapters/level_zero/usm.cpp

@@ -178,7 +178,7 @@ static ur_result_t USMDeviceAllocImpl(void **ResultPtr,
  ZeDesc.flags = 0;
  ZeDesc.ordinal = 0;
 
- if (Device->useOptimized32bitAccess() == 0 &&
+ if (Device->useRelaxedAllocationLimits() == 1 &&
  (Size > Device->ZeDeviceProperties->maxMemAllocSize)) {
  // Tell Level-Zero to accept Size > maxMemAllocSize if
  // large allocations are used.