Raise large scratch allocation limit for RCCL.

Temporary workaround for 2.10 release.  RCCL, compiler, or firmware
must be corrected and this code reverted before another ASIC release.

Change-Id: I27851353289b93df9acb72d28b8c6ccb9f7f7d7a


[ROCm/ROCR-Runtime commit: 35c1ffa863]

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp

@@ -958,7 +958,8 @@ void GpuAgent::AcquireQueueScratch(ScratchInfo& scratch) {
   ScopedAcquire<KernelMutex> lock(&scratch_lock_);
   // Limit to 1/8th of scratch pool for small scratch and 1/4 of that for a single queue.
   size_t small_limit = scratch_pool_.size() >> 3;
-  size_t single_limit = small_limit >> 2;
+  // Lift limit for 2.10 release RCCL workaround.
+  size_t single_limit = 146800640; //small_limit >> 2;
   bool large = (scratch.size > single_limit) ||
       (scratch_pool_.size() - scratch_pool_.remaining() + scratch.size > small_limit);
   large = (isa_->GetMajorVersion() < 8) ? false : large;