SWDEV-301667 - Increase kern arg pool

Change-Id: Ie4b087ae4aec08fccaaa7958cdf545e4e27ac5c1


[ROCm/clr commit: 6a77f73050]

projects/clr/rocclr/device/rocm/rocvirtual.hpp

@@ -511,7 +511,7 @@ class VirtualGPU : public device::VirtualDevice {
   HwQueueTracker  barriers_;      //!< Tracks active barriers in ROCr
 
   //!< The number of chunks the kernel arg pool will be divided
-  static constexpr uint32_t KernelArgPoolNumSignal = 8;
+  static constexpr uint32_t KernelArgPoolNumSignal = 4;
   address   kernarg_pool_base_;
   uint32_t  kernarg_pool_size_;
   uint32_t  kernarg_pool_chunk_end_;    //!< The end offset of the current chunck

projects/clr/rocclr/utils/flags.hpp

@@ -134,7 +134,7 @@ release(uint, OPENCL_VERSION, (IS_BRAHMA ? 120 : 200),                        \
         "Force GPU opencl verison")                                           \
 release(bool, HSA_LOCAL_MEMORY_ENABLE, true,                                  \
         "Enable HSA device local memory usage")                               \
-release(uint, HSA_KERNARG_POOL_SIZE, 512 * 1024,                              \
+release(uint, HSA_KERNARG_POOL_SIZE, 1024 * 1024,                             \
         "Kernarg pool size")                                                  \
 release(bool, HSA_ENABLE_COARSE_GRAIN_SVM, true,                              \
         "Enable device memory for coarse grain SVM allocations")              \