diff options
| author | Georgios Pinitas <georgios.pinitas@arm.com> | 2018-03-19 19:57:01 +0000 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:49:16 +0000 |
| commit | ceff0f9a991b693f568c25b1e0933582301082e7 (patch) | |
| tree | 3938e9b2151fc6b3d42d2bab9f66efa4709a5da1 /src/runtime/CL/CLTensorAllocator.cpp | |
| parent | 7db3a30cd15659af75938cd964eebb17a12c6e81 (diff) | |
| download | ComputeLibrary-ceff0f9a991b693f568c25b1e0933582301082e7.tar.gz | |
COMPMID-1016: Optimize kernel reconfiguration
Optimizes kernel reconfiguration when memory manager is used.
Note that this works only if every sub-sequent reconfigurations leads to
sizes less than the first one.
Change-Id: I08898e99929c3756147a02979b726c2380b6e11d
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/125114
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/runtime/CL/CLTensorAllocator.cpp')
| -rw-r--r-- | src/runtime/CL/CLTensorAllocator.cpp | 9 |
1 files changed, 2 insertions, 7 deletions
diff --git a/src/runtime/CL/CLTensorAllocator.cpp b/src/runtime/CL/CLTensorAllocator.cpp index 705c4edd60..c5524b1ccb 100644 --- a/src/runtime/CL/CLTensorAllocator.cpp +++ b/src/runtime/CL/CLTensorAllocator.cpp @@ -24,7 +24,6 @@ #include "arm_compute/runtime/CL/CLTensorAllocator.h" #include "arm_compute/core/Error.h" -#include "arm_compute/core/Log.h" #include "arm_compute/core/TensorInfo.h" #include "arm_compute/runtime/CL/CLMemoryGroup.h" #include "arm_compute/runtime/CL/CLScheduler.h" @@ -58,11 +57,7 @@ void *SVMMemory::allocate(cl_context context, size_t size, cl_svm_mem_flags flag ARM_COMPUTE_ERROR_ON(_ptr != nullptr); ARM_COMPUTE_ERROR_ON(size > CL_DEVICE_MAX_MEM_ALLOC_SIZE); _ptr = clSVMAlloc(context, flags, size, alignment); - if(_ptr == nullptr) - { - ARM_COMPUTE_LOG_INFO_MSG_CORE("Call to clSVMAlloc() failed."); - } - else + if(_ptr != nullptr) { _size = size; _fine_grain = static_cast<bool>(flags & CL_MEM_SVM_FINE_GRAIN_BUFFER); @@ -76,9 +71,9 @@ void *CLTensorAllocator::svm_ptr() void CLTensorAllocator::allocate() { - ARM_COMPUTE_ERROR_ON(_buffer.get() != nullptr); if(_associated_memory_group == nullptr) { + ARM_COMPUTE_ERROR_ON(_buffer.get() != nullptr); if(_svm_memory.allocate(CLScheduler::get().context()(), CL_MEM_READ_WRITE | CL_MEM_SVM_FINE_GRAIN_BUFFER, info().total_size(), 0) == nullptr) { // try at coarse grain svm memory |