From 2ef733b40e841e377edb603ad07ea810f25dcef0 Mon Sep 17 00:00:00 2001 From: Vladimir Paramuzov Date: Wed, 7 Aug 2024 15:50:32 +0400 Subject: [PATCH] [GPU] Remove property --- .../openvino/runtime/intel_gpu/properties.hpp | 9 -------- .../impls/ocl/kernel_selector_helper.cpp | 1 - .../kernel_selector/kernel_base_opencl.cpp | 3 +-- .../kernel_selector/kernel_selector_params.h | 1 - .../intel_gpu/src/plugin/program_builder.cpp | 23 ------------------- .../src/runtime/execution_config.cpp | 2 -- .../tests/unit/test_cases/reduce_gpu_test.cpp | 3 ++- 7 files changed, 3 insertions(+), 39 deletions(-) diff --git a/src/inference/include/openvino/runtime/intel_gpu/properties.hpp b/src/inference/include/openvino/runtime/intel_gpu/properties.hpp index 8cfa0bc81c71c3..185195e288805c 100644 --- a/src/inference/include/openvino/runtime/intel_gpu/properties.hpp +++ b/src/inference/include/openvino/runtime/intel_gpu/properties.hpp @@ -123,15 +123,6 @@ static constexpr Property available_device_mem{"AVAILABLE_DEVICE_MEM_SI * @ingroup ov_runtime_ocl_gpu_prop_cpp_api */ static constexpr Property enable_sdpa_optimization{"GPU_ENABLE_SDPA_OPTIMIZATION"}; - -/** - * @brief Turning on this key switches addressing mode to allow allocations larger than 4GB - * as described here - * https://github.com/intel/compute-runtime/blob/master/programmers-guide/ALLOCATIONS_GREATER_THAN_4GB.md#creating-allocations-greater-than-4GB - * Note: On some platform performance of primitives with i64 data type may be lower with this option enabled. - * @ingroup ov_runtime_ocl_gpu_prop_cpp_api - */ -static constexpr Property enable_large_allocations{"GPU_ENABLE_LARGE_ALLOCATIONS"}; } // namespace hint /** diff --git a/src/plugins/intel_gpu/src/graph/impls/ocl/kernel_selector_helper.cpp b/src/plugins/intel_gpu/src/graph/impls/ocl/kernel_selector_helper.cpp index 8cd54607c8ba8d..2ebcebd3b0b48f 100644 --- a/src/plugins/intel_gpu/src/graph/impls/ocl/kernel_selector_helper.cpp +++ b/src/plugins/intel_gpu/src/graph/impls/ocl/kernel_selector_helper.cpp @@ -1120,7 +1120,6 @@ void set_params(const kernel_impl_params& param_info, kernel_selector::params& p params.engineInfo.supports_imad = device_info.supports_imad; params.engineInfo.supports_immad = device_info.supports_immad; params.engineInfo.enable_sub_groups_emulation = true; - params.engineInfo.enable_large_allocations = config.get_property(ov::intel_gpu::hint::enable_large_allocations); params.engineInfo.bOptHintsSupport = false; params.engineInfo.bLocalBlockIOSupport = query_local_block_io_supported(engine, config); diff --git a/src/plugins/intel_gpu/src/kernel_selector/kernel_base_opencl.cpp b/src/plugins/intel_gpu/src/kernel_selector/kernel_base_opencl.cpp index 1811ae30afacbe..dc4daa983e6d79 100644 --- a/src/plugins/intel_gpu/src/kernel_selector/kernel_base_opencl.cpp +++ b/src/plugins/intel_gpu/src/kernel_selector/kernel_base_opencl.cpp @@ -173,8 +173,7 @@ std::shared_ptr KernelBaseOpenCL::GetKernelString(const std::strin kernel_string->undefs = jit.second; if (engine_info.vendor_id == cldnn::INTEL_VENDOR_ID) { kernel_string->options = exe_mode + " -cl-mad-enable"; - if (engine_info.enable_large_allocations) - kernel_string->options += " -cl-intel-greater-than-4GB-buffer-required"; + kernel_string->options += " -cl-intel-greater-than-4GB-buffer-required"; if (engine_info.bOptHintsSupport) kernel_string->options += " -DOPT_HINTS_SUPPORTED=1"; if (engine_info.bLocalBlockIOSupport) diff --git a/src/plugins/intel_gpu/src/kernel_selector/kernel_selector_params.h b/src/plugins/intel_gpu/src/kernel_selector/kernel_selector_params.h index 6730c6c6255ceb..7daccaf99a2fa6 100644 --- a/src/plugins/intel_gpu/src/kernel_selector/kernel_selector_params.h +++ b/src/plugins/intel_gpu/src/kernel_selector/kernel_selector_params.h @@ -390,7 +390,6 @@ struct EngineInfo { bool supports_imad = false; bool supports_immad = false; bool enable_sub_groups_emulation = false; - bool enable_large_allocations = false; bool bOptHintsSupport = false; bool bLocalBlockIOSupport = false; bool supports_microkernels = false; diff --git a/src/plugins/intel_gpu/src/plugin/program_builder.cpp b/src/plugins/intel_gpu/src/plugin/program_builder.cpp index 7e4306fef3f7ef..57e9bfdf444bf9 100644 --- a/src/plugins/intel_gpu/src/plugin/program_builder.cpp +++ b/src/plugins/intel_gpu/src/plugin/program_builder.cpp @@ -2,7 +2,6 @@ // SPDX-License-Identifier: Apache-2.0 // -#include "openvino/core/shape.hpp" #include "openvino/op/constant.hpp" #include "openvino/op/split.hpp" #include "openvino/op/variadic_split.hpp" @@ -17,7 +16,6 @@ #include "intel_gpu/primitives/data.hpp" #include "intel_gpu/op/fully_connected_compressed.hpp" #include "intel_gpu/op/placeholder.hpp" -#include "openvino/runtime/intel_gpu/properties.hpp" #include "openvino/util/pp.hpp" #ifdef __linux__ @@ -28,18 +26,6 @@ #include #endif -namespace { -bool requires_large_allocations(const std::shared_ptr& op, size_t max_alloc_size) { - for (size_t i = 0; i < op->get_output_size(); i++) { - const auto& out_pshape = op->get_output_partial_shape(i); - if (out_pshape.is_static() && ov::shape_size(out_pshape.to_shape()) * op->get_output_element_type(i).size() >= max_alloc_size) { - return true; - } - } - - return false; -} -} // namespace namespace ov { namespace intel_gpu { @@ -168,14 +154,6 @@ std::shared_ptr ProgramBuilder::build(const std::vector ProgramBuilder::build(const std::vector @@ -52,7 +51,6 @@ void ExecutionConfig::set_default() { std::make_tuple(ov::intel_gpu::hint::queue_throttle, ov::intel_gpu::hint::ThrottleLevel::MEDIUM), std::make_tuple(ov::intel_gpu::hint::queue_priority, ov::hint::Priority::MEDIUM), std::make_tuple(ov::intel_gpu::hint::enable_sdpa_optimization, true), - std::make_tuple(ov::intel_gpu::hint::enable_large_allocations, false), std::make_tuple(ov::intel_gpu::enable_loop_unrolling, true), std::make_tuple(ov::intel_gpu::disable_winograd_convolution, false), std::make_tuple(ov::internal::exclusive_async_requests, false), diff --git a/src/plugins/intel_gpu/tests/unit/test_cases/reduce_gpu_test.cpp b/src/plugins/intel_gpu/tests/unit/test_cases/reduce_gpu_test.cpp index a4526b73d3f1be..d835af19fdb858 100644 --- a/src/plugins/intel_gpu/tests/unit/test_cases/reduce_gpu_test.cpp +++ b/src/plugins/intel_gpu/tests/unit/test_cases/reduce_gpu_test.cpp @@ -2215,9 +2215,10 @@ TEST(reduce_f32_fw_gpu, large_buffer) { layout in_l = { sz_8gb, data_types::f32, format::bfyx }; + auto config = get_test_default_config(*engine); topology topology(input_layout("input", in_l), reduce("reduce", input_info("input"), reduce_mode::mean, {2}, true)); - network network(*engine, topology, get_test_default_config(*engine)); + network network(*engine, topology, config); auto input = network.get_output_memory("input"); { mem_lock l(input, get_test_stream());