Home
last modified time | relevance | path

Searched refs:gpu_info (Results 1 – 25 of 161) sorted by relevance

1234567

/external/tensorflow/tensorflow/lite/delegates/gpu/common/selectors/default/
Dconvolution_selector.cc38 const GpuInfo& gpu_info, const OperationDef& op_def, in SelectConvolutionAdreno() argument
40 if (IsConvConstantsSupported(gpu_info, op_def, attr)) { in SelectConvolutionAdreno()
41 GPUOperation conv = CreateConvConstants(gpu_info, op_def, attr); in SelectConvolutionAdreno()
44 ConvPowerVR conv = CreateConvPowerVR(gpu_info, op_def, attr, &dst_shape); in SelectConvolutionAdreno()
51 const GpuInfo& gpu_info, const OperationDef& op_def, in SelectConvolutionWinogradAdreno() argument
54 CreateConvPowerVRWino4x4To6x6(gpu_info, op_def, attr, &dst_shape); in SelectConvolutionWinogradAdreno()
60 const BHWC& dst_shape, const GpuInfo& gpu_info, in SelectConvolutionDynamicWeightsAdreno() argument
64 gpu_info, op_def, attr, weights_shape, &dst_shape); in SelectConvolutionDynamicWeightsAdreno()
71 const GpuInfo& gpu_info, const OperationDef& op_def) { in SelectConvolutionNVidia() argument
72 if (IsConvConstantsSupported(gpu_info, op_def, attr)) { in SelectConvolutionNVidia()
[all …]
Dconvolution_transposed_selector.cc29 const ConvolutionTransposedAttributes& attr, const GpuInfo& gpu_info, in SelectConvolutionTransposedAdreno() argument
33 CreateConvolutionTransposedThin(gpu_info, op_def, attr); in SelectConvolutionTransposedAdreno()
37 CreateConvolutionTransposed3x3Thin(gpu_info, op_def, attr); in SelectConvolutionTransposedAdreno()
41 CreateConvolutionTransposed(gpu_info, op_def, attr); in SelectConvolutionTransposedAdreno()
47 const ConvolutionTransposedAttributes& attr, const GpuInfo& gpu_info, in SelectConvolutionTransposedPowerVR() argument
51 CreateConvolutionTransposedThin(gpu_info, op_def, attr); in SelectConvolutionTransposedPowerVR()
55 CreateConvolutionTransposed3x3Thin(gpu_info, op_def, attr); in SelectConvolutionTransposedPowerVR()
59 CreateConvolutionTransposed3x3(gpu_info, op_def, attr); in SelectConvolutionTransposedPowerVR()
63 CreateConvolutionTransposed4x4(gpu_info, op_def, attr); in SelectConvolutionTransposedPowerVR()
67 CreateConvolutionTransposed(gpu_info, op_def, attr); in SelectConvolutionTransposedPowerVR()
[all …]
Ddw_convolution_selector.cc27 const DepthwiseConvolution2DAttributes& attr, const GpuInfo& gpu_info, in SelectDWConvolutionAdreno() argument
31 CreateDepthwiseConv3x3(gpu_info, op_def, attr)); in SelectDWConvolutionAdreno()
34 CreateDepthwiseConvolution2D(gpu_info, op_def, attr)); in SelectDWConvolutionAdreno()
39 const DepthwiseConvolution2DAttributes& attr, const GpuInfo& gpu_info, in SelectDWConvolutionPowerVR() argument
43 CreateDepthwiseConv3x3(gpu_info, op_def, attr)); in SelectDWConvolutionPowerVR()
46 CreateDepthwiseConvolution2D(gpu_info, op_def, attr)); in SelectDWConvolutionPowerVR()
51 const DepthwiseConvolution2DAttributes& attr, const GpuInfo& gpu_info, in SelectDWConvolutionMali() argument
56 const MaliInfo mali_info = gpu_info.mali_info; in SelectDWConvolutionMali()
60 CreateDepthwiseConv3x3(gpu_info, op_def, attr)); in SelectDWConvolutionMali()
63 CreateDepthwiseConvolution2D(gpu_info, op_def, attr)); in SelectDWConvolutionMali()
[all …]
Dfully_connected_selector.cc28 const FullyConnectedAttributes& attr, const GpuInfo& gpu_info, in SelectFullyConnectedGeneric() argument
32 ConvPowerVR conv = CreateConvPowerVR(gpu_info, op_def, attr, &dst_shape); in SelectFullyConnectedGeneric()
35 FullyConnected fc = CreateFullyConnected(gpu_info, op_def, attr); in SelectFullyConnectedGeneric()
41 const FullyConnectedAttributes& attr, const GpuInfo& gpu_info, in SelectFullyConnectedAdreno() argument
45 ConvPowerVR conv = CreateConvPowerVR(gpu_info, op_def, attr, &dst_shape); in SelectFullyConnectedAdreno()
48 FullyConnected fc = CreateFullyConnected(gpu_info, op_def, attr); in SelectFullyConnectedAdreno()
54 const FullyConnectedAttributes& attr, const GpuInfo& gpu_info, in SelectFullyConnectedPowerVR() argument
57 ConvPowerVR conv = CreateConvPowerVR(gpu_info, op_def, attr); in SelectFullyConnectedPowerVR()
60 FullyConnected fc = CreateFullyConnected(gpu_info, op_def, attr); in SelectFullyConnectedPowerVR()
66 const FullyConnectedAttributes& attr, const GpuInfo& gpu_info, in SelectFullyConnectedMali() argument
[all …]
/external/tensorflow/tensorflow/lite/delegates/gpu/common/task/
Dstorage_type_util.cc26 bool CanCreateTensorWithShape(const GpuInfo& gpu_info, const BHWDC& shape, in CanCreateTensorWithShape() argument
35 return buffer_size <= gpu_info.GetMaxBufferSize(); in CanCreateTensorWithShape()
39 gpu_info.GetMaxImageBufferWidth(); in CanCreateTensorWithShape()
41 if (gpu_info.opencl_info.cl_version < OpenClVersion::kCl1_2 && in CanCreateTensorWithShape()
47 return shape.w * shape.b <= gpu_info.GetMaxImage3DWidth() && in CanCreateTensorWithShape()
48 shape.h <= gpu_info.GetMaxImage3DHeight() && in CanCreateTensorWithShape()
49 slices * shape.d <= gpu_info.GetMaxImage3DDepth(); in CanCreateTensorWithShape()
52 if (slices == 1 && gpu_info.IsAdreno() && in CanCreateTensorWithShape()
53 !gpu_info.adreno_info.support_one_layer_texture_array) { in CanCreateTensorWithShape()
56 return shape.w * shape.b <= gpu_info.GetMaxImage2DWidth() && in CanCreateTensorWithShape()
[all …]
Dwork_group_picking.cc37 const GpuInfo& gpu_info, WorkGroupSizeAlignment z_alignment) { in GenerateWorkGroupSizesXYMultipleOf() argument
54 if (x <= gpu_info.GetMaxWorkGroupSizeForX() && in GenerateWorkGroupSizesXYMultipleOf()
55 y <= gpu_info.GetMaxWorkGroupSizeForY() && in GenerateWorkGroupSizesXYMultipleOf()
56 z <= gpu_info.GetMaxWorkGroupSizeForZ()) { in GenerateWorkGroupSizesXYMultipleOf()
67 const GpuInfo& gpu_info, WorkGroupSizeAlignment z_alignment) { in GenerateWorkGroupSizesXMultipleOf() argument
80 if (x <= gpu_info.GetMaxWorkGroupSizeForX() && in GenerateWorkGroupSizesXMultipleOf()
81 y <= gpu_info.GetMaxWorkGroupSizeForY() && in GenerateWorkGroupSizesXMultipleOf()
82 z <= gpu_info.GetMaxWorkGroupSizeForZ() && in GenerateWorkGroupSizesXMultipleOf()
92 void GetWorkGroupsAlignedToGrid(const GpuInfo& gpu_info, in GetWorkGroupsAlignedToGrid() argument
96 max_wg_size.x = gpu_info.GetMaxWorkGroupSizeForX(); in GetWorkGroupsAlignedToGrid()
[all …]
/external/tensorflow/tensorflow/lite/delegates/gpu/common/tasks/
Dfully_connected.cc34 bool UseBufferForWeights(const GpuInfo& gpu_info) { in UseBufferForWeights() argument
35 return gpu_info.IsAdreno() || gpu_info.IsAMD() || gpu_info.IsMali() || in UseBufferForWeights()
36 gpu_info.IsApple(); in UseBufferForWeights()
41 const GpuInfo& gpu_info) in FullyConnected() argument
43 if (gpu_info.IsAdreno()) { in FullyConnected()
44 if (gpu_info.adreno_info.IsAdreno3xx()) { in FullyConnected()
46 } else if (gpu_info.adreno_info.IsAdreno4xx()) { in FullyConnected()
51 } else if (gpu_info.IsIntel() || gpu_info.IsNvidia() || in FullyConnected()
52 gpu_info.IsPowerVR() || gpu_info.IsApple()) { in FullyConnected()
57 code_ = GetFullyConnectedKernelCode(definition_, gpu_info); in FullyConnected()
[all …]
Dmean_stddev_normalization.cc33 std::string GetReduceCode(const GpuInfo& gpu_info, int reduction_size) { in GetReduceCode() argument
46 if (gpu_info.IsApiOpenCl()) { in GetReduceCode()
91 const GpuInfo& gpu_info, in MeanStdDevNormalization() argument
100 std::min(tensor_slices, gpu_info.GetMaxWorkGroupSizeForX()); in MeanStdDevNormalization()
101 if (gpu_info.IsMali()) { in MeanStdDevNormalization()
107 if (gpu_info.IsAdreno()) { in MeanStdDevNormalization()
108 AdrenoInfo info = gpu_info.adreno_info; in MeanStdDevNormalization()
131 if (gpu_info.IsPowerVR()) { in MeanStdDevNormalization()
134 if (gpu_info.IsApple()) { in MeanStdDevNormalization()
143 code_ = GetNormalizationCode(gpu_info); in MeanStdDevNormalization()
[all …]
Dconvolution_transposed_4x4.cc29 const GpuInfo& gpu_info) { in GetBestWeightsUploadType() argument
32 if (gpu_info.IsPowerVR()) { in GetBestWeightsUploadType()
35 } else if (gpu_info.IsNvidia() || gpu_info.IsIntel()) { in GetBestWeightsUploadType()
38 } else if (gpu_info.IsAMD()) { in GetBestWeightsUploadType()
50 const OperationDef& definition, const GpuInfo& gpu_info) in ConvolutionTransposed4x4() argument
53 if (gpu_info.IsApple()) { in ConvolutionTransposed4x4()
57 if (gpu_info.IsApple()) { in ConvolutionTransposed4x4()
63 code_ = GenerateConvolutionTransposedCode(gpu_info, definition_, in ConvolutionTransposed4x4()
64 GetBestWeightsUploadType(gpu_info)); in ConvolutionTransposed4x4()
66 gpu_info.IsPowerVR()) { in ConvolutionTransposed4x4()
[all …]
Dconv_powervr.h44 TuningType tuning_type, const GpuInfo& gpu_info,
111 const Convolution2DAttributes& attr, const GpuInfo& gpu_info,
115 const GpuInfo& gpu_info, const BHWC* dst_shape = nullptr);
117 const FullyConnectedAttributes& attr, const GpuInfo& gpu_info,
121 const Convolution3DAttributes& attr, const GpuInfo& gpu_info,
124 void GenerateCode(const GpuInfo& gpu_info);
142 friend ConvPowerVR CreateConvPowerVR(const GpuInfo& gpu_info,
147 friend ConvPowerVR CreateConvPowerVR(const GpuInfo& gpu_info,
153 const GpuInfo& gpu_info, const OperationDef& definition,
158 const GpuInfo& gpu_info, const OperationDef& definition,
[all …]
Dwinograd.h35 TuningType tuning_type, const GpuInfo& gpu_info, in GetPossibleKernelWorkGroups() argument
65 const Padding2D& padding, const GpuInfo& gpu_info);
69 TuningType tuning_type, const GpuInfo& gpu_info,
81 const GpuInfo& gpu_info, const OperationDef& definition,
95 const GpuInfo& gpu_info, const OperationDef& definition,
102 TuningType tuning_type, const GpuInfo& gpu_info, in GetPossibleKernelWorkGroups() argument
131 const GpuInfo& gpu_info);
135 TuningType tuning_type, const GpuInfo& gpu_info,
148 const GpuInfo& gpu_info, const OperationDef& definition,
160 const GpuInfo& gpu_info, const OperationDef& definition,
Dconv_powervr.cc172 const GpuInfo& gpu_info, const BHWC* dst_shape) in ConvPowerVR() argument
178 conv_params_(GuessBestParams(gpu_info, definition, attr, dst_shape)) {} in ConvPowerVR()
182 const BHWC& weights_shape, const GpuInfo& gpu_info, in ConvPowerVR() argument
189 conv_params_(GuessBestParams(gpu_info, definition, attr, weights_shape, in ConvPowerVR()
194 const GpuInfo& gpu_info, const BHWC* dst_shape) in ConvPowerVR() argument
200 conv_params_(GuessBestParams(gpu_info, definition, attr, dst_shape)) {} in ConvPowerVR()
219 const GpuInfo& gpu_info, const BHWDC* dst_shape) in ConvPowerVR() argument
227 conv_params_(GuessBestParams(gpu_info, definition, attr, dst_shape)) {} in ConvPowerVR()
241 void ConvPowerVR::GenerateCode(const GpuInfo& gpu_info) { in GenerateCode() argument
249 code_ = GenerateConv(gpu_info, definition_, stride_correction, conv_params_); in GenerateCode()
[all …]
Dconvolution_transposed_3x3.h41 TuningType tuning_type, const GpuInfo& gpu_info,
70 const GpuInfo& gpu_info, int2 padding);
72 const GpuInfo& gpu_info, const OperationDef& definition,
75 const GpuInfo& gpu_info, const OperationDef& definition,
84 const GpuInfo& gpu_info, const OperationDef& op_def,
98 const GpuInfo& gpu_info, const OperationDef& definition,
102 const GpuInfo& gpu_info, const OperationDef& definition,
Dconvolution_transposed_4x4.h41 TuningType tuning_type, const GpuInfo& gpu_info, in GetPossibleKernelWorkGroups() argument
72 const GpuInfo& gpu_info);
75 const GpuInfo& gpu_info, const OperationDef& definition,
78 const GpuInfo& gpu_info, const OperationDef& definition,
88 const GpuInfo& gpu_info, const OperationDef& op_def,
99 const GpuInfo& gpu_info, const OperationDef& definition,
103 const GpuInfo& gpu_info, const OperationDef& definition,
Dconvolution_transposed.cc34 const GpuInfo& gpu_info, bool weights_are_buffer) in ConvolutionTransposed() argument
39 if (gpu_info.IsApple()) { in ConvolutionTransposed()
45 if (gpu_info.IsApple()) { in ConvolutionTransposed()
52 if (gpu_info.IsMali()) { in ConvolutionTransposed()
53 if (gpu_info.mali_info.IsMidgard()) { in ConvolutionTransposed()
61 if (!gpu_info.IsMali()) { in ConvolutionTransposed()
73 code_ = GenerateConvolutionTransposedCode(definition_, gpu_info, in ConvolutionTransposed()
79 const ConvolutionTransposed3DAttributes& attr, const GpuInfo& gpu_info, in ConvolutionTransposed() argument
85 if (gpu_info.IsApple()) { in ConvolutionTransposed()
91 if (gpu_info.IsApple()) { in ConvolutionTransposed()
[all …]
/external/tensorflow/tensorflow/lite/delegates/gpu/cl/
Denvironment.cc51 bool IsGpuSupportsStorageType(const GpuInfo& gpu_info, in IsGpuSupportsStorageType() argument
55 return !gpu_info.IsAMD(); in IsGpuSupportsStorageType()
59 return !gpu_info.IsAMD() && gpu_info.SupportsTextureArray(); in IsGpuSupportsStorageType()
61 return (gpu_info.IsAdreno() || gpu_info.IsAMD() || gpu_info.IsNvidia()) && in IsGpuSupportsStorageType()
62 gpu_info.SupportsImageBuffer(); in IsGpuSupportsStorageType()
64 return !gpu_info.IsAMD() && gpu_info.SupportsImage3D(); in IsGpuSupportsStorageType()
73 bool IsGpuSupportsPrecision(const GpuInfo& gpu_info, in IsGpuSupportsPrecision() argument
78 return gpu_info.SupportsFP16(); in IsGpuSupportsPrecision()
187 TensorStorageType GetFastestStorageType(const GpuInfo& gpu_info) { in GetFastestStorageType() argument
188 if (gpu_info.IsAdreno()) { in GetFastestStorageType()
[all …]
/external/tensorflow/tensorflow/lite/delegates/gpu/metal/
Dmetal_device.cc26 GpuInfo gpu_info; in CreateGpuInfoFromMetalDevice() local
27 GetGpuInfoFromDeviceDescription(device_name, GpuApi::kMetal, &gpu_info); in CreateGpuInfoFromMetalDevice()
31 gpu_info.metal_info.max_work_group_size_x = threadsPerGroup.width; in CreateGpuInfoFromMetalDevice()
32 gpu_info.metal_info.max_work_group_size_y = threadsPerGroup.height; in CreateGpuInfoFromMetalDevice()
33 gpu_info.metal_info.max_work_group_size_z = threadsPerGroup.depth; in CreateGpuInfoFromMetalDevice()
35 gpu_info.metal_info.max_work_group_size_x = 256; in CreateGpuInfoFromMetalDevice()
36 gpu_info.metal_info.max_work_group_size_y = 256; in CreateGpuInfoFromMetalDevice()
37 gpu_info.metal_info.max_work_group_size_z = 64; in CreateGpuInfoFromMetalDevice()
41 gpu_info.metal_info.buffer_max_size = [device maxBufferLength]; in CreateGpuInfoFromMetalDevice()
44 gpu_info.metal_info.buffer_max_size = 256 * 1024 * 1024; in CreateGpuInfoFromMetalDevice()
[all …]
/external/tensorflow/tensorflow/lite/delegates/gpu/common/tasks/special/
Dfc_fc_add.cc31 bool UseBufferForWeights(const GpuInfo& gpu_info) { in UseBufferForWeights() argument
32 return gpu_info.IsAdreno() || gpu_info.IsAMD() || gpu_info.IsMali(); in UseBufferForWeights()
36 FCFCAdd::FCFCAdd(const OperationDef& definition, const GpuInfo& gpu_info) in FCFCAdd() argument
38 if (gpu_info.IsAdreno()) { in FCFCAdd()
39 if (gpu_info.adreno_info.IsAdreno3xx()) { in FCFCAdd()
41 } else if (gpu_info.adreno_info.IsAdreno4xx()) { in FCFCAdd()
46 } else if (gpu_info.IsIntel()) { in FCFCAdd()
48 } else if (gpu_info.IsNvidia()) { in FCFCAdd()
50 } else if (gpu_info.IsPowerVR()) { in FCFCAdd()
55 code_ = GetFCFCAddKernelCode(definition_, gpu_info); in FCFCAdd()
[all …]
/external/tensorflow/tensorflow/lite/delegates/gpu/gl/workgroups/
Dcalculator_from_metadata.cc43 const GpuInfo& gpu_info) in WorkgroupsCalculatorFromMetadata() argument
44 : WorkgroupsCalculator(gpu_info), in WorkgroupsCalculatorFromMetadata()
45 default_calculator_(NewDefaultWorkgroupsCalculator(gpu_info)) { in WorkgroupsCalculatorFromMetadata()
70 const data::CustomWorkgroups& workgroups, const GpuInfo& gpu_info) { in FindWorkgroups() argument
72 if (workgroup->gpu_info()->c_str() == gpu_info.opengl_info.renderer_name) { in FindWorkgroups()
82 const uint8_t* metadata, const GpuInfo& gpu_info) { in NewWorkgroupsCalculatorFromMetadata() argument
88 FindWorkgroups(*flow_metadata->workgroups(), gpu_info); in NewWorkgroupsCalculatorFromMetadata()
91 gpu_info); in NewWorkgroupsCalculatorFromMetadata()
97 const uint8_t* metadata, const GpuInfo& gpu_info) {
Ddefault_calculator.cc29 explicit DefaultWorkgroupsCalculator(const GpuInfo& gpu_info) in DefaultWorkgroupsCalculator() argument
30 : WorkgroupsCalculator(gpu_info) {} in DefaultWorkgroupsCalculator()
57 explicit WorkgroupsCalculatorForMali(const GpuInfo& gpu_info) in WorkgroupsCalculatorForMali() argument
58 : WorkgroupsCalculator(gpu_info) {} in WorkgroupsCalculatorForMali()
83 const GpuInfo& gpu_info) { in NewDefaultWorkgroupsCalculator() argument
84 if (gpu_info.IsMali()) { in NewDefaultWorkgroupsCalculator()
85 return absl::make_unique<WorkgroupsCalculatorForMali>(gpu_info); in NewDefaultWorkgroupsCalculator()
87 return absl::make_unique<DefaultWorkgroupsCalculator>(gpu_info); in NewDefaultWorkgroupsCalculator()
Dcalculator.cc31 void MaybeShrinkWorkgroup(const GpuInfo& gpu_info, uint3* wg) { in MaybeShrinkWorkgroup() argument
32 while (wg->x > gpu_info.GetMaxWorkGroupSizeForX()) { in MaybeShrinkWorkgroup()
36 while (wg->y > gpu_info.GetMaxWorkGroupSizeForY()) { in MaybeShrinkWorkgroup()
40 while (wg->z > gpu_info.GetMaxWorkGroupSizeForZ()) { in MaybeShrinkWorkgroup()
51 while (CalculateProduct(*wg) > gpu_info.GetMaxWorkGroupTotalSize()) { in MaybeShrinkWorkgroup()
61 WorkgroupsCalculator::WorkgroupsCalculator(const GpuInfo& gpu_info) in WorkgroupsCalculator() argument
62 : gpu_info_{gpu_info} {} in WorkgroupsCalculator()
/external/tensorflow/tensorflow/lite/delegates/gpu/gl/kernels/
Dmean.cc35 bool UseSubgroupBasedImpl(const GpuInfo& gpu_info) { in UseSubgroupBasedImpl() argument
36 return gpu_info.IsApiVulkan() && in UseSubgroupBasedImpl()
37 (gpu_info.vulkan_info.api_version_major > 1 || in UseSubgroupBasedImpl()
38 gpu_info.vulkan_info.api_version_minor >= 1) && in UseSubgroupBasedImpl()
39 gpu_info.vulkan_info.subgroup_size >= 32 && in UseSubgroupBasedImpl()
40 gpu_info.vulkan_info.supports_subgroup_arithmetic; in UseSubgroupBasedImpl()
98 const uint32_t subgroup_size = ctx.gpu_info->vulkan_info.subgroup_size; in GenerateSubgroupBasedMean()
99 const uint32_t max_wg_size_x = ctx.gpu_info->GetMaxWorkGroupSizeForX(); in GenerateSubgroupBasedMean()
100 const uint32_t max_wg_size_y = ctx.gpu_info->GetMaxWorkGroupSizeForY(); in GenerateSubgroupBasedMean()
104 std::min(static_cast<uint32_t>(ctx.gpu_info->GetMaxWorkGroupTotalSize()), in GenerateSubgroupBasedMean()
[all …]
/external/tensorflow/tensorflow/lite/delegates/gpu/common/selectors/
Dsimple_selectors.cc50 const GpuInfo& gpu_info) { in SelectLSTM() argument
51 return absl::make_unique<GPUOperation>(CreateLSTM(op_def, gpu_info)); in SelectLSTM()
60 const GpuInfo& gpu_info, in SelectPReLU() argument
62 return absl::make_unique<GPUOperation>(CreatePReLU(gpu_info, op_def, attr)); in SelectPReLU()
91 const OperationDef& op_def, const GpuInfo& gpu_info, in SelectConcat() argument
95 GPUOperation operation = CreateConcatZ(op_def, channels, gpu_info); in SelectConcat()
113 const DepthwiseConvolution2DAttributes& attr, const GpuInfo& gpu_info, in SelectDWConvolutionDynamicWeights() argument
116 CreateDepthwiseConvolution2DDynamicWeights(gpu_info, op_def, attr)); in SelectDWConvolutionDynamicWeights()
165 const GpuInfo& gpu_info) { in SelectReduce() argument
167 CreateReduce(axis_to_reduce, src_shape, op_type, op_def, gpu_info)); in SelectReduce()
[all …]
Doperation_selector.cc44 const GpuInfo& gpu_info, in IsRecommendedForWinograd4x4To6x6() argument
52 if (gpu_info.IsAdreno() || gpu_info.IsAMD()) { in IsRecommendedForWinograd4x4To6x6()
56 if (gpu_info.IsAdreno()) { in IsRecommendedForWinograd4x4To6x6()
57 if (gpu_info.adreno_info.IsAdreno6xx()) { in IsRecommendedForWinograd4x4To6x6()
63 if (gpu_info.IsAMD()) { in IsRecommendedForWinograd4x4To6x6()
79 absl::Status WinogradFromNode(const GpuInfo& gpu_info, in WinogradFromNode() argument
89 if (!IsRecommendedForWinograd4x4To6x6(attr, gpu_info, output_shape)) { in WinogradFromNode()
99 gpu_info, shape_0, op_def.src_tensors[0].storage_type, in WinogradFromNode()
105 gpu_info, shape_1, op_def.src_tensors[0].storage_type, in WinogradFromNode()
119 SelectWinograd4x4To36(gpu_info, attr.padding, winograd_up_def); in WinogradFromNode()
[all …]
/external/tensorflow/tensorflow/lite/delegates/gpu/gl/
Dcompiler.cc61 bool ExceedsMaxSize(const Object& object, const GpuInfo& gpu_info) { in ExceedsMaxSize() argument
64 int2(gpu_info.GetMaxImage2DWidth(), gpu_info.GetMaxImage2DHeight()); in ExceedsMaxSize()
65 size_checker.max_z_size = gpu_info.GetMaxImage2DArrayLayers(); in ExceedsMaxSize()
69 ObjectType ChooseFastestObjectType(const GpuInfo& gpu_info) { in ChooseFastestObjectType() argument
70 return gpu_info.IsAdreno() ? ObjectType::TEXTURE : ObjectType::BUFFER; in ChooseFastestObjectType()
73 ObjectType ChooseFastestRefObjectType(const GpuInfo& gpu_info, in ChooseFastestRefObjectType() argument
75 if (!gpu_info.IsAdreno()) { in ChooseFastestRefObjectType()
78 if (gpu_info.adreno_info.adreno_gpu == AdrenoGpu::kAdreno630) { in ChooseFastestRefObjectType()
95 CompilerImpl(const NodeShader* node_shader, const GpuInfo* gpu_info, in CompilerImpl() argument
97 : node_shader_(*node_shader), gpu_info_(*gpu_info), options_(options) { in CompilerImpl()
[all …]

1234567