| /aosp_15_r20/external/pytorch/aten/src/ATen/native/quantized/cpu/ |
| H A D | qembeddingbag_unpack.cpp | 23 auto packed_weight = packed_w; in unpack() local
27 const auto input_rows = packed_weight.size(0); in unpack()
28 const auto input_columns = packed_weight.size(1); in unpack()
40 const auto* input = packed_weight.const_data_ptr<uint8_t>(); in unpack()
106 Tensor& qembeddingbag_byte_unpack_out(Tensor& output, const Tensor& packed_weight) { in qembeddingbag_byte_unpack_out() argument
121 const auto packed_weight_sizes = packed_weight.sizes(); in qembeddingbag_byte_unpack_out()
128 const auto* input_data = packed_weight.const_data_ptr<uint8_t>(); in qembeddingbag_byte_unpack_out()
161 Tensor qembeddingbag_byte_unpack(const Tensor& packed_weight) { in qembeddingbag_byte_unpack() argument
164 packed_weight.options().dtype(kFloat), in qembeddingbag_byte_unpack()
165 packed_weight.suggest_memory_format()); in qembeddingbag_byte_unpack()
[all …]
|
| H A D | qlinear.cpp | 986 auto packed_weight = at::native::itensor_from_mkldnn(onednn_weight); in linear_int8_with_onednn_weight() local
987 int64_t K = input.size(dim - 1), M = input.numel() / K, N = packed_weight.get_dim(1); in linear_int8_with_onednn_weight()
1025 auto weights_desc = packed_weight.get_desc(); in linear_int8_with_onednn_weight()
1066 auto expected_weight = packed_weight.reorder_if_differ_in(primitive_desc.weights_desc()); in linear_int8_with_onednn_weight()
1113 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight, in run() argument
1117 return packed_weight->apply_relu( in run()
1120 return packed_weight->apply( in run()
1130 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight, in run() argument
1139 return dynamic_cast<PackedLinearWeightsOnednn*>(packed_weight.get())->apply_leaky_relu( in run()
1155 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight, in run() argument
[all …]
|
| H A D | qconv_dynamic.cpp | 194 packed_weight, in run()
196 return packed_weight->apply_dynamic(input, reduce_range); in run()
205 const c10::intrusive_ptr<ConvPackedParamsBase<2>>& packed_weight, in run() argument
210 output = packed_weight->apply_dynamic(input, reduce_range); in run()
|
| H A D | qconv.cpp | 1551 auto packed_weight = at::native::itensor_from_mkldnn(weight); in _quantized_convolution_onednn() local
1587 const auto& kernel_size = packed_weight.get_dims(); in _quantized_convolution_onednn()
1623 auto weight_grouped = packed_weight.make_grouped_weights(groups, /* is_deconv */false); in _quantized_convolution_onednn()
1712 params, src, packed_weight, expected_bias, dst_dims, dst, in _quantized_convolution_onednn()
1720 ideep::tensor expected_weight = packed_weight.reorder_if_differ_in(expected_weight_desc); in _quantized_convolution_onednn()
1776 const c10::intrusive_ptr<ConvPackedParamsBase<kSpatialDim>>& packed_weight, in run() argument
1780 return packed_weight->apply_relu(act, output_scale, output_zero_point); in run()
1782 return packed_weight->apply(act, output_scale, output_zero_point); in run()
1793 const c10::intrusive_ptr<ConvPackedParamsBase<kSpatialDim>>& packed_weight, in run() argument
1802 … return dynamic_cast<PackedConvWeightsOnednn<kSpatialDim>*>(packed_weight.get())->apply_add_relu( in run()
[all …]
|
| H A D | qlinear_dynamic.cpp | 630 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight, in run() argument
633 return packed_weight->apply_dynamic_relu(std::move(input), reduce_range); in run()
635 return packed_weight->apply_dynamic(std::move(input), reduce_range); in run()
646 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight) { in run() argument
653 auto output = packed_weight->apply_dynamic(std::move(input)); in run()
692 auto packed_weight = PackedLinearWeightFp16::prepack(weight, bias); in run() local
693 auto output = packed_weight->apply_dynamic(std::move(input)); in run()
|
| H A D | qlinear_prepack.cpp | 305 auto packed_weight = at::native::new_with_itensor_mkldnn( in pack_weight_to_onednn_tensor() local
309 return packed_weight; in pack_weight_to_onednn_tensor()
479 const at::Tensor& packed_weight,
486 const at::Tensor& packed_weight, in _wrapped_quantized_linear_prepacked() argument
499 packed_weight); in _wrapped_quantized_linear_prepacked()
536 [[maybe_unused]] const at::Tensor& packed_weight,
544 [[maybe_unused]] const at::Tensor& packed_weight, in _wrapped_quantized_linear_prepacked_meta() argument
|
| H A D | qembeddingbag.cpp | 1039 const c10::intrusive_ptr<EmbeddingPackedParamsBase>& packed_weight, in run() argument
1049 return packed_weight->embeddingbag_byte( in run()
1058 return packed_weight->embeddingbag_4bit( in run()
1078 const c10::intrusive_ptr<EmbeddingPackedParamsBase>& packed_weight, in run() argument
1086 return packed_weight->embeddingbag_byte( in run()
1095 return packed_weight->embeddingbag_4bit( in run()
|
| /aosp_15_r20/external/pytorch/aten/src/ATen/native/quantized/ |
| H A D | qconv_unpack.cpp | 53 const c10::intrusive_ptr<ConvPackedParamsBase<kSpatialDim>>& packed_weight) { in run() argument
59 return packed_weight->unpack(); in run()
69 return packed_weight->unpack(); in run()
75 return packed_weight->unpack(); in run()
89 const c10::intrusive_ptr<ConvPackedParamsBase<2>>& packed_weight) { in run() argument
96 std::tie(weight, bias) = packed_weight->unpack(); in run()
104 std::tie(weight, bias) = packed_weight->unpack(); in run()
113 std::tie(weight, bias) = packed_weight->unpack(); in run()
131 const c10::intrusive_ptr<ConvPackedParamsBase<kSpatialDim>>& packed_weight) { in run() argument
132 return packed_weight->stride(); in run()
[all …]
|
| H A D | qlinear_unpack.cpp | 25 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight) { in run() argument
26 return packed_weight->unpack(); in run()
33 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight) { in run() argument
41 return packed_weight->unpack(); in run()
48 const at::Tensor& packed_weight) { in run() argument
59 const at::Tensor& packed_weight) { in run() argument
|
| /aosp_15_r20/external/pytorch/torch/_inductor/ |
| H A D | mkldnn_lowerings.py | 369 packed_weight: TensorBox,
389 packed_weight,
416 packed_weight: TensorBox,
452 packed_weight,
476 packed_weight: TensorBox,
526 *_, layout, x, packed_weight = mm_args(
527 x, packed_weight, layout=layout, out_dtype=output_dtype
543 and use_cpp_packed_gemm_template(layout, x, packed_weight)
545 W_tensor = V.graph.constants[packed_weight.get_name()].to_dense()
549 name=packed_weight.get_name() + "_BMatrixCompens",
[all …]
|
| H A D | mkldnn_ir.py | 639 packed_weight = args[1]
681 packed_weight,
882 packed_weight = args[1]
939 packed_weight,
1339 packed_weight = args[1]
1392 packed_weight,
1557 packed_weight = args[1]
1630 packed_weight,
|
| /aosp_15_r20/external/pytorch/aten/src/ATen/native/ |
| H A D | QuantizedLinear.cpp | 406 c10::intrusive_ptr<LinearPackedParamsBase> packed_weight = in fbgemm_pack_gemm_matrix_fp16() local
409 std::make_unique<decltype(packed_weight)>(std::move(packed_weight)); in fbgemm_pack_gemm_matrix_fp16()
416 const Tensor& packed_weight, in fbgemm_linear_fp16_weight_fp32_activation() argument
433 c10::intrusive_ptr<LinearPackedParamsBase>>(packed_weight)) in fbgemm_linear_fp16_weight_fp32_activation()
464 const Tensor& packed_weight, in fbgemm_linear_fp16_weight() argument
467 input, packed_weight, bias); in fbgemm_linear_fp16_weight()
558 const Tensor& packed_weight,
572 const Tensor& packed_weight,
|
| /aosp_15_r20/external/pytorch/aten/src/ATen/native/quantized/cuda/ |
| H A D | EmbeddingBag.cu | 172 Tensor qembeddingbag_byte_unpack(const Tensor& packed_weight) { in qembeddingbag_byte_unpack() argument
173 const auto packed_weight_sizes = packed_weight.sizes(); in qembeddingbag_byte_unpack()
184 packed_weight.options().dtype(kFloat), in qembeddingbag_byte_unpack()
185 packed_weight.suggest_memory_format()); in qembeddingbag_byte_unpack()
544 Tensor qembeddingbag_4bit_unpack(const Tensor& packed_weight) { in qembeddingbag_4bit_unpack() argument
546 const auto input_rows = packed_weight.size(0); in qembeddingbag_4bit_unpack()
547 const auto input_columns = packed_weight.size(1); in qembeddingbag_4bit_unpack()
548 const auto* input_data = packed_weight.const_data_ptr<uint8_t>(); in qembeddingbag_4bit_unpack()
560 packed_weight.options().dtype(kFloat), in qembeddingbag_4bit_unpack()
561 packed_weight.suggest_memory_format()); in qembeddingbag_4bit_unpack()
|
| /aosp_15_r20/external/pytorch/benchmarks/operator_benchmark/pt/ |
| H A D | qembedding_pack_test.py | 73 def forward(self, packed_weight): argument
74 return self.op_func(packed_weight)
85 def forward(self, packed_weight): argument
86 return self.op_func(packed_weight)
|
| /aosp_15_r20/external/pytorch/aten/src/ATen/native/quantized/cudnn/ |
| H A D | Conv.cpp | 354 const c10::intrusive_ptr<ConvPackedParamsBase<2>>& packed_weight, in run() argument
363 output = packed_weight->apply_relu(act, output_scale, output_zero_point); in run()
365 output = packed_weight->apply(act, output_scale, output_zero_point); in run()
377 const c10::intrusive_ptr<ConvPackedParamsBase<kSpatialDim>>& packed_weight, in run() argument
384 return packed_weight->apply_relu(act, output_scale, output_zero_point); in run()
386 return packed_weight->apply(act, output_scale, output_zero_point); in run()
|
| H A D | Linear.cpp | 347 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight, in run() argument
352 return packed_weight->apply_relu(std::move(act), output_scale, output_zero_point); in run()
354 return packed_weight->apply(std::move(act), output_scale, output_zero_point); in run()
|
| /aosp_15_r20/external/pytorch/torch/ao/quantization/fx/ |
| H A D | _lower_to_native_backend.py | 420 packed_weight = getattr(self, attr_name)
421 destination[prefix + attr_name] = packed_weight
467 packed_weight = prepacking_module()
468 packed_weights[node.name] = packed_weight
480 packed_weight = packed_weights[node.name]
488 setattr(quantized_model, packed_weight_name, packed_weight)
916 packed_weight = model.graph.create_node(
929 packed_weight,
1058 packed_weight = model.graph.create_node(
1065 func_node.args = (pattern_input, packed_weight, reduce_range_node)
[all …]
|
| /aosp_15_r20/external/pytorch/aten/src/ATen/native/ao_sparse/quantized/cpu/ |
| H A D | qlinear_dynamic.cpp | 170 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight) { in run() argument
175 return packed_weight->apply_dynamic_relu(input); in run()
177 return packed_weight->apply_dynamic(input); in run()
|
| H A D | qlinear.cpp | 238 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight, in run() argument
242 return packed_weight->apply_relu(input, output_scale, output_zero_point); in run()
244 return packed_weight->apply(input, output_scale, output_zero_point); in run()
|
| H A D | qlinear_unpack.cpp | 130 const c10::intrusive_ptr<LinearPackedParamsBase>& packed_weight) { in run() argument
131 return packed_weight->unpack(); in run()
|
| /aosp_15_r20/external/pytorch/torch/csrc/jit/runtime/static/ |
| H A D | ops.cpp | 2221 c10::intrusive_ptr<LinearPackedParamsBase> packed_weight; in __anon11f46a8b6e02() local
2223 packed_weight = w->toCustomClass<LinearPackedParamsBase>(); in __anon11f46a8b6e02()
2225 return [packed_weight](ProcessedNode* p_node) { in __anon11f46a8b6e02()
2244 if (packed_weight) { in __anon11f46a8b6e02()
2245 packed_weight->apply_out(input, output_scale, output_zero_point, out_t); in __anon11f46a8b6e02()
2266 c10::intrusive_ptr<LinearPackedParamsBase> packed_weight; in __anon11f46a8b7002() local
2268 packed_weight = w->toCustomClass<LinearPackedParamsBase>(); in __anon11f46a8b7002()
2270 return [packed_weight](ProcessedNode* p_node) { in __anon11f46a8b7002()
2290 if (packed_weight) { in __anon11f46a8b7002()
2291 packed_weight->apply_out( in __anon11f46a8b7002()
[all …]
|
| /aosp_15_r20/external/pytorch/aten/src/ATen/native/mkldnn/ |
| H A D | ConvPrepack.cpp | 83 ideep::tensor packed_weight; in create() local
84 packed_weight.init(expected_weight_desc); in create()
85 packed_weight.feed_from(w); in create()
88 std::move(packed_weight), in create()
|
| /aosp_15_r20/external/pytorch/torch/ao/nn/quantized/dynamic/modules/ |
| H A D | rnn.py | 47 packed_weight = torch.ops.quantized.linear_prepack(qweight, bias)
49 return packed_weight
55 packed_weight = torch.ops.quantized.linear_prepack_fp16(qweight, bias)
57 return packed_weight
431 packed_weight = torch.ops.quantized.linear_prepack(qweight, b)
432 return packed_weight
|
| /aosp_15_r20/external/pytorch/torch/_inductor/fx_passes/ |
| H A D | quantization.py | 324 packed_weight, w_scale, w_zp = (
354 packed_weight,
412 packed_weight, w_scale, w_zp = (
432 packed_weight,
478 packed_weight, w_scale, w_zp = (
508 packed_weight,
644 packed_weight, w_scale, w_zp = (
674 packed_weight,
|
| /aosp_15_r20/external/pytorch/torch/csrc/jit/passes/ |
| H A D | mkldnn_rewrite.cpp | 178 Value* packed_weight = graph->insertConstant(weak_class_obj) in PrePackingOpsFolder() local
180 prepack_op_value->replaceAllUsesWith(packed_weight); in PrePackingOpsFolder()
|