Revert D33716039: [pytorch][PR] Add ONEDNN quantization backend

Test Plan: revert-hammer

Differential Revision:
D33716039 (pytorch@989b248)

Original commit changeset: 6f7bb807e857

Original Phabricator Diff: D33716039 (pytorch@989b248)

fbshipit-source-id: ed233c5b99d4edb7d5a9d6c600825c78555f16d0
(cherry picked from commit d3e1f82)

aten/src/ATen/Context.cpp

@@ -236,10 +236,6 @@ const std::vector<at::QEngine>& Context::supportedQEngines() {
     engines.push_back(at::kNoQEngine);
 #endif // C10_MOBILE
 
-#if AT_MKLDNN_ENABLED()
-    engines.push_back(at::kONEDNN);
-#endif
-
 #ifdef USE_FBGEMM
     if (fbgemm::fbgemmSupportedCPU()) {
       engines.push_back(at::kFBGEMM);

aten/src/ATen/native/quantized/cpu/conv_serialization.h

@@ -4,7 +4,6 @@
 #include <ATen/core/List.h>
 #include <ATen/native/quantized/cpu/fbgemm_utils.h>
 #include <ATen/native/quantized/cpu/qnnpack_utils.h>
-#include <ATen/native/quantized/cpu/onednn_utils.h>
 #include <c10/util/irange.h>
 
 #include <tuple>
@@ -359,20 +358,6 @@ c10::intrusive_ptr<ConvPackedParamsBase<kSpatialDim>> deserialize_conv(
     );
   }
 #endif // USE_PYTORCH_QNNPACK
-#if AT_MKLDNN_ENABLED()
-  if (ctx.qEngine() == at::QEngine::ONEDNN) {
-    return PackedConvWeightsOnednn<kSpatialDim>::prepack(
-      weight.value(),
-      bias,
-      stride,
-      padding,
-      output_padding,
-      dilation,
-      groups,
-      transpose
-    );
-  }
-#endif // AT_MKLDNN_ENABLED()
 TORCH_CHECK(
   false,
   "Didn't find engine for when deserializing ConvPackedParams: ",

aten/src/ATen/native/quantized/cpu/fbgemm_utils.cpp

@@ -4,7 +4,6 @@
 #include <ATen/native/quantized/cpu/embedding_packed_params.h>
 #include <ATen/native/quantized/cpu/fbgemm_utils.h>
 #include <ATen/native/quantized/cpu/qnnpack_utils.h>
-#include <ATen/native/quantized/cpu/onednn_utils.h>
 #include <ATen/native/TensorFactories.h>
 #include <ATen/quantized/QTensorImpl.h>
 #include <ATen/quantized/Quantizer.h>
@@ -471,16 +470,6 @@ int register_linear_params() {
                       std::move(weight), std::move(bias));
                 }
 #endif // USE_PYTORCH_QNNPACK
-#if AT_MKLDNN_ENABLED()
-                if (at::globalContext().qEngine() == at::QEngine::ONEDNN) {
-                  TORCH_CHECK(
-                      weight.scalar_type() == at::kQInt8,
-                      "ONEDNN only supports INT8 bit width currently. Got ",
-                      c10::toString(weight.scalar_type()));
-                  return PackedLinearWeightsOnednn::prepack(
-                      std::move(weight), std::move(bias));
-                }
-#endif // #if AT_MKLDNN_ENABLED()
                 TORCH_CHECK(false, "Unknown qengine");
               })
               .def("bias", [](const c10::intrusive_ptr<LinearPackedParamsBase>& self) {