TFE_TensorHandle* m = TestMatrixTensorHandle(ctx);
  TFE_Op* matmul = TFE_NewOp(ctx, "MatMul", status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_TensorHandle* retvals[1];
  int num_retvals = 1;
  for (auto s : state) {
    TFE_OpReset(matmul, "MatMul", nullptr, status);
    CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
    TFE_OpAddInput(matmul, m, status);

                                   "gradients/MatMul", false, true);
    TF_Operation* matmul2 = MatMul(expected_graph_, s_, const0, const3,
                                   "gradients/MatMul_1", true, false);
    expected_grad_outputs[0] = {matmul1, 0};
    expected_grad_outputs[1] = {matmul2, 0};
  }

  TF_Tensor* FloatTensor2x2(const float* values) {

  TFE_TensorHandle* h0_task0 = TestMatrixTensorHandle(ctx);

  TFE_Op* matmul = MatMulOp(ctx, h0_task0, h0_task0);
  TFE_OpSetDevice(matmul, remote_device_name, status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  TFE_TensorHandle* retvals[1];
  int num_retvals = 1;
  TFE_Execute(matmul, &retvals[0], &num_retvals, status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  TFE_TensorHandle* m = TestMatrixTensorHandle(ctx);
  TFE_Op* matmul = MatMulOp(ctx, m, m);
  TFE_TensorHandle* retvals[2] = {nullptr, nullptr};
  int num_retvals = 2;
  TFE_Execute(matmul, &retvals[0], &num_retvals, status);
  EXPECT_EQ(1, num_retvals);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_DeleteOp(matmul);
  TFE_DeleteTensorHandle(m);

  TFE_ContextOptions* tfe_context_options_;
  TFE_Context* tfe_context_;
};

TEST_F(ShapeInferenceTest, InfersShapesFromInputShapes) {
  TFE_Op* matmul_op;
  matmul_op = TFE_NewOp(tfe_context_, "MatMul", status_);
  CHECK_EQ(TF_OK, TF_GetCode(status_)) << TF_Message(status_);

  // Infer shape when everything is known.
  CheckOutputShapes(matmul_op,

  TF_DeleteStatus(status);
  TFE_OpSetAttrType(op, "T", TFE_TensorHandleDataType(a));

  return op;
}

TFE_Op* MatMulOp(TFE_Context* ctx, TFE_TensorHandle* a, TFE_TensorHandle* b) {
  TF_Status* status = TF_NewStatus();

  TFE_Op* op = TFE_NewOp(ctx, "MatMul", status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_OpAddInput(op, a, status);

* `tf.config.experimental.enable_tensor_float_32_execution`

 Ze=null;const Je=R({},["audio","video","img","source","image","track"]);let Qe=null;const et=R({},["alt","class","for","id","label","name","pattern","placeholder","role","summary","title","value","style","xmlns"]),tt="http://www.w3.org/1998/Math/MathML",nt="http://www.w3.org/2000/svg",ot="http://www.w3.org/1999/xhtml";let rt=ot,it=!1,at=null;const lt=R({},[tt,nt,ot],g);let ct=R({},["mi","mo","mn","ms","mtext"]),st=R({},["annotation-xml"]);const ut=R({},["title","style","font","a","script"]);let...

performance. t = f[j+1] f[j+1] = fieldAdd(t, flen[j+1]) flen[j+1] = fieldMontgomeryMulSu(zeta, flen[j+1], t) } } } for i := range f { f[i] = fieldMontgomeryMul(f[i], 16382) // 16382 = 256⁻¹ * R mod q } return ringElement(f) } // nttMul multiplies two nttElements. func nttMul(a, b nttElement) (p nttElement) { for i := range p { p[i] = fieldMontgomeryMul(a[i], b[i]) } return p } // sampleNTT samples an nttElement uniformly at random from the seed rho and the // indices s and r. It implements Step 3 of...