From e89a3bccbe4bbf431ff0816d085686276f3f3ccc Mon Sep 17 00:00:00 2001
From: Hiroshi Horii <horii@jp.ibm.com>
Date: Sun, 1 Sep 2019 02:52:44 +0900
Subject: [PATCH 1/2] support cu1 fusion

---
 src/transpile/fusion.hpp | 34 ++++++++++++++++++++++------------
 1 file changed, 22 insertions(+), 12 deletions(-)

diff --git a/src/transpile/fusion.hpp b/src/transpile/fusion.hpp
index fe83d9875f..ce326e0d80 100644
--- a/src/transpile/fusion.hpp
+++ b/src/transpile/fusion.hpp
@@ -15,6 +15,8 @@
 #ifndef _aer_transpile_fusion_hpp_
 #define _aer_transpile_fusion_hpp_
 
+//#define DEBUG
+
 #include "transpile/circuitopt.hpp"
 
 namespace AER {
@@ -116,6 +118,7 @@ const std::vector<std::string> Fusion::supported_gates({
   // Two-qubit gates
   "CX",   // Controlled-X gate (CNOT)
   "cx",   // Controlled-X gate (CNOT)
+  "cu1",  // Controlled-U1 gate
   "cz",   // Controlled-Z gate
   "swap" // SWAP gate
   // Three-qubit gates
@@ -514,21 +517,21 @@ bool Fusion::only_u1(const std::vector<op_t>& ops,
                      const uint_t until) const {
 
   for (uint_t i = from; i <= until; ++i) {
-    if (ops[i].name == "u1")
-      continue;
-    if (from < i && (i + 2) <= until
-        && ops[i - 1].name == "u1"
-        && ops[i    ].name == "cx"
-        && ops[i + 1].name == "u1"
-        && ops[i + 2].name == "cx"
-        && ops[i - 1].qubits[0] == ops[i    ].qubits[1]
-        && ops[i    ].qubits[1] == ops[i + 1].qubits[0]
-        && ops[i + 1].qubits[0] == ops[i + 2].qubits[1]
-        && ops[i    ].qubits[0] == ops[i + 2].qubits[0] )
+    if ((i + 3) <= until
+        && ops[i    ].name == "u1"
+        && ops[i + 1].name == "cx"
+        && ops[i + 2].name == "u1"
+        && ops[i + 3].name == "cx"
+        && ops[i    ].qubits[0] == ops[i + 1].qubits[1]
+        && ops[i + 1].qubits[1] == ops[i + 2].qubits[0]
+        && ops[i + 2].qubits[0] == ops[i + 3].qubits[1]
+        && ops[i + 1].qubits[0] == ops[i + 3].qubits[0] )
     {
-      i += 2;
+      i += 3;
       continue;
     }
+    if (ops[i].name == "u1" || ops[i].name == "cu1")
+      continue;
     return false;
   }
   return true;
@@ -579,6 +582,13 @@ cmatrix_t Fusion::matrix(const op_t& op) const {
         { {1, 0}, {0, 0} },
         { {0, 0}, std::exp( complex_t(0, 1.) * std::real(op.params[0])) }}
       );
+    } else if (op.name == "cu1") {   // zero-X90 pulse waltz gate
+      return Utils::make_matrix<complex_t>( {
+        { {1, 0}, {0, 0}, {0, 0}, {0, 0} },
+        { {0, 0}, {1, 0}, {0, 0}, {0, 0} },
+        { {0, 0}, {0, 0}, {1, 0}, {0, 0} },
+        { {0, 0}, {0, 0}, {0, 0}, std::exp( complex_t(0, 1.) * std::real(op.params[0])) }}
+      );
     } else if (op.name == "u2") {   // single-X90 pulse waltz gate
       return Utils::Matrix::u3( M_PI / 2., std::real(op.params[0]), std::real(op.params[1]));
     } else if (op.name == "u3" || op.name == "U") {   // two X90 pulse waltz gate

From f3c075e9c52fd11f9accdcaa558ec2b053d17463 Mon Sep 17 00:00:00 2001
From: Hiroshi Horii <horii@jp.ibm.com>
Date: Fri, 6 Sep 2019 21:44:58 +0900
Subject: [PATCH 2/2] simplify a logic to identify an unrolled cu1

---
 src/transpile/fusion.hpp | 40 +++++++++++++++++++++-------------------
 1 file changed, 21 insertions(+), 19 deletions(-)

diff --git a/src/transpile/fusion.hpp b/src/transpile/fusion.hpp
index ce326e0d80..9c6cd2a349 100644
--- a/src/transpile/fusion.hpp
+++ b/src/transpile/fusion.hpp
@@ -15,8 +15,6 @@
 #ifndef _aer_transpile_fusion_hpp_
 #define _aer_transpile_fusion_hpp_
 
-//#define DEBUG
-
 #include "transpile/circuitopt.hpp"
 
 namespace AER {
@@ -73,9 +71,9 @@ class Fusion : public CircuitOptimization {
                           const reg_t& dst_sorted_qubits,
                           const cmatrix_t& mat) const;
 
-  bool only_u1(const oplist_t& ops,
-               const uint_t from,
-               const uint_t until) const;
+  bool is_diagonal(const oplist_t& ops,
+                   const uint_t from,
+                   const uint_t until) const;
 
   double estimate_cost(const oplist_t& ops,
                        const uint_t from,
@@ -512,22 +510,26 @@ cmatrix_t Fusion::sort_matrix(const reg_t &src,
   return ret;
 }
 
-bool Fusion::only_u1(const std::vector<op_t>& ops,
-                     const uint_t from,
-                     const uint_t until) const {
+bool Fusion::is_diagonal(const std::vector<op_t>& ops,
+                         const uint_t from,
+                         const uint_t until) const {
+
+  // check unitary matrix of ops between "from" and "to" is a diagonal matrix
 
   for (uint_t i = from; i <= until; ++i) {
-    if ((i + 3) <= until
-        && ops[i    ].name == "u1"
-        && ops[i + 1].name == "cx"
-        && ops[i + 2].name == "u1"
-        && ops[i + 3].name == "cx"
-        && ops[i    ].qubits[0] == ops[i + 1].qubits[1]
-        && ops[i + 1].qubits[1] == ops[i + 2].qubits[0]
-        && ops[i + 2].qubits[0] == ops[i + 3].qubits[1]
-        && ops[i + 1].qubits[0] == ops[i + 3].qubits[0] )
+    //   ┌───┐┌────┐┌───┐
+    //  ─┤ X ├┤ U1 ├┤ X ├
+    //   └─┬─┘└────┘└─┬─┘
+    //  ───■──────────■─-
+    if ((i + 2) <= until
+        && ops[i + 0].name == "cx"
+        && ops[i + 1].name == "u1"
+        && ops[i + 2].name == "cx"
+        && ops[i + 0].qubits[1] == ops[i + 1].qubits[0]
+        && ops[i + 1].qubits[0] == ops[i + 2].qubits[1]
+        && ops[i + 0].qubits[0] == ops[i + 2].qubits[0] )
     {
-      i += 3;
+      i += 2;
       continue;
     }
     if (ops[i].name == "u1" || ops[i].name == "cu1")
@@ -540,7 +542,7 @@ bool Fusion::only_u1(const std::vector<op_t>& ops,
 double Fusion::estimate_cost(const std::vector<op_t>& ops,
                              const uint_t from,
                              const uint_t until) const {
-  if (only_u1(ops, from, until))
+  if (is_diagonal(ops, from, until))
     return cost_factor_;
 
   reg_t fusion_qubits;