diff --git a/src/broadcom/compiler/vir_to_qpu.c b/src/broadcom/compiler/vir_to_qpu.c
index c8b6e0a91a0..08970d52954 100644
--- a/src/broadcom/compiler/vir_to_qpu.c
+++ b/src/broadcom/compiler/vir_to_qpu.c
@@ -129,19 +129,8 @@ set_src(struct v3d_qpu_instr *instr, enum v3d_qpu_mux *mux, struct qpu_reg src)
 }
 
 static bool
-is_no_op_mov(struct qinst *qinst)
+v3d33_mov_src_and_dst_equal(struct qinst *qinst)
 {
-        static const struct v3d_qpu_sig no_sig = {0};
-
-        /* Make sure it's just a lone MOV. */
-        if (qinst->qpu.type != V3D_QPU_INSTR_TYPE_ALU ||
-            qinst->qpu.alu.mul.op != V3D_QPU_M_MOV ||
-            qinst->qpu.alu.add.op != V3D_QPU_A_NOP ||
-            memcmp(&qinst->qpu.sig, &no_sig, sizeof(no_sig)) != 0) {
-                return false;
-        }
-
-        /* Check if it's a MOV from a register to itself. */
         enum v3d_qpu_waddr waddr = qinst->qpu.alu.mul.waddr;
         if (qinst->qpu.alu.mul.magic_write) {
                 if (waddr < V3D_QPU_WADDR_R0 || waddr > V3D_QPU_WADDR_R4)
@@ -168,6 +157,57 @@ is_no_op_mov(struct qinst *qinst)
                         return false;
         }
 
+        return true;
+}
+
+static bool
+v3d71_mov_src_and_dst_equal(struct qinst *qinst)
+{
+        if (qinst->qpu.alu.mul.magic_write)
+                return false;
+
+        enum v3d_qpu_waddr waddr = qinst->qpu.alu.mul.waddr;
+        int raddr;
+
+        raddr = qinst->qpu.alu.mul.a.raddr;
+        if (raddr != waddr)
+                return false;
+
+        return true;
+}
+
+static bool
+mov_src_and_dst_equal(struct qinst *qinst,
+                      const struct v3d_device_info *devinfo)
+{
+        if (devinfo->ver < 71)
+                return v3d33_mov_src_and_dst_equal(qinst);
+        else
+                return v3d71_mov_src_and_dst_equal(qinst);
+}
+
+
+static bool
+is_no_op_mov(struct qinst *qinst,
+             const struct v3d_device_info *devinfo)
+{
+        static const struct v3d_qpu_sig no_sig = {0};
+
+        /* Make sure it's just a lone MOV. We only check for M_MOV. Although
+         * for V3D 7.x there is also A_MOV, we don't need to check for it as
+         * we always emit using M_MOV. We could use A_MOV later on the
+         * squedule to improve performance
+         */
+        if (qinst->qpu.type != V3D_QPU_INSTR_TYPE_ALU ||
+            qinst->qpu.alu.mul.op != V3D_QPU_M_MOV ||
+            qinst->qpu.alu.add.op != V3D_QPU_A_NOP ||
+            memcmp(&qinst->qpu.sig, &no_sig, sizeof(no_sig)) != 0) {
+                return false;
+        }
+
+        if (!mov_src_and_dst_equal(qinst, devinfo))
+                return false;
+
         /* No packing or flags updates, or we need to execute the
          * instruction.
          */
@@ -324,7 +364,7 @@ v3d_generate_code_block(struct v3d_compile *c,
                                 qinst->qpu.alu.mul.waddr = dst.index;
                                 qinst->qpu.alu.mul.magic_write = dst.magic;
 
-                                if (is_no_op_mov(qinst)) {
+                                if (is_no_op_mov(qinst, c->devinfo)) {
                                         vir_remove_instruction(c, qinst);
                                         continue;
                                 }