diff --git a/jtag/include/tap.h b/jtag/include/tap.h
index 765126d6..75742d00 100644
--- a/jtag/include/tap.h
+++ b/jtag/include/tap.h
@@ -31,6 +31,8 @@
 void tap_reset( chain_t *chain );
 void tap_capture_dr( chain_t *chain );
 void tap_capture_ir( chain_t *chain );
+void tap_defer_shift_register( chain_t *chain, const tap_register *in, tap_register *out, int exit );
+void tap_shift_register_output( chain_t *chain, const tap_register *in, tap_register *out, int exit );
 void tap_shift_register( chain_t *chain, const tap_register *in, tap_register *out, int exit );
 
 #endif /* TAP_H */
diff --git a/jtag/src/tap/chain.c b/jtag/src/tap/chain.c
index 64eb0a53..654c9573 100644
--- a/jtag/src/tap/chain.c
+++ b/jtag/src/tap/chain.c
@@ -178,7 +178,7 @@ chain_shift_data_registers_mode( chain_t *chain, int capture_output, int capture
 				capture_output ? ps->parts[i]->active_instruction->data_register->out : NULL,
 				(i + 1) == ps->len ? exit : EXITMODE_SHIFT );
 	}
-#else
+#elif 1
 	{
 		/* new implementation:
 		   combine the data registers of all parts in the chain into one temporary register,
@@ -219,6 +219,26 @@ chain_shift_data_registers_mode( chain_t *chain, int capture_output, int capture
 
 		data_register_free( temp_reg );
 	}
+#else
+	/* new^2 implementation: split into defer + retrieve part
+	   shift the data register of each part in the chain one by one */
+
+	for (i = 0; i < ps->len; i++) {
+		puts("tap_defer_shift_register");
+		tap_defer_shift_register( chain, ps->parts[i]->active_instruction->data_register->in,
+				capture_output ? ps->parts[i]->active_instruction->data_register->out : NULL,
+				(i + 1) == ps->len ? exit : EXITMODE_SHIFT );
+	}
+
+	if(capture_output)
+	{
+		for (i = 0; i < ps->len; i++) {
+			puts("tap_shift_register_output");
+			tap_shift_register_output( chain, ps->parts[i]->active_instruction->data_register->in,
+				ps->parts[i]->active_instruction->data_register->out,
+				(i + 1) == ps->len ? exit : EXITMODE_SHIFT );
+		}
+	}
 #endif
 }
 
diff --git a/jtag/src/tap/tap.c b/jtag/src/tap/tap.c
index 9e88d94c..5ba467ef 100644
--- a/jtag/src/tap/tap.c
+++ b/jtag/src/tap/tap.c
@@ -43,9 +43,9 @@ tap_reset( chain_t *chain )
 }
 
 void
-tap_shift_register( chain_t *chain, const tap_register *in, tap_register *out, int exit )
+tap_defer_shift_register( chain_t *chain, const tap_register *in, tap_register *out, int exit )
 {
-	int i,j;
+	int i;
 
 	if (!(tap_state( chain ) & TAPSTAT_SHIFT))
 		printf( _("%s: Invalid state: %2X\n"), "tap_shift_register", tap_state( chain ) );
@@ -63,7 +63,6 @@ tap_shift_register( chain_t *chain, const tap_register *in, tap_register *out, i
 	else
 		cable_defer_transfer( chain->cable, i, in->data, NULL );
 
-	j = i;
 	for (; i < in->len; i++) {
 		if (out != NULL && (i < out->len))
 			out->data[i] = cable_defer_get_tdo( chain->cable );
@@ -75,16 +74,35 @@ tap_shift_register( chain_t *chain, const tap_register *in, tap_register *out, i
 		chain_defer_clock( chain, 1, 0, 1 );	/* Update-DR or Update-IR */
 		chain_defer_clock( chain, 0, 0, 1 );	/* Run-Test/Idle */
 	}
+}
 
+void
+tap_shift_register_output( chain_t *chain, const tap_register *in, tap_register *out, int exit )
+{
 	if(out != NULL)
 	{
-		/* Asking for the result of the cable transfer actually flushes the queue */
+		int j;
+
+		j = in->len;
+		if(exit) j--;
+		if(out && out->len < j) j = out->len;
+
+		/* Asking for the result of the cable transfer
+         * actually flushes the queue */
+
 		(void)cable_transfer_late( chain->cable, out->data );
 		for (; j < in->len && j < out->len; j++) 
 				out->data[j] = cable_get_tdo_late( chain->cable );
 	}
 }
 
+void
+tap_shift_register( chain_t *chain, const tap_register *in, tap_register *out, int exit )
+{
+	tap_defer_shift_register( chain, in, out, exit );
+	tap_shift_register_output( chain, in, out, exit );
+}
+
 void
 tap_capture_dr( chain_t *chain )
 {