diff --git a/kernel/vortex_test.elf b/kernel/vortex_test.elf
index 27379902..ad965d57 100755
Binary files a/kernel/vortex_test.elf and b/kernel/vortex_test.elf differ
diff --git a/kernel/vx_os/vx_back/vx_back.s b/kernel/vx_os/vx_back/vx_back.s
index 1421181d..48656abf 100644
--- a/kernel/vx_os/vx_back/vx_back.s
+++ b/kernel/vx_os/vx_back/vx_back.s
@@ -8,7 +8,7 @@
 _start:
 #     li a0, 4
 #     la a1, SPAWN
-#     .word 0x00b5106b
+#     .word 0x00b5106b # wspawn a0(numWarps), a1(PC SPAWN)
 #     j SPAWN
 #     nop
 #     nop
@@ -21,8 +21,10 @@ _start:
 # SPAWN:
 #     li a2, 7
 #     li a0, 0
+#     li a1, 4
+#     .word 0x00b5406b # barrier a0(barrier id), a1(numWarps)
 #     .word 0x0005006b # tmc a0
-    ###########################
+    ##########################
 #     li a0, 4
 #     .word 0x0005006b    # tmc a0
 #     csrr a1, 0x20       # read thread IDs
diff --git a/rtl/VX_define.v b/rtl/VX_define.v
index 344f6129..eb95bddd 100644
--- a/rtl/VX_define.v
+++ b/rtl/VX_define.v
@@ -10,6 +10,8 @@
 // `define ONLY
 
 
+`define NUM_BARRIERS 4
+
 `define R_INST 7'd51
 `define L_INST 7'd3
 `define ALU_INST 7'd19
diff --git a/rtl/VX_fetch.v b/rtl/VX_fetch.v
index 62276d2c..0f707061 100644
--- a/rtl/VX_fetch.v
+++ b/rtl/VX_fetch.v
@@ -30,6 +30,12 @@ module VX_fetch (
 			.clk              (clk),
 			.reset            (reset),
 			.stall            (pipe_stall),
+
+			.is_barrier       (VX_warp_ctl.is_barrier),
+			.barrier_id       (VX_warp_ctl.barrier_id),
+			.num_warps        (VX_warp_ctl.num_warps),
+			.barrier_warp_num (VX_warp_ctl.warp_num),
+
 			// Wspawn
 			.wspawn           (VX_warp_ctl.wspawn),
 			.wsapwn_pc        (VX_warp_ctl.wspawn_pc),
diff --git a/rtl/VX_gpgpu_inst.v b/rtl/VX_gpgpu_inst.v
index 9cf8333f..c468dc39 100644
--- a/rtl/VX_gpgpu_inst.v
+++ b/rtl/VX_gpgpu_inst.v
@@ -37,6 +37,9 @@ module VX_gpgpu_inst (
 	end
 
 
+	assign VX_warp_ctl.is_barrier = VX_gpu_inst_req.is_barrier && valid_inst;
+	assign VX_warp_ctl.barrier_id = VX_gpu_inst_req.a_reg_data[0];
+	assign VX_warp_ctl.num_warps  = VX_gpu_inst_req.rd2 - 1;
 
 	assign VX_warp_ctl.wspawn            = wspawn;
 	assign VX_warp_ctl.wspawn_pc         = wspawn_pc;
diff --git a/rtl/VX_warp_scheduler.v b/rtl/VX_warp_scheduler.v
index d37825c3..d5e09584 100644
--- a/rtl/VX_warp_scheduler.v
+++ b/rtl/VX_warp_scheduler.v
@@ -18,6 +18,11 @@ module VX_warp_scheduler (
 	input  wire           whalt,
 	input  wire[`NW_M1:0] whalt_warp_num,
 
+	input wire            is_barrier,
+	input wire[31:0]      barrier_id,
+	input wire[`NW_M1:0]  num_warps,
+	input wire[`NW_M1:0]  barrier_warp_num,
+
 	// WSTALL
 	input  wire           wstall,
 	input  wire[`NW_M1:0] wstall_warp_num,
@@ -72,10 +77,16 @@ module VX_warp_scheduler (
 	reg[`NW-1:0] visible_active;
 	wire[`NW-1:0] use_active;
 
+	wire wstall_this_cycle;
 
 	reg[`NT_M1:0] thread_masks[`NW-1:0];
 	reg[31:0]     warp_pcs[`NW-1:0];
 
+	// barriers
+	reg[`NW-1:0] barrier_stall_mask[(`NUM_BARRIERS-1):0];
+	wire         reached_barrier_limit;
+	wire[`NW-1:0] curr_barrier_mask;
+	wire[($clog2(`NUM_BARRIERS)-1):0] curr_barrier_count;
 
 	// wsapwn
 	reg[31:0]    use_wsapwn_pc;
@@ -91,27 +102,23 @@ module VX_warp_scheduler (
 
 	wire[31:0] new_pc;
 
+	reg[`NW-1:0] total_barrier_stall;
+
 	/* verilator lint_off UNUSED */
 	wire[`NW_M1:0] num_active;
 	/* verilator lint_on UNUSED */
 
-	reg[1:0] start;
-	// initial begin
-	// 	warp_pcs[0] = (32'h80000000 - 4);
-	// 	start = 0;
-	// 	warp_active[0]     = 1; // Activating first warp
-	// 	visible_active[0]  = 1; // Activating first warp
-	// 	thread_masks[0][0] = 1; // Activating first thread in first warp
-	// end
-
 	integer curr_w_help;
 	always @(posedge clk or posedge reset) begin
 		if (reset) begin
-			start              <= 0;
-			warp_pcs[0]        <= (32'h80000000 - 4);
-			warp_active[0]     <= 1; // Activating first warp
-			visible_active[0]  <= 1; // Activating first warp
-			thread_masks[0]    <= 1; // Activating first thread in first warp
+			barrier_stall_mask[0] <= 0;
+			barrier_stall_mask[1] <= 0;
+			use_wsapwn_pc         <= 0;
+			use_wsapwn            <= 0;
+			warp_pcs[0]           <= (32'h80000000 - 4);
+			warp_active[0]        <= 1; // Activating first warp
+			visible_active[0]     <= 1; // Activating first warp
+			thread_masks[0]       <= 1; // Activating first thread in first warp
 			for (curr_w_help = 1; curr_w_help < `NW; curr_w_help=curr_w_help+1) begin
 				warp_pcs[curr_w_help]        <= 0;
 				warp_active[curr_w_help]     <= 0; // Activating first warp
@@ -127,6 +134,15 @@ module VX_warp_scheduler (
 				use_wsapwn     <= wspawn_new_active & (~`NW'b1);
 			end
 
+			if (is_barrier) begin
+				warp_stalled[barrier_warp_num]       <= 0;
+				if (reached_barrier_limit) begin
+					barrier_stall_mask[barrier_id] <= 0;
+				end else begin
+					barrier_stall_mask[barrier_id][barrier_warp_num] <= 1;
+				end
+			end
+
 			if (update_use_wspawn) begin
 				use_wsapwn[warp_to_schedule] <= 0;
 			end
@@ -162,7 +178,7 @@ module VX_warp_scheduler (
 
 			// Refilling active warps
 			if (update_visible_active) begin
-				visible_active <= warp_active & (~warp_stalled);
+				visible_active <= warp_active & (~warp_stalled) & (~total_barrier_stall);
 			end
 
 			// Don't change state if stall
@@ -185,8 +201,23 @@ module VX_warp_scheduler (
 		end
 	end
 
+	assign curr_barrier_mask     = barrier_stall_mask[barrier_id][`NW-1:0];
+	assign curr_barrier_count    = $countones(curr_barrier_mask);
+	assign reached_barrier_limit = curr_barrier_count == (num_warps);
 
-	assign update_visible_active = ($countones(visible_active) < 1) && !(stall || wstall || hazard || is_join);
+	assign wstall_this_cycle = wstall && (wstall_warp_num == warp_to_schedule); // Maybe bug
+
+	genvar curr_b;
+	always @(*) begin
+		total_barrier_stall = 0;
+		for (curr_b = 0; curr_b < `NUM_BARRIERS; curr_b=curr_b+1)
+		begin
+			total_barrier_stall[`NW-1:0] = total_barrier_stall[`NW-1:0] | barrier_stall_mask[curr_b[($clog2(`NUM_BARRIERS)-1):0]][`NW-1:0];
+		end
+	end
+
+
+	assign update_visible_active = ($countones(visible_active) < 1) && !(stall || wstall_this_cycle || hazard || is_join);
 
 	wire[(1+32+`NT_M1):0] q1 = {1'b1, 32'b0                   , thread_masks[split_warp_num]};
 	wire[(1+32+`NT_M1):0] q2 = {1'b0, split_save_pc           , split_later_mask};
@@ -221,9 +252,9 @@ module VX_warp_scheduler (
 
 	assign hazard = (should_jal || should_bra) && schedule;
 
-	assign real_schedule = schedule && !warp_stalled[warp_to_schedule];
+	assign real_schedule = schedule && !warp_stalled[warp_to_schedule] && !total_barrier_stall[warp_to_schedule];
 
-	assign global_stall = (stall || wstall || hazard || !real_schedule || is_join);
+	assign global_stall = (stall || wstall_this_cycle || hazard || !real_schedule || is_join);
 
 
 	wire real_use_wspawn = use_wsapwn[warp_to_schedule];
@@ -237,7 +268,7 @@ module VX_warp_scheduler (
 	assign new_pc = warp_pc + 4;
 
 
-	assign use_active = (num_active < 1) ? (warp_active & (~warp_stalled)) : visible_active;
+	assign use_active = (num_active < 1) ? (warp_active & (~warp_stalled) & (~total_barrier_stall)) : visible_active;
 
 	// Choosing a warp to schedule
 	VX_priority_encoder choose_schedule(
diff --git a/rtl/interfaces/VX_gpu_inst_req_inter.v b/rtl/interfaces/VX_gpu_inst_req_inter.v
index 8ce4b343..1d24c960 100644
--- a/rtl/interfaces/VX_gpu_inst_req_inter.v
+++ b/rtl/interfaces/VX_gpu_inst_req_inter.v
@@ -11,6 +11,7 @@ interface VX_gpu_inst_req_inter();
 	wire                 is_wspawn;
 	wire                 is_tmc;   
 	wire                 is_split; 
+
 	wire                 is_barrier;
 
 	wire[31:0]           pc_next;
diff --git a/rtl/interfaces/VX_warp_ctl_inter.v b/rtl/interfaces/VX_warp_ctl_inter.v
index 84979972..a6697de9 100644
--- a/rtl/interfaces/VX_warp_ctl_inter.v
+++ b/rtl/interfaces/VX_warp_ctl_inter.v
@@ -17,6 +17,10 @@ interface VX_warp_ctl_inter ();
 
 	wire           ebreak;
 
+	// barrier
+	wire           is_barrier;
+	wire[31:0]     barrier_id;
+	wire[`NW_M1:0]  num_warps;
 
 	wire           is_split;
 	wire[`NW_M1:0] split_warp_num;