Optimize asm matmul dma software path

rtl/ip/APB_UART/URT/uart_defines.h

@@ -113,7 +113,7 @@ THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 // FIFO parameter defines
 
 `define UART_FIFO_WIDTH	     8
-`define UART_FIFO_DEPTH	     32
+`define UART_FIFO_DEPTH	     16
-`define UART_FIFO_POINTER_W	 5
+`define UART_FIFO_POINTER_W	 4
-`define UART_FIFO_COUNTER_W	 6
+`define UART_FIFO_COUNTER_W	 5
 `define UART_FIFO_REC_WIDTH  11

rtl/ip/Bus_interconnects/axi2sram_sp_external.v

@@ -20,8 +20,7 @@
 module axi2sram_sp_external #(
     parameter AXI_ID_WIDTH      = 5,
     parameter AXI_ADDR_WIDTH    = 32,
-    parameter AXI_DATA_WIDTH    = 32,
+    parameter AXI_DATA_WIDTH    = 32
-    parameter FAST_READ_BURST   = 1
 )(
     input                         clk,
     input                         resetn,
@@ -65,7 +64,6 @@ module axi2sram_sp_external #(
     output  reg                       req_o,
     output  reg                       we_o,
     output  reg [AXI_ADDR_WIDTH-1:0]   addr_o,
-    output  reg                       bank_o,
     output  reg [AXI_DATA_WIDTH/8-1:0] be_o,
     output  reg [AXI_DATA_WIDTH-1:0]   data_o,
     input   [AXI_DATA_WIDTH-1:0]   data_i
@@ -102,34 +100,40 @@ module axi2sram_sp_external #(
     localparam              FIXED       = 2'b00;
     localparam              INCR        = 2'b01;
     localparam              WRAP        = 2'b10;
-    localparam [AXI_ADDR_WIDTH-1:0] ADDR_INCR =
-        {{(AXI_ADDR_WIDTH-1){1'b0}}, 1'b1} << LOG_NR_BYTES;
     
     reg [AXI_ADDR_WIDTH-1:0] req_addr_d, req_addr_q;
-    reg [8:0]                cnt_d, cnt_q;
+    reg [7:0]                cnt_d, cnt_q;
-    reg [AXI_ADDR_WIDTH-1:0] wrap_mask_d, wrap_mask_q;
-    reg                      req_bank_d, req_bank_q;
-    reg [AXI_DATA_WIDTH-1:0] read_data_q;
-    reg                      read_capture;
 
-    wire [AXI_ADDR_WIDTH-1:0] ar_burst_bytes =
+    function automatic [AXI_ADDR_WIDTH-1:0] get_wrap_boundary;
-        ({{(AXI_ADDR_WIDTH-8){1'b0}}, s_arlen} + {{(AXI_ADDR_WIDTH-1){1'b0}}, 1'b1}) << LOG_NR_BYTES;
+        input [AXI_ADDR_WIDTH-1:0] unaligned_address;
-    wire [AXI_ADDR_WIDTH-1:0] aw_burst_bytes =
+        input [7:0] len;
-        ({{(AXI_ADDR_WIDTH-8){1'b0}}, s_awlen} + {{(AXI_ADDR_WIDTH-1){1'b0}}, 1'b1}) << LOG_NR_BYTES;
+    begin
-    wire [AXI_ADDR_WIDTH-1:0] ar_wrap_mask = ar_burst_bytes - {{(AXI_ADDR_WIDTH-1){1'b0}}, 1'b1};
+        get_wrap_boundary = 'h0;
-    wire [AXI_ADDR_WIDTH-1:0] aw_wrap_mask = aw_burst_bytes - {{(AXI_ADDR_WIDTH-1){1'b0}}, 1'b1};
+        //  for wrapping transfers ax_len can only be of size 1, 3, 7 or 15
+        if (len == 4'b1)
+            get_wrap_boundary[AXI_ADDR_WIDTH-1:1+LOG_NR_BYTES] = unaligned_address[AXI_ADDR_WIDTH-1:1+LOG_NR_BYTES];
+        else if (len == 4'b11)
+            get_wrap_boundary[AXI_ADDR_WIDTH-1:2+LOG_NR_BYTES] = unaligned_address[AXI_ADDR_WIDTH-1:2+LOG_NR_BYTES];
+        else if (len == 4'b111)
+            get_wrap_boundary[AXI_ADDR_WIDTH-1:3+LOG_NR_BYTES] = unaligned_address[AXI_ADDR_WIDTH-3:2+LOG_NR_BYTES];
+        else if (len == 4'b1111)
+            get_wrap_boundary[AXI_ADDR_WIDTH-1:4+LOG_NR_BYTES] = unaligned_address[AXI_ADDR_WIDTH-3:4+LOG_NR_BYTES];
+    end
+    endfunction
 
-    reg [AXI_ADDR_WIDTH-1:0] next_addr;
+    reg [AXI_ADDR_WIDTH-1:0] aligned_address;
-    reg [AXI_ADDR_WIDTH-1:0] seq_addr;
+    reg [AXI_ADDR_WIDTH-1:0] wrap_boundary;
+    reg [AXI_ADDR_WIDTH-1:0] upper_wrap_boundary;
+    reg [AXI_ADDR_WIDTH-1:0] cons_addr;
 
     always @ (*) begin
-        seq_addr = req_addr_q + ADDR_INCR;
+        // address generation
-        case (ax_req_q_burst)
+        aligned_address = {ax_req_q_addr[AXI_ADDR_WIDTH-1:LOG_NR_BYTES], {{LOG_NR_BYTES}{1'b0}}};
-            FIXED: next_addr = req_addr_q;
+        wrap_boundary = get_wrap_boundary(ax_req_q_addr, ax_req_q_len);
-            INCR:  next_addr = seq_addr;
+        // this will overflow
-            WRAP:  next_addr = (req_addr_q & ~wrap_mask_q) | (seq_addr & wrap_mask_q);
+        upper_wrap_boundary = wrap_boundary + ((ax_req_q_len + 1) << LOG_NR_BYTES);
-            default: next_addr = seq_addr;
+        // calculate consecutive address
-        endcase
+        cons_addr = aligned_address + (cnt_q << LOG_NR_BYTES);
 
         // Transaction attributes
         // default assignments
@@ -141,16 +145,12 @@ module axi2sram_sp_external #(
         ax_req_d_burst  = ax_req_q_burst;
         req_addr_d      = req_addr_q;
         cnt_d           = cnt_q;
-        wrap_mask_d     = wrap_mask_q;
-        req_bank_d      = req_bank_q;
-        read_capture    = 1'b0;
         // Memory default assignments
         data_o = s_wdata;
         be_o   = s_wstrb;
         we_o   = 1'b0;
         req_o  = 1'b0;
         addr_o = 'h0;
-        bank_o = req_bank_q;
         // AXI assignments
         // request
         s_awready = 1'b0;
@@ -183,17 +183,14 @@ module axi2sram_sp_external #(
                     ax_req_d_len    = s_arlen;
                     ax_req_d_size   = s_arsize;
                     ax_req_d_burst  = s_arburst;
-                    wrap_mask_d     = ar_wrap_mask;
+                    state_d        = READ;
-                    req_bank_d      = s_araddr[22];
-                    state_d        = READ_ADDR;
                     //  we can request the first address, this saves us time
                     req_o          = 1'b1;
                     addr_o         = s_araddr;
-                    bank_o         = s_araddr[22];
                     // save the address
                     req_addr_d     = s_araddr;
                     // save the ar_len
-                    cnt_d          = 0;
+                    cnt_d          = 1;
                 // ------------
                 // Write
                 // ------------
@@ -207,9 +204,6 @@ module axi2sram_sp_external #(
                     ax_req_d_len    = s_awlen;
                     ax_req_d_size   = s_awsize;
                     ax_req_d_burst  = s_awburst;
-                    wrap_mask_d     = aw_wrap_mask;
-                    req_bank_d      = s_awaddr[22];
-                    bank_o          = s_awaddr[22];
                     // we've got our first w_valid so start the write process
                     if (s_wvalid) begin
                         req_o          = 1'b1;
@@ -236,65 +230,56 @@ module axi2sram_sp_external #(
             end
 
             READ: begin
+                // keep request to memory high
+                req_o  = 1'b1;
+                addr_o = req_addr_q;
                 // send the response
                 s_rvalid = 1'b1;
-                s_rdata  = read_data_q;
+                s_rdata  = data_i;
                 s_rid    = ax_req_q_id;
-                s_rlast  = (cnt_q == ({1'b0, ax_req_q_len} + 9'd1));
+                s_rlast  = (cnt_q == ax_req_q_len + 1);
-                if (FAST_READ_BURST && !s_rlast) begin
-                    req_o  = 1'b1;
-                    addr_o = next_addr;
-                end
                 // check that the master is ready, the slave must not wait on this
                 if (s_rready) begin
                     // we sent the last byte -> go back to idle
                     if (s_rlast) begin
-                        if (s_arvalid) begin
-                            // Accept the next burst without an intervening
-                            // IDLE cycle and start its SRAM address phase.
-                            s_arready       = 1'b1;
-                            ax_req_d_id     = s_arid;
-                            ax_req_d_addr   = s_araddr;
-                            ax_req_d_len    = s_arlen;
-                            ax_req_d_size   = s_arsize;
-                            ax_req_d_burst  = s_arburst;
-                            wrap_mask_d     = ar_wrap_mask;
-                            req_bank_d      = s_araddr[22];
-                            req_o           = 1'b1;
-                            addr_o          = s_araddr;
-                            bank_o          = s_araddr[22];
-                            req_addr_d      = s_araddr;
-                            // ARVALID was presented during the preceding
-                            // beat, so the async-SRAM address has a full
-                            // cycle to settle before this capture edge.
-                            read_capture    = 1'b1;
-                            cnt_d           = 1;
-                            state_d         = READ;
-                        end else begin
                         state_d = IDLE;
-                        end
+                        // we already got everything
                     end
                     else begin
-                        if (FAST_READ_BURST) begin
-                            read_capture = 1'b1;
-                            req_addr_d = next_addr;
-                            cnt_d   = cnt_q + 1;
-                            state_d = READ;
-                        end else begin
-                            req_addr_d = next_addr;
                         state_d = READ_ADDR;
                     end
                 end
             end
-            end
 
             READ_ADDR: begin
-                // Fill the read data register after the async-SRAM address has
+                // keep request to memory high
-                // been stable for one full sys_clk cycle.
                 req_o  = 1'b1;
-                addr_o = req_addr_q;
+                // send the response
-                read_capture = 1'b1;
+                s_rvalid = 1'b0;
+                // ----------------------------
+                // Next address generation
+                // ----------------------------
+                // handle the correct burst type
+                case (ax_req_q_burst)
+                    FIXED, INCR: addr_o = cons_addr;
+                    WRAP:  begin
+                        // check if the address reached warp boundary
+                        if (cons_addr == upper_wrap_boundary) begin
+                            addr_o = wrap_boundary;
+                        // address warped beyond boundary
+                        end else if (cons_addr > upper_wrap_boundary) begin
+                            addr_o = ax_req_q_addr + ((cnt_q - ax_req_q_len) << LOG_NR_BYTES);
+                        // we are still in the incremental regime
+                        end else begin
+                            addr_o = cons_addr;
+                        end
+                    end
+                endcase
+                // we need to change the address here for the upcoming request
+                // we can decrease the counter as the master has consumed the read data
                 cnt_d = cnt_q + 1;
+                // save the request address for the next cycle
+                req_addr_d = addr_o;
                 state_d = READ;
             end
 
@@ -312,9 +297,24 @@ module axi2sram_sp_external #(
                     // Next address generation
                     // ----------------------------
                     // handle the correct burst type
-                    addr_o = next_addr;
+                    case (ax_req_q_burst)
+
+                        FIXED, INCR: addr_o = cons_addr;
+                        WRAP:  begin
+                            // check if the address reached warp boundary
+                            if (cons_addr == upper_wrap_boundary) begin
+                                addr_o = wrap_boundary;
+                            // address warped beyond boundary
+                            end else if (cons_addr > upper_wrap_boundary) begin
+                                addr_o = ax_req_q_addr + ((cnt_q - ax_req_q_len) << LOG_NR_BYTES);
+                            // we are still in the incremental regime
+                            end else begin
+                                addr_o = cons_addr;
+                            end
+                        end
+                    endcase
                     // save the request address for the next cycle
-                    req_addr_d = next_addr;
+                    req_addr_d = addr_o;
                     // we can decrease the counter as the master has consumed the read data
                     cnt_d = cnt_q + 1;
 
@@ -345,10 +345,7 @@ module axi2sram_sp_external #(
             ax_req_q_len    <= 8'h0;
             ax_req_q_size   <= 3'h0;
             req_addr_q      <= 'h0;
-            cnt_q           <= 9'h0;
+            cnt_q           <= 8'h0;
-            wrap_mask_q     <= 'h0;
-            req_bank_q      <= 1'b0;
-            read_data_q     <= 'h0;
         end else begin
             state_q         <= state_d;
             ax_req_q_addr   <= ax_req_d_addr;
@@ -358,10 +355,6 @@ module axi2sram_sp_external #(
             ax_req_q_size   <= ax_req_d_size;
             req_addr_q      <= req_addr_d;
             cnt_q           <= cnt_d;
-            wrap_mask_q     <= wrap_mask_d;
-            req_bank_q      <= req_bank_d;
-            if (read_capture)
-                read_data_q <= data_i;
         end
     end
 

rtl/ip/PLL_2019_2/clk_pll.xci

@@ -101,7 +101,7 @@
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT0_ACTUAL_FREQ">32.97872</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT1_1">0000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT1_2">0000</spirit:configurableElementValue>
-        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT1_ACTUAL_FREQ">59.61538</spirit:configurableElementValue>
+        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT1_ACTUAL_FREQ">50.00000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT1_DRIVES">BUFG</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT1_DUTY_CYCLE">50.0</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT1_MATCHED_ROUTING">false</spirit:configurableElementValue>
@@ -117,10 +117,10 @@
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_DRIVES">BUFG</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_DUTY_CYCLE">50.0</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_MATCHED_ROUTING">false</spirit:configurableElementValue>
-        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_OUT_FREQ">59.61538</spirit:configurableElementValue>
+        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_OUT_FREQ">50.00000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_PHASE">0.000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_REQUESTED_DUTY_CYCLE">50.000</spirit:configurableElementValue>
-        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_REQUESTED_OUT_FREQ">60</spirit:configurableElementValue>
+        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_REQUESTED_OUT_FREQ">50.000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_REQUESTED_PHASE">0.000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_SEQUENCE_NUMBER">1</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_CLKOUT2_USED">1</spirit:configurableElementValue>
@@ -203,7 +203,7 @@
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_DIN_PORT">din</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_DIVCLK">0000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_DIVIDE1_AUTO">1</spirit:configurableElementValue>
-        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_DIVIDE2_AUTO">0.55</spirit:configurableElementValue>
+        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_DIVIDE2_AUTO">0.66</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_DIVIDE3_AUTO">0.33</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_DIVIDE4_AUTO">0.33</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_DIVIDE5_AUTO">0.33</spirit:configurableElementValue>
@@ -255,7 +255,7 @@
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_MMCM_CLKOUT0_DUTY_CYCLE">0.500</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_MMCM_CLKOUT0_PHASE">0.000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_MMCM_CLKOUT0_USE_FINE_PS">FALSE</spirit:configurableElementValue>
-        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_MMCM_CLKOUT1_DIVIDE">26</spirit:configurableElementValue>
+        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_MMCM_CLKOUT1_DIVIDE">31</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_MMCM_CLKOUT1_DUTY_CYCLE">0.500</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_MMCM_CLKOUT1_PHASE">0.000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_MMCM_CLKOUT1_USE_FINE_PS">FALSE</spirit:configurableElementValue>
@@ -293,7 +293,7 @@
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_OUTCLK_SUM_ROW0A"> Output     Output      Phase    Duty Cycle   Pk-to-Pk     Phase</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_OUTCLK_SUM_ROW0B">  Clock     Freq (MHz)  (degrees)    (%)     Jitter (ps)  Error (ps)</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_OUTCLK_SUM_ROW1">_cpu_clk__32.97872______0.000______50.0______130.093____123.600</spirit:configurableElementValue>
-        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_OUTCLK_SUM_ROW2">_sys_clk__59.61538______0.000______50.0______118.808____123.600</spirit:configurableElementValue>
+        <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_OUTCLK_SUM_ROW2">_sys_clk__50.00000______0.000______50.0______122.035____123.600</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_OUTCLK_SUM_ROW3">no_CLK_OUT3_output</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_OUTCLK_SUM_ROW4">no_CLK_OUT4_output</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="MODELPARAM_VALUE.C_OUTCLK_SUM_ROW5">no_CLK_OUT5_output</spirit:configurableElementValue>
@@ -423,11 +423,11 @@
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT1_SEQUENCE_NUMBER">1</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT1_USED">true</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_DRIVES">BUFG</spirit:configurableElementValue>
-        <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_JITTER">118.808</spirit:configurableElementValue>
+        <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_JITTER">122.035</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_MATCHED_ROUTING">false</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_PHASE_ERROR">123.600</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_REQUESTED_DUTY_CYCLE">50.000</spirit:configurableElementValue>
-        <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_REQUESTED_OUT_FREQ">60</spirit:configurableElementValue>
+        <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_REQUESTED_OUT_FREQ">50.000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_REQUESTED_PHASE">0.000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_SEQUENCE_NUMBER">1</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.CLKOUT2_USED">true</spirit:configurableElementValue>
@@ -534,7 +534,7 @@
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.MMCM_CLKOUT0_DUTY_CYCLE">0.500</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.MMCM_CLKOUT0_PHASE">0.000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.MMCM_CLKOUT0_USE_FINE_PS">false</spirit:configurableElementValue>
-        <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.MMCM_CLKOUT1_DIVIDE">26</spirit:configurableElementValue>
+        <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.MMCM_CLKOUT1_DIVIDE">31</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.MMCM_CLKOUT1_DUTY_CYCLE">0.500</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.MMCM_CLKOUT1_PHASE">0.000</spirit:configurableElementValue>
         <spirit:configurableElementValue spirit:referenceId="PARAM_VALUE.MMCM_CLKOUT1_USE_FINE_PS">false</spirit:configurableElementValue>

rtl/ip/matmul/matmul_axi_slave.v

@@ -1,5 +1,7 @@
+// ═══════════════════════════════════════════════════════
 //  3-stage pipelined multiplier (LUT-based, no DSP)
 //  latency=3, throughput=1 result/cycle
+// ═══════════════════════════════════════════════════════
 (* use_dsp = "no", use_dsp48 = "no", syn_multstyle = "logic" *)
 module pipelined_mult_32x32 (
     input             clk,
@@ -41,7 +43,9 @@ module pipelined_mult_32x32 (
 endmodule
 
 
-// AXI Burst Slave + Ping-Pong + Stream CRC32
+// ═══════════════════════════════════════════════════════
+//  矩阵乘法器顶层模块 (AXI Burst Slave + Ping-Pong + Stream CRC32)
+// ═══════════════════════════════════════════════════════
 module matmul_axi_slave (
     input             clk,
     input             resetn,
@@ -92,6 +96,21 @@ module matmul_axi_slave (
     output reg        s_axi_rvalid,
     input             s_axi_rready,
 
+    // CDMA local endpoint port
+    input             ep_wr_valid,
+    input      [3:0]  ep_wr_id,
+    input      [3:0]  ep_wr_bank,
+    input      [7:0]  ep_wr_word_addr,
+    input      [31:0] ep_wr_data,
+    output            ep_wr_ready,
+
+    input             ep_rd_valid,
+    input      [3:0]  ep_rd_id,
+    input      [3:0]  ep_rd_bank,
+    input      [7:0]  ep_rd_word_addr,
+    output reg [31:0] ep_rd_data,
+    output            ep_rd_ready,
+
     // AXI master port for autonomous matrix stream load/store
     output reg [3:0]  m_axi_awid,
     output reg [31:0] m_axi_awaddr,
@@ -138,7 +157,7 @@ module matmul_axi_slave (
 );
 
     // ==========================================
-    // Memory Map
+    // 内存地址映射 (Memory Map)
     // ==========================================
     localparam ADDR_CTRL    = 12'h000;
     localparam ADDR_STATUS  = 12'h004;
@@ -148,7 +167,6 @@ module matmul_axi_slave (
     localparam ADDR_DMA_GROUPS     = 12'h028;
     localparam ADDR_DMA_SRC_STRIDE = 12'h02C;
     localparam ADDR_DMA_DST_STRIDE = 12'h030;
-    localparam ADDR_DMA_CYCLES     = 12'h034;
     
     localparam ADDR_B0_A    = 12'h100; 
     localparam ADDR_B0_B    = 12'h140; 
@@ -159,7 +177,7 @@ module matmul_axi_slave (
     localparam ADDR_B1_C    = 12'h380; 
 
     // ==========================================
-    // Registers and data buffers
+    // 寄存器与数据缓冲
     // ==========================================
     reg        req_0, req_1;
     reg        busy;
@@ -180,11 +198,8 @@ module matmul_axi_slave (
     reg [31:0] dma_groups;
     reg [31:0] dma_src_stride;
     reg [31:0] dma_dst_stride;
-    // Full sys_clk cycles spent in dma_busy; held after DMA completion.
-    reg [31:0] dma_cycle_count;
     reg        dma_busy;
     reg        dma_done;
-    reg        dma_compute_complete;
     reg [31:0] load_group;
     reg [31:0] store_group;
     reg [31:0] dma_next_rd_addr;
@@ -202,46 +217,32 @@ module matmul_axi_slave (
     reg        dma_req0_pulse;
     reg        dma_req1_pulse;
 
-    function [3:0] c_elem_from_word;
+    assign ep_wr_ready = 1'b1;
-        input [5:0] word_idx;
+    assign ep_rd_ready = 1'b1;
-        begin
-            case (word_idx)
-                6'd0,  6'd1,  6'd2:  c_elem_from_word = 4'd0;
-                6'd3,  6'd4,  6'd5:  c_elem_from_word = 4'd1;
-                6'd6,  6'd7,  6'd8:  c_elem_from_word = 4'd2;
-                6'd9,  6'd10, 6'd11: c_elem_from_word = 4'd3;
-                6'd12, 6'd13, 6'd14: c_elem_from_word = 4'd4;
-                6'd15, 6'd16, 6'd17: c_elem_from_word = 4'd5;
-                6'd18, 6'd19, 6'd20: c_elem_from_word = 4'd6;
-                6'd21, 6'd22, 6'd23: c_elem_from_word = 4'd7;
-                6'd24, 6'd25, 6'd26: c_elem_from_word = 4'd8;
-                6'd27, 6'd28, 6'd29: c_elem_from_word = 4'd9;
-                6'd30, 6'd31, 6'd32: c_elem_from_word = 4'd10;
-                6'd33, 6'd34, 6'd35: c_elem_from_word = 4'd11;
-                6'd36, 6'd37, 6'd38: c_elem_from_word = 4'd12;
-                6'd39, 6'd40, 6'd41: c_elem_from_word = 4'd13;
-                6'd42, 6'd43, 6'd44: c_elem_from_word = 4'd14;
-                default:             c_elem_from_word = 4'd15;
-            endcase
-        end
-    endfunction
 
-    function [1:0] c_part_from_word;
+    wire [3:0] ep_wr_lo_idx = ep_wr_word_addr[3:0];
-        input [5:0] word_idx;
+    wire [3:0] ep_wr_hi_idx = ep_wr_word_addr[3:0];
-        begin
+    wire [7:0] ep_rd_elem_raw = ep_rd_word_addr / 8'd3;
-            case (word_idx)
+    wire [7:0] ep_rd_part_raw = ep_rd_word_addr % 8'd3;
-                6'd0,  6'd3,  6'd6,  6'd9,
+    wire [3:0] ep_rd_elem = ep_rd_elem_raw[3:0];
-                6'd12, 6'd15, 6'd18, 6'd21,
+    wire [1:0] ep_rd_part = ep_rd_part_raw[1:0];
-                6'd24, 6'd27, 6'd30, 6'd33,
+    wire [65:0] ep_rd_c0 = b0_c[ep_rd_elem];
-                6'd36, 6'd39, 6'd42, 6'd45: c_part_from_word = 2'd0;
+    wire [65:0] ep_rd_c1 = b1_c[ep_rd_elem];
-                6'd1,  6'd4,  6'd7,  6'd10,
+
-                6'd13, 6'd16, 6'd19, 6'd22,
+    always @(*) begin
-                6'd25, 6'd28, 6'd31, 6'd34,
+        ep_rd_data = 32'b0;
-                6'd37, 6'd40, 6'd43, 6'd46: c_part_from_word = 2'd1;
+        if (ep_rd_word_addr < 8'd48) begin
-                default:                     c_part_from_word = 2'd2;
+            if (ep_rd_bank[0] == 1'b0) begin
-            endcase
+                if (ep_rd_part == 2'd0)      ep_rd_data = ep_rd_c0[31:0];
+                else if (ep_rd_part == 2'd1) ep_rd_data = ep_rd_c0[63:32];
+                else                         ep_rd_data = {30'b0, ep_rd_c0[65:64]};
+            end else begin
+                if (ep_rd_part == 2'd0)      ep_rd_data = ep_rd_c1[31:0];
+                else if (ep_rd_part == 2'd1) ep_rd_data = ep_rd_c1[63:32];
+                else                         ep_rd_data = {30'b0, ep_rd_c1[65:64]};
+            end
+        end
     end
-    endfunction
 
     function [31:0] apply_wstrb;
         input [31:0] old_value;
@@ -258,70 +259,29 @@ module matmul_axi_slave (
 
     // ==========================================
     // CRC32 Engine (IEEE 802.3, polynomial 0xEDB88320)
-    //
-    // AXI-stream: single-cycle crc32_update_word (4 LUT lookup depth, ok for
-    //   AXI timing—the bus runs at or below internal compute rate).
-    // DMA-calc: two 6-byte CRC steps per 66-bit C-element.  A result FIFO
-    //   absorbs the 16-cycle compute burst so the recursive CRC state never
-    //   backpressures the multiplier under the normal DMA load cadence.
     // ==========================================
     reg [31:0] crc32_table [0:255];
     reg [31:0] crc_reg;
 
-    // ---- AXI-stream CRC (single-cycle) ----------------------------------
+    wire        crc_dma_valid;
-    wire [31:0] crc_stream_data = s_axi_rdata;
+    wire [31:0] crc_dma_data;
+    wire [31:0] crc_stream_data = crc_dma_valid ? crc_dma_data :
+                                  (ep_rd_valid ? ep_rd_data : s_axi_rdata);
     wire        crc_axi_valid = (r_state == R_DATA) && s_axi_rvalid && s_axi_rready &&
                                 ((r_offset >= ADDR_B0_C && r_offset < (ADDR_B0_C + 12'h0C0)) ||
                                  (r_offset >= ADDR_B1_C && r_offset < (ADDR_B1_C + 12'h0C0)));
-    wire        crc_stream_valid = crc_axi_valid;
+    wire        crc_ep_valid = ep_rd_valid && ep_rd_ready && (ep_rd_word_addr < 8'd48);
-    wire [31:0] crc_next = crc32_update_word(crc_reg, crc_stream_data);
+    wire        crc_stream_valid = crc_axi_valid || crc_ep_valid || crc_dma_valid;
 
-    function [31:0] crc32_update_word;
+    wire [31:0] crc_t0, crc_c0, crc_t1, crc_c1, crc_t2, crc_c2, crc_t3, crc_next;
-        input [31:0] crc_in;
+    assign crc_t0 = crc32_table[(crc_reg[7:0]   ^ crc_stream_data[7:0])];
-        input [31:0] data_in;
+    assign crc_c0 = crc_t0 ^ {8'h00, crc_reg[31:8]};
-        reg [31:0] c0;
+    assign crc_t1 = crc32_table[(crc_c0[7:0]    ^ crc_stream_data[15:8])];
-        reg [31:0] c1;
+    assign crc_c1 = crc_t1 ^ {8'h00, crc_c0[31:8]};
-        reg [31:0] c2;
+    assign crc_t2 = crc32_table[(crc_c1[7:0]    ^ crc_stream_data[23:16])];
-        begin
+    assign crc_c2 = crc_t2 ^ {8'h00, crc_c1[31:8]};
-            c0 = crc32_table[(crc_in[7:0] ^ data_in[7:0])] ^ {8'h00, crc_in[31:8]};
+    assign crc_t3 = crc32_table[(crc_c2[7:0]    ^ crc_stream_data[31:24])];
-            c1 = crc32_table[(c0[7:0]     ^ data_in[15:8])] ^ {8'h00, c0[31:8]};
+    assign crc_next = crc_t3 ^ {8'h00, crc_c2[31:8]};
-            c2 = crc32_table[(c1[7:0]     ^ data_in[23:16])] ^ {8'h00, c1[31:8]};
-            crc32_update_word = crc32_table[(c2[7:0] ^ data_in[31:24])] ^ {8'h00, c2[31:8]};
-        end
-    endfunction
-
-    function [31:0] crc32_update_6bytes;
-        input [31:0] crc_in;
-        input [47:0] data_in;
-        reg [31:0] c;
-        integer byte_idx;
-        begin
-            c = crc_in;
-            for (byte_idx = 0; byte_idx < 6; byte_idx = byte_idx + 1)
-                c = crc32_table[(c[7:0] ^ data_in[byte_idx*8 +: 8])] ^
-                    {8'h00, c[31:8]};
-            crc32_update_6bytes = c;
-        end
-    endfunction
-
-    // ---- DMA result FIFO and two-cycle recursive CRC consumer -----------
-    localparam CRC_FIFO_DEPTH = 32;
-    reg [65:0] crc_fifo [0:CRC_FIFO_DEPTH-1];
-    reg [4:0]  crc_fifo_wr_ptr;
-    reg [4:0]  crc_fifo_rd_ptr;
-    reg [5:0]  crc_fifo_count;
-    reg        crc_c66_busy;
-    reg [65:0] crc_c66_data;
-    reg [31:0] crc_c66_half;
-
-    wire        crc_fifo_empty = (crc_fifo_count == 6'd0);
-    wire        crc_fifo_push = dma_busy && acc_valid_d4;
-    wire        crc_fifo_pop = !crc_c66_busy && !crc_fifo_empty;
-    wire [31:0] crc_c66_first_next =
-        crc32_update_6bytes(crc_reg, crc_fifo[crc_fifo_rd_ptr][47:0]);
-    wire [47:0] crc_c66_second_data = {30'b0, crc_c66_data[65:48]};
-    wire [31:0] crc_c66_final_next =
-        crc32_update_6bytes(crc_c66_half, crc_c66_second_data);
 
     initial begin
         crc32_table[0  ] = 32'h00000000; crc32_table[1  ] = 32'h77073096;
@@ -455,7 +415,7 @@ module matmul_axi_slave (
     end
 
     // ==========================================
-    // AXI Write (INCR Burst)
+    // AXI Write 通道逻辑 (支持 INCR Burst)
     // ==========================================
     localparam W_IDLE = 2'd0;
     localparam W_DATA = 2'd1;
@@ -471,6 +431,7 @@ module matmul_axi_slave (
     assign s_axi_bid   = w_id;
     assign s_axi_bresp = 2'b00;
 
+    // 前向声明引擎的状态
     localparam C_IDLE      = 2'd0;
     localparam C_RUN       = 2'd1;
     reg [1:0] state;
@@ -481,19 +442,10 @@ module matmul_axi_slave (
     localparam DMA_RD_ADDR = 2'd1;
     localparam DMA_RD_DATA = 2'd2;
     reg [1:0] dma_rd_state;
-    reg [7:0] dma_rd_count;
+    reg [4:0] dma_rd_count;
     reg       dma_rd_bank;
-    wire [4:0] dma_rd_group_word_idx = dma_rd_count[4:0];
     wire [3:0] dma_rd_word_idx = dma_rd_count[3:0];
-    wire       dma_rd_fill_bank = dma_rd_bank ^ dma_rd_count[5];
     wire       dma_rd_commit = (dma_rd_state == DMA_RD_DATA) && m_axi_rvalid && m_axi_rready;
-    wire       dma_bank0_free = !bank_loaded[0] && !bank_compute_pending[0] && !req_0 &&
-                                !(state == C_RUN && cur_bank == 1'b0);
-    wire       dma_bank1_free = !bank_loaded[1] && !bank_compute_pending[1] && !req_1 &&
-                                !(state == C_RUN && cur_bank == 1'b1);
-    wire       dma_load_bank_free = dma_load_bank ? dma_bank1_free : dma_bank0_free;
-    wire       dma_rd_fill_bank_free = dma_rd_fill_bank ? dma_bank1_free : dma_bank0_free;
-    wire       dma_rd_next_bank_free = dma_rd_fill_bank ? dma_bank0_free : dma_bank1_free;
 
     localparam DMA_WR_IDLE = 2'd0;
     localparam DMA_WR_ADDR = 2'd1;
@@ -502,6 +454,31 @@ module matmul_axi_slave (
     reg [1:0] dma_wr_state;
     reg [5:0] dma_wr_count;
 
+    wire [7:0] dma_wr_elem_raw = dma_wr_count / 6'd3;
+    wire [7:0] dma_wr_part_raw = dma_wr_count % 6'd3;
+    wire [3:0] dma_wr_elem = dma_wr_elem_raw[3:0];
+    wire [1:0] dma_wr_part = dma_wr_part_raw[1:0];
+    wire [65:0] dma_wr_c0 = b0_c[dma_wr_elem];
+    wire [65:0] dma_wr_c1 = b1_c[dma_wr_elem];
+    wire [65:0] dma_wr_c = dma_store_bank ? dma_wr_c1 : dma_wr_c0;
+    wire [31:0] dma_wr_word =
+        (dma_wr_part == 2'd0) ? dma_wr_c[31:0] :
+        (dma_wr_part == 2'd1) ? dma_wr_c[63:32] :
+                                {30'b0, dma_wr_c[65:64]};
+    wire [5:0] dma_wr_next_count = dma_wr_count + 6'd1;
+    wire [7:0] dma_wr_next_elem_raw = dma_wr_next_count / 6'd3;
+    wire [7:0] dma_wr_next_part_raw = dma_wr_next_count % 6'd3;
+    wire [3:0] dma_wr_next_elem = dma_wr_next_elem_raw[3:0];
+    wire [1:0] dma_wr_next_part = dma_wr_next_part_raw[1:0];
+    wire [65:0] dma_wr_next_c0 = b0_c[dma_wr_next_elem];
+    wire [65:0] dma_wr_next_c1 = b1_c[dma_wr_next_elem];
+    wire [65:0] dma_wr_next_c = dma_store_bank ? dma_wr_next_c1 : dma_wr_next_c0;
+    wire [31:0] dma_wr_word_next =
+        (dma_wr_next_part == 2'd0) ? dma_wr_next_c[31:0] :
+        (dma_wr_next_part == 2'd1) ? dma_wr_next_c[63:32] :
+                                     {30'b0, dma_wr_next_c[65:64]};
+    
+    // 【修改点 1】单独引出一个写脉冲信号给 CRC 清零用
     wire crc_clear_pulse = (w_state == W_DATA) && s_axi_wvalid && s_axi_wready && (w_offset == ADDR_CRC);
     wire done_clear_pulse = (w_state == W_DATA) && s_axi_wvalid && s_axi_wready &&
                             (w_offset == ADDR_STATUS) && s_axi_wdata[1];
@@ -527,14 +504,32 @@ module matmul_axi_slave (
             dma_start_pulse <= 1'b0;
             dma_done_clear_pulse <= 1'b0;
 
+            if (ep_wr_valid && ep_wr_ready) begin
+                if (ep_wr_bank[0] == 1'b0) begin
+                    if (ep_wr_word_addr < 8'd16)
+                        b0_a[ep_wr_lo_idx] <= ep_wr_data;
+                    else if (ep_wr_word_addr < 8'd32)
+                        b0_b[ep_wr_hi_idx] <= ep_wr_data;
+                    else
+                        error <= 1'b1;
+                end else begin
+                    if (ep_wr_word_addr < 8'd16)
+                        b1_a[ep_wr_lo_idx] <= ep_wr_data;
+                    else if (ep_wr_word_addr < 8'd32)
+                        b1_b[ep_wr_hi_idx] <= ep_wr_data;
+                    else
+                        error <= 1'b1;
+                end
+            end
+
             if (dma_rd_commit) begin
-                if (dma_rd_fill_bank == 1'b0) begin
+                if (dma_rd_bank == 1'b0) begin
-                    if (dma_rd_group_word_idx < 5'd16)
+                    if (dma_rd_count < 5'd16)
                         b0_a[dma_rd_word_idx] <= m_axi_rdata;
                     else
                         b0_b[dma_rd_word_idx] <= m_axi_rdata;
                 end else begin
-                    if (dma_rd_group_word_idx < 5'd16)
+                    if (dma_rd_count < 5'd16)
                         b1_a[dma_rd_word_idx] <= m_axi_rdata;
                     else
                         b1_b[dma_rd_word_idx] <= m_axi_rdata;
@@ -579,7 +574,7 @@ module matmul_axi_slave (
                             dma_src_stride <= apply_wstrb(dma_src_stride, s_axi_wdata, s_axi_wstrb);
                         else if (w_offset == ADDR_DMA_DST_STRIDE)
                             dma_dst_stride <= apply_wstrb(dma_dst_stride, s_axi_wdata, s_axi_wstrb);
-                        // CRC register is now managed by the pipelined CRC engine
+                        // 【修改点 2】删除 else if (w_offset == ADDR_CRC) { crc_reg <= ... } 防止多重驱动
                         else if (w_offset >= ADDR_B0_A && w_offset < ADDR_B0_B)
                             b0_a[(w_offset - ADDR_B0_A) >> 2] <= apply_wstrb(b0_a[(w_offset - ADDR_B0_A) >> 2], s_axi_wdata, s_axi_wstrb);
                         else if (w_offset >= ADDR_B0_B && w_offset < ADDR_B0_C)
@@ -620,7 +615,7 @@ module matmul_axi_slave (
     end
 
     // ==========================================
-    // AXI Read (INCR Burst)
+    // AXI Read 通道逻辑 (支持 INCR Burst)
     // ==========================================
     localparam R_IDLE = 2'd0;
     localparam R_DATA = 2'd1;
@@ -635,13 +630,13 @@ module matmul_axi_slave (
     assign s_axi_rresp = 2'b00;
 
     wire [5:0] c0_offset = (r_offset - ADDR_B0_C) >> 2;
-    wire [3:0] c0_elem   = c_elem_from_word(c0_offset);
+    wire [3:0] c0_elem   = c0_offset / 3;
-    wire [1:0] c0_part   = c_part_from_word(c0_offset);
+    wire [1:0] c0_part   = c0_offset % 3;
     wire [65:0] target_c0 = b0_c[c0_elem];
 
     wire [5:0] c1_offset = (r_offset - ADDR_B1_C) >> 2;
-    wire [3:0] c1_elem   = c_elem_from_word(c1_offset);
+    wire [3:0] c1_elem   = c1_offset / 3;
-    wire [1:0] c1_part   = c_part_from_word(c1_offset);
+    wire [1:0] c1_part   = c1_offset % 3;
     wire [65:0] target_c1 = b1_c[c1_elem];
 
     wire crc_read_en = (r_offset >= ADDR_B0_C && r_offset < (ADDR_B0_C + 12'h0C0)) ||
@@ -672,6 +667,7 @@ module matmul_axi_slave (
                 end
                 R_DATA: begin
                     if (s_axi_rvalid && s_axi_rready) begin
+                        // 【修改点 3】删掉直接赋值 crc_stream_valid 的逻辑，移至专用 Pipeline 块
                         if (r_cnt == r_len) begin
                             s_axi_rvalid  <= 1'b0;
                             s_axi_rlast   <= 1'b0;
@@ -699,8 +695,7 @@ module matmul_axi_slave (
             else if (current_r_offset == ADDR_CRC)
                 s_axi_rdata = crc_reg ^ 32'hFFFFFFFF;
             else if (current_r_offset == ADDR_STATUS)
-                s_axi_rdata = {26'b0, (crc_c66_busy || !crc_fifo_empty),
+                s_axi_rdata = {27'b0, dma_done, dma_busy, error, done, busy};
-                               dma_done, dma_busy, error, done, busy};
             else if (current_r_offset == ADDR_DMA_SRC_BASE)
                 s_axi_rdata = dma_src_base;
             else if (current_r_offset == ADDR_DMA_DST_BASE)
@@ -711,8 +706,6 @@ module matmul_axi_slave (
                 s_axi_rdata = dma_src_stride;
             else if (current_r_offset == ADDR_DMA_DST_STRIDE)
                 s_axi_rdata = dma_dst_stride;
-            else if (current_r_offset == ADDR_DMA_CYCLES)
-                s_axi_rdata = dma_cycle_count;
             else if (current_r_offset >= ADDR_B0_A && current_r_offset < ADDR_B0_B)
                 s_axi_rdata = b0_a[(current_r_offset - ADDR_B0_A) >> 2];
             else if (current_r_offset >= ADDR_B0_B && current_r_offset < ADDR_B0_C)
@@ -734,8 +727,19 @@ module matmul_axi_slave (
         end
     end
     
+    // CRC register: update on each C data read beat, clear on write to 0x010
+    always @(posedge clk) begin
+        if (!resetn) begin
+            crc_reg <= 32'hFFFFFFFF;
+        end else if (crc_clear_pulse) begin
+            crc_reg <= 32'hFFFFFFFF;
+        end else if (crc_stream_valid) begin
+            crc_reg <= crc_next;
+        end
+    end
+
     // ==========================================
-    // Compute FSM
+    // 计算引擎 (Compute FSM)
     // ==========================================
     wire [63:0] mult_p    [0:3];
     reg  [31:0] m_in_a    [0:3];
@@ -759,56 +763,6 @@ module matmul_axi_slave (
     reg       acc_bank_d1, acc_bank_d2, acc_bank_d3, acc_bank_d4;
     reg       acc_valid_d1, acc_valid_d2, acc_valid_d3, acc_valid_d4;
     reg [4:0] pipe_cnt;
-    wire [65:0] acc_result = ({2'b0, mult_p[0]} + {2'b0, mult_p[1]}) +
-                             ({2'b0, mult_p[2]} + {2'b0, mult_p[3]});
-
-    // Leave eight entries for results already in the multiplier pipeline.
-    // Under the normal DMA cadence the FIFO stays well below this threshold.
-    wire dma_crc_issue_ready = !dma_busy || (crc_fifo_count < 6'd24);
-
-    // DMA CRC takes priority over the software-driven AXI read stream.
-    always @(posedge clk) begin
-        if (!resetn) begin
-            crc_reg         <= 32'hFFFFFFFF;
-            crc_fifo_wr_ptr <= 5'd0;
-            crc_fifo_rd_ptr <= 5'd0;
-            crc_fifo_count  <= 6'd0;
-            crc_c66_busy    <= 1'b0;
-            crc_c66_data    <= 66'd0;
-            crc_c66_half    <= 32'd0;
-        end else if (crc_clear_pulse) begin
-            crc_reg         <= 32'hFFFFFFFF;
-            crc_fifo_wr_ptr <= 5'd0;
-            crc_fifo_rd_ptr <= 5'd0;
-            crc_fifo_count  <= 6'd0;
-            crc_c66_busy    <= 1'b0;
-            crc_c66_data    <= 66'd0;
-            crc_c66_half    <= 32'd0;
-        end else begin
-            if (crc_fifo_push) begin
-                crc_fifo[crc_fifo_wr_ptr] <= acc_result;
-                crc_fifo_wr_ptr <= crc_fifo_wr_ptr + 5'd1;
-            end
-
-            case ({crc_fifo_push, crc_fifo_pop})
-                2'b10: crc_fifo_count <= crc_fifo_count + 6'd1;
-                2'b01: crc_fifo_count <= crc_fifo_count - 6'd1;
-                default: crc_fifo_count <= crc_fifo_count;
-            endcase
-
-            if (crc_c66_busy) begin
-                crc_reg <= crc_c66_final_next;
-                crc_c66_busy <= 1'b0;
-            end else if (crc_fifo_pop) begin
-                crc_c66_data <= crc_fifo[crc_fifo_rd_ptr];
-                crc_c66_half <= crc_c66_first_next;
-                crc_c66_busy <= 1'b1;
-                crc_fifo_rd_ptr <= crc_fifo_rd_ptr + 5'd1;
-            end else if (crc_stream_valid) begin
-                crc_reg <= crc_next;
-            end
-        end
-    end
 
     task start_calc_bank;
         input bank;
@@ -856,7 +810,7 @@ module matmul_axi_slave (
         end else begin
             calc_done_pulse <= 1'b0;
 
-            // Pipeline shift (4 cycles total)
+            // Pipeline shift: issue_idx → d1 → d2 → d3 → d4 (4 cycles total)
             acc_idx_d4  <= acc_idx_d3;
             acc_idx_d3  <= acc_idx_d2;
             acc_idx_d2  <= acc_idx_d1;
@@ -883,10 +837,8 @@ module matmul_axi_slave (
                 end
 
                 C_RUN: begin
-                    // Issue one C element per cycle unless the CRC FIFO is
+                    // Issue 4 new operands every cycle (pipe_cnt 0-15, 16 rounds)
-                    // approaching capacity.
                     if (pipe_cnt < 5'd16) begin
-                        if (dma_crc_issue_ready) begin
                         acc_idx_d1   <= pipe_cnt[3:0];
                         acc_bank_d1  <= cur_bank;
                         acc_valid_d1 <= 1'b1;
@@ -901,36 +853,24 @@ module matmul_axi_slave (
                             m_in_a[2] <= b1_a[{pipe_cnt[3:2], 2'd2}]; m_in_b[2] <= b1_b[{2'd2, pipe_cnt[1:0]}];
                             m_in_a[3] <= b1_a[{pipe_cnt[3:2], 2'd3}]; m_in_b[3] <= b1_b[{2'd3, pipe_cnt[1:0]}];
                         end
-                            pipe_cnt <= pipe_cnt + 5'd1;
-                        end
                     end
 
                     // Accumulate results from 4 cycles ago.
                     if (acc_valid_d4) begin
                         if (acc_bank_d4 == 1'b0)
-                            b0_c[acc_idx_d4] <= acc_result;
+                            b0_c[acc_idx_d4] <= ({2'b0, mult_p[0]} + {2'b0, mult_p[1]}) +
+                                                ({2'b0, mult_p[2]} + {2'b0, mult_p[3]});
                         else
-                            b1_c[acc_idx_d4] <= acc_result;
+                            b1_c[acc_idx_d4] <= ({2'b0, mult_p[0]} + {2'b0, mult_p[1]}) +
+                                                ({2'b0, mult_p[2]} + {2'b0, mult_p[3]});
                     end
 
                     if (acc_valid_d4 && acc_idx_d4 == 4'd15) begin
                         calc_done_pulse <= 1'b1;
                     end
 
-                    if (pipe_cnt >= 5'd16) begin
+                    if (pipe_cnt < 5'd16) begin
-                        if (dma_busy) begin
-                            if (acc_valid_d4 && acc_idx_d4 == 4'd15) begin
-                                if (cur_bank == 1'b0 && req_1)
-                                    start_calc_bank(1'b1);
-                                else if (cur_bank == 1'b1 && req_0)
-                                    start_calc_bank(1'b0);
-                                else begin
-                                    busy  <= 1'b0;
-                                    state <= C_IDLE;
-                                end
-                            end else begin
                         pipe_cnt <= pipe_cnt + 5'd1;
-                            end
                     end else if (cur_bank == 1'b0 && req_1) begin
                         start_calc_bank(1'b1);
                     end else if (cur_bank == 1'b1 && req_0) begin
@@ -942,20 +882,20 @@ module matmul_axi_slave (
                         pipe_cnt <= pipe_cnt + 5'd1;
                     end
                 end
-                end
             endcase
         end
     end
 
     // ==========================================
-    // Autonomous AXI master loader with inline result CRC.
+    // Autonomous AXI master loader/storer
     // ==========================================
+    assign crc_dma_valid = (dma_wr_state == DMA_WR_DATA) && m_axi_wvalid && m_axi_wready;
+    assign crc_dma_data  = m_axi_wdata;
+
     always @(posedge clk) begin
         if (!resetn) begin
             dma_busy           <= 1'b0;
             dma_done           <= 1'b0;
-            dma_compute_complete <= 1'b0;
-            dma_cycle_count    <= 32'd0;
             load_group         <= 32'd0;
             store_group        <= 32'd0;
             dma_next_rd_addr   <= 32'd0;
@@ -969,7 +909,7 @@ module matmul_axi_slave (
             dma_load_bank      <= 1'b0;
             dma_store_bank     <= 1'b0;
             dma_rd_state       <= DMA_RD_IDLE;
-            dma_rd_count       <= 8'd0;
+            dma_rd_count       <= 5'd0;
             dma_rd_bank        <= 1'b0;
             dma_wr_state       <= DMA_WR_IDLE;
             dma_wr_count       <= 6'd0;
@@ -1011,8 +951,6 @@ module matmul_axi_slave (
             if (dma_start_pulse && !dma_busy && dma_groups != 32'd0) begin
                 dma_busy           <= 1'b1;
                 dma_done           <= 1'b0;
-                dma_compute_complete <= 1'b0;
-                dma_cycle_count    <= 32'd0;
                 load_group         <= 32'd0;
                 store_group        <= 32'd0;
                 dma_next_rd_addr   <= dma_src_base;
@@ -1027,21 +965,19 @@ module matmul_axi_slave (
                 dma_store_bank     <= 1'b0;
                 dma_rd_state       <= DMA_RD_IDLE;
                 dma_wr_state       <= DMA_WR_IDLE;
-            end else if (dma_busy) begin
-                dma_cycle_count <= dma_cycle_count + 32'd1;
             end
 
             if (dma_busy) begin
                 if (state == C_RUN)
                     bank_compute_pending[cur_bank] <= 1'b0;
 
-                if (bank_loaded[0] && !req_0 &&
+                if (bank_loaded[0] && !req_0 && !bank_store_pending[0] && !bank_storing[0] &&
                     !bank_compute_pending[0] && !(state == C_RUN && cur_bank == 1'b0)) begin
                     dma_req0_pulse <= 1'b1;
                     bank_loaded[0] <= 1'b0;
                     bank_compute_pending[0] <= 1'b1;
                     dma_compute_bank <= 1'b0;
-                end else if (bank_loaded[1] && !req_1 &&
+                end else if (bank_loaded[1] && !req_1 && !bank_store_pending[1] && !bank_storing[1] &&
                              !bank_compute_pending[1] && !(state == C_RUN && cur_bank == 1'b1)) begin
                     dma_req1_pulse <= 1'b1;
                     bank_loaded[1] <= 1'b0;
@@ -1051,28 +987,21 @@ module matmul_axi_slave (
             end
 
             if (calc_done_pulse && dma_busy) begin
-                store_group <= store_group + 32'd1;
+                bank_store_pending[cur_bank] <= 1'b1;
-                completed_groups <= completed_groups + 32'd1;
-                if ((completed_groups + 32'd1) >= dma_groups)
-                    dma_compute_complete <= 1'b1;
-            end
-
-            if (dma_compute_complete && crc_fifo_empty && !crc_c66_busy) begin
-                dma_busy <= 1'b0;
-                dma_done <= 1'b1;
-                dma_compute_complete <= 1'b0;
             end
 
             case (dma_rd_state)
                 DMA_RD_IDLE: begin
                     m_axi_arvalid <= 1'b0;
                     m_axi_rready  <= 1'b0;
-                    if (dma_busy && load_group < dma_groups && dma_load_bank_free) begin
+                    if (dma_busy && load_group < dma_groups && !bank_loaded[dma_load_bank] &&
+                        !bank_compute_pending[dma_load_bank] &&
+                        !((dma_load_bank == 1'b0) ? req_0 : req_1) &&
+                        !(state == C_RUN && cur_bank == dma_load_bank)) begin
                         dma_rd_bank   <= dma_load_bank;
-                        dma_rd_count  <= 8'd0;
+                        dma_rd_count  <= 5'd0;
                         m_axi_araddr  <= dma_next_rd_addr;
-                        m_axi_arlen   <= ((dma_src_stride == 32'd128) &&
+                        m_axi_arlen   <= 8'd31;
-                                          ((dma_groups - load_group) >= 32'd8)) ? 8'd255 : 8'd31;
                         m_axi_arsize  <= 3'd2;
                         m_axi_arburst <= 2'b01;
                         m_axi_arvalid <= 1'b1;
@@ -1087,54 +1016,16 @@ module matmul_axi_slave (
                     end
                 end
                 DMA_RD_DATA: begin
-                    if (!m_axi_rready && dma_rd_fill_bank_free)
-                        m_axi_rready <= 1'b1;
-
                     if (m_axi_rvalid && m_axi_rready) begin
-                        // Present the next request one beat early so the SRAM
+                        if (m_axi_rlast) begin
-                        // slave can accept it together with the current RLAST.
+                            m_axi_rready <= 1'b0;
-                        if ((dma_rd_count == 8'd254) && dma_busy &&
+                            bank_loaded[dma_rd_bank] <= 1'b1;
-                            ((load_group + 32'd1) < dma_groups) && dma_rd_next_bank_free) begin
-                            m_axi_araddr  <= dma_next_rd_addr + dma_src_stride;
-                            m_axi_arlen   <= ((dma_src_stride == 32'd128) &&
-                                              ((dma_groups - (load_group + 32'd1)) >= 32'd8)) ?
-                                             8'd255 : 8'd31;
-                            m_axi_arvalid <= 1'b1;
-                        end
-
-                        if (dma_rd_group_word_idx == 5'd31) begin
-                            bank_loaded[dma_rd_fill_bank] <= 1'b1;
                             load_group <= load_group + 32'd1;
                             dma_next_rd_addr <= dma_next_rd_addr + dma_src_stride;
                             dma_load_bank <= ~dma_load_bank;
-                        end
-
-                        if (m_axi_rlast) begin
-                            m_axi_rready <= 1'b0;
-                            if (dma_busy && ((load_group + 32'd1) < dma_groups) &&
-                                dma_rd_next_bank_free) begin
-                                dma_rd_bank   <= ~dma_rd_fill_bank;
-                                dma_rd_count  <= 8'd0;
-                                if (m_axi_arvalid && m_axi_arready) begin
-                                    m_axi_arvalid <= 1'b0;
-                                    m_axi_rready  <= 1'b1;
-                                    dma_rd_state  <= DMA_RD_DATA;
-                                end else begin
-                                    m_axi_araddr  <= dma_next_rd_addr + dma_src_stride;
-                                    m_axi_arlen   <= ((dma_src_stride == 32'd128) &&
-                                                      ((dma_groups - (load_group + 32'd1)) >= 32'd8)) ?
-                                                     8'd255 : 8'd31;
-                                    m_axi_arvalid <= 1'b1;
-                                    dma_rd_state  <= DMA_RD_ADDR;
-                                end
-                            end else begin
-                                m_axi_arvalid <= 1'b0;
                             dma_rd_state <= DMA_RD_IDLE;
-                            end
                         end else begin
-                            dma_rd_count <= dma_rd_count + 8'd1;
+                            dma_rd_count <= dma_rd_count + 5'd1;
-                            if ((dma_rd_group_word_idx == 5'd31) && !dma_rd_next_bank_free)
-                                m_axi_rready <= 1'b0;
                         end
                     end
                 end
@@ -1146,32 +1037,56 @@ module matmul_axi_slave (
                     m_axi_awvalid <= 1'b0;
                     m_axi_wvalid  <= 1'b0;
                     m_axi_bready  <= 1'b0;
+                    if (dma_busy && bank_store_pending[dma_store_bank]) begin
+                        dma_wr_count <= 6'd0;
+                        bank_store_pending[dma_store_bank] <= 1'b0;
+                        bank_storing[dma_store_bank] <= 1'b1;
+                        m_axi_awaddr  <= dma_next_wr_addr;
+                        m_axi_awlen   <= 8'd47;
+                        m_axi_awsize  <= 3'd2;
+                        m_axi_awburst <= 2'b01;
+                        m_axi_awvalid <= 1'b1;
+                        dma_wr_state  <= DMA_WR_ADDR;
+                    end
                 end
                 DMA_WR_ADDR: begin
+                    if (m_axi_awvalid && m_axi_awready) begin
                         m_axi_awvalid <= 1'b0;
+                        m_axi_wvalid  <= 1'b1;
+                        m_axi_wdata   <= dma_wr_word;
+                        m_axi_wlast   <= 1'b0;
                         dma_wr_state  <= DMA_WR_DATA;
                     end
+                end
                 DMA_WR_DATA: begin
+                    if (m_axi_wvalid && m_axi_wready) begin
+                        if (dma_wr_count == 6'd47) begin
                             m_axi_wvalid <= 1'b0;
                             m_axi_wlast  <= 1'b0;
-                    if (dma_wr_count == 6'd47) begin
+                            m_axi_bready <= 1'b1;
+                            dma_wr_state <= DMA_WR_RESP;
+                        end else begin
+                            dma_wr_count <= dma_wr_count + 6'd1;
+                            m_axi_wdata  <= dma_wr_word_next;
+                            m_axi_wlast  <= (dma_wr_count == 6'd46);
+                        end
+                    end
+                end
+                DMA_WR_RESP: begin
+                    if (m_axi_bvalid && m_axi_bready) begin
+                        m_axi_bready <= 1'b0;
                         bank_storing[dma_store_bank] <= 1'b0;
                         dma_store_bank <= ~dma_store_bank;
                         store_group <= store_group + 32'd1;
+                        dma_next_wr_addr <= dma_next_wr_addr + dma_dst_stride;
                         completed_groups <= completed_groups + 32'd1;
                         if ((completed_groups + 32'd1) >= dma_groups) begin
                             dma_busy <= 1'b0;
                             dma_done <= 1'b1;
                         end
                         dma_wr_state <= DMA_WR_IDLE;
-                    end else begin
-                        dma_wr_count <= dma_wr_count + 6'd1;
                     end
                 end
-                DMA_WR_RESP: begin
-                    m_axi_bready <= 1'b0;
-                    dma_wr_state <= DMA_WR_IDLE;
-                end
                 default: dma_wr_state <= DMA_WR_IDLE;
             endcase
         end

rtl/ip/ram_wrap/axi_wrap_ram_sp_external.v

@@ -75,21 +75,21 @@ module axi_wrap_ram_sp_external (
     output        axi_bvalid ,
     input         axi_bready ,
 
-    // BaseRAM
+    //BaseRAM信号
-    inout  [31:0] base_ram_data,
+    inout  [31:0] base_ram_data,  //BaseRAM数据，低8位与CPLD串口控制器共享
-    output [19:0] base_ram_addr,
+    output [19:0] base_ram_addr, //BaseRAM地址
-    output [ 3:0] base_ram_be_n,
+    output [ 3:0] base_ram_be_n,  //BaseRAM字节使能，低有效。如果不使用字节使能，请保持为0
-    output        base_ram_ce_n,
+    output  base_ram_ce_n,       //BaseRAM片选，低有效
-    output        base_ram_oe_n,
+    output  base_ram_oe_n,       //BaseRAM读使能，低有效
-    output        base_ram_we_n,
+    output  base_ram_we_n,       //BaseRAM写使能，低有效
 
-    // ExtRAM
+    //ExtRAM信号
-    inout  [31:0] ext_ram_data,
+    inout  [31:0] ext_ram_data,  //ExtRAM数据
-    output [19:0] ext_ram_addr,
+    output [19:0] ext_ram_addr, //ExtRAM地址
-    output [ 3:0] ext_ram_be_n,
+    output [ 3:0] ext_ram_be_n,  //ExtRAM字节使能，低有效。如果不使用字节使能，请保持为0
-    output        ext_ram_ce_n,
+    output  ext_ram_ce_n,       //ExtRAM片选，低有效
-    output        ext_ram_oe_n,
+    output  ext_ram_oe_n,       //ExtRAM读使能，低有效
-    output        ext_ram_we_n
+    output  ext_ram_we_n       //ExtRAM写使能，低有效
 );
 
 
@@ -139,7 +139,6 @@ wire        ram_bready ;
 wire  [31:0]    soc_sram_addr;
 wire            soc_sram_cs;
 wire            soc_sram_we;
-wire            soc_sram_bank;
 wire  [3:0]     soc_sram_be;
 wire  [31:0]    soc_sram_wdata;
 wire  [31:0]    soc_sram_rdata;
@@ -189,8 +188,7 @@ assign ram_bready = axi_bready ;
 axi2sram_sp_external #(
     .AXI_ID_WIDTH   ( 5  ),
     .AXI_ADDR_WIDTH ( 32 ),
-    .AXI_DATA_WIDTH ( 32 ),
+    .AXI_DATA_WIDTH ( 32 ))
-    .FAST_READ_BURST( 1  ))
  u_axi_sram_sp (
     .clk                     ( aclk         ),
     .resetn                  ( aresetn      ),
@@ -234,13 +232,12 @@ axi2sram_sp_external #(
     .req_o                   ( soc_sram_cs       ),
     .we_o                    ( soc_sram_we       ),
     .addr_o                  ( soc_sram_addr     ),
-    .bank_o                  ( soc_sram_bank     ),
     .be_o                    ( soc_sram_be       ),
     .data_o                  ( soc_sram_wdata    ),
     .data_i                  ( soc_sram_rdata    )
 );
 
-wire choose_sram = soc_sram_bank;//1:ExtRAM 0:BaseRAM
+wire choose_sram = soc_sram_addr[22];//1:ExtRAM 0:BaseRAM
 wire [3:0] be_out = soc_sram_we ? soc_sram_be : 4'b1111;
 
 assign base_ram_addr = soc_sram_addr[21:2];

rtl/soc_top.v

@@ -44,38 +44,39 @@ module soc_top #(parameter SIMULATION=1'b0)
     input           clk,
     input           reset,
 
-    output [2:0]    video_red,
+    //图像输出信号
-    output [2:0]    video_green,
+    output [2:0]    video_red,          //红色像素，3位
-    output [1:0]    video_blue,
+    output [2:0]    video_green,        //绿色像素，3位
-    output          video_hsync,
+    output [1:0]    video_blue,         //蓝色像素，2位
-    output          video_vsync,
+    output          video_hsync,        //行同步（水平同步）信号
-    output          video_clk,
+    output          video_vsync,        //场同步（垂直同步）信号
-    output          video_de,
+    output          video_clk,          //像素时钟输出
+    output          video_de,           //行数据有效信号，用于区分消隐区
 
-    input  [3:0]    touch_btn,
+    input  [3:0]    touch_btn,          //BTN1~BTN4，按钮开关，按下时为1
-    input  [31:0]   dip_sw,
+    input  [31:0]   dip_sw,             //32位拨码开关，拨到“ON”时为1
-    output [15:0]   leds,
+    output [15:0]   leds,               //16位LED，输出时1点亮
-    output [7:0]    dpy0,
+    output [7:0]    dpy0,               //数码管低位信号，包括小数点，输出1点亮
-    output [7:0]    dpy1,
+    output [7:0]    dpy1,               //数码管高位信号，包括小数点，输出1点亮
 
-    // BaseRAM signals
+    //BaseRAM信号
-    inout  [31:0]   base_ram_data,
+    inout  [31:0]   base_ram_data,      //BaseRAM数据，低8位与CPLD串口控制器共享
-    output [19:0]   base_ram_addr,
+    output [19:0]   base_ram_addr,      //BaseRAM地址
-    output [ 3:0]   base_ram_be_n,
+    output [ 3:0]   base_ram_be_n,      //BaseRAM字节使能，低有效。如果不使用字节使能，请保持为0
-    output          base_ram_ce_n,
+    output          base_ram_ce_n,      //BaseRAM片选，低有效
-    output          base_ram_oe_n,
+    output          base_ram_oe_n,      //BaseRAM读使能，低有效
-    output          base_ram_we_n,
+    output          base_ram_we_n,      //BaseRAM写使能，低有效
-    // ExtRAM signals
+    //ExtRAM信号
-    inout  [31:0]   ext_ram_data,
+    inout  [31:0]   ext_ram_data,       //ExtRAM数据
-    output [19:0]   ext_ram_addr,
+    output [19:0]   ext_ram_addr,       //ExtRAM地址
-    output [ 3:0]   ext_ram_be_n,
+    output [ 3:0]   ext_ram_be_n,       //ExtRAM字节使能，低有效。如果不使用字节使能，请保持为0
-    output          ext_ram_ce_n,
+    output          ext_ram_ce_n,       //ExtRAM片选，低有效
-    output          ext_ram_oe_n,
+    output          ext_ram_oe_n,       //ExtRAM读使能，低有效
-    output          ext_ram_we_n,
+    output          ext_ram_we_n,       //ExtRAM写使能，低有效
 
     //------uart-------
-    inout           UART_RX,
+    inout           UART_RX,            //串口RX接收
-    inout           UART_TX
+    inout           UART_TX             //串口TX发送
 );
 
 wire cpu_clk;
@@ -359,6 +360,44 @@ wire [3 :0] dma_m_bid    ;
 wire [1 :0] dma_m_bresp  ;
 wire        dma_m_bvalid ;
 wire        dma_m_bready ;
+
+wire [3 :0] cdma_m_arid   ;
+wire [31:0] cdma_m_araddr ;
+wire [7 :0] cdma_m_arlen  ;
+wire [2 :0] cdma_m_arsize ;
+wire [1 :0] cdma_m_arburst;
+wire        cdma_m_arlock ;
+wire [3 :0] cdma_m_arcache;
+wire [2 :0] cdma_m_arprot ;
+wire        cdma_m_arvalid;
+wire        cdma_m_arready = 1'b0;
+wire [3 :0] cdma_m_rid    = 4'd0;
+wire [31:0] cdma_m_rdata  = 32'd0;
+wire [1 :0] cdma_m_rresp  = 2'd0;
+wire        cdma_m_rlast  = 1'b0;
+wire        cdma_m_rvalid = 1'b0;
+wire        cdma_m_rready ;
+wire [3 :0] cdma_m_awid   ;
+wire [31:0] cdma_m_awaddr ;
+wire [7 :0] cdma_m_awlen  ;
+wire [2 :0] cdma_m_awsize ;
+wire [1 :0] cdma_m_awburst;
+wire        cdma_m_awlock ;
+wire [3 :0] cdma_m_awcache;
+wire [2 :0] cdma_m_awprot ;
+wire        cdma_m_awvalid;
+wire        cdma_m_awready = 1'b0;
+wire [3 :0] cdma_m_wid    ;
+wire [31:0] cdma_m_wdata  ;
+wire [3 :0] cdma_m_wstrb  ;
+wire        cdma_m_wlast  ;
+wire        cdma_m_wvalid ;
+wire        cdma_m_wready = 1'b0;
+wire [3 :0] cdma_m_bid    = 4'd0;
+wire [1 :0] cdma_m_bresp  = 2'd0;
+wire        cdma_m_bvalid = 1'b0;
+wire        cdma_m_bready ;
+
 wire [4 :0] dma_s_arid   ;
 wire [31:0] dma_s_araddr ;
 wire [7 :0] dma_s_arlen  ;
@@ -394,7 +433,7 @@ wire [4 :0] dma_s_bid    ;
 wire [1 :0] dma_s_bresp  ;
 wire        dma_s_bvalid ;
 wire        dma_s_bready ;
-wire        dma_finish = 1'b0;
+wire        dma_finish   ;
 
 // reserved
 wire [4 :0] axiOut_1_arid   ;
@@ -602,6 +641,7 @@ wire        axiOut_7_bready ;
 
 wire confreg_int;
 wire matmul_intr;
+wire mlkem_intr;
 
 AxiCrossbar_2x8  u_AxiCrossbar_2x8 (
     .clk                     ( sys_clk             ),
@@ -1033,8 +1073,8 @@ AxiCrossbar_2x8  u_AxiCrossbar_2x8 (
 
 
 core_top u_cpu(
-    .intrpt          ({4'd0, 1'b0, matmul_intr, dma_finish, confreg_int}), //high active
+    .intrpt          ({4'd0, mlkem_intr, matmul_intr, dma_finish, confreg_int}), //high active
-    // intrpt[2]=matmul, intrpt[1]=dma, intrpt[0]=confreg
+    // intrpt[3]=mlkem, intrpt[2]=matmul, intrpt[1]=dma, intrpt[0]=confreg
 
     .aclk            (cpu_clk ),
     .aresetn         (cpu_resetn ), //low active
@@ -1309,12 +1349,13 @@ axi_uart_controller u_axi_uart_controller
 confreg #(
     .SIMULATION(SIMULATION)
 ) u_confreg (
-    .aclk       (sys_clk),
+    // 时钟与复位
-    .aresetn    (sys_resetn),
+    .aclk       (sys_clk),    // AXI总线时钟
-    .cpu_clk    (cpu_clk),
+    .aresetn    (sys_resetn), // AXI总线复位
-    .cpu_resetn (cpu_resetn),
+    .cpu_clk    (cpu_clk),    // CPU时钟
+    .cpu_resetn (cpu_resetn), // CPU复位
 
-    // AXI Write Address
+    // AXI写地址通道
     .s_awid     (confreg_awid),
     .s_awaddr   (confreg_awaddr),
     .s_awlen    (confreg_awlen),
@@ -1326,7 +1367,7 @@ confreg #(
     .s_awvalid  (confreg_awvalid),
     .s_awready  (confreg_awready),
 
-    // AXI Write Data
+    // AXI写数据通道
     .s_wid      (confreg_wid),
     .s_wdata    (confreg_wdata),
     .s_wstrb    (confreg_wstrb),
@@ -1334,13 +1375,13 @@ confreg #(
     .s_wvalid   (confreg_wvalid),
     .s_wready   (confreg_wready),
 
-    // AXI Write Response
+    // AXI写响应通道
     .s_bid      (confreg_bid),
     .s_bresp    (confreg_bresp),
     .s_bvalid   (confreg_bvalid),
     .s_bready   (confreg_bready),
 
-    // AXI Read Address
+    // AXI读地址通道
     .s_arid     (confreg_arid),
     .s_araddr   (confreg_araddr),
     .s_arlen    (confreg_arlen),
@@ -1352,7 +1393,7 @@ confreg #(
     .s_arvalid  (confreg_arvalid),
     .s_arready  (confreg_arready),
 
-    // AXI Read Data
+    // AXI读数据通道
     .s_rid      (confreg_rid),
     .s_rdata    (confreg_rdata),
     .s_rresp    (confreg_rresp),
@@ -1360,17 +1401,19 @@ confreg #(
     .s_rvalid   (confreg_rvalid),
     .s_rready   (confreg_rready),
 
-    // IO
+    // IO外设接口 (与soc_top顶层引脚相连)
-    .led        (leds),
+    .led        (leds),             // 顶层引脚名称为
-    .dpy0       (dpy0),
+    .dpy0       (dpy0),             // 低位数码管
-    .dpy1       (dpy1),
+    .dpy1       (dpy1),             // 高位数码管
-    .switch     (dip_sw),
+    .switch     (dip_sw),           // 拨码开关，顶层名称为 dip_sw
-    .touch_btn  (touch_btn),
+    .touch_btn  (touch_btn),        // 触摸按钮
 
-    .uart0_int  (uart0_int),
+    // 中断与状态信号
-    .fft_finish (fft_finish),
+    .uart0_int  (uart0_int),        // UART0中断
-    .confreg_int(confreg_int)
+    .fft_finish (fft_finish),       // FFT完成信号
+    .confreg_int(confreg_int)       // 输出中断给CPU
 );
+
 axi_dvi u_axi_dvi (
     .s_awvalid ( dvi_awvalid ),
     .s_awaddr ( dvi_awaddr ),
@@ -1419,20 +1462,69 @@ axi_dvi u_axi_dvi (
     .video_blue ( video_blue )
 );
 
-// slave 6 reserved (mlkem removed)
+// slave 6 - ML-KEM accelerator (0x1f400000, virtual 0xbf400000)
-assign fft_arready = 1'b1;
+// Bus runs at sys_clk; the core is internally clocked at sys_clk/2 (25 MHz).
-assign fft_rid    = 5'b0;
+mlkem_axi_slave u_mlkem_axi_slave (
-assign fft_rdata  = 32'b0;
+    .clk            (sys_clk),
-assign fft_rresp  = 2'b0;
+    .resetn         (sys_resetn),
-assign fft_rlast  = 1'b0;
+
-assign fft_rvalid = 1'b0;
+    .s_axi_awid     (fft_awid   ),
-assign fft_awready = 1'b1;
+    .s_axi_awaddr   (fft_awaddr ),
-assign fft_wready = 1'b1;
+    .s_axi_awlen    (fft_awlen  ),
-assign fft_bid    = 5'b0;
+    .s_axi_awsize   (fft_awsize ),
-assign fft_bresp  = 2'b0;
+    .s_axi_awburst  (fft_awburst),
-assign fft_bvalid = 1'b0;
+    .s_axi_awlock   (fft_awlock ),
+    .s_axi_awcache  (fft_awcache),
+    .s_axi_awprot   (fft_awprot ),
+    .s_axi_awvalid  (fft_awvalid),
+    .s_axi_awready  (fft_awready),
+
+    .s_axi_wid      (fft_wid   ),
+    .s_axi_wdata    (fft_wdata ),
+    .s_axi_wstrb    (fft_wstrb ),
+    .s_axi_wlast    (fft_wlast ),
+    .s_axi_wvalid   (fft_wvalid),
+    .s_axi_wready   (fft_wready),
+
+    .s_axi_bid      (fft_bid   ),
+    .s_axi_bresp    (fft_bresp ),
+    .s_axi_bvalid   (fft_bvalid),
+    .s_axi_bready   (fft_bready),
+
+    .s_axi_arid     (fft_arid   ),
+    .s_axi_araddr   (fft_araddr ),
+    .s_axi_arlen    (fft_arlen  ),
+    .s_axi_arsize   (fft_arsize ),
+    .s_axi_arburst  (fft_arburst),
+    .s_axi_arlock   (fft_arlock ),
+    .s_axi_arcache  (fft_arcache),
+    .s_axi_arprot   (fft_arprot ),
+    .s_axi_arvalid  (fft_arvalid),
+    .s_axi_arready  (fft_arready),
+
+    .s_axi_rid      (fft_rid   ),
+    .s_axi_rdata    (fft_rdata ),
+    .s_axi_rresp    (fft_rresp ),
+    .s_axi_rlast    (fft_rlast ),
+    .s_axi_rvalid   (fft_rvalid),
+    .s_axi_rready   (fft_rready),
+    .intr           (mlkem_intr)
+);
 
 //slave 7 - matmul accelerator
+wire        dma_ep_wr_valid;
+wire [3:0]  dma_ep_wr_id;
+wire [3:0]  dma_ep_wr_bank;
+wire [7:0]  dma_ep_wr_word_addr;
+wire [31:0] dma_ep_wr_data;
+wire        dma_ep_wr_ready;
+wire        dma_ep_rd_valid;
+wire [3:0]  dma_ep_rd_id;
+wire [3:0]  dma_ep_rd_bank;
+wire [7:0]  dma_ep_rd_word_addr;
+wire [31:0] dma_ep_rd_data;
+wire        dma_ep_rd_ready;
+
 matmul_axi_slave u_matmul_axi_slave (
     .clk            (sys_clk),
     .resetn         (sys_resetn),
@@ -1478,6 +1570,19 @@ matmul_axi_slave u_matmul_axi_slave (
     .s_axi_rvalid   (axiOut_7_rvalid),
     .s_axi_rready   (axiOut_7_rready),
 
+    .ep_wr_valid    (dma_ep_wr_valid),
+    .ep_wr_id       (dma_ep_wr_id),
+    .ep_wr_bank     (dma_ep_wr_bank),
+    .ep_wr_word_addr(dma_ep_wr_word_addr),
+    .ep_wr_data     (dma_ep_wr_data),
+    .ep_wr_ready    (dma_ep_wr_ready),
+    .ep_rd_valid    (dma_ep_rd_valid),
+    .ep_rd_id       (dma_ep_rd_id),
+    .ep_rd_bank     (dma_ep_rd_bank),
+    .ep_rd_word_addr(dma_ep_rd_word_addr),
+    .ep_rd_data     (dma_ep_rd_data),
+    .ep_rd_ready    (dma_ep_rd_ready),
+
     .m_axi_awid     (dma_m_awid),
     .m_axi_awaddr   (dma_m_awaddr),
     .m_axi_awlen    (dma_m_awlen),
@@ -1517,16 +1622,186 @@ matmul_axi_slave u_matmul_axi_slave (
     .intr           (matmul_intr)
 );
 
-//slave 5 reserved (cdma adapter removed)
+wire [31:0] cdma_s_awaddr;
-assign dma_s_arready = 1'b1;
+wire [2:0]  cdma_s_awprot;
-assign dma_s_rid    = 5'b0;
+wire        cdma_s_awvalid;
-assign dma_s_rdata  = 32'b0;
+wire        cdma_s_awready;
-assign dma_s_rresp  = 2'b0;
+wire [31:0] cdma_s_wdata;
-assign dma_s_rlast  = 1'b0;
+wire [3:0]  cdma_s_wstrb;
-assign dma_s_rvalid = 1'b0;
+wire        cdma_s_wvalid;
-assign dma_s_awready = 1'b1;
+wire        cdma_s_wready;
-assign dma_s_wready = 1'b1;
+wire [1:0]  cdma_s_bresp;
-assign dma_s_bid    = 5'b0;
+wire        cdma_s_bvalid;
-assign dma_s_bresp  = 2'b0;
+wire        cdma_s_bready;
-assign dma_s_bvalid = 1'b0;
+wire [31:0] cdma_s_araddr;
+wire [2:0]  cdma_s_arprot;
+wire        cdma_s_arvalid;
+wire        cdma_s_arready;
+wire [31:0] cdma_s_rdata;
+wire [1:0]  cdma_s_rresp;
+wire        cdma_s_rvalid;
+wire        cdma_s_rready;
+
+axi_axil_adapter #(
+    .ADDR_WIDTH(32),
+    .AXI_DATA_WIDTH(32),
+    .AXI_ID_WIDTH(5),
+    .AXIL_DATA_WIDTH(32),
+    .CONVERT_BURST(1),
+    .CONVERT_NARROW_BURST(0)
+)
+u_axi_axil_adapter(
+    .clk(sys_clk),
+    .rst(~sys_resetn),
+    .s_axi_awid(dma_s_awid),
+    .s_axi_awaddr(dma_s_awaddr),
+    .s_axi_awlen(dma_s_awlen),
+    .s_axi_awsize(dma_s_awsize),
+    .s_axi_awburst(dma_s_awburst),
+    .s_axi_awlock(dma_s_awlock),
+    .s_axi_awcache(dma_s_awcache),
+    .s_axi_awprot(dma_s_awprot),
+    .s_axi_awvalid(dma_s_awvalid),
+    .s_axi_awready(dma_s_awready),
+    .s_axi_wdata(dma_s_wdata),
+    .s_axi_wstrb(dma_s_wstrb),
+    .s_axi_wlast(dma_s_wlast),
+    .s_axi_wvalid(dma_s_wvalid),
+    .s_axi_wready(dma_s_wready),
+    .s_axi_bid(dma_s_bid),
+    .s_axi_bresp(dma_s_bresp),
+    .s_axi_bvalid(dma_s_bvalid),
+    .s_axi_bready(dma_s_bready),
+    .s_axi_arid(dma_s_arid),
+    .s_axi_araddr(dma_s_araddr),
+    .s_axi_arlen(dma_s_arlen),
+    .s_axi_arsize(dma_s_arsize),
+    .s_axi_arburst(dma_s_arburst),
+    .s_axi_arlock(dma_s_arlock),
+    .s_axi_arcache(dma_s_arcache),
+    .s_axi_arprot(dma_s_arprot),
+    .s_axi_arvalid(dma_s_arvalid),
+    .s_axi_arready(dma_s_arready),
+    .s_axi_rid(dma_s_rid),
+    .s_axi_rdata(dma_s_rdata),
+    .s_axi_rresp(dma_s_rresp),
+    .s_axi_rlast(dma_s_rlast),
+    .s_axi_rvalid(dma_s_rvalid),
+    .s_axi_rready(dma_s_rready),
+
+    .m_axil_awaddr(cdma_s_awaddr),
+    .m_axil_awprot(cdma_s_awprot),
+    .m_axil_awvalid(cdma_s_awvalid),
+    .m_axil_awready(cdma_s_awready),
+    .m_axil_wdata(cdma_s_wdata),
+    .m_axil_wstrb(cdma_s_wstrb),
+    .m_axil_wvalid(cdma_s_wvalid),
+    .m_axil_wready(cdma_s_wready),
+    .m_axil_bresp(cdma_s_bresp),
+    .m_axil_bvalid(cdma_s_bvalid),
+    .m_axil_bready(cdma_s_bready),
+    .m_axil_araddr(cdma_s_araddr),
+    .m_axil_arprot(cdma_s_arprot),
+    .m_axil_arvalid(cdma_s_arvalid),
+    .m_axil_arready(cdma_s_arready),
+    .m_axil_rdata(cdma_s_rdata),
+    .m_axil_rresp(cdma_s_rresp),
+    .m_axil_rvalid(cdma_s_rvalid),
+    .m_axil_rready(cdma_s_rready)
+);
+
+// 8 通道 DMA
+snix_axil_cdma_mux #(
+    .ADDR_WIDTH      (32),
+    .DATA_WIDTH      (32), // 适配你的 32-bit 总线
+    .AXIL_ADDR_WIDTH (32),
+    .AXIL_DATA_WIDTH (32),
+    .ID_WIDTH        (4),  // 匹配 Crossbar 的 Master ID 宽度
+    .USER_WIDTH      (1),
+    .PORTS           (4),  // 8 个通道
+    .FIFO_DEPTH      (64)
+) u_snix_axil_cdma_mux_8ch (
+    .clk             (sys_clk),
+    .rst_n           (sys_resetn), // 低电平复位
+
+    // AXI-Lite 从机接口 (接 CPU 发来的配置请求)
+    .s_axil_awaddr   (cdma_s_awaddr),
+    .s_axil_awvalid  (cdma_s_awvalid),
+    .s_axil_awready  (cdma_s_awready),
+    .s_axil_wdata    (cdma_s_wdata),
+    .s_axil_wstrb    (cdma_s_wstrb),
+    .s_axil_wvalid   (cdma_s_wvalid),
+    .s_axil_wready   (cdma_s_wready),
+    .s_axil_bresp    (cdma_s_bresp),
+    .s_axil_bvalid   (cdma_s_bvalid),
+    .s_axil_bready   (cdma_s_bready),
+    .s_axil_araddr   (cdma_s_araddr),
+    .s_axil_arvalid  (cdma_s_arvalid),
+    .s_axil_arready  (cdma_s_arready),
+    .s_axil_rdata    (cdma_s_rdata),
+    .s_axil_rresp    (cdma_s_rresp),
+    .s_axil_rvalid   (cdma_s_rvalid),
+    .s_axil_rready   (cdma_s_rready),
+
+    // AXI4 主机接口 (接 Crossbar 去搬运数据)
+    .mm2mm_awid      (cdma_m_awid),
+    .mm2mm_awaddr    (cdma_m_awaddr),
+    .mm2mm_awlen     (cdma_m_awlen),
+    .mm2mm_awsize    (cdma_m_awsize),
+    .mm2mm_awburst   (cdma_m_awburst),
+    .mm2mm_awlock    (cdma_m_awlock),
+    .mm2mm_awcache   (cdma_m_awcache),
+    .mm2mm_awprot    (cdma_m_awprot),
+    .mm2mm_awqos     (),             // 悬空即可
+    .mm2mm_awuser    (),             // 悬空即可
+    .mm2mm_awvalid   (cdma_m_awvalid),
+    .mm2mm_awready   (cdma_m_awready),
+    .mm2mm_wdata     (cdma_m_wdata),
+    .mm2mm_wstrb     (cdma_m_wstrb),
+    .mm2mm_wlast     (cdma_m_wlast),
+    .mm2mm_wuser     (),             // 悬空即可
+    .mm2mm_wvalid    (cdma_m_wvalid),
+    .mm2mm_wready    (cdma_m_wready),
+    .mm2mm_bid       (cdma_m_bid),
+    .mm2mm_bresp     (cdma_m_bresp),
+    .mm2mm_buser     (1'b0),
+    .mm2mm_bvalid    (cdma_m_bvalid),
+    .mm2mm_bready    (cdma_m_bready),
+    .mm2mm_arid      (cdma_m_arid),
+    .mm2mm_araddr    (cdma_m_araddr),
+    .mm2mm_arlen     (cdma_m_arlen),
+    .mm2mm_arsize    (cdma_m_arsize),
+    .mm2mm_arburst   (cdma_m_arburst),
+    .mm2mm_arlock    (cdma_m_arlock),
+    .mm2mm_arcache   (cdma_m_arcache),
+    .mm2mm_arprot    (cdma_m_arprot),
+    .mm2mm_arqos     (),             // 悬空即可
+    .mm2mm_aruser    (),             // 悬空即可
+    .mm2mm_arvalid   (cdma_m_arvalid),
+    .mm2mm_arready   (cdma_m_arready),
+    .mm2mm_rid       (cdma_m_rid),
+    .mm2mm_rdata     (cdma_m_rdata),
+    .mm2mm_rresp     (cdma_m_rresp),
+    .mm2mm_rlast     (cdma_m_rlast),
+    .mm2mm_ruser     (1'b0),
+    .mm2mm_rvalid    (cdma_m_rvalid),
+    .mm2mm_rready    (cdma_m_rready),
+
+    .ep_wr_valid     (dma_ep_wr_valid),
+    .ep_wr_id        (dma_ep_wr_id),
+    .ep_wr_bank      (dma_ep_wr_bank),
+    .ep_wr_word_addr (dma_ep_wr_word_addr),
+    .ep_wr_data      (dma_ep_wr_data),
+    .ep_wr_ready     (dma_ep_wr_ready),
+    .ep_rd_valid     (dma_ep_rd_valid),
+    .ep_rd_id        (dma_ep_rd_id),
+    .ep_rd_bank      (dma_ep_rd_bank),
+    .ep_rd_word_addr (dma_ep_rd_word_addr),
+    .ep_rd_data      (dma_ep_rd_data),
+    .ep_rd_ready     (dma_ep_rd_ready),
+
+    // 全局中断输出
+    .dma_finish      (dma_finish)
+);
+
 endmodule

sdk/axi_ram.mif

@@ -24,7 +24,7 @@
 00010101011111100100000111101110
 00000011100101000000000111001110
 00101000100000000000000111001110
-00000011100000001000000000001101
+00000011100000000110110000001101
 01011000000000000000100000001110
 00000011100000000000010000001101
 00101001000000000000000110001101
@@ -39,142 +39,160 @@
 01010000000000000000000000000000
 00000000000000000000000000000000
 00010101011111100000000000001100
-00000010100000010011010000001111
+00000010100000010011010000010000
-00101001000000000000000110001111
-00000010100000010000010000010001
-00101001000000000000000110010001
-00000010100000010101000000010000
 00101001000000000000000110010000
-00101001000000000000000110001111
+00000010100000010000010000010010
-00000010100000010101010000000101
-00101001000000000000000110000101
-00000010100000010011000000000100
-00101001000000000000000110000100
-00000010100000010111110000010100
-00101001000000000000000110010100
-00000010100000010100110000000111
-00101001000000000000000110000111
-00101001000000000000000110010000
-00101001000000000000000110010001
-00000010100000010100100000010011
-00101001000000000000000110010011
-00010101011111101010000000001110
-00101001000000000000000110010000
-00000011100000000100000111010010
-00101001100000000000001001000000
-00000010100000000101100000000110
-00000011100000000001000111001101
-00101001100000000000000110100110
-00000011100000001000000111001000
-00010100001110001000000000001001
-00010100000000000000000000110010
-00101001100000000000000100001001
-00000011100000001010000111000111
-00000011100011100010001001010010
-00101001100000000000000011110010
-00000011100000001011000111000110
-00000010100000100000000000000111
-00101001100000000000000011000111
-00000010100001000000000000001001
-00101001100000000000000111001001
-00101001000000000000000110001111
-00101001000000000000000110010001
-00101001000000000000000110010000
-00101001000000000000000110001111
-00101001000000000000000110000101
-00101001000000000000000110000100
-00101001000000000000000110010100
-00000010100000010000110000010010
 00101001000000000000000110010010
-00101001000000000000000110010011
+00000010100000010101000000001111
-00101001000000000000000110010010
+00101001000000000000000110001111
-00000010100000001100110000001000
+00101001000000000000000110010000
-00101001000000000000000110001000
+00000010100000010101010000000110
-00000010100000001100100000000111
-00000010100000001111010000000110
-00101001000000000000000110000111
-00000010101111111100000001100011
 00101001000000000000000110000110
-00101000100000000000000110101100
+00000010100000010011000000001010
+00101001000000000000000110001010
+00000010100000010111110000000101
+00101001000000000000000110000101
+00000010100000010100110000010100
+00101001000000000000000110010100
+00101001000000000000000110001111
+00101001000000000000000110010010
+00000010100000010100100000000100
+00101001000000000000000110000100
+00010101011111101010000000001101
+00101001000000000000000110001111
+00000011100000000100000110110011
+00101001100000000000001001100000
+00000011100000000001000110101110
+00000010100000000101100000001001
+00101001100000000000000111001001
+00000011100000001000000110100111
+00010100001110001000000000001000
+00010100001110001001001110010001
+00101001100000000000000011101000
+00000011100000001001000110100110
+00000011100100000000001000110001
+00010100000000000000000000110000
+00101001100000000000000011010001
+00000011100000001010000110100101
+00000011100011100010001000010000
+00101001100000000000000010110000
+00000011100000001011000110110100
+00000010100000100000000000000100
+00101001100000000000001010000100
+00000011100000001100000110110010
+00000010100000110000000000010011
+00000010100001000000000000001100
+00101001100000000000001001010011
+00000010101111111100000001100011
+00010101011111101010000000001111
+00101001100000000000000110101100
+00101000100000000000000111001100
 00000011010000000100000110001100
 01011011111111111111100110000000
-00000010100000000100000000001111
+00000010100000000100000000010000
-00000011100000000100000111001100
+00000011100000000100000111101100
-00101001100000000000000110101111
+00101001100000000000000111010000
-00101000100000000000000110001101
+00101000100000000000000110001110
-00010101011111100000000000001110
+00010101011111100000000000001101
-00000011100000000001010111001110
+00000011100000000001010110101111
-00101010000000000000000111001100
+00101010000000000000000111101100
 00000011010000001000000110001100
 01011011111111111111100110000000
-00000000010001001111000110101100
+00000010100000010011010000010000
+00101001000000000000000110110000
+00000010100000010000010000001100
+00101001000000000000000110101100
+00000010100000010101000000010001
+00101001000000000000000110110001
+00000000010001001111000111001100
 00101001100000000011000001101100
-00101010000000000011000001101111
+00101010000000000011000001110001
-00000010100000000010010000010000
+00000010100000010101010000001000
-00000010100000010101110111101110
+00101001000000000000000110110000
-01101000000000000000101000001100
+00101001000000000000000110101000
-00000010100000001100000111101110
+00000010100000010011000000000111
-00010101011111100000000000010000
+00101001000000000000000110100111
-00000000010001001110000110101100
+00000010100000010111110000000110
+00101001000000000000000110100110
+00000010100000010000110000001111
+00101001000000000000000110101111
+00000010100000010100100000000101
+00101001000000000000000110100101
+00101001000000000000000110101111
+00000010100000001100110000000100
+00101001000000000000000110100100
+00000010100000001100100000010100
+00000010100000001111010000010011
+00101001000000000000000110110100
+00000010100000000010010000010010
+00101001000000000000000110110011
+00000010100000010101111000110000
+01101000000000000000101001001100
+00000010100000001100001000110000
+00010101011111100000000000001101
+00000000010001001110000111001100
 00000011010000000011110110001100
-00101001000000000000001000001110
+00101001000000000000000110110000
 00000010100000000010010000001111
-00000010100000010101110110001110
+00000010100000010101110110001101
 01101000000000000000100111101100
-00000010100000001100000110001110
+00000010100000001100000110001101
-00010101011111100000000000010000
-00000000010001001101000110101100
-00000011010000000011110110001100
-00101001000000000000001000001110
-00000010100000000010010000001111
-00000010100000010101110110001110
-01101000000000000000100111101100
-00000010100000001100000110001110
-00010101011111100000000000010000
-00000000010001001100000110101100
-00000011010000000011110110001100
-00101001000000000000001000001110
-00000010100000000010010000001111
-00000010100000010101110110001110
-01101000000000000000100111101100
-00000010100000001100000110001110
-00010101011111100000000000010000
-00000000010001001011000110101100
-00000011010000000011110110001100
-00101001000000000000001000001110
-00000010100000000010010000001111
-00000010100000010101110110001110
-01101000000000000000100111101100
-00000010100000001100000110001110
-00010101011111100000000000010000
-00000000010001001010000110101100
-00000011010000000011110110001100
-00101001000000000000001000001110
-00000010100000000010010000001111
-00000010100000010101110110001110
-01101000000000000000100111101100
-00000010100000001100000110001110
-00010101011111100000000000010000
-00000000010001001001000110101100
-00000011010000000011110110001100
-00101001000000000000001000001110
-00000010100000000010010000001111
-00000010100000010101110110001110
-01101000000000000000100111101100
-00000010100000001100000110001110
 00010101011111100000000000001111
-00000011010000000011110110101101
+00000000010001001101000111001100
-00101001000000000000000111101110
+00000011010000000011110110001100
+00101001000000000000000111101101
+00000010100000000010010000010000
+00000010100000010101110110001111
+01101100000000010001101000001100
+00010101011111100000000000001101
+00101001000000000000000110101111
+00000011100000000001010110101101
+00101010000000000000000110101100
+00000011010000001000000110001100
+01011011111111111111100110000000
+00000000010001001100000111001100
+00000011010000000011110110001100
+00000010100000000010010000001111
+00000010100000010101110110001101
+01101000000000000000100111101100
+00000010100000001100000110001101
+00010101011111100000000000010000
+00000000010001001011000111001100
+00000011010000000011110110001100
+00101001000000000000001000001101
+00000010100000000010010000001111
+00000010100000010101110110001101
+01101000000000000000100111101100
+00000010100000001100000110001101
+00010101011111100000000000010000
+00000000010001001010000111001100
+00000011010000000011110110001100
+00101001000000000000001000001101
+00000010100000000010010000001111
+00000010100000010101110110001101
+01101000000000000000100111101100
+00000010100000001100000110001101
+00010101011111100000000000010000
+00000000010001001001000111001100
+00000011010000000011110110001100
+00101001000000000000001000001101
+00000010100000000010010000001111
+00000010100000010101110110001101
+01101000000000000000100111101100
+00000010100000001100000110001101
+00010101011111100000000000001111
+00000011010000000011110111001110
+00101001000000000000000111101101
 00000010100000000010010000001100
-00000010100000010101110110101110
+00000010100000010101110111001111
-01101000000000000000100110001101
+01101000000000000000100110001110
-00000010100000001100000110101110
+00000010100000001100000111001111
 00010101011111100000000000001100
-00101001000000000000000110001110
+00101001000000000000000110001111
 00000010100000010011010000001101
 00101001000000000000000110001101
-00000010100000010000010000000100
+00000010100000010000010000001110
-00101001000000000000000110000100
+00101001000000000000000110001110
 00000010100000010101000000000101
 00101001000000000000000110000101
 00101001000000000000000110001101
@@ -195,6 +213,8 @@
 00000000000101010000000000000100
 00000010100000000100000001100011
 01001100000000000000000000100000
+00000010100000001100000110001111
+01010011111111101110101111111111
 00000011010000000000000000000000
 00000011010000000000000000000000
 00000011010000000000000000000000

sdk/software/bsp/env/start.S

@@ -118,7 +118,7 @@ _start:
 	#WRITE(li.wne,OFS_DIVISOR_LSB, divisor & 0xff);
     li.w        t3, 0xbf20f500
     ld.w        t3, t3, 0
-    li.w        t2, 0x20
+    li.w        t2, 0x1b
     beq         zero, t3, 1f
     li.w        t2, 0x1
 1:

sdk/software/examples/asm/Makefile

@@ -18,7 +18,6 @@ BSP_USE_TRAP := 0
 BSP_USE_DRIVERS := 0
 endif
 
-LA32RSOC_WINDOWS_HOME?=../../../..
 include ../../bsp/common.mk
 
 # Strip picolibc: no -lc -lm -lg -lsemihost, keep -lgcc only

sdk/software/examples/asm/start_matmul_fast.S

@@ -36,7 +36,7 @@ _start:
     st.b    zero, t0, 1
     li.w    t2, 0xbf20f500
     ld.w    t2, t2, 0
-    li.w    t1, 0x20
+    li.w    t1, 0x1b
     beq     zero, t2, 1f
     li.w    t1, 0x1
 1:

sdk/software/examples/asm/user-sample.c

@@ -1,7 +1,7 @@
 /**
  * user-sample - CIC-IEC 2026 Stage 3 Evaluation Program
  *
- * Matmul AXI-master CRC-only path with fast startup and low-overhead UART output.
+ * Matmul AXI-master path with fast startup and low-overhead UART output.
  */
 
 #include <dma.h>
@@ -33,23 +33,6 @@ static void uart_put_matmul_start_fifo(void)
     UART_THR = 'T';
 }
 
-static void uart_put_crc_prefix_fifo(void)
-{
-    UART_THR = 'M';
-    UART_THR = 'A';
-    UART_THR = 'T';
-    UART_THR = 'M';
-    UART_THR = 'U';
-    UART_THR = 'L';
-    UART_THR = '_';
-    UART_THR = 'C';
-    UART_THR = 'R';
-    UART_THR = 'C';
-    UART_THR = '3';
-    UART_THR = '2';
-    UART_THR = '=';
-}
-
 static inline void uart_wait_tx_empty(void)
 {
     while ((UART_LSR & LSR_THRE) == 0) {
@@ -66,9 +49,24 @@ static inline unsigned char hex_digit(unsigned int val, int shift)
 static void uart_put_result_fifo(unsigned int crc)
 {
     uart_wait_tx_empty();
+    UART_THR = 'M';
+    UART_THR = 'A';
+    UART_THR = 'T';
+    UART_THR = 'M';
+    UART_THR = 'U';
+    UART_THR = 'L';
+    UART_THR = '_';
+    UART_THR = 'C';
+    UART_THR = 'R';
+    UART_THR = 'C';
+    UART_THR = '3';
+    UART_THR = '2';
+    UART_THR = '=';
     UART_THR = hex_digit(crc, 28);
     UART_THR = hex_digit(crc, 24);
     UART_THR = hex_digit(crc, 20);
+
+    uart_wait_tx_empty();
     UART_THR = hex_digit(crc, 16);
     UART_THR = hex_digit(crc, 12);
     UART_THR = hex_digit(crc, 8);
@@ -99,7 +97,11 @@ unsigned long UART_BASE              = 0xbf000000;
 #define EXTRAM_BASE_UNCACHED (0xA0000000 | (EXTRAM_BASE_PHYS & 0x1FFFFFFF))
 
 #define SRC_BASE_WORDS    0
+#define DST_BASE_WORDS    (SRC_BASE_WORDS + MATMUL_GROUP_NUM * 32)
+#define RESULT_WORDS      (MATMUL_GROUP_NUM * 48)
+
 #define EXTRAM_PHYS_AB(g) (EXTRAM_BASE_PHYS + (SRC_BASE_WORDS * 4) + (g) * 128)
+#define EXTRAM_PHYS_C(g)  (EXTRAM_BASE_PHYS + (DST_BASE_WORDS * 4) + (g) * 192)
 
 #define MATMUL_BASE_PHYS     0x1f500000
 #define MATMUL_BASE_UNCACHED 0xbf500000
@@ -107,8 +109,10 @@ unsigned long UART_BASE              = 0xbf000000;
 #define MATMUL_CTRL       (MATMUL_BASE_UNCACHED + 0x000)
 #define MATMUL_STATUS     (MATMUL_BASE_UNCACHED + 0x004)
 #define MATMUL_DMA_SRC_BASE   (MATMUL_BASE_UNCACHED + 0x020)
+#define MATMUL_DMA_DST_BASE   (MATMUL_BASE_UNCACHED + 0x024)
 #define MATMUL_DMA_GROUPS     (MATMUL_BASE_UNCACHED + 0x028)
 #define MATMUL_DMA_SRC_STRIDE (MATMUL_BASE_UNCACHED + 0x02C)
+#define MATMUL_DMA_DST_STRIDE (MATMUL_BASE_UNCACHED + 0x030)
 
 #define CTRL_START_B0     0x1
 #define CTRL_START_B1     0x2
@@ -140,10 +144,11 @@ int main(void)
     *(volatile unsigned int *)MATMUL_CRC = 0;
     *mat_status = STATUS_DMA_DONE | STATUS_DONE | STATUS_ERROR;
     *(volatile unsigned int *)MATMUL_DMA_SRC_BASE = EXTRAM_PHYS_AB(0);
+    *(volatile unsigned int *)MATMUL_DMA_DST_BASE = EXTRAM_PHYS_C(0);
     *(volatile unsigned int *)MATMUL_DMA_GROUPS = MATMUL_GROUP_NUM;
     *(volatile unsigned int *)MATMUL_DMA_SRC_STRIDE = 128;
+    *(volatile unsigned int *)MATMUL_DMA_DST_STRIDE = 192;
     *mat_ctrl = CTRL_DMA_START;
-    uart_put_crc_prefix_fifo();
     while (!(*mat_status & STATUS_DMA_DONE)) {}
     *mat_status = STATUS_DMA_DONE;