URL
https://opencores.org/ocsvn/mpmc8/mpmc8/trunk
Subversion Repositories mpmc8
[/] [mpmc8/] [trunk/] [rtl/] [mpmc10/] [mpmc10_wb.sv] - Rev 7
Go to most recent revision | Compare with Previous | Blame | View Log
`timescale 1ns / 1ps
// ============================================================================
// __
// \\__/ o\ (C) 2015-2022 Robert Finch, Waterloo
// \ __ / All rights reserved.
// \/_// robfinch<remove>@finitron.ca
// ||
//
// BSD 3-Clause License
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
// 1. Redistributions of source code must retain the above copyright notice, this
// list of conditions and the following disclaimer.
//
// 2. Redistributions in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// 3. Neither the name of the copyright holder nor the names of its
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
// DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
// FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
// SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
// OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
// 16000 LUTs, 130 BRAM (64kB cache)
//
// Read channels always wait until there is valid data in the cache.
// ============================================================================
//
//`define RED_SCREEN 1'b1
import wishbone_pkg::*;
import mpmc10_pkg::*;
module mpmc10_wb(
input rst,
input clk100MHz,
input mem_ui_rst,
input mem_ui_clk,
input calib_complete,
output reg rstn,
output [31:0] app_waddr,
input app_rdy,
output app_en,
output [2:0] app_cmd,
output [28:0] app_addr,
input app_rd_data_valid,
output [15:0] app_wdf_mask,
output reg [127:0] app_wdf_data,
input app_wdf_rdy,
output app_wdf_wren,
output app_wdf_end,
input [127:0] app_rd_data,
input app_rd_data_end,
input ch0clk, ch1clk, ch2clk, ch3clk, ch4clk, ch5clk, ch6clk, ch7clk,
input wb_write_request128_t ch0i,
output wb_read_response128_t ch0o,
input wb_write_request128_t ch1i,
output wb_read_response128_t ch1o,
input wb_write_request128_t ch2i,
output wb_read_response128_t ch2o,
input wb_write_request128_t ch3i,
output wb_read_response128_t ch3o,
input wb_write_request128_t ch4i,
output wb_read_response128_t ch4o,
input wb_write_request128_t ch5i,
output wb_read_response128_t ch5o,
input wb_write_request128_t ch6i,
output wb_read_response128_t ch6o,
input wb_write_request128_t ch7i,
output wb_read_response128_t ch7o,
output mpmc10_state_t state
);
parameter NAR = 2; // Number of address reservations
parameter CL = 3'd4; // Cache read latency
parameter STREAM0 = TRUE;
parameter STREAM1 = FALSE;
parameter STREAM2 = FALSE;
parameter STREAM3 = FALSE;
parameter STREAM4 = FALSE;
parameter STREAM5 = TRUE;
parameter STREAM6 = FALSE;
parameter STREAM7 = FALSE;
wb_write_request128_t ch0is;
wb_write_request128_t ch0is2;
wb_write_request128_t ch1is;
wb_write_request128_t ch1is2;
wb_write_request128_t ch2is;
wb_write_request128_t ch2is2;
wb_write_request128_t ch3is;
wb_write_request128_t ch3is2;
wb_write_request128_t ch4is;
wb_write_request128_t ch4is2;
wb_write_request128_t ch5is;
wb_write_request128_t ch5is2;
wb_write_request128_t ch6is;
wb_write_request128_t ch6is2;
wb_write_request128_t ch7is;
wb_write_request128_t ch7is2;
wb_read_response128_t ch0oa, ch0ob;
wb_read_response128_t ch1oa, ch1ob;
wb_read_response128_t ch2oa, ch2ob;
wb_read_response128_t ch3oa, ch3ob;
wb_read_response128_t ch4oa, ch4ob;
wb_read_response128_t ch5oa, ch5ob;
wb_read_response128_t ch6oa, ch6ob;
wb_read_response128_t ch7oa, ch7ob;
assign ch0o = STREAM0 ? ch0ob : ch0oa;
assign ch1o = STREAM1 ? ch1ob : ch1oa;
assign ch2o = STREAM2 ? ch2ob : ch2oa;
assign ch3o = STREAM3 ? ch3ob : ch3oa;
assign ch4o = STREAM4 ? ch4ob : ch4oa;
assign ch5o = STREAM5 ? ch5ob : ch5oa;
assign ch6o = STREAM6 ? ch6ob : ch6oa;
assign ch7o = STREAM7 ? ch7ob : ch7oa;
wb_write_request128_t req_fifoi;
wb_write_request128_t req_fifoo;
wb_write_request128_t ld;
wb_write_request128_t fifo_mask;
wb_write_request128_t fifoo = req_fifoo & fifo_mask;
genvar g;
integer n1,n2,n3;
wire almost_full;
wire [4:0] cnt;
reg wr_fifo;
mpmc10_state_t prev_state;
wire rd_fifo; // from state machine
reg [5:0] num_strips; // from fifo
wire [5:0] req_strip_cnt;
wire [5:0] resp_strip_cnt;
wire [15:0] tocnt;
wire [31:0] app_waddr;
reg [31:0] adr;
reg [3:0] uch; // update channel
wire [15:0] wmask;
wire [15:0] mem_wdf_mask2;
reg [127:0] dat128;
wire [127:0] dat256;
wire [3:0] resv_ch [0:NAR-1];
wire [31:0] resv_adr [0:NAR-1];
wire rb1;
reg [7:0] req;
reg [127:0] rd_data_r;
reg rd_data_valid_r;
wire ch0_hit_s, ch1_hit_s, ch2_hit_s, ch3_hit_s;
wire ch4_hit_s, ch5_hit_s, ch6_hit_s, ch7_hit_s;
wire ch0_hit_ne, ch5_hit_ne;
always_ff @(posedge mem_ui_clk)
rd_data_r <= app_rd_data;
always_ff @(posedge mem_ui_clk)
rd_data_valid_r <= app_rd_data_valid;
reg [19:0] rst_ctr;
always @(posedge clk100MHz)
if (rst)
rst_ctr <= 24'd0;
else begin
if (!rst_ctr[15])
rst_ctr <= rst_ctr + 2'd1;
rstn <= rst_ctr[15];
end
reg [7:0] stb [0:7];
always_comb stb[0] = ch0is.stb;
always_comb stb[1] = ch1is.stb;
always_comb stb[2] = ch2is.stb;
always_comb stb[3] = ch3is.stb;
always_comb stb[4] = ch4is.stb;
always_comb stb[5] = ch5is.stb;
always_comb stb[6] = ch6is.stb;
always_comb stb[7] = ch7is.stb;
reg [2:0] chcnt [0:7];
always_ff @(posedge mem_ui_clk)
if (rst) begin
for (n2 = 0; n2 < 8; n2 = n2 + 1)
chcnt[n2] <= 'd0;
end
else begin
for (n2 = 0; n2 < 8; n2 = n2 + 1)
if (stb[n2]) begin
if (chcnt[n2] < CL)
chcnt[n2] <= chcnt[n2] + 2'd1;
end
else
chcnt[n2] <= 'd0;
end
wire [7:0] pe_req;
reg [7:0] chack;
always_comb chack[0] = ch0o.ack;
always_comb chack[1] = ch1o.ack;
always_comb chack[2] = ch2o.ack;
always_comb chack[3] = ch3o.ack;
always_comb chack[4] = ch4o.ack;
always_comb chack[5] = ch5o.ack;
always_comb chack[6] = ch6o.ack;
always_comb chack[7] = ch7o.ack;
reg [7:0] reqa;
always_comb reqa[1] = (!ch1o.ack && ch1is.stb && !ch1is.we && chcnt[1]==CL) || (ch1is.we && ch1is.stb);
always_comb reqa[5] = (!ch5o.ack && ch5is.stb && !ch5is.we && chcnt[5]==CL) || (ch5is.we && ch5is.stb);
wire rste = mem_ui_rst||rst||!calib_complete;
edge_det edch0 (
.rst(rste),
.clk(mem_ui_clk),
.ce(1'b1),
.i((!ch0o.ack && ch0is.stb && !ch0is.we && chcnt[0]==CL) || (ch0is.we && ch0is.stb)),
.pe(pe_req[0]),
.ne(),
.ee()
);
edge_det edch1 (
.rst(rste),
.clk(mem_ui_clk),
.ce(1'b1),
.i(reqa[1]),
.pe(pe_req[1]),
.ne(),
.ee()
);
edge_det edch2 (
.rst(rste),
.clk(mem_ui_clk),
.ce(1'b1),
.i((!ch2o.ack && ch2is.stb && !ch2is.we && chcnt[2]==CL) || (ch2is.we && ch2is.stb)),
.pe(pe_req[2]),
.ne(),
.ee()
);
edge_det edch3 (
.rst(rste),
.clk(mem_ui_clk),
.ce(1'b1),
.i((!ch3o.ack && ch3is.stb && !ch3is.we && chcnt[3]==CL) || (ch3is.we && ch3is.stb)),
.pe(pe_req[3]),
.ne(),
.ee()
);
edge_det edch4 (
.rst(rste),
.clk(mem_ui_clk),
.ce(1'b1),
.i((!ch4o.ack && ch4is.stb && !ch4is.we && chcnt[4]==CL) || (ch4is.we && ch4is.stb)),
.pe(pe_req[4]),
.ne(),
.ee()
);
edge_det edch5 (
.rst(rste),
.clk(mem_ui_clk),
.ce(1'b1),
.i(reqa[5]),
.pe(pe_req[5]),
.ne(),
.ee()
);
edge_det edch6 (
.rst(rste),
.clk(mem_ui_clk),
.ce(1'b1),
.i((!ch6o.ack && ch6is.stb && !ch6is.we && chcnt[6]==CL) || (ch6is.we && ch6is.stb)),
.pe(pe_req[6]),
.ne(),
.ee()
);
edge_det edch7 (
.rst(rste),
.clk(mem_ui_clk),
.ce(1'b1),
.i((!ch7o.ack && ch7is.stb && !ch7is.we && chcnt[7]==CL) || (ch7is.we && ch7is.stb)),
.pe(pe_req[7]),
.ne(),
.ee()
);
wire [3:0] req_sel;
always_ff @(posedge mem_ui_clk)
for (n3 = 0; n3 < 8; n3 = n3 + 1)
if (pe_req[n3])
req[n3] <= 1'b1;
else if ((req_sel==n3[3:0]) || chack[n3])
req[n3] <= 1'b0;
// Register signals onto mem_ui_clk domain
mpmc10_sync128_wb usyn0
(
.clk(mem_ui_clk),
.i(ch0i),
.o(ch0is)
);
mpmc10_sync128_wb usyn1
(
.clk(mem_ui_clk),
.i(ch1i),
.o(ch1is)
);
mpmc10_sync128_wb usyn2
(
.clk(mem_ui_clk),
.i(ch2i),
.o(ch2is)
);
mpmc10_sync128_wb usyn3
(
.clk(mem_ui_clk),
.i(ch3i),
.o(ch3is)
);
mpmc10_sync128_wb usyn4
(
.clk(mem_ui_clk),
.i(ch4i),
.o(ch4is)
);
mpmc10_sync128_wb usyn5
(
.clk(mem_ui_clk),
.i(ch5i),
.o(ch5is)
);
mpmc10_sync128_wb usyn6
(
.clk(mem_ui_clk),
.i(ch6i),
.o(ch6is)
);
mpmc10_sync128_wb usyn7
(
.clk(mem_ui_clk),
.i(ch7i),
.o(ch7is)
);
// Streaming channels have a burst length of 64. Round the address to the burst
// length.
always_comb
begin
ch0is2 <= ch0is;
ch0is2.adr <= {ch0is.adr[31:10],10'b0};
end
always_comb
begin
ch1is2 <= ch1is;
ch1is2.adr <= {ch1is.adr[31:10],10'b0};
end
always_comb
begin
ch2is2 <= ch2is;
ch2is2.adr <= {ch2is.adr[31:10],10'b0};
end
always_comb
begin
ch3is2 <= ch3is;
ch3is2.adr <= {ch3is.adr[31:10],10'b0};
end
always_comb
begin
ch4is2 <= ch4is;
ch4is2.adr <= {ch4is.adr[31:10],10'b0};
end
always_comb
begin
ch5is2 <= ch5is;
ch5is2.adr <= {ch5is.adr[31:10],10'b0};
end
always_comb
begin
ch6is2 <= ch6is;
ch6is2.adr <= {ch6is.adr[31:10],10'b0};
end
always_comb
begin
ch7is2 <= ch7is;
ch7is2.adr <= {ch7is.adr[31:10],10'b0};
end
always_comb
begin
ld.bte <= wishbone_pkg::LINEAR;
ld.cti <= wishbone_pkg::CLASSIC;
ld.blen <= 'd0;
ld.cyc <= fifoo.stb && !fifoo.we && rd_data_valid_r && (uch!=4'd0 && uch!=4'd5 && uch!=4'd15);
ld.stb <= fifoo.stb && !fifoo.we && rd_data_valid_r && (uch!=4'd0 && uch!=4'd5 && uch!=4'd15);
ld.we <= 1'b0;
ld.adr <= {app_waddr[31:4],4'h0};
ld.dat <= {app_waddr[31:14],8'h00,rd_data_r}; // modified=false,tag = high order address bits
ld.sel <= {36{1'b1}}; // update all bytes
end
reg ch0wack;
reg ch1wack;
reg ch2wack;
reg ch3wack;
reg ch4wack;
reg ch5wack;
reg ch6wack;
reg ch7wack;
always_ff @(posedge mem_ui_clk)
begin
if (!ch0i.stb) ch0wack <= 1'b0;
if (!ch1i.stb) ch1wack <= 1'b0;
if (!ch2i.stb) ch2wack <= 1'b0;
if (!ch3i.stb) ch3wack <= 1'b0;
if (!ch4i.stb) ch4wack <= 1'b0;
if (!ch5i.stb) ch5wack <= 1'b0;
if (!ch6i.stb) ch6wack <= 1'b0;
if (!ch7i.stb) ch7wack <= 1'b0;
if (state==WRITE_DATA3)
case(uch)
4'd0: ch0wack <= 1'b1;
4'd1: ch1wack <= 1'b1;
4'd2: ch2wack <= 1'b1;
4'd3: ch3wack <= 1'b1;
4'd4: ch4wack <= 1'b1;
4'd5: ch5wack <= 1'b1;
4'd6: ch6wack <= 1'b1;
4'd7: ch7wack <= 1'b1;
default: ;
endcase
end
mpmc10_cache_wb ucache1
(
.rst(mem_ui_rst),
.wclk(mem_ui_clk),
.inv(1'b0),
.wchi(fifoo),
.wcho(),
.ld(ld),
.ch0clk(STREAM0 ? 1'b0 : ch0clk),
.ch1clk(STREAM1 ? 1'b0 : ch1clk),
.ch2clk(STREAM2 ? 1'b0 : ch2clk),
.ch3clk(STREAM3 ? 1'b0 : ch3clk),
.ch4clk(STREAM4 ? 1'b0 : ch4clk),
.ch5clk(STREAM5 ? 1'b0 : ch5clk),
.ch6clk(STREAM6 ? 1'b0 : ch6clk),
.ch7clk(STREAM7 ? 1'b0 : ch7clk),
.ch0i(STREAM0 ? 'd0 : ch0is),
.ch1i(STREAM1 ? 'd0 : ch1is),
.ch2i(STREAM2 ? 'd0 : ch2is),
.ch3i(STREAM3 ? 'd0 : ch3is),
.ch4i(STREAM4 ? 'd0 : ch4is),
.ch5i(STREAM5 ? 'd0 : ch5is),
.ch6i(STREAM6 ? 'd0 : ch6is),
.ch7i(STREAM7 ? 'd0 : ch7is),
.ch0wack(ch0wack),
.ch1wack(ch1wack),
.ch2wack(ch2wack),
.ch3wack(ch3wack),
.ch4wack(ch4wack),
.ch5wack(ch5wack),
.ch6wack(ch6wack),
.ch7wack(ch7wack),
.ch0o(ch0oa),
.ch1o(ch1oa),
.ch2o(ch2oa),
.ch3o(ch3oa),
.ch4o(ch4oa),
.ch5o(ch5oa),
.ch6o(ch6oa),
.ch7o(ch7oa)
);
mpmc10_strm_read_cache ustrm0
(
.rst(rst),
.wclk(mem_ui_clk),
.wr(uch==4'd0 && rd_data_valid_r),
.wadr({app_waddr[31:4],4'h0}),
.wdat(rd_data_r),
// .inv(1'b0),
.rclk(mem_ui_clk),
.rd(ch0is.stb & ~ch0is.we),
.radr({ch0is.adr[31:4],4'h0}),
.rdat(ch0ob.dat),
.hit(ch0_hit_s)
);
mpmc10_strm_read_cache ustrm1
(
.rst(rst),
.wclk(mem_ui_clk),
.wr(uch==4'd1 && rd_data_valid_r),
.wadr({app_waddr[31:4],4'h0}),
.wdat(rd_data_r),
// .inv(1'b0),
.rclk(mem_ui_clk),
.rd(ch1is.stb & ~ch1is.we),
.radr({ch1is.adr[31:4],4'h0}),
.rdat(ch1ob.dat),
.hit(ch1_hit_s)
);
mpmc10_strm_read_cache ustrm2
(
.rst(rst),
.wclk(mem_ui_clk),
.wr(uch==4'd2 && rd_data_valid_r),
.wadr({app_waddr[31:4],4'h0}),
.wdat(rd_data_r),
// .inv(1'b0),
.rclk(mem_ui_clk),
.rd(ch2is.stb & ~ch2is.we),
.radr({ch2is.adr[31:4],4'h0}),
.rdat(ch2ob.dat),
.hit(ch2_hit_s)
);
mpmc10_strm_read_cache ustrm3
(
.rst(rst),
.wclk(mem_ui_clk),
.wr(uch==4'd3 && rd_data_valid_r),
.wadr({app_waddr[31:4],4'h0}),
.wdat(rd_data_r),
// .inv(1'b0),
.rclk(mem_ui_clk),
.rd(ch3is.stb & ~ch3is.we),
.radr({ch3is.adr[31:4],4'h0}),
.rdat(ch3ob.dat),
.hit(ch3_hit_s)
);
mpmc10_strm_read_cache ustrm4
(
.rst(rst),
.wclk(mem_ui_clk),
.wr(uch==4'd4 && rd_data_valid_r),
.wadr({app_waddr[31:4],4'h0}),
.wdat(rd_data_r),
// .inv(1'b0),
.rclk(mem_ui_clk),
.rd(ch4is.stb & ~ch4is.we),
.radr({ch4is.adr[31:4],4'h0}),
.rdat(ch4ob.dat),
.hit(ch4_hit_s)
);
mpmc10_strm_read_cache ustrm5
(
.rst(rst),
.wclk(mem_ui_clk),
.wr(uch==4'd5 && rd_data_valid_r),
.wadr({app_waddr[31:4],4'h0}),
.wdat(rd_data_r),
// .inv(1'b0),
.rclk(mem_ui_clk),
.rd(ch5is.stb & ~ch5is.we),
.radr({ch5is.adr[31:4],4'h0}),
.rdat(ch5ob.dat),
.hit(ch5_hit_s)
);
mpmc10_strm_read_cache ustrm6
(
.rst(rst),
.wclk(mem_ui_clk),
.wr(uch==4'd6 && rd_data_valid_r),
.wadr({app_waddr[31:4],4'h0}),
.wdat(rd_data_r),
// .inv(1'b0),
.rclk(mem_ui_clk),
.rd(ch6is.stb & ~ch6is.we),
.radr({ch6is.adr[31:4],4'h0}),
.rdat(ch6ob.dat),
.hit(ch6_hit_s)
);
mpmc10_strm_read_cache ustrm7
(
.rst(rst),
.wclk(mem_ui_clk),
.wr(uch==4'd7 && rd_data_valid_r),
.wadr({app_waddr[31:4],4'h0}),
.wdat(rd_data_r),
// .inv(1'b0),
.rclk(mem_ui_clk),
.rd(ch7is.stb & ~ch7is.we),
.radr({ch7is.adr[31:4],4'h0}),
.rdat(ch7ob.dat),
.hit(ch7_hit_s)
);
always_comb ch0ob.ack = ch0_hit_s & ch0i.stb;
always_comb ch1ob.ack = ch1_hit_s & ch1i.stb;
always_comb ch2ob.ack = ch2_hit_s & ch2i.stb;
always_comb ch3ob.ack = ch3_hit_s & ch3i.stb;
always_comb ch4ob.ack = ch4_hit_s & ch4i.stb;
always_comb ch5ob.ack = ch5_hit_s & ch5i.stb;
always_comb ch6ob.ack = ch6_hit_s & ch6i.stb;
always_comb ch7ob.ack = ch7_hit_s & ch7i.stb;
wire [7:0] sel;
wire rd_rst_busy;
wire wr_rst_busy;
wire cd_sel;
change_det #(.WID(8)) ucdsel (.rst(rst), .ce(1'b1), .clk(mem_ui_clk), .i(sel), .cd(cd_sel));
always_comb //ff @(posedge mem_ui_clk)
wr_fifo = |sel & ~almost_full & ~wr_rst_busy & cd_sel;
roundRobin rr1
(
.rst(rst),
.clk(mem_ui_clk),
.ce(1'b1),//~|req || chack[req_sel]),
.req(req),
.lock(8'h00),
.sel(sel),
.sel_enc(req_sel)
);
always_comb
case(req_sel)
4'd0: req_fifoi <= STREAM0 ? ch0is2 : ch0is;
4'd1: req_fifoi <= STREAM1 ? ch1is2 : ch1is;
4'd2: req_fifoi <= STREAM2 ? ch2is2 : ch2is;
4'd3: req_fifoi <= STREAM3 ? ch3is2 : ch3is;
4'd4: req_fifoi <= STREAM4 ? ch4is2 : ch4is;
4'd5: req_fifoi <= STREAM5 ? ch5is2 : ch5is;
4'd6: req_fifoi <= STREAM6 ? ch6is2 : ch6is;
4'd7: req_fifoi <= STREAM7 ? ch7is2 : ch7is;
default:
begin
req_fifoi <= 'd0;
req_fifoi.cid <= 4'd15;
end
endcase
mpmc10_fifo ufifo1
(
.rst(rst),
.clk(mem_ui_clk),
.rd_fifo(rd_fifo),
.wr_fifo(wr_fifo),
.req_fifoi(req_fifoi),
.req_fifoo(req_fifoo),
.v(v),
.full(full),
.empty(empty),
.almost_full(almost_full),
.rd_rst_busy(rd_rst_busy),
.wr_rst_busy(wr_rst_busy),
.cnt(cnt)
);
always_comb
uch <= fifoo.cid;
always_comb
num_strips <= fifoo.blen;
always_comb
adr <= fifoo.adr;
wire [2:0] app_addr3; // dummy to make up 32-bits
mpmc10_addr_gen uag1
(
.rst(mem_ui_rst),
.clk(mem_ui_clk),
.state(state),
.rdy(app_rdy),
.num_strips(num_strips),
.strip_cnt(req_strip_cnt),
.addr_base(adr),
.addr({app_addr3,app_addr})
);
mpmc10_waddr_gen uwag1
(
.rst(mem_ui_rst),
.clk(mem_ui_clk),
.state(state),
.valid(rd_data_valid_r),
.num_strips(num_strips),
.strip_cnt(resp_strip_cnt),
.addr_base(adr),
.addr(app_waddr)
);
mpmc10_set_write_mask_wb uswm1
(
.clk(mem_ui_clk),
.state(state),
.we(fifoo.we),
.sel(req_fifoo.sel[15:0]),
.adr(adr|{req_strip_cnt[0],4'h0}),
.mask(wmask)
);
mpmc10_mask_select unsks1
(
.rst(mem_ui_rst),
.clk(mem_ui_clk),
.state(state),
.wmask(wmask),
.mask(app_wdf_mask),
.mask2(mem_wdf_mask2)
);
mpmc10_data_select #(.WID(128)) uds1
(
.clk(mem_ui_clk),
.state(state),
.dati(req_fifoo.dat),
.dato(dat256)
);
always_comb
dat128 <= dat256;//req_strip_cnt[0] ? dat256[255:128] : dat256[127:0];
// Setting the data value. Unlike reads there is only a single strip involved.
// Force unselected byte lanes to $FF
reg [127:0] dat128x;
generate begin
for (g = 0; g < 16; g = g + 1)
always_comb
if (mem_wdf_mask2[g])
dat128x[g*8+7:g*8] = 8'hFF;
else
dat128x[g*8+7:g*8] = dat128[g*8+7:g*8];
end
endgenerate
always_ff @(posedge mem_ui_clk)
if (mem_ui_rst)
app_wdf_data <= 128'd0;
else begin
if (state==PRESET3)
app_wdf_data <= dat128x;
// else if (state==WRITE_TRAMP1)
// app_wdf_data <= rmw_data;
end
mpmc10_rd_fifo_gen urdf1
(
.rst(rst|mem_ui_rst),
.clk(mem_ui_clk),
.state(state),
.empty(empty),
.rd_rst_busy(rd_rst_busy),
.calib_complete(calib_complete),
.rd(rd_fifo)
);
always_ff @(posedge mem_ui_clk)
if (rst)
fifo_mask <= 'd0;
else begin
if (rd_fifo)
fifo_mask <= {$bits(fifo_mask){1'b1}};
else if (state==IDLE)
fifo_mask <= 'd0;
end
mpmc10_state_machine_wb usm1
(
.rst(rst|mem_ui_rst),
.clk(mem_ui_clk),
.calib_complete(calib_complete),
.to(tocnt[9]),
.rdy(app_rdy),
.wdf_rdy(app_wdf_rdy),
.fifo_empty(empty),
.rd_rst_busy(rd_rst_busy),
.fifo_out(req_fifoo),
.state(state),
.num_strips(num_strips),
.req_strip_cnt(req_strip_cnt),
.resp_strip_cnt(resp_strip_cnt),
.rd_data_valid(rd_data_valid_r)
);
mpmc10_to_cnt utoc1
(
.clk(mem_ui_clk),
.state(state),
.prev_state(prev_state),
.to_cnt(tocnt)
);
mpmc10_prev_state upst1
(
.clk(mem_ui_clk),
.state(state),
.prev_state(prev_state)
);
mpmc10_app_en_gen ueng1
(
.clk(mem_ui_clk),
.state(state),
.rdy(app_rdy),
.strip_cnt(req_strip_cnt),
.num_strips(num_strips),
.en(app_en)
);
mpmc10_app_cmd_gen ucg1
(
.clk(mem_ui_clk),
.state(state),
.cmd(app_cmd)
);
mpmc10_app_wdf_wren_gen uwreng1
(
.clk(mem_ui_clk),
.state(state),
.rdy(app_wdf_rdy),
.wren(app_wdf_wren)
);
mpmc10_app_wdf_end_gen uwendg1
(
.clk(mem_ui_clk),
.state(state),
.rdy(app_wdf_rdy),
.strip_cnt(req_strip_cnt),
.num_strips(num_strips),
.wend(app_wdf_end)
);
mpmc10_req_strip_cnt ursc1
(
.clk(mem_ui_clk),
.state(state),
.wdf_rdy(app_wdf_rdy),
.rdy(app_rdy),
.num_strips(num_strips),
.strip_cnt(req_strip_cnt)
);
mpmc10_resp_strip_cnt urespsc1
(
.clk(mem_ui_clk),
.state(state),
.valid(rd_data_valid_r),
.num_strips(num_strips),
.strip_cnt(resp_strip_cnt)
);
// Reservation status bit
mpmc10_resv_bit ursb1
(
.clk(mem_ui_clk),
.state(state),
.wch(fifoo.cid),
.we(fifoo.stb & fifoo.we),
.cr(fifoo.csr & fifoo.we),
.adr(fifoo.adr),
.resv_ch(resv_ch),
.resv_adr(resv_adr),
.rb(rb1)
);
mpmc10_addr_resv_man #(.NAR(NAR)) ursvm1
(
.rst(mem_ui_rst),
.clk(mem_ui_clk),
.state(state),
.adr0(32'h0),
.adr1(ch1is.adr),
.adr2(ch2is.adr),
.adr3(ch3is.adr),
.adr4(ch4is.adr),
.adr5(32'h0),
.adr6(ch6is.adr),
.adr7(ch7is.adr),
.sr0(1'b0),
.sr1(ch1is.csr & ch1is.stb & ~ch1is.we),
.sr2(ch2is.csr & ch2is.stb & ~ch2is.we),
.sr3(ch3is.csr & ch3is.stb & ~ch3is.we),
.sr4(ch4is.csr & ch4is.stb & ~ch4is.we),
.sr5(1'b0),
.sr6(ch6is.csr & ch6is.stb & ~ch6is.we),
.sr7(ch7is.csr & ch7is.stb & ~ch7is.we),
.wch(fifoo.stb ? fifoo.cid : 4'd15),
.we(fifoo.stb & fifoo.we),
.wadr(fifoo.adr),
.cr(fifoo.csr & fifoo.stb & fifoo.we),
.resv_ch(resv_ch),
.resv_adr(resv_adr)
);
endmodule
Go to most recent revision | Compare with Previous | Blame | View Log