Commit 6e79acaf authored by Andrey Filippov's avatar Andrey Filippov

address generation for different types of macroblocks

parent 8b1599eb
...@@ -23,234 +23,159 @@ ...@@ -23,234 +23,159 @@
`timescale 1ns/1ps `timescale 1ns/1ps
module cmprs_pixel_buf_iface #( module cmprs_pixel_buf_iface #(
parameter CMPRS_PREEND_EARLY = 8, // TODO: adjust according to cmprs_macroblock_buf_iface latency. In
// color18 mode this should be later than end of address run - (6*64>18*18)
// "0" would generate pulse at eth same time as next macro mb_pre_start
parameter CMPRS_RELEASE_EARLY = 16, // set to minimal actual latency in memory read, but not more than
parameter CMPRS_BUF_EXTRA_LATENCY = 0 // extra register layers insered between the buffer and this module parameter CMPRS_BUF_EXTRA_LATENCY = 0 // extra register layers insered between the buffer and this module
)( )(
input rst,
input xclk, // global clock input, compressor single clock rate input xclk, // global clock input, compressor single clock rate
input frame_en, // if 0 - will reset logic immediately (but not page number)
input mclk, // global clock for commands (posedge) and write side of the memory buffer (negedge)
// buffer interface, DDR3 memory read // buffer interface, DDR3 memory read
input xfer_reset_page_rd, // @ negedge mclk - reset ddr3 memory buffer. Use it to reset the read buffer too
input page_ready_chn, // single mclk (posedge)
output next_page_chn, // single mclk (posedge): Done with the page in the buffer, memory controller may read more data
input [ 7:0] buf_di, // data from the buffer input [ 7:0] buf_di, // data from the buffer
output [11:0] buf_ra, // buffer read address (1 MSB - page number) output [11:0] buf_ra, // buffer read address (2 MSB - page number)
output [ 1:0] buf_rd, // buf {regen, re} output [ 1:0] buf_rd, // buf {regen, re}
input frame_en, // if 0 - will reset logic immediately (but not page number)
input frame_go, // start frame: if idle, will start reading data (if available),
// if running - will not restart a new frame if 0. // if running - will not restart a new frame if 0.
input [ 6:0] mode, // TODO: adjust width. Color mode that determins address mapping input [ 2:0] converter_type, // 0 - color18, 1 - color20, 2 - mono, 3 - jp4, 4 - jp4-diff, 7 - mono8 (not yet implemented)
input [ 4:0] left_marg, // left margin (for not-yet-implemented) mono JPEG (8 lines tile row) can need 7 bits (mod 32 - tile) input [ 5:0] mb_w_m1, // macroblock width minus 1 // 3 LSB not used, SHOULD BE SET to 3'b111
input [12:0] n_blocks_in_row_m1, // number of macroblocks in a macroblock row minus 1 input [ 5:0] mb_h_m1, // macroblock horizontal period (8/16) // 3 LSB not used SHOULD BE SET to 3'b111
input [12:0] n_block_rows_m1, // number of macroblock rows in a frame minus 1 input [ 1:0] tile_width, // memory tile width (can be 128 for monochrome JPEG) Can be 32/64/128: 0 - 16, 1 - 32, 2 - 64, 3 - 128
input tile_col_width, // 0 - 16 pixels, 1 -32 pixels
// Tiles/macroblocks level (from cmprs_macroblock_buf_iface)
output mb_pre_end, // just in time to start a new macroblock w/o gaps
output mb_release_buf, // send required "next_page" pulses to buffer. Having rather long minimal latency in the memory
// controller this can just be the same as mb_pre_end_in
input mb_pre_start, // 1 clock cycle before stream of addresses to the buffer
input [ 1:0] start_page, // page to read next tile from (or first of several pages)
input [ 6:0] macroblock_x, // macroblock left pixel x relative to a tile (page) Maximal page - 128 bytes wide
output [ 7:0] data_out, // output [ 7:0] data_out, //
output pre_first_out, // For each macroblock in a frame output pre_first_out, // For each macroblock in a frame
output data_valid // output data_valid //
); );
localparam CMPRS_MB_DLY=5; localparam PERIOD_COLOR18 = 384; // >18*18, limited by 6*64 (macroblocks)
wire buf_re_w; localparam PERIOD_COLOR20 = 400; // limited by the 20x20 padded macroblock
reg [CMPRS_BUF_EXTRA_LATENCY+2:0] buf_re; localparam PERIOD_MONO16 = 384; // 6*64 - sends 2 of zeroed blobks
reg [ 7:0] do_r; localparam PERIOD_JP4 = 256; // 4*64 - exact match
wire reset_page_rd; localparam PERIOD_JP4DIFF = 256; // TODO: see if correct
wire page_ready; localparam PERIOD_MONO8 = 64; // 1*64 - exact match - not yet implemented (notmal mono JPEG)
// wire next_page; // @ posedge xclk - source
// wire busy_next_page; // do not send next_page -previous is crossing clock boundaries
wire frame_end_w; // calculated
wire frame_en_w;
reg frame_en_r;
wire pre_frame_start;
wire frame_start;
wire en;
reg pre_en;
reg [11:0] mb_start; // full adderss of the next macroblock start;
reg [11:0] mbr_start; // start address of the next macroblock row
reg [11:0] buf_ra_r; // buffer read address
reg [ 5:0] mb_w_m1; // macroblock width minus 1
reg [ 5:0] mb_h_m1; // macroblock height minus 1
// reg [ 4:0] left_marg; // left margin of the leftmost macroblock column relative to read tiles (aligned to 32 bytes)
reg [ 4:3] mb_hper; // macroblock horizontal period (8/16)
reg [ 1:0] tile_width; // memory tile width (can be 128 for monochrome JPEG) Can be 32/64/128: 0 - 16, 1 - 32, 2 - 64, 3 - 128
reg [12:0] mb_col; // number of a macroblock in a row
reg [12:0] mb_row; // number of a macroblock row
wire [ 6:0] mbl_x; // macroblock left pixel x relative to a tile (page) Maximal page - 128 bytes wide
reg [ 6:3] mbl_x_r; // macroblock left pixel x relative to a tile (page) (3 low don't change)
reg [ 6:3] mbl_x_next_r; // macroblock left pixel x relative to a tile (page), not valid for first column (3 low don't change)
reg [ 7:3] mbl_x_inc_r; // intermediate register for calculating mbl_x_next_r and add_invalid
reg [ 7:3] mbl_x_last_r; // intermediate register for calculating needed_page
wire mb_pre_end_w; // start mb_pre_end sequence
reg [ 3:0] mb_pre_end; // 1-hot macroblock pre end calcualtions
// wire mb_pre_end_done = mb_pre_end[2]; // overlap
reg [1:0] pre_advance_tiles; // advance tiles by this for same row of macroblocks
wire mb_pre_start_w; // same timing as mb_pre_end[1] reg [CMPRS_BUF_EXTRA_LATENCY+3:0] buf_re=0;
reg [ 4:0] mb_pre_start; // 1-hot macroblock pre start calcualtions - TODO: adjust width reg [ 7:0] do_r;
wire [ 2:0] buf_diff; // difference between page needed and next valid - should be negative to have it ready reg [11:0] bufa_r; // buffer read address (2 MSB - page number)
wire buf_ready_w; // External memory buffer has all the pages needed reg [11:0] row_sa; // row start address
reg [ 9:0] tile_sa; // tile start address for the same row (w/o page number) for continuing row
reg mb_first_in_row; // to the next tile. Valid @ first column (first column is always from the start tile)
reg mb_last_in_row; reg [ 9:4] col_inc; // address increment when crossing tile column (1 + (macroblock_height - 1) * tile_column_width)
// inc by 1 - always
reg [ 2:0] next_valid; // number of next valid page (only 2 LSB are actual page number) reg [ 5:0] cols_left;
reg [ 2:0] next_invalid; // oldest valid page reg [ 5:0] rows_left;
reg [ 1:0] add_invalid; // advance next_invalid pointer by this value, send next_page pulses reg [ 6:0] tile_x; // horizontal position in a tile
reg we_invalid; // advance next_invalid pointer, send next_page pulses reg [ 4:0] column_x; // horizontal position in a column (0..31 or 0..15)
reg [ 2:0] used_pages; // number of pages simultaneously used for the last macroblock reg last_col;
reg [ 2:0] needed_page; // calculate at MB start reg first_col;
// reg [ 1:0] need_pages_m1; // number of tiles needed for macroblock being started minus 1 (0 - just a single macroblock) reg last_row;
assign buf_rd = buf_re[1:0];
assign data_out = do_r;
assign frame_en_w = frame_en && frame_go;
assign pre_frame_start=frame_en_w && (en?frame_end_w:(!frame_en_r));
assign mbl_x={mbl_x_r[6:3], left_marg[2:0]}; wire addr_run_end; // generate last cycle of address run
wire [ 6:0] tile_width_or; // set unused msb to all 1
wire [ 4:0] column_width_or;// set unused msb to all 1
wire last_in_col; // last pixel in a tile column
wire last_in_tile; // last pixel in a tile
assign buf_diff = needed_page - next_valid; reg [ 8:0] period_cntr;
assign buf_ready_w = buf_diff[2]; reg mb_pre_end_r;
reg mb_release_buf_r;
always @ (posedge xclk) begin reg pre_first_out_r;
if (!frame_en) frame_en_r <= 0;
else frame_en_r <= frame_en_w;
if (!frame_en) pre_en <= 0;
else if ( frame_end_w || !pre_en) pre_en <= frame_go;
if (!en) buf_re <= 0; assign buf_ra = bufa_r;
else buf_re <= {buf_re[CMPRS_BUF_EXTRA_LATENCY+1:0],buf_re_w}; assign tile_width_or= tile_width[1]?(tile_width[0]?0:'h40):(tile_width[0]?'h60:'h70);
assign column_width_or = tile_col_width? 0: 'h10;
assign last_in_col = &column_x;
assign last_in_tile = &tile_x;
assign addr_run_end = last_col && last_row;
assign mb_pre_end = mb_pre_end_r;
assign mb_release_buf = mb_release_buf_r;
assign buf_rd = buf_re[1:0];
assign data_out = do_r;
assign pre_first_out = pre_first_out_r;
assign data_valid = buf_re[CMPRS_BUF_EXTRA_LATENCY+2];
always @(posedge xclk) begin
if (!frame_en) buf_re[0] <= 0;
else if (mb_pre_start) buf_re[0] <= 1'b1;
else if (addr_run_end) buf_re[0] <= 1'b0;
if (!frame_en) buf_re[CMPRS_BUF_EXTRA_LATENCY+3:1] <= 0;
else buf_re[CMPRS_BUF_EXTRA_LATENCY+3:1] <= {buf_re[CMPRS_BUF_EXTRA_LATENCY+2:0]};
// Buffer data read:
if (buf_re[CMPRS_BUF_EXTRA_LATENCY+2]) do_r <= buf_di; if (buf_re[CMPRS_BUF_EXTRA_LATENCY+2]) do_r <= buf_di;
// pages read from the external memory, previous one is the last in the buffer if (!frame_en) pre_first_out_r <= 0;
if (reset_page_rd) next_valid <= 0; else pre_first_out_r <= buf_re[CMPRS_BUF_EXTRA_LATENCY+1] && ! buf_re[CMPRS_BUF_EXTRA_LATENCY+2];
else if (page_ready) next_valid <= next_valid + 1;
// at the end of each macroblock - calculate start page increment (and after delay - advance invalidate_next) if (mb_pre_start) rows_left <= mb_h_m1;
// calculate next start X in page (regardless of emd of macroblock row - selection will be at macroblock start) else if (last_col) rows_left <= rows_left - 1;
if (mb_pre_end_w) mb_pre_end <= 1;
else mb_pre_end <= mb_pre_end << 1; if (mb_pre_start || last_col) cols_left <= mb_w_m1;
if (mb_pre_end[0]) mbl_x_inc_r[7:3] <= {1'b0,mbl_x_r[6:3]} + {3'b0,mb_hper[4:3]}; else if (buf_re[0]) cols_left <= cols_left - 1;
if (mb_pre_end[1]) begin
case (tile_width)
2'b00: begin
mbl_x_next_r[6:3] <= {3'b0,mbl_x_inc_r[3]};
pre_advance_tiles[1:0] <= mbl_x_inc_r[5:4];
end
2'b01: begin
mbl_x_next_r[6:3] <= {2'b0,mbl_x_inc_r[4:3]};
pre_advance_tiles[1:0] <= mbl_x_inc_r[6:5];
end
2'b10: begin
mbl_x_next_r[6:3] <= {1'b0,mbl_x_inc_r[5:3]};
pre_advance_tiles[1:0] <= mbl_x_inc_r[7:6];
end
2'b11: begin
mbl_x_next_r[6:3] <= { mbl_x_inc_r[6:3]};
pre_advance_tiles[1:0] <= {1'b0, mbl_x_inc_r[7]};
end
endcase
used_pages <= needed_page - next_invalid +1;
end
if (mb_pre_end[2]) begin // TODO: apply after delay, regardless last or not
if (mb_last_in_row) add_invalid <= used_pages[1:0];
else add_invalid <= pre_advance_tiles;
end
// pages already processed by compressor - they can be reused for reading new tiles
if (reset_page_rd) next_invalid <= 0;
else if (mb_pre_end[3]) next_invalid <= next_invalid + {1'b0, add_invalid}; // TODO: Send next_page after delay
if (!frame_en) buf_re[CMPRS_BUF_EXTRA_LATENCY+2:1] <= 0;
// calculate before starting each macroblock (will wait if buffer is not ready) (TODO: align mb_pre_start[0] to mb_pre_end[2] - same)
//mb_pre_start_w
if (mb_pre_start_w) mb_pre_start <= 1;
else if (!mb_pre_start[3] || buf_ready_w) mb_pre_start <= mb_pre_start << 1;
if (mb_pre_start[0]) mbl_x_r[6:3] <= mb_first_in_row? {2'b0,left_marg[4:3]} : mbl_x_next_r[6:3];
if (mb_pre_start[1]) mbl_x_last_r[7:3] <= {1'b0,mbl_x_r[6:3]} + {2'b0,mb_w_m1[5:3]};
if (mb_pre_start[2]) begin
case (tile_width)
2'b00: needed_page[2:0] <= next_invalid[2:0]+{1'b0, mbl_x_last_r[5:4]};
2'b01: needed_page[2:0] <= next_invalid[2:0]+{1'b0, mbl_x_last_r[6:5]};
2'b10: needed_page[2:0] <= next_invalid[2:0]+{1'b0, mbl_x_last_r[7:6]};
2'b11: needed_page[2:0] <= next_invalid[2:0]+{2'b0, mbl_x_last_r[7]};
endcase
end
//need_pages_m1[1:0] <= ;
if (buf_re[0]) last_col <= 0;
else last_col <= (cols_left == 1);
/* if (buf_re[0]) last_row <= 0;
if (mb_pre_end_w) mb_pre_end <= 1; else if (last_col) last_row <= (rows_left == 1);
else mb_pre_end <= mb_pre_end << 1;
reg [ 3:0] mb_pre_start; // 1-hot macroblock pre start calcualtions - TODO: adjust width
wire buf_ready_w; // External memory buffer has all the pages needed
reg mb_first_in_row;
reg [ 2:0] next_valid; // number of next valid page (only 2 LSB are actual page number)
reg [ 2:0] next_invalid; // oldest valid page
reg [ 2:0] needed_page; // calculate at MB start
reg [ 7:3] mbl_x_last_r; // intermediate register for calculating needed_page first_col <= (mb_pre_start || (last_col && !last_row));
reg [ 5:0] mb_w_m1; // macroblock width minus 1 if (mb_pre_start) row_sa <= {start_page,3'b0,macroblock_x};
else if (first_col) row_sa <= row_sa + (tile_col_width ? 12'h20:12'h10);
wire [ 6:0] mbl_x; // macroblock left pixel x relative to a tile (page) Maximal page - 128 bytes wide
reg [ 6:3] mbl_x_r; // macroblock left pixel x relative to a tile (page) (3 low don't change)
reg [ 6:3] mbl_x_next_r; // macroblock left pixel x relative to a tile (page), not valid for first column (3 low don't change)
reg [ 1:0] add_invalid; // advance next_invalid pointer by this value, send next_page pulses
reg we_invalid; // advance next_invalid pointer, send next_page pulses
reg [ 4:3] mb_hper; // macroblock horizontal period (8/16)
reg [ 4:0] mb_pre_end; // 1-hot macroblock pre end calcualtions - TODO: adjust width
wire [ 6:0] mbl_x; // macroblock left pixel x relative to a tile (page) Maximal page - 128 bytes wide
reg [ 6:3] mbl_x_r; // macroblock left pixel x relative to a tile (page) (3 low don't change)
reg [ 6:3] mbl_x_next_r; // macroblock left pixel x relative to a tile (page), not valid for first column (3 low don't change)
reg [ 7:3] mbl_x_inc_r; // intermediate register for calculatingm bl_x_next_r and add_invalid
reg [ 4:0] mb_hper; // macroblock horizontal period (8/16) if (mb_pre_start) tile_sa <= 0;
reg [ 1:0] tile_width; // memory tile width (can be 128 for monochrome JPEG) Can be 32/64/128: 0 - 16, 1 - 32, 2 - 64, 3 - 128 else if (last_col) tile_sa <= tile_sa + (tile_col_width ? 10'h20:10'h10);
if (mb_pre_start) col_inc[9:4] <= (tile_col_width ?{mb_h_m1[4:0],1'b0} : {mb_h_m1}); // valid at first column
if (mb_pre_start || last_col) column_x <= macroblock_x[4:0] | column_width_or;
else if (buf_re[0]) column_x <= (column_x + 1) | column_width_or;
reg mb_first_in_row; if (mb_pre_start || last_col) tile_x <= {2'b0,macroblock_x[4:0]} | tile_width_or;
reg mb_last_in_row; else if (buf_re[0]) tile_x <= (tile_x+1) | tile_width_or;
reg [ 1:0] add_invalid; // advance next_invalid pointer by this value, send next_page pulses
reg we_invalid; // advance next_invalid pointer, send next_page pulses
*/
if (mb_pre_start || last_col) bufa_r[11:10] <= start_page;
else if (last_in_tile) bufa_r[11:10] <= bufa_r[11:10] + 1;
// Most time critical - calculation of the buffer address
if (mb_pre_start) bufa_r[9:0] <= {3'b0,macroblock_x};
else if (last_col) bufa_r[9:0] <= row_sa[9:0];
else if (last_in_tile) bufa_r[9:0] <= tile_sa;
else if (buf_re[0]) bufa_r[9:0] <= bufa_r[9:0] + {last_in_col?col_inc[9:4]:6'b0,4'b1};
// now just generate delayed outputs
if (!frame_en) period_cntr <= 0;
else if (mb_pre_start) begin
case (converter_type[2:0])
3'h0: period_cntr <= PERIOD_COLOR18 - 1;
3'h1: period_cntr <= PERIOD_COLOR20 - 1;
3'h2: period_cntr <= PERIOD_MONO16 - 1;
3'h3: period_cntr <= PERIOD_JP4 - 1;
3'h4: period_cntr <= PERIOD_JP4DIFF - 1;
3'h7: period_cntr <= PERIOD_MONO8 - 1;
default: period_cntr <= 'bx;
endcase
end else if (|period_cntr) period_cntr <= period_cntr - 1;
if (!frame_en) mb_pre_end_r <= 0;
else mb_pre_end_r <= (period_cntr == (CMPRS_PREEND_EARLY+1));
if (!frame_en) mb_release_buf_r <= 0;
else mb_release_buf_r <= (period_cntr == (CMPRS_RELEASE_EARLY+1));
end
end
// synchronization between mclk and xclk clock domains
// negedge mclk -> xclk (verify clock inversion is absorbed)
pulse_cross_clock reset_page_rd_i (.rst(rst), .src_clk(~mclk),.dst_clk(xclk), .in_pulse(xfer_reset_page_rd), .out_pulse(reset_page_rd),.busy());
// mclk -> xclk
pulse_cross_clock page_ready_i (.rst(rst), .src_clk(mclk), .dst_clk(xclk), .in_pulse(page_ready_chn), .out_pulse(page_ready),.busy());
// xclk -> mclk
// pulse_cross_clock next_page_i (.rst(rst), .src_clk(xclk), .dst_clk(mclk), .in_pulse(next_page), .out_pulse(next_page_chn),.busy(busy_next_page));
dly_16 #(.WIDTH(1)) dly_16_i (.clk(xclk),.rst(rst), .dly(CMPRS_MB_DLY), .din(pre_frame_start), .dout(frame_start));
dly_16 #(.WIDTH(1)) dly_16_i (.clk(xclk),.rst(!frame_en),.dly(CMPRS_MB_DLY), .din(pre_en), .dout(en));
multipulse_cross_clock #(
.WIDTH(3),
.EXTRA_DLY(0)
) multipulse_cross_clock_i (
.rst (rst), // input
.src_clk (xclk), // input
.dst_clk (mclk), // input
.num_pulses ({1'b0,add_invalid}), // input[0:0]
.we (we_invalid), // input
.out_pulse (next_page_chn), // output
.busy () // output
);
endmodule endmodule
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment