zhouhua/jpeg_ls_for_7v690t_ref_to_charls
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Expand JPEG-LS parameter helpers toward NEAR 255

Browse Source
This commit is contained in:
zhouhua
2026-04-16 23:27:47 +08:00
parent 96a60ff32b
commit 827d925b4a
6 changed files with 278 additions and 324 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
fpga/sim/tb_jls_coding_params.sv
View File

@@ -12,11 +12,11 @@
module tb_jls_coding_params;
// NEAR stimulus shared by all parameterized instances.
logic [5:0] near_8;
logic [5:0] near_10;
logic [5:0] near_12;
logic [5:0] near_14;
logic [5:0] near_16;
logic [7:0] near_8;
logic [7:0] near_10;
logic [7:0] near_12;
logic [7:0] near_14;
logic [7:0] near_16;
// DUT outputs.
logic [16:0] range_8;
@@ -81,11 +81,11 @@ module tb_jls_coding_params;
);
initial begin
near_8 = 6'd0;
near_10 = 6'd16;
near_12 = 6'd31;
near_14 = 6'd8;
near_16 = 6'd0;
near_8 = 8'd0;
near_10 = 8'd16;
near_12 = 8'd31;
near_14 = 8'd8;
near_16 = 8'd0;
#1;
if (range_8 !== 17'd256 || qbpp_8 !== 5'd8 || limit_8 !== 7'd32) begin
@@ -108,11 +108,11 @@ module tb_jls_coding_params;
$fatal(1, "16-bit NEAR=0 coding params mismatch");
end
near_8 = 6'd63;
near_16 = 6'd31;
near_8 = 8'd63;
near_16 = 8'd31;
#1;
if (range_8 !== 17'd6 || qbpp_8 !== 5'd3 || limit_8 !== 7'd32) begin
if (range_8 !== 17'd4 || qbpp_8 !== 5'd2 || limit_8 !== 7'd32) begin
$fatal(1, "8-bit defensive NEAR clamp mismatch");
end

28
fpga/sim/tb_jls_preset_defaults.sv
View File

@@ -12,12 +12,12 @@
module tb_jls_preset_defaults;
// Test NEAR inputs.
logic [5:0] near_8b_0;
logic [5:0] near_8b_31;
logic [5:0] near_10b_4;
logic [5:0] near_12b_1;
logic [5:0] near_14b_2;
logic [5:0] near_16b_63;
logic [7:0] near_8b_0;
logic [7:0] near_8b_31;
logic [7:0] near_10b_4;
logic [7:0] near_12b_1;
logic [7:0] near_14b_2;
logic [7:0] near_16b_63;
// Preset outputs for each instance under test.
logic [15:0] maxval_8b_0;
@@ -118,12 +118,12 @@ module tb_jls_preset_defaults;
);
initial begin
near_8b_0 = 6'd0;
near_8b_31 = 6'd31;
near_10b_4 = 6'd4;
near_12b_1 = 6'd1;
near_14b_2 = 6'd2;
near_16b_63 = 6'd63;
near_8b_0 = 8'd0;
near_8b_31 = 8'd31;
near_10b_4 = 8'd4;
near_12b_1 = 8'd1;
near_14b_2 = 8'd2;
near_16b_63 = 8'd63;
#1;
@@ -157,8 +157,8 @@ module tb_jls_preset_defaults;
$fatal(1, "14-bit NEAR=2 defaults mismatch");
end
if (maxval_16b_63 !== 16'hFFFF || t1_16b_63 !== 16'd111 ||
t2_16b_63 !== 16'd222 || t3_16b_63 !== 16'd493 ||
if (maxval_16b_63 !== 16'hFFFF || t1_16b_63 !== 16'd207 ||
t2_16b_63 !== 16'd382 || t3_16b_63 !== 16'd717 ||
reset_16b_63 !== 16'd64) begin
$fatal(1, "16-bit NEAR clamp defaults mismatch");
end

296
fpga/verilog/jls_coding_params.sv
View File

@@ -6,9 +6,10 @@
// Trace : docs/jls_traceability.md#jls-coding-parameters
// Example : PIX_WIDTH=8,NEAR=0 gives RANGE=256,qbpp=8,LIMIT=32.
//
// JPEG-LS coding parameter lookup. RANGE and qbpp depend on NEAR, but NEAR is
// limited to 0..31 in this project. A lookup table avoids a synthesized
// runtime divider and keeps this strip-level control path timing friendly.
// JPEG-LS coding parameter helper. RANGE depends on NEAR through
// RANGE = floor((MAXVAL + 2*NEAR)/(2*NEAR+1)) + 1. This is a strip-level
// control path, so the generic arithmetic is acceptable here and keeps the
// pixel pipeline free from runtime division logic.
`default_nettype none
@@ -17,7 +18,7 @@ module jls_coding_params #(
parameter int PIX_WIDTH = 16
) (
// JPEG-LS NEAR parameter for the current strip frame.
input var logic [5:0] NEAR,
input var logic [7:0] NEAR,
// JPEG-LS RANGE parameter.
output logic [16:0] RANGE,
@@ -29,221 +30,120 @@ module jls_coding_params #(
output logic [6:0] LIMIT
);
// Defensive clamp even though upstream NEAR is already limited to 31.
logic [5:0] near_clamped;
logic [16:0] range_next;
logic [4:0] qbpp_next;
logic [6:0] limit_next;
integer maximum_sample_value_int;
integer maximum_near_int;
integer near_clamped_int;
integer denominator_int;
integer range_int;
integer qbpp_int;
integer limit_int;
always_comb begin
near_clamped = NEAR;
if (NEAR > 6'd31) begin
near_clamped = 6'd31;
end
end
always_comb begin
range_next = 17'd65536;
qbpp_next = 5'd16;
limit_next = 7'd64;
maximum_sample_value_int = 65535;
limit_int = 64;
case (PIX_WIDTH)
8: begin
limit_next = 7'd32;
case (near_clamped)
6'd0: begin range_next = 17'd256; qbpp_next = 5'd8; end
6'd1: begin range_next = 17'd86; qbpp_next = 5'd7; end
6'd2: begin range_next = 17'd52; qbpp_next = 5'd6; end
6'd3: begin range_next = 17'd38; qbpp_next = 5'd6; end
6'd4: begin range_next = 17'd30; qbpp_next = 5'd5; end
6'd5: begin range_next = 17'd25; qbpp_next = 5'd5; end
6'd6: begin range_next = 17'd21; qbpp_next = 5'd5; end
6'd7: begin range_next = 17'd18; qbpp_next = 5'd5; end
6'd8: begin range_next = 17'd16; qbpp_next = 5'd4; end
6'd9: begin range_next = 17'd15; qbpp_next = 5'd4; end
6'd10: begin range_next = 17'd14; qbpp_next = 5'd4; end
6'd11: begin range_next = 17'd13; qbpp_next = 5'd4; end
6'd12: begin range_next = 17'd12; qbpp_next = 5'd4; end
6'd13: begin range_next = 17'd11; qbpp_next = 5'd4; end
6'd14: begin range_next = 17'd10; qbpp_next = 5'd4; end
6'd15: begin range_next = 17'd10; qbpp_next = 5'd4; end
6'd16: begin range_next = 17'd9; qbpp_next = 5'd4; end
6'd17: begin range_next = 17'd9; qbpp_next = 5'd4; end
6'd18: begin range_next = 17'd8; qbpp_next = 5'd3; end
6'd19: begin range_next = 17'd8; qbpp_next = 5'd3; end
6'd20: begin range_next = 17'd8; qbpp_next = 5'd3; end
6'd21: begin range_next = 17'd7; qbpp_next = 5'd3; end
6'd22: begin range_next = 17'd7; qbpp_next = 5'd3; end
6'd23: begin range_next = 17'd7; qbpp_next = 5'd3; end
6'd24: begin range_next = 17'd7; qbpp_next = 5'd3; end
6'd25: begin range_next = 17'd6; qbpp_next = 5'd3; end
6'd26: begin range_next = 17'd6; qbpp_next = 5'd3; end
6'd27: begin range_next = 17'd6; qbpp_next = 5'd3; end
6'd28: begin range_next = 17'd6; qbpp_next = 5'd3; end
6'd29: begin range_next = 17'd6; qbpp_next = 5'd3; end
6'd30: begin range_next = 17'd6; qbpp_next = 5'd3; end
default: begin range_next = 17'd6; qbpp_next = 5'd3; end
endcase
maximum_sample_value_int = 255;
limit_int = 32;
end
10: begin
limit_next = 7'd40;
case (near_clamped)
6'd0: begin range_next = 17'd1024; qbpp_next = 5'd10; end
6'd1: begin range_next = 17'd342; qbpp_next = 5'd9; end
6'd2: begin range_next = 17'd206; qbpp_next = 5'd8; end
6'd3: begin range_next = 17'd148; qbpp_next = 5'd8; end
6'd4: begin range_next = 17'd115; qbpp_next = 5'd7; end
6'd5: begin range_next = 17'd94; qbpp_next = 5'd7; end
6'd6: begin range_next = 17'd80; qbpp_next = 5'd7; end
6'd7: begin range_next = 17'd70; qbpp_next = 5'd7; end
6'd8: begin range_next = 17'd62; qbpp_next = 5'd6; end
6'd9: begin range_next = 17'd55; qbpp_next = 5'd6; end
6'd10: begin range_next = 17'd50; qbpp_next = 5'd6; end
6'd11: begin range_next = 17'd46; qbpp_next = 5'd6; end
6'd12: begin range_next = 17'd42; qbpp_next = 5'd6; end
6'd13: begin range_next = 17'd39; qbpp_next = 5'd6; end
6'd14: begin range_next = 17'd37; qbpp_next = 5'd6; end
6'd15: begin range_next = 17'd34; qbpp_next = 5'd6; end
6'd16: begin range_next = 17'd32; qbpp_next = 5'd5; end
6'd17: begin range_next = 17'd31; qbpp_next = 5'd5; end
6'd18: begin range_next = 17'd29; qbpp_next = 5'd5; end
6'd19: begin range_next = 17'd28; qbpp_next = 5'd5; end
6'd20: begin range_next = 17'd26; qbpp_next = 5'd5; end
6'd21: begin range_next = 17'd25; qbpp_next = 5'd5; end
6'd22: begin range_next = 17'd24; qbpp_next = 5'd5; end
6'd23: begin range_next = 17'd23; qbpp_next = 5'd5; end
6'd24: begin range_next = 17'd22; qbpp_next = 5'd5; end
6'd25: begin range_next = 17'd22; qbpp_next = 5'd5; end
6'd26: begin range_next = 17'd21; qbpp_next = 5'd5; end
6'd27: begin range_next = 17'd20; qbpp_next = 5'd5; end
6'd28: begin range_next = 17'd19; qbpp_next = 5'd5; end
6'd29: begin range_next = 17'd19; qbpp_next = 5'd5; end
6'd30: begin range_next = 17'd18; qbpp_next = 5'd5; end
default: begin range_next = 17'd18; qbpp_next = 5'd5; end
endcase
maximum_sample_value_int = 1023;
limit_int = 40;
end
12: begin
limit_next = 7'd48;
case (near_clamped)
6'd0: begin range_next = 17'd4096; qbpp_next = 5'd12; end
6'd1: begin range_next = 17'd1366; qbpp_next = 5'd11; end
6'd2: begin range_next = 17'd820; qbpp_next = 5'd10; end
6'd3: begin range_next = 17'd586; qbpp_next = 5'd10; end
6'd4: begin range_next = 17'd456; qbpp_next = 5'd9; end
6'd5: begin range_next = 17'd374; qbpp_next = 5'd9; end
6'd6: begin range_next = 17'd316; qbpp_next = 5'd9; end
6'd7: begin range_next = 17'd274; qbpp_next = 5'd9; end
6'd8: begin range_next = 17'd242; qbpp_next = 5'd8; end
6'd9: begin range_next = 17'd217; qbpp_next = 5'd8; end
6'd10: begin range_next = 17'd196; qbpp_next = 5'd8; end
6'd11: begin range_next = 17'd180; qbpp_next = 5'd8; end
6'd12: begin range_next = 17'd165; qbpp_next = 5'd8; end
6'd13: begin range_next = 17'd153; qbpp_next = 5'd8; end
6'd14: begin range_next = 17'd143; qbpp_next = 5'd8; end
6'd15: begin range_next = 17'd134; qbpp_next = 5'd8; end
6'd16: begin range_next = 17'd126; qbpp_next = 5'd7; end
6'd17: begin range_next = 17'd118; qbpp_next = 5'd7; end
6'd18: begin range_next = 17'd112; qbpp_next = 5'd7; end
6'd19: begin range_next = 17'd106; qbpp_next = 5'd7; end
6'd20: begin range_next = 17'd101; qbpp_next = 5'd7; end
6'd21: begin range_next = 17'd97; qbpp_next = 5'd7; end
6'd22: begin range_next = 17'd92; qbpp_next = 5'd7; end
6'd23: begin range_next = 17'd89; qbpp_next = 5'd7; end
6'd24: begin range_next = 17'd85; qbpp_next = 5'd7; end
6'd25: begin range_next = 17'd82; qbpp_next = 5'd7; end
6'd26: begin range_next = 17'd79; qbpp_next = 5'd7; end
6'd27: begin range_next = 17'd76; qbpp_next = 5'd7; end
6'd28: begin range_next = 17'd73; qbpp_next = 5'd7; end
6'd29: begin range_next = 17'd71; qbpp_next = 5'd7; end
6'd30: begin range_next = 17'd69; qbpp_next = 5'd7; end
default: begin range_next = 17'd66; qbpp_next = 5'd7; end
endcase
maximum_sample_value_int = 4095;
limit_int = 48;
end
14: begin
limit_next = 7'd56;
case (near_clamped)
6'd0: begin range_next = 17'd16384; qbpp_next = 5'd14; end
6'd1: begin range_next = 17'd5462; qbpp_next = 5'd13; end
6'd2: begin range_next = 17'd3278; qbpp_next = 5'd12; end
6'd3: begin range_next = 17'd2342; qbpp_next = 5'd12; end
6'd4: begin range_next = 17'd1822; qbpp_next = 5'd11; end
6'd5: begin range_next = 17'd1491; qbpp_next = 5'd11; end
6'd6: begin range_next = 17'd1262; qbpp_next = 5'd11; end
6'd7: begin range_next = 17'd1094; qbpp_next = 5'd11; end
6'd8: begin range_next = 17'd965; qbpp_next = 5'd10; end
6'd9: begin range_next = 17'd864; qbpp_next = 5'd10; end
6'd10: begin range_next = 17'd782; qbpp_next = 5'd10; end
6'd11: begin range_next = 17'd714; qbpp_next = 5'd10; end
6'd12: begin range_next = 17'd657; qbpp_next = 5'd10; end
6'd13: begin range_next = 17'd608; qbpp_next = 5'd10; end
6'd14: begin range_next = 17'd566; qbpp_next = 5'd10; end
6'd15: begin range_next = 17'd530; qbpp_next = 5'd10; end
6'd16: begin range_next = 17'd498; qbpp_next = 5'd9; end
6'd17: begin range_next = 17'd470; qbpp_next = 5'd9; end
6'd18: begin range_next = 17'd444; qbpp_next = 5'd9; end
6'd19: begin range_next = 17'd422; qbpp_next = 5'd9; end
6'd20: begin range_next = 17'd401; qbpp_next = 5'd9; end
6'd21: begin range_next = 17'd382; qbpp_next = 5'd9; end
6'd22: begin range_next = 17'd366; qbpp_next = 5'd9; end
6'd23: begin range_next = 17'd350; qbpp_next = 5'd9; end
6'd24: begin range_next = 17'd336; qbpp_next = 5'd9; end
6'd25: begin range_next = 17'd323; qbpp_next = 5'd9; end
6'd26: begin range_next = 17'd311; qbpp_next = 5'd9; end
6'd27: begin range_next = 17'd299; qbpp_next = 5'd9; end
6'd28: begin range_next = 17'd289; qbpp_next = 5'd9; end
6'd29: begin range_next = 17'd279; qbpp_next = 5'd9; end
6'd30: begin range_next = 17'd270; qbpp_next = 5'd9; end
default: begin range_next = 17'd262; qbpp_next = 5'd9; end
endcase
maximum_sample_value_int = 16383;
limit_int = 56;
end
default: begin
limit_next = 7'd64;
case (near_clamped)
6'd0: begin range_next = 17'd65536; qbpp_next = 5'd16; end
6'd1: begin range_next = 17'd21846; qbpp_next = 5'd15; end
6'd2: begin range_next = 17'd13108; qbpp_next = 5'd14; end
6'd3: begin range_next = 17'd9364; qbpp_next = 5'd14; end
6'd4: begin range_next = 17'd7283; qbpp_next = 5'd13; end
6'd5: begin range_next = 17'd5959; qbpp_next = 5'd13; end
6'd6: begin range_next = 17'd5043; qbpp_next = 5'd13; end
6'd7: begin range_next = 17'd4370; qbpp_next = 5'd13; end
6'd8: begin range_next = 17'd3856; qbpp_next = 5'd12; end
6'd9: begin range_next = 17'd3451; qbpp_next = 5'd12; end
6'd10: begin range_next = 17'd3122; qbpp_next = 5'd12; end
6'd11: begin range_next = 17'd2851; qbpp_next = 5'd12; end
6'd12: begin range_next = 17'd2623; qbpp_next = 5'd12; end
6'd13: begin range_next = 17'd2429; qbpp_next = 5'd12; end
6'd14: begin range_next = 17'd2261; qbpp_next = 5'd12; end
6'd15: begin range_next = 17'd2116; qbpp_next = 5'd12; end
6'd16: begin range_next = 17'd1987; qbpp_next = 5'd11; end
6'd17: begin range_next = 17'd1874; qbpp_next = 5'd11; end
6'd18: begin range_next = 17'd1773; qbpp_next = 5'd11; end
6'd19: begin range_next = 17'd1682; qbpp_next = 5'd11; end
6'd20: begin range_next = 17'd1600; qbpp_next = 5'd11; end
6'd21: begin range_next = 17'd1526; qbpp_next = 5'd11; end
6'd22: begin range_next = 17'd1458; qbpp_next = 5'd11; end
6'd23: begin range_next = 17'd1396; qbpp_next = 5'd11; end
6'd24: begin range_next = 17'd1339; qbpp_next = 5'd11; end
6'd25: begin range_next = 17'd1286; qbpp_next = 5'd11; end
6'd26: begin range_next = 17'd1238; qbpp_next = 5'd11; end
6'd27: begin range_next = 17'd1193; qbpp_next = 5'd11; end
6'd28: begin range_next = 17'd1151; qbpp_next = 5'd11; end
6'd29: begin range_next = 17'd1112; qbpp_next = 5'd11; end
6'd30: begin range_next = 17'd1076; qbpp_next = 5'd11; end
default: begin range_next = 17'd1042; qbpp_next = 5'd11; end
endcase
maximum_sample_value_int = 65535;
limit_int = 64;
end
endcase
end
always_comb begin
RANGE = range_next;
qbpp = qbpp_next;
LIMIT = limit_next;
maximum_near_int = maximum_sample_value_int / 2;
if (maximum_near_int > 255) begin
maximum_near_int = 255;
end
end
always_comb begin
near_clamped_int = NEAR;
if (near_clamped_int > maximum_near_int) begin
near_clamped_int = maximum_near_int;
end
end
always_comb begin
denominator_int = (2 * near_clamped_int) + 1;
range_int = ((maximum_sample_value_int + (2 * near_clamped_int)) / denominator_int) + 1;
end
always_comb begin
qbpp_int = 0;
if (range_int > 1) begin
qbpp_int = 1;
if (range_int > 2) begin
qbpp_int = 2;
end
if (range_int > 4) begin
qbpp_int = 3;
end
if (range_int > 8) begin
qbpp_int = 4;
end
if (range_int > 16) begin
qbpp_int = 5;
end
if (range_int > 32) begin
qbpp_int = 6;
end
if (range_int > 64) begin
qbpp_int = 7;
end
if (range_int > 128) begin
qbpp_int = 8;
end
if (range_int > 256) begin
qbpp_int = 9;
end
if (range_int > 512) begin
qbpp_int = 10;
end
if (range_int > 1024) begin
qbpp_int = 11;
end
if (range_int > 2048) begin
qbpp_int = 12;
end
if (range_int > 4096) begin
qbpp_int = 13;
end
if (range_int > 8192) begin
qbpp_int = 14;
end
if (range_int > 16384) begin
qbpp_int = 15;
end
if (range_int > 32768) begin
qbpp_int = 16;
end
end
end
always_comb begin
RANGE = range_int[16:0];
qbpp = qbpp_int[4:0];
LIMIT = limit_int[6:0];
end
endmodule

4
fpga/verilog/jls_common_pkg.sv
View File

@@ -20,8 +20,8 @@ package jls_common_pkg;
// Output FIFO byte plus original-image-start sideband bit.
localparam int JLS_OFIFO_WIDTH = 9;
// JPEG-LS NEAR is clamped to 0..31 in this IP.
localparam int JLS_NEAR_WIDTH = 6;
// JPEG-LS NEAR transport width. The standard valid range is 0..min(255,MAXVAL/2).
localparam int JLS_NEAR_WIDTH = 8;
// JPEG marker byte used by SOI/SOF55/LSE/SOS/EOI marker generation.
localparam logic [7:0] JLS_MARKER_PREFIX = 8'hFF;

51
fpga/verilog/jls_near_scale_mul.sv
View File

@@ -4,14 +4,11 @@
// Table : N/A
// Pseudocode : product = multiplicand * (2 * NEAR + 1)
// Trace : docs/jls_traceability.md#regular-error-quantization
// Example : multiplicand=5 and near_scale=5 gives 25.
// Example : multiplicand=5 and near_scale=9 gives 45.
//
// The JPEG-LS NEAR scale is always an odd 6-bit positive value in the range
// 1..63. Vivado tended to map these narrow-scale multiplies into cascaded
// DSP48E1 structures, which put PCOUT->PCIN on the top timing path. This
// helper keeps the operation in carry chains with a fixed three-adder shape:
// one partial sum for bits [2:0], one partial sum for bits [5:3], then a final
// add. The caller provides the surrounding pipeline registers.
// The JPEG-LS NEAR scale is always an odd positive value in the range 1..511.
// This helper keeps the multiply in carry-chain style by grouping the odd-scale
// bits into three 3-bit slices: [2:0], [5:3], and [8:6].
`default_nettype none
@@ -20,7 +17,7 @@ module jls_near_scale_mul #(
parameter int OUTPUT_WIDTH = 41
) (
input var logic signed [INPUT_WIDTH-1:0] multiplicand_i,
input var logic [5:0] near_scale_i,
input var logic [8:0] near_scale_i,
output logic signed [OUTPUT_WIDTH-1:0] product_o
);
@@ -32,7 +29,11 @@ module jls_near_scale_mul #(
logic signed [OUTPUT_WIDTH-1:0] multiplicand_shift_3;
logic signed [OUTPUT_WIDTH-1:0] multiplicand_shift_4;
logic signed [OUTPUT_WIDTH-1:0] multiplicand_shift_5;
logic signed [OUTPUT_WIDTH-1:0] multiplicand_shift_6;
logic signed [OUTPUT_WIDTH-1:0] multiplicand_shift_7;
logic signed [OUTPUT_WIDTH-1:0] multiplicand_shift_8;
logic signed [OUTPUT_WIDTH-1:0] partial_lo;
logic signed [OUTPUT_WIDTH-1:0] partial_mid;
logic signed [OUTPUT_WIDTH-1:0] partial_hi;
always_comb begin
@@ -42,6 +43,9 @@ module jls_near_scale_mul #(
multiplicand_shift_3 = multiplicand_ext <<< 3;
multiplicand_shift_4 = multiplicand_ext <<< 4;
multiplicand_shift_5 = multiplicand_ext <<< 5;
multiplicand_shift_6 = multiplicand_ext <<< 6;
multiplicand_shift_7 = multiplicand_ext <<< 7;
multiplicand_shift_8 = multiplicand_ext <<< 8;
end
always_comb begin
@@ -59,19 +63,32 @@ module jls_near_scale_mul #(
always_comb begin
case (near_scale_i[5:3])
3'b000: partial_hi = {OUTPUT_WIDTH{1'b0}};
3'b001: partial_hi = multiplicand_shift_3;
3'b010: partial_hi = multiplicand_shift_4;
3'b011: partial_hi = multiplicand_shift_3 + multiplicand_shift_4;
3'b100: partial_hi = multiplicand_shift_5;
3'b101: partial_hi = multiplicand_shift_3 + multiplicand_shift_5;
3'b110: partial_hi = multiplicand_shift_4 + multiplicand_shift_5;
default: partial_hi = multiplicand_shift_3 + multiplicand_shift_4 + multiplicand_shift_5;
3'b000: partial_mid = {OUTPUT_WIDTH{1'b0}};
3'b001: partial_mid = multiplicand_shift_3;
3'b010: partial_mid = multiplicand_shift_4;
3'b011: partial_mid = multiplicand_shift_3 + multiplicand_shift_4;
3'b100: partial_mid = multiplicand_shift_5;
3'b101: partial_mid = multiplicand_shift_3 + multiplicand_shift_5;
3'b110: partial_mid = multiplicand_shift_4 + multiplicand_shift_5;
default: partial_mid = multiplicand_shift_3 + multiplicand_shift_4 + multiplicand_shift_5;
endcase
end
always_comb begin
product_o = partial_lo + partial_hi;
case (near_scale_i[8:6])
3'b000: partial_hi = {OUTPUT_WIDTH{1'b0}};
3'b001: partial_hi = multiplicand_shift_6;
3'b010: partial_hi = multiplicand_shift_7;
3'b011: partial_hi = multiplicand_shift_6 + multiplicand_shift_7;
3'b100: partial_hi = multiplicand_shift_8;
3'b101: partial_hi = multiplicand_shift_6 + multiplicand_shift_8;
3'b110: partial_hi = multiplicand_shift_7 + multiplicand_shift_8;
default: partial_hi = multiplicand_shift_6 + multiplicand_shift_7 + multiplicand_shift_8;
endcase
end
always_comb begin
product_o = partial_lo + partial_mid + partial_hi;
end
endmodule

197
fpga/verilog/jls_preset_defaults.sv
View File

@@ -2,18 +2,15 @@
// Clause : Annex C.2.4.1.1 preset coding parameters
// Figure : C.3 clamping function, referenced by default threshold rules
// Table : Table C.1 valid preset parameters, Table C.2 RESET, Table C.3 defaults
// Pseudocode : Default threshold calculation for MAXVAL >= 128
// Pseudocode : Default threshold calculation with MAXVAL and NEAR
// Trace : docs/jls_traceability.md#jls-preset-defaults
// Example : PIX_WIDTH=8, NEAR=0 gives MAXVAL=255, T1=3, T2=7, T3=21.
//
// JPEG-LS default preset coding parameter helper. The first RTL version only
// supports 8/10/12/14/16-bit grayscale samples and NEAR is clamped to 0..31.
// For all supported sample precisions MAXVAL >= 128. With NEAR <= 31 the
// default thresholds do not hit MAXVAL, so the standard C.2.4.1.1 equations
// reduce to shallow shift-add expressions:
// T1 = FACTOR * 1 + 2 + 3*NEAR
// T2 = FACTOR * 4 + 3 + 5*NEAR
// T3 = FACTOR * 17 + 4 + 7*NEAR
// JPEG-LS default preset coding parameter helper. This version follows the
// full CharLS/ISO formula and supports NEAR up to min(255, MAXVAL/2). The
// work is strip-level control, not per-pixel, so one combinational divide is
// acceptable here and keeps the pixel pipeline independent of the wider NEAR
// range.
`default_nettype none
@@ -21,8 +18,8 @@ module jls_preset_defaults #(
// Compile-time grayscale sample precision. Legal values: 8, 10, 12, 14, 16.
parameter int PIX_WIDTH = 16
) (
// Requested NEAR value. Values above 31 are clamped defensively.
input var logic [5:0] near,
// Requested NEAR value. Values above min(255, MAXVAL/2) are clamped.
input var logic [7:0] near,
// JPEG-LS LSE MAXVAL preset coding parameter.
output logic [15:0] preset_maxval,
@@ -40,97 +37,137 @@ module jls_preset_defaults #(
output logic [15:0] preset_reset
);
// Default RESET value from T.87 Table C.2.
localparam logic [15:0] DEFAULT_RESET_VALUE = 16'd64;
localparam int DEFAULT_RESET_VALUE = 64;
localparam int BASIC_T1 = 3;
localparam int BASIC_T2 = 7;
localparam int BASIC_T3 = 21;
// Defensive NEAR clamp for the project maximum.
logic [5:0] near_clamped;
// Shift-add terms for 3*NEAR, 5*NEAR, and 7*NEAR.
logic [15:0] near_ext;
logic [15:0] near_times_2;
logic [15:0] near_times_3;
logic [15:0] near_times_4;
logic [15:0] near_times_5;
logic [15:0] near_times_7;
// Base threshold values after applying the standard FACTOR term.
logic [15:0] base_t1;
logic [15:0] base_t2;
logic [15:0] base_t3;
integer maximum_sample_value_int;
integer maximum_near_int;
integer near_clamped_int;
integer factor_int;
integer threshold1_candidate;
integer threshold2_candidate;
integer threshold3_candidate;
integer threshold1_int;
integer threshold2_int;
integer threshold3_int;
always_comb begin
near_clamped = near;
if (near > 6'd31) begin
near_clamped = 6'd31;
end
end
always_comb begin
near_ext = {10'd0, near_clamped};
near_times_2 = {near_ext[14:0], 1'b0};
near_times_3 = near_times_2 + near_ext;
near_times_4 = {near_ext[13:0], 2'b00};
near_times_5 = near_times_4 + near_ext;
near_times_7 = near_times_4 + near_times_2 + near_ext;
end
always_comb begin
preset_maxval = 16'hFFFF;
base_t1 = 16'd18;
base_t2 = 16'd67;
base_t3 = 16'd276;
maximum_sample_value_int = 65535;
case (PIX_WIDTH)
8: begin
preset_maxval = 16'd255;
base_t1 = 16'd3;
base_t2 = 16'd7;
base_t3 = 16'd21;
maximum_sample_value_int = 255;
end
10: begin
preset_maxval = 16'd1023;
base_t1 = 16'd6;
base_t2 = 16'd19;
base_t3 = 16'd72;
maximum_sample_value_int = 1023;
end
12: begin
preset_maxval = 16'd4095;
base_t1 = 16'd18;
base_t2 = 16'd67;
base_t3 = 16'd276;
maximum_sample_value_int = 4095;
end
14: begin
preset_maxval = 16'd16383;
base_t1 = 16'd18;
base_t2 = 16'd67;
base_t3 = 16'd276;
end
16: begin
preset_maxval = 16'hFFFF;
base_t1 = 16'd18;
base_t2 = 16'd67;
base_t3 = 16'd276;
maximum_sample_value_int = 16383;
end
default: begin
preset_maxval = 16'hFFFF;
base_t1 = 16'd18;
base_t2 = 16'd67;
base_t3 = 16'd276;
maximum_sample_value_int = 65535;
end
endcase
end
always_comb begin
preset_t1 = base_t1 + near_times_3;
preset_t2 = base_t2 + near_times_5;
preset_t3 = base_t3 + near_times_7;
preset_reset = DEFAULT_RESET_VALUE;
maximum_near_int = maximum_sample_value_int / 2;
if (maximum_near_int > 255) begin
maximum_near_int = 255;
end
end
always_comb begin
near_clamped_int = near;
if (near_clamped_int > maximum_near_int) begin
near_clamped_int = maximum_near_int;
end
end
always_comb begin
factor_int = 1;
if (maximum_sample_value_int >= 128) begin
factor_int = maximum_sample_value_int;
if (factor_int > 4095) begin
factor_int = 4095;
end
factor_int = (factor_int + 128) / 256;
threshold1_candidate = (factor_int * (BASIC_T1 - 2)) + 2 + (3 * near_clamped_int);
if ((threshold1_candidate > maximum_sample_value_int) ||
(threshold1_candidate < (near_clamped_int + 1))) begin
threshold1_int = near_clamped_int + 1;
end else begin
threshold1_int = threshold1_candidate;
end
threshold2_candidate = (factor_int * (BASIC_T2 - 3)) + 3 + (5 * near_clamped_int);
if ((threshold2_candidate > maximum_sample_value_int) ||
(threshold2_candidate < threshold1_int)) begin
threshold2_int = threshold1_int;
end else begin
threshold2_int = threshold2_candidate;
end
threshold3_candidate = (factor_int * (BASIC_T3 - 4)) + 4 + (7 * near_clamped_int);
if ((threshold3_candidate > maximum_sample_value_int) ||
(threshold3_candidate < threshold2_int)) begin
threshold3_int = threshold2_int;
end else begin
threshold3_int = threshold3_candidate;
end
end else begin
factor_int = 256 / (maximum_sample_value_int + 1);
threshold1_candidate = (BASIC_T1 / factor_int) + (3 * near_clamped_int);
if (threshold1_candidate < 2) begin
threshold1_candidate = 2;
end
if ((threshold1_candidate > maximum_sample_value_int) ||
(threshold1_candidate < (near_clamped_int + 1))) begin
threshold1_int = near_clamped_int + 1;
end else begin
threshold1_int = threshold1_candidate;
end
threshold2_candidate = (BASIC_T2 / factor_int) + (5 * near_clamped_int);
if (threshold2_candidate < 3) begin
threshold2_candidate = 3;
end
if ((threshold2_candidate > maximum_sample_value_int) ||
(threshold2_candidate < threshold1_int)) begin
threshold2_int = threshold1_int;
end else begin
threshold2_int = threshold2_candidate;
end
threshold3_candidate = (BASIC_T3 / factor_int) + (7 * near_clamped_int);
if (threshold3_candidate < 4) begin
threshold3_candidate = 4;
end
if ((threshold3_candidate > maximum_sample_value_int) ||
(threshold3_candidate < threshold2_int)) begin
threshold3_int = threshold2_int;
end else begin
threshold3_int = threshold3_candidate;
end
end
end
always_comb begin
preset_maxval = maximum_sample_value_int[15:0];
preset_t1 = threshold1_int[15:0];
preset_t2 = threshold2_int[15:0];
preset_t3 = threshold3_int[15:0];
preset_reset = DEFAULT_RESET_VALUE[15:0];
end
endmodule