thejayman77/retroDE_ps2
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
fog-fidelity-timing
retroDE_ps2/sim/tb/gif_gs/tb_gs_fog.sv
T
thejayman77 ba74bbd5aa Snapshot: fog implementation + fidelity tooling baseline (pre bilinear-clamp fix) 
Per-vertex GS fog end-to-end (gs_stub emit incl. persp_emit5, gs_prim_list_feeder
XYZ2->XYZF2 on PRIM.FGE, gs_make_sh3_scheduler_fixture.py F/FGE packing), new fog
TBs, fidelity attribution tooling. Functional baseline before removing the dead
bilinear lerp8 clamps (Codex: 161-node comb loop -> -0.042ns setup fail).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-20 19:56:46 -04:00

446 lines
19 KiB
Systemverilog
Raw Permalink Blame History

// retroDE_ps2 — tb_gs_fog
//
// Focused white-box TB for GS per-vertex FOG in the reduced rasterizer.
//
// PS2 GS fog: final_color_channel = (C * F + FOGCOL * (255 - F)) >> 8, per
// RGB channel, where F is the 8-bit per-vertex fog coefficient (XYZF2 bits
// [63:56]) interpolated affinely across the primitive (flat for a SPRITE),
// gated on PRIM.FGE (bit 5). FOGCOL is GIF reg 0x3D (low 24 bits = 0xBBGGRR).
// Alpha is NOT fogged. When FGE=0 the emit is byte-identical to the no-fog
// path — the hard invariant.
//
// Tests:
// T1 — FGE=1 triangle, FLAT color, FLAT F. Assert the emitted RGB equals the
// EXACT hand-computed fog blend at several interior pixels, and the
// alpha is preserved unchanged.
// T2 — FGE=1 triangle, FLAT color, DISTINCT per-vertex F. Probe the white-box
// s2_fog_f at interior S2 pixels and assert it matches the barycentric
// (ground-truth) interpolated F within 1 LSB — proving F rides the
// shared affine gradient engine like the colour/Z attributes.
// T3 — FGE=0 triangle, SAME geometry/color as T1 with FOGCOL still set.
// Assert the emitted color is BYTE-IDENTICAL to the raw vertex color
// (no fog applied) — the non-negotiable invariant.
// T4 — FGE=1 SPRITE, flat color, flat F. Assert the exact fog blend on the
// flat-sprite emit path (s2_sprite_color64 chokepoint).
`timescale 1ns/1ps
module tb_gs_fog;
logic clk;
logic rst_n;
initial clk = 1'b0;
always #5 clk = ~clk;
logic gif_reg_wr_en;
logic [7:0] gif_reg_num;
logic [63:0] gif_reg_data;
logic [7:0] bg_r, bg_g, bg_b;
logic [63:0] prim_q, rgbaq_q, xyz2_q, xyzf2_q, frame_1_q, zbuf_1_q;
logic prim_complete;
logic [31:0] prim_complete_count;
logic [63:0] prim_v0_q, prim_v1_q, prim_v2_q;
logic [63:0] prim_color_q;
logic [63:0] prim_color_v0_q, prim_color_v1_q, prim_color_v2_q;
trace_pkg::vertex_t prim_v0_decoded_q, prim_v1_decoded_q, prim_v2_decoded_q;
trace_pkg::color_t prim_v0_color_decoded_q, prim_v1_color_decoded_q, prim_v2_color_decoded_q;
logic pixel_emit;
logic [31:0] pixel_emit_count;
logic [11:0] pixel_x_q, pixel_y_q;
logic [63:0] pixel_color_q;
logic [8:0] pixel_fbp_q;
logic [5:0] pixel_fbw_q, pixel_psm_q;
logic [31:0] pixel_fb_addr_q;
logic raster_pixel_emit;
logic [31:0] raster_pixel_emit_count;
logic [11:0] raster_pixel_x_q, raster_pixel_y_q;
logic [63:0] raster_pixel_color_q;
logic [31:0] raster_pixel_fb_addr_q;
logic [3:0] raster_pixel_be_q;
logic [31:0] raster_pixel_mask_q;
logic [5:0] raster_pixel_psm_q;
logic raster_active;
logic raster_overflow;
logic raster_fifo_full;
logic raster_degenerate;
logic tex_rd_en; logic [31:0] tex_rd_addr;
logic fb_rd_en; logic [31:0] fb_rd_addr;
logic z_rd_en; logic [31:0] z_rd_addr;
logic ev_valid;
trace_pkg::subsys_e ev_subsys;
trace_pkg::event_e ev_event;
logic [63:0] ev_arg0, ev_arg1, ev_arg2, ev_arg3;
logic [31:0] ev_flags;
gs_stub u_gs (
.clk(clk), .rst_n(rst_n),
.reg_wr_en(1'b0), .reg_wr_addr(16'd0), .reg_wr_data(64'd0),
.gif_reg_wr_en(gif_reg_wr_en),
.gif_reg_num(gif_reg_num),
.gif_reg_data(gif_reg_data),
.bg_r(bg_r), .bg_g(bg_g), .bg_b(bg_b),
.prim_q(prim_q), .rgbaq_q(rgbaq_q),
.xyz2_q(xyz2_q), .xyzf2_q(xyzf2_q),
.frame_1_q(frame_1_q), .zbuf_1_q(zbuf_1_q),
.prim_complete(prim_complete),
.prim_complete_count(prim_complete_count),
.prim_v0_q(prim_v0_q), .prim_v1_q(prim_v1_q), .prim_v2_q(prim_v2_q),
.prim_color_q(prim_color_q),
.prim_color_v0_q(prim_color_v0_q),
.prim_color_v1_q(prim_color_v1_q),
.prim_color_v2_q(prim_color_v2_q),
.prim_v0_decoded_q(prim_v0_decoded_q),
.prim_v1_decoded_q(prim_v1_decoded_q),
.prim_v2_decoded_q(prim_v2_decoded_q),
.prim_v0_color_decoded_q(prim_v0_color_decoded_q),
.prim_v1_color_decoded_q(prim_v1_color_decoded_q),
.prim_v2_color_decoded_q(prim_v2_color_decoded_q),
.pixel_emit(pixel_emit),
.pixel_emit_count(pixel_emit_count),
.pixel_x_q(pixel_x_q), .pixel_y_q(pixel_y_q),
.pixel_color_q(pixel_color_q),
.pixel_fbp_q(pixel_fbp_q),
.pixel_fbw_q(pixel_fbw_q),
.pixel_psm_q(pixel_psm_q),
.pixel_fb_addr_q(pixel_fb_addr_q),
.raster_pixel_emit(raster_pixel_emit),
.raster_pixel_emit_count(raster_pixel_emit_count),
.raster_pixel_x_q(raster_pixel_x_q),
.raster_pixel_y_q(raster_pixel_y_q),
.raster_pixel_color_q(raster_pixel_color_q),
.raster_pixel_fb_addr_q(raster_pixel_fb_addr_q),
.raster_pixel_be_q(raster_pixel_be_q),
.raster_pixel_mask_q(raster_pixel_mask_q),
.raster_pixel_psm_q(raster_pixel_psm_q),
.raster_active(raster_active),
.raster_overflow(raster_overflow),
.raster_fifo_full(raster_fifo_full),
.raster_degenerate(raster_degenerate),
.tex_rd_en(tex_rd_en), .tex_rd_addr(tex_rd_addr), .tex_rd_data(32'd0),
.fb_rd_en(fb_rd_en), .fb_rd_addr(fb_rd_addr), .fb_rd_data(32'd0),
.z_rd_en(z_rd_en), .z_rd_addr(z_rd_addr), .z_rd_data(32'd0),
.ev_valid(ev_valid), .ev_subsys(ev_subsys), .ev_event(ev_event),
.ev_arg0(ev_arg0), .ev_arg1(ev_arg1),
.ev_arg2(ev_arg2), .ev_arg3(ev_arg3),
.ev_flags(ev_flags)
);
// ----- Drive helpers -----
task automatic drive_reg(input logic [7:0] num, input logic [63:0] data);
@(negedge clk);
gif_reg_wr_en = 1'b1; gif_reg_num = num; gif_reg_data = data;
@(posedge clk);
endtask
task automatic drive_idle();
@(negedge clk);
gif_reg_wr_en = 1'b0; gif_reg_num = 8'd0; gif_reg_data = 64'd0;
@(posedge clk);
endtask
// XYZF2 (reg 0x04): X=[15:0] (12.4), Y=[31:16] (12.4), Z=[55:32] (24b),
// F=[63:56] (8b fog). Screen coords carry no fractional bits here.
function automatic logic [63:0] xyzf2(input int x, input int y,
input int z, input int f);
return {8'(f), 24'(z), 12'(y), 4'd0, 12'(x), 4'd0};
endfunction
function automatic logic [63:0] rgbaq(input int r, input int g, input int b, input int a);
return {32'd0, 8'(a), 8'(b), 8'(g), 8'(r)};
endfunction
localparam logic [7:0] R_PRIM = 8'h00;
localparam logic [7:0] R_RGBAQ = 8'h01;
localparam logic [7:0] R_XYZF2 = 8'h04;
localparam logic [7:0] R_FOGCOL = 8'h3D;
localparam logic [7:0] R_FRAME_1 = 8'h4C;
localparam logic [63:0] PRIM_TRI = 64'd3; // TRI
localparam logic [63:0] PRIM_TRI_FGE = 64'd3 | (64'd1 << 5); // TRI + FGE
localparam logic [63:0] PRIM_SPRITE_FGE = 64'd6 | (64'd1 << 5); // SPRITE + FGE
localparam logic [63:0] FRAME_1_VAL = 64'h0000_0000_0001_0000; // FBW=1, PSMCT32
// FOGCOL = 0xBBGGRR
localparam int FOG_R = 8'h20;
localparam int FOG_G = 8'h40;
localparam int FOG_B = 8'h60;
localparam logic [63:0] FOGCOL_VAL = {40'd0, 8'(FOG_B), 8'(FOG_G), 8'(FOG_R)};
// Reference fog blend (matches RTL fog_blend_abgr exactly).
function automatic int fog_ch(input int c, input int f, input int fc);
return (c * f + fc * (255 - f)) >> 8;
endfunction
function automatic int absdiff(input int a, input int b);
return (a > b) ? (a - b) : (b - a);
endfunction
// ----- Independent triangle reference (edge fn, fill rule, barycentric) -----
int vx0, vy0, vx1, vy1, vx2, vy2;
int vf0, vf1, vf2;
int ref_det;
int sx1, sy1, sx2, sy2, sf1, sf2;
function automatic int edge_f(input int px, input int py,
input int ax, input int ay,
input int bx, input int by);
return (px - ax) * (by - ay) - (py - ay) * (bx - ax);
endfunction
task automatic setup_ref();
int sa;
sa = (vx1 - vx0) * (vy2 - vy0) - (vy1 - vy0) * (vx2 - vx0);
if (sa < 0) begin
sx1 = vx2; sy1 = vy2; sf1 = vf2;
sx2 = vx1; sy2 = vy1; sf2 = vf1;
ref_det = -sa;
end else begin
sx1 = vx1; sy1 = vy1; sf1 = vf1;
sx2 = vx2; sy2 = vy2; sf2 = vf2;
ref_det = sa;
end
endtask
function automatic bit tol(input int ax, input int ay, input int bx, input int by);
int dx, dy;
dx = bx - ax; dy = by - ay;
return (dy > 0) || ((dy == 0) && (dx > 0));
endfunction
function automatic bit ref_inside(input int px, input int py);
int e0, e1, e2, b0, b1, b2;
e0 = edge_f(px, py, vx0, vy0, sx1, sy1);
e1 = edge_f(px, py, sx1, sy1, sx2, sy2);
e2 = edge_f(px, py, sx2, sy2, vx0, vy0);
b0 = tol(vx0, vy0, sx1, sy1) ? 0 : 1;
b1 = tol(sx1, sy1, sx2, sy2) ? 0 : 1;
b2 = tol(sx2, sy2, vx0, vy0) ? 0 : 1;
return ((e0 + b0) <= 0) && ((e1 + b1) <= 0) && ((e2 + b2) <= 0);
endfunction
function automatic int ref_attr(input int px, input int py,
input int a0, input int a1, input int a2);
int L0, L1, L2, num;
L0 = -edge_f(px, py, sx1, sy1, sx2, sy2);
L1 = -edge_f(px, py, sx2, sy2, vx0, vy0);
L2 = -edge_f(px, py, vx0, vy0, sx1, sy1);
num = L0 * a0 + L1 * a1 + L2 * a2;
if (ref_det == 0) return 0;
return num / ref_det;
endfunction
int errors;
// ----- Emit capture (full 32-bit ABGR) -----
bit covered [0:15][0:15];
logic [31:0] cap_c [0:15][0:15];
bit cap_armed;
task automatic clear_cov();
for (int y = 0; y < 16; y++)
for (int x = 0; x < 16; x++) begin
covered[y][x] = 1'b0; cap_c[y][x] = 32'd0;
end
endtask
always_ff @(posedge clk) begin
if (rst_n && cap_armed && raster_pixel_emit
&& raster_pixel_x_q < 16 && raster_pixel_y_q < 16) begin
covered[raster_pixel_y_q][raster_pixel_x_q] <= 1'b1;
cap_c [raster_pixel_y_q][raster_pixel_x_q] <= raster_pixel_color_q[31:0];
end
end
// ----- T2 F-interpolation probe: white-box s2_fog_f at interior pixels -----
int f_probe_checks;
bit f_probe_armed;
always_ff @(posedge clk) begin
if (rst_n && f_probe_armed
&& u_gs.s2_valid_q && u_gs.s2_inside_q && u_gs.ras_tri_active) begin
int px, py, ef;
px = int'(u_gs.s2_x_q);
py = int'(u_gs.s2_y_q);
if (px < 16 && py < 16 && ref_inside(px, py)) begin
ef = ref_attr(px, py, vf0, sf1, sf2);
if (absdiff(int'(u_gs.s2_fog_f), ef) > 1) begin
$error("[T2 F] (%0d,%0d) s2_fog_f=%0d expected ~%0d",
px, py, u_gs.s2_fog_f, ef);
errors = errors + 1;
end else begin
f_probe_checks = f_probe_checks + 1;
end
end
end
end
// Exact fog-blend spot check on a captured emitted pixel.
task automatic chk_fog(input int x, input int y,
input int cr, input int cg, input int cb, input int ca,
input int f);
int er, eg, eb;
er = fog_ch(cr, f, FOG_R);
eg = fog_ch(cg, f, FOG_G);
eb = fog_ch(cb, f, FOG_B);
if (!covered[y][x]) begin
$error("[fog] (%0d,%0d) expected covered but was not", x, y);
errors = errors + 1;
end else if (cap_c[y][x][7:0] !== er[7:0] ||
cap_c[y][x][15:8] !== eg[7:0] ||
cap_c[y][x][23:16] !== eb[7:0]) begin
$error("[fog] (%0d,%0d) got (%0d,%0d,%0d) expected EXACT (%0d,%0d,%0d)",
x, y, cap_c[y][x][7:0], cap_c[y][x][15:8], cap_c[y][x][23:16], er, eg, eb);
errors = errors + 1;
end else if (cap_c[y][x][31:24] !== ca[7:0]) begin
$error("[fog] (%0d,%0d) alpha fogged: got %0d expected %0d (unchanged)",
x, y, cap_c[y][x][31:24], ca);
errors = errors + 1;
end else begin
$display("[fog] (%0d,%0d) got (%0d,%0d,%0d,a=%0d) EXACT OK",
x, y, cap_c[y][x][7:0], cap_c[y][x][15:8], cap_c[y][x][23:16],
cap_c[y][x][31:24]);
end
endtask
// Exact no-fog (FGE=0 invariant): emitted RGB == raw color, alpha == raw.
task automatic chk_nofog(input int x, input int y,
input int cr, input int cg, input int cb, input int ca);
if (!covered[y][x]) begin
$error("[nofog] (%0d,%0d) expected covered but was not", x, y);
errors = errors + 1;
end else if (cap_c[y][x][7:0] !== cr[7:0] ||
cap_c[y][x][15:8] !== cg[7:0] ||
cap_c[y][x][23:16] !== cb[7:0] ||
cap_c[y][x][31:24] !== ca[7:0]) begin
$error("[nofog] (%0d,%0d) FGE=0 NOT byte-identical: got %08x expected raw (%0d,%0d,%0d,a=%0d)",
x, y, cap_c[y][x], cr, cg, cb, ca);
errors = errors + 1;
end else begin
$display("[nofog] (%0d,%0d) FGE=0 byte-identical raw color OK", x, y);
end
endtask
// Flat triangle constants.
localparam int TC_R = 8'hC0, TC_G = 8'h80, TC_B = 8'h30, TC_A = 8'hFF;
initial begin
errors = 0; f_probe_checks = 0; cap_armed = 1'b0; f_probe_armed = 1'b0;
rst_n = 1'b0; gif_reg_wr_en = 1'b0; gif_reg_num = 8'd0; gif_reg_data = 64'd0;
clear_cov();
repeat (4) @(posedge clk);
rst_n = 1'b1;
repeat (2) @(posedge clk);
// ================================================================
// T1 — FGE=1 triangle, FLAT color, FLAT F. Exact fog blend.
// ================================================================
vx0=2; vy0=1; vf0=8'h40;
vx1=13;vy1=2; vf1=8'h40;
vx2=5; vy2=7; vf2=8'h40;
setup_ref();
clear_cov(); cap_armed = 1'b1;
drive_reg(R_FOGCOL, FOGCOL_VAL);
drive_reg(R_PRIM, PRIM_TRI_FGE);
drive_reg(R_FRAME_1, FRAME_1_VAL);
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(vx0,vy0,0,vf0));
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(vx1,vy1,0,vf1));
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(vx2,vy2,0,vf2)); // closes
drive_idle();
repeat (300) @(posedge clk);
cap_armed = 1'b0;
@(posedge clk);
chk_fog(7, 3, TC_R, TC_G, TC_B, TC_A, 8'h40);
chk_fog(5, 4, TC_R, TC_G, TC_B, TC_A, 8'h40);
chk_fog(6, 2, TC_R, TC_G, TC_B, TC_A, 8'h40);
if (raster_overflow || raster_degenerate) begin
$error("[T1] raster anomaly"); errors = errors + 1;
end
// ================================================================
// T2 — FGE=1 triangle, FLAT color, DISTINCT per-vertex F. Probe the
// white-box interpolated s2_fog_f vs barycentric ground truth.
// ================================================================
vx0=2; vy0=1; vf0=8'h10;
vx1=13;vy1=2; vf1=8'hF0;
vx2=5; vy2=7; vf2=8'h80;
setup_ref();
f_probe_armed = 1'b1;
drive_reg(R_PRIM, PRIM_TRI_FGE);
drive_reg(R_FRAME_1, FRAME_1_VAL);
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(vx0,vy0,0,vf0));
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(vx1,vy1,0,vf1));
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(vx2,vy2,0,vf2)); // closes
drive_idle();
repeat (320) @(posedge clk);
f_probe_armed = 1'b0;
@(posedge clk);
if (f_probe_checks == 0) begin
$error("[T2] no interior F pixels observed"); errors = errors + 1;
end
// ================================================================
// T3 — FGE=0 triangle, SAME geometry/color as T1, FOGCOL still set.
// Emitted color MUST be byte-identical to the raw vertex color.
// ================================================================
vx0=2; vy0=1; vf0=8'h40;
vx1=13;vy1=2; vf1=8'h40;
vx2=5; vy2=7; vf2=8'h40;
setup_ref();
clear_cov(); cap_armed = 1'b1;
drive_reg(R_PRIM, PRIM_TRI); // FGE = 0
drive_reg(R_FRAME_1, FRAME_1_VAL);
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(vx0,vy0,0,vf0));
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(vx1,vy1,0,vf1));
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(vx2,vy2,0,vf2)); // closes
drive_idle();
repeat (300) @(posedge clk);
cap_armed = 1'b0;
@(posedge clk);
chk_nofog(7, 3, TC_R, TC_G, TC_B, TC_A);
chk_nofog(5, 4, TC_R, TC_G, TC_B, TC_A);
chk_nofog(6, 2, TC_R, TC_G, TC_B, TC_A);
// ================================================================
// T4 — FGE=1 SPRITE, flat color, flat F. Exact fog blend on the
// sprite emit chokepoint (s2_sprite_color64).
// ================================================================
clear_cov(); cap_armed = 1'b1;
drive_reg(R_PRIM, PRIM_SPRITE_FGE);
drive_reg(R_FRAME_1, FRAME_1_VAL);
drive_reg(R_RGBAQ, rgbaq(TC_R,TC_G,TC_B,TC_A));
drive_reg(R_XYZF2, xyzf2(1, 1, 0, 8'hA0)); // sprite origin vertex
drive_reg(R_XYZF2, xyzf2(5, 5, 0, 8'hA0)); // closing vertex -> flat F=0xA0
drive_idle();
repeat (200) @(posedge clk);
cap_armed = 1'b0;
@(posedge clk);
// Interior sprite pixels (rect [1,5)x[1,5)).
chk_fog(2, 2, TC_R, TC_G, TC_B, TC_A, 8'hA0);
chk_fog(3, 4, TC_R, TC_G, TC_B, TC_A, 8'hA0);
chk_fog(4, 1, TC_R, TC_G, TC_B, TC_A, 8'hA0);
$display("[tb_gs_fog] T2 f_checks=%0d errors=%0d", f_probe_checks, errors);
if (errors == 0) $display("[tb_gs_fog] PASS");
else $display("[tb_gs_fog] FAIL");
$finish;
end
initial begin
#5000000;
$error("[tb_gs_fog] timeout");
$finish;
end
endmodule : tb_gs_fog
Reference in New Issue View Git Blame Copy Permalink