thejayman77/retroDE_ps2
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add FOG_ENABLE param, prune fog cone on the board profile (fix -6.165ns setup)

Browse Source 
The fog blend's tex_color*F multiply landed on the texture->color critical path
(u_texcache RAM -> u_tex|tex_color -> mult_126 -> raster_pixel_color_q, 46ns,
-6.165ns setup). Fog is a proven no-op for the current board scene (96.5% F=255),
so pay zero for it here.

FOG_ENABLE (default 1, byte-identical fog) added to gs_stub; each fog mux gated
(FOG_ENABLE && ras_fge) so FOG_ENABLE=0 constant-folds the entire fog cone away
(multiply, FOGCOL adders, s2_fog_f/persp_fog_f5 interp). Threaded through
top_psmct32_raster_demo_bram; set FOG_ENABLE(1'b0) on the active GS_SH3_LPDDR_FB
board arm (elsif at :1086, the one the QSF profile compiles). Fog stays default-on
everywhere else.

Verified: FOG_ENABLE=1 fog TBs unchanged PASS; FOG_ENABLE=0 (new tb_gs_fog_disabled
+ board scene TB) emits raw color = pre-fog datapath; board elaborates clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
thejayman77
2026-07-20 21:31:14 -04:00
parent dd75d78491
commit a72aaded22
6 changed files with 178 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rtl/gif_gs/gs_stub.sv
+13 -5
View File
@@ -450,6 +450,14 @@ module gs_stub
// out, so every existing TB / the flat production build is byte-identical
// (zero DSP/ALM/M20K cost). Only a perspective demo build sets it 1.
parameter bit PERSPECTIVE_CORRECT = 1'b0,
// GS per-vertex FOG compile gate. Default 1 = fog blend active (byte-
// identical to the fog build). 0 = every fog application collapses to the
// pre-fog color at elaboration (FOG_ENABLE is a constant), so the
// synthesizer constant-folds the entire fog cone — the texel*F/color*F
// multiplies, the FOGCOL adders/mux, AND the s2_fog_f / persp_fog_f5
// interpolation pipes that feed ONLY fog — off the texture->color critical
// path. Set 0 on board profiles whose scene fog is a functional no-op.
parameter bit FOG_ENABLE = 1'b1,
// Preserve the low four XYZ2/XYZF2 screen-coordinate bits and evaluate
// triangle coverage at the GS 12.4 pixel center. Default off retains the
// integer/corner raster contract used by every pre-Ch416 fixture.
@@ -5307,7 +5315,7 @@ module gs_stub
// consumer (s2_natural_abgr pack, the PSMCT32 emit, the Z-tested emit color)
// sees the fogged color. Gated on ras_fge → FGE=0 is byte-identical.
logic [63:0] s2_sprite_color64;
assign s2_sprite_color64 = ras_fge
assign s2_sprite_color64 = (FOG_ENABLE && ras_fge)
? {s2_sprite_color64_raw[63:32],
fog_blend_abgr(s2_sprite_color64_raw[31:0], s2_fog_f, fogcol_q)}
: s2_sprite_color64_raw;
@@ -5326,7 +5334,7 @@ module gs_stub
// GS per-vertex FOG — blend the post-texture/post-Gouraud RGB toward FOGCOL
// when FGE=1 (see the SPRITE note above). Gated → FGE=0 byte-identical.
logic [63:0] s2_tri_color64;
assign s2_tri_color64 = ras_fge
assign s2_tri_color64 = (FOG_ENABLE && ras_fge)
? {s2_tri_color64_raw[63:32],
fog_blend_abgr(s2_tri_color64_raw[31:0], s2_fog_f, fogcol_q)}
: s2_tri_color64_raw;
@@ -6982,7 +6990,7 @@ module gs_stub
// GS per-vertex FOG — fog the held bilinear-perspective texel
// with its captured +5-aligned F. Gated → FGE=0 byte-identical.
raster_pixel_color_q <= {32'd0,
ras_fge
(FOG_ENABLE && ras_fge)
? fog_blend_abgr(comb_modulate ? persp_bili_mod_color
: s1_tex_color,
persp_bili_fog_q, fogcol_q)
@@ -7076,7 +7084,7 @@ module gs_stub
// GS per-vertex FOG — fog the source RGB BEFORE the blend
// when FGE=1 (sprite fog is flat: s2_fog_f == ras_f0).
// Source alpha is NOT fogged. FGE=0 → byte-identical.
s3_cs_q <= ras_fge
s3_cs_q <= (FOG_ENABLE && ras_fge)
? {ras_color[63:32],
fog_blend_abgr(ras_color[31:0], s2_fog_f, fogcol_q)}
: ras_color;
@@ -7161,7 +7169,7 @@ module gs_stub
// F (persp_fog_f5). Gated on ras_fge → FGE=0 byte-identical
// to the pre-fog perspective emit.
raster_pixel_color_q <= {32'd0,
ras_fge
(FOG_ENABLE && ras_fge)
? fog_blend_abgr(comb_modulate ? modulated_s1_tex_color
: s1_tex_color,
persp_fog_f5, fogcol_q)
rtl/top/de25_nano_psmct32_raster_demo_top.sv
+6
View File
@@ -1108,6 +1108,12 @@ module de25_nano_psmct32_raster_demo_top (
.BIN_BUFFER_ENABLE (1'b0),
.HEARTBEAT_SPLICE_ENABLE (1'b0),
.PERSPECTIVE_CORRECT(1'b1),
// GS per-vertex FOG pruned for the board scene: 96.5% of this frame's
// vertices have F=255 (fog is a functional no-op here), so gate it OUT
// to keep the texel*F multiply off the tex-cache->raster_pixel_color
// critical path (recovers the -6.165 ns setup fail). Covers the
// FB/_Z/_640/_SCHED/_MULTI variants, which all share this instance.
.FOG_ENABLE (1'b0),
.SUBPIXEL_XY (1'b1), // Ch416 — preserve native XYZ 12.4 and sample pixel centers
.PERSP_RECIP_IDX_BITS(11),
.BILINEAR_ENABLE (1'b1),
rtl/top/top_psmct32_raster_demo_bram.sv
+5
View File
@@ -195,6 +195,10 @@ module top_psmct32_raster_demo_bram
// Default 0 generate-guards all perspective logic out (zero cost); only the
// GS_PERSP_DEMO board profile sets it 1.
parameter bit PERSPECTIVE_CORRECT = 1'b0,
// GS per-vertex FOG compile gate, forwarded to gs_stub. Default 1 (fog
// active). Board profiles whose scene fog is a functional no-op set 0 to
// prune the fog cone off the texture->color critical path.
parameter bit FOG_ENABLE = 1'b1,
parameter bit SUBPIXEL_XY = 1'b0,
parameter int PERSP_RECIP_IDX_BITS = 8, // Ch351 — perspective reciprocal LUT width (far-W -> 11)
parameter int GRAD_DIV_CYCLES = 1, // Ch352 — triangle-setup divide settle cycles (board fits -> 4)
@@ -1017,6 +1021,7 @@ module top_psmct32_raster_demo_bram
.FB_RD_REGISTERED (1'b1),
.Z_RD_REGISTERED (1'b1),
.PERSPECTIVE_CORRECT(PERSPECTIVE_CORRECT), // Ch301 — forwarded board param (default 0)
.FOG_ENABLE (FOG_ENABLE), // per-vertex fog gate; 0 prunes the fog cone
.SUBPIXEL_XY (SUBPIXEL_XY), // Ch416 — native GS 12.4 coverage (default 0)
.PERSP_RECIP_IDX_BITS(PERSP_RECIP_IDX_BITS), // Ch351 — perspective reciprocal LUT width (far-W -> 11)
.GRAD_DIV_CYCLES(GRAD_DIV_CYCLES), // Ch352 — triangle-setup divide settle cycles (board -> 4)