Skip to content

Free games for graphing calculators: NumWorks, TI-Nspire CX, Casio fx-CG50 and soon TI-84 Plus CE. Balatro, Celeste, Hollow Knight, Minecraft and more.

NumBlocks CE: the pixel loops run from the archive

Mason36344b499ac011c
diff --git a/games/numblocks/ce/flash/rsf.s b/games/numblocks/ce/flash/rsf.s
new file mode 100644
index 00000000..5eb1dba6
--- /dev/null
+++ b/games/numblocks/ce/flash/rsf.s
@@ -0,0 +1,110 @@
+; The 3D view's busiest loops again, to run from the archive (appended to NBCETEX): on the CE, code read from the
+; flash's cache costs 2 or 3 cycles a byte, from RAM 4. Each at a fixed offset (rs.h: RSF_*), without an absolute
+; address of its own (relative jumps only, what it needs handed in registers): they run wherever the appvar is.
+ .assume adl=1
+ .section .text
+
+; ------------------------------------------------------------------------------------------------- 0: span_plain
+; (rs.s's span sets up: de dst, hl the texture page, bc the shade row; other set: hl u, de du, bc dv; ix v;
+; iy the end's low byte; ix pushed)
+f_span_plain:
+ ld a, (de)
+ or a, a
+ jr nz, .Lp_skip
+ exx
+ ld a, h
+ exx
+ ld l, a
+ ld a, ixh
+ and a, 0xF0
+ or a, l
+ ld l, a
+ ld c, (hl)
+ ld a, (bc)
+ ld (de), a
+.Lp_skip:
+ inc de
+ exx
+ add hl, de
+ add ix, bc
+ exx
+ ld a, e
+ cp a, (iy)
+ jr nz, f_span_plain
+ pop ix
+ ret
+
+; ------------------------------------------------------------------------------------------------- 64: span_alpha
+ .org 64
+f_span_alpha:
+ ld a, (de)
+ or a, a
+ jr nz, .La_skip
+ exx
+ ld a, h
+ exx
+ ld l, a
+ ld a, ixh
+ and a, 0xF0
+ or a, l
+ ld l, a
+ ld a, (hl)
+ or a, a
+ jr z, .La_skip
+ ld c, a
+ ld a, (bc)
+ ld (de), a
+.La_skip:
+ inc de
+ exx
+ add hl, de
+ add ix, bc
+ exx
+ ld a, e
+ cp a, (iy)
+ jr nz, f_span_alpha
+ pop ix
+ ret
+
+; ------------------------------------------------------------------------------------------------- 128: double
+; b pixels from hl doubled to de (0: c, the row's sky), hl and de after them
+ .org 128
+f_double:
+ ld a, (hl)
+ inc hl
+ or a, a
+ jr nz, .Ld_set
+ ld a, c
+.Ld_set:
+ ld (de), a
+ inc de
+ ld (de), a
+ inc de
+ ld a, (hl)
+ inc hl
+ or a, a
+ jr nz, .Ld_set2
+ ld a, c
+.Ld_set2:
+ ld (de), a
+ inc de
+ ld (de), a
+ inc de
+ djnz f_double
+ ret
+
+; ------------------------------------------------------------------------------------------------- 192: overlay
+; b pixels from hl over de where not 0 (the HUD's layer), hl and de after them
+ .org 192
+f_overlay:
+ ld a, (hl)
+ inc hl
+ or a, a
+ jr z, .Lo_skip
+ ld (de), a
+.Lo_skip:
+ inc de
+ djnz f_overlay
+ ret
+
+ .org 224
diff --git a/games/numblocks/ce/makefile b/games/numblocks/ce/makefile
index e4b0c537..db54c671 100644
--- a/games/numblocks/ce/makefile
+++ b/games/numblocks/ce/makefile
@@ -16,3 +16,12 @@ target: bin/NBCEDAT.8xv bin/NBCETEX.8xv
bin/%.8xv: gfx/%.bin
@mkdir -p bin
convbin -j bin -k 8xv -r -n $* -i $< -o $@
+# the textures' appvar carries the busiest loops too (flash/rsf.s), run from the archive
+obj/rsf.bin: flash/rsf.s
+ @mkdir -p obj
+ $(AS) -march=ez80+full $< -o obj/rsf.o
+ $(OBJCOPY) -O binary -j .text obj/rsf.o $@
+bin/NBCETEX.8xv: gfx/NBCETEX.bin obj/rsf.bin
+ @mkdir -p bin
+ cat $^ > obj/NBCETEX.bin
+ convbin -j bin -k 8xv -r -n NBCETEX -i obj/NBCETEX.bin -o $@
diff --git a/games/numblocks/ce/src/main.c b/games/numblocks/ce/src/main.c
index 7bec6658..36ca1373 100644
--- a/games/numblocks/ce/src/main.c
+++ b/games/numblocks/ce/src/main.c
@@ -267,6 +267,26 @@ int main(void) {
plat_exit();
}
#endif
+#if NB_BENCH == 9
+ {
+ /* the picture with the archive's loops and without: the same? */
+ in_game = true, gui = GUI_NONE;
+ extern const uint8_t *rs_flash;
+ const uint8_t *fl = rs_flash;
+ uint32_t ha = 0, hb = 0;
+ for (int pass = 0; pass < 2; pass++) {
+ rs_flash = pass ? 0 : fl;
+ render_stale = 1;
+ render_frame();
+ uint32_t h = 0;
+ for (unsigned i = 0; i < RW * RH; i++) h = h * 31 + rbuf[i];
+ if (pass) hb = h; else ha = h;
+ }
+ n += sprintf(out + n, "flash %p a %lx b %lx\n", (void *)fl, (unsigned long)ha, (unsigned long)hb);
+ plat_save("NBBENCH", out, (unsigned)n, 0, 0);
+ plat_exit();
+ }
+#endif
#if NB_BENCH == 7
{
/* one picture, as the computer build makes it (to compare) */
@@ -274,7 +294,7 @@ int main(void) {
render_frame();
hud_draw();
compose(true, true);
- for (uint32_t t = plat_ms(); plat_ms() - t < 3000;) plat_frame();
+ for (uint32_t t = plat_ms(); plat_ms() - t < 6000;) plat_frame();
plat_exit();
}
#endif
diff --git a/games/numblocks/ce/src/plat_ce.c b/games/numblocks/ce/src/plat_ce.c
index 51e46762..e2457af1 100644
--- a/games/numblocks/ce/src/plat_ce.c
+++ b/games/numblocks/ce/src/plat_ce.c
@@ -15,21 +15,26 @@
#include <ti/screen.h>
#include <time.h>
#include "nb.h"
+#include "rs.h"
uint8_t *scr = (uint8_t *)lcd_Ram;
static uint8_t pal_live[256 * 3];
bool vram_lost; /* the screen memory was cleared (a garbage collection): the window must be made again */
-static const uint8_t *appvar(const char *name, unsigned size) {
+static const uint8_t *appvar(const char *name, unsigned size, unsigned *got) {
uint8_t h = ti_Open(name, "r");
if (!h) return 0;
- const uint8_t *p = ti_GetSize(h) == size ? ti_GetDataPtr(h) : 0;
+ *got = ti_GetSize(h);
+ const uint8_t *p = *got >= size ? ti_GetDataPtr(h) : 0;
ti_Close(h);
return p;
}
static void gfx_reload(void) {
- gfx = appvar("NBCEDAT", GFX_SIZE);
- texdat = appvar("NBCETEX", TEXDAT_SIZE);
+ unsigned n;
+ gfx = appvar("NBCEDAT", GFX_SIZE, &n);
+ texdat = appvar("NBCETEX", TEXDAT_SIZE, &n);
+ /* (the busiest loops' copy, after the tables: run where they are, in the archive) */
+ rs_flash = texdat && n >= TEXDAT_SIZE + RSF_SIZE ? texdat + TEXDAT_SIZE : 0;
}
static void gc_before(void) { gfx_End(); }
static void gc_after(void) {
diff --git a/games/numblocks/ce/src/rs.c b/games/numblocks/ce/src/rs.c
index e4e5ddb1..75abf361 100644
--- a/games/numblocks/ce/src/rs.c
+++ b/games/numblocks/ce/src/rs.c
@@ -255,6 +255,7 @@ uint8_t RS(rs_cells)(const uint8_t *mask, uint8_t slot, uint8_t up, uint8_t down
}
#ifndef __ez80__
+const uint8_t *rs_flash;
int rs_jY[65], rs_jZ[65], rs_jV[65];
uint16_t rs_jS[128];
uint8_t rs_fog[257];
diff --git a/games/numblocks/ce/src/rs.h b/games/numblocks/ce/src/rs.h
index 9634058a..d6cb2169 100644
--- a/games/numblocks/ce/src/rs.h
+++ b/games/numblocks/ce/src/rs.h
@@ -57,6 +57,9 @@ extern const uint16_t *rs_qsqi, *rs_qsqk;
extern uint8_t rs_qsx, rs_qsz, rs_qxs0, rs_qzs0, rs_vd;
extern int rs_x, rs_z, rs_ci, rs_X0;
extern uint8_t rs_xs, rs_zs;
+/* the busiest loops' copy in the archive (flash/rsf.s, after NBCETEX's tables; 0: none), RSF_SIZE bytes */
+extern const uint8_t *rs_flash;
+#define RSF_SIZE 224
/* 2^28 / k for k 16..2560 (3 bytes each, from k = 16), and f * 256 / d for d 1..15, f < d ([d * 16 + f]): the
* tables rs_poly divides with (in NBCETEX) */
extern const uint8_t *rs_recip, *rs_ratio;
diff --git a/games/numblocks/ce/src/rs.s b/games/numblocks/ce/src/rs.s
index e324a9c7..02e5cb00 100644
--- a/games/numblocks/ce/src/rs.s
+++ b/games/numblocks/ce/src/rs.s
@@ -15,6 +15,9 @@
.equ W_DIR, 20 ; 1 or n - 1
.equ W_SIZE, 21
.equ RW, 160
+ .equ RSF_SPAN_ALPHA, 64 ; (flash/rsf.s's offsets)
+ .equ RSF_DOUBLE, 128
+ .equ RSF_OVERLAY, 192
; ------------------------------------------------------------------------------------------------- rs_setup
; void rs_setup(const uint8_t *tex, const uint8_t *cm, uint8_t alpha)
@@ -28,12 +31,25 @@ _rs_setup:
ld (r_tex), hl
ld hl, (iy+6)
ld (r_cm), hl
+ ; the loop: the archive's copy when there is one (flash/rsf.s: faster), else this one
+ ld de, (_rs_flash)
+ or a, a
+ sbc hl, hl
+ adc hl, de
+ jr z, .Lsetup_ram
+ ld a, (iy+9)
+ or a, a
+ jr z, .Lsetup_set
+ ld de, RSF_SPAN_ALPHA
+ add hl, de
+ jr .Lsetup_set
+.Lsetup_ram:
ld hl, span_plain
ld a, (iy+9)
or a, a
- jr z, .Lsetup_plain
+ jr z, .Lsetup_set
ld hl, span_alpha
-.Lsetup_plain:
+.Lsetup_set:
ld (r_loop), hl
ret
@@ -1103,6 +1119,29 @@ h_px:
.equ HUD_W, 187
_rs_expand:
push ix
+ ; the loops: the archive's when there is one (flash/rsf.s), written into the calls below
+ ld de, (_rs_flash)
+ or a, a
+ sbc hl, hl
+ adc hl, de
+ jr z, .Lex_ram
+ ld de, RSF_DOUBLE
+ add hl, de
+ ld (.Lex_c1 + 1), hl
+ ld (.Lex_c2 + 1), hl
+ ld de, RSF_OVERLAY - RSF_DOUBLE
+ add hl, de
+ ld (.Lex_c3 + 1), hl
+ ld (.Lex_c4 + 1), hl
+ jr .Lex_go
+.Lex_ram:
+ ld hl, r_double
+ ld (.Lex_c1 + 1), hl
+ ld (.Lex_c2 + 1), hl
+ ld hl, r_overlay
+ ld (.Lex_c3 + 1), hl
+ ld (.Lex_c4 + 1), hl
+.Lex_go:
ld iy, 0
add iy, sp
ld hl, (iy+15)
@@ -1126,19 +1165,9 @@ _rs_expand:
cp a, (iy)
jr c, .Lex_hudrow
ld c, (ix+0)
- ld b, RW
-.Lex_px:
- ld a, (hl)
- inc hl
- or a, a
- jr nz, .Lex_set
- ld a, c
-.Lex_set:
- ld (de), a
- inc de
- ld (de), a
- inc de
- djnz .Lex_px
+ ld b, RW / 2
+.Lex_c1:
+ call r_double
push hl
ld hl, -320
add hl, de
@@ -1158,19 +1187,9 @@ _rs_expand:
push de
ld de, r_line
ld c, (ix+0)
- ld b, RW
-.Lex_hpx:
- ld a, (hl)
- inc hl
- or a, a
- jr nz, .Lex_hset
- ld a, c
-.Lex_hset:
- ld (de), a
- inc de
- ld (de), a
- inc de
- djnz .Lex_hpx
+ ld b, RW / 2
+.Lex_c2:
+ call r_double
push hl
ld hl, r_line
ld de, r_line2
@@ -1178,8 +1197,12 @@ _rs_expand:
ldir
ld de, r_line + HUD_X
ld hl, (r_hud)
+ ld b, HUD_W
+.Lex_c3:
call r_overlay
ld de, r_line2 + HUD_X
+ ld b, HUD_W
+.Lex_c4:
call r_overlay
ld (r_hud), hl
pop hl
@@ -1191,10 +1214,33 @@ _rs_expand:
pop hl
jr .Lex_next
-; HUD_W pixels of the layer (hl) over de, 0 see-through; hl after them
+; b pairs of pixels from hl doubled to de (0: c, the row's sky); hl and de after them (flash/rsf.s has it too)
+r_double:
+ ld a, (hl)
+ inc hl
+ or a, a
+ jr nz, .Lrd_set
+ ld a, c
+.Lrd_set:
+ ld (de), a
+ inc de
+ ld (de), a
+ inc de
+ ld a, (hl)
+ inc hl
+ or a, a
+ jr nz, .Lrd_set2
+ ld a, c
+.Lrd_set2:
+ ld (de), a
+ inc de
+ ld (de), a
+ inc de
+ djnz r_double
+ ret
+
+; b pixels of the layer (hl) over de, 0 see-through; hl and de after them
r_overlay:
- ld b, HUD_W
-.Lov_px:
ld a, (hl)
inc hl
or a, a
@@ -1202,7 +1248,7 @@ r_overlay:
ld (de), a
.Lov_skip:
inc de
- djnz .Lov_px
+ djnz r_overlay
ret
@@ -2342,6 +2388,8 @@ _rs_eys: .db 0
_rs_up: .db 0
_rs_down: .db 0
_rs_jbase: .db 0
+ .global _rs_flash
+_rs_flash: .d24 0
.global _rs_qXi
.global _rs_qYi
.global _rs_qZi
diff --git a/tools/ce_emu/ce_emu.c b/tools/ce_emu/ce_emu.c
index cc556c9c..b2f7f66e 100644
--- a/tools/ce_emu/ce_emu.c
+++ b/tools/ce_emu/ce_emu.c
@@ -139,12 +139,12 @@ static uint64_t cycles_now(void) { return sched_total_cycles(); }
uint64_t emu_ms(void) { return now_ms; }
bool emu_key_held(int group, int bit) { return group >= 0 && group < 8 && bit >= 0 && bit < 8 && held[group][bit]; }
uint32_t emu_key_presses(int group, int bit) { return group >= 0 && group < 8 && bit >= 0 && bit < 8 ? presses[group][bit] : 0; }
-/* an OS routine's time: in steps of a millisecond, the scheduler's events between them (timers, the LCD, the
- * seconds), as they go on while a calculator runs a long routine such as archiving; charged at once, a second's
- * worth left events behind the second and stopped the core (sched_second's assertion) */
+/* an OS routine's time; a long one's (from 0.1 s: archiving, a garbage collection) in steps of a millisecond, the
+ * scheduler's events between them (timers, the LCD, the seconds), as they go on while a calculator runs it: charged
+ * at once, a second's worth left events behind the second and stopped the core (sched_second's assertion) */
static bool in_hook;
void emu_charge(uint32_t cycles) {
- if (!in_hook) {
+ if (!in_hook || cycles < 4800000) {
debug.cpuCycles += cycles;
return;
}