Slyvtt
/
mario-kart-casio
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

ASM implementation of the 3D track (slower for now)

This commit is contained in:
Heath Mitchell 2022-12-03 15:32:36 +00:00
parent 900837e36b
commit 189dc9b102
6 changed files with 468 additions and 77 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
CMakeLists.txt
View File

@ -2,7 +2,7 @@
# toolchain file and module path of the fxSDK
cmake_minimum_required(VERSION 3.15)
project(MarioKart)
project(MarioKart LANGUAGES C ASM)
set(SOURCES
data-headers/track.c
@ -25,6 +25,7 @@ set(SOURCES
if (DEFINED FXSDK_PLATFORM_LONG)
set(SOURCES ${SOURCES}
src/platforms/gint.c
# src/tilemap.S
)
include(GenerateG1A)
@ -115,7 +116,7 @@ if (DEFINED FXSDK_PLATFORM_LONG)
fxconv_declare_assets(${ASSETS} ${ASSETS_fx} ${ASSETS_cg} WITH_METADATA)
add_executable(mariokart ${SOURCES} ${ASSETS} ${ASSETS_${FXSDK_PLATFORM}})
target_compile_options(mariokart PRIVATE -Wall -Wextra -Werror-implicit-function-declaration -Ofast -flto -g)
target_compile_options(mariokart PRIVATE -DUSE_ASM -Wall -Wextra -Werror-implicit-function-declaration -Ofast -flto -g)
target_link_libraries(mariokart Gint::Gint)
target_link_libraries(mariokart LibProf::LibProf)

207
src/3d.c
View File

@ -12,6 +12,10 @@
#include "platforms/gint.h"
// #include "./3d-bg-dat.h"
// #ifdef PROFILING_ENABLED
// #include "libprof.h"
// #endif
#define lowResCutoff (LCD_HEIGHT_PX * 3 / 4) - 30
inline void setPixel(int x, int y, color_t color) {
@ -20,61 +24,61 @@ inline void setPixel(int x, int y, color_t color) {
int hFovModifier = 1 << 12;
void normalFov() {
int angleCos = fpcos(angle);
int angleSin = fpsin(angle);
for (unsigned short y = horizon + 2; y < /*LCD_HEIGHT_PX*/lowResCutoff; y++) {
int dist = newLut[y - horizon];
int wx = -(LCD_WIDTH_PX / 2) / 2 * dist;
for (unsigned short x = 0; x < LCD_WIDTH_PX / 2; x++) {
int x2 = wx >> 6;
int y2 = dist;
// void __attribute__ ((noinline, hot)) normalFov() {
// int angleCos = fpcos(angle);
// int angleSin = fpsin(angle);
// for (unsigned short y = horizon + 2; y < /*LCD_HEIGHT_PX*/lowResCutoff; y++) {
// int dist = newLut[y - horizon];
// int wx = -(LCD_WIDTH_PX / 2) / 2 * dist;
// for (unsigned short x = 0; x < LCD_WIDTH_PX / 2; x++) {
// int x2 = wx >> 6;
// int y2 = dist;
// Rotate by angle
int newX = ((x2 * angleCos) >> 15) + ((y2 * angleSin) >> 15);
int newY = ((y2 * angleCos) >> 15) - ((x2 * angleSin) >> 15);
// // Rotate by angle
// int newX = ((x2 * angleCos) >> 15) + ((y2 * angleSin) >> 15);
// int newY = ((y2 * angleCos) >> 15) - ((x2 * angleSin) >> 15);
color_t col = samplePixel(newX >> 1, newY >> 1);
// if (col == 0) {
// color_t* img_data = (color_t*) data_3d_bg + 4;
// col = img_data[(LCD_WIDTH_PX * (y - (horizon + 2))) + (x << 1)];
// }
// setPixel(x * 2, y, col);
// setPixel(x * 2 + 1, y, col);
// Cast to an unsigned int array so two pixels are stored at once.
((unsigned int *)VRAM)[y * (LCD_WIDTH_PX / 2) + x] = (col << 16 | col);
// color_t col = samplePixel(newX >> 1, newY >> 1);
// // if (col == 0) {
// // color_t* img_data = (color_t*) data_3d_bg + 4;
// // col = img_data[(LCD_WIDTH_PX * (y - (horizon + 2))) + (x << 1)];
// // }
// // setPixel(x * 2, y, col);
// // setPixel(x * 2 + 1, y, col);
// // Cast to an unsigned int array so two pixels are stored at once.
// ((unsigned int *)VRAM)[y * (LCD_WIDTH_PX / 2) + x] = (col << 16 | col);
wx += dist;
}
}
// Fill in the low resolution area at half vertical resolution.
// TODO: remove duplicate code
for (unsigned short y = lowResCutoff; y < LCD_HEIGHT_PX; y += 2) {
int dist = newLut[y - horizon];
int wx = -(LCD_WIDTH_PX / 2) / 2 * dist;
for (unsigned short x = 0; x < LCD_WIDTH_PX / 2; x++) {
int x2 = wx >> 6;
int y2 = dist;
// wx += dist;
// }
// }
// // Fill in the low resolution area at half vertical resolution.
// // TODO: remove duplicate code
// for (unsigned short y = lowResCutoff; y < LCD_HEIGHT_PX; y += 2) {
// int dist = newLut[y - horizon];
// int wx = -(LCD_WIDTH_PX / 2) / 2 * dist;
// for (unsigned short x = 0; x < LCD_WIDTH_PX / 2; x++) {
// int x2 = wx >> 6;
// int y2 = dist;
// Rotate by angle
int newX = ((x2 * angleCos) >> 15) + ((y2 * angleSin) >> 15);
int newY = ((y2 * angleCos) >> 15) - ((x2 * angleSin) >> 15);
// // Rotate by angle
// int newX = ((x2 * angleCos) >> 15) + ((y2 * angleSin) >> 15);
// int newY = ((y2 * angleCos) >> 15) - ((x2 * angleSin) >> 15);
color_t col = samplePixel(newX >> 1, newY >> 1);
// if (col == 0) {
// color_t* img_data = (color_t*) data_3d_bg + 4;
// col = img_data[(LCD_WIDTH_PX * (y - (horizon + 2))) + (x << 1)];
// }
// setPixel(x * 2, y, col);
// setPixel(x * 2 + 1, y, col);
// Cast to an unsigned int array so two pixels are stored at once.
((unsigned int *)VRAM)[y * (LCD_WIDTH_PX / 2) + x] = (col << 16 | col);
((unsigned int *)VRAM)[(y + 1) * (LCD_WIDTH_PX / 2) + x] = (col << 16 | col);
// color_t col = samplePixel(newX >> 1, newY >> 1);
// // if (col == 0) {
// // color_t* img_data = (color_t*) data_3d_bg + 4;
// // col = img_data[(LCD_WIDTH_PX * (y - (horizon + 2))) + (x << 1)];
// // }
// // setPixel(x * 2, y, col);
// // setPixel(x * 2 + 1, y, col);
// // Cast to an unsigned int array so two pixels are stored at once.
// ((unsigned int *)VRAM)[y * (LCD_WIDTH_PX / 2) + x] = (col << 16 | col);
// ((unsigned int *)VRAM)[(y + 1) * (LCD_WIDTH_PX / 2) + x] = (col << 16 | col);
wx += dist;
}
}
}
// wx += dist;
// }
// }
// }
// void normalFov() {
// for (unsigned short y = horizon + 2; y < LCD_HEIGHT_PX; y++) {
@ -205,36 +209,101 @@ void worldToScreenSpace(int worldX, int worldY, int* x, int* y, int* dist) {
printf("x: %d\n", *x);
}
void fullRes() {
// Original
// void __attribute__ ((noinline)) fullRes() {
// int angleCos = fpcos(angle);
// int angleSin = fpsin(angle);
// for (unsigned short y = horizon + 2; y < LCD_HEIGHT_PX; y++) {
// int dist = newLut[y - horizon];
// int wx = -(LCD_WIDTH_PX / 2) / 2 * dist;
// for (unsigned short x = 0; x < LCD_WIDTH_PX / 2; x++) {
// int x2 = wx >> 6;
// int y2 = dist;
// // Rotate by angle
// int newX = ((x2 * angleCos) >> 15) + ((y2 * angleSin) >> 15);
// int newY = ((y2 * angleCos) >> 15) - ((x2 * angleSin) >> 15);
// color_t col = samplePixel(newX >> 1, newY >> 1);
// // if (col == 0) {
// // color_t* img_data = (color_t*) data_3d_bg + 4;
// // col = img_data[(LCD_WIDTH_PX * (y - (horizon + 2))) + (x << 1)];
// // }
// // setPixel(x * 2, y, col);
// // setPixel(x * 2 + 1, y, col);
// // Cast to an unsigned int array so two pixels are stored at once.
// ((unsigned int *)VRAM)[y * (LCD_WIDTH_PX / 2) + x] = (col << 16 | col);
// wx += dist;
// }
// }
// }
// dx/dy version
// void __attribute__ ((noinline)) fullRes() {
// int angleCos = fpcos(angle);
// int angleSin = fpsin(angle);
// for (unsigned short y = horizon + 2; y < LCD_HEIGHT_PX; y++) {
// int dist = newLut[y - horizon];
// int wx = -(LCD_WIDTH_PX / 2) / 2 * dist;
// int y2 = dist;
// int newX = (wx * (angleCos >> 6)) + (y2 * angleSin);
// int newY = (y2 * angleCos) - (wx * (angleSin >> 6));
// int dx = (dist * (angleCos >> 6));
// int dy = -(dist * (angleSin >> 6));
// for (unsigned short x = 0; x < LCD_WIDTH_PX / 2; x++) {
// color_t col = samplePixel(newX >> 16, newY >> 16);
// ((unsigned int *)VRAM)[y * (LCD_WIDTH_PX / 2) + x] = (col << 16 | col);
// newX += dx;
// newY += dy;
// }
// }
// }
// TODO: Put this in fast on-chip memory?
unsigned int* vramLine;
#ifndef USE_ASM
void draw3DLine(int x, int y, int dx, int dy/*, unsigned int* vramLine*/) {
for (unsigned short x2 = 0; x2 < LCD_WIDTH_PX / 2; x2++) {
color_t col = samplePixel(x >> 16, y >> 16);
*vramLine = (col << 16 | col);
vramLine++;
x += dx;
y += dy;
}
}
#endif
void __attribute__ ((noinline)) fullRes() {
int angleCos = fpcos(angle);
int angleSin = fpsin(angle);
for (unsigned short y = horizon + 2; y < LCD_HEIGHT_PX; y++) {
int dist = newLut[y - horizon];
int wx = -(LCD_WIDTH_PX / 2) / 2 * dist;
for (unsigned short x = 0; x < LCD_WIDTH_PX / 2; x++) {
int x2 = wx >> 6;
int y2 = dist;
int y2 = dist;
// Rotate by angle
int newX = ((x2 * angleCos) >> 15) + ((y2 * angleSin) >> 15);
int newY = ((y2 * angleCos) >> 15) - ((x2 * angleSin) >> 15);
int newX = (wx * (angleCos >> 6)) + (y2 * angleSin);
int newY = (y2 * angleCos) - (wx * (angleSin >> 6));
color_t col = samplePixel(newX >> 1, newY >> 1);
// if (col == 0) {
// color_t* img_data = (color_t*) data_3d_bg + 4;
// col = img_data[(LCD_WIDTH_PX * (y - (horizon + 2))) + (x << 1)];
// }
// setPixel(x * 2, y, col);
// setPixel(x * 2 + 1, y, col);
// Cast to an unsigned int array so two pixels are stored at once.
((unsigned int *)VRAM)[y * (LCD_WIDTH_PX / 2) + x] = (col << 16 | col);
int dx = (dist * (angleCos >> 6));
int dy = -(dist * (angleSin >> 6));
wx += dist;
}
vramLine = (unsigned int *)VRAM + y * (LCD_WIDTH_PX / 2);
draw3DLine(newX, newY, dx, dy/*, vramLine*/);
}
}
void changedFov() {
#define normalFov fullRes
void __attribute__ ((noinline)) changedFov() {
int angleCos = fpcos(angle);
int angleSin = fpsin(angle);
for (unsigned short y = horizon + 2; y < /*LCD_HEIGHT_PX*/lowResCutoff; y++) {
@ -293,7 +362,7 @@ void changedFov() {
}
}
void draw3D(bool highQuality) {
void __attribute__ ((noinline)) draw3D(bool highQuality) {
if (hFovModifier == 1 << 12) {
if (highQuality) {
fullRes();

2
src/3d.h
View File

@ -9,4 +9,6 @@ void worldToScreenSpace(int worldX, int worldY, int* x, int* y, int* dist);
void draw3D(bool highQuality);
extern int hFovModifier;
void draw3DLine(int x, int y, int dx, int dy/*, unsigned int* vramLine*/);
#endif // _3D_H_

314
src/tilemap.S Normal file
View File

@ -0,0 +1,314 @@
.global _getTileID
.type _getTileID, @function
_getTileID: # (r4: xPos, r5: yPos) -> r0
mov r4, r0
or r5, r0
shlr8 r0
tst #0b11111000, r0
# T is now 0 if we want to return 0
bf .return0
# otherwise, continue
mov #-3, r1
shad r1, r4
shad r1, r5
# multiply r5 by 256
shll8 r5
add r5, r4
# read from the tilemap
mov.l .tilemap, r0
# now r0 = _tilemap symbol = address of the variable, which is an array
# shll2 r4
mov.b @(r0, r4), r0
rts
extu.b r0, r0
.return0:
rts
mov #0, r0
# .align 4
# .tilemap:
# .long _tilemap
# extern unsigned char tileset[256][64];
# extern unsigned short palette[256];
# extern unsigned char tilemap[256 * 256];
#
# unsigned short __attribute__ ((hot)) samplePixel(int xPos, int yPos) {
# xPos += xOffset;
# yPos += yOffset;
#
# // Divide by 4
# xPos = xPos >> 2;
# yPos = yPos >> 2;
#
# // Get the position of the pixel in the tile
# int xPixelInTile = xPos & (8 - 1);
# int yPixelInTile = yPos & (8 - 1);
#
# // Get the colour of the pixel in the tile
# unsigned char tileID = getTileID(xPos, yPos);
# unsigned char index = tileset[tileID][(yPixelInTile * 8) + xPixelInTile];
# return palette[index];
# }
.global _samplePixel
.type _samplePixel, @function
_samplePixel: # (r4: xPos, r5: yPos) -> r0
# TODO: When inlining this into the loop, we can make sure to only do these loads once if we have the registers available
# add xOffset to xPos
mov.l .xOffset, r1
mov.w @r1, r1
add r1, r4
# add yOffset to yPos
mov.l .yOffset, r1
mov.w @r1, r1
add r1, r5
# divide by 4
shlr2 r4
shlr2 r5
# get the position of the pixel in the tile
# copy them into other registers so we can use the original ones for the tileID
mov r4, r6
mov #0b111, r0
and r0, r6
mov r5, r7
and r0, r7
# get the colour of the pixel in the tile
# get the tile ID by calling getTileID
# Inlined version of getTileID
mov r4, r0
or r5, r0
shlr8 r0
tst #0b11111000, r0
# T is now 0 if we want to return 0
bf .return0v2
# otherwise, continue
mov #-3, r1
shad r1, r4
shad r1, r5
# multiply r5 by 256
shll8 r5
add r5, r4
# read from the tilemap
mov.l .tilemap, r0
# now r0 = _tilemap symbol = address of the variable, which is an array
# shll2 r4
mov.b @(r0, r4), r0
bra .end
extu.b r0, r0
.return0v2:
# rts
mov #0, r0
.end:
# r0 now contains the tile ID
# multiply r7 by 8
# Happens in branch delay slot
shll2 r7
add r7, r7
# add r6 to r7
add r6, r7
# multiply r0 by 64
shll8 r0
shlr2 r0
# add r0 to r7
add r0, r7
# read from the tileset
mov.l .tileset, r0
# now r0 = _tileset symbol = address of the variable, which is an array
mov.b @(r0, r7), r0
extu.b r0, r0
# read from the palette
mov.l .palette, r1
# now r1 = _palette symbol = address of the variable, which is an array
# multiply r0 by 2
add r0, r0
# read from the palette
mov.w @(r0, r1), r0
rts
extu.w r0, r0
# void draw3DLine(int x, int y, int dx, int dy, int y2,/*, unsigned int* vramLine*/) {
# for (unsigned short x2 = 0; x2 < 198; x2++) {
# color_t col = samplePixel(x >> 16, y >> 16);
# *vramLine = (col << 16 | col);
# vramLine++;
#
# x += dx;
# y += dy;
# }
# }
.global _draw3DLine
.type _draw3DLine, @function
_draw3DLine: # (r4: x, r5: y, r6: dx, r7: dy) -> r0
# vramLine is stored in a global because I don't want to deal with the stack
# Save registers
mov.l r8,@-r15
mov.l r9,@-r15
mov.l r10,@-r15
mov.l r11,@-r15
mov.l r12,@-r15
mov.l r13,@-r15
mov.l r14,@-r15
# Copy dx and dy into r13 and r14 (otherwise they get overwritten)
mov r6, r13
mov r7, r14
# Load vramLine into r8
mov.l .vramLine, r8
mov.l @r8, r8
# Use r9 as the loop counter
mov #0, r9
.loop:
# TODO: Avoid the mov by making r10/r11 the shifted values, and modify the inlined samplePixel to use r10/r11 instead of r4/r5?
# Set r4 to x >> 16
mov r4, r10
# shlr16 r4
swap.w r4, r4
exts.w r4, r4
# Set r5 to y >> 16
mov r5, r11
# shlr16 r5
swap.w r5, r5
exts.w r5, r5
# Call samplePixel
# TODO: When inlining this into the loop, we can make sure to only do these loads once if we have the registers available
# add xOffset to xPos
mov.l .xOffset, r1
mov.w @r1, r1
add r1, r4
# add yOffset to yPos
mov.l .yOffset, r1
mov.w @r1, r1
add r1, r5
# divide by 4
shlr2 r4
shlr2 r5
# get the position of the pixel in the tile
# copy them into other registers so we can use the original ones for the tileID
mov r4, r6
mov #0b111, r0
and r0, r6
mov r5, r7
and r0, r7
# get the colour of the pixel in the tile
# get the tile ID by calling getTileID
# Inlined version of getTileID
mov r4, r0
or r5, r0
shlr8 r0
tst #0b11111000, r0
# T is now 0 if we want to return 0
bf .return0v3
# otherwise, continue
mov #-3, r1
shad r1, r4
shad r1, r5
# multiply r5 by 256
shll8 r5
add r5, r4
# read from the tilemap
mov.l .tilemap, r0
# now r0 = _tilemap symbol = address of the variable, which is an array
# shll2 r4
mov.b @(r0, r4), r0
bra .endv2
extu.b r0, r0
.return0v3:
# rts
mov #0, r0
.endv2:
# r0 now contains the tile ID
# multiply r7 by 8
# Happens in branch delay slot
shll2 r7
add r7, r7
# add r6 to r7
add r6, r7
# multiply r0 by 64
shll8 r0
shlr2 r0
# add r0 to r7
add r0, r7
# read from the tileset
mov.l .tileset, r0
# now r0 = _tileset symbol = address of the variable, which is an array
mov.b @(r0, r7), r0
extu.b r0, r0
# read from the palette
mov.l .palette, r1
# now r1 = _palette symbol = address of the variable, which is an array
# multiply r0 by 2
add r0, r0
# read from the palette
mov.w @(r0, r1), r0
extu.w r0, r0
# End of samplePixel
# Set r12 to the result
mov r0, r12
# Shift r12 left by 16
shll16 r12
# OR r12 with r0
or r0, r12
# Store r12 in vramLine
mov.l r12, @r8
# Increment vramLine
add #4, r8
# Put the original values of r4 and r5 back
mov r10, r4
mov r11, r5
# Increment x and y by dx and dy
add r13, r4
add r14, r5
# Increment the loop counter
add #1, r9
# Check if we're done
# Load 198 from .halfWidth
mov.w .halfWidth, r10
cmp/eq r9, r10
bf .loop
# Restore registers
mov.l @r15+,r14
mov.l @r15+,r13
mov.l @r15+,r12
mov.l @r15+,r11
mov.l @r15+,r10
mov.l @r15+,r9
rts
mov.l @r15+,r8
.align 2
.halfWidth:
.word 198
.align 4
.vramLine:
.long _vramLine
.align 4
.xOffset:
.long _xOffset
.align 4
.yOffset:
.long _yOffset
.align 4
.tilemap:
.long _tilemap
.align 4
.tileset:
.long _tileset
.align 4
.palette:
.long _palette

12
src/tilemap.c
View File

@ -9,18 +9,20 @@
#define trackImageWidth 256 * tileSize
#define trackImageHeight 256 * tileSize
unsigned char getTileID(short xPos, short yPos) {
#ifndef USE_ASM
unsigned char __attribute__ ((hot)) getTileID(int xPos, int yPos) {
// __builtin_expect(xPos < 0 || xPos >= trackImageWidth || yPos < 0 || yPos >= trackImageHeight, 0);
if((unsigned short) xPos >= trackImageWidth || (unsigned short) yPos >= trackImageHeight) {
if ((unsigned short) xPos >= trackImageWidth || (unsigned short) yPos >= trackImageHeight) {
return 0; // Grass
} else {
// Divide by 8
int xPixel = xPos >> 3;
int yPixel = yPos >> 3;
return /*track.*/tilemap[((yPixel * (trackImageWidth / tileSize)) + xPixel)];
return /*track.*/tilemap[(yPixel * (trackImageWidth / tileSize)) + xPixel];
}
}
#endif
// unsigned char getTileID(short l0, short l1) {
// unsigned int si0, si1, si2;
@ -55,7 +57,8 @@ enum TileType getTileType(int tileID) {
return (enum TileType) track.tileTypes[tileID];
}
unsigned short samplePixel(short xPos, short yPos) {
#ifndef USE_ASM
unsigned short __attribute__ ((hot)) samplePixel(int xPos, int yPos) {
xPos += xOffset;
yPos += yOffset;
@ -72,6 +75,7 @@ unsigned short samplePixel(short xPos, short yPos) {
return /*track.*/palette[index];
// return index;
}
#endif
// unsigned short samplePixelFast(short xPos, short yPos) {
// xPos += xOffset;

5
src/tilemap.h
View File

@ -3,8 +3,9 @@
#include <stdbool.h>
unsigned char getTileID(short xPos, short yPos);
// unsigned char getTileID(short xPos, short yPos);
unsigned char getTileID(int xPos, int yPos);
enum TileType getTileType(int tileID);
unsigned short samplePixel(short xPos, short yPos);
unsigned short samplePixel(int xPos, int yPos);
#endif // _TILEMAP_H