wgroeneveld
/
gba-bitmap-engine
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

move to toncs faster line rendering algorithms

This commit is contained in:
wgroeneveld 2020-07-10 20:14:47 +02:00
parent 6855ce19f2
commit 0e6f2db4ed
8 changed files with 510 additions and 94 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
engine/CMakeLists.txt
View File

@ -1,16 +1,23 @@
project(gba-bitmap-engine)
set_property(SOURCE src/gba/sin_lut.s PROPERTY LANGUAGE C)
set_property(SOURCE src/gba/tonc_bios.s PROPERTY LANGUAGE C)
set_property(SOURCE src/background/tonc_font.s PROPERTY LANGUAGE C)
set_source_files_properties(src/gba/tonc_bios.s PROPERTIES COMPILE_FLAGS "-x assembler-with-cpp")
set_source_files_properties(src/background/tonc_font.s PROPERTIES COMPILE_FLAGS "-x assembler-with-cpp")
set_property(SOURCE
src/gba/sin_lut.s
src/gba/tonc_bios.s
src/gba/tonc_memset.s
src/background/tonc_font.s PROPERTY LANGUAGE C)
set_source_files_properties(
src/gba/tonc_bios.s
src/gba/tonc_memset.s
src/background/tonc_font.s PROPERTIES COMPILE_FLAGS "-x assembler-with-cpp")
set_source_files_properties(src/gba/tonc_bmp8.cpp PROPERTIES COMPILE_FLAGS "-Wno-pointer-arith")
add_library(${PROJECT_NAME}
src/palette/palette_manager.cpp
src/palette/combined_palette.cpp
src/background/text_stream.cpp
src/gba/sin_lut.s
src/gba/tonc_memset.s
src/gba/tonc_bios.s
src/gba/tonc_bmp8.cpp
src/background/tonc_font.s
src/gba_engine.cpp
src/math.cpp

55
engine/include/libgba-sprite-engine/gba/tonc_core.h
View File

@ -148,58 +148,11 @@ INLINE void *toncset32(void *dst, u32 src, uint count);
// Fast memcpy/set
void memset16(void *dst, u16 hw, uint hwcount);
void memcpy16(void *dst, const void* src, uint hwcount);
extern "C" void memset16(void *dst, u16 hw, uint hwcount);
extern "C" void memcpy16(void *dst, const void* src, uint hwcount);
IWRAM_CODE void memset32(void *dst, u32 wd, uint wcount);
IWRAM_CODE void memcpy32(void *dst, const void* src, uint wcount);
//! Fastfill for halfwords, analogous to memset()
/*! Uses <code>memset32()</code> if \a hwcount>5
* \param dst Destination address.
* \param hw Source halfword (not address).
* \param hwcount Number of halfwords to fill.
* \note \a dst <b>must</b> be halfword aligned.
* \note \a r0 returns as \a dst + \a hwcount*2.
*/
void memset16(void *dst, u16 hw, uint hwcount);
//! \brief Copy for halfwords.
/*! Uses <code>memcpy32()</code> if \a hwn>6 and
\a src and \a dst are aligned equally.
\param dst Destination address.
\param src Source address.
\param hwcount Number of halfwords to fill.
\note \a dst and \a src <b>must</b> be halfword aligned.
\note \a r0 and \a r1 return as
\a dst + \a hwcount*2 and \a src + \a hwcount*2.
*/
void memcpy16(void *dst, const void* src, uint hwcount);
//! Fast-fill by words, analogous to memset()
/*! Like CpuFastSet(), only without the requirement of
32byte chunks and no awkward store-value-in-memory-first issue.
\param dst Destination address.
\param wd Fill word (not address).
\param wdcount Number of words to fill.
\note \a dst <b>must</b> be word aligned.
\note \a r0 returns as \a dst + \a wdcount*4.
*/
IWRAM_CODE void memset32(void *dst, u32 wd, uint wdcount);
//! \brief Fast-copy by words.
/*! Like CpuFastFill(), only without the requirement of 32byte chunks
\param dst Destination address.
\param src Source address.
\param wdcount Number of words.
\note \a src and \a dst <b>must</b> be word aligned.
\note \a r0 and \a r1 return as
\a dst + \a wdcount*4 and \a src + \a wdcount*4.
*/
IWRAM_CODE void memcpy32(void *dst, const void* src, uint wdcount);
extern "C" IWRAM_CODE void memset32(void *dst, u32 wd, uint wcount);
extern "C" IWRAM_CODE void memcpy32(void *dst, const void* src, uint wcount);
//\}

49
engine/include/libgba-sprite-engine/gba/tonc_video.h Normal file
View File

@ -0,0 +1,49 @@
//
// Created by Wouter Groeneveld on 10/07/20.
// Excerpts from tonc_video - only taken which was needed (M4)
//
#ifndef GBA_BITMAP_ENGINE_PROJECT_TONC_VIDEO_H
#define GBA_BITMAP_ENGINE_PROJECT_TONC_VIDEO_H
#include <libgba-sprite-engine/gba/tonc_types.h>
#include <libgba-sprite-engine/gba/toolbox.h>
void bmp8_plot(int x, int y, u32 clr, void *dstBase, uint dstP);
void bmp8_hline(int x1, int y, int x2, u32 clr, void *dstBase, uint dstP);
void bmp8_vline(int x, int y1, int y2, u32 clr, void *dstBase, uint dstP);
void bmp8_line(int x1, int y1, int x2, int y2, u32 clr,
void *dstBase, uint dstP);
INLINE void m4_hline(int x1, int y, int x2, u8 clrid);
INLINE void m4_vline(int x, int y1, int y2, u8 clrid);
INLINE void m4_line(int x1, int y1, int x2, int y2, u8 clrid);
INLINE void m4_plot(int x, int y, u8 clrid);
//! Plot a \a clrid pixel on the current mode 4 backbuffer
INLINE void m4_plot(int x, int y, u8 clrid)
{
u16 *dst= &vid_page[(y*M4_WIDTH+x)>>1];
if(x&1)
*dst= (*dst& 0xFF) | (clrid<<8);
else
*dst= (*dst&~0xFF) | clrid;
}
//! Draw a \a clrid colored horizontal line in mode 4.
INLINE void m4_hline(int x1, int y, int x2, u8 clrid)
{ bmp8_hline(x1, y, x2, clrid, vid_page, M4_WIDTH); }
//! Draw a \a clrid colored vertical line in mode 4.
INLINE void m4_vline(int x, int y1, int y2, u8 clrid)
{ bmp8_vline(x, y1, y2, clrid, vid_page, M4_WIDTH); }
//! Draw a \a clrid colored line in mode 4.
INLINE void m4_line(int x1, int y1, int x2, int y2, u8 clrid)
{ bmp8_line(x1, y1, x2, y2, clrid, vid_page, M4_WIDTH); }
#endif //GBA_BITMAP_ENGINE_PROJECT_TONC_VIDEO_H

18
engine/include/libgba-sprite-engine/gba/toolbox.h Normal file
View File

@ -0,0 +1,18 @@
//
// Created by Wouter Groeneveld on 10/07/20.
//
#ifndef GBA_BITMAP_ENGINE_PROJECT_TOOLBOX_H
#define GBA_BITMAP_ENGINE_PROJECT_TOOLBOX_H
#define GBA_SCREEN_WIDTH 240
#define GBA_SCREEN_WIDTH_FX GBA_SCREEN_WIDTH << 8
#define GBA_SCREEN_HEIGHT 160
#define GBA_SCREEN_HEIGHT_FX GBA_SCREEN_HEIGHT << 8
#define M4_WIDTH GBA_SCREEN_WIDTH
#define M4_HEIGHT GBA_SCREEN_HEIGHT
extern COLOR *vid_page;
#endif //GBA_BITMAP_ENGINE_PROJECT_TOOLBOX_H

9
engine/include/libgba-sprite-engine/gba_engine.h
View File

@ -15,16 +15,8 @@
#include "sound_control.h"
#include "timer.h"
#define GBA_SCREEN_WIDTH 240
#define GBA_SCREEN_WIDTH_FX GBA_SCREEN_WIDTH << 8
#define GBA_SCREEN_HEIGHT 160
#define GBA_SCREEN_HEIGHT_FX GBA_SCREEN_HEIGHT << 8
#define M4_WIDTH 240
const unsigned int black[VRAM_PAGE_SIZE] = {};
extern u16 *vid_page;
class GBAEngine {
private:
@ -74,6 +66,7 @@ public:
for(int i = 0; i < times; i++){}
}
inline void plotPixel(int x, int y, u8 clrId);
inline void plotPixel(const VectorPx &pixel, u8 clrId);
inline void plotLine(const VectorPx &point0, const VectorPx &point1, u8 clrId);
};

206
engine/src/gba/tonc_bmp8.cpp Normal file
View File

@ -0,0 +1,206 @@
//
// Created by Wouter Groeneveld on 10/07/20.
//
//! Plot a single pixel on a 8-bit buffer
#include <libgba-sprite-engine/gba/tonc_types.h>
#include <libgba-sprite-engine/gba/tonc_core.h>
#include <libgba-sprite-engine/gba/toolbox.h>
/*!
\param x X-coord.
\param y Y-coord.
\param clr Color.
\param dstBase Canvas pointer (halfword-aligned plz).
\param dstP Canvas pitch in bytes.
\note Slow as fuck. Inline plotting functionality if possible.
*/
void bmp8_plot(int x, int y, u32 clr, void *dstBase, uint dstP)
{
u16 *dstD= (u16*)(dstBase+y*dstP+(x&~1));
if(x&1)
*dstD= (*dstD& 0xFF) | (clr<<8);
else
*dstD= (*dstD&~0xFF) | (clr&0xFF);
}
//! Draw a horizontal line on an 8bit buffer
/*!
\param x1 First X-coord.
\param y Y-coord.
\param x2 Second X-coord.
\param clr Color index.
\param dstBase Canvas pointer (halfword-aligned plz).
\param dstP canvas pitch in bytes.
\note Does normalization, but not bounds checks.
*/
void bmp8_hline(int x1, int y, int x2, u32 clr, void *dstBase, uint dstP)
{
// --- Normalize ---
clr &= 0xFF;
if(x2<x1)
{ int tmp= x1; x1= x2; x2= tmp; }
uint width= x2-x1+1;
u16 *dstL= (u16*)(dstBase+y*dstP + (x1&~1));
// --- Left unaligned pixel ---
if(x1&1)
{
*dstL= (*dstL & 0xFF) + (clr<<8);
width--;
dstL++;
}
// --- Right unaligned pixel ---
if(width&1)
dstL[width/2]= (dstL[width/2]&~0xFF) + clr;
width /= 2;
// --- Aligned line ---
if(width)
memset16(dstL, dup8(clr), width);
}
//! Draw a vertical line on an 8bit buffer
/*!
\param x X-coord.
\param y1 First Y-coord.
\param y2 Second Y-coord.
\param clr Color index.
\param dstBase Canvas pointer (halfword-aligned plz).
\param dstP canvas pitch in bytes.
\note Does normalization, but not bounds checks.
*/
void bmp8_vline(int x, int y1, int y2, u32 clr, void *dstBase, uint dstP)
{
// --- Normalize ---
if(y2<y1)
{ int tmp= y1; y1= y2; y2= tmp; }
uint height= y2-y1+1;
u16 *dstL= (u16*)(dstBase+y1*dstP + (x&~1));
dstP /= 2;
if(x&1)
{
clr <<= 8;
while(height--)
{ *dstL= (*dstL& 0xFF) + clr; dstL += dstP; }
}
else
{
clr &= 0xFF;
while(height--)
{ *dstL= (*dstL&~0xFF) + clr; dstL += dstP; }
}
}
//! Draw a line on an 8bit buffer
/*!
\param x1 First X-coord.
\param y1 First Y-coord.
\param x2 Second X-coord.
\param y2 Second Y-coord.
\param clr Color index.
\param dstBase Canvas pointer (halfword-aligned plz).
\param dstP Canvas pitch in bytes.
\note Does normalization, but not bounds checks.
*/
void bmp8_line(int x1, int y1, int x2, int y2, u32 clr,
void *dstBase, uint dstP)
{
// Trivial lines: horz and vertical
if(y1 == y2) // Horizontal
{
// Adjust for inclusive ends
if(x2 == x1)
{ bmp8_plot(x1, y1, clr, dstBase, dstP); return; }
bmp8_hline(x1, y1, x2, clr, dstBase, dstP);
return;
}
else if(x1 == x2) // Vertical
{
// Adjust for inclusive ends
if(y2 == y1)
{ bmp8_plot(x1, y1, clr, dstBase, dstP); return; }
bmp8_vline(x1, y1, y2, clr, dstBase, dstP);
return;
}
// --- Slogging through the diagonal ---
int ii, dx, dy, xstep, ystep, dd;
u32 addr= (u32)(dstBase + y1*dstP + x1), mask= 255;
u16 *dstL;
clr &= mask;
clr |= clr<<8;
if(x1 & 1)
mask= ~mask;
// --- Normalization ---
if(x1>x2)
{ xstep= -1; dx= x1-x2; }
else
{ xstep= +1; dx= x2-x1; }
if(y1>y2)
{ ystep= -dstP; dy= y1-y2; }
else
{ ystep= +dstP; dy= y2-y1; }
// --- Drawing ---
// NOTE: because xstep is alternating, you can do marvels
// with mask-flips
// NOTE: (mask>>31) is equivalent to (x&1) ? 0 : 1
if(dx>=dy) // Diagonal, slope <= 1
{
dd= 2*dy - dx;
for(ii=dx; ii>=0; ii--)
{
dstL= (u16*)(addr - (mask>>31));
*dstL= (*dstL &~ mask) | (clr & mask);
if(dd >= 0)
{ dd -= 2*dx; addr += ystep; }
dd += 2*dy;
addr += xstep;
mask = ~mask;
}
}
else // # Diagonal, slope > 1
{
dd= 2*dx - dy;
for(ii=dy; ii>=0; ii--)
{
dstL= (u16*)(addr - (mask>>31));
*dstL= (*dstL &~ mask) | (clr & mask);
if(dd >= 0)
{
dd -= 2*dy;
addr += xstep;
mask = ~mask;
}
dd += 2*dx;
addr += ystep;
}
}
}
INLINE void m4_line(int x1, int y1, int x2, int y2, u8 clrid)
{ bmp8_line(x1, y1, x2, y2, clrid, vid_page, M4_WIDTH); }

216
engine/src/gba/tonc_memset.s Normal file
View File

@ -0,0 +1,216 @@
//
// Alignment-safe and fast memset routines
//
//! \file tonc_memcpy.s
//! \author J Vijn
//! \date 20060508 - 20090801
//
// === NOTES ===
@ * 20050924: Lower overhead for all; reduced i-count for u16 loops.
@ * These are 16/32bit memset and memcpy. The 32bit versions are in
@ iwram for maximum effect and pretty much do what CpuFastSet does,
@ except that it'll work for non multiples of 8 words too. Speed
@ is as good as CpuFastSet, but with a little less overhead.
@ * The 16bit versions call the 32bit ones if possible and/or desirable.
@ They are thumb/ROM functions but did them in asm anyway because
@ GCC goes haywire with the use of registers resulting in a much
@ higher overhead (i.e., detrimental for low counts)
@ * Crossover with inline while(nn--) loops (not for(ii++), which are
@ much slower):
@ memset32: ~5
@ memset16: ~8
.file "tonc_memset.s"
#define DEF_SIZE(_name) .size _name, .-_name
//! \name Section definitions for assembly.
//\{
#define CSEC_TEXT .text //!< Standard code section directive.
#define CSEC_EWRAM .section .ewram , "ax", %progbits //!< EWRAM code section directive.
#define CSEC_IWRAM .section .iwram, "ax", %progbits //!< IWRAM code section directive.
#define DSEC_DATA .data //<! Standard data section directive.
#define DSEC_ROM .section .rodata //!< ROM data section directive.
#define DSEC_BSS .section .bss //!< Uninited data (RAM) section directive.
#define DSEC_SBSS .section .sbss //!< Uninited data (DTCM?) section directive.
#define ARM_FUNC .arm //!< Indicates an ARM function.
#define THUMB_FUNC .thumb_func //!< Indicates a Thumb function.
//# NOTE: because these use commas, I can't pass them through CPP macros.
//# Yes, this is stupid, but do you have a better idea?
#undef CSEC_EWRAM
.macro CSEC_EWRAM
.section .ewram , "ax", %progbits
.endm
#undef CSEC_IWRAM
.macro CSEC_IWRAM
.section .iwram , "ax", %progbits
.endm
//\}
//! \name Function definition macros.
//\{
//! Start an assembly function.
/*!
\param _name Name of function.
\param _section Section to place function in (like .text)
*/
#define BEGIN_FUNC(_name, _section, _iset) \
_section; \
_iset; \
.align 2; \
.global _name; \
.type _name STT_FUNC; \
_name:
//! End of a function.
#define END_FUNC(_name) DEF_SIZE(_name)
//! Begin an ARM function
/*!
\param _name Name of function.
\param _section Section to place function in (like .text)
*/
#define BEGIN_FUNC_ARM(_name, _section) BEGIN_FUNC(_name, _section, ARM_FUNC)
//! Begin a THUMB function.
/*!
\param _name Name of function.
\param _section Section to place function in (like .text)
*/
#define BEGIN_FUNC_THUMB(_name, _section) BEGIN_FUNC(_name, _section, THUMB_FUNC)
//\}
//! \name Data definition macros.
//\{
#define BEGIN_SYMBOL(_name, _section) \
_section; \
.align; \
.global _name; \
_name:
#define END_SYMBOL(_name) DEF_SIZE(_name)
//\}
// --------------------------------------------------------------------
// CONSTANTS
// --------------------------------------------------------------------
//! \name TSurface member offsets.
//\{
#define TSRF_data 0
#define TSRF_pitch 4
#define TSRF_width 8
#define TSRF_height 10
#define TSRF_bpp 12
#define TSRF_type 13
#define TSRF_palSize 14
#define TSRF_pal 16
//\}
@ === void memset32(void *dst, u32 src, u32 wdn); =====================
/*! \fn void memset32(void *dst, u32 src, u32 wdn) IWRAM_CODE;
\brief Fast-fill by words.
\param dst Destination address.
\param src Fill word (not address).
\param wdn Number of words to fill.
\note \a dst <b>must</b> be word aligned.
\note \a r0 returns as \a dst + \a wdn.
*/
/* Reglist:
r0, r1: dst, src
r2: wdn, then wdn>>3
r3-r10: data buffer
r12: wdn&7
*/
BEGIN_FUNC_ARM(memset32, CSEC_IWRAM)
and r12, r2, #7
movs r2, r2, lsr #3
beq .Lres_set32
push {r4-r9}
@ set 32byte chunks with 8fold xxmia
mov r3, r1
mov r4, r1
mov r5, r1
mov r6, r1
mov r7, r1
mov r8, r1
mov r9, r1
.Lmain_set32:
stmia r0!, {r1, r3-r9}
subs r2, r2, #1
bhi .Lmain_set32
pop {r4-r9}
@ residual 0-7 words
.Lres_set32:
subs r12, r12, #1
stmhsia r0!, {r1}
bhi .Lres_set32
bx lr
END_FUNC(memset32)
@ === void memset16(void *dst, u16 src, u32 hwn); =====================
/*! \fn void memset16(void *dst, u16 src, u32 hwn);
\brief Fill for halfwords.
Uses <code>memset32()</code> if \a hwn>5
\param dst Destination address.
\param src Source halfword (not address).
\param wdn Number of halfwords to fill.
\note \a dst <b>must</b> be halfword aligned.
\note \a r0 returns as \a dst + \a hwn.
*/
/* Reglist:
r0, r1: dst, src
r2, r4: wdn
r3: tmp; and data buffer
*/
BEGIN_FUNC_THUMB(memset16, CSEC_TEXT)
push {r4, lr}
@ under 6 hwords -> std set
cmp r2, #5
bls .Ltail_set16
@ dst not word aligned: copy 1 hword and align
lsl r3, r0, #31
bcc .Lmain_set16
strh r1, [r0]
add r0, #2
sub r2, r2, #1
@ Again, memset32 does the real work
.Lmain_set16:
lsl r4, r1, #16
orr r1, r4
lsl r4, r2, #31
lsr r2, r2, #1
ldr r3, =memset32
bl .Llong_bl
@ NOTE: r0 is altered by memset32, but in exactly the right
@ way, so we can use is as is. r1 is now doubled though.
lsr r2, r4, #31
beq .Lend_set16
lsr r1, #16
.Ltail_set16:
sub r2, #1
bcc .Lend_set16 @ r2 was 0, bug out
lsl r2, r2, #1
.Lres_set16:
strh r1, [r0, r2]
sub r2, r2, #2
bcs .Lres_set16
.Lend_set16:
pop {r4}
pop {r3}
.Llong_bl:
bx r3
END_FUNC(memset16)
@ EOF

34
engine/src/gba_engine.cpp
View File

@ -3,6 +3,7 @@
//
#include <libgba-sprite-engine/gba/tonc_memdef.h>
#include <libgba-sprite-engine/gba/tonc_video.h>
#include <libgba-sprite-engine/gba_engine.h>
#include <cstring>
#include <libgba-sprite-engine/gba/tonc_core.h>
@ -145,42 +146,15 @@ void GBAEngine::flipPage() {
REG_DISPCNT ^= DCNT_PAGE;
}
// http://www.coranac.com/tonc/text/bitmaps.htm
// this thing is supposed to be very slow. see link above.
inline void GBAEngine::plotPixel(const VectorPx &pixel, u8 clrId) {
u16 *dst = &vid_page[(pixel.y() * M4_WIDTH + pixel.x()) / 2];
if(pixel.x() & 1) {
*dst = (*dst & 0xFF) | (clrId << 8);
} else {
*dst = (*dst & ~0xFF) | clrId;
}
m4_plot(pixel.x(), pixel.y(), clrId);
}
// more or less 1-to-1:
// https://www.davrous.com/2013/06/14/tutorial-part-2-learning-how-to-write-a-3d-soft-engine-from-scratch-in-c-ts-or-js-drawing-lines-triangles/
inline void GBAEngine::plotLine(const VectorPx &point0, const VectorPx &point1, u8 clrId) {
int x0 = point0.x();
int y0 = point0.y();
int x1 = point1.x();
int y1 = point1.y();
int dx = ABS(x1 - x0);
int dy = ABS(y1 - y0);
int sx = (x0 < x1) ? 1 : -1;
int sy = (y0 < y1) ? 1 : -1;
int err = dx - dy;
while (true) {
plotPixel(VectorPx(x0, y0), clrId);
if ((x0 == x1) && (y0 == y1)) break;
auto e2 = 2 * err;
if (e2 > -dy) { err -= dy; x0 += sx; }
if (e2 < dx) { err += dx; y0 += sy; }
}
// uses tonc's optimalization tricks to get 10 FPS extra compared to standard bline algorithms
m4_line(point0.x(), point0.y(), point1.x(), point1.y(), clrId);
}
inline VectorPx GBAEngine::project(const VectorFx &coord, const MatrixFx &transMat) {
auto point = MatrixFx::transformCoordinates(coord, transMat);