etc/uxn
0
0
Fork 0
mirror of https://git.sr.ht/~rabbits/uxn synced 2024-11-04 21:35:04 +00:00
Code Issues Projects Releases Wiki Activity

go full-on aarch64 mode, with clang as the default (and only supported) compiler

Browse source
This commit is contained in:
Sigrid Solveig Haflínudóttir 2021-12-27 13:49:30 +01:00
parent 84f17059f7
commit 94985be51b
2 changed files with 32 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
build.sh
View file

@ -30,7 +30,7 @@ then
fi fi
mkdir -p bin mkdir -p bin
CC="${CC:-cc}" CC="${CC:-clang}"
CFLAGS="${CFLAGS:--std=c89 -Wall -Wno-unknown-pragmas}" CFLAGS="${CFLAGS:--std=c89 -Wall -Wno-unknown-pragmas}"
case "$(uname -s 2>/dev/null)" in case "$(uname -s 2>/dev/null)" in
MSYS_NT*|MINGW*) # MSYS2 on Windows MSYS_NT*|MINGW*) # MSYS2 on Windows

31
src/devices/ppu_aarch64.c Normal file
View file

@ -0,0 +1,31 @@
#include <arm_neon.h>
#include "ppu.h"
void
ppu_redraw(Ppu *p, Uint32 *screen)
{
uint8x16x4_t pal = vld4q_u8((Uint8*)p->palette);
Uint8 *fg = p->fg.pixels;
Uint8 *bg = p->bg.pixels;
int i;
p->fg.changed = p->bg.changed = 0;
__builtin_assume(p->width > 0 && p->height > 0);
for(i = 0; i < (p->width * p->height & ~15); i += 16, fg += 16, bg += 16, screen += 16) {
uint8x16_t fg8 = vld1q_u8(fg);
uint8x16_t bg8 = vld1q_u8(bg);
uint8x16_t px8 = vbslq_u8(vceqzq_u8(fg8), bg8, fg8);
uint8x16x4_t px = {
vqtbl1q_u8(pal.val[0], px8),
vqtbl1q_u8(pal.val[1], px8),
vqtbl1q_u8(pal.val[2], px8),
vdupq_n_u8(0xff),
};
vst4q_u8((uint8_t*)screen, px);
}
for(; i < p->width * p->height; i++)
screen[i] = p->palette[*fg ? *fg : *bg];
}