Lephenixnoir
/
Azur
2
2
Fork 1
Code Issues Pull Requests 1 Projects Releases Wiki Activity

azrp: improve the triangle shader (speedup x1.2) 

It's quite optimized now.
This commit is contained in:
Lephenixnoir 2022-08-20 20:56:18 +02:00
parent 8448467cd4
commit 4ee1537084
Signed by: Lephenixnoir
GPG Key ID: 1BBA026E13FC0495
2 changed files with 34 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
azur/src/gint/shaders/triangle.S
View File

@ -37,42 +37,50 @@ _azrp_shader_triangle:
mov.w @r5+, _color /* color */
mulu.w r4, r0
add #2, r5
nop
mov.l r8, @-r15
sub r2, r3
add #2, r5
mov.l r9, @-r15
add #1, r3
sub r2, r3
mov.l r10, @-r15
sts macl, r0
add #1, r3
mov.l r11, @-r15
add r0, _frag
sts macl, r0
ldrs 2f
add r2, _frag
add r0, _frag
ldre 3f
add r2, _frag
mov.l r12, @-r15
sub r3, r4
add r2, _frag
mov.l r13, @-r15
sub r3, r4
mov.l r14, @-r15
mov #-31, r14
sub r3, r4
mov.l @r5, _u
mov #-31, r14
mov.l @(4, r5), _v
add #-2, _frag
mov.l @(8, r5), _w
nop
mov.l @(12, r5), _du_x
not _u, _u
mov.l @(16, r5), _dv_x
not _v, _v
mov.l @(20, r5), _dw_x
not _w, _w
/* Loop on y */
1: ldrc r3
@ -80,30 +88,24 @@ _azrp_shader_triangle:
/* Loop on x */
2: mov _u, r0
or _v, r0
and _v, r0
or _w, r0
nop
and _w, r0
mov.w @_frag+, r2
shad r14, r0
nop
not r0, r0
nop
and _frag, r0
nop
add #2, _frag
sub _du_x, _u
nop
add _du_x, _u
sub _dv_x, _v
mov.w _color, @r0
add _dv_x, _v
nop
add _dw_x, _w
sub _dw_x, _w
3: nop
/* Sloooow */
@ -113,20 +115,24 @@ _azrp_shader_triangle:
dt r1
mov.l @(28, r5), r2
add r0, _u
sub r0, _u
mov.l @(32, r5), r0
add r2, _v
sub r2, _v
nop
bf.s 1b
add r0, _w
sub r0, _w
add #-12, r5
mov #0, r0
/* 1 cycle lost */
not _u, _u
not _v, _v
not _w, _w
mov.l _u, @(12, r5)
mov.l _v, @(16, r5)
mov.l _w, @(20, r5)

4
azur/src/gint/shaders/triangle.c
View File

@ -107,8 +107,10 @@ void azrp_triangle(int x1, int y1, int x2, int y2, int x3, int y3, int color)
int min_y = max(0, min(y1, min(y2, y3)));
int max_y = min(azrp_height-1, max(y1, max(y2, y3)));
if(min_x >= max_x || min_y >= max_y)
if(min_x >= max_x || min_y >= max_y) {
prof_leave(azrp_perf_cmdgen);
return;
}
/* TODO: Have a proper way to do optimized-division by azrp_frag_height
TODO: Also account for first-fragment offset */