mrq/GLdc
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Faster inversion

Browse Source
This commit is contained in:
Luke Benstead 2022-09-14 19:20:24 +01:00
parent 3caaf798e7
commit 6be3e6185d
3 changed files with 18 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
GL/platforms/sh4.c
View File

@ -47,8 +47,13 @@ void SceneListBegin(GPUList list) {
pvr_list_begin(list); pvr_list_begin(list);
} }
GL_FORCE_INLINE float _glFastInvert(float x) {
const float sgn = (x > 0) - (x < 0);
return sgn * MATH_fsrra(x * x);
}
GL_FORCE_INLINE void _glPerspectiveDivideVertex(Vertex* vertex, const float h) { GL_FORCE_INLINE void _glPerspectiveDivideVertex(Vertex* vertex, const float h) {
const float f = MATH_Fast_Invert(vertex->w); const float f = _glFastInvert(vertex->w);
/* Convert to NDC and apply viewport */ /* Convert to NDC and apply viewport */
vertex->xyz[0] = __builtin_fmaf( vertex->xyz[0] = __builtin_fmaf(
@ -65,7 +70,7 @@ GL_FORCE_INLINE void _glPerspectiveDivideVertex(Vertex* vertex, const float h) {
avoid a divide by zero. avoid a divide by zero.
*/ */
if(unlikely(vertex->w == 1.0f)) { if(unlikely(vertex->w == 1.0f)) {
vertex->xyz[2] = MATH_Fast_Invert(1.0001f + vertex->xyz[2]); vertex->xyz[2] = _glFastInvert(1.0001f + vertex->xyz[2]);
} else { } else {
vertex->xyz[2] = f; vertex->xyz[2] = f;
} }

10
GL/texture.c
View File

@ -161,9 +161,11 @@ static void GPUTextureTwiddle8PPP(void* src, void* dst, uint32_t w, uint32_t h)
for(y = 0; y < h; y += 2) { for(y = 0; y < h; y += 2) {
yout = y; yout = y;
for(x = 0; x < w; x++) { for(x = 0; x < w; x++) {
vtex[TWIDOUT((yout & mask) / 2, x & mask) + int32_t idx = TWIDOUT((yout & mask) / 2, x & mask) +
(x / min + yout / min)*min * min / 2] = (x / min + yout / min)*min * min / 2;
pixels[y * w + x] | (pixels[(y + 1) * w + x] << 8);
gl_assert(idx < (w * h));
vtex[idx] = pixels[y * w + x] | (pixels[(y + 1) * w + x] << 8);
} }
} }
} }
@ -1719,7 +1721,7 @@ GLAPI void APIENTRY glColorTableEXT(GLenum target, GLenum internalFormat, GLsize
sharedPaletteUsed = GL_TRUE; sharedPaletteUsed = GL_TRUE;
} }
for (GLbyte i = 1; i < MAX_GLDC_SHARED_PALETTES; ++i) { for (GLubyte i = 1; i < MAX_GLDC_SHARED_PALETTES; ++i) {
if (target == GL_SHARED_TEXTURE_PALETTE_0_KOS + i) { if (target == GL_SHARED_TEXTURE_PALETTE_0_KOS + i) {
palette = SHARED_PALETTES[i]; palette = SHARED_PALETTES[i];
sharedPaletteUsed = GL_TRUE; sharedPaletteUsed = GL_TRUE;

5
samples/quadmark/main.c
View File

@ -68,6 +68,7 @@ int check_start() {
} }
void setup() { void setup() {
//PVR needs to warm up for a frame, or results will be low
glKosInit(); glKosInit();
glMatrixMode(GL_MODELVIEW); glMatrixMode(GL_MODELVIEW);
glLoadIdentity(); glLoadIdentity();
@ -76,6 +77,10 @@ void setup() {
glLoadIdentity(); glLoadIdentity();
glDisable(GL_NEARZ_CLIPPING_KOS); glDisable(GL_NEARZ_CLIPPING_KOS);
pvr_wait_ready();
pvr_scene_begin();
pvr_scene_finish();
} }
void do_frame() { void do_frame() {