MGislv/GLideN64
1
0
Fork 0
mirror of https://github.com/blawar/GLideN64.git synced 2024-07-04 10:03:36 +00:00
Code Issues Projects Releases Wiki Activity

arm neon: add gSPLightVertex and gSPLightVertex4 neon version

Browse Source
This commit is contained in:
gizmo98 2017-05-06 11:26:38 +02:00 committed by Sergey Lipskiy
parent 542d083840
commit e52f39aebe
2 changed files with 209 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

192
src/Neon/gSPNeon.cpp
View File

@ -23,6 +23,8 @@
#include "DisplayWindow.h"
#include <arm_neon.h>
using namespace std;
void gSPTransformVertex4NEON(u32 v, float mtx[4][4])
{
GraphicsDrawer & drawer = dwnd().getDrawer();
@ -134,3 +136,193 @@ void gSPTransformVertex_NEON(float vtx[4], float mtx[4][4])
// Store vtx
vst1q_f32(vtx, _mtx0);
}
void DotProductMax7FullNeon( float v0[3], float v1[7][3], float lights[7][3], float _vtx[3])
{
asm volatile (
"pld [%0] \n\t" //preload lights
"pld [%0, #64] \n\t"
"pld [%3] \n\t" //preload vtx
"vld3.32 {d2[0],d3[0],d4[0]}, [%1] \n\t" //load v0
"vld3.32 {d6,d8,d10}, [%2]! \n\t" //load v1
"vld3.32 {d7,d9,d11}, [%2]! \n\t"
"vld3.32 {d18,d20,d22}, [%2]! \n\t"
"vld3.32 {d19[0],d21[0],d23[0]}, [%2] \n\t"
"vmul.f32 q0, q3, d2[0] \n\t" //q0=v0[0]*v1[i][0]
"vmul.f32 q6, q9, d2[0] \n\t" //q6=v0[0]*v1[i+4][0]
"vmla.f32 q0, q4, d3[0] \n\t" //q0+=v0[0]*v1[i][1]
"vmla.f32 q6, q10, d3[0] \n\t" //q6+=v0[0]*v1[i+4][1]
"vmov.f32 q15, #0.0 \n\t" //q15={0.0f,0.0f,0.0f,0.0f}
"vmla.f32 q0, q5, d4[0] \n\t" //q0+=v0[0]*v1[i][2]
"vmla.f32 q6, q11, d4[0] \n\t" //q6+=v0[0]*v1[i+4][2]
"vld3.32 {d4,d6,d8}, [%0]! \n\t" //load lights
"vld3.32 {d5,d7,d9}, [%0]! \n\t"
"vmax.f32 q0, q0, q15 \n\t" //q0=max(q0,q15)
"vmov.f32 d11, #0.0 \n\t" //d11={0.0f,0.0f}
"vmov.f32 d15, #0.0 \n\t" //d15={0.0f,0.0f}
"vmax.f32 q1, q6, q15 \n\t" //q1=max(q6,q15)
"vmov.f32 d13, #0.0 \n\t" //d13={0.0f,0.0f}
"vld3.32 {d10,d12,d14}, [%0]! \n\t" //d10={x1,y1}
"vld3.32 {d11[0],d13[0],d15[0]}, [%0] \n\t" //d10={x1,y1}
"vmul.f32 q2, q2, q0 \n\t" //q2=light.x*intensity
"vmul.f32 q3, q3, q0 \n\t" //q3=light.y*intensity
"vmul.f32 q4, q4, q0 \n\t" //q4=light.z*intensity
"vmul.f32 q5, q5, q1 \n\t" //q5=(light.x+4)*intensity
"vmul.f32 q6, q6, q1 \n\t" //q6=(light.y+4)*intensity
"vmul.f32 q7, q7, q1 \n\t" //q7=(light.z+4)*intensity
"vld3.32 {d22[0],d23[0],d24[0]}, [%3] \n\t" //load vtx
"vadd.f32 d4,d4,d5 \n\t" //add everything to vtx
"vadd.f32 d6,d6,d7 \n\t"
"vadd.f32 d8,d8,d9 \n\t"
"vadd.f32 d10,d10,d11 \n\t"
"vadd.f32 d12,d12,d13 \n\t"
"vadd.f32 d14,d14,d15 \n\t"
"vpadd.f32 d4,d4,d4 \n\t"
"vpadd.f32 d10,d10,d10 \n\t"
"vpadd.f32 d5,d6,d6 \n\t"
"vpadd.f32 d11,d12,d12 \n\t"
"vpadd.f32 d6,d8,d8 \n\t"
"vpadd.f32 d12,d14,d14 \n\t"
"vadd.f32 d4,d4,d10 \n\t"
"vadd.f32 d5,d5,d11 \n\t"
"vadd.f32 d6,d6,d12 \n\t"
"vadd.f32 d4,d4,d22 \n\t"
"vadd.f32 d5,d5,d23 \n\t"
"vadd.f32 d6,d6,d24 \n\t"
"vst3.32 {d4[0],d5[0],d6[0]}, [%3] \n\t"
: "+r"(lights), "+r"(v0), "+r"(v1), "+r"(_vtx):
: "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d8", "d9", "d10", "d11", "d12", "d13", "d14",
"d15", "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", "d24", "memory"
);
}
void DotProductMax4FullNeon( float v0[3], float v1[4][3], float _lights[4][3], float _vtx[3])
{
asm volatile (
"vld3.32 {d0[0],d2[0],d4[0]}, [%1] \n\t" //load v0
"vld3.32 {d6,d8,d10}, [%2]! \n\t" //load v1
"vld3.32 {d7,d9,d11}, [%2] \n\t" //
"vmul.f32 q0, q3, d0[0] \n\t" //product=v0[0]*v1[0]
"vld3.32 {d12,d14,d16}, [%0]! \n\t" //load lights
"vmla.f32 q0, q4, d2[0] \n\t" //product+=v0[1]*v1[1]
"vld3.32 {d13,d15,d17}, [%0] \n\t" //load lights +2
"vmov.f32 q11, #0.0 \n\t"
"vmla.f32 q0, q5, d4[0] \n\t" //product+=v0[2]*v1[2]
"vmax.f32 q0, q0, q11 \n\t" //product=max(product,0.0f)
"vld3.32 {d18[0],d19[0],d20[0]}, [%3]\n\t" //load vtx
"vmul.f32 q6, q6, q0 \n\t" //lights.r = lights.r * intensity
"vmul.f32 q7, q7, q0 \n\t" //lights.g = lights.g * intensity
"vmul.f32 q8, q8, q0 \n\t" //lights.b = lights.b * intensity
"vadd.f32 d12,d12,d13 \n\t" //add all values
"vadd.f32 d14,d14,d15 \n\t"
"vadd.f32 d16,d16,d17 \n\t"
"vpadd.f32 d12,d12,d12 \n\t"
"vpadd.f32 d13,d14,d14 \n\t"
"vpadd.f32 d14,d16,d16 \n\t"
"vadd.f32 d12,d12,d18 \n\t"
"vadd.f32 d13,d13,d19 \n\t"
"vadd.f32 d14,d14,d20 \n\t"
"vst3.32 {d12[0],d13[0],d14[0]}, [%3]\n\t" //store vtx
: "+r"(_lights), "+r"(v0), "+r"(v1), "+r"(_vtx):
: "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d8", "d9", "d10", "d11", "d12", "d13", "memory"
);
}
void gSPLightVertex4_NEON(u32 v)
{
GraphicsDrawer & drawer = dwnd().getDrawer();
if (!config.generalEmulation.enableHWLighting) {
for(int j = 0; j < 4; ++j) {
SPVertex & vtx = drawer.getVertex(v+j);
vtx.r = gSP.lights.rgb[gSP.numLights][R];
vtx.g = gSP.lights.rgb[gSP.numLights][G];
vtx.b = gSP.lights.rgb[gSP.numLights][B];
vtx.HWLight = 0;
s32 count = gSP.numLights-1;
while (count >= 6) {
DotProductMax7FullNeon(&vtx.nx,(float (*)[3])gSP.lights.i_xyz[gSP.numLights - count - 1],(float (*)[3])gSP.lights.rgb[gSP.numLights - count - 1],&vtx.r);
count -= 7;
}
while (count >= 3) {
DotProductMax4FullNeon(&vtx.nx,(float (*)[3])gSP.lights.i_xyz[gSP.numLights - count - 1],(float (*)[3])gSP.lights.rgb[gSP.numLights - count - 1],&vtx.r);
count -= 4;
}
while (count >= 0)
{
f32 intensity = DotProduct( &vtx.nx, gSP.lights.i_xyz[gSP.numLights - count - 1] );
if (intensity < 0.0f)
intensity = 0.0f;
vtx.r += gSP.lights.rgb[gSP.numLights - count - 1][R] * intensity;
vtx.g += gSP.lights.rgb[gSP.numLights - count - 1][G] * intensity;
vtx.b += gSP.lights.rgb[gSP.numLights - count - 1][B] * intensity;
count -= 1;
}
vtx.r = min(1.0f, vtx.r);
vtx.g = min(1.0f, vtx.g);
vtx.b = min(1.0f, vtx.b);
}
} else {
for(int j = 0; j < 4; ++j) {
SPVertex & vtx = drawer.getVertex(v+j);
vtx.HWLight = gSP.numLights;
vtx.r = vtx.nx;
vtx.g = vtx.ny;
vtx.b = vtx.nz;
}
}
}
void gSPLightVertex_NEON(SPVertex & _vtx)
{
if (config.generalEmulation.enableHWLighting == 0) {
_vtx.HWLight = 0;
_vtx.r = gSP.lights.rgb[gSP.numLights][R];
_vtx.g = gSP.lights.rgb[gSP.numLights][G];
_vtx.b = gSP.lights.rgb[gSP.numLights][B];
s32 count = gSP.numLights-1;
while (count >= 6) {
DotProductMax7FullNeon(&_vtx.nx,(float (*)[3])gSP.lights.i_xyz[gSP.numLights - count - 1],(float (*)[3])gSP.lights.rgb[gSP.numLights - count - 1],&_vtx.r);
count -= 7;
}
while (count >= 3) {
DotProductMax4FullNeon(&_vtx.nx,(float (*)[3])gSP.lights.i_xyz[gSP.numLights - count - 1],(float (*)[3])gSP.lights.rgb[gSP.numLights - count - 1],&_vtx.r);
count -= 4;
}
while (count >= 0)
{
f32 intensity = DotProduct( &_vtx.nx, gSP.lights.i_xyz[gSP.numLights - count - 1] );
if (intensity < 0.0f)
intensity = 0.0f;
_vtx.r += gSP.lights.rgb[gSP.numLights - count - 1][R] * intensity;
_vtx.g += gSP.lights.rgb[gSP.numLights - count - 1][G] * intensity;
_vtx.b += gSP.lights.rgb[gSP.numLights - count - 1][B] * intensity;
count -= 1;
}
_vtx.r = min(1.0f, _vtx.r);
_vtx.g = min(1.0f, _vtx.g);
_vtx.b = min(1.0f, _vtx.b);
} else {
_vtx.HWLight = gSP.numLights;
_vtx.r = _vtx.nx;
_vtx.g = _vtx.ny;
_vtx.b = _vtx.nz;
}
}

19
src/gSP.cpp
View File

@ -2567,18 +2567,21 @@ void gSPObjRendermode(u32 _mode)
void gSPTransformVertex4NEON(u32 v, float mtx[4][4]);
void gSPBillboardVertex4NEON(u32 v);
void gSPTransformVertex_NEON(float vtx[4], float mtx[4][4]);
void gSPLightVertex_NEON(SPVertex & _vtx);
void gSPLightVertex4_NEON(u32 v);
#endif //__NEON_OPT
#ifdef __VEC4_OPT
#ifndef __NEON_OPT
void (*gSPTransformVertex4)(u32 v, float mtx[4][4]) = gSPTransformVertex4_default;
void (*gSPBillboardVertex4)(u32 v) = gSPBillboardVertex4_default;
void (*gSPLightVertex4)(u32 v) = gSPLightVertex4_default;
#else
void (*gSPTransformVertex4)(u32 v, float mtx[4][4]) = gSPTransformVertex4NEON;
void (*gSPBillboardVertex4)(u32 v) = gSPBillboardVertex4NEON;
void (*gSPLightVertex4)(u32 v) = gSPLightVertex4_NEON;
#endif
void (*gSPLightVertex4)(u32 v) = gSPLightVertex4_default;
void (*gSPPointLightVertex4)(u32 v, float _vPos[4][3]) = gSPPointLightVertex4_default;
#endif
@ -2587,11 +2590,14 @@ void (*gSPPointLightVertex4)(u32 v, float _vPos[4][3]) = gSPPointLightVertex4_de
#ifndef __NEON_OPT
void (*gSPTransformVertex)(float vtx[4], float mtx[4][4]) =
gSPTransformVertex_default;
void (*gSPLightVertex)(SPVertex & _vtx) =
gSPLightVertex_default;
#else
void (*gSPTransformVertex)(float vtx[4], float mtx[4][4]) =
gSPTransformVertex_NEON;
void (*gSPLightVertex)(SPVertex & _vtx) =
gSPLightVertex_NEON;
#endif
void (*gSPLightVertex)(SPVertex & _vtx) = gSPLightVertex_default;
void (*gSPPointLightVertex)(SPVertex & _vtx, float * _vPos) = gSPPointLightVertex_default;
void (*gSPBillboardVertex)(u32 v, u32 i) = gSPBillboardVertex_default;
@ -2600,10 +2606,19 @@ void gSPSetupFunctions()
if (GBI.getMicrocodeType() != F3DEX2CBFD) {
#ifdef __VEC4_OPT
#ifndef __NEON_OPT
gSPLightVertex4 = gSPLightVertex4_default;
#else
gSPLightVertex4 = gSPLightVertex4_NEON;
#endif
gSPPointLightVertex4 = gSPPointLightVertex4_default;
#endif
#ifndef __NEON_OPT
gSPLightVertex = gSPLightVertex_default;
#else
gSPLightVertex = gSPLightVertex_NEON;
#endif
gSPPointLightVertex = gSPPointLightVertex_default;
return;
}