[XCX] Bloom pixelation and scale fix

Smoother bloom
This commit is contained in:
getdls 2017-12-07 00:19:27 +01:00
parent c4db15cc03
commit 80f032414a
4 changed files with 290 additions and 7 deletions

View File

@ -9,18 +9,50 @@ $scaleFactorY = always_decimal_format($fullHeight / 720.0);
#version 420
#extension GL_ARB_texture_gather : enable
// shader 8c1e55fd967b0496
// ver/horz alignment tweak, just a magic value
// To-do. Go back and re-center all passes instead.
// 1/4 -> 1/16 bloom pyramid . Pixelated unless scaled but still needs blur for light bleed..
// To-do. Check if screen res is * samples stable
const float resScale = <?=$scaleFactorX?>;
//const int sampleScale = 1;
//const float resScale = 4.0;
const float repositionBloom = 1.00125;
highp float lineRand(vec2 co)
{
highp float a = 12.9898;
highp float b = 78.233;
highp float c = 43758.5453;
highp float dt = dot(co.xy, vec2(a, b));
highp float sn = mod(dt, 3.14);
return fract(sin(sn) * c);
}
uniform ivec4 uf_remappedPS[5];
layout(binding = 0) uniform sampler2D textureUnitPS0;// Tex0 addr 0xf4e76000 res 320x180x1 dim 1 tm: 4 format 0816 compSel: 0 1 2 5 mipView: 0x0 (num 0x1) sliceView: 0x0 (num 0x1) Sampler0 ClampX/Y/Z: 6 6 6 border: 0
layout(location = 0) in vec4 passParameterSem0;
layout(location = 0) out vec4 passPixelColor0;
uniform vec2 uf_fragCoordScale;
// FabriceNeyret2, single pass gaussian by intermediate MIPmap level. https://www.shadertoy.com/view/ltScRG
// I hereby pledge my loyalty to the FabriceNeyret2 fanclub, this is bloody beautiful!
const int samples = 8, //check if must scale to pascal levels
LOD = 2, // gaussian done on MIPmap at scale LOD
sLOD = 1 << LOD; // tile size = 2^LOD
const float sigma = float(samples) * .25;
float gaussian(vec2 i) {
return exp(-.5* dot(i /= sigma, i)) / (6.28 * sigma*sigma);
}
vec4 blur(sampler2D sp, vec2 U, vec2 scale) {
vec4 O = vec4(0);
int s = samples / sLOD;
for (int i = 0; i < s*s; i++) {
vec2 d = vec2(i%s, i / s)*float(sLOD) - float(samples) / 2.;
O += gaussian(d) * textureLod(sp, U + scale * d, float(LOD));
}
return O / O.a;
}
int clampFI32(int v)
@ -54,7 +86,8 @@ void main()
vec3 cubeMapSTM;
int cubeMapFaceId;
R0f = passParameterSem0;
//R0f.xy = R0f.xy - (lineRand(gl_FragCoord.xy)*0.015);
//R0f.xy = vec2((passParameterSem0.x + passParameterSem0.z), (passParameterSem0.y + passParameterSem0.w));
// 0
R1f.x = R0f.x + (intBitsToFloat(uf_remappedPS[0].x) / resScale);
R1f.y = R0f.y + (intBitsToFloat(uf_remappedPS[0].y) / resScale);
@ -74,12 +107,25 @@ void main()
R0f.y = backupReg1f + (intBitsToFloat(uf_remappedPS[4].y) / resScale);
R4f.xyz = (texture(textureUnitPS0, R1f.xy).xyz);
vec2 coord = passParameterSem0.xy*textureSize(textureUnitPS0, 0); // R0f.xy;// vec2(0.5, 0.5);
vec2 ps = vec2(1.0) / textureSize(textureUnitPS0, 0);
vec2 uv = coord * ps;
R4f.xyz = blur(textureUnitPS0, uv, ps).xyz;
//R4f.xyz = (texture(textureUnitPS0, R1f.xy).xyz);
R3f.xyz = (texture(textureUnitPS0, R0f.zw).xyz);
R2f.xyz = (texture(textureUnitPS0, R2f.xy).xyz);
R2f.xyz = (texture(textureUnitPS0, R2f.xy).xyz);// prob?
R1f.xyz = (texture(textureUnitPS0, R1f.zw).xyz);
R0f.xyz = (texture(textureUnitPS0, R0f.xy).xyz);
/* //org
R4f.xyz = (texture(textureUnitPS0, R1f.xy).xyz);
R3f.xyz = (texture(textureUnitPS0, R0f.zw).xyz);
R2f.xyz = (texture(textureUnitPS0, R2f.xy).xyz);// prob?
R1f.xyz = (texture(textureUnitPS0, R1f.zw).xyz);
R0f.xyz = (texture(textureUnitPS0, R0f.xy).xyz);
*/
// 0
PV0f.x = R4f.y * intBitsToFloat(uf_remappedPS[0].z);
PV0f.y = R4f.x * intBitsToFloat(uf_remappedPS[0].z);

View File

@ -13,7 +13,7 @@ $scaleFactorY = always_decimal_format($fullHeight / 720.0);
//to do - extra edge smooth pass,
//const float resScale = 2.0;// 2.0 looks good
const float resScale = <?=$scaleFactorX?>;
const float shadowExposure = 0.75; //By truncating shadow, we decrease overall exposure during day only.
const float shadowExposure = 0.85; //By truncating shadow, we decrease overall exposure during day only.
uniform ivec4 uf_remappedPS[31];
layout(binding = 0) uniform sampler2D textureUnitPS0;// Tex0 addr 0xf4386000 res 1280x720x1 dim 1 tm: 4 format 001a compSel: 0 1 2 3 mipView: 0x0 (num 0x1) sliceView: 0x0 (num 0x1) Sampler0 ClampX/Y/Z: 2 2 2 border: 0
layout(binding = 1) uniform sampler2D textureUnitPS1;// Tex1 addr 0xf599a000 res 1024x1024x1 dim 1 tm: 4 format 0005 compSel: 0 4 4 5 mipView: 0x0 (num 0x1) sliceView: 0x0 (num 0x1) Sampler1 ClampX/Y/Z: 6 6 6 border: 2

View File

@ -0,0 +1,237 @@
<?php
include 'Source/functions.php';
$fullWidth = $argv[1];
$fullHeight = $argv[2];
$scaleFactorX = always_decimal_format($fullWidth / 1280.0);
$scaleFactorY = always_decimal_format($fullHeight / 720.0);
?>
#version 420
#extension GL_ARB_texture_gather : enable
// shader b3fb199c73caa796 // bloom pyramid sample 1st
const float resScale = <?=$scaleFactorX?>;
//const float resScale = 3.0;
uniform ivec4 uf_remappedPS[4];
layout(binding = 0) uniform sampler2D textureUnitPS0;// Tex0 addr 0xf551a000 res 1280x720x1 dim 1 tm: 4 format 001a compSel: 0 1 2 3 mipView: 0x0 (num 0x1) sliceView: 0x0 (num 0x1) Sampler0 ClampX/Y/Z: 2 2 2 border: 0
layout(binding = 1) uniform sampler2D textureUnitPS1;// Tex1 addr 0xf470a000 res 1280x720x1 dim 1 tm: 4 format 0816 compSel: 0 1 2 5 mipView: 0x0 (num 0x1) sliceView: 0x0 (num 0x1) Sampler1 ClampX/Y/Z: 2 2 2 border: 0
layout(location = 0) in vec4 passParameterSem0;
layout(location = 0) out vec4 passPixelColor0;
uniform vec2 uf_fragCoordScale;
const int samples = 8, //check if must scale to pascal levels 8 balances 13 is smooth.
LOD = 2, // gaussian done on MIPmap at scale LOD
sLOD = 1 << LOD; // tile size = 2^LOD
const float sigma = float(samples) * .25;
float gaussian(vec2 i) {
return exp(-.5* dot(i /= sigma, i)) / (6.28 * sigma*sigma);
}
vec4 blur(sampler2D sp, vec2 U, vec2 scale) {
vec4 O = vec4(0);
int s = samples / sLOD;
for (int i = 0; i < s*s; i++) {
vec2 d = vec2(i%s, i / s)*float(sLOD) - float(samples) / 2.;
O += gaussian(d) * textureLod(sp, U + scale * d, float(LOD));
}
return O / O.a;
}
int clampFI32(int v)
{
if( v == 0x7FFFFFFF )
return floatBitsToInt(1.0);
else if( v == 0xFFFFFFFF )
return floatBitsToInt(0.0);
return floatBitsToInt(clamp(intBitsToFloat(v), 0.0, 1.0));
}
float mul_nonIEEE(float a, float b){ return min(a*b,min(abs(a)*3.40282347E+38F,abs(b)*3.40282347E+38F)); }
void main()
{
vec4 R0f = vec4(0.0);
vec4 R1f = vec4(0.0);
vec4 R2f = vec4(0.0);
vec4 R3f = vec4(0.0);
vec4 R4f = vec4(0.0);
vec4 R5f = vec4(0.0);
vec4 R6f = vec4(0.0);
vec4 R7f = vec4(0.0);
vec4 R123f = vec4(0.0);
vec4 R124f = vec4(0.0);
vec4 R125f = vec4(0.0);
vec4 R126f = vec4(0.0);
vec4 R127f = vec4(0.0);
float backupReg0f, backupReg1f, backupReg2f, backupReg3f, backupReg4f;
vec4 PV0f = vec4(0.0), PV1f = vec4(0.0);
float PS0f = 0.0, PS1f = 0.0;
vec4 tempf = vec4(0.0);
float tempResultf;
int tempResulti;
ivec4 ARi = ivec4(0);
bool predResult = true;
vec3 cubeMapSTM;
int cubeMapFaceId;
R0f = passParameterSem0;
// 0
R1f.x = R0f.x + intBitsToFloat(uf_remappedPS[0].x)/resScale;
R1f.y = R0f.y + intBitsToFloat(uf_remappedPS[0].y)/resScale;
R0f.z = R0f.x + intBitsToFloat(uf_remappedPS[0].z)/resScale;
R0f.w = R0f.y + intBitsToFloat(uf_remappedPS[0].w)/resScale;
// 1
backupReg0f = R0f.x;
backupReg1f = R0f.y;
backupReg0f = R0f.x;
backupReg1f = R0f.y;
R0f.x = backupReg0f + intBitsToFloat(uf_remappedPS[1].x)/resScale;
R0f.y = backupReg1f + intBitsToFloat(uf_remappedPS[1].y) /resScale;
R1f.z = backupReg0f + intBitsToFloat(uf_remappedPS[1].z) /resScale;
R1f.w = backupReg1f + intBitsToFloat(uf_remappedPS[1].w) /resScale;
vec2 coord = passParameterSem0.xy*textureSize(textureUnitPS0, 0); // R0f.xy;// vec2(0.5, 0.5);
vec2 ps = vec2(1.0) / textureSize(textureUnitPS0, 0);
vec2 uv = coord * ps;
R2f.xyzw = blur(textureUnitPS1, uv, ps);
R3f = R2f;
R4f = R2f;
R5f = R2f;
/*
R2f.xyzw = (texture(textureUnitPS1, R1f.xy).xyzw);
R3f.xyzw = (texture(textureUnitPS1, R0f.zw).xyzw);
R4f.xyzw = (texture(textureUnitPS1, R0f.xy).xyzw);
R5f.xyzw = (texture(textureUnitPS1, R1f.zw).xyzw);
*/
//R1f.xyzw = blur(textureUnitPS0, uv, ps);
//R7f = R6f;
//R0f = R6f;
//R1f = R6f;
R6f.xyzw = (texture(textureUnitPS0, R1f.xy).xyzw);
R7f.xyzw = (texture(textureUnitPS0, R0f.zw).xyzw);
R0f.xyzw = (texture(textureUnitPS0, R0f.xy).xyzw);
R1f.xyzw = (texture(textureUnitPS0, R1f.zw).xyzw);
// 0
PV0f.x = R2f.z + R3f.z;
PV0f.y = R2f.y + R3f.y;
PV0f.z = R2f.x + R3f.x;
PV0f.w = R2f.w + R3f.w;
// 1
PV1f.x = PV0f.x + R4f.z;
PV1f.y = PV0f.y + R4f.y;
PV1f.z = PV0f.z + R4f.x;
PV1f.w = PV0f.w + R4f.w;
R127f.x = R6f.x + R7f.x;
PS1f = R127f.x;
// 2
PV0f.x = PV1f.x + R5f.z;
PV0f.y = PV1f.y + R5f.y;
PV0f.z = PV1f.z + R5f.x;
PV0f.w = PV1f.w + R5f.w;
R127f.w = R6f.y + R7f.y;
PS0f = R127f.w;
// 3
PV1f.x = PV0f.x * intBitsToFloat(uf_remappedPS[2].z);
PV1f.y = PV0f.y * intBitsToFloat(uf_remappedPS[2].z);
PV1f.z = PV0f.z * intBitsToFloat(uf_remappedPS[2].z);
PV1f.w = PV0f.w * intBitsToFloat(uf_remappedPS[2].z);
R127f.z = R6f.z + R7f.z;
PS1f = R127f.z;
// 4
PV0f.x = max(PV1f.x, 0.0);
PV0f.y = max(PV1f.y, 0.0);
PV0f.z = max(PV1f.z, 0.0);
PV0f.w = max(PV1f.w, 0.0);
R126f.y = R6f.w + R7f.w;
PS0f = R126f.y;
// 5
R6f.x = min(PV0f.x, 4.0);
PV1f.x = R6f.x;
R125f.y = min(PV0f.y, 4.0);
PV1f.y = R125f.y;
R126f.z = min(PV0f.z, 4.0);
PV1f.z = R126f.z;
R125f.w = min(PV0f.w, 4.0);
R124f.x = R127f.x + R0f.x;
PS1f = R124f.x;
// 6
PV0f.x = PV1f.z + -(intBitsToFloat(uf_remappedPS[2].x));
PV0f.y = PV1f.y + -(intBitsToFloat(uf_remappedPS[3].x));
PV0f.z = PV1f.z + -(intBitsToFloat(uf_remappedPS[3].x));
PV0f.w = PV1f.x + -(intBitsToFloat(uf_remappedPS[3].x));
R127f.y = PV1f.y + -(intBitsToFloat(uf_remappedPS[2].x));
PS0f = R127f.y;
// 7
R127f.x = R6f.x + -(intBitsToFloat(uf_remappedPS[2].x));
PV1f.y = max(PV0f.z, 0.0);
PV1f.z = max(PV0f.y, 0.0);
PV1f.w = max(PV0f.w, 0.0);
R126f.x = max(PV0f.x, 0.0);
PS1f = R126f.x;
// 8
backupReg0f = R127f.y;
R125f.x = dot(vec4(intBitsToFloat(uf_remappedPS[3].y),intBitsToFloat(uf_remappedPS[3].y),intBitsToFloat(uf_remappedPS[3].y),-0.0),vec4(PV1f.y,PV1f.z,PV1f.w,0.0));
PV0f.x = R125f.x;
PV0f.y = R125f.x;
PV0f.z = R125f.x;
PV0f.w = R125f.x;
R127f.y = max(backupReg0f, 0.0);
PS0f = R127f.y;
// 9
backupReg0f = R126f.y;
backupReg1f = R127f.x;
backupReg2f = R127f.w;
R127f.x = R127f.z + R0f.z;
R126f.y = backupReg0f + R0f.w;
PV1f.z = max(backupReg1f, 0.0);
R127f.w = backupReg2f + R0f.y;
R126f.w = R126f.z * PV0f.x;
PS1f = R126f.w;
// 10
tempf.x = dot(vec4(intBitsToFloat(uf_remappedPS[2].y),intBitsToFloat(uf_remappedPS[2].y),intBitsToFloat(uf_remappedPS[2].y),-0.0),vec4(R126f.x,R127f.y,PV1f.z,0.0));
PV0f.x = tempf.x;
PV0f.y = tempf.x;
PV0f.z = tempf.x;
PV0f.w = tempf.x;
R124f.w = tempf.x;
R127f.z = R125f.y * R125f.x;
PS0f = R127f.z;
// 11
PV1f.x = R125f.w * R125f.x;
PV1f.y = R6f.x * R125f.x;
R125f.z = R127f.w + R1f.y;
R127f.w = R127f.x + R1f.z;
R0f.w = R126f.y + R1f.w;
PS1f = R0f.w;
// 12
backupReg0f = R124f.x;
R124f.x = (R125f.y * R124f.w + R127f.z);
R125f.y = (R6f.x * R124f.w + PV1f.y);
PV0f.z = backupReg0f + R1f.x;
R123f.w = (R126f.z * R124f.w + R126f.w);
PV0f.w = R123f.w;
R127f.x = (R125f.w * R124f.w + PV1f.x);
PS0f = R127f.x;
// 13
R125f.x = R127f.w * intBitsToFloat(uf_remappedPS[3].z);
R126f.y = R125f.z * intBitsToFloat(uf_remappedPS[3].z);
PV1f.z = PV0f.z * intBitsToFloat(uf_remappedPS[3].z);
R127f.w = R0f.w * intBitsToFloat(uf_remappedPS[3].z);
PS1f = min(PV0f.w, intBitsToFloat(uf_remappedPS[2].w));
// 14
PV0f.x = min(R127f.x, intBitsToFloat(uf_remappedPS[2].w));
PV0f.y = min(R125f.y, intBitsToFloat(uf_remappedPS[2].w));
PV0f.z = min(R124f.x, intBitsToFloat(uf_remappedPS[2].w));
R0f.x = max(PV1f.z, PS1f);
PS0f = R0f.x;
// 15
R0f.y = max(R126f.y, PV0f.z);
R0f.z = max(R125f.x, PV0f.y);
R0f.w = max(R127f.w, PV0f.x);
// export
passPixelColor0 = vec4(R0f.x, R0f.y, R0f.z, R0f.w)*1.5;
}

View File

@ -12,7 +12,7 @@ $scaleFactorY = always_decimal_format($fullHeight / 720.0);
// shadows scaling, ie 4096 = x4
// To do - some nice edge smoothing at higher res
const float resScale = <?=$scaleFactorX?>;
const float shadowExposure = 0.75;
const float shadowExposure = 0.85;
//const float resScale = 4;
uniform ivec4 uf_remappedPS[23];
layout(binding = 0) uniform sampler2D textureUnitPS0;// Tex0 addr 0xf4386000 res 1280x720x1 dim 1 tm: 4 format 001a compSel: 0 1 2 3 mipView: 0x0 (num 0x1) sliceView: 0x0 (num 0x1) Sampler0 ClampX/Y/Z: 2 2 2 border: 0