pcsx2/plugins/zzogl-pg/opengl/ps2hw_gl4.glsl

916 lines
24 KiB
GLSL

//#version 420 Keep it for text editor detection
// ZZ Open GL graphics plugin
// Copyright (c)2009-2010 zeydlitz@gmail.com, arcum42@gmail.com, gregory.hainaut@gmail.com
// Based on Zerofrog's ZeroGS KOSMOS (c)2005-2008
//
// This program is free software; you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation; either version 2 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program; if not, write to the Free Software
// Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
// divides by z for every pixel, instead of in vertex shader
// fixes kh textures
#extension GL_ARB_texture_rectangle: require
#extension GL_ARB_shading_language_420pack: require
#extension GL_ARB_separate_shader_objects : require
//#define TEST_AEM // tests AEM for black pixels
//#define REGION_REPEAT // set if texture wrapping mode is region repeat
//#define WRITE_DEPTH // set if depth is also written in a MRT
//#define ACCURATE_DECOMPRESSION // set for less capable hardware ATI Radeon 9000 series
//#define EXACT_COLOR // make sure the output color is clamped to 1/255 boundaries (for alpha testing)
//#define NO_LOGZ // disable logz
#define PERSPECTIVE_CORRECT_TEX
#ifdef PERSPECTIVE_CORRECT_TEX
#define TEX_XY tex.xy/tex.z
#define TEX_DECL vec3
#else
#define TEX_XY tex.xy
#define TEX_DECL vec2
#endif
// NVidia CG-data types
#define half2 vec2
#define half3 vec3
#define half4 vec4
#define float2 vec2
#define float3 vec3
#define float4 vec4
////////////////////////////////////////////////////////////////////
// INPUT/OUTPUT
////////////////////////////////////////////////////////////////////
// NOTE: Future optimization tex.w is normally useless (in cg it is a float3) so it can contains the fog value
struct vertex
{
vec4 color;
TEX_DECL tex;
vec4 Z;
float fog;
};
#ifdef VERTEX_SHADER
out gl_PerVertex {
invariant vec4 gl_Position;
float gl_PointSize;
float gl_ClipDistance[];
};
layout(location = 0) in ivec4 Vert;
layout(location = 1) in vec4 Color;
layout(location = 2) in vec4 SecondaryColor;
layout(location = 3) in vec3 TexCoord;
layout(location = 0) out vertex VSout;
#endif
#ifdef FRAGMENT_SHADER
layout(location = 0) in vertex PSin;
// FIXME: host only do glDrawBuffers of 1 buffers not 2. I think this is a major bug
layout(location = 0) out vec4 FragData0;
layout(location = 1) out vec4 FragData1;
#endif
////////////////////////////////////////////////////////////////////
// Texture SAMPLER
////////////////////////////////////////////////////////////////////
// // main ps2 memory, each pixel is stored in 32bit color
// uniform sampler2DRect g_sMemory[2];
//
// // used to get the tiled offset into a page given the linear offset
// uniform sampler2DRect g_sSrcFinal;
// uniform sampler2D g_sBlocks;
// uniform sampler2D g_sBilinearBlocks;
// uniform sampler2D g_sConv16to32;
// uniform sampler3D g_sConv32to16;
// uniform sampler2DRect g_sBitwiseANDX;
// uniform sampler2DRect g_sBitwiseANDY;
// uniform sampler2DRect g_sInterlace;
//
// // used only on rare cases where the render target is PSMT8H
// uniform sampler2D g_sCLUT;
// main ps2 memory, each pixel is stored in 32bit color
layout(binding = 10) uniform sampler2DRect g_sMemory; // dual context
// used to get the tiled offset into a page given the linear offset
layout(binding = 1) uniform sampler2DRect g_sSrcFinal;
layout(binding = 2) uniform sampler2D g_sBlocks;
layout(binding = 3) uniform sampler2D g_sBilinearBlocks;
layout(binding = 4) uniform sampler2D g_sConv16to32;
layout(binding = 5) uniform sampler3D g_sConv32to16;
layout(binding = 6) uniform sampler2DRect g_sBitwiseANDX;
layout(binding = 7) uniform sampler2DRect g_sBitwiseANDY;
layout(binding = 8) uniform sampler2DRect g_sInterlace;
// used only on rare cases where the render target is PSMT8H
layout(binding = 9) uniform sampler2D g_sCLUT;
////////////////////////////////////////////////////////////////////
// UNIFORM BUFFER
////////////////////////////////////////////////////////////////////
layout(std140, binding = 0) uniform constant_buffer
{
// Both shader
// .z is used for the addressing fn
// FIXME: not same value between c and shader...
// float4 g_fBilinear = float4(-0.7f, -0.65f, 0.9,1/32767.0f);
float4 g_fBilinear;
float4 g_fZBias;
float4 g_fc0;
float4 g_fMult;
// Vertex
float4 g_fZ; // transforms d3dcolor z into float z
// Pixel
half4 g_fExactColor;
};
layout(std140, binding = 1) uniform common_buffer
{
float4 g_fPosXY;
float4 g_fFogColor;
};
layout(std140, binding = 2) uniform vertex_buffer
{
float4 g_fBitBltPos;
float4 g_fBitBltTex;
float4 g_fBitBltTrans;
};
layout(std140, binding = 3) uniform fragment_buffer
{
half4 fTexAlpha2;
float4 g_fTexOffset; // converts the page and block offsets into the mem addr/1024
float4 g_fTexDims; // mult by tex dims when accessing the block texture
float4 g_fTexBlock;
float4 g_fClampExts; // if clamping the texture, use (minu, minv, maxu, maxv)
float4 TexWrapMode; // 0 - repeat/clamp, 1 - region rep (use fRegRepMask)
float4 g_fRealTexDims; // tex dims used for linear filtering (w,h,1/w,1/h)
// (alpha0, alpha1, 1 if highlight2 and tcc is rgba, 1-y)
half4 g_fTestBlack; // used for aem bit
float4 g_fPageOffset;
half4 fTexAlpha;
float4 g_fInvTexDims; // similar to g_fClutOff
// used for rectblitting
float4 g_fBitBltZ;
half4 g_fOneColor; // col*.xxxy+.zzzw
};
// given a local tex coord, returns the coord in the memory
float2 ps2memcoord(float2 realtex)
{
float4 off;
// block off
realtex.xy = realtex.xy * g_fTexDims.xy + g_fTexDims.zw;
realtex.xy = (realtex.xy - fract(realtex.xy)) * g_fMult.zw;
float2 fblock = fract(realtex.xy);
off.xy = realtex.xy-fblock.xy;
#ifdef ACCURATE_DECOMPRESSION
off.zw = texture(g_sBlocks, g_fTexBlock.xy*fblock + g_fTexBlock.zw).ar;
off.x = dot(off.xy, g_fTexOffset.xy);
float r = g_fTexOffset.w;
float f = fract(off.x);
float fadd = g_fTexOffset.z * off.z;
off.w = off.x + fadd + r;
off.x = fract(f + fadd + r);
off.w -= off.x ;
#else
off.z = texture(g_sBlocks, g_fTexBlock.xy*fblock + g_fTexBlock.zw).a;
// combine the two
off.x = dot(off.xyz, g_fTexOffset.xyz)+g_fTexOffset.w;
off.x = modf(off.x, off.w);
#endif
off.xy = off.xw * g_fPageOffset.zy + g_fPageOffset.wx;
//off.y = off.w * g_fPageOffset.y + g_fPageOffset.x;
return off.xy;
}
// find all texcoords for bilinear filtering
// assume that orgtex are already on boundaries
void ps2memcoord4(float4 orgtex, out float4 off0, out float4 off1)
{
//float4 off0, off1, off2, off3;
float4 realtex;
// block off
realtex = (orgtex * g_fTexDims.xyxy + g_fTexDims.zwzw);// * g_fMult.zwzw;
float4 fblock = fract(realtex.xyzw);
float4 ftransblock = g_fTexBlock.xyxy*fblock + g_fTexBlock.zwzw;
realtex -= fblock;
float4 transvals = g_fTexOffset.x * realtex.xzxz + g_fTexOffset.y * realtex.yyww + g_fTexOffset.w;
float4 colors;// = texture(g_sBilinearBlocks, ftransblock.xy);
// this is faster on ffx ingame
colors.x = texture(g_sBlocks, ftransblock.xy).a;
colors.y = texture(g_sBlocks, ftransblock.zy).a;
colors.z = texture(g_sBlocks, ftransblock.xw).a;
colors.w = texture(g_sBlocks, ftransblock.zw).a;
float4 fr, rem;
#ifdef ACCURATE_DECOMPRESSION
fr = fract(transvals);
float4 fadd = colors * g_fTexOffset.z;
rem = transvals + fadd;
fr = fract(fr + fadd);
rem -= fr;
#else
transvals += colors * g_fTexOffset.z;
fr = modf(transvals, rem);
#endif
rem = rem * g_fPageOffset.y + g_fPageOffset.x;
fr = fr * g_fPageOffset.z + g_fPageOffset.w;
// combine
off0 = g_fc0.yxyx * fr.xxyy + g_fc0.xyxy * rem.xxyy;
off1 = g_fc0.yxyx * fr.zzww + g_fc0.xyxy * rem.zzww;
}
void ps2memcoord4_fast(float4 orgtex, out float4 off0, out float4 off1)
{
float4 realtex;
realtex = (orgtex * g_fTexDims.xyxy + g_fTexDims.zwzw);// * g_fMult.zwzw;
float4 fblock = fract(realtex.xyzw);
float2 ftransblock = g_fTexBlock.xy*fblock.xy + g_fTexBlock.zw;
realtex -= fblock;
float4 transvals = g_fTexOffset.x * realtex.xzxz + g_fTexOffset.y * realtex.yyww + g_fTexOffset.w;
float4 colors = texture(g_sBilinearBlocks, ftransblock.xy);
float4 fr, rem;
#ifdef ACCURATE_DECOMPRESSION
fr = fract(transvals);
float4 fadd = colors * g_fTexOffset.z;
rem = transvals + fadd;
fr = fract(fr + fadd);
rem -= fr;
#else
transvals += colors * g_fTexOffset.z;
fr = modf(transvals, rem);
#endif
rem = rem * g_fPageOffset.y + g_fPageOffset.x;
fr = fr * g_fPageOffset.z;
off0 = g_fc0.yxyx * fr.xxyy + g_fc0.xyxy * rem.xxyy;
off1 = g_fc0.yxyx * fr.zzww + g_fc0.xyxy * rem.zzww;
}
// Wrapping modes
#if defined(REPEAT)
float2 ps2addr(float2 coord)
{
return fract(coord.xy);
}
#elif defined(CLAMP)
float2 ps2addr(float2 coord)
{
return clamp(coord.xy, g_fClampExts.xy, g_fClampExts.zw);
}
#elif defined(REGION_REPEAT)
// computes the local tex coord along with addressing modes
float2 ps2addr(float2 coord)
{
float2 final = fract(clamp(coord.xy, g_fClampExts.xy, g_fClampExts.zw));
if( TexWrapMode.x > g_fBilinear.z ) // region repeat mode for x (umsk&x)|ufix
final.x = texture(g_sBitwiseANDX, abs(coord.x)*TexWrapMode.zx).x * g_fClampExts.x + g_fClampExts.z;
if( TexWrapMode.y > g_fBilinear.z ) // region repeat mode for x (vmsk&x)|vfix
final.y = texture(g_sBitwiseANDY, abs(coord.y)*TexWrapMode.wy).x * g_fClampExts.y + g_fClampExts.w;
return final;
}
#else
float2 ps2addr(float2 coord)
{
return fract(clamp(coord.xy, g_fClampExts.xy, g_fClampExts.zw));
}
#endif
half4 tex2DPS_32(float2 tex0)
{
return texture(g_sMemory, ps2memcoord(tex0));
}
// use when texture is not tiled -- shader 1
half4 tex2DPS_tex32(float2 tex0)
{
return texture(g_sMemory, g_fTexDims.xy*tex0+g_fTexDims.zw)*g_fZBias.zzzw+g_fPageOffset.w;
}
// use when texture is not tiled -- shader 2
half4 tex2DPS_clut32(float2 tex0)
{
float index = texture(g_sMemory, g_fTexDims.xy*tex0+g_fTexDims.zw).a+g_fPageOffset.w;
return texture(g_sCLUT, index*g_fExactColor.xz+g_fExactColor.yz);
}
// Shader 3
// use when texture is not tiled and converting from 32bit to 16bit
// don't convert on the block level, only on the column level
// so every other 8 pixels, use the upper bits instead of lower
half4 tex2DPS_tex32to16(float2 tex0)
{
bool upper = false;
tex0.y += g_fPageOffset.z;
float2 ffrac = mod(tex0, g_fTexOffset.xy);
tex0.xy = g_fc0.ww * (tex0.xy + ffrac);
if( ffrac.x > g_fTexOffset.z ) {
tex0.x -= g_fTexOffset.z;
upper = true;
}
if( ffrac.y >= g_fTexOffset.w ) {
tex0.y -= g_fTexOffset.w;
tex0.x += g_fc0.w;
}
half4 color = texture(g_sMemory, g_fTexDims.xy*tex0+g_fTexDims.zw)*g_fZBias.zzzw+g_fPageOffset.w;
float2 uv = upper ? color.xw : color.zy;
return texture(g_sConv16to32, uv+g_fPageOffset.xy);
}
// Shader 4
// used when a 16 bit texture is used an 8h
half4 tex2DPS_tex16to8h(float2 tex0)
{
float4 final;
float2 ffrac = mod(tex0+g_fPageOffset.zw, g_fTexOffset.xy);
tex0.xy = g_fPageOffset.xy * tex0.xy - ffrac * g_fc0.yw;
if( ffrac.x > g_fTexOffset.x*g_fc0.w )
tex0.x += g_fTexOffset.x*g_fc0.w;
if( tex0.x >= g_fc0.y ) tex0 += g_fTexOffset.zw;
float4 upper = texture(g_sMemory, g_fTexDims.xy*tex0+g_fTexDims.zw);
// only need alpha
float index = texture(g_sConv32to16, upper.zyx-g_fc0.z).y + upper.w*g_fc0.w*g_fc0.w;
return texture(g_sCLUT, index+g_fExactColor.yz);
}
// Shader 5
// used when a 16 bit texture is used a 32bit one
half4 tex2DPS_tex16to32(float2 tex0)
{
float4 final;
float2 ffrac = mod(tex0+g_fPageOffset.zw, g_fTexOffset.xy);
//tex0.xy = g_fPageOffset.xy * tex0.xy - ffrac * g_fc0.yw;
tex0.y += g_fPageOffset.y * ffrac.y;
if( ffrac.x > g_fTexOffset.z ) {
tex0.x -= g_fTexOffset.z;
tex0.y += g_fTexOffset.w;
}
float fconst = g_fc0.w*g_fc0.w;
float4 lower = texture(g_sSrcFinal, g_fTexDims.xy*tex0);
float4 upper = texture(g_sMemory, g_fTexDims.xy*tex0+g_fTexDims.zw);
final.zy = texture(g_sConv32to16, lower.zyx).xy + lower.ww*fconst;
final.xw = texture(g_sConv32to16, upper.zyx).xy + upper.ww*fconst;
return final;
}
half4 tex2DPS_tex16to32h(float2 tex0)
{
float4 final = vec4(0.0, 0.0, 0.0, 0.0);
return final;
}
//half4 f;
//f.w = old.y > (127.2f/255.0f) ? 1 : 0;
//old.y -= 0.5f * f.w;
//f.xyz = fract(old.yyx*half3(2.002*255.0f/256.0f, 64.025f*255.0f/256.0f, 8.002*255.0f/256.0f));
//f.y += old.x * (0.25f*255.0f/256.0f);
////////////////////////////////
// calculates the texture color
////////////////////////////////
#define decl_ps2shade(num) \
decl_ps2shade_##num(_32) \
decl_ps2shade_##num(_tex32) \
decl_ps2shade_##num(_clut32) \
decl_ps2shade_##num(_tex32to16) \
decl_ps2shade_##num(_tex16to8h) \
decl_ps2shade_##num(_tex16to32h)
// nearest
#define decl_ps2shade_0(bit) \
float4 ps2shade0##bit( TEX_DECL tex) \
{ \
return tex2DPS##bit( ps2addr(TEX_XY)); \
}
// do fast memcoord4 calcs when textures behave well
#ifdef REPEAT
#define PS2MEMCOORD4 ps2memcoord4
#else
#define PS2MEMCOORD4 ps2memcoord4
#endif
#define decl_BilinearFilter(bit, addrfn) \
half4 BilinearFilter##bit(float2 tex0) \
{ \
float4 off0, off1; \
float4 ftex; \
float2 ffrac; \
ftex.xy = tex0 + g_fBilinear.xy * g_fRealTexDims.zw; \
ffrac = fract(ftex.xy*g_fRealTexDims.xy); \
ftex.xy -= ffrac.xy * g_fRealTexDims.zw; \
\
ftex.zw = ps2addr(ftex.xy + g_fRealTexDims.zw); \
ftex.xy = ps2addr(ftex.xy); \
\
PS2MEMCOORD4(ftex, off0, off1); \
half4 c0 = texture(g_sMemory, off0.xy); \
half4 c1 = texture(g_sMemory, off0.zw); \
half4 c2 = texture(g_sMemory, off1.xy); \
half4 c3 = texture(g_sMemory, off1.zw); \
return mix( mix(c0, c1, vec4(ffrac.x)), mix(c2, c3, ffrac.x), vec4(ffrac.y) ); \
}
decl_BilinearFilter(_32, ps2addr)
decl_BilinearFilter(_tex32, ps2addr)
decl_BilinearFilter(_clut32, ps2addr)
decl_BilinearFilter(_tex32to16, ps2addr)
decl_BilinearFilter(_tex16to8h, ps2addr)
decl_BilinearFilter(_tex16to32h, ps2addr)
//TODO! For mip maps, only apply when LOD >= 0
// lcm == 0, LOD = log(1/Q)*L + K, lcm == 1, LOD = K
// bilinear
#define decl_ps2shade_1(bit) \
half4 ps2shade1##bit(TEX_DECL tex) \
{ \
return BilinearFilter##bit(TEX_XY); \
}
// nearest, mip nearest
#define decl_ps2shade_2(bit) \
half4 ps2shade2##bit(TEX_DECL tex) \
{ \
return tex2DPS##bit( ps2addr(TEX_XY)); \
}
// nearest, mip linear
#define decl_ps2shade_3(bit) \
half4 ps2shade3##bit(TEX_DECL tex) \
{ \
return tex2DPS##bit(ps2addr(TEX_XY)); \
}
// linear, mip nearest
#define decl_ps2shade_4(bit) \
half4 ps2shade4##bit(TEX_DECL tex) \
{ \
return BilinearFilter##bit(TEX_XY); \
}
// linear, mip linear
#define decl_ps2shade_5(bit) \
half4 ps2shade5##bit(TEX_DECL tex) \
{ \
return BilinearFilter##bit(TEX_XY); \
}
decl_ps2shade(0)
decl_ps2shade(1)
decl_ps2shade(2)
decl_ps2shade(3)
decl_ps2shade(4)
decl_ps2shade(5)
half4 ps2CalcShade(half4 texcol, half4 color)
{
#ifdef TEST_AEM
if( dot(texcol.xyzw, g_fTestBlack.xyzw) <= g_fc0.z )
texcol.w = g_fc0.x;
else
#endif
texcol.w = texcol.w * fTexAlpha.y + fTexAlpha.x;
texcol = texcol * (fTexAlpha2.zzzw * color + fTexAlpha2.xxxy) + fTexAlpha.zzzw * color.wwww;
return texcol;
}
// final ops on the color
#ifdef EXACT_COLOR
half4 ps2FinalColor(half4 col)
{
// g_fOneColor has to scale by 255
half4 temp = col * g_fOneColor.xxxy + g_fOneColor.zzzw;
temp.w = floor(temp.w)*g_fExactColor.w;
return temp;
}
#else
half4 ps2FinalColor(half4 col)
{
return col * g_fOneColor.xxxy + g_fOneColor.zzzw;
}
#endif
#ifdef FRAGMENT_SHADER // This is code only for FRAGMENTS (pixel shader)
#ifdef WRITE_DEPTH
void write_depth_target()
{
FragData1 = PSin.Z;
}
#else
void write_depth_target() { }
#endif
void RegularPS() {
// whenever outputting depth, make sure to mult by 255/256 and 1
FragData0 = ps2FinalColor(PSin.color);
write_depth_target();
}
#define DECL_TEXPS(num, bit) \
void Texture##num##bit##PS() \
{ \
FragData0 = ps2FinalColor(ps2CalcShade(ps2shade##num##bit(PSin.tex), PSin.color)); \
write_depth_target(); \
}
#define DECL_TEXPS_(num) \
DECL_TEXPS(num, _32) \
DECL_TEXPS(num, _tex32) \
DECL_TEXPS(num, _clut32) \
DECL_TEXPS(num, _tex32to16) \
DECL_TEXPS(num, _tex16to8h)
DECL_TEXPS_(0)
DECL_TEXPS_(1)
DECL_TEXPS_(2)
DECL_TEXPS_(3)
DECL_TEXPS_(4)
DECL_TEXPS_(5)
void RegularFogPS() {
half4 c;
c.xyz = mix(g_fFogColor.xyz, PSin.color.xyz, vec3(PSin.fog));
c.w = PSin.color.w;
FragData0 = ps2FinalColor(c);
write_depth_target();
}
#define DECL_TEXFOGPS(num, bit) \
void TextureFog##num##bit##PS() \
{ \
half4 c = ps2CalcShade(ps2shade##num##bit(PSin.tex), PSin.color); \
c.xyz = mix(g_fFogColor.xyz, c.xyz, vec3(PSin.fog)); \
FragData0 = ps2FinalColor(c); \
write_depth_target(); \
}
#define DECL_TEXFOGPS_(num) \
DECL_TEXFOGPS(num, _32) \
DECL_TEXFOGPS(num, _tex32) \
DECL_TEXFOGPS(num, _clut32) \
DECL_TEXFOGPS(num, _tex32to16) \
DECL_TEXFOGPS(num, _tex16to8h)
DECL_TEXFOGPS_(0)
DECL_TEXFOGPS_(1)
DECL_TEXFOGPS_(2)
DECL_TEXFOGPS_(3)
DECL_TEXFOGPS_(4)
DECL_TEXFOGPS_(5)
//-------------------------------------------------------
// Techniques not related to the main primitive commands
half4 BilinearBitBlt(float2 tex0)
{
float4 ftex;
float2 ffrac;
ffrac.xy = fract(tex0*g_fRealTexDims.xy);
ftex.xy = tex0 - ffrac.xy * g_fRealTexDims.zw;
ftex.zw = ftex.xy + g_fRealTexDims.zw;
float4 off0, off1;
ps2memcoord4_fast(ftex, off0, off1);
half4 c0 = texture(g_sMemory, off0.xy);
half4 c1 = texture(g_sMemory, off0.zw);
half4 c2 = texture(g_sMemory, off1.xy);
half4 c3 = texture(g_sMemory, off1.zw);
return mix( mix(c0, c1, vec4(ffrac.x)), mix(c2, c3, vec4(ffrac.x)), vec4(ffrac.y) );
}
void BitBltPS() {
FragData0 = texture(g_sMemory, ps2memcoord(PSin.tex.xy))*g_fOneColor.xxxy;
}
// used when AA
void BitBltAAPS() {
FragData0 = BilinearBitBlt(PSin.tex.xy) * g_fOneColor.xxxy;
}
void BitBltDepthPS() {
vec4 data;
data = texture(g_sMemory, ps2memcoord(PSin.tex.xy));
FragData0 = data + g_fZBias.y;
#ifdef NO_LOGZ
gl_FragDepth = dot(data, g_fBitBltZ);
#else
gl_FragDepth = log(g_fc0.y + dot(data, g_fBitBltZ)) * g_fOneColor.w;
#endif
}
void BitBltDepthMRTPS() {
vec4 data;
data = texture(g_sMemory, ps2memcoord(PSin.tex.xy));
FragData0 = data + g_fZBias.y;
FragData1.x = g_fc0.x;
#ifdef NO_LOGZ
gl_FragDepth = dot(data, g_fBitBltZ);
#else
gl_FragDepth = log(g_fc0.y + dot(data, g_fBitBltZ)) * g_fOneColor.w;
#endif
}
// static const float BlurKernel[9] = {
// 0.027601,
// 0.066213,
// 0.123701,
// 0.179952,
// 0.205065,
// 0.179952,
// 0.123701,
// 0.066213,
// 0.027601
// };
half4 BilinearFloat16(float2 tex0)
{
return texture(g_sSrcFinal, tex0.xy);
}
void CRTCTargInterPS() {
float finter = texture(g_sInterlace, PSin.Z.yy).x * g_fOneColor.z + g_fOneColor.w + g_fc0.w;
float4 c = BilinearFloat16(PSin.tex.xy);
c.w = ( g_fc0.w*c.w * g_fOneColor.x + g_fOneColor.y ) * finter;
FragData0 = c;
}
void CRTCTargPS() {
float4 c = BilinearFloat16(PSin.tex.xy);
// FIXME DEBUG: to validate tex coord on blit
//c = vec4(PSin.tex.x/512.0f, PSin.tex.y/512.0f, 0.0, 1.0);
c.w = g_fc0.w * c.w * g_fOneColor.x + g_fOneColor.y;
FragData0 = c;
}
void CRTCInterPS() {
float finter = texture(g_sInterlace, PSin.Z.yy).x * g_fOneColor.z + g_fOneColor.w + g_fc0.w;
float2 filtcoord = trunc(PSin.tex.xy) * g_fInvTexDims.xy + g_fInvTexDims.zw;
half4 c = BilinearBitBlt(filtcoord);
c.w = (c.w * g_fOneColor.x + g_fOneColor.y)*finter;
FragData0 = c;
}
// simpler
void CRTCInterPS_Nearest() {
float finter = texture(g_sInterlace, PSin.Z.yy).x * g_fOneColor.z + g_fOneColor.w + g_fc0.w;
half4 c = texture(g_sMemory, ps2memcoord(PSin.tex.xy));
c.w = (c.w * g_fOneColor.x + g_fOneColor.y)*finter;
FragData0 = c;
}
void CRTCPS() {
float2 filtcoord = PSin.tex.xy * g_fInvTexDims.xy+g_fInvTexDims.zw;
half4 c = BilinearBitBlt(filtcoord);
c.w = c.w * g_fOneColor.x + g_fOneColor.y;
FragData0 = c;
}
// simpler
void CRTCPS_Nearest() {
half4 c = texture(g_sMemory, ps2memcoord(PSin.tex.xy));
c.w = c.w * g_fOneColor.x + g_fOneColor.y;
FragData0 = c;
}
void CRTC24InterPS() {
float finter = texture(g_sInterlace, PSin.Z.yy).x * g_fOneColor.z + g_fOneColor.w + g_fc0.w;
float2 filtcoord = trunc(PSin.tex.xy) * g_fInvTexDims.xy + g_fInvTexDims.zw;
half4 c = texture(g_sMemory, ps2memcoord(filtcoord));
c.w = (c.w * g_fOneColor.x + g_fOneColor.y)*finter;
FragData0 = c;
}
void CRTC24PS() {
float2 filtcoord = trunc(PSin.tex.xy) * g_fInvTexDims.xy + g_fInvTexDims.zw;
half4 c = texture(g_sMemory, ps2memcoord(filtcoord));
c.w = c.w * g_fOneColor.x + g_fOneColor.y;
FragData0 = c;
}
void ZeroPS() {
FragData0 = g_fOneColor;
}
void ZeroDebugPS() {
FragData0 = vec4(PSin.tex.x, PSin.tex.y, PSin.tex.z, 0.5);
}
void ZeroDebug2PS() {
vec2 xy = ps2memcoord(fract(PSin.tex.xy/PSin.tex.z)) * vec2(1/4096.0f, 1/48.0f);
FragData0 = vec4(xy.x, xy.y, 0.0, 0.5);
}
void ZeroDebug3PS() {
//FragData0 = vec4(PSin.position.x/2.0f + 0.5f, PSin.position.y/2.0f + 0.5f, 1.0, 0.5);
}
void BaseTexturePS() {
FragData0 = texture(g_sSrcFinal, PSin.tex.xy) * g_fOneColor;
}
void Convert16to32PS() {
float4 final;
float2 ffrac = mod ( PSin.tex.xy + g_fTexDims.zw, g_fTexOffset.xy);
float2 tex0 = g_fTexDims.xy * PSin.tex.xy - ffrac * g_fc0.yw;
if (ffrac.x > g_fTexOffset.x*g_fc0.w)
tex0.x += g_fTexOffset.x*g_fc0.w;
if (tex0.x >= g_fc0.y)
tex0 += g_fTexOffset.zw;
float4 lower = texture(g_sSrcFinal, tex0);
float4 upper = texture(g_sSrcFinal, tex0 + g_fPageOffset.xy);
final.zy = texture(g_sConv32to16, lower.zyx).xy + lower.ww*g_fPageOffset.zw;
final.xw = texture(g_sConv32to16, upper.zyx).xy + upper.ww*g_fPageOffset.zw;
FragData0= final;
}
// use when texture is not tiled and converting from 32bit to 16bit
// don't convert on the block level, only on the column level
// so every other 8 pixels, use the upper bits instead of lower
void Convert32to16PS() {
bool upper = false;
float2 ffrac = mod(PSin.tex.xy + g_fTexDims.zw, g_fTexOffset.xy);
float2 tex0 = g_fc0.ww * (PSin.tex.xy + ffrac);
if( ffrac.x > g_fTexOffset.z ) {
tex0.x -= g_fTexOffset.z;
upper = true;
}
if( ffrac.y >= g_fTexOffset.w ) {
tex0.y -= g_fTexOffset.w;
tex0.x += g_fc0.w;
}
half4 color = texture(g_sSrcFinal, tex0*g_fTexDims.xy)*g_fc0.yyyw;
float2 uv = upper ? color.xw : color.zy;
FragData0 = texture(g_sConv16to32, uv*g_fPageOffset.xy+g_fPageOffset.zw)*g_fTexDims.xxxy;
}
#endif //FRAGMENT_SHADER
#ifdef VERTEX_SHADER
void SetColor() {
VSout.color = Color;
}
void SetTex() {
#ifdef PERSPECTIVE_CORRECT_TEX
VSout.tex = TexCoord;
#else
VSout.tex.xy = TexCoord.xy/TexCoord.z;
#endif
}
void SetZ() {
#ifdef WRITE_DEPTH
VSout.Z = SecondaryColor * g_fZBias.x + g_fZBias.y;
VSout.Z.w = 1.0f;
#endif
}
void SetPosition() {
float4 position;
position.xy = vec2(Vert.xy) * g_fPosXY.xy + g_fPosXY.zw;
// FIXME: the factor in normal mode seem bogus. They don't have same order than in log mode. Or I failed to understand the logic
//// normal mode.
// -> dot(g_fZ, SecondaryColor.zyxw)) => reconstruct a float from normalized char. The float range from 0 to 1
// position.z = dot(g_fZ, SecondaryColor.zyxw);
//// logz mode
// -> dot(g_fZ, SecondaryColor.zyxw)) => reconstruct a float from normalized char. The float range from 0 to 2**32
// position.z = log(g_fc0.y + dot(g_fZ, SecondaryColor.zyxw)) * g_fZNorm.x
// position.z = log(1 + Z_INT) * 0.999f / (32 * log(2.0)) = log2(1 + Z_INT) * 0.999f / 32
// log2(...) will range from 0 to 32
// position.z = (log(g_fc0.y + dot(g_fZ, SecondaryColor.zyxw)) * g_fZNorm.x + g_fZNorm.y) * g_fZMin.y + dot(g_fZ, SecondaryColor.zyxw) * g_fZMin.x ;
#ifdef NO_LOGZ
position.z = dot(g_fZ, SecondaryColor.zyxw);
#else
position.z = log2(1.0f + dot(g_fZ, SecondaryColor.zyxw)) * 0.999f/32.0f;
#endif
position.w = 1.0f;
gl_Position = position;
}
void SetFog() {
VSout.fog = float(Vert.z) * g_fBilinear.w;
}
// just smooth shadering
void RegularVS() {
SetPosition();
SetColor();
SetZ();
}
// diffuse texture mapping
void TextureVS() {
SetPosition();
SetColor();
SetTex();
SetZ();
}
void RegularFogVS() {
SetPosition();
SetColor();
SetZ();
SetFog();
}
void TextureFogVS() {
SetPosition();
SetColor();
SetTex();
SetZ();
SetFog();
}
void BitBltVS() {
vec4 position;
position.xy = vec2(Vert.xy) * g_fBitBltPos.xy + g_fBitBltPos.zw;
position.zw = g_fc0.xy;
gl_Position = position;
VSout.tex.xy = TexCoord.xy * g_fBitBltTex.xy + g_fBitBltTex.zw;
VSout.Z.xy = position.xy * g_fBitBltTrans.xy + g_fBitBltTrans.zw;
}
#endif