dlls/wined3d/pixelshader.c

   1 /*
   2  * shaders implementation
   3  *
   4  * Copyright 2002-2003 Jason Edmeades
   5  * Copyright 2002-2003 Raphael Junqueira
   6  * Copyright 2004 Christian Costa
   7  * Copyright 2005 Oliver Stieber
   8  * Copyright 2006 Ivan Gyurdiev
   9  * Copyright 2007-2008 Stefan Dösinger for CodeWeavers
  10  *
  11  * This library is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2.1 of the License, or (at your option) any later version.
  15  *
  16  * This library is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with this library; if not, write to the Free Software
  23  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  24  */
  25
  26 #include "config.h"
  27
  28 #include <math.h>
  29 #include <stdio.h>
  30
  31 #include "wined3d_private.h"
  32
  33 WINE_DEFAULT_DEBUG_CHANNEL(d3d_shader);
  34
  35 #define GLINFO_LOCATION ((IWineD3DDeviceImpl *) This->baseShader.device)->adapter->gl_info
  36
  37 #define GLNAME_REQUIRE_GLSL  ((const char *)1)
  38
  39 static HRESULT  WINAPI IWineD3DPixelShaderImpl_QueryInterface(IWineD3DPixelShader *iface, REFIID riid, LPVOID *ppobj) {
  40     return IWineD3DBaseShaderImpl_QueryInterface((IWineD3DBaseShader *) iface, riid, ppobj);
  41 }
  42
  43 static ULONG  WINAPI IWineD3DPixelShaderImpl_AddRef(IWineD3DPixelShader *iface) {
  44     return IWineD3DBaseShaderImpl_AddRef((IWineD3DBaseShader *) iface);
  45 }
  46
  47 static ULONG  WINAPI IWineD3DPixelShaderImpl_Release(IWineD3DPixelShader *iface) {
  48     return IWineD3DBaseShaderImpl_Release((IWineD3DBaseShader *) iface);
  49 }
  50
  51 /* *******************************************
  52    IWineD3DPixelShader IWineD3DPixelShader parts follow
  53    ******************************************* */
  54
  55 static HRESULT  WINAPI IWineD3DPixelShaderImpl_GetParent(IWineD3DPixelShader *iface, IUnknown** parent){
  56     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
  57
  58     *parent = This->parent;
  59     IUnknown_AddRef(*parent);
  60     TRACE("(%p) : returning %p\n", This, *parent);
  61     return WINED3D_OK;
  62 }
  63
  64 static HRESULT  WINAPI IWineD3DPixelShaderImpl_GetDevice(IWineD3DPixelShader* iface, IWineD3DDevice **pDevice){
  65     IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)iface;
  66     IWineD3DDevice_AddRef(This->baseShader.device);
  67     *pDevice = This->baseShader.device;
  68     TRACE("(%p) returning %p\n", This, *pDevice);
  69     return WINED3D_OK;
  70 }
  71
  72
  73 static HRESULT  WINAPI IWineD3DPixelShaderImpl_GetFunction(IWineD3DPixelShader* impl, VOID* pData, UINT* pSizeOfData) {
  74   IWineD3DPixelShaderImpl *This = (IWineD3DPixelShaderImpl *)impl;
  75   TRACE("(%p) : pData(%p), pSizeOfData(%p)\n", This, pData, pSizeOfData);
  76
  77   if (NULL == pData) {
  78     *pSizeOfData = This->baseShader.functionLength;
  79     return WINED3D_OK;
  80   }
  81   if (*pSizeOfData < This->baseShader.functionLength) {
  82     /* MSDN claims (for d3d8 at least) that if *pSizeOfData is smaller
  83      * than the required size we should write the required size and
  84      * return D3DERR_MOREDATA. That's not actually true. */
  85     return WINED3DERR_INVALIDCALL;
  86   }
  87   if (NULL == This->baseShader.function) { /* no function defined */
  88     TRACE("(%p) : GetFunction no User Function defined using NULL to %p\n", This, pData);
  89     (*(DWORD **) pData) = NULL;
  90   } else {
  91     if (This->baseShader.functionLength == 0) {
  92
  93     }
  94     TRACE("(%p) : GetFunction copying to %p\n", This, pData);
  95     memcpy(pData, This->baseShader.function, This->baseShader.functionLength);
  96   }
  97   return WINED3D_OK;
  98 }
  99
 100 CONST SHADER_OPCODE IWineD3DPixelShaderImpl_shader_ins[] = {
 101     /* Arithmetic */
 102     {WINED3DSIO_NOP,          "nop",          "NOP",               0, 0, WINED3DSIH_NOP,          0,                      0                     },
 103     {WINED3DSIO_MOV,          "mov",          "MOV",               1, 2, WINED3DSIH_MOV,          0,                      0                     },
 104     {WINED3DSIO_ADD,          "add",          "ADD",               1, 3, WINED3DSIH_ADD,          0,                      0                     },
 105     {WINED3DSIO_SUB,          "sub",          "SUB",               1, 3, WINED3DSIH_SUB,          0,                      0                     },
 106     {WINED3DSIO_MAD,          "mad",          "MAD",               1, 4, WINED3DSIH_MAD,          0,                      0                     },
 107     {WINED3DSIO_MUL,          "mul",          "MUL",               1, 3, WINED3DSIH_MUL,          0,                      0                     },
 108     {WINED3DSIO_RCP,          "rcp",          "RCP",               1, 2, WINED3DSIH_RCP,          0,                      0                     },
 109     {WINED3DSIO_RSQ,          "rsq",          "RSQ",               1, 2, WINED3DSIH_RSQ,          0,                      0                     },
 110     {WINED3DSIO_DP3,          "dp3",          "DP3",               1, 3, WINED3DSIH_DP3,          0,                      0                     },
 111     {WINED3DSIO_DP4,          "dp4",          "DP4",               1, 3, WINED3DSIH_DP4,          0,                      0                     },
 112     {WINED3DSIO_MIN,          "min",          "MIN",               1, 3, WINED3DSIH_MIN,          0,                      0                     },
 113     {WINED3DSIO_MAX,          "max",          "MAX",               1, 3, WINED3DSIH_MAX,          0,                      0                     },
 114     {WINED3DSIO_SLT,          "slt",          "SLT",               1, 3, WINED3DSIH_SLT,          0,                      0                     },
 115     {WINED3DSIO_SGE,          "sge",          "SGE",               1, 3, WINED3DSIH_SGE,          0,                      0                     },
 116     {WINED3DSIO_ABS,          "abs",          "ABS",               1, 2, WINED3DSIH_ABS,          0,                      0                     },
 117     {WINED3DSIO_EXP,          "exp",          "EX2",               1, 2, WINED3DSIH_EXP,          0,                      0                     },
 118     {WINED3DSIO_LOG,          "log",          "LG2",               1, 2, WINED3DSIH_LOG,          0,                      0                     },
 119     {WINED3DSIO_EXPP,         "expp",         "EXP",               1, 2, WINED3DSIH_EXPP,         0,                      0                     },
 120     {WINED3DSIO_LOGP,         "logp",         "LOG",               1, 2, WINED3DSIH_LOGP,         0,                      0                     },
 121     {WINED3DSIO_DST,          "dst",          "DST",               1, 3, WINED3DSIH_DST,          0,                      0                     },
 122     {WINED3DSIO_LRP,          "lrp",          "LRP",               1, 4, WINED3DSIH_LRP,          0,                      0                     },
 123     {WINED3DSIO_FRC,          "frc",          "FRC",               1, 2, WINED3DSIH_FRC,          0,                      0                     },
 124     {WINED3DSIO_CND,          "cnd",          NULL,                1, 4, WINED3DSIH_CND,          WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,4)},
 125     {WINED3DSIO_CMP,          "cmp",          NULL,                1, 4, WINED3DSIH_CMP,          WINED3DPS_VERSION(1,2), WINED3DPS_VERSION(3,0)},
 126     {WINED3DSIO_POW,          "pow",          "POW",               1, 3, WINED3DSIH_POW,          0,                      0                     },
 127     {WINED3DSIO_CRS,          "crs",          "XPD",               1, 3, WINED3DSIH_CRS,          0,                      0                     },
 128     {WINED3DSIO_NRM,          "nrm",          NULL,                1, 2, WINED3DSIH_NRM,          0,                      0                     },
 129     {WINED3DSIO_SINCOS,       "sincos",       NULL,                1, 4, WINED3DSIH_SINCOS,       WINED3DPS_VERSION(2,0), WINED3DPS_VERSION(2,1)},
 130     {WINED3DSIO_SINCOS,       "sincos",       "SCS",               1, 2, WINED3DSIH_SINCOS,       WINED3DPS_VERSION(3,0), -1                    },
 131     {WINED3DSIO_DP2ADD,       "dp2add",       NULL,                1, 4, WINED3DSIH_DP2ADD,       WINED3DPS_VERSION(2,0), -1                    },
 132     /* Matrix */
 133     {WINED3DSIO_M4x4,         "m4x4",         "undefined",         1, 3, WINED3DSIH_M4x4,         0,                      0                     },
 134     {WINED3DSIO_M4x3,         "m4x3",         "undefined",         1, 3, WINED3DSIH_M4x3,         0,                      0                     },
 135     {WINED3DSIO_M3x4,         "m3x4",         "undefined",         1, 3, WINED3DSIH_M3x4,         0,                      0                     },
 136     {WINED3DSIO_M3x3,         "m3x3",         "undefined",         1, 3, WINED3DSIH_M3x3,         0,                      0                     },
 137     {WINED3DSIO_M3x2,         "m3x2",         "undefined",         1, 3, WINED3DSIH_M3x2,         0,                      0                     },
 138     /* Register declarations */
 139     {WINED3DSIO_DCL,          "dcl",          NULL,                0, 2, WINED3DSIH_DCL,          0,                      0                     },
 140     /* Flow control - requires GLSL or software shaders */
 141     {WINED3DSIO_REP ,         "rep",          NULL,                0, 1, WINED3DSIH_REP,          WINED3DPS_VERSION(2,1), -1                    },
 142     {WINED3DSIO_ENDREP,       "endrep",       NULL,                0, 0, WINED3DSIH_ENDREP,       WINED3DPS_VERSION(2,1), -1                    },
 143     {WINED3DSIO_IF,           "if",           NULL,                0, 1, WINED3DSIH_IF,           WINED3DPS_VERSION(2,1), -1                    },
 144     {WINED3DSIO_IFC,          "ifc",          NULL,                0, 2, WINED3DSIH_IFC,          WINED3DPS_VERSION(2,1), -1                    },
 145     {WINED3DSIO_ELSE,         "else",         NULL,                0, 0, WINED3DSIH_ELSE,         WINED3DPS_VERSION(2,1), -1                    },
 146     {WINED3DSIO_ENDIF,        "endif",        NULL,                0, 0, WINED3DSIH_ENDIF,        WINED3DPS_VERSION(2,1), -1                    },
 147     {WINED3DSIO_BREAK,        "break",        NULL,                0, 0, WINED3DSIH_BREAK,        WINED3DPS_VERSION(2,1), -1                    },
 148     {WINED3DSIO_BREAKC,       "breakc",       NULL,                0, 2, WINED3DSIH_BREAKC,       WINED3DPS_VERSION(2,1), -1                    },
 149     {WINED3DSIO_BREAKP,       "breakp",       GLNAME_REQUIRE_GLSL, 0, 1, WINED3DSIH_BREAKP,       0,                      0                     },
 150     {WINED3DSIO_CALL,         "call",         NULL,                0, 1, WINED3DSIH_CALL,         WINED3DPS_VERSION(2,1), -1                    },
 151     {WINED3DSIO_CALLNZ,       "callnz",       NULL,                0, 2, WINED3DSIH_CALLNZ,       WINED3DPS_VERSION(2,1), -1                    },
 152     {WINED3DSIO_LOOP,         "loop",         NULL,                0, 2, WINED3DSIH_LOOP,         WINED3DPS_VERSION(3,0), -1                    },
 153     {WINED3DSIO_RET,          "ret",          NULL,                0, 0, WINED3DSIH_RET,          WINED3DPS_VERSION(2,1), -1                    },
 154     {WINED3DSIO_ENDLOOP,      "endloop",      NULL,                0, 0, WINED3DSIH_ENDLOOP,      WINED3DPS_VERSION(3,0), -1                    },
 155     {WINED3DSIO_LABEL,        "label",        NULL,                0, 1, WINED3DSIH_LABEL,        WINED3DPS_VERSION(2,1), -1                    },
 156     /* Constant definitions */
 157     {WINED3DSIO_DEF,          "def",          "undefined",         1, 5, WINED3DSIH_DEF,          0,                      0                     },
 158     {WINED3DSIO_DEFB,         "defb",         GLNAME_REQUIRE_GLSL, 1, 2, WINED3DSIH_DEFB,         0,                      0                     },
 159     {WINED3DSIO_DEFI,         "defi",         GLNAME_REQUIRE_GLSL, 1, 5, WINED3DSIH_DEFI,         0,                      0                     },
 160     /* Texture */
 161     {WINED3DSIO_TEXCOORD,     "texcoord",     "undefined",         1, 1, WINED3DSIH_TEXCOORD,     0,                      WINED3DPS_VERSION(1,3)},
 162     {WINED3DSIO_TEXCOORD,     "texcrd",       "undefined",         1, 2, WINED3DSIH_TEXCOORD,     WINED3DPS_VERSION(1,4), WINED3DPS_VERSION(1,4)},
 163     {WINED3DSIO_TEXKILL,      "texkill",      "KIL",               1, 1, WINED3DSIH_TEXKILL,      WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(3,0)},
 164     {WINED3DSIO_TEX,          "tex",          "undefined",         1, 1, WINED3DSIH_TEX,          0,                      WINED3DPS_VERSION(1,3)},
 165     {WINED3DSIO_TEX,          "texld",        "undefined",         1, 2, WINED3DSIH_TEX,          WINED3DPS_VERSION(1,4), WINED3DPS_VERSION(1,4)},
 166     {WINED3DSIO_TEX,          "texld",        "undefined",         1, 3, WINED3DSIH_TEX,          WINED3DPS_VERSION(2,0), -1                    },
 167     {WINED3DSIO_TEXBEM,       "texbem",       "undefined",         1, 2, WINED3DSIH_TEXBEM,       0,                      WINED3DPS_VERSION(1,3)},
 168     {WINED3DSIO_TEXBEML,      "texbeml",      GLNAME_REQUIRE_GLSL, 1, 2, WINED3DSIH_TEXBEML,      WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,3)},
 169     {WINED3DSIO_TEXREG2AR,    "texreg2ar",    "undefined",         1, 2, WINED3DSIH_TEXREG2AR,    WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,3)},
 170     {WINED3DSIO_TEXREG2GB,    "texreg2gb",    "undefined",         1, 2, WINED3DSIH_TEXREG2GB,    WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,3)},
 171     {WINED3DSIO_TEXREG2RGB,   "texreg2rgb",   "undefined",         1, 2, WINED3DSIH_TEXREG2RGB,   WINED3DPS_VERSION(1,2), WINED3DPS_VERSION(1,3)},
 172     {WINED3DSIO_TEXM3x2PAD,   "texm3x2pad",   "undefined",         1, 2, WINED3DSIH_TEXM3x2PAD,   WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,3)},
 173     {WINED3DSIO_TEXM3x2TEX,   "texm3x2tex",   "undefined",         1, 2, WINED3DSIH_TEXM3x2TEX,   WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,3)},
 174     {WINED3DSIO_TEXM3x3PAD,   "texm3x3pad",   "undefined",         1, 2, WINED3DSIH_TEXM3x3PAD,   WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,3)},
 175     {WINED3DSIO_TEXM3x3DIFF,  "texm3x3diff",  GLNAME_REQUIRE_GLSL, 1, 2, WINED3DSIH_TEXM3x3DIFF,  WINED3DPS_VERSION(0,0), WINED3DPS_VERSION(0,0)},
 176     {WINED3DSIO_TEXM3x3SPEC,  "texm3x3spec",  "undefined",         1, 3, WINED3DSIH_TEXM3x3SPEC,  WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,3)},
 177     {WINED3DSIO_TEXM3x3VSPEC, "texm3x3vspec", "undefined",         1, 2, WINED3DSIH_TEXM3x3VSPEC, WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,3)},
 178     {WINED3DSIO_TEXM3x3TEX,   "texm3x3tex",   "undefined",         1, 2, WINED3DSIH_TEXM3x3TEX,   WINED3DPS_VERSION(1,0), WINED3DPS_VERSION(1,3)},
 179     {WINED3DSIO_TEXDP3TEX,    "texdp3tex",    NULL,                1, 2, WINED3DSIH_TEXDP3TEX,    WINED3DPS_VERSION(1,2), WINED3DPS_VERSION(1,3)},
 180     {WINED3DSIO_TEXM3x2DEPTH, "texm3x2depth", GLNAME_REQUIRE_GLSL, 1, 2, WINED3DSIH_TEXM3x2DEPTH, WINED3DPS_VERSION(1,3), WINED3DPS_VERSION(1,3)},
 181     {WINED3DSIO_TEXDP3,       "texdp3",       NULL,                1, 2, WINED3DSIH_TEXDP3,       WINED3DPS_VERSION(1,2), WINED3DPS_VERSION(1,3)},
 182     {WINED3DSIO_TEXM3x3,      "texm3x3",      NULL,                1, 2, WINED3DSIH_TEXM3x3,      WINED3DPS_VERSION(1,2), WINED3DPS_VERSION(1,3)},
 183     {WINED3DSIO_TEXDEPTH,     "texdepth",     NULL,                1, 1, WINED3DSIH_TEXDEPTH,     WINED3DPS_VERSION(1,4), WINED3DPS_VERSION(1,4)},
 184     {WINED3DSIO_BEM,          "bem",          "undefined",         1, 3, WINED3DSIH_BEM,          WINED3DPS_VERSION(1,4), WINED3DPS_VERSION(1,4)},
 185     {WINED3DSIO_DSX,          "dsx",          NULL,                1, 2, WINED3DSIH_DSX,          WINED3DPS_VERSION(2,1), -1                    },
 186     {WINED3DSIO_DSY,          "dsy",          NULL,                1, 2, WINED3DSIH_DSY,          WINED3DPS_VERSION(2,1), -1                    },
 187     {WINED3DSIO_TEXLDD,       "texldd",       GLNAME_REQUIRE_GLSL, 1, 5, WINED3DSIH_TEXLDD,       WINED3DPS_VERSION(2,1), -1                    },
 188     {WINED3DSIO_SETP,         "setp",         GLNAME_REQUIRE_GLSL, 1, 3, WINED3DSIH_SETP,         0,                      0                     },
 189     {WINED3DSIO_TEXLDL,       "texldl",       NULL,                1, 3, WINED3DSIH_TEXLDL,       WINED3DPS_VERSION(3,0), -1                    },
 190     {WINED3DSIO_PHASE,        "phase",        GLNAME_REQUIRE_GLSL, 0, 0, WINED3DSIH_PHASE,        0,                      0                     },
 191     {0,                       NULL,           NULL,                0, 0, 0,                       0,                      0                     }
 192 };
 193
 194 static void pshader_set_limits(
 195       IWineD3DPixelShaderImpl *This) {
 196
 197       This->baseShader.limits.attributes = 0;
 198       This->baseShader.limits.address = 0;
 199       This->baseShader.limits.packed_output = 0;
 200
 201       switch (This->baseShader.hex_version) {
 202           case WINED3DPS_VERSION(1,0):
 203           case WINED3DPS_VERSION(1,1):
 204           case WINED3DPS_VERSION(1,2):
 205           case WINED3DPS_VERSION(1,3):
 206                    This->baseShader.limits.temporary = 2;
 207                    This->baseShader.limits.constant_float = 8;
 208                    This->baseShader.limits.constant_int = 0;
 209                    This->baseShader.limits.constant_bool = 0;
 210                    This->baseShader.limits.texcoord = 4;
 211                    This->baseShader.limits.sampler = 4;
 212                    This->baseShader.limits.packed_input = 0;
 213                    This->baseShader.limits.label = 0;
 214                    break;
 215
 216           case WINED3DPS_VERSION(1,4):
 217                    This->baseShader.limits.temporary = 6;
 218                    This->baseShader.limits.constant_float = 8;
 219                    This->baseShader.limits.constant_int = 0;
 220                    This->baseShader.limits.constant_bool = 0;
 221                    This->baseShader.limits.texcoord = 6;
 222                    This->baseShader.limits.sampler = 6;
 223                    This->baseShader.limits.packed_input = 0;
 224                    This->baseShader.limits.label = 0;
 225                    break;
 226
 227           /* FIXME: temporaries must match D3DPSHADERCAPS2_0.NumTemps */
 228           case WINED3DPS_VERSION(2,0):
 229                    This->baseShader.limits.temporary = 32;
 230                    This->baseShader.limits.constant_float = 32;
 231                    This->baseShader.limits.constant_int = 16;
 232                    This->baseShader.limits.constant_bool = 16;
 233                    This->baseShader.limits.texcoord = 8;
 234                    This->baseShader.limits.sampler = 16;
 235                    This->baseShader.limits.packed_input = 0;
 236                    break;
 237
 238           case WINED3DPS_VERSION(2,1):
 239                    This->baseShader.limits.temporary = 32;
 240                    This->baseShader.limits.constant_float = 32;
 241                    This->baseShader.limits.constant_int = 16;
 242                    This->baseShader.limits.constant_bool = 16;
 243                    This->baseShader.limits.texcoord = 8;
 244                    This->baseShader.limits.sampler = 16;
 245                    This->baseShader.limits.packed_input = 0;
 246                    This->baseShader.limits.label = 16;
 247                    break;
 248
 249           case WINED3DPS_VERSION(3,0):
 250                    This->baseShader.limits.temporary = 32;
 251                    This->baseShader.limits.constant_float = 224;
 252                    This->baseShader.limits.constant_int = 16;
 253                    This->baseShader.limits.constant_bool = 16;
 254                    This->baseShader.limits.texcoord = 0;
 255                    This->baseShader.limits.sampler = 16;
 256                    This->baseShader.limits.packed_input = 12;
 257                    This->baseShader.limits.label = 16; /* FIXME: 2048 */
 258                    break;
 259
 260           default: This->baseShader.limits.temporary = 32;
 261                    This->baseShader.limits.constant_float = 32;
 262                    This->baseShader.limits.constant_int = 16;
 263                    This->baseShader.limits.constant_bool = 16;
 264                    This->baseShader.limits.texcoord = 8;
 265                    This->baseShader.limits.sampler = 16;
 266                    This->baseShader.limits.packed_input = 0;
 267                    This->baseShader.limits.label = 0;
 268                    FIXME("Unrecognized pixel shader version %#x\n",
 269                        This->baseShader.hex_version);
 270       }
 271 }
 272
 273 /** Generate a pixel shader string using either GL_FRAGMENT_PROGRAM_ARB
 274     or GLSL and send it to the card */
 275 static inline GLuint IWineD3DPixelShaderImpl_GenerateShader(
 276     IWineD3DPixelShaderImpl *This) {
 277     SHADER_BUFFER buffer;
 278     GLuint shader;
 279
 280     buffer.buffer = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, SHADER_PGMSIZE);
 281     buffer.bsize = 0;
 282     buffer.lineNo = 0;
 283     buffer.newline = TRUE;
 284
 285     shader = ((IWineD3DDeviceImpl *)This->baseShader.device)->shader_backend->shader_generate_pshader((IWineD3DPixelShader *)This, &buffer);
 286
 287     HeapFree(GetProcessHeap(), 0, buffer.buffer);
 288
 289     return shader;
 290 }
 291
 292 static HRESULT WINAPI IWineD3DPixelShaderImpl_SetFunction(IWineD3DPixelShader *iface, CONST DWORD *pFunction) {
 293
 294     IWineD3DPixelShaderImpl *This =(IWineD3DPixelShaderImpl *)iface;
 295     IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl *) This->baseShader.device;
 296
 297     TRACE("(%p) : pFunction %p\n", iface, pFunction);
 298
 299     /* First pass: trace shader */
 300     shader_trace_init((IWineD3DBaseShader*) This, pFunction);
 301     pshader_set_limits(This);
 302
 303     /* Initialize immediate constant lists */
 304     list_init(&This->baseShader.constantsF);
 305     list_init(&This->baseShader.constantsB);
 306     list_init(&This->baseShader.constantsI);
 307
 308     if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) > 1) {
 309         shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
 310         HRESULT hr;
 311         unsigned int i, j, highest_reg_used = 0, num_regs_used = 0;
 312
 313         /* Second pass: figure out which registers are used, what the semantics are, etc.. */
 314         memset(reg_maps, 0, sizeof(shader_reg_maps));
 315         hr = shader_get_registers_used((IWineD3DBaseShader*) This, reg_maps,
 316             This->semantics_in, NULL, pFunction, NULL);
 317         if (FAILED(hr)) return hr;
 318         /* FIXME: validate reg_maps against OpenGL */
 319
 320         for(i = 0; i < MAX_REG_INPUT; i++) {
 321             if(This->input_reg_used[i]) {
 322                 num_regs_used++;
 323                 highest_reg_used = i;
 324             }
 325         }
 326
 327         /* Don't do any register mapping magic if it is not needed, or if we can't
 328          * achieve anything anyway
 329          */
 330         if(highest_reg_used < (GL_LIMITS(glsl_varyings) / 4) ||
 331            num_regs_used > (GL_LIMITS(glsl_varyings) / 4) ) {
 332             if(num_regs_used > (GL_LIMITS(glsl_varyings) / 4)) {
 333                 /* This happens with relative addressing. The input mapper function
 334                  * warns about this if the higher registers are declared too, so
 335                  * don't write a FIXME here
 336                  */
 337                 WARN("More varying registers used than supported\n");
 338             }
 339
 340             for(i = 0; i < MAX_REG_INPUT; i++) {
 341                 This->input_reg_map[i] = i;
 342             }
 343             This->declared_in_count = highest_reg_used + 1;
 344         } else {
 345             j = 0;
 346             for(i = 0; i < MAX_REG_INPUT; i++) {
 347                 if(This->input_reg_used[i]) {
 348                     This->input_reg_map[i] = j;
 349                     j++;
 350                 } else {
 351                     This->input_reg_map[i] = -1;
 352                 }
 353             }
 354             This->declared_in_count = j;
 355         }
 356     }
 357     This->baseShader.load_local_constsF = FALSE;
 358
 359     This->baseShader.shader_mode = deviceImpl->ps_selected_mode;
 360
 361     TRACE("(%p) : Copying the function\n", This);
 362     if (NULL != pFunction) {
 363         void *function;
 364
 365         function = HeapAlloc(GetProcessHeap(), HEAP_ZERO_MEMORY, This->baseShader.functionLength);
 366         if (!function) return E_OUTOFMEMORY;
 367         memcpy(function, pFunction, This->baseShader.functionLength);
 368         This->baseShader.function = function;
 369     } else {
 370         This->baseShader.function = NULL;
 371     }
 372
 373     return WINED3D_OK;
 374 }
 375
 376 static GLuint pixelshader_compile(IWineD3DPixelShaderImpl *This, const struct ps_compile_args *args)
 377 {
 378     CONST DWORD *function = This->baseShader.function;
 379     HRESULT hr;
 380     GLuint retval;
 381
 382     TRACE("(%p) : function %p\n", This, function);
 383
 384     hr = IWineD3DPixelShader_UpdateSamplers((IWineD3DPixelShader *) This);
 385     if(FAILED(hr)) {
 386         ERR("Failed to update sampler information\n");
 387         return 0;
 388     }
 389
 390     /* Reset fields tracking stateblock values being hardcoded in the shader */
 391     This->baseShader.num_sampled_samplers = 0;
 392
 393     /* Generate the HW shader */
 394     TRACE("(%p) : Generating hardware program\n", This);
 395     retval = IWineD3DPixelShaderImpl_GenerateShader(This);
 396
 397     This->baseShader.is_compiled = TRUE;
 398
 399     return retval;
 400 }
 401
 402 static HRESULT WINAPI IWineD3DPixelShaderImpl_UpdateSamplers(IWineD3DPixelShader *iface) {
 403     IWineD3DPixelShaderImpl *This =(IWineD3DPixelShaderImpl *)iface;
 404
 405     if (WINED3DSHADER_VERSION_MAJOR(This->baseShader.hex_version) == 1) {
 406         IWineD3DDeviceImpl *deviceImpl = (IWineD3DDeviceImpl*) This->baseShader.device;
 407         shader_reg_maps *reg_maps = &This->baseShader.reg_maps;
 408         HRESULT hr;
 409
 410         /* Second pass: figure out which registers are used, what the semantics are, etc.. */
 411         memset(reg_maps, 0, sizeof(shader_reg_maps));
 412         hr = shader_get_registers_used((IWineD3DBaseShader*)This, reg_maps,
 413                                         This->semantics_in, NULL, This->baseShader.function, deviceImpl->stateBlock);
 414         return hr;
 415         /* FIXME: validate reg_maps against OpenGL */
 416     } else {
 417         return WINED3D_OK;
 418     }
 419 }
 420
 421 const IWineD3DPixelShaderVtbl IWineD3DPixelShader_Vtbl =
 422 {
 423     /*** IUnknown methods ***/
 424     IWineD3DPixelShaderImpl_QueryInterface,
 425     IWineD3DPixelShaderImpl_AddRef,
 426     IWineD3DPixelShaderImpl_Release,
 427     /*** IWineD3DBase methods ***/
 428     IWineD3DPixelShaderImpl_GetParent,
 429     /*** IWineD3DBaseShader methods ***/
 430     IWineD3DPixelShaderImpl_SetFunction,
 431     /*** IWineD3DPixelShader methods ***/
 432     IWineD3DPixelShaderImpl_UpdateSamplers,
 433     IWineD3DPixelShaderImpl_GetDevice,
 434     IWineD3DPixelShaderImpl_GetFunction
 435 };
 436
 437 void find_ps_compile_args(IWineD3DPixelShaderImpl *shader, IWineD3DStateBlockImpl *stateblock, struct ps_compile_args *args) {
 438     UINT i, sampler;
 439     IWineD3DBaseTextureImpl *tex;
 440
 441     args->srgb_correction = stateblock->renderState[WINED3DRS_SRGBWRITEENABLE] ? 1 : 0;
 442
 443     memset(args->color_fixup, 0, sizeof(args->color_fixup));
 444     for(i = 0; i < shader->baseShader.num_sampled_samplers; i++) {
 445         sampler = shader->baseShader.sampled_samplers[i];
 446         tex = (IWineD3DBaseTextureImpl *) stateblock->textures[sampler];
 447         if(!tex) {
 448             args->color_fixup[sampler] = COLOR_FIXUP_IDENTITY;
 449             continue;
 450         }
 451         args->color_fixup[sampler] = tex->baseTexture.shader_color_fixup;
 452     }
 453     if(shader->baseShader.hex_version >= WINED3DPS_VERSION(3,0)) {
 454         if(((IWineD3DDeviceImpl *) shader->baseShader.device)->strided_streams.u.s.position_transformed) {
 455             args->vp_mode = pretransformed;
 456         } else if(use_vs((IWineD3DDeviceImpl *) shader->baseShader.device)) {
 457             args->vp_mode = vertexshader;
 458         } else {
 459             args->vp_mode = fixedfunction;
 460         }
 461     } else {
 462         args->vp_mode = vertexshader;
 463     }
 464 }
 465
 466 GLuint find_gl_pshader(IWineD3DPixelShaderImpl *shader, const struct ps_compile_args *args)
 467 {
 468     UINT i;
 469     struct ps_compiled_shader *old_array;
 470
 471     /* Usually we have very few GL shaders for each d3d shader(just 1 or maybe 2),
 472      * so a linear search is more performant than a hashmap
 473      */
 474     for(i = 0; i < shader->num_gl_shaders; i++) {
 475         if(memcmp(&shader->gl_shaders[i].args, args, sizeof(*args)) == 0) {
 476             return shader->gl_shaders[i].prgId;
 477         }
 478     }
 479
 480     TRACE("No matching GL shader found, compiling a new shader\n");
 481     old_array = shader->gl_shaders;
 482     if(old_array) {
 483         shader->gl_shaders = HeapReAlloc(GetProcessHeap(), 0, old_array,
 484                                          (shader->num_gl_shaders + 1) * sizeof(*shader->gl_shaders));
 485     } else {
 486         shader->gl_shaders = HeapAlloc(GetProcessHeap(), 0, sizeof(*shader->gl_shaders));
 487     }
 488
 489     if(!shader->gl_shaders) {
 490         ERR("Out of memory\n");
 491         return 0;
 492     }
 493
 494     shader->gl_shaders[shader->num_gl_shaders].args = *args;
 495     shader->gl_shaders[shader->num_gl_shaders].prgId = pixelshader_compile(shader, args);
 496     return shader->gl_shaders[shader->num_gl_shaders++].prgId;
 497 }