src/mesa/drivers/dri/r600/r700_fragprog.c

   1 /*
   2  * Copyright (C) 2008-2009  Advanced Micro Devices, Inc.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice shall be included
  12  * in all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  15  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  18  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  19  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  20  */
  21
  22 /*
  23  * Authors:
  24  *   Richard Li <RichardZ.Li@amd.com>, <richardradeon@gmail.com>
  25  *   CooperYuan <cooper.yuan@amd.com>, <cooperyuan@gmail.com>
  26  */
  27
  28 #include <stdio.h>
  29 #include <stdarg.h>
  30 #include <stdlib.h>
  31 #include <string.h>
  32 #include <math.h>
  33
  34 #include "main/imports.h"
  35 #include "shader/prog_parameter.h"
  36 #include "shader/prog_statevars.h"
  37
  38 #include "r600_context.h"
  39 #include "r600_cmdbuf.h"
  40
  41 #include "r700_fragprog.h"
  42
  43 #include "r700_debug.h"
  44
  45 //TODO : Validate FP input with VP output.
  46 void Map_Fragment_Program(r700_AssemblerBase         *pAsm,
  47                                                   struct gl_fragment_program *mesa_fp)
  48 {
  49         unsigned int unBit;
  50     unsigned int i;
  51     GLuint       ui;
  52
  53         pAsm->number_used_registers = 0;
  54
  55 //Input mapping : mesa_fp->Base.InputsRead set the flag, set in
  56         //The flags parsed in parse_attrib_binding. FRAG_ATTRIB_COLx, FRAG_ATTRIB_TEXx, ...
  57         //MUST match order in Map_Vertex_Output
  58         unBit = 1 << FRAG_ATTRIB_COL0;
  59         if(mesa_fp->Base.InputsRead & unBit)
  60         {
  61                 pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL0] = pAsm->number_used_registers++;
  62         }
  63
  64         unBit = 1 << FRAG_ATTRIB_COL1;
  65         if(mesa_fp->Base.InputsRead & unBit)
  66         {
  67                 pAsm->uiFP_AttributeMap[FRAG_ATTRIB_COL1] = pAsm->number_used_registers++;
  68         }
  69
  70         for(i=0; i<8; i++)
  71         {
  72                 unBit = 1 << (FRAG_ATTRIB_TEX0 + i);
  73                 if(mesa_fp->Base.InputsRead & unBit)
  74                 {
  75                         pAsm->uiFP_AttributeMap[FRAG_ATTRIB_TEX0 + i] = pAsm->number_used_registers++;
  76                 }
  77         }
  78
  79 /* Map temporary registers (GPRs) */
  80     pAsm->starting_temp_register_number = pAsm->number_used_registers;
  81
  82     if(mesa_fp->Base.NumNativeTemporaries >= mesa_fp->Base.NumTemporaries)
  83     {
  84             pAsm->number_used_registers += mesa_fp->Base.NumNativeTemporaries;
  85     }
  86     else
  87     {
  88         pAsm->number_used_registers += mesa_fp->Base.NumTemporaries;
  89     }
  90
  91 /* Output mapping */
  92         pAsm->number_of_exports = 0;
  93         pAsm->number_of_colorandz_exports = 0; /* don't include stencil and mask out. */
  94         pAsm->starting_export_register_number = pAsm->number_used_registers;
  95         unBit = 1 << FRAG_RESULT_COLOR;
  96         if(mesa_fp->Base.OutputsWritten & unBit)
  97         {
  98                 pAsm->uiFP_OutputMap[FRAG_RESULT_COLOR] = pAsm->number_used_registers++;
  99                 pAsm->number_of_exports++;
 100                 pAsm->number_of_colorandz_exports++;
 101         }
 102         unBit = 1 << FRAG_RESULT_DEPTH;
 103         if(mesa_fp->Base.OutputsWritten & unBit)
 104         {
 105         pAsm->depth_export_register_number = pAsm->number_used_registers;
 106                 pAsm->uiFP_OutputMap[FRAG_RESULT_DEPTH] = pAsm->number_used_registers++;
 107                 pAsm->number_of_exports++;
 108                 pAsm->number_of_colorandz_exports++;
 109         }
 110
 111     pAsm->pucOutMask = (unsigned char*) MALLOC(pAsm->number_of_exports);
 112     for(ui=0; ui<pAsm->number_of_exports; ui++)
 113     {
 114         pAsm->pucOutMask[ui] = 0x0;
 115     }
 116
 117         pAsm->uFirstHelpReg = pAsm->number_used_registers;
 118 }
 119
 120 GLboolean Find_Instruction_Dependencies_fp(struct r700_fragment_program *fp,
 121                                                                 struct gl_fragment_program   *mesa_fp)
 122 {
 123     GLuint i, j;
 124     GLint * puiTEMPwrites;
 125     struct prog_instruction * pILInst;
 126     InstDeps         *pInstDeps;
 127     struct prog_instruction * texcoord_DepInst;
 128     GLint              nDepInstID;
 129
 130     puiTEMPwrites = (GLint*) MALLOC(sizeof(GLuint)*mesa_fp->Base.NumTemporaries);
 131     for(i=0; i<mesa_fp->Base.NumTemporaries; i++)
 132     {
 133         puiTEMPwrites[i] = -1;
 134     }
 135
 136     pInstDeps = (InstDeps*)MALLOC(sizeof(InstDeps)*mesa_fp->Base.NumInstructions);
 137
 138     for(i=0; i<mesa_fp->Base.NumInstructions; i++)
 139     {
 140         pInstDeps[i].nDstDep = -1;
 141         pILInst = &(mesa_fp->Base.Instructions[i]);
 142
 143         //Dst
 144         if(pILInst->DstReg.File == PROGRAM_TEMPORARY)
 145         {
 146             //Set lastwrite for the temp
 147             puiTEMPwrites[pILInst->DstReg.Index] = i;
 148         }
 149
 150         //Src
 151         for(j=0; j<3; j++)
 152         {
 153             if(pILInst->SrcReg[j].File == PROGRAM_TEMPORARY)
 154             {
 155                 //Set dep.
 156                 pInstDeps[i].nSrcDeps[j] = puiTEMPwrites[pILInst->SrcReg[j].Index];
 157             }
 158             else
 159             {
 160                 pInstDeps[i].nSrcDeps[j] = -1;
 161             }
 162         }
 163     }
 164
 165     fp->r700AsmCode.pInstDeps = pInstDeps;
 166
 167     FREE(puiTEMPwrites);
 168
 169     //Find dep for tex inst
 170     for(i=0; i<mesa_fp->Base.NumInstructions; i++)
 171     {
 172         pILInst = &(mesa_fp->Base.Instructions[i]);
 173
 174         if(GL_TRUE == IsTex(pILInst->Opcode))
 175         {   //src0 is the tex coord register, src1 is texunit, src2 is textype
 176             nDepInstID = pInstDeps[i].nSrcDeps[0];
 177             if(nDepInstID >= 0)
 178             {
 179                 texcoord_DepInst = &(mesa_fp->Base.Instructions[nDepInstID]);
 180                 if(GL_TRUE == IsAlu(texcoord_DepInst->Opcode) )
 181                 {
 182                     pInstDeps[nDepInstID].nDstDep = i;
 183                     pInstDeps[i].nDstDep = i;
 184                 }
 185                 else if(GL_TRUE == IsTex(texcoord_DepInst->Opcode) )
 186                 {
 187                     pInstDeps[i].nDstDep = i;
 188                 }
 189                 else
 190                 {   //... other deps?
 191                 }
 192             }
 193         }
 194         }
 195
 196     return GL_TRUE;
 197 }
 198
 199 GLboolean r700TranslateFragmentShader(struct r700_fragment_program *fp,
 200                                                              struct gl_fragment_program   *mesa_fp)
 201 {
 202         GLuint    number_of_colors_exported;
 203         GLboolean z_enabled = GL_FALSE;
 204         GLuint    unBit;
 205
 206     //Init_Program
 207         Init_r700_AssemblerBase( SPT_FP, &(fp->r700AsmCode), &(fp->r700Shader) );
 208         Map_Fragment_Program(&(fp->r700AsmCode), mesa_fp);
 209
 210     if( GL_FALSE == Find_Instruction_Dependencies_fp(fp, mesa_fp) )
 211         {
 212                 return GL_FALSE;
 213     }
 214
 215         if( GL_FALSE == AssembleInstr(mesa_fp->Base.NumInstructions,
 216                                   &(mesa_fp->Base.Instructions[0]),
 217                                   &(fp->r700AsmCode)) )
 218         {
 219                 return GL_FALSE;
 220         }
 221
 222     if(GL_FALSE == Process_Fragment_Exports(&(fp->r700AsmCode), mesa_fp->Base.OutputsWritten) )
 223     {
 224         return GL_FALSE;
 225     }
 226
 227     fp->r700Shader.nRegs = (fp->r700AsmCode.number_used_registers == 0) ? 0
 228                          : (fp->r700AsmCode.number_used_registers - 1);
 229
 230         fp->r700Shader.nParamExports = fp->r700AsmCode.number_of_exports;
 231
 232         number_of_colors_exported = fp->r700AsmCode.number_of_colorandz_exports;
 233
 234         unBit = 1 << FRAG_RESULT_DEPTH;
 235         if(mesa_fp->Base.OutputsWritten & unBit)
 236         {
 237                 z_enabled = GL_TRUE;
 238                 number_of_colors_exported--;
 239         }
 240
 241         fp->r700Shader.exportMode = number_of_colors_exported << 1 | z_enabled;
 242
 243     fp->translated = GL_TRUE;
 244
 245         return GL_TRUE;
 246 }
 247
 248 void * r700GetActiveFpShaderBo(GLcontext * ctx)
 249 {
 250     struct r700_fragment_program *fp = (struct r700_fragment_program *)
 251                                            (ctx->FragmentProgram._Current);
 252
 253     return fp->shaderbo;
 254 }
 255
 256 GLboolean r700SetupFragmentProgram(GLcontext * ctx)
 257 {
 258     context_t *context = R700_CONTEXT(ctx);
 259     BATCH_LOCALS(&context->radeon);
 260
 261     R700_CHIP_CONTEXT *r700 = (R700_CHIP_CONTEXT*)(&context->hw);
 262
 263     struct r700_fragment_program *fp = (struct r700_fragment_program *)
 264                                            (ctx->FragmentProgram._Current);
 265
 266     struct gl_program_parameter_list *paramList;
 267     unsigned int unNumParamData;
 268     unsigned int ui;
 269
 270     unsigned int unNumOfReg;
 271
 272     if(GL_FALSE == fp->loaded)
 273     {
 274         if(fp->r700Shader.bNeedsAssembly == GL_TRUE)
 275             {
 276                     Assemble( &(fp->r700Shader) );
 277             }
 278
 279         /* Load fp to gpu */
 280         r600EmitShader(ctx,
 281                        &(fp->shaderbo),
 282                        (GLvoid *)(fp->r700Shader.pProgram),
 283                        fp->r700Shader.uShaderBinaryDWORDSize,
 284                        "FS");
 285
 286         fp->loaded = GL_TRUE;
 287     }
 288
 289     DumpHwBinary(DUMP_PIXEL_SHADER, (GLvoid *)(fp->r700Shader.pProgram),
 290                  fp->r700Shader.uShaderBinaryDWORDSize);
 291
 292     /* TODO : enable this after MemUse fixed *=
 293     (context->chipobj.MemUse)(context, fp->shadercode.buf->id);
 294     */
 295
 296     r700->ps.SQ_PGM_START_PS.u32All = 0; /* set from buffer obj */
 297
 298     unNumOfReg = fp->r700Shader.nRegs + 1;
 299
 300     ui = (r700->SPI_PS_IN_CONTROL_0.u32All & NUM_INTERP_mask) / (1 << NUM_INTERP_shift);
 301
 302     ui = ui ? ui : unNumOfReg;
 303
 304     SETfield(r700->ps.SQ_PGM_RESOURCES_PS.u32All, ui, NUM_GPRS_shift, NUM_GPRS_mask);
 305
 306     CLEARbit(r700->ps.SQ_PGM_RESOURCES_PS.u32All, UNCACHED_FIRST_INST_bit);
 307
 308     if(fp->r700Shader.uStackSize) /* we don't use branch for now, it should be zero. */
 309         {
 310         SETfield(r700->ps.SQ_PGM_RESOURCES_PS.u32All, fp->r700Shader.uStackSize,
 311                  STACK_SIZE_shift, STACK_SIZE_mask);
 312     }
 313
 314     SETfield(r700->ps.SQ_PGM_EXPORTS_PS.u32All, fp->r700Shader.exportMode,
 315              EXPORT_MODE_shift, EXPORT_MODE_mask);
 316
 317     if(fp->r700Shader.killIsUsed)
 318     {
 319             SETbit(r700->DB_SHADER_CONTROL.u32All, KILL_ENABLE_bit);
 320     }
 321     else
 322     {
 323         CLEARbit(r700->DB_SHADER_CONTROL.u32All, KILL_ENABLE_bit);
 324     }
 325
 326     if(fp->r700Shader.depthIsExported)
 327     {
 328             SETbit(r700->DB_SHADER_CONTROL.u32All, Z_EXPORT_ENABLE_bit);
 329     }
 330     else
 331     {
 332         CLEARbit(r700->DB_SHADER_CONTROL.u32All, Z_EXPORT_ENABLE_bit);
 333     }
 334
 335     /* sent out shader constants. */
 336
 337     paramList = fp->mesa_program.Base.Parameters;
 338
 339     if(NULL != paramList)
 340     {
 341         _mesa_load_state_parameters(ctx, paramList);
 342
 343         unNumParamData = paramList->NumParameters * 4;
 344
 345         BEGIN_BATCH_NO_AUTOSTATE(2 + unNumParamData);
 346
 347         R600_OUT_BATCH(CP_PACKET3(R600_IT_SET_ALU_CONST, unNumParamData));
 348
 349         /* assembler map const from very beginning. */
 350         R600_OUT_BATCH(SQ_ALU_CONSTANT_PS_OFFSET * 4);
 351
 352         unNumParamData = paramList->NumParameters;
 353
 354         for(ui=0; ui<unNumParamData; ui++)
 355         {
 356             R600_OUT_BATCH(*((unsigned int*)&(paramList->ParameterValues[ui][0])));
 357             R600_OUT_BATCH(*((unsigned int*)&(paramList->ParameterValues[ui][1])));
 358             R600_OUT_BATCH(*((unsigned int*)&(paramList->ParameterValues[ui][2])));
 359             R600_OUT_BATCH(*((unsigned int*)&(paramList->ParameterValues[ui][3])));
 360         }
 361         END_BATCH();
 362         COMMIT_BATCH();
 363     }
 364
 365     return GL_TRUE;
 366 }
 367
 368
 369