src/mesa/drivers/dri/r300/r300_fragprog_common.c

   1 /*
   2  * Copyright (C) 2009 Maciej Cencora <m.cencora@gmail.com>
   3  *
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining
   7  * a copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sublicense, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial
  16  * portions of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  */
  27
  28 /**
  29  * \file
  30  *
  31  * Fragment program compiler. Perform transformations on the intermediate
  32  * representation until the program is in a form where we can translate
  33  * it more or less directly into machine-readable form.
  34  *
  35  * \author Ben Skeggs <darktama@iinet.net.au>
  36  * \author Jerome Glisse <j.glisse@gmail.com>
  37  */
  38
  39 #include "r300_fragprog_common.h"
  40
  41 #include "shader/program.h"
  42 #include "shader/prog_parameter.h"
  43 #include "shader/prog_print.h"
  44
  45 #include "r300_state.h"
  46 #include "r300_fragprog.h"
  47 #include "r300_fragprog_swizzle.h"
  48 #include "r500_fragprog.h"
  49
  50 #include "radeon_program.h"
  51 #include "radeon_program_alu.h"
  52
  53 static void update_params(GLcontext *ctx, struct gl_fragment_program *fp)
  54 {
  55         /* Ask Mesa nicely to fill in ParameterValues for us */
  56         if (fp->Base.Parameters)
  57                 _mesa_load_state_parameters(ctx, fp->Base.Parameters);
  58 }
  59
  60 static void nqssadce_init(struct nqssadce_state* s)
  61 {
  62         s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW;
  63         s->Outputs[FRAG_RESULT_DEPTH].Sourced = WRITEMASK_W;
  64 }
  65
  66 /**
  67  * Transform the program to support fragment.position.
  68  *
  69  * Introduce a small fragment at the start of the program that will be
  70  * the only code that directly reads the FRAG_ATTRIB_WPOS input.
  71  * All other code pieces that reference that input will be rewritten
  72  * to read from a newly allocated temporary.
  73  *
  74  */
  75 static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler)
  76 {
  77         GLuint InputsRead = compiler->fp->Base.Base.InputsRead;
  78
  79         if (!(InputsRead & FRAG_BIT_WPOS))
  80                 return;
  81
  82         static gl_state_index tokens[STATE_LENGTH] = {
  83                 STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0
  84         };
  85         struct prog_instruction *fpi;
  86         GLuint window_index;
  87         int i = 0;
  88         GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY);
  89
  90         _mesa_insert_instructions(compiler->program, 0, 3);
  91         fpi = compiler->program->Instructions;
  92
  93         /* perspective divide */
  94         fpi[i].Opcode = OPCODE_RCP;
  95
  96         fpi[i].DstReg.File = PROGRAM_TEMPORARY;
  97         fpi[i].DstReg.Index = tempregi;
  98         fpi[i].DstReg.WriteMask = WRITEMASK_W;
  99         fpi[i].DstReg.CondMask = COND_TR;
 100
 101         fpi[i].SrcReg[0].File = PROGRAM_INPUT;
 102         fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
 103         fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW;
 104         i++;
 105
 106         fpi[i].Opcode = OPCODE_MUL;
 107
 108         fpi[i].DstReg.File = PROGRAM_TEMPORARY;
 109         fpi[i].DstReg.Index = tempregi;
 110         fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
 111         fpi[i].DstReg.CondMask = COND_TR;
 112
 113         fpi[i].SrcReg[0].File = PROGRAM_INPUT;
 114         fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
 115         fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
 116
 117         fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY;
 118         fpi[i].SrcReg[1].Index = tempregi;
 119         fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW;
 120         i++;
 121
 122         /* viewport transformation */
 123         window_index = _mesa_add_state_reference(compiler->program->Parameters, tokens);
 124
 125         fpi[i].Opcode = OPCODE_MAD;
 126
 127         fpi[i].DstReg.File = PROGRAM_TEMPORARY;
 128         fpi[i].DstReg.Index = tempregi;
 129         fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
 130         fpi[i].DstReg.CondMask = COND_TR;
 131
 132         fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY;
 133         fpi[i].SrcReg[0].Index = tempregi;
 134         fpi[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
 135
 136         fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR;
 137         fpi[i].SrcReg[1].Index = window_index;
 138         fpi[i].SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
 139
 140         fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR;
 141         fpi[i].SrcReg[2].Index = window_index;
 142         fpi[i].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
 143         i++;
 144
 145         for (; i < compiler->program->NumInstructions; ++i) {
 146                 int reg;
 147                 for (reg = 0; reg < 3; reg++) {
 148                         if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT &&
 149                             fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) {
 150                                 fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY;
 151                                 fpi[i].SrcReg[reg].Index = tempregi;
 152                         }
 153                 }
 154         }
 155 }
 156
 157 static GLuint build_dtm(GLuint depthmode)
 158 {
 159         switch(depthmode) {
 160         default:
 161         case GL_LUMINANCE: return 0;
 162         case GL_INTENSITY: return 1;
 163         case GL_ALPHA: return 2;
 164         }
 165 }
 166
 167 static GLuint build_func(GLuint comparefunc)
 168 {
 169         return comparefunc - GL_NEVER;
 170 }
 171
 172 /**
 173  * Collect all external state that is relevant for compiling the given
 174  * fragment program.
 175  */
 176 static void build_state(
 177         r300ContextPtr r300,
 178         struct r300_fragment_program *fp,
 179         struct r300_fragment_program_external_state *state)
 180 {
 181         int unit;
 182
 183         _mesa_bzero(state, sizeof(*state));
 184
 185         for(unit = 0; unit < 16; ++unit) {
 186                 if (fp->Base.Base.ShadowSamplers & (1 << unit)) {
 187                         struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current;
 188
 189                         state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode);
 190                         state->unit[unit].texture_compare_func = build_func(tex->CompareFunc);
 191                 }
 192         }
 193 }
 194
 195 void r300TranslateFragmentShader(GLcontext *ctx, struct gl_fragment_program *fp)
 196 {
 197         r300ContextPtr r300 = R300_CONTEXT(ctx);
 198         struct r300_fragment_program *r300_fp = (struct r300_fragment_program *)fp;
 199         struct r300_fragment_program_external_state state;
 200
 201         build_state(r300, r300_fp, &state);
 202         if (_mesa_memcmp(&r300_fp->state, &state, sizeof(state))) {
 203                 /* TODO: cache compiled programs */
 204                 r300_fp->translated = GL_FALSE;
 205                 _mesa_memcpy(&r300_fp->state, &state, sizeof(state));
 206         }
 207
 208         if (!r300_fp->translated) {
 209                 struct r300_fragment_program_compiler compiler;
 210
 211                 compiler.r300 = r300;
 212                 compiler.fp = r300_fp;
 213                 compiler.code = &r300_fp->code;
 214                 compiler.program = _mesa_clone_program(ctx, &fp->Base);
 215
 216                 if (RADEON_DEBUG & DEBUG_PIXEL) {
 217                         fflush(stdout);
 218                         _mesa_printf("Fragment Program: Initial program:\n");
 219                         _mesa_print_program(compiler.program);
 220                         fflush(stdout);
 221                 }
 222
 223                 insert_WPOS_trailer(&compiler);
 224
 225                 if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
 226                         struct radeon_program_transformation transformations[] = {
 227                                 { &r500_transform_TEX, &compiler },
 228                                 { &radeonTransformALU, 0 },
 229                                 { &radeonTransformDeriv, 0 },
 230                                 { &radeonTransformTrigScale, 0 }
 231                         };
 232                         radeonLocalTransform(ctx, compiler.program, 4, transformations);
 233                 } else {
 234                         struct radeon_program_transformation transformations[] = {
 235                                 { &r300_transform_TEX, &compiler },
 236                                 { &radeonTransformALU, 0 },
 237                                 { &radeonTransformTrigSimple, 0 }
 238                         };
 239                         radeonLocalTransform(ctx, compiler.program, 3, transformations);
 240                 }
 241
 242                 if (RADEON_DEBUG & DEBUG_PIXEL) {
 243                         _mesa_printf("Fragment Program: After native rewrite:\n");
 244                         _mesa_print_program(compiler.program);
 245                         fflush(stdout);
 246                 }
 247
 248                 if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
 249                         struct radeon_nqssadce_descr nqssadce = {
 250                                 .Init = &nqssadce_init,
 251                                 .IsNativeSwizzle = &r500FPIsNativeSwizzle,
 252                                 .BuildSwizzle = &r500FPBuildSwizzle,
 253                                 .RewriteDepthOut = GL_TRUE
 254                         };
 255                         radeonNqssaDce(ctx, compiler.program, &nqssadce);
 256                 } else {
 257                         struct radeon_nqssadce_descr nqssadce = {
 258                                 .Init = &nqssadce_init,
 259                                 .IsNativeSwizzle = &r300FPIsNativeSwizzle,
 260                                 .BuildSwizzle = &r300FPBuildSwizzle,
 261                                 .RewriteDepthOut = GL_TRUE
 262                         };
 263                         radeonNqssaDce(ctx, compiler.program, &nqssadce);
 264                 }
 265
 266                 if (RADEON_DEBUG & DEBUG_PIXEL) {
 267                         _mesa_printf("Compiler: after NqSSA-DCE:\n");
 268                         _mesa_print_program(compiler.program);
 269                         fflush(stdout);
 270                 }
 271
 272                 if (!r300->vtbl.BuildFragmentProgramHwCode(&compiler))
 273                         r300_fp->error = GL_TRUE;
 274
 275                 /* Subtle: Rescue any parameters that have been added during transformations */
 276                 _mesa_free_parameter_list(fp->Base.Parameters);
 277                 fp->Base.Parameters = compiler.program->Parameters;
 278                 compiler.program->Parameters = 0;
 279
 280                 _mesa_reference_program(ctx, &compiler.program, NULL);
 281
 282                 r300_fp->translated = GL_TRUE;
 283
 284                 r300UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
 285
 286                 if (r300_fp->error || (RADEON_DEBUG & DEBUG_PIXEL))
 287                         r300->vtbl.FragmentProgramDump(&r300_fp->code);
 288         }
 289
 290         update_params(ctx, fp);
 291 }