docs/shading.html

   1 <HTML>
   2
   3 <TITLE>Shading Language Support</TITLE>
   4
   5 <link rel="stylesheet" type="text/css" href="mesa.css"></head>
   6
   7 <BODY>
   8
   9 <H1>Shading Language Support</H1>
  10
  11 <p>
  12 This page describes the features and status of Mesa's support for the
  13 <a href="http://opengl.org/documentation/glsl/" target="_parent">
  14 OpenGL Shading Language</a>.
  15 </p>
  16
  17 <p>
  18 Last updated on 15 December 2008.
  19 </p>
  20
  21 <p>
  22 Contents
  23 </p>
  24 <ul>
  25 <li><a href="#envvars">Environment variables</a>
  26 <li><a href="#120">GLSL 1.20 support</a>
  27 <li><a href="#unsup">Unsupported Features</a>
  28 <li><a href="#notes">Implementation Notes</a>
  29 <li><a href="#hints">Programming Hints</a>
  30 <li><a href="#standalone">Stand-alone GLSL Compiler</a>
  31 <li><a href="#implementation">Compiler Implementation</a>
  32 <li><a href="#validation">Compiler Validation</a>
  33 </ul>
  34
  35
  36
  37 <a name="envvars">
  38 <h2>Environment Variables</h2>
  39
  40 <p>
  41 The <b>MESA_GLSL</b> environment variable can be set to a comma-separated
  42 list of keywords to control some aspects of the GLSL compiler:
  43 </p>
  44 <ul>
  45 <li>dump - print GLSL shader code to stdout at link time
  46 <li>log - log all GLSL shaders to files.
  47     The filenames will be "shader_X.vert" or "shader_X.frag" where X
  48     the shader ID.
  49 <li>nopt - disable compiler optimizations
  50 <li>opt - force compiler optimizations
  51 <li>uniform - print message to stdout when glUniform is called
  52 </ul>
  53 <p>
  54 Example:  export MESA_GLSL=dump,nopt
  55 </p>
  56
  57
  58 <a name="120">
  59 <h2>GLSL 1.20 support</h2>
  60
  61 <p>
  62 GLSL version 1.20 is supported in Mesa 7.3 and later.
  63 Among the features/differences of GLSL 1.20 are:
  64 <ul>
  65 <li><code>mat2x3, mat2x4</code>, etc. types and functions
  66 <li><code>transpose(), outerProduct(), matrixCompMult()</code> functions
  67 (but untested)
  68 <li>precision qualifiers (lowp, mediump, highp)
  69 <li><code>invariant</code> qualifier
  70 <li><code>array.length()</code> method
  71 <li><code>float[5] a;</code> array syntax
  72 <li><code>centroid</code> qualifier
  73 <li>unsized array constructors
  74 <li>initializers for uniforms
  75 <li>const initializers calling built-in functions
  76 </ul>
  77
  78
  79
  80 <a name="unsup">
  81 <h2>Unsupported Features</h2>
  82
  83 <p>
  84 The following features of the shading language are not yet fully supported
  85 in Mesa:
  86 </p>
  87
  88 <ul>
  89 <li>Linking of multiple shaders does not always work.  Currently, linking
  90     is implemented through shader concatenation and re-compiling.  This
  91     doesn't always work because of some #pragma and preprocessor issues.
  92 <li>gl_ClipVertex
  93 <li>The gl_Color and gl_SecondaryColor varying vars are interpolated
  94     without perspective correction
  95 </ul>
  96
  97 <p>
  98 All other major features of the shading language should function.
  99 </p>
 100
 101
 102 <a name="notes">
 103 <h2>Implementation Notes</h2>
 104
 105 <ul>
 106 <li>Shading language programs are compiled into low-level programs
 107     very similar to those of GL_ARB_vertex/fragment_program.
 108 <li>All vector types (vec2, vec3, vec4, bvec2, etc) currently occupy full
 109     float[4] registers.
 110 <li>Float constants and variables are packed so that up to four floats
 111     can occupy one program parameter/register.
 112 <li>All function calls are inlined.
 113 <li>Shaders which use too many registers will not compile.
 114 <li>The quality of generated code is pretty good, register usage is fair.
 115 <li>Shader error detection and reporting of errors (InfoLog) is not
 116     very good yet.
 117 <li>The ftransform() function doesn't necessarily match the results of
 118     fixed-function transformation.
 119 </ul>
 120
 121 <p>
 122 These issues will be addressed/resolved in the future.
 123 </p>
 124
 125
 126 <a name="hints">
 127 <h2>Programming Hints</h2>
 128
 129 <ul>
 130 <li>Declare <em>in</em> function parameters as <em>const</em> whenever possible.
 131     This improves the efficiency of function inlining.
 132 </li>
 133 <br>
 134 <li>To reduce register usage, declare variables within smaller scopes.
 135     For example, the following code:
 136 <pre>
 137     void main()
 138     {
 139        vec4 a1, a2, b1, b2;
 140        gl_Position = expression using a1, a2.
 141        gl_Color = expression using b1, b2;
 142     }
 143 </pre>
 144     Can be rewritten as follows to use half as many registers:
 145 <pre>
 146     void main()
 147     {
 148        {
 149           vec4 a1, a2;
 150           gl_Position = expression using a1, a2.
 151        }
 152        {
 153           vec4 b1, b2;
 154           gl_Color = expression using b1, b2;
 155        }
 156     }
 157 </pre>
 158     Alternately, rather than using several float variables, use
 159     a vec4 instead.  Use swizzling and writemasks to access the
 160     components of the vec4 as floats.
 161 </li>
 162 <br>
 163 <li>Use the built-in library functions whenever possible.
 164     For example, instead of writing this:
 165 <pre>
 166         float x = 1.0 / sqrt(y);
 167 </pre>
 168     Write this:
 169 <pre>
 170         float x = inversesqrt(y);
 171 </pre>
 172 <li>
 173    Use ++i when possible as it's more efficient than i++
 174 </li>
 175 </ul>
 176
 177
 178 <a name="standalone">
 179 <h2>Stand-alone GLSL Compiler</h2>
 180
 181 <p>
 182 A unique stand-alone GLSL compiler driver has been added to Mesa.
 183 <p>
 184
 185 <p>
 186 The stand-alone compiler (like a conventional command-line compiler)
 187 is a tool that accepts Shading Language programs and emits low-level
 188 GPU programs.
 189 </p>
 190
 191 <p>
 192 This tool is useful for:
 193 <p>
 194 <ul>
 195 <li>Inspecting GPU code to gain insight into compilation
 196 <li>Generating initial GPU code for subsequent hand-tuning
 197 <li>Debugging the GLSL compiler itself
 198 </ul>
 199
 200 <p>
 201 After building Mesa, the glslcompiler can be built by manually running:
 202 </p>
 203 <pre>
 204     make realclean
 205     make linux
 206     cd src/mesa/drivers/glslcompiler
 207     make
 208 </pre>
 209
 210
 211 <p>
 212 Here's an example of using the compiler to compile a vertex shader and
 213 emit GL_ARB_vertex_program-style instructions:
 214 </p>
 215 <pre>
 216     bin/glslcompiler --debug --numbers --fs progs/glsl/CH06-brick.frag.txt
 217 </pre>
 218 <p>
 219 results in:
 220 </p>
 221 <pre>
 222 # Fragment Program/Shader
 223   0: RCP TEMP[4].x, UNIFORM[2].xxxx;
 224   1: RCP TEMP[4].y, UNIFORM[2].yyyy;
 225   2: MUL TEMP[3].xy, VARYING[0], TEMP[4];
 226   3: MOV TEMP[1], TEMP[3];
 227   4: MUL TEMP[0].w, TEMP[1].yyyy, CONST[4].xxxx;
 228   5: FRC TEMP[1].z, TEMP[0].wwww;
 229   6: SGT.C TEMP[0].w, TEMP[1].zzzz, CONST[4].xxxx;
 230   7: IF (NE.wwww); # (if false, goto 9);
 231   8:    ADD TEMP[1].x, TEMP[1].xxxx, CONST[4].xxxx;
 232   9: ENDIF;
 233  10: FRC TEMP[1].xy, TEMP[1];
 234  11: SGT TEMP[2].xy, UNIFORM[3], TEMP[1];
 235  12: MUL TEMP[1].z, TEMP[2].xxxx, TEMP[2].yyyy;
 236  13: LRP TEMP[0], TEMP[1].zzzz, UNIFORM[0], UNIFORM[1];
 237  14: MUL TEMP[0].xyz, TEMP[0], VARYING[1].xxxx;
 238  15: MOV OUTPUT[0].xyz, TEMP[0];
 239  16: MOV OUTPUT[0].w, CONST[4].yyyy;
 240  17: END
 241 </pre>
 242
 243 <p>
 244 Note that some shading language constructs (such as uniform and varying
 245 variables) aren't expressible in ARB or NV-style programs.
 246 Therefore, the resulting output is not always legal by definition of
 247 those program languages.
 248 </p>
 249 <p>
 250 Also note that this compiler driver is still under development.
 251 Over time, the correctness of the GPU programs, with respect to the ARB
 252 and NV languagues, should improve.
 253 </p>
 254
 255
 256
 257 <a name="implementation">
 258 <h2>Compiler Implementation</h2>
 259
 260 <p>
 261 The source code for Mesa's shading language compiler is in the
 262 <code>src/mesa/shader/slang/</code> directory.
 263 </p>
 264
 265 <p>
 266 The compiler follows a fairly standard design and basically works as follows:
 267 </p>
 268 <ul>
 269 <li>The input string is tokenized (see grammar.c) and parsed
 270 (see slang_compiler_*.c) to produce an Abstract Syntax Tree (AST).
 271 The nodes in this tree are slang_operation structures
 272 (see slang_compile_operation.h).
 273 The nodes are decorated with symbol table, scoping and datatype information.
 274 <li>The AST is converted into an Intermediate representation (IR) tree
 275 (see the slang_codegen.c file).
 276 The IR nodes represent basic GPU instructions, like add, dot product,
 277 move, etc.
 278 The IR tree is mostly a binary tree, but a few nodes have three or four
 279 children.
 280 In principle, the IR tree could be executed by doing an in-order traversal.
 281 <li>The IR tree is traversed in-order to emit code (see slang_emit.c).
 282 This is also when registers are allocated to store variables and temps.
 283 <li>In the future, a pattern-matching code generator-generator may be
 284 used for code generation.
 285 Programs such as L-BURG (Bottom-Up Rewrite Generator) and Twig look for
 286 patterns in IR trees, compute weights for subtrees and use the weights
 287 to select the best instructions to represent the sub-tree.
 288 <li>The emitted GPU instructions (see prog_instruction.h) are stored in a
 289 gl_program object (see mtypes.h).
 290 <li>When a fragment shader and vertex shader are linked (see slang_link.c)
 291 the varying vars are matched up, uniforms are merged, and vertex
 292 attributes are resolved (rewriting instructions as needed).
 293 </ul>
 294
 295 <p>
 296 The final vertex and fragment programs may be interpreted in software
 297 (see prog_execute.c) or translated into a specific hardware architecture
 298 (see drivers/dri/i915/i915_fragprog.c for example).
 299 </p>
 300
 301 <h3>Code Generation Options</h3>
 302
 303 <p>
 304 Internally, there are several options that control the compiler's code
 305 generation and instruction selection.
 306 These options are seen in the gl_shader_state struct and may be set
 307 by the device driver to indicate its preferences:
 308
 309 <pre>
 310 struct gl_shader_state
 311 {
 312    ...
 313    /** Driver-selectable options: */
 314    GLboolean EmitHighLevelInstructions;
 315    GLboolean EmitCondCodes;
 316    GLboolean EmitComments;
 317 };
 318 </pre>
 319
 320 <ul>
 321 <li>EmitHighLevelInstructions
 322 <br>
 323 This option controls instruction selection for loops and conditionals.
 324 If the option is set high-level IF/ELSE/ENDIF, LOOP/ENDLOOP, CONT/BRK
 325 instructions will be emitted.
 326 Otherwise, those constructs will be implemented with BRA instructions.
 327 </li>
 328
 329 <li>EmitCondCodes
 330 <br>
 331 If set, condition codes (ala GL_NV_fragment_program) will be used for
 332 branching and looping.
 333 Otherwise, ordinary registers will be used (the IF instruction will
 334 examine the first operand's X component and do the if-part if non-zero).
 335 This option is only relevant if EmitHighLevelInstructions is set.
 336 </li>
 337
 338 <li>EmitComments
 339 <br>
 340 If set, instructions will be annoted with comments to help with debugging.
 341 Extra NOP instructions will also be inserted.
 342 </br>
 343
 344 </ul>
 345
 346
 347 <a name="validation">
 348 <h2>Compiler Validation</h2>
 349
 350 <p>
 351 A <a href="http://glean.sf.net" target="_parent">Glean</a> test has
 352 been create to exercise the GLSL compiler.
 353 </p>
 354 <p>
 355 The <em>glsl1</em> test runs over 170 sub-tests to check that the language
 356 features and built-in functions work properly.
 357 This test should be run frequently while working on the compiler to catch
 358 regressions.
 359 </p>
 360 <p>
 361 The test coverage is reasonably broad and complete but additional tests
 362 should be added.
 363 </p>
 364
 365
 366 </BODY>
 367 </HTML>