src/mesa/drivers/dri/i965/brw_shader.cpp

   1 /*
   2  * Copyright © 2010 Intel Corporation
   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 (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 extern "C" {
  25 #include "main/macros.h"
  26 #include "brw_context.h"
  27 #include "brw_vs.h"
  28 }
  29 #include "brw_fs.h"
  30 #include "glsl/ir_optimization.h"
  31 #include "glsl/ir_print_visitor.h"
  32
  33 struct gl_shader *
  34 brw_new_shader(struct gl_context *ctx, GLuint name, GLuint type)
  35 {
  36    struct brw_shader *shader;
  37
  38    shader = rzalloc(NULL, struct brw_shader);
  39    if (shader) {
  40       shader->base.Type = type;
  41       shader->base.Name = name;
  42       _mesa_init_shader(ctx, &shader->base);
  43    }
  44
  45    return &shader->base;
  46 }
  47
  48 struct gl_shader_program *
  49 brw_new_shader_program(struct gl_context *ctx, GLuint name)
  50 {
  51    struct brw_shader_program *prog;
  52    prog = rzalloc(NULL, struct brw_shader_program);
  53    if (prog) {
  54       prog->base.Name = name;
  55       _mesa_init_shader_program(ctx, &prog->base);
  56    }
  57    return &prog->base;
  58 }
  59
  60 /**
  61  * Performs a compile of the shader stages even when we don't know
  62  * what non-orthogonal state will be set, in the hope that it reflects
  63  * the eventual NOS used, and thus allows us to produce link failures.
  64  */
  65 bool
  66 brw_shader_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
  67 {
  68    struct brw_context *brw = brw_context(ctx);
  69
  70    if (brw->precompile && !brw_fs_precompile(ctx, prog))
  71       return false;
  72
  73    if (!brw_vs_precompile(ctx, prog))
  74       return false;
  75
  76    return true;
  77 }
  78
  79 GLboolean
  80 brw_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
  81 {
  82    struct brw_context *brw = brw_context(ctx);
  83    struct intel_context *intel = &brw->intel;
  84    unsigned int stage;
  85
  86    if (!_mesa_ir_link_shader(ctx, prog))
  87       return false;
  88
  89    for (stage = 0; stage < ARRAY_SIZE(prog->_LinkedShaders); stage++) {
  90       struct brw_shader *shader =
  91          (struct brw_shader *)prog->_LinkedShaders[stage];
  92
  93       if (!shader)
  94          continue;
  95
  96       void *mem_ctx = ralloc_context(NULL);
  97       bool progress;
  98
  99       if (shader->ir)
 100          ralloc_free(shader->ir);
 101       shader->ir = new(shader) exec_list;
 102       clone_ir_list(mem_ctx, shader->ir, shader->base.ir);
 103
 104       do_mat_op_to_vec(shader->ir);
 105       lower_instructions(shader->ir,
 106                          MOD_TO_FRACT |
 107                          DIV_TO_MUL_RCP |
 108                          SUB_TO_ADD_NEG |
 109                          EXP_TO_EXP2 |
 110                          LOG_TO_LOG2);
 111
 112       /* Pre-gen6 HW can only nest if-statements 16 deep.  Beyond this,
 113        * if-statements need to be flattened.
 114        */
 115       if (intel->gen < 6)
 116          lower_if_to_cond_assign(shader->ir, 16);
 117
 118       do_lower_texture_projection(shader->ir);
 119       do_vec_index_to_cond_assign(shader->ir);
 120       brw_do_cubemap_normalize(shader->ir);
 121       lower_noise(shader->ir);
 122       lower_quadop_vector(shader->ir, false);
 123
 124       bool input = true;
 125       bool output = stage == MESA_SHADER_FRAGMENT;
 126       bool temp = stage == MESA_SHADER_FRAGMENT;
 127       bool uniform = stage == MESA_SHADER_FRAGMENT;
 128
 129       lower_variable_index_to_cond_assign(shader->ir,
 130                                           input, output, temp, uniform);
 131
 132       do {
 133          progress = false;
 134
 135          if (stage == MESA_SHADER_FRAGMENT) {
 136             brw_do_channel_expressions(shader->ir);
 137             brw_do_vector_splitting(shader->ir);
 138          }
 139
 140          progress = do_lower_jumps(shader->ir, true, true,
 141                                    true, /* main return */
 142                                    false, /* continue */
 143                                    false /* loops */
 144                                    ) || progress;
 145
 146          progress = do_common_optimization(shader->ir, true, true, 32)
 147            || progress;
 148       } while (progress);
 149
 150       validate_ir_tree(shader->ir);
 151
 152       reparent_ir(shader->ir, shader->ir);
 153       ralloc_free(mem_ctx);
 154    }
 155
 156    if (!brw_shader_precompile(ctx, prog))
 157       return false;
 158
 159    return true;
 160 }
 161
 162
 163 int
 164 brw_type_for_base_type(const struct glsl_type *type)
 165 {
 166    switch (type->base_type) {
 167    case GLSL_TYPE_FLOAT:
 168       return BRW_REGISTER_TYPE_F;
 169    case GLSL_TYPE_INT:
 170    case GLSL_TYPE_BOOL:
 171       return BRW_REGISTER_TYPE_D;
 172    case GLSL_TYPE_UINT:
 173       return BRW_REGISTER_TYPE_UD;
 174    case GLSL_TYPE_ARRAY:
 175       return brw_type_for_base_type(type->fields.array);
 176    case GLSL_TYPE_STRUCT:
 177    case GLSL_TYPE_SAMPLER:
 178       /* These should be overridden with the type of the member when
 179        * dereferenced into.  BRW_REGISTER_TYPE_UD seems like a likely
 180        * way to trip up if we don't.
 181        */
 182       return BRW_REGISTER_TYPE_UD;
 183    default:
 184       assert(!"not reached");
 185       return BRW_REGISTER_TYPE_F;
 186    }
 187 }
 188
 189 uint32_t
 190 brw_conditional_for_comparison(unsigned int op)
 191 {
 192    switch (op) {
 193    case ir_binop_less:
 194       return BRW_CONDITIONAL_L;
 195    case ir_binop_greater:
 196       return BRW_CONDITIONAL_G;
 197    case ir_binop_lequal:
 198       return BRW_CONDITIONAL_LE;
 199    case ir_binop_gequal:
 200       return BRW_CONDITIONAL_GE;
 201    case ir_binop_equal:
 202    case ir_binop_all_equal: /* same as equal for scalars */
 203       return BRW_CONDITIONAL_Z;
 204    case ir_binop_nequal:
 205    case ir_binop_any_nequal: /* same as nequal for scalars */
 206       return BRW_CONDITIONAL_NZ;
 207    default:
 208       assert(!"not reached: bad operation for comparison");
 209       return BRW_CONDITIONAL_NZ;
 210    }
 211 }
 212
 213 uint32_t
 214 brw_math_function(enum opcode op)
 215 {
 216    switch (op) {
 217    case SHADER_OPCODE_RCP:
 218       return BRW_MATH_FUNCTION_INV;
 219    case SHADER_OPCODE_RSQ:
 220       return BRW_MATH_FUNCTION_RSQ;
 221    case SHADER_OPCODE_SQRT:
 222       return BRW_MATH_FUNCTION_SQRT;
 223    case SHADER_OPCODE_EXP2:
 224       return BRW_MATH_FUNCTION_EXP;
 225    case SHADER_OPCODE_LOG2:
 226       return BRW_MATH_FUNCTION_LOG;
 227    case SHADER_OPCODE_POW:
 228       return BRW_MATH_FUNCTION_POW;
 229    case SHADER_OPCODE_SIN:
 230       return BRW_MATH_FUNCTION_SIN;
 231    case SHADER_OPCODE_COS:
 232       return BRW_MATH_FUNCTION_COS;
 233    case SHADER_OPCODE_INT_QUOTIENT:
 234       return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
 235    case SHADER_OPCODE_INT_REMAINDER:
 236       return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
 237    default:
 238       assert(!"not reached: unknown math function");
 239       return 0;
 240    }
 241 }
 242
 243 uint32_t
 244 brw_texture_offset(ir_constant *offset)
 245 {
 246    assert(offset != NULL);
 247
 248    signed char offsets[3];
 249    for (unsigned i = 0; i < offset->type->vector_elements; i++)
 250       offsets[i] = (signed char) offset->value.i[i];
 251
 252    /* Combine all three offsets into a single unsigned dword:
 253     *
 254     *    bits 11:8 - U Offset (X component)
 255     *    bits  7:4 - V Offset (Y component)
 256     *    bits  3:0 - R Offset (Z component)
 257     */
 258    unsigned offset_bits = 0;
 259    for (unsigned i = 0; i < offset->type->vector_elements; i++) {
 260       const unsigned shift = 4 * (2 - i);
 261       offset_bits |= (offsets[i] << shift) & (0xF << shift);
 262    }
 263    return offset_bits;
 264 }