src/intel/compiler/brw_fs_surface_builder.cpp

   1 /*
   2  * Copyright © 2013-2015 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 #include "isl/isl.h"
  25 #include "brw_fs_surface_builder.h"
  26 #include "brw_fs.h"
  27
  28 using namespace brw;
  29
  30 namespace brw {
  31    namespace surface_access {
  32       namespace {
  33          /**
  34           * Generate a logical send opcode for a surface message and return
  35           * the result.
  36           */
  37          fs_reg
  38          emit_send(const fs_builder &bld, enum opcode opcode,
  39                    const fs_reg &addr, const fs_reg &src, const fs_reg &surface,
  40                    unsigned dims, unsigned arg, unsigned rsize,
  41                    brw_predicate pred = BRW_PREDICATE_NONE)
  42          {
  43             /* Reduce the dynamically uniform surface index to a single
  44              * scalar.
  45              */
  46             const fs_reg usurface = bld.emit_uniformize(surface);
  47             const fs_reg srcs[] = {
  48                addr, src, usurface, brw_imm_ud(dims), brw_imm_ud(arg)
  49             };
  50             const fs_reg dst = bld.vgrf(BRW_REGISTER_TYPE_UD, rsize);
  51             fs_inst *inst = bld.emit(opcode, dst, srcs, ARRAY_SIZE(srcs));
  52
  53             inst->size_written = rsize * dst.component_size(inst->exec_size);
  54             inst->predicate = pred;
  55             return dst;
  56          }
  57       }
  58
  59       /**
  60        * Emit an untyped surface read opcode.  \p dims determines the number
  61        * of components of the address and \p size the number of components of
  62        * the returned value.
  63        */
  64       fs_reg
  65       emit_untyped_read(const fs_builder &bld,
  66                         const fs_reg &surface, const fs_reg &addr,
  67                         unsigned dims, unsigned size,
  68                         brw_predicate pred)
  69       {
  70          return emit_send(bld, SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL,
  71                           addr, fs_reg(), surface, dims, size, size, pred);
  72       }
  73
  74       /**
  75        * Emit an untyped surface write opcode.  \p dims determines the number
  76        * of components of the address and \p size the number of components of
  77        * the argument.
  78        */
  79       void
  80       emit_untyped_write(const fs_builder &bld, const fs_reg &surface,
  81                          const fs_reg &addr, const fs_reg &src,
  82                          unsigned dims, unsigned size,
  83                          brw_predicate pred)
  84       {
  85          emit_send(bld, SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL,
  86                    addr, src, surface, dims, size, 0, pred);
  87       }
  88
  89       /**
  90        * Emit an untyped surface atomic opcode.  \p dims determines the number
  91        * of components of the address and \p rsize the number of components of
  92        * the returned value (either zero or one).
  93        */
  94       fs_reg
  95       emit_untyped_atomic(const fs_builder &bld,
  96                           const fs_reg &surface, const fs_reg &addr,
  97                           const fs_reg &src0, const fs_reg &src1,
  98                           unsigned dims, unsigned rsize, unsigned op,
  99                           brw_predicate pred)
 100       {
 101          /* FINISHME: Factor out this frequently recurring pattern into a
 102           * helper function.
 103           */
 104          const unsigned n = (src0.file != BAD_FILE) + (src1.file != BAD_FILE);
 105          const fs_reg srcs[] = { src0, src1 };
 106          const fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UD, n);
 107          bld.LOAD_PAYLOAD(tmp, srcs, n, 0);
 108
 109          return emit_send(bld, SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL,
 110                           addr, tmp, surface, dims, op, rsize, pred);
 111       }
 112
 113       /**
 114        * Emit an untyped surface atomic float opcode.  \p dims determines the
 115        * number of components of the address and \p rsize the number of
 116        * components of the returned value (either zero or one).
 117        */
 118       fs_reg
 119       emit_untyped_atomic_float(const fs_builder &bld,
 120                                 const fs_reg &surface, const fs_reg &addr,
 121                                 const fs_reg &src0, const fs_reg &src1,
 122                                 unsigned dims, unsigned rsize, unsigned op,
 123                                 brw_predicate pred)
 124       {
 125          /* FINISHME: Factor out this frequently recurring pattern into a
 126           * helper function.
 127           */
 128          const unsigned n = (src0.file != BAD_FILE) + (src1.file != BAD_FILE);
 129          const fs_reg srcs[] = { src0, src1 };
 130          const fs_reg tmp = bld.vgrf(src0.type, n);
 131          bld.LOAD_PAYLOAD(tmp, srcs, n, 0);
 132
 133          return emit_send(bld, SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL,
 134                           addr, tmp, surface, dims, op, rsize, pred);
 135       }
 136
 137       /**
 138        * Emit a typed surface read opcode.  \p dims determines the number of
 139        * components of the address and \p size the number of components of the
 140        * returned value.
 141        */
 142       fs_reg
 143       emit_typed_read(const fs_builder &bld, const fs_reg &surface,
 144                       const fs_reg &addr, unsigned dims, unsigned size)
 145       {
 146          return emit_send(bld, SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL,
 147                           addr, fs_reg(), surface, dims, size, size);
 148       }
 149
 150       /**
 151        * Emit a typed surface write opcode.  \p dims determines the number of
 152        * components of the address and \p size the number of components of the
 153        * argument.
 154        */
 155       void
 156       emit_typed_write(const fs_builder &bld, const fs_reg &surface,
 157                        const fs_reg &addr, const fs_reg &src,
 158                        unsigned dims, unsigned size)
 159       {
 160          emit_send(bld, SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL,
 161                    addr, src, surface, dims, size, 0);
 162       }
 163
 164       /**
 165        * Emit a typed surface atomic opcode.  \p dims determines the number of
 166        * components of the address and \p rsize the number of components of
 167        * the returned value (either zero or one).
 168        */
 169       fs_reg
 170       emit_typed_atomic(const fs_builder &bld, const fs_reg &surface,
 171                         const fs_reg &addr,
 172                         const fs_reg &src0, const fs_reg &src1,
 173                         unsigned dims, unsigned rsize, unsigned op,
 174                         brw_predicate pred)
 175       {
 176          /* FINISHME: Factor out this frequently recurring pattern into a
 177           * helper function.
 178           */
 179          const unsigned n = (src0.file != BAD_FILE) + (src1.file != BAD_FILE);
 180          const fs_reg srcs[] = { src0, src1 };
 181          const fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UD, n);
 182          bld.LOAD_PAYLOAD(tmp, srcs, n, 0);
 183
 184          return emit_send(bld, SHADER_OPCODE_TYPED_ATOMIC_LOGICAL,
 185                           addr, tmp, surface, dims, op, rsize);
 186       }
 187
 188       fs_reg
 189       emit_byte_scattered_read(const fs_builder &bld,
 190                                const fs_reg &surface, const fs_reg &addr,
 191                                unsigned dims, unsigned size,
 192                                unsigned bit_size, brw_predicate pred)
 193       {
 194          return emit_send(bld, SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL,
 195                           addr, fs_reg(), surface, dims, bit_size, size, pred);
 196       }
 197
 198       void
 199       emit_byte_scattered_write(const fs_builder &bld, const fs_reg &surface,
 200                                 const fs_reg &addr, const fs_reg &src,
 201                                 unsigned dims,
 202                                 unsigned bit_size, brw_predicate pred)
 203       {
 204          emit_send(bld, SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL,
 205                    addr, src, surface, dims, bit_size, 0, pred);
 206       }
 207    }
 208 }