src/amd/compiler/tests/test_to_hw_instr.cpp

   1 /*
   2  * Copyright © 2020 Valve 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 "helpers.h"
  25
  26 using namespace aco;
  27
  28 BEGIN_TEST(to_hw_instr.swap_subdword)
  29    for (unsigned i = GFX8; i <= GFX9; i++) {
  30       if (!setup_cs(NULL, (chip_class)i))
  31          continue;
  32
  33       PhysReg v0_lo{256};
  34       PhysReg v0_hi{256};
  35       PhysReg v0_b1{256};
  36       PhysReg v0_b3{256};
  37       PhysReg v1_lo{257};
  38       PhysReg v1_hi{257};
  39       PhysReg v1_b1{257};
  40       PhysReg v1_b3{257};
  41       v0_hi.reg_b += 2;
  42       v1_hi.reg_b += 2;
  43       v0_b1.reg_b += 1;
  44       v1_b1.reg_b += 1;
  45       v0_b3.reg_b += 3;
  46       v1_b3.reg_b += 3;
  47
  48       //>> p_unit_test 0
  49       //~gfx8! v2b: %0:v[0][16:32] = v_xor_b32 %0:v[0][16:32], %0:v[0][0:16] dst_preserve
  50       //~gfx8! v2b: %0:v[0][0:16] = v_xor_b32 %0:v[0][16:32], %0:v[0][0:16] dst_preserve
  51       //~gfx8! v2b: %0:v[0][16:32] = v_xor_b32 %0:v[0][16:32], %0:v[0][0:16] dst_preserve
  52       //~gfx9! v1: %0:v[0] = v_pk_add_u16 %0:v[0].yx, 0
  53       bld.pseudo(aco_opcode::p_unit_test, Operand(0u));
  54       bld.pseudo(aco_opcode::p_parallelcopy,
  55                  Definition(v0_lo, v2b), Definition(v0_hi, v2b),
  56                  Operand(v0_hi, v2b), Operand(v0_lo, v2b));
  57
  58       //! p_unit_test 1
  59       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
  60       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
  61       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
  62       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
  63       //! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][16:32] dst_preserve
  64       bld.pseudo(aco_opcode::p_unit_test, Operand(1u));
  65       bld.pseudo(aco_opcode::p_parallelcopy,
  66                  Definition(v0_lo, v1), Definition(v1_lo, v2b),
  67                  Operand(v1_lo, v1), Operand(v0_lo, v2b));
  68
  69       //! p_unit_test 2
  70       //! v2b: %0:v[0][16:32] = v_mov_b32 %0:v[1][16:32] dst_preserve
  71       //! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][0:16] dst_preserve
  72       //! v2b: %0:v[1][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_preserve
  73       //! v2b: %0:v[0][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_preserve
  74       //! v2b: %0:v[1][0:16] = v_xor_b32 %0:v[1][0:16], %0:v[0][0:16] dst_preserve
  75       bld.pseudo(aco_opcode::p_unit_test, Operand(2u));
  76       bld.pseudo(aco_opcode::p_parallelcopy,
  77                  Definition(v0_lo, v1), Definition(v1_lo, v2b), Definition(v1_hi, v2b),
  78                  Operand(v1_lo, v1), Operand(v0_lo, v2b), Operand(v0_lo, v2b));
  79
  80       //! p_unit_test 3
  81       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
  82       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
  83       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
  84       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
  85       //! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][0:16] dst_preserve
  86       //! v1b: %0:v[1][16:24] = v_mov_b32 %0:v[0][16:24] dst_preserve
  87       bld.pseudo(aco_opcode::p_unit_test, Operand(3u));
  88       bld.pseudo(aco_opcode::p_parallelcopy,
  89                  Definition(v0_lo, v1), Definition(v1_b3, v1b),
  90                  Operand(v1_lo, v1), Operand(v0_b3, v1b));
  91
  92       //! p_unit_test 4
  93       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
  94       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
  95       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
  96       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
  97       //! v1b: %0:v[1][8:16] = v_mov_b32 %0:v[0][8:16] dst_preserve
  98       //! v2b: %0:v[1][16:32] = v_mov_b32 %0:v[0][16:32] dst_preserve
  99       bld.pseudo(aco_opcode::p_unit_test, Operand(4u));
 100       bld.pseudo(aco_opcode::p_parallelcopy,
 101                  Definition(v0_lo, v1), Definition(v1_lo, v1b),
 102                  Operand(v1_lo, v1), Operand(v0_lo, v1b));
 103
 104       //! p_unit_test 5
 105       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[0], %0:v[1]
 106       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[0], %0:v[1]
 107       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[0], %0:v[1]
 108       //~gfx9! v1: %0:v[1],  v1: %0:v[0] = v_swap_b32 %0:v[0], %0:v[1]
 109       //! v1b: %0:v[0][8:16] = v_mov_b32 %0:v[1][8:16] dst_preserve
 110       //! v1b: %0:v[0][24:32] = v_mov_b32 %0:v[1][24:32] dst_preserve
 111       bld.pseudo(aco_opcode::p_unit_test, Operand(5u));
 112       bld.pseudo(aco_opcode::p_parallelcopy,
 113                  Definition(v0_lo, v1b), Definition(v0_hi, v1b), Definition(v1_lo, v1),
 114                  Operand(v1_lo, v1b), Operand(v1_hi, v1b), Operand(v0_lo, v1));
 115
 116       //! p_unit_test 6
 117       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
 118       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
 119       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
 120       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
 121       bld.pseudo(aco_opcode::p_unit_test, Operand(6u));
 122       bld.pseudo(aco_opcode::p_parallelcopy,
 123                  Definition(v0_lo, v2b), Definition(v0_hi, v2b), Definition(v1_lo, v1),
 124                  Operand(v1_lo, v2b), Operand(v1_hi, v2b), Operand(v0_lo, v1));
 125
 126       //! p_unit_test 7
 127       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[0], %0:v[1]
 128       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[0], %0:v[1]
 129       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[0], %0:v[1]
 130       //~gfx8! v2b: %0:v[0][16:32] = v_xor_b32 %0:v[0][16:32], %0:v[0][0:16] dst_preserve
 131       //~gfx8! v2b: %0:v[0][0:16] = v_xor_b32 %0:v[0][16:32], %0:v[0][0:16] dst_preserve
 132       //~gfx8! v2b: %0:v[0][16:32] = v_xor_b32 %0:v[0][16:32], %0:v[0][0:16] dst_preserve
 133       //~gfx9! v1: %0:v[1],  v1: %0:v[0] = v_swap_b32 %0:v[0], %0:v[1]
 134       //~gfx9! v1: %0:v[0] = v_pk_add_u16 %0:v[0].yx, 0
 135       bld.pseudo(aco_opcode::p_unit_test, Operand(7u));
 136       bld.pseudo(aco_opcode::p_parallelcopy,
 137                  Definition(v0_lo, v2b), Definition(v0_hi, v2b), Definition(v1_lo, v1),
 138                  Operand(v1_hi, v2b), Operand(v1_lo, v2b), Operand(v0_lo, v1));
 139
 140       //! p_unit_test 8
 141       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
 142       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
 143       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
 144       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
 145       //! v1b: %0:v[1][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_preserve
 146       //! v1b: %0:v[0][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_preserve
 147       //! v1b: %0:v[1][24:32] = v_xor_b32 %0:v[1][24:32], %0:v[0][24:32] dst_preserve
 148       bld.pseudo(aco_opcode::p_unit_test, Operand(8u));
 149       bld.pseudo(aco_opcode::p_parallelcopy,
 150                  Definition(v0_lo, v3b), Definition(v1_lo, v3b),
 151                  Operand(v1_lo, v3b), Operand(v0_lo, v3b));
 152
 153       //! p_unit_test 9
 154       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
 155       //~gfx8! v1: %0:v[0] = v_xor_b32 %0:v[1], %0:v[0]
 156       //~gfx8! v1: %0:v[1] = v_xor_b32 %0:v[1], %0:v[0]
 157       //~gfx9! v1: %0:v[0],  v1: %0:v[1] = v_swap_b32 %0:v[1], %0:v[0]
 158       //! v1b: %0:v[1][24:32] = v_mov_b32 %0:v[0][24:32] dst_preserve
 159       bld.pseudo(aco_opcode::p_unit_test, Operand(9u));
 160       bld.pseudo(aco_opcode::p_parallelcopy,
 161                  Definition(v0_lo, v3b), Definition(v1_lo, v3b), Definition(v0_b3, v1b),
 162                  Operand(v1_lo, v3b), Operand(v0_lo, v3b), Operand(v1_b3, v1b));
 163
 164       //! p_unit_test 10
 165       //! v1b: %0:v[1][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_preserve
 166       //! v1b: %0:v[0][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_preserve
 167       //! v1b: %0:v[1][8:16] = v_xor_b32 %0:v[1][8:16], %0:v[0][8:16] dst_preserve
 168       //! v1b: %0:v[1][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_preserve
 169       //! v1b: %0:v[0][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_preserve
 170       //! v1b: %0:v[1][16:24] = v_xor_b32 %0:v[1][16:24], %0:v[0][16:24] dst_preserve
 171       bld.pseudo(aco_opcode::p_unit_test, Operand(10u));
 172       bld.pseudo(aco_opcode::p_parallelcopy,
 173                  Definition(v0_b1, v2b), Definition(v1_b1, v2b),
 174                  Operand(v1_b1, v2b), Operand(v0_b1, v2b));
 175
 176       //! p_unit_test 11
 177       //! v2b: %0:v[1][0:16] = v_mov_b32 %0:v[0][16:32] dst_preserve
 178       //! v1: %0:v[0] = v_mov_b32 42
 179       bld.pseudo(aco_opcode::p_unit_test, Operand(11u));
 180       bld.pseudo(aco_opcode::p_parallelcopy,
 181                  Definition(v0_lo, v1), Definition(v1_lo, v2b),
 182                  Operand(42u), Operand(v0_hi, v2b));
 183
 184       //! s_endpgm
 185
 186       finish_to_hw_instr_test();
 187    }
 188 END_TEST