1 /* Copyright (c) 2018-2019 Alyssa Rosenzweig ([email protected]) 2 * Copyright (C) 2019-2020 Collabora, Ltd. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a copy 5 * of this software and associated documentation files (the "Software"), to deal 6 * in the Software without restriction, including without limitation the rights 7 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 8 * copies of the Software, and to permit persons to whom the Software is 9 * furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 17 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 19 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 20 * THE SOFTWARE. 21 */ 22 23 #include "midgard.h" 24 25 /* Include the definitions of the macros and such */ 26 27 #define MIDGARD_OPS_TABLE 28 #include "helpers.h" 29 #undef MIDGARD_OPS_TABLE 30 31 #include "midgard_ops.h" 32 33 /* 34 * Table of mapping opcodes to accompanying properties. This is used for both 35 * the disassembler and the compiler. It is placed in a .c file like this to 36 * avoid duplications in the binary. 37 */ 38 39 /* clang-format off */ 40 struct mir_op_props alu_opcode_props[256] = { 41 [midgard_alu_op_fadd] = {"FADD", UNITS_ADD | OP_COMMUTES}, 42 [midgard_alu_op_fadd_rtz] = {"FADD.rtz", UNITS_ADD | OP_COMMUTES}, 43 [midgard_alu_op_fadd_rtn] = {"FADD.rtn", UNITS_ADD | OP_COMMUTES}, 44 [midgard_alu_op_fadd_rtp] = {"FADD.rtp", UNITS_ADD | OP_COMMUTES}, 45 [midgard_alu_op_fmul] = {"FMUL", UNITS_MUL | UNIT_VLUT | OP_COMMUTES}, 46 [midgard_alu_op_fmul_rtz] = {"FMUL.rtz", UNITS_MUL | UNIT_VLUT | OP_COMMUTES}, 47 [midgard_alu_op_fmul_rtn] = {"FMUL.rtn", UNITS_MUL | UNIT_VLUT | OP_COMMUTES}, 48 [midgard_alu_op_fmul_rtp] = {"FMUL.rtp", UNITS_MUL | UNIT_VLUT | OP_COMMUTES}, 49 [midgard_alu_op_fmin] = {"FMIN", UNITS_MOST | OP_COMMUTES}, 50 [midgard_alu_op_fmin_nan] = {"FMIN.nan", UNITS_MOST | OP_COMMUTES}, 51 [midgard_alu_op_fabsmin] = {"FABSMIN", UNITS_MOST | OP_COMMUTES}, 52 [midgard_alu_op_fabsmin_nan] = {"FABSMIN.nan", UNITS_MOST | OP_COMMUTES}, 53 [midgard_alu_op_fmax] = {"FMAX", UNITS_MOST | OP_COMMUTES}, 54 [midgard_alu_op_fmax_nan] = {"FMAX.nan", UNITS_MOST | OP_COMMUTES}, 55 [midgard_alu_op_fabsmax] = {"FABSMAX", UNITS_MOST | OP_COMMUTES}, 56 [midgard_alu_op_fabsmax_nan] = {"FABSMAX.nan", UNITS_MOST | OP_COMMUTES}, 57 [midgard_alu_op_imin] = {"MIN", UNITS_MOST | OP_COMMUTES}, 58 [midgard_alu_op_imax] = {"MAX", UNITS_MOST | OP_COMMUTES}, 59 [midgard_alu_op_umin] = {"MIN", UNITS_MOST | OP_COMMUTES}, 60 [midgard_alu_op_umax] = {"MAX", UNITS_MOST | OP_COMMUTES}, 61 [midgard_alu_op_iavg] = {"AVG.rtz", UNITS_ADD | OP_COMMUTES}, 62 [midgard_alu_op_uavg] = {"AVG.rtz", UNITS_ADD | OP_COMMUTES}, 63 [midgard_alu_op_iravg] = {"AVG.round", UNITS_ADD | OP_COMMUTES}, 64 [midgard_alu_op_uravg] = {"AVG.round", UNITS_ADD | OP_COMMUTES}, 65 66 [midgard_alu_op_fmov] = {"FMOV", UNITS_ALL | QUIRK_FLIPPED_R24}, 67 [midgard_alu_op_fmov_rtz] = {"FMOV.rtz", UNITS_ALL | QUIRK_FLIPPED_R24}, 68 [midgard_alu_op_fmov_rtn] = {"FMOV.rtn", UNITS_ALL | QUIRK_FLIPPED_R24}, 69 [midgard_alu_op_fmov_rtp] = {"FMOV.rtp", UNITS_ALL | QUIRK_FLIPPED_R24}, 70 [midgard_alu_op_froundaway] = {"FROUNDAWAY", UNITS_ADD}, 71 [midgard_alu_op_froundeven] = {"FROUNDEVEN", UNITS_ADD}, 72 [midgard_alu_op_ftrunc] = {"FTRUNC", UNITS_ADD}, 73 [midgard_alu_op_ffloor] = {"FFLOOR", UNITS_ADD}, 74 [midgard_alu_op_fceil] = {"FCEIL", UNITS_ADD}, 75 76 /* Multiplies the X/Y components of the first arg and adds the second 77 * arg. Like other LUTs, it must be scalarized. */ 78 [midgard_alu_op_ffma] = {"FMA", UNIT_VLUT}, 79 [midgard_alu_op_ffma_rtz] = {"FMA.rtz", UNIT_VLUT}, 80 [midgard_alu_op_ffma_rtn] = {"FMA.rtn", UNIT_VLUT}, 81 [midgard_alu_op_ffma_rtp] = {"FMA.rtp", UNIT_VLUT}, 82 83 /* Though they output a scalar, they need to run on a vector unit 84 * since they process vectors */ 85 [midgard_alu_op_fdot3] = {"FDOT3", UNIT_VMUL | OP_CHANNEL_COUNT(3) | OP_COMMUTES}, 86 [midgard_alu_op_fdot3r] = {"FDOT3R", UNIT_VMUL | OP_CHANNEL_COUNT(3) | OP_COMMUTES}, 87 [midgard_alu_op_fdot4] = {"FDOT4", UNIT_VMUL | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 88 89 /* Incredibly, iadd can run on vmul, etc */ 90 [midgard_alu_op_iadd] = {"ADD", UNITS_MOST | OP_COMMUTES}, 91 [midgard_alu_op_ishladd] = {"ADD", UNITS_MUL}, 92 [midgard_alu_op_iaddsat] = {"ADDSAT", UNITS_ADD | OP_COMMUTES}, 93 [midgard_alu_op_uaddsat] = {"ADDSAT", UNITS_ADD | OP_COMMUTES}, 94 [midgard_alu_op_uabsdiff] = {"ABSDIFF", UNITS_ADD}, 95 [midgard_alu_op_iabsdiff] = {"ABSDIFF", UNITS_ADD}, 96 [midgard_alu_op_ichoose] = {"CHOOSE", UNITS_ADD}, 97 [midgard_alu_op_isub] = {"SUB", UNITS_MOST}, 98 [midgard_alu_op_ishlsub] = {"SUB", UNITS_MUL}, 99 [midgard_alu_op_isubsat] = {"SUBSAT", UNITS_ADD}, 100 [midgard_alu_op_usubsat] = {"SUBSAT", UNITS_ADD}, 101 [midgard_alu_op_imul] = {"MUL", UNITS_MUL | OP_COMMUTES}, 102 [midgard_alu_op_iwmul] = {"WMUL.s", UNIT_VMUL | OP_COMMUTES}, 103 [midgard_alu_op_uwmul] = {"WMUL.u", UNIT_VMUL | OP_COMMUTES}, 104 [midgard_alu_op_iuwmul] = {"WMUL.su", UNIT_VMUL | OP_COMMUTES}, 105 [midgard_alu_op_imov] = {"MOV", UNITS_ALL | QUIRK_FLIPPED_R24}, 106 107 /* For vector comparisons, use ball etc */ 108 [midgard_alu_op_feq] = {"FCMP.eq", UNITS_MOST | OP_TYPE_CONVERT | OP_COMMUTES}, 109 [midgard_alu_op_fne] = {"FCMP.ne", UNITS_MOST | OP_TYPE_CONVERT | OP_COMMUTES}, 110 [midgard_alu_op_fle] = {"FCMP.le", UNITS_MOST | OP_TYPE_CONVERT}, 111 [midgard_alu_op_flt] = {"FCMP.lt", UNITS_MOST | OP_TYPE_CONVERT}, 112 [midgard_alu_op_ieq] = {"CMP.eq", UNITS_MOST | OP_COMMUTES}, 113 [midgard_alu_op_ine] = {"CMP.ne", UNITS_MOST | OP_COMMUTES}, 114 [midgard_alu_op_ilt] = {"CMP.lt", UNITS_MOST}, 115 [midgard_alu_op_ile] = {"CMP.le", UNITS_MOST}, 116 [midgard_alu_op_ult] = {"CMP.lt", UNITS_MOST}, 117 [midgard_alu_op_ule] = {"CMP.le", UNITS_MOST}, 118 119 /* CSEL (MUX) runs in the second pipeline stage, sourcing its selector 120 * the previous scalar or vector stage as indicated in the opcode. It 121 * muxes individual bits based on the selector, implementing both 122 * bit_select and bcsel (the latter because CMP returns 0/~0 booleans). 123 * 124 * It is legal to schedule (F)CSEL.vector to the scalar unit, but it 125 * isn't usually useful. Our scheduler does not handle that case, so 126 * don't try to and fall over. 127 */ 128 [midgard_alu_op_icsel] = {"CSEL.scalar", UNIT_VADD | UNIT_SMUL}, 129 [midgard_alu_op_icsel_v] = {"CSEL.vector", UNIT_VADD}, 130 [midgard_alu_op_fcsel_v] = {"FCSEL.vector", UNIT_VADD}, 131 [midgard_alu_op_fcsel] = {"FCSEL.scalar", UNIT_VADD | UNIT_SMUL}, 132 133 [midgard_alu_op_frcp] = {"FRCP", UNIT_VLUT}, 134 [midgard_alu_op_frsqrt] = {"FRSQRT", UNIT_VLUT}, 135 [midgard_alu_op_fsqrt] = {"FSQRT", UNIT_VLUT}, 136 [midgard_alu_op_fpow_pt1] = {"FPOW_PT1", UNIT_VLUT}, 137 [midgard_alu_op_fpown_pt1] = {"FPOWN_PT1", UNIT_VLUT}, 138 [midgard_alu_op_fpowr_pt1] = {"FPOWR_PT1", UNIT_VLUT}, 139 [midgard_alu_op_fexp2] = {"FEXP2", UNIT_VLUT}, 140 [midgard_alu_op_flog2] = {"FLOG2", UNIT_VLUT}, 141 142 [midgard_alu_op_f2i_rte] = {"F2I", UNITS_ADD | OP_TYPE_CONVERT | MIDGARD_ROUNDS}, 143 [midgard_alu_op_f2i_rtz] = {"F2I.rtz", UNITS_ADD | OP_TYPE_CONVERT}, 144 [midgard_alu_op_f2i_rtn] = {"F2I.rtn", UNITS_ADD | OP_TYPE_CONVERT}, 145 [midgard_alu_op_f2i_rtp] = {"F2I.rtp", UNITS_ADD | OP_TYPE_CONVERT}, 146 [midgard_alu_op_f2u_rte] = {"F2U", UNITS_ADD | OP_TYPE_CONVERT | MIDGARD_ROUNDS}, 147 [midgard_alu_op_f2u_rtz] = {"F2U.rtz", UNITS_ADD | OP_TYPE_CONVERT}, 148 [midgard_alu_op_f2u_rtn] = {"F2U.rtn", UNITS_ADD | OP_TYPE_CONVERT}, 149 [midgard_alu_op_f2u_rtp] = {"F2U.rtp", UNITS_ADD | OP_TYPE_CONVERT}, 150 [midgard_alu_op_i2f_rte] = {"I2F", UNITS_ADD | OP_TYPE_CONVERT}, 151 [midgard_alu_op_i2f_rtz] = {"I2F.rtz", UNITS_ADD | OP_TYPE_CONVERT}, 152 [midgard_alu_op_i2f_rtn] = {"I2F.rtn", UNITS_ADD | OP_TYPE_CONVERT}, 153 [midgard_alu_op_i2f_rtp] = {"I2F.rtp", UNITS_ADD | OP_TYPE_CONVERT}, 154 [midgard_alu_op_u2f_rte] = {"U2F", UNITS_ADD | OP_TYPE_CONVERT}, 155 [midgard_alu_op_u2f_rtz] = {"U2F.rtz", UNITS_ADD | OP_TYPE_CONVERT}, 156 [midgard_alu_op_u2f_rtn] = {"U2F.rtn", UNITS_ADD | OP_TYPE_CONVERT}, 157 [midgard_alu_op_u2f_rtp] = {"U2F.rtp", UNITS_ADD | OP_TYPE_CONVERT}, 158 159 [midgard_alu_op_fsinpi] = {"FSINPI", UNIT_VLUT}, 160 [midgard_alu_op_fcospi] = {"FCOSPI", UNIT_VLUT}, 161 162 [midgard_alu_op_iand] = {"AND", UNITS_MOST | OP_COMMUTES}, 163 [midgard_alu_op_iandnot] = {"ANDNOT", UNITS_MOST}, 164 165 [midgard_alu_op_ior] = {"OR", UNITS_MOST | OP_COMMUTES}, 166 [midgard_alu_op_iornot] = {"ORNOT", UNITS_MOST | OP_COMMUTES}, 167 [midgard_alu_op_inor] = {"NOR", UNITS_MOST | OP_COMMUTES}, 168 [midgard_alu_op_ixor] = {"XOR", UNITS_MOST | OP_COMMUTES}, 169 [midgard_alu_op_inxor] = {"NXOR", UNITS_MOST | OP_COMMUTES}, 170 [midgard_alu_op_iclz] = {"CLZ", UNITS_ADD}, 171 [midgard_alu_op_ipopcnt] = {"POPCNT", UNIT_VADD}, 172 [midgard_alu_op_inand] = {"NAND", UNITS_MOST}, 173 [midgard_alu_op_ishl] = {"SHL", UNITS_ADD}, 174 [midgard_alu_op_ishlsat] = {"SHL.sat", UNITS_ADD}, 175 [midgard_alu_op_ushlsat] = {"SHL.sat", UNITS_ADD}, 176 [midgard_alu_op_iasr] = {"ASR", UNITS_ADD}, 177 [midgard_alu_op_ilsr] = {"LSR", UNITS_ADD}, 178 179 [midgard_alu_op_fball_eq] = {"FCMP.all.eq", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES | OP_TYPE_CONVERT}, 180 [midgard_alu_op_fball_neq] = {"FCMP.all.ne", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES | OP_TYPE_CONVERT}, 181 [midgard_alu_op_fball_lt] = {"FCMP.all.lt", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES | OP_TYPE_CONVERT}, 182 [midgard_alu_op_fball_lte] = {"FCMP.all.le", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES | OP_TYPE_CONVERT}, 183 184 [midgard_alu_op_fbany_eq] = {"FCMP.any.eq", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES | OP_TYPE_CONVERT}, 185 [midgard_alu_op_fbany_neq] = {"FCMP.any.ne", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES | OP_TYPE_CONVERT}, 186 [midgard_alu_op_fbany_lt] = {"FCMP.any.lt", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES | OP_TYPE_CONVERT}, 187 [midgard_alu_op_fbany_lte] = {"FCMP.any.le", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES | OP_TYPE_CONVERT}, 188 189 [midgard_alu_op_iball_eq] = {"CMP.all.eq", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 190 [midgard_alu_op_iball_neq] = {"CMP.all.ne", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 191 [midgard_alu_op_iball_lt] = {"CMP.all.lt", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 192 [midgard_alu_op_iball_lte] = {"CMP.all.le", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 193 [midgard_alu_op_uball_lt] = {"CMP.all.lt", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 194 [midgard_alu_op_uball_lte] = {"CMP.all.le", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 195 196 [midgard_alu_op_ibany_eq] = {"CMP.any.eq", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 197 [midgard_alu_op_ibany_neq] = {"CMP.any.ne", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 198 [midgard_alu_op_ibany_lt] = {"CMP.any.lt", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 199 [midgard_alu_op_ibany_lte] = {"CMP.any.le", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 200 [midgard_alu_op_ubany_lt] = {"CMP.any.lt", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 201 [midgard_alu_op_ubany_lte] = {"CMP.any.le", UNITS_VECTOR | OP_CHANNEL_COUNT(4) | OP_COMMUTES}, 202 203 [midgard_alu_op_fatan2_pt1] = {"FATAN2_PT1", UNIT_VLUT}, 204 [midgard_alu_op_fatan2_pt2] = {"FATAN2_PT2", UNIT_VLUT}, 205 206 /* Haven't seen in a while */ 207 [midgard_alu_op_freduce] = {"FREDUCE", 0}, 208 }; 209 210 /* Define shorthands */ 211 212 #define M8 midgard_reg_mode_8 213 #define M16 midgard_reg_mode_16 214 #define M32 midgard_reg_mode_32 215 #define M64 midgard_reg_mode_64 216 217 struct mir_ldst_op_props load_store_opcode_props[256] = { 218 [midgard_op_unpack_colour_f32] = {"UNPACK.f32", M32}, 219 [midgard_op_unpack_colour_f16] = {"UNPACK.f16", M32}, 220 [midgard_op_unpack_colour_u32] = {"UNPACK.u32", M32}, 221 [midgard_op_unpack_colour_s32] = {"UNPACK.s32", M32}, 222 [midgard_op_pack_colour_f32] = {"PACK.f32", M32}, 223 [midgard_op_pack_colour_f16] = {"PACK.f16", M32}, 224 [midgard_op_pack_colour_u32] = {"PACK.u32", M32}, 225 [midgard_op_pack_colour_s32] = {"PACK.s32", M32}, 226 [midgard_op_lea] = {"LEA", M32 | LDST_ADDRESS }, 227 [midgard_op_lea_image] = {"LEA_IMAGE", M32 | LDST_ATTRIB }, 228 [midgard_op_ld_cubemap_coords] = {"CUBEMAP", M32}, 229 [midgard_op_ldst_mov] = {"LDST_MOV", M32}, 230 [midgard_op_ldst_perspective_div_y] = {"LDST_PERSPECTIVE_DIV_Y", M32}, 231 [midgard_op_ldst_perspective_div_z] = {"LDST_PERSPECTIVE_DIV_Z", M32}, 232 [midgard_op_ldst_perspective_div_w] = {"LDST_PERSPECTIVE_DIV_W", M32}, 233 234 [midgard_op_atomic_add] = {"AADD.32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 235 [midgard_op_atomic_and] = {"AAND.32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 236 [midgard_op_atomic_or] = {"AOR.32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 237 [midgard_op_atomic_xor] = {"AXOR.32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 238 [midgard_op_atomic_imin] = {"AMIN.s32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 239 [midgard_op_atomic_umin] = {"AMIN.u32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 240 [midgard_op_atomic_imax] = {"AMAX.s32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 241 [midgard_op_atomic_umax] = {"AMAX.u32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 242 [midgard_op_atomic_xchg] = {"XCHG.32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 243 [midgard_op_atomic_cmpxchg] = {"CMPXCHG.32", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 244 245 [midgard_op_atomic_add64] = {"AADD.64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 246 [midgard_op_atomic_and64] = {"AAND.64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 247 [midgard_op_atomic_or64] = {"AOR.64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 248 [midgard_op_atomic_xor64] = {"AXOR.64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 249 [midgard_op_atomic_imin64] = {"AMIN.s64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 250 [midgard_op_atomic_umin64] = {"AMIN.u64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 251 [midgard_op_atomic_imax64] = {"AMAX.s64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 252 [midgard_op_atomic_umax64] = {"AMAX.u64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 253 [midgard_op_atomic_xchg64] = {"XCHG.64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 254 [midgard_op_atomic_cmpxchg64] = {"CMPXCHG.64", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 255 256 [midgard_op_atomic_add_be] = {"AADD.32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 257 [midgard_op_atomic_and_be] = {"AAND.32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 258 [midgard_op_atomic_or_be] = {"AOR.32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 259 [midgard_op_atomic_xor_be] = {"AXOR.32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 260 [midgard_op_atomic_imin_be] = {"AMIN.s32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 261 [midgard_op_atomic_umin_be] = {"AMIN.u32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 262 [midgard_op_atomic_imax_be] = {"AMAX.s32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 263 [midgard_op_atomic_umax_be] = {"AMAX.u32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 264 [midgard_op_atomic_xchg_be] = {"XCHG.32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 265 [midgard_op_atomic_cmpxchg_be] = {"CMPXCHG.32.be", M32 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 266 267 [midgard_op_atomic_add64] = {"AADD.64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 268 [midgard_op_atomic_and64] = {"AAND.64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 269 [midgard_op_atomic_or64] = {"AOR.64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 270 [midgard_op_atomic_xor64] = {"AXOR.64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 271 [midgard_op_atomic_imin64] = {"AMIN.s64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 272 [midgard_op_atomic_umin64] = {"AMIN.u64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 273 [midgard_op_atomic_imax64] = {"AMAX.s64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 274 [midgard_op_atomic_umax64] = {"AMAX.u64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 275 [midgard_op_atomic_xchg64] = {"XCHG.64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 276 [midgard_op_atomic_cmpxchg64] = {"CMPXCHG.64.be", M64 | LDST_SIDE_FX | LDST_ADDRESS | LDST_ATOMIC}, 277 278 [midgard_op_ld_u8] = {"LD.u8", M32 | LDST_ADDRESS}, 279 [midgard_op_ld_i8] = {"LD.s8", M32 | LDST_ADDRESS}, 280 [midgard_op_ld_u16] = {"LD.u16", M32 | LDST_ADDRESS}, 281 [midgard_op_ld_i16] = {"LD.s16", M32 | LDST_ADDRESS}, 282 [midgard_op_ld_u16_be] = {"LD.u16.be", M32 | LDST_ADDRESS}, 283 [midgard_op_ld_i16_be] = {"LD.s16.be", M32 | LDST_ADDRESS}, 284 [midgard_op_ld_32] = {"LD.32", M32 | LDST_ADDRESS}, 285 [midgard_op_ld_32_bswap2] = {"LD.32.bswap2", M32 | LDST_ADDRESS}, 286 [midgard_op_ld_32_bswap4] = {"LD.32.bswap4", M32 | LDST_ADDRESS}, 287 [midgard_op_ld_64] = {"LD.64", M32 | LDST_ADDRESS}, 288 [midgard_op_ld_64_bswap2] = {"LD.64.bswap2", M32 | LDST_ADDRESS}, 289 [midgard_op_ld_64_bswap4] = {"LD.64.bswap4", M32 | LDST_ADDRESS}, 290 [midgard_op_ld_64_bswap8] = {"LD.64.bswap8", M32 | LDST_ADDRESS}, 291 [midgard_op_ld_128] = {"LD.128", M32 | LDST_ADDRESS}, 292 [midgard_op_ld_128_bswap2] = {"LD.128.bswap2", M32 | LDST_ADDRESS}, 293 [midgard_op_ld_128_bswap4] = {"LD.128.bswap4", M32 | LDST_ADDRESS}, 294 [midgard_op_ld_128_bswap8] = {"LD.128.bswap8", M32 | LDST_ADDRESS}, 295 296 [midgard_op_ld_attr_32] = {"LD_ATTR.f32", M32 | LDST_ATTRIB}, 297 [midgard_op_ld_attr_32i] = {"LD_ATTR.s32", M32 | LDST_ATTRIB}, 298 [midgard_op_ld_attr_32u] = {"LD_ATTR.u32", M32 | LDST_ATTRIB}, 299 [midgard_op_ld_attr_16] = {"LD_ATTR.f16", M32 | LDST_ATTRIB}, 300 301 [midgard_op_ld_vary_32] = {"LD_VARY.f32", M32 | LDST_ATTRIB}, 302 [midgard_op_ld_vary_16] = {"LD_VARY.f16", M32 | LDST_ATTRIB}, 303 [midgard_op_ld_vary_32i] = {"LD_VARY.s32", M32 | LDST_ATTRIB}, 304 [midgard_op_ld_vary_32u] = {"LD_VARY.u32", M32 | LDST_ATTRIB}, 305 306 [midgard_op_ld_special_32f] = {"LD_SPECIAL.f32", M32 | LDST_SPECIAL_MASK}, 307 [midgard_op_ld_special_16f] = {"LD_SPECIAL.f16", M16 | LDST_SPECIAL_MASK}, 308 [midgard_op_ld_special_32u] = {"LD_SPECIAL.u32", M32}, 309 [midgard_op_ld_special_32i] = {"LD_SPECIAL.s32", M32}, 310 311 [midgard_op_ld_tilebuffer_32f] = {"LD_TILEBUFFER.f32", M32}, 312 [midgard_op_ld_tilebuffer_16f] = {"LD_TILEBUFFER.f16", M16}, 313 [midgard_op_ld_tilebuffer_raw] = {"LD_TILEBUFFER.raw", M32}, 314 315 [midgard_op_ld_ubo_u8] = {"LD_UBO.u8", M32}, 316 [midgard_op_ld_ubo_i8] = {"LD_UBO.s8", M32}, 317 [midgard_op_ld_ubo_u16] = {"LD_UBO.u16", M16}, 318 [midgard_op_ld_ubo_i16] = {"LD_UBO.s16", M16}, 319 [midgard_op_ld_ubo_u16_be] = {"LD_UBO.u16.be", M16}, 320 [midgard_op_ld_ubo_i16_be] = {"LD_UBO.s16.be", M16}, 321 [midgard_op_ld_ubo_32] = {"LD_UBO.32", M32}, 322 [midgard_op_ld_ubo_32_bswap2] = {"LD_UBO.32.bswap2", M32}, 323 [midgard_op_ld_ubo_32_bswap4] = {"LD_UBO.32.bswap4", M32}, 324 [midgard_op_ld_ubo_64] = {"LD_UBO.64", M32}, 325 [midgard_op_ld_ubo_64_bswap2] = {"LD_UBO.64.bswap2", M32}, 326 [midgard_op_ld_ubo_64_bswap4] = {"LD_UBO.64.bswap4", M32}, 327 [midgard_op_ld_ubo_64_bswap8] = {"LD_UBO.64.bswap8", M32}, 328 [midgard_op_ld_ubo_128] = {"LD_UBO.128", M32}, 329 [midgard_op_ld_ubo_128_bswap2] = {"LD_UBO.128.bswap2", M32}, 330 [midgard_op_ld_ubo_128_bswap4] = {"LD_UBO.128.bswap4", M32}, 331 [midgard_op_ld_ubo_128_bswap8] = {"LD_UBO.128.bswap8", M32}, 332 333 [midgard_op_ld_image_32f] = {"LD_IMAGE.f32", M32 | LDST_ATTRIB}, 334 [midgard_op_ld_image_16f] = {"LD_IMAGE.f16", M16 | LDST_ATTRIB}, 335 [midgard_op_ld_image_32i] = {"LD_IMAGE.s32", M32 | LDST_ATTRIB}, 336 [midgard_op_ld_image_32u] = {"LD_IMAGE.u32", M32 | LDST_ATTRIB}, 337 338 [midgard_op_st_u8] = {"ST.u8", M32 | LDST_STORE | LDST_ADDRESS}, 339 [midgard_op_st_i8] = {"ST.s8", M32 | LDST_STORE | LDST_ADDRESS}, 340 [midgard_op_st_u16] = {"ST.u16", M32 | LDST_STORE | LDST_ADDRESS}, 341 [midgard_op_st_i16] = {"ST.s16", M32 | LDST_STORE | LDST_ADDRESS}, 342 [midgard_op_st_u16_be] = {"ST.u16.be", M32 | LDST_STORE | LDST_ADDRESS}, 343 [midgard_op_st_i16_be] = {"ST.s16.be", M32 | LDST_STORE | LDST_ADDRESS}, 344 [midgard_op_st_32] = {"ST.32", M32 | LDST_STORE | LDST_ADDRESS}, 345 [midgard_op_st_32_bswap2] = {"ST.32.bswap2", M32 | LDST_STORE | LDST_ADDRESS}, 346 [midgard_op_st_32_bswap4] = {"ST.32.bswap4", M32 | LDST_STORE | LDST_ADDRESS}, 347 [midgard_op_st_64] = {"ST.64", M32 | LDST_STORE | LDST_ADDRESS}, 348 [midgard_op_st_64_bswap2] = {"ST.64.bswap2", M32 | LDST_STORE | LDST_ADDRESS}, 349 [midgard_op_st_64_bswap4] = {"ST.64.bswap4", M32 | LDST_STORE | LDST_ADDRESS}, 350 [midgard_op_st_64_bswap8] = {"ST.64.bswap8", M32 | LDST_STORE | LDST_ADDRESS}, 351 [midgard_op_st_128] = {"ST.128", M32 | LDST_STORE | LDST_ADDRESS}, 352 [midgard_op_st_128_bswap2] = {"ST.128.bswap2", M32 | LDST_STORE | LDST_ADDRESS}, 353 [midgard_op_st_128_bswap4] = {"ST.128.bswap4", M32 | LDST_STORE | LDST_ADDRESS}, 354 [midgard_op_st_128_bswap8] = {"ST.128.bswap8", M32 | LDST_STORE | LDST_ADDRESS}, 355 356 [midgard_op_st_vary_32] = {"ST_VARY.f32", M32 | LDST_STORE | LDST_ATTRIB}, 357 [midgard_op_st_vary_32i] = {"ST_VARY.s32", M32 | LDST_STORE | LDST_ATTRIB}, 358 [midgard_op_st_vary_32u] = {"ST_VARY.u32", M32 | LDST_STORE | LDST_ATTRIB}, 359 [midgard_op_st_vary_16] = {"ST_VARY.f16", M16 | LDST_STORE | LDST_ATTRIB}, 360 361 [midgard_op_st_image_32f] = {"ST_IMAGE.f32", M32 | LDST_STORE | LDST_ATTRIB}, 362 [midgard_op_st_image_16f] = {"ST_IMAGE.f16", M16 | LDST_STORE | LDST_ATTRIB}, 363 [midgard_op_st_image_32i] = {"ST_IMAGE.u32", M32 | LDST_STORE | LDST_ATTRIB}, 364 [midgard_op_st_image_32u] = {"ST_IMAGE.s32", M32 | LDST_STORE | LDST_ATTRIB}, 365 366 [midgard_op_st_special_32f] = {"ST_SPECIAL.f32", M32}, 367 [midgard_op_st_special_16f] = {"ST_SPECIAL.f16", M16}, 368 [midgard_op_st_special_32u] = {"ST_SPECIAL.u32", M32}, 369 [midgard_op_st_special_32i] = {"ST_SPECIAL.s32", M32}, 370 371 [midgard_op_st_tilebuffer_32f] = {"ST_TILEBUFFER.f32", M32}, 372 [midgard_op_st_tilebuffer_16f] = {"ST_TILEBUFFER.f16", M16}, 373 [midgard_op_st_tilebuffer_raw] = {"ST_TILEBUFFER.raw", M32}, 374 }; 375 376 struct mir_tex_op_props tex_opcode_props[16] = { 377 [midgard_tex_op_normal] = {"TEX", M32}, 378 [midgard_tex_op_gradient] = {"TEX_GRAD", M32}, 379 [midgard_tex_op_fetch] = {"TEX_FETCH", M32}, 380 [midgard_tex_op_grad_from_derivative] = {"DER_TO_GRAD", M32}, 381 [midgard_tex_op_grad_from_coords] = {"COORDS_TO_GRAD", M32}, 382 [midgard_tex_op_mov] = {"MOV", M32}, 383 [midgard_tex_op_barrier] = {"BARRIER", M32}, 384 [midgard_tex_op_derivative] = {"DERIVATIVE", M32} 385 }; 386 387 #undef M8 388 #undef M16 389 #undef M32 390 #undef M64 391 392 struct mir_tag_props midgard_tag_props[16] = { 393 [TAG_INVALID] = {"invalid", 0}, 394 [TAG_BREAK] = {"break", 0}, 395 [TAG_TEXTURE_4_VTX] = {"tex/vt", 1}, 396 [TAG_TEXTURE_4] = {"tex", 1}, 397 [TAG_TEXTURE_4_BARRIER] = {"tex/bar", 1}, 398 [TAG_LOAD_STORE_4] = {"ldst", 1}, 399 [TAG_UNKNOWN_1] = {"unk1", 1}, 400 [TAG_UNKNOWN_2] = {"unk2", 1}, 401 [TAG_ALU_4] = {"alu/4", 1}, 402 [TAG_ALU_8] = {"alu/8", 2}, 403 [TAG_ALU_12] = {"alu/12", 3}, 404 [TAG_ALU_16] = {"alu/16", 4}, 405 [TAG_ALU_4_WRITEOUT] = {"aluw/4", 1}, 406 [TAG_ALU_8_WRITEOUT] = {"aluw/8", 2}, 407 [TAG_ALU_12_WRITEOUT] = {"aluw/12", 3}, 408 [TAG_ALU_16_WRITEOUT] = {"aluw/16", 4} 409 }; 410 /* clang-format on */ 411