llvm/test/MC/AMDGPU/gfx11-promotions.s

   1 // RUN: llvm-mc -triple=amdgcn -show-encoding -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 %s | FileCheck --check-prefix=GFX11 %s
   2
   3 // Check opcode promotions and forced suffices.
   4 // 1. When a suffix is optional, check that it may be omitted.
   5 // 2. When a suffix is optional, check that it may be specified w/o any effect.
   6 // 3. When a suffix is required, check that specifying it enforces opcode promotion.
   7 // 4. When a suffix is required, check that omitting the suffix results in a different encoding.
   8
   9 //===----------------------------------------------------------------------===//
  10 // VOP1.
  11 //===----------------------------------------------------------------------===//
  12
  13 v_mov_b32 v0, v1
  14 // GFX11: encoding: [0x01,0x03,0x00,0x7e]
  15
  16 v_mov_b32_e32 v0, v1
  17 // GFX11: encoding: [0x01,0x03,0x00,0x7e]
  18
  19 //===----------------------------------------------------------------------===//
  20 // VOP2.
  21 //===----------------------------------------------------------------------===//
  22
  23 v_add_f16 v5, v1, v2
  24 // GFX11: encoding: [0x01,0x05,0x0a,0x64]
  25
  26 v_add_f16_e32 v5, v1, v2
  27 // GFX11: encoding: [0x01,0x05,0x0a,0x64]
  28
  29 //===----------------------------------------------------------------------===//
  30 // VOPC.
  31 //===----------------------------------------------------------------------===//
  32
  33 v_cmp_lt_f32 vcc_lo, v1, v2
  34 // GFX11: encoding: [0x01,0x05,0x22,0x7c]
  35
  36 v_cmp_lt_f32_e32 vcc_lo, v1, v2
  37 // GFX11: encoding: [0x01,0x05,0x22,0x7c]
  38
  39 //===----------------------------------------------------------------------===//
  40 // VOPCX.
  41 //===----------------------------------------------------------------------===//
  42
  43 v_cmpx_class_f16 v1, v2
  44 // GFX11: encoding: [0x01,0x05,0xfa,0x7d]
  45
  46 v_cmpx_class_f16_e32 v1, v2
  47 // GFX11: encoding: [0x01,0x05,0xfa,0x7d]
  48
  49 //===----------------------------------------------------------------------===//
  50 // VOP1.DPP8.
  51 //===----------------------------------------------------------------------===//
  52
  53 v_bfrev_b32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
  54 // GFX11: encoding: [0xe9,0x70,0x0a,0x7e,0x01,0x77,0x39,0x05]
  55
  56 v_bfrev_b32_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
  57 // GFX11: encoding: [0xe9,0x70,0x0a,0x7e,0x01,0x77,0x39,0x05]
  58
  59 //===----------------------------------------------------------------------===//
  60 // VOP1.DPP16.
  61 //===----------------------------------------------------------------------===//
  62
  63 v_bfrev_b32 v5, v1 quad_perm:[3,2,1,0]
  64 // GFX11: encoding: [0xfa,0x70,0x0a,0x7e,0x01,0x1b,0x00,0xff]
  65
  66 v_bfrev_b32_dpp v5, v1 quad_perm:[3,2,1,0]
  67 // GFX11: encoding: [0xfa,0x70,0x0a,0x7e,0x01,0x1b,0x00,0xff]
  68
  69 //===----------------------------------------------------------------------===//
  70 // VOP2.DPP8.
  71 //===----------------------------------------------------------------------===//
  72
  73 v_add_f16 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0]
  74 // GFX11: encoding: [0xe9,0x04,0x0a,0x64,0x01,0x77,0x39,0x05]
  75
  76 v_add_f16_dpp v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0]
  77 // GFX11: encoding: [0xe9,0x04,0x0a,0x64,0x01,0x77,0x39,0x05]
  78
  79 //===----------------------------------------------------------------------===//
  80 // VOP2.DPP16.
  81 //===----------------------------------------------------------------------===//
  82
  83 v_add_f16 v5, v1, v2 quad_perm:[3,2,1,0]
  84 // GFX11: encoding: [0xfa,0x04,0x0a,0x64,0x01,0x1b,0x00,0xff]
  85
  86 v_add_f16_dpp v5, v1, v2 quad_perm:[3,2,1,0]
  87 // GFX11: encoding: [0xfa,0x04,0x0a,0x64,0x01,0x1b,0x00,0xff]
  88
  89 //===----------------------------------------------------------------------===//
  90 // VOPC.DPP8.
  91 //===----------------------------------------------------------------------===//
  92
  93 v_cmp_le_u16 v1, v2 dpp8:[7,7,7,3,4,4,6,7] fi:1
  94 // GFX11: encoding: [0xea,0x04,0x76,0x7c,0x01,0xff,0x47,0xfa]
  95
  96 v_cmp_le_u16_dpp v1, v2 dpp8:[7,7,7,3,4,4,6,7] fi:1
  97 // GFX11: encoding: [0xea,0x04,0x76,0x7c,0x01,0xff,0x47,0xfa]
  98
  99 //===----------------------------------------------------------------------===//
 100 // VOPC.DPP16.
 101 //===----------------------------------------------------------------------===//
 102
 103 v_cmp_gt_u16 v1, v2 row_shl:0x7 row_mask:0x0 bank_mask:0x0 fi:1
 104 // GFX11: encoding: [0xfa,0x04,0x78,0x7c,0x01,0x07,0x05,0x00]
 105
 106 v_cmp_gt_u16_dpp v1, v2 row_shl:0x7 row_mask:0x0 bank_mask:0x0 fi:1
 107 // GFX11: encoding: [0xfa,0x04,0x78,0x7c,0x01,0x07,0x05,0x00]
 108
 109 //===----------------------------------------------------------------------===//
 110 // VOPCX.DPP8.
 111 //===----------------------------------------------------------------------===//
 112
 113 v_cmpx_class_f16 v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 114 // GFX11: encoding: [0xe9,0x04,0xfa,0x7d,0x01,0x77,0x39,0x05]
 115
 116 v_cmpx_class_f16_dpp v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 117 // GFX11: encoding: [0xe9,0x04,0xfa,0x7d,0x01,0x77,0x39,0x05]
 118
 119 //===----------------------------------------------------------------------===//
 120 // VOPCX.DPP16.
 121 //===----------------------------------------------------------------------===//
 122
 123 v_cmpx_class_f16 v1, v2 quad_perm:[3,2,1,0]
 124 // GFX11: encoding: [0xfa,0x04,0xfa,0x7d,0x01,0x1b,0x00,0xff]
 125
 126 v_cmpx_class_f16_dpp v1, v2 quad_perm:[3,2,1,0]
 127 // GFX11: encoding: [0xfa,0x04,0xfa,0x7d,0x01,0x1b,0x00,0xff]
 128
 129 //===----------------------------------------------------------------------===//
 130 // VOP1 -> VOP3.
 131 //===----------------------------------------------------------------------===//
 132
 133 v_sin_f32 v5, 0.5 mul:2
 134 // GFX11: encoding: [0x05,0x00,0xb5,0xd5,0xf0,0x00,0x00,0x08]
 135
 136 v_sin_f32_e64 v5, 0.5 mul:2
 137 // GFX11: encoding: [0x05,0x00,0xb5,0xd5,0xf0,0x00,0x00,0x08]
 138
 139 v_sin_f32_e64 v5, v1
 140 // GFX11: encoding: [0x05,0x00,0xb5,0xd5,0x01,0x01,0x00,0x00]
 141
 142 v_sin_f32 v5, v1
 143 // GFX11: encoding: [0x01,0x6b,0x0a,0x7e]
 144
 145 //===----------------------------------------------------------------------===//
 146 // VOP2 -> VOP3.
 147 //===----------------------------------------------------------------------===//
 148
 149 v_add_f32 v5, v1, -v2
 150 // GFX11: encoding: [0x05,0x00,0x03,0xd5,0x01,0x05,0x02,0x40]
 151
 152 v_add_f32_e64 v5, v1, -v2
 153 // GFX11: encoding: [0x05,0x00,0x03,0xd5,0x01,0x05,0x02,0x40]
 154
 155 v_add_f32_e64 v5, v1, v2
 156 // GFX11: encoding: [0x05,0x00,0x03,0xd5,0x01,0x05,0x02,0x00]
 157
 158 v_add_f32 v5, v1, v2
 159 // GFX11: encoding: [0x01,0x05,0x0a,0x06]
 160
 161 //===----------------------------------------------------------------------===//
 162 // VOPC -> VOP3.
 163 //===----------------------------------------------------------------------===//
 164
 165 v_cmp_f_f32 s10, -v1, v2
 166 // GFX11: encoding: [0x0a,0x00,0x10,0xd4,0x01,0x05,0x02,0x20]
 167
 168 v_cmp_f_f32_e64 s10, -v1, v2
 169 // GFX11: encoding: [0x0a,0x00,0x10,0xd4,0x01,0x05,0x02,0x20]
 170
 171 v_cmp_f_f32_e64 vcc_lo, v1, v2
 172 // GFX11: encoding: [0x6a,0x00,0x10,0xd4,0x01,0x05,0x02,0x00]
 173
 174 v_cmp_f_f32 vcc_lo, v1, v2
 175 // GFX11: encoding: [0x01,0x05,0x20,0x7c]
 176
 177 //===----------------------------------------------------------------------===//
 178 // VOPCX -> VOP3.
 179 //===----------------------------------------------------------------------===//
 180
 181 v_cmpx_f_f32 -v1, v2
 182 // GFX11: encoding: [0x7e,0x00,0x90,0xd4,0x01,0x05,0x02,0x20]
 183
 184 v_cmpx_f_f32_e64 -v1, v2
 185 // GFX11: encoding: [0x7e,0x00,0x90,0xd4,0x01,0x05,0x02,0x20]
 186
 187 v_cmpx_f_f32_e64 v1, v2
 188 // GFX11: encoding: [0x7e,0x00,0x90,0xd4,0x01,0x05,0x02,0x00]
 189
 190 v_cmpx_f_f32 v1, v2
 191 // GFX11: encoding: [0x01,0x05,0x20,0x7d]
 192
 193 //===----------------------------------------------------------------------===//
 194 // VOP3.
 195 //===----------------------------------------------------------------------===//
 196
 197 v_add3_u32 v5, v1, v2, s3
 198 // GFX11: encoding: [0x05,0x00,0x55,0xd6,0x01,0x05,0x0e,0x00]
 199
 200 v_add3_u32_e64 v5, v1, v2, s3
 201 // GFX11: encoding: [0x05,0x00,0x55,0xd6,0x01,0x05,0x0e,0x00]
 202
 203 //===----------------------------------------------------------------------===//
 204 // VOP1 -> VOP3.DPP8.
 205 //===----------------------------------------------------------------------===//
 206
 207 v_sin_f32 v5, v1 div:2 dpp8:[0,0,0,0,0,0,0,0]
 208 // GFX11: encoding: [0x05,0x00,0xb5,0xd5,0xe9,0x00,0x00,0x18,0x01,0x00,0x00,0x00]
 209
 210 v_sin_f32_e64_dpp v5, v1 div:2 dpp8:[0,0,0,0,0,0,0,0]
 211 // GFX11: encoding: [0x05,0x00,0xb5,0xd5,0xe9,0x00,0x00,0x18,0x01,0x00,0x00,0x00]
 212
 213 v_sin_f32_e64_dpp v5, v1 dpp8:[7,6,5,4,3,2,1,0]
 214 // GFX11: encoding: [0x05,0x00,0xb5,0xd5,0xe9,0x00,0x00,0x00,0x01,0x77,0x39,0x05]
 215
 216 v_sin_f32 v5, v1 dpp8:[7,6,5,4,3,2,1,0]
 217 // GFX11: encoding: [0xe9,0x6a,0x0a,0x7e,0x01,0x77,0x39,0x05]
 218
 219 //===----------------------------------------------------------------------===//
 220 // VOP2 -> VOP3.DPP8.
 221 //===----------------------------------------------------------------------===//
 222
 223 v_add_f32 v5, v1, v2 mul:4 dpp8:[7,6,5,4,3,2,1,0]
 224 // GFX11: encoding: [0x05,0x00,0x03,0xd5,0xe9,0x04,0x02,0x10,0x01,0x77,0x39,0x05]
 225
 226 v_add_f32_e64_dpp v5, v1, v2 mul:4 dpp8:[7,6,5,4,3,2,1,0]
 227 // GFX11: encoding: [0x05,0x00,0x03,0xd5,0xe9,0x04,0x02,0x10,0x01,0x77,0x39,0x05]
 228
 229 v_add_f32_e64_dpp v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 230 // GFX11: encoding: [0x05,0x00,0x03,0xd5,0xe9,0x04,0x02,0x00,0x01,0x77,0x39,0x05]
 231
 232 v_add_f32 v5, v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 233 // GFX11: encoding: [0xe9,0x04,0x0a,0x06,0x01,0x77,0x39,0x05]
 234
 235 //===----------------------------------------------------------------------===//
 236 // VOPC -> VOP3.DPP8.
 237 //===----------------------------------------------------------------------===//
 238
 239 v_cmp_class_f32 s5, v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 240 // GFX11: encoding: [0x05,0x00,0x7e,0xd4,0xe9,0x04,0x02,0x00,0x01,0x77,0x39,0x05]
 241
 242 v_cmp_class_f32_e64_dpp s5, v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 243 // GFX11: encoding: [0x05,0x00,0x7e,0xd4,0xe9,0x04,0x02,0x00,0x01,0x77,0x39,0x05]
 244
 245 v_cmp_class_f32_e64_dpp vcc_lo, v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 246 // GFX11: encoding: [0x6a,0x00,0x7e,0xd4,0xe9,0x04,0x02,0x00,0x01,0x77,0x39,0x05]
 247
 248 v_cmp_class_f32 vcc_lo, v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 249 // GFX11: encoding: [0xe9,0x04,0xfc,0x7c,0x01,0x77,0x39,0x05]
 250
 251 //===----------------------------------------------------------------------===//
 252 // VOPCX -> VOP3.DPP8.
 253 //===----------------------------------------------------------------------===//
 254
 255 v_cmpx_class_f32 -v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 256 // GFX11: encoding: [0x7e,0x00,0xfe,0xd4,0xe9,0x04,0x02,0x20,0x01,0x77,0x39,0x05]
 257
 258 v_cmpx_class_f32_e64_dpp -v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 259 // GFX11: encoding: [0x7e,0x00,0xfe,0xd4,0xe9,0x04,0x02,0x20,0x01,0x77,0x39,0x05]
 260
 261 v_cmpx_class_f32_e64_dpp v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 262 // GFX11: encoding: [0x7e,0x00,0xfe,0xd4,0xe9,0x04,0x02,0x00,0x01,0x77,0x39,0x05]
 263
 264 v_cmpx_class_f32 v1, v2 dpp8:[7,6,5,4,3,2,1,0]
 265 // GFX11: encoding: [0xe9,0x04,0xfc,0x7d,0x01,0x77,0x39,0x05]
 266
 267 //===----------------------------------------------------------------------===//
 268 // VOP1 -> VOP3.DPP16.
 269 //===----------------------------------------------------------------------===//
 270
 271 v_sin_f32 v5, v1 div:2 row_xmask:15
 272 // GFX11: encoding: [0x05,0x00,0xb5,0xd5,0xfa,0x00,0x00,0x18,0x01,0x6f,0x01,0xff]
 273
 274 v_sin_f32_e64_dpp v5, v1 div:2 row_xmask:15
 275 // GFX11: encoding: [0x05,0x00,0xb5,0xd5,0xfa,0x00,0x00,0x18,0x01,0x6f,0x01,0xff]
 276
 277 v_sin_f32_e64_dpp v5, v1 quad_perm:[3,2,1,0]
 278 // GFX11: encoding: [0x05,0x00,0xb5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]
 279
 280 v_sin_f32 v5, v1 quad_perm:[3,2,1,0]
 281 // GFX11: encoding: [0xfa,0x6a,0x0a,0x7e,0x01,0x1b,0x00,0xff]
 282
 283 //===----------------------------------------------------------------------===//
 284 // VOP2 -> VOP3.DPP16.
 285 //===----------------------------------------------------------------------===//
 286
 287 v_add_f32 v5, v1, v2 div:2 quad_perm:[3,2,1,0]
 288 // GFX11: encoding: [0x05,0x00,0x03,0xd5,0xfa,0x04,0x02,0x18,0x01,0x1b,0x00,0xff]
 289
 290 v_add_f32_e64_dpp v5, v1, v2 div:2 quad_perm:[3,2,1,0]
 291 // GFX11: encoding: [0x05,0x00,0x03,0xd5,0xfa,0x04,0x02,0x18,0x01,0x1b,0x00,0xff]
 292
 293 v_add_f32_e64_dpp v5, v1, v2 quad_perm:[3,2,1,0]
 294 // GFX11: encoding: [0x05,0x00,0x03,0xd5,0xfa,0x04,0x02,0x00,0x01,0x1b,0x00,0xff]
 295
 296 v_add_f32 v5, v1, v2 quad_perm:[3,2,1,0]
 297 // GFX11: encoding: [0xfa,0x04,0x0a,0x06,0x01,0x1b,0x00,0xff]
 298
 299 //===----------------------------------------------------------------------===//
 300 // VOPC -> VOP3.DPP16.
 301 //===----------------------------------------------------------------------===//
 302
 303 v_cmp_class_f32 s5, v1, v2 quad_perm:[3,2,1,0]
 304 // GFX11: encoding: [0x05,0x00,0x7e,0xd4,0xfa,0x04,0x02,0x00,0x01,0x1b,0x00,0xff]
 305
 306 v_cmp_class_f32_e64_dpp s5, v1, v2 quad_perm:[3,2,1,0]
 307 // GFX11: encoding: [0x05,0x00,0x7e,0xd4,0xfa,0x04,0x02,0x00,0x01,0x1b,0x00,0xff]
 308
 309 v_cmp_class_f32_e64_dpp vcc_lo, v1, v2 row_share:0 row_mask:0xf bank_mask:0xf
 310 // GFX11: encoding: [0x6a,0x00,0x7e,0xd4,0xfa,0x04,0x02,0x00,0x01,0x50,0x01,0xff]
 311
 312 v_cmp_class_f32 vcc_lo, v1, v2 row_share:0 row_mask:0xf bank_mask:0xf
 313 // GFX11: encoding: [0xfa,0x04,0xfc,0x7c,0x01,0x50,0x01,0xff]
 314
 315 //===----------------------------------------------------------------------===//
 316 // VOPCX -> VOP3.DPP16.
 317 //===----------------------------------------------------------------------===//
 318
 319 v_cmpx_class_f32_e64_dpp v1, v2 quad_perm:[3,2,1,0]
 320 // GFX11: encoding: [0x7e,0x00,0xfe,0xd4,0xfa,0x04,0x02,0x00,0x01,0x1b,0x00,0xff]
 321
 322 v_cmpx_class_f32 v1, v2 quad_perm:[3,2,1,0]
 323 // GFX11: encoding: [0xfa,0x04,0xfc,0x7d,0x01,0x1b,0x00,0xff]
 324
 325 //===----------------------------------------------------------------------===//
 326 // VOP3P.
 327 //===----------------------------------------------------------------------===//
 328
 329 v_dot2_f32_f16 v0, v1, v2, v3
 330 // GFX11: encoding: [0x00,0x40,0x13,0xcc,0x01,0x05,0x0e,0x1c]
 331
 332 v_dot2_f32_f16_e64 v0, v1, v2, v3
 333 // GFX11: encoding: [0x00,0x40,0x13,0xcc,0x01,0x05,0x0e,0x1c]
 334
 335 //===----------------------------------------------------------------------===//
 336 // VOP3P.DPP8.
 337 //===----------------------------------------------------------------------===//
 338
 339 v_dot2_f32_f16 v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0]
 340 // GFX11: encoding: [0x00,0x00,0x13,0xcc,0xe9,0x04,0x0e,0x04,0x01,0x77,0x39,0x05]
 341
 342 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0]
 343 // GFX11: encoding: [0x00,0x00,0x13,0xcc,0xe9,0x04,0x0e,0x04,0x01,0x77,0x39,0x05]
 344
 345 //===----------------------------------------------------------------------===//
 346 // VOP3P.DPP16.
 347 //===----------------------------------------------------------------------===//
 348
 349 v_dot2_f32_f16 v0, v1, v2, v3 quad_perm:[1,2,3,0]
 350 // GFX11: encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x39,0x00,0xff]
 351
 352 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0]
 353 // GFX11: encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x39,0x00,0xff]