llvm/test/MC/AMDGPU/gfx1030_new.s

   1 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1030 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
   2 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1031 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
   3 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1032 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
   4 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1033 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
   5 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1034 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
   6 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1035 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
   7 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1036 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
   8
   9 global_load_dword_addtid v1, s[2:3] offset:16
  10 // GFX10: encoding: [0x10,0x80,0x58,0xdc,0x00,0x00,0x02,0x01]
  11
  12 global_load_dword_addtid v1, s[2:3] offset:16 glc slc dlc
  13 // GFX10: encoding: [0x10,0x90,0x5b,0xdc,0x00,0x00,0x02,0x01]
  14
  15 global_store_dword_addtid v1, s[2:3] offset:16 glc slc dlc
  16 // GFX10: encoding: [0x10,0x90,0x5f,0xdc,0x00,0x01,0x02,0x00]
  17
  18 global_load_dword_addtid v1, off offset:16
  19 // GFX10: encoding: [0x10,0x80,0x58,0xdc,0x00,0x00,0x7d,0x01]
  20
  21 global_load_dword_addtid v1, off offset:16 glc slc dlc
  22 // GFX10: encoding: [0x10,0x90,0x5b,0xdc,0x00,0x00,0x7d,0x01]
  23
  24 global_store_dword_addtid v1, off offset:16 glc slc dlc
  25 // GFX10: encoding: [0x10,0x90,0x5f,0xdc,0x00,0x01,0x7d,0x00]
  26
  27 global_store_dword v254, v1, s[2:3] offset:16
  28 // GFX10: encoding: [0x10,0x80,0x70,0xdc,0xfe,0x01,0x02,0x00]
  29
  30 global_atomic_csub v2, v[0:1], v2, off offset:100 glc slc
  31 // GFX10: encoding: [0x64,0x80,0xd3,0xdc,0x00,0x02,0x7d,0x02]
  32
  33 global_atomic_csub v[0:1], v2, off offset:100 slc
  34 // GFX10: encoding: [0x64,0x80,0xd2,0xdc,0x00,0x02,0x7d,0x00]
  35
  36 global_atomic_csub v2, v[0:1], v2, off glc
  37 // GFX10: encoding: [0x00,0x80,0xd1,0xdc,0x00,0x02,0x7d,0x02]
  38
  39 global_atomic_csub v[0:1], v2, off
  40 // GFX10: encoding: [0x00,0x80,0xd0,0xdc,0x00,0x02,0x7d,0x00]
  41
  42 global_atomic_csub v2, v0, v2, s[2:3] glc
  43 // GFX10: encoding: [0x00,0x80,0xd1,0xdc,0x00,0x02,0x02,0x02]
  44
  45 global_atomic_csub v0, v2, s[2:3]
  46 // GFX10: encoding: [0x00,0x80,0xd0,0xdc,0x00,0x02,0x02,0x00]
  47
  48 global_atomic_csub v2, v0, v2, s[2:3] offset:100 glc slc
  49 // GFX10: encoding: [0x64,0x80,0xd3,0xdc,0x00,0x02,0x02,0x02]
  50
  51 buffer_atomic_csub v5, off, s[8:11], s3 glc
  52 // GFX10: encoding: [0x00,0x40,0xd0,0xe0,0x00,0x05,0x02,0x03]
  53
  54 buffer_atomic_csub v5, off, s[8:11], s3
  55 // GFX10: encoding: [0x00,0x00,0xd0,0xe0,0x00,0x05,0x02,0x03]
  56
  57 buffer_atomic_csub v5, off, s[8:11], s3 offset:4095 glc
  58 // GFX10: encoding: [0xff,0x4f,0xd0,0xe0,0x00,0x05,0x02,0x03]
  59
  60 buffer_atomic_csub v5, off, s[8:11], -1 offset:4095 glc
  61 // GFX10: encoding: [0xff,0x4f,0xd0,0xe0,0x00,0x05,0x02,0xc1]
  62
  63 buffer_atomic_csub v5, v0, s[8:11], s3 offen offset:4095 glc
  64 // GFX10: encoding: [0xff,0x5f,0xd0,0xe0,0x00,0x05,0x02,0x03]
  65
  66 buffer_atomic_csub v5, v0, s[8:11], s3 offen offset:4095
  67 // GFX10: encoding: [0xff,0x1f,0xd0,0xe0,0x00,0x05,0x02,0x03]
  68
  69 buffer_atomic_csub v5, v0, s[8:11], s3 idxen offset:4095 glc
  70 // GFX10: encoding: [0xff,0x6f,0xd0,0xe0,0x00,0x05,0x02,0x03]
  71
  72 buffer_atomic_csub v5, v0, s[8:11], s3 idxen offset:4095
  73 // GFX10: encoding: [0xff,0x2f,0xd0,0xe0,0x00,0x05,0x02,0x03]
  74
  75 buffer_atomic_csub v5, off, s[8:11], s3 glc slc
  76 // GFX10: encoding: [0x00,0x40,0xd0,0xe0,0x00,0x05,0x42,0x03]
  77
  78 buffer_atomic_csub v5, off, s[8:11], s3 slc
  79 // GFX10: encoding: [0x00,0x00,0xd0,0xe0,0x00,0x05,0x42,0x03]
  80
  81 s_getreg_b32 s2, hwreg(HW_REG_SHADER_CYCLES)
  82 // GFX10: encoding: [0x1d,0xf8,0x02,0xb9]
  83
  84 s_getreg_b32 s2, 29
  85 // GFX10: s_getreg_b32 s2, hwreg(HW_REG_SHADER_CYCLES, 0, 1) ; encoding: [0x1d,0x00,0x02,0xb9]
  86
  87 s_getreg_b32 s2, hwreg(22)
  88 // GFX10: s_getreg_b32 s2, hwreg(22) ; encoding: [0x16,0xf8,0x02,0xb9]
  89
  90 v_fma_legacy_f32 v0, v1, v2, v3
  91 // GFX10: encoding: [0x00,0x00,0x40,0xd5,0x01,0x05,0x0e,0x04]
  92
  93 v_fma_legacy_f32 v0, v1, |v2|, -v3
  94 // GFX10: encoding: [0x00,0x02,0x40,0xd5,0x01,0x05,0x0e,0x84]
  95
  96 v_fma_legacy_f32 v0, s1, 2.0, -v3
  97 // GFX10: encoding: [0x00,0x00,0x40,0xd5,0x01,0xe8,0x0d,0x84]
  98
  99 v_fmac_legacy_f32 v0, v1, v2
 100 // GFX10: encoding: [0x01,0x05,0x00,0x0c]
 101
 102 v_fmac_legacy_f32 v0, |v1|, -v2
 103 // GFX10: encoding: [0x00,0x01,0x06,0xd5,0x01,0x05,0x02,0x40]
 104
 105 v_fmac_legacy_f32 v0, s1, 2.0
 106 // GFX10: encoding: [0x00,0x00,0x06,0xd5,0x01,0xe8,0x01,0x00]
 107
 108 image_bvh_intersect_ray v[4:7], v[9:19], s[4:7]
 109 // GFX10: encoding: [0x01,0x9f,0x98,0xf1,0x09,0x04,0x01,0x00]
 110
 111 image_bvh_intersect_ray v[4:7], v[9:19], s[4:7] noa16
 112 // GFX10: encoding: [0x01,0x9f,0x98,0xf1,0x09,0x04,0x01,0x00]
 113
 114 image_bvh_intersect_ray v[4:7], v[9:16], s[4:7] a16
 115 // GFX10: encoding: [0x01,0x9f,0x98,0xf1,0x09,0x04,0x01,0x40]
 116
 117 image_bvh64_intersect_ray v[4:7], v[9:20], s[4:7]
 118 // GFX10: encoding: [0x01,0x9f,0x9c,0xf1,0x09,0x04,0x01,0x00]
 119
 120 image_bvh64_intersect_ray v[4:7], v[9:17], s[4:7] a16
 121 // GFX10: encoding: [0x01,0x9f,0x9c,0xf1,0x09,0x04,0x01,0x40]
 122
 123 image_bvh_intersect_ray v[39:42], [v50, v46, v23, v17, v16, v15, v21, v20, v19, v37, v40], s[12:15]
 124 // GFX10: encoding: [0x07,0x9f,0x98,0xf1,0x32,0x27,0x03,0x00,0x2e,0x17,0x11,0x10,0x0f,0x15,0x14,0x13,0x25,0x28,0x00,0x00]
 125
 126 image_bvh_intersect_ray v[39:42], [v50, v46, v23, v17, v16, v15, v21, v20, v19, v37, v40], s[12:15] noa16
 127 // GFX10: encoding: [0x07,0x9f,0x98,0xf1,0x32,0x27,0x03,0x00,0x2e,0x17,0x11,0x10,0x0f,0x15,0x14,0x13,0x25,0x28,0x00,0x00]
 128
 129 image_bvh_intersect_ray v[39:42], [v50, v46, v23, v17, v16, v15, v21, v20], s[12:15] a16
 130 // GFX10: encoding: [0x05,0x9f,0x98,0xf1,0x32,0x27,0x03,0x40,0x2e,0x17,0x11,0x10,0x0f,0x15,0x14,0x00]
 131
 132 image_bvh64_intersect_ray v[39:42], [v50, v46, v23, v17, v16, v15, v21, v20, v19, v37, v40, v42], s[12:15]
 133 // GFX10: encoding: [0x07,0x9f,0x9c,0xf1,0x32,0x27,0x03,0x00,0x2e,0x17,0x11,0x10,0x0f,0x15,0x14,0x13,0x25,0x28,0x2a,0x00]
 134
 135 image_bvh64_intersect_ray v[39:42], [v50, v46, v23, v17, v16, v15, v21, v20, v19], s[12:15] a16
 136 // GFX10: encoding: [0x05,0x9f,0x9c,0xf1,0x32,0x27,0x03,0x40,0x2e,0x17,0x11,0x10,0x0f,0x15,0x14,0x13]
 137
 138 image_msaa_load v[1:4], v[5:7], s[8:15] dmask:0xf dim:SQ_RSRC_IMG_2D_MSAA
 139 // GFX10: encoding: [0x31,0x0f,0x00,0xf0,0x05,0x01,0x02,0x00]
 140
 141 image_msaa_load v[1:4], v[5:7], s[8:15] dmask:0xf dim:SQ_RSRC_IMG_2D_MSAA glc
 142 // GFX10: encoding: [0x31,0x2f,0x00,0xf0,0x05,0x01,0x02,0x00]
 143
 144 image_msaa_load v5, v[1:3], s[8:15] dmask:0x1 dim:SQ_RSRC_IMG_2D_MSAA d16
 145 // GFX10: encoding: [0x31,0x01,0x00,0xf0,0x01,0x05,0x02,0x80]
 146
 147 image_msaa_load v[1:4], v[5:8], s[8:15] dmask:0xf dim:SQ_RSRC_IMG_2D_MSAA_ARRAY
 148 // GFX10: encoding: [0x39,0x0f,0x00,0xf0,0x05,0x01,0x02,0x00]
 149
 150 image_msaa_load v14, [v204,v11,v14,v19], s[40:47] dmask:0x1 dim:SQ_RSRC_IMG_2D_MSAA_ARRAY
 151 // GFX10: encoding: [0x3b,0x01,0x00,0xf0,0xcc,0x0e,0x0a,0x00,0x0b,0x0e,0x13,0x00]
 152
 153 //===----------------------------------------------------------------------===//
 154 // s_waitcnt_depctr.
 155 //===----------------------------------------------------------------------===//
 156
 157 s_waitcnt_depctr -32768
 158 // GFX10: encoding: [0x00,0x80,0xa3,0xbf]
 159
 160 s_waitcnt_depctr 65535
 161 // GFX10: encoding: [0xff,0xff,0xa3,0xbf]
 162
 163 s_waitcnt_depctr depctr_hold_cnt(0)
 164 // GFX10: encoding: [0x1f,0xff,0xa3,0xbf]
 165
 166 s_waitcnt_depctr depctr_hold_cnt(1)
 167 // GFX10: encoding: [0x9f,0xff,0xa3,0xbf]
 168
 169 s_waitcnt_depctr depctr_sa_sdst(0)
 170 // GFX10: encoding: [0x9e,0xff,0xa3,0xbf]
 171
 172 s_waitcnt_depctr depctr_sa_sdst(1)
 173 // GFX10: encoding: [0x9f,0xff,0xa3,0xbf]
 174
 175 s_waitcnt_depctr depctr_va_vdst(0)
 176 // GFX10: encoding: [0x9f,0x0f,0xa3,0xbf]
 177
 178 s_waitcnt_depctr depctr_va_vdst(1)
 179 // GFX10: encoding: [0x9f,0x1f,0xa3,0xbf]
 180
 181 s_waitcnt_depctr depctr_va_vdst(14)
 182 // GFX10: encoding: [0x9f,0xef,0xa3,0xbf]
 183
 184 s_waitcnt_depctr depctr_va_vdst(15)
 185 // GFX10: encoding: [0x9f,0xff,0xa3,0xbf]
 186
 187 s_waitcnt_depctr depctr_va_sdst(0)
 188 // GFX10: encoding: [0x9f,0xf1,0xa3,0xbf]
 189
 190 s_waitcnt_depctr depctr_va_sdst(1)
 191 // GFX10: encoding: [0x9f,0xf3,0xa3,0xbf]
 192
 193 s_waitcnt_depctr depctr_va_sdst(6)
 194 // GFX10: encoding: [0x9f,0xfd,0xa3,0xbf]
 195
 196 s_waitcnt_depctr depctr_va_sdst(7)
 197 // GFX10: encoding: [0x9f,0xff,0xa3,0xbf]
 198
 199 s_waitcnt_depctr depctr_va_ssrc(0)
 200 // GFX10: encoding: [0x9f,0xfe,0xa3,0xbf]
 201
 202 s_waitcnt_depctr depctr_va_ssrc(1)
 203 // GFX10: encoding: [0x9f,0xff,0xa3,0xbf]
 204
 205 s_waitcnt_depctr depctr_va_vcc(0)
 206 // GFX10: encoding: [0x9d,0xff,0xa3,0xbf]
 207
 208 s_waitcnt_depctr depctr_va_vcc(1)
 209 // GFX10: encoding: [0x9f,0xff,0xa3,0xbf]
 210
 211 s_waitcnt_depctr depctr_vm_vsrc(0)
 212 // GFX10: encoding: [0x83,0xff,0xa3,0xbf]
 213
 214 s_waitcnt_depctr depctr_vm_vsrc(1)
 215 // GFX10: encoding: [0x87,0xff,0xa3,0xbf]
 216
 217 s_waitcnt_depctr depctr_vm_vsrc(6)
 218 // GFX10: encoding: [0x9b,0xff,0xa3,0xbf]
 219
 220 s_waitcnt_depctr depctr_vm_vsrc(7)
 221 // GFX10: encoding: [0x9f,0xff,0xa3,0xbf]
 222
 223 s_waitcnt_depctr depctr_hold_cnt(0) depctr_sa_sdst(0) depctr_va_vdst(0) depctr_va_sdst(0) depctr_va_ssrc(0) depctr_va_vcc(0) depctr_vm_vsrc(0)
 224 // GFX10: encoding: [0x00,0x00,0xa3,0xbf]
 225
 226 s_waitcnt_depctr depctr_hold_cnt(1) depctr_sa_sdst(1) depctr_va_vdst(15) depctr_va_sdst(7) depctr_va_ssrc(1) depctr_va_vcc(1) depctr_vm_vsrc(7)
 227 // GFX10: encoding: [0x9f,0xff,0xa3,0xbf]
 228
 229 s_waitcnt_depctr depctr_hold_cnt(1) & depctr_sa_sdst(1) & depctr_va_vdst(1) & depctr_va_sdst(1) & depctr_va_ssrc(1) & depctr_va_vcc(1) & depctr_vm_vsrc(1)
 230 // GFX10: encoding: [0x87,0x13,0xa3,0xbf]
 231
 232 s_waitcnt_depctr depctr_hold_cnt(1), depctr_sa_sdst(1), depctr_va_vdst(14), depctr_va_sdst(6), depctr_va_ssrc(1), depctr_va_vcc(1), depctr_vm_vsrc(6)
 233 // GFX10: encoding: [0x9b,0xed,0xa3,0xbf]
 234
 235 scratch_load_dword off, off offset:1024 lds
 236 // GFX10: [0x00,0x64,0x30,0xdc,0x00,0x00,0x7f,0x00]
 237
 238 scratch_load_ubyte off, off offset:1024 lds
 239 // GFX10: [0x00,0x64,0x20,0xdc,0x00,0x00,0x7f,0x00]
 240
 241 scratch_load_sbyte off, off offset:1024 lds
 242 // GFX10: [0x00,0x64,0x24,0xdc,0x00,0x00,0x7f,0x00]
 243
 244 scratch_load_ushort off, off offset:1024 lds
 245 // GFX10: [0x00,0x64,0x28,0xdc,0x00,0x00,0x7f,0x00]
 246
 247 scratch_load_sshort off, off offset:1024 lds
 248 // GFX10: [0x00,0x64,0x2c,0xdc,0x00,0x00,0x7f,0x00]