// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize32 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize64 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
//===----------------------------------------------------------------------===//
// ENC_FLAT.
//===----------------------------------------------------------------------===//
flat_load_ubyte v5, v[1:2]
// GFX10: encoding: [0x00,0x00,0x20,0xdc,0x01,0x00,0x7d,0x05]
flat_load_sbyte v5, v[1:2]
// GFX10: encoding: [0x00,0x00,0x24,0xdc,0x01,0x00,0x7d,0x05]
flat_load_ushort v5, v[1:2]
// GFX10: encoding: [0x00,0x00,0x28,0xdc,0x01,0x00,0x7d,0x05]
flat_load_sshort v5, v[1:2]
// GFX10: encoding: [0x00,0x00,0x2c,0xdc,0x01,0x00,0x7d,0x05]
flat_load_dword v5, v[1:2]
// GFX10: encoding: [0x00,0x00,0x30,0xdc,0x01,0x00,0x7d,0x05]
flat_load_dwordx2 v[5:6], v[1:2]
// GFX10: encoding: [0x00,0x00,0x34,0xdc,0x01,0x00,0x7d,0x05]
flat_load_dwordx3 v[5:7], v[1:2]
// GFX10: encoding: [0x00,0x00,0x3c,0xdc,0x01,0x00,0x7d,0x05]
flat_load_dwordx4 v[5:8], v[1:2]
// GFX10: encoding: [0x00,0x00,0x38,0xdc,0x01,0x00,0x7d,0x05]
flat_store_byte v[1:2], v2
// GFX10: encoding: [0x00,0x00,0x60,0xdc,0x01,0x02,0x7d,0x00]
flat_store_short v[1:2], v2
// GFX10: encoding: [0x00,0x00,0x68,0xdc,0x01,0x02,0x7d,0x00]
flat_store_dword v[1:2], v2
// GFX10: encoding: [0x00,0x00,0x70,0xdc,0x01,0x02,0x7d,0x00]
flat_store_dwordx2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x74,0xdc,0x01,0x02,0x7d,0x00]
flat_store_dwordx3 v[1:2], v[2:4]
// GFX10: encoding: [0x00,0x00,0x7c,0xdc,0x01,0x02,0x7d,0x00]
flat_store_dwordx4 v[1:2], v[2:5]
// GFX10: encoding: [0x00,0x00,0x78,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_swap v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xc0,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_cmpswap v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0xc4,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_add v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xc8,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_sub v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xcc,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_smin v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xd4,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_umin v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xd8,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_smax v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xdc,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_umax v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xe0,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_and v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xe4,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_or v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xe8,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_xor v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xec,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_inc v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xf0,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_dec v[1:2], v2
// GFX10: encoding: [0x00,0x00,0xf4,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_swap_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x40,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_cmpswap_x2 v[1:2], v[2:5]
// GFX10: encoding: [0x00,0x00,0x44,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_add_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x48,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_sub_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x4c,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_smin_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x54,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_umin_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x58,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_smax_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x5c,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_umax_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x60,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_and_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x64,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_or_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x68,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_xor_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x6c,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_inc_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x70,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_dec_x2 v[1:2], v[2:3]
// GFX10: encoding: [0x00,0x00,0x74,0xdd,0x01,0x02,0x7d,0x00]
//===----------------------------------------------------------------------===//
// ENC_FLAT: dlc support for atomics.
//===----------------------------------------------------------------------===//
flat_atomic_add v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xc8,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_add_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x48,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_and v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xe4,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_and_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x64,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_cmpswap v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0xc4,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_cmpswap_x2 v[1:2], v[2:5] dlc
// GFX10: [0x00,0x10,0x44,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_dec v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xf4,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_dec_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x74,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_fcmpswap v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0xf8,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_fcmpswap_x2 v[1:2], v[2:5] dlc
// GFX10: [0x00,0x10,0x78,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_fmax v[1:2], v2 dlc
// GFX10: [0x00,0x10,0x00,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_fmax_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x80,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_fmin v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xfc,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_fmin_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x7c,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_inc v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xf0,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_inc_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x70,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_or v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xe8,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_or_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x68,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_smax v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xdc,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_smax_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x5c,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_smin v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xd4,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_smin_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x54,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_sub v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xcc,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_sub_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x4c,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_swap v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xc0,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_swap_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x40,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_umax v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xe0,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_umax_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x60,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_umin v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xd8,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_umin_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x58,0xdd,0x01,0x02,0x7d,0x00]
flat_atomic_xor v[1:2], v2 dlc
// GFX10: [0x00,0x10,0xec,0xdc,0x01,0x02,0x7d,0x00]
flat_atomic_xor_x2 v[1:2], v[2:3] dlc
// GFX10: [0x00,0x10,0x6c,0xdd,0x01,0x02,0x7d,0x00]
//===----------------------------------------------------------------------===//
// ENC_FLAT_GLOBAL: fcmpswap.
//===----------------------------------------------------------------------===//
global_atomic_fcmpswap v[1:2], v[2:3], off offset:-1
// GFX10: [0xff,0x8f,0xf8,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_fcmpswap v5, v[1:2], v[2:3], off offset:-1 glc
// GFX10: [0xff,0x8f,0xf9,0xdc,0x01,0x02,0x7d,0x05]
global_atomic_fcmpswap v[1:2], v[2:3], off offset:2047
// GFX10: [0xff,0x87,0xf8,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_fcmpswap v[1:2], v[2:3], off offset:-2048
// GFX10: [0x00,0x88,0xf8,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_fcmpswap v[1:2], v[2:3], off offset:-1 slc
// GFX10: [0xff,0x8f,0xfa,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_fcmpswap v[1:2], v[2:3], off offset:-1 dlc
// GFX10: [0xff,0x9f,0xf8,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_fcmpswap_x2 v[1:2], v[2:5], off offset:-1
// GFX10: [0xff,0x8f,0x78,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_fcmpswap_x2 v[5:6], v[1:2], v[2:5], off offset:-1 glc
// GFX10: [0xff,0x8f,0x79,0xdd,0x01,0x02,0x7d,0x05]
global_atomic_fcmpswap_x2 v[1:2], v[2:5], off offset:2047
// GFX10: [0xff,0x87,0x78,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_fcmpswap_x2 v[1:2], v[2:5], off offset:-2048
// GFX10: [0x00,0x88,0x78,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_fcmpswap_x2 v[1:2], v[2:5], off offset:-1 slc
// GFX10: [0xff,0x8f,0x7a,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_fcmpswap_x2 v[1:2], v[2:5], off offset:-1 dlc
// GFX10: [0xff,0x9f,0x78,0xdd,0x01,0x02,0x7d,0x00]
//===----------------------------------------------------------------------===//
// ENC_FLAT_GLOBAL: dlc support for atomics.
//===----------------------------------------------------------------------===//
global_atomic_add v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xc8,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_add_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x48,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_and v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xe4,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_and_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x64,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_cmpswap v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0xc4,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_cmpswap_x2 v[1:2], v[2:5], off dlc
// GFX10: [0x00,0x90,0x44,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_dec v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xf4,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_dec_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x74,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_fmax v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0x00,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_fmax_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x80,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_fmin v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xfc,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_fmin_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x7c,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_inc v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xf0,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_inc_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x70,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_or v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xe8,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_or_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x68,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_smax v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xdc,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_smax_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x5c,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_smin v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xd4,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_smin_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x54,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_sub v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xcc,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_sub_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x4c,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_swap v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xc0,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_swap_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x40,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_umax v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xe0,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_umax_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x60,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_umin v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xd8,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_umin_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x58,0xdd,0x01,0x02,0x7d,0x00]
global_atomic_xor v[1:2], v2, off dlc
// GFX10: [0x00,0x90,0xec,0xdc,0x01,0x02,0x7d,0x00]
global_atomic_xor_x2 v[1:2], v[2:3], off dlc
// GFX10: [0x00,0x90,0x6c,0xdd,0x01,0x02,0x7d,0x00]
global_load_dword v2, s[4:5] offset:1024 lds
// GFX10: [0x00,0xa4,0x30,0xdc,0x02,0x00,0x04,0x00]
global_load_dword v2, s[4:5] offset:1024 lds
// GFX10: [0x00,0xa4,0x30,0xdc,0x02,0x00,0x04,0x00]
global_load_ubyte v2, s[4:5] offset:1024 lds
// GFX10: [0x00,0xa4,0x20,0xdc,0x02,0x00,0x04,0x00]
global_load_sbyte v2, s[4:5] offset:1024 lds
// GFX10: [0x00,0xa4,0x24,0xdc,0x02,0x00,0x04,0x00]
global_load_ushort v2, s[4:5] offset:1024 lds
// GFX10: [0x00,0xa4,0x28,0xdc,0x02,0x00,0x04,0x00]
global_load_sshort v2, s[4:5] offset:1024 lds
// GFX10: [0x00,0xa4,0x2c,0xdc,0x02,0x00,0x04,0x00]
global_load_dword v[2:3], off offset:1024 lds
// GFX10: [0x00,0xa4,0x30,0xdc,0x02,0x00,0x7d,0x00]
global_load_ubyte v[2:3], off offset:1024 lds
// GFX10: [0x00,0xa4,0x20,0xdc,0x02,0x00,0x7d,0x00]
global_load_sbyte v[2:3], off offset:1024 lds
// GFX10: [0x00,0xa4,0x24,0xdc,0x02,0x00,0x7d,0x00]
global_load_ushort v[2:3], off offset:1024 lds
// GFX10: [0x00,0xa4,0x28,0xdc,0x02,0x00,0x7d,0x00]
global_load_sshort v[2:3], off offset:1024 lds
// GFX10: [0x00,0xa4,0x2c,0xdc,0x02,0x00,0x7d,0x00]
scratch_load_dword v2, off offset:1024 lds
// GFX10: [0x00,0x64,0x30,0xdc,0x02,0x00,0x7d,0x00]
scratch_load_ubyte v2, off offset:1024 lds
// GFX10: [0x00,0x64,0x20,0xdc,0x02,0x00,0x7d,0x00]
scratch_load_sbyte v2, off offset:1024 lds
// GFX10: [0x00,0x64,0x24,0xdc,0x02,0x00,0x7d,0x00]
scratch_load_ushort v2, off offset:1024 lds
// GFX10: [0x00,0x64,0x28,0xdc,0x02,0x00,0x7d,0x00]
scratch_load_sshort v2, off offset:1024 lds
// GFX10: [0x00,0x64,0x2c,0xdc,0x02,0x00,0x7d,0x00]
scratch_load_dword off, s4 offset:1024 lds
// GFX10: [0x00,0x64,0x30,0xdc,0x00,0x00,0x04,0x00]
scratch_load_ubyte off, s4 offset:1024 lds
// GFX10: [0x00,0x64,0x20,0xdc,0x00,0x00,0x04,0x00]
scratch_load_sbyte off, s4 offset:1024 lds
// GFX10: [0x00,0x64,0x24,0xdc,0x00,0x00,0x04,0x00]
scratch_load_ushort off, s4 offset:1024 lds
// GFX10: [0x00,0x64,0x28,0xdc,0x00,0x00,0x04,0x00]
scratch_load_sshort off, s4 offset:1024 lds
// GFX10: [0x00,0x64,0x2c,0xdc,0x00,0x00,0x04,0x00]
//===----------------------------------------------------------------------===//
// Also see flat-gfx10.s, flat-global.s, flat-scratch-instructions.s.
//===----------------------------------------------------------------------===//
Codebase Browser
llvm