Codebase Browser
llvm
Go to App

llvm/llvm/test/MC/AMDGPU/gfx11_asm_vop3_dpp16_from_vop1-fake16.s 

// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-real-true16,+wavefrontsize32 -show-encoding %s | FileCheck --check-prefixes=GFX11 %s

v_ceil_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]

v_ceil_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]

v_ceil_f16_e64_dpp v5, v1 row_mirror
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]

v_ceil_f16_e64_dpp v5, v1 row_half_mirror
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]

v_ceil_f16_e64_dpp v5, v1 row_shl:1
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]

v_ceil_f16_e64_dpp v5, v1 row_shl:15
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]

v_ceil_f16_e64_dpp v5, v1 row_shr:1
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]

v_ceil_f16_e64_dpp v5, v1 row_shr:15
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]

v_ceil_f16_e64_dpp v5, v1 row_ror:1
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]

v_ceil_f16_e64_dpp v5, v1 row_ror:15
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]

v_ceil_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]

v_ceil_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]

v_ceil_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
// GFX11: [0x05,0x00,0xdc,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]

v_ceil_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX11: [0xff,0x81,0xdc,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]

v_floor_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]

v_floor_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]

v_floor_f16_e64_dpp v5, v1 row_mirror
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]

v_floor_f16_e64_dpp v5, v1 row_half_mirror
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]

v_floor_f16_e64_dpp v5, v1 row_shl:1
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]

v_floor_f16_e64_dpp v5, v1 row_shl:15
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]

v_floor_f16_e64_dpp v5, v1 row_shr:1
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]

v_floor_f16_e64_dpp v5, v1 row_shr:15
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]

v_floor_f16_e64_dpp v5, v1 row_ror:1
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]

v_floor_f16_e64_dpp v5, v1 row_ror:15
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]

v_floor_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]

v_floor_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]

v_floor_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
// GFX11: [0x05,0x00,0xdb,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]

v_floor_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX11: [0xff,0x81,0xdb,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]

v_rcp_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]

v_rcp_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]

v_rcp_f16_e64_dpp v5, v1 row_mirror
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]

v_rcp_f16_e64_dpp v5, v1 row_half_mirror
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]

v_rcp_f16_e64_dpp v5, v1 row_shl:1
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]

v_rcp_f16_e64_dpp v5, v1 row_shl:15
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]

v_rcp_f16_e64_dpp v5, v1 row_shr:1
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]

v_rcp_f16_e64_dpp v5, v1 row_shr:15
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]

v_rcp_f16_e64_dpp v5, v1 row_ror:1
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]

v_rcp_f16_e64_dpp v5, v1 row_ror:15
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]

v_rcp_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]

v_rcp_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]

v_rcp_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
// GFX11: [0x05,0x00,0xd4,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]

v_rcp_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX11: [0xff,0x81,0xd4,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]

v_sqrt_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]

v_sqrt_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]

v_sqrt_f16_e64_dpp v5, v1 row_mirror
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]

v_sqrt_f16_e64_dpp v5, v1 row_half_mirror
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]

v_sqrt_f16_e64_dpp v5, v1 row_shl:1
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]

v_sqrt_f16_e64_dpp v5, v1 row_shl:15
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]

v_sqrt_f16_e64_dpp v5, v1 row_shr:1
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]

v_sqrt_f16_e64_dpp v5, v1 row_shr:15
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]

v_sqrt_f16_e64_dpp v5, v1 row_ror:1
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]

v_sqrt_f16_e64_dpp v5, v1 row_ror:15
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]

v_sqrt_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]

v_sqrt_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]

v_sqrt_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
// GFX11: [0x05,0x00,0xd5,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]

v_sqrt_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX11: [0xff,0x81,0xd5,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]

v_rsq_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]

v_rsq_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]

v_rsq_f16_e64_dpp v5, v1 row_mirror
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]

v_rsq_f16_e64_dpp v5, v1 row_half_mirror
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]

v_rsq_f16_e64_dpp v5, v1 row_shl:1
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]

v_rsq_f16_e64_dpp v5, v1 row_shl:15
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]

v_rsq_f16_e64_dpp v5, v1 row_shr:1
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]

v_rsq_f16_e64_dpp v5, v1 row_shr:15
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]

v_rsq_f16_e64_dpp v5, v1 row_ror:1
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]

v_rsq_f16_e64_dpp v5, v1 row_ror:15
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]

v_rsq_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]

v_rsq_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]

v_rsq_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
// GFX11: [0x05,0x00,0xd6,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]

v_rsq_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX11: [0xff,0x81,0xd6,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]

v_log_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]

v_log_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]

v_log_f16_e64_dpp v5, v1 row_mirror
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]

v_log_f16_e64_dpp v5, v1 row_half_mirror
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]

v_log_f16_e64_dpp v5, v1 row_shl:1
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]

v_log_f16_e64_dpp v5, v1 row_shl:15
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]

v_log_f16_e64_dpp v5, v1 row_shr:1
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]

v_log_f16_e64_dpp v5, v1 row_shr:15
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]

v_log_f16_e64_dpp v5, v1 row_ror:1
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]

v_log_f16_e64_dpp v5, v1 row_ror:15
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]

v_log_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]

v_log_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]

v_log_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
// GFX11: [0x05,0x00,0xd7,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]

v_log_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX11: [0xff,0x81,0xd7,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]

v_exp_f16_e64_dpp v5, v1 quad_perm:[3,2,1,0]
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1b,0x00,0xff]

v_exp_f16_e64_dpp v5, v1 quad_perm:[0,1,2,3]
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0xe4,0x00,0xff]

v_exp_f16_e64_dpp v5, v1 row_mirror
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x40,0x01,0xff]

v_exp_f16_e64_dpp v5, v1 row_half_mirror
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x41,0x01,0xff]

v_exp_f16_e64_dpp v5, v1 row_shl:1
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x01,0x01,0xff]

v_exp_f16_e64_dpp v5, v1 row_shl:15
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x0f,0x01,0xff]

v_exp_f16_e64_dpp v5, v1 row_shr:1
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x11,0x01,0xff]

v_exp_f16_e64_dpp v5, v1 row_shr:15
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x1f,0x01,0xff]

v_exp_f16_e64_dpp v5, v1 row_ror:1
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x21,0x01,0xff]

v_exp_f16_e64_dpp v5, v1 row_ror:15
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x2f,0x01,0xff]

v_exp_f16_e64_dpp v5, v1 row_share:0 row_mask:0xf bank_mask:0xf
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x00,0x01,0x50,0x01,0xff]

v_exp_f16_e64_dpp v5, v1 mul:2 row_share:15 row_mask:0x0 bank_mask:0x1
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x08,0x01,0x5f,0x01,0x01]

v_exp_f16_e64_dpp v5, v1 mul:4 row_xmask:0 row_mask:0x1 bank_mask:0x3 bound_ctrl:1 fi:0
// GFX11: [0x05,0x00,0xd8,0xd5,0xfa,0x00,0x00,0x10,0x01,0x60,0x09,0x13]

v_exp_f16_e64_dpp v255, -|v255| clamp div:2 row_xmask:15 row_mask:0x3 bank_mask:0x0 bound_ctrl:0 fi:1
// GFX11: [0xff,0x81,0xd8,0xd5,0xfa,0x00,0x00,0x38,0xff,0x6f,0x05,0x30]