blob: 2f2fd81e0c21d198fcd835fda7e3d2637b090aec [file] [log] [blame]
# RUN: llvm-mc -arch=amdgcn -mcpu=gfx1011 -disassemble -show-encoding < %s | FileCheck %s
# RUN: llvm-mc -arch=amdgcn -mcpu=gfx1012 -disassemble -show-encoding < %s | FileCheck %s
# CHECK: v_dot2c_f32_f16_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x04]
0x01,0x05,0x0a,0x04
# CHECK: v_dot2c_f32_f16_e32 v255, v1, v2 ; encoding: [0x01,0x05,0xfe,0x05]
0x01,0x05,0xfe,0x05
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x00]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x00
# CHECK: v_dot2c_f32_f16_dpp v255, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0xfe,0x05,0x01,0xe4,0x00,0x00]
0xfa,0x04,0xfe,0x05,0x01,0xe4,0x00,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v255, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0xff,0xe4,0x00,0x00]
0xfa,0x04,0x0a,0x04,0xff,0xe4,0x00,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v255 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0xfe,0x0b,0x04,0x01,0xe4,0x00,0x00]
0xfa,0xfe,0x0b,0x04,0x01,0xe4,0x00,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0x1b,0x00,0x00]
0xfa,0x04,0x0a,0x04,0x01,0x1b,0x00,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 row_mirror row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0x40,0x01,0x00]
0xfa,0x04,0x0a,0x04,0x01,0x40,0x01,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 row_half_mirror row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0x41,0x01,0x00]
0xfa,0x04,0x0a,0x04,0x01,0x41,0x01,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 row_shl:1 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0x01,0x01,0x00]
0xfa,0x04,0x0a,0x04,0x01,0x01,0x01,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 row_shl:15 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0x0f,0x01,0x00]
0xfa,0x04,0x0a,0x04,0x01,0x0f,0x01,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 row_shr:1 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0x11,0x01,0x00]
0xfa,0x04,0x0a,0x04,0x01,0x11,0x01,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 row_shr:15 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0x1f,0x01,0x00]
0xfa,0x04,0x0a,0x04,0x01,0x1f,0x01,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 row_ror:1 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0x21,0x01,0x00]
0xfa,0x04,0x0a,0x04,0x01,0x21,0x01,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 row_ror:15 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0x2f,0x01,0x00]
0xfa,0x04,0x0a,0x04,0x01,0x2f,0x01,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x1 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x10]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x10
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x3 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x30]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x30
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0xf0]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0xf0
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0xf0]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0xf0
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x01]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x01
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x3 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x03]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x03
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x0f]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x0f
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x0f]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x00,0x0f
# CHECK: v_dot2c_f32_f16_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 bound_ctrl:1 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x08,0x00]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x08,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, -v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x10,0x00]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x10,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, |v1|, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x20,0x00]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x20,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, -v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x40,0x00]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x40,0x00
# CHECK: v_dot2c_f32_f16_dpp v5, v1, |v2| quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x04,0x01,0xe4,0x80,0x00]
0xfa,0x04,0x0a,0x04,0x01,0xe4,0x80,0x00
# CHECK: v_dot4c_i32_i8_e32 v5, v1, v2 ; encoding: [0x01,0x05,0x0a,0x1a]
0x01,0x05,0x0a,0x1a
# CHECK: v_dot4c_i32_i8_e32 v255, v1, v2 ; encoding: [0x01,0x05,0xfe,0x1b]
0x01,0x05,0xfe,0x1b
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x00
# CHECK: v_dot4c_i32_i8_dpp v255, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0xfe,0x1b,0x01,0xe4,0x00,0x00]
0xfa,0x04,0xfe,0x1b,0x01,0xe4,0x00,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v255, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0xff,0xe4,0x00,0x00]
0xfa,0x04,0x0a,0x1a,0xff,0xe4,0x00,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v255 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0xfe,0x0b,0x1a,0x01,0xe4,0x00,0x00]
0xfa,0xfe,0x0b,0x1a,0x01,0xe4,0x00,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[3,2,1,0] row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0x1b,0x00,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0x1b,0x00,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 row_mirror row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0x40,0x01,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0x40,0x01,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 row_half_mirror row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0x41,0x01,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0x41,0x01,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 row_shl:1 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0x01,0x01,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0x01,0x01,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 row_shl:15 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0x0f,0x01,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0x0f,0x01,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 row_shr:1 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0x11,0x01,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0x11,0x01,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 row_shr:15 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0x1f,0x01,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0x1f,0x01,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 row_ror:1 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0x21,0x01,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0x21,0x01,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 row_ror:15 row_mask:0x0 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0x2f,0x01,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0x2f,0x01,0x00
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x1 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x10]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x10
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x3 bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x30]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x30
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0xf0]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0xf0
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0xf bank_mask:0x0 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0xf0]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0xf0
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x1 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x01]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x01
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x3 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x03]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x03
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x0f]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x0f
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0xf ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x0f]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x00,0x0f
# CHECK: v_dot4c_i32_i8_dpp v5, v1, v2 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0 bound_ctrl:1 ; encoding: [0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x08,0x00]
0xfa,0x04,0x0a,0x1a,0x01,0xe4,0x08,0x00