1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 2; RUN: llc < %s -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs -show-mc-encoding | FileCheck -enable-var-scope -check-prefixes=UNPACKED %s 3; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx810 -verify-machineinstrs | FileCheck -enable-var-scope -check-prefixes=PACKED %s 4; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs | FileCheck -enable-var-scope -check-prefixes=PACKED %s 5 6define amdgpu_ps half @buffer_load_format_d16_x(ptr addrspace(8) inreg %rsrc) { 7; UNPACKED-LABEL: buffer_load_format_d16_x: 8; UNPACKED: ; %bb.0: ; %main_body 9; UNPACKED-NEXT: v_mov_b32_e32 v0, 0 ; encoding: [0x80,0x02,0x00,0x7e] 10; UNPACKED-NEXT: buffer_load_format_d16_x v0, v0, s[0:3], 0 idxen ; encoding: [0x00,0x20,0x20,0xe0,0x00,0x00,0x00,0x80] 11; UNPACKED-NEXT: s_waitcnt vmcnt(0) ; encoding: [0x70,0x0f,0x8c,0xbf] 12; UNPACKED-NEXT: ; return to shader part epilog 13; 14; PACKED-LABEL: buffer_load_format_d16_x: 15; PACKED: ; %bb.0: ; %main_body 16; PACKED-NEXT: v_mov_b32_e32 v0, 0 17; PACKED-NEXT: buffer_load_format_d16_x v0, v0, s[0:3], 0 idxen 18; PACKED-NEXT: s_waitcnt vmcnt(0) 19; PACKED-NEXT: ; return to shader part epilog 20main_body: 21 %data = call half @llvm.amdgcn.struct.ptr.buffer.load.format.f16(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0, i32 0) 22 ret half %data 23} 24 25define amdgpu_ps half @buffer_load_format_d16_xy(ptr addrspace(8) inreg %rsrc) { 26; UNPACKED-LABEL: buffer_load_format_d16_xy: 27; UNPACKED: ; %bb.0: ; %main_body 28; UNPACKED-NEXT: v_mov_b32_e32 v0, 0 ; encoding: [0x80,0x02,0x00,0x7e] 29; UNPACKED-NEXT: buffer_load_format_d16_xy v[0:1], v0, s[0:3], 0 idxen ; encoding: [0x00,0x20,0x24,0xe0,0x00,0x00,0x00,0x80] 30; UNPACKED-NEXT: s_waitcnt vmcnt(0) ; encoding: [0x70,0x0f,0x8c,0xbf] 31; UNPACKED-NEXT: v_mov_b32_e32 v0, v1 ; encoding: [0x01,0x03,0x00,0x7e] 32; UNPACKED-NEXT: ; return to shader part epilog 33; 34; PACKED-LABEL: buffer_load_format_d16_xy: 35; PACKED: ; %bb.0: ; %main_body 36; PACKED-NEXT: v_mov_b32_e32 v0, 0 37; PACKED-NEXT: buffer_load_format_d16_xy v0, v0, s[0:3], 0 idxen 38; PACKED-NEXT: s_waitcnt vmcnt(0) 39; PACKED-NEXT: v_lshrrev_b32_e32 v0, 16, v0 40; PACKED-NEXT: ; return to shader part epilog 41main_body: 42 %data = call <2 x half> @llvm.amdgcn.struct.ptr.buffer.load.format.v2f16(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0, i32 0) 43 %elt = extractelement <2 x half> %data, i32 1 44 ret half %elt 45} 46 47define amdgpu_ps half @buffer_load_format_d16_xyz(ptr addrspace(8) inreg %rsrc) { 48; UNPACKED-LABEL: buffer_load_format_d16_xyz: 49; UNPACKED: ; %bb.0: ; %main_body 50; UNPACKED-NEXT: v_mov_b32_e32 v0, 0 ; encoding: [0x80,0x02,0x00,0x7e] 51; UNPACKED-NEXT: buffer_load_format_d16_xyz v[0:2], v0, s[0:3], 0 idxen ; encoding: [0x00,0x20,0x28,0xe0,0x00,0x00,0x00,0x80] 52; UNPACKED-NEXT: s_waitcnt vmcnt(0) ; encoding: [0x70,0x0f,0x8c,0xbf] 53; UNPACKED-NEXT: v_mov_b32_e32 v0, v2 ; encoding: [0x02,0x03,0x00,0x7e] 54; UNPACKED-NEXT: ; return to shader part epilog 55; 56; PACKED-LABEL: buffer_load_format_d16_xyz: 57; PACKED: ; %bb.0: ; %main_body 58; PACKED-NEXT: v_mov_b32_e32 v0, 0 59; PACKED-NEXT: buffer_load_format_d16_xyz v[0:1], v0, s[0:3], 0 idxen 60; PACKED-NEXT: s_waitcnt vmcnt(0) 61; PACKED-NEXT: v_mov_b32_e32 v0, v1 62; PACKED-NEXT: ; return to shader part epilog 63main_body: 64 %data = call <3 x half> @llvm.amdgcn.struct.ptr.buffer.load.format.v3f16(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0, i32 0) 65 %elt = extractelement <3 x half> %data, i32 2 66 ret half %elt 67} 68 69define amdgpu_ps half @buffer_load_format_d16_xyzw(ptr addrspace(8) inreg %rsrc) { 70; UNPACKED-LABEL: buffer_load_format_d16_xyzw: 71; UNPACKED: ; %bb.0: ; %main_body 72; UNPACKED-NEXT: v_mov_b32_e32 v0, 0 ; encoding: [0x80,0x02,0x00,0x7e] 73; UNPACKED-NEXT: buffer_load_format_d16_xyzw v[0:3], v0, s[0:3], 0 idxen ; encoding: [0x00,0x20,0x2c,0xe0,0x00,0x00,0x00,0x80] 74; UNPACKED-NEXT: s_waitcnt vmcnt(0) ; encoding: [0x70,0x0f,0x8c,0xbf] 75; UNPACKED-NEXT: v_mov_b32_e32 v0, v3 ; encoding: [0x03,0x03,0x00,0x7e] 76; UNPACKED-NEXT: ; return to shader part epilog 77; 78; PACKED-LABEL: buffer_load_format_d16_xyzw: 79; PACKED: ; %bb.0: ; %main_body 80; PACKED-NEXT: v_mov_b32_e32 v0, 0 81; PACKED-NEXT: buffer_load_format_d16_xyzw v[0:1], v0, s[0:3], 0 idxen 82; PACKED-NEXT: s_waitcnt vmcnt(0) 83; PACKED-NEXT: v_lshrrev_b32_e32 v0, 16, v1 84; PACKED-NEXT: ; return to shader part epilog 85main_body: 86 %data = call <4 x half> @llvm.amdgcn.struct.ptr.buffer.load.format.v4f16(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0, i32 0) 87 %elt = extractelement <4 x half> %data, i32 3 88 ret half %elt 89} 90 91define amdgpu_ps half @buffer_load_format_i16_x(ptr addrspace(8) inreg %rsrc) { 92; UNPACKED-LABEL: buffer_load_format_i16_x: 93; UNPACKED: ; %bb.0: ; %main_body 94; UNPACKED-NEXT: v_mov_b32_e32 v0, 0 ; encoding: [0x80,0x02,0x00,0x7e] 95; UNPACKED-NEXT: buffer_load_format_d16_x v0, v0, s[0:3], 0 idxen ; encoding: [0x00,0x20,0x20,0xe0,0x00,0x00,0x00,0x80] 96; UNPACKED-NEXT: s_waitcnt vmcnt(0) ; encoding: [0x70,0x0f,0x8c,0xbf] 97; UNPACKED-NEXT: ; return to shader part epilog 98; 99; PACKED-LABEL: buffer_load_format_i16_x: 100; PACKED: ; %bb.0: ; %main_body 101; PACKED-NEXT: v_mov_b32_e32 v0, 0 102; PACKED-NEXT: buffer_load_format_d16_x v0, v0, s[0:3], 0 idxen 103; PACKED-NEXT: s_waitcnt vmcnt(0) 104; PACKED-NEXT: ; return to shader part epilog 105main_body: 106 %data = call i16 @llvm.amdgcn.struct.ptr.buffer.load.format.i16(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0, i32 0) 107 %fdata = bitcast i16 %data to half 108 ret half %fdata 109} 110 111declare half @llvm.amdgcn.struct.ptr.buffer.load.format.f16(ptr addrspace(8), i32, i32, i32, i32) 112declare <2 x half> @llvm.amdgcn.struct.ptr.buffer.load.format.v2f16(ptr addrspace(8), i32, i32, i32, i32) 113declare <3 x half> @llvm.amdgcn.struct.ptr.buffer.load.format.v3f16(ptr addrspace(8), i32, i32, i32, i32) 114declare <4 x half> @llvm.amdgcn.struct.ptr.buffer.load.format.v4f16(ptr addrspace(8), i32, i32, i32, i32) 115declare i16 @llvm.amdgcn.struct.ptr.buffer.load.format.i16(ptr addrspace(8), i32, i32, i32, i32) 116