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: buffer_load_format_d16_x v0, off, s[0:3], 0 ; encoding: [0x00,0x00,0x20,0xe0,0x00,0x00,0x00,0x80] 10; UNPACKED-NEXT: s_waitcnt vmcnt(0) ; encoding: [0x70,0x0f,0x8c,0xbf] 11; UNPACKED-NEXT: ; return to shader part epilog 12; 13; PACKED-LABEL: buffer_load_format_d16_x: 14; PACKED: ; %bb.0: ; %main_body 15; PACKED-NEXT: buffer_load_format_d16_x v0, off, s[0:3], 0 16; PACKED-NEXT: s_waitcnt vmcnt(0) 17; PACKED-NEXT: ; return to shader part epilog 18main_body: 19 %data = call half @llvm.amdgcn.raw.ptr.buffer.load.format.f16(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0) 20 ret half %data 21} 22 23define amdgpu_ps half @buffer_load_format_d16_xy(ptr addrspace(8) inreg %rsrc) { 24; UNPACKED-LABEL: buffer_load_format_d16_xy: 25; UNPACKED: ; %bb.0: ; %main_body 26; UNPACKED-NEXT: buffer_load_format_d16_xy v[0:1], off, s[0:3], 0 ; encoding: [0x00,0x00,0x24,0xe0,0x00,0x00,0x00,0x80] 27; UNPACKED-NEXT: s_waitcnt vmcnt(0) ; encoding: [0x70,0x0f,0x8c,0xbf] 28; UNPACKED-NEXT: v_mov_b32_e32 v0, v1 ; encoding: [0x01,0x03,0x00,0x7e] 29; UNPACKED-NEXT: ; return to shader part epilog 30; 31; PACKED-LABEL: buffer_load_format_d16_xy: 32; PACKED: ; %bb.0: ; %main_body 33; PACKED-NEXT: buffer_load_format_d16_xy v0, off, s[0:3], 0 34; PACKED-NEXT: s_waitcnt vmcnt(0) 35; PACKED-NEXT: v_lshrrev_b32_e32 v0, 16, v0 36; PACKED-NEXT: ; return to shader part epilog 37main_body: 38 %data = call <2 x half> @llvm.amdgcn.raw.ptr.buffer.load.format.v2f16(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0) 39 %elt = extractelement <2 x half> %data, i32 1 40 ret half %elt 41} 42 43define amdgpu_ps half @buffer_load_format_d16_xyz(ptr addrspace(8) inreg %rsrc) { 44; UNPACKED-LABEL: buffer_load_format_d16_xyz: 45; UNPACKED: ; %bb.0: ; %main_body 46; UNPACKED-NEXT: buffer_load_format_d16_xyz v[0:2], off, s[0:3], 0 ; encoding: [0x00,0x00,0x28,0xe0,0x00,0x00,0x00,0x80] 47; UNPACKED-NEXT: s_waitcnt vmcnt(0) ; encoding: [0x70,0x0f,0x8c,0xbf] 48; UNPACKED-NEXT: v_mov_b32_e32 v0, v2 ; encoding: [0x02,0x03,0x00,0x7e] 49; UNPACKED-NEXT: ; return to shader part epilog 50; 51; PACKED-LABEL: buffer_load_format_d16_xyz: 52; PACKED: ; %bb.0: ; %main_body 53; PACKED-NEXT: buffer_load_format_d16_xyz v[0:1], off, s[0:3], 0 54; PACKED-NEXT: s_waitcnt vmcnt(0) 55; PACKED-NEXT: v_mov_b32_e32 v0, v1 56; PACKED-NEXT: ; return to shader part epilog 57main_body: 58 %data = call <3 x half> @llvm.amdgcn.raw.ptr.buffer.load.format.v3f16(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0) 59 %elt = extractelement <3 x half> %data, i32 2 60 ret half %elt 61} 62 63define amdgpu_ps half @buffer_load_format_d16_xyzw(ptr addrspace(8) inreg %rsrc) { 64; UNPACKED-LABEL: buffer_load_format_d16_xyzw: 65; UNPACKED: ; %bb.0: ; %main_body 66; UNPACKED-NEXT: buffer_load_format_d16_xyzw v[0:3], off, s[0:3], 0 ; encoding: [0x00,0x00,0x2c,0xe0,0x00,0x00,0x00,0x80] 67; UNPACKED-NEXT: s_waitcnt vmcnt(0) ; encoding: [0x70,0x0f,0x8c,0xbf] 68; UNPACKED-NEXT: v_mov_b32_e32 v0, v3 ; encoding: [0x03,0x03,0x00,0x7e] 69; UNPACKED-NEXT: ; return to shader part epilog 70; 71; PACKED-LABEL: buffer_load_format_d16_xyzw: 72; PACKED: ; %bb.0: ; %main_body 73; PACKED-NEXT: buffer_load_format_d16_xyzw v[0:1], off, s[0:3], 0 74; PACKED-NEXT: s_waitcnt vmcnt(0) 75; PACKED-NEXT: v_lshrrev_b32_e32 v0, 16, v1 76; PACKED-NEXT: ; return to shader part epilog 77main_body: 78 %data = call <4 x half> @llvm.amdgcn.raw.ptr.buffer.load.format.v4f16(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0) 79 %elt = extractelement <4 x half> %data, i32 3 80 ret half %elt 81} 82 83declare half @llvm.amdgcn.raw.ptr.buffer.load.format.f16(ptr addrspace(8), i32, i32, i32) 84declare <2 x half> @llvm.amdgcn.raw.ptr.buffer.load.format.v2f16(ptr addrspace(8), i32, i32, i32) 85declare <3 x half> @llvm.amdgcn.raw.ptr.buffer.load.format.v3f16(ptr addrspace(8), i32, i32, i32) 86declare <4 x half> @llvm.amdgcn.raw.ptr.buffer.load.format.v4f16(ptr addrspace(8), i32, i32, i32) 87