113 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 22; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn < %s | FileCheck -check-prefixes=GFX6 %s3; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefixes=GFX8 %s4; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 -mattr=-flat-for-global < %s | FileCheck -check-prefixes=GFX11-TRUE16 %s5; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 -mattr=-flat-for-global < %s | FileCheck -check-prefixes=GFX11-FAKE16 %s6; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=r600 -mcpu=cypress < %s | FileCheck -check-prefixes=CYPRESS %s7 8declare i16 @llvm.convert.to.fp16.f32(float) nounwind readnone9 10define amdgpu_kernel void @test_convert_fp32_to_fp16(ptr addrspace(1) noalias %out, ptr addrspace(1) noalias %in) nounwind {11; GFX6-LABEL: test_convert_fp32_to_fp16:12; GFX6: ; %bb.0:13; GFX6-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x914; GFX6-NEXT: s_mov_b32 s7, 0xf00015; GFX6-NEXT: s_mov_b32 s6, -116; GFX6-NEXT: s_mov_b32 s10, s617; GFX6-NEXT: s_mov_b32 s11, s718; GFX6-NEXT: s_waitcnt lgkmcnt(0)19; GFX6-NEXT: s_mov_b32 s8, s220; GFX6-NEXT: s_mov_b32 s9, s321; GFX6-NEXT: buffer_load_dword v0, off, s[8:11], 022; GFX6-NEXT: s_mov_b32 s4, s023; GFX6-NEXT: s_mov_b32 s5, s124; GFX6-NEXT: s_waitcnt vmcnt(0)25; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v026; GFX6-NEXT: buffer_store_short v0, off, s[4:7], 027; GFX6-NEXT: s_endpgm28;29; GFX8-LABEL: test_convert_fp32_to_fp16:30; GFX8: ; %bb.0:31; GFX8-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x2432; GFX8-NEXT: s_mov_b32 s7, 0xf00033; GFX8-NEXT: s_mov_b32 s6, -134; GFX8-NEXT: s_mov_b32 s10, s635; GFX8-NEXT: s_mov_b32 s11, s736; GFX8-NEXT: s_waitcnt lgkmcnt(0)37; GFX8-NEXT: s_mov_b32 s8, s238; GFX8-NEXT: s_mov_b32 s9, s339; GFX8-NEXT: buffer_load_dword v0, off, s[8:11], 040; GFX8-NEXT: s_mov_b32 s4, s041; GFX8-NEXT: s_mov_b32 s5, s142; GFX8-NEXT: s_waitcnt vmcnt(0)43; GFX8-NEXT: v_cvt_f16_f32_e32 v0, v044; GFX8-NEXT: buffer_store_short v0, off, s[4:7], 045; GFX8-NEXT: s_endpgm46;47; GFX11-TRUE16-LABEL: test_convert_fp32_to_fp16:48; GFX11-TRUE16: ; %bb.0:49; GFX11-TRUE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x2450; GFX11-TRUE16-NEXT: s_mov_b32 s6, -151; GFX11-TRUE16-NEXT: s_mov_b32 s7, 0x3101600052; GFX11-TRUE16-NEXT: s_mov_b32 s10, s653; GFX11-TRUE16-NEXT: s_mov_b32 s11, s754; GFX11-TRUE16-NEXT: s_waitcnt lgkmcnt(0)55; GFX11-TRUE16-NEXT: s_mov_b32 s8, s256; GFX11-TRUE16-NEXT: s_mov_b32 s9, s357; GFX11-TRUE16-NEXT: s_mov_b32 s4, s058; GFX11-TRUE16-NEXT: buffer_load_b32 v0, off, s[8:11], 059; GFX11-TRUE16-NEXT: s_mov_b32 s5, s160; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)61; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.l, v062; GFX11-TRUE16-NEXT: buffer_store_b16 v0, off, s[4:7], 063; GFX11-TRUE16-NEXT: s_endpgm64;65; GFX11-FAKE16-LABEL: test_convert_fp32_to_fp16:66; GFX11-FAKE16: ; %bb.0:67; GFX11-FAKE16-NEXT: s_load_b128 s[0:3], s[4:5], 0x2468; GFX11-FAKE16-NEXT: s_mov_b32 s6, -169; GFX11-FAKE16-NEXT: s_mov_b32 s7, 0x3101600070; GFX11-FAKE16-NEXT: s_mov_b32 s10, s671; GFX11-FAKE16-NEXT: s_mov_b32 s11, s772; GFX11-FAKE16-NEXT: s_waitcnt lgkmcnt(0)73; GFX11-FAKE16-NEXT: s_mov_b32 s8, s274; GFX11-FAKE16-NEXT: s_mov_b32 s9, s375; GFX11-FAKE16-NEXT: s_mov_b32 s4, s076; GFX11-FAKE16-NEXT: buffer_load_b32 v0, off, s[8:11], 077; GFX11-FAKE16-NEXT: s_mov_b32 s5, s178; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)79; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v0, v080; GFX11-FAKE16-NEXT: buffer_store_b16 v0, off, s[4:7], 081; GFX11-FAKE16-NEXT: s_endpgm82;83; CYPRESS-LABEL: test_convert_fp32_to_fp16:84; CYPRESS: ; %bb.0:85; CYPRESS-NEXT: ALU 0, @8, KC0[CB0:0-32], KC1[]86; CYPRESS-NEXT: TEX 0 @687; CYPRESS-NEXT: ALU 11, @9, KC0[CB0:0-32], KC1[]88; CYPRESS-NEXT: MEM_RAT MSKOR T0.XW, T1.X89; CYPRESS-NEXT: CF_END90; CYPRESS-NEXT: PAD91; CYPRESS-NEXT: Fetch clause starting at 6:92; CYPRESS-NEXT: VTX_READ_32 T0.X, T0.X, 0, #193; CYPRESS-NEXT: ALU clause starting at 8:94; CYPRESS-NEXT: MOV * T0.X, KC0[2].Z,95; CYPRESS-NEXT: ALU clause starting at 9:96; CYPRESS-NEXT: AND_INT * T0.W, KC0[2].Y, literal.x,97; CYPRESS-NEXT: 3(4.203895e-45), 0(0.000000e+00)98; CYPRESS-NEXT: FLT32_TO_FLT16 T1.W, T0.X,99; CYPRESS-NEXT: LSHL * T0.W, PV.W, literal.x,100; CYPRESS-NEXT: 3(4.203895e-45), 0(0.000000e+00)101; CYPRESS-NEXT: LSHL T0.X, PV.W, PS,102; CYPRESS-NEXT: LSHL * T0.W, literal.x, PS,103; CYPRESS-NEXT: 65535(9.183409e-41), 0(0.000000e+00)104; CYPRESS-NEXT: MOV T0.Y, 0.0,105; CYPRESS-NEXT: MOV * T0.Z, 0.0,106; CYPRESS-NEXT: LSHR * T1.X, KC0[2].Y, literal.x,107; CYPRESS-NEXT: 2(2.802597e-45), 0(0.000000e+00)108 %val = load float, ptr addrspace(1) %in, align 4109 %cvt = call i16 @llvm.convert.to.fp16.f32(float %val) nounwind readnone110 store i16 %cvt, ptr addrspace(1) %out, align 2111 ret void112}113