Matt Arsenault | b005548 | 2015-01-21 18:18:25 +0000 | [diff] [blame^] | 1 | ; RUN: llc -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s |
| 2 | |
| 3 | ; FUNC-LABEL: {{^}}round_f64: |
| 4 | ; SI: s_endpgm |
| 5 | define void @round_f64(double addrspace(1)* %out, double %x) #0 { |
| 6 | %result = call double @llvm.round.f64(double %x) #1 |
| 7 | store double %result, double addrspace(1)* %out |
| 8 | ret void |
| 9 | } |
| 10 | |
| 11 | ; This is a pretty large function, so just test a few of the |
| 12 | ; instructions that are necessary. |
| 13 | |
| 14 | ; FUNC-LABEL: {{^}}v_round_f64: |
| 15 | ; SI: buffer_load_dwordx2 |
| 16 | ; SI: v_bfe_u32 [[EXP:v[0-9]+]], v{{[0-9]+}}, 20, 11 |
| 17 | |
| 18 | ; SI: v_not_b32_e32 |
| 19 | ; SI: v_not_b32_e32 |
| 20 | |
| 21 | ; SI: v_cmp_eq_i32 |
| 22 | |
| 23 | ; SI: s_mov_b32 [[BFIMASK:s[0-9]+]], 0x7fffffff |
| 24 | ; SI: v_bfi_b32 [[COPYSIGN:v[0-9]+]], [[BFIMASK]] |
| 25 | |
| 26 | ; SI: v_cmp_lt_i32_e64 |
| 27 | ; SI: v_cmp_gt_i32_e64 |
| 28 | |
| 29 | |
| 30 | ; SI: buffer_store_dwordx2 |
| 31 | ; SI: s_endpgm |
| 32 | define void @v_round_f64(double addrspace(1)* %out, double addrspace(1)* %in) #0 { |
| 33 | %tid = call i32 @llvm.r600.read.tidig.x() #1 |
| 34 | %gep = getelementptr double addrspace(1)* %in, i32 %tid |
| 35 | %out.gep = getelementptr double addrspace(1)* %out, i32 %tid |
| 36 | %x = load double addrspace(1)* %gep |
| 37 | %result = call double @llvm.round.f64(double %x) #1 |
| 38 | store double %result, double addrspace(1)* %out.gep |
| 39 | ret void |
| 40 | } |
| 41 | |
| 42 | ; FUNC-LABEL: {{^}}round_v2f64: |
| 43 | ; SI: s_endpgm |
| 44 | define void @round_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %in) #0 { |
| 45 | %result = call <2 x double> @llvm.round.v2f64(<2 x double> %in) #1 |
| 46 | store <2 x double> %result, <2 x double> addrspace(1)* %out |
| 47 | ret void |
| 48 | } |
| 49 | |
| 50 | ; FUNC-LABEL: {{^}}round_v4f64: |
| 51 | ; SI: s_endpgm |
| 52 | define void @round_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %in) #0 { |
| 53 | %result = call <4 x double> @llvm.round.v4f64(<4 x double> %in) #1 |
| 54 | store <4 x double> %result, <4 x double> addrspace(1)* %out |
| 55 | ret void |
| 56 | } |
| 57 | |
| 58 | ; FUNC-LABEL: {{^}}round_v8f64: |
| 59 | ; SI: s_endpgm |
| 60 | define void @round_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %in) #0 { |
| 61 | %result = call <8 x double> @llvm.round.v8f64(<8 x double> %in) #1 |
| 62 | store <8 x double> %result, <8 x double> addrspace(1)* %out |
| 63 | ret void |
| 64 | } |
| 65 | |
| 66 | declare i32 @llvm.r600.read.tidig.x() #1 |
| 67 | |
| 68 | declare double @llvm.round.f64(double) #1 |
| 69 | declare <2 x double> @llvm.round.v2f64(<2 x double>) #1 |
| 70 | declare <4 x double> @llvm.round.v4f64(<4 x double>) #1 |
| 71 | declare <8 x double> @llvm.round.v8f64(<8 x double>) #1 |
| 72 | |
| 73 | attributes #0 = { nounwind } |
| 74 | attributes #1 = { nounwind readnone } |