1 ; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s 3 4 5 declare i32 @llvm.r600.read.tidig.x() readnone 6 7 ; FUNC-LABEL: {{^}}test_sub_i32: 8 ; EG: SUB_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} 9 10 ; SI: v_subrev_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}} 11 define void @test_sub_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 12 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 13 %a = load i32, i32 addrspace(1)* %in 14 %b = load i32, i32 addrspace(1)* %b_ptr 15 %result = sub i32 %a, %b 16 store i32 %result, i32 addrspace(1)* %out 17 ret void 18 } 19 20 21 ; FUNC-LABEL: {{^}}test_sub_v2i32: 22 ; EG: SUB_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} 23 ; EG: SUB_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} 24 25 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}} 26 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}} 27 28 define void @test_sub_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { 29 %b_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1 30 %a = load <2 x i32>, <2 x i32> addrspace(1) * %in 31 %b = load <2 x i32>, <2 x i32> addrspace(1) * %b_ptr 32 %result = sub <2 x i32> %a, %b 33 store <2 x i32> %result, <2 x i32> addrspace(1)* %out 34 ret void 35 } 36 37 ; FUNC-LABEL: {{^}}test_sub_v4i32: 38 ; EG: SUB_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} 39 ; EG: SUB_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} 40 ; EG: SUB_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} 41 ; EG: SUB_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}} 42 43 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}} 44 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}} 45 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}} 46 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}} 47 48 define void @test_sub_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { 49 %b_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1 50 %a = load <4 x i32>, <4 x i32> addrspace(1) * %in 51 %b = load <4 x i32>, <4 x i32> addrspace(1) * %b_ptr 52 %result = sub <4 x i32> %a, %b 53 store <4 x i32> %result, <4 x i32> addrspace(1)* %out 54 ret void 55 } 56 57 ; FUNC-LABEL: {{^}}s_sub_i64: 58 ; SI: s_sub_u32 59 ; SI: s_subb_u32 60 61 ; EG: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+}}.XY 62 ; EG-DAG: SUB_INT {{[* ]*}} 63 ; EG-DAG: SUBB_UINT 64 ; EG-DAG: SUB_INT 65 ; EG-DAG: SUB_INT {{[* ]*}} 66 define void @s_sub_i64(i64 addrspace(1)* noalias %out, i64 %a, i64 %b) nounwind { 67 %result = sub i64 %a, %b 68 store i64 %result, i64 addrspace(1)* %out, align 8 69 ret void 70 } 71 72 ; FUNC-LABEL: {{^}}v_sub_i64: 73 ; SI: v_sub_i32_e32 74 ; SI: v_subb_u32_e32 75 76 ; EG: MEM_RAT_CACHELESS STORE_RAW T{{[0-9]+}}.XY 77 ; EG-DAG: SUB_INT {{[* ]*}} 78 ; EG-DAG: SUBB_UINT 79 ; EG-DAG: SUB_INT 80 ; EG-DAG: SUB_INT {{[* ]*}} 81 define void @v_sub_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %inA, i64 addrspace(1)* noalias %inB) nounwind { 82 %tid = call i32 @llvm.r600.read.tidig.x() readnone 83 %a_ptr = getelementptr i64, i64 addrspace(1)* %inA, i32 %tid 84 %b_ptr = getelementptr i64, i64 addrspace(1)* %inB, i32 %tid 85 %a = load i64, i64 addrspace(1)* %a_ptr 86 %b = load i64, i64 addrspace(1)* %b_ptr 87 %result = sub i64 %a, %b 88 store i64 %result, i64 addrspace(1)* %out, align 8 89 ret void 90 } 91 92 ; FUNC-LABEL: {{^}}v_test_sub_v2i64: 93 ; SI: v_sub_i32_e32 94 ; SI: v_subb_u32_e32 95 ; SI: v_sub_i32_e32 96 ; SI: v_subb_u32_e32 97 define void @v_test_sub_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* noalias %inA, <2 x i64> addrspace(1)* noalias %inB) { 98 %tid = call i32 @llvm.r600.read.tidig.x() readnone 99 %a_ptr = getelementptr <2 x i64>, <2 x i64> addrspace(1)* %inA, i32 %tid 100 %b_ptr = getelementptr <2 x i64>, <2 x i64> addrspace(1)* %inB, i32 %tid 101 %a = load <2 x i64>, <2 x i64> addrspace(1)* %a_ptr 102 %b = load <2 x i64>, <2 x i64> addrspace(1)* %b_ptr 103 %result = sub <2 x i64> %a, %b 104 store <2 x i64> %result, <2 x i64> addrspace(1)* %out 105 ret void 106 } 107 108 ; FUNC-LABEL: {{^}}v_test_sub_v4i64: 109 ; SI: v_subrev_i32_e32 110 ; SI: v_subb_u32_e32 111 ; SI: v_subrev_i32_e32 112 ; SI: v_subb_u32_e32 113 ; SI: v_subrev_i32_e32 114 ; SI: v_subb_u32_e32 115 ; SI: v_subrev_i32_e32 116 ; SI: v_subb_u32_e32 117 define void @v_test_sub_v4i64(<4 x i64> addrspace(1)* %out, <4 x i64> addrspace(1)* noalias %inA, <4 x i64> addrspace(1)* noalias %inB) { 118 %tid = call i32 @llvm.r600.read.tidig.x() readnone 119 %a_ptr = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %inA, i32 %tid 120 %b_ptr = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %inB, i32 %tid 121 %a = load <4 x i64>, <4 x i64> addrspace(1)* %a_ptr 122 %b = load <4 x i64>, <4 x i64> addrspace(1)* %b_ptr 123 %result = sub <4 x i64> %a, %b 124 store <4 x i64> %result, <4 x i64> addrspace(1)* %out 125 ret void 126 } 127
