1; RUN: llc -march=amdgcn < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 3; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s 4 5; The code generated by sdiv is long and complex and may frequently change. 6; The goal of this test is to make sure the ISel doesn't fail. 7; 8; This program was previously failing to compile when one of the selectcc 9; opcodes generated by the sdiv lowering was being legalized and optimized to: 10; selectcc Remainder -1, 0, -1, SETGT 11; This was fixed by adding an additional pattern in R600Instructions.td to 12; match this pattern with a CNDGE_INT. 13 14; FUNC-LABEL: {{^}}sdiv_i32: 15; EG: CF_END 16define void @sdiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 17 %den_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 18 %num = load i32, i32 addrspace(1) * %in 19 %den = load i32, i32 addrspace(1) * %den_ptr 20 %result = sdiv i32 %num, %den 21 store i32 %result, i32 addrspace(1)* %out 22 ret void 23} 24 25; FUNC-LABEL: {{^}}sdiv_i32_4: 26define void @sdiv_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 27 %num = load i32, i32 addrspace(1) * %in 28 %result = sdiv i32 %num, 4 29 store i32 %result, i32 addrspace(1)* %out 30 ret void 31} 32 33; Multiply by a weird constant to make sure setIntDivIsCheap is 34; working. 35 36; FUNC-LABEL: {{^}}slow_sdiv_i32_3435: 37; SI-DAG: buffer_load_dword [[VAL:v[0-9]+]], 38; SI-DAG: v_mov_b32_e32 [[MAGIC:v[0-9]+]], 0x98a1930b 39; SI: v_mul_hi_i32 [[TMP:v[0-9]+]], [[MAGIC]], [[VAL]] 40; SI: v_add_i32 41; SI: v_lshrrev_b32 42; SI: v_ashrrev_i32 43; SI: v_add_i32 44; SI: buffer_store_dword 45; SI: s_endpgm 46define void @slow_sdiv_i32_3435(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 47 %num = load i32, i32 addrspace(1) * %in 48 %result = sdiv i32 %num, 3435 49 store i32 %result, i32 addrspace(1)* %out 50 ret void 51} 52 53define void @sdiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { 54 %den_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1 55 %num = load <2 x i32>, <2 x i32> addrspace(1) * %in 56 %den = load <2 x i32>, <2 x i32> addrspace(1) * %den_ptr 57 %result = sdiv <2 x i32> %num, %den 58 store <2 x i32> %result, <2 x i32> addrspace(1)* %out 59 ret void 60} 61 62define void @sdiv_v2i32_4(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { 63 %num = load <2 x i32>, <2 x i32> addrspace(1) * %in 64 %result = sdiv <2 x i32> %num, <i32 4, i32 4> 65 store <2 x i32> %result, <2 x i32> addrspace(1)* %out 66 ret void 67} 68 69define void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { 70 %den_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1 71 %num = load <4 x i32>, <4 x i32> addrspace(1) * %in 72 %den = load <4 x i32>, <4 x i32> addrspace(1) * %den_ptr 73 %result = sdiv <4 x i32> %num, %den 74 store <4 x i32> %result, <4 x i32> addrspace(1)* %out 75 ret void 76} 77 78define void @sdiv_v4i32_4(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { 79 %num = load <4 x i32>, <4 x i32> addrspace(1) * %in 80 %result = sdiv <4 x i32> %num, <i32 4, i32 4, i32 4, i32 4> 81 store <4 x i32> %result, <4 x i32> addrspace(1)* %out 82 ret void 83} 84 85; FUNC-LABEL: {{^}}v_sdiv_i8: 86; SI: v_rcp_f32 87; SI: v_bfe_i32 [[BFE:v[0-9]+]], v{{[0-9]+}}, 0, 8 88; SI: buffer_store_dword [[BFE]] 89define void @v_sdiv_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %in) { 90 %den_ptr = getelementptr i8, i8 addrspace(1)* %in, i8 1 91 %num = load i8, i8 addrspace(1) * %in 92 %den = load i8, i8 addrspace(1) * %den_ptr 93 %result = sdiv i8 %num, %den 94 %result.ext = sext i8 %result to i32 95 store i32 %result.ext, i32 addrspace(1)* %out 96 ret void 97} 98 99; FUNC-LABEL: {{^}}v_sdiv_i23: 100; SI: v_rcp_f32 101; SI: v_bfe_i32 [[BFE:v[0-9]+]], v{{[0-9]+}}, 0, 23 102; SI: buffer_store_dword [[BFE]] 103define void @v_sdiv_i23(i32 addrspace(1)* %out, i23 addrspace(1)* %in) { 104 %den_ptr = getelementptr i23, i23 addrspace(1)* %in, i23 1 105 %num = load i23, i23 addrspace(1) * %in 106 %den = load i23, i23 addrspace(1) * %den_ptr 107 %result = sdiv i23 %num, %den 108 %result.ext = sext i23 %result to i32 109 store i32 %result.ext, i32 addrspace(1)* %out 110 ret void 111} 112 113; FUNC-LABEL: {{^}}v_sdiv_i24: 114; SI: v_rcp_f32 115; SI: v_bfe_i32 [[BFE:v[0-9]+]], v{{[0-9]+}}, 0, 24 116; SI: buffer_store_dword [[BFE]] 117define void @v_sdiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)* %in) { 118 %den_ptr = getelementptr i24, i24 addrspace(1)* %in, i24 1 119 %num = load i24, i24 addrspace(1) * %in 120 %den = load i24, i24 addrspace(1) * %den_ptr 121 %result = sdiv i24 %num, %den 122 %result.ext = sext i24 %result to i32 123 store i32 %result.ext, i32 addrspace(1)* %out 124 ret void 125} 126 127; FUNC-LABEL: {{^}}v_sdiv_i25: 128; SI-NOT: v_rcp_f32 129define void @v_sdiv_i25(i32 addrspace(1)* %out, i25 addrspace(1)* %in) { 130 %den_ptr = getelementptr i25, i25 addrspace(1)* %in, i25 1 131 %num = load i25, i25 addrspace(1) * %in 132 %den = load i25, i25 addrspace(1) * %den_ptr 133 %result = sdiv i25 %num, %den 134 %result.ext = sext i25 %result to i32 135 store i32 %result.ext, i32 addrspace(1)* %out 136 ret void 137} 138 139; Tests for 64-bit divide bypass. 140; define void @test_get_quotient(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind { 141; %result = sdiv i64 %a, %b 142; store i64 %result, i64 addrspace(1)* %out, align 8 143; ret void 144; } 145 146; define void @test_get_remainder(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind { 147; %result = srem i64 %a, %b 148; store i64 %result, i64 addrspace(1)* %out, align 8 149; ret void 150; } 151 152; define void @test_get_quotient_and_remainder(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind { 153; %resultdiv = sdiv i64 %a, %b 154; %resultrem = srem i64 %a, %b 155; %result = add i64 %resultdiv, %resultrem 156; store i64 %result, i64 addrspace(1)* %out, align 8 157; ret void 158; } 159