1; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s -check-prefix=CHECK -check-prefix=ENABLED 2; RUN: llc --disable-x86-lea-opt < %s -mtriple=x86_64-linux | FileCheck %s -check-prefix=CHECK -check-prefix=DISABLED 3 4%struct.anon1 = type { i32, i32, i32 } 5%struct.anon2 = type { i32, [32 x i32], i32 } 6 7@arr1 = external global [65 x %struct.anon1], align 16 8@arr2 = external global [65 x %struct.anon2], align 16 9 10define void @test1(i64 %x) nounwind { 11entry: 12 %a = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 0 13 %tmp = load i32, i32* %a, align 4 14 %b = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 1 15 %tmp1 = load i32, i32* %b, align 4 16 %sub = sub i32 %tmp, %tmp1 17 %c = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 2 18 %tmp2 = load i32, i32* %c, align 4 19 %add = add nsw i32 %sub, %tmp2 20 switch i32 %add, label %sw.epilog [ 21 i32 1, label %sw.bb.1 22 i32 2, label %sw.bb.2 23 ] 24 25sw.bb.1: ; preds = %entry 26 store i32 111, i32* %b, align 4 27 store i32 222, i32* %c, align 4 28 br label %sw.epilog 29 30sw.bb.2: ; preds = %entry 31 store i32 333, i32* %b, align 4 32 store i32 444, i32* %c, align 4 33 br label %sw.epilog 34 35sw.epilog: ; preds = %sw.bb.2, %sw.bb.1, %entry 36 ret void 37; CHECK-LABEL: test1: 38; CHECK: shlq $2, [[REG1:%[a-z]+]] 39; CHECK: movl arr1([[REG1]],[[REG1]],2), {{.*}} 40; CHECK: leaq arr1+4([[REG1]],[[REG1]],2), [[REG2:%[a-z]+]] 41; CHECK: subl arr1+4([[REG1]],[[REG1]],2), {{.*}} 42; DISABLED: leaq arr1+8([[REG1]],[[REG1]],2), [[REG3:%[a-z]+]] 43; CHECK: addl arr1+8([[REG1]],[[REG1]],2), {{.*}} 44; CHECK: movl ${{[1-4]+}}, ([[REG2]]) 45; ENABLED: movl ${{[1-4]+}}, 4([[REG2]]) 46; DISABLED: movl ${{[1-4]+}}, ([[REG3]]) 47; CHECK: movl ${{[1-4]+}}, ([[REG2]]) 48; ENABLED: movl ${{[1-4]+}}, 4([[REG2]]) 49; DISABLED: movl ${{[1-4]+}}, ([[REG3]]) 50} 51 52define void @test2(i64 %x) nounwind optsize { 53entry: 54 %a = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 0 55 %tmp = load i32, i32* %a, align 4 56 %b = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 1 57 %tmp1 = load i32, i32* %b, align 4 58 %sub = sub i32 %tmp, %tmp1 59 %c = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 2 60 %tmp2 = load i32, i32* %c, align 4 61 %add = add nsw i32 %sub, %tmp2 62 switch i32 %add, label %sw.epilog [ 63 i32 1, label %sw.bb.1 64 i32 2, label %sw.bb.2 65 ] 66 67sw.bb.1: ; preds = %entry 68 store i32 111, i32* %b, align 4 69 store i32 222, i32* %c, align 4 70 br label %sw.epilog 71 72sw.bb.2: ; preds = %entry 73 store i32 333, i32* %b, align 4 74 store i32 444, i32* %c, align 4 75 br label %sw.epilog 76 77sw.epilog: ; preds = %sw.bb.2, %sw.bb.1, %entry 78 ret void 79; CHECK-LABEL: test2: 80; CHECK: shlq $2, [[REG1:%[a-z]+]] 81; DISABLED: movl arr1([[REG1]],[[REG1]],2), {{.*}} 82; CHECK: leaq arr1+4([[REG1]],[[REG1]],2), [[REG2:%[a-z]+]] 83; ENABLED: movl -4([[REG2]]), {{.*}} 84; ENABLED: subl ([[REG2]]), {{.*}} 85; ENABLED: addl 4([[REG2]]), {{.*}} 86; DISABLED: subl arr1+4([[REG1]],[[REG1]],2), {{.*}} 87; DISABLED: leaq arr1+8([[REG1]],[[REG1]],2), [[REG3:%[a-z]+]] 88; DISABLED: addl arr1+8([[REG1]],[[REG1]],2), {{.*}} 89; CHECK: movl ${{[1-4]+}}, ([[REG2]]) 90; ENABLED: movl ${{[1-4]+}}, 4([[REG2]]) 91; DISABLED: movl ${{[1-4]+}}, ([[REG3]]) 92; CHECK: movl ${{[1-4]+}}, ([[REG2]]) 93; ENABLED: movl ${{[1-4]+}}, 4([[REG2]]) 94; DISABLED: movl ${{[1-4]+}}, ([[REG3]]) 95} 96 97; Check that LEA optimization pass takes into account a resultant address 98; displacement when choosing a LEA instruction for replacing a redundant 99; address recalculation. 100 101define void @test3(i64 %x) nounwind optsize { 102entry: 103 %a = getelementptr inbounds [65 x %struct.anon2], [65 x %struct.anon2]* @arr2, i64 0, i64 %x, i32 2 104 %tmp = load i32, i32* %a, align 4 105 %b = getelementptr inbounds [65 x %struct.anon2], [65 x %struct.anon2]* @arr2, i64 0, i64 %x, i32 0 106 %tmp1 = load i32, i32* %b, align 4 107 %add = add nsw i32 %tmp, %tmp1 108 switch i32 %add, label %sw.epilog [ 109 i32 1, label %sw.bb.1 110 i32 2, label %sw.bb.2 111 ] 112 113sw.bb.1: ; preds = %entry 114 store i32 111, i32* %a, align 4 115 store i32 222, i32* %b, align 4 116 br label %sw.epilog 117 118sw.bb.2: ; preds = %entry 119 store i32 333, i32* %a, align 4 120 ; Make sure the REG3's definition LEA won't be removed as redundant. 121 %cvt = ptrtoint i32* %b to i32 122 store i32 %cvt, i32* %b, align 4 123 br label %sw.epilog 124 125sw.epilog: ; preds = %sw.bb.2, %sw.bb.1, %entry 126 ret void 127; CHECK-LABEL: test3: 128; CHECK: imulq {{.*}}, [[REG1:%[a-z]+]] 129; CHECK: leaq arr2+132([[REG1]]), [[REG2:%[a-z]+]] 130; CHECK: leaq arr2([[REG1]]), [[REG3:%[a-z]+]] 131 132; REG3's definition is closer to movl than REG2's, but the pass still chooses 133; REG2 because it provides the resultant address displacement fitting 1 byte. 134 135; ENABLED: movl ([[REG2]]), {{.*}} 136; ENABLED: addl ([[REG3]]), {{.*}} 137; DISABLED: movl arr2+132([[REG1]]), {{.*}} 138; DISABLED: addl arr2([[REG1]]), {{.*}} 139; CHECK: movl ${{[1-4]+}}, ([[REG2]]) 140; CHECK: movl ${{[1-4]+}}, ([[REG3]]) 141; CHECK: movl ${{[1-4]+}}, ([[REG2]]) 142; CHECK: movl {{.*}}, ([[REG3]]) 143} 144 145define void @test4(i64 %x) nounwind minsize { 146entry: 147 %a = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 0 148 %tmp = load i32, i32* %a, align 4 149 %b = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 1 150 %tmp1 = load i32, i32* %b, align 4 151 %sub = sub i32 %tmp, %tmp1 152 %c = getelementptr inbounds [65 x %struct.anon1], [65 x %struct.anon1]* @arr1, i64 0, i64 %x, i32 2 153 %tmp2 = load i32, i32* %c, align 4 154 %add = add nsw i32 %sub, %tmp2 155 switch i32 %add, label %sw.epilog [ 156 i32 1, label %sw.bb.1 157 i32 2, label %sw.bb.2 158 ] 159 160sw.bb.1: ; preds = %entry 161 store i32 111, i32* %b, align 4 162 store i32 222, i32* %c, align 4 163 br label %sw.epilog 164 165sw.bb.2: ; preds = %entry 166 store i32 333, i32* %b, align 4 167 store i32 444, i32* %c, align 4 168 br label %sw.epilog 169 170sw.epilog: ; preds = %sw.bb.2, %sw.bb.1, %entry 171 ret void 172; CHECK-LABEL: test4: 173; CHECK: imulq {{.*}}, [[REG1:%[a-z]+]] 174; DISABLED: movl arr1([[REG1]]), {{.*}} 175; CHECK: leaq arr1+4([[REG1]]), [[REG2:%[a-z]+]] 176; ENABLED: movl -4([[REG2]]), {{.*}} 177; ENABLED: subl ([[REG2]]), {{.*}} 178; ENABLED: addl 4([[REG2]]), {{.*}} 179; DISABLED: subl arr1+4([[REG1]]), {{.*}} 180; DISABLED: leaq arr1+8([[REG1]]), [[REG3:%[a-z]+]] 181; DISABLED: addl arr1+8([[REG1]]), {{.*}} 182; CHECK: movl ${{[1-4]+}}, ([[REG2]]) 183; ENABLED: movl ${{[1-4]+}}, 4([[REG2]]) 184; DISABLED: movl ${{[1-4]+}}, ([[REG3]]) 185; CHECK: movl ${{[1-4]+}}, ([[REG2]]) 186; ENABLED: movl ${{[1-4]+}}, 4([[REG2]]) 187; DISABLED: movl ${{[1-4]+}}, ([[REG3]]) 188} 189