292 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -mtriple=riscv64-unknown-elf -passes='separate-const-offset-from-gep,early-cse' \3; RUN: -S | FileCheck %s4 5; Several tests for separate-const-offset-from-gep. The transformation6; heavily relies on TargetTransformInfo, so we put these tests under7; target-specific folders.8 9; Simple case when GEPs should be optimized.10define i64 @test1(ptr %array, i64 %i, i64 %j) {11; CHECK-LABEL: @test1(12; CHECK-NEXT: entry:13; CHECK-NEXT: [[ADD:%.*]] = add nsw i64 [[I:%.*]], 514; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i64, ptr [[ARRAY:%.*]], i64 [[I]]15; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[TMP0]], i64 4016; CHECK-NEXT: store i64 [[J:%.*]], ptr [[GEP4]], align 817; CHECK-NEXT: [[GEP26:%.*]] = getelementptr i8, ptr [[TMP0]], i64 4818; CHECK-NEXT: store i64 [[J]], ptr [[GEP26]], align 819; CHECK-NEXT: [[GEP38:%.*]] = getelementptr i8, ptr [[TMP0]], i64 28020; CHECK-NEXT: store i64 [[ADD]], ptr [[GEP38]], align 821; CHECK-NEXT: ret i64 undef22;23entry:24 %add = add nsw i64 %i, 525 %gep = getelementptr inbounds i64, ptr %array, i64 %add26 store i64 %j, ptr %gep27 %add2 = add nsw i64 %i, 628 %gep2 = getelementptr inbounds i64, ptr %array, i64 %add229 store i64 %j, ptr %gep230 %add3 = add nsw i64 %i, 3531 %gep3 = getelementptr inbounds i64, ptr %array, i64 %add332 store i64 %add, ptr %gep333 ret i64 undef34}35 36; Optimize GEPs when there sext instructions are needed to cast index value to expected type.37define i32 @test2(ptr %array, i32 %i, i32 %j) {38; CHECK-LABEL: @test2(39; CHECK-NEXT: entry:40; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[I:%.*]], 541; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[I]] to i6442; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[ARRAY:%.*]], i64 [[TMP0]]43; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[TMP1]], i64 2044; CHECK-NEXT: store i32 [[J:%.*]], ptr [[GEP2]], align 445; CHECK-NEXT: [[GEP54:%.*]] = getelementptr i8, ptr [[TMP1]], i64 2446; CHECK-NEXT: store i32 [[J]], ptr [[GEP54]], align 447; CHECK-NEXT: [[GEP86:%.*]] = getelementptr i8, ptr [[TMP1]], i64 14048; CHECK-NEXT: store i32 [[ADD]], ptr [[GEP86]], align 449; CHECK-NEXT: ret i32 undef50;51entry:52 %add = add nsw i32 %i, 553 %sext = sext i32 %add to i6454 %gep = getelementptr inbounds i32, ptr %array, i64 %sext55 store i32 %j, ptr %gep56 %add3 = add nsw i32 %i, 657 %sext4 = sext i32 %add3 to i6458 %gep5 = getelementptr inbounds i32, ptr %array, i64 %sext459 store i32 %j, ptr %gep560 %add6 = add nsw i32 %i, 3561 %sext7 = sext i32 %add6 to i6462 %gep8 = getelementptr inbounds i32, ptr %array, i64 %sext763 store i32 %add, ptr %gep864 ret i32 undef65}66 67; No need to modify because all values are also used in other expressions.68; Modification doesn't decrease register pressure.69define i32 @test3(ptr %array, i32 %i) {70; CHECK-LABEL: @test3(71; CHECK-NEXT: entry:72; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[I:%.*]], 573; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[I]] to i6474; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[ARRAY:%.*]], i64 [[TMP0]]75; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[TMP1]], i64 2076; CHECK-NEXT: store i32 [[ADD]], ptr [[GEP2]], align 477; CHECK-NEXT: [[ADD3:%.*]] = add nsw i32 [[I]], 678; CHECK-NEXT: [[GEP54:%.*]] = getelementptr i8, ptr [[TMP1]], i64 2479; CHECK-NEXT: store i32 [[ADD3]], ptr [[GEP54]], align 480; CHECK-NEXT: [[ADD6:%.*]] = add nsw i32 [[I]], 3581; CHECK-NEXT: [[GEP86:%.*]] = getelementptr i8, ptr [[TMP1]], i64 14082; CHECK-NEXT: store i32 [[ADD6]], ptr [[GEP86]], align 483; CHECK-NEXT: ret i32 undef84;85entry:86 %add = add nsw i32 %i, 587 %sext = sext i32 %add to i6488 %gep = getelementptr inbounds i32, ptr %array, i64 %sext89 store i32 %add, ptr %gep90 %add3 = add nsw i32 %i, 691 %sext4 = sext i32 %add3 to i6492 %gep5 = getelementptr inbounds i32, ptr %array, i64 %sext493 store i32 %add3, ptr %gep594 %add6 = add nsw i32 %i, 3595 %sext7 = sext i32 %add6 to i6496 %gep8 = getelementptr inbounds i32, ptr %array, i64 %sext797 store i32 %add6, ptr %gep898 ret i32 undef99}100 101; Optimized GEPs for multidimensional array with same base102define i32 @test4(ptr %array2, i32 %i) {103; CHECK-LABEL: @test4(104; CHECK-NEXT: entry:105; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[I:%.*]], 5106; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[I]] to i64107; CHECK-NEXT: [[TMP1:%.*]] = getelementptr [50 x i32], ptr [[ARRAY2:%.*]], i64 [[TMP0]], i64 [[TMP0]]108; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[TMP1]], i64 1020109; CHECK-NEXT: store i32 [[I]], ptr [[GEP3]], align 4110; CHECK-NEXT: [[GEP56:%.*]] = getelementptr i8, ptr [[TMP1]], i64 1024111; CHECK-NEXT: store i32 [[ADD]], ptr [[GEP56]], align 4112; CHECK-NEXT: [[GEP89:%.*]] = getelementptr i8, ptr [[TMP1]], i64 1140113; CHECK-NEXT: store i32 [[I]], ptr [[GEP89]], align 4114; CHECK-NEXT: ret i32 undef115;116entry:117 %add = add nsw i32 %i, 5118 %sext = sext i32 %add to i64119 %gep = getelementptr inbounds [50 x i32], ptr %array2, i64 %sext, i64 %sext120 store i32 %i, ptr %gep121 %add3 = add nsw i32 %i, 6122 %sext4 = sext i32 %add3 to i64123 %gep5 = getelementptr inbounds [50 x i32], ptr %array2, i64 %sext, i64 %sext4124 store i32 %add, ptr %gep5125 %add6 = add nsw i32 %i, 35126 %sext7 = sext i32 %add6 to i64127 %gep8 = getelementptr inbounds [50 x i32], ptr %array2, i64 %sext, i64 %sext7128 store i32 %i, ptr %gep8129 ret i32 undef130}131 132; Don't optimize GEPs for multidimensional array with same base because RISC-V doesn't support the addressing mode133define i32 @test5(ptr %array2, i32 %i, i64 %j) {134; CHECK-LABEL: @test5(135; CHECK-NEXT: entry:136; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[I:%.*]], 5137; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[I]] to i64138; CHECK-NEXT: [[TMP1:%.*]] = getelementptr [50 x i32], ptr [[ARRAY2:%.*]], i64 [[TMP0]], i64 [[TMP0]]139; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[TMP1]], i64 1020140; CHECK-NEXT: store i32 [[ADD]], ptr [[GEP3]], align 4141; CHECK-NEXT: [[TMP2:%.*]] = getelementptr [50 x i32], ptr [[ARRAY2]], i64 [[TMP0]], i64 [[J:%.*]]142; CHECK-NEXT: [[GEP55:%.*]] = getelementptr i8, ptr [[TMP2]], i64 1200143; CHECK-NEXT: store i32 [[I]], ptr [[GEP55]], align 4144; CHECK-NEXT: [[ADD6:%.*]] = add nsw i32 [[I]], 35145; CHECK-NEXT: [[SEXT7:%.*]] = sext i32 [[ADD6]] to i64146; CHECK-NEXT: [[GEP8:%.*]] = getelementptr inbounds [50 x i32], ptr [[ARRAY2]], i64 [[SEXT7]], i64 [[J]]147; CHECK-NEXT: store i32 [[I]], ptr [[GEP8]], align 4148; CHECK-NEXT: ret i32 undef149;150entry:151 %add = add nsw i32 %i, 5152 %sext = sext i32 %add to i64153 %gep = getelementptr inbounds [50 x i32], ptr %array2, i64 %sext, i64 %sext154 store i32 %add, ptr %gep155 %add3 = add nsw i32 %i, 6156 %sext4 = sext i32 %add3 to i64157 %gep5 = getelementptr inbounds [50 x i32], ptr %array2, i64 %sext4, i64 %j158 store i32 %i, ptr %gep5159 %add6 = add nsw i32 %i, 35160 %sext7 = sext i32 %add6 to i64161 %gep8 = getelementptr inbounds [50 x i32], ptr %array2, i64 %sext7, i64 %j162 store i32 %i, ptr %gep8163 ret i32 undef164}165 166; No need to optimize GEPs, because there is critical amount with non-constant offsets.167define i64 @test6(ptr %array, i64 %i, i64 %j) {168; CHECK-LABEL: @test6(169; CHECK-NEXT: entry:170; CHECK-NEXT: [[ADD:%.*]] = add nsw i64 [[I:%.*]], 5171; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i64, ptr [[ARRAY:%.*]], i64 [[J:%.*]]172; CHECK-NEXT: store i64 [[ADD]], ptr [[GEP]], align 8173; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i64, ptr [[ARRAY]], i64 [[I]]174; CHECK-NEXT: [[GEP52:%.*]] = getelementptr i8, ptr [[TMP0]], i64 48175; CHECK-NEXT: store i64 [[I]], ptr [[GEP52]], align 8176; CHECK-NEXT: store i64 [[I]], ptr [[TMP0]], align 8177; CHECK-NEXT: ret i64 undef178;179entry:180 %add = add nsw i64 %i, 5181 %gep = getelementptr inbounds i64, ptr %array, i64 %j182 store i64 %add, ptr %gep183 %add3 = add nsw i64 %i, 6184 %gep5 = getelementptr inbounds i64, ptr %array, i64 %add3185 store i64 %i, ptr %gep5186 %add6 = add nsw i64 %i, 35187 %gep8 = getelementptr inbounds i64, ptr %array, i64 %i188 store i64 %i, ptr %gep8189 ret i64 undef190}191 192; No need to optimize GEPs, because the base variable is different.193define i32 @test7(ptr %array, i32 %i, i32 %j, i32 %k) {194; CHECK-LABEL: @test7(195; CHECK-NEXT: entry:196; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[I:%.*]], 5197; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[I]] to i64198; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[ARRAY:%.*]], i64 [[TMP0]]199; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[TMP1]], i64 20200; CHECK-NEXT: store i32 [[ADD]], ptr [[GEP2]], align 4201; CHECK-NEXT: [[TMP2:%.*]] = sext i32 [[K:%.*]] to i64202; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i32, ptr [[ARRAY]], i64 [[TMP2]]203; CHECK-NEXT: [[GEP54:%.*]] = getelementptr i8, ptr [[TMP3]], i64 24204; CHECK-NEXT: store i32 [[I]], ptr [[GEP54]], align 4205; CHECK-NEXT: [[TMP4:%.*]] = sext i32 [[J:%.*]] to i64206; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i32, ptr [[ARRAY]], i64 [[TMP4]]207; CHECK-NEXT: [[GEP86:%.*]] = getelementptr i8, ptr [[TMP5]], i64 140208; CHECK-NEXT: store i32 [[I]], ptr [[GEP86]], align 4209; CHECK-NEXT: ret i32 undef210;211entry:212 %add = add nsw i32 %i, 5213 %sext = sext i32 %add to i64214 %gep = getelementptr inbounds i32, ptr %array, i64 %sext215 store i32 %add, ptr %gep216 %add3 = add nsw i32 %k, 6217 %sext4 = sext i32 %add3 to i64218 %gep5 = getelementptr inbounds i32, ptr %array, i64 %sext4219 store i32 %i, ptr %gep5220 %add6 = add nsw i32 %j, 35221 %sext7 = sext i32 %add6 to i64222 %gep8 = getelementptr inbounds i32, ptr %array, i64 %sext7223 store i32 %i, ptr %gep8224 ret i32 undef225}226 227; No need to optimize GEPs, because the base of GEP instructions is different.228define i32 @test8(ptr %array, ptr %array2, ptr %array3, i32 %i) {229; CHECK-LABEL: @test8(230; CHECK-NEXT: entry:231; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[I:%.*]], 5232; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[I]] to i64233; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[ARRAY:%.*]], i64 [[TMP0]]234; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[TMP1]], i64 20235; CHECK-NEXT: store i32 [[ADD]], ptr [[GEP2]], align 4236; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[ARRAY2:%.*]], i64 [[TMP0]]237; CHECK-NEXT: [[GEP54:%.*]] = getelementptr i8, ptr [[TMP2]], i64 24238; CHECK-NEXT: store i32 [[I]], ptr [[GEP54]], align 4239; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i32, ptr [[ARRAY3:%.*]], i64 [[TMP0]]240; CHECK-NEXT: [[GEP86:%.*]] = getelementptr i8, ptr [[TMP3]], i64 140241; CHECK-NEXT: store i32 [[I]], ptr [[GEP86]], align 4242; CHECK-NEXT: ret i32 undef243;244entry:245 %add = add nsw i32 %i, 5246 %sext = sext i32 %add to i64247 %gep = getelementptr inbounds i32, ptr %array, i64 %sext248 store i32 %add, ptr %gep249 %add3 = add nsw i32 %i, 6250 %sext4 = sext i32 %add3 to i64251 %gep5 = getelementptr inbounds i32, ptr %array2, i64 %sext4252 store i32 %i, ptr %gep5253 %add6 = add nsw i32 %i, 35254 %sext7 = sext i32 %add6 to i64255 %gep8 = getelementptr inbounds i32, ptr %array3, i64 %sext7256 store i32 %i, ptr %gep8257 ret i32 undef258}259 260; No need to optimize GEPs of multidimensional array, because the base of GEP instructions is different.261define i32 @test9(ptr %array, i32 %i) {262; CHECK-LABEL: @test9(263; CHECK-NEXT: entry:264; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[I:%.*]], 5265; CHECK-NEXT: [[TMP0:%.*]] = sext i32 [[I]] to i64266; CHECK-NEXT: [[TMP1:%.*]] = getelementptr [50 x i32], ptr [[ARRAY:%.*]], i64 0, i64 [[TMP0]]267; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[TMP1]], i64 20268; CHECK-NEXT: store i32 [[ADD]], ptr [[GEP2]], align 4269; CHECK-NEXT: [[TMP2:%.*]] = getelementptr [50 x i32], ptr [[ARRAY]], i64 [[TMP0]], i64 [[TMP0]]270; CHECK-NEXT: [[GEP54:%.*]] = getelementptr i8, ptr [[TMP2]], i64 24271; CHECK-NEXT: store i32 [[I]], ptr [[GEP54]], align 4272; CHECK-NEXT: [[GEP87:%.*]] = getelementptr i8, ptr [[TMP2]], i64 1340273; CHECK-NEXT: store i32 [[I]], ptr [[GEP87]], align 4274; CHECK-NEXT: ret i32 undef275;276entry:277 %add = add nsw i32 %i, 5278 %sext = sext i32 %add to i64279 %gep = getelementptr inbounds [50 x i32], ptr %array, i64 0, i64 %sext280 store i32 %add, ptr %gep281 %add3 = add nsw i32 %i, 6282 %sext4 = sext i32 %add3 to i64283 %int = sext i32 %i to i64284 %gep5 = getelementptr inbounds [50 x i32], ptr %array, i64 %int, i64 %sext4285 store i32 %i, ptr %gep5286 %add6 = add nsw i32 %i, 35287 %sext7 = sext i32 %add6 to i64288 %gep8 = getelementptr inbounds [50 x i32], ptr %array, i64 %sext4, i64 %sext7289 store i32 %i, ptr %gep8290 ret i32 undef291}292