1; RUN: opt %loadNPMPolly -polly-parallel -polly-parallel-force '-passes=print<polly-ast>' -disable-output < %s | FileCheck %s -check-prefix=AST 2; RUN: opt %loadNPMPolly -polly-parallel -polly-parallel-force -passes=polly-codegen -S -verify-dom-info < %s | FileCheck %s -check-prefix=IR 3 4; RUN: opt %loadNPMPolly -polly-parallel -polly-parallel-force '-passes=polly-import-jscop,print<polly-ast>' -disable-output < %s | FileCheck %s -check-prefix=AST-STRIDE4 5; RUN: opt %loadNPMPolly -polly-parallel -polly-parallel-force '-passes=polly-import-jscop,polly-codegen' -S < %s | FileCheck %s -check-prefix=IR-STRIDE4 6 7; RUN: opt %loadNPMPolly -polly-parallel -polly-parallel-force -passes=polly-codegen -polly-omp-backend=LLVM -polly-scheduling=static -polly-scheduling-chunksize=43 -S -verify-dom-info < %s | FileCheck %s -check-prefix=LIBOMP-IR-STATIC-CHUNKED 8; RUN: opt %loadNPMPolly -polly-parallel -polly-parallel-force -passes=polly-codegen -polly-omp-backend=LLVM -polly-scheduling=static -S -verify-dom-info < %s | FileCheck %s -check-prefix=LIBOMP-IR-STATIC 9; RUN: opt %loadNPMPolly -polly-parallel -polly-parallel-force -passes=polly-codegen -polly-omp-backend=LLVM -polly-scheduling=dynamic -S -verify-dom-info < %s | FileCheck %s -check-prefix=LIBOMP-IR-DYNAMIC 10; RUN: opt %loadNPMPolly -polly-parallel -polly-parallel-force -passes=polly-codegen -polly-omp-backend=LLVM -polly-scheduling=dynamic -polly-scheduling-chunksize=4 -S -verify-dom-info < %s | FileCheck %s -check-prefix=LIBOMP-IR-DYNAMIC-FOUR 11; RUN: opt %loadNPMPolly -polly-parallel -polly-parallel-force '-passes=polly-import-jscop,polly-codegen' -polly-omp-backend=LLVM -S < %s | FileCheck %s -check-prefix=LIBOMP-IR-STRIDE4 12 13; This extensive test case tests the creation of the full set of OpenMP calls 14; as well as the subfunction creation using a trivial loop as example. 15; 16; #define N 1024 17; float A[N]; 18; 19; void single_parallel_loop(void) { 20; for (long i = 0; i < N; i++) 21; A[i] = 1; 22; } 23 24; AST: #pragma simd 25; AST: #pragma omp parallel for 26; AST: for (int c0 = 0; c0 <= 1023; c0 += 1) 27; AST: Stmt_S(c0); 28 29; AST-STRIDE4: #pragma omp parallel for 30; AST-STRIDE4: for (int c0 = 0; c0 <= 1023; c0 += 4) 31; AST-STRIDE4: #pragma simd 32; AST-STRIDE4: for (int c1 = c0; c1 <= c0 + 3; c1 += 1) 33; AST-STRIDE4: Stmt_S(c1); 34 35; IR-LABEL: single_parallel_loop() 36; IR-NEXT: entry 37; IR-NEXT: %polly.par.userContext = alloca 38 39; IR-LABEL: polly.parallel.for: 40; IR-NEXT: call void @GOMP_parallel_loop_runtime_start(ptr @single_parallel_loop_polly_subfn, ptr %polly.par.userContext, i32 0, i64 0, i64 1024, i64 1) 41; IR-NEXT: call void @single_parallel_loop_polly_subfn(ptr %polly.par.userContext) 42; IR-NEXT: call void @GOMP_parallel_end() 43; IR-NEXT: br label %polly.exiting 44 45; IR: define internal void @single_parallel_loop_polly_subfn(ptr %polly.par.userContext) #1 46; IR-LABEL: polly.par.setup: 47; IR-NEXT: %polly.par.LBPtr = alloca i64 48; IR-NEXT: %polly.par.UBPtr = alloca i64 49; IR: br label %polly.par.checkNext 50 51; IR-LABEL: polly.par.exit: 52; IR-NEXT: call void @GOMP_loop_end_nowait() 53; IR-NEXT: ret void 54 55; IR-LABEL: polly.par.checkNext: 56; IR-NEXT: %[[parnext:[._a-zA-Z0-9]*]] = call i8 @GOMP_loop_runtime_next(ptr %polly.par.LBPtr, ptr %polly.par.UBPtr) 57; IR-NEXT: %[[cmp:[._a-zA-Z0-9]*]] = icmp ne i8 %[[parnext]], 0 58; IR-NEXT: br i1 %[[cmp]], label %polly.par.loadIVBounds, label %polly.par.exit 59 60; IR-LABEL: polly.par.loadIVBounds: 61; IR-NEXT: %polly.par.LB = load i64, ptr %polly.par.LBPtr 62; IR-NEXT: %polly.par.UB = load i64, ptr %polly.par.UBPtr 63; IR-NEXT: %polly.par.UBAdjusted = sub i64 %polly.par.UB, 1 64; IR-NEXT: br label %polly.loop_preheader 65 66; IR-LABEL: polly.loop_exit: 67; IR-NEXT: br label %polly.par.checkNext 68 69; IR-LABEL: polly.loop_header: 70; IR-NEXT: %polly.indvar = phi i64 [ %polly.par.LB, %polly.loop_preheader ], [ %polly.indvar_next, %polly.stmt.S ] 71; IR-NEXT: br label %polly.stmt.S 72 73; IR-LABEL: polly.stmt.S: 74; IR-NEXT: %[[offset:[0-9]+]] = shl i64 %polly.indvar, 2 75; IR-NEXT: %[[gep:[._a-zA-Z0-9]*]] = getelementptr i8, ptr {{.*}}, i64 %[[offset]] 76; IR-NEXT: store float 1.000000e+00, ptr %[[gep]] 77; IR-NEXT: %polly.indvar_next = add nsw i64 %polly.indvar, 1 78; IR-NEXT: %polly.loop_cond = icmp sle i64 %polly.indvar_next, %polly.par.UBAdjusted 79; IR-NEXT: br i1 %polly.loop_cond, label %polly.loop_header, label %polly.loop_exit 80 81; IR-LABEL: polly.loop_preheader: 82; IR-NEXT: br label %polly.loop_header 83 84; IR: attributes #1 = { "polly.skip.fn" } 85 86; IR-STRIDE4: call void @GOMP_parallel_loop_runtime_start(ptr @single_parallel_loop_polly_subfn, ptr %polly.par.userContext, i32 0, i64 0, i64 1024, i64 4) 87; IR-STRIDE4: add nsw i64 %polly.indvar, 3 88; IR-STRIDE4: %polly.indvar_next = add nsw i64 %polly.indvar, 4 89; IR-STRIDE4 %polly.adjust_ub = sub i64 %polly.par.UBAdjusted, 4 90 91; LIBOMP-IR-STATIC-CHUNKED: %struct.ident_t = type { i32, i32, i32, i32, ptr } 92 93; LIBOMP-IR-STATIC-CHUNKED-LABEL: single_parallel_loop() 94; LIBOMP-IR-STATIC-CHUNKED-NEXT: entry 95; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.par.userContext = alloca 96 97; LIBOMP-IR-STATIC-CHUNKED-LABEL: polly.parallel.for: 98; LIBOMP-IR-STATIC-CHUNKED-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @.loc.dummy, i32 4, ptr @single_parallel_loop_polly_subfn, i64 0, i64 1024, i64 1, ptr %polly.par.userContext) 99; LIBOMP-IR-STATIC-CHUNKED-NEXT: br label %polly.exiting 100 101; LIBOMP-IR-STATIC-CHUNKED: define internal void @single_parallel_loop_polly_subfn(ptr %polly.kmpc.global_tid, ptr %polly.kmpc.bound_tid, i64 %polly.kmpc.lb, i64 %polly.kmpc.ub, i64 %polly.kmpc.inc, ptr %polly.kmpc.shared) 102; LIBOMP-IR-STATIC-CHUNKED-LABEL: polly.par.setup: 103; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.par.LBPtr = alloca i64 104; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.par.UBPtr = alloca i64 105; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.par.lastIterPtr = alloca i32 106; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.par.StridePtr = alloca i64 107; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.par.global_tid = load i32, ptr %polly.kmpc.global_tid 108; LIBOMP-IR-STATIC-CHUNKED-NEXT: store i64 %polly.kmpc.lb, ptr %polly.par.LBPtr 109; LIBOMP-IR-STATIC-CHUNKED-NEXT: store i64 %polly.kmpc.ub, ptr %polly.par.UBPtr 110; LIBOMP-IR-STATIC-CHUNKED-NEXT: store i32 0, ptr %polly.par.lastIterPtr 111; LIBOMP-IR-STATIC-CHUNKED-NEXT: store i64 %polly.kmpc.inc, ptr %polly.par.StridePtr 112; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.UBAdjusted = add i64 %polly.kmpc.ub, -1 113; LIBOMP-IR-STATIC-CHUNKED-NEXT: store i64 %polly.indvar.UBAdjusted, ptr %polly.par.UBPtr, align 8 114; LIBOMP-IR-STATIC-CHUNKED-NEXT: call void @__kmpc_for_static_init_{{[4|8]}}(ptr @.loc.dummy{{[.0-9]*}}, i32 %polly.par.global_tid, i32 33, ptr %polly.par.lastIterPtr, ptr %polly.par.LBPtr, ptr %polly.par.UBPtr, ptr %polly.par.StridePtr, i64 1, i64 43) 115; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.kmpc.stride = load i64, ptr %polly.par.StridePtr, align 8 116; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.LB = load i64, ptr %polly.par.LBPtr 117; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.UB.temp = load i64, ptr %polly.par.UBPtr 118; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.UB.inRange = icmp sle i64 %polly.indvar.UB.temp, %polly.indvar.UBAdjusted 119; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.UB = select i1 %polly.indvar.UB.inRange, i64 %polly.indvar.UB.temp, i64 %polly.indvar.UBAdjusted 120; LIBOMP-IR-STATIC-CHUNKED-NEXT: store i64 %polly.indvar.UB, ptr %polly.par.UBPtr, align 8 121; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.hasIteration = icmp sle i64 %polly.indvar.LB, %polly.indvar.UB 122; LIBOMP-IR-STATIC-CHUNKED: br i1 %polly.hasIteration, label %polly.par.loadIVBounds, label %polly.par.exit 123 124; LIBOMP-IR-STATIC-CHUNKED-LABEL: polly.par.exit: 125; LIBOMP-IR-STATIC-CHUNKED-NEXT: call void @__kmpc_for_static_fini(ptr @.loc.dummy, i32 %polly.par.global_tid) 126; LIBOMP-IR-STATIC-CHUNKED-NEXT: ret void 127 128; LIBOMP-IR-STATIC-CHUNKED-LABEL: polly.par.checkNext: 129; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.nextLB = add i64 %polly.indvar.LB.entry, %polly.kmpc.stride 130; LIBOMP-IR-STATIC-CHUNKED-NEXT: %{{[0-9]+}} = add i64 %polly.indvar.UB.entry, %polly.kmpc.stride 131; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.nextUB.outOfBounds = icmp sgt i64 %{{[0-9]+}}, %polly.indvar.UBAdjusted 132; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.nextUB = select i1 %polly.indvar.nextUB.outOfBounds, i64 %polly.indvar.UBAdjusted, i64 %{{[0-9]+}} 133; LIBOMP-IR-STATIC-CHUNKED-NEXT: store i64 %polly.indvar.nextLB, ptr %polly.par.LBPtr, align 8 134; LIBOMP-IR-STATIC-CHUNKED-NEXT: store i64 %polly.indvar.nextUB, ptr %polly.par.UBPtr, align 8 135; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.hasWork = icmp sle i64 %polly.indvar.nextLB, %polly.indvar.UBAdjusted 136; LIBOMP-IR-STATIC-CHUNKED-NEXT: br i1 %polly.hasWork, label %polly.par.loadIVBounds, label %polly.par.exit 137 138; LIBOMP-IR-STATIC-CHUNKED-LABEL: polly.par.loadIVBounds: 139; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.LB.entry = load i64, ptr %polly.par.LBPtr, align 8 140; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar.UB.entry = load i64, ptr %polly.par.UBPtr, align 8 141; LIBOMP-IR-STATIC-CHUNKED-NEXT: br label %polly.loop_preheader 142 143; LIBOMP-IR-STATIC-CHUNKED-LABEL: polly.loop_exit: 144; LIBOMP-IR-STATIC-CHUNKED-NEXT: br label %polly.par.checkNext 145 146; LIBOMP-IR-STATIC-CHUNKED-LABEL: polly.loop_header: 147; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar = phi i64 [ %polly.indvar.LB.entry, %polly.loop_preheader ], [ %polly.indvar_next, %polly.stmt.S ] 148; LIBOMP-IR-STATIC-CHUNKED-NEXT: br label %polly.stmt.S 149 150; LIBOMP-IR-STATIC-CHUNKED-LABEL: polly.stmt.S: 151; LIBOMP-IR-STATIC-CHUNKED-NEXT: %[[offset:[0-9]+]] = shl i64 %polly.indvar, 2 152; LIBOMP-IR-STATIC-CHUNKED-NEXT: %[[gep:[._a-zA-Z0-9]*]] = getelementptr i8, ptr {{.*}}, i64 %[[offset]] 153; LIBOMP-IR-STATIC-CHUNKED-NEXT: store float 1.000000e+00, ptr %[[gep]] 154; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.indvar_next = add nsw i64 %polly.indvar, 1 155; LIBOMP-IR-STATIC-CHUNKED-NEXT: %polly.loop_cond = icmp sle i64 %polly.indvar_next, %polly.indvar.UB.entry 156; LIBOMP-IR-STATIC-CHUNKED-NEXT: br i1 %polly.loop_cond, label %polly.loop_header, label %polly.loop_exit 157 158; LIBOMP-IR-STATIC-CHUNKED-LABEL: polly.loop_preheader: 159; LIBOMP-IR-STATIC-CHUNKED-NEXT: br label %polly.loop_header 160 161; LIBOMP-IR-STATIC-CHUNKED: attributes #1 = { "polly.skip.fn" } 162 163; LIBOMP-IR-STATIC: define internal void @single_parallel_loop_polly_subfn(ptr %polly.kmpc.global_tid, ptr %polly.kmpc.bound_tid, i64 %polly.kmpc.lb, i64 %polly.kmpc.ub, i64 %polly.kmpc.inc, ptr %polly.kmpc.shared) 164; LIBOMP-IR-STATIC-LABEL: polly.par.setup: 165; LIBOMP-IR-STATIC: call void @__kmpc_for_static_init_{{[4|8]}}(ptr @.loc.dummy{{[.0-9]*}}, i32 %polly.par.global_tid, i32 34, ptr %polly.par.lastIterPtr, ptr %polly.par.LBPtr, ptr %polly.par.UBPtr, ptr %polly.par.StridePtr, i64 1, i64 1) 166; LIBOMP-IR-STATIC: br i1 %polly.hasIteration, label %polly.par.loadIVBounds, label %polly.par.exit 167 168; LIBOMP-IR-STATIC-LABEL: polly.par.exit: 169; LIBOMP-IR-STATIC-NEXT: call void @__kmpc_for_static_fini(ptr @.loc.dummy, i32 %polly.par.global_tid) 170; LIBOMP-IR-STATIC-NEXT: ret void 171 172; LIBOMP-IR-STATIC-LABEL: polly.par.checkNext: 173; LIBOMP-IR-STATIC-NEXT: br label %polly.par.exit 174 175; LIBOMP-IR-STATIC-LABEL: polly.par.loadIVBounds: 176; LIBOMP-IR-STATIC-NEXT: br label %polly.loop_preheader 177 178; LIBOMP-IR-STATIC-LABEL: polly.loop_exit: 179; LIBOMP-IR-STATIC-NEXT: br label %polly.par.checkNext 180 181; LIBOMP-IR-STATIC-LABEL: polly.loop_header: 182; LIBOMP-IR-STATIC-NEXT: %polly.indvar = phi i64 [ %polly.indvar.LB, %polly.loop_preheader ], [ %polly.indvar_next, %polly.stmt.S ] 183; LIBOMP-IR-STATIC-NEXT: br label %polly.stmt.S 184 185; LIBOMP-IR-STATIC-LABEL: polly.stmt.S: 186; LIBOMP-IR-STATIC-NEXT: %[[offset:[0-9]+]] = shl i64 %polly.indvar, 2 187; LIBOMP-IR-STATIC-NEXT: %[[gep:[._a-zA-Z0-9]*]] = getelementptr i8, ptr {{.*}}, i64 %[[offset]] 188; LIBOMP-IR-STATIC-NEXT: store float 1.000000e+00, ptr %[[gep]] 189; LIBOMP-IR-STATIC-NEXT: %polly.indvar_next = add nsw i64 %polly.indvar, 1 190; LIBOMP-IR-STATIC-NEXT: %polly.loop_cond = icmp sle i64 %polly.indvar_next, %polly.indvar.UB 191; LIBOMP-IR-STATIC-NEXT: br i1 %polly.loop_cond, label %polly.loop_header, label %polly.loop_exit 192 193; LIBOMP-IR-STATIC-LABEL: polly.loop_preheader: 194; LIBOMP-IR-STATIC-NEXT: br label %polly.loop_header 195 196; LIBOMP-IR-DYNAMIC: call void @__kmpc_dispatch_init_{{[4|8]}}(ptr @.loc.dummy, i32 %polly.par.global_tid, i32 35, i64 %polly.kmpc.lb, i64 %polly.indvar.UBAdjusted, i64 %polly.kmpc.inc, i64 1) 197; LIBOMP-IR-DYNAMIC-NEXT: %{{[0-9]+}} = call i32 @__kmpc_dispatch_next_{{[4|8]}}(ptr @.loc.dummy, i32 %polly.par.global_tid, ptr %polly.par.lastIterPtr, ptr %polly.par.LBPtr, ptr %polly.par.UBPtr, ptr %polly.par.StridePtr) 198; LIBOMP-IR-DYNAMIC-NEXT: %polly.hasIteration = icmp eq i32 %{{[0-9]+}}, 1 199; LIBOMP-IR-DYNAMIC-NEXT: br i1 %polly.hasIteration, label %polly.par.loadIVBounds, label %polly.par.exit 200 201; LIBOMP-IR-DYNAMIC-LABEL: polly.par.exit: 202; LIBOMP-IR-DYNAMIC-NEXT: ret void 203 204; LIBOMP-IR-DYNAMIC-LABEL: polly.par.checkNext: 205; LIBOMP-IR-DYNAMIC-NEXT: %{{[0-9]+}} = call i32 @__kmpc_dispatch_next_{{[4|8]}}(ptr @.loc.dummy, i32 %polly.par.global_tid, ptr %polly.par.lastIterPtr, ptr %polly.par.LBPtr, ptr %polly.par.UBPtr, ptr %polly.par.StridePtr) 206; LIBOMP-IR-DYNAMIC-NEXT: %polly.hasWork = icmp eq i32 %{{[0-9]+}}, 1 207; LIBOMP-IR-DYNAMIC-NEXT: br i1 %polly.hasWork, label %polly.par.loadIVBounds, label %polly.par.exit 208 209; LIBOMP-IR-DYNAMIC-LABEL: polly.par.loadIVBounds: 210; LIBOMP-IR-DYNAMIC-NEXT: %polly.indvar.LB = load i64, ptr %polly.par.LBPtr 211; LIBOMP-IR-DYNAMIC-NEXT: %polly.indvar.UB = load i64, ptr %polly.par.UBPtr 212; LIBOMP-IR-DYNAMIC-NEXT: br label %polly.loop_preheader 213 214; LIBOMP-IR-DYNAMIC-FOUR: call void @__kmpc_dispatch_init_{{[4|8]}}(ptr @.loc.dummy, i32 %polly.par.global_tid, i32 35, i64 %polly.kmpc.lb, i64 %polly.indvar.UBAdjusted, i64 %polly.kmpc.inc, i64 4) 215 216; LIBOMP-IR-STRIDE4: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @.loc.dummy, i32 4, ptr @single_parallel_loop_polly_subfn, i64 0, i64 1024, i64 4, ptr %polly.par.userContext) 217 218target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64" 219 220@A = common global [1024 x float] zeroinitializer, align 16 221 222define void @single_parallel_loop() nounwind { 223entry: 224 br label %for.i 225 226for.i: 227 %indvar = phi i64 [ %indvar.next, %for.inc], [ 0, %entry ] 228 %scevgep = getelementptr [1024 x float], ptr @A, i64 0, i64 %indvar 229 %exitcond = icmp ne i64 %indvar, 1024 230 br i1 %exitcond, label %S, label %exit 231 232S: 233 store float 1.0, ptr %scevgep 234 br label %for.inc 235 236for.inc: 237 %indvar.next = add i64 %indvar, 1 238 br label %for.i 239 240exit: 241 ret void 242} 243