brintos

brintos / llvm-project-archived public Read only

0
0
Text · 4.4 KiB · 6d1592c Raw
99 lines · plain
1; RUN: opt -S %loadNPMPolly -aa-pipeline=basic-aa -polly-pattern-matching-based-opts=false -polly-vectorizer=stripmine '-passes=polly-custom<opt-isl;ast>' -polly-print-ast -disable-output < %s | FileCheck %s2; RUN: opt -S %loadNPMPolly -aa-pipeline=basic-aa -polly-pattern-matching-based-opts=false -polly-vectorizer=stripmine -polly-prevect-width=16 '-passes=polly-custom<opt-isl;ast>' -polly-print-ast -disable-output < %s | FileCheck %s -check-prefix=VEC163 4target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"5 6@C = common global [1536 x [1536 x float]] zeroinitializer, align 167@A = common global [1536 x [1536 x float]] zeroinitializer, align 168@B = common global [1536 x [1536 x float]] zeroinitializer, align 169 10; Function Attrs: nounwind uwtable11define void @foo() {12entry:13  br label %entry.split14 15entry.split:                                      ; preds = %entry16  br label %for.cond1.preheader17 18for.cond1.preheader:                              ; preds = %entry.split, %for.inc2819  %indvar4 = phi i64 [ 0, %entry.split ], [ %indvar.next5, %for.inc28 ]20  br label %for.body321 22for.body3:                                        ; preds = %for.cond1.preheader, %for.inc2523  %indvar6 = phi i64 [ 0, %for.cond1.preheader ], [ %indvar.next7, %for.inc25 ]24  %arrayidx24 = getelementptr [1536 x [1536 x float]], ptr @C, i64 0, i64 %indvar4, i64 %indvar625  store float 0.000000e+00, ptr %arrayidx24, align 426  br label %for.body827 28for.body8:                                        ; preds = %for.body3, %for.body829  %indvar = phi i64 [ 0, %for.body3 ], [ %indvar.next, %for.body8 ]30  %arrayidx16 = getelementptr [1536 x [1536 x float]], ptr @A, i64 0, i64 %indvar4, i64 %indvar31  %arrayidx20 = getelementptr [1536 x [1536 x float]], ptr @B, i64 0, i64 %indvar, i64 %indvar632  %0 = load float, ptr %arrayidx24, align 433  %1 = load float, ptr %arrayidx16, align 434  %2 = load float, ptr %arrayidx20, align 435  %mul = fmul float %1, %236  %add = fadd float %0, %mul37  store float %add, ptr %arrayidx24, align 438  %indvar.next = add i64 %indvar, 139  %exitcond = icmp ne i64 %indvar.next, 153640  br i1 %exitcond, label %for.body8, label %for.inc2541 42for.inc25:                                        ; preds = %for.body843  %indvar.next7 = add i64 %indvar6, 144  %exitcond8 = icmp ne i64 %indvar.next7, 153645  br i1 %exitcond8, label %for.body3, label %for.inc2846 47for.inc28:                                        ; preds = %for.inc2548  %indvar.next5 = add i64 %indvar4, 149  %exitcond9 = icmp ne i64 %indvar.next5, 153650  br i1 %exitcond9, label %for.cond1.preheader, label %for.end3051 52for.end30:                                        ; preds = %for.inc2853  ret void54}55 56; CHECK: #pragma known-parallel57; CHECK: for (int c0 = 0; c0 <= 47; c0 += 1)58; CHECK:   for (int c1 = 0; c1 <= 47; c1 += 1)59; CHECK:     for (int c2 = 0; c2 <= 31; c2 += 1)60; CHECK:       for (int c3 = 0; c3 <= 7; c3 += 1)61; CHECK:         // SIMD62; CHECK:         for (int c4 = 0; c4 <= 3; c4 += 1)63; CHECK:           Stmt_for_body3(32 * c0 + c2, 32 * c1 + 4 * c3 + c4);64; CHECK: #pragma known-parallel65; CHECK: for (int c0 = 0; c0 <= 47; c0 += 1)66; CHECK:   for (int c1 = 0; c1 <= 47; c1 += 1)67; CHECK:     for (int c2 = 0; c2 <= 47; c2 += 1)68; CHECK:       for (int c3 = 0; c3 <= 31; c3 += 1)69; CHECK:         for (int c4 = 0; c4 <= 7; c4 += 1)70; CHECK:           for (int c5 = 0; c5 <= 31; c5 += 1)71; CHECK:             // SIMD72; CHECK:             for (int c6 = 0; c6 <= 3; c6 += 1)73; CHECK:               Stmt_for_body8(32 * c0 + c3, 32 * c1 + 4 * c4 + c6, 32 * c2 + c5);74 75; VEC16: {76; VEC16:   #pragma known-parallel77; VEC16:   for (int c0 = 0; c0 <= 47; c0 += 1)78; VEC16:     for (int c1 = 0; c1 <= 47; c1 += 1)79; VEC16:       for (int c2 = 0; c2 <= 31; c2 += 1)80; VEC16:         for (int c3 = 0; c3 <= 1; c3 += 1)81; VEC16:           // SIMD82; VEC16:           for (int c4 = 0; c4 <= 15; c4 += 1)83; VEC16:             Stmt_for_body3(32 * c0 + c2, 32 * c1 + 16 * c3 + c4);84; VEC16:   #pragma known-parallel85; VEC16:   for (int c0 = 0; c0 <= 47; c0 += 1)86; VEC16:     for (int c1 = 0; c1 <= 47; c1 += 1)87; VEC16:       for (int c2 = 0; c2 <= 47; c2 += 1)88; VEC16:         for (int c3 = 0; c3 <= 31; c3 += 1)89; VEC16:           for (int c4 = 0; c4 <= 1; c4 += 1)90; VEC16:             for (int c5 = 0; c5 <= 31; c5 += 1)91; VEC16:               // SIMD92; VEC16:               for (int c6 = 0; c6 <= 15; c6 += 1)93; VEC16:                 Stmt_for_body8(32 * c0 + c3, 32 * c1 + 16 * c4 + c6, 32 * c2 + c5);94; VEC16: }95 96!llvm.ident = !{!0}97 98!0 = !{!"clang version 3.5.0 "}99