113 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<import-jscop;simplify-0>' -polly-print-simplify -polly-import-jscop-postfix=transformed -disable-output < %s | FileCheck %s2;3; void gemm(float A[][1024], float B[][1024], float C[][1024]) {4; for (long i = 0; i < 1024; i++)5; for (long j = 0; j < 1024; j++) {6; float tmp = C[i][j];7; for (long k = 0; k < 1024; k++)8; tmp += A[i][k] * B[k][j];9; C[i][j] = tmp;10; }11; }12 13; CHECK: After accesses {14; CHECK-NEXT: Stmt_bb1315; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 1]16; CHECK-NEXT: { Stmt_bb13[i0, i1, i2] -> MemRef_tmp_0__phi[] };17; CHECK-NEXT: new: { Stmt_bb13[i0, i1, i2] -> MemRef_C[i0, i1] };18; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]19; CHECK-NEXT: { Stmt_bb13[i0, i1, i2] -> MemRef_A[i0, i2] };20; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]21; CHECK-NEXT: { Stmt_bb13[i0, i1, i2] -> MemRef_B[i2, i1] };22; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 1]23; CHECK-NEXT: { Stmt_bb13[i0, i1, i2] -> MemRef_tmp_0[] };24; CHECK-NEXT: new: { Stmt_bb13[i0, i1, i2] -> MemRef_C[i0, i1] };25; CHECK-NEXT: }26 27target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"28target triple = "x86_64-pc-linux-gnu"29 30define void @gemm(ptr %A, ptr %B, ptr %C) {31bb:32 br label %bb333 34bb3: ; preds = %bb26, %bb35 %i.0 = phi i64 [ 0, %bb ], [ %tmp27, %bb26 ]36 %exitcond2 = icmp ne i64 %i.0, 102437 br i1 %exitcond2, label %bb5, label %bb438 39bb4: ; preds = %bb340 br label %bb2841 42bb5: ; preds = %bb343 br label %bb644 45bb6: ; preds = %bb23, %bb546 %j.0 = phi i64 [ 0, %bb5 ], [ %tmp24, %bb23 ]47 %exitcond1 = icmp ne i64 %j.0, 102448 br i1 %exitcond1, label %bb8, label %bb749 50bb7: ; preds = %bb651 br label %bb2552 53bb8: ; preds = %bb654 %tmp = getelementptr inbounds [1024 x float], ptr %C, i64 %i.0, i64 %j.055 %tmp9 = load float, ptr %tmp, align 4, !tbaa !156 br label %bb1057 58bb10: ; preds = %bb13, %bb859 %tmp.0 = phi float [ %tmp9, %bb8 ], [ %tmp19, %bb13 ]60 %k.0 = phi i64 [ 0, %bb8 ], [ %tmp20, %bb13 ]61 %exitcond = icmp ne i64 %k.0, 102462 br i1 %exitcond, label %bb12, label %bb1163 64bb11: ; preds = %bb1065 %tmp.0.lcssa = phi float [ %tmp.0, %bb10 ]66 br label %bb2167 68bb12: ; preds = %bb1069 br label %bb1370 71bb13: ; preds = %bb1272 %tmp14 = getelementptr inbounds [1024 x float], ptr %A, i64 %i.0, i64 %k.073 %tmp15 = load float, ptr %tmp14, align 4, !tbaa !174 %tmp16 = getelementptr inbounds [1024 x float], ptr %B, i64 %k.0, i64 %j.075 %tmp17 = load float, ptr %tmp16, align 4, !tbaa !176 %tmp18 = fmul float %tmp15, %tmp1777 %tmp19 = fadd float %tmp.0, %tmp1878 %tmp20 = add nuw nsw i64 %k.0, 179 br label %bb1080 81bb21: ; preds = %bb1182 %tmp22 = getelementptr inbounds [1024 x float], ptr %C, i64 %i.0, i64 %j.083 store float %tmp.0.lcssa, ptr %tmp22, align 4, !tbaa !184 br label %bb2385 86bb23: ; preds = %bb2187 %tmp24 = add nuw nsw i64 %j.0, 188 br label %bb689 90bb25: ; preds = %bb791 br label %bb2692 93bb26: ; preds = %bb2594 %tmp27 = add nuw nsw i64 %i.0, 195 br label %bb396 97bb28: ; preds = %bb498 ret void99}100 101declare void @llvm.lifetime.start(i64, ptr nocapture)102 103declare void @llvm.lifetime.end(i64, ptr nocapture)104 105 106!llvm.ident = !{!0}107 108!0 = !{!"Ubuntu clang version 3.7.1-3ubuntu4 (tags/RELEASE_371/final) (based on LLVM 3.7.1)"}109!1 = !{!2, !2, i64 0}110!2 = !{!"float", !3, i64 0}111!3 = !{!"omnipotent char", !4, i64 0}112!4 = !{!"Simple C/C++ TBAA"}113