55 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<import-jscop;opt-isl;ast;codegen>' -polly-target-throughput-vector-fma=1 -polly-target-latency-vector-fma=8 -polly-target-1st-cache-level-associativity=8 -polly-target-2nd-cache-level-associativity=8 -polly-target-1st-cache-level-size=32768 -polly-target-vector-register-bitwidth=256 -polly-target-2nd-cache-level-size=262144 -polly-import-jscop-postfix=transformed -S < %s | FileCheck %s2;3; Check that we disable the Loop Vectorizer.4;5; CHECK: !{!"llvm.loop.vectorize.enable", i1 false}6;7target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"8target triple = "x86_64-unknown-unknown"9 10define void @kernel_gemm(i32 %ni, i32 %nj, i32 %nk, ptr %A, ptr %B, ptr %C, ptr %C1) {11entry:12 br label %entry.split13 14entry.split: ; preds = %entry15 br label %for.body16 17for.body: ; preds = %for.inc22, %entry.split18 %indvars.iv43 = phi i64 [ 0, %entry.split ], [ %indvars.iv.next44, %for.inc22 ]19 br label %for.body320 21for.body3: ; preds = %for.inc19, %for.body22 %indvars.iv40 = phi i64 [ 0, %for.body ], [ %indvars.iv.next41, %for.inc19 ]23 br label %for.body624 25for.body6: ; preds = %for.body6, %for.body326 %indvars.iv = phi i64 [ 0, %for.body3 ], [ %indvars.iv.next, %for.body6 ]27 %tmp = load double, ptr %C1, align 828 %arrayidx9 = getelementptr inbounds [1024 x double], ptr %A, i64 %indvars.iv43, i64 %indvars.iv29 %tmp1 = load double, ptr %arrayidx9, align 830 %arrayidx13 = getelementptr inbounds [1024 x double], ptr %B, i64 %indvars.iv, i64 %indvars.iv4031 %tmp2 = load double, ptr %arrayidx13, align 832 %mul = fmul double %tmp1, %tmp233 %add = fadd double %tmp, %mul34 %arrayidx17 = getelementptr inbounds [1024 x double], ptr %C, i64 %indvars.iv43, i64 %indvars.iv4035 %tmp3 = load double, ptr %arrayidx17, align 836 %add18 = fadd double %tmp3, %add37 store double %add18, ptr %arrayidx17, align 838 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 139 %exitcond = icmp ne i64 %indvars.iv.next, 102440 br i1 %exitcond, label %for.body6, label %for.inc1941 42for.inc19: ; preds = %for.body643 %indvars.iv.next41 = add nuw nsw i64 %indvars.iv40, 144 %exitcond42 = icmp ne i64 %indvars.iv.next41, 102445 br i1 %exitcond42, label %for.body3, label %for.inc2246 47for.inc22: ; preds = %for.inc1948 %indvars.iv.next44 = add nuw nsw i64 %indvars.iv43, 149 %exitcond45 = icmp ne i64 %indvars.iv.next44, 102450 br i1 %exitcond45, label %for.body, label %for.end2451 52for.end24: ; preds = %for.inc2253 ret void54}55