55 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<import-jscop;opt-isl>' -polly-import-jscop-postfix=transformed -polly-pattern-matching-based-opts=true -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 -debug -polly-tc-opt=true -disable-output < %s 2>&1 | FileCheck %s2; REQUIRES: asserts3;4; Check that the pattern matching detects the matrix multiplication pattern5; in case scalar memory accesses were replaced by accesses to newly created6; arrays.7;8; CHECK: The tensor contraction pattern was detected9; CHECK: The matrix multiplication pattern was detected10;11target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"12target triple = "x86_64-unknown-unknown"13 14define void @kernel_gemm(i32 %ni, i32 %nj, i32 %nk, double %A, ptr %B, ptr %C) {15entry:16 br label %entry.split17 18entry.split: ; preds = %entry19 br label %for.cond1.preheader20 21for.cond1.preheader: ; preds = %for.inc16, %entry.split22 %indvars.iv35 = phi i64 [ 0, %entry.split ], [ %indvars.iv.next36, %for.inc16 ]23 br label %for.cond4.preheader24 25for.cond4.preheader: ; preds = %for.inc13, %for.cond1.preheader26 %indvars.iv32 = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next33, %for.inc13 ]27 br label %for.body628 29for.body6: ; preds = %for.body6, %for.cond4.preheader30 %indvars.iv = phi i64 [ 0, %for.cond4.preheader ], [ %indvars.iv.next, %for.body6 ]31 %arrayidx8 = getelementptr inbounds [1024 x double], ptr %B, i64 %indvars.iv, i64 %indvars.iv3232 %tmp = load double, ptr %arrayidx8, align 833 %mul = fmul double %tmp, %A34 %arrayidx12 = getelementptr inbounds [1024 x double], ptr %C, i64 %indvars.iv35, i64 %indvars.iv3235 %tmp1 = load double, ptr %arrayidx12, align 836 %add = fadd double %tmp1, %mul37 store double %add, ptr %arrayidx12, 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.inc1341 42for.inc13: ; preds = %for.body643 %indvars.iv.next33 = add nuw nsw i64 %indvars.iv32, 144 %exitcond34 = icmp ne i64 %indvars.iv.next33, 102445 br i1 %exitcond34, label %for.cond4.preheader, label %for.inc1646 47for.inc16: ; preds = %for.inc1348 %indvars.iv.next36 = add nuw nsw i64 %indvars.iv35, 149 %exitcond37 = icmp ne i64 %indvars.iv.next36, 102450 br i1 %exitcond37, label %for.cond1.preheader, label %for.end1851 52for.end18: ; preds = %for.inc1653 ret void54}55