49 lines · plain
1; RUN: opt -mtriple=armv8m.main-none-none-eabi -mattr=+dsp < %s -arm-parallel-dsp -S | FileCheck %s2;3; The loop header is not the loop latch.4;5; CHECK: call i32 @llvm.arm.smlad6;7define dso_local i32 @test(i32 %arg, ptr nocapture readnone %arg1, ptr nocapture readonly %arg2, ptr nocapture readonly %arg3) {8entry:9 %cmp24 = icmp sgt i32 %arg, 010 br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup11 12for.body.preheader:13 %.pre = load i16, ptr %arg3, align 214 %.pre27 = load i16, ptr %arg2, align 215 br label %for.body16 17for.cond.cleanup:18 %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]19 ret i32 %mac1.0.lcssa20 21; This is the loop header:22for.body:23 %mac1.026 = phi i32 [ %add11, %for.body2 ], [ 0, %for.body.preheader ]24 %i.025 = phi i32 [ %add, %for.body2 ], [ 0, %for.body.preheader ]25 %arrayidx = getelementptr inbounds i16, ptr %arg3, i32 %i.02526 %0 = load i16, ptr %arrayidx, align 227 %add = add nuw nsw i32 %i.025, 128 %arrayidx1 = getelementptr inbounds i16, ptr %arg3, i32 %add29 %1 = load i16, ptr %arrayidx1, align 230 %arrayidx3 = getelementptr inbounds i16, ptr %arg2, i32 %i.02531 %2 = load i16, ptr %arrayidx3, align 232 %conv = sext i16 %2 to i3233 %conv4 = sext i16 %0 to i3234 %mul = mul nsw i32 %conv, %conv435 %arrayidx6 = getelementptr inbounds i16, ptr %arg2, i32 %add36 %3 = load i16, ptr %arrayidx6, align 237 %conv7 = sext i16 %3 to i3238 %conv8 = sext i16 %1 to i3239 %mul9 = mul nsw i32 %conv7, %conv840 %add10 = add i32 %mul, %mac1.02641 %add11 = add i32 %mul9, %add1042 %exitcond = icmp ne i32 %add, %arg43 br i1 %exitcond, label %for.body2, label %for.cond.cleanup44 45; And this is the loop latch:46for.body2:47 br label %for.body48}49