1; REQUIRES: asserts 2; RUN: opt -mtriple=armv8m.main-none-none-eabi -mattr=+dsp < %s -arm-parallel-dsp -S -stats 2>&1 | FileCheck %s 3; 4; A more complicated chain: 4 mul operations, so we expect 2 smlad calls. 5; 6; CHECK: %mac1{{\.}}054 = phi i32 [ [[V17:%[0-9]+]], %for.body ], [ 0, %for.body.preheader ] 7; CHECK: [[V11:%[0-9]+]] = load i32, ptr %arrayidx, align 2 8; CHECK: [[V16:%[0-9]+]] = load i32, ptr %arrayidx4, align 2 9; CHECK: [[V9:%[0-9]+]] = load i32, ptr %arrayidx8, align 2 10; CHECK: [[ACC:%[0-9]+]] = call i32 @llvm.arm.smlad(i32 [[V9]], i32 [[V11]], i32 %mac1{{\.}}054) 11; CHECK: [[V14:%[0-9]+]] = load i32, ptr %arrayidx17, align 2 12; CHECK: [[V12:%[0-9]+]] = call i32 @llvm.arm.smlad(i32 [[V14]], i32 [[V16]], i32 [[ACC]]) 13; 14; And we don't want to see a 3rd smlad: 15; CHECK-NOT: call i32 @llvm.arm.smlad 16; 17; CHECK: 2 arm-parallel-dsp - Number of smlad instructions generated 18; 19define dso_local i32 @test(i32 %arg, ptr nocapture readnone %arg1, ptr nocapture readonly %arg2, ptr nocapture readonly %arg3) { 20entry: 21 %cmp52 = icmp sgt i32 %arg, 0 22 br i1 %cmp52, label %for.body.preheader, label %for.cond.cleanup 23 24for.cond.cleanup: 25 %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add28, %for.body ] 26 ret i32 %mac1.0.lcssa 27 28for.body.preheader: 29 br label %for.body 30 31for.body: 32 %mac1.054 = phi i32 [ %add28, %for.body ], [ 0, %for.body.preheader ] 33 %i.053 = phi i32 [ %add29, %for.body ], [ 0, %for.body.preheader ] 34 %arrayidx = getelementptr inbounds i16, ptr %arg3, i32 %i.053 35 %0 = load i16, ptr %arrayidx, align 2 36 %add1 = or disjoint i32 %i.053, 1 37 %arrayidx2 = getelementptr inbounds i16, ptr %arg3, i32 %add1 38 %1 = load i16, ptr %arrayidx2, align 2 39 %add3 = or disjoint i32 %i.053, 2 40 %arrayidx4 = getelementptr inbounds i16, ptr %arg3, i32 %add3 41 %2 = load i16, ptr %arrayidx4, align 2 42 %add5 = or disjoint i32 %i.053, 3 43 %arrayidx6 = getelementptr inbounds i16, ptr %arg3, i32 %add5 44 %3 = load i16, ptr %arrayidx6, align 2 45 %arrayidx8 = getelementptr inbounds i16, ptr %arg2, i32 %i.053 46 %4 = load i16, ptr %arrayidx8, align 2 47 %conv = sext i16 %4 to i32 48 %conv9 = sext i16 %0 to i32 49 %mul = mul nsw i32 %conv, %conv9 50 %arrayidx11 = getelementptr inbounds i16, ptr %arg2, i32 %add1 51 %5 = load i16, ptr %arrayidx11, align 2 52 %conv12 = sext i16 %5 to i32 53 %conv13 = sext i16 %1 to i32 54 %mul14 = mul nsw i32 %conv12, %conv13 55 %arrayidx17 = getelementptr inbounds i16, ptr %arg2, i32 %add3 56 %6 = load i16, ptr %arrayidx17, align 2 57 %conv18 = sext i16 %6 to i32 58 %conv19 = sext i16 %2 to i32 59 %mul20 = mul nsw i32 %conv18, %conv19 60 %arrayidx23 = getelementptr inbounds i16, ptr %arg2, i32 %add5 61 %7 = load i16, ptr %arrayidx23, align 2 62 %conv24 = sext i16 %7 to i32 63 %conv25 = sext i16 %3 to i32 64 %mul26 = mul nsw i32 %conv24, %conv25 65 %add15 = add i32 %mul, %mac1.054 66 %add21 = add i32 %add15, %mul14 67 %add27 = add i32 %add21, %mul20 68 %add28 = add i32 %add27, %mul26 69 %add29 = add nuw nsw i32 %i.053, 4 70 %cmp = icmp slt i32 %add29, %arg 71 br i1 %cmp, label %for.body, label %for.cond.cleanup 72} 73