1 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s 2 ; 3 ; Mul with operands that are not simple load and sext/zext chains: this is not 4 ; yet supported so the rewrite shouldn't trigger (but we do want to support this 5 ; soon). 6 ; 7 ; CHECK-NOT: call i32 @llvm.arm.smlad 8 ; 9 define dso_local i32 @test(i32 %arg, i32* nocapture readnone %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3, i16* %arg4) { 10 entry: 11 %cmp24 = icmp sgt i32 %arg, 0 12 br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup 13 14 for.body.preheader: 15 %.pre = load i16, i16* %arg3, align 2 16 %.pre27 = load i16, i16* %arg2, align 2 17 %gep0 = getelementptr inbounds i16, i16* %arg4, i32 0 18 %gep1 = getelementptr inbounds i16, i16* %arg4, i32 1 19 %.add4 = load i16, i16* %gep0, align 2 20 %.add5 = load i16, i16* %gep1, align 2 21 %.zext4 = zext i16 %.add4 to i32 22 %.zext5 = zext i16 %.add5 to i32 23 br label %for.body 24 25 for.cond.cleanup: 26 %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ] 27 ret i32 %mac1.0.lcssa 28 29 for.body: 30 %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ] 31 %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ] 32 %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025 33 %0 = load i16, i16* %arrayidx, align 2 34 %add = add nuw nsw i32 %i.025, 1 35 %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add 36 %1 = load i16, i16* %arrayidx1, align 2 37 %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025 38 %2 = load i16, i16* %arrayidx3, align 2 39 %conv = sext i16 %2 to i32 40 %conv4 = sext i16 %0 to i32 41 %add1 = add i32 %conv, %.zext4 42 43 ; This mul has a more complicated pattern as an operand, %add1 44 ; is another add and load, which we don't support for now. 45 %mul = mul nsw i32 %add1, %conv4 46 %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add 47 %3 = load i16, i16* %arrayidx6, align 2 48 %conv7 = sext i16 %3 to i32 49 %conv8 = sext i16 %1 to i32 50 %add2 = add i32 %conv7, %.zext5 51 52 ; Same here 53 %mul9 = mul nsw i32 %add2, %conv8 54 %add10 = add i32 %mul, %mac1.026 55 56 %add11 = add i32 %mul9, %add10 57 %exitcond = icmp ne i32 %add, %arg 58 br i1 %exitcond, label %for.body, label %for.cond.cleanup 59 } 60
