<div dir="auto">X86 and did. :)<div dir="auto"><br></div><div dir="auto">Few more messages from Craig and I in the thread. </div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Tue, Jun 25, 2019, 12:37 AM Simon Pilgrim <<a href="mailto:llvm-dev@redking.me.uk">llvm-dev@redking.me.uk</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Is this on X86 or some other target? The SimplifyDemanded* functions can <br>
be a little quick to change opcodes without proper legal op testing....<br>
<br>
If you can send me a repro I'll take a look.<br>
<br>
Simon.<br>
<br>
On 25/06/2019 00:58, Eric Christopher wrote:<br>
> Hi Simon,<br>
><br>
> I'm seeing at least one occurrence of "fatal error: error in backend:<br>
> Cannot emit physreg copy instruction" after this patch attempting to<br>
> build an mp4 library. I'm not entirely sure from where or how (inline<br>
> assembly, normal code, etc), but wanted to let you know as we<br>
> investigate. The code has, of course, been building up to this point<br>
> so my inclination is to revert and get a testcase reduced out of the<br>
> sources.<br>
><br>
> Thoughts?<br>
><br>
> -eric<br>
><br>
> On Wed, Jun 19, 2019 at 6:54 AM Simon Pilgrim via llvm-commits<br>
> <<a href="mailto:llvm-commits@lists.llvm.org" target="_blank" rel="noreferrer">llvm-commits@lists.llvm.org</a>> wrote:<br>
>> Author: rksimon<br>
>> Date: Wed Jun 19 06:58:02 2019<br>
>> New Revision: 363802<br>
>><br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project?rev=363802&view=rev" rel="noreferrer noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project?rev=363802&view=rev</a><br>
>> Log:<br>
>> [TargetLowering] SimplifyDemandedBits SIGN_EXTEND_VECTOR_INREG -> ANY/ZERO_EXTEND_VECTOR_INREG<br>
>><br>
>> Simplify SIGN_EXTEND_VECTOR_INREG if the extended bits are not required/known zero.<br>
>><br>
>> Matches what we already do for SIGN_EXTEND.<br>
>><br>
>> Modified:<br>
>> llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp<br>
>> llvm/trunk/test/CodeGen/X86/pmul.ll<br>
>> llvm/trunk/test/CodeGen/X86/xop-ifma.ll<br>
>><br>
>> Modified: llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp<br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp?rev=363802&r1=363801&r2=363802&view=diff" rel="noreferrer noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp?rev=363802&r1=363801&r2=363802&view=diff</a><br>
>> ==============================================================================<br>
>> --- llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp (original)<br>
>> +++ llvm/trunk/lib/CodeGen/SelectionDAG/TargetLowering.cpp Wed Jun 19 06:58:02 2019<br>
>> @@ -1413,9 +1413,11 @@ bool TargetLowering::SimplifyDemandedBit<br>
>> bool IsVecInReg = Op.getOpcode() == ISD::SIGN_EXTEND_VECTOR_INREG;<br>
>><br>
>> // If none of the top bits are demanded, convert this into an any_extend.<br>
>> - // TODO: Add SIGN_EXTEND_VECTOR_INREG - ANY_EXTEND_VECTOR_INREG fold.<br>
>> - if (DemandedBits.getActiveBits() <= InBits && !IsVecInReg)<br>
>> - return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::ANY_EXTEND, dl, VT, Src));<br>
>> + if (DemandedBits.getActiveBits() <= InBits)<br>
>> + return TLO.CombineTo(<br>
>> + Op, TLO.DAG.getNode(IsVecInReg ? ISD::ANY_EXTEND_VECTOR_INREG<br>
>> + : ISD::ANY_EXTEND,<br>
>> + dl, VT, Src));<br>
>><br>
>> APInt InDemandedBits = DemandedBits.trunc(InBits);<br>
>> APInt InDemandedElts = DemandedElts.zextOrSelf(InElts);<br>
>> @@ -1434,9 +1436,11 @@ bool TargetLowering::SimplifyDemandedBit<br>
>> Known = Known.sext(BitWidth);<br>
>><br>
>> // If the sign bit is known zero, convert this to a zero extend.<br>
>> - // TODO: Add SIGN_EXTEND_VECTOR_INREG - ZERO_EXTEND_VECTOR_INREG fold.<br>
>> - if (Known.isNonNegative() && !IsVecInReg)<br>
>> - return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::ZERO_EXTEND, dl, VT, Src));<br>
>> + if (Known.isNonNegative())<br>
>> + return TLO.CombineTo(<br>
>> + Op, TLO.DAG.getNode(IsVecInReg ? ISD::ZERO_EXTEND_VECTOR_INREG<br>
>> + : ISD::ZERO_EXTEND,<br>
>> + dl, VT, Src));<br>
>> break;<br>
>> }<br>
>> case ISD::ANY_EXTEND: {<br>
>><br>
>> Modified: llvm/trunk/test/CodeGen/X86/pmul.ll<br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pmul.ll?rev=363802&r1=363801&r2=363802&view=diff" rel="noreferrer noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pmul.ll?rev=363802&r1=363801&r2=363802&view=diff</a><br>
>> ==============================================================================<br>
>> --- llvm/trunk/test/CodeGen/X86/pmul.ll (original)<br>
>> +++ llvm/trunk/test/CodeGen/X86/pmul.ll Wed Jun 19 06:58:02 2019<br>
>> @@ -1326,15 +1326,13 @@ define <8 x i64> @mul_v8i64_sext(<8 x i1<br>
>> ; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm0[1,1,2,3]<br>
>> ; SSE41-NEXT: pmovsxwq %xmm3, %xmm6<br>
>> ; SSE41-NEXT: pmovsxwq %xmm0, %xmm7<br>
>> -; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]<br>
>> -; SSE41-NEXT: pmovsxdq %xmm0, %xmm3<br>
>> +; SSE41-NEXT: pshufd {{.*#+}} xmm3 = xmm2[2,2,3,3]<br>
>> ; SSE41-NEXT: pmuldq %xmm4, %xmm3<br>
>> -; SSE41-NEXT: pmovsxdq %xmm2, %xmm2<br>
>> +; SSE41-NEXT: pmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero<br>
>> ; SSE41-NEXT: pmuldq %xmm5, %xmm2<br>
>> -; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]<br>
>> -; SSE41-NEXT: pmovsxdq %xmm0, %xmm4<br>
>> +; SSE41-NEXT: pshufd {{.*#+}} xmm4 = xmm1[2,2,3,3]<br>
>> ; SSE41-NEXT: pmuldq %xmm6, %xmm4<br>
>> -; SSE41-NEXT: pmovsxdq %xmm1, %xmm0<br>
>> +; SSE41-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero<br>
>> ; SSE41-NEXT: pmuldq %xmm7, %xmm0<br>
>> ; SSE41-NEXT: movdqa %xmm4, %xmm1<br>
>> ; SSE41-NEXT: retq<br>
>><br>
>> Modified: llvm/trunk/test/CodeGen/X86/xop-ifma.ll<br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/xop-ifma.ll?rev=363802&r1=363801&r2=363802&view=diff" rel="noreferrer noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/xop-ifma.ll?rev=363802&r1=363801&r2=363802&view=diff</a><br>
>> ==============================================================================<br>
>> --- llvm/trunk/test/CodeGen/X86/xop-ifma.ll (original)<br>
>> +++ llvm/trunk/test/CodeGen/X86/xop-ifma.ll Wed Jun 19 06:58:02 2019<br>
>> @@ -67,12 +67,10 @@ define <8 x i32> @test_mul_v8i32_add_v8i<br>
>> define <4 x i64> @test_mulx_v4i32_add_v4i64(<4 x i32> %a0, <4 x i32> %a1, <4 x i64> %a2) {<br>
>> ; XOP-AVX1-LABEL: test_mulx_v4i32_add_v4i64:<br>
>> ; XOP-AVX1: # %bb.0:<br>
>> -; XOP-AVX1-NEXT: vpmovsxdq %xmm0, %xmm3<br>
>> -; XOP-AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]<br>
>> -; XOP-AVX1-NEXT: vpmovsxdq %xmm0, %xmm0<br>
>> -; XOP-AVX1-NEXT: vpmovsxdq %xmm1, %xmm4<br>
>> -; XOP-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]<br>
>> -; XOP-AVX1-NEXT: vpmovsxdq %xmm1, %xmm1<br>
>> +; XOP-AVX1-NEXT: vpmovzxdq {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero<br>
>> +; XOP-AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,1,3,3]<br>
>> +; XOP-AVX1-NEXT: vpmovzxdq {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero<br>
>> +; XOP-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]<br>
>> ; XOP-AVX1-NEXT: vextractf128 $1, %ymm2, %xmm5<br>
>> ; XOP-AVX1-NEXT: vpmacsdql %xmm5, %xmm1, %xmm0, %xmm0<br>
>> ; XOP-AVX1-NEXT: vpmacsdql %xmm2, %xmm4, %xmm3, %xmm1<br>
>><br>
>><br>
>> _______________________________________________<br>
>> llvm-commits mailing list<br>
>> <a href="mailto:llvm-commits@lists.llvm.org" target="_blank" rel="noreferrer">llvm-commits@lists.llvm.org</a><br>
>> <a href="https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer noreferrer" target="_blank">https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>
</blockquote></div>