<div dir="ltr">Hi Rafael and Bob,<div><br></div><div>Sorry about that! I've just reverted 216066.</div><div><br></div><div>Bob, do you also see infinite loop?</div><div><br></div><div>Thanks,</div><div>-Jiangning</div>
</div><div class="gmail_extra"><br><br><div class="gmail_quote">2014-08-21 8:54 GMT+08:00 Bob Wilson <span dir="ltr"><<a href="mailto:bob.wilson@apple.com" target="_blank">bob.wilson@apple.com</a>></span>:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
We are seeing similar massive slowdowns, although we had not yet identified the culprit.<br>
<div class="HOEnZb"><div class="h5"><br>
> On Aug 20, 2014, at 3:41 PM, Rafael Espíndola <<a href="mailto:rafael.espindola@gmail.com">rafael.espindola@gmail.com</a>> wrote:<br>
><br>
> I am seeing some massive slowdowns (infinite loop?) in a lto build of<br>
> clang after this.<br>
><br>
> With 216065 running llc (release+asserts) on the combined optimized<br>
> module takes 543.097 seconds. With this revision (216066) I killed it<br>
> after 39 minutes :-(<br>
><br>
> Would you mind reverting this for now? I will try to put the .bc file<br>
> somewhere you can access.<br>
><br>
><br>
> On 20 August 2014 08:05, Jiangning Liu <<a href="mailto:jiangning.liu@arm.com">jiangning.liu@arm.com</a>> wrote:<br>
>> Author: jiangning<br>
>> Date: Wed Aug 20 07:05:15 2014<br>
>> New Revision: 216066<br>
>><br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project?rev=216066&view=rev" target="_blank">http://llvm.org/viewvc/llvm-project?rev=216066&view=rev</a><br>
>> Log:<br>
>> Optimize ZERO_EXTEND and SIGN_EXTEND in both SelectionDAG Builder and type<br>
>> legalization stage. With those two optimizations, fewer signed/zero extension<br>
>> instructions can be inserted, and then we can expose more opportunities to<br>
>> Machine CSE pass in back-end.<br>
>><br>
>><br>
>> Added:<br>
>> llvm/trunk/test/CodeGen/AArch64/rm_redundant_cmp.ll<br>
>> Modified:<br>
>> llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp<br>
>> llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp<br>
>> llvm/trunk/test/CodeGen/AArch64/atomic-ops.ll<br>
>><br>
>> Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp<br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp?rev=216066&r1=216065&r2=216066&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp?rev=216066&r1=216065&r2=216066&view=diff</a><br>
>> ==============================================================================<br>
>> --- llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp (original)<br>
>> +++ llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp Wed Aug 20 07:05:15 2014<br>
>> @@ -861,7 +861,28 @@ void DAGTypeLegalizer::PromoteSetCCOpera<br>
>> switch (CCCode) {<br>
>> default: llvm_unreachable("Unknown integer comparison!");<br>
>> case ISD::SETEQ:<br>
>> - case ISD::SETNE:<br>
>> + case ISD::SETNE: {<br>
>> + SDValue OpL, OpR;<br>
>> +<br>
>> + OpL = GetPromotedInteger(NewLHS);<br>
>> + OpR = GetPromotedInteger(NewRHS);<br>
>> +<br>
>> + // We would prefer to promote the comparison operand with sign extension,<br>
>> + // if we find the operand is actually to truncate an AssertSext. With this<br>
>> + // optimization, we can avoid inserting real truncate instruction, which<br>
>> + // is redudant eventually.<br>
>> + if (OpL->getOpcode() == ISD::AssertSext &&<br>
>> + cast<VTSDNode>(OpL->getOperand(1))->getVT() == NewLHS.getValueType() &&<br>
>> + OpR->getOpcode() == ISD::AssertSext &&<br>
>> + cast<VTSDNode>(OpR->getOperand(1))->getVT() == NewRHS.getValueType()) {<br>
>> + NewLHS = OpL;<br>
>> + NewRHS = OpR;<br>
>> + } else {<br>
>> + NewLHS = ZExtPromotedInteger(NewLHS);<br>
>> + NewRHS = ZExtPromotedInteger(NewRHS);<br>
>> + }<br>
>> + break;<br>
>> + }<br>
>> case ISD::SETUGE:<br>
>> case ISD::SETUGT:<br>
>> case ISD::SETULE:<br>
>><br>
>> Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp<br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp?rev=216066&r1=216065&r2=216066&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp?rev=216066&r1=216065&r2=216066&view=diff</a><br>
>> ==============================================================================<br>
>> --- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp (original)<br>
>> +++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp Wed Aug 20 07:05:15 2014<br>
>> @@ -757,6 +757,28 @@ SDValue RegsForValue::getCopyFromRegs(Se<br>
>> return DAG.getNode(ISD::MERGE_VALUES, dl, DAG.getVTList(ValueVTs), Values);<br>
>> }<br>
>><br>
>> +static ISD::NodeType getPreferredExtendForValue(const Value *V) {<br>
>> + // For the users of the source value being used for compare instruction, if<br>
>> + // the number of signed predicate is greater than unsigned predicate, we<br>
>> + // prefer to use SIGN_EXTEND.<br>
>> + //<br>
>> + // With this optimization, we would be able to reduce some redundant sign or<br>
>> + // zero extension instruction, and eventually more machine CSE opportunities<br>
>> + // can be exposed.<br>
>> + ISD::NodeType ExtendKind = ISD::ANY_EXTEND;<br>
>> + unsigned int NumOfSigned = 0, NumOfUnsigned = 0;<br>
>> + for (const User *U : V->users()) {<br>
>> + if (const CmpInst *CI = dyn_cast<CmpInst>(U)) {<br>
>> + NumOfSigned += CI->isSigned();<br>
>> + NumOfUnsigned += CI->isUnsigned();<br>
>> + }<br>
>> + }<br>
>> + if (NumOfSigned > NumOfUnsigned)<br>
>> + ExtendKind = ISD::SIGN_EXTEND;<br>
>> +<br>
>> + return ExtendKind;<br>
>> +}<br>
>> +<br>
>> /// getCopyToRegs - Emit a series of CopyToReg nodes that copies the<br>
>> /// specified value into the registers specified by this object. This uses<br>
>> /// Chain/Flag as the input and updates them for the output Chain/Flag.<br>
>> @@ -765,6 +787,7 @@ void RegsForValue::getCopyToRegs(SDValue<br>
>> SDValue &Chain, SDValue *Flag,<br>
>> const Value *V) const {<br>
>> const TargetLowering &TLI = DAG.getTargetLoweringInfo();<br>
>> + ISD::NodeType ExtendKind = getPreferredExtendForValue(V);<br>
>><br>
>> // Get the list of the values's legal parts.<br>
>> unsigned NumRegs = Regs.size();<br>
>> @@ -773,8 +796,9 @@ void RegsForValue::getCopyToRegs(SDValue<br>
>> EVT ValueVT = ValueVTs[Value];<br>
>> unsigned NumParts = TLI.getNumRegisters(*DAG.getContext(), ValueVT);<br>
>> MVT RegisterVT = RegVTs[Value];<br>
>> - ISD::NodeType ExtendKind =<br>
>> - TLI.isZExtFree(Val, RegisterVT)? ISD::ZERO_EXTEND: ISD::ANY_EXTEND;<br>
>> +<br>
>> + if (ExtendKind == ISD::ANY_EXTEND && TLI.isZExtFree(Val, RegisterVT))<br>
>> + ExtendKind = ISD::ZERO_EXTEND;<br>
>><br>
>> getCopyToParts(DAG, dl, Val.getValue(Val.getResNo() + Value),<br>
>> &Parts[Part], NumParts, RegisterVT, V, ExtendKind);<br>
>><br>
>> Modified: llvm/trunk/test/CodeGen/AArch64/atomic-ops.ll<br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/atomic-ops.ll?rev=216066&r1=216065&r2=216066&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/atomic-ops.ll?rev=216066&r1=216065&r2=216066&view=diff</a><br>
>> ==============================================================================<br>
>> --- llvm/trunk/test/CodeGen/AArch64/atomic-ops.ll (original)<br>
>> +++ llvm/trunk/test/CodeGen/AArch64/atomic-ops.ll Wed Aug 20 07:05:15 2014<br>
>> @@ -493,6 +493,7 @@ define i8 @test_atomic_load_min_i8(i8 %o<br>
>> ; CHECK-LABEL: test_atomic_load_min_i8:<br>
>> %old = atomicrmw min i8* @var8, i8 %offset acquire<br>
>> ; CHECK-NOT: dmb<br>
>> +; CHECK: sxtb w[[TMP:[0-9]+]], w0<br>
>> ; CHECK: adrp [[TMPADDR:x[0-9]+]], var8<br>
>> ; CHECK: add x[[ADDR:[0-9]+]], [[TMPADDR]], {{#?}}:lo12:var8<br>
>><br>
>> @@ -502,14 +503,13 @@ define i8 @test_atomic_load_min_i8(i8 %o<br>
>> ; function there.<br>
>><br>
>> ; CHECK-NEXT: sxtb w[[OLD_EXT:[0-9]+]], w[[OLD]]<br>
>> -; CHECK-NEXT: cmp w[[OLD_EXT]], w0, sxtb<br>
>> -; CHECK-NEXT: csel [[NEW:w[0-9]+]], w[[OLD]], w0, le<br>
>> +; CHECK-NEXT: cmp w[[OLD_EXT]], w[[TMP]]<br>
>> +; CHECK-NEXT: csel [[NEW:w[0-9]+]], w[[OLD]], w[[TMP]], le<br>
>><br>
>> ; CHECK-NEXT: stxrb [[STATUS:w[0-9]+]], [[NEW]], [x[[ADDR]]]<br>
>> ; CHECK-NEXT: cbnz [[STATUS]], .LBB{{[0-9]+}}_1<br>
>> ; CHECK-NOT: dmb<br>
>><br>
>> -; CHECK: mov {{[xw]}}0, {{[xw]}}[[OLD]]<br>
>> ret i8 %old<br>
>> }<br>
>><br>
>> @@ -517,6 +517,7 @@ define i16 @test_atomic_load_min_i16(i16<br>
>> ; CHECK-LABEL: test_atomic_load_min_i16:<br>
>> %old = atomicrmw min i16* @var16, i16 %offset release<br>
>> ; CHECK-NOT: dmb<br>
>> +; CHECK: sxth w[[TMP:[0-9]+]], w0<br>
>> ; CHECK: adrp [[TMPADDR:x[0-9]+]], var16<br>
>> ; CHECK: add x[[ADDR:[0-9]+]], [[TMPADDR]], {{#?}}:lo12:var16<br>
>><br>
>> @@ -526,15 +527,14 @@ define i16 @test_atomic_load_min_i16(i16<br>
>> ; function there.<br>
>><br>
>> ; CHECK-NEXT: sxth w[[OLD_EXT:[0-9]+]], w[[OLD]]<br>
>> -; CHECK-NEXT: cmp w[[OLD_EXT]], w0, sxth<br>
>> -; CHECK-NEXT: csel [[NEW:w[0-9]+]], w[[OLD]], w0, le<br>
>> +; CHECK-NEXT: cmp w[[OLD_EXT]], w[[TMP]]<br>
>> +; CHECK-NEXT: csel [[NEW:w[0-9]+]], w[[OLD]], w[[TMP]], le<br>
>><br>
>><br>
>> ; CHECK-NEXT: stlxrh [[STATUS:w[0-9]+]], [[NEW]], [x[[ADDR]]]<br>
>> ; CHECK-NEXT: cbnz [[STATUS]], .LBB{{[0-9]+}}_1<br>
>> ; CHECK-NOT: dmb<br>
>><br>
>> -; CHECK: mov {{[xw]}}0, {{[xw]}}[[OLD]]<br>
>> ret i16 %old<br>
>> }<br>
>><br>
>> @@ -590,6 +590,7 @@ define i8 @test_atomic_load_max_i8(i8 %o<br>
>> ; CHECK-LABEL: test_atomic_load_max_i8:<br>
>> %old = atomicrmw max i8* @var8, i8 %offset seq_cst<br>
>> ; CHECK-NOT: dmb<br>
>> +; CHECK: sxtb w[[TMP:[0-9]+]], w0<br>
>> ; CHECK: adrp [[TMPADDR:x[0-9]+]], var8<br>
>> ; CHECK: add x[[ADDR:[0-9]+]], [[TMPADDR]], {{#?}}:lo12:var8<br>
>><br>
>> @@ -599,15 +600,14 @@ define i8 @test_atomic_load_max_i8(i8 %o<br>
>> ; function there.<br>
>><br>
>> ; CHECK-NEXT: sxtb w[[OLD_EXT:[0-9]+]], w[[OLD]]<br>
>> -; CHECK-NEXT: cmp w[[OLD_EXT]], w0, sxtb<br>
>> -; CHECK-NEXT: csel [[NEW:w[0-9]+]], w[[OLD]], w0, gt<br>
>> +; CHECK-NEXT: cmp w[[OLD_EXT]], w[[TMP]]<br>
>> +; CHECK-NEXT: csel [[NEW:w[0-9]+]], w[[OLD]], w[[TMP]], gt<br>
>><br>
>><br>
>> ; CHECK-NEXT: stlxrb [[STATUS:w[0-9]+]], [[NEW]], [x[[ADDR]]]<br>
>> ; CHECK-NEXT: cbnz [[STATUS]], .LBB{{[0-9]+}}_1<br>
>> ; CHECK-NOT: dmb<br>
>><br>
>> -; CHECK: mov {{[xw]}}0, {{[xw]}}[[OLD]]<br>
>> ret i8 %old<br>
>> }<br>
>><br>
>> @@ -615,6 +615,7 @@ define i16 @test_atomic_load_max_i16(i16<br>
>> ; CHECK-LABEL: test_atomic_load_max_i16:<br>
>> %old = atomicrmw max i16* @var16, i16 %offset acquire<br>
>> ; CHECK-NOT: dmb<br>
>> +; CHECK: sxth w[[TMP:[0-9]+]], w0<br>
>> ; CHECK: adrp [[TMPADDR:x[0-9]+]], var16<br>
>> ; CHECK: add x[[ADDR:[0-9]+]], [[TMPADDR]], {{#?}}:lo12:var16<br>
>><br>
>> @@ -624,15 +625,14 @@ define i16 @test_atomic_load_max_i16(i16<br>
>> ; function there.<br>
>><br>
>> ; CHECK-NEXT: sxth w[[OLD_EXT:[0-9]+]], w[[OLD]]<br>
>> -; CHECK-NEXT: cmp w[[OLD_EXT]], w0, sxth<br>
>> -; CHECK-NEXT: csel [[NEW:w[0-9]+]], w[[OLD]], w0, gt<br>
>> +; CHECK-NEXT: cmp w[[OLD_EXT]], w[[TMP]]<br>
>> +; CHECK-NEXT: csel [[NEW:w[0-9]+]], w[[OLD]], w[[TMP]], gt<br>
>><br>
>><br>
>> ; CHECK-NEXT: stxrh [[STATUS:w[0-9]+]], [[NEW]], [x[[ADDR]]]<br>
>> ; CHECK-NEXT: cbnz [[STATUS]], .LBB{{[0-9]+}}_1<br>
>> ; CHECK-NOT: dmb<br>
>><br>
>> -; CHECK: mov {{[xw]}}0, {{[xw]}}[[OLD]]<br>
>> ret i16 %old<br>
>> }<br>
>><br>
>><br>
>> Added: llvm/trunk/test/CodeGen/AArch64/rm_redundant_cmp.ll<br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/rm_redundant_cmp.ll?rev=216066&view=auto" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/rm_redundant_cmp.ll?rev=216066&view=auto</a><br>
>> ==============================================================================<br>
>> --- llvm/trunk/test/CodeGen/AArch64/rm_redundant_cmp.ll (added)<br>
>> +++ llvm/trunk/test/CodeGen/AArch64/rm_redundant_cmp.ll Wed Aug 20 07:05:15 2014<br>
>> @@ -0,0 +1,254 @@<br>
>> +; RUN: llc < %s -mtriple=aarch64-linux-gnuabi -O2 | FileCheck %s<br>
>> +<br>
>> +; The following cases are for i16<br>
>> +<br>
>> +%struct.s_signed_i16 = type { i16, i16, i16 }<br>
>> +%struct.s_unsigned_i16 = type { i16, i16, i16 }<br>
>> +<br>
>> +@cost_s_i8_i16 = common global %struct.s_signed_i16 zeroinitializer, align 2<br>
>> +@cost_u_i16 = common global %struct.s_unsigned_i16 zeroinitializer, align 2<br>
>> +<br>
>> +define void @test_i16_2cmp_signed_1() {<br>
>> +; CHECK-LABEL: test_i16_2cmp_signed_1<br>
>> +; CHECK: cmp {{w[0-9]+}}, {{w[0-9]+}}<br>
>> +; CHECK-NEXT: <a href="http://b.gt" target="_blank">b.gt</a><br>
>> +; CHECK-NOT: cmp<br>
>> +; CHECK: <a href="http://b.ne" target="_blank">b.ne</a><br>
>> +entry:<br>
>> + %0 = load i16* getelementptr inbounds (%struct.s_signed_i16* @cost_s_i8_i16, i64 0, i32 1), align 2<br>
>> + %1 = load i16* getelementptr inbounds (%struct.s_signed_i16* @cost_s_i8_i16, i64 0, i32 2), align 2<br>
>> + %cmp = icmp sgt i16 %0, %1<br>
>> + br i1 %cmp, label %if.then, label %if.else<br>
>> +<br>
>> +if.then: ; preds = %entry<br>
>> + store i16 %0, i16* getelementptr inbounds (%struct.s_signed_i16* @cost_s_i8_i16, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.else: ; preds = %entry<br>
>> + %cmp5 = icmp eq i16 %0, %1<br>
>> + br i1 %cmp5, label %if.then7, label %if.end8<br>
>> +<br>
>> +if.then7: ; preds = %if.else<br>
>> + store i16 %0, i16* getelementptr inbounds (%struct.s_signed_i16* @cost_s_i8_i16, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.end8: ; preds = %if.else, %if.then7, %if.then<br>
>> + ret void<br>
>> +}<br>
>> +<br>
>> +define void @test_i16_2cmp_signed_2() {<br>
>> +; CHECK-LABEL: test_i16_2cmp_signed_2<br>
>> +; CHECK: cmp {{w[0-9]+}}, {{w[0-9]+}}<br>
>> +; CHECK-NEXT: b.le<br>
>> +; CHECK-NOT: cmp<br>
>> +; CHECK: <a href="http://b.ge" target="_blank">b.ge</a><br>
>> +entry:<br>
>> + %0 = load i16* getelementptr inbounds (%struct.s_signed_i16* @cost_s_i8_i16, i64 0, i32 1), align 2<br>
>> + %1 = load i16* getelementptr inbounds (%struct.s_signed_i16* @cost_s_i8_i16, i64 0, i32 2), align 2<br>
>> + %cmp = icmp sgt i16 %0, %1<br>
>> + br i1 %cmp, label %if.then, label %if.else<br>
>> +<br>
>> +if.then: ; preds = %entry<br>
>> + store i16 %0, i16* getelementptr inbounds (%struct.s_signed_i16* @cost_s_i8_i16, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.else: ; preds = %entry<br>
>> + %cmp5 = icmp slt i16 %0, %1<br>
>> + br i1 %cmp5, label %if.then7, label %if.end8<br>
>> +<br>
>> +if.then7: ; preds = %if.else<br>
>> + store i16 %1, i16* getelementptr inbounds (%struct.s_signed_i16* @cost_s_i8_i16, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.end8: ; preds = %if.else, %if.then7, %if.then<br>
>> + ret void<br>
>> +}<br>
>> +<br>
>> +define void @test_i16_2cmp_unsigned_1() {<br>
>> +; CHECK-LABEL: test_i16_2cmp_unsigned_1<br>
>> +; CHECK: cmp {{w[0-9]+}}, {{w[0-9]+}}<br>
>> +; CHECK-NEXT: b.hi<br>
>> +; CHECK-NOT: cmp<br>
>> +; CHECK: <a href="http://b.ne" target="_blank">b.ne</a><br>
>> +entry:<br>
>> + %0 = load i16* getelementptr inbounds (%struct.s_unsigned_i16* @cost_u_i16, i64 0, i32 1), align 2<br>
>> + %1 = load i16* getelementptr inbounds (%struct.s_unsigned_i16* @cost_u_i16, i64 0, i32 2), align 2<br>
>> + %cmp = icmp ugt i16 %0, %1<br>
>> + br i1 %cmp, label %if.then, label %if.else<br>
>> +<br>
>> +if.then: ; preds = %entry<br>
>> + store i16 %0, i16* getelementptr inbounds (%struct.s_unsigned_i16* @cost_u_i16, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.else: ; preds = %entry<br>
>> + %cmp5 = icmp eq i16 %0, %1<br>
>> + br i1 %cmp5, label %if.then7, label %if.end8<br>
>> +<br>
>> +if.then7: ; preds = %if.else<br>
>> + store i16 %0, i16* getelementptr inbounds (%struct.s_unsigned_i16* @cost_u_i16, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.end8: ; preds = %if.else, %if.then7, %if.then<br>
>> + ret void<br>
>> +}<br>
>> +<br>
>> +define void @test_i16_2cmp_unsigned_2() {<br>
>> +; CHECK-LABEL: test_i16_2cmp_unsigned_2<br>
>> +; CHECK: cmp {{w[0-9]+}}, {{w[0-9]+}}<br>
>> +; CHECK-NEXT: <a href="http://b.ls" target="_blank">b.ls</a><br>
>> +; CHECK-NOT: cmp<br>
>> +; CHECK: b.hs<br>
>> +entry:<br>
>> + %0 = load i16* getelementptr inbounds (%struct.s_unsigned_i16* @cost_u_i16, i64 0, i32 1), align 2<br>
>> + %1 = load i16* getelementptr inbounds (%struct.s_unsigned_i16* @cost_u_i16, i64 0, i32 2), align 2<br>
>> + %cmp = icmp ugt i16 %0, %1<br>
>> + br i1 %cmp, label %if.then, label %if.else<br>
>> +<br>
>> +if.then: ; preds = %entry<br>
>> + store i16 %0, i16* getelementptr inbounds (%struct.s_unsigned_i16* @cost_u_i16, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.else: ; preds = %entry<br>
>> + %cmp5 = icmp ult i16 %0, %1<br>
>> + br i1 %cmp5, label %if.then7, label %if.end8<br>
>> +<br>
>> +if.then7: ; preds = %if.else<br>
>> + store i16 %1, i16* getelementptr inbounds (%struct.s_unsigned_i16* @cost_u_i16, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.end8: ; preds = %if.else, %if.then7, %if.then<br>
>> + ret void<br>
>> +}<br>
>> +<br>
>> +; The following cases are for i8<br>
>> +<br>
>> +%struct.s_signed_i8 = type { i8, i8, i8 }<br>
>> +%struct.s_unsigned_i8 = type { i8, i8, i8 }<br>
>> +<br>
>> +@cost_s = common global %struct.s_signed_i8 zeroinitializer, align 2<br>
>> +@cost_u_i8 = common global %struct.s_unsigned_i8 zeroinitializer, align 2<br>
>> +<br>
>> +<br>
>> +define void @test_i8_2cmp_signed_1() {<br>
>> +; CHECK-LABEL: test_i8_2cmp_signed_1<br>
>> +; CHECK: cmp {{w[0-9]+}}, {{w[0-9]+}}<br>
>> +; CHECK-NEXT: <a href="http://b.gt" target="_blank">b.gt</a><br>
>> +; CHECK-NOT: cmp<br>
>> +; CHECK: <a href="http://b.ne" target="_blank">b.ne</a><br>
>> +entry:<br>
>> + %0 = load i8* getelementptr inbounds (%struct.s_signed_i8* @cost_s, i64 0, i32 1), align 2<br>
>> + %1 = load i8* getelementptr inbounds (%struct.s_signed_i8* @cost_s, i64 0, i32 2), align 2<br>
>> + %cmp = icmp sgt i8 %0, %1<br>
>> + br i1 %cmp, label %if.then, label %if.else<br>
>> +<br>
>> +if.then: ; preds = %entry<br>
>> + store i8 %0, i8* getelementptr inbounds (%struct.s_signed_i8* @cost_s, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.else: ; preds = %entry<br>
>> + %cmp5 = icmp eq i8 %0, %1<br>
>> + br i1 %cmp5, label %if.then7, label %if.end8<br>
>> +<br>
>> +if.then7: ; preds = %if.else<br>
>> + store i8 %0, i8* getelementptr inbounds (%struct.s_signed_i8* @cost_s, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.end8: ; preds = %if.else, %if.then7, %if.then<br>
>> + ret void<br>
>> +}<br>
>> +<br>
>> +define void @test_i8_2cmp_signed_2() {<br>
>> +; CHECK-LABEL: test_i8_2cmp_signed_2<br>
>> +; CHECK: cmp {{w[0-9]+}}, {{w[0-9]+}}<br>
>> +; CHECK-NEXT: b.le<br>
>> +; CHECK-NOT: cmp<br>
>> +; CHECK: <a href="http://b.ge" target="_blank">b.ge</a><br>
>> +entry:<br>
>> + %0 = load i8* getelementptr inbounds (%struct.s_signed_i8* @cost_s, i64 0, i32 1), align 2<br>
>> + %1 = load i8* getelementptr inbounds (%struct.s_signed_i8* @cost_s, i64 0, i32 2), align 2<br>
>> + %cmp = icmp sgt i8 %0, %1<br>
>> + br i1 %cmp, label %if.then, label %if.else<br>
>> +<br>
>> +if.then: ; preds = %entry<br>
>> + store i8 %0, i8* getelementptr inbounds (%struct.s_signed_i8* @cost_s, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.else: ; preds = %entry<br>
>> + %cmp5 = icmp slt i8 %0, %1<br>
>> + br i1 %cmp5, label %if.then7, label %if.end8<br>
>> +<br>
>> +if.then7: ; preds = %if.else<br>
>> + store i8 %1, i8* getelementptr inbounds (%struct.s_signed_i8* @cost_s, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.end8: ; preds = %if.else, %if.then7, %if.then<br>
>> + ret void<br>
>> +}<br>
>> +<br>
>> +define void @test_i8_2cmp_unsigned_1() {<br>
>> +; CHECK-LABEL: test_i8_2cmp_unsigned_1<br>
>> +; CHECK: cmp {{w[0-9]+}}, {{w[0-9]+}}<br>
>> +; CHECK-NEXT: b.hi<br>
>> +; CHECK-NOT: cmp<br>
>> +; CHECK: <a href="http://b.ne" target="_blank">b.ne</a><br>
>> +entry:<br>
>> + %0 = load i8* getelementptr inbounds (%struct.s_unsigned_i8* @cost_u_i8, i64 0, i32 1), align 2<br>
>> + %1 = load i8* getelementptr inbounds (%struct.s_unsigned_i8* @cost_u_i8, i64 0, i32 2), align 2<br>
>> + %cmp = icmp ugt i8 %0, %1<br>
>> + br i1 %cmp, label %if.then, label %if.else<br>
>> +<br>
>> +if.then: ; preds = %entry<br>
>> + store i8 %0, i8* getelementptr inbounds (%struct.s_unsigned_i8* @cost_u_i8, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.else: ; preds = %entry<br>
>> + %cmp5 = icmp eq i8 %0, %1<br>
>> + br i1 %cmp5, label %if.then7, label %if.end8<br>
>> +<br>
>> +if.then7: ; preds = %if.else<br>
>> + store i8 %0, i8* getelementptr inbounds (%struct.s_unsigned_i8* @cost_u_i8, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.end8: ; preds = %if.else, %if.then7, %if.then<br>
>> + ret void<br>
>> +}<br>
>> +<br>
>> +define void @test_i8_2cmp_unsigned_2() {<br>
>> +; CHECK-LABEL: test_i8_2cmp_unsigned_2<br>
>> +; CHECK: cmp {{w[0-9]+}}, {{w[0-9]+}}<br>
>> +; CHECK-NEXT: <a href="http://b.ls" target="_blank">b.ls</a><br>
>> +; CHECK-NOT: cmp<br>
>> +; CHECK: b.hs<br>
>> +entry:<br>
>> + %0 = load i8* getelementptr inbounds (%struct.s_unsigned_i8* @cost_u_i8, i64 0, i32 1), align 2<br>
>> + %1 = load i8* getelementptr inbounds (%struct.s_unsigned_i8* @cost_u_i8, i64 0, i32 2), align 2<br>
>> + %cmp = icmp ugt i8 %0, %1<br>
>> + br i1 %cmp, label %if.then, label %if.else<br>
>> +<br>
>> +if.then: ; preds = %entry<br>
>> + store i8 %0, i8* getelementptr inbounds (%struct.s_unsigned_i8* @cost_u_i8, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.else: ; preds = %entry<br>
>> + %cmp5 = icmp ult i8 %0, %1<br>
>> + br i1 %cmp5, label %if.then7, label %if.end8<br>
>> +<br>
>> +if.then7: ; preds = %if.else<br>
>> + store i8 %1, i8* getelementptr inbounds (%struct.s_unsigned_i8* @cost_u_i8, i64 0, i32 0), align 2<br>
>> + br label %if.end8<br>
>> +<br>
>> +if.end8: ; preds = %if.else, %if.then7, %if.then<br>
>> + ret void<br>
>> +}<br>
>> +<br>
>> +; Make sure the case below won't crash.<br>
>> +<br>
>> +; The optimization of ZERO_EXTEND and SIGN_EXTEND in type legalization stage can't assert<br>
>> +; the operand of a set_cc is always a TRUNCATE.<br>
>> +<br>
>> +define i1 @foo(float %inl, float %inr) {<br>
>> + %lval = fptosi float %inl to i8<br>
>> + %rval = fptosi float %inr to i8<br>
>> + %sum = icmp eq i8 %lval, %rval<br>
>> + ret i1 %sum<br>
>> +}<br>
>><br>
>><br>
>> _______________________________________________<br>
>> llvm-commits mailing list<br>
>> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
>> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
> _______________________________________________<br>
> llvm-commits mailing list<br>
> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
<br>
<br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
</div></div></blockquote></div><br></div>