<div dir="ltr">Could you send me more details on how to reproduce the regression? My suspension is that the branching overhead is much higher on arm than x86. So you probably want to try a lower machine-sink-split-probability-threshold to prevent from excessive critical-edge breaking?<div><br></div><div>Thanks,</div><div>Dehao</div></div><div class="gmail_extra"><br><div class="gmail_quote">On Sun, Oct 23, 2016 at 12:15 AM, Keith Walker <span dir="ltr"><<a href="mailto:Keith.Walker@arm.com" target="_blank">Keith.Walker@arm.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Dehao,<br>
<br>
Just to let you know that unfortunately this change has had a severe impact<br>
on a third party embedded benchmarking suite running on AArch64, ARM and Thumb<br>
targets with regressions in multiple benchmarks ranging from about 10% to 48%<br>
<br>
I will try to get additional details for you when I have them available.<br>
<span class="HOEnZb"><font color="#888888"><br>
Keith<br>
</font></span><div class="HOEnZb"><div class="h5"><br>
> -----Original Message-----<br>
> From: llvm-commits [mailto:<a href="mailto:llvm-commits-bounces@lists.llvm.org">llvm-commits-bounces@<wbr>lists.llvm.org</a>] On<br>
> Behalf Of Dehao Chen via llvm-commits<br>
> Sent: 20 October 2016 19:07<br>
> To: <a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a><br>
> Subject: [llvm] r284757 - Using branch probability to guide critical<br>
> edge splitting.<br>
><br>
> Author: dehao<br>
> Date: Thu Oct 20 13:06:52 2016<br>
> New Revision: 284757<br>
><br>
> URL: <a href="http://llvm.org/viewvc/llvm-project?rev=284757&view=rev" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-<wbr>project?rev=284757&view=rev</a><br>
> Log:<br>
> Using branch probability to guide critical edge splitting.<br>
><br>
> Summary:<br>
> The original heuristic to break critical edge during machine sink is<br>
> relatively conservertive: when there is only one instruction sinkable to<br>
> the critical edge, it is likely that the machine sink pass will not<br>
> break the critical edge. This leads to many speculative instructions<br>
> executed at runtime. However, with profile info, we could model the<br>
> splitting benefits: if the critical edge has 50% taken rate, it would<br>
> always be beneficial to split the critical edge to avoid the speculated<br>
> runtime instructions. This patch uses profile to guide critical edge<br>
> splitting in machine sink pass.<br>
><br>
> The performance impact on speccpu2006 on Intel sandybridge machines:<br>
><br>
> spec/2006/fp/C++/444.namd 25.3 +0.26%<br>
> spec/2006/fp/C++/447.dealII 45.96 -0.10%<br>
> spec/2006/fp/C++/450.soplex 41.97 +1.49%<br>
> spec/2006/fp/C++/453.povray 36.83 -0.96%<br>
> spec/2006/fp/C/433.milc 23.81 +0.32%<br>
> spec/2006/fp/C/470.lbm 41.17 +0.34%<br>
> spec/2006/fp/C/482.sphinx3 48.13 +0.69%<br>
> spec/2006/int/C++/471.omnetpp 22.45 +3.25%<br>
> spec/2006/int/C++/473.astar 21.35 -2.06%<br>
> spec/2006/int/C++/483.<wbr>xalancbmk 36.02 -2.39%<br>
> spec/2006/int/C/400.perlbench 33.7 -0.17%<br>
> spec/2006/int/C/401.bzip2 22.9 +0.52%<br>
> spec/2006/int/C/403.gcc 32.42 -0.54%<br>
> spec/2006/int/C/429.mcf 39.59 +0.19%<br>
> spec/2006/int/C/445.gobmk 26.98 -0.00%<br>
> spec/2006/int/C/456.hmmer 24.52 -0.18%<br>
> spec/2006/int/C/458.sjeng 28.26 +0.02%<br>
> spec/2006/int/C/462.libquantum 55.44 +3.74%<br>
> spec/2006/int/C/464.h264ref 46.67 -0.39%<br>
><br>
> geometric mean +0.20%<br>
><br>
> Manually checked 473 and 471 to verify the diff is in the noise range.<br>
><br>
> Reviewers: rengolin, davidxl<br>
><br>
> Subscribers: llvm-commits<br>
><br>
> Differential Revision: <a href="https://reviews.llvm.org/D24818" rel="noreferrer" target="_blank">https://reviews.llvm.org/<wbr>D24818</a><br>
><br>
> Added:<br>
> llvm/trunk/test/CodeGen/X86/<wbr>machine-sink.ll<br>
> Modified:<br>
> llvm/trunk/lib/CodeGen/<wbr>MachineSink.cpp<br>
> llvm/trunk/test/CodeGen/ARM/<wbr>atomic-cmpxchg.ll<br>
> llvm/trunk/test/CodeGen/ARM/<wbr>code-placement.ll<br>
> llvm/trunk/test/CodeGen/X86/<wbr>block-placement.ll<br>
> llvm/trunk/test/CodeGen/X86/<wbr>clz.ll<br>
> llvm/trunk/test/CodeGen/X86/<wbr>loop-search.ll<br>
> llvm/trunk/test/CodeGen/X86/<wbr>phys_subreg_coalesce-2.ll<br>
> llvm/trunk/test/CodeGen/X86/<wbr>pr2659.ll<br>
> llvm/trunk/test/DebugInfo/<wbr>COFF/pieces.ll<br>
><br>
> Modified: llvm/trunk/lib/CodeGen/<wbr>MachineSink.cpp<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/lib/<wbr>CodeGen/MachineSink.cpp?rev=<wbr>284757&r1=284756&r2=2<br>
> 84757&view=diff<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/lib/CodeGen/<wbr>MachineSink.cpp (original)<br>
> +++ llvm/trunk/lib/CodeGen/<wbr>MachineSink.cpp Thu Oct 20 13:06:52 2016<br>
> @@ -24,6 +24,7 @@<br>
> #include "llvm/Analysis/AliasAnalysis.<wbr>h"<br>
> #include "llvm/CodeGen/<wbr>MachineBasicBlock.h"<br>
> #include "llvm/CodeGen/<wbr>MachineBlockFrequencyInfo.h"<br>
> +#include "llvm/CodeGen/<wbr>MachineBranchProbabilityInfo.<wbr>h"<br>
> #include "llvm/CodeGen/<wbr>MachineDominators.h"<br>
> #include "llvm/CodeGen/MachineFunction.<wbr>h"<br>
> #include "llvm/CodeGen/<wbr>MachineFunctionPass.h"<br>
> @@ -60,6 +61,15 @@ UseBlockFreqInfo("machine-<wbr>sink-bfi",<br>
> cl::desc("Use block frequency info to find successors to<br>
> sink"),<br>
> cl::init(true), cl::Hidden);<br>
><br>
> +static cl::opt<unsigned> SplitEdgeProbabilityThreshold(<br>
> + "machine-sink-split-<wbr>probability-threshold",<br>
> + cl::desc(<br>
> + "Percentage threshold for splitting single-instruction critical<br>
> edge. "<br>
> + "If the branch threshold is higher than this threshold, we<br>
> allow "<br>
> + "speculative execution of up to 1 instruction to avoid<br>
> branching to "<br>
> + "splitted critical edge"),<br>
> + cl::init(40), cl::Hidden);<br>
> +<br>
> STATISTIC(NumSunk, "Number of machine instructions sunk");<br>
> STATISTIC(NumSplit, "Number of critical edges split");<br>
> STATISTIC(NumCoalesces, "Number of copies coalesced");<br>
> @@ -74,6 +84,7 @@ namespace {<br>
> MachinePostDominatorTree *PDT; // Machine post dominator tree<br>
> MachineLoopInfo *LI;<br>
> const MachineBlockFrequencyInfo *MBFI;<br>
> + const MachineBranchProbabilityInfo *MBPI;<br>
> AliasAnalysis *AA;<br>
><br>
> // Remember which edges have been considered for breaking.<br>
> @@ -105,6 +116,7 @@ namespace {<br>
> AU.addRequired<<wbr>MachineDominatorTree>();<br>
> AU.addRequired<<wbr>MachinePostDominatorTree>();<br>
> AU.addRequired<<wbr>MachineLoopInfo>();<br>
> + AU.addRequired<<wbr>MachineBranchProbabilityInfo>(<wbr>);<br>
> AU.addPreserved<<wbr>MachineDominatorTree>();<br>
> AU.addPreserved<<wbr>MachinePostDominatorTree>();<br>
> AU.addPreserved<<wbr>MachineLoopInfo>();<br>
> @@ -163,6 +175,7 @@ char MachineSinking::ID = 0;<br>
> char &llvm::MachineSinkingID = MachineSinking::ID;<br>
> INITIALIZE_PASS_BEGIN(<wbr>MachineSinking, "machine-sink",<br>
> "Machine code sinking", false, false)<br>
> +INITIALIZE_PASS_DEPENDENCY(<wbr>MachineBranchProbabilityInfo)<br>
> INITIALIZE_PASS_DEPENDENCY(<wbr>MachineDominatorTree)<br>
> INITIALIZE_PASS_DEPENDENCY(<wbr>MachineLoopInfo)<br>
> INITIALIZE_PASS_DEPENDENCY(<wbr>AAResultsWrapperPass)<br>
> @@ -283,6 +296,7 @@ bool MachineSinking::<wbr>runOnMachineFunctio<br>
> PDT = &getAnalysis<<wbr>MachinePostDominatorTree>();<br>
> LI = &getAnalysis<MachineLoopInfo>(<wbr>);<br>
> MBFI = UseBlockFreqInfo ? &getAnalysis<<wbr>MachineBlockFrequencyInfo>() :<br>
> nullptr;<br>
> + MBPI = &getAnalysis<<wbr>MachineBranchProbabilityInfo>(<wbr>);<br>
> AA = &getAnalysis<<wbr>AAResultsWrapperPass>().<wbr>getAAResults();<br>
><br>
> bool EverMadeChange = false;<br>
> @@ -383,6 +397,10 @@ bool MachineSinking::<wbr>isWorthBreakingCrit<br>
> if (!MI.isCopy() && !TII->isAsCheapAsAMove(MI))<br>
> return true;<br>
><br>
> + if (From->isSuccessor(To) && MBPI->getEdgeProbability(From, To) <=<br>
> + BranchProbability(<wbr>SplitEdgeProbabilityThreshold, 100))<br>
> + return true;<br>
> +<br>
> // MI is cheap, we probably don't want to break the critical edge for<br>
> it.<br>
> // However, if this would allow some definitions of its source<br>
> operands<br>
> // to be sunk then it's probably worth it.<br>
><br>
> Modified: llvm/trunk/test/CodeGen/ARM/<wbr>atomic-cmpxchg.ll<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/test/<wbr>CodeGen/ARM/atomic-<br>
> cmpxchg.ll?rev=284757&r1=<wbr>284756&r2=284757&view=diff<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/test/CodeGen/ARM/<wbr>atomic-cmpxchg.ll (original)<br>
> +++ llvm/trunk/test/CodeGen/ARM/<wbr>atomic-cmpxchg.ll Thu Oct 20 13:06:52<br>
> 2016<br>
> @@ -38,16 +38,14 @@ entry:<br>
> ; CHECK-ARMV6-NEXT: uxtb [[DESIRED:r[0-9]+]], r1<br>
> ; CHECK-ARMV6-NEXT: [[TRY:.LBB[0-9_]+]]:<br>
> ; CHECK-ARMV6-NEXT: ldrexb [[LD:r[0-9]+]], [r0]<br>
> -; CHECK-ARMV6-NEXT: mov [[RES:r[0-9]+]], #0<br>
> ; CHECK-ARMV6-NEXT: cmp [[LD]], [[DESIRED]]<br>
> -; CHECK-ARMV6-NEXT: bne [[END:.LBB[0-9_]+]]<br>
> +; CHECK-ARMV6-NEXT: movne [[RES:r[0-9]+]], #0<br>
> +; CHECK-ARMV6-NEXT: bxne lr<br>
> ; CHECK-ARMV6-NEXT: strexb [[SUCCESS:r[0-9]+]], r2, [r0]<br>
> -; CHECK-ARMV6-NEXT: mov [[RES]], #1<br>
> ; CHECK-ARMV6-NEXT: cmp [[SUCCESS]], #0<br>
> -; CHECK-ARMV6-NEXT: bne [[TRY]]<br>
> -; CHECK-ARMV6-NEXT: [[END]]:<br>
> -; CHECK-ARMV6-NEXT: mov r0, [[RES]]<br>
> -; CHECK-ARMV6-NEXT: bx lr<br>
> +; CHECK-ARMV6-NEXT: moveq [[RES]], #1<br>
> +; CHECK-ARMV6-NEXT: bxeq lr<br>
> +; CHECK-ARMV6-NEXT: b [[TRY]]<br>
><br>
> ; CHECK-THUMBV6-LABEL: test_cmpxchg_res_i8:<br>
> ; CHECK-THUMBV6: mov [[EXPECTED:r[0-9]+]], r1<br>
> @@ -64,20 +62,18 @@ entry:<br>
> ; CHECK-ARMV7-LABEL: test_cmpxchg_res_i8:<br>
> ; CHECK-ARMV7-NEXT: .fnstart<br>
> ; CHECK-ARMV7-NEXT: uxtb [[DESIRED:r[0-9]+]], r1<br>
> -; CHECK-ARMV7-NEXT: [[TRY:.LBB[0-9_]+]]:<br>
> -; CHECK-ARMV7-NEXT: ldrexb [[LD:r[0-9]+]], [r0]<br>
> -; CHECK-ARMV7-NEXT: cmp [[LD]], [[DESIRED]]<br>
> -; CHECK-ARMV7-NEXT: bne [[FAIL:.LBB[0-9_]+]]<br>
> +; CHECK-ARMV7-NEXT: b [[TRY:.LBB[0-9_]+]]<br>
> +; CHECK-ARMV7-NEXT: [[HEAD:.LBB[0-9_]+]]:<br>
> ; CHECK-ARMV7-NEXT: strexb [[SUCCESS:r[0-9]+]], r2, [r0]<br>
> -; CHECK-ARMV7-NEXT: mov [[RES:r[0-9]+]], #1<br>
> ; CHECK-ARMV7-NEXT: cmp [[SUCCESS]], #0<br>
> -; CHECK-ARMV7-NEXT: bne [[TRY]]<br>
> -; CHECK-ARMV7-NEXT: b [[END:.LBB[0-9_]+]]<br>
> -; CHECK-ARMV7-NEXT: [[FAIL]]:<br>
> +; CHECK-ARMV7-NEXT: moveq [[RES:r[0-9]+]], #1<br>
> +; CHECK-ARMV7-NEXT: bxeq lr<br>
> +; CHECK-ARMV7-NEXT: [[TRY]]:<br>
> +; CHECK-ARMV7-NEXT: ldrexb [[LD:r[0-9]+]], [r0]<br>
> +; CHECK-ARMV7-NEXT: cmp [[LD]], [[DESIRED]]<br>
> +; CHECK-ARMV7-NEXT: beq [[HEAD]]<br>
> ; CHECK-ARMV7-NEXT: clrex<br>
> ; CHECK-ARMV7-NEXT: mov [[RES]], #0<br>
> -; CHECK-ARMV7-NEXT: [[END]]:<br>
> -; CHECK-ARMV7-NEXT: mov r0, [[RES]]<br>
> ; CHECK-ARMV7-NEXT: bx lr<br>
><br>
> ; CHECK-THUMBV7-LABEL: test_cmpxchg_res_i8:<br>
><br>
> Modified: llvm/trunk/test/CodeGen/ARM/<wbr>code-placement.ll<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/test/<wbr>CodeGen/ARM/code-<br>
> placement.ll?rev=284757&r1=<wbr>284756&r2=284757&view=diff<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/test/CodeGen/ARM/<wbr>code-placement.ll (original)<br>
> +++ llvm/trunk/test/CodeGen/ARM/<wbr>code-placement.ll Thu Oct 20 13:06:52<br>
> 2016<br>
> @@ -12,9 +12,9 @@ entry:<br>
> br i1 %0, label %bb2, label %bb<br>
><br>
> bb:<br>
> -; CHECK: LBB0_2:<br>
> -; CHECK: bne LBB0_2<br>
> -; CHECK-NOT: b LBB0_2<br>
> +; CHECK: LBB0_[[LABEL:[0-9]]]:<br>
> +; CHECK: bne LBB0_[[LABEL]]<br>
> +; CHECK-NOT: b LBB0_[[LABEL]]<br>
> ; CHECK: bx lr<br>
> %list_addr.05 = phi %struct.list_head* [ %2, %bb ], [ %list, %entry ]<br>
> %next.04 = phi %struct.list_head* [ %list_addr.05, %bb ], [ null,<br>
> %entry ]<br>
> @@ -34,14 +34,13 @@ bb2:<br>
> define i32 @t2(i32 %passes, i32* nocapture %src, i32 %size) nounwind<br>
> readonly {<br>
> entry:<br>
> ; CHECK-LABEL: t2:<br>
> -; CHECK: beq LBB1_[[RET:.]]<br>
> %0 = icmp eq i32 %passes, 0 ; <i1> [#uses=1]<br>
> br i1 %0, label %bb5, label %bb.nph15<br>
><br>
> -; CHECK: LBB1_[[PREHDR:.]]: @ %bb2.preheader<br>
> bb1: ; preds =<br>
> %bb2.preheader, %bb1<br>
> -; CHECK: LBB1_[[BB1:.]]: @ %bb1<br>
> -; CHECK: bne LBB1_[[BB1]]<br>
> +; CHECK: LBB1_[[BB3:.]]: @ %bb3<br>
> +; CHECK: LBB1_[[PREHDR:.]]: @ %bb2.preheader<br>
> +; CHECK: blt LBB1_[[BB3]]<br>
> %indvar = phi i32 [ %indvar.next, %bb1 ], [ 0, %bb2.preheader ] ;<br>
> <i32> [#uses=2]<br>
> %sum.08 = phi i32 [ %2, %bb1 ], [ %sum.110, %bb2.preheader ] ; <i32><br>
> [#uses=1]<br>
> %tmp17 = sub i32 %i.07, %indvar ; <i32> [#uses=1]<br>
> @@ -53,9 +52,9 @@ bb1:<br>
> br i1 %exitcond, label %bb3, label %bb1<br>
><br>
> bb3: ; preds = %bb1,<br>
> %bb2.preheader<br>
> -; CHECK: LBB1_[[BB3:.]]: @ %bb3<br>
> -; CHECK: bne LBB1_[[PREHDR]]<br>
> -; CHECK-NOT: b LBB1_<br>
> +; CHECK: LBB1_[[BB1:.]]: @ %bb1<br>
> +; CHECK: bne LBB1_[[BB1]]<br>
> +; CHECK: b LBB1_[[BB3]]<br>
> %sum.0.lcssa = phi i32 [ %sum.110, %bb2.preheader ], [ %2, %bb1 ] ;<br>
> <i32> [#uses=2]<br>
> %3 = add i32 %pass.011, 1 ; <i32> [#uses=2]<br>
> %exitcond18 = icmp eq i32 %3, %passes ; <i1> [#uses=1]<br>
> @@ -71,8 +70,6 @@ bb2.preheader:<br>
> %sum.110 = phi i32 [ 0, %bb.nph15 ], [ %sum.0.lcssa, %bb3 ] ; <i32><br>
> [#uses=2]<br>
> br i1 %4, label %bb1, label %bb3<br>
><br>
> -; CHECK: LBB1_[[RET]]: @ %bb5<br>
> -; CHECK: pop<br>
> bb5: ; preds = %bb3,<br>
> %entry<br>
> %sum.1.lcssa = phi i32 [ 0, %entry ], [ %sum.0.lcssa, %bb3 ] ; <i32><br>
> [#uses=1]<br>
> ret i32 %sum.1.lcssa<br>
><br>
> Modified: llvm/trunk/test/CodeGen/X86/<wbr>block-placement.ll<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/test/<wbr>CodeGen/X86/block-<br>
> placement.ll?rev=284757&r1=<wbr>284756&r2=284757&view=diff<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/test/CodeGen/X86/<wbr>block-placement.ll (original)<br>
> +++ llvm/trunk/test/CodeGen/X86/<wbr>block-placement.ll Thu Oct 20 13:06:52<br>
> 2016<br>
> @@ -478,12 +478,12 @@ define void @fpcmp_unanalyzable_branch(i<br>
> ; CHECK-LABEL: fpcmp_unanalyzable_branch:<br>
> ; CHECK: # BB#0: # %entry<br>
> ; CHECK: # BB#1: # %entry.if.then_crit_edge<br>
> -; CHECK: .LBB10_4: # %if.then<br>
> -; CHECK: .LBB10_5: # %if.end<br>
> +; CHECK: .LBB10_5: # %if.then<br>
> +; CHECK: .LBB10_6: # %if.end<br>
> ; CHECK: # BB#3: # %exit<br>
> ; CHECK: jne .LBB10_4<br>
> -; CHECK-NEXT: jnp .LBB10_5<br>
> -; CHECK-NEXT: jmp .LBB10_4<br>
> +; CHECK-NEXT: jnp .LBB10_6<br>
> +; CHECK: jmp .LBB10_5<br>
><br>
> entry:<br>
> ; Note that this branch must be strongly biased toward<br>
><br>
> Modified: llvm/trunk/test/CodeGen/X86/<wbr>clz.ll<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/test/<wbr>CodeGen/X86/clz.ll?rev=284757&<wbr>r1=284756&r2=28475<br>
> 7&view=diff<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/test/CodeGen/X86/<wbr>clz.ll (original)<br>
> +++ llvm/trunk/test/CodeGen/X86/<wbr>clz.ll Thu Oct 20 13:06:52 2016<br>
> @@ -279,28 +279,32 @@ define i64 @ctlz_i64(i64 %x) {<br>
> define i8 @ctlz_i8_zero_test(i8 %n) {<br>
> ; X32-LABEL: ctlz_i8_zero_test:<br>
> ; X32: # BB#0:<br>
> -; X32-NEXT: movb {{[0-9]+}}(%esp), %cl<br>
> -; X32-NEXT: movb $8, %al<br>
> -; X32-NEXT: testb %cl, %cl<br>
> -; X32-NEXT: je .LBB8_2<br>
> -; X32-NEXT: # BB#1: # %cond.false<br>
> -; X32-NEXT: movzbl %cl, %eax<br>
> +; X32-NEXT: movb {{[0-9]+}}(%esp), %al<br>
> +; X32-NEXT: testb %al, %al<br>
> +; X32-NEXT: je .LBB8_1<br>
> +; X32-NEXT: # BB#2: # %cond.false<br>
> +; X32-NEXT: movzbl %al, %eax<br>
> ; X32-NEXT: bsrl %eax, %eax<br>
> ; X32-NEXT: xorl $7, %eax<br>
> -; X32-NEXT: .LBB8_2: # %cond.end<br>
> +; X32-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill><br>
> +; X32-NEXT: retl<br>
> +; X32-NEXT: .LBB8_1:<br>
> +; X32-NEXT: movb $8, %al<br>
> ; X32-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill><br>
> ; X32-NEXT: retl<br>
> ;<br>
> ; X64-LABEL: ctlz_i8_zero_test:<br>
> ; X64: # BB#0:<br>
> -; X64-NEXT: movb $8, %al<br>
> ; X64-NEXT: testb %dil, %dil<br>
> -; X64-NEXT: je .LBB8_2<br>
> -; X64-NEXT: # BB#1: # %cond.false<br>
> +; X64-NEXT: je .LBB8_1<br>
> +; X64-NEXT: # BB#2: # %cond.false<br>
> ; X64-NEXT: movzbl %dil, %eax<br>
> ; X64-NEXT: bsrl %eax, %eax<br>
> ; X64-NEXT: xorl $7, %eax<br>
> -; X64-NEXT: .LBB8_2: # %cond.end<br>
> +; X64-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill><br>
> +; X64-NEXT: retq<br>
> +; X64-NEXT: .LBB8_1:<br>
> +; X64-NEXT: movb $8, %al<br>
> ; X64-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill><br>
> ; X64-NEXT: retq<br>
> ;<br>
> @@ -327,26 +331,30 @@ define i8 @ctlz_i8_zero_test(i8 %n) {<br>
> define i16 @ctlz_i16_zero_test(i16 %n) {<br>
> ; X32-LABEL: ctlz_i16_zero_test:<br>
> ; X32: # BB#0:<br>
> -; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx<br>
> -; X32-NEXT: movw $16, %ax<br>
> -; X32-NEXT: testw %cx, %cx<br>
> -; X32-NEXT: je .LBB9_2<br>
> -; X32-NEXT: # BB#1: # %cond.false<br>
> -; X32-NEXT: bsrw %cx, %ax<br>
> +; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax<br>
> +; X32-NEXT: testw %ax, %ax<br>
> +; X32-NEXT: je .LBB9_1<br>
> +; X32-NEXT: # BB#2: # %cond.false<br>
> +; X32-NEXT: bsrw %ax, %ax<br>
> ; X32-NEXT: xorl $15, %eax<br>
> -; X32-NEXT: .LBB9_2: # %cond.end<br>
> +; X32-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill><br>
> +; X32-NEXT: retl<br>
> +; X32-NEXT: .LBB9_1:<br>
> +; X32-NEXT: movw $16, %ax<br>
> ; X32-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill><br>
> ; X32-NEXT: retl<br>
> ;<br>
> ; X64-LABEL: ctlz_i16_zero_test:<br>
> ; X64: # BB#0:<br>
> -; X64-NEXT: movw $16, %ax<br>
> ; X64-NEXT: testw %di, %di<br>
> -; X64-NEXT: je .LBB9_2<br>
> -; X64-NEXT: # BB#1: # %cond.false<br>
> +; X64-NEXT: je .LBB9_1<br>
> +; X64-NEXT: # BB#2: # %cond.false<br>
> ; X64-NEXT: bsrw %di, %ax<br>
> ; X64-NEXT: xorl $15, %eax<br>
> -; X64-NEXT: .LBB9_2: # %cond.end<br>
> +; X64-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill><br>
> +; X64-NEXT: retq<br>
> +; X64-NEXT: .LBB9_1:<br>
> +; X64-NEXT: movw $16, %ax<br>
> ; X64-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill><br>
> ; X64-NEXT: retq<br>
> ;<br>
> @@ -367,25 +375,27 @@ define i16 @ctlz_i16_zero_test(i16 %n) {<br>
> define i32 @ctlz_i32_zero_test(i32 %n) {<br>
> ; X32-LABEL: ctlz_i32_zero_test:<br>
> ; X32: # BB#0:<br>
> -; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx<br>
> -; X32-NEXT: movl $32, %eax<br>
> -; X32-NEXT: testl %ecx, %ecx<br>
> -; X32-NEXT: je .LBB10_2<br>
> -; X32-NEXT: # BB#1: # %cond.false<br>
> -; X32-NEXT: bsrl %ecx, %eax<br>
> +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax<br>
> +; X32-NEXT: testl %eax, %eax<br>
> +; X32-NEXT: je .LBB10_1<br>
> +; X32-NEXT: # BB#2: # %cond.false<br>
> +; X32-NEXT: bsrl %eax, %eax<br>
> ; X32-NEXT: xorl $31, %eax<br>
> -; X32-NEXT: .LBB10_2: # %cond.end<br>
> +; X32-NEXT: retl<br>
> +; X32-NEXT: .LBB10_1:<br>
> +; X32-NEXT: movl $32, %eax<br>
> ; X32-NEXT: retl<br>
> ;<br>
> ; X64-LABEL: ctlz_i32_zero_test:<br>
> ; X64: # BB#0:<br>
> -; X64-NEXT: movl $32, %eax<br>
> ; X64-NEXT: testl %edi, %edi<br>
> -; X64-NEXT: je .LBB10_2<br>
> -; X64-NEXT: # BB#1: # %cond.false<br>
> +; X64-NEXT: je .LBB10_1<br>
> +; X64-NEXT: # BB#2: # %cond.false<br>
> ; X64-NEXT: bsrl %edi, %eax<br>
> ; X64-NEXT: xorl $31, %eax<br>
> -; X64-NEXT: .LBB10_2: # %cond.end<br>
> +; X64-NEXT: retq<br>
> +; X64-NEXT: .LBB10_1:<br>
> +; X64-NEXT: movl $32, %eax<br>
> ; X64-NEXT: retq<br>
> ;<br>
> ; X32-CLZ-LABEL: ctlz_i32_zero_test:<br>
> @@ -464,26 +474,30 @@ define i64 @ctlz_i64_zero_test(i64 %n) {<br>
> define i8 @cttz_i8_zero_test(i8 %n) {<br>
> ; X32-LABEL: cttz_i8_zero_test:<br>
> ; X32: # BB#0:<br>
> -; X32-NEXT: movb {{[0-9]+}}(%esp), %cl<br>
> -; X32-NEXT: movb $8, %al<br>
> -; X32-NEXT: testb %cl, %cl<br>
> -; X32-NEXT: je .LBB12_2<br>
> -; X32-NEXT: # BB#1: # %cond.false<br>
> -; X32-NEXT: movzbl %cl, %eax<br>
> +; X32-NEXT: movb {{[0-9]+}}(%esp), %al<br>
> +; X32-NEXT: testb %al, %al<br>
> +; X32-NEXT: je .LBB12_1<br>
> +; X32-NEXT: # BB#2: # %cond.false<br>
> +; X32-NEXT: movzbl %al, %eax<br>
> ; X32-NEXT: bsfl %eax, %eax<br>
> -; X32-NEXT: .LBB12_2: # %cond.end<br>
> +; X32-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill><br>
> +; X32-NEXT: retl<br>
> +; X32-NEXT: .LBB12_1<br>
> +; X32-NEXT: movb $8, %al<br>
> ; X32-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill><br>
> ; X32-NEXT: retl<br>
> ;<br>
> ; X64-LABEL: cttz_i8_zero_test:<br>
> ; X64: # BB#0:<br>
> -; X64-NEXT: movb $8, %al<br>
> ; X64-NEXT: testb %dil, %dil<br>
> -; X64-NEXT: je .LBB12_2<br>
> -; X64-NEXT: # BB#1: # %cond.false<br>
> +; X64-NEXT: je .LBB12_1<br>
> +; X64-NEXT: # BB#2: # %cond.false<br>
> ; X64-NEXT: movzbl %dil, %eax<br>
> ; X64-NEXT: bsfl %eax, %eax<br>
> -; X64-NEXT: .LBB12_2: # %cond.end<br>
> +; X64-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill><br>
> +; X64-NEXT: retq<br>
> +; X64-NEXT: .LBB12_1:<br>
> +; X64-NEXT: movb $8, %al<br>
> ; X64-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill><br>
> ; X64-NEXT: retq<br>
> ;<br>
> @@ -510,23 +524,25 @@ define i8 @cttz_i8_zero_test(i8 %n) {<br>
> define i16 @cttz_i16_zero_test(i16 %n) {<br>
> ; X32-LABEL: cttz_i16_zero_test:<br>
> ; X32: # BB#0:<br>
> -; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx<br>
> +; X32-NEXT: movzwl {{[0-9]+}}(%esp), %eax<br>
> +; X32-NEXT: testw %ax, %ax<br>
> +; X32-NEXT: je .LBB13_1<br>
> +; X32-NEXT: # BB#2: # %cond.false<br>
> +; X32-NEXT: bsfw %ax, %ax<br>
> +; X32-NEXT: retl<br>
> +; X32-NEXT: .LBB13_1<br>
> ; X32-NEXT: movw $16, %ax<br>
> -; X32-NEXT: testw %cx, %cx<br>
> -; X32-NEXT: je .LBB13_2<br>
> -; X32-NEXT: # BB#1: # %cond.false<br>
> -; X32-NEXT: bsfw %cx, %ax<br>
> -; X32-NEXT: .LBB13_2: # %cond.end<br>
> ; X32-NEXT: retl<br>
> ;<br>
> ; X64-LABEL: cttz_i16_zero_test:<br>
> ; X64: # BB#0:<br>
> -; X64-NEXT: movw $16, %ax<br>
> ; X64-NEXT: testw %di, %di<br>
> -; X64-NEXT: je .LBB13_2<br>
> -; X64-NEXT: # BB#1: # %cond.false<br>
> +; X64-NEXT: je .LBB13_1<br>
> +; X64-NEXT: # BB#2: # %cond.false<br>
> ; X64-NEXT: bsfw %di, %ax<br>
> -; X64-NEXT: .LBB13_2: # %cond.end<br>
> +; X64-NEXT: retq<br>
> +; X64-NEXT: .LBB13_1:<br>
> +; X64-NEXT: movw $16, %ax<br>
> ; X64-NEXT: retq<br>
> ;<br>
> ; X32-CLZ-LABEL: cttz_i16_zero_test:<br>
> @@ -546,23 +562,25 @@ define i16 @cttz_i16_zero_test(i16 %n) {<br>
> define i32 @cttz_i32_zero_test(i32 %n) {<br>
> ; X32-LABEL: cttz_i32_zero_test:<br>
> ; X32: # BB#0:<br>
> -; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx<br>
> +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax<br>
> +; X32-NEXT: testl %eax, %eax<br>
> +; X32-NEXT: je .LBB14_1<br>
> +; X32-NEXT: # BB#2: # %cond.false<br>
> +; X32-NEXT: bsfl %eax, %eax<br>
> +; X32-NEXT: retl<br>
> +; X32-NEXT: .LBB14_1<br>
> ; X32-NEXT: movl $32, %eax<br>
> -; X32-NEXT: testl %ecx, %ecx<br>
> -; X32-NEXT: je .LBB14_2<br>
> -; X32-NEXT: # BB#1: # %cond.false<br>
> -; X32-NEXT: bsfl %ecx, %eax<br>
> -; X32-NEXT: .LBB14_2: # %cond.end<br>
> ; X32-NEXT: retl<br>
> ;<br>
> ; X64-LABEL: cttz_i32_zero_test:<br>
> ; X64: # BB#0:<br>
> -; X64-NEXT: movl $32, %eax<br>
> ; X64-NEXT: testl %edi, %edi<br>
> -; X64-NEXT: je .LBB14_2<br>
> -; X64-NEXT: # BB#1: # %cond.false<br>
> +; X64-NEXT: je .LBB14_1<br>
> +; X64-NEXT: # BB#2: # %cond.false<br>
> ; X64-NEXT: bsfl %edi, %eax<br>
> -; X64-NEXT: .LBB14_2: # %cond.end<br>
> +; X64-NEXT: retq<br>
> +; X64-NEXT: .LBB14_1:<br>
> +; X64-NEXT: movl $32, %eax<br>
> ; X64-NEXT: retq<br>
> ;<br>
> ; X32-CLZ-LABEL: cttz_i32_zero_test:<br>
> @@ -642,25 +660,27 @@ define i64 @cttz_i64_zero_test(i64 %n) {<br>
> define i32 @ctlz_i32_fold_cmov(i32 %n) {<br>
> ; X32-LABEL: ctlz_i32_fold_cmov:<br>
> ; X32: # BB#0:<br>
> -; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx<br>
> -; X32-NEXT: orl $1, %ecx<br>
> -; X32-NEXT: movl $32, %eax<br>
> -; X32-NEXT: je .LBB16_2<br>
> -; X32-NEXT: # BB#1: # %cond.false<br>
> -; X32-NEXT: bsrl %ecx, %eax<br>
> +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax<br>
> +; X32-NEXT: orl $1, %eax<br>
> +; X32-NEXT: je .LBB16_1<br>
> +; X32-NEXT: # BB#2: # %cond.false<br>
> +; X32-NEXT: bsrl %eax, %eax<br>
> ; X32-NEXT: xorl $31, %eax<br>
> -; X32-NEXT: .LBB16_2: # %cond.end<br>
> +; X32-NEXT: retl<br>
> +; X32-NEXT: .LBB16_1<br>
> +; X32-NEXT: movl $32, %eax<br>
> ; X32-NEXT: retl<br>
> ;<br>
> ; X64-LABEL: ctlz_i32_fold_cmov:<br>
> ; X64: # BB#0:<br>
> ; X64-NEXT: orl $1, %edi<br>
> -; X64-NEXT: movl $32, %eax<br>
> -; X64-NEXT: je .LBB16_2<br>
> -; X64-NEXT: # BB#1: # %cond.false<br>
> +; X64-NEXT: je .LBB16_1<br>
> +; X64-NEXT: # BB#2: # %cond.false<br>
> ; X64-NEXT: bsrl %edi, %eax<br>
> ; X64-NEXT: xorl $31, %eax<br>
> -; X64-NEXT: .LBB16_2: # %cond.end<br>
> +; X64-NEXT: retq<br>
> +; X64-NEXT: .LBB16_1:<br>
> +; X64-NEXT: movl $32, %eax<br>
> ; X64-NEXT: retq<br>
> ;<br>
> ; X32-CLZ-LABEL: ctlz_i32_fold_cmov:<br>
> @@ -716,26 +736,30 @@ define i32 @ctlz_bsr(i32 %n) {<br>
> define i32 @ctlz_bsr_zero_test(i32 %n) {<br>
> ; X32-LABEL: ctlz_bsr_zero_test:<br>
> ; X32: # BB#0:<br>
> -; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx<br>
> -; X32-NEXT: movl $32, %eax<br>
> -; X32-NEXT: testl %ecx, %ecx<br>
> -; X32-NEXT: je .LBB18_2<br>
> -; X32-NEXT: # BB#1: # %cond.false<br>
> -; X32-NEXT: bsrl %ecx, %eax<br>
> +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax<br>
> +; X32-NEXT: testl %eax, %eax<br>
> +; X32-NEXT: je .LBB18_1<br>
> +; X32-NEXT: # BB#2: # %cond.false<br>
> +; X32-NEXT: bsrl %eax, %eax<br>
> ; X32-NEXT: xorl $31, %eax<br>
> -; X32-NEXT: .LBB18_2: # %cond.end<br>
> +; X32-NEXT: xorl $31, %eax<br>
> +; X32-NEXT: retl<br>
> +; X32-NEXT: .LBB18_1:<br>
> +; X32-NEXT: movl $32, %eax<br>
> ; X32-NEXT: xorl $31, %eax<br>
> ; X32-NEXT: retl<br>
> ;<br>
> ; X64-LABEL: ctlz_bsr_zero_test:<br>
> ; X64: # BB#0:<br>
> -; X64-NEXT: movl $32, %eax<br>
> ; X64-NEXT: testl %edi, %edi<br>
> -; X64-NEXT: je .LBB18_2<br>
> -; X64-NEXT: # BB#1: # %cond.false<br>
> +; X64-NEXT: je .LBB18_1<br>
> +; X64-NEXT: # BB#2: # %cond.false<br>
> ; X64-NEXT: bsrl %edi, %eax<br>
> ; X64-NEXT: xorl $31, %eax<br>
> -; X64-NEXT: .LBB18_2: # %cond.end<br>
> +; X64-NEXT: xorl $31, %eax<br>
> +; X64-NEXT: retq<br>
> +; X64-NEXT: .LBB18_1:<br>
> +; X64-NEXT: movl $32, %eax<br>
> ; X64-NEXT: xorl $31, %eax<br>
> ; X64-NEXT: retq<br>
> ;<br>
><br>
> Modified: llvm/trunk/test/CodeGen/X86/<wbr>loop-search.ll<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/test/<wbr>CodeGen/X86/loop-<br>
> search.ll?rev=284757&r1=<wbr>284756&r2=284757&view=diff<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/test/CodeGen/X86/<wbr>loop-search.ll (original)<br>
> +++ llvm/trunk/test/CodeGen/X86/<wbr>loop-search.ll Thu Oct 20 13:06:52 2016<br>
> @@ -10,19 +10,17 @@ define zeroext i1 @search(i32 %needle, i<br>
> ; CHECK-NEXT: testl %edx, %edx<br>
> ; CHECK-NEXT: jle LBB0_1<br>
> ; CHECK-NEXT: ## BB#4: ## %for.body.preheader<br>
> -; CHECK-NEXT: movslq %edx, %rcx<br>
> -; CHECK-NEXT: xorl %edx, %edx<br>
> +; CHECK-NEXT: movslq %edx, %rax<br>
> +; CHECK-NEXT: xorl %ecx, %ecx<br>
> ; CHECK-NEXT: .p2align 4, 0x90<br>
> ; CHECK-NEXT: LBB0_5: ## %for.body<br>
> ; CHECK-NEXT: ## =>This Inner Loop Header: Depth=1<br>
> -; ### FIXME: This loop invariant should be hoisted<br>
> -; CHECK-NEXT: movb $1, %al<br>
> -; CHECK-NEXT: cmpl %edi, (%rsi,%rdx,4)<br>
> +; CHECK-NEXT: cmpl %edi, (%rsi,%rcx,4)<br>
> ; CHECK-NEXT: je LBB0_6<br>
> ; CHECK-NEXT: ## BB#2: ## %for.cond<br>
> ; CHECK-NEXT: ## in Loop: Header=BB0_5 Depth=1<br>
> -; CHECK-NEXT: incq %rdx<br>
> -; CHECK-NEXT: cmpq %rcx, %rdx<br>
> +; CHECK-NEXT: incq %rcx<br>
> +; CHECK-NEXT: cmpq %rax, %rcx<br>
> ; CHECK-NEXT: jl LBB0_5<br>
> ; ### FIXME: BB#3 and LBB0_1 should be merged<br>
> ; CHECK-NEXT: ## BB#3:<br>
> @@ -33,7 +31,8 @@ define zeroext i1 @search(i32 %needle, i<br>
> ; CHECK-NEXT: xorl %eax, %eax<br>
> ; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill><br>
> ; CHECK-NEXT: retq<br>
> -; CHECK-NEXT: LBB0_6: ## %cleanup<br>
> +; CHECK-NEXT: LBB0_6:<br>
> +; CHECK-NEXT: movb $1, %al<br>
> ; CHECK-NEXT: ## kill: %AL<def> %AL<kill> %EAX<kill><br>
> ; CHECK-NEXT: retq<br>
> ;<br>
><br>
> Added: llvm/trunk/test/CodeGen/X86/<wbr>machine-sink.ll<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/test/<wbr>CodeGen/X86/machine-sink.ll?<wbr>rev=284757&view=auto<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/test/CodeGen/X86/<wbr>machine-sink.ll (added)<br>
> +++ llvm/trunk/test/CodeGen/X86/<wbr>machine-sink.ll Thu Oct 20 13:06:52 2016<br>
> @@ -0,0 +1,21 @@<br>
> +; RUN: llc < %s -mtriple=x86_64-pc-linux | FileCheck %s<br>
> +<br>
> +; Checks if movl $1 is sinked to critical edge.<br>
> +; CHECK-NOT: movl $1<br>
> +; CHECK: jbe<br>
> +; CHECK: movl $1<br>
> +define i32 @test(i32 %n, i32 %k) nounwind {<br>
> +entry:<br>
> + %cmp = icmp ugt i32 %k, %n<br>
> + br i1 %cmp, label %ifthen, label %ifend, !prof !1<br>
> +<br>
> +ifthen:<br>
> + %y = add i32 %k, 2<br>
> + br label %ifend<br>
> +<br>
> +ifend:<br>
> + %ret = phi i32 [ 1, %entry ] , [ %y, %ifthen]<br>
> + ret i32 %ret<br>
> +}<br>
> +<br>
> +!1 = !{!"branch_weights", i32 100, i32 1}<br>
><br>
> Modified: llvm/trunk/test/CodeGen/X86/<wbr>phys_subreg_coalesce-2.ll<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/test/<wbr>CodeGen/X86/phys_subreg_<wbr>coalesce-<br>
> 2.ll?rev=284757&r1=284756&r2=<wbr>284757&view=diff<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/test/CodeGen/X86/<wbr>phys_subreg_coalesce-2.ll (original)<br>
> +++ llvm/trunk/test/CodeGen/X86/<wbr>phys_subreg_coalesce-2.ll Thu Oct 20<br>
> 13:06:52 2016<br>
> @@ -14,7 +14,9 @@ forcond.preheader: ; preds = %entry<br>
> ifthen: ; preds = %entry<br>
> ret i32 0<br>
> ; CHECK: forbody{{$}}<br>
> +; There should be no mov instruction in the for body.<br>
> ; CHECK-NOT: mov<br>
> +; CHECK: jbe<br>
> forbody: ; preds = %forbody, %forcond.preheader<br>
> %indvar = phi i32 [ 0, %forcond.preheader ], [ %divisor.02,<br>
> %forbody ] ; <i32> [#uses=3]<br>
> %accumulator.01 = phi i32 [ 1, %forcond.preheader ], [ %div,<br>
> %forbody ] ; <i32> [#uses=1]<br>
><br>
> Modified: llvm/trunk/test/CodeGen/X86/<wbr>pr2659.ll<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/test/<wbr>CodeGen/X86/pr2659.ll?rev=<wbr>284757&r1=284756&r2=28<br>
> 4757&view=diff<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/test/CodeGen/X86/<wbr>pr2659.ll (original)<br>
> +++ llvm/trunk/test/CodeGen/X86/<wbr>pr2659.ll Thu Oct 20 13:06:52 2016<br>
> @@ -14,7 +14,7 @@ forcond.preheader: ; preds<br>
> br i1 %cmp44, label %afterfor, label %forbody<br>
><br>
> ; CHECK: %forcond.preheader<br>
> -; CHECK: movl $1<br>
> +; CHECK: testl<br>
> ; CHECK-NOT: xorl<br>
> ; CHECK-NOT: movl<br>
> ; CHECK-NOT: LBB<br>
> @@ -24,6 +24,7 @@ forcond.preheader: ; preds<br>
> ; CHECK: %forbody{{$}}<br>
> ; CHECK-NOT: mov<br>
> ; CHECK: jbe<br>
> +; CHECK: movl $1<br>
><br>
> ifthen: ; preds = %entry<br>
> ret i32 0<br>
><br>
> Modified: llvm/trunk/test/DebugInfo/<wbr>COFF/pieces.ll<br>
> URL: <a href="http://llvm.org/viewvc/llvm-" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-</a><br>
> project/llvm/trunk/test/<wbr>DebugInfo/COFF/pieces.ll?rev=<wbr>284757&r1=284756&r2<br>
> =284757&view=diff<br>
> ==============================<wbr>==============================<wbr>============<br>
> ======<br>
> --- llvm/trunk/test/DebugInfo/<wbr>COFF/pieces.ll (original)<br>
> +++ llvm/trunk/test/DebugInfo/<wbr>COFF/pieces.ll Thu Oct 20 13:06:52 2016<br>
> @@ -37,11 +37,11 @@<br>
> ; ASM-LABEL: loop_csr: # @loop_csr<br>
> ; ASM: #DEBUG_VALUE: loop_csr:o [bit_piece offset=0 size=32] <-<br>
> 0<br>
> ; ASM: #DEBUG_VALUE: loop_csr:o [bit_piece offset=32 size=32] <-<br>
> 0<br>
> -; ASM: # BB#1: # %for.body.preheader<br>
> +; ASM: # BB#2: # %for.body.preheader<br>
> ; ASM: xorl %edi, %edi<br>
> ; ASM: xorl %esi, %esi<br>
> ; ASM: .p2align 4, 0x90<br>
> -; ASM: .LBB0_2: # %for.body<br>
> +; ASM: .LBB0_3: # %for.body<br>
> ; ASM: [[ox_start:\.Ltmp[0-9]+]]:<br>
> ; ASM: #DEBUG_VALUE: loop_csr:o [bit_piece offset=0 size=32] <-<br>
> %EDI<br>
> ; ASM: .cv_loc 0 1 13 11 # t.c:13:11<br>
> @@ -57,7 +57,7 @@<br>
> ; ASM: movl %eax, %esi<br>
> ; ASM: #DEBUG_VALUE: loop_csr:o [bit_piece offset=32 size=32]<br>
> <- %ESI<br>
> ; ASM: cmpl n(%rip), %eax<br>
> -; ASM: jl .LBB0_2<br>
> +; ASM: jl .LBB0_3<br>
> ; ASM: [[oy_end:\.Ltmp[0-9]+]]:<br>
> ; ASM: addl %edi, %esi<br>
> ; ASM: movl %esi, %eax<br>
><br>
><br>
> ______________________________<wbr>_________________<br>
> llvm-commits mailing list<br>
> <a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a><br>
> <a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/<wbr>mailman/listinfo/llvm-commits</a><br>
</div></div></blockquote></div><br></div>