replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
hanchenye-llvm-project/llvm/test/Transforms/LICM/scalar_promote.ll

194 lines
5.3 KiB
LLVM
Raw Normal View History

Fix LICM's memory promotion optimization to preserve TBAA tags when promoting a store in a loop. This was noticed when working on PR14753, but isn't directly related. llvm-svn: 171281
2012-12-31 16:37:17 +08:00
; RUN: opt < %s -basicaa -tbaa -licm -S | FileCheck %s
licm is wasting time hoisting constant foldable operations, instead of hoisting them, just fold them away. This occurs in the testcase for PR8041, for example. llvm-svn: 112669
2010-09-01 07:00:16 +08:00
target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128"
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
@X = global i32 7 ; <i32*> [#uses=4]
New testcase to check to see if LICM is performing scalar promotion llvm-svn: 5607
2003-02-23 06:20:19 +08:00
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
define void @test1(i32 %i) {
Entry:
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
br label %Loop
Update Transforms tests to use CHECK-LABEL for easier debugging. No functionality change. This update was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_]*\):\( *\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186268
2013-07-14 09:42:54 +08:00
; CHECK-LABEL: @test1(
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
; CHECK: Entry:
; CHECK-NEXT: load i32* @X
; CHECK-NEXT: br label %Loop
New testcase to check to see if LICM is performing scalar promotion llvm-svn: 5607
2003-02-23 06:20:19 +08:00
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
Loop: ; preds = %Loop, %0
%j = phi i32 [ 0, %Entry ], [ %Next, %Loop ] ; <i32> [#uses=1]
%x = load i32* @X ; <i32> [#uses=1]
%x2 = add i32 %x, 1 ; <i32> [#uses=1]
store i32 %x2, i32* @X
%Next = add i32 %j, 1 ; <i32> [#uses=2]
%cond = icmp eq i32 %Next, 0 ; <i1> [#uses=1]
br i1 %cond, label %Out, label %Loop
New testcase to check to see if LICM is performing scalar promotion llvm-svn: 5607
2003-02-23 06:20:19 +08:00
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
Out:
ret void
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
; CHECK: Out:
[LPM] Make LCSSA a utility with a FunctionPass that applies it to all the loops in a function, and teach LICM to work in the presance of LCSSA. Previously, LCSSA was a loop pass. That made passes requiring it also be loop passes and unable to depend on function analysis passes easily. It also caused outer loops to have a different "canonical" form from inner loops during analysis. Instead, we go into LCSSA form and preserve it through the loop pass manager run. Note that this has the same problem as LoopSimplify that prevents enabling its verification -- loop passes which run at the end of the loop pass manager and don't preserve these are valid, but the subsequent loop pass runs of outer loops that do preserve this pass trigger too much verification and fail because the inner loop no longer verifies. The other problem this exposed is that LICM was completely unable to handle LCSSA form. It didn't preserve it and it actually would give up on moving instructions in many cases when they were used by an LCSSA phi node. I've taught LICM to support detecting LCSSA-form PHI nodes and to hoist and sink around them. This may actually let LICM fire significantly more because we put everything into LCSSA form to rotate the loop before running LICM. =/ Now LICM should handle that fine and preserve it correctly. The down side is that LICM has to require LCSSA in order to preserve it. This is just a fact of life for LCSSA. It's entirely possible we should completely remove LCSSA from the optimizer. The test updates are essentially accomodating LCSSA phi nodes in the output of LICM, and the fact that we now completely sink every instruction in ashr-crash below the loop bodies prior to unrolling. With this change, LCSSA is computed only three times in the pass pipeline. One of them could be removed (and potentially a SCEV run and a separate LoopPassManager entirely!) if we had a LoopPass variant of InstCombine that ran InstCombine on the loop body but refused to combine away LCSSA PHI nodes. Currently, this also prevents loop unrolling from being in the same loop pass manager is rotate, LICM, and unswitch. There is one thing that I *really* don't like -- preserving LCSSA in LICM is quite expensive. We end up having to re-run LCSSA twice for some loops after LICM runs because LICM can undo LCSSA both in the current loop and the parent loop. I don't really see good solutions to this other than to completely move away from LCSSA and using tools like SSAUpdater instead. llvm-svn: 200067
2014-01-25 12:07:24 +08:00
; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2
; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* @X
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
; CHECK-NEXT: ret void
New testcase to check to see if LICM is performing scalar promotion llvm-svn: 5607
2003-02-23 06:20:19 +08:00
}
New testcase for scalar promotion llvm-svn: 5611
2003-02-24 11:52:04 +08:00
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
define void @test2(i32 %i) {
Entry:
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
br label %Loop
Update Transforms tests to use CHECK-LABEL for easier debugging. No functionality change. This update was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_]*\):\( *\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186268
2013-07-14 09:42:54 +08:00
; CHECK-LABEL: @test2(
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
; CHECK: Entry:
licm is wasting time hoisting constant foldable operations, instead of hoisting them, just fold them away. This occurs in the testcase for PR8041, for example. llvm-svn: 112669
2010-09-01 07:00:16 +08:00
; CHECK-NEXT: %.promoted = load i32* getelementptr inbounds (i32* @X, i64 1)
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
; CHECK-NEXT: br label %Loop
New testcase for scalar promotion llvm-svn: 5611
2003-02-24 11:52:04 +08:00
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
Loop: ; preds = %Loop, %0
%X1 = getelementptr i32* @X, i64 1 ; <i32*> [#uses=1]
%A = load i32* %X1 ; <i32> [#uses=1]
%V = add i32 %A, 1 ; <i32> [#uses=1]
%X2 = getelementptr i32* @X, i64 1 ; <i32*> [#uses=1]
store i32 %V, i32* %X2
br i1 false, label %Loop, label %Exit
New testcase for scalar promotion llvm-svn: 5611
2003-02-24 11:52:04 +08:00
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
Exit: ; preds = %Loop
ret void
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
; CHECK: Exit:
[LPM] Make LCSSA a utility with a FunctionPass that applies it to all the loops in a function, and teach LICM to work in the presance of LCSSA. Previously, LCSSA was a loop pass. That made passes requiring it also be loop passes and unable to depend on function analysis passes easily. It also caused outer loops to have a different "canonical" form from inner loops during analysis. Instead, we go into LCSSA form and preserve it through the loop pass manager run. Note that this has the same problem as LoopSimplify that prevents enabling its verification -- loop passes which run at the end of the loop pass manager and don't preserve these are valid, but the subsequent loop pass runs of outer loops that do preserve this pass trigger too much verification and fail because the inner loop no longer verifies. The other problem this exposed is that LICM was completely unable to handle LCSSA form. It didn't preserve it and it actually would give up on moving instructions in many cases when they were used by an LCSSA phi node. I've taught LICM to support detecting LCSSA-form PHI nodes and to hoist and sink around them. This may actually let LICM fire significantly more because we put everything into LCSSA form to rotate the loop before running LICM. =/ Now LICM should handle that fine and preserve it correctly. The down side is that LICM has to require LCSSA in order to preserve it. This is just a fact of life for LCSSA. It's entirely possible we should completely remove LCSSA from the optimizer. The test updates are essentially accomodating LCSSA phi nodes in the output of LICM, and the fact that we now completely sink every instruction in ashr-crash below the loop bodies prior to unrolling. With this change, LCSSA is computed only three times in the pass pipeline. One of them could be removed (and potentially a SCEV run and a separate LoopPassManager entirely!) if we had a LoopPass variant of InstCombine that ran InstCombine on the loop body but refused to combine away LCSSA PHI nodes. Currently, this also prevents loop unrolling from being in the same loop pass manager is rotate, LICM, and unswitch. There is one thing that I *really* don't like -- preserving LCSSA in LICM is quite expensive. We end up having to re-run LCSSA twice for some loops after LICM runs because LICM can undo LCSSA both in the current loop and the parent loop. I don't really see good solutions to this other than to completely move away from LCSSA and using tools like SSAUpdater instead. llvm-svn: 200067
2014-01-25 12:07:24 +08:00
; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %V
; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* getelementptr inbounds (i32* @X, i64 1)
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
; CHECK-NEXT: ret void
New testcase for scalar promotion llvm-svn: 5611
2003-02-24 11:52:04 +08:00
}
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
define void @test3(i32 %i) {
Update Transforms tests to use CHECK-LABEL for easier debugging. No functionality change. This update was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_]*\):\( *\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186268
2013-07-14 09:42:54 +08:00
; CHECK-LABEL: @test3(
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
br label %Loop
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
Loop:
; Should not promote this to a register
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
%x = load volatile i32* @X
%x2 = add i32 %x, 1
store i32 %x2, i32* @X
br i1 true, label %Out, label %Loop
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
; CHECK: Loop:
Move "atomic" and "volatile" designations on instructions after the opcode of the instruction. Note that this change affects the existing non-atomic load and store instructions; the parser now accepts both forms, and the change is noted in the release notes. llvm-svn: 137527
2011-08-13 06:50:01 +08:00
; CHECK-NEXT: load volatile
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
Out: ; preds = %Loop
ret void
convert scalar_promote to filecheck style and merge 2003-12-13-VolatilePromote.ll into it. llvm-svn: 80521
2009-08-31 06:08:19 +08:00
}
fix more AST updating bugs, correcting miscompilation in PR8041 llvm-svn: 112878
2010-09-03 06:19:10 +08:00
; PR8041
define void @test4(i8* %x, i8 %n) {
Update Transforms tests to use CHECK-LABEL for easier debugging. No functionality change. This update was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_]*\):\( *\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186268
2013-07-14 09:42:54 +08:00
; CHECK-LABEL: @test4(
fix more AST updating bugs, correcting miscompilation in PR8041 llvm-svn: 112878
2010-09-03 06:19:10 +08:00
%handle1 = alloca i8*
%handle2 = alloca i8*
store i8* %x, i8** %handle1
br label %loop
loop:
%tmp = getelementptr i8* %x, i64 8
store i8* %tmp, i8** %handle2
br label %subloop
subloop:
%count = phi i8 [ 0, %loop ], [ %nextcount, %subloop ]
%offsetx2 = load i8** %handle2
store i8 %n, i8* %offsetx2
%newoffsetx2 = getelementptr i8* %offsetx2, i64 -1
store i8* %newoffsetx2, i8** %handle2
%nextcount = add i8 %count, 1
%innerexitcond = icmp sge i8 %nextcount, 8
br i1 %innerexitcond, label %innerexit, label %subloop
; Should have promoted 'handle2' accesses.
; CHECK: subloop:
; CHECK-NEXT: phi i8* [
; CHECK-NEXT: %count = phi i8 [
; CHECK-NEXT: store i8 %n
; CHECK-NOT: store
; CHECK: br i1
innerexit:
%offsetx1 = load i8** %handle1
%val = load i8* %offsetx1
%cond = icmp eq i8 %val, %n
br i1 %cond, label %exit, label %loop
; Should not have promoted offsetx1 loads.
; CHECK: innerexit:
; CHECK: %val = load i8* %offsetx1
; CHECK: %cond = icmp eq i8 %val, %n
; CHECK: br i1 %cond, label %exit, label %loop
exit:
ret void
}
Enhance LICM to promote alias sets whose pointers themselves are stored, which doesn't affect the memory address being promoted. llvm-svn: 122172
2010-12-19 13:57:25 +08:00
define void @test5(i32 %i, i32** noalias %P2) {
Entry:
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
br label %Loop
Update Transforms tests to use CHECK-LABEL for easier debugging. No functionality change. This update was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$NAME" if ! grep -q "^; *RUN: *llc" $NAME; then TEMP=`mktemp -t temp` cp $NAME $TEMP sed -n "s/^define [^@]*@\([A-Za-z0-9_]*\)(.*$/\1/p" < $NAME | \ while read FUNC; do sed -i '' "s/;\(.*\)\([A-Za-z0-9_]*\):\( *\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186268
2013-07-14 09:42:54 +08:00
; CHECK-LABEL: @test5(
Enhance LICM to promote alias sets whose pointers themselves are stored, which doesn't affect the memory address being promoted. llvm-svn: 122172
2010-12-19 13:57:25 +08:00
; CHECK: Entry:
; CHECK-NEXT: load i32* @X
; CHECK-NEXT: br label %Loop
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
Loop: ; preds = %Loop, %0
%j = phi i32 [ 0, %Entry ], [ %Next, %Loop ] ; <i32> [#uses=1]
%x = load i32* @X ; <i32> [#uses=1]
%x2 = add i32 %x, 1 ; <i32> [#uses=1]
store i32 %x2, i32* @X
Upgrade syntax of tests using volatile instructions to use 'load volatile' instead of 'volatile load', which is archaic. llvm-svn: 145171
2011-11-27 14:54:59 +08:00
store volatile i32* @X, i32** %P2
Fix LICM's memory promotion optimization to preserve TBAA tags when promoting a store in a loop. This was noticed when working on PR14753, but isn't directly related. llvm-svn: 171281
2012-12-31 16:37:17 +08:00
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
%Next = add i32 %j, 1 ; <i32> [#uses=2]
%cond = icmp eq i32 %Next, 0 ; <i1> [#uses=1]
br i1 %cond, label %Out, label %Loop
Enhance LICM to promote alias sets whose pointers themselves are stored, which doesn't affect the memory address being promoted. llvm-svn: 122172
2010-12-19 13:57:25 +08:00
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
Out:
ret void
Enhance LICM to promote alias sets whose pointers themselves are stored, which doesn't affect the memory address being promoted. llvm-svn: 122172
2010-12-19 13:57:25 +08:00
; CHECK: Out:
[LPM] Make LCSSA a utility with a FunctionPass that applies it to all the loops in a function, and teach LICM to work in the presance of LCSSA. Previously, LCSSA was a loop pass. That made passes requiring it also be loop passes and unable to depend on function analysis passes easily. It also caused outer loops to have a different "canonical" form from inner loops during analysis. Instead, we go into LCSSA form and preserve it through the loop pass manager run. Note that this has the same problem as LoopSimplify that prevents enabling its verification -- loop passes which run at the end of the loop pass manager and don't preserve these are valid, but the subsequent loop pass runs of outer loops that do preserve this pass trigger too much verification and fail because the inner loop no longer verifies. The other problem this exposed is that LICM was completely unable to handle LCSSA form. It didn't preserve it and it actually would give up on moving instructions in many cases when they were used by an LCSSA phi node. I've taught LICM to support detecting LCSSA-form PHI nodes and to hoist and sink around them. This may actually let LICM fire significantly more because we put everything into LCSSA form to rotate the loop before running LICM. =/ Now LICM should handle that fine and preserve it correctly. The down side is that LICM has to require LCSSA in order to preserve it. This is just a fact of life for LCSSA. It's entirely possible we should completely remove LCSSA from the optimizer. The test updates are essentially accomodating LCSSA phi nodes in the output of LICM, and the fact that we now completely sink every instruction in ashr-crash below the loop bodies prior to unrolling. With this change, LCSSA is computed only three times in the pass pipeline. One of them could be removed (and potentially a SCEV run and a separate LoopPassManager entirely!) if we had a LoopPass variant of InstCombine that ran InstCombine on the loop body but refused to combine away LCSSA PHI nodes. Currently, this also prevents loop unrolling from being in the same loop pass manager is rotate, LICM, and unswitch. There is one thing that I *really* don't like -- preserving LCSSA in LICM is quite expensive. We end up having to re-run LCSSA twice for some loops after LICM runs because LICM can undo LCSSA both in the current loop and the parent loop. I don't really see good solutions to this other than to completely move away from LCSSA and using tools like SSAUpdater instead. llvm-svn: 200067
2014-01-25 12:07:24 +08:00
; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %x2
; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* @X
Enhance LICM to promote alias sets whose pointers themselves are stored, which doesn't affect the memory address being promoted. llvm-svn: 122172
2010-12-19 13:57:25 +08:00
; CHECK-NEXT: ret void
}
Fix LICM's memory promotion optimization to preserve TBAA tags when promoting a store in a loop. This was noticed when working on PR14753, but isn't directly related. llvm-svn: 171281
2012-12-31 16:37:17 +08:00
; PR14753 - Preserve TBAA tags when promoting values in a loop.
define void @test6(i32 %n, float* nocapture %a, i32* %gi) {
entry:
store i32 0, i32* %gi, align 4, !tbaa !0
%cmp1 = icmp slt i32 0, %n
br i1 %cmp1, label %for.body.lr.ph, label %for.end
for.body.lr.ph: ; preds = %entry
br label %for.body
for.body: ; preds = %for.body.lr.ph, %for.body
%storemerge2 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
%idxprom = sext i32 %storemerge2 to i64
%arrayidx = getelementptr inbounds float* %a, i64 %idxprom
store float 0.000000e+00, float* %arrayidx, align 4, !tbaa !3
%0 = load i32* %gi, align 4, !tbaa !0
%inc = add nsw i32 %0, 1
store i32 %inc, i32* %gi, align 4, !tbaa !0
%cmp = icmp slt i32 %inc, %n
br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge
for.cond.for.end_crit_edge: ; preds = %for.body
br label %for.end
for.end: ; preds = %for.cond.for.end_crit_edge, %entry
ret void
tabs-to-spaces llvm-svn: 171550
2013-01-05 07:11:35 +08:00
Fix LICM's memory promotion optimization to preserve TBAA tags when promoting a store in a loop. This was noticed when working on PR14753, but isn't directly related. llvm-svn: 171281
2012-12-31 16:37:17 +08:00
; CHECK: for.body.lr.ph:
; CHECK-NEXT: %gi.promoted = load i32* %gi, align 4, !tbaa !0
; CHECK: for.cond.for.end_crit_edge:
[LPM] Make LCSSA a utility with a FunctionPass that applies it to all the loops in a function, and teach LICM to work in the presance of LCSSA. Previously, LCSSA was a loop pass. That made passes requiring it also be loop passes and unable to depend on function analysis passes easily. It also caused outer loops to have a different "canonical" form from inner loops during analysis. Instead, we go into LCSSA form and preserve it through the loop pass manager run. Note that this has the same problem as LoopSimplify that prevents enabling its verification -- loop passes which run at the end of the loop pass manager and don't preserve these are valid, but the subsequent loop pass runs of outer loops that do preserve this pass trigger too much verification and fail because the inner loop no longer verifies. The other problem this exposed is that LICM was completely unable to handle LCSSA form. It didn't preserve it and it actually would give up on moving instructions in many cases when they were used by an LCSSA phi node. I've taught LICM to support detecting LCSSA-form PHI nodes and to hoist and sink around them. This may actually let LICM fire significantly more because we put everything into LCSSA form to rotate the loop before running LICM. =/ Now LICM should handle that fine and preserve it correctly. The down side is that LICM has to require LCSSA in order to preserve it. This is just a fact of life for LCSSA. It's entirely possible we should completely remove LCSSA from the optimizer. The test updates are essentially accomodating LCSSA phi nodes in the output of LICM, and the fact that we now completely sink every instruction in ashr-crash below the loop bodies prior to unrolling. With this change, LCSSA is computed only three times in the pass pipeline. One of them could be removed (and potentially a SCEV run and a separate LoopPassManager entirely!) if we had a LoopPass variant of InstCombine that ran InstCombine on the loop body but refused to combine away LCSSA PHI nodes. Currently, this also prevents loop unrolling from being in the same loop pass manager is rotate, LICM, and unswitch. There is one thing that I *really* don't like -- preserving LCSSA in LICM is quite expensive. We end up having to re-run LCSSA twice for some loops after LICM runs because LICM can undo LCSSA both in the current loop and the parent loop. I don't really see good solutions to this other than to completely move away from LCSSA and using tools like SSAUpdater instead. llvm-svn: 200067
2014-01-25 12:07:24 +08:00
; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %inc
; CHECK-NEXT: store i32 %[[LCSSAPHI]], i32* %gi, align 4, !tbaa !0
Fix LICM's memory promotion optimization to preserve TBAA tags when promoting a store in a loop. This was noticed when working on PR14753, but isn't directly related. llvm-svn: 171281
2012-12-31 16:37:17 +08:00
}
TBAA: handle scalar TBAA format and struct-path aware TBAA format. Remove the command line argument "struct-path-tbaa" since we should not depend on command line argument to decide which format the IR file is using. Instead, we check the first operand of the tbaa tag node, if it is a MDNode, we treat it as struct-path aware TBAA format, otherwise, we treat it as scalar TBAA format. When clang starts to use struct-path aware TBAA format no matter whether struct-path-tbaa is no, and we can auto-upgrade existing bc files, the support for scalar TBAA format can be dropped. Existing testing cases are updated to use the struct-path aware TBAA format. llvm-svn: 191538
2013-09-28 02:34:27 +08:00
!0 = metadata !{metadata !4, metadata !4, i64 0}
Fix LICM's memory promotion optimization to preserve TBAA tags when promoting a store in a loop. This was noticed when working on PR14753, but isn't directly related. llvm-svn: 171281
2012-12-31 16:37:17 +08:00
!1 = metadata !{metadata !"omnipotent char", metadata !2}
!2 = metadata !{metadata !"Simple C/C++ TBAA"}
TBAA: handle scalar TBAA format and struct-path aware TBAA format. Remove the command line argument "struct-path-tbaa" since we should not depend on command line argument to decide which format the IR file is using. Instead, we check the first operand of the tbaa tag node, if it is a MDNode, we treat it as struct-path aware TBAA format, otherwise, we treat it as scalar TBAA format. When clang starts to use struct-path aware TBAA format no matter whether struct-path-tbaa is no, and we can auto-upgrade existing bc files, the support for scalar TBAA format can be dropped. Existing testing cases are updated to use the struct-path aware TBAA format. llvm-svn: 191538
2013-09-28 02:34:27 +08:00
!3 = metadata !{metadata !5, metadata !5, i64 0}
!4 = metadata !{metadata !"int", metadata !1}
!5 = metadata !{metadata !"float", metadata !1}