mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-10-04 00:54:53 +00:00
3d69cf57e1
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. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@200067 91177308-0d34-0410-b5e6-96231b3b80d8
255 lines
6.9 KiB
LLVM
255 lines
6.9 KiB
LLVM
; RUN: opt < %s -basicaa -licm -S | FileCheck %s
|
|
|
|
declare i32 @strlen(i8*) readonly
|
|
|
|
declare void @foo()
|
|
|
|
; Sink readonly function.
|
|
define i32 @test1(i8* %P) {
|
|
br label %Loop
|
|
|
|
Loop: ; preds = %Loop, %0
|
|
%A = call i32 @strlen( i8* %P ) readonly
|
|
br i1 false, label %Loop, label %Out
|
|
|
|
Out: ; preds = %Loop
|
|
ret i32 %A
|
|
; CHECK-LABEL: @test1(
|
|
; CHECK: Out:
|
|
; CHECK-NEXT: call i32 @strlen
|
|
; CHECK-NEXT: ret i32 %A
|
|
}
|
|
|
|
declare double @sin(double) readnone
|
|
|
|
; Sink readnone function out of loop with unknown memory behavior.
|
|
define double @test2(double %X) {
|
|
br label %Loop
|
|
|
|
Loop: ; preds = %Loop, %0
|
|
call void @foo( )
|
|
%A = call double @sin( double %X ) readnone
|
|
br i1 true, label %Loop, label %Out
|
|
|
|
Out: ; preds = %Loop
|
|
ret double %A
|
|
; CHECK-LABEL: @test2(
|
|
; CHECK: Out:
|
|
; CHECK-NEXT: call double @sin
|
|
; CHECK-NEXT: ret double %A
|
|
}
|
|
|
|
; This testcase checks to make sure the sinker does not cause problems with
|
|
; critical edges.
|
|
define void @test3() {
|
|
Entry:
|
|
br i1 false, label %Loop, label %Exit
|
|
Loop:
|
|
%X = add i32 0, 1
|
|
br i1 false, label %Loop, label %Exit
|
|
Exit:
|
|
%Y = phi i32 [ 0, %Entry ], [ %X, %Loop ]
|
|
ret void
|
|
|
|
; CHECK-LABEL: @test3(
|
|
; CHECK: Exit.loopexit:
|
|
; CHECK-NEXT: %X = add i32 0, 1
|
|
; CHECK-NEXT: br label %Exit
|
|
|
|
}
|
|
|
|
; If the result of an instruction is only used outside of the loop, sink
|
|
; the instruction to the exit blocks instead of executing it on every
|
|
; iteration of the loop.
|
|
;
|
|
define i32 @test4(i32 %N) {
|
|
Entry:
|
|
br label %Loop
|
|
Loop: ; preds = %Loop, %Entry
|
|
%N_addr.0.pn = phi i32 [ %dec, %Loop ], [ %N, %Entry ]
|
|
%tmp.6 = mul i32 %N, %N_addr.0.pn ; <i32> [#uses=1]
|
|
%tmp.7 = sub i32 %tmp.6, %N ; <i32> [#uses=1]
|
|
%dec = add i32 %N_addr.0.pn, -1 ; <i32> [#uses=1]
|
|
%tmp.1 = icmp ne i32 %N_addr.0.pn, 1 ; <i1> [#uses=1]
|
|
br i1 %tmp.1, label %Loop, label %Out
|
|
Out: ; preds = %Loop
|
|
ret i32 %tmp.7
|
|
; CHECK-LABEL: @test4(
|
|
; CHECK: Out:
|
|
; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %N_addr.0.pn
|
|
; CHECK-NEXT: mul i32 %N, %[[LCSSAPHI]]
|
|
; CHECK-NEXT: sub i32 %tmp.6, %N
|
|
; CHECK-NEXT: ret i32
|
|
}
|
|
|
|
; To reduce register pressure, if a load is hoistable out of the loop, and the
|
|
; result of the load is only used outside of the loop, sink the load instead of
|
|
; hoisting it!
|
|
;
|
|
@X = global i32 5 ; <i32*> [#uses=1]
|
|
|
|
define i32 @test5(i32 %N) {
|
|
Entry:
|
|
br label %Loop
|
|
Loop: ; preds = %Loop, %Entry
|
|
%N_addr.0.pn = phi i32 [ %dec, %Loop ], [ %N, %Entry ]
|
|
%tmp.6 = load i32* @X ; <i32> [#uses=1]
|
|
%dec = add i32 %N_addr.0.pn, -1 ; <i32> [#uses=1]
|
|
%tmp.1 = icmp ne i32 %N_addr.0.pn, 1 ; <i1> [#uses=1]
|
|
br i1 %tmp.1, label %Loop, label %Out
|
|
Out: ; preds = %Loop
|
|
ret i32 %tmp.6
|
|
; CHECK-LABEL: @test5(
|
|
; CHECK: Out:
|
|
; CHECK-NEXT: %tmp.6 = load i32* @X
|
|
; CHECK-NEXT: ret i32 %tmp.6
|
|
}
|
|
|
|
|
|
|
|
; The loop sinker was running from the bottom of the loop to the top, causing
|
|
; it to miss opportunities to sink instructions that depended on sinking other
|
|
; instructions from the loop. Instead they got hoisted, which is better than
|
|
; leaving them in the loop, but increases register pressure pointlessly.
|
|
|
|
%Ty = type { i32, i32 }
|
|
@X2 = external global %Ty
|
|
|
|
define i32 @test6() {
|
|
br label %Loop
|
|
Loop:
|
|
%dead = getelementptr %Ty* @X2, i64 0, i32 0
|
|
%sunk2 = load i32* %dead
|
|
br i1 false, label %Loop, label %Out
|
|
Out: ; preds = %Loop
|
|
ret i32 %sunk2
|
|
; CHECK-LABEL: @test6(
|
|
; CHECK: Out:
|
|
; CHECK-NEXT: %dead = getelementptr %Ty* @X2, i64 0, i32 0
|
|
; CHECK-NEXT: %sunk2 = load i32* %dead
|
|
; CHECK-NEXT: ret i32 %sunk2
|
|
}
|
|
|
|
|
|
|
|
; This testcase ensures that we can sink instructions from loops with
|
|
; multiple exits.
|
|
;
|
|
define i32 @test7(i32 %N, i1 %C) {
|
|
Entry:
|
|
br label %Loop
|
|
Loop: ; preds = %ContLoop, %Entry
|
|
%N_addr.0.pn = phi i32 [ %dec, %ContLoop ], [ %N, %Entry ]
|
|
%tmp.6 = mul i32 %N, %N_addr.0.pn
|
|
%tmp.7 = sub i32 %tmp.6, %N ; <i32> [#uses=2]
|
|
%dec = add i32 %N_addr.0.pn, -1 ; <i32> [#uses=1]
|
|
br i1 %C, label %ContLoop, label %Out1
|
|
ContLoop:
|
|
%tmp.1 = icmp ne i32 %N_addr.0.pn, 1
|
|
br i1 %tmp.1, label %Loop, label %Out2
|
|
Out1: ; preds = %Loop
|
|
ret i32 %tmp.7
|
|
Out2: ; preds = %ContLoop
|
|
ret i32 %tmp.7
|
|
; CHECK-LABEL: @test7(
|
|
; CHECK: Out1:
|
|
; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %N_addr.0.pn
|
|
; CHECK-NEXT: mul i32 %N, %[[LCSSAPHI]]
|
|
; CHECK-NEXT: sub i32 %tmp.6, %N
|
|
; CHECK-NEXT: ret
|
|
; CHECK: Out2:
|
|
; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %N_addr.0.pn
|
|
; CHECK-NEXT: mul i32 %N, %[[LCSSAPHI]]
|
|
; CHECK-NEXT: sub i32 %tmp.6.le, %N
|
|
; CHECK-NEXT: ret
|
|
}
|
|
|
|
|
|
; This testcase checks to make sure we can sink values which are only live on
|
|
; some exits out of the loop, and that we can do so without breaking dominator
|
|
; info.
|
|
define i32 @test8(i1 %C1, i1 %C2, i32* %P, i32* %Q) {
|
|
Entry:
|
|
br label %Loop
|
|
Loop: ; preds = %Cont, %Entry
|
|
br i1 %C1, label %Cont, label %exit1
|
|
Cont: ; preds = %Loop
|
|
%X = load i32* %P ; <i32> [#uses=2]
|
|
store i32 %X, i32* %Q
|
|
%V = add i32 %X, 1 ; <i32> [#uses=1]
|
|
br i1 %C2, label %Loop, label %exit2
|
|
exit1: ; preds = %Loop
|
|
ret i32 0
|
|
exit2: ; preds = %Cont
|
|
ret i32 %V
|
|
; CHECK-LABEL: @test8(
|
|
; CHECK: exit1:
|
|
; CHECK-NEXT: ret i32 0
|
|
; CHECK: exit2:
|
|
; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %X
|
|
; CHECK-NEXT: %V = add i32 %[[LCSSAPHI]], 1
|
|
; CHECK-NEXT: ret i32 %V
|
|
}
|
|
|
|
|
|
define void @test9() {
|
|
loopentry.2.i:
|
|
br i1 false, label %no_exit.1.i.preheader, label %loopentry.3.i.preheader
|
|
no_exit.1.i.preheader: ; preds = %loopentry.2.i
|
|
br label %no_exit.1.i
|
|
no_exit.1.i: ; preds = %endif.8.i, %no_exit.1.i.preheader
|
|
br i1 false, label %return.i, label %endif.8.i
|
|
endif.8.i: ; preds = %no_exit.1.i
|
|
%inc.1.i = add i32 0, 1 ; <i32> [#uses=1]
|
|
br i1 false, label %no_exit.1.i, label %loopentry.3.i.preheader.loopexit
|
|
loopentry.3.i.preheader.loopexit: ; preds = %endif.8.i
|
|
br label %loopentry.3.i.preheader
|
|
loopentry.3.i.preheader: ; preds = %loopentry.3.i.preheader.loopexit, %loopentry.2.i
|
|
%arg_num.0.i.ph13000 = phi i32 [ 0, %loopentry.2.i ], [ %inc.1.i, %loopentry.3.i.preheader.loopexit ] ; <i32> [#uses=0]
|
|
ret void
|
|
return.i: ; preds = %no_exit.1.i
|
|
ret void
|
|
|
|
; CHECK-LABEL: @test9(
|
|
; CHECK: loopentry.3.i.preheader.loopexit:
|
|
; CHECK-NEXT: %inc.1.i = add i32 0, 1
|
|
; CHECK-NEXT: br label %loopentry.3.i.preheader
|
|
}
|
|
|
|
|
|
; Potentially trapping instructions may be sunk as long as they are guaranteed
|
|
; to be executed.
|
|
define i32 @test10(i32 %N) {
|
|
Entry:
|
|
br label %Loop
|
|
Loop: ; preds = %Loop, %Entry
|
|
%N_addr.0.pn = phi i32 [ %dec, %Loop ], [ %N, %Entry ] ; <i32> [#uses=3]
|
|
%tmp.6 = sdiv i32 %N, %N_addr.0.pn ; <i32> [#uses=1]
|
|
%dec = add i32 %N_addr.0.pn, -1 ; <i32> [#uses=1]
|
|
%tmp.1 = icmp ne i32 %N_addr.0.pn, 0 ; <i1> [#uses=1]
|
|
br i1 %tmp.1, label %Loop, label %Out
|
|
Out: ; preds = %Loop
|
|
ret i32 %tmp.6
|
|
|
|
; CHECK-LABEL: @test10(
|
|
; CHECK: Out:
|
|
; CHECK-NEXT: %[[LCSSAPHI:.*]] = phi i32 [ %N_addr.0.pn
|
|
; CHECK-NEXT: %tmp.6 = sdiv i32 %N, %[[LCSSAPHI]]
|
|
; CHECK-NEXT: ret i32 %tmp.6
|
|
}
|
|
|
|
; Should delete, not sink, dead instructions.
|
|
define void @test11() {
|
|
br label %Loop
|
|
Loop:
|
|
%dead = getelementptr %Ty* @X2, i64 0, i32 0
|
|
br i1 false, label %Loop, label %Out
|
|
Out:
|
|
ret void
|
|
; CHECK-LABEL: @test11(
|
|
; CHECK: Out:
|
|
; CHECK-NEXT: ret void
|
|
}
|
|
|
|
|