mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-10-23 04:52:54 +02:00
99ee90db82
Currently, this pass only focuses on *trivial* loop unswitching. At that reduced problem it remains significantly better than the current loop unswitch: - Old pass is worse than cubic complexity. New pass is (I think) linear. - New pass is much simpler in its design by focusing on full unswitching. (See below for details on this). - New pass doesn't carry state for thresholds between pass iterations. - New pass doesn't carry state for correctness (both miscompile and infloop) between pass iterations. - New pass produces substantially better code after unswitching. - New pass can handle more trivial unswitch cases. - New pass doesn't recompute the dominator tree for the entire function and instead incrementally updates it. I've ported all of the trivial unswitching test cases from the old pass to the new one to make sure that major functionality isn't lost in the process. For several of the test cases I've worked to improve the precision and rigor of the CHECKs, but for many I've just updated them to handle the new IR produced. My initial motivation was the fact that the old pass carried state in very unreliable ways between pass iterations, and these mechansims were incompatible with the new pass manager. However, I discovered many more improvements to make along the way. This pass makes two very significant assumptions that enable most of these improvements: 1) Focus on *full* unswitching -- that is, completely removing whatever control flow construct is being unswitched from the loop. In the case of trivial unswitching, this means removing the trivial (exiting) edge. In non-trivial unswitching, this means removing the branch or switch itself. This is in opposition to *partial* unswitching where some part of the unswitched control flow remains in the loop. Partial unswitching only really applies to switches and to folded branches. These are very similar to full unrolling and partial unrolling. The full form is an effective canonicalization, the partial form needs a complex cost model, cannot be iterated, isn't canonicalizing, and should be a separate pass that runs very late (much like unrolling). 2) Leverage LLVM's Loop machinery to the fullest. The original unswitch dates from a time when a great deal of LLVM's loop infrastructure was missing, ineffective, and/or unreliable. As a consequence, a lot of complexity was added which we no longer need. With these two overarching principles, I think we can build a fast and effective unswitcher that fits in well in the new PM and in the canonicalization pipeline. Some of the remaining functionality around partial unswitching may not be relevant today (not many test cases or benchmarks I can find) but if they are I'd like to add support for them as a separate layer that runs very late in the pipeline. Purely to make reviewing and introducing this code more manageable, I've split this into first a trivial-unswitch-only pass and in the next patch I'll add support for full non-trivial unswitching against a *fixed* threshold, exactly like full unrolling. I even plan to re-use the unrolling thresholds, as these are incredibly similar cost tradeoffs: we're cloning a loop body in order to end up with simplified control flow. We should only do that when the total growth is reasonably small. One of the biggest changes with this pass compared to the previous one is that previously, each individual trivial exiting edge from a switch was unswitched separately as a branch. Now, we unswitch the entire switch at once, with cases going to the various destinations. This lets us unswitch multiple exiting edges in a single operation and also avoids numerous extremely bad behaviors, where we would introduce 1000s of branches to test for thousands of possible values, all of which would take the exact same exit path bypassing the loop. Now we will use a switch with 1000s of cases that can be efficiently lowered into a jumptable. This avoids relying on somehow forming a switch out of the branches or getting horrible code if that fails for any reason. Another significant change is that this pass actively updates the CFG based on unswitching. For trivial unswitching, this is actually very easy because of the definition of loop simplified form. Doing this makes the code coming out of loop unswitch dramatically more friendly. We still should run loop-simplifycfg (at the least) after this to clean up, but it will have to do a lot less work. Finally, this pass makes much fewer attempts to simplify instructions based on the unswitch. Something like loop-instsimplify, instcombine, or GVN can be used to do increasingly powerful simplifications based on the now dominating predicate. The old simplifications are things that something like loop-instsimplify should get today or a very, very basic loop-instcombine could get. Keeping that logic separate is a big simplifying technique. Most of the code in this pass that isn't in the old one has to do with achieving specific goals: - Updating the dominator tree as we go - Unswitching all cases in a switch in a single step. I think it is still shorter than just the trivial unswitching code in the old pass despite having this functionality. Differential Revision: https://reviews.llvm.org/D32409 llvm-svn: 301576
130 lines
5.1 KiB
LLVM
130 lines
5.1 KiB
LLVM
; RUN: opt -simple-loop-unswitch -verify-loop-info -verify-dom-info -disable-output < %s
|
|
|
|
; Loop unswitch should be able to unswitch these loops and
|
|
; preserve LCSSA and LoopSimplify forms.
|
|
|
|
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:64:64-v128:128:128-a0:0:64"
|
|
target triple = "armv6-apple-darwin9"
|
|
|
|
@delim1 = external global i32 ; <i32*> [#uses=1]
|
|
@delim2 = external global i32 ; <i32*> [#uses=1]
|
|
|
|
define i32 @ineqn(i8* %s, i8* %p) nounwind readonly {
|
|
entry:
|
|
%0 = load i32, i32* @delim1, align 4 ; <i32> [#uses=1]
|
|
%1 = load i32, i32* @delim2, align 4 ; <i32> [#uses=1]
|
|
br label %bb8.outer
|
|
|
|
bb: ; preds = %bb8
|
|
%2 = icmp eq i8* %p_addr.0, %s ; <i1> [#uses=1]
|
|
br i1 %2, label %bb10, label %bb2
|
|
|
|
bb2: ; preds = %bb
|
|
%3 = getelementptr inbounds i8, i8* %p_addr.0, i32 1 ; <i8*> [#uses=3]
|
|
switch i32 %ineq.0.ph, label %bb8.backedge [
|
|
i32 0, label %bb3
|
|
i32 1, label %bb6
|
|
]
|
|
|
|
bb8.backedge: ; preds = %bb6, %bb5, %bb2
|
|
br label %bb8
|
|
|
|
bb3: ; preds = %bb2
|
|
%4 = icmp eq i32 %8, %0 ; <i1> [#uses=1]
|
|
br i1 %4, label %bb8.outer.loopexit, label %bb5
|
|
|
|
bb5: ; preds = %bb3
|
|
br i1 %6, label %bb6, label %bb8.backedge
|
|
|
|
bb6: ; preds = %bb5, %bb2
|
|
%5 = icmp eq i32 %8, %1 ; <i1> [#uses=1]
|
|
br i1 %5, label %bb7, label %bb8.backedge
|
|
|
|
bb7: ; preds = %bb6
|
|
%.lcssa1 = phi i8* [ %3, %bb6 ] ; <i8*> [#uses=1]
|
|
br label %bb8.outer.backedge
|
|
|
|
bb8.outer.backedge: ; preds = %bb8.outer.loopexit, %bb7
|
|
%.lcssa2 = phi i8* [ %.lcssa1, %bb7 ], [ %.lcssa, %bb8.outer.loopexit ] ; <i8*> [#uses=1]
|
|
%ineq.0.ph.be = phi i32 [ 0, %bb7 ], [ 1, %bb8.outer.loopexit ] ; <i32> [#uses=1]
|
|
br label %bb8.outer
|
|
|
|
bb8.outer.loopexit: ; preds = %bb3
|
|
%.lcssa = phi i8* [ %3, %bb3 ] ; <i8*> [#uses=1]
|
|
br label %bb8.outer.backedge
|
|
|
|
bb8.outer: ; preds = %bb8.outer.backedge, %entry
|
|
%ineq.0.ph = phi i32 [ 0, %entry ], [ %ineq.0.ph.be, %bb8.outer.backedge ] ; <i32> [#uses=3]
|
|
%p_addr.0.ph = phi i8* [ %p, %entry ], [ %.lcssa2, %bb8.outer.backedge ] ; <i8*> [#uses=1]
|
|
%6 = icmp eq i32 %ineq.0.ph, 1 ; <i1> [#uses=1]
|
|
br label %bb8
|
|
|
|
bb8: ; preds = %bb8.outer, %bb8.backedge
|
|
%p_addr.0 = phi i8* [ %p_addr.0.ph, %bb8.outer ], [ %3, %bb8.backedge ] ; <i8*> [#uses=3]
|
|
%7 = load i8, i8* %p_addr.0, align 1 ; <i8> [#uses=2]
|
|
%8 = sext i8 %7 to i32 ; <i32> [#uses=2]
|
|
%9 = icmp eq i8 %7, 0 ; <i1> [#uses=1]
|
|
br i1 %9, label %bb10, label %bb
|
|
|
|
bb10: ; preds = %bb8, %bb
|
|
%.0 = phi i32 [ %ineq.0.ph, %bb ], [ 0, %bb8 ] ; <i32> [#uses=1]
|
|
ret i32 %.0
|
|
}
|
|
|
|
; This is a simplified form of ineqn from above. It triggers some
|
|
; different cases in the loop-unswitch code.
|
|
|
|
define void @simplified_ineqn() nounwind readonly {
|
|
entry:
|
|
br label %bb8.outer
|
|
|
|
bb8.outer: ; preds = %bb6, %bb2, %entry
|
|
%x = phi i32 [ 0, %entry ], [ 0, %bb6 ], [ 1, %bb2 ] ; <i32> [#uses=1]
|
|
br i1 undef, label %return, label %bb2
|
|
|
|
bb2: ; preds = %bb
|
|
switch i32 %x, label %bb6 [
|
|
i32 0, label %bb8.outer
|
|
]
|
|
|
|
bb6: ; preds = %bb2
|
|
br i1 undef, label %bb8.outer, label %bb2
|
|
|
|
return: ; preds = %bb8, %bb
|
|
ret void
|
|
}
|
|
|
|
; This function requires special handling to preserve LCSSA form.
|
|
; PR4934
|
|
|
|
define void @pnp_check_irq() nounwind noredzone {
|
|
entry:
|
|
%conv56 = trunc i64 undef to i32 ; <i32> [#uses=1]
|
|
br label %while.cond.i
|
|
|
|
while.cond.i: ; preds = %while.cond.i.backedge, %entry
|
|
%call.i25 = call i8* @pci_get_device() nounwind noredzone ; <i8*> [#uses=2]
|
|
br i1 undef, label %if.then65, label %while.body.i
|
|
|
|
while.body.i: ; preds = %while.cond.i
|
|
br i1 undef, label %if.then31.i.i, label %while.cond.i.backedge
|
|
|
|
while.cond.i.backedge: ; preds = %if.then31.i.i, %while.body.i
|
|
br label %while.cond.i
|
|
|
|
if.then31.i.i: ; preds = %while.body.i
|
|
switch i32 %conv56, label %while.cond.i.backedge [
|
|
i32 14, label %if.then42.i.i
|
|
i32 15, label %if.then42.i.i
|
|
]
|
|
|
|
if.then42.i.i: ; preds = %if.then31.i.i, %if.then31.i.i
|
|
%call.i25.lcssa48 = phi i8* [ %call.i25, %if.then31.i.i ], [ %call.i25, %if.then31.i.i ] ; <i8*> [#uses=0]
|
|
unreachable
|
|
|
|
if.then65: ; preds = %while.cond.i
|
|
unreachable
|
|
}
|
|
|
|
declare i8* @pci_get_device() noredzone
|