llvm/test/CodeGen/ARM/2011-08-25-ldmia_ret.ll
Will Dietz e4b44c1617 MachineSink: Fix and tweak critical-edge breaking heuristic.
Per original comment, the intention of this loop
is to go ahead and break the critical edge
(in order to sink this instruction) if there's
reason to believe doing so might "unblock" the
sinking of additional instructions that define
registers used by this one.  The idea is that if
we have a few instructions to sink "together"
breaking the edge might be worthwhile.

This commit makes a few small changes
to help better realize this goal:

First, modify the loop to ignore registers
defined by this instruction.  We don't
sink definitions of physical registers,
and sinking an SSA definition isn't
going to unblock an upstream instruction.

Second, ignore uses of physical registers.
Instructions that define physical registers are
rejected for sinking, and so moving this one
won't enable moving any defining instructions.
As an added bonus, while virtual register
use-def chains are generally small due
to SSA goodness, iteration over the uses
and definitions (used by hasOneNonDBGUse)
for physical registers like EFLAGS
can be rather expensive in practice.
(This is the original reason for looking at this)

Finally, to keep things simple continue
to only consider this trick for registers that
have a single use (via hasOneNonDBGUse),
but to avoid spuriously breaking critical edges
only do so if the definition resides
in the same MBB and therefore this one directly
blocks it from being sunk as well.
If sinking them together is meant to be,
let the iterative nature of this pass
sink the definition into this block first.

Update tests to accomodate this change,
add new testcase where sinking avoids pipeline stalls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192608 91177308-0d34-0410-b5e6-96231b3b80d8
2013-10-14 16:57:17 +00:00

101 lines
3.7 KiB
LLVM

; RUN: llc < %s -mtriple=thumbv7-apple-ios -mcpu=cortex-a9 | FileCheck %s
; Test that ldmia_ret preserves implicit operands for return values.
;
; This CFG is reduced from a benchmark miscompile. With current
; if-conversion heuristics, one of the return paths is if-converted
; into sw.bb18 resulting in an ldmia_ret in the middle of the
; block. The postra scheduler needs to know that the return implicitly
; uses the return register, otherwise its antidep breaker scavenges
; the register in order to hoist the constant load required to test
; the switch.
declare i32 @getint()
declare i1 @getbool()
declare void @foo(i32)
declare i32 @bar(i32)
define i32 @test(i32 %in1, i32 %in2) nounwind {
entry:
%call = tail call zeroext i1 @getbool() nounwind
br i1 %call, label %sw.bb18, label %sw.bb2
sw.bb2: ; preds = %entry
%cmp = tail call zeroext i1 @getbool() nounwind
br i1 %cmp, label %sw.epilog58, label %land.lhs.true
land.lhs.true: ; preds = %sw.bb2
%cmp13 = tail call zeroext i1 @getbool() nounwind
br i1 %cmp13, label %if.then, label %sw.epilog58
if.then: ; preds = %land.lhs.true
tail call void @foo(i32 %in1) nounwind
br label %sw.epilog58
; load the return value
; CHECK: movs [[RRET:r.]], #2
; hoist the switch constant without clobbering RRET
; CHECK: movw
; CHECK-NOT: [[RRET]]
; CHECK: , #63707
; CHECK-NOT: [[RRET]]
; CHECK: tst
; If-convert the return
; CHECK: it ne
; Fold the CSR+return into a pop
; CHECK: pop {r4, r5, r7, pc}
sw.bb18:
%call20 = tail call i32 @bar(i32 %in2) nounwind
switch i32 %call20, label %sw.default56 [
i32 168, label %sw.bb21
i32 165, label %sw.bb21
i32 261, label %sw.epilog58
i32 188, label %sw.epilog58
i32 187, label %sw.epilog58
i32 186, label %sw.epilog58
i32 185, label %sw.epilog58
i32 184, label %sw.epilog58
i32 175, label %sw.epilog58
i32 174, label %sw.epilog58
i32 173, label %sw.epilog58
i32 172, label %sw.epilog58
i32 171, label %sw.epilog58
i32 167, label %sw.epilog58
i32 166, label %sw.epilog58
i32 164, label %sw.epilog58
i32 163, label %sw.epilog58
i32 161, label %sw.epilog58
i32 160, label %sw.epilog58
i32 -1, label %sw.bb33
]
sw.bb21: ; preds = %sw.bb18, %sw.bb18
tail call void @foo(i32 %in2) nounwind
%call28 = tail call i32 @getint() nounwind
%tobool = icmp eq i32 %call28, 0
br i1 %tobool, label %if.then29, label %sw.epilog58
if.then29: ; preds = %sw.bb21
tail call void @foo(i32 %in2) nounwind
br label %sw.epilog58
sw.bb33: ; preds = %sw.bb18
%cmp42 = tail call zeroext i1 @getbool() nounwind
br i1 %cmp42, label %sw.default56, label %land.lhs.true44
land.lhs.true44: ; preds = %sw.bb33
%call50 = tail call i32 @getint() nounwind
%cmp51 = icmp slt i32 %call50, 0
br i1 %cmp51, label %if.then53, label %sw.default56
if.then53: ; preds = %land.lhs.true44
tail call void @foo(i32 %in2) nounwind
br label %sw.default56
sw.default56: ; preds = %sw.bb33, %land.lhs.true44, %if.then53, %sw.bb18
br label %sw.epilog58
sw.epilog58:
%retval.0 = phi i32 [ 4, %sw.default56 ], [ 2, %sw.bb21 ], [ 2, %if.then29 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb18 ], [ 2, %sw.bb2 ], [ 2, %land.lhs.true ], [ 2, %if.then ]
ret i32 %retval.0
}