llvm-6502/test/CodeGen/ARM/global-merge.ll
Ahmed Bougacha c9ad3ab624 [AArch64, ARM] Enable GlobalMerge with -O3 rather than -O1.
The pass used to be enabled by default with CodeGenOpt::Less (-O1).
This is too aggressive, considering the pass indiscriminately merges
all globals together.

Currently, performance doesn't always improve, and, on code that uses
few globals (e.g., the odd file- or function- static), more often than
not is degraded by the optimization.  Lengthy discussion can be found
on llvmdev (AArch64-focused;  ARM has similar problems):
  http://lists.cs.uiuc.edu/pipermail/llvmdev/2015-February/082800.html
Also, it makes tooling and debuggers less useful when dealing with
globals and data sections.

GlobalMerge needs to better identify those cases that benefit, and this
will be done separately.  In the meantime, move the pass to run with
-O3 rather than -O1, on both ARM and AArch64.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@233024 91177308-0d34-0410-b5e6-96231b3b80d8
2015-03-23 21:17:36 +00:00

75 lines
3.0 KiB
LLVM

; RUN: llc < %s -mtriple=thumb-apple-darwin -O3 -global-merge-on-const=true | FileCheck %s
; Test the ARMGlobalMerge pass. Use -march=thumb because it has a small
; value for the maximum offset (127).
; A local array that exceeds the maximum offset should not be merged.
; CHECK: g0:
@g0 = internal global [32 x i32] [ i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 1, i32 2 ]
; Global variables marked with "used" attribute must be kept
; CHECK: g8
@g8 = internal global i32 0
@llvm.used = appending global [1 x i8*] [i8* bitcast (i32* @g8 to i8*)], section "llvm.metadata"
; Global used in landing pad instruction must be kept
; CHECK: ZTIi
@_ZTIi = internal global i8* null
define i32 @_Z9exceptioni(i32 %arg) {
bb:
%tmp = invoke i32 @_Z14throwSomethingi(i32 %arg)
to label %bb9 unwind label %bb1
bb1: ; preds = %bb
%tmp2 = landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__gxx_personality_sj0 to i8*)
catch i8* bitcast (i8** @_ZTIi to i8*)
%tmp3 = extractvalue { i8*, i32 } %tmp2, 1
%tmp4 = tail call i32 @llvm.eh.typeid.for(i8* bitcast (i8** @_ZTIi to i8*))
%tmp5 = icmp eq i32 %tmp3, %tmp4
br i1 %tmp5, label %bb6, label %bb10
bb6: ; preds = %bb1
%tmp7 = extractvalue { i8*, i32 } %tmp2, 0
%tmp8 = tail call i8* @__cxa_begin_catch(i8* %tmp7)
tail call void @__cxa_end_catch()
br label %bb9
bb9: ; preds = %bb6, %bb
%res.0 = phi i32 [ 0, %bb6 ], [ %tmp, %bb ]
ret i32 %res.0
bb10: ; preds = %bb1
resume { i8*, i32 } %tmp2
}
declare i32 @_Z14throwSomethingi(i32)
declare i32 @__gxx_personality_sj0(...)
declare i32 @llvm.eh.typeid.for(i8*)
declare i8* @__cxa_begin_catch(i8*)
declare void @__cxa_end_catch()
; CHECK: _MergedGlobals:
@g1 = internal global i32 1
@g2 = internal global i32 2
; Make sure that the complete variable fits within the range of the maximum
; offset. Having the starting offset in range is not sufficient.
; When this works properly, @g3 is placed in a separate chunk of merged globals.
; CHECK: _MergedGlobals1:
@g3 = internal global [30 x i32] [ i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10 ], align 4
; Global variables that can be placed in BSS should be kept together in a
; separate pool of merged globals.
; CHECK: _MergedGlobals2
@g4 = internal global i32 0
@g5 = internal global i32 0
; Global variables that are constant can be merged together
; CHECK: _MergedGlobals3
@g6 = internal constant [12 x i32] zeroinitializer, align 4
@g7 = internal constant [12 x i32] zeroinitializer, align 4