maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/Transforms/ObjCARC/basic.ll

3075 lines
87 KiB
LLVM
Raw Normal View History

Make provenance checking conservative in cases when pointers-to-strong-pointers may be in play. These can lead to retains and releases happening in unstructured ways, foiling the optimizer. This fixes rdar://12150909. llvm-svn: 163180
2012-09-05 07:16:20 +08:00
; RUN: opt -basicaa -objc-arc -S < %s | FileCheck %s
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
target datalayout = "e-p:64:64:64"
declare i8* @objc_retain(i8*)
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
declare i8* @objc_retainAutoreleasedReturnValue(i8*)
Add support for objc_unsafeClaimAutoreleasedReturnValue to the ObjC ARC Optimizer. The main implication of this is: 1. Ensuring that we treat it conservatively in terms of optimization. 2. We put the ASM marker on it so that the runtime can recognize objc_unsafeClaimAutoreleasedReturnValue from releaseRV. <rdar://problem/21567064> Patch by Michael Gottesman! llvm-svn: 258970
2016-01-28 03:05:08 +08:00
declare i8* @objc_unsafeClaimAutoreleasedReturnValue(i8*)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
declare void @objc_release(i8*)
declare i8* @objc_autorelease(i8*)
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
declare i8* @objc_autoreleaseReturnValue(i8*)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
declare void @objc_autoreleasePoolPop(i8*)
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
declare i8* @objc_autoreleasePoolPush()
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
declare i8* @objc_retainBlock(i8*)
declare i8* @objc_retainedObject(i8*)
declare i8* @objc_unretainedObject(i8*)
declare i8* @objc_unretainedPointer(i8*)
declare void @use_pointer(i8*)
declare void @callee()
declare void @callee_fnptr(void ()*)
declare void @invokee()
declare i8* @returner()
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
declare void @bar(i32 ()*)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
Remove the obsolete offset parameter from @llvm.dbg.value There is no situation where this rarely-used argument cannot be substituted with a DIExpression and removing it allows us to simplify the DWARF backend. Note that this patch does not yet remove any of the newly dead code. rdar://problem/33580047 Differential Revision: https://reviews.llvm.org/D35951 llvm-svn: 309426
2017-07-29 04:21:02 +08:00
declare void @llvm.dbg.value(metadata, metadata, metadata)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
declare i8* @objc_msgSend(i8*, i8*, ...)
; Simple retain+release pair deletion, with some intervening control
; flow and harmless instructions.
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: define void @test0_precise(i32* %x, i1 %p) [[NUW:#[0-9]+]] {
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: @objc_retain
; CHECK: @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test0_precise(i32* %x, i1 %p) nounwind {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %return
f:
store i32 7, i32* %x
br label %return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind
ret void
}
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: define void @test0_imprecise(i32* %x, i1 %p) [[NUW]] {
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test0_imprecise(i32* %x, i1 %p) nounwind {
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %return
f:
store i32 7, i32* %x
br label %return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind, !clang.imprecise_release !0
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; Like test0 but the release isn't always executed when the retain is,
; so the optimization is not safe.
; TODO: Make the objc_release's argument be %0.
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: define void @test1_precise(i32* %x, i1 %p, i1 %q) [[NUW]] {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @objc_retain(i8* %a)
; CHECK: @objc_release
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test1_precise(i32* %x, i1 %p, i1 %q) nounwind {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %return
f:
store i32 7, i32* %x
call void @callee()
br i1 %q, label %return, label %alt_return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind
ret void
alt_return:
ret void
}
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: define void @test1_imprecise(i32* %x, i1 %p, i1 %q) [[NUW]] {
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: @objc_retain(i8* %a)
; CHECK: @objc_release
; CHECK: }
define void @test1_imprecise(i32* %x, i1 %p, i1 %q) nounwind {
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %return
f:
store i32 7, i32* %x
call void @callee()
br i1 %q, label %return, label %alt_return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind, !clang.imprecise_release !0
ret void
alt_return:
ret void
}
Suppress partial retain+release elimination when there's a possibility that it will span multiple CFG diamonds/triangles which could have different controlling predicates. rdar://10282956 llvm-svn: 142222
2011-10-18 02:48:25 +08:00
; Don't do partial elimination into two different CFG diamonds.
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: define void @test1b_precise(i8* %x, i1 %p, i1 %q) {
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: entry:
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: tail call i8* @objc_retain(i8* %x) [[NUW]]
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NOT: @objc_
; CHECK: if.end5:
; CHECK: tail call void @objc_release(i8* %x) [[NUW]]
; CHECK-NOT: @objc_
; CHECK: }
define void @test1b_precise(i8* %x, i1 %p, i1 %q) {
entry:
tail call i8* @objc_retain(i8* %x) nounwind
br i1 %p, label %if.then, label %if.end
if.then: ; preds = %entry
tail call void @callee()
br label %if.end
if.end: ; preds = %if.then, %entry
br i1 %q, label %if.then3, label %if.end5
if.then3: ; preds = %if.end
tail call void @use_pointer(i8* %x)
br label %if.end5
if.end5: ; preds = %if.then3, %if.end
tail call void @objc_release(i8* %x) nounwind
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test1b_imprecise(
Suppress partial retain+release elimination when there's a possibility that it will span multiple CFG diamonds/triangles which could have different controlling predicates. rdar://10282956 llvm-svn: 142222
2011-10-18 02:48:25 +08:00
; CHECK: entry:
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: tail call i8* @objc_retain(i8* %x) [[NUW:#[0-9]+]]
Suppress partial retain+release elimination when there's a possibility that it will span multiple CFG diamonds/triangles which could have different controlling predicates. rdar://10282956 llvm-svn: 142222
2011-10-18 02:48:25 +08:00
; CHECK-NOT: @objc_
; CHECK: if.end5:
Debug Info: update testing cases to specify the debug info version number. We are going to drop debug info without a version number or with a different version number, to make sure we don't crash when we see bitcode files with different debug info metadata format. Make tests more robust by removing hard-coded metadata numbers in CHECK lines. llvm-svn: 195535
2013-11-23 09:16:29 +08:00
; CHECK: tail call void @objc_release(i8* %x) [[NUW]], !clang.imprecise_release ![[RELEASE:[0-9]+]]
Suppress partial retain+release elimination when there's a possibility that it will span multiple CFG diamonds/triangles which could have different controlling predicates. rdar://10282956 llvm-svn: 142222
2011-10-18 02:48:25 +08:00
; CHECK-NOT: @objc_
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test1b_imprecise(i8* %x, i1 %p, i1 %q) {
Suppress partial retain+release elimination when there's a possibility that it will span multiple CFG diamonds/triangles which could have different controlling predicates. rdar://10282956 llvm-svn: 142222
2011-10-18 02:48:25 +08:00
entry:
tail call i8* @objc_retain(i8* %x) nounwind
br i1 %p, label %if.then, label %if.end
if.then: ; preds = %entry
tail call void @callee()
br label %if.end
if.end: ; preds = %if.then, %entry
br i1 %q, label %if.then3, label %if.end5
if.then3: ; preds = %if.end
tail call void @use_pointer(i8* %x)
br label %if.end5
if.end5: ; preds = %if.then3, %if.end
tail call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
ret void
}
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; Like test0 but the pointer is passed to an intervening call,
; so the optimization is not safe.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test2_precise(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @objc_retain(i8* %a)
; CHECK: @objc_release
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test2_precise(i32* %x, i1 %p) nounwind {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %return
f:
store i32 7, i32* %x
call void @use_pointer(i8* %0)
%d = bitcast i32* %x to float*
store float 3.0, float* %d
br label %return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test2_imprecise(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: @objc_retain(i8* %a)
; CHECK: @objc_release
; CHECK: }
define void @test2_imprecise(i32* %x, i1 %p) nounwind {
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %return
f:
store i32 7, i32* %x
call void @use_pointer(i8* %0)
%d = bitcast i32* %x to float*
store float 3.0, float* %d
br label %return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind, !clang.imprecise_release !0
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; Like test0 but the release is in a loop,
; so the optimization is not safe.
; TODO: For now, assume this can't happen.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test3_precise(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; TODO: @objc_retain(i8* %a)
; TODO: @objc_release
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test3_precise(i32* %x, i1* %q) nounwind {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br label %loop
loop:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%j = load volatile i1, i1* %q
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
br i1 %j, label %loop, label %return
return:
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test3_imprecise(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; TODO: @objc_retain(i8* %a)
; TODO: @objc_release
; CHECK: }
define void @test3_imprecise(i32* %x, i1* %q) nounwind {
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br label %loop
loop:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind, !clang.imprecise_release !0
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%j = load volatile i1, i1* %q
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
br i1 %j, label %loop, label %return
return:
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; TODO: For now, assume this can't happen.
; Like test0 but the retain is in a loop,
; so the optimization is not safe.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test4_precise(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; TODO: @objc_retain(i8* %a)
; TODO: @objc_release
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test4_precise(i32* %x, i1* %q) nounwind {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
br label %loop
loop:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%j = load volatile i1, i1* %q
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
br i1 %j, label %loop, label %return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test4_imprecise(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; TODO: @objc_retain(i8* %a)
; TODO: @objc_release
; CHECK: }
define void @test4_imprecise(i32* %x, i1* %q) nounwind {
entry:
br label %loop
loop:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%j = load volatile i1, i1* %q
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
br i1 %j, label %loop, label %return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind, !clang.imprecise_release !0
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; Like test0 but the pointer is conditionally passed to an intervening call,
; so the optimization is not safe.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test5a(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @objc_retain(i8*
; CHECK: @objc_release
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test5a(i32* %x, i1 %q, i8* %y) nounwind {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
%s = select i1 %q, i8* %y, i8* %0
call void @use_pointer(i8* %s)
store i32 7, i32* %x
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test5b(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: @objc_retain(i8*
; CHECK: @objc_release
; CHECK: }
define void @test5b(i32* %x, i1 %q, i8* %y) nounwind {
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
%s = select i1 %q, i8* %y, i8* %0
call void @use_pointer(i8* %s)
store i32 7, i32* %x
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind, !clang.imprecise_release !0
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; retain+release pair deletion, where the release happens on two different
; flow paths.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test6a(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: entry:
; CHECK: tail call i8* @objc_retain(
; CHECK: t:
; CHECK: call void @objc_release(
; CHECK: f:
; CHECK: call void @objc_release(
; CHECK: return:
; CHECK: }
define void @test6a(i32* %x, i1 %p) nounwind {
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
%ct = bitcast i32* %x to i8*
call void @objc_release(i8* %ct) nounwind
br label %return
f:
store i32 7, i32* %x
call void @callee()
%cf = bitcast i32* %x to i8*
call void @objc_release(i8* %cf) nounwind
br label %return
return:
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test6b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test6b(i32* %x, i1 %p) nounwind {
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
%ct = bitcast i32* %x to i8*
call void @objc_release(i8* %ct) nounwind, !clang.imprecise_release !0
br label %return
f:
store i32 7, i32* %x
call void @callee()
%cf = bitcast i32* %x to i8*
call void @objc_release(i8* %cf) nounwind, !clang.imprecise_release !0
br label %return
return:
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test6c(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: entry:
; CHECK: tail call i8* @objc_retain(
; CHECK: t:
; CHECK: call void @objc_release(
; CHECK: f:
; CHECK: call void @objc_release(
; CHECK: return:
; CHECK: }
define void @test6c(i32* %x, i1 %p) nounwind {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
%ct = bitcast i32* %x to i8*
call void @objc_release(i8* %ct) nounwind
br label %return
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
f:
store i32 7, i32* %x
call void @callee()
%cf = bitcast i32* %x to i8*
call void @objc_release(i8* %cf) nounwind, !clang.imprecise_release !0
br label %return
return:
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test6d(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: entry:
; CHECK: tail call i8* @objc_retain(
; CHECK: t:
; CHECK: call void @objc_release(
; CHECK: f:
; CHECK: call void @objc_release(
; CHECK: return:
; CHECK: }
define void @test6d(i32* %x, i1 %p) nounwind {
entry:
%a = bitcast i32* %x to i8*
%0 = call i8* @objc_retain(i8* %a) nounwind
br i1 %p, label %t, label %f
t:
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
%ct = bitcast i32* %x to i8*
call void @objc_release(i8* %ct) nounwind, !clang.imprecise_release !0
br label %return
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
f:
store i32 7, i32* %x
call void @callee()
%cf = bitcast i32* %x to i8*
call void @objc_release(i8* %cf) nounwind
br label %return
return:
ret void
}
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; retain+release pair deletion, where the retain happens on two different
; flow paths.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test7(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: entry:
; CHECK-NOT: objc_
; CHECK: t:
; CHECK: call i8* @objc_retain
; CHECK: f:
; CHECK: call i8* @objc_retain
; CHECK: return:
; CHECK: call void @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define void @test7(i32* %x, i1 %p) nounwind {
entry:
%a = bitcast i32* %x to i8*
br i1 %p, label %t, label %f
t:
%0 = call i8* @objc_retain(i8* %a) nounwind
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %return
f:
%1 = call i8* @objc_retain(i8* %a) nounwind
store i32 7, i32* %x
call void @callee()
br label %return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test7b(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test7b(i32* %x, i1 %p) nounwind {
entry:
%a = bitcast i32* %x to i8*
br i1 %p, label %t, label %f
t:
%0 = call i8* @objc_retain(i8* %a) nounwind
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %return
f:
%1 = call i8* @objc_retain(i8* %a) nounwind
store i32 7, i32* %x
call void @callee()
br label %return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind, !clang.imprecise_release !0
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; Like test7, but there's a retain/retainBlock mismatch. Don't delete!
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test7c(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: t:
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: call i8* @objc_retainBlock
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: f:
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: call i8* @objc_retain
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: return:
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: call void @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test7c(i32* %x, i1 %p) nounwind {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%a = bitcast i32* %x to i8*
br i1 %p, label %t, label %f
t:
%0 = call i8* @objc_retainBlock(i8* %a) nounwind
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %return
f:
%1 = call i8* @objc_retain(i8* %a) nounwind
store i32 7, i32* %x
call void @callee()
br label %return
return:
%c = bitcast i32* %x to i8*
call void @objc_release(i8* %c) nounwind
ret void
}
; retain+release pair deletion, where the retain and release both happen on
; different flow paths. Wild!
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test8a(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: entry:
; CHECK: t:
; CHECK: @objc_retain
; CHECK: f:
; CHECK: @objc_retain
; CHECK: mid:
; CHECK: u:
; CHECK: @objc_release
; CHECK: g:
; CHECK: @objc_release
; CHECK: return:
; CHECK: }
define void @test8a(i32* %x, i1 %p, i1 %q) nounwind {
entry:
%a = bitcast i32* %x to i8*
br i1 %p, label %t, label %f
t:
%0 = call i8* @objc_retain(i8* %a) nounwind
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %mid
f:
%1 = call i8* @objc_retain(i8* %a) nounwind
store i32 7, i32* %x
br label %mid
mid:
br i1 %q, label %u, label %g
u:
call void @callee()
%cu = bitcast i32* %x to i8*
call void @objc_release(i8* %cu) nounwind
br label %return
g:
%cg = bitcast i32* %x to i8*
call void @objc_release(i8* %cg) nounwind
br label %return
return:
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test8b(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test8b(i32* %x, i1 %p, i1 %q) nounwind {
entry:
%a = bitcast i32* %x to i8*
br i1 %p, label %t, label %f
t:
%0 = call i8* @objc_retain(i8* %a) nounwind
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %mid
f:
%1 = call i8* @objc_retain(i8* %a) nounwind
store i32 7, i32* %x
br label %mid
mid:
br i1 %q, label %u, label %g
u:
call void @callee()
%cu = bitcast i32* %x to i8*
call void @objc_release(i8* %cu) nounwind, !clang.imprecise_release !0
br label %return
g:
%cg = bitcast i32* %x to i8*
call void @objc_release(i8* %cg) nounwind, !clang.imprecise_release !0
br label %return
return:
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test8c(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: entry:
; CHECK: t:
; CHECK: @objc_retain
; CHECK: f:
; CHECK: @objc_retain
; CHECK: mid:
; CHECK: u:
; CHECK: @objc_release
; CHECK: g:
; CHECK: @objc_release
; CHECK: return:
; CHECK: }
define void @test8c(i32* %x, i1 %p, i1 %q) nounwind {
entry:
%a = bitcast i32* %x to i8*
br i1 %p, label %t, label %f
t:
%0 = call i8* @objc_retain(i8* %a) nounwind
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %mid
f:
%1 = call i8* @objc_retain(i8* %a) nounwind
store i32 7, i32* %x
br label %mid
mid:
br i1 %q, label %u, label %g
u:
call void @callee()
%cu = bitcast i32* %x to i8*
call void @objc_release(i8* %cu) nounwind
br label %return
g:
%cg = bitcast i32* %x to i8*
call void @objc_release(i8* %cg) nounwind, !clang.imprecise_release !0
br label %return
return:
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test8d(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: entry:
; CHECK: t:
; CHECK: @objc_retain
; CHECK: f:
; CHECK: @objc_retain
; CHECK: mid:
; CHECK: u:
; CHECK: @objc_release
; CHECK: g:
; CHECK: @objc_release
; CHECK: return:
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test8d(i32* %x, i1 %p, i1 %q) nounwind {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%a = bitcast i32* %x to i8*
br i1 %p, label %t, label %f
t:
%0 = call i8* @objc_retain(i8* %a) nounwind
store i8 3, i8* %a
%b = bitcast i32* %x to float*
store float 2.0, float* %b
br label %mid
f:
%1 = call i8* @objc_retain(i8* %a) nounwind
store i32 7, i32* %x
br label %mid
mid:
br i1 %q, label %u, label %g
u:
call void @callee()
%cu = bitcast i32* %x to i8*
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call void @objc_release(i8* %cu) nounwind, !clang.imprecise_release !0
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
br label %return
g:
%cg = bitcast i32* %x to i8*
call void @objc_release(i8* %cg) nounwind
br label %return
return:
ret void
}
; Trivial retain+release pair deletion.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test9(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test9(i8* %x) nounwind {
entry:
%0 = call i8* @objc_retain(i8* %x) nounwind
call void @objc_release(i8* %0) nounwind
ret void
}
; Retain+release pair, but on an unknown pointer relationship. Don't delete!
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test9b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @objc_retain(i8* %x)
; CHECK: @objc_release(i8* %s)
; CHECK: }
define void @test9b(i8* %x, i1 %j, i8* %p) nounwind {
entry:
%0 = call i8* @objc_retain(i8* %x) nounwind
%s = select i1 %j, i8* %x, i8* %p
call void @objc_release(i8* %s) nounwind
ret void
}
; Trivial retain+release pair with intervening calls - don't delete!
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test10(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @objc_retain(i8* %x)
Make a few tests slightly more strict. llvm-svn: 138241
2011-08-23 01:27:02 +08:00
; CHECK: @callee
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @use_pointer
; CHECK: @objc_release
; CHECK: }
define void @test10(i8* %x) nounwind {
entry:
%0 = call i8* @objc_retain(i8* %x) nounwind
Make a few tests slightly more strict. llvm-svn: 138241
2011-08-23 01:27:02 +08:00
call void @callee()
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @use_pointer(i8* %x)
call void @objc_release(i8* %0) nounwind
ret void
}
; Trivial retain+autoreleaserelease pair. Don't delete!
; Also, add a tail keyword, since objc_retain can never be passed
; a stack argument.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test11(
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: tail call i8* @objc_retain(i8* %x) [[NUW]]
; CHECK: call i8* @objc_autorelease(i8* %0) [[NUW]]
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define void @test11(i8* %x) nounwind {
entry:
%0 = call i8* @objc_retain(i8* %x) nounwind
call i8* @objc_autorelease(i8* %0) nounwind
call void @use_pointer(i8* %x)
ret void
}
; Same as test11 but with no use_pointer call. Delete the pair!
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test11a(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: entry:
; CHECK-NEXT: ret void
; CHECK: }
define void @test11a(i8* %x) nounwind {
entry:
%0 = call i8* @objc_retain(i8* %x) nounwind
call i8* @objc_autorelease(i8* %0) nounwind
ret void
}
Remove an optimization where we were changing an objc_autorelease into an objc_autoreleaseReturnValue. The semantics of ARC implies that a pointer passed into an objc_autorelease must live until some point (potentially down the stack) where an autorelease pool is popped. On the other hand, an objc_autoreleaseReturnValue just signifies that the object must live until the end of the given function at least. Thus objc_autorelease is stronger than objc_autoreleaseReturnValue in terms of the semantics of ARC* implying that performing the given strength reduction without any knowledge of how this relates to the autorelease pool pop that is further up the stack violates the semantics of ARC. *Even though objc_autoreleaseReturnValue if you know that no RV optimization will occur is more computationally expensive. llvm-svn: 178612
2013-04-03 10:57:24 +08:00
; Same as test11 but the value is returned. Do not perform an RV optimization
; since if the frontend emitted code for an __autoreleasing variable, we may
; want it to be in the autorelease pool.
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define i8* @test11b(
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: tail call i8* @objc_retain(i8* %x) [[NUW]]
Remove an optimization where we were changing an objc_autorelease into an objc_autoreleaseReturnValue. The semantics of ARC implies that a pointer passed into an objc_autorelease must live until some point (potentially down the stack) where an autorelease pool is popped. On the other hand, an objc_autoreleaseReturnValue just signifies that the object must live until the end of the given function at least. Thus objc_autorelease is stronger than objc_autoreleaseReturnValue in terms of the semantics of ARC* implying that performing the given strength reduction without any knowledge of how this relates to the autorelease pool pop that is further up the stack violates the semantics of ARC. *Even though objc_autoreleaseReturnValue if you know that no RV optimization will occur is more computationally expensive. llvm-svn: 178612
2013-04-03 10:57:24 +08:00
; CHECK: call i8* @objc_autorelease(i8* %0) [[NUW]]
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define i8* @test11b(i8* %x) nounwind {
entry:
%0 = call i8* @objc_retain(i8* %x) nounwind
call i8* @objc_autorelease(i8* %0) nounwind
ret i8* %x
}
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; We can not delete this retain, release since we do not have a post-dominating
; use of the release.
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test12(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: entry:
; CHECK-NEXT: @objc_retain(i8* %x)
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: @objc_retain
; CHECK: @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define void @test12(i8* %x, i64 %n) {
entry:
call i8* @objc_retain(i8* %x) nounwind
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
ret void
}
; Trivial retain,autorelease pair. Don't delete!
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test13(
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: tail call i8* @objc_retain(i8* %x) [[NUW]]
; CHECK: tail call i8* @objc_retain(i8* %x) [[NUW]]
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @use_pointer(i8* %x)
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: call i8* @objc_autorelease(i8* %x) [[NUW]]
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define void @test13(i8* %x, i64 %n) {
entry:
call i8* @objc_retain(i8* %x) nounwind
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call i8* @objc_autorelease(i8* %x) nounwind
ret void
}
; Delete the retain+release pair.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test13b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: entry:
; CHECK-NEXT: @objc_retain(i8* %x)
; CHECK-NEXT: @use_pointer
; CHECK-NEXT: @use_pointer
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: @use_pointer
; CHECK-NEXT: @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: ret void
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK-NEXT: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define void @test13b(i8* %x, i64 %n) {
entry:
call i8* @objc_retain(i8* %x) nounwind
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
ret void
}
; Don't delete the retain+release pair because there's an
; autoreleasePoolPop in the way.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test13c(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @objc_retain(i8* %x)
; CHECK: @objc_autoreleasePoolPop
; CHECK: @objc_retain(i8* %x)
; CHECK: @use_pointer
; CHECK: @objc_release
; CHECK: }
define void @test13c(i8* %x, i64 %n) {
entry:
call i8* @objc_retain(i8* %x) nounwind
call void @objc_autoreleasePoolPop(i8* undef)
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
ret void
}
; Like test13c, but there's an autoreleasePoolPush in the way, but that
; doesn't matter.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test13d(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: entry:
; CHECK-NEXT: @objc_retain(i8* %x)
; CHECK-NEXT: @objc_autoreleasePoolPush
; CHECK-NEXT: @use_pointer
; CHECK-NEXT: @use_pointer
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: @use_pointer
; CHECK-NEXT: @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: ret void
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK-NEXT: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define void @test13d(i8* %x, i64 %n) {
entry:
call i8* @objc_retain(i8* %x) nounwind
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
call i8* @objc_autoreleasePoolPush()
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
ret void
}
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; Trivial retain,release pair with intervening call, and it's post-dominated by
; another release. But it is not known safe in the top down direction. We can
; not eliminate it.
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test14(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: entry:
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: @objc_retain
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: @use_pointer
; CHECK-NEXT: @use_pointer
; CHECK-NEXT: @objc_release
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test14(i8* %x, i64 %n) {
entry:
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
call void @objc_release(i8* %x) nounwind
ret void
}
; Trivial retain,autorelease pair with intervening call, but it's post-dominated
; by another release. Don't delete anything.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test15(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: entry:
; CHECK-NEXT: @objc_retain(i8* %x)
; CHECK-NEXT: @use_pointer
; CHECK-NEXT: @objc_autorelease(i8* %x)
; CHECK-NEXT: @objc_release
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test15(i8* %x, i64 %n) {
entry:
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call i8* @objc_autorelease(i8* %x) nounwind
call void @objc_release(i8* %x) nounwind
ret void
}
; Trivial retain,autorelease pair, post-dominated
; by another release. Delete the retain and release.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test15b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: entry:
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NEXT: @objc_retain
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: @objc_autorelease
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NEXT: @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test15b(i8* %x, i64 %n) {
entry:
call i8* @objc_retain(i8* %x) nounwind
call i8* @objc_autorelease(i8* %x) nounwind
call void @objc_release(i8* %x) nounwind
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test15c(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NEXT: entry:
; CHECK-NEXT: @objc_autorelease
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test15c(i8* %x, i64 %n) {
entry:
call i8* @objc_retain(i8* %x) nounwind
call i8* @objc_autorelease(i8* %x) nounwind
call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; Retain+release pairs in diamonds, all dominated by a retain.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test16a(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: @objc_retain(i8* %x)
; CHECK-NOT: @objc
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK: purple:
; CHECK: @use_pointer
; CHECK: @objc_release
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: }
define void @test16a(i1 %a, i1 %b, i8* %x) {
entry:
call i8* @objc_retain(i8* %x) nounwind
br i1 %a, label %red, label %orange
red:
call i8* @objc_retain(i8* %x) nounwind
br label %yellow
orange:
call i8* @objc_retain(i8* %x) nounwind
br label %yellow
yellow:
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
br i1 %b, label %green, label %blue
green:
call void @objc_release(i8* %x) nounwind
br label %purple
blue:
call void @objc_release(i8* %x) nounwind
br label %purple
purple:
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test16b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @objc_retain(i8* %x)
; CHECK-NOT: @objc
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK: purple:
; CHECK-NEXT: @use_pointer
; CHECK-NEXT: @use_pointer
; CHECK-NEXT: @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test16b(i1 %a, i1 %b, i8* %x) {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
call i8* @objc_retain(i8* %x) nounwind
br i1 %a, label %red, label %orange
red:
call i8* @objc_retain(i8* %x) nounwind
br label %yellow
orange:
call i8* @objc_retain(i8* %x) nounwind
br label %yellow
yellow:
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
br i1 %b, label %green, label %blue
green:
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
br label %purple
blue:
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @objc_release(i8* %x) nounwind
br label %purple
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
purple:
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test16c(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: @objc_retain(i8* %x)
; CHECK-NOT: @objc
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK: purple:
; CHECK: @use_pointer
; CHECK: @objc_release
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: }
define void @test16c(i1 %a, i1 %b, i8* %x) {
entry:
call i8* @objc_retain(i8* %x) nounwind
br i1 %a, label %red, label %orange
red:
call i8* @objc_retain(i8* %x) nounwind
br label %yellow
orange:
call i8* @objc_retain(i8* %x) nounwind
br label %yellow
yellow:
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
br i1 %b, label %green, label %blue
green:
call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
br label %purple
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
blue:
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
br label %purple
purple:
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test16d(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: @objc_retain(i8* %x)
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK: @objc
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: }
define void @test16d(i1 %a, i1 %b, i8* %x) {
entry:
call i8* @objc_retain(i8* %x) nounwind
br i1 %a, label %red, label %orange
red:
call i8* @objc_retain(i8* %x) nounwind
br label %yellow
orange:
call i8* @objc_retain(i8* %x) nounwind
br label %yellow
yellow:
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
br i1 %b, label %green, label %blue
green:
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @objc_release(i8* %x) nounwind
br label %purple
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
blue:
call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
br label %purple
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
purple:
ret void
}
; Delete no-ops.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test18(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test18() {
call i8* @objc_retain(i8* null)
call void @objc_release(i8* null)
call i8* @objc_autorelease(i8* null)
ret void
}
; Delete no-ops where undef can be assumed to be null.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test18b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test18b() {
call i8* @objc_retain(i8* undef)
call void @objc_release(i8* undef)
call i8* @objc_autorelease(i8* undef)
ret void
}
; Replace uses of arguments with uses of return values, to reduce
; register pressure.
; CHECK: define void @test19(i32* %y) {
; CHECK: %z = bitcast i32* %y to i8*
; CHECK: %0 = bitcast i32* %y to i8*
; CHECK: %1 = tail call i8* @objc_retain(i8* %0)
; CHECK: call void @use_pointer(i8* %z)
; CHECK: call void @use_pointer(i8* %z)
; CHECK: %2 = bitcast i32* %y to i8*
; CHECK: call void @objc_release(i8* %2)
; CHECK: ret void
; CHECK: }
define void @test19(i32* %y) {
entry:
%x = bitcast i32* %y to i8*
%0 = call i8* @objc_retain(i8* %x) nounwind
%z = bitcast i32* %y to i8*
call void @use_pointer(i8* %z)
call void @use_pointer(i8* %z)
call void @objc_release(i8* %x)
ret void
}
; Bitcast insertion
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test20(
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: %tmp1 = tail call i8* @objc_retain(i8* %tmp) [[NUW]]
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: invoke
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: }
Move the personality function from LandingPadInst to Function The personality routine currently lives in the LandingPadInst. This isn't desirable because: - All LandingPadInsts in the same function must have the same personality routine. This means that each LandingPadInst beyond the first has an operand which produces no additional information. - There is ongoing work to introduce EH IR constructs other than LandingPadInst. Moving the personality routine off of any one particular Instruction and onto the parent function seems a lot better than have N different places a personality function can sneak onto an exceptional function. Differential Revision: http://reviews.llvm.org/D10429 llvm-svn: 239940
2015-06-18 04:52:32 +08:00
define void @test20(double* %self) personality i32 (...)* @__gxx_personality_v0 {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
if.then12:
%tmp = bitcast double* %self to i8*
%tmp1 = call i8* @objc_retain(i8* %tmp) nounwind
invoke void @invokee()
to label %invoke.cont23 unwind label %lpad20
invoke.cont23: ; preds = %if.then12
invoke void @invokee()
to label %if.end unwind label %lpad20
lpad20: ; preds = %invoke.cont23, %if.then12
%tmp502 = phi double* [ undef, %invoke.cont23 ], [ %self, %if.then12 ]
Move the personality function from LandingPadInst to Function The personality routine currently lives in the LandingPadInst. This isn't desirable because: - All LandingPadInsts in the same function must have the same personality routine. This means that each LandingPadInst beyond the first has an operand which produces no additional information. - There is ongoing work to introduce EH IR constructs other than LandingPadInst. Moving the personality routine off of any one particular Instruction and onto the parent function seems a lot better than have N different places a personality function can sneak onto an exceptional function. Differential Revision: http://reviews.llvm.org/D10429 llvm-svn: 239940
2015-06-18 04:52:32 +08:00
%exn = landingpad {i8*, i32}
Update more tests to the new EH scheme. llvm-svn: 138894
2011-09-01 05:04:11 +08:00
cleanup
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
unreachable
if.end: ; preds = %invoke.cont23
ret void
}
; Delete a redundant retain,autorelease when forwaring a call result
; directly to a return value.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define i8* @test21(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: call i8* @returner()
; CHECK-NEXT: ret i8* %call
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK-NEXT: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define i8* @test21() {
entry:
%call = call i8* @returner()
%0 = call i8* @objc_retain(i8* %call) nounwind
%1 = call i8* @objc_autorelease(i8* %0) nounwind
ret i8* %1
}
; Move an objc call up through a phi that has null operands.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test22(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: B:
; CHECK: %1 = bitcast double* %p to i8*
; CHECK: call void @objc_release(i8* %1)
; CHECK: br label %C
; CHECK: C: ; preds = %B, %A
; CHECK-NOT: @objc_release
; CHECK: }
define void @test22(double* %p, i1 %a) {
br i1 %a, label %A, label %B
A:
br label %C
B:
br label %C
[ObjCARC] Do not move a release that has the clang.imprecise_release tag above PHI instructions. ARC optimizer has an optimization that moves a call to an ObjC runtime function above a phi instruction when the phi has a null operand and is an argument passed to the function call. This optimization should not kick in when the runtime function is an objc_release that releases an object with precise lifetime semantics. rdar://problem/34959669 llvm-svn: 315914
2017-10-17 00:46:59 +08:00
C:
%h = phi double* [ null, %A ], [ %p, %B ]
%c = bitcast double* %h to i8*
call void @objc_release(i8* %c), !clang.imprecise_release !0
ret void
}
; Do not move an objc_release that doesn't have the clang.imprecise_release tag.
; CHECK-LABEL: define void @test22_precise(
; CHECK: %[[P0:.*]] = phi double*
; CHECK: %[[V0:.*]] = bitcast double* %[[P0]] to i8*
; CHECK: call void @objc_release(i8* %[[V0]])
; CHECK: ret void
define void @test22_precise(double* %p, i1 %a) {
br i1 %a, label %A, label %B
A:
br label %C
B:
br label %C
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
C:
%h = phi double* [ null, %A ], [ %p, %B ]
%c = bitcast double* %h to i8*
call void @objc_release(i8* %c)
ret void
}
; Any call can decrement a retain count.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test24(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: @objc_retain(i8* %a)
; CHECK: @objc_release
; CHECK: }
define void @test24(i8* %r, i8* %a) {
call i8* @objc_retain(i8* %a)
call void @use_pointer(i8* %r)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%q = load i8, i8* %a
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @objc_release(i8* %a)
ret void
}
; Don't move a retain/release pair if the release can be moved
; but the retain can't be moved to balance it.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test25(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: entry:
; CHECK: call i8* @objc_retain(i8* %p)
; CHECK: true:
; CHECK: done:
; CHECK: call void @objc_release(i8* %p)
; CHECK: }
define void @test25(i8* %p, i1 %x) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
call void @callee()
br i1 %x, label %true, label %done
true:
store i8 0, i8* %p
br label %done
done:
call void @objc_release(i8* %p)
ret void
}
; Don't move a retain/release pair if the retain can be moved
; but the release can't be moved to balance it.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test26(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: entry:
; CHECK: call i8* @objc_retain(i8* %p)
; CHECK: true:
; CHECK: done:
; CHECK: call void @objc_release(i8* %p)
; CHECK: }
define void @test26(i8* %p, i1 %x) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
call void @callee()
br label %done
done:
store i8 0, i8* %p
call void @objc_release(i8* %p)
ret void
}
; Don't sink the retain,release into the loop.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test27(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: entry:
; CHECK: call i8* @objc_retain(i8* %p)
; CHECK: loop:
; CHECK-NOT: @objc_
; CHECK: done:
; CHECK: call void @objc_release
; CHECK: }
define void @test27(i8* %p, i1 %x, i1 %y) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %loop, label %done
loop:
call void @callee()
store i8 0, i8* %p
br i1 %y, label %done, label %loop
done:
call void @objc_release(i8* %p)
ret void
}
; Trivial code motion case: Triangle.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test28(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: true:
; CHECK: call i8* @objc_retain(
; CHECK: call void @callee()
; CHECK: store
; CHECK: call void @objc_release
; CHECK: done:
; CHECK-NOT: @objc_
; CHECK: }
define void @test28(i8* %p, i1 %x) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
call void @callee()
store i8 0, i8* %p
br label %done
done:
call void @objc_release(i8* %p), !clang.imprecise_release !0
ret void
}
; Trivial code motion case: Triangle, but no metadata. Don't move past
; unrelated memory references!
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test28b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: call i8* @objc_retain(
; CHECK: true:
; CHECK-NOT: @objc_
; CHECK: call void @callee()
; CHECK-NOT: @objc_
; CHECK: store
; CHECK-NOT: @objc_
; CHECK: done:
; CHECK: @objc_release
; CHECK: }
define void @test28b(i8* %p, i1 %x, i8* noalias %t) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
call void @callee()
store i8 0, i8* %p
br label %done
done:
store i8 0, i8* %t
call void @objc_release(i8* %p)
ret void
}
; Trivial code motion case: Triangle, with metadata. Do move past
; unrelated memory references! And preserve the metadata.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test28c(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: true:
; CHECK: call i8* @objc_retain(
; CHECK: call void @callee()
; CHECK: store
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: call void @objc_release(i8* %p) [[NUW]], !clang.imprecise_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: done:
; CHECK-NOT: @objc_
; CHECK: }
define void @test28c(i8* %p, i1 %x, i8* noalias %t) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
call void @callee()
store i8 0, i8* %p
br label %done
done:
store i8 0, i8* %t
call void @objc_release(i8* %p), !clang.imprecise_release !0
ret void
}
; Like test28. but with two releases.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test29(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: true:
; CHECK: call i8* @objc_retain(
; CHECK: call void @callee()
; CHECK: store
; CHECK: call void @objc_release
; CHECK-NOT: @objc_release
; CHECK: done:
; CHECK-NOT: @objc_
; CHECK: ohno:
; CHECK-NOT: @objc_
; CHECK: }
define void @test29(i8* %p, i1 %x, i1 %y) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
call void @callee()
store i8 0, i8* %p
br i1 %y, label %done, label %ohno
done:
call void @objc_release(i8* %p)
ret void
ohno:
call void @objc_release(i8* %p)
ret void
}
; Basic case with the use and call in a diamond
; with an extra release.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test30(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: true:
; CHECK: call i8* @objc_retain(
; CHECK: call void @callee()
; CHECK: store
; CHECK: call void @objc_release
; CHECK-NOT: @objc_release
; CHECK: false:
; CHECK-NOT: @objc_
; CHECK: done:
; CHECK-NOT: @objc_
; CHECK: ohno:
; CHECK-NOT: @objc_
; CHECK: }
define void @test30(i8* %p, i1 %x, i1 %y, i1 %z) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %false
true:
call void @callee()
store i8 0, i8* %p
br i1 %y, label %done, label %ohno
false:
br i1 %z, label %done, label %ohno
done:
call void @objc_release(i8* %p)
ret void
ohno:
call void @objc_release(i8* %p)
ret void
}
; Basic case with a mergeable release.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test31(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: call i8* @objc_retain(i8* %p)
; CHECK: call void @callee()
; CHECK: store
; CHECK: call void @objc_release
; CHECK-NOT: @objc_release
; CHECK: true:
; CHECK-NOT: @objc_release
; CHECK: false:
; CHECK-NOT: @objc_release
; CHECK: ret void
; CHECK-NOT: @objc_release
; CHECK: }
define void @test31(i8* %p, i1 %x) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
call void @callee()
store i8 0, i8* %p
br i1 %x, label %true, label %false
true:
call void @objc_release(i8* %p)
ret void
false:
call void @objc_release(i8* %p)
ret void
}
; Don't consider bitcasts or getelementptrs direct uses.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test32(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: true:
; CHECK: call i8* @objc_retain(
; CHECK: call void @callee()
; CHECK: store
; CHECK: call void @objc_release
; CHECK: done:
; CHECK-NOT: @objc_
; CHECK: }
define void @test32(i8* %p, i1 %x) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
call void @callee()
store i8 0, i8* %p
br label %done
done:
%g = bitcast i8* %p to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%h = getelementptr i8, i8* %g, i64 0
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @objc_release(i8* %g)
ret void
}
; Do consider icmps to be direct uses.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test33(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: true:
; CHECK: call i8* @objc_retain(
; CHECK: call void @callee()
; CHECK: icmp
; CHECK: call void @objc_release
; CHECK: done:
; CHECK-NOT: @objc_
; CHECK: }
define void @test33(i8* %p, i1 %x, i8* %y) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
call void @callee()
%v = icmp eq i8* %p, %y
br label %done
done:
%g = bitcast i8* %p to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%h = getelementptr i8, i8* %g, i64 0
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @objc_release(i8* %g)
ret void
}
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; Delete retain,release if there's just a possible dec and we have imprecise
; releases.
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test34a(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: call i8* @objc_retain
; CHECK: true:
; CHECK: done:
; CHECK: call void @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test34a(i8* %p, i1 %x, i8* %y) {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
call void @callee()
br label %done
done:
%g = bitcast i8* %p to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%h = getelementptr i8, i8* %g, i64 0
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @objc_release(i8* %g)
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test34b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test34b(i8* %p, i1 %x, i8* %y) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
call void @callee()
br label %done
done:
%g = bitcast i8* %p to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%h = getelementptr i8, i8* %g, i64 0
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call void @objc_release(i8* %g), !clang.imprecise_release !0
ret void
}
; Delete retain,release if there's just a use and we do not have a precise
; release.
; Precise.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test35a(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: entry:
; CHECK: call i8* @objc_retain
; CHECK: true:
; CHECK: done:
; CHECK: call void @objc_release
; CHECK: }
define void @test35a(i8* %p, i1 %x, i8* %y) {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
%v = icmp eq i8* %p, %y
br label %done
done:
%g = bitcast i8* %p to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%h = getelementptr i8, i8* %g, i64 0
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @objc_release(i8* %g)
ret void
}
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; Imprecise.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test35b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: }
define void @test35b(i8* %p, i1 %x, i8* %y) {
entry:
%f0 = call i8* @objc_retain(i8* %p)
br i1 %x, label %true, label %done
true:
%v = icmp eq i8* %p, %y
br label %done
done:
%g = bitcast i8* %p to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%h = getelementptr i8, i8* %g, i64 0
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call void @objc_release(i8* %g), !clang.imprecise_release !0
ret void
}
; Delete a retain,release if there's no actual use and we have precise release.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test36a(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: @objc_retain
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: call void @callee()
; CHECK-NOT: @objc_
; CHECK: call void @callee()
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test36a(i8* %p) {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
call i8* @objc_retain(i8* %p)
call void @callee()
call void @callee()
call void @objc_release(i8* %p)
ret void
}
; Like test36, but with metadata.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test36b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test36b(i8* %p) {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
entry:
call i8* @objc_retain(i8* %p)
call void @callee()
call void @callee()
call void @objc_release(i8* %p), !clang.imprecise_release !0
ret void
}
; Be aggressive about analyzing phis to eliminate possible uses.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test38(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test38(i8* %p, i1 %u, i1 %m, i8* %z, i8* %y, i8* %x, i8* %w) {
entry:
call i8* @objc_retain(i8* %p)
br i1 %u, label %true, label %false
true:
br i1 %m, label %a, label %b
false:
br i1 %m, label %c, label %d
a:
br label %e
b:
br label %e
c:
br label %f
d:
br label %f
e:
%j = phi i8* [ %z, %a ], [ %y, %b ]
br label %g
f:
%k = phi i8* [ %w, %c ], [ %x, %d ]
br label %g
g:
%h = phi i8* [ %j, %e ], [ %k, %f ]
call void @use_pointer(i8* %h)
call void @objc_release(i8* %p), !clang.imprecise_release !0
ret void
}
; Delete retain,release pairs around loops.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test39(
Fix broken check lines. I really need to find a way to automate this, but I can't come up with a regex that has no false positives while handling tricky cases like custom check prefixes. llvm-svn: 162097
2012-08-17 20:28:26 +08:00
; CHECK-NOT: @objc_
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define void @test39(i8* %p) {
entry:
%0 = call i8* @objc_retain(i8* %p)
br label %loop
loop: ; preds = %loop, %entry
br i1 undef, label %loop, label %exit
exit: ; preds = %loop
call void @objc_release(i8* %0), !clang.imprecise_release !0
ret void
}
; Delete retain,release pairs around loops containing uses.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test39b(
Fix broken check lines. I really need to find a way to automate this, but I can't come up with a regex that has no false positives while handling tricky cases like custom check prefixes. llvm-svn: 162097
2012-08-17 20:28:26 +08:00
; CHECK-NOT: @objc_
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define void @test39b(i8* %p) {
entry:
%0 = call i8* @objc_retain(i8* %p)
br label %loop
loop: ; preds = %loop, %entry
store i8 0, i8* %0
br i1 undef, label %loop, label %exit
exit: ; preds = %loop
call void @objc_release(i8* %0), !clang.imprecise_release !0
ret void
}
; Delete retain,release pairs around loops containing potential decrements.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test39c(
Fix broken check lines. I really need to find a way to automate this, but I can't come up with a regex that has no false positives while handling tricky cases like custom check prefixes. llvm-svn: 162097
2012-08-17 20:28:26 +08:00
; CHECK-NOT: @objc_
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define void @test39c(i8* %p) {
entry:
%0 = call i8* @objc_retain(i8* %p)
br label %loop
loop: ; preds = %loop, %entry
call void @use_pointer(i8* %0)
br i1 undef, label %loop, label %exit
exit: ; preds = %loop
call void @objc_release(i8* %0), !clang.imprecise_release !0
ret void
}
; Delete retain,release pairs around loops even if
; the successors are in a different order.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test40(
Fix broken check lines. I really need to find a way to automate this, but I can't come up with a regex that has no false positives while handling tricky cases like custom check prefixes. llvm-svn: 162097
2012-08-17 20:28:26 +08:00
; CHECK-NOT: @objc_
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define void @test40(i8* %p) {
entry:
%0 = call i8* @objc_retain(i8* %p)
br label %loop
loop: ; preds = %loop, %entry
call void @use_pointer(i8* %0)
br i1 undef, label %exit, label %loop
exit: ; preds = %loop
call void @objc_release(i8* %0), !clang.imprecise_release !0
ret void
}
; Do the known-incremented retain+release elimination even if the pointer
; is also autoreleased.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test42(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: entry:
; CHECK-NEXT: call i8* @objc_retain(i8* %p)
; CHECK-NEXT: call i8* @objc_autorelease(i8* %p)
; CHECK-NEXT: call void @use_pointer(i8* %p)
; CHECK-NEXT: call void @use_pointer(i8* %p)
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: call void @use_pointer(i8* %p)
; CHECK-NEXT: call void @use_pointer(i8* %p)
; CHECK-NEXT: call void @objc_release(i8* %p)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test42(i8* %p) {
entry:
call i8* @objc_retain(i8* %p)
call i8* @objc_autorelease(i8* %p)
call i8* @objc_retain(i8* %p)
call void @use_pointer(i8* %p)
call void @use_pointer(i8* %p)
call void @objc_release(i8* %p)
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call void @use_pointer(i8* %p)
call void @use_pointer(i8* %p)
call void @objc_release(i8* %p)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
ret void
}
; Don't the known-incremented retain+release elimination if the pointer is
; autoreleased and there's an autoreleasePoolPop.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test43(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: entry:
; CHECK-NEXT: call i8* @objc_retain(i8* %p)
; CHECK-NEXT: call i8* @objc_autorelease(i8* %p)
; CHECK-NEXT: call i8* @objc_retain
; CHECK-NEXT: call void @use_pointer(i8* %p)
; CHECK-NEXT: call void @use_pointer(i8* %p)
; CHECK-NEXT: call void @objc_autoreleasePoolPop(i8* undef)
; CHECK-NEXT: call void @objc_release
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test43(i8* %p) {
entry:
call i8* @objc_retain(i8* %p)
call i8* @objc_autorelease(i8* %p)
call i8* @objc_retain(i8* %p)
call void @use_pointer(i8* %p)
call void @use_pointer(i8* %p)
call void @objc_autoreleasePoolPop(i8* undef)
call void @objc_release(i8* %p)
ret void
}
; Do the known-incremented retain+release elimination if the pointer is
; autoreleased and there's an autoreleasePoolPush.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test43b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: entry:
; CHECK-NEXT: call i8* @objc_retain(i8* %p)
; CHECK-NEXT: call i8* @objc_autorelease(i8* %p)
; CHECK-NEXT: call void @use_pointer(i8* %p)
; CHECK-NEXT: call void @use_pointer(i8* %p)
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
; CHECK-NEXT: call i8* @objc_autoreleasePoolPush()
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: call void @use_pointer(i8* %p)
; CHECK-NEXT: call void @objc_release
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test43b(i8* %p) {
entry:
call i8* @objc_retain(i8* %p)
call i8* @objc_autorelease(i8* %p)
call i8* @objc_retain(i8* %p)
call void @use_pointer(i8* %p)
call void @use_pointer(i8* %p)
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
call i8* @objc_autoreleasePoolPush()
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @objc_release(i8* %p)
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call void @use_pointer(i8* %p)
call void @objc_release(i8* %p)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
ret void
}
; Do retain+release elimination for non-provenance pointers.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test44(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: objc_
; CHECK: }
define void @test44(i8** %pp) {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%p = load i8*, i8** %pp
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%q = call i8* @objc_retain(i8* %p)
call void @objc_release(i8* %q)
ret void
}
; Don't delete retain+release with an unknown-provenance
; may-alias objc_release between them.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test45(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: call i8* @objc_retain(i8* %p)
; CHECK: call void @objc_release(i8* %q)
; CHECK: call void @use_pointer(i8* %p)
; CHECK: call void @objc_release(i8* %p)
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define void @test45(i8** %pp, i8** %qq) {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%p = load i8*, i8** %pp
%q = load i8*, i8** %qq
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call i8* @objc_retain(i8* %p)
call void @objc_release(i8* %q)
call void @use_pointer(i8* %p)
call void @objc_release(i8* %p)
ret void
}
; Don't delete retain and autorelease here.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test46(
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: tail call i8* @objc_retain(i8* %p) [[NUW]]
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: true:
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: call i8* @objc_autorelease(i8* %p) [[NUW]]
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define void @test46(i8* %p, i1 %a) {
entry:
call i8* @objc_retain(i8* %p)
br i1 %a, label %true, label %false
true:
call i8* @objc_autorelease(i8* %p)
call void @use_pointer(i8* %p)
ret void
false:
ret void
}
; Delete no-op cast calls.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define i8* @test47(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: call
; CHECK: ret i8* %p
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define i8* @test47(i8* %p) nounwind {
%x = call i8* @objc_retainedObject(i8* %p)
ret i8* %x
}
; Delete no-op cast calls.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define i8* @test48(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: call
; CHECK: ret i8* %p
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define i8* @test48(i8* %p) nounwind {
%x = call i8* @objc_unretainedObject(i8* %p)
ret i8* %x
}
; Delete no-op cast calls.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define i8* @test49(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: call
; CHECK: ret i8* %p
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define i8* @test49(i8* %p) nounwind {
%x = call i8* @objc_unretainedPointer(i8* %p)
ret i8* %x
}
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; Do delete retain+release with intervening stores of the address value if we
; have imprecise release attached to objc_release.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test50a(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NEXT: call i8* @objc_retain
; CHECK-NEXT: call void @callee
; CHECK-NEXT: store
; CHECK-NEXT: call void @objc_release
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test50a(i8* %p, i8** %pp) {
call i8* @objc_retain(i8* %p)
call void @callee()
store i8* %p, i8** %pp
call void @objc_release(i8* %p)
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test50b(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test50b(i8* %p, i8** %pp) {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call i8* @objc_retain(i8* %p)
call void @callee()
store i8* %p, i8** %pp
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call void @objc_release(i8* %p), !clang.imprecise_release !0
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
ret void
}
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; Don't delete retain+release with intervening stores through the
; address value.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test51a(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: call i8* @objc_retain(i8* %p)
; CHECK: call void @objc_release(i8* %p)
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: ret void
; CHECK: }
define void @test51a(i8* %p) {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call i8* @objc_retain(i8* %p)
call void @callee()
store i8 0, i8* %p
call void @objc_release(i8* %p)
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test51b(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: call i8* @objc_retain(i8* %p)
; CHECK: call void @objc_release(i8* %p)
; CHECK: ret void
; CHECK: }
define void @test51b(i8* %p) {
call i8* @objc_retain(i8* %p)
call void @callee()
store i8 0, i8* %p
call void @objc_release(i8* %p), !clang.imprecise_release !0
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; Don't delete retain+release with intervening use of a pointer of
; unknown provenance.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test52a(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: call i8* @objc_retain
; CHECK: call void @callee()
; CHECK: call void @use_pointer(i8* %z)
; CHECK: call void @objc_release
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: ret void
; CHECK: }
define void @test52a(i8** %zz, i8** %pp) {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%p = load i8*, i8** %pp
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%1 = call i8* @objc_retain(i8* %p)
call void @callee()
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%z = load i8*, i8** %zz
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @use_pointer(i8* %z)
call void @objc_release(i8* %p)
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test52b(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: call i8* @objc_retain
; CHECK: call void @callee()
; CHECK: call void @use_pointer(i8* %z)
; CHECK: call void @objc_release
; CHECK: ret void
; CHECK: }
define void @test52b(i8** %zz, i8** %pp) {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%p = load i8*, i8** %pp
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
%1 = call i8* @objc_retain(i8* %p)
call void @callee()
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%z = load i8*, i8** %zz
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call void @use_pointer(i8* %z)
call void @objc_release(i8* %p), !clang.imprecise_release !0
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; Like test52, but the pointer has function type, so it's assumed to
; be not reference counted.
It turns out that clang does use pointer-to-function types to point to ARC-managed pointers sometimes. This fixes rdar://10551239. llvm-svn: 146577
2011-12-15 03:10:53 +08:00
; Oops. That's wrong. Clang sometimes uses function types gratuitously.
; See rdar://10551239.
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test53(
It turns out that clang does use pointer-to-function types to point to ARC-managed pointers sometimes. This fixes rdar://10551239. llvm-svn: 146577
2011-12-15 03:10:53 +08:00
; CHECK: @objc_
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: }
define void @test53(void ()** %zz, i8** %pp) {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%p = load i8*, i8** %pp
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%1 = call i8* @objc_retain(i8* %p)
call void @callee()
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%z = load void ()*, void ()** %zz
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
call void @callee_fnptr(void ()* %z)
call void @objc_release(i8* %p)
ret void
}
; Convert autorelease to release if the value is unused.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test54(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: call i8* @returner()
Debug Info: update testing cases to specify the debug info version number. We are going to drop debug info without a version number or with a different version number, to make sure we don't crash when we see bitcode files with different debug info metadata format. Make tests more robust by removing hard-coded metadata numbers in CHECK lines. llvm-svn: 195535
2013-11-23 09:16:29 +08:00
; CHECK-NEXT: call void @objc_release(i8* %t) [[NUW]], !clang.imprecise_release ![[RELEASE]]
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: ret void
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define void @test54() {
%t = call i8* @returner()
call i8* @objc_autorelease(i8* %t)
ret void
}
; Nested retain+release pairs. Delete them both.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test55(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc
; CHECK: }
define void @test55(i8* %x) {
entry:
%0 = call i8* @objc_retain(i8* %x) nounwind
%1 = call i8* @objc_retain(i8* %x) nounwind
call void @objc_release(i8* %x) nounwind
call void @objc_release(i8* %x) nounwind
ret void
}
; Nested retain+release pairs where the inner pair depends
; on the outer pair to be removed, and then the outer pair
; can be partially eliminated. Plus an extra outer pair to
; eliminate, for fun.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test56(
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc
; CHECK: if.then:
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK-NEXT: %0 = tail call i8* @objc_retain(i8* %x) [[NUW]]
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: tail call void @use_pointer(i8* %x)
; CHECK-NEXT: tail call void @use_pointer(i8* %x)
Debug Info: update testing cases to specify the debug info version number. We are going to drop debug info without a version number or with a different version number, to make sure we don't crash when we see bitcode files with different debug info metadata format. Make tests more robust by removing hard-coded metadata numbers in CHECK lines. llvm-svn: 195535
2013-11-23 09:16:29 +08:00
; CHECK-NEXT: tail call void @objc_release(i8* %x) [[NUW]], !clang.imprecise_release ![[RELEASE]]
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NEXT: br label %if.end
; CHECK-NOT: @objc
; CHECK: }
define void @test56(i8* %x, i32 %n) {
entry:
%0 = tail call i8* @objc_retain(i8* %x) nounwind
%1 = tail call i8* @objc_retain(i8* %0) nounwind
%tobool = icmp eq i32 %n, 0
br i1 %tobool, label %if.end, label %if.then
if.then: ; preds = %entry
%2 = tail call i8* @objc_retain(i8* %1) nounwind
tail call void @use_pointer(i8* %2)
tail call void @use_pointer(i8* %2)
tail call void @objc_release(i8* %2) nounwind, !clang.imprecise_release !0
br label %if.end
if.end: ; preds = %entry, %if.then
tail call void @objc_release(i8* %1) nounwind, !clang.imprecise_release !0
tail call void @objc_release(i8* %0) nounwind, !clang.imprecise_release !0
ret void
}
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; When there are adjacent retain+release pairs, the first one is known
; unnecessary because the presence of the second one means that the first one
; won't be deleting the object.
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test57(
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
; CHECK-NEXT: entry:
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: tail call i8* @objc_retain(i8* %x) [[NUW]]
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
; CHECK-NEXT: call void @use_pointer(i8* %x)
; CHECK-NEXT: call void @use_pointer(i8* %x)
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: tail call i8* @objc_retain(i8* %x) [[NUW]]
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
; CHECK-NEXT: call void @use_pointer(i8* %x)
; CHECK-NEXT: call void @use_pointer(i8* %x)
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK-NEXT: call void @objc_release(i8* %x) [[NUW]]
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test57(i8* %x) nounwind {
entry:
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call i8* @objc_retain(i8* %x) nounwind
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
ret void
}
; An adjacent retain+release pair is sufficient even if it will be
; removed itself.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test58(
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
; CHECK-NEXT: entry:
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NEXT: @objc_retain
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
; CHECK-NEXT: call void @use_pointer(i8* %x)
; CHECK-NEXT: call void @use_pointer(i8* %x)
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test58(i8* %x) nounwind {
entry:
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call i8* @objc_retain(i8* %x) nounwind
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
call i8* @objc_retain(i8* %x) nounwind
call void @objc_release(i8* %x) nounwind
ret void
}
; Don't delete the second retain+release pair in an adjacent set.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test59(
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
; CHECK-NEXT: entry:
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK-NEXT: %0 = tail call i8* @objc_retain(i8* %x) [[NUW]]
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
; CHECK-NEXT: call void @use_pointer(i8* %x)
; CHECK-NEXT: call void @use_pointer(i8* %x)
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK-NEXT: call void @objc_release(i8* %x) [[NUW]]
Track a retain+release nesting level independently of the known-incremented level, because the two concepts can be used to prove the saftey of a retain+release removal in different ways. llvm-svn: 138016
2011-08-19 08:26:36 +08:00
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test59(i8* %x) nounwind {
entry:
%a = call i8* @objc_retain(i8* %x) nounwind
call void @objc_release(i8* %x) nounwind
%b = call i8* @objc_retain(i8* %x) nounwind
call void @use_pointer(i8* %x)
call void @use_pointer(i8* %x)
call void @objc_release(i8* %x) nounwind
ret void
}
Constant pointers to objects don't need reference counting. llvm-svn: 138242
2011-08-23 01:29:11 +08:00
; Constant pointers to objects don't need reference counting.
@constptr = external constant i8*
@something = external global i8*
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; We have a precise lifetime retain/release here. We can not remove them since
; @something is not constant.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test60a(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: call i8* @objc_retain
; CHECK: call void @objc_release
; CHECK: }
define void @test60a() {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%t = load i8*, i8** @constptr
%s = load i8*, i8** @something
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call i8* @objc_retain(i8* %s)
call void @callee()
call void @use_pointer(i8* %t)
call void @objc_release(i8* %s)
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test60b(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK: call i8* @objc_retain
; CHECK-NOT: call i8* @objc_retain
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
; CHECK-NOT: call i8* @objc_release
Constant pointers to objects don't need reference counting. llvm-svn: 138242
2011-08-23 01:29:11 +08:00
; CHECK: }
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
define void @test60b() {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%t = load i8*, i8** @constptr
%s = load i8*, i8** @something
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call i8* @objc_retain(i8* %t)
call i8* @objc_retain(i8* %t)
Constant pointers to objects don't need reference counting. llvm-svn: 138242
2011-08-23 01:29:11 +08:00
call void @callee()
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call void @use_pointer(i8* %s)
call void @objc_release(i8* %t)
Constant pointers to objects don't need reference counting. llvm-svn: 138242
2011-08-23 01:29:11 +08:00
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test60c(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test60c() {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%t = load i8*, i8** @constptr
%s = load i8*, i8** @something
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call i8* @objc_retain(i8* %t)
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call void @callee()
[objc-arc] Make the ARC optimizer more conservative by forcing it to be non-safe in both direction, but mitigate the problem by noting that we just care if there was a further use. The problem here is the infamous one direction known safe. I was hesitant to turn it off before b/c of the potential for regressions without an actual bug from users hitting the problem. This is that bug ; ). The main performance impact of having known safe in both directions is that often times it is very difficult to find two releases without a use in-between them since we are so conservative with determining potential uses. The one direction known safe gets around that problem by taking advantage of many situations where we have two retains in a row, allowing us to avoid that problem. That being said, the one direction known safe is unsafe. Consider the following situation: retain(x) retain(x) call(x) call(x) release(x) Then we know the following about the reference count of x: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 1 (since we can not release a deallocated pointer). release(x) // rc(x) >= 0 That is all the information that we can know statically. That means that we know that A(x), B(x) together can release (x) at most N+1 times. Lets say that we remove the inner retain, release pair. // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 release(x) // rc(x) >= 0 We knew before that A(x), B(x) could release x up to N+1 times meaning that rc(x) may be zero at the release(x). That is not safe. On the other hand, consider the following situation where we have a must use of release(x) that x must be kept alive for after the release(x)**. Then we know that: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 retain(x) // rc(x) == N+2 call A(x) call B(x) // rc(x) >= 2 (since we know that we are going to release x and that that release can not be the last use of x). release(x) // rc(x) >= 1 (since we can not deallocate the pointer since we have a must use after x). … // rc(x) >= 1 use(x) Thus we know that statically the calls to A(x), B(x) can together only release rc(x) N times. Thus if we remove the inner retain, release pair: // rc(x) == N (for some N). retain(x) // rc(x) == N+1 call A(x) call B(x) // rc(x) >= 1 … // rc(x) >= 1 use(x) We are still safe unless in the final … there are unbalanced retains, releases which would have caused the program to blow up anyways even before optimization occurred. The simplest form of must use is an additional release that has not been paired up with any retain (if we had paired the release with a retain and removed it we would not have the additional use). This fits nicely into the ARC framework since basically what you do is say that given any nested releases regardless of what is in between, the inner release is known safe. This enables us to get back the lost performance. <rdar://problem/19023795> llvm-svn: 232351
2015-03-16 15:02:36 +08:00
call void @use_pointer(i8* %s)
call void @objc_release(i8* %t), !clang.imprecise_release !0
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test60d(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test60d() {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%t = load i8*, i8** @constptr
%s = load i8*, i8** @something
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call i8* @objc_retain(i8* %t)
call void @callee()
call void @use_pointer(i8* %s)
call void @objc_release(i8* %t)
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test60e(
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test60e() {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%t = load i8*, i8** @constptr
%s = load i8*, i8** @something
Properly model precise lifetime when given an incomplete dataflow sequence. The normal dataflow sequence in the ARC optimizer consists of the following states: Retain -> CanRelease -> Use -> Release The optimizer before this patch stored the uses that determine the lifetime of the retainable object pointer when it bottom up hits a retain or when top down it hits a release. This is correct for an imprecise lifetime scenario since what we are trying to do is remove retains/releases while making sure that no ``CanRelease'' (which is usually a call) deallocates the given pointer before we get to the ``Use'' (since that would cause a segfault). If we are considering the precise lifetime scenario though, this is not correct. In such a situation, we *DO* care about the previous sequence, but additionally, we wish to track the uses resulting from the following incomplete sequences: Retain -> CanRelease -> Release (TopDown) Retain <- Use <- Release (BottomUp) *NOTE* This patch looks large but the most of it consists of updating test cases. Additionally this fix exposed an additional bug. I removed the test case that expressed said bug and will recommit it with the fix in a little bit. llvm-svn: 178921
2013-04-06 06:54:28 +08:00
call i8* @objc_retain(i8* %t)
call void @callee()
call void @use_pointer(i8* %s)
call void @objc_release(i8* %t), !clang.imprecise_release !0
ret void
}
Constant pointers to objects don't need reference counting. llvm-svn: 138242
2011-08-23 01:29:11 +08:00
; Constant pointers to objects don't need to be considered related to other
; pointers.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test61(
Constant pointers to objects don't need reference counting. llvm-svn: 138242
2011-08-23 01:29:11 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test61() {
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%t = load i8*, i8** @constptr
Constant pointers to objects don't need reference counting. llvm-svn: 138242
2011-08-23 01:29:11 +08:00
call i8* @objc_retain(i8* %t)
call void @callee()
call void @use_pointer(i8* %t)
call void @objc_release(i8* %t)
ret void
}
When identifying exit nodes for the reverse-CFG reverse-post-order traversal, consider nodes for which the only successors are backedges which the traversal is ignoring to be exit nodes. This fixes a problem where the bottom-up traversal was failing to visit split blocks along split loop backedges. This fixes rdar://10989035. llvm-svn: 152421
2012-03-10 02:50:52 +08:00
; Delete a retain matched by releases when one is inside the loop and the
; other is outside the loop.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test62(
When identifying exit nodes for the reverse-CFG reverse-post-order traversal, consider nodes for which the only successors are backedges which the traversal is ignoring to be exit nodes. This fixes a problem where the bottom-up traversal was failing to visit split blocks along split loop backedges. This fixes rdar://10989035. llvm-svn: 152421
2012-03-10 02:50:52 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define void @test62(i8* %x, i1* %p) nounwind {
entry:
br label %loop
loop:
call i8* @objc_retain(i8* %x)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%q = load i1, i1* %p
When identifying exit nodes for the reverse-CFG reverse-post-order traversal, consider nodes for which the only successors are backedges which the traversal is ignoring to be exit nodes. This fixes a problem where the bottom-up traversal was failing to visit split blocks along split loop backedges. This fixes rdar://10989035. llvm-svn: 152421
2012-03-10 02:50:52 +08:00
br i1 %q, label %loop.more, label %exit
loop.more:
call void @objc_release(i8* %x)
br label %loop
exit:
call void @objc_release(i8* %x)
ret void
}
; Like test62 but with no release in exit.
; Don't delete anything!
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test63(
When identifying exit nodes for the reverse-CFG reverse-post-order traversal, consider nodes for which the only successors are backedges which the traversal is ignoring to be exit nodes. This fixes a problem where the bottom-up traversal was failing to visit split blocks along split loop backedges. This fixes rdar://10989035. llvm-svn: 152421
2012-03-10 02:50:52 +08:00
; CHECK: loop:
; CHECK: tail call i8* @objc_retain(i8* %x)
; CHECK: loop.more:
; CHECK: call void @objc_release(i8* %x)
; CHECK: }
define void @test63(i8* %x, i1* %p) nounwind {
entry:
br label %loop
loop:
call i8* @objc_retain(i8* %x)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%q = load i1, i1* %p
When identifying exit nodes for the reverse-CFG reverse-post-order traversal, consider nodes for which the only successors are backedges which the traversal is ignoring to be exit nodes. This fixes a problem where the bottom-up traversal was failing to visit split blocks along split loop backedges. This fixes rdar://10989035. llvm-svn: 152421
2012-03-10 02:50:52 +08:00
br i1 %q, label %loop.more, label %exit
loop.more:
call void @objc_release(i8* %x)
br label %loop
exit:
ret void
}
; Like test62 but with no release in loop.more.
; Don't delete anything!
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test64(
When identifying exit nodes for the reverse-CFG reverse-post-order traversal, consider nodes for which the only successors are backedges which the traversal is ignoring to be exit nodes. This fixes a problem where the bottom-up traversal was failing to visit split blocks along split loop backedges. This fixes rdar://10989035. llvm-svn: 152421
2012-03-10 02:50:52 +08:00
; CHECK: loop:
; CHECK: tail call i8* @objc_retain(i8* %x)
; CHECK: exit:
; CHECK: call void @objc_release(i8* %x)
; CHECK: }
define void @test64(i8* %x, i1* %p) nounwind {
entry:
br label %loop
loop:
call i8* @objc_retain(i8* %x)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%q = load i1, i1* %p
When identifying exit nodes for the reverse-CFG reverse-post-order traversal, consider nodes for which the only successors are backedges which the traversal is ignoring to be exit nodes. This fixes a problem where the bottom-up traversal was failing to visit split blocks along split loop backedges. This fixes rdar://10989035. llvm-svn: 152421
2012-03-10 02:50:52 +08:00
br i1 %q, label %loop.more, label %exit
loop.more:
br label %loop
exit:
call void @objc_release(i8* %x)
ret void
}
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
; Move an autorelease past a phi with a null.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define i8* @test65(
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
; CHECK: if.then:
; CHECK: call i8* @objc_autorelease(
; CHECK: return:
; CHECK-NOT: @objc_autorelease
; CHECK: }
define i8* @test65(i1 %x) {
entry:
br i1 %x, label %return, label %if.then
if.then: ; preds = %entry
%c = call i8* @returner()
%s = call i8* @objc_retainAutoreleasedReturnValue(i8* %c) nounwind
br label %return
return: ; preds = %if.then, %entry
%retval = phi i8* [ %s, %if.then ], [ null, %entry ]
%q = call i8* @objc_autorelease(i8* %retval) nounwind
ret i8* %retval
}
; Don't move an autorelease past an autorelease pool boundary.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define i8* @test65b(
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
; CHECK: if.then:
; CHECK-NOT: @objc_autorelease
; CHECK: return:
; CHECK: call i8* @objc_autorelease(
; CHECK: }
define i8* @test65b(i1 %x) {
entry:
%t = call i8* @objc_autoreleasePoolPush()
br i1 %x, label %return, label %if.then
if.then: ; preds = %entry
%c = call i8* @returner()
%s = call i8* @objc_retainAutoreleasedReturnValue(i8* %c) nounwind
br label %return
return: ; preds = %if.then, %entry
%retval = phi i8* [ %s, %if.then ], [ null, %entry ]
call void @objc_autoreleasePoolPop(i8* %t)
%q = call i8* @objc_autorelease(i8* %retval) nounwind
ret i8* %retval
}
; Don't move an autoreleaseReuturnValue, which would break
; the RV optimization.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define i8* @test65c(
Don't move objc_autorelease calls past autorelease pool boundaries when optimizing autorelease calls on phi nodes with null operands. This fixes rdar://11207070. llvm-svn: 154642
2012-04-13 08:59:57 +08:00
; CHECK: if.then:
; CHECK-NOT: @objc_autorelease
; CHECK: return:
; CHECK: call i8* @objc_autoreleaseReturnValue(
; CHECK: }
define i8* @test65c(i1 %x) {
entry:
br i1 %x, label %return, label %if.then
if.then: ; preds = %entry
%c = call i8* @returner()
%s = call i8* @objc_retainAutoreleasedReturnValue(i8* %c) nounwind
br label %return
return: ; preds = %if.then, %entry
%retval = phi i8* [ %s, %if.then ], [ null, %entry ]
%q = call i8* @objc_autoreleaseReturnValue(i8* %retval) nounwind
ret i8* %retval
}
Add support for objc_unsafeClaimAutoreleasedReturnValue to the ObjC ARC Optimizer. The main implication of this is: 1. Ensuring that we treat it conservatively in terms of optimization. 2. We put the ASM marker on it so that the runtime can recognize objc_unsafeClaimAutoreleasedReturnValue from releaseRV. <rdar://problem/21567064> Patch by Michael Gottesman! llvm-svn: 258970
2016-01-28 03:05:08 +08:00
; CHECK-LABEL: define i8* @test65d(
; CHECK: if.then:
; CHECK-NOT: @objc_autorelease
; CHECK: return:
; CHECK: call i8* @objc_autoreleaseReturnValue(
; CHECK: }
define i8* @test65d(i1 %x) {
entry:
br i1 %x, label %return, label %if.then
if.then: ; preds = %entry
%c = call i8* @returner()
%s = call i8* @objc_unsafeClaimAutoreleasedReturnValue(i8* %c) nounwind
br label %return
return: ; preds = %if.then, %entry
%retval = phi i8* [ %s, %if.then ], [ null, %entry ]
%q = call i8* @objc_autoreleaseReturnValue(i8* %retval) nounwind
ret i8* %retval
}
An objc_retain can serve as a use for a different pointer. This is the counterpart to commit r160637, except it performs the action in the bottomup portion of the data flow analysis. llvm-svn: 178922
2013-04-06 06:54:32 +08:00
; An objc_retain can serve as a may-use for a different pointer.
; rdar://11931823
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test66a(
An objc_retain can serve as a use for a different pointer. This is the counterpart to commit r160637, except it performs the action in the bottomup portion of the data flow analysis. llvm-svn: 178922
2013-04-06 06:54:32 +08:00
; CHECK: tail call i8* @objc_retain(i8* %cond) [[NUW]]
; CHECK: tail call void @objc_release(i8* %call) [[NUW]]
; CHECK: tail call i8* @objc_retain(i8* %tmp8) [[NUW]]
; CHECK: tail call void @objc_release(i8* %cond) [[NUW]]
; CHECK: }
define void @test66a(i8* %tmp5, i8* %bar, i1 %tobool, i1 %tobool1, i8* %call) {
entry:
br i1 %tobool, label %cond.true, label %cond.end
cond.true:
br label %cond.end
cond.end: ; preds = %cond.true, %entry
%cond = phi i8* [ %tmp5, %cond.true ], [ %call, %entry ]
%tmp7 = tail call i8* @objc_retain(i8* %cond) nounwind
tail call void @objc_release(i8* %call) nounwind
%tmp8 = select i1 %tobool1, i8* %cond, i8* %bar
%tmp9 = tail call i8* @objc_retain(i8* %tmp8) nounwind
tail call void @objc_release(i8* %cond) nounwind
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test66b(
An objc_retain can serve as a use for a different pointer. This is the counterpart to commit r160637, except it performs the action in the bottomup portion of the data flow analysis. llvm-svn: 178922
2013-04-06 06:54:32 +08:00
; CHECK: tail call i8* @objc_retain(i8* %cond) [[NUW]]
; CHECK: tail call void @objc_release(i8* %call) [[NUW]]
; CHECK: tail call i8* @objc_retain(i8* %tmp8) [[NUW]]
; CHECK: tail call void @objc_release(i8* %cond) [[NUW]]
; CHECK: }
define void @test66b(i8* %tmp5, i8* %bar, i1 %tobool, i1 %tobool1, i8* %call) {
entry:
br i1 %tobool, label %cond.true, label %cond.end
cond.true:
br label %cond.end
cond.end: ; preds = %cond.true, %entry
%cond = phi i8* [ %tmp5, %cond.true ], [ %call, %entry ]
%tmp7 = tail call i8* @objc_retain(i8* %cond) nounwind
tail call void @objc_release(i8* %call) nounwind, !clang.imprecise_release !0
%tmp8 = select i1 %tobool1, i8* %cond, i8* %bar
%tmp9 = tail call i8* @objc_retain(i8* %tmp8) nounwind
tail call void @objc_release(i8* %cond) nounwind
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test66c(
An objc_retain can serve as a use for a different pointer. This is the counterpart to commit r160637, except it performs the action in the bottomup portion of the data flow analysis. llvm-svn: 178922
2013-04-06 06:54:32 +08:00
; CHECK: tail call i8* @objc_retain(i8* %cond) [[NUW]]
; CHECK: tail call void @objc_release(i8* %call) [[NUW]]
; CHECK: tail call i8* @objc_retain(i8* %tmp8) [[NUW]]
; CHECK: tail call void @objc_release(i8* %cond) [[NUW]]
; CHECK: }
define void @test66c(i8* %tmp5, i8* %bar, i1 %tobool, i1 %tobool1, i8* %call) {
entry:
br i1 %tobool, label %cond.true, label %cond.end
cond.true:
br label %cond.end
cond.end: ; preds = %cond.true, %entry
%cond = phi i8* [ %tmp5, %cond.true ], [ %call, %entry ]
%tmp7 = tail call i8* @objc_retain(i8* %cond) nounwind
tail call void @objc_release(i8* %call) nounwind
%tmp8 = select i1 %tobool1, i8* %cond, i8* %bar
%tmp9 = tail call i8* @objc_retain(i8* %tmp8) nounwind, !clang.imprecise_release !0
tail call void @objc_release(i8* %cond) nounwind
ret void
}
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion was done with the following bash script: find test/Transforms -name "*.ll" | \ while read NAME; do echo "$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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test66d(
An objc_retain can serve as a use for a different pointer. This is the counterpart to commit r160637, except it performs the action in the bottomup portion of the data flow analysis. llvm-svn: 178922
2013-04-06 06:54:32 +08:00
; CHECK: tail call i8* @objc_retain(i8* %cond) [[NUW]]
; CHECK: tail call void @objc_release(i8* %call) [[NUW]]
; CHECK: tail call i8* @objc_retain(i8* %tmp8) [[NUW]]
; CHECK: tail call void @objc_release(i8* %cond) [[NUW]]
; CHECK: }
define void @test66d(i8* %tmp5, i8* %bar, i1 %tobool, i1 %tobool1, i8* %call) {
entry:
br i1 %tobool, label %cond.true, label %cond.end
cond.true:
br label %cond.end
cond.end: ; preds = %cond.true, %entry
%cond = phi i8* [ %tmp5, %cond.true ], [ %call, %entry ]
%tmp7 = tail call i8* @objc_retain(i8* %cond) nounwind
tail call void @objc_release(i8* %call) nounwind, !clang.imprecise_release !0
%tmp8 = select i1 %tobool1, i8* %cond, i8* %bar
%tmp9 = tail call i8* @objc_retain(i8* %tmp8) nounwind
tail call void @objc_release(i8* %cond) nounwind, !clang.imprecise_release !0
ret void
}
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; A few real-world testcases.
@.str4 = private unnamed_addr constant [33 x i8] c"-[A z] = { %f, %f, { %f, %f } }\0A\00"
@"OBJC_IVAR_$_A.myZ" = global i64 20, section "__DATA, __objc_const", align 8
declare i32 @printf(i8* nocapture, ...) nounwind
declare i32 @puts(i8* nocapture) nounwind
@str = internal constant [16 x i8] c"-[ Top0 _getX ]\00"
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: define { <2 x float>, <2 x float> } @"\01-[A z]"({}* %self, i8* nocapture %_cmd) [[NUW]] {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define {<2 x float>, <2 x float>} @"\01-[A z]"({}* %self, i8* nocapture %_cmd) nounwind {
invoke.cont:
%0 = bitcast {}* %self to i8*
%1 = tail call i8* @objc_retain(i8* %0) nounwind
Remove the obsolete offset parameter from @llvm.dbg.value There is no situation where this rarely-used argument cannot be substituted with a DIExpression and removing it allows us to simplify the DWARF backend. Note that this patch does not yet remove any of the newly dead code. rdar://problem/33580047 Differential Revision: https://reviews.llvm.org/D35951 llvm-svn: 309426
2017-07-29 04:21:02 +08:00
tail call void @llvm.dbg.value(metadata {}* %self, metadata !DILocalVariable(scope: !2), metadata !DIExpression()), !dbg !DILocation(scope: !2)
tail call void @llvm.dbg.value(metadata {}* %self, metadata !DILocalVariable(scope: !2), metadata !DIExpression()), !dbg !DILocation(scope: !2)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%ivar = load i64, i64* @"OBJC_IVAR_$_A.myZ", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%add.ptr = getelementptr i8, i8* %0, i64 %ivar
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp1 = bitcast i8* %add.ptr to float*
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp2 = load float, float* %tmp1, align 4
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%conv = fpext float %tmp2 to double
%add.ptr.sum = add i64 %ivar, 4
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%tmp6 = getelementptr inbounds i8, i8* %0, i64 %add.ptr.sum
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%2 = bitcast i8* %tmp6 to float*
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp7 = load float, float* %2, align 4
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%conv8 = fpext float %tmp7 to double
%add.ptr.sum36 = add i64 %ivar, 8
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%tmp12 = getelementptr inbounds i8, i8* %0, i64 %add.ptr.sum36
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%arrayidx = bitcast i8* %tmp12 to float*
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp13 = load float, float* %arrayidx, align 4
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%conv14 = fpext float %tmp13 to double
%tmp12.sum = add i64 %ivar, 12
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx19 = getelementptr inbounds i8, i8* %0, i64 %tmp12.sum
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%3 = bitcast i8* %arrayidx19 to float*
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp20 = load float, float* %3, align 4
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%conv21 = fpext float %tmp20 to double
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call = tail call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([33 x i8], [33 x i8]* @.str4, i64 0, i64 0), double %conv, double %conv8, double %conv14, double %conv21)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%ivar23 = load i64, i64* @"OBJC_IVAR_$_A.myZ", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%add.ptr24 = getelementptr i8, i8* %0, i64 %ivar23
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%4 = bitcast i8* %add.ptr24 to i128*
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%srcval = load i128, i128* %4, align 4
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
tail call void @objc_release(i8* %0) nounwind
%tmp29 = trunc i128 %srcval to i64
%tmp30 = bitcast i64 %tmp29 to <2 x float>
%tmp31 = insertvalue {<2 x float>, <2 x float>} undef, <2 x float> %tmp30, 0
%tmp32 = lshr i128 %srcval, 64
%tmp33 = trunc i128 %tmp32 to i64
%tmp34 = bitcast i64 %tmp33 to <2 x float>
%tmp35 = insertvalue {<2 x float>, <2 x float>} %tmp31, <2 x float> %tmp34, 1
ret {<2 x float>, <2 x float>} %tmp35
}
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: @"\01-[Top0 _getX]"({}* %self, i8* nocapture %_cmd) [[NUW]] {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: @objc_
; CHECK: }
define i32 @"\01-[Top0 _getX]"({}* %self, i8* nocapture %_cmd) nounwind {
invoke.cont:
%0 = bitcast {}* %self to i8*
%1 = tail call i8* @objc_retain(i8* %0) nounwind
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
%puts = tail call i32 @puts(i8* getelementptr inbounds ([16 x i8], [16 x i8]* @str, i64 0, i64 0))
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
tail call void @objc_release(i8* %0) nounwind
ret i32 0
}
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
@"\01L_OBJC_METH_VAR_NAME_" = internal global [5 x i8] c"frob\00", section "__TEXT,__cstring,cstring_literals", align 1@"\01L_OBJC_SELECTOR_REFERENCES_" = internal global i8* getelementptr inbounds ([5 x i8], [5 x i8]* @"\01L_OBJC_METH_VAR_NAME_", i64 0, i64 0), section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
@"\01L_OBJC_IMAGE_INFO" = internal constant [2 x i32] [i32 0, i32 16], section "__DATA, __objc_imageinfo, regular, no_dead_strip"
[opaque pointer type] Add textual IR support for explicit type parameter to gep operator Similar to gep (r230786) and load (r230794) changes. Similar migration script can be used to update test cases, which successfully migrated all of LLVM and Polly, but about 4 test cases needed manually changes in Clang. (this script will read the contents of stdin and massage it into stdout - wrap it in the 'apply.sh' script shown in previous commits + xargs to apply it over a large set of test cases) import fileinput import sys import re rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL) def conv(match): line = match.group(1) line += match.group(4) line += ", " line += match.group(2) return line line = sys.stdin.read() off = 0 for match in re.finditer(rep, line): sys.stdout.write(line[off:match.start()]) sys.stdout.write(conv(match)) off = match.end() sys.stdout.write(line[off:]) llvm-svn: 232184
2015-03-14 02:20:45 +08:00
@llvm.used = appending global [3 x i8*] [i8* getelementptr inbounds ([5 x i8], [5 x i8]* @"\01L_OBJC_METH_VAR_NAME_", i32 0, i32 0), i8* bitcast (i8** @"\01L_OBJC_SELECTOR_REFERENCES_" to i8*), i8* bitcast ([2 x i32]* @"\01L_OBJC_IMAGE_INFO" to i8*)], section "llvm.metadata"
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; A simple loop. Eliminate the retain and release inside of it!
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: define void @loop(i8* %x, i64 %n) {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK: for.body:
; CHECK-NOT: @objc_
; CHECK: @objc_msgSend
; CHECK-NOT: @objc_
; CHECK: for.end:
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: }
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
define void @loop(i8* %x, i64 %n) {
entry:
%0 = tail call i8* @objc_retain(i8* %x) nounwind
%cmp9 = icmp sgt i64 %n, 0
br i1 %cmp9, label %for.body, label %for.end
for.body: ; preds = %entry, %for.body
%i.010 = phi i64 [ %inc, %for.body ], [ 0, %entry ]
%1 = tail call i8* @objc_retain(i8* %x) nounwind
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp5 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call = tail call i8* (i8*, i8*, ...) @objc_msgSend(i8* %1, i8* %tmp5)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
tail call void @objc_release(i8* %1) nounwind, !clang.imprecise_release !0
%inc = add nsw i64 %i.010, 1
%exitcond = icmp eq i64 %inc, %n
br i1 %exitcond, label %for.end, label %for.body
for.end: ; preds = %for.body, %entry
tail call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
ret void
}
; ObjCARCOpt can delete the retain,release on self.
[objc-arc] Test cleanups. Mainly adding paranoid checks for the closing brace of a function to help with FileCheck error readability. Also some other minor changes. No actual CHECK changes. llvm-svn: 180668
2013-04-27 13:25:54 +08:00
; CHECK: define void @TextEditTest(%2* %self, %3* %pboard) {
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
; CHECK-NOT: call i8* @objc_retain(i8* %tmp7)
; CHECK: }
%0 = type { i8* (i8*, %struct._message_ref_t*, ...)*, i8* }
%1 = type opaque
%2 = type opaque
%3 = type opaque
%4 = type opaque
%5 = type opaque
%struct.NSConstantString = type { i32*, i32, i8*, i64 }
%struct._NSRange = type { i64, i64 }
%struct.__CFString = type opaque
%struct.__method_list_t = type { i32, i32, [0 x %struct._objc_method] }
%struct._class_ro_t = type { i32, i32, i32, i8*, i8*, %struct.__method_list_t*, %struct._objc_protocol_list*, %struct._ivar_list_t*, i8*, %struct._prop_list_t* }
%struct._class_t = type { %struct._class_t*, %struct._class_t*, %struct._objc_cache*, i8* (i8*, i8*)**, %struct._class_ro_t* }
%struct._ivar_list_t = type { i32, i32, [0 x %struct._ivar_t] }
%struct._ivar_t = type { i64*, i8*, i8*, i32, i32 }
%struct._message_ref_t = type { i8*, i8* }
%struct._objc_cache = type opaque
%struct._objc_method = type { i8*, i8*, i8* }
%struct._objc_protocol_list = type { i64, [0 x %struct._protocol_t*] }
%struct._prop_list_t = type { i32, i32, [0 x %struct._message_ref_t] }
%struct._protocol_t = type { i8*, i8*, %struct._objc_protocol_list*, %struct.__method_list_t*, %struct.__method_list_t*, %struct.__method_list_t*, %struct.__method_list_t*, %struct._prop_list_t*, i32, i32 }
@"\01L_OBJC_CLASSLIST_REFERENCES_$_17" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
@kUTTypePlainText = external constant %struct.__CFString*
@"\01L_OBJC_SELECTOR_REFERENCES_19" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_21" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_23" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_25" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_CLASSLIST_REFERENCES_$_26" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
@"\01L_OBJC_SELECTOR_REFERENCES_28" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_CLASSLIST_REFERENCES_$_29" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
@"\01L_OBJC_SELECTOR_REFERENCES_31" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_33" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_35" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_37" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_CLASSLIST_REFERENCES_$_38" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
@"\01L_OBJC_SELECTOR_REFERENCES_40" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_42" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@_unnamed_cfstring_44 = external hidden constant %struct.NSConstantString, section "__DATA,__cfstring"
@"\01L_OBJC_SELECTOR_REFERENCES_46" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_48" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01l_objc_msgSend_fixup_isEqual_" = external hidden global %0, section "__DATA, __objc_msgrefs, coalesced", align 16
@"\01L_OBJC_CLASSLIST_REFERENCES_$_50" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
@NSCocoaErrorDomain = external constant %1*
@"\01L_OBJC_CLASSLIST_REFERENCES_$_51" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
@NSFilePathErrorKey = external constant %1*
@"\01L_OBJC_SELECTOR_REFERENCES_53" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_55" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_CLASSLIST_REFERENCES_$_56" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
@"\01L_OBJC_SELECTOR_REFERENCES_58" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
@"\01L_OBJC_SELECTOR_REFERENCES_60" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
declare %1* @truncatedString(%1*, i64)
define void @TextEditTest(%2* %self, %3* %pboard) {
entry:
%err = alloca %4*, align 8
%tmp7 = bitcast %2* %self to i8*
%tmp8 = call i8* @objc_retain(i8* %tmp7) nounwind
store %4* null, %4** %err, align 8
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp1 = load %struct._class_t*, %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_17", align 8
%tmp2 = load %struct.__CFString*, %struct.__CFString** @kUTTypePlainText, align 8
%tmp3 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_19", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp4 = bitcast %struct._class_t* %tmp1 to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call5 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp4, i8* %tmp3, %struct.__CFString* %tmp2)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp5 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_21", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp6 = bitcast %3* %pboard to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call76 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp6, i8* %tmp5, i8* %call5)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp9 = call i8* @objc_retain(i8* %call76) nounwind
%tobool = icmp eq i8* %tmp9, null
br i1 %tobool, label %end, label %land.lhs.true
land.lhs.true: ; preds = %entry
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp11 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_23", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call137 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp6, i8* %tmp11, i8* %tmp9)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp = bitcast i8* %call137 to %1*
%tmp10 = call i8* @objc_retain(i8* %call137) nounwind
call void @objc_release(i8* null) nounwind
%tmp12 = call i8* @objc_retain(i8* %call137) nounwind
call void @objc_release(i8* null) nounwind
%tobool16 = icmp eq i8* %call137, null
br i1 %tobool16, label %end, label %if.then
if.then: ; preds = %land.lhs.true
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp19 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_25", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%call21 = call signext i8 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8 (i8*, i8*)*)(i8* %call137, i8* %tmp19)
%tobool22 = icmp eq i8 %call21, 0
br i1 %tobool22, label %if.then44, label %land.lhs.true23
land.lhs.true23: ; preds = %if.then
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp24 = load %struct._class_t*, %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_26", align 8
%tmp26 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_28", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp27 = bitcast %struct._class_t* %tmp24 to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call2822 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp27, i8* %tmp26, i8* %call137)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp13 = bitcast i8* %call2822 to %5*
%tmp14 = call i8* @objc_retain(i8* %call2822) nounwind
call void @objc_release(i8* null) nounwind
%tobool30 = icmp eq i8* %call2822, null
br i1 %tobool30, label %if.then44, label %if.end
if.end: ; preds = %land.lhs.true23
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp32 = load %struct._class_t*, %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_29", align 8
%tmp33 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_31", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp34 = bitcast %struct._class_t* %tmp32 to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call35 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp34, i8* %tmp33)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp37 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_33", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call3923 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %call35, i8* %tmp37, i8* %call2822, i32 signext 1, %4** %err)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%cmp = icmp eq i8* %call3923, null
br i1 %cmp, label %if.then44, label %end
if.then44: ; preds = %if.end, %land.lhs.true23, %if.then
%url.025 = phi %5* [ %tmp13, %if.end ], [ %tmp13, %land.lhs.true23 ], [ null, %if.then ]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp49 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_35", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%call51 = call %struct._NSRange bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %struct._NSRange (i8*, i8*, i64, i64)*)(i8* %call137, i8* %tmp49, i64 0, i64 0)
%call513 = extractvalue %struct._NSRange %call51, 0
%call514 = extractvalue %struct._NSRange %call51, 1
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp52 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_37", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call548 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %call137, i8* %tmp52, i64 %call513, i64 %call514)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp55 = load %struct._class_t*, %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_38", align 8
%tmp56 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_40", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp57 = bitcast %struct._class_t* %tmp55 to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call58 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp57, i8* %tmp56)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp59 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_42", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call6110 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %call548, i8* %tmp59, i8* %call58)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp15 = call i8* @objc_retain(i8* %call6110) nounwind
call void @objc_release(i8* %call137) nounwind
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp64 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_46", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%call66 = call signext i8 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8 (i8*, i8*, %1*)*)(i8* %call6110, i8* %tmp64, %1* bitcast (%struct.NSConstantString* @_unnamed_cfstring_44 to %1*))
%tobool67 = icmp eq i8 %call66, 0
br i1 %tobool67, label %if.end74, label %if.then68
if.then68: ; preds = %if.then44
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp70 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_48", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call7220 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %call6110, i8* %tmp70)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp16 = call i8* @objc_retain(i8* %call7220) nounwind
call void @objc_release(i8* %call6110) nounwind
br label %if.end74
if.end74: ; preds = %if.then68, %if.then44
%filename.0.in = phi i8* [ %call7220, %if.then68 ], [ %call6110, %if.then44 ]
%filename.0 = bitcast i8* %filename.0.in to %1*
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp17 = load i8*, i8** bitcast (%0* @"\01l_objc_msgSend_fixup_isEqual_" to i8**), align 16
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp18 = bitcast i8* %tmp17 to i8 (i8*, %struct._message_ref_t*, i8*, ...)*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call78 = call signext i8 (i8*, %struct._message_ref_t*, i8*, ...) %tmp18(i8* %call137, %struct._message_ref_t* bitcast (%0* @"\01l_objc_msgSend_fixup_isEqual_" to %struct._message_ref_t*), i8* %filename.0.in)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tobool79 = icmp eq i8 %call78, 0
br i1 %tobool79, label %land.lhs.true80, label %if.then109
land.lhs.true80: ; preds = %if.end74
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp82 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_25", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%call84 = call signext i8 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8 (i8*, i8*)*)(i8* %filename.0.in, i8* %tmp82)
%tobool86 = icmp eq i8 %call84, 0
br i1 %tobool86, label %if.then109, label %if.end106
if.end106: ; preds = %land.lhs.true80
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp88 = load %struct._class_t*, %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_26", align 8
%tmp90 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_28", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp91 = bitcast %struct._class_t* %tmp88 to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call9218 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp91, i8* %tmp90, i8* %filename.0.in)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp20 = bitcast i8* %call9218 to %5*
%tmp21 = call i8* @objc_retain(i8* %call9218) nounwind
%tmp22 = bitcast %5* %url.025 to i8*
call void @objc_release(i8* %tmp22) nounwind
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp94 = load %struct._class_t*, %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_29", align 8
%tmp95 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_31", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp96 = bitcast %struct._class_t* %tmp94 to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call97 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp96, i8* %tmp95)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp99 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_33", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call10119 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %call97, i8* %tmp99, i8* %call9218, i32 signext 1, %4** %err)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%phitmp = icmp eq i8* %call10119, null
br i1 %phitmp, label %if.then109, label %end
if.then109: ; preds = %if.end106, %land.lhs.true80, %if.end74
%url.129 = phi %5* [ %tmp20, %if.end106 ], [ %url.025, %if.end74 ], [ %url.025, %land.lhs.true80 ]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp110 = load %4*, %4** %err, align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tobool111 = icmp eq %4* %tmp110, null
br i1 %tobool111, label %if.then112, label %if.end125
if.then112: ; preds = %if.then109
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp113 = load %struct._class_t*, %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_50", align 8
%tmp114 = load %1*, %1** @NSCocoaErrorDomain, align 8
%tmp115 = load %struct._class_t*, %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_51", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%call117 = call %1* @truncatedString(%1* %filename.0, i64 1034)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp118 = load %1*, %1** @NSFilePathErrorKey, align 8
%tmp119 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_53", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp120 = bitcast %struct._class_t* %tmp115 to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call12113 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp120, i8* %tmp119, %1* %call117, %1* %tmp118, i8* null)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp122 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_55", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp123 = bitcast %struct._class_t* %tmp113 to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call12414 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp123, i8* %tmp122, %1* %tmp114, i64 258, i8* %call12113)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp23 = call i8* @objc_retain(i8* %call12414) nounwind
%tmp25 = call i8* @objc_autorelease(i8* %tmp23) nounwind
%tmp28 = bitcast i8* %tmp25 to %4*
store %4* %tmp28, %4** %err, align 8
br label %if.end125
if.end125: ; preds = %if.then112, %if.then109
%tmp127 = phi %4* [ %tmp110, %if.then109 ], [ %tmp28, %if.then112 ]
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp126 = load %struct._class_t*, %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_56", align 8
%tmp128 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_58", align 8
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
%tmp129 = bitcast %struct._class_t* %tmp126 to i8*
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call13015 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %tmp129, i8* %tmp128, %4* %tmp127)
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%tmp131 = load i8*, i8** @"\01L_OBJC_SELECTOR_REFERENCES_60", align 8
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 07:24:18 +08:00
%call13317 = call i8* (i8*, i8*, ...) @objc_msgSend(i8* %call13015, i8* %tmp131)
The ARC language-specific optimizer. Credit to Dan Gohman. llvm-svn: 133108
2011-06-16 07:37:01 +08:00
br label %end
end: ; preds = %if.end125, %if.end106, %if.end, %land.lhs.true, %entry
%filename.2 = phi %1* [ %filename.0, %if.end106 ], [ %filename.0, %if.end125 ], [ %tmp, %land.lhs.true ], [ null, %entry ], [ %tmp, %if.end ]
%origFilename.0 = phi %1* [ %tmp, %if.end106 ], [ %tmp, %if.end125 ], [ %tmp, %land.lhs.true ], [ null, %entry ], [ %tmp, %if.end ]
%url.2 = phi %5* [ %tmp20, %if.end106 ], [ %url.129, %if.end125 ], [ null, %land.lhs.true ], [ null, %entry ], [ %tmp13, %if.end ]
call void @objc_release(i8* %tmp9) nounwind, !clang.imprecise_release !0
%tmp29 = bitcast %5* %url.2 to i8*
call void @objc_release(i8* %tmp29) nounwind, !clang.imprecise_release !0
%tmp30 = bitcast %1* %origFilename.0 to i8*
call void @objc_release(i8* %tmp30) nounwind, !clang.imprecise_release !0
%tmp31 = bitcast %1* %filename.2 to i8*
call void @objc_release(i8* %tmp31) nounwind, !clang.imprecise_release !0
call void @objc_release(i8* %tmp7) nounwind, !clang.imprecise_release !0
ret void
}
Update more tests to the new EH scheme. llvm-svn: 138894
2011-09-01 05:04:11 +08:00
declare i32 @__gxx_personality_v0(...)
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
[objc-arc] Teach the ARC optimizer that objc_sync_enter/objc_sync_exit do not modify the ref count of an objc object and additionally are inert for modref purposes. llvm-svn: 185769
2013-07-07 09:52:55 +08:00
declare i32 @objc_sync_enter(i8*)
declare i32 @objc_sync_exit(i8*)
; Make sure that we understand that objc_sync_{enter,exit} are IC_User not
; IC_Call/IC_CallOrUser.
Catch more CHECK that can be converted to CHECK-LABEL in Transforms for easier debugging. No functionality change. This conversion 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/;\(.*\)\([A-Za-z0-9_]*\):\( *\)define\([^@]*\)@$FUNC\([( ]*\)\$/;\1\2-LABEL:\3define\4@$FUNC(/g" $TEMP done mv $TEMP $NAME fi done llvm-svn: 186269
2013-07-14 09:50:49 +08:00
; CHECK-LABEL: define void @test67(
[objc-arc] Teach the ARC optimizer that objc_sync_enter/objc_sync_exit do not modify the ref count of an objc object and additionally are inert for modref purposes. llvm-svn: 185769
2013-07-07 09:52:55 +08:00
; CHECK-NEXT: call i32 @objc_sync_enter(i8* %x)
; CHECK-NEXT: call i32 @objc_sync_exit(i8* %x)
; CHECK-NEXT: ret void
; CHECK-NEXT: }
define void @test67(i8* %x) {
call i8* @objc_retain(i8* %x)
call i32 @objc_sync_enter(i8* %x)
call i32 @objc_sync_exit(i8* %x)
call void @objc_release(i8* %x), !clang.imprecise_release !0
ret void
}
Debug Info: update testing cases to specify the debug info version number. We are going to drop debug info without a version number or with a different version number, to make sure we don't crash when we see bitcode files with different debug info metadata format. Make tests more robust by removing hard-coded metadata numbers in CHECK lines. llvm-svn: 195535
2013-11-23 09:16:29 +08:00
!llvm.module.flags = !{!1}
[DebugInfo] Fix tests so that each subprogram belongs to a CU. llvm-svn: 265490
2016-04-06 07:37:08 +08:00
!llvm.dbg.cu = !{!3}
Debug Info: update testing cases to specify the debug info version number. We are going to drop debug info without a version number or with a different version number, to make sure we don't crash when we see bitcode files with different debug info metadata format. Make tests more robust by removing hard-coded metadata numbers in CHECK lines. llvm-svn: 195535
2013-11-23 09:16:29 +08:00
IR: Make metadata typeless in assembly Now that `Metadata` is typeless, reflect that in the assembly. These are the matching assembly changes for the metadata/value split in r223802. - Only use the `metadata` type when referencing metadata from a call intrinsic -- i.e., only when it's used as a `Value`. - Stop pretending that `ValueAsMetadata` is wrapped in an `MDNode` when referencing it from call intrinsics. So, assembly like this: define @foo(i32 %v) { call void @llvm.foo(metadata !{i32 %v}, metadata !0) call void @llvm.foo(metadata !{i32 7}, metadata !0) call void @llvm.foo(metadata !1, metadata !0) call void @llvm.foo(metadata !3, metadata !0) call void @llvm.foo(metadata !{metadata !3}, metadata !0) ret void, !bar !2 } !0 = metadata !{metadata !2} !1 = metadata !{i32* @global} !2 = metadata !{metadata !3} !3 = metadata !{} turns into this: define @foo(i32 %v) { call void @llvm.foo(metadata i32 %v, metadata !0) call void @llvm.foo(metadata i32 7, metadata !0) call void @llvm.foo(metadata i32* @global, metadata !0) call void @llvm.foo(metadata !3, metadata !0) call void @llvm.foo(metadata !{!3}, metadata !0) ret void, !bar !2 } !0 = !{!2} !1 = !{i32* @global} !2 = !{!3} !3 = !{} I wrote an upgrade script that handled almost all of the tests in llvm and many of the tests in cfe (even handling many `CHECK` lines). I've attached it (or will attach it in a moment if you're speedy) to PR21532 to help everyone update their out-of-tree testcases. This is part of PR21532. llvm-svn: 224257
2014-12-16 03:07:53 +08:00
!0 = !{}
DebugInfo: Move new hierarchy into place Move the specialized metadata nodes for the new debug info hierarchy into place, finishing off PR22464. I've done bootstraps (and all that) and I'm confident this commit is NFC as far as DWARF output is concerned. Let me know if I'm wrong :). The code changes are fairly mechanical: - Bumped the "Debug Info Version". - `DIBuilder` now creates the appropriate subclass of `MDNode`. - Subclasses of DIDescriptor now expect to hold their "MD" counterparts (e.g., `DIBasicType` expects `MDBasicType`). - Deleted a ton of dead code in `AsmWriter.cpp` and `DebugInfo.cpp` for printing comments. - Big update to LangRef to describe the nodes in the new hierarchy. Feel free to make it better. Testcase changes are enormous. There's an accompanying clang commit on its way. If you have out-of-tree debug info testcases, I just broke your build. - `upgrade-specialized-nodes.sh` is attached to PR22564. I used it to update all the IR testcases. - Unfortunately I failed to find way to script the updates to CHECK lines, so I updated all of these by hand. This was fairly painful, since the old CHECKs are difficult to reason about. That's one of the benefits of the new hierarchy. This work isn't quite finished, BTW. The `DIDescriptor` subclasses are almost empty wrappers, but not quite: they still have loose casting checks (see the `RETURN_FROM_RAW()` macro). Once they're completely gutted, I'll rename the "MD" classes to "DI" and kill the wrappers. I also expect to make a few schema changes now that it's easier to reason about everything. llvm-svn: 231082
2015-03-04 01:24:31 +08:00
!1 = !{i32 1, !"Debug Info Version", i32 3}
[PR27284] Reverse the ownership between DICompileUnit and DISubprogram. Currently each Function points to a DISubprogram and DISubprogram has a scope field. For member functions the scope is a DICompositeType. DIScopes point to the DICompileUnit to facilitate type uniquing. Distinct DISubprograms (with isDefinition: true) are not part of the type hierarchy and cannot be uniqued. This change removes the subprograms list from DICompileUnit and instead adds a pointer to the owning compile unit to distinct DISubprograms. This would make it easy for ThinLTO to strip unneeded DISubprograms and their transitively referenced debug info. Motivation ---------- Materializing DISubprograms is currently the most expensive operation when doing a ThinLTO build of clang. We want the DISubprogram to be stored in a separate Bitcode block (or the same block as the function body) so we can avoid having to expensively deserialize all DISubprograms together with the global metadata. If a function has been inlined into another subprogram we need to store a reference the block containing the inlined subprogram. Attached to https://llvm.org/bugs/show_bug.cgi?id=27284 is a python script that updates LLVM IR testcases to the new format. http://reviews.llvm.org/D19034 <rdar://problem/25256815> llvm-svn: 266446
2016-04-15 23:57:41 +08:00
!2 = distinct !DISubprogram(unit: !3)
[DebugInfo] Fix tests so that each subprogram belongs to a CU. llvm-svn: 265490
2016-04-06 07:37:08 +08:00
!3 = distinct !DICompileUnit(language: DW_LANG_C99, producer: "clang",
file: !4,
isOptimized: true, flags: "-O2",
[PR27284] Reverse the ownership between DICompileUnit and DISubprogram. Currently each Function points to a DISubprogram and DISubprogram has a scope field. For member functions the scope is a DICompositeType. DIScopes point to the DICompileUnit to facilitate type uniquing. Distinct DISubprograms (with isDefinition: true) are not part of the type hierarchy and cannot be uniqued. This change removes the subprograms list from DICompileUnit and instead adds a pointer to the owning compile unit to distinct DISubprograms. This would make it easy for ThinLTO to strip unneeded DISubprograms and their transitively referenced debug info. Motivation ---------- Materializing DISubprograms is currently the most expensive operation when doing a ThinLTO build of clang. We want the DISubprogram to be stored in a separate Bitcode block (or the same block as the function body) so we can avoid having to expensively deserialize all DISubprograms together with the global metadata. If a function has been inlined into another subprogram we need to store a reference the block containing the inlined subprogram. Attached to https://llvm.org/bugs/show_bug.cgi?id=27284 is a python script that updates LLVM IR testcases to the new format. http://reviews.llvm.org/D19034 <rdar://problem/25256815> llvm-svn: 266446
2016-04-15 23:57:41 +08:00
splitDebugFilename: "abc.debug", emissionKind: 2)
[DebugInfo] Fix tests so that each subprogram belongs to a CU. llvm-svn: 265490
2016-04-06 07:37:08 +08:00
!4 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
!5 = !{i32 2, !"Debug Info Version", i32 3}
[objc-arc] Teach the ARC optimizer that objc_sync_enter/objc_sync_exit do not modify the ref count of an objc object and additionally are inert for modref purposes. llvm-svn: 185769
2013-07-07 09:52:55 +08:00
Replace hardcoded intrinsic list with speculatable attribute. No change in which intrinsics should be speculated. llvm-svn: 301995
2017-05-03 10:26:10 +08:00
; CHECK: attributes #0 = { nounwind readnone speculatable }
Use references to attribute groups on the call/invoke instructions. Listing all of the attributes for the callee of a call/invoke instruction is way too much and makes the IR unreadable. Use references to attributes instead. llvm-svn: 175877
2013-02-22 17:09:42 +08:00
; CHECK: attributes [[NUW]] = { nounwind }
IR: Make metadata typeless in assembly Now that `Metadata` is typeless, reflect that in the assembly. These are the matching assembly changes for the metadata/value split in r223802. - Only use the `metadata` type when referencing metadata from a call intrinsic -- i.e., only when it's used as a `Value`. - Stop pretending that `ValueAsMetadata` is wrapped in an `MDNode` when referencing it from call intrinsics. So, assembly like this: define @foo(i32 %v) { call void @llvm.foo(metadata !{i32 %v}, metadata !0) call void @llvm.foo(metadata !{i32 7}, metadata !0) call void @llvm.foo(metadata !1, metadata !0) call void @llvm.foo(metadata !3, metadata !0) call void @llvm.foo(metadata !{metadata !3}, metadata !0) ret void, !bar !2 } !0 = metadata !{metadata !2} !1 = metadata !{i32* @global} !2 = metadata !{metadata !3} !3 = metadata !{} turns into this: define @foo(i32 %v) { call void @llvm.foo(metadata i32 %v, metadata !0) call void @llvm.foo(metadata i32 7, metadata !0) call void @llvm.foo(metadata i32* @global, metadata !0) call void @llvm.foo(metadata !3, metadata !0) call void @llvm.foo(metadata !{!3}, metadata !0) ret void, !bar !2 } !0 = !{!2} !1 = !{i32* @global} !2 = !{!3} !3 = !{} I wrote an upgrade script that handled almost all of the tests in llvm and many of the tests in cfe (even handling many `CHECK` lines). I've attached it (or will attach it in a moment if you're speedy) to PR21532 to help everyone update their out-of-tree testcases. This is part of PR21532. llvm-svn: 224257
2014-12-16 03:07:53 +08:00
; CHECK: ![[RELEASE]] = !{}