[PowerPC][GlobalMerge] Reduce TOC usage by merging internal and private global data #101224
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
amy-kwan
requested review from
MaskRay,
chenzheng1030,
mandlebug,
scui-ibm,
efriedma-quic and
stefanp-synopsys
|
@llvm/pr-subscribers-backend-powerpc Author: Amy Kwan (amy-kwan) ChangesThis patch aims to reduce TOC usage by merging internal and private global data. Moreover, we also add the GlobalMerge pass within the PPCTargetMachine pipeline, which is disabled by default. This transformation can be enabled by -ppc-global-merge. Full diff: https://github.com/llvm/llvm-project/pull/101224.diff 5 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/GlobalMerge.h b/llvm/include/llvm/CodeGen/GlobalMerge.h
index 13ad67d4544bc..ef767d548dc6e 100644
--- a/llvm/include/llvm/CodeGen/GlobalMerge.h
+++ b/llvm/include/llvm/CodeGen/GlobalMerge.h
@@ -28,6 +28,10 @@ struct GlobalMergeOptions {
bool MergeConst = false;
/// Whether we should merge global variables that have external linkage.
bool MergeExternal = true;
+ /// Whether we should merge global variables that have private linkage.
+ bool MergePrivateGlobals = false;
+ /// Whether we should merge constant global variables.
+ bool MergeConstantGlobals = false;
/// Whether we should try to optimize for size only.
/// Currently, this applies a dead simple heuristic: only consider globals
/// used in minsize functions for merging.
diff --git a/llvm/include/llvm/CodeGen/Passes.h b/llvm/include/llvm/CodeGen/Passes.h
index cafb9781698a2..b401a8d9f10fd 100644
--- a/llvm/include/llvm/CodeGen/Passes.h
+++ b/llvm/include/llvm/CodeGen/Passes.h
@@ -476,7 +476,9 @@ namespace llvm {
///
Pass *createGlobalMergePass(const TargetMachine *TM, unsigned MaximalOffset,
bool OnlyOptimizeForSize = false,
- bool MergeExternalByDefault = false);
+ bool MergeExternalByDefault = false,
+ bool MergePrivateByDefault = false,
+ bool MergeConstantByDefault = false);
/// This pass splits the stack into a safe stack and an unsafe stack to
/// protect against stack-based overflow vulnerabilities.
diff --git a/llvm/lib/CodeGen/GlobalMerge.cpp b/llvm/lib/CodeGen/GlobalMerge.cpp
index e420c2bb7a25e..4056860fddd93 100644
--- a/llvm/lib/CodeGen/GlobalMerge.cpp
+++ b/llvm/lib/CodeGen/GlobalMerge.cpp
@@ -196,11 +196,14 @@ class GlobalMerge : public FunctionPass {
}
explicit GlobalMerge(const TargetMachine *TM, unsigned MaximalOffset,
- bool OnlyOptimizeForSize, bool MergeExternalGlobals)
+ bool OnlyOptimizeForSize, bool MergeExternalGlobals,
+ bool MergePrivateGlobals, bool MergeConstantGlobals)
: FunctionPass(ID), TM(TM) {
Opt.MaxOffset = MaximalOffset;
Opt.SizeOnly = OnlyOptimizeForSize;
Opt.MergeExternal = MergeExternalGlobals;
+ Opt.MergePrivateGlobals = MergePrivateGlobals;
+ Opt.MergeConstantGlobals = MergeConstantGlobals;
initializeGlobalMergePass(*PassRegistry::getPassRegistry());
}
@@ -475,7 +478,8 @@ bool GlobalMergeImpl::doMerge(const SmallVectorImpl<GlobalVariable *> &Globals,
auto &DL = M.getDataLayout();
LLVM_DEBUG(dbgs() << " Trying to merge set, starts with #"
- << GlobalSet.find_first() << "\n");
+ << GlobalSet.find_first() << ", total of " << Globals.size()
+ << "\n");
bool Changed = false;
ssize_t i = GlobalSet.find_first();
@@ -551,6 +555,8 @@ bool GlobalMergeImpl::doMerge(const SmallVectorImpl<GlobalVariable *> &Globals,
MergedGV->setAlignment(MaxAlign);
MergedGV->setSection(Globals[i]->getSection());
+ LLVM_DEBUG(dbgs() << "MergedGV: " << *MergedGV << "\n");
+
const StructLayout *MergedLayout = DL.getStructLayout(MergedTy);
for (ssize_t k = i, idx = 0; k != j; k = GlobalSet.find_next(k), ++idx) {
GlobalValue::LinkageTypes Linkage = Globals[k]->getLinkage();
@@ -700,6 +706,11 @@ bool GlobalMergeImpl::run(Module &M) {
else
Globals[{AddressSpace, Section}].push_back(&GV);
}
+ LLVM_DEBUG(dbgs() << "GV "
+ << ((DL.getTypeAllocSize(Ty) < Opt.MaxOffset)
+ ? "to merge: "
+ : "not to merge: ")
+ << GV << "\n");
}
for (auto &P : Globals)
@@ -710,7 +721,7 @@ bool GlobalMergeImpl::run(Module &M) {
if (P.second.size() > 1)
Changed |= doMerge(P.second, M, false, P.first.first);
- if (EnableGlobalMergeOnConst)
+ if (Opt.MergeConstantGlobals)
for (auto &P : ConstGlobals)
if (P.second.size() > 1)
Changed |= doMerge(P.second, M, true, P.first.first);
@@ -720,8 +731,14 @@ bool GlobalMergeImpl::run(Module &M) {
Pass *llvm::createGlobalMergePass(const TargetMachine *TM, unsigned Offset,
bool OnlyOptimizeForSize,
- bool MergeExternalByDefault) {
+ bool MergeExternalByDefault,
+ bool MergePrivateByDefault,
+ bool MergeConstantByDefault) {
bool MergeExternal = (EnableGlobalMergeOnExternal == cl::BOU_UNSET) ?
MergeExternalByDefault : (EnableGlobalMergeOnExternal == cl::BOU_TRUE);
- return new GlobalMerge(TM, Offset, OnlyOptimizeForSize, MergeExternal);
+ bool MergeConstant = EnableGlobalMergeOnConst.getNumOccurrences() > 0
+ ? EnableGlobalMergeOnConst
+ : MergeConstantByDefault;
+ return new GlobalMerge(TM, Offset, OnlyOptimizeForSize, MergeExternal,
+ MergePrivateByDefault, MergeConstant);
}
diff --git a/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp b/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
index 1ef891d1b677a..e4045ec304435 100644
--- a/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
+++ b/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
@@ -111,6 +111,15 @@ static cl::opt<bool> EnablePPCGenScalarMASSEntries(
"(scalar) entries"),
cl::Hidden);
+static cl::opt<bool>
+ EnableGlobalMerge("ppc-global-merge", cl::Hidden, cl::init(false),
+ cl::desc("Enable the global merge pass"));
+
+static cl::opt<unsigned>
+ GlobalMergeMaxOffset("ppc-global-merge-max-offset", cl::Hidden,
+ cl::init(0x7fff),
+ cl::desc("Maximum global merge offset"));
+
extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializePowerPCTarget() {
// Register the targets
RegisterTargetMachine<PPCTargetMachine> A(getThePPC32Target());
@@ -491,6 +500,10 @@ void PPCPassConfig::addIRPasses() {
}
bool PPCPassConfig::addPreISel() {
+ if (EnableGlobalMerge)
+ addPass(createGlobalMergePass(TM, GlobalMergeMaxOffset, false, false, true,
+ true));
+
if (MergeStringPool && getOptLevel() != CodeGenOptLevel::None)
addPass(createPPCMergeStringPoolPass());
diff --git a/llvm/test/CodeGen/PowerPC/merge-private.ll b/llvm/test/CodeGen/PowerPC/merge-private.ll
index 6cf276990d7ea..6ed2d6dfc542b 100644
--- a/llvm/test/CodeGen/PowerPC/merge-private.ll
+++ b/llvm/test/CodeGen/PowerPC/merge-private.ll
@@ -1,15 +1,15 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -verify-machineinstrs -mtriple powerpc64-ibm-aix-xcoff -mcpu=pwr8 \
-; RUN: -ppc-asm-full-reg-names < %s | FileCheck %s \
+; RUN: -ppc-asm-full-reg-names -ppc-global-merge=true < %s | FileCheck %s \
; RUN: --check-prefix=AIX64
; RUN: llc -verify-machineinstrs -mtriple powerpc-ibm-aix-xcoff -mcpu=pwr8 \
-; RUN: -ppc-asm-full-reg-names < %s | FileCheck %s \
+; RUN: -ppc-asm-full-reg-names -ppc-global-merge=true < %s | FileCheck %s \
; RUN: --check-prefix=AIX32
; RUN: llc -verify-machineinstrs -mtriple powerpc64le-unknown-linux -mcpu=pwr8 \
-; RUN: -ppc-asm-full-reg-names < %s | FileCheck %s \
+; RUN: -ppc-asm-full-reg-names -ppc-global-merge=true < %s | FileCheck %s \
; RUN: --check-prefix=LINUX64LE
; RUN: llc -verify-machineinstrs -mtriple powerpc64-unknown-linux -mcpu=pwr8 \
-; RUN: -ppc-asm-full-reg-names < %s | FileCheck %s \
+; RUN: -ppc-asm-full-reg-names -ppc-global-merge=true < %s | FileCheck %s \
; RUN: --check-prefix=LINUX64BE
@.str = private unnamed_addr constant [15 x i8] c"Private global\00", align 1
@@ -24,7 +24,7 @@ define dso_local void @print_func() {
; AIX64-NEXT: stdu r1, -128(r1)
; AIX64-NEXT: std r0, 144(r1)
; AIX64-NEXT: std r31, 120(r1) # 8-byte Folded Spill
-; AIX64-NEXT: ld r31, L..C0(r2) # @__ModuleStringPool
+; AIX64-NEXT: ld r31, L..C0(r2) # @_MergedGlobals
; AIX64-NEXT: mr r3, r31
; AIX64-NEXT: bl .puts[PR]
; AIX64-NEXT: nop
@@ -43,7 +43,7 @@ define dso_local void @print_func() {
; AIX32-NEXT: stwu r1, -64(r1)
; AIX32-NEXT: stw r0, 72(r1)
; AIX32-NEXT: stw r31, 60(r1) # 4-byte Folded Spill
-; AIX32-NEXT: lwz r31, L..C0(r2) # @__ModuleStringPool
+; AIX32-NEXT: lwz r31, L..C0(r2) # @_MergedGlobals
; AIX32-NEXT: mr r3, r31
; AIX32-NEXT: bl .puts[PR]
; AIX32-NEXT: nop
@@ -64,9 +64,9 @@ define dso_local void @print_func() {
; LINUX64LE-NEXT: .cfi_offset r30, -16
; LINUX64LE-NEXT: std r30, -16(r1) # 8-byte Folded Spill
; LINUX64LE-NEXT: stdu r1, -48(r1)
-; LINUX64LE-NEXT: addis r3, r2, .L__ModuleStringPool@toc@ha
+; LINUX64LE-NEXT: addis r3, r2, .L_MergedGlobals@toc@ha
; LINUX64LE-NEXT: std r0, 64(r1)
-; LINUX64LE-NEXT: addi r30, r3, .L__ModuleStringPool@toc@l
+; LINUX64LE-NEXT: addi r30, r3, .L_MergedGlobals@toc@l
; LINUX64LE-NEXT: mr r3, r30
; LINUX64LE-NEXT: bl puts
; LINUX64LE-NEXT: nop
@@ -87,9 +87,9 @@ define dso_local void @print_func() {
; LINUX64BE-NEXT: .cfi_def_cfa_offset 128
; LINUX64BE-NEXT: .cfi_offset lr, 16
; LINUX64BE-NEXT: .cfi_offset r30, -16
-; LINUX64BE-NEXT: addis r3, r2, .L__ModuleStringPool@toc@ha
+; LINUX64BE-NEXT: addis r3, r2, .L_MergedGlobals@toc@ha
; LINUX64BE-NEXT: std r30, 112(r1) # 8-byte Folded Spill
-; LINUX64BE-NEXT: addi r30, r3, .L__ModuleStringPool@toc@l
+; LINUX64BE-NEXT: addi r30, r3, .L_MergedGlobals@toc@l
; LINUX64BE-NEXT: mr r3, r30
; LINUX64BE-NEXT: bl puts
; LINUX64BE-NEXT: nop
|
| @@ -28,6 +28,10 @@ struct GlobalMergeOptions { | |||
| bool MergeConst = false; | |||
| /// Whether we should merge global variables that have external linkage. | |||
| bool MergeExternal = true; | |||
| /// Whether we should merge global variables that have private linkage. | |||
| bool MergePrivateGlobals = false; | |||
There was a problem hiding this comment.
MergePrivateGlobals is unused?
There was a problem hiding this comment.
+1
Like the internal global variables, private global variables should always be merged and should not be guarded under an option. You've already done this in #101222 :)
redstar
reviewed
Jul 30, 2024
llvm/lib/CodeGen/GlobalMerge.cpp
Outdated
Comment on lines
739
to
741
| bool MergeConstant = EnableGlobalMergeOnConst.getNumOccurrences() > 0 | ||
| ? EnableGlobalMergeOnConst | ||
| : MergeConstantByDefault; |
There was a problem hiding this comment.
Suggested change
| bool MergeConstant = EnableGlobalMergeOnConst.getNumOccurrences() > 0 | |
| ? EnableGlobalMergeOnConst | |
| : MergeConstantByDefault; | |
| bool MergeConstant = EnableGlobalMergeOnConst || MergeConstantByDefault; |
amy-kwan
force-pushed
the
users/amy-kwan/ppc-global-merge-private-internal
branch
from
61a41df to
d14bd7c
Compare
amy-kwan
force-pushed
the
users/amy-kwan/ppc-global-merge-private-internal
branch
from
d14bd7c to
10d3137
Compare
chenzheng1030
approved these changes
Aug 5, 2024
Base automatically changed from
users/amy-kwan/update-globalmerge-private-condition
to
main
August 13, 2024 14:13
amy-kwan
force-pushed
the
users/amy-kwan/ppc-global-merge-private-internal
branch
from
10d3137 to
10af3fb
Compare
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
…te global data This patch aims to reduce TOC usage by merging internal and private global data. Moreover, we also add the GlobalMerge pass within the PPCTargetMachine pipeline, which is disabled by default. This transformation can be enabled by -ppc-global-merge.
amy-kwan
force-pushed
the
users/amy-kwan/ppc-global-merge-private-internal
branch
from
10af3fb to
40e0221
Compare
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/60/builds/4942 Here is the relevant piece of the build log for the reference: |
qiaojbao
pushed a commit
to GPUOpen-Drivers/llvm-project
that referenced
this pull request
Aug 29, 2024
…b32e22b9d Local branch amd-gfx e0cb32e Merged main:98119718603c6e623b34e3d2160b65b26d9bfcdc into amd-gfx:a22f3eb12095 Remote branch main 5e990b0 [PowerPC][GlobalMerge] Reduce TOC usage by merging internal and private global data (llvm#101224)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This patch aims to reduce TOC usage by merging internal and private global data.
Moreover, we also add the GlobalMerge pass within the PPCTargetMachine pipeline, which is disabled by default. This transformation can be enabled by -ppc-global-merge.