[llvm] [CodeGenPrepare] Handle address sinking obtained from invoke (PR #143566)
Evgenii Kudriashov via llvm-commits
llvm-commits at lists.llvm.org
Fri Jun 13 10:19:28 PDT 2025
https://github.com/e-kud updated https://github.com/llvm/llvm-project/pull/143566
>From b7f4b17c69c0fd021aea986c3b802d3645c9a1ba Mon Sep 17 00:00:00 2001
From: Evgenii Kudriashov <evgenii.kudriashov at intel.com>
Date: Tue, 10 Jun 2025 09:00:35 -0700
Subject: [PATCH 1/5] [CodeGenPrepare] Handle address sinking obtained from
invoke
---
llvm/lib/CodeGen/CodeGenPrepare.cpp | 3 +-
.../CodeGenPrepare/X86/sink-addr-reuse.ll | 35 +++++++++++++++++++
2 files changed, 37 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp
index 32348a899683d..7529ead8af202 100644
--- a/llvm/lib/CodeGen/CodeGenPrepare.cpp
+++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp
@@ -5790,7 +5790,8 @@ static BasicBlock::iterator findInsertPos(Value *Addr, Instruction *MemoryInst,
// instruction after it.
if (SunkAddr) {
if (Instruction *AddrInst = dyn_cast<Instruction>(SunkAddr))
- return std::next(AddrInst->getIterator());
+ return AddrInst->isTerminator() ? MemoryInst->getIterator()
+ : std::next(AddrInst->getIterator());
}
// Find the first user of Addr in current BB.
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-reuse.ll b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-reuse.ll
index 019f311406550..5f084fc7ea2b5 100644
--- a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-reuse.ll
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-reuse.ll
@@ -42,3 +42,38 @@ bb3:
bb7:
ret void
}
+
+; Test a scenario when the address is the last instruction in the basic block.
+
+define void @adress_terminator() personality ptr null {
+; CHECK-LABEL: define void @adress_terminator() personality ptr null {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[PTR:%.*]] = invoke ptr null(i64 0)
+; CHECK-NEXT: to label %[[BODY_1:.*]] unwind label %[[EHCLEANUP:.*]]
+; CHECK: [[EHCLEANUP]]:
+; CHECK-NEXT: [[TMP0:%.*]] = cleanuppad within none []
+; CHECK-NEXT: cleanupret from [[TMP0]] unwind to caller
+; CHECK: [[BODY_1]]:
+; CHECK-NEXT: [[TMP1:%.*]] = bitcast ptr [[PTR]] to ptr
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[TMP1]], align 4
+; CHECK-NEXT: [[V0:%.*]] = load <4 x i32>, ptr [[PTR]], align 4
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[PTR]], align 4
+; CHECK-NEXT: ret void
+;
+entry:
+ %ptr = invoke ptr null(i64 0) to label %body.1 unwind label %ehcleanup
+
+body.2:
+ %v0 = load <4 x i32>, ptr %2, align 4
+ store <4 x i32> zeroinitializer, ptr %2, align 4
+ ret void
+
+ehcleanup:
+ %1 = cleanuppad within none []
+ cleanupret from %1 unwind to caller
+
+body.1:
+ %2 = getelementptr { i32 }, ptr %ptr, i64 0, i32 0
+ store <4 x i32> zeroinitializer, ptr %2, align 4
+ br label %body.2
+}
>From 8d928bc05274e347b92de88db5f9aeff858dd317 Mon Sep 17 00:00:00 2001
From: Evgenii Kudriashov <evgenii.kudriashov at intel.com>
Date: Wed, 11 Jun 2025 16:50:00 -0700
Subject: [PATCH 2/5] Don't try to sink addresses that we can't recreate
---
llvm/lib/CodeGen/CodeGenPrepare.cpp | 10 ++++-
.../CodeGenPrepare/X86/sink-addr-recreate.ll | 42 +++++++++++++++++++
.../CodeGenPrepare/X86/sink-addr-reuse.ll | 35 ----------------
3 files changed, 50 insertions(+), 37 deletions(-)
create mode 100644 llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp
index 7529ead8af202..a3b7e09156b85 100644
--- a/llvm/lib/CodeGen/CodeGenPrepare.cpp
+++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp
@@ -5790,8 +5790,7 @@ static BasicBlock::iterator findInsertPos(Value *Addr, Instruction *MemoryInst,
// instruction after it.
if (SunkAddr) {
if (Instruction *AddrInst = dyn_cast<Instruction>(SunkAddr))
- return AddrInst->isTerminator() ? MemoryInst->getIterator()
- : std::next(AddrInst->getIterator());
+ return std::next(AddrInst->getIterator());
}
// Find the first user of Addr in current BB.
@@ -6100,6 +6099,13 @@ bool CodeGenPrepare::optimizeMemoryInst(Instruction *MemoryInst, Value *Addr,
}
if (!ResultIndex) {
+ auto PtrInst = dyn_cast<Instruction>(ResultPtr);
+ // Here we know that we have just a pointer without any offsets. If
+ // this pointer comes from a different from the current basic block we
+ // need to know how to recreate it in another basic block.
+ // Currently we don't support recreation of any of instruction.
+ if (PtrInst && PtrInst->getParent() != MemoryInst->getParent())
+ return Modified;
SunkAddr = ResultPtr;
} else {
if (ResultPtr->getType() != I8PtrTy)
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
new file mode 100644
index 0000000000000..2cb92d413c5ff
--- /dev/null
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
@@ -0,0 +1,42 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -p 'require<profile-summary>,codegenprepare' -cgpp-huge-func=0 < %s | FileCheck %s
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-grtev4-linux-gnu"
+
+; Test a scenario when the address cannot be recreated in the current basic block
+
+declare ptr @get_ptr(i64)
+define void @address_terminator() personality ptr null {
+; CHECK-LABEL: define void @address_terminator() personality ptr null {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[PTR:%.*]] = invoke ptr @get_ptr(i64 0)
+; CHECK-NEXT: to label %[[BODY_1:.*]] unwind label %[[EHCLEANUP:.*]]
+; CHECK: [[EHCLEANUP]]:
+; CHECK-NEXT: [[PAD:%.*]] = cleanuppad within none []
+; CHECK-NEXT: cleanupret from [[PAD]] unwind to caller
+; CHECK: [[BODY_1]]:
+; CHECK-NEXT: [[GEP1:%.*]] = bitcast ptr [[PTR]] to ptr
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[GEP1]], align 4
+; CHECK-NEXT: [[TMP0:%.*]] = bitcast ptr [[PTR]] to ptr
+; CHECK-NEXT: [[UNUSED:%.*]] = load <4 x i32>, ptr [[TMP0]], align 4
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[TMP0]], align 4
+; CHECK-NEXT: ret void
+;
+entry:
+ %ptr = invoke ptr @get_ptr(i64 0) to label %body.1 unwind label %ehcleanup
+
+body.2:
+ %unused = load <4 x i32>, ptr %gep, align 4
+ store <4 x i32> zeroinitializer, ptr %gep, align 4
+ ret void
+
+ehcleanup:
+ %pad = cleanuppad within none []
+ cleanupret from %pad unwind to caller
+
+body.1:
+ %gep = getelementptr { i32 }, ptr %ptr, i64 0, i32 0
+ store <4 x i32> zeroinitializer, ptr %gep, align 4
+ br label %body.2
+}
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-reuse.ll b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-reuse.ll
index 5f084fc7ea2b5..019f311406550 100644
--- a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-reuse.ll
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-reuse.ll
@@ -42,38 +42,3 @@ bb3:
bb7:
ret void
}
-
-; Test a scenario when the address is the last instruction in the basic block.
-
-define void @adress_terminator() personality ptr null {
-; CHECK-LABEL: define void @adress_terminator() personality ptr null {
-; CHECK-NEXT: [[ENTRY:.*:]]
-; CHECK-NEXT: [[PTR:%.*]] = invoke ptr null(i64 0)
-; CHECK-NEXT: to label %[[BODY_1:.*]] unwind label %[[EHCLEANUP:.*]]
-; CHECK: [[EHCLEANUP]]:
-; CHECK-NEXT: [[TMP0:%.*]] = cleanuppad within none []
-; CHECK-NEXT: cleanupret from [[TMP0]] unwind to caller
-; CHECK: [[BODY_1]]:
-; CHECK-NEXT: [[TMP1:%.*]] = bitcast ptr [[PTR]] to ptr
-; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[TMP1]], align 4
-; CHECK-NEXT: [[V0:%.*]] = load <4 x i32>, ptr [[PTR]], align 4
-; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[PTR]], align 4
-; CHECK-NEXT: ret void
-;
-entry:
- %ptr = invoke ptr null(i64 0) to label %body.1 unwind label %ehcleanup
-
-body.2:
- %v0 = load <4 x i32>, ptr %2, align 4
- store <4 x i32> zeroinitializer, ptr %2, align 4
- ret void
-
-ehcleanup:
- %1 = cleanuppad within none []
- cleanupret from %1 unwind to caller
-
-body.1:
- %2 = getelementptr { i32 }, ptr %ptr, i64 0, i32 0
- store <4 x i32> zeroinitializer, ptr %2, align 4
- br label %body.2
-}
>From 0c804d3d8b06b4ff03c03f577397f3f32d70947b Mon Sep 17 00:00:00 2001
From: Evgenii Kudriashov <evgenii.kudriashov at intel.com>
Date: Wed, 11 Jun 2025 16:56:35 -0700
Subject: [PATCH 3/5] Rename the test
---
llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
index 2cb92d413c5ff..3cd5a9b161624 100644
--- a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
@@ -7,8 +7,8 @@ target triple = "x86_64-grtev4-linux-gnu"
; Test a scenario when the address cannot be recreated in the current basic block
declare ptr @get_ptr(i64)
-define void @address_terminator() personality ptr null {
-; CHECK-LABEL: define void @address_terminator() personality ptr null {
+define void @addr_from_invoke() personality ptr null {
+; CHECK-LABEL: define void @addr_from_invoke() personality ptr null {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[PTR:%.*]] = invoke ptr @get_ptr(i64 0)
; CHECK-NEXT: to label %[[BODY_1:.*]] unwind label %[[EHCLEANUP:.*]]
>From a5b5b9abcb61e02901355d0dd33dee569c948506 Mon Sep 17 00:00:00 2001
From: Evgenii Kudriashov <evgenii.kudriashov at intel.com>
Date: Fri, 13 Jun 2025 08:41:21 -0700
Subject: [PATCH 4/5] Add tests for args and globals
---
.../CodeGenPrepare/X86/sink-addr-recreate.ll | 68 ++++++++++++++++++-
1 file changed, 66 insertions(+), 2 deletions(-)
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
index 3cd5a9b161624..3c87396168e0a 100644
--- a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
@@ -4,9 +4,11 @@
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
target triple = "x86_64-grtev4-linux-gnu"
-; Test a scenario when the address cannot be recreated in the current basic block
-
+ at globalptr = external global ptr
declare ptr @get_ptr(i64)
+
+; Can't recreate invoke instruction
+
define void @addr_from_invoke() personality ptr null {
; CHECK-LABEL: define void @addr_from_invoke() personality ptr null {
; CHECK-NEXT: [[ENTRY:.*:]]
@@ -40,3 +42,65 @@ body.1:
store <4 x i32> zeroinitializer, ptr %gep, align 4
br label %body.2
}
+
+define void @addr_from_arg(ptr %ptr, i1 %p) {
+; CHECK-LABEL: define void @addr_from_arg(
+; CHECK-SAME: ptr [[PTR:%.*]], i1 [[P:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: br i1 [[P]], label %[[BODY_1:.*]], label %[[EXIT:.*]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: ret void
+; CHECK: [[BODY_1]]:
+; CHECK-NEXT: [[SUNKADDR:%.*]] = bitcast ptr [[PTR]] to ptr
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[SUNKADDR]], align 4
+; CHECK-NEXT: [[UNUSED:%.*]] = load <4 x i32>, ptr [[PTR]], align 4
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[PTR]], align 4
+; CHECK-NEXT: ret void
+;
+entry:
+ br i1 %p, label %body.1, label %exit
+
+body.2:
+ %unused = load <4 x i32>, ptr %gep, align 4
+ store <4 x i32> zeroinitializer, ptr %gep, align 4
+ ret void
+
+exit:
+ ret void
+
+body.1:
+ %gep = getelementptr { i32 }, ptr %ptr, i64 0, i32 0
+ store <4 x i32> zeroinitializer, ptr %gep, align 4
+ br label %body.2
+}
+
+define void @addr_from_global(i1 %p) {
+; CHECK-LABEL: define void @addr_from_global(
+; CHECK-SAME: i1 [[P:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: br i1 [[P]], label %[[BODY_1:.*]], label %[[EXIT:.*]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: ret void
+; CHECK: [[BODY_1]]:
+; CHECK-NEXT: [[GEP1:%.*]] = bitcast ptr @globalptr to ptr
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[GEP1]], align 4
+; CHECK-NEXT: [[UNUSED:%.*]] = load <4 x i32>, ptr @globalptr, align 4
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr @globalptr, align 4
+; CHECK-NEXT: ret void
+;
+entry:
+ br i1 %p, label %body.1, label %exit
+
+body.2:
+ %unused = load <4 x i32>, ptr %gep, align 4
+ store <4 x i32> zeroinitializer, ptr %gep, align 4
+ ret void
+
+exit:
+ ret void
+
+body.1:
+ %gep = getelementptr { i32 }, ptr @globalptr, i64 0, i32 0
+ store <4 x i32> zeroinitializer, ptr %gep, align 4
+ br label %body.2
+}
>From 81ddd177f0723f9316030788309a8ac6b1bdbbcb Mon Sep 17 00:00:00 2001
From: Evgenii Kudriashov <evgenii.kudriashov at intel.com>
Date: Fri, 13 Jun 2025 08:48:01 -0700
Subject: [PATCH 5/5] Skip anything that is not an instruction from current
basic block
---
llvm/lib/CodeGen/CodeGenPrepare.cpp | 10 +++++-----
.../CodeGenPrepare/X86/sink-addr-recreate.ll | 9 +++++----
2 files changed, 10 insertions(+), 9 deletions(-)
diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp
index a3b7e09156b85..1a3a5bff984dc 100644
--- a/llvm/lib/CodeGen/CodeGenPrepare.cpp
+++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp
@@ -6100,11 +6100,11 @@ bool CodeGenPrepare::optimizeMemoryInst(Instruction *MemoryInst, Value *Addr,
if (!ResultIndex) {
auto PtrInst = dyn_cast<Instruction>(ResultPtr);
- // Here we know that we have just a pointer without any offsets. If
- // this pointer comes from a different from the current basic block we
- // need to know how to recreate it in another basic block.
- // Currently we don't support recreation of any of instruction.
- if (PtrInst && PtrInst->getParent() != MemoryInst->getParent())
+ // We know that we have a pointer without any offsets. If this pointer
+ // originates from a different basic block than the current one, we
+ // must be able to recreate it in the current basic block.
+ // We do not support the recreation of any instructions yet.
+ if (!PtrInst || PtrInst->getParent() != MemoryInst->getParent())
return Modified;
SunkAddr = ResultPtr;
} else {
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
index 3c87396168e0a..af801a7a7a3f9 100644
--- a/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
@@ -53,8 +53,9 @@ define void @addr_from_arg(ptr %ptr, i1 %p) {
; CHECK: [[BODY_1]]:
; CHECK-NEXT: [[SUNKADDR:%.*]] = bitcast ptr [[PTR]] to ptr
; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[SUNKADDR]], align 4
-; CHECK-NEXT: [[UNUSED:%.*]] = load <4 x i32>, ptr [[PTR]], align 4
-; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[PTR]], align 4
+; CHECK-NEXT: [[TMP0:%.*]] = bitcast ptr [[PTR]] to ptr
+; CHECK-NEXT: [[UNUSED:%.*]] = load <4 x i32>, ptr [[TMP0]], align 4
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[TMP0]], align 4
; CHECK-NEXT: ret void
;
entry:
@@ -84,8 +85,8 @@ define void @addr_from_global(i1 %p) {
; CHECK: [[BODY_1]]:
; CHECK-NEXT: [[GEP1:%.*]] = bitcast ptr @globalptr to ptr
; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[GEP1]], align 4
-; CHECK-NEXT: [[UNUSED:%.*]] = load <4 x i32>, ptr @globalptr, align 4
-; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr @globalptr, align 4
+; CHECK-NEXT: [[UNUSED:%.*]] = load <4 x i32>, ptr [[GEP1]], align 4
+; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[GEP1]], align 4
; CHECK-NEXT: ret void
;
entry:
More information about the llvm-commits
mailing list