[llvm] PowerPC32:PIC: Update to bcl to fix branch prediction mis-predict issue (PR #134140)
Lei Huang via llvm-commits
llvm-commits at lists.llvm.org
Fri Apr 4 13:45:55 PDT 2025
https://github.com/lei137 updated https://github.com/llvm/llvm-project/pull/134140
>From 6d0cb30449ecf6304ccd3b7eddf67366a826eecf Mon Sep 17 00:00:00 2001
From: Lei Huang <lei at ca.ibm.com>
Date: Wed, 2 Apr 2025 14:54:52 -0400
Subject: [PATCH 1/4] PowerPC32:PIC: Update to bcl to fix branch prediction
mis-predict issue
Update `bl` to `bcl 20, 31, .+4` for 32bit PIC code gen
so we don't corrupt the link stack and cause mis-predict for
the branch predictor.
fixes: https://github.com/llvm/llvm-project/issues/128644
---
llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp | 8 +++---
llvm/test/CodeGen/PowerPC/mcm-5.ll | 4 +--
llvm/test/CodeGen/PowerPC/ppc32-pic-bcl.ll | 29 ++++++++++++++++++++
llvm/test/CodeGen/PowerPC/ppc32-pic-large.ll | 2 +-
llvm/test/CodeGen/PowerPC/ppc32-pic.ll | 4 +--
5 files changed, 38 insertions(+), 9 deletions(-)
create mode 100644 llvm/test/CodeGen/PowerPC/ppc32-pic-bcl.ll
diff --git a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
index ba6653e10bdc5..0aca8ec7567af 100644
--- a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
+++ b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
@@ -955,8 +955,8 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) {
OutContext),
MCConstantExpr::create(4, OutContext), OutContext);
- // Emit the 'bl'.
- EmitToStreamer(*OutStreamer, MCInstBuilder(PPC::BL).addExpr(OffsExpr));
+ // Emit 'bcl 20,31,.+4' so the link stack is not corrupted.
+ EmitToStreamer(*OutStreamer, MCInstBuilder(PPC::BCLalways).addExpr(OffsExpr));
return;
}
case PPC::MovePCtoLR:
@@ -967,9 +967,9 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) {
// L1$pb:
MCSymbol *PICBase = MF->getPICBaseSymbol();
- // Emit the 'bl'.
+ // Emit 'bcl 20,31,.+4' so the link stack is not corrupted.
EmitToStreamer(*OutStreamer,
- MCInstBuilder(PPC::BL)
+ MCInstBuilder(PPC::BCLalways)
// FIXME: We would like an efficient form for this, so we
// don't have to do a lot of extra uniquing.
.addExpr(MCSymbolRefExpr::create(PICBase, OutContext)));
diff --git a/llvm/test/CodeGen/PowerPC/mcm-5.ll b/llvm/test/CodeGen/PowerPC/mcm-5.ll
index b88f405211b05..f9629b5710d72 100644
--- a/llvm/test/CodeGen/PowerPC/mcm-5.ll
+++ b/llvm/test/CodeGen/PowerPC/mcm-5.ll
@@ -51,7 +51,7 @@ sw.epilog: ; preds = %sw.bb3, %sw.default
ret i32 %5
}
; CHECK-LABEL: test_jump_table:
-; CHECK-NOT: bl .L0$pb
+; CHECK-NOT: bcl 20, 31, .L0$pb
; CHECK: addis [[REG1:[0-9]+]], 2, .LC[[TOCNUM:[0-9]+]]@toc at ha
; CHECK: ld [[REG2:[0-9]+]], .LC[[TOCNUM]]@toc at l([[REG1]])
@@ -64,7 +64,7 @@ sw.epilog: ; preds = %sw.bb3, %sw.default
; CHECK-NEXT: .long .LBB0_{{[0-9]+}}-.LJTI0_0
; LARGE-LABEL: test_jump_table:
-; LARGE: bl .L0$pb
+; LARGE: bcl 20, 31, .L0$pb
; LARGE-NEXT: .L0$pb:
; LARGE: mflr [[REGBASE:[0-9]+]]
diff --git a/llvm/test/CodeGen/PowerPC/ppc32-pic-bcl.ll b/llvm/test/CodeGen/PowerPC/ppc32-pic-bcl.ll
new file mode 100644
index 0000000000000..1e938b135fbfd
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/ppc32-pic-bcl.ll
@@ -0,0 +1,29 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=powerpc -relocation-model=pic | \
+; RUN: FileCheck -check-prefixes=SMALL %s
+
+ at val = global i8 0, align 1
+
+define zeroext i8 @testbcl() nounwind {
+; SMALL-LABEL: testbcl:
+; SMALL: # %bb.0: # %entry
+; SMALL-NEXT: mflr 0
+; SMALL-NEXT: stwu 1, -16(1)
+; SMALL-NEXT: stw 30, 8(1)
+; SMALL-NEXT: stw 0, 20(1)
+; SMALL-NEXT: bcl 20, 31, .L0$pb
+; SMALL-NEXT: .L0$pb:
+; SMALL-NEXT: mflr 30
+; SMALL-NEXT: lwz 3, .L0$poff-.L0$pb(30)
+; SMALL-NEXT: add 30, 3, 30
+; SMALL-NEXT: lwz 3, .LC0-.LTOC(30)
+; SMALL-NEXT: lbz 3, 0(3)
+; SMALL-NEXT: lwz 0, 20(1)
+; SMALL-NEXT: lwz 30, 8(1)
+; SMALL-NEXT: addi 1, 1, 16
+; SMALL-NEXT: mtlr 0
+; SMALL-NEXT: blr
+entry:
+ %0 = load i8, ptr @val, align 1
+ ret i8 %0
+}
diff --git a/llvm/test/CodeGen/PowerPC/ppc32-pic-large.ll b/llvm/test/CodeGen/PowerPC/ppc32-pic-large.ll
index 2f0b92964c13b..7be1a80b7af43 100644
--- a/llvm/test/CodeGen/PowerPC/ppc32-pic-large.ll
+++ b/llvm/test/CodeGen/PowerPC/ppc32-pic-large.ll
@@ -55,7 +55,7 @@ entry:
; LARGE-BSS-NEXT: foo:
; LARGE-BSS: stwu 1, -32(1)
; LARGE-BSS: stw 30, 24(1)
-; LARGE-BSS: bl [[PB]]
+; LARGE-BSS: bcl 20, 31, [[PB]]
; LARGE-BSS-NEXT: [[PB]]:
; LARGE-BSS: mflr 30
; LARGE-BSS: lwz [[REG:[0-9]+]], [[POFF]]-[[PB]](30)
diff --git a/llvm/test/CodeGen/PowerPC/ppc32-pic.ll b/llvm/test/CodeGen/PowerPC/ppc32-pic.ll
index aed994144940c..9ffd7395ffeb5 100644
--- a/llvm/test/CodeGen/PowerPC/ppc32-pic.ll
+++ b/llvm/test/CodeGen/PowerPC/ppc32-pic.ll
@@ -18,8 +18,8 @@ entry:
; SMALL-LABEL: foo:
; SMALL: stwu 1, -32(1)
; SMALL: stw 30, 24(1)
-; SMALL-BSS: bl _GLOBAL_OFFSET_TABLE_ at local-4
-; SMALL-SECURE: bl .L0$pb
+; SMALL-BSS: bcl 20, 31, _GLOBAL_OFFSET_TABLE_ at local-4
+; SMALL-SECURE: bcl 20, 31, .L0$pb
; SMALL: mflr 30
; SMALL-SECURE: addis 30, 30, _GLOBAL_OFFSET_TABLE_-.L0$pb at ha
; SMALL-SECURE: addi 30, 30, _GLOBAL_OFFSET_TABLE_-.L0$pb at l
>From 58906f473af33e87a2d297d443a138ad343c5532 Mon Sep 17 00:00:00 2001
From: Lei Huang <lei at ca.ibm.com>
Date: Wed, 2 Apr 2025 15:13:00 -0400
Subject: [PATCH 2/4] run clang-format
---
llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
index 0aca8ec7567af..7ff91238f8608 100644
--- a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
+++ b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
@@ -956,7 +956,8 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) {
MCConstantExpr::create(4, OutContext), OutContext);
// Emit 'bcl 20,31,.+4' so the link stack is not corrupted.
- EmitToStreamer(*OutStreamer, MCInstBuilder(PPC::BCLalways).addExpr(OffsExpr));
+ EmitToStreamer(*OutStreamer,
+ MCInstBuilder(PPC::BCLalways).addExpr(OffsExpr));
return;
}
case PPC::MovePCtoLR:
>From eb0e07d7ef1bbfb4b3902d962a39f79669fd9c14 Mon Sep 17 00:00:00 2001
From: Lei Huang <lei at ca.ibm.com>
Date: Fri, 4 Apr 2025 16:22:47 -0400
Subject: [PATCH 3/4] Remove update for PPC::MoveGOTtoLR pseudo
---
llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp | 5 ++---
llvm/test/CodeGen/PowerPC/ppc32-pic.ll | 4 ++--
2 files changed, 4 insertions(+), 5 deletions(-)
diff --git a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
index 7ff91238f8608..f07331bf6c6b5 100644
--- a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
+++ b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
@@ -955,9 +955,8 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) {
OutContext),
MCConstantExpr::create(4, OutContext), OutContext);
- // Emit 'bcl 20,31,.+4' so the link stack is not corrupted.
- EmitToStreamer(*OutStreamer,
- MCInstBuilder(PPC::BCLalways).addExpr(OffsExpr));
+ // Emit the 'bl'.
+ EmitToStreamer(*OutStreamer, MCInstBuilder(PPC::BL).addExpr(OffsExpr));
return;
}
case PPC::MovePCtoLR:
diff --git a/llvm/test/CodeGen/PowerPC/ppc32-pic.ll b/llvm/test/CodeGen/PowerPC/ppc32-pic.ll
index 9ffd7395ffeb5..aed994144940c 100644
--- a/llvm/test/CodeGen/PowerPC/ppc32-pic.ll
+++ b/llvm/test/CodeGen/PowerPC/ppc32-pic.ll
@@ -18,8 +18,8 @@ entry:
; SMALL-LABEL: foo:
; SMALL: stwu 1, -32(1)
; SMALL: stw 30, 24(1)
-; SMALL-BSS: bcl 20, 31, _GLOBAL_OFFSET_TABLE_ at local-4
-; SMALL-SECURE: bcl 20, 31, .L0$pb
+; SMALL-BSS: bl _GLOBAL_OFFSET_TABLE_ at local-4
+; SMALL-SECURE: bl .L0$pb
; SMALL: mflr 30
; SMALL-SECURE: addis 30, 30, _GLOBAL_OFFSET_TABLE_-.L0$pb at ha
; SMALL-SECURE: addi 30, 30, _GLOBAL_OFFSET_TABLE_-.L0$pb at l
>From aa188a35d3696a15fc0b4201e35dc49304a33af8 Mon Sep 17 00:00:00 2001
From: Lei Huang <lei at ca.ibm.com>
Date: Fri, 4 Apr 2025 16:45:41 -0400
Subject: [PATCH 4/4] restore ck for conversion of bl .+4
---
llvm/test/CodeGen/PowerPC/ppc32-pic.ll | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/test/CodeGen/PowerPC/ppc32-pic.ll b/llvm/test/CodeGen/PowerPC/ppc32-pic.ll
index aed994144940c..f7d8df9133306 100644
--- a/llvm/test/CodeGen/PowerPC/ppc32-pic.ll
+++ b/llvm/test/CodeGen/PowerPC/ppc32-pic.ll
@@ -19,7 +19,7 @@ entry:
; SMALL: stwu 1, -32(1)
; SMALL: stw 30, 24(1)
; SMALL-BSS: bl _GLOBAL_OFFSET_TABLE_ at local-4
-; SMALL-SECURE: bl .L0$pb
+; SMALL-SECURE: bcl 20, 31, .L0$pb
; SMALL: mflr 30
; SMALL-SECURE: addis 30, 30, _GLOBAL_OFFSET_TABLE_-.L0$pb at ha
; SMALL-SECURE: addi 30, 30, _GLOBAL_OFFSET_TABLE_-.L0$pb at l
More information about the llvm-commits
mailing list