Skip to content

Commit d7028da

Browse files
vchuravyvchuravy
authored
Merge pull request #42041 from JuliaLang/vc/llvm_13
Allow Julia to build on LLVM 13
2 parents 1709f79 + 9e0d8e9 commit d7028da

12 files changed

+923
-6
lines changed

‎deps/llvm.mk

Lines changed: 17 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -508,16 +508,31 @@ $(eval $(call LLVM_PATCH,llvm-D88630-clang-cmake))
508508
endif
509509
$(eval $(call LLVM_PATCH,llvm-invalid-addrspacecast-sink)) # Still being upstreamed as D92210
510510
$(eval $(call LLVM_PATCH,llvm-11-D93154-globalisel-as))
511-
$(eval $(call LLVM_PATCH,llvm-11-D94813-mergeicmps))
511+
$(eval $(call LLVM_PATCH,llvm-11-D94813-mergeicmps)) # remove for LLVM 14
512512
$(eval $(call LLVM_PROJ_PATCH,llvm-11-AArch64-FastIsel-bug))
513513
$(eval $(call LLVM_PATCH,llvm-12-D97435-AArch64-movaddrreg))
514514
$(eval $(call LLVM_PROJ_PATCH,llvm-11-D97571-AArch64-loh)) # remove for LLVM 13
515515
$(eval $(call LLVM_PROJ_PATCH,llvm-11-aarch64-addrspace)) # remove for LLVM 13
516516
$(eval $(call LLVM_PROJ_PATCH,llvm-12-fde-symbols-aarch64)) # remove for LLVM 13
517517
$(eval $(call LLVM_PROJ_PATCH,llvm-12-force-eh_frame-aarch64)) # remove for LLVM 13
518-
$(eval $(call LLVM_PROJ_PATCH,llvm-12-D109203-stackprobe-x86)) # remove for LLVM 13
518+
$(eval $(call LLVM_PROJ_PATCH,llvm-12-D109203-stackprobe-x86)) # remove for LLVM 14
519519
endif # LLVM_VER 12.0
520520

521+
ifeq ($(LLVM_VER_SHORT),13.0)
522+
$(eval $(call LLVM_PATCH,llvm7-revert-D44485)) # Needs upstreaming
523+
# $(eval $(call LLVM_PROJ_PATCH,llvm-13-D75072-SCEV-add-type)) # Should no longer be necessary
524+
$(eval $(call LLVM_PATCH,llvm-julia-tsan-custom-as))
525+
ifeq ($(BUILD_LLVM_CLANG),1)
526+
$(eval $(call LLVM_PATCH,llvm-D88630-clang-cmake))
527+
endif
528+
$(eval $(call LLVM_PATCH,llvm-invalid-addrspacecast-sink)) # Still being upstreamed as D92210
529+
$(eval $(call LLVM_PATCH,llvm-11-D93154-globalisel-as))
530+
$(eval $(call LLVM_PATCH,llvm-11-D94813-mergeicmps)) # remove for LLVM 14
531+
$(eval $(call LLVM_PATCH,llvm-13-AArch64-FastIsel-bug))
532+
$(eval $(call LLVM_PATCH,llvm-13-D97435-AArch64-movaddrreg))
533+
$(eval $(call LLVM_PROJ_PATCH,llvm-13-D109203-stackprobe-x86)) # remove for LLVM 14 (nominated for backport)
534+
endif # LLVM_VER 13.0
535+
521536
# Add a JL prefix to the version map. DO NOT REMOVE
522537
ifneq ($(LLVM_VER), svn)
523538
$(eval $(call LLVM_PATCH,llvm7-symver-jlprefix))

‎deps/patches/llvm-13-AArch64-FastIsel-bug.patch

Lines changed: 42 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,42 @@
1+
From d82434785ecab16223a0bd4ec8ede020bf244003 Mon Sep 17 00:00:00 2001
2+
From: Keno Fischer <[email protected]>
3+
Date: Mon, 1 Mar 2021 16:42:05 -0500
4+
Subject: [PATCH 4/4] AArch64: Remove Bad optimization
5+
6+
Removes the code responsible for causing https://bugs.llvm.org/show_bug.cgi?id=49357.
7+
A fix is in progress upstream, but I don't think it's easy, so this
8+
fixes the bug in the meantime. The optimization it does is minor.
9+
---
10+
llvm/lib/Target/AArch64/AArch64FastISel.cpp | 19 ---------------------
11+
1 file changed, 19 deletions(-)
12+
13+
diff --git llvm/lib/Target/AArch64/AArch64FastISel.cpp llvm/lib/Target/AArch64/AArch64FastISel.cpp
14+
index 9acda17b816f..5f0c7922569b 100644
15+
--- llvm/lib/Target/AArch64/AArch64FastISel.cpp
16+
+++ llvm/lib/Target/AArch64/AArch64FastISel.cpp
17+
@@ -4494,25 +4494,6 @@ bool AArch64FastISel::selectIntExt(const Instruction *I) {
18+
if (!SrcReg)
19+
return false;
20+
21+
- // Try to optimize already sign-/zero-extended values from function arguments.
22+
- bool IsZExt = isa<ZExtInst>(I);
23+
- if (const auto *Arg = dyn_cast<Argument>(I->getOperand(0))) {
24+
- if ((IsZExt && Arg->hasZExtAttr()) || (!IsZExt && Arg->hasSExtAttr())) {
25+
- if (RetVT == MVT::i64 && SrcVT != MVT::i64) {
26+
- unsigned ResultReg = createResultReg(&AArch64::GPR64RegClass);
27+
- BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
28+
- TII.get(AArch64::SUBREG_TO_REG), ResultReg)
29+
- .addImm(0)
30+
- .addReg(SrcReg)
31+
- .addImm(AArch64::sub_32);
32+
- SrcReg = ResultReg;
33+
- }
34+
-
35+
- updateValueMap(I, SrcReg);
36+
- return true;
37+
- }
38+
- }
39+
-
40+
unsigned ResultReg = emitIntExt(SrcVT, SrcReg, RetVT, IsZExt);
41+
if (!ResultReg)
42+
return false;

‎deps/patches/llvm-13-D109203-stackprobe-x86.patch

Lines changed: 157 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,157 @@
1+
From d914f133f8d6770cfd230e1b55aef67c16d19473 Mon Sep 17 00:00:00 2001
2+
From: Elliot Saba <[email protected]>
3+
Date: Tue, 7 Sep 2021 10:10:49 -0400
4+
Subject: [PATCH] [X86] Don't clobber EBX in stackprobes
5+
6+
On X86, the stackprobe emission code chooses the `R11D` register, which
7+
is illegal on i686. This ends up wrapping around to `EBX`, which does
8+
not get properly callee-saved within the stack probing prologue,
9+
clobbering the register for the callers.
10+
11+
We fix this by explicitly using `EAX` as the stack probe register.
12+
13+
Reviewed By: pengfei
14+
15+
Differential Revision: https://reviews.llvm.org/D109203
16+
---
17+
llvm/lib/Target/X86/X86FrameLowering.cpp | 8 +-
18+
llvm/test/CodeGen/X86/stack-clash-large.ll | 97 +++++++++++++---------
19+
2 files changed, 64 insertions(+), 41 deletions(-)
20+
21+
diff --git a/llvm/lib/Target/X86/X86FrameLowering.cpp b/llvm/lib/Target/X86/X86FrameLowering.cpp
22+
index 4cde7971e597..86cb86b19d62 100644
23+
--- a/llvm/lib/Target/X86/X86FrameLowering.cpp
24+
+++ b/llvm/lib/Target/X86/X86FrameLowering.cpp
25+
@@ -671,7 +671,9 @@ void X86FrameLowering::emitStackProbeInlineGenericLoop(
26+
MF.insert(MBBIter, testMBB);
27+
MF.insert(MBBIter, tailMBB);
28+
29+
- Register FinalStackProbed = Uses64BitFramePtr ? X86::R11 : X86::R11D;
30+
+ Register FinalStackProbed = Uses64BitFramePtr ? X86::R11
31+
+ : Is64Bit ? X86::R11D
32+
+ : X86::EAX;
33+
BuildMI(MBB, MBBI, DL, TII.get(TargetOpcode::COPY), FinalStackProbed)
34+
.addReg(StackPtr)
35+
.setMIFlag(MachineInstr::FrameSetup);
36+
@@ -1092,7 +1094,9 @@ void X86FrameLowering::BuildStackAlignAND(MachineBasicBlock &MBB,
37+
MF.insert(MBBIter, bodyMBB);
38+
MF.insert(MBBIter, footMBB);
39+
const unsigned MovMIOpc = Is64Bit ? X86::MOV64mi32 : X86::MOV32mi;
40+
- Register FinalStackProbed = Uses64BitFramePtr ? X86::R11 : X86::R11D;
41+
+ Register FinalStackProbed = Uses64BitFramePtr ? X86::R11
42+
+ : Is64Bit ? X86::R11D
43+
+ : X86::EAX;
44+
45+
// Setup entry block
46+
{
47+
diff --git a/llvm/test/CodeGen/X86/stack-clash-large.ll b/llvm/test/CodeGen/X86/stack-clash-large.ll
48+
index 9129e4ed40fd..00c7843b54f5 100644
49+
--- a/llvm/test/CodeGen/X86/stack-clash-large.ll
50+
+++ b/llvm/test/CodeGen/X86/stack-clash-large.ll
51+
@@ -1,45 +1,64 @@
52+
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --no_x86_scrub_sp
53+
-; RUN: llc -mtriple=x86_64-linux-android < %s | FileCheck -check-prefix=CHECK-X86-64 %s
54+
-; RUN: llc -mtriple=i686-linux-android < %s | FileCheck -check-prefix=CHECK-X86-32 %s
55+
+; RUN: llc -mtriple=x86_64-linux-android < %s | FileCheck -check-prefix=CHECK-X64 %s
56+
+; RUN: llc -mtriple=i686-linux-android < %s | FileCheck -check-prefix=CHECK-X86 %s
57+
+; RUN: llc -mtriple=x86_64-linux-gnux32 < %s | FileCheck -check-prefix=CHECK-X32 %s
58+
59+
define i32 @foo() local_unnamed_addr #0 {
60+
-; CHECK-X86-64-LABEL: foo:
61+
-; CHECK-X86-64: # %bb.0:
62+
-; CHECK-X86-64-NEXT: movq %rsp, %r11
63+
-; CHECK-X86-64-NEXT: subq $69632, %r11 # imm = 0x11000
64+
-; CHECK-X86-64-NEXT: .LBB0_1: # =>This Inner Loop Header: Depth=1
65+
-; CHECK-X86-64-NEXT: subq $4096, %rsp # imm = 0x1000
66+
-; CHECK-X86-64-NEXT: movq $0, (%rsp)
67+
-; CHECK-X86-64-NEXT: cmpq %r11, %rsp
68+
-; CHECK-X86-64-NEXT: jne .LBB0_1
69+
-; CHECK-X86-64-NEXT: # %bb.2:
70+
-; CHECK-X86-64-NEXT: subq $2248, %rsp # imm = 0x8C8
71+
-; CHECK-X86-64-NEXT: .cfi_def_cfa_offset 71888
72+
-; CHECK-X86-64-NEXT: movl $1, 264(%rsp)
73+
-; CHECK-X86-64-NEXT: movl $1, 28664(%rsp)
74+
-; CHECK-X86-64-NEXT: movl -128(%rsp), %eax
75+
-; CHECK-X86-64-NEXT: addq $71880, %rsp # imm = 0x118C8
76+
-; CHECK-X86-64-NEXT: .cfi_def_cfa_offset 8
77+
-; CHECK-X86-64-NEXT: retq
78+
+; CHECK-X64-LABEL: foo:
79+
+; CHECK-X64: # %bb.0:
80+
+; CHECK-X64-NEXT: movq %rsp, %r11
81+
+; CHECK-X64-NEXT: subq $69632, %r11 # imm = 0x11000
82+
+; CHECK-X64-NEXT: .LBB0_1: # =>This Inner Loop Header: Depth=1
83+
+; CHECK-X64-NEXT: subq $4096, %rsp # imm = 0x1000
84+
+; CHECK-X64-NEXT: movq $0, (%rsp)
85+
+; CHECK-X64-NEXT: cmpq %r11, %rsp
86+
+; CHECK-X64-NEXT: jne .LBB0_1
87+
+; CHECK-X64-NEXT: # %bb.2:
88+
+; CHECK-X64-NEXT: subq $2248, %rsp # imm = 0x8C8
89+
+; CHECK-X64-NEXT: .cfi_def_cfa_offset 71888
90+
+; CHECK-X64-NEXT: movl $1, 264(%rsp)
91+
+; CHECK-X64-NEXT: movl $1, 28664(%rsp)
92+
+; CHECK-X64-NEXT: movl -128(%rsp), %eax
93+
+; CHECK-X64-NEXT: addq $71880, %rsp # imm = 0x118C8
94+
+; CHECK-X64-NEXT: .cfi_def_cfa_offset 8
95+
+; CHECK-X64-NEXT: retq
96+
;
97+
-; CHECK-X86-32-LABEL: foo:
98+
-; CHECK-X86-32: # %bb.0:
99+
-; CHECK-X86-32-NEXT: movl %esp, %r11d
100+
-; CHECK-X86-32-NEXT: subl $69632, %r11d # imm = 0x11000
101+
-; CHECK-X86-32-NEXT: .LBB0_1: # =>This Inner Loop Header: Depth=1
102+
-; CHECK-X86-32-NEXT: subl $4096, %esp # imm = 0x1000
103+
-; CHECK-X86-32-NEXT: movl $0, (%esp)
104+
-; CHECK-X86-32-NEXT: cmpl %r11d, %esp
105+
-; CHECK-X86-32-NEXT: jne .LBB0_1
106+
-; CHECK-X86-32-NEXT: # %bb.2:
107+
-; CHECK-X86-32-NEXT: subl $2380, %esp # imm = 0x94C
108+
-; CHECK-X86-32-NEXT: .cfi_def_cfa_offset 72016
109+
-; CHECK-X86-32-NEXT: movl $1, 392(%esp)
110+
-; CHECK-X86-32-NEXT: movl $1, 28792(%esp)
111+
-; CHECK-X86-32-NEXT: movl (%esp), %eax
112+
-; CHECK-X86-32-NEXT: addl $72012, %esp # imm = 0x1194C
113+
-; CHECK-X86-32-NEXT: .cfi_def_cfa_offset 4
114+
-; CHECK-X86-32-NEXT: retl
115+
+; CHECK-X86-LABEL: foo:
116+
+; CHECK-X86: # %bb.0:
117+
+; CHECK-X86-NEXT: movl %esp, %eax
118+
+; CHECK-X86-NEXT: subl $69632, %eax # imm = 0x11000
119+
+; CHECK-X86-NEXT: .LBB0_1: # =>This Inner Loop Header: Depth=1
120+
+; CHECK-X86-NEXT: subl $4096, %esp # imm = 0x1000
121+
+; CHECK-X86-NEXT: movl $0, (%esp)
122+
+; CHECK-X86-NEXT: cmpl %eax, %esp
123+
+; CHECK-X86-NEXT: jne .LBB0_1
124+
+; CHECK-X86-NEXT: # %bb.2:
125+
+; CHECK-X86-NEXT: subl $2380, %esp # imm = 0x94C
126+
+; CHECK-X86-NEXT: .cfi_def_cfa_offset 72016
127+
+; CHECK-X86-NEXT: movl $1, 392(%esp)
128+
+; CHECK-X86-NEXT: movl $1, 28792(%esp)
129+
+; CHECK-X86-NEXT: movl (%esp), %eax
130+
+; CHECK-X86-NEXT: addl $72012, %esp # imm = 0x1194C
131+
+; CHECK-X86-NEXT: .cfi_def_cfa_offset 4
132+
+; CHECK-X86-NEXT: retl
133+
+;
134+
+; CHECK-X32-LABEL: foo:
135+
+; CHECK-X32: # %bb.0:
136+
+; CHECK-X32-NEXT: movl %esp, %r11d
137+
+; CHECK-X32-NEXT: subl $69632, %r11d # imm = 0x11000
138+
+; CHECK-X32-NEXT: .LBB0_1: # =>This Inner Loop Header: Depth=1
139+
+; CHECK-X32-NEXT: subl $4096, %esp # imm = 0x1000
140+
+; CHECK-X32-NEXT: movq $0, (%esp)
141+
+; CHECK-X32-NEXT: cmpl %r11d, %esp
142+
+; CHECK-X32-NEXT: jne .LBB0_1
143+
+; CHECK-X32-NEXT: # %bb.2:
144+
+; CHECK-X32-NEXT: subl $2248, %esp # imm = 0x8C8
145+
+; CHECK-X32-NEXT: .cfi_def_cfa_offset 71888
146+
+; CHECK-X32-NEXT: movl $1, 264(%esp)
147+
+; CHECK-X32-NEXT: movl $1, 28664(%esp)
148+
+; CHECK-X32-NEXT: movl -128(%esp), %eax
149+
+; CHECK-X32-NEXT: addl $71880, %esp # imm = 0x118C8
150+
+; CHECK-X32-NEXT: .cfi_def_cfa_offset 8
151+
+; CHECK-X32-NEXT: retq
152+
%a = alloca i32, i64 18000, align 16
153+
%b0 = getelementptr inbounds i32, i32* %a, i64 98
154+
%b1 = getelementptr inbounds i32, i32* %a, i64 7198
155+
--
156+
2.33.0
157+

0 commit comments

Comments
 (0)