[llvm-branch-commits] [lld] [lld][LoongArch] Relax TLSDESC code sequence. (PR #123677)
Zhaoxin Yang via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Mon Jan 20 16:39:36 PST 2025
https://github.com/ylzsx created https://github.com/llvm/llvm-project/pull/123677
Relax TLSDESC code sequence.
Original code sequence:
* pcalau12i $a0, %desc_pc_hi20(sym_desc)
* addi.d $a0, $a0, %desc_pc_lo12(sym_desc)
* ld.d $ra, $a0, %desc_ld(sym_desc)
* jirl $ra, $ra, %desc_call(sym_desc)
Cannot convert to LE/IE, but relax:
* pcaddi $a0, %desc_pcrel_20(sym_desc)
* ld.d $ra, $a0, %desc_ld(sym_desc)
* jirl $ra, $ra, %desc_call(sym_desc)
FIXME: The conversion of TLSDESC GD/LD to LE/IE will implement in a future patch.
>From 56c24f9746ef42c449a4d1d5caf10f7cd1dd7d81 Mon Sep 17 00:00:00 2001
From: yangzhaoxin <yangzhaoxin at loongson.cn>
Date: Tue, 31 Dec 2024 15:51:43 +0800
Subject: [PATCH 1/6] Relax TLSDESC code sequence.
Original code sequence:
* pcalau12i $a0, %desc_pc_hi20(sym_desc)
* addi.d $a0, $a0, %desc_pc_lo12(sym_desc)
* ld.d $ra, $a0, %desc_ld(sym_desc)
* jirl $ra, $ra, %desc_call(sym_desc)
Cannot convert to LE/IE, but relax:
* pcaddi $a0, %desc_pcrel_20(sym_desc)
* ld.d $ra, $a0, %desc_ld(sym_desc)
* jirl $ra, $ra, %desc_call(sym_desc)
FIXME: The conversion of TLSDESC GD/LD to LE/IE will implement in a
future patch.
---
lld/ELF/Arch/LoongArch.cpp | 44 +++++++++++++++++++++++++++++++++++++-
1 file changed, 43 insertions(+), 1 deletion(-)
diff --git a/lld/ELF/Arch/LoongArch.cpp b/lld/ELF/Arch/LoongArch.cpp
index 2d6d86d2ca63b2..a6db15bbf1efd4 100644
--- a/lld/ELF/Arch/LoongArch.cpp
+++ b/lld/ELF/Arch/LoongArch.cpp
@@ -782,7 +782,9 @@ static void relaxPCHi20Lo12(Ctx &ctx, const InputSection &sec, size_t i,
(rHi20.type == R_LARCH_TLS_GD_PC_HI20 &&
rLo12.type == R_LARCH_GOT_PC_LO12) ||
(rHi20.type == R_LARCH_TLS_LD_PC_HI20 &&
- rLo12.type == R_LARCH_GOT_PC_LO12)))
+ rLo12.type == R_LARCH_GOT_PC_LO12) ||
+ (rHi20.type == R_LARCH_TLS_DESC_PC_HI20 &&
+ rLo12.type == R_LARCH_TLS_DESC_PC_LO12)))
return;
// GOT references to absolute symbols can't be relaxed to use pcaddi in
@@ -804,6 +806,8 @@ static void relaxPCHi20Lo12(Ctx &ctx, const InputSection &sec, size_t i,
symBase = rHi20.sym->getVA(ctx);
else if (rHi20.expr == RE_LOONGARCH_TLSGD_PAGE_PC)
symBase = ctx.in.got->getGlobalDynAddr(*rHi20.sym);
+ else if (rHi20.expr == RE_LOONGARCH_TLSDESC_PAGE_PC)
+ symBase = ctx.in.got->getTlsDescAddr(*rHi20.sym);
else {
Err(ctx) << getErrorLoc(ctx, (const uint8_t *)loc) << "unknown expr ("
<< rHi20.expr << ") against symbol " << rHi20.sym
@@ -837,6 +841,8 @@ static void relaxPCHi20Lo12(Ctx &ctx, const InputSection &sec, size_t i,
sec.relaxAux->relocTypes[i + 2] = R_LARCH_TLS_GD_PCREL20_S2;
else if (rHi20.type == R_LARCH_TLS_LD_PC_HI20)
sec.relaxAux->relocTypes[i + 2] = R_LARCH_TLS_LD_PCREL20_S2;
+ else if (rHi20.type == R_LARCH_TLS_DESC_PC_HI20)
+ sec.relaxAux->relocTypes[i + 2] = R_LARCH_TLS_DESC_PCREL20_S2;
else
sec.relaxAux->relocTypes[i + 2] = R_LARCH_PCREL20_S2;
sec.relaxAux->writes.push_back(insn(PCADDI, getD5(nextInsn), 0, 0));
@@ -903,6 +909,33 @@ static void relaxTlsLe(Ctx &ctx, const InputSection &sec, size_t i,
}
}
+// Relax TLSDESC code sequence. In LoongArch, the conversion of TLSDESC GD/LD to
+// LE/IE is closely tied to relaxation, similar to how GCC handles it. (Due to
+// the lack of an efficient way for handling conversions in the extreme code
+// model and the difficulty in determining whether the extreme code model is
+// being used in handleTlsRelocation, this approach may seem like a workaround).
+// Consequently, the resulting code sequence depends on whether the conversion
+// to LE/IE is performed.
+//
+// Original code sequence:
+// * pcalau12i $a0, %desc_pc_hi20(sym_desc)
+// * addi.d $a0, $a0, %desc_pc_lo12(sym_desc)
+// * ld.d $ra, $a0, %desc_ld(sym_desc)
+// * jirl $ra, $ra, %desc_call(sym_desc)
+//
+// Cannot convert to LE/IE, but relax:
+// * pcaddi $a0, %desc_pcrel_20(sym_desc)
+// * ld.d $ra, $a0, %desc_ld(sym_desc)
+// * jirl $ra, $ra, %desc_call(sym_desc)
+//
+// FIXME: Implement TLSDESC GD/LD to LE/IE.
+static void relaxTlsdesc(Ctx &ctx, const InputSection &sec, size_t i,
+ uint64_t loc, Relocation &rHi20, Relocation &rLo12,
+ uint32_t &remove) {
+ if (ctx.arg.shared && rHi20.type == R_LARCH_TLS_DESC_PC_HI20)
+ return relaxPCHi20Lo12(ctx, sec, i, loc, rHi20, rLo12, remove);
+}
+
static bool relax(Ctx &ctx, InputSection &sec) {
const uint64_t secAddr = sec.getVA();
const MutableArrayRef<Relocation> relocs = sec.relocs();
@@ -959,6 +992,10 @@ static bool relax(Ctx &ctx, InputSection &sec) {
if (relaxable(relocs, i))
relaxTlsLe(ctx, sec, i, loc, r, remove);
break;
+ case R_LARCH_TLS_DESC_PC_HI20:
+ if (isPairRelaxable(relocs, i))
+ relaxTlsdesc(ctx, sec, i, loc, r, relocs[i + 2], remove);
+ break;
}
// For all anchors whose offsets are <= r.offset, they are preceded by
@@ -1078,6 +1115,11 @@ void LoongArch::finalizeRelax(int passes) const {
write32le(p, aux.writes[writesIdx++]);
r.expr = R_TLSGD_PC;
break;
+ case R_LARCH_TLS_DESC_PCREL20_S2:
+ skip = 4;
+ write32le(p, aux.writes[writesIdx++]);
+ r.expr = R_TLSDESC_PC;
+ break;
default:
llvm_unreachable("unsupported type");
}
>From 880523cdb13c1a1216cbdd08faf3180e79587110 Mon Sep 17 00:00:00 2001
From: yangzhaoxin <yangzhaoxin at loongson.cn>
Date: Tue, 31 Dec 2024 22:30:40 +0800
Subject: [PATCH 2/6] Add test loongarch-relax-tlsdesc.s
---
lld/test/ELF/loongarch-relax-tlsdesc.s | 189 +++++++++++++++++++++++++
1 file changed, 189 insertions(+)
create mode 100644 lld/test/ELF/loongarch-relax-tlsdesc.s
diff --git a/lld/test/ELF/loongarch-relax-tlsdesc.s b/lld/test/ELF/loongarch-relax-tlsdesc.s
new file mode 100644
index 00000000000000..6470bdf35cb223
--- /dev/null
+++ b/lld/test/ELF/loongarch-relax-tlsdesc.s
@@ -0,0 +1,189 @@
+# REQUIRES: loongarch
+# RUN: rm -rf %t && split-file %s %t && cd %t
+# RUN: llvm-mc -filetype=obj -triple=loongarch64 -mattr=+relax a.s -o a.64.o
+# RUN: llvm-mc -filetype=obj -triple=loongarch64 -mattr=+relax c.s -o c.64.o
+# RUN: ld.lld -shared -soname=c.64.so c.64.o -o c.64.so
+# RUN: llvm-mc -filetype=obj -triple=loongarch32 -mattr=+relax --defsym ELF32=1 a.s -o a.32.o
+# RUN: llvm-mc -filetype=obj -triple=loongarch32 -mattr=+relax --defsym ELF32=1 c.s -o c.32.o
+# RUN: ld.lld -shared -soname=c.32.so c.32.o -o c.32.so
+
+## Test the TLSDESC relaxation. Also check --emit-relocs.
+# RUN: ld.lld -shared --emit-relocs -z now a.64.o c.64.o -o a.64.so
+# RUN: llvm-readobj -r -x .got a.64.so | FileCheck --check-prefix=GD64-RELA %s
+# RUN: llvm-objdump --no-show-raw-insn -dr a.64.so | FileCheck %s --check-prefix=GD64
+
+## FIXME: The transition frome TLSDESC to IE/LE has not yet been implemented.
+## Keep the dynamic relocations and hand them over to dynamic linker.
+
+# RUN: ld.lld -e 0 -z now --emit-relocs a.64.o c.64.o -o a.64.le
+# RUN: llvm-readobj -r -x .got a.64.le | FileCheck --check-prefix=LE64-RELA %s
+# RUN: llvm-objdump --no-show-raw-insn -d a.64.le | FileCheck %s --check-prefix=LE64
+
+# RUN: ld.lld -e 0 -z now --emit-relocs a.64.o c.64.so -o a.64.ie
+# RUN: llvm-readobj -r -x .got a.64.ie | FileCheck --check-prefix=IE64-RELA %s
+# RUN: llvm-objdump --no-show-raw-insn -d a.64.ie | FileCheck %s --check-prefix=IE64
+
+## 32-bit code is mostly the same. We only test a few variants.
+
+# RUN: ld.lld -shared -z now a.32.o c.32.o -o rel.32.so -z rel
+# RUN: llvm-readobj -r -x .got rel.32.so | FileCheck --check-prefix=GD32-REL %s
+
+# GD64-RELA: .rela.dyn {
+# GD64-RELA-NEXT: 0x20448 R_LARCH_TLS_DESC64 - 0x7FF
+# GD64-RELA-NEXT: 0x20418 R_LARCH_TLS_DESC64 a 0x0
+# GD64-RELA-NEXT: 0x20428 R_LARCH_TLS_DESC64 c 0x0
+# GD64-RELA-NEXT: 0x20438 R_LARCH_TLS_DESC64 d 0x0
+# GD64-RELA-NEXT: }
+# GD64-RELA: Hex dump of section '.got':
+# GD64-RELA-NEXT: 0x00020418 00000000 00000000 00000000 00000000 .
+# GD64-RELA-NEXT: 0x00020428 00000000 00000000 00000000 00000000 .
+# GD64-RELA-NEXT: 0x00020438 00000000 00000000 00000000 00000000 .
+# GD64-RELA-NEXT: 0x00020448 00000000 00000000 00000000 00000000 .
+
+## &.got[a]-. = 0x20418 - 0x10318 = 16448<<2
+# GD64: 10318: pcaddi $a0, 16448
+# GD64-NEXT: ld.d $ra, $a0, 0
+# GD64-NEXT: jirl $ra, $ra, 0
+# GD64-NEXT: add.d $a1, $a0, $tp
+
+## &.got[b]-. = 0x20418+48 - 0x10328 = 16456<<2
+# GD64: 10328: pcaddi $a0, 16456
+# GD64-NEXT: ld.d $ra, $a0, 0
+# GD64-NEXT: jirl $ra, $ra, 0
+# GD64-NEXT: add.d $a2, $a0, $tp
+
+## &.got[c]-. = 0x20418+16 - 0x10338 = 16444<<2
+# GD64: 10338: pcaddi $a0, 16
+# GD64-NEXT: ld.d $ra, $a0, 0
+# GD64-NEXT: jirl $ra, $ra, 0
+# GD64-NEXT: add.d $a3, $a0, $tp
+
+## &.got[d]-. = 0x20418+32 - 0x10348 = 16444<<2
+# GD64: 10348: pcaddi $a0, 16444
+# GD64-NEXT: ld.d $ra, $a0, 0
+# GD64-NEXT: jirl $ra, $ra, 0
+# GD64-NEXT: add.d $a4, $a0, $tp
+
+# LE64-RELA: .rela.dyn {
+# LE64-RELA-NEXT: 0x30278 R_LARCH_TLS_DESC64 - 0x8
+# LE64-RELA-NEXT: 0x30288 R_LARCH_TLS_DESC64 - 0x800
+# LE64-RELA-NEXT: 0x30298 R_LARCH_TLS_DESC64 - 0x1000
+# LE64-RELA-NEXT: 0x302A8 R_LARCH_TLS_DESC64 - 0x7FF
+# LE64-RELA-NEXT: }
+# LE64-RELA: Hex dump of section '.got':
+# LE64-RELA-NEXT: 0x00030278 00000000 00000000 00000000 00000000 .
+# LE64-RELA-NEXT: 0x00030288 00000000 00000000 00000000 00000000 .
+# LE64-RELA-NEXT: 0x00030298 00000000 00000000 00000000 00000000 .
+# LE64-RELA-NEXT: 0x000302a8 00000000 00000000 00000000 00000000 .
+
+# LE64-LABEL: <.text>:
+## &.got[a]-. = 0x30278 - 0x20228: 0x10 pages, page offset 0x278
+# LE64-NEXT: 20228: pcalau12i $a0, 16
+# LE64-NEXT: addi.d $a0, $a0, 632
+# LE64-NEXT: ld.d $ra, $a0, 0
+# LE64-NEXT: jirl $ra, $ra, 0
+# LE64-NEXT: add.d $a1, $a0, $tp
+## &.got[b]-. = 0x302a8 - 0x2023c: 0x10 pages, page offset 0x2a8
+# LE64-NEXT: 2023c: pcalau12i $a0, 16
+# LE64-NEXT: addi.d $a0, $a0, 680
+# LE64-NEXT: ld.d $ra, $a0, 0
+# LE64-NEXT: jirl $ra, $ra, 0
+# LE64-NEXT: add.d $a2, $a0, $tp
+## &.got[c]-. = 0x30288 - 0x20250: 0x10 pages, page offset 0x288
+# LE64-NEXT: 20250: pcalau12i $a0, 16
+# LE64-NEXT: addi.d $a0, $a0, 648
+# LE64-NEXT: ld.d $ra, $a0, 0
+# LE64-NEXT: jirl $ra, $ra, 0
+# LE64-NEXT: add.d $a3, $a0, $tp
+## &.got[d]-. = 0x30298 - 0x20264: 0x10 pages, page offset 0x298
+# LE64-NEXT: 20264: pcalau12i $a0, 16
+# LE64-NEXT: addi.d $a0, $a0, 664
+# LE64-NEXT: ld.d $ra, $a0, 0
+# LE64-NEXT: jirl $ra, $ra, 0
+# LE64-NEXT: add.d $a4, $a0, $tp
+
+# IE64-RELA: .rela.dyn {
+# IE64-RELA-NEXT: 0x30428 R_LARCH_TLS_DESC64 - 0x8
+# IE64-RELA-NEXT: 0x30458 R_LARCH_TLS_DESC64 - 0x7FF
+# IE64-RELA-NEXT: 0x30438 R_LARCH_TLS_DESC64 c 0x0
+# IE64-RELA-NEXT: 0x30448 R_LARCH_TLS_DESC64 d 0x0
+# IE64-RELA-NEXT: }
+# IE64-RELA: Hex dump of section '.got':
+# IE64-RELA-NEXT: 0x00030428 00000000 00000000 00000000 00000000 .
+# IE64-RELA-NEXT: 0x00030438 00000000 00000000 00000000 00000000 .
+# IE64-RELA-NEXT: 0x00030448 00000000 00000000 00000000 00000000 .
+# IE64-RELA-NEXT: 0x00030458 00000000 00000000 00000000 00000000 .
+
+## a and b are optimized to use LE. c and d are optimized to IE.
+# IE64-LABEL: <.text>:
+## &.got[a]-. = 0x30428 - 0x202f8: 0x10 pages, page offset 0x428
+# IE64-NEXT: 202f8: pcalau12i $a0, 16
+# IE64-NEXT: addi.d $a0, $a0, 1064
+# IE64-NEXT: ld.d $ra, $a0, 0
+# IE64-NEXT: jirl $ra, $ra, 0
+# IE64-NEXT: add.d $a1, $a0, $tp
+## &.got[b]-. = 0x30458 - 0x2030c: 0x10 pages, page offset 0x458
+# IE64-NEXT: 2030c: pcalau12i $a0, 16
+# IE64-NEXT: addi.d $a0, $a0, 1112
+# IE64-NEXT: ld.d $ra, $a0, 0
+# IE64-NEXT: jirl $ra, $ra, 0
+# IE64-NEXT: add.d $a2, $a0, $tp
+## &.got[c]-. = 0x30438 - 0x20320: 0x10 pages, page offset 0x438
+# IE64-NEXT: 20320: pcalau12i $a0, 16
+# IE64-NEXT: addi.d $a0, $a0, 1080
+# IE64-NEXT: ld.d $ra, $a0, 0
+# IE64-NEXT: jirl $ra, $ra, 0
+# IE64-NEXT: add.d $a3, $a0, $tp
+## &.got[d]-. = 0x30448 - 0x20334: 0x10 pages, page offset 0x448
+# IE64-NEXT: 20334: pcalau12i $a0, 16
+# IE64-NEXT: addi.d $a0, $a0, 1096
+# IE64-NEXT: ld.d $ra, $a0, 0
+# IE64-NEXT: jirl $ra, $ra, 0
+# IE64-NEXT: add.d $a4, $a0, $tp
+
+# GD32-REL: .rel.dyn {
+# GD32-REL-NEXT: 0x202A4 R_LARCH_TLS_DESC32 -
+# GD32-REL-NEXT: 0x2028C R_LARCH_TLS_DESC32 a
+# GD32-REL-NEXT: 0x20294 R_LARCH_TLS_DESC32 c
+# GD32-REL-NEXT: 0x2029C R_LARCH_TLS_DESC32 d
+# GD32-REL-NEXT: }
+# GD32-REL: Hex dump of section '.got':
+# GD32-REL-NEXT: 0x0002028c 00000000 00000000 00000000 00000000 .
+# GD32-REL-NEXT: 0x0002029c 00000000 00000000 00000000 ff070000 .
+
+#--- a.s
+.macro add dst, src1, src2
+.ifdef ELF32
+add.w \dst, \src1, \src2
+.else
+add.d \dst, \src1, \src2
+.endif
+.endm
+
+la.tls.desc $a0, a
+add $a1, $a0, $tp
+
+la.tls.desc $a0, b
+add $a2, $a0, $tp
+
+la.tls.desc $a0, c
+add $a3, $a0, $tp
+
+la.tls.desc $a0, d
+add $a4, $a0, $tp
+
+.section .tbss,"awT", at nobits
+.globl a
+.zero 8
+a:
+.zero 2039 ## Place b at 0x7ff
+b:
+.zero 1
+
+#--- c.s
+.section .tbss,"awT", at nobits
+.globl c, d
+c:
+.zero 2048 ## Place d at 0x1000
+d:
+.zero 4
>From a48a2b989d747754b0aa33f4f8d0f13a7e7c696c Mon Sep 17 00:00:00 2001
From: yangzhaoxin <yangzhaoxin at loongson.cn>
Date: Thu, 2 Jan 2025 08:48:52 +0800
Subject: [PATCH 3/6] Relax desc_pc_hi20+desc_pc_lo12 ==> desc_pcrel_20
---
lld/ELF/Arch/LoongArch.cpp | 37 +++-------------
lld/test/ELF/loongarch-relax-tlsdesc.s | 60 +++++++++++---------------
2 files changed, 31 insertions(+), 66 deletions(-)
diff --git a/lld/ELF/Arch/LoongArch.cpp b/lld/ELF/Arch/LoongArch.cpp
index a6db15bbf1efd4..ec09437404eddc 100644
--- a/lld/ELF/Arch/LoongArch.cpp
+++ b/lld/ELF/Arch/LoongArch.cpp
@@ -762,9 +762,12 @@ static bool isPairRelaxable(ArrayRef<Relocation> relocs, size_t i) {
// Relax code sequence.
// From:
// pcalau12i $a0, %pc_hi20(sym) | %ld_pc_hi20(sym) | %gd_pc_hi20(sym)
+// | %desc_pc_hi20(sym)
// addi.w/d $a0, $a0, %pc_lo12(sym) | %got_pc_lo12(sym) | %got_pc_lo12(sym)
+// | %desc_pc_lo12(sym)
// To:
-// pcaddi $a0, %pc_lo12(sym) | %got_pc_lo12(sym) | %got_pc_lo12(sym)
+// pcaddi $a0, %pc_lo12(sym) | %got_pc_lo12(sym) | %got_pc_lo12(sym)
+// | %desc_pcrel_20(sym)
//
// From:
// pcalau12i $a0, %got_pc_hi20(sym_got)
@@ -909,33 +912,6 @@ static void relaxTlsLe(Ctx &ctx, const InputSection &sec, size_t i,
}
}
-// Relax TLSDESC code sequence. In LoongArch, the conversion of TLSDESC GD/LD to
-// LE/IE is closely tied to relaxation, similar to how GCC handles it. (Due to
-// the lack of an efficient way for handling conversions in the extreme code
-// model and the difficulty in determining whether the extreme code model is
-// being used in handleTlsRelocation, this approach may seem like a workaround).
-// Consequently, the resulting code sequence depends on whether the conversion
-// to LE/IE is performed.
-//
-// Original code sequence:
-// * pcalau12i $a0, %desc_pc_hi20(sym_desc)
-// * addi.d $a0, $a0, %desc_pc_lo12(sym_desc)
-// * ld.d $ra, $a0, %desc_ld(sym_desc)
-// * jirl $ra, $ra, %desc_call(sym_desc)
-//
-// Cannot convert to LE/IE, but relax:
-// * pcaddi $a0, %desc_pcrel_20(sym_desc)
-// * ld.d $ra, $a0, %desc_ld(sym_desc)
-// * jirl $ra, $ra, %desc_call(sym_desc)
-//
-// FIXME: Implement TLSDESC GD/LD to LE/IE.
-static void relaxTlsdesc(Ctx &ctx, const InputSection &sec, size_t i,
- uint64_t loc, Relocation &rHi20, Relocation &rLo12,
- uint32_t &remove) {
- if (ctx.arg.shared && rHi20.type == R_LARCH_TLS_DESC_PC_HI20)
- return relaxPCHi20Lo12(ctx, sec, i, loc, rHi20, rLo12, remove);
-}
-
static bool relax(Ctx &ctx, InputSection &sec) {
const uint64_t secAddr = sec.getVA();
const MutableArrayRef<Relocation> relocs = sec.relocs();
@@ -978,6 +954,7 @@ static bool relax(Ctx &ctx, InputSection &sec) {
case R_LARCH_GOT_PC_HI20:
case R_LARCH_TLS_GD_PC_HI20:
case R_LARCH_TLS_LD_PC_HI20:
+ case R_LARCH_TLS_DESC_PC_HI20:
// The overflow check for i+2 will be carried out in isPairRelaxable.
if (isPairRelaxable(relocs, i))
relaxPCHi20Lo12(ctx, sec, i, loc, r, relocs[i + 2], remove);
@@ -992,10 +969,6 @@ static bool relax(Ctx &ctx, InputSection &sec) {
if (relaxable(relocs, i))
relaxTlsLe(ctx, sec, i, loc, r, remove);
break;
- case R_LARCH_TLS_DESC_PC_HI20:
- if (isPairRelaxable(relocs, i))
- relaxTlsdesc(ctx, sec, i, loc, r, relocs[i + 2], remove);
- break;
}
// For all anchors whose offsets are <= r.offset, they are preceded by
diff --git a/lld/test/ELF/loongarch-relax-tlsdesc.s b/lld/test/ELF/loongarch-relax-tlsdesc.s
index 6470bdf35cb223..f9f509d8d93f90 100644
--- a/lld/test/ELF/loongarch-relax-tlsdesc.s
+++ b/lld/test/ELF/loongarch-relax-tlsdesc.s
@@ -65,78 +65,70 @@
# GD64-NEXT: add.d $a4, $a0, $tp
# LE64-RELA: .rela.dyn {
-# LE64-RELA-NEXT: 0x30278 R_LARCH_TLS_DESC64 - 0x8
-# LE64-RELA-NEXT: 0x30288 R_LARCH_TLS_DESC64 - 0x800
-# LE64-RELA-NEXT: 0x30298 R_LARCH_TLS_DESC64 - 0x1000
-# LE64-RELA-NEXT: 0x302A8 R_LARCH_TLS_DESC64 - 0x7FF
+# LE64-RELA-NEXT: 0x30268 R_LARCH_TLS_DESC64 - 0x8
+# LE64-RELA-NEXT: 0x30278 R_LARCH_TLS_DESC64 - 0x800
+# LE64-RELA-NEXT: 0x30288 R_LARCH_TLS_DESC64 - 0x1000
+# LE64-RELA-NEXT: 0x30298 R_LARCH_TLS_DESC64 - 0x7FF
# LE64-RELA-NEXT: }
# LE64-RELA: Hex dump of section '.got':
+# LE64-RELA-NEXT: 0x00030268 00000000 00000000 00000000 00000000 .
# LE64-RELA-NEXT: 0x00030278 00000000 00000000 00000000 00000000 .
# LE64-RELA-NEXT: 0x00030288 00000000 00000000 00000000 00000000 .
# LE64-RELA-NEXT: 0x00030298 00000000 00000000 00000000 00000000 .
-# LE64-RELA-NEXT: 0x000302a8 00000000 00000000 00000000 00000000 .
# LE64-LABEL: <.text>:
-## &.got[a]-. = 0x30278 - 0x20228: 0x10 pages, page offset 0x278
-# LE64-NEXT: 20228: pcalau12i $a0, 16
-# LE64-NEXT: addi.d $a0, $a0, 632
+## &.got[a]-. = 0x30268 - 0x20228 = 16400<<2
+# LE64-NEXT: 20228: pcaddi $a0, 16400
# LE64-NEXT: ld.d $ra, $a0, 0
# LE64-NEXT: jirl $ra, $ra, 0
# LE64-NEXT: add.d $a1, $a0, $tp
-## &.got[b]-. = 0x302a8 - 0x2023c: 0x10 pages, page offset 0x2a8
-# LE64-NEXT: 2023c: pcalau12i $a0, 16
-# LE64-NEXT: addi.d $a0, $a0, 680
+## &.got[b]-. = 0x30298 - 0x20238 = 16408<<2
+# LE64-NEXT: 20238: pcaddi $a0, 16408
# LE64-NEXT: ld.d $ra, $a0, 0
# LE64-NEXT: jirl $ra, $ra, 0
# LE64-NEXT: add.d $a2, $a0, $tp
-## &.got[c]-. = 0x30288 - 0x20250: 0x10 pages, page offset 0x288
-# LE64-NEXT: 20250: pcalau12i $a0, 16
-# LE64-NEXT: addi.d $a0, $a0, 648
+## &.got[c]-. = 0x30278 - 0x20248 = 16396<<2
+# LE64-NEXT: 20248: pcaddi $a0, 16396
# LE64-NEXT: ld.d $ra, $a0, 0
# LE64-NEXT: jirl $ra, $ra, 0
# LE64-NEXT: add.d $a3, $a0, $tp
-## &.got[d]-. = 0x30298 - 0x20264: 0x10 pages, page offset 0x298
-# LE64-NEXT: 20264: pcalau12i $a0, 16
-# LE64-NEXT: addi.d $a0, $a0, 664
+## &.got[d]-. = 0x30288 - 0x20258 = 16396<<2
+# LE64-NEXT: 20258: pcaddi $a0, 16396
# LE64-NEXT: ld.d $ra, $a0, 0
# LE64-NEXT: jirl $ra, $ra, 0
# LE64-NEXT: add.d $a4, $a0, $tp
# IE64-RELA: .rela.dyn {
-# IE64-RELA-NEXT: 0x30428 R_LARCH_TLS_DESC64 - 0x8
-# IE64-RELA-NEXT: 0x30458 R_LARCH_TLS_DESC64 - 0x7FF
-# IE64-RELA-NEXT: 0x30438 R_LARCH_TLS_DESC64 c 0x0
-# IE64-RELA-NEXT: 0x30448 R_LARCH_TLS_DESC64 d 0x0
+# IE64-RELA-NEXT: 0x30418 R_LARCH_TLS_DESC64 - 0x8
+# IE64-RELA-NEXT: 0x30448 R_LARCH_TLS_DESC64 - 0x7FF
+# IE64-RELA-NEXT: 0x30428 R_LARCH_TLS_DESC64 c 0x0
+# IE64-RELA-NEXT: 0x30438 R_LARCH_TLS_DESC64 d 0x0
# IE64-RELA-NEXT: }
# IE64-RELA: Hex dump of section '.got':
+# IE64-RELA-NEXT: 0x00030418 00000000 00000000 00000000 00000000 .
# IE64-RELA-NEXT: 0x00030428 00000000 00000000 00000000 00000000 .
# IE64-RELA-NEXT: 0x00030438 00000000 00000000 00000000 00000000 .
# IE64-RELA-NEXT: 0x00030448 00000000 00000000 00000000 00000000 .
-# IE64-RELA-NEXT: 0x00030458 00000000 00000000 00000000 00000000 .
## a and b are optimized to use LE. c and d are optimized to IE.
# IE64-LABEL: <.text>:
-## &.got[a]-. = 0x30428 - 0x202f8: 0x10 pages, page offset 0x428
-# IE64-NEXT: 202f8: pcalau12i $a0, 16
-# IE64-NEXT: addi.d $a0, $a0, 1064
+## &.got[a]-. = 0x30418 - 0x202f8 = 16456<<2
+# IE64-NEXT: 202f8: pcaddi $a0, 16456
# IE64-NEXT: ld.d $ra, $a0, 0
# IE64-NEXT: jirl $ra, $ra, 0
# IE64-NEXT: add.d $a1, $a0, $tp
-## &.got[b]-. = 0x30458 - 0x2030c: 0x10 pages, page offset 0x458
-# IE64-NEXT: 2030c: pcalau12i $a0, 16
-# IE64-NEXT: addi.d $a0, $a0, 1112
+## &.got[b]-. = 0x30448 - 0x20308 = 16464<<2
+# IE64-NEXT: 20308: pcaddi $a0, 16464
# IE64-NEXT: ld.d $ra, $a0, 0
# IE64-NEXT: jirl $ra, $ra, 0
# IE64-NEXT: add.d $a2, $a0, $tp
-## &.got[c]-. = 0x30438 - 0x20320: 0x10 pages, page offset 0x438
-# IE64-NEXT: 20320: pcalau12i $a0, 16
-# IE64-NEXT: addi.d $a0, $a0, 1080
+## &.got[c]-. = 0x30428 - 0x20318 = 16452<<2
+# IE64-NEXT: 20318: pcaddi $a0, 16452
# IE64-NEXT: ld.d $ra, $a0, 0
# IE64-NEXT: jirl $ra, $ra, 0
# IE64-NEXT: add.d $a3, $a0, $tp
-## &.got[d]-. = 0x30448 - 0x20334: 0x10 pages, page offset 0x448
-# IE64-NEXT: 20334: pcalau12i $a0, 16
-# IE64-NEXT: addi.d $a0, $a0, 1096
+## &.got[d]-. = 0x30438 - 0x20328 = 16452<<2
+# IE64-NEXT: 20328: pcaddi $a0, 16452
# IE64-NEXT: ld.d $ra, $a0, 0
# IE64-NEXT: jirl $ra, $ra, 0
# IE64-NEXT: add.d $a4, $a0, $tp
>From 576773c8e2c67bc89cd653f3a52b894446ef590e Mon Sep 17 00:00:00 2001
From: yangzhaoxin <yangzhaoxin at loongson.cn>
Date: Sat, 4 Jan 2025 09:47:34 +0800
Subject: [PATCH 4/6] Modify test loongarch-relax-tlsdesc.s
---
lld/test/ELF/loongarch-relax-tlsdesc.s | 319 ++++++++++++++++---------
1 file changed, 209 insertions(+), 110 deletions(-)
diff --git a/lld/test/ELF/loongarch-relax-tlsdesc.s b/lld/test/ELF/loongarch-relax-tlsdesc.s
index f9f509d8d93f90..6561066ebeb5ca 100644
--- a/lld/test/ELF/loongarch-relax-tlsdesc.s
+++ b/lld/test/ELF/loongarch-relax-tlsdesc.s
@@ -3,166 +3,265 @@
# RUN: llvm-mc -filetype=obj -triple=loongarch64 -mattr=+relax a.s -o a.64.o
# RUN: llvm-mc -filetype=obj -triple=loongarch64 -mattr=+relax c.s -o c.64.o
# RUN: ld.lld -shared -soname=c.64.so c.64.o -o c.64.so
-# RUN: llvm-mc -filetype=obj -triple=loongarch32 -mattr=+relax --defsym ELF32=1 a.s -o a.32.o
-# RUN: llvm-mc -filetype=obj -triple=loongarch32 -mattr=+relax --defsym ELF32=1 c.s -o c.32.o
-# RUN: ld.lld -shared -soname=c.32.so c.32.o -o c.32.so
## Test the TLSDESC relaxation. Also check --emit-relocs.
# RUN: ld.lld -shared --emit-relocs -z now a.64.o c.64.o -o a.64.so
# RUN: llvm-readobj -r -x .got a.64.so | FileCheck --check-prefix=GD64-RELA %s
-# RUN: llvm-objdump --no-show-raw-insn -dr a.64.so | FileCheck %s --check-prefix=GD64
+# RUN: llvm-objdump --no-show-raw-insn -dr -h a.64.so | FileCheck %s --check-prefix=GD64
## FIXME: The transition frome TLSDESC to IE/LE has not yet been implemented.
## Keep the dynamic relocations and hand them over to dynamic linker.
# RUN: ld.lld -e 0 -z now --emit-relocs a.64.o c.64.o -o a.64.le
# RUN: llvm-readobj -r -x .got a.64.le | FileCheck --check-prefix=LE64-RELA %s
-# RUN: llvm-objdump --no-show-raw-insn -d a.64.le | FileCheck %s --check-prefix=LE64
+# RUN: llvm-objdump --no-show-raw-insn -dr -h a.64.le | FileCheck %s --check-prefix=LE64
+
+# RUN: ld.lld -e 0 -z now --no-relax a.64.o c.64.o -o a.64.le.norelax
+# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.le.norelax | FileCheck %s --check-prefix=LE64-NORELAX
# RUN: ld.lld -e 0 -z now --emit-relocs a.64.o c.64.so -o a.64.ie
# RUN: llvm-readobj -r -x .got a.64.ie | FileCheck --check-prefix=IE64-RELA %s
-# RUN: llvm-objdump --no-show-raw-insn -d a.64.ie | FileCheck %s --check-prefix=IE64
-
-## 32-bit code is mostly the same. We only test a few variants.
+# RUN: llvm-objdump --no-show-raw-insn -dr -h a.64.ie | FileCheck %s --check-prefix=IE64
-# RUN: ld.lld -shared -z now a.32.o c.32.o -o rel.32.so -z rel
-# RUN: llvm-readobj -r -x .got rel.32.so | FileCheck --check-prefix=GD32-REL %s
+# RUN: ld.lld -e 0 -z now --no-relax a.64.o c.64.so -o a.64.ie.norelax
+# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.ie.norelax | FileCheck %s --check-prefix=IE64-NORELAX
# GD64-RELA: .rela.dyn {
-# GD64-RELA-NEXT: 0x20448 R_LARCH_TLS_DESC64 - 0x7FF
-# GD64-RELA-NEXT: 0x20418 R_LARCH_TLS_DESC64 a 0x0
-# GD64-RELA-NEXT: 0x20428 R_LARCH_TLS_DESC64 c 0x0
-# GD64-RELA-NEXT: 0x20438 R_LARCH_TLS_DESC64 d 0x0
+# GD64-RELA-NEXT: 0x20460 R_LARCH_TLS_DESC64 - 0x7FF
+# GD64-RELA-NEXT: 0x20430 R_LARCH_TLS_DESC64 a 0x0
+# GD64-RELA-NEXT: 0x20440 R_LARCH_TLS_DESC64 c 0x0
+# GD64-RELA-NEXT: 0x20450 R_LARCH_TLS_DESC64 d 0x0
# GD64-RELA-NEXT: }
# GD64-RELA: Hex dump of section '.got':
-# GD64-RELA-NEXT: 0x00020418 00000000 00000000 00000000 00000000 .
-# GD64-RELA-NEXT: 0x00020428 00000000 00000000 00000000 00000000 .
-# GD64-RELA-NEXT: 0x00020438 00000000 00000000 00000000 00000000 .
-# GD64-RELA-NEXT: 0x00020448 00000000 00000000 00000000 00000000 .
+# GD64-RELA-NEXT: 0x00020430 00000000 00000000 00000000 00000000 .
+# GD64-RELA-NEXT: 0x00020440 00000000 00000000 00000000 00000000 .
+# GD64-RELA-NEXT: 0x00020450 00000000 00000000 00000000 00000000 .
+# GD64-RELA-NEXT: 0x00020460 00000000 00000000 00000000 00000000 .
+
+# GD64: .got 00000040 0000000000020430
-## &.got[a]-. = 0x20418 - 0x10318 = 16448<<2
-# GD64: 10318: pcaddi $a0, 16448
+## &.got[a]-. = 0x20430 - 0x10318 = 16454<<2
+# GD64: 10318: pcaddi $a0, 16454
# GD64-NEXT: ld.d $ra, $a0, 0
# GD64-NEXT: jirl $ra, $ra, 0
# GD64-NEXT: add.d $a1, $a0, $tp
-## &.got[b]-. = 0x20418+48 - 0x10328 = 16456<<2
-# GD64: 10328: pcaddi $a0, 16456
+## &.got[b]-. = 0x20430+48 - 0x10328: 0x10 pages, page offset 0x460
+## R_LARCH_RELAX does not appear in pairs. No relaxation.
+# GD64: 10328: pcalau12i $a0, 16
+# GD64-NEXT: addi.d $a0, $a0, 1120
# GD64-NEXT: ld.d $ra, $a0, 0
# GD64-NEXT: jirl $ra, $ra, 0
# GD64-NEXT: add.d $a2, $a0, $tp
-## &.got[c]-. = 0x20418+16 - 0x10338 = 16444<<2
-# GD64: 10338: pcaddi $a0, 16
+## &.got[c]-. = 0x20430+16 - 0x1033c: 0x10 pages, page offset 0x440
+## Without R_LARCH_RELAX relocation. No relaxation.
+# GD64: 1033c: pcalau12i $a0, 16
+# GD64-NEXT: addi.d $t0, $zero, 0
+# GD64-NEXT: addi.d $a0, $a0, 1088
+# GD64-NEXT: addi.d $t0, $t0, 1
# GD64-NEXT: ld.d $ra, $a0, 0
+# GD64-NEXT: addi.d $t0, $t0, 1
# GD64-NEXT: jirl $ra, $ra, 0
# GD64-NEXT: add.d $a3, $a0, $tp
-## &.got[d]-. = 0x20418+32 - 0x10348 = 16444<<2
-# GD64: 10348: pcaddi $a0, 16444
+## &.got[d]-. = 0x20430+32 - 0x1035c = 16445<<2
+# GD64: 1035c: pcaddi $a0, 16445
# GD64-NEXT: ld.d $ra, $a0, 0
# GD64-NEXT: jirl $ra, $ra, 0
# GD64-NEXT: add.d $a4, $a0, $tp
# LE64-RELA: .rela.dyn {
-# LE64-RELA-NEXT: 0x30268 R_LARCH_TLS_DESC64 - 0x8
-# LE64-RELA-NEXT: 0x30278 R_LARCH_TLS_DESC64 - 0x800
-# LE64-RELA-NEXT: 0x30288 R_LARCH_TLS_DESC64 - 0x1000
-# LE64-RELA-NEXT: 0x30298 R_LARCH_TLS_DESC64 - 0x7FF
+# LE64-RELA-NEXT: 0x30280 R_LARCH_TLS_DESC64 - 0x8
+# LE64-RELA-NEXT: 0x30290 R_LARCH_TLS_DESC64 - 0x800
+# LE64-RELA-NEXT: 0x302A0 R_LARCH_TLS_DESC64 - 0x1000
+# LE64-RELA-NEXT: 0x302B0 R_LARCH_TLS_DESC64 - 0x7FF
# LE64-RELA-NEXT: }
# LE64-RELA: Hex dump of section '.got':
-# LE64-RELA-NEXT: 0x00030268 00000000 00000000 00000000 00000000 .
-# LE64-RELA-NEXT: 0x00030278 00000000 00000000 00000000 00000000 .
-# LE64-RELA-NEXT: 0x00030288 00000000 00000000 00000000 00000000 .
-# LE64-RELA-NEXT: 0x00030298 00000000 00000000 00000000 00000000 .
-
-# LE64-LABEL: <.text>:
-## &.got[a]-. = 0x30268 - 0x20228 = 16400<<2
-# LE64-NEXT: 20228: pcaddi $a0, 16400
-# LE64-NEXT: ld.d $ra, $a0, 0
-# LE64-NEXT: jirl $ra, $ra, 0
-# LE64-NEXT: add.d $a1, $a0, $tp
-## &.got[b]-. = 0x30298 - 0x20238 = 16408<<2
-# LE64-NEXT: 20238: pcaddi $a0, 16408
-# LE64-NEXT: ld.d $ra, $a0, 0
-# LE64-NEXT: jirl $ra, $ra, 0
-# LE64-NEXT: add.d $a2, $a0, $tp
-## &.got[c]-. = 0x30278 - 0x20248 = 16396<<2
-# LE64-NEXT: 20248: pcaddi $a0, 16396
-# LE64-NEXT: ld.d $ra, $a0, 0
-# LE64-NEXT: jirl $ra, $ra, 0
-# LE64-NEXT: add.d $a3, $a0, $tp
-## &.got[d]-. = 0x30288 - 0x20258 = 16396<<2
-# LE64-NEXT: 20258: pcaddi $a0, 16396
-# LE64-NEXT: ld.d $ra, $a0, 0
-# LE64-NEXT: jirl $ra, $ra, 0
-# LE64-NEXT: add.d $a4, $a0, $tp
+# LE64-RELA-NEXT: 0x00030280 00000000 00000000 00000000 00000000 .
+# LE64-RELA-NEXT: 0x00030290 00000000 00000000 00000000 00000000 .
+# LE64-RELA-NEXT: 0x000302a0 00000000 00000000 00000000 00000000 .
+# LE64-RELA-NEXT: 0x000302b0 00000000 00000000 00000000 00000000 .
+
+# LE64: .got 00000040 0000000000030280
+
+## &.got[a]-. = 0x30280 - 0x20228 = 16406<<2
+# LE64: 20228: pcaddi $a0, 16406
+# LE64-NEXT: ld.d $ra, $a0, 0
+# LE64-NEXT: jirl $ra, $ra, 0
+# LE64-NEXT: add.d $a1, $a0, $tp
+
+## &.got[b]-. = 0x30280+48 - 0x20238: 0x10 pages, page offset 0x2b0
+## R_LARCH_RELAX does not appear in pairs. No relaxation.
+# LE64: 20238: pcalau12i $a0, 16
+# LE64-NEXT: addi.d $a0, $a0, 688
+# LE64-NEXT: ld.d $ra, $a0, 0
+# LE64-NEXT: jirl $ra, $ra, 0
+# LE64-NEXT: add.d $a2, $a0, $tp
+
+## &.got[c]-. = 0x30280+16 - 0x2024c: 0x10 pages, page offset 0x290
+## Without R_LARCH_RELAX relocation. No relaxation.
+# LE64: 2024c: pcalau12i $a0, 16
+# LE64-NEXT: addi.d $t0, $zero, 0
+# LE64-NEXT: addi.d $a0, $a0, 656
+# LE64-NEXT: addi.d $t0, $t0, 1
+# LE64-NEXT: ld.d $ra, $a0, 0
+# LE64-NEXT: addi.d $t0, $t0, 1
+# LE64-NEXT: jirl $ra, $ra, 0
+# LE64-NEXT: add.d $a3, $a0, $tp
+
+## &.got[d]-. = 0x30280+32 - 0x2026c = 16397<<2
+# LE64: 2026c: pcaddi $a0, 16397
+# LE64-NEXT: ld.d $ra, $a0, 0
+# LE64-NEXT: jirl $ra, $ra, 0
+# LE64-NEXT: add.d $a4, $a0, $tp
+
+# LE64-NORELAX: .got 00000040 0000000000030288
+
+## &.got[a]-. = 0x30288 - 0x20228 = 0x10 pages, page offset 0x288
+# LE64-NORELAX: 20228: pcalau12i $a0, 16
+# LE64-NORELAX-NEXT: addi.d $a0, $a0, 648
+# LE64-NORELAX-NEXT: ld.d $ra, $a0, 0
+# LE64-NORELAX-NEXT: jirl $ra, $ra, 0
+# LE64-NORELAX-NEXT: add.d $a1, $a0, $tp
+
+## &.got[b]-. = 0x30288+48 - 0x2023c: 0x10 pages, page offset 0x2b8
+## R_LARCH_RELAX does not appear in pairs. No relaxation.
+# LE64-NORELAX: 2023c: pcalau12i $a0, 16
+# LE64-NORELAX-NEXT: addi.d $a0, $a0, 696
+# LE64-NORELAX-NEXT: ld.d $ra, $a0, 0
+# LE64-NORELAX-NEXT: jirl $ra, $ra, 0
+# LE64-NORELAX-NEXT: add.d $a2, $a0, $tp
+
+## &.got[c]-. = 0x30288+16 - 0x20250: 0x10 pages, page offset 0x298
+## Without R_LARCH_RELAX relocation. No relaxation.
+# LE64-NORELAX: 20250: pcalau12i $a0, 16
+# LE64-NORELAX-NEXT: addi.d $t0, $zero, 0
+# LE64-NORELAX-NEXT: addi.d $a0, $a0, 664
+# LE64-NORELAX-NEXT: addi.d $t0, $t0, 1
+# LE64-NORELAX-NEXT: ld.d $ra, $a0, 0
+# LE64-NORELAX-NEXT: addi.d $t0, $t0, 1
+# LE64-NORELAX-NEXT: jirl $ra, $ra, 0
+# LE64-NORELAX-NEXT: add.d $a3, $a0, $tp
+
+## &.got[d]-. = 0x30288+32 - 0x20270: 0x10 pages, page offset 0x2a8
+# LE64-NORELAX: 20270: pcalau12i $a0, 16
+# LE64-NORELAX-NEXT: addi.d $a0, $a0, 680
+# LE64-NORELAX-NEXT: ld.d $ra, $a0, 0
+# LE64-NORELAX-NEXT: jirl $ra, $ra, 0
+# LE64-NORELAX-NEXT: add.d $a4, $a0, $tp
# IE64-RELA: .rela.dyn {
-# IE64-RELA-NEXT: 0x30418 R_LARCH_TLS_DESC64 - 0x8
-# IE64-RELA-NEXT: 0x30448 R_LARCH_TLS_DESC64 - 0x7FF
-# IE64-RELA-NEXT: 0x30428 R_LARCH_TLS_DESC64 c 0x0
-# IE64-RELA-NEXT: 0x30438 R_LARCH_TLS_DESC64 d 0x0
+# IE64-RELA-NEXT: 0x30430 R_LARCH_TLS_DESC64 - 0x8
+# IE64-RELA-NEXT: 0x30460 R_LARCH_TLS_DESC64 - 0x7FF
+# IE64-RELA-NEXT: 0x30440 R_LARCH_TLS_DESC64 c 0x0
+# IE64-RELA-NEXT: 0x30450 R_LARCH_TLS_DESC64 d 0x0
# IE64-RELA-NEXT: }
# IE64-RELA: Hex dump of section '.got':
-# IE64-RELA-NEXT: 0x00030418 00000000 00000000 00000000 00000000 .
-# IE64-RELA-NEXT: 0x00030428 00000000 00000000 00000000 00000000 .
-# IE64-RELA-NEXT: 0x00030438 00000000 00000000 00000000 00000000 .
-# IE64-RELA-NEXT: 0x00030448 00000000 00000000 00000000 00000000 .
+# IE64-RELA-NEXT: 0x00030430 00000000 00000000 00000000 00000000 .
+# IE64-RELA-NEXT: 0x00030440 00000000 00000000 00000000 00000000 .
+# IE64-RELA-NEXT: 0x00030450 00000000 00000000 00000000 00000000 .
+# IE64-RELA-NEXT: 0x00030460 00000000 00000000 00000000 00000000 .
+
+# IE64: .got 00000040 0000000000030430
## a and b are optimized to use LE. c and d are optimized to IE.
-# IE64-LABEL: <.text>:
-## &.got[a]-. = 0x30418 - 0x202f8 = 16456<<2
-# IE64-NEXT: 202f8: pcaddi $a0, 16456
-# IE64-NEXT: ld.d $ra, $a0, 0
-# IE64-NEXT: jirl $ra, $ra, 0
-# IE64-NEXT: add.d $a1, $a0, $tp
-## &.got[b]-. = 0x30448 - 0x20308 = 16464<<2
-# IE64-NEXT: 20308: pcaddi $a0, 16464
-# IE64-NEXT: ld.d $ra, $a0, 0
-# IE64-NEXT: jirl $ra, $ra, 0
-# IE64-NEXT: add.d $a2, $a0, $tp
-## &.got[c]-. = 0x30428 - 0x20318 = 16452<<2
-# IE64-NEXT: 20318: pcaddi $a0, 16452
-# IE64-NEXT: ld.d $ra, $a0, 0
-# IE64-NEXT: jirl $ra, $ra, 0
-# IE64-NEXT: add.d $a3, $a0, $tp
-## &.got[d]-. = 0x30438 - 0x20328 = 16452<<2
-# IE64-NEXT: 20328: pcaddi $a0, 16452
-# IE64-NEXT: ld.d $ra, $a0, 0
-# IE64-NEXT: jirl $ra, $ra, 0
-# IE64-NEXT: add.d $a4, $a0, $tp
-
-# GD32-REL: .rel.dyn {
-# GD32-REL-NEXT: 0x202A4 R_LARCH_TLS_DESC32 -
-# GD32-REL-NEXT: 0x2028C R_LARCH_TLS_DESC32 a
-# GD32-REL-NEXT: 0x20294 R_LARCH_TLS_DESC32 c
-# GD32-REL-NEXT: 0x2029C R_LARCH_TLS_DESC32 d
-# GD32-REL-NEXT: }
-# GD32-REL: Hex dump of section '.got':
-# GD32-REL-NEXT: 0x0002028c 00000000 00000000 00000000 00000000 .
-# GD32-REL-NEXT: 0x0002029c 00000000 00000000 00000000 ff070000 .
+## &.got[a]-. = 0x30430 - 0x202f8 = 16462<<2
+# IE64: 202f8: pcaddi $a0, 16462
+# IE64-NEXT: ld.d $ra, $a0, 0
+# IE64-NEXT: jirl $ra, $ra, 0
+# IE64-NEXT: add.d $a1, $a0, $tp
-#--- a.s
-.macro add dst, src1, src2
-.ifdef ELF32
-add.w \dst, \src1, \src2
-.else
-add.d \dst, \src1, \src2
-.endif
-.endm
+## &.got[b]-. = 0x30430+48 - 0x20308: 0x10 pages, page offset 0x460
+## R_LARCH_RELAX does not appear in pairs. No relaxation.
+# IE64: 20308: pcalau12i $a0, 16
+# IE64-NEXT: addi.d $a0, $a0, 1120
+# IE64-NEXT: ld.d $ra, $a0, 0
+# IE64-NEXT: jirl $ra, $ra, 0
+# IE64-NEXT: add.d $a2, $a0, $tp
+
+## &.got[c]-. = 0x30430+16 - 0x2031c: 0x10 pages, page offset 0x440
+## Without R_LARCH_RELAX relocation. No relaxation.
+# IE64: 2031c: pcalau12i $a0, 16
+# IE64-NEXT: addi.d $t0, $zero, 0
+# IE64-NEXT: addi.d $a0, $a0, 1088
+# IE64-NEXT: addi.d $t0, $t0, 1
+# IE64-NEXT: ld.d $ra, $a0, 0
+# IE64-NEXT: addi.d $t0, $t0, 1
+# IE64-NEXT: jirl $ra, $ra, 0
+# IE64-NEXT: add.d $a3, $a0, $tp
+
+## &.got[d]-. = 0x30430+32 - 0x2033c = 16453<<2
+# IE64: 2033c: pcaddi $a0, 16453
+# IE64-NEXT: ld.d $ra, $a0, 0
+# IE64-NEXT: jirl $ra, $ra, 0
+# IE64-NEXT: add.d $a4, $a0, $tp
+
+# IE64-NORELAX: .got 00000040 0000000000030438
+## &.got[a]-. = 0x30438 - 0x202f8 = 0x10 pages, page offset 0x438
+# IE64-NORELAX: 202f8: pcalau12i $a0, 16
+# IE64-NORELAX-NEXT: addi.d $a0, $a0, 1080
+# IE64-NORELAX-NEXT: ld.d $ra, $a0, 0
+# IE64-NORELAX-NEXT: jirl $ra, $ra, 0
+# IE64-NORELAX-NEXT: add.d $a1, $a0, $tp
+
+## &.got[b]-. = 0x30438+48 - 0x2030c: 0x10 pages, page offset 0x468
+## R_LARCH_RELAX does not appear in pairs. No relaxation.
+# IE64-NORELAX: 2030c: pcalau12i $a0, 16
+# IE64-NORELAX-NEXT: addi.d $a0, $a0, 1128
+# IE64-NORELAX-NEXT: ld.d $ra, $a0, 0
+# IE64-NORELAX-NEXT: jirl $ra, $ra, 0
+# IE64-NORELAX-NEXT: add.d $a2, $a0, $tp
+
+## &.got[c]-. = 0x30438+16 - 0x20320: 0x10 pages, page offset 0x448
+## Without R_LARCH_RELAX relocation. No relaxation.
+# IE64-NORELAX: 20320: pcalau12i $a0, 16
+# IE64-NORELAX-NEXT: addi.d $t0, $zero, 0
+# IE64-NORELAX-NEXT: addi.d $a0, $a0, 1096
+# IE64-NORELAX-NEXT: addi.d $t0, $t0, 1
+# IE64-NORELAX-NEXT: ld.d $ra, $a0, 0
+# IE64-NORELAX-NEXT: addi.d $t0, $t0, 1
+# IE64-NORELAX-NEXT: jirl $ra, $ra, 0
+# IE64-NORELAX-NEXT: add.d $a3, $a0, $tp
+
+## &.got[d]-. = 0x30438+32 - 0x20340: 0x10 pages, page offset 0x458
+# IE64-NORELAX: 20340: pcalau12i $a0, 16
+# IE64-NORELAX-NEXT: addi.d $a0, $a0, 1112
+# IE64-NORELAX-NEXT: ld.d $ra, $a0, 0
+# IE64-NORELAX-NEXT: jirl $ra, $ra, 0
+# IE64-NORELAX-NEXT: add.d $a4, $a0, $tp
+
+#--- a.s
la.tls.desc $a0, a
-add $a1, $a0, $tp
+add.d $a1, $a0, $tp
-la.tls.desc $a0, b
-add $a2, $a0, $tp
+# ADDI.D does not have R_LARCH_RELAX. No relaxation.
+pcalau12i $a0, %desc_pc_hi20(b)
+.reloc .-4, R_LARCH_RELAX, 0
+addi.d $a0, $a0, %desc_pc_lo12(b)
+ld.d $ra, $a0, %desc_ld(b)
+jirl $ra, $ra, %desc_call(b)
+add.d $a2, $a0, $tp
-la.tls.desc $a0, c
-add $a3, $a0, $tp
+# TLSDESC to LE. No relaxation.
+pcalau12i $a0, %desc_pc_hi20(c)
+addi.d $t0, $zero, 0
+addi.d $a0, $a0, %desc_pc_lo12(c)
+addi.d $t0, $t0, 1
+ld.d $ra, $a0, %desc_ld(c)
+addi.d $t0, $t0, 1
+jirl $ra, $ra, %desc_call(c)
+add.d $a3, $a0, $tp
-la.tls.desc $a0, d
-add $a4, $a0, $tp
+# PCALAU12I and ADDI.D have R_LARCH_RELAX. We preform relaxation.
+pcalau12i $a0, %desc_pc_hi20(d)
+.reloc .-4, R_LARCH_RELAX, 0
+addi.d $a0, $a0, %desc_pc_lo12(d)
+.reloc .-4, R_LARCH_RELAX, 0
+ld.d $ra, $a0, %desc_ld(d)
+jirl $ra, $ra, %desc_call(d)
+add.d $a4, $a0, $tp
.section .tbss,"awT", at nobits
.globl a
>From b2f11e3b33a0c2650c165992c8497fd6be851113 Mon Sep 17 00:00:00 2001
From: yangzhaoxin <yangzhaoxin at loongson.cn>
Date: Sat, 4 Jan 2025 12:35:54 +0800
Subject: [PATCH 5/6] Remove --emit-relocs in loongarch-relax-tlsdesc.s
llvm-objdump -dr cannot print the relocations, however objudmp works
well. It also occurs for RISCV.
---
lld/test/ELF/loongarch-relax-tlsdesc.s | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/lld/test/ELF/loongarch-relax-tlsdesc.s b/lld/test/ELF/loongarch-relax-tlsdesc.s
index 6561066ebeb5ca..5a8cf7acbb7de0 100644
--- a/lld/test/ELF/loongarch-relax-tlsdesc.s
+++ b/lld/test/ELF/loongarch-relax-tlsdesc.s
@@ -4,24 +4,24 @@
# RUN: llvm-mc -filetype=obj -triple=loongarch64 -mattr=+relax c.s -o c.64.o
# RUN: ld.lld -shared -soname=c.64.so c.64.o -o c.64.so
-## Test the TLSDESC relaxation. Also check --emit-relocs.
-# RUN: ld.lld -shared --emit-relocs -z now a.64.o c.64.o -o a.64.so
+## Test the TLSDESC relaxation.
+# RUN: ld.lld -shared -z now a.64.o c.64.o -o a.64.so
# RUN: llvm-readobj -r -x .got a.64.so | FileCheck --check-prefix=GD64-RELA %s
# RUN: llvm-objdump --no-show-raw-insn -dr -h a.64.so | FileCheck %s --check-prefix=GD64
## FIXME: The transition frome TLSDESC to IE/LE has not yet been implemented.
## Keep the dynamic relocations and hand them over to dynamic linker.
-# RUN: ld.lld -e 0 -z now --emit-relocs a.64.o c.64.o -o a.64.le
+# RUN: ld.lld -e 0 -z now a.64.o c.64.o -o a.64.le
# RUN: llvm-readobj -r -x .got a.64.le | FileCheck --check-prefix=LE64-RELA %s
-# RUN: llvm-objdump --no-show-raw-insn -dr -h a.64.le | FileCheck %s --check-prefix=LE64
+# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.le | FileCheck %s --check-prefix=LE64
# RUN: ld.lld -e 0 -z now --no-relax a.64.o c.64.o -o a.64.le.norelax
# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.le.norelax | FileCheck %s --check-prefix=LE64-NORELAX
-# RUN: ld.lld -e 0 -z now --emit-relocs a.64.o c.64.so -o a.64.ie
+# RUN: ld.lld -e 0 -z now a.64.o c.64.so -o a.64.ie
# RUN: llvm-readobj -r -x .got a.64.ie | FileCheck --check-prefix=IE64-RELA %s
-# RUN: llvm-objdump --no-show-raw-insn -dr -h a.64.ie | FileCheck %s --check-prefix=IE64
+# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.ie | FileCheck %s --check-prefix=IE64
# RUN: ld.lld -e 0 -z now --no-relax a.64.o c.64.so -o a.64.ie.norelax
# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.ie.norelax | FileCheck %s --check-prefix=IE64-NORELAX
>From d5c9d649c92235eab855cd7bad34ab3e1580a6fa Mon Sep 17 00:00:00 2001
From: yangzhaoxin <yangzhaoxin at loongson.cn>
Date: Fri, 17 Jan 2025 08:49:30 +0800
Subject: [PATCH 6/6] Modify test. Add --relax option.
---
lld/test/ELF/loongarch-relax-tlsdesc.s | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/lld/test/ELF/loongarch-relax-tlsdesc.s b/lld/test/ELF/loongarch-relax-tlsdesc.s
index 5a8cf7acbb7de0..f9d984ad6387a3 100644
--- a/lld/test/ELF/loongarch-relax-tlsdesc.s
+++ b/lld/test/ELF/loongarch-relax-tlsdesc.s
@@ -2,28 +2,28 @@
# RUN: rm -rf %t && split-file %s %t && cd %t
# RUN: llvm-mc -filetype=obj -triple=loongarch64 -mattr=+relax a.s -o a.64.o
# RUN: llvm-mc -filetype=obj -triple=loongarch64 -mattr=+relax c.s -o c.64.o
-# RUN: ld.lld -shared -soname=c.64.so c.64.o -o c.64.so
+# RUN: ld.lld --relax -shared -soname=c.64.so c.64.o -o c.64.so
## Test the TLSDESC relaxation.
-# RUN: ld.lld -shared -z now a.64.o c.64.o -o a.64.so
+# RUN: ld.lld --relax -shared -z now a.64.o c.64.o -o a.64.so
# RUN: llvm-readobj -r -x .got a.64.so | FileCheck --check-prefix=GD64-RELA %s
# RUN: llvm-objdump --no-show-raw-insn -dr -h a.64.so | FileCheck %s --check-prefix=GD64
## FIXME: The transition frome TLSDESC to IE/LE has not yet been implemented.
## Keep the dynamic relocations and hand them over to dynamic linker.
-# RUN: ld.lld -e 0 -z now a.64.o c.64.o -o a.64.le
+# RUN: ld.lld --relax -e 0 -z now a.64.o c.64.o -o a.64.le
# RUN: llvm-readobj -r -x .got a.64.le | FileCheck --check-prefix=LE64-RELA %s
# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.le | FileCheck %s --check-prefix=LE64
-# RUN: ld.lld -e 0 -z now --no-relax a.64.o c.64.o -o a.64.le.norelax
+# RUN: ld.lld --no-relax -e 0 -z now a.64.o c.64.o -o a.64.le.norelax
# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.le.norelax | FileCheck %s --check-prefix=LE64-NORELAX
-# RUN: ld.lld -e 0 -z now a.64.o c.64.so -o a.64.ie
+# RUN: ld.lld --relax -e 0 -z now a.64.o c.64.so -o a.64.ie
# RUN: llvm-readobj -r -x .got a.64.ie | FileCheck --check-prefix=IE64-RELA %s
# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.ie | FileCheck %s --check-prefix=IE64
-# RUN: ld.lld -e 0 -z now --no-relax a.64.o c.64.so -o a.64.ie.norelax
+# RUN: ld.lld --no-relax -e 0 -z now a.64.o c.64.so -o a.64.ie.norelax
# RUN: llvm-objdump --no-show-raw-insn -d -h a.64.ie.norelax | FileCheck %s --check-prefix=IE64-NORELAX
# GD64-RELA: .rela.dyn {
More information about the llvm-branch-commits
mailing list