caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[lld][LoongArch] Support TLSDESC GD/LD to IE/LE (#123715)

Browse Source 
Support TLSDESC to initial-exec or local-exec optimizations. Introduce a
new hook RE_LOONGARCH_RELAX_TLS_GD_TO_IE_PAGE_PC and use existing
R_RELAX_TLS_GD_TO_IE_ABS to support TLSDESC => IE, while use existing
R_RELAX_TLS_GD_TO_LE to support TLSDESC => LE.
    
In normal or medium code model, there are two forms of code sequences:
* 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)
------
* pcaddi     $a0, %desc_pcrel_20(sym_desc)
* ld.d       $ra, $a0, %desc_ld(sym_desc)
* jirl       $ra, $ra, %desc_call(sym_desc)
    
Convert to IE:
* pcalau12i $a0, %ie_pc_hi20(sym_ie)
* ld.[wd]   $a0, $a0, %ie_pc_lo12(sym_ie)

Convert to LE:
* lu12i.w $a0, %le_hi20(sym_le) # le_hi20 != 0, otherwise NOP
* ori $a0 src, %le_lo12(sym_le) # le_hi20 != 0, src = $a0, otherwise src = $zero

Simplicity, whether tlsdescToIe or tlsdescToLe, we always tend to
convert the preceding instructions to NOPs, due to both forms of code
sequence (corresponding to relocation combinations:
R_LARCH_TLS_DESC_PC_HI20+R_LARCH_TLS_DESC_PC_LO12 and
R_LARCH_TLS_DESC_PCREL20_S2) have same process.
    
TODO: When relaxation enables, redundant NOPs can be removed. It will be
implemented in a future patch.
    
Note: All forms of TLSDESC code sequences should not appear interleaved
in the normal, medium or extreme code model, which compilers do not
generate and lld is unsupported. This is thanks to the guard in
PostRASchedulerList.cpp in llvm.
```
Calls are not scheduling boundaries before register allocation,
but post-ra we don't gain anything by scheduling across calls
since we don't need to worry about register pressure.
```
This commit is contained in:
Zhaoxin Yang
2025-07-02 16:09:51 +08:00
committed by GitHub
parent f1cc0b607b
commit 2c1900860c
7 changed files with 399 additions and 170 deletions
Download Patch File Download Diff File Expand all files Collapse all files

151
lld/ELF/Arch/LoongArch.cpp
View File

@@ -39,8 +39,13 @@ public:
void relocate(uint8_t *loc, const Relocation &rel,
uint64_t val) const override;
bool relaxOnce(int pass) const override;
RelExpr adjustTlsExpr(RelType type, RelExpr expr) const override;
void relocateAlloc(InputSectionBase &sec, uint8_t *buf) const override;
void finalizeRelax(int passes) const override;
private:
void tlsdescToIe(uint8_t *loc, const Relocation &rel, uint64_t val) const;
void tlsdescToLe(uint8_t *loc, const Relocation &rel, uint64_t val) const;
};
} // end anonymous namespace
@@ -58,6 +63,7 @@ enum Op {
LU12I_W = 0x14000000,
PCADDI = 0x18000000,
PCADDU12I = 0x1c000000,
PCALAU12I = 0x1a000000,
LD_W = 0x28800000,
LD_D = 0x28c00000,
JIRL = 0x4c000000,
@@ -69,6 +75,7 @@ enum Reg {
R_ZERO = 0,
R_RA = 1,
R_TP = 2,
R_A0 = 4,
R_T0 = 12,
R_T1 = 13,
R_T2 = 14,
@@ -961,7 +968,8 @@ static bool relax(Ctx &ctx, InputSection &sec) {
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))
if (r.expr != RE_LOONGARCH_RELAX_TLS_GD_TO_IE_PAGE_PC &&
r.expr != R_RELAX_TLS_GD_TO_LE && isPairRelaxable(relocs, i))
relaxPCHi20Lo12(ctx, sec, i, loc, r, relocs[i + 2], remove);
break;
case R_LARCH_CALL36:
@@ -1046,6 +1054,104 @@ static void tlsIeToLe(uint8_t *loc, const Relocation &rel, uint64_t val) {
}
}
// Convert TLSDESC GD/LD to IE.
// In normal or medium code model, there are two forms of code sequences:
// * 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)
// ------
// * pcaddi $a0, %desc_pcrel_20(a)
// * load $ra, $a0, %desc_ld(a)
// * jirl $ra, $ra, %desc_call(a)
//
// The code sequence obtained is as follows:
// * pcalau12i $a0, %ie_pc_hi20(sym_ie)
// * ld.[wd] $a0, $a0, %ie_pc_lo12(sym_ie)
//
// Simplicity, whether tlsdescToIe or tlsdescToLe, we always tend to convert the
// preceding instructions to NOPs, due to both forms of code sequence
// (corresponding to relocation combinations:
// R_LARCH_TLS_DESC_PC_HI20+R_LARCH_TLS_DESC_PC_LO12 and
// R_LARCH_TLS_DESC_PCREL20_S2) have same process.
//
// When relaxation enables, redundant NOPs can be removed.
void LoongArch::tlsdescToIe(uint8_t *loc, const Relocation &rel,
uint64_t val) const {
switch (rel.type) {
case R_LARCH_TLS_DESC_PC_HI20:
case R_LARCH_TLS_DESC_PC_LO12:
case R_LARCH_TLS_DESC_PCREL20_S2:
write32le(loc, insn(ANDI, R_ZERO, R_ZERO, 0)); // nop
break;
case R_LARCH_TLS_DESC_LD:
write32le(loc, insn(PCALAU12I, R_A0, 0, 0)); // pcalau12i $a0, %ie_pc_hi20
relocateNoSym(loc, R_LARCH_TLS_IE_PC_HI20, val);
break;
case R_LARCH_TLS_DESC_CALL:
write32le(loc, insn(ctx.arg.is64 ? LD_D : LD_W, R_A0, R_A0,
0)); // ld.[wd] $a0, $a0, %ie_pc_lo12
relocateNoSym(loc, R_LARCH_TLS_IE_PC_LO12, val);
break;
default:
llvm_unreachable("unsupported relocation for TLSDESC to IE");
}
}
// Convert TLSDESC GD/LD to LE.
// The code sequence obtained in the normal or medium code model is as follows:
// * lu12i.w $a0, %le_hi20(sym) # le_hi20 != 0, otherwise NOP
// * ori $a0, src, %le_lo12(sym) # le_hi20 != 0, src = $a0,
// # otherwise, src = $zero
// See the comment in tlsdescToIe for detailed information.
void LoongArch::tlsdescToLe(uint8_t *loc, const Relocation &rel,
uint64_t val) const {
assert(isInt<32>(val) &&
"val exceeds the range of medium code model in tlsdescToLe");
bool isUInt12 = isUInt<12>(val);
switch (rel.type) {
case R_LARCH_TLS_DESC_PC_HI20:
case R_LARCH_TLS_DESC_PC_LO12:
case R_LARCH_TLS_DESC_PCREL20_S2:
write32le(loc, insn(ANDI, R_ZERO, R_ZERO, 0)); // nop
break;
case R_LARCH_TLS_DESC_LD:
if (isUInt12)
write32le(loc, insn(ANDI, R_ZERO, R_ZERO, 0)); // nop
else
write32le(loc, insn(LU12I_W, R_A0, extractBits(val, 31, 12),
0)); // lu12i.w $a0, %le_hi20
break;
case R_LARCH_TLS_DESC_CALL:
if (isUInt12)
write32le(loc, insn(ORI, R_A0, R_ZERO, val)); // ori $a0, $zero, %le_lo12
else
write32le(loc,
insn(ORI, R_A0, R_A0, lo12(val))); // ori $a0, $a0, %le_lo12
break;
default:
llvm_unreachable("unsupported relocation for TLSDESC to LE");
}
}
// During TLSDESC GD_TO_IE, the converted code sequence always includes an
// instruction related to the Lo12 relocation (ld.[wd]). To obtain correct val
// in `getRelocTargetVA`, expr of this instruction should be adjusted to
// R_RELAX_TLS_GD_TO_IE_ABS, while expr of other instructions related to the
// Hi20 relocation (pcalau12i) should be adjusted to
// RE_LOONGARCH_RELAX_TLS_GD_TO_IE_PAGE_PC. Specifically, in the normal or
// medium code model, the instruction with relocation R_LARCH_TLS_DESC_CALL is
// the candidate of Lo12 relocation.
RelExpr LoongArch::adjustTlsExpr(RelType type, RelExpr expr) const {
if (expr == R_RELAX_TLS_GD_TO_IE) {
if (type != R_LARCH_TLS_DESC_CALL)
return RE_LOONGARCH_RELAX_TLS_GD_TO_IE_PAGE_PC;
return R_RELAX_TLS_GD_TO_IE_ABS;
}
return expr;
}
void LoongArch::relocateAlloc(InputSectionBase &sec, uint8_t *buf) const {
const unsigned bits = ctx.arg.is64 ? 64 : 32;
uint64_t secAddr = sec.getOutputSection()->addr;
@@ -1074,7 +1180,7 @@ void LoongArch::relocateAlloc(InputSectionBase &sec, uint8_t *buf) const {
// * i+2 -- R_LARCH_TLS_IE64_PC_LO20
// * i+3 -- R_LARCH_TLS_IE64_PC_HI12
isExtreme =
(i + 2 < size && relocs[i + 2].type == R_LARCH_TLS_IE64_PC_LO20);
i + 2 < size && relocs[i + 2].type == R_LARCH_TLS_IE64_PC_LO20;
}
if (isExtreme) {
rel.expr = getRelExpr(rel.type, *rel.sym, loc);
@@ -1088,6 +1194,47 @@ void LoongArch::relocateAlloc(InputSectionBase &sec, uint8_t *buf) const {
tlsIeToLe(loc, rel, val);
}
continue;
case RE_LOONGARCH_RELAX_TLS_GD_TO_IE_PAGE_PC:
if (rel.type == R_LARCH_TLS_DESC_PC_HI20) {
// LoongArch does not support TLSDESC GD/LD to LE/IE optimization in the
// extreme code model. In these cases, the relocs are as follows:
//
// * i -- R_LARCH_TLS_DESC_PC_HI20
// * i+1 -- R_LARCH_TLS_DESC_PC_LO12
// * i+2 -- R_LARCH_TLS_DESC64_PC_LO20
// * i+3 -- R_LARCH_TLS_DESC64_PC_HI12
isExtreme =
i + 2 < size && relocs[i + 2].type == R_LARCH_TLS_DESC64_PC_LO20;
}
[[fallthrough]];
case R_RELAX_TLS_GD_TO_IE_ABS:
if (isExtreme) {
if (rel.type == R_LARCH_TLS_DESC_CALL)
continue;
rel.expr = getRelExpr(rel.type, *rel.sym, loc);
val = SignExtend64(sec.getRelocTargetVA(ctx, rel, secAddr + rel.offset),
bits);
relocateNoSym(loc, rel.type, val);
} else {
tlsdescToIe(loc, rel, val);
}
continue;
case R_RELAX_TLS_GD_TO_LE:
if (rel.type == R_LARCH_TLS_DESC_PC_HI20) {
isExtreme =
i + 2 < size && relocs[i + 2].type == R_LARCH_TLS_DESC64_PC_LO20;
}
if (isExtreme) {
if (rel.type == R_LARCH_TLS_DESC_CALL)
continue;
rel.expr = getRelExpr(rel.type, *rel.sym, loc);
val = SignExtend64(sec.getRelocTargetVA(ctx, rel, secAddr + rel.offset),
bits);
relocateNoSym(loc, rel.type, val);
} else {
tlsdescToLe(loc, rel, val);
}
continue;
default:
break;
}

1
lld/ELF/InputSection.cpp
View File

@@ -830,6 +830,7 @@ uint64_t InputSectionBase::getRelocTargetVA(Ctx &ctx, const Relocation &r,
case R_GOTPLT_PC:
return r.sym->getGotPltVA(ctx) + a - p;
case RE_LOONGARCH_GOT_PAGE_PC:
case RE_LOONGARCH_RELAX_TLS_GD_TO_IE_PAGE_PC:
if (r.sym->hasFlag(NEEDS_TLSGD))
return getLoongArchPageDelta(ctx.in.got->getGlobalDynAddr(*r.sym) + a, p,
r.type);

38
lld/ELF/Relocations.cpp
View File

@@ -1340,22 +1340,10 @@ unsigned RelocationScanner::handleTlsRelocation(RelExpr expr, RelType type,
if (ctx.arg.emachine == EM_MIPS)
return handleMipsTlsRelocation(ctx, type, sym, *sec, offset, addend, expr);
// LoongArch does not yet implement transition from TLSDESC to LE/IE, so
// generate TLSDESC dynamic relocation for the dynamic linker to handle.
if (ctx.arg.emachine == EM_LOONGARCH &&
oneof<RE_LOONGARCH_TLSDESC_PAGE_PC, R_TLSDESC, R_TLSDESC_PC,
R_TLSDESC_CALL>(expr)) {
if (expr != R_TLSDESC_CALL) {
sym.setFlags(NEEDS_TLSDESC);
sec->addReloc({expr, type, offset, addend, &sym});
}
return 1;
}
bool isRISCV = ctx.arg.emachine == EM_RISCV;
if (oneof<RE_AARCH64_TLSDESC_PAGE, R_TLSDESC, R_TLSDESC_CALL, R_TLSDESC_PC,
R_TLSDESC_GOTPLT>(expr) &&
R_TLSDESC_GOTPLT, RE_LOONGARCH_TLSDESC_PAGE_PC>(expr) &&
ctx.arg.shared) {
// R_RISCV_TLSDESC_{LOAD_LO12,ADD_LO12_I,CALL} reference a label. Do not
// set NEEDS_TLSDESC on the label.
@@ -1369,10 +1357,14 @@ unsigned RelocationScanner::handleTlsRelocation(RelExpr expr, RelType type,
return 1;
}
// LoongArch supports IE to LE optimization in non-extreme code model.
// LoongArch supports IE to LE, DESC GD/LD to IE/LE optimizations in
// non-extreme code model.
bool execOptimizeInLoongArch =
ctx.arg.emachine == EM_LOONGARCH &&
(type == R_LARCH_TLS_IE_PC_HI20 || type == R_LARCH_TLS_IE_PC_LO12);
(type == R_LARCH_TLS_IE_PC_HI20 || type == R_LARCH_TLS_IE_PC_LO12 ||
type == R_LARCH_TLS_DESC_PC_HI20 || type == R_LARCH_TLS_DESC_PC_LO12 ||
type == R_LARCH_TLS_DESC_LD || type == R_LARCH_TLS_DESC_CALL ||
type == R_LARCH_TLS_DESC_PCREL20_S2);
// ARM, Hexagon, LoongArch and RISC-V do not support GD/LD to IE/LE
// optimizations.
@@ -1431,9 +1423,23 @@ unsigned RelocationScanner::handleTlsRelocation(RelExpr expr, RelType type,
return 1;
}
// LoongArch does not support transition from TLSDESC to LE/IE in the extreme
// code model, in which NEEDS_TLSDESC should set, rather than NEEDS_TLSGD. So
// we check independently.
if (ctx.arg.emachine == EM_LOONGARCH &&
oneof<RE_LOONGARCH_TLSDESC_PAGE_PC, R_TLSDESC, R_TLSDESC_PC,
R_TLSDESC_CALL>(expr) &&
!execOptimize) {
if (expr != R_TLSDESC_CALL) {
sym.setFlags(NEEDS_TLSDESC);
sec->addReloc({expr, type, offset, addend, &sym});
}
return 1;
}
if (oneof<RE_AARCH64_TLSDESC_PAGE, R_TLSDESC, R_TLSDESC_CALL, R_TLSDESC_PC,
R_TLSDESC_GOTPLT, R_TLSGD_GOT, R_TLSGD_GOTPLT, R_TLSGD_PC,
RE_LOONGARCH_TLSGD_PAGE_PC>(expr)) {
RE_LOONGARCH_TLSGD_PAGE_PC, RE_LOONGARCH_TLSDESC_PAGE_PC>(expr)) {
if (!execOptimize) {
sym.setFlags(NEEDS_TLSGD);
sec->addReloc({expr, type, offset, addend, &sym});

1
lld/ELF/Relocations.h
View File

@@ -131,6 +131,7 @@ enum RelExpr {
RE_LOONGARCH_GOT_PAGE_PC,
RE_LOONGARCH_TLSGD_PAGE_PC,
RE_LOONGARCH_TLSDESC_PAGE_PC,
RE_LOONGARCH_RELAX_TLS_GD_TO_IE_PAGE_PC,
};
// Architecture-neutral representation of relocation.

254
lld/test/ELF/loongarch-relax-tlsdesc.s
View File

@@ -9,19 +9,18 @@
# 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 from TLSDESC to IE/LE has not yet been implemented.
## Keep the dynamic relocations and hand them over to dynamic linker.
# 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
## FIXME: IE/LE relaxation have not yet been implemented, --relax/--no-relax obtain the same results.
## Transition from TLSDESC to IE/LE. Also check --emit-relocs.
# 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 2>&1 | FileCheck --check-prefix=LE64-RELA %s
# RUN: llvm-objdump --no-show-raw-insn -dr -h a.64.le | FileCheck %s --check-prefix=LE64
# 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 --relax -e 0 -z now a.64.o c.64.so -o a.64.ie
# RUN: ld.lld --relax -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 -h a.64.ie | FileCheck %s --check-prefix=IE64
# RUN: llvm-objdump --no-show-raw-insn -dr -h a.64.ie | FileCheck %s --check-prefix=IE64
# 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
@@ -71,172 +70,199 @@
# GD64-NEXT: jirl $ra, $ra, 0
# GD64-NEXT: add.d $a4, $a0, $tp
# LE64-RELA: .rela.dyn {
# 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: 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-RELA: could not find section '.got'
# 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
## a@tprel = 0x8
# LE64: 20158: nop
# LE64-NEXT: R_LARCH_TLS_DESC_PC_HI20 a
# LE64-NEXT: R_LARCH_RELAX *ABS*
# LE64-NEXT: nop
# LE64-NEXT: R_LARCH_TLS_DESC_PC_LO12 a
# LE64-NEXT: R_LARCH_RELAX *ABS*
# LE64-NEXT: nop
# LE64-NEXT: R_LARCH_TLS_DESC_LD a
# LE64-NEXT: R_LARCH_RELAX *ABS*
# LE64-NEXT: ori $a0, $zero, 8
# LE64-NEXT: R_LARCH_TLS_DESC_CALL a
# LE64-NEXT: R_LARCH_RELAX *ABS*
# 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
## b@tprel = 0x7ff
# LE64: 2016c: nop
# LE64-NEXT: R_LARCH_TLS_DESC_PC_HI20 b
# LE64-NEXT: R_LARCH_RELAX *ABS*
# LE64-NEXT: nop
# LE64-NEXT: R_LARCH_TLS_DESC_PC_LO12 b
# LE64-NEXT: nop
# LE64-NEXT: R_LARCH_TLS_DESC_LD b
# LE64-NEXT: ori $a0, $zero, 2047
# LE64-NEXT: R_LARCH_TLS_DESC_CALL b
# LE64-NEXT: add.d $a2, $a0, $tp
## &.got[c]-. = 0x30280+16 - 0x2024c: 0x10 pages, page offset 0x290
## c@tprel = 0x800
## Without R_LARCH_RELAX relocation. No relaxation.
# LE64: 2024c: pcalau12i $a0, 16
# LE64: 20180: nop
# LE64-NEXT: R_LARCH_TLS_DESC_PC_HI20 c
# LE64-NEXT: addi.d $t0, $zero, 0
# LE64-NEXT: addi.d $a0, $a0, 656
# LE64-NEXT: nop
# LE64-NEXT: R_LARCH_TLS_DESC_PC_LO12 c
# LE64-NEXT: addi.d $t0, $t0, 1
# LE64-NEXT: ld.d $ra, $a0, 0
# LE64-NEXT: nop
# LE64-NEXT: R_LARCH_TLS_DESC_LD c
# LE64-NEXT: addi.d $t0, $t0, 1
# LE64-NEXT: jirl $ra, $ra, 0
# LE64-NEXT: ori $a0, $zero, 2048
# LE64-NEXT: R_LARCH_TLS_DESC_CALL c
# 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
## d@tprel = 0x1000
# LE64: 201a0: nop
# LE64-NEXT: R_LARCH_TLS_DESC_PC_HI20 d
# LE64-NEXT: R_LARCH_RELAX *ABS*
# LE64-NEXT: nop
# LE64-NEXT: R_LARCH_TLS_DESC_PC_LO12 d
# LE64-NEXT: R_LARCH_RELAX *ABS*
# LE64-NEXT: lu12i.w $a0, 1
# LE64-NEXT: R_LARCH_TLS_DESC_LD d
# LE64-NEXT: ori $a0, $a0, 0
# LE64-NEXT: R_LARCH_TLS_DESC_CALL d
# 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
## a@tprel = 0x8
# LE64-NORELAX: 20158: nop
# LE64-NORELAX-NEXT: nop
# LE64-NORELAX-NEXT: nop
# LE64-NORELAX-NEXT: ori $a0, $zero, 8
# 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
## b@tprel = 0x7ff
# LE64-NORELAX: 2016c: nop
# LE64-NORELAX-NEXT: nop
# LE64-NORELAX-NEXT: nop
# LE64-NORELAX-NEXT: ori $a0, $zero, 2047
# LE64-NORELAX-NEXT: add.d $a2, $a0, $tp
## &.got[c]-. = 0x30288+16 - 0x20250: 0x10 pages, page offset 0x298
## c@tprel = 0x800
## Without R_LARCH_RELAX relocation. No relaxation.
# LE64-NORELAX: 20250: pcalau12i $a0, 16
# LE64-NORELAX: 20180: nop
# LE64-NORELAX-NEXT: addi.d $t0, $zero, 0
# LE64-NORELAX-NEXT: addi.d $a0, $a0, 664
# LE64-NORELAX-NEXT: nop
# LE64-NORELAX-NEXT: addi.d $t0, $t0, 1
# LE64-NORELAX-NEXT: ld.d $ra, $a0, 0
# LE64-NORELAX-NEXT: nop
# LE64-NORELAX-NEXT: addi.d $t0, $t0, 1
# LE64-NORELAX-NEXT: jirl $ra, $ra, 0
# LE64-NORELAX-NEXT: ori $a0, $zero, 2048
# 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
## d@tprel = 0x1000
# LE64-NORELAX: 201a0: nop
# LE64-NORELAX-NEXT: nop
# LE64-NORELAX-NEXT: lu12i.w $a0, 1
# LE64-NORELAX-NEXT: ori $a0, $a0, 0
# LE64-NORELAX-NEXT: add.d $a4, $a0, $tp
# IE64-RELA: .rela.dyn {
# 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: 0x30408 R_LARCH_TLS_TPREL64 c 0x0
# IE64-RELA-NEXT: 0x30410 R_LARCH_TLS_TPREL64 d 0x0
# IE64-RELA-NEXT: }
# IE64-RELA: Hex dump of section '.got':
# 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-RELA-NEXT: 0x00030408 00000000 00000000 00000000 00000000 .
# IE64: .got 00000040 0000000000030430
# IE64: .got 00000010 0000000000030408
## a and b are optimized to use LE. c and d are optimized to IE.
## &.got[a]-. = 0x30430 - 0x202f8 = 16462<<2
# IE64: 202f8: pcaddi $a0, 16462
# IE64-NEXT: ld.d $ra, $a0, 0
# IE64-NEXT: jirl $ra, $ra, 0
## a@tprel = 0x8
# IE64: 202c8: nop
# IE64-NEXT: R_LARCH_TLS_DESC_PC_HI20 a
# IE64-NEXT: R_LARCH_RELAX *ABS*
# IE64-NEXT: nop
# IE64-NEXT: R_LARCH_TLS_DESC_PC_LO12 a
# IE64-NEXT: R_LARCH_RELAX *ABS*
# IE64-NEXT: nop
# IE64-NEXT: R_LARCH_TLS_DESC_LD a
# IE64-NEXT: R_LARCH_RELAX *ABS*
# IE64-NEXT: ori $a0, $zero, 8
# IE64-NEXT: R_LARCH_TLS_DESC_CALL a
# IE64-NEXT: R_LARCH_RELAX *ABS*
# IE64-NEXT: add.d $a1, $a0, $tp
## &.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
## b@tprel = 0x7ff
# IE64: 202dc: nop
# IE64-NEXT: R_LARCH_TLS_DESC_PC_HI20 b
# IE64-NEXT: R_LARCH_RELAX *ABS*
# IE64-NEXT: nop
# IE64-NEXT: R_LARCH_TLS_DESC_PC_LO12 b
# IE64-NEXT: nop
# IE64-NEXT: R_LARCH_TLS_DESC_LD b
# IE64-NEXT: ori $a0, $zero, 2047
# IE64-NEXT: R_LARCH_TLS_DESC_CALL b
# IE64-NEXT: add.d $a2, $a0, $tp
## &.got[c]-. = 0x30430+16 - 0x2031c: 0x10 pages, page offset 0x440
## &.got[c]-. = 0x30408 - 0x20300: 0x10 pages, page offset 0x408
## Without R_LARCH_RELAX relocation. No relaxation.
# IE64: 2031c: pcalau12i $a0, 16
# IE64: 202f0: nop
# IE64-NEXT: R_LARCH_TLS_DESC_PC_HI20 c
# IE64-NEXT: addi.d $t0, $zero, 0
# IE64-NEXT: addi.d $a0, $a0, 1088
# IE64-NEXT: nop
# IE64-NEXT: R_LARCH_TLS_DESC_PC_LO12 c
# IE64-NEXT: addi.d $t0, $t0, 1
# IE64-NEXT: ld.d $ra, $a0, 0
# IE64-NEXT: pcalau12i $a0, 16
# IE64-NEXT: R_LARCH_TLS_DESC_LD c
# IE64-NEXT: addi.d $t0, $t0, 1
# IE64-NEXT: jirl $ra, $ra, 0
# IE64-NEXT: ld.d $a0, $a0, 1032
# IE64-NEXT: R_LARCH_TLS_DESC_CALL c
# 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
## &.got[d]-. = 0x30408+8 - 0x20318: 0x10 pages, page offset 0x410
# IE64: 20310: nop
# IE64-NEXT: R_LARCH_TLS_DESC_PC_HI20 d
# IE64-NEXT: R_LARCH_RELAX *ABS*
# IE64-NEXT: nop
# IE64-NEXT: R_LARCH_TLS_DESC_PC_LO12 d
# IE64-NEXT: R_LARCH_RELAX *ABS*
# IE64-NEXT: pcalau12i $a0, 16
# IE64-NEXT: R_LARCH_TLS_DESC_LD d
# IE64-NEXT: ld.d $a0, $a0, 1040
# IE64-NEXT: R_LARCH_TLS_DESC_CALL d
# IE64-NEXT: add.d $a4, $a0, $tp
# IE64-NORELAX: .got 00000040 0000000000030438
# IE64-NORELAX: .got 00000010 0000000000030408
## &.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
## a@tprel = 0x8
# IE64-NORELAX: 202c8: nop
# IE64-NORELAX-NEXT: nop
# IE64-NORELAX-NEXT: nop
# IE64-NORELAX-NEXT: ori $a0, $zero, 8
# 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
## b@tprel = 0x7ff
# IE64-NORELAX: 202dc: nop
# IE64-NORELAX-NEXT: nop
# IE64-NORELAX-NEXT: nop
# IE64-NORELAX-NEXT: ori $a0, $zero, 2047
# IE64-NORELAX-NEXT: add.d $a2, $a0, $tp
## &.got[c]-. = 0x30438+16 - 0x20320: 0x10 pages, page offset 0x448
## &.got[c]-. = 0x30408 - 0x20300: 0x10 pages, page offset 0x408
## Without R_LARCH_RELAX relocation. No relaxation.
# IE64-NORELAX: 20320: pcalau12i $a0, 16
# IE64-NORELAX: 202f0: nop
# IE64-NORELAX-NEXT: addi.d $t0, $zero, 0
# IE64-NORELAX-NEXT: addi.d $a0, $a0, 1096
# IE64-NORELAX-NEXT: nop
# IE64-NORELAX-NEXT: addi.d $t0, $t0, 1
# IE64-NORELAX-NEXT: ld.d $ra, $a0, 0
# IE64-NORELAX-NEXT: pcalau12i $a0, 16
# IE64-NORELAX-NEXT: addi.d $t0, $t0, 1
# IE64-NORELAX-NEXT: jirl $ra, $ra, 0
# IE64-NORELAX-NEXT: ld.d $a0, $a0, 1032
# 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
## &.got[d]-. = 0x30408+8 - 0x20318: 0x10 pages, page offset 0x410
# IE64-NORELAX: 20310: nop
# IE64-NORELAX-NEXT: nop
# IE64-NORELAX-NEXT: pcalau12i $a0, 16
# IE64-NORELAX-NEXT: ld.d $a0, $a0, 1040
# IE64-NORELAX-NEXT: add.d $a4, $a0, $tp
#--- a.s
la.tls.desc $a0, a
add.d $a1, $a0, $tp
# ADDI.D does not have R_LARCH_RELAX. No relaxation.
# ADDI.D does not have R_LARCH_RELAX. No relaxation when it is not optimized to IE/LE (--shared).
pcalau12i $a0, %desc_pc_hi20(b)
.reloc .-4, R_LARCH_RELAX, 0
addi.d $a0, $a0, %desc_pc_lo12(b)

59
lld/test/ELF/loongarch-tlsdesc-pcrel20-s2.s
View File

@@ -14,14 +14,14 @@
# RUN: ld.lld -shared -z now a.64.o c.64.o -o rel.64.so -z rel
# RUN: llvm-readobj -r -x .got rel.64.so | FileCheck --check-prefix=GD64-REL %s
## FIXME: The transition frome TLSDESC to IE/LE has not yet been implemented.
## Keep the dynamic relocations and hand them over to dynamic linker.
## Transition from TLSDESC to IE/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-readobj -r -x .got a.64.le 2>&1 | FileCheck --check-prefix=LE64-RELA %s
# RUN: llvm-objdump --no-show-raw-insn -d a.64.le | FileCheck --check-prefix=LE64 %s
# 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 -d a.64.ie | FileCheck --check-prefix=IE64 %s
## 32-bit code is mostly the same. We only test a few variants.
@@ -68,25 +68,46 @@
# GD64-NEXT: jirl $ra, $ra, 0
# GD64-NEXT: add.d $a3, $a0, $tp
# LE64-RELA: .rela.dyn {
# LE64-RELA-NEXT: 0x30240 R_LARCH_TLS_DESC64 - 0x8
# LE64-RELA-NEXT: 0x30250 R_LARCH_TLS_DESC64 - 0x800
# LE64-RELA-NEXT: 0x30260 R_LARCH_TLS_DESC64 - 0x7FF
# LE64-RELA-NEXT: }
# LE64-RELA: Hex dump of section '.got':
# LE64-RELA-NEXT: 0x00030240 00000000 00000000 00000000 00000000 .
# LE64-RELA-NEXT: 0x00030250 00000000 00000000 00000000 00000000 .
# LE64-RELA-NEXT: 0x00030260 00000000 00000000 00000000 00000000 .
# LE64-RELA: could not find section '.got'
# LE64-LABEL: <.text>:
## st_value(a) = 8
# LE64-NEXT: nop
# LE64-NEXT: nop
# LE64-NEXT: ori $a0, $zero, 8
# LE64-NEXT: add.d $a1, $a0, $tp
## st_value(b) = 2047
# LE64-NEXT: nop
# LE64-NEXT: nop
# LE64-NEXT: ori $a0, $zero, 2047
# LE64-NEXT: add.d $a2, $a0, $tp
## st_value(c) = 2048
# LE64-NEXT: nop
# LE64-NEXT: nop
# LE64-NEXT: ori $a0, $zero, 2048
# LE64-NEXT: add.d $a3, $a0, $tp
# IE64-RELA: .rela.dyn {
# IE64-RELA-NEXT: 0x303C8 R_LARCH_TLS_DESC64 - 0x8
# IE64-RELA-NEXT: 0x303E8 R_LARCH_TLS_DESC64 - 0x7FF
# IE64-RELA-NEXT: 0x303D8 R_LARCH_TLS_DESC64 c 0x0
# IE64-RELA-NEXT: 0x30398 R_LARCH_TLS_TPREL64 c 0x0
# IE64-RELA-NEXT: }
# IE64-RELA: Hex dump of section '.got':
# IE64-RELA-NEXT: 0x000303c8 00000000 00000000 00000000 00000000 .
# IE64-RELA-NEXT: 0x000303d8 00000000 00000000 00000000 00000000 .
# IE64-RELA-NEXT: 0x000303e8 00000000 00000000 00000000 00000000 .
# IE64-RELA-NEXT: 0x00030398 00000000 00000000 .
## a and b are optimized to use LE. c is optimized to IE.
# IE64-LABEL: <.text>:
# IE64-NEXT: nop
# IE64-NEXT: nop
# IE64-NEXT: ori $a0, $zero, 8
# IE64-NEXT: add.d $a1, $a0, $tp
# IE64-NEXT: nop
# IE64-NEXT: nop
# IE64-NEXT: ori $a0, $zero, 2047
# IE64-NEXT: add.d $a2, $a0, $tp
## &.got[c]-. = 0x30398 - 0x202ac: 0x10 pages, page offset 0x398
# IE64-NEXT: nop
# IE64-NEXT: 202ac: pcalau12i $a0, 16
# IE64-NEXT: ld.d $a0, $a0, 920
# IE64-NEXT: add.d $a3, $a0, $tp
# GD32-REL: .rel.dyn {
# GD32-REL-NEXT: 0x20264 R_LARCH_TLS_DESC32 -

65
lld/test/ELF/loongarch-tlsdesc.s
View File

@@ -14,14 +14,14 @@
# RUN: ld.lld -shared -z now a.64.o c.64.o -o rel.64.so -z rel
# RUN: llvm-readobj -r -x .got rel.64.so | FileCheck --check-prefix=GD64-REL %s
## FIXME: The transition frome TLSDESC to IE/LE has not yet been implemented.
## Keep the dynamic relocations and hand them over to dynamic linker.
## Transition from TLSDESC to IE/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-readobj -r -x .got a.64.le 2>&1 | FileCheck --check-prefix=LE64-RELA %s
# RUN: llvm-objdump --no-show-raw-insn -d a.64.le | FileCheck --check-prefix=LE64 %s
# 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 -d a.64.ie | FileCheck --check-prefix=IE64 %s
## 32-bit code is mostly the same. We only test a few variants.
@@ -71,25 +71,52 @@
# GD64-NEXT: jirl $ra, $ra, 0
# GD64-NEXT: add.d $a3, $a0, $tp
# LE64-RELA: .rela.dyn {
# LE64-RELA-NEXT: 0x30250 R_LARCH_TLS_DESC64 - 0x8
# LE64-RELA-NEXT: 0x30260 R_LARCH_TLS_DESC64 - 0x800
# LE64-RELA-NEXT: 0x30270 R_LARCH_TLS_DESC64 - 0x7FF
# LE64-RELA-NEXT: }
# LE64-RELA: Hex dump of section '.got':
# LE64-RELA-NEXT: 0x00030250 00000000 00000000 00000000 00000000 .
# LE64-RELA-NEXT: 0x00030260 00000000 00000000 00000000 00000000 .
# LE64-RELA-NEXT: 0x00030270 00000000 00000000 00000000 00000000 .
# LE64-RELA: could not find section '.got'
# LE64-LABEL: <.text>:
## st_value(a) = 8
# LE64-NEXT: nop
# LE64-NEXT: nop
# LE64-NEXT: nop
# LE64-NEXT: ori $a0, $zero, 8
# LE64-NEXT: add.d $a1, $a0, $tp
## st_value(b) = 2047
# LE64-NEXT: nop
# LE64-NEXT: nop
# LE64-NEXT: nop
# LE64-NEXT: ori $a0, $zero, 2047
# LE64-NEXT: add.d $a2, $a0, $tp
## st_value(c) = 2048
# LE64-NEXT: nop
# LE64-NEXT: nop
# LE64-NEXT: nop
# LE64-NEXT: ori $a0, $zero, 2048
# LE64-NEXT: add.d $a3, $a0, $tp
# IE64-RELA: .rela.dyn {
# IE64-RELA-NEXT: 0x303D8 R_LARCH_TLS_DESC64 - 0x8
# IE64-RELA-NEXT: 0x303F8 R_LARCH_TLS_DESC64 - 0x7FF
# IE64-RELA-NEXT: 0x303E8 R_LARCH_TLS_DESC64 c 0x0
# IE64-RELA-NEXT: 0x303A8 R_LARCH_TLS_TPREL64 c 0x0
# IE64-RELA-NEXT: }
# IE64-RELA: Hex dump of section '.got':
# IE64-RELA-NEXT: 0x000303d8 00000000 00000000 00000000 00000000 .
# IE64-RELA-NEXT: 0x000303e8 00000000 00000000 00000000 00000000 .
# IE64-RELA-NEXT: 0x000303f8 00000000 00000000 00000000 00000000 .
# IE64-RELA-NEXT: 0x000303a8 00000000 00000000 .
## a and b are optimized to use LE. c is optimized to IE.
# IE64-LABEL: <.text>:
# IE64-NEXT: nop
# IE64-NEXT: nop
# IE64-NEXT: nop
# IE64-NEXT: ori $a0, $zero, 8
# IE64-NEXT: add.d $a1, $a0, $tp
# IE64-NEXT: nop
# IE64-NEXT: nop
# IE64-NEXT: nop
# IE64-NEXT: ori $a0, $zero, 2047
# IE64-NEXT: add.d $a2, $a0, $tp
## &.got[c]-. = 0x303a8 - 0x202b8: 0x10 pages, page offset 0x3a8
# IE64-NEXT: nop
# IE64-NEXT: nop
# IE64-NEXT: 202b8: pcalau12i $a0, 16
# IE64-NEXT: ld.d $a0, $a0, 936
# IE64-NEXT: add.d $a3, $a0, $tp
# GD32-REL: .rel.dyn {
# GD32-REL-NEXT: 0x20270 R_LARCH_TLS_DESC32 -