[PPC64] toc-indirect to toc-relative relaxation
This is based on D54720 by Sean Fertile.
When accessing a global symbol which is not defined in the translation unit,
compilers will generate instructions that load the address from the toc entry.
If the symbol is defined, non-preemptable, and addressable with a 32-bit
signed offset from the toc pointer, the address can be computed
directly. e.g.
addis 3, 2, .LC0@toc@ha # R_PPC64_TOC16_HA
ld 3, .LC0@toc@l(3) # R_PPC64_TOC16_LO_DS, load the address from a .toc entry
ld/lwa 3, 0(3) # load the value from the address
.section .toc,"aw",@progbits
.LC0: .tc var[TC],var
can be relaxed to
addis 3,2,var@toc@ha # this may be relaxed to a nop,
addi 3,3,var@toc@l # then this becomes addi 3,2,var@toc
ld/lwa 3, 0(3) # load the value from the address
We can delete the test ppc64-got-indirect.s as its purpose is covered by
newly added ppc64-toc-relax.s and ppc64-toc-relax-constants.s
Reviewed By: ruiu, sfertile
Differential Revision: https://reviews.llvm.org/D60958
llvm-svn: 360112
2019-05-07 12:26:05 +08:00
|
|
|
# REQUIRES: ppc
|
|
|
|
|
# RUN: llvm-mc -filetype=obj -triple=powerpc64le-unknown-linux %s -o %t.o
|
|
|
|
|
# RUN: ld.lld %t.o -o %t
|
|
|
|
|
# RUN: llvm-readelf -S %t | FileCheck --check-prefixes=SECTIONS %s
|
|
|
|
|
# RUN: llvm-readelf -x .toc %t | FileCheck --check-prefixes=HEX-LE %s
|
|
|
|
|
# RUN: llvm-objdump -d %t | FileCheck --check-prefixes=CHECK %s
|
|
|
|
|
|
|
|
|
|
# RUN: llvm-mc -filetype=obj -triple=powerpc64-unknown-linux %s -o %t.o
|
|
|
|
|
# RUN: ld.lld %t.o -o %t
|
|
|
|
|
# RUN: llvm-readelf -S %t | FileCheck --check-prefixes=SECTIONS %s
|
|
|
|
|
# RUN: llvm-readelf -x .toc %t | FileCheck --check-prefixes=HEX-BE %s
|
|
|
|
|
# RUN: llvm-objdump -d %t | FileCheck --check-prefixes=CHECK %s
|
|
|
|
|
|
|
|
|
|
# .LJT is a local symbol (non-preemptable).
|
|
|
|
|
# Test we can perform the toc-indirect to toc-relative relaxation.
|
|
|
|
|
|
|
|
|
|
# SECTIONS: .rodata PROGBITS 00000000100001c8
|
|
|
|
|
|
|
|
|
|
# HEX-LE: section '.toc':
|
[ELF][PPC] Allow PT_LOAD to have overlapping p_offset ranges
This change affects the non-linker script case (precisely, when the
`SECTIONS` command is not used). It deletes 3 alignments at PT_LOAD
boundaries for the default case: the size of a powerpc64 binary can be
decreased by at most 192kb. The technique can be ported to other
targets.
Let me demonstrate the idea with a maxPageSize=65536 example:
When assigning the address to the first output section of a new PT_LOAD,
if the end p_vaddr of the previous PT_LOAD is 0x10020, we advance to
the next multiple of maxPageSize: 0x20000. The new PT_LOAD will thus
have p_vaddr=0x20000. Because p_offset and p_vaddr are congruent modulo
maxPageSize, p_offset will be 0x20000, leaving a p_offset gap [0x10020,
0x20000) in the output.
Alternatively, if we advance to 0x20020, the new PT_LOAD will have
p_vaddr=0x20020. We can pick either 0x10020 or 0x20020 for p_offset!
Obviously 0x10020 is the choice because it leaves no gap. At runtime,
p_vaddr will be rounded down by pagesize (65536 if
pagesize=maxPageSize). This PT_LOAD will load additional initial
contents from p_offset ranges [0x10000,0x10020), which will also be
loaded by the previous PT_LOAD. This is fine if -z noseparate-code is in
effect or if we are not transiting between executable and non-executable
segments.
ld.bfd -z noseparate-code leverages this technique to keep output small.
This patch implements the technique in lld, which is mostly effective on
targets with large defaultMaxPageSize (AArch64/MIPS/PPC: 65536). The 3
removed alignments can save almost 3*65536 bytes.
Two places that rely on p_vaddr%pagesize = 0 have to be updated.
1) We used to round p_memsz(PT_GNU_RELRO) up to commonPageSize (defaults
to 4096 on all targets). Now p_vaddr%commonPageSize may be non-zero.
The updated formula takes account of that factor.
2) Our TP offsets formulae are only correct if p_vaddr%p_align = 0.
Fix them. See the updated comments in InputSection.cpp for details.
On targets that we enable the technique (only PPC64 now),
we can potentially make `p_vaddr(PT_TLS)%p_align(PT_TLS) != 0`
if `sh_addralign(.tdata) < sh_addralign(.tbss)`
This exposes many problems in ld.so implementations, especially the
offsets of dynamic TLS blocks. Known issues:
FreeBSD 13.0-CURRENT rtld-elf (i386/amd64/powerpc/arm64)
glibc (HEAD) i386 and x86_64 https://sourceware.org/bugzilla/show_bug.cgi?id=24606
musl<=1.1.22 on TLS Variant I architectures (aarch64/powerpc64/...)
So, force p_vaddr%p_align = 0 by rounding dot up to p_align(PT_TLS).
The technique will be enabled (with updated tests) for other targets in
subsequent patches.
Reviewed By: ruiu
Differential Revision: https://reviews.llvm.org/D64906
llvm-svn: 369343
2019-08-20 16:34:25 +08:00
|
|
|
# HEX-LE-NEXT: 10020228 c8010010 00000000
|
[PPC64] toc-indirect to toc-relative relaxation
This is based on D54720 by Sean Fertile.
When accessing a global symbol which is not defined in the translation unit,
compilers will generate instructions that load the address from the toc entry.
If the symbol is defined, non-preemptable, and addressable with a 32-bit
signed offset from the toc pointer, the address can be computed
directly. e.g.
addis 3, 2, .LC0@toc@ha # R_PPC64_TOC16_HA
ld 3, .LC0@toc@l(3) # R_PPC64_TOC16_LO_DS, load the address from a .toc entry
ld/lwa 3, 0(3) # load the value from the address
.section .toc,"aw",@progbits
.LC0: .tc var[TC],var
can be relaxed to
addis 3,2,var@toc@ha # this may be relaxed to a nop,
addi 3,3,var@toc@l # then this becomes addi 3,2,var@toc
ld/lwa 3, 0(3) # load the value from the address
We can delete the test ppc64-got-indirect.s as its purpose is covered by
newly added ppc64-toc-relax.s and ppc64-toc-relax-constants.s
Reviewed By: ruiu, sfertile
Differential Revision: https://reviews.llvm.org/D60958
llvm-svn: 360112
2019-05-07 12:26:05 +08:00
|
|
|
|
|
|
|
|
# HEX-BE: section '.toc':
|
[ELF][PPC] Allow PT_LOAD to have overlapping p_offset ranges
This change affects the non-linker script case (precisely, when the
`SECTIONS` command is not used). It deletes 3 alignments at PT_LOAD
boundaries for the default case: the size of a powerpc64 binary can be
decreased by at most 192kb. The technique can be ported to other
targets.
Let me demonstrate the idea with a maxPageSize=65536 example:
When assigning the address to the first output section of a new PT_LOAD,
if the end p_vaddr of the previous PT_LOAD is 0x10020, we advance to
the next multiple of maxPageSize: 0x20000. The new PT_LOAD will thus
have p_vaddr=0x20000. Because p_offset and p_vaddr are congruent modulo
maxPageSize, p_offset will be 0x20000, leaving a p_offset gap [0x10020,
0x20000) in the output.
Alternatively, if we advance to 0x20020, the new PT_LOAD will have
p_vaddr=0x20020. We can pick either 0x10020 or 0x20020 for p_offset!
Obviously 0x10020 is the choice because it leaves no gap. At runtime,
p_vaddr will be rounded down by pagesize (65536 if
pagesize=maxPageSize). This PT_LOAD will load additional initial
contents from p_offset ranges [0x10000,0x10020), which will also be
loaded by the previous PT_LOAD. This is fine if -z noseparate-code is in
effect or if we are not transiting between executable and non-executable
segments.
ld.bfd -z noseparate-code leverages this technique to keep output small.
This patch implements the technique in lld, which is mostly effective on
targets with large defaultMaxPageSize (AArch64/MIPS/PPC: 65536). The 3
removed alignments can save almost 3*65536 bytes.
Two places that rely on p_vaddr%pagesize = 0 have to be updated.
1) We used to round p_memsz(PT_GNU_RELRO) up to commonPageSize (defaults
to 4096 on all targets). Now p_vaddr%commonPageSize may be non-zero.
The updated formula takes account of that factor.
2) Our TP offsets formulae are only correct if p_vaddr%p_align = 0.
Fix them. See the updated comments in InputSection.cpp for details.
On targets that we enable the technique (only PPC64 now),
we can potentially make `p_vaddr(PT_TLS)%p_align(PT_TLS) != 0`
if `sh_addralign(.tdata) < sh_addralign(.tbss)`
This exposes many problems in ld.so implementations, especially the
offsets of dynamic TLS blocks. Known issues:
FreeBSD 13.0-CURRENT rtld-elf (i386/amd64/powerpc/arm64)
glibc (HEAD) i386 and x86_64 https://sourceware.org/bugzilla/show_bug.cgi?id=24606
musl<=1.1.22 on TLS Variant I architectures (aarch64/powerpc64/...)
So, force p_vaddr%p_align = 0 by rounding dot up to p_align(PT_TLS).
The technique will be enabled (with updated tests) for other targets in
subsequent patches.
Reviewed By: ruiu
Differential Revision: https://reviews.llvm.org/D64906
llvm-svn: 369343
2019-08-20 16:34:25 +08:00
|
|
|
# HEX-BE-NEXT: 10020228 00000000 100001c8
|
[PPC64] toc-indirect to toc-relative relaxation
This is based on D54720 by Sean Fertile.
When accessing a global symbol which is not defined in the translation unit,
compilers will generate instructions that load the address from the toc entry.
If the symbol is defined, non-preemptable, and addressable with a 32-bit
signed offset from the toc pointer, the address can be computed
directly. e.g.
addis 3, 2, .LC0@toc@ha # R_PPC64_TOC16_HA
ld 3, .LC0@toc@l(3) # R_PPC64_TOC16_LO_DS, load the address from a .toc entry
ld/lwa 3, 0(3) # load the value from the address
.section .toc,"aw",@progbits
.LC0: .tc var[TC],var
can be relaxed to
addis 3,2,var@toc@ha # this may be relaxed to a nop,
addi 3,3,var@toc@l # then this becomes addi 3,2,var@toc
ld/lwa 3, 0(3) # load the value from the address
We can delete the test ppc64-got-indirect.s as its purpose is covered by
newly added ppc64-toc-relax.s and ppc64-toc-relax-constants.s
Reviewed By: ruiu, sfertile
Differential Revision: https://reviews.llvm.org/D60958
llvm-svn: 360112
2019-05-07 12:26:05 +08:00
|
|
|
|
|
|
|
|
# CHECK-LABEL: _start
|
|
|
|
|
# CHECK: clrldi 3, 3, 62
|
[ELF][PPC] Allow PT_LOAD to have overlapping p_offset ranges
This change affects the non-linker script case (precisely, when the
`SECTIONS` command is not used). It deletes 3 alignments at PT_LOAD
boundaries for the default case: the size of a powerpc64 binary can be
decreased by at most 192kb. The technique can be ported to other
targets.
Let me demonstrate the idea with a maxPageSize=65536 example:
When assigning the address to the first output section of a new PT_LOAD,
if the end p_vaddr of the previous PT_LOAD is 0x10020, we advance to
the next multiple of maxPageSize: 0x20000. The new PT_LOAD will thus
have p_vaddr=0x20000. Because p_offset and p_vaddr are congruent modulo
maxPageSize, p_offset will be 0x20000, leaving a p_offset gap [0x10020,
0x20000) in the output.
Alternatively, if we advance to 0x20020, the new PT_LOAD will have
p_vaddr=0x20020. We can pick either 0x10020 or 0x20020 for p_offset!
Obviously 0x10020 is the choice because it leaves no gap. At runtime,
p_vaddr will be rounded down by pagesize (65536 if
pagesize=maxPageSize). This PT_LOAD will load additional initial
contents from p_offset ranges [0x10000,0x10020), which will also be
loaded by the previous PT_LOAD. This is fine if -z noseparate-code is in
effect or if we are not transiting between executable and non-executable
segments.
ld.bfd -z noseparate-code leverages this technique to keep output small.
This patch implements the technique in lld, which is mostly effective on
targets with large defaultMaxPageSize (AArch64/MIPS/PPC: 65536). The 3
removed alignments can save almost 3*65536 bytes.
Two places that rely on p_vaddr%pagesize = 0 have to be updated.
1) We used to round p_memsz(PT_GNU_RELRO) up to commonPageSize (defaults
to 4096 on all targets). Now p_vaddr%commonPageSize may be non-zero.
The updated formula takes account of that factor.
2) Our TP offsets formulae are only correct if p_vaddr%p_align = 0.
Fix them. See the updated comments in InputSection.cpp for details.
On targets that we enable the technique (only PPC64 now),
we can potentially make `p_vaddr(PT_TLS)%p_align(PT_TLS) != 0`
if `sh_addralign(.tdata) < sh_addralign(.tbss)`
This exposes many problems in ld.so implementations, especially the
offsets of dynamic TLS blocks. Known issues:
FreeBSD 13.0-CURRENT rtld-elf (i386/amd64/powerpc/arm64)
glibc (HEAD) i386 and x86_64 https://sourceware.org/bugzilla/show_bug.cgi?id=24606
musl<=1.1.22 on TLS Variant I architectures (aarch64/powerpc64/...)
So, force p_vaddr%p_align = 0 by rounding dot up to p_align(PT_TLS).
The technique will be enabled (with updated tests) for other targets in
subsequent patches.
Reviewed By: ruiu
Differential Revision: https://reviews.llvm.org/D64906
llvm-svn: 369343
2019-08-20 16:34:25 +08:00
|
|
|
# CHECK-NEXT: addis 4, 2, -3
|
|
|
|
|
# CHECK-NEXT: addi 4, 4, 32680
|
[PPC64] toc-indirect to toc-relative relaxation
This is based on D54720 by Sean Fertile.
When accessing a global symbol which is not defined in the translation unit,
compilers will generate instructions that load the address from the toc entry.
If the symbol is defined, non-preemptable, and addressable with a 32-bit
signed offset from the toc pointer, the address can be computed
directly. e.g.
addis 3, 2, .LC0@toc@ha # R_PPC64_TOC16_HA
ld 3, .LC0@toc@l(3) # R_PPC64_TOC16_LO_DS, load the address from a .toc entry
ld/lwa 3, 0(3) # load the value from the address
.section .toc,"aw",@progbits
.LC0: .tc var[TC],var
can be relaxed to
addis 3,2,var@toc@ha # this may be relaxed to a nop,
addi 3,3,var@toc@l # then this becomes addi 3,2,var@toc
ld/lwa 3, 0(3) # load the value from the address
We can delete the test ppc64-got-indirect.s as its purpose is covered by
newly added ppc64-toc-relax.s and ppc64-toc-relax-constants.s
Reviewed By: ruiu, sfertile
Differential Revision: https://reviews.llvm.org/D60958
llvm-svn: 360112
2019-05-07 12:26:05 +08:00
|
|
|
# CHECK-NEXT: sldi 3, 3, 2
|
|
|
|
|
|
|
|
|
|
.text
|
|
|
|
|
.global _start
|
|
|
|
|
.type _start, @function
|
|
|
|
|
_start:
|
|
|
|
|
.Lstart_gep:
|
|
|
|
|
addis 2, 12, .TOC.-.Lstart_gep@ha
|
|
|
|
|
addi 2, 2, .TOC.-.Lstart_gep@l
|
|
|
|
|
.Lstart_lep:
|
|
|
|
|
.localentry _start, .Lstart_lep-.Lstart_gep
|
|
|
|
|
rldicl 3, 3, 0, 62
|
|
|
|
|
addis 4, 2, .LJTI_TE@toc@ha
|
|
|
|
|
ld 4, .LJTI_TE@toc@l(4)
|
|
|
|
|
sldi 3, 3, 2
|
|
|
|
|
lwax 3, 3, 4
|
|
|
|
|
add 3, 3, 4
|
|
|
|
|
mtctr 3
|
|
|
|
|
bctr
|
|
|
|
|
|
|
|
|
|
.LBB1:
|
|
|
|
|
li 3, 0
|
|
|
|
|
blr
|
|
|
|
|
.LBB2:
|
|
|
|
|
li 3, 10
|
|
|
|
|
blr
|
|
|
|
|
.LBB3:
|
|
|
|
|
li 3, 55
|
|
|
|
|
blr
|
|
|
|
|
.LBB4:
|
|
|
|
|
li 3, 255
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
.section .rodata,"a",@progbits
|
|
|
|
|
.p2align 2
|
|
|
|
|
.LJT:
|
|
|
|
|
.long .LBB1-.LJT
|
|
|
|
|
.long .LBB2-.LJT
|
|
|
|
|
.long .LBB3-.LJT
|
|
|
|
|
.long .LBB4-.LJT
|
|
|
|
|
|
|
|
|
|
.section .toc,"aw",@progbits
|
|
|
|
|
# TOC entry for the jumptable address.
|
|
|
|
|
.LJTI_TE:
|
|
|
|
|
.tc .LJT[TC],.LJT
|