maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/lld/test/ELF/ppc64-toc-addis-nop.s

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

276 lines
6.9 KiB
ArmAsm
Raw Normal View History

[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# REQUIRES: ppc
# RUN: llvm-mc -filetype=obj -triple=powerpc64le-unknown-linux %s -o %t.o
[PPC][PPC64] Improve some llvm-objdump -d -D tests Various improvement: Some offsets in disassembly are incorrect after several layout adjustment. Fix them. llvm-objdump -D should not be used. -D dumps unrelated non-text sections. Replace them with llvm-objdump -d, llvm-readelf -x, etc Many llvm-objdump -d tests use {{.*}} . Add the option --no-show-raw-insn to avoid check hex bytes. ppc64-long-branch.s does not need a shared object. Delete it. Make ppc64-ifunc.s check 2 ifuncs. Reviewers: ruiu, espindola Subscribers: emaste, nemanjai, arichardson, kbarton, jsji, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D60998 llvm-svn: 358975
2019-04-23 19:47:28 +08:00
# RUN: llvm-readelf -r %t.o | FileCheck --check-prefix=InputRelocs %s
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# RUN: llvm-mc -filetype=obj -triple=powerpc64le-unknown-linux %p/Inputs/shared-ppc64.s -o %t2.o
[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
# RUN: ld.lld -shared -soname=t2.so %t2.o -o %t2.so
## Place all sections in the same segment so that .text and .TOC. are on the same page.
# RUN: echo 'PHDRS { all PT_LOAD; }' > %t.script
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
#
[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
# RUN: ld.lld %t2.so %t.o -T %t.script -o %t
[PPC][PPC64] Improve some llvm-objdump -d -D tests Various improvement: Some offsets in disassembly are incorrect after several layout adjustment. Fix them. llvm-objdump -D should not be used. -D dumps unrelated non-text sections. Replace them with llvm-objdump -d, llvm-readelf -x, etc Many llvm-objdump -d tests use {{.*}} . Add the option --no-show-raw-insn to avoid check hex bytes. ppc64-long-branch.s does not need a shared object. Delete it. Make ppc64-ifunc.s check 2 ifuncs. Reviewers: ruiu, espindola Subscribers: emaste, nemanjai, arichardson, kbarton, jsji, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D60998 llvm-svn: 358975
2019-04-23 19:47:28 +08:00
# RUN: llvm-objdump -d --no-show-raw-insn %t | FileCheck --check-prefix=Dis %s
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
#
[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
# RUN: ld.lld %t2.so %t.o -T %t.script --no-toc-optimize -o %t
[PPC][PPC64] Improve some llvm-objdump -d -D tests Various improvement: Some offsets in disassembly are incorrect after several layout adjustment. Fix them. llvm-objdump -D should not be used. -D dumps unrelated non-text sections. Replace them with llvm-objdump -d, llvm-readelf -x, etc Many llvm-objdump -d tests use {{.*}} . Add the option --no-show-raw-insn to avoid check hex bytes. ppc64-long-branch.s does not need a shared object. Delete it. Make ppc64-ifunc.s check 2 ifuncs. Reviewers: ruiu, espindola Subscribers: emaste, nemanjai, arichardson, kbarton, jsji, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D60998 llvm-svn: 358975
2019-04-23 19:47:28 +08:00
# RUN: llvm-objdump -d --no-show-raw-insn %t | FileCheck --check-prefix=NoOpt %s
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# InputRelocs: Relocation section '.rela.text'
# InputRelocs: R_PPC64_TOC16_HA
# InputRelocs: R_PPC64_TOC16_LO
# InputRelocs: R_PPC64_TOC16_LO_DS
.text
.abiversion 2
.global bytes
.p2align 4
.type bytes,@function
bytes:
.Lbytes_gep:
addis 2, 12, .TOC.-.Lbytes_gep@ha
addi 2, 2, .TOC.-.Lbytes_gep@l
.Lbytes_lep:
.localentry bytes, .Lbytes_lep-.Lbytes_gep
addis 3, 2, byteLd@toc@ha
lbz 3, byteLd@toc@l(3)
addis 4, 2, byteSt@toc@ha
stb 3, byteSt@toc@l(4)
blr
[llvm-objdump] -d: print `00000000 <foo>:` instead of `00000000 foo:` The new behavior matches GNU objdump. A pair of angle brackets makes tests slightly easier. `.foo:` is not unique and thus cannot be used in a `CHECK-LABEL:` directive. Without `-LABEL`, the CHECK line can match the `Disassembly of section` line and causes the next `CHECK-NEXT:` to fail. ``` Disassembly of section .foo: 0000000000001634 .foo: ``` Bdragon: <> has metalinguistic connotation. it just "feels right" Reviewed By: rupprecht Differential Revision: https://reviews.llvm.org/D75713
2020-03-06 06:18:38 +08:00
# Dis-LABEL: <bytes>:
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: addis
# Dis-NEXT: addi
# Dis-NEXT: nop
[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
# Dis-NEXT: lbz 3, -32752(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: stb 3, -32751(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: blr
[llvm-objdump] -d: print `00000000 <foo>:` instead of `00000000 foo:` The new behavior matches GNU objdump. A pair of angle brackets makes tests slightly easier. `.foo:` is not unique and thus cannot be used in a `CHECK-LABEL:` directive. Without `-LABEL`, the CHECK line can match the `Disassembly of section` line and causes the next `CHECK-NEXT:` to fail. ``` Disassembly of section .foo: 0000000000001634 .foo: ``` Bdragon: <> has metalinguistic connotation. it just "feels right" Reviewed By: rupprecht Differential Revision: https://reviews.llvm.org/D75713
2020-03-06 06:18:38 +08:00
# NoOpt-LABEL: <bytes>:
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis
# NoOpt-NEXT: addi
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: lbz 3, -32752(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 4, 2, 0
[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
# NoOpt-NEXT: stb 3, -32751(4)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: blr
.global halfs
.p2align 4
.type halfs,@function
halfs:
.Lhalfs_gep:
addis 2, 12, .TOC.-.Lhalfs_gep@ha
addi 2, 2, .TOC.-.Lhalfs_gep@l
.Lhalfs_lep:
.localentry halfs, .Lhalfs_lep-.Lhalfs_gep
addis 3, 2, halfLd@toc@ha
lhz 3, halfLd@toc@l(3)
addis 4, 2, halfLd@toc@ha
lha 4, halfLd@toc@l(4)
addis 5, 2, halfSt@toc@ha
sth 4, halfSt@toc@l(5)
blr
[llvm-objdump] -d: print `00000000 <foo>:` instead of `00000000 foo:` The new behavior matches GNU objdump. A pair of angle brackets makes tests slightly easier. `.foo:` is not unique and thus cannot be used in a `CHECK-LABEL:` directive. Without `-LABEL`, the CHECK line can match the `Disassembly of section` line and causes the next `CHECK-NEXT:` to fail. ``` Disassembly of section .foo: 0000000000001634 .foo: ``` Bdragon: <> has metalinguistic connotation. it just "feels right" Reviewed By: rupprecht Differential Revision: https://reviews.llvm.org/D75713
2020-03-06 06:18:38 +08:00
# Dis-LABEL: <halfs>:
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: addis
# Dis-NEXT: addi
# Dis-NEXT: nop
[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
# Dis-NEXT: lhz 3, -32750(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: lha 4, -32750(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: sth 4, -32748(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: blr
[llvm-objdump] -d: print `00000000 <foo>:` instead of `00000000 foo:` The new behavior matches GNU objdump. A pair of angle brackets makes tests slightly easier. `.foo:` is not unique and thus cannot be used in a `CHECK-LABEL:` directive. Without `-LABEL`, the CHECK line can match the `Disassembly of section` line and causes the next `CHECK-NEXT:` to fail. ``` Disassembly of section .foo: 0000000000001634 .foo: ``` Bdragon: <> has metalinguistic connotation. it just "feels right" Reviewed By: rupprecht Differential Revision: https://reviews.llvm.org/D75713
2020-03-06 06:18:38 +08:00
# NoOpt-LABEL: <halfs>:
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis
# NoOpt-NEXT: addi
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: lhz 3, -32750(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 4, 2, 0
[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
# NoOpt-NEXT: lha 4, -32750(4)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 5, 2, 0
[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
# NoOpt-NEXT: sth 4, -32748(5)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: blr
.global words
.p2align 4
.type words,@function
words:
.Lwords_gep:
addis 2, 12, .TOC.-.Lwords_gep@ha
addi 2, 2, .TOC.-.Lwords_gep@l
.Lwords_lep:
.localentry words, .Lwords_lep-.Lwords_gep
addis 3, 2, wordLd@toc@ha
lwz 3, wordLd@toc@l(3)
addis 4, 2, wordLd@toc@ha
lwa 4, wordLd@toc@l(4)
addis 5, 2, wordSt@toc@ha
stw 4, wordSt@toc@l(5)
blr
# Dis-LABEL: words
# Dis-NEXT: addis
# Dis-NEXT: addi
# Dis-NEXT: nop
[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
# Dis-NEXT: lwz 3, -32744(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: lwa 4, -32744(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: stw 4, -32740(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: blr
# NoOpt-LABEL: words
# NoOpt-NEXT: addis
# NoOpt-NEXT: addi
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: lwz 3, -32744(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 4, 2, 0
[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
# NoOpt-NEXT: lwa 4, -32744(4)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 5, 2, 0
[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
# NoOpt-NEXT: stw 4, -32740(5)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: blr
.global doublewords
.p2align 4
.type doublewords,@function
doublewords:
.Ldoublewords_gep:
addis 2, 12, .TOC.-.Ldoublewords_gep@ha
addi 2, 2, .TOC.-.Ldoublewords_gep@l
.Ldoublewords_lep:
.localentry doublewords, .Ldoublewords_lep-.Ldoublewords_gep
addis 3, 2, dwordLd@toc@ha
ld 3, dwordLd@toc@l(3)
addis 4, 2, dwordSt@toc@ha
std 3, dwordSt@toc@l(4)
blr
# Dis-LABEL: doublewords
# Dis-NEXT: addis
# Dis-NEXT: addi
# Dis-NEXT: nop
[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
# Dis-NEXT: ld 3, -32736(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: std 3, -32728(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: blr
# NoOpt-LABEL: doublewords
# NoOpt-NEXT: addis
# NoOpt-NEXT: addi
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: ld 3, -32736(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 4, 2, 0
[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
# NoOpt-NEXT: std 3, -32728(4)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: blr
.global vec_dq
.p2align 4
.type vec_dq,@function
vec_dq:
.Lvec_dq_gep:
addis 2, 12, .TOC.-.Lvec_dq_gep@ha
addi 2, 2, .TOC.-.Lvec_dq_gep@l
.Lvec_dq_lep:
.localentry vec_dq, .Lvec_dq_lep-.Lvec_dq_gep
addis 3, 2, vecLd@toc@ha
lxv 3, vecLd@toc@l(3)
addis 3, 2, vecSt@toc@ha
stxv 3, vecSt@toc@l(3)
blr
[llvm-objdump] -d: print `00000000 <foo>:` instead of `00000000 foo:` The new behavior matches GNU objdump. A pair of angle brackets makes tests slightly easier. `.foo:` is not unique and thus cannot be used in a `CHECK-LABEL:` directive. Without `-LABEL`, the CHECK line can match the `Disassembly of section` line and causes the next `CHECK-NEXT:` to fail. ``` Disassembly of section .foo: 0000000000001634 .foo: ``` Bdragon: <> has metalinguistic connotation. it just "feels right" Reviewed By: rupprecht Differential Revision: https://reviews.llvm.org/D75713
2020-03-06 06:18:38 +08:00
# Dis-LABEL: <vec_dq>:
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: addis
# Dis-NEXT: addi
# Dis-NEXT: nop
[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
# Dis-NEXT: lxv 3, -32720(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: stxv 3, -32704(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: blr
[llvm-objdump] -d: print `00000000 <foo>:` instead of `00000000 foo:` The new behavior matches GNU objdump. A pair of angle brackets makes tests slightly easier. `.foo:` is not unique and thus cannot be used in a `CHECK-LABEL:` directive. Without `-LABEL`, the CHECK line can match the `Disassembly of section` line and causes the next `CHECK-NEXT:` to fail. ``` Disassembly of section .foo: 0000000000001634 .foo: ``` Bdragon: <> has metalinguistic connotation. it just "feels right" Reviewed By: rupprecht Differential Revision: https://reviews.llvm.org/D75713
2020-03-06 06:18:38 +08:00
# NoOpt-LABEL: <vec_dq>:
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis
# NoOpt-NEXT: addi
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: lxv 3, -32720(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: stxv 3, -32704(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: blr
.global vec_ds
.p2align 4
.type vec_ds,@function
vec_ds:
.Lvec_ds_gep:
addis 2, 12, .TOC.-.Lvec_ds_gep@ha
addi 2, 2, .TOC.-.Lvec_ds_gep@l
.Lvec_ds_lep:
.localentry vec_ds, .Lvec_dq_lep-.Lvec_dq_gep
addis 3, 2, vecLd@toc@ha
lxsd 3, vecLd@toc@l(3)
addis 3, 2, vecSt@toc@ha
stxsd 3, vecSt@toc@l(3)
addis 3, 2, vecLd@toc@ha
lxssp 3, vecLd@toc@l(3)
addis 3, 2, vecSt@toc@ha
stxssp 3, vecSt@toc@l(3)
blr
[llvm-objdump] -d: print `00000000 <foo>:` instead of `00000000 foo:` The new behavior matches GNU objdump. A pair of angle brackets makes tests slightly easier. `.foo:` is not unique and thus cannot be used in a `CHECK-LABEL:` directive. Without `-LABEL`, the CHECK line can match the `Disassembly of section` line and causes the next `CHECK-NEXT:` to fail. ``` Disassembly of section .foo: 0000000000001634 .foo: ``` Bdragon: <> has metalinguistic connotation. it just "feels right" Reviewed By: rupprecht Differential Revision: https://reviews.llvm.org/D75713
2020-03-06 06:18:38 +08:00
# Dis-LABEL: <vec_ds>:
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: addis
# Dis-NEXT: addi
# Dis-NEXT: nop
[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
# Dis-NEXT: lxsd 3, -32720(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: stxsd 3, -32704(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: lxssp 3, -32720(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: nop
[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
# Dis-NEXT: stxssp 3, -32704(2)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# Dis-NEXT: blr
[llvm-objdump] -d: print `00000000 <foo>:` instead of `00000000 foo:` The new behavior matches GNU objdump. A pair of angle brackets makes tests slightly easier. `.foo:` is not unique and thus cannot be used in a `CHECK-LABEL:` directive. Without `-LABEL`, the CHECK line can match the `Disassembly of section` line and causes the next `CHECK-NEXT:` to fail. ``` Disassembly of section .foo: 0000000000001634 .foo: ``` Bdragon: <> has metalinguistic connotation. it just "feels right" Reviewed By: rupprecht Differential Revision: https://reviews.llvm.org/D75713
2020-03-06 06:18:38 +08:00
# NoOpt-LABEL: <vec_ds>:
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis
# NoOpt-NEXT: addi
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: lxsd 3, -32720(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: stxsd 3, -32704(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: lxssp 3, -32720(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: addis 3, 2, 0
[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
# NoOpt-NEXT: stxssp 3, -32704(3)
[PPC64] Optimize redundant instructions in global access sequences. The access sequence for global variables in the medium and large code models use 2 instructions to add an offset to the toc-pointer. If the offset fits whithin 16-bits then the instruction that sets the high 16 bits is redundant. This patch adds the --toc-optimize option, (on by default) and enables rewriting of 2 instruction global variable accesses into 1 when the offset from the TOC-pointer to the variable (or .got entry) fits in 16 signed bits. eg addis %r3, %r2, 0 --> nop addi %r3, %r3, -0x8000 --> addi %r3, %r2, -0x8000 This rewriting can be disabled with the --no-toc-optimize flag Differential Revision: https://reviews.llvm.org/D49237 llvm-svn: 342602
2018-09-20 08:26:44 +08:00
# NoOpt-NEXT: blr
.global byteLd
.lcomm byteLd, 1, 1
.global byteSt
.lcomm byteSt, 1, 1
.global halfLd
.lcomm halfLd, 2, 2
.global halfSt
.lcomm halfSt, 2, 2
.global wordLd
.lcomm wordLd, 4, 4
.global wordSt
.lcomm wordSt, 4, 4
.global dwordLd
.lcomm dwordLd, 8, 8
.global dwordSt
.lcomm dwordSt, 8, 8
.global vecLd
.lcomm vecLd, 16, 16
.global vecSt
.lcomm vecSt, 16, 16