Improve dead code elimination

This commit is contained in:
Justine Tunney 2021-02-08 04:04:42 -08:00
parent 760db8c5ad
commit 0e36cb3ac4
6606 changed files with 9685 additions and 9854 deletions

View file

@ -47,9 +47,6 @@
#define USE_SYMBOL_HACK 0
.source "NOTICE"
.source "ape/ape.S"
.source "ape/ape.lds"
.section .text,"ax",@progbits
.align __SIZEOF_POINTER__
.previous
@ -108,6 +105,8 @@ rlstr: .endobj rlstr,globl,hidden # ←for gdb readibility
αcτµαlly pδrταblε εxεcµταblε § the old technology
*/
#if SupportsWindows() || SupportsXnu()
/ MZ Literally Executable Header
/
/ This is the beginning of the program file and it can serve as an
@ -116,7 +115,7 @@ rlstr: .endobj rlstr,globl,hidden # ←for gdb readibility
/ a machine, this header may need to morph itself to say the magic
/ words, e.g. ELF, which also works fine as a generic entrypoint.
/
/ @see www.delorie.com/djgpp/doc/exe/
/ @param dl is drive number
/ @noreturn
ape.mz: .ascii "MZ" # Mark 'Zibo' Joseph Zbikowski
jno 2f # MZ: bytes on last page
@ -136,10 +135,48 @@ ape.mz: .ascii "MZ" # Mark 'Zibo' Joseph Zbikowski
.ascii "JT" # MZ: OEM identifier
.short 0 # MZ: OEM information
.org 0x40-4 # MZ: bytes reserved for you
#if SupportsWindows()
.long RVA(ape.pe) # PE: the new technology
#else
.long 0
#endif
.endfn ape.mz,globl,hidden
#else /* !(SupportsWindows() || SupportsXnu()) */
/ ELF Literally Executable Header
/
/ If we don't need to support Microsoft or Apple then we can
/ produce a conventional executable without the shell script
/
/ @param dl is drive number
/ @noreturn
.ascii "\177ELF" # 0x0: ELF
.byte ELFCLASS64 # 4: long mode
.byte ELFDATA2LSB # 5: little endian
.byte 1 # 6: elf v1.o
.byte ELFOSABI_FREEBSD # 7: FreeBSD
.byte 0 # 8: os/abi ver.
.align 8,0 # 9: padding
.short ET_EXEC # 10: εxεcµταblε
.short EM_NEXGEN32E # 12: NexGen32e
.long 1 # 14: elf v1.o
.quad .Lape.elf.entry # 18: e_entry
.quad .Lape.elf.phoff # 20: e_phoff
.quad .Lape.elf.shoff # 28: e_shoff
.long 0 # 30: e_flags
.short 64 # 34: e_ehsize
.short 56 # 36: e_phentsize
.short .Lape.elf.phnum # 38: e_phnum
.short 0 # 3a: e_shentsize
.short .Lape.elf.shnum # 3c: e_shnum
.short .Lape.elf.shstrndx # 3e: e_shstrndx
#endif /* SupportsWindows() || SupportsXnu() */
/ Disk Operating System Stub
/
/ @param dl is drive number
/ @noreturn
.org 0x40 # mz/elf header length
stub: mov $0x40,%dl # *literally* dos
@ -187,6 +224,7 @@ pc: cld
xor %cx,%cx
mov %cx,%es
rlstack %di,%cx
#if SupportsMetal()
push %cs # memcpy() [relocate this page]
pop %ds
call 1f
@ -391,6 +429,8 @@ ape_disk:
.short BOOTSIG
.endobj ape_disk
#endif /* SupportsMetal() */
/*
@ -432,8 +472,10 @@ ape_disk:
the bourne executable & linkable format */
#if SupportsWindows() || SupportsXnu()
apesh: .ascii "'\n#'\"\n" # sixth edition shebang
.ascii "o=\"$(command -v \"$0\")\"\n"
#if SupportsXnu()
.ascii "if [ -d /Applications ]; then\n"
.ascii "dd if=\"$o\""
.ascii " of=\"$o\""
@ -443,7 +485,9 @@ apesh: .ascii "'\n#'\"\n" # sixth edition shebang
.ascii "\" count=\""
.shstub .Lape.macho.dd.count,2
.ascii "\" conv=notrunc 2>/dev/null\n"
.ascii "elif exec 7<> \"$o\"; then\n"
.ascii "el"
#endif
.ascii "if exec 7<> \"$o\"; then\n"
.ascii "printf '"
.ascii "\\177ELF" # 0x0: ELF
.ascii "\\2" # 4: long mode
@ -483,6 +527,7 @@ apesh: .ascii "'\n#'\"\n" # sixth edition shebang
.ascii "fi\n"
.ascii "exit $R\n"
.endobj apesh
#endif /* SupportsWindows() || SupportsXnu() */
.section .elf.phdrs,"a",@progbits
.align __SIZEOF_POINTER__
@ -506,6 +551,7 @@ ape.phdrs:
.stub .Lape.ram.filesz,quad
.stub .Lape.ram.memsz,quad
.stub .Lape.ram.align,quad
#if SupportsLinux()
/ Linux ignores mprotect() and returns 0 without this lool
/ It has nothing to do with the stack, which is still exec
.align __SIZEOF_POINTER__
@ -517,6 +563,8 @@ ape.phdrs:
.quad 0 # p_filesz
.quad 0 # p_memsz
.quad 16 # p_align
#endif
#if SupportsOpenbsd() || SupportsNetbsd()
.align __SIZEOF_POINTER__
.long PT_NOTE # notes
.long PF_R
@ -526,8 +574,10 @@ ape.phdrs:
.stub .Lape.note.filesz,quad
.stub .Lape.note.memsz,quad
.stub .Lape.note.align,quad
#endif
.previous
#if SupportsOpenbsd()
.section .note.openbsd.ident,"a",@progbits
.Lopenbsd.ident:
.long 2f-1f
@ -539,7 +589,9 @@ ape.phdrs:
4: .size .Lopenbsd.ident,.-.Lopenbsd.ident
.type .Lopenbsd.ident,@object
.previous
#endif /* SupportsOpenbsd() */
#if SupportsNetbsd()
.section .note.netbsd.ident,"a",@progbits
.Lnetbsd.ident:
.long 2f-1f
@ -551,6 +603,7 @@ ape.phdrs:
4: .size .Lnetbsd.ident,.-.Lnetbsd.ident
.type .Lnetbsd.ident,@object
.previous
#endif /* SupportsNetbsd() */
/*
@ -583,6 +636,7 @@ ape.phdrs:
@see System V Application Binary Interface NexGen32e Architecture
Processor Supplement, Version 1.0, December 5th, 2018 */
#if SupportsXnu()
.section .macho,"a",@progbits
.align __SIZEOF_POINTER__
@ -654,8 +708,8 @@ ape.macho:
.long 0,0,0 # reserved
40: .long MAC_LC_UUID
.long 50f-40b
.stub uuid1_,quad
.stub uuid2_,quad
.stub .Lape.uuid1,quad
.stub .Lape.uuid2,quad
50: .long MAC_LC_UNIXTHREAD
.long 60f-50b # cmdsize
.long MAC_THREAD_NEXGEN32E # flavaflav
@ -686,6 +740,7 @@ ape.macho:
.endobj ape.macho,globl,hidden
.previous /* .macho */
#endif /* SupportsXnu() */
/*
@ -763,6 +818,8 @@ ape.macho:
.LPEDATA = 0b11000000000000000000000011000000
.LPEIMPS = 0b11000000000000000000000001000000
#if SupportsWindows()
.section .pe.header,"a",@progbits
.align __SIZEOF_POINTER__
ape.pe: .ascin "PE",4
@ -849,6 +906,8 @@ ape.pe: .ascin "PE",4
.long .LPEDATA # Flags
.previous
#endif /* SupportsWindows() */
.section .idata.ro.idt.1,"a",@progbits
.type idata.idtend,@object
.type idata.idt,@object
@ -878,6 +937,7 @@ idata.iat:
idata.iatend:
.previous
#if SupportsMetal()
/*
αcτµαlly pδrταblε εxεcµταblε § early-stage read-only data
@ -1486,6 +1546,8 @@ metal:
jmp _start
.endfn metal
#endif /* SupportsMetal() */
/ Avoid linker script variables appearing as code in objdump.
.macro .ldsvar name:req
.type \name,@object

View file

@ -176,6 +176,7 @@
#ifdef __LINKER__
#include "ape/macros.internal.h"
#include "ape/config.h"
#include "libc/dce.h"
#include "libc/nt/pedef.internal.h"
#include "libc/zip.h"
@ -245,7 +246,7 @@ SECTIONS {
. += 1;
KEEP(*(.ape.pad.head))
. = ALIGN(4096); /* alignments only mandatory when impossible otherwise */
. = ALIGN(SupportsWindows() ? PAGESIZE : 16);
HIDDEN(_ehead = .);
} AT>SmallCode :Head
@ -345,6 +346,22 @@ SECTIONS {
KEEP(*(SORT_BY_NAME(.idata.ro.*)))
. += 1;
. = ALIGN(__SIZEOF_POINTER__);
PROVIDE_HIDDEN(__init_array_start = .);
KEEP(*(SORT_BY_INIT_PRIORITY(.init_array.*)
SORT_BY_INIT_PRIORITY(.ctors.*)))
KEEP(*(SORT_NONE(.ctors)))
KEEP(*(SORT_NONE(.init_array)))
KEEP(*(SORT_NONE(.preinit_array)))
PROVIDE_HIDDEN(__init_array_end = .);
. = ALIGN(__SIZEOF_POINTER__);
PROVIDE_HIDDEN(__fini_array_start = .);
KEEP(*(SORT_BY_INIT_PRIORITY(.fini_array.*)
SORT_BY_INIT_PRIORITY(.dtors.*)))
KEEP(*(SORT_NONE(.dtors)))
PROVIDE_HIDDEN(__fini_array_end = .);
/* Encoded Data Structures w/ Linear Initialization Order */
KEEP(*(.initroprologue))
KEEP(*(SORT_BY_NAME(.initro.*)))
@ -379,24 +396,6 @@ SECTIONS {
. = ALIGN(.Lape.piro.align);
HIDDEN(__piro_start = .);
QUAD(IMAGE_BASE_VIRTUAL);
PROVIDE_HIDDEN(__init_array_start = .);
KEEP(*(SORT_BY_INIT_PRIORITY(.init_array.*)
SORT_BY_INIT_PRIORITY(.ctors.*)))
KEEP(*(SORT_NONE(.ctors)))
KEEP(*(SORT_NONE(.init_array)))
KEEP(*(SORT_NONE(.preinit_array)))
PROVIDE_HIDDEN(__init_array_end = .);
. += 1;
. = ALIGN(__SIZEOF_POINTER__);
PROVIDE_HIDDEN(__fini_array_start = .);
KEEP(*(SORT_BY_INIT_PRIORITY(.fini_array.*)
SORT_BY_INIT_PRIORITY(.dtors.*)))
KEEP(*(SORT_NONE(.dtors)))
PROVIDE_HIDDEN(__fini_array_end = .);
KEEP(*(SORT_BY_NAME(.piro.relo.sort.*)))
PROVIDE_HIDDEN(__relo_end = .);
KEEP(*(SORT_BY_NAME(.piro.data.sort.*)))
@ -481,25 +480,17 @@ PFSTUB8(.Lape.elf.shoff, 0);
PFSTUB4(.Lape.elf.phnum, (.Lape.phdrs.end - ape.phdrs) / 56);
PFSTUB4(.Lape.elf.shnum, 0);
PFSTUB4(.Lape.elf.shstrndx, 0);
SHSTUB2(.Lape.macho.dd.skip, RVA(ape.macho) / 8);
SHSTUB2(.Lape.macho.dd.count, (.Lape.macho.end - ape.macho) / 8);
PFSTUB4(.Lape.pe.offset, ape.pe - ape.mz);
HIDDEN(__privileged_addr = ROUNDDOWN(__privileged_start, PAGESIZE));
HIDDEN(__privileged_size = (ROUNDUP(__privileged_end, PAGESIZE) -
ROUNDDOWN(__privileged_start, PAGESIZE)));
HIDDEN(.Lape.pe.optsz = .Lape.pe.sections - (ape.pe + 24));
HIDDEN(.Lape.pe.shnum = (.Lape.pe.sections_end - .Lape.pe.sections) / 40);
HIDDEN(.Lidata.idtsize = idata.idtend - idata.idt);
HIDDEN(.Lidata.iatsize = idata.iatend - idata.iat);
HIDDEN(.Lape.rom.offset = 0);
HIDDEN(.Lape.rom.vaddr = ADDR(.head));
HIDDEN(.Lape.rom.paddr = LOADADDR(.head));
HIDDEN(.Lape.rom.filesz = LOADADDR(.data) - .Lape.rom.paddr);
HIDDEN(.Lape.rom.memsz = ADDR(.data) - ADDR(.head));
HIDDEN(.Lape.rom.align = 0x1000);
HIDDEN(.Lape.rom.align = .Lape.data.align);
HIDDEN(.Lape.rom.rva = RVA(.Lape.rom.vaddr));
HIDDEN(.Lape.ram.offset = .Lape.rom.offset + .Lape.rom.filesz);
@ -507,7 +498,7 @@ HIDDEN(.Lape.ram.vaddr = ADDR(.data));
HIDDEN(.Lape.ram.paddr = LOADADDR(.data));
HIDDEN(.Lape.ram.filesz = LOADADDR(.bss) - LOADADDR(.data));
HIDDEN(.Lape.ram.memsz = ADDR(.bss) + SIZEOF(.bss) - .Lape.ram.vaddr);
HIDDEN(.Lape.ram.align = 0x1000);
HIDDEN(.Lape.ram.align = .Lape.data.align);
HIDDEN(.Lape.ram.rva = RVA(.Lape.ram.vaddr));
HIDDEN(.Lape.note.offset = .Lape.rom.offset + (.Lape.note - .Lape.rom.vaddr));
@ -530,7 +521,7 @@ HIDDEN(.Lape.data.paddr = LOADADDR(.data));
HIDDEN(.Lape.data.vaddr = ADDR(.data));
HIDDEN(.Lape.data.filesz = SIZEOF(.data));
HIDDEN(.Lape.data.memsz = SIZEOF(.data));
HIDDEN(.Lape.data.align = 0x1000);
HIDDEN(.Lape.data.align = .Lape.data.align);
HIDDEN(.Lape.data.rva = RVA(.Lape.data.vaddr));
HIDDEN(.Lape.bss.offset = .Lape.ram.offset + LOADADDR(.bss) - .Lape.ram.paddr);
@ -540,18 +531,30 @@ HIDDEN(.Lape.bss.filesz = 0);
HIDDEN(.Lape.bss.memsz = SIZEOF(.bss));
HIDDEN(.Lape.bss.align = .Lape.data.align);
/* Program Loader Auto-Tune */
#if SupportsXnu()
SHSTUB2(.Lape.macho.dd.skip, RVA(ape.macho) / 8);
SHSTUB2(.Lape.macho.dd.count, (.Lape.macho.end - ape.macho) / 8);
#endif
#if SupportsWindows()
PFSTUB4(.Lape.pe.offset, ape.pe - ape.mz);
HIDDEN(.Lape.pe.optsz = .Lape.pe.sections - (ape.pe + 24));
HIDDEN(.Lape.pe.shnum = (.Lape.pe.sections_end - .Lape.pe.sections) / 40);
HIDDEN(.Lidata.idtsize = idata.idtend - idata.idt);
HIDDEN(.Lidata.iatsize = idata.iatend - idata.iat);
HIDDEN(v_ntsubsystem = (DEFINED(GetMessage)
? kNtImageSubsystemWindowsGui
: kNtImageSubsystemWindowsCui));
#endif
#if SupportsMetal()
HIDDEN(v_ape_realsectors =
MIN(REAL_SCRATCH_AREA - IMAGE_BASE_REAL,
ROUNDUP(RVA(_edata), 4096)) / 512);
HIDDEN(v_ape_realpages = v_ape_realsectors / (4096 / 512));
HIDDEN(v_ape_highsectors =
(ROUNDUP(RVA(_edata), 512) / 512) - v_ape_realsectors);
/* Windows NT Auto-Subsystem Embedding */
HIDDEN(v_ntsubsystem = (DEFINED(GetMessage)
? kNtImageSubsystemWindowsGui
: kNtImageSubsystemWindowsCui));
#endif
/* ZIP End of Central Directory header */
#define ZIPCONST(NAME, VAL) HIDDEN(NAME = DEFINED(__zip_start) ? VAL : 0);
@ -560,7 +563,8 @@ ZIPCONST(v_zip_cdirsize, __zip_end - __zip_start);
ZIPCONST(v_zip_records, v_zip_cdirsize / kZipCdirHdrLinkableSize);
ZIPCONST(v_zip_commentsize, _edata - __zip_end - kZipCdirHdrMinSize);
/* Generates deterministic ID for Mach-O. */
#if SupportsXnu()
/* Generates deterministic ID. */
#define PHI 0x9e3779b9925d4c17
#define XOR(X,Y) ((X | Y) - (X & Y))
#define XORSHIFT(X,Y) \
@ -573,14 +577,12 @@ ZIPCONST(v_zip_commentsize, _edata - __zip_end - kZipCdirHdrMinSize);
X = (X + (Y >> 020) & 0xFF) * PHI; \
X = (X + (Y >> 030) & 0xFF) * PHI
#define CHURN(X) \
XORSHIFT(uuid1_, X); \
KMH(uuid1_, X); \
XORSHIFT(uuid2_, X); \
KMH(uuid2_, X)
HIDDEN(uuid1_ = 88172645463325252);
HIDDEN(uuid2_ = 88172645463325252);
XORSHIFT(.Lape.uuid1, X); \
KMH(.Lape.uuid1, X); \
XORSHIFT(.Lape.uuid2, X); \
KMH(.Lape.uuid2, X)
HIDDEN(.Lape.uuid1 = 88172645463325252);
HIDDEN(.Lape.uuid2 = 88172645463325252);
CHURN(.Lape.bss.align);
CHURN(.Lape.bss.filesz);
CHURN(.Lape.bss.memsz);
@ -608,12 +610,6 @@ CHURN(.Lape.note.memsz);
CHURN(.Lape.note.offset);
CHURN(.Lape.note.paddr);
CHURN(.Lape.note.vaddr);
CHURN(.Lape.pe.offset);
CHURN(.Lape.pe.optsz);
CHURN(.Lape.pe.sections);
CHURN(.Lape.pe.sections_end);
CHURN(.Lape.pe.shnum);
CHURN(.Lape.phdrs.end);
CHURN(.Lape.ram.align);
CHURN(.Lape.ram.filesz);
CHURN(.Lape.ram.memsz);
@ -636,15 +632,28 @@ CHURN(.Lape.text.paddr);
CHURN(.Lape.text.rva);
CHURN(.Lape.text.vaddr);
CHURN(ADDR(.bss));
CHURN(WinMain);
CHURN(_start);
CHURN(ape.phdrs);
#if SupportsMetal()
CHURN(v_ape_realsectors);
#endif
#if SupportsXnu()
CHURN(ape.macho);
#endif
#if SupportsWindows()
CHURN(ape.mz);
CHURN(ape.pe);
CHURN(ape.phdrs);
CHURN(v_ape_realsectors);
CHURN(.Lape.pe.offset);
CHURN(.Lape.pe.optsz);
CHURN(.Lape.pe.sections);
CHURN(.Lape.pe.sections_end);
CHURN(.Lape.pe.shnum);
CHURN(.Lape.phdrs.end);
CHURN(WinMain);
#endif /* SupportsWindows() */
#endif /* SupportsXnu() */
ASSERT(ape.mz == IMAGE_BASE_VIRTUAL, "linker panic");
ASSERT(DEFINED(ape.mz) ? ape.mz == IMAGE_BASE_VIRTUAL : 1, "linker panic");
ASSERT((DEFINED(__init_bss_end) ? __init_bss_end : 0) % __SIZEOF_POINTER__ == 0,
"__init_bss misalign");
ASSERT(((DEFINED(__init_rodata_end) ? __init_rodata_end : 0) %

View file

@ -21,7 +21,7 @@ APE = $(APE_DEPS) \
APELINK = \
ACTION=LINK.ape \
$(MKDIR) $(@D) && \
$(COMPILE) \
$(LINK) \
$(LINKARGS) \
$(OUTPUT_OPTION) && \
@ -39,6 +39,13 @@ APE_OBJS = $(APE_SRCS:%.S=o/$(MODE)/%.o)
APE_DEPS = $(APE_LIB)
APE_CHECKS = $(APE_HDRS:%=o/%.ok)
o/$(MODE)/ape/ape.lds: \
ape/ape.lds \
ape/config.h \
ape/macros.internal.h \
libc/dce.h \
libc/zip.h
o/ape/idata.inc: \
ape/idata.internal.h \
ape/relocations.h