diff options
| author | Adhemerval Zanella <adhemerval.zanella@linaro.org> | 2017-12-12 19:15:45 -0200 |
|---|---|---|
| committer | Adhemerval Zanella <adhemerval.zanella@linaro.org> | 2017-12-20 16:55:24 -0200 |
| commit | ed95f6114928f00f5f8fe76ccf2ec9e77872cbad (patch) | |
| tree | 0161d2bc2599666e4dab69cfb07a75a445760f79 /sysdeps/tile | |
| parent | 4735850f7a4f07f05ee9ff405238332f4bb3ab50 (diff) | |
| download | glibc-ed95f6114928f00f5f8fe76ccf2ec9e77872cbad.tar.gz glibc-ed95f6114928f00f5f8fe76ccf2ec9e77872cbad.tar.xz glibc-ed95f6114928f00f5f8fe76ccf2ec9e77872cbad.zip | |
Remove tilepro-*-linux-gnu support
As from previous discussions [1] this patch removes tileprox-*-linux-gnu support from GLIBC. This patch is a straigthfoward one, which just remove tilepro specific implementation and configurations (no sysdep simplfication or reorganization is done). * README: Remove tilepro-*-linux-gnu from supported architecture. * scripts/build-many-glibcs.py: Likewise. * sysdeps/tile/__tls_get_addr.S (__tls_get_addr): Likewise. * sysdeps/tile/crti.S (PREINIT_FUNCTION): Likewise. * sysdeps/tile/dl-machine.h (ELF_MACHINE_NAME, elf_machine_matches_host, elf_machine_dynamic, elf_machine_load_address, elf_machine_runtime_setup, reloc_howto howto, elf_machine_rela): Likewise * sysdeps/tile/dl-start.S (_start): Likewise. * sysdeps/tile/memcmp.c (DBLALIGN, REVBYTES): Likewise. * sysdeps/tile/memcopy.h (MEMCPY_OK_FOR_FWD_MEMMOVE, MEMCPY_OK_FOR_FWD_MEMMOVE, op_t): Likewise. * sysdeps/tile/nptl/pthread_spin_lock.c (TNS, CMPTNS): Likewise. * sysdeps/tile/nptl/pthread_spin_trylock.c (TNS): Likewise. * sysdeps/tile/nptl/pthread_spin_unlock.c (pthread_spin_unlock): Likewise. * sysdeps/tile/nptl/tls.h (DB_THREAD_SELF): Likewise. * sysdeps/tile/preconfigure: Likewise. * sysdeps/tile/stackguard-macros.h (STACK_CHK_GUARD, POINTER_CHK_GUARD): Likewise. * sysdeps/tile/stackinfo.h (__stackinfo_sub): Likewise. * sysdeps/tile/start.S (_start): Likewise. * sysdeps/tile/tls-macros.h (TLS_GD_OFFSET, TLS_IE_OFFSET, _TLS_LE): Likewise. * sysdeps/tile/sysdep.h (REGSIZE): Likewise. (LD, LD4U, ST, ST4, BNEZ, BEQZ, BEQZT, BGTZ, CMPEQI, CMPEQ, CMOVEQZ, CMOVNEZ): Remove. * sysdeps/unix/sysv/linux/tile/bits/environments.h (__ILP32_OFF32_CFLAGS, __ILP32_OFFBIG_CFLAGS, __ILP32_OFF32_LDFLAGS, __ILP32_OFFBIG_LDFLAGS, __LP64_OFF64_CFLAGS, __LP64_OFF64_LDFLAGS): Likewise. * sysdeps/tile/wordcopy.c (DBLALIGN): Likewise. * sysdeps/tile/tilepro/Implies: Remove file. * sysdeps/tile/tilepro/atomic-machine.h: Likewise. * sysdeps/tile/tilepro/bits/wordsize.h: Likewise. * sysdeps/tile/tilepro/memchr.c: Likewise. * sysdeps/tile/tilepro/memcpy.S: Likewise. * sysdeps/tile/tilepro/memset.c: Likewise. * sysdeps/tile/tilepro/memusage.h: Likewise. * sysdeps/tile/tilepro/rawmemchr.c: Likewise. * sysdeps/tile/tilepro/strchr.c: Likewise. * sysdeps/tile/tilepro/strchrnul.c: Likewise. * sysdeps/tile/tilepro/strlen.c: Likewise. * sysdeps/tile/tilepro/strrchr.c: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/Implies: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/c++-types.data: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/jmp_buf-macros.h: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/ld.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/ldconfig.h: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libBrokenLocale.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libanl.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libc.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libcrypt.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libdl.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libnsl.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libpthread.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libresolv.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/librt.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libthread_db.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libutil.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/register-dump.h: Likewise. * sysdeps/unix/sysv/linux/tile/sysconf.c (linux_sysconf): Remove tilepro mention in comment. [1] https://sourceware.org/ml/libc-alpha/2017-12/msg00038.html
Diffstat (limited to 'sysdeps/tile')
29 files changed, 27 insertions, 1437 deletions
diff --git a/sysdeps/tile/__tls_get_addr.S b/sysdeps/tile/__tls_get_addr.S index 3b2b95522f..2402bd1e9a 100644 --- a/sysdeps/tile/__tls_get_addr.S +++ b/sysdeps/tile/__tls_get_addr.S @@ -37,7 +37,6 @@ ENTRY (__tls_get_addr) ADDI_PTR r27, tp, DTV_OFFSET } .Llnk: -#ifdef __tilegx__ { LD_PTR r27, r27 /* r27 = THREAD_DTV() */ moveli r26, hw1_last(_rtld_local + TLS_GENERATION_OFFSET - .Llnk) @@ -47,16 +46,6 @@ ENTRY (__tls_get_addr) ADD_PTR r25, r25, r26 LD_PTR r26, r0 /* r26 = ti_module */ } -#else - { - LD_PTR r27, r27 /* r27 = THREAD_DTV() */ - addli r25, r25, lo16(_rtld_local + TLS_GENERATION_OFFSET - .Llnk) - } - { - auli r25, r25, ha16(_rtld_local + TLS_GENERATION_OFFSET - .Llnk) - LD_PTR r26, r0 /* r26 = ti_module */ - } -#endif LD_PTR r25, r25 /* r25 = DL(dl_tls_generation) */ { LD_PTR r28, r27 /* r28 = THREAD_DTV()->counter */ diff --git a/sysdeps/tile/crti.S b/sysdeps/tile/crti.S index 522373aff5..1d714ab483 100644 --- a/sysdeps/tile/crti.S +++ b/sysdeps/tile/crti.S @@ -70,7 +70,6 @@ _init: #if PREINIT_FUNCTION_WEAK lnk r2 0: -# ifdef __tilegx__ moveli r1, hw2_last(_GLOBAL_OFFSET_TABLE_ - 0b) { shl16insli r1, r1, hw1(_GLOBAL_OFFSET_TABLE_ - 0b) @@ -80,22 +79,12 @@ _init: shl16insli r1, r1, hw0(_GLOBAL_OFFSET_TABLE_ - 0b) shl16insli r0, r0, hw0_got(PREINIT_FUNCTION) } -# else - { - moveli r1, lo16(_GLOBAL_OFFSET_TABLE_ - 0b) - moveli r0, got_lo16(PREINIT_FUNCTION) - } - { - auli r1, r1, ha16(_GLOBAL_OFFSET_TABLE_ - 0b) - auli r0, r0, got_ha16(PREINIT_FUNCTION) - } -# endif ADD_PTR r0, r0, r1 ADD_PTR r0, r0, r2 LD_PTR r0, r0 BEQZ r0, .Lno_weak_fn jalr r0 -#elif defined(__tilegx__) && !defined(NO_PLT_PCREL) +#elif !defined(NO_PLT_PCREL) /* Since we are calling from the start of the object to the PLT, call by loading the full address into a register. */ lnk r2 diff --git a/sysdeps/tile/dl-machine.h b/sysdeps/tile/dl-machine.h index f7af7c3848..d745a9c07f 100644 --- a/sysdeps/tile/dl-machine.h +++ b/sysdeps/tile/dl-machine.h @@ -20,11 +20,7 @@ #ifndef dl_machine_h #define dl_machine_h -#ifdef __tilegx__ #define ELF_MACHINE_NAME "tilegx" -#else -#define ELF_MACHINE_NAME "tilepro" -#endif #include <sys/param.h> #include <string.h> @@ -37,18 +33,12 @@ static inline int elf_machine_matches_host (const ElfW(Ehdr) *ehdr) { -#if defined __tilegx__ if (ehdr->e_machine != EM_TILEGX) return 0; -# if __WORDSIZE == 32 +#if __WORDSIZE == 32 return (ehdr->e_ident[EI_CLASS] == ELFCLASS32); -# else - return (ehdr->e_ident[EI_CLASS] == ELFCLASS64); -# endif -#elif defined __tilepro__ - return ehdr->e_machine == EM_TILEPRO; #else -# error "Unknown tile architecture." + return (ehdr->e_ident[EI_CLASS] == ELFCLASS64); #endif } @@ -62,19 +52,12 @@ elf_machine_dynamic (void) { ElfW(Addr) *got; -#ifdef __tilegx__ ElfW(Addr) tmp; asm( " { lnk %0; moveli %1, hw2_last(_GLOBAL_OFFSET_TABLE_ - 1f) }\n" "1: shl16insli %1, %1, hw1(_GLOBAL_OFFSET_TABLE_ - 1b)\n" " shl16insli %1, %1, hw0(_GLOBAL_OFFSET_TABLE_ - 1b)\n" " add %0, %0, %1" : "=r" (got), "=r" (tmp)); -#else - asm( " lnk %0\n" - "1: addli %0, %0, lo16(_GLOBAL_OFFSET_TABLE_ - 1b)\n" - " auli %0, %0, ha16(_GLOBAL_OFFSET_TABLE_ - 1b)" - : "=r" (got)); -#endif return *got; } @@ -86,9 +69,8 @@ elf_machine_load_address (void) { ElfW(Addr) *got; ElfW(Addr) dynamic; - -#ifdef __tilegx__ ElfW(Addr) tmp; + asm( " lnk %2\n" "1: {\n" " moveli %0, hw2_last(_GLOBAL_OFFSET_TABLE_ - 1b)\n" @@ -107,18 +89,6 @@ elf_machine_load_address (void) " add %1, %1, %2\n" " }" : "=r" (got), "=r" (dynamic), "=r" (tmp)); -#else - asm( " lnk %0\n" - "1: {\n" - " addli %0, %0, lo16(_GLOBAL_OFFSET_TABLE_ - 1b)\n" - " addli %1, %0, lo16(_DYNAMIC - 1b)\n" - " }\n" - " {\n" - " auli %0, %0, ha16(_GLOBAL_OFFSET_TABLE_ - 1b)\n" - " auli %1, %1, ha16(_DYNAMIC - 1b)\n" - " }\n" - : "=r" (got), "=r" (dynamic)); -#endif return dynamic - *got; } @@ -197,15 +167,10 @@ elf_machine_runtime_setup (struct link_map *l, int lazy, int profile) #endif /* Wrap a generic Tilera relocation type. */ -#ifdef __tilegx__ #define R_TILE(x) R_TILEGX_##x #define __R_TILE_TLS(x,c) R_TILEGX_TLS_##x##c #define _R_TILE_TLS(x,c) __R_TILE_TLS(x,c) #define R_TILE_TLS(x) _R_TILE_TLS(x,__ELF_NATIVE_CLASS) -#else -#define R_TILE(x) R_TILEPRO_##x -#define R_TILE_TLS(x) R_TILEPRO_TLS_##x##32 -#endif /* ELF_RTYPE_CLASS_PLT iff TYPE describes relocation of a PLT entry or TLS variable, so undefined references should not be allowed to @@ -272,13 +237,8 @@ struct reloc_howto /* Right shift operand by this number of bits. */ unsigned char right_shift; -#ifdef __tilegx__ /* If nonzero, this is updating a code bundle. */ unsigned char is_bundle_update; -#else - /* If nonzero, add 0x8000 to the value. */ - unsigned char add_0x8000; -#endif /* If nonzero, subtract the containing address from the address. */ unsigned char is_pcrel; @@ -291,19 +251,17 @@ struct reloc_howto because then the table would not be position-independent. */ static const struct reloc_howto howto[] = { -#ifdef __tilegx__ - -# if __WORDSIZE == 32 +#if __WORDSIZE == 32 /* The GX -m32 loader only handles 32-bit types, so it will be confused by shifts larger than that. We convert them to just sign-extend; they usually indicate a program bug or missed optimization, but we have to handle them correctly anyway. */ -# define S32 31 -# define S48 31 -# else -# define S32 32 -# define S48 48 -# endif +# define S32 31 +# define S48 31 +#else +# define S32 32 +# define S48 48 +#endif /* R_TILEGX_NONE */ { 0, 0, 0, 0 }, /* R_TILEGX_64 */ { 0, 0, 0, 8 }, @@ -417,93 +375,6 @@ static const struct reloc_howto howto[] = /* R_TILEGX_TLS_DTPMOD32 */ { 0, 0, 0, 0 }, /* R_TILEGX_TLS_DTPOFF32 */ { 0, 0, 0, 0 }, /* R_TILEGX_TLS_TPOFF32 */ { 0, 0, 0, 0 } -#else - /* R_TILEPRO_NONE */ { 0, 0, 0, 0 }, - /* R_TILEPRO_32 */ { 0, 0, 0, 4 }, - /* R_TILEPRO_16 */ { 0, 0, 0, 2 }, - /* R_TILEPRO_8 */ { 0, 0, 0, 1 }, - /* R_TILEPRO_32_PCREL */ { 0, 0, 1, 4 }, - /* R_TILEPRO_16_PCREL */ { 0, 0, 1, 2 }, - /* R_TILEPRO_8_PCREL */ { 0, 0, 1, 1 }, - /* R_TILEPRO_LO16 */ { 0, 0, 0, 2 }, - /* R_TILEPRO_HI16 */ { 16, 0, 0, 2 }, - /* R_TILEPRO_HA16 */ { 16, 1, 0, 2 }, - /* R_TILEPRO_COPY */ { 0, 0, 0, 0 }, - /* R_TILEPRO_GLOB_DAT */ { 0, 0, 0, 4 }, - /* R_TILEPRO_JMP_SLOT */ { 0, 0, 0, 0 }, - /* R_TILEPRO_RELATIVE */ { 0, 0, 0, 0 }, - /* R_TILEPRO_BROFF_X1 */ { 3, 0, 1, 8 }, - /* R_TILEPRO_JOFFLONG_X1 */ { 3, 0, 1, 8 }, - /* R_TILEPRO_JOFFLONG_X1_PLT */ { 3, 0, 1, 8 }, - /* R_TILEPRO_IMM8_X0 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM8_Y0 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM8_X1 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM8_Y1 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_MT_IMM15_X1 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_MF_IMM15_X1 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X1 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0_LO */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X1_LO */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0_HI */ { 16, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X1_HI */ { 16, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0_HA */ { 16, 1, 0, 8 }, - /* R_TILEPRO_IMM16_X1_HA */ { 16, 1, 0, 8 }, - /* R_TILEPRO_IMM16_X0_PCREL */ { 0, 0, 1, 8 }, - /* R_TILEPRO_IMM16_X1_PCREL */ { 0, 0, 1, 8 }, - /* R_TILEPRO_IMM16_X0_LO_PCREL */ { 0, 0, 1, 8 }, - /* R_TILEPRO_IMM16_X1_LO_PCREL */ { 0, 0, 1, 8 }, - /* R_TILEPRO_IMM16_X0_HI_PCREL */ { 16, 0, 1, 8 }, - /* R_TILEPRO_IMM16_X1_HI_PCREL */ { 16, 0, 1, 8 }, - /* R_TILEPRO_IMM16_X0_HA_PCREL */ { 16, 1, 1, 8 }, - /* R_TILEPRO_IMM16_X1_HA_PCREL */ { 16, 1, 1, 8 }, - /* R_TILEPRO_IMM16_X0_GOT */ { 0, 0, 0, 0 }, - /* R_TILEPRO_IMM16_X1_GOT */ { 0, 0, 0, 0 }, - /* R_TILEPRO_IMM16_X0_GOT_LO */ { 0, 0, 0, 0 }, - /* R_TILEPRO_IMM16_X1_GOT_LO */ { 0, 0, 0, 0 }, - /* R_TILEPRO_IMM16_X0_GOT_HI */ { 0, 0, 0, 0 }, - /* R_TILEPRO_IMM16_X1_GOT_HI */ { 0, 0, 0, 0 }, - /* R_TILEPRO_IMM16_X0_GOT_HA */ { 0, 0, 0, 0 }, - /* R_TILEPRO_IMM16_X1_GOT_HA */ { 0, 0, 0, 0 }, - /* R_TILEPRO_MMSTART_X0 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_MMEND_X0 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_MMSTART_X1 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_MMEND_X1 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_SHAMT_X0 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_SHAMT_X1 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_SHAMT_Y0 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_SHAMT_Y1 */ { 0, 0, 0, 8 }, - /* R_TILEPRO_SN_BROFF */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_IMM8 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_UIMM8 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_BYTE0 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_BYTE1 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_BYTE2 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_BYTE3 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_SPCREL0 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_SPCREL1 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_SPCREL2 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_SN_SPCREL3 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_IMM16_X0_TLS_GD */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X1_TLS_GD */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0_TLS_GD_LO */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X1_TLS_GD_LO */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0_TLS_GD_HI */ { 16, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X1_TLS_GD_HI */ { 16, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0_TLS_GD_HA */ { 16, 1, 0, 8 }, - /* R_TILEPRO_IMM16_X1_TLS_GD_HA */ { 16, 1, 0, 8 }, - /* R_TILEPRO_IMM16_X0_TLS_IE */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X1_TLS_IE */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0_TLS_IE_LO */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X1_TLS_IE_LO */ { 0, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0_TLS_IE_HI */ { 16, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X1_TLS_IE_HI */ { 16, 0, 0, 8 }, - /* R_TILEPRO_IMM16_X0_TLS_IE_HA */ { 16, 1, 0, 8 }, - /* R_TILEPRO_IMM16_X1_TLS_IE_HA */ { 16, 1, 0, 8 }, - /* R_TILEPRO_TLS_DTPMOD32 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_TLS_DTPOFF32 */ { 0, 0, 0, 0 }, - /* R_TILEPRO_TLS_TPOFF32 */ { 0, 0, 0, 0 }, -#endif }; #if __ELF_NATIVE_CLASS == 32 @@ -653,11 +524,6 @@ elf_machine_rela (struct link_map *map, const ElfW(Rela) *reloc, if (h->is_pcrel) value -= (ElfW(Addr)) reloc_addr; -#ifndef __tilegx__ - if (h->add_0x8000) - value += 0x8000; -#endif - value = ((long) value) >> h->right_shift; switch (h->byte_size) @@ -671,14 +537,12 @@ elf_machine_rela (struct link_map *map, const ElfW(Rela) *reloc, case 4: *(int *) reloc_addr = value; return; -#ifdef __tilegx__ case 8: if (!h->is_bundle_update) { *(ElfW(Addr) *) reloc_addr = value; return; } -#endif } /* We are updating a bundle, so use the function pointer that @@ -704,7 +568,6 @@ elf_machine_rela (struct link_map *map, const ElfW(Rela) *reloc, switch (r_type) { -#ifdef __tilegx__ case R_TILEGX_BROFF_X1: MUNGE_SIGNED (BrOff_X1, 17); break; @@ -796,97 +659,6 @@ elf_machine_rela (struct link_map *map, const ElfW(Rela) *reloc, case R_TILEGX_SHAMT_Y1: MUNGE (ShAmt_Y1); break; -#else - case R_TILEPRO_BROFF_X1: - MUNGE_SIGNED (BrOff_X1, 17); - break; - case R_TILEPRO_JOFFLONG_X1: - case R_TILEPRO_JOFFLONG_X1_PLT: - MUNGE_NOCHECK (JOffLong_X1); /* holds full 32-bit value */ - break; - case R_TILEPRO_IMM8_X0: - MUNGE_SIGNED (Imm8_X0, 8); - break; - case R_TILEPRO_IMM8_Y0: - MUNGE_SIGNED (Imm8_Y0, 8); - break; - case R_TILEPRO_IMM8_X1: - MUNGE_SIGNED (Imm8_X1, 8); - break; - case R_TILEPRO_IMM8_Y1: - MUNGE_SIGNED (Imm8_Y1, 8); - break; - case R_TILEPRO_MT_IMM15_X1: - MUNGE (MT_Imm15_X1); - break; - case R_TILEPRO_MF_IMM15_X1: - MUNGE (MF_Imm15_X1); - break; - case R_TILEPRO_IMM16_X0_LO: - case R_TILEPRO_IMM16_X0_HI: - case R_TILEPRO_IMM16_X0_HA: - case R_TILEPRO_IMM16_X0_LO_PCREL: - case R_TILEPRO_IMM16_X0_HI_PCREL: - case R_TILEPRO_IMM16_X0_HA_PCREL: - case R_TILEPRO_IMM16_X0_TLS_GD_LO: - case R_TILEPRO_IMM16_X0_TLS_GD_HI: - case R_TILEPRO_IMM16_X0_TLS_GD_HA: - case R_TILEPRO_IMM16_X0_TLS_IE_LO: - case R_TILEPRO_IMM16_X0_TLS_IE_HI: - case R_TILEPRO_IMM16_X0_TLS_IE_HA: - MUNGE_NOCHECK (Imm16_X0); - break; - case R_TILEPRO_IMM16_X0: - case R_TILEPRO_IMM16_X0_PCREL: - case R_TILEPRO_IMM16_X0_TLS_GD: - case R_TILEPRO_IMM16_X0_TLS_IE: - MUNGE_SIGNED (Imm16_X0, 16); - break; - case R_TILEPRO_IMM16_X1_LO: - case R_TILEPRO_IMM16_X1_HI: - case R_TILEPRO_IMM16_X1_HA: - case R_TILEPRO_IMM16_X1_LO_PCREL: - case R_TILEPRO_IMM16_X1_HI_PCREL: - case R_TILEPRO_IMM16_X1_HA_PCREL: - case R_TILEPRO_IMM16_X1_TLS_GD_LO: - case R_TILEPRO_IMM16_X1_TLS_GD_HI: - case R_TILEPRO_IMM16_X1_TLS_GD_HA: - case R_TILEPRO_IMM16_X1_TLS_IE_LO: - case R_TILEPRO_IMM16_X1_TLS_IE_HI: - case R_TILEPRO_IMM16_X1_TLS_IE_HA: - MUNGE_NOCHECK (Imm16_X1); - break; - case R_TILEPRO_IMM16_X1: - case R_TILEPRO_IMM16_X1_PCREL: - case R_TILEPRO_IMM16_X1_TLS_GD: - case R_TILEPRO_IMM16_X1_TLS_IE: - MUNGE_SIGNED (Imm16_X1, 16); - break; - case R_TILEPRO_MMSTART_X0: - MUNGE (MMStart_X0); - break; - case R_TILEPRO_MMEND_X0: - MUNGE (MMEnd_X0); - break; - case R_TILEPRO_MMSTART_X1: - MUNGE (MMStart_X1); - break; - case R_TILEPRO_MMEND_X1: - MUNGE (MMEnd_X1); - break; - case R_TILEPRO_SHAMT_X0: - MUNGE (ShAmt_X0); - break; - case R_TILEPRO_SHAMT_X1: - MUNGE (ShAmt_X1); - break; - case R_TILEPRO_SHAMT_Y0: - MUNGE (ShAmt_Y0); - break; - case R_TILEPRO_SHAMT_Y1: - MUNGE (ShAmt_Y1); - break; -#endif } #undef MUNGE *p = bits; diff --git a/sysdeps/tile/dl-start.S b/sysdeps/tile/dl-start.S index 4f807a9e64..1782c9cd00 100644 --- a/sysdeps/tile/dl-start.S +++ b/sysdeps/tile/dl-start.S @@ -20,14 +20,9 @@ /* Get address of "sym" in "reg" assuming r51 holds ".Llink". */ .macro pic_addr reg, sym -#ifdef __tilegx__ moveli \reg, hw1_last(\sym - .Llink) shl16insli \reg, \reg, hw0(\sym - .Llink) ADD_PTR \reg, r51, \reg -#else - ADDLI_PTR \reg, r51, lo16(\sym - .Llink) - auli \reg, \reg, ha16(\sym - .Llink) -#endif .endm .text diff --git a/sysdeps/tile/memcmp.c b/sysdeps/tile/memcmp.c index c3ed88bca7..07305d04f2 100644 --- a/sysdeps/tile/memcmp.c +++ b/sysdeps/tile/memcmp.c @@ -59,13 +59,8 @@ typedef unsigned char byte; /* Provide the appropriate builtins to shift two registers based on the alignment of a pointer held in a third register, and to reverse the bytes in a word. */ -#ifdef __tilegx__ #define DBLALIGN __insn_dblalign #define REVBYTES __insn_revbytes -#else -#define DBLALIGN __insn_dword_align -#define REVBYTES __insn_bytex -#endif #ifdef WORDS_BIGENDIAN # define CMP_LT_OR_GT(a, b) ((a) > (b) ? 1 : -1) diff --git a/sysdeps/tile/memcopy.h b/sysdeps/tile/memcopy.h index d934165d98..1579beeb7a 100644 --- a/sysdeps/tile/memcopy.h +++ b/sysdeps/tile/memcopy.h @@ -20,14 +20,12 @@ #include <bits/wordsize.h> /* The tilegx implementation of memcpy is safe to use for memmove. */ -#ifdef __tilegx__ -# undef MEMCPY_OK_FOR_FWD_MEMMOVE -# define MEMCPY_OK_FOR_FWD_MEMMOVE 1 -#endif +#undef MEMCPY_OK_FOR_FWD_MEMMOVE +#define MEMCPY_OK_FOR_FWD_MEMMOVE 1 /* Support more efficient copying on tilegx32, which supports long long as a native 64-bit type. */ -#if defined (__tilegx__) && __WORDSIZE == 32 +#if __WORDSIZE == 32 # undef op_t # define op_t unsigned long long int #endif diff --git a/sysdeps/tile/nptl/pthread_spin_lock.c b/sysdeps/tile/nptl/pthread_spin_lock.c index 16ef021de0..21c5ed4c0e 100644 --- a/sysdeps/tile/nptl/pthread_spin_lock.c +++ b/sysdeps/tile/nptl/pthread_spin_lock.c @@ -26,16 +26,10 @@ /* Initial cycle delay for exponential backoff */ #define BACKOFF_START 32 -#ifdef __tilegx__ /* Use cmpexch() after the initial fast-path exch to avoid invalidating the cache line of the lock holder. */ -# define TNS(p) atomic_exchange_acq((p), 1) -# define CMPTNS(p) atomic_compare_and_exchange_val_acq((p), 1, 0) -#else -# define TNS(p) __insn_tns(p) -# define CMPTNS(p) __insn_tns(p) -# define SPR_CYCLE SPR_CYCLE_LOW /* The low 32 bits are sufficient. */ -#endif +#define TNS(p) atomic_exchange_acq((p), 1) +#define CMPTNS(p) atomic_compare_and_exchange_val_acq((p), 1, 0) int pthread_spin_lock (pthread_spinlock_t *lock) diff --git a/sysdeps/tile/nptl/pthread_spin_trylock.c b/sysdeps/tile/nptl/pthread_spin_trylock.c index c2d6766fd1..6974a09983 100644 --- a/sysdeps/tile/nptl/pthread_spin_trylock.c +++ b/sysdeps/tile/nptl/pthread_spin_trylock.c @@ -19,11 +19,7 @@ #include "pthreadP.h" #include <errno.h> -#ifdef __tilegx__ #define TNS(p) __insn_exch4((p), 1) -#else -#define TNS(p) __insn_tns(p) -#endif int pthread_spin_trylock (pthread_spinlock_t *lock) diff --git a/sysdeps/tile/nptl/pthread_spin_unlock.c b/sysdeps/tile/nptl/pthread_spin_unlock.c index 63bc814985..be23478e39 100644 --- a/sysdeps/tile/nptl/pthread_spin_unlock.c +++ b/sysdeps/tile/nptl/pthread_spin_unlock.c @@ -22,12 +22,7 @@ int pthread_spin_unlock (pthread_spinlock_t *lock) { -#ifdef __tilegx__ /* Use exchange() to bypass the write buffer. */ atomic_exchange_rel (lock, 0); -#else - atomic_full_barrier (); - *lock = 0; -#endif return 0; } diff --git a/sysdeps/tile/nptl/tls.h b/sysdeps/tile/nptl/tls.h index 7314843cfb..5ddae18d15 100644 --- a/sysdeps/tile/nptl/tls.h +++ b/sysdeps/tile/nptl/tls.h @@ -113,13 +113,8 @@ register void *__thread_pointer asm ("tp"); - TLS_TCB_OFFSET - TLS_PRE_TCB_SIZE)) /* Magic for libthread_db to know how to do THREAD_SELF. */ -#ifdef __tilegx__ # define DB_THREAD_SELF \ - REGISTER (64, 64, REG_TP * 8, - TLS_TCB_OFFSET - TLS_PRE_TCB_SIZE) -#else -# define DB_THREAD_SELF \ - REGISTER (32, 32, REG_TP * 4, - TLS_TCB_OFFSET - TLS_PRE_TCB_SIZE) -#endif + REGISTER (64, 64, REG_TP * 8, - TLS_TCB_OFFSET - TLS_PRE_TCB_SIZE) /* Read member of the thread descriptor directly. */ # define THREAD_GETMEM(descr, member) (descr->member) diff --git a/sysdeps/tile/preconfigure b/sysdeps/tile/preconfigure index f3e5d7edae..d2165c301d 100644 --- a/sysdeps/tile/preconfigure +++ b/sysdeps/tile/preconfigure @@ -1,7 +1,5 @@ # This is a -*- sh -*- case "$machine" in - tilepro) - base_machine=tile machine=tile/tilepro ;; tilegx*) base_machine=tile if $CC $CFLAGS $CPPFLAGS -E -dM -xc /dev/null | grep -q __LP64__; then diff --git a/sysdeps/tile/stackguard-macros.h b/sysdeps/tile/stackguard-macros.h index f2e041b99b..c3ea0add70 100644 --- a/sysdeps/tile/stackguard-macros.h +++ b/sysdeps/tile/stackguard-macros.h @@ -1,20 +1,13 @@ #include <bits/wordsize.h> -#ifdef __tilegx__ -# if __WORDSIZE == 64 -# define STACK_CHK_GUARD \ +#if __WORDSIZE == 64 +# define STACK_CHK_GUARD \ ({ uintptr_t x; asm ("addi %0, tp, -16; ld %0, %0" : "=r" (x)); x; }) -# define POINTER_CHK_GUARD \ +# define POINTER_CHK_GUARD \ ({ uintptr_t x; asm ("addi %0, tp, -24; ld %0, %0" : "=r" (x)); x; }) -# else -# define STACK_CHK_GUARD \ - ({ uintptr_t x; asm ("addi %0, tp, -8; ld4s %0, %0" : "=r" (x)); x; }) -# define POINTER_CHK_GUARD \ - ({ uintptr_t x; asm ("addi %0, tp, -12; ld4s %0, %0" : "=r" (x)); x; }) -# endif #else # define STACK_CHK_GUARD \ - ({ uintptr_t x; asm ("addi %0, tp, -8; lw %0, %0" : "=r" (x)); x; }) + ({ uintptr_t x; asm ("addi %0, tp, -8; ld4s %0, %0" : "=r" (x)); x; }) # define POINTER_CHK_GUARD \ - ({ uintptr_t x; asm ("addi %0, tp, -12; lw %0, %0" : "=r" (x)); x; }) + ({ uintptr_t x; asm ("addi %0, tp, -12; ld4s %0, %0" : "=r" (x)); x; }) #endif diff --git a/sysdeps/tile/stackinfo.h b/sysdeps/tile/stackinfo.h index 0bef407941..d79c6b0570 100644 --- a/sysdeps/tile/stackinfo.h +++ b/sysdeps/tile/stackinfo.h @@ -35,7 +35,7 @@ (unnecessary) parameters. */ #define stackinfo_get_sp() \ ({ void *p__; asm volatile ("move %0, sp" : "=r" (p__)); p__; }) -#if defined __tilegx__ && __WORDSIZE == 32 +#if __WORDSIZE == 32 #define __stackinfo_sub "subx" #else #define __stackinfo_sub "sub" diff --git a/sysdeps/tile/start.S b/sysdeps/tile/start.S index c790c265e3..da3ba9c013 100644 --- a/sysdeps/tile/start.S +++ b/sysdeps/tile/start.S @@ -129,7 +129,6 @@ _start: /* Pass address of main() in r0, and of our own entry points to .fini and .init in r3 and r4. */ -#ifdef __tilegx__ moveli r0, hw2_last(main - .Lmy_pc) } { @@ -162,30 +161,12 @@ _start: } { ADD_PTR r4, r4, r13 -# ifdef NO_PLT_PCREL +#ifdef NO_PLT_PCREL j plt(__libc_start_main) -# else - jr r12 -# endif - } #else - addli r0, r13, lo16(main - .Lmy_pc) - } - { - auli r0, r0, ha16(main - .Lmy_pc) - addli r3, r13, lo16(__libc_csu_init - .Lmy_pc) - } - { - auli r3, r3, ha16(__libc_csu_init - .Lmy_pc) - addli r4, r13, lo16(__libc_csu_fini - .Lmy_pc) - } - { - auli r4, r4, ha16(__libc_csu_fini - .Lmy_pc) - /* Call the user's main function, and exit with its value. - But let the libc call main. */ - j plt(__libc_start_main) - } + jr r12 #endif + } { /* Tell backtracer to give up (_start has no caller). */ diff --git a/sysdeps/tile/sysdep.h b/sysdeps/tile/sysdep.h index cd94e7745b..0f4b64d9f7 100644 --- a/sysdeps/tile/sysdep.h +++ b/sysdeps/tile/sysdep.h @@ -60,34 +60,11 @@ #define L(name) .L##name /* Specify the size in bytes of a machine register. */ -#ifdef __tilegx__ #define REGSIZE 8 -#else -#define REGSIZE 4 -#endif - -/* Support a limited form of shared assembly between tilepro and tilegx. - The presumption is that LD/ST are used for manipulating registers. - Since opcode parsing is case-insensitive, we don't need to provide - definitions for these on tilegx. */ -#ifndef __tilegx__ -#define LD lw -#define LD4U lw -#define ST sw -#define ST4 sw -#define BNEZ bnz -#define BEQZ bz -#define BEQZT bzt -#define BGTZ bgz -#define CMPEQI seqi -#define CMPEQ seq -#define CMOVEQZ mvz -#define CMOVNEZ mvnz -#endif /* Provide "pointer-oriented" instruction variants. These differ not just for tilepro vs tilegx, but also for tilegx -m64 vs -m32. */ -#if defined __tilegx__ && __WORDSIZE == 32 +#if __WORDSIZE == 32 #define ADD_PTR addx #define ADDI_PTR addxi #define ADDLI_PTR addxli @@ -100,11 +77,7 @@ #define ADDLI_PTR addli #define LD_PTR LD #define ST_PTR ST -#ifdef __tilegx__ #define SHL_PTR_ADD shl3add -#else -#define SHL_PTR_ADD s2a -#endif #endif #endif /* __ASSEMBLER__ */ diff --git a/sysdeps/tile/tilepro/Implies b/sysdeps/tile/tilepro/Implies deleted file mode 100644 index 709e1dc122..0000000000 --- a/sysdeps/tile/tilepro/Implies +++ /dev/null @@ -1,2 +0,0 @@ -tile -wordsize-32 diff --git a/sysdeps/tile/tilepro/atomic-machine.h b/sysdeps/tile/tilepro/atomic-machine.h deleted file mode 100644 index 45e36de1e4..0000000000 --- a/sysdeps/tile/tilepro/atomic-machine.h +++ /dev/null @@ -1,99 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#ifndef _ATOMIC_MACHINE_H -#define _ATOMIC_MACHINE_H 1 - -#include <asm/unistd.h> - -#define __HAVE_64B_ATOMICS 0 -#define USE_ATOMIC_COMPILER_BUILTINS 0 -#define ATOMIC_EXCHANGE_USES_CAS 0 - -/* 32-bit integer compare-and-exchange. */ -static __inline __attribute__ ((always_inline)) -int __atomic_cmpxchg_32 (volatile int *mem, int newval, int oldval) -{ - int result; - __asm__ __volatile__ ("swint1" - : "=R00" (result), "=m" (*mem) - : "R10" (__NR_FAST_cmpxchg), "R00" (mem), - "R01" (oldval), "R02" (newval), "m" (*mem) - : "r20", "r21", "r22", "r23", "r24", - "r25", "r26", "r27", "r28", "r29", "memory"); - return result; -} - -#define atomic_compare_and_exchange_val_acq(mem, n, o) \ - ({ \ - if (sizeof (*(mem)) != 4) \ - __atomic_error_bad_argument_size (); \ - (__typeof (*(mem))) \ - __atomic_cmpxchg_32 ((int *) (mem), (int) (n), (int) (o)); \ - }) - -/* Atomically compute: - int old = *ptr; - *ptr = (old & mask) + addend; - return old; */ - -static __inline __attribute__ ((always_inline)) -int __atomic_update_32 (volatile int *mem, int mask, int addend) -{ - int result; - __asm__ __volatile__ ("swint1" - : "=R00" (result), "=m" (*mem) - : "R10" (__NR_FAST_atomic_update), "R00" (mem), - "R01" (mask), "R02" (addend), "m" (*mem) - : "r20", "r21", "r22", "r23", "r24", - "r25", "r26", "r27", "r28", "r29", "memory"); - return result; -} - -/* Size-checked verson of __atomic_update_32. */ -#define __atomic_update(mem, mask, addend) \ - ({ \ - if (sizeof (*(mem)) != 4) \ - __atomic_error_bad_argument_size (); \ - (__typeof (*(mem))) \ - __atomic_update_32 ((int *) (mem), (int) (mask), (int) (addend)); \ - }) - -#define atomic_exchange_acq(mem, newvalue) \ - __atomic_update ((mem), 0, (newvalue)) -#define atomic_exchange_and_add(mem, value) \ - __atomic_update ((mem), -1, (value)) -#define atomic_and_val(mem, mask) \ - __atomic_update ((mem), (mask), 0) -#define atomic_or_val(mem, mask) \ - ({ __typeof (mask) __att1_v = (mask); \ - __atomic_update ((mem), ~__att1_v, __att1_v); }) - -/* - * We must use the kernel atomics for atomic_store, since otherwise an - * unsynchronized store could become visible after another core's - * kernel-atomic implementation had read the memory word in question, - * but before it had written the updated value to it, which would - * cause the unsynchronized store to be lost. - */ -#define atomic_store_relaxed(mem, val) atomic_exchange_acq (mem, val) -#define atomic_store_release(mem, val) atomic_exchange_rel (mem, val) - -#include <sysdeps/tile/atomic-machine.h> - -#endif /* atomic-machine.h */ diff --git a/sysdeps/tile/tilepro/bits/wordsize.h b/sysdeps/tile/tilepro/bits/wordsize.h deleted file mode 100644 index d2c4a354a7..0000000000 --- a/sysdeps/tile/tilepro/bits/wordsize.h +++ /dev/null @@ -1,6 +0,0 @@ -/* Determine the wordsize from the preprocessor defines. */ - -#define __WORDSIZE 32 -#define __WORDSIZE_TIME64_COMPAT32 0 -#define __WORDSIZE32_SIZE_ULONG 0 -#define __WORDSIZE32_PTRDIFF_LONG 0 diff --git a/sysdeps/tile/tilepro/memchr.c b/sysdeps/tile/tilepro/memchr.c deleted file mode 100644 index fba1f70c2c..0000000000 --- a/sysdeps/tile/tilepro/memchr.c +++ /dev/null @@ -1,76 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#include <string.h> -#include <stdint.h> - -void * -__memchr (const void *s, int c, size_t n) -{ - const uint32_t *last_word_ptr; - const uint32_t *p; - const char *last_byte_ptr; - uintptr_t s_int; - uint32_t goal, before_mask, v, bits; - char *ret; - - if (__builtin_expect (n == 0, 0)) - { - /* Don't dereference any memory if the array is empty. */ - return NULL; - } - - /* Get an aligned pointer. */ - s_int = (uintptr_t) s; - p = (const uint32_t *) (s_int & -4); - - /* Create four copies of the byte for which we are looking. */ - goal = 0x01010101 * (uint8_t) c; - - /* Read the first word, but munge it so that bytes before the array - will not match goal. Note that this shift count expression works - because we know shift counts are taken mod 32. */ - before_mask = (1 << (s_int << 3)) - 1; - v = (*p | before_mask) ^ (goal & before_mask); - - /* Compute the address of the last byte. */ - last_byte_ptr = (const char *) s + n - 1; - - /* Handle possible addition overflow. */ - if (__glibc_unlikely ((uintptr_t) last_byte_ptr < (uintptr_t) s)) - last_byte_ptr = (const char *) UINTPTR_MAX; - - /* Compute the address of the word containing the last byte. */ - last_word_ptr = (const uint32_t *) ((uintptr_t) last_byte_ptr & -4); - - while ((bits = __insn_seqb (v, goal)) == 0) - { - if (__builtin_expect (p == last_word_ptr, 0)) - { - /* We already read the last word in the array, so give up. */ - return NULL; - } - v = *++p; - } - - /* We found a match, but it might be in a byte past the end of the array. */ - ret = ((char *) p) + (__insn_ctz (bits) >> 3); - return (ret <= last_byte_ptr) ? ret : NULL; -} -weak_alias (__memchr, memchr) -libc_hidden_builtin_def (memchr) diff --git a/sysdeps/tile/tilepro/memcpy.S b/sysdeps/tile/tilepro/memcpy.S deleted file mode 100644 index 0f73ccbc2f..0000000000 --- a/sysdeps/tile/tilepro/memcpy.S +++ /dev/null @@ -1,397 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#include <arch/chip.h> -#include <sysdep.h> - - .text -ENTRY (__memcpy) - FEEDBACK_ENTER(__memcpy) - - /* r0 is the dest, r1 is the source, r2 is the size. */ - - /* Save aside original dest so we can return it at the end. */ - { sw sp, lr; move r23, r0; or r4, r0, r1 } - cfi_offset (lr, 0) - - /* Check for an empty size. */ - { bz r2, .Ldone; andi r4, r4, 3 } - - /* Check for an unaligned source or dest. */ - { bnz r4, .Lcopy_unaligned_maybe_many; addli r4, r2, -256 } - -.Lcheck_aligned_copy_size: - /* If we are copying < 256 bytes, branch to simple case. */ - { blzt r4, .Lcopy_8_check; slti_u r8, r2, 8 } - - /* Copying >= 256 bytes, so jump to complex prefetching loop. */ - { andi r6, r1, 63; j .Lcopy_many } - -/* Aligned 4 byte at a time copy loop. */ - -.Lcopy_8_loop: - /* Copy two words at a time to hide load latency. */ - { lw r3, r1; addi r1, r1, 4; slti_u r8, r2, 16 } - { lw r4, r1; addi r1, r1, 4 } - { sw r0, r3; addi r0, r0, 4; addi r2, r2, -4 } - { sw r0, r4; addi r0, r0, 4; addi r2, r2, -4 } -.Lcopy_8_check: - { bzt r8, .Lcopy_8_loop; slti_u r4, r2, 4 } - - /* Copy odd leftover word, if any. */ - { bnzt r4, .Lcheck_odd_stragglers } - { lw r3, r1; addi r1, r1, 4 } - { sw r0, r3; addi r0, r0, 4; addi r2, r2, -4 } - -.Lcheck_odd_stragglers: - { bnz r2, .Lcopy_unaligned_few } - -.Ldone: - { move r0, r23; jrp lr } - -/* Prefetching multiple cache line copy handler (for large transfers). */ - - /* Copy words until r1 is cache-line-aligned. */ -.Lalign_loop: - { lw r3, r1; addi r1, r1, 4 } - { andi r6, r1, 63 } - { sw r0, r3; addi r0, r0, 4; addi r2, r2, -4 } -.Lcopy_many: - { bnzt r6, .Lalign_loop; addi r9, r0, 63 } - - { addi r3, r1, 60; andi r9, r9, -64 } - - /* No need to prefetch dst, we'll just do the wh64 - right before we copy a line. */ - { lw r5, r3; addi r3, r3, 64; movei r4, 1 } - /* Intentionally stall for a few cycles to leave L2 cache alone. */ - { bnzt zero, .; move r27, lr } - { lw r6, r3; addi r3, r3, 64 } - /* Intentionally stall for a few cycles to leave L2 cache alone. */ - { bnzt zero, . } - { lw r7, r3; addi r3, r3, 64 } - /* Intentionally stall for a few cycles to leave L2 cache alone. */ - { bz zero, .Lbig_loop2 } - - /* On entry to this loop: - - r0 points to the start of dst line 0 - - r1 points to start of src line 0 - - r2 >= (256 - 60), only the first time the loop trips. - - r3 contains r1 + 128 + 60 [pointer to end of source line 2] - This is our prefetch address. When we get near the end - rather than prefetching off the end this is changed to point - to some "safe" recently loaded address. - - r5 contains *(r1 + 60) [i.e. last word of source line 0] - - r6 contains *(r1 + 64 + 60) [i.e. last word of source line 1] - - r9 contains ((r0 + 63) & -64) - [start of next dst cache line.] */ - -.Lbig_loop: - { jal .Lcopy_line2; add r15, r1, r2 } - -.Lbig_loop2: - /* Copy line 0, first stalling until r5 is ready. */ - { move r12, r5; lw r16, r1 } - { bz r4, .Lcopy_8_check; slti_u r8, r2, 8 } - /* Prefetch several lines ahead. */ - { lw r5, r3; addi r3, r3, 64 } - { jal .Lcopy_line } - - /* Copy line 1, first stalling until r6 is ready. */ - { move r12, r6; lw r16, r1 } - { bz r4, .Lcopy_8_check; slti_u r8, r2, 8 } - /* Prefetch several lines ahead. */ - { lw r6, r3; addi r3, r3, 64 } - { jal .Lcopy_line } - - /* Copy line 2, first stalling until r7 is ready. */ - { move r12, r7; lw r16, r1 } - { bz r4, .Lcopy_8_check; slti_u r8, r2, 8 } - /* Prefetch several lines ahead. */ - { lw r7, r3; addi r3, r3, 64 } - /* Use up a caches-busy cycle by jumping back to the top of the - loop. Might as well get it out of the way now. */ - { j .Lbig_loop } - - - /* On entry: - - r0 points to the destination line. - - r1 points to the source line. - - r3 is the next prefetch address. - - r9 holds the last address used for wh64. - - r12 = WORD_15 - - r16 = WORD_0. - - r17 == r1 + 16. - - r27 holds saved lr to restore. - - On exit: - - r0 is incremented by 64. - - r1 is incremented by 64, unless that would point to a word - beyond the end of the source array, in which case it is redirected - to point to an arbitrary word already in the cache. - - r2 is decremented by 64. - - r3 is unchanged, unless it points to a word beyond the - end of the source array, in which case it is redirected - to point to an arbitrary word already in the cache. - Redirecting is OK since if we are that close to the end - of the array we will not come back to this subroutine - and use the contents of the prefetched address. - - r4 is nonzero iff r2 >= 64. - - r9 is incremented by 64, unless it points beyond the - end of the last full destination cache line, in which - case it is redirected to a "safe address" that can be - clobbered (sp - 64) - - lr contains the value in r27. */ - -/* r26 unused */ - -.Lcopy_line: - /* TODO: when r3 goes past the end, we would like to redirect it - to prefetch the last partial cache line (if any) just once, for the - benefit of the final cleanup loop. But we don't want to - prefetch that line more than once, or subsequent prefetches - will go into the RTF. But then .Lbig_loop should unconditionally - branch to top of loop to execute final prefetch, and its - nop should become a conditional branch. */ - - /* We need two non-memory cycles here to cover the resources - used by the loads initiated by the caller. */ - { add r15, r1, r2 } -.Lcopy_line2: - { slt_u r13, r3, r15; addi r17, r1, 16 } - - /* NOTE: this will stall for one cycle as L1 is busy. */ - - /* Fill second L1D line. */ - { lw r17, r17; addi r1, r1, 48; mvz r3, r13, r1 } /* r17 = WORD_4 */ - - /* Prepare destination line for writing. */ - { wh64 r9; addi r9, r9, 64 } - - /* Load seven words that are L1D hits to cover wh64 L2 usage. */ - - /* Load the three remaining words from the last L1D line, which - we know has already filled the L1D. */ - { lw r4, r1; addi r1, r1, 4; addi r20, r1, 16 } /* r4 = WORD_12 */ - { lw r8, r1; addi r1, r1, 4; slt_u r13, r20, r15 }/* r8 = WORD_13 */ - { lw r11, r1; addi r1, r1, -52; mvz r20, r13, r1 } /* r11 = WORD_14 */ - - /* Load the three remaining words from the first L1D line, first - stalling until it has filled by "looking at" r16. */ - { lw r13, r1; addi r1, r1, 4; move zero, r16 } /* r13 = WORD_1 */ - { lw r14, r1; addi r1, r1, 4 } /* r14 = WORD_2 */ - { lw r15, r1; addi r1, r1, 8; addi r10, r0, 60 } /* r15 = WORD_3 */ - - /* Load second word from the second L1D line, first - stalling until it has filled by "looking at" r17. */ - { lw r19, r1; addi r1, r1, 4; move zero, r17 } /* r19 = WORD_5 */ - - /* Store last word to the destination line, potentially dirtying it - for the first time, which keeps the L2 busy for two cycles. */ - { sw r10, r12 } /* store(WORD_15) */ - - /* Use two L1D hits to cover the sw L2 access above. */ - { lw r10, r1; addi r1, r1, 4 } /* r10 = WORD_6 */ - { lw r12, r1; addi r1, r1, 4 } /* r12 = WORD_7 */ - - /* Fill third L1D line. */ - { lw r18, r1; addi r1, r1, 4 } /* r18 = WORD_8 */ - - /* Store first L1D line. */ - { sw r0, r16; addi r0, r0, 4; add r16, r0, r2 } /* store(WORD_0) */ - { sw r0, r13; addi r0, r0, 4; andi r16, r16, -64 } /* store(WORD_1) */ - { sw r0, r14; addi r0, r0, 4; slt_u r16, r9, r16 } /* store(WORD_2) */ - { sw r0, r15; addi r0, r0, 4; addi r13, sp, -64 } /* store(WORD_3) */ - - /* Store second L1D line. */ - { sw r0, r17; addi r0, r0, 4; mvz r9, r16, r13 }/* store(WORD_4) */ - { sw r0, r19; addi r0, r0, 4 } /* store(WORD_5) */ - { sw r0, r10; addi r0, r0, 4 } /* store(WORD_6) */ - { sw r0, r12; addi r0, r0, 4 } /* store(WORD_7) */ - - { lw r13, r1; addi r1, r1, 4; move zero, r18 } /* r13 = WORD_9 */ - { lw r14, r1; addi r1, r1, 4 } /* r14 = WORD_10 */ - { lw r15, r1; move r1, r20 } /* r15 = WORD_11 */ - - /* Store third L1D line. */ - { sw r0, r18; addi r0, r0, 4 } /* store(WORD_8) */ - { sw r0, r13; addi r0, r0, 4 } /* store(WORD_9) */ - { sw r0, r14; addi r0, r0, 4 } /* store(WORD_10) */ - { sw r0, r15; addi r0, r0, 4 } /* store(WORD_11) */ - - /* Store rest of fourth L1D line. */ - { sw r0, r4; addi r0, r0, 4 } /* store(WORD_12) */ - { - sw r0, r8 /* store(WORD_13) */ - addi r0, r0, 4 - /* Will r2 be > 64 after we subtract 64 below? */ - shri r4, r2, 7 - } - { - sw r0, r11 /* store(WORD_14) */ - addi r0, r0, 8 - /* Record 64 bytes successfully copied. */ - addi r2, r2, -64 - } - - { jrp lr; move lr, r27 } - - /* Convey to the backtrace library that the stack frame is - size zero, and the real return address is on the stack - rather than in 'lr'. */ - { info 8 } - - .align 64 -.Lcopy_unaligned_maybe_many: - /* Skip the setup overhead if we aren't copying many bytes. */ - { slti_u r8, r2, 20; sub r4, zero, r0 } - { bnzt r8, .Lcopy_unaligned_few; andi r4, r4, 3 } - { bz r4, .Ldest_is_word_aligned; add r18, r1, r2 } - -/* Unaligned 4 byte at a time copy handler. */ - - /* Copy single bytes until r0 == 0 mod 4, so we can store words. */ -.Lalign_dest_loop: - { lb_u r3, r1; addi r1, r1, 1; addi r4, r4, -1 } - { sb r0, r3; addi r0, r0, 1; addi r2, r2, -1 } - { bnzt r4, .Lalign_dest_loop; andi r3, r1, 3 } - - /* If source and dest are now *both* aligned, do an aligned copy. */ - { bz r3, .Lcheck_aligned_copy_size; addli r4, r2, -256 } - -.Ldest_is_word_aligned: - - { andi r8, r0, 63; lwadd_na r6, r1, 4} - { slti_u r9, r2, 64; bz r8, .Ldest_is_L2_line_aligned } - - /* This copies unaligned words until either there are fewer - than 4 bytes left to copy, or until the destination pointer - is cache-aligned, whichever comes first. - - On entry: - - r0 is the next store address. - - r1 points 4 bytes past the load address corresponding to r0. - - r2 >= 4 - - r6 is the next aligned word loaded. */ -.Lcopy_unaligned_src_words: - { lwadd_na r7, r1, 4; slti_u r8, r2, 4 + 4 } - /* stall */ - { dword_align r6, r7, r1; slti_u r9, r2, 64 + 4 } - { swadd r0, r6, 4; addi r2, r2, -4 } - { bnz r8, .Lcleanup_unaligned_words; andi r8, r0, 63 } - { bnzt r8, .Lcopy_unaligned_src_words; move r6, r7 } - - /* On entry: - - r0 is the next store address. - - r1 points 4 bytes past the load address corresponding to r0. - - r2 >= 4 (# of bytes left to store). - - r6 is the next aligned src word value. - - r9 = (r2 < 64U). - - r18 points one byte past the end of source memory. */ -.Ldest_is_L2_line_aligned: - - { - /* Not a full cache line remains. */ - bnz r9, .Lcleanup_unaligned_words - move r7, r6 - } - - /* r2 >= 64 */ - - /* Kick off two prefetches, but don't go past the end. */ - { addi r3, r1, 63 - 4; addi r8, r1, 64 + 63 - 4 } - { prefetch r3; move r3, r8; slt_u r8, r8, r18 } - { mvz r3, r8, r1; addi r8, r3, 64 } - { prefetch r3; move r3, r8; slt_u r8, r8, r18 } - { mvz r3, r8, r1; movei r17, 0 } - -.Lcopy_unaligned_line: - /* Prefetch another line. */ - { prefetch r3; addi r15, r1, 60; addi r3, r3, 64 } - /* Fire off a load of the last word we are about to copy. */ - { lw_na r15, r15; slt_u r8, r3, r18 } - - { mvz r3, r8, r1; wh64 r0 } - - /* This loop runs twice. - - On entry: - - r17 is even before the first iteration, and odd before - the second. It is incremented inside the loop. Encountering - an even value at the end of the loop makes it stop. */ -.Lcopy_half_an_unaligned_line: - { - /* Stall until the last byte is ready. In the steady state this - guarantees all words to load below will be in the L2 cache, which - avoids shunting the loads to the RTF. */ - move zero, r15 - lwadd_na r7, r1, 16 - } - { lwadd_na r11, r1, 12 } - { lwadd_na r14, r1, -24 } - { lwadd_na r8, r1, 4 } - { lwadd_na r9, r1, 4 } - { - lwadd_na r10, r1, 8 - /* r16 = (r2 < 64), after we subtract 32 from r2 below. */ - slti_u r16, r2, 64 + 32 - } - { lwadd_na r12, r1, 4; addi r17, r17, 1 } - { lwadd_na r13, r1, 8; dword_align r6, r7, r1 } - { swadd r0, r6, 4; dword_align r7, r8, r1 } - { swadd r0, r7, 4; dword_align r8, r9, r1 } - { swadd r0, r8, 4; dword_align r9, r10, r1 } - { swadd r0, r9, 4; dword_align r10, r11, r1 } - { swadd r0, r10, 4; dword_align r11, r12, r1 } - { swadd r0, r11, 4; dword_align r12, r13, r1 } - { swadd r0, r12, 4; dword_align r13, r14, r1 } - { swadd r0, r13, 4; addi r2, r2, -32 } - { move r6, r14; bbst r17, .Lcopy_half_an_unaligned_line } - - { bzt r16, .Lcopy_unaligned_line; move r7, r6 } - - /* On entry: - - r0 is the next store address. - - r1 points 4 bytes past the load address corresponding to r0. - - r2 >= 0 (# of bytes left to store). - - r7 is the next aligned src word value. */ -.Lcleanup_unaligned_words: - /* Handle any trailing bytes. */ - { bz r2, .Lcopy_unaligned_done; slti_u r8, r2, 4 } - { bzt r8, .Lcopy_unaligned_src_words; move r6, r7 } - - /* Move r1 back to the point where it corresponds to r0. */ - { addi r1, r1, -4 } - - /* Fall through */ - -/* 1 byte at a time copy handler. */ - -.Lcopy_unaligned_few: - { lb_u r3, r1; addi r1, r1, 1 } - { sb r0, r3; addi r0, r0, 1; addi r2, r2, -1 } - { bnzt r2, .Lcopy_unaligned_few } - -.Lcopy_unaligned_done: - - { move r0, r23; jrp lr } - -END (__memcpy) - -weak_alias (__memcpy, memcpy) -libc_hidden_builtin_def (memcpy) diff --git a/sysdeps/tile/tilepro/memset.c b/sysdeps/tile/tilepro/memset.c deleted file mode 100644 index aaeaacaacc..0000000000 --- a/sysdeps/tile/tilepro/memset.c +++ /dev/null @@ -1,151 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#include <string.h> -#include <stdint.h> -#include <arch/chip.h> - -void * inhibit_loop_to_libcall -__memset (void *s, int c, size_t n) -{ - uint32_t *out32; - int n32; - uint32_t v16, v32; - uint8_t *out8 = s; - int to_align32; - - /* Experimentation shows that a trivial tight loop is a win up until - around a size of 20, where writing a word at a time starts to win. */ -#define BYTE_CUTOFF 20 - -#if BYTE_CUTOFF < 3 - /* This must be at least at least this big, or some code later - on doesn't work. */ -# error "BYTE_CUTOFF is too small." -#endif - - if (n < BYTE_CUTOFF) - { - /* Strangely, this turns out to be the tightest way to write - this loop. */ - if (n != 0) - { - do - { - /* Strangely, combining these into one line performs worse. */ - *out8 = c; - out8++; - } - while (--n != 0); - } - - return s; - } - - /* Align 'out8'. We know n >= 3 so this won't write past the end. */ - while (((uintptr_t) out8 & 3) != 0) - { - *out8++ = c; - --n; - } - - /* Align 'n'. */ - while (n & 3) - out8[--n] = c; - - out32 = (uint32_t *) out8; - n32 = n >> 2; - - /* Tile input byte out to 32 bits. */ - v16 = __insn_intlb (c, c); - v32 = __insn_intlh (v16, v16); - - /* This must be at least 8 or the following loop doesn't work. */ -#define CACHE_LINE_SIZE_IN_WORDS (CHIP_L2_LINE_SIZE() / 4) - - /* Determine how many words we need to emit before the 'out32' - pointer becomes aligned modulo the cache line size. */ - to_align32 = (-((uintptr_t) out32 >> 2)) & (CACHE_LINE_SIZE_IN_WORDS - 1); - - /* Only bother aligning and using wh64 if there is at least one full - cache line to process. This check also prevents overrunning the - end of the buffer with alignment words. */ - if (to_align32 <= n32 - CACHE_LINE_SIZE_IN_WORDS) - { - int lines_left; - - /* Align out32 mod the cache line size so we can use wh64. */ - n32 -= to_align32; - for (; to_align32 != 0; to_align32--) - { - *out32 = v32; - out32++; - } - - /* Use unsigned divide to turn this into a right shift. */ - lines_left = (unsigned) n32 / CACHE_LINE_SIZE_IN_WORDS; - - do - { - /* Only wh64 a few lines at a time, so we don't exceed the - maximum number of victim lines. */ - int x = ((lines_left < CHIP_MAX_OUTSTANDING_VICTIMS ())? lines_left - : CHIP_MAX_OUTSTANDING_VICTIMS ()); - uint32_t *wh = out32; - int i = x; - int j; - - lines_left -= x; - - do - { - __insn_wh64 (wh); - wh += CACHE_LINE_SIZE_IN_WORDS; - } - while (--i); - - for (j = x * (CACHE_LINE_SIZE_IN_WORDS / 4); j != 0; j--) - { - *out32++ = v32; - *out32++ = v32; - *out32++ = v32; - *out32++ = v32; - } - } - while (lines_left != 0); - - /* We processed all full lines above, so only this many words - remain to be processed. */ - n32 &= CACHE_LINE_SIZE_IN_WORDS - 1; - } - - /* Now handle any leftover values. */ - if (n32 != 0) - { - do - { - *out32 = v32; - out32++; - } - while (--n32 != 0); - } - - return s; -} -weak_alias (__memset, memset) -libc_hidden_builtin_def (memset) diff --git a/sysdeps/tile/tilepro/memusage.h b/sysdeps/tile/tilepro/memusage.h deleted file mode 100644 index 9ed4600f13..0000000000 --- a/sysdeps/tile/tilepro/memusage.h +++ /dev/null @@ -1,29 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#include <arch/spr_def.h> - -#define GETSP() ({ register uintptr_t stack_ptr asm ("sp"); stack_ptr; }) - -#define GETTIME(low,high) \ - { \ - low = __insn_mfspr (SPR_CYCLE_LOW); \ - high = __insn_mfspr (SPR_CYCLE_HIGH); \ - } - -#include <sysdeps/generic/memusage.h> diff --git a/sysdeps/tile/tilepro/rawmemchr.c b/sysdeps/tile/tilepro/rawmemchr.c deleted file mode 100644 index a5e714407d..0000000000 --- a/sysdeps/tile/tilepro/rawmemchr.c +++ /dev/null @@ -1,45 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#include <string.h> -#include <stdint.h> - -void * -__rawmemchr (const void *s, int c) -{ - /* Get an aligned pointer. */ - const uintptr_t s_int = (uintptr_t) s; - const uint32_t *p = (const uint32_t *) (s_int & -4); - - /* Create four copies of the byte for which we are looking. */ - const uint32_t goal = 0x01010101 * (uint8_t) c; - - /* Read the first word, but munge it so that bytes before the array - will not match goal. Note that this shift count expression works - because we know shift counts are taken mod 32. */ - const uint32_t before_mask = (1 << (s_int << 3)) - 1; - uint32_t v = (*p | before_mask) ^ (goal & before_mask); - - uint32_t bits; - while ((bits = __insn_seqb (v, goal)) == 0) - v = *++p; - - return ((char *) p) + (__insn_ctz (bits) >> 3); -} -libc_hidden_def (__rawmemchr) -weak_alias (__rawmemchr, rawmemchr) diff --git a/sysdeps/tile/tilepro/strchr.c b/sysdeps/tile/tilepro/strchr.c deleted file mode 100644 index 11f8cda418..0000000000 --- a/sysdeps/tile/tilepro/strchr.c +++ /dev/null @@ -1,68 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#include <string.h> -#include <stdint.h> - -#undef strchr - -char * -strchr (const char *s, int c) -{ - int z, g; - - /* Get an aligned pointer. */ - const uintptr_t s_int = (uintptr_t) s; - const uint32_t *p = (const uint32_t *) (s_int & -4); - - /* Create four copies of the byte for which we are looking. */ - const uint32_t goal = 0x01010101 * (uint8_t) c; - - /* Read the first aligned word, but force bytes before the string to - match neither zero nor goal (we make sure the high bit of each - byte is 1, and the low 7 bits are all the opposite of the goal - byte). Note that this shift count expression works because we - know shift counts are taken mod 32. */ - const uint32_t before_mask = (1 << (s_int << 3)) - 1; - uint32_t v = (*p | before_mask) ^ (goal & __insn_shrib (before_mask, 1)); - - uint32_t zero_matches, goal_matches; - while (1) - { - /* Look for a terminating '\0'. */ - zero_matches = __insn_seqb (v, 0); - - /* Look for the goal byte. */ - goal_matches = __insn_seqb (v, goal); - - if (__builtin_expect ((zero_matches | goal_matches) != 0, 0)) - break; - - v = *++p; - } - - z = __insn_ctz (zero_matches); - g = __insn_ctz (goal_matches); - - /* If we found c before '\0' we got a match. Note that if c == '\0' - then g == z, and we correctly return the address of the '\0' - rather than NULL. */ - return (g <= z) ? ((char *) p) + (g >> 3) : NULL; -} -weak_alias (strchr, index) -libc_hidden_builtin_def (strchr) diff --git a/sysdeps/tile/tilepro/strchrnul.c b/sysdeps/tile/tilepro/strchrnul.c deleted file mode 100644 index 3220ad7970..0000000000 --- a/sysdeps/tile/tilepro/strchrnul.c +++ /dev/null @@ -1,65 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#include <string.h> -#include <stdint.h> - -char * -__strchrnul (const char *s, int c) -{ - int z, g; - - /* Get an aligned pointer. */ - const uintptr_t s_int = (uintptr_t) s; - const uint32_t *p = (const uint32_t *) (s_int & -4); - - /* Create four copies of the byte for which we are looking. */ - const uint32_t goal = 0x01010101 * (uint8_t) c; - - /* Read the first aligned word, but force bytes before the string to - match neither zero nor goal (we make sure the high bit of each - byte is 1, and the low 7 bits are all the opposite of the goal - byte). Note that this shift count expression works because we - know shift counts are taken mod 32. */ - const uint32_t before_mask = (1 << (s_int << 3)) - 1; - uint32_t v = (*p | before_mask) ^ (goal & __insn_shrib (before_mask, 1)); - - uint32_t zero_matches, goal_matches; - while (1) - { - /* Look for a terminating '\0'. */ - zero_matches = __insn_seqb (v, 0); - - /* Look for the goal byte. */ - goal_matches = __insn_seqb (v, goal); - - if (__builtin_expect ((zero_matches | goal_matches) != 0, 0)) - break; - - v = *++p; - } - - z = __insn_ctz (zero_matches); - g = __insn_ctz (goal_matches); - - /* Return a pointer to the NUL or goal, whichever is first. */ - if (z < g) - g = z; - return ((char *) p) + (g >> 3); -} -weak_alias (__strchrnul, strchrnul) diff --git a/sysdeps/tile/tilepro/strlen.c b/sysdeps/tile/tilepro/strlen.c deleted file mode 100644 index adad3d80e4..0000000000 --- a/sysdeps/tile/tilepro/strlen.c +++ /dev/null @@ -1,39 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#include <string.h> -#include <stdint.h> - -size_t -strlen (const char *s) -{ - /* Get an aligned pointer. */ - const uintptr_t s_int = (uintptr_t) s; - const uint32_t *p = (const uint32_t *) (s_int & -4); - - /* Read the first word, but force bytes before the string to be nonzero. - This expression works because we know shift counts are taken mod 32. */ - uint32_t v = *p | ((1 << (s_int << 3)) - 1); - - uint32_t bits; - while ((bits = __insn_seqb (v, 0)) == 0) - v = *++p; - - return ((const char *) p) + (__insn_ctz (bits) >> 3) - s; -} -libc_hidden_builtin_def (strlen) diff --git a/sysdeps/tile/tilepro/strrchr.c b/sysdeps/tile/tilepro/strrchr.c deleted file mode 100644 index b30d259773..0000000000 --- a/sysdeps/tile/tilepro/strrchr.c +++ /dev/null @@ -1,73 +0,0 @@ -/* Copyright (C) 2011-2017 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - <http://www.gnu.org/licenses/>. */ - -#include <string.h> -#include <stdint.h> - -char * -strrchr (const char *s, int c) -{ - /* Get an aligned pointer. */ - const uintptr_t s_int = (uintptr_t) s; - const uint32_t *p = (const uint32_t *) (s_int & -4); - - /* Create four copies of the byte for which we are looking. */ - const uint32_t goal = 0x01010101 * (uint8_t) c; - - /* Read the first aligned word, but force bytes before the string to - match neither zero nor goal (we make sure the high bit of each - byte is 1, and the low 7 bits are all the opposite of the goal - byte). Note that this shift count expression works because we - know shift counts are taken mod 32. */ - const uint32_t before_mask = (1 << (s_int << 3)) - 1; - uint32_t v = (*p | before_mask) ^ (goal & __insn_shrib (before_mask, 1)); - const char *found = NULL; - uint32_t zero_matches, goal_matches; - while (1) - { - /* Look for a terminating '\0'. */ - zero_matches = __insn_seqb (v, 0); - - /* Look for the goal byte. */ - goal_matches = __insn_seqb (v, goal); - - /* If we found the goal, record the last offset. */ - if (__builtin_expect (goal_matches != 0, 0)) - { - if (__builtin_expect (zero_matches != 0, 0)) - { - /* Clear any goal after the first zero. */ - int first_nul = __insn_ctz (zero_matches); - /* The number of top bits we need to clear is - 32 - (first_nul + 8). */ - int shift_amt = (24 - first_nul); - goal_matches <<= shift_amt; - goal_matches >>= shift_amt; - } - if (__builtin_expect (goal_matches != 0, 1)) - found = ((char *) p) + 3 - (__insn_clz (goal_matches) >> 3); - } - - if (__builtin_expect (zero_matches != 0, 0)) - return (char *) found; - - v = *++p; - } -} -weak_alias (strrchr, rindex) -libc_hidden_builtin_def (strrchr) diff --git a/sysdeps/tile/tls-macros.h b/sysdeps/tile/tls-macros.h index 878358a731..47689b38f9 100644 --- a/sysdeps/tile/tls-macros.h +++ b/sysdeps/tile/tls-macros.h @@ -16,16 +16,10 @@ License along with the GNU C Library. If not, see <http://www.gnu.org/licenses/>. */ -#ifdef __tilegx__ #define TLS_GD_OFFSET(x) \ "moveli r0, hw1_last_tls_gd(" #x ")\n\t" \ "shl16insli r0, r0, hw0_tls_gd(" #x ")\n\t" \ "addi r0, %1, tls_add(" #x ")\n\t" -#else -#define TLS_GD_OFFSET(x) \ - "auli r0, %1, tls_gd_ha16(" #x ")\n\t" \ - "addli r0, r0, tls_gd_lo16(" #x ")\n\t" -#endif #define TLS_GD(x) \ ({ \ @@ -42,18 +36,11 @@ /* No special support for LD mode. */ #define TLS_LD TLS_GD -#ifdef __tilegx__ #define TLS_IE_OFFSET(x) \ "moveli %0, hw1_last_tls_ie(" #x ")\n\t" \ "shl16insli %0, %0, hw0_tls_ie(" #x ")\n\t" \ "addi %0, %1, tls_add(" #x ")\n\t" #define LD_TLS "ld_tls" -#else -#define TLS_IE_OFFSET(x) \ - "auli %0, %1, tls_ie_ha16(" #x ")\n\t" \ - "addli %0, %0, tls_ie_lo16(" #x ")\n\t" -#define LD_TLS "lw_tls" -#endif #define TLS_IE(x) \ ({ \ @@ -66,16 +53,10 @@ "=&r" (__retval) : "r" (_GLOBAL_OFFSET_TABLE_)); \ __retval; }) -#ifdef __tilegx__ #define _TLS_LE(x) \ "moveli %0, hw1_last_tls_le(" #x ")\n\t" \ "shl16insli %0, %0, hw0_tls_le(" #x ")\n\t" \ "add %0, %0, tp" -#else -#define _TLS_LE(x) \ - "auli %0, tp, tls_le_ha16(" #x ")\n\t" \ - "addli %0, %0, tls_le_lo16(" #x ")\n\t" -#endif #define TLS_LE(x) \ ({ \ diff --git a/sysdeps/tile/wordcopy.c b/sysdeps/tile/wordcopy.c index 2f93e5a5d1..027a4d6be6 100644 --- a/sysdeps/tile/wordcopy.c +++ b/sysdeps/tile/wordcopy.c @@ -31,11 +31,7 @@ /* Provide the appropriate dblalign builtin to shift two registers based on the alignment of a pointer held in a third register. */ -#ifdef __tilegx__ #define DBLALIGN __insn_dblalign -#else -#define DBLALIGN __insn_dword_align -#endif /* _wordcopy_fwd_aligned -- Copy block beginning at SRCP to block beginning at DSTP with LEN `op_t' words (not LEN bytes!). |