/* Output routines for GCC for ARM.
Copyright (C) 1991, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
- 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
+ 2002, 2003, 2004, 2005, 2006, 2007 Free Software Foundation, Inc.
Contributed by Pieter `Tiggr' Schoenmakers (rcpieter@win.tue.nl)
and Martin Simmons (@harleqn.co.uk).
More major hacks by Richard Earnshaw (rearnsha@arm.com).
#include "target-def.h"
#include "debug.h"
#include "langhooks.h"
+#include "df.h"
/* Forward definitions of types. */
typedef struct minipool_node Mnode;
static unsigned bit_count (unsigned long);
static int arm_address_register_rtx_p (rtx, int);
static int arm_legitimate_index_p (enum machine_mode, rtx, RTX_CODE, int);
-static int thumb_base_register_rtx_p (rtx, enum machine_mode, int);
-inline static int thumb_index_register_rtx_p (rtx, int);
+static int thumb2_legitimate_index_p (enum machine_mode, rtx, int);
+static int thumb1_base_register_rtx_p (rtx, enum machine_mode, int);
+inline static int thumb1_index_register_rtx_p (rtx, int);
static int thumb_far_jump_used_p (void);
static bool thumb_force_lr_save (void);
+static unsigned long thumb1_compute_save_reg_mask (void);
static int const_ok_for_op (HOST_WIDE_INT, enum rtx_code);
static rtx emit_sfm (int, int);
static int arm_size_return_regs (void);
rtx);
static void arm_reorg (void);
static bool note_invalid_constants (rtx, HOST_WIDE_INT, int);
-static int current_file_function_operand (rtx);
static unsigned long arm_compute_save_reg0_reg12_mask (void);
static unsigned long arm_compute_save_reg_mask (void);
static unsigned long arm_isr_value (tree);
#endif
static void arm_output_function_epilogue (FILE *, HOST_WIDE_INT);
static void arm_output_function_prologue (FILE *, HOST_WIDE_INT);
-static void thumb_output_function_prologue (FILE *, HOST_WIDE_INT);
+static void thumb1_output_function_prologue (FILE *, HOST_WIDE_INT);
static int arm_comp_type_attributes (tree, tree);
static void arm_set_default_type_attributes (tree);
static int arm_adjust_cost (rtx, rtx, rtx, int);
static rtx arm_expand_unop_builtin (enum insn_code, tree, rtx, int);
static rtx arm_expand_builtin (tree, rtx, rtx, enum machine_mode, int);
static void emit_constant_insn (rtx cond, rtx pattern);
+static rtx emit_set_insn (rtx, rtx);
static int arm_arg_partial_bytes (CUMULATIVE_ARGS *, enum machine_mode,
tree, bool);
#ifdef OBJECT_FORMAT_ELF
-static void arm_elf_asm_constructor (rtx, int);
+static void arm_elf_asm_constructor (rtx, int) ATTRIBUTE_UNUSED;
+static void arm_elf_asm_destructor (rtx, int) ATTRIBUTE_UNUSED;
#endif
#ifndef ARM_PE
static void arm_encode_section_info (tree, rtx, int);
#endif
static void arm_file_end (void);
+static void arm_file_start (void);
#ifdef AOF_ASSEMBLER
static void aof_globalize_label (FILE *, const char *);
static void aof_dump_pic_table (FILE *);
static void aof_file_start (void);
static void aof_file_end (void);
+static void aof_asm_init_sections (void);
#endif
-static rtx arm_struct_value_rtx (tree, int);
static void arm_setup_incoming_varargs (CUMULATIVE_ARGS *, enum machine_mode,
tree, int *, int);
static bool arm_pass_by_reference (CUMULATIVE_ARGS *,
static void arm_unwind_emit (FILE *, rtx);
static bool arm_output_ttype (rtx);
#endif
+static void arm_dwarf_handle_frame_unspec (const char *, rtx, int);
static tree arm_cxx_guard_type (void);
static bool arm_cxx_guard_mask_bit (void);
static void arm_init_libfuncs (void);
static bool arm_handle_option (size_t, const char *, int);
static unsigned HOST_WIDE_INT arm_shift_truncation_mask (enum machine_mode);
+static bool arm_cannot_copy_insn_p (rtx);
+static bool arm_tls_symbol_p (rtx x);
+static void arm_output_dwarf_dtprel (FILE *, int, rtx) ATTRIBUTE_UNUSED;
+
\f
/* Initialize the GCC target structure. */
#if TARGET_DLLIMPORT_DECL_ATTRIBUTES
#undef TARGET_ATTRIBUTE_TABLE
#define TARGET_ATTRIBUTE_TABLE arm_attribute_table
+#undef TARGET_ASM_FILE_START
+#define TARGET_ASM_FILE_START arm_file_start
#undef TARGET_ASM_FILE_END
#define TARGET_ASM_FILE_END arm_file_end
#undef TARGET_ARG_PARTIAL_BYTES
#define TARGET_ARG_PARTIAL_BYTES arm_arg_partial_bytes
-#undef TARGET_STRUCT_VALUE_RTX
-#define TARGET_STRUCT_VALUE_RTX arm_struct_value_rtx
-
#undef TARGET_SETUP_INCOMING_VARARGS
#define TARGET_SETUP_INCOMING_VARARGS arm_setup_incoming_varargs
#undef TARGET_ALIGN_ANON_BITFIELD
#define TARGET_ALIGN_ANON_BITFIELD arm_align_anon_bitfield
+#undef TARGET_NARROW_VOLATILE_BITFIELD
+#define TARGET_NARROW_VOLATILE_BITFIELD hook_bool_void_false
+
#undef TARGET_CXX_GUARD_TYPE
#define TARGET_CXX_GUARD_TYPE arm_cxx_guard_type
#define TARGET_ARM_EABI_UNWINDER true
#endif /* TARGET_UNWIND_INFO */
+#undef TARGET_DWARF_HANDLE_FRAME_UNSPEC
+#define TARGET_DWARF_HANDLE_FRAME_UNSPEC arm_dwarf_handle_frame_unspec
+
+#undef TARGET_CANNOT_COPY_INSN_P
+#define TARGET_CANNOT_COPY_INSN_P arm_cannot_copy_insn_p
+
+#ifdef HAVE_AS_TLS
+#undef TARGET_HAVE_TLS
+#define TARGET_HAVE_TLS true
+#endif
+
+#undef TARGET_CANNOT_FORCE_CONST_MEM
+#define TARGET_CANNOT_FORCE_CONST_MEM arm_tls_referenced_p
+
+#ifdef HAVE_AS_TLS
+#undef TARGET_ASM_OUTPUT_DWARF_DTPREL
+#define TARGET_ASM_OUTPUT_DWARF_DTPREL arm_output_dwarf_dtprel
+#endif
+
struct gcc_target targetm = TARGET_INITIALIZER;
\f
/* Obstack for minipool constant handling. */
/* The processor for which instructions should be scheduled. */
enum processor_type arm_tune = arm_none;
+/* The default processor used if not overridden by commandline. */
+static enum processor_type arm_default_cpu = arm_none;
+
/* Which floating point model to use. */
enum arm_fp_model arm_fp_model;
/* Which ABI to use. */
enum arm_abi_type arm_abi;
+/* Which thread pointer model to use. */
+enum arm_tp_type target_thread_pointer = TP_AUTO;
+
/* Used to parse -mstructure_size_boundary command line option. */
int arm_structure_size_boundary = DEFAULT_STRUCTURE_SIZE_BOUNDARY;
#define FL_VFPV2 (1 << 13) /* Vector Floating Point V2. */
#define FL_WBUF (1 << 14) /* Schedule for write buffer ops.
Note: ARM6 & 7 derivatives only. */
+#define FL_ARCH6K (1 << 15) /* Architecture rel 6 K extensions. */
+#define FL_THUMB2 (1 << 16) /* Thumb-2. */
+#define FL_NOTM (1 << 17) /* Instructions not present in the 'M'
+ profile. */
+#define FL_DIV (1 << 18) /* Hardware divide. */
+#define FL_VFPV3 (1 << 19) /* Vector Floating Point V3. */
#define FL_IWMMXT (1 << 29) /* XScale v2 or "Intel Wireless MMX technology". */
-#define FL_FOR_ARCH2 0
-#define FL_FOR_ARCH3 FL_MODE32
+#define FL_FOR_ARCH2 FL_NOTM
+#define FL_FOR_ARCH3 (FL_FOR_ARCH2 | FL_MODE32)
#define FL_FOR_ARCH3M (FL_FOR_ARCH3 | FL_ARCH3M)
#define FL_FOR_ARCH4 (FL_FOR_ARCH3M | FL_ARCH4)
#define FL_FOR_ARCH4T (FL_FOR_ARCH4 | FL_THUMB)
#define FL_FOR_ARCH5TEJ FL_FOR_ARCH5TE
#define FL_FOR_ARCH6 (FL_FOR_ARCH5TE | FL_ARCH6)
#define FL_FOR_ARCH6J FL_FOR_ARCH6
-#define FL_FOR_ARCH6K FL_FOR_ARCH6
+#define FL_FOR_ARCH6K (FL_FOR_ARCH6 | FL_ARCH6K)
#define FL_FOR_ARCH6Z FL_FOR_ARCH6
-#define FL_FOR_ARCH6ZK FL_FOR_ARCH6
+#define FL_FOR_ARCH6ZK FL_FOR_ARCH6K
+#define FL_FOR_ARCH6T2 (FL_FOR_ARCH6 | FL_THUMB2)
+#define FL_FOR_ARCH7 (FL_FOR_ARCH6T2 &~ FL_NOTM)
+#define FL_FOR_ARCH7A (FL_FOR_ARCH7 | FL_NOTM)
+#define FL_FOR_ARCH7R (FL_FOR_ARCH7A | FL_DIV)
+#define FL_FOR_ARCH7M (FL_FOR_ARCH7 | FL_DIV)
/* The bits in this mask specify which
instructions we are allowed to generate. */
/* Nonzero if this chip supports the ARM Architecture 6 extensions. */
int arm_arch6 = 0;
+/* Nonzero if this chip supports the ARM 6K extensions. */
+int arm_arch6k = 0;
+
+/* Nonzero if instructions not present in the 'M' profile can be used. */
+int arm_arch_notm = 0;
+
/* Nonzero if this chip can benefit from load scheduling. */
int arm_ld_sched = 0;
/* Nonzero if tuning for XScale */
int arm_tune_xscale = 0;
-/* Nonzero if we want to tune for stores that access the write-buffer.
+/* Nonzero if we want to tune for stores that access the write-buffer.
This typically means an ARM6 or ARM7 with MMU or MPU. */
int arm_tune_wbuf = 0;
interworking clean. */
int arm_cpp_interwork = 0;
+/* Nonzero if chip supports Thumb 2. */
+int arm_arch_thumb2;
+
+/* Nonzero if chip supports integer division instruction. */
+int arm_arch_hwdiv;
+
/* In case of a PRE_INC, POST_INC, PRE_DEC, POST_DEC memory reference, we
must report the mode of the memory reference from PRINT_OPERAND to
PRINT_OPERAND_ADDRESS. */
enum machine_mode output_memory_reference_mode;
/* The register number to be used for the PIC offset register. */
-int arm_pic_register = INVALID_REGNUM;
+unsigned arm_pic_register = INVALID_REGNUM;
/* Set to 1 when a return insn is output, this means that the epilogue
is not needed. */
/* For an explanation of these variables, see final_prescan_insn below. */
int arm_ccfsm_state;
+/* arm_current_cc is also used for Thumb-2 cond_exec blocks. */
enum arm_cond_code arm_current_cc;
rtx arm_target_insn;
int arm_target_label;
+/* The number of conditionally executed insns, including the current insn. */
+int arm_condexec_count = 0;
+/* A bitmask specifying the patterns for the IT block.
+ Zero means do not output an IT block before this insn. */
+int arm_condexec_mask = 0;
+/* The number of bits used in arm_condexec_mask. */
+int arm_condexec_masklen = 0;
/* The condition codes of the ARM, and the inverse function. */
static const char * const arm_condition_codes[] =
"hi", "ls", "ge", "lt", "gt", "le", "al", "nv"
};
+#define ARM_LSL_NAME (TARGET_UNIFIED_ASM ? "lsl" : "asl")
#define streq(string1, string2) (strcmp (string1, string2) == 0)
+
+#define THUMB2_WORK_REGS (0xff & ~( (1 << THUMB_HARD_FRAME_POINTER_REGNUM) \
+ | (1 << SP_REGNUM) | (1 << PC_REGNUM) \
+ | (1 << PIC_OFFSET_TABLE_REGNUM)))
\f
/* Initialization code. */
{"armv6k", mpcore, "6K", FL_CO_PROC | FL_FOR_ARCH6K, NULL},
{"armv6z", arm1176jzs, "6Z", FL_CO_PROC | FL_FOR_ARCH6Z, NULL},
{"armv6zk", arm1176jzs, "6ZK", FL_CO_PROC | FL_FOR_ARCH6ZK, NULL},
+ {"armv6t2", arm1156t2s, "6T2", FL_CO_PROC | FL_FOR_ARCH6T2, NULL},
+ {"armv7", cortexa8, "7", FL_CO_PROC | FL_FOR_ARCH7, NULL},
+ {"armv7-a", cortexa8, "7A", FL_CO_PROC | FL_FOR_ARCH7A, NULL},
+ {"armv7-r", cortexr4, "7R", FL_CO_PROC | FL_FOR_ARCH7R, NULL},
+ {"armv7-m", cortexm3, "7M", FL_CO_PROC | FL_FOR_ARCH7M, NULL},
{"ep9312", ep9312, "4T", FL_LDSCHED | FL_CIRRUS | FL_FOR_ARCH4, NULL},
{"iwmmxt", iwmmxt, "5TE", FL_LDSCHED | FL_STRONG | FL_FOR_ARCH5TE | FL_XSCALE | FL_IWMMXT , NULL},
{NULL, arm_none, NULL, 0 , NULL}
#define ARM_OPT_SET_ARCH 1
#define ARM_OPT_SET_TUNE 2
-/* The name of the proprocessor macro to define for this architecture. */
+/* The name of the preprocessor macro to define for this architecture. */
char arm_arch_name[] = "__ARM_ARCH_0UNK__";
};
-/* Available values for for -mfpu=. */
+/* Available values for -mfpu=. */
static const struct fpu_desc all_fpus[] =
{
{"fpe2", FPUTYPE_FPA_EMU2},
{"fpe3", FPUTYPE_FPA_EMU2},
{"maverick", FPUTYPE_MAVERICK},
- {"vfp", FPUTYPE_VFP}
+ {"vfp", FPUTYPE_VFP},
+ {"vfp3", FPUTYPE_VFP3},
};
ARM_FP_MODEL_FPA, /* FPUTYPE_FPA_EMU2 */
ARM_FP_MODEL_FPA, /* FPUTYPE_FPA_EMU3 */
ARM_FP_MODEL_MAVERICK, /* FPUTYPE_MAVERICK */
- ARM_FP_MODEL_VFP /* FPUTYPE_VFP */
+ ARM_FP_MODEL_VFP, /* FPUTYPE_VFP */
+ ARM_FP_MODEL_VFP /* FPUTYPE_VFP3 */
};
{"apcs-gnu", ARM_ABI_APCS},
{"atpcs", ARM_ABI_ATPCS},
{"aapcs", ARM_ABI_AAPCS},
- {"iwmmxt", ARM_ABI_IWMMXT}
+ {"iwmmxt", ARM_ABI_IWMMXT},
+ {"aapcs-linux", ARM_ABI_AAPCS_LINUX}
+};
+
+/* Supported TLS relocations. */
+
+enum tls_reloc {
+ TLS_GD32,
+ TLS_LDM32,
+ TLS_LDO32,
+ TLS_IE32,
+ TLS_LE32
};
+/* Emit an insn that's a simple single-set. Both the operands must be known
+ to be valid. */
+inline static rtx
+emit_set_insn (rtx x, rtx y)
+{
+ return emit_insn (gen_rtx_SET (VOIDmode, x, y));
+}
+
/* Return the number of bits set in VALUE. */
static unsigned
bit_count (unsigned long value)
routines. */
set_optab_libfunc (sdiv_optab, DImode, "__aeabi_ldivmod");
set_optab_libfunc (udiv_optab, DImode, "__aeabi_uldivmod");
- set_optab_libfunc (sdiv_optab, SImode, "__aeabi_idivmod");
- set_optab_libfunc (udiv_optab, SImode, "__aeabi_uidivmod");
+
+ /* For SImode division the ABI provides div-without-mod routines,
+ which are faster. */
+ set_optab_libfunc (sdiv_optab, SImode, "__aeabi_idiv");
+ set_optab_libfunc (udiv_optab, SImode, "__aeabi_uidiv");
/* We don't have mod libcalls. Fortunately gcc knows how to use the
divmod libcalls instead. */
options. */
if (i == ARM_OPT_SET_ARCH)
target_arch_cpu = sel->core;
-
+
if (i != ARM_OPT_SET_TUNE)
{
/* If we have been given an architecture and a processor
insn_flags = sel->flags;
}
sprintf (arm_arch_name, "__ARM_ARCH_%s__", sel->arch);
+ arm_default_cpu = (enum processor_type) (sel - all_cores);
if (arm_tune == arm_none)
- arm_tune = (enum processor_type) (sel - all_cores);
+ arm_tune = arm_default_cpu;
}
/* The processor for which we should tune should now have been
/* Make sure that the processor choice does not conflict with any of the
other command line choices. */
+ if (TARGET_ARM && !(insn_flags & FL_NOTM))
+ error ("target CPU does not support ARM mode");
+
if (TARGET_INTERWORK && !(insn_flags & FL_THUMB))
{
warning (0, "target CPU does not support interworking" );
&& (TARGET_DEFAULT & MASK_APCS_FRAME))
warning (0, "-g with -mno-apcs-frame may not give sensible debugging");
- /* If stack checking is disabled, we can use r10 as the PIC register,
- which keeps r9 available. */
- if (flag_pic)
- arm_pic_register = TARGET_APCS_STACK ? 9 : 10;
-
if (TARGET_APCS_FLOAT)
warning (0, "passing floating point arguments in fp regs not yet supported");
arm_arch5 = (insn_flags & FL_ARCH5) != 0;
arm_arch5e = (insn_flags & FL_ARCH5E) != 0;
arm_arch6 = (insn_flags & FL_ARCH6) != 0;
+ arm_arch6k = (insn_flags & FL_ARCH6K) != 0;
+ arm_arch_notm = (insn_flags & FL_NOTM) != 0;
+ arm_arch_thumb2 = (insn_flags & FL_THUMB2) != 0;
arm_arch_xscale = (insn_flags & FL_XSCALE) != 0;
arm_arch_cirrus = (insn_flags & FL_CIRRUS) != 0;
arm_tune_wbuf = (tune_flags & FL_WBUF) != 0;
arm_tune_xscale = (tune_flags & FL_XSCALE) != 0;
arm_arch_iwmmxt = (insn_flags & FL_IWMMXT) != 0;
+ arm_arch_hwdiv = (insn_flags & FL_DIV) != 0;
/* V5 code we generate is completely interworking capable, so we turn off
TARGET_INTERWORK here to avoid many tests later on. */
if (arm_float_abi == ARM_FLOAT_ABI_HARD && TARGET_VFP)
sorry ("-mfloat-abi=hard and VFP");
+ /* FPA and iWMMXt are incompatible because the insn encodings overlap.
+ VFP and iWMMXt can theoretically coexist, but it's unlikely such silicon
+ will ever exist. GCC makes no attempt to support this combination. */
+ if (TARGET_IWMMXT && !TARGET_SOFT_FLOAT)
+ sorry ("iWMMXt and hardware floating point");
+
+ /* ??? iWMMXt insn patterns need auditing for Thumb-2. */
+ if (TARGET_THUMB2 && TARGET_IWMMXT)
+ sorry ("Thumb-2 iWMMXt");
+
/* If soft-float is specified then don't use FPU. */
if (TARGET_SOFT_FLOAT)
arm_fpu_arch = FPUTYPE_NONE;
&& (tune_flags & FL_MODE32) == 0)
flag_schedule_insns = flag_schedule_insns_after_reload = 0;
+ if (target_thread_switch)
+ {
+ if (strcmp (target_thread_switch, "soft") == 0)
+ target_thread_pointer = TP_SOFT;
+ else if (strcmp (target_thread_switch, "auto") == 0)
+ target_thread_pointer = TP_AUTO;
+ else if (strcmp (target_thread_switch, "cp15") == 0)
+ target_thread_pointer = TP_CP15;
+ else
+ error ("invalid thread pointer option: -mtp=%s", target_thread_switch);
+ }
+
+ /* Use the cp15 method if it is available. */
+ if (target_thread_pointer == TP_AUTO)
+ {
+ if (arm_arch6k && !TARGET_THUMB)
+ target_thread_pointer = TP_CP15;
+ else
+ target_thread_pointer = TP_SOFT;
+ }
+
+ if (TARGET_HARD_TP && TARGET_THUMB1)
+ error ("can not use -mtp=cp15 with 16-bit Thumb");
+
/* Override the default structure alignment for AAPCS ABI. */
- if (arm_abi == ARM_ABI_AAPCS)
+ if (TARGET_AAPCS_BASED)
arm_structure_size_boundary = 8;
if (structure_size_string != NULL)
ARM_DOUBLEWORD_ALIGN ? "8, 32 or 64": "8 or 32");
}
+ if (!TARGET_ARM && TARGET_VXWORKS_RTP && flag_pic)
+ {
+ error ("RTP PIC is incompatible with Thumb");
+ flag_pic = 0;
+ }
+
+ /* If stack checking is disabled, we can use r10 as the PIC register,
+ which keeps r9 available. The EABI specifies r9 as the PIC register. */
+ if (flag_pic && TARGET_SINGLE_PIC_BASE)
+ {
+ if (TARGET_VXWORKS_RTP)
+ warning (0, "RTP PIC is incompatible with -msingle-pic-base");
+ arm_pic_register = (TARGET_APCS_STACK || TARGET_AAPCS_BASED) ? 9 : 10;
+ }
+
+ if (flag_pic && TARGET_VXWORKS_RTP)
+ arm_pic_register = 9;
+
if (arm_pic_register_string != NULL)
{
int pic_register = decode_reg_name (arm_pic_register_string);
else if (pic_register < 0 || call_used_regs[pic_register]
|| pic_register == HARD_FRAME_POINTER_REGNUM
|| pic_register == STACK_POINTER_REGNUM
- || pic_register >= PC_REGNUM)
+ || pic_register >= PC_REGNUM
+ || (TARGET_VXWORKS_RTP
+ && (unsigned int) pic_register != arm_pic_register))
error ("unable to use '%s' for PIC register", arm_pic_register_string);
else
arm_pic_register = pic_register;
}
+ /* ??? We might want scheduling for thumb2. */
if (TARGET_THUMB && flag_schedule_insns)
{
/* Don't warn since it's on by default in -O2. */
if (optimize_size)
{
- /* There's some dispute as to whether this should be 1 or 2. However,
- experiments seem to show that in pathological cases a setting of
- 1 degrades less severely than a setting of 2. This could change if
- other parts of the compiler change their behavior. */
arm_constant_limit = 1;
/* If optimizing for size, bump the number of instructions that we
const isr_attribute_arg * ptr;
const char * arg;
+ if (!arm_arch_notm)
+ return ARM_FT_NORMAL | ARM_FT_STACKALIGN;
+
/* No argument - default to IRQ. */
if (argument == NULL_TREE)
return ARM_FT_ISR;
register values that will never be needed again. This optimization
was added to speed up context switching in a kernel application. */
if (optimize > 0
- && TREE_NOTHROW (current_function_decl)
+ && (TREE_NOTHROW (current_function_decl)
+ || !(flag_unwind_tables
+ || (flag_exceptions && !USING_SJLJ_EXCEPTIONS)))
&& TREE_THIS_VOLATILE (current_function_decl))
type |= ARM_FT_VOLATILE;
func_type = arm_current_func_type ();
- /* Naked functions and volatile functions need special
+ /* Naked, volatile and stack alignment functions need special
consideration. */
- if (func_type & (ARM_FT_VOLATILE | ARM_FT_NAKED))
+ if (func_type & (ARM_FT_VOLATILE | ARM_FT_NAKED | ARM_FT_STACKALIGN))
return 0;
- /* So do interrupt functions that use the frame pointer. */
- if (IS_INTERRUPT (func_type) && frame_pointer_needed)
+ /* So do interrupt functions that use the frame pointer and Thumb
+ interrupt functions. */
+ if (IS_INTERRUPT (func_type) && (frame_pointer_needed || TARGET_THUMB))
return 0;
offsets = arm_get_frame_offsets ();
We test for !arm_arch5 here, because code for any architecture
less than this could potentially be run on one of the buggy
chips. */
- if (stack_adjust == 4 && !arm_arch5)
+ if (stack_adjust == 4 && !arm_arch5 && TARGET_ARM)
{
/* Validate that r3 is a call-clobbered register (always true in
the default abi) ... */
/* Can't be done if interworking with Thumb, and any registers have been
stacked. */
- if (TARGET_INTERWORK && saved_int_regs != 0)
+ if (TARGET_INTERWORK && saved_int_regs != 0 && !IS_INTERRUPT(func_type))
return 0;
/* On StrongARM, conditional returns are expensive if they aren't
if (saved_int_regs != 0 && saved_int_regs != (1 << LR_REGNUM))
return 0;
- if (flag_pic && regs_ever_live[PIC_OFFSET_TABLE_REGNUM])
+ if (flag_pic
+ && arm_pic_register != INVALID_REGNUM
+ && df_regs_ever_live_p (PIC_OFFSET_TABLE_REGNUM))
return 0;
}
since this also requires an insn. */
if (TARGET_HARD_FLOAT && TARGET_FPA)
for (regno = FIRST_FPA_REGNUM; regno <= LAST_FPA_REGNUM; regno++)
- if (regs_ever_live[regno] && !call_used_regs[regno])
+ if (df_regs_ever_live_p (regno) && !call_used_regs[regno])
return 0;
/* Likewise VFP regs. */
if (TARGET_HARD_FLOAT && TARGET_VFP)
for (regno = FIRST_VFP_REGNUM; regno <= LAST_VFP_REGNUM; regno++)
- if (regs_ever_live[regno] && !call_used_regs[regno])
+ if (df_regs_ever_live_p (regno) && !call_used_regs[regno])
return 0;
if (TARGET_REALLY_IWMMXT)
for (regno = FIRST_IWMMXT_REGNUM; regno <= LAST_IWMMXT_REGNUM; regno++)
- if (regs_ever_live[regno] && ! call_used_regs [regno])
+ if (df_regs_ever_live_p (regno) && ! call_used_regs[regno])
return 0;
return 1;
const_ok_for_arm (HOST_WIDE_INT i)
{
int lowbit;
-
+
/* For machines with >32 bit HOST_WIDE_INT, the bits above bit 31 must
be all zero, or all one. */
if ((i & ~(unsigned HOST_WIDE_INT) 0xffffffff) != 0
return FALSE;
i &= (unsigned HOST_WIDE_INT) 0xffffffff;
-
+
/* Fast return for 0 and small values. We must do this for zero, since
the code below can't handle that one case. */
if ((i & ~(unsigned HOST_WIDE_INT) 0xff) == 0)
return TRUE;
- /* Get the number of trailing zeros, rounded down to the nearest even
- number. */
- lowbit = (ffs ((int) i) - 1) & ~1;
+ /* Get the number of trailing zeros. */
+ lowbit = ffs((int) i) - 1;
+
+ /* Only even shifts are allowed in ARM mode so round down to the
+ nearest even number. */
+ if (TARGET_ARM)
+ lowbit &= ~1;
if ((i & ~(((unsigned HOST_WIDE_INT) 0xff) << lowbit)) == 0)
return TRUE;
- else if (lowbit <= 4
+
+ if (TARGET_ARM)
+ {
+ /* Allow rotated constants in ARM mode. */
+ if (lowbit <= 4
&& ((i & ~0xc000003f) == 0
|| (i & ~0xf000000f) == 0
|| (i & ~0xfc000003) == 0))
- return TRUE;
+ return TRUE;
+ }
+ else
+ {
+ HOST_WIDE_INT v;
+
+ /* Allow repeated pattern. */
+ v = i & 0xff;
+ v |= v << 16;
+ if (i == v || i == (v | (v << 8)))
+ return TRUE;
+ }
return FALSE;
}
either produce a simpler sequence, or we will want to cse the values.
Return value is the number of insns emitted. */
+/* ??? Tweak this for thumb2. */
int
arm_split_constant (enum rtx_code code, enum machine_mode mode, rtx insn,
HOST_WIDE_INT val, rtx target, rtx source, int subtargets)
{
/* Currently SET is the only monadic value for CODE, all
the rest are diadic. */
- emit_insn (gen_rtx_SET (VOIDmode, target, GEN_INT (val)));
+ emit_set_insn (target, GEN_INT (val));
return 1;
}
else
{
rtx temp = subtargets ? gen_reg_rtx (mode) : target;
- emit_insn (gen_rtx_SET (VOIDmode, temp, GEN_INT (val)));
+ emit_set_insn (temp, GEN_INT (val));
/* For MINUS, the value is subtracted from, since we never
have subtraction of a constant. */
if (code == MINUS)
- emit_insn (gen_rtx_SET (VOIDmode, target,
- gen_rtx_MINUS (mode, temp, source)));
+ emit_set_insn (target, gen_rtx_MINUS (mode, temp, source));
else
- emit_insn (gen_rtx_SET (VOIDmode, target,
- gen_rtx_fmt_ee (code, mode, source, temp)));
+ emit_set_insn (target,
+ gen_rtx_fmt_ee (code, mode, source, temp));
return 2;
}
}
1);
}
+/* Return the number of ARM instructions required to synthesize the given
+ constant. */
static int
count_insns_for_constant (HOST_WIDE_INT remainder, int i)
{
/* As above, but extra parameter GENERATE which, if clear, suppresses
RTL generation. */
+/* ??? This needs more work for thumb2. */
static int
arm_gen_constant (enum rtx_code code, enum machine_mode mode, rtx cond,
gen_rtx_SET (VOIDmode, target, source));
return 1;
}
-
+
/* We don't know how to handle other cases yet. */
gcc_assert (remainder == 0xffffffff);
-
+
if (generate)
emit_constant_insn (cond,
gen_rtx_SET (VOIDmode, target,
switch (code)
{
case SET:
+ /* See if we can use movw. */
+ if (arm_arch_thumb2 && (remainder & 0xffff0000) == 0)
+ {
+ if (generate)
+ emit_constant_insn (cond, gen_rtx_SET (VOIDmode, target,
+ GEN_INT (val)));
+ return 1;
+ }
+
/* See if we can do this by sign_extending a constant that is known
to be negative. This is a good, way of doing it, since the shift
may well merge into a subsequent insn. */
temp1 = 0x80000000 >> (topshift - 1);
temp2 = ARM_SIGN_EXTEND (temp1 - remainder);
-
+
if (const_ok_for_arm (temp2))
{
if (generate)
We start by looking for the largest block of zeros that are aligned on
a 2-bit boundary, we then fill up the temps, wrapping around to the
top of the word when we drop off the bottom.
- In the worst case this code should produce no more than four insns. */
+ In the worst case this code should produce no more than four insns.
+ Thumb-2 constants are shifted, not rotated, so the MSB is always the
+ best place to start. */
+
+ /* ??? Use thumb2 replicated constants when the high and low halfwords are
+ the same. */
{
int best_start = 0;
- int best_consecutive_zeros = 0;
-
- for (i = 0; i < 32; i += 2)
+ if (!TARGET_THUMB2)
{
- int consecutive_zeros = 0;
+ int best_consecutive_zeros = 0;
- if (!(remainder & (3 << i)))
+ for (i = 0; i < 32; i += 2)
{
- while ((i < 32) && !(remainder & (3 << i)))
- {
- consecutive_zeros += 2;
- i += 2;
- }
- if (consecutive_zeros > best_consecutive_zeros)
+ int consecutive_zeros = 0;
+
+ if (!(remainder & (3 << i)))
{
- best_consecutive_zeros = consecutive_zeros;
- best_start = i - consecutive_zeros;
+ while ((i < 32) && !(remainder & (3 << i)))
+ {
+ consecutive_zeros += 2;
+ i += 2;
+ }
+ if (consecutive_zeros > best_consecutive_zeros)
+ {
+ best_consecutive_zeros = consecutive_zeros;
+ best_start = i - consecutive_zeros;
+ }
+ i -= 2;
}
- i -= 2;
}
- }
-
- /* So long as it won't require any more insns to do so, it's
- desirable to emit a small constant (in bits 0...9) in the last
- insn. This way there is more chance that it can be combined with
- a later addressing insn to form a pre-indexed load or store
- operation. Consider:
-
- *((volatile int *)0xe0000100) = 1;
- *((volatile int *)0xe0000110) = 2;
-
- We want this to wind up as:
- mov rA, #0xe0000000
- mov rB, #1
- str rB, [rA, #0x100]
- mov rB, #2
- str rB, [rA, #0x110]
-
- rather than having to synthesize both large constants from scratch.
-
- Therefore, we calculate how many insns would be required to emit
- the constant starting from `best_start', and also starting from
- zero (i.e. with bit 31 first to be output). If `best_start' doesn't
- yield a shorter sequence, we may as well use zero. */
- if (best_start != 0
- && ((((unsigned HOST_WIDE_INT) 1) << best_start) < remainder)
- && (count_insns_for_constant (remainder, 0) <=
- count_insns_for_constant (remainder, best_start)))
- best_start = 0;
+ /* So long as it won't require any more insns to do so, it's
+ desirable to emit a small constant (in bits 0...9) in the last
+ insn. This way there is more chance that it can be combined with
+ a later addressing insn to form a pre-indexed load or store
+ operation. Consider:
+
+ *((volatile int *)0xe0000100) = 1;
+ *((volatile int *)0xe0000110) = 2;
+
+ We want this to wind up as:
+
+ mov rA, #0xe0000000
+ mov rB, #1
+ str rB, [rA, #0x100]
+ mov rB, #2
+ str rB, [rA, #0x110]
+
+ rather than having to synthesize both large constants from scratch.
+
+ Therefore, we calculate how many insns would be required to emit
+ the constant starting from `best_start', and also starting from
+ zero (i.e. with bit 31 first to be output). If `best_start' doesn't
+ yield a shorter sequence, we may as well use zero. */
+ if (best_start != 0
+ && ((((unsigned HOST_WIDE_INT) 1) << best_start) < remainder)
+ && (count_insns_for_constant (remainder, 0) <=
+ count_insns_for_constant (remainder, best_start)))
+ best_start = 0;
+ }
/* Now start emitting the insns. */
i = best_start;
code = PLUS;
insns++;
- i -= 6;
+ if (TARGET_ARM)
+ i -= 6;
+ else
+ i -= 7;
}
- i -= 2;
+ /* Arm allows rotates by a multiple of two. Thumb-2 allows arbitrary
+ shifts. */
+ if (TARGET_ARM)
+ i -= 2;
+ else
+ i--;
}
while (remainder);
}
mode = mode_for_size (size * BITS_PER_UNIT, MODE_INT, 0);
}
}
-
+
return LIBCALL_VALUE(mode);
}
-/* Determine the amount of memory needed to store the possible return
+/* Determine the amount of memory needed to store the possible return
registers of an untyped call. */
int
arm_apply_result_size (void)
tree fndecl ATTRIBUTE_UNUSED)
{
/* On the ARM, the offset starts at 0. */
- pcum->nregs = ((fntype && aggregate_value_p (TREE_TYPE (fntype), fntype)) ? 1 : 0);
+ pcum->nregs = 0;
pcum->iwmmxt_nregs = 0;
pcum->can_split = true;
- pcum->call_cookie = CALL_NORMAL;
-
- if (TARGET_LONG_CALLS)
- pcum->call_cookie = CALL_LONG;
-
- /* Check for long call/short call attributes. The attributes
- override any command line option. */
- if (fntype)
- {
- if (lookup_attribute ("short_call", TYPE_ATTRIBUTES (fntype)))
- pcum->call_cookie = CALL_SHORT;
- else if (lookup_attribute ("long_call", TYPE_ATTRIBUTES (fntype)))
- pcum->call_cookie = CALL_LONG;
- }
-
/* Varargs vectors are treated the same as long long.
named_count avoids having to change the way arm handles 'named' */
pcum->named_count = 0;
pcum->nregs++;
if (mode == VOIDmode)
- /* Compute operand 2 of the call insn. */
- return GEN_INT (pcum->call_cookie);
+ /* Pick an arbitrary value for operand 2 of the call insn. */
+ return const0_rtx;
/* Only allow splitting an arg between regs and memory if all preceding
args were allocated to regs. For args passed by reference we only count
/* Encode the current state of the #pragma [no_]long_calls. */
typedef enum
{
- OFF, /* No #pramgma [no_]long_calls is in effect. */
+ OFF, /* No #pragma [no_]long_calls is in effect. */
LONG, /* #pragma long_calls is in effect. */
SHORT /* #pragma no_long_calls is in effect. */
} arm_pragma_enum;
attribute. */
static tree
-arm_handle_notshared_attribute (tree *node,
- tree name ATTRIBUTE_UNUSED,
- tree args ATTRIBUTE_UNUSED,
- int flags ATTRIBUTE_UNUSED,
+arm_handle_notshared_attribute (tree *node,
+ tree name ATTRIBUTE_UNUSED,
+ tree args ATTRIBUTE_UNUSED,
+ int flags ATTRIBUTE_UNUSED,
bool *no_add_attrs)
{
tree decl = TYPE_NAME (*node);
return 1;
}
-/* Encode long_call or short_call attribute by prefixing
- symbol name in DECL with a special character FLAG. */
-void
-arm_encode_call_attribute (tree decl, int flag)
-{
- const char * str = XSTR (XEXP (DECL_RTL (decl), 0), 0);
- int len = strlen (str);
- char * newstr;
-
- /* Do not allow weak functions to be treated as short call. */
- if (DECL_WEAK (decl) && flag == SHORT_CALL_FLAG_CHAR)
- return;
-
- newstr = alloca (len + 2);
- newstr[0] = flag;
- strcpy (newstr + 1, str);
-
- newstr = (char *) ggc_alloc_string (newstr, len + 1);
- XSTR (XEXP (DECL_RTL (decl), 0), 0) = newstr;
-}
-
/* Assigns default attributes to newly defined type. This is used to
set short_call/long_call attributes for function types of
functions defined inside corresponding #pragma scopes. */
}
}
\f
-/* Return 1 if the operand is a SYMBOL_REF for a function known to be
- defined within the current compilation unit. If this cannot be
- determined, then 0 is returned. */
-static int
-current_file_function_operand (rtx sym_ref)
+/* Return true if DECL is known to be linked into section SECTION. */
+
+static bool
+arm_function_in_section_p (tree decl, section *section)
{
- /* This is a bit of a fib. A function will have a short call flag
- applied to its name if it has the short call attribute, or it has
- already been defined within the current compilation unit. */
- if (ENCODED_SHORT_CALL_ATTR_P (XSTR (sym_ref, 0)))
- return 1;
+ /* We can only be certain about functions defined in the same
+ compilation unit. */
+ if (!TREE_STATIC (decl))
+ return false;
- /* The current function is always defined within the current compilation
- unit. If it s a weak definition however, then this may not be the real
- definition of the function, and so we have to say no. */
- if (sym_ref == XEXP (DECL_RTL (current_function_decl), 0)
- && !DECL_WEAK (current_function_decl))
- return 1;
+ /* Make sure that SYMBOL always binds to the definition in this
+ compilation unit. */
+ if (!targetm.binds_local_p (decl))
+ return false;
- /* We cannot make the determination - default to returning 0. */
- return 0;
+ /* If DECL_SECTION_NAME is set, assume it is trustworthy. */
+ if (!DECL_SECTION_NAME (decl))
+ {
+ /* Only cater for unit-at-a-time mode, where we know that the user
+ cannot later specify a section for DECL. */
+ if (!flag_unit_at_a_time)
+ return false;
+
+ /* Make sure that we will not create a unique section for DECL. */
+ if (flag_function_sections || DECL_ONE_ONLY (decl))
+ return false;
+ }
+
+ return function_section (decl) == section;
}
-/* Return nonzero if a 32 bit "long_call" should be generated for
- this call. We generate a long_call if the function:
+/* Return nonzero if a 32-bit "long_call" should be generated for
+ a call from the current function to DECL. We generate a long_call
+ if the function:
a. has an __attribute__((long call))
or b. is within the scope of a #pragma long_calls
or c. the -mlong-calls command line switch has been specified
- . and either:
- 1. -ffunction-sections is in effect
- or 2. the current function has __attribute__ ((section))
- or 3. the target function has __attribute__ ((section))
However we do not generate a long call if the function:
d. has an __attribute__ ((short_call))
or e. is inside the scope of a #pragma no_long_calls
- or f. is defined within the current compilation unit.
-
- This function will be called by C fragments contained in the machine
- description file. SYM_REF and CALL_COOKIE correspond to the matched
- rtl operands. CALL_SYMBOL is used to distinguish between
- two different callers of the function. It is set to 1 in the
- "call_symbol" and "call_symbol_value" patterns and to 0 in the "call"
- and "call_value" patterns. This is because of the difference in the
- SYM_REFs passed by these patterns. */
-int
-arm_is_longcall_p (rtx sym_ref, int call_cookie, int call_symbol)
-{
- if (!call_symbol)
- {
- if (GET_CODE (sym_ref) != MEM)
- return 0;
+ or f. is defined in the same section as the current function. */
- sym_ref = XEXP (sym_ref, 0);
- }
+bool
+arm_is_long_call_p (tree decl)
+{
+ tree attrs;
- if (GET_CODE (sym_ref) != SYMBOL_REF)
- return 0;
+ if (!decl)
+ return TARGET_LONG_CALLS;
- if (call_cookie & CALL_SHORT)
- return 0;
+ attrs = TYPE_ATTRIBUTES (TREE_TYPE (decl));
+ if (lookup_attribute ("short_call", attrs))
+ return false;
- if (TARGET_LONG_CALLS)
- {
- if (flag_function_sections
- || DECL_SECTION_NAME (current_function_decl))
- /* c.3 is handled by the definition of the
- ARM_DECLARE_FUNCTION_SIZE macro. */
- return 1;
- }
+ /* For "f", be conservative, and only cater for cases in which the
+ whole of the current function is placed in the same section. */
+ if (!flag_reorder_blocks_and_partition
+ && arm_function_in_section_p (decl, current_function_section ()))
+ return false;
- if (current_file_function_operand (sym_ref))
- return 0;
+ if (lookup_attribute ("long_call", attrs))
+ return true;
- return (call_cookie & CALL_LONG)
- || ENCODED_LONG_CALL_ATTR_P (XSTR (sym_ref, 0))
- || TARGET_LONG_CALLS;
+ return TARGET_LONG_CALLS;
}
/* Return nonzero if it is ok to make a tail-call to DECL. */
static bool
arm_function_ok_for_sibcall (tree decl, tree exp ATTRIBUTE_UNUSED)
{
- int call_type = TARGET_LONG_CALLS ? CALL_LONG : CALL_NORMAL;
+ unsigned long func_type;
if (cfun->machine->sibcall_blocked)
return false;
if (decl == NULL || TARGET_THUMB)
return false;
- /* Get the calling method. */
- if (lookup_attribute ("short_call", TYPE_ATTRIBUTES (TREE_TYPE (decl))))
- call_type = CALL_SHORT;
- else if (lookup_attribute ("long_call", TYPE_ATTRIBUTES (TREE_TYPE (decl))))
- call_type = CALL_LONG;
+ /* The PIC register is live on entry to VxWorks PLT entries, so we
+ must make the call before restoring the PIC register. */
+ if (TARGET_VXWORKS_RTP && flag_pic && !targetm.binds_local_p (decl))
+ return false;
/* Cannot tail-call to long calls, since these are out of range of
- a branch instruction. However, if not compiling PIC, we know
- we can reach the symbol if it is in this compilation unit. */
- if (call_type == CALL_LONG && (flag_pic || !TREE_ASM_WRITTEN (decl)))
+ a branch instruction. */
+ if (arm_is_long_call_p (decl))
return false;
/* If we are interworking and the function is not declared static
if (TARGET_INTERWORK && TREE_PUBLIC (decl) && !TREE_ASM_WRITTEN (decl))
return false;
+ func_type = arm_current_func_type ();
/* Never tailcall from an ISR routine - it needs a special exit sequence. */
- if (IS_INTERRUPT (arm_current_func_type ()))
+ if (IS_INTERRUPT (func_type))
+ return false;
+
+ /* Never tailcall if function may be called with a misaligned SP. */
+ if (IS_STACKALIGN (func_type))
return false;
/* Everything else is ok. */
/* Addressing mode support functions. */
/* Return nonzero if X is a legitimate immediate operand when compiling
- for PIC. */
+ for PIC. We know that X satisfies CONSTANT_P and flag_pic is true. */
int
legitimate_pic_operand_p (rtx x)
{
- if (CONSTANT_P (x)
- && flag_pic
- && (GET_CODE (x) == SYMBOL_REF
- || (GET_CODE (x) == CONST
- && GET_CODE (XEXP (x, 0)) == PLUS
- && GET_CODE (XEXP (XEXP (x, 0), 0)) == SYMBOL_REF)))
+ if (GET_CODE (x) == SYMBOL_REF
+ || (GET_CODE (x) == CONST
+ && GET_CODE (XEXP (x, 0)) == PLUS
+ && GET_CODE (XEXP (XEXP (x, 0), 0)) == SYMBOL_REF))
return 0;
return 1;
}
+/* Record that the current function needs a PIC register. Initialize
+ cfun->machine->pic_reg if we have not already done so. */
+
+static void
+require_pic_register (void)
+{
+ /* A lot of the logic here is made obscure by the fact that this
+ routine gets called as part of the rtx cost estimation process.
+ We don't want those calls to affect any assumptions about the real
+ function; and further, we can't call entry_of_function() until we
+ start the real expansion process. */
+ if (!current_function_uses_pic_offset_table)
+ {
+ gcc_assert (!no_new_pseudos);
+ if (arm_pic_register != INVALID_REGNUM)
+ {
+ cfun->machine->pic_reg = gen_rtx_REG (Pmode, arm_pic_register);
+
+ /* Play games to avoid marking the function as needing pic
+ if we are being called as part of the cost-estimation
+ process. */
+ if (current_ir_type () != IR_GIMPLE)
+ current_function_uses_pic_offset_table = 1;
+ }
+ else
+ {
+ rtx seq;
+
+ cfun->machine->pic_reg = gen_reg_rtx (Pmode);
+
+ /* Play games to avoid marking the function as needing pic
+ if we are being called as part of the cost-estimation
+ process. */
+ if (current_ir_type () != IR_GIMPLE)
+ {
+ current_function_uses_pic_offset_table = 1;
+ start_sequence ();
+
+ arm_load_pic_register (0UL);
+
+ seq = get_insns ();
+ end_sequence ();
+ emit_insn_after (seq, entry_of_function ());
+ }
+ }
+ }
+}
+
rtx
legitimize_pic_address (rtx orig, enum machine_mode mode, rtx reg)
{
rtx insn;
int subregs = 0;
+ /* If this function doesn't have a pic register, create one now. */
+ require_pic_register ();
+
if (reg == 0)
{
gcc_assert (!no_new_pseudos);
if (TARGET_ARM)
emit_insn (gen_pic_load_addr_arm (address, orig));
- else
- emit_insn (gen_pic_load_addr_thumb (address, orig));
-
+ else if (TARGET_THUMB2)
+ emit_insn (gen_pic_load_addr_thumb2 (address, orig));
+ else /* TARGET_THUMB1 */
+ emit_insn (gen_pic_load_addr_thumb1 (address, orig));
+
+ /* VxWorks does not impose a fixed gap between segments; the run-time
+ gap can be different from the object-file gap. We therefore can't
+ use GOTOFF unless we are absolutely sure that the symbol is in the
+ same segment as the GOT. Unfortunately, the flexibility of linker
+ scripts means that we can't be sure of that in general, so assume
+ that GOTOFF is never valid on VxWorks. */
if ((GET_CODE (orig) == LABEL_REF
|| (GET_CODE (orig) == SYMBOL_REF &&
SYMBOL_REF_LOCAL_P (orig)))
- && NEED_GOT_RELOC)
- pic_ref = gen_rtx_PLUS (Pmode, pic_offset_table_rtx, address);
+ && NEED_GOT_RELOC
+ && !TARGET_VXWORKS_RTP)
+ pic_ref = gen_rtx_PLUS (Pmode, cfun->machine->pic_reg, address);
else
{
pic_ref = gen_const_mem (Pmode,
- gen_rtx_PLUS (Pmode, pic_offset_table_rtx,
+ gen_rtx_PLUS (Pmode, cfun->machine->pic_reg,
address));
}
insn = emit_move_insn (reg, pic_ref);
#endif
- current_function_uses_pic_offset_table = 1;
/* Put a REG_EQUAL note on this insn, so that it can be optimized
by loop. */
- REG_NOTES (insn) = gen_rtx_EXPR_LIST (REG_EQUAL, orig,
- REG_NOTES (insn));
+ set_unique_reg_note (insn, REG_EQUAL, orig);
+
return reg;
}
else if (GET_CODE (orig) == CONST)
rtx base, offset;
if (GET_CODE (XEXP (orig, 0)) == PLUS
- && XEXP (XEXP (orig, 0), 0) == pic_offset_table_rtx)
+ && XEXP (XEXP (orig, 0), 0) == cfun->machine->pic_reg)
+ return orig;
+
+ if (GET_CODE (XEXP (orig, 0)) == UNSPEC
+ && XINT (XEXP (orig, 0), 1) == UNSPEC_TLS)
return orig;
if (reg == 0)
}
gcc_assert (GET_CODE (XEXP (orig, 0)) == PLUS);
-
+
base = legitimize_pic_address (XEXP (XEXP (orig, 0), 0), Pmode, reg);
offset = legitimize_pic_address (XEXP (XEXP (orig, 0), 1), Pmode,
base == reg ? 0 : reg);
}
-/* Find a spare low register to use during the prolog of a function. */
+/* Find a spare register to use during the prolog of a function. */
static int
thumb_find_work_register (unsigned long pushed_regs_mask)
register allocation order means that sometimes r3 might be used
but earlier argument registers might not, so check them all. */
for (reg = LAST_ARG_REGNUM; reg >= 0; reg --)
- if (!regs_ever_live[reg])
+ if (!df_regs_ever_live_p (reg))
return reg;
/* Before going on to check the call-saved registers we can try a couple
&& current_function_args_size <= (LAST_ARG_REGNUM * UNITS_PER_WORD)
&& cfun->args_info.nregs < 4)
return LAST_ARG_REGNUM;
-
+
/* Otherwise look for a call-saved register that is going to be pushed. */
for (reg = LAST_LO_REGNUM; reg > LAST_ARG_REGNUM; reg --)
if (pushed_regs_mask & (1 << reg))
return reg;
+ if (TARGET_THUMB2)
+ {
+ /* Thumb-2 can use high regs. */
+ for (reg = FIRST_HI_REGNUM; reg < 15; reg ++)
+ if (pushed_regs_mask & (1 << reg))
+ return reg;
+ }
/* Something went wrong - thumb_compute_save_reg_mask()
should have arranged for a suitable register to be pushed. */
gcc_unreachable ();
}
+static GTY(()) int pic_labelno;
/* Generate code to load the PIC register. In thumb mode SCRATCH is a
low register. */
void
-arm_load_pic_register (unsigned int scratch)
+arm_load_pic_register (unsigned long saved_regs ATTRIBUTE_UNUSED)
{
#ifndef AOF_ASSEMBLER
- rtx l1, pic_tmp, pic_tmp2, pic_rtx;
+ rtx l1, labelno, pic_tmp, pic_tmp2, pic_rtx, pic_reg;
rtx global_offset_table;
if (current_function_uses_pic_offset_table == 0 || TARGET_SINGLE_PIC_BASE)
gcc_assert (flag_pic);
- l1 = gen_label_rtx ();
-
- global_offset_table = gen_rtx_SYMBOL_REF (Pmode, "_GLOBAL_OFFSET_TABLE_");
- /* On the ARM the PC register contains 'dot + 8' at the time of the
- addition, on the Thumb it is 'dot + 4'. */
- pic_tmp = plus_constant (gen_rtx_LABEL_REF (Pmode, l1), TARGET_ARM ? 8 : 4);
- if (GOT_PCREL)
- pic_tmp2 = gen_rtx_CONST (VOIDmode,
- gen_rtx_PLUS (Pmode, global_offset_table, pc_rtx));
- else
- pic_tmp2 = gen_rtx_CONST (VOIDmode, global_offset_table);
+ pic_reg = cfun->machine->pic_reg;
+ if (TARGET_VXWORKS_RTP)
+ {
+ pic_rtx = gen_rtx_SYMBOL_REF (Pmode, VXWORKS_GOTT_BASE);
+ pic_rtx = gen_rtx_CONST (Pmode, pic_rtx);
+ emit_insn (gen_pic_load_addr_arm (pic_reg, pic_rtx));
- pic_rtx = gen_rtx_CONST (Pmode, gen_rtx_MINUS (Pmode, pic_tmp2, pic_tmp));
+ emit_insn (gen_rtx_SET (Pmode, pic_reg, gen_rtx_MEM (Pmode, pic_reg)));
- if (TARGET_ARM)
- {
- emit_insn (gen_pic_load_addr_arm (pic_offset_table_rtx, pic_rtx));
- emit_insn (gen_pic_add_dot_plus_eight (pic_offset_table_rtx, l1));
+ pic_tmp = gen_rtx_SYMBOL_REF (Pmode, VXWORKS_GOTT_INDEX);
+ emit_insn (gen_pic_offset_arm (pic_reg, pic_reg, pic_tmp));
}
else
{
- if (REGNO (pic_offset_table_rtx) > LAST_LO_REGNUM)
+ /* We use an UNSPEC rather than a LABEL_REF because this label
+ never appears in the code stream. */
+
+ labelno = GEN_INT (pic_labelno++);
+ l1 = gen_rtx_UNSPEC (Pmode, gen_rtvec (1, labelno), UNSPEC_PIC_LABEL);
+ l1 = gen_rtx_CONST (VOIDmode, l1);
+
+ global_offset_table
+ = gen_rtx_SYMBOL_REF (Pmode, "_GLOBAL_OFFSET_TABLE_");
+ /* On the ARM the PC register contains 'dot + 8' at the time of the
+ addition, on the Thumb it is 'dot + 4'. */
+ pic_tmp = plus_constant (l1, TARGET_ARM ? 8 : 4);
+ if (GOT_PCREL)
{
- /* We will have pushed the pic register, so should always be
- able to find a work register. */
- pic_tmp = gen_rtx_REG (SImode, scratch);
- emit_insn (gen_pic_load_addr_thumb (pic_tmp, pic_rtx));
- emit_insn (gen_movsi (pic_offset_table_rtx, pic_tmp));
+ pic_tmp2 = gen_rtx_PLUS (Pmode, global_offset_table, pc_rtx);
+ pic_tmp2 = gen_rtx_CONST (VOIDmode, pic_tmp2);
}
else
- emit_insn (gen_pic_load_addr_thumb (pic_offset_table_rtx, pic_rtx));
- emit_insn (gen_pic_add_dot_plus_four (pic_offset_table_rtx, l1));
+ pic_tmp2 = gen_rtx_CONST (VOIDmode, global_offset_table);
+
+ pic_rtx = gen_rtx_MINUS (Pmode, pic_tmp2, pic_tmp);
+ pic_rtx = gen_rtx_CONST (Pmode, pic_rtx);
+
+ if (TARGET_ARM)
+ {
+ emit_insn (gen_pic_load_addr_arm (pic_reg, pic_rtx));
+ emit_insn (gen_pic_add_dot_plus_eight (pic_reg, pic_reg, labelno));
+ }
+ else if (TARGET_THUMB2)
+ {
+ /* Thumb-2 only allows very limited access to the PC. Calculate the
+ address in a temporary register. */
+ if (arm_pic_register != INVALID_REGNUM)
+ {
+ pic_tmp = gen_rtx_REG (SImode,
+ thumb_find_work_register (saved_regs));
+ }
+ else
+ {
+ gcc_assert (!no_new_pseudos);
+ pic_tmp = gen_reg_rtx (Pmode);
+ }
+
+ emit_insn (gen_pic_load_addr_thumb2 (pic_reg, pic_rtx));
+ emit_insn (gen_pic_load_dot_plus_four (pic_tmp, labelno));
+ emit_insn (gen_addsi3 (pic_reg, pic_reg, pic_tmp));
+ }
+ else /* TARGET_THUMB1 */
+ {
+ if (arm_pic_register != INVALID_REGNUM
+ && REGNO (pic_reg) > LAST_LO_REGNUM)
+ {
+ /* We will have pushed the pic register, so we should always be
+ able to find a work register. */
+ pic_tmp = gen_rtx_REG (SImode,
+ thumb_find_work_register (saved_regs));
+ emit_insn (gen_pic_load_addr_thumb1 (pic_tmp, pic_rtx));
+ emit_insn (gen_movsi (pic_offset_table_rtx, pic_tmp));
+ }
+ else
+ emit_insn (gen_pic_load_addr_thumb1 (pic_reg, pic_rtx));
+ emit_insn (gen_pic_add_dot_plus_four (pic_reg, pic_reg, labelno));
+ }
}
/* Need to emit this whether or not we obey regdecls,
since setjmp/longjmp can cause life info to screw up. */
- emit_insn (gen_rtx_USE (VOIDmode, pic_offset_table_rtx));
+ emit_insn (gen_rtx_USE (VOIDmode, pic_reg));
#endif /* AOF_ASSEMBLER */
}
|| regno == ARG_POINTER_REGNUM);
}
+/* Return TRUE if this rtx is the difference of a symbol and a label,
+ and will reduce to a PC-relative relocation in the object file.
+ Expressions like this can be left alone when generating PIC, rather
+ than forced through the GOT. */
+static int
+pcrel_constant_p (rtx x)
+{
+ if (GET_CODE (x) == MINUS)
+ return symbol_mentioned_p (XEXP (x, 0)) && label_mentioned_p (XEXP (x, 1));
+
+ return FALSE;
+}
+
/* Return nonzero if X is a valid ARM state address operand. */
int
arm_legitimate_address_p (enum machine_mode mode, rtx x, RTX_CODE outer,
&& code == SYMBOL_REF
&& CONSTANT_POOL_ADDRESS_P (x)
&& ! (flag_pic
- && symbol_mentioned_p (get_pool_constant (x))))
+ && symbol_mentioned_p (get_pool_constant (x))
+ && ! pcrel_constant_p (get_pool_constant (x))))
return 1;
return 0;
}
-/* Return nonzero if INDEX is valid for an address index operand in
- ARM state. */
-static int
-arm_legitimate_index_p (enum machine_mode mode, rtx index, RTX_CODE outer,
- int strict_p)
+/* Return nonzero if X is a valid Thumb-2 address operand. */
+int
+thumb2_legitimate_address_p (enum machine_mode mode, rtx x, int strict_p)
{
- HOST_WIDE_INT range;
- enum rtx_code code = GET_CODE (index);
-
- /* Standard coprocessor addressing modes. */
- if (TARGET_HARD_FLOAT
- && (TARGET_FPA || TARGET_MAVERICK)
- && (GET_MODE_CLASS (mode) == MODE_FLOAT
- || (TARGET_MAVERICK && mode == DImode)))
- return (code == CONST_INT && INTVAL (index) < 1024
- && INTVAL (index) > -1024
- && (INTVAL (index) & 3) == 0);
+ bool use_ldrd;
+ enum rtx_code code = GET_CODE (x);
+
+ if (arm_address_register_rtx_p (x, strict_p))
+ return 1;
- if (TARGET_REALLY_IWMMXT && VALID_IWMMXT_REG_MODE (mode))
- return (code == CONST_INT
- && INTVAL (index) < 1024
- && INTVAL (index) > -1024
- && (INTVAL (index) & 3) == 0);
+ use_ldrd = (TARGET_LDRD
+ && (mode == DImode
+ || (mode == DFmode && (TARGET_SOFT_FLOAT || TARGET_VFP))));
- if (arm_address_register_rtx_p (index, strict_p)
- && (GET_MODE_SIZE (mode) <= 4))
- return 1;
+ if (code == POST_INC || code == PRE_DEC
+ || ((code == PRE_INC || code == POST_DEC)
+ && (use_ldrd || GET_MODE_SIZE (mode) <= 4)))
+ return arm_address_register_rtx_p (XEXP (x, 0), strict_p);
- if (mode == DImode || mode == DFmode)
+ else if ((code == POST_MODIFY || code == PRE_MODIFY)
+ && arm_address_register_rtx_p (XEXP (x, 0), strict_p)
+ && GET_CODE (XEXP (x, 1)) == PLUS
+ && rtx_equal_p (XEXP (XEXP (x, 1), 0), XEXP (x, 0)))
{
- if (code == CONST_INT)
- {
- HOST_WIDE_INT val = INTVAL (index);
+ /* Thumb-2 only has autoincrement by constant. */
+ rtx addend = XEXP (XEXP (x, 1), 1);
+ HOST_WIDE_INT offset;
+
+ if (GET_CODE (addend) != CONST_INT)
+ return 0;
+
+ offset = INTVAL(addend);
+ if (GET_MODE_SIZE (mode) <= 4)
+ return (offset > -256 && offset < 256);
+
+ return (use_ldrd && offset > -1024 && offset < 1024
+ && (offset & 3) == 0);
+ }
+
+ /* After reload constants split into minipools will have addresses
+ from a LABEL_REF. */
+ else if (reload_completed
+ && (code == LABEL_REF
+ || (code == CONST
+ && GET_CODE (XEXP (x, 0)) == PLUS
+ && GET_CODE (XEXP (XEXP (x, 0), 0)) == LABEL_REF
+ && GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT)))
+ return 1;
+
+ else if (mode == TImode)
+ return 0;
+
+ else if (code == PLUS)
+ {
+ rtx xop0 = XEXP (x, 0);
+ rtx xop1 = XEXP (x, 1);
+
+ return ((arm_address_register_rtx_p (xop0, strict_p)
+ && thumb2_legitimate_index_p (mode, xop1, strict_p))
+ || (arm_address_register_rtx_p (xop1, strict_p)
+ && thumb2_legitimate_index_p (mode, xop0, strict_p)));
+ }
+
+ else if (GET_MODE_CLASS (mode) != MODE_FLOAT
+ && code == SYMBOL_REF
+ && CONSTANT_POOL_ADDRESS_P (x)
+ && ! (flag_pic
+ && symbol_mentioned_p (get_pool_constant (x))
+ && ! pcrel_constant_p (get_pool_constant (x))))
+ return 1;
+
+ return 0;
+}
+
+/* Return nonzero if INDEX is valid for an address index operand in
+ ARM state. */
+static int
+arm_legitimate_index_p (enum machine_mode mode, rtx index, RTX_CODE outer,
+ int strict_p)
+{
+ HOST_WIDE_INT range;
+ enum rtx_code code = GET_CODE (index);
+
+ /* Standard coprocessor addressing modes. */
+ if (TARGET_HARD_FLOAT
+ && (TARGET_FPA || TARGET_MAVERICK)
+ && (GET_MODE_CLASS (mode) == MODE_FLOAT
+ || (TARGET_MAVERICK && mode == DImode)))
+ return (code == CONST_INT && INTVAL (index) < 1024
+ && INTVAL (index) > -1024
+ && (INTVAL (index) & 3) == 0);
+
+ if (TARGET_REALLY_IWMMXT && VALID_IWMMXT_REG_MODE (mode))
+ return (code == CONST_INT
+ && INTVAL (index) < 1024
+ && INTVAL (index) > -1024
+ && (INTVAL (index) & 3) == 0);
+
+ if (arm_address_register_rtx_p (index, strict_p)
+ && (GET_MODE_SIZE (mode) <= 4))
+ return 1;
+
+ if (mode == DImode || mode == DFmode)
+ {
+ if (code == CONST_INT)
+ {
+ HOST_WIDE_INT val = INTVAL (index);
if (TARGET_LDRD)
return val > -256 && val < 256;
&& INTVAL (index) > -range);
}
-/* Return nonzero if X is valid as a Thumb state base register. */
+/* Return true if OP is a valid index scaling factor for Thumb-2 address
+ index operand. i.e. 1, 2, 4 or 8. */
+static bool
+thumb2_index_mul_operand (rtx op)
+{
+ HOST_WIDE_INT val;
+
+ if (GET_CODE(op) != CONST_INT)
+ return false;
+
+ val = INTVAL(op);
+ return (val == 1 || val == 2 || val == 4 || val == 8);
+}
+
+/* Return nonzero if INDEX is a valid Thumb-2 address index operand. */
+static int
+thumb2_legitimate_index_p (enum machine_mode mode, rtx index, int strict_p)
+{
+ enum rtx_code code = GET_CODE (index);
+
+ /* ??? Combine arm and thumb2 coprocessor addressing modes. */
+ /* Standard coprocessor addressing modes. */
+ if (TARGET_HARD_FLOAT
+ && (TARGET_FPA || TARGET_MAVERICK)
+ && (GET_MODE_CLASS (mode) == MODE_FLOAT
+ || (TARGET_MAVERICK && mode == DImode)))
+ return (code == CONST_INT && INTVAL (index) < 1024
+ && INTVAL (index) > -1024
+ && (INTVAL (index) & 3) == 0);
+
+ if (TARGET_REALLY_IWMMXT && VALID_IWMMXT_REG_MODE (mode))
+ {
+ /* For DImode assume values will usually live in core regs
+ and only allow LDRD addressing modes. */
+ if (!TARGET_LDRD || mode != DImode)
+ return (code == CONST_INT
+ && INTVAL (index) < 1024
+ && INTVAL (index) > -1024
+ && (INTVAL (index) & 3) == 0);
+ }
+
+ if (arm_address_register_rtx_p (index, strict_p)
+ && (GET_MODE_SIZE (mode) <= 4))
+ return 1;
+
+ if (mode == DImode || mode == DFmode)
+ {
+ HOST_WIDE_INT val = INTVAL (index);
+ /* ??? Can we assume ldrd for thumb2? */
+ /* Thumb-2 ldrd only has reg+const addressing modes. */
+ if (code != CONST_INT)
+ return 0;
+
+ /* ldrd supports offsets of +-1020.
+ However the ldr fallback does not. */
+ return val > -256 && val < 256 && (val & 3) == 0;
+ }
+
+ if (code == MULT)
+ {
+ rtx xiop0 = XEXP (index, 0);
+ rtx xiop1 = XEXP (index, 1);
+
+ return ((arm_address_register_rtx_p (xiop0, strict_p)
+ && thumb2_index_mul_operand (xiop1))
+ || (arm_address_register_rtx_p (xiop1, strict_p)
+ && thumb2_index_mul_operand (xiop0)));
+ }
+ else if (code == ASHIFT)
+ {
+ rtx op = XEXP (index, 1);
+
+ return (arm_address_register_rtx_p (XEXP (index, 0), strict_p)
+ && GET_CODE (op) == CONST_INT
+ && INTVAL (op) > 0
+ && INTVAL (op) <= 3);
+ }
+
+ return (code == CONST_INT
+ && INTVAL (index) < 4096
+ && INTVAL (index) > -256);
+}
+
+/* Return nonzero if X is valid as a 16-bit Thumb state base register. */
static int
-thumb_base_register_rtx_p (rtx x, enum machine_mode mode, int strict_p)
+thumb1_base_register_rtx_p (rtx x, enum machine_mode mode, int strict_p)
{
int regno;
regno = REGNO (x);
if (strict_p)
- return THUMB_REGNO_MODE_OK_FOR_BASE_P (regno, mode);
+ return THUMB1_REGNO_MODE_OK_FOR_BASE_P (regno, mode);
return (regno <= LAST_LO_REGNUM
|| regno > LAST_VIRTUAL_REGISTER
/* Return nonzero if x is a legitimate index register. This is the case
for any base register that can access a QImode object. */
inline static int
-thumb_index_register_rtx_p (rtx x, int strict_p)
+thumb1_index_register_rtx_p (rtx x, int strict_p)
{
- return thumb_base_register_rtx_p (x, QImode, strict_p);
+ return thumb1_base_register_rtx_p (x, QImode, strict_p);
}
-/* Return nonzero if x is a legitimate Thumb-state address.
+/* Return nonzero if x is a legitimate 16-bit Thumb-state address.
The AP may be eliminated to either the SP or the FP, so we use the
least common denominator, e.g. SImode, and offsets from 0 to 64.
reload pass starts. This is so that eliminating such addresses
into stack based ones won't produce impossible code. */
int
-thumb_legitimate_address_p (enum machine_mode mode, rtx x, int strict_p)
+thumb1_legitimate_address_p (enum machine_mode mode, rtx x, int strict_p)
{
/* ??? Not clear if this is right. Experiment. */
if (GET_MODE_SIZE (mode) < 4
return 0;
/* Accept any base register. SP only in SImode or larger. */
- else if (thumb_base_register_rtx_p (x, mode, strict_p))
+ else if (thumb1_base_register_rtx_p (x, mode, strict_p))
return 1;
/* This is PC relative data before arm_reorg runs. */
else if (GET_MODE_SIZE (mode) >= 4 && CONSTANT_P (x)
&& GET_CODE (x) == SYMBOL_REF
- && CONSTANT_POOL_ADDRESS_P (x) && ! flag_pic)
+ && CONSTANT_POOL_ADDRESS_P (x) && !flag_pic)
return 1;
/* This is PC relative data after arm_reorg runs. */
/* Post-inc indexing only supported for SImode and larger. */
else if (GET_CODE (x) == POST_INC && GET_MODE_SIZE (mode) >= 4
- && thumb_index_register_rtx_p (XEXP (x, 0), strict_p))
+ && thumb1_index_register_rtx_p (XEXP (x, 0), strict_p))
return 1;
else if (GET_CODE (x) == PLUS)
if (GET_MODE_SIZE (mode) <= 4
&& XEXP (x, 0) != frame_pointer_rtx
&& XEXP (x, 1) != frame_pointer_rtx
- && thumb_index_register_rtx_p (XEXP (x, 0), strict_p)
- && thumb_index_register_rtx_p (XEXP (x, 1), strict_p))
+ && thumb1_index_register_rtx_p (XEXP (x, 0), strict_p)
+ && thumb1_index_register_rtx_p (XEXP (x, 1), strict_p))
return 1;
/* REG+const has 5-7 bit offset for non-SP registers. */
- else if ((thumb_index_register_rtx_p (XEXP (x, 0), strict_p)
+ else if ((thumb1_index_register_rtx_p (XEXP (x, 0), strict_p)
|| XEXP (x, 0) == arg_pointer_rtx)
&& GET_CODE (XEXP (x, 1)) == CONST_INT
&& thumb_legitimate_offset_p (mode, INTVAL (XEXP (x, 1))))
return 1;
- /* REG+const has 10 bit offset for SP, but only SImode and
+ /* REG+const has 10-bit offset for SP, but only SImode and
larger is supported. */
/* ??? Should probably check for DI/DFmode overflow here
just like GO_IF_LEGITIMATE_OFFSET does. */
return 1;
else if (GET_CODE (XEXP (x, 0)) == REG
- && REGNO (XEXP (x, 0)) == FRAME_POINTER_REGNUM
+ && (REGNO (XEXP (x, 0)) == FRAME_POINTER_REGNUM
+ || REGNO (XEXP (x, 0)) == ARG_POINTER_REGNUM
+ || (REGNO (XEXP (x, 0)) >= FIRST_VIRTUAL_REGISTER
+ && REGNO (XEXP (x, 0)) <= LAST_VIRTUAL_REGISTER))
&& GET_MODE_SIZE (mode) >= 4
&& GET_CODE (XEXP (x, 1)) == CONST_INT
&& (INTVAL (XEXP (x, 1)) & 3) == 0)
&& GET_MODE_SIZE (mode) == 4
&& GET_CODE (x) == SYMBOL_REF
&& CONSTANT_POOL_ADDRESS_P (x)
- && !(flag_pic
- && symbol_mentioned_p (get_pool_constant (x))))
+ && ! (flag_pic
+ && symbol_mentioned_p (get_pool_constant (x))
+ && ! pcrel_constant_p (get_pool_constant (x))))
return 1;
return 0;
}
}
+/* Build the SYMBOL_REF for __tls_get_addr. */
+
+static GTY(()) rtx tls_get_addr_libfunc;
+
+static rtx
+get_tls_get_addr (void)
+{
+ if (!tls_get_addr_libfunc)
+ tls_get_addr_libfunc = init_one_libfunc ("__tls_get_addr");
+ return tls_get_addr_libfunc;
+}
+
+static rtx
+arm_load_tp (rtx target)
+{
+ if (!target)
+ target = gen_reg_rtx (SImode);
+
+ if (TARGET_HARD_TP)
+ {
+ /* Can return in any reg. */
+ emit_insn (gen_load_tp_hard (target));
+ }
+ else
+ {
+ /* Always returned in r0. Immediately copy the result into a pseudo,
+ otherwise other uses of r0 (e.g. setting up function arguments) may
+ clobber the value. */
+
+ rtx tmp;
+
+ emit_insn (gen_load_tp_soft ());
+
+ tmp = gen_rtx_REG (SImode, 0);
+ emit_move_insn (target, tmp);
+ }
+ return target;
+}
+
+static rtx
+load_tls_operand (rtx x, rtx reg)
+{
+ rtx tmp;
+
+ if (reg == NULL_RTX)
+ reg = gen_reg_rtx (SImode);
+
+ tmp = gen_rtx_CONST (SImode, x);
+
+ emit_move_insn (reg, tmp);
+
+ return reg;
+}
+
+static rtx
+arm_call_tls_get_addr (rtx x, rtx reg, rtx *valuep, int reloc)
+{
+ rtx insns, label, labelno, sum;
+
+ start_sequence ();
+
+ labelno = GEN_INT (pic_labelno++);
+ label = gen_rtx_UNSPEC (Pmode, gen_rtvec (1, labelno), UNSPEC_PIC_LABEL);
+ label = gen_rtx_CONST (VOIDmode, label);
+
+ sum = gen_rtx_UNSPEC (Pmode,
+ gen_rtvec (4, x, GEN_INT (reloc), label,
+ GEN_INT (TARGET_ARM ? 8 : 4)),
+ UNSPEC_TLS);
+ reg = load_tls_operand (sum, reg);
+
+ if (TARGET_ARM)
+ emit_insn (gen_pic_add_dot_plus_eight (reg, reg, labelno));
+ else if (TARGET_THUMB2)
+ {
+ rtx tmp;
+ /* Thumb-2 only allows very limited access to the PC. Calculate
+ the address in a temporary register. */
+ tmp = gen_reg_rtx (SImode);
+ emit_insn (gen_pic_load_dot_plus_four (tmp, labelno));
+ emit_insn (gen_addsi3(reg, reg, tmp));
+ }
+ else /* TARGET_THUMB1 */
+ emit_insn (gen_pic_add_dot_plus_four (reg, reg, labelno));
+
+ *valuep = emit_library_call_value (get_tls_get_addr (), NULL_RTX, LCT_PURE, /* LCT_CONST? */
+ Pmode, 1, reg, Pmode);
+
+ insns = get_insns ();
+ end_sequence ();
+
+ return insns;
+}
+
+rtx
+legitimize_tls_address (rtx x, rtx reg)
+{
+ rtx dest, tp, label, labelno, sum, insns, ret, eqv, addend;
+ unsigned int model = SYMBOL_REF_TLS_MODEL (x);
+
+ switch (model)
+ {
+ case TLS_MODEL_GLOBAL_DYNAMIC:
+ insns = arm_call_tls_get_addr (x, reg, &ret, TLS_GD32);
+ dest = gen_reg_rtx (Pmode);
+ emit_libcall_block (insns, dest, ret, x);
+ return dest;
+
+ case TLS_MODEL_LOCAL_DYNAMIC:
+ insns = arm_call_tls_get_addr (x, reg, &ret, TLS_LDM32);
+
+ /* Attach a unique REG_EQUIV, to allow the RTL optimizers to
+ share the LDM result with other LD model accesses. */
+ eqv = gen_rtx_UNSPEC (Pmode, gen_rtvec (1, const1_rtx),
+ UNSPEC_TLS);
+ dest = gen_reg_rtx (Pmode);
+ emit_libcall_block (insns, dest, ret, eqv);
+
+ /* Load the addend. */
+ addend = gen_rtx_UNSPEC (Pmode, gen_rtvec (2, x, GEN_INT (TLS_LDO32)),
+ UNSPEC_TLS);
+ addend = force_reg (SImode, gen_rtx_CONST (SImode, addend));
+ return gen_rtx_PLUS (Pmode, dest, addend);
+
+ case TLS_MODEL_INITIAL_EXEC:
+ labelno = GEN_INT (pic_labelno++);
+ label = gen_rtx_UNSPEC (Pmode, gen_rtvec (1, labelno), UNSPEC_PIC_LABEL);
+ label = gen_rtx_CONST (VOIDmode, label);
+ sum = gen_rtx_UNSPEC (Pmode,
+ gen_rtvec (4, x, GEN_INT (TLS_IE32), label,
+ GEN_INT (TARGET_ARM ? 8 : 4)),
+ UNSPEC_TLS);
+ reg = load_tls_operand (sum, reg);
+
+ if (TARGET_ARM)
+ emit_insn (gen_tls_load_dot_plus_eight (reg, reg, labelno));
+ else if (TARGET_THUMB2)
+ {
+ rtx tmp;
+ /* Thumb-2 only allows very limited access to the PC. Calculate
+ the address in a temporary register. */
+ tmp = gen_reg_rtx (SImode);
+ emit_insn (gen_pic_load_dot_plus_four (tmp, labelno));
+ emit_insn (gen_addsi3(reg, reg, tmp));
+ emit_move_insn (reg, gen_const_mem (SImode, reg));
+ }
+ else
+ {
+ emit_insn (gen_pic_add_dot_plus_four (reg, reg, labelno));
+ emit_move_insn (reg, gen_const_mem (SImode, reg));
+ }
+
+ tp = arm_load_tp (NULL_RTX);
+
+ return gen_rtx_PLUS (Pmode, tp, reg);
+
+ case TLS_MODEL_LOCAL_EXEC:
+ tp = arm_load_tp (NULL_RTX);
+
+ reg = gen_rtx_UNSPEC (Pmode,
+ gen_rtvec (2, x, GEN_INT (TLS_LE32)),
+ UNSPEC_TLS);
+ reg = force_reg (SImode, gen_rtx_CONST (SImode, reg));
+
+ return gen_rtx_PLUS (Pmode, tp, reg);
+
+ default:
+ abort ();
+ }
+}
+
/* Try machine-dependent ways of modifying an illegitimate address
to be legitimate. If we find one, return the new, valid address. */
rtx
arm_legitimize_address (rtx x, rtx orig_x, enum machine_mode mode)
{
+ if (arm_tls_symbol_p (x))
+ return legitimize_tls_address (x, NULL_RTX);
+
if (GET_CODE (x) == PLUS)
{
rtx xop0 = XEXP (x, 0);
}
base_reg = gen_reg_rtx (SImode);
- val = force_operand (gen_rtx_PLUS (SImode, xop0,
- GEN_INT (n)), NULL_RTX);
+ val = force_operand (plus_constant (xop0, n), NULL_RTX);
emit_move_insn (base_reg, val);
- x = (low_n == 0 ? base_reg
- : gen_rtx_PLUS (SImode, base_reg, GEN_INT (low_n)));
+ x = plus_constant (base_reg, low_n);
}
else if (xop0 != XEXP (x, 0) || xop1 != XEXP (x, 1))
x = gen_rtx_PLUS (SImode, xop0, xop1);
x = gen_rtx_MINUS (SImode, xop0, xop1);
}
+ /* Make sure to take full advantage of the pre-indexed addressing mode
+ with absolute addresses which often allows for the base register to
+ be factorized for multiple adjacent memory references, and it might
+ even allows for the mini pool to be avoided entirely. */
+ else if (GET_CODE (x) == CONST_INT && optimize > 0)
+ {
+ unsigned int bits;
+ HOST_WIDE_INT mask, base, index;
+ rtx base_reg;
+
+ /* ldr and ldrb can use a 12-bit index, ldrsb and the rest can only
+ use a 8-bit index. So let's use a 12-bit index for SImode only and
+ hope that arm_gen_constant will enable ldrb to use more bits. */
+ bits = (mode == SImode) ? 12 : 8;
+ mask = (1 << bits) - 1;
+ base = INTVAL (x) & ~mask;
+ index = INTVAL (x) & mask;
+ if (bit_count (base & 0xffffffff) > (32 - bits)/2)
+ {
+ /* It'll most probably be more efficient to generate the base
+ with more bits set and use a negative index instead. */
+ base |= mask;
+ index -= mask;
+ }
+ base_reg = force_reg (SImode, GEN_INT (base));
+ x = plus_constant (base_reg, index);
+ }
+
if (flag_pic)
{
/* We need to find and carefully transform any SYMBOL and LABEL
rtx
thumb_legitimize_address (rtx x, rtx orig_x, enum machine_mode mode)
{
+ if (arm_tls_symbol_p (x))
+ return legitimize_tls_address (x, NULL_RTX);
+
if (GET_CODE (x) == PLUS
&& GET_CODE (XEXP (x, 1)) == CONST_INT
&& (INTVAL (XEXP (x, 1)) >= 32 * GET_MODE_SIZE (mode)
return x;
}
-\f
+rtx
+thumb_legitimize_reload_address (rtx *x_p,
+ enum machine_mode mode,
+ int opnum, int type,
+ int ind_levels ATTRIBUTE_UNUSED)
+{
+ rtx x = *x_p;
+
+ if (GET_CODE (x) == PLUS
+ && GET_MODE_SIZE (mode) < 4
+ && REG_P (XEXP (x, 0))
+ && XEXP (x, 0) == stack_pointer_rtx
+ && GET_CODE (XEXP (x, 1)) == CONST_INT
+ && !thumb_legitimate_offset_p (mode, INTVAL (XEXP (x, 1))))
+ {
+ rtx orig_x = x;
+
+ x = copy_rtx (x);
+ push_reload (orig_x, NULL_RTX, x_p, NULL, MODE_BASE_REG_CLASS (mode),
+ Pmode, VOIDmode, 0, 0, opnum, type);
+ return x;
+ }
+
+ /* If both registers are hi-regs, then it's better to reload the
+ entire expression rather than each register individually. That
+ only requires one reload register rather than two. */
+ if (GET_CODE (x) == PLUS
+ && REG_P (XEXP (x, 0))
+ && REG_P (XEXP (x, 1))
+ && !REG_MODE_OK_FOR_REG_BASE_P (XEXP (x, 0), mode)
+ && !REG_MODE_OK_FOR_REG_BASE_P (XEXP (x, 1), mode))
+ {
+ rtx orig_x = x;
+
+ x = copy_rtx (x);
+ push_reload (orig_x, NULL_RTX, x_p, NULL, MODE_BASE_REG_CLASS (mode),
+ Pmode, VOIDmode, 0, 0, opnum, type);
+ return x;
+ }
+
+ return NULL;
+}
+
+/* Test for various thread-local symbols. */
+
+/* Return TRUE if X is a thread-local symbol. */
+
+static bool
+arm_tls_symbol_p (rtx x)
+{
+ if (! TARGET_HAVE_TLS)
+ return false;
+
+ if (GET_CODE (x) != SYMBOL_REF)
+ return false;
+
+ return SYMBOL_REF_TLS_MODEL (x) != 0;
+}
+
+/* Helper for arm_tls_referenced_p. */
+
+static int
+arm_tls_operand_p_1 (rtx *x, void *data ATTRIBUTE_UNUSED)
+{
+ if (GET_CODE (*x) == SYMBOL_REF)
+ return SYMBOL_REF_TLS_MODEL (*x) != 0;
+
+ /* Don't recurse into UNSPEC_TLS looking for TLS symbols; these are
+ TLS offsets, not real symbol references. */
+ if (GET_CODE (*x) == UNSPEC
+ && XINT (*x, 1) == UNSPEC_TLS)
+ return -1;
+
+ return 0;
+}
+
+/* Return TRUE if X contains any TLS symbol references. */
+
+bool
+arm_tls_referenced_p (rtx x)
+{
+ if (! TARGET_HAVE_TLS)
+ return false;
+ return for_each_rtx (&x, arm_tls_operand_p_1, NULL);
+}
+\f
#define REG_OR_SUBREG_REG(X) \
(GET_CODE (X) == REG \
|| (GET_CODE (X) == SUBREG && GET_CODE (SUBREG_REG (X)) == REG))
#define COSTS_N_INSNS(N) ((N) * 4 - 2)
#endif
static inline int
-thumb_rtx_costs (rtx x, enum rtx_code code, enum rtx_code outer)
+thumb1_rtx_costs (rtx x, enum rtx_code code, enum rtx_code outer)
{
enum machine_mode mode = GET_MODE (x);
/* Worker routine for arm_rtx_costs. */
+/* ??? This needs updating for thumb2. */
static inline int
arm_rtx_costs_1 (rtx x, enum rtx_code code, enum rtx_code outer)
{
? 0 : 4));
case MINUS:
+ if (GET_CODE (XEXP (x, 1)) == MULT && mode == SImode && arm_arch_thumb2)
+ {
+ extra_cost = rtx_cost (XEXP (x, 1), code);
+ if (!REG_OR_SUBREG_REG (XEXP (x, 0)))
+ extra_cost += 4 * ARM_NUM_REGS (mode);
+ return extra_cost;
+ }
+
if (mode == DImode)
return (4 + (REG_OR_SUBREG_REG (XEXP (x, 1)) ? 0 : 8)
+ ((REG_OR_SUBREG_REG (XEXP (x, 0))
/* Fall through */
case PLUS:
+ if (GET_CODE (XEXP (x, 0)) == MULT)
+ {
+ extra_cost = rtx_cost (XEXP (x, 0), code);
+ if (!REG_OR_SUBREG_REG (XEXP (x, 1)))
+ extra_cost += 4 * ARM_NUM_REGS (mode);
+ return extra_cost;
+ }
+
if (GET_MODE_CLASS (mode) == MODE_FLOAT)
return (2 + (REG_OR_SUBREG_REG (XEXP (x, 0)) ? 0 : 8)
+ ((REG_OR_SUBREG_REG (XEXP (x, 1))
return 4 + (mode == DImode ? 4 : 0);
case SIGN_EXTEND:
+ /* ??? value extensions are cheaper on armv6. */
if (GET_MODE (XEXP (x, 0)) == QImode)
return (4 + (mode == DImode ? 4 : 0)
+ (GET_CODE (XEXP (x, 0)) == MEM ? 10 : 0));
return 6;
case CONST_DOUBLE:
- if (arm_const_double_rtx (x))
+ if (arm_const_double_rtx (x) || vfp3_const_double_rtx (x))
return outer == SET ? 2 : -1;
else if ((outer == COMPARE || outer == PLUS)
&& neg_const_double_rtx_ok_for_fpa (x))
if (TARGET_THUMB)
{
/* XXX TBD. For now, use the standard costs. */
- *total = thumb_rtx_costs (x, code, outer_code);
+ *total = thumb1_rtx_costs (x, code, outer_code);
return true;
}
}
}
-/* RTX costs for cores with a slow MUL implementation. */
+/* RTX costs for cores with a slow MUL implementation. Thumb-2 is not
+ supported on any "slowmul" cores, so it can be ignored. */
static bool
arm_slowmul_rtx_costs (rtx x, int code, int outer_code, int *total)
if (TARGET_THUMB)
{
- *total = thumb_rtx_costs (x, code, outer_code);
+ *total = thumb1_rtx_costs (x, code, outer_code);
return true;
}
{
enum machine_mode mode = GET_MODE (x);
- if (TARGET_THUMB)
+ if (TARGET_THUMB1)
{
- *total = thumb_rtx_costs (x, code, outer_code);
+ *total = thumb1_rtx_costs (x, code, outer_code);
return true;
}
+ /* ??? should thumb2 use different costs? */
switch (code)
{
case MULT:
}
-/* RTX cost for XScale CPUs. */
+/* RTX cost for XScale CPUs. Thumb-2 is not supported on any xscale cores,
+ so it can be ignored. */
static bool
arm_xscale_rtx_costs (rtx x, int code, int outer_code, int *total)
if (TARGET_THUMB)
{
- *total = thumb_rtx_costs (x, code, outer_code);
+ *total = thumb1_rtx_costs (x, code, outer_code);
return true;
}
int nonreg_cost;
int cost;
- if (TARGET_THUMB)
+ if (TARGET_THUMB1)
{
switch (code)
{
return true;
default:
- *total = thumb_rtx_costs (x, code, outer_code);
+ *total = thumb1_rtx_costs (x, code, outer_code);
return true;
}
}
static int
arm_address_cost (rtx x)
{
- return TARGET_ARM ? arm_arm_address_cost (x) : arm_thumb_address_cost (x);
+ return TARGET_32BIT ? arm_arm_address_cost (x) : arm_thumb_address_cost (x);
}
static int
return 0;
}
-\f
-/* Predicates for `match_operand' and `match_operator'. */
-/* Return nonzero if OP is a valid Cirrus memory address pattern. */
-int
-cirrus_memory_offset (rtx op)
-{
- /* Reject eliminable registers. */
- if (! (reload_in_progress || reload_completed)
- && ( reg_mentioned_p (frame_pointer_rtx, op)
- || reg_mentioned_p (arg_pointer_rtx, op)
- || reg_mentioned_p (virtual_incoming_args_rtx, op)
- || reg_mentioned_p (virtual_outgoing_args_rtx, op)
- || reg_mentioned_p (virtual_stack_dynamic_rtx, op)
- || reg_mentioned_p (virtual_stack_vars_rtx, op)))
- return 0;
- if (GET_CODE (op) == MEM)
- {
- rtx ind;
+/* VFPv3 has a fairly wide range of representable immediates, formed from
+ "quarter-precision" floating-point values. These can be evaluated using this
+ formula (with ^ for exponentiation):
+
+ -1^s * n * 2^-r
+
+ Where 's' is a sign bit (0/1), 'n' and 'r' are integers such that
+ 16 <= n <= 31 and 0 <= r <= 7.
+
+ These values are mapped onto an 8-bit integer ABCDEFGH s.t.
+
+ - A (most-significant) is the sign bit.
+ - BCD are the exponent (encoded as r XOR 3).
+ - EFGH are the mantissa (encoded as n - 16).
+*/
+
+/* Return an integer index for a VFPv3 immediate operand X suitable for the
+ fconst[sd] instruction, or -1 if X isn't suitable. */
+static int
+vfp3_const_double_index (rtx x)
+{
+ REAL_VALUE_TYPE r, m;
+ int sign, exponent;
+ unsigned HOST_WIDE_INT mantissa, mant_hi;
+ unsigned HOST_WIDE_INT mask;
+ int point_pos = 2 * HOST_BITS_PER_WIDE_INT - 1;
+
+ if (!TARGET_VFP3 || GET_CODE (x) != CONST_DOUBLE)
+ return -1;
+
+ REAL_VALUE_FROM_CONST_DOUBLE (r, x);
+
+ /* We can't represent these things, so detect them first. */
+ if (REAL_VALUE_ISINF (r) || REAL_VALUE_ISNAN (r) || REAL_VALUE_MINUS_ZERO (r))
+ return -1;
+
+ /* Extract sign, exponent and mantissa. */
+ sign = REAL_VALUE_NEGATIVE (r) ? 1 : 0;
+ r = REAL_VALUE_ABS (r);
+ exponent = REAL_EXP (&r);
+ /* For the mantissa, we expand into two HOST_WIDE_INTS, apart from the
+ highest (sign) bit, with a fixed binary point at bit point_pos.
+ WARNING: If there's ever a VFP version which uses more than 2 * H_W_I - 1
+ bits for the mantissa, this may fail (low bits would be lost). */
+ real_ldexp (&m, &r, point_pos - exponent);
+ REAL_VALUE_TO_INT (&mantissa, &mant_hi, m);
+
+ /* If there are bits set in the low part of the mantissa, we can't
+ represent this value. */
+ if (mantissa != 0)
+ return -1;
+
+ /* Now make it so that mantissa contains the most-significant bits, and move
+ the point_pos to indicate that the least-significant bits have been
+ discarded. */
+ point_pos -= HOST_BITS_PER_WIDE_INT;
+ mantissa = mant_hi;
+
+ /* We can permit four significant bits of mantissa only, plus a high bit
+ which is always 1. */
+ mask = ((unsigned HOST_WIDE_INT)1 << (point_pos - 5)) - 1;
+ if ((mantissa & mask) != 0)
+ return -1;
+
+ /* Now we know the mantissa is in range, chop off the unneeded bits. */
+ mantissa >>= point_pos - 5;
+
+ /* The mantissa may be zero. Disallow that case. (It's possible to load the
+ floating-point immediate zero with Neon using an integer-zero load, but
+ that case is handled elsewhere.) */
+ if (mantissa == 0)
+ return -1;
+
+ gcc_assert (mantissa >= 16 && mantissa <= 31);
+
+ /* The value of 5 here would be 4 if GCC used IEEE754-like encoding (where
+ normalised significands are in the range [1, 2). (Our mantissa is shifted
+ left 4 places at this point relative to normalised IEEE754 values). GCC
+ internally uses [0.5, 1) (see real.c), so the exponent returned from
+ REAL_EXP must be altered. */
+ exponent = 5 - exponent;
+
+ if (exponent < 0 || exponent > 7)
+ return -1;
+
+ /* Sign, mantissa and exponent are now in the correct form to plug into the
+ formulae described in the comment above. */
+ return (sign << 7) | ((exponent ^ 3) << 4) | (mantissa - 16);
+}
+
+/* Return TRUE if rtx X is a valid immediate VFPv3 constant. */
+int
+vfp3_const_double_rtx (rtx x)
+{
+ if (!TARGET_VFP3)
+ return 0;
+
+ return vfp3_const_double_index (x) != -1;
+}
+
+\f
+/* Predicates for `match_operand' and `match_operator'. */
+
+/* Return nonzero if OP is a valid Cirrus memory address pattern. */
+int
+cirrus_memory_offset (rtx op)
+{
+ /* Reject eliminable registers. */
+ if (! (reload_in_progress || reload_completed)
+ && ( reg_mentioned_p (frame_pointer_rtx, op)
+ || reg_mentioned_p (arg_pointer_rtx, op)
+ || reg_mentioned_p (virtual_incoming_args_rtx, op)
+ || reg_mentioned_p (virtual_outgoing_args_rtx, op)
+ || reg_mentioned_p (virtual_stack_dynamic_rtx, op)
+ || reg_mentioned_p (virtual_stack_vars_rtx, op)))
+ return 0;
+
+ if (GET_CODE (op) == MEM)
+ {
+ rtx ind;
ind = XEXP (op, 0);
}
/* Return TRUE if OP is a valid coprocessor memory address pattern.
- WB if true if writeback address modes are allowed. */
+ WB is true if full writeback address modes are allowed and is false
+ if limited writeback address modes (POST_INC and PRE_DEC) are
+ allowed. */
int
arm_coproc_mem_operand (rtx op, bool wb)
if (GET_CODE (ind) == REG)
return arm_address_register_rtx_p (ind, 0);
- /* Autoincremment addressing modes. */
- if (wb
- && (GET_CODE (ind) == PRE_INC
- || GET_CODE (ind) == POST_INC
- || GET_CODE (ind) == PRE_DEC
- || GET_CODE (ind) == POST_DEC))
+ /* Autoincremment addressing modes. POST_INC and PRE_DEC are
+ acceptable in any case (subject to verification by
+ arm_address_register_rtx_p). We need WB to be true to accept
+ PRE_INC and POST_DEC. */
+ if (GET_CODE (ind) == POST_INC
+ || GET_CODE (ind) == PRE_DEC
+ || (wb
+ && (GET_CODE (ind) == PRE_INC
+ || GET_CODE (ind) == POST_DEC)))
return arm_address_register_rtx_p (XEXP (ind, 0), 0);
if (wb
}
/* Return GENERAL_REGS if a scratch register required to reload x to/from
- VFP registers. Otherwise return NO_REGS. */
+ coprocessor registers. Otherwise return NO_REGS. */
enum reg_class
-vfp_secondary_reload_class (enum machine_mode mode, rtx x)
+coproc_secondary_reload_class (enum machine_mode mode, rtx x, bool wb)
{
- if (arm_coproc_mem_operand (x, FALSE) || s_register_operand (x, mode))
+ if (arm_coproc_mem_operand (x, wb) || s_register_operand (x, mode))
return NO_REGS;
return GENERAL_REGS;
if (GET_CODE (x) == SYMBOL_REF)
return 1;
+ /* UNSPEC_TLS entries for a symbol include the SYMBOL_REF, but they
+ are constant offsets, not symbols. */
+ if (GET_CODE (x) == UNSPEC && XINT (x, 1) == UNSPEC_TLS)
+ return 0;
+
fmt = GET_RTX_FORMAT (GET_CODE (x));
for (i = GET_RTX_LENGTH (GET_CODE (x)) - 1; i >= 0; i--)
if (GET_CODE (x) == LABEL_REF)
return 1;
+ /* UNSPEC_TLS entries for a symbol include a LABEL_REF for the referencing
+ instruction, but they are constant offsets, not symbols. */
+ if (GET_CODE (x) == UNSPEC && XINT (x, 1) == UNSPEC_TLS)
+ return 0;
+
fmt = GET_RTX_FORMAT (GET_CODE (x));
for (i = GET_RTX_LENGTH (GET_CODE (x)) - 1; i >= 0; i--)
{
return 0;
}
+int
+tls_mentioned_p (rtx x)
+{
+ switch (GET_CODE (x))
+ {
+ case CONST:
+ return tls_mentioned_p (XEXP (x, 0));
+
+ case UNSPEC:
+ if (XINT (x, 1) == UNSPEC_TLS)
+ return 1;
+
+ default:
+ return 0;
+ }
+}
+
+/* Must not copy a SET whose source operand is PC-relative. */
+
+static bool
+arm_cannot_copy_insn_p (rtx insn)
+{
+ rtx pat = PATTERN (insn);
+
+ if (GET_CODE (pat) == SET)
+ {
+ rtx rhs = SET_SRC (pat);
+
+ if (GET_CODE (rhs) == UNSPEC
+ && XINT (rhs, 1) == UNSPEC_PIC_BASE)
+ return TRUE;
+
+ if (GET_CODE (rhs) == MEM
+ && GET_CODE (XEXP (rhs, 0)) == UNSPEC
+ && XINT (XEXP (rhs, 0), 1) == UNSPEC_PIC_BASE)
+ return TRUE;
+ }
+
+ return FALSE;
+}
+
enum rtx_code
minmax_code (rtx x)
{
if (unsorted_offsets[order[0]] == 0)
return 1; /* ldmia */
- if (unsorted_offsets[order[0]] == 4)
+ if (TARGET_ARM && unsorted_offsets[order[0]] == 4)
return 2; /* ldmib */
- if (unsorted_offsets[order[nops - 1]] == 0)
+ if (TARGET_ARM && unsorted_offsets[order[nops - 1]] == 0)
return 3; /* ldmda */
if (unsorted_offsets[order[nops - 1]] == -4)
switch (load_multiple_sequence (operands, nops, regs, &base_reg, &offset))
{
case 1:
- strcpy (buf, "ldm%?ia\t");
+ strcpy (buf, "ldm%(ia%)\t");
break;
case 2:
- strcpy (buf, "ldm%?ib\t");
+ strcpy (buf, "ldm%(ib%)\t");
break;
case 3:
- strcpy (buf, "ldm%?da\t");
+ strcpy (buf, "ldm%(da%)\t");
break;
case 4:
- strcpy (buf, "ldm%?db\t");
+ strcpy (buf, "ldm%(db%)\t");
break;
case 5:
(long) -offset);
output_asm_insn (buf, operands);
base_reg = regs[0];
- strcpy (buf, "ldm%?ia\t");
+ strcpy (buf, "ldm%(ia%)\t");
break;
default:
switch (store_multiple_sequence (operands, nops, regs, &base_reg, &offset))
{
case 1:
- strcpy (buf, "stm%?ia\t");
+ strcpy (buf, "stm%(ia%)\t");
break;
case 2:
- strcpy (buf, "stm%?ib\t");
+ strcpy (buf, "stm%(ib%)\t");
break;
case 3:
- strcpy (buf, "stm%?da\t");
+ strcpy (buf, "stm%(da%)\t");
break;
case 4:
- strcpy (buf, "stm%?db\t");
+ strcpy (buf, "stm%(db%)\t");
break;
default:
output_asm_insn (buf, operands);
return "";
}
-
\f
/* Routines for use in generating RTL. */
if (write_back)
{
XVECEXP (result, 0, 0)
- = gen_rtx_SET (GET_MODE (from), from,
- plus_constant (from, count * 4 * sign));
+ = gen_rtx_SET (VOIDmode, from, plus_constant (from, count * 4 * sign));
i = 1;
count++;
}
if (write_back)
{
XVECEXP (result, 0, 0)
- = gen_rtx_SET (GET_MODE (to), to,
+ = gen_rtx_SET (VOIDmode, to,
plus_constant (to, count * 4 * sign));
i = 1;
count++;
return 1;
}
-/* Generate a memory reference for a half word, such that it will be loaded
- into the top 16 bits of the word. We can assume that the address is
- known to be alignable and of the form reg, or plus (reg, const). */
-
-rtx
-arm_gen_rotated_half_load (rtx memref)
-{
- HOST_WIDE_INT offset = 0;
- rtx base = XEXP (memref, 0);
-
- if (GET_CODE (base) == PLUS)
- {
- offset = INTVAL (XEXP (base, 1));
- base = XEXP (base, 0);
- }
-
- /* If we aren't allowed to generate unaligned addresses, then fail. */
- if ((BYTES_BIG_ENDIAN ? 1 : 0) ^ ((offset & 2) == 0))
- return NULL;
-
- base = gen_rtx_MEM (SImode, plus_constant (base, offset & ~2));
-
- if ((BYTES_BIG_ENDIAN ? 1 : 0) ^ ((offset & 2) == 2))
- return base;
-
- return gen_rtx_ROTATE (SImode, base, GEN_INT (16));
-}
-
/* Select a dominance comparison mode if possible for a test of the general
form (OP (COND_OR (X) (Y)) (const_int 0)). We support three forms.
COND_OR == DOM_CC_X_AND_Y => (X && Y)
case LT:
if (cond_or == DOM_CC_X_AND_Y)
return CC_DLTmode;
-
+
switch (cond2)
{
case LT:
case GTU:
if (cond_or == DOM_CC_X_AND_Y)
return CC_DGTUmode;
-
+
switch (cond2)
{
case GTU:
/* An operation (on Thumb) where we want to test for a single bit.
This is done by shifting that bit up into the top bit of a
scratch register; we can then branch on the sign bit. */
- if (TARGET_THUMB
+ if (TARGET_THUMB1
&& GET_MODE (x) == SImode
&& (op == EQ || op == NE)
- && (GET_CODE (x) == ZERO_EXTRACT))
+ && GET_CODE (x) == ZERO_EXTRACT
+ && XEXP (x, 1) == const1_rtx)
return CC_Nmode;
/* An operation that sets the condition codes as a side-effect, the
V flag is not set correctly, so we can only use comparisons where
this doesn't matter. (For LT and GE we can use "mi" and "pl"
instead.) */
+ /* ??? Does the ZERO_EXTRACT case really apply to thumb2? */
if (GET_MODE (x) == SImode
&& y == const0_rtx
&& (op == EQ || op == NE || op == LT || op == GE)
|| GET_CODE (x) == LSHIFTRT
|| GET_CODE (x) == ASHIFT || GET_CODE (x) == ASHIFTRT
|| GET_CODE (x) == ROTATERT
- || (TARGET_ARM && GET_CODE (x) == ZERO_EXTRACT)))
+ || (TARGET_32BIT && GET_CODE (x) == ZERO_EXTRACT)))
return CC_NOOVmode;
if (GET_MODE (x) == QImode && (op == EQ || op == NE))
enum machine_mode mode = SELECT_CC_MODE (code, x, y);
rtx cc_reg = gen_rtx_REG (mode, CC_REGNUM);
- emit_insn (gen_rtx_SET (VOIDmode, cc_reg,
- gen_rtx_COMPARE (mode, x, y)));
+ emit_set_insn (cc_reg, gen_rtx_COMPARE (mode, x, y));
return cc_reg;
}
{
rtx base_plus = gen_rtx_REG (SImode, REGNO (operands[2]) + 1);
- emit_insn (gen_rtx_SET (VOIDmode, base_plus, base));
+ emit_set_insn (base_plus, base);
base = base_plus;
}
else if (GET_CODE (base) == PLUS)
plus_constant (base,
offset + 1))));
if (!BYTES_BIG_ENDIAN)
- emit_insn (gen_rtx_SET (VOIDmode, gen_rtx_SUBREG (SImode, operands[0], 0),
- gen_rtx_IOR (SImode,
- gen_rtx_ASHIFT
- (SImode,
- gen_rtx_SUBREG (SImode, operands[0], 0),
- GEN_INT (8)),
- scratch)));
+ emit_set_insn (gen_rtx_SUBREG (SImode, operands[0], 0),
+ gen_rtx_IOR (SImode,
+ gen_rtx_ASHIFT
+ (SImode,
+ gen_rtx_SUBREG (SImode, operands[0], 0),
+ GEN_INT (8)),
+ scratch));
else
- emit_insn (gen_rtx_SET (VOIDmode, gen_rtx_SUBREG (SImode, operands[0], 0),
- gen_rtx_IOR (SImode,
- gen_rtx_ASHIFT (SImode, scratch,
- GEN_INT (8)),
- gen_rtx_SUBREG (SImode, operands[0],
- 0))));
+ emit_set_insn (gen_rtx_SUBREG (SImode, operands[0], 0),
+ gen_rtx_IOR (SImode,
+ gen_rtx_ASHIFT (SImode, scratch,
+ GEN_INT (8)),
+ gen_rtx_SUBREG (SImode, operands[0], 0)));
}
/* Handle storing a half-word to memory during reload by synthesizing as two
}
}
- emit_insn (gen_rtx_SET (VOIDmode, base_plus, base));
+ emit_set_insn (base_plus, base);
base = base_plus;
}
else if (GET_CODE (base) == PLUS)
return !BYTES_BIG_ENDIAN;
}
-
\f
/* Print a symbolic form of X to the debug file, F. */
static void
static Mnode * minipool_vector_head;
static Mnode * minipool_vector_tail;
static rtx minipool_vector_label;
+static int minipool_pad;
/* The linked list of all minipool fixes required for this function. */
Mfix * minipool_fix_head;
{
/* ADDR_VECs only take room if read-only data does into the text
section. */
- if (JUMP_TABLES_IN_TEXT_SECTION
-#if !defined(READONLY_DATA_SECTION) && !defined(READONLY_DATA_SECTION_ASM_OP)
- || 1
-#endif
- )
+ if (JUMP_TABLES_IN_TEXT_SECTION || readonly_data_section == text_section)
{
rtx body = PATTERN (insn);
int elt = GET_CODE (body) == ADDR_DIFF_VEC ? 1 : 0;
+ HOST_WIDE_INT size;
+ HOST_WIDE_INT modesize;
- return GET_MODE_SIZE (GET_MODE (body)) * XVECLEN (body, elt);
+ modesize = GET_MODE_SIZE (GET_MODE (body));
+ size = modesize * XVECLEN (body, elt);
+ switch (modesize)
+ {
+ case 1:
+ /* Round up size of TBB table to a halfword boundary. */
+ size = (size + 1) & ~(HOST_WIDE_INT)1;
+ break;
+ case 2:
+ /* No padding necessary for TBH. */
+ break;
+ case 4:
+ /* Add two bytes for alignment on Thumb. */
+ if (TARGET_THUMB)
+ size += 2;
+ break;
+ default:
+ gcc_unreachable ();
+ }
+ return size;
}
return 0;
/* If set, max_mp is the first pool_entry that has a lower
constraint than the one we are trying to add. */
Mnode * max_mp = NULL;
- HOST_WIDE_INT max_address = fix->address + fix->forwards;
+ HOST_WIDE_INT max_address = fix->address + fix->forwards - minipool_pad;
Mnode * mp;
- /* If this fix's address is greater than the address of the first
- entry, then we can't put the fix in this pool. We subtract the
- size of the current fix to ensure that if the table is fully
- packed we still have enough room to insert this value by shuffling
- the other fixes forwards. */
+ /* If the minipool starts before the end of FIX->INSN then this FIX
+ can not be placed into the current pool. Furthermore, adding the
+ new constant pool entry may cause the pool to start FIX_SIZE bytes
+ earlier. */
if (minipool_vector_head &&
- fix->address >= minipool_vector_head->max_address - fix->fix_size)
+ (fix->address + get_attr_length (fix->insn)
+ >= minipool_vector_head->max_address - fix->fix_size))
return NULL;
/* Scan the pool to see if a constant with the same value has
any existing entry. Otherwise, we insert the new fix before
MAX_MP and, if necessary, adjust the constraints on the other
entries. */
- mp = xmalloc (sizeof (* mp));
+ mp = XNEW (Mnode);
mp->fix_size = fix->fix_size;
mp->mode = fix->mode;
mp->value = fix->value;
}
/* We need to create a new entry. */
- mp = xmalloc (sizeof (* mp));
+ mp = XNEW (Mnode);
mp->fix_size = fix->fix_size;
mp->mode = fix->mode;
mp->value = fix->value;
HOST_WIDE_INT count = 0;
rtx barrier;
rtx from = fix->insn;
- rtx selected = from;
+ /* The instruction after which we will insert the jump. */
+ rtx selected = NULL;
int selected_cost;
+ /* The address at which the jump instruction will be placed. */
HOST_WIDE_INT selected_address;
Mfix * new_fix;
HOST_WIDE_INT max_count = max_address - fix->address;
still put the pool after the table. */
new_cost = arm_barrier_cost (from);
- if (count < max_count && new_cost <= selected_cost)
+ if (count < max_count
+ && (!selected || new_cost <= selected_cost))
{
selected = tmp;
selected_cost = new_cost;
new_cost = arm_barrier_cost (from);
- if (count < max_count && new_cost <= selected_cost)
+ if (count < max_count
+ && (!selected || new_cost <= selected_cost))
{
selected = from;
selected_cost = new_cost;
from = NEXT_INSN (from);
}
+ /* Make sure that we found a place to insert the jump. */
+ gcc_assert (selected);
+
/* Create a new JUMP_INSN that branches around a barrier. */
from = emit_jump_insn_after (gen_jump (label), selected);
JUMP_LABEL (from) = label;
to generate duff assembly code. */
gcc_assert (fix->forwards || fix->backwards);
- /* With AAPCS/iWMMXt enabled, the pool is aligned to an 8-byte boundary.
- So there might be an empty word before the start of the pool.
- Hence we reduce the forward range by 4 to allow for this
- possibility. */
+ /* If an entry requires 8-byte alignment then assume all constant pools
+ require 4 bytes of padding. Trying to do this later on a per-pool
+ basis is awkward because existing pool entries have to be modified. */
if (ARM_DOUBLEWORD_ALIGN && fix->fix_size == 8)
- fix->forwards -= 4;
+ minipool_pad = 4;
if (dump_file)
{
{
rtx lowpart, highpart;
enum machine_mode mode;
-
+
mode = GET_MODE (val);
if (mode == VOIDmode)
mode = DImode;
gcc_assert (GET_MODE_SIZE (mode) == 8);
-
+
lowpart = gen_lowpart (SImode, val);
highpart = gen_highpart_mode (SImode, mode, val);
-
+
gcc_assert (GET_CODE (lowpart) == CONST_INT);
gcc_assert (GET_CODE (highpart) == CONST_INT);
if (mode == VOIDmode)
mode = DImode;
-
+
part = gen_highpart_mode (SImode, mode, val);
-
+
gcc_assert (GET_CODE (part) == CONST_INT);
-
+
if (const_ok_for_arm (INTVAL (part))
|| const_ok_for_arm (~INTVAL (part)))
return true;
-
+
part = gen_lowpart (SImode, val);
-
+
gcc_assert (GET_CODE (part) == CONST_INT);
-
+
if (const_ok_for_arm (INTVAL (part))
|| const_ok_for_arm (~INTVAL (part)))
return true;
-
+
return false;
}
scan it properly. */
insn = get_insns ();
gcc_assert (GET_CODE (insn) == NOTE);
+ minipool_pad = 0;
/* Scan all the insns and record the operands that will need fixing. */
for (insn = next_nonnote_insn (insn); insn; insn = next_nonnote_insn (insn))
/* Check that there isn't another fix that is in range that
we couldn't fit into this pool because the pool was
already too large: we need to put the pool before such an
- instruction. */
+ instruction. The pool itself may come just after the
+ fix because create_fix_barrier also allows space for a
+ jump instruction. */
if (ftmp->address < max_address)
- max_address = ftmp->address;
+ max_address = ftmp->address + 1;
last_barrier = create_fix_barrier (last_added_fix, max_address);
}
/* Output the operands of a LDM/STM instruction to STREAM.
MASK is the ARM register set mask of which only bits 0-15 are important.
REG is the base register, either the frame pointer or the stack pointer,
- INSTR is the possibly suffixed load or store instruction. */
+ INSTR is the possibly suffixed load or store instruction.
+ RFE is nonzero if the instruction should also copy spsr to cpsr. */
static void
print_multi_reg (FILE *stream, const char *instr, unsigned reg,
- unsigned long mask)
+ unsigned long mask, int rfe)
{
unsigned i;
bool not_first = FALSE;
+ gcc_assert (!rfe || (mask & (1 << PC_REGNUM)));
fputc ('\t', stream);
asm_fprintf (stream, instr, reg);
- fputs (", {", stream);
+ fputc ('{', stream);
for (i = 0; i <= LAST_ARM_REGNUM; i++)
if (mask & (1 << i))
not_first = TRUE;
}
- fprintf (stream, "}\n");
+ if (rfe)
+ fprintf (stream, "}^\n");
+ else
+ fprintf (stream, "}\n");
}
-/* Output a FLDMX instruction to STREAM.
+/* Output a FLDMD instruction to STREAM.
BASE if the register containing the address.
REG and COUNT specify the register range.
- Extra registers may be added to avoid hardware bugs. */
+ Extra registers may be added to avoid hardware bugs.
+
+ We output FLDMD even for ARMv5 VFP implementations. Although
+ FLDMD is technically not supported until ARMv6, it is believed
+ that all VFP implementations support its use in this context. */
static void
-arm_output_fldmx (FILE * stream, unsigned int base, int reg, int count)
+vfp_output_fldmd (FILE * stream, unsigned int base, int reg, int count)
{
int i;
count++;
}
+ /* FLDMD may not load more than 16 doubleword registers at a time. Split the
+ load into multiple parts if we have to handle more than 16 registers. */
+ if (count > 16)
+ {
+ vfp_output_fldmd (stream, base, reg, 16);
+ vfp_output_fldmd (stream, base, reg + 16, count - 16);
+ return;
+ }
+
fputc ('\t', stream);
- asm_fprintf (stream, "fldmfdx\t%r!, {", base);
+ asm_fprintf (stream, "fldmfdd\t%r!, {", base);
for (i = reg; i < reg + count; i++)
{
/* Output the assembly for a store multiple. */
const char *
-vfp_output_fstmx (rtx * operands)
+vfp_output_fstmd (rtx * operands)
{
char pattern[100];
int p;
int base;
int i;
- strcpy (pattern, "fstmfdx\t%m0!, {%P1");
+ strcpy (pattern, "fstmfdd\t%m0!, {%P1");
p = strlen (pattern);
gcc_assert (GET_CODE (operands[1]) == REG);
number of bytes pushed. */
static int
-vfp_emit_fstmx (int base_reg, int count)
+vfp_emit_fstmd (int base_reg, int count)
{
rtx par;
rtx dwarf;
count++;
}
- /* ??? The frame layout is implementation defined. We describe
- standard format 1 (equivalent to a FSTMD insn and unused pad word).
- We really need some way of representing the whole block so that the
- unwinder can figure it out at runtime. */
+ /* FSTMD may not store more than 16 doubleword registers at once. Split
+ larger stores into multiple parts (up to a maximum of two, in
+ practice). */
+ if (count > 16)
+ {
+ int saved;
+ /* NOTE: base_reg is an internal register number, so each D register
+ counts as 2. */
+ saved = vfp_emit_fstmd (base_reg + 32, count - 16);
+ saved += vfp_emit_fstmd (base_reg, 16);
+ return saved;
+ }
+
par = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (count));
dwarf = gen_rtx_SEQUENCE (VOIDmode, rtvec_alloc (count + 1));
XVECEXP (par, 0, 0)
= gen_rtx_SET (VOIDmode,
- gen_rtx_MEM (BLKmode,
- gen_rtx_PRE_DEC (BLKmode, stack_pointer_rtx)),
+ gen_frame_mem (BLKmode,
+ gen_rtx_PRE_DEC (BLKmode,
+ stack_pointer_rtx)),
gen_rtx_UNSPEC (BLKmode,
gen_rtvec (1, reg),
UNSPEC_PUSH_MULT));
tmp = gen_rtx_SET (VOIDmode, stack_pointer_rtx,
- gen_rtx_PLUS (SImode, stack_pointer_rtx,
- GEN_INT (-(count * 8 + 4))));
+ plus_constant (stack_pointer_rtx, -(count * 8)));
RTX_FRAME_RELATED_P (tmp) = 1;
XVECEXP (dwarf, 0, 0) = tmp;
tmp = gen_rtx_SET (VOIDmode,
- gen_rtx_MEM (DFmode, stack_pointer_rtx),
+ gen_frame_mem (DFmode, stack_pointer_rtx),
reg);
RTX_FRAME_RELATED_P (tmp) = 1;
XVECEXP (dwarf, 0, 1) = tmp;
XVECEXP (par, 0, i) = gen_rtx_USE (VOIDmode, reg);
tmp = gen_rtx_SET (VOIDmode,
- gen_rtx_MEM (DFmode,
- gen_rtx_PLUS (SImode,
- stack_pointer_rtx,
- GEN_INT (i * 8))),
+ gen_frame_mem (DFmode,
+ plus_constant (stack_pointer_rtx,
+ i * 8)),
reg);
RTX_FRAME_RELATED_P (tmp) = 1;
XVECEXP (dwarf, 0, i + 1) = tmp;
REG_NOTES (par));
RTX_FRAME_RELATED_P (par) = 1;
- return count * 8 + 4;
+ return count * 8;
}
+/* Emit a call instruction with pattern PAT. ADDR is the address of
+ the call target. */
+
+void
+arm_emit_call_insn (rtx pat, rtx addr)
+{
+ rtx insn;
+
+ insn = emit_call_insn (pat);
+
+ /* The PIC register is live on entry to VxWorks PIC PLT entries.
+ If the call might use such an entry, add a use of the PIC register
+ to the instruction's CALL_INSN_FUNCTION_USAGE. */
+ if (TARGET_VXWORKS_RTP
+ && flag_pic
+ && GET_CODE (addr) == SYMBOL_REF
+ && (SYMBOL_REF_DECL (addr)
+ ? !targetm.binds_local_p (SYMBOL_REF_DECL (addr))
+ : !SYMBOL_REF_LOCAL_P (addr)))
+ {
+ require_pic_register ();
+ use_reg (&CALL_INSN_FUNCTION_USAGE (insn), cfun->machine->pic_reg);
+ }
+}
/* Output a 'call' insn. */
const char *
ops[1] = gen_rtx_REG (SImode, 1 + arm_reg0);
ops[2] = gen_rtx_REG (SImode, 2 + arm_reg0);
- output_asm_insn ("stm%?fd\t%|sp!, {%0, %1, %2}", ops);
+ output_asm_insn ("stm%(fd%)\t%|sp!, {%0, %1, %2}", ops);
output_asm_insn ("ldf%?e\t%0, [%|sp], #12", operands);
return "";
ops[2] = gen_rtx_REG (SImode, 2 + arm_reg0);
output_asm_insn ("stf%?e\t%1, [%|sp, #-12]!", operands);
- output_asm_insn ("ldm%?fd\t%|sp!, {%0, %1, %2}", ops);
+ output_asm_insn ("ldm%(fd%)\t%|sp!, {%0, %1, %2}", ops);
return "";
}
ops[0] = gen_rtx_REG (SImode, arm_reg0);
ops[1] = gen_rtx_REG (SImode, 1 + arm_reg0);
- output_asm_insn ("stm%?fd\t%|sp!, {%0, %1}", ops);
+ output_asm_insn ("stm%(fd%)\t%|sp!, {%0, %1}", ops);
output_asm_insn ("ldf%?d\t%0, [%|sp], #8", operands);
return "";
}
ops[0] = gen_rtx_REG (SImode, arm_reg0);
ops[1] = gen_rtx_REG (SImode, 1 + arm_reg0);
output_asm_insn ("stf%?d\t%1, [%|sp, #-8]!", operands);
- output_asm_insn ("ldm%?fd\t%|sp!, {%0, %1}", ops);
+ output_asm_insn ("ldm%(fd%)\t%|sp!, {%0, %1}", ops);
return "";
}
switch (GET_CODE (XEXP (operands[1], 0)))
{
case REG:
- output_asm_insn ("ldm%?ia\t%m1, %M0", operands);
+ output_asm_insn ("ldm%(ia%)\t%m1, %M0", operands);
break;
-
+
case PRE_INC:
gcc_assert (TARGET_LDRD);
- output_asm_insn ("ldr%?d\t%0, [%m1, #8]!", operands);
+ output_asm_insn ("ldr%(d%)\t%0, [%m1, #8]!", operands);
break;
-
+
case PRE_DEC:
- output_asm_insn ("ldm%?db\t%m1!, %M0", operands);
+ if (TARGET_LDRD)
+ output_asm_insn ("ldr%(d%)\t%0, [%m1, #-8]!", operands);
+ else
+ output_asm_insn ("ldm%(db%)\t%m1!, %M0", operands);
break;
-
+
case POST_INC:
- output_asm_insn ("ldm%?ia\t%m1!, %M0", operands);
+ output_asm_insn ("ldm%(ia%)\t%m1!, %M0", operands);
break;
-
+
case POST_DEC:
gcc_assert (TARGET_LDRD);
- output_asm_insn ("ldr%?d\t%0, [%m1], #-8", operands);
+ output_asm_insn ("ldr%(d%)\t%0, [%m1], #-8", operands);
break;
-
+
case PRE_MODIFY:
case POST_MODIFY:
otherops[0] = operands[0];
otherops[1] = XEXP (XEXP (XEXP (operands[1], 0), 1), 0);
otherops[2] = XEXP (XEXP (XEXP (operands[1], 0), 1), 1);
-
+
if (GET_CODE (XEXP (operands[1], 0)) == PRE_MODIFY)
{
if (reg_overlap_mentioned_p (otherops[0], otherops[2]))
{
/* Registers overlap so split out the increment. */
output_asm_insn ("add%?\t%1, %1, %2", otherops);
- output_asm_insn ("ldr%?d\t%0, [%1] @split", otherops);
+ output_asm_insn ("ldr%(d%)\t%0, [%1] @split", otherops);
}
else
- output_asm_insn ("ldr%?d\t%0, [%1, %2]!", otherops);
+ {
+ /* IWMMXT allows offsets larger than ldrd can handle,
+ fix these up with a pair of ldr. */
+ if (GET_CODE (otherops[2]) == CONST_INT
+ && (INTVAL(otherops[2]) <= -256
+ || INTVAL(otherops[2]) >= 256))
+ {
+ output_asm_insn ("ldr%?\t%0, [%1, %2]!", otherops);
+ otherops[0] = gen_rtx_REG (SImode, 1 + reg0);
+ output_asm_insn ("ldr%?\t%0, [%1, #4]", otherops);
+ }
+ else
+ output_asm_insn ("ldr%(d%)\t%0, [%1, %2]!", otherops);
+ }
}
else
{
- /* We only allow constant increments, so this is safe. */
- output_asm_insn ("ldr%?d\t%0, [%1], %2", otherops);
+ /* IWMMXT allows offsets larger than ldrd can handle,
+ fix these up with a pair of ldr. */
+ if (GET_CODE (otherops[2]) == CONST_INT
+ && (INTVAL(otherops[2]) <= -256
+ || INTVAL(otherops[2]) >= 256))
+ {
+ otherops[0] = gen_rtx_REG (SImode, 1 + reg0);
+ output_asm_insn ("ldr%?\t%0, [%1, #4]", otherops);
+ otherops[0] = operands[0];
+ output_asm_insn ("ldr%?\t%0, [%1], %2", otherops);
+ }
+ else
+ /* We only allow constant increments, so this is safe. */
+ output_asm_insn ("ldr%(d%)\t%0, [%1], %2", otherops);
}
break;
-
+
case LABEL_REF:
case CONST:
output_asm_insn ("adr%?\t%0, %1", operands);
- output_asm_insn ("ldm%?ia\t%0, %M0", operands);
+ output_asm_insn ("ldm%(ia%)\t%0, %M0", operands);
break;
-
+
+ /* ??? This needs checking for thumb2. */
default:
if (arm_add_operand (XEXP (XEXP (operands[1], 0), 1),
GET_MODE (XEXP (XEXP (operands[1], 0), 1))))
otherops[0] = operands[0];
otherops[1] = XEXP (XEXP (operands[1], 0), 0);
otherops[2] = XEXP (XEXP (operands[1], 0), 1);
-
+
if (GET_CODE (XEXP (operands[1], 0)) == PLUS)
{
if (GET_CODE (otherops[2]) == CONST_INT)
switch ((int) INTVAL (otherops[2]))
{
case -8:
- output_asm_insn ("ldm%?db\t%1, %M0", otherops);
+ output_asm_insn ("ldm%(db%)\t%1, %M0", otherops);
return "";
case -4:
- output_asm_insn ("ldm%?da\t%1, %M0", otherops);
+ if (TARGET_THUMB2)
+ break;
+ output_asm_insn ("ldm%(da%)\t%1, %M0", otherops);
return "";
case 4:
- output_asm_insn ("ldm%?ib\t%1, %M0", otherops);
+ if (TARGET_THUMB2)
+ break;
+ output_asm_insn ("ldm%(ib%)\t%1, %M0", otherops);
return "";
}
}
avoid a conflict. */
otherops[1] = XEXP (XEXP (operands[1], 0), 1);
otherops[2] = XEXP (XEXP (operands[1], 0), 0);
-
}
/* If both registers conflict, it will usually
have been fixed by a splitter. */
if (reg_overlap_mentioned_p (otherops[0], otherops[2]))
{
output_asm_insn ("add%?\t%1, %1, %2", otherops);
- output_asm_insn ("ldr%?d\t%0, [%1]",
+ output_asm_insn ("ldr%(d%)\t%0, [%1]",
otherops);
}
else
- output_asm_insn ("ldr%?d\t%0, [%1, %2]", otherops);
+ output_asm_insn ("ldr%(d%)\t%0, [%1, %2]", otherops);
return "";
}
-
+
if (GET_CODE (otherops[2]) == CONST_INT)
{
if (!(const_ok_for_arm (INTVAL (otherops[2]))))
else
output_asm_insn ("sub%?\t%0, %1, %2", otherops);
- return "ldm%?ia\t%0, %M0";
+ return "ldm%(ia%)\t%0, %M0";
}
else
{
switch (GET_CODE (XEXP (operands[0], 0)))
{
case REG:
- output_asm_insn ("stm%?ia\t%m0, %M1", operands);
+ output_asm_insn ("stm%(ia%)\t%m0, %M1", operands);
break;
case PRE_INC:
gcc_assert (TARGET_LDRD);
- output_asm_insn ("str%?d\t%1, [%m0, #8]!", operands);
+ output_asm_insn ("str%(d%)\t%1, [%m0, #8]!", operands);
break;
case PRE_DEC:
- output_asm_insn ("stm%?db\t%m0!, %M1", operands);
+ if (TARGET_LDRD)
+ output_asm_insn ("str%(d%)\t%1, [%m0, #-8]!", operands);
+ else
+ output_asm_insn ("stm%(db%)\t%m0!, %M1", operands);
break;
case POST_INC:
- output_asm_insn ("stm%?ia\t%m0!, %M1", operands);
+ output_asm_insn ("stm%(ia%)\t%m0!, %M1", operands);
break;
case POST_DEC:
gcc_assert (TARGET_LDRD);
- output_asm_insn ("str%?d\t%1, [%m0], #-8", operands);
+ output_asm_insn ("str%(d%)\t%1, [%m0], #-8", operands);
break;
case PRE_MODIFY:
otherops[1] = XEXP (XEXP (XEXP (operands[0], 0), 1), 0);
otherops[2] = XEXP (XEXP (XEXP (operands[0], 0), 1), 1);
- if (GET_CODE (XEXP (operands[0], 0)) == PRE_MODIFY)
- output_asm_insn ("str%?d\t%0, [%1, %2]!", otherops);
+ /* IWMMXT allows offsets larger than ldrd can handle,
+ fix these up with a pair of ldr. */
+ if (GET_CODE (otherops[2]) == CONST_INT
+ && (INTVAL(otherops[2]) <= -256
+ || INTVAL(otherops[2]) >= 256))
+ {
+ rtx reg1;
+ reg1 = gen_rtx_REG (SImode, 1 + REGNO (operands[1]));
+ if (GET_CODE (XEXP (operands[0], 0)) == PRE_MODIFY)
+ {
+ output_asm_insn ("ldr%?\t%0, [%1, %2]!", otherops);
+ otherops[0] = reg1;
+ output_asm_insn ("ldr%?\t%0, [%1, #4]", otherops);
+ }
+ else
+ {
+ otherops[0] = reg1;
+ output_asm_insn ("ldr%?\t%0, [%1, #4]", otherops);
+ otherops[0] = operands[1];
+ output_asm_insn ("ldr%?\t%0, [%1], %2", otherops);
+ }
+ }
+ else if (GET_CODE (XEXP (operands[0], 0)) == PRE_MODIFY)
+ output_asm_insn ("str%(d%)\t%0, [%1, %2]!", otherops);
else
- output_asm_insn ("str%?d\t%0, [%1], %2", otherops);
+ output_asm_insn ("str%(d%)\t%0, [%1], %2", otherops);
break;
case PLUS:
switch ((int) INTVAL (XEXP (XEXP (operands[0], 0), 1)))
{
case -8:
- output_asm_insn ("stm%?db\t%m0, %M1", operands);
+ output_asm_insn ("stm%(db%)\t%m0, %M1", operands);
return "";
case -4:
- output_asm_insn ("stm%?da\t%m0, %M1", operands);
+ if (TARGET_THUMB2)
+ break;
+ output_asm_insn ("stm%(da%)\t%m0, %M1", operands);
return "";
case 4:
- output_asm_insn ("stm%?ib\t%m0, %M1", operands);
+ if (TARGET_THUMB2)
+ break;
+ output_asm_insn ("stm%(ib%)\t%m0, %M1", operands);
return "";
}
}
{
otherops[0] = operands[1];
otherops[1] = XEXP (XEXP (operands[0], 0), 0);
- output_asm_insn ("str%?d\t%0, [%1, %2]", otherops);
+ output_asm_insn ("str%(d%)\t%0, [%1, %2]", otherops);
return "";
}
/* Fall through */
return "";
}
+/* Output a VFP load or store instruction. */
+
+const char *
+output_move_vfp (rtx *operands)
+{
+ rtx reg, mem, addr, ops[2];
+ int load = REG_P (operands[0]);
+ int dp = GET_MODE_SIZE (GET_MODE (operands[0])) == 8;
+ int integer_p = GET_MODE_CLASS (GET_MODE (operands[0])) == MODE_INT;
+ const char *template;
+ char buff[50];
+
+ reg = operands[!load];
+ mem = operands[load];
+
+ gcc_assert (REG_P (reg));
+ gcc_assert (IS_VFP_REGNUM (REGNO (reg)));
+ gcc_assert (GET_MODE (reg) == SFmode
+ || GET_MODE (reg) == DFmode
+ || GET_MODE (reg) == SImode
+ || GET_MODE (reg) == DImode);
+ gcc_assert (MEM_P (mem));
+
+ addr = XEXP (mem, 0);
+
+ switch (GET_CODE (addr))
+ {
+ case PRE_DEC:
+ template = "f%smdb%c%%?\t%%0!, {%%%s1}%s";
+ ops[0] = XEXP (addr, 0);
+ ops[1] = reg;
+ break;
+
+ case POST_INC:
+ template = "f%smia%c%%?\t%%0!, {%%%s1}%s";
+ ops[0] = XEXP (addr, 0);
+ ops[1] = reg;
+ break;
+
+ default:
+ template = "f%s%c%%?\t%%%s0, %%1%s";
+ ops[0] = reg;
+ ops[1] = mem;
+ break;
+ }
+
+ sprintf (buff, template,
+ load ? "ld" : "st",
+ dp ? 'd' : 's',
+ dp ? "P" : "",
+ integer_p ? "\t%@ int" : "");
+ output_asm_insn (buff, ops);
+
+ return "";
+}
+
/* Output an ADD r, s, #n where n may be too big for one instruction.
If adding zero to one register, output nothing. */
const char *
return "";
}
+/* Return the name of a shifter operation. */
+static const char *
+arm_shift_nmem(enum rtx_code code)
+{
+ switch (code)
+ {
+ case ASHIFT:
+ return ARM_LSL_NAME;
+
+ case ASHIFTRT:
+ return "asr";
+
+ case LSHIFTRT:
+ return "lsr";
+
+ case ROTATERT:
+ return "ror";
+
+ default:
+ abort();
+ }
+}
+
/* Return the appropriate ARM instruction for the operation code.
The returned result should not be overwritten. OP is the rtx of the
operation. SHIFT_FIRST_ARG is TRUE if the first argument of the operator
case AND:
return "and";
+ case ASHIFT:
+ case ASHIFTRT:
+ case LSHIFTRT:
+ case ROTATERT:
+ return arm_shift_nmem(GET_CODE(op));
+
default:
gcc_unreachable ();
}
switch (code)
{
- case ASHIFT:
- mnem = "asl";
- break;
-
- case ASHIFTRT:
- mnem = "asr";
- break;
-
- case LSHIFTRT:
- mnem = "lsr";
- break;
-
case ROTATE:
gcc_assert (*amountp != -1);
*amountp = 32 - *amountp;
+ code = ROTATERT;
/* Fall through. */
+ case ASHIFT:
+ case ASHIFTRT:
+ case LSHIFTRT:
case ROTATERT:
- mnem = "ror";
+ mnem = arm_shift_nmem(code);
break;
case MULT:
power of 2, since this case can never be reloaded from a reg. */
gcc_assert (*amountp != -1);
*amountp = int_log2 (*amountp);
- return "asl";
+ return ARM_LSL_NAME;
default:
gcc_unreachable ();
{
/* This is not 100% correct, but follows from the desire to merge
multiplication by a power of 2 with the recognizer for a
- shift. >=32 is not a valid shift for "asl", so we must try and
+ shift. >=32 is not a valid shift for "lsl", so we must try and
output a shift that produces the correct arithmetical result.
Using lsr #32 is identical except for the fact that the carry bit
is not set correctly if we set the flags; but we never use the
max_reg = 12;
for (reg = 0; reg <= max_reg; reg++)
- if (regs_ever_live[reg]
- || (! current_function_is_leaf && call_used_regs [reg]))
+ if (df_regs_ever_live_p (reg)
+ || (! current_function_is_leaf && call_used_regs[reg]))
save_reg_mask |= (1 << reg);
/* Also save the pic base register if necessary. */
if (flag_pic
&& !TARGET_SINGLE_PIC_BASE
+ && arm_pic_register != INVALID_REGNUM
&& current_function_uses_pic_offset_table)
save_reg_mask |= 1 << PIC_OFFSET_TABLE_REGNUM;
}
else
{
+ /* In arm mode we handle r11 (FP) as a special case. */
+ unsigned last_reg = TARGET_ARM ? 10 : 11;
+
/* In the normal case we only need to save those registers
which are call saved and which are used by this function. */
- for (reg = 0; reg <= 10; reg++)
- if (regs_ever_live[reg] && ! call_used_regs [reg])
+ for (reg = 0; reg <= last_reg; reg++)
+ if (df_regs_ever_live_p (reg) && ! call_used_regs[reg])
save_reg_mask |= (1 << reg);
/* Handle the frame pointer as a special case. */
if (! TARGET_APCS_FRAME
&& ! frame_pointer_needed
- && regs_ever_live[HARD_FRAME_POINTER_REGNUM]
+ && df_regs_ever_live_p (HARD_FRAME_POINTER_REGNUM)
&& ! call_used_regs[HARD_FRAME_POINTER_REGNUM])
save_reg_mask |= 1 << HARD_FRAME_POINTER_REGNUM;
+ else if (! TARGET_APCS_FRAME
+ && ! frame_pointer_needed
+ && df_regs_ever_live_p (HARD_FRAME_POINTER_REGNUM)
+ && ! call_used_regs[HARD_FRAME_POINTER_REGNUM])
+ save_reg_mask |= 1 << HARD_FRAME_POINTER_REGNUM;
/* If we aren't loading the PIC register,
don't stack it even though it may be live. */
if (flag_pic
- && !TARGET_SINGLE_PIC_BASE
- && (regs_ever_live[PIC_OFFSET_TABLE_REGNUM]
+ && !TARGET_SINGLE_PIC_BASE
+ && arm_pic_register != INVALID_REGNUM
+ && (df_regs_ever_live_p (PIC_OFFSET_TABLE_REGNUM)
|| current_function_uses_pic_offset_table))
save_reg_mask |= 1 << PIC_OFFSET_TABLE_REGNUM;
+
+ /* The prologue will copy SP into R0, so save it. */
+ if (IS_STACKALIGN (func_type))
+ save_reg_mask |= 1;
}
/* Save registers so the exception handler can modify them. */
return save_reg_mask;
}
+
/* Compute a bit mask of which registers need to be
saved on the stack for the current function. */
{
unsigned int save_reg_mask = 0;
unsigned long func_type = arm_current_func_type ();
+ unsigned int reg;
if (IS_NAKED (func_type))
/* This should never really happen. */
/* If we are creating a stack frame, then we must save the frame pointer,
IP (which will hold the old stack pointer), LR and the PC. */
- if (frame_pointer_needed)
+ if (frame_pointer_needed && TARGET_ARM)
save_reg_mask |=
(1 << ARM_HARD_FRAME_POINTER_REGNUM)
| (1 << IP_REGNUM)
now and then popping it back into the PC. This incurs extra memory
accesses though, so we only do it when optimizing for size, and only
if we know that we will not need a fancy return sequence. */
- if (regs_ever_live [LR_REGNUM]
- || (save_reg_mask
- && optimize_size
- && ARM_FUNC_TYPE (func_type) == ARM_FT_NORMAL
- && !current_function_calls_eh_return))
+ if (df_regs_ever_live_p (LR_REGNUM)
+ || (save_reg_mask
+ && optimize_size
+ && ARM_FUNC_TYPE (func_type) == ARM_FT_NORMAL
+ && !current_function_calls_eh_return))
save_reg_mask |= 1 << LR_REGNUM;
if (cfun->machine->lr_save_eliminated)
&& ((bit_count (save_reg_mask)
+ ARM_NUM_INTS (current_function_pretend_args_size)) % 2) != 0)
{
- unsigned int reg;
-
/* The total number of registers that are going to be pushed
onto the stack is odd. We need to ensure that the stack
is 64-bit aligned before we start to save iWMMXt registers,
}
}
+ /* We may need to push an additional register for use initializing the
+ PIC base register. */
+ if (TARGET_THUMB2 && IS_NESTED (func_type) && flag_pic
+ && (save_reg_mask & THUMB2_WORK_REGS) == 0)
+ {
+ reg = thumb_find_work_register (1 << 4);
+ if (!call_used_regs[reg])
+ save_reg_mask |= (1 << reg);
+ }
+
return save_reg_mask;
}
/* Compute a bit mask of which registers need to be
saved on the stack for the current function. */
static unsigned long
-thumb_compute_save_reg_mask (void)
+thumb1_compute_save_reg_mask (void)
{
unsigned long mask;
unsigned reg;
mask = 0;
for (reg = 0; reg < 12; reg ++)
- if (regs_ever_live[reg] && !call_used_regs[reg])
+ if (df_regs_ever_live_p (reg) && !call_used_regs[reg])
mask |= 1 << reg;
if (flag_pic
&& !TARGET_SINGLE_PIC_BASE
+ && arm_pic_register != INVALID_REGNUM
&& current_function_uses_pic_offset_table)
mask |= 1 << PIC_OFFSET_TABLE_REGNUM;
regno < LAST_VFP_REGNUM;
regno += 2)
{
- if ((!regs_ever_live[regno] || call_used_regs[regno])
- && (!regs_ever_live[regno + 1] || call_used_regs[regno + 1]))
+ if ((!df_regs_ever_live_p (regno) || call_used_regs[regno])
+ && (!df_regs_ever_live_p (regno + 1) || call_used_regs[regno + 1]))
{
if (count > 0)
{
/* Workaround ARM10 VFPr1 bug. */
if (count == 2 && !arm_arch6)
count++;
- saved += count * 8 + 4;
+ saved += count * 8;
}
count = 0;
}
{
if (count == 2 && !arm_arch6)
count++;
- saved += count * 8 + 4;
+ saved += count * 8;
}
}
return saved;
const char * return_reg;
/* If we do not have any special requirements for function exit
- (e.g. interworking, or ISR) then we can load the return address
+ (e.g. interworking) then we can load the return address
directly into the PC. Otherwise we must load it into LR. */
if (really_return
- && ! TARGET_INTERWORK)
+ && (IS_INTERRUPT (func_type) || !TARGET_INTERWORK))
return_reg = reg_names[PC_REGNUM];
else
return_reg = reg_names[LR_REGNUM];
stack_adjust = offsets->outgoing_args - offsets->saved_regs;
gcc_assert (stack_adjust == 0 || stack_adjust == 4);
- if (stack_adjust && arm_arch5)
+ if (stack_adjust && arm_arch5 && TARGET_ARM)
sprintf (instr, "ldm%sib\t%%|sp, {", conditional);
else
{
{
case ARM_FT_ISR:
case ARM_FT_FIQ:
+ /* ??? This is wrong for unified assembly syntax. */
sprintf (instr, "sub%ss\t%%|pc, %%|lr, #4", conditional);
break;
break;
case ARM_FT_EXCEPTION:
+ /* ??? This is wrong for unified assembly syntax. */
sprintf (instr, "mov%ss\t%%|pc, %%|lr", conditional);
break;
{
unsigned long func_type;
- if (!TARGET_ARM)
+ if (TARGET_THUMB1)
{
- thumb_output_function_prologue (f, frame_size);
+ thumb1_output_function_prologue (f, frame_size);
return;
}
if (IS_NESTED (func_type))
asm_fprintf (f, "\t%@ Nested: function declared inside another function.\n");
+ if (IS_STACKALIGN (func_type))
+ asm_fprintf (f, "\t%@ Stack Align: May be called with mis-aligned SP.\n");
asm_fprintf (f, "\t%@ args = %d, pretend = %d, frame = %wd\n",
current_function_args_size,
if (saved_regs_mask & (1 << reg))
floats_offset += 4;
- if (frame_pointer_needed)
+ if (frame_pointer_needed && TARGET_ARM)
{
/* This variable is for the Virtual Frame Pointer, not VFP regs. */
int vfp_offset = offsets->frame;
if (arm_fpu_arch == FPUTYPE_FPA_EMU2)
{
for (reg = LAST_FPA_REGNUM; reg >= FIRST_FPA_REGNUM; reg--)
- if (regs_ever_live[reg] && !call_used_regs[reg])
+ if (df_regs_ever_live_p (reg) && !call_used_regs[reg])
{
floats_offset += 12;
asm_fprintf (f, "\tldfe\t%r, [%r, #-%d]\n",
for (reg = LAST_FPA_REGNUM; reg >= FIRST_FPA_REGNUM; reg--)
{
- if (regs_ever_live[reg] && !call_used_regs[reg])
+ if (df_regs_ever_live_p (reg) && !call_used_regs[reg])
{
floats_offset += 12;
{
int saved_size;
- /* The fldmx insn does not have base+offset addressing modes,
- so we use IP to hold the address. */
+ /* The fldmd insns do not have base+offset addressing
+ modes, so we use IP to hold the address. */
saved_size = arm_get_vfp_saved_size ();
if (saved_size > 0)
start_reg = FIRST_VFP_REGNUM;
for (reg = FIRST_VFP_REGNUM; reg < LAST_VFP_REGNUM; reg += 2)
{
- if ((!regs_ever_live[reg] || call_used_regs[reg])
- && (!regs_ever_live[reg + 1] || call_used_regs[reg + 1]))
+ if ((!df_regs_ever_live_p (reg) || call_used_regs[reg])
+ && (!df_regs_ever_live_p (reg + 1) || call_used_regs[reg + 1]))
{
if (start_reg != reg)
- arm_output_fldmx (f, IP_REGNUM,
+ vfp_output_fldmd (f, IP_REGNUM,
(start_reg - FIRST_VFP_REGNUM) / 2,
(reg - start_reg) / 2);
start_reg = reg + 2;
}
}
if (start_reg != reg)
- arm_output_fldmx (f, IP_REGNUM,
+ vfp_output_fldmd (f, IP_REGNUM,
(start_reg - FIRST_VFP_REGNUM) / 2,
(reg - start_reg) / 2);
}
lrm_count += (lrm_count % 2 ? 2 : 1);
for (reg = LAST_IWMMXT_REGNUM; reg >= FIRST_IWMMXT_REGNUM; reg--)
- if (regs_ever_live[reg] && !call_used_regs[reg])
+ if (df_regs_ever_live_p (reg) && !call_used_regs[reg])
{
asm_fprintf (f, "\twldrd\t%r, [%r, #-%d]\n",
reg, FP_REGNUM, lrm_count * 4);
|| current_function_calls_alloca)
asm_fprintf (f, "\tsub\t%r, %r, #%d\n", SP_REGNUM, FP_REGNUM,
4 * bit_count (saved_regs_mask));
- print_multi_reg (f, "ldmfd\t%r", SP_REGNUM, saved_regs_mask);
+ print_multi_reg (f, "ldmfd\t%r, ", SP_REGNUM, saved_regs_mask, 0);
if (IS_INTERRUPT (func_type))
/* Interrupt handlers will have pushed the
IP onto the stack, so restore it now. */
- print_multi_reg (f, "ldmfd\t%r!", SP_REGNUM, 1 << IP_REGNUM);
+ print_multi_reg (f, "ldmfd\t%r!, ", SP_REGNUM, 1 << IP_REGNUM, 0);
}
else
{
+ HOST_WIDE_INT amount;
+ int rfe;
/* Restore stack pointer if necessary. */
- if (offsets->outgoing_args != offsets->saved_regs)
+ if (frame_pointer_needed)
{
- operands[0] = operands[1] = stack_pointer_rtx;
- operands[2] = GEN_INT (offsets->outgoing_args - offsets->saved_regs);
+ /* For Thumb-2 restore sp from the frame pointer.
+ Operand restrictions mean we have to increment FP, then copy
+ to SP. */
+ amount = offsets->locals_base - offsets->saved_regs;
+ operands[0] = hard_frame_pointer_rtx;
+ }
+ else
+ {
+ operands[0] = stack_pointer_rtx;
+ amount = offsets->outgoing_args - offsets->saved_regs;
+ }
+
+ if (amount)
+ {
+ operands[1] = operands[0];
+ operands[2] = GEN_INT (amount);
output_add_immediate (operands);
}
+ if (frame_pointer_needed)
+ asm_fprintf (f, "\tmov\t%r, %r\n",
+ SP_REGNUM, HARD_FRAME_POINTER_REGNUM);
if (arm_fpu_arch == FPUTYPE_FPA_EMU2)
{
for (reg = FIRST_FPA_REGNUM; reg <= LAST_FPA_REGNUM; reg++)
- if (regs_ever_live[reg] && !call_used_regs[reg])
+ if (df_regs_ever_live_p (reg) && !call_used_regs[reg])
asm_fprintf (f, "\tldfe\t%r, [%r], #12\n",
reg, SP_REGNUM);
}
for (reg = FIRST_FPA_REGNUM; reg <= LAST_FPA_REGNUM; reg++)
{
- if (regs_ever_live[reg] && !call_used_regs[reg])
+ if (df_regs_ever_live_p (reg) && !call_used_regs[reg])
{
if (reg - start_reg == 3)
{
start_reg = FIRST_VFP_REGNUM;
for (reg = FIRST_VFP_REGNUM; reg < LAST_VFP_REGNUM; reg += 2)
{
- if ((!regs_ever_live[reg] || call_used_regs[reg])
- && (!regs_ever_live[reg + 1] || call_used_regs[reg + 1]))
+ if ((!df_regs_ever_live_p (reg) || call_used_regs[reg])
+ && (!df_regs_ever_live_p (reg + 1) || call_used_regs[reg + 1]))
{
if (start_reg != reg)
- arm_output_fldmx (f, SP_REGNUM,
+ vfp_output_fldmd (f, SP_REGNUM,
(start_reg - FIRST_VFP_REGNUM) / 2,
(reg - start_reg) / 2);
start_reg = reg + 2;
}
}
if (start_reg != reg)
- arm_output_fldmx (f, SP_REGNUM,
+ vfp_output_fldmd (f, SP_REGNUM,
(start_reg - FIRST_VFP_REGNUM) / 2,
(reg - start_reg) / 2);
}
if (TARGET_IWMMXT)
for (reg = FIRST_IWMMXT_REGNUM; reg <= LAST_IWMMXT_REGNUM; reg++)
- if (regs_ever_live[reg] && !call_used_regs[reg])
+ if (df_regs_ever_live_p (reg) && !call_used_regs[reg])
asm_fprintf (f, "\twldrd\t%r, [%r], #8\n", reg, SP_REGNUM);
/* If we can, restore the LR into the PC. */
- if (ARM_FUNC_TYPE (func_type) == ARM_FT_NORMAL
+ if (ARM_FUNC_TYPE (func_type) != ARM_FT_INTERWORKED
+ && (TARGET_ARM || ARM_FUNC_TYPE (func_type) == ARM_FT_NORMAL)
+ && !IS_STACKALIGN (func_type)
&& really_return
&& current_function_pretend_args_size == 0
&& saved_regs_mask & (1 << LR_REGNUM)
{
saved_regs_mask &= ~ (1 << LR_REGNUM);
saved_regs_mask |= (1 << PC_REGNUM);
+ rfe = IS_INTERRUPT (func_type);
}
+ else
+ rfe = 0;
/* Load the registers off the stack. If we only have one register
- to load use the LDR instruction - it is faster. */
- if (saved_regs_mask == (1 << LR_REGNUM))
+ to load use the LDR instruction - it is faster. For Thumb-2
+ always use pop and the assembler will pick the best instruction.*/
+ if (TARGET_ARM && saved_regs_mask == (1 << LR_REGNUM)
+ && !IS_INTERRUPT(func_type))
{
asm_fprintf (f, "\tldr\t%r, [%r], #4\n", LR_REGNUM, SP_REGNUM);
}
(i.e. "ldmfd sp!..."). We know that the stack pointer is
in the list of registers and if we add writeback the
instruction becomes UNPREDICTABLE. */
- print_multi_reg (f, "ldmfd\t%r", SP_REGNUM, saved_regs_mask);
+ print_multi_reg (f, "ldmfd\t%r, ", SP_REGNUM, saved_regs_mask,
+ rfe);
+ else if (TARGET_ARM)
+ print_multi_reg (f, "ldmfd\t%r!, ", SP_REGNUM, saved_regs_mask,
+ rfe);
else
- print_multi_reg (f, "ldmfd\t%r!", SP_REGNUM, saved_regs_mask);
+ print_multi_reg (f, "pop\t", SP_REGNUM, saved_regs_mask, 0);
}
if (current_function_pretend_args_size)
break;
default:
+ if (IS_STACKALIGN (func_type))
+ {
+ /* See comment in arm_expand_prologue. */
+ asm_fprintf (f, "\tmov\t%r, %r\n", SP_REGNUM, 0);
+ }
if (arm_arch5 || arm_arch4t)
asm_fprintf (f, "\tbx\t%r\n", LR_REGNUM);
else
{
arm_stack_offsets *offsets;
- if (TARGET_THUMB)
+ if (TARGET_THUMB1)
{
int regno;
if (label != NULL)
{
- function_section (current_function_decl);
+ switch_to_section (function_section (current_function_decl));
targetm.asm_out.internal_label (asm_out_file, "L",
CODE_LABEL_NUMBER (label));
asm_fprintf (asm_out_file, "\tbx\t%r\n", regno);
RTL for it. This does not happen for inline functions. */
return_used_this_function = 0;
}
- else
+ else /* TARGET_32BIT */
{
/* We need to take into account any stack-frame rounding. */
offsets = arm_get_frame_offsets ();
XVECEXP (par, 0, 0)
= gen_rtx_SET (VOIDmode,
- gen_rtx_MEM (BLKmode,
- gen_rtx_PRE_DEC (BLKmode,
- stack_pointer_rtx)),
+ gen_frame_mem (BLKmode,
+ gen_rtx_PRE_DEC (BLKmode,
+ stack_pointer_rtx)),
gen_rtx_UNSPEC (BLKmode,
gen_rtvec (1, reg),
UNSPEC_PUSH_MULT));
if (i != PC_REGNUM)
{
tmp = gen_rtx_SET (VOIDmode,
- gen_rtx_MEM (SImode, stack_pointer_rtx),
+ gen_frame_mem (SImode, stack_pointer_rtx),
reg);
RTX_FRAME_RELATED_P (tmp) = 1;
XVECEXP (dwarf, 0, dwarf_par_index) = tmp;
if (i != PC_REGNUM)
{
- tmp = gen_rtx_SET (VOIDmode,
- gen_rtx_MEM (SImode,
+ tmp
+ = gen_rtx_SET (VOIDmode,
+ gen_frame_mem (SImode,
plus_constant (stack_pointer_rtx,
4 * j)),
- reg);
+ reg);
RTX_FRAME_RELATED_P (tmp) = 1;
XVECEXP (dwarf, 0, dwarf_par_index++) = tmp;
}
par = emit_insn (par);
- tmp = gen_rtx_SET (SImode,
+ tmp = gen_rtx_SET (VOIDmode,
stack_pointer_rtx,
- gen_rtx_PLUS (SImode,
- stack_pointer_rtx,
- GEN_INT (-4 * num_regs)));
+ plus_constant (stack_pointer_rtx, -4 * num_regs));
RTX_FRAME_RELATED_P (tmp) = 1;
XVECEXP (dwarf, 0, 0) = tmp;
XVECEXP (par, 0, 0)
= gen_rtx_SET (VOIDmode,
- gen_rtx_MEM (BLKmode,
- gen_rtx_PRE_DEC (BLKmode, stack_pointer_rtx)),
+ gen_frame_mem (BLKmode,
+ gen_rtx_PRE_DEC (BLKmode,
+ stack_pointer_rtx)),
gen_rtx_UNSPEC (BLKmode,
gen_rtvec (1, reg),
UNSPEC_PUSH_MULT));
tmp = gen_rtx_SET (VOIDmode,
- gen_rtx_MEM (XFmode, stack_pointer_rtx), reg);
+ gen_frame_mem (XFmode, stack_pointer_rtx), reg);
RTX_FRAME_RELATED_P (tmp) = 1;
XVECEXP (dwarf, 0, 1) = tmp;
XVECEXP (par, 0, i) = gen_rtx_USE (VOIDmode, reg);
tmp = gen_rtx_SET (VOIDmode,
- gen_rtx_MEM (XFmode,
- plus_constant (stack_pointer_rtx,
- i * 12)),
+ gen_frame_mem (XFmode,
+ plus_constant (stack_pointer_rtx,
+ i * 12)),
reg);
RTX_FRAME_RELATED_P (tmp) = 1;
XVECEXP (dwarf, 0, i + 1) = tmp;
tmp = gen_rtx_SET (VOIDmode,
stack_pointer_rtx,
- gen_rtx_PLUS (SImode,
- stack_pointer_rtx,
- GEN_INT (-12 * count)));
+ plus_constant (stack_pointer_rtx, -12 * count));
+
RTX_FRAME_RELATED_P (tmp) = 1;
XVECEXP (dwarf, 0, 0) = tmp;
return !cfun->machine->lr_save_eliminated
&& (!leaf_function_p ()
|| thumb_far_jump_used_p ()
- || regs_ever_live [LR_REGNUM]);
+ || df_regs_ever_live_p (LR_REGNUM));
}
| | \
| | local
| | variables
- | | /
+ locals base pointer -> | | /
--
| | \
| | outgoing
/* Space for variadic functions. */
offsets->saved_args = current_function_pretend_args_size;
+ /* In Thumb mode this is incorrect, but never used. */
offsets->frame = offsets->saved_args + (frame_pointer_needed ? 4 : 0);
- if (TARGET_ARM)
+ if (TARGET_32BIT)
{
unsigned int regno;
for (regno = FIRST_IWMMXT_REGNUM;
regno <= LAST_IWMMXT_REGNUM;
regno++)
- if (regs_ever_live [regno] && ! call_used_regs [regno])
+ if (df_regs_ever_live_p (regno) && ! call_used_regs[regno])
saved += 8;
}
{
/* Space for saved FPA registers. */
for (regno = FIRST_FPA_REGNUM; regno <= LAST_FPA_REGNUM; regno++)
- if (regs_ever_live[regno] && ! call_used_regs[regno])
+ if (df_regs_ever_live_p (regno) && ! call_used_regs[regno])
saved += 12;
/* Space for saved VFP registers. */
saved += arm_get_vfp_saved_size ();
}
}
- else /* TARGET_THUMB */
+ else /* TARGET_THUMB1 */
{
- saved = bit_count (thumb_compute_save_reg_mask ()) * 4;
+ saved = bit_count (thumb1_compute_save_reg_mask ()) * 4;
if (TARGET_BACKTRACE)
saved += 16;
}
if (leaf && frame_size == 0)
{
offsets->outgoing_args = offsets->soft_frame;
+ offsets->locals_base = offsets->soft_frame;
return offsets;
}
&& (offsets->soft_frame & 7))
offsets->soft_frame += 4;
- offsets->outgoing_args = offsets->soft_frame + frame_size
- + current_function_outgoing_args_size;
+ offsets->locals_base = offsets->soft_frame + frame_size;
+ offsets->outgoing_args = (offsets->locals_base
+ + current_function_outgoing_args_size);
if (ARM_DOUBLEWORD_ALIGN)
{
}
-/* Generate the prologue instructions for entry into an ARM function. */
+/* Emit RTL to save coprocessor registers on function entry. Returns the
+ number of bytes pushed. */
+
+static int
+arm_save_coproc_regs(void)
+{
+ int saved_size = 0;
+ unsigned reg;
+ unsigned start_reg;
+ rtx insn;
+
+ for (reg = LAST_IWMMXT_REGNUM; reg >= FIRST_IWMMXT_REGNUM; reg--)
+ if (df_regs_ever_live_p (reg) && ! call_used_regs[reg])
+ {
+ insn = gen_rtx_PRE_DEC (V2SImode, stack_pointer_rtx);
+ insn = gen_rtx_MEM (V2SImode, insn);
+ insn = emit_set_insn (insn, gen_rtx_REG (V2SImode, reg));
+ RTX_FRAME_RELATED_P (insn) = 1;
+ saved_size += 8;
+ }
+
+ /* Save any floating point call-saved registers used by this
+ function. */
+ if (arm_fpu_arch == FPUTYPE_FPA_EMU2)
+ {
+ for (reg = LAST_FPA_REGNUM; reg >= FIRST_FPA_REGNUM; reg--)
+ if (df_regs_ever_live_p (reg) && !call_used_regs[reg])
+ {
+ insn = gen_rtx_PRE_DEC (XFmode, stack_pointer_rtx);
+ insn = gen_rtx_MEM (XFmode, insn);
+ insn = emit_set_insn (insn, gen_rtx_REG (XFmode, reg));
+ RTX_FRAME_RELATED_P (insn) = 1;
+ saved_size += 12;
+ }
+ }
+ else
+ {
+ start_reg = LAST_FPA_REGNUM;
+
+ for (reg = LAST_FPA_REGNUM; reg >= FIRST_FPA_REGNUM; reg--)
+ {
+ if (df_regs_ever_live_p (reg) && !call_used_regs[reg])
+ {
+ if (start_reg - reg == 3)
+ {
+ insn = emit_sfm (reg, 4);
+ RTX_FRAME_RELATED_P (insn) = 1;
+ saved_size += 48;
+ start_reg = reg - 1;
+ }
+ }
+ else
+ {
+ if (start_reg != reg)
+ {
+ insn = emit_sfm (reg + 1, start_reg - reg);
+ RTX_FRAME_RELATED_P (insn) = 1;
+ saved_size += (start_reg - reg) * 12;
+ }
+ start_reg = reg - 1;
+ }
+ }
+
+ if (start_reg != reg)
+ {
+ insn = emit_sfm (reg + 1, start_reg - reg);
+ saved_size += (start_reg - reg) * 12;
+ RTX_FRAME_RELATED_P (insn) = 1;
+ }
+ }
+ if (TARGET_HARD_FLOAT && TARGET_VFP)
+ {
+ start_reg = FIRST_VFP_REGNUM;
+
+ for (reg = FIRST_VFP_REGNUM; reg < LAST_VFP_REGNUM; reg += 2)
+ {
+ if ((!df_regs_ever_live_p (reg) || call_used_regs[reg])
+ && (!df_regs_ever_live_p (reg + 1) || call_used_regs[reg + 1]))
+ {
+ if (start_reg != reg)
+ saved_size += vfp_emit_fstmd (start_reg,
+ (reg - start_reg) / 2);
+ start_reg = reg + 2;
+ }
+ }
+ if (start_reg != reg)
+ saved_size += vfp_emit_fstmd (start_reg,
+ (reg - start_reg) / 2);
+ }
+ return saved_size;
+}
+
+
+/* Set the Thumb frame pointer from the stack pointer. */
+
+static void
+thumb_set_frame_pointer (arm_stack_offsets *offsets)
+{
+ HOST_WIDE_INT amount;
+ rtx insn, dwarf;
+
+ amount = offsets->outgoing_args - offsets->locals_base;
+ if (amount < 1024)
+ insn = emit_insn (gen_addsi3 (hard_frame_pointer_rtx,
+ stack_pointer_rtx, GEN_INT (amount)));
+ else
+ {
+ emit_insn (gen_movsi (hard_frame_pointer_rtx, GEN_INT (amount)));
+ insn = emit_insn (gen_addsi3 (hard_frame_pointer_rtx,
+ hard_frame_pointer_rtx,
+ stack_pointer_rtx));
+ dwarf = gen_rtx_SET (VOIDmode, hard_frame_pointer_rtx,
+ plus_constant (stack_pointer_rtx, amount));
+ RTX_FRAME_RELATED_P (dwarf) = 1;
+ REG_NOTES (insn) = gen_rtx_EXPR_LIST (REG_FRAME_RELATED_EXPR, dwarf,
+ REG_NOTES (insn));
+ }
+
+ RTX_FRAME_RELATED_P (insn) = 1;
+}
+
+/* Generate the prologue instructions for entry into an ARM or Thumb-2
+ function. */
void
arm_expand_prologue (void)
{
- int reg;
rtx amount;
rtx insn;
rtx ip_rtx;
ip_rtx = gen_rtx_REG (SImode, IP_REGNUM);
- if (frame_pointer_needed)
+ if (IS_STACKALIGN (func_type))
+ {
+ rtx dwarf;
+ rtx r0;
+ rtx r1;
+ /* Handle a word-aligned stack pointer. We generate the following:
+
+ mov r0, sp
+ bic r1, r0, #7
+ mov sp, r1
+ <save and restore r0 in normal prologue/epilogue>
+ mov sp, r0
+ bx lr
+
+ The unwinder doesn't need to know about the stack realignment.
+ Just tell it we saved SP in r0. */
+ gcc_assert (TARGET_THUMB2 && !arm_arch_notm && args_to_push == 0);
+
+ r0 = gen_rtx_REG (SImode, 0);
+ r1 = gen_rtx_REG (SImode, 1);
+ dwarf = gen_rtx_UNSPEC (SImode, NULL_RTVEC, UNSPEC_STACK_ALIGN);
+ dwarf = gen_rtx_SET (VOIDmode, r0, dwarf);
+ insn = gen_movsi (r0, stack_pointer_rtx);
+ RTX_FRAME_RELATED_P (insn) = 1;
+ REG_NOTES (insn) = gen_rtx_EXPR_LIST (REG_FRAME_RELATED_EXPR,
+ dwarf, REG_NOTES (insn));
+ emit_insn (insn);
+ emit_insn (gen_andsi3 (r1, r0, GEN_INT (~(HOST_WIDE_INT)7)));
+ emit_insn (gen_movsi (stack_pointer_rtx, r1));
+ }
+
+ if (frame_pointer_needed && TARGET_ARM)
{
if (IS_INTERRUPT (func_type))
{
doesn't need to be unwound, as it doesn't contain a value
inherited from the caller. */
- if (regs_ever_live[3] == 0)
- {
- insn = gen_rtx_REG (SImode, 3);
- insn = gen_rtx_SET (SImode, insn, ip_rtx);
- insn = emit_insn (insn);
- }
+ if (df_regs_ever_live_p (3) == false)
+ insn = emit_set_insn (gen_rtx_REG (SImode, 3), ip_rtx);
else if (args_to_push == 0)
{
rtx dwarf;
- insn = gen_rtx_PRE_DEC (SImode, stack_pointer_rtx);
- insn = gen_rtx_MEM (SImode, insn);
- insn = gen_rtx_SET (VOIDmode, insn, ip_rtx);
- insn = emit_insn (insn);
+ insn = gen_rtx_PRE_DEC (SImode, stack_pointer_rtx);
+ insn = emit_set_insn (gen_frame_mem (SImode, insn), ip_rtx);
fp_offset = 4;
/* Just tell the dwarf backend that we adjusted SP. */
dwarf = gen_rtx_SET (VOIDmode, stack_pointer_rtx,
- gen_rtx_PLUS (SImode, stack_pointer_rtx,
- GEN_INT (-fp_offset)));
+ plus_constant (stack_pointer_rtx,
+ -fp_offset));
RTX_FRAME_RELATED_P (insn) = 1;
REG_NOTES (insn) = gen_rtx_EXPR_LIST (REG_FRAME_RELATED_EXPR,
dwarf, REG_NOTES (insn));
args_to_push = 0;
/* Now reuse r3 to preserve IP. */
- insn = gen_rtx_REG (SImode, 3);
- insn = gen_rtx_SET (SImode, insn, ip_rtx);
- (void) emit_insn (insn);
+ emit_set_insn (gen_rtx_REG (SImode, 3), ip_rtx);
}
}
- if (fp_offset)
- {
- insn = gen_rtx_PLUS (SImode, stack_pointer_rtx, GEN_INT (fp_offset));
- insn = gen_rtx_SET (SImode, ip_rtx, insn);
- }
- else
- insn = gen_movsi (ip_rtx, stack_pointer_rtx);
-
- insn = emit_insn (insn);
+ insn = emit_set_insn (ip_rtx,
+ plus_constant (stack_pointer_rtx, fp_offset));
RTX_FRAME_RELATED_P (insn) = 1;
}
can be done with a single instruction. */
if ((func_type == ARM_FT_ISR || func_type == ARM_FT_FIQ)
&& (live_regs_mask & (1 << LR_REGNUM)) != 0
- && ! frame_pointer_needed)
- emit_insn (gen_rtx_SET (SImode,
- gen_rtx_REG (SImode, LR_REGNUM),
- gen_rtx_PLUS (SImode,
- gen_rtx_REG (SImode, LR_REGNUM),
- GEN_INT (-4))));
+ && ! frame_pointer_needed
+ && TARGET_ARM)
+ {
+ rtx lr = gen_rtx_REG (SImode, LR_REGNUM);
+
+ emit_set_insn (lr, plus_constant (lr, -4));
+ }
if (live_regs_mask)
{
RTX_FRAME_RELATED_P (insn) = 1;
}
- if (TARGET_IWMMXT)
- for (reg = LAST_IWMMXT_REGNUM; reg >= FIRST_IWMMXT_REGNUM; reg--)
- if (regs_ever_live[reg] && ! call_used_regs [reg])
- {
- insn = gen_rtx_PRE_DEC (V2SImode, stack_pointer_rtx);
- insn = gen_rtx_MEM (V2SImode, insn);
- insn = emit_insn (gen_rtx_SET (VOIDmode, insn,
- gen_rtx_REG (V2SImode, reg)));
- RTX_FRAME_RELATED_P (insn) = 1;
- saved_regs += 8;
- }
-
if (! IS_VOLATILE (func_type))
- {
- int start_reg;
-
- /* Save any floating point call-saved registers used by this
- function. */
- if (arm_fpu_arch == FPUTYPE_FPA_EMU2)
- {
- for (reg = LAST_FPA_REGNUM; reg >= FIRST_FPA_REGNUM; reg--)
- if (regs_ever_live[reg] && !call_used_regs[reg])
- {
- insn = gen_rtx_PRE_DEC (XFmode, stack_pointer_rtx);
- insn = gen_rtx_MEM (XFmode, insn);
- insn = emit_insn (gen_rtx_SET (VOIDmode, insn,
- gen_rtx_REG (XFmode, reg)));
- RTX_FRAME_RELATED_P (insn) = 1;
- saved_regs += 12;
- }
- }
- else
- {
- start_reg = LAST_FPA_REGNUM;
-
- for (reg = LAST_FPA_REGNUM; reg >= FIRST_FPA_REGNUM; reg--)
- {
- if (regs_ever_live[reg] && !call_used_regs[reg])
- {
- if (start_reg - reg == 3)
- {
- insn = emit_sfm (reg, 4);
- RTX_FRAME_RELATED_P (insn) = 1;
- saved_regs += 48;
- start_reg = reg - 1;
- }
- }
- else
- {
- if (start_reg != reg)
- {
- insn = emit_sfm (reg + 1, start_reg - reg);
- RTX_FRAME_RELATED_P (insn) = 1;
- saved_regs += (start_reg - reg) * 12;
- }
- start_reg = reg - 1;
- }
- }
+ saved_regs += arm_save_coproc_regs ();
- if (start_reg != reg)
- {
- insn = emit_sfm (reg + 1, start_reg - reg);
- saved_regs += (start_reg - reg) * 12;
- RTX_FRAME_RELATED_P (insn) = 1;
- }
- }
- if (TARGET_HARD_FLOAT && TARGET_VFP)
+ if (frame_pointer_needed && TARGET_ARM)
+ {
+ /* Create the new frame pointer. */
{
- start_reg = FIRST_VFP_REGNUM;
+ insn = GEN_INT (-(4 + args_to_push + fp_offset));
+ insn = emit_insn (gen_addsi3 (hard_frame_pointer_rtx, ip_rtx, insn));
+ RTX_FRAME_RELATED_P (insn) = 1;
- for (reg = FIRST_VFP_REGNUM; reg < LAST_VFP_REGNUM; reg += 2)
+ if (IS_NESTED (func_type))
{
- if ((!regs_ever_live[reg] || call_used_regs[reg])
- && (!regs_ever_live[reg + 1] || call_used_regs[reg + 1]))
+ /* Recover the static chain register. */
+ if (!df_regs_ever_live_p (3)
+ || saved_pretend_args)
+ insn = gen_rtx_REG (SImode, 3);
+ else /* if (current_function_pretend_args_size == 0) */
{
- if (start_reg != reg)
- saved_regs += vfp_emit_fstmx (start_reg,
- (reg - start_reg) / 2);
- start_reg = reg + 2;
+ insn = plus_constant (hard_frame_pointer_rtx, 4);
+ insn = gen_frame_mem (SImode, insn);
}
+ emit_set_insn (ip_rtx, insn);
+ /* Add a USE to stop propagate_one_insn() from barfing. */
+ emit_insn (gen_prologue_use (ip_rtx));
}
- if (start_reg != reg)
- saved_regs += vfp_emit_fstmx (start_reg,
- (reg - start_reg) / 2);
- }
- }
-
- if (frame_pointer_needed)
- {
- /* Create the new frame pointer. */
- insn = GEN_INT (-(4 + args_to_push + fp_offset));
- insn = emit_insn (gen_addsi3 (hard_frame_pointer_rtx, ip_rtx, insn));
- RTX_FRAME_RELATED_P (insn) = 1;
-
- if (IS_NESTED (func_type))
- {
- /* Recover the static chain register. */
- if (regs_ever_live [3] == 0
- || saved_pretend_args)
- insn = gen_rtx_REG (SImode, 3);
- else /* if (current_function_pretend_args_size == 0) */
- {
- insn = gen_rtx_PLUS (SImode, hard_frame_pointer_rtx,
- GEN_INT (4));
- insn = gen_rtx_MEM (SImode, insn);
- }
-
- emit_insn (gen_rtx_SET (SImode, ip_rtx, insn));
- /* Add a USE to stop propagate_one_insn() from barfing. */
- emit_insn (gen_prologue_use (ip_rtx));
}
}
}
- if (flag_pic)
- arm_load_pic_register (INVALID_REGNUM);
+ if (frame_pointer_needed && TARGET_THUMB2)
+ thumb_set_frame_pointer (offsets);
+
+ if (flag_pic && arm_pic_register != INVALID_REGNUM)
+ {
+ unsigned long mask;
+
+ mask = live_regs_mask;
+ mask &= THUMB2_WORK_REGS;
+ if (!IS_NESTED (func_type))
+ mask |= (1 << IP_REGNUM);
+ arm_load_pic_register (mask);
+ }
/* If we are profiling, make sure no instructions are scheduled before
the call to mcount. Similarly if the user has requested no
- scheduling in the prolog. */
- if (current_function_profile || !TARGET_SCHED_PROLOG)
+ scheduling in the prolog. Similarly if we want non-call exceptions
+ using the EABI unwinder, to prevent faulting instructions from being
+ swapped with a stack adjustment. */
+ if (current_function_profile || !TARGET_SCHED_PROLOG
+ || (ARM_EABI_UNWIND_TABLES && flag_non_call_exceptions))
emit_insn (gen_blockage ());
/* If the link register is being kept alive, with the return address in it,
then make sure that it does not get reused by the ce2 pass. */
if ((live_regs_mask & (1 << LR_REGNUM)) == 0)
+ cfun->machine->lr_save_eliminated = 1;
+}
+\f
+/* Print condition code to STREAM. Helper function for arm_print_operand. */
+static void
+arm_print_condition (FILE *stream)
+{
+ if (arm_ccfsm_state == 3 || arm_ccfsm_state == 4)
+ {
+ /* Branch conversion is not implemented for Thumb-2. */
+ if (TARGET_THUMB)
+ {
+ output_operand_lossage ("predicated Thumb instruction");
+ return;
+ }
+ if (current_insn_predicate != NULL)
+ {
+ output_operand_lossage
+ ("predicated instruction in conditional sequence");
+ return;
+ }
+
+ fputs (arm_condition_codes[arm_current_cc], stream);
+ }
+ else if (current_insn_predicate)
{
- emit_insn (gen_prologue_use (gen_rtx_REG (SImode, LR_REGNUM)));
- cfun->machine->lr_save_eliminated = 1;
+ enum arm_cond_code code;
+
+ if (TARGET_THUMB1)
+ {
+ output_operand_lossage ("predicated Thumb instruction");
+ return;
+ }
+
+ code = get_arm_condition_code (current_insn_predicate);
+ fputs (arm_condition_codes[code], stream);
}
}
-\f
+
+
/* If CODE is 'd', then the X is a condition operand and the instruction
should only be executed if the condition is true.
if CODE is 'D', then the X is a condition operand and the instruction
return;
case '?':
- if (arm_ccfsm_state == 3 || arm_ccfsm_state == 4)
- {
- if (TARGET_THUMB)
- {
- output_operand_lossage ("predicated Thumb instruction");
- break;
- }
- if (current_insn_predicate != NULL)
- {
- output_operand_lossage
- ("predicated instruction in conditional sequence");
- break;
- }
+ arm_print_condition (stream);
+ return;
- fputs (arm_condition_codes[arm_current_cc], stream);
+ case '(':
+ /* Nothing in unified syntax, otherwise the current condition code. */
+ if (!TARGET_UNIFIED_ASM)
+ arm_print_condition (stream);
+ break;
+
+ case ')':
+ /* The current condition code in unified syntax, otherwise nothing. */
+ if (TARGET_UNIFIED_ASM)
+ arm_print_condition (stream);
+ break;
+
+ case '.':
+ /* The current condition code for a condition code setting instruction.
+ Preceded by 's' in unified syntax, otherwise followed by 's'. */
+ if (TARGET_UNIFIED_ASM)
+ {
+ fputc('s', stream);
+ arm_print_condition (stream);
}
- else if (current_insn_predicate)
+ else
{
- enum arm_cond_code code;
-
- if (TARGET_THUMB)
- {
- output_operand_lossage ("predicated Thumb instruction");
- break;
- }
-
- code = get_arm_condition_code (current_insn_predicate);
- fputs (arm_condition_codes[code], stream);
+ arm_print_condition (stream);
+ fputc('s', stream);
}
return;
+ case '!':
+ /* If the instruction is conditionally executed then print
+ the current condition code, otherwise print 's'. */
+ gcc_assert (TARGET_THUMB2 && TARGET_UNIFIED_ASM);
+ if (current_insn_predicate)
+ arm_print_condition (stream);
+ else
+ fputc('s', stream);
+ break;
+
case 'N':
{
REAL_VALUE_TYPE r;
}
return;
+ case 'L':
+ /* The low 16 bits of an immediate constant. */
+ fprintf (stream, HOST_WIDE_INT_PRINT_DEC, INTVAL(x) & 0xffff);
+ return;
+
case 'i':
fprintf (stream, "%s", arithmetic_instr (x, 1));
return;
case 'S':
{
HOST_WIDE_INT val;
- const char * shift = shift_op (x, &val);
+ const char *shift;
+
+ if (!shift_operator (x, SImode))
+ {
+ output_operand_lossage ("invalid shift operand");
+ break;
+ }
+
+ shift = shift_op (x, &val);
if (shift)
{
- fprintf (stream, ", %s ", shift_op (x, &val));
+ fprintf (stream, ", %s ", shift);
if (val == -1)
arm_print_operand (stream, XEXP (x, 1), 0);
else
want to do that. */
if (x == const_true_rtx)
{
- output_operand_lossage ("instruction never exectued");
+ output_operand_lossage ("instruction never executed");
return;
}
if (!COMPARISON_P (x))
}
return;
+ /* Print a VFPv3 floating-point constant, represented as an integer
+ index. */
+ case 'G':
+ {
+ int index = vfp3_const_double_index (x);
+ gcc_assert (index != -1);
+ fprintf (stream, "%d", index);
+ }
+ return;
+
default:
if (x == 0)
{
if (NEED_GOT_RELOC && flag_pic && making_const_table &&
(GET_CODE (x) == SYMBOL_REF || GET_CODE (x) == LABEL_REF))
{
- if (GET_CODE (x) == SYMBOL_REF
- && (CONSTANT_POOL_ADDRESS_P (x)
- || SYMBOL_REF_LOCAL_P (x)))
- fputs ("(GOTOFF)", asm_out_file);
- else if (GET_CODE (x) == LABEL_REF)
- fputs ("(GOTOFF)", asm_out_file);
- else
+ /* See legitimize_pic_address for an explanation of the
+ TARGET_VXWORKS_RTP check. */
+ if (TARGET_VXWORKS_RTP
+ || (GET_CODE (x) == SYMBOL_REF && !SYMBOL_REF_LOCAL_P (x)))
fputs ("(GOT)", asm_out_file);
+ else
+ fputs ("(GOTOFF)", asm_out_file);
}
fputc ('\n', asm_out_file);
return true;
return default_assemble_integer (x, size, aligned_p);
}
-
-/* Add a function to the list of static constructors. */
-
static void
-arm_elf_asm_constructor (rtx symbol, int priority ATTRIBUTE_UNUSED)
+arm_elf_asm_cdtor (rtx symbol, int priority, bool is_ctor)
{
+ section *s;
+
if (!TARGET_AAPCS_BASED)
{
- default_named_section_asm_out_constructor (symbol, priority);
+ (is_ctor ?
+ default_named_section_asm_out_constructor
+ : default_named_section_asm_out_destructor) (symbol, priority);
return;
}
/* Put these in the .init_array section, using a special relocation. */
- ctors_section ();
+ if (priority != DEFAULT_INIT_PRIORITY)
+ {
+ char buf[18];
+ sprintf (buf, "%s.%.5u",
+ is_ctor ? ".init_array" : ".fini_array",
+ priority);
+ s = get_section (buf, SECTION_WRITE, NULL_TREE);
+ }
+ else if (is_ctor)
+ s = ctors_section;
+ else
+ s = dtors_section;
+
+ switch_to_section (s);
assemble_align (POINTER_SIZE);
fputs ("\t.word\t", asm_out_file);
output_addr_const (asm_out_file, symbol);
fputs ("(target1)\n", asm_out_file);
}
+
+/* Add a function to the list of static constructors. */
+
+static void
+arm_elf_asm_constructor (rtx symbol, int priority)
+{
+ arm_elf_asm_cdtor (symbol, priority, /*is_ctor=*/true);
+}
+
+/* Add a function to the list of static destructors. */
+
+static void
+arm_elf_asm_destructor (rtx symbol, int priority)
+{
+ arm_elf_asm_cdtor (symbol, priority, /*is_ctor=*/false);
+}
#endif
\f
/* A finite state machine takes care of noticing whether or not instructions
time. But then, I want to reduce the code size to somewhere near what
/bin/cc produces. */
+/* In addition to this, state is maintained for Thumb-2 COND_EXEC
+ instructions. When a COND_EXEC instruction is seen the subsequent
+ instructions are scanned so that multiple conditional instructions can be
+ combined into a single IT block. arm_condexec_count and arm_condexec_mask
+ specify the length and true/false mask for the IT block. These will be
+ decremented/zeroed by arm_asm_output_opcode as the insns are output. */
+
/* Returns the index of the ARM condition code string in
`arm_condition_codes'. COMPARISON should be an rtx like
`(eq (...) (...))'. */
default: gcc_unreachable ();
}
- case CC_Cmode:
- switch (comp_code)
- {
- case LTU: return ARM_CS;
- case GEU: return ARM_CC;
- default: gcc_unreachable ();
- }
+ case CC_Cmode:
+ switch (comp_code)
+ {
+ case LTU: return ARM_CS;
+ case GEU: return ARM_CC;
+ default: gcc_unreachable ();
+ }
+
+ case CCmode:
+ switch (comp_code)
+ {
+ case NE: return ARM_NE;
+ case EQ: return ARM_EQ;
+ case GE: return ARM_GE;
+ case GT: return ARM_GT;
+ case LE: return ARM_LE;
+ case LT: return ARM_LT;
+ case GEU: return ARM_CS;
+ case GTU: return ARM_HI;
+ case LEU: return ARM_LS;
+ case LTU: return ARM_CC;
+ default: gcc_unreachable ();
+ }
+
+ default: gcc_unreachable ();
+ }
+}
+
+/* Tell arm_asm_output_opcode to output IT blocks for conditionally executed
+ instructions. */
+void
+thumb2_final_prescan_insn (rtx insn)
+{
+ rtx first_insn = insn;
+ rtx body = PATTERN (insn);
+ rtx predicate;
+ enum arm_cond_code code;
+ int n;
+ int mask;
+
+ /* Remove the previous insn from the count of insns to be output. */
+ if (arm_condexec_count)
+ arm_condexec_count--;
+
+ /* Nothing to do if we are already inside a conditional block. */
+ if (arm_condexec_count)
+ return;
+
+ if (GET_CODE (body) != COND_EXEC)
+ return;
+
+ /* Conditional jumps are implemented directly. */
+ if (GET_CODE (insn) == JUMP_INSN)
+ return;
+
+ predicate = COND_EXEC_TEST (body);
+ arm_current_cc = get_arm_condition_code (predicate);
+
+ n = get_attr_ce_count (insn);
+ arm_condexec_count = 1;
+ arm_condexec_mask = (1 << n) - 1;
+ arm_condexec_masklen = n;
+ /* See if subsequent instructions can be combined into the same block. */
+ for (;;)
+ {
+ insn = next_nonnote_insn (insn);
+
+ /* Jumping into the middle of an IT block is illegal, so a label or
+ barrier terminates the block. */
+ if (GET_CODE (insn) != INSN && GET_CODE(insn) != JUMP_INSN)
+ break;
+
+ body = PATTERN (insn);
+ /* USE and CLOBBER aren't really insns, so just skip them. */
+ if (GET_CODE (body) == USE
+ || GET_CODE (body) == CLOBBER)
+ continue;
+
+ /* ??? Recognize conditional jumps, and combine them with IT blocks. */
+ if (GET_CODE (body) != COND_EXEC)
+ break;
+ /* Allow up to 4 conditionally executed instructions in a block. */
+ n = get_attr_ce_count (insn);
+ if (arm_condexec_masklen + n > 4)
+ break;
+
+ predicate = COND_EXEC_TEST (body);
+ code = get_arm_condition_code (predicate);
+ mask = (1 << n) - 1;
+ if (arm_current_cc == code)
+ arm_condexec_mask |= (mask << arm_condexec_masklen);
+ else if (arm_current_cc != ARM_INVERSE_CONDITION_CODE(code))
+ break;
- case CCmode:
- switch (comp_code)
- {
- case NE: return ARM_NE;
- case EQ: return ARM_EQ;
- case GE: return ARM_GE;
- case GT: return ARM_GT;
- case LE: return ARM_LE;
- case LT: return ARM_LT;
- case GEU: return ARM_CS;
- case GTU: return ARM_HI;
- case LEU: return ARM_LS;
- case LTU: return ARM_CC;
- default: gcc_unreachable ();
- }
+ arm_condexec_count++;
+ arm_condexec_masklen += n;
- default: gcc_unreachable ();
+ /* A jump must be the last instruction in a conditional block. */
+ if (GET_CODE(insn) == JUMP_INSN)
+ break;
}
+ /* Restore recog_data (getting the attributes of other insns can
+ destroy this array, but final.c assumes that it remains intact
+ across this call). */
+ extract_constrain_insn_cached (first_insn);
}
void
else
{
gcc_assert (seeking_return || arm_ccfsm_state == 2);
-
+
while (this_insn && GET_CODE (PATTERN (this_insn)) == USE)
{
this_insn = next_nonnote_insn (this_insn);
if (!this_insn)
{
/* Oh, dear! we ran off the end.. give up. */
- recog (PATTERN (insn), insn, NULL);
+ extract_constrain_insn_cached (insn);
arm_ccfsm_state = 0;
arm_target_insn = NULL;
return;
/* Restore recog_data (getting the attributes of other insns can
destroy this array, but final.c assumes that it remains intact
- across this call; since the insn has been recognized already we
- call recog direct). */
- recog (PATTERN (insn), insn, NULL);
+ across this call. */
+ extract_constrain_insn_cached (insn);
+ }
+}
+
+/* Output IT instructions. */
+void
+thumb2_asm_output_opcode (FILE * stream)
+{
+ char buff[5];
+ int n;
+
+ if (arm_condexec_mask)
+ {
+ for (n = 0; n < arm_condexec_masklen; n++)
+ buff[n] = (arm_condexec_mask & (1 << n)) ? 't' : 'e';
+ buff[n] = 0;
+ asm_fprintf(stream, "i%s\t%s\n\t", buff,
+ arm_condition_codes[arm_current_cc]);
+ arm_condexec_mask = 0;
}
}
arm_hard_regno_mode_ok (unsigned int regno, enum machine_mode mode)
{
if (GET_MODE_CLASS (mode) == MODE_CC)
- return regno == CC_REGNUM || regno == VFPCC_REGNUM;
+ return (regno == CC_REGNUM
+ || (TARGET_HARD_FLOAT && TARGET_VFP
+ && regno == VFPCC_REGNUM));
- if (TARGET_THUMB)
+ if (TARGET_THUMB1)
/* For the Thumb we only allow values bigger than SImode in
registers 0 - 6, so that there is always a second low
register available to hold the upper part of the value.
start of an even numbered register pair. */
return (ARM_NUM_REGS (mode) < 2) || (regno < LAST_LO_REGNUM);
- if (IS_CIRRUS_REGNUM (regno))
+ if (TARGET_HARD_FLOAT && TARGET_MAVERICK
+ && IS_CIRRUS_REGNUM (regno))
/* We have outlawed SI values in Cirrus registers because they
reside in the lower 32 bits, but SF values reside in the
upper 32 bits. This causes gcc all sorts of grief. We can't
get sign extended to 64bits-- aldyh. */
return (GET_MODE_CLASS (mode) == MODE_FLOAT) || (mode == DImode);
- if (IS_VFP_REGNUM (regno))
+ if (TARGET_HARD_FLOAT && TARGET_VFP
+ && IS_VFP_REGNUM (regno))
{
if (mode == SFmode || mode == SImode)
- return TRUE;
+ return VFP_REGNO_OK_FOR_SINGLE (regno);
- /* DFmode values are only valid in even register pairs. */
if (mode == DFmode)
- return ((regno - FIRST_VFP_REGNUM) & 1) == 0;
+ return VFP_REGNO_OK_FOR_DOUBLE (regno);
return FALSE;
}
- if (IS_IWMMXT_GR_REGNUM (regno))
- return mode == SImode;
-
- if (IS_IWMMXT_REGNUM (regno))
- return VALID_IWMMXT_REG_MODE (mode);
+ if (TARGET_REALLY_IWMMXT)
+ {
+ if (IS_IWMMXT_GR_REGNUM (regno))
+ return mode == SImode;
+ if (IS_IWMMXT_REGNUM (regno))
+ return VALID_IWMMXT_REG_MODE (mode);
+ }
+
/* We allow any value to be stored in the general registers.
Restrict doubleword quantities to even register pairs so that we can
use ldrd. */
if (regno <= LAST_ARM_REGNUM)
return !(TARGET_LDRD && GET_MODE_SIZE (mode) > 4 && (regno & 1) != 0);
- if ( regno == FRAME_POINTER_REGNUM
+ if (regno == FRAME_POINTER_REGNUM
|| regno == ARG_POINTER_REGNUM)
/* We only allow integers in the fake hard registers. */
return GET_MODE_CLASS (mode) == MODE_INT;
/* The only registers left are the FPA registers
which we only allow to hold FP values. */
- return GET_MODE_CLASS (mode) == MODE_FLOAT
- && regno >= FIRST_FPA_REGNUM
- && regno <= LAST_FPA_REGNUM;
+ return (TARGET_HARD_FLOAT && TARGET_FPA
+ && GET_MODE_CLASS (mode) == MODE_FLOAT
+ && regno >= FIRST_FPA_REGNUM
+ && regno <= LAST_FPA_REGNUM);
}
+/* For efficiency and historical reasons LO_REGS, HI_REGS and CC_REGS are
+ not used in arm mode. */
int
arm_regno_class (int regno)
{
- if (TARGET_THUMB)
+ if (TARGET_THUMB1)
{
if (regno == STACK_POINTER_REGNUM)
return STACK_REG;
return HI_REGS;
}
+ if (TARGET_THUMB2 && regno < 8)
+ return LO_REGS;
+
if ( regno <= LAST_ARM_REGNUM
|| regno == FRAME_POINTER_REGNUM
|| regno == ARG_POINTER_REGNUM)
- return GENERAL_REGS;
+ return TARGET_THUMB2 ? HI_REGS : GENERAL_REGS;
if (regno == CC_REGNUM || regno == VFPCC_REGNUM)
- return NO_REGS;
+ return TARGET_THUMB2 ? CC_REG : NO_REGS;
if (IS_CIRRUS_REGNUM (regno))
return CIRRUS_REGS;
if (IS_VFP_REGNUM (regno))
- return VFP_REGS;
+ {
+ if (regno <= D7_VFP_REGNUM)
+ return VFP_D0_D7_REGS;
+ else if (regno <= LAST_LO_VFP_REGNUM)
+ return VFP_LO_REGS;
+ else
+ return VFP_HI_REGS;
+ }
if (IS_IWMMXT_REGNUM (regno))
return IWMMXT_REGS;
/* If we are using the stack pointer to point at the
argument, then an offset of 0 is correct. */
+ /* ??? Check this is consistent with thumb2 frame layout. */
if ((TARGET_THUMB || !frame_pointer_needed)
&& REGNO (addr) == SP_REGNUM)
return 0;
do \
{ \
if ((MASK) & insn_flags) \
- lang_hooks.builtin_function ((NAME), (TYPE), (CODE), \
- BUILT_IN_MD, NULL, NULL_TREE); \
+ add_builtin_function ((NAME), (TYPE), (CODE), \
+ BUILT_IN_MD, NULL, NULL_TREE); \
} \
while (0)
}
static void
+arm_init_tls_builtins (void)
+{
+ tree ftype;
+ tree nothrow = tree_cons (get_identifier ("nothrow"), NULL, NULL);
+ tree const_nothrow = tree_cons (get_identifier ("const"), NULL, nothrow);
+
+ ftype = build_function_type (ptr_type_node, void_list_node);
+ add_builtin_function ("__builtin_thread_pointer", ftype,
+ ARM_BUILTIN_THREAD_POINTER, BUILT_IN_MD,
+ NULL, const_nothrow);
+}
+
+static void
arm_init_builtins (void)
{
+ arm_init_tls_builtins ();
+
if (TARGET_REALLY_IWMMXT)
arm_init_iwmmxt_builtins ();
}
static rtx
arm_expand_binop_builtin (enum insn_code icode,
- tree arglist, rtx target)
+ tree exp, rtx target)
{
rtx pat;
- tree arg0 = TREE_VALUE (arglist);
- tree arg1 = TREE_VALUE (TREE_CHAIN (arglist));
- rtx op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
- rtx op1 = expand_expr (arg1, NULL_RTX, VOIDmode, 0);
+ tree arg0 = CALL_EXPR_ARG (exp, 0);
+ tree arg1 = CALL_EXPR_ARG (exp, 1);
+ rtx op0 = expand_normal (arg0);
+ rtx op1 = expand_normal (arg1);
enum machine_mode tmode = insn_data[icode].operand[0].mode;
enum machine_mode mode0 = insn_data[icode].operand[1].mode;
enum machine_mode mode1 = insn_data[icode].operand[2].mode;
static rtx
arm_expand_unop_builtin (enum insn_code icode,
- tree arglist, rtx target, int do_load)
+ tree exp, rtx target, int do_load)
{
rtx pat;
- tree arg0 = TREE_VALUE (arglist);
- rtx op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
+ tree arg0 = CALL_EXPR_ARG (exp, 0);
+ rtx op0 = expand_normal (arg0);
enum machine_mode tmode = insn_data[icode].operand[0].mode;
enum machine_mode mode0 = insn_data[icode].operand[1].mode;
{
const struct builtin_description * d;
enum insn_code icode;
- tree fndecl = TREE_OPERAND (TREE_OPERAND (exp, 0), 0);
- tree arglist = TREE_OPERAND (exp, 1);
+ tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
tree arg0;
tree arg1;
tree arg2;
: fcode == ARM_BUILTIN_TEXTRMUH ? CODE_FOR_iwmmxt_textrmuh
: CODE_FOR_iwmmxt_textrmw);
- arg0 = TREE_VALUE (arglist);
- arg1 = TREE_VALUE (TREE_CHAIN (arglist));
- op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
- op1 = expand_expr (arg1, NULL_RTX, VOIDmode, 0);
+ arg0 = CALL_EXPR_ARG (exp, 0);
+ arg1 = CALL_EXPR_ARG (exp, 1);
+ op0 = expand_normal (arg0);
+ op1 = expand_normal (arg1);
tmode = insn_data[icode].operand[0].mode;
mode0 = insn_data[icode].operand[1].mode;
mode1 = insn_data[icode].operand[2].mode;
icode = (fcode == ARM_BUILTIN_TINSRB ? CODE_FOR_iwmmxt_tinsrb
: fcode == ARM_BUILTIN_TINSRH ? CODE_FOR_iwmmxt_tinsrh
: CODE_FOR_iwmmxt_tinsrw);
- arg0 = TREE_VALUE (arglist);
- arg1 = TREE_VALUE (TREE_CHAIN (arglist));
- arg2 = TREE_VALUE (TREE_CHAIN (TREE_CHAIN (arglist)));
- op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
- op1 = expand_expr (arg1, NULL_RTX, VOIDmode, 0);
- op2 = expand_expr (arg2, NULL_RTX, VOIDmode, 0);
+ arg0 = CALL_EXPR_ARG (exp, 0);
+ arg1 = CALL_EXPR_ARG (exp, 1);
+ arg2 = CALL_EXPR_ARG (exp, 2);
+ op0 = expand_normal (arg0);
+ op1 = expand_normal (arg1);
+ op2 = expand_normal (arg2);
tmode = insn_data[icode].operand[0].mode;
mode0 = insn_data[icode].operand[1].mode;
mode1 = insn_data[icode].operand[2].mode;
return target;
case ARM_BUILTIN_SETWCX:
- arg0 = TREE_VALUE (arglist);
- arg1 = TREE_VALUE (TREE_CHAIN (arglist));
- op0 = force_reg (SImode, expand_expr (arg0, NULL_RTX, VOIDmode, 0));
- op1 = expand_expr (arg1, NULL_RTX, VOIDmode, 0);
+ arg0 = CALL_EXPR_ARG (exp, 0);
+ arg1 = CALL_EXPR_ARG (exp, 1);
+ op0 = force_reg (SImode, expand_normal (arg0));
+ op1 = expand_normal (arg1);
emit_insn (gen_iwmmxt_tmcr (op1, op0));
return 0;
case ARM_BUILTIN_GETWCX:
- arg0 = TREE_VALUE (arglist);
- op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
+ arg0 = CALL_EXPR_ARG (exp, 0);
+ op0 = expand_normal (arg0);
target = gen_reg_rtx (SImode);
emit_insn (gen_iwmmxt_tmrc (target, op0));
return target;
case ARM_BUILTIN_WSHUFH:
icode = CODE_FOR_iwmmxt_wshufh;
- arg0 = TREE_VALUE (arglist);
- arg1 = TREE_VALUE (TREE_CHAIN (arglist));
- op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
- op1 = expand_expr (arg1, NULL_RTX, VOIDmode, 0);
+ arg0 = CALL_EXPR_ARG (exp, 0);
+ arg1 = CALL_EXPR_ARG (exp, 1);
+ op0 = expand_normal (arg0);
+ op1 = expand_normal (arg1);
tmode = insn_data[icode].operand[0].mode;
mode1 = insn_data[icode].operand[1].mode;
mode2 = insn_data[icode].operand[2].mode;
return target;
case ARM_BUILTIN_WSADB:
- return arm_expand_binop_builtin (CODE_FOR_iwmmxt_wsadb, arglist, target);
+ return arm_expand_binop_builtin (CODE_FOR_iwmmxt_wsadb, exp, target);
case ARM_BUILTIN_WSADH:
- return arm_expand_binop_builtin (CODE_FOR_iwmmxt_wsadh, arglist, target);
+ return arm_expand_binop_builtin (CODE_FOR_iwmmxt_wsadh, exp, target);
case ARM_BUILTIN_WSADBZ:
- return arm_expand_binop_builtin (CODE_FOR_iwmmxt_wsadbz, arglist, target);
+ return arm_expand_binop_builtin (CODE_FOR_iwmmxt_wsadbz, exp, target);
case ARM_BUILTIN_WSADHZ:
- return arm_expand_binop_builtin (CODE_FOR_iwmmxt_wsadhz, arglist, target);
+ return arm_expand_binop_builtin (CODE_FOR_iwmmxt_wsadhz, exp, target);
/* Several three-argument builtins. */
case ARM_BUILTIN_WMACS:
: fcode == ARM_BUILTIN_TMIATB ? CODE_FOR_iwmmxt_tmiatb
: fcode == ARM_BUILTIN_TMIATT ? CODE_FOR_iwmmxt_tmiatt
: CODE_FOR_iwmmxt_walign);
- arg0 = TREE_VALUE (arglist);
- arg1 = TREE_VALUE (TREE_CHAIN (arglist));
- arg2 = TREE_VALUE (TREE_CHAIN (TREE_CHAIN (arglist)));
- op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
- op1 = expand_expr (arg1, NULL_RTX, VOIDmode, 0);
- op2 = expand_expr (arg2, NULL_RTX, VOIDmode, 0);
+ arg0 = CALL_EXPR_ARG (exp, 0);
+ arg1 = CALL_EXPR_ARG (exp, 1);
+ arg2 = CALL_EXPR_ARG (exp, 2);
+ op0 = expand_normal (arg0);
+ op1 = expand_normal (arg1);
+ op2 = expand_normal (arg2);
tmode = insn_data[icode].operand[0].mode;
mode0 = insn_data[icode].operand[1].mode;
mode1 = insn_data[icode].operand[2].mode;
emit_insn (gen_iwmmxt_clrdi (target));
return target;
+ case ARM_BUILTIN_THREAD_POINTER:
+ return arm_load_tp (target);
+
default:
break;
}
for (i = 0, d = bdesc_2arg; i < ARRAY_SIZE (bdesc_2arg); i++, d++)
if (d->code == (const enum arm_builtins) fcode)
- return arm_expand_binop_builtin (d->icode, arglist, target);
+ return arm_expand_binop_builtin (d->icode, exp, target);
for (i = 0, d = bdesc_1arg; i < ARRAY_SIZE (bdesc_1arg); i++, d++)
if (d->code == (const enum arm_builtins) fcode)
- return arm_expand_unop_builtin (d->icode, arglist, target, 0);
+ return arm_expand_unop_builtin (d->icode, exp, target, 0);
/* @@@ Should really do something sensible here. */
return NULL_RTX;
{
/* If we can deduce the registers used from the function's
return value. This is more reliable that examining
- regs_ever_live[] because that will be set if the register is
+ df_regs_ever_live_p () because that will be set if the register is
ever used in the function, not just if the register is used
to hold a return value. */
\f
void
-thumb_final_prescan_insn (rtx insn)
+thumb1_final_prescan_insn (rtx insn)
{
if (flag_print_asm_name)
asm_fprintf (asm_out_file, "%@ 0x%04x\n",
If we need doubleword stack alignment this could affect the other
elimination offsets so we can't risk getting it wrong. */
- if (regs_ever_live [ARG_POINTER_REGNUM])
+ if (df_regs_ever_live_p (ARG_POINTER_REGNUM))
cfun->machine->arg_pointer_live = 1;
else if (!cfun->machine->arg_pointer_live)
return 0;
if (IS_NAKED (arm_current_func_type ()))
return "";
- live_regs_mask = thumb_compute_save_reg_mask ();
+ live_regs_mask = thumb1_compute_save_reg_mask ();
high_regs_pushed = bit_count (live_regs_mask & 0x0f00);
/* If we can deduce the registers used from the function's return value.
- This is more reliable that examining regs_ever_live[] because that
+ This is more reliable that examining df_regs_ever_live_p () because that
will be set if the register is ever used in the function, not just if
the register is used to hold a return value. */
size = arm_size_return_regs ();
}
-/* Like arm_compute_initial_elimination offset. Simpler because
- THUMB_HARD_FRAME_POINTER isn't actually the ABI specified frame pointer. */
+/* Like arm_compute_initial_elimination offset. Simpler because there
+ isn't an ABI specified frame pointer for Thumb. Instead, we set it
+ to point at the base of the local variables after static stack
+ space for a function has been allocated. */
HOST_WIDE_INT
thumb_compute_initial_elimination_offset (unsigned int from, unsigned int to)
case FRAME_POINTER_REGNUM:
return offsets->soft_frame - offsets->saved_args;
- case THUMB_HARD_FRAME_POINTER_REGNUM:
case ARM_HARD_FRAME_POINTER_REGNUM:
return offsets->saved_regs - offsets->saved_args;
+ case THUMB_HARD_FRAME_POINTER_REGNUM:
+ return offsets->locals_base - offsets->saved_args;
+
default:
gcc_unreachable ();
}
case STACK_POINTER_REGNUM:
return offsets->outgoing_args - offsets->soft_frame;
- case THUMB_HARD_FRAME_POINTER_REGNUM:
case ARM_HARD_FRAME_POINTER_REGNUM:
return offsets->saved_regs - offsets->soft_frame;
+ case THUMB_HARD_FRAME_POINTER_REGNUM:
+ return offsets->locals_base - offsets->soft_frame;
+
default:
gcc_unreachable ();
}
}
}
-
/* Generate the rest of a function's prologue. */
void
-thumb_expand_prologue (void)
+thumb1_expand_prologue (void)
{
rtx insn, dwarf;
return;
}
- live_regs_mask = thumb_compute_save_reg_mask ();
+ live_regs_mask = thumb1_compute_save_reg_mask ();
/* Load the pic register before setting the frame pointer,
so we can use r7 as a temporary work register. */
- if (flag_pic)
- arm_load_pic_register (thumb_find_work_register (live_regs_mask));
+ if (flag_pic && arm_pic_register != INVALID_REGNUM)
+ arm_load_pic_register (live_regs_mask);
- offsets = arm_get_frame_offsets ();
-
- if (frame_pointer_needed)
- {
- insn = emit_insn (gen_movsi (hard_frame_pointer_rtx,
- stack_pointer_rtx));
- RTX_FRAME_RELATED_P (insn) = 1;
- }
- else if (CALLER_INTERWORKING_SLOT_SIZE > 0)
+ if (!frame_pointer_needed && CALLER_INTERWORKING_SLOT_SIZE > 0)
emit_move_insn (gen_rtx_REG (Pmode, ARM_HARD_FRAME_POINTER_REGNUM),
stack_pointer_rtx);
+ offsets = arm_get_frame_offsets ();
amount = offsets->outgoing_args - offsets->saved_regs;
if (amount)
{
insn = emit_insn (gen_addsi3 (stack_pointer_rtx,
stack_pointer_rtx, reg));
RTX_FRAME_RELATED_P (insn) = 1;
- dwarf = gen_rtx_SET (SImode, stack_pointer_rtx,
+ dwarf = gen_rtx_SET (VOIDmode, stack_pointer_rtx,
plus_constant (stack_pointer_rtx,
-amount));
RTX_FRAME_RELATED_P (dwarf) = 1;
insn = emit_insn (gen_addsi3 (stack_pointer_rtx,
stack_pointer_rtx, reg));
RTX_FRAME_RELATED_P (insn) = 1;
- dwarf = gen_rtx_SET (SImode, stack_pointer_rtx,
+ dwarf = gen_rtx_SET (VOIDmode, stack_pointer_rtx,
plus_constant (stack_pointer_rtx,
-amount));
RTX_FRAME_RELATED_P (dwarf) = 1;
REG_NOTES (insn));
}
}
- /* If the frame pointer is needed, emit a special barrier that
- will prevent the scheduler from moving stores to the frame
- before the stack adjustment. */
- if (frame_pointer_needed)
- emit_insn (gen_stack_tie (stack_pointer_rtx,
- hard_frame_pointer_rtx));
}
- if (current_function_profile || !TARGET_SCHED_PROLOG)
+ if (frame_pointer_needed)
+ thumb_set_frame_pointer (offsets);
+
+ /* If we are profiling, make sure no instructions are scheduled before
+ the call to mcount. Similarly if the user has requested no
+ scheduling in the prolog. Similarly if we want non-call exceptions
+ using the EABI unwinder, to prevent faulting instructions from being
+ swapped with a stack adjustment. */
+ if (current_function_profile || !TARGET_SCHED_PROLOG
+ || (ARM_EABI_UNWIND_TABLES && flag_non_call_exceptions))
emit_insn (gen_blockage ());
cfun->machine->lr_save_eliminated = !thumb_force_lr_save ();
if (live_regs_mask & 0xff)
cfun->machine->lr_save_eliminated = 0;
-
- /* If the link register is being kept alive, with the return address in it,
- then make sure that it does not get reused by the ce2 pass. */
- if (cfun->machine->lr_save_eliminated)
- emit_insn (gen_prologue_use (gen_rtx_REG (SImode, LR_REGNUM)));
}
void
-thumb_expand_epilogue (void)
+thumb1_expand_epilogue (void)
{
HOST_WIDE_INT amount;
arm_stack_offsets *offsets;
amount = offsets->outgoing_args - offsets->saved_regs;
if (frame_pointer_needed)
- emit_insn (gen_movsi (stack_pointer_rtx, hard_frame_pointer_rtx));
- else if (amount)
+ {
+ emit_insn (gen_movsi (stack_pointer_rtx, hard_frame_pointer_rtx));
+ amount = offsets->locals_base - offsets->saved_regs;
+ }
+
+ gcc_assert (amount >= 0);
+ if (amount)
{
if (amount < 512)
emit_insn (gen_addsi3 (stack_pointer_rtx, stack_pointer_rtx,
/* Emit a clobber for each insn that will be restored in the epilogue,
so that flow2 will get register lifetimes correct. */
for (regno = 0; regno < 13; regno++)
- if (regs_ever_live[regno] && !call_used_regs[regno])
+ if (df_regs_ever_live_p (regno) && !call_used_regs[regno])
emit_insn (gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (SImode, regno)));
- if (! regs_ever_live[LR_REGNUM])
+ if (! df_regs_ever_live_p (LR_REGNUM))
emit_insn (gen_rtx_USE (VOIDmode, gen_rtx_REG (SImode, LR_REGNUM)));
}
static void
-thumb_output_function_prologue (FILE *f, HOST_WIDE_INT size ATTRIBUTE_UNUSED)
+thumb1_output_function_prologue (FILE *f, HOST_WIDE_INT size ATTRIBUTE_UNUSED)
{
unsigned long live_regs_mask = 0;
unsigned long l_mask;
}
/* Get the registers we are going to push. */
- live_regs_mask = thumb_compute_save_reg_mask ();
+ live_regs_mask = thumb1_compute_save_reg_mask ();
/* Extract a mask of the ones we can give to the Thumb's push instruction. */
l_mask = live_regs_mask & 0x40ff;
/* Then count how many other high registers will need to be pushed. */
switch (GET_CODE (addr))
{
case REG:
- operands[2] = gen_rtx_MEM (SImode,
- plus_constant (XEXP (operands[1], 0), 4));
+ operands[2] = adjust_address (operands[1], SImode, 4);
if (REGNO (operands[0]) == REGNO (addr))
{
case CONST:
/* Compute <address> + 4 for the high order load. */
- operands[2] = gen_rtx_MEM (SImode,
- plus_constant (XEXP (operands[1], 0), 4));
+ operands[2] = adjust_address (operands[1], SImode, 4);
output_asm_insn ("ldr\t%0, %1", operands);
output_asm_insn ("ldr\t%H0, %2", operands);
else
{
/* Compute <address> + 4 for the high order load. */
- operands[2] = gen_rtx_MEM (SImode,
- plus_constant (XEXP (operands[1], 0), 4));
+ operands[2] = adjust_address (operands[1], SImode, 4);
/* If the computed address is held in the low order register
then load the high order register first, otherwise always
case LABEL_REF:
/* With no registers to worry about we can just load the value
directly. */
- operands[2] = gen_rtx_MEM (SImode,
- plus_constant (XEXP (operands[1], 0), 4));
+ operands[2] = adjust_address (operands[1], SImode, 4);
output_asm_insn ("ldr\t%H0, %2", operands);
output_asm_insn ("ldr\t%0, %1", operands);
/* If we are in the normal text section we can use a single instance
per compilation unit. If we are doing function sections, then we need
an entry per section, since we can't rely on reachability. */
- if (in_text_section ())
+ if (in_section == text_section)
{
thumb_call_reg_needed = 1;
}
static void
+arm_file_start (void)
+{
+ int val;
+
+ if (TARGET_UNIFIED_ASM)
+ asm_fprintf (asm_out_file, "\t.syntax unified\n");
+
+ if (TARGET_BPABI)
+ {
+ const char *fpu_name;
+ if (arm_select[0].string)
+ asm_fprintf (asm_out_file, "\t.cpu %s\n", arm_select[0].string);
+ else if (arm_select[1].string)
+ asm_fprintf (asm_out_file, "\t.arch %s\n", arm_select[1].string);
+ else
+ asm_fprintf (asm_out_file, "\t.cpu %s\n",
+ all_cores[arm_default_cpu].name);
+
+ if (TARGET_SOFT_FLOAT)
+ {
+ if (TARGET_VFP)
+ fpu_name = "softvfp";
+ else
+ fpu_name = "softfpa";
+ }
+ else
+ {
+ int set_float_abi_attributes = 0;
+ switch (arm_fpu_arch)
+ {
+ case FPUTYPE_FPA:
+ fpu_name = "fpa";
+ break;
+ case FPUTYPE_FPA_EMU2:
+ fpu_name = "fpe2";
+ break;
+ case FPUTYPE_FPA_EMU3:
+ fpu_name = "fpe3";
+ break;
+ case FPUTYPE_MAVERICK:
+ fpu_name = "maverick";
+ break;
+ case FPUTYPE_VFP:
+ fpu_name = "vfp";
+ set_float_abi_attributes = 1;
+ break;
+ case FPUTYPE_VFP3:
+ fpu_name = "vfp3";
+ set_float_abi_attributes = 1;
+ break;
+ default:
+ abort();
+ }
+ if (set_float_abi_attributes)
+ {
+ if (TARGET_HARD_FLOAT)
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 27, 3\n");
+ if (TARGET_HARD_FLOAT_ABI)
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 28, 1\n");
+ }
+ }
+ asm_fprintf (asm_out_file, "\t.fpu %s\n", fpu_name);
+
+ /* Some of these attributes only apply when the corresponding features
+ are used. However we don't have any easy way of figuring this out.
+ Conservatively record the setting that would have been used. */
+
+ /* Tag_ABI_PCS_wchar_t. */
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 18, %d\n",
+ (int)WCHAR_TYPE_SIZE / BITS_PER_UNIT);
+
+ /* Tag_ABI_FP_rounding. */
+ if (flag_rounding_math)
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 19, 1\n");
+ if (!flag_unsafe_math_optimizations)
+ {
+ /* Tag_ABI_FP_denomal. */
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 20, 1\n");
+ /* Tag_ABI_FP_exceptions. */
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 21, 1\n");
+ }
+ /* Tag_ABI_FP_user_exceptions. */
+ if (flag_signaling_nans)
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 22, 1\n");
+ /* Tag_ABI_FP_number_model. */
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 23, %d\n",
+ flag_finite_math_only ? 1 : 3);
+
+ /* Tag_ABI_align8_needed. */
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 24, 1\n");
+ /* Tag_ABI_align8_preserved. */
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 25, 1\n");
+ /* Tag_ABI_enum_size. */
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 26, %d\n",
+ flag_short_enums ? 1 : 2);
+
+ /* Tag_ABI_optimization_goals. */
+ if (optimize_size)
+ val = 4;
+ else if (optimize >= 2)
+ val = 2;
+ else if (optimize)
+ val = 1;
+ else
+ val = 6;
+ asm_fprintf (asm_out_file, "\t.eabi_attribute 30, %d\n", val);
+ }
+ default_file_start();
+}
+
+static void
arm_file_end (void)
{
int regno;
+ if (NEED_INDICATE_EXEC_STACK)
+ /* Add .note.GNU-stack. */
+ file_end_indicate_exec_stack ();
+
if (! thumb_call_reg_needed)
return;
- text_section ();
+ switch_to_section (text_section);
asm_fprintf (asm_out_file, "\t.code 16\n");
ASM_OUTPUT_ALIGN (asm_out_file, 1);
int arm_text_section_count = 1;
-char *
-aof_text_section (void )
+/* A get_unnamed_section callback for switching to the text section. */
+
+static void
+aof_output_text_section_asm_op (const void *data ATTRIBUTE_UNUSED)
{
- static char buf[100];
- sprintf (buf, "\tAREA |C$$code%d|, CODE, READONLY",
+ fprintf (asm_out_file, "\tAREA |C$$code%d|, CODE, READONLY",
arm_text_section_count++);
if (flag_pic)
- strcat (buf, ", PIC, REENTRANT");
- return buf;
+ fprintf (asm_out_file, ", PIC, REENTRANT");
+ fprintf (asm_out_file, "\n");
}
static int arm_data_section_count = 1;
-char *
-aof_data_section (void)
+/* A get_unnamed_section callback for switching to the data section. */
+
+static void
+aof_output_data_section_asm_op (const void *data ATTRIBUTE_UNUSED)
+{
+ fprintf (asm_out_file, "\tAREA |C$$data%d|, DATA\n",
+ arm_data_section_count++);
+}
+
+/* Implement TARGET_ASM_INIT_SECTIONS.
+
+ AOF Assembler syntax is a nightmare when it comes to areas, since once
+ we change from one area to another, we can't go back again. Instead,
+ we must create a new area with the same attributes and add the new output
+ to that. Unfortunately, there is nothing we can do here to guarantee that
+ two areas with the same attributes will be linked adjacently in the
+ resulting executable, so we have to be careful not to do pc-relative
+ addressing across such boundaries. */
+
+static void
+aof_asm_init_sections (void)
+{
+ text_section = get_unnamed_section (SECTION_CODE,
+ aof_output_text_section_asm_op, NULL);
+ data_section = get_unnamed_section (SECTION_WRITE,
+ aof_output_data_section_asm_op, NULL);
+ readonly_data_section = text_section;
+}
+
+void
+zero_init_section (void)
{
- static char buf[100];
- sprintf (buf, "\tAREA |C$$data%d|, DATA", arm_data_section_count++);
- return buf;
+ static int zero_init_count = 1;
+
+ fprintf (asm_out_file, "\tAREA |C$$zidata%d|,NOINIT\n", zero_init_count++);
+ in_section = NULL;
}
/* The AOF assembler is religiously strict about declarations of
automagically. */
if (arm_main_function)
{
- text_section ();
+ switch_to_section (text_section);
fputs ("\tIMPORT __main\n", f);
fputs ("\tDCD __main\n", f);
}
fputs ("__f5\tFN\t5\n", asm_out_file);
fputs ("__f6\tFN\t6\n", asm_out_file);
fputs ("__f7\tFN\t7\n", asm_out_file);
- text_section ();
+ switch_to_section (text_section);
}
static void
SYMBOL_REF_FLAG (XEXP (rtl, 0)) = 1;
#endif
- /* If we are referencing a function that is weak then encode a long call
- flag in the function name, otherwise if the function is static or
- or known to be defined in this file then encode a short call flag. */
- if (first && DECL_P (decl))
- {
- if (TREE_CODE (decl) == FUNCTION_DECL && DECL_WEAK (decl))
- arm_encode_call_attribute (decl, LONG_CALL_FLAG_CHAR);
- else if (! TREE_PUBLIC (decl))
- arm_encode_call_attribute (decl, SHORT_CALL_FLAG_CHAR);
- }
+ default_encode_section_info (decl, rtl, first);
}
#endif /* !ARM_PE */
{
static int thunk_label = 0;
char label[256];
+ char labelpc[256];
int mi_delta = delta;
const char *const mi_op = mi_delta < 0 ? "sub" : "add";
int shift = 0;
? 1 : 0);
if (mi_delta < 0)
mi_delta = - mi_delta;
- if (TARGET_THUMB)
+ /* When generating 16-bit thumb code, thunks are entered in arm mode. */
+ if (TARGET_THUMB1)
{
int labelno = thunk_label++;
ASM_GENERATE_INTERNAL_LABEL (label, "LTHUMBFUNC", labelno);
fputs ("\tldr\tr12, ", file);
assemble_name (file, label);
fputc ('\n', file);
+ if (flag_pic)
+ {
+ /* If we are generating PIC, the ldr instruction below loads
+ "(target - 7) - .LTHUNKPCn" into r12. The pc reads as
+ the address of the add + 8, so we have:
+
+ r12 = (target - 7) - .LTHUNKPCn + (.LTHUNKPCn + 8)
+ = target + 1.
+
+ Note that we have "+ 1" because some versions of GNU ld
+ don't set the low bit of the result for R_ARM_REL32
+ relocations against thumb function symbols. */
+ ASM_GENERATE_INTERNAL_LABEL (labelpc, "LTHUNKPC", labelno);
+ assemble_name (file, labelpc);
+ fputs (":\n", file);
+ fputs ("\tadd\tr12, pc, r12\n", file);
+ }
}
+ /* TODO: Use movw/movt for large constants when available. */
while (mi_delta != 0)
{
if ((mi_delta & (3 << shift)) == 0)
shift += 8;
}
}
- if (TARGET_THUMB)
+ if (TARGET_THUMB1)
{
fprintf (file, "\tbx\tr12\n");
ASM_OUTPUT_ALIGN (file, 2);
assemble_name (file, label);
fputs (":\n", file);
- assemble_integer (XEXP (DECL_RTL (function), 0), 4, BITS_PER_WORD, 1);
+ if (flag_pic)
+ {
+ /* Output ".word .LTHUNKn-7-.LTHUNKPCn". */
+ rtx tem = XEXP (DECL_RTL (function), 0);
+ tem = gen_rtx_PLUS (GET_MODE (tem), tem, GEN_INT (-7));
+ tem = gen_rtx_MINUS (GET_MODE (tem),
+ tem,
+ gen_rtx_SYMBOL_REF (Pmode,
+ ggc_strdup (labelpc)));
+ assemble_integer (tem, 4, BITS_PER_WORD, 1);
+ }
+ else
+ /* Output ".word .LTHUNKn". */
+ assemble_integer (XEXP (DECL_RTL (function), 0), 4, BITS_PER_WORD, 1);
}
else
{
return "";
}
-static rtx
-arm_struct_value_rtx (tree fntype ATTRIBUTE_UNUSED,
- int incoming ATTRIBUTE_UNUSED)
-{
-#if 0
- /* FIXME: The ARM backend has special code to handle structure
- returns, and will reserve its own hidden first argument. So
- if this macro is enabled a *second* hidden argument will be
- reserved, which will break binary compatibility with old
- toolchains and also thunk handling. One day this should be
- fixed. */
- return 0;
-#else
- /* Register in which address to store a structure value
- is passed to a function. */
- return gen_rtx_REG (Pmode, ARG_REGISTER (1));
-#endif
-}
-
/* Worker function for TARGET_SETUP_INCOMING_VARARGS.
On the ARM, PRETEND_SIZE is set in order to have the prologue push the last
static bool
arm_default_short_enums (void)
{
- return TARGET_AAPCS_BASED;
+ return TARGET_AAPCS_BASED && arm_abi != ARM_ABI_AAPCS_LINUX;
}
DECL_VISIBILITY (decl) = VISIBILITY_DEFAULT;
DECL_VISIBILITY_SPECIFIED (decl) = 1;
}
-
+
static bool
arm_cxx_class_data_always_comdat (void)
{
addr = plus_constant (addr, delta);
}
- emit_move_insn (gen_rtx_MEM (Pmode, addr), source);
+ emit_move_insn (gen_frame_mem (Pmode, addr), source);
}
}
{
arm_stack_offsets *offsets;
HOST_WIDE_INT delta;
+ HOST_WIDE_INT limit;
int reg;
rtx addr;
unsigned long mask;
emit_insn (gen_rtx_USE (VOIDmode, source));
- mask = thumb_compute_save_reg_mask ();
+ mask = thumb1_compute_save_reg_mask ();
if (mask & (1 << LR_REGNUM))
{
offsets = arm_get_frame_offsets ();
+ limit = 1024;
/* Find the saved regs. */
if (frame_pointer_needed)
{
delta = offsets->soft_frame - offsets->saved_args;
reg = THUMB_HARD_FRAME_POINTER_REGNUM;
+ if (TARGET_THUMB1)
+ limit = 128;
}
else
{
reg = SP_REGNUM;
}
/* Allow for the stack frame. */
- if (TARGET_BACKTRACE)
+ if (TARGET_THUMB1 && TARGET_BACKTRACE)
delta -= 16;
/* The link register is always the first saved register. */
delta -= 4;
/* Construct the address. */
addr = gen_rtx_REG (SImode, reg);
- if ((reg != SP_REGNUM && delta >= 128)
- || delta >= 1024)
+ if (delta > limit)
{
emit_insn (gen_movsi (scratch, GEN_INT (delta)));
emit_insn (gen_addsi3 (scratch, scratch, stack_pointer_rtx));
else
addr = plus_constant (addr, delta);
- emit_move_insn (gen_rtx_MEM (Pmode, addr), source);
+ emit_move_insn (gen_frame_mem (Pmode, addr), source);
}
else
emit_move_insn (gen_rtx_REG (Pmode, LR_REGNUM), source);
if (IS_FPA_REGNUM (regno))
return (TARGET_AAPCS_BASED ? 96 : 16) + regno - FIRST_FPA_REGNUM;
+ /* FIXME: VFPv3 register numbering. */
if (IS_VFP_REGNUM (regno))
return 64 + regno - FIRST_VFP_REGNUM;
#ifdef TARGET_UNWIND_INFO
-/* Emit unwind directives for a store-multiple instruction. This should
- only ever be generated by the function prologue code, so we expect it
- to have a particular form. */
+/* Emit unwind directives for a store-multiple instruction or stack pointer
+ push during alignment.
+ These should only ever be generated by the function prologue code, so
+ expect them to have a particular form. */
static void
-arm_unwind_emit_stm (FILE * asm_out_file, rtx p)
+arm_unwind_emit_sequence (FILE * asm_out_file, rtx p)
{
int i;
HOST_WIDE_INT offset;
unsigned lastreg;
rtx e;
- /* First insn will adjust the stack pointer. */
e = XVECEXP (p, 0, 0);
+ if (GET_CODE (e) != SET)
+ abort ();
+
+ /* First insn will adjust the stack pointer. */
if (GET_CODE (e) != SET
|| GET_CODE (XEXP (e, 0)) != REG
|| REGNO (XEXP (e, 0)) != SP_REGNUM
offset -= 4;
}
reg_size = 4;
+ fprintf (asm_out_file, "\t.save {");
}
else if (IS_VFP_REGNUM (reg))
{
- /* FPA register saves use an additional word. */
- offset -= 4;
reg_size = 8;
+ fprintf (asm_out_file, "\t.vsave {");
}
else if (reg >= FIRST_FPA_REGNUM && reg <= LAST_FPA_REGNUM)
{
if (offset != nregs * reg_size)
abort ();
- fprintf (asm_out_file, "\t.save {");
-
offset = 0;
lastreg = 0;
/* The remaining insns will describe the stores. */
|| GET_CODE (XEXP (e, 0)) != MEM
|| GET_CODE (XEXP (e, 1)) != REG)
abort ();
-
+
reg = REGNO (XEXP (e, 1));
if (reg < lastreg)
abort ();
-
+
if (i != 1)
fprintf (asm_out_file, ", ");
/* We can't use %r for vfp because we need to use the
{
rtx e0;
rtx e1;
+ unsigned reg;
e0 = XEXP (p, 0);
e1 = XEXP (p, 1);
else if (REGNO (e0) == HARD_FRAME_POINTER_REGNUM)
{
HOST_WIDE_INT offset;
- unsigned reg;
-
+
if (GET_CODE (e1) == PLUS)
{
if (GET_CODE (XEXP (e1, 0)) != REG
/* Move from sp to reg. */
asm_fprintf (asm_out_file, "\t.movsp %r\n", REGNO (e0));
}
+ else if (GET_CODE (e1) == PLUS
+ && GET_CODE (XEXP (e1, 0)) == REG
+ && REGNO (XEXP (e1, 0)) == SP_REGNUM
+ && GET_CODE (XEXP (e1, 1)) == CONST_INT)
+ {
+ /* Set reg to offset from sp. */
+ asm_fprintf (asm_out_file, "\t.movsp %r, #%d\n",
+ REGNO (e0), (int)INTVAL(XEXP (e1, 1)));
+ }
+ else if (GET_CODE (e1) == UNSPEC && XINT (e1, 1) == UNSPEC_STACK_ALIGN)
+ {
+ /* Stack pointer save before alignment. */
+ reg = REGNO (e0);
+ asm_fprintf (asm_out_file, "\t.unwind_raw 0, 0x%x @ vsp = r%d\n",
+ reg + 0x90, reg);
+ }
else
abort ();
break;
case SEQUENCE:
/* Store multiple. */
- arm_unwind_emit_stm (asm_out_file, pat);
+ arm_unwind_emit_sequence (asm_out_file, pat);
break;
default:
#endif /* TARGET_UNWIND_INFO */
+/* Handle UNSPEC DWARF call frame instructions. These are needed for dynamic
+ stack alignment. */
+
+static void
+arm_dwarf_handle_frame_unspec (const char *label, rtx pattern, int index)
+{
+ rtx unspec = SET_SRC (pattern);
+ gcc_assert (GET_CODE (unspec) == UNSPEC);
+
+ switch (index)
+ {
+ case UNSPEC_STACK_ALIGN:
+ /* ??? We should set the CFA = (SP & ~7). At this point we haven't
+ put anything on the stack, so hopefully it won't matter.
+ CFA = SP will be correct after alignment. */
+ dwarf2out_reg_save_reg (label, stack_pointer_rtx,
+ SET_DEST (pattern));
+ break;
+ default:
+ gcc_unreachable ();
+ }
+}
+
+
/* Output unwind directives for the start/end of a function. */
void
else
fputs ("\t.fnend\n", f);
}
+
+static bool
+arm_emit_tls_decoration (FILE *fp, rtx x)
+{
+ enum tls_reloc reloc;
+ rtx val;
+
+ val = XVECEXP (x, 0, 0);
+ reloc = INTVAL (XVECEXP (x, 0, 1));
+
+ output_addr_const (fp, val);
+
+ switch (reloc)
+ {
+ case TLS_GD32:
+ fputs ("(tlsgd)", fp);
+ break;
+ case TLS_LDM32:
+ fputs ("(tlsldm)", fp);
+ break;
+ case TLS_LDO32:
+ fputs ("(tlsldo)", fp);
+ break;
+ case TLS_IE32:
+ fputs ("(gottpoff)", fp);
+ break;
+ case TLS_LE32:
+ fputs ("(tpoff)", fp);
+ break;
+ default:
+ gcc_unreachable ();
+ }
+
+ switch (reloc)
+ {
+ case TLS_GD32:
+ case TLS_LDM32:
+ case TLS_IE32:
+ fputs (" + (. - ", fp);
+ output_addr_const (fp, XVECEXP (x, 0, 2));
+ fputs (" - ", fp);
+ output_addr_const (fp, XVECEXP (x, 0, 3));
+ fputc (')', fp);
+ break;
+ default:
+ break;
+ }
+
+ return TRUE;
+}
+
+/* ARM implementation of TARGET_ASM_OUTPUT_DWARF_DTPREL. */
+
+static void
+arm_output_dwarf_dtprel (FILE *file, int size, rtx x)
+{
+ gcc_assert (size == 4);
+ fputs ("\t.word\t", file);
+ output_addr_const (file, x);
+ fputs ("(tlsldo)", file);
+}
+
+bool
+arm_output_addr_const_extra (FILE *fp, rtx x)
+{
+ if (GET_CODE (x) == UNSPEC && XINT (x, 1) == UNSPEC_TLS)
+ return arm_emit_tls_decoration (fp, x);
+ else if (GET_CODE (x) == UNSPEC && XINT (x, 1) == UNSPEC_PIC_LABEL)
+ {
+ char label[256];
+ int labelno = INTVAL (XVECEXP (x, 0, 0));
+
+ ASM_GENERATE_INTERNAL_LABEL (label, "LPIC", labelno);
+ assemble_name_raw (fp, label);
+
+ return TRUE;
+ }
+ else if (GET_CODE (x) == CONST_VECTOR)
+ return arm_emit_vector_const (fp, x);
+
+ return FALSE;
+}
+
+/* Output assembly for a shift instruction.
+ SET_FLAGS determines how the instruction modifies the condition codes.
+ 0 - Do not set condition codes.
+ 1 - Set condition codes.
+ 2 - Use smallest instruction. */
+const char *
+arm_output_shift(rtx * operands, int set_flags)
+{
+ char pattern[100];
+ static const char flag_chars[3] = {'?', '.', '!'};
+ const char *shift;
+ HOST_WIDE_INT val;
+ char c;
+
+ c = flag_chars[set_flags];
+ if (TARGET_UNIFIED_ASM)
+ {
+ shift = shift_op(operands[3], &val);
+ if (shift)
+ {
+ if (val != -1)
+ operands[2] = GEN_INT(val);
+ sprintf (pattern, "%s%%%c\t%%0, %%1, %%2", shift, c);
+ }
+ else
+ sprintf (pattern, "mov%%%c\t%%0, %%1", c);
+ }
+ else
+ sprintf (pattern, "mov%%%c\t%%0, %%1%%S3", c);
+ output_asm_insn (pattern, operands);
+ return "";
+}
+
+/* Output a Thumb-2 casesi instruction. */
+const char *
+thumb2_output_casesi (rtx *operands)
+{
+ rtx diff_vec = PATTERN (next_real_insn (operands[2]));
+
+ gcc_assert (GET_CODE (diff_vec) == ADDR_DIFF_VEC);
+
+ output_asm_insn ("cmp\t%0, %1", operands);
+ output_asm_insn ("bhi\t%l3", operands);
+ switch (GET_MODE(diff_vec))
+ {
+ case QImode:
+ return "tbb\t[%|pc, %0]";
+ case HImode:
+ return "tbh\t[%|pc, %0, lsl #1]";
+ case SImode:
+ if (flag_pic)
+ {
+ output_asm_insn ("adr\t%4, %l2", operands);
+ output_asm_insn ("ldr\t%5, [%4, %0, lsl #2]", operands);
+ output_asm_insn ("add\t%4, %4, %5", operands);
+ return "bx\t%4";
+ }
+ else
+ {
+ output_asm_insn ("adr\t%4, %l2", operands);
+ return "ldr\t%|pc, [%4, %0, lsl #2]";
+ }
+ default:
+ gcc_unreachable ();
+ }
+}
+
+#include "gt-arm.h"
OSDN Git Service
