| /* Target machine subroutines for Altera Nios II. |
| Copyright (C) 2012-2021 Free Software Foundation, Inc. |
| Contributed by Jonah Graham (jgraham@altera.com), |
| Will Reece (wreece@altera.com), and Jeff DaSilva (jdasilva@altera.com). |
| Contributed by Mentor Graphics, Inc. |
| |
| This file is part of GCC. |
| |
| GCC is free software; you can redistribute it and/or modify it |
| under the terms of the GNU General Public License as published |
| by the Free Software Foundation; either version 3, or (at your |
| option) any later version. |
| |
| GCC is distributed in the hope that it will be useful, but WITHOUT |
| ANY WARRANTY; without even the implied warranty of MERCHANTABILITY |
| or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public |
| License for more details. |
| |
| You should have received a copy of the GNU General Public License |
| along with GCC; see the file COPYING3. If not see |
| <http://www.gnu.org/licenses/>. */ |
| |
| #define IN_TARGET_CODE 1 |
| |
| #include "config.h" |
| #include "system.h" |
| #include "coretypes.h" |
| #include "backend.h" |
| #include "target.h" |
| #include "rtl.h" |
| #include "tree.h" |
| #include "stringpool.h" |
| #include "attribs.h" |
| #include "df.h" |
| #include "memmodel.h" |
| #include "tm_p.h" |
| #include "optabs.h" |
| #include "regs.h" |
| #include "emit-rtl.h" |
| #include "recog.h" |
| #include "diagnostic-core.h" |
| #include "output.h" |
| #include "insn-attr.h" |
| #include "flags.h" |
| #include "explow.h" |
| #include "calls.h" |
| #include "varasm.h" |
| #include "expr.h" |
| #include "toplev.h" |
| #include "langhooks.h" |
| #include "stor-layout.h" |
| #include "builtins.h" |
| #include "tree-pass.h" |
| #include "xregex.h" |
| |
| /* This file should be included last. */ |
| #include "target-def.h" |
| |
| /* Forward function declarations. */ |
| static bool nios2_symbolic_constant_p (rtx); |
| static bool prologue_saved_reg_p (unsigned); |
| static void nios2_load_pic_register (void); |
| static void nios2_register_custom_code (unsigned int, enum nios2_ccs_code, int); |
| static const char *nios2_unspec_reloc_name (int); |
| static void nios2_register_builtin_fndecl (unsigned, tree); |
| static rtx nios2_ldst_parallel (bool, bool, bool, rtx, int, |
| unsigned HOST_WIDE_INT, bool); |
| static int nios2_address_cost (rtx, machine_mode, addr_space_t, bool); |
| |
| /* Threshold for data being put into the small data/bss area, instead |
| of the normal data area (references to the small data/bss area take |
| 1 instruction, and use the global pointer, references to the normal |
| data area takes 2 instructions). */ |
| unsigned HOST_WIDE_INT nios2_section_threshold = NIOS2_DEFAULT_GVALUE; |
| |
| struct GTY (()) machine_function |
| { |
| /* Current frame information, to be filled in by nios2_compute_frame_layout |
| with register save masks, and offsets for the current function. */ |
| |
| /* Mask of registers to save. */ |
| unsigned int save_mask; |
| /* Number of bytes that the entire frame takes up. */ |
| int total_size; |
| /* Number of bytes that variables take up. */ |
| int var_size; |
| /* Number of bytes that outgoing arguments take up. */ |
| int args_size; |
| /* Number of bytes needed to store registers in frame. */ |
| int save_reg_size; |
| /* Number of bytes used to store callee-saved registers. */ |
| int callee_save_reg_size; |
| /* Offset from new stack pointer to store registers. */ |
| int save_regs_offset; |
| /* Offset from save_regs_offset to store frame pointer register. */ |
| int fp_save_offset; |
| /* != 0 if function has a variable argument list. */ |
| int uses_anonymous_args; |
| /* != 0 if frame layout already calculated. */ |
| int initialized; |
| }; |
| |
| /* State to track the assignment of custom codes to FPU/custom builtins. */ |
| static enum nios2_ccs_code custom_code_status[256]; |
| static int custom_code_index[256]; |
| /* Set to true if any conflicts (re-use of a code between 0-255) are found. */ |
| static bool custom_code_conflict = false; |
| |
| /* State for command-line options. */ |
| regex_t nios2_gprel_sec_regex; |
| regex_t nios2_r0rel_sec_regex; |
| |
| |
| /* Definition of builtin function types for nios2. */ |
| |
| #define N2_FTYPES \ |
| N2_FTYPE(1, (SF)) \ |
| N2_FTYPE(1, (VOID)) \ |
| N2_FTYPE(2, (DF, DF)) \ |
| N2_FTYPE(3, (DF, DF, DF)) \ |
| N2_FTYPE(2, (DF, SF)) \ |
| N2_FTYPE(2, (DF, SI)) \ |
| N2_FTYPE(2, (DF, UI)) \ |
| N2_FTYPE(2, (SF, DF)) \ |
| N2_FTYPE(2, (SF, SF)) \ |
| N2_FTYPE(3, (SF, SF, SF)) \ |
| N2_FTYPE(2, (SF, SI)) \ |
| N2_FTYPE(2, (SF, UI)) \ |
| N2_FTYPE(2, (SI, CVPTR)) \ |
| N2_FTYPE(2, (SI, DF)) \ |
| N2_FTYPE(3, (SI, DF, DF)) \ |
| N2_FTYPE(2, (SI, SF)) \ |
| N2_FTYPE(3, (SI, SF, SF)) \ |
| N2_FTYPE(2, (SI, SI)) \ |
| N2_FTYPE(3, (SI, SI, SI)) \ |
| N2_FTYPE(3, (SI, VPTR, SI)) \ |
| N2_FTYPE(2, (UI, CVPTR)) \ |
| N2_FTYPE(2, (UI, DF)) \ |
| N2_FTYPE(2, (UI, SF)) \ |
| N2_FTYPE(2, (VOID, DF)) \ |
| N2_FTYPE(2, (VOID, SF)) \ |
| N2_FTYPE(2, (VOID, SI)) \ |
| N2_FTYPE(3, (VOID, SI, SI)) \ |
| N2_FTYPE(2, (VOID, VPTR)) \ |
| N2_FTYPE(3, (VOID, VPTR, SI)) |
| |
| #define N2_FTYPE_OP1(R) N2_FTYPE_ ## R ## _VOID |
| #define N2_FTYPE_OP2(R, A1) N2_FTYPE_ ## R ## _ ## A1 |
| #define N2_FTYPE_OP3(R, A1, A2) N2_FTYPE_ ## R ## _ ## A1 ## _ ## A2 |
| |
| /* Expand ftcode enumeration. */ |
| enum nios2_ftcode { |
| #define N2_FTYPE(N,ARGS) N2_FTYPE_OP ## N ARGS, |
| N2_FTYPES |
| #undef N2_FTYPE |
| N2_FTYPE_MAX |
| }; |
| |
| /* Return the tree function type, based on the ftcode. */ |
| static tree |
| nios2_ftype (enum nios2_ftcode ftcode) |
| { |
| static tree types[(int) N2_FTYPE_MAX]; |
| |
| tree N2_TYPE_SF = float_type_node; |
| tree N2_TYPE_DF = double_type_node; |
| tree N2_TYPE_SI = integer_type_node; |
| tree N2_TYPE_UI = unsigned_type_node; |
| tree N2_TYPE_VOID = void_type_node; |
| |
| static const_tree N2_TYPE_CVPTR, N2_TYPE_VPTR; |
| if (!N2_TYPE_CVPTR) |
| { |
| /* const volatile void *. */ |
| N2_TYPE_CVPTR |
| = build_pointer_type (build_qualified_type (void_type_node, |
| (TYPE_QUAL_CONST |
| | TYPE_QUAL_VOLATILE))); |
| /* volatile void *. */ |
| N2_TYPE_VPTR |
| = build_pointer_type (build_qualified_type (void_type_node, |
| TYPE_QUAL_VOLATILE)); |
| } |
| if (types[(int) ftcode] == NULL_TREE) |
| switch (ftcode) |
| { |
| #define N2_FTYPE_ARGS1(R) N2_TYPE_ ## R |
| #define N2_FTYPE_ARGS2(R,A1) N2_TYPE_ ## R, N2_TYPE_ ## A1 |
| #define N2_FTYPE_ARGS3(R,A1,A2) N2_TYPE_ ## R, N2_TYPE_ ## A1, N2_TYPE_ ## A2 |
| #define N2_FTYPE(N,ARGS) \ |
| case N2_FTYPE_OP ## N ARGS: \ |
| types[(int) ftcode] \ |
| = build_function_type_list (N2_FTYPE_ARGS ## N ARGS, NULL_TREE); \ |
| break; |
| N2_FTYPES |
| #undef N2_FTYPE |
| default: gcc_unreachable (); |
| } |
| return types[(int) ftcode]; |
| } |
| |
| |
| /* Definition of FPU instruction descriptions. */ |
| |
| struct nios2_fpu_insn_info |
| { |
| const char *name; |
| int num_operands, *optvar; |
| int opt, no_opt; |
| #define N2F_DF 0x1 |
| #define N2F_DFREQ 0x2 |
| #define N2F_UNSAFE 0x4 |
| #define N2F_FINITE 0x8 |
| #define N2F_NO_ERRNO 0x10 |
| unsigned int flags; |
| enum insn_code icode; |
| enum nios2_ftcode ftcode; |
| }; |
| |
| /* Base macro for defining FPU instructions. */ |
| #define N2FPU_INSN_DEF_BASE(insn, nop, flags, icode, args) \ |
| { #insn, nop, &nios2_custom_ ## insn, OPT_mcustom_##insn##_, \ |
| OPT_mno_custom_##insn, flags, CODE_FOR_ ## icode, \ |
| N2_FTYPE_OP ## nop args } |
| |
| /* Arithmetic and math functions; 2 or 3 operand FP operations. */ |
| #define N2FPU_OP2(mode) (mode, mode) |
| #define N2FPU_OP3(mode) (mode, mode, mode) |
| #define N2FPU_INSN_DEF(code, icode, nop, flags, m, M) \ |
| N2FPU_INSN_DEF_BASE (f ## code ## m, nop, flags, \ |
| icode ## m ## f ## nop, N2FPU_OP ## nop (M ## F)) |
| #define N2FPU_INSN_SF(code, nop, flags) \ |
| N2FPU_INSN_DEF (code, code, nop, flags, s, S) |
| #define N2FPU_INSN_DF(code, nop, flags) \ |
| N2FPU_INSN_DEF (code, code, nop, flags | N2F_DF, d, D) |
| |
| /* Compare instructions, 3 operand FP operation with a SI result. */ |
| #define N2FPU_CMP_DEF(code, flags, m, M) \ |
| N2FPU_INSN_DEF_BASE (fcmp ## code ## m, 3, flags, \ |
| nios2_s ## code ## m ## f, (SI, M ## F, M ## F)) |
| #define N2FPU_CMP_SF(code) N2FPU_CMP_DEF (code, 0, s, S) |
| #define N2FPU_CMP_DF(code) N2FPU_CMP_DEF (code, N2F_DF, d, D) |
| |
| /* The order of definition needs to be maintained consistent with |
| enum n2fpu_code in nios2-opts.h. */ |
| struct nios2_fpu_insn_info nios2_fpu_insn[] = |
| { |
| /* Single precision instructions. */ |
| N2FPU_INSN_SF (add, 3, 0), |
| N2FPU_INSN_SF (sub, 3, 0), |
| N2FPU_INSN_SF (mul, 3, 0), |
| N2FPU_INSN_SF (div, 3, 0), |
| /* Due to textual difference between min/max and smin/smax. */ |
| N2FPU_INSN_DEF (min, smin, 3, N2F_FINITE, s, S), |
| N2FPU_INSN_DEF (max, smax, 3, N2F_FINITE, s, S), |
| N2FPU_INSN_SF (neg, 2, 0), |
| N2FPU_INSN_SF (abs, 2, 0), |
| N2FPU_INSN_SF (sqrt, 2, 0), |
| N2FPU_INSN_SF (sin, 2, N2F_UNSAFE), |
| N2FPU_INSN_SF (cos, 2, N2F_UNSAFE), |
| N2FPU_INSN_SF (tan, 2, N2F_UNSAFE), |
| N2FPU_INSN_SF (atan, 2, N2F_UNSAFE), |
| N2FPU_INSN_SF (exp, 2, N2F_UNSAFE), |
| N2FPU_INSN_SF (log, 2, N2F_UNSAFE), |
| /* Single precision compares. */ |
| N2FPU_CMP_SF (eq), N2FPU_CMP_SF (ne), |
| N2FPU_CMP_SF (lt), N2FPU_CMP_SF (le), |
| N2FPU_CMP_SF (gt), N2FPU_CMP_SF (ge), |
| |
| /* Double precision instructions. */ |
| N2FPU_INSN_DF (add, 3, 0), |
| N2FPU_INSN_DF (sub, 3, 0), |
| N2FPU_INSN_DF (mul, 3, 0), |
| N2FPU_INSN_DF (div, 3, 0), |
| /* Due to textual difference between min/max and smin/smax. */ |
| N2FPU_INSN_DEF (min, smin, 3, N2F_FINITE, d, D), |
| N2FPU_INSN_DEF (max, smax, 3, N2F_FINITE, d, D), |
| N2FPU_INSN_DF (neg, 2, 0), |
| N2FPU_INSN_DF (abs, 2, 0), |
| N2FPU_INSN_DF (sqrt, 2, 0), |
| N2FPU_INSN_DF (sin, 2, N2F_UNSAFE), |
| N2FPU_INSN_DF (cos, 2, N2F_UNSAFE), |
| N2FPU_INSN_DF (tan, 2, N2F_UNSAFE), |
| N2FPU_INSN_DF (atan, 2, N2F_UNSAFE), |
| N2FPU_INSN_DF (exp, 2, N2F_UNSAFE), |
| N2FPU_INSN_DF (log, 2, N2F_UNSAFE), |
| /* Double precision compares. */ |
| N2FPU_CMP_DF (eq), N2FPU_CMP_DF (ne), |
| N2FPU_CMP_DF (lt), N2FPU_CMP_DF (le), |
| N2FPU_CMP_DF (gt), N2FPU_CMP_DF (ge), |
| |
| /* Conversion instructions. */ |
| N2FPU_INSN_DEF_BASE (floatis, 2, 0, floatsisf2, (SF, SI)), |
| N2FPU_INSN_DEF_BASE (floatus, 2, 0, floatunssisf2, (SF, UI)), |
| N2FPU_INSN_DEF_BASE (floatid, 2, 0, floatsidf2, (DF, SI)), |
| N2FPU_INSN_DEF_BASE (floatud, 2, 0, floatunssidf2, (DF, UI)), |
| N2FPU_INSN_DEF_BASE (round, 2, N2F_NO_ERRNO, lroundsfsi2, (SI, SF)), |
| N2FPU_INSN_DEF_BASE (fixsi, 2, 0, fix_truncsfsi2, (SI, SF)), |
| N2FPU_INSN_DEF_BASE (fixsu, 2, 0, fixuns_truncsfsi2, (UI, SF)), |
| N2FPU_INSN_DEF_BASE (fixdi, 2, 0, fix_truncdfsi2, (SI, DF)), |
| N2FPU_INSN_DEF_BASE (fixdu, 2, 0, fixuns_truncdfsi2, (UI, DF)), |
| N2FPU_INSN_DEF_BASE (fextsd, 2, 0, extendsfdf2, (DF, SF)), |
| N2FPU_INSN_DEF_BASE (ftruncds, 2, 0, truncdfsf2, (SF, DF)), |
| |
| /* X, Y access instructions. */ |
| N2FPU_INSN_DEF_BASE (fwrx, 2, N2F_DFREQ, nios2_fwrx, (VOID, DF)), |
| N2FPU_INSN_DEF_BASE (fwry, 2, N2F_DFREQ, nios2_fwry, (VOID, SF)), |
| N2FPU_INSN_DEF_BASE (frdxlo, 1, N2F_DFREQ, nios2_frdxlo, (SF)), |
| N2FPU_INSN_DEF_BASE (frdxhi, 1, N2F_DFREQ, nios2_frdxhi, (SF)), |
| N2FPU_INSN_DEF_BASE (frdy, 1, N2F_DFREQ, nios2_frdy, (SF)) |
| }; |
| |
| /* Some macros for ease of access. */ |
| #define N2FPU(code) nios2_fpu_insn[(int) code] |
| #define N2FPU_ENABLED_P(code) (N2FPU_N(code) >= 0) |
| #define N2FPU_N(code) (*N2FPU(code).optvar) |
| #define N2FPU_NAME(code) (N2FPU(code).name) |
| #define N2FPU_ICODE(code) (N2FPU(code).icode) |
| #define N2FPU_FTCODE(code) (N2FPU(code).ftcode) |
| #define N2FPU_FINITE_P(code) (N2FPU(code).flags & N2F_FINITE) |
| #define N2FPU_UNSAFE_P(code) (N2FPU(code).flags & N2F_UNSAFE) |
| #define N2FPU_NO_ERRNO_P(code) (N2FPU(code).flags & N2F_NO_ERRNO) |
| #define N2FPU_DOUBLE_P(code) (N2FPU(code).flags & N2F_DF) |
| #define N2FPU_DOUBLE_REQUIRED_P(code) (N2FPU(code).flags & N2F_DFREQ) |
| |
| /* Same as above, but for cases where using only the op part is shorter. */ |
| #define N2FPU_OP(op) N2FPU(n2fpu_ ## op) |
| #define N2FPU_OP_NAME(op) N2FPU_NAME(n2fpu_ ## op) |
| #define N2FPU_OP_ENABLED_P(op) N2FPU_ENABLED_P(n2fpu_ ## op) |
| |
| /* Export the FPU insn enabled predicate to nios2.md. */ |
| bool |
| nios2_fpu_insn_enabled (enum n2fpu_code code) |
| { |
| return N2FPU_ENABLED_P (code); |
| } |
| |
| /* Return true if COND comparison for mode MODE is enabled under current |
| settings. */ |
| |
| static bool |
| nios2_fpu_compare_enabled (enum rtx_code cond, machine_mode mode) |
| { |
| if (mode == SFmode) |
| switch (cond) |
| { |
| case EQ: return N2FPU_OP_ENABLED_P (fcmpeqs); |
| case NE: return N2FPU_OP_ENABLED_P (fcmpnes); |
| case GT: return N2FPU_OP_ENABLED_P (fcmpgts); |
| case GE: return N2FPU_OP_ENABLED_P (fcmpges); |
| case LT: return N2FPU_OP_ENABLED_P (fcmplts); |
| case LE: return N2FPU_OP_ENABLED_P (fcmples); |
| default: break; |
| } |
| else if (mode == DFmode) |
| switch (cond) |
| { |
| case EQ: return N2FPU_OP_ENABLED_P (fcmpeqd); |
| case NE: return N2FPU_OP_ENABLED_P (fcmpned); |
| case GT: return N2FPU_OP_ENABLED_P (fcmpgtd); |
| case GE: return N2FPU_OP_ENABLED_P (fcmpged); |
| case LT: return N2FPU_OP_ENABLED_P (fcmpltd); |
| case LE: return N2FPU_OP_ENABLED_P (fcmpled); |
| default: break; |
| } |
| return false; |
| } |
| |
| /* Stack layout and calling conventions. */ |
| |
| #define NIOS2_STACK_ALIGN(LOC) \ |
| (((LOC) + ((PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT) - 1)) \ |
| & ~((PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT) - 1)) |
| |
| /* Return the bytes needed to compute the frame pointer from the current |
| stack pointer. */ |
| static int |
| nios2_compute_frame_layout (void) |
| { |
| unsigned int regno; |
| unsigned int save_mask = 0; |
| int total_size; |
| int var_size; |
| int out_args_size; |
| int save_reg_size; |
| int callee_save_reg_size; |
| |
| if (cfun->machine->initialized) |
| return cfun->machine->total_size; |
| |
| /* Calculate space needed for gp registers. */ |
| save_reg_size = 0; |
| for (regno = 0; regno <= LAST_GP_REG; regno++) |
| if (prologue_saved_reg_p (regno)) |
| { |
| save_mask |= 1 << regno; |
| save_reg_size += 4; |
| } |
| |
| /* If we are saving any callee-save register, then assume |
| push.n/pop.n should be used. Make sure RA is saved, and |
| contiguous registers starting from r16-- are all saved. */ |
| if (TARGET_HAS_CDX && save_reg_size != 0) |
| { |
| if ((save_mask & (1 << RA_REGNO)) == 0) |
| { |
| save_mask |= 1 << RA_REGNO; |
| save_reg_size += 4; |
| } |
| |
| for (regno = 23; regno >= 16; regno--) |
| if ((save_mask & (1 << regno)) != 0) |
| { |
| /* Starting from highest numbered callee-saved |
| register that is used, make sure all regs down |
| to r16 is saved, to maintain contiguous range |
| for push.n/pop.n. */ |
| unsigned int i; |
| for (i = regno - 1; i >= 16; i--) |
| if ((save_mask & (1 << i)) == 0) |
| { |
| save_mask |= 1 << i; |
| save_reg_size += 4; |
| } |
| break; |
| } |
| } |
| |
| callee_save_reg_size = save_reg_size; |
| |
| /* If we call eh_return, we need to save the EH data registers. */ |
| if (crtl->calls_eh_return) |
| { |
| unsigned i; |
| unsigned r; |
| |
| for (i = 0; (r = EH_RETURN_DATA_REGNO (i)) != INVALID_REGNUM; i++) |
| if (!(save_mask & (1 << r))) |
| { |
| save_mask |= 1 << r; |
| save_reg_size += 4; |
| } |
| } |
| |
| cfun->machine->fp_save_offset = 0; |
| if (save_mask & (1 << HARD_FRAME_POINTER_REGNUM)) |
| { |
| int fp_save_offset = 0; |
| for (regno = 0; regno < HARD_FRAME_POINTER_REGNUM; regno++) |
| if (save_mask & (1 << regno)) |
| fp_save_offset += 4; |
| |
| cfun->machine->fp_save_offset = fp_save_offset; |
| } |
| |
| var_size = NIOS2_STACK_ALIGN (get_frame_size ()); |
| out_args_size = NIOS2_STACK_ALIGN (crtl->outgoing_args_size); |
| total_size = var_size + out_args_size; |
| |
| save_reg_size = NIOS2_STACK_ALIGN (save_reg_size); |
| total_size += save_reg_size; |
| total_size += NIOS2_STACK_ALIGN (crtl->args.pretend_args_size); |
| |
| /* Save other computed information. */ |
| cfun->machine->save_mask = save_mask; |
| cfun->machine->total_size = total_size; |
| cfun->machine->var_size = var_size; |
| cfun->machine->args_size = out_args_size; |
| cfun->machine->save_reg_size = save_reg_size; |
| cfun->machine->callee_save_reg_size = callee_save_reg_size; |
| cfun->machine->initialized = reload_completed; |
| cfun->machine->save_regs_offset = out_args_size + var_size; |
| |
| return total_size; |
| } |
| |
| /* Generate save/restore of register REGNO at SP + OFFSET. Used by the |
| prologue/epilogue expand routines. */ |
| static void |
| save_reg (int regno, unsigned offset) |
| { |
| rtx reg = gen_rtx_REG (SImode, regno); |
| rtx addr = plus_constant (Pmode, stack_pointer_rtx, offset, false); |
| rtx_insn *insn = emit_move_insn (gen_frame_mem (Pmode, addr), reg); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| } |
| |
| static void |
| restore_reg (int regno, unsigned offset) |
| { |
| rtx reg = gen_rtx_REG (SImode, regno); |
| rtx addr = plus_constant (Pmode, stack_pointer_rtx, offset, false); |
| rtx_insn *insn = emit_move_insn (reg, gen_frame_mem (Pmode, addr)); |
| /* Tag epilogue unwind note. */ |
| add_reg_note (insn, REG_CFA_RESTORE, reg); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| } |
| |
| /* This routine tests for the base register update SET in load/store |
| multiple RTL insns, used in pop_operation_p and ldstwm_operation_p. */ |
| static bool |
| base_reg_adjustment_p (rtx set, rtx *base_reg, rtx *offset) |
| { |
| if (GET_CODE (set) == SET |
| && REG_P (SET_DEST (set)) |
| && GET_CODE (SET_SRC (set)) == PLUS |
| && REG_P (XEXP (SET_SRC (set), 0)) |
| && rtx_equal_p (SET_DEST (set), XEXP (SET_SRC (set), 0)) |
| && CONST_INT_P (XEXP (SET_SRC (set), 1))) |
| { |
| *base_reg = XEXP (SET_SRC (set), 0); |
| *offset = XEXP (SET_SRC (set), 1); |
| return true; |
| } |
| return false; |
| } |
| |
| /* Does the CFA note work for push/pop prologue/epilogue instructions. */ |
| static void |
| nios2_create_cfa_notes (rtx_insn *insn, bool epilogue_p) |
| { |
| int i = 0; |
| rtx base_reg, offset, elt, pat = PATTERN (insn); |
| if (epilogue_p) |
| { |
| elt = XVECEXP (pat, 0, 0); |
| if (GET_CODE (elt) == RETURN) |
| i++; |
| elt = XVECEXP (pat, 0, i); |
| if (base_reg_adjustment_p (elt, &base_reg, &offset)) |
| { |
| add_reg_note (insn, REG_CFA_ADJUST_CFA, copy_rtx (elt)); |
| i++; |
| } |
| for (; i < XVECLEN (pat, 0); i++) |
| { |
| elt = SET_DEST (XVECEXP (pat, 0, i)); |
| gcc_assert (REG_P (elt)); |
| add_reg_note (insn, REG_CFA_RESTORE, elt); |
| } |
| } |
| else |
| { |
| /* Tag each of the prologue sets. */ |
| for (i = 0; i < XVECLEN (pat, 0); i++) |
| RTX_FRAME_RELATED_P (XVECEXP (pat, 0, i)) = 1; |
| } |
| } |
| |
| /* Temp regno used inside prologue/epilogue. */ |
| #define TEMP_REG_NUM 8 |
| |
| /* Emit conditional trap for checking stack limit. SIZE is the number of |
| additional bytes required. |
| |
| GDB prologue analysis depends on this generating a direct comparison |
| to the SP register, so the adjustment to add SIZE needs to be done on |
| the other operand to the comparison. Use TEMP_REG_NUM as a temporary, |
| if necessary. */ |
| static void |
| nios2_emit_stack_limit_check (int size) |
| { |
| rtx sum = NULL_RTX; |
| |
| if (GET_CODE (stack_limit_rtx) == SYMBOL_REF) |
| { |
| /* This generates a %hiadj/%lo pair with the constant size |
| add handled by the relocations. */ |
| sum = gen_rtx_REG (Pmode, TEMP_REG_NUM); |
| emit_move_insn (sum, plus_constant (Pmode, stack_limit_rtx, size)); |
| } |
| else if (!REG_P (stack_limit_rtx)) |
| sorry ("Unknown form for stack limit expression"); |
| else if (size == 0) |
| sum = stack_limit_rtx; |
| else if (SMALL_INT (size)) |
| { |
| sum = gen_rtx_REG (Pmode, TEMP_REG_NUM); |
| emit_move_insn (sum, plus_constant (Pmode, stack_limit_rtx, size)); |
| } |
| else |
| { |
| sum = gen_rtx_REG (Pmode, TEMP_REG_NUM); |
| emit_move_insn (sum, gen_int_mode (size, Pmode)); |
| emit_insn (gen_add2_insn (sum, stack_limit_rtx)); |
| } |
| |
| emit_insn (gen_ctrapsi4 (gen_rtx_LTU (VOIDmode, stack_pointer_rtx, sum), |
| stack_pointer_rtx, sum, GEN_INT (3))); |
| } |
| |
| static rtx_insn * |
| nios2_emit_add_constant (rtx reg, HOST_WIDE_INT immed) |
| { |
| rtx_insn *insn; |
| if (SMALL_INT (immed)) |
| insn = emit_insn (gen_add2_insn (reg, gen_int_mode (immed, Pmode))); |
| else |
| { |
| rtx tmp = gen_rtx_REG (Pmode, TEMP_REG_NUM); |
| emit_move_insn (tmp, gen_int_mode (immed, Pmode)); |
| insn = emit_insn (gen_add2_insn (reg, tmp)); |
| } |
| return insn; |
| } |
| |
| static rtx_insn * |
| nios2_adjust_stack (int sp_adjust, bool epilogue_p) |
| { |
| enum reg_note note_kind = REG_NOTE_MAX; |
| rtx_insn *insn = NULL; |
| if (sp_adjust) |
| { |
| if (SMALL_INT (sp_adjust)) |
| insn = emit_insn (gen_add2_insn (stack_pointer_rtx, |
| gen_int_mode (sp_adjust, Pmode))); |
| else |
| { |
| rtx tmp = gen_rtx_REG (Pmode, TEMP_REG_NUM); |
| emit_move_insn (tmp, gen_int_mode (sp_adjust, Pmode)); |
| insn = emit_insn (gen_add2_insn (stack_pointer_rtx, tmp)); |
| /* Attach a note indicating what happened. */ |
| if (!epilogue_p) |
| note_kind = REG_FRAME_RELATED_EXPR; |
| } |
| if (epilogue_p) |
| note_kind = REG_CFA_ADJUST_CFA; |
| if (note_kind != REG_NOTE_MAX) |
| { |
| rtx cfa_adj = gen_rtx_SET (stack_pointer_rtx, |
| plus_constant (Pmode, stack_pointer_rtx, |
| sp_adjust)); |
| add_reg_note (insn, note_kind, cfa_adj); |
| } |
| RTX_FRAME_RELATED_P (insn) = 1; |
| } |
| return insn; |
| } |
| |
| void |
| nios2_expand_prologue (void) |
| { |
| unsigned int regno; |
| int total_frame_size, save_offset; |
| int sp_offset; /* offset from base_reg to final stack value. */ |
| int save_regs_base; /* offset from base_reg to register save area. */ |
| rtx_insn *insn; |
| |
| total_frame_size = nios2_compute_frame_layout (); |
| |
| if (flag_stack_usage_info) |
| current_function_static_stack_size = total_frame_size; |
| |
| /* When R2 CDX push.n/stwm is available, arrange for stack frame to be built |
| using them. */ |
| if (TARGET_HAS_CDX |
| && (cfun->machine->save_reg_size != 0 |
| || cfun->machine->uses_anonymous_args)) |
| { |
| unsigned int regmask = cfun->machine->save_mask; |
| unsigned int callee_save_regs = regmask & 0xffff0000; |
| unsigned int caller_save_regs = regmask & 0x0000ffff; |
| int push_immed = 0; |
| int pretend_args_size = NIOS2_STACK_ALIGN (crtl->args.pretend_args_size); |
| rtx stack_mem = |
| gen_frame_mem (SImode, plus_constant (Pmode, stack_pointer_rtx, -4)); |
| |
| /* Check that there is room for the entire stack frame before doing |
| any SP adjustments or pushes. */ |
| if (crtl->limit_stack) |
| nios2_emit_stack_limit_check (total_frame_size); |
| |
| if (pretend_args_size) |
| { |
| if (cfun->machine->uses_anonymous_args) |
| { |
| /* Emit a stwm to push copy of argument registers onto |
| the stack for va_arg processing. */ |
| unsigned int r, mask = 0, n = pretend_args_size / 4; |
| for (r = LAST_ARG_REGNO - n + 1; r <= LAST_ARG_REGNO; r++) |
| mask |= (1 << r); |
| insn = emit_insn (nios2_ldst_parallel |
| (false, false, false, stack_mem, |
| -pretend_args_size, mask, false)); |
| /* Tag first SP adjustment as frame-related. */ |
| RTX_FRAME_RELATED_P (XVECEXP (PATTERN (insn), 0, 0)) = 1; |
| RTX_FRAME_RELATED_P (insn) = 1; |
| } |
| else |
| nios2_adjust_stack (-pretend_args_size, false); |
| } |
| if (callee_save_regs) |
| { |
| /* Emit a push.n to save registers and optionally allocate |
| push_immed extra bytes on the stack. */ |
| int sp_adjust; |
| if (caller_save_regs) |
| /* Can't allocate extra stack space yet. */ |
| push_immed = 0; |
| else if (cfun->machine->save_regs_offset <= 60) |
| /* Stack adjustment fits entirely in the push.n. */ |
| push_immed = cfun->machine->save_regs_offset; |
| else if (frame_pointer_needed |
| && cfun->machine->fp_save_offset == 0) |
| /* Deferring the entire stack adjustment until later |
| allows us to use a mov.n instead of a 32-bit addi |
| instruction to set the frame pointer. */ |
| push_immed = 0; |
| else |
| /* Splitting the stack adjustment between the push.n |
| and an explicit adjustment makes it more likely that |
| we can use spdeci.n for the explicit part. */ |
| push_immed = 60; |
| sp_adjust = -(cfun->machine->callee_save_reg_size + push_immed); |
| insn = emit_insn (nios2_ldst_parallel (false, false, false, |
| stack_mem, sp_adjust, |
| callee_save_regs, false)); |
| nios2_create_cfa_notes (insn, false); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| } |
| |
| if (caller_save_regs) |
| { |
| /* Emit a stwm to save the EH data regs, r4-r7. */ |
| int caller_save_size = (cfun->machine->save_reg_size |
| - cfun->machine->callee_save_reg_size); |
| gcc_assert ((caller_save_regs & ~0xf0) == 0); |
| insn = emit_insn (nios2_ldst_parallel |
| (false, false, false, stack_mem, |
| -caller_save_size, caller_save_regs, false)); |
| nios2_create_cfa_notes (insn, false); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| } |
| |
| save_regs_base = push_immed; |
| sp_offset = -(cfun->machine->save_regs_offset - push_immed); |
| } |
| /* The non-CDX cases decrement the stack pointer, to prepare for individual |
| register saves to the stack. */ |
| else if (!SMALL_INT (total_frame_size)) |
| { |
| /* We need an intermediary point, this will point at the spill block. */ |
| nios2_adjust_stack (cfun->machine->save_regs_offset - total_frame_size, |
| false); |
| save_regs_base = 0; |
| sp_offset = -cfun->machine->save_regs_offset; |
| if (crtl->limit_stack) |
| nios2_emit_stack_limit_check (cfun->machine->save_regs_offset); |
| } |
| else if (total_frame_size) |
| { |
| nios2_adjust_stack (-total_frame_size, false); |
| save_regs_base = cfun->machine->save_regs_offset; |
| sp_offset = 0; |
| if (crtl->limit_stack) |
| nios2_emit_stack_limit_check (0); |
| } |
| else |
| save_regs_base = sp_offset = 0; |
| |
| /* Save the registers individually in the non-CDX case. */ |
| if (!TARGET_HAS_CDX) |
| { |
| save_offset = save_regs_base + cfun->machine->save_reg_size; |
| |
| for (regno = LAST_GP_REG; regno > 0; regno--) |
| if (cfun->machine->save_mask & (1 << regno)) |
| { |
| save_offset -= 4; |
| save_reg (regno, save_offset); |
| } |
| } |
| |
| /* Set the hard frame pointer. */ |
| if (frame_pointer_needed) |
| { |
| int fp_save_offset = save_regs_base + cfun->machine->fp_save_offset; |
| insn = |
| (fp_save_offset == 0 |
| ? emit_move_insn (hard_frame_pointer_rtx, stack_pointer_rtx) |
| : emit_insn (gen_add3_insn (hard_frame_pointer_rtx, |
| stack_pointer_rtx, |
| gen_int_mode (fp_save_offset, Pmode)))); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| } |
| |
| /* Allocate sp_offset more bytes in the stack frame. */ |
| nios2_adjust_stack (sp_offset, false); |
| |
| /* Load the PIC register if needed. */ |
| if (crtl->uses_pic_offset_table) |
| nios2_load_pic_register (); |
| |
| /* If we are profiling, make sure no instructions are scheduled before |
| the call to mcount. */ |
| if (crtl->profile) |
| emit_insn (gen_blockage ()); |
| } |
| |
| void |
| nios2_expand_epilogue (bool sibcall_p) |
| { |
| rtx_insn *insn; |
| rtx cfa_adj; |
| int total_frame_size; |
| int sp_adjust, save_offset; |
| unsigned int regno; |
| |
| if (!sibcall_p && nios2_can_use_return_insn ()) |
| { |
| emit_jump_insn (gen_return ()); |
| return; |
| } |
| |
| emit_insn (gen_blockage ()); |
| |
| total_frame_size = nios2_compute_frame_layout (); |
| if (frame_pointer_needed) |
| { |
| /* Recover the stack pointer. */ |
| insn = |
| (cfun->machine->fp_save_offset == 0 |
| ? emit_move_insn (stack_pointer_rtx, hard_frame_pointer_rtx) |
| : emit_insn (gen_add3_insn |
| (stack_pointer_rtx, hard_frame_pointer_rtx, |
| gen_int_mode (-cfun->machine->fp_save_offset, Pmode)))); |
| cfa_adj = plus_constant (Pmode, stack_pointer_rtx, |
| (total_frame_size |
| - cfun->machine->save_regs_offset)); |
| add_reg_note (insn, REG_CFA_DEF_CFA, cfa_adj); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| |
| save_offset = 0; |
| sp_adjust = total_frame_size - cfun->machine->save_regs_offset; |
| } |
| else if (!SMALL_INT (total_frame_size)) |
| { |
| nios2_adjust_stack (cfun->machine->save_regs_offset, true); |
| save_offset = 0; |
| sp_adjust = total_frame_size - cfun->machine->save_regs_offset; |
| } |
| else |
| { |
| save_offset = cfun->machine->save_regs_offset; |
| sp_adjust = total_frame_size; |
| } |
| |
| if (!TARGET_HAS_CDX) |
| { |
| /* Generate individual register restores. */ |
| save_offset += cfun->machine->save_reg_size; |
| |
| for (regno = LAST_GP_REG; regno > 0; regno--) |
| if (cfun->machine->save_mask & (1 << regno)) |
| { |
| save_offset -= 4; |
| restore_reg (regno, save_offset); |
| } |
| nios2_adjust_stack (sp_adjust, true); |
| } |
| else if (cfun->machine->save_reg_size == 0) |
| { |
| /* Nothing to restore, just recover the stack position. */ |
| nios2_adjust_stack (sp_adjust, true); |
| } |
| else |
| { |
| /* Emit CDX pop.n/ldwm to restore registers and optionally return. */ |
| unsigned int regmask = cfun->machine->save_mask; |
| unsigned int callee_save_regs = regmask & 0xffff0000; |
| unsigned int caller_save_regs = regmask & 0x0000ffff; |
| int callee_save_size = cfun->machine->callee_save_reg_size; |
| int caller_save_size = cfun->machine->save_reg_size - callee_save_size; |
| int pretend_args_size = NIOS2_STACK_ALIGN (crtl->args.pretend_args_size); |
| bool ret_p = (!pretend_args_size && !crtl->calls_eh_return |
| && !sibcall_p); |
| |
| if (!ret_p || caller_save_size > 0) |
| sp_adjust = save_offset; |
| else |
| sp_adjust = (save_offset > 60 ? save_offset - 60 : 0); |
| |
| save_offset -= sp_adjust; |
| |
| nios2_adjust_stack (sp_adjust, true); |
| |
| if (caller_save_regs) |
| { |
| /* Emit a ldwm to restore EH data regs. */ |
| rtx stack_mem = gen_frame_mem (SImode, stack_pointer_rtx); |
| insn = emit_insn (nios2_ldst_parallel |
| (true, true, true, stack_mem, |
| caller_save_size, caller_save_regs, false)); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| nios2_create_cfa_notes (insn, true); |
| } |
| |
| if (callee_save_regs) |
| { |
| int sp_adjust = save_offset + callee_save_size; |
| rtx stack_mem; |
| if (ret_p) |
| { |
| /* Emit a pop.n to restore regs and return. */ |
| stack_mem = |
| gen_frame_mem (SImode, |
| gen_rtx_PLUS (Pmode, stack_pointer_rtx, |
| gen_int_mode (sp_adjust - 4, |
| Pmode))); |
| insn = |
| emit_jump_insn (nios2_ldst_parallel (true, false, false, |
| stack_mem, sp_adjust, |
| callee_save_regs, ret_p)); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| /* No need to attach CFA notes since we cannot step over |
| a return. */ |
| return; |
| } |
| else |
| { |
| /* If no return, we have to use the ldwm form. */ |
| stack_mem = gen_frame_mem (SImode, stack_pointer_rtx); |
| insn = |
| emit_insn (nios2_ldst_parallel (true, true, true, |
| stack_mem, sp_adjust, |
| callee_save_regs, ret_p)); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| nios2_create_cfa_notes (insn, true); |
| } |
| } |
| |
| if (pretend_args_size) |
| nios2_adjust_stack (pretend_args_size, true); |
| } |
| |
| /* Add in the __builtin_eh_return stack adjustment. */ |
| if (crtl->calls_eh_return) |
| emit_insn (gen_add2_insn (stack_pointer_rtx, EH_RETURN_STACKADJ_RTX)); |
| |
| if (!sibcall_p) |
| emit_jump_insn (gen_simple_return ()); |
| } |
| |
| bool |
| nios2_expand_return (void) |
| { |
| /* If CDX is available, generate a pop.n instruction to do both |
| the stack pop and return. */ |
| if (TARGET_HAS_CDX) |
| { |
| int total_frame_size = nios2_compute_frame_layout (); |
| int sp_adjust = (cfun->machine->save_regs_offset |
| + cfun->machine->callee_save_reg_size); |
| gcc_assert (sp_adjust == total_frame_size); |
| if (sp_adjust != 0) |
| { |
| rtx mem = |
| gen_frame_mem (SImode, |
| plus_constant (Pmode, stack_pointer_rtx, |
| sp_adjust - 4, false)); |
| rtx_insn *insn = |
| emit_jump_insn (nios2_ldst_parallel (true, false, false, |
| mem, sp_adjust, |
| cfun->machine->save_mask, |
| true)); |
| RTX_FRAME_RELATED_P (insn) = 1; |
| /* No need to create CFA notes since we can't step over |
| a return. */ |
| return true; |
| } |
| } |
| return false; |
| } |
| |
| /* Implement RETURN_ADDR_RTX. Note, we do not support moving |
| back to a previous frame. */ |
| rtx |
| nios2_get_return_address (int count) |
| { |
| if (count != 0) |
| return const0_rtx; |
| |
| return get_hard_reg_initial_val (Pmode, RA_REGNO); |
| } |
| |
| /* Emit code to change the current function's return address to |
| ADDRESS. SCRATCH is available as a scratch register, if needed. |
| ADDRESS and SCRATCH are both word-mode GPRs. */ |
| void |
| nios2_set_return_address (rtx address, rtx scratch) |
| { |
| nios2_compute_frame_layout (); |
| if (cfun->machine->save_mask & (1 << RA_REGNO)) |
| { |
| unsigned offset = cfun->machine->save_reg_size - 4; |
| rtx base; |
| |
| if (frame_pointer_needed) |
| base = hard_frame_pointer_rtx; |
| else |
| { |
| base = stack_pointer_rtx; |
| offset += cfun->machine->save_regs_offset; |
| |
| if (!SMALL_INT (offset)) |
| { |
| emit_move_insn (scratch, gen_int_mode (offset, Pmode)); |
| emit_insn (gen_add2_insn (scratch, base)); |
| base = scratch; |
| offset = 0; |
| } |
| } |
| if (offset) |
| base = plus_constant (Pmode, base, offset); |
| emit_move_insn (gen_rtx_MEM (Pmode, base), address); |
| } |
| else |
| emit_move_insn (gen_rtx_REG (Pmode, RA_REGNO), address); |
| } |
| |
| /* Implement FUNCTION_PROFILER macro. */ |
| void |
| nios2_function_profiler (FILE *file, int labelno ATTRIBUTE_UNUSED) |
| { |
| fprintf (file, "\tmov\tr8, ra\n"); |
| if (flag_pic == 1) |
| { |
| fprintf (file, "\tnextpc\tr2\n"); |
| fprintf (file, "\t1: movhi\tr3, %%hiadj(_gp_got - 1b)\n"); |
| fprintf (file, "\taddi\tr3, r3, %%lo(_gp_got - 1b)\n"); |
| fprintf (file, "\tadd\tr2, r2, r3\n"); |
| fprintf (file, "\tldw\tr2, %%call(_mcount)(r2)\n"); |
| fprintf (file, "\tcallr\tr2\n"); |
| } |
| else if (flag_pic == 2) |
| { |
| fprintf (file, "\tnextpc\tr2\n"); |
| fprintf (file, "\t1: movhi\tr3, %%hiadj(_gp_got - 1b)\n"); |
| fprintf (file, "\taddi\tr3, r3, %%lo(_gp_got - 1b)\n"); |
| fprintf (file, "\tadd\tr2, r2, r3\n"); |
| fprintf (file, "\tmovhi\tr3, %%call_hiadj(_mcount)\n"); |
| fprintf (file, "\taddi\tr3, r3, %%call_lo(_mcount)\n"); |
| fprintf (file, "\tadd\tr3, r2, r3\n"); |
| fprintf (file, "\tldw\tr2, 0(r3)\n"); |
| fprintf (file, "\tcallr\tr2\n"); |
| } |
| else |
| fprintf (file, "\tcall\t_mcount\n"); |
| fprintf (file, "\tmov\tra, r8\n"); |
| } |
| |
| /* Dump stack layout. */ |
| static void |
| nios2_dump_frame_layout (FILE *file) |
| { |
| fprintf (file, "\t%s Current Frame Info\n", ASM_COMMENT_START); |
| fprintf (file, "\t%s total_size = %d\n", ASM_COMMENT_START, |
| cfun->machine->total_size); |
| fprintf (file, "\t%s var_size = %d\n", ASM_COMMENT_START, |
| cfun->machine->var_size); |
| fprintf (file, "\t%s args_size = %d\n", ASM_COMMENT_START, |
| cfun->machine->args_size); |
| fprintf (file, "\t%s save_reg_size = %d\n", ASM_COMMENT_START, |
| cfun->machine->save_reg_size); |
| fprintf (file, "\t%s initialized = %d\n", ASM_COMMENT_START, |
| cfun->machine->initialized); |
| fprintf (file, "\t%s save_regs_offset = %d\n", ASM_COMMENT_START, |
| cfun->machine->save_regs_offset); |
| fprintf (file, "\t%s is_leaf = %d\n", ASM_COMMENT_START, |
| crtl->is_leaf); |
| fprintf (file, "\t%s frame_pointer_needed = %d\n", ASM_COMMENT_START, |
| frame_pointer_needed); |
| fprintf (file, "\t%s pretend_args_size = %d\n", ASM_COMMENT_START, |
| crtl->args.pretend_args_size); |
| } |
| |
| /* Return true if REGNO should be saved in the prologue. */ |
| static bool |
| prologue_saved_reg_p (unsigned regno) |
| { |
| gcc_assert (GP_REG_P (regno)); |
| |
| if (df_regs_ever_live_p (regno) && !call_used_or_fixed_reg_p (regno)) |
| return true; |
| |
| if (regno == HARD_FRAME_POINTER_REGNUM && frame_pointer_needed) |
| return true; |
| |
| if (regno == PIC_OFFSET_TABLE_REGNUM && crtl->uses_pic_offset_table) |
| return true; |
| |
| if (regno == RA_REGNO && df_regs_ever_live_p (RA_REGNO)) |
| return true; |
| |
| return false; |
| } |
| |
| /* Implement TARGET_CAN_ELIMINATE. */ |
| static bool |
| nios2_can_eliminate (const int from ATTRIBUTE_UNUSED, const int to) |
| { |
| if (to == STACK_POINTER_REGNUM) |
| return !frame_pointer_needed; |
| return true; |
| } |
| |
| /* Implement INITIAL_ELIMINATION_OFFSET macro. */ |
| int |
| nios2_initial_elimination_offset (int from, int to) |
| { |
| int offset; |
| |
| nios2_compute_frame_layout (); |
| |
| /* Set OFFSET to the offset from the stack pointer. */ |
| switch (from) |
| { |
| case FRAME_POINTER_REGNUM: |
| /* This is the high end of the local variable storage, not the |
| hard frame pointer. */ |
| offset = cfun->machine->args_size + cfun->machine->var_size; |
| break; |
| |
| case ARG_POINTER_REGNUM: |
| offset = cfun->machine->total_size; |
| offset -= crtl->args.pretend_args_size; |
| break; |
| |
| default: |
| gcc_unreachable (); |
| } |
| |
| /* If we are asked for the frame pointer offset, then adjust OFFSET |
| by the offset from the frame pointer to the stack pointer. */ |
| if (to == HARD_FRAME_POINTER_REGNUM) |
| offset -= (cfun->machine->save_regs_offset |
| + cfun->machine->fp_save_offset); |
| |
| return offset; |
| } |
| |
| /* Return nonzero if this function is known to have a null epilogue. |
| This allows the optimizer to omit jumps to jumps if no stack |
| was created. */ |
| int |
| nios2_can_use_return_insn (void) |
| { |
| int total_frame_size; |
| |
| if (!reload_completed || crtl->profile) |
| return 0; |
| |
| total_frame_size = nios2_compute_frame_layout (); |
| |
| /* If CDX is available, check if we can return using a |
| single pop.n instruction. */ |
| if (TARGET_HAS_CDX |
| && !frame_pointer_needed |
| && cfun->machine->save_regs_offset <= 60 |
| && (cfun->machine->save_mask & 0x80000000) != 0 |
| && (cfun->machine->save_mask & 0xffff) == 0 |
| && crtl->args.pretend_args_size == 0) |
| return true; |
| |
| return total_frame_size == 0; |
| } |
| |
| |
| /* Check and signal some warnings/errors on FPU insn options. */ |
| static void |
| nios2_custom_check_insns (void) |
| { |
| unsigned int i, j; |
| bool errors = false; |
| |
| for (i = 0; i < ARRAY_SIZE (nios2_fpu_insn); i++) |
| if (N2FPU_ENABLED_P (i) && N2FPU_DOUBLE_P (i)) |
| { |
| for (j = 0; j < ARRAY_SIZE (nios2_fpu_insn); j++) |
| if (N2FPU_DOUBLE_REQUIRED_P (j) && ! N2FPU_ENABLED_P (j)) |
| { |
| error ("switch %<-mcustom-%s%> is required for " |
| "double-precision floating-point", N2FPU_NAME (j)); |
| errors = true; |
| } |
| break; |
| } |
| |
| if (errors || custom_code_conflict) |
| fatal_error (input_location, |
| "conflicting use of %<-mcustom%> switches, " |
| "target attributes, " |
| "and/or %<__builtin_custom_%> functions"); |
| } |
| |
| static void |
| nios2_set_fpu_custom_code (enum n2fpu_code code, int n, bool override_p) |
| { |
| if (override_p || N2FPU_N (code) == -1) |
| N2FPU_N (code) = n; |
| nios2_register_custom_code (n, CCS_FPU, (int) code); |
| } |
| |
| /* Type to represent a standard FPU config. */ |
| struct nios2_fpu_config |
| { |
| const char *name; |
| bool set_sp_constants; |
| int code[n2fpu_code_num]; |
| }; |
| |
| #define NIOS2_FPU_CONFIG_NUM 4 |
| static struct nios2_fpu_config custom_fpu_config[NIOS2_FPU_CONFIG_NUM]; |
| |
| static void |
| nios2_init_fpu_configs (void) |
| { |
| struct nios2_fpu_config* cfg; |
| int i = 0; |
| #define NEXT_FPU_CONFIG \ |
| do { \ |
| cfg = &custom_fpu_config[i++]; \ |
| memset (cfg, -1, sizeof (struct nios2_fpu_config));\ |
| } while (0) |
| |
| NEXT_FPU_CONFIG; |
| cfg->name = "60-1"; |
| cfg->set_sp_constants = true; |
| cfg->code[n2fpu_fmuls] = 252; |
| cfg->code[n2fpu_fadds] = 253; |
| cfg->code[n2fpu_fsubs] = 254; |
| |
| NEXT_FPU_CONFIG; |
| cfg->name = "60-2"; |
| cfg->set_sp_constants = true; |
| cfg->code[n2fpu_fmuls] = 252; |
| cfg->code[n2fpu_fadds] = 253; |
| cfg->code[n2fpu_fsubs] = 254; |
| cfg->code[n2fpu_fdivs] = 255; |
| |
| NEXT_FPU_CONFIG; |
| cfg->name = "72-3"; |
| cfg->set_sp_constants = true; |
| cfg->code[n2fpu_floatus] = 243; |
| cfg->code[n2fpu_fixsi] = 244; |
| cfg->code[n2fpu_floatis] = 245; |
| cfg->code[n2fpu_fcmpgts] = 246; |
| cfg->code[n2fpu_fcmples] = 249; |
| cfg->code[n2fpu_fcmpeqs] = 250; |
| cfg->code[n2fpu_fcmpnes] = 251; |
| cfg->code[n2fpu_fmuls] = 252; |
| cfg->code[n2fpu_fadds] = 253; |
| cfg->code[n2fpu_fsubs] = 254; |
| cfg->code[n2fpu_fdivs] = 255; |
| |
| NEXT_FPU_CONFIG; |
| cfg->name = "fph2"; |
| cfg->code[n2fpu_fabss] = 224; |
| cfg->code[n2fpu_fnegs] = 225; |
| cfg->code[n2fpu_fcmpnes] = 226; |
| cfg->code[n2fpu_fcmpeqs] = 227; |
| cfg->code[n2fpu_fcmpges] = 228; |
| cfg->code[n2fpu_fcmpgts] = 229; |
| cfg->code[n2fpu_fcmples] = 230; |
| cfg->code[n2fpu_fcmplts] = 231; |
| cfg->code[n2fpu_fmaxs] = 232; |
| cfg->code[n2fpu_fmins] = 233; |
| cfg->code[n2fpu_round] = 248; |
| cfg->code[n2fpu_fixsi] = 249; |
| cfg->code[n2fpu_floatis] = 250; |
| cfg->code[n2fpu_fsqrts] = 251; |
| cfg->code[n2fpu_fmuls] = 252; |
| cfg->code[n2fpu_fadds] = 253; |
| cfg->code[n2fpu_fsubs] = 254; |
| cfg->code[n2fpu_fdivs] = 255; |
| |
| #undef NEXT_FPU_CONFIG |
| gcc_assert (i == NIOS2_FPU_CONFIG_NUM); |
| } |
| |
| static struct nios2_fpu_config * |
| nios2_match_custom_fpu_cfg (const char *cfgname, const char *endp) |
| { |
| int i; |
| for (i = 0; i < NIOS2_FPU_CONFIG_NUM; i++) |
| { |
| bool match = !(endp != NULL |
| ? strncmp (custom_fpu_config[i].name, cfgname, |
| endp - cfgname) |
| : strcmp (custom_fpu_config[i].name, cfgname)); |
| if (match) |
| return &custom_fpu_config[i]; |
| } |
| return NULL; |
| } |
| |
| /* Use CFGNAME to lookup FPU config, ENDP if not NULL marks end of string. |
| OVERRIDE is true if loaded config codes should overwrite current state. */ |
| static void |
| nios2_handle_custom_fpu_cfg (const char *cfgname, const char *endp, |
| bool override) |
| { |
| struct nios2_fpu_config *cfg = nios2_match_custom_fpu_cfg (cfgname, endp); |
| if (cfg) |
| { |
| unsigned int i; |
| for (i = 0; i < ARRAY_SIZE (nios2_fpu_insn); i++) |
| if (cfg->code[i] >= 0) |
| nios2_set_fpu_custom_code ((enum n2fpu_code) i, cfg->code[i], |
| override); |
| if (cfg->set_sp_constants) |
| flag_single_precision_constant = 1; |
| } |
| else |
| warning (0, "ignoring unrecognized switch %<-mcustom-fpu-cfg%> " |
| "value %<%s%>", cfgname); |
| |
| /* Guard against errors in the standard configurations. */ |
| nios2_custom_check_insns (); |
| } |
| |
| /* Check individual FPU insn options, and register custom code. */ |
| static void |
| nios2_handle_custom_fpu_insn_option (int fpu_insn_index) |
| { |
| int param = N2FPU_N (fpu_insn_index); |
| |
| if (param >= 0 && param <= 255) |
| nios2_register_custom_code (param, CCS_FPU, fpu_insn_index); |
| |
| /* Valid values are 0-255, but also allow -1 so that the |
| -mno-custom-<opt> switches work. */ |
| else if (param != -1) |
| error ("switch %<-mcustom-%s%> value %d must be between 0 and 255", |
| N2FPU_NAME (fpu_insn_index), param); |
| } |
| |
| /* Allocate a chunk of memory for per-function machine-dependent data. */ |
| static struct machine_function * |
| nios2_init_machine_status (void) |
| { |
| return ggc_cleared_alloc<machine_function> (); |
| } |
| |
| /* Implement TARGET_OPTION_OVERRIDE. */ |
| static void |
| nios2_option_override (void) |
| { |
| unsigned int i; |
| |
| #ifdef SUBTARGET_OVERRIDE_OPTIONS |
| SUBTARGET_OVERRIDE_OPTIONS; |
| #endif |
| |
| /* Check for unsupported options. */ |
| if (flag_pic && !TARGET_LINUX_ABI) |
| sorry ("position-independent code requires the Linux ABI"); |
| if (flag_pic && stack_limit_rtx |
| && GET_CODE (stack_limit_rtx) == SYMBOL_REF) |
| sorry ("PIC support for %<-fstack-limit-symbol%>"); |
| |
| /* Function to allocate machine-dependent function status. */ |
| init_machine_status = &nios2_init_machine_status; |
| |
| nios2_section_threshold |
| = (global_options_set.x_g_switch_value |
| ? g_switch_value : NIOS2_DEFAULT_GVALUE); |
| |
| if (nios2_gpopt_option == gpopt_unspecified) |
| { |
| /* Default to -mgpopt unless -fpic or -fPIC. */ |
| if (flag_pic) |
| nios2_gpopt_option = gpopt_none; |
| else |
| nios2_gpopt_option = gpopt_local; |
| } |
| |
| /* GP-relative and r0-relative addressing don't make sense for PIC. */ |
| if (flag_pic) |
| { |
| if (nios2_gpopt_option != gpopt_none) |
| error ("%<-mgpopt%> not supported with PIC"); |
| if (nios2_gprel_sec) |
| error ("%<-mgprel-sec=%> not supported with PIC"); |
| if (nios2_r0rel_sec) |
| error ("%<-mr0rel-sec=%> not supported with PIC"); |
| } |
| |
| /* Process -mgprel-sec= and -m0rel-sec=. */ |
| if (nios2_gprel_sec) |
| { |
| if (regcomp (&nios2_gprel_sec_regex, nios2_gprel_sec, |
| REG_EXTENDED | REG_NOSUB)) |
| error ("%<-mgprel-sec=%> argument is not a valid regular expression"); |
| } |
| if (nios2_r0rel_sec) |
| { |
| if (regcomp (&nios2_r0rel_sec_regex, nios2_r0rel_sec, |
| REG_EXTENDED | REG_NOSUB)) |
| error ("%<-mr0rel-sec=%> argument is not a valid regular expression"); |
| } |
| |
| /* If we don't have mul, we don't have mulx either! */ |
| if (!TARGET_HAS_MUL && TARGET_HAS_MULX) |
| target_flags &= ~MASK_HAS_MULX; |
| |
| /* Optional BMX and CDX instructions only make sense for R2. */ |
| if (!TARGET_ARCH_R2) |
| { |
| if (TARGET_HAS_BMX) |
| error ("BMX instructions are only supported with R2 architecture"); |
| if (TARGET_HAS_CDX) |
| error ("CDX instructions are only supported with R2 architecture"); |
| } |
| |
| /* R2 is little-endian only. */ |
| if (TARGET_ARCH_R2 && TARGET_BIG_ENDIAN) |
| error ("R2 architecture is little-endian only"); |
| |
| /* Initialize default FPU configurations. */ |
| nios2_init_fpu_configs (); |
| |
| /* Set up default handling for floating point custom instructions. |
| |
| Putting things in this order means that the -mcustom-fpu-cfg= |
| switch will always be overridden by individual -mcustom-fadds= |
| switches, regardless of the order in which they were specified |
| on the command line. |
| |
| This behavior of prioritization of individual -mcustom-<insn>= |
| options before the -mcustom-fpu-cfg= switch is maintained for |
| compatibility. */ |
| if (nios2_custom_fpu_cfg_string && *nios2_custom_fpu_cfg_string) |
| nios2_handle_custom_fpu_cfg (nios2_custom_fpu_cfg_string, NULL, false); |
| |
| /* Handle options for individual FPU insns. */ |
| for (i = 0; i < ARRAY_SIZE (nios2_fpu_insn); i++) |
| nios2_handle_custom_fpu_insn_option (i); |
| |
| nios2_custom_check_insns (); |
| |
| /* Save the initial options in case the user does function specific |
| options. */ |
| target_option_default_node = target_option_current_node |
| = build_target_option_node (&global_options, &global_options_set); |
| } |
| |
| |
| /* Return true if CST is a constant within range of movi/movui/movhi. */ |
| static bool |
| nios2_simple_const_p (const_rtx cst) |
| { |
| if (!CONST_INT_P (cst)) |
| return false; |
| HOST_WIDE_INT val = INTVAL (cst); |
| return SMALL_INT (val) || SMALL_INT_UNSIGNED (val) || UPPER16_INT (val); |
| } |
| |
| /* Compute a (partial) cost for rtx X. Return true if the complete |
| cost has been computed, and false if subexpressions should be |
| scanned. In either case, *TOTAL contains the cost result. */ |
| static bool |
| nios2_rtx_costs (rtx x, machine_mode mode, |
| int outer_code, |
| int opno, |
| int *total, bool speed) |
| { |
| int code = GET_CODE (x); |
| |
| switch (code) |
| { |
| case CONST_INT: |
| if (INTVAL (x) == 0 || nios2_simple_const_p (x)) |
| { |
| *total = COSTS_N_INSNS (0); |
| return true; |
| } |
| else |
| { |
| /* High + lo_sum. */ |
| *total = COSTS_N_INSNS (1); |
| return true; |
| } |
| |
| case LABEL_REF: |
| case SYMBOL_REF: |
| case CONST: |
| case CONST_DOUBLE: |
| if (gprel_constant_p (x) || r0rel_constant_p (x)) |
| { |
| *total = COSTS_N_INSNS (1); |
| return true; |
| } |
| else |
| { |
| /* High + lo_sum. */ |
| *total = COSTS_N_INSNS (1); |
| return true; |
| } |
| |
| case HIGH: |
| { |
| /* This is essentially a constant. */ |
| *total = COSTS_N_INSNS (0); |
| return true; |
| } |
| |
| case LO_SUM: |
| { |
| *total = COSTS_N_INSNS (0); |
| return true; |
| } |
| |
| case AND: |
| { |
| /* Recognize 'nor' insn pattern. */ |
| if (GET_CODE (XEXP (x, 0)) == NOT |
| && GET_CODE (XEXP (x, 1)) == NOT) |
| { |
| *total = COSTS_N_INSNS (1); |
| return true; |
| } |
| return false; |
| } |
| |
| /* For insns that have an execution latency (3 cycles), don't |
| penalize by the full amount since we can often schedule |
| to avoid it. */ |
| case MULT: |
| { |
| if (!TARGET_HAS_MUL) |
| *total = COSTS_N_INSNS (5); /* Guess? */ |
| else if (speed) |
| *total = COSTS_N_INSNS (2); /* Latency adjustment. */ |
| else |
| *total = COSTS_N_INSNS (1); |
| if (TARGET_HAS_MULX && GET_MODE (x) == DImode) |
| { |
| enum rtx_code c0 = GET_CODE (XEXP (x, 0)); |
| enum rtx_code c1 = GET_CODE (XEXP (x, 1)); |
| if ((c0 == SIGN_EXTEND && c1 == SIGN_EXTEND) |
| || (c0 == ZERO_EXTEND && c1 == ZERO_EXTEND)) |
| /* This is the <mul>sidi3 pattern, which expands into 4 insns, |
| 2 multiplies and 2 moves. */ |
| { |
| *total = *total * 2 + COSTS_N_INSNS (2); |
| return true; |
| } |
| } |
| return false; |
| } |
| |
| case DIV: |
| { |
| if (!TARGET_HAS_DIV) |
| *total = COSTS_N_INSNS (5); /* Guess? */ |
| else if (speed) |
| *total = COSTS_N_INSNS (2); /* Latency adjustment. */ |
| else |
| *total = COSTS_N_INSNS (1); |
| return false; |
| } |
| |
| case ASHIFT: |
| case ASHIFTRT: |
| case LSHIFTRT: |
| case ROTATE: |
| { |
| if (!speed) |
| *total = COSTS_N_INSNS (1); |
| else |
| *total = COSTS_N_INSNS (2); /* Latency adjustment. */ |
| return false; |
| } |
| |
| case ZERO_EXTRACT: |
| if (TARGET_HAS_BMX) |
| { |
| *total = COSTS_N_INSNS (1); |
| return true; |
| } |
| return false; |
| |
| case SIGN_EXTEND: |
| { |
| if (MEM_P (XEXP (x, 0))) |
| *total = COSTS_N_INSNS (1); |
| else |
| *total = COSTS_N_INSNS (3); |
| return false; |
| } |
| |
| case MEM: |
| { |
| rtx addr = XEXP (x, 0); |
| |
| /* Account for cost of different addressing modes. */ |
| *total = nios2_address_cost (addr, mode, ADDR_SPACE_GENERIC, speed); |
| |
| if (outer_code == SET && opno == 0) |
| /* Stores execute in 1 cycle accounted for by |
| the outer SET. */ |
| ; |
| else if (outer_code == SET || outer_code == SIGN_EXTEND |
| || outer_code == ZERO_EXTEND) |
| /* Latency adjustment. */ |
| { |
| if (speed) |
| *total += COSTS_N_INSNS (1); |
| } |
| else |
| /* This is going to have to be split into a load. */ |
| *total += COSTS_N_INSNS (speed ? 2 : 1); |
| return true; |
| } |
| |
| default: |
| return false; |
| } |
| } |
| |
| /* Implement TARGET_PREFERRED_RELOAD_CLASS. */ |
| static reg_class_t |
| nios2_preferred_reload_class (rtx x ATTRIBUTE_UNUSED, reg_class_t regclass) |
| { |
| return regclass == NO_REGS ? GENERAL_REGS : regclass; |
| } |
| |
| /* Emit a call to __tls_get_addr. TI is the argument to this function. |
| RET is an RTX for the return value location. The entire insn sequence |
| is returned. */ |
| static GTY(()) rtx nios2_tls_symbol; |
| |
| static rtx |
| nios2_call_tls_get_addr (rtx ti) |
| { |
| rtx arg = gen_rtx_REG (Pmode, FIRST_ARG_REGNO); |
| rtx ret = gen_rtx_REG (Pmode, FIRST_RETVAL_REGNO); |
| rtx fn; |
| rtx_insn *insn; |
| |
| if (!nios2_tls_symbol) |
| nios2_tls_symbol = init_one_libfunc ("__tls_get_addr"); |
| |
| emit_move_insn (arg, ti); |
| fn = gen_rtx_MEM (QImode, nios2_tls_symbol); |
| insn = emit_call_insn (gen_call_value (ret, fn, const0_rtx)); |
| RTL_CONST_CALL_P (insn) = 1; |
| use_reg (&CALL_INSN_FUNCTION_USAGE (insn), ret); |
| use_reg (&CALL_INSN_FUNCTION_USAGE (insn), arg); |
| |
| return ret; |
| } |
| |
| /* Return true for large offsets requiring hiadj/lo relocation pairs. */ |
| static bool |
| nios2_large_offset_p (int unspec) |
| { |
| gcc_assert (nios2_unspec_reloc_name (unspec) != NULL); |
| |
| if (flag_pic == 2 |
| /* FIXME: TLS GOT offset relocations will eventually also get this |
| treatment, after binutils support for those are also completed. */ |
| && (unspec == UNSPEC_PIC_SYM || unspec == UNSPEC_PIC_CALL_SYM)) |
| return true; |
| |
| /* 'gotoff' offsets are always hiadj/lo. */ |
| if (unspec == UNSPEC_PIC_GOTOFF_SYM) |
| return true; |
| |
| return false; |
| } |
| |
| /* Return true for conforming unspec relocations. Also used in |
| constraints.md and predicates.md. */ |
| bool |
| nios2_unspec_reloc_p (rtx op) |
| { |
| return (GET_CODE (op) == CONST |
| && GET_CODE (XEXP (op, 0)) == UNSPEC |
| && ! nios2_large_offset_p (XINT (XEXP (op, 0), 1))); |
| } |
| |
| static bool |
| nios2_large_unspec_reloc_p (rtx op) |
| { |
| return (GET_CODE (op) == CONST |
| && GET_CODE (XEXP (op, 0)) == UNSPEC |
| && nios2_large_offset_p (XINT (XEXP (op, 0), 1))); |
| } |
| |
| /* Helper to generate unspec constant. */ |
| static rtx |
| nios2_unspec_offset (rtx loc, int unspec) |
| { |
| return gen_rtx_CONST (Pmode, gen_rtx_UNSPEC (Pmode, gen_rtvec (1, loc), |
| unspec)); |
| } |
| |
| /* Generate GOT pointer based address with large offset. */ |
| static rtx |
| nios2_large_got_address (rtx offset, rtx tmp) |
| { |
| if (!tmp) |
| tmp = gen_reg_rtx (Pmode); |
| emit_move_insn (tmp, offset); |
| return gen_rtx_PLUS (Pmode, tmp, pic_offset_table_rtx); |
| } |
| |
| /* Generate a GOT pointer based address. */ |
| static rtx |
| nios2_got_address (rtx loc, int unspec) |
| { |
| rtx offset = nios2_unspec_offset (loc, unspec); |
| crtl->uses_pic_offset_table = 1; |
| |
| if (nios2_large_offset_p (unspec)) |
| return force_reg (Pmode, nios2_large_got_address (offset, NULL_RTX)); |
| |
| return gen_rtx_PLUS (Pmode, pic_offset_table_rtx, offset); |
| } |
| |
| /* Generate the code to access LOC, a thread local SYMBOL_REF. The |
| return value will be a valid address and move_operand (either a REG |
| or a LO_SUM). */ |
| static rtx |
| nios2_legitimize_tls_address (rtx loc) |
| { |
| rtx tmp, mem, tp; |
| enum tls_model model = SYMBOL_REF_TLS_MODEL (loc); |
| |
| switch (model) |
| { |
| case TLS_MODEL_GLOBAL_DYNAMIC: |
| tmp = gen_reg_rtx (Pmode); |
| emit_move_insn (tmp, nios2_got_address (loc, UNSPEC_ADD_TLS_GD)); |
| return nios2_call_tls_get_addr (tmp); |
| |
| case TLS_MODEL_LOCAL_DYNAMIC: |
| tmp = gen_reg_rtx (Pmode); |
| emit_move_insn (tmp, nios2_got_address (loc, UNSPEC_ADD_TLS_LDM)); |
| return gen_rtx_PLUS (Pmode, nios2_call_tls_get_addr (tmp), |
| nios2_unspec_offset (loc, UNSPEC_ADD_TLS_LDO)); |
| |
| case TLS_MODEL_INITIAL_EXEC: |
| tmp = gen_reg_rtx (Pmode); |
| mem = gen_const_mem (Pmode, nios2_got_address (loc, UNSPEC_LOAD_TLS_IE)); |
| emit_move_insn (tmp, mem); |
| tp = gen_rtx_REG (Pmode, TP_REGNO); |
| return gen_rtx_PLUS (Pmode, tp, tmp); |
| |
| case TLS_MODEL_LOCAL_EXEC: |
| tp = gen_rtx_REG (Pmode, TP_REGNO); |
| return gen_rtx_PLUS (Pmode, tp, |
| nios2_unspec_offset (loc, UNSPEC_ADD_TLS_LE)); |
| default: |
| gcc_unreachable (); |
| } |
| } |
| |
| /* Divide Support |
| |
| If -O3 is used, we want to output a table lookup for |
| divides between small numbers (both num and den >= 0 |
| and < 0x10). The overhead of this method in the worst |
| case is 40 bytes in the text section (10 insns) and |
| 256 bytes in the data section. Additional divides do |
| not incur additional penalties in the data section. |
| |
| Code speed is improved for small divides by about 5x |
| when using this method in the worse case (~9 cycles |
| vs ~45). And in the worst case divides not within the |
| table are penalized by about 10% (~5 cycles vs ~45). |
| However in the typical case the penalty is not as bad |
| because doing the long divide in only 45 cycles is |
| quite optimistic. |
| |
| ??? would be nice to have some benchmarks other |
| than Dhrystone to back this up. |
| |
| This bit of expansion is to create this instruction |
| sequence as rtl. |
| or $8, $4, $5 |
| slli $9, $4, 4 |
| cmpgeui $3, $8, 16 |
| beq $3, $0, .L3 |
| or $10, $9, $5 |
| add $12, $11, divide_table |
| ldbu $2, 0($12) |
| br .L1 |
| .L3: |
| call slow_div |
| .L1: |
| # continue here with result in $2 |
| |
| ??? Ideally I would like the libcall block to contain all |
| of this code, but I don't know how to do that. What it |
| means is that if the divide can be eliminated, it may not |
| completely disappear. |
| |
| ??? The __divsi3_table label should ideally be moved out |
| of this block and into a global. If it is placed into the |
| sdata section we can save even more cycles by doing things |
| gp relative. */ |
| void |
| nios2_emit_expensive_div (rtx *operands, machine_mode mode) |
| { |
| rtx or_result, shift_left_result; |
| rtx lookup_value; |
| rtx_code_label *lab1, *lab3; |
| rtx_insn *insns; |
| rtx libfunc; |
| rtx final_result; |
| rtx_insn *tmp; |
| rtx table; |
| |
| /* It may look a little generic, but only SImode is supported for now. */ |
| gcc_assert (mode == SImode); |
| libfunc = optab_libfunc (sdiv_optab, SImode); |
| |
| lab1 = gen_label_rtx (); |
| lab3 = gen_label_rtx (); |
| |
| or_result = expand_simple_binop (SImode, IOR, |
| operands[1], operands[2], |
| 0, 0, OPTAB_LIB_WIDEN); |
| |
| emit_cmp_and_jump_insns (or_result, GEN_INT (15), GTU, 0, |
| GET_MODE (or_result), 0, lab3); |
| JUMP_LABEL (get_last_insn ()) = lab3; |
| |
| shift_left_result = expand_simple_binop (SImode, ASHIFT, |
| operands[1], GEN_INT (4), |
| 0, 0, OPTAB_LIB_WIDEN); |
| |
| lookup_value = expand_simple_binop (SImode, IOR, |
| shift_left_result, operands[2], |
| 0, 0, OPTAB_LIB_WIDEN); |
| table = gen_rtx_PLUS (SImode, lookup_value, |
| gen_rtx_SYMBOL_REF (SImode, "__divsi3_table")); |
| convert_move (operands[0], gen_rtx_MEM (QImode, table), 1); |
| |
| tmp = emit_jump_insn (gen_jump (lab1)); |
| JUMP_LABEL (tmp) = lab1; |
| emit_barrier (); |
| |
| emit_label (lab3); |
| LABEL_NUSES (lab3) = 1; |
| |
| start_sequence (); |
| final_result = emit_library_call_value (libfunc, NULL_RTX, |
| LCT_CONST, SImode, |
| operands[1], SImode, |
| operands[2], SImode); |
| |
| insns = get_insns (); |
| end_sequence (); |
| emit_libcall_block (insns, operands[0], final_result, |
| gen_rtx_DIV (SImode, operands[1], operands[2])); |
| |
| emit_label (lab1); |
| LABEL_NUSES (lab1) = 1; |
| } |
| |
| |
| /* Branches and compares. */ |
| |
| /* Return in *ALT_CODE and *ALT_OP, an alternate equivalent constant |
| comparison, e.g. >= 1 into > 0. */ |
| static void |
| nios2_alternate_compare_const (enum rtx_code code, rtx op, |
| enum rtx_code *alt_code, rtx *alt_op, |
| machine_mode mode) |
| { |
| gcc_assert (CONST_INT_P (op)); |
| |
| HOST_WIDE_INT opval = INTVAL (op); |
| enum rtx_code scode = signed_condition (code); |
| bool dec_p = (scode == LT || scode == GE); |
| |
| if (code == EQ || code == NE) |
| { |
| *alt_code = code; |
| *alt_op = op; |
| return; |
| } |
| |
| *alt_op = (dec_p |
| ? gen_int_mode (opval - 1, mode) |
| : gen_int_mode (opval + 1, mode)); |
| |
| /* The required conversion between [>,>=] and [<,<=] is captured |
| by a reverse + swap of condition codes. */ |
| *alt_code = reverse_condition (swap_condition (code)); |
| |
| { |
| /* Test if the incremented/decremented value crosses the over/underflow |
| boundary. Supposedly, such boundary cases should already be transformed |
| into always-true/false or EQ conditions, so use an assertion here. */ |
| unsigned HOST_WIDE_INT alt_opval = INTVAL (*alt_op); |
| if (code == scode) |
| alt_opval ^= (1 << (GET_MODE_BITSIZE (mode) - 1)); |
| alt_opval &= GET_MODE_MASK (mode); |
| gcc_assert (dec_p ? alt_opval != GET_MODE_MASK (mode) : alt_opval != 0); |
| } |
| } |
| |
| /* Return true if the constant comparison is supported by nios2. */ |
| static bool |
| nios2_valid_compare_const_p (enum rtx_code code, rtx op) |
| { |
| gcc_assert (CONST_INT_P (op)); |
| switch (code) |
| { |
| case EQ: case NE: case GE: case LT: |
| return SMALL_INT (INTVAL (op)); |
| case GEU: case LTU: |
| return SMALL_INT_UNSIGNED (INTVAL (op)); |
| default: |
| return false; |
| } |
| } |
| |
| /* Checks if the FPU comparison in *CMP, *OP1, and *OP2 can be supported in |
| the current configuration. Perform modifications if MODIFY_P is true. |
| Returns true if FPU compare can be done. */ |
| |
| bool |
| nios2_validate_fpu_compare (machine_mode mode, rtx *cmp, rtx *op1, rtx *op2, |
| bool modify_p) |
| { |
| bool rev_p = false; |
| enum rtx_code code = GET_CODE (*cmp); |
| |
| if (!nios2_fpu_compare_enabled (code, mode)) |
| { |
| code = swap_condition (code); |
| if (nios2_fpu_compare_enabled (code, mode)) |
| rev_p = true; |
| else |
| return false; |
| } |
| |
| if (modify_p) |
| { |
| if (rev_p) |
| { |
| rtx tmp = *op1; |
| *op1 = *op2; |
| *op2 = tmp; |
| } |
| *op1 = force_reg (mode, *op1); |
| *op2 = force_reg (mode, *op2); |
| *cmp = gen_rtx_fmt_ee (code, mode, *op1, *op2); |
| } |
| return true; |
| } |
| |
| /* Checks and modifies the comparison in *CMP, *OP1, and *OP2 into valid |
| nios2 supported form. Returns true if success. */ |
| bool |
| nios2_validate_compare (machine_mode mode, rtx *cmp, rtx *op1, rtx *op2) |
| { |
| enum rtx_code code = GET_CODE (*cmp); |
| enum rtx_code alt_code; |
| rtx alt_op2; |
| |
| if (GET_MODE_CLASS (mode) == MODE_FLOAT) |
| return nios2_validate_fpu_compare (mode, cmp, op1, op2, true); |
| |
| if (CONST_INT_P (*op2) && *op2 != const0_rtx) |
| { |
| /* Create alternate constant compare. */ |
| nios2_alternate_compare_const (code, *op2, &alt_code, &alt_op2, mode); |
| |
| /* If alterate op2 is zero(0), we can use it directly, possibly |
| swapping the compare code. */ |
| if (alt_op2 == const0_rtx) |
| { |
| code = alt_code; |
| *op2 = alt_op2; |
| goto check_rebuild_cmp; |
| } |
| |
| /* Check if either constant compare can be used. */ |
| if (nios2_valid_compare_const_p (code, *op2)) |
| return true; |
| else if (nios2_valid_compare_const_p (alt_code, alt_op2)) |
| { |
| code = alt_code; |
| *op2 = alt_op2; |
| goto rebuild_cmp; |
| } |
| |
| /* We have to force op2 into a register now. Try to pick one |
| with a lower cost. */ |
| if (! nios2_simple_const_p (*op2) |
| && nios2_simple_const_p (alt_op2)) |
| { |
| code = alt_code; |
| *op2 = alt_op2; |
| } |
| *op2 = force_reg (mode, *op2); |
| } |
| else if (!reg_or_0_operand (*op2, mode)) |
| *op2 = force_reg (mode, *op2); |
| |
| check_rebuild_cmp: |
| if (code == GT || code == GTU || code == LE || code == LEU) |
| { |
| rtx t = *op1; *op1 = *op2; *op2 = t; |
| code = swap_condition (code); |
| } |
| rebuild_cmp: |
| *cmp = gen_rtx_fmt_ee (code, mode, *op1, *op2); |
| return true; |
| } |
| |
| |
| /* Addressing modes and constants. */ |
| |
| /* Symbol references and other 32-bit constants are split into |
| high/lo_sum pairs during the split1 pass. After that, they are not |
| considered legitimate addresses. |
| This function returns true if in a pre-split context where these |
| constants are allowed. */ |
| static bool |
| nios2_large_constant_allowed (void) |
| { |
| /* The reload_completed check is for the benefit of |
| nios2_asm_output_mi_thunk and perhaps other places that try to |
| emulate a post-reload pass. */ |
| return !(cfun->curr_properties & PROP_rtl_split_insns) && !reload_completed; |
| } |
| |
| /* Return true if X is constant expression with a reference to an |
| "ordinary" symbol; not GOT-relative, not GP-relative, not TLS. */ |
| static bool |
| nios2_symbolic_constant_p (rtx x) |
| { |
| rtx base, offset; |
| |
| if (flag_pic) |
| return false; |
| if (GET_CODE (x) == LABEL_REF) |
| return true; |
| else if (CONSTANT_P (x)) |
| { |
| split_const (x, &base, &offset); |
| return (SYMBOL_REF_P (base) |
| && !SYMBOL_REF_TLS_MODEL (base) |
| && !gprel_constant_p (base) |
| && !r0rel_constant_p (base) |
| && SMALL_INT (INTVAL (offset))); |
| } |
| return false; |
| } |
| |
| /* Return true if X is an expression of the form |
| (PLUS reg large_constant). */ |
| static bool |
| nios2_plus_large_constant_p (rtx x) |
| { |
| return (GET_CODE (x) == PLUS |
| && REG_P (XEXP (x, 0)) |
| && nios2_large_constant_p (XEXP (x, 1))); |
| } |
| |
| /* Implement TARGET_LEGITIMATE_CONSTANT_P. */ |
| static bool |
| nios2_legitimate_constant_p (machine_mode mode ATTRIBUTE_UNUSED, rtx x) |
| { |
| rtx base, offset; |
| split_const (x, &base, &offset); |
| return GET_CODE (base) != SYMBOL_REF || !SYMBOL_REF_TLS_MODEL (base); |
| } |
| |
| /* Implement TARGET_CANNOT_FORCE_CONST_MEM. */ |
| static bool |
| nios2_cannot_force_const_mem (machine_mode mode ATTRIBUTE_UNUSED, rtx x) |
| { |
| return nios2_legitimate_constant_p (mode, x) == false; |
| } |
| |
| /* Return true if register REGNO is a valid base register. |
| STRICT_P is true if REG_OK_STRICT is in effect. */ |
| |
| bool |
| nios2_regno_ok_for_base_p (int regno, bool strict_p) |
| { |
| if (!HARD_REGISTER_NUM_P (regno)) |
| { |
| if (!strict_p) |
| return true; |
| |
| if (!reg_renumber) |
| return false; |
| |
| regno = reg_renumber[regno]; |
| } |
| |
| /* The fake registers will be eliminated to either the stack or |
| hard frame pointer, both of which are usually valid base registers. |
| Reload deals with the cases where the eliminated form isn't valid. */ |
| return (GP_REG_P (regno) |
| || regno == FRAME_POINTER_REGNUM |
| || regno == ARG_POINTER_REGNUM); |
| } |
| |
| /* Return true if OFFSET is permitted in a load/store address expression. |
| Normally any 16-bit value is permitted, but on R2 if we may be emitting |
| the IO forms of these instructions we must restrict the offset to fit |
| in a 12-bit field instead. */ |
| |
| static bool |
| nios2_valid_addr_offset_p (rtx offset) |
| { |
| return (CONST_INT_P (offset) |
| && ((TARGET_ARCH_R2 && (TARGET_BYPASS_CACHE |
| || TARGET_BYPASS_CACHE_VOLATILE)) |
| ? SMALL_INT12 (INTVAL (offset)) |
| : SMALL_INT (INTVAL (offset)))); |
| } |
| |
| /* Return true if the address expression formed by BASE + OFFSET is |
| valid. */ |
| static bool |
| nios2_valid_addr_expr_p (rtx base, rtx offset, bool strict_p) |
| { |
| if (!strict_p && GET_CODE (base) == SUBREG) |
| base = SUBREG_REG (base); |
| return (REG_P (base) |
| && nios2_regno_ok_for_base_p (REGNO (base), strict_p) |
| && (offset == NULL_RTX |
| || nios2_valid_addr_offset_p (offset) |
| || (nios2_large_constant_allowed () |
| && nios2_symbolic_constant_p (offset)) |
| || nios2_unspec_reloc_p (offset))); |
| } |
| |
| /* Implement TARGET_LEGITIMATE_ADDRESS_P. */ |
| static bool |
| nios2_legitimate_address_p (machine_mode mode ATTRIBUTE_UNUSED, |
| rtx operand, bool strict_p) |
| { |
| switch (GET_CODE (operand)) |
| { |
| /* Direct. */ |
| case SYMBOL_REF: |
| if (SYMBOL_REF_TLS_MODEL (operand)) |
| return false; |
| |
| /* Else, fall through. */ |
| case CONST: |
| if (gprel_constant_p (operand) || r0rel_constant_p (operand)) |
| return true; |
| |
| /* Else, fall through. */ |
| case LABEL_REF: |
| if (nios2_large_constant_allowed () |
| && nios2_symbolic_constant_p (operand)) |
| return true; |
| return false; |
| |
| case CONST_INT: |
| if (r0rel_constant_p (operand)) |
| return true; |
| return nios2_large_constant_allowed (); |
| |
| case CONST_DOUBLE: |
| return false; |
| |
| /* Register indirect. */ |
| case REG: |
| return nios2_regno_ok_for_base_p (REGNO (operand), strict_p); |
| |
| /* Register indirect with displacement. */ |
| case PLUS: |
| { |
| rtx op0 = XEXP (operand, 0); |
| rtx op1 = XEXP (operand, 1); |
| |
| if (nios2_valid_addr_expr_p (op0, op1, strict_p) |
| || nios2_valid_addr_expr_p (op1, op0, strict_p)) |
| return true; |
| } |
| break; |
| |
| /* %lo(constant)(reg) |
| This requires a 16-bit relocation and isn't valid with R2 |
| io-variant load/stores. */ |
| case LO_SUM: |
| if (TARGET_ARCH_R2 |
| && (TARGET_BYPASS_CACHE || TARGET_BYPASS_CACHE_VOLATILE)) |
| return false; |
| else |
| { |
| rtx op0 = XEXP (operand, 0); |
| rtx op1 = XEXP (operand, 1); |
| |
| return (REG_P (op0) |
| && nios2_regno_ok_for_base_p (REGNO (op0), strict_p) |
| && nios2_large_constant_p (op1)); |
| } |
| |
| default: |
| break; |
| } |
| return false; |
| } |
| |
| /* Implement TARGET_ADDRESS_COST. |
| Experimentation has shown that we get better code by penalizing the |
| the (plus reg symbolic_constant) and (plus reg (const ...)) forms |
| but giving (plus reg symbol_ref) address modes the same cost as those |
| that don't require splitting. Also, from a theoretical point of view: |
| - This is in line with the recommendation in the GCC internals |
| documentation to make address forms involving multiple |
| registers more expensive than single-register forms. |
| - OTOH it still encourages fwprop1 to propagate constants into |
| address expressions more aggressively. |
| - We should discourage splitting (symbol + offset) into hi/lo pairs |
| to allow CSE'ing the symbol when it's used with more than one offset, |
| but not so heavily as to avoid this addressing mode at all. */ |
| static int |
| nios2_address_cost (rtx address, |
| machine_mode mode ATTRIBUTE_UNUSED, |
| addr_space_t as ATTRIBUTE_UNUSED, |
| bool speed ATTRIBUTE_UNUSED) |
| { |
| if (nios2_plus_large_constant_p (address)) |
| return COSTS_N_INSNS (1); |
| if (nios2_large_constant_p (address)) |
| { |
| if (GET_CODE (address) == CONST) |
| return COSTS_N_INSNS (1); |
| else |
| return COSTS_N_INSNS (0); |
| } |
| return COSTS_N_INSNS (0); |
| } |
| |
| /* Return true if X is a MEM whose address expression involves a large (32-bit) |
| constant. */ |
| bool |
| nios2_large_constant_memory_operand_p (rtx x) |
| { |
| rtx addr; |
| |
| if (GET_CODE (x) != MEM) |
| return false; |
| addr = XEXP (x, 0); |
| |
| return (nios2_large_constant_p (addr) |
| || nios2_plus_large_constant_p (addr)); |
| } |
| |
| |
| /* Return true if X is something that needs to be split into a |
| high/lo_sum pair. */ |
| bool |
| nios2_large_constant_p (rtx x) |
| { |
| return (nios2_symbolic_constant_p (x) |
| || nios2_large_unspec_reloc_p (x) |
| || (CONST_INT_P (x) && !SMALL_INT (INTVAL (x)))); |
| } |
| |
| /* Given an RTX X that satisfies nios2_large_constant_p, split it into |
| high and lo_sum parts using TEMP as a scratch register. Emit the high |
| instruction and return the lo_sum expression. |
| Also handle special cases involving constant integers. */ |
| rtx |
| nios2_split_large_constant (rtx x, rtx temp) |
| { |
| if (CONST_INT_P (x)) |
| { |
| HOST_WIDE_INT val = INTVAL (x); |
| if (SMALL_INT (val)) |
| return x; |
| else if (SMALL_INT_UNSIGNED (val) || UPPER16_INT (val)) |
| { |
| emit_move_insn (temp, x); |
| return temp; |
| } |
| else |
| { |
| HOST_WIDE_INT high = (val + 0x8000) & ~0xffff; |
| HOST_WIDE_INT low = val - high; |
| emit_move_insn (temp, gen_int_mode (high, Pmode)); |
| return gen_rtx_PLUS (Pmode, temp, gen_int_mode (low, Pmode)); |
| } |
| } |
| |
| emit_insn (gen_rtx_SET (temp, gen_rtx_HIGH (Pmode, copy_rtx (x)))); |
| return gen_rtx_LO_SUM (Pmode, temp, copy_rtx (x)); |
| } |
| |
| /* Split an RTX of the form |
| (plus op0 op1) |
| where op1 is a large constant into |
| (set temp (high op1)) |
| (set temp (plus op0 temp)) |
| (lo_sum temp op1) |
| returning the lo_sum expression as the value. */ |
| static rtx |
| nios2_split_plus_large_constant (rtx op0, rtx op1) |
| { |
| rtx temp = gen_reg_rtx (Pmode); |
| op0 = force_reg (Pmode, op0); |
| |
| emit_insn (gen_rtx_SET (temp, gen_rtx_HIGH (Pmode, copy_rtx (op1)))); |
| emit_insn (gen_rtx_SET (temp, gen_rtx_PLUS (Pmode, op0, temp))); |
| return gen_rtx_LO_SUM (Pmode, temp, copy_rtx (op1)); |
| } |
| |
| /* Given a MEM OP with an address that includes a splittable symbol or |
| other large constant, emit some instructions to do the split and |
| return a new MEM. */ |
| rtx |
| nios2_split_large_constant_memory_operand (rtx op) |
| { |
| rtx addr = XEXP (op, 0); |
| |
| if (nios2_large_constant_p (addr)) |
| addr = nios2_split_large_constant (addr, gen_reg_rtx (Pmode)); |
| else if (nios2_plus_large_constant_p (addr)) |
| addr = nios2_split_plus_large_constant (XEXP (addr, 0), XEXP (addr, 1)); |
| else |
| gcc_unreachable (); |
| return replace_equiv_address (op, addr, false); |
| } |
| |
| /* Return true if SECTION is a small section name. */ |
| static bool |
| nios2_small_section_name_p (const char *section) |
| { |
| return (strcmp (section, ".sbss") == 0 |
| || startswith (section, ".sbss.") |
| || strcmp (section, ".sdata") == 0 |
| || startswith (section, ".sdata.") |
| || (nios2_gprel_sec |
| && regexec (&nios2_gprel_sec_regex, section, 0, NULL, 0) == 0)); |
| } |
| |
| /* Return true if SECTION is a r0-relative section name. */ |
| static bool |
| nios2_r0rel_section_name_p (const char *section) |
| { |
| return (nios2_r0rel_sec |
| && regexec (&nios2_r0rel_sec_regex, section, 0, NULL, 0) == 0); |
| } |
| |
| /* Return true if EXP should be placed in the small data section. */ |
| static bool |
| nios2_in_small_data_p (const_tree exp) |
| { |
| /* We want to merge strings, so we never consider them small data. */ |
| if (TREE_CODE (exp) == STRING_CST) |
| return false; |
| |
| if (TREE_CODE (exp) == VAR_DECL) |
| { |
| if (DECL_SECTION_NAME (exp)) |
| { |
| const char *section = DECL_SECTION_NAME (exp); |
| if (nios2_small_section_name_p (section)) |
| return true; |
| } |
| else if (flexible_array_type_p (TREE_TYPE (exp)) |
| && (!TREE_PUBLIC (exp) || DECL_EXTERNAL (exp))) |
| { |
| /* We really should not consider any objects of any flexibly-sized |
| type to be small data, but pre-GCC 10 did not test |
| for this and just fell through to the next case. Thus older |
| code compiled with -mgpopt=global could contain GP-relative |
| accesses to objects defined in this compilation unit with |
| external linkage. We retain the possible small-data treatment |
| of such definitions for backward ABI compatibility, but |
| no longer generate GP-relative accesses for external |
| references (so that the ABI could be changed in the future |
| with less potential impact), or objects with internal |
| linkage. */ |
| return false; |
| } |
| else |
| { |
| HOST_WIDE_INT size = int_size_in_bytes (TREE_TYPE (exp)); |
| |
| /* If this is an incomplete type with size 0, then we can't put it |
| in sdata because it might be too big when completed. */ |
| if (size > 0 |
| && (unsigned HOST_WIDE_INT) size <= nios2_section_threshold) |
| return true; |
| } |
| } |
| |
| return false; |
| } |
| |
| /* Return true if symbol is in small data section. */ |
| |
| static bool |
| nios2_symbol_ref_in_small_data_p (rtx sym) |
| { |
| tree decl; |
| |
| gcc_assert (GET_CODE (sym) == SYMBOL_REF); |
| decl = SYMBOL_REF_DECL (sym); |
| |
| /* TLS variables are not accessed through the GP. */ |
| if (SYMBOL_REF_TLS_MODEL (sym) != 0) |
| return false; |
| |
| /* On Nios II R2, there is no GP-relative relocation that can be |
| used with "io" instructions. So, if we are implicitly generating |
| those instructions, we cannot emit GP-relative accesses. */ |
| if (TARGET_ARCH_R2 |
| && (TARGET_BYPASS_CACHE || TARGET_BYPASS_CACHE_VOLATILE)) |
| return false; |
| |
| /* If the user has explicitly placed the symbol in a small data section |
| via an attribute, generate gp-relative addressing even if the symbol |
| is external, weak, or larger than we'd automatically put in the |
| small data section. OTOH, if the symbol is located in some |
| non-small-data section, we can't use gp-relative accesses on it |
| unless the user has requested gpopt_data or gpopt_all. */ |
| |
| switch (nios2_gpopt_option) |
| { |
| case gpopt_none: |
| /* Don't generate a gp-relative addressing mode if that's been |
| disabled. */ |
| return false; |
| |
| case gpopt_local: |
| /* Use GP-relative addressing for small data symbols that are |
| not external or weak or uninitialized common, plus any symbols |
| that have explicitly been placed in a small data section. */ |
| if (decl && DECL_SECTION_NAME (decl)) |
| return nios2_small_section_name_p (DECL_SECTION_NAME (decl)); |
| return (SYMBOL_REF_SMALL_P (sym) |
| && !SYMBOL_REF_EXTERNAL_P (sym) |
| && !(decl && DECL_WEAK (decl)) |
| && !(decl && DECL_COMMON (decl) |
| && (DECL_INITIAL (decl) == NULL |
| || (!in_lto_p |
| && DECL_INITIAL (decl) == error_mark_node)))); |
| |
| case gpopt_global: |
| /* Use GP-relative addressing for small data symbols, even if |
| they are external or weak. Note that SYMBOL_REF_SMALL_P |
| is also true of symbols that have explicitly been placed |
| in a small data section. */ |
| return SYMBOL_REF_SMALL_P (sym); |
| |
| case gpopt_data: |
| /* Use GP-relative addressing for all data symbols regardless |
| of the object size, but not for code symbols. This option |
| is equivalent to the user asserting that the entire data |
| section is accessible from the GP. */ |
| return !SYMBOL_REF_FUNCTION_P (sym); |
| |
| case gpopt_all: |
| /* Use GP-relative addressing for everything, including code. |
| Effectively, the user has asserted that the entire program |
| fits within the 64K range of the GP offset. */ |
| return true; |
| |
| default: |
| /* We shouldn't get here. */ |
| return false; |
| } |
| } |
| |
| /* Likewise for r0-relative addressing. */ |
| static bool |
| nios2_symbol_ref_in_r0rel_data_p (rtx sym) |
| { |
| tree decl; |
| |
| gcc_assert (GET_CODE (sym) == SYMBOL_REF); |
| decl = SYMBOL_REF_DECL (sym); |
| |
| /* TLS variables are not accessed through r0. */ |
| if (SYMBOL_REF_TLS_MODEL (sym) != 0) |
| return false; |
| |
| /* On Nios II R2, there is no r0-relative relocation that can be |
| used with "io" instructions. So, if we are implicitly generating |
| those instructions, we cannot emit r0-relative accesses. */ |
| if (TARGET_ARCH_R2 |
| && (TARGET_BYPASS_CACHE || TARGET_BYPASS_CACHE_VOLATILE)) |
| return false; |
| |
| /* If the user has explicitly placed the symbol in a r0rel section |
| via an attribute, generate r0-relative addressing. */ |
| if (decl && DECL_SECTION_NAME (decl)) |
| return nios2_r0rel_section_name_p (DECL_SECTION_NAME (decl)); |
| return false; |
| } |
| |
| /* Implement TARGET_SECTION_TYPE_FLAGS. */ |
| |
| static unsigned int |
| nios2_section_type_flags (tree decl, const char *name, int reloc) |
| { |
| unsigned int flags; |
| |
| flags = default_section_type_flags (decl, name, reloc); |
| |
| if (nios2_small_section_name_p (name)) |
| flags |= SECTION_SMALL; |
| |
| return flags; |
| } |
| |
| /* Return true if SYMBOL_REF X binds locally. */ |
| |
| static bool |
| nios2_symbol_binds_local_p (const_rtx x) |
| { |
| return (SYMBOL_REF_DECL (x) |
| ? targetm.binds_local_p (SYMBOL_REF_DECL (x)) |
| : SYMBOL_REF_LOCAL_P (x)); |
| } |
| |
| /* Position independent code related. */ |
| |
| /* Emit code to load the PIC register. */ |
| static void |
| nios2_load_pic_register (void) |
| { |
| rtx tmp = gen_rtx_REG (Pmode, TEMP_REG_NUM); |
| |
| emit_insn (gen_load_got_register (pic_offset_table_rtx, tmp)); |
| emit_insn (gen_add3_insn (pic_offset_table_rtx, pic_offset_table_rtx, tmp)); |
| } |
| |
| /* Generate a PIC address as a MEM rtx. */ |
| static rtx |
| nios2_load_pic_address (rtx sym, int unspec, rtx tmp) |
| { |
| if (flag_pic == 2 |
| && GET_CODE (sym) == SYMBOL_REF |
| && nios2_symbol_binds_local_p (sym)) |
| /* Under -fPIC, generate a GOTOFF address for local symbols. */ |
| { |
| rtx offset = nios2_unspec_offset (sym, UNSPEC_PIC_GOTOFF_SYM); |
| crtl->uses_pic_offset_table = 1; |
| return nios2_large_got_address (offset, tmp); |
| } |
| |
| return gen_const_mem (Pmode, nios2_got_address (sym, unspec)); |
| } |
| |
| /* Nonzero if the constant value X is a legitimate general operand |
| when generating PIC code. It is given that flag_pic is on and |
| that X satisfies CONSTANT_P or is a CONST_DOUBLE. */ |
| bool |
| nios2_legitimate_pic_operand_p (rtx x) |
| { |
| if (nios2_large_unspec_reloc_p (x)) |
| return true; |
| |
| return ! (GET_CODE (x) == SYMBOL_REF |
| || GET_CODE (x) == LABEL_REF || GET_CODE (x) == CONST); |
| } |
| |
| /* Return TRUE if X is a thread-local symbol. */ |
| static bool |
| nios2_tls_symbol_p (rtx x) |
| { |
| return (targetm.have_tls && GET_CODE (x) == SYMBOL_REF |
| && SYMBOL_REF_TLS_MODEL (x) != 0); |
| } |
| |
| /* Legitimize addresses that are CONSTANT_P expressions. */ |
| static rtx |
| nios2_legitimize_constant_address (rtx addr) |
| { |
| rtx base, offset; |
| split_const (addr, &base, &offset); |
| |
| if (nios2_tls_symbol_p (base)) |
| base = nios2_legitimize_tls_address (base); |
| else if (flag_pic) |
| base = nios2_load_pic_address (base, UNSPEC_PIC_SYM, NULL_RTX); |
| else if (!nios2_large_constant_allowed () |
| && nios2_symbolic_constant_p (addr)) |
| return nios2_split_large_constant (addr, gen_reg_rtx (Pmode)); |
| else if (CONST_INT_P (addr)) |
| { |
| HOST_WIDE_INT val = INTVAL (addr); |
| if (SMALL_INT (val)) |
| /* Use r0-relative addressing. */ |
| return addr; |
| else if (!nios2_large_constant_allowed ()) |
| /* Split into high/lo pair. */ |
| return nios2_split_large_constant (addr, gen_reg_rtx (Pmode)); |
| } |
| else |
| return addr; |
| |
| if (offset != const0_rtx) |
| { |
| gcc_assert (can_create_pseudo_p ()); |
| return gen_rtx_PLUS (Pmode, force_reg (Pmode, base), |
| (CONST_INT_P (offset) |
| ? (SMALL_INT (INTVAL (offset)) |
| ? offset : force_reg (Pmode, offset)) |
| : offset)); |
| } |
| return base; |
| } |
| |
| /* Implement TARGET_LEGITIMIZE_ADDRESS. */ |
| static rtx |
| nios2_legitimize_address (rtx x, rtx oldx ATTRIBUTE_UNUSED, |
| machine_mode mode ATTRIBUTE_UNUSED) |
| { |
| rtx op0, op1; |
| |
| if (CONSTANT_P (x)) |
| return nios2_legitimize_constant_address (x); |
| |
| /* Remaining cases all involve something + a constant. */ |
| if (GET_CODE (x) != PLUS) |
| return x; |
| |
| op0 = XEXP (x, 0); |
| op1 = XEXP (x, 1); |
| |
| /* Target-independent code turns (exp + constant) into plain |
| register indirect. Although subsequent optimization passes will |
| eventually sort that out, ivopts uses the unoptimized form for |
| computing its cost model, so we get better results by generating |
| the correct form from the start. */ |
| if (nios2_valid_addr_offset_p (op1)) |
| return gen_rtx_PLUS (Pmode, force_reg (Pmode, op0), copy_rtx (op1)); |
| |
| /* We may need to split symbolic constants now. */ |
| else if (nios2_symbolic_constant_p (op1)) |
| { |
| if (nios2_large_constant_allowed ()) |
| return gen_rtx_PLUS (Pmode, force_reg (Pmode, op0), copy_rtx (op1)); |
| else |
| return nios2_split_plus_large_constant (op0, op1); |
| } |
| |
| /* For the TLS LE (Local Exec) model, the compiler may try to |
| combine constant offsets with unspec relocs, creating address RTXs |
| looking like this: |
| (plus:SI (reg:SI 23 r23) |
| (const:SI |
| (plus:SI |
| (unspec:SI [(symbol_ref:SI ("var"))] UNSPEC_ADD_TLS_LE) |
| (const_int 48 [0x30])))) |
| |
| This usually happens when 'var' is a thread-local struct variable, |
| and access of a field in var causes the addend. |
| |
| We typically want this combining, so transform the above into this |
| form, which is allowed: |
| (plus:SI (reg:SI 23 r23) |
| (const:SI |
| (unspec:SI |
| [(const:SI |
| (plus:SI (symbol_ref:SI ("var")) |
| (const_int 48 [0x30])))] UNSPEC_ADD_TLS_LE))) |
| |
| Which will be output as '%tls_le(var+48)(r23)' in assembly. */ |
| else if (GET_CODE (op1) == CONST) |
| { |
| rtx unspec, offset; |
| split_const (op1, &unspec, &offset); |
| if (GET_CODE (unspec) == UNSPEC |
| && !nios2_large_offset_p (XINT (unspec, 1)) |
| && offset != const0_rtx) |
| { |
| rtx reg = force_reg (Pmode, op0); |
| unspec = copy_rtx (unspec); |
| XVECEXP (unspec, 0, 0) |
| = plus_constant (Pmode, XVECEXP (unspec, 0, 0), INTVAL (offset)); |
| return gen_rtx_PLUS (Pmode, reg, gen_rtx_CONST (Pmode, unspec)); |
| } |
| } |
| |
| return x; |
| } |
| |
| static rtx |
| nios2_delegitimize_address (rtx x) |
| { |
| x = delegitimize_mem_from_attrs (x); |
| |
| if (GET_CODE (x) == CONST && GET_CODE (XEXP (x, 0)) == UNSPEC) |
| { |
| switch (XINT (XEXP (x, 0), 1)) |
| { |
| case UNSPEC_PIC_SYM: |
| case UNSPEC_PIC_CALL_SYM: |
| case UNSPEC_PIC_GOTOFF_SYM: |
| case UNSPEC_ADD_TLS_GD: |
| case UNSPEC_ADD_TLS_LDM: |
| case UNSPEC_LOAD_TLS_IE: |
| case UNSPEC_ADD_TLS_LE: |
| x = XVECEXP (XEXP (x, 0), 0, 0); |
| gcc_assert (CONSTANT_P (x)); |
| break; |
| } |
| } |
| return x; |
| } |
| |
| /* Main expander function for RTL moves. */ |
| bool |
| nios2_emit_move_sequence (rtx *operands, machine_mode mode) |
| { |
| rtx to = operands[0]; |
| rtx from = operands[1]; |
| |
| if (!register_operand (to, mode) && !reg_or_0_operand (from, mode)) |
| { |
| gcc_assert (can_create_pseudo_p ()); |
| from = copy_to_mode_reg (mode, from); |
| } |
| |
| if (CONSTANT_P (from)) |
| { |
| if (CONST_INT_P (from)) |
| { |
| if (!SMALL_INT (INTVAL (from)) |
| && !SMALL_INT_UNSIGNED (INTVAL (from)) |
| && !UPPER16_INT (INTVAL (from))) |
| { |
| HOST_WIDE_INT high = (INTVAL (from) + 0x8000) & ~0xffff; |
| HOST_WIDE_INT low = INTVAL (from) & 0xffff; |
| emit_move_insn (to, gen_int_mode (high, SImode)); |
| emit_insn (gen_add2_insn (to, gen_int_mode (low, HImode))); |
| set_unique_reg_note (get_last_insn (), REG_EQUAL, |
| copy_rtx (from)); |
| return true; |
| } |
| } |
| else if (gprel_constant_p (from) || r0rel_constant_p (from)) |
| /* Handled directly by movsi_internal as gp + offset |
| or r0 + offset. */ |
| ; |
| else if (nios2_large_constant_p (from)) |
| /* This case covers either a regular symbol reference or an UNSPEC |
| representing a 32-bit offset. We split the former |
| only conditionally and the latter always. */ |
| { |
| if (!nios2_large_constant_allowed () |
| || nios2_large_unspec_reloc_p (from)) |
| { |
| rtx lo = nios2_split_large_constant (from, to); |
| emit_insn (gen_rtx_SET (to, lo)); |
| set_unique_reg_note (get_last_insn (), REG_EQUAL, |
| copy_rtx (operands[1])); |
| return true; |
| } |
| } |
| else |
| /* This is a TLS or PIC symbol. */ |
| { |
| from = nios2_legitimize_constant_address (from); |
| if (CONSTANT_P (from)) |
| { |
| emit_insn (gen_rtx_SET (to, |
| gen_rtx_HIGH (Pmode, copy_rtx (from)))); |
| emit_insn (gen_rtx_SET (to, gen_rtx_LO_SUM (Pmode, to, from))); |
| set_unique_reg_note (get_last_insn (), REG_EQUAL, |
| copy_rtx (operands[1])); |
| return true; |
| } |
| } |
| } |
| |
| operands[0] = to; |
| operands[1] = from; |
| return false; |
| } |
| |
| /* The function with address *ADDR is being called. If the address |
| needs to be loaded from the GOT, emit the instruction to do so and |
| update *ADDR to point to the rtx for the loaded value. |
| If REG != NULL_RTX, it is used as the target/scratch register in the |
| GOT address calculation. */ |
| void |
| nios2_adjust_call_address (rtx *call_op, rtx reg) |
| { |
| if (MEM_P (*call_op)) |
| call_op = &XEXP (*call_op, 0); |
| |
| rtx addr = *call_op; |
| if (flag_pic && CONSTANT_P (addr)) |
| { |
| rtx tmp = reg ? reg : NULL_RTX; |
| if (!reg) |
| reg = gen_reg_rtx (Pmode); |
| addr = nios2_load_pic_address (addr, UNSPEC_PIC_CALL_SYM, tmp); |
| emit_insn (gen_rtx_SET (reg, addr)); |
| *call_op = reg; |
| } |
| } |
| |
| |
| /* Output assembly language related definitions. */ |
| |
| /* Implement TARGET_PRINT_OPERAND_PUNCT_VALID_P. */ |
| static bool |
| nios2_print_operand_punct_valid_p (unsigned char code) |
| { |
| return (code == '.' || code == '!'); |
| } |
| |
| |
| /* Print the operand OP to file stream FILE modified by LETTER. |
| LETTER can be one of: |
| |
| i: print i/hi/ui suffixes (used for mov instruction variants), |
| when OP is the appropriate immediate operand. |
| |
| u: like 'i', except without "ui" suffix case (used for cmpgeu/cmpltu) |
| |
| o: print "io" if OP needs volatile access (due to TARGET_BYPASS_CACHE |
| or TARGET_BYPASS_CACHE_VOLATILE). |
| |
| x: print i/hi/ci/chi suffixes for the and instruction, |
| when OP is the appropriate immediate operand. |
| |
| z: prints the third register immediate operand in assembly |
| instructions. Outputs const0_rtx as the 'zero' register |
| instead of '0'. |
| |
| y: same as 'z', but for specifically for logical instructions, |
| where the processing for immediates are slightly different. |
| |
| H: for %hiadj |
| L: for %lo |
| D: for the upper 32-bits of a 64-bit double value |
| R: prints reverse condition. |
| A: prints (reg) operand for ld[s]ex and st[s]ex. |
| |
| .: print .n suffix for 16-bit instructions. |
| !: print r.n suffix for 16-bit instructions. Used for jmpr.n. |
| */ |
| static void |
| nios2_print_operand (FILE *file, rtx op, int letter) |
| { |
| |
| /* First take care of the format letters that just insert a string |
| into the output stream. */ |
| switch (letter) |
| { |
| case '.': |
| if (current_output_insn && get_attr_length (current_output_insn) == 2) |
| fprintf (file, ".n"); |
| return; |
| |
| case '!': |
| if (current_output_insn && get_attr_length (current_output_insn) == 2) |
| fprintf (file, "r.n"); |
| return; |
| |
| case 'x': |
| if (CONST_INT_P (op)) |
| { |
| HOST_WIDE_INT val = INTVAL (op); |
| HOST_WIDE_INT low = val & 0xffff; |
| HOST_WIDE_INT high = (val >> 16) & 0xffff; |
| |
| if (val != 0) |
| { |
| if (high != 0) |
| { |
| if (low != 0) |
| { |
| gcc_assert (TARGET_ARCH_R2); |
| if (high == 0xffff) |
| fprintf (file, "c"); |
| else if (low == 0xffff) |
| fprintf (file, "ch"); |
| else |
| gcc_unreachable (); |
| } |
| else |
| fprintf (file, "h"); |
| } |
| fprintf (file, "i"); |
| } |
| } |
| return; |
| |
| case 'u': |
| case 'i': |
| if (CONST_INT_P (op)) |
| { |
| HOST_WIDE_INT val = INTVAL (op); |
| HOST_WIDE_INT low = val & 0xffff; |
| HOST_WIDE_INT high = (val >> 16) & 0xffff; |
| if (val != 0) |
| { |
| if (low == 0 && high != 0) |
| fprintf (file, "h"); |
| else if (high == 0 && (low & 0x8000) != 0 && letter != 'u') |
| fprintf (file, "u"); |
| } |
| } |
| if (CONSTANT_P (op) && op != const0_rtx) |
| fprintf (file, "i"); |
| return; |
| |
| case 'o': |
| if (GET_CODE (op) == MEM |
| && ((MEM_VOLATILE_P (op) && TARGET_BYPASS_CACHE_VOLATILE) |
| || TARGET_BYPASS_CACHE)) |
| { |
| gcc_assert (current_output_insn |
| && get_attr_length (current_output_insn) == 4); |
| fprintf (file, "io"); |
| } |
| return; |
| |
| default: |
| break; |
| } |
| |
| /* Handle comparison operator names. */ |
| if (comparison_operator (op, VOIDmode)) |
| { |
| enum rtx_code cond = GET_CODE (op); |
| if (letter == 0) |
| { |
| fprintf (file, "%s", GET_RTX_NAME (cond)); |
| return; |
| } |
| if (letter == 'R') |
| { |
| fprintf (file, "%s", GET_RTX_NAME (reverse_condition (cond))); |
| return; |
| } |
| } |
| |
| /* Now handle the cases where we actually need to format an operand. */ |
| switch (GET_CODE (op)) |
| { |
| case REG: |
| if (letter == 0 || letter == 'z' || letter == 'y') |
| { |
| fprintf (file, "%s", reg_names[REGNO (op)]); |
| return; |
| } |
| else if (letter == 'D') |
| { |
| fprintf (file, "%s", reg_names[REGNO (op)+1]); |
| return; |
| } |
| break; |
| |
| case CONST_INT: |
| { |
| rtx int_rtx = op; |
| HOST_WIDE_INT val = INTVAL (int_rtx); |
| HOST_WIDE_INT low = val & 0xffff; |
| HOST_WIDE_INT high = (val >> 16) & 0xffff; |
| |
| if (letter == 'y') |
| { |
| if (val == 0) |
| fprintf (file, "zero"); |
| else |
| { |
| if (high != 0) |
| { |
| if (low != 0) |
| { |
| gcc_assert (TARGET_ARCH_R2); |
| if (high == 0xffff) |
| /* andci. */ |
| int_rtx = gen_int_mode (low, SImode); |
| else if (low == 0xffff) |
| /* andchi. */ |
| int_rtx = gen_int_mode (high, SImode); |
| else |
| gcc_unreachable (); |
| } |
| else |
| /* andhi. */ |
| int_rtx = gen_int_mode (high, SImode); |
| } |
| else |
| /* andi. */ |
| int_rtx = gen_int_mode (low, SImode); |
| output_addr_const (file, int_rtx); |
| } |
| return; |
| } |
| else if (letter == 'z') |
| { |
| if (val == 0) |
| fprintf (file, "zero"); |
| else |
| { |
| if (low == 0 && high != 0) |
| int_rtx = gen_int_mode (high, SImode); |
| else if (low != 0) |
| { |
| gcc_assert (high == 0 || high == 0xffff); |
| int_rtx = gen_int_mode (low, high == 0 ? SImode : HImode); |
| } |
| else |
| gcc_unreachable (); |
| output_addr_const (file, int_rtx); |
| } |
| return; |
| } |
| } |
| |
| /* Else, fall through. */ |
| |
| case CONST: |
| case LABEL_REF: |
| case SYMBOL_REF: |
| case CONST_DOUBLE: |
| if (letter == 0 || letter == 'z') |
| { |
| output_addr_const (file, op); |
| return; |
| } |
| else if (letter == 'H' || letter == 'L') |
| { |
| fprintf (file, "%%"); |
| if (GET_CODE (op) == CONST |
| && GET_CODE (XEXP (op, 0)) == UNSPEC) |
| { |
| rtx unspec = XEXP (op, 0); |
| int unspec_reloc = XINT (unspec, 1); |
| gcc_assert (nios2_large_offset_p (unspec_reloc)); |
| fprintf (file, "%s_", nios2_unspec_reloc_name (unspec_reloc)); |
| op = XVECEXP (unspec, 0, 0); |
| } |
| fprintf (file, letter == 'H' ? "hiadj(" : "lo("); |
| output_addr_const (file, op); |
| fprintf (file, ")"); |
| return; |
| } |
| break; |
| |
| case SUBREG: |
| case MEM: |
| if (letter == 'A') |
| { |
| /* Address of '(reg)' form, with no index. */ |
| fprintf (file, "(%s)", reg_names[REGNO (XEXP (op, 0))]); |
| return; |
| } |
| if (letter == 0) |
| { |
| output_address (VOIDmode, op); |
| return; |
| } |
| break; |
| |
| case CODE_LABEL: |
| if (letter == 0) |
| { |
| output_addr_const (file, op); |
| return; |
| } |
| break; |
| |
| default: |
| break; |
| } |
| |
| debug_rtx (op); |
| output_operand_lossage ("Unsupported operand for code '%c'", letter); |
| gcc_unreachable (); |
| } |
| |
| /* Return true if this is a GP-relative accessible reference. */ |
| bool |
| gprel_constant_p (rtx op) |
| { |
| if (GET_CODE (op) == SYMBOL_REF |
| && nios2_symbol_ref_in_small_data_p (op)) |
| return true; |
| else if (GET_CODE (op) == CONST |
| && GET_CODE (XEXP (op, 0)) == PLUS) |
| return gprel_constant_p (XEXP (XEXP (op, 0), 0)); |
| |
| return false; |
| } |
| |
| /* Likewise if this is a zero-relative accessible reference. */ |
| bool |
| r0rel_constant_p (rtx op) |
| { |
| if (GET_CODE (op) == SYMBOL_REF |
| && nios2_symbol_ref_in_r0rel_data_p (op)) |
| return true; |
| else if (GET_CODE (op) == CONST |
| && GET_CODE |