/* Subroutines used for code generation on IBM S/390 and zSeries
- Copyright (C) 1999-2015 Free Software Foundation, Inc.
+ Copyright (C) 1999-2016 Free Software Foundation, Inc.
Contributed by Hartmut Penner (hpenner@de.ibm.com) and
Ulrich Weigand (uweigand@de.ibm.com) and
Andreas Krebbel (Andreas.Krebbel@de.ibm.com).
#include "config.h"
#include "system.h"
#include "coretypes.h"
-#include "tm.h"
+#include "backend.h"
+#include "target.h"
+#include "target-globals.h"
#include "rtl.h"
-#include "hash-set.h"
-#include "machmode.h"
-#include "vec.h"
-#include "double-int.h"
-#include "input.h"
-#include "alias.h"
-#include "symtab.h"
-#include "wide-int.h"
-#include "inchash.h"
#include "tree.h"
+#include "gimple.h"
+#include "cfghooks.h"
+#include "cfgloop.h"
+#include "df.h"
+#include "tm_p.h"
+#include "stringpool.h"
+#include "expmed.h"
+#include "optabs.h"
+#include "regs.h"
+#include "emit-rtl.h"
+#include "recog.h"
+#include "cgraph.h"
+#include "diagnostic-core.h"
+#include "diagnostic.h"
+#include "alias.h"
#include "fold-const.h"
#include "print-tree.h"
-#include "stringpool.h"
#include "stor-layout.h"
#include "varasm.h"
#include "calls.h"
-#include "tm_p.h"
-#include "regs.h"
-#include "hard-reg-set.h"
-#include "insn-config.h"
#include "conditions.h"
#include "output.h"
#include "insn-attr.h"
#include "flags.h"
#include "except.h"
-#include "function.h"
-#include "recog.h"
-#include "hashtab.h"
-#include "statistics.h"
-#include "real.h"
-#include "fixed-value.h"
-#include "expmed.h"
#include "dojump.h"
#include "explow.h"
-#include "emit-rtl.h"
#include "stmt.h"
#include "expr.h"
#include "reload.h"
-#include "diagnostic-core.h"
-#include "predict.h"
-#include "dominance.h"
-#include "cfg.h"
#include "cfgrtl.h"
#include "cfganal.h"
#include "lcm.h"
#include "cfgbuild.h"
#include "cfgcleanup.h"
-#include "basic-block.h"
-#include "ggc.h"
-#include "target.h"
-#include "target-def.h"
#include "debug.h"
#include "langhooks.h"
-#include "insn-codes.h"
-#include "optabs.h"
-#include "hash-table.h"
-#include "tree-ssa-alias.h"
#include "internal-fn.h"
#include "gimple-fold.h"
#include "tree-eh.h"
-#include "gimple-expr.h"
-#include "is-a.h"
-#include "gimple.h"
#include "gimplify.h"
-#include "df.h"
#include "params.h"
-#include "cfgloop.h"
#include "opts.h"
#include "tree-pass.h"
#include "context.h"
#include "builtins.h"
#include "rtl-iter.h"
#include "intl.h"
-#include "plugin-api.h"
-#include "ipa-ref.h"
-#include "cgraph.h"
+#include "tm-constrs.h"
+
+/* This file should be included last. */
+#include "target-def.h"
+
+/* Remember the last target of s390_set_current_function. */
+static GTY(()) tree s390_previous_fndecl;
/* Define the specific costs for a given cpu. */
const int dsgr;
};
-const struct processor_costs *s390_cost;
+#define s390_cost ((const struct processor_costs *)(s390_cost_pointer))
static const
struct processor_costs z900_cost =
COSTS_N_INSNS (160), /* DSGR cracked */
};
+static struct
+{
+ const char *const name;
+ const enum processor_type processor;
+ const struct processor_costs *cost;
+}
+const processor_table[] =
+{
+ { "g5", PROCESSOR_9672_G5, &z900_cost },
+ { "g6", PROCESSOR_9672_G6, &z900_cost },
+ { "z900", PROCESSOR_2064_Z900, &z900_cost },
+ { "z990", PROCESSOR_2084_Z990, &z990_cost },
+ { "z9-109", PROCESSOR_2094_Z9_109, &z9_109_cost },
+ { "z9-ec", PROCESSOR_2094_Z9_EC, &z9_109_cost },
+ { "z10", PROCESSOR_2097_Z10, &z10_cost },
+ { "z196", PROCESSOR_2817_Z196, &z196_cost },
+ { "zEC12", PROCESSOR_2827_ZEC12, &zEC12_cost },
+ { "z13", PROCESSOR_2964_Z13, &zEC12_cost },
+ { "native", PROCESSOR_NATIVE, NULL }
+};
+
extern int reload_completed;
/* Kept up to date using the SCHED_VARIABLE_ISSUE hook. */
be saved to.
0 - does not need to be saved at all
-1 - stack slot */
+#define SAVE_SLOT_NONE 0
+#define SAVE_SLOT_STACK -1
signed char gpr_save_slots[16];
/* Number of first and last gpr to be saved, restored. */
/* True if the current function may contain a tbegin clobbering
FPRs. */
bool tbegin_p;
+
+ /* For -fsplit-stack support: A stack local which holds a pointer to
+ the stack arguments for a function with a variable number of
+ arguments. This is set at the start of the function and is used
+ to initialize the overflow_arg_area field of the va_list
+ structure. */
+ rtx split_stack_varargs_pointer;
};
/* Few accessor macros for struct cfun->machine->s390_frame_layout. */
#define PREDICT_DISTANCE (TARGET_Z10 ? 384 : 2048)
+/* Indicate which ABI has been used for passing vector args.
+ 0 - no vector type arguments have been passed where the ABI is relevant
+ 1 - the old ABI has been used
+ 2 - a vector type argument has been passed either in a vector register
+ or on the stack by value */
+static int s390_vector_abi = 0;
+
+/* Set the vector ABI marker if TYPE is subject to the vector ABI
+ switch. The vector ABI affects only vector data types. There are
+ two aspects of the vector ABI relevant here:
+
+ 1. vectors >= 16 bytes have an alignment of 8 bytes with the new
+ ABI and natural alignment with the old.
+
+ 2. vector <= 16 bytes are passed in VRs or by value on the stack
+ with the new ABI but by reference on the stack with the old.
+
+ If ARG_P is true TYPE is used for a function argument or return
+ value. The ABI marker then is set for all vector data types. If
+ ARG_P is false only type 1 vectors are being checked. */
+
+static void
+s390_check_type_for_vector_abi (const_tree type, bool arg_p, bool in_struct_p)
+{
+ static hash_set<const_tree> visited_types_hash;
+
+ if (s390_vector_abi)
+ return;
+
+ if (type == NULL_TREE || TREE_CODE (type) == ERROR_MARK)
+ return;
+
+ if (visited_types_hash.contains (type))
+ return;
+
+ visited_types_hash.add (type);
+
+ if (VECTOR_TYPE_P (type))
+ {
+ int type_size = int_size_in_bytes (type);
+
+ /* Outside arguments only the alignment is changing and this
+ only happens for vector types >= 16 bytes. */
+ if (!arg_p && type_size < 16)
+ return;
+
+ /* In arguments vector types > 16 are passed as before (GCC
+ never enforced the bigger alignment for arguments which was
+ required by the old vector ABI). However, it might still be
+ ABI relevant due to the changed alignment if it is a struct
+ member. */
+ if (arg_p && type_size > 16 && !in_struct_p)
+ return;
+
+ s390_vector_abi = TARGET_VX_ABI ? 2 : 1;
+ }
+ else if (POINTER_TYPE_P (type) || TREE_CODE (type) == ARRAY_TYPE)
+ {
+ /* ARRAY_TYPE: Since with neither of the ABIs we have more than
+ natural alignment there will never be ABI dependent padding
+ in an array type. That's why we do not set in_struct_p to
+ true here. */
+ s390_check_type_for_vector_abi (TREE_TYPE (type), arg_p, in_struct_p);
+ }
+ else if (TREE_CODE (type) == FUNCTION_TYPE || TREE_CODE (type) == METHOD_TYPE)
+ {
+ tree arg_chain;
+
+ /* Check the return type. */
+ s390_check_type_for_vector_abi (TREE_TYPE (type), true, false);
+
+ for (arg_chain = TYPE_ARG_TYPES (type);
+ arg_chain;
+ arg_chain = TREE_CHAIN (arg_chain))
+ s390_check_type_for_vector_abi (TREE_VALUE (arg_chain), true, false);
+ }
+ else if (RECORD_OR_UNION_TYPE_P (type))
+ {
+ tree field;
+
+ for (field = TYPE_FIELDS (type); field; field = DECL_CHAIN (field))
+ {
+ if (TREE_CODE (field) != FIELD_DECL)
+ continue;
+
+ s390_check_type_for_vector_abi (TREE_TYPE (field), arg_p, true);
+ }
+ }
+}
+
+
/* System z builtins. */
#include "s390-builtins.h"
-const unsigned int flags_builtin[S390_BUILTIN_MAX + 1] =
+const unsigned int bflags_builtin[S390_BUILTIN_MAX + 1] =
+ {
+#undef B_DEF
+#undef OB_DEF
+#undef OB_DEF_VAR
+#define B_DEF(NAME, PATTERN, ATTRS, BFLAGS, ...) BFLAGS,
+#define OB_DEF(...)
+#define OB_DEF_VAR(...)
+#include "s390-builtins.def"
+ 0
+ };
+
+const unsigned int opflags_builtin[S390_BUILTIN_MAX + 1] =
{
#undef B_DEF
#undef OB_DEF
#undef OB_DEF_VAR
-#define B_DEF(NAME, PATTERN, ATTRS, FLAGS, FNTYPE) FLAGS,
+#define B_DEF(NAME, PATTERN, ATTRS, BFLAGS, OPFLAGS, ...) OPFLAGS,
#define OB_DEF(...)
#define OB_DEF_VAR(...)
#include "s390-builtins.def"
0
};
-const unsigned int flags_overloaded_builtin_var[S390_OVERLOADED_BUILTIN_VAR_MAX + 1] =
+const unsigned int bflags_overloaded_builtin[S390_OVERLOADED_BUILTIN_MAX + 1] =
+ {
+#undef B_DEF
+#undef OB_DEF
+#undef OB_DEF_VAR
+#define B_DEF(...)
+#define OB_DEF(NAME, FIRST_VAR_NAME, LAST_VAR_NAME, BFLAGS, ...) BFLAGS,
+#define OB_DEF_VAR(...)
+#include "s390-builtins.def"
+ 0
+ };
+
+const unsigned int
+opflags_overloaded_builtin_var[S390_OVERLOADED_BUILTIN_VAR_MAX + 1] =
{
#undef B_DEF
#undef OB_DEF
c_uint64_type_node = long_long_unsigned_type_node;
#undef DEF_TYPE
-#define DEF_TYPE(INDEX, NODE, CONST_P) \
- s390_builtin_types[INDEX] = (!CONST_P) ? \
- (NODE) : build_type_variant ((NODE), 1, 0);
+#define DEF_TYPE(INDEX, BFLAGS, NODE, CONST_P) \
+ if (s390_builtin_types[INDEX] == NULL) \
+ s390_builtin_types[INDEX] = (!CONST_P) ? \
+ (NODE) : build_type_variant ((NODE), 1, 0);
#undef DEF_POINTER_TYPE
-#define DEF_POINTER_TYPE(INDEX, INDEX_BASE) \
- s390_builtin_types[INDEX] = \
- build_pointer_type (s390_builtin_types[INDEX_BASE]);
+#define DEF_POINTER_TYPE(INDEX, BFLAGS, INDEX_BASE) \
+ if (s390_builtin_types[INDEX] == NULL) \
+ s390_builtin_types[INDEX] = \
+ build_pointer_type (s390_builtin_types[INDEX_BASE]);
#undef DEF_DISTINCT_TYPE
-#define DEF_DISTINCT_TYPE(INDEX, INDEX_BASE) \
- s390_builtin_types[INDEX] = \
- build_distinct_type_copy (s390_builtin_types[INDEX_BASE]);
+#define DEF_DISTINCT_TYPE(INDEX, BFLAGS, INDEX_BASE) \
+ if (s390_builtin_types[INDEX] == NULL) \
+ s390_builtin_types[INDEX] = \
+ build_distinct_type_copy (s390_builtin_types[INDEX_BASE]);
#undef DEF_VECTOR_TYPE
-#define DEF_VECTOR_TYPE(INDEX, INDEX_BASE, ELEMENTS) \
- s390_builtin_types[INDEX] = \
- build_vector_type (s390_builtin_types[INDEX_BASE], ELEMENTS);
+#define DEF_VECTOR_TYPE(INDEX, BFLAGS, INDEX_BASE, ELEMENTS) \
+ if (s390_builtin_types[INDEX] == NULL) \
+ s390_builtin_types[INDEX] = \
+ build_vector_type (s390_builtin_types[INDEX_BASE], ELEMENTS);
#undef DEF_OPAQUE_VECTOR_TYPE
-#define DEF_OPAQUE_VECTOR_TYPE(INDEX, INDEX_BASE, ELEMENTS) \
- s390_builtin_types[INDEX] = \
- build_opaque_vector_type (s390_builtin_types[INDEX_BASE], ELEMENTS);
+#define DEF_OPAQUE_VECTOR_TYPE(INDEX, BFLAGS, INDEX_BASE, ELEMENTS) \
+ if (s390_builtin_types[INDEX] == NULL) \
+ s390_builtin_types[INDEX] = \
+ build_opaque_vector_type (s390_builtin_types[INDEX_BASE], ELEMENTS);
#undef DEF_FN_TYPE
-#define DEF_FN_TYPE(INDEX, args...) \
- s390_builtin_fn_types[INDEX] = \
- build_function_type_list (args, NULL_TREE);
+#define DEF_FN_TYPE(INDEX, BFLAGS, args...) \
+ if (s390_builtin_fn_types[INDEX] == NULL) \
+ s390_builtin_fn_types[INDEX] = \
+ build_function_type_list (args, NULL_TREE);
#undef DEF_OV_TYPE
#define DEF_OV_TYPE(...)
#include "s390-builtin-types.def"
#undef B_DEF
-#define B_DEF(NAME, PATTERN, ATTRS, FLAGS, FNTYPE) \
- s390_builtin_decls[S390_BUILTIN_##NAME] = \
- add_builtin_function ("__builtin_" #NAME, \
- s390_builtin_fn_types[FNTYPE], \
- S390_BUILTIN_##NAME, \
- BUILT_IN_MD, \
- NULL, \
- ATTRS);
+#define B_DEF(NAME, PATTERN, ATTRS, BFLAGS, OPFLAGS, FNTYPE) \
+ if (s390_builtin_decls[S390_BUILTIN_##NAME] == NULL) \
+ s390_builtin_decls[S390_BUILTIN_##NAME] = \
+ add_builtin_function ("__builtin_" #NAME, \
+ s390_builtin_fn_types[FNTYPE], \
+ S390_BUILTIN_##NAME, \
+ BUILT_IN_MD, \
+ NULL, \
+ ATTRS);
#undef OB_DEF
-#define OB_DEF(NAME, FIRST_VAR_NAME, LAST_VAR_NAME, FNTYPE) \
- s390_builtin_decls[S390_OVERLOADED_BUILTIN_##NAME + S390_BUILTIN_MAX] = \
- add_builtin_function ("__builtin_" #NAME, \
- s390_builtin_fn_types[FNTYPE], \
- S390_OVERLOADED_BUILTIN_##NAME + S390_BUILTIN_MAX, \
- BUILT_IN_MD, \
- NULL, \
- 0);
+#define OB_DEF(NAME, FIRST_VAR_NAME, LAST_VAR_NAME, BFLAGS, FNTYPE) \
+ if (s390_builtin_decls[S390_OVERLOADED_BUILTIN_##NAME + S390_BUILTIN_MAX] \
+ == NULL) \
+ s390_builtin_decls[S390_OVERLOADED_BUILTIN_##NAME + S390_BUILTIN_MAX] = \
+ add_builtin_function ("__builtin_" #NAME, \
+ s390_builtin_fn_types[FNTYPE], \
+ S390_OVERLOADED_BUILTIN_##NAME + S390_BUILTIN_MAX, \
+ BUILT_IN_MD, \
+ NULL, \
+ 0);
#undef OB_DEF_VAR
#define OB_DEF_VAR(...)
#include "s390-builtins.def"
HOST_WIDE_INT_PRINT_DEC ".."
HOST_WIDE_INT_PRINT_DEC ")",
argnum, decl,
- -(HOST_WIDE_INT)1 << (bitwidth - 1),
+ -((HOST_WIDE_INT)1 << (bitwidth - 1)),
((HOST_WIDE_INT)1 << (bitwidth - 1)) - 1);
return false;
}
bool nonvoid;
tree arg;
call_expr_arg_iterator iter;
- unsigned int all_op_flags = flags_for_builtin (fcode);
+ unsigned int all_op_flags = opflags_for_builtin (fcode);
machine_mode last_vec_mode = VOIDmode;
if (TARGET_DEBUG_ARG)
{
fprintf (stderr,
- "s390_expand_builtin, code = %4d, %s\n",
- (int)fcode, IDENTIFIER_POINTER (DECL_NAME (fndecl)));
+ "s390_expand_builtin, code = %4d, %s, bflags = 0x%x\n",
+ (int)fcode, IDENTIFIER_POINTER (DECL_NAME (fndecl)),
+ bflags_for_builtin (fcode));
}
+ if (S390_USE_TARGET_ATTRIBUTE)
+ {
+ unsigned int bflags;
+ bflags = bflags_for_builtin (fcode);
+ if ((bflags & B_HTM) && !TARGET_HTM)
+ {
+ error ("Builtin %qF is not supported without -mhtm "
+ "(default with -march=zEC12 and higher).", fndecl);
+ return const0_rtx;
+ }
+ if ((bflags & B_VX) && !TARGET_VX)
+ {
+ error ("Builtin %qF is not supported without -mvx "
+ "(default with -march=z13 and higher).", fndecl);
+ return const0_rtx;
+ }
+ }
if (fcode >= S390_OVERLOADED_BUILTIN_VAR_OFFSET
&& fcode < S390_ALL_BUILTIN_MAX)
{
insn_op = &insn_data[icode].operand[arity + nonvoid];
op[arity] = expand_expr (arg, NULL_RTX, insn_op->mode, EXPAND_NORMAL);
+ /* expand_expr truncates constants to the target mode only if it
+ is "convenient". However, our checks below rely on this
+ being done. */
+ if (CONST_INT_P (op[arity])
+ && SCALAR_INT_MODE_P (insn_op->mode)
+ && GET_MODE (op[arity]) != insn_op->mode)
+ op[arity] = GEN_INT (trunc_int_for_mode (INTVAL (op[arity]),
+ insn_op->mode));
+
/* Wrap the expanded RTX for pointer types into a MEM expr with
the proper mode. This allows us to use e.g. (match_operand
"memory_operand"..) in the insn patterns instead of (mem
arity++;
}
- if (last_vec_mode != VOIDmode && !TARGET_VX)
- {
- error ("Vector type builtin %qF is not supported without -mvx "
- "(default with -march=z13).",
- fndecl);
- return const0_rtx;
- }
-
switch (arity)
{
case 0:
{
int bit0, bit1;
- /* ??? Fixme: should work on CONST_DOUBLE as well. */
+ /* ??? Fixme: should work on CONST_WIDE_INT as well. */
if (GET_CODE (op1) != CONST_INT || GET_CODE (op2) != CONST_INT)
return VOIDmode;
{
unsigned HOST_WIDE_INT mask;
int length, size;
+ rtx elt;
- if (!VECTOR_MODE_P (GET_MODE (op))
- || GET_CODE (op) != CONST_VECTOR
- || !CONST_INT_P (XVECEXP (op, 0, 0)))
+ if (!const_vec_duplicate_p (op, &elt)
+ || !CONST_INT_P (elt))
return false;
- if (GET_MODE_NUNITS (GET_MODE (op)) > 1)
- {
- int i;
+ size = GET_MODE_UNIT_BITSIZE (GET_MODE (op));
- for (i = 1; i < GET_MODE_NUNITS (GET_MODE (op)); ++i)
- if (!rtx_equal_p (XVECEXP (op, 0, i), XVECEXP (op, 0, 0)))
- return false;
- }
+ /* We cannot deal with V1TI/V1TF. This would require a vgmq. */
+ if (size > 64)
+ return false;
- size = GET_MODE_UNIT_BITSIZE (GET_MODE (op));
- mask = UINTVAL (XVECEXP (op, 0, 0));
+ mask = UINTVAL (elt);
if (s390_contiguous_bitmask_p (mask, size, start,
end != NULL ? &length : NULL))
{
/* 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.
- CODE contains GET_CODE (x), OUTER_CODE contains the code
- of the superexpression of x. */
+ OUTER_CODE contains the code of the superexpression of x. */
static bool
-s390_rtx_costs (rtx x, int code, int outer_code, int opno ATTRIBUTE_UNUSED,
+s390_rtx_costs (rtx x, machine_mode mode, int outer_code,
+ int opno ATTRIBUTE_UNUSED,
int *total, bool speed ATTRIBUTE_UNUSED)
{
+ int code = GET_CODE (x);
switch (code)
{
case CONST:
case LABEL_REF:
case SYMBOL_REF:
case CONST_DOUBLE:
+ case CONST_WIDE_INT:
case MEM:
*total = 0;
return true;
+ case IOR:
+ /* risbg */
+ if (GET_CODE (XEXP (x, 0)) == AND
+ && GET_CODE (XEXP (x, 1)) == ASHIFT
+ && REG_P (XEXP (XEXP (x, 0), 0))
+ && REG_P (XEXP (XEXP (x, 1), 0))
+ && CONST_INT_P (XEXP (XEXP (x, 0), 1))
+ && CONST_INT_P (XEXP (XEXP (x, 1), 1))
+ && (UINTVAL (XEXP (XEXP (x, 0), 1)) ==
+ (1UL << UINTVAL (XEXP (XEXP (x, 1), 1))) - 1))
+ {
+ *total = COSTS_N_INSNS (2);
+ return true;
+ }
case ASHIFT:
case ASHIFTRT:
case LSHIFTRT:
case ROTATE:
case ROTATERT:
case AND:
- case IOR:
case XOR:
case NEG:
case NOT:
return false;
case MULT:
- switch (GET_MODE (x))
+ switch (mode)
{
case SImode:
{
return false;
case FMA:
- switch (GET_MODE (x))
+ switch (mode)
{
case DFmode:
*total = s390_cost->madbr;
/* Negate in the third argument is free: FMSUB. */
if (GET_CODE (XEXP (x, 2)) == NEG)
{
- *total += (rtx_cost (XEXP (x, 0), FMA, 0, speed)
- + rtx_cost (XEXP (x, 1), FMA, 1, speed)
- + rtx_cost (XEXP (XEXP (x, 2), 0), FMA, 2, speed));
+ *total += (rtx_cost (XEXP (x, 0), mode, FMA, 0, speed)
+ + rtx_cost (XEXP (x, 1), mode, FMA, 1, speed)
+ + rtx_cost (XEXP (XEXP (x, 2), 0), mode, FMA, 2, speed));
return true;
}
return false;
case UDIV:
case UMOD:
- if (GET_MODE (x) == TImode) /* 128 bit division */
+ if (mode == TImode) /* 128 bit division */
*total = s390_cost->dlgr;
- else if (GET_MODE (x) == DImode)
+ else if (mode == DImode)
{
rtx right = XEXP (x, 1);
if (GET_CODE (right) == ZERO_EXTEND) /* 64 by 32 bit division */
else /* 64 by 64 bit division */
*total = s390_cost->dlgr;
}
- else if (GET_MODE (x) == SImode) /* 32 bit division */
+ else if (mode == SImode) /* 32 bit division */
*total = s390_cost->dlr;
return false;
case DIV:
case MOD:
- if (GET_MODE (x) == DImode)
+ if (mode == DImode)
{
rtx right = XEXP (x, 1);
if (GET_CODE (right) == ZERO_EXTEND) /* 64 by 32 bit division */
else /* 64 by 64 bit division */
*total = s390_cost->dsgr;
}
- else if (GET_MODE (x) == SImode) /* 32 bit division */
+ else if (mode == SImode) /* 32 bit division */
*total = s390_cost->dlr;
- else if (GET_MODE (x) == SFmode)
+ else if (mode == SFmode)
{
*total = s390_cost->debr;
}
- else if (GET_MODE (x) == DFmode)
+ else if (mode == DFmode)
{
*total = s390_cost->ddbr;
}
- else if (GET_MODE (x) == TFmode)
+ else if (mode == TFmode)
{
*total = s390_cost->dxbr;
}
return false;
case SQRT:
- if (GET_MODE (x) == SFmode)
+ if (mode == SFmode)
*total = s390_cost->sqebr;
- else if (GET_MODE (x) == DFmode)
+ else if (mode == DFmode)
*total = s390_cost->sqdbr;
else /* TFmode */
*total = s390_cost->sqxbr;
/* Return true if OP is a legitimate general operand when
generating PIC code. It is given that flag_pic is on
- and that OP satisfies CONSTANT_P or is a CONST_DOUBLE. */
+ and that OP satisfies CONSTANT_P. */
int
legitimate_pic_operand_p (rtx op)
}
/* Returns true if the constant value OP is a legitimate general operand.
- It is given that OP satisfies CONSTANT_P or is a CONST_DOUBLE. */
+ It is given that OP satisfies CONSTANT_P. */
static bool
s390_legitimate_constant_p (machine_mode mode, rtx op)
{
- if (VECTOR_MODE_P (mode) && GET_CODE (op) == CONST_VECTOR)
+ if (TARGET_VX && VECTOR_MODE_P (mode) && GET_CODE (op) == CONST_VECTOR)
{
if (GET_MODE_SIZE (mode) != 16)
return 0;
- if (!const0_operand (op, mode)
- && !s390_contiguous_bitmask_vector_p (op, NULL, NULL)
- && !s390_bytemask_vector_p (op, NULL))
+ if (!satisfies_constraint_j00 (op)
+ && !satisfies_constraint_jm1 (op)
+ && !satisfies_constraint_jKK (op)
+ && !satisfies_constraint_jxx (op)
+ && !satisfies_constraint_jyy (op))
return 0;
}
{
case CONST_INT:
case CONST_DOUBLE:
+ case CONST_WIDE_INT:
case CONST_VECTOR:
/* Accept all non-symbolic constants. */
return false;
return true;
/* Accept double-word operands that can be split. */
- if (GET_CODE (op) == CONST_INT
- && trunc_int_for_mode (INTVAL (op), word_mode) != INTVAL (op))
+ if (GET_CODE (op) == CONST_WIDE_INT
+ || (GET_CODE (op) == CONST_INT
+ && trunc_int_for_mode (INTVAL (op), word_mode) != INTVAL (op)))
{
machine_mode dword_mode = word_mode == SImode ? DImode : TImode;
rtx hi = operand_subword (op, 0, 0, dword_mode);
static bool
legitimate_reload_vector_constant_p (rtx op)
{
- /* FIXME: Support constant vectors with all the same 16 bit unsigned
- operands. These can be loaded with vrepi. */
-
if (TARGET_VX && GET_MODE_SIZE (GET_MODE (op)) == 16
- && (const0_operand (op, GET_MODE (op))
- || constm1_operand (op, GET_MODE (op))
- || s390_contiguous_bitmask_vector_p (op, NULL, NULL)
- || s390_bytemask_vector_p (op, NULL)))
+ && (satisfies_constraint_j00 (op)
+ || satisfies_constraint_jm1 (op)
+ || satisfies_constraint_jKK (op)
+ || satisfies_constraint_jxx (op)
+ || satisfies_constraint_jyy (op)))
return true;
return false;
case CONST_VECTOR:
case CONST_DOUBLE:
case CONST_INT:
+ case CONST_WIDE_INT:
if (reg_class_subset_p (GENERAL_REGS, rclass)
&& legitimate_reload_constant_p (op))
return GENERAL_REGS;
HOST_WIDE_INT addend;
rtx symref;
+ /* The "required alignment" might be 0 (e.g. for certain structs
+ accessed via BLKmode). Early abort in this case, as well as when
+ an alignment > 8 is required. */
+ if (alignment < 2 || alignment > 8)
+ return false;
+
if (!s390_loadrelative_operand_p (addr, &symref, &addend))
return false;
if (addend & (alignment - 1))
return false;
- if (GET_CODE (symref) == SYMBOL_REF
- && !SYMBOL_REF_NOT_NATURALLY_ALIGNED_P (symref))
- return true;
+ if (GET_CODE (symref) == SYMBOL_REF)
+ {
+ /* We have load-relative instructions for 2-byte, 4-byte, and
+ 8-byte alignment so allow only these. */
+ switch (alignment)
+ {
+ case 8: return !SYMBOL_FLAG_NOTALIGN8_P (symref);
+ case 4: return !SYMBOL_FLAG_NOTALIGN4_P (symref);
+ case 2: return !SYMBOL_FLAG_NOTALIGN2_P (symref);
+ default: return false;
+ }
+ }
if (GET_CODE (symref) == UNSPEC
&& alignment <= UNITS_PER_LONG)
/* Reload might have pulled a constant out of the literal pool.
Force it back in. */
if (CONST_INT_P (mem) || GET_CODE (mem) == CONST_DOUBLE
+ || GET_CODE (mem) == CONST_WIDE_INT
|| GET_CODE (mem) == CONST_VECTOR
|| GET_CODE (mem) == CONST)
mem = force_const_mem (GET_MODE (reg), mem);
if (in_p
&& s390_loadrelative_operand_p (x, &symref, &offset)
&& mode == Pmode
- && !SYMBOL_REF_ALIGN1_P (symref)
+ && !SYMBOL_FLAG_NOTALIGN2_P (symref)
&& (offset & 1) == 1)
sri->icode = ((mode == DImode) ? CODE_FOR_reloaddi_larl_odd_addend_z10
: CODE_FOR_reloadsi_larl_odd_addend_z10);
if (MEM_P (x)
&& s390_loadrelative_operand_p (XEXP (x, 0), NULL, NULL)
&& (mode == QImode
- || !reg_classes_intersect_p (GENERAL_REGS, rclass)
+ || !reg_class_subset_p (rclass, GENERAL_REGS)
|| GET_MODE_SIZE (mode) > UNITS_PER_WORD
|| !s390_check_symref_alignment (XEXP (x, 0),
GET_MODE_SIZE (mode))))
else if (TARGET_MVCLE)
{
val = force_not_mem (convert_modes (Pmode, QImode, val, 1));
- emit_insn (gen_setmem_long (dst, convert_to_mode (Pmode, len, 1), val));
+ if (TARGET_64BIT)
+ emit_insn (gen_setmem_long_di (dst, convert_to_mode (Pmode, len, 1),
+ val));
+ else
+ emit_insn (gen_setmem_long_si (dst, convert_to_mode (Pmode, len, 1),
+ val));
}
else
if (mode_s == VOIDmode)
{
- /* Assume const_int etc already in the proper mode. */
- src = force_reg (mode, src);
+ /* For constant zero values the representation with AND
+ appears to be folded in more situations than the (set
+ (zero_extract) ...).
+ We only do this when the start and end of the bitfield
+ remain in the same SImode chunk. That way nihf or nilf
+ can be used.
+ The AND patterns might still generate a risbg for this. */
+ if (src == const0_rtx && bitpos / 32 == (bitpos + bitsize - 1) / 32)
+ return false;
+ else
+ src = force_reg (mode, src);
}
else if (mode_s != mode)
{
machine_mode result_mode;
rtx result_target;
+ machine_mode target_mode = GET_MODE (target);
+ machine_mode cmp_mode = GET_MODE (cmp_op1);
+ rtx op = (cond == LT) ? els : then;
+
+ /* Try to optimize x < 0 ? -1 : 0 into (signed) x >> 31
+ and x < 0 ? 1 : 0 into (unsigned) x >> 31. Likewise
+ for short and byte (x >> 15 and x >> 7 respectively). */
+ if ((cond == LT || cond == GE)
+ && target_mode == cmp_mode
+ && cmp_op2 == CONST0_RTX (cmp_mode)
+ && op == CONST0_RTX (target_mode)
+ && s390_vector_mode_supported_p (target_mode)
+ && GET_MODE_CLASS (target_mode) == MODE_VECTOR_INT)
+ {
+ rtx negop = (cond == LT) ? then : els;
+
+ int shift = GET_MODE_BITSIZE (GET_MODE_INNER (target_mode)) - 1;
+
+ /* if x < 0 ? 1 : 0 or if x >= 0 ? 0 : 1 */
+ if (negop == CONST1_RTX (target_mode))
+ {
+ rtx res = expand_simple_binop (cmp_mode, LSHIFTRT, cmp_op1,
+ GEN_INT (shift), target,
+ 1, OPTAB_DIRECT);
+ if (res != target)
+ emit_move_insn (target, res);
+ return;
+ }
+
+ /* if x < 0 ? -1 : 0 or if x >= 0 ? 0 : -1 */
+ else if (all_ones_operand (negop, target_mode))
+ {
+ rtx res = expand_simple_binop (cmp_mode, ASHIFTRT, cmp_op1,
+ GEN_INT (shift), target,
+ 0, OPTAB_DIRECT);
+ if (res != target)
+ emit_move_insn (target, res);
+ return;
+ }
+ }
+
/* We always use an integral type vector to hold the comparison
result. */
- result_mode = GET_MODE (cmp_op1) == V2DFmode ? V2DImode : GET_MODE (cmp_op1);
+ result_mode = cmp_mode == V2DFmode ? V2DImode : cmp_mode;
result_target = gen_reg_rtx (result_mode);
- /* Alternatively this could be done by reload by lowering the cmp*
- predicates. But it appears to be better for scheduling etc. to
- have that in early. */
+ /* We allow vector immediates as comparison operands that
+ can be handled by the optimization above but not by the
+ following code. Hence, force them into registers here. */
if (!REG_P (cmp_op1))
- cmp_op1 = force_reg (GET_MODE (target), cmp_op1);
+ cmp_op1 = force_reg (target_mode, cmp_op1);
if (!REG_P (cmp_op2))
- cmp_op2 = force_reg (GET_MODE (target), cmp_op2);
+ cmp_op2 = force_reg (target_mode, cmp_op2);
s390_expand_vec_compare (result_target, cond,
cmp_op1, cmp_op2);
/* If the results are supposed to be either -1 or 0 we are done
since this is what our compare instructions generate anyway. */
- if (constm1_operand (then, GET_MODE (then))
+ if (all_ones_operand (then, GET_MODE (then))
&& const0_operand (els, GET_MODE (els)))
{
- emit_move_insn (target, gen_rtx_SUBREG (GET_MODE (target),
+ emit_move_insn (target, gen_rtx_SUBREG (target_mode,
result_target, 0));
return;
}
/* This gets triggered e.g.
with gcc.c-torture/compile/pr53410-1.c */
if (!REG_P (then))
- then = force_reg (GET_MODE (target), then);
+ then = force_reg (target_mode, then);
if (!REG_P (els))
- els = force_reg (GET_MODE (target), els);
+ els = force_reg (target_mode, els);
tmp = gen_rtx_fmt_ee (EQ, VOIDmode,
result_target,
/* We compared the result against zero above so we have to swap then
and els here. */
- tmp = gen_rtx_IF_THEN_ELSE (GET_MODE (target), tmp, els, then);
+ tmp = gen_rtx_IF_THEN_ELSE (target_mode, tmp, els, then);
- gcc_assert (GET_MODE (target) == GET_MODE (then));
+ gcc_assert (target_mode == GET_MODE (then));
emit_insn (gen_rtx_SET (target, tmp));
}
case SET:
if (ac.aligned && MEM_P (val))
store_bit_field (new_rtx, GET_MODE_BITSIZE (mode), 0,
- 0, 0, SImode, val);
+ 0, 0, SImode, val, false);
else
{
new_rtx = expand_simple_binop (SImode, AND, new_rtx, ac.modemaski,
{
machine_mode save_mode = default_dwarf_frame_reg_mode (regno);
+ /* Make sure not to return DImode for any GPR with -m31 -mzarch. */
+ if (GENERAL_REGNO_P (regno))
+ save_mode = Pmode;
+
/* The rightmost 64 bits of vector registers are call-clobbered. */
if (GET_MODE_SIZE (save_mode) > 8)
save_mode = DImode;
}
}
+/* Write the current .machine and .machinemode specification to the assembler
+ file. */
+
+#ifdef HAVE_AS_MACHINE_MACHINEMODE
+static void
+s390_asm_output_machine_for_arch (FILE *asm_out_file)
+{
+ fprintf (asm_out_file, "\t.machinemode %s\n",
+ (TARGET_ZARCH) ? "zarch" : "esa");
+ fprintf (asm_out_file, "\t.machine \"%s", processor_table[s390_arch].name);
+ if (S390_USE_ARCHITECTURE_MODIFIERS)
+ {
+ int cpu_flags;
+
+ cpu_flags = processor_flags_table[(int) s390_arch];
+ if (TARGET_HTM && !(cpu_flags & PF_TX))
+ fprintf (asm_out_file, "+htm");
+ else if (!TARGET_HTM && (cpu_flags & PF_TX))
+ fprintf (asm_out_file, "+nohtm");
+ if (TARGET_VX && !(cpu_flags & PF_VX))
+ fprintf (asm_out_file, "+vx");
+ else if (!TARGET_VX && (cpu_flags & PF_VX))
+ fprintf (asm_out_file, "+novx");
+ }
+ fprintf (asm_out_file, "\"\n");
+}
+
+/* Write an extra function header before the very start of the function. */
+
+void
+s390_asm_output_function_prefix (FILE *asm_out_file,
+ const char *fnname ATTRIBUTE_UNUSED)
+{
+ if (DECL_FUNCTION_SPECIFIC_TARGET (current_function_decl) == NULL)
+ return;
+ /* Since only the function specific options are saved but not the indications
+ which options are set, it's too much work here to figure out which options
+ have actually changed. Thus, generate .machine and .machinemode whenever a
+ function has the target attribute or pragma. */
+ fprintf (asm_out_file, "\t.machinemode push\n");
+ fprintf (asm_out_file, "\t.machine push\n");
+ s390_asm_output_machine_for_arch (asm_out_file);
+}
+
+/* Write an extra function footer after the very end of the function. */
+
+void
+s390_asm_declare_function_size (FILE *asm_out_file,
+ const char *fnname, tree decl)
+{
+ if (!flag_inhibit_size_directive)
+ ASM_OUTPUT_MEASURED_SIZE (asm_out_file, fnname);
+ if (DECL_FUNCTION_SPECIFIC_TARGET (decl) == NULL)
+ return;
+ fprintf (asm_out_file, "\t.machine pop\n");
+ fprintf (asm_out_file, "\t.machinemode pop\n");
+}
+#endif
+
/* Write the extra assembler code needed to declare a function properly. */
void
ASM_OUTPUT_ALIGN (asm_out_file, floor_log2 (function_alignment));
}
+ if (S390_USE_TARGET_ATTRIBUTE && TARGET_DEBUG_ARG)
+ {
+ asm_fprintf (asm_out_file, "\t# fn:%s ar%d\n", fname, s390_arch);
+ asm_fprintf (asm_out_file, "\t# fn:%s tu%d\n", fname, s390_tune);
+ asm_fprintf (asm_out_file, "\t# fn:%s sg%d\n", fname, s390_stack_guard);
+ asm_fprintf (asm_out_file, "\t# fn:%s ss%d\n", fname, s390_stack_size);
+ asm_fprintf (asm_out_file, "\t# fn:%s bc%d\n", fname, s390_branch_cost);
+ asm_fprintf (asm_out_file, "\t# fn:%s wf%d\n", fname,
+ s390_warn_framesize);
+ asm_fprintf (asm_out_file, "\t# fn:%s ba%d\n", fname, TARGET_BACKCHAIN);
+ asm_fprintf (asm_out_file, "\t# fn:%s hd%d\n", fname, TARGET_HARD_DFP);
+ asm_fprintf (asm_out_file, "\t# fn:%s hf%d\n", fname, !TARGET_SOFT_FLOAT);
+ asm_fprintf (asm_out_file, "\t# fn:%s ht%d\n", fname, TARGET_OPT_HTM);
+ asm_fprintf (asm_out_file, "\t# fn:%s vx%d\n", fname, TARGET_OPT_VX);
+ asm_fprintf (asm_out_file, "\t# fn:%s ps%d\n", fname,
+ TARGET_PACKED_STACK);
+ asm_fprintf (asm_out_file, "\t# fn:%s se%d\n", fname, TARGET_SMALL_EXEC);
+ asm_fprintf (asm_out_file, "\t# fn:%s mv%d\n", fname, TARGET_MVCLE);
+ asm_fprintf (asm_out_file, "\t# fn:%s zv%d\n", fname, TARGET_ZVECTOR);
+ asm_fprintf (asm_out_file, "\t# fn:%s wd%d\n", fname,
+ s390_warn_dynamicstack_p);
+ }
ASM_OUTPUT_LABEL (asm_out_file, fname);
if (hw_after > 0)
asm_fprintf (asm_out_file,
break;
case MEM:
- output_address (XEXP (x, 0));
+ output_address (GET_MODE (x), XEXP (x, 0));
break;
case CONST:
fprintf (file, HOST_WIDE_INT_PRINT_DEC, ival);
break;
- case CONST_DOUBLE:
- gcc_assert (GET_MODE (x) == VOIDmode);
+ case CONST_WIDE_INT:
if (code == 'b')
- fprintf (file, HOST_WIDE_INT_PRINT_DEC, CONST_DOUBLE_LOW (x) & 0xff);
+ fprintf (file, HOST_WIDE_INT_PRINT_DEC,
+ CONST_WIDE_INT_ELT (x, 0) & 0xff);
else if (code == 'x')
- fprintf (file, HOST_WIDE_INT_PRINT_DEC, CONST_DOUBLE_LOW (x) & 0xffff);
+ fprintf (file, HOST_WIDE_INT_PRINT_DEC,
+ CONST_WIDE_INT_ELT (x, 0) & 0xffff);
else if (code == 'h')
fprintf (file, HOST_WIDE_INT_PRINT_DEC,
- ((CONST_DOUBLE_LOW (x) & 0xffff) ^ 0x8000) - 0x8000);
+ ((CONST_WIDE_INT_ELT (x, 0) & 0xffff) ^ 0x8000) - 0x8000);
else
{
if (code == 0)
case CONST_VECTOR:
switch (code)
{
+ case 'h':
+ gcc_assert (const_vec_duplicate_p (x));
+ fprintf (file, HOST_WIDE_INT_PRINT_DEC,
+ ((INTVAL (XVECEXP (x, 0, 0)) & 0xffff) ^ 0x8000) - 0x8000);
+ break;
case 'e':
case 's':
{
if (! INSN_P (insn))
return priority;
- if (s390_tune != PROCESSOR_2084_Z990
- && s390_tune != PROCESSOR_2094_Z9_109
- && s390_tune != PROCESSOR_2097_Z10
- && s390_tune != PROCESSOR_2817_Z196
- && s390_tune != PROCESSOR_2827_ZEC12
- && s390_tune != PROCESSOR_2964_Z13)
+ if (s390_tune <= PROCESSOR_2064_Z900)
return priority;
switch (s390_safe_attr_type (insn))
{
case PROCESSOR_2084_Z990:
case PROCESSOR_2094_Z9_109:
+ case PROCESSOR_2094_Z9_EC:
case PROCESSOR_2817_Z196:
return 3;
case PROCESSOR_2097_Z10:
return 2;
+ case PROCESSOR_9672_G5:
+ case PROCESSOR_9672_G6:
+ case PROCESSOR_2064_Z900:
/* Starting with EC12 we use the sched_reorder hook to take care
of instruction dispatch constraints. The algorithm only
picks the best instruction and assumes only a single
instruction gets issued per cycle. */
case PROCESSOR_2827_ZEC12:
+ case PROCESSOR_2964_Z13:
default:
return 1;
}
/* We keep a list of constants which we have to add to internal
constant tables in the middle of large functions. */
-#define NR_C_MODES 31
+#define NR_C_MODES 32
machine_mode constant_modes[NR_C_MODES] =
{
TFmode, TImode, TDmode,
- V16QImode, V8HImode, V4SImode, V2DImode, V4SFmode, V2DFmode, V1TFmode,
+ V16QImode, V8HImode, V4SImode, V2DImode, V1TImode,
+ V4SFmode, V2DFmode, V1TFmode,
DFmode, DImode, DDmode,
V8QImode, V4HImode, V2SImode, V1DImode, V2SFmode, V1DFmode,
SFmode, SImode, SDmode,
void
s390_output_pool_entry (rtx exp, machine_mode mode, unsigned int align)
{
- REAL_VALUE_TYPE r;
-
switch (GET_MODE_CLASS (mode))
{
case MODE_FLOAT:
case MODE_DECIMAL_FLOAT:
gcc_assert (GET_CODE (exp) == CONST_DOUBLE);
- REAL_VALUE_FROM_CONST_DOUBLE (r, exp);
- assemble_real (r, mode, align);
+ assemble_real (*CONST_DOUBLE_REAL_VALUE (exp), mode, align);
break;
case MODE_INT:
for (i = 15; i >= 6; i--)
{
- if (cfun_gpr_save_slot (i) == 0)
+ if (cfun_gpr_save_slot (i) == SAVE_SLOT_NONE)
continue;
/* Advance to the next FP register which can be used as a
case we ran out of FPR save slots. */
for (j = 6; j <= 15; j++)
if (FP_REGNO_P (cfun_gpr_save_slot (j)))
- cfun_gpr_save_slot (j) = -1;
+ cfun_gpr_save_slot (j) = SAVE_SLOT_STACK;
break;
}
cfun_gpr_save_slot (i) = save_reg_slot++;
return;
min_fpr = crtl->args.info.fprs;
- max_fpr = min_fpr + cfun->va_list_fpr_size;
- if (max_fpr > FP_ARG_NUM_REG)
- max_fpr = FP_ARG_NUM_REG;
+ max_fpr = min_fpr + cfun->va_list_fpr_size - 1;
+ if (max_fpr >= FP_ARG_NUM_REG)
+ max_fpr = FP_ARG_NUM_REG - 1;
+
+ /* FPR argument regs start at f0. */
+ min_fpr += FPR0_REGNUM;
+ max_fpr += FPR0_REGNUM;
- for (i = min_fpr; i < max_fpr; i++)
- cfun_set_fpr_save (i + FPR0_REGNUM);
+ for (i = min_fpr; i <= max_fpr; i++)
+ cfun_set_fpr_save (i);
}
/* Reserve the GPR save slots for GPRs which need to be saved due to
return;
min_gpr = crtl->args.info.gprs;
- max_gpr = min_gpr + cfun->va_list_gpr_size;
- if (max_gpr > GP_ARG_NUM_REG)
- max_gpr = GP_ARG_NUM_REG;
+ max_gpr = min_gpr + cfun->va_list_gpr_size - 1;
+ if (max_gpr >= GP_ARG_NUM_REG)
+ max_gpr = GP_ARG_NUM_REG - 1;
+
+ /* GPR argument regs start at r2. */
+ min_gpr += GPR2_REGNUM;
+ max_gpr += GPR2_REGNUM;
+
+ /* If r6 was supposed to be saved into an FPR and now needs to go to
+ the stack for vararg we have to adjust the restore range to make
+ sure that the restore is done from stack as well. */
+ if (FP_REGNO_P (cfun_gpr_save_slot (GPR6_REGNUM))
+ && min_gpr <= GPR6_REGNUM
+ && max_gpr >= GPR6_REGNUM)
+ {
+ if (cfun_frame_layout.first_restore_gpr == -1
+ || cfun_frame_layout.first_restore_gpr > GPR6_REGNUM)
+ cfun_frame_layout.first_restore_gpr = GPR6_REGNUM;
+ if (cfun_frame_layout.last_restore_gpr == -1
+ || cfun_frame_layout.last_restore_gpr < GPR6_REGNUM)
+ cfun_frame_layout.last_restore_gpr = GPR6_REGNUM;
+ }
+
+ if (cfun_frame_layout.first_save_gpr == -1
+ || cfun_frame_layout.first_save_gpr > min_gpr)
+ cfun_frame_layout.first_save_gpr = min_gpr;
+
+ if (cfun_frame_layout.last_save_gpr == -1
+ || cfun_frame_layout.last_save_gpr < max_gpr)
+ cfun_frame_layout.last_save_gpr = max_gpr;
+
+ for (i = min_gpr; i <= max_gpr; i++)
+ cfun_gpr_save_slot (i) = SAVE_SLOT_STACK;
+}
+
+/* Calculate the save and restore ranges for stm(g) and lm(g) in the
+ prologue and epilogue. */
+
+static void
+s390_register_info_set_ranges ()
+{
+ int i, j;
- for (i = min_gpr; i < max_gpr; i++)
- cfun_gpr_save_slot (2 + i) = -1;
+ /* Find the first and the last save slot supposed to use the stack
+ to set the restore range.
+ Vararg regs might be marked as save to stack but only the
+ call-saved regs really need restoring (i.e. r6). This code
+ assumes that the vararg regs have not yet been recorded in
+ cfun_gpr_save_slot. */
+ for (i = 0; i < 16 && cfun_gpr_save_slot (i) != SAVE_SLOT_STACK; i++);
+ for (j = 15; j > i && cfun_gpr_save_slot (j) != SAVE_SLOT_STACK; j--);
+ cfun_frame_layout.first_restore_gpr = (i == 16) ? -1 : i;
+ cfun_frame_layout.last_restore_gpr = (i == 16) ? -1 : j;
+ cfun_frame_layout.first_save_gpr = (i == 16) ? -1 : i;
+ cfun_frame_layout.last_save_gpr = (i == 16) ? -1 : j;
}
/* The GPR and FPR save slots in cfun->machine->frame_layout are set
static void
s390_register_info ()
{
- int i, j;
+ int i;
char clobbered_regs[32];
gcc_assert (!epilogue_completed);
cfun_frame_layout.high_fprs++;
}
- if (flag_pic)
- clobbered_regs[PIC_OFFSET_TABLE_REGNUM]
- |= !!df_regs_ever_live_p (PIC_OFFSET_TABLE_REGNUM);
+ /* Register 12 is used for GOT address, but also as temp in prologue
+ for split-stack stdarg functions (unless r14 is available). */
+ clobbered_regs[12]
+ |= ((flag_pic && df_regs_ever_live_p (PIC_OFFSET_TABLE_REGNUM))
+ || (flag_split_stack && cfun->stdarg
+ && (crtl->is_leaf || TARGET_TPF_PROFILING
+ || has_hard_reg_initial_val (Pmode, RETURN_REGNUM))));
clobbered_regs[BASE_REGNUM]
|= (cfun->machine->base_reg
|| (reload_completed && cfun_frame_layout.frame_size > 0)
|| cfun->calls_alloca);
- memset (cfun_frame_layout.gpr_save_slots, 0, 16);
+ memset (cfun_frame_layout.gpr_save_slots, SAVE_SLOT_NONE, 16);
for (i = 6; i < 16; i++)
if (clobbered_regs[i])
- cfun_gpr_save_slot (i) = -1;
+ cfun_gpr_save_slot (i) = SAVE_SLOT_STACK;
s390_register_info_stdarg_fpr ();
s390_register_info_gprtofpr ();
-
- /* First find the range of GPRs to be restored. Vararg regs don't
- need to be restored so we do it before assigning slots to the
- vararg GPRs. */
- for (i = 0; i < 16 && cfun_gpr_save_slot (i) != -1; i++);
- for (j = 15; j > i && cfun_gpr_save_slot (j) != -1; j--);
- cfun_frame_layout.first_restore_gpr = (i == 16) ? -1 : i;
- cfun_frame_layout.last_restore_gpr = (i == 16) ? -1 : j;
-
+ s390_register_info_set_ranges ();
/* stdarg functions might need to save GPRs 2 to 6. This might
- override the GPR->FPR save decision made above for r6 since
- vararg regs must go to the stack. */
+ override the GPR->FPR save decision made by
+ s390_register_info_gprtofpr for r6 since vararg regs must go to
+ the stack. */
s390_register_info_stdarg_gpr ();
-
- /* Now the range of GPRs which need saving. */
- for (i = 0; i < 16 && cfun_gpr_save_slot (i) != -1; i++);
- for (j = 15; j > i && cfun_gpr_save_slot (j) != -1; j--);
- cfun_frame_layout.first_save_gpr = (i == 16) ? -1 : i;
- cfun_frame_layout.last_save_gpr = (i == 16) ? -1 : j;
}
/* This function is called by s390_optimize_prologue in order to get
s390_optimize_register_info ()
{
char clobbered_regs[32];
- int i, j;
+ int i;
gcc_assert (epilogue_completed);
gcc_assert (!cfun->machine->split_branches_pending_p);
|| cfun_frame_layout.save_return_addr_p
|| crtl->calls_eh_return);
- memset (cfun_frame_layout.gpr_save_slots, 0, 6);
+ memset (cfun_frame_layout.gpr_save_slots, SAVE_SLOT_NONE, 6);
for (i = 6; i < 16; i++)
if (!clobbered_regs[i])
- cfun_gpr_save_slot (i) = 0;
-
- for (i = 0; i < 16 && cfun_gpr_save_slot (i) != -1; i++);
- for (j = 15; j > i && cfun_gpr_save_slot (j) != -1; j--);
- cfun_frame_layout.first_restore_gpr = (i == 16) ? -1 : i;
- cfun_frame_layout.last_restore_gpr = (i == 16) ? -1 : j;
+ cfun_gpr_save_slot (i) = SAVE_SLOT_NONE;
+ s390_register_info_set_ranges ();
s390_register_info_stdarg_gpr ();
-
- for (i = 0; i < 16 && cfun_gpr_save_slot (i) != -1; i++);
- for (j = 15; j > i && cfun_gpr_save_slot (j) != -1; j--);
- cfun_frame_layout.first_save_gpr = (i == 16) ? -1 : i;
- cfun_frame_layout.last_save_gpr = (i == 16) ? -1 : j;
}
/* Fill cfun->machine with info about frame of current function. */
HOST_WIDE_INT frame_size;
int base_used;
- gcc_assert (!reload_completed);
+ /* After LRA the frame layout is supposed to be read-only and should
+ not be re-computed. */
+ if (reload_completed)
+ return;
/* On S/390 machines, we may need to perform branch splitting, which
will require both base and return address register. We have no
as base register to avoid save/restore overhead. */
if (!base_used)
cfun->machine->base_reg = NULL_RTX;
- else if (crtl->is_leaf && !df_regs_ever_live_p (5))
- cfun->machine->base_reg = gen_rtx_REG (Pmode, 5);
else
- cfun->machine->base_reg = gen_rtx_REG (Pmode, BASE_REGNUM);
+ {
+ int br = 0;
+
+ if (crtl->is_leaf)
+ /* Prefer r5 (most likely to be free). */
+ for (br = 5; br >= 2 && df_regs_ever_live_p (br); br--)
+ ;
+ cfun->machine->base_reg =
+ gen_rtx_REG (Pmode, (br >= 2) ? br : BASE_REGNUM);
+ }
s390_register_info ();
s390_frame_info ();
regrename manually about it. */
if (GENERAL_REGNO_P (new_reg)
&& !call_really_used_regs[new_reg]
- && cfun_gpr_save_slot (new_reg) == 0)
+ && cfun_gpr_save_slot (new_reg) == SAVE_SLOT_NONE)
return false;
return true;
/* See s390_hard_regno_rename_ok. */
if (GENERAL_REGNO_P (regno)
&& !call_really_used_regs[regno]
- && cfun_gpr_save_slot (regno) == 0)
+ && cfun_gpr_save_slot (regno) == SAVE_SLOT_NONE)
return false;
return true;
emit_move_insn (gen_rtx_REG (DImode, cfun_gpr_save_slot (i)),
gen_rtx_REG (DImode, i));
RTX_FRAME_RELATED_P (insn) = 1;
+ /* This prevents dwarf2cfi from interpreting the set. Doing
+ so it might emit def_cfa_register infos setting an FPR as
+ new CFA. */
+ add_reg_note (insn, REG_CFA_REGISTER, PATTERN (insn));
}
}
}
int next_fpr = 0;
/* Choose best register to use for temp use within prologue.
- See below for why TPF must use the register 1. */
+ TPF with profiling must avoid the register 14 - the tracing function
+ needs the original contents of r14 to be preserved. */
if (!has_hard_reg_initial_val (Pmode, RETURN_REGNUM)
&& !crtl->is_leaf
&& !TARGET_TPF_PROFILING)
temp_reg = gen_rtx_REG (Pmode, RETURN_REGNUM);
+ else if (flag_split_stack && cfun->stdarg)
+ temp_reg = gen_rtx_REG (Pmode, 12);
else
temp_reg = gen_rtx_REG (Pmode, 1);
current_function_name(), cfun_frame_layout.frame_size,
s390_stack_size);
emit_insn (gen_trap ());
+ emit_barrier ();
}
else
{
be in between two GPRs which need saving.) Otherwise it
would be difficult to take that decision back in
s390_optimize_prologue. */
- if (cfun_gpr_save_slot (RETURN_REGNUM) == -1)
+ if (cfun_gpr_save_slot (RETURN_REGNUM) == SAVE_SLOT_STACK)
{
int return_regnum = find_unused_clobbered_reg();
if (!return_regnum)
SET_HARD_REG_BIT (regs->set, REGNO (cfun->machine->base_reg));
}
-/* Return true if the function can use simple_return to return outside
- of a shrink-wrapped region. At present shrink-wrapping is supported
- in all cases. */
+/* -fsplit-stack support. */
-bool
-s390_can_use_simple_return_insn (void)
-{
- return true;
-}
+/* A SYMBOL_REF for __morestack. */
+static GTY(()) rtx morestack_ref;
-/* Return true if the epilogue is guaranteed to contain only a return
- instruction and if a direct return can therefore be used instead.
- One of the main advantages of using direct return instructions
- is that we can then use conditional returns. */
+/* When using -fsplit-stack, the allocation routines set a field in
+ the TCB to the bottom of the stack plus this much space, measured
+ in bytes. */
-bool
-s390_can_use_return_insn (void)
-{
- int i;
+#define SPLIT_STACK_AVAILABLE 1024
- if (!reload_completed)
- return false;
+/* Emit -fsplit-stack prologue, which goes before the regular function
+ prologue. */
- if (crtl->profile)
- return false;
+void
+s390_expand_split_stack_prologue (void)
+{
+ rtx r1, guard, cc = NULL;
+ rtx_insn *insn;
+ /* Offset from thread pointer to __private_ss. */
+ int psso = TARGET_64BIT ? 0x38 : 0x20;
+ /* Pointer size in bytes. */
+ /* Frame size and argument size - the two parameters to __morestack. */
+ HOST_WIDE_INT frame_size = cfun_frame_layout.frame_size;
+ /* Align argument size to 8 bytes - simplifies __morestack code. */
+ HOST_WIDE_INT args_size = crtl->args.size >= 0
+ ? ((crtl->args.size + 7) & ~7)
+ : 0;
+ /* Label to be called by __morestack. */
+ rtx_code_label *call_done = NULL;
+ rtx_code_label *parm_base = NULL;
+ rtx tmp;
- if (TARGET_TPF_PROFILING)
- return false;
+ gcc_assert (flag_split_stack && reload_completed);
+ if (!TARGET_CPU_ZARCH)
+ {
+ sorry ("CPUs older than z900 are not supported for -fsplit-stack");
+ return;
+ }
- for (i = 0; i < 16; i++)
- if (cfun_gpr_save_slot (i))
- return false;
+ r1 = gen_rtx_REG (Pmode, 1);
- /* For 31 bit this is not covered by the frame_size check below
- since f4, f6 are saved in the register save area without needing
- additional stack space. */
+ /* If no stack frame will be allocated, don't do anything. */
+ if (!frame_size)
+ {
+ if (cfun->machine->split_stack_varargs_pointer != NULL_RTX)
+ {
+ /* If va_start is used, just use r15. */
+ emit_move_insn (r1,
+ gen_rtx_PLUS (Pmode, stack_pointer_rtx,
+ GEN_INT (STACK_POINTER_OFFSET)));
+
+ }
+ return;
+ }
+
+ if (morestack_ref == NULL_RTX)
+ {
+ morestack_ref = gen_rtx_SYMBOL_REF (Pmode, "__morestack");
+ SYMBOL_REF_FLAGS (morestack_ref) |= (SYMBOL_FLAG_LOCAL
+ | SYMBOL_FLAG_FUNCTION);
+ }
+
+ if (CONST_OK_FOR_K (frame_size) || CONST_OK_FOR_Op (frame_size))
+ {
+ /* If frame_size will fit in an add instruction, do a stack space
+ check, and only call __morestack if there's not enough space. */
+
+ /* Get thread pointer. r1 is the only register we can always destroy - r0
+ could contain a static chain (and cannot be used to address memory
+ anyway), r2-r6 can contain parameters, and r6-r15 are callee-saved. */
+ emit_move_insn (r1, gen_rtx_REG (Pmode, TP_REGNUM));
+ /* Aim at __private_ss. */
+ guard = gen_rtx_MEM (Pmode, plus_constant (Pmode, r1, psso));
+
+ /* If less that 1kiB used, skip addition and compare directly with
+ __private_ss. */
+ if (frame_size > SPLIT_STACK_AVAILABLE)
+ {
+ emit_move_insn (r1, guard);
+ if (TARGET_64BIT)
+ emit_insn (gen_adddi3 (r1, r1, GEN_INT (frame_size)));
+ else
+ emit_insn (gen_addsi3 (r1, r1, GEN_INT (frame_size)));
+ guard = r1;
+ }
+
+ /* Compare the (maybe adjusted) guard with the stack pointer. */
+ cc = s390_emit_compare (LT, stack_pointer_rtx, guard);
+ }
+
+ call_done = gen_label_rtx ();
+ parm_base = gen_label_rtx ();
+
+ /* Emit the parameter block. */
+ tmp = gen_split_stack_data (parm_base, call_done,
+ GEN_INT (frame_size),
+ GEN_INT (args_size));
+ insn = emit_insn (tmp);
+ add_reg_note (insn, REG_LABEL_OPERAND, call_done);
+ LABEL_NUSES (call_done)++;
+ add_reg_note (insn, REG_LABEL_OPERAND, parm_base);
+ LABEL_NUSES (parm_base)++;
+
+ /* %r1 = litbase. */
+ insn = emit_move_insn (r1, gen_rtx_LABEL_REF (VOIDmode, parm_base));
+ add_reg_note (insn, REG_LABEL_OPERAND, parm_base);
+ LABEL_NUSES (parm_base)++;
+
+ /* Now, we need to call __morestack. It has very special calling
+ conventions: it preserves param/return/static chain registers for
+ calling main function body, and looks for its own parameters at %r1. */
+
+ if (cc != NULL)
+ {
+ tmp = gen_split_stack_cond_call (morestack_ref, cc, call_done);
+
+ insn = emit_jump_insn (tmp);
+ JUMP_LABEL (insn) = call_done;
+ LABEL_NUSES (call_done)++;
+
+ /* Mark the jump as very unlikely to be taken. */
+ add_int_reg_note (insn, REG_BR_PROB, REG_BR_PROB_BASE / 100);
+
+ if (cfun->machine->split_stack_varargs_pointer != NULL_RTX)
+ {
+ /* If va_start is used, and __morestack was not called, just use
+ r15. */
+ emit_move_insn (r1,
+ gen_rtx_PLUS (Pmode, stack_pointer_rtx,
+ GEN_INT (STACK_POINTER_OFFSET)));
+ }
+ }
+ else
+ {
+ tmp = gen_split_stack_call (morestack_ref, call_done);
+ insn = emit_jump_insn (tmp);
+ JUMP_LABEL (insn) = call_done;
+ LABEL_NUSES (call_done)++;
+ emit_barrier ();
+ }
+
+ /* __morestack will call us here. */
+
+ emit_label (call_done);
+}
+
+/* We may have to tell the dataflow pass that the split stack prologue
+ is initializing a register. */
+
+static void
+s390_live_on_entry (bitmap regs)
+{
+ if (cfun->machine->split_stack_varargs_pointer != NULL_RTX)
+ {
+ gcc_assert (flag_split_stack);
+ bitmap_set_bit (regs, 1);
+ }
+}
+
+/* Return true if the function can use simple_return to return outside
+ of a shrink-wrapped region. At present shrink-wrapping is supported
+ in all cases. */
+
+bool
+s390_can_use_simple_return_insn (void)
+{
+ return true;
+}
+
+/* Return true if the epilogue is guaranteed to contain only a return
+ instruction and if a direct return can therefore be used instead.
+ One of the main advantages of using direct return instructions
+ is that we can then use conditional returns. */
+
+bool
+s390_can_use_return_insn (void)
+{
+ int i;
+
+ if (!reload_completed)
+ return false;
+
+ if (crtl->profile)
+ return false;
+
+ if (TARGET_TPF_PROFILING)
+ return false;
+
+ for (i = 0; i < 16; i++)
+ if (cfun_gpr_save_slot (i) != SAVE_SLOT_NONE)
+ return false;
+
+ /* For 31 bit this is not covered by the frame_size check below
+ since f4, f6 are saved in the register save area without needing
+ additional stack space. */
if (!TARGET_64BIT
&& (cfun_fpr_save_p (FPR4_REGNUM) || cfun_fpr_save_p (FPR6_REGNUM)))
return false;
{
CUMULATIVE_ARGS *cum = get_cumulative_args (cum_v);
+ if (!named)
+ s390_check_type_for_vector_abi (type, true, false);
if (s390_function_arg_vector (mode, type))
{
expand_expr (t, const0_rtx, VOIDmode, EXPAND_NORMAL);
}
+ if (flag_split_stack
+ && (lookup_attribute ("no_split_stack", DECL_ATTRIBUTES (cfun->decl))
+ == NULL)
+ && cfun->machine->split_stack_varargs_pointer == NULL_RTX)
+ {
+ rtx reg;
+ rtx_insn *seq;
+
+ reg = gen_reg_rtx (Pmode);
+ cfun->machine->split_stack_varargs_pointer = reg;
+
+ start_sequence ();
+ emit_move_insn (reg, gen_rtx_REG (Pmode, 1));
+ seq = get_insns ();
+ end_sequence ();
+
+ push_topmost_sequence ();
+ emit_insn_after (seq, entry_of_function ());
+ pop_topmost_sequence ();
+ }
+
/* Find the overflow area.
FIXME: This currently is too pessimistic when the vector ABI is
enabled. In that case we *always* set up the overflow area
|| n_fpr + cfun->va_list_fpr_size > FP_ARG_NUM_REG
|| TARGET_VX_ABI)
{
- t = make_tree (TREE_TYPE (ovf), virtual_incoming_args_rtx);
+ if (cfun->machine->split_stack_varargs_pointer == NULL_RTX)
+ t = make_tree (TREE_TYPE (ovf), virtual_incoming_args_rtx);
+ else
+ t = make_tree (TREE_TYPE (ovf), cfun->machine->split_stack_varargs_pointer);
off = INTVAL (crtl->args.arg_offset_rtx);
off = off < 0 ? 0 : off;
size = int_size_in_bytes (type);
+ s390_check_type_for_vector_abi (type, true, false);
+
if (pass_by_reference (NULL, TYPE_MODE (type), type, false))
{
if (TARGET_DEBUG_ARG)
}
if (clobber_fprs_p)
- emit_insn (gen_tbegin_1 (gen_rtx_CONST_INT (VOIDmode, TBEGIN_MASK), tdb));
+ {
+ if (TARGET_VX)
+ emit_insn (gen_tbegin_1_z13 (gen_rtx_CONST_INT (VOIDmode, TBEGIN_MASK),
+ tdb));
+ else
+ emit_insn (gen_tbegin_1 (gen_rtx_CONST_INT (VOIDmode, TBEGIN_MASK),
+ tdb));
+ }
else
emit_insn (gen_tbegin_nofloat_1 (gen_rtx_CONST_INT (VOIDmode, TBEGIN_MASK),
tdb));
output_asm_insn ("brasl\t%0,%4", op);
output_asm_insn ("lg\t%0,%1", op);
}
+ else if (TARGET_CPU_ZARCH)
+ {
+ output_asm_insn ("st\t%0,%1", op);
+ output_asm_insn ("larl\t%2,%3", op);
+ output_asm_insn ("brasl\t%0,%4", op);
+ output_asm_insn ("l\t%0,%1", op);
+ }
else if (!flag_pic)
{
op[6] = gen_label_rtx ();
if (TREE_CODE (decl) == VAR_DECL)
{
- /* If a variable has a forced alignment to < 2 bytes, mark it
- with SYMBOL_FLAG_ALIGN1 to prevent it from being used as LARL
- operand. */
- if (DECL_USER_ALIGN (decl) && DECL_ALIGN (decl) < 16)
- SYMBOL_REF_FLAGS (XEXP (rtl, 0)) |= SYMBOL_FLAG_ALIGN1;
- if (!DECL_SIZE (decl)
- || !DECL_ALIGN (decl)
- || !tree_fits_shwi_p (DECL_SIZE (decl))
- || (DECL_ALIGN (decl) <= 64
- && DECL_ALIGN (decl) != tree_to_shwi (DECL_SIZE (decl))))
- SYMBOL_REF_FLAGS (XEXP (rtl, 0)) |= SYMBOL_FLAG_NOT_NATURALLY_ALIGNED;
+ /* Store the alignment to be able to check if we can use
+ a larl/load-relative instruction. We only handle the cases
+ that can go wrong (i.e. no FUNC_DECLs). If a symref does
+ not have any flag we assume it to be correctly aligned. */
+
+ if (DECL_ALIGN (decl) % 64)
+ SYMBOL_FLAG_SET_NOTALIGN8 (XEXP (rtl, 0));
+
+ if (DECL_ALIGN (decl) % 32)
+ SYMBOL_FLAG_SET_NOTALIGN4 (XEXP (rtl, 0));
+
+ if (DECL_ALIGN (decl) == 0 || DECL_ALIGN (decl) % 16)
+ SYMBOL_FLAG_SET_NOTALIGN2 (XEXP (rtl, 0));
}
/* Literal pool references don't have a decl so they are handled
differently here. We rely on the information in the MEM_ALIGN
- entry to decide upon natural alignment. */
+ entry to decide upon the alignment. */
if (MEM_P (rtl)
&& GET_CODE (XEXP (rtl, 0)) == SYMBOL_REF
&& TREE_CONSTANT_POOL_ADDRESS_P (XEXP (rtl, 0))
- && (MEM_ALIGN (rtl) == 0
- || GET_MODE_BITSIZE (GET_MODE (rtl)) == 0
- || MEM_ALIGN (rtl) < GET_MODE_BITSIZE (GET_MODE (rtl))))
- SYMBOL_REF_FLAGS (XEXP (rtl, 0)) |= SYMBOL_FLAG_NOT_NATURALLY_ALIGNED;
+ && MEM_ALIGN (rtl) != 0
+ && GET_MODE_BITSIZE (GET_MODE (rtl)) != 0)
+ {
+ if (MEM_ALIGN (rtl) % 64)
+ SYMBOL_FLAG_SET_NOTALIGN8 (XEXP (rtl, 0));
+
+ if (MEM_ALIGN (rtl) % 32)
+ SYMBOL_FLAG_SET_NOTALIGN4 (XEXP (rtl, 0));
+
+ if (MEM_ALIGN (rtl) == 0 || MEM_ALIGN (rtl) % 16)
+ SYMBOL_FLAG_SET_NOTALIGN2 (XEXP (rtl, 0));
+ }
}
/* Output thunk to FILE that implements a C++ virtual function call (with
replace the symbol itself with the PLT stub. */
if (flag_pic && !SYMBOL_REF_LOCAL_P (addr_location))
{
- if (retaddr_reg != NULL_RTX)
+ if (TARGET_64BIT || retaddr_reg != NULL_RTX)
{
addr_location = gen_rtx_UNSPEC (Pmode,
gen_rtvec (1, addr_location),
/* It must not happen that what we once saved in an FPR now
needs a stack slot. */
- gcc_assert (cfun_gpr_save_slot (gpr_regno) != -1);
+ gcc_assert (cfun_gpr_save_slot (gpr_regno) != SAVE_SLOT_STACK);
- if (cfun_gpr_save_slot (gpr_regno) == 0)
+ if (cfun_gpr_save_slot (gpr_regno) == SAVE_SLOT_NONE)
{
remove_insn (insn);
continue;
s390_optimize_prologue ();
/* Walk over the insns and do some >=z10 specific changes. */
- if (s390_tune == PROCESSOR_2097_Z10
- || s390_tune == PROCESSOR_2817_Z196
- || s390_tune == PROCESSOR_2827_ZEC12
- || s390_tune == PROCESSOR_2964_Z13)
+ if (s390_tune >= PROCESSOR_2097_Z10)
{
rtx_insn *insn;
bool insn_added_p = false;
/* Insert NOPs for hotpatching. */
for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
- {
- if (NOTE_P (insn) && NOTE_KIND (insn) == NOTE_INSN_FUNCTION_BEG)
- break;
- }
- gcc_assert (insn);
- /* Output a series of NOPs after the NOTE_INSN_FUNCTION_BEG. */
- while (hw_after > 0)
+ /* Emit NOPs
+ 1. inside the area covered by debug information to allow setting
+ breakpoints at the NOPs,
+ 2. before any insn which results in an asm instruction,
+ 3. before in-function labels to avoid jumping to the NOPs, for
+ example as part of a loop,
+ 4. before any barrier in case the function is completely empty
+ (__builtin_unreachable ()) and has neither internal labels nor
+ active insns.
+ */
+ if (active_insn_p (insn) || BARRIER_P (insn) || LABEL_P (insn))
+ break;
+ /* Output a series of NOPs before the first active insn. */
+ while (insn && hw_after > 0)
{
if (hw_after >= 3 && TARGET_CPU_ZARCH)
{
- insn = emit_insn_after (gen_nop_6_byte (), insn);
+ emit_insn_before (gen_nop_6_byte (), insn);
hw_after -= 3;
}
else if (hw_after >= 2)
{
- insn = emit_insn_after (gen_nop_4_byte (), insn);
+ emit_insn_before (gen_nop_4_byte (), insn);
hw_after -= 2;
}
else
{
- insn = emit_insn_after (gen_nop_2_byte (), insn);
+ emit_insn_before (gen_nop_2_byte (), insn);
hw_after -= 1;
}
}
- gcc_assert (hw_after == 0);
}
}
s390_sched_reorder (FILE *file, int verbose,
rtx_insn **ready, int *nreadyp, int clock ATTRIBUTE_UNUSED)
{
- if (s390_tune == PROCESSOR_2097_Z10)
- if (reload_completed && *nreadyp > 1)
- s390_z10_prevent_earlyload_conflicts (ready, nreadyp);
+ if (s390_tune == PROCESSOR_2097_Z10
+ && reload_completed
+ && *nreadyp > 1)
+ s390_z10_prevent_earlyload_conflicts (ready, nreadyp);
- if ((s390_tune == PROCESSOR_2827_ZEC12
- || s390_tune == PROCESSOR_2964_Z13)
+ if (s390_tune >= PROCESSOR_2827_ZEC12
&& reload_completed
&& *nreadyp > 1)
{
{
last_scheduled_insn = insn;
- if ((s390_tune == PROCESSOR_2827_ZEC12
- || s390_tune == PROCESSOR_2964_Z13)
+ if (s390_tune >= PROCESSOR_2827_ZEC12
&& reload_completed
&& recog_memoized (insn) >= 0)
{
The loop is analyzed for memory accesses by calling check_dpu for
each rtx of the loop. Depending on the loop_depth and the amount of
memory accesses a new number <=nunroll is returned to improve the
- behaviour of the hardware prefetch unit. */
+ behavior of the hardware prefetch unit. */
static unsigned
s390_loop_unroll_adjust (unsigned nunroll, struct loop *loop)
{
unsigned i;
unsigned mem_count = 0;
- if (s390_tune != PROCESSOR_2097_Z10
- && s390_tune != PROCESSOR_2817_Z196
- && s390_tune != PROCESSOR_2827_ZEC12
- && s390_tune != PROCESSOR_2964_Z13)
+ if (s390_tune < PROCESSOR_2097_Z10)
return nunroll;
/* Count the number of memory references within the loop body. */
}
}
+/* Restore the current options. This is a hook function and also called
+ internally. */
+
static void
-s390_option_override (void)
+s390_function_specific_restore (struct gcc_options *opts,
+ struct cl_target_option *ptr ATTRIBUTE_UNUSED)
{
- unsigned int i;
- cl_deferred_option *opt;
- vec<cl_deferred_option> *v =
- (vec<cl_deferred_option> *) s390_deferred_options;
+ opts->x_s390_cost_pointer = (long)processor_table[opts->x_s390_tune].cost;
+}
- if (v)
- FOR_EACH_VEC_ELT (*v, i, opt)
- {
- switch (opt->opt_index)
- {
- case OPT_mhotpatch_:
- {
- int val1;
- int val2;
- char s[256];
- char *t;
+static void
+s390_option_override_internal (bool main_args_p,
+ struct gcc_options *opts,
+ const struct gcc_options *opts_set)
+{
+ const char *prefix;
+ const char *suffix;
- strncpy (s, opt->arg, 256);
- s[255] = 0;
- t = strchr (s, ',');
- if (t != NULL)
- {
- *t = 0;
- t++;
- val1 = integral_argument (s);
- val2 = integral_argument (t);
- }
- else
- {
- val1 = -1;
- val2 = -1;
- }
- if (val1 == -1 || val2 == -1)
- {
- /* argument is not a plain number */
- error ("arguments to %qs should be non-negative integers",
- "-mhotpatch=n,m");
- break;
- }
- else if (val1 > s390_hotpatch_hw_max
- || val2 > s390_hotpatch_hw_max)
- {
- error ("argument to %qs is too large (max. %d)",
- "-mhotpatch=n,m", s390_hotpatch_hw_max);
- break;
- }
- s390_hotpatch_hw_before_label = val1;
- s390_hotpatch_hw_after_label = val2;
- break;
- }
- default:
- gcc_unreachable ();
- }
- }
+ /* Set up prefix/suffix so the error messages refer to either the command
+ line argument, or the attribute(target). */
+ if (main_args_p)
+ {
+ prefix = "-m";
+ suffix = "";
+ }
+ else
+ {
+ prefix = "option(\"";
+ suffix = "\")";
+ }
- /* Set up function hooks. */
- init_machine_status = s390_init_machine_status;
/* Architecture mode defaults according to ABI. */
- if (!(target_flags_explicit & MASK_ZARCH))
+ if (!(opts_set->x_target_flags & MASK_ZARCH))
{
if (TARGET_64BIT)
- target_flags |= MASK_ZARCH;
+ opts->x_target_flags |= MASK_ZARCH;
else
- target_flags &= ~MASK_ZARCH;
+ opts->x_target_flags &= ~MASK_ZARCH;
}
- /* Set the march default in case it hasn't been specified on
- cmdline. */
- if (s390_arch == PROCESSOR_max)
- {
- s390_arch_string = TARGET_ZARCH? "z900" : "g5";
- s390_arch = TARGET_ZARCH ? PROCESSOR_2064_Z900 : PROCESSOR_9672_G5;
- s390_arch_flags = processor_flags_table[(int)s390_arch];
- }
+ /* Set the march default in case it hasn't been specified on cmdline. */
+ if (!opts_set->x_s390_arch)
+ opts->x_s390_arch = PROCESSOR_2064_Z900;
+ else if (opts->x_s390_arch == PROCESSOR_9672_G5
+ || opts->x_s390_arch == PROCESSOR_9672_G6)
+ warning (OPT_Wdeprecated, "%sarch=%s%s is deprecated and will be removed "
+ "in future releases; use at least %sarch=z900%s",
+ prefix, opts->x_s390_arch == PROCESSOR_9672_G5 ? "g5" : "g6",
+ suffix, prefix, suffix);
+
+ opts->x_s390_arch_flags = processor_flags_table[(int) opts->x_s390_arch];
/* Determine processor to tune for. */
- if (s390_tune == PROCESSOR_max)
- {
- s390_tune = s390_arch;
- s390_tune_flags = s390_arch_flags;
- }
+ if (!opts_set->x_s390_tune)
+ opts->x_s390_tune = opts->x_s390_arch;
+ else if (opts->x_s390_tune == PROCESSOR_9672_G5
+ || opts->x_s390_tune == PROCESSOR_9672_G6)
+ warning (OPT_Wdeprecated, "%stune=%s%s is deprecated and will be removed "
+ "in future releases; use at least %stune=z900%s",
+ prefix, opts->x_s390_tune == PROCESSOR_9672_G5 ? "g5" : "g6",
+ suffix, prefix, suffix);
+
+ opts->x_s390_tune_flags = processor_flags_table[opts->x_s390_tune];
/* Sanity checks. */
- if (TARGET_ZARCH && !TARGET_CPU_ZARCH)
- error ("z/Architecture mode not supported on %s", s390_arch_string);
- if (TARGET_64BIT && !TARGET_ZARCH)
+ if (opts->x_s390_arch == PROCESSOR_NATIVE
+ || opts->x_s390_tune == PROCESSOR_NATIVE)
+ gcc_unreachable ();
+ if (TARGET_ZARCH_P (opts->x_target_flags) && !TARGET_CPU_ZARCH_P (opts))
+ error ("z/Architecture mode not supported on %s",
+ processor_table[(int)opts->x_s390_arch].name);
+ if (TARGET_64BIT && !TARGET_ZARCH_P (opts->x_target_flags))
error ("64-bit ABI not supported in ESA/390 mode");
- /* Use hardware DFP if available and not explicitly disabled by
- user. E.g. with -m31 -march=z10 -mzarch */
- if (!(target_flags_explicit & MASK_HARD_DFP) && TARGET_DFP)
- target_flags |= MASK_HARD_DFP;
-
/* Enable hardware transactions if available and not explicitly
disabled by user. E.g. with -m31 -march=zEC12 -mzarch */
- if (!(target_flags_explicit & MASK_OPT_HTM) && TARGET_CPU_HTM && TARGET_ZARCH)
- target_flags |= MASK_OPT_HTM;
+ if (!TARGET_OPT_HTM_P (opts_set->x_target_flags))
+ {
+ if (TARGET_CPU_HTM_P (opts) && TARGET_ZARCH_P (opts->x_target_flags))
+ opts->x_target_flags |= MASK_OPT_HTM;
+ else
+ opts->x_target_flags &= ~MASK_OPT_HTM;
+ }
- if (target_flags_explicit & MASK_OPT_VX)
+ if (TARGET_OPT_VX_P (opts_set->x_target_flags))
{
- if (TARGET_OPT_VX)
+ if (TARGET_OPT_VX_P (opts->x_target_flags))
{
- if (!TARGET_CPU_VX)
+ if (!TARGET_CPU_VX_P (opts))
error ("hardware vector support not available on %s",
- s390_arch_string);
- if (TARGET_SOFT_FLOAT)
+ processor_table[(int)opts->x_s390_arch].name);
+ if (TARGET_SOFT_FLOAT_P (opts->x_target_flags))
error ("hardware vector support not available with -msoft-float");
}
}
- else if (TARGET_CPU_VX)
- /* Enable vector support if available and not explicitly disabled
- by user. E.g. with -m31 -march=z13 -mzarch */
- target_flags |= MASK_OPT_VX;
+ else
+ {
+ if (TARGET_CPU_VX_P (opts))
+ /* Enable vector support if available and not explicitly disabled
+ by user. E.g. with -m31 -march=z13 -mzarch */
+ opts->x_target_flags |= MASK_OPT_VX;
+ else
+ opts->x_target_flags &= ~MASK_OPT_VX;
+ }
+
+ /* Use hardware DFP if available and not explicitly disabled by
+ user. E.g. with -m31 -march=z10 -mzarch */
+ if (!TARGET_HARD_DFP_P (opts_set->x_target_flags))
+ {
+ if (TARGET_DFP_P (opts))
+ opts->x_target_flags |= MASK_HARD_DFP;
+ else
+ opts->x_target_flags &= ~MASK_HARD_DFP;
+ }
- if (TARGET_HARD_DFP && !TARGET_DFP)
+ if (TARGET_HARD_DFP_P (opts->x_target_flags) && !TARGET_DFP_P (opts))
{
- if (target_flags_explicit & MASK_HARD_DFP)
+ if (TARGET_HARD_DFP_P (opts_set->x_target_flags))
{
- if (!TARGET_CPU_DFP)
+ if (!TARGET_CPU_DFP_P (opts))
error ("hardware decimal floating point instructions"
- " not available on %s", s390_arch_string);
- if (!TARGET_ZARCH)
+ " not available on %s",
+ processor_table[(int)opts->x_s390_arch].name);
+ if (!TARGET_ZARCH_P (opts->x_target_flags))
error ("hardware decimal floating point instructions"
" not available in ESA/390 mode");
}
else
- target_flags &= ~MASK_HARD_DFP;
+ opts->x_target_flags &= ~MASK_HARD_DFP;
}
- if ((target_flags_explicit & MASK_SOFT_FLOAT) && TARGET_SOFT_FLOAT)
+ if (TARGET_SOFT_FLOAT_P (opts_set->x_target_flags)
+ && TARGET_SOFT_FLOAT_P (opts->x_target_flags))
{
- if ((target_flags_explicit & MASK_HARD_DFP) && TARGET_HARD_DFP)
+ if (TARGET_HARD_DFP_P (opts_set->x_target_flags)
+ && TARGET_HARD_DFP_P (opts->x_target_flags))
error ("-mhard-dfp can%'t be used in conjunction with -msoft-float");
- target_flags &= ~MASK_HARD_DFP;
+ opts->x_target_flags &= ~MASK_HARD_DFP;
}
- /* Set processor cost function. */
- switch (s390_tune)
- {
- case PROCESSOR_2084_Z990:
- s390_cost = &z990_cost;
- break;
- case PROCESSOR_2094_Z9_109:
- s390_cost = &z9_109_cost;
- break;
- case PROCESSOR_2097_Z10:
- s390_cost = &z10_cost;
- break;
- case PROCESSOR_2817_Z196:
- s390_cost = &z196_cost;
- break;
- case PROCESSOR_2827_ZEC12:
- case PROCESSOR_2964_Z13:
- s390_cost = &zEC12_cost;
- break;
- default:
- s390_cost = &z900_cost;
- }
-
- if (TARGET_BACKCHAIN && TARGET_PACKED_STACK && TARGET_HARD_FLOAT)
+ if (TARGET_BACKCHAIN_P (opts->x_target_flags)
+ && TARGET_PACKED_STACK_P (opts->x_target_flags)
+ && TARGET_HARD_FLOAT_P (opts->x_target_flags))
error ("-mbackchain -mpacked-stack -mhard-float are not supported "
"in combination");
- if (s390_stack_size)
+ if (opts->x_s390_stack_size)
{
- if (s390_stack_guard >= s390_stack_size)
+ if (opts->x_s390_stack_guard >= opts->x_s390_stack_size)
error ("stack size must be greater than the stack guard value");
- else if (s390_stack_size > 1 << 16)
+ else if (opts->x_s390_stack_size > 1 << 16)
error ("stack size must not be greater than 64k");
}
- else if (s390_stack_guard)
+ else if (opts->x_s390_stack_guard)
error ("-mstack-guard implies use of -mstack-size");
#ifdef TARGET_DEFAULT_LONG_DOUBLE_128
- if (!(target_flags_explicit & MASK_LONG_DOUBLE_128))
- target_flags |= MASK_LONG_DOUBLE_128;
+ if (!TARGET_LONG_DOUBLE_128_P (opts_set->x_target_flags))
+ opts->x_target_flags |= MASK_LONG_DOUBLE_128;
#endif
- if (s390_tune == PROCESSOR_2097_Z10
- || s390_tune == PROCESSOR_2817_Z196
- || s390_tune == PROCESSOR_2827_ZEC12
- || s390_tune == PROCESSOR_2964_Z13)
+ if (opts->x_s390_tune >= PROCESSOR_2097_Z10)
{
maybe_set_param_value (PARAM_MAX_UNROLLED_INSNS, 100,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
maybe_set_param_value (PARAM_MAX_UNROLL_TIMES, 32,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
maybe_set_param_value (PARAM_MAX_COMPLETELY_PEELED_INSNS, 2000,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
maybe_set_param_value (PARAM_MAX_COMPLETELY_PEEL_TIMES, 64,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
}
maybe_set_param_value (PARAM_MAX_PENDING_LIST_LENGTH, 256,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
/* values for loop prefetching */
maybe_set_param_value (PARAM_L1_CACHE_LINE_SIZE, 256,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
maybe_set_param_value (PARAM_L1_CACHE_SIZE, 128,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
/* s390 has more than 2 levels and the size is much larger. Since
we are always running virtualized assume that we only get a small
part of the caches above l1. */
maybe_set_param_value (PARAM_L2_CACHE_SIZE, 1500,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
maybe_set_param_value (PARAM_PREFETCH_MIN_INSN_TO_MEM_RATIO, 2,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
maybe_set_param_value (PARAM_SIMULTANEOUS_PREFETCHES, 6,
- global_options.x_param_values,
- global_options_set.x_param_values);
+ opts->x_param_values,
+ opts_set->x_param_values);
+
+ /* Use the alternative scheduling-pressure algorithm by default. */
+ maybe_set_param_value (PARAM_SCHED_PRESSURE_ALGORITHM, 2,
+ opts->x_param_values,
+ opts_set->x_param_values);
+
+ /* Call target specific restore function to do post-init work. At the moment,
+ this just sets opts->x_s390_cost_pointer. */
+ s390_function_specific_restore (opts, NULL);
+}
+
+static void
+s390_option_override (void)
+{
+ unsigned int i;
+ cl_deferred_option *opt;
+ vec<cl_deferred_option> *v =
+ (vec<cl_deferred_option> *) s390_deferred_options;
+
+ if (v)
+ FOR_EACH_VEC_ELT (*v, i, opt)
+ {
+ switch (opt->opt_index)
+ {
+ case OPT_mhotpatch_:
+ {
+ int val1;
+ int val2;
+ char s[256];
+ char *t;
+
+ strncpy (s, opt->arg, 256);
+ s[255] = 0;
+ t = strchr (s, ',');
+ if (t != NULL)
+ {
+ *t = 0;
+ t++;
+ val1 = integral_argument (s);
+ val2 = integral_argument (t);
+ }
+ else
+ {
+ val1 = -1;
+ val2 = -1;
+ }
+ if (val1 == -1 || val2 == -1)
+ {
+ /* argument is not a plain number */
+ error ("arguments to %qs should be non-negative integers",
+ "-mhotpatch=n,m");
+ break;
+ }
+ else if (val1 > s390_hotpatch_hw_max
+ || val2 > s390_hotpatch_hw_max)
+ {
+ error ("argument to %qs is too large (max. %d)",
+ "-mhotpatch=n,m", s390_hotpatch_hw_max);
+ break;
+ }
+ s390_hotpatch_hw_before_label = val1;
+ s390_hotpatch_hw_after_label = val2;
+ break;
+ }
+ default:
+ gcc_unreachable ();
+ }
+ }
+
+ /* Set up function hooks. */
+ init_machine_status = s390_init_machine_status;
+
+ s390_option_override_internal (true, &global_options, &global_options_set);
+
+ /* Save the initial options in case the user does function specific
+ options. */
+ target_option_default_node = build_target_option_node (&global_options);
+ target_option_current_node = target_option_default_node;
/* This cannot reside in s390_option_optimization_table since HAVE_prefetch
requires the arch flags to be evaluated already. Since prefetching
if (flag_prefetch_loop_arrays < 0 && HAVE_prefetch && optimize >= 3)
flag_prefetch_loop_arrays = 1;
- /* Use the alternative scheduling-pressure algorithm by default. */
- maybe_set_param_value (PARAM_SCHED_PRESSURE_ALGORITHM, 2,
- global_options.x_param_values,
- global_options_set.x_param_values);
-
if (TARGET_TPF)
{
/* Don't emit DWARF3/4 unless specifically selected. The TPF
register_pass (&insert_pass_s390_early_mach);
}
+#if S390_USE_TARGET_ATTRIBUTE
+/* Inner function to process the attribute((target(...))), take an argument and
+ set the current options from the argument. If we have a list, recursively go
+ over the list. */
+
+static bool
+s390_valid_target_attribute_inner_p (tree args,
+ struct gcc_options *opts,
+ struct gcc_options *new_opts_set,
+ bool force_pragma)
+{
+ char *next_optstr;
+ bool ret = true;
+
+#define S390_ATTRIB(S,O,A) { S, sizeof (S)-1, O, A, 0 }
+#define S390_PRAGMA(S,O,A) { S, sizeof (S)-1, O, A, 1 }
+ static const struct
+ {
+ const char *string;
+ size_t len;
+ int opt;
+ int has_arg;
+ int only_as_pragma;
+ } attrs[] = {
+ /* enum options */
+ S390_ATTRIB ("arch=", OPT_march_, 1),
+ S390_ATTRIB ("tune=", OPT_mtune_, 1),
+ /* uinteger options */
+ S390_ATTRIB ("stack-guard=", OPT_mstack_guard_, 1),
+ S390_ATTRIB ("stack-size=", OPT_mstack_size_, 1),
+ S390_ATTRIB ("branch-cost=", OPT_mbranch_cost_, 1),
+ S390_ATTRIB ("warn-framesize=", OPT_mwarn_framesize_, 1),
+ /* flag options */
+ S390_ATTRIB ("backchain", OPT_mbackchain, 0),
+ S390_ATTRIB ("hard-dfp", OPT_mhard_dfp, 0),
+ S390_ATTRIB ("hard-float", OPT_mhard_float, 0),
+ S390_ATTRIB ("htm", OPT_mhtm, 0),
+ S390_ATTRIB ("vx", OPT_mvx, 0),
+ S390_ATTRIB ("packed-stack", OPT_mpacked_stack, 0),
+ S390_ATTRIB ("small-exec", OPT_msmall_exec, 0),
+ S390_ATTRIB ("soft-float", OPT_msoft_float, 0),
+ S390_ATTRIB ("mvcle", OPT_mmvcle, 0),
+ S390_PRAGMA ("zvector", OPT_mzvector, 0),
+ /* boolean options */
+ S390_ATTRIB ("warn-dynamicstack", OPT_mwarn_dynamicstack, 0),
+ };
+#undef S390_ATTRIB
+#undef S390_PRAGMA
+
+ /* If this is a list, recurse to get the options. */
+ if (TREE_CODE (args) == TREE_LIST)
+ {
+ bool ret = true;
+ int num_pragma_values;
+ int i;
+
+ /* Note: attribs.c:decl_attributes prepends the values from
+ current_target_pragma to the list of target attributes. To determine
+ whether we're looking at a value of the attribute or the pragma we
+ assume that the first [list_length (current_target_pragma)] values in
+ the list are the values from the pragma. */
+ num_pragma_values = (!force_pragma && current_target_pragma != NULL)
+ ? list_length (current_target_pragma) : 0;
+ for (i = 0; args; args = TREE_CHAIN (args), i++)
+ {
+ bool is_pragma;
+
+ is_pragma = (force_pragma || i < num_pragma_values);
+ if (TREE_VALUE (args)
+ && !s390_valid_target_attribute_inner_p (TREE_VALUE (args),
+ opts, new_opts_set,
+ is_pragma))
+ {
+ ret = false;
+ }
+ }
+ return ret;
+ }
+
+ else if (TREE_CODE (args) != STRING_CST)
+ {
+ error ("attribute %<target%> argument not a string");
+ return false;
+ }
+
+ /* Handle multiple arguments separated by commas. */
+ next_optstr = ASTRDUP (TREE_STRING_POINTER (args));
+
+ while (next_optstr && *next_optstr != '\0')
+ {
+ char *p = next_optstr;
+ char *orig_p = p;
+ char *comma = strchr (next_optstr, ',');
+ size_t len, opt_len;
+ int opt;
+ bool opt_set_p;
+ char ch;
+ unsigned i;
+ int mask = 0;
+ enum cl_var_type var_type;
+ bool found;
+
+ if (comma)
+ {
+ *comma = '\0';
+ len = comma - next_optstr;
+ next_optstr = comma + 1;
+ }
+ else
+ {
+ len = strlen (p);
+ next_optstr = NULL;
+ }
+
+ /* Recognize no-xxx. */
+ if (len > 3 && p[0] == 'n' && p[1] == 'o' && p[2] == '-')
+ {
+ opt_set_p = false;
+ p += 3;
+ len -= 3;
+ }
+ else
+ opt_set_p = true;
+
+ /* Find the option. */
+ ch = *p;
+ found = false;
+ for (i = 0; i < ARRAY_SIZE (attrs); i++)
+ {
+ opt_len = attrs[i].len;
+ if (ch == attrs[i].string[0]
+ && ((attrs[i].has_arg) ? len > opt_len : len == opt_len)
+ && memcmp (p, attrs[i].string, opt_len) == 0)
+ {
+ opt = attrs[i].opt;
+ if (!opt_set_p && cl_options[opt].cl_reject_negative)
+ continue;
+ mask = cl_options[opt].var_value;
+ var_type = cl_options[opt].var_type;
+ found = true;
+ break;
+ }
+ }
+
+ /* Process the option. */
+ if (!found)
+ {
+ error ("attribute(target(\"%s\")) is unknown", orig_p);
+ return false;
+ }
+ else if (attrs[i].only_as_pragma && !force_pragma)
+ {
+ /* Value is not allowed for the target attribute. */
+ error ("Value %qs is not supported by attribute %<target%>",
+ attrs[i].string);
+ return false;
+ }
+
+ else if (var_type == CLVC_BIT_SET || var_type == CLVC_BIT_CLEAR)
+ {
+ if (var_type == CLVC_BIT_CLEAR)
+ opt_set_p = !opt_set_p;
+
+ if (opt_set_p)
+ opts->x_target_flags |= mask;
+ else
+ opts->x_target_flags &= ~mask;
+ new_opts_set->x_target_flags |= mask;
+ }
+
+ else if (cl_options[opt].var_type == CLVC_BOOLEAN)
+ {
+ int value;
+
+ if (cl_options[opt].cl_uinteger)
+ {
+ /* Unsigned integer argument. Code based on the function
+ decode_cmdline_option () in opts-common.c. */
+ value = integral_argument (p + opt_len);
+ }
+ else
+ value = (opt_set_p) ? 1 : 0;
+
+ if (value != -1)
+ {
+ struct cl_decoded_option decoded;
+
+ /* Value range check; only implemented for numeric and boolean
+ options at the moment. */
+ generate_option (opt, NULL, value, CL_TARGET, &decoded);
+ s390_handle_option (opts, new_opts_set, &decoded, input_location);
+ set_option (opts, new_opts_set, opt, value,
+ p + opt_len, DK_UNSPECIFIED, input_location,
+ global_dc);
+ }
+ else
+ {
+ error ("attribute(target(\"%s\")) is unknown", orig_p);
+ ret = false;
+ }
+ }
+
+ else if (cl_options[opt].var_type == CLVC_ENUM)
+ {
+ bool arg_ok;
+ int value;
+
+ arg_ok = opt_enum_arg_to_value (opt, p + opt_len, &value, CL_TARGET);
+ if (arg_ok)
+ set_option (opts, new_opts_set, opt, value,
+ p + opt_len, DK_UNSPECIFIED, input_location,
+ global_dc);
+ else
+ {
+ error ("attribute(target(\"%s\")) is unknown", orig_p);
+ ret = false;
+ }
+ }
+
+ else
+ gcc_unreachable ();
+ }
+ return ret;
+}
+
+/* Return a TARGET_OPTION_NODE tree of the target options listed or NULL. */
+
+tree
+s390_valid_target_attribute_tree (tree args,
+ struct gcc_options *opts,
+ const struct gcc_options *opts_set,
+ bool force_pragma)
+{
+ tree t = NULL_TREE;
+ struct gcc_options new_opts_set;
+
+ memset (&new_opts_set, 0, sizeof (new_opts_set));
+
+ /* Process each of the options on the chain. */
+ if (! s390_valid_target_attribute_inner_p (args, opts, &new_opts_set,
+ force_pragma))
+ return error_mark_node;
+
+ /* If some option was set (even if it has not changed), rerun
+ s390_option_override_internal, and then save the options away. */
+ if (new_opts_set.x_target_flags
+ || new_opts_set.x_s390_arch
+ || new_opts_set.x_s390_tune
+ || new_opts_set.x_s390_stack_guard
+ || new_opts_set.x_s390_stack_size
+ || new_opts_set.x_s390_branch_cost
+ || new_opts_set.x_s390_warn_framesize
+ || new_opts_set.x_s390_warn_dynamicstack_p)
+ {
+ const unsigned char *src = (const unsigned char *)opts_set;
+ unsigned char *dest = (unsigned char *)&new_opts_set;
+ unsigned int i;
+
+ /* Merge the original option flags into the new ones. */
+ for (i = 0; i < sizeof(*opts_set); i++)
+ dest[i] |= src[i];
+
+ /* Do any overrides, such as arch=xxx, or tune=xxx support. */
+ s390_option_override_internal (false, opts, &new_opts_set);
+ /* Save the current options unless we are validating options for
+ #pragma. */
+ t = build_target_option_node (opts);
+ }
+ return t;
+}
+
+/* Hook to validate attribute((target("string"))). */
+
+static bool
+s390_valid_target_attribute_p (tree fndecl,
+ tree ARG_UNUSED (name),
+ tree args,
+ int ARG_UNUSED (flags))
+{
+ struct gcc_options func_options;
+ tree new_target, new_optimize;
+ bool ret = true;
+
+ /* attribute((target("default"))) does nothing, beyond
+ affecting multi-versioning. */
+ if (TREE_VALUE (args)
+ && TREE_CODE (TREE_VALUE (args)) == STRING_CST
+ && TREE_CHAIN (args) == NULL_TREE
+ && strcmp (TREE_STRING_POINTER (TREE_VALUE (args)), "default") == 0)
+ return true;
+
+ tree old_optimize = build_optimization_node (&global_options);
+
+ /* Get the optimization options of the current function. */
+ tree func_optimize = DECL_FUNCTION_SPECIFIC_OPTIMIZATION (fndecl);
+
+ if (!func_optimize)
+ func_optimize = old_optimize;
+
+ /* Init func_options. */
+ memset (&func_options, 0, sizeof (func_options));
+ init_options_struct (&func_options, NULL);
+ lang_hooks.init_options_struct (&func_options);
+
+ cl_optimization_restore (&func_options, TREE_OPTIMIZATION (func_optimize));
+
+ /* Initialize func_options to the default before its target options can
+ be set. */
+ cl_target_option_restore (&func_options,
+ TREE_TARGET_OPTION (target_option_default_node));
+
+ new_target = s390_valid_target_attribute_tree (args, &func_options,
+ &global_options_set,
+ (args ==
+ current_target_pragma));
+ new_optimize = build_optimization_node (&func_options);
+ if (new_target == error_mark_node)
+ ret = false;
+ else if (fndecl && new_target)
+ {
+ DECL_FUNCTION_SPECIFIC_TARGET (fndecl) = new_target;
+ if (old_optimize != new_optimize)
+ DECL_FUNCTION_SPECIFIC_OPTIMIZATION (fndecl) = new_optimize;
+ }
+ return ret;
+}
+
+/* Restore targets globals from NEW_TREE and invalidate s390_previous_fndecl
+ cache. */
+
+void
+s390_activate_target_options (tree new_tree)
+{
+ cl_target_option_restore (&global_options, TREE_TARGET_OPTION (new_tree));
+ if (TREE_TARGET_GLOBALS (new_tree))
+ restore_target_globals (TREE_TARGET_GLOBALS (new_tree));
+ else if (new_tree == target_option_default_node)
+ restore_target_globals (&default_target_globals);
+ else
+ TREE_TARGET_GLOBALS (new_tree) = save_target_globals_default_opts ();
+ s390_previous_fndecl = NULL_TREE;
+}
+
+/* Establish appropriate back-end context for processing the function
+ FNDECL. The argument might be NULL to indicate processing at top
+ level, outside of any function scope. */
+static void
+s390_set_current_function (tree fndecl)
+{
+ /* Only change the context if the function changes. This hook is called
+ several times in the course of compiling a function, and we don't want to
+ slow things down too much or call target_reinit when it isn't safe. */
+ if (fndecl == s390_previous_fndecl)
+ return;
+
+ tree old_tree;
+ if (s390_previous_fndecl == NULL_TREE)
+ old_tree = target_option_current_node;
+ else if (DECL_FUNCTION_SPECIFIC_TARGET (s390_previous_fndecl))
+ old_tree = DECL_FUNCTION_SPECIFIC_TARGET (s390_previous_fndecl);
+ else
+ old_tree = target_option_default_node;
+
+ if (fndecl == NULL_TREE)
+ {
+ if (old_tree != target_option_current_node)
+ s390_activate_target_options (target_option_current_node);
+ return;
+ }
+
+ tree new_tree = DECL_FUNCTION_SPECIFIC_TARGET (fndecl);
+ if (new_tree == NULL_TREE)
+ new_tree = target_option_default_node;
+
+ if (old_tree != new_tree)
+ s390_activate_target_options (new_tree);
+ s390_previous_fndecl = fndecl;
+}
+#endif
+
/* Implement TARGET_USE_BY_PIECES_INFRASTRUCTURE_P. */
static bool
tree sfpc = s390_builtin_decls[S390_BUILTIN_s390_sfpc];
tree efpc = s390_builtin_decls[S390_BUILTIN_s390_efpc];
tree call_efpc = build_call_expr (efpc, 0);
- tree fenv_var = create_tmp_var (unsigned_type_node);
+ tree fenv_var = create_tmp_var_raw (unsigned_type_node);
#define FPC_EXCEPTION_MASK HOST_WIDE_INT_UC (0xf8000000)
#define FPC_FLAGS_MASK HOST_WIDE_INT_UC (0x00f80000)
__builtin_s390_sfpc (fenv_var);
__atomic_feraiseexcept ((old_fpc & FPC_FLAGS_MASK) >> FPC_FLAGS_SHIFT); */
- old_fpc = create_tmp_var (unsigned_type_node);
+ old_fpc = create_tmp_var_raw (unsigned_type_node);
tree store_old_fpc = build2 (MODIFY_EXPR, void_type_node,
old_fpc, call_efpc);
int misalignment ATTRIBUTE_UNUSED,
bool is_packed ATTRIBUTE_UNUSED)
{
- return true;
+ if (TARGET_VX)
+ return true;
+
+ return default_builtin_support_vector_misalignment (mode, type, misalignment,
+ is_packed);
}
/* The vector ABI requires vector types to be aligned on an 8 byte
return MIN (64, tree_to_shwi (TYPE_SIZE (type)));
}
+#ifdef HAVE_AS_MACHINE_MACHINEMODE
+/* Implement TARGET_ASM_FILE_START. */
+static void
+s390_asm_file_start (void)
+{
+ s390_asm_output_machine_for_arch (asm_out_file);
+}
+#endif
+
+/* Implement TARGET_ASM_FILE_END. */
+static void
+s390_asm_file_end (void)
+{
+#ifdef HAVE_AS_GNU_ATTRIBUTE
+ varpool_node *vnode;
+ cgraph_node *cnode;
+
+ FOR_EACH_VARIABLE (vnode)
+ if (TREE_PUBLIC (vnode->decl))
+ s390_check_type_for_vector_abi (TREE_TYPE (vnode->decl), false, false);
+
+ FOR_EACH_FUNCTION (cnode)
+ if (TREE_PUBLIC (cnode->decl))
+ s390_check_type_for_vector_abi (TREE_TYPE (cnode->decl), false, false);
+
+
+ if (s390_vector_abi != 0)
+ fprintf (asm_out_file, "\t.gnu_attribute 8, %d\n",
+ s390_vector_abi);
+#endif
+ file_end_indicate_exec_stack ();
+
+ if (flag_split_stack)
+ file_end_indicate_split_stack ();
+}
/* Return true if TYPE is a vector bool type. */
static inline bool
#undef TARGET_SET_UP_BY_PROLOGUE
#define TARGET_SET_UP_BY_PROLOGUE s300_set_up_by_prologue
+#undef TARGET_EXTRA_LIVE_ON_ENTRY
+#define TARGET_EXTRA_LIVE_ON_ENTRY s390_live_on_entry
+
#undef TARGET_USE_BY_PIECES_INFRASTRUCTURE_P
#define TARGET_USE_BY_PIECES_INFRASTRUCTURE_P \
s390_use_by_pieces_infrastructure_p
#undef TARGET_INVALID_BINARY_OP
#define TARGET_INVALID_BINARY_OP s390_invalid_binary_op
+#ifdef HAVE_AS_MACHINE_MACHINEMODE
+#undef TARGET_ASM_FILE_START
+#define TARGET_ASM_FILE_START s390_asm_file_start
+#endif
+
+#undef TARGET_ASM_FILE_END
+#define TARGET_ASM_FILE_END s390_asm_file_end
+
+#if S390_USE_TARGET_ATTRIBUTE
+#undef TARGET_SET_CURRENT_FUNCTION
+#define TARGET_SET_CURRENT_FUNCTION s390_set_current_function
+
+#undef TARGET_OPTION_VALID_ATTRIBUTE_P
+#define TARGET_OPTION_VALID_ATTRIBUTE_P s390_valid_target_attribute_p
+#endif
+
+#undef TARGET_OPTION_RESTORE
+#define TARGET_OPTION_RESTORE s390_function_specific_restore
+
struct gcc_target targetm = TARGET_INITIALIZER;
#include "gt-s390.h"