#define TARGET_CPU_sparcv9 7 /* alias */
#define TARGET_CPU_sparc64 7 /* alias */
#define TARGET_CPU_ultrasparc 8
+#define TARGET_CPU_ultrasparc3 9
#if TARGET_CPU_DEFAULT == TARGET_CPU_v9 \
- || TARGET_CPU_DEFAULT == TARGET_CPU_ultrasparc
+ || TARGET_CPU_DEFAULT == TARGET_CPU_ultrasparc \
+ || TARGET_CPU_DEFAULT == TARGET_CPU_ultrasparc3
#define CPP_CPU32_DEFAULT_SPEC ""
#define ASM_CPU32_DEFAULT_SPEC ""
#define CPP_CPU64_DEFAULT_SPEC "-D__sparc_v9__"
#define ASM_CPU64_DEFAULT_SPEC "-Av9a"
#endif
+#if TARGET_CPU_DEFAULT == TARGET_CPU_ultrasparc3
+#define CPP_CPU64_DEFAULT_SPEC "-D__sparc_v9__"
+#define ASM_CPU64_DEFAULT_SPEC "-Av9b"
+#endif
#else
%{mcpu=sparclite86x:-D__sparclite86x__} \
%{mcpu=v9:-D__sparc_v9__} \
%{mcpu=ultrasparc:-D__sparc_v9__} \
+%{mcpu=ultrasparc3:-D__sparc_v9__} \
%{!mcpu*:%{!mcypress:%{!msparclite:%{!mf930:%{!mf934:%{!mv8:%{!msupersparc:%(cpp_cpu_default)}}}}}}} \
"
sparc64 in 32 bit environments, so for now we only use `sparc64' in
64 bit environments. */
-#ifdef SPARC_BI_ARCH
-
-#define CPP_ARCH32_SPEC "-D__SIZE_TYPE__=unsigned\\ int -D__PTRDIFF_TYPE__=int \
--D__GCC_NEW_VARARGS__ -Acpu=sparc -Amachine=sparc"
-#define CPP_ARCH64_SPEC "-D__SIZE_TYPE__=long\\ unsigned\\ int -D__PTRDIFF_TYPE__=long\\ int \
--D__arch64__ -Acpu=sparc64 -Amachine=sparc64"
-
-#else
-
#define CPP_ARCH32_SPEC "-D__GCC_NEW_VARARGS__ -Acpu=sparc -Amachine=sparc"
#define CPP_ARCH64_SPEC "-D__arch64__ -Acpu=sparc64 -Amachine=sparc64"
-#endif
-
#define CPP_ARCH_DEFAULT_SPEC \
(DEFAULT_ARCH32_P ? CPP_ARCH32_SPEC : CPP_ARCH64_SPEC)
%{mv8plus:-Av8plus} \
%{mcpu=v9:-Av9} \
%{mcpu=ultrasparc:%{!mv8plus:-Av9a}} \
+%{mcpu=ultrasparc3:%{!mv8plus:-Av9b}} \
%{!mcpu*:%{!mcypress:%{!msparclite:%{!mf930:%{!mf934:%{!mv8:%{!msupersparc:%(asm_cpu_default)}}}}}}} \
"
SUBTARGET_EXTRA_SPECS
#define SUBTARGET_EXTRA_SPECS
+
+/* Because libgcc can generate references back to libc (via .umul etc.) we have
+ to list libc again after the second libgcc. */
+#define LINK_GCC_C_SEQUENCE_SPEC "%G %L %G %L"
+
\f
-#ifdef SPARC_BI_ARCH
-#define NO_BUILTIN_PTRDIFF_TYPE
-#define NO_BUILTIN_SIZE_TYPE
-#endif
#define PTRDIFF_TYPE (TARGET_ARCH64 ? "long int" : "int")
#define SIZE_TYPE (TARGET_ARCH64 ? "long unsigned int" : "unsigned int")
/* ??? This should be 32 bits for v9 but what can we do? */
#define WCHAR_TYPE "short unsigned int"
#define WCHAR_TYPE_SIZE 16
-#define MAX_WCHAR_TYPE_SIZE 16
/* Show we can debug even without a frame pointer. */
#define CAN_DEBUG_WITHOUT_FP
-/* To make profiling work with -f{pic,PIC}, we need to emit the profiling
- code into the rtl. Also, if we are profiling, we cannot eliminate
- the frame pointer (because the return address will get smashed). */
-
-#define OVERRIDE_OPTIONS \
- do { \
- if (profile_flag || profile_arc_flag) \
- { \
- if (flag_pic) \
- { \
- const char *const pic_string = (flag_pic == 1) ? "-fpic" : "-fPIC";\
- warning ("%s and profiling conflict: disabling %s", \
- pic_string, pic_string); \
- flag_pic = 0; \
- } \
- flag_omit_frame_pointer = 0; \
- } \
- sparc_override_options (); \
- SUBTARGET_OVERRIDE_OPTIONS; \
- } while (0)
-
-/* This is meant to be redefined in the host dependent files. */
-#define SUBTARGET_OVERRIDE_OPTIONS
+#define OVERRIDE_OPTIONS sparc_override_options ()
/* Generate DBX debugging information. */
#define MASK_FPU 1
#define TARGET_FPU (target_flags & MASK_FPU)
-/* Nonzero if we should use function_epilogue(). Otherwise, we
- use fast return insns, but lose some generality. */
-#define MASK_EPILOGUE 2
-#define TARGET_EPILOGUE (target_flags & MASK_EPILOGUE)
-
/* Nonzero if we should assume that double pointers might be unaligned.
This can happen when linking gcc compiled code with other compilers,
because the ABI only guarantees 4 byte alignment. */
{"soft-float", -MASK_FPU, \
N_("Do not use hardware fp") }, \
{"soft-float", MASK_FPU_SET, NULL }, \
- {"epilogue", MASK_EPILOGUE, \
- N_("Use function_epilogue()") }, \
- {"no-epilogue", -MASK_EPILOGUE, \
- N_("Do not use function_epilogue()") }, \
{"unaligned-doubles", MASK_UNALIGNED_DOUBLES, \
N_("Assume possible double misalignment") }, \
{"no-unaligned-doubles", -MASK_UNALIGNED_DOUBLES, \
/* MASK_APP_REGS must always be the default because that's what
FIXED_REGISTERS is set to and -ffixed- is processed before
CONDITIONAL_REGISTER_USAGE is called (where we process -mno-app-regs). */
-#define TARGET_DEFAULT (MASK_APP_REGS + MASK_EPILOGUE + MASK_FPU)
+#define TARGET_DEFAULT (MASK_APP_REGS + MASK_FPU)
/* This is meant to be redefined in target specific files. */
#define SUBTARGET_SWITCHES
PROCESSOR_SPARCLET,
PROCESSOR_TSC701,
PROCESSOR_V9,
- PROCESSOR_ULTRASPARC
+ PROCESSOR_ULTRASPARC,
+ PROCESSOR_ULTRASPARC3
};
/* This is set from -m{cpu,tune}=xxx. */
/* Width of a word, in units (bytes). */
#define UNITS_PER_WORD (TARGET_ARCH64 ? 8 : 4)
+#ifdef IN_LIBGCC2
+#define MIN_UNITS_PER_WORD UNITS_PER_WORD
+#else
#define MIN_UNITS_PER_WORD 4
+#endif
/* Now define the sizes of the C data types. */
#ifndef SUNOS4_SHARED_LIBRARIES
#define SUNOS4_SHARED_LIBRARIES 0
#endif
-
-
-/* Use text section for a constant
- unless we need more alignment than that offers. */
-/* This is defined differently for v9 in a cover file. */
-#define SELECT_RTX_SECTION(MODE, X, ALIGN) \
-{ \
- if (GET_MODE_BITSIZE (MODE) <= MAX_TEXT_ALIGN \
- && ! (flag_pic && (symbolic_operand ((X), (MODE)) || SUNOS4_SHARED_LIBRARIES))) \
- text_section (); \
- else \
- data_section (); \
-}
\f
/* Standard register usage. */
fixed_regs[4] = 0; \
if (TARGET_FLAT) \
{ \
+ int regno; \
/* Let the compiler believe the frame pointer is still \
%fp, but output it as %i7. */ \
fixed_regs[31] = 1; \
reg_names[HARD_FRAME_POINTER_REGNUM] = "%i7"; \
/* Disable leaf functions */ \
memset (sparc_leaf_regs, 0, FIRST_PSEUDO_REGISTER); \
+ /* Make LEAF_REG_REMAP a noop. */ \
+ for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++) \
+ leaf_reg_remap [regno] = regno; \
} \
} \
while (0)
/* This is the order in which to allocate registers normally.
- We put %f0/%f1 last among the float registers, so as to make it more
+ We put %f0-%f7 last among the float registers, so as to make it more
likely that a pseudo-register which dies in the float return register
- will get allocated to the float return register, thus saving a move
- instruction at the end of the function. */
+ area will get allocated to the float return register, thus saving a move
+ instruction at the end of the function.
+
+ Similarly for integer return value registers.
+
+ We know in this case that we will not end up with a leaf function.
+
+ The register allocater is given the global and out registers first
+ because these registers are call clobbered and thus less useful to
+ global register allocation.
+
+ Next we list the local and in registers. They are not call clobbered
+ and thus very useful for global register allocation. We list the input
+ registers before the locals so that it is more likely the incoming
+ arguments received in those registers can just stay there and not be
+ reloaded. */
#define REG_ALLOC_ORDER \
-{ 8, 9, 10, 11, 12, 13, 2, 3, \
- 15, 16, 17, 18, 19, 20, 21, 22, \
- 23, 24, 25, 26, 27, 28, 29, 31, \
- 34, 35, 36, 37, 38, 39, /* %f2-%f7 */ \
+{ 1, 2, 3, 4, 5, 6, 7, /* %g1-%g7 */ \
+ 13, 12, 11, 10, 9, 8, /* %o5-%o0 */ \
+ 15, /* %o7 */ \
+ 16, 17, 18, 19, 20, 21, 22, 23, /* %l0-%l7 */ \
+ 29, 28, 27, 26, 25, 24, 31, /* %i5-%i0,%i7 */\
40, 41, 42, 43, 44, 45, 46, 47, /* %f8-%f15 */ \
48, 49, 50, 51, 52, 53, 54, 55, /* %f16-%f23 */ \
56, 57, 58, 59, 60, 61, 62, 63, /* %f24-%f31 */ \
72, 73, 74, 75, 76, 77, 78, 79, /* %f40-%f47 */ \
80, 81, 82, 83, 84, 85, 86, 87, /* %f48-%f55 */ \
88, 89, 90, 91, 92, 93, 94, 95, /* %f56-%f63 */ \
- 32, 33, /* %f0,%f1 */ \
- 96, 97, 98, 99, 100, /* %fcc0-3, %icc */ \
- 1, 4, 5, 6, 7, 0, 14, 30, 101}
+ 39, 38, 37, 36, 35, 34, 33, 32, /* %f7-%f0 */ \
+ 96, 97, 98, 99, /* %fcc0-3 */ \
+ 100, 0, 14, 30, 101} /* %icc, %g0, %o6, %i6, %sfp */
/* This is the order in which to allocate registers for
- leaf functions. If all registers can fit in the "gi" registers,
- then we have the possibility of having a leaf function. */
+ leaf functions. If all registers can fit in the global and
+ output registers, then we have the possibility of having a leaf
+ function.
+
+ The macro actually mentioned the input registers first,
+ because they get renumbered into the output registers once
+ we know really do have a leaf function.
+
+ To be more precise, this register allocation order is used
+ when %o7 is found to not be clobbered right before register
+ allocation. Normally, the reason %o7 would be clobbered is
+ due to a call which could not be transformed into a sibling
+ call.
+
+ As a consequence, it is possible to use the leaf register
+ allocation order and not end up with a leaf function. We will
+ not get suboptimal register allocation in that case because by
+ definition of being potentially leaf, there were no function
+ calls. Therefore, allocation order within the local register
+ window is not critical like it is when we do have function calls. */
#define REG_LEAF_ALLOC_ORDER \
-{ 2, 3, 24, 25, 26, 27, 28, 29, \
- 4, 5, 6, 7, 1, \
- 15, 8, 9, 10, 11, 12, 13, \
- 16, 17, 18, 19, 20, 21, 22, 23, \
- 34, 35, 36, 37, 38, 39, \
- 40, 41, 42, 43, 44, 45, 46, 47, \
- 48, 49, 50, 51, 52, 53, 54, 55, \
- 56, 57, 58, 59, 60, 61, 62, 63, \
- 64, 65, 66, 67, 68, 69, 70, 71, \
- 72, 73, 74, 75, 76, 77, 78, 79, \
- 80, 81, 82, 83, 84, 85, 86, 87, \
- 88, 89, 90, 91, 92, 93, 94, 95, \
- 32, 33, \
- 96, 97, 98, 99, 100, \
- 0, 14, 30, 31, 101}
+{ 1, 2, 3, 4, 5, 6, 7, /* %g1-%g7 */ \
+ 29, 28, 27, 26, 25, 24, /* %i5-%i0 */ \
+ 15, /* %o7 */ \
+ 13, 12, 11, 10, 9, 8, /* %o5-%o0 */ \
+ 16, 17, 18, 19, 20, 21, 22, 23, /* %l0-%l7 */ \
+ 40, 41, 42, 43, 44, 45, 46, 47, /* %f8-%f15 */ \
+ 48, 49, 50, 51, 52, 53, 54, 55, /* %f16-%f23 */ \
+ 56, 57, 58, 59, 60, 61, 62, 63, /* %f24-%f31 */ \
+ 64, 65, 66, 67, 68, 69, 70, 71, /* %f32-%f39 */ \
+ 72, 73, 74, 75, 76, 77, 78, 79, /* %f40-%f47 */ \
+ 80, 81, 82, 83, 84, 85, 86, 87, /* %f48-%f55 */ \
+ 88, 89, 90, 91, 92, 93, 94, 95, /* %f56-%f63 */ \
+ 39, 38, 37, 36, 35, 34, 33, 32, /* %f7-%f0 */ \
+ 96, 97, 98, 99, /* %fcc0-3 */ \
+ 100, 0, 14, 30, 31, 101} /* %icc, %g0, %o6, %i6, %i7, %sfp */
#define ORDER_REGS_FOR_LOCAL_ALLOC order_regs_for_local_alloc ()
extern char sparc_leaf_regs[];
#define LEAF_REGISTERS sparc_leaf_regs
-extern const char leaf_reg_remap[];
+extern char leaf_reg_remap[];
#define LEAF_REG_REMAP(REGNO) (leaf_reg_remap[REGNO])
/* The class value for index registers, and the one for base regs. */
We need a temporary when loading/storing a HImode/QImode value
between memory and the FPU registers. This can happen when combine puts
- a paradoxical subreg in a float/fix conversion insn. */
+ a paradoxical subreg in a float/fix conversion insn.
+
+ We need a temporary when loading/storing a DFmode value between
+ unaligned memory and the upper FPU registers. */
#define SECONDARY_INPUT_RELOAD_CLASS(CLASS, MODE, IN) \
((FP_REG_CLASS_P (CLASS) \
|| ((GET_CODE (IN) == REG || GET_CODE (IN) == SUBREG) \
&& true_regnum (IN) == -1))) \
? GENERAL_REGS \
- : (((TARGET_CM_MEDANY \
- && symbolic_operand ((IN), (MODE))) \
- || (TARGET_CM_EMBMEDANY \
- && text_segment_operand ((IN), (MODE)))) \
- && !flag_pic) \
- ? GENERAL_REGS \
- : NO_REGS)
+ : ((CLASS) == EXTRA_FP_REGS && (MODE) == DFmode \
+ && GET_CODE (IN) == MEM && TARGET_ARCH32 \
+ && ! mem_min_alignment ((IN), 8)) \
+ ? FP_REGS \
+ : (((TARGET_CM_MEDANY \
+ && symbolic_operand ((IN), (MODE))) \
+ || (TARGET_CM_EMBMEDANY \
+ && text_segment_operand ((IN), (MODE)))) \
+ && !flag_pic) \
+ ? GENERAL_REGS \
+ : NO_REGS)
#define SECONDARY_OUTPUT_RELOAD_CLASS(CLASS, MODE, IN) \
- ((FP_REG_CLASS_P (CLASS) \
+ ((FP_REG_CLASS_P (CLASS) \
&& ((MODE) == HImode || (MODE) == QImode) \
&& (GET_CODE (IN) == MEM \
|| ((GET_CODE (IN) == REG || GET_CODE (IN) == SUBREG) \
&& true_regnum (IN) == -1))) \
- ? GENERAL_REGS \
- : (((TARGET_CM_MEDANY \
- && symbolic_operand ((IN), (MODE))) \
- || (TARGET_CM_EMBMEDANY \
- && text_segment_operand ((IN), (MODE)))) \
- && !flag_pic) \
- ? GENERAL_REGS \
- : NO_REGS)
+ ? GENERAL_REGS \
+ : ((CLASS) == EXTRA_FP_REGS && (MODE) == DFmode \
+ && GET_CODE (IN) == MEM && TARGET_ARCH32 \
+ && ! mem_min_alignment ((IN), 8)) \
+ ? FP_REGS \
+ : (((TARGET_CM_MEDANY \
+ && symbolic_operand ((IN), (MODE))) \
+ || (TARGET_CM_EMBMEDANY \
+ && text_segment_operand ((IN), (MODE)))) \
+ && !flag_pic) \
+ ? GENERAL_REGS \
+ : NO_REGS)
/* On SPARC it is not possible to directly move data between
GENERAL_REGS and FP_REGS. */
stored from the compare operation. Note that we can't use "rtx" here
since it hasn't been defined! */
-extern struct rtx_def *sparc_compare_op0, *sparc_compare_op1;
+extern GTY(()) rtx sparc_compare_op0;
+extern GTY(()) rtx sparc_compare_op1;
\f
/* Generate the special assembly code needed to tell the assembler whatever
#endif
\f
-/* Output assembler code to FILE to increment profiler label # LABELNO
- for profiling a function entry. */
+/* Emit rtl for profiling. */
+#define PROFILE_HOOK(LABEL) sparc_profile_hook (LABEL)
-#define FUNCTION_PROFILER(FILE, LABELNO) \
- sparc_function_profiler(FILE, LABELNO)
+/* All the work done in PROFILE_HOOK, but still required. */
+#define FUNCTION_PROFILER(FILE, LABELNO) do { } while (0)
/* Set the name of the mcount function for the system. */
-
#define MCOUNT_FUNCTION "*mcount"
\f
/* EXIT_IGNORE_STACK should be nonzero if, when returning from a function,
#define STRICT_ARGUMENT_NAMING TARGET_V9
/* We do not allow sibling calls if -mflat, nor
- we do not allow indirect calls to be optimized into sibling calls. */
-#define FUNCTION_OK_FOR_SIBCALL(DECL) (DECL && ! TARGET_FLAT)
+ we do not allow indirect calls to be optimized into sibling calls.
+
+ Also, on sparc 32-bit we cannot emit a sibling call when the
+ current function returns a structure. This is because the "unimp
+ after call" convention would cause the callee to return to the
+ wrong place. The generic code already disallows cases where the
+ function being called returns a structure.
+
+ It may seem strange how this last case could occur. Usually there
+ is code after the call which jumps to epilogue code which dumps the
+ return value into the struct return area. That ought to invalidate
+ the sibling call right? Well, in the c++ case we can end up passing
+ the pointer to the struct return area to a constructor (which returns
+ void) and then nothing else happens. Such a sibling call would look
+ valid without the added check here. */
+#define FUNCTION_OK_FOR_SIBCALL(DECL) \
+ (DECL \
+ && ! TARGET_FLAT \
+ && (TARGET_ARCH64 || ! current_function_returns_struct))
/* Generate RTL to flush the register windows so as to make arbitrary frames
available. */
If assembler and linker properly support .uaword %r_disp32(foo),
then use PC relative 32-bit relocations instead of absolute relocs
for shared libraries. On sparc64, use pc relative 32-bit relocs even
- for binaries, to save memory. */
+ for binaries, to save memory.
+
+ binutils 2.12 would emit a R_SPARC_DISP32 dynamic relocation if the
+ symbol %r_disp32() is against was not local, but .hidden. In that
+ case, we have to use DW_EH_PE_absptr for pic personality. */
#ifdef HAVE_AS_SPARC_UA_PCREL
+#ifdef HAVE_AS_SPARC_UA_PCREL_HIDDEN
#define ASM_PREFERRED_EH_DATA_FORMAT(CODE,GLOBAL) \
(flag_pic \
? (GLOBAL ? DW_EH_PE_indirect : 0) | DW_EH_PE_pcrel | DW_EH_PE_sdata4\
: ((TARGET_ARCH64 && ! GLOBAL) \
? (DW_EH_PE_pcrel | DW_EH_PE_sdata4) \
: DW_EH_PE_absptr))
+#else
+#define ASM_PREFERRED_EH_DATA_FORMAT(CODE,GLOBAL) \
+ (flag_pic \
+ ? (GLOBAL ? DW_EH_PE_absptr : (DW_EH_PE_pcrel | DW_EH_PE_sdata4)) \
+ : ((TARGET_ARCH64 && ! GLOBAL) \
+ ? (DW_EH_PE_pcrel | DW_EH_PE_sdata4) \
+ : DW_EH_PE_absptr))
+#endif
/* Emit a PC-relative relocation. */
#define ASM_OUTPUT_DWARF_PCREL(FILE, SIZE, LABEL) \
be at least 8 bytes.
`U' handles all pseudo registers or a hard even numbered
- integer register, needed for ldd/std instructions. */
+ integer register, needed for ldd/std instructions.
-#define EXTRA_CONSTRAINT_BASE(OP, C) \
- ((C) == 'Q' ? fp_sethi_p(OP) \
- : (C) == 'R' ? fp_mov_p(OP) \
- : (C) == 'S' ? fp_high_losum_p(OP) \
- : 0)
+ 'W' handles the memory operand when moving operands in/out
+ of 'e' constraint floating point registers. */
#ifndef REG_OK_STRICT
or if it is a pseudo reg. */
#define REG_OK_FOR_BASE_P(X) REG_OK_FOR_INDEX_P (X)
-/* 'T', 'U' are for aligned memory loads which aren't needed for arch64. */
+/* 'T', 'U' are for aligned memory loads which aren't needed for arch64.
+ 'W' is like 'T' but is assumed true on arch64.
-#define EXTRA_CONSTRAINT(OP, C) \
- (EXTRA_CONSTRAINT_BASE(OP, C) \
- || ((! TARGET_ARCH64 && (C) == 'T') \
- ? (mem_min_alignment (OP, 8)) \
- : ((! TARGET_ARCH64 && (C) == 'U') \
- ? (register_ok_for_ldd (OP)) \
- : 0)))
+ Remember to accept pseudo-registers for memory constraints if reload is
+ in progress. */
+
+#define EXTRA_CONSTRAINT(OP, C) \
+ sparc_extra_constraint_check(OP, C, 0)
#else
/* Nonzero if X is a hard reg that can be used as a base reg. */
#define REG_OK_FOR_BASE_P(X) REGNO_OK_FOR_BASE_P (REGNO (X))
-#define EXTRA_CONSTRAINT(OP, C) \
- (EXTRA_CONSTRAINT_BASE(OP, C) \
- || ((! TARGET_ARCH64 && (C) == 'T') \
- ? mem_min_alignment (OP, 8) && strict_memory_address_p (Pmode, XEXP (OP, 0)) \
- : ((! TARGET_ARCH64 && (C) == 'U') \
- ? (GET_CODE (OP) == REG \
- && (REGNO (OP) < FIRST_PSEUDO_REGISTER \
- || reg_renumber[REGNO (OP)] >= 0) \
- && register_ok_for_ldd (OP)) \
- : 0)))
+#define EXTRA_CONSTRAINT(OP, C) \
+ sparc_extra_constraint_check(OP, C, 1)
#endif
\f
/* Decompose SImode constants into hi+lo_sum. We do have to \
rerecognize what we produce, so be careful. */ \
if (CONSTANT_P (X) \
- && (MODE != TFmode || TARGET_V9) \
+ && (MODE != TFmode || TARGET_ARCH64) \
&& GET_MODE (X) == SImode \
&& GET_CODE (X) != LO_SUM && GET_CODE (X) != HIGH \
&& ! (flag_pic \
&& (symbolic_operand (X, Pmode) \
- || pic_address_needs_scratch (X)))) \
+ || pic_address_needs_scratch (X))) \
+ && sparc_cmodel <= CM_MEDLOW) \
{ \
X = gen_rtx_LO_SUM (GET_MODE (X), \
gen_rtx_HIGH (GET_MODE (X), X), X); \
On the SPARC this is never true. */
#define GO_IF_MODE_DEPENDENT_ADDRESS(ADDR,LABEL)
-
-/* If we are referencing a function make the SYMBOL_REF special.
- In the Embedded Medium/Anywhere code model, %g4 points to the data segment
- so we must not add it to function addresses. */
-
-#define ENCODE_SECTION_INFO(DECL, FIRST) \
- do { \
- if (TARGET_CM_EMBMEDANY && TREE_CODE (DECL) == FUNCTION_DECL) \
- SYMBOL_REF_FLAG (XEXP (DECL_RTL (DECL), 0)) = 1; \
- } while (0)
\f
/* Specify the machine mode that this machine uses
for the index in the tablejump instruction. */
#define LTTF2_LIBCALL "_Q_flt"
#define LETF2_LIBCALL "_Q_fle"
+/* Assume by default that the _Qp_* 64-bit libcalls are implemented such
+ that the inputs are fully consumed before the output memory is clobbered. */
+
+#define TARGET_BUGGY_QP_LIB 0
+
/* We can define the TFmode sqrt optab only if TARGET_FPU. This is because
with soft-float, the SFmode and DFmode sqrt instructions will be absent,
and the compiler will notice and try to use the TFmode sqrt instruction
of the libgcc2 functions is used. */
#define FLOAT_LIB_COMPARE_RETURNS_BOOL(MODE, COMPARISON) ((MODE) == TFmode)
-/* Compute the cost of computing a constant rtl expression RTX
- whose rtx-code is CODE. The body of this macro is a portion
- of a switch statement. If the code is computed here,
- return it with a return statement. Otherwise, break from the switch. */
-
-#define CONST_COSTS(RTX,CODE,OUTER_CODE) \
- case CONST_INT: \
- if (INTVAL (RTX) < 0x1000 && INTVAL (RTX) >= -0x1000) \
- return 0; \
- case HIGH: \
- return 2; \
- case CONST: \
- case LABEL_REF: \
- case SYMBOL_REF: \
- return 4; \
- case CONST_DOUBLE: \
- if (GET_MODE (RTX) == DImode) \
- if ((XINT (RTX, 3) == 0 \
- && (unsigned) XINT (RTX, 2) < 0x1000) \
- || (XINT (RTX, 3) == -1 \
- && XINT (RTX, 2) < 0 \
- && XINT (RTX, 2) >= -0x1000)) \
- return 0; \
- return 8;
-
-#define ADDRESS_COST(RTX) 1
-
/* Compute extra cost of moving data between one register class
and another. */
#define GENERAL_OR_I64(C) ((C) == GENERAL_REGS || (C) == I64_REGS)
(((FP_REG_CLASS_P (CLASS1) && GENERAL_OR_I64 (CLASS2)) \
|| (GENERAL_OR_I64 (CLASS1) && FP_REG_CLASS_P (CLASS2)) \
|| (CLASS1) == FPCC_REGS || (CLASS2) == FPCC_REGS) \
- ? (sparc_cpu == PROCESSOR_ULTRASPARC ? 12 : 6) : 2)
+ ? ((sparc_cpu == PROCESSOR_ULTRASPARC \
+ || sparc_cpu == PROCESSOR_ULTRASPARC3) ? 12 : 6) : 2)
+
+/* Provide the cost of a branch. For pre-v9 processors we use
+ a value of 3 to take into account the potential annulling of
+ the delay slot (which ends up being a bubble in the pipeline slot)
+ plus a cycle to take into consideration the instruction cache
+ effects.
+
+ On v9 and later, which have branch prediction facilities, we set
+ it to the depth of the pipeline as that is the cost of a
+ mispredicted branch. */
+
+#define BRANCH_COST \
+ ((sparc_cpu == PROCESSOR_V9 \
+ || sparc_cpu == PROCESSOR_ULTRASPARC) \
+ ? 7 \
+ : (sparc_cpu == PROCESSOR_ULTRASPARC3 \
+ ? 9 : 3))
+
+/* The cases that RTX_COSTS handles. */
+
+#define RTX_COSTS_CASES \
+case PLUS: case MINUS: case ABS: case NEG: \
+case FLOAT: case UNSIGNED_FLOAT: \
+case FIX: case UNSIGNED_FIX: \
+case FLOAT_EXTEND: case FLOAT_TRUNCATE: \
+case SQRT: \
+case COMPARE: case IF_THEN_ELSE: \
+case MEM: \
+case MULT: case DIV: case UDIV: case MOD: case UMOD: \
+case CONST_INT: case HIGH: case CONST: \
+case LABEL_REF: case SYMBOL_REF: case CONST_DOUBLE:
/* Provide the costs of a rtl expression. This is in the body of a
- switch on CODE. The purpose for the cost of MULT is to encourage
- `synth_mult' to find a synthetic multiply when reasonable.
-
- If we need more than 12 insns to do a multiply, then go out-of-line,
- since the call overhead will be < 10% of the cost of the multiply. */
+ switch on CODE. */
#define RTX_COSTS(X,CODE,OUTER_CODE) \
- case MULT: \
- if (sparc_cpu == PROCESSOR_ULTRASPARC) \
- return (GET_MODE (X) == DImode ? \
- COSTS_N_INSNS (34) : COSTS_N_INSNS (19)); \
- return TARGET_HARD_MUL ? COSTS_N_INSNS (5) : COSTS_N_INSNS (25); \
- case DIV: \
- case UDIV: \
- case MOD: \
- case UMOD: \
- if (sparc_cpu == PROCESSOR_ULTRASPARC) \
- return (GET_MODE (X) == DImode ? \
- COSTS_N_INSNS (68) : COSTS_N_INSNS (37)); \
- return COSTS_N_INSNS (25); \
- /* Make FLOAT and FIX more expensive than CONST_DOUBLE,\
- so that cse will favor the latter. */ \
- case FLOAT: \
- case FIX: \
- return 19;
-
-/* Conditional branches with empty delay slots have a length of two. */
-#define ADJUST_INSN_LENGTH(INSN, LENGTH) \
-do { \
- if (GET_CODE (INSN) == CALL_INSN \
- || (GET_CODE (INSN) == JUMP_INSN && ! simplejump_p (insn))) \
- LENGTH += 1; \
-} while (0)
+ RTX_COSTS_CASES \
+ return sparc_rtx_costs(X,CODE,OUTER_CODE);
+
+#define ADDRESS_COST(RTX) 1
+
+#define PREFETCH_BLOCK \
+ ((sparc_cpu == PROCESSOR_ULTRASPARC \
+ || sparc_cpu == PROCESSOR_ULTRASPARC3) \
+ ? 64 : 32)
+
+#define SIMULTANEOUS_PREFETCHES \
+ ((sparc_cpu == PROCESSOR_ULTRASPARC) \
+ ? 2 \
+ : (sparc_cpu == PROCESSOR_ULTRASPARC3 \
+ ? 8 : 3))
\f
/* Control the assembler format that we output. */
/* Output code to add DELTA to the first argument, and then jump to FUNCTION.
Used for C++ multiple inheritance. */
-#define ASM_OUTPUT_MI_THUNK(FILE, THUNK_FNDECL, DELTA, FUNCTION) \
-do { \
- int reg = 0; \
- \
- if (TARGET_ARCH64 \
- && aggregate_value_p (TREE_TYPE (TREE_TYPE (FUNCTION)))) \
- reg = 1; \
- if ((DELTA) >= 4096 || (DELTA) < -4096) \
- fprintf (FILE, "\tset\t%d, %%g1\n\tadd\t%%o%d, %%g1, %%o%d\n", \
- (int)(DELTA), reg, reg); \
- else \
- fprintf (FILE, "\tadd\t%%o%d, %d, %%o%d\n", reg, (int)(DELTA), reg);\
- fprintf (FILE, "\tor\t%%o7, %%g0, %%g1\n"); \
- fprintf (FILE, "\tcall\t"); \
- assemble_name (FILE, XSTR (XEXP (DECL_RTL (FUNCTION), 0), 0)); \
- fprintf (FILE, ", 0\n"); \
- fprintf (FILE, "\t or\t%%g1, %%g0, %%o7\n"); \
-} while (0)
+#define ASM_OUTPUT_MI_THUNK(FILE, THUNK_FNDECL, DELTA, FUNCTION) \
+ sparc_output_mi_thunk (FILE, THUNK_FNDECL, DELTA, FUNCTION)
#define PRINT_OPERAND_PUNCT_VALID_P(CHAR) \
((CHAR) == '#' || (CHAR) == '*' || (CHAR) == '^' || (CHAR) == '(' || (CHAR) == '_')
#define PREDICATE_CODES \
{"reg_or_0_operand", {SUBREG, REG, CONST_INT, CONST_DOUBLE}}, \
+{"const1_operand", {CONST_INT}}, \
{"fp_zero_operand", {CONST_DOUBLE}}, \
+{"fp_register_operand", {SUBREG, REG}}, \
{"intreg_operand", {SUBREG, REG}}, \
{"fcc_reg_operand", {REG}}, \
{"fcc0_reg_operand", {REG}}, \