config.gcc (with_fpu): Allow --with-fpu=vfp3.

	gcc/
	* config.gcc (with_fpu): Allow --with-fpu=vfp3.
	* config/arm/aout.h (REGISTER_NAMES): Add D16-D31.
	* config/arm/aof.h (REGISTER_NAMES): Add D16-D31.
	* config/arm/arm.c (FL_VFPV3): New flag for VFPv3 processor
	capability.
	(all_fpus): Add FPUTYPE_VFP3.
	(fp_model_for_fpu): Add VFPv3 field.
	(arm_rtx_costs_1): Give cost to VFPv3 constants.
	(vfp3_const_double_index): New function. Return integer index of
	VFPv3 constant suitable for fconst[sd] insns, or -1 if constant
	isn't suitable.
	(vfp3_const_double_rtx): New function. True if VFPv3 is enabled
	and argument represents a valid RTX for a VFPv3 constant.
	(vfp_output_fldmd): Split fldmd with > 16 registers in the list into
	two instructions.
	(vfp_emit_fstmd): Similar, for fstmd.
	(arm_print_operand): Implement new code 'G' for VFPv3 floating-point
	constants, represented as integer indices.
	(arm_hard_regno_mode_ok): Use VFP_REGNO_OK_FOR_SINGLE,
	VFP_REGNO_OK_FOR_DOUBLE macros.
	(arm_regno_class): Handle VFPv3 d0-d7, low, high register split.
	(arm_file_start): Set float-abi attribute for VFPv3, and output
	correct ".fpu" assembler directive.
	(arm_dbx_register_numbering): Add FIXME.
	* config/arm/arm.h (TARGET_VFP3): New macro. Target supports VFPv3.
	(fputype): Add FPUTYPE_VFP3.
	(FIXED_REGISTERS): Add 32 registers for D16-D31.
	(CALL_USED_REGISTERS): Likewise.
	(CONDITIONAL_REGISTER_USAGE): Add note about conditional definition
	of LAST_VFP_REGNUM. Make D16-D31 caller-saved, if present.
	(LAST_VFP_REGNUM): Extend available VFP registers for VFPv3.
	(D7_VFP_REGNUM): New.
	(LAST_LO_VFP_REGNUM, FIRST_HI_VFP_REGNUM, LAST_HI_VFP_REGNUM)
	(VFP_REGNO_OK_FOR_SINGLE, VFP_REGNO_OK_FOR_SINGLE)
	(VFP_REGNO_OK_FOR_DOUBLE): Define new macros.
	(FIRST_PSEUDO_REGISTER): Shift up to 128 to accommodate VFPv3.
	(REG_ALLOC_ORDER): Adjust for VFPv3.
	(reg_class): Add VFP_D0_D7_REGS, VFP_LO_REGS, VFP_HI_REGS.
	(REG_CLASS_NAMES): Add entries corresponding to VFP_D0_D7_REGS,
	VFP_LO_REGS, VFP_HI_REGS.
	(REG_CLASS_CONTENTS): Likewise. Extend contents for VFP_REGS.
	(IS_VFP_CLASS): Define macro.
	(SECONDARY_OUTPUT_RELOAD_CLASS, SECONDARY_INPUT_RELOAD_CLASS): Use
	IS_VFP_CLASS.
	(REGISTER_MOVE_COST): Likewise.
	* config/arm/arm-protos.h (vfp3_const_double_rtx): Add prototype.
	* config/arm/vfp.md (VFPCC_REGNUM): Redefine as 127.
	(*arm_movsi_vfp, *thumb2_movsi_vfp, *movsfcc_vfp)
	(*thumb2_movsfcc_vfp, *abssf2_vfp, *negsf2_vfp, *addsf3_vfp)
	(*subsf3_vfp, *divsf_vfp, *mulsf_vfp, *mulsf3negsf_vfp)
	(*mulsf3addsf_vfp, *mulsf3subsf_vfp, *mulsf3negsfaddsf_vfp)
	(*extendsfdf2_vfp, *truncdfsf2_vfp, *truncsisf2_vfp)
	(*truncsidf2_vfp, fixuns_truncsfsi2, fixuns_truncdfsi2)
	(*floatsisf2_vfp, *floatsidf2_vfp, floatunssisf2)
	(floatunssidf2, *sqrtsf2_vfp, *cmpsf_split_vfp)
	(*cmpsf_trap_split_vfp, *cmpsf_vfp, *cmpsf_trap_vfp): Use 't'
	where appropriate for single-word registers.
	(*movsf_vfp, *thumb2_movsf_vfp, *movdf_vfp, *thumb2_movdf_vfp):
	As above. Fix type attributes.
	* config/arm/constraints.md (register_contraint "t"): Define.
	(register_constraint "w"): Change to D0-D15, or D0-D31 for
	VFPv3/NEON.
	(register_constraint "x"): Define.
	(constraint "Dv"): Define.

From-SVN: r126272

gcc/ChangeLog

+2007-07-03  Julian Brown  <julian@codesourcery.com>
+
+	* config.gcc (with_fpu): Allow --with-fpu=vfp3.
+	* config/arm/aout.h (REGISTER_NAMES): Add D16-D31.
+	* config/arm/aof.h (REGISTER_NAMES): Add D16-D31.
+	* config/arm/arm.c (FL_VFPV3): New flag for VFPv3 processor
+	capability.
+	(all_fpus): Add FPUTYPE_VFP3.
+	(fp_model_for_fpu): Add VFPv3 field.
+	(arm_rtx_costs_1): Give cost to VFPv3 constants.
+	(vfp3_const_double_index): New function. Return integer index of
+	VFPv3 constant suitable for fconst[sd] insns, or -1 if constant
+	isn't suitable.
+	(vfp3_const_double_rtx): New function. True if VFPv3 is enabled
+	and argument represents a valid RTX for a VFPv3 constant.
+	(vfp_output_fldmd): Split fldmd with > 16 registers in the list into
+	two instructions.
+	(vfp_emit_fstmd): Similar, for fstmd.
+	(arm_print_operand): Implement new code 'G' for VFPv3 floating-point
+	constants, represented as integer indices.
+	(arm_hard_regno_mode_ok): Use VFP_REGNO_OK_FOR_SINGLE,
+	VFP_REGNO_OK_FOR_DOUBLE macros.
+	(arm_regno_class): Handle VFPv3 d0-d7, low, high register split.
+	(arm_file_start): Set float-abi attribute for VFPv3, and output
+	correct ".fpu" assembler directive.
+	(arm_dbx_register_numbering): Add FIXME.
+	* config/arm/arm.h (TARGET_VFP3): New macro. Target supports VFPv3.
+	(fputype): Add FPUTYPE_VFP3.
+	(FIXED_REGISTERS): Add 32 registers for D16-D31.
+	(CALL_USED_REGISTERS): Likewise.
+	(CONDITIONAL_REGISTER_USAGE): Add note about conditional definition
+	of LAST_VFP_REGNUM. Make D16-D31 caller-saved, if present.
+	(LAST_VFP_REGNUM): Extend available VFP registers for VFPv3.
+	(D7_VFP_REGNUM): New.
+	(LAST_LO_VFP_REGNUM, FIRST_HI_VFP_REGNUM, LAST_HI_VFP_REGNUM)
+	(VFP_REGNO_OK_FOR_SINGLE, VFP_REGNO_OK_FOR_SINGLE)
+	(VFP_REGNO_OK_FOR_DOUBLE): Define new macros.
+	(FIRST_PSEUDO_REGISTER): Shift up to 128 to accommodate VFPv3.
+	(REG_ALLOC_ORDER): Adjust for VFPv3.
+	(reg_class): Add VFP_D0_D7_REGS, VFP_LO_REGS, VFP_HI_REGS.
+	(REG_CLASS_NAMES): Add entries corresponding to VFP_D0_D7_REGS,
+	VFP_LO_REGS, VFP_HI_REGS.
+	(REG_CLASS_CONTENTS): Likewise. Extend contents for VFP_REGS.
+	(IS_VFP_CLASS): Define macro.
+	(SECONDARY_OUTPUT_RELOAD_CLASS, SECONDARY_INPUT_RELOAD_CLASS): Use
+	IS_VFP_CLASS.
+	(REGISTER_MOVE_COST): Likewise.
+	* config/arm/arm-protos.h (vfp3_const_double_rtx): Add prototype.
+	* config/arm/vfp.md (VFPCC_REGNUM): Redefine as 127.
+	(*arm_movsi_vfp, *thumb2_movsi_vfp, *movsfcc_vfp)
+	(*thumb2_movsfcc_vfp, *abssf2_vfp, *negsf2_vfp, *addsf3_vfp)
+	(*subsf3_vfp, *divsf_vfp, *mulsf_vfp, *mulsf3negsf_vfp)
+	(*mulsf3addsf_vfp, *mulsf3subsf_vfp, *mulsf3negsfaddsf_vfp)
+	(*extendsfdf2_vfp, *truncdfsf2_vfp, *truncsisf2_vfp)
+	(*truncsidf2_vfp, fixuns_truncsfsi2, fixuns_truncdfsi2)
+	(*floatsisf2_vfp, *floatsidf2_vfp, floatunssisf2)
+	(floatunssidf2, *sqrtsf2_vfp, *cmpsf_split_vfp)
+	(*cmpsf_trap_split_vfp, *cmpsf_vfp, *cmpsf_trap_vfp): Use 't'
+	where appropriate for single-word registers.
+	(*movsf_vfp, *thumb2_movsf_vfp, *movdf_vfp, *thumb2_movdf_vfp):
+	As above. Fix type attributes.
+	* config/arm/constraints.md (register_contraint "t"): Define.
+	(register_constraint "w"): Change to D0-D15, or D0-D31 for
+	VFPv3/NEON.
+	(register_constraint "x"): Define.
+	(constraint "Dv"): Define.
+
 2007-07-03  Geoffrey Keating  <geoffk@apple.com>
 
 	* tree.h (DECL_ALIGN): Prevent use on a FUNCTION_DECL.

gcc/config.gcc

@@ -2835,7 +2835,7 @@ case "${target}" in
 
 		case "$with_fpu" in
 		"" \
-		| fpa | fpe2 | fpe3 | maverick | vfp )
+		| fpa | fpe2 | fpe3 | maverick | vfp | vfp3 )
 			# OK
 			;;
 		*)

gcc/config/arm/aof.h

@@ -187,7 +187,10 @@ do {					\
 #define CTORS_SECTION_ASM_OP "\tAREA\t|C$$gnu_ctorsvec|, DATA, READONLY"
 #define DTORS_SECTION_ASM_OP "\tAREA\t|C$$gnu_dtorsvec|, DATA, READONLY"
 
-/* Output of Assembler Instructions.  */
+/* Output of Assembler Instructions.  Note that the ?xx registers are
+   there so that VFPv3/NEON registers D16-D31 have the same spacing as D0-D15
+   (each of which is overlaid on two S registers), although there are no
+   actual single-precision registers which correspond to D16-D31.  */
 
 #define REGISTER_NAMES				\
 {						\
@@ -210,7 +213,11 @@ do {					\
   "s0",  "s1",  "s2",  "s3",  "s4",  "s5",  "s6",  "s7",  \
   "s8",  "s9",  "s10", "s11", "s12", "s13", "s14", "s15", \
   "s16", "s17", "s18", "s19", "s20", "s21", "s22", "s23", \
-  "s24", "s25", "s26", "s27", "s28", "s29", "s30", "s31",  \
+  "s24", "s25", "s26", "s27", "s28", "s29", "s30", "s31", \
+  "d16", "?16", "d17", "?17", "d18", "?18", "d19", "?19", \
+  "d20", "?20", "d21", "?21", "d22", "?22", "d23", "?23", \
+  "d24", "?24", "d25", "?25", "d26", "?26", "d27", "?27", \
+  "d28", "?28", "d29", "?29", "d30", "?30", "d31", "?31", \
   "vfpcc"					\
 }
 

gcc/config/arm/aout.h

@@ -47,7 +47,10 @@
 #define LOCAL_LABEL_PREFIX 	""
 #endif
 
-/* The assembler's names for the registers.  */
+/* The assembler's names for the registers.  Note that the ?xx registers are
+   there so that VFPv3/NEON registers D16-D31 have the same spacing as D0-D15
+   (each of which is overlaid on two S registers), although there are no
+   actual single-precision registers which correspond to D16-D31.  */
 #ifndef REGISTER_NAMES
 #define REGISTER_NAMES				   \
 {				                   \
@@ -68,6 +71,10 @@
   "s8",  "s9",  "s10", "s11", "s12", "s13", "s14", "s15", \
   "s16", "s17", "s18", "s19", "s20", "s21", "s22", "s23", \
   "s24", "s25", "s26", "s27", "s28", "s29", "s30", "s31", \
+  "d16", "?16", "d17", "?17", "d18", "?18", "d19", "?19", \
+  "d20", "?20", "d21", "?21", "d22", "?22", "d23", "?23", \
+  "d24", "?24", "d25", "?25", "d26", "?26", "d27", "?27", \
+  "d28", "?28", "d29", "?29", "d30", "?30", "d31", "?31", \
   "vfpcc"					   \
 }
 #endif

gcc/config/arm/arm-protos.h

@@ -67,6 +67,7 @@ extern rtx thumb_legitimize_reload_address (rtx *, enum machine_mode, int, int,
 					    int);
 extern int arm_const_double_rtx (rtx);
 extern int neg_const_double_rtx_ok_for_fpa (rtx);
+extern int vfp3_const_double_rtx (rtx);
 extern enum reg_class coproc_secondary_reload_class (enum machine_mode, rtx,
 						     bool);
 extern bool arm_tls_referenced_p (rtx);

gcc/config/arm/arm.c

@@ -457,6 +457,7 @@ static int thumb_call_reg_needed;
 #define FL_NOTM	      (1 << 17)	      /* Instructions not present in the 'M'
 					 profile.  */
 #define FL_DIV	      (1 << 18)	      /* Hardware divide.  */
+#define FL_VFPV3      (1 << 19)       /* Vector Floating Point V3.  */
 
 #define FL_IWMMXT     (1 << 29)	      /* XScale v2 or "Intel Wireless MMX technology".  */
 
@@ -700,7 +701,8 @@ static const struct fpu_desc all_fpus[] =
   {"fpe2",	FPUTYPE_FPA_EMU2},
   {"fpe3",	FPUTYPE_FPA_EMU2},
   {"maverick",	FPUTYPE_MAVERICK},
-  {"vfp",	FPUTYPE_VFP}
+  {"vfp",	FPUTYPE_VFP},
+  {"vfp3",	FPUTYPE_VFP3},
 };
 
 
@@ -715,7 +717,8 @@ static const enum fputype fp_model_for_fpu[] =
   ARM_FP_MODEL_FPA,		/* FPUTYPE_FPA_EMU2  */
   ARM_FP_MODEL_FPA,		/* FPUTYPE_FPA_EMU3  */
   ARM_FP_MODEL_MAVERICK,	/* FPUTYPE_MAVERICK  */
-  ARM_FP_MODEL_VFP		/* FPUTYPE_VFP  */
+  ARM_FP_MODEL_VFP,		/* FPUTYPE_VFP  */
+  ARM_FP_MODEL_VFP		/* FPUTYPE_VFP3  */
 };
 
 
@@ -4950,7 +4953,7 @@ arm_rtx_costs_1 (rtx x, enum rtx_code code, enum rtx_code outer)
       return 6;
 
     case CONST_DOUBLE:
-      if (arm_const_double_rtx (x))
+      if (arm_const_double_rtx (x) || vfp3_const_double_rtx (x))
 	return outer == SET ? 2 : -1;
       else if ((outer == COMPARE || outer == PLUS)
 	       && neg_const_double_rtx_ok_for_fpa (x))
@@ -5649,6 +5652,108 @@ neg_const_double_rtx_ok_for_fpa (rtx x)
 
   return 0;
 }
+
+
+/* VFPv3 has a fairly wide range of representable immediates, formed from
+   "quarter-precision" floating-point values. These can be evaluated using this
+   formula (with ^ for exponentiation):
+
+     -1^s * n * 2^-r
+
+   Where 's' is a sign bit (0/1), 'n' and 'r' are integers such that
+   16 <= n <= 31 and 0 <= r <= 7.
+
+   These values are mapped onto an 8-bit integer ABCDEFGH s.t.
+
+     - A (most-significant) is the sign bit.
+     - BCD are the exponent (encoded as r XOR 3).
+     - EFGH are the mantissa (encoded as n - 16).
+*/
+
+/* Return an integer index for a VFPv3 immediate operand X suitable for the
+   fconst[sd] instruction, or -1 if X isn't suitable.  */
+static int
+vfp3_const_double_index (rtx x)
+{
+  REAL_VALUE_TYPE r, m;
+  int sign, exponent;
+  unsigned HOST_WIDE_INT mantissa, mant_hi;
+  unsigned HOST_WIDE_INT mask;
+  int point_pos = 2 * HOST_BITS_PER_WIDE_INT - 1;
+
+  if (!TARGET_VFP3 || GET_CODE (x) != CONST_DOUBLE)
+    return -1;
+
+  REAL_VALUE_FROM_CONST_DOUBLE (r, x);
+
+  /* We can't represent these things, so detect them first.  */
+  if (REAL_VALUE_ISINF (r) || REAL_VALUE_ISNAN (r) || REAL_VALUE_MINUS_ZERO (r))
+    return -1;
+
+  /* Extract sign, exponent and mantissa.  */
+  sign = REAL_VALUE_NEGATIVE (r) ? 1 : 0;
+  r = REAL_VALUE_ABS (r);
+  exponent = REAL_EXP (&r);
+  /* For the mantissa, we expand into two HOST_WIDE_INTS, apart from the
+     highest (sign) bit, with a fixed binary point at bit point_pos.
+     WARNING: If there's ever a VFP version which uses more than 2 * H_W_I - 1
+     bits for the mantissa, this may fail (low bits would be lost).  */
+  real_ldexp (&m, &r, point_pos - exponent);
+  REAL_VALUE_TO_INT (&mantissa, &mant_hi, m);
+
+  /* If there are bits set in the low part of the mantissa, we can't
+     represent this value.  */
+  if (mantissa != 0)
+    return -1;
+
+  /* Now make it so that mantissa contains the most-significant bits, and move
+     the point_pos to indicate that the least-significant bits have been
+     discarded.  */
+  point_pos -= HOST_BITS_PER_WIDE_INT;
+  mantissa = mant_hi;
+
+  /* We can permit four significant bits of mantissa only, plus a high bit
+     which is always 1.  */
+  mask = ((unsigned HOST_WIDE_INT)1 << (point_pos - 5)) - 1;
+  if ((mantissa & mask) != 0)
+    return -1;
+
+  /* Now we know the mantissa is in range, chop off the unneeded bits.  */
+  mantissa >>= point_pos - 5;
+
+  /* The mantissa may be zero. Disallow that case. (It's possible to load the
+     floating-point immediate zero with Neon using an integer-zero load, but
+     that case is handled elsewhere.)  */
+  if (mantissa == 0)
+    return -1;
+
+  gcc_assert (mantissa >= 16 && mantissa <= 31);
+
+  /* The value of 5 here would be 4 if GCC used IEEE754-like encoding (where
+     normalised significands are in the range [1, 2). (Our mantissa is shifted
+     left 4 places at this point relative to normalised IEEE754 values).  GCC
+     internally uses [0.5, 1) (see real.c), so the exponent returned from
+     REAL_EXP must be altered.  */
+  exponent = 5 - exponent;
+
+  if (exponent < 0 || exponent > 7)
+    return -1;
+
+  /* Sign, mantissa and exponent are now in the correct form to plug into the
+     formulae described in the comment above.  */
+  return (sign << 7) | ((exponent ^ 3) << 4) | (mantissa - 16);
+}
+
+/* Return TRUE if rtx X is a valid immediate VFPv3 constant.  */
+int
+vfp3_const_double_rtx (rtx x)
+{
+  if (!TARGET_VFP3)
+    return 0;
+
+  return vfp3_const_double_index (x) != -1;
+}
+
 
 /* Predicates for `match_operand' and `match_operator'.  */
 
@@ -8808,6 +8913,15 @@ vfp_output_fldmd (FILE * stream, unsigned int base, int reg, int count)
       count++;
     }
 
+  /* FLDMD may not load more than 16 doubleword registers at a time. Split the
+     load into multiple parts if we have to handle more than 16 registers.  */
+  if (count > 16)
+    {
+      vfp_output_fldmd (stream, base, reg, 16);
+      vfp_output_fldmd (stream, base, reg + 16, count - 16);
+      return;
+    }
+
   fputc ('\t', stream);
   asm_fprintf (stream, "fldmfdd\t%r!, {", base);
 
@@ -8870,6 +8984,19 @@ vfp_emit_fstmd (int base_reg, int count)
       count++;
     }
 
+  /* FSTMD may not store more than 16 doubleword registers at once.  Split
+     larger stores into multiple parts (up to a maximum of two, in
+     practice).  */
+  if (count > 16)
+    {
+      int saved;
+      /* NOTE: base_reg is an internal register number, so each D register
+         counts as 2.  */
+      saved = vfp_emit_fstmd (base_reg + 32, count - 16);
+      saved += vfp_emit_fstmd (base_reg, 16);
+      return saved;
+    }
+
   par = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (count));
   dwarf = gen_rtx_SEQUENCE (VOIDmode, rtvec_alloc (count + 1));
 
@@ -11982,6 +12109,16 @@ arm_print_operand (FILE *stream, rtx x, int code)
       }
       return;
 
+    /* Print a VFPv3 floating-point constant, represented as an integer
+       index.  */
+    case 'G':
+      {
+        int index = vfp3_const_double_index (x);
+	gcc_assert (index != -1);
+	fprintf (stream, "%d", index);
+      }
+      return;
+
     default:
       if (x == 0)
 	{
@@ -12761,11 +12898,10 @@ arm_hard_regno_mode_ok (unsigned int regno, enum machine_mode mode)
       && IS_VFP_REGNUM (regno))
     {
       if (mode == SFmode || mode == SImode)
-	return TRUE;
+	return VFP_REGNO_OK_FOR_SINGLE (regno);
 
-      /* DFmode values are only valid in even register pairs.  */
       if (mode == DFmode)
-	return ((regno - FIRST_VFP_REGNUM) & 1) == 0;
+	return VFP_REGNO_OK_FOR_DOUBLE (regno);
       return FALSE;
     }
 
@@ -12828,7 +12964,14 @@ arm_regno_class (int regno)
     return CIRRUS_REGS;
 
   if (IS_VFP_REGNUM (regno))
-    return VFP_REGS;
+    {
+      if (regno <= D7_VFP_REGNUM)
+	return VFP_D0_D7_REGS;
+      else if (regno <= LAST_LO_VFP_REGNUM)
+        return VFP_LO_REGS;
+      else
+        return VFP_HI_REGS;
+    }
 
   if (IS_IWMMXT_REGNUM (regno))
     return IWMMXT_REGS;
@@ -15270,6 +15413,7 @@ arm_file_start (void)
 	}
       else
 	{
+	  int set_float_abi_attributes = 0;
 	  switch (arm_fpu_arch)
 	    {
 	    case FPUTYPE_FPA:
@@ -15285,15 +15429,23 @@ arm_file_start (void)
 	      fpu_name = "maverick";
 	      break;
 	    case FPUTYPE_VFP:
-	      if (TARGET_HARD_FLOAT)
-		asm_fprintf (asm_out_file, "\t.eabi_attribute 27, 3\n");
-	      if (TARGET_HARD_FLOAT_ABI)
-		asm_fprintf (asm_out_file, "\t.eabi_attribute 28, 1\n");
 	      fpu_name = "vfp";
+	      set_float_abi_attributes = 1;
+	      break;
+	    case FPUTYPE_VFP3:
+	      fpu_name = "vfp3";
+	      set_float_abi_attributes = 1;
 	      break;
 	    default:
 	      abort();
 	    }
+	  if (set_float_abi_attributes)
+	    {
+	      if (TARGET_HARD_FLOAT)
+		asm_fprintf (asm_out_file, "\t.eabi_attribute 27, 3\n");
+	      if (TARGET_HARD_FLOAT_ABI)
+		asm_fprintf (asm_out_file, "\t.eabi_attribute 28, 1\n");
+	    }
 	}
       asm_fprintf (asm_out_file, "\t.fpu %s\n", fpu_name);
 
@@ -16172,6 +16324,7 @@ arm_dbx_register_number (unsigned int regno)
   if (IS_FPA_REGNUM (regno))
     return (TARGET_AAPCS_BASED ? 96 : 16) + regno - FIRST_FPA_REGNUM;
 
+  /* FIXME: VFPv3 register numbering.  */
   if (IS_VFP_REGNUM (regno))
     return 64 + regno - FIRST_VFP_REGNUM;
 

gcc/config/arm/arm.h

@@ -206,6 +206,11 @@ extern GTY(()) rtx aof_pic_label;
 /* 32-bit Thumb-2 code.  */
 #define TARGET_THUMB2			(TARGET_THUMB && arm_arch_thumb2)
 
+/* FPU is VFPv3 (with twice the number of D registers).  Setting the FPU to
+   Neon automatically enables VFPv3 too.  */
+#define TARGET_VFP3 (arm_fp_model == ARM_FP_MODEL_VFP \
+		     && (arm_fpu_arch == FPUTYPE_VFP3))
+
 /* "DSP" multiply instructions, eg. SMULxy.  */
 #define TARGET_DSP_MULTIPLY \
   (TARGET_32BIT && arm_arch5e && arm_arch_notm)
@@ -275,7 +280,9 @@ enum fputype
   /* Cirrus Maverick floating point co-processor.  */
   FPUTYPE_MAVERICK,
   /* VFP.  */
-  FPUTYPE_VFP
+  FPUTYPE_VFP,
+  /* VFPv3.  */
+  FPUTYPE_VFP3
 };
 
 /* Recast the floating point class to be the floating point attribute.  */
@@ -643,6 +650,10 @@ extern int arm_structure_size_boundary;
   1,1,1,1,1,1,1,1,	\
   1,1,1,1,1,1,1,1,	\
   1,1,1,1,1,1,1,1,	\
+  1,1,1,1,1,1,1,1,	\
+  1,1,1,1,1,1,1,1,	\
+  1,1,1,1,1,1,1,1,	\
+  1,1,1,1,1,1,1,1,	\
   1			\
 }
 
@@ -669,6 +680,10 @@ extern int arm_structure_size_boundary;
   1,1,1,1,1,1,1,1,	     \
   1,1,1,1,1,1,1,1,	     \
   1,1,1,1,1,1,1,1,	     \
+  1,1,1,1,1,1,1,1,	     \
+  1,1,1,1,1,1,1,1,	     \
+  1,1,1,1,1,1,1,1,	     \
+  1,1,1,1,1,1,1,1,	     \
   1			     \
 }
 
@@ -720,11 +735,15 @@ extern int arm_structure_size_boundary;
 	}							\
       if (TARGET_VFP)						\
 	{							\
+	  /* VFPv3 registers are disabled when earlier VFP	\
+	     versions are selected due to the definition of	\
+	     LAST_VFP_REGNUM.  */				\
 	  for (regno = FIRST_VFP_REGNUM;			\
 	       regno <= LAST_VFP_REGNUM; ++ regno)		\
 	    {							\
 	      fixed_regs[regno] = 0;				\
-	      call_used_regs[regno] = regno < FIRST_VFP_REGNUM + 16; \
+	      call_used_regs[regno] = regno < FIRST_VFP_REGNUM + 16 \
+	      	|| regno >= FIRST_VFP_REGNUM + 32;		\
 	    }							\
 	}							\
     }								\
@@ -898,15 +917,35 @@ extern int arm_structure_size_boundary;
   (((REGNUM) >= FIRST_CIRRUS_FP_REGNUM) && ((REGNUM) <= LAST_CIRRUS_FP_REGNUM))
 
 #define FIRST_VFP_REGNUM	63
-#define LAST_VFP_REGNUM		94
+#define D7_VFP_REGNUM		78  /* Registers 77 and 78 == VFP reg D7.  */
+#define LAST_VFP_REGNUM	\
+  (TARGET_VFP3 ? LAST_HI_VFP_REGNUM : LAST_LO_VFP_REGNUM)
+
 #define IS_VFP_REGNUM(REGNUM) \
   (((REGNUM) >= FIRST_VFP_REGNUM) && ((REGNUM) <= LAST_VFP_REGNUM))
 
+/* VFP registers are split into two types: those defined by VFP versions < 3
+   have D registers overlaid on consecutive pairs of S registers. VFP version 3
+   defines 16 new D registers (d16-d31) which, for simplicity and correctness
+   in various parts of the backend, we implement as "fake" single-precision
+   registers (which would be S32-S63, but cannot be used in that way).  The
+   following macros define these ranges of registers.  */
+#define LAST_LO_VFP_REGNUM	94
+#define FIRST_HI_VFP_REGNUM	95
+#define LAST_HI_VFP_REGNUM	126
+
+#define VFP_REGNO_OK_FOR_SINGLE(REGNUM) \
+  ((REGNUM) <= LAST_LO_VFP_REGNUM)
+
+/* DFmode values are only valid in even register pairs.  */
+#define VFP_REGNO_OK_FOR_DOUBLE(REGNUM) \
+  ((((REGNUM) - FIRST_VFP_REGNUM) & 1) == 0)
+
 /* The number of hard registers is 16 ARM + 8 FPA + 1 CC + 1 SFP + 1 AFP.  */
 /* + 16 Cirrus registers take us up to 43.  */
 /* Intel Wireless MMX Technology registers add 16 + 4 more.  */
-/* VFP adds 32 + 1 more.  */
-#define FIRST_PSEUDO_REGISTER   96
+/* VFP (VFP3) adds 32 (64) + 1 more.  */
+#define FIRST_PSEUDO_REGISTER   128
 
 #define DBX_REGISTER_NUMBER(REGNO) arm_dbx_register_number (REGNO)
 
@@ -960,24 +999,32 @@ extern int arm_structure_size_boundary;
    function parameters.  It is quite good to use lr since other calls may
    clobber it anyway.  Allocate r0 through r3 in reverse order since r3 is
    least likely to contain a function parameter; in addition results are
-   returned in r0.  */
-
-#define REG_ALLOC_ORDER  	    \
-{                                   \
-     3,  2,  1,  0, 12, 14,  4,  5, \
-     6,  7,  8, 10,  9, 11, 13, 15, \
-    16, 17, 18, 19, 20, 21, 22, 23, \
-    27, 28, 29, 30, 31, 32, 33, 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,		    \
-    24, 25, 26,			    \
-    78, 77, 76, 75, 74, 73, 72, 71, \
-    70, 69, 68, 67, 66, 65, 64, 63, \
-    79, 80, 81, 82, 83, 84, 85, 86, \
-    87, 88, 89, 90, 91, 92, 93, 94, \
-    95				    \
+   returned in r0.
+   For VFP/VFPv3, allocate D16-D31 first, then caller-saved registers (D0-D7),
+   then D8-D15.  The reason for doing this is to attempt to reduce register
+   pressure when both single- and double-precision registers are used in a
+   function.  */
+
+#define REG_ALLOC_ORDER				\
+{						\
+     3,  2,  1,  0, 12, 14,  4,  5,		\
+     6,  7,  8, 10,  9, 11, 13, 15,		\
+    16, 17, 18, 19, 20, 21, 22, 23,		\
+    27, 28, 29, 30, 31, 32, 33, 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,				\
+    24, 25, 26,					\
+    95,  96,  97,  98,  99, 100, 101, 102,	\
+   103, 104, 105, 106, 107, 108, 109, 110,	\
+   111, 112, 113, 114, 115, 116, 117, 118,	\
+   119, 120, 121, 122, 123, 124, 125, 126,	\
+    78,  77,  76,  75,  74,  73,  72,  71,	\
+    70,  69,  68,  67,  66,  65,  64,  63,	\
+    79,  80,  81,  82,  83,  84,  85,  86,	\
+    87,  88,  89,  90,  91,  92,  93,  94,	\
+   127						\
 }
 
 /* Interrupt functions can only use registers that have already been
@@ -996,6 +1043,9 @@ enum reg_class
   NO_REGS,
   FPA_REGS,
   CIRRUS_REGS,
+  VFP_D0_D7_REGS,
+  VFP_LO_REGS,
+  VFP_HI_REGS,
   VFP_REGS,
   IWMMXT_GR_REGS,
   IWMMXT_REGS,
@@ -1018,6 +1068,9 @@ enum reg_class
   "NO_REGS",		\
   "FPA_REGS",		\
   "CIRRUS_REGS",	\
+  "VFP_D0_D7_REGS",	\
+  "VFP_LO_REGS",	\
+  "VFP_HI_REGS",	\
   "VFP_REGS",		\
   "IWMMXT_GR_REGS",	\
   "IWMMXT_REGS",	\
@@ -1034,24 +1087,32 @@ enum reg_class
 /* Define which registers fit in which classes.
    This is an initializer for a vector of HARD_REG_SET
    of length N_REG_CLASSES.  */
-#define REG_CLASS_CONTENTS					\
-{								\
-  { 0x00000000, 0x00000000, 0x00000000 }, /* NO_REGS  */	\
-  { 0x00FF0000, 0x00000000, 0x00000000 }, /* FPA_REGS */	\
-  { 0xF8000000, 0x000007FF, 0x00000000 }, /* CIRRUS_REGS */	\
-  { 0x00000000, 0x80000000, 0x7FFFFFFF }, /* VFP_REGS  */	\
-  { 0x00000000, 0x00007800, 0x00000000 }, /* IWMMXT_GR_REGS */	\
-  { 0x00000000, 0x7FFF8000, 0x00000000 }, /* IWMMXT_REGS */	\
-  { 0x000000FF, 0x00000000, 0x00000000 }, /* LO_REGS */		\
-  { 0x00002000, 0x00000000, 0x00000000 }, /* STACK_REG */	\
-  { 0x000020FF, 0x00000000, 0x00000000 }, /* BASE_REGS */	\
-  { 0x0000FF00, 0x00000000, 0x00000000 }, /* HI_REGS */		\
-  { 0x01000000, 0x00000000, 0x00000000 }, /* CC_REG */		\
-  { 0x00000000, 0x00000000, 0x80000000 }, /* VFPCC_REG */	\
-  { 0x0200FFFF, 0x00000000, 0x00000000 }, /* GENERAL_REGS */	\
-  { 0xFAFFFFFF, 0xFFFFFFFF, 0x7FFFFFFF }  /* ALL_REGS */	\
+#define REG_CLASS_CONTENTS						\
+{									\
+  { 0x00000000, 0x00000000, 0x00000000, 0x00000000 }, /* NO_REGS  */	\
+  { 0x00FF0000, 0x00000000, 0x00000000, 0x00000000 }, /* FPA_REGS */	\
+  { 0xF8000000, 0x000007FF, 0x00000000, 0x00000000 }, /* CIRRUS_REGS */	\
+  { 0x00000000, 0x80000000, 0x00007FFF, 0x00000000 }, /* VFP_D0_D7_REGS  */ \
+  { 0x00000000, 0x80000000, 0x7FFFFFFF, 0x00000000 }, /* VFP_LO_REGS  */ \
+  { 0x00000000, 0x00000000, 0x80000000, 0x7FFFFFFF }, /* VFP_HI_REGS  */ \
+  { 0x00000000, 0x80000000, 0xFFFFFFFF, 0x7FFFFFFF }, /* VFP_REGS  */	\
+  { 0x00000000, 0x00007800, 0x00000000, 0x00000000 }, /* IWMMXT_GR_REGS */ \
+  { 0x00000000, 0x7FFF8000, 0x00000000, 0x00000000 }, /* IWMMXT_REGS */	\
+  { 0x000000FF, 0x00000000, 0x00000000, 0x00000000 }, /* LO_REGS */	\
+  { 0x00002000, 0x00000000, 0x00000000, 0x00000000 }, /* STACK_REG */	\
+  { 0x000020FF, 0x00000000, 0x00000000, 0x00000000 }, /* BASE_REGS */	\
+  { 0x0000FF00, 0x00000000, 0x00000000, 0x00000000 }, /* HI_REGS */	\
+  { 0x01000000, 0x00000000, 0x00000000, 0x00000000 }, /* CC_REG */	\
+  { 0x00000000, 0x00000000, 0x00000000, 0x80000000 }, /* VFPCC_REG */	\
+  { 0x0200FFFF, 0x00000000, 0x00000000, 0x00000000 }, /* GENERAL_REGS */ \
+  { 0xFAFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x7FFFFFFF }  /* ALL_REGS */	\
 }
 
+/* Any of the VFP register classes.  */
+#define IS_VFP_CLASS(X) \
+  ((X) == VFP_D0_D7_REGS || (X) == VFP_LO_REGS \
+   || (X) == VFP_HI_REGS || (X) == VFP_REGS)
+
 /* The same information, inverted:
    Return the class number of the smallest class containing
    reg number REGNO.  This could be a conditional expression
@@ -1125,7 +1186,7 @@ enum reg_class
 #define SECONDARY_OUTPUT_RELOAD_CLASS(CLASS, MODE, X)		\
   /* Restrict which direct reloads are allowed for VFP/iWMMXt regs.  */ \
   ((TARGET_VFP && TARGET_HARD_FLOAT				\
-    && (CLASS) == VFP_REGS)					\
+    && IS_VFP_CLASS (CLASS))					\
    ? coproc_secondary_reload_class (MODE, X, FALSE)		\
    : (TARGET_IWMMXT && (CLASS) == IWMMXT_REGS)			\
    ? coproc_secondary_reload_class (MODE, X, TRUE)		\
@@ -1138,7 +1199,7 @@ enum reg_class
 #define SECONDARY_INPUT_RELOAD_CLASS(CLASS, MODE, X)		\
   /* Restrict which direct reloads are allowed for VFP/iWMMXt regs.  */ \
   ((TARGET_VFP && TARGET_HARD_FLOAT				\
-    && (CLASS) == VFP_REGS)					\
+    && IS_VFP_CLASS (CLASS))					\
     ? coproc_secondary_reload_class (MODE, X, FALSE) :		\
     (TARGET_IWMMXT && (CLASS) == IWMMXT_REGS) ?			\
     coproc_secondary_reload_class (MODE, X, TRUE) :		\
@@ -1257,8 +1318,8 @@ do {									      \
   (TARGET_32BIT ?						\
    ((FROM) == FPA_REGS && (TO) != FPA_REGS ? 20 :	\
     (FROM) != FPA_REGS && (TO) == FPA_REGS ? 20 :	\
-    (FROM) == VFP_REGS && (TO) != VFP_REGS ? 10 :  \
-    (FROM) != VFP_REGS && (TO) == VFP_REGS ? 10 :  \
+    IS_VFP_CLASS (FROM) && !IS_VFP_CLASS (TO) ? 10 :	\
+    !IS_VFP_CLASS (FROM) && IS_VFP_CLASS (TO) ? 10 :	\
     (FROM) == IWMMXT_REGS && (TO) != IWMMXT_REGS ? 4 :  \
     (FROM) != IWMMXT_REGS && (TO) == IWMMXT_REGS ? 4 :  \
     (FROM) == IWMMXT_GR_REGS || (TO) == IWMMXT_GR_REGS ? 20 :  \

gcc/config/arm/constraints.md

@@ -20,7 +20,7 @@
 ;; Boston, MA 02110-1301, USA.
 
 ;; The following register constraints have been used:
-;; - in ARM/Thumb-2 state: f, v, w, y, z
+;; - in ARM/Thumb-2 state: f, t, v, w, x, y, z
 ;; - in Thumb state: h, k, b
 ;; - in both states: l, c
 ;; In ARM state, 'l' is an alias for 'r'
@@ -30,7 +30,7 @@
 ;; in Thumb-1 state: I, J, K, L, M, N, O
 
 ;; The following multi-letter normal constraints have been used:
-;; in ARM/Thumb-2 state: Da, Db, Dc
+;; in ARM/Thumb-2 state: Da, Db, Dc, Dv
 
 ;; The following memory constraints have been used:
 ;; in ARM/Thumb-2 state: Q, Uv, Uy
@@ -40,11 +40,18 @@
 (define_register_constraint "f" "TARGET_ARM ? FPA_REGS : NO_REGS"
  "Legacy FPA registers @code{f0}-@code{f7}.")
 
+(define_register_constraint "t" "TARGET_32BIT ? VFP_LO_REGS : NO_REGS"
+ "The VFP registers @code{s0}-@code{s31}.")
+
 (define_register_constraint "v" "TARGET_ARM ? CIRRUS_REGS : NO_REGS"
  "The Cirrus Maverick co-processor registers.")
 
-(define_register_constraint "w" "TARGET_ARM ? VFP_REGS : NO_REGS"
- "The VFP registers @code{s0}-@code{s31}.")
+(define_register_constraint "w"
+  "TARGET_32BIT ? (TARGET_VFP3 ? VFP_REGS : VFP_LO_REGS) : NO_REGS"
+ "The VFP registers @code{d0}-@code{d15}, or @code{d0}-@code{d31} for VFPv3.")
+
+(define_register_constraint "x" "TARGET_32BIT ? VFP_D0_D7_REGS : NO_REGS"
+ "The VFP registers @code{d0}-@code{d7}.")
 
 (define_register_constraint "y" "TARGET_REALLY_IWMMXT ? IWMMXT_REGS : NO_REGS"
  "The Intel iWMMX co-processor registers.")
@@ -157,6 +164,13 @@
       (match_test "TARGET_32BIT && arm_const_double_inline_cost (op) == 4
 		   && !(optimize_size || arm_ld_sched)")))
 
+(define_constraint "Dv"
+ "@internal
+  In ARM/Thumb-2 state a const_double which can be used with a VFP fconsts
+  or fconstd instruction."
+ (and (match_code "const_double")
+      (match_test "TARGET_32BIT && vfp3_const_double_rtx (op)")))
+
 (define_memory_constraint "Uv"
  "@internal
   In ARM/Thumb-2 state a valid VFP load/store address."