static int rs6000_sched_reorder2 (FILE *, int, rtx *, int *, int);
static int rs6000_use_sched_lookahead (void);
static int rs6000_use_sched_lookahead_guard (rtx);
+static tree rs6000_builtin_reciprocal (unsigned int, bool, bool);
static tree rs6000_builtin_mask_for_load (void);
static tree rs6000_builtin_mul_widen_even (tree);
static tree rs6000_builtin_mul_widen_odd (tree);
static void rs6000_common_init_builtins (void);
static void rs6000_init_libfuncs (void);
+static void paired_init_builtins (void);
+static rtx paired_expand_builtin (tree, rtx, bool *);
+static rtx paired_expand_lv_builtin (enum insn_code, tree, rtx);
+static rtx paired_expand_stv_builtin (enum insn_code, tree);
+static rtx paired_expand_predicate_builtin (enum insn_code, tree, rtx);
+
static void enable_mask_for_builtins (struct builtin_description *, int,
enum rs6000_builtins,
enum rs6000_builtins);
#endif
static tree rs6000_build_builtin_va_list (void);
+static void rs6000_va_start (tree, rtx);
static tree rs6000_gimplify_va_arg (tree, tree, tree *, tree *);
static bool rs6000_must_pass_in_stack (enum machine_mode, const_tree);
static bool rs6000_scalar_mode_supported_p (enum machine_mode);
#undef TARGET_BUILD_BUILTIN_VA_LIST
#define TARGET_BUILD_BUILTIN_VA_LIST rs6000_build_builtin_va_list
+#undef TARGET_EXPAND_BUILTIN_VA_START
+#define TARGET_EXPAND_BUILTIN_VA_START rs6000_va_start
+
#undef TARGET_GIMPLIFY_VA_ARG_EXPR
#define TARGET_GIMPLIFY_VA_ARG_EXPR rs6000_gimplify_va_arg
#undef TARGET_USE_BLOCKS_FOR_CONSTANT_P
#define TARGET_USE_BLOCKS_FOR_CONSTANT_P rs6000_use_blocks_for_constant_p
+#undef TARGET_BUILTIN_RECIPROCAL
+#define TARGET_BUILTIN_RECIPROCAL rs6000_builtin_reciprocal
+
struct gcc_target targetm = TARGET_INITIALIZER;
\f
This excludes the 32-bit decimal float mode for now. */
if (FP_REGNO_P (regno))
return
- (SCALAR_FLOAT_MODE_P (mode)
+ ((SCALAR_FLOAT_MODE_P (mode)
&& (mode != TDmode || (regno % 2) == 0)
&& mode != SDmode
&& FP_REGNO_P (regno + HARD_REGNO_NREGS (regno, mode) - 1))
|| (GET_MODE_CLASS (mode) == MODE_INT
- && GET_MODE_SIZE (mode) == UNITS_PER_FP_WORD);
+ && GET_MODE_SIZE (mode) == UNITS_PER_FP_WORD)
+ || (PAIRED_SIMD_REGNO_P (regno) && TARGET_PAIRED_FLOAT
+ && PAIRED_VECTOR_MODE (mode)));
/* The CR register can only hold CC modes. */
if (CR_REGNO_P (regno))
rs6000_ieeequad = 1;
#endif
- /* Set Altivec ABI as default for powerpc64 linux. */
+ /* Enable Altivec ABI for AIX -maltivec. */
+ if (TARGET_XCOFF && TARGET_ALTIVEC)
+ rs6000_altivec_abi = 1;
+
+ /* Set Altivec ABI as default for PowerPC64 Linux. */
if (TARGET_ELF && TARGET_64BIT)
{
rs6000_altivec_abi = 1;
set_param_value ("l1-cache-line-size", rs6000_cost->cache_line_size);
if (!PARAM_SET_P (PARAM_L2_CACHE_SIZE))
set_param_value ("l2-cache-size", rs6000_cost->l2_cache_size);
+
+ /* If using typedef char *va_list, signal that __builtin_va_start (&ap, 0)
+ can be optimized to ap = __builtin_next_arg (0). */
+ if (DEFAULT_ABI != ABI_V4)
+ targetm.expand_builtin_va_start = NULL;
}
/* Implement targetm.vectorize.builtin_mask_for_load. */
return "li %0,%1\n\tevmergelo %0,%0,%0\n\tli %0,%2";
}
+/* Initialize TARGET of vector PAIRED to VALS. */
+
+void
+paired_expand_vector_init (rtx target, rtx vals)
+{
+ enum machine_mode mode = GET_MODE (target);
+ int n_elts = GET_MODE_NUNITS (mode);
+ int n_var = 0;
+ rtx x, new, tmp, constant_op, op1, op2;
+ int i;
+
+ for (i = 0; i < n_elts; ++i)
+ {
+ x = XVECEXP (vals, 0, i);
+ if (!CONSTANT_P (x))
+ ++n_var;
+ }
+ if (n_var == 0)
+ {
+ /* Load from constant pool. */
+ emit_move_insn (target, gen_rtx_CONST_VECTOR (mode, XVEC (vals, 0)));
+ return;
+ }
+
+ if (n_var == 2)
+ {
+ /* The vector is initialized only with non-constants. */
+ new = gen_rtx_VEC_CONCAT (V2SFmode, XVECEXP (vals, 0, 0),
+ XVECEXP (vals, 0, 1));
+
+ emit_move_insn (target, new);
+ return;
+ }
+
+ /* One field is non-constant and the other one is a constant. Load the
+ constant from the constant pool and use ps_merge instruction to
+ construct the whole vector. */
+ op1 = XVECEXP (vals, 0, 0);
+ op2 = XVECEXP (vals, 0, 1);
+
+ constant_op = (CONSTANT_P (op1)) ? op1 : op2;
+
+ tmp = gen_reg_rtx (GET_MODE (constant_op));
+ emit_move_insn (tmp, constant_op);
+
+ if (CONSTANT_P (op1))
+ new = gen_rtx_VEC_CONCAT (V2SFmode, tmp, op2);
+ else
+ new = gen_rtx_VEC_CONCAT (V2SFmode, op1, tmp);
+
+ emit_move_insn (target, new);
+}
+
+void
+paired_expand_vector_move (rtx operands[])
+{
+ rtx op0 = operands[0], op1 = operands[1];
+
+ emit_move_insn (op0, op1);
+}
+
+/* Emit vector compare for code RCODE. DEST is destination, OP1 and
+ OP2 are two VEC_COND_EXPR operands, CC_OP0 and CC_OP1 are the two
+ operands for the relation operation COND. This is a recursive
+ function. */
+
+static void
+paired_emit_vector_compare (enum rtx_code rcode,
+ rtx dest, rtx op0, rtx op1,
+ rtx cc_op0, rtx cc_op1)
+{
+ rtx tmp = gen_reg_rtx (V2SFmode);
+ rtx tmp1, max, min, equal_zero;
+
+ gcc_assert (TARGET_PAIRED_FLOAT);
+ gcc_assert (GET_MODE (op0) == GET_MODE (op1));
+
+ switch (rcode)
+ {
+ case LT:
+ case LTU:
+ paired_emit_vector_compare (GE, dest, op1, op0, cc_op0, cc_op1);
+ return;
+ case GE:
+ case GEU:
+ emit_insn (gen_subv2sf3 (tmp, cc_op0, cc_op1));
+ emit_insn (gen_selv2sf4 (dest, tmp, op0, op1, CONST0_RTX (SFmode)));
+ return;
+ case LE:
+ case LEU:
+ paired_emit_vector_compare (GE, dest, op0, op1, cc_op1, cc_op0);
+ return;
+ case GT:
+ paired_emit_vector_compare (LE, dest, op1, op0, cc_op0, cc_op1);
+ return;
+ case EQ:
+ tmp1 = gen_reg_rtx (V2SFmode);
+ max = gen_reg_rtx (V2SFmode);
+ min = gen_reg_rtx (V2SFmode);
+ equal_zero = gen_reg_rtx (V2SFmode);
+
+ emit_insn (gen_subv2sf3 (tmp, cc_op0, cc_op1));
+ emit_insn (gen_selv2sf4
+ (max, tmp, cc_op0, cc_op1, CONST0_RTX (SFmode)));
+ emit_insn (gen_subv2sf3 (tmp, cc_op1, cc_op0));
+ emit_insn (gen_selv2sf4
+ (min, tmp, cc_op0, cc_op1, CONST0_RTX (SFmode)));
+ emit_insn (gen_subv2sf3 (tmp1, min, max));
+ emit_insn (gen_selv2sf4 (dest, tmp1, op0, op1, CONST0_RTX (SFmode)));
+ return;
+ case NE:
+ paired_emit_vector_compare (EQ, dest, op1, op0, cc_op0, cc_op1);
+ return;
+ case UNLE:
+ paired_emit_vector_compare (LE, dest, op1, op0, cc_op0, cc_op1);
+ return;
+ case UNLT:
+ paired_emit_vector_compare (LT, dest, op1, op0, cc_op0, cc_op1);
+ return;
+ case UNGE:
+ paired_emit_vector_compare (GE, dest, op1, op0, cc_op0, cc_op1);
+ return;
+ case UNGT:
+ paired_emit_vector_compare (GT, dest, op1, op0, cc_op0, cc_op1);
+ return;
+ default:
+ gcc_unreachable ();
+ }
+
+ return;
+}
+
+/* Emit vector conditional expression.
+ DEST is destination. OP1 and OP2 are two VEC_COND_EXPR operands.
+ CC_OP0 and CC_OP1 are the two operands for the relation operation COND. */
+
+int
+paired_emit_vector_cond_expr (rtx dest, rtx op1, rtx op2,
+ rtx cond, rtx cc_op0, rtx cc_op1)
+{
+ enum rtx_code rcode = GET_CODE (cond);
+
+ if (!TARGET_PAIRED_FLOAT)
+ return 0;
+
+ paired_emit_vector_compare (rcode, dest, op1, op2, cc_op0, cc_op1);
+
+ return 1;
+}
+
/* Initialize vector TARGET to VALS. */
void
&& (mode == SImode || mode == DImode || mode == TImode)
&& REG_P (SUBREG_REG (op))
&& (GET_MODE (SUBREG_REG (op)) == DFmode
- || GET_MODE (SUBREG_REG (op)) == TFmode))
+ || GET_MODE (SUBREG_REG (op)) == TFmode
+ || GET_MODE (SUBREG_REG (op)) == DDmode
+ || GET_MODE (SUBREG_REG (op)) == TDmode))
return true;
/* Reject (subreg:DF (reg:DI)); likewise with subreg:TF and
reg:TI. */
if (GET_CODE (op) == SUBREG
- && (mode == DFmode || mode == TFmode)
+ && (mode == DFmode || mode == TFmode
+ || mode == DDmode || mode == TDmode)
&& REG_P (SUBREG_REG (op))
&& (GET_MODE (SUBREG_REG (op)) == DImode
|| GET_MODE (SUBREG_REG (op)) == TImode))
case V4SFmode:
case V4SImode:
/* AltiVec vector modes. Only reg+reg addressing is valid and
- constant offset zero should not occur due to canonicalization.
- Allow any offset when not strict before reload. */
- return !strict;
+ constant offset zero should not occur due to canonicalization. */
+ return false;
case V4HImode:
case V2SImode:
case V1DImode:
case V2SFmode:
+ /* Paired vector modes. Only reg+reg addressing is valid and
+ constant offset zero should not occur due to canonicalization. */
+ if (TARGET_PAIRED_FLOAT)
+ return false;
/* SPE vector modes. */
return SPE_CONST_OFFSET_OK (offset);
break;
case TFmode:
+ case TDmode:
if (TARGET_E500_DOUBLE)
return (SPE_CONST_OFFSET_OK (offset)
&& SPE_CONST_OFFSET_OK (offset + 8));
case TImode:
- case TDmode:
if (mode == TFmode || mode == TDmode || !TARGET_POWERPC64)
extra = 12;
else if (offset & 3)
return false;
/* Restrict addressing for DI because of our SUBREG hackery. */
if (TARGET_E500_DOUBLE && (mode == DFmode || mode == TFmode
+ || mode == DDmode || mode == TDmode
|| mode == DImode))
return false;
x = XEXP (x, 1);
return false;
if (GET_MODE_BITSIZE (mode) > 64
|| (GET_MODE_BITSIZE (mode) > 32 && !TARGET_POWERPC64
- && !(TARGET_HARD_FLOAT && TARGET_FPRS && mode == DFmode)))
+ && !(TARGET_HARD_FLOAT && TARGET_FPRS
+ && (mode == DFmode || mode == DDmode))))
return false;
return CONSTANT_P (x);
if (GET_CODE (x) == PLUS
&& GET_CODE (XEXP (x, 0)) == REG
&& GET_CODE (XEXP (x, 1)) == CONST_INT
- && (unsigned HOST_WIDE_INT) (INTVAL (XEXP (x, 1)) + 0x8000) >= 0x10000)
+ && (unsigned HOST_WIDE_INT) (INTVAL (XEXP (x, 1)) + 0x8000) >= 0x10000
+ && !(SPE_VECTOR_MODE (mode)
+ || (TARGET_E500_DOUBLE && (mode == DFmode || mode == TFmode
+ || mode == DImode))))
{
HOST_WIDE_INT high_int, low_int;
rtx sum;
&& CONSTANT_P (x)
&& GET_MODE_NUNITS (mode) == 1
&& (GET_MODE_BITSIZE (mode) <= 32
- || ((TARGET_HARD_FLOAT && TARGET_FPRS) && mode == DFmode)))
+ || ((TARGET_HARD_FLOAT && TARGET_FPRS)
+ && (mode == DFmode || mode == DDmode))))
{
rtx reg = gen_reg_rtx (Pmode);
emit_insn (gen_elf_high (reg, x));
&& GET_CODE (x) != CONST_INT
&& GET_CODE (x) != CONST_DOUBLE
&& CONSTANT_P (x)
- && ((TARGET_HARD_FLOAT && TARGET_FPRS) || mode != DFmode)
+ && ((TARGET_HARD_FLOAT && TARGET_FPRS)
+ || (mode != DFmode && mode != DDmode))
&& mode != DImode
&& mode != TImode)
{
&& GET_CODE (XEXP (x, 1)) == CONST_INT
&& !SPE_VECTOR_MODE (mode)
&& !(TARGET_E500_DOUBLE && (mode == DFmode || mode == TFmode
+ || mode == DDmode || mode == TDmode
|| mode == DImode))
&& !ALTIVEC_VECTOR_MODE (mode))
{
&& !flag_pic
#endif
/* Don't do this for TFmode or TDmode, since the result isn't offsettable.
- The same goes for DImode without 64-bit gprs and DFmode
+ The same goes for DImode without 64-bit gprs and DFmode and DDmode
without fprs. */
&& mode != TFmode
&& mode != TDmode
&& (mode != DImode || TARGET_POWERPC64)
- && (mode != DFmode || TARGET_POWERPC64
+ && ((mode != DFmode && mode != DDmode) || TARGET_POWERPC64
|| (TARGET_FPRS && TARGET_HARD_FLOAT)))
{
#if TARGET_MACHO
refers to a constant pool entry of an address (or the sum of it
plus a constant), a short (16-bit signed) constant plus a register,
the sum of two registers, or a register indirect, possibly with an
- auto-increment. For DFmode and DImode with a constant plus register,
- we must ensure that both words are addressable or PowerPC64 with offset
- word aligned.
+ auto-increment. For DFmode, DDmode and DImode with a constant plus
+ register, we must ensure that both words are addressable or PowerPC64
+ with offset word aligned.
- For modes spanning multiple registers (DFmode in 32-bit GPRs,
+ For modes spanning multiple registers (DFmode and DDmode in 32-bit GPRs,
32-bit DImode, TImode, TFmode, TDmode), indexed addressing cannot be used
because adjacent memory cells are accessed by adding word-sized offsets
during assembly output. */
&& mode != TFmode
&& mode != TDmode
/* Restrict addressing for DI because of our SUBREG hackery. */
- && !(TARGET_E500_DOUBLE && (mode == DFmode || mode == TFmode
- || mode == DImode))
+ && !(TARGET_E500_DOUBLE
+ && (mode == DFmode || mode == DDmode || mode == DImode))
&& TARGET_UPDATE
&& legitimate_indirect_address_p (XEXP (x, 0), reg_ok_strict))
return 1;
&& mode != TDmode
&& ((TARGET_HARD_FLOAT && TARGET_FPRS)
|| TARGET_POWERPC64
- || ((mode != DFmode || TARGET_E500_DOUBLE) && mode != TFmode))
+ || ((mode != DFmode && mode != DDmode) || TARGET_E500_DOUBLE))
&& (TARGET_POWERPC64 || mode != DImode)
&& legitimate_indexed_address_p (x, reg_ok_strict))
return 1;
&& mode != TDmode
&& ((TARGET_HARD_FLOAT && TARGET_FPRS)
|| TARGET_POWERPC64
- || ((mode != DFmode || TARGET_E500_DOUBLE) && mode != TFmode))
+ || ((mode != DFmode && mode != DDmode) || TARGET_E500_DOUBLE))
&& (TARGET_POWERPC64 || mode != DImode)
&& !ALTIVEC_VECTOR_MODE (mode)
&& !SPE_VECTOR_MODE (mode)
/* Restrict addressing for DI because of our SUBREG hackery. */
- && !(TARGET_E500_DOUBLE && (mode == DFmode || mode == DImode))
+ && !(TARGET_E500_DOUBLE
+ && (mode == DFmode || mode == DDmode || mode == DImode))
&& TARGET_UPDATE
&& legitimate_indirect_address_p (XEXP (x, 0), reg_ok_strict)
&& (rs6000_legitimate_offset_address_p (mode, XEXP (x, 1), reg_ok_strict)
fixed_regs[RS6000_PIC_OFFSET_TABLE_REGNUM]
= call_used_regs[RS6000_PIC_OFFSET_TABLE_REGNUM] = 1;
- if (TARGET_ALTIVEC)
- global_regs[VSCR_REGNO] = 1;
-
if (TARGET_SPE)
{
global_regs[SPEFSCR_REGNO] = 1;
= call_really_used_regs[14] = 1;
}
- if (! TARGET_ALTIVEC)
+ if (!TARGET_ALTIVEC)
{
for (i = FIRST_ALTIVEC_REGNO; i <= LAST_ALTIVEC_REGNO; ++i)
fixed_regs[i] = call_used_regs[i] = call_really_used_regs[i] = 1;
call_really_used_regs[VRSAVE_REGNO] = 1;
}
+ if (TARGET_ALTIVEC)
+ global_regs[VSCR_REGNO] = 1;
+
if (TARGET_ALTIVEC_ABI)
- for (i = FIRST_ALTIVEC_REGNO; i < FIRST_ALTIVEC_REGNO + 20; ++i)
- call_used_regs[i] = call_really_used_regs[i] = 1;
+ {
+ for (i = FIRST_ALTIVEC_REGNO; i < FIRST_ALTIVEC_REGNO + 20; ++i)
+ call_used_regs[i] = call_really_used_regs[i] = 1;
+
+ /* AIX reserves VR20:31 in non-extended ABI mode. */
+ if (TARGET_XCOFF)
+ for (i = FIRST_ALTIVEC_REGNO + 20; i < FIRST_ALTIVEC_REGNO + 32; ++i)
+ fixed_regs[i] = call_used_regs[i] = call_really_used_regs[i] = 1;
+ }
}
\f
/* Try to output insns to set TARGET equal to the constant C if it can
else
{
cum->fregno = FP_ARG_V4_MAX_REG + 1;
- if (mode == DFmode || mode == TFmode || mode == DDmode || mode == TDmode)
+ if (mode == DFmode || mode == TFmode
+ || mode == DDmode || mode == TDmode)
cum->words += cum->words & 1;
cum->words += rs6000_arg_size (mode, type);
}
switch (mode)
{
case DFmode:
+ case DDmode:
r1 = gen_rtx_REG (DImode, gregno);
r1 = gen_rtx_EXPR_LIST (VOIDmode, r1, const0_rtx);
return gen_rtx_PARALLEL (mode, gen_rtvec (1, r1));
case DCmode:
case TFmode:
+ case TDmode:
r1 = gen_rtx_REG (DImode, gregno);
r1 = gen_rtx_EXPR_LIST (VOIDmode, r1, const0_rtx);
r3 = gen_rtx_REG (DImode, gregno + 2);
/* On E500 v2, double arithmetic is done on the full 64-bit GPR, but
are passed and returned in a pair of GPRs for ABI compatibility. */
- if (TARGET_E500_DOUBLE && (mode == DFmode || mode == DCmode
- || mode == TFmode || mode == TCmode))
+ if (TARGET_E500_DOUBLE && (mode == DFmode || mode == TFmode
+ || mode == DDmode || mode == TDmode
+ || mode == DCmode || mode == TCmode))
{
int n_words = rs6000_arg_size (mode, type);
/* Doubles go in an odd/even register pair (r5/r6, etc). */
- if (mode == DFmode)
+ if (mode == DFmode || mode == DDmode)
gregno += (1 - gregno) & 1;
/* Multi-reg args are not split between registers and stack. */
/* Implement va_start. */
-void
+static void
rs6000_va_start (tree valist, rtx nextarg)
{
HOST_WIDE_INT words, n_gpr, n_fpr;
static void
def_builtin (int mask, const char *name, tree type, int code)
{
- if (mask & target_flags)
+ if ((mask & target_flags) || TARGET_PAIRED_FLOAT)
{
if (rs6000_builtin_decls[code])
abort ();
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_nmsub", ALTIVEC_BUILTIN_VEC_NMSUB },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_perm", ALTIVEC_BUILTIN_VEC_PERM },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_sel", ALTIVEC_BUILTIN_VEC_SEL },
+
+ { 0, CODE_FOR_paired_msub, "__builtin_paired_msub", PAIRED_BUILTIN_MSUB },
+ { 0, CODE_FOR_paired_madd, "__builtin_paired_madd", PAIRED_BUILTIN_MADD },
+ { 0, CODE_FOR_paired_madds0, "__builtin_paired_madds0", PAIRED_BUILTIN_MADDS0 },
+ { 0, CODE_FOR_paired_madds1, "__builtin_paired_madds1", PAIRED_BUILTIN_MADDS1 },
+ { 0, CODE_FOR_paired_nmsub, "__builtin_paired_nmsub", PAIRED_BUILTIN_NMSUB },
+ { 0, CODE_FOR_paired_nmadd, "__builtin_paired_nmadd", PAIRED_BUILTIN_NMADD },
+ { 0, CODE_FOR_paired_sum0, "__builtin_paired_sum0", PAIRED_BUILTIN_SUM0 },
+ { 0, CODE_FOR_paired_sum1, "__builtin_paired_sum1", PAIRED_BUILTIN_SUM1 },
+ { 0, CODE_FOR_selv2sf4, "__builtin_paired_selv2sf4", PAIRED_BUILTIN_SELV2SF4 },
};
/* DST operations: void foo (void *, const int, const char). */
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_sums", ALTIVEC_BUILTIN_VEC_SUMS },
{ MASK_ALTIVEC, CODE_FOR_nothing, "__builtin_vec_xor", ALTIVEC_BUILTIN_VEC_XOR },
+ { 0, CODE_FOR_divv2sf3, "__builtin_paired_divv2sf3", PAIRED_BUILTIN_DIVV2SF3 },
+ { 0, CODE_FOR_addv2sf3, "__builtin_paired_addv2sf3", PAIRED_BUILTIN_ADDV2SF3 },
+ { 0, CODE_FOR_subv2sf3, "__builtin_paired_subv2sf3", PAIRED_BUILTIN_SUBV2SF3 },
+ { 0, CODE_FOR_mulv2sf3, "__builtin_paired_mulv2sf3", PAIRED_BUILTIN_MULV2SF3 },
+ { 0, CODE_FOR_paired_muls0, "__builtin_paired_muls0", PAIRED_BUILTIN_MULS0 },
+ { 0, CODE_FOR_paired_muls1, "__builtin_paired_muls1", PAIRED_BUILTIN_MULS1 },
+ { 0, CODE_FOR_paired_merge00, "__builtin_paired_merge00", PAIRED_BUILTIN_MERGE00 },
+ { 0, CODE_FOR_paired_merge01, "__builtin_paired_merge01", PAIRED_BUILTIN_MERGE01 },
+ { 0, CODE_FOR_paired_merge10, "__builtin_paired_merge10", PAIRED_BUILTIN_MERGE10 },
+ { 0, CODE_FOR_paired_merge11, "__builtin_paired_merge11", PAIRED_BUILTIN_MERGE11 },
+
/* Place holder, leave as first spe builtin. */
{ 0, CODE_FOR_spe_evaddw, "__builtin_spe_evaddw", SPE_BUILTIN_EVADDW },
{ 0, CODE_FOR_spe_evand, "__builtin_spe_evand", SPE_BUILTIN_EVAND },
{ 0, CODE_FOR_spe_evfststeq, "__builtin_spe_evsel_fststeq", SPE_BUILTIN_EVSEL_FSTSTEQ },
};
+/* PAIRED predicates. */
+static const struct builtin_description bdesc_paired_preds[] =
+{
+ /* Place-holder. Leave as first. */
+ { 0, CODE_FOR_paired_cmpu0, "__builtin_paired_cmpu0", PAIRED_BUILTIN_CMPU0 },
+ /* Place-holder. Leave as last. */
+ { 0, CODE_FOR_paired_cmpu1, "__builtin_paired_cmpu1", PAIRED_BUILTIN_CMPU1 },
+};
+
/* ABS* operations. */
static const struct builtin_description bdesc_abs[] =
{ 0, CODE_FOR_spe_evsubfumiaaw, "__builtin_spe_evsubfumiaaw", SPE_BUILTIN_EVSUBFUMIAAW },
/* Place-holder. Leave as last unary SPE builtin. */
- { 0, CODE_FOR_spe_evsubfusiaaw, "__builtin_spe_evsubfusiaaw", SPE_BUILTIN_EVSUBFUSIAAW }
+ { 0, CODE_FOR_spe_evsubfusiaaw, "__builtin_spe_evsubfusiaaw", SPE_BUILTIN_EVSUBFUSIAAW },
+
+ { 0, CODE_FOR_absv2sf2, "__builtin_paired_absv2sf2", PAIRED_BUILTIN_ABSV2SF2 },
+ { 0, CODE_FOR_nabsv2sf2, "__builtin_paired_nabsv2sf2", PAIRED_BUILTIN_NABSV2SF2 },
+ { 0, CODE_FOR_negv2sf2, "__builtin_paired_negv2sf2", PAIRED_BUILTIN_NEGV2SF2 },
+ { 0, CODE_FOR_sqrtv2sf2, "__builtin_paired_sqrtv2sf2", PAIRED_BUILTIN_SQRTV2SF2 },
+ { 0, CODE_FOR_resv2sf2, "__builtin_paired_resv2sf2", PAIRED_BUILTIN_RESV2SF2 }
};
static rtx
}
static rtx
+paired_expand_lv_builtin (enum insn_code icode, tree exp, rtx target)
+{
+ rtx pat, addr;
+ tree arg0 = CALL_EXPR_ARG (exp, 0);
+ tree arg1 = CALL_EXPR_ARG (exp, 1);
+ enum machine_mode tmode = insn_data[icode].operand[0].mode;
+ enum machine_mode mode0 = Pmode;
+ enum machine_mode mode1 = Pmode;
+ rtx op0 = expand_normal (arg0);
+ rtx op1 = expand_normal (arg1);
+
+ if (icode == CODE_FOR_nothing)
+ /* Builtin not supported on this processor. */
+ return 0;
+
+ /* If we got invalid arguments bail out before generating bad rtl. */
+ if (arg0 == error_mark_node || arg1 == error_mark_node)
+ return const0_rtx;
+
+ if (target == 0
+ || GET_MODE (target) != tmode
+ || ! (*insn_data[icode].operand[0].predicate) (target, tmode))
+ target = gen_reg_rtx (tmode);
+
+ op1 = copy_to_mode_reg (mode1, op1);
+
+ if (op0 == const0_rtx)
+ {
+ addr = gen_rtx_MEM (tmode, op1);
+ }
+ else
+ {
+ op0 = copy_to_mode_reg (mode0, op0);
+ addr = gen_rtx_MEM (tmode, gen_rtx_PLUS (Pmode, op0, op1));
+ }
+
+ pat = GEN_FCN (icode) (target, addr);
+
+ if (! pat)
+ return 0;
+ emit_insn (pat);
+
+ return target;
+}
+
+static rtx
altivec_expand_lv_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat, addr;
}
static rtx
+paired_expand_stv_builtin (enum insn_code icode, tree exp)
+{
+ tree arg0 = CALL_EXPR_ARG (exp, 0);
+ tree arg1 = CALL_EXPR_ARG (exp, 1);
+ tree arg2 = CALL_EXPR_ARG (exp, 2);
+ rtx op0 = expand_normal (arg0);
+ rtx op1 = expand_normal (arg1);
+ rtx op2 = expand_normal (arg2);
+ rtx pat, addr;
+ enum machine_mode tmode = insn_data[icode].operand[0].mode;
+ enum machine_mode mode1 = Pmode;
+ enum machine_mode mode2 = Pmode;
+
+ /* Invalid arguments. Bail before doing anything stoopid! */
+ if (arg0 == error_mark_node
+ || arg1 == error_mark_node
+ || arg2 == error_mark_node)
+ return const0_rtx;
+
+ if (! (*insn_data[icode].operand[1].predicate) (op0, tmode))
+ op0 = copy_to_mode_reg (tmode, op0);
+
+ op2 = copy_to_mode_reg (mode2, op2);
+
+ if (op1 == const0_rtx)
+ {
+ addr = gen_rtx_MEM (tmode, op2);
+ }
+ else
+ {
+ op1 = copy_to_mode_reg (mode1, op1);
+ addr = gen_rtx_MEM (tmode, gen_rtx_PLUS (Pmode, op1, op2));
+ }
+
+ pat = GEN_FCN (icode) (addr, op0);
+ if (pat)
+ emit_insn (pat);
+ return NULL_RTX;
+}
+
+static rtx
altivec_expand_stv_builtin (enum insn_code icode, tree exp)
{
tree arg0 = CALL_EXPR_ARG (exp, 0);
if (! (*insn_data[icode].operand[3].predicate) (op2, mode2))
op2 = copy_to_mode_reg (mode2, op2);
- pat = GEN_FCN (icode) (target, op0, op1, op2);
+ if (TARGET_PAIRED_FLOAT && icode == CODE_FOR_selv2sf4)
+ pat = GEN_FCN (icode) (target, op0, op1, op2, CONST0_RTX (SFmode));
+ else
+ pat = GEN_FCN (icode) (target, op0, op1, op2);
if (! pat)
return 0;
emit_insn (pat);
return NULL_RTX;
}
+/* Expand the builtin in EXP and store the result in TARGET. Store
+ true in *EXPANDEDP if we found a builtin to expand. */
+static rtx
+paired_expand_builtin (tree exp, rtx target, bool * expandedp)
+{
+ tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
+ unsigned int fcode = DECL_FUNCTION_CODE (fndecl);
+ const struct builtin_description *d;
+ size_t i;
+
+ *expandedp = true;
+
+ switch (fcode)
+ {
+ case PAIRED_BUILTIN_STX:
+ return paired_expand_stv_builtin (CODE_FOR_paired_stx, exp);
+ case PAIRED_BUILTIN_LX:
+ return paired_expand_lv_builtin (CODE_FOR_paired_lx, exp, target);
+ default:
+ break;
+ /* Fall through. */
+ }
+
+ /* Expand the paired predicates. */
+ d = bdesc_paired_preds;
+ for (i = 0; i < ARRAY_SIZE (bdesc_paired_preds); i++, d++)
+ if (d->code == fcode)
+ return paired_expand_predicate_builtin (d->icode, exp, target);
+
+ *expandedp = false;
+ return NULL_RTX;
+}
+
/* Binops that need to be initialized manually, but can be expanded
automagically by rs6000_expand_binop_builtin. */
static struct builtin_description bdesc_2arg_spe[] =
}
static rtx
+paired_expand_predicate_builtin (enum insn_code icode, tree exp, rtx target)
+{
+ rtx pat, scratch, tmp;
+ tree form = CALL_EXPR_ARG (exp, 0);
+ tree arg0 = CALL_EXPR_ARG (exp, 1);
+ tree arg1 = CALL_EXPR_ARG (exp, 2);
+ rtx op0 = expand_normal (arg0);
+ rtx op1 = expand_normal (arg1);
+ enum machine_mode mode0 = insn_data[icode].operand[1].mode;
+ enum machine_mode mode1 = insn_data[icode].operand[2].mode;
+ int form_int;
+ enum rtx_code code;
+
+ if (TREE_CODE (form) != INTEGER_CST)
+ {
+ error ("argument 1 of __builtin_paired_predicate must be a constant");
+ return const0_rtx;
+ }
+ else
+ form_int = TREE_INT_CST_LOW (form);
+
+ gcc_assert (mode0 == mode1);
+
+ if (arg0 == error_mark_node || arg1 == error_mark_node)
+ return const0_rtx;
+
+ if (target == 0
+ || GET_MODE (target) != SImode
+ || !(*insn_data[icode].operand[0].predicate) (target, SImode))
+ target = gen_reg_rtx (SImode);
+ if (!(*insn_data[icode].operand[1].predicate) (op0, mode0))
+ op0 = copy_to_mode_reg (mode0, op0);
+ if (!(*insn_data[icode].operand[2].predicate) (op1, mode1))
+ op1 = copy_to_mode_reg (mode1, op1);
+
+ scratch = gen_reg_rtx (CCFPmode);
+
+ pat = GEN_FCN (icode) (scratch, op0, op1);
+ if (!pat)
+ return const0_rtx;
+
+ emit_insn (pat);
+
+ switch (form_int)
+ {
+ /* LT bit. */
+ case 0:
+ code = LT;
+ break;
+ /* GT bit. */
+ case 1:
+ code = GT;
+ break;
+ /* EQ bit. */
+ case 2:
+ code = EQ;
+ break;
+ /* UN bit. */
+ case 3:
+ emit_insn (gen_move_from_CR_ov_bit (target, scratch));
+ return target;
+ default:
+ error ("argument 1 of __builtin_paired_predicate is out of range");
+ return const0_rtx;
+ }
+
+ tmp = gen_rtx_fmt_ee (code, SImode, scratch, const0_rtx);
+ emit_move_insn (target, tmp);
+ return target;
+}
+
+static rtx
spe_expand_predicate_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat, scratch, tmp;
rtx ret;
bool success;
+ if (fcode == RS6000_BUILTIN_RECIP)
+ return rs6000_expand_binop_builtin (CODE_FOR_recipdf3, exp, target);
+
+ if (fcode == RS6000_BUILTIN_RECIPF)
+ return rs6000_expand_binop_builtin (CODE_FOR_recipsf3, exp, target);
+
+ if (fcode == RS6000_BUILTIN_RSQRTF)
+ return rs6000_expand_unop_builtin (CODE_FOR_rsqrtsf2, exp, target);
+
if (fcode == ALTIVEC_BUILTIN_MASK_FOR_LOAD
|| fcode == ALTIVEC_BUILTIN_MASK_FOR_STORE)
{
if (success)
return ret;
}
+ if (TARGET_PAIRED_FLOAT)
+ {
+ ret = paired_expand_builtin (exp, target, &success);
- gcc_assert (TARGET_ALTIVEC || TARGET_SPE);
+ if (success)
+ return ret;
+ }
+
+ gcc_assert (TARGET_ALTIVEC || TARGET_SPE || TARGET_PAIRED_FLOAT);
/* Handle simple unary operations. */
d = (struct builtin_description *) bdesc_1arg;
get_identifier ("__vector __pixel"),
pixel_V8HI_type_node));
+ if (TARGET_PAIRED_FLOAT)
+ paired_init_builtins ();
if (TARGET_SPE)
spe_init_builtins ();
if (TARGET_ALTIVEC)
altivec_init_builtins ();
- if (TARGET_ALTIVEC || TARGET_SPE)
+ if (TARGET_ALTIVEC || TARGET_SPE || TARGET_PAIRED_FLOAT)
rs6000_common_init_builtins ();
+ if (TARGET_PPC_GFXOPT)
+ {
+ tree ftype = build_function_type_list (float_type_node,
+ float_type_node,
+ float_type_node,
+ NULL_TREE);
+ def_builtin (MASK_PPC_GFXOPT, "__builtin_recipdivf", ftype,
+ RS6000_BUILTIN_RECIPF);
+
+ ftype = build_function_type_list (float_type_node,
+ float_type_node,
+ NULL_TREE);
+ def_builtin (MASK_PPC_GFXOPT, "__builtin_rsqrtf", ftype,
+ RS6000_BUILTIN_RSQRTF);
+ }
+ if (TARGET_POPCNTB)
+ {
+ tree ftype = build_function_type_list (double_type_node,
+ double_type_node,
+ double_type_node,
+ NULL_TREE);
+ def_builtin (MASK_POPCNTB, "__builtin_recipdiv", ftype,
+ RS6000_BUILTIN_RECIP);
+
+ }
#if TARGET_XCOFF
/* AIX libm provides clog as __clog. */
}
static void
+paired_init_builtins (void)
+{
+ const struct builtin_description *d;
+ size_t i;
+ tree endlink = void_list_node;
+
+ tree int_ftype_int_v2sf_v2sf
+ = build_function_type
+ (integer_type_node,
+ tree_cons (NULL_TREE, integer_type_node,
+ tree_cons (NULL_TREE, V2SF_type_node,
+ tree_cons (NULL_TREE, V2SF_type_node,
+ endlink))));
+ tree pcfloat_type_node =
+ build_pointer_type (build_qualified_type
+ (float_type_node, TYPE_QUAL_CONST));
+
+ tree v2sf_ftype_long_pcfloat = build_function_type_list (V2SF_type_node,
+ long_integer_type_node,
+ pcfloat_type_node,
+ NULL_TREE);
+ tree void_ftype_v2sf_long_pcfloat =
+ build_function_type_list (void_type_node,
+ V2SF_type_node,
+ long_integer_type_node,
+ pcfloat_type_node,
+ NULL_TREE);
+
+
+ def_builtin (0, "__builtin_paired_lx", v2sf_ftype_long_pcfloat,
+ PAIRED_BUILTIN_LX);
+
+
+ def_builtin (0, "__builtin_paired_stx", void_ftype_v2sf_long_pcfloat,
+ PAIRED_BUILTIN_STX);
+
+ /* Predicates. */
+ d = bdesc_paired_preds;
+ for (i = 0; i < ARRAY_SIZE (bdesc_paired_preds); ++i, d++)
+ {
+ tree type;
+
+ switch (insn_data[d->icode].operand[1].mode)
+ {
+ case V2SFmode:
+ type = int_ftype_int_v2sf_v2sf;
+ break;
+ default:
+ gcc_unreachable ();
+ }
+
+ def_builtin (d->mask, d->name, type, d->code);
+ }
+}
+
+static void
altivec_init_builtins (void)
{
const struct builtin_description *d;
const struct builtin_description *d;
size_t i;
+ tree v2sf_ftype_v2sf_v2sf_v2sf
+ = build_function_type_list (V2SF_type_node,
+ V2SF_type_node, V2SF_type_node,
+ V2SF_type_node, NULL_TREE);
+
tree v4sf_ftype_v4sf_v4sf_v16qi
= build_function_type_list (V4SF_type_node,
V4SF_type_node, V4SF_type_node,
opaque_V2SI_type_node,
opaque_V2SI_type_node, NULL_TREE);
- tree v2sf_ftype_v2sf_v2sf
+ tree v2sf_ftype_v2sf_v2sf_spe
= build_function_type_list (opaque_V2SF_type_node,
opaque_V2SF_type_node,
opaque_V2SF_type_node, NULL_TREE);
+ tree v2sf_ftype_v2sf_v2sf
+ = build_function_type_list (V2SF_type_node,
+ V2SF_type_node,
+ V2SF_type_node, NULL_TREE);
+
+
tree v2si_ftype_int_int
= build_function_type_list (opaque_V2SI_type_node,
integer_type_node, integer_type_node,
= build_function_type_list (opaque_V2SI_type_node,
opaque_V2SI_type_node, NULL_TREE);
- tree v2sf_ftype_v2sf
+ tree v2sf_ftype_v2sf_spe
= build_function_type_list (opaque_V2SF_type_node,
opaque_V2SF_type_node, NULL_TREE);
+ tree v2sf_ftype_v2sf
+ = build_function_type_list (V2SF_type_node,
+ V2SF_type_node, NULL_TREE);
+
tree v2sf_ftype_v2si
= build_function_type_list (opaque_V2SF_type_node,
opaque_V2SI_type_node, NULL_TREE);
case V16QImode:
type = v16qi_ftype_v16qi_v16qi_v16qi;
break;
+ case V2SFmode:
+ type = v2sf_ftype_v2sf_v2sf_v2sf;
+ break;
default:
gcc_unreachable ();
}
case V2SImode:
type = v2si_ftype_v2si_v2si;
break;
- case V2SFmode:
- type = v2sf_ftype_v2sf_v2sf;
+ case V2SFmode:
+ if (TARGET_PAIRED_FLOAT)
+ type = v2sf_ftype_v2sf_v2sf;
+ else
+ type = v2sf_ftype_v2sf_v2sf_spe;
break;
case SImode:
type = int_ftype_int_int;
else if (mode0 == V2SImode && mode1 == V2SImode)
type = v2si_ftype_v2si;
else if (mode0 == V2SFmode && mode1 == V2SFmode)
- type = v2sf_ftype_v2sf;
+ {
+ if (TARGET_PAIRED_FLOAT)
+ type = v2sf_ftype_v2sf;
+ else
+ type = v2sf_ftype_v2sf_spe;
+ }
else if (mode0 == V2SFmode && mode1 == V2SImode)
type = v2sf_ftype_v2si;
else if (mode0 == V2SImode && mode1 == V2SFmode)
ishift = GET_MODE_BITSIZE (SImode) - GET_MODE_BITSIZE (mode);
shift = GEN_INT (ishift);
+ used_m = change_address (used_m, SImode, 0);
}
else
{
rtx addrSI, aligned_addr;
int shift_mask = mode == QImode ? 0x18 : 0x10;
- addrSI = force_reg (SImode, gen_lowpart_common (SImode,
- XEXP (used_m, 0)));
+ addrSI = gen_lowpart_common (SImode,
+ force_reg (Pmode, XEXP (used_m, 0)));
+ addrSI = force_reg (SImode, addrSI);
shift = gen_reg_rtx (SImode);
emit_insn (gen_rlwinm (shift, addrSI, GEN_INT (3),
1, OPTAB_LIB_WIDEN);
used_m = change_address (used_m, SImode, aligned_addr);
set_mem_align (used_m, 32);
- /* It's safe to keep the old alias set of USED_M, because
- the operation is atomic and only affects the original
- USED_M. */
- if (GET_CODE (m) == NOT)
- m = gen_rtx_NOT (SImode, used_m);
- else
- m = used_m;
}
+ /* It's safe to keep the old alias set of USED_M, because
+ the operation is atomic and only affects the original
+ USED_M. */
+ if (GET_CODE (m) == NOT)
+ m = gen_rtx_NOT (SImode, used_m);
+ else
+ m = used_m;
if (GET_CODE (op) == NOT)
{
reg_mode = DECIMAL_FLOAT_MODE_P (mode) ? DDmode : DFmode;
else if (ALTIVEC_REGNO_P (reg))
reg_mode = V16QImode;
- else if (TARGET_E500_DOUBLE && mode == TFmode)
- reg_mode = DFmode;
+ else if (TARGET_E500_DOUBLE && (mode == TFmode || mode == TDmode))
+ reg_mode = DECIMAL_FLOAT_MODE_P (mode) ? DDmode : DFmode;
else
reg_mode = word_mode;
reg_mode_size = GET_MODE_SIZE (reg_mode);
int reg_size = TARGET_32BIT ? 4 : 8;
int ehrd_size;
int save_align;
+ int first_gp;
HOST_WIDE_INT non_fixed_size;
memset (&info, 0, sizeof (info));
/* Calculate which registers need to be saved & save area size. */
info_ptr->first_gp_reg_save = first_reg_to_save ();
/* Assume that we will have to save RS6000_PIC_OFFSET_TABLE_REGNUM,
- even if it currently looks like we won't. */
+ even if it currently looks like we won't. Reload may need it to
+ get at a constant; if so, it will have already created a constant
+ pool entry for it. */
if (((TARGET_TOC && TARGET_MINIMAL_TOC)
|| (flag_pic == 1 && DEFAULT_ABI == ABI_V4)
|| (flag_pic && DEFAULT_ABI == ABI_DARWIN))
+ && current_function_uses_const_pool
&& info_ptr->first_gp_reg_save > RS6000_PIC_OFFSET_TABLE_REGNUM)
- info_ptr->gp_size = reg_size * (32 - RS6000_PIC_OFFSET_TABLE_REGNUM);
+ first_gp = RS6000_PIC_OFFSET_TABLE_REGNUM;
else
- info_ptr->gp_size = reg_size * (32 - info_ptr->first_gp_reg_save);
+ first_gp = info_ptr->first_gp_reg_save;
+
+ info_ptr->gp_size = reg_size * (32 - first_gp);
/* For the SPE, we have an additional upper 32-bits on each GPR.
Ideally we should save the entire 64-bits only when the upper
+ info_ptr->parm_size);
if (TARGET_SPE_ABI && info_ptr->spe_64bit_regs_used != 0)
- info_ptr->spe_gp_size = 8 * (32 - info_ptr->first_gp_reg_save);
+ info_ptr->spe_gp_size = 8 * (32 - first_gp);
else
info_ptr->spe_gp_size = 0;
if (SPE_VECTOR_MODE (mode))
return true;
- if (TARGET_E500_DOUBLE && (mode == DFmode || mode == TFmode))
+ if (TARGET_E500_DOUBLE && (mode == DFmode || mode == TFmode
+ || mode == DDmode || mode == TDmode))
return true;
}
}
/* Some cases that need register indexed addressing. */
if ((TARGET_ALTIVEC_ABI && ALTIVEC_VECTOR_MODE (mode))
- || (TARGET_E500_DOUBLE && mode == DFmode)
+ || (TARGET_E500_DOUBLE && (mode == DFmode || mode == DDmode))
|| (TARGET_SPE_ABI
&& SPE_VECTOR_MODE (mode)
&& !SPE_CONST_OFFSET_OK (offset)))
int_rtx = GEN_INT (offset);
if ((TARGET_SPE_ABI && SPE_VECTOR_MODE (mode))
- || (TARGET_E500_DOUBLE && mode == DFmode))
+ || (TARGET_E500_DOUBLE && (mode == DFmode || mode == DDmode)))
{
offset_rtx = gen_rtx_REG (Pmode, FIXED_SCRATCH);
emit_move_insn (offset_rtx, int_rtx);
if (spe_regs_addressable_via_sp)
{
- spe_save_area_ptr = sp_reg_rtx;
+ spe_save_area_ptr = frame_reg_rtx;
spe_offset = info->spe_gp_save_offset + sp_offset;
}
else
}
spe_save_area_ptr = gen_rtx_REG (Pmode, 11);
- emit_insn (gen_addsi3 (spe_save_area_ptr, sp_reg_rtx,
+ emit_insn (gen_addsi3 (spe_save_area_ptr, frame_reg_rtx,
GEN_INT (info->spe_gp_save_offset + sp_offset)));
spe_offset = 0;
return;
}
- /* Set sp_offset based on the stack push from the prologue. */
- if (info->total_size < 32767)
+ /* frame_reg_rtx + sp_offset points to the top of this stack frame. */
+ if (info->push_p)
sp_offset = info->total_size;
/* Restore AltiVec registers if needed. */
}
}
- /* Restore VRSAVE if needed. */
- if (TARGET_ALTIVEC && TARGET_ALTIVEC_VRSAVE
- && info->vrsave_mask != 0)
- {
- rtx addr, mem, reg;
-
- addr = gen_rtx_PLUS (Pmode, frame_reg_rtx,
- GEN_INT (info->vrsave_save_offset + sp_offset));
- mem = gen_frame_mem (SImode, addr);
- reg = gen_rtx_REG (SImode, 12);
- emit_move_insn (reg, mem);
-
- emit_insn (generate_set_vrsave (reg, info, 1));
- }
-
- sp_offset = 0;
-
/* If we have a frame pointer, a call to alloca, or a large stack
frame, restore the old stack pointer using the backchain. Otherwise,
we know what size to update it with. */
emit_move_insn (frame_reg_rtx,
gen_rtx_MEM (Pmode, sp_reg_rtx));
+ sp_offset = 0;
}
- else if (info->push_p)
+ else if (info->push_p
+ && DEFAULT_ABI != ABI_V4
+ && !current_function_calls_eh_return)
{
- if (DEFAULT_ABI == ABI_V4
- || current_function_calls_eh_return)
- sp_offset = info->total_size;
- else
- {
- emit_insn (TARGET_32BIT
- ? gen_addsi3 (sp_reg_rtx, sp_reg_rtx,
- GEN_INT (info->total_size))
- : gen_adddi3 (sp_reg_rtx, sp_reg_rtx,
- GEN_INT (info->total_size)));
- }
+ emit_insn (TARGET_32BIT
+ ? gen_addsi3 (sp_reg_rtx, sp_reg_rtx,
+ GEN_INT (info->total_size))
+ : gen_adddi3 (sp_reg_rtx, sp_reg_rtx,
+ GEN_INT (info->total_size)));
+ sp_offset = 0;
+ }
+
+ /* Restore VRSAVE if needed. */
+ if (TARGET_ALTIVEC && TARGET_ALTIVEC_VRSAVE
+ && info->vrsave_mask != 0)
+ {
+ rtx addr, mem, reg;
+
+ addr = gen_rtx_PLUS (Pmode, frame_reg_rtx,
+ GEN_INT (info->vrsave_save_offset + sp_offset));
+ mem = gen_frame_mem (SImode, addr);
+ reg = gen_rtx_REG (SImode, 12);
+ emit_move_insn (reg, mem);
+
+ emit_insn (generate_set_vrsave (reg, info, 1));
}
/* Get the old lr if we saved it. */
&& info->spe_64bit_regs_used != 0
&& info->first_gp_reg_save != 32)
{
- rtx spe_save_area_ptr;
/* Determine whether we can address all of the registers that need
to be saved with an offset from the stack pointer that fits in
the small const field for SPE memory instructions. */
int spe_offset;
if (spe_regs_addressable_via_sp)
- {
- spe_save_area_ptr = frame_reg_rtx;
- spe_offset = info->spe_gp_save_offset + sp_offset;
- }
+ spe_offset = info->spe_gp_save_offset + sp_offset;
else
{
+ rtx old_frame_reg_rtx = frame_reg_rtx;
/* Make r11 point to the start of the SPE save area. We worried about
not clobbering it when we were saving registers in the prologue.
There's no need to worry here because the static chain is passed
anew to every function. */
- spe_save_area_ptr = gen_rtx_REG (Pmode, 11);
-
- emit_insn (gen_addsi3 (spe_save_area_ptr, frame_reg_rtx,
+ if (frame_reg_rtx == sp_reg_rtx)
+ frame_reg_rtx = gen_rtx_REG (Pmode, 11);
+ emit_insn (gen_addsi3 (frame_reg_rtx, old_frame_reg_rtx,
GEN_INT (info->spe_gp_save_offset + sp_offset)));
+ /* Keep the invariant that frame_reg_rtx + sp_offset points
+ at the top of the stack frame. */
+ sp_offset = -info->spe_gp_save_offset;
spe_offset = 0;
}
gcc_assert (SPE_CONST_OFFSET_OK (spe_offset + reg_size * i));
offset = GEN_INT (spe_offset + reg_size * i);
- addr = gen_rtx_PLUS (Pmode, spe_save_area_ptr, offset);
+ addr = gen_rtx_PLUS (Pmode, frame_reg_rtx, offset);
mem = gen_rtx_MEM (V2SImode, addr);
emit_move_insn (gen_rtx_REG (reg_mode, info->first_gp_reg_save + i),
/* This blockage is needed so that sched doesn't decide to move
the sp change before the register restores. */
rs6000_emit_stack_tie ();
- if (TARGET_SPE_ABI
- && info->spe_64bit_regs_used != 0
- && info->first_gp_reg_save != 32)
- emit_insn (gen_addsi3 (sp_reg_rtx, gen_rtx_REG (Pmode, 11),
- GEN_INT (-(info->spe_gp_save_offset + sp_offset))));
+ if (sp_offset != 0)
+ emit_insn (gen_addsi3 (sp_reg_rtx, frame_reg_rtx,
+ GEN_INT (sp_offset)));
else
emit_move_insn (sp_reg_rtx, frame_reg_rtx);
}
|| (GET_CODE (XEXP (b, 0)) == PLUS
&& GET_CODE (XEXP (XEXP (b, 0), 1)) == CONST_INT)))
{
- HOST_WIDE_INT val0 = 0, val1 = 0;
+ HOST_WIDE_INT val0 = 0, val1 = 0, val_diff;
rtx reg0, reg1;
- int val_diff;
if (GET_CODE (XEXP (a, 0)) == PLUS)
{
val_diff = val1 - val0;
return ((REGNO (reg0) == REGNO (reg1))
- && (val_diff == INTVAL (MEM_SIZE (a))
- || val_diff == -INTVAL (MEM_SIZE (b))));
+ && ((MEM_SIZE (a) && val_diff == INTVAL (MEM_SIZE (a)))
+ || (MEM_SIZE (b) && val_diff == -INTVAL (MEM_SIZE (b)))));
}
return false;
int i, j;
bool ret = false;
+ /* stack_tie does not produce any real memory traffic. */
+ if (GET_CODE (pat) == UNSPEC
+ && XINT (pat, 1) == UNSPEC_TIE)
+ return false;
+
if (GET_CODE (pat) == MEM)
return true;
}
}
+static inline bool
+compare_section_name (const char *section, const char *template)
+{
+ int len;
+
+ len = strlen (template);
+ return (strncmp (section, template, len) == 0
+ && (section[len] == 0 || section[len] == '.'));
+}
+
bool
rs6000_elf_in_small_data_p (const_tree decl)
{
if (TREE_CODE (decl) == VAR_DECL && DECL_SECTION_NAME (decl))
{
const char *section = TREE_STRING_POINTER (DECL_SECTION_NAME (decl));
- if (strcmp (section, ".sdata") == 0
- || strcmp (section, ".sdata2") == 0
- || strcmp (section, ".sbss") == 0
- || strcmp (section, ".sbss2") == 0
+ if (compare_section_name (section, ".sdata")
+ || compare_section_name (section, ".sdata2")
+ || compare_section_name (section, ".gnu.linkonce.s")
+ || compare_section_name (section, ".sbss")
+ || compare_section_name (section, ".sbss2")
+ || compare_section_name (section, ".gnu.linkonce.sb")
|| strcmp (section, ".PPC.EMB.sdata0") == 0
|| strcmp (section, ".PPC.EMB.sbss0") == 0)
return true;
static void
rs6000_xcoff_output_readonly_section_asm_op (const void *directive)
{
- fprintf (asm_out_file, "\t.csect %s[RO],3\n",
- *(const char *const *) directive);
+ fprintf (asm_out_file, "\t.csect %s[RO],%s\n",
+ *(const char *const *) directive,
+ XCOFF_CSECT_DEFAULT_ALIGNMENT_STR);
}
/* Likewise for read-write sections. */
static void
rs6000_xcoff_output_readwrite_section_asm_op (const void *directive)
{
- fprintf (asm_out_file, "\t.csect %s[RW],3\n",
- *(const char *const *) directive);
+ fprintf (asm_out_file, "\t.csect %s[RW],%s\n",
+ *(const char *const *) directive,
+ XCOFF_CSECT_DEFAULT_ALIGNMENT_STR);
}
/* A get_unnamed_section callback, used for switching to toc_section. */
return 4 + rs6000_register_move_cost (mode, class, GENERAL_REGS);
}
+/* Returns a code for a target-specific builtin that implements
+ reciprocal of the function, or NULL_TREE if not available. */
+
+static tree
+rs6000_builtin_reciprocal (unsigned int fn, bool md_fn,
+ bool sqrt ATTRIBUTE_UNUSED)
+{
+ if (! (TARGET_RECIP && TARGET_PPC_GFXOPT && !optimize_size
+ && flag_finite_math_only && !flag_trapping_math
+ && flag_unsafe_math_optimizations))
+ return NULL_TREE;
+
+ if (md_fn)
+ return NULL_TREE;
+ else
+ switch (fn)
+ {
+ case BUILT_IN_SQRTF:
+ return rs6000_builtin_decls[RS6000_BUILTIN_RSQRTF];
+
+ default:
+ return NULL_TREE;
+ }
+}
+
/* Newton-Raphson approximation of single-precision floating point divide n/d.
Assumes no trapping math and finite arguments. */
void
-rs6000_emit_swdivsf (rtx res, rtx n, rtx d)
+rs6000_emit_swdivsf (rtx dst, rtx n, rtx d)
{
rtx x0, e0, e1, y1, u0, v0, one;
emit_insn (gen_rtx_SET (VOIDmode, v0,
gen_rtx_MINUS (SFmode, n,
gen_rtx_MULT (SFmode, d, u0))));
- /* res = u0 + v0 * y1 */
- emit_insn (gen_rtx_SET (VOIDmode, res,
+ /* dst = u0 + v0 * y1 */
+ emit_insn (gen_rtx_SET (VOIDmode, dst,
gen_rtx_PLUS (SFmode,
gen_rtx_MULT (SFmode, v0, y1), u0)));
}
Assumes no trapping math and finite arguments. */
void
-rs6000_emit_swdivdf (rtx res, rtx n, rtx d)
+rs6000_emit_swdivdf (rtx dst, rtx n, rtx d)
{
rtx x0, e0, e1, e2, y1, y2, y3, u0, v0, one;
emit_insn (gen_rtx_SET (VOIDmode, v0,
gen_rtx_MINUS (DFmode, n,
gen_rtx_MULT (DFmode, d, u0))));
- /* res = u0 + v0 * y3 */
- emit_insn (gen_rtx_SET (VOIDmode, res,
+ /* dst = u0 + v0 * y3 */
+ emit_insn (gen_rtx_SET (VOIDmode, dst,
gen_rtx_PLUS (DFmode,
gen_rtx_MULT (DFmode, v0, y3), u0)));
}
+/* Newton-Raphson approximation of single-precision floating point rsqrt.
+ Assumes no trapping math and finite arguments. */
+
+void
+rs6000_emit_swrsqrtsf (rtx dst, rtx src)
+{
+ rtx x0, x1, x2, y1, u0, u1, u2, v0, v1, v2, t0,
+ half, one, halfthree, c1, cond, label;
+
+ x0 = gen_reg_rtx (SFmode);
+ x1 = gen_reg_rtx (SFmode);
+ x2 = gen_reg_rtx (SFmode);
+ y1 = gen_reg_rtx (SFmode);
+ u0 = gen_reg_rtx (SFmode);
+ u1 = gen_reg_rtx (SFmode);
+ u2 = gen_reg_rtx (SFmode);
+ v0 = gen_reg_rtx (SFmode);
+ v1 = gen_reg_rtx (SFmode);
+ v2 = gen_reg_rtx (SFmode);
+ t0 = gen_reg_rtx (SFmode);
+ halfthree = gen_reg_rtx (SFmode);
+ cond = gen_rtx_REG (CCFPmode, CR1_REGNO);
+ label = gen_rtx_LABEL_REF (VOIDmode, gen_label_rtx ());
+
+ /* check 0.0, 1.0, NaN, Inf by testing src * src = src */
+ emit_insn (gen_rtx_SET (VOIDmode, t0,
+ gen_rtx_MULT (SFmode, src, src)));
+
+ emit_insn (gen_rtx_SET (VOIDmode, cond,
+ gen_rtx_COMPARE (CCFPmode, t0, src)));
+ c1 = gen_rtx_EQ (VOIDmode, cond, const0_rtx);
+ emit_unlikely_jump (c1, label);
+
+ half = force_reg (SFmode, CONST_DOUBLE_FROM_REAL_VALUE (dconsthalf, SFmode));
+ one = force_reg (SFmode, CONST_DOUBLE_FROM_REAL_VALUE (dconst1, SFmode));
+
+ /* halfthree = 1.5 = 1.0 + 0.5 */
+ emit_insn (gen_rtx_SET (VOIDmode, halfthree,
+ gen_rtx_PLUS (SFmode, one, half)));
+
+ /* x0 = rsqrt estimate */
+ emit_insn (gen_rtx_SET (VOIDmode, x0,
+ gen_rtx_UNSPEC (SFmode, gen_rtvec (1, src),
+ UNSPEC_RSQRT)));
+
+ /* y1 = 0.5 * src = 1.5 * src - src -> fewer constants */
+ emit_insn (gen_rtx_SET (VOIDmode, y1,
+ gen_rtx_MINUS (SFmode,
+ gen_rtx_MULT (SFmode, src, halfthree),
+ src)));
+
+ /* x1 = x0 * (1.5 - y1 * (x0 * x0)) */
+ emit_insn (gen_rtx_SET (VOIDmode, u0,
+ gen_rtx_MULT (SFmode, x0, x0)));
+ emit_insn (gen_rtx_SET (VOIDmode, v0,
+ gen_rtx_MINUS (SFmode,
+ halfthree,
+ gen_rtx_MULT (SFmode, y1, u0))));
+ emit_insn (gen_rtx_SET (VOIDmode, x1,
+ gen_rtx_MULT (SFmode, x0, v0)));
+
+ /* x2 = x1 * (1.5 - y1 * (x1 * x1)) */
+ emit_insn (gen_rtx_SET (VOIDmode, u1,
+ gen_rtx_MULT (SFmode, x1, x1)));
+ emit_insn (gen_rtx_SET (VOIDmode, v1,
+ gen_rtx_MINUS (SFmode,
+ halfthree,
+ gen_rtx_MULT (SFmode, y1, u1))));
+ emit_insn (gen_rtx_SET (VOIDmode, x2,
+ gen_rtx_MULT (SFmode, x1, v1)));
+
+ /* dst = x2 * (1.5 - y1 * (x2 * x2)) */
+ emit_insn (gen_rtx_SET (VOIDmode, u2,
+ gen_rtx_MULT (SFmode, x2, x2)));
+ emit_insn (gen_rtx_SET (VOIDmode, v2,
+ gen_rtx_MINUS (SFmode,
+ halfthree,
+ gen_rtx_MULT (SFmode, y1, u2))));
+ emit_insn (gen_rtx_SET (VOIDmode, dst,
+ gen_rtx_MULT (SFmode, x2, v2)));
+
+ emit_label (XEXP (label, 0));
+}
+
/* Emit popcount intrinsic on TARGET_POPCNTB targets. DST is the
target, and SRC is the argument operand. */
&& ALTIVEC_VECTOR_MODE (mode))
regno = ALTIVEC_ARG_RETURN;
else if (TARGET_E500_DOUBLE && TARGET_HARD_FLOAT
- && (mode == DFmode || mode == DCmode
- || mode == TFmode || mode == TCmode))
+ && (mode == DFmode || mode == DDmode || mode == DCmode
+ || mode == TFmode || mode == TDmode || mode == TCmode))
return spe_build_register_parallel (mode, GP_ARG_RETURN);
else
regno = GP_ARG_RETURN;
else if (COMPLEX_MODE_P (mode) && targetm.calls.split_complex_arg)
return rs6000_complex_function_value (mode);
else if (TARGET_E500_DOUBLE && TARGET_HARD_FLOAT
- && (mode == DFmode || mode == DCmode
- || mode == TFmode || mode == TCmode))
+ && (mode == DFmode || mode == DDmode || mode == DCmode
+ || mode == TFmode || mode == TDmode || mode == TCmode))
return spe_build_register_parallel (mode, GP_ARG_RETURN);
else
regno = GP_ARG_RETURN;
if (TARGET_SPE
&& (SPE_VECTOR_MODE (GET_MODE (reg))
- || (TARGET_E500_DOUBLE && GET_MODE (reg) == DFmode)))
+ || (TARGET_E500_DOUBLE
+ && (GET_MODE (reg) == DFmode || GET_MODE (reg) == DDmode))))
;
else
return NULL_RTX;
rs6000_vector_mode_supported_p (enum machine_mode mode)
{
+ if (TARGET_PAIRED_FLOAT && PAIRED_VECTOR_MODE (mode))
+ return true;
+
if (TARGET_SPE && SPE_VECTOR_MODE (mode))
return true;
OSDN Git Service
