can be optimized to ap = __builtin_next_arg (0). */
if (!TARGET_64BIT || TARGET_64BIT_MS_ABI)
targetm.expand_builtin_va_start = NULL;
+
+#ifdef USE_IX86_CLD
+ /* Use -mcld by default for 32-bit code if configured with --enable-cld. */
+ if (!TARGET_64BIT)
+ target_flags |= MASK_CLD & ~target_flags_explicit;
+#endif
}
\f
/* Return true if this goes in large data/bss. */
{
int align;
if (type)
- align = TYPE_ALIGN (type);
+ {
+ if (TYPE_STRUCTURAL_EQUALITY_P (type))
+ align = TYPE_ALIGN (type);
+ else
+ align = TYPE_ALIGN (TYPE_CANONICAL (type));
+ }
else
align = GET_MODE_ALIGNMENT (mode);
if (align < PARM_BOUNDARY)
/* Return true iff type is returned in memory. */
-static int
+static int ATTRIBUTE_UNUSED
return_in_memory_32 (const_tree type, enum machine_mode mode)
{
HOST_WIDE_INT size;
return 0;
}
-static int
+static int ATTRIBUTE_UNUSED
return_in_memory_64 (const_tree type, enum machine_mode mode)
{
int needed_intregs, needed_sseregs;
return !examine_argument (mode, type, 1, &needed_intregs, &needed_sseregs);
}
-static int
+static int ATTRIBUTE_UNUSED
return_in_memory_ms_64 (const_tree type, enum machine_mode mode)
{
HOST_WIDE_INT size = int_size_in_bytes (type);
return (size != 1 && size != 2 && size != 4 && size != 8);
}
-int
-ix86_return_in_memory (const_tree type)
+static bool
+ix86_return_in_memory (const_tree type, const_tree fntype ATTRIBUTE_UNUSED)
{
- const enum machine_mode mode = type_natural_mode (type);
-
+#ifdef SUBTARGET_RETURN_IN_MEMORY
+ return SUBTARGET_RETURN_IN_MEMORY (type, fntype);
+#else
+ const enum machine_mode mode = type_natural_mode (type);
+
if (TARGET_64BIT_MS_ABI)
- return return_in_memory_ms_64 (type, mode);
- else if (TARGET_64BIT)
- return return_in_memory_64 (type, mode);
- else
- return return_in_memory_32 (type, mode);
+ return return_in_memory_ms_64 (type, mode);
+ else if (TARGET_64BIT)
+ return return_in_memory_64 (type, mode);
+ else
+ return return_in_memory_32 (type, mode);
+#endif
}
/* Return false iff TYPE is returned in memory. This version is used
but differs notably in that when MMX is available, 8-byte vectors
are returned in memory, rather than in MMX registers. */
-int
-ix86_sol10_return_in_memory (const_tree type)
+bool
+ix86_sol10_return_in_memory (const_tree type, const_tree fntype ATTRIBUTE_UNUSED)
{
int size;
enum machine_mode mode = type_natural_mode (type);
Choose the STRUCT_VALUE_RTX hook because that's (at present) only
called in response to actually generating a caller or callee that
- uses such a type. As opposed to RETURN_IN_MEMORY, which is called
+ uses such a type. As opposed to TARGET_RETURN_IN_MEMORY, which is called
via aggregate_value_p for general type probing from tree-ssa. */
static rtx
emit_insn (gen_prologue_use (pic_offset_table_rtx));
emit_insn (gen_blockage ());
}
+
+ /* Emit cld instruction if stringops are used in the function. */
+ if (TARGET_CLD && ix86_current_function_needs_cld)
+ emit_insn (gen_cld ());
}
/* Emit code to restore saved registers using MOV insns. First register
insns = get_insns ();
end_sequence ();
- CONST_OR_PURE_CALL_P (insns) = 1;
+ RTL_CONST_CALL_P (insns) = 1;
emit_libcall_block (insns, dest, rax, x);
}
else if (TARGET_64BIT && TARGET_GNU2_TLS)
note = gen_rtx_EXPR_LIST (VOIDmode, const0_rtx, NULL);
note = gen_rtx_EXPR_LIST (VOIDmode, ix86_tls_get_addr (), note);
- CONST_OR_PURE_CALL_P (insns) = 1;
+ RTL_CONST_CALL_P (insns) = 1;
emit_libcall_block (insns, base, rax, note);
}
else if (TARGET_64BIT && TARGET_GNU2_TLS)
&& standard_sse_constant_p (op1) <= 0)
op1 = validize_mem (force_const_mem (mode, op1));
- /* TDmode values are passed as TImode on the stack. TImode values
- are moved via xmm registers, and moving them to stack can result in
- unaligned memory access. Use ix86_expand_vector_move_misalign()
- if memory operand is not aligned correctly. */
+ /* We need to check memory alignment for SSE mode since attribute
+ can make operands unaligned. */
if (can_create_pseudo_p ()
- && (mode == TImode) && !TARGET_64BIT
+ && SSE_REG_MODE_P (mode)
&& ((MEM_P (op0) && (MEM_ALIGN (op0) < align))
|| (MEM_P (op1) && (MEM_ALIGN (op1) < align))))
{
/* Canonicalize operand order. */
if (ix86_swap_binary_operands_p (code, mode, operands))
{
- rtx temp = src1;
+ rtx temp;
+
+ /* It is invalid to swap operands of different modes. */
+ gcc_assert (GET_MODE (src1) == GET_MODE (src2));
+
+ temp = src1;
src1 = src2;
src2 = temp;
}
enum machine_mode mode = GET_MODE (dest);
rtx t2, t3, x;
- if (TARGET_SSE5)
- {
- rtx pcmov = gen_rtx_SET (mode, dest,
- gen_rtx_IF_THEN_ELSE (mode, cmp,
- op_true,
- op_false));
- emit_insn (pcmov);
- }
- else if (op_false == CONST0_RTX (mode))
+ if (op_false == CONST0_RTX (mode))
{
op_true = force_reg (mode, op_true);
x = gen_rtx_AND (mode, cmp, op_true);
x = gen_rtx_AND (mode, x, op_false);
emit_insn (gen_rtx_SET (VOIDmode, dest, x));
}
+ else if (TARGET_SSE5)
+ {
+ rtx pcmov = gen_rtx_SET (mode, dest,
+ gen_rtx_IF_THEN_ELSE (mode, cmp,
+ op_true,
+ op_false));
+ emit_insn (pcmov);
+ }
else
{
op_true = force_reg (mode, op_true);
cop0 = operands[4];
cop1 = operands[5];
- /* Canonicalize the comparison to EQ, GT, GTU. */
- switch (code)
- {
- case EQ:
- case GT:
- case GTU:
- break;
-
- case NE:
- case LE:
- case LEU:
- code = reverse_condition (code);
- negate = true;
- break;
-
- case GE:
- case GEU:
- code = reverse_condition (code);
- negate = true;
- /* FALLTHRU */
-
- case LT:
- case LTU:
- code = swap_condition (code);
- x = cop0, cop0 = cop1, cop1 = x;
- break;
-
- default:
- gcc_unreachable ();
- }
-
- /* Only SSE4.1/SSE4.2 supports V2DImode. */
- if (mode == V2DImode)
+ /* SSE5 supports all of the comparisons on all vector int types. */
+ if (!TARGET_SSE5)
{
+ /* Canonicalize the comparison to EQ, GT, GTU. */
switch (code)
{
case EQ:
- /* SSE4.1 supports EQ. */
- if (!TARGET_SSE4_1)
- return false;
- break;
-
case GT:
case GTU:
- /* SSE4.2 supports GT/GTU. */
- if (!TARGET_SSE4_2)
- return false;
+ break;
+
+ case NE:
+ case LE:
+ case LEU:
+ code = reverse_condition (code);
+ negate = true;
+ break;
+
+ case GE:
+ case GEU:
+ code = reverse_condition (code);
+ negate = true;
+ /* FALLTHRU */
+
+ case LT:
+ case LTU:
+ code = swap_condition (code);
+ x = cop0, cop0 = cop1, cop1 = x;
break;
default:
gcc_unreachable ();
}
- }
- /* Unsigned parallel compare is not supported by the hardware. Play some
- tricks to turn this into a signed comparison against 0. */
- if (code == GTU)
- {
- cop0 = force_reg (mode, cop0);
+ /* Only SSE4.1/SSE4.2 supports V2DImode. */
+ if (mode == V2DImode)
+ {
+ switch (code)
+ {
+ case EQ:
+ /* SSE4.1 supports EQ. */
+ if (!TARGET_SSE4_1)
+ return false;
+ break;
- switch (mode)
+ case GT:
+ case GTU:
+ /* SSE4.2 supports GT/GTU. */
+ if (!TARGET_SSE4_2)
+ return false;
+ break;
+
+ default:
+ gcc_unreachable ();
+ }
+ }
+
+ /* Unsigned parallel compare is not supported by the hardware. Play some
+ tricks to turn this into a signed comparison against 0. */
+ if (code == GTU)
{
- case V4SImode:
- case V2DImode:
- {
- rtx t1, t2, mask;
-
- /* Perform a parallel modulo subtraction. */
- t1 = gen_reg_rtx (mode);
- emit_insn ((mode == V4SImode
- ? gen_subv4si3
- : gen_subv2di3) (t1, cop0, cop1));
-
- /* Extract the original sign bit of op0. */
- mask = ix86_build_signbit_mask (GET_MODE_INNER (mode),
- true, false);
- t2 = gen_reg_rtx (mode);
- emit_insn ((mode == V4SImode
- ? gen_andv4si3
- : gen_andv2di3) (t2, cop0, mask));
-
- /* XOR it back into the result of the subtraction. This results
- in the sign bit set iff we saw unsigned underflow. */
- x = gen_reg_rtx (mode);
- emit_insn ((mode == V4SImode
- ? gen_xorv4si3
- : gen_xorv2di3) (x, t1, t2));
-
- code = GT;
- }
- break;
+ cop0 = force_reg (mode, cop0);
- case V16QImode:
- case V8HImode:
- /* Perform a parallel unsigned saturating subtraction. */
- x = gen_reg_rtx (mode);
- emit_insn (gen_rtx_SET (VOIDmode, x,
- gen_rtx_US_MINUS (mode, cop0, cop1)));
+ switch (mode)
+ {
+ case V4SImode:
+ case V2DImode:
+ {
+ rtx t1, t2, mask;
+
+ /* Perform a parallel modulo subtraction. */
+ t1 = gen_reg_rtx (mode);
+ emit_insn ((mode == V4SImode
+ ? gen_subv4si3
+ : gen_subv2di3) (t1, cop0, cop1));
+
+ /* Extract the original sign bit of op0. */
+ mask = ix86_build_signbit_mask (GET_MODE_INNER (mode),
+ true, false);
+ t2 = gen_reg_rtx (mode);
+ emit_insn ((mode == V4SImode
+ ? gen_andv4si3
+ : gen_andv2di3) (t2, cop0, mask));
+
+ /* XOR it back into the result of the subtraction. This results
+ in the sign bit set iff we saw unsigned underflow. */
+ x = gen_reg_rtx (mode);
+ emit_insn ((mode == V4SImode
+ ? gen_xorv4si3
+ : gen_xorv2di3) (x, t1, t2));
+
+ code = GT;
+ }
+ break;
- code = EQ;
- negate = !negate;
- break;
+ case V16QImode:
+ case V8HImode:
+ /* Perform a parallel unsigned saturating subtraction. */
+ x = gen_reg_rtx (mode);
+ emit_insn (gen_rtx_SET (VOIDmode, x,
+ gen_rtx_US_MINUS (mode, cop0, cop1)));
- default:
- gcc_unreachable ();
- }
+ code = EQ;
+ negate = !negate;
+ break;
+
+ default:
+ gcc_unreachable ();
+ }
- cop0 = x;
- cop1 = CONST0_RTX (mode);
+ cop0 = x;
+ cop1 = CONST0_RTX (mode);
+ }
}
x = ix86_expand_sse_cmp (operands[0], code, cop0, cop1,
}
/* This function performs the same task as ix86_expand_sse_unpack,
- but with amdfam15 instructions. */
-
-#define PPERM_SRC 0x00 /* copy source */
-#define PPERM_INVERT 0x20 /* invert source */
-#define PPERM_REVERSE 0x40 /* bit reverse source */
-#define PPERM_REV_INV 0x60 /* bit reverse & invert src */
-#define PPERM_ZERO 0x80 /* all 0's */
-#define PPERM_ONES 0xa0 /* all 1's */
-#define PPERM_SIGN 0xc0 /* propagate sign bit */
-#define PPERM_INV_SIGN 0xe0 /* invert & propagate sign */
-
-#define PPERM_SRC1 0x00 /* use first source byte */
-#define PPERM_SRC2 0x10 /* use second source byte */
+ but with sse5 instructions. */
void
ix86_expand_sse5_unpack (rtx operands[2], bool unsigned_p, bool high_p)
return align;
}
-/* Compute the alignment for a local variable.
- TYPE is the data type, and ALIGN is the alignment that
- the object would ordinarily have. The value of this macro is used
- instead of that alignment to align the object. */
+/* Compute the alignment for a local variable or a stack slot. TYPE is
+ the data type, MODE is the widest mode available and ALIGN is the
+ alignment that the object would ordinarily have. The value of this
+ macro is used instead of that alignment to align the object. */
-int
-ix86_local_alignment (tree type, int align)
+unsigned int
+ix86_local_alignment (tree type, enum machine_mode mode,
+ unsigned int align)
{
+ /* If TYPE is NULL, we are allocating a stack slot for caller-save
+ register in MODE. We will return the largest alignment of XF
+ and DF. */
+ if (!type)
+ {
+ if (mode == XFmode && align < GET_MODE_ALIGNMENT (DFmode))
+ align = GET_MODE_ALIGNMENT (DFmode);
+ return align;
+ }
+
/* x86-64 ABI requires arrays greater than 16 bytes to be aligned
to 16byte boundary. */
if (TARGET_64BIT)
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdneq", IX86_BUILTIN_UCOMINEQSD, LTGT, 0 },
};
-static const struct builtin_description bdesc_ptest[] =
-{
- /* SSE4.1 */
- { OPTION_MASK_ISA_ROUND, CODE_FOR_sse4_1_ptest, "__builtin_ia32_ptestz128", IX86_BUILTIN_PTESTZ, EQ, 0 },
- { OPTION_MASK_ISA_ROUND, CODE_FOR_sse4_1_ptest, "__builtin_ia32_ptestc128", IX86_BUILTIN_PTESTC, LTU, 0 },
- { OPTION_MASK_ISA_ROUND, CODE_FOR_sse4_1_ptest, "__builtin_ia32_ptestnzc128", IX86_BUILTIN_PTESTNZC, GTU, 0 },
-};
-
static const struct builtin_description bdesc_pcmpestr[] =
{
/* SSE4.2 */
FLOAT128_FTYPE_FLOAT128,
FLOAT_FTYPE_FLOAT,
FLOAT128_FTYPE_FLOAT128_FLOAT128,
+ INT_FTYPE_V2DI_V2DI_PTEST,
INT64_FTYPE_V4SF,
INT64_FTYPE_V2DF,
INT_FTYPE_V16QI,
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_movntv4sf, "__builtin_ia32_movntps", IX86_BUILTIN_MOVNTPS, UNKNOWN, (int) VOID_FTYPE_PFLOAT_V4SF },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_movups, "__builtin_ia32_loadups", IX86_BUILTIN_LOADUPS, UNKNOWN, (int) V4SF_FTYPE_PCFLOAT },
- { OPTION_MASK_ISA_SSE, CODE_FOR_sse_loadhps, "__builtin_ia32_loadhps", IX86_BUILTIN_LOADHPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_PCV2SF },
- { OPTION_MASK_ISA_SSE, CODE_FOR_sse_loadlps, "__builtin_ia32_loadlps", IX86_BUILTIN_LOADLPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_PCV2SF },
+ { OPTION_MASK_ISA_SSE, CODE_FOR_sse_loadhps_exp, "__builtin_ia32_loadhps", IX86_BUILTIN_LOADHPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_PCV2SF },
+ { OPTION_MASK_ISA_SSE, CODE_FOR_sse_loadlps_exp, "__builtin_ia32_loadlps", IX86_BUILTIN_LOADLPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_PCV2SF },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_storehps, "__builtin_ia32_storehps", IX86_BUILTIN_STOREHPS, UNKNOWN, (int) VOID_FTYPE_PV2SF_V4SF },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_storelps, "__builtin_ia32_storelps", IX86_BUILTIN_STORELPS, UNKNOWN, (int) VOID_FTYPE_PV2SF_V4SF },
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_movupd, "__builtin_ia32_loadupd", IX86_BUILTIN_LOADUPD, UNKNOWN, (int) V2DF_FTYPE_PCDOUBLE },
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_movdqu, "__builtin_ia32_loaddqu", IX86_BUILTIN_LOADDQU, UNKNOWN, (int) V16QI_FTYPE_PCCHAR },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_loadhpd, "__builtin_ia32_loadhpd", IX86_BUILTIN_LOADHPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_PCDOUBLE },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_loadlpd, "__builtin_ia32_loadlpd", IX86_BUILTIN_LOADLPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_PCDOUBLE },
+ { OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_loadhpd_exp, "__builtin_ia32_loadhpd", IX86_BUILTIN_LOADHPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_PCDOUBLE },
+ { OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_loadlpd_exp, "__builtin_ia32_loadlpd", IX86_BUILTIN_LOADLPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_PCDOUBLE },
/* SSE3 */
{ OPTION_MASK_ISA_SSE3, CODE_FOR_sse3_lddqu, "__builtin_ia32_lddqu", IX86_BUILTIN_LDDQU, UNKNOWN, (int) V16QI_FTYPE_PCCHAR },
{ OPTION_MASK_ISA_SSE, CODE_FOR_xorv4sf3, "__builtin_ia32_xorps", IX86_BUILTIN_XORPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_movss, "__builtin_ia32_movss", IX86_BUILTIN_MOVSS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF },
- { OPTION_MASK_ISA_SSE, CODE_FOR_sse_movhlps, "__builtin_ia32_movhlps", IX86_BUILTIN_MOVHLPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF },
- { OPTION_MASK_ISA_SSE, CODE_FOR_sse_movlhps, "__builtin_ia32_movlhps", IX86_BUILTIN_MOVLHPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF },
+ { OPTION_MASK_ISA_SSE, CODE_FOR_sse_movhlps_exp, "__builtin_ia32_movhlps", IX86_BUILTIN_MOVHLPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF },
+ { OPTION_MASK_ISA_SSE, CODE_FOR_sse_movlhps_exp, "__builtin_ia32_movlhps", IX86_BUILTIN_MOVLHPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_unpckhps, "__builtin_ia32_unpckhps", IX86_BUILTIN_UNPCKHPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF },
{ OPTION_MASK_ISA_SSE, CODE_FOR_sse_unpcklps, "__builtin_ia32_unpcklps", IX86_BUILTIN_UNPCKLPS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF },
{ OPTION_MASK_ISA_SSE2, CODE_FOR_xorv2df3, "__builtin_ia32_xorpd", IX86_BUILTIN_XORPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF },
{ OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_movsd, "__builtin_ia32_movsd", IX86_BUILTIN_MOVSD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_unpckhpd, "__builtin_ia32_unpckhpd", IX86_BUILTIN_UNPCKHPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF },
- { OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_unpcklpd, "__builtin_ia32_unpcklpd", IX86_BUILTIN_UNPCKLPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF },
+ { OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_unpckhpd_exp, "__builtin_ia32_unpckhpd", IX86_BUILTIN_UNPCKHPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF },
+ { OPTION_MASK_ISA_SSE2, CODE_FOR_sse2_unpcklpd_exp, "__builtin_ia32_unpcklpd", IX86_BUILTIN_UNPCKLPD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF },
{ OPTION_MASK_ISA_SSE2, CODE_FOR_vec_pack_sfix_v2df, "__builtin_ia32_vec_pack_sfix", IX86_BUILTIN_VEC_PACK_SFIX, UNKNOWN, (int) V4SI_FTYPE_V2DF_V2DF },
{ OPTION_MASK_ISA_ROUND, CODE_FOR_sse4_1_roundsd, "__builtin_ia32_roundsd", IX86_BUILTIN_ROUNDSD, UNKNOWN, (int) V2DF_FTYPE_V2DF_V2DF_INT },
{ OPTION_MASK_ISA_ROUND, CODE_FOR_sse4_1_roundss, "__builtin_ia32_roundss", IX86_BUILTIN_ROUNDSS, UNKNOWN, (int) V4SF_FTYPE_V4SF_V4SF_INT },
+ { OPTION_MASK_ISA_ROUND, CODE_FOR_sse4_1_ptest, "__builtin_ia32_ptestz128", IX86_BUILTIN_PTESTZ, EQ, (int) INT_FTYPE_V2DI_V2DI_PTEST },
+ { OPTION_MASK_ISA_ROUND, CODE_FOR_sse4_1_ptest, "__builtin_ia32_ptestc128", IX86_BUILTIN_PTESTC, LTU, (int) INT_FTYPE_V2DI_V2DI_PTEST },
+ { OPTION_MASK_ISA_ROUND, CODE_FOR_sse4_1_ptest, "__builtin_ia32_ptestnzc128", IX86_BUILTIN_PTESTNZC, GTU, (int) INT_FTYPE_V2DI_V2DI_PTEST },
+
/* SSE4.2 */
{ OPTION_MASK_ISA_SSE4_2, CODE_FOR_sse4_2_gtv2di3, "__builtin_ia32_pcmpgtq", IX86_BUILTIN_PCMPGTQ, UNKNOWN, (int) V2DI_FTYPE_V2DI_V2DI },
{ OPTION_MASK_ISA_SSE4_2, CODE_FOR_sse4_2_crc32qi, "__builtin_ia32_crc32qi", IX86_BUILTIN_CRC32QI, UNKNOWN, (int) UINT_FTYPE_UINT_UCHAR },
{ OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_pmacsdqh, "__builtin_ia32_pmacsdqh", IX86_BUILTIN_PMACSDQH, 0, (int)MULTI_ARG_3_SI_DI },
{ OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_pmadcsswd, "__builtin_ia32_pmadcsswd", IX86_BUILTIN_PMADCSSWD, 0, (int)MULTI_ARG_3_HI_SI },
{ OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_pmadcswd, "__builtin_ia32_pmadcswd", IX86_BUILTIN_PMADCSWD, 0, (int)MULTI_ARG_3_HI_SI },
- { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_rotlv2di3, "__builtin_ia32_protq", IX86_BUILTIN_PROTQ, 0, (int)MULTI_ARG_2_DI },
- { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_rotlv4si3, "__builtin_ia32_protd", IX86_BUILTIN_PROTD, 0, (int)MULTI_ARG_2_SI },
- { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_rotlv8hi3, "__builtin_ia32_protw", IX86_BUILTIN_PROTW, 0, (int)MULTI_ARG_2_HI },
- { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_rotlv16qi3, "__builtin_ia32_protb", IX86_BUILTIN_PROTB, 0, (int)MULTI_ARG_2_QI },
- { OPTION_MASK_ISA_SSE5, CODE_FOR_rotlv2di3, "__builtin_ia32_protqi", IX86_BUILTIN_PROTQ_IMM, 0, (int)MULTI_ARG_2_DI_IMM },
- { OPTION_MASK_ISA_SSE5, CODE_FOR_rotlv4si3, "__builtin_ia32_protdi", IX86_BUILTIN_PROTD_IMM, 0, (int)MULTI_ARG_2_SI_IMM },
- { OPTION_MASK_ISA_SSE5, CODE_FOR_rotlv8hi3, "__builtin_ia32_protwi", IX86_BUILTIN_PROTW_IMM, 0, (int)MULTI_ARG_2_HI_IMM },
- { OPTION_MASK_ISA_SSE5, CODE_FOR_rotlv16qi3, "__builtin_ia32_protbi", IX86_BUILTIN_PROTB_IMM, 0, (int)MULTI_ARG_2_QI_IMM },
+ { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_vrotlv2di3, "__builtin_ia32_protq", IX86_BUILTIN_PROTQ, 0, (int)MULTI_ARG_2_DI },
+ { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_vrotlv4si3, "__builtin_ia32_protd", IX86_BUILTIN_PROTD, 0, (int)MULTI_ARG_2_SI },
+ { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_vrotlv8hi3, "__builtin_ia32_protw", IX86_BUILTIN_PROTW, 0, (int)MULTI_ARG_2_HI },
+ { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_vrotlv16qi3, "__builtin_ia32_protb", IX86_BUILTIN_PROTB, 0, (int)MULTI_ARG_2_QI },
+ { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_rotlv2di3, "__builtin_ia32_protqi", IX86_BUILTIN_PROTQ_IMM, 0, (int)MULTI_ARG_2_DI_IMM },
+ { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_rotlv4si3, "__builtin_ia32_protdi", IX86_BUILTIN_PROTD_IMM, 0, (int)MULTI_ARG_2_SI_IMM },
+ { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_rotlv8hi3, "__builtin_ia32_protwi", IX86_BUILTIN_PROTW_IMM, 0, (int)MULTI_ARG_2_HI_IMM },
+ { OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_rotlv16qi3, "__builtin_ia32_protbi", IX86_BUILTIN_PROTB_IMM, 0, (int)MULTI_ARG_2_QI_IMM },
{ OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_ashlv2di3, "__builtin_ia32_pshaq", IX86_BUILTIN_PSHAQ, 0, (int)MULTI_ARG_2_DI },
{ OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_ashlv4si3, "__builtin_ia32_pshad", IX86_BUILTIN_PSHAD, 0, (int)MULTI_ARG_2_SI },
{ OPTION_MASK_ISA_SSE5, CODE_FOR_sse5_ashlv8hi3, "__builtin_ia32_pshaw", IX86_BUILTIN_PSHAW, 0, (int)MULTI_ARG_2_HI },
case FLOAT_FTYPE_FLOAT:
type = float_ftype_float;
break;
+ case INT_FTYPE_V2DI_V2DI_PTEST:
+ type = int_ftype_v2di_v2di;
+ break;
case INT64_FTYPE_V4SF:
type = int64_ftype_v4sf;
break;
else
def_builtin_const (d->mask, d->name, int_ftype_v4sf_v4sf, d->code);
- /* ptest insns. */
- for (i = 0, d = bdesc_ptest; i < ARRAY_SIZE (bdesc_ptest); i++, d++)
- def_builtin_const (d->mask, d->name, int_ftype_v2di_v2di, d->code);
-
/* SSE */
def_builtin (OPTION_MASK_ISA_SSE, "__builtin_ia32_ldmxcsr", void_ftype_unsigned, IX86_BUILTIN_LDMXCSR);
def_builtin (OPTION_MASK_ISA_SSE, "__builtin_ia32_stmxcsr", unsigned_ftype_void, IX86_BUILTIN_STMXCSR);
static rtx
ix86_expand_binop_builtin (enum insn_code icode, tree exp, rtx target)
{
- rtx pat, xops[3];
+ rtx pat;
tree arg0 = CALL_EXPR_ARG (exp, 0);
tree arg1 = CALL_EXPR_ARG (exp, 1);
rtx op0 = expand_normal (arg0);
if (!(*insn_data[icode].operand[2].predicate) (op1, mode1))
op1 = copy_to_mode_reg (mode1, op1);
- /* ??? Using ix86_fixup_binary_operands is problematic when
- we've got mismatched modes. Fake it. */
-
- xops[0] = target;
- xops[1] = op0;
- xops[2] = op1;
-
- if (tmode == mode0 && tmode == mode1)
- {
- target = ix86_fixup_binary_operands (UNKNOWN, tmode, xops);
- op0 = xops[1];
- op1 = xops[2];
- }
- else if (optimize || !ix86_binary_operator_ok (UNKNOWN, tmode, xops))
- {
- op0 = force_reg (mode0, op0);
- op1 = force_reg (mode1, op1);
- target = gen_reg_rtx (tmode);
- }
-
pat = GEN_FCN (icode) (target, op0, op1);
if (! pat)
return 0;
+
emit_insn (pat);
+
return target;
}
return target;
}
-/* Subroutine of ix86_expand_builtin to take care of insns with
- variable number of operands. */
+/* Subroutine of ix86_expand_builtin to take care of comi insns. */
static rtx
-ix86_expand_args_builtin (const struct builtin_description *d,
- tree exp, rtx target)
+ix86_expand_sse_comi (const struct builtin_description *d, tree exp,
+ rtx target)
{
- rtx pat, real_target;
- unsigned int i, nargs;
- unsigned int nargs_constant = 0;
- int num_memory = 0;
- struct
- {
- rtx op;
- enum machine_mode mode;
- } args[4];
- bool last_arg_count = false;
- enum insn_code icode = d->icode;
- const struct insn_data *insn_p = &insn_data[icode];
- enum machine_mode tmode = insn_p->operand[0].mode;
- enum machine_mode rmode = VOIDmode;
- bool swap = false;
+ rtx pat;
+ tree arg0 = CALL_EXPR_ARG (exp, 0);
+ tree arg1 = CALL_EXPR_ARG (exp, 1);
+ rtx op0 = expand_normal (arg0);
+ rtx op1 = expand_normal (arg1);
+ enum machine_mode mode0 = insn_data[d->icode].operand[0].mode;
+ enum machine_mode mode1 = insn_data[d->icode].operand[1].mode;
enum rtx_code comparison = d->comparison;
- switch ((enum ix86_builtin_type) d->flag)
+ if (VECTOR_MODE_P (mode0))
+ op0 = safe_vector_operand (op0, mode0);
+ if (VECTOR_MODE_P (mode1))
+ op1 = safe_vector_operand (op1, mode1);
+
+ /* Swap operands if we have a comparison that isn't available in
+ hardware. */
+ if (d->flag & BUILTIN_DESC_SWAP_OPERANDS)
{
- case FLOAT128_FTYPE_FLOAT128:
- case FLOAT_FTYPE_FLOAT:
- case INT64_FTYPE_V4SF:
- case INT64_FTYPE_V2DF:
- case INT_FTYPE_V16QI:
- case INT_FTYPE_V8QI:
- case INT_FTYPE_V4SF:
- case INT_FTYPE_V2DF:
- case V16QI_FTYPE_V16QI:
- case V8HI_FTYPE_V8HI:
- case V8HI_FTYPE_V16QI:
- case V8QI_FTYPE_V8QI:
- case V4SI_FTYPE_V4SI:
- case V4SI_FTYPE_V16QI:
- case V4SI_FTYPE_V4SF:
- case V4SI_FTYPE_V8HI:
- case V4SI_FTYPE_V2DF:
- case V4HI_FTYPE_V4HI:
- case V4SF_FTYPE_V4SF:
- case V4SF_FTYPE_V4SI:
- case V4SF_FTYPE_V2DF:
- case V2DI_FTYPE_V2DI:
- case V2DI_FTYPE_V16QI:
- case V2DI_FTYPE_V8HI:
- case V2DI_FTYPE_V4SI:
- case V2DF_FTYPE_V2DF:
- case V2DF_FTYPE_V4SI:
- case V2DF_FTYPE_V4SF:
- case V2DF_FTYPE_V2SI:
- case V2SI_FTYPE_V2SI:
- case V2SI_FTYPE_V4SF:
- case V2SI_FTYPE_V2SF:
- case V2SI_FTYPE_V2DF:
- case V2SF_FTYPE_V2SF:
- case V2SF_FTYPE_V2SI:
- nargs = 1;
- break;
- case V4SF_FTYPE_V4SF_VEC_MERGE:
- case V2DF_FTYPE_V2DF_VEC_MERGE:
- return ix86_expand_unop_vec_merge_builtin (icode, exp, target);
- case FLOAT128_FTYPE_FLOAT128_FLOAT128:
- case V16QI_FTYPE_V16QI_V16QI:
- case V16QI_FTYPE_V8HI_V8HI:
- case V8QI_FTYPE_V8QI_V8QI:
- case V8QI_FTYPE_V4HI_V4HI:
- case V8HI_FTYPE_V8HI_V8HI:
- case V8HI_FTYPE_V16QI_V16QI:
- case V8HI_FTYPE_V4SI_V4SI:
- case V4SI_FTYPE_V4SI_V4SI:
- case V4SI_FTYPE_V8HI_V8HI:
- case V4SI_FTYPE_V4SF_V4SF:
- case V4SI_FTYPE_V2DF_V2DF:
- case V4HI_FTYPE_V4HI_V4HI:
- case V4HI_FTYPE_V8QI_V8QI:
- case V4HI_FTYPE_V2SI_V2SI:
- case V4SF_FTYPE_V4SF_V4SF:
- case V4SF_FTYPE_V4SF_V2SI:
- case V4SF_FTYPE_V4SF_V2DF:
- case V4SF_FTYPE_V4SF_DI:
- case V4SF_FTYPE_V4SF_SI:
- case V2DI_FTYPE_V2DI_V2DI:
- case V2DI_FTYPE_V16QI_V16QI:
- case V2DI_FTYPE_V4SI_V4SI:
- case V2DI_FTYPE_V2DI_V16QI:
- case V2DI_FTYPE_V2DF_V2DF:
- case V2SI_FTYPE_V2SI_V2SI:
- case V2SI_FTYPE_V4HI_V4HI:
- case V2SI_FTYPE_V2SF_V2SF:
- case V2DF_FTYPE_V2DF_V2DF:
- case V2DF_FTYPE_V2DF_V4SF:
- case V2DF_FTYPE_V2DF_DI:
- case V2DF_FTYPE_V2DF_SI:
- case V2SF_FTYPE_V2SF_V2SF:
- case V1DI_FTYPE_V1DI_V1DI:
- case V1DI_FTYPE_V8QI_V8QI:
- case V1DI_FTYPE_V2SI_V2SI:
- if (comparison == UNKNOWN)
- return ix86_expand_binop_builtin (icode, exp, target);
- nargs = 2;
- break;
- case V4SF_FTYPE_V4SF_V4SF_SWAP:
- case V2DF_FTYPE_V2DF_V2DF_SWAP:
- gcc_assert (comparison != UNKNOWN);
- nargs = 2;
- swap = true;
- break;
- case V8HI_FTYPE_V8HI_V8HI_COUNT:
- case V8HI_FTYPE_V8HI_SI_COUNT:
- case V4SI_FTYPE_V4SI_V4SI_COUNT:
- case V4SI_FTYPE_V4SI_SI_COUNT:
- case V4HI_FTYPE_V4HI_V4HI_COUNT:
- case V4HI_FTYPE_V4HI_SI_COUNT:
- case V2DI_FTYPE_V2DI_V2DI_COUNT:
- case V2DI_FTYPE_V2DI_SI_COUNT:
- case V2SI_FTYPE_V2SI_V2SI_COUNT:
- case V2SI_FTYPE_V2SI_SI_COUNT:
- case V1DI_FTYPE_V1DI_V1DI_COUNT:
- case V1DI_FTYPE_V1DI_SI_COUNT:
- nargs = 2;
- last_arg_count = true;
- break;
- case UINT64_FTYPE_UINT64_UINT64:
- case UINT_FTYPE_UINT_UINT:
- case UINT_FTYPE_UINT_USHORT:
- case UINT_FTYPE_UINT_UCHAR:
- nargs = 2;
- break;
- case V2DI2TI_FTYPE_V2DI_INT:
- nargs = 2;
- rmode = V2DImode;
- nargs_constant = 1;
- break;
- case V8HI_FTYPE_V8HI_INT:
- case V4SI_FTYPE_V4SI_INT:
- case V4HI_FTYPE_V4HI_INT:
- case V4SF_FTYPE_V4SF_INT:
- case V2DI_FTYPE_V2DI_INT:
- case V2DF_FTYPE_V2DF_INT:
- nargs = 2;
- nargs_constant = 1;
- break;
- case V16QI_FTYPE_V16QI_V16QI_V16QI:
- case V4SF_FTYPE_V4SF_V4SF_V4SF:
- case V2DF_FTYPE_V2DF_V2DF_V2DF:
- nargs = 3;
- break;
- case V16QI_FTYPE_V16QI_V16QI_INT:
- case V8HI_FTYPE_V8HI_V8HI_INT:
- case V4SI_FTYPE_V4SI_V4SI_INT:
- case V4SF_FTYPE_V4SF_V4SF_INT:
- case V2DI_FTYPE_V2DI_V2DI_INT:
- case V2DF_FTYPE_V2DF_V2DF_INT:
- nargs = 3;
- nargs_constant = 1;
- break;
- case V2DI2TI_FTYPE_V2DI_V2DI_INT:
- nargs = 3;
- rmode = V2DImode;
- nargs_constant = 1;
- break;
- case V1DI2DI_FTYPE_V1DI_V1DI_INT:
- nargs = 3;
- rmode = DImode;
- nargs_constant = 1;
- break;
- case V2DI_FTYPE_V2DI_UINT_UINT:
- nargs = 3;
- nargs_constant = 2;
- break;
- case V2DI_FTYPE_V2DI_V2DI_UINT_UINT:
- nargs = 4;
- nargs_constant = 2;
- break;
- default:
- gcc_unreachable ();
- }
-
- gcc_assert (nargs <= ARRAY_SIZE (args));
-
- if (comparison != UNKNOWN)
- {
- gcc_assert (nargs == 2);
- return ix86_expand_sse_compare (d, exp, target, swap);
+ rtx tmp = op1;
+ op1 = op0;
+ op0 = tmp;
}
- if (rmode == VOIDmode || rmode == tmode)
- {
- if (optimize
- || target == 0
- || GET_MODE (target) != tmode
- || ! (*insn_p->operand[0].predicate) (target, tmode))
- target = gen_reg_rtx (tmode);
- real_target = target;
- }
- else
- {
- target = gen_reg_rtx (rmode);
- real_target = simplify_gen_subreg (tmode, target, rmode, 0);
- }
+ target = gen_reg_rtx (SImode);
+ emit_move_insn (target, const0_rtx);
+ target = gen_rtx_SUBREG (QImode, target, 0);
- for (i = 0; i < nargs; i++)
- {
- tree arg = CALL_EXPR_ARG (exp, i);
- rtx op = expand_normal (arg);
- enum machine_mode mode = insn_p->operand[i + 1].mode;
- bool match = (*insn_p->operand[i + 1].predicate) (op, mode);
+ if ((optimize && !register_operand (op0, mode0))
+ || !(*insn_data[d->icode].operand[0].predicate) (op0, mode0))
+ op0 = copy_to_mode_reg (mode0, op0);
+ if ((optimize && !register_operand (op1, mode1))
+ || !(*insn_data[d->icode].operand[1].predicate) (op1, mode1))
+ op1 = copy_to_mode_reg (mode1, op1);
- if (last_arg_count && (i + 1) == nargs)
- {
- /* SIMD shift insns take either an 8-bit immediate or
- register as count. But builtin functions take int as
- count. If count doesn't match, we put it in register. */
- if (!match)
- {
- op = simplify_gen_subreg (SImode, op, GET_MODE (op), 0);
- if (!(*insn_p->operand[i + 1].predicate) (op, mode))
- op = copy_to_reg (op);
- }
- }
- else if ((nargs - i) <= nargs_constant)
- {
- if (!match)
- switch (icode)
- {
- case CODE_FOR_sse4_1_roundpd:
- case CODE_FOR_sse4_1_roundps:
- case CODE_FOR_sse4_1_roundsd:
- case CODE_FOR_sse4_1_roundss:
- case CODE_FOR_sse4_1_blendps:
- error ("the last argument must be a 4-bit immediate");
- return const0_rtx;
+ pat = GEN_FCN (d->icode) (op0, op1);
+ if (! pat)
+ return 0;
+ emit_insn (pat);
+ emit_insn (gen_rtx_SET (VOIDmode,
+ gen_rtx_STRICT_LOW_PART (VOIDmode, target),
+ gen_rtx_fmt_ee (comparison, QImode,
+ SET_DEST (pat),
+ const0_rtx)));
- case CODE_FOR_sse4_1_blendpd:
- error ("the last argument must be a 2-bit immediate");
- return const0_rtx;
+ return SUBREG_REG (target);
+}
- default:
- switch (nargs_constant)
- {
- case 2:
- if ((nargs - i) == nargs_constant)
- {
- error ("the next to last argument must be an 8-bit immediate");
- break;
- }
- case 1:
- error ("the last argument must be an 8-bit immediate");
- break;
- default:
- gcc_unreachable ();
- }
- return const0_rtx;
- }
- }
- else
- {
- if (VECTOR_MODE_P (mode))
- op = safe_vector_operand (op, mode);
+/* Subroutine of ix86_expand_builtin to take care of ptest insns. */
- /* If we aren't optimizing, only allow one memory operand to
- be generated. */
- if (memory_operand (op, mode))
- num_memory++;
+static rtx
+ix86_expand_sse_ptest (const struct builtin_description *d, tree exp,
+ rtx target)
+{
+ rtx pat;
+ tree arg0 = CALL_EXPR_ARG (exp, 0);
+ tree arg1 = CALL_EXPR_ARG (exp, 1);
+ rtx op0 = expand_normal (arg0);
+ rtx op1 = expand_normal (arg1);
+ enum machine_mode mode0 = insn_data[d->icode].operand[0].mode;
+ enum machine_mode mode1 = insn_data[d->icode].operand[1].mode;
+ enum rtx_code comparison = d->comparison;
- if (GET_MODE (op) == mode || GET_MODE (op) == VOIDmode)
- {
- if (optimize || !match || num_memory > 1)
- op = copy_to_mode_reg (mode, op);
- }
- else
- {
- op = copy_to_reg (op);
- op = simplify_gen_subreg (mode, op, GET_MODE (op), 0);
- }
- }
+ if (VECTOR_MODE_P (mode0))
+ op0 = safe_vector_operand (op0, mode0);
+ if (VECTOR_MODE_P (mode1))
+ op1 = safe_vector_operand (op1, mode1);
- args[i].op = op;
- args[i].mode = mode;
- }
+ target = gen_reg_rtx (SImode);
+ emit_move_insn (target, const0_rtx);
+ target = gen_rtx_SUBREG (QImode, target, 0);
- switch (nargs)
- {
- case 1:
- pat = GEN_FCN (icode) (real_target, args[0].op);
- break;
- case 2:
- pat = GEN_FCN (icode) (real_target, args[0].op, args[1].op);
- break;
- case 3:
- pat = GEN_FCN (icode) (real_target, args[0].op, args[1].op,
- args[2].op);
- break;
- case 4:
- pat = GEN_FCN (icode) (real_target, args[0].op, args[1].op,
- args[2].op, args[3].op);
- break;
- default:
- gcc_unreachable ();
- }
+ if ((optimize && !register_operand (op0, mode0))
+ || !(*insn_data[d->icode].operand[0].predicate) (op0, mode0))
+ op0 = copy_to_mode_reg (mode0, op0);
+ if ((optimize && !register_operand (op1, mode1))
+ || !(*insn_data[d->icode].operand[1].predicate) (op1, mode1))
+ op1 = copy_to_mode_reg (mode1, op1);
+ pat = GEN_FCN (d->icode) (op0, op1);
if (! pat)
return 0;
-
emit_insn (pat);
- return target;
+ emit_insn (gen_rtx_SET (VOIDmode,
+ gen_rtx_STRICT_LOW_PART (VOIDmode, target),
+ gen_rtx_fmt_ee (comparison, QImode,
+ SET_DEST (pat),
+ const0_rtx)));
+
+ return SUBREG_REG (target);
}
-/* Subroutine of ix86_expand_builtin to take care of special insns
- with variable number of operands. */
-
-static rtx
-ix86_expand_special_args_builtin (const struct builtin_description *d,
- tree exp, rtx target)
-{
- tree arg;
- rtx pat, op;
- unsigned int i, nargs, arg_adjust, memory;
- struct
- {
- rtx op;
- enum machine_mode mode;
- } args[2];
- enum insn_code icode = d->icode;
- bool last_arg_constant = false;
- const struct insn_data *insn_p = &insn_data[icode];
- enum machine_mode tmode = insn_p->operand[0].mode;
- enum { load, store } class;
-
- switch ((enum ix86_special_builtin_type) d->flag)
- {
- case VOID_FTYPE_VOID:
- emit_insn (GEN_FCN (icode) (target));
- return 0;
- case V2DI_FTYPE_PV2DI:
- case V16QI_FTYPE_PCCHAR:
- case V4SF_FTYPE_PCFLOAT:
- case V2DF_FTYPE_PCDOUBLE:
- nargs = 1;
- class = load;
- memory = 0;
- break;
- case VOID_FTYPE_PV2SF_V4SF:
- case VOID_FTYPE_PV2DI_V2DI:
- case VOID_FTYPE_PCHAR_V16QI:
- case VOID_FTYPE_PFLOAT_V4SF:
- case VOID_FTYPE_PDOUBLE_V2DF:
- case VOID_FTYPE_PDI_DI:
- case VOID_FTYPE_PINT_INT:
- nargs = 1;
- class = store;
- /* Reserve memory operand for target. */
- memory = ARRAY_SIZE (args);
- break;
- case V4SF_FTYPE_V4SF_PCV2SF:
- case V2DF_FTYPE_V2DF_PCDOUBLE:
- nargs = 2;
- class = load;
- memory = 1;
- break;
- default:
- gcc_unreachable ();
- }
-
- gcc_assert (nargs <= ARRAY_SIZE (args));
-
- if (class == store)
- {
- arg = CALL_EXPR_ARG (exp, 0);
- op = expand_normal (arg);
- gcc_assert (target == 0);
- target = gen_rtx_MEM (tmode, copy_to_mode_reg (Pmode, op));
- arg_adjust = 1;
- }
- else
- {
- arg_adjust = 0;
- if (optimize
- || target == 0
- || GET_MODE (target) != tmode
- || ! (*insn_p->operand[0].predicate) (target, tmode))
- target = gen_reg_rtx (tmode);
- }
-
- for (i = 0; i < nargs; i++)
- {
- enum machine_mode mode = insn_p->operand[i + 1].mode;
- bool match;
-
- arg = CALL_EXPR_ARG (exp, i + arg_adjust);
- op = expand_normal (arg);
- match = (*insn_p->operand[i + 1].predicate) (op, mode);
-
- if (last_arg_constant && (i + 1) == nargs)
- {
- if (!match)
- switch (icode)
- {
- default:
- error ("the last argument must be an 8-bit immediate");
- return const0_rtx;
- }
- }
- else
- {
- if (i == memory)
- {
- /* This must be the memory operand. */
- op = gen_rtx_MEM (mode, copy_to_mode_reg (Pmode, op));
- gcc_assert (GET_MODE (op) == mode
- || GET_MODE (op) == VOIDmode);
- }
- else
- {
- /* This must be register. */
- if (VECTOR_MODE_P (mode))
- op = safe_vector_operand (op, mode);
-
- gcc_assert (GET_MODE (op) == mode
- || GET_MODE (op) == VOIDmode);
- op = copy_to_mode_reg (mode, op);
- }
- }
-
- args[i].op = op;
- args[i].mode = mode;
- }
-
- switch (nargs)
- {
- case 1:
- pat = GEN_FCN (icode) (target, args[0].op);
- break;
- case 2:
- pat = GEN_FCN (icode) (target, args[0].op, args[1].op);
- break;
- default:
- gcc_unreachable ();
- }
-
- if (! pat)
- return 0;
- emit_insn (pat);
- return class == store ? 0 : target;
-}
-
-/* Subroutine of ix86_expand_builtin to take care of comi insns. */
-
-static rtx
-ix86_expand_sse_comi (const struct builtin_description *d, tree exp,
- rtx target)
-{
- rtx pat;
- tree arg0 = CALL_EXPR_ARG (exp, 0);
- tree arg1 = CALL_EXPR_ARG (exp, 1);
- rtx op0 = expand_normal (arg0);
- rtx op1 = expand_normal (arg1);
- enum machine_mode mode0 = insn_data[d->icode].operand[0].mode;
- enum machine_mode mode1 = insn_data[d->icode].operand[1].mode;
- enum rtx_code comparison = d->comparison;
-
- if (VECTOR_MODE_P (mode0))
- op0 = safe_vector_operand (op0, mode0);
- if (VECTOR_MODE_P (mode1))
- op1 = safe_vector_operand (op1, mode1);
-
- /* Swap operands if we have a comparison that isn't available in
- hardware. */
- if (d->flag & BUILTIN_DESC_SWAP_OPERANDS)
- {
- rtx tmp = op1;
- op1 = op0;
- op0 = tmp;
- }
-
- target = gen_reg_rtx (SImode);
- emit_move_insn (target, const0_rtx);
- target = gen_rtx_SUBREG (QImode, target, 0);
-
- if ((optimize && !register_operand (op0, mode0))
- || !(*insn_data[d->icode].operand[0].predicate) (op0, mode0))
- op0 = copy_to_mode_reg (mode0, op0);
- if ((optimize && !register_operand (op1, mode1))
- || !(*insn_data[d->icode].operand[1].predicate) (op1, mode1))
- op1 = copy_to_mode_reg (mode1, op1);
-
- pat = GEN_FCN (d->icode) (op0, op1);
- if (! pat)
- return 0;
- emit_insn (pat);
- emit_insn (gen_rtx_SET (VOIDmode,
- gen_rtx_STRICT_LOW_PART (VOIDmode, target),
- gen_rtx_fmt_ee (comparison, QImode,
- SET_DEST (pat),
- const0_rtx)));
-
- return SUBREG_REG (target);
-}
-
-/* Subroutine of ix86_expand_builtin to take care of ptest insns. */
-
-static rtx
-ix86_expand_sse_ptest (const struct builtin_description *d, tree exp,
- rtx target)
-{
- rtx pat;
- tree arg0 = CALL_EXPR_ARG (exp, 0);
- tree arg1 = CALL_EXPR_ARG (exp, 1);
- rtx op0 = expand_normal (arg0);
- rtx op1 = expand_normal (arg1);
- enum machine_mode mode0 = insn_data[d->icode].operand[0].mode;
- enum machine_mode mode1 = insn_data[d->icode].operand[1].mode;
- enum rtx_code comparison = d->comparison;
-
- if (VECTOR_MODE_P (mode0))
- op0 = safe_vector_operand (op0, mode0);
- if (VECTOR_MODE_P (mode1))
- op1 = safe_vector_operand (op1, mode1);
-
- target = gen_reg_rtx (SImode);
- emit_move_insn (target, const0_rtx);
- target = gen_rtx_SUBREG (QImode, target, 0);
-
- if ((optimize && !register_operand (op0, mode0))
- || !(*insn_data[d->icode].operand[0].predicate) (op0, mode0))
- op0 = copy_to_mode_reg (mode0, op0);
- if ((optimize && !register_operand (op1, mode1))
- || !(*insn_data[d->icode].operand[1].predicate) (op1, mode1))
- op1 = copy_to_mode_reg (mode1, op1);
-
- pat = GEN_FCN (d->icode) (op0, op1);
- if (! pat)
- return 0;
- emit_insn (pat);
- emit_insn (gen_rtx_SET (VOIDmode,
- gen_rtx_STRICT_LOW_PART (VOIDmode, target),
- gen_rtx_fmt_ee (comparison, QImode,
- SET_DEST (pat),
- const0_rtx)));
-
- return SUBREG_REG (target);
-}
-
-/* Subroutine of ix86_expand_builtin to take care of pcmpestr[im] insns. */
+/* Subroutine of ix86_expand_builtin to take care of pcmpestr[im] insns. */
static rtx
ix86_expand_sse_pcmpestr (const struct builtin_description *d,
if (! (*insn_data[d->icode].operand[5].predicate) (op3, modei5))
op3 = copy_to_mode_reg (modei5, op3);
- if (! (*insn_data[d->icode].operand[6].predicate) (op4, modeimm))
+ if (! (*insn_data[d->icode].operand[6].predicate) (op4, modeimm))
+ {
+ error ("the fifth argument must be a 8-bit immediate");
+ return const0_rtx;
+ }
+
+ if (d->code == IX86_BUILTIN_PCMPESTRI128)
+ {
+ if (optimize || !target
+ || GET_MODE (target) != tmode0
+ || ! (*insn_data[d->icode].operand[0].predicate) (target, tmode0))
+ target = gen_reg_rtx (tmode0);
+
+ scratch1 = gen_reg_rtx (tmode1);
+
+ pat = GEN_FCN (d->icode) (target, scratch1, op0, op1, op2, op3, op4);
+ }
+ else if (d->code == IX86_BUILTIN_PCMPESTRM128)
+ {
+ if (optimize || !target
+ || GET_MODE (target) != tmode1
+ || ! (*insn_data[d->icode].operand[1].predicate) (target, tmode1))
+ target = gen_reg_rtx (tmode1);
+
+ scratch0 = gen_reg_rtx (tmode0);
+
+ pat = GEN_FCN (d->icode) (scratch0, target, op0, op1, op2, op3, op4);
+ }
+ else
+ {
+ gcc_assert (d->flag);
+
+ scratch0 = gen_reg_rtx (tmode0);
+ scratch1 = gen_reg_rtx (tmode1);
+
+ pat = GEN_FCN (d->icode) (scratch0, scratch1, op0, op1, op2, op3, op4);
+ }
+
+ if (! pat)
+ return 0;
+
+ emit_insn (pat);
+
+ if (d->flag)
+ {
+ target = gen_reg_rtx (SImode);
+ emit_move_insn (target, const0_rtx);
+ target = gen_rtx_SUBREG (QImode, target, 0);
+
+ emit_insn
+ (gen_rtx_SET (VOIDmode, gen_rtx_STRICT_LOW_PART (VOIDmode, target),
+ gen_rtx_fmt_ee (EQ, QImode,
+ gen_rtx_REG ((enum machine_mode) d->flag,
+ FLAGS_REG),
+ const0_rtx)));
+ return SUBREG_REG (target);
+ }
+ else
+ return target;
+}
+
+
+/* Subroutine of ix86_expand_builtin to take care of pcmpistr[im] insns. */
+
+static rtx
+ix86_expand_sse_pcmpistr (const struct builtin_description *d,
+ tree exp, rtx target)
+{
+ rtx pat;
+ tree arg0 = CALL_EXPR_ARG (exp, 0);
+ tree arg1 = CALL_EXPR_ARG (exp, 1);
+ tree arg2 = CALL_EXPR_ARG (exp, 2);
+ rtx scratch0, scratch1;
+ rtx op0 = expand_normal (arg0);
+ rtx op1 = expand_normal (arg1);
+ rtx op2 = expand_normal (arg2);
+ enum machine_mode tmode0, tmode1, modev2, modev3, modeimm;
+
+ tmode0 = insn_data[d->icode].operand[0].mode;
+ tmode1 = insn_data[d->icode].operand[1].mode;
+ modev2 = insn_data[d->icode].operand[2].mode;
+ modev3 = insn_data[d->icode].operand[3].mode;
+ modeimm = insn_data[d->icode].operand[4].mode;
+
+ if (VECTOR_MODE_P (modev2))
+ op0 = safe_vector_operand (op0, modev2);
+ if (VECTOR_MODE_P (modev3))
+ op1 = safe_vector_operand (op1, modev3);
+
+ if (! (*insn_data[d->icode].operand[2].predicate) (op0, modev2))
+ op0 = copy_to_mode_reg (modev2, op0);
+ if ((optimize && !register_operand (op1, modev3))
+ || !(*insn_data[d->icode].operand[3].predicate) (op1, modev3))
+ op1 = copy_to_mode_reg (modev3, op1);
+
+ if (! (*insn_data[d->icode].operand[4].predicate) (op2, modeimm))
+ {
+ error ("the third argument must be a 8-bit immediate");
+ return const0_rtx;
+ }
+
+ if (d->code == IX86_BUILTIN_PCMPISTRI128)
+ {
+ if (optimize || !target
+ || GET_MODE (target) != tmode0
+ || ! (*insn_data[d->icode].operand[0].predicate) (target, tmode0))
+ target = gen_reg_rtx (tmode0);
+
+ scratch1 = gen_reg_rtx (tmode1);
+
+ pat = GEN_FCN (d->icode) (target, scratch1, op0, op1, op2);
+ }
+ else if (d->code == IX86_BUILTIN_PCMPISTRM128)
+ {
+ if (optimize || !target
+ || GET_MODE (target) != tmode1
+ || ! (*insn_data[d->icode].operand[1].predicate) (target, tmode1))
+ target = gen_reg_rtx (tmode1);
+
+ scratch0 = gen_reg_rtx (tmode0);
+
+ pat = GEN_FCN (d->icode) (scratch0, target, op0, op1, op2);
+ }
+ else
+ {
+ gcc_assert (d->flag);
+
+ scratch0 = gen_reg_rtx (tmode0);
+ scratch1 = gen_reg_rtx (tmode1);
+
+ pat = GEN_FCN (d->icode) (scratch0, scratch1, op0, op1, op2);
+ }
+
+ if (! pat)
+ return 0;
+
+ emit_insn (pat);
+
+ if (d->flag)
+ {
+ target = gen_reg_rtx (SImode);
+ emit_move_insn (target, const0_rtx);
+ target = gen_rtx_SUBREG (QImode, target, 0);
+
+ emit_insn
+ (gen_rtx_SET (VOIDmode, gen_rtx_STRICT_LOW_PART (VOIDmode, target),
+ gen_rtx_fmt_ee (EQ, QImode,
+ gen_rtx_REG ((enum machine_mode) d->flag,
+ FLAGS_REG),
+ const0_rtx)));
+ return SUBREG_REG (target);
+ }
+ else
+ return target;
+}
+
+/* Subroutine of ix86_expand_builtin to take care of insns with
+ variable number of operands. */
+
+static rtx
+ix86_expand_args_builtin (const struct builtin_description *d,
+ tree exp, rtx target)
+{
+ rtx pat, real_target;
+ unsigned int i, nargs;
+ unsigned int nargs_constant = 0;
+ int num_memory = 0;
+ struct
+ {
+ rtx op;
+ enum machine_mode mode;
+ } args[4];
+ bool last_arg_count = false;
+ enum insn_code icode = d->icode;
+ const struct insn_data *insn_p = &insn_data[icode];
+ enum machine_mode tmode = insn_p->operand[0].mode;
+ enum machine_mode rmode = VOIDmode;
+ bool swap = false;
+ enum rtx_code comparison = d->comparison;
+
+ switch ((enum ix86_builtin_type) d->flag)
+ {
+ case INT_FTYPE_V2DI_V2DI_PTEST:
+ return ix86_expand_sse_ptest (d, exp, target);
+ case FLOAT128_FTYPE_FLOAT128:
+ case FLOAT_FTYPE_FLOAT:
+ case INT64_FTYPE_V4SF:
+ case INT64_FTYPE_V2DF:
+ case INT_FTYPE_V16QI:
+ case INT_FTYPE_V8QI:
+ case INT_FTYPE_V4SF:
+ case INT_FTYPE_V2DF:
+ case V16QI_FTYPE_V16QI:
+ case V8HI_FTYPE_V8HI:
+ case V8HI_FTYPE_V16QI:
+ case V8QI_FTYPE_V8QI:
+ case V4SI_FTYPE_V4SI:
+ case V4SI_FTYPE_V16QI:
+ case V4SI_FTYPE_V4SF:
+ case V4SI_FTYPE_V8HI:
+ case V4SI_FTYPE_V2DF:
+ case V4HI_FTYPE_V4HI:
+ case V4SF_FTYPE_V4SF:
+ case V4SF_FTYPE_V4SI:
+ case V4SF_FTYPE_V2DF:
+ case V2DI_FTYPE_V2DI:
+ case V2DI_FTYPE_V16QI:
+ case V2DI_FTYPE_V8HI:
+ case V2DI_FTYPE_V4SI:
+ case V2DF_FTYPE_V2DF:
+ case V2DF_FTYPE_V4SI:
+ case V2DF_FTYPE_V4SF:
+ case V2DF_FTYPE_V2SI:
+ case V2SI_FTYPE_V2SI:
+ case V2SI_FTYPE_V4SF:
+ case V2SI_FTYPE_V2SF:
+ case V2SI_FTYPE_V2DF:
+ case V2SF_FTYPE_V2SF:
+ case V2SF_FTYPE_V2SI:
+ nargs = 1;
+ break;
+ case V4SF_FTYPE_V4SF_VEC_MERGE:
+ case V2DF_FTYPE_V2DF_VEC_MERGE:
+ return ix86_expand_unop_vec_merge_builtin (icode, exp, target);
+ case FLOAT128_FTYPE_FLOAT128_FLOAT128:
+ case V16QI_FTYPE_V16QI_V16QI:
+ case V16QI_FTYPE_V8HI_V8HI:
+ case V8QI_FTYPE_V8QI_V8QI:
+ case V8QI_FTYPE_V4HI_V4HI:
+ case V8HI_FTYPE_V8HI_V8HI:
+ case V8HI_FTYPE_V16QI_V16QI:
+ case V8HI_FTYPE_V4SI_V4SI:
+ case V4SI_FTYPE_V4SI_V4SI:
+ case V4SI_FTYPE_V8HI_V8HI:
+ case V4SI_FTYPE_V4SF_V4SF:
+ case V4SI_FTYPE_V2DF_V2DF:
+ case V4HI_FTYPE_V4HI_V4HI:
+ case V4HI_FTYPE_V8QI_V8QI:
+ case V4HI_FTYPE_V2SI_V2SI:
+ case V4SF_FTYPE_V4SF_V4SF:
+ case V4SF_FTYPE_V4SF_V2SI:
+ case V4SF_FTYPE_V4SF_V2DF:
+ case V4SF_FTYPE_V4SF_DI:
+ case V4SF_FTYPE_V4SF_SI:
+ case V2DI_FTYPE_V2DI_V2DI:
+ case V2DI_FTYPE_V16QI_V16QI:
+ case V2DI_FTYPE_V4SI_V4SI:
+ case V2DI_FTYPE_V2DI_V16QI:
+ case V2DI_FTYPE_V2DF_V2DF:
+ case V2SI_FTYPE_V2SI_V2SI:
+ case V2SI_FTYPE_V4HI_V4HI:
+ case V2SI_FTYPE_V2SF_V2SF:
+ case V2DF_FTYPE_V2DF_V2DF:
+ case V2DF_FTYPE_V2DF_V4SF:
+ case V2DF_FTYPE_V2DF_DI:
+ case V2DF_FTYPE_V2DF_SI:
+ case V2SF_FTYPE_V2SF_V2SF:
+ case V1DI_FTYPE_V1DI_V1DI:
+ case V1DI_FTYPE_V8QI_V8QI:
+ case V1DI_FTYPE_V2SI_V2SI:
+ if (comparison == UNKNOWN)
+ return ix86_expand_binop_builtin (icode, exp, target);
+ nargs = 2;
+ break;
+ case V4SF_FTYPE_V4SF_V4SF_SWAP:
+ case V2DF_FTYPE_V2DF_V2DF_SWAP:
+ gcc_assert (comparison != UNKNOWN);
+ nargs = 2;
+ swap = true;
+ break;
+ case V8HI_FTYPE_V8HI_V8HI_COUNT:
+ case V8HI_FTYPE_V8HI_SI_COUNT:
+ case V4SI_FTYPE_V4SI_V4SI_COUNT:
+ case V4SI_FTYPE_V4SI_SI_COUNT:
+ case V4HI_FTYPE_V4HI_V4HI_COUNT:
+ case V4HI_FTYPE_V4HI_SI_COUNT:
+ case V2DI_FTYPE_V2DI_V2DI_COUNT:
+ case V2DI_FTYPE_V2DI_SI_COUNT:
+ case V2SI_FTYPE_V2SI_V2SI_COUNT:
+ case V2SI_FTYPE_V2SI_SI_COUNT:
+ case V1DI_FTYPE_V1DI_V1DI_COUNT:
+ case V1DI_FTYPE_V1DI_SI_COUNT:
+ nargs = 2;
+ last_arg_count = true;
+ break;
+ case UINT64_FTYPE_UINT64_UINT64:
+ case UINT_FTYPE_UINT_UINT:
+ case UINT_FTYPE_UINT_USHORT:
+ case UINT_FTYPE_UINT_UCHAR:
+ nargs = 2;
+ break;
+ case V2DI2TI_FTYPE_V2DI_INT:
+ nargs = 2;
+ rmode = V2DImode;
+ nargs_constant = 1;
+ break;
+ case V8HI_FTYPE_V8HI_INT:
+ case V4SI_FTYPE_V4SI_INT:
+ case V4HI_FTYPE_V4HI_INT:
+ case V4SF_FTYPE_V4SF_INT:
+ case V2DI_FTYPE_V2DI_INT:
+ case V2DF_FTYPE_V2DF_INT:
+ nargs = 2;
+ nargs_constant = 1;
+ break;
+ case V16QI_FTYPE_V16QI_V16QI_V16QI:
+ case V4SF_FTYPE_V4SF_V4SF_V4SF:
+ case V2DF_FTYPE_V2DF_V2DF_V2DF:
+ nargs = 3;
+ break;
+ case V16QI_FTYPE_V16QI_V16QI_INT:
+ case V8HI_FTYPE_V8HI_V8HI_INT:
+ case V4SI_FTYPE_V4SI_V4SI_INT:
+ case V4SF_FTYPE_V4SF_V4SF_INT:
+ case V2DI_FTYPE_V2DI_V2DI_INT:
+ case V2DF_FTYPE_V2DF_V2DF_INT:
+ nargs = 3;
+ nargs_constant = 1;
+ break;
+ case V2DI2TI_FTYPE_V2DI_V2DI_INT:
+ nargs = 3;
+ rmode = V2DImode;
+ nargs_constant = 1;
+ break;
+ case V1DI2DI_FTYPE_V1DI_V1DI_INT:
+ nargs = 3;
+ rmode = DImode;
+ nargs_constant = 1;
+ break;
+ case V2DI_FTYPE_V2DI_UINT_UINT:
+ nargs = 3;
+ nargs_constant = 2;
+ break;
+ case V2DI_FTYPE_V2DI_V2DI_UINT_UINT:
+ nargs = 4;
+ nargs_constant = 2;
+ break;
+ default:
+ gcc_unreachable ();
+ }
+
+ gcc_assert (nargs <= ARRAY_SIZE (args));
+
+ if (comparison != UNKNOWN)
{
- error ("the fifth argument must be a 8-bit immediate");
- return const0_rtx;
+ gcc_assert (nargs == 2);
+ return ix86_expand_sse_compare (d, exp, target, swap);
}
- if (d->code == IX86_BUILTIN_PCMPESTRI128)
+ if (rmode == VOIDmode || rmode == tmode)
{
- if (optimize || !target
- || GET_MODE (target) != tmode0
- || ! (*insn_data[d->icode].operand[0].predicate) (target, tmode0))
- target = gen_reg_rtx (tmode0);
-
- scratch1 = gen_reg_rtx (tmode1);
-
- pat = GEN_FCN (d->icode) (target, scratch1, op0, op1, op2, op3, op4);
+ if (optimize
+ || target == 0
+ || GET_MODE (target) != tmode
+ || ! (*insn_p->operand[0].predicate) (target, tmode))
+ target = gen_reg_rtx (tmode);
+ real_target = target;
}
- else if (d->code == IX86_BUILTIN_PCMPESTRM128)
+ else
{
- if (optimize || !target
- || GET_MODE (target) != tmode1
- || ! (*insn_data[d->icode].operand[1].predicate) (target, tmode1))
- target = gen_reg_rtx (tmode1);
-
- scratch0 = gen_reg_rtx (tmode0);
-
- pat = GEN_FCN (d->icode) (scratch0, target, op0, op1, op2, op3, op4);
+ target = gen_reg_rtx (rmode);
+ real_target = simplify_gen_subreg (tmode, target, rmode, 0);
}
- else
+
+ for (i = 0; i < nargs; i++)
{
- gcc_assert (d->flag);
+ tree arg = CALL_EXPR_ARG (exp, i);
+ rtx op = expand_normal (arg);
+ enum machine_mode mode = insn_p->operand[i + 1].mode;
+ bool match = (*insn_p->operand[i + 1].predicate) (op, mode);
- scratch0 = gen_reg_rtx (tmode0);
- scratch1 = gen_reg_rtx (tmode1);
+ if (last_arg_count && (i + 1) == nargs)
+ {
+ /* SIMD shift insns take either an 8-bit immediate or
+ register as count. But builtin functions take int as
+ count. If count doesn't match, we put it in register. */
+ if (!match)
+ {
+ op = simplify_gen_subreg (SImode, op, GET_MODE (op), 0);
+ if (!(*insn_p->operand[i + 1].predicate) (op, mode))
+ op = copy_to_reg (op);
+ }
+ }
+ else if ((nargs - i) <= nargs_constant)
+ {
+ if (!match)
+ switch (icode)
+ {
+ case CODE_FOR_sse4_1_roundpd:
+ case CODE_FOR_sse4_1_roundps:
+ case CODE_FOR_sse4_1_roundsd:
+ case CODE_FOR_sse4_1_roundss:
+ case CODE_FOR_sse4_1_blendps:
+ error ("the last argument must be a 4-bit immediate");
+ return const0_rtx;
- pat = GEN_FCN (d->icode) (scratch0, scratch1, op0, op1, op2, op3, op4);
+ case CODE_FOR_sse4_1_blendpd:
+ error ("the last argument must be a 2-bit immediate");
+ return const0_rtx;
+
+ default:
+ switch (nargs_constant)
+ {
+ case 2:
+ if ((nargs - i) == nargs_constant)
+ {
+ error ("the next to last argument must be an 8-bit immediate");
+ break;
+ }
+ case 1:
+ error ("the last argument must be an 8-bit immediate");
+ break;
+ default:
+ gcc_unreachable ();
+ }
+ return const0_rtx;
+ }
+ }
+ else
+ {
+ if (VECTOR_MODE_P (mode))
+ op = safe_vector_operand (op, mode);
+
+ /* If we aren't optimizing, only allow one memory operand to
+ be generated. */
+ if (memory_operand (op, mode))
+ num_memory++;
+
+ if (GET_MODE (op) == mode || GET_MODE (op) == VOIDmode)
+ {
+ if (optimize || !match || num_memory > 1)
+ op = copy_to_mode_reg (mode, op);
+ }
+ else
+ {
+ op = copy_to_reg (op);
+ op = simplify_gen_subreg (mode, op, GET_MODE (op), 0);
+ }
+ }
+
+ args[i].op = op;
+ args[i].mode = mode;
+ }
+
+ switch (nargs)
+ {
+ case 1:
+ pat = GEN_FCN (icode) (real_target, args[0].op);
+ break;
+ case 2:
+ pat = GEN_FCN (icode) (real_target, args[0].op, args[1].op);
+ break;
+ case 3:
+ pat = GEN_FCN (icode) (real_target, args[0].op, args[1].op,
+ args[2].op);
+ break;
+ case 4:
+ pat = GEN_FCN (icode) (real_target, args[0].op, args[1].op,
+ args[2].op, args[3].op);
+ break;
+ default:
+ gcc_unreachable ();
}
if (! pat)
return 0;
emit_insn (pat);
-
- if (d->flag)
- {
- target = gen_reg_rtx (SImode);
- emit_move_insn (target, const0_rtx);
- target = gen_rtx_SUBREG (QImode, target, 0);
-
- emit_insn
- (gen_rtx_SET (VOIDmode, gen_rtx_STRICT_LOW_PART (VOIDmode, target),
- gen_rtx_fmt_ee (EQ, QImode,
- gen_rtx_REG ((enum machine_mode) d->flag,
- FLAGS_REG),
- const0_rtx)));
- return SUBREG_REG (target);
- }
- else
- return target;
+ return target;
}
-
-/* Subroutine of ix86_expand_builtin to take care of pcmpistr[im] insns. */
+/* Subroutine of ix86_expand_builtin to take care of special insns
+ with variable number of operands. */
static rtx
-ix86_expand_sse_pcmpistr (const struct builtin_description *d,
- tree exp, rtx target)
+ix86_expand_special_args_builtin (const struct builtin_description *d,
+ tree exp, rtx target)
{
- rtx pat;
- tree arg0 = CALL_EXPR_ARG (exp, 0);
- tree arg1 = CALL_EXPR_ARG (exp, 1);
- tree arg2 = CALL_EXPR_ARG (exp, 2);
- rtx scratch0, scratch1;
- rtx op0 = expand_normal (arg0);
- rtx op1 = expand_normal (arg1);
- rtx op2 = expand_normal (arg2);
- enum machine_mode tmode0, tmode1, modev2, modev3, modeimm;
-
- tmode0 = insn_data[d->icode].operand[0].mode;
- tmode1 = insn_data[d->icode].operand[1].mode;
- modev2 = insn_data[d->icode].operand[2].mode;
- modev3 = insn_data[d->icode].operand[3].mode;
- modeimm = insn_data[d->icode].operand[4].mode;
+ tree arg;
+ rtx pat, op;
+ unsigned int i, nargs, arg_adjust, memory;
+ struct
+ {
+ rtx op;
+ enum machine_mode mode;
+ } args[2];
+ enum insn_code icode = d->icode;
+ bool last_arg_constant = false;
+ const struct insn_data *insn_p = &insn_data[icode];
+ enum machine_mode tmode = insn_p->operand[0].mode;
+ enum { load, store } class;
- if (VECTOR_MODE_P (modev2))
- op0 = safe_vector_operand (op0, modev2);
- if (VECTOR_MODE_P (modev3))
- op1 = safe_vector_operand (op1, modev3);
+ switch ((enum ix86_special_builtin_type) d->flag)
+ {
+ case VOID_FTYPE_VOID:
+ emit_insn (GEN_FCN (icode) (target));
+ return 0;
+ case V2DI_FTYPE_PV2DI:
+ case V16QI_FTYPE_PCCHAR:
+ case V4SF_FTYPE_PCFLOAT:
+ case V2DF_FTYPE_PCDOUBLE:
+ nargs = 1;
+ class = load;
+ memory = 0;
+ break;
+ case VOID_FTYPE_PV2SF_V4SF:
+ case VOID_FTYPE_PV2DI_V2DI:
+ case VOID_FTYPE_PCHAR_V16QI:
+ case VOID_FTYPE_PFLOAT_V4SF:
+ case VOID_FTYPE_PDOUBLE_V2DF:
+ case VOID_FTYPE_PDI_DI:
+ case VOID_FTYPE_PINT_INT:
+ nargs = 1;
+ class = store;
+ /* Reserve memory operand for target. */
+ memory = ARRAY_SIZE (args);
+ break;
+ case V4SF_FTYPE_V4SF_PCV2SF:
+ case V2DF_FTYPE_V2DF_PCDOUBLE:
+ nargs = 2;
+ class = load;
+ memory = 1;
+ break;
+ default:
+ gcc_unreachable ();
+ }
- if (! (*insn_data[d->icode].operand[2].predicate) (op0, modev2))
- op0 = copy_to_mode_reg (modev2, op0);
- if ((optimize && !register_operand (op1, modev3))
- || !(*insn_data[d->icode].operand[3].predicate) (op1, modev3))
- op1 = copy_to_mode_reg (modev3, op1);
+ gcc_assert (nargs <= ARRAY_SIZE (args));
- if (! (*insn_data[d->icode].operand[4].predicate) (op2, modeimm))
+ if (class == store)
{
- error ("the third argument must be a 8-bit immediate");
- return const0_rtx;
+ arg = CALL_EXPR_ARG (exp, 0);
+ op = expand_normal (arg);
+ gcc_assert (target == 0);
+ target = gen_rtx_MEM (tmode, copy_to_mode_reg (Pmode, op));
+ arg_adjust = 1;
+ }
+ else
+ {
+ arg_adjust = 0;
+ if (optimize
+ || target == 0
+ || GET_MODE (target) != tmode
+ || ! (*insn_p->operand[0].predicate) (target, tmode))
+ target = gen_reg_rtx (tmode);
}
- if (d->code == IX86_BUILTIN_PCMPISTRI128)
+ for (i = 0; i < nargs; i++)
{
- if (optimize || !target
- || GET_MODE (target) != tmode0
- || ! (*insn_data[d->icode].operand[0].predicate) (target, tmode0))
- target = gen_reg_rtx (tmode0);
+ enum machine_mode mode = insn_p->operand[i + 1].mode;
+ bool match;
- scratch1 = gen_reg_rtx (tmode1);
+ arg = CALL_EXPR_ARG (exp, i + arg_adjust);
+ op = expand_normal (arg);
+ match = (*insn_p->operand[i + 1].predicate) (op, mode);
- pat = GEN_FCN (d->icode) (target, scratch1, op0, op1, op2);
- }
- else if (d->code == IX86_BUILTIN_PCMPISTRM128)
- {
- if (optimize || !target
- || GET_MODE (target) != tmode1
- || ! (*insn_data[d->icode].operand[1].predicate) (target, tmode1))
- target = gen_reg_rtx (tmode1);
+ if (last_arg_constant && (i + 1) == nargs)
+ {
+ if (!match)
+ switch (icode)
+ {
+ default:
+ error ("the last argument must be an 8-bit immediate");
+ return const0_rtx;
+ }
+ }
+ else
+ {
+ if (i == memory)
+ {
+ /* This must be the memory operand. */
+ op = gen_rtx_MEM (mode, copy_to_mode_reg (Pmode, op));
+ gcc_assert (GET_MODE (op) == mode
+ || GET_MODE (op) == VOIDmode);
+ }
+ else
+ {
+ /* This must be register. */
+ if (VECTOR_MODE_P (mode))
+ op = safe_vector_operand (op, mode);
- scratch0 = gen_reg_rtx (tmode0);
+ gcc_assert (GET_MODE (op) == mode
+ || GET_MODE (op) == VOIDmode);
+ op = copy_to_mode_reg (mode, op);
+ }
+ }
- pat = GEN_FCN (d->icode) (scratch0, target, op0, op1, op2);
+ args[i].op = op;
+ args[i].mode = mode;
}
- else
- {
- gcc_assert (d->flag);
-
- scratch0 = gen_reg_rtx (tmode0);
- scratch1 = gen_reg_rtx (tmode1);
- pat = GEN_FCN (d->icode) (scratch0, scratch1, op0, op1, op2);
+ switch (nargs)
+ {
+ case 1:
+ pat = GEN_FCN (icode) (target, args[0].op);
+ break;
+ case 2:
+ pat = GEN_FCN (icode) (target, args[0].op, args[1].op);
+ break;
+ default:
+ gcc_unreachable ();
}
if (! pat)
return 0;
-
emit_insn (pat);
-
- if (d->flag)
- {
- target = gen_reg_rtx (SImode);
- emit_move_insn (target, const0_rtx);
- target = gen_rtx_SUBREG (QImode, target, 0);
-
- emit_insn
- (gen_rtx_SET (VOIDmode, gen_rtx_STRICT_LOW_PART (VOIDmode, target),
- gen_rtx_fmt_ee (EQ, QImode,
- gen_rtx_REG ((enum machine_mode) d->flag,
- FLAGS_REG),
- const0_rtx)));
- return SUBREG_REG (target);
- }
- else
- return target;
+ return class == store ? 0 : target;
}
/* Return the integer constant in ARG. Constrain it to be in the range
if (d->code == fcode)
return ix86_expand_sse_comi (d, exp, target);
- for (i = 0, d = bdesc_ptest; i < ARRAY_SIZE (bdesc_ptest); i++, d++)
- if (d->code == fcode)
- return ix86_expand_sse_ptest (d, exp, target);
-
for (i = 0, d = bdesc_pcmpestr;
i < ARRAY_SIZE (bdesc_pcmpestr);
i++, d++)
return regclass;
}
+static enum reg_class
+ix86_secondary_reload (bool in_p, rtx x, enum reg_class class,
+ enum machine_mode mode,
+ secondary_reload_info *sri ATTRIBUTE_UNUSED)
+{
+ /* QImode spills from non-QI registers require
+ intermediate register on 32bit targets. */
+ if (!in_p && mode == QImode && !TARGET_64BIT
+ && (class == GENERAL_REGS
+ || class == LEGACY_REGS
+ || class == INDEX_REGS))
+ {
+ int regno;
+
+ if (REG_P (x))
+ regno = REGNO (x);
+ else
+ regno = -1;
+
+ if (regno >= FIRST_PSEUDO_REGISTER || GET_CODE (x) == SUBREG)
+ regno = true_regnum (x);
+
+ /* Return Q_REGS if the operand is in memory. */
+ if (regno == -1)
+ return Q_REGS;
+ }
+
+ return NO_REGS;
+}
+
/* If we are copying between general and FP registers, we need a memory
location. The same is true for SSE and MMX registers.
if (TARGET_64BIT || TARGET_ALIGN_DOUBLE)
return computed;
- mode = TYPE_MODE (TREE_CODE (type) == ARRAY_TYPE
- ? get_inner_array_type (type) : type);
+ mode = TYPE_MODE (strip_array_types (type));
if (mode == DFmode || mode == DCmode
|| GET_MODE_CLASS (mode) == MODE_INT
|| GET_MODE_CLASS (mode) == MODE_COMPLEX_INT)
enum machine_mode vsimode;
rtx new_target;
rtx x, tmp;
+ bool use_vector_set = false;
+
+ switch (mode)
+ {
+ case V2DImode:
+ use_vector_set = TARGET_64BIT && TARGET_SSE4_1;
+ break;
+ case V16QImode:
+ case V4SImode:
+ case V4SFmode:
+ use_vector_set = TARGET_SSE4_1;
+ break;
+ case V8HImode:
+ use_vector_set = TARGET_SSE2;
+ break;
+ case V4HImode:
+ use_vector_set = TARGET_SSE || TARGET_3DNOW_A;
+ break;
+ default:
+ break;
+ }
+
+ if (use_vector_set)
+ {
+ emit_insn (gen_rtx_SET (VOIDmode, target, CONST0_RTX (mode)));
+ var = force_reg (GET_MODE_INNER (mode), var);
+ ix86_expand_vector_set (mmx_ok, target, var, one_var);
+ return true;
+ }
switch (mode)
{
break;
case V16QImode:
+ if (TARGET_SSE4_1)
+ break;
wmode = V8HImode;
goto widen;
case V8QImode:
return true;
}
+/* A subroutine of ix86_expand_vector_init_general. Use vector
+ concatenate to handle the most general case: all values variable,
+ and none identical. */
+
+static void
+ix86_expand_vector_init_concat (enum machine_mode mode,
+ rtx target, rtx *ops, int n)
+{
+ enum machine_mode cmode, hmode = VOIDmode;
+ rtx first[4], second[2];
+ rtvec v;
+ int i, j;
+
+ switch (n)
+ {
+ case 2:
+ switch (mode)
+ {
+ case V4SImode:
+ cmode = V2SImode;
+ break;
+ case V4SFmode:
+ cmode = V2SFmode;
+ break;
+ case V2DImode:
+ cmode = DImode;
+ break;
+ case V2SImode:
+ cmode = SImode;
+ break;
+ case V2DFmode:
+ cmode = DFmode;
+ break;
+ case V2SFmode:
+ cmode = SFmode;
+ break;
+ default:
+ gcc_unreachable ();
+ }
+
+ if (!register_operand (ops[1], cmode))
+ ops[1] = force_reg (cmode, ops[1]);
+ if (!register_operand (ops[0], cmode))
+ ops[0] = force_reg (cmode, ops[0]);
+ emit_insn (gen_rtx_SET (VOIDmode, target,
+ gen_rtx_VEC_CONCAT (mode, ops[0],
+ ops[1])));
+ break;
+
+ case 4:
+ switch (mode)
+ {
+ case V4SImode:
+ cmode = V2SImode;
+ break;
+ case V4SFmode:
+ cmode = V2SFmode;
+ break;
+ default:
+ gcc_unreachable ();
+ }
+ goto half;
+
+half:
+ /* FIXME: We process inputs backward to help RA. PR 36222. */
+ i = n - 1;
+ j = (n >> 1) - 1;
+ for (; i > 0; i -= 2, j--)
+ {
+ first[j] = gen_reg_rtx (cmode);
+ v = gen_rtvec (2, ops[i - 1], ops[i]);
+ ix86_expand_vector_init (false, first[j],
+ gen_rtx_PARALLEL (cmode, v));
+ }
+
+ n >>= 1;
+ if (n > 2)
+ {
+ gcc_assert (hmode != VOIDmode);
+ for (i = j = 0; i < n; i += 2, j++)
+ {
+ second[j] = gen_reg_rtx (hmode);
+ ix86_expand_vector_init_concat (hmode, second [j],
+ &first [i], 2);
+ }
+ n >>= 1;
+ ix86_expand_vector_init_concat (mode, target, second, n);
+ }
+ else
+ ix86_expand_vector_init_concat (mode, target, first, n);
+ break;
+
+ default:
+ gcc_unreachable ();
+ }
+}
+
+/* A subroutine of ix86_expand_vector_init_general. Use vector
+ interleave to handle the most general case: all values variable,
+ and none identical. */
+
+static void
+ix86_expand_vector_init_interleave (enum machine_mode mode,
+ rtx target, rtx *ops, int n)
+{
+ enum machine_mode first_imode, second_imode, third_imode;
+ int i, j;
+ rtx op0, op1;
+ rtx (*gen_load_even) (rtx, rtx, rtx);
+ rtx (*gen_interleave_first_low) (rtx, rtx, rtx);
+ rtx (*gen_interleave_second_low) (rtx, rtx, rtx);
+
+ switch (mode)
+ {
+ case V8HImode:
+ gen_load_even = gen_vec_setv8hi;
+ gen_interleave_first_low = gen_vec_interleave_lowv4si;
+ gen_interleave_second_low = gen_vec_interleave_lowv2di;
+ first_imode = V4SImode;
+ second_imode = V2DImode;
+ third_imode = VOIDmode;
+ break;
+ case V16QImode:
+ gen_load_even = gen_vec_setv16qi;
+ gen_interleave_first_low = gen_vec_interleave_lowv8hi;
+ gen_interleave_second_low = gen_vec_interleave_lowv4si;
+ first_imode = V8HImode;
+ second_imode = V4SImode;
+ third_imode = V2DImode;
+ break;
+ default:
+ gcc_unreachable ();
+ }
+
+ for (i = 0; i < n; i++)
+ {
+ /* Extend the odd elment to SImode using a paradoxical SUBREG. */
+ op0 = gen_reg_rtx (SImode);
+ emit_move_insn (op0, gen_lowpart (SImode, ops [i + i]));
+
+ /* Insert the SImode value as low element of V4SImode vector. */
+ op1 = gen_reg_rtx (V4SImode);
+ op0 = gen_rtx_VEC_MERGE (V4SImode,
+ gen_rtx_VEC_DUPLICATE (V4SImode,
+ op0),
+ CONST0_RTX (V4SImode),
+ const1_rtx);
+ emit_insn (gen_rtx_SET (VOIDmode, op1, op0));
+
+ /* Cast the V4SImode vector back to a vector in orignal mode. */
+ op0 = gen_reg_rtx (mode);
+ emit_move_insn (op0, gen_lowpart (mode, op1));
+
+ /* Load even elements into the second positon. */
+ emit_insn ((*gen_load_even) (op0, ops [i + i + 1],
+ const1_rtx));
+
+ /* Cast vector to FIRST_IMODE vector. */
+ ops[i] = gen_reg_rtx (first_imode);
+ emit_move_insn (ops[i], gen_lowpart (first_imode, op0));
+ }
+
+ /* Interleave low FIRST_IMODE vectors. */
+ for (i = j = 0; i < n; i += 2, j++)
+ {
+ op0 = gen_reg_rtx (first_imode);
+ emit_insn ((*gen_interleave_first_low) (op0, ops[i], ops[i + 1]));
+
+ /* Cast FIRST_IMODE vector to SECOND_IMODE vector. */
+ ops[j] = gen_reg_rtx (second_imode);
+ emit_move_insn (ops[j], gen_lowpart (second_imode, op0));
+ }
+
+ /* Interleave low SECOND_IMODE vectors. */
+ switch (second_imode)
+ {
+ case V4SImode:
+ for (i = j = 0; i < n / 2; i += 2, j++)
+ {
+ op0 = gen_reg_rtx (second_imode);
+ emit_insn ((*gen_interleave_second_low) (op0, ops[i],
+ ops[i + 1]));
+
+ /* Cast the SECOND_IMODE vector to the THIRD_IMODE
+ vector. */
+ ops[j] = gen_reg_rtx (third_imode);
+ emit_move_insn (ops[j], gen_lowpart (third_imode, op0));
+ }
+ second_imode = V2DImode;
+ gen_interleave_second_low = gen_vec_interleave_lowv2di;
+ /* FALLTHRU */
+
+ case V2DImode:
+ op0 = gen_reg_rtx (second_imode);
+ emit_insn ((*gen_interleave_second_low) (op0, ops[0],
+ ops[1]));
+
+ /* Cast the SECOND_IMODE vector back to a vector on original
+ mode. */
+ emit_insn (gen_rtx_SET (VOIDmode, target,
+ gen_lowpart (mode, op0)));
+ break;
+
+ default:
+ gcc_unreachable ();
+ }
+}
+
/* A subroutine of ix86_expand_vector_init. Handle the most general case:
all values variable, and none identical. */
ix86_expand_vector_init_general (bool mmx_ok, enum machine_mode mode,
rtx target, rtx vals)
{
- enum machine_mode half_mode = GET_MODE_INNER (mode);
- rtx op0 = NULL, op1 = NULL;
- bool use_vec_concat = false;
+ rtx ops[16];
+ int n, i;
switch (mode)
{
break;
/* FALLTHRU */
- case V2DFmode:
- case V2DImode:
- /* For the two element vectors, we always implement VEC_CONCAT. */
- op0 = XVECEXP (vals, 0, 0);
- op1 = XVECEXP (vals, 0, 1);
- use_vec_concat = true;
- break;
-
case V4SFmode:
- half_mode = V2SFmode;
- goto half;
case V4SImode:
- half_mode = V2SImode;
- goto half;
- half:
- {
- rtvec v;
-
- /* For V4SF and V4SI, we implement a concat of two V2 vectors.
- Recurse to load the two halves. */
+ case V2DFmode:
+ case V2DImode:
+ n = GET_MODE_NUNITS (mode);
+ for (i = 0; i < n; i++)
+ ops[i] = XVECEXP (vals, 0, i);
+ ix86_expand_vector_init_concat (mode, target, ops, n);
+ return;
- op0 = gen_reg_rtx (half_mode);
- v = gen_rtvec (2, XVECEXP (vals, 0, 0), XVECEXP (vals, 0, 1));
- ix86_expand_vector_init (false, op0, gen_rtx_PARALLEL (half_mode, v));
+ case V16QImode:
+ if (!TARGET_SSE4_1)
+ break;
+ /* FALLTHRU */
- op1 = gen_reg_rtx (half_mode);
- v = gen_rtvec (2, XVECEXP (vals, 0, 2), XVECEXP (vals, 0, 3));
- ix86_expand_vector_init (false, op1, gen_rtx_PARALLEL (half_mode, v));
+ case V8HImode:
+ if (!TARGET_SSE2)
+ break;
- use_vec_concat = true;
- }
- break;
+ n = GET_MODE_NUNITS (mode);
+ for (i = 0; i < n; i++)
+ ops[i] = XVECEXP (vals, 0, i);
+ ix86_expand_vector_init_interleave (mode, target, ops, n >> 1);
+ return;
- case V8HImode:
- case V16QImode:
case V4HImode:
case V8QImode:
break;
gcc_unreachable ();
}
- if (use_vec_concat)
- {
- if (!register_operand (op0, half_mode))
- op0 = force_reg (half_mode, op0);
- if (!register_operand (op1, half_mode))
- op1 = force_reg (half_mode, op1);
-
- emit_insn (gen_rtx_SET (VOIDmode, target,
- gen_rtx_VEC_CONCAT (mode, op0, op1)));
- }
- else
{
int i, j, n_elts, n_words, n_elt_per_word;
enum machine_mode inner_mode;
else if (n_words == 4)
{
rtx tmp = gen_reg_rtx (V4SImode);
+ gcc_assert (word_mode == SImode);
vals = gen_rtx_PARALLEL (V4SImode, gen_rtvec_v (4, words));
ix86_expand_vector_init_general (false, V4SImode, tmp, vals);
emit_move_insn (target, gen_lowpart (mode, tmp));
NUM is the number of operands.
USES_OC0 is true if the instruction uses OC0 and provides 4 variants.
NUM_MEMORY is the maximum number of memory operands to accept. */
+
bool
-ix86_sse5_valid_op_p (rtx operands[], rtx insn, int num, bool uses_oc0, int num_memory)
+ix86_sse5_valid_op_p (rtx operands[], rtx insn ATTRIBUTE_UNUSED, int num,
+ bool uses_oc0, int num_memory)
{
int mem_mask;
int mem_count;
}
}
+ /* Special case pmacsdq{l,h} where we allow the 3rd argument to be
+ a memory operation. */
+ if (num_memory < 0)
+ {
+ num_memory = -num_memory;
+ if ((mem_mask & (1 << (num-1))) != 0)
+ {
+ mem_mask &= ~(1 << (num-1));
+ mem_count--;
+ }
+ }
+
/* If there were no memory operations, allow the insn */
if (mem_mask == 0)
return true;
}
/* Initialize the GCC target structure. */
+#undef TARGET_RETURN_IN_MEMORY
+#define TARGET_RETURN_IN_MEMORY ix86_return_in_memory
+
#undef TARGET_ATTRIBUTE_TABLE
#define TARGET_ATTRIBUTE_TABLE ix86_attribute_table
#if TARGET_DLLIMPORT_DECL_ATTRIBUTES
#undef TARGET_FUNCTION_VALUE
#define TARGET_FUNCTION_VALUE ix86_function_value
+#undef TARGET_SECONDARY_RELOAD
+#define TARGET_SECONDARY_RELOAD ix86_secondary_reload
+
#undef TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST
#define TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST x86_builtin_vectorization_cost
OSDN Git Service
