+Tue Oct 22 00:04:20 CEST 2002 Jan Hubicka <jh@suse.cz>
+
+ * i386.c (builtin_description): Add punpcklqdq and movdq2q
+ (ix86_init_mmx_sse_builtins): Add v2di_ftype_void, di_ftype_v2di,
+ v16qi_ftype_pchar, void_ftype_pchar_v16qi, v4si_ftype_pchar,
+ void_ftype_pchar_v4si; Initialize __builtin_ia32_movdq2q,
+ __builtin_ia32_loaddqa, __builtin_ia32_loaddqu, __builtin_ia32_loadd
+ __builtin_ia32_storedqa, __builtin_ia32_storedqu, __builtin_ia32_stored
+ __builtin_ia32_setzero128.
+ (ix86_expand_builtin): Handle IX86_BUILTIN_CLRTI, IX86_BUILTIN_LOADDQA,
+ IX86_BUILTIN_LOADDQU, IX86_BUILTIN_LOADD, IX86_BUILTIN_STOREDQA,
+ IX86_BUILTIN_STOREDQU, IX86_BUILTIN_STORED, Ix86_BUILTIN_MOVQ.
+ * i386.h (ix86_builtins): Add IX86_BUILTIN_LOADDQA, IX86_BUILTIN_LOADDQU,
+ IX86_BUILTIN_STOREDQA, IX86_BUILTIN_STOREDQU, IX86_BUILTIN_LOADD,
+ IX86_BUILTIN_STORED, IX86_BUILTIN_CLRTI, IX86_BUILTIN_MOVDQ2Q,
+ IX86_BUILTIN_PUNPCKLQDQ128, Ix86_BUILTIN_MOVQ.
+ * i386.md (sse2_punpcklqdq, sse2_movqsse2_loadd, sse2_stored,
+ sse2_movq): New patterns.
+ (sse2_movdqa, sse2_movdqu, sse2_movdq2q): Fix.
+ * xmmintrin.h (_mm_load_si128, _mm_loadu_si128, _mm_loadl_epi64,
+ _mm_store_si128, _mm_storeu_si128, _mm_storel_epi64,
+ _mm_setzero_si128, _mm_set_epi64, _mm_set_epi32, _mm_set_epi16,
+ _mm_set_epi8, _mm_set1_epi64, _mm_set1_epi32, _mm_set1_epi16,
+ _mm_set1_epi8, _mm_setr_epi64, _mm_setr_epi32, _mm_setr_epi16,
+ _mm_setr_epi8, _mm_unpacklo_epi64,_mm_set_moveq): New functions.
+ (_mm_insert_epi16): Fix.
+
2002-10-21 Dale Johannesen <dalej@apple.com>
* config/rs6000/rs6000.c (rs6000_reverse_condition): Handle
static tree ix86_handle_cdecl_attribute PARAMS ((tree *, tree, tree, int, bool *));
static tree ix86_handle_regparm_attribute PARAMS ((tree *, tree, tree, int, bool *));
static int ix86_value_regno PARAMS ((enum machine_mode));
+static bool contains_128bit_aligned_vector_p PARAMS ((tree));
#if defined (DO_GLOBAL_CTORS_BODY) && defined (HAS_INIT_SECTION)
static void ix86_svr3_asm_out_constructor PARAMS ((rtx, int));
{ MASK_SSE2, CODE_FOR_sse2_punpcklbw, "__builtin_ia32_punpcklbw128", IX86_BUILTIN_PUNPCKLBW128, 0, 0 },
{ MASK_SSE2, CODE_FOR_sse2_punpcklwd, "__builtin_ia32_punpcklwd128", IX86_BUILTIN_PUNPCKLWD128, 0, 0 },
{ MASK_SSE2, CODE_FOR_sse2_punpckldq, "__builtin_ia32_punpckldq128", IX86_BUILTIN_PUNPCKLDQ128, 0, 0 },
+ { MASK_SSE2, CODE_FOR_sse2_punpcklqdq, "__builtin_ia32_punpcklqdq128", IX86_BUILTIN_PUNPCKLQDQ128, 0, 0 },
{ MASK_SSE2, CODE_FOR_sse2_packsswb, "__builtin_ia32_packsswb128", IX86_BUILTIN_PACKSSWB128, 0, 0 },
{ MASK_SSE2, CODE_FOR_sse2_packssdw, "__builtin_ia32_packssdw128", IX86_BUILTIN_PACKSSDW128, 0, 0 },
{ MASK_SSE2, CODE_FOR_sse2_pmovmskb, 0, IX86_BUILTIN_PMOVMSKB128, 0, 0 },
{ MASK_SSE2, CODE_FOR_sse2_movmskpd, 0, IX86_BUILTIN_MOVMSKPD, 0, 0 },
{ MASK_SSE2, CODE_FOR_sse2_movq2dq, 0, IX86_BUILTIN_MOVQ2DQ, 0, 0 },
+ { MASK_SSE2, CODE_FOR_sse2_movdq2q, 0, IX86_BUILTIN_MOVDQ2Q, 0, 0 },
{ MASK_SSE2, CODE_FOR_sqrtv2df2, 0, IX86_BUILTIN_SQRTPD, 0, 0 },
{ MASK_SSE2, CODE_FOR_cvtps2dq, 0, IX86_BUILTIN_CVTPS2DQ, 0, 0 },
{ MASK_SSE2, CODE_FOR_cvtps2pd, 0, IX86_BUILTIN_CVTPS2PD, 0, 0 },
- { MASK_SSE2, CODE_FOR_cvttps2dq, 0, IX86_BUILTIN_CVTTPS2DQ, 0, 0 }
+ { MASK_SSE2, CODE_FOR_cvttps2dq, 0, IX86_BUILTIN_CVTTPS2DQ, 0, 0 },
+
+ { MASK_SSE2, CODE_FOR_sse2_movq, 0, IX86_BUILTIN_MOVQ, 0, 0 }
};
void
tree ti_ftype_void
= build_function_type (intTI_type_node, void_list_node);
+ tree v2di_ftype_void
+ = build_function_type (V2DI_type_node, void_list_node);
tree ti_ftype_ti_ti
= build_function_type_list (intTI_type_node,
intTI_type_node, intTI_type_node, NULL_TREE);
tree v2di_ftype_di
= build_function_type_list (V2DI_type_node,
long_long_unsigned_type_node, NULL_TREE);
+ tree di_ftype_v2di
+ = build_function_type_list (long_long_unsigned_type_node,
+ V2DI_type_node, NULL_TREE);
tree v4sf_ftype_v4si
= build_function_type_list (V4SF_type_node, V4SI_type_node, NULL_TREE);
tree v4si_ftype_v4sf
V16QI_type_node, V16QI_type_node, NULL_TREE);
tree int_ftype_v16qi
= build_function_type_list (integer_type_node, V16QI_type_node, NULL_TREE);
+ tree v16qi_ftype_pchar
+ = build_function_type_list (V16QI_type_node, pchar_type_node, NULL_TREE);
+ tree void_ftype_pchar_v16qi
+ = build_function_type_list (void_type_node,
+ pchar_type_node, V16QI_type_node, NULL_TREE);
+ tree v4si_ftype_pchar
+ = build_function_type_list (V4SI_type_node, pchar_type_node, NULL_TREE);
+ tree void_ftype_pchar_v4si
+ = build_function_type_list (void_type_node,
+ pchar_type_node, V4SI_type_node, NULL_TREE);
+ tree v2di_ftype_v2di
+ = build_function_type_list (V2DI_type_node, V2DI_type_node, NULL_TREE);
/* Add all builtins that are more or less simple operations on two
operands. */
def_builtin (MASK_SSE2, "__builtin_ia32_maskmovdqu", void_ftype_v16qi_v16qi_pchar, IX86_BUILTIN_MASKMOVDQU);
def_builtin (MASK_SSE2, "__builtin_ia32_movq2dq", v2di_ftype_di, IX86_BUILTIN_MOVQ2DQ);
+ def_builtin (MASK_SSE2, "__builtin_ia32_movdq2q", di_ftype_v2di, IX86_BUILTIN_MOVDQ2Q);
def_builtin (MASK_SSE2, "__builtin_ia32_loadapd", v2df_ftype_pdouble, IX86_BUILTIN_LOADAPD);
def_builtin (MASK_SSE2, "__builtin_ia32_loadupd", v2df_ftype_pdouble, IX86_BUILTIN_LOADUPD);
def_builtin (MASK_SSE2, "__builtin_ia32_lfence", void_ftype_void, IX86_BUILTIN_LFENCE);
def_builtin (MASK_SSE2, "__builtin_ia32_mfence", void_ftype_void, IX86_BUILTIN_MFENCE);
+ def_builtin (MASK_SSE2, "__builtin_ia32_loaddqa", v16qi_ftype_pchar, IX86_BUILTIN_LOADDQA);
+ def_builtin (MASK_SSE2, "__builtin_ia32_loaddqu", v16qi_ftype_pchar, IX86_BUILTIN_LOADDQU);
+ def_builtin (MASK_SSE2, "__builtin_ia32_loadd", v4si_ftype_pchar, IX86_BUILTIN_LOADD);
+ def_builtin (MASK_SSE2, "__builtin_ia32_storedqa", void_ftype_pchar_v16qi, IX86_BUILTIN_STOREDQA);
+ def_builtin (MASK_SSE2, "__builtin_ia32_storedqu", void_ftype_pchar_v16qi, IX86_BUILTIN_STOREDQU);
+ def_builtin (MASK_SSE2, "__builtin_ia32_stored", void_ftype_pchar_v4si, IX86_BUILTIN_STORED);
+ def_builtin (MASK_SSE2, "__builtin_ia32_movq", v2di_ftype_v2di, IX86_BUILTIN_MOVQ);
+
+ def_builtin (MASK_SSE1, "__builtin_ia32_setzero128", v2di_ftype_void, IX86_BUILTIN_CLRTI);
+
def_builtin (MASK_SSE2, "__builtin_ia32_psllw128", v8hi_ftype_v8hi_v2di, IX86_BUILTIN_PSLLW128);
def_builtin (MASK_SSE2, "__builtin_ia32_pslld128", v4si_ftype_v4si_v2di, IX86_BUILTIN_PSLLD128);
def_builtin (MASK_SSE2, "__builtin_ia32_psllq128", v2di_ftype_v2di_v2di, IX86_BUILTIN_PSLLQ128);
case IX86_BUILTIN_STOREAPS:
return ix86_expand_store_builtin (CODE_FOR_sse_movaps, arglist);
+
case IX86_BUILTIN_STOREUPS:
return ix86_expand_store_builtin (CODE_FOR_sse_movups, arglist);
emit_insn (gen_mmx_clrdi (target));
return target;
+ case IX86_BUILTIN_CLRTI:
+ target = gen_reg_rtx (V2DImode);
+ emit_insn (gen_sse2_clrti (simplify_gen_subreg (TImode, target, V2DImode, 0)));
+ return target;
+
+
case IX86_BUILTIN_SQRTSD:
return ix86_expand_unop1_builtin (CODE_FOR_vmsqrtv2df2, arglist, target);
case IX86_BUILTIN_LOADAPD:
case IX86_BUILTIN_MOVNTI:
return ix86_expand_store_builtin (CODE_FOR_sse2_movntsi, arglist);
+ case IX86_BUILTIN_LOADDQA:
+ return ix86_expand_unop_builtin (CODE_FOR_sse2_movdqa, arglist, target, 1);
+ case IX86_BUILTIN_LOADDQU:
+ return ix86_expand_unop_builtin (CODE_FOR_sse2_movdqu, arglist, target, 1);
+ case IX86_BUILTIN_LOADD:
+ return ix86_expand_unop_builtin (CODE_FOR_sse2_loadd, arglist, target, 1);
+
+ case IX86_BUILTIN_STOREDQA:
+ return ix86_expand_store_builtin (CODE_FOR_sse2_movdqa, arglist);
+ case IX86_BUILTIN_STOREDQU:
+ return ix86_expand_store_builtin (CODE_FOR_sse2_movdqu, arglist);
+ case IX86_BUILTIN_STORED:
+ return ix86_expand_store_builtin (CODE_FOR_sse2_stored, arglist);
+
default:
break;
}
IX86_BUILTIN_MOVNTPS,
IX86_BUILTIN_MOVNTQ,
+ IX86_BUILTIN_LOADDQA,
+ IX86_BUILTIN_LOADDQU,
+ IX86_BUILTIN_STOREDQA,
+ IX86_BUILTIN_STOREDQU,
+ IX86_BUILTIN_MOVQ,
+ IX86_BUILTIN_LOADD,
+ IX86_BUILTIN_STORED,
+
+ IX86_BUILTIN_CLRTI,
+
IX86_BUILTIN_PACKSSWB,
IX86_BUILTIN_PACKSSDW,
IX86_BUILTIN_PACKUSWB,
IX86_BUILTIN_MOVMSKPD,
IX86_BUILTIN_PMOVMSKB128,
IX86_BUILTIN_MOVQ2DQ,
+ IX86_BUILTIN_MOVDQ2Q,
IX86_BUILTIN_PACKSSWB128,
IX86_BUILTIN_PACKSSDW128,
IX86_BUILTIN_PUNPCKLBW128,
IX86_BUILTIN_PUNPCKLWD128,
IX86_BUILTIN_PUNPCKLDQ128,
+ IX86_BUILTIN_PUNPCKLQDQ128,
IX86_BUILTIN_CLFLUSH,
IX86_BUILTIN_MFENCE,
[(set_attr "type" "ssecvt")
(set_attr "mode" "TI")])
+(define_insn "sse2_punpcklqdq"
+ [(set (match_operand:V2DI 0 "register_operand" "=x")
+ (vec_merge:V2DI
+ (match_operand:V2DI 1 "register_operand" "0")
+ (vec_select:V2DI (match_operand:V2DI 2 "register_operand" "x")
+ (parallel [(const_int 1)
+ (const_int 0)]))
+ (const_int 1)))]
+ "TARGET_SSE2"
+ "punpcklqdq\t{%2, %0|%0, %2}"
+ [(set_attr "type" "ssecvt")
+ (set_attr "mode" "TI")])
+
;; SSE2 moves
(define_insn "sse2_movapd"
(set_attr "mode" "V2DF")])
(define_insn "sse2_movdqa"
- [(set (match_operand:TI 0 "nonimmediate_operand" "=x,m")
- (unspec:TI [(match_operand:TI 1 "general_operand" "xm,x")]
- UNSPEC_MOVA))]
+ [(set (match_operand:V16QI 0 "nonimmediate_operand" "=x,m")
+ (unspec:V16QI [(match_operand:V16QI 1 "general_operand" "xm,x")]
+ UNSPEC_MOVA))]
"TARGET_SSE2"
"@
movdqa\t{%1, %0|%0, %1}
(set_attr "mode" "TI")])
(define_insn "sse2_movdqu"
- [(set (match_operand:TI 0 "nonimmediate_operand" "=x,m")
- (unspec:TI [(match_operand:TI 1 "general_operand" "xm,x")]
- UNSPEC_MOVU))]
+ [(set (match_operand:V16QI 0 "nonimmediate_operand" "=x,m")
+ (unspec:V16QI [(match_operand:V16QI 1 "general_operand" "xm,x")]
+ UNSPEC_MOVU))]
"TARGET_SSE2"
"@
movdqu\t{%1, %0|%0, %1}
(set_attr "mode" "TI")])
(define_insn "sse2_movdq2q"
- [(set (match_operand:DI 0 "nonimmediate_operand" "=y")
- (vec_select:DI (match_operand:V2DI 1 "general_operand" "x")
+ [(set (match_operand:DI 0 "nonimmediate_operand" "=m,y")
+ (vec_select:DI (match_operand:V2DI 1 "register_operand" "x,x")
(parallel [(const_int 0)])))]
"TARGET_SSE2"
- "movdq2q\t{%1, %0|%0, %1}"
+ "@
+ movq\t{%1, %0|%0, %1}
+ movdq2q\t{%1, %0|%0, %1}"
[(set_attr "type" "ssecvt")
(set_attr "mode" "TI")])
(define_insn "sse2_movq2dq"
- [(set (match_operand:V2DI 0 "nonimmediate_operand" "=x")
- (vec_concat:V2DI (match_operand:DI 1 "general_operand" "y")
- (const_vector:DI [(const_int 0)])))]
+ [(set (match_operand:V2DI 0 "register_operand" "=x,?x")
+ (vec_concat:V2DI (match_operand:DI 1 "nonimmediate_operand" "m,y")
+ (const_int 0)))]
"TARGET_SSE2"
- "movq2dq\t{%1, %0|%0, %1}"
- [(set_attr "type" "ssecvt")
+ "@
+ movq\t{%1, %0|%0, %1}
+ movq2dq\t{%1, %0|%0, %1}"
+ [(set_attr "type" "ssecvt,ssemov")
+ (set_attr "mode" "TI")])
+
+(define_insn "sse2_movq"
+ [(set (match_operand:V2DI 0 "register_operand" "=x")
+ (vec_concat:V2DI (vec_select:DI
+ (match_operand:V2DI 1 "nonimmediate_operand" "xm")
+ (parallel [(const_int 0)]))
+ (const_int 0)))]
+ "TARGET_SSE2"
+ "movq\t{%1, %0|%0, %1}"
+ [(set_attr "type" "ssemov")
+ (set_attr "mode" "TI")])
+
+(define_insn "sse2_loadd"
+ [(set (match_operand:V4SI 0 "register_operand" "=x")
+ (vec_merge:V4SI
+ (vec_duplicate:V4HI (match_operand:SI 1 "nonimmediate_operand" "mr"))
+ (const_vector:V4SI [(const_int 0)
+ (const_int 0)
+ (const_int 0)
+ (const_int 0)])
+ (const_int 1)))]
+ "TARGET_SSE2"
+ "movd\t{%1, %0|%0, %1}"
+ [(set_attr "type" "ssemov")
+ (set_attr "mode" "TI")])
+
+(define_insn "sse2_stored"
+ [(set (match_operand:SI 0 "nonimmediate_operand" "=mr")
+ (vec_select:SI
+ (match_operand:V4SI 1 "register_operand" "x")
+ (parallel [(const_int 0)])))]
+ "TARGET_SSE2"
+ "movd\t{%1, %0|%0, %1}"
+ [(set_attr "type" "ssemov")
(set_attr "mode" "TI")])
(define_insn "sse2_movhpd"
return __builtin_ia32_ucomisdneq ((__v2df)__A, (__v2df)__B);
}
+/* Create a vector with element 0 as *P and the rest zero. */
+
+static __inline __m128i
+_mm_load_si128 (__m128i const *__P)
+{
+ return (__m128i) __builtin_ia32_loaddqa (__P);
+}
+
+static __inline __m128i
+_mm_loadu_si128 (__m128i const *__P)
+{
+ return (__m128i) __builtin_ia32_loaddqu (__P);
+}
+
+static __inline __m128i
+_mm_loadl_epi64 (__m128i const *__P)
+{
+ return (__m128i) __builtin_ia32_movq2dq (*(unsigned long long *)__P);
+}
+
+static __inline void
+_mm_store_si128 (__m128i *__P, __m128i __B)
+{
+ __builtin_ia32_storedqa (__P, (__v16qi)__B);
+}
+
+static __inline void
+_mm_storeu_si128 (__m128i *__P, __m128i __B)
+{
+ __builtin_ia32_storedqu (__P, (__v16qi)__B);
+}
+
+static __inline void
+_mm_storel_epi64 (__m128i *__P, __m128i __B)
+{
+ *(long long *)__P = __builtin_ia32_movdq2q ((__v2di)__B);
+}
+
+static __inline __m128i
+_mm_move_epi64 (__m128i __A)
+{
+ return (__m128i) __builtin_ia32_movq ((__v2di)__A);
+}
+
+/* Create a vector of zeros. */
+static __inline __m128i
+_mm_setzero_si128 (void)
+{
+ return (__m128i) __builtin_ia32_setzero128 ();
+}
+
+static __inline __m128i
+_mm_set_epi64 (__m64 __A, __m64 __B)
+{
+ __v2di __tmp = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__A);
+ __v2di __tmp2 = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__B);
+ return (__m128i)__builtin_ia32_punpcklqdq128 (__tmp2, __tmp);
+}
+
+/* Create the vector [Z Y X W]. */
+static __inline __m128i
+_mm_set_epi32 (int __Z, int __Y, int __X, int __W)
+{
+ union {
+ int __a[4];
+ __m128i __v;
+ } __u;
+
+ __u.__a[0] = __W;
+ __u.__a[1] = __X;
+ __u.__a[2] = __Y;
+ __u.__a[3] = __Z;
+
+ return __u.__v;
+}
+/* Create the vector [S T U V Z Y X W]. */
+static __inline __m128i
+_mm_set_epi16 (short __Z, short __Y, short __X, short __W,
+ short __V, short __U, short __T, short __S)
+{
+ union {
+ short __a[8];
+ __m128i __v;
+ } __u;
+
+ __u.__a[0] = __S;
+ __u.__a[1] = __T;
+ __u.__a[2] = __U;
+ __u.__a[3] = __V;
+ __u.__a[4] = __W;
+ __u.__a[5] = __X;
+ __u.__a[6] = __Y;
+ __u.__a[7] = __Z;
+
+ return __u.__v;
+}
+
+/* Create the vector [S T U V Z Y X W]. */
+static __inline __m128i
+_mm_set_epi8 (char __Z, char __Y, char __X, char __W,
+ char __V, char __U, char __T, char __S,
+ char __Z1, char __Y1, char __X1, char __W1,
+ char __V1, char __U1, char __T1, char __S1)
+{
+ union {
+ char __a[16];
+ __m128i __v;
+ } __u;
+
+ __u.__a[0] = __S1;
+ __u.__a[1] = __T1;
+ __u.__a[2] = __U1;
+ __u.__a[3] = __V1;
+ __u.__a[4] = __W1;
+ __u.__a[5] = __X1;
+ __u.__a[6] = __Y1;
+ __u.__a[7] = __Z1;
+ __u.__a[8] = __S;
+ __u.__a[9] = __T;
+ __u.__a[10] = __U;
+ __u.__a[11] = __V;
+ __u.__a[12] = __W;
+ __u.__a[13] = __X;
+ __u.__a[14] = __Y;
+ __u.__a[15] = __Z;
+
+ return __u.__v;
+}
+
+static __inline __m128i
+_mm_set1_epi64 (__m64 __A)
+{
+ __v2di __tmp = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__A);
+ return (__m128i)__builtin_ia32_punpcklqdq128 (__tmp, __tmp);
+}
+
+static __inline __m128i
+_mm_set1_epi32 (int __A)
+{
+ __v4si __tmp = (__v4si)__builtin_ia32_loadd (&__A);
+ return (__m128i) __builtin_ia32_pshufd ((__v4si)__tmp, _MM_SHUFFLE (0,0,0,0));
+}
+
+static __inline __m128i
+_mm_set1_epi16 (short __A)
+{
+ int __Acopy = (unsigned short)__A;
+ __v4si __tmp = (__v4si)__builtin_ia32_loadd (&__Acopy);
+ __tmp = (__v4si)__builtin_ia32_punpcklwd128 ((__v8hi)__tmp, (__v8hi)__tmp);
+ return (__m128i) __builtin_ia32_pshufd ((__v4si)__tmp, _MM_SHUFFLE (0,0,0,0));
+}
+
+static __inline __m128i
+_mm_set1_epi8 (char __A)
+{
+ int __Acopy = (unsigned char)__A;
+ __v4si __tmp = (__v4si)__builtin_ia32_loadd (&__Acopy);
+ __tmp = (__v4si)__builtin_ia32_punpcklbw128 ((__v16qi)__tmp, (__v16qi)__tmp);
+ __tmp = (__v4si)__builtin_ia32_punpcklbw128 ((__v16qi)__tmp, (__v16qi)__tmp);
+ return (__m128i) __builtin_ia32_pshufd ((__v4si)__tmp, _MM_SHUFFLE (0,0,0,0));
+}
+
+static __inline __m128i
+_mm_setr_epi64 (__m64 __A, __m64 __B)
+{
+ __v2di __tmp = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__A);
+ __v2di __tmp2 = (__v2di)__builtin_ia32_movq2dq ((unsigned long long)__B);
+ return (__m128i)__builtin_ia32_punpcklqdq128 (__tmp, __tmp2);
+}
+
+/* Create the vector [Z Y X W]. */
+static __inline __m128i
+_mm_setr_epi32 (int __W, int __X, int __Y, int __Z)
+{
+ union {
+ int __a[4];
+ __m128i __v;
+ } __u;
+
+ __u.__a[0] = __W;
+ __u.__a[1] = __X;
+ __u.__a[2] = __Y;
+ __u.__a[3] = __Z;
+
+ return __u.__v;
+}
+/* Create the vector [S T U V Z Y X W]. */
+static __inline __m128i
+_mm_setr_epi16 (short __S, short __T, short __U, short __V,
+ short __W, short __X, short __Y, short __Z)
+{
+ union {
+ short __a[8];
+ __m128i __v;
+ } __u;
+
+ __u.__a[0] = __S;
+ __u.__a[1] = __T;
+ __u.__a[2] = __U;
+ __u.__a[3] = __V;
+ __u.__a[4] = __W;
+ __u.__a[5] = __X;
+ __u.__a[6] = __Y;
+ __u.__a[7] = __Z;
+
+ return __u.__v;
+}
+
+/* Create the vector [S T U V Z Y X W]. */
+static __inline __m128i
+_mm_setr_epi8 (char __S1, char __T1, char __U1, char __V1,
+ char __W1, char __X1, char __Y1, char __Z1,
+ char __S, char __T, char __U, char __V,
+ char __W, char __X, char __Y, char __Z)
+{
+ union {
+ char __a[16];
+ __m128i __v;
+ } __u;
+
+ __u.__a[0] = __S1;
+ __u.__a[1] = __T1;
+ __u.__a[2] = __U1;
+ __u.__a[3] = __V1;
+ __u.__a[4] = __W1;
+ __u.__a[5] = __X1;
+ __u.__a[6] = __Y1;
+ __u.__a[7] = __Z1;
+ __u.__a[8] = __S;
+ __u.__a[9] = __T;
+ __u.__a[10] = __U;
+ __u.__a[11] = __V;
+ __u.__a[12] = __W;
+ __u.__a[13] = __X;
+ __u.__a[14] = __Y;
+ __u.__a[15] = __Z;
+
+ return __u.__v;
+}
+
static __inline __m128d
_mm_cvtepi32_pd (__m128i __A)
{
return (__m128i)__builtin_ia32_punpckldq128 ((__v4si)__A, (__v4si)__B);
}
+static __inline __m128i
+_mm_unpacklo_epi64 (__m128i __A, __m128i __B)
+{
+ return (__m128i)__builtin_ia32_punpcklqdq128 ((__v2di)__A, (__v2di)__B);
+}
+
static __inline __m128i
_mm_add_epi8 (__m128i __A, __m128i __B)
{
#define _mm_extract_epi16(__A, __B) __builtin_ia32_pextrw128 ((__v8hi)__A, __B)
-#define _mm_insert_epi16 (__A, __B, __C) ((__m128i)__builtin_ia32_pinsrw128 ((__v8hi)__A, __B, __C))
+#define _mm_insert_epi16(__A, __B, __C) ((__m128i)__builtin_ia32_pinsrw128 ((__v8hi)__A, __B, __C))
static __inline __m128i
_mm_max_epi16 (__m128i __A, __m128i __B)