i386.c (IX86_BUILTIN_MOVQ, [...]): Remove.

* config/i386/i386.c (IX86_BUILTIN_MOVQ, IX86_BUILTIN_LOADD,
        IX86_BUILTIN_STORED, IX86_BUILTIN_MOVQ2DQ,
        IX86_BUILTIN_MOVDQ2Q): Remove.
        (IX86_BUILTIN_VEC_EXT_V4SI): New.
        (ix86_init_mmx_sse_builtins, ix86_expand_builtin): Update to match.
        (ix86_expand_vector_extract): For V4S[FI], extract element 0 after
        shuffling.
        * config/i386/sse.md (sse_concatv2sf): Accept zero operand 2.
        (sse2_pextrw): Fix immediate constraint.
        (sse2_loadq, sse2_loadq_rex64): Remove.
        * config/i386/emmintrin.h (_mm_cvtsi128_si32, _mm_cvtsi128_si64x):
        Use __builtin_ia32_vec_ext_<size>.
        (_mm_cvtsi32_si128, _mm_cvtsi64x_si128): Use _mm_set_epi<size>.

From-SVN: r93604
This commit is contained in:
Richard Henderson 2005-01-13 09:51:34 -08:00 committed by Richard Henderson
parent 172847595a
commit ed9b539671
4 changed files with 39 additions and 88 deletions

View File

@ -1,3 +1,19 @@
2005-01-13 Richard Henderson <rth@redhat.com>
* config/i386/i386.c (IX86_BUILTIN_MOVQ, IX86_BUILTIN_LOADD,
IX86_BUILTIN_STORED, IX86_BUILTIN_MOVQ2DQ,
IX86_BUILTIN_MOVDQ2Q): Remove.
(IX86_BUILTIN_VEC_EXT_V4SI): New.
(ix86_init_mmx_sse_builtins, ix86_expand_builtin): Update to match.
(ix86_expand_vector_extract): For V4S[FI], extract element 0 after
shuffling.
* config/i386/sse.md (sse_concatv2sf): Accept zero operand 2.
(sse2_pextrw): Fix immediate constraint.
(sse2_loadq, sse2_loadq_rex64): Remove.
* config/i386/emmintrin.h (_mm_cvtsi128_si32, _mm_cvtsi128_si64x):
Use __builtin_ia32_vec_ext_<size>.
(_mm_cvtsi32_si128, _mm_cvtsi64x_si128): Use _mm_set_epi<size>.
2005-01-13 Aldy Hernandez <aldyh@redhat.com>
* function.c (assign_parm_setup_block): Look inside original

View File

@ -195,20 +195,17 @@ _mm_storer_pd (double *__P, __m128d __A)
static __inline int
_mm_cvtsi128_si32 (__m128i __A)
{
int __tmp;
__builtin_ia32_stored (&__tmp, (__v4si)__A);
return __tmp;
return __builtin_ia32_vec_ext_v4si ((__v4si)__A, 0);
}
#ifdef __x86_64__
static __inline long long
_mm_cvtsi128_si64x (__m128i __A)
{
return __builtin_ia32_movdq2q ((__v2di)__A);
return __builtin_ia32_vec_ext_v2di ((__v2di)__A, 0);
}
#endif
static __inline __m128d
_mm_add_pd (__m128d __A, __m128d __B)
{
@ -1377,14 +1374,14 @@ _mm_mfence (void)
static __inline __m128i
_mm_cvtsi32_si128 (int __A)
{
return (__m128i) __builtin_ia32_loadd (&__A);
return _mm_set_epi32 (0, 0, 0, __A);
}
#ifdef __x86_64__
static __inline __m128i
_mm_cvtsi64x_si128 (long long __A)
{
return (__m128i) __builtin_ia32_movq2dq (__A);
return _mm_set_epi64x (0, __A);
}
#endif

View File

@ -12265,9 +12265,6 @@ enum ix86_builtins
IX86_BUILTIN_LOADDQU,
IX86_BUILTIN_STOREDQU,
IX86_BUILTIN_MOVQ,
IX86_BUILTIN_LOADD,
IX86_BUILTIN_STORED,
IX86_BUILTIN_PACKSSWB,
IX86_BUILTIN_PACKSSDW,
@ -12498,8 +12495,6 @@ enum ix86_builtins
IX86_BUILTIN_MASKMOVDQU,
IX86_BUILTIN_MOVMSKPD,
IX86_BUILTIN_PMOVMSKB128,
IX86_BUILTIN_MOVQ2DQ,
IX86_BUILTIN_MOVDQ2Q,
IX86_BUILTIN_PACKSSWB128,
IX86_BUILTIN_PACKSSDW128,
@ -12607,6 +12602,7 @@ enum ix86_builtins
IX86_BUILTIN_VEC_EXT_V2DF,
IX86_BUILTIN_VEC_EXT_V2DI,
IX86_BUILTIN_VEC_EXT_V4SF,
IX86_BUILTIN_VEC_EXT_V4SI,
IX86_BUILTIN_VEC_EXT_V8HI,
IX86_BUILTIN_VEC_EXT_V4HI,
IX86_BUILTIN_VEC_SET_V8HI,
@ -13154,8 +13150,6 @@ ix86_init_mmx_sse_builtins (void)
= build_function_type_list (V2SI_type_node,
V2SF_type_node, V2SF_type_node, NULL_TREE);
tree pint_type_node = build_pointer_type (integer_type_node);
tree pcint_type_node = build_pointer_type (
build_type_variant (integer_type_node, 1, 0));
tree pdouble_type_node = build_pointer_type (double_type_node);
tree pcdouble_type_node = build_pointer_type (
build_type_variant (double_type_node, 1, 0));
@ -13168,12 +13162,6 @@ ix86_init_mmx_sse_builtins (void)
intTI_type_node, intTI_type_node, NULL_TREE);
tree void_ftype_pcvoid
= build_function_type_list (void_type_node, const_ptr_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
@ -13285,13 +13273,6 @@ ix86_init_mmx_sse_builtins (void)
tree void_ftype_pchar_v16qi
= build_function_type_list (void_type_node,
pchar_type_node, V16QI_type_node, NULL_TREE);
tree v4si_ftype_pcint
= build_function_type_list (V4SI_type_node, pcint_type_node, NULL_TREE);
tree void_ftype_pcint_v4si
= build_function_type_list (void_type_node,
pcint_type_node, V4SI_type_node, NULL_TREE);
tree v2di_ftype_v2di
= build_function_type_list (V2DI_type_node, V2DI_type_node, NULL_TREE);
tree float80_type;
tree float128_type;
@ -13479,8 +13460,6 @@ ix86_init_mmx_sse_builtins (void)
/* SSE2 */
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_loadupd", v2df_ftype_pcdouble, IX86_BUILTIN_LOADUPD);
def_builtin (MASK_SSE2, "__builtin_ia32_storeupd", void_ftype_pdouble_v2df, IX86_BUILTIN_STOREUPD);
@ -13534,10 +13513,7 @@ ix86_init_mmx_sse_builtins (void)
def_builtin (MASK_SSE2, "__builtin_ia32_mfence", void_ftype_void, IX86_BUILTIN_MFENCE);
def_builtin (MASK_SSE2, "__builtin_ia32_loaddqu", v16qi_ftype_pcchar, IX86_BUILTIN_LOADDQU);
def_builtin (MASK_SSE2, "__builtin_ia32_loadd", v4si_ftype_pcint, IX86_BUILTIN_LOADD);
def_builtin (MASK_SSE2, "__builtin_ia32_storedqu", void_ftype_pchar_v16qi, IX86_BUILTIN_STOREDQU);
def_builtin (MASK_SSE2, "__builtin_ia32_stored", void_ftype_pcint_v4si, IX86_BUILTIN_STORED);
def_builtin (MASK_SSE2, "__builtin_ia32_movq", v2di_ftype_v2di, IX86_BUILTIN_MOVQ);
def_builtin (MASK_SSE2, "__builtin_ia32_pmuludq", di_ftype_v2si_v2si, IX86_BUILTIN_PMULUDQ);
def_builtin (MASK_SSE2, "__builtin_ia32_pmuludq128", v2di_ftype_v4si_v4si, IX86_BUILTIN_PMULUDQ128);
@ -13622,6 +13598,11 @@ ix86_init_mmx_sse_builtins (void)
def_builtin (MASK_SSE, "__builtin_ia32_vec_ext_v4sf",
ftype, IX86_BUILTIN_VEC_EXT_V4SF);
ftype = build_function_type_list (intSI_type_node, V4SI_type_node,
integer_type_node, NULL_TREE);
def_builtin (MASK_SSE, "__builtin_ia32_vec_ext_v4si",
ftype, IX86_BUILTIN_VEC_EXT_V4SI);
ftype = build_function_type_list (intHI_type_node, V8HI_type_node,
integer_type_node, NULL_TREE);
def_builtin (MASK_SSE, "__builtin_ia32_vec_ext_v8hi",
@ -14399,13 +14380,8 @@ ix86_expand_builtin (tree exp, rtx target, rtx subtarget ATTRIBUTE_UNUSED,
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_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);
case IX86_BUILTIN_MONITOR:
arg0 = TREE_VALUE (arglist);
@ -14447,6 +14423,7 @@ ix86_expand_builtin (tree exp, rtx target, rtx subtarget ATTRIBUTE_UNUSED,
case IX86_BUILTIN_VEC_EXT_V2DF:
case IX86_BUILTIN_VEC_EXT_V2DI:
case IX86_BUILTIN_VEC_EXT_V4SF:
case IX86_BUILTIN_VEC_EXT_V4SI:
case IX86_BUILTIN_VEC_EXT_V8HI:
case IX86_BUILTIN_VEC_EXT_V4HI:
return ix86_expand_vec_ext_builtin (arglist, target);
@ -14480,8 +14457,7 @@ ix86_expand_builtin (tree exp, rtx target, rtx subtarget ATTRIBUTE_UNUSED,
if (d->code == fcode)
return ix86_expand_sse_comi (d, arglist, target);
/* @@@ Should really do something sensible here. */
return 0;
gcc_unreachable ();
}
/* Store OPERAND to the memory after reload is completed. This means
@ -16402,6 +16378,7 @@ ix86_expand_vector_extract (bool mmx_ok, rtx target, rtx vec, int elt)
}
vec = tmp;
use_vec_extr = true;
elt = 0;
break;
case V4SImode:
@ -16431,6 +16408,7 @@ ix86_expand_vector_extract (bool mmx_ok, rtx target, rtx vec, int elt)
}
vec = tmp;
use_vec_extr = true;
elt = 0;
}
else
{

View File

@ -1067,16 +1067,18 @@
;; nonimmediate_operand for operand 2 and *not* allowing memory for the SSE
;; alternatives pretty much forces the MMX alternative to be chosen.
(define_insn "*sse_concatv2sf"
[(set (match_operand:V2SF 0 "register_operand" "=x,*y")
[(set (match_operand:V2SF 0 "register_operand" "=x,x,*y,*y")
(vec_concat:V2SF
(match_operand:SF 1 "register_operand" " 0, 0")
(match_operand:SF 2 "register_operand" " x,*y")))]
(match_operand:SF 1 "nonimmediate_operand" " 0,m, 0, m")
(match_operand:SF 2 "vector_move_operand" " x,C,*y, C")))]
"TARGET_SSE"
"@
unpcklps\t{%2, %0|%0, %2}
punpckldq\t{%2, %0|%0, %2}"
[(set_attr "type" "sselog,mmxcvt")
(set_attr "mode" "V4SF,DI")])
movss\t{%1, %0|%0, %1}
punpckldq\t{%2, %0|%0, %2}
movd\t{%1, %0|%0, %1}"
[(set_attr "type" "sselog,ssemov,mmxcvt,mmxmov")
(set_attr "mode" "V4SF,SF,DI,DI")])
(define_insn "*sse_concatv4sf"
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
@ -2671,7 +2673,7 @@
(zero_extend:SI
(vec_select:HI
(match_operand:V8HI 1 "register_operand" "x")
(parallel [(match_operand:SI 2 "const_0_to_7_operand" "0")]))))]
(parallel [(match_operand:SI 2 "const_0_to_7_operand" "n")]))))]
"TARGET_SSE2"
"pextrw\t{%2, %1, %0|%0, %1, %2}"
[(set_attr "type" "sselog")
@ -2865,48 +2867,6 @@
operands[1] = gen_rtx_REG (DImode, REGNO (operands[1]));
})
(define_expand "sse2_loadq"
[(set (match_operand:V2DI 0 "register_operand" "")
(vec_merge:V2DI
(vec_duplicate:V2DI
(match_operand:DI 1 "nonimmediate_operand" ""))
(match_dup 2)
(const_int 1)))]
"TARGET_SSE"
"operands[2] = CONST0_RTX (V2DImode);")
(define_insn "*sse2_loadq"
[(set (match_operand:V2DI 0 "register_operand" "=Y,?Y,Y,x")
(vec_merge:V2DI
(vec_duplicate:V2DI
(match_operand:DI 1 "nonimmediate_operand" " m,*y,Y,0"))
(match_operand:V2DI 2 "vector_move_operand" " C, C,0,x")
(const_int 1)))]
"TARGET_SSE && !TARGET_64BIT"
"@
movq\t{%1, %0|%0, %1}
movq2dq\t{%1, %0|%0, %1}
movq\t{%1, %0|%0, %1}
shufps\t{$0xe4, %1, %0|%0, %1, 0xe4}"
[(set_attr "type" "ssemov,ssemov,ssemov,sselog")
(set_attr "mode" "TI,TI,TI,V4SF")])
(define_insn "*sse2_loadq_rex64"
[(set (match_operand:V2DI 0 "register_operand" "=x,?x,?x,x")
(vec_merge:V2DI
(vec_duplicate:V2DI
(match_operand:DI 1 "nonimmediate_operand" " m,*y, r,x"))
(match_operand:V2DI 2 "vector_move_operand" " C, C, C,0")
(const_int 1)))]
"TARGET_SSE2 && TARGET_64BIT"
"@
movq\t{%1, %0|%0, %1}
movq2dq\t{%1, %0|%0, %1}
movd\t{%1, %0|%0, %1}
movq\t{%1, %0|%0, %1}"
[(set_attr "type" "ssemov")
(set_attr "mode" "TI")])
(define_insn "*vec_dupv4si"
[(set (match_operand:V4SI 0 "register_operand" "=Y,x")
(vec_duplicate:V4SI