This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
SSE fix 9 - MMX _m64 type
- From: Jan Hubicka <jh at suse dot cz>
- To: gcc-patches at gcc dot gnu dot org, rth at cygnus dot com,radkver at atrey dot karlin dot mff dot cuni dot cz
- Date: Fri, 18 Oct 2002 11:22:12 +0200
- Subject: SSE fix 9 - MMX _m64 type
Hi,
this makes __mm64 type vector that is required to get argument passing
convention right for both i386 and x86-64.
We've discussed the change earlier, so hope it is safe even when bit
intrusive.
Honza
Fri Oct 18 01:58:36 CEST 2002 Jan Hubicka <jh@suse.cz>
* mmintrin.h (__m64): typedef it to v2si.
(_mm_cvtsi32_si64, _mm_cvtsi32_si64_mm_sll_pi16,
_mm_sll_pi32, _mm_sll_pi64, _mm_slli_pi64, _mm_sra_pi16,
_mm_sra_pi32, _mm_srl_pi16, _mm_srl_pi32, _mm_srl_pi64,
_mm_srli_pi64, _mm_and_si64, _mm_andnot_si64,
_mm_or_si64, _mm_xor_si64): Add neccesary casts.
* xmmintrin.h (_mm_setzero_si64): Likewise.
*** /p1/dfa/egcs/gcc/config/i386/mmintrin.h Fri Feb 15 21:33:04 2002
--- mmintrin.h Fri Oct 18 01:54:59 2002
***************
*** 31,37 ****
#define _MMINTRIN_H_INCLUDED
/* The data type intended for user use. */
! typedef unsigned long long __m64;
/* Internal data types for implementing the intrinsics. */
typedef int __v2si __attribute__ ((__mode__ (__V2SI__)));
--- 31,37 ----
#define _MMINTRIN_H_INCLUDED
/* The data type intended for user use. */
! typedef int __m64 __attribute__ ((__mode__ (__V2SI__)));
/* Internal data types for implementing the intrinsics. */
typedef int __v2si __attribute__ ((__mode__ (__V2SI__)));
*************** _mm_empty (void)
*** 49,62 ****
static __inline __m64
_mm_cvtsi32_si64 (int __i)
{
! return (unsigned int) __i;
}
/* Convert the lower 32 bits of the __m64 object into an integer. */
static __inline int
_mm_cvtsi64_si32 (__m64 __i)
{
! return __i;
}
/* Pack the four 16-bit values from M1 into the lower four 8-bit values of
--- 49,64 ----
static __inline __m64
_mm_cvtsi32_si64 (int __i)
{
! long long __tmp = (unsigned int)__i;
! return (__m64) __tmp;
}
/* Convert the lower 32 bits of the __m64 object into an integer. */
static __inline int
_mm_cvtsi64_si32 (__m64 __i)
{
! long long __tmp = (long long)__i;
! return __tmp;
}
/* Pack the four 16-bit values from M1 into the lower four 8-bit values of
*************** _mm_mullo_pi16 (__m64 __m1, __m64 __m2)
*** 269,275 ****
static __inline __m64
_mm_sll_pi16 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psllw ((__v4hi)__m, __count);
}
static __inline __m64
--- 271,277 ----
static __inline __m64
_mm_sll_pi16 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psllw ((__v4hi)__m, (long long)__count);
}
static __inline __m64
*************** _mm_slli_pi16 (__m64 __m, int __count)
*** 282,288 ****
static __inline __m64
_mm_sll_pi32 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_pslld ((__v2si)__m, __count);
}
static __inline __m64
--- 284,290 ----
static __inline __m64
_mm_sll_pi32 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_pslld ((__v2si)__m, (long long)__count);
}
static __inline __m64
*************** _mm_slli_pi32 (__m64 __m, int __count)
*** 295,314 ****
static __inline __m64
_mm_sll_pi64 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psllq (__m, __count);
}
static __inline __m64
_mm_slli_pi64 (__m64 __m, int __count)
{
! return (__m64) __builtin_ia32_psllq (__m, __count);
}
/* Shift four 16-bit values in M right by COUNT; shift in the sign bit. */
static __inline __m64
_mm_sra_pi16 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psraw ((__v4hi)__m, __count);
}
static __inline __m64
--- 297,316 ----
static __inline __m64
_mm_sll_pi64 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psllq ((long long)__m, (long long)__count);
}
static __inline __m64
_mm_slli_pi64 (__m64 __m, int __count)
{
! return (__m64) __builtin_ia32_psllq ((long long)__m, (long long)__count);
}
/* Shift four 16-bit values in M right by COUNT; shift in the sign bit. */
static __inline __m64
_mm_sra_pi16 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psraw ((__v4hi)__m, (long long)__count);
}
static __inline __m64
*************** _mm_srai_pi16 (__m64 __m, int __count)
*** 321,327 ****
static __inline __m64
_mm_sra_pi32 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psrad ((__v2si)__m, __count);
}
static __inline __m64
--- 323,329 ----
static __inline __m64
_mm_sra_pi32 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psrad ((__v2si)__m, (long long)__count);
}
static __inline __m64
*************** _mm_srai_pi32 (__m64 __m, int __count)
*** 334,340 ****
static __inline __m64
_mm_srl_pi16 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, __count);
}
static __inline __m64
--- 336,342 ----
static __inline __m64
_mm_srl_pi16 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psrlw ((__v4hi)__m, (long long)__count);
}
static __inline __m64
*************** _mm_srli_pi16 (__m64 __m, int __count)
*** 347,353 ****
static __inline __m64
_mm_srl_pi32 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psrld ((__v2si)__m, __count);
}
static __inline __m64
--- 349,355 ----
static __inline __m64
_mm_srl_pi32 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psrld ((__v2si)__m, (long long)__count);
}
static __inline __m64
*************** _mm_srli_pi32 (__m64 __m, int __count)
*** 360,379 ****
static __inline __m64
_mm_srl_pi64 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psrlq (__m, __count);
}
static __inline __m64
_mm_srli_pi64 (__m64 __m, int __count)
{
! return (__m64) __builtin_ia32_psrlq (__m, __count);
}
/* Bit-wise AND the 64-bit values in M1 and M2. */
static __inline __m64
_mm_and_si64 (__m64 __m1, __m64 __m2)
{
! return __builtin_ia32_pand (__m1, __m2);
}
/* Bit-wise complement the 64-bit value in M1 and bit-wise AND it with the
--- 362,381 ----
static __inline __m64
_mm_srl_pi64 (__m64 __m, __m64 __count)
{
! return (__m64) __builtin_ia32_psrlq ((long long)__m, (long long)__count);
}
static __inline __m64
_mm_srli_pi64 (__m64 __m, int __count)
{
! return (__m64) __builtin_ia32_psrlq ((long long)__m, (long long)__count);
}
/* Bit-wise AND the 64-bit values in M1 and M2. */
static __inline __m64
_mm_and_si64 (__m64 __m1, __m64 __m2)
{
! return (__m64) __builtin_ia32_pand ((long long)__m1, (long long)__m2);
}
/* Bit-wise complement the 64-bit value in M1 and bit-wise AND it with the
*************** _mm_and_si64 (__m64 __m1, __m64 __m2)
*** 381,401 ****
static __inline __m64
_mm_andnot_si64 (__m64 __m1, __m64 __m2)
{
! return __builtin_ia32_pandn (__m1, __m2);
}
/* Bit-wise inclusive OR the 64-bit values in M1 and M2. */
static __inline __m64
_mm_or_si64 (__m64 __m1, __m64 __m2)
{
! return __builtin_ia32_por (__m1, __m2);
}
/* Bit-wise exclusive OR the 64-bit values in M1 and M2. */
static __inline __m64
_mm_xor_si64 (__m64 __m1, __m64 __m2)
{
! return __builtin_ia32_pxor (__m1, __m2);
}
/* Compare eight 8-bit values. The result of the comparison is 0xFF if the
--- 383,403 ----
static __inline __m64
_mm_andnot_si64 (__m64 __m1, __m64 __m2)
{
! return (__m64) __builtin_ia32_pandn ((long long)__m1, (long long)__m2);
}
/* Bit-wise inclusive OR the 64-bit values in M1 and M2. */
static __inline __m64
_mm_or_si64 (__m64 __m1, __m64 __m2)
{
! return (__m64)__builtin_ia32_por ((long long)__m1, (long long)__m2);
}
/* Bit-wise exclusive OR the 64-bit values in M1 and M2. */
static __inline __m64
_mm_xor_si64 (__m64 __m1, __m64 __m2)
{
! return (__m64)__builtin_ia32_pxor ((long long)__m1, (long long)__m2);
}
/* Compare eight 8-bit values. The result of the comparison is 0xFF if the
*************** _mm_cmpgt_pi32 (__m64 __m1, __m64 __m2)
*** 444,450 ****
static __inline __m64
_mm_setzero_si64 (void)
{
! return __builtin_ia32_mmx_zero ();
}
/* Creates a vector of two 32-bit values; I0 is least significant. */
--- 446,452 ----
static __inline __m64
_mm_setzero_si64 (void)
{
! return (__m64)__builtin_ia32_mmx_zero ();
}
/* Creates a vector of two 32-bit values; I0 is least significant. */