* [PATCH, rs6000] Replace comparison operators with vec_cmp* in intrinsic compatibility headers
@ 2018-10-24 20:22 Bill Schmidt
2018-10-25 18:43 ` Segher Boessenkool
0 siblings, 1 reply; 2+ messages in thread
From: Bill Schmidt @ 2018-10-24 20:22 UTC (permalink / raw)
To: GCC Patches; +Cc: Segher Boessenkool
Hi,
Due to some unfortunate history, not all compilers currently have correct result
types produced for comparison operators on vector types. For compatibility
purposes, this patch replaces those with vec_cmp* built-ins. It also cleans up
some silly type usages and formatting in the affected functions.
Bootstrapped and tested on powerpc64le-linux-gnu with no regressions. Is this
okay for trunk?
Thanks,
Bill
2018-10-24 Bill Schmidt <wschmidt@linux.ibm.com>
Jinsong Ji <jji@us.ibm.com>
* gcc/config/rs6000/emmintrin.h (_mm_sll_epi16): Replace
comparison operators with vec_cmp* for compatibility due to
unfortunate history; clean up formatting and use types more
appropriately.
(_mm_sll_epi32): Likewise.
(_mm_sll_epi64): Likewise.
(_mm_srl_epi16): Likewise.
(_mm_srl_epi32): Likewise.
(_mm_srl_epi64): Likewise.
Index: gcc/config/rs6000/emmintrin.h
===================================================================
--- gcc/config/rs6000/emmintrin.h (revision 265464)
+++ gcc/config/rs6000/emmintrin.h (working copy)
@@ -1725,18 +1725,19 @@ _mm_srli_epi64 (__m128i __A, int __B)
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sll_epi16 (__m128i __A, __m128i __B)
{
- __v8hu lshift, shmask;
+ __v8hu lshift;
+ __vector __bool short shmask;
const __v8hu shmax = { 15, 15, 15, 15, 15, 15, 15, 15 };
__v8hu result;
#ifdef __LITTLE_ENDIAN__
- lshift = vec_splat ((__v8hu)__B, 0);
+ lshift = vec_splat ((__v8hu) __B, 0);
#elif __BIG_ENDIAN__
- lshift = vec_splat ((__v8hu)__B, 3);
+ lshift = vec_splat ((__v8hu) __B, 3);
#endif
- shmask = lshift <= shmax;
+ shmask = vec_cmple (lshift, shmax);
result = vec_vslh ((__v8hu) __A, lshift);
- result = vec_sel (shmask, result, shmask);
+ result = vec_sel ((__v8hu) shmask, result, shmask);
return (__m128i) result;
}
@@ -1744,17 +1745,18 @@ _mm_sll_epi16 (__m128i __A, __m128i __B)
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sll_epi32 (__m128i __A, __m128i __B)
{
- __v4su lshift, shmask;
+ __v4su lshift;
+ __vector __bool int shmask;
const __v4su shmax = { 32, 32, 32, 32 };
__v4su result;
#ifdef __LITTLE_ENDIAN__
- lshift = vec_splat ((__v4su)__B, 0);
+ lshift = vec_splat ((__v4su) __B, 0);
#elif __BIG_ENDIAN__
- lshift = vec_splat ((__v4su)__B, 1);
+ lshift = vec_splat ((__v4su) __B, 1);
#endif
- shmask = lshift < shmax;
+ shmask = vec_cmplt (lshift, shmax);
result = vec_vslw ((__v4su) __A, lshift);
- result = vec_sel (shmask, result, shmask);
+ result = vec_sel ((__v4su) shmask, result, shmask);
return (__m128i) result;
}
@@ -1763,15 +1765,15 @@ _mm_sll_epi32 (__m128i __A, __m128i __B)
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_sll_epi64 (__m128i __A, __m128i __B)
{
- __v2du lshift, shmask;
+ __v2du lshift;
+ __vector __bool long long shmask;
const __v2du shmax = { 64, 64 };
__v2du result;
- lshift = (__v2du) vec_splat ((__v2du)__B, 0);
- shmask = lshift < shmax;
+ lshift = vec_splat ((__v2du) __B, 0);
+ shmask = vec_cmplt (lshift, shmax);
result = vec_vsld ((__v2du) __A, lshift);
- result = (__v2du) vec_sel ((__v2df) shmask, (__v2df) result,
- (__v2df) shmask);
+ result = vec_sel ((__v2du) shmask, result, shmask);
return (__m128i) result;
}
@@ -1816,18 +1818,19 @@ _mm_sra_epi32 (__m128i __A, __m128i __B)
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srl_epi16 (__m128i __A, __m128i __B)
{
- __v8hu rshift, shmask;
+ __v8hu rshift;
+ __vector __bool short shmask;
const __v8hu shmax = { 15, 15, 15, 15, 15, 15, 15, 15 };
__v8hu result;
#ifdef __LITTLE_ENDIAN__
- rshift = vec_splat ((__v8hu)__B, 0);
+ rshift = vec_splat ((__v8hu) __B, 0);
#elif __BIG_ENDIAN__
- rshift = vec_splat ((__v8hu)__B, 3);
+ rshift = vec_splat ((__v8hu) __B, 3);
#endif
- shmask = rshift <= shmax;
+ shmask = vec_cmple (rshift, shmax);
result = vec_vsrh ((__v8hu) __A, rshift);
- result = vec_sel (shmask, result, shmask);
+ result = vec_sel ((__v8hu) shmask, result, shmask);
return (__m128i) result;
}
@@ -1835,18 +1838,19 @@ _mm_srl_epi16 (__m128i __A, __m128i __B)
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srl_epi32 (__m128i __A, __m128i __B)
{
- __v4su rshift, shmask;
+ __v4su rshift;
+ __vector __bool int shmask;
const __v4su shmax = { 32, 32, 32, 32 };
__v4su result;
#ifdef __LITTLE_ENDIAN__
- rshift = vec_splat ((__v4su)__B, 0);
+ rshift = vec_splat ((__v4su) __B, 0);
#elif __BIG_ENDIAN__
- rshift = vec_splat ((__v4su)__B, 1);
+ rshift = vec_splat ((__v4su) __B, 1);
#endif
- shmask = rshift < shmax;
+ shmask = vec_cmplt (rshift, shmax);
result = vec_vsrw ((__v4su) __A, rshift);
- result = vec_sel (shmask, result, shmask);
+ result = vec_sel ((__v4su) shmask, result, shmask);
return (__m128i) result;
}
@@ -1855,14 +1859,15 @@ _mm_srl_epi32 (__m128i __A, __m128i __B)
extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
_mm_srl_epi64 (__m128i __A, __m128i __B)
{
- __v2du rshift, shmask;
+ __v2du rshift;
+ __vector __bool long long shmask;
const __v2du shmax = { 64, 64 };
__v2du result;
- rshift = (__v2du) vec_splat ((__v2du)__B, 0);
- shmask = rshift < shmax;
+ rshift = vec_splat ((__v2du) __B, 0);
+ shmask = vec_cmplt (rshift, shmax);
result = vec_vsrd ((__v2du) __A, rshift);
- result = (__v2du)vec_sel ((__v2du)shmask, (__v2du)result, (__v2du)shmask);
+ result = vec_sel ((__v2du) shmask, result, shmask);
return (__m128i) result;
}
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH, rs6000] Replace comparison operators with vec_cmp* in intrinsic compatibility headers
2018-10-24 20:22 [PATCH, rs6000] Replace comparison operators with vec_cmp* in intrinsic compatibility headers Bill Schmidt
@ 2018-10-25 18:43 ` Segher Boessenkool
0 siblings, 0 replies; 2+ messages in thread
From: Segher Boessenkool @ 2018-10-25 18:43 UTC (permalink / raw)
To: Bill Schmidt; +Cc: GCC Patches
Hi!
On Wed, Oct 24, 2018 at 02:30:10PM -0500, Bill Schmidt wrote:
> Due to some unfortunate history, not all compilers currently have correct result
> types produced for comparison operators on vector types. For compatibility
> purposes, this patch replaces those with vec_cmp* built-ins. It also cleans up
> some silly type usages and formatting in the affected functions.
Those compilers really need to be fixed, this isn't a tiny detail they
get wrong.
But, this patch makes the code more readable, so okay. Thanks!
Segher
> 2018-10-24 Bill Schmidt <wschmidt@linux.ibm.com>
> Jinsong Ji <jji@us.ibm.com>
>
> * gcc/config/rs6000/emmintrin.h (_mm_sll_epi16): Replace
> comparison operators with vec_cmp* for compatibility due to
> unfortunate history; clean up formatting and use types more
> appropriately.
> (_mm_sll_epi32): Likewise.
> (_mm_sll_epi64): Likewise.
> (_mm_srl_epi16): Likewise.
> (_mm_srl_epi32): Likewise.
> (_mm_srl_epi64): Likewise.
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2018-10-25 17:29 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-10-24 20:22 [PATCH, rs6000] Replace comparison operators with vec_cmp* in intrinsic compatibility headers Bill Schmidt
2018-10-25 18:43 ` Segher Boessenkool
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).