* [PATCH] Improve *avx_cvtp?2??256_2
@ 2016-05-04 19:36 Jakub Jelinek
2016-05-06 11:46 ` Kirill Yukhin
0 siblings, 1 reply; 2+ messages in thread
From: Jakub Jelinek @ 2016-05-04 19:36 UTC (permalink / raw)
To: Uros Bizjak, Kirill Yukhin; +Cc: gcc-patches
Hi!
Not sure how to easily construct a testcase for this (these insns are
usually used for vectorization, and then it really depends on register
pressure).
But in any case, looking at documentation it seems all the used insns are
available (generally even for further patches, what I'm looking for is
whether the insns are available already in AVX512F, or, if all the operands
are 128-bit or 256-bit vectors, in AVX512VL, or if they need further ISA
extensions; HARD_REGNO_MODE_OK should guarantee that the 128-bit and 256-bit
vectors would not be assigned to xmm16+ unless -mavx512vl).
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2016-05-04 Jakub Jelinek <jakub@redhat.com>
* config/i386/sse.md (*avx_cvtpd2dq256_2, *avx_cvtps2pd256_2): Use
v constraint instead of x.
--- gcc/config/i386/sse.md.jj 2016-05-04 14:36:08.000000000 +0200
+++ gcc/config/i386/sse.md 2016-05-04 15:16:44.180894303 +0200
@@ -4735,9 +4735,9 @@ (define_expand "avx_cvtpd2dq256_2"
"operands[2] = CONST0_RTX (V4SImode);")
(define_insn "*avx_cvtpd2dq256_2"
- [(set (match_operand:V8SI 0 "register_operand" "=x")
+ [(set (match_operand:V8SI 0 "register_operand" "=v")
(vec_concat:V8SI
- (unspec:V4SI [(match_operand:V4DF 1 "nonimmediate_operand" "xm")]
+ (unspec:V4SI [(match_operand:V4DF 1 "nonimmediate_operand" "vm")]
UNSPEC_FIX_NOTRUNC)
(match_operand:V4SI 2 "const0_operand")))]
"TARGET_AVX"
@@ -5050,10 +5050,10 @@ (define_insn "<sse2_avx_avx512f>_cvtps2p
(set_attr "mode" "<MODE>")])
(define_insn "*avx_cvtps2pd256_2"
- [(set (match_operand:V4DF 0 "register_operand" "=x")
+ [(set (match_operand:V4DF 0 "register_operand" "=v")
(float_extend:V4DF
(vec_select:V4SF
- (match_operand:V8SF 1 "nonimmediate_operand" "xm")
+ (match_operand:V8SF 1 "nonimmediate_operand" "vm")
(parallel [(const_int 0) (const_int 1)
(const_int 2) (const_int 3)]))))]
"TARGET_AVX"
Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Improve *avx_cvtp?2??256_2
2016-05-04 19:36 [PATCH] Improve *avx_cvtp?2??256_2 Jakub Jelinek
@ 2016-05-06 11:46 ` Kirill Yukhin
0 siblings, 0 replies; 2+ messages in thread
From: Kirill Yukhin @ 2016-05-06 11:46 UTC (permalink / raw)
To: Jakub Jelinek; +Cc: Uros Bizjak, gcc-patches
On 04 May 21:35, Jakub Jelinek wrote:
> Hi!
>
> Not sure how to easily construct a testcase for this (these insns are
> usually used for vectorization, and then it really depends on register
> pressure).
> But in any case, looking at documentation it seems all the used insns are
> available (generally even for further patches, what I'm looking for is
> whether the insns are available already in AVX512F, or, if all the operands
> are 128-bit or 256-bit vectors, in AVX512VL, or if they need further ISA
> extensions; HARD_REGNO_MODE_OK should guarantee that the 128-bit and 256-bit
> vectors would not be assigned to xmm16+ unless -mavx512vl).
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
OK for trunk.
--
Thanks, K
>
> 2016-05-04 Jakub Jelinek <jakub@redhat.com>
>
> * config/i386/sse.md (*avx_cvtpd2dq256_2, *avx_cvtps2pd256_2): Use
> v constraint instead of x.
>
> --- gcc/config/i386/sse.md.jj 2016-05-04 14:36:08.000000000 +0200
> +++ gcc/config/i386/sse.md 2016-05-04 15:16:44.180894303 +0200
> @@ -4735,9 +4735,9 @@ (define_expand "avx_cvtpd2dq256_2"
> "operands[2] = CONST0_RTX (V4SImode);")
>
> (define_insn "*avx_cvtpd2dq256_2"
> - [(set (match_operand:V8SI 0 "register_operand" "=x")
> + [(set (match_operand:V8SI 0 "register_operand" "=v")
> (vec_concat:V8SI
> - (unspec:V4SI [(match_operand:V4DF 1 "nonimmediate_operand" "xm")]
> + (unspec:V4SI [(match_operand:V4DF 1 "nonimmediate_operand" "vm")]
> UNSPEC_FIX_NOTRUNC)
> (match_operand:V4SI 2 "const0_operand")))]
> "TARGET_AVX"
> @@ -5050,10 +5050,10 @@ (define_insn "<sse2_avx_avx512f>_cvtps2p
> (set_attr "mode" "<MODE>")])
>
> (define_insn "*avx_cvtps2pd256_2"
> - [(set (match_operand:V4DF 0 "register_operand" "=x")
> + [(set (match_operand:V4DF 0 "register_operand" "=v")
> (float_extend:V4DF
> (vec_select:V4SF
> - (match_operand:V8SF 1 "nonimmediate_operand" "xm")
> + (match_operand:V8SF 1 "nonimmediate_operand" "vm")
> (parallel [(const_int 0) (const_int 1)
> (const_int 2) (const_int 3)]))))]
> "TARGET_AVX"
>
> Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2016-05-06 11:46 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-05-04 19:36 [PATCH] Improve *avx_cvtp?2??256_2 Jakub Jelinek
2016-05-06 11:46 ` Kirill Yukhin
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).