* [PATCH] i386: Use movss to implement V2SImode VEC_PERM.
@ 2023-03-14 17:43 Uros Bizjak
0 siblings, 0 replies; only message in thread
From: Uros Bizjak @ 2023-03-14 17:43 UTC (permalink / raw)
To: gcc-patches
[-- Attachment #1: Type: text/plain, Size: 868 bytes --]
Perform V2SI vector permutation in the same way as existing V2SF for
TARGET_MMX_WITH_SSE targets. The testcase:
typedef unsigned int v2si __attribute__((vector_size(8)));
v2si foo(v2si x, v2si y) { return (v2si){y[0], x[1]}; }
is currently compiled to (-O2):
foo:
movdqa %xmm0, %xmm2
movdqa %xmm1, %xmm0
pshufd $0xe5, %xmm2, %xmm2
punpckldq %xmm2, %xmm0
ret
and with the patched compiler:
foo:
movss %xmm1, %xmm0
ret
The functionality is already tested in gcc.target/i386/vperm-v2si.c
gcc/ChangeLog:
* config/i386/i386-expand.cc (expand_vec_perm_movs):
Handle V2SImode for TARGET_MMX_WITH_SSE.
* config/i386/mmx.md (*mmx_movss_<mode>): Rename from *mmx_movss
using V2FI mode iterator to handle both V2SI and V2SF modes.
Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.
Pushed to master.
Uros.
[-- Attachment #2: p.diff.txt --]
[-- Type: text/plain, Size: 1569 bytes --]
diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc
index 1094ece8b6d..6cc8bd5c80c 100644
--- a/gcc/config/i386/i386-expand.cc
+++ b/gcc/config/i386/i386-expand.cc
@@ -18949,11 +18949,9 @@ expand_vec_perm_movs (struct expand_vec_perm_d *d)
if (d->one_operand_p)
return false;
- if (!(TARGET_SSE && vmode == V4SFmode)
- && !(TARGET_SSE && vmode == V4SImode)
- && !(TARGET_MMX_WITH_SSE && vmode == V2SFmode)
- && !(TARGET_SSE2 && vmode == V2DFmode)
- && !(TARGET_SSE2 && vmode == V2DImode))
+ if (!(TARGET_SSE && (vmode == V4SFmode || vmode == V4SImode))
+ && !(TARGET_MMX_WITH_SSE && (vmode == V2SFmode || vmode == V2SImode))
+ && !(TARGET_SSE2 && (vmode == V2DFmode || vmode == V2DImode)))
return false;
/* Only the first element is changed. */
diff --git a/gcc/config/i386/mmx.md b/gcc/config/i386/mmx.md
index 0deccc2d2f4..f9c66115f81 100644
--- a/gcc/config/i386/mmx.md
+++ b/gcc/config/i386/mmx.md
@@ -1518,11 +1518,11 @@
(set_attr "prefix" "*,maybe_vex,orig")
(set_attr "mode" "DI,V4SF,V4SF")])
-(define_insn "*mmx_movss"
- [(set (match_operand:V2SF 0 "register_operand" "=x,v")
- (vec_merge:V2SF
- (match_operand:V2SF 2 "register_operand" " x,v")
- (match_operand:V2SF 1 "register_operand" " 0,v")
+(define_insn "*mmx_movss_<mode>"
+ [(set (match_operand:V2FI 0 "register_operand" "=x,v")
+ (vec_merge:V2FI
+ (match_operand:V2FI 2 "register_operand" " x,v")
+ (match_operand:V2FI 1 "register_operand" " 0,v")
(const_int 1)))]
"TARGET_MMX_WITH_SSE"
"@
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2023-03-14 17:43 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-03-14 17:43 [PATCH] i386: Use movss to implement V2SImode VEC_PERM Uros Bizjak
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).