* [PATCH] target/100312 - make x86 masked load builtins pure
@ 2021-04-28 11:02 Richard Biener
2021-04-28 11:23 ` Uros Bizjak
0 siblings, 1 reply; 8+ messages in thread
From: Richard Biener @ 2021-04-28 11:02 UTC (permalink / raw)
To: gcc-patches; +Cc: jakub, ubizjak
This arranges for the x86 AVX and AVX2 masked load builtins to be
pure to enable dead code elimination and more appropriate alias
analysis.
Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
Thanks,
Richard.
2021-04-28 Richard Biener <rguenther@suse.de>
PR target/100312
* config/i386/i386-builtin.def (IX86_BUILTIN_MASKLOADPD,
IX86_BUILTIN_MASKLOADPS, IX86_BUILTIN_MASKLOADPD256,
IX86_BUILTIN_MASKLOADPS256, IX86_BUILTIN_MASKLOADD,
IX86_BUILTIN_MASKLOADQ, IX86_BUILTIN_MASKLOADD256,
IX86_BUILTIN_MASKLOADQ256): Remove.
* config/i386/i386-builtins.h (ix86_builtins): Add entries for
IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256.
* config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
Initialize IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
as pure builtins.
---
gcc/config/i386/i386-builtin.def | 8 --------
gcc/config/i386/i386-builtins.c | 19 +++++++++++++++++++
gcc/config/i386/i386-builtins.h | 8 ++++++++
3 files changed, 27 insertions(+), 8 deletions(-)
diff --git a/gcc/config/i386/i386-builtin.def b/gcc/config/i386/i386-builtin.def
index 4dbd4f23647..dc0452de645 100644
--- a/gcc/config/i386/i386-builtin.def
+++ b/gcc/config/i386/i386-builtin.def
@@ -187,10 +187,6 @@ BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_movntv4di, "__builtin_ia32_movntdq25
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_movntv4df, "__builtin_ia32_movntpd256", IX86_BUILTIN_MOVNTPD256, UNKNOWN, (int) VOID_FTYPE_PDOUBLE_V4DF)
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_movntv8sf, "__builtin_ia32_movntps256", IX86_BUILTIN_MOVNTPS256, UNKNOWN, (int) VOID_FTYPE_PFLOAT_V8SF)
-BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadpd, "__builtin_ia32_maskloadpd", IX86_BUILTIN_MASKLOADPD, UNKNOWN, (int) V2DF_FTYPE_PCV2DF_V2DI)
-BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadps, "__builtin_ia32_maskloadps", IX86_BUILTIN_MASKLOADPS, UNKNOWN, (int) V4SF_FTYPE_PCV4SF_V4SI)
-BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadpd256, "__builtin_ia32_maskloadpd256", IX86_BUILTIN_MASKLOADPD256, UNKNOWN, (int) V4DF_FTYPE_PCV4DF_V4DI)
-BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadps256, "__builtin_ia32_maskloadps256", IX86_BUILTIN_MASKLOADPS256, UNKNOWN, (int) V8SF_FTYPE_PCV8SF_V8SI)
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstorepd, "__builtin_ia32_maskstorepd", IX86_BUILTIN_MASKSTOREPD, UNKNOWN, (int) VOID_FTYPE_PV2DF_V2DI_V2DF)
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstoreps, "__builtin_ia32_maskstoreps", IX86_BUILTIN_MASKSTOREPS, UNKNOWN, (int) VOID_FTYPE_PV4SF_V4SI_V4SF)
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstorepd256, "__builtin_ia32_maskstorepd256", IX86_BUILTIN_MASKSTOREPD256, UNKNOWN, (int) VOID_FTYPE_PV4DF_V4DI_V4DF)
@@ -198,10 +194,6 @@ BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstoreps256, "__builtin_ia32_mask
/* AVX2 */
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_movntdqa, "__builtin_ia32_movntdqa256", IX86_BUILTIN_MOVNTDQA256, UNKNOWN, (int) V4DI_FTYPE_PV4DI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadd, "__builtin_ia32_maskloadd", IX86_BUILTIN_MASKLOADD, UNKNOWN, (int) V4SI_FTYPE_PCV4SI_V4SI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadq, "__builtin_ia32_maskloadq", IX86_BUILTIN_MASKLOADQ, UNKNOWN, (int) V2DI_FTYPE_PCV2DI_V2DI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadd256, "__builtin_ia32_maskloadd256", IX86_BUILTIN_MASKLOADD256, UNKNOWN, (int) V8SI_FTYPE_PCV8SI_V8SI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadq256, "__builtin_ia32_maskloadq256", IX86_BUILTIN_MASKLOADQ256, UNKNOWN, (int) V4DI_FTYPE_PCV4DI_V4DI)
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskstored, "__builtin_ia32_maskstored", IX86_BUILTIN_MASKSTORED, UNKNOWN, (int) VOID_FTYPE_PV4SI_V4SI_V4SI)
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskstoreq, "__builtin_ia32_maskstoreq", IX86_BUILTIN_MASKSTOREQ, UNKNOWN, (int) VOID_FTYPE_PV2DI_V2DI_V2DI)
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskstored256, "__builtin_ia32_maskstored256", IX86_BUILTIN_MASKSTORED256, UNKNOWN, (int) VOID_FTYPE_PV8SI_V8SI_V8SI)
diff --git a/gcc/config/i386/i386-builtins.c b/gcc/config/i386/i386-builtins.c
index 128bd39816c..2ffcbe5cd3e 100644
--- a/gcc/config/i386/i386-builtins.c
+++ b/gcc/config/i386/i386-builtins.c
@@ -667,7 +667,26 @@ ix86_init_mmx_sse_builtins (void)
"__builtin_ia32_rdrand64_step", INT_FTYPE_PULONGLONG,
IX86_BUILTIN_RDRAND64_STEP);
+ /* AVX */
+ def_builtin_pure (OPTION_MASK_ISA_AVX, 0, "__builtin_ia32_maskloadpd",
+ V2DF_FTYPE_PCV2DF_V2DI, IX86_BUILTIN_MASKLOADPD);
+ def_builtin_pure (OPTION_MASK_ISA_AVX, 0, "__builtin_ia32_maskloadps",
+ V4SF_FTYPE_PCV4SF_V4SI, IX86_BUILTIN_MASKLOADPS);
+ def_builtin_pure (OPTION_MASK_ISA_AVX, 0, "__builtin_ia32_maskloadpd256",
+ V4DF_FTYPE_PCV4DF_V4DI, IX86_BUILTIN_MASKLOADPD256);
+ def_builtin_pure (OPTION_MASK_ISA_AVX, 0, "__builtin_ia32_maskloadps256",
+ V8SF_FTYPE_PCV8SF_V8SI, IX86_BUILTIN_MASKLOADPS256);
+
/* AVX2 */
+ def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_maskloadd",
+ V4SI_FTYPE_PCV4SI_V4SI, IX86_BUILTIN_MASKLOADD);
+ def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_maskloadq",
+ V2DI_FTYPE_PCV2DI_V2DI, IX86_BUILTIN_MASKLOADQ);
+ def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_maskloadd256",
+ V8SI_FTYPE_PCV8SI_V8SI, IX86_BUILTIN_MASKLOADD256);
+ def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_maskloadq256",
+ V4DI_FTYPE_PCV4DI_V4DI, IX86_BUILTIN_MASKLOADQ256);
+
def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_gathersiv2df",
V2DF_FTYPE_V2DF_PCDOUBLE_V4SI_V2DF_INT,
IX86_BUILTIN_GATHERSIV2DF);
diff --git a/gcc/config/i386/i386-builtins.h b/gcc/config/i386/i386-builtins.h
index 0641808c7a7..6a5e249e382 100644
--- a/gcc/config/i386/i386-builtins.h
+++ b/gcc/config/i386/i386-builtins.h
@@ -60,6 +60,14 @@ enum ix86_builtins
IX86_BUILTIN_VEC_SET_V8HI,
IX86_BUILTIN_VEC_SET_V4HI,
IX86_BUILTIN_VEC_SET_V16QI,
+ IX86_BUILTIN_MASKLOADPD,
+ IX86_BUILTIN_MASKLOADPS,
+ IX86_BUILTIN_MASKLOADPD256,
+ IX86_BUILTIN_MASKLOADPS256,
+ IX86_BUILTIN_MASKLOADD,
+ IX86_BUILTIN_MASKLOADQ,
+ IX86_BUILTIN_MASKLOADD256,
+ IX86_BUILTIN_MASKLOADQ256,
IX86_BUILTIN_GATHERSIV2DF,
IX86_BUILTIN_GATHERSIV4DF,
IX86_BUILTIN_GATHERDIV2DF,
--
2.26.2
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] target/100312 - make x86 masked load builtins pure
2021-04-28 11:02 [PATCH] target/100312 - make x86 masked load builtins pure Richard Biener
@ 2021-04-28 11:23 ` Uros Bizjak
2021-04-28 11:46 ` Richard Biener
0 siblings, 1 reply; 8+ messages in thread
From: Uros Bizjak @ 2021-04-28 11:23 UTC (permalink / raw)
To: Richard Biener; +Cc: gcc-patches, Jakub Jelinek
On Wed, Apr 28, 2021 at 1:02 PM Richard Biener <rguenther@suse.de> wrote:
>
> This arranges for the x86 AVX and AVX2 masked load builtins to be
> pure to enable dead code elimination and more appropriate alias
> analysis.
>
> Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
>
> Thanks,
> Richard.
>
> 2021-04-28 Richard Biener <rguenther@suse.de>
>
> PR target/100312
> * config/i386/i386-builtin.def (IX86_BUILTIN_MASKLOADPD,
> IX86_BUILTIN_MASKLOADPS, IX86_BUILTIN_MASKLOADPD256,
> IX86_BUILTIN_MASKLOADPS256, IX86_BUILTIN_MASKLOADD,
> IX86_BUILTIN_MASKLOADQ, IX86_BUILTIN_MASKLOADD256,
> IX86_BUILTIN_MASKLOADQ256): Remove.
> * config/i386/i386-builtins.h (ix86_builtins): Add entries for
> IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256.
> * config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
> Initialize IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
> as pure builtins.
OK.
Thanks,
Uros.
> ---
> gcc/config/i386/i386-builtin.def | 8 --------
> gcc/config/i386/i386-builtins.c | 19 +++++++++++++++++++
> gcc/config/i386/i386-builtins.h | 8 ++++++++
> 3 files changed, 27 insertions(+), 8 deletions(-)
>
> diff --git a/gcc/config/i386/i386-builtin.def b/gcc/config/i386/i386-builtin.def
> index 4dbd4f23647..dc0452de645 100644
> --- a/gcc/config/i386/i386-builtin.def
> +++ b/gcc/config/i386/i386-builtin.def
> @@ -187,10 +187,6 @@ BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_movntv4di, "__builtin_ia32_movntdq25
> BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_movntv4df, "__builtin_ia32_movntpd256", IX86_BUILTIN_MOVNTPD256, UNKNOWN, (int) VOID_FTYPE_PDOUBLE_V4DF)
> BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_movntv8sf, "__builtin_ia32_movntps256", IX86_BUILTIN_MOVNTPS256, UNKNOWN, (int) VOID_FTYPE_PFLOAT_V8SF)
>
> -BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadpd, "__builtin_ia32_maskloadpd", IX86_BUILTIN_MASKLOADPD, UNKNOWN, (int) V2DF_FTYPE_PCV2DF_V2DI)
> -BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadps, "__builtin_ia32_maskloadps", IX86_BUILTIN_MASKLOADPS, UNKNOWN, (int) V4SF_FTYPE_PCV4SF_V4SI)
> -BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadpd256, "__builtin_ia32_maskloadpd256", IX86_BUILTIN_MASKLOADPD256, UNKNOWN, (int) V4DF_FTYPE_PCV4DF_V4DI)
> -BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadps256, "__builtin_ia32_maskloadps256", IX86_BUILTIN_MASKLOADPS256, UNKNOWN, (int) V8SF_FTYPE_PCV8SF_V8SI)
> BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstorepd, "__builtin_ia32_maskstorepd", IX86_BUILTIN_MASKSTOREPD, UNKNOWN, (int) VOID_FTYPE_PV2DF_V2DI_V2DF)
> BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstoreps, "__builtin_ia32_maskstoreps", IX86_BUILTIN_MASKSTOREPS, UNKNOWN, (int) VOID_FTYPE_PV4SF_V4SI_V4SF)
> BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstorepd256, "__builtin_ia32_maskstorepd256", IX86_BUILTIN_MASKSTOREPD256, UNKNOWN, (int) VOID_FTYPE_PV4DF_V4DI_V4DF)
> @@ -198,10 +194,6 @@ BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstoreps256, "__builtin_ia32_mask
>
> /* AVX2 */
> BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_movntdqa, "__builtin_ia32_movntdqa256", IX86_BUILTIN_MOVNTDQA256, UNKNOWN, (int) V4DI_FTYPE_PV4DI)
> -BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadd, "__builtin_ia32_maskloadd", IX86_BUILTIN_MASKLOADD, UNKNOWN, (int) V4SI_FTYPE_PCV4SI_V4SI)
> -BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadq, "__builtin_ia32_maskloadq", IX86_BUILTIN_MASKLOADQ, UNKNOWN, (int) V2DI_FTYPE_PCV2DI_V2DI)
> -BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadd256, "__builtin_ia32_maskloadd256", IX86_BUILTIN_MASKLOADD256, UNKNOWN, (int) V8SI_FTYPE_PCV8SI_V8SI)
> -BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadq256, "__builtin_ia32_maskloadq256", IX86_BUILTIN_MASKLOADQ256, UNKNOWN, (int) V4DI_FTYPE_PCV4DI_V4DI)
> BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskstored, "__builtin_ia32_maskstored", IX86_BUILTIN_MASKSTORED, UNKNOWN, (int) VOID_FTYPE_PV4SI_V4SI_V4SI)
> BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskstoreq, "__builtin_ia32_maskstoreq", IX86_BUILTIN_MASKSTOREQ, UNKNOWN, (int) VOID_FTYPE_PV2DI_V2DI_V2DI)
> BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskstored256, "__builtin_ia32_maskstored256", IX86_BUILTIN_MASKSTORED256, UNKNOWN, (int) VOID_FTYPE_PV8SI_V8SI_V8SI)
> diff --git a/gcc/config/i386/i386-builtins.c b/gcc/config/i386/i386-builtins.c
> index 128bd39816c..2ffcbe5cd3e 100644
> --- a/gcc/config/i386/i386-builtins.c
> +++ b/gcc/config/i386/i386-builtins.c
> @@ -667,7 +667,26 @@ ix86_init_mmx_sse_builtins (void)
> "__builtin_ia32_rdrand64_step", INT_FTYPE_PULONGLONG,
> IX86_BUILTIN_RDRAND64_STEP);
>
> + /* AVX */
> + def_builtin_pure (OPTION_MASK_ISA_AVX, 0, "__builtin_ia32_maskloadpd",
> + V2DF_FTYPE_PCV2DF_V2DI, IX86_BUILTIN_MASKLOADPD);
> + def_builtin_pure (OPTION_MASK_ISA_AVX, 0, "__builtin_ia32_maskloadps",
> + V4SF_FTYPE_PCV4SF_V4SI, IX86_BUILTIN_MASKLOADPS);
> + def_builtin_pure (OPTION_MASK_ISA_AVX, 0, "__builtin_ia32_maskloadpd256",
> + V4DF_FTYPE_PCV4DF_V4DI, IX86_BUILTIN_MASKLOADPD256);
> + def_builtin_pure (OPTION_MASK_ISA_AVX, 0, "__builtin_ia32_maskloadps256",
> + V8SF_FTYPE_PCV8SF_V8SI, IX86_BUILTIN_MASKLOADPS256);
> +
> /* AVX2 */
> + def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_maskloadd",
> + V4SI_FTYPE_PCV4SI_V4SI, IX86_BUILTIN_MASKLOADD);
> + def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_maskloadq",
> + V2DI_FTYPE_PCV2DI_V2DI, IX86_BUILTIN_MASKLOADQ);
> + def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_maskloadd256",
> + V8SI_FTYPE_PCV8SI_V8SI, IX86_BUILTIN_MASKLOADD256);
> + def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_maskloadq256",
> + V4DI_FTYPE_PCV4DI_V4DI, IX86_BUILTIN_MASKLOADQ256);
> +
> def_builtin_pure (OPTION_MASK_ISA_AVX2, 0, "__builtin_ia32_gathersiv2df",
> V2DF_FTYPE_V2DF_PCDOUBLE_V4SI_V2DF_INT,
> IX86_BUILTIN_GATHERSIV2DF);
> diff --git a/gcc/config/i386/i386-builtins.h b/gcc/config/i386/i386-builtins.h
> index 0641808c7a7..6a5e249e382 100644
> --- a/gcc/config/i386/i386-builtins.h
> +++ b/gcc/config/i386/i386-builtins.h
> @@ -60,6 +60,14 @@ enum ix86_builtins
> IX86_BUILTIN_VEC_SET_V8HI,
> IX86_BUILTIN_VEC_SET_V4HI,
> IX86_BUILTIN_VEC_SET_V16QI,
> + IX86_BUILTIN_MASKLOADPD,
> + IX86_BUILTIN_MASKLOADPS,
> + IX86_BUILTIN_MASKLOADPD256,
> + IX86_BUILTIN_MASKLOADPS256,
> + IX86_BUILTIN_MASKLOADD,
> + IX86_BUILTIN_MASKLOADQ,
> + IX86_BUILTIN_MASKLOADD256,
> + IX86_BUILTIN_MASKLOADQ256,
> IX86_BUILTIN_GATHERSIV2DF,
> IX86_BUILTIN_GATHERSIV4DF,
> IX86_BUILTIN_GATHERDIV2DF,
> --
> 2.26.2
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] target/100312 - make x86 masked load builtins pure
2021-04-28 11:23 ` Uros Bizjak
@ 2021-04-28 11:46 ` Richard Biener
2021-04-28 12:11 ` Uros Bizjak
0 siblings, 1 reply; 8+ messages in thread
From: Richard Biener @ 2021-04-28 11:46 UTC (permalink / raw)
To: Uros Bizjak; +Cc: gcc-patches, Jakub Jelinek
On Wed, 28 Apr 2021, Uros Bizjak wrote:
> On Wed, Apr 28, 2021 at 1:02 PM Richard Biener <rguenther@suse.de> wrote:
> >
> > This arranges for the x86 AVX and AVX2 masked load builtins to be
> > pure to enable dead code elimination and more appropriate alias
> > analysis.
> >
> > Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
> >
> > Thanks,
> > Richard.
> >
> > 2021-04-28 Richard Biener <rguenther@suse.de>
> >
> > PR target/100312
> > * config/i386/i386-builtin.def (IX86_BUILTIN_MASKLOADPD,
> > IX86_BUILTIN_MASKLOADPS, IX86_BUILTIN_MASKLOADPD256,
> > IX86_BUILTIN_MASKLOADPS256, IX86_BUILTIN_MASKLOADD,
> > IX86_BUILTIN_MASKLOADQ, IX86_BUILTIN_MASKLOADD256,
> > IX86_BUILTIN_MASKLOADQ256): Remove.
> > * config/i386/i386-builtins.h (ix86_builtins): Add entries for
> > IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256.
> > * config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
> > Initialize IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
> > as pure builtins.
>
> OK.
Whoops - somehow I posted the wrong version. The version posted
ICEs because I failed to realize I'd have to explicitely handle
those builtins in ix86_expand_builtin. What bootstrapped and
tested OK is the version below.
Is that version also OK?
Thanks,
Richard.
From 44fb1b12c4b5ec649804d8ea40ebc087723ccc7f Mon Sep 17 00:00:00 2001
From: Richard Biener <rguenther@suse.de>
Date: Wed, 28 Apr 2021 12:09:03 +0200
Subject: [PATCH] target/100312 - make x86 masked load builtins pure
To: gcc-patches@gcc.gnu.org
This arranges for the x86 AVX and AVX2 masked load builtins to be
pure to enable dead code elimination and more appropriate alias
analysis.
2021-04-28 Richard Biener <rguenther@suse.de>
PR target/100312
* config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
Arrange for IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
to be defined as pure builtins.
---
gcc/config/i386/i386-builtins.c | 12 +++++++++++-
1 file changed, 11 insertions(+), 1 deletion(-)
diff --git a/gcc/config/i386/i386-builtins.c b/gcc/config/i386/i386-builtins.c
index 128bd39816c..fa05cd6cf65 100644
--- a/gcc/config/i386/i386-builtins.c
+++ b/gcc/config/i386/i386-builtins.c
@@ -521,7 +521,17 @@ ix86_init_mmx_sse_builtins (void)
continue;
ftype = (enum ix86_builtin_func_type) d->flag;
- def_builtin (d->mask, d->mask2, d->name, ftype, d->code);
+ if (d->code == IX86_BUILTIN_MASKLOADPD
+ || d->code == IX86_BUILTIN_MASKLOADPS
+ || d->code == IX86_BUILTIN_MASKLOADPD256
+ || d->code == IX86_BUILTIN_MASKLOADPS256
+ || d->code == IX86_BUILTIN_MASKLOADD
+ || d->code == IX86_BUILTIN_MASKLOADQ
+ || d->code == IX86_BUILTIN_MASKLOADD256
+ || d->code == IX86_BUILTIN_MASKLOADQ256)
+ def_builtin_pure (d->mask, d->mask2, d->name, ftype, d->code);
+ else
+ def_builtin (d->mask, d->mask2, d->name, ftype, d->code);
}
BDESC_VERIFYS (IX86_BUILTIN__BDESC_SPECIAL_ARGS_LAST,
IX86_BUILTIN__BDESC_SPECIAL_ARGS_FIRST,
--
2.26.2
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] target/100312 - make x86 masked load builtins pure
2021-04-28 11:46 ` Richard Biener
@ 2021-04-28 12:11 ` Uros Bizjak
2021-04-28 13:32 ` Richard Biener
0 siblings, 1 reply; 8+ messages in thread
From: Uros Bizjak @ 2021-04-28 12:11 UTC (permalink / raw)
To: Richard Biener; +Cc: gcc-patches, Jakub Jelinek
On Wed, Apr 28, 2021 at 1:46 PM Richard Biener <rguenther@suse.de> wrote:
>
> On Wed, 28 Apr 2021, Uros Bizjak wrote:
>
> > On Wed, Apr 28, 2021 at 1:02 PM Richard Biener <rguenther@suse.de> wrote:
> > >
> > > This arranges for the x86 AVX and AVX2 masked load builtins to be
> > > pure to enable dead code elimination and more appropriate alias
> > > analysis.
> > >
> > > Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
> > >
> > > Thanks,
> > > Richard.
> > >
> > > 2021-04-28 Richard Biener <rguenther@suse.de>
> > >
> > > PR target/100312
> > > * config/i386/i386-builtin.def (IX86_BUILTIN_MASKLOADPD,
> > > IX86_BUILTIN_MASKLOADPS, IX86_BUILTIN_MASKLOADPD256,
> > > IX86_BUILTIN_MASKLOADPS256, IX86_BUILTIN_MASKLOADD,
> > > IX86_BUILTIN_MASKLOADQ, IX86_BUILTIN_MASKLOADD256,
> > > IX86_BUILTIN_MASKLOADQ256): Remove.
> > > * config/i386/i386-builtins.h (ix86_builtins): Add entries for
> > > IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256.
> > > * config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
> > > Initialize IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
> > > as pure builtins.
> >
> > OK.
>
> Whoops - somehow I posted the wrong version. The version posted
> ICEs because I failed to realize I'd have to explicitely handle
> those builtins in ix86_expand_builtin. What bootstrapped and
> tested OK is the version below.
>
> Is that version also OK?
Hm, your previous version removed mentioned builtins from builtins.def
and initialized them in ix86_init_mmx_sse_builtins by hand (this seems
to be the preferred way). This should work, so I really don't see why
the previous approach would result in ICE.
Uros.
>
> Thanks,
> Richard.
>
> From 44fb1b12c4b5ec649804d8ea40ebc087723ccc7f Mon Sep 17 00:00:00 2001
> From: Richard Biener <rguenther@suse.de>
> Date: Wed, 28 Apr 2021 12:09:03 +0200
> Subject: [PATCH] target/100312 - make x86 masked load builtins pure
> To: gcc-patches@gcc.gnu.org
>
> This arranges for the x86 AVX and AVX2 masked load builtins to be
> pure to enable dead code elimination and more appropriate alias
> analysis.
>
> 2021-04-28 Richard Biener <rguenther@suse.de>
>
> PR target/100312
> * config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
> Arrange for IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
> to be defined as pure builtins.
> ---
> gcc/config/i386/i386-builtins.c | 12 +++++++++++-
> 1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/gcc/config/i386/i386-builtins.c b/gcc/config/i386/i386-builtins.c
> index 128bd39816c..fa05cd6cf65 100644
> --- a/gcc/config/i386/i386-builtins.c
> +++ b/gcc/config/i386/i386-builtins.c
> @@ -521,7 +521,17 @@ ix86_init_mmx_sse_builtins (void)
> continue;
>
> ftype = (enum ix86_builtin_func_type) d->flag;
> - def_builtin (d->mask, d->mask2, d->name, ftype, d->code);
> + if (d->code == IX86_BUILTIN_MASKLOADPD
> + || d->code == IX86_BUILTIN_MASKLOADPS
> + || d->code == IX86_BUILTIN_MASKLOADPD256
> + || d->code == IX86_BUILTIN_MASKLOADPS256
> + || d->code == IX86_BUILTIN_MASKLOADD
> + || d->code == IX86_BUILTIN_MASKLOADQ
> + || d->code == IX86_BUILTIN_MASKLOADD256
> + || d->code == IX86_BUILTIN_MASKLOADQ256)
> + def_builtin_pure (d->mask, d->mask2, d->name, ftype, d->code);
> + else
> + def_builtin (d->mask, d->mask2, d->name, ftype, d->code);
> }
> BDESC_VERIFYS (IX86_BUILTIN__BDESC_SPECIAL_ARGS_LAST,
> IX86_BUILTIN__BDESC_SPECIAL_ARGS_FIRST,
> --
> 2.26.2
>
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] target/100312 - make x86 masked load builtins pure
2021-04-28 12:11 ` Uros Bizjak
@ 2021-04-28 13:32 ` Richard Biener
2021-04-28 15:07 ` Uros Bizjak
0 siblings, 1 reply; 8+ messages in thread
From: Richard Biener @ 2021-04-28 13:32 UTC (permalink / raw)
To: Uros Bizjak; +Cc: gcc-patches, Jakub Jelinek
On Wed, 28 Apr 2021, Uros Bizjak wrote:
> On Wed, Apr 28, 2021 at 1:46 PM Richard Biener <rguenther@suse.de> wrote:
> >
> > On Wed, 28 Apr 2021, Uros Bizjak wrote:
> >
> > > On Wed, Apr 28, 2021 at 1:02 PM Richard Biener <rguenther@suse.de> wrote:
> > > >
> > > > This arranges for the x86 AVX and AVX2 masked load builtins to be
> > > > pure to enable dead code elimination and more appropriate alias
> > > > analysis.
> > > >
> > > > Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
> > > >
> > > > Thanks,
> > > > Richard.
> > > >
> > > > 2021-04-28 Richard Biener <rguenther@suse.de>
> > > >
> > > > PR target/100312
> > > > * config/i386/i386-builtin.def (IX86_BUILTIN_MASKLOADPD,
> > > > IX86_BUILTIN_MASKLOADPS, IX86_BUILTIN_MASKLOADPD256,
> > > > IX86_BUILTIN_MASKLOADPS256, IX86_BUILTIN_MASKLOADD,
> > > > IX86_BUILTIN_MASKLOADQ, IX86_BUILTIN_MASKLOADD256,
> > > > IX86_BUILTIN_MASKLOADQ256): Remove.
> > > > * config/i386/i386-builtins.h (ix86_builtins): Add entries for
> > > > IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > > > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > > > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > > > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256.
> > > > * config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
> > > > Initialize IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > > > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > > > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > > > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
> > > > as pure builtins.
> > >
> > > OK.
> >
> > Whoops - somehow I posted the wrong version. The version posted
> > ICEs because I failed to realize I'd have to explicitely handle
> > those builtins in ix86_expand_builtin. What bootstrapped and
> > tested OK is the version below.
> >
> > Is that version also OK?
>
> Hm, your previous version removed mentioned builtins from builtins.def
> and initialized them in ix86_init_mmx_sse_builtins by hand (this seems
> to be the preferred way). This should work, so I really don't see why
> the previous approach would result in ICE.
It eventually runs into the gcc_unreachable () at the very end of
ix86_expand_builtin since IX86_BUILTIN_MASKLOADD and friends are
no longer in the range of IX86_BUILTIN__BDESC_SPECIAL_ARGS_FIRST
to IX86_BUILTIN__BDESC_SPECIAL_ARGS_LAST.
Richard.
> Uros.
>
> >
> > Thanks,
> > Richard.
> >
> > From 44fb1b12c4b5ec649804d8ea40ebc087723ccc7f Mon Sep 17 00:00:00 2001
> > From: Richard Biener <rguenther@suse.de>
> > Date: Wed, 28 Apr 2021 12:09:03 +0200
> > Subject: [PATCH] target/100312 - make x86 masked load builtins pure
> > To: gcc-patches@gcc.gnu.org
> >
> > This arranges for the x86 AVX and AVX2 masked load builtins to be
> > pure to enable dead code elimination and more appropriate alias
> > analysis.
> >
> > 2021-04-28 Richard Biener <rguenther@suse.de>
> >
> > PR target/100312
> > * config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
> > Arrange for IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
> > to be defined as pure builtins.
> > ---
> > gcc/config/i386/i386-builtins.c | 12 +++++++++++-
> > 1 file changed, 11 insertions(+), 1 deletion(-)
> >
> > diff --git a/gcc/config/i386/i386-builtins.c b/gcc/config/i386/i386-builtins.c
> > index 128bd39816c..fa05cd6cf65 100644
> > --- a/gcc/config/i386/i386-builtins.c
> > +++ b/gcc/config/i386/i386-builtins.c
> > @@ -521,7 +521,17 @@ ix86_init_mmx_sse_builtins (void)
> > continue;
> >
> > ftype = (enum ix86_builtin_func_type) d->flag;
> > - def_builtin (d->mask, d->mask2, d->name, ftype, d->code);
> > + if (d->code == IX86_BUILTIN_MASKLOADPD
> > + || d->code == IX86_BUILTIN_MASKLOADPS
> > + || d->code == IX86_BUILTIN_MASKLOADPD256
> > + || d->code == IX86_BUILTIN_MASKLOADPS256
> > + || d->code == IX86_BUILTIN_MASKLOADD
> > + || d->code == IX86_BUILTIN_MASKLOADQ
> > + || d->code == IX86_BUILTIN_MASKLOADD256
> > + || d->code == IX86_BUILTIN_MASKLOADQ256)
> > + def_builtin_pure (d->mask, d->mask2, d->name, ftype, d->code);
> > + else
> > + def_builtin (d->mask, d->mask2, d->name, ftype, d->code);
> > }
> > BDESC_VERIFYS (IX86_BUILTIN__BDESC_SPECIAL_ARGS_LAST,
> > IX86_BUILTIN__BDESC_SPECIAL_ARGS_FIRST,
> > --
> > 2.26.2
> >
>
--
Richard Biener <rguenther@suse.de>
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Felix Imendörffer; HRB 36809 (AG Nuernberg)
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] target/100312 - make x86 masked load builtins pure
2021-04-28 13:32 ` Richard Biener
@ 2021-04-28 15:07 ` Uros Bizjak
2021-04-29 7:25 ` Richard Biener
0 siblings, 1 reply; 8+ messages in thread
From: Uros Bizjak @ 2021-04-28 15:07 UTC (permalink / raw)
To: Richard Biener; +Cc: gcc-patches, Jakub Jelinek
[-- Attachment #1: Type: text/plain, Size: 2765 bytes --]
On Wed, Apr 28, 2021 at 3:32 PM Richard Biener <rguenther@suse.de> wrote:
>
> On Wed, 28 Apr 2021, Uros Bizjak wrote:
>
> > On Wed, Apr 28, 2021 at 1:46 PM Richard Biener <rguenther@suse.de> wrote:
> > >
> > > On Wed, 28 Apr 2021, Uros Bizjak wrote:
> > >
> > > > On Wed, Apr 28, 2021 at 1:02 PM Richard Biener <rguenther@suse.de> wrote:
> > > > >
> > > > > This arranges for the x86 AVX and AVX2 masked load builtins to be
> > > > > pure to enable dead code elimination and more appropriate alias
> > > > > analysis.
> > > > >
> > > > > Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
> > > > >
> > > > > Thanks,
> > > > > Richard.
> > > > >
> > > > > 2021-04-28 Richard Biener <rguenther@suse.de>
> > > > >
> > > > > PR target/100312
> > > > > * config/i386/i386-builtin.def (IX86_BUILTIN_MASKLOADPD,
> > > > > IX86_BUILTIN_MASKLOADPS, IX86_BUILTIN_MASKLOADPD256,
> > > > > IX86_BUILTIN_MASKLOADPS256, IX86_BUILTIN_MASKLOADD,
> > > > > IX86_BUILTIN_MASKLOADQ, IX86_BUILTIN_MASKLOADD256,
> > > > > IX86_BUILTIN_MASKLOADQ256): Remove.
> > > > > * config/i386/i386-builtins.h (ix86_builtins): Add entries for
> > > > > IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > > > > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > > > > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > > > > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256.
> > > > > * config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
> > > > > Initialize IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > > > > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > > > > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > > > > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
> > > > > as pure builtins.
> > > >
> > > > OK.
> > >
> > > Whoops - somehow I posted the wrong version. The version posted
> > > ICEs because I failed to realize I'd have to explicitely handle
> > > those builtins in ix86_expand_builtin. What bootstrapped and
> > > tested OK is the version below.
> > >
> > > Is that version also OK?
> >
> > Hm, your previous version removed mentioned builtins from builtins.def
> > and initialized them in ix86_init_mmx_sse_builtins by hand (this seems
> > to be the preferred way). This should work, so I really don't see why
> > the previous approach would result in ICE.
>
> It eventually runs into the gcc_unreachable () at the very end of
> ix86_expand_builtin since IX86_BUILTIN_MASKLOADD and friends are
> no longer in the range of IX86_BUILTIN__BDESC_SPECIAL_ARGS_FIRST
> to IX86_BUILTIN__BDESC_SPECIAL_ARGS_LAST.
How about the attached (untested) patch then?
Uros.
[-- Attachment #2: p.diff.txt --]
[-- Type: text/plain, Size: 7326 bytes --]
diff --git a/gcc/config/i386/i386-builtin.def b/gcc/config/i386/i386-builtin.def
index 4dbd4f23647..80c2a2c0294 100644
--- a/gcc/config/i386/i386-builtin.def
+++ b/gcc/config/i386/i386-builtin.def
@@ -187,10 +187,6 @@ BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_movntv4di, "__builtin_ia32_movntdq25
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_movntv4df, "__builtin_ia32_movntpd256", IX86_BUILTIN_MOVNTPD256, UNKNOWN, (int) VOID_FTYPE_PDOUBLE_V4DF)
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_movntv8sf, "__builtin_ia32_movntps256", IX86_BUILTIN_MOVNTPS256, UNKNOWN, (int) VOID_FTYPE_PFLOAT_V8SF)
-BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadpd, "__builtin_ia32_maskloadpd", IX86_BUILTIN_MASKLOADPD, UNKNOWN, (int) V2DF_FTYPE_PCV2DF_V2DI)
-BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadps, "__builtin_ia32_maskloadps", IX86_BUILTIN_MASKLOADPS, UNKNOWN, (int) V4SF_FTYPE_PCV4SF_V4SI)
-BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadpd256, "__builtin_ia32_maskloadpd256", IX86_BUILTIN_MASKLOADPD256, UNKNOWN, (int) V4DF_FTYPE_PCV4DF_V4DI)
-BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadps256, "__builtin_ia32_maskloadps256", IX86_BUILTIN_MASKLOADPS256, UNKNOWN, (int) V8SF_FTYPE_PCV8SF_V8SI)
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstorepd, "__builtin_ia32_maskstorepd", IX86_BUILTIN_MASKSTOREPD, UNKNOWN, (int) VOID_FTYPE_PV2DF_V2DI_V2DF)
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstoreps, "__builtin_ia32_maskstoreps", IX86_BUILTIN_MASKSTOREPS, UNKNOWN, (int) VOID_FTYPE_PV4SF_V4SI_V4SF)
BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstorepd256, "__builtin_ia32_maskstorepd256", IX86_BUILTIN_MASKSTOREPD256, UNKNOWN, (int) VOID_FTYPE_PV4DF_V4DI_V4DF)
@@ -198,10 +194,6 @@ BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskstoreps256, "__builtin_ia32_mask
/* AVX2 */
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_movntdqa, "__builtin_ia32_movntdqa256", IX86_BUILTIN_MOVNTDQA256, UNKNOWN, (int) V4DI_FTYPE_PV4DI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadd, "__builtin_ia32_maskloadd", IX86_BUILTIN_MASKLOADD, UNKNOWN, (int) V4SI_FTYPE_PCV4SI_V4SI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadq, "__builtin_ia32_maskloadq", IX86_BUILTIN_MASKLOADQ, UNKNOWN, (int) V2DI_FTYPE_PCV2DI_V2DI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadd256, "__builtin_ia32_maskloadd256", IX86_BUILTIN_MASKLOADD256, UNKNOWN, (int) V8SI_FTYPE_PCV8SI_V8SI)
-BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadq256, "__builtin_ia32_maskloadq256", IX86_BUILTIN_MASKLOADQ256, UNKNOWN, (int) V4DI_FTYPE_PCV4DI_V4DI)
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskstored, "__builtin_ia32_maskstored", IX86_BUILTIN_MASKSTORED, UNKNOWN, (int) VOID_FTYPE_PV4SI_V4SI_V4SI)
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskstoreq, "__builtin_ia32_maskstoreq", IX86_BUILTIN_MASKSTOREQ, UNKNOWN, (int) VOID_FTYPE_PV2DI_V2DI_V2DI)
BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskstored256, "__builtin_ia32_maskstored256", IX86_BUILTIN_MASKSTORED256, UNKNOWN, (int) VOID_FTYPE_PV8SI_V8SI_V8SI)
@@ -473,7 +465,22 @@ BDESC (0, OPTION_MASK_ISA2_WIDEKL, CODE_FOR_nothing, "__builtin_ia32_aesdecwide2
BDESC (0, OPTION_MASK_ISA2_WIDEKL, CODE_FOR_nothing, "__builtin_ia32_aesencwide128kl_u8", IX86_BUILTIN_AESENCWIDE128KLU8, UNKNOWN, (int) UINT8_FTYPE_PV2DI_PCV2DI_PCVOID)
BDESC (0, OPTION_MASK_ISA2_WIDEKL, CODE_FOR_nothing, "__builtin_ia32_aesencwide256kl_u8", IX86_BUILTIN_AESENCWIDE256KLU8, UNKNOWN, (int) UINT8_FTYPE_PV2DI_PCV2DI_PCVOID)
-BDESC_END (SPECIAL_ARGS, ARGS)
+BDESC_END (SPECIAL_ARGS, PURE_ARGS)
+
+/* AVX */
+BDESC_FIRST (pure_args, PURE_ARGS,
+ OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadpd, "__builtin_ia32_maskloadpd", IX86_BUILTIN_MASKLOADPD, UNKNOWN, (int) V2DF_FTYPE_PCV2DF_V2DI)
+BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadps, "__builtin_ia32_maskloadps", IX86_BUILTIN_MASKLOADPS, UNKNOWN, (int) V4SF_FTYPE_PCV4SF_V4SI)
+BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadpd256, "__builtin_ia32_maskloadpd256", IX86_BUILTIN_MASKLOADPD256, UNKNOWN, (int) V4DF_FTYPE_PCV4DF_V4DI)
+BDESC (OPTION_MASK_ISA_AVX, 0, CODE_FOR_avx_maskloadps256, "__builtin_ia32_maskloadps256", IX86_BUILTIN_MASKLOADPS256, UNKNOWN, (int) V8SF_FTYPE_PCV8SF_V8SI)
+
+/* AVX2 */
+BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadd, "__builtin_ia32_maskloadd", IX86_BUILTIN_MASKLOADD, UNKNOWN, (int) V4SI_FTYPE_PCV4SI_V4SI)
+BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadq, "__builtin_ia32_maskloadq", IX86_BUILTIN_MASKLOADQ, UNKNOWN, (int) V2DI_FTYPE_PCV2DI_V2DI)
+BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadd256, "__builtin_ia32_maskloadd256", IX86_BUILTIN_MASKLOADD256, UNKNOWN, (int) V8SI_FTYPE_PCV8SI_V8SI)
+BDESC (OPTION_MASK_ISA_AVX2, 0, CODE_FOR_avx2_maskloadq256, "__builtin_ia32_maskloadq256", IX86_BUILTIN_MASKLOADQ256, UNKNOWN, (int) V4DI_FTYPE_PCV4DI_V4DI)
+
+BDESC_END (PURE_ARGS, ARGS)
/* Builtins with variable number of arguments. */
BDESC_FIRST (args, ARGS,
diff --git a/gcc/config/i386/i386-builtins.c b/gcc/config/i386/i386-builtins.c
index 128bd39816c..b66911082ab 100644
--- a/gcc/config/i386/i386-builtins.c
+++ b/gcc/config/i386/i386-builtins.c
@@ -108,8 +108,10 @@ BDESC_VERIFYS (IX86_BUILTIN__BDESC_PCMPISTR_FIRST,
IX86_BUILTIN__BDESC_PCMPESTR_LAST, 1);
BDESC_VERIFYS (IX86_BUILTIN__BDESC_SPECIAL_ARGS_FIRST,
IX86_BUILTIN__BDESC_PCMPISTR_LAST, 1);
-BDESC_VERIFYS (IX86_BUILTIN__BDESC_ARGS_FIRST,
+BDESC_VERIFYS (IX86_BUILTIN__BDESC_PURE_ARGS_FIRST,
IX86_BUILTIN__BDESC_SPECIAL_ARGS_LAST, 1);
+BDESC_VERIFYS (IX86_BUILTIN__BDESC_ARGS_FIRST,
+ IX86_BUILTIN__BDESC_PURE_ARGS_LAST, 1);
BDESC_VERIFYS (IX86_BUILTIN__BDESC_ROUND_ARGS_FIRST,
IX86_BUILTIN__BDESC_ARGS_LAST, 1);
BDESC_VERIFYS (IX86_BUILTIN__BDESC_MULTI_ARG_FIRST,
@@ -527,7 +529,23 @@ ix86_init_mmx_sse_builtins (void)
IX86_BUILTIN__BDESC_SPECIAL_ARGS_FIRST,
ARRAY_SIZE (bdesc_special_args) - 1);
- /* Add all builtins with variable number of operands. */
+ /* Add all pure builtins with variable number of operands. */
+ for (i = 0, d = bdesc_pure_args;
+ i < ARRAY_SIZE (bdesc_pure_args);
+ i++, d++)
+ {
+ BDESC_VERIFY (d->code, IX86_BUILTIN__BDESC_PURE_ARGS_FIRST, i);
+ if (d->name == 0)
+ continue;
+
+ ftype = (enum ix86_builtin_func_type) d->flag;
+ def_builtin_pure (d->mask, d->mask2, d->name, ftype, d->code);
+ }
+ BDESC_VERIFYS (IX86_BUILTIN__BDESC_PURE_ARGS_LAST,
+ IX86_BUILTIN__BDESC_PURE_ARGS_FIRST,
+ ARRAY_SIZE (bdesc_pure_args) - 1);
+
+ /* Add all const builtins with variable number of operands. */
for (i = 0, d = bdesc_args;
i < ARRAY_SIZE (bdesc_args);
i++, d++)
diff --git a/gcc/config/i386/i386-expand.c b/gcc/config/i386/i386-expand.c
index 516440eb5c1..76990628150 100644
--- a/gcc/config/i386/i386-expand.c
+++ b/gcc/config/i386/i386-expand.c
@@ -13240,6 +13240,14 @@ rdseed_step:
target);
}
+ if (fcode >= IX86_BUILTIN__BDESC_PURE_ARGS_FIRST
+ && fcode <= IX86_BUILTIN__BDESC_PURE_ARGS_LAST)
+ {
+ i = fcode - IX86_BUILTIN__BDESC_PURE_ARGS_FIRST;
+ return ix86_expand_special_args_builtin (bdesc_pure_args + i, exp,
+ target);
+ }
+
if (fcode >= IX86_BUILTIN__BDESC_ARGS_FIRST
&& fcode <= IX86_BUILTIN__BDESC_ARGS_LAST)
{
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] target/100312 - make x86 masked load builtins pure
2021-04-28 15:07 ` Uros Bizjak
@ 2021-04-29 7:25 ` Richard Biener
2021-04-29 14:48 ` Uros Bizjak
0 siblings, 1 reply; 8+ messages in thread
From: Richard Biener @ 2021-04-29 7:25 UTC (permalink / raw)
To: Uros Bizjak; +Cc: gcc-patches, Jakub Jelinek
On Wed, 28 Apr 2021, Uros Bizjak wrote:
> On Wed, Apr 28, 2021 at 3:32 PM Richard Biener <rguenther@suse.de> wrote:
> >
> > On Wed, 28 Apr 2021, Uros Bizjak wrote:
> >
> > > On Wed, Apr 28, 2021 at 1:46 PM Richard Biener <rguenther@suse.de> wrote:
> > > >
> > > > On Wed, 28 Apr 2021, Uros Bizjak wrote:
> > > >
> > > > > On Wed, Apr 28, 2021 at 1:02 PM Richard Biener <rguenther@suse.de> wrote:
> > > > > >
> > > > > > This arranges for the x86 AVX and AVX2 masked load builtins to be
> > > > > > pure to enable dead code elimination and more appropriate alias
> > > > > > analysis.
> > > > > >
> > > > > > Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk?
> > > > > >
> > > > > > Thanks,
> > > > > > Richard.
> > > > > >
> > > > > > 2021-04-28 Richard Biener <rguenther@suse.de>
> > > > > >
> > > > > > PR target/100312
> > > > > > * config/i386/i386-builtin.def (IX86_BUILTIN_MASKLOADPD,
> > > > > > IX86_BUILTIN_MASKLOADPS, IX86_BUILTIN_MASKLOADPD256,
> > > > > > IX86_BUILTIN_MASKLOADPS256, IX86_BUILTIN_MASKLOADD,
> > > > > > IX86_BUILTIN_MASKLOADQ, IX86_BUILTIN_MASKLOADD256,
> > > > > > IX86_BUILTIN_MASKLOADQ256): Remove.
> > > > > > * config/i386/i386-builtins.h (ix86_builtins): Add entries for
> > > > > > IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > > > > > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > > > > > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > > > > > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256.
> > > > > > * config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
> > > > > > Initialize IX86_BUILTIN_MASKLOADPD, IX86_BUILTIN_MASKLOADPS,
> > > > > > IX86_BUILTIN_MASKLOADPD256, IX86_BUILTIN_MASKLOADPS256,
> > > > > > IX86_BUILTIN_MASKLOADD, IX86_BUILTIN_MASKLOADQ,
> > > > > > IX86_BUILTIN_MASKLOADD256, and IX86_BUILTIN_MASKLOADQ256
> > > > > > as pure builtins.
> > > > >
> > > > > OK.
> > > >
> > > > Whoops - somehow I posted the wrong version. The version posted
> > > > ICEs because I failed to realize I'd have to explicitely handle
> > > > those builtins in ix86_expand_builtin. What bootstrapped and
> > > > tested OK is the version below.
> > > >
> > > > Is that version also OK?
> > >
> > > Hm, your previous version removed mentioned builtins from builtins.def
> > > and initialized them in ix86_init_mmx_sse_builtins by hand (this seems
> > > to be the preferred way). This should work, so I really don't see why
> > > the previous approach would result in ICE.
> >
> > It eventually runs into the gcc_unreachable () at the very end of
> > ix86_expand_builtin since IX86_BUILTIN_MASKLOADD and friends are
> > no longer in the range of IX86_BUILTIN__BDESC_SPECIAL_ARGS_FIRST
> > to IX86_BUILTIN__BDESC_SPECIAL_ARGS_LAST.
>
> How about the attached (untested) patch then?
Ah yeah - good idea.
The patch works, bootstrapped and tested on x86_64-unknown-linux-gnu.
As expected it reduces the number of masked load builtins for the
PR99912 testcase from ~800 to ~350 (at the GIMPLE level).
Thanks,
Richard.
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH] target/100312 - make x86 masked load builtins pure
2021-04-29 7:25 ` Richard Biener
@ 2021-04-29 14:48 ` Uros Bizjak
0 siblings, 0 replies; 8+ messages in thread
From: Uros Bizjak @ 2021-04-29 14:48 UTC (permalink / raw)
To: Richard Biener; +Cc: gcc-patches, Jakub Jelinek
On Thu, Apr 29, 2021 at 9:25 AM Richard Biener <rguenther@suse.de> wrote:
> > > It eventually runs into the gcc_unreachable () at the very end of
> > > ix86_expand_builtin since IX86_BUILTIN_MASKLOADD and friends are
> > > no longer in the range of IX86_BUILTIN__BDESC_SPECIAL_ARGS_FIRST
> > > to IX86_BUILTIN__BDESC_SPECIAL_ARGS_LAST.
> >
> > How about the attached (untested) patch then?
>
> Ah yeah - good idea.
>
> The patch works, bootstrapped and tested on x86_64-unknown-linux-gnu.
>
> As expected it reduces the number of masked load builtins for the
> PR99912 testcase from ~800 to ~350 (at the GIMPLE level).
Thanks, pushed with the following ChangeLog:
i386: Mark x86 masked load builtins pure [PR100312]
Mark x86 AVX and AVX2 masked load builtins pure to enable dead code
elimination and more appropriate alias analysis.
2021-04-29 Uroš Bizjak <ubizjak@gmail.com>
Richard Biener <rguenther@suse.de>
gcc/
PR target/100312
* config/i386/i386-builtin.def (IX86_BUILTIN_MASKLOADPD)
(IX86_BUILTIN_MASKLOADPS, IX86_BUILTIN_MASKLOADPD256)
(IX86_BUILTIN_MASKLOADPS256, IX86_BUILTIN_MASKLOADD)
(IX86_BUILTIN_MASKLOADQ, IX86_BUILTIN_MASKLOADD256)
(IX86_BUILTIN_MASKLOADQ256): Move from SPECIAL_ARGS
to PURE_ARGS category.
* config/i386/i386-builtins.c (ix86_init_mmx_sse_builtins):
Handle PURE_ARGS category.
* config/i386/i386-expand.c (ix86_expand_builtin): Ditto.
Uros.
> Thanks,
> Richard.
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2021-04-29 14:48 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-04-28 11:02 [PATCH] target/100312 - make x86 masked load builtins pure Richard Biener
2021-04-28 11:23 ` Uros Bizjak
2021-04-28 11:46 ` Richard Biener
2021-04-28 12:11 ` Uros Bizjak
2021-04-28 13:32 ` Richard Biener
2021-04-28 15:07 ` Uros Bizjak
2021-04-29 7:25 ` Richard Biener
2021-04-29 14:48 ` 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).