[x86] PR target/106060: Improved SSE vector constant materialization.
Checks
| Context |
Check |
Description |
| linaro-tcwg-bot/tcwg_gcc_build--master-aarch64 |
success
|
Testing passed
|
| linaro-tcwg-bot/tcwg_gcc_build--master-arm |
success
|
Testing passed
|
| linaro-tcwg-bot/tcwg_gcc_check--master-aarch64 |
success
|
Testing passed
|
| linaro-tcwg-bot/tcwg_gcc_check--master-arm |
success
|
Testing passed
|
Commit Message
I thought I'd just missed the bug fixing season of stage3, but there
appears to a little latitude in early stage4 (for vector patches), so
I'll post this now.
This patch resolves PR target/106060 by providing efficient methods for
materializing/synthesizing special "vector" constants on x86. Currently
there are three methods of materializing a vector constant; the most
general is to load a vector from the constant pool, secondly "duplicated"
constants can be synthesized by moving an integer between units and
broadcasting (or shuffling it), and finally the special cases of the
all-zeros vector and all-ones vectors can be loaded via a single SSE
instruction. This patch handles additional cases that can be synthesized
in two instructions, loading an all-ones vector followed by another SSE
instruction. Following my recent patch for PR target/112992, there's
conveniently a single place in i386-expand.cc where these special cases
can be handled.
Two examples are given in the original bugzilla PR for 106060.
__m256i
should_be_cmpeq_abs ()
{
return _mm256_set1_epi8 (1);
}
is now generated (with -O3 -march=x86-64-v3) as:
vpcmpeqd %ymm0, %ymm0, %ymm0
vpabsb %ymm0, %ymm0
ret
and
__m256i
should_be_cmpeq_add ()
{
return _mm256_set1_epi8 (-2);
}
is now generated as:
vpcmpeqd %ymm0, %ymm0, %ymm0
vpaddb %ymm0, %ymm0, %ymm0
ret
This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
and make -k check, both with and without --target_board=unix{-m32}
with no new failures. Ok for mainline?
2024-01-16 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
PR target/106060
* config/i386/i386-expand.cc (enum ix86_vec_bcast_alg): New.
(struct ix86_vec_bcast_map_simode_t): New type for table below.
(ix86_vec_bcast_map_simode): Table of SImode constants that may
be efficiently synthesized by a ix86_vec_bcast_alg method.
(ix86_vec_bcast_map_simode_cmp): New comparator for bsearch.
(ix86_vector_duplicate_simode_const): Efficiently synthesize
V4SImode and V8SImode constants that duplicate special constants.
(ix86_vector_duplicate_value): Attempt to synthesize "special"
vector constants using ix86_vector_duplicate_simode_const.
* config/i386/i386.cc (ix86_rtx_costs) <case ABS>: ABS of a
vector integer mode costs with a single SSE instruction.
gcc/testsuite/ChangeLog
PR target/106060
* gcc.target/i386/auto-init-8.c: Update test case.
* gcc.target/i386/avx512fp16-3.c: Likewise.
* gcc.target/i386/pr100865-9a.c: Likewise.
* gcc.target/i386/pr106060-1.c: New test case.
* gcc.target/i386/pr106060-2.c: Likewise.
* gcc.target/i386/pr106060-3.c: Likewise.
* gcc.target/i386/pr70314-3.c: Update test case.
* gcc.target/i386/vect-shiftv4qi.c: Likewise.
* gcc.target/i386/vect-shiftv8qi.c: Likewise.
Thanks in advance,
Roger
--
Comments
On Wed, Jan 17, 2024 at 5:59 AM Roger Sayle <roger@nextmovesoftware.com> wrote:
>
>
> I thought I'd just missed the bug fixing season of stage3, but there
> appears to a little latitude in early stage4 (for vector patches), so
> I'll post this now.
>
> This patch resolves PR target/106060 by providing efficient methods for
> materializing/synthesizing special "vector" constants on x86. Currently
> there are three methods of materializing a vector constant; the most
> general is to load a vector from the constant pool, secondly "duplicated"
> constants can be synthesized by moving an integer between units and
> broadcasting (or shuffling it), and finally the special cases of the
> all-zeros vector and all-ones vectors can be loaded via a single SSE
> instruction. This patch handles additional cases that can be synthesized
> in two instructions, loading an all-ones vector followed by another SSE
> instruction. Following my recent patch for PR target/112992, there's
> conveniently a single place in i386-expand.cc where these special cases
> can be handled.
>
> Two examples are given in the original bugzilla PR for 106060.
>
> __m256i
> should_be_cmpeq_abs ()
> {
> return _mm256_set1_epi8 (1);
> }
>
> is now generated (with -O3 -march=x86-64-v3) as:
>
> vpcmpeqd %ymm0, %ymm0, %ymm0
> vpabsb %ymm0, %ymm0
> ret
>
> and
>
> __m256i
> should_be_cmpeq_add ()
> {
> return _mm256_set1_epi8 (-2);
> }
>
> is now generated as:
>
> vpcmpeqd %ymm0, %ymm0, %ymm0
> vpaddb %ymm0, %ymm0, %ymm0
> ret
>
> This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
> and make -k check, both with and without --target_board=unix{-m32}
> with no new failures. Ok for mainline?
>
>
> 2024-01-16 Roger Sayle <roger@nextmovesoftware.com>
>
> gcc/ChangeLog
> PR target/106060
> * config/i386/i386-expand.cc (enum ix86_vec_bcast_alg): New.
> (struct ix86_vec_bcast_map_simode_t): New type for table below.
> (ix86_vec_bcast_map_simode): Table of SImode constants that may
> be efficiently synthesized by a ix86_vec_bcast_alg method.
> (ix86_vec_bcast_map_simode_cmp): New comparator for bsearch.
> (ix86_vector_duplicate_simode_const): Efficiently synthesize
> V4SImode and V8SImode constants that duplicate special constants.
> (ix86_vector_duplicate_value): Attempt to synthesize "special"
> vector constants using ix86_vector_duplicate_simode_const.
> * config/i386/i386.cc (ix86_rtx_costs) <case ABS>: ABS of a
> vector integer mode costs with a single SSE instruction.
>
+ switch (entry->alg)
+ {
+ case VEC_BCAST_PXOR:
+ if (mode == V8SImode && !TARGET_AVX2)
+ return false;
+ emit_move_insn (target, CONST0_RTX (mode));
+ return true;
+ case VEC_BCAST_PCMPEQ:
+ if ((mode == V4SImode && !TARGET_SSE2)
+ || (mode == V8SImode && !TARGET_AVX2))
+ return false;
+ emit_move_insn (target, CONSTM1_RTX (mode));
+ return true;
I think we need to prevent those standard_sse_constant_p getting in
ix86_expand_vector_init_duplicate by below codes.
/* If all values are identical, broadcast the value. */
if (all_same
&& (nvars != 0 || !standard_sse_constant_p (gen_rtx_CONST_VECTOR
(mode, XVEC (vals, 0)), mode))
&& ix86_expand_vector_init_duplicate (mmx_ok, mode, target,
XVECEXP (vals, 0, 0)))
return;
+ case VEC_BCAST_PABSB:
+ if (mode == V4SImode)
+ {
+ tmp1 = gen_reg_rtx (V16QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
+ tmp2 = gen_reg_rtx (V16QImode);
+ emit_insn (gen_absv16qi2 (tmp2, tmp1));
Shouldn't it rely on TARGET_SSE2?
+ case VEC_BCAST_PADDB:
+ if (mode == V4SImode)
+ {
+ tmp1 = gen_reg_rtx (V16QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
+ tmp2 = gen_reg_rtx (V16QImode);
+ emit_insn (gen_addv16qi3 (tmp2, tmp1, tmp1));
Ditto here and for all logic shift cases.
+ }
+
+ if ((mode == V4SImode || mode == V8SImode)
+ && CONST_INT_P (val)
+ && ix86_vector_duplicate_simode_const (mode, target, INTVAL (val)))
+ return true;
+
The alternative way is adding a pre_reload define_insn_and_split to
match specific const_vector and splitt it into new instructions.
In theoritically, the constant info can be retained before combine and
will enable more simplication.
Also the patch can be extend to V16SImode, but it can be a separate patch.
> gcc/testsuite/ChangeLog
> PR target/106060
> * gcc.target/i386/auto-init-8.c: Update test case.
> * gcc.target/i386/avx512fp16-3.c: Likewise.
> * gcc.target/i386/pr100865-9a.c: Likewise.
> * gcc.target/i386/pr106060-1.c: New test case.
> * gcc.target/i386/pr106060-2.c: Likewise.
> * gcc.target/i386/pr106060-3.c: Likewise.
> * gcc.target/i386/pr70314-3.c: Update test case.
> * gcc.target/i386/vect-shiftv4qi.c: Likewise.
> * gcc.target/i386/vect-shiftv8qi.c: Likewise.
>
>
> Thanks in advance,
> Roger
> --
>
Hi Hongtao,
Many thanks for the review. Here's a revised version of my patch
that addresses (most of) the issues you've raised. Firstly the
handling of zero and all_ones in this function is mostly for
completeness/documentation, these standard_sse_constant_p
values are (currently/normally) handled elsewhere. But I have
added an "n_var == 0" optimization to ix86_expand_vector_init.
As you've suggested I've added explicit TARGET_SSE2 tests where
required, and for consistency I've also added support for AVX512's
V16SImode.
As you've predicted, the eventual goal is to move this after combine
(or reload) using define_insn_and_split, but that requires a significant
restructuring that should be done in steps. This also interacts with
a similar planned reorganization of TImode constant handling. If
all 128-bit (vector) constants are acceptable before combine, then
STV has the freedom to chose V1TImode (and this broadcast
functionality) to implement TImode operations on immediate
constants.
This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
and make -k check, both with and without --target_board=unix{-m32}
with no new failures. Ok for mainline (in stage 1)?
2024-01-25 Roger Sayle <roger@nextmovesoftware.com>
Hongtao Liu <hongtao.liu@intel.com>
gcc/ChangeLog
PR target/106060
* config/i386/i386-expand.cc (enum ix86_vec_bcast_alg): New.
(struct ix86_vec_bcast_map_simode_t): New type for table below.
(ix86_vec_bcast_map_simode): Table of SImode constants that may
be efficiently synthesized by a ix86_vec_bcast_alg method.
(ix86_vec_bcast_map_simode_cmp): New comparator for bsearch.
(ix86_vector_duplicate_simode_const): Efficiently synthesize
V4SImode and V8SImode constants that duplicate special constants.
(ix86_vector_duplicate_value): Attempt to synthesize "special"
vector constants using ix86_vector_duplicate_simode_const.
* config/i386/i386.cc (ix86_rtx_costs) <case ABS>: ABS of a
vector integer mode costs with a single SSE instruction.
gcc/testsuite/ChangeLog
PR target/106060
* gcc.target/i386/auto-init-8.c: Update test case.
* gcc.target/i386/avx512fp16-3.c: Likewise.
* gcc.target/i386/pr100865-9a.c: Likewise.
* gcc.target/i386/pr101796-1.c: Likewise.
* gcc.target/i386/pr106060-1.c: New test case.
* gcc.target/i386/pr106060-2.c: Likewise.
* gcc.target/i386/pr106060-3.c: Likewise.
* gcc.target/i386/pr70314.c: Update test case.
* gcc.target/i386/vect-shiftv4qi.c: Likewise.
* gcc.target/i386/vect-shiftv8qi.c: Likewise.
Roger
--
> -----Original Message-----
> From: Hongtao Liu <crazylht@gmail.com>
> Sent: 17 January 2024 03:13
> To: Roger Sayle <roger@nextmovesoftware.com>
> Cc: gcc-patches@gcc.gnu.org; Uros Bizjak <ubizjak@gmail.com>
> Subject: Re: [x86 PATCH] PR target/106060: Improved SSE vector constant
> materialization.
>
> On Wed, Jan 17, 2024 at 5:59 AM Roger Sayle <roger@nextmovesoftware.com>
> wrote:
> >
> >
> > I thought I'd just missed the bug fixing season of stage3, but there
> > appears to a little latitude in early stage4 (for vector patches), so
> > I'll post this now.
> >
> > This patch resolves PR target/106060 by providing efficient methods
> > for materializing/synthesizing special "vector" constants on x86.
> > Currently there are three methods of materializing a vector constant;
> > the most general is to load a vector from the constant pool, secondly
> "duplicated"
> > constants can be synthesized by moving an integer between units and
> > broadcasting (or shuffling it), and finally the special cases of the
> > all-zeros vector and all-ones vectors can be loaded via a single SSE
> > instruction. This patch handles additional cases that can be synthesized
> > in two instructions, loading an all-ones vector followed by another
> > SSE instruction. Following my recent patch for PR target/112992,
> > there's conveniently a single place in i386-expand.cc where these
> > special cases can be handled.
> >
> > Two examples are given in the original bugzilla PR for 106060.
> >
> > __m256i
> > should_be_cmpeq_abs ()
> > {
> > return _mm256_set1_epi8 (1);
> > }
> >
> > is now generated (with -O3 -march=x86-64-v3) as:
> >
> > vpcmpeqd %ymm0, %ymm0, %ymm0
> > vpabsb %ymm0, %ymm0
> > ret
> >
> > and
> >
> > __m256i
> > should_be_cmpeq_add ()
> > {
> > return _mm256_set1_epi8 (-2);
> > }
> >
> > is now generated as:
> >
> > vpcmpeqd %ymm0, %ymm0, %ymm0
> > vpaddb %ymm0, %ymm0, %ymm0
> > ret
> >
> > This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
> > and make -k check, both with and without --target_board=unix{-m32}
> > with no new failures. Ok for mainline?
> >
> >
> > 2024-01-16 Roger Sayle <roger@nextmovesoftware.com>
> >
> > gcc/ChangeLog
> > PR target/106060
> > * config/i386/i386-expand.cc (enum ix86_vec_bcast_alg): New.
> > (struct ix86_vec_bcast_map_simode_t): New type for table below.
> > (ix86_vec_bcast_map_simode): Table of SImode constants that may
> > be efficiently synthesized by a ix86_vec_bcast_alg method.
> > (ix86_vec_bcast_map_simode_cmp): New comparator for bsearch.
> > (ix86_vector_duplicate_simode_const): Efficiently synthesize
> > V4SImode and V8SImode constants that duplicate special constants.
> > (ix86_vector_duplicate_value): Attempt to synthesize "special"
> > vector constants using ix86_vector_duplicate_simode_const.
> > * config/i386/i386.cc (ix86_rtx_costs) <case ABS>: ABS of a
> > vector integer mode costs with a single SSE instruction.
> >
>
> + switch (entry->alg)
> + {
> + case VEC_BCAST_PXOR:
> + if (mode == V8SImode && !TARGET_AVX2) return false;
> + emit_move_insn (target, CONST0_RTX (mode));
> + return true;
> + case VEC_BCAST_PCMPEQ:
> + if ((mode == V4SImode && !TARGET_SSE2)
> + || (mode == V8SImode && !TARGET_AVX2)) return false;
> + emit_move_insn (target, CONSTM1_RTX (mode));
> + return true;
>
> I think we need to prevent those standard_sse_constant_p getting in
> ix86_expand_vector_init_duplicate by below codes.
>
> /* If all values are identical, broadcast the value. */
> if (all_same
> && (nvars != 0 || !standard_sse_constant_p (gen_rtx_CONST_VECTOR
> (mode, XVEC (vals, 0)), mode))
> && ix86_expand_vector_init_duplicate (mmx_ok, mode, target,
> XVECEXP (vals, 0, 0)))
> return;
>
> + case VEC_BCAST_PABSB:
> + if (mode == V4SImode)
> + {
> + tmp1 = gen_reg_rtx (V16QImode);
> + emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
> + tmp2 = gen_reg_rtx (V16QImode);
> + emit_insn (gen_absv16qi2 (tmp2, tmp1));
> Shouldn't it rely on TARGET_SSE2?
>
> + case VEC_BCAST_PADDB:
> + if (mode == V4SImode)
> + {
> + tmp1 = gen_reg_rtx (V16QImode);
> + emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
> + tmp2 = gen_reg_rtx (V16QImode);
> + emit_insn (gen_addv16qi3 (tmp2, tmp1, tmp1));
> Ditto here and for all logic shift cases.
> + }
>
> +
> + if ((mode == V4SImode || mode == V8SImode)
> + && CONST_INT_P (val)
> + && ix86_vector_duplicate_simode_const (mode, target, INTVAL (val)))
> + return true;
> +
> The alternative way is adding a pre_reload define_insn_and_split to match
> specific const_vector and splitt it into new instructions.
> In theoritically, the constant info can be retained before combine and will enable
> more simplication.
>
> Also the patch can be extend to V16SImode, but it can be a separate patch.
>
> > gcc/testsuite/ChangeLog
> > PR target/106060
> > * gcc.target/i386/auto-init-8.c: Update test case.
> > * gcc.target/i386/avx512fp16-3.c: Likewise.
> > * gcc.target/i386/pr100865-9a.c: Likewise.
> > * gcc.target/i386/pr106060-1.c: New test case.
> > * gcc.target/i386/pr106060-2.c: Likewise.
> > * gcc.target/i386/pr106060-3.c: Likewise.
> > * gcc.target/i386/pr70314-3.c: Update test case.
> > * gcc.target/i386/vect-shiftv4qi.c: Likewise.
> > * gcc.target/i386/vect-shiftv8qi.c: Likewise.
> >
> >
> > Thanks in advance,
> > Roger
> > --
> >
>
>
> --
> BR,
> Hongtao
diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc
index 52754e1..f802b8f 100644
--- a/gcc/config/i386/i386-expand.cc
+++ b/gcc/config/i386/i386-expand.cc
@@ -15638,6 +15638,332 @@ s4fma_expand:
gcc_unreachable ();
}
+/* See below where shifts are handled for explanation of this enum. */
+enum ix86_vec_bcast_alg
+{
+ VEC_BCAST_PXOR,
+ VEC_BCAST_PCMPEQ,
+ VEC_BCAST_PABSB,
+ VEC_BCAST_PADDB,
+ VEC_BCAST_PSRLW,
+ VEC_BCAST_PSRLD,
+ VEC_BCAST_PSLLW,
+ VEC_BCAST_PSLLD
+};
+
+struct ix86_vec_bcast_map_simode_t
+{
+ unsigned int key;
+ enum ix86_vec_bcast_alg alg;
+ unsigned int arg;
+};
+
+/* This table must be kept sorted as values are looked-up using bsearch. */
+static const ix86_vec_bcast_map_simode_t ix86_vec_bcast_map_simode[] = {
+ { 0x00000000, VEC_BCAST_PXOR, 0 },
+ { 0x00000001, VEC_BCAST_PSRLD, 31 },
+ { 0x00000003, VEC_BCAST_PSRLD, 30 },
+ { 0x00000007, VEC_BCAST_PSRLD, 29 },
+ { 0x0000000f, VEC_BCAST_PSRLD, 28 },
+ { 0x0000001f, VEC_BCAST_PSRLD, 27 },
+ { 0x0000003f, VEC_BCAST_PSRLD, 26 },
+ { 0x0000007f, VEC_BCAST_PSRLD, 25 },
+ { 0x000000ff, VEC_BCAST_PSRLD, 24 },
+ { 0x000001ff, VEC_BCAST_PSRLD, 23 },
+ { 0x000003ff, VEC_BCAST_PSRLD, 22 },
+ { 0x000007ff, VEC_BCAST_PSRLD, 21 },
+ { 0x00000fff, VEC_BCAST_PSRLD, 20 },
+ { 0x00001fff, VEC_BCAST_PSRLD, 19 },
+ { 0x00003fff, VEC_BCAST_PSRLD, 18 },
+ { 0x00007fff, VEC_BCAST_PSRLD, 17 },
+ { 0x0000ffff, VEC_BCAST_PSRLD, 16 },
+ { 0x00010001, VEC_BCAST_PSRLW, 15 },
+ { 0x0001ffff, VEC_BCAST_PSRLD, 15 },
+ { 0x00030003, VEC_BCAST_PSRLW, 14 },
+ { 0x0003ffff, VEC_BCAST_PSRLD, 14 },
+ { 0x00070007, VEC_BCAST_PSRLW, 13 },
+ { 0x0007ffff, VEC_BCAST_PSRLD, 13 },
+ { 0x000f000f, VEC_BCAST_PSRLW, 12 },
+ { 0x000fffff, VEC_BCAST_PSRLD, 12 },
+ { 0x001f001f, VEC_BCAST_PSRLW, 11 },
+ { 0x001fffff, VEC_BCAST_PSRLD, 11 },
+ { 0x003f003f, VEC_BCAST_PSRLW, 10 },
+ { 0x003fffff, VEC_BCAST_PSRLD, 10 },
+ { 0x007f007f, VEC_BCAST_PSRLW, 9 },
+ { 0x007fffff, VEC_BCAST_PSRLD, 9 },
+ { 0x00ff00ff, VEC_BCAST_PSRLW, 8 },
+ { 0x00ffffff, VEC_BCAST_PSRLD, 8 },
+ { 0x01010101, VEC_BCAST_PABSB, 0 },
+ { 0x01ff01ff, VEC_BCAST_PSRLW, 7 },
+ { 0x01ffffff, VEC_BCAST_PSRLD, 7 },
+ { 0x03ff03ff, VEC_BCAST_PSRLW, 6 },
+ { 0x03ffffff, VEC_BCAST_PSRLD, 6 },
+ { 0x07ff07ff, VEC_BCAST_PSRLW, 5 },
+ { 0x07ffffff, VEC_BCAST_PSRLD, 5 },
+ { 0x0fff0fff, VEC_BCAST_PSRLW, 4 },
+ { 0x0fffffff, VEC_BCAST_PSRLD, 4 },
+ { 0x1fff1fff, VEC_BCAST_PSRLW, 3 },
+ { 0x1fffffff, VEC_BCAST_PSRLD, 3 },
+ { 0x3fff3fff, VEC_BCAST_PSRLW, 2 },
+ { 0x3fffffff, VEC_BCAST_PSRLD, 2 },
+ { 0x7fff7fff, VEC_BCAST_PSRLW, 1 },
+ { 0x7fffffff, VEC_BCAST_PSRLD, 1 },
+ { 0x80000000, VEC_BCAST_PSLLD, 31 },
+ { 0x80008000, VEC_BCAST_PSLLW, 15 },
+ { 0xc0000000, VEC_BCAST_PSLLD, 30 },
+ { 0xc000c000, VEC_BCAST_PSLLW, 14 },
+ { 0xe0000000, VEC_BCAST_PSLLD, 29 },
+ { 0xe000e000, VEC_BCAST_PSLLW, 13 },
+ { 0xf0000000, VEC_BCAST_PSLLD, 28 },
+ { 0xf000f000, VEC_BCAST_PSLLW, 12 },
+ { 0xf8000000, VEC_BCAST_PSLLD, 27 },
+ { 0xf800f800, VEC_BCAST_PSLLW, 11 },
+ { 0xfc000000, VEC_BCAST_PSLLD, 26 },
+ { 0xfc00fc00, VEC_BCAST_PSLLW, 10 },
+ { 0xfe000000, VEC_BCAST_PSLLD, 25 },
+ { 0xfe00fe00, VEC_BCAST_PSLLW, 9 },
+ { 0xfefefefe, VEC_BCAST_PADDB, 0 },
+ { 0xff000000, VEC_BCAST_PSLLD, 24 },
+ { 0xff00ff00, VEC_BCAST_PSLLW, 8 },
+ { 0xff800000, VEC_BCAST_PSLLD, 23 },
+ { 0xff80ff80, VEC_BCAST_PSLLW, 7 },
+ { 0xffc00000, VEC_BCAST_PSLLD, 22 },
+ { 0xffc0ffc0, VEC_BCAST_PSLLW, 6 },
+ { 0xffe00000, VEC_BCAST_PSLLD, 21 },
+ { 0xffe0ffe0, VEC_BCAST_PSLLW, 5 },
+ { 0xfff00000, VEC_BCAST_PSLLD, 20 },
+ { 0xfff0fff0, VEC_BCAST_PSLLW, 4 },
+ { 0xfff80000, VEC_BCAST_PSLLD, 19 },
+ { 0xfff8fff8, VEC_BCAST_PSLLW, 3 },
+ { 0xfffc0000, VEC_BCAST_PSLLD, 18 },
+ { 0xfffcfffc, VEC_BCAST_PSLLW, 2 },
+ { 0xfffe0000, VEC_BCAST_PSLLD, 17 },
+ { 0xfffefffe, VEC_BCAST_PSLLW, 1 },
+ { 0xffff0000, VEC_BCAST_PSLLD, 16 },
+ { 0xffff8000, VEC_BCAST_PSLLD, 15 },
+ { 0xffffc000, VEC_BCAST_PSLLD, 14 },
+ { 0xffffe000, VEC_BCAST_PSLLD, 13 },
+ { 0xfffff000, VEC_BCAST_PSLLD, 12 },
+ { 0xfffff800, VEC_BCAST_PSLLD, 11 },
+ { 0xfffffc00, VEC_BCAST_PSLLD, 10 },
+ { 0xfffffe00, VEC_BCAST_PSLLD, 9 },
+ { 0xffffff00, VEC_BCAST_PSLLD, 8 },
+ { 0xffffff80, VEC_BCAST_PSLLD, 7 },
+ { 0xffffffc0, VEC_BCAST_PSLLD, 6 },
+ { 0xffffffe0, VEC_BCAST_PSLLD, 5 },
+ { 0xfffffff0, VEC_BCAST_PSLLD, 4 },
+ { 0xfffffff8, VEC_BCAST_PSLLD, 3 },
+ { 0xfffffffc, VEC_BCAST_PSLLD, 2 },
+ { 0xfffffffe, VEC_BCAST_PSLLD, 1 },
+ { 0xffffffff, VEC_BCAST_PCMPEQ, 0 }
+};
+
+/* Comparator for bsearch on ix86_vec_bcast_map. */
+static int
+ix86_vec_bcast_map_simode_cmp (const void *key, const void *entry)
+{
+ return (*(const unsigned int*)key)
+ - ((const ix86_vec_bcast_map_simode_t*)entry)->key;
+}
+
+/* A subroutine of ix86_vector_duplicate_value. Tries to efficiently
+ materialize V4SImode, V8SImode and V16SImode vectors from SImode
+ integer constants. */
+static bool
+ix86_vector_duplicate_simode_const (machine_mode mode, rtx target,
+ unsigned int val)
+{
+ const ix86_vec_bcast_map_simode_t *entry;
+ rtx tmp1, tmp2;
+
+ entry = (const ix86_vec_bcast_map_simode_t*)
+ bsearch(&val, ix86_vec_bcast_map_simode,
+ ARRAY_SIZE (ix86_vec_bcast_map_simode),
+ sizeof (ix86_vec_bcast_map_simode_t),
+ ix86_vec_bcast_map_simode_cmp);
+ if (!entry)
+ return false;
+
+ switch (entry->alg)
+ {
+ case VEC_BCAST_PXOR:
+ if ((mode == V8SImode && !TARGET_AVX2)
+ || (mode == V16SImode && !(TARGET_AVX512F && TARGET_EVEX512)))
+ return false;
+ emit_move_insn (target, CONST0_RTX (mode));
+ return true;
+
+ case VEC_BCAST_PCMPEQ:
+ if ((mode == V4SImode && !TARGET_SSE2)
+ || (mode == V8SImode && !TARGET_AVX2)
+ || (mode == V16SImode && !(TARGET_AVX512F && TARGET_EVEX512)))
+ return false;
+ emit_move_insn (target, CONSTM1_RTX (mode));
+ return true;
+
+ case VEC_BCAST_PABSB:
+ if (mode == V4SImode && TARGET_SSE2)
+ {
+ tmp1 = gen_reg_rtx (V16QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
+ tmp2 = gen_reg_rtx (V16QImode);
+ emit_insn (gen_absv16qi2 (tmp2, tmp1));
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V32QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V32QImode));
+ tmp2 = gen_reg_rtx (V32QImode);
+ emit_insn (gen_absv32qi2 (tmp2, tmp1));
+ }
+ else if (mode == V16SImode && TARGET_AVX512BW && TARGET_EVEX512)
+ {
+ tmp1 = gen_reg_rtx (V64QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V64QImode));
+ tmp2 = gen_reg_rtx (V64QImode);
+ emit_insn (gen_absv64qi2 (tmp2, tmp1));
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PADDB:
+ if (mode == V4SImode && TARGET_SSE2)
+ {
+ tmp1 = gen_reg_rtx (V16QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
+ tmp2 = gen_reg_rtx (V16QImode);
+ emit_insn (gen_addv16qi3 (tmp2, tmp1, tmp1));
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V32QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V32QImode));
+ tmp2 = gen_reg_rtx (V32QImode);
+ emit_insn (gen_addv32qi3 (tmp2, tmp1, tmp1));
+ }
+ else if (mode == V16SImode && TARGET_AVX512BW && TARGET_EVEX512)
+ {
+ tmp1 = gen_reg_rtx (V64QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V64QImode));
+ tmp2 = gen_reg_rtx (V64QImode);
+ emit_insn (gen_addv64qi3 (tmp2, tmp1, tmp1));
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PSRLW:
+ if (mode == V4SImode && TARGET_SSE2)
+ {
+ tmp1 = gen_reg_rtx (V8HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V8HImode));
+ tmp2 = gen_reg_rtx (V8HImode);
+ emit_insn (gen_lshrv8hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V16HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16HImode));
+ tmp2 = gen_reg_rtx (V16HImode);
+ emit_insn (gen_lshrv16hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else if (mode == V16SImode && TARGET_AVX512BW && TARGET_EVEX512)
+ {
+ tmp1 = gen_reg_rtx (V32HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V32HImode));
+ tmp2 = gen_reg_rtx (V32HImode);
+ emit_insn (gen_lshrv32hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PSRLD:
+ if (mode == V4SImode && TARGET_SSE2)
+ {
+ tmp1 = gen_reg_rtx (V4SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V4SImode));
+ emit_insn (gen_lshrv4si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V8SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V8SImode));
+ emit_insn (gen_lshrv8si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else if (mode == V16SImode && TARGET_AVX512F && TARGET_EVEX512)
+ {
+ tmp1 = gen_reg_rtx (V16SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16SImode));
+ emit_insn (gen_lshrv16si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PSLLW:
+ if (mode == V4SImode && TARGET_SSE2)
+ {
+ tmp1 = gen_reg_rtx (V8HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V8HImode));
+ tmp2 = gen_reg_rtx (V8HImode);
+ emit_insn (gen_ashlv8hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V16HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16HImode));
+ tmp2 = gen_reg_rtx (V16HImode);
+ emit_insn (gen_ashlv16hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else if (mode == V16SImode && TARGET_AVX512BW && TARGET_EVEX512)
+ {
+ tmp1 = gen_reg_rtx (V32HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V32HImode));
+ tmp2 = gen_reg_rtx (V32HImode);
+ emit_insn (gen_ashlv32hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PSLLD:
+ if (mode == V4SImode && TARGET_SSE2)
+ {
+ tmp1 = gen_reg_rtx (V4SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V4SImode));
+ emit_insn (gen_ashlv4si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V8SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V8SImode));
+ emit_insn (gen_ashlv8si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else if (mode == V16SImode && TARGET_AVX512F && TARGET_EVEX512)
+ {
+ tmp1 = gen_reg_rtx (V16SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16SImode));
+ emit_insn (gen_ashlv16si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else
+ return false;
+
+ default:
+ return false;
+ }
+
+ emit_move_insn (target, gen_lowpart (mode, tmp2));
+ return true;
+}
+
/* A subroutine of ix86_expand_vector_init_duplicate. Tries to
fill target with val via vec_duplicate. */
@@ -15647,6 +15973,12 @@ ix86_vector_duplicate_value (machine_mode mode, rtx target, rtx val)
bool ok;
rtx_insn *insn;
rtx dup;
+
+ if ((mode == V4SImode || mode == V8SImode || mode == V16SImode)
+ && CONST_INT_P (val)
+ && ix86_vector_duplicate_simode_const (mode, target, INTVAL (val)))
+ return true;
+
/* Save/restore recog_data in case this is called from splitters
or other routines where recog_data needs to stay valid across
force_reg. See PR106577. */
@@ -15743,6 +16075,24 @@ ix86_expand_vector_init_duplicate (bool mmx_ok, machine_mode mode,
}
return ix86_vector_duplicate_value (mode, target, val);
+ case E_V8DImode:
+ if (CONST_INT_P (val))
+ {
+ int tmp = (int)INTVAL (val);
+ if (tmp == (int)(INTVAL (val) >> 32))
+ {
+ rtx reg = gen_reg_rtx (V16SImode);
+ ok = ix86_vector_duplicate_value (V16SImode, reg,
+ GEN_INT (tmp));
+ if (ok)
+ {
+ emit_move_insn (target, gen_lowpart (V8DImode, reg));
+ return true;
+ }
+ }
+ }
+ return ix86_vector_duplicate_value (mode, target, val);
+
case E_V2SImode:
case E_V2SFmode:
if (!mmx_ok)
@@ -15756,7 +16106,6 @@ ix86_expand_vector_init_duplicate (bool mmx_ok, machine_mode mode,
case E_V4SFmode:
case E_V4SImode:
case E_V16SImode:
- case E_V8DImode:
case E_V16SFmode:
case E_V8DFmode:
return ix86_vector_duplicate_value (mode, target, val);
@@ -15961,9 +16310,13 @@ ix86_expand_vector_init_duplicate (bool mmx_ok, machine_mode mode,
return true;
case E_V32HImode:
+ case E_V64QImode:
+ if (CONST_INT_P (val))
+ goto widen;
+ /* FALLTHRU */
+
case E_V32HFmode:
case E_V32BFmode:
- case E_V64QImode:
gcc_assert (TARGET_EVEX512);
if (TARGET_AVX512BW)
return ix86_vector_duplicate_value (mode, target, val);
@@ -16963,6 +17316,13 @@ ix86_expand_vector_init (bool mmx_ok, rtx target, rtx vals)
all_same = false;
}
+ /* Handle the zero vector as special case. */
+ if (n_var == 0 && all_const_zero)
+ {
+ emit_move_insn (target, CONST0_RTX (mode));
+ return;
+ }
+
/* If all values are identical, broadcast the value. */
if (all_same
&& ix86_expand_vector_init_duplicate (mmx_ok, mode, target,
diff --git a/gcc/config/i386/i386.cc b/gcc/config/i386/i386.cc
index c5eaeed..5b0e8f9 100644
--- a/gcc/config/i386/i386.cc
+++ b/gcc/config/i386/i386.cc
@@ -22076,6 +22076,8 @@ ix86_rtx_costs (rtx x, machine_mode mode, int outer_code_i, int opno,
*total = cost->fabs;
else if (FLOAT_MODE_P (mode))
*total = ix86_vec_cost (mode, cost->sse_op);
+ else if (GET_MODE_CLASS (mode) == MODE_VECTOR_INT)
+ *total = cost->sse_op;
return false;
case SQRT:
diff --git a/gcc/testsuite/gcc.target/i386/auto-init-8.c b/gcc/testsuite/gcc.target/i386/auto-init-8.c
index 7023d72..666ee14 100644
--- a/gcc/testsuite/gcc.target/i386/auto-init-8.c
+++ b/gcc/testsuite/gcc.target/i386/auto-init-8.c
@@ -29,7 +29,7 @@ double foo()
return result;
}
-/* { dg-final { scan-rtl-dump-times "0xfffffffffefefefe" 3 "expand" } } */
+/* { dg-final { scan-rtl-dump-times "0xfffffffffefefefe" 1 "expand" } } */
/* { dg-final { scan-rtl-dump-times "\\\[0xfefefefefefefefe\\\]" 2 "expand" } } */
/* { dg-final { scan-rtl-dump-times "0xfffffffffffffffe\\\]\\\) repeated x16" 2 "expand" } } */
diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-13.c b/gcc/testsuite/gcc.target/i386/avx512fp16-13.c
index f431b8a..1cd9a07 100644
--- a/gcc/testsuite/gcc.target/i386/avx512fp16-13.c
+++ b/gcc/testsuite/gcc.target/i386/avx512fp16-13.c
@@ -116,7 +116,6 @@ abs512_ph (__m512h a)
return _mm512_abs_ph (a);
}
-/* { dg-final { scan-assembler-times "vpbroadcastd\[^\n\]*%zmm\[0-9\]+" 1 { target { ! ia32 } } } } */
/* { dg-final { scan-assembler-times "vpandd\[^\n\]*%zmm\[0-9\]+" 1 } } */
__m256h
@@ -126,7 +125,6 @@ abs256_ph (__m256h a)
return _mm256_abs_ph (a);
}
-/* { dg-final { scan-assembler-times "vpbroadcastd\[^\n\]*%ymm\[0-9\]+" 1 { target { ! ia32 } } } } */
/* { dg-final { scan-assembler-times "vpand\[^\n\]*%ymm\[0-9\]+" 1 } } */
__m128h
@@ -136,5 +134,4 @@ abs_ph (__m128h a)
return _mm_abs_ph (a);
}
-/* { dg-final { scan-assembler-times "vpbroadcastd\[^\n\]*%xmm\[0-9\]+" 1 { target { ! ia32 } } } } */
/* { dg-final { scan-assembler-times "vpand\[^\n\]*%xmm\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr100865-9a.c b/gcc/testsuite/gcc.target/i386/pr100865-9a.c
index f2ac1bd..91cfeda 100644
--- a/gcc/testsuite/gcc.target/i386/pr100865-9a.c
+++ b/gcc/testsuite/gcc.target/i386/pr100865-9a.c
@@ -18,7 +18,7 @@ foo (void)
{
int i;
for (i = 0; i < sizeof (array) / sizeof (array[0]); i++)
- array[i] = MK_CONST128_BROADCAST (0x1fff);
+ array[i] = MK_CONST128_BROADCAST (0x1234);
}
/* { dg-final { scan-assembler-times "vpbroadcastd\[\\t \]+%xmm\[0-9\]+, %xmm\[0-9\]+" 1 } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr101796-1.c b/gcc/testsuite/gcc.target/i386/pr101796-1.c
index b25464d..09532f9 100644
--- a/gcc/testsuite/gcc.target/i386/pr101796-1.c
+++ b/gcc/testsuite/gcc.target/i386/pr101796-1.c
@@ -15,7 +15,7 @@ volatile __m512i a, b;
void
foo()
{
- b = _mm512_srlv_epi16 (a, _mm512_set1_epi16 (3));
- b = _mm512_sllv_epi16 (a, _mm512_set1_epi16 (4));
- b = _mm512_srav_epi16 (a, _mm512_set1_epi16 (5));
+ b = _mm512_srlv_epi16 (a, _mm512_set1_epi16 (4));
+ b = _mm512_sllv_epi16 (a, _mm512_set1_epi16 (5));
+ b = _mm512_srav_epi16 (a, _mm512_set1_epi16 (6));
}
diff --git a/gcc/testsuite/gcc.target/i386/pr106060-1.c b/gcc/testsuite/gcc.target/i386/pr106060-1.c
new file mode 100644
index 0000000..a734d56
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr106060-1.c
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=x86-64-v3" } */
+#include <immintrin.h>
+
+__m256i
+foo ()
+{
+ /* shouldnt_have_movabs */
+ return _mm256_set1_epi8 (123);
+}
+
+/* { dg-final { scan-assembler-not "movabs" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr106060-2.c b/gcc/testsuite/gcc.target/i386/pr106060-2.c
new file mode 100644
index 0000000..23933ab
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr106060-2.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=x86-64-v3" } */
+#include <immintrin.h>
+
+__m256i
+foo ()
+{
+ /* should_be_cmpeq_abs */
+ return _mm256_set1_epi8 (1);
+}
+
+/* { dg-final { scan-assembler "pcmpeq" } } */
+/* { dg-final { scan-assembler "pabsb" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr106060-3.c b/gcc/testsuite/gcc.target/i386/pr106060-3.c
new file mode 100644
index 0000000..59c128c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr106060-3.c
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=x86-64-v3" } */
+#include <immintrin.h>
+
+__m256i
+foo ()
+{
+ /* should_be_cmpeq_add */
+ return _mm256_set1_epi8 (-2);
+}
+
+/* { dg-final { scan-assembler "pcmpeq" } } */
+/* { dg-final { scan-assembler "paddb" } } */
+
diff --git a/gcc/testsuite/gcc.target/i386/pr70314.c b/gcc/testsuite/gcc.target/i386/pr70314.c
index aad8dd9..181d2b4 100644
--- a/gcc/testsuite/gcc.target/i386/pr70314.c
+++ b/gcc/testsuite/gcc.target/i386/pr70314.c
@@ -1,6 +1,6 @@
/* { dg-do compile } */
/* { dg-options "-march=skylake-avx512 -O2" } */
-/* { dg-final { scan-assembler-times "cmp" 2 } } */
+/* { dg-final { scan-assembler-times "cmp\[dq\]" 2 } } */
/* { dg-final { scan-assembler-not "and" } } */
typedef long vec __attribute__((vector_size(16)));
diff --git a/gcc/testsuite/gcc.target/i386/vect-shiftv4qi.c b/gcc/testsuite/gcc.target/i386/vect-shiftv4qi.c
index c6a6390..b7e45c2 100644
--- a/gcc/testsuite/gcc.target/i386/vect-shiftv4qi.c
+++ b/gcc/testsuite/gcc.target/i386/vect-shiftv4qi.c
@@ -28,7 +28,7 @@ __vu srl_c (__vu a)
return a >> 5;
}
-/* { dg-final { scan-assembler-times "psrlw" 2 } } */
+/* { dg-final { scan-assembler-times "psrlw" 5 } } */
__vi sra (__vi a, int n)
{
diff --git a/gcc/testsuite/gcc.target/i386/vect-shiftv8qi.c b/gcc/testsuite/gcc.target/i386/vect-shiftv8qi.c
index 244b0db..2471e6e 100644
--- a/gcc/testsuite/gcc.target/i386/vect-shiftv8qi.c
+++ b/gcc/testsuite/gcc.target/i386/vect-shiftv8qi.c
@@ -28,7 +28,7 @@ __vu srl_c (__vu a)
return a >> 5;
}
-/* { dg-final { scan-assembler-times "psrlw" 2 } } */
+/* { dg-final { scan-assembler-times "psrlw" 5 } } */
__vi sra (__vi a, int n)
{
On Fri, Jan 26, 2024 at 3:03 AM Roger Sayle <roger@nextmovesoftware.com> wrote:
>
>
> Hi Hongtao,
> Many thanks for the review. Here's a revised version of my patch
> that addresses (most of) the issues you've raised. Firstly the
> handling of zero and all_ones in this function is mostly for
> completeness/documentation, these standard_sse_constant_p
> values are (currently/normally) handled elsewhere. But I have
> added an "n_var == 0" optimization to ix86_expand_vector_init.
>
> As you've suggested I've added explicit TARGET_SSE2 tests where
> required, and for consistency I've also added support for AVX512's
> V16SImode.
>
> As you've predicted, the eventual goal is to move this after combine
> (or reload) using define_insn_and_split, but that requires a significant
> restructuring that should be done in steps. This also interacts with
> a similar planned reorganization of TImode constant handling. If
> all 128-bit (vector) constants are acceptable before combine, then
> STV has the freedom to chose V1TImode (and this broadcast
> functionality) to implement TImode operations on immediate
> constants.
>
> This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
> and make -k check, both with and without --target_board=unix{-m32}
> with no new failures. Ok for mainline (in stage 1)?
Ok, thanks for handling this.
>
>
> 2024-01-25 Roger Sayle <roger@nextmovesoftware.com>
> Hongtao Liu <hongtao.liu@intel.com>
>
> gcc/ChangeLog
> PR target/106060
> * config/i386/i386-expand.cc (enum ix86_vec_bcast_alg): New.
> (struct ix86_vec_bcast_map_simode_t): New type for table below.
> (ix86_vec_bcast_map_simode): Table of SImode constants that may
> be efficiently synthesized by a ix86_vec_bcast_alg method.
> (ix86_vec_bcast_map_simode_cmp): New comparator for bsearch.
> (ix86_vector_duplicate_simode_const): Efficiently synthesize
> V4SImode and V8SImode constants that duplicate special constants.
> (ix86_vector_duplicate_value): Attempt to synthesize "special"
> vector constants using ix86_vector_duplicate_simode_const.
> * config/i386/i386.cc (ix86_rtx_costs) <case ABS>: ABS of a
> vector integer mode costs with a single SSE instruction.
>
> gcc/testsuite/ChangeLog
> PR target/106060
> * gcc.target/i386/auto-init-8.c: Update test case.
> * gcc.target/i386/avx512fp16-3.c: Likewise.
> * gcc.target/i386/pr100865-9a.c: Likewise.
> * gcc.target/i386/pr101796-1.c: Likewise.
> * gcc.target/i386/pr106060-1.c: New test case.
> * gcc.target/i386/pr106060-2.c: Likewise.
> * gcc.target/i386/pr106060-3.c: Likewise.
> * gcc.target/i386/pr70314.c: Update test case.
> * gcc.target/i386/vect-shiftv4qi.c: Likewise.
> * gcc.target/i386/vect-shiftv8qi.c: Likewise.
>
>
> Roger
> --
>
> > -----Original Message-----
> > From: Hongtao Liu <crazylht@gmail.com>
> > Sent: 17 January 2024 03:13
> > To: Roger Sayle <roger@nextmovesoftware.com>
> > Cc: gcc-patches@gcc.gnu.org; Uros Bizjak <ubizjak@gmail.com>
> > Subject: Re: [x86 PATCH] PR target/106060: Improved SSE vector constant
> > materialization.
> >
> > On Wed, Jan 17, 2024 at 5:59 AM Roger Sayle <roger@nextmovesoftware.com>
> > wrote:
> > >
> > >
> > > I thought I'd just missed the bug fixing season of stage3, but there
> > > appears to a little latitude in early stage4 (for vector patches), so
> > > I'll post this now.
> > >
> > > This patch resolves PR target/106060 by providing efficient methods
> > > for materializing/synthesizing special "vector" constants on x86.
> > > Currently there are three methods of materializing a vector constant;
> > > the most general is to load a vector from the constant pool, secondly
> > "duplicated"
> > > constants can be synthesized by moving an integer between units and
> > > broadcasting (or shuffling it), and finally the special cases of the
> > > all-zeros vector and all-ones vectors can be loaded via a single SSE
> > > instruction. This patch handles additional cases that can be synthesized
> > > in two instructions, loading an all-ones vector followed by another
> > > SSE instruction. Following my recent patch for PR target/112992,
> > > there's conveniently a single place in i386-expand.cc where these
> > > special cases can be handled.
> > >
> > > Two examples are given in the original bugzilla PR for 106060.
> > >
> > > __m256i
> > > should_be_cmpeq_abs ()
> > > {
> > > return _mm256_set1_epi8 (1);
> > > }
> > >
> > > is now generated (with -O3 -march=x86-64-v3) as:
> > >
> > > vpcmpeqd %ymm0, %ymm0, %ymm0
> > > vpabsb %ymm0, %ymm0
> > > ret
> > >
> > > and
> > >
> > > __m256i
> > > should_be_cmpeq_add ()
> > > {
> > > return _mm256_set1_epi8 (-2);
> > > }
> > >
> > > is now generated as:
> > >
> > > vpcmpeqd %ymm0, %ymm0, %ymm0
> > > vpaddb %ymm0, %ymm0, %ymm0
> > > ret
> > >
> > > This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
> > > and make -k check, both with and without --target_board=unix{-m32}
> > > with no new failures. Ok for mainline?
> > >
> > >
> > > 2024-01-16 Roger Sayle <roger@nextmovesoftware.com>
> > >
> > > gcc/ChangeLog
> > > PR target/106060
> > > * config/i386/i386-expand.cc (enum ix86_vec_bcast_alg): New.
> > > (struct ix86_vec_bcast_map_simode_t): New type for table below.
> > > (ix86_vec_bcast_map_simode): Table of SImode constants that may
> > > be efficiently synthesized by a ix86_vec_bcast_alg method.
> > > (ix86_vec_bcast_map_simode_cmp): New comparator for bsearch.
> > > (ix86_vector_duplicate_simode_const): Efficiently synthesize
> > > V4SImode and V8SImode constants that duplicate special constants.
> > > (ix86_vector_duplicate_value): Attempt to synthesize "special"
> > > vector constants using ix86_vector_duplicate_simode_const.
> > > * config/i386/i386.cc (ix86_rtx_costs) <case ABS>: ABS of a
> > > vector integer mode costs with a single SSE instruction.
> > >
> >
> > + switch (entry->alg)
> > + {
> > + case VEC_BCAST_PXOR:
> > + if (mode == V8SImode && !TARGET_AVX2) return false;
> > + emit_move_insn (target, CONST0_RTX (mode));
> > + return true;
> > + case VEC_BCAST_PCMPEQ:
> > + if ((mode == V4SImode && !TARGET_SSE2)
> > + || (mode == V8SImode && !TARGET_AVX2)) return false;
> > + emit_move_insn (target, CONSTM1_RTX (mode));
> > + return true;
> >
> > I think we need to prevent those standard_sse_constant_p getting in
> > ix86_expand_vector_init_duplicate by below codes.
> >
> > /* If all values are identical, broadcast the value. */
> > if (all_same
> > && (nvars != 0 || !standard_sse_constant_p (gen_rtx_CONST_VECTOR
> > (mode, XVEC (vals, 0)), mode))
> > && ix86_expand_vector_init_duplicate (mmx_ok, mode, target,
> > XVECEXP (vals, 0, 0)))
> > return;
> >
> > + case VEC_BCAST_PABSB:
> > + if (mode == V4SImode)
> > + {
> > + tmp1 = gen_reg_rtx (V16QImode);
> > + emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
> > + tmp2 = gen_reg_rtx (V16QImode);
> > + emit_insn (gen_absv16qi2 (tmp2, tmp1));
> > Shouldn't it rely on TARGET_SSE2?
> >
> > + case VEC_BCAST_PADDB:
> > + if (mode == V4SImode)
> > + {
> > + tmp1 = gen_reg_rtx (V16QImode);
> > + emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
> > + tmp2 = gen_reg_rtx (V16QImode);
> > + emit_insn (gen_addv16qi3 (tmp2, tmp1, tmp1));
> > Ditto here and for all logic shift cases.
> > + }
> >
> > +
> > + if ((mode == V4SImode || mode == V8SImode)
> > + && CONST_INT_P (val)
> > + && ix86_vector_duplicate_simode_const (mode, target, INTVAL (val)))
> > + return true;
> > +
> > The alternative way is adding a pre_reload define_insn_and_split to match
> > specific const_vector and splitt it into new instructions.
> > In theoritically, the constant info can be retained before combine and will enable
> > more simplication.
> >
> > Also the patch can be extend to V16SImode, but it can be a separate patch.
> >
> > > gcc/testsuite/ChangeLog
> > > PR target/106060
> > > * gcc.target/i386/auto-init-8.c: Update test case.
> > > * gcc.target/i386/avx512fp16-3.c: Likewise.
> > > * gcc.target/i386/pr100865-9a.c: Likewise.
> > > * gcc.target/i386/pr106060-1.c: New test case.
> > > * gcc.target/i386/pr106060-2.c: Likewise.
> > > * gcc.target/i386/pr106060-3.c: Likewise.
> > > * gcc.target/i386/pr70314-3.c: Update test case.
> > > * gcc.target/i386/vect-shiftv4qi.c: Likewise.
> > > * gcc.target/i386/vect-shiftv8qi.c: Likewise.
> > >
> > >
> > > Thanks in advance,
> > > Roger
> > > --
> > >
> >
> >
> > --
> > BR,
> > Hongtao
@@ -15638,6 +15638,288 @@ s4fma_expand:
gcc_unreachable ();
}
+/* See below where shifts are handled for explanation of this enum. */
+enum ix86_vec_bcast_alg
+{
+ VEC_BCAST_PXOR,
+ VEC_BCAST_PCMPEQ,
+ VEC_BCAST_PABSB,
+ VEC_BCAST_PADDB,
+ VEC_BCAST_PSRLW,
+ VEC_BCAST_PSRLD,
+ VEC_BCAST_PSLLW,
+ VEC_BCAST_PSLLD
+};
+
+struct ix86_vec_bcast_map_simode_t
+{
+ unsigned int key;
+ enum ix86_vec_bcast_alg alg;
+ unsigned int arg;
+};
+
+/* This table must be kept sorted as values are looked-up using bsearch. */
+static const ix86_vec_bcast_map_simode_t ix86_vec_bcast_map_simode[] = {
+ { 0x00000000, VEC_BCAST_PXOR, 0 },
+ { 0x00000001, VEC_BCAST_PSRLD, 31 },
+ { 0x00000003, VEC_BCAST_PSRLD, 30 },
+ { 0x00000007, VEC_BCAST_PSRLD, 29 },
+ { 0x0000000f, VEC_BCAST_PSRLD, 28 },
+ { 0x0000001f, VEC_BCAST_PSRLD, 27 },
+ { 0x0000003f, VEC_BCAST_PSRLD, 26 },
+ { 0x0000007f, VEC_BCAST_PSRLD, 25 },
+ { 0x000000ff, VEC_BCAST_PSRLD, 24 },
+ { 0x000001ff, VEC_BCAST_PSRLD, 23 },
+ { 0x000003ff, VEC_BCAST_PSRLD, 22 },
+ { 0x000007ff, VEC_BCAST_PSRLD, 21 },
+ { 0x00000fff, VEC_BCAST_PSRLD, 20 },
+ { 0x00001fff, VEC_BCAST_PSRLD, 19 },
+ { 0x00003fff, VEC_BCAST_PSRLD, 18 },
+ { 0x00007fff, VEC_BCAST_PSRLD, 17 },
+ { 0x0000ffff, VEC_BCAST_PSRLD, 16 },
+ { 0x00010001, VEC_BCAST_PSRLW, 15 },
+ { 0x0001ffff, VEC_BCAST_PSRLD, 15 },
+ { 0x00030003, VEC_BCAST_PSRLW, 14 },
+ { 0x0003ffff, VEC_BCAST_PSRLD, 14 },
+ { 0x00070007, VEC_BCAST_PSRLW, 13 },
+ { 0x0007ffff, VEC_BCAST_PSRLD, 13 },
+ { 0x000f000f, VEC_BCAST_PSRLW, 12 },
+ { 0x000fffff, VEC_BCAST_PSRLD, 12 },
+ { 0x001f001f, VEC_BCAST_PSRLW, 11 },
+ { 0x001fffff, VEC_BCAST_PSRLD, 11 },
+ { 0x003f003f, VEC_BCAST_PSRLW, 10 },
+ { 0x003fffff, VEC_BCAST_PSRLD, 10 },
+ { 0x007f007f, VEC_BCAST_PSRLW, 9 },
+ { 0x007fffff, VEC_BCAST_PSRLD, 9 },
+ { 0x00ff00ff, VEC_BCAST_PSRLW, 8 },
+ { 0x00ffffff, VEC_BCAST_PSRLD, 8 },
+ { 0x01010101, VEC_BCAST_PABSB, 0 },
+ { 0x01ff01ff, VEC_BCAST_PSRLW, 7 },
+ { 0x01ffffff, VEC_BCAST_PSRLD, 7 },
+ { 0x03ff03ff, VEC_BCAST_PSRLW, 6 },
+ { 0x03ffffff, VEC_BCAST_PSRLD, 6 },
+ { 0x07ff07ff, VEC_BCAST_PSRLW, 5 },
+ { 0x07ffffff, VEC_BCAST_PSRLD, 5 },
+ { 0x0fff0fff, VEC_BCAST_PSRLW, 4 },
+ { 0x0fffffff, VEC_BCAST_PSRLD, 4 },
+ { 0x1fff1fff, VEC_BCAST_PSRLW, 3 },
+ { 0x1fffffff, VEC_BCAST_PSRLD, 3 },
+ { 0x3fff3fff, VEC_BCAST_PSRLW, 2 },
+ { 0x3fffffff, VEC_BCAST_PSRLD, 2 },
+ { 0x7fff7fff, VEC_BCAST_PSRLW, 1 },
+ { 0x7fffffff, VEC_BCAST_PSRLD, 1 },
+ { 0x80000000, VEC_BCAST_PSLLD, 31 },
+ { 0x80008000, VEC_BCAST_PSLLW, 15 },
+ { 0xc0000000, VEC_BCAST_PSLLD, 30 },
+ { 0xc000c000, VEC_BCAST_PSLLW, 14 },
+ { 0xe0000000, VEC_BCAST_PSLLD, 29 },
+ { 0xe000e000, VEC_BCAST_PSLLW, 13 },
+ { 0xf0000000, VEC_BCAST_PSLLD, 28 },
+ { 0xf000f000, VEC_BCAST_PSLLW, 12 },
+ { 0xf8000000, VEC_BCAST_PSLLD, 27 },
+ { 0xf800f800, VEC_BCAST_PSLLW, 11 },
+ { 0xfc000000, VEC_BCAST_PSLLD, 26 },
+ { 0xfc00fc00, VEC_BCAST_PSLLW, 10 },
+ { 0xfe000000, VEC_BCAST_PSLLD, 25 },
+ { 0xfe00fe00, VEC_BCAST_PSLLW, 9 },
+ { 0xfefefefe, VEC_BCAST_PADDB, 0 },
+ { 0xff000000, VEC_BCAST_PSLLD, 24 },
+ { 0xff00ff00, VEC_BCAST_PSLLW, 8 },
+ { 0xff800000, VEC_BCAST_PSLLD, 23 },
+ { 0xff80ff80, VEC_BCAST_PSLLW, 7 },
+ { 0xffc00000, VEC_BCAST_PSLLD, 22 },
+ { 0xffc0ffc0, VEC_BCAST_PSLLW, 6 },
+ { 0xffe00000, VEC_BCAST_PSLLD, 21 },
+ { 0xffe0ffe0, VEC_BCAST_PSLLW, 5 },
+ { 0xfff00000, VEC_BCAST_PSLLD, 20 },
+ { 0xfff0fff0, VEC_BCAST_PSLLW, 4 },
+ { 0xfff80000, VEC_BCAST_PSLLD, 19 },
+ { 0xfff8fff8, VEC_BCAST_PSLLW, 3 },
+ { 0xfffc0000, VEC_BCAST_PSLLD, 18 },
+ { 0xfffcfffc, VEC_BCAST_PSLLW, 2 },
+ { 0xfffe0000, VEC_BCAST_PSLLD, 17 },
+ { 0xfffefffe, VEC_BCAST_PSLLW, 1 },
+ { 0xffff0000, VEC_BCAST_PSLLD, 16 },
+ { 0xffff8000, VEC_BCAST_PSLLD, 15 },
+ { 0xffffc000, VEC_BCAST_PSLLD, 14 },
+ { 0xffffe000, VEC_BCAST_PSLLD, 13 },
+ { 0xfffff000, VEC_BCAST_PSLLD, 12 },
+ { 0xfffff800, VEC_BCAST_PSLLD, 11 },
+ { 0xfffffc00, VEC_BCAST_PSLLD, 10 },
+ { 0xfffffe00, VEC_BCAST_PSLLD, 9 },
+ { 0xffffff00, VEC_BCAST_PSLLD, 8 },
+ { 0xffffff80, VEC_BCAST_PSLLD, 7 },
+ { 0xffffffc0, VEC_BCAST_PSLLD, 6 },
+ { 0xffffffe0, VEC_BCAST_PSLLD, 5 },
+ { 0xfffffff0, VEC_BCAST_PSLLD, 4 },
+ { 0xfffffff8, VEC_BCAST_PSLLD, 3 },
+ { 0xfffffffc, VEC_BCAST_PSLLD, 2 },
+ { 0xfffffffe, VEC_BCAST_PSLLD, 1 },
+ { 0xffffffff, VEC_BCAST_PCMPEQ, 0 }
+};
+
+/* Comparator for bsearch on ix86_vec_bcast_map. */
+static int
+ix86_vec_bcast_map_simode_cmp (const void *key, const void *entry)
+{
+ return (*(const unsigned int*)key)
+ - ((const ix86_vec_bcast_map_simode_t*)entry)->key;
+}
+
+/* A subroutine of ix86_vector_duplicate_value. Tries to efficiently
+ materialize V4SImode and V8SImode vectors from SImode integer
+ constants. */
+static bool
+ix86_vector_duplicate_simode_const (machine_mode mode, rtx target,
+ unsigned int val)
+{
+ const ix86_vec_bcast_map_simode_t *entry;
+ rtx tmp1, tmp2;
+
+ entry = (const ix86_vec_bcast_map_simode_t*)
+ bsearch(&val, ix86_vec_bcast_map_simode,
+ ARRAY_SIZE (ix86_vec_bcast_map_simode),
+ sizeof (ix86_vec_bcast_map_simode_t),
+ ix86_vec_bcast_map_simode_cmp);
+ if (!entry)
+ return false;
+
+ switch (entry->alg)
+ {
+ case VEC_BCAST_PXOR:
+ if (mode == V8SImode && !TARGET_AVX2)
+ return false;
+ emit_move_insn (target, CONST0_RTX (mode));
+ return true;
+
+ case VEC_BCAST_PCMPEQ:
+ if ((mode == V4SImode && !TARGET_SSE2)
+ || (mode == V8SImode && !TARGET_AVX2))
+ return false;
+ emit_move_insn (target, CONSTM1_RTX (mode));
+ return true;
+
+ case VEC_BCAST_PABSB:
+ if (mode == V4SImode)
+ {
+ tmp1 = gen_reg_rtx (V16QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
+ tmp2 = gen_reg_rtx (V16QImode);
+ emit_insn (gen_absv16qi2 (tmp2, tmp1));
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V32QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V32QImode));
+ tmp2 = gen_reg_rtx (V32QImode);
+ emit_insn (gen_absv32qi2 (tmp2, tmp1));
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PADDB:
+ if (mode == V4SImode)
+ {
+ tmp1 = gen_reg_rtx (V16QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16QImode));
+ tmp2 = gen_reg_rtx (V16QImode);
+ emit_insn (gen_addv16qi3 (tmp2, tmp1, tmp1));
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V32QImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V32QImode));
+ tmp2 = gen_reg_rtx (V32QImode);
+ emit_insn (gen_addv32qi3 (tmp2, tmp1, tmp1));
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PSRLW:
+ if (mode == V4SImode)
+ {
+ tmp1 = gen_reg_rtx (V8HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V8HImode));
+ tmp2 = gen_reg_rtx (V8HImode);
+ emit_insn (gen_lshrv8hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V16HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16HImode));
+ tmp2 = gen_reg_rtx (V16HImode);
+ emit_insn (gen_lshrv16hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PSRLD:
+ if (mode == V4SImode)
+ {
+ tmp1 = gen_reg_rtx (V4SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V4SImode));
+ emit_insn (gen_lshrv4si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V8SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V8SImode));
+ emit_insn (gen_lshrv8si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PSLLW:
+ if (mode == V4SImode)
+ {
+ tmp1 = gen_reg_rtx (V8HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V8HImode));
+ tmp2 = gen_reg_rtx (V8HImode);
+ emit_insn (gen_ashlv8hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V16HImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V16HImode));
+ tmp2 = gen_reg_rtx (V16HImode);
+ emit_insn (gen_ashlv16hi3 (tmp2, tmp1, GEN_INT (entry->arg)));
+ }
+ else
+ return false;
+ break;
+
+ case VEC_BCAST_PSLLD:
+ if (mode == V4SImode)
+ {
+ tmp1 = gen_reg_rtx (V4SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V4SImode));
+ emit_insn (gen_ashlv4si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else if (mode == V8SImode && TARGET_AVX2)
+ {
+ tmp1 = gen_reg_rtx (V8SImode);
+ emit_move_insn (tmp1, CONSTM1_RTX (V8SImode));
+ emit_insn (gen_ashlv8si3 (target, tmp1, GEN_INT (entry->arg)));
+ return true;
+ }
+ else
+ return false;
+
+ default:
+ return false;
+ }
+
+ emit_move_insn (target, gen_lowpart (mode, tmp2));
+ return true;
+}
+
/* A subroutine of ix86_expand_vector_init_duplicate. Tries to
fill target with val via vec_duplicate. */
@@ -15647,6 +15929,12 @@ ix86_vector_duplicate_value (machine_mode mode, rtx target, rtx val)
bool ok;
rtx_insn *insn;
rtx dup;
+
+ if ((mode == V4SImode || mode == V8SImode)
+ && CONST_INT_P (val)
+ && ix86_vector_duplicate_simode_const (mode, target, INTVAL (val)))
+ return true;
+
/* Save/restore recog_data in case this is called from splitters
or other routines where recog_data needs to stay valid across
force_reg. See PR106577. */
@@ -22076,6 +22076,8 @@ ix86_rtx_costs (rtx x, machine_mode mode, int outer_code_i, int opno,
*total = cost->fabs;
else if (FLOAT_MODE_P (mode))
*total = ix86_vec_cost (mode, cost->sse_op);
+ else if (GET_MODE_CLASS (mode) == MODE_VECTOR_INT)
+ *total = cost->sse_op;
return false;
case SQRT:
@@ -29,7 +29,7 @@ double foo()
return result;
}
-/* { dg-final { scan-rtl-dump-times "0xfffffffffefefefe" 3 "expand" } } */
+/* { dg-final { scan-rtl-dump-times "0xfffffffffefefefe" 1 "expand" } } */
/* { dg-final { scan-rtl-dump-times "\\\[0xfefefefefefefefe\\\]" 2 "expand" } } */
/* { dg-final { scan-rtl-dump-times "0xfffffffffffffffe\\\]\\\) repeated x16" 2 "expand" } } */
@@ -126,7 +126,6 @@ abs256_ph (__m256h a)
return _mm256_abs_ph (a);
}
-/* { dg-final { scan-assembler-times "vpbroadcastd\[^\n\]*%ymm\[0-9\]+" 1 { target { ! ia32 } } } } */
/* { dg-final { scan-assembler-times "vpand\[^\n\]*%ymm\[0-9\]+" 1 } } */
__m128h
@@ -136,5 +135,4 @@ abs_ph (__m128h a)
return _mm_abs_ph (a);
}
-/* { dg-final { scan-assembler-times "vpbroadcastd\[^\n\]*%xmm\[0-9\]+" 1 { target { ! ia32 } } } } */
/* { dg-final { scan-assembler-times "vpand\[^\n\]*%xmm\[0-9\]+" 1 } } */
@@ -18,7 +18,7 @@ foo (void)
{
int i;
for (i = 0; i < sizeof (array) / sizeof (array[0]); i++)
- array[i] = MK_CONST128_BROADCAST (0x1fff);
+ array[i] = MK_CONST128_BROADCAST (0x1234);
}
/* { dg-final { scan-assembler-times "vpbroadcastd\[\\t \]+%xmm\[0-9\]+, %xmm\[0-9\]+" 1 } } */
new file mode 100644
@@ -0,0 +1,12 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=x86-64-v3" } */
+#include <immintrin.h>
+
+__m256i
+foo ()
+{
+ /* shouldnt_have_movabs */
+ return _mm256_set1_epi8 (123);
+}
+
+/* { dg-final { scan-assembler-not "movabs" } } */
new file mode 100644
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=x86-64-v3" } */
+#include <immintrin.h>
+
+__m256i
+foo ()
+{
+ /* should_be_cmpeq_abs */
+ return _mm256_set1_epi8 (1);
+}
+
+/* { dg-final { scan-assembler "pcmpeq" } } */
+/* { dg-final { scan-assembler "pabsb" } } */
new file mode 100644
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 -march=x86-64-v3" } */
+#include <immintrin.h>
+
+__m256i
+foo ()
+{
+ /* should_be_cmpeq_add */
+ return _mm256_set1_epi8 (-2);
+}
+
+/* { dg-final { scan-assembler "pcmpeq" } } */
+/* { dg-final { scan-assembler "paddb" } } */
+
@@ -1,6 +1,6 @@
/* { dg-do compile } */
/* { dg-options "-march=skylake-avx512 -O2" } */
-/* { dg-final { scan-assembler-times "cmp" 2 } } */
+/* { dg-final { scan-assembler-times "cmp\[dq\]" 2 } } */
/* { dg-final { scan-assembler-not "and" } } */
typedef long vec __attribute__((vector_size(16)));
@@ -28,7 +28,7 @@ __vu srl_c (__vu a)
return a >> 5;
}
-/* { dg-final { scan-assembler-times "psrlw" 2 } } */
+/* { dg-final { scan-assembler-times "psrlw" 5 } } */
__vi sra (__vi a, int n)
{
@@ -28,7 +28,7 @@ __vu srl_c (__vu a)
return a >> 5;
}
-/* { dg-final { scan-assembler-times "psrlw" 2 } } */
+/* { dg-final { scan-assembler-times "psrlw" 5 } } */
__vi sra (__vi a, int n)
{