https://gcc.gnu.org/g:a59c4e496fa916cb9a484a649aa1b4cebd6550f2
commit r15-2430-ga59c4e496fa916cb9a484a649aa1b4cebd6550f2
Author: Hongyu Wang
Date: Fri Jul 26 08:27:01 2024 +0800
i386: Mark target option with optimization when enabled with opt level
[PR116065]
When introducing
Richard Biener 于2024年7月26日周五 19:45写道:
>
> On Fri, Jul 26, 2024 at 10:50 AM Hongyu Wang wrote:
> >
> > Hi,
> >
> > When introducing munroll-only-small-loops, the option was marked as
> > Target Save and added to -O2 default which makes attribute(optimize)
Hi,
When introducing munroll-only-small-loops, the option was marked as
Target Save and added to -O2 default which makes attribute(optimize)
resets target option and causing error when cmdline has O1 and
funciton attribute has O2 and other target options. Mark this option
as Optimization to fix.
https://gcc.gnu.org/g:681ff5ccca153864eb86099eed201838d8d98bc2
commit r15-2037-g681ff5ccca153864eb86099eed201838d8d98bc2
Author: Hongyu Wang
Date: Thu Apr 18 16:53:26 2024 +0800
[APX NF] Add a pass to convert legacy insn to NF insns
For APX ccmp, current infrastructure
https://gcc.gnu.org/g:02a3bf5e2f0c18078bf67fc0002219edba1d76ff
commit r15-2030-g02a3bf5e2f0c18078bf67fc0002219edba1d76ff
Author: Hongyu Wang
Date: Sat Jul 13 11:45:31 2024 +0800
AVX512BF16: Do not allow permutation with vcvtne2ps2bf16 [PR115889]
According to the instruction spec
> Could you just git revert 6d0b7b69d143025f271d0041cfa29cf26e6c343b?
We can still deal with BFmode permutation the same way as HFmode, so
the change in ix86_vectorize_vec_perm_const can be preserved.
Hongtao Liu 于2024年7月15日周一 09:40写道:
>
> On Sat, Jul 13, 2024 at 3:44 PM Hongyu Wa
Hi,
According to the instruction spec of AVX512BF16, the convert from float
to BF16 is not a simple truncation. It has special handling for
denormal/nan, even for normal float it will add an extra bias according
to the least significant bit for bf number. This means we cannot use the
Hi,
For APX ccmp, current infrastructure will always generate cstore for
the ccmp flag user, like
cmpe%rcx, %r8
ccmpnel %rax, %rbx
seta%dil
add %rcx, %r9
add %r9, %rdx
testb %dil, %dil
je .L2
For such case, the
https://gcc.gnu.org/g:8e72b1bb3896f6e8d4f4679cbcfbc2a8212d04f9
commit r15-1833-g8e72b1bb3896f6e8d4f4679cbcfbc2a8212d04f9
Author: Hongyu Wang
Date: Wed Feb 7 14:42:58 2024 +0800
[APX PPX] Avoid generating unmatched pushp/popp in pro/epilogue
According to APX spec, the pushp/popp
apx spec, the mismatched
pushp/popp pair does confused the fast-forwarding logic and turns off
the PPX optimization. We just need to make sure every pushp for a
certain reg has corresponding popp for that reg.
Richard Biener 于2024年7月2日周二 16:18写道:
>
> On Tue, Jul 2, 2024 at 5:24 AM Hongyu Wang wr
Hi,
According to APX spec, the pushp/popp pairs should be matched,
otherwise the PPX hint cannot take effect and cause performance loss.
In the ix86_expand_epilogue, there are several optimizations that may
cause the epilogue using mov to restore the regs. Check if PPX applied
and prevent usage
https://gcc.gnu.org/g:4867cc815531ede8bc356a2507f1c35ee6e6399c
commit r15-1469-g4867cc815531ede8bc356a2507f1c35ee6e6399c
Author: Hongyu Wang
Date: Mon Jun 17 10:34:01 2024 +0800
i386: Fix some ISA bit test in option_override
Adjust several new feature check
Hi,
This patch adjusts several new feature check in ix86_option_override_interal
that directly use TARGET_* instead of TARGET_*_P (opts->ix86_isa_flags),
which caused cmdline option overrides target_attribute isa flag.
Bootstrapped && regtested on x86_64-pc-linux-gnu.
Ok for trunk?
https://gcc.gnu.org/g:83a765768510d1f329887116757d6818d7846717
commit r15-1293-g83a765768510d1f329887116757d6818d7846717
Author: Hongyu Wang
Date: Thu Jun 13 00:18:32 2024 +0800
[APX CCMP] Add targetm.have_ccmp hook [PR115370]
In cfgexpand, there is an optimization for branch
Thanks, this it the patch I'm going to check-in.
Richard Sandiford 于2024年6月13日周四 17:04写道:
>
> Hongyu Wang writes:
> > Hi,
> >
> > In cfgexpand, there is an optimization for branch which tests
> > targetm.gen_ccmp_first == NULL. However for target like x86-64
Sorry for breaking the original logic, and very appreciate for your
patch!! It does makes the logic more clear on top of opts and
opts_set.
I think the function name can be like ix86_unroll_flag_adjust instead
of ix86_override_options_after_change_1, like the previous 2 functions
which declares
https://gcc.gnu.org/g:99e6cf404e37655be303e71f20df03c284c7989e
commit r15-1242-g99e6cf404e37655be303e71f20df03c284c7989e
Author: Hongyu Wang
Date: Thu May 9 10:12:16 2024 +0800
[APX CCMP] Use ctestcc when comparing to const 0
For CTEST, we don't have conditional AND so there's
> Perhaps the constraint can be slightly optimized to avoid repeating
> (,) pairs.
>
> ",m,"
> "C ,,"
Yes, will check-in with this change. Thanks!
Uros Bizjak 于2024年6月13日周四 14:06写道:
>
> On Thu, Jun 13, 2024 at 3:44 AM Hongyu Wang wrote:
>
Hi,
In cfgexpand, there is an optimization for branch which tests
targetm.gen_ccmp_first == NULL. However for target like x86-64, the
hook was implemented but it does not indicate that ccmp was enabled.
Add a new target hook TARGET_HAVE_CCMP and replace the middle-end
check for the existance of
Thanks for the advice, updated patch in attachment.
Bootstrapped/regtested on x86-64-pc-linux-gnu. Ok for trunk?
Uros Bizjak 于2024年6月12日周三 18:12写道:
>
> On Wed, Jun 12, 2024 at 12:00 PM Uros Bizjak wrote:
> >
> > On Wed, Jun 12, 2024 at 5:12 AM Hongyu Wang wro
Hi,
For CTEST, we don't have conditional AND so there's no optimization
opportunity to write a new ctest pattern. Emit ctest when ccmp did
comparison to const 0 to save bytes.
Bootstrapped & regtested under x86-64-pc-linux-gnu.
Ok for trunk?
gcc/ChangeLog:
* config/i386/i386.md
https://gcc.gnu.org/g:0b6cea8783b9e1b86c5c7c277c301cb5931bc5e0
commit r15-1060-g0b6cea8783b9e1b86c5c7c277c301cb5931bc5e0
Author: Hongyu Wang
Date: Wed May 8 11:08:42 2024 +0800
[APX CCMP] Support ccmp for float compare
The ccmp insn itself doesn't support fp compare, but x86 has
https://gcc.gnu.org/g:23db87301b623ecf162c9df718ce82ed9aa354a8
commit r15-1059-g23db87301b623ecf162c9df718ce82ed9aa354a8
Author: Hongyu Wang
Date: Tue Apr 9 16:05:26 2024 +0800
[APX CCMP] Adjust startegy for selecting ccmp candidates
For general ccmp scenario, the tree sequence
https://gcc.gnu.org/g:c989e59fc99d994159114304d4e715c72bedff0a
commit r15-1058-gc989e59fc99d994159114304d4e715c72bedff0a
Author: Hongyu Wang
Date: Wed Mar 27 10:13:06 2024 +0800
[APX CCMP] Support APX CCMP
APX CCMP feature implements conditional compare which executes compare
https://gcc.gnu.org/g:f46d54a2a76acb94356989fb187853e5b58c3098
commit r15-1057-gf46d54a2a76acb94356989fb187853e5b58c3098
Author: Hongyu Wang
Date: Thu Jun 6 13:00:26 2024 +0800
[APX] Adjust target-support check [PR 115341]
Current target apxf check does not specify sub-features
st. The costs are not
+ meaningful for failed expansions. */
+
+ if (ret2 && (!ret || cost2 < cost1))
{
*prep_seq = prep_seq_2;
*gen_seq = gen_seq_2;
--
2.31.1
Richard Sandiford 于2024年6月5日周三 17:21写道:
>
> Hongyu Wang writes:
> > CC'd Richard for ccmp
Current target apxf check does not specify sub-features that assembler
supports, so the check with older binutils will fail at assemble stage
for new apx features like NF,CCMP or CFCMOV. Adjust the assembler check
for latest apx subfeatures.
Bootstrapped & regtested on x86-64-pc-linux-gnu with
https://gcc.gnu.org/g:173f8763a66622f2a70ad66f60573fcff7d6b49e
commit r13-8811-g173f8763a66622f2a70ad66f60573fcff7d6b49e
Author: Hongyu Wang
Date: Wed May 15 11:24:34 2024 +0800
i386: Fix ix86_option override after change [PR 113719]
In ix86_override_options_after_change, calls
https://gcc.gnu.org/g:cd161b335c2723d0dce1cab00ad216b423ec2767
commit r14-10262-gcd161b335c2723d0dce1cab00ad216b423ec2767
Author: Hongyu Wang
Date: Wed May 15 11:24:34 2024 +0800
i386: Fix ix86_option override after change [PR 113719]
In ix86_override_options_after_change, calls
Gently ping :)
Hi Richard, Is it OK to adopt the ccmp change? Or did you know who can
help to review this part?
Thanks.
Hongyu Wang 于2024年5月23日周四 16:27写道:
>
> Gently ping for this :)
> Hi Richard, Is it OK to adopt the ccmp change? Or did you know who can
> help to review this pa
https://gcc.gnu.org/g:499d00127d39ba894b0f7216d73660b380bdc325
commit r15-893-g499d00127d39ba894b0f7216d73660b380bdc325
Author: Hongyu Wang
Date: Wed May 15 11:24:34 2024 +0800
i386: Fix ix86_option override after change [PR 113719]
In ix86_override_options_after_change, calls
Gently ping for this :)
Hi Richard, Is it OK to adopt the ccmp change? Or did you know who can
help to review this part?
Thanks.
Hongyu Wang 于2024年5月15日周三 16:25写道:
>
> CC'd Richard for ccmp part as previously it is added only for aarch64.
> The original logic will not interrup
Richard Biener 于2024年5月16日周四 15:05写道:
>
> On Thu, May 16, 2024 at 8:25 AM Hongyu Wang wrote:
> >
> > Hi,
> >
> > In ix86_override_options_after_change, calls to ix86_default_align
> > and ix86_recompute_optlev_based_flags will cause mism
Hi,
In ix86_override_options_after_change, calls to ix86_default_align
and ix86_recompute_optlev_based_flags will cause mismatched target
opt_set when doing cl_optimization_restore. Move them back to
ix86_option_override_internal to solve the issue.
Bootstrapped & regtested on
supports but ccmp not,
so ret/ret2 will all be valid when comparing cost.
Thanks in advance.
Hongyu Wang 于2024年5月15日周三 16:22写道:
>
> For general ccmp scenario, the tree sequence is like
>
> _1 = (a < b)
> _2 = (c < d)
> _3 = _1 & _2
>
> current ccmp expanding wil
APX CCMP feature implements conditional compare which executes compare
when EFLAGS matches certain condition.
CCMP introduces default flags value (dfv), when conditional compare does
not execute, it will directly set the flags according to dfv.
The instruction goes like
ccmpeq {dfv=sf,of,cf,zf}
For general ccmp scenario, the tree sequence is like
_1 = (a < b)
_2 = (c < d)
_3 = _1 & _2
current ccmp expanding will try to swap compare order for _1 and _2,
compare the cost/cost2 between compare _1 and _2 first, then return the
sequence with lower cost.
For x86 ccmp, we don't support FP
The ccmp insn itself doesn't support fp compare, but x86 has fp comi
insn that changes EFLAG which can be the scc input to ccmp. Allow
scalar fp compare in ix86_gen_ccmp_first except ORDERED/UNORDERD
compare which can not be identified in ccmp.
gcc/ChangeLog:
* config/i386/i386-expand.cc
html
Hongyu Wang (3):
[APX CCMP] Support APX CCMP
[APX CCMP] Adjust startegy for selecting ccmp candidates
[APX CCMP] Support ccmp for float compare
gcc/ccmp.cc| 12 +-
gcc/config/i386/i386-expand.cc | 164 +
gcc/config/i386/
https://gcc.gnu.org/g:ea665f90260acb3ffd2e39fcd2e200e702ee0ead
commit r14-9882-gea665f90260acb3ffd2e39fcd2e200e702ee0ead
Author: Hongyu Wang
Date: Tue Apr 9 09:50:11 2024 +0800
[APX] Prohibit SHA/KEYLOCKER usage of EGPR when APX enabled
The latest APX spec announced removal
The latest APX spec announced removal of SHA/KEYLOCKER evex promotion [1],
which means the SHA/KEYLOCKER insn does not support EGPR when APX
enabled. Update the corresponding constraints to their EGPR-disabled
counterparts.
Bootstrapped and regtested on x86-64-pc-linux-gnu.
Ok for trunk?
Thanks for fixing this! Didn't notice that the pointer conversion can
cause this issue...
Was it possible to use local array like
char a[64] = (char *)p
__asm__ volatile ("ldtilecfg\t%X0" :: "m" (a)));
If not, for the two patterns we can use "m" instead of "jm" as APX
supports EGPR extension
I'm going to check-in this if no objection
Hongyu Wang 于2024年1月9日周二 15:14写道:
>
> Hi,
>
> This patch adds missing description for inline asm behavior and related
> compiler switch for APX.
>
> Ok for gcc-wwwdocs?
>
> ---
> htdocs/gcc-14/changes.html | 6 ++
&
Thanks, this is the patch I'm going to check-in
Hongtao Liu 于2024年1月10日周三 16:02写道:
>
> On Tue, Jan 9, 2024 at 3:09 PM Hongyu Wang wrote:
> >
> > Hi,
> >
> > For APX, the inline asm behavior was not mentioned in any document
> > before. Add description for i
Hi,
This patch adds missing description for inline asm behavior and related
compiler switch for APX.
Ok for gcc-wwwdocs?
---
htdocs/gcc-14/changes.html | 6 ++
1 file changed, 6 insertions(+)
diff --git a/htdocs/gcc-14/changes.html b/htdocs/gcc-14/changes.html
index e3a68998..73a90d30
Hi,
For APX, the inline asm behavior was not mentioned in any document
before. Add description for it.
Ok for trunk?
gcc/ChangeLog:
* config/i386/i386.opt: Adjust document.
* doc/invoke.texi: Add description for
-mapx-inline-asm-use-gpr32.
---
gcc/config/i386/i386.opt
Hi,
The supported sub-features for APX was missing in option document and
target attribute section. Add those missing ones.
Ok for trunk?
gcc/ChangeLog:
* config/i386/i386.opt: Add supported sub-features.
* doc/extend.texi: Add description for target attribute.
---
Hi,
As Coudert points out, this test fails on darwin as it does not
support _Decimal64, so require dfp for it.
Pushed as obvious fix.
gcc/testsuite/ChangeLog:
* gcc.target/i386/pr112943.c: Require dfp.
---
gcc/testsuite/gcc.target/i386/pr112943.c | 2 +-
1 file changed, 1
Hi,
Currently move_max follows the tuning feature first, but ideally it
should sync with prefer-vector-width when it is explicitly set to keep
vector move and operation with same vector size.
Bootstrapped/regtested on x86-64-pc-linux-gnu{-m32,}
OK for trunk?
gcc/ChangeLog:
PR
> > +__int128 u128_2 = (9223372036854775808 << 4) * foo0_u8_0; /* {
> > dg-warning "integer constant is so large that it is unsigned" "so large" }
> > */
>
> Just you can use (9223372036854775807LL + (__int128) 1) instead of
> 9223372036854775808
> to avoid the warning.
> The testcase will
Hi,
The ashl/lshr/ashr expanders calls ix86_expand_binary_operator, while
they will be called for some post-reload split, and TARGET_APX_NDD is
required for these calls to avoid force-load to memory at postreload
stage.
Bootstrapped/regtested on x86-64-pc-linux-gnu{-m32,}
Ok for master?
gcc/ChangeLog:
* config/i386/i386.md (*3_1): Extend with a new
alternative to support NDD for SI/DI rotate, and adjust output
template.
(*si3_1_zext): Likewise.
(*3_1): Likewise for QI/HI modes.
(rcrsi2): Likewise, and use nonimmediate_operand for
For TImode shifts, they are splitted by splitter functions, which assume
operands[0] and operands[1] to be the same. For the NDD alternative the
assumption may not be true so add split functions for NDD to emit the NDD
form instructions, and omit the handling of !64bit target split.
Although the
From: Kong Lingling
gcc/ChangeLog:
* config/i386/i386-expand.cc (ix86_fixup_binary_operands_no_copy):
Add use_ndd parameter and parse it.
* config/i386/i386-protos.h (ix86_fixup_binary_operands_no_copy):
Change define.
* config/i386/i386.md (sub3): Add
Similar to LSHIFT, rshift do not need to omit $1 for NDD form.
gcc/ChangeLog:
* config/i386/i386.md (ashr3_cvt): Extend with new
alternatives to support NDD, and adjust output templates.
(*ashr3_1): Likewise for SI/DI mode.
(*lshr3_1): Likewise.
For left shift, there is an optimization TARGET_DOUBLE_WITH_ADD that shl
1 can be optimized to add. As NDD form of add requires src operand to
be register since NDD cannot take 2 memory src, we currently just keep
using NDD form shift instead of add.
The optimization TARGET_SHIFT1 will try to
From: Kong Lingling
gcc/ChangeLog:
* config/i386/i386.md: (addsi_1_zext): Add new alternatives for
NDD and adjust output templates.
(*add_2): Likewise.
(*addsi_2_zext): Likewise.
(*add_3): Likewise.
(*addsi_3_zext): Likewise.
(*adddi_4):
For shld/shrd insns, the old pattern use match_dup 0 as its shift src and use
+r*m as its constraint. To support NDD we added new define_insns to handle NDD
form pattern with extra input and dest operand to be fixed in register.
gcc/ChangeLog:
* config/i386/i386.md (x86_64_shld_ndd): New
gcc/ChangeLog:
* config/i386/i386.md (*movcc_noc): Extend with new constraints
to support NDD.
(*movsicc_noc_zext): Likewise.
(*movsicc_noc_zext_1): Likewise.
(*movqicc_noc): Likewise.
gcc/testsuite/ChangeLog:
* gcc.target/i386/apx-ndd-cmov.c: New
From: Kong Lingling
For NDD form AND insn, there are three splitter fixes after extending legacy
patterns.
1. APX NDD does not support high QImode registers like ah, bh, ch, dh, so for
some optimization splitters that generates highpart zero_extract for QImode
need to be prohibited under NDD
From: Kong Lingling
gcc/ChangeLog:
* config/i386/i386-expand.cc (ix86_expand_unary_operator): Add use_ndd
parameter and adjust for NDD.
* config/i386/i386-protos.h: Add use_ndd parameter for
ix86_unary_operator_ok and ix86_expand_unary_operator.
*
From: Kong Lingling
Similar to *add3_doubleword, operands[1] may not equal to operands[0] so
extra move and earlyclobber are required.
gcc/ChangeLog:
* config/i386/i386.md (*sub3_doubleword): Add new alternative for
NDD, adopt '&' modifier to NDD dest and emit move when
From: Kong Lingling
Legacy adc patterns are commonly adopted to TImode add, when extending TImode
add to NDD version, operands[0] and operands[1] can be different, so extra move
should be emitted if those patterns have optimization when adding const0_rtx.
For TImode insn, there could be
From: Kong Lingling
Similar to AND insn, two splitters need to be adjusted to prevent
misoptimizaiton for NDD OR/XOR.
Also adjust *one_cmplsi2_2_zext and its corresponding splitter that will
generate xor insn.
gcc/ChangeLog:
* config/i386/i386.md (3): Add new alternative for NDD
NDD uses evex prefix, so when segment prefix is also applied, the instruction
could excceed its 15byte limit, especially adding immediates. This could happen
when "e" constraint accepts any UNSPEC_TPOFF/UNSPEC_NTPOFF constant and it will
add the offset to segment register, which will be encoded
From: Kong Lingling
APX NDD provides an extra destination register operand for several gpr
related legacy insns, so a new alternative can be adopted to operand1
with "r" constraint.
This first patch supports NDD for add instruction, and keeps to use lea
when all operands are registers since lea
From: Kong Lingling
For *one_cmplsi2_2_zext, it will be splitted to xor, so its NDD form will be
added together with xor NDD support.
gcc/ChangeLog:
* config/i386/i386.md (one_cmpl2): Add new constraints for NDD
and adjust output template.
(*one_cmpl2_1): Likewise.
) == ISA_APX_NDD instead of
checking alternative at asm output stage.
Bootstrapped & regtested on x86_64-pc-linux-gnu{-m32,} and sde.
Ok for master?
Hongyu Wang (7):
[APX NDD] Disable seg_prefixed memory usage for NDD add
[APX NDD] Support APX NDD for left shift insns
[APX NDD] Support
Uros Bizjak 于2023年12月5日周二 18:46写道:
>
> On Tue, Dec 5, 2023 at 3:29 AM Hongyu Wang wrote:
> >
> > Under APX NDD, previous TImode allocation will have issue that it was
> > originally allocated using continuous pair, like rax:rdi, rdi:rdx.
> >
> > This will cau
From: Kong Lingling
For NDD form AND insn, there are three splitter fixes after extending legacy
patterns.
1. APX NDD does not support high QImode registers like ah, bh, ch, dh, so for
some optimization splitters that generates highpart zero_extract for QImode
need to be prohibited under NDD
For TImode shifts, they are splitted by splitter functions, which assume
operands[0] and operands[1] to be the same. For the NDD alternative the
assumption may not be true so add split functions for NDD to emit the NDD
form instructions, and omit the handling of !64bit target split.
Although the
From: Kong Lingling
Similar to AND insn, two splitters need to be adjusted to prevent
misoptimizaiton for NDD OR/XOR.
Also adjust *one_cmplsi2_2_zext and its corresponding splitter that will
generate xor insn.
gcc/ChangeLog:
* config/i386/i386.md (3): Add new alternative for NDD
From: Kong Lingling
APX NDD provides an extra destination register operand for several gpr
related legacy insns, so a new alternative can be adopted to operand1
with "r" constraint.
This first patch supports NDD for add instruction, and keeps to use lea
when all operands are registers since lea
From: Kong Lingling
Similar to *add3_doubleword, operands[1] may not equal to operands[0] so
extra move is required.
gcc/ChangeLog:
* config/i386/i386.md (*sub3_doubleword): Add new alternative for
NDD, and emit move when operands[0] not equal to operands[1].
From: Kong Lingling
Legacy adc patterns are commonly adopted to TImode add, when extending TImode
add to NDD version, operands[0] and operands[1] can be different, so extra move
should be emitted if those patterns have optimization when adding const0_rtx.
NDD instructions will automatically
gcc/ChangeLog:
* config/i386/i386.md (*3_1): Extend with a new
alternative to support NDD for SI/DI rotate, and adjust output
template.
(*si3_1_zext): Likewise.
(*3_1): Likewise for QI/HI modes.
(rcrsi2): Likewise, and use nonimmediate_operand for
For left shift, there is an optimization TARGET_DOUBLE_WITH_ADD that shl
1 can be optimized to add. As NDD form of add requires src operand to
be register since NDD cannot take 2 memory src, we currently just keep
using NDD form shift instead of add.
The optimization TARGET_SHIFT1 will try to
For shld/shrd insns, the old pattern use match_dup 0 as its shift src and use
+r*m as its constraint. To support NDD we added new define_insns to handle NDD
form pattern with extra input and dest operand to be fixed in register.
gcc/ChangeLog:
* config/i386/i386.md (x86_64_shld_ndd): New
Similar to LSHIFT, rshift do not need to omit $1 for NDD form.
gcc/ChangeLog:
* config/i386/i386.md (ashr3_cvt): Extend with new
alternatives to support NDD, and adjust output templates.
(*ashr3_1): Likewise for SI/DI mode.
(*lshr3_1): Likewise.
From: Kong Lingling
gcc/ChangeLog:
* config/i386/i386-expand.cc (ix86_fixup_binary_operands_no_copy):
Add use_ndd parameter and parse it.
* config/i386/i386-protos.h (ix86_fixup_binary_operands_no_copy):
Change define.
* config/i386/i386.md (sub3): Add
gcc/ChangeLog:
* config/i386/i386.md (*movcc_noc): Extend with new constraints
to support NDD.
(*movsicc_noc_zext): Likewise.
(*movsicc_noc_zext_1): Likewise.
(*movqicc_noc): Likewise.
gcc/testsuite/ChangeLog:
* gcc.target/i386/apx-ndd-cmov.c: New
NDD uses evex prefix, so when segment prefix is also applied, the instruction
could excceed its 15byte limit, especially adding immediates. This could happen
when "e" constraint accepts any UNSPEC_TPOFF/UNSPEC_NTPOFF constant and it will
add the offset to segment register, which will be encoded
From: Kong Lingling
gcc/ChangeLog:
* config/i386/i386.md: (addsi_1_zext): Add new alternatives for
NDD and adjust output templates.
(*add_2): Likewise.
(*addsi_2_zext): Likewise.
(*add_3): Likewise.
(*addsi_3_zext): Likewise.
(*adddi_4):
Under APX NDD, previous TImode allocation will have issue that it was
originally allocated using continuous pair, like rax:rdi, rdi:rdx.
This will cause issue for all TImode NDD patterns. For NDD we will not
assume the arithmetic operations like add have dependency between dest
and src1, then
From: Kong Lingling
gcc/ChangeLog:
* config/i386/i386-expand.cc (ix86_expand_unary_operator): Add use_ndd
parameter and adjust for NDD.
* config/i386/i386-protos.h: Add use_ndd parameter for
ix86_unary_operator_ok and ix86_expand_unary_operator.
*
-gnu{-m32,} and sde.
OK for trunk?
Hongyu Wang (8):
[APX NDD] Restrict TImode register usage when NDD enabled
[APX NDD] Disable seg_prefixed memory usage for NDD add
[APX NDD] Support APX NDD for left shift insns
[APX NDD] Support APX NDD for right shift insns
[APX NDD] Support APX NDD
From: Kong Lingling
For *one_cmplsi2_2_zext, it will be splitted to xor, so its NDD form will be
added together with xor NDD support.
gcc/ChangeLog:
* config/i386/i386.md (one_cmpl2): Add new constraints for NDD
and adjust output template.
(*one_cmpl2_1): Likewise.
Hi,
On linux x86-64, -fomit-frame-pointer was by default enabled so the
push2pop2 tests cfi scans are based on it. On other target with
-fno-omit-frame-pointer the cfi scan will be wrong as the frame pointer
is pushed at first. Add -fomit-frame-pointer to these tests that related
to cfi scan.
OK
Hi,
The push2/pop2 operand order does not match the binutils implementation
for AT syntax that it will first push operands[2] then operands[1].
Correct it by reverse operand order for AT syntax.
Bootstrapped/regtested on x86-64-linux-pc-gnu{-m32,}
Ok for master?
gcc/ChangeLog:
*
then,
thanks for the suggestion.
Updated patch with just 1 new UNSPEC and removed cfa handling.
Hongtao Liu 于2023年11月20日周一 14:46写道:
>
> On Fri, Nov 17, 2023 at 3:26 PM Hongyu Wang wrote:
> >
> > Intel APX PPX feature has been released in [1].
> >
> > PPX stands for Push-Pop Accelera
Intel APX PPX feature has been released in [1].
PPX stands for Push-Pop Acceleration. PUSH/PUSH2 and its corresponding POP
can be marked with a 1-bit hint to indicate that the POP reads the
value written by the PUSH from the stack. The processor tracks these marked
instructions internally and
Similar to LSHIFT, rshift should also emit $1 for NDD form with CX_REG as
operands[1].
gcc/ChangeLog:
* config/i386/i386-expand.cc (ix86_can_use_ndd_p): Add LSHIFTRT
and RSHIFTRT.
* config/i386/i386.md (ashr3_cvt): Extend with new
alternatives to support NDD, and
For shld/shrd insns, the old pattern use match_dup 0 as its shift src and use
+r*m as its constraint. To support NDD we added new define_insns to handle NDD
form pattern with extra input and dest operand to be fixed in register.
gcc/ChangeLog:
* config/i386/i386.md (x86_64_shld_ndd): New
From: Kong Lingling
Similar to AND insn, two splitters need to be adjusted to prevent
misoptimizaiton for NDD OR/XOR.
gcc/ChangeLog:
* config/i386/i386-expand.cc (ix86_can_use_ndd_p): Add IOR/XOR
support.
* config/i386/i386.md (3): Add NDD alternative and adjust
From: Kong Lingling
For NDD form AND insn, there are three splitter fixes after extending legacy
patterns.
1. APX NDD does not support high QImode registers like ah, bh, ch, dh, so for
some optimization splitters that generates highpart zero_extract for QImode
need to be prohibited under NDD
gcc/ChangeLog:
* config/i386/i386.md (*movcc_noc): Extend with new constraints
to support NDD.
(*movsicc_noc_zext): Likewise.
(*movsicc_noc_zext_1): Likewise.
(*movqicc_noc): Likewise.
gcc/testsuite/ChangeLog:
* gcc.target/i386/apx-ndd-cmov.c: New
From: Kong Lingling
Similar to *add3_doubleword, operands[1] may not equal to operands[0] so
extra move is required.
gcc/ChangeLog:
* config/i386/i386.md (*sub3_doubleword): Add ndd constraints, and
emit move when operands[0] not equal to operands[1].
From: Kong Lingling
Legacy adc patterns are commonly adopted to TImode add, when extending TImode
add to NDD version, operands[0] and operands[1] can be different, so extra move
should be emitted if those patterns have optimization when adding const0_rtx.
gcc/ChangeLog:
*
For left shift, there is an optimization TARGET_DOUBLE_WITH_ADD that shl
1 can be optimized to add. As NDD form of add requires src operand to
be register since NDD cannot take 2 memory src, we currently just keep
using NDD form shift instead of add.
The optimization TARGET_SHIFT1 will try to
From: Kong Lingling
gcc/ChangeLog:
* config/i386/i386-expand.cc (ix86_can_use_ndd_p): Add NOT
support.
* config/i386/i386.md (one_cmpl2): Add NDD constraints, adjust
output template.
(*one_cmpl2_1): Likewise.
(*one_cmplqi2_1): Likewise.
1 - 100 of 303 matches
Mail list logo