| Summary: | Target implements obsolete vcond{,u,eq} expanders | ||
|---|---|---|---|
| Product: | gcc | Reporter: | Richard Biener <rguenth> |
| Component: | target | Assignee: | Not yet assigned to anyone <unassigned> |
| Status: | NEW --- | ||
| Severity: | normal | CC: | ams, chenglulu, ebotcazou, liuhongt, pinskia, rearnsha |
| Priority: | P3 | Keywords: | internal-improvement |
| Version: | 15.0 | ||
| Target Milestone: | --- | ||
| See Also: | https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109955 | ||
| Host: | Target: | sparc, mips, ia64, gcn, arm, aarch64, | |
| Build: | Known to work: | ||
| Known to fail: | Last reconfirmed: | 2024-06-17 00:00:00 | |
| Bug Depends on: | 115519, 115683, 115490, 115517, 115659, 116041 | ||
| Bug Blocks: | |||
| Attachments: | patch to remove use of the legacy optabs | ||
The following shows them all:
grep 'vcond[<eu]' gcc/config/*/*.md
ia64 and sparc are the only targets that do not implement any vec_cmp{,u,eq}
pattern.
mips, sparc and ia64 do not implement any vcond_mask pattern.
Created attachment 58426 [details]
patch to remove use of the legacy optabs
arm is OK - https://ci.linaro.org/job/tcwg_gcc_check--master-arm-precommit/7755/artifact/artifacts/artifacts.precommit/notify/mail-body.txt loongarch is also reportedly OK aarch64 reports just FAIL: gcc.target/aarch64/if-compare_2.c check-function-bodies bar1 FAIL: gcc.target/aarch64/if-compare_2.c check-function-bodies bar2 . The master branch has been updated by hongtao Liu <liuhongt@gcc.gnu.org>: https://gcc.gnu.org/g:aac00d09859cc5934bd0f7493d537b8430337773 commit r15-1638-gaac00d09859cc5934bd0f7493d537b8430337773 Author: liuhongt <hongtao.liu@intel.com> Date: Thu Jun 20 12:41:13 2024 +0800 Optimize a < 0 ? -1 : 0 to (signed)a >> 31. Try to optimize x < 0 ? -1 : 0 into (signed) x >> 31 and x < 0 ? 1 : 0 into (unsigned) x >> 31. Move the optimization did in ix86_expand_int_vcond to match.pd gcc/ChangeLog: PR target/114189 * match.pd: Simplify a < 0 ? -1 : 0 to (signed) >> 31 and a < 0 ? 1 : 0 to (unsigned) a >> 31 for vector integer type. gcc/testsuite/ChangeLog: * gcc.target/i386/avx2-pr115517.c: New test. * gcc.target/i386/avx512-pr115517.c: New test. * g++.target/i386/avx2-pr115517.C: New test. * g++.target/i386/avx512-pr115517.C: New test. * g++.dg/tree-ssa/pr88152-1.C: Adjust testcase. powerpc has removed the obsolete expanders. There are 2 entries for vcond_mask in the internal manual:
'vcond_mask_MN'
Similar to 'vcondMN' but operand 3 holds a pre-computed result of
vector comparison.
'vcond_mask_MN'
Set each element of operand 0 to the corresponding element of
operand 2 or operand 3. Choose operand 2 if both the element index
is less than operand 4 plus operand 5 and the corresponding element
of operand 1 is nonzero:
for (i = 0; i < GET_MODE_NUNITS (M); i++)
op0[i] = i < op4 + op5 && op1[i] ? op2[i] : op3[i];
Operands 0, 2 and 3 have mode M. Operand 1 has mode N. Operands 4
and 5 have a target-dependent scalar integer mode
On Fri, 12 Jul 2024, ebotcazou at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114189 > > Eric Botcazou <ebotcazou at gcc dot gnu.org> changed: > > What |Removed |Added > ---------------------------------------------------------------------------- > CC| |ebotcazou at gcc dot gnu.org > > --- Comment #8 from Eric Botcazou <ebotcazou at gcc dot gnu.org> --- > There are 2 entries for vcond_mask in the internal manual: > > 'vcond_mask_MN' > Similar to 'vcondMN' but operand 3 holds a pre-computed result of > vector comparison. > > 'vcond_mask_MN' This should be vcond_mask_len_MN I think. > Set each element of operand 0 to the corresponding element of > operand 2 or operand 3. Choose operand 2 if both the element index > is less than operand 4 plus operand 5 and the corresponding element > of operand 1 is nonzero: > > for (i = 0; i < GET_MODE_NUNITS (M); i++) > op0[i] = i < op4 + op5 && op1[i] ? op2[i] : op3[i]; > > Operands 0, 2 and 3 have mode M. Operand 1 has mode N. Operands 4 > and 5 have a target-dependent scalar integer mode The master branch has been updated by Stefan Schulze Frielinghaus <stefansf@gcc.gnu.org>: https://gcc.gnu.org/g:75c0bf997d2808561451e62aa6b7ae7c8e32b9e9 commit r15-2058-g75c0bf997d2808561451e62aa6b7ae7c8e32b9e9 Author: Stefan Schulze Frielinghaus <stefansf@gcc.gnu.org> Date: Tue Jul 16 10:41:52 2024 +0200 s390: Drop vcond{,u} expanders Optabs vcond{,u} will be removed for GCC 15. Since regtest shows no fallout, dropping the expanders, now. gcc/ChangeLog: PR target/114189 * config/s390/vector.md (V_HW2): Remove. (vcond<V_HW:mode><V_HW2:mode>): Remove. (vcondu<V_HW:mode><V_HW2:mode>): Remove. (In reply to Richard Biener from comment #4) > aarch64 reports just > > FAIL: gcc.target/aarch64/if-compare_2.c check-function-bodies bar1 > FAIL: gcc.target/aarch64/if-compare_2.c check-function-bodies bar2 I think these 2 are now fixed after the fixes for PR 115659. There is also missing andc/iorc patterns. Which I am testing a change for them. I am doing it as part of using andc/iorc for scalars change due to be able to use them for vectors too. (In reply to Andrew Pinski from comment #11) > (In reply to Richard Biener from comment #4) > > aarch64 reports just > > > > FAIL: gcc.target/aarch64/if-compare_2.c check-function-bodies bar1 > > FAIL: gcc.target/aarch64/if-compare_2.c check-function-bodies bar2 > > I think these 2 are now fixed after the fixes for PR 115659. > There is also missing andc/iorc patterns. > Which I am testing a change for them. I am doing it as part of using > andc/iorc for scalars change due to be able to use them for vectors too. Actually I take that back, these are no longer vectorized. Let me file that as a seperate issue. The i386 backend is also clear of vcond{,u,eq} patterns, there's some XFAILed fallout when just SSE is available (PR115683).
CCing maintainers for loongarch, arm/aarch64 and gcn. The master branch has been updated by LuluCheng <chenglulu@gcc.gnu.org>: https://gcc.gnu.org/g:0498f8bda1d08b8ed8100d759917792baf2da15f commit r15-2878-g0498f8bda1d08b8ed8100d759917792baf2da15f Author: Lulu Cheng <chenglulu@loongson.cn> Date: Thu Aug 8 10:39:54 2024 +0800 LoongArch: Drop vcond{,u} expanders. Optabs vcond{,u} will be removed for GCC 15. Since regtest shows no fallout, dropping the expanders, now. gcc/ChangeLog: PR target/114189 * config/loongarch/lasx.md (vcondu<LASX:mode><ILASX:mode>): Delete. (vcond<LASX:mode><LASX_2:mode>): Likewise. * config/loongarch/lsx.md (vcondu<LSX:mode><ILSX:mode>): Likewise. (vcond<LSX:mode><LSX_2:mode>): Likewise. |
The listed targets implement any of vcond, vcondu or vcondeq expanders where the modern way of exposing vector conditional instructions is using mask generating vec_cmp{,u,eq} and vcond_mask expanders. vcond, vcondu and vcondeq optabs will be retired in the future.