https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102227
Bug ID: 102227
Summary: [12 Regression] Likely wrong code since
r12-3376-g13beaf9e8d2d8264c0ad8f6504793fdcf26f3f73
Product: gcc
Version: 12.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: marxin at gcc dot gnu.org
CC: hjl.tools at gmail dot com
Blocks: 26163
Target Milestone: ---
Since the revision, 434.zeusmp and other SPEC benchmarks do fail with
-march=native -Ofast on AMD znver1 and Intel skylake.
I isolated that the affected source file is tranx3.f, where I can see a couple
of assembly changes like:
Before:
vmulsd gridcomr_+44688(,%rax,8), %xmm2, %xmm2
vaddsd %xmm2, %xmm0, %xmm1
vmulsd %xmm2, %xmm0, %xmm0
vxorpd %xmm2, %xmm2, %xmm2
vmaxsd %xmm2, %xmm0, %xmm0
vmovsd %xmm1, %xmm1, %xmm2
vandpd .LC1(%rip), %xmm2, %xmm2
vxorpd %xmm0, %xmm2, %xmm2
vandpd .LC2(%rip), %xmm1, %xmm0
vmovsd .LC3(%rip), %xmm1
vmaxsd %xmm1, %xmm0, %xmm0
vdivsd %xmm0, %xmm2, %xmm0
vmovsd %xmm0, 856(%rsp,%rbx,8)
.L102:
After:
vmulsd gridcomr_+44688(,%rbx,8), %xmm0, %xmm0
vsubsd %xmm2, %xmm1, %xmm2
vmulsd gridcomr_+44688(,%rax,8), %xmm2, %xmm2
vaddsd %xmm2, %xmm0, %xmm1
vandpd .LC1(%rip), %xmm1, %xmm1
vmulsd %xmm2, %xmm0, %xmm0
vxorpd %xmm2, %xmm2, %xmm2
vmaxsd %xmm2, %xmm0, %xmm0
vmovsd .LC3(%rip), %xmm2
vxorpd %xmm0, %xmm1, %xmm0
vandpd .LC2(%rip), %xmm1, %xmm1
vmaxsd %xmm2, %xmm1, %xmm1
vdivsd %xmm1, %xmm0, %xmm0
vmovsd %xmm0, 856(%rsp,%rbx,8)
.L102:
@H.J. Can you please take a look? It's not easy to reproduce a smaller
test-case ..
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=26163
[Bug 26163] [meta-bug] missed optimization in SPEC (2k17, 2k and 2k6 and 95)
