[Bug tree-optimization/80155] New: [7 regression] Performance regression with code hoisting enabled

[thopre01 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80155

            Bug ID: 80155
           Summary: [7 regression] Performance regression with code
                    hoisting enabled
           Product: gcc
           Version: 7.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: thopre01 at gcc dot gnu.org
                CC: avieira at gcc dot gnu.org, rguenth at gcc dot gnu.org
  Target Milestone: ---
            Target: arm-none-eabi

Created attachment 41022
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=41022&action=edit
Reproducers for Cortex-M0+ and Cortex-M7

Hi,

Performance of a popular embedded benchmark regress in GCC 7 at -O2
optimization level on arm-none-eabi targets (at least on Cortex-M0+ or
Cortex-M7). They key to trigger it is to have -fcode-hoisting enabled. If
disabled, performance is improved.

Attached to this bug report are reduced testcase for Cortex-M0+ and Cortex-M7.
Compile respectively with:

arm-none-eabi-gcc -S -O2 -mcpu=cortex-m0plus
arm-none-eabi-gcc -S -O2 -mcpu=cortex-m7

and compare to the same command line with the extra -fno-code-hoisting option
added to see the amount of register push decrease.