aboutsummaryrefslogtreecommitdiff
path: root/gcc/tree-ssa-loop-prefetch.cc
diff options
context:
space:
mode:
authorMartin Liska <mliska@suse.cz>2022-01-14 16:56:44 +0100
committerMartin Liska <mliska@suse.cz>2022-01-17 22:12:04 +0100
commit5c69acb32329d49e58c26fa41ae74229a52b9106 (patch)
treeddb05f9d73afb6f998457d2ac4b720e3b3b60483 /gcc/tree-ssa-loop-prefetch.cc
parent490e23032baaece71f2ec09fa1805064b150fbc2 (diff)
downloadgcc-5c69acb32329d49e58c26fa41ae74229a52b9106.zip
gcc-5c69acb32329d49e58c26fa41ae74229a52b9106.tar.gz
gcc-5c69acb32329d49e58c26fa41ae74229a52b9106.tar.bz2
Rename .c files to .cc files.
gcc/ada/ChangeLog: * adadecode.c: Moved to... * adadecode.cc: ...here. * affinity.c: Moved to... * affinity.cc: ...here. * argv-lynxos178-raven-cert.c: Moved to... * argv-lynxos178-raven-cert.cc: ...here. * argv.c: Moved to... * argv.cc: ...here. * aux-io.c: Moved to... * aux-io.cc: ...here. * cio.c: Moved to... * cio.cc: ...here. * cstreams.c: Moved to... * cstreams.cc: ...here. * env.c: Moved to... * env.cc: ...here. * exit.c: Moved to... * exit.cc: ...here. * expect.c: Moved to... * expect.cc: ...here. * final.c: Moved to... * final.cc: ...here. * gcc-interface/cuintp.c: Moved to... * gcc-interface/cuintp.cc: ...here. * gcc-interface/decl.c: Moved to... * gcc-interface/decl.cc: ...here. * gcc-interface/misc.c: Moved to... * gcc-interface/misc.cc: ...here. * gcc-interface/targtyps.c: Moved to... * gcc-interface/targtyps.cc: ...here. * gcc-interface/trans.c: Moved to... * gcc-interface/trans.cc: ...here. * gcc-interface/utils.c: Moved to... * gcc-interface/utils.cc: ...here. * gcc-interface/utils2.c: Moved to... * gcc-interface/utils2.cc: ...here. * init.c: Moved to... * init.cc: ...here. * initialize.c: Moved to... * initialize.cc: ...here. * libgnarl/thread.c: Moved to... * libgnarl/thread.cc: ...here. * link.c: Moved to... * link.cc: ...here. * locales.c: Moved to... * locales.cc: ...here. * mkdir.c: Moved to... * mkdir.cc: ...here. * raise.c: Moved to... * raise.cc: ...here. * rtfinal.c: Moved to... * rtfinal.cc: ...here. * rtinit.c: Moved to... * rtinit.cc: ...here. * seh_init.c: Moved to... * seh_init.cc: ...here. * sigtramp-armdroid.c: Moved to... * sigtramp-armdroid.cc: ...here. * sigtramp-ios.c: Moved to... * sigtramp-ios.cc: ...here. * sigtramp-qnx.c: Moved to... * sigtramp-qnx.cc: ...here. * sigtramp-vxworks.c: Moved to... * sigtramp-vxworks.cc: ...here. * socket.c: Moved to... * socket.cc: ...here. * tracebak.c: Moved to... * tracebak.cc: ...here. * version.c: Moved to... * version.cc: ...here. * vx_stack_info.c: Moved to... * vx_stack_info.cc: ...here. gcc/ChangeLog: * adjust-alignment.c: Moved to... * adjust-alignment.cc: ...here. * alias.c: Moved to... * alias.cc: ...here. * alloc-pool.c: Moved to... * alloc-pool.cc: ...here. * asan.c: Moved to... * asan.cc: ...here. * attribs.c: Moved to... * attribs.cc: ...here. * auto-inc-dec.c: Moved to... * auto-inc-dec.cc: ...here. * auto-profile.c: Moved to... * auto-profile.cc: ...here. * bb-reorder.c: Moved to... * bb-reorder.cc: ...here. * bitmap.c: Moved to... * bitmap.cc: ...here. * btfout.c: Moved to... * btfout.cc: ...here. * builtins.c: Moved to... * builtins.cc: ...here. * caller-save.c: Moved to... * caller-save.cc: ...here. * calls.c: Moved to... * calls.cc: ...here. * ccmp.c: Moved to... * ccmp.cc: ...here. * cfg.c: Moved to... * cfg.cc: ...here. * cfganal.c: Moved to... * cfganal.cc: ...here. * cfgbuild.c: Moved to... * cfgbuild.cc: ...here. * cfgcleanup.c: Moved to... * cfgcleanup.cc: ...here. * cfgexpand.c: Moved to... * cfgexpand.cc: ...here. * cfghooks.c: Moved to... * cfghooks.cc: ...here. * cfgloop.c: Moved to... * cfgloop.cc: ...here. * cfgloopanal.c: Moved to... * cfgloopanal.cc: ...here. * cfgloopmanip.c: Moved to... * cfgloopmanip.cc: ...here. * cfgrtl.c: Moved to... * cfgrtl.cc: ...here. * cgraph.c: Moved to... * cgraph.cc: ...here. * cgraphbuild.c: Moved to... * cgraphbuild.cc: ...here. * cgraphclones.c: Moved to... * cgraphclones.cc: ...here. * cgraphunit.c: Moved to... * cgraphunit.cc: ...here. * collect-utils.c: Moved to... * collect-utils.cc: ...here. * collect2-aix.c: Moved to... * collect2-aix.cc: ...here. * collect2.c: Moved to... * collect2.cc: ...here. * combine-stack-adj.c: Moved to... * combine-stack-adj.cc: ...here. * combine.c: Moved to... * combine.cc: ...here. * common/common-targhooks.c: Moved to... * common/common-targhooks.cc: ...here. * common/config/aarch64/aarch64-common.c: Moved to... * common/config/aarch64/aarch64-common.cc: ...here. * common/config/alpha/alpha-common.c: Moved to... * common/config/alpha/alpha-common.cc: ...here. * common/config/arc/arc-common.c: Moved to... * common/config/arc/arc-common.cc: ...here. * common/config/arm/arm-common.c: Moved to... * common/config/arm/arm-common.cc: ...here. * common/config/avr/avr-common.c: Moved to... * common/config/avr/avr-common.cc: ...here. * common/config/bfin/bfin-common.c: Moved to... * common/config/bfin/bfin-common.cc: ...here. * common/config/bpf/bpf-common.c: Moved to... * common/config/bpf/bpf-common.cc: ...here. * common/config/c6x/c6x-common.c: Moved to... * common/config/c6x/c6x-common.cc: ...here. * common/config/cr16/cr16-common.c: Moved to... * common/config/cr16/cr16-common.cc: ...here. * common/config/cris/cris-common.c: Moved to... * common/config/cris/cris-common.cc: ...here. * common/config/csky/csky-common.c: Moved to... * common/config/csky/csky-common.cc: ...here. * common/config/default-common.c: Moved to... * common/config/default-common.cc: ...here. * common/config/epiphany/epiphany-common.c: Moved to... * common/config/epiphany/epiphany-common.cc: ...here. * common/config/fr30/fr30-common.c: Moved to... * common/config/fr30/fr30-common.cc: ...here. * common/config/frv/frv-common.c: Moved to... * common/config/frv/frv-common.cc: ...here. * common/config/gcn/gcn-common.c: Moved to... * common/config/gcn/gcn-common.cc: ...here. * common/config/h8300/h8300-common.c: Moved to... * common/config/h8300/h8300-common.cc: ...here. * common/config/i386/i386-common.c: Moved to... * common/config/i386/i386-common.cc: ...here. * common/config/ia64/ia64-common.c: Moved to... * common/config/ia64/ia64-common.cc: ...here. * common/config/iq2000/iq2000-common.c: Moved to... * common/config/iq2000/iq2000-common.cc: ...here. * common/config/lm32/lm32-common.c: Moved to... * common/config/lm32/lm32-common.cc: ...here. * common/config/m32r/m32r-common.c: Moved to... * common/config/m32r/m32r-common.cc: ...here. * common/config/m68k/m68k-common.c: Moved to... * common/config/m68k/m68k-common.cc: ...here. * common/config/mcore/mcore-common.c: Moved to... * common/config/mcore/mcore-common.cc: ...here. * common/config/microblaze/microblaze-common.c: Moved to... * common/config/microblaze/microblaze-common.cc: ...here. * common/config/mips/mips-common.c: Moved to... * common/config/mips/mips-common.cc: ...here. * common/config/mmix/mmix-common.c: Moved to... * common/config/mmix/mmix-common.cc: ...here. * common/config/mn10300/mn10300-common.c: Moved to... * common/config/mn10300/mn10300-common.cc: ...here. * common/config/msp430/msp430-common.c: Moved to... * common/config/msp430/msp430-common.cc: ...here. * common/config/nds32/nds32-common.c: Moved to... * common/config/nds32/nds32-common.cc: ...here. * common/config/nios2/nios2-common.c: Moved to... * common/config/nios2/nios2-common.cc: ...here. * common/config/nvptx/nvptx-common.c: Moved to... * common/config/nvptx/nvptx-common.cc: ...here. * common/config/or1k/or1k-common.c: Moved to... * common/config/or1k/or1k-common.cc: ...here. * common/config/pa/pa-common.c: Moved to... * common/config/pa/pa-common.cc: ...here. * common/config/pdp11/pdp11-common.c: Moved to... * common/config/pdp11/pdp11-common.cc: ...here. * common/config/pru/pru-common.c: Moved to... * common/config/pru/pru-common.cc: ...here. * common/config/riscv/riscv-common.c: Moved to... * common/config/riscv/riscv-common.cc: ...here. * common/config/rs6000/rs6000-common.c: Moved to... * common/config/rs6000/rs6000-common.cc: ...here. * common/config/rx/rx-common.c: Moved to... * common/config/rx/rx-common.cc: ...here. * common/config/s390/s390-common.c: Moved to... * common/config/s390/s390-common.cc: ...here. * common/config/sh/sh-common.c: Moved to... * common/config/sh/sh-common.cc: ...here. * common/config/sparc/sparc-common.c: Moved to... * common/config/sparc/sparc-common.cc: ...here. * common/config/tilegx/tilegx-common.c: Moved to... * common/config/tilegx/tilegx-common.cc: ...here. * common/config/tilepro/tilepro-common.c: Moved to... * common/config/tilepro/tilepro-common.cc: ...here. * common/config/v850/v850-common.c: Moved to... * common/config/v850/v850-common.cc: ...here. * common/config/vax/vax-common.c: Moved to... * common/config/vax/vax-common.cc: ...here. * common/config/visium/visium-common.c: Moved to... * common/config/visium/visium-common.cc: ...here. * common/config/xstormy16/xstormy16-common.c: Moved to... * common/config/xstormy16/xstormy16-common.cc: ...here. * common/config/xtensa/xtensa-common.c: Moved to... * common/config/xtensa/xtensa-common.cc: ...here. * compare-elim.c: Moved to... * compare-elim.cc: ...here. * config/aarch64/aarch64-bti-insert.c: Moved to... * config/aarch64/aarch64-bti-insert.cc: ...here. * config/aarch64/aarch64-builtins.c: Moved to... * config/aarch64/aarch64-builtins.cc: ...here. * config/aarch64/aarch64-c.c: Moved to... * config/aarch64/aarch64-c.cc: ...here. * config/aarch64/aarch64-d.c: Moved to... * config/aarch64/aarch64-d.cc: ...here. * config/aarch64/aarch64.c: Moved to... * config/aarch64/aarch64.cc: ...here. * config/aarch64/cortex-a57-fma-steering.c: Moved to... * config/aarch64/cortex-a57-fma-steering.cc: ...here. * config/aarch64/driver-aarch64.c: Moved to... * config/aarch64/driver-aarch64.cc: ...here. * config/aarch64/falkor-tag-collision-avoidance.c: Moved to... * config/aarch64/falkor-tag-collision-avoidance.cc: ...here. * config/aarch64/host-aarch64-darwin.c: Moved to... * config/aarch64/host-aarch64-darwin.cc: ...here. * config/alpha/alpha.c: Moved to... * config/alpha/alpha.cc: ...here. * config/alpha/driver-alpha.c: Moved to... * config/alpha/driver-alpha.cc: ...here. * config/arc/arc-c.c: Moved to... * config/arc/arc-c.cc: ...here. * config/arc/arc.c: Moved to... * config/arc/arc.cc: ...here. * config/arc/driver-arc.c: Moved to... * config/arc/driver-arc.cc: ...here. * config/arm/aarch-common.c: Moved to... * config/arm/aarch-common.cc: ...here. * config/arm/arm-builtins.c: Moved to... * config/arm/arm-builtins.cc: ...here. * config/arm/arm-c.c: Moved to... * config/arm/arm-c.cc: ...here. * config/arm/arm-d.c: Moved to... * config/arm/arm-d.cc: ...here. * config/arm/arm.c: Moved to... * config/arm/arm.cc: ...here. * config/arm/driver-arm.c: Moved to... * config/arm/driver-arm.cc: ...here. * config/avr/avr-c.c: Moved to... * config/avr/avr-c.cc: ...here. * config/avr/avr-devices.c: Moved to... * config/avr/avr-devices.cc: ...here. * config/avr/avr-log.c: Moved to... * config/avr/avr-log.cc: ...here. * config/avr/avr.c: Moved to... * config/avr/avr.cc: ...here. * config/avr/driver-avr.c: Moved to... * config/avr/driver-avr.cc: ...here. * config/avr/gen-avr-mmcu-specs.c: Moved to... * config/avr/gen-avr-mmcu-specs.cc: ...here. * config/avr/gen-avr-mmcu-texi.c: Moved to... * config/avr/gen-avr-mmcu-texi.cc: ...here. * config/bfin/bfin.c: Moved to... * config/bfin/bfin.cc: ...here. * config/bpf/bpf.c: Moved to... * config/bpf/bpf.cc: ...here. * config/bpf/coreout.c: Moved to... * config/bpf/coreout.cc: ...here. * config/c6x/c6x.c: Moved to... * config/c6x/c6x.cc: ...here. * config/cr16/cr16.c: Moved to... * config/cr16/cr16.cc: ...here. * config/cris/cris.c: Moved to... * config/cris/cris.cc: ...here. * config/csky/csky.c: Moved to... * config/csky/csky.cc: ...here. * config/darwin-c.c: Moved to... * config/darwin-c.cc: ...here. * config/darwin-d.c: Moved to... * config/darwin-d.cc: ...here. * config/darwin-driver.c: Moved to... * config/darwin-driver.cc: ...here. * config/darwin-f.c: Moved to... * config/darwin-f.cc: ...here. * config/darwin.c: Moved to... * config/darwin.cc: ...here. * config/default-c.c: Moved to... * config/default-c.cc: ...here. * config/default-d.c: Moved to... * config/default-d.cc: ...here. * config/dragonfly-d.c: Moved to... * config/dragonfly-d.cc: ...here. * config/epiphany/epiphany.c: Moved to... * config/epiphany/epiphany.cc: ...here. * config/epiphany/mode-switch-use.c: Moved to... * config/epiphany/mode-switch-use.cc: ...here. * config/epiphany/resolve-sw-modes.c: Moved to... * config/epiphany/resolve-sw-modes.cc: ...here. * config/fr30/fr30.c: Moved to... * config/fr30/fr30.cc: ...here. * config/freebsd-d.c: Moved to... * config/freebsd-d.cc: ...here. * config/frv/frv.c: Moved to... * config/frv/frv.cc: ...here. * config/ft32/ft32.c: Moved to... * config/ft32/ft32.cc: ...here. * config/gcn/driver-gcn.c: Moved to... * config/gcn/driver-gcn.cc: ...here. * config/gcn/gcn-run.c: Moved to... * config/gcn/gcn-run.cc: ...here. * config/gcn/gcn-tree.c: Moved to... * config/gcn/gcn-tree.cc: ...here. * config/gcn/gcn.c: Moved to... * config/gcn/gcn.cc: ...here. * config/gcn/mkoffload.c: Moved to... * config/gcn/mkoffload.cc: ...here. * config/glibc-c.c: Moved to... * config/glibc-c.cc: ...here. * config/glibc-d.c: Moved to... * config/glibc-d.cc: ...here. * config/h8300/h8300.c: Moved to... * config/h8300/h8300.cc: ...here. * config/host-darwin.c: Moved to... * config/host-darwin.cc: ...here. * config/host-hpux.c: Moved to... * config/host-hpux.cc: ...here. * config/host-linux.c: Moved to... * config/host-linux.cc: ...here. * config/host-netbsd.c: Moved to... * config/host-netbsd.cc: ...here. * config/host-openbsd.c: Moved to... * config/host-openbsd.cc: ...here. * config/host-solaris.c: Moved to... * config/host-solaris.cc: ...here. * config/i386/djgpp.c: Moved to... * config/i386/djgpp.cc: ...here. * config/i386/driver-i386.c: Moved to... * config/i386/driver-i386.cc: ...here. * config/i386/driver-mingw32.c: Moved to... * config/i386/driver-mingw32.cc: ...here. * config/i386/gnu-property.c: Moved to... * config/i386/gnu-property.cc: ...here. * config/i386/host-cygwin.c: Moved to... * config/i386/host-cygwin.cc: ...here. * config/i386/host-i386-darwin.c: Moved to... * config/i386/host-i386-darwin.cc: ...here. * config/i386/host-mingw32.c: Moved to... * config/i386/host-mingw32.cc: ...here. * config/i386/i386-builtins.c: Moved to... * config/i386/i386-builtins.cc: ...here. * config/i386/i386-c.c: Moved to... * config/i386/i386-c.cc: ...here. * config/i386/i386-d.c: Moved to... * config/i386/i386-d.cc: ...here. * config/i386/i386-expand.c: Moved to... * config/i386/i386-expand.cc: ...here. * config/i386/i386-features.c: Moved to... * config/i386/i386-features.cc: ...here. * config/i386/i386-options.c: Moved to... * config/i386/i386-options.cc: ...here. * config/i386/i386.c: Moved to... * config/i386/i386.cc: ...here. * config/i386/intelmic-mkoffload.c: Moved to... * config/i386/intelmic-mkoffload.cc: ...here. * config/i386/msformat-c.c: Moved to... * config/i386/msformat-c.cc: ...here. * config/i386/winnt-cxx.c: Moved to... * config/i386/winnt-cxx.cc: ...here. * config/i386/winnt-d.c: Moved to... * config/i386/winnt-d.cc: ...here. * config/i386/winnt-stubs.c: Moved to... * config/i386/winnt-stubs.cc: ...here. * config/i386/winnt.c: Moved to... * config/i386/winnt.cc: ...here. * config/i386/x86-tune-sched-atom.c: Moved to... * config/i386/x86-tune-sched-atom.cc: ...here. * config/i386/x86-tune-sched-bd.c: Moved to... * config/i386/x86-tune-sched-bd.cc: ...here. * config/i386/x86-tune-sched-core.c: Moved to... * config/i386/x86-tune-sched-core.cc: ...here. * config/i386/x86-tune-sched.c: Moved to... * config/i386/x86-tune-sched.cc: ...here. * config/ia64/ia64-c.c: Moved to... * config/ia64/ia64-c.cc: ...here. * config/ia64/ia64.c: Moved to... * config/ia64/ia64.cc: ...here. * config/iq2000/iq2000.c: Moved to... * config/iq2000/iq2000.cc: ...here. * config/linux.c: Moved to... * config/linux.cc: ...here. * config/lm32/lm32.c: Moved to... * config/lm32/lm32.cc: ...here. * config/m32c/m32c-pragma.c: Moved to... * config/m32c/m32c-pragma.cc: ...here. * config/m32c/m32c.c: Moved to... * config/m32c/m32c.cc: ...here. * config/m32r/m32r.c: Moved to... * config/m32r/m32r.cc: ...here. * config/m68k/m68k.c: Moved to... * config/m68k/m68k.cc: ...here. * config/mcore/mcore.c: Moved to... * config/mcore/mcore.cc: ...here. * config/microblaze/microblaze-c.c: Moved to... * config/microblaze/microblaze-c.cc: ...here. * config/microblaze/microblaze.c: Moved to... * config/microblaze/microblaze.cc: ...here. * config/mips/driver-native.c: Moved to... * config/mips/driver-native.cc: ...here. * config/mips/frame-header-opt.c: Moved to... * config/mips/frame-header-opt.cc: ...here. * config/mips/mips-d.c: Moved to... * config/mips/mips-d.cc: ...here. * config/mips/mips.c: Moved to... * config/mips/mips.cc: ...here. * config/mmix/mmix.c: Moved to... * config/mmix/mmix.cc: ...here. * config/mn10300/mn10300.c: Moved to... * config/mn10300/mn10300.cc: ...here. * config/moxie/moxie.c: Moved to... * config/moxie/moxie.cc: ...here. * config/msp430/driver-msp430.c: Moved to... * config/msp430/driver-msp430.cc: ...here. * config/msp430/msp430-c.c: Moved to... * config/msp430/msp430-c.cc: ...here. * config/msp430/msp430-devices.c: Moved to... * config/msp430/msp430-devices.cc: ...here. * config/msp430/msp430.c: Moved to... * config/msp430/msp430.cc: ...here. * config/nds32/nds32-cost.c: Moved to... * config/nds32/nds32-cost.cc: ...here. * config/nds32/nds32-fp-as-gp.c: Moved to... * config/nds32/nds32-fp-as-gp.cc: ...here. * config/nds32/nds32-intrinsic.c: Moved to... * config/nds32/nds32-intrinsic.cc: ...here. * config/nds32/nds32-isr.c: Moved to... * config/nds32/nds32-isr.cc: ...here. * config/nds32/nds32-md-auxiliary.c: Moved to... * config/nds32/nds32-md-auxiliary.cc: ...here. * config/nds32/nds32-memory-manipulation.c: Moved to... * config/nds32/nds32-memory-manipulation.cc: ...here. * config/nds32/nds32-pipelines-auxiliary.c: Moved to... * config/nds32/nds32-pipelines-auxiliary.cc: ...here. * config/nds32/nds32-predicates.c: Moved to... * config/nds32/nds32-predicates.cc: ...here. * config/nds32/nds32-relax-opt.c: Moved to... * config/nds32/nds32-relax-opt.cc: ...here. * config/nds32/nds32-utils.c: Moved to... * config/nds32/nds32-utils.cc: ...here. * config/nds32/nds32.c: Moved to... * config/nds32/nds32.cc: ...here. * config/netbsd-d.c: Moved to... * config/netbsd-d.cc: ...here. * config/netbsd.c: Moved to... * config/netbsd.cc: ...here. * config/nios2/nios2.c: Moved to... * config/nios2/nios2.cc: ...here. * config/nvptx/mkoffload.c: Moved to... * config/nvptx/mkoffload.cc: ...here. * config/nvptx/nvptx-c.c: Moved to... * config/nvptx/nvptx-c.cc: ...here. * config/nvptx/nvptx.c: Moved to... * config/nvptx/nvptx.cc: ...here. * config/openbsd-d.c: Moved to... * config/openbsd-d.cc: ...here. * config/or1k/or1k.c: Moved to... * config/or1k/or1k.cc: ...here. * config/pa/pa-d.c: Moved to... * config/pa/pa-d.cc: ...here. * config/pa/pa.c: Moved to... * config/pa/pa.cc: ...here. * config/pdp11/pdp11.c: Moved to... * config/pdp11/pdp11.cc: ...here. * config/pru/pru-passes.c: Moved to... * config/pru/pru-passes.cc: ...here. * config/pru/pru-pragma.c: Moved to... * config/pru/pru-pragma.cc: ...here. * config/pru/pru.c: Moved to... * config/pru/pru.cc: ...here. * config/riscv/riscv-builtins.c: Moved to... * config/riscv/riscv-builtins.cc: ...here. * config/riscv/riscv-c.c: Moved to... * config/riscv/riscv-c.cc: ...here. * config/riscv/riscv-d.c: Moved to... * config/riscv/riscv-d.cc: ...here. * config/riscv/riscv-shorten-memrefs.c: Moved to... * config/riscv/riscv-shorten-memrefs.cc: ...here. * config/riscv/riscv-sr.c: Moved to... * config/riscv/riscv-sr.cc: ...here. * config/riscv/riscv.c: Moved to... * config/riscv/riscv.cc: ...here. * config/rl78/rl78-c.c: Moved to... * config/rl78/rl78-c.cc: ...here. * config/rl78/rl78.c: Moved to... * config/rl78/rl78.cc: ...here. * config/rs6000/driver-rs6000.c: Moved to... * config/rs6000/driver-rs6000.cc: ...here. * config/rs6000/host-darwin.c: Moved to... * config/rs6000/host-darwin.cc: ...here. * config/rs6000/host-ppc64-darwin.c: Moved to... * config/rs6000/host-ppc64-darwin.cc: ...here. * config/rs6000/rbtree.c: Moved to... * config/rs6000/rbtree.cc: ...here. * config/rs6000/rs6000-c.c: Moved to... * config/rs6000/rs6000-c.cc: ...here. * config/rs6000/rs6000-call.c: Moved to... * config/rs6000/rs6000-call.cc: ...here. * config/rs6000/rs6000-d.c: Moved to... * config/rs6000/rs6000-d.cc: ...here. * config/rs6000/rs6000-gen-builtins.c: Moved to... * config/rs6000/rs6000-gen-builtins.cc: ...here. * config/rs6000/rs6000-linux.c: Moved to... * config/rs6000/rs6000-linux.cc: ...here. * config/rs6000/rs6000-logue.c: Moved to... * config/rs6000/rs6000-logue.cc: ...here. * config/rs6000/rs6000-p8swap.c: Moved to... * config/rs6000/rs6000-p8swap.cc: ...here. * config/rs6000/rs6000-pcrel-opt.c: Moved to... * config/rs6000/rs6000-pcrel-opt.cc: ...here. * config/rs6000/rs6000-string.c: Moved to... * config/rs6000/rs6000-string.cc: ...here. * config/rs6000/rs6000.c: Moved to... * config/rs6000/rs6000.cc: ...here. * config/rx/rx.c: Moved to... * config/rx/rx.cc: ...here. * config/s390/driver-native.c: Moved to... * config/s390/driver-native.cc: ...here. * config/s390/s390-c.c: Moved to... * config/s390/s390-c.cc: ...here. * config/s390/s390-d.c: Moved to... * config/s390/s390-d.cc: ...here. * config/s390/s390.c: Moved to... * config/s390/s390.cc: ...here. * config/sh/divtab-sh4-300.c: Moved to... * config/sh/divtab-sh4-300.cc: ...here. * config/sh/divtab-sh4.c: Moved to... * config/sh/divtab-sh4.cc: ...here. * config/sh/divtab.c: Moved to... * config/sh/divtab.cc: ...here. * config/sh/sh-c.c: Moved to... * config/sh/sh-c.cc: ...here. * config/sh/sh.c: Moved to... * config/sh/sh.cc: ...here. * config/sol2-c.c: Moved to... * config/sol2-c.cc: ...here. * config/sol2-cxx.c: Moved to... * config/sol2-cxx.cc: ...here. * config/sol2-d.c: Moved to... * config/sol2-d.cc: ...here. * config/sol2-stubs.c: Moved to... * config/sol2-stubs.cc: ...here. * config/sol2.c: Moved to... * config/sol2.cc: ...here. * config/sparc/driver-sparc.c: Moved to... * config/sparc/driver-sparc.cc: ...here. * config/sparc/sparc-c.c: Moved to... * config/sparc/sparc-c.cc: ...here. * config/sparc/sparc-d.c: Moved to... * config/sparc/sparc-d.cc: ...here. * config/sparc/sparc.c: Moved to... * config/sparc/sparc.cc: ...here. * config/stormy16/stormy16.c: Moved to... * config/stormy16/stormy16.cc: ...here. * config/tilegx/mul-tables.c: Moved to... * config/tilegx/mul-tables.cc: ...here. * config/tilegx/tilegx-c.c: Moved to... * config/tilegx/tilegx-c.cc: ...here. * config/tilegx/tilegx.c: Moved to... * config/tilegx/tilegx.cc: ...here. * config/tilepro/mul-tables.c: Moved to... * config/tilepro/mul-tables.cc: ...here. * config/tilepro/tilepro-c.c: Moved to... * config/tilepro/tilepro-c.cc: ...here. * config/tilepro/tilepro.c: Moved to... * config/tilepro/tilepro.cc: ...here. * config/v850/v850-c.c: Moved to... * config/v850/v850-c.cc: ...here. * config/v850/v850.c: Moved to... * config/v850/v850.cc: ...here. * config/vax/vax.c: Moved to... * config/vax/vax.cc: ...here. * config/visium/visium.c: Moved to... * config/visium/visium.cc: ...here. * config/vms/vms-c.c: Moved to... * config/vms/vms-c.cc: ...here. * config/vms/vms-f.c: Moved to... * config/vms/vms-f.cc: ...here. * config/vms/vms.c: Moved to... * config/vms/vms.cc: ...here. * config/vxworks-c.c: Moved to... * config/vxworks-c.cc: ...here. * config/vxworks.c: Moved to... * config/vxworks.cc: ...here. * config/winnt-c.c: Moved to... * config/winnt-c.cc: ...here. * config/xtensa/xtensa.c: Moved to... * config/xtensa/xtensa.cc: ...here. * context.c: Moved to... * context.cc: ...here. * convert.c: Moved to... * convert.cc: ...here. * coverage.c: Moved to... * coverage.cc: ...here. * cppbuiltin.c: Moved to... * cppbuiltin.cc: ...here. * cppdefault.c: Moved to... * cppdefault.cc: ...here. * cprop.c: Moved to... * cprop.cc: ...here. * cse.c: Moved to... * cse.cc: ...here. * cselib.c: Moved to... * cselib.cc: ...here. * ctfc.c: Moved to... * ctfc.cc: ...here. * ctfout.c: Moved to... * ctfout.cc: ...here. * data-streamer-in.c: Moved to... * data-streamer-in.cc: ...here. * data-streamer-out.c: Moved to... * data-streamer-out.cc: ...here. * data-streamer.c: Moved to... * data-streamer.cc: ...here. * dbgcnt.c: Moved to... * dbgcnt.cc: ...here. * dbxout.c: Moved to... * dbxout.cc: ...here. * dce.c: Moved to... * dce.cc: ...here. * ddg.c: Moved to... * ddg.cc: ...here. * debug.c: Moved to... * debug.cc: ...here. * df-core.c: Moved to... * df-core.cc: ...here. * df-problems.c: Moved to... * df-problems.cc: ...here. * df-scan.c: Moved to... * df-scan.cc: ...here. * dfp.c: Moved to... * dfp.cc: ...here. * diagnostic-color.c: Moved to... * diagnostic-color.cc: ...here. * diagnostic-show-locus.c: Moved to... * diagnostic-show-locus.cc: ...here. * diagnostic-spec.c: Moved to... * diagnostic-spec.cc: ...here. * diagnostic.c: Moved to... * diagnostic.cc: ...here. * dojump.c: Moved to... * dojump.cc: ...here. * dominance.c: Moved to... * dominance.cc: ...here. * domwalk.c: Moved to... * domwalk.cc: ...here. * double-int.c: Moved to... * double-int.cc: ...here. * dse.c: Moved to... * dse.cc: ...here. * dumpfile.c: Moved to... * dumpfile.cc: ...here. * dwarf2asm.c: Moved to... * dwarf2asm.cc: ...here. * dwarf2cfi.c: Moved to... * dwarf2cfi.cc: ...here. * dwarf2ctf.c: Moved to... * dwarf2ctf.cc: ...here. * dwarf2out.c: Moved to... * dwarf2out.cc: ...here. * early-remat.c: Moved to... * early-remat.cc: ...here. * edit-context.c: Moved to... * edit-context.cc: ...here. * emit-rtl.c: Moved to... * emit-rtl.cc: ...here. * errors.c: Moved to... * errors.cc: ...here. * et-forest.c: Moved to... * et-forest.cc: ...here. * except.c: Moved to... * except.cc: ...here. * explow.c: Moved to... * explow.cc: ...here. * expmed.c: Moved to... * expmed.cc: ...here. * expr.c: Moved to... * expr.cc: ...here. * fibonacci_heap.c: Moved to... * fibonacci_heap.cc: ...here. * file-find.c: Moved to... * file-find.cc: ...here. * file-prefix-map.c: Moved to... * file-prefix-map.cc: ...here. * final.c: Moved to... * final.cc: ...here. * fixed-value.c: Moved to... * fixed-value.cc: ...here. * fold-const-call.c: Moved to... * fold-const-call.cc: ...here. * fold-const.c: Moved to... * fold-const.cc: ...here. * fp-test.c: Moved to... * fp-test.cc: ...here. * function-tests.c: Moved to... * function-tests.cc: ...here. * function.c: Moved to... * function.cc: ...here. * fwprop.c: Moved to... * fwprop.cc: ...here. * gcc-ar.c: Moved to... * gcc-ar.cc: ...here. * gcc-main.c: Moved to... * gcc-main.cc: ...here. * gcc-rich-location.c: Moved to... * gcc-rich-location.cc: ...here. * gcc.c: Moved to... * gcc.cc: ...here. * gcov-dump.c: Moved to... * gcov-dump.cc: ...here. * gcov-io.c: Moved to... * gcov-io.cc: ...here. * gcov-tool.c: Moved to... * gcov-tool.cc: ...here. * gcov.c: Moved to... * gcov.cc: ...here. * gcse-common.c: Moved to... * gcse-common.cc: ...here. * gcse.c: Moved to... * gcse.cc: ...here. * genattr-common.c: Moved to... * genattr-common.cc: ...here. * genattr.c: Moved to... * genattr.cc: ...here. * genattrtab.c: Moved to... * genattrtab.cc: ...here. * genautomata.c: Moved to... * genautomata.cc: ...here. * gencfn-macros.c: Moved to... * gencfn-macros.cc: ...here. * gencheck.c: Moved to... * gencheck.cc: ...here. * genchecksum.c: Moved to... * genchecksum.cc: ...here. * gencodes.c: Moved to... * gencodes.cc: ...here. * genconditions.c: Moved to... * genconditions.cc: ...here. * genconfig.c: Moved to... * genconfig.cc: ...here. * genconstants.c: Moved to... * genconstants.cc: ...here. * genemit.c: Moved to... * genemit.cc: ...here. * genenums.c: Moved to... * genenums.cc: ...here. * generic-match-head.c: Moved to... * generic-match-head.cc: ...here. * genextract.c: Moved to... * genextract.cc: ...here. * genflags.c: Moved to... * genflags.cc: ...here. * gengenrtl.c: Moved to... * gengenrtl.cc: ...here. * gengtype-parse.c: Moved to... * gengtype-parse.cc: ...here. * gengtype-state.c: Moved to... * gengtype-state.cc: ...here. * gengtype.c: Moved to... * gengtype.cc: ...here. * genhooks.c: Moved to... * genhooks.cc: ...here. * genmatch.c: Moved to... * genmatch.cc: ...here. * genmddeps.c: Moved to... * genmddeps.cc: ...here. * genmddump.c: Moved to... * genmddump.cc: ...here. * genmodes.c: Moved to... * genmodes.cc: ...here. * genopinit.c: Moved to... * genopinit.cc: ...here. * genoutput.c: Moved to... * genoutput.cc: ...here. * genpeep.c: Moved to... * genpeep.cc: ...here. * genpreds.c: Moved to... * genpreds.cc: ...here. * genrecog.c: Moved to... * genrecog.cc: ...here. * gensupport.c: Moved to... * gensupport.cc: ...here. * gentarget-def.c: Moved to... * gentarget-def.cc: ...here. * genversion.c: Moved to... * genversion.cc: ...here. * ggc-common.c: Moved to... * ggc-common.cc: ...here. * ggc-none.c: Moved to... * ggc-none.cc: ...here. * ggc-page.c: Moved to... * ggc-page.cc: ...here. * ggc-tests.c: Moved to... * ggc-tests.cc: ...here. * gimple-builder.c: Moved to... * gimple-builder.cc: ...here. * gimple-expr.c: Moved to... * gimple-expr.cc: ...here. * gimple-fold.c: Moved to... * gimple-fold.cc: ...here. * gimple-iterator.c: Moved to... * gimple-iterator.cc: ...here. * gimple-laddress.c: Moved to... * gimple-laddress.cc: ...here. * gimple-loop-jam.c: Moved to... * gimple-loop-jam.cc: ...here. * gimple-low.c: Moved to... * gimple-low.cc: ...here. * gimple-match-head.c: Moved to... * gimple-match-head.cc: ...here. * gimple-pretty-print.c: Moved to... * gimple-pretty-print.cc: ...here. * gimple-ssa-backprop.c: Moved to... * gimple-ssa-backprop.cc: ...here. * gimple-ssa-evrp-analyze.c: Moved to... * gimple-ssa-evrp-analyze.cc: ...here. * gimple-ssa-evrp.c: Moved to... * gimple-ssa-evrp.cc: ...here. * gimple-ssa-isolate-paths.c: Moved to... * gimple-ssa-isolate-paths.cc: ...here. * gimple-ssa-nonnull-compare.c: Moved to... * gimple-ssa-nonnull-compare.cc: ...here. * gimple-ssa-split-paths.c: Moved to... * gimple-ssa-split-paths.cc: ...here. * gimple-ssa-sprintf.c: Moved to... * gimple-ssa-sprintf.cc: ...here. * gimple-ssa-store-merging.c: Moved to... * gimple-ssa-store-merging.cc: ...here. * gimple-ssa-strength-reduction.c: Moved to... * gimple-ssa-strength-reduction.cc: ...here. * gimple-ssa-warn-alloca.c: Moved to... * gimple-ssa-warn-alloca.cc: ...here. * gimple-ssa-warn-restrict.c: Moved to... * gimple-ssa-warn-restrict.cc: ...here. * gimple-streamer-in.c: Moved to... * gimple-streamer-in.cc: ...here. * gimple-streamer-out.c: Moved to... * gimple-streamer-out.cc: ...here. * gimple-walk.c: Moved to... * gimple-walk.cc: ...here. * gimple-warn-recursion.c: Moved to... * gimple-warn-recursion.cc: ...here. * gimple.c: Moved to... * gimple.cc: ...here. * gimplify-me.c: Moved to... * gimplify-me.cc: ...here. * gimplify.c: Moved to... * gimplify.cc: ...here. * godump.c: Moved to... * godump.cc: ...here. * graph.c: Moved to... * graph.cc: ...here. * graphds.c: Moved to... * graphds.cc: ...here. * graphite-dependences.c: Moved to... * graphite-dependences.cc: ...here. * graphite-isl-ast-to-gimple.c: Moved to... * graphite-isl-ast-to-gimple.cc: ...here. * graphite-optimize-isl.c: Moved to... * graphite-optimize-isl.cc: ...here. * graphite-poly.c: Moved to... * graphite-poly.cc: ...here. * graphite-scop-detection.c: Moved to... * graphite-scop-detection.cc: ...here. * graphite-sese-to-poly.c: Moved to... * graphite-sese-to-poly.cc: ...here. * graphite.c: Moved to... * graphite.cc: ...here. * haifa-sched.c: Moved to... * haifa-sched.cc: ...here. * hash-map-tests.c: Moved to... * hash-map-tests.cc: ...here. * hash-set-tests.c: Moved to... * hash-set-tests.cc: ...here. * hash-table.c: Moved to... * hash-table.cc: ...here. * hooks.c: Moved to... * hooks.cc: ...here. * host-default.c: Moved to... * host-default.cc: ...here. * hw-doloop.c: Moved to... * hw-doloop.cc: ...here. * hwint.c: Moved to... * hwint.cc: ...here. * ifcvt.c: Moved to... * ifcvt.cc: ...here. * inchash.c: Moved to... * inchash.cc: ...here. * incpath.c: Moved to... * incpath.cc: ...here. * init-regs.c: Moved to... * init-regs.cc: ...here. * input.c: Moved to... * input.cc: ...here. * internal-fn.c: Moved to... * internal-fn.cc: ...here. * intl.c: Moved to... * intl.cc: ...here. * ipa-comdats.c: Moved to... * ipa-comdats.cc: ...here. * ipa-cp.c: Moved to... * ipa-cp.cc: ...here. * ipa-devirt.c: Moved to... * ipa-devirt.cc: ...here. * ipa-fnsummary.c: Moved to... * ipa-fnsummary.cc: ...here. * ipa-icf-gimple.c: Moved to... * ipa-icf-gimple.cc: ...here. * ipa-icf.c: Moved to... * ipa-icf.cc: ...here. * ipa-inline-analysis.c: Moved to... * ipa-inline-analysis.cc: ...here. * ipa-inline-transform.c: Moved to... * ipa-inline-transform.cc: ...here. * ipa-inline.c: Moved to... * ipa-inline.cc: ...here. * ipa-modref-tree.c: Moved to... * ipa-modref-tree.cc: ...here. * ipa-modref.c: Moved to... * ipa-modref.cc: ...here. * ipa-param-manipulation.c: Moved to... * ipa-param-manipulation.cc: ...here. * ipa-polymorphic-call.c: Moved to... * ipa-polymorphic-call.cc: ...here. * ipa-predicate.c: Moved to... * ipa-predicate.cc: ...here. * ipa-profile.c: Moved to... * ipa-profile.cc: ...here. * ipa-prop.c: Moved to... * ipa-prop.cc: ...here. * ipa-pure-const.c: Moved to... * ipa-pure-const.cc: ...here. * ipa-ref.c: Moved to... * ipa-ref.cc: ...here. * ipa-reference.c: Moved to... * ipa-reference.cc: ...here. * ipa-split.c: Moved to... * ipa-split.cc: ...here. * ipa-sra.c: Moved to... * ipa-sra.cc: ...here. * ipa-utils.c: Moved to... * ipa-utils.cc: ...here. * ipa-visibility.c: Moved to... * ipa-visibility.cc: ...here. * ipa.c: Moved to... * ipa.cc: ...here. * ira-build.c: Moved to... * ira-build.cc: ...here. * ira-color.c: Moved to... * ira-color.cc: ...here. * ira-conflicts.c: Moved to... * ira-conflicts.cc: ...here. * ira-costs.c: Moved to... * ira-costs.cc: ...here. * ira-emit.c: Moved to... * ira-emit.cc: ...here. * ira-lives.c: Moved to... * ira-lives.cc: ...here. * ira.c: Moved to... * ira.cc: ...here. * jump.c: Moved to... * jump.cc: ...here. * langhooks.c: Moved to... * langhooks.cc: ...here. * lcm.c: Moved to... * lcm.cc: ...here. * lists.c: Moved to... * lists.cc: ...here. * loop-doloop.c: Moved to... * loop-doloop.cc: ...here. * loop-init.c: Moved to... * loop-init.cc: ...here. * loop-invariant.c: Moved to... * loop-invariant.cc: ...here. * loop-iv.c: Moved to... * loop-iv.cc: ...here. * loop-unroll.c: Moved to... * loop-unroll.cc: ...here. * lower-subreg.c: Moved to... * lower-subreg.cc: ...here. * lra-assigns.c: Moved to... * lra-assigns.cc: ...here. * lra-coalesce.c: Moved to... * lra-coalesce.cc: ...here. * lra-constraints.c: Moved to... * lra-constraints.cc: ...here. * lra-eliminations.c: Moved to... * lra-eliminations.cc: ...here. * lra-lives.c: Moved to... * lra-lives.cc: ...here. * lra-remat.c: Moved to... * lra-remat.cc: ...here. * lra-spills.c: Moved to... * lra-spills.cc: ...here. * lra.c: Moved to... * lra.cc: ...here. * lto-cgraph.c: Moved to... * lto-cgraph.cc: ...here. * lto-compress.c: Moved to... * lto-compress.cc: ...here. * lto-opts.c: Moved to... * lto-opts.cc: ...here. * lto-section-in.c: Moved to... * lto-section-in.cc: ...here. * lto-section-out.c: Moved to... * lto-section-out.cc: ...here. * lto-streamer-in.c: Moved to... * lto-streamer-in.cc: ...here. * lto-streamer-out.c: Moved to... * lto-streamer-out.cc: ...here. * lto-streamer.c: Moved to... * lto-streamer.cc: ...here. * lto-wrapper.c: Moved to... * lto-wrapper.cc: ...here. * main.c: Moved to... * main.cc: ...here. * mcf.c: Moved to... * mcf.cc: ...here. * mode-switching.c: Moved to... * mode-switching.cc: ...here. * modulo-sched.c: Moved to... * modulo-sched.cc: ...here. * multiple_target.c: Moved to... * multiple_target.cc: ...here. * omp-expand.c: Moved to... * omp-expand.cc: ...here. * omp-general.c: Moved to... * omp-general.cc: ...here. * omp-low.c: Moved to... * omp-low.cc: ...here. * omp-offload.c: Moved to... * omp-offload.cc: ...here. * omp-simd-clone.c: Moved to... * omp-simd-clone.cc: ...here. * opt-suggestions.c: Moved to... * opt-suggestions.cc: ...here. * optabs-libfuncs.c: Moved to... * optabs-libfuncs.cc: ...here. * optabs-query.c: Moved to... * optabs-query.cc: ...here. * optabs-tree.c: Moved to... * optabs-tree.cc: ...here. * optabs.c: Moved to... * optabs.cc: ...here. * opts-common.c: Moved to... * opts-common.cc: ...here. * opts-global.c: Moved to... * opts-global.cc: ...here. * opts.c: Moved to... * opts.cc: ...here. * passes.c: Moved to... * passes.cc: ...here. * plugin.c: Moved to... * plugin.cc: ...here. * postreload-gcse.c: Moved to... * postreload-gcse.cc: ...here. * postreload.c: Moved to... * postreload.cc: ...here. * predict.c: Moved to... * predict.cc: ...here. * prefix.c: Moved to... * prefix.cc: ...here. * pretty-print.c: Moved to... * pretty-print.cc: ...here. * print-rtl-function.c: Moved to... * print-rtl-function.cc: ...here. * print-rtl.c: Moved to... * print-rtl.cc: ...here. * print-tree.c: Moved to... * print-tree.cc: ...here. * profile-count.c: Moved to... * profile-count.cc: ...here. * profile.c: Moved to... * profile.cc: ...here. * read-md.c: Moved to... * read-md.cc: ...here. * read-rtl-function.c: Moved to... * read-rtl-function.cc: ...here. * read-rtl.c: Moved to... * read-rtl.cc: ...here. * real.c: Moved to... * real.cc: ...here. * realmpfr.c: Moved to... * realmpfr.cc: ...here. * recog.c: Moved to... * recog.cc: ...here. * ree.c: Moved to... * ree.cc: ...here. * reg-stack.c: Moved to... * reg-stack.cc: ...here. * regcprop.c: Moved to... * regcprop.cc: ...here. * reginfo.c: Moved to... * reginfo.cc: ...here. * regrename.c: Moved to... * regrename.cc: ...here. * regstat.c: Moved to... * regstat.cc: ...here. * reload.c: Moved to... * reload.cc: ...here. * reload1.c: Moved to... * reload1.cc: ...here. * reorg.c: Moved to... * reorg.cc: ...here. * resource.c: Moved to... * resource.cc: ...here. * rtl-error.c: Moved to... * rtl-error.cc: ...here. * rtl-tests.c: Moved to... * rtl-tests.cc: ...here. * rtl.c: Moved to... * rtl.cc: ...here. * rtlanal.c: Moved to... * rtlanal.cc: ...here. * rtlhash.c: Moved to... * rtlhash.cc: ...here. * rtlhooks.c: Moved to... * rtlhooks.cc: ...here. * rtx-vector-builder.c: Moved to... * rtx-vector-builder.cc: ...here. * run-rtl-passes.c: Moved to... * run-rtl-passes.cc: ...here. * sancov.c: Moved to... * sancov.cc: ...here. * sanopt.c: Moved to... * sanopt.cc: ...here. * sbitmap.c: Moved to... * sbitmap.cc: ...here. * sched-deps.c: Moved to... * sched-deps.cc: ...here. * sched-ebb.c: Moved to... * sched-ebb.cc: ...here. * sched-rgn.c: Moved to... * sched-rgn.cc: ...here. * sel-sched-dump.c: Moved to... * sel-sched-dump.cc: ...here. * sel-sched-ir.c: Moved to... * sel-sched-ir.cc: ...here. * sel-sched.c: Moved to... * sel-sched.cc: ...here. * selftest-diagnostic.c: Moved to... * selftest-diagnostic.cc: ...here. * selftest-rtl.c: Moved to... * selftest-rtl.cc: ...here. * selftest-run-tests.c: Moved to... * selftest-run-tests.cc: ...here. * selftest.c: Moved to... * selftest.cc: ...here. * sese.c: Moved to... * sese.cc: ...here. * shrink-wrap.c: Moved to... * shrink-wrap.cc: ...here. * simplify-rtx.c: Moved to... * simplify-rtx.cc: ...here. * sparseset.c: Moved to... * sparseset.cc: ...here. * spellcheck-tree.c: Moved to... * spellcheck-tree.cc: ...here. * spellcheck.c: Moved to... * spellcheck.cc: ...here. * sreal.c: Moved to... * sreal.cc: ...here. * stack-ptr-mod.c: Moved to... * stack-ptr-mod.cc: ...here. * statistics.c: Moved to... * statistics.cc: ...here. * stmt.c: Moved to... * stmt.cc: ...here. * stor-layout.c: Moved to... * stor-layout.cc: ...here. * store-motion.c: Moved to... * store-motion.cc: ...here. * streamer-hooks.c: Moved to... * streamer-hooks.cc: ...here. * stringpool.c: Moved to... * stringpool.cc: ...here. * substring-locations.c: Moved to... * substring-locations.cc: ...here. * symtab.c: Moved to... * symtab.cc: ...here. * target-globals.c: Moved to... * target-globals.cc: ...here. * targhooks.c: Moved to... * targhooks.cc: ...here. * timevar.c: Moved to... * timevar.cc: ...here. * toplev.c: Moved to... * toplev.cc: ...here. * tracer.c: Moved to... * tracer.cc: ...here. * trans-mem.c: Moved to... * trans-mem.cc: ...here. * tree-affine.c: Moved to... * tree-affine.cc: ...here. * tree-call-cdce.c: Moved to... * tree-call-cdce.cc: ...here. * tree-cfg.c: Moved to... * tree-cfg.cc: ...here. * tree-cfgcleanup.c: Moved to... * tree-cfgcleanup.cc: ...here. * tree-chrec.c: Moved to... * tree-chrec.cc: ...here. * tree-complex.c: Moved to... * tree-complex.cc: ...here. * tree-data-ref.c: Moved to... * tree-data-ref.cc: ...here. * tree-dfa.c: Moved to... * tree-dfa.cc: ...here. * tree-diagnostic.c: Moved to... * tree-diagnostic.cc: ...here. * tree-dump.c: Moved to... * tree-dump.cc: ...here. * tree-eh.c: Moved to... * tree-eh.cc: ...here. * tree-emutls.c: Moved to... * tree-emutls.cc: ...here. * tree-if-conv.c: Moved to... * tree-if-conv.cc: ...here. * tree-inline.c: Moved to... * tree-inline.cc: ...here. * tree-into-ssa.c: Moved to... * tree-into-ssa.cc: ...here. * tree-iterator.c: Moved to... * tree-iterator.cc: ...here. * tree-loop-distribution.c: Moved to... * tree-loop-distribution.cc: ...here. * tree-nested.c: Moved to... * tree-nested.cc: ...here. * tree-nrv.c: Moved to... * tree-nrv.cc: ...here. * tree-object-size.c: Moved to... * tree-object-size.cc: ...here. * tree-outof-ssa.c: Moved to... * tree-outof-ssa.cc: ...here. * tree-parloops.c: Moved to... * tree-parloops.cc: ...here. * tree-phinodes.c: Moved to... * tree-phinodes.cc: ...here. * tree-predcom.c: Moved to... * tree-predcom.cc: ...here. * tree-pretty-print.c: Moved to... * tree-pretty-print.cc: ...here. * tree-profile.c: Moved to... * tree-profile.cc: ...here. * tree-scalar-evolution.c: Moved to... * tree-scalar-evolution.cc: ...here. * tree-sra.c: Moved to... * tree-sra.cc: ...here. * tree-ssa-address.c: Moved to... * tree-ssa-address.cc: ...here. * tree-ssa-alias.c: Moved to... * tree-ssa-alias.cc: ...here. * tree-ssa-ccp.c: Moved to... * tree-ssa-ccp.cc: ...here. * tree-ssa-coalesce.c: Moved to... * tree-ssa-coalesce.cc: ...here. * tree-ssa-copy.c: Moved to... * tree-ssa-copy.cc: ...here. * tree-ssa-dce.c: Moved to... * tree-ssa-dce.cc: ...here. * tree-ssa-dom.c: Moved to... * tree-ssa-dom.cc: ...here. * tree-ssa-dse.c: Moved to... * tree-ssa-dse.cc: ...here. * tree-ssa-forwprop.c: Moved to... * tree-ssa-forwprop.cc: ...here. * tree-ssa-ifcombine.c: Moved to... * tree-ssa-ifcombine.cc: ...here. * tree-ssa-live.c: Moved to... * tree-ssa-live.cc: ...here. * tree-ssa-loop-ch.c: Moved to... * tree-ssa-loop-ch.cc: ...here. * tree-ssa-loop-im.c: Moved to... * tree-ssa-loop-im.cc: ...here. * tree-ssa-loop-ivcanon.c: Moved to... * tree-ssa-loop-ivcanon.cc: ...here. * tree-ssa-loop-ivopts.c: Moved to... * tree-ssa-loop-ivopts.cc: ...here. * tree-ssa-loop-manip.c: Moved to... * tree-ssa-loop-manip.cc: ...here. * tree-ssa-loop-niter.c: Moved to... * tree-ssa-loop-niter.cc: ...here. * tree-ssa-loop-prefetch.c: Moved to... * tree-ssa-loop-prefetch.cc: ...here. * tree-ssa-loop-split.c: Moved to... * tree-ssa-loop-split.cc: ...here. * tree-ssa-loop-unswitch.c: Moved to... * tree-ssa-loop-unswitch.cc: ...here. * tree-ssa-loop.c: Moved to... * tree-ssa-loop.cc: ...here. * tree-ssa-math-opts.c: Moved to... * tree-ssa-math-opts.cc: ...here. * tree-ssa-operands.c: Moved to... * tree-ssa-operands.cc: ...here. * tree-ssa-phiopt.c: Moved to... * tree-ssa-phiopt.cc: ...here. * tree-ssa-phiprop.c: Moved to... * tree-ssa-phiprop.cc: ...here. * tree-ssa-pre.c: Moved to... * tree-ssa-pre.cc: ...here. * tree-ssa-propagate.c: Moved to... * tree-ssa-propagate.cc: ...here. * tree-ssa-reassoc.c: Moved to... * tree-ssa-reassoc.cc: ...here. * tree-ssa-sccvn.c: Moved to... * tree-ssa-sccvn.cc: ...here. * tree-ssa-scopedtables.c: Moved to... * tree-ssa-scopedtables.cc: ...here. * tree-ssa-sink.c: Moved to... * tree-ssa-sink.cc: ...here. * tree-ssa-strlen.c: Moved to... * tree-ssa-strlen.cc: ...here. * tree-ssa-structalias.c: Moved to... * tree-ssa-structalias.cc: ...here. * tree-ssa-tail-merge.c: Moved to... * tree-ssa-tail-merge.cc: ...here. * tree-ssa-ter.c: Moved to... * tree-ssa-ter.cc: ...here. * tree-ssa-threadbackward.c: Moved to... * tree-ssa-threadbackward.cc: ...here. * tree-ssa-threadedge.c: Moved to... * tree-ssa-threadedge.cc: ...here. * tree-ssa-threadupdate.c: Moved to... * tree-ssa-threadupdate.cc: ...here. * tree-ssa-uncprop.c: Moved to... * tree-ssa-uncprop.cc: ...here. * tree-ssa-uninit.c: Moved to... * tree-ssa-uninit.cc: ...here. * tree-ssa.c: Moved to... * tree-ssa.cc: ...here. * tree-ssanames.c: Moved to... * tree-ssanames.cc: ...here. * tree-stdarg.c: Moved to... * tree-stdarg.cc: ...here. * tree-streamer-in.c: Moved to... * tree-streamer-in.cc: ...here. * tree-streamer-out.c: Moved to... * tree-streamer-out.cc: ...here. * tree-streamer.c: Moved to... * tree-streamer.cc: ...here. * tree-switch-conversion.c: Moved to... * tree-switch-conversion.cc: ...here. * tree-tailcall.c: Moved to... * tree-tailcall.cc: ...here. * tree-vect-data-refs.c: Moved to... * tree-vect-data-refs.cc: ...here. * tree-vect-generic.c: Moved to... * tree-vect-generic.cc: ...here. * tree-vect-loop-manip.c: Moved to... * tree-vect-loop-manip.cc: ...here. * tree-vect-loop.c: Moved to... * tree-vect-loop.cc: ...here. * tree-vect-patterns.c: Moved to... * tree-vect-patterns.cc: ...here. * tree-vect-slp-patterns.c: Moved to... * tree-vect-slp-patterns.cc: ...here. * tree-vect-slp.c: Moved to... * tree-vect-slp.cc: ...here. * tree-vect-stmts.c: Moved to... * tree-vect-stmts.cc: ...here. * tree-vector-builder.c: Moved to... * tree-vector-builder.cc: ...here. * tree-vectorizer.c: Moved to... * tree-vectorizer.cc: ...here. * tree-vrp.c: Moved to... * tree-vrp.cc: ...here. * tree.c: Moved to... * tree.cc: ...here. * tsan.c: Moved to... * tsan.cc: ...here. * typed-splay-tree.c: Moved to... * typed-splay-tree.cc: ...here. * ubsan.c: Moved to... * ubsan.cc: ...here. * valtrack.c: Moved to... * valtrack.cc: ...here. * value-prof.c: Moved to... * value-prof.cc: ...here. * var-tracking.c: Moved to... * var-tracking.cc: ...here. * varasm.c: Moved to... * varasm.cc: ...here. * varpool.c: Moved to... * varpool.cc: ...here. * vec-perm-indices.c: Moved to... * vec-perm-indices.cc: ...here. * vec.c: Moved to... * vec.cc: ...here. * vmsdbgout.c: Moved to... * vmsdbgout.cc: ...here. * vr-values.c: Moved to... * vr-values.cc: ...here. * vtable-verify.c: Moved to... * vtable-verify.cc: ...here. * web.c: Moved to... * web.cc: ...here. * xcoffout.c: Moved to... * xcoffout.cc: ...here. gcc/c-family/ChangeLog: * c-ada-spec.c: Moved to... * c-ada-spec.cc: ...here. * c-attribs.c: Moved to... * c-attribs.cc: ...here. * c-common.c: Moved to... * c-common.cc: ...here. * c-cppbuiltin.c: Moved to... * c-cppbuiltin.cc: ...here. * c-dump.c: Moved to... * c-dump.cc: ...here. * c-format.c: Moved to... * c-format.cc: ...here. * c-gimplify.c: Moved to... * c-gimplify.cc: ...here. * c-indentation.c: Moved to... * c-indentation.cc: ...here. * c-lex.c: Moved to... * c-lex.cc: ...here. * c-omp.c: Moved to... * c-omp.cc: ...here. * c-opts.c: Moved to... * c-opts.cc: ...here. * c-pch.c: Moved to... * c-pch.cc: ...here. * c-ppoutput.c: Moved to... * c-ppoutput.cc: ...here. * c-pragma.c: Moved to... * c-pragma.cc: ...here. * c-pretty-print.c: Moved to... * c-pretty-print.cc: ...here. * c-semantics.c: Moved to... * c-semantics.cc: ...here. * c-ubsan.c: Moved to... * c-ubsan.cc: ...here. * c-warn.c: Moved to... * c-warn.cc: ...here. * cppspec.c: Moved to... * cppspec.cc: ...here. * stub-objc.c: Moved to... * stub-objc.cc: ...here. gcc/c/ChangeLog: * c-aux-info.c: Moved to... * c-aux-info.cc: ...here. * c-convert.c: Moved to... * c-convert.cc: ...here. * c-decl.c: Moved to... * c-decl.cc: ...here. * c-errors.c: Moved to... * c-errors.cc: ...here. * c-fold.c: Moved to... * c-fold.cc: ...here. * c-lang.c: Moved to... * c-lang.cc: ...here. * c-objc-common.c: Moved to... * c-objc-common.cc: ...here. * c-parser.c: Moved to... * c-parser.cc: ...here. * c-typeck.c: Moved to... * c-typeck.cc: ...here. * gccspec.c: Moved to... * gccspec.cc: ...here. * gimple-parser.c: Moved to... * gimple-parser.cc: ...here. gcc/cp/ChangeLog: * call.c: Moved to... * call.cc: ...here. * class.c: Moved to... * class.cc: ...here. * constexpr.c: Moved to... * constexpr.cc: ...here. * cp-gimplify.c: Moved to... * cp-gimplify.cc: ...here. * cp-lang.c: Moved to... * cp-lang.cc: ...here. * cp-objcp-common.c: Moved to... * cp-objcp-common.cc: ...here. * cp-ubsan.c: Moved to... * cp-ubsan.cc: ...here. * cvt.c: Moved to... * cvt.cc: ...here. * cxx-pretty-print.c: Moved to... * cxx-pretty-print.cc: ...here. * decl.c: Moved to... * decl.cc: ...here. * decl2.c: Moved to... * decl2.cc: ...here. * dump.c: Moved to... * dump.cc: ...here. * error.c: Moved to... * error.cc: ...here. * except.c: Moved to... * except.cc: ...here. * expr.c: Moved to... * expr.cc: ...here. * friend.c: Moved to... * friend.cc: ...here. * g++spec.c: Moved to... * g++spec.cc: ...here. * init.c: Moved to... * init.cc: ...here. * lambda.c: Moved to... * lambda.cc: ...here. * lex.c: Moved to... * lex.cc: ...here. * mangle.c: Moved to... * mangle.cc: ...here. * method.c: Moved to... * method.cc: ...here. * name-lookup.c: Moved to... * name-lookup.cc: ...here. * optimize.c: Moved to... * optimize.cc: ...here. * parser.c: Moved to... * parser.cc: ...here. * pt.c: Moved to... * pt.cc: ...here. * ptree.c: Moved to... * ptree.cc: ...here. * rtti.c: Moved to... * rtti.cc: ...here. * search.c: Moved to... * search.cc: ...here. * semantics.c: Moved to... * semantics.cc: ...here. * tree.c: Moved to... * tree.cc: ...here. * typeck.c: Moved to... * typeck.cc: ...here. * typeck2.c: Moved to... * typeck2.cc: ...here. * vtable-class-hierarchy.c: Moved to... * vtable-class-hierarchy.cc: ...here. gcc/fortran/ChangeLog: * arith.c: Moved to... * arith.cc: ...here. * array.c: Moved to... * array.cc: ...here. * bbt.c: Moved to... * bbt.cc: ...here. * check.c: Moved to... * check.cc: ...here. * class.c: Moved to... * class.cc: ...here. * constructor.c: Moved to... * constructor.cc: ...here. * convert.c: Moved to... * convert.cc: ...here. * cpp.c: Moved to... * cpp.cc: ...here. * data.c: Moved to... * data.cc: ...here. * decl.c: Moved to... * decl.cc: ...here. * dependency.c: Moved to... * dependency.cc: ...here. * dump-parse-tree.c: Moved to... * dump-parse-tree.cc: ...here. * error.c: Moved to... * error.cc: ...here. * expr.c: Moved to... * expr.cc: ...here. * f95-lang.c: Moved to... * f95-lang.cc: ...here. * frontend-passes.c: Moved to... * frontend-passes.cc: ...here. * gfortranspec.c: Moved to... * gfortranspec.cc: ...here. * interface.c: Moved to... * interface.cc: ...here. * intrinsic.c: Moved to... * intrinsic.cc: ...here. * io.c: Moved to... * io.cc: ...here. * iresolve.c: Moved to... * iresolve.cc: ...here. * match.c: Moved to... * match.cc: ...here. * matchexp.c: Moved to... * matchexp.cc: ...here. * misc.c: Moved to... * misc.cc: ...here. * module.c: Moved to... * module.cc: ...here. * openmp.c: Moved to... * openmp.cc: ...here. * options.c: Moved to... * options.cc: ...here. * parse.c: Moved to... * parse.cc: ...here. * primary.c: Moved to... * primary.cc: ...here. * resolve.c: Moved to... * resolve.cc: ...here. * scanner.c: Moved to... * scanner.cc: ...here. * simplify.c: Moved to... * simplify.cc: ...here. * st.c: Moved to... * st.cc: ...here. * symbol.c: Moved to... * symbol.cc: ...here. * target-memory.c: Moved to... * target-memory.cc: ...here. * trans-array.c: Moved to... * trans-array.cc: ...here. * trans-common.c: Moved to... * trans-common.cc: ...here. * trans-const.c: Moved to... * trans-const.cc: ...here. * trans-decl.c: Moved to... * trans-decl.cc: ...here. * trans-expr.c: Moved to... * trans-expr.cc: ...here. * trans-intrinsic.c: Moved to... * trans-intrinsic.cc: ...here. * trans-io.c: Moved to... * trans-io.cc: ...here. * trans-openmp.c: Moved to... * trans-openmp.cc: ...here. * trans-stmt.c: Moved to... * trans-stmt.cc: ...here. * trans-types.c: Moved to... * trans-types.cc: ...here. * trans.c: Moved to... * trans.cc: ...here. gcc/go/ChangeLog: * go-backend.c: Moved to... * go-backend.cc: ...here. * go-lang.c: Moved to... * go-lang.cc: ...here. * gospec.c: Moved to... * gospec.cc: ...here. gcc/jit/ChangeLog: * dummy-frontend.c: Moved to... * dummy-frontend.cc: ...here. * jit-builtins.c: Moved to... * jit-builtins.cc: ...here. * jit-logging.c: Moved to... * jit-logging.cc: ...here. * jit-playback.c: Moved to... * jit-playback.cc: ...here. * jit-recording.c: Moved to... * jit-recording.cc: ...here. * jit-result.c: Moved to... * jit-result.cc: ...here. * jit-spec.c: Moved to... * jit-spec.cc: ...here. * jit-tempdir.c: Moved to... * jit-tempdir.cc: ...here. * jit-w32.c: Moved to... * jit-w32.cc: ...here. * libgccjit.c: Moved to... * libgccjit.cc: ...here. gcc/lto/ChangeLog: * common.c: Moved to... * common.cc: ...here. * lto-common.c: Moved to... * lto-common.cc: ...here. * lto-dump.c: Moved to... * lto-dump.cc: ...here. * lto-lang.c: Moved to... * lto-lang.cc: ...here. * lto-object.c: Moved to... * lto-object.cc: ...here. * lto-partition.c: Moved to... * lto-partition.cc: ...here. * lto-symtab.c: Moved to... * lto-symtab.cc: ...here. * lto.c: Moved to... * lto.cc: ...here. gcc/objc/ChangeLog: * objc-act.c: Moved to... * objc-act.cc: ...here. * objc-encoding.c: Moved to... * objc-encoding.cc: ...here. * objc-gnu-runtime-abi-01.c: Moved to... * objc-gnu-runtime-abi-01.cc: ...here. * objc-lang.c: Moved to... * objc-lang.cc: ...here. * objc-map.c: Moved to... * objc-map.cc: ...here. * objc-next-runtime-abi-01.c: Moved to... * objc-next-runtime-abi-01.cc: ...here. * objc-next-runtime-abi-02.c: Moved to... * objc-next-runtime-abi-02.cc: ...here. * objc-runtime-shared-support.c: Moved to... * objc-runtime-shared-support.cc: ...here. gcc/objcp/ChangeLog: * objcp-decl.c: Moved to... * objcp-decl.cc: ...here. * objcp-lang.c: Moved to... * objcp-lang.cc: ...here. libcpp/ChangeLog: * charset.c: Moved to... * charset.cc: ...here. * directives.c: Moved to... * directives.cc: ...here. * errors.c: Moved to... * errors.cc: ...here. * expr.c: Moved to... * expr.cc: ...here. * files.c: Moved to... * files.cc: ...here. * identifiers.c: Moved to... * identifiers.cc: ...here. * init.c: Moved to... * init.cc: ...here. * lex.c: Moved to... * lex.cc: ...here. * line-map.c: Moved to... * line-map.cc: ...here. * macro.c: Moved to... * macro.cc: ...here. * makeucnid.c: Moved to... * makeucnid.cc: ...here. * mkdeps.c: Moved to... * mkdeps.cc: ...here. * pch.c: Moved to... * pch.cc: ...here. * symtab.c: Moved to... * symtab.cc: ...here. * traditional.c: Moved to... * traditional.cc: ...here.
Diffstat (limited to 'gcc/tree-ssa-loop-prefetch.cc')
-rw-r--r--gcc/tree-ssa-loop-prefetch.cc2108
1 files changed, 2108 insertions, 0 deletions
diff --git a/gcc/tree-ssa-loop-prefetch.cc b/gcc/tree-ssa-loop-prefetch.cc
new file mode 100644
index 0000000..aebd7c9
--- /dev/null
+++ b/gcc/tree-ssa-loop-prefetch.cc
@@ -0,0 +1,2108 @@
+/* Array prefetching.
+ Copyright (C) 2005-2022 Free Software Foundation, Inc.
+
+This file is part of GCC.
+
+GCC is free software; you can redistribute it and/or modify it
+under the terms of the GNU General Public License as published by the
+Free Software Foundation; either version 3, or (at your option) any
+later version.
+
+GCC is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+for more details.
+
+You should have received a copy of the GNU General Public License
+along with GCC; see the file COPYING3. If not see
+<http://www.gnu.org/licenses/>. */
+
+#include "config.h"
+#include "system.h"
+#include "coretypes.h"
+#include "backend.h"
+#include "target.h"
+#include "rtl.h"
+#include "tree.h"
+#include "gimple.h"
+#include "predict.h"
+#include "tree-pass.h"
+#include "gimple-ssa.h"
+#include "optabs-query.h"
+#include "tree-pretty-print.h"
+#include "fold-const.h"
+#include "stor-layout.h"
+#include "gimplify.h"
+#include "gimple-iterator.h"
+#include "gimplify-me.h"
+#include "tree-ssa-loop-ivopts.h"
+#include "tree-ssa-loop-manip.h"
+#include "tree-ssa-loop-niter.h"
+#include "tree-ssa-loop.h"
+#include "ssa.h"
+#include "tree-into-ssa.h"
+#include "cfgloop.h"
+#include "tree-scalar-evolution.h"
+#include "langhooks.h"
+#include "tree-inline.h"
+#include "tree-data-ref.h"
+#include "diagnostic-core.h"
+#include "dbgcnt.h"
+
+/* This pass inserts prefetch instructions to optimize cache usage during
+ accesses to arrays in loops. It processes loops sequentially and:
+
+ 1) Gathers all memory references in the single loop.
+ 2) For each of the references it decides when it is profitable to prefetch
+ it. To do it, we evaluate the reuse among the accesses, and determines
+ two values: PREFETCH_BEFORE (meaning that it only makes sense to do
+ prefetching in the first PREFETCH_BEFORE iterations of the loop) and
+ PREFETCH_MOD (meaning that it only makes sense to prefetch in the
+ iterations of the loop that are zero modulo PREFETCH_MOD). For example
+ (assuming cache line size is 64 bytes, char has size 1 byte and there
+ is no hardware sequential prefetch):
+
+ char *a;
+ for (i = 0; i < max; i++)
+ {
+ a[255] = ...; (0)
+ a[i] = ...; (1)
+ a[i + 64] = ...; (2)
+ a[16*i] = ...; (3)
+ a[187*i] = ...; (4)
+ a[187*i + 50] = ...; (5)
+ }
+
+ (0) obviously has PREFETCH_BEFORE 1
+ (1) has PREFETCH_BEFORE 64, since (2) accesses the same memory
+ location 64 iterations before it, and PREFETCH_MOD 64 (since
+ it hits the same cache line otherwise).
+ (2) has PREFETCH_MOD 64
+ (3) has PREFETCH_MOD 4
+ (4) has PREFETCH_MOD 1. We do not set PREFETCH_BEFORE here, since
+ the cache line accessed by (5) is the same with probability only
+ 7/32.
+ (5) has PREFETCH_MOD 1 as well.
+
+ Additionally, we use data dependence analysis to determine for each
+ reference the distance till the first reuse; this information is used
+ to determine the temporality of the issued prefetch instruction.
+
+ 3) We determine how much ahead we need to prefetch. The number of
+ iterations needed is time to fetch / time spent in one iteration of
+ the loop. The problem is that we do not know either of these values,
+ so we just make a heuristic guess based on a magic (possibly)
+ target-specific constant and size of the loop.
+
+ 4) Determine which of the references we prefetch. We take into account
+ that there is a maximum number of simultaneous prefetches (provided
+ by machine description). We prefetch as many prefetches as possible
+ while still within this bound (starting with those with lowest
+ prefetch_mod, since they are responsible for most of the cache
+ misses).
+
+ 5) We unroll and peel loops so that we are able to satisfy PREFETCH_MOD
+ and PREFETCH_BEFORE requirements (within some bounds), and to avoid
+ prefetching nonaccessed memory.
+ TODO -- actually implement peeling.
+
+ 6) We actually emit the prefetch instructions. ??? Perhaps emit the
+ prefetch instructions with guards in cases where 5) was not sufficient
+ to satisfy the constraints?
+
+ A cost model is implemented to determine whether or not prefetching is
+ profitable for a given loop. The cost model has three heuristics:
+
+ 1. Function trip_count_to_ahead_ratio_too_small_p implements a
+ heuristic that determines whether or not the loop has too few
+ iterations (compared to ahead). Prefetching is not likely to be
+ beneficial if the trip count to ahead ratio is below a certain
+ minimum.
+
+ 2. Function mem_ref_count_reasonable_p implements a heuristic that
+ determines whether the given loop has enough CPU ops that can be
+ overlapped with cache missing memory ops. If not, the loop
+ won't benefit from prefetching. In the implementation,
+ prefetching is not considered beneficial if the ratio between
+ the instruction count and the mem ref count is below a certain
+ minimum.
+
+ 3. Function insn_to_prefetch_ratio_too_small_p implements a
+ heuristic that disables prefetching in a loop if the prefetching
+ cost is above a certain limit. The relative prefetching cost is
+ estimated by taking the ratio between the prefetch count and the
+ total intruction count (this models the I-cache cost).
+
+ The limits used in these heuristics are defined as parameters with
+ reasonable default values. Machine-specific default values will be
+ added later.
+
+ Some other TODO:
+ -- write and use more general reuse analysis (that could be also used
+ in other cache aimed loop optimizations)
+ -- make it behave sanely together with the prefetches given by user
+ (now we just ignore them; at the very least we should avoid
+ optimizing loops in that user put his own prefetches)
+ -- we assume cache line size alignment of arrays; this could be
+ improved. */
+
+/* Magic constants follow. These should be replaced by machine specific
+ numbers. */
+
+/* True if write can be prefetched by a read prefetch. */
+
+#ifndef WRITE_CAN_USE_READ_PREFETCH
+#define WRITE_CAN_USE_READ_PREFETCH 1
+#endif
+
+/* True if read can be prefetched by a write prefetch. */
+
+#ifndef READ_CAN_USE_WRITE_PREFETCH
+#define READ_CAN_USE_WRITE_PREFETCH 0
+#endif
+
+/* The size of the block loaded by a single prefetch. Usually, this is
+ the same as cache line size (at the moment, we only consider one level
+ of cache hierarchy). */
+
+#ifndef PREFETCH_BLOCK
+#define PREFETCH_BLOCK param_l1_cache_line_size
+#endif
+
+/* Do we have a forward hardware sequential prefetching? */
+
+#ifndef HAVE_FORWARD_PREFETCH
+#define HAVE_FORWARD_PREFETCH 0
+#endif
+
+/* Do we have a backward hardware sequential prefetching? */
+
+#ifndef HAVE_BACKWARD_PREFETCH
+#define HAVE_BACKWARD_PREFETCH 0
+#endif
+
+/* In some cases we are only able to determine that there is a certain
+ probability that the two accesses hit the same cache line. In this
+ case, we issue the prefetches for both of them if this probability
+ is less then (1000 - ACCEPTABLE_MISS_RATE) per thousand. */
+
+#ifndef ACCEPTABLE_MISS_RATE
+#define ACCEPTABLE_MISS_RATE 50
+#endif
+
+#define L1_CACHE_SIZE_BYTES ((unsigned) (param_l1_cache_size * 1024))
+#define L2_CACHE_SIZE_BYTES ((unsigned) (param_l2_cache_size * 1024))
+
+/* We consider a memory access nontemporal if it is not reused sooner than
+ after L2_CACHE_SIZE_BYTES of memory are accessed. However, we ignore
+ accesses closer than L1_CACHE_SIZE_BYTES / NONTEMPORAL_FRACTION,
+ so that we use nontemporal prefetches e.g. if single memory location
+ is accessed several times in a single iteration of the loop. */
+#define NONTEMPORAL_FRACTION 16
+
+/* In case we have to emit a memory fence instruction after the loop that
+ uses nontemporal stores, this defines the builtin to use. */
+
+#ifndef FENCE_FOLLOWING_MOVNT
+#define FENCE_FOLLOWING_MOVNT NULL_TREE
+#endif
+
+/* It is not profitable to prefetch when the trip count is not at
+ least TRIP_COUNT_TO_AHEAD_RATIO times the prefetch ahead distance.
+ For example, in a loop with a prefetch ahead distance of 10,
+ supposing that TRIP_COUNT_TO_AHEAD_RATIO is equal to 4, it is
+ profitable to prefetch when the trip count is greater or equal to
+ 40. In that case, 30 out of the 40 iterations will benefit from
+ prefetching. */
+
+#ifndef TRIP_COUNT_TO_AHEAD_RATIO
+#define TRIP_COUNT_TO_AHEAD_RATIO 4
+#endif
+
+/* The group of references between that reuse may occur. */
+
+struct mem_ref_group
+{
+ tree base; /* Base of the reference. */
+ tree step; /* Step of the reference. */
+ struct mem_ref *refs; /* References in the group. */
+ struct mem_ref_group *next; /* Next group of references. */
+ unsigned int uid; /* Group UID, used only for debugging. */
+};
+
+/* Assigned to PREFETCH_BEFORE when all iterations are to be prefetched. */
+
+#define PREFETCH_ALL HOST_WIDE_INT_M1U
+
+/* Do not generate a prefetch if the unroll factor is significantly less
+ than what is required by the prefetch. This is to avoid redundant
+ prefetches. For example, when prefetch_mod is 16 and unroll_factor is
+ 2, prefetching requires unrolling the loop 16 times, but
+ the loop is actually unrolled twice. In this case (ratio = 8),
+ prefetching is not likely to be beneficial. */
+
+#ifndef PREFETCH_MOD_TO_UNROLL_FACTOR_RATIO
+#define PREFETCH_MOD_TO_UNROLL_FACTOR_RATIO 4
+#endif
+
+/* Some of the prefetch computations have quadratic complexity. We want to
+ avoid huge compile times and, therefore, want to limit the amount of
+ memory references per loop where we consider prefetching. */
+
+#ifndef PREFETCH_MAX_MEM_REFS_PER_LOOP
+#define PREFETCH_MAX_MEM_REFS_PER_LOOP 200
+#endif
+
+/* The memory reference. */
+
+struct mem_ref
+{
+ gimple *stmt; /* Statement in that the reference appears. */
+ tree mem; /* The reference. */
+ HOST_WIDE_INT delta; /* Constant offset of the reference. */
+ struct mem_ref_group *group; /* The group of references it belongs to. */
+ unsigned HOST_WIDE_INT prefetch_mod;
+ /* Prefetch only each PREFETCH_MOD-th
+ iteration. */
+ unsigned HOST_WIDE_INT prefetch_before;
+ /* Prefetch only first PREFETCH_BEFORE
+ iterations. */
+ unsigned reuse_distance; /* The amount of data accessed before the first
+ reuse of this value. */
+ struct mem_ref *next; /* The next reference in the group. */
+ unsigned int uid; /* Ref UID, used only for debugging. */
+ unsigned write_p : 1; /* Is it a write? */
+ unsigned independent_p : 1; /* True if the reference is independent on
+ all other references inside the loop. */
+ unsigned issue_prefetch_p : 1; /* Should we really issue the prefetch? */
+ unsigned storent_p : 1; /* True if we changed the store to a
+ nontemporal one. */
+};
+
+/* Dumps information about memory reference */
+static void
+dump_mem_details (FILE *file, tree base, tree step,
+ HOST_WIDE_INT delta, bool write_p)
+{
+ fprintf (file, "(base ");
+ print_generic_expr (file, base, TDF_SLIM);
+ fprintf (file, ", step ");
+ if (cst_and_fits_in_hwi (step))
+ fprintf (file, HOST_WIDE_INT_PRINT_DEC, int_cst_value (step));
+ else
+ print_generic_expr (file, step, TDF_SLIM);
+ fprintf (file, ")\n");
+ fprintf (file, " delta " HOST_WIDE_INT_PRINT_DEC "\n", delta);
+ fprintf (file, " %s\n\n", write_p ? "write" : "read");
+}
+
+/* Dumps information about reference REF to FILE. */
+
+static void
+dump_mem_ref (FILE *file, struct mem_ref *ref)
+{
+ fprintf (file, "reference %u:%u (", ref->group->uid, ref->uid);
+ print_generic_expr (file, ref->mem, TDF_SLIM);
+ fprintf (file, ")\n");
+}
+
+/* Finds a group with BASE and STEP in GROUPS, or creates one if it does not
+ exist. */
+
+static struct mem_ref_group *
+find_or_create_group (struct mem_ref_group **groups, tree base, tree step)
+{
+ /* Global count for setting struct mem_ref_group->uid. */
+ static unsigned int last_mem_ref_group_uid = 0;
+
+ struct mem_ref_group *group;
+
+ for (; *groups; groups = &(*groups)->next)
+ {
+ if (operand_equal_p ((*groups)->step, step, 0)
+ && operand_equal_p ((*groups)->base, base, 0))
+ return *groups;
+
+ /* If step is an integer constant, keep the list of groups sorted
+ by decreasing step. */
+ if (cst_and_fits_in_hwi ((*groups)->step) && cst_and_fits_in_hwi (step)
+ && int_cst_value ((*groups)->step) < int_cst_value (step))
+ break;
+ }
+
+ group = XNEW (struct mem_ref_group);
+ group->base = base;
+ group->step = step;
+ group->refs = NULL;
+ group->uid = ++last_mem_ref_group_uid;
+ group->next = *groups;
+ *groups = group;
+
+ return group;
+}
+
+/* Records a memory reference MEM in GROUP with offset DELTA and write status
+ WRITE_P. The reference occurs in statement STMT. */
+
+static void
+record_ref (struct mem_ref_group *group, gimple *stmt, tree mem,
+ HOST_WIDE_INT delta, bool write_p)
+{
+ unsigned int last_mem_ref_uid = 0;
+ struct mem_ref **aref;
+
+ /* Do not record the same address twice. */
+ for (aref = &group->refs; *aref; aref = &(*aref)->next)
+ {
+ last_mem_ref_uid = (*aref)->uid;
+
+ /* It does not have to be possible for write reference to reuse the read
+ prefetch, or vice versa. */
+ if (!WRITE_CAN_USE_READ_PREFETCH
+ && write_p
+ && !(*aref)->write_p)
+ continue;
+ if (!READ_CAN_USE_WRITE_PREFETCH
+ && !write_p
+ && (*aref)->write_p)
+ continue;
+
+ if ((*aref)->delta == delta)
+ return;
+ }
+
+ (*aref) = XNEW (struct mem_ref);
+ (*aref)->stmt = stmt;
+ (*aref)->mem = mem;
+ (*aref)->delta = delta;
+ (*aref)->write_p = write_p;
+ (*aref)->prefetch_before = PREFETCH_ALL;
+ (*aref)->prefetch_mod = 1;
+ (*aref)->reuse_distance = 0;
+ (*aref)->issue_prefetch_p = false;
+ (*aref)->group = group;
+ (*aref)->next = NULL;
+ (*aref)->independent_p = false;
+ (*aref)->storent_p = false;
+ (*aref)->uid = last_mem_ref_uid + 1;
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ {
+ dump_mem_ref (dump_file, *aref);
+
+ fprintf (dump_file, " group %u ", group->uid);
+ dump_mem_details (dump_file, group->base, group->step, delta,
+ write_p);
+ }
+}
+
+/* Release memory references in GROUPS. */
+
+static void
+release_mem_refs (struct mem_ref_group *groups)
+{
+ struct mem_ref_group *next_g;
+ struct mem_ref *ref, *next_r;
+
+ for (; groups; groups = next_g)
+ {
+ next_g = groups->next;
+ for (ref = groups->refs; ref; ref = next_r)
+ {
+ next_r = ref->next;
+ free (ref);
+ }
+ free (groups);
+ }
+}
+
+/* A structure used to pass arguments to idx_analyze_ref. */
+
+struct ar_data
+{
+ class loop *loop; /* Loop of the reference. */
+ gimple *stmt; /* Statement of the reference. */
+ tree *step; /* Step of the memory reference. */
+ HOST_WIDE_INT *delta; /* Offset of the memory reference. */
+};
+
+/* Analyzes a single INDEX of a memory reference to obtain information
+ described at analyze_ref. Callback for for_each_index. */
+
+static bool
+idx_analyze_ref (tree base, tree *index, void *data)
+{
+ struct ar_data *ar_data = (struct ar_data *) data;
+ tree ibase, step, stepsize;
+ HOST_WIDE_INT idelta = 0, imult = 1;
+ affine_iv iv;
+
+ if (!simple_iv (ar_data->loop, loop_containing_stmt (ar_data->stmt),
+ *index, &iv, true))
+ return false;
+ ibase = iv.base;
+ step = iv.step;
+
+ if (TREE_CODE (ibase) == POINTER_PLUS_EXPR
+ && cst_and_fits_in_hwi (TREE_OPERAND (ibase, 1)))
+ {
+ idelta = int_cst_value (TREE_OPERAND (ibase, 1));
+ ibase = TREE_OPERAND (ibase, 0);
+ }
+ if (cst_and_fits_in_hwi (ibase))
+ {
+ idelta += int_cst_value (ibase);
+ ibase = build_int_cst (TREE_TYPE (ibase), 0);
+ }
+
+ if (TREE_CODE (base) == ARRAY_REF)
+ {
+ stepsize = array_ref_element_size (base);
+ if (!cst_and_fits_in_hwi (stepsize))
+ return false;
+ imult = int_cst_value (stepsize);
+ step = fold_build2 (MULT_EXPR, sizetype,
+ fold_convert (sizetype, step),
+ fold_convert (sizetype, stepsize));
+ idelta *= imult;
+ }
+
+ if (*ar_data->step == NULL_TREE)
+ *ar_data->step = step;
+ else
+ *ar_data->step = fold_build2 (PLUS_EXPR, sizetype,
+ fold_convert (sizetype, *ar_data->step),
+ fold_convert (sizetype, step));
+ *ar_data->delta += idelta;
+ *index = ibase;
+
+ return true;
+}
+
+/* Tries to express REF_P in shape &BASE + STEP * iter + DELTA, where DELTA and
+ STEP are integer constants and iter is number of iterations of LOOP. The
+ reference occurs in statement STMT. Strips nonaddressable component
+ references from REF_P. */
+
+static bool
+analyze_ref (class loop *loop, tree *ref_p, tree *base,
+ tree *step, HOST_WIDE_INT *delta,
+ gimple *stmt)
+{
+ struct ar_data ar_data;
+ tree off;
+ HOST_WIDE_INT bit_offset;
+ tree ref = *ref_p;
+
+ *step = NULL_TREE;
+ *delta = 0;
+
+ /* First strip off the component references. Ignore bitfields.
+ Also strip off the real and imagine parts of a complex, so that
+ they can have the same base. */
+ if (TREE_CODE (ref) == REALPART_EXPR
+ || TREE_CODE (ref) == IMAGPART_EXPR
+ || (TREE_CODE (ref) == COMPONENT_REF
+ && DECL_NONADDRESSABLE_P (TREE_OPERAND (ref, 1))))
+ {
+ if (TREE_CODE (ref) == IMAGPART_EXPR)
+ *delta += int_size_in_bytes (TREE_TYPE (ref));
+ ref = TREE_OPERAND (ref, 0);
+ }
+
+ *ref_p = ref;
+
+ for (; TREE_CODE (ref) == COMPONENT_REF; ref = TREE_OPERAND (ref, 0))
+ {
+ off = DECL_FIELD_BIT_OFFSET (TREE_OPERAND (ref, 1));
+ bit_offset = TREE_INT_CST_LOW (off);
+ gcc_assert (bit_offset % BITS_PER_UNIT == 0);
+
+ *delta += bit_offset / BITS_PER_UNIT;
+ }
+
+ *base = unshare_expr (ref);
+ ar_data.loop = loop;
+ ar_data.stmt = stmt;
+ ar_data.step = step;
+ ar_data.delta = delta;
+ return for_each_index (base, idx_analyze_ref, &ar_data);
+}
+
+/* Record a memory reference REF to the list REFS. The reference occurs in
+ LOOP in statement STMT and it is write if WRITE_P. Returns true if the
+ reference was recorded, false otherwise. */
+
+static bool
+gather_memory_references_ref (class loop *loop, struct mem_ref_group **refs,
+ tree ref, bool write_p, gimple *stmt)
+{
+ tree base, step;
+ HOST_WIDE_INT delta;
+ struct mem_ref_group *agrp;
+
+ if (get_base_address (ref) == NULL)
+ return false;
+
+ if (!analyze_ref (loop, &ref, &base, &step, &delta, stmt))
+ return false;
+ /* If analyze_ref fails the default is a NULL_TREE. We can stop here. */
+ if (step == NULL_TREE)
+ return false;
+
+ /* Stop if the address of BASE could not be taken. */
+ if (may_be_nonaddressable_p (base))
+ return false;
+
+ /* Limit non-constant step prefetching only to the innermost loops and
+ only when the step is loop invariant in the entire loop nest. */
+ if (!cst_and_fits_in_hwi (step))
+ {
+ if (loop->inner != NULL)
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ {
+ fprintf (dump_file, "Memory expression %p\n",(void *) ref );
+ print_generic_expr (dump_file, ref, TDF_SLIM);
+ fprintf (dump_file,":");
+ dump_mem_details (dump_file, base, step, delta, write_p);
+ fprintf (dump_file,
+ "Ignoring %p, non-constant step prefetching is "
+ "limited to inner most loops \n",
+ (void *) ref);
+ }
+ return false;
+ }
+ else
+ {
+ if (!expr_invariant_in_loop_p (loop_outermost (loop), step))
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ {
+ fprintf (dump_file, "Memory expression %p\n",(void *) ref );
+ print_generic_expr (dump_file, ref, TDF_SLIM);
+ fprintf (dump_file,":");
+ dump_mem_details (dump_file, base, step, delta, write_p);
+ fprintf (dump_file,
+ "Not prefetching, ignoring %p due to "
+ "loop variant step\n",
+ (void *) ref);
+ }
+ return false;
+ }
+ }
+ }
+
+ /* Now we know that REF = &BASE + STEP * iter + DELTA, where DELTA and STEP
+ are integer constants. */
+ agrp = find_or_create_group (refs, base, step);
+ record_ref (agrp, stmt, ref, delta, write_p);
+
+ return true;
+}
+
+/* Record the suitable memory references in LOOP. NO_OTHER_REFS is set to
+ true if there are no other memory references inside the loop. */
+
+static struct mem_ref_group *
+gather_memory_references (class loop *loop, bool *no_other_refs, unsigned *ref_count)
+{
+ basic_block *body = get_loop_body_in_dom_order (loop);
+ basic_block bb;
+ unsigned i;
+ gimple_stmt_iterator bsi;
+ gimple *stmt;
+ tree lhs, rhs;
+ struct mem_ref_group *refs = NULL;
+
+ *no_other_refs = true;
+ *ref_count = 0;
+
+ /* Scan the loop body in order, so that the former references precede the
+ later ones. */
+ for (i = 0; i < loop->num_nodes; i++)
+ {
+ bb = body[i];
+ if (bb->loop_father != loop)
+ continue;
+
+ for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
+ {
+ stmt = gsi_stmt (bsi);
+
+ if (gimple_code (stmt) != GIMPLE_ASSIGN)
+ {
+ if (gimple_vuse (stmt)
+ || (is_gimple_call (stmt)
+ && !(gimple_call_flags (stmt) & ECF_CONST)))
+ *no_other_refs = false;
+ continue;
+ }
+
+ if (! gimple_vuse (stmt))
+ continue;
+
+ lhs = gimple_assign_lhs (stmt);
+ rhs = gimple_assign_rhs1 (stmt);
+
+ if (REFERENCE_CLASS_P (rhs))
+ {
+ *no_other_refs &= gather_memory_references_ref (loop, &refs,
+ rhs, false, stmt);
+ *ref_count += 1;
+ }
+ if (REFERENCE_CLASS_P (lhs))
+ {
+ *no_other_refs &= gather_memory_references_ref (loop, &refs,
+ lhs, true, stmt);
+ *ref_count += 1;
+ }
+ }
+ }
+ free (body);
+
+ return refs;
+}
+
+/* Prune the prefetch candidate REF using the self-reuse. */
+
+static void
+prune_ref_by_self_reuse (struct mem_ref *ref)
+{
+ HOST_WIDE_INT step;
+ bool backward;
+
+ /* If the step size is non constant, we cannot calculate prefetch_mod. */
+ if (!cst_and_fits_in_hwi (ref->group->step))
+ return;
+
+ step = int_cst_value (ref->group->step);
+
+ backward = step < 0;
+
+ if (step == 0)
+ {
+ /* Prefetch references to invariant address just once. */
+ ref->prefetch_before = 1;
+ return;
+ }
+
+ if (backward)
+ step = -step;
+
+ if (step > PREFETCH_BLOCK)
+ return;
+
+ if ((backward && HAVE_BACKWARD_PREFETCH)
+ || (!backward && HAVE_FORWARD_PREFETCH))
+ {
+ ref->prefetch_before = 1;
+ return;
+ }
+
+ ref->prefetch_mod = PREFETCH_BLOCK / step;
+}
+
+/* Divides X by BY, rounding down. */
+
+static HOST_WIDE_INT
+ddown (HOST_WIDE_INT x, unsigned HOST_WIDE_INT by)
+{
+ gcc_assert (by > 0);
+
+ if (x >= 0)
+ return x / (HOST_WIDE_INT) by;
+ else
+ return (x + (HOST_WIDE_INT) by - 1) / (HOST_WIDE_INT) by;
+}
+
+/* Given a CACHE_LINE_SIZE and two inductive memory references
+ with a common STEP greater than CACHE_LINE_SIZE and an address
+ difference DELTA, compute the probability that they will fall
+ in different cache lines. Return true if the computed miss rate
+ is not greater than the ACCEPTABLE_MISS_RATE. DISTINCT_ITERS is the
+ number of distinct iterations after which the pattern repeats itself.
+ ALIGN_UNIT is the unit of alignment in bytes. */
+
+static bool
+is_miss_rate_acceptable (unsigned HOST_WIDE_INT cache_line_size,
+ HOST_WIDE_INT step, HOST_WIDE_INT delta,
+ unsigned HOST_WIDE_INT distinct_iters,
+ int align_unit)
+{
+ unsigned align, iter;
+ int total_positions, miss_positions, max_allowed_miss_positions;
+ int address1, address2, cache_line1, cache_line2;
+
+ /* It always misses if delta is greater than or equal to the cache
+ line size. */
+ if (delta >= (HOST_WIDE_INT) cache_line_size)
+ return false;
+
+ miss_positions = 0;
+ total_positions = (cache_line_size / align_unit) * distinct_iters;
+ max_allowed_miss_positions = (ACCEPTABLE_MISS_RATE * total_positions) / 1000;
+
+ /* Iterate through all possible alignments of the first
+ memory reference within its cache line. */
+ for (align = 0; align < cache_line_size; align += align_unit)
+
+ /* Iterate through all distinct iterations. */
+ for (iter = 0; iter < distinct_iters; iter++)
+ {
+ address1 = align + step * iter;
+ address2 = address1 + delta;
+ cache_line1 = address1 / cache_line_size;
+ cache_line2 = address2 / cache_line_size;
+ if (cache_line1 != cache_line2)
+ {
+ miss_positions += 1;
+ if (miss_positions > max_allowed_miss_positions)
+ return false;
+ }
+ }
+ return true;
+}
+
+/* Prune the prefetch candidate REF using the reuse with BY.
+ If BY_IS_BEFORE is true, BY is before REF in the loop. */
+
+static void
+prune_ref_by_group_reuse (struct mem_ref *ref, struct mem_ref *by,
+ bool by_is_before)
+{
+ HOST_WIDE_INT step;
+ bool backward;
+ HOST_WIDE_INT delta_r = ref->delta, delta_b = by->delta;
+ HOST_WIDE_INT delta = delta_b - delta_r;
+ HOST_WIDE_INT hit_from;
+ unsigned HOST_WIDE_INT prefetch_before, prefetch_block;
+ HOST_WIDE_INT reduced_step;
+ unsigned HOST_WIDE_INT reduced_prefetch_block;
+ tree ref_type;
+ int align_unit;
+
+ /* If the step is non constant we cannot calculate prefetch_before. */
+ if (!cst_and_fits_in_hwi (ref->group->step)) {
+ return;
+ }
+
+ step = int_cst_value (ref->group->step);
+
+ backward = step < 0;
+
+
+ if (delta == 0)
+ {
+ /* If the references has the same address, only prefetch the
+ former. */
+ if (by_is_before)
+ ref->prefetch_before = 0;
+
+ return;
+ }
+
+ if (!step)
+ {
+ /* If the reference addresses are invariant and fall into the
+ same cache line, prefetch just the first one. */
+ if (!by_is_before)
+ return;
+
+ if (ddown (ref->delta, PREFETCH_BLOCK)
+ != ddown (by->delta, PREFETCH_BLOCK))
+ return;
+
+ ref->prefetch_before = 0;
+ return;
+ }
+
+ /* Only prune the reference that is behind in the array. */
+ if (backward)
+ {
+ if (delta > 0)
+ return;
+
+ /* Transform the data so that we may assume that the accesses
+ are forward. */
+ delta = - delta;
+ step = -step;
+ delta_r = PREFETCH_BLOCK - 1 - delta_r;
+ delta_b = PREFETCH_BLOCK - 1 - delta_b;
+ }
+ else
+ {
+ if (delta < 0)
+ return;
+ }
+
+ /* Check whether the two references are likely to hit the same cache
+ line, and how distant the iterations in that it occurs are from
+ each other. */
+
+ if (step <= PREFETCH_BLOCK)
+ {
+ /* The accesses are sure to meet. Let us check when. */
+ hit_from = ddown (delta_b, PREFETCH_BLOCK) * PREFETCH_BLOCK;
+ prefetch_before = (hit_from - delta_r + step - 1) / step;
+
+ /* Do not reduce prefetch_before if we meet beyond cache size. */
+ if (prefetch_before > absu_hwi (L2_CACHE_SIZE_BYTES / step))
+ prefetch_before = PREFETCH_ALL;
+ if (prefetch_before < ref->prefetch_before)
+ ref->prefetch_before = prefetch_before;
+
+ return;
+ }
+
+ /* A more complicated case with step > prefetch_block. First reduce
+ the ratio between the step and the cache line size to its simplest
+ terms. The resulting denominator will then represent the number of
+ distinct iterations after which each address will go back to its
+ initial location within the cache line. This computation assumes
+ that PREFETCH_BLOCK is a power of two. */
+ prefetch_block = PREFETCH_BLOCK;
+ reduced_prefetch_block = prefetch_block;
+ reduced_step = step;
+ while ((reduced_step & 1) == 0
+ && reduced_prefetch_block > 1)
+ {
+ reduced_step >>= 1;
+ reduced_prefetch_block >>= 1;
+ }
+
+ prefetch_before = delta / step;
+ delta %= step;
+ ref_type = TREE_TYPE (ref->mem);
+ align_unit = TYPE_ALIGN (ref_type) / 8;
+ if (is_miss_rate_acceptable (prefetch_block, step, delta,
+ reduced_prefetch_block, align_unit))
+ {
+ /* Do not reduce prefetch_before if we meet beyond cache size. */
+ if (prefetch_before > L2_CACHE_SIZE_BYTES / PREFETCH_BLOCK)
+ prefetch_before = PREFETCH_ALL;
+ if (prefetch_before < ref->prefetch_before)
+ ref->prefetch_before = prefetch_before;
+
+ return;
+ }
+
+ /* Try also the following iteration. */
+ prefetch_before++;
+ delta = step - delta;
+ if (is_miss_rate_acceptable (prefetch_block, step, delta,
+ reduced_prefetch_block, align_unit))
+ {
+ if (prefetch_before < ref->prefetch_before)
+ ref->prefetch_before = prefetch_before;
+
+ return;
+ }
+
+ /* The ref probably does not reuse by. */
+ return;
+}
+
+/* Prune the prefetch candidate REF using the reuses with other references
+ in REFS. */
+
+static void
+prune_ref_by_reuse (struct mem_ref *ref, struct mem_ref *refs)
+{
+ struct mem_ref *prune_by;
+ bool before = true;
+
+ prune_ref_by_self_reuse (ref);
+
+ for (prune_by = refs; prune_by; prune_by = prune_by->next)
+ {
+ if (prune_by == ref)
+ {
+ before = false;
+ continue;
+ }
+
+ if (!WRITE_CAN_USE_READ_PREFETCH
+ && ref->write_p
+ && !prune_by->write_p)
+ continue;
+ if (!READ_CAN_USE_WRITE_PREFETCH
+ && !ref->write_p
+ && prune_by->write_p)
+ continue;
+
+ prune_ref_by_group_reuse (ref, prune_by, before);
+ }
+}
+
+/* Prune the prefetch candidates in GROUP using the reuse analysis. */
+
+static void
+prune_group_by_reuse (struct mem_ref_group *group)
+{
+ struct mem_ref *ref_pruned;
+
+ for (ref_pruned = group->refs; ref_pruned; ref_pruned = ref_pruned->next)
+ {
+ prune_ref_by_reuse (ref_pruned, group->refs);
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ {
+ dump_mem_ref (dump_file, ref_pruned);
+
+ if (ref_pruned->prefetch_before == PREFETCH_ALL
+ && ref_pruned->prefetch_mod == 1)
+ fprintf (dump_file, " no restrictions");
+ else if (ref_pruned->prefetch_before == 0)
+ fprintf (dump_file, " do not prefetch");
+ else if (ref_pruned->prefetch_before <= ref_pruned->prefetch_mod)
+ fprintf (dump_file, " prefetch once");
+ else
+ {
+ if (ref_pruned->prefetch_before != PREFETCH_ALL)
+ {
+ fprintf (dump_file, " prefetch before ");
+ fprintf (dump_file, HOST_WIDE_INT_PRINT_DEC,
+ ref_pruned->prefetch_before);
+ }
+ if (ref_pruned->prefetch_mod != 1)
+ {
+ fprintf (dump_file, " prefetch mod ");
+ fprintf (dump_file, HOST_WIDE_INT_PRINT_DEC,
+ ref_pruned->prefetch_mod);
+ }
+ }
+ fprintf (dump_file, "\n");
+ }
+ }
+}
+
+/* Prune the list of prefetch candidates GROUPS using the reuse analysis. */
+
+static void
+prune_by_reuse (struct mem_ref_group *groups)
+{
+ for (; groups; groups = groups->next)
+ prune_group_by_reuse (groups);
+}
+
+/* Returns true if we should issue prefetch for REF. */
+
+static bool
+should_issue_prefetch_p (struct mem_ref *ref)
+{
+ /* Do we want to issue prefetches for non-constant strides? */
+ if (!cst_and_fits_in_hwi (ref->group->step)
+ && param_prefetch_dynamic_strides == 0)
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file,
+ "Skipping non-constant step for reference %u:%u\n",
+ ref->group->uid, ref->uid);
+ return false;
+ }
+
+ /* Some processors may have a hardware prefetcher that may conflict with
+ prefetch hints for a range of strides. Make sure we don't issue
+ prefetches for such cases if the stride is within this particular
+ range. */
+ if (cst_and_fits_in_hwi (ref->group->step)
+ && abs_hwi (int_cst_value (ref->group->step))
+ < (HOST_WIDE_INT) param_prefetch_minimum_stride)
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file,
+ "Step for reference %u:%u (" HOST_WIDE_INT_PRINT_DEC
+ ") is less than the mininum required stride of %d\n",
+ ref->group->uid, ref->uid, int_cst_value (ref->group->step),
+ param_prefetch_minimum_stride);
+ return false;
+ }
+
+ /* For now do not issue prefetches for only first few of the
+ iterations. */
+ if (ref->prefetch_before != PREFETCH_ALL)
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Ignoring reference %u:%u due to prefetch_before\n",
+ ref->group->uid, ref->uid);
+ return false;
+ }
+
+ /* Do not prefetch nontemporal stores. */
+ if (ref->storent_p)
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Ignoring nontemporal store reference %u:%u\n", ref->group->uid, ref->uid);
+ return false;
+ }
+
+ return true;
+}
+
+/* Decide which of the prefetch candidates in GROUPS to prefetch.
+ AHEAD is the number of iterations to prefetch ahead (which corresponds
+ to the number of simultaneous instances of one prefetch running at a
+ time). UNROLL_FACTOR is the factor by that the loop is going to be
+ unrolled. Returns true if there is anything to prefetch. */
+
+static bool
+schedule_prefetches (struct mem_ref_group *groups, unsigned unroll_factor,
+ unsigned ahead)
+{
+ unsigned remaining_prefetch_slots, n_prefetches, prefetch_slots;
+ unsigned slots_per_prefetch;
+ struct mem_ref *ref;
+ bool any = false;
+
+ /* At most param_simultaneous_prefetches should be running
+ at the same time. */
+ remaining_prefetch_slots = param_simultaneous_prefetches;
+
+ /* The prefetch will run for AHEAD iterations of the original loop, i.e.,
+ AHEAD / UNROLL_FACTOR iterations of the unrolled loop. In each iteration,
+ it will need a prefetch slot. */
+ slots_per_prefetch = (ahead + unroll_factor / 2) / unroll_factor;
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Each prefetch instruction takes %u prefetch slots.\n",
+ slots_per_prefetch);
+
+ /* For now we just take memory references one by one and issue
+ prefetches for as many as possible. The groups are sorted
+ starting with the largest step, since the references with
+ large step are more likely to cause many cache misses. */
+
+ for (; groups; groups = groups->next)
+ for (ref = groups->refs; ref; ref = ref->next)
+ {
+ if (!should_issue_prefetch_p (ref))
+ continue;
+
+ /* The loop is far from being sufficiently unrolled for this
+ prefetch. Do not generate prefetch to avoid many redudant
+ prefetches. */
+ if (ref->prefetch_mod / unroll_factor > PREFETCH_MOD_TO_UNROLL_FACTOR_RATIO)
+ continue;
+
+ /* If we need to prefetch the reference each PREFETCH_MOD iterations,
+ and we unroll the loop UNROLL_FACTOR times, we need to insert
+ ceil (UNROLL_FACTOR / PREFETCH_MOD) instructions in each
+ iteration. */
+ n_prefetches = ((unroll_factor + ref->prefetch_mod - 1)
+ / ref->prefetch_mod);
+ prefetch_slots = n_prefetches * slots_per_prefetch;
+
+ /* If more than half of the prefetches would be lost anyway, do not
+ issue the prefetch. */
+ if (2 * remaining_prefetch_slots < prefetch_slots)
+ continue;
+
+ /* Stop prefetching if debug counter is activated. */
+ if (!dbg_cnt (prefetch))
+ continue;
+
+ ref->issue_prefetch_p = true;
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Decided to issue prefetch for reference %u:%u\n",
+ ref->group->uid, ref->uid);
+
+ if (remaining_prefetch_slots <= prefetch_slots)
+ return true;
+ remaining_prefetch_slots -= prefetch_slots;
+ any = true;
+ }
+
+ return any;
+}
+
+/* Return TRUE if no prefetch is going to be generated in the given
+ GROUPS. */
+
+static bool
+nothing_to_prefetch_p (struct mem_ref_group *groups)
+{
+ struct mem_ref *ref;
+
+ for (; groups; groups = groups->next)
+ for (ref = groups->refs; ref; ref = ref->next)
+ if (should_issue_prefetch_p (ref))
+ return false;
+
+ return true;
+}
+
+/* Estimate the number of prefetches in the given GROUPS.
+ UNROLL_FACTOR is the factor by which LOOP was unrolled. */
+
+static int
+estimate_prefetch_count (struct mem_ref_group *groups, unsigned unroll_factor)
+{
+ struct mem_ref *ref;
+ unsigned n_prefetches;
+ int prefetch_count = 0;
+
+ for (; groups; groups = groups->next)
+ for (ref = groups->refs; ref; ref = ref->next)
+ if (should_issue_prefetch_p (ref))
+ {
+ n_prefetches = ((unroll_factor + ref->prefetch_mod - 1)
+ / ref->prefetch_mod);
+ prefetch_count += n_prefetches;
+ }
+
+ return prefetch_count;
+}
+
+/* Issue prefetches for the reference REF into loop as decided before.
+ HEAD is the number of iterations to prefetch ahead. UNROLL_FACTOR
+ is the factor by which LOOP was unrolled. */
+
+static void
+issue_prefetch_ref (struct mem_ref *ref, unsigned unroll_factor, unsigned ahead)
+{
+ HOST_WIDE_INT delta;
+ tree addr, addr_base, write_p, local, forward;
+ gcall *prefetch;
+ gimple_stmt_iterator bsi;
+ unsigned n_prefetches, ap;
+ bool nontemporal = ref->reuse_distance >= L2_CACHE_SIZE_BYTES;
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Issued%s prefetch for reference %u:%u.\n",
+ nontemporal ? " nontemporal" : "",
+ ref->group->uid, ref->uid);
+
+ bsi = gsi_for_stmt (ref->stmt);
+
+ n_prefetches = ((unroll_factor + ref->prefetch_mod - 1)
+ / ref->prefetch_mod);
+ addr_base = build_fold_addr_expr_with_type (ref->mem, ptr_type_node);
+ addr_base = force_gimple_operand_gsi (&bsi, unshare_expr (addr_base),
+ true, NULL, true, GSI_SAME_STMT);
+ write_p = ref->write_p ? integer_one_node : integer_zero_node;
+ local = nontemporal ? integer_zero_node : integer_three_node;
+
+ for (ap = 0; ap < n_prefetches; ap++)
+ {
+ if (cst_and_fits_in_hwi (ref->group->step))
+ {
+ /* Determine the address to prefetch. */
+ delta = (ahead + ap * ref->prefetch_mod) *
+ int_cst_value (ref->group->step);
+ addr = fold_build_pointer_plus_hwi (addr_base, delta);
+ addr = force_gimple_operand_gsi (&bsi, unshare_expr (addr), true,
+ NULL, true, GSI_SAME_STMT);
+ }
+ else
+ {
+ /* The step size is non-constant but loop-invariant. We use the
+ heuristic to simply prefetch ahead iterations ahead. */
+ forward = fold_build2 (MULT_EXPR, sizetype,
+ fold_convert (sizetype, ref->group->step),
+ fold_convert (sizetype, size_int (ahead)));
+ addr = fold_build_pointer_plus (addr_base, forward);
+ addr = force_gimple_operand_gsi (&bsi, unshare_expr (addr), true,
+ NULL, true, GSI_SAME_STMT);
+ }
+
+ if (addr_base != addr
+ && TREE_CODE (addr_base) == SSA_NAME
+ && TREE_CODE (addr) == SSA_NAME)
+ {
+ duplicate_ssa_name_ptr_info (addr, SSA_NAME_PTR_INFO (addr_base));
+ /* As this isn't a plain copy we have to reset alignment
+ information. */
+ if (SSA_NAME_PTR_INFO (addr))
+ mark_ptr_info_alignment_unknown (SSA_NAME_PTR_INFO (addr));
+ }
+
+ /* Create the prefetch instruction. */
+ prefetch = gimple_build_call (builtin_decl_explicit (BUILT_IN_PREFETCH),
+ 3, addr, write_p, local);
+ gsi_insert_before (&bsi, prefetch, GSI_SAME_STMT);
+ }
+}
+
+/* Issue prefetches for the references in GROUPS into loop as decided before.
+ HEAD is the number of iterations to prefetch ahead. UNROLL_FACTOR is the
+ factor by that LOOP was unrolled. */
+
+static void
+issue_prefetches (struct mem_ref_group *groups,
+ unsigned unroll_factor, unsigned ahead)
+{
+ struct mem_ref *ref;
+
+ for (; groups; groups = groups->next)
+ for (ref = groups->refs; ref; ref = ref->next)
+ if (ref->issue_prefetch_p)
+ issue_prefetch_ref (ref, unroll_factor, ahead);
+}
+
+/* Returns true if REF is a memory write for that a nontemporal store insn
+ can be used. */
+
+static bool
+nontemporal_store_p (struct mem_ref *ref)
+{
+ machine_mode mode;
+ enum insn_code code;
+
+ /* REF must be a write that is not reused. We require it to be independent
+ on all other memory references in the loop, as the nontemporal stores may
+ be reordered with respect to other memory references. */
+ if (!ref->write_p
+ || !ref->independent_p
+ || ref->reuse_distance < L2_CACHE_SIZE_BYTES)
+ return false;
+
+ /* Check that we have the storent instruction for the mode. */
+ mode = TYPE_MODE (TREE_TYPE (ref->mem));
+ if (mode == BLKmode)
+ return false;
+
+ code = optab_handler (storent_optab, mode);
+ return code != CODE_FOR_nothing;
+}
+
+/* If REF is a nontemporal store, we mark the corresponding modify statement
+ and return true. Otherwise, we return false. */
+
+static bool
+mark_nontemporal_store (struct mem_ref *ref)
+{
+ if (!nontemporal_store_p (ref))
+ return false;
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Marked reference %u:%u as a nontemporal store.\n",
+ ref->group->uid, ref->uid);
+
+ gimple_assign_set_nontemporal_move (ref->stmt, true);
+ ref->storent_p = true;
+
+ return true;
+}
+
+/* Issue a memory fence instruction after LOOP. */
+
+static void
+emit_mfence_after_loop (class loop *loop)
+{
+ auto_vec<edge> exits = get_loop_exit_edges (loop);
+ edge exit;
+ gcall *call;
+ gimple_stmt_iterator bsi;
+ unsigned i;
+
+ FOR_EACH_VEC_ELT (exits, i, exit)
+ {
+ call = gimple_build_call (FENCE_FOLLOWING_MOVNT, 0);
+
+ if (!single_pred_p (exit->dest)
+ /* If possible, we prefer not to insert the fence on other paths
+ in cfg. */
+ && !(exit->flags & EDGE_ABNORMAL))
+ split_loop_exit_edge (exit);
+ bsi = gsi_after_labels (exit->dest);
+
+ gsi_insert_before (&bsi, call, GSI_NEW_STMT);
+ }
+
+ update_ssa (TODO_update_ssa_only_virtuals);
+}
+
+/* Returns true if we can use storent in loop, false otherwise. */
+
+static bool
+may_use_storent_in_loop_p (class loop *loop)
+{
+ bool ret = true;
+
+ if (loop->inner != NULL)
+ return false;
+
+ /* If we must issue a mfence insn after using storent, check that there
+ is a suitable place for it at each of the loop exits. */
+ if (FENCE_FOLLOWING_MOVNT != NULL_TREE)
+ {
+ auto_vec<edge> exits = get_loop_exit_edges (loop);
+ unsigned i;
+ edge exit;
+
+ FOR_EACH_VEC_ELT (exits, i, exit)
+ if ((exit->flags & EDGE_ABNORMAL)
+ && exit->dest == EXIT_BLOCK_PTR_FOR_FN (cfun))
+ ret = false;
+ }
+
+ return ret;
+}
+
+/* Marks nontemporal stores in LOOP. GROUPS contains the description of memory
+ references in the loop. */
+
+static void
+mark_nontemporal_stores (class loop *loop, struct mem_ref_group *groups)
+{
+ struct mem_ref *ref;
+ bool any = false;
+
+ if (!may_use_storent_in_loop_p (loop))
+ return;
+
+ for (; groups; groups = groups->next)
+ for (ref = groups->refs; ref; ref = ref->next)
+ any |= mark_nontemporal_store (ref);
+
+ if (any && FENCE_FOLLOWING_MOVNT != NULL_TREE)
+ emit_mfence_after_loop (loop);
+}
+
+/* Determines whether we can profitably unroll LOOP FACTOR times, and if
+ this is the case, fill in DESC by the description of number of
+ iterations. */
+
+static bool
+should_unroll_loop_p (class loop *loop, class tree_niter_desc *desc,
+ unsigned factor)
+{
+ if (!can_unroll_loop_p (loop, factor, desc))
+ return false;
+
+ /* We only consider loops without control flow for unrolling. This is not
+ a hard restriction -- tree_unroll_loop works with arbitrary loops
+ as well; but the unrolling/prefetching is usually more profitable for
+ loops consisting of a single basic block, and we want to limit the
+ code growth. */
+ if (loop->num_nodes > 2)
+ return false;
+
+ return true;
+}
+
+/* Determine the coefficient by that unroll LOOP, from the information
+ contained in the list of memory references REFS. Description of
+ number of iterations of LOOP is stored to DESC. NINSNS is the number of
+ insns of the LOOP. EST_NITER is the estimated number of iterations of
+ the loop, or -1 if no estimate is available. */
+
+static unsigned
+determine_unroll_factor (class loop *loop, struct mem_ref_group *refs,
+ unsigned ninsns, class tree_niter_desc *desc,
+ HOST_WIDE_INT est_niter)
+{
+ unsigned upper_bound;
+ unsigned nfactor, factor, mod_constraint;
+ struct mem_ref_group *agp;
+ struct mem_ref *ref;
+
+ /* First check whether the loop is not too large to unroll. We ignore
+ PARAM_MAX_UNROLL_TIMES, because for small loops, it prevented us
+ from unrolling them enough to make exactly one cache line covered by each
+ iteration. Also, the goal of PARAM_MAX_UNROLL_TIMES is to prevent
+ us from unrolling the loops too many times in cases where we only expect
+ gains from better scheduling and decreasing loop overhead, which is not
+ the case here. */
+ upper_bound = param_max_unrolled_insns / ninsns;
+
+ /* If we unrolled the loop more times than it iterates, the unrolled version
+ of the loop would be never entered. */
+ if (est_niter >= 0 && est_niter < (HOST_WIDE_INT) upper_bound)
+ upper_bound = est_niter;
+
+ if (upper_bound <= 1)
+ return 1;
+
+ /* Choose the factor so that we may prefetch each cache just once,
+ but bound the unrolling by UPPER_BOUND. */
+ factor = 1;
+ for (agp = refs; agp; agp = agp->next)
+ for (ref = agp->refs; ref; ref = ref->next)
+ if (should_issue_prefetch_p (ref))
+ {
+ mod_constraint = ref->prefetch_mod;
+ nfactor = least_common_multiple (mod_constraint, factor);
+ if (nfactor <= upper_bound)
+ factor = nfactor;
+ }
+
+ if (!should_unroll_loop_p (loop, desc, factor))
+ return 1;
+
+ return factor;
+}
+
+/* Returns the total volume of the memory references REFS, taking into account
+ reuses in the innermost loop and cache line size. TODO -- we should also
+ take into account reuses across the iterations of the loops in the loop
+ nest. */
+
+static unsigned
+volume_of_references (struct mem_ref_group *refs)
+{
+ unsigned volume = 0;
+ struct mem_ref_group *gr;
+ struct mem_ref *ref;
+
+ for (gr = refs; gr; gr = gr->next)
+ for (ref = gr->refs; ref; ref = ref->next)
+ {
+ /* Almost always reuses another value? */
+ if (ref->prefetch_before != PREFETCH_ALL)
+ continue;
+
+ /* If several iterations access the same cache line, use the size of
+ the line divided by this number. Otherwise, a cache line is
+ accessed in each iteration. TODO -- in the latter case, we should
+ take the size of the reference into account, rounding it up on cache
+ line size multiple. */
+ volume += param_l1_cache_line_size / ref->prefetch_mod;
+ }
+ return volume;
+}
+
+/* Returns the volume of memory references accessed across VEC iterations of
+ loops, whose sizes are described in the LOOP_SIZES array. N is the number
+ of the loops in the nest (length of VEC and LOOP_SIZES vectors). */
+
+static unsigned
+volume_of_dist_vector (lambda_vector vec, unsigned *loop_sizes, unsigned n)
+{
+ unsigned i;
+
+ for (i = 0; i < n; i++)
+ if (vec[i] != 0)
+ break;
+
+ if (i == n)
+ return 0;
+
+ gcc_assert (vec[i] > 0);
+
+ /* We ignore the parts of the distance vector in subloops, since usually
+ the numbers of iterations are much smaller. */
+ return loop_sizes[i] * vec[i];
+}
+
+/* Add the steps of ACCESS_FN multiplied by STRIDE to the array STRIDE
+ at the position corresponding to the loop of the step. N is the depth
+ of the considered loop nest, and, LOOP is its innermost loop. */
+
+static void
+add_subscript_strides (tree access_fn, unsigned stride,
+ HOST_WIDE_INT *strides, unsigned n, class loop *loop)
+{
+ class loop *aloop;
+ tree step;
+ HOST_WIDE_INT astep;
+ unsigned min_depth = loop_depth (loop) - n;
+
+ while (TREE_CODE (access_fn) == POLYNOMIAL_CHREC)
+ {
+ aloop = get_chrec_loop (access_fn);
+ step = CHREC_RIGHT (access_fn);
+ access_fn = CHREC_LEFT (access_fn);
+
+ if ((unsigned) loop_depth (aloop) <= min_depth)
+ continue;
+
+ if (tree_fits_shwi_p (step))
+ astep = tree_to_shwi (step);
+ else
+ astep = param_l1_cache_line_size;
+
+ strides[n - 1 - loop_depth (loop) + loop_depth (aloop)] += astep * stride;
+
+ }
+}
+
+/* Returns the volume of memory references accessed between two consecutive
+ self-reuses of the reference DR. We consider the subscripts of DR in N
+ loops, and LOOP_SIZES contains the volumes of accesses in each of the
+ loops. LOOP is the innermost loop of the current loop nest. */
+
+static unsigned
+self_reuse_distance (data_reference_p dr, unsigned *loop_sizes, unsigned n,
+ class loop *loop)
+{
+ tree stride, access_fn;
+ HOST_WIDE_INT *strides, astride;
+ vec<tree> access_fns;
+ tree ref = DR_REF (dr);
+ unsigned i, ret = ~0u;
+
+ /* In the following example:
+
+ for (i = 0; i < N; i++)
+ for (j = 0; j < N; j++)
+ use (a[j][i]);
+ the same cache line is accessed each N steps (except if the change from
+ i to i + 1 crosses the boundary of the cache line). Thus, for self-reuse,
+ we cannot rely purely on the results of the data dependence analysis.
+
+ Instead, we compute the stride of the reference in each loop, and consider
+ the innermost loop in that the stride is less than cache size. */
+
+ strides = XCNEWVEC (HOST_WIDE_INT, n);
+ access_fns = DR_ACCESS_FNS (dr);
+
+ FOR_EACH_VEC_ELT (access_fns, i, access_fn)
+ {
+ /* Keep track of the reference corresponding to the subscript, so that we
+ know its stride. */
+ while (handled_component_p (ref) && TREE_CODE (ref) != ARRAY_REF)
+ ref = TREE_OPERAND (ref, 0);
+
+ if (TREE_CODE (ref) == ARRAY_REF)
+ {
+ stride = TYPE_SIZE_UNIT (TREE_TYPE (ref));
+ if (tree_fits_uhwi_p (stride))
+ astride = tree_to_uhwi (stride);
+ else
+ astride = param_l1_cache_line_size;
+
+ ref = TREE_OPERAND (ref, 0);
+ }
+ else
+ astride = 1;
+
+ add_subscript_strides (access_fn, astride, strides, n, loop);
+ }
+
+ for (i = n; i-- > 0; )
+ {
+ unsigned HOST_WIDE_INT s;
+
+ s = strides[i] < 0 ? -strides[i] : strides[i];
+
+ if (s < (unsigned) param_l1_cache_line_size
+ && (loop_sizes[i]
+ > (unsigned) (L1_CACHE_SIZE_BYTES / NONTEMPORAL_FRACTION)))
+ {
+ ret = loop_sizes[i];
+ break;
+ }
+ }
+
+ free (strides);
+ return ret;
+}
+
+/* Determines the distance till the first reuse of each reference in REFS
+ in the loop nest of LOOP. NO_OTHER_REFS is true if there are no other
+ memory references in the loop. Return false if the analysis fails. */
+
+static bool
+determine_loop_nest_reuse (class loop *loop, struct mem_ref_group *refs,
+ bool no_other_refs)
+{
+ class loop *nest, *aloop;
+ vec<data_reference_p> datarefs = vNULL;
+ vec<ddr_p> dependences = vNULL;
+ struct mem_ref_group *gr;
+ struct mem_ref *ref, *refb;
+ auto_vec<loop_p> vloops;
+ unsigned *loop_data_size;
+ unsigned i, j, n;
+ unsigned volume, dist, adist;
+ HOST_WIDE_INT vol;
+ data_reference_p dr;
+ ddr_p dep;
+
+ if (loop->inner)
+ return true;
+
+ /* Find the outermost loop of the loop nest of loop (we require that
+ there are no sibling loops inside the nest). */
+ nest = loop;
+ while (1)
+ {
+ aloop = loop_outer (nest);
+
+ if (aloop == current_loops->tree_root
+ || aloop->inner->next)
+ break;
+
+ nest = aloop;
+ }
+
+ /* For each loop, determine the amount of data accessed in each iteration.
+ We use this to estimate whether the reference is evicted from the
+ cache before its reuse. */
+ find_loop_nest (nest, &vloops);
+ n = vloops.length ();
+ loop_data_size = XNEWVEC (unsigned, n);
+ volume = volume_of_references (refs);
+ i = n;
+ while (i-- != 0)
+ {
+ loop_data_size[i] = volume;
+ /* Bound the volume by the L2 cache size, since above this bound,
+ all dependence distances are equivalent. */
+ if (volume > L2_CACHE_SIZE_BYTES)
+ continue;
+
+ aloop = vloops[i];
+ vol = estimated_stmt_executions_int (aloop);
+ if (vol == -1)
+ vol = expected_loop_iterations (aloop);
+ volume *= vol;
+ }
+
+ /* Prepare the references in the form suitable for data dependence
+ analysis. We ignore unanalyzable data references (the results
+ are used just as a heuristics to estimate temporality of the
+ references, hence we do not need to worry about correctness). */
+ for (gr = refs; gr; gr = gr->next)
+ for (ref = gr->refs; ref; ref = ref->next)
+ {
+ dr = create_data_ref (loop_preheader_edge (nest),
+ loop_containing_stmt (ref->stmt),
+ ref->mem, ref->stmt, !ref->write_p, false);
+
+ if (dr)
+ {
+ ref->reuse_distance = volume;
+ dr->aux = ref;
+ datarefs.safe_push (dr);
+ }
+ else
+ no_other_refs = false;
+ }
+
+ FOR_EACH_VEC_ELT (datarefs, i, dr)
+ {
+ dist = self_reuse_distance (dr, loop_data_size, n, loop);
+ ref = (struct mem_ref *) dr->aux;
+ if (ref->reuse_distance > dist)
+ ref->reuse_distance = dist;
+
+ if (no_other_refs)
+ ref->independent_p = true;
+ }
+
+ if (!compute_all_dependences (datarefs, &dependences, vloops, true))
+ return false;
+
+ FOR_EACH_VEC_ELT (dependences, i, dep)
+ {
+ if (DDR_ARE_DEPENDENT (dep) == chrec_known)
+ continue;
+
+ ref = (struct mem_ref *) DDR_A (dep)->aux;
+ refb = (struct mem_ref *) DDR_B (dep)->aux;
+
+ if (DDR_ARE_DEPENDENT (dep) == chrec_dont_know
+ || DDR_COULD_BE_INDEPENDENT_P (dep)
+ || DDR_NUM_DIST_VECTS (dep) == 0)
+ {
+ /* If the dependence cannot be analyzed, assume that there might be
+ a reuse. */
+ dist = 0;
+
+ ref->independent_p = false;
+ refb->independent_p = false;
+ }
+ else
+ {
+ /* The distance vectors are normalized to be always lexicographically
+ positive, hence we cannot tell just from them whether DDR_A comes
+ before DDR_B or vice versa. However, it is not important,
+ anyway -- if DDR_A is close to DDR_B, then it is either reused in
+ DDR_B (and it is not nontemporal), or it reuses the value of DDR_B
+ in cache (and marking it as nontemporal would not affect
+ anything). */
+
+ dist = volume;
+ for (j = 0; j < DDR_NUM_DIST_VECTS (dep); j++)
+ {
+ adist = volume_of_dist_vector (DDR_DIST_VECT (dep, j),
+ loop_data_size, n);
+
+ /* If this is a dependence in the innermost loop (i.e., the
+ distances in all superloops are zero) and it is not
+ the trivial self-dependence with distance zero, record that
+ the references are not completely independent. */
+ if (lambda_vector_zerop (DDR_DIST_VECT (dep, j), n - 1)
+ && (ref != refb
+ || DDR_DIST_VECT (dep, j)[n-1] != 0))
+ {
+ ref->independent_p = false;
+ refb->independent_p = false;
+ }
+
+ /* Ignore accesses closer than
+ L1_CACHE_SIZE_BYTES / NONTEMPORAL_FRACTION,
+ so that we use nontemporal prefetches e.g. if single memory
+ location is accessed several times in a single iteration of
+ the loop. */
+ if (adist < L1_CACHE_SIZE_BYTES / NONTEMPORAL_FRACTION)
+ continue;
+
+ if (adist < dist)
+ dist = adist;
+ }
+ }
+
+ if (ref->reuse_distance > dist)
+ ref->reuse_distance = dist;
+ if (refb->reuse_distance > dist)
+ refb->reuse_distance = dist;
+ }
+
+ free_dependence_relations (dependences);
+ free_data_refs (datarefs);
+ free (loop_data_size);
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ {
+ fprintf (dump_file, "Reuse distances:\n");
+ for (gr = refs; gr; gr = gr->next)
+ for (ref = gr->refs; ref; ref = ref->next)
+ fprintf (dump_file, " reference %u:%u distance %u\n",
+ ref->group->uid, ref->uid, ref->reuse_distance);
+ }
+
+ return true;
+}
+
+/* Determine whether or not the trip count to ahead ratio is too small based
+ on prefitablility consideration.
+ AHEAD: the iteration ahead distance,
+ EST_NITER: the estimated trip count. */
+
+static bool
+trip_count_to_ahead_ratio_too_small_p (unsigned ahead, HOST_WIDE_INT est_niter)
+{
+ /* Assume trip count to ahead ratio is big enough if the trip count could not
+ be estimated at compile time. */
+ if (est_niter < 0)
+ return false;
+
+ if (est_niter < (HOST_WIDE_INT) (TRIP_COUNT_TO_AHEAD_RATIO * ahead))
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file,
+ "Not prefetching -- loop estimated to roll only %d times\n",
+ (int) est_niter);
+ return true;
+ }
+
+ return false;
+}
+
+/* Determine whether or not the number of memory references in the loop is
+ reasonable based on the profitablity and compilation time considerations.
+ NINSNS: estimated number of instructions in the loop,
+ MEM_REF_COUNT: total number of memory references in the loop. */
+
+static bool
+mem_ref_count_reasonable_p (unsigned ninsns, unsigned mem_ref_count)
+{
+ int insn_to_mem_ratio;
+
+ if (mem_ref_count == 0)
+ return false;
+
+ /* Miss rate computation (is_miss_rate_acceptable) and dependence analysis
+ (compute_all_dependences) have high costs based on quadratic complexity.
+ To avoid huge compilation time, we give up prefetching if mem_ref_count
+ is too large. */
+ if (mem_ref_count > PREFETCH_MAX_MEM_REFS_PER_LOOP)
+ return false;
+
+ /* Prefetching improves performance by overlapping cache missing
+ memory accesses with CPU operations. If the loop does not have
+ enough CPU operations to overlap with memory operations, prefetching
+ won't give a significant benefit. One approximate way of checking
+ this is to require the ratio of instructions to memory references to
+ be above a certain limit. This approximation works well in practice.
+ TODO: Implement a more precise computation by estimating the time
+ for each CPU or memory op in the loop. Time estimates for memory ops
+ should account for cache misses. */
+ insn_to_mem_ratio = ninsns / mem_ref_count;
+
+ if (insn_to_mem_ratio < param_prefetch_min_insn_to_mem_ratio)
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file,
+ "Not prefetching -- instruction to memory reference ratio (%d) too small\n",
+ insn_to_mem_ratio);
+ return false;
+ }
+
+ return true;
+}
+
+/* Determine whether or not the instruction to prefetch ratio in the loop is
+ too small based on the profitablity consideration.
+ NINSNS: estimated number of instructions in the loop,
+ PREFETCH_COUNT: an estimate of the number of prefetches,
+ UNROLL_FACTOR: the factor to unroll the loop if prefetching. */
+
+static bool
+insn_to_prefetch_ratio_too_small_p (unsigned ninsns, unsigned prefetch_count,
+ unsigned unroll_factor)
+{
+ int insn_to_prefetch_ratio;
+
+ /* Prefetching most likely causes performance degradation when the instruction
+ to prefetch ratio is too small. Too many prefetch instructions in a loop
+ may reduce the I-cache performance.
+ (unroll_factor * ninsns) is used to estimate the number of instructions in
+ the unrolled loop. This implementation is a bit simplistic -- the number
+ of issued prefetch instructions is also affected by unrolling. So,
+ prefetch_mod and the unroll factor should be taken into account when
+ determining prefetch_count. Also, the number of insns of the unrolled
+ loop will usually be significantly smaller than the number of insns of the
+ original loop * unroll_factor (at least the induction variable increases
+ and the exit branches will get eliminated), so it might be better to use
+ tree_estimate_loop_size + estimated_unrolled_size. */
+ insn_to_prefetch_ratio = (unroll_factor * ninsns) / prefetch_count;
+ if (insn_to_prefetch_ratio < param_min_insn_to_prefetch_ratio)
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file,
+ "Not prefetching -- instruction to prefetch ratio (%d) too small\n",
+ insn_to_prefetch_ratio);
+ return true;
+ }
+
+ return false;
+}
+
+
+/* Issue prefetch instructions for array references in LOOP. Returns
+ true if the LOOP was unrolled. */
+
+static bool
+loop_prefetch_arrays (class loop *loop)
+{
+ struct mem_ref_group *refs;
+ unsigned ahead, ninsns, time, unroll_factor;
+ HOST_WIDE_INT est_niter;
+ class tree_niter_desc desc;
+ bool unrolled = false, no_other_refs;
+ unsigned prefetch_count;
+ unsigned mem_ref_count;
+
+ if (optimize_loop_nest_for_size_p (loop))
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, " ignored (cold area)\n");
+ return false;
+ }
+
+ /* FIXME: the time should be weighted by the probabilities of the blocks in
+ the loop body. */
+ time = tree_num_loop_insns (loop, &eni_time_weights);
+ if (time == 0)
+ return false;
+
+ ahead = (param_prefetch_latency + time - 1) / time;
+ est_niter = estimated_stmt_executions_int (loop);
+ if (est_niter == -1)
+ est_niter = likely_max_stmt_executions_int (loop);
+
+ /* Prefetching is not likely to be profitable if the trip count to ahead
+ ratio is too small. */
+ if (trip_count_to_ahead_ratio_too_small_p (ahead, est_niter))
+ return false;
+
+ ninsns = tree_num_loop_insns (loop, &eni_size_weights);
+
+ /* Step 1: gather the memory references. */
+ refs = gather_memory_references (loop, &no_other_refs, &mem_ref_count);
+
+ /* Give up prefetching if the number of memory references in the
+ loop is not reasonable based on profitablity and compilation time
+ considerations. */
+ if (!mem_ref_count_reasonable_p (ninsns, mem_ref_count))
+ goto fail;
+
+ /* Step 2: estimate the reuse effects. */
+ prune_by_reuse (refs);
+
+ if (nothing_to_prefetch_p (refs))
+ goto fail;
+
+ if (!determine_loop_nest_reuse (loop, refs, no_other_refs))
+ goto fail;
+
+ /* Step 3: determine unroll factor. */
+ unroll_factor = determine_unroll_factor (loop, refs, ninsns, &desc,
+ est_niter);
+
+ /* Estimate prefetch count for the unrolled loop. */
+ prefetch_count = estimate_prefetch_count (refs, unroll_factor);
+ if (prefetch_count == 0)
+ goto fail;
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Ahead %d, unroll factor %d, trip count "
+ HOST_WIDE_INT_PRINT_DEC "\n"
+ "insn count %d, mem ref count %d, prefetch count %d\n",
+ ahead, unroll_factor, est_niter,
+ ninsns, mem_ref_count, prefetch_count);
+
+ /* Prefetching is not likely to be profitable if the instruction to prefetch
+ ratio is too small. */
+ if (insn_to_prefetch_ratio_too_small_p (ninsns, prefetch_count,
+ unroll_factor))
+ goto fail;
+
+ mark_nontemporal_stores (loop, refs);
+
+ /* Step 4: what to prefetch? */
+ if (!schedule_prefetches (refs, unroll_factor, ahead))
+ goto fail;
+
+ /* Step 5: unroll the loop. TODO -- peeling of first and last few
+ iterations so that we do not issue superfluous prefetches. */
+ if (unroll_factor != 1)
+ {
+ tree_unroll_loop (loop, unroll_factor, &desc);
+ unrolled = true;
+ }
+
+ /* Step 6: issue the prefetches. */
+ issue_prefetches (refs, unroll_factor, ahead);
+
+fail:
+ release_mem_refs (refs);
+ return unrolled;
+}
+
+/* Issue prefetch instructions for array references in loops. */
+
+unsigned int
+tree_ssa_prefetch_arrays (void)
+{
+ bool unrolled = false;
+ int todo_flags = 0;
+
+ if (!targetm.have_prefetch ()
+ /* It is possible to ask compiler for say -mtune=i486 -march=pentium4.
+ -mtune=i486 causes us having PREFETCH_BLOCK 0, since this is part
+ of processor costs and i486 does not have prefetch, but
+ -march=pentium4 causes targetm.have_prefetch to be true. Ugh. */
+ || PREFETCH_BLOCK == 0)
+ return 0;
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ {
+ fprintf (dump_file, "Prefetching parameters:\n");
+ fprintf (dump_file, " simultaneous prefetches: %d\n",
+ param_simultaneous_prefetches);
+ fprintf (dump_file, " prefetch latency: %d\n", param_prefetch_latency);
+ fprintf (dump_file, " prefetch block size: %d\n", PREFETCH_BLOCK);
+ fprintf (dump_file, " L1 cache size: %d lines, %d kB\n",
+ L1_CACHE_SIZE_BYTES / param_l1_cache_line_size,
+ param_l1_cache_size);
+ fprintf (dump_file, " L1 cache line size: %d\n",
+ param_l1_cache_line_size);
+ fprintf (dump_file, " L2 cache size: %d kB\n", param_l2_cache_size);
+ fprintf (dump_file, " min insn-to-prefetch ratio: %d \n",
+ param_min_insn_to_prefetch_ratio);
+ fprintf (dump_file, " min insn-to-mem ratio: %d \n",
+ param_prefetch_min_insn_to_mem_ratio);
+ fprintf (dump_file, "\n");
+ }
+
+ initialize_original_copy_tables ();
+
+ if (!builtin_decl_explicit_p (BUILT_IN_PREFETCH))
+ {
+ tree type = build_function_type_list (void_type_node,
+ const_ptr_type_node, NULL_TREE);
+ tree decl = add_builtin_function ("__builtin_prefetch", type,
+ BUILT_IN_PREFETCH, BUILT_IN_NORMAL,
+ NULL, NULL_TREE);
+ DECL_IS_NOVOPS (decl) = true;
+ set_builtin_decl (BUILT_IN_PREFETCH, decl, false);
+ }
+
+ for (auto loop : loops_list (cfun, LI_FROM_INNERMOST))
+ {
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "Processing loop %d:\n", loop->num);
+
+ unrolled |= loop_prefetch_arrays (loop);
+
+ if (dump_file && (dump_flags & TDF_DETAILS))
+ fprintf (dump_file, "\n\n");
+ }
+
+ if (unrolled)
+ {
+ scev_reset ();
+ todo_flags |= TODO_cleanup_cfg;
+ }
+
+ free_original_copy_tables ();
+ return todo_flags;
+}
+
+/* Prefetching. */
+
+namespace {
+
+const pass_data pass_data_loop_prefetch =
+{
+ GIMPLE_PASS, /* type */
+ "aprefetch", /* name */
+ OPTGROUP_LOOP, /* optinfo_flags */
+ TV_TREE_PREFETCH, /* tv_id */
+ ( PROP_cfg | PROP_ssa ), /* properties_required */
+ 0, /* properties_provided */
+ 0, /* properties_destroyed */
+ 0, /* todo_flags_start */
+ 0, /* todo_flags_finish */
+};
+
+class pass_loop_prefetch : public gimple_opt_pass
+{
+public:
+ pass_loop_prefetch (gcc::context *ctxt)
+ : gimple_opt_pass (pass_data_loop_prefetch, ctxt)
+ {}
+
+ /* opt_pass methods: */
+ virtual bool gate (function *) { return flag_prefetch_loop_arrays > 0; }
+ virtual unsigned int execute (function *);
+
+}; // class pass_loop_prefetch
+
+unsigned int
+pass_loop_prefetch::execute (function *fun)
+{
+ if (number_of_loops (fun) <= 1)
+ return 0;
+
+ if ((PREFETCH_BLOCK & (PREFETCH_BLOCK - 1)) != 0)
+ {
+ static bool warned = false;
+
+ if (!warned)
+ {
+ warning (OPT_Wdisabled_optimization,
+ "%<l1-cache-size%> parameter is not a power of two %d",
+ PREFETCH_BLOCK);
+ warned = true;
+ }
+ return 0;
+ }
+
+ return tree_ssa_prefetch_arrays ();
+}
+
+} // anon namespace
+
+gimple_opt_pass *
+make_pass_loop_prefetch (gcc::context *ctxt)
+{
+ return new pass_loop_prefetch (ctxt);
+}
+
+
generated by cgit v1.1 at 2025-05-21 09:54:55 +0000