https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78166
Bug ID: 78166
Summary: [6 Regression] hash.c:1887:1: error: unrecognizable
insn
Product: gcc
Version: 6.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: danglin at gcc dot gnu.org
CC: law at gcc dot gnu.org
Target Milestone: ---
Host: hppa-unknown-linux-gnu
Target: hppa-unknown-linux-gnu
Build: hppa-unknown-linux-gnu
Created attachment 39924
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=39924&action=edit
Patch
Building racket-6.7, we have the following error:
gcc -g -O2 -fdebug-prefix-map=/<<PKGBUILDDIR>>=. -Wformat
-Werror=format-security -Wall -I./..
-I/<<PKGBUILDDIR>>/src/racket/src/../include -Wdate-time -D_FORTIFY_SOURCE=2
-DUSE_SENORA_GC -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -c
/<<PKGBUILDDIR>>/src/racket/src/hash.c -o hash.o
/<<PKGBUILDDIR>>/src/racket/src/hash.c: In function 'equal_hash_key':
/<<PKGBUILDDIR>>/src/racket/src/hash.c:1887:1: error: unrecognizable insn:
}
^
(insn 1891 425 1572 62 (set (reg:SI 21 %r21)
(plus:SI (mult:SI (reg:SI 4 %r4 [orig:163 ivtmp.808 ] [163])
(const_int 8 [0x8]))
(reg/v/f:SI 3 %r3 [orig:137 o ] [137]))) -1
(nil))
/<<PKGBUILDDIR>>/src/racket/src/hash.c:1887:1: internal compiler error: in
extract_insn, at recog.c:2287
This insn doesn't exist. However, we do have the corresponding shadd
patterns.
We have the following after ira:
(insn 424 435 425 62 (set (reg/v:DF 227 [ d ])
(mem:DF (plus:SI (mult:SI (reg:SI 163 [ ivtmp.808 ])
(const_int 8 [0x8]))
(reg/v/f:SI 137 [ o ])) [36 MEM[base: o_4, index:
ivtmp.808_501,
step: 8, offset: 0B]+0 S8 A64]))
/home/dave/debian/racket/racket-6.7/src/racket
/src/hash.c:1608 74 {*pa.md:3808}
(expr_list:REG_EQUIV (mem:DF (plus:SI (mult:SI (reg:SI 163 [ ivtmp.808 ])
(const_int 8 [0x8]))
(reg/v/f:SI 137 [ o ])) [36 MEM[base: o_4, index:
ivtmp.808_501,
step: 8, offset: 0B]+0 S8 A64])
(nil)))
(debug_insn 425 424 1572 62 (var_location:DF d (reg/v:DF 227 [ d ]))
/home/dave/
debian/racket/racket-6.7/src/racket/src/hash.c:1608 -1
(nil))
(insn 1572 425 1573 62 (set (reg:SI 715 [ _479 ])
(subreg:SI (reg/v:DF 227 [ d ]) 0)) 42 {*pa.md:2184}
(nil))
(insn 1573 1572 437 62 (set (reg:SI 716 [ _479+4 ])
(subreg:SI (reg/v:DF 227 [ d ]) 4)) 42 {*pa.md:2184}
(nil))
Then, during reload, we have:
Spilling for insn 1572.
Using reg 28 for reload 0
Spilling for insn 1573.
Using reg 28 for reload 0
Using reg 28 for reload 1
Reloads for insn # 1572
Reload 0: reload_in (SI) = (plus:SI (mult:SI (reg:SI 4 %r4 [orig:163 ivtmp.808
]
[163])
(const_int 8 [0x8]))
(reg/v/f:SI 3 %r3 [orig:137
o ] [137]))
GENERAL_REGS, RELOAD_FOR_OPERAND_ADDRESS (opnum = 1)
reload_in_reg: (plus:SI (mult:SI (reg:SI 4 %r4 [orig:163 ivtmp.808 ]
[163])
(const_int 8 [0x8]))
(reg/v/f:SI 3 %r3 [orig:137
o ] [137]))
reload_reg_rtx: (reg:SI 21 %r21)
Reloads for insn # 1573
Reload 0: reload_in (SI) = (plus:SI (reg/v/f:SI 3 %r3 [orig:137 o ] [137])
(const_int 4 [0x4]))
GENERAL_REGS, RELOAD_FOR_OPERAND_ADDRESS (opnum = 1)
reload_in_reg: (plus:SI (reg/v/f:SI 3 %r3 [orig:137 o ] [137])
(const_int 4 [0x4]))
reload_reg_rtx: (reg:SI 22 %r22)Reload 1: reload_in (SI) = (plus:SI
(mult:SI (reg:SI 4 %r4 [orig:163 ivtmp.808 ]
[163])
(const_int 8 [0x8]))
(plus:SI (reg/v/f:SI 3 %r3
[orig:137 o ] [137])
(const_int 4 [0x4])))
GENERAL_REGS, RELOAD_FOR_OPERAND_ADDRESS (opnum = 1)
reload_in_reg: (plus:SI (mult:SI (reg:SI 4 %r4 [orig:163 ivtmp.808 ]
[163])
(const_int 8 [0x8]))
(plus:SI (reg/v/f:SI 3 %r3
[orig:137 o ] [137])
(const_int 4 [0x4])))
reload_reg_rtx: (reg:SI 23 %r23)
Reloads for insn # 443
Reload 0: reload_in (SI) = (plus:SI (mult:SI (reg:SI 4 %r4 [orig:163 ivtmp.808
] [163])
(const_int 8 [0x8]))
(reg/v/f:SI 3 %r3 [orig:137
o ] [137]))
GENERAL_REGS, RELOAD_FOR_INPUT_ADDRESS (opnum = 1)
reload_in_reg: (plus:SI (mult:SI (reg:SI 4 %r4 [orig:163 ivtmp.808 ]
[163])
(const_int 8 [0x8]))
(reg/v/f:SI 3 %r3 [orig:137
o ] [137]))
reload_reg_rtx: (reg:SI 24 %r24)
Reload 1: reload_in (SI) = (plus:SI (reg/v/f:SI 3 %r3 [orig:137 o ] [137])
(const_int 4 [0x4]))
GENERAL_REGS, RELOAD_FOR_INPUT_ADDRESS (opnum = 2)
reload_in_reg: (plus:SI (reg/v/f:SI 3 %r3 [orig:137 o ] [137])
(const_int 4 [0x4]))
reload_reg_rtx: (reg:SI 24 %r24)
Reload 2: reload_in (SI) = (plus:SI (mult:SI (reg:SI 4 %r4 [orig:163 ivtmp.808
] [163])
(const_int 8 [0x8]))
(plus:SI (reg/v/f:SI 3 %r3
[
orig:137 o ] [137])
(const_int 4 [0x4])))
GENERAL_REGS, RELOAD_FOR_INPUT_ADDRESS (opnum = 2)
reload_in_reg: (plus:SI (mult:SI (reg:SI 4 %r4 [orig:163 ivtmp.808 ]
[163])
(const_int 8 [0x8]))
(plus:SI (reg/v/f:SI 3 %r3
[orig:137 o ] [137])
(const_int 4 [0x4])))
reload_reg_rtx: (reg:SI 29 %r29)
Reload 3: reload_in (SI) = (mem:SI (plus:SI (mult:SI (reg:SI 4 %r4 [orig:163
ivtmp.808 ] [163])
(const_int 8
[0x8]))
(reg/v/f:SI 3 %r3
[orig:137 o ] [137])) [36 MEM[base: o_4, index: ivtmp.808_501, step: 8, offset:
0B]+0 S4 A64])
GENERAL_REGS, RELOAD_FOR_INPUT (opnum = 1), can't combine
reload_in_reg: (subreg:SI (reg/v:DF 227 [ d ]) 0)
reload_reg_rtx: (reg:SI 31 %r31)
Reload 4: reload_in (SI) = (mem:SI (plus:SI (mult:SI (reg:SI 4 %r4 [orig:163
ivtmp.808 ] [163])
(const_int 8
[0x8]))
(plus:SI (reg/v/f:SI 3
%r3 [orig:137 o ] [137])
(const_int 4
[0x4]))) [36 MEM[base: o_4, index: ivtmp.808_501, step: 8, offset: 0B]+4 S4
A32])
GENERAL_REGS, RELOAD_FOR_INPUT (opnum = 2), can't combine
reload_in_reg: (subreg:SI (reg/v:DF 227 [ d ]) 4)
reload_reg_rtx: (reg:SI 24 %r24)
The above was with gcc-6:
dave@mx3210:~/debian/racket/racket-6.7$ gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/lib/gcc/hppa-linux-gnu/6/lto-wrapper
Target: hppa-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Debian 6.2.0-10'
--with-bugurl=file:///usr/share/doc/gcc-6/README.Bugs
--enable-languages=c,ada,c++,java,d,fortran,objc,obj-c++ --prefix=/usr
--program-suffix=-6 --program-prefix=hppa-linux-gnu- --enable-shared
--enable-linker-build-id --libexecdir=/usr/lib --without-included-gettext
--enable-threads=posix --libdir=/usr/lib --enable-nls --with-sysroot=/
--enable-clocale=gnu --enable-libstdcxx-debug --enable-libstdcxx-time=yes
--with-default-libstdcxx-abi=new --enable-gnu-unique-object --disable-libssp
--disable-libitm --disable-libsanitizer --disable-libquadmath --enable-plugin
--with-system-zlib --disable-browser-plugin --enable-java-awt=gtk
--enable-gtk-cairo --with-java-home=/usr/lib/jvm/java-1.5.0-gcj-6-hppa/jre
--enable-java-home --with-jvm-root-dir=/usr/lib/jvm/java-1.5.0-gcj-6-hppa
--with-jvm-jar-dir=/usr/lib/jvm-exports/java-1.5.0-gcj-6-hppa
--with-arch-directory=parisc --with-ecj-jar=/usr/share/java/eclipse-ecj.jar
--enable-objc-gc --enable-multiarch --disable-libstdcxx-pch
--enable-checking=release --build=hppa-linux-gnu --host=hppa-linux-gnu
--target=hppa-linux-gnu
Thread model: posix
gcc version 6.2.0 20161027 (Debian 6.2.0-10)
I believe this problem was introduced by the following change:
2015-05-22 Jeff Law <[email protected]>
* config/pa/pa.md (integer_indexed_store splitters): Use
mem_shadd_operand. Use ASHIFT rather than MULT in the resulting
insns -- adjusting the constant 2nd operand accordingly.
* combine.c (try_combine): Canonicalize (plus (mult X pow2) Y) into
(plus (ashift X log2) Y) if it is a split point.
* config/pa/pa.c (mem_shadd_or_shadd_rtx_p): New function factoredx
out of hppa_legitimize_address to handle both forms of a multiply
by 2, 4 or 8.
(hppa_legitimize_address): Use mem_shadd_or_shadd_rtx_p.
Always generate the ASHIFT variant as the result is not directly
used in a MEM. Update comments and refactor slightly to improve
readability.
At least for gcc-6, the safest thing seems to be to add the missing MULT
patterns as they are needed by reload. However, the goal of Jeff's change
was to use ASHIFT rather than MULT.
Testing attached change.
