Merge lp:~uweigand/gcc-linaro/neon-extendsidi-4.7 into lp:gcc-linaro/4.7

[Andrew Stubbs]

Better optimize SI->DI extends that move from core to NEON registers.

Patch posted upstream here:
http://gcc.gnu.org/ml/gcc-patches/2012-02/msg01213.html

UPDATE 20120327: the shifts patch should be fixed now. This patch needs that patch as prerequisite.

UPDATE 20120404: fixed misapplied patch.

[Michael Hope (michaelh1)]

cbuild has taken a snapshot of this branch at r114970 and queued it for build.

The diff against the ancestor r114969 is available at:
 http://builds.linaro.org/toolchain/snapshots/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7.diff

and will be built on the following builders:
 a9-builder armv5-builder i686 x86_64

You can track the build queue at:
 http://ex.seabright.co.nz/helpers/scheduler

cbuild-snapshot: gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7
cbuild-ancestor: lp:gcc-linaro/4.7+bzr114969
cbuild-state: check

[Michael Hope (michaelh1)]

cbuild had trouble building this on armv7l-natty-cbuild259-tcpanda02-cortexa9r1.

See the *failed.txt logs under the build results at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild259-tcpanda02-cortexa9r1

The test suite results were not checked.

cbuild-checked: armv7l-natty-cbuild259-tcpanda02-cortexa9r1

review: Needs Fixing

[Michael Hope (michaelh1)]

cbuild successfully built this on x86_64-natty-cbuild259-oort1-x86_64r1.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild259-oort1-x86_64r1

The testsuite results are the same as the branch point lp:gcc-linaro/4.7+bzr114969

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild259-oort1-x86_64r1/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild259-oort1-x86_64r1/gcc-testsuite.txt

cbuild-checked: x86_64-natty-cbuild259-oort1-x86_64r1

[Andrew Stubbs (ams-codesourcery)]

I forgot that this patch depends on the neon-shifts patch! I'll resubmit when that one works.

review: Needs Fixing

[Michael Hope (michaelh1)]

cbuild successfully built this on armv7l-natty-cbuild259-tcpanda03-armv5r2.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild259-tcpanda03-armv5r2

-PASS: gcc.dg/vect/vect-cond-7.c (test for excess errors)
-PASS: gcc.dg/vect/vect-cond-7.c -flto (test for excess errors)
-PASS: gcc.dg/vect/vect-cond-7.c -flto execution test
-PASS: gcc.dg/vect/vect-cond-7.c execution test
+FAIL: gcc.dg/vect/vect-cond-7.c (internal compiler error)
+FAIL: gcc.dg/vect/vect-cond-7.c (test for excess errors)
+FAIL: gcc.dg/vect/vect-cond-7.c -flto (internal compiler error)
+FAIL: gcc.dg/vect/vect-cond-7.c -flto (test for excess errors)
+UNRESOLVED: gcc.dg/vect/vect-cond-7.c -flto compilation failed to produce executable
+UNRESOLVED: gcc.dg/vect/vect-cond-7.c compilation failed to produce executable
+FAIL: gcc.target/arm/neon-extend-1.c (internal compiler error)
+FAIL: gcc.target/arm/neon-extend-1.c (test for excess errors)
+UNRESOLVED: gcc.target/arm/neon-extend-1.c scan-assembler vdup.32
+UNRESOLVED: gcc.target/arm/neon-extend-1.c scan-assembler vshr.u64
+FAIL: gcc.target/arm/neon-extend-2.c (internal compiler error)
+FAIL: gcc.target/arm/neon-extend-2.c (test for excess errors)
+UNRESOLVED: gcc.target/arm/neon-extend-2.c scan-assembler vdup.32
+UNRESOLVED: gcc.target/arm/neon-extend-2.c scan-assembler vshr.s64

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild259-tcpanda03-armv5r2/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild259-tcpanda03-armv5r2/gcc-testsuite.txt

cbuild-checked: armv7l-natty-cbuild259-tcpanda03-armv5r2

[Michael Hope (michaelh1)]

cbuild successfully built this on i686-natty-cbuild259-oort2-i686r1.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/i686-natty-cbuild259-oort2-i686r1

The testsuite results are the same as the branch point lp:gcc-linaro/4.7+bzr114969

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/i686-natty-cbuild259-oort2-i686r1/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114970~ams-codesourcery~neon-extendsidi-4.7/logs/i686-natty-cbuild259-oort2-i686r1/gcc-testsuite.txt

cbuild-checked: i686-natty-cbuild259-oort2-i686r1

[Michael Hope (michaelh1)]

cbuild has taken a snapshot of this branch at r114971 and queued it for build.

The diff against the ancestor r114969 is available at:
 http://builds.linaro.org/toolchain/snapshots/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7.diff

and will be built on the following builders:
 a9-builder armv5-builder i686 x86_64

You can track the build queue at:
 http://ex.seabright.co.nz/helpers/scheduler

cbuild-snapshot: gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7
cbuild-ancestor: lp:gcc-linaro/4.7+bzr114969
cbuild-state: check

[Michael Hope (michaelh1)]

cbuild successfully built this on x86_64-natty-cbuild280-oort1-x86_64r1.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild280-oort1-x86_64r1

The testsuite results are the same as the branch point lp:gcc-linaro/4.7+bzr114969

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild280-oort1-x86_64r1/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild280-oort1-x86_64r1/gcc-testsuite.txt

cbuild-checked: x86_64-natty-cbuild280-oort1-x86_64r1

[Michael Hope (michaelh1)]

cbuild had trouble building this on armv7l-natty-cbuild280-tcpanda05-cortexa9r1.

See the *failed.txt logs under the build results at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild280-tcpanda05-cortexa9r1

The test suite results were not checked.

cbuild-checked: armv7l-natty-cbuild280-tcpanda05-cortexa9r1

review: Needs Fixing

[Michael Hope (michaelh1)]

cbuild successfully built this on armv7l-natty-cbuild280-tcpanda05-armv5r2.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild280-tcpanda05-armv5r2

+PASS: gcc.target/arm/neon-extend-1.c (test for excess errors)
+PASS: gcc.target/arm/neon-extend-1.c scan-assembler vdup.32
+PASS: gcc.target/arm/neon-extend-1.c scan-assembler vshr.u64
+PASS: gcc.target/arm/neon-extend-2.c (test for excess errors)
+PASS: gcc.target/arm/neon-extend-2.c scan-assembler vdup.32
+PASS: gcc.target/arm/neon-extend-2.c scan-assembler vshr.s64
+UNSUPPORTED: gcc.target/arm/thumb-16bit-ops.c
+UNSUPPORTED: gcc.target/arm/thumb-ifcvt.c

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild280-tcpanda05-armv5r2/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild280-tcpanda05-armv5r2/gcc-testsuite.txt

cbuild-checked: armv7l-natty-cbuild280-tcpanda05-armv5r2

[Michael Hope (michaelh1)]

cbuild successfully built this on i686-natty-cbuild280-oort2-i686r1.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/i686-natty-cbuild280-oort2-i686r1

The testsuite results are the same as the branch point lp:gcc-linaro/4.7+bzr114969

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/i686-natty-cbuild280-oort2-i686r1/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114971~ams-codesourcery~neon-extendsidi-4.7/logs/i686-natty-cbuild280-oort2-i686r1/gcc-testsuite.txt

cbuild-checked: i686-natty-cbuild280-oort2-i686r1

[Michael Hope (michaelh1)]

This seems to also include the 16 bit instructions patch:

+UNSUPPORTED: gcc.target/arm/thumb-16bit-ops.c
+UNSUPPORTED: gcc.target/arm/thumb-ifcvt.c

2012-03-08 Andrew Stubbs <email address hidden>

[Michael Hope (michaelh1)]

The Cortex-A9 build fails with a split ICE:

../../../../gcc-linaro-4.7/libgcc/fixed-bit.c: In function '__gnu_satfractqqdq2':
../../../../gcc-linaro-4.7/libgcc/fixed-bit.c:873:1: error: could not split insn
(insn:TI 25 192 193 4 (set (reg/v:DI 95 d16 [orig:137 utemp ] [137])
        (sign_extend:DI (reg:QI 0 r0 [orig:135 D.8823 ] [135]))) ../../../../gcc-linaro-4.7/libgcc/fixed-bit.c:803 156 {extendqidi2}
     (nil))
../../../../gcc-linaro-4.7/libgcc/fixed-bit.c:873:1: internal compiler error: in final_scan_insn, at final.c:2716

review: Disapprove

[Andrew Stubbs (ams-codesourcery)]

This patch depends on the 16-bit instructions patch (or more accurately, it depends on the neon-shifts patch, which in turn depends on the 16-bit instruction patch), hence the presence of the test cases.

The build failure is surprising. I saw that failure myself, and fixed it already, so I don't know why it's recurred. Maybe I misapplied the patch?

[Andrew Stubbs (ams-codesourcery)]

Yeah, I misapplied the patch update. Somehow the QI->DI and HI->DI extends are missing.

I've updated the branch now, but I'll wait until the neon-shifts patch is working before resubmitting this for more testing.

review: Needs Fixing

[Andrew Stubbs (ams-codesourcery)]

Neon shifts appears good, after all. I'll resubmit this now.

[Michael Hope (michaelh1)]

cbuild has taken a snapshot of this branch at r114972 and queued it for build.

The diff against the ancestor r114969 is available at:
 http://builds.linaro.org/toolchain/snapshots/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7.diff

and will be built on the following builders:
 a9-builder armv5-builder i686 x86_64

You can track the build queue at:
 http://ex.seabright.co.nz/helpers/scheduler

cbuild-snapshot: gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7
cbuild-ancestor: lp:gcc-linaro+bzr114969
cbuild-state: check

[Michael Hope (michaelh1)]

cbuild successfully built this on x86_64-natty-cbuild285-oort1-x86_64r1.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild285-oort1-x86_64r1

The testsuite results are the same as the branch point lp:gcc-linaro+bzr114969

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild285-oort1-x86_64r1/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild285-oort1-x86_64r1/gcc-testsuite.txt

cbuild-checked: x86_64-natty-cbuild285-oort1-x86_64r1

[Michael Hope (michaelh1)]

cbuild successfully built this on x86_64-natty-cbuild285-oort1-x86_64r1.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild285-oort1-x86_64r1

The testsuite results are the same as the branch point lp:gcc-linaro+bzr114969

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild285-oort1-x86_64r1/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/x86_64-natty-cbuild285-oort1-x86_64r1/gcc-testsuite.txt

cbuild-checked: x86_64-natty-cbuild285-oort1-x86_64r1

[Michael Hope (michaelh1)]

cbuild successfully built this on i686-natty-cbuild285-oort4-i686r1.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/i686-natty-cbuild285-oort4-i686r1

The testsuite results are the same as the branch point lp:gcc-linaro+bzr114969

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/i686-natty-cbuild285-oort4-i686r1/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/i686-natty-cbuild285-oort4-i686r1/gcc-testsuite.txt

cbuild-checked: i686-natty-cbuild285-oort4-i686r1

[Michael Hope (michaelh1)]

cbuild had trouble building this on armv7l-natty-cbuild286-ursa3-cortexa9r1.

See the *failed.txt logs under the build results at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild286-ursa3-cortexa9r1

The test suite results were not checked.

cbuild-checked: armv7l-natty-cbuild286-ursa3-cortexa9r1

review: Needs Fixing

[Michael Hope (michaelh1)]

cbuild successfully built this on armv7l-natty-cbuild286-ursa4-armv5r2.

The build results are available at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild286-ursa4-armv5r2

+PASS: gcc.target/arm/neon-extend-1.c (test for excess errors)
+PASS: gcc.target/arm/neon-extend-1.c scan-assembler vdup.32
+PASS: gcc.target/arm/neon-extend-1.c scan-assembler vshr.u64
+PASS: gcc.target/arm/neon-extend-2.c (test for excess errors)
+PASS: gcc.target/arm/neon-extend-2.c scan-assembler vdup.32
+PASS: gcc.target/arm/neon-extend-2.c scan-assembler vshr.s64
+UNSUPPORTED: gcc.target/arm/thumb-16bit-ops.c
+UNSUPPORTED: gcc.target/arm/thumb-ifcvt.c

The full diff is at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild286-ursa4-armv5r2/testsuite-diff.txt

The full testsuite results are at:
 http://ex.seabright.co.nz/build/gcc-linaro-4.7+bzr114972~ams-codesourcery~neon-extendsidi-4.7/logs/armv7l-natty-cbuild286-ursa4-armv5r2/gcc-testsuite.txt

cbuild-checked: armv7l-natty-cbuild286-ursa4-armv5r2

[Andrew Stubbs (ams-codesourcery)]

The build failure is this:

../../../../gcc-linaro-4.7/libgfortran/io/write.c: In function 'extract_int':
../../../../gcc-linaro-4.7/libgfortran/io/write.c:466:1: error: insn does not satisfy its constraints:
(insn 20 19 21 4 (set (reg/v:DI 2 r2 [orig:134 i ] [134])
        (sign_extend:DI (mem:QI (reg:SI 3 r3) [0 MEM[(char * {ref-all})p_4(D)]+0 S1 A8]))) ../../../../gcc-linaro-4.7/libgfortran/io/write.c:428 156 {extendqidi2}
     (nil))
../../../../gcc-linaro-4.7/libgfortran/io/write.c:466:1: internal compiler error: in reload_cse_simplify_operands, at postreload.c:403
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

This appears to be caused by a latent bug: the extendqidi2 pattern (actually "extend<mode>di2") has a predicate that allows memory loads in both ARM and Thumb modes, but the constraints only permit mems in ARM mode. Presumably this hasn't been a problem before because a) when optimizing the pattern has always been split before register allocation (i.e. before the constraints really mean anything), and b) when not optimizing it's too stupid to do load and extend in one insn.

It didn't show up in my own test builds. This might be because I have a slightly different baseline, or it might be because I wasn't building fortran.

I can fix this by adding a proper alternative for thumb mode, I expect.