[Bug target/83245] New: [8 regression] several tests fail starting with r255258

["seurer at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83245

            Bug ID: 83245
           Summary: [8 regression] several tests fail starting with
                    r255258
           Product: gcc
           Version: 8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: seurer at gcc dot gnu.org
  Target Milestone: ---

Starting with r255258 there are 5 test cases that fail on powerpc64 BE (LE is
fine).

> FAIL: gcc.dg/torture/stackalign/builtin-apply-2.c   -Os -fpic (test for excess errors)
> FAIL: gcc.dg/torture/stackalign/builtin-return-1.c   -Os -fpic (test for excess errors)
> FAIL: gfortran.dg/alloc_comp_assign_2.f90   -Os  (test for excess errors)
> FAIL: gfortran.dg/defined_assignment_2.f90   -Os  execution test
> FAIL: libgomp.fortran/allocatable9.f90   -Os  (test for excess errors)


They all seem to be failing for the same reason:


spawn -ignore SIGHUP
/home/seurer/gcc/build/gcc-bisect/gcc/testsuite/gfortran/../../gfortran
-B/home/seurer/gcc/build/gcc-bisect/gcc/testsuite/gfortran/../../
-B/home/seurer/gcc/build/gcc-bisect/powerpc64-unknown-linux-gnu/./libgfortran/
/home/seurer/gcc/gcc-bisect/gcc/testsuite/gfortran.dg/alloc_comp_assign_2.f90
-fno-diagnostics-show-caret -fdiagnostics-color=never -Os -pedantic-errors
-B/home/seurer/gcc/build/gcc-bisect/powerpc64-unknown-linux-gnu/./libgfortran/.libs
-L/home/seurer/gcc/build/gcc-bisect/powerpc64-unknown-linux-gnu/./libgfortran/.libs
-L/home/seurer/gcc/build/gcc-bisect/powerpc64-unknown-linux-gnu/./libgfortran/.libs
-L/home/seurer/gcc/build/gcc-bisect/powerpc64-unknown-linux-gnu/./libatomic/.libs
-B/home/seurer/gcc/build/gcc-bisect/powerpc64-unknown-linux-gnu/./libquadmath/.libs
-L/home/seurer/gcc/build/gcc-bisect/powerpc64-unknown-linux-gnu/./libquadmath/.libs
-L/home/seurer/gcc/build/gcc-bisect/powerpc64-unknown-linux-gnu/./libquadmath/.libs
-lm -o ./alloc_comp_assign_2.exe
/tmp/ccAojICz.s: Assembler messages:
/tmp/ccAojICz.s:703: Error: address register in load range
compiler exited with status 1
output is:
/tmp/ccAojICz.s: Assembler messages:
/tmp/ccAojICz.s:703: Error: address register in load range

FAIL: gfortran.dg/alloc_comp_assign_2.f90   -Os  (test for excess errors)
Excess errors:
/tmp/ccAojICz.s:703:0: Error: address register in load range


spawn -ignore SIGHUP /home/seurer/gcc/build/gcc-bisect/gcc/xgcc
-B/home/seurer/gcc/build/gcc-bisect/gcc/
/home/seurer/gcc/gcc-bisect/gcc/testsuite/gcc.dg/torture/stackalign/builtin-apply-2.c
-fno-diagnostics-show-caret -fdiagnostics-color=never -Os -fpic -lm -o
./builtin-apply-2.exe
/tmp/ccmQaGmG.s: Assembler messages:
/tmp/ccmQaGmG.s:116: Error: address register in load range
compiler exited with status 1
output is:
/tmp/ccmQaGmG.s: Assembler messages:
/tmp/ccmQaGmG.s:116: Error: address register in load range

FAIL: gcc.dg/torture/stackalign/builtin-apply-2.c   -Os -fpic (test for excess
errors)
Excess errors:
/tmp/ccmQaGmG.s:116: Error: address register in load range


spawn -ignore SIGHUP /home/seurer/gcc/build/gcc-bisect/gcc/xgcc
-B/home/seurer/gcc/build/gcc-bisect/gcc/
/home/seurer/gcc/gcc-bisect/gcc/testsuite/gcc.dg/torture/stackalign/builtin-return-1.c
-fno-diagnostics-show-caret -fdiagnostics-color=never -Os -fpic -lm -o
./builtin-return-1.exe
/tmp/ccbLTFT6.s: Assembler messages:
/tmp/ccbLTFT6.s:112: Error: address register in load range
compiler exited with status 1
output is:
/tmp/ccbLTFT6.s: Assembler messages:
/tmp/ccbLTFT6.s:112: Error: address register in load range

FAIL: gcc.dg/torture/stackalign/builtin-return-1.c   -Os -fpic (test for excess
errors)
Excess errors:
/tmp/ccbLTFT6.s:112: Error: address register in load range


Looking in detail at builtin-apply-2.s line 116 is:

        lswi 5,9,0

The previous revision used r4 here instead of r9. r9 is in the range of
registers that lswi is loading so the instruction is invalid.


Load String Word Immediate
lswi RT,RA,NB

Let the effective address (EA) be (RA|0). Let n = NB if
NB≠0, n = 32 if NB=0; n is the number of bytes to load.
Let nr=CEIL(n/4); nr is the number of registers to
receive data.

n consecutive bytes starting at EA are loaded into
GPRs RT through RT+nr-1.

If RA is in the range of registers to be loaded, including
the case in which RA=0, the instruction form is invalid.



seurer@makalu-lp1:~/gcc/build/gcc-test$ diff builtin-apply-2.s.255257
builtin-apply-2.s.255258
33c33
<       stdu 1,-832(1)
---
> 	stdu 1,-816(1)
51d50
<       std 3,56(31)
52a52
> 	std 3,56(31)
66c66,67
<       addi 3,31,880
---
> 	addi 3,31,864
> 	addi 4,31,48
103,104d103
<       addi 9,31,48
<       std 9,672(31)
106d104
<       std 9,592(31)
107a106
> 	std 9,592(31)
111,112c110,111
<       mr 4,0
<       addi 0,31,912
---
> 	mr 9,0
> 	addi 0,31,896
114,115c113,114
<       stswi 5,4,0
<       mr 4,0
---
> 	stswi 5,9,0
> 	mr 9,0
117c116
<       lswi 5,4,0
---
> 	lswi 5,9,0
119c118,125
<       ld 10,672(31)
---
> 	mr 10,4
> 	ld 11,608(31)
> 	ld 3,8(4)
> 	ld 5,24(10)
> 	ld 6,32(10)
> 	ld 7,40(10)
> 	ld 8,48(10)
> 	ld 9,56(10)
122a129,130
> 	ld 4,16(4)
> 	ld 10,64(10)
126,132d133
<       ld 3,8(10)
<       ld 4,16(10)
<       ld 5,24(10)
<       ld 6,32(10)
<       ld 7,40(10)
<       ld 8,48(10)
<       ld 9,56(10)
134d134
<       ld 10,64(10)
145d144
<       ld 11,608(31)
174c173
<       addi 1,31,832
---
> 	addi 1,31,816
228c227
<       .ident  "GCC: (GNU) 8.0.0 20171129 (experimental) [trunk revision
255257]"
---
> 	.ident	"GCC: (GNU) 8.0.0 20171129 (experimental) [trunk revision 255258]"