Compiling FFTW 3.1.1 on a new Intel Mac, gcc 4.0.1 generates unassemblable code
for several of the main files, e.g. the following,
[stp:/Users/stp/Code/MacLibs/fftw-3.1.1/kernel] make trig.lo
gcc -DHAVE_CONFIG_H -I. -I. -I.. -I../simd -O3 -fomit-frame-pointer
-malign-double -fstrict-aliasing -ffast-math -march=pentiumpro -D_THREAD_SAFE
-MT trig.lo -MD -MP -MF .deps/trig.Tpo -c trig.c -o trig.o
/var/tmp//ccL7WCuH.s:79:operands given don't match any known 386 instruction
/var/tmp//ccL7WCuH.s:83:operands given don't match any known 386 instruction
/var/tmp//ccL7WCuH.s:112:operands given don't match any known 386 instruction
/var/tmp//ccL7WCuH.s:114:operands given don't match any known 386 instruction
/var/tmp//ccL7WCuH.s:138:operands given don't match any known 386 instruction
/var/tmp//ccL7WCuH.s:144:operands given don't match any known 386 instruction
make: *** [trig.lo] Error 1
I'm attaching the broken assembly file below.
If I compile with no flags, it succeeds; as in,
[stp:/Users/stp/Code/MacLibs/fftw-3.1.1/kernel] make CFLAGS="-O2" trig.lo
gcc -DHAVE_CONFIG_H -I. -I. -I.. -I../simd -O2 -MT trig.lo -MD -MP -MF
.deps/trig.Tpo -c trig.c -o trig.o
which works fine.
Here's the gcc version:
Target: i686-apple-darwin8
Configured with: /private/var/tmp/gcc/gcc-5250.obj~12/src/configure
--disable-checking -enable-werror --prefix=/usr --mandir=/share/man
--enable-languages=c,objc,c++,obj-c++
--program-transform-name=/^[cg][^.-]*$/s/$/-4.0/
--with-gxx-include-dir=/include/c++/4.0.0 --build=powerpc-apple-darwin8
--with-arch=pentium-m --with-tune=prescott --program-prefix=
--host=i686-apple-darwin8 --target=i686-apple-darwin8
Thread model: posix
gcc version 4.0.1 (Apple Computer, Inc. build 5250)
Here's the broken trig.s file
.text
_cexpl_sqrtn_table:
pushl %edi
pushl %esi
subl $4, %esp
movl 16(%esp), %ecx
movl 20(%esp), %edx
movl 24(%esp), %edi
movl %edx, %eax
shrl $31, %eax
imull 32(%ecx), %eax
addl %eax, %edx
movl 24(%ecx), %esi
movl %edx, %eax
andl 20(%ecx), %eax
sall $4, %eax
fldl (%esi,%eax)
fldl 8(%esi,%eax)
movl 28(%ecx), %eax
movl 12(%ecx), %ecx
sarl %cl, %edx
sall $4, %edx
fldl (%eax,%edx)
fldl 8(%eax,%edx)
fld %st(3)
fmul %st(2), %st
fld %st(3)
fmul %st(2), %st
fsubrp %st, %st(1)
fstpl (%edi)
fmulp %st, %st(3)
fmulp %st, %st(1)
faddp %st, %st(1)
fstpl 8(%edi)
addl $4, %esp
popl %esi
popl %edi
ret
_rotate_sqrtn_table:
pushl %edi
pushl %esi
subl $4, %esp
movl 16(%esp), %ecx
movl 20(%esp), %edx
movl 32(%esp), %edi
movl %edx, %eax
shrl $31, %eax
imull 32(%ecx), %eax
addl %eax, %edx
movl 24(%ecx), %esi
movl %edx, %eax
andl 20(%ecx), %eax
sall $4, %eax
fldl (%esi,%eax)
fldl 8(%esi,%eax)
movl 28(%ecx), %eax
movl 12(%ecx), %ecx
sarl %cl, %edx
sall $4, %edx
fldl (%eax,%edx)
fldl 8(%eax,%edx)
fld %st(3)
fmul %st(2), %st
fld %st(3)
fmul %st(2), %st
fsubrp %st, %st(1)
fxch %st(4)
fmulp %st, %st(1)
fxch %st(2)
fmulp %st, %st(1)
faddp %st, %st(1)
flds 24(%esp)
flds 28(%esp)
fld %st(3)
fmul %st(2), %st
fld %st(3)
fmul %st(2), %st
faddp %st, %st(1)
fstp (%edi)
fmulp %st, %st(3)
fmulp %st, %st(1)
fsubrp %st, %st(1)
fstp 4(%edi)
addl $4, %esp
popl %esi
popl %edi
ret
_cexp_zero:
movl 12(%esp), %edx
xorl %eax, %eax
movl %eax, (%edx)
movl %eax, 4(%edx)
ret
_cexpl_zero:
movl 12(%esp), %eax
fldz
fstl (%eax)
fstpl 8(%eax)
ret
_cexp_generic:
pushl %esi
subl $40, %esp
movl 48(%esp), %edx
movl 56(%esp), %esi
leal 16(%esp), %eax
movl %eax, 8(%esp)
movl 52(%esp), %eax
movl %eax, 4(%esp)
movl %edx, (%esp)
call *4(%edx)
fldl 16(%esp)
fstp (%esi)
fldl 24(%esp)
fstp 4(%esi)
addl $40, %esp
popl %esi
ret
_rotate_generic:
pushl %esi
subl $40, %esp
movl 48(%esp), %edx
movl 64(%esp), %esi
leal 16(%esp), %eax
movl %eax, 8(%esp)
movl 52(%esp), %eax
movl %eax, 4(%esp)
movl %edx, (%esp)
call *4(%edx)
flds 56(%esp)
fldl 16(%esp)
flds 60(%esp)
fldl 24(%esp)
fld %st(3)
fmul %st(3), %st
fld %st(2)
fmul %st(2), %st
faddp %st, %st(1)
fstp (%esi)
fxch %st(2)
fmulp %st, %st(1)
fxch %st(1)
fmulp %st, %st(2)
fsubp %st, %st(1)
fstp 4(%esi)
addl $40, %esp
popl %esi
ret
.globl _fftwf_triggen_destroy
_fftwf_triggen_destroy:
pushl %esi
subl $24, %esp
movl 32(%esp), %esi
movl 24(%esi), %eax
movl %eax, (%esp)
call L_fftwf_ifree0$stub
movl 28(%esi), %eax
movl %eax, (%esp)
call L_fftwf_ifree0$stub
movl %esi, 32(%esp)
addl $24, %esp
popl %esi
jmp L_fftwf_ifree$stub
.literal8
.align 3
LC2:
.long 1413754136
.long 1075388923
.text
_cexpl_sincos:
pushl %ebp
pushl %edi
pushl %esi
pushl %ebx
subl $4, %esp
call ___i686.get_pc_thunk.bx
"L00000000001$pb":
movl 24(%esp), %eax
movl 32(%eax), %esi
leal 0(,%esi,4), %edi
movl 28(%esp), %edx
sall $2, %edx
leal (%edx,%edi), %eax
cmpl $-1, %edx
cmovle %eax, %edx
movl %edi, %ecx
subl %edx, %ecx
cmpl %ecx, %edx
jle L18
movl $4, %ebp
movl %ecx, %eax
subl %esi, %eax
testl %eax, %eax
jle L36
L21:
orl $2, %ebp
L23:
movl %esi, %edx
subl %eax, %edx
cmpl %edx, %eax
jg L24
movl %eax, %edx
L26:
fldl LC2-"L00000000001$pb"(%ebx)
pushl %edx
fimull (%esp)
movl %edi, (%esp)
fidivl (%esp)
addl $4, %esp
fsincos
fld %st(0)
testl $1, %ebp
jne L37
fstp %st(0)
fxch %st(1)
L27:
testl $2, %ebp
je L31
fchs
fxch %st(1)
L31:
andl $4, %ebp
je L32
fchs
L32:
fxch %st(1)
movl 32(%esp), %eax
fstpl (%eax)
fstpl 8(%eax)
addl $4, %esp
popl %ebx
popl %esi
popl %edi
popl %ebp
ret
L37:
fstp %st(1)
jmp L27
L24:
orl $1, %ebp
jmp L26
L18:
movl %edx, %ecx
xorl %ebp, %ebp
movl %ecx, %eax
subl %esi, %eax
testl %eax, %eax
jg L21
L36:
movl %ecx, %eax
jmp L23
.literal8
.align 3
LC5:
.long 1413754136
.long 1075388923
.text
.globl _fftwf_mktriggen
_fftwf_mktriggen:
pushl %ebp
pushl %edi
pushl %esi
pushl %ebx
subl $76, %esp
call ___i686.get_pc_thunk.bx
"L00000000002$pb":
movl 96(%esp), %esi
movl $36, (%esp)
call L_fftwf_malloc_plain$stub
movl %eax, %edi
movl 100(%esp), %eax
movl %eax, 32(%edi)
movl $0, 28(%edi)
movl $0, 24(%edi)
movl $0, (%edi)
movl $0, 8(%edi)
cmpl $2, %esi
je L41
cmpl $3, %esi
je L42
subl $1, %esi
je L100
L115:
movl (%edi), %ecx
testl %ecx, %ecx
je L117
L89:
movl 8(%edi), %edx
testl %edx, %edx
je L118
movl %edi, %eax
addl $76, %esp
popl %ebx
popl %esi
popl %edi
popl %ebp
ret
L118:
leal _rotate_generic-"L00000000002$pb"(%ebx), %eax
movl %eax, 8(%edi)
movl %edi, %eax
addl $76, %esp
popl %ebx
popl %esi
popl %edi
popl %ebp
ret
L41:
movl 100(%esp), %eax
testl %eax, %eax
jle L103
movl 100(%esp), %eax
xorl %ecx, %ecx
L44:
addl $1, %ecx
sarl $2, %eax
testl %eax, %eax
jg L44
movl $1, 48(%esp)
sall %cl, 48(%esp)
L46:
movl %ecx, 12(%edi)
movl 48(%esp), %edx
movl %edx, 16(%edi)
movl %edx, %eax
subl $1, %eax
movl %eax, 20(%edi)
movl 100(%esp), %ecx
leal -1(%ecx,%edx), %esi
movl %esi, %eax
cltd
idivl 48(%esp)
movl %eax, 32(%esp)
movl 48(%esp), %eax
sall $4, %eax
movl %eax, (%esp)
call L_fftwf_malloc_plain$stub
movl %eax, 24(%edi)
movl 32(%esp), %eax
sall $4, %eax
movl %eax, (%esp)
call L_fftwf_malloc_plain$stub
movl %eax, 40(%esp)
movl %eax, 28(%edi)
movl 48(%esp), %ebp
testl %ebp, %ebp
jle L47
movl 100(%esp), %edx
sall $2, %edx
movl %edx, 52(%esp)
fildl 52(%esp)
movl $0, 44(%esp)
movl $0, 56(%esp)
fldl LC5-"L00000000002$pb"(%ebx)
movl 24(%edi), %ebp
L49:
movl 56(%esp), %edx
testl %edx, %edx
js L105
L50:
movl 52(%esp), %eax
subl %edx, %eax
cmpl %eax, %edx
jg L106
movl %edx, %eax
xorl %esi, %esi
movl %eax, %ecx
subl 100(%esp), %ecx
testl %ecx, %ecx
jle L104
L55:
orl $2, %esi
movl 100(%esp), %edx
subl %ecx, %edx
cmpl %edx, %ecx
jg L58
L107:
movl %ecx, %edx
fld %st(0)
pushl %edx
fimull (%esp)
addl $4, %esp
fdiv %st(2), %st
fsincos
fld %st(0)
testl $1, %esi
jne L111
L120:
fstp %st(0)
fxch %st(1)
L61:
testl $2, %esi
je L65
fchs
fxch %st(1)
L65:
andl $4, %esi
je L66
fchs
L66:
fxch %st(1)
fstpl (%ebp)
fstpl 8(%ebp)
addl $1, 44(%esp)
addl $4, 56(%esp)
addl $16, %ebp
movl 44(%esp), %ecx
cmpl %ecx, 48(%esp)
jne L49
fstp %st(0)
fstp %st(0)
L47:
movl 32(%esp), %esi
testl %esi, %esi
jle L68
movl 100(%esp), %ebp
sall $2, %ebp
pushl %ebp
fildl (%esp)
addl $4, %esp
movl 16(%edi), %eax
movl %eax, 36(%esp)
movl 40(%esp), %esi
movl $0, 60(%esp)
fldl LC5-"L00000000002$pb"(%ebx)
movl $0, 28(%esp)
L70:
movl 28(%esp), %edx
sall $2, %edx
leal (%edx,%ebp), %eax
cmpl $-1, %edx
cmovle %eax, %edx
movl %ebp, %eax
subl %edx, %eax
cmpl %eax, %edx
jg L109
movl %edx, %eax
xorl %ecx, %ecx
movl %eax, %edx
subl 100(%esp), %edx
testl %edx, %edx
jle L108
L76:
orl $2, %ecx
movl 100(%esp), %eax
subl %edx, %eax
cmpl %eax, %edx
jg L79
L110:
movl %edx, %eax
fld %st(0)
pushl %eax
fimull (%esp)
addl $4, %esp
fdiv %st(2), %st
fsincos
fld %st(0)
testb $1, %cl
jne L113
L119:
fstp %st(0)
fxch %st(1)
L82:
testb $2, %cl
je L86
fchs
fxch %st(1)
L86:
andb $4, %cl
je L87
fchs
L87:
fxch %st(1)
fstpl (%esi)
fstpl 8(%esi)
addl $1, 60(%esp)
movl 36(%esp), %edx
addl %edx, 28(%esp)
addl $16, %esi
movl 60(%esp), %ecx
cmpl %ecx, 32(%esp)
jne L70
fstp %st(0)
fstp %st(0)
L68:
leal _cexpl_sqrtn_table-"L00000000002$pb"(%ebx), %eax
movl %eax, 4(%edi)
leal _rotate_sqrtn_table-"L00000000002$pb"(%ebx), %eax
movl %eax, 8(%edi)
movl (%edi), %ecx
testl %ecx, %ecx
jne L89
L117:
leal _cexp_generic-"L00000000002$pb"(%ebx), %eax
movl %eax, (%edi)
jmp L89
L42:
leal _cexpl_sincos-"L00000000002$pb"(%ebx), %eax
movl %eax, 4(%edi)
jmp L115
L100:
leal _cexp_zero-"L00000000002$pb"(%ebx), %eax
movl %eax, (%edi)
leal _cexpl_zero-"L00000000002$pb"(%ebx), %eax
movl %eax, 4(%edi)
jmp L115
L109:
movl $4, %ecx
movl %eax, %edx
subl 100(%esp), %edx
testl %edx, %edx
jg L76
L108:
movl %eax, %edx
movl 100(%esp), %eax
subl %edx, %eax
cmpl %eax, %edx
jle L110
L79:
orl $1, %ecx
fld %st(0)
pushl %eax
fimull (%esp)
addl $4, %esp
fdiv %st(2), %st
fsincos
fld %st(0)
testb $1, %cl
je L119
L113:
fstp %st(1)
jmp L82
L106:
movl $4, %esi
movl %eax, %ecx
subl 100(%esp), %ecx
testl %ecx, %ecx
jg L55
L104:
movl %eax, %ecx
movl 100(%esp), %edx
subl %ecx, %edx
cmpl %edx, %ecx
jle L107
L58:
orl $1, %esi
fld %st(0)
pushl %edx
fimull (%esp)
addl $4, %esp
fdiv %st(2), %st
fsincos
fld %st(0)
testl $1, %esi
je L120
L111:
fstp %st(1)
jmp L61
L105:
addl 52(%esp), %edx
jmp L50
L103:
xorl %ecx, %ecx
movl $1, 48(%esp)
jmp L46
.section
__IMPORT,__jump_table,symbol_stubs,self_modifying_code+pure_instructions,5
L_fftwf_ifree$stub:
.indirect_symbol _fftwf_ifree
hlt ; hlt ; hlt ; hlt ; hlt
L_fftwf_ifree0$stub:
.indirect_symbol _fftwf_ifree0
hlt ; hlt ; hlt ; hlt ; hlt
L_fftwf_malloc_plain$stub:
.indirect_symbol _fftwf_malloc_plain
hlt ; hlt ; hlt ; hlt ; hlt
.subsections_via_symbols
.section __TEXT,__textcoal_nt,coalesced,pure_instructions
.weak_definition ___i686.get_pc_thunk.bx
.private_extern ___i686.get_pc_thunk.bx
___i686.get_pc_thunk.bx:
movl (%esp), %ebx
ret
----END----
--
Summary: gcc 4.0.1 creates unassembleable code for FFTW library
Product: gcc
Version: 4.0.1
Status: UNCONFIRMED
Severity: major
Priority: P3
Component: c
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: stp at create dot ucsb dot edu
GCC build triplet: powerpc-apple-darwin8
GCC host triplet: i686-apple-darwin8 (Intel Mac)
GCC target triplet: i686-apple-darwin8
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27264
