On OpenBSD 7.5/i386, the two tests still fail, except when compiling
without optimization (e.g. with CFLAGS="-ggdb").
So, it's a compiler optimization bug, as can be seen by the code
generated by the OpenBSD compiler (a clang 16.0.6 derivative).
Cf. the attached totalorderf.O0.s, totalorderf.O2.s.
The patch below avoids the compiler bug, cf. the attached
totalorderf.O2-fixed.s.
2024-04-09 Bruno Haible <br...@clisp.org>
totalorder, totalorderf: Avoid miscompilation by clang on OpenBSD/i386.
* lib/totalorder.c (totalorder): Insert a compiler optimization barrier.
* lib/totalorderf.c (totalorderf): Likewise.
diff --git a/lib/totalorder.c b/lib/totalorder.c
index 635e3cb276..eaef9a582d 100644
--- a/lib/totalorder.c
+++ b/lib/totalorder.c
@@ -57,6 +57,11 @@ totalorder (double const *x, double const *y)
xu.f = *x;
yu.f = *y;
#else
+# ifdef __clang__
+ /* Prevent clang 16.0.6 on OpenBSD 7.5 from reusing the values of *x and *y
+ (fetched above) in optimized inlined memcpy expansions. */
+ __asm__ __volatile__ ("" : : : "memory");
+# endif
/* On 32-bit x86 processors, as well as on x86_64 processors with
CC="gcc -mfpmath=387", the evaluation of *x and *y above is done through
an 'fldl' instruction, which converts a signalling NaN to a quiet NaN. See
diff --git a/lib/totalorderf.c b/lib/totalorderf.c
index 75024b6839..f5f1b40112 100644
--- a/lib/totalorderf.c
+++ b/lib/totalorderf.c
@@ -57,6 +57,11 @@ totalorderf (float const *x, float const *y)
xu.f = *x;
yu.f = *y;
#else
+# ifdef __clang__
+ /* Prevent clang 16.0.6 on OpenBSD 7.5 from reusing the values of *x and *y
+ (fetched above) in optimized inlined memcpy expansions. */
+ __asm__ __volatile__ ("" : : : "memory");
+# endif
/* On 32-bit x86 processors, as well as on x86_64 processors with
CC="gcc -mfpmath=387", the evaluation of *x and *y above is done through
an 'flds' instruction, which converts a signalling NaN to a quiet NaN. See
.text
.file "totalorderf.c"
.globl totalorderf # -- Begin function totalorderf
.p2align 4, 0xcc
.type totalorderf,@function
totalorderf: # @totalorderf
.cfi_startproc
# %bb.0:
pushl %ebp
.cfi_def_cfa_offset 8
.cfi_offset %ebp, -8
movl %esp, %ebp
.cfi_def_cfa_register %ebp
pushl %ebx
subl $60, %esp
.cfi_offset %ebx, -12
calll .L0$pb
.L0$pb:
popl %eax
.Ltmp0:
addl $_GLOBAL_OFFSET_TABLE_+(.Ltmp0-.L0$pb), %eax
movl %eax, -48(%ebp) # 4-byte Spill
movl 12(%ebp), %ecx
movl 8(%ebp), %ecx
movl __guard_local@GOTOFF(%eax), %eax
movl %eax, -8(%ebp)
movl 8(%ebp), %eax
movl (%eax), %eax
shrl $31, %eax
movl %eax, -24(%ebp)
movl 12(%ebp), %eax
flds (%eax)
fstps -44(%ebp)
movl -44(%ebp), %eax
cmpl $0, %eax
setl %al
andb $1, %al
movzbl %al, %eax
movl %eax, -28(%ebp)
cmpl $0, -24(%ebp)
setne %al
xorb $-1, %al
andb $1, %al
movzbl %al, %eax
cmpl $0, -28(%ebp)
setne %cl
xorb $-1, %cl
andb $1, %cl
movzbl %cl, %ecx
cmpl %ecx, %eax
je .LBB0_2
# %bb.1:
movl -24(%ebp), %eax
movl %eax, -20(%ebp)
jmp .LBB0_7
.LBB0_2:
movl 8(%ebp), %eax
flds (%eax)
fucomp %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setp %al
movzbl %al, %eax
movl %eax, -32(%ebp)
movl 12(%ebp), %eax
flds (%eax)
fucomp %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setp %al
andb $1, %al
movzbl %al, %eax
movl %eax, -36(%ebp)
cmpl $0, -32(%ebp)
setne %al
xorb $-1, %al
andb $1, %al
movzbl %al, %eax
cmpl $0, -36(%ebp)
setne %cl
xorb $-1, %cl
andb $1, %cl
movzbl %cl, %ecx
cmpl %ecx, %eax
je .LBB0_4
# %bb.3:
cmpl $0, -32(%ebp)
setne %al
xorb $-1, %al
andb $1, %al
movzbl %al, %eax
cmpl $0, -24(%ebp)
setne %cl
xorb $-1, %cl
andb $1, %cl
movzbl %cl, %ecx
cmpl %ecx, %eax
sete %al
andb $1, %al
movzbl %al, %eax
movl %eax, -20(%ebp)
jmp .LBB0_7
.LBB0_4:
cmpl $0, -32(%ebp)
jne .LBB0_6
# %bb.5:
movl 8(%ebp), %eax
flds (%eax)
movl 12(%ebp), %eax
flds (%eax)
fucompp
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setae %al
andb $1, %al
movzbl %al, %eax
movl %eax, -20(%ebp)
jmp .LBB0_7
.LBB0_6:
movl -48(%ebp), %ebx # 4-byte Reload
cmpl $0, -24(%ebp)
setne %al
xorb $-1, %al
xorb $-1, %al
andb $1, %al
movzbl %al, %ecx
xorl %eax, %eax
subl %ecx, %eax
movl %eax, -40(%ebp)
leal -12(%ebp), %eax
xorl %ecx, %ecx
movl %eax, (%esp)
movl $0, 4(%esp)
movl $4, 8(%esp)
calll memset@PLT
movl -48(%ebp), %ebx # 4-byte Reload
leal -16(%ebp), %eax
xorl %ecx, %ecx
movl %eax, (%esp)
movl $0, 4(%esp)
movl $4, 8(%esp)
calll memset@PLT
movl 8(%ebp), %eax
movl (%eax), %eax
movl %eax, -12(%ebp)
movl 12(%ebp), %eax
movl (%eax), %eax
movl %eax, -16(%ebp)
movl -12(%ebp), %eax
xorl -40(%ebp), %eax
movl -16(%ebp), %ecx
xorl -40(%ebp), %ecx
cmpl %ecx, %eax
setbe %al
andb $1, %al
movzbl %al, %eax
movl %eax, -20(%ebp)
.LBB0_7:
movl -48(%ebp), %eax # 4-byte Reload
movl -20(%ebp), %ecx
movl %ecx, -52(%ebp) # 4-byte Spill
movl __guard_local@GOTOFF(%eax), %eax
movl -8(%ebp), %ecx
cmpl %ecx, %eax
jne .LBB0_9
# %bb.8:
movl -52(%ebp), %eax # 4-byte Reload
addl $60, %esp
popl %ebx
popl %ebp
.cfi_def_cfa %esp, 4
retl
.LBB0_9:
.cfi_def_cfa %ebp, 8
movl -48(%ebp), %ebx # 4-byte Reload
leal .LSSH@GOTOFF(%ebx), %eax
movl %eax, (%esp)
calll __stack_smash_handler@PLT
.Lfunc_end0:
.size totalorderf, .Lfunc_end0-totalorderf
.cfi_endproc
# -- End function
.hidden __guard_local
.type .LSSH,@object # @SSH
.section .rodata.str1.1,"aMS",@progbits,1
.LSSH:
.asciz "totalorderf"
.size .LSSH, 12
.section ".note.GNU-stack","",@progbits
.addrsig
.addrsig_sym __stack_smash_handler
.addrsig_sym __guard_local
.text
.file "totalorderf.c"
.globl totalorderf # -- Begin function totalorderf
.p2align 4, 0xcc
.type totalorderf,@function
totalorderf: # @totalorderf
.cfi_startproc
# %bb.0:
pushl %ebp
.cfi_def_cfa_offset 8
.cfi_offset %ebp, -8
movl %esp, %ebp
.cfi_def_cfa_register %ebp
pushl %ebx
subl $8, %esp
.cfi_offset %ebx, -12
movl 12(%ebp), %eax
movl 8(%ebp), %ecx
flds (%ecx)
fsts -8(%ebp)
movl -8(%ebp), %ecx
flds (%eax)
fsts -12(%ebp)
movl -12(%ebp), %edx
movl %edx, %eax
xorl %ecx, %eax
js .LBB0_1
# %bb.2:
fxch %st(1)
fucom %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setnp %bl
fxch %st(1)
fucom %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setp %al
xorb %bl, %al
je .LBB0_3
# %bb.4:
fxch %st(1)
fucom %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
jp .LBB0_6
# %bb.5:
fxch %st(1)
fucompp
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setae %al
jmp .LBB0_7
.LBB0_1:
fstp %st(1)
fstp %st(0)
shrl $31, %ecx
jmp .LBB0_8
.LBB0_3:
fstp %st(0)
fucomp %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setnp %dl
testl %ecx, %ecx
sets %al
xorb %dl, %al
.LBB0_7:
movzbl %al, %ecx
.LBB0_8:
movl %ecx, %eax
addl $8, %esp
popl %ebx
popl %ebp
.cfi_def_cfa %esp, 4
retl
.LBB0_6:
.cfi_def_cfa %ebp, 8
fstp %st(0)
fstp %st(0)
movl %ecx, %eax
sarl $31, %eax
xorl %eax, %ecx
xorl %edx, %eax
cmpl %eax, %ecx
setbe %al
jmp .LBB0_7
.Lfunc_end0:
.size totalorderf, .Lfunc_end0-totalorderf
.cfi_endproc
# -- End function
.section ".note.GNU-stack","",@progbits
.addrsig
.text
.file "totalorderf.c"
.globl totalorderf # -- Begin function totalorderf
.p2align 4, 0xcc
.type totalorderf,@function
totalorderf: # @totalorderf
.cfi_startproc
# %bb.0:
pushl %ebp
.cfi_def_cfa_offset 8
.cfi_offset %ebp, -8
movl %esp, %ebp
.cfi_def_cfa_register %ebp
pushl %ebx
pushl %esi
subl $8, %esp
.cfi_offset %esi, -16
.cfi_offset %ebx, -12
movl 12(%ebp), %edx
movl 8(%ebp), %esi
flds (%esi)
fsts -12(%ebp)
movl -12(%ebp), %ecx
flds (%edx)
fsts -16(%ebp)
movl -16(%ebp), %eax
xorl %ecx, %eax
js .LBB0_1
# %bb.2:
fxch %st(1)
fucom %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setnp %bl
fxch %st(1)
fucom %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setp %al
xorb %bl, %al
je .LBB0_3
# %bb.4:
fxch %st(1)
fucom %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
jp .LBB0_6
# %bb.5:
fxch %st(1)
fucompp
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setae %al
jmp .LBB0_7
.LBB0_1:
fstp %st(1)
fstp %st(0)
shrl $31, %ecx
jmp .LBB0_8
.LBB0_3:
fstp %st(0)
fucomp %st(0)
fnstsw %ax
# kill: def $ah killed $ah killed $ax
sahf
setnp %dl
testl %ecx, %ecx
sets %al
xorb %dl, %al
.LBB0_7:
movzbl %al, %ecx
.LBB0_8:
movl %ecx, %eax
addl $8, %esp
popl %esi
popl %ebx
popl %ebp
.cfi_def_cfa %esp, 4
retl
.LBB0_6:
.cfi_def_cfa %ebp, 8
fstp %st(0)
fstp %st(0)
sarl $31, %ecx
#APP
#NO_APP
movl (%esi), %eax
xorl %ecx, %eax
xorl (%edx), %ecx
cmpl %ecx, %eax
setbe %al
jmp .LBB0_7
.Lfunc_end0:
.size totalorderf, .Lfunc_end0-totalorderf
.cfi_endproc
# -- End function
.section ".note.GNU-stack","",@progbits
.addrsig
