101 lines
2.6 KiB
ArmAsm
101 lines
2.6 KiB
ArmAsm
#if defined(__SUNPRO_C) && defined(__sparcv9)
|
|
# define ABI64 /* They've said -xarch=v9 at command line */
|
|
#elif defined(__GNUC__) && defined(__arch64__)
|
|
# define ABI64 /* They've said -m64 at command line */
|
|
#endif
|
|
|
|
#ifdef ABI64
|
|
.register %g2,#scratch
|
|
.register %g3,#scratch
|
|
# define FRAME -192
|
|
# define BIAS 2047
|
|
#else
|
|
# define FRAME -96
|
|
# define BIAS 0
|
|
#endif
|
|
|
|
.global _sparcv9_vis1_probe
|
|
.align 8
|
|
_sparcv9_vis1_probe:
|
|
add %sp,BIAS+2,%o1
|
|
.word 0xc19a5a40 !ldda [%o1]ASI_FP16_P,%f0
|
|
retl
|
|
.word 0x81b00d80 !fxor %f0,%f0,%f0
|
|
.type _sparcv9_vis1_probe,#function
|
|
.size _sparcv9_vis1_probe,.-_sparcv9_vis1_probe
|
|
|
|
! Probe and instrument VIS1 instruction. Output is number of cycles it
|
|
! takes to execute rdtick and pair of VIS1 instructions. US-Tx VIS unit
|
|
! is slow (documented to be 6 cycles on T2) and the core is in-order
|
|
! single-issue, it should be possible to distinguish Tx reliably...
|
|
! Observed return values are:
|
|
!
|
|
! UltraSPARC IIe 7
|
|
! UltraSPARC III 7
|
|
! UltraSPARC T1 24
|
|
!
|
|
! Numbers for T2 and SPARC64 V-VII are more than welcomed.
|
|
!
|
|
! It would be possible to detect specifically US-T1 by instrumenting
|
|
! fmul8ulx16, which is emulated on T1 and as such accounts for quite
|
|
! a lot of %tick-s, couple of thousand on Linux...
|
|
.global _sparcv9_vis1_instrument
|
|
.align 8
|
|
_sparcv9_vis1_instrument:
|
|
.word 0x91410000 !rd %tick,%o0
|
|
.word 0x81b00d80 !fxor %f0,%f0,%f0
|
|
.word 0x85b08d82 !fxor %f2,%f2,%f2
|
|
.word 0x93410000 !rd %tick,%o1
|
|
.word 0x81b00d80 !fxor %f0,%f0,%f0
|
|
.word 0x85b08d82 !fxor %f2,%f2,%f2
|
|
.word 0x95410000 !rd %tick,%o2
|
|
.word 0x81b00d80 !fxor %f0,%f0,%f0
|
|
.word 0x85b08d82 !fxor %f2,%f2,%f2
|
|
.word 0x97410000 !rd %tick,%o3
|
|
.word 0x81b00d80 !fxor %f0,%f0,%f0
|
|
.word 0x85b08d82 !fxor %f2,%f2,%f2
|
|
.word 0x99410000 !rd %tick,%o4
|
|
|
|
! calculate intervals
|
|
sub %o1,%o0,%o0
|
|
sub %o2,%o1,%o1
|
|
sub %o3,%o2,%o2
|
|
sub %o4,%o3,%o3
|
|
|
|
! find minimum value
|
|
cmp %o0,%o1
|
|
.word 0x38680002 !bgu,a %xcc,.+8
|
|
mov %o1,%o0
|
|
cmp %o0,%o2
|
|
.word 0x38680002 !bgu,a %xcc,.+8
|
|
mov %o2,%o0
|
|
cmp %o0,%o3
|
|
.word 0x38680002 !bgu,a %xcc,.+8
|
|
mov %o3,%o0
|
|
|
|
retl
|
|
nop
|
|
.type _sparcv9_vis1_instrument,#function
|
|
.size _sparcv9_vis1_instrument,.-_sparcv9_vis1_instrument
|
|
|
|
.global _sparcv9_vis2_probe
|
|
.align 8
|
|
_sparcv9_vis2_probe:
|
|
retl
|
|
.word 0x81b00980 !bshuffle %f0,%f0,%f0
|
|
.type _sparcv9_vis2_probe,#function
|
|
.size _sparcv9_vis2_probe,.-_sparcv9_vis2_probe
|
|
|
|
.global _sparcv9_fmadd_probe
|
|
.align 8
|
|
_sparcv9_fmadd_probe:
|
|
.word 0x81b00d80 !fxor %f0,%f0,%f0
|
|
.word 0x85b08d82 !fxor %f2,%f2,%f2
|
|
retl
|
|
.word 0x81b80440 !fmaddd %f0,%f0,%f2,%f0
|
|
.type _sparcv9_fmadd_probe,#function
|
|
.size _sparcv9_fmadd_probe,.-_sparcv9_fmadd_probe
|
|
|
|
.section ".init",#alloc,#execinstr
|
|
call OPENSSL_cpuid_setup
|
|
nop
|