549 lines
33 KiB
ArmAsm
549 lines
33 KiB
ArmAsm
|
.text
|
||
|
|
.file "vecadd.cu"
|
||
|
|
.globl _Z21__device_stub__vecAddPdS_S_i # -- Begin function _Z21__device_stub__vecAddPdS_S_i
|
||
|
|
.p2align 4, 0x90
|
||
|
|
.type _Z21__device_stub__vecAddPdS_S_i,@function
|
||
|
|
_Z21__device_stub__vecAddPdS_S_i: # @_Z21__device_stub__vecAddPdS_S_i
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rbp
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
.cfi_offset %rbp, -16
|
||
|
|
movq %rsp, %rbp
|
||
|
|
.cfi_def_cfa_register %rbp
|
||
|
|
subq $160, %rsp
|
||
|
|
movq %rdi, -8(%rbp)
|
||
|
|
movq %rsi, -16(%rbp)
|
||
|
|
movq %rdx, -24(%rbp)
|
||
|
|
movl %ecx, -28(%rbp)
|
||
|
|
leaq -8(%rbp), %rax
|
||
|
|
movq %rax, -144(%rbp)
|
||
|
|
leaq -16(%rbp), %rax
|
||
|
|
movq %rax, -136(%rbp)
|
||
|
|
leaq -24(%rbp), %rax
|
||
|
|
movq %rax, -128(%rbp)
|
||
|
|
leaq -28(%rbp), %rax
|
||
|
|
movq %rax, -120(%rbp)
|
||
|
|
leaq -40(%rbp), %rdi
|
||
|
|
leaq -56(%rbp), %rsi
|
||
|
|
leaq -64(%rbp), %rdx
|
||
|
|
leaq -72(%rbp), %rcx
|
||
|
|
callq __cudaPopCallConfiguration@PLT
|
||
|
|
movq -64(%rbp), %r10
|
||
|
|
movq -72(%rbp), %rax
|
||
|
|
movq -40(%rbp), %rcx
|
||
|
|
movq %rcx, -88(%rbp)
|
||
|
|
movl -32(%rbp), %ecx
|
||
|
|
movl %ecx, -80(%rbp)
|
||
|
|
movq -88(%rbp), %rsi
|
||
|
|
movl -80(%rbp), %edx
|
||
|
|
movq -56(%rbp), %rcx
|
||
|
|
movq %rcx, -104(%rbp)
|
||
|
|
movl -48(%rbp), %ecx
|
||
|
|
movl %ecx, -96(%rbp)
|
||
|
|
movq -104(%rbp), %rcx
|
||
|
|
movl -96(%rbp), %r8d
|
||
|
|
leaq _Z21__device_stub__vecAddPdS_S_i(%rip), %rdi
|
||
|
|
leaq -144(%rbp), %r9
|
||
|
|
movq %r10, (%rsp)
|
||
|
|
movq %rax, 8(%rsp)
|
||
|
|
callq cudaLaunchKernel@PLT
|
||
|
|
# %bb.1:
|
||
|
|
addq $160, %rsp
|
||
|
|
popq %rbp
|
||
|
|
.cfi_def_cfa %rsp, 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end0:
|
||
|
|
.size _Z21__device_stub__vecAddPdS_S_i, .Lfunc_end0-_Z21__device_stub__vecAddPdS_S_i
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.section .rodata.cst8,"aM",@progbits,8
|
||
|
|
.p2align 3 # -- Begin function main
|
||
|
|
.LCPI1_0:
|
||
|
|
.quad 0x3eb0c6f7a0b5ed8d # double 9.9999999999999995E-7
|
||
|
|
.LCPI1_1:
|
||
|
|
.quad 0x3ff0000000000000 # double 1
|
||
|
|
.text
|
||
|
|
.globl main
|
||
|
|
.p2align 4, 0x90
|
||
|
|
.type main,@function
|
||
|
|
main: # @main
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rbp
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
.cfi_offset %rbp, -16
|
||
|
|
movq %rsp, %rbp
|
||
|
|
.cfi_def_cfa_register %rbp
|
||
|
|
subq $176, %rsp
|
||
|
|
movl $0, -4(%rbp)
|
||
|
|
movl %edi, -8(%rbp)
|
||
|
|
movq %rsi, -16(%rbp)
|
||
|
|
movl $100000, -20(%rbp) # imm = 0x186A0
|
||
|
|
movslq -20(%rbp), %rax
|
||
|
|
shlq $3, %rax
|
||
|
|
movq %rax, -80(%rbp)
|
||
|
|
movq -80(%rbp), %rdi
|
||
|
|
callq malloc@PLT
|
||
|
|
movq %rax, -32(%rbp)
|
||
|
|
movq -80(%rbp), %rdi
|
||
|
|
callq malloc@PLT
|
||
|
|
movq %rax, -40(%rbp)
|
||
|
|
movq -80(%rbp), %rdi
|
||
|
|
callq malloc@PLT
|
||
|
|
movq %rax, -48(%rbp)
|
||
|
|
movq -80(%rbp), %rsi
|
||
|
|
leaq -56(%rbp), %rdi
|
||
|
|
callq _ZL10cudaMallocIdE9cudaErrorPPT_m
|
||
|
|
movq -80(%rbp), %rsi
|
||
|
|
leaq -64(%rbp), %rdi
|
||
|
|
callq _ZL10cudaMallocIdE9cudaErrorPPT_m
|
||
|
|
movq -80(%rbp), %rsi
|
||
|
|
leaq -72(%rbp), %rdi
|
||
|
|
callq _ZL10cudaMallocIdE9cudaErrorPPT_m
|
||
|
|
movl $0, -84(%rbp)
|
||
|
|
.LBB1_1: # =>This Inner Loop Header: Depth=1
|
||
|
|
movl -84(%rbp), %eax
|
||
|
|
cmpl -20(%rbp), %eax
|
||
|
|
jge .LBB1_4
|
||
|
|
# %bb.2: # in Loop: Header=BB1_1 Depth=1
|
||
|
|
movl -84(%rbp), %edi
|
||
|
|
callq _ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
movsd %xmm0, -176(%rbp) # 8-byte Spill
|
||
|
|
movl -84(%rbp), %edi
|
||
|
|
callq _ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
movaps %xmm0, %xmm1
|
||
|
|
movsd -176(%rbp), %xmm0 # 8-byte Reload
|
||
|
|
# xmm0 = mem[0],zero
|
||
|
|
mulsd %xmm1, %xmm0
|
||
|
|
movq -32(%rbp), %rax
|
||
|
|
movslq -84(%rbp), %rcx
|
||
|
|
movsd %xmm0, (%rax,%rcx,8)
|
||
|
|
movl -84(%rbp), %edi
|
||
|
|
callq _ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
movsd %xmm0, -168(%rbp) # 8-byte Spill
|
||
|
|
movl -84(%rbp), %edi
|
||
|
|
callq _ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
movaps %xmm0, %xmm1
|
||
|
|
movsd -168(%rbp), %xmm0 # 8-byte Reload
|
||
|
|
# xmm0 = mem[0],zero
|
||
|
|
mulsd %xmm1, %xmm0
|
||
|
|
movq -40(%rbp), %rax
|
||
|
|
movslq -84(%rbp), %rcx
|
||
|
|
movsd %xmm0, (%rax,%rcx,8)
|
||
|
|
# %bb.3: # in Loop: Header=BB1_1 Depth=1
|
||
|
|
movl -84(%rbp), %eax
|
||
|
|
addl $1, %eax
|
||
|
|
movl %eax, -84(%rbp)
|
||
|
|
jmp .LBB1_1
|
||
|
|
.LBB1_4:
|
||
|
|
movq -56(%rbp), %rdi
|
||
|
|
movq -32(%rbp), %rsi
|
||
|
|
movq -80(%rbp), %rdx
|
||
|
|
movl $1, %ecx
|
||
|
|
callq cudaMemcpy@PLT
|
||
|
|
movq -64(%rbp), %rdi
|
||
|
|
movq -40(%rbp), %rsi
|
||
|
|
movq -80(%rbp), %rdx
|
||
|
|
movl $1, %ecx
|
||
|
|
callq cudaMemcpy@PLT
|
||
|
|
movl $1024, -88(%rbp) # imm = 0x400
|
||
|
|
cvtsi2ssl -20(%rbp), %xmm0
|
||
|
|
cvtsi2ssl -88(%rbp), %xmm1
|
||
|
|
divss %xmm1, %xmm0
|
||
|
|
callq _ZSt4ceilf
|
||
|
|
cvttss2si %xmm0, %eax
|
||
|
|
movl %eax, -92(%rbp)
|
||
|
|
movl -92(%rbp), %esi
|
||
|
|
leaq -104(%rbp), %rdi
|
||
|
|
movl $1, %ecx
|
||
|
|
movl %ecx, %edx
|
||
|
|
callq _ZN4dim3C2Ejjj
|
||
|
|
movl -88(%rbp), %esi
|
||
|
|
leaq -120(%rbp), %rdi
|
||
|
|
movl $1, %ecx
|
||
|
|
movl %ecx, %edx
|
||
|
|
callq _ZN4dim3C2Ejjj
|
||
|
|
movq -104(%rbp), %rax
|
||
|
|
movq %rax, -136(%rbp)
|
||
|
|
movl -96(%rbp), %eax
|
||
|
|
movl %eax, -128(%rbp)
|
||
|
|
movq -136(%rbp), %rdi
|
||
|
|
movl -128(%rbp), %esi
|
||
|
|
movq -120(%rbp), %rax
|
||
|
|
movq %rax, -152(%rbp)
|
||
|
|
movl -112(%rbp), %eax
|
||
|
|
movl %eax, -144(%rbp)
|
||
|
|
movq -152(%rbp), %rdx
|
||
|
|
movl -144(%rbp), %ecx
|
||
|
|
xorl %eax, %eax
|
||
|
|
movl %eax, %r9d
|
||
|
|
movq %r9, %r8
|
||
|
|
callq __cudaPushCallConfiguration@PLT
|
||
|
|
cmpl $0, %eax
|
||
|
|
jne .LBB1_6
|
||
|
|
# %bb.5:
|
||
|
|
movq -56(%rbp), %rdi
|
||
|
|
movq -64(%rbp), %rsi
|
||
|
|
movq -72(%rbp), %rdx
|
||
|
|
movl -20(%rbp), %ecx
|
||
|
|
callq _Z21__device_stub__vecAddPdS_S_i
|
||
|
|
.LBB1_6:
|
||
|
|
movq -48(%rbp), %rdi
|
||
|
|
movq -72(%rbp), %rsi
|
||
|
|
movq -80(%rbp), %rdx
|
||
|
|
movl $2, %ecx
|
||
|
|
callq cudaMemcpy@PLT
|
||
|
|
xorps %xmm0, %xmm0
|
||
|
|
movsd %xmm0, -160(%rbp)
|
||
|
|
movl $0, -84(%rbp)
|
||
|
|
.LBB1_7: # =>This Inner Loop Header: Depth=1
|
||
|
|
movl -84(%rbp), %eax
|
||
|
|
cmpl -20(%rbp), %eax
|
||
|
|
jge .LBB1_10
|
||
|
|
# %bb.8: # in Loop: Header=BB1_7 Depth=1
|
||
|
|
movq -48(%rbp), %rax
|
||
|
|
movslq -84(%rbp), %rcx
|
||
|
|
movsd (%rax,%rcx,8), %xmm0 # xmm0 = mem[0],zero
|
||
|
|
addsd -160(%rbp), %xmm0
|
||
|
|
movsd %xmm0, -160(%rbp)
|
||
|
|
# %bb.9: # in Loop: Header=BB1_7 Depth=1
|
||
|
|
movl -84(%rbp), %eax
|
||
|
|
addl $1, %eax
|
||
|
|
movl %eax, -84(%rbp)
|
||
|
|
jmp .LBB1_7
|
||
|
|
.LBB1_10:
|
||
|
|
cvtsi2sdl -20(%rbp), %xmm1
|
||
|
|
movsd -160(%rbp), %xmm0 # xmm0 = mem[0],zero
|
||
|
|
divsd %xmm1, %xmm0
|
||
|
|
movsd %xmm0, -160(%rbp)
|
||
|
|
movsd -160(%rbp), %xmm0 # xmm0 = mem[0],zero
|
||
|
|
movsd .LCPI1_1(%rip), %xmm1 # xmm1 = mem[0],zero
|
||
|
|
subsd %xmm1, %xmm0
|
||
|
|
callq _ZSt3absd
|
||
|
|
movaps %xmm0, %xmm1
|
||
|
|
movsd .LCPI1_0(%rip), %xmm0 # xmm0 = mem[0],zero
|
||
|
|
ucomisd %xmm1, %xmm0
|
||
|
|
jbe .LBB1_12
|
||
|
|
# %bb.11:
|
||
|
|
leaq .L.str(%rip), %rdi
|
||
|
|
movb $0, %al
|
||
|
|
callq printf@PLT
|
||
|
|
jmp .LBB1_13
|
||
|
|
.LBB1_12:
|
||
|
|
leaq .L.str.1(%rip), %rdi
|
||
|
|
movb $0, %al
|
||
|
|
callq printf@PLT
|
||
|
|
.LBB1_13:
|
||
|
|
movq -56(%rbp), %rdi
|
||
|
|
callq cudaFree@PLT
|
||
|
|
movq -64(%rbp), %rdi
|
||
|
|
callq cudaFree@PLT
|
||
|
|
movq -72(%rbp), %rdi
|
||
|
|
callq cudaFree@PLT
|
||
|
|
movq -32(%rbp), %rdi
|
||
|
|
callq free@PLT
|
||
|
|
movq -40(%rbp), %rdi
|
||
|
|
callq free@PLT
|
||
|
|
movq -48(%rbp), %rdi
|
||
|
|
callq free@PLT
|
||
|
|
xorl %eax, %eax
|
||
|
|
addq $176, %rsp
|
||
|
|
popq %rbp
|
||
|
|
.cfi_def_cfa %rsp, 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end1:
|
||
|
|
.size main, .Lfunc_end1-main
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.p2align 4, 0x90 # -- Begin function _ZL10cudaMallocIdE9cudaErrorPPT_m
|
||
|
|
.type _ZL10cudaMallocIdE9cudaErrorPPT_m,@function
|
||
|
|
_ZL10cudaMallocIdE9cudaErrorPPT_m: # @_ZL10cudaMallocIdE9cudaErrorPPT_m
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rbp
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
.cfi_offset %rbp, -16
|
||
|
|
movq %rsp, %rbp
|
||
|
|
.cfi_def_cfa_register %rbp
|
||
|
|
subq $16, %rsp
|
||
|
|
movq %rdi, -8(%rbp)
|
||
|
|
movq %rsi, -16(%rbp)
|
||
|
|
movq -8(%rbp), %rdi
|
||
|
|
movq -16(%rbp), %rsi
|
||
|
|
callq cudaMalloc@PLT
|
||
|
|
addq $16, %rsp
|
||
|
|
popq %rbp
|
||
|
|
.cfi_def_cfa %rsp, 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end2:
|
||
|
|
.size _ZL10cudaMallocIdE9cudaErrorPPT_m, .Lfunc_end2-_ZL10cudaMallocIdE9cudaErrorPPT_m
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.section .text._ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_,"axG",@progbits,_ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_,comdat
|
||
|
|
.weak _ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_ # -- Begin function _ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
.p2align 4, 0x90
|
||
|
|
.type _ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_,@function
|
||
|
|
_ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_: # @_ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rbp
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
.cfi_offset %rbp, -16
|
||
|
|
movq %rsp, %rbp
|
||
|
|
.cfi_def_cfa_register %rbp
|
||
|
|
subq $16, %rsp
|
||
|
|
movl %edi, -4(%rbp)
|
||
|
|
cvtsi2sdl -4(%rbp), %xmm0
|
||
|
|
callq sin@PLT
|
||
|
|
addq $16, %rsp
|
||
|
|
popq %rbp
|
||
|
|
.cfi_def_cfa %rsp, 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end3:
|
||
|
|
.size _ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_, .Lfunc_end3-_ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.section .text._ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_,"axG",@progbits,_ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_,comdat
|
||
|
|
.weak _ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_ # -- Begin function _ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
.p2align 4, 0x90
|
||
|
|
.type _ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_,@function
|
||
|
|
_ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_: # @_ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rbp
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
.cfi_offset %rbp, -16
|
||
|
|
movq %rsp, %rbp
|
||
|
|
.cfi_def_cfa_register %rbp
|
||
|
|
subq $16, %rsp
|
||
|
|
movl %edi, -4(%rbp)
|
||
|
|
cvtsi2sdl -4(%rbp), %xmm0
|
||
|
|
callq cos@PLT
|
||
|
|
addq $16, %rsp
|
||
|
|
popq %rbp
|
||
|
|
.cfi_def_cfa %rsp, 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end4:
|
||
|
|
.size _ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_, .Lfunc_end4-_ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.section .text._ZSt4ceilf,"axG",@progbits,_ZSt4ceilf,comdat
|
||
|
|
.weak _ZSt4ceilf # -- Begin function _ZSt4ceilf
|
||
|
|
.p2align 4, 0x90
|
||
|
|
.type _ZSt4ceilf,@function
|
||
|
|
_ZSt4ceilf: # @_ZSt4ceilf
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rbp
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
.cfi_offset %rbp, -16
|
||
|
|
movq %rsp, %rbp
|
||
|
|
.cfi_def_cfa_register %rbp
|
||
|
|
subq $16, %rsp
|
||
|
|
movss %xmm0, -4(%rbp)
|
||
|
|
movss -4(%rbp), %xmm0 # xmm0 = mem[0],zero,zero,zero
|
||
|
|
callq ceilf@PLT
|
||
|
|
addq $16, %rsp
|
||
|
|
popq %rbp
|
||
|
|
.cfi_def_cfa %rsp, 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end5:
|
||
|
|
.size _ZSt4ceilf, .Lfunc_end5-_ZSt4ceilf
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.section .text._ZN4dim3C2Ejjj,"axG",@progbits,_ZN4dim3C2Ejjj,comdat
|
||
|
|
.weak _ZN4dim3C2Ejjj # -- Begin function _ZN4dim3C2Ejjj
|
||
|
|
.p2align 4, 0x90
|
||
|
|
.type _ZN4dim3C2Ejjj,@function
|
||
|
|
_ZN4dim3C2Ejjj: # @_ZN4dim3C2Ejjj
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rbp
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
.cfi_offset %rbp, -16
|
||
|
|
movq %rsp, %rbp
|
||
|
|
.cfi_def_cfa_register %rbp
|
||
|
|
movq %rdi, -8(%rbp)
|
||
|
|
movl %esi, -12(%rbp)
|
||
|
|
movl %edx, -16(%rbp)
|
||
|
|
movl %ecx, -20(%rbp)
|
||
|
|
movq -8(%rbp), %rax
|
||
|
|
movl -12(%rbp), %ecx
|
||
|
|
movl %ecx, (%rax)
|
||
|
|
movl -16(%rbp), %ecx
|
||
|
|
movl %ecx, 4(%rax)
|
||
|
|
movl -20(%rbp), %ecx
|
||
|
|
movl %ecx, 8(%rax)
|
||
|
|
popq %rbp
|
||
|
|
.cfi_def_cfa %rsp, 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end6:
|
||
|
|
.size _ZN4dim3C2Ejjj, .Lfunc_end6-_ZN4dim3C2Ejjj
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.section .rodata.cst16,"aM",@progbits,16
|
||
|
|
.p2align 4 # -- Begin function _ZSt3absd
|
||
|
|
.LCPI7_0:
|
||
|
|
.quad 0x7fffffffffffffff # double NaN
|
||
|
|
.quad 0x7fffffffffffffff # double NaN
|
||
|
|
.section .text._ZSt3absd,"axG",@progbits,_ZSt3absd,comdat
|
||
|
|
.weak _ZSt3absd
|
||
|
|
.p2align 4, 0x90
|
||
|
|
.type _ZSt3absd,@function
|
||
|
|
_ZSt3absd: # @_ZSt3absd
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rbp
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
.cfi_offset %rbp, -16
|
||
|
|
movq %rsp, %rbp
|
||
|
|
.cfi_def_cfa_register %rbp
|
||
|
|
movsd %xmm0, -8(%rbp)
|
||
|
|
movsd -8(%rbp), %xmm0 # xmm0 = mem[0],zero
|
||
|
|
movaps .LCPI7_0(%rip), %xmm1 # xmm1 = [NaN,NaN]
|
||
|
|
pand %xmm1, %xmm0
|
||
|
|
popq %rbp
|
||
|
|
.cfi_def_cfa %rsp, 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end7:
|
||
|
|
.size _ZSt3absd, .Lfunc_end7-_ZSt3absd
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.text
|
||
|
|
.p2align 4, 0x90 # -- Begin function __cuda_register_globals
|
||
|
|
.type __cuda_register_globals,@function
|
||
|
|
__cuda_register_globals: # @__cuda_register_globals
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
subq $40, %rsp
|
||
|
|
.cfi_def_cfa_offset 48
|
||
|
|
leaq _Z21__device_stub__vecAddPdS_S_i(%rip), %rsi
|
||
|
|
leaq .L__unnamed_1(%rip), %rcx
|
||
|
|
movl $4294967295, %r8d # imm = 0xFFFFFFFF
|
||
|
|
xorl %eax, %eax
|
||
|
|
movl %eax, %r9d
|
||
|
|
movq %rcx, %rdx
|
||
|
|
movq $0, (%rsp)
|
||
|
|
movq $0, 8(%rsp)
|
||
|
|
movq $0, 16(%rsp)
|
||
|
|
movq $0, 24(%rsp)
|
||
|
|
callq __cudaRegisterFunction@PLT
|
||
|
|
addq $40, %rsp
|
||
|
|
.cfi_def_cfa_offset 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end8:
|
||
|
|
.size __cuda_register_globals, .Lfunc_end8-__cuda_register_globals
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.p2align 4, 0x90 # -- Begin function __cuda_module_ctor
|
||
|
|
.type __cuda_module_ctor,@function
|
||
|
|
__cuda_module_ctor: # @__cuda_module_ctor
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rax
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
leaq __cuda_fatbin_wrapper(%rip), %rdi
|
||
|
|
callq __cudaRegisterFatBinary@PLT
|
||
|
|
movq %rax, %rdi
|
||
|
|
movq %rdi, (%rsp) # 8-byte Spill
|
||
|
|
movq %rdi, __cuda_gpubin_handle(%rip)
|
||
|
|
callq __cuda_register_globals
|
||
|
|
movq (%rsp), %rdi # 8-byte Reload
|
||
|
|
callq __cudaRegisterFatBinaryEnd@PLT
|
||
|
|
leaq __cuda_module_dtor(%rip), %rdi
|
||
|
|
callq atexit@PLT
|
||
|
|
popq %rax
|
||
|
|
.cfi_def_cfa_offset 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end9:
|
||
|
|
.size __cuda_module_ctor, .Lfunc_end9-__cuda_module_ctor
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.p2align 4, 0x90 # -- Begin function __cuda_module_dtor
|
||
|
|
.type __cuda_module_dtor,@function
|
||
|
|
__cuda_module_dtor: # @__cuda_module_dtor
|
||
|
|
.cfi_startproc
|
||
|
|
# %bb.0:
|
||
|
|
pushq %rax
|
||
|
|
.cfi_def_cfa_offset 16
|
||
|
|
movq __cuda_gpubin_handle(%rip), %rdi
|
||
|
|
callq __cudaUnregisterFatBinary@PLT
|
||
|
|
popq %rax
|
||
|
|
.cfi_def_cfa_offset 8
|
||
|
|
retq
|
||
|
|
.Lfunc_end10:
|
||
|
|
.size __cuda_module_dtor, .Lfunc_end10-__cuda_module_dtor
|
||
|
|
.cfi_endproc
|
||
|
|
# -- End function
|
||
|
|
.type .L.str,@object # @.str
|
||
|
|
.section .rodata.str1.1,"aMS",@progbits,1
|
||
|
|
.L.str:
|
||
|
|
.asciz "PASS\n"
|
||
|
|
.size .L.str, 6
|
||
|
|
|
||
|
|
.type .L.str.1,@object # @.str.1
|
||
|
|
.L.str.1:
|
||
|
|
.asciz "FAIL\n"
|
||
|
|
.size .L.str.1, 6
|
||
|
|
|
||
|
|
.type .L__unnamed_1,@object # @0
|
||
|
|
.L__unnamed_1:
|
||
|
|
.asciz "_Z6vecAddPdS_S_i"
|
||
|
|
.size .L__unnamed_1, 17
|
||
|
|
|
||
|
|
.type .L__unnamed_2,@object # @1
|
||
|
|
.section .nv_fatbin,"a",@progbits
|
||
|
|
.p2align 3
|
||
|
|
.L__unnamed_2:
|
||
|
|
.asciz "P\355U\272\001\000\020\000\330\024\000\000\000\000\000\000\002\000\001\001@\000\000\000@\021\000\000\000\000\000\000\000\000\000\000\000\000\000\000\007\000\001\0002\000\000\000\000\000\000\000\000\000\000\000\021\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\177ELF\002\001\0013\007\000\000\000\000\000\000\000\002\000\276\000u\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\300\016\000\000\000\000\000\0002\0052\000@\000\000\000\000\000@\000\n\000\001\000\000.shstrtab\000.strtab\000.symtab\000.symtab_shndx\000.nv.info\000.text._Z6vecAddPdS_S_i\000.nv.info._Z6vecAddPdS_S_i\000.nv.shared._Z6vecAddPdS_S_i\000.nv.global\000.nv.constant0._Z6vecAddPdS_S_i\000.nv.rel.action\000\000.shstrtab\000.strtab\000.symtab\000.symtab_shndx\000.nv.info\000_Z6vecAddPdS_S_i\000.text._Z6vecAddPdS_S_i\000.nv.info._Z6vecAddPdS_S_i\000.nv.shared._Z6vecAddPdS_S_i\000.nv.global\000blockIdx\000blockDim\000threadIdx\000.nv.constant0._Z6vecAddPdS_S_i\000_param\000.nv.rel.action\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000C\000\000\000\003\000\b\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\220\000\000\000\003\000\t\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\233\000\000\000\001\000\t\000\001\000\000\000\000\000\000\000\001\000\000\000\000\000\000\000\244\000\000\000\001\000\t\000\002\000\000\000\000\000\000\000\001\000\000\000\000\000\000\000\255\000\000\000\001\000\t\000\000\000\000\000\000\000\000\000\001\000\000\000\000\000\000\000\267\000\000\000\003\000\007\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\335\000\000\000\003\000\006\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\0002\000\000\000\022\020\b\000\000\000\000\000\000\000\000\000\000\t\000\000\000\000\000\000\004/\b\000\b\000\000\000\r\000\000\000\004#\b\000\b\000\000\000\000\000\000\000\004\022\b\000\b\000\000\000 \000\000\000\004\021\b\000\b\000\000\000 \000\000\000\0047\004\000u\000\000\000\0010\000\000\001*\000\000\004\n\b\000\006\000\000\000@\001\034\000\003\031\034\000\004\027\f\000\000\000\000\000\003\000\030\000\000\360\021\000\004\027\f\000\000\000\000\000\002\000\020\000\000\360!\000\004\027\f\000\000\000\000\000\001\000\b\000\000\360!\000\004\027\f\000\000\000\000\000\000\000\000\000\000\360!\000\003\033\377\000\004\035\004\000\350\003\000\000\004\034\004\000\270\b\000\000\004\036\004\000 \000\000\000\000\000\000\000K\000\000\000\000\000\000\000\000\002\002\b\020\n/\"\000\000\000\b\000\000\000\000\000\000\b\b\000\000\000\000\000\000\020\b\000\000\000\000\000\000\030\b\000\000\000\000\000\000 \b\000\000\000\000\000\000(\b\000\000\000\000\000\0000\b\000\000\000\000\000\0008\b\000\000\000\000\001\000\000\b\000\000\000\000\001\000\b\b\000\000\000\000\001\000\020\b\000\000\000\000\001\000\030\b\000\000\000\000\001\000 \b\000\000\000\000\001\000(\b\000\000\000\000\001\0000\b\000\000\000\000\001\0008\b\000\000\000\000\002\000\000\b\000\000\000\000\002\000\b\b\000\000\000\000\002\000\020\b\000\000\000\000\002\000\030\b\000\000\000\000\002\000 \b\000\000\000\000\002\000(\b\000\000\000\000\002\0000\b\000\000\000\000\002\0008\b\000\000\000\000\000\000\000\024,\000\000\000\t\000\000\f\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\
|
||
|
|
.size .L__unnamed_2, 5353
|
||
|
|
|
||
|
|
.type __cuda_fatbin_wrapper,@object # @__cuda_fatbin_wrapper
|
||
|
|
.section .nvFatBinSegment,"aw",@progbits
|
||
|
|
.p2align 3
|
||
|
|
__cuda_fatbin_wrapper:
|
||
|
|
.long 1180844977 # 0x466243b1
|
||
|
|
.long 1 # 0x1
|
||
|
|
.quad .L__unnamed_2
|
||
|
|
.quad 0
|
||
|
|
.size __cuda_fatbin_wrapper, 24
|
||
|
|
|
||
|
|
.type __cuda_gpubin_handle,@object # @__cuda_gpubin_handle
|
||
|
|
.local __cuda_gpubin_handle
|
||
|
|
.comm __cuda_gpubin_handle,8,8
|
||
|
|
.section .init_array,"aw",@init_array
|
||
|
|
.p2align 3
|
||
|
|
.quad __cuda_module_ctor
|
||
|
|
.ident "Ubuntu clang version 14.0.0-1ubuntu1.1"
|
||
|
|
.section ".note.GNU-stack","",@progbits
|
||
|
|
.addrsig
|
||
|
|
.addrsig_sym _Z21__device_stub__vecAddPdS_S_i
|
||
|
|
.addrsig_sym __cudaPopCallConfiguration
|
||
|
|
.addrsig_sym cudaLaunchKernel
|
||
|
|
.addrsig_sym malloc
|
||
|
|
.addrsig_sym _ZL10cudaMallocIdE9cudaErrorPPT_m
|
||
|
|
.addrsig_sym _ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
.addrsig_sym _ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_
|
||
|
|
.addrsig_sym cudaMemcpy
|
||
|
|
.addrsig_sym _ZSt4ceilf
|
||
|
|
.addrsig_sym __cudaPushCallConfiguration
|
||
|
|
.addrsig_sym _ZSt3absd
|
||
|
|
.addrsig_sym printf
|
||
|
|
.addrsig_sym cudaFree
|
||
|
|
.addrsig_sym free
|
||
|
|
.addrsig_sym sin
|
||
|
|
.addrsig_sym cos
|
||
|
|
.addrsig_sym cudaMalloc
|
||
|
|
.addrsig_sym __cuda_register_globals
|
||
|
|
.addrsig_sym __cudaRegisterFunction
|
||
|
|
.addrsig_sym __cudaRegisterFatBinary
|
||
|
|
.addrsig_sym __cuda_module_ctor
|
||
|
|
.addrsig_sym __cudaRegisterFatBinaryEnd
|
||
|
|
.addrsig_sym __cudaUnregisterFatBinary
|
||
|
|
.addrsig_sym __cuda_module_dtor
|
||
|
|
.addrsig_sym atexit
|
||
|
|
.addrsig_sym .L__unnamed_2
|
||
|
|
.addrsig_sym __cuda_fatbin_wrapper
|
||
|
|
.addrsig_sym __cuda_gpubin_handle
|