.text .file "vecadd.cu" .globl _Z21__device_stub__vecAddPdS_S_i_host # -- Begin function _Z21__device_stub__vecAddPdS_S_i_host .p2align 4, 0x90 .type _Z21__device_stub__vecAddPdS_S_i_host,@function _Z21__device_stub__vecAddPdS_S_i_host: # @_Z21__device_stub__vecAddPdS_S_i_host .cfi_startproc # %bb.0: pushq %rbp .cfi_def_cfa_offset 16 .cfi_offset %rbp, -16 movq %rsp, %rbp .cfi_def_cfa_register %rbp pushq %r15 pushq %r14 pushq %r13 pushq %r12 pushq %rbx subq $88, %rsp .cfi_offset %rbx, -56 .cfi_offset %r12, -48 .cfi_offset %r13, -40 .cfi_offset %r14, -32 .cfi_offset %r15, -24 movl %ecx, -44(%rbp) # 4-byte Spill movq %rdx, -120(%rbp) # 8-byte Spill movq %rsi, -112(%rbp) # 8-byte Spill movq %rdi, -104(%rbp) # 8-byte Spill movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, %r14 movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, %r15 movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, %r12 movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, %rbx movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, -96(%rbp) # 8-byte Spill movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, -88(%rbp) # 8-byte Spill movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, -80(%rbp) # 8-byte Spill movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, -56(%rbp) # 8-byte Spill movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, -64(%rbp) # 8-byte Spill movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, -72(%rbp) # 8-byte Spill movq -104(%rbp), %rax # 8-byte Reload movq %rax, (%r14) movq -112(%rbp), %rax # 8-byte Reload movq %rax, (%r15) movq -120(%rbp), %rax # 8-byte Reload movq %rax, (%r12) movl -44(%rbp), %eax # 4-byte Reload movl %eax, (%rbx) movl $256, %edi # imm = 0x100 callq malloc@PLT movq %rax, %r13 movq %r14, (%rax) movq %r15, 8(%rax) movq %r12, 16(%rax) movq %rbx, 24(%rax) movq -96(%rbp), %rbx # 8-byte Reload movq %rbx, %rdi movq -88(%rbp), %r14 # 8-byte Reload movq %r14, %rsi movq -80(%rbp), %r15 # 8-byte Reload movq %r15, %rdx movq -56(%rbp), %r12 # 8-byte Reload movq %r12, %rcx callq __cudaPopCallConfiguration@PLT movq (%r15), %r10 movq (%r12), %rax movq (%rbx), %rcx movq -64(%rbp), %rdx # 8-byte Reload movq %rcx, (%rdx) movl 8(%rbx), %ecx movl %ecx, 8(%rdx) movq (%rdx), %rsi movl 8(%rdx), %edx movl 8(%r14), %r8d movq -72(%rbp), %rdi # 8-byte Reload movl %r8d, 8(%rdi) movq (%r14), %rcx movq %rcx, (%rdi) leaq vecAddPdS_S_i_wrapper(%rip), %rdi movq %r13, %r9 pushq %rax pushq %r10 callq cudaLaunchKernel@PLT addq $16, %rsp callq cudaDeviceSynchronize@PLT addq $88, %rsp popq %rbx popq %r12 popq %r13 popq %r14 popq %r15 popq %rbp .cfi_def_cfa %rsp, 8 retq .Lfunc_end0: .size _Z21__device_stub__vecAddPdS_S_i_host, .Lfunc_end0-_Z21__device_stub__vecAddPdS_S_i_host .cfi_endproc # -- End function .section .rodata.cst4,"aM",@progbits,4 .p2align 2 # -- Begin function main .LCPI1_0: .long 0x44800000 # float 1024 .section .rodata.cst8,"aM",@progbits,8 .p2align 3 .LCPI1_1: .quad 0xbff0000000000000 # double -1 .LCPI1_2: .quad 0x3eb0c6f7a0b5ed8d # double 9.9999999999999995E-7 .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, -124(%rbp) movl %edi, -120(%rbp) movq %rsi, -168(%rbp) movl $100000, -8(%rbp) # imm = 0x186A0 movq $800000, -16(%rbp) # imm = 0xC3500 movl $800000, %edi # imm = 0xC3500 callq malloc@PLT movq %rax, -80(%rbp) movq -16(%rbp), %rdi callq malloc@PLT movq %rax, -72(%rbp) movq -16(%rbp), %rdi callq malloc@PLT movq %rax, -64(%rbp) movq -16(%rbp), %rsi leaq -56(%rbp), %rdi callq _ZL10cudaMallocIdE9cudaErrorPPT_m movq -16(%rbp), %rsi leaq -48(%rbp), %rdi callq _ZL10cudaMallocIdE9cudaErrorPPT_m movq -16(%rbp), %rsi leaq -40(%rbp), %rdi callq _ZL10cudaMallocIdE9cudaErrorPPT_m movl $0, -4(%rbp) .p2align 4, 0x90 .LBB1_1: # =>This Inner Loop Header: Depth=1 movl -4(%rbp), %eax cmpl -8(%rbp), %eax jge .LBB1_3 # %bb.2: # in Loop: Header=BB1_1 Depth=1 movl -4(%rbp), %edi callq _ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_ movsd %xmm0, -32(%rbp) # 8-byte Spill movl -4(%rbp), %edi callq _ZSt3sinIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_ mulsd -32(%rbp), %xmm0 # 8-byte Folded Reload movq -80(%rbp), %rax movslq -4(%rbp), %rcx movsd %xmm0, (%rax,%rcx,8) movl -4(%rbp), %edi callq _ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_ movsd %xmm0, -32(%rbp) # 8-byte Spill movl -4(%rbp), %edi callq _ZSt3cosIiEN9__gnu_cxx11__enable_ifIXsr12__is_integerIT_EE7__valueEdE6__typeES2_ mulsd -32(%rbp), %xmm0 # 8-byte Folded Reload movq -72(%rbp), %rax movslq -4(%rbp), %rcx movsd %xmm0, (%rax,%rcx,8) addl $1, -4(%rbp) jmp .LBB1_1 .LBB1_3: movq -56(%rbp), %rdi movq -80(%rbp), %rsi movq -16(%rbp), %rdx movl $1, %ecx callq cudaMemcpy@PLT movq -48(%rbp), %rdi movq -72(%rbp), %rsi movq -16(%rbp), %rdx movl $1, %ecx callq cudaMemcpy@PLT xorps %xmm0, %xmm0 cvtsi2ssl -8(%rbp), %xmm0 movl $1024, -84(%rbp) # imm = 0x400 divss .LCPI1_0(%rip), %xmm0 callq _ZSt4ceilf cvttss2si %xmm0, %esi movl %esi, -116(%rbp) leaq -112(%rbp), %rdi movl $1, %edx movl $1, %ecx callq _ZN4dim3C2Ejjj movl -84(%rbp), %esi leaq -96(%rbp), %rdi movl $1, %edx movl $1, %ecx callq _ZN4dim3C2Ejjj movq -112(%rbp), %rdi movq %rdi, -160(%rbp) movl -104(%rbp), %esi movl %esi, -152(%rbp) movq -96(%rbp), %rdx movq %rdx, -144(%rbp) movl -88(%rbp), %ecx movl %ecx, -136(%rbp) xorl %r8d, %r8d xorl %r9d, %r9d callq __cudaPushCallConfiguration@PLT testl %eax, %eax jne .LBB1_5 # %bb.4: movq -56(%rbp), %rdi movq -48(%rbp), %rsi movq -40(%rbp), %rdx movl -8(%rbp), %ecx callq _Z21__device_stub__vecAddPdS_S_i_host .LBB1_5: movq -64(%rbp), %rdi movq -40(%rbp), %rsi movq -16(%rbp), %rdx movl $2, %ecx callq cudaMemcpy@PLT movq $0, -24(%rbp) movl $0, -4(%rbp) .p2align 4, 0x90 .LBB1_6: # =>This Inner Loop Header: Depth=1 movl -4(%rbp), %eax cmpl -8(%rbp), %eax jge .LBB1_8 # %bb.7: # in Loop: Header=BB1_6 Depth=1 movq -64(%rbp), %rax movslq -4(%rbp), %rcx movsd -24(%rbp), %xmm0 # xmm0 = mem[0],zero addsd (%rax,%rcx,8), %xmm0 movsd %xmm0, -24(%rbp) leal 1(%rcx), %eax movl %eax, -4(%rbp) jmp .LBB1_6 .LBB1_8: cvtsi2sdl -8(%rbp), %xmm1 movsd -24(%rbp), %xmm0 # xmm0 = mem[0],zero divsd %xmm1, %xmm0 movsd %xmm0, -24(%rbp) addsd .LCPI1_1(%rip), %xmm0 callq _ZSt3absd movsd .LCPI1_2(%rip), %xmm1 # xmm1 = mem[0],zero ucomisd %xmm0, %xmm1 jbe .LBB1_10 # %bb.9: leaq .L.str(%rip), %rdi jmp .LBB1_11 .LBB1_10: leaq .L.str.1(%rip), %rdi .LBB1_11: xorl %eax, %eax callq printf@PLT movq -56(%rbp), %rdi callq cudaFree@PLT movq -48(%rbp), %rdi callq cudaFree@PLT movq -40(%rbp), %rdi callq cudaFree@PLT movq -80(%rbp), %rdi callq free@PLT movq -72(%rbp), %rdi callq free@PLT movq -64(%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, -16(%rbp) movq %rsi, -8(%rbp) 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) cvtsi2sd %edi, %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) cvtsi2sd %edi, %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 movss %xmm0, -4(%rbp) popq %rbp .cfi_def_cfa %rsp, 8 jmp ceilf@PLT # TAILCALL .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, -24(%rbp) movl %esi, -12(%rbp) movl %edx, -8(%rbp) movl %ecx, -4(%rbp) movl %esi, (%rdi) movl -8(%rbp), %eax movl %eax, 4(%rdi) movl -4(%rbp), %eax movl %eax, 8(%rdi) 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) andps .LCPI7_0(%rip), %xmm0 popq %rbp .cfi_def_cfa %rsp, 8 retq .Lfunc_end7: .size _ZSt3absd, .Lfunc_end7-_ZSt3absd .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\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\357\037\340\375\003\"\000\21132 %r<9>3\000Urd<18\023\000\020f\023\000\323fd<4>;\n\nmov.uW\000\033,\212\000b;\ncvta\262\000\004%\000\023,\201\000\"ld\363\000\001\362\000o%r1, [\370\000\005\030].\000\002\217\000\0373/\000\007\0372/\000\000\0372/\000\007\0371/\000\000\017\215\000\b#0]\325\000#tok\002\0045\000 4,\006\000\0233\037\000\n\034\000\0215\034\000\0374;\000\005\0216\037\000\0372;\000\002\0217\034\000\0376;\000\005\0218\037\000\0371;\000\002\0219\034\000Q8;\nst\023\000q[%SP+0]\026\000\0329\026\000\0228\026\000\0327\026\000\"16\027\000\0225\027\000\"32\027\000!24\027\000\"1;\375\001\001\300\001\2702, %ctaid.x\027\000c3, %nt\026\000qul.lo.s\031\000#4,5\000(r30\000\000)\001\003/\0003add,\000$6,1\000\f\211\000\002\267\000\0216\302\001\002A\000%7,\033\000\007\026\000%8,\272\000\222;\nsetp.ge]\0002p1,6\000\362\016%r8;\n@%p1 bra LBB0_2;\nbra.uni\020\00021;\n\b\000\021:Z\000\002e\001410,Y\001\001q\000\002\263\0008d11\211\0004shlI\003412, \000\0233\345\000\003\031\000$3,P\000\000\007\000\0212N\000\003k\003\001M\000\000#\000(];{\000$4,\276\001\nJ\000$5,\037\000\rJ\000\0232J\000\02353\000$rn\032\000\000\177\000\001j\000\000&\000\tg\000%6,\017\002\nh\000$7, \000\003h\000!stN\000\001b\000!7]K\000\0333E\001\0232E\001\3002:\nret;\n\n}\n\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 .ident "Ubuntu clang version 14.0.0-1ubuntu1.1" .section ".note.GNU-stack","",@progbits