Untitled

.version 2.0
    .target sm_20
    // compiled with /usr/local/cuda/bin/../open64/lib//be
    // nvopencc 3.0 built on 2010-03-11

    //-----------------------------------------------------------
    // Compiling /tmp/tmpxft_0000e98a_00000000-13_vectorAdd.compute_20.cpp3.i (/var/folders/fZ/fZpMfXGjEIefAb1MGdDCpk+++TI/-Tmp-/ccBI#.A9sVnm)
    //-----------------------------------------------------------

    //-----------------------------------------------------------
    // Options:
    //-----------------------------------------------------------
    //  Target:ptx, ISA:sm_20, Endian:little, Pointer Size:32
    //  -O3 (Optimization level)
    //  -g0 (Debug level)
    //  -m2 (Report advisories)
    //-----------------------------------------------------------

    .file   1   "<command-line>"
    .file   2   "/tmp/tmpxft_0000e98a_00000000-12_vectorAdd.compute_20.cudafe2.gpu"
    .file   3   "/usr/lib/gcc/i686-apple-darwin10/4.2.1/include/stddef.h"
    .file   4   "/usr/local/cuda/bin/../include/crt/device_runtime.h"
    .file   5   "/usr/local/cuda/bin/../include/host_defines.h"
    .file   6   "/usr/local/cuda/bin/../include/builtin_types.h"
    .file   7   "/usr/local/cuda/bin/../include/device_types.h"
    .file   8   "/usr/local/cuda/bin/../include/driver_types.h"
    .file   9   "/usr/local/cuda/bin/../include/texture_types.h"
    .file   10  "/usr/local/cuda/bin/../include/vector_types.h"
    .file   11  "/usr/local/cuda/bin/../include/device_launch_parameters.h"
    .file   12  "/usr/local/cuda/bin/../include/crt/storage_class.h"
    .file   13  "/usr/include/i386/_types.h"
    .file   14  "/usr/include/time.h"
    .file   15  "/usr/local/cuda/bin/../include/texture_fetch_functions.h"
    .file   16  "/usr/local/cuda/bin/../include/common_functions.h"
    .file   17  "/usr/local/cuda/bin/../include/crt/func_macro.h"
    .file   18  "/usr/local/cuda/bin/../include/math_functions.h"
    .file   19  "/usr/local/cuda/bin/../include/device_functions.h"
    .file   20  "/usr/local/cuda/bin/../include/math_constants.h"
    .file   21  "/usr/local/cuda/bin/../include/sm_11_atomic_functions.h"
    .file   22  "/usr/local/cuda/bin/../include/sm_12_atomic_functions.h"
    .file   23  "/usr/local/cuda/bin/../include/sm_13_double_functions.h"
    .file   24  "/usr/local/cuda/bin/../include/sm_20_atomic_functions.h"
    .file   25  "/usr/local/cuda/bin/../include/sm_20_intrinsics.h"
    .file   26  "/usr/local/cuda/bin/../include/math_functions_dbl_ptx3.h"
    .file   27  "vectorAdd.cu"


    .entry _Z6VecAddPKfS0_Pfi (
        .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_A,
        .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_B,
        .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_C,
        .param .s32 __cudaparm__Z6VecAddPKfS0_Pfi_N)
    {
    .reg .u32 %r<15>;
    .reg .f32 %f<5>;
    .reg .pred %p<3>;
    .loc    27  43  0
$LBB1__Z6VecAddPKfS0_Pfi:
    mov.u32     %r1, %ctaid.x;
    mov.u32     %r2, %ntid.x;
    mul.lo.u32  %r3, %r1, %r2;
    mov.u32     %r4, %tid.x;
    add.u32     %r5, %r4, %r3;
    ld.param.s32    %r6, [__cudaparm__Z6VecAddPKfS0_Pfi_N];
    setp.le.s32     %p1, %r6, %r5;
    @%p1 bra    $Lt_0_1026;
    .loc    27  47  0
    mul.lo.u32  %r7, %r5, 4;
    ld.param.u32    %r8, [__cudaparm__Z6VecAddPKfS0_Pfi_A];
    add.u32     %r9, %r8, %r7;
    ld.global.f32   %f1, [%r9+0];
    ld.param.u32    %r10, [__cudaparm__Z6VecAddPKfS0_Pfi_B];
    add.u32     %r11, %r10, %r7;
    ld.global.f32   %f2, [%r11+0];
    add.f32     %f3, %f1, %f2;
    ld.param.u32    %r12, [__cudaparm__Z6VecAddPKfS0_Pfi_C];
    add.u32     %r13, %r12, %r7;
    st.global.f32   [%r13+0], %f3;
$Lt_0_1026:
    .loc    27  48  0
    exit;
$LDWend__Z6VecAddPKfS0_Pfi:
    } // _Z6VecAddPKfS0_Pfi


.version 1.4
    .target sm_10, map_f64_to_f32
    // compiled with /usr/local/cuda/bin/../open64/lib//be
    // nvopencc 3.0 built on 2010-03-11

    //-----------------------------------------------------------
    // Compiling /tmp/tmpxft_0000e98a_00000000-9_vectorAdd.compute_10.cpp3.i (/var/folders/fZ/fZpMfXGjEIefAb1MGdDCpk+++TI/-Tmp-/ccBI#.ZvYJOx)
    //-----------------------------------------------------------

    //-----------------------------------------------------------
    // Options:
    //-----------------------------------------------------------
    //  Target:ptx, ISA:sm_10, Endian:little, Pointer Size:32
    //  -O3 (Optimization level)
    //  -g0 (Debug level)
    //  -m2 (Report advisories)
    //-----------------------------------------------------------

    .file   1   "<command-line>"
    .file   2   "/tmp/tmpxft_0000e98a_00000000-8_vectorAdd.compute_10.cudafe2.gpu"
    .file   3   "/usr/lib/gcc/i686-apple-darwin10/4.2.1/include/stddef.h"
    .file   4   "/usr/local/cuda/bin/../include/crt/device_runtime.h"
    .file   5   "/usr/local/cuda/bin/../include/host_defines.h"
    .file   6   "/usr/local/cuda/bin/../include/builtin_types.h"
    .file   7   "/usr/local/cuda/bin/../include/device_types.h"
    .file   8   "/usr/local/cuda/bin/../include/driver_types.h"
    .file   9   "/usr/local/cuda/bin/../include/texture_types.h"
    .file   10  "/usr/local/cuda/bin/../include/vector_types.h"
    .file   11  "/usr/local/cuda/bin/../include/device_launch_parameters.h"
    .file   12  "/usr/local/cuda/bin/../include/crt/storage_class.h"
    .file   13  "/usr/include/i386/_types.h"
    .file   14  "/usr/include/time.h"
    .file   15  "/usr/local/cuda/bin/../include/texture_fetch_functions.h"
    .file   16  "/usr/local/cuda/bin/../include/common_functions.h"
    .file   17  "/usr/local/cuda/bin/../include/crt/func_macro.h"
    .file   18  "/usr/local/cuda/bin/../include/math_functions.h"
    .file   19  "/usr/local/cuda/bin/../include/device_functions.h"
    .file   20  "/usr/local/cuda/bin/../include/math_constants.h"
    .file   21  "/usr/local/cuda/bin/../include/sm_11_atomic_functions.h"
    .file   22  "/usr/local/cuda/bin/../include/sm_12_atomic_functions.h"
    .file   23  "/usr/local/cuda/bin/../include/sm_13_double_functions.h"
    .file   24  "/usr/local/cuda/bin/../include/common_types.h"
    .file   25  "/usr/local/cuda/bin/../include/sm_20_atomic_functions.h"
    .file   26  "/usr/local/cuda/bin/../include/sm_20_intrinsics.h"
    .file   27  "/usr/local/cuda/bin/../include/math_functions_dbl_ptx1.h"
    .file   28  "vectorAdd.cu"


    .entry _Z6VecAddPKfS0_Pfi (
        .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_A,
        .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_B,
        .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_C,
        .param .s32 __cudaparm__Z6VecAddPKfS0_Pfi_N)
    {
    .reg .u16 %rh<4>;
    .reg .u32 %r<13>;
    .reg .f32 %f<5>;
    .reg .pred %p<3>;
    .loc    28  43  0
$LBB1__Z6VecAddPKfS0_Pfi:
    mov.u16     %rh1, %ctaid.x;
    mov.u16     %rh2, %ntid.x;
    mul.wide.u16    %r1, %rh1, %rh2;
    cvt.u32.u16     %r2, %tid.x;
    add.u32     %r3, %r2, %r1;
    ld.param.s32    %r4, [__cudaparm__Z6VecAddPKfS0_Pfi_N];
    setp.le.s32     %p1, %r4, %r3;
    @%p1 bra    $Lt_0_1026;
    .loc    28  47  0
    mul.lo.u32  %r5, %r3, 4;
    ld.param.u32    %r6, [__cudaparm__Z6VecAddPKfS0_Pfi_A];
    add.u32     %r7, %r6, %r5;
    ld.global.f32   %f1, [%r7+0];
    ld.param.u32    %r8, [__cudaparm__Z6VecAddPKfS0_Pfi_B];
    add.u32     %r9, %r8, %r5;
    ld.global.f32   %f2, [%r9+0];
    add.f32     %f3, %f1, %f2;
    ld.param.u32    %r10, [__cudaparm__Z6VecAddPKfS0_Pfi_C];
    add.u32     %r11, %r10, %r5;
    st.global.f32   [%r11+0], %f3;
$Lt_0_1026:
    .loc    28  48  0
    exit;
$LDWend__Z6VecAddPKfS0_Pfi:
    } // _Z6VecAddPKfS0_Pfi


// Found 2 PTX assemblies.