72 lines
2.8 KiB
Plaintext
72 lines
2.8 KiB
Plaintext
|
.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_00000eab_00000000-7_hello.cpp3.i (/var/folders/KD/KDnx4D80Eh0fsORqNrFWBE+++TI/-Tmp-/ccBI#.AYqbdQ)
|
||
|
|
//-----------------------------------------------------------
|
||
|
|
|
||
|
|
//-----------------------------------------------------------
|
||
|
|
// 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_00000eab_00000000-6_hello.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 "hello.cu"
|
||
|
|
|
||
|
|
|
||
|
|
.entry helloWorld (
|
||
|
|
.param .u32 __cudaparm_helloWorld_str)
|
||
|
|
{
|
||
|
|
.reg .u16 %rh<4>;
|
||
|
|
.reg .u32 %r<9>;
|
||
|
|
.loc 28 58 0
|
||
|
|
$LBB1_helloWorld:
|
||
|
|
.loc 28 64 0
|
||
|
|
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.u32 %r4, [__cudaparm_helloWorld_str];
|
||
|
|
add.u32 %r5, %r4, %r3;
|
||
|
|
ld.global.s8 %r6, [%r5+0];
|
||
|
|
add.s32 %r7, %r6, %r3;
|
||
|
|
st.global.s8 [%r5+0], %r7;
|
||
|
|
.loc 28 65 0
|
||
|
|
exit;
|
||
|
|
$LDWend_helloWorld:
|
||
|
|
} // helloWorld
|
||
|
|
|