SHARE
TWEET

Untitled

a guest Apr 26th, 2010 704 Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
  1. .version 2.0
  2.         .target sm_20
  3.         // compiled with /usr/local/cuda/bin/../open64/lib//be
  4.         // nvopencc 3.0 built on 2010-03-11
  5.  
  6.         //-----------------------------------------------------------
  7.         // Compiling /tmp/tmpxft_0000e98a_00000000-13_vectorAdd.compute_20.cpp3.i (/var/folders/fZ/fZpMfXGjEIefAb1MGdDCpk+++TI/-Tmp-/ccBI#.A9sVnm)
  8.         //-----------------------------------------------------------
  9.  
  10.         //-----------------------------------------------------------
  11.         // Options:
  12.         //-----------------------------------------------------------
  13.         //  Target:ptx, ISA:sm_20, Endian:little, Pointer Size:32
  14.         //  -O3 (Optimization level)
  15.         //  -g0 (Debug level)
  16.         //  -m2 (Report advisories)
  17.         //-----------------------------------------------------------
  18.  
  19.         .file   1       "<command-line>"
  20.         .file   2       "/tmp/tmpxft_0000e98a_00000000-12_vectorAdd.compute_20.cudafe2.gpu"
  21.         .file   3       "/usr/lib/gcc/i686-apple-darwin10/4.2.1/include/stddef.h"
  22.         .file   4       "/usr/local/cuda/bin/../include/crt/device_runtime.h"
  23.         .file   5       "/usr/local/cuda/bin/../include/host_defines.h"
  24.         .file   6       "/usr/local/cuda/bin/../include/builtin_types.h"
  25.         .file   7       "/usr/local/cuda/bin/../include/device_types.h"
  26.         .file   8       "/usr/local/cuda/bin/../include/driver_types.h"
  27.         .file   9       "/usr/local/cuda/bin/../include/texture_types.h"
  28.         .file   10      "/usr/local/cuda/bin/../include/vector_types.h"
  29.         .file   11      "/usr/local/cuda/bin/../include/device_launch_parameters.h"
  30.         .file   12      "/usr/local/cuda/bin/../include/crt/storage_class.h"
  31.         .file   13      "/usr/include/i386/_types.h"
  32.         .file   14      "/usr/include/time.h"
  33.         .file   15      "/usr/local/cuda/bin/../include/texture_fetch_functions.h"
  34.         .file   16      "/usr/local/cuda/bin/../include/common_functions.h"
  35.         .file   17      "/usr/local/cuda/bin/../include/crt/func_macro.h"
  36.         .file   18      "/usr/local/cuda/bin/../include/math_functions.h"
  37.         .file   19      "/usr/local/cuda/bin/../include/device_functions.h"
  38.         .file   20      "/usr/local/cuda/bin/../include/math_constants.h"
  39.         .file   21      "/usr/local/cuda/bin/../include/sm_11_atomic_functions.h"
  40.         .file   22      "/usr/local/cuda/bin/../include/sm_12_atomic_functions.h"
  41.         .file   23      "/usr/local/cuda/bin/../include/sm_13_double_functions.h"
  42.         .file   24      "/usr/local/cuda/bin/../include/sm_20_atomic_functions.h"
  43.         .file   25      "/usr/local/cuda/bin/../include/sm_20_intrinsics.h"
  44.         .file   26      "/usr/local/cuda/bin/../include/math_functions_dbl_ptx3.h"
  45.         .file   27      "vectorAdd.cu"
  46.  
  47.  
  48.         .entry _Z6VecAddPKfS0_Pfi (
  49.                 .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_A,
  50.                 .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_B,
  51.                 .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_C,
  52.                 .param .s32 __cudaparm__Z6VecAddPKfS0_Pfi_N)
  53.         {
  54.         .reg .u32 %r<15>;
  55.         .reg .f32 %f<5>;
  56.         .reg .pred %p<3>;
  57.         .loc    27      43      0
  58. $LBB1__Z6VecAddPKfS0_Pfi:
  59.         mov.u32         %r1, %ctaid.x;
  60.         mov.u32         %r2, %ntid.x;
  61.         mul.lo.u32      %r3, %r1, %r2;
  62.         mov.u32         %r4, %tid.x;
  63.         add.u32         %r5, %r4, %r3;
  64.         ld.param.s32    %r6, [__cudaparm__Z6VecAddPKfS0_Pfi_N];
  65.         setp.le.s32     %p1, %r6, %r5;
  66.         @%p1 bra        $Lt_0_1026;
  67.         .loc    27      47      0
  68.         mul.lo.u32      %r7, %r5, 4;
  69.         ld.param.u32    %r8, [__cudaparm__Z6VecAddPKfS0_Pfi_A];
  70.         add.u32         %r9, %r8, %r7;
  71.         ld.global.f32   %f1, [%r9+0];
  72.         ld.param.u32    %r10, [__cudaparm__Z6VecAddPKfS0_Pfi_B];
  73.         add.u32         %r11, %r10, %r7;
  74.         ld.global.f32   %f2, [%r11+0];
  75.         add.f32         %f3, %f1, %f2;
  76.         ld.param.u32    %r12, [__cudaparm__Z6VecAddPKfS0_Pfi_C];
  77.         add.u32         %r13, %r12, %r7;
  78.         st.global.f32   [%r13+0], %f3;
  79. $Lt_0_1026:
  80.         .loc    27      48      0
  81.         exit;
  82. $LDWend__Z6VecAddPKfS0_Pfi:
  83.         } // _Z6VecAddPKfS0_Pfi
  84.  
  85.  
  86. .version 1.4
  87.         .target sm_10, map_f64_to_f32
  88.         // compiled with /usr/local/cuda/bin/../open64/lib//be
  89.         // nvopencc 3.0 built on 2010-03-11
  90.  
  91.         //-----------------------------------------------------------
  92.         // Compiling /tmp/tmpxft_0000e98a_00000000-9_vectorAdd.compute_10.cpp3.i (/var/folders/fZ/fZpMfXGjEIefAb1MGdDCpk+++TI/-Tmp-/ccBI#.ZvYJOx)
  93.         //-----------------------------------------------------------
  94.  
  95.         //-----------------------------------------------------------
  96.         // Options:
  97.         //-----------------------------------------------------------
  98.         //  Target:ptx, ISA:sm_10, Endian:little, Pointer Size:32
  99.         //  -O3 (Optimization level)
  100.         //  -g0 (Debug level)
  101.         //  -m2 (Report advisories)
  102.         //-----------------------------------------------------------
  103.  
  104.         .file   1       "<command-line>"
  105.         .file   2       "/tmp/tmpxft_0000e98a_00000000-8_vectorAdd.compute_10.cudafe2.gpu"
  106.         .file   3       "/usr/lib/gcc/i686-apple-darwin10/4.2.1/include/stddef.h"
  107.         .file   4       "/usr/local/cuda/bin/../include/crt/device_runtime.h"
  108.         .file   5       "/usr/local/cuda/bin/../include/host_defines.h"
  109.         .file   6       "/usr/local/cuda/bin/../include/builtin_types.h"
  110.         .file   7       "/usr/local/cuda/bin/../include/device_types.h"
  111.         .file   8       "/usr/local/cuda/bin/../include/driver_types.h"
  112.         .file   9       "/usr/local/cuda/bin/../include/texture_types.h"
  113.         .file   10      "/usr/local/cuda/bin/../include/vector_types.h"
  114.         .file   11      "/usr/local/cuda/bin/../include/device_launch_parameters.h"
  115.         .file   12      "/usr/local/cuda/bin/../include/crt/storage_class.h"
  116.         .file   13      "/usr/include/i386/_types.h"
  117.         .file   14      "/usr/include/time.h"
  118.         .file   15      "/usr/local/cuda/bin/../include/texture_fetch_functions.h"
  119.         .file   16      "/usr/local/cuda/bin/../include/common_functions.h"
  120.         .file   17      "/usr/local/cuda/bin/../include/crt/func_macro.h"
  121.         .file   18      "/usr/local/cuda/bin/../include/math_functions.h"
  122.         .file   19      "/usr/local/cuda/bin/../include/device_functions.h"
  123.         .file   20      "/usr/local/cuda/bin/../include/math_constants.h"
  124.         .file   21      "/usr/local/cuda/bin/../include/sm_11_atomic_functions.h"
  125.         .file   22      "/usr/local/cuda/bin/../include/sm_12_atomic_functions.h"
  126.         .file   23      "/usr/local/cuda/bin/../include/sm_13_double_functions.h"
  127.         .file   24      "/usr/local/cuda/bin/../include/common_types.h"
  128.         .file   25      "/usr/local/cuda/bin/../include/sm_20_atomic_functions.h"
  129.         .file   26      "/usr/local/cuda/bin/../include/sm_20_intrinsics.h"
  130.         .file   27      "/usr/local/cuda/bin/../include/math_functions_dbl_ptx1.h"
  131.         .file   28      "vectorAdd.cu"
  132.  
  133.  
  134.         .entry _Z6VecAddPKfS0_Pfi (
  135.                 .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_A,
  136.                 .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_B,
  137.                 .param .u32 __cudaparm__Z6VecAddPKfS0_Pfi_C,
  138.                 .param .s32 __cudaparm__Z6VecAddPKfS0_Pfi_N)
  139.         {
  140.         .reg .u16 %rh<4>;
  141.         .reg .u32 %r<13>;
  142.         .reg .f32 %f<5>;
  143.         .reg .pred %p<3>;
  144.         .loc    28      43      0
  145. $LBB1__Z6VecAddPKfS0_Pfi:
  146.         mov.u16         %rh1, %ctaid.x;
  147.         mov.u16         %rh2, %ntid.x;
  148.         mul.wide.u16    %r1, %rh1, %rh2;
  149.         cvt.u32.u16     %r2, %tid.x;
  150.         add.u32         %r3, %r2, %r1;
  151.         ld.param.s32    %r4, [__cudaparm__Z6VecAddPKfS0_Pfi_N];
  152.         setp.le.s32     %p1, %r4, %r3;
  153.         @%p1 bra        $Lt_0_1026;
  154.         .loc    28      47      0
  155.         mul.lo.u32      %r5, %r3, 4;
  156.         ld.param.u32    %r6, [__cudaparm__Z6VecAddPKfS0_Pfi_A];
  157.         add.u32         %r7, %r6, %r5;
  158.         ld.global.f32   %f1, [%r7+0];
  159.         ld.param.u32    %r8, [__cudaparm__Z6VecAddPKfS0_Pfi_B];
  160.         add.u32         %r9, %r8, %r5;
  161.         ld.global.f32   %f2, [%r9+0];
  162.         add.f32         %f3, %f1, %f2;
  163.         ld.param.u32    %r10, [__cudaparm__Z6VecAddPKfS0_Pfi_C];
  164.         add.u32         %r11, %r10, %r5;
  165.         st.global.f32   [%r11+0], %f3;
  166. $Lt_0_1026:
  167.         .loc    28      48      0
  168.         exit;
  169. $LDWend__Z6VecAddPKfS0_Pfi:
  170.         } // _Z6VecAddPKfS0_Pfi
  171.  
  172.  
  173. // Found 2 PTX assemblies.
RAW Paste Data
Top