A problem of PTX code of Cublas64_50_35 of Cuda5.0 release

owen123 · November 5, 2012, 7:27am

Hi, I can’t get the ptx code of Cublas64_50_35.dll by cuobjdump tool. My command is cuobjdump.exe -ptx cublas64_50_35.dll. I am a student and want to study all functions of cublas of CUDA_50 release. And who can help me? Thank you very much

owen123 · November 5, 2012, 8:50am

My OS is Windows7. My hardware is GTX640.

sjiagc · November 5, 2012, 9:07am

It is odd! I could get it. So your machine and OS is 64-bit? Could you get ptx from vecAdd - one of the CUDA sample?

owen123 · November 6, 2012, 2:24am

Yes, My machine and OS is 64-bit. I can get ptx from vecAdd and cublas32_50_35.dll, but not cublas64_50_35.dll.

owen123 · November 6, 2012, 3:14am

I have installed the software of Cuda5.0 release named ‘cuda_5.0.35_winvista_win7_win8_general_64’

sjiagc · November 6, 2012, 5:45am

Could you paste the output of your cuobjdump for cublas 64-bit? That may help us.

owen123 · November 6, 2012, 11:12am

Hi, the output of cuobjdump on cublas64_50_35.dll as follow. And there are some sm_35’s ptx code but no sm_30. As you know, There are many kernels in cublas64_50_35.dll.

Fatbin elf code:
================
arch = sm_10
code version = [1,2]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas_v1.cu

Fatbin elf code:
================
arch = sm_13
code version = [1,2]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas_v1.cu

Fatbin elf code:
================
arch = sm_20
code version = [1,6]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas_v1.cu

Fatbin elf code:
================
arch = sm_30
code version = [1,6]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas_v1.cu

Fatbin elf code:
================
arch = sm_35
code version = [1,6]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas_v1.cu

Fatbin ptx code:
================
arch = sm_35
code version = [3,1]
producer = cuda
host = windows
compile_size = 64bit
compressed
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas_v1.cu
ptxasOptions = 












.version 3.1

.target sm_35

.address_size 64



.file	1 "C:/DOCUME~1/BUILDM~1.COM/LOCALS~1/Temp/tmpxft_000013e4_00000000-39_cublas_v1.compute_35.cpp3.i"

.file	2 "d:\bld\rel\gpgpu\toolkit\r5.0\cuda\tools\cudart\../cnprt/cuda_device_runtime_api.h"



.weak .func (.param .b32 func_retval0) cudaMalloc(

.param .b64 cudaMalloc_param_0,

.param .b64 cudaMalloc_param_1

)

{

.reg .s32 %r;





mov.u32 %r1, 30;

st.param.b32	[func_retval0+0], %r1;

.loc 2 66 3

ret;

}



.weak .func (.param .b32 func_retval0) cudaFuncGetAttributes(

.param .b64 cudaFuncGetAttributes_param_0,

.param .b64 cudaFuncGetAttributes_param_1

)

{

.reg .s32 %r;





mov.u32 %r1, 30;

st.param.b32	[func_retval0+0], %r1;

.loc 2 71 3

ret;

}






Fatbin elf code:
================
arch = sm_10
code version = [1,2]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas.cu

Fatbin elf code:
================
arch = sm_13
code version = [1,2]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas.cu

Fatbin elf code:
================
arch = sm_20
code version = [1,6]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas.cu

Fatbin elf code:
================
arch = sm_30
code version = [1,6]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas.cu

Fatbin elf code:
================
arch = sm_35
code version = [1,6]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas.cu

Fatbin ptx code:
================
arch = sm_35
code version = [3,1]
producer = cuda
host = windows
compile_size = 64bit
compressed
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/cublas.cu
ptxasOptions = 












.version 3.1

.target sm_35

.address_size 64



.file	1 "C:/DOCUME~1/BUILDM~1.COM/LOCALS~1/Temp/tmpxft_00000d08_00000000-39_cublas.compute_35.cpp3.i"

.file	2 "d:\bld\rel\gpgpu\toolkit\r5.0\cuda\tools\cudart\../cnprt/cuda_device_runtime_api.h"



.weak .func (.param .b32 func_retval0) cudaMalloc(

.param .b64 cudaMalloc_param_0,

.param .b64 cudaMalloc_param_1

)

{

.reg .s32 %r;





mov.u32 %r1, 30;

st.param.b32	[func_retval0+0], %r1;

.loc 2 66 3

ret;

}



.weak .func (.param .b32 func_retval0) cudaFuncGetAttributes(

.param .b64 cudaFuncGetAttributes_param_0,

.param .b64 cudaFuncGetAttributes_param_1

)

{

.reg .s32 %r;





mov.u32 %r1, 30;

st.param.b32	[func_retval0+0], %r1;

.loc 2 71 3

ret;

}






Fatbin elf code:
================
arch = sm_10
code version = [1,2]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/xerbla.cu

Fatbin elf code:
================
arch = sm_13
code version = [1,2]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/xerbla.cu

Fatbin elf code:
================
arch = sm_20
code version = [1,6]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/xerbla.cu

Fatbin elf code:
================
arch = sm_30
code version = [1,6]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/xerbla.cu

Fatbin elf code:
================
arch = sm_35
code version = [1,6]
producer = cuda
host = windows
compile_size = 64bit
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/xerbla.cu

Fatbin ptx code:
================
arch = sm_35
code version = [3,1]
producer = cuda
host = windows
compile_size = 64bit
compressed
identifier = D:/Bld/rel/gpgpu/toolkit/r5.0/cublas/src/xerbla.cu
ptxasOptions = 












.version 3.1

.target sm_35

.address_size 64



.file	1 "C:/DOCUME~1/BUILDM~1.COM/LOCALS~1/Temp/tmpxft_000000a8_00000000-39_xerbla.compute_35.cpp3.i"

.file	2 "d:\bld\rel\gpgpu\toolkit\r5.0\cuda\tools\cudart\../cnprt/cuda_device_runtime_api.h"



.weak .func (.param .b32 func_retval0) cudaMalloc(

.param .b64 cudaMalloc_param_0,

.param .b64 cudaMalloc_param_1

)

{

.reg .s32 %r;





mov.u32 %r1, 30;

st.param.b32	[func_retval0+0], %r1;

.loc 2 66 3

ret;

}



.weak .func (.param .b32 func_retval0) cudaFuncGetAttributes(

.param .b64 cudaFuncGetAttributes_param_0,

.param .b64 cudaFuncGetAttributes_param_1

)

{

.reg .s32 %r;





mov.u32 %r1, 30;

st.param.b32	[func_retval0+0], %r1;

.loc 2 71 3

ret;

}

owen123 · November 6, 2012, 11:14am

There are no the ptx code of the kernel Sgem and dgem.

sjiagc · November 7, 2012, 8:11am

Maybe there is something wrong here. Are you willing to file a bug? Here is the steps:

If not registered, please register first and then apply for program “CUDA/GPU Computing Registered Developer Program” in registered developer programs;
Open page https://developer.nvidia.com/rdp/bugs/cudagpu-bug-reporting;
After seeing the bug report page, please fill the required itmes, other items are optional, but detailed information will help us to target and fix the issue a lot;
If necessary, an attachment should be uploaded;
For Linux system, it is better to attach an nvidia-bug-report;
If an issue is related to specific code pattern, a sample code and instructions to compile it are desired for reproduction.

If you have problem in filing a bug or you want me to file a bug for you, please feel free to let me know.

Topic		Replies	Views
Help!! I can't get my NVidia GeForce GT 525M to load in a single CUDA PTX kernel!! CUDA Programming and Performance	11	5794	November 16, 2012
cublas for 10.1 is missing CUDA Setup and Installation	23	64095	October 24, 2020
cuBLAS call from kernel in CUDA 10.0 GPU-Accelerated Libraries	9	4821	April 7, 2021
How to call cublas library into my cuda fortran code? Legacy PGI Compilers	5	5935	December 8, 2011
Cublas Bug GPU-Accelerated Libraries cublas	8	2027	June 21, 2022
cannot compile with cublas code in kernel CUDA Programming and Performance	4	998	July 8, 2019
Cannot run any CUDA kernels CUDA runtime doesn't recognize NVIDIA GPU CUDA Programming and Performance	26	12333	August 24, 2010
Error in my code... CUDA Programming and Performance	11	2538	December 19, 2014
Matlab 2007b and CUDA Can't compile CUDA example with Matlab CUDA Programming and Performance	19	20100	June 25, 2009
Undefined reference to cublas_ -- but library is there! Legacy PGI Compilers	5	12474	September 1, 2010

A problem of PTX code of Cublas64_50_35 of Cuda5.0 release

Related topics