CUBLAS abort

[fernandoFernandeSantos]

I'm trying to run the opcode_hist on 0_Simple/matrixMulCUBLAS from NVIDIA samples on Tesla K40c. This sample uses sgemm from CUBLAS to perform matrix multiplication. However the I'm having trouble with it. When I try to execute the following line:

eval LD_PRELOAD=.../nvbit_release/tools/opcode_hist/opcode_hist.so ./matrixMulCUBLAS

The output is as follows:

------------- NVBit (NVidia Binary Instrumentation Tool v1.4) Loaded --------------
NVBit core environment variables (mostly for nvbit-devs):
            NVDISASM = nvdisasm - override default nvdisasm found in PATH
            NOBANNER = 0 - if set, does not print this banner
---------------------------------------------------------------------------------
         INSTR_BEGIN = 0 - Beginning of the instruction interval where to apply instrumentation
           INSTR_END = 4294967295 - End of the instruction interval where to apply instrumentation
        KERNEL_BEGIN = 0 - Beginning of the kernel launch interval where to apply instrumentation
          KERNEL_END = 4294967295 - End of the kernel launch interval where to apply instrumentation
        TOOL_VERBOSE = 0 - Enable verbosity inside the tool
    COUNT_WARP_LEVEL = 1 - Count warp level or thread level instructions
    EXCLUDE_PRED_OFF = 0 - Exclude predicated off instruction from count
----------------------------------------------------------------------------------------------------
[Matrix Multiply CUBLAS] - Starting...
GPU Device 0: "Tesla K40c" with compute capability 3.5

GPU Device 0: "Tesla K40c" with compute capability 3.5

MatrixA(640,480), MatrixB(480,320), MatrixC(640,320)
Computing result using CUBLAS...matrixMulCUBLAS: arch/gk11x_hal.cpp:173: void set_imm_relative_control_flow(uint64_t*, int64_t): Assertion `!((((imm)&0xFF000000) != 0) && (((imm)&0xFF000000) != 0xFF000000))' failed.
Aborted (core dumped)

I'm using nvcc 10.1. Without the instrumentation, the code runs as expected.

[ovilla]

It is definitely a bug, and possible a bad one to fix. Thanks for reporting.
SM3.5 is relatively old in GPU terms and a lot of code was not properly tested on it.
We will take a look for sure and try to fix, but it is possible this will not get solved and we will simply deprecate SM3.5 in NVBit.

[AjinkyaBankar]

Hi Nvbit team,
I am facing a similar problem and get the following error:

lenet: arch/gm10x_hal.cpp:181: void set_imm_relative_control_flow(uint64_t*, int64_t): Assertion `!IS_LARGER_THAN_24BIT(imm)' failed.
Aborted (core dumped)

I have NVIDIA GeForce 1080Ti on my machine. Are you able to get any solution? Thanks.

[ovilla]

version 1.5.5 should have fixed this issue, closing!