nvlabs / xmp

CUDA accelerated(X) Multi-Precision library

XMP

CUDA accelerated(X) Multi-Precision library

Please see documentation folder for more details

xmp's Issues

Compile Error

Hello, I am now using the xmp lib for my project. When I download the example code. (simple_rsa.cc) I could not compile the code perfectly, no metter using gcc, g++ or nvcc. Here is my command:
g++ simple_rsa.cc -L lib/libxmp.so lib/libxmp.a
And I got this errors:

simple_rsa.cc: In function ‘int main()’:
simple_rsa.cc:117:81: error: invalid conversion from ‘uint32_t* {aka unsigned int*}’ to ‘uint32_t {aka unsigned int}’ [-fpermissive]
   XMP_CHECK_ERROR(xmpIntegersImport(handle,n,limbs,-1,sizeof(uint32_t),0,0,h_n,1));
                                                                                 ^
simple_rsa.cc:29:20: note: in definition of macro ‘XMP_CHECK_ERROR’
   xmpError_t error=fun;     \
                    ^~~
In file included from simple_rsa.cc:1:0:
/home/pi/Documents/xmp/include/xmp.h:127:19: note:   initializing argument 8 of ‘xmpError_t xmpIntegersImport(xmpHandle_t, xmpIntegers_t, uint32_t, uint32_t, int32_t, size_t, int32_t, uint32_t, void*)’
 xmpError_t XMPAPI xmpIntegersImport(xmpHandle_t handle, xmpIntegers_t out, uint32_t count, uint32_t words, int32_t order, size_t size, int32_t endian, uint32_t nails, void* in);
                   ^~~~~~~~~~~~~~~~~
simple_rsa.cc:117:81: error: invalid conversion from ‘int’ to ‘void*’ [-fpermissive]
   XMP_CHECK_ERROR(xmpIntegersImport(handle,n,limbs,-1,sizeof(uint32_t),0,0,h_n,1));
                                                                                 ^
simple_rsa.cc:29:20: note: in definition of macro ‘XMP_CHECK_ERROR’
   xmpError_t error=fun;     \
                    ^~~
In file included from simple_rsa.cc:1:0:
/home/pi/Documents/xmp/include/xmp.h:127:19: note:   initializing argument 9 of ‘xmpError_t xmpIntegersImport(xmpHandle_t, xmpIntegers_t, uint32_t, uint32_t, int32_t, size_t, int32_t, uint32_t, void*)’
 xmpError_t XMPAPI xmpIntegersImport(xmpHandle_t handle, xmpIntegers_t out, uint32_t count, uint32_t words, int32_t order, size_t size, int32_t endian, uint32_t nails, void* in);
                   ^~~~~~~~~~~~~~~~~
simple_rsa.cc:118:77: error: invalid conversion from ‘uint32_t* {aka unsigned int*}’ to ‘uint32_t {aka unsigned int}’ [-fpermissive]
   XMP_CHECK_ERROR(xmpIntegersImport(handle,e,1,-1,sizeof(uint32_t),0,0,h_e,1));
                                                                             ^
simple_rsa.cc:29:20: note: in definition of macro ‘XMP_CHECK_ERROR’
   xmpError_t error=fun;     \
                    ^~~
In file included from simple_rsa.cc:1:0:
/home/pi/Documents/xmp/include/xmp.h:127:19: note:   initializing argument 8 of ‘xmpError_t xmpIntegersImport(xmpHandle_t, xmpIntegers_t, uint32_t, uint32_t, int32_t, size_t, int32_t, uint32_t, void*)’
 xmpError_t XMPAPI xmpIntegersImport(xmpHandle_t handle, xmpIntegers_t out, uint32_t count, uint32_t words, int32_t order, size_t size, int32_t endian, uint32_t nails, void* in);
                   ^~~~~~~~~~~~~~~~~
simple_rsa.cc:118:77: error: invalid conversion from ‘int’ to ‘void*’ [-fpermissive]
   XMP_CHECK_ERROR(xmpIntegersImport(handle,e,1,-1,sizeof(uint32_t),0,0,h_e,1));
                                                                             ^
simple_rsa.cc:29:20: note: in definition of macro ‘XMP_CHECK_ERROR’
   xmpError_t error=fun;     \
                    ^~~
In file included from simple_rsa.cc:1:0:
/home/pi/Documents/xmp/include/xmp.h:127:19: note:   initializing argument 9 of ‘xmpError_t xmpIntegersImport(xmpHandle_t, xmpIntegers_t, uint32_t, uint32_t, int32_t, size_t, int32_t, uint32_t, void*)’
 xmpError_t XMPAPI xmpIntegersImport(xmpHandle_t handle, xmpIntegers_t out, uint32_t count, uint32_t words, int32_t order, size_t size, int32_t endian, uint32_t nails, void* in);
                   ^~~~~~~~~~~~~~~~~
simple_rsa.cc:119:81: error: invalid conversion from ‘uint32_t* {aka unsigned int*}’ to ‘uint32_t {aka unsigned int}’ [-fpermissive]
   XMP_CHECK_ERROR(xmpIntegersImport(handle,d,limbs,-1,sizeof(uint32_t),0,0,h_d,1));
                                                                                 ^
simple_rsa.cc:29:20: note: in definition of macro ‘XMP_CHECK_ERROR’
   xmpError_t error=fun;     \
                    ^~~
In file included from simple_rsa.cc:1:0:
/home/pi/Documents/xmp/include/xmp.h:127:19: note:   initializing argument 8 of ‘xmpError_t xmpIntegersImport(xmpHandle_t, xmpIntegers_t, uint32_t, uint32_t, int32_t, size_t, int32_t, uint32_t, void*)’
 xmpError_t XMPAPI xmpIntegersImport(xmpHandle_t handle, xmpIntegers_t out, uint32_t count, uint32_t words, int32_t order, size_t size, int32_t endian, uint32_t nails, void* in);
                   ^~~~~~~~~~~~~~~~~
simple_rsa.cc:119:81: error: invalid conversion from ‘int’ to ‘void*’ [-fpermissive]
   XMP_CHECK_ERROR(xmpIntegersImport(handle,d,limbs,-1,sizeof(uint32_t),0,0,h_d,1));
                                                                                 ^
simple_rsa.cc:29:20: note: in definition of macro ‘XMP_CHECK_ERROR’
   xmpError_t error=fun;     \
                    ^~~
In file included from simple_rsa.cc:1:0:
/home/pi/Documents/xmp/include/xmp.h:127:19: note:   initializing argument 9 of ‘xmpError_t xmpIntegersImport(xmpHandle_t, xmpIntegers_t, uint32_t, uint32_t, int32_t, size_t, int32_t, uint32_t, void*)’
 xmpError_t XMPAPI xmpIntegersImport(xmpHandle_t handle, xmpIntegers_t out, uint32_t count, uint32_t words, int32_t order, size_t size, int32_t endian, uint32_t nails, void* in);
                   ^~~~~~~~~~~~~~~~~
simple_rsa.cc:120:85: error: invalid conversion from ‘uint32_t* {aka unsigned int*}’ to ‘uint32_t {aka unsigned int}’ [-fpermissive]
 MP_CHECK_ERROR(xmpIntegersImport(handle,m,limbs,-1,sizeof(uint32_t),0,0,h_m,count));
                                                                                  ^
simple_rsa.cc:29:20: note: in definition of macro ‘XMP_CHECK_ERROR’
   xmpError_t error=fun;     \
                    ^~~
In file included from simple_rsa.cc:1:0:
/home/pi/Documents/xmp/include/xmp.h:127:19: note:   initializing argument 8 of ‘xmpError_t xmpIntegersImport(xmpHandle_t, xmpIntegers_t, uint32_t, uint32_t, int32_t, size_t, int32_t, uint32_t, void*)’
 xmpError_t XMPAPI xmpIntegersImport(xmpHandle_t handle, xmpIntegers_t out, uint32_t count, uint32_t words, int32_t order, size_t size, int32_t endian, uint32_t nails, void* in);
                   ^~~~~~~~~~~~~~~~~
simple_rsa.cc:120:85: error: invalid conversion from ‘int’ to ‘void*’ [-fpermissive]
 MP_CHECK_ERROR(xmpIntegersImport(handle,m,limbs,-1,sizeof(uint32_t),0,0,h_m,count));
                                                                                  ^
simple_rsa.cc:29:20: note: in definition of macro ‘XMP_CHECK_ERROR’
   xmpError_t error=fun;     \
                    ^~~
In file included from simple_rsa.cc:1:0:
/home/pi/Documents/xmp/include/xmp.h:127:19: note:   initializing argument 9 of ‘xmpError_t xmpIntegersImport(xmpHandle_t, xmpIntegers_t, uint32_t, uint32_t, int32_t, size_t, int32_t, uint32_t, void*)’
 xmpError_t XMPAPI xmpIntegersImport(xmpHandle_t handle, xmpIntegers_t out, uint32_t count, uint32_t words, int32_t order, size_t size, int32_t endian, uint32_t nails, void* in);
                   ^~~~~~~~~~~~~~~~~

Is there any method to fix this problem?

Compute 7_0 errors

I'm getting the following errors when attempting to compile with CUDA 10.1 V10.1.243:

ptxas /tmp/tmpxft_00004b30_00000000-5_distributedmp.compute_70.ptx, line 41537; error : Instruction 'shfl' without '.sync' is not supported on .target sm_70 and higher from PTX ISA version 6.4

I've fixed the errors by using shfl.sync i.e.

#if __CUDA_ARCH__ >= 700
    asm volatile ("shfl.sync.up.b32 %0|%%c1, %1, 1, %2, 0xffffffff;" : "=r"(x) :\
 "r"(r1), "r"((256-_width)*256));
#else
    asm volatile ("shfl.up.b32 %0|%%c1, %1, 1, %2;" : "=r"(x) : "r"(r1), "r"((25\
6-_width)*256));
#endif

and

#if CUDA_ARCH >= 700
asm volatile ("setp.eq.and.u32 %%c2,%1,0xFFFFFFFF,%%c1;\n\t"
"vote.sync.ballot.b32 %0,%%c2, 0xFFFFFFFF;"
: "=r"(p) : "r"(x));
asm volatile ("setp.eq.u32 %%c2,%1,1;\n\t"
"vote.sync.ballot.b32 %0,%%c2, 0xFFFFFFFF;"
: "=r"(g) : "r"(r1));
#else
asm volatile ("setp.eq.and.u32 %%c2,%1,0xFFFFFFFF,%%c1;\n\t"
"vote.ballot.b32 %0,%%c2;" : "=r"(p) : "r"(x));
asm volatile ("setp.eq.u32 %%c2,%1,1;\n\t"
"vote.ballot.b32 %0,%%c2;" : "=r"(g) : "r"(r1));
#endif

Wherever needed, however, it was a mechanical Turk kind of thing because I never programmed in CUDA and certainly not low level CUDA, so I'm not sure if the implementation is correct.

Newer CUDA?

Any chance this repo might be updated to a newer version of CUDA?

Feature request: add GMP API compatibility, gmpbench results

Hi,
Just seen project on github but no source code still..
Would be nice if similar to how cublas, cufft provide standard API with popular CPU libraries like FFTW this cuda multiprecision library has API like GMP so codes using GMP doesn't need to be ported.. Assuming no performance is lost using this API..

error: ‘memcpy’ was not declared in this scope

I got the error shown below compiling the library on my system, which has the following
characteristics:

• Ubuntu 16.04 LTS
• Cuda compilations tools v7.5.17
• gcc version 5.3.1

/usr/include/string.h: In function ‘void* mempcpy_inline(void, const void, size_t)’:
/usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this scope
return (char ) memcpy (__dest, __src, __n) + __n;
^
Makefile:20: recipe for target 'xmp.o' failed
make: ** [xmp.o] Error 1

I actually fixed it by adding the _FORCE_INLINES definition on nvcc compiler flags

NVCC_FLAGS=-D_FORCE_INLINES ...

Calling functions from device

Is it safe to call functions like regmp_add_kernel from a device kernel function by including implementation.h? Or should functions only be called from CPU side? Any possibility of adding a device API in the future?

Error in doc

there is copy-paste error in example in https://github.com/NVlabs/xmp/blob/master/docs/xmp.docx

XMP_CHECK_ERROR(xmpIntegersImport(handle,base,N,limbs,-1,sizeof(uint32_t),0,0,b));
XMP_CHECK_ERROR(xmpIntegersImport(handle,exp,1,limbs,-1,sizeof(uint32_t),0,0,b));
XMP_CHECK_ERROR(xmpIntegersImport(handle,mod,1,limbs,-1,sizeof(uint32_t),0,0,b));

last parameter here should be b/e/m rather than b/b/b

Can provide Windows binaries until Cuda 8.0 for Maxwell testing?

crt_rsa sample gets stuck for large number of iterations

When running crt_rsa.exe on AWS P3.2XLarge and P2.XLarge with CUDA 10 the test gets stuck. If the number of iterations is decreased the test completes.

Thanks,
Yuval.