Ninjastic
PostEdited versionsQuotes to this post
Post
Topic
Board Mining (Altcoins)
Re: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels.
by
salfter
on 20/01/2018, 22:10:37 UTC
While trying to get the latest version built on Gentoo, I'm getting these errors:

Code:
quark/cuda_skein512.cu: In function ‘void precalc()’:
quark/cuda_skein512.cu:2712:50: error: narrowing conversion of ‘2147483648u’ from ‘unsigned int’ to ‘int’ inside { } [-Wnarrowing]
   0x08b2b050, 0x9d7c4c27, 0x0ce2a393, 0x88e6e1ea, 0xa52b4335, 0x67a16f49, 0xd732016f, 0x4eeb2e91,
                                                  ^
quark/cuda_skein512.cu:2712:50: error: narrowing conversion of ‘2147483648u’ from ‘unsigned int’ to ‘int’ inside { } [-Wnarrowing]
quark/cuda_skein512.cu:2712:50: error: narrowing conversion of ‘2684354592u’ from ‘unsigned int’ to ‘int’ inside { } [-Wnarrowing]
quark/cuda_skein512.cu:2712:50: error: narrowing conversion of ‘4202700544u’ from ‘unsigned int’ to ‘int’ inside { } [-Wnarrowing]
quark/cuda_skein512.cu:2712:50: error: narrowing conversion of ‘3543279056u’ from ‘unsigned int’ to ‘int’ inside { } [-Wnarrowing]
quark/cuda_skein512.cu:2712:50: error: narrowing conversion of ‘4142317530u’ from ‘unsigned int’ to ‘int’ inside { } [-Wnarrowing]
quark/cuda_skein512.cu:2712:50: error: narrowing conversion of ‘3003913545u’ from ‘unsigned int’ to ‘int’ inside { } [-Wnarrowing]
...

This is on a new build I just set up last night, building against CUDA SDK 9.0.176.  I already have tpruvot-ccminer (on which sp-ccminer is based) built and running. 

I've never written any CUDA code, but it looks like unsigned values are being written into a couple of signed arrays.  I'm trying a build now with this patch:

Code:
--- quark/cuda_skein512.cu~     2017-01-01 12:38:31.000000000 -0800
+++ quark/cuda_skein512.cu      2018-01-20 14:05:06.764299663 -0800
@@ -2704,7 +2704,7 @@
 //     buffer[8] = 0x7000000000000040ull; //t2;
        CUDA_SAFE_CALL(cudaMemcpyToSymbol(precalcvalues, buffer, sizeof(buffer), 0, cudaMemcpyHostToDevice));
 
-       int endingTable[] = {
+       unsigned int endingTable[] = {
                0x80000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
                0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000200,
                0x80000000, 0x01400000, 0x00205000, 0x00005088, 0x22000800, 0x22550014, 0x05089742, 0xa0000020,
@@ -2715,7 +2715,7 @@
                0x4f0d0f04, 0x2627484e, 0x310128d2, 0xc668b434, 0x420841cc, 0x62d311b8, 0xe59ba771, 0x85a7a484
        };
 
-       int constantTable[64] = {
+       unsigned int constantTable[64] = {
                0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
                0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
                0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,

This appeared to build OK, but then it conked out again:

Code:
nvcc -gencode=arch=compute_52,code=\"sm_52,compute_52\" -gencode=arch=compute_50,code=\"sm_50,compute_50\" -I/usr/local/cuda/include -I.  --ptxas-options="-v" -gencode=arch=compute_20,code=\"sm_21,compute_20\" --maxrregcount=80 -o scrypt/salsa_kernel.o -c scrypt/salsa_kernel.cu
nvcc fatal   : Unsupported gpu architecture 'compute_20'

I'd continue, but I'll need to look into this later as I have somewhere to be in a bit. Smiley