Open acejim opened 5 years ago
use CFLAGS
Hi I also got this error, have you fixed this?
use CFLAGS
Hello, have you resolved this issue?
First, should use CUDA 9.0 not CUDA 11.0, and downgrade GCC & G++(So I directly use ubuntu 16.04, whose default GCC version is 5.4.0). Then there will no errors like above It still has... But it seems that it doesn't affect simulation.Warning: cannot find deviceFun
.
After change to CUDA 9.0, when run this repo's cutlass, there may be error like:
cutlass-test.1.sm_70.ptx:347 Syntax error:
wmma.load.a.sync.col.m16n16k16.f16 {%r590, %r589, %r588, %r587, %r586, %r585, %r584, %r583}, [%rd66], %r374;
My solution is to modify the src/cuda-sim/ptx.l
, Line 172:
<INITIAL,NOT_OPCODE,IN_INST,IN_FUNC_DECL>{
\.a\.sync\.aligned TC; yylval->int_value = LOAD_A; return WMMA_DIRECTIVE;
\.b\.sync\.aligned TC; yylval->int_value = LOAD_B; return WMMA_DIRECTIVE;
\.c\.sync\.aligned TC; yylval->int_value = LOAD_C; return WMMA_DIRECTIVE;
\.d\.sync\.aligned TC; yylval->int_value = STORE_D; return WMMA_DIRECTIVE;
+ \.a\.sync TC; yylval->int_value = LOAD_A; return WMMA_DIRECTIVE;
+ \.b\.sync TC; yylval->int_value = LOAD_B; return WMMA_DIRECTIVE;
+ \.c\.sync TC; yylval->int_value = LOAD_C; return WMMA_DIRECTIVE;
+ \.d\.sync TC; yylval->int_value = STORE_D; return WMMA_DIRECTIVE;
\.mma\.sync\.aligned TC;yylval->int_value=MMA; return WMMA_DIRECTIVE;
+ \.mma\.sync TC;yylval->int_value=MMA; return WMMA_DIRECTIVE;
Then it can run in my environment.
Hi there,
I have successfully run gpgpu-sim, and followed the steps to run cutlass-test with gpgpu-sim, but I got the output like this:
Could you tell me how to solve it? Thank you.