When building the code using CUDA 12.6 and g++ 13, I get several warnings about declared variables which are never referenced, e.g.:
nvcc -Xptxas -dlcm=cv -Xptxas -dscm=wt --define-macro ILPconfig=2,ITERS=999,MEAN=0.0,STDDEV=1.0 -gencode=arch=compute_89,code=\"sm_89,compute_89\" mma_m8n8k4_fp16fp32.cu -o mma_m8n8k4_fp16fp32.app -I /usr/local/cuda/samples/common/inc/ -L -lcudart
mma_m8n8k16_int8.cu(66): warning #177-D: variable "fpuA" was declared but never referenced
float fpuA = frag_A[0];
^
Remark: The warnings can be suppressed with "-diag-suppress <warning-number>"
mma_m8n8k16_int8.cu(67): warning #177-D: variable "fpuB" was declared but never referenced
float fpuB = frag_B[0];
^
mma_m8n8k16_int8.cu(70): warning #177-D: variable "intA" was declared but never referenced
int intA = threadIdx.x;
^
mma_m8n8k16_int8.cu(71): warning #177-D: variable "intB" was declared but never referenced
int intB = threadIdx.x + 1;
^
and it does look like these variables are not referenced. So, should these lines be deleted? Commented-out? Is there some subtle bug that's being exposed here?
When building the code using CUDA 12.6 and g++ 13, I get several warnings about declared variables which are never referenced, e.g.:
and it does look like these variables are not referenced. So, should these lines be deleted? Commented-out? Is there some subtle bug that's being exposed here?