add witness struct

spmvm/spmvm.cuh

@@ -99,6 +99,17 @@ struct spmvm_host_t
     size_t block_size;
 };
 
+template <typename scalar_t>
+struct witness_t
+{
+    const scalar_t *W;
+    const scalar_t *u;
+    const scalar_t *U;
+
+    size_t nW;
+    size_t nU;
+};
+
 template <typename scalar_t>
 class spmvm_t
 {
@@ -209,16 +220,21 @@ public:
     }
 
 public:
-    RustError invoke(spmvm_host_t<scalar_t> *csr, const scalar_t scalars[], scalar_t out[], size_t nthreads)
+    RustError invoke(spmvm_host_t<scalar_t> *csr, const witness_t<scalar_t> *witness, scalar_t out[], size_t nthreads)
     {
         assert(csr->num_rows == context->num_rows);
         assert(csr->num_cols == context->num_cols);
         assert(csr->nnz == context->nnz);
+        assert(witness->nW + witness->nU + 1 == csr->num_cols);
 
         try
         {
-            if (scalars)
-                gpu[2].HtoD(&context->d_scalars[0], &scalars[0], context->num_cols);
+            if (witness->W)
+                gpu[2].HtoD(&context->d_scalars[0], &witness->W[0], witness->nW);
+            gpu[2].HtoD(&context->d_scalars[witness->nW], witness->u, 1);
+            if (witness->U)
+                gpu[2].HtoD(&context->d_scalars[witness->nW + 1], &witness->U[0], witness->nU);
+
             cudaMemsetAsync(&context->d_out[0], 0, context->num_rows * sizeof(scalar_t), gpu[2]);
 
             size_t start_row = 0;
@@ -258,6 +274,8 @@ public:
 
                 csr_vector_mul<scalar_t><<<gpu.sm_count(), nthreads, 0, gpu[i&1]>>>(d_context);
                 CUDA_OK(cudaGetLastError());
+
+                gpu[i&1].sync();
             }
 
             gpu[2].DtoH(&out[0], &context->d_out[0], csr->num_rows);
@@ -278,12 +296,12 @@ public:
 };
 
 template <typename scalar_t>
-static RustError sparse_matrix_mul(spmvm_host_t<scalar_t> *csr, const scalar_t *scalars, scalar_t *out, size_t nthreads)
+static RustError sparse_matrix_mul(spmvm_host_t<scalar_t> *csr, const witness_t<scalar_t> *witness, scalar_t *out, size_t nthreads)
 {
     try
     {
         spmvm_t<scalar_t> spmvm{csr};
-        return spmvm.invoke(csr, scalars, out, nthreads);
+        return spmvm.invoke(csr, witness, out, nthreads);
     }
     catch (const cuda_error &e)
     {