Save some WIP

Author: caugonnet

c/experimental/stf/include/cccl/c/experimental/stf/stf.h

@@ -41,12 +41,12 @@ void stf_task_add_dep(stf_task_handle t, stf_logical_data_handle ld, stf_access_
 void stf_task_start(stf_task_handle t);
 void stf_task_end(stf_task_handle t);
 cudaStream_t stf_task_get_stream(stf_task_handle t);
-void *stf_task_get(stf_task_handle t, size_t submitted_index);
+void* stf_task_get(stf_task_handle t, size_t submitted_index);
 void stf_task_destroy(stf_task_handle t);
 
-typedef struct stf_kernel_desc_handle_t *stf_kernel_desc_handle;
+typedef struct stf_kernel_desc_handle_t* stf_kernel_desc_handle;
 
-void stf_kernel_create(stf_kernel_desc_handle *d);
+void stf_kernel_create(stf_kernel_desc_handle* d);
 void stf_kernel_destroy(stf_kernel_desc_handle d);
 // TODO stf_cuda_kernel_desc : symbol, deps, args... ?
 // void stf_kernel_set_symbol((stf_kernel_handle k, const char* symbol)

c/experimental/stf/src/stf.cu

@@ -38,8 +38,8 @@ void stf_ctx_finalize(stf_ctx_handle ctx)
 
 cudaStream_t stf_fence(stf_ctx_handle ctx)
 {
-   assert(ctx);
-   return ctx->ctx.fence();
+  assert(ctx);
+  return ctx->ctx.fence();
 }
 
 void stf_logical_data(stf_ctx_handle ctx, stf_logical_data_handle* ld, void* addr, size_t sz)
@@ -107,4 +107,70 @@ void stf_task_destroy(stf_task_handle t)
   assert(t);
   delete t;
 }
+
+/**
+ * Low level example of cuda_kernel(_chain)
+ *   auto t = ctx.cuda_kernel_chain();
+     t.add_deps(lX.read());
+     t.add_deps(lY.rw());
+     t->*[&]() {
+     auto dX = t.template get<slice<double>>(0);
+     auto dY = t.template get<slice<double>>(1);
+     return std::vector<cuda_kernel_desc> {
+         { axpy, 16, 128, 0, alpha, dX, dY },
+         { axpy, 16, 128, 0, beta, dX, dY },
+         { axpy, 16, 128, 0, gamma, dX, dY }
+     };
+  };
+
+ *
+ */
+struct stf_cuda_kernel_handle_t
+{
+  // return type of ctx.cuda_kernel()
+  using kernel_type = decltype(::std::declval<context>().cuda_kernel());
+  kernel_type k;
+};
+
+void stf_cuda_kernel_create(stf_ctx_handle ctx, stf_cuda_kernel_handle* k)
+{
+  assert(k);
+  assert(ctx);
+
+  *k = new stf_cuda_kernel_handle_t{ctx->ctx.cuda_kernel()};
+}
+
+void stf_cuda_kernel_set_symbol(stf_cuda_kernel_handle k, const char* symbol)
+{
+  assert(k);
+  assert(symbol);
+
+  k->k.set_symbol(symbol);
+}
+
+void stf_cuda_kernel_add_dep(stf_cuda_kernel_handle k, stf_logical_data_handle ld, stf_access_mode m)
+{
+  assert(k);
+  assert(ld);
+
+  k->k.add_deps(cuda_kernel_dep_untyped(ld->ld, access_mode(m)));
+}
+
+// void stf_cuda_kernel_start(stf_cuda_kernel_handle k)
+// {
+//   assert(k);
+//   k->k.start();
+// }
+//
+// void stf_cuda_kernel_end(stf_cuda_kernel_handle k)
+// {
+//   assert(k);
+//   k->k.end();
+// }
+
+void stf_cuda_kernel_destroy(stf_cuda_kernel_handle t)
+{
+  assert(t);
+  delete t;
+}
 }

c/experimental/stf/test/test_task.cpp

@@ -9,8 +9,8 @@
 //===----------------------------------------------------------------------===//
 
 #include <cuda_runtime.h>
-#include <c2h/catch2_test_helper.h>
 
+#include <c2h/catch2_test_helper.h>
 #include <cccl/c/experimental/stf/stf.h>
 
 C2H_TEST("empty stf tasks", "[task]")
@@ -23,13 +23,13 @@ C2H_TEST("empty stf tasks", "[task]")
   stf_logical_data_handle lX, lY, lZ;
 
   float *X, *Y, *Z;
-  X = (float *)malloc(N*sizeof(float));
-  Y = (float *)malloc(N*sizeof(float));
-  Z = (float *)malloc(N*sizeof(float));
+  X = (float*) malloc(N * sizeof(float));
+  Y = (float*) malloc(N * sizeof(float));
+  Z = (float*) malloc(N * sizeof(float));
 
-  stf_logical_data(ctx, &lX, X, N*sizeof(float));
-  stf_logical_data(ctx, &lY, Y, N*sizeof(float));
-  stf_logical_data(ctx, &lZ, Z, N*sizeof(float));
+  stf_logical_data(ctx, &lX, X, N * sizeof(float));
+  stf_logical_data(ctx, &lY, Y, N * sizeof(float));
+  stf_logical_data(ctx, &lZ, Z, N * sizeof(float));
 
   stf_logical_data_set_symbol(lX, "X");
   stf_logical_data_set_symbol(lY, "Y");