[runtime] Implement most basic offloading support (#10)

With these changes we now have the DM core acting as our "host" that
runs IREE and the compute cores parked until a kernel should be
executed. The DM core then wakes the kernels using a cluster interrupt
to start executing a workgroup. This should additionally enable
multicore processing once the Codegen backend is capable of tiling and
distributing tiles to workgroups.

runtime/runtime/src/Quidditch/CMakeLists.txt

@@ -1,21 +1,5 @@
-add_subdirectory(registration)
-
-iree_cc_library(
-    NAME
-    device
-    SRCS
-    device.c
-    event.c
-    semaphore.c
-    DEPS
-    snRuntime
-    iree::base
-    iree::base::internal
-    iree::base::internal::arena
-    iree::base::internal::synchronization
-    iree::hal::utils::deferred_command_buffer
-    iree::hal::utils::file_transfer
-    iree::hal::utils::memory_file
-    iree::hal::utils::semaphore_base
-    PUBLIC
-)
+add_subdirectory(command_buffer)
+add_subdirectory(device)
+add_subdirectory(dispatch)
+add_subdirectory(executable)
+add_subdirectory(loader)

runtime/runtime/src/Quidditch/command_buffer/CMakeLists.txt

@@ -0,0 +1,18 @@
+
+iree_cc_library(
+    NAME
+    command_buffer
+    HDRS
+    "command_buffer.h"
+    SRCS
+    "command_buffer.c"
+    DEPS
+    snRuntime
+    iree::base
+    iree::base::internal
+    iree::base::internal::cpu
+    iree::base::internal::fpu_state
+    iree::hal
+    Quidditch::executable::executable
+    PUBLIC
+)