out-of-order command buffer support and sample (#60)

* merge improvements and reduce code deltas

* add support for out-of-order command buffers

* actually create an out-of-order command buffer

* fix README

* add more command line options

* update README, add command buffer picture

Author: bashbaug

include/layer_util.hpp

@@ -8,6 +8,7 @@
 #include <CL/cl_layer.h>
 
 #include <cstring>
+#include <cctype>
 #include <vector>
 
 template<class T>
@@ -76,6 +77,37 @@ static inline cl_int writeStringToMemory(
     return CL_SUCCESS;
 }
 
+static cl_uint getOpenCLVersionFromString(
+    const char* str)
+{
+    cl_uint major = 0;
+    cl_uint minor = 0;
+
+    // The device version string has the form:
+    //   OpenCL <Major>.<Minor> <Vendor Specific Info>
+    const char* prefix = "OpenCL ";
+    size_t sz = strlen(prefix);
+    if (strlen(str) > sz &&
+        strncmp(str, prefix, sz) == 0) {
+        const char* check = str + sz;
+        while (isdigit(check[0])) {
+            major *= 10;
+            major += check[0] - '0';
+            ++check;
+        }
+        if (check[0] == '.') {
+            ++check;
+        }
+        while (isdigit(check[0])) {
+            minor *= 10;
+            minor += check[0] - '0';
+            ++check;
+        }
+    }
+
+    return (major << 16) | minor;
+}
+
 static inline bool checkStringForExtension(
     const char* str,
     const char* extensionName )

layers/10_cmdbufemu/README.md

@@ -12,6 +12,11 @@ The functionality in this emulation layer is sufficient to run the command buffe
 
 Please note that the emulated command buffers are intended to be functional, but unlike a native implementation of `cl_khr_command_buffer`, they may not provide any performance benefit over similar code without using command buffers.
 
+## Layer Requirement
+
+Because this layer calls `clCloneKernel` when recording a command buffer it requires an OpenCL 2.1 or newer device.
+If an older device is detected then the layer will not advertise support for the `cl_khr_command_buffer` extension.
+
 ## Key APIs and Concepts
 
 The most important concepts to understand from this sample are how to intercept `clGetExtensionFunctionAddressForPlatform` to return emulated functions for an extension.
@@ -27,3 +32,4 @@ This section describes some of the limitations of the emulated `cl_khr_command_b
 
 * The event associated with `clEnqueueCommandBufferKHR` will not have the proper event command type `CL_COMMAND_COMMAND_BUFFER_KHR`.
 * Event profiling for `clEnqueueCommandBufferKHR` will not be correct for the `QUEUED`, `SUBMITTED`, and `START` times.
+* Many error conditions are not properly checked for and returned.