All enums in the C API are now prefixed with CLBlast to avoid potential name clashes with other projects

CHANGELOG

@@ -1,6 +1,7 @@
 
 Development version (next release)
 - Updated to version 8.0 of the CLCudaAPI C++11 OpenCL header
+- Changed the enums in the C API to avoid potential name clashes with external code
 - Greatly improved the way exceptions are handled in the library (thanks to 'intelfx')
 - Improved performance of GEMM kernels for small sizes by using a direct single-kernel implementation
 - Fixed a bug in the tests and samples related to waiting for an invalid event

samples/cache.c

@@ -106,13 +106,13 @@ void run_example_routine(const cl_device_id device) {
   clock_t start = clock();
 
   // Calls an example routine
-  StatusCode status = CLBlastSasum(n,
-                                   device_output, 0,
-                                   device_input, 0, 1,
-                                   &queue, &event);
+  CLBlastStatusCode status = CLBlastSasum(n,
+                                          device_output, 0,
+                                          device_input, 0, 1,
+                                          &queue, &event);
 
   // Wait for completion
-  if (status == kSuccess) {
+  if (status == CLBlastSuccess) {
     clWaitForEvents(1, &event);
     clReleaseEvent(event);
   }

samples/dgemv.c

@@ -74,17 +74,17 @@ int main(void) {
   clEnqueueWriteBuffer(queue, device_y, CL_TRUE, 0, m*sizeof(double), host_y, 0, NULL, NULL);
 
   // Call the DGEMV routine.
-  StatusCode status = CLBlastDgemv(kRowMajor, kNo,
-                                   m, n,
-                                   alpha,
-                                   device_a, 0, a_ld,
-                                   device_x, 0, 1,
-                                   beta,
-                                   device_y, 0, 1,
-                                   &queue, &event);
+  CLBlastStatusCode status = CLBlastDgemv(CLBlastLayoutRowMajor, CLBlastTransposeNo,
+                                          m, n,
+                                          alpha,
+                                          device_a, 0, a_ld,
+                                          device_x, 0, 1,
+                                          beta,
+                                          device_y, 0, 1,
+                                          &queue, &event);
 
   // Wait for completion
-  if (status == kSuccess) {
+  if (status == CLBlastSuccess) {
     clWaitForEvents(1, &event);
     clReleaseEvent(event);
   }

samples/haxpy.c

@@ -71,13 +71,13 @@ int main(void) {
   clEnqueueWriteBuffer(queue, device_b, CL_TRUE, 0, n*sizeof(cl_half), host_b, 0, NULL, NULL);
 
   // Call the HAXPY routine.
-  StatusCode status = CLBlastHaxpy(n, alpha,
-                                   device_a, 0, 1,
-                                   device_b, 0, 1,
-                                   &queue, &event);
+  CLBlastStatusCode status = CLBlastHaxpy(n, alpha,
+                                          device_a, 0, 1,
+                                          device_b, 0, 1,
+                                          &queue, &event);
 
   // Wait for completion
-  if (status == kSuccess) {
+  if (status == CLBlastSuccess) {
     clWaitForEvents(1, &event);
     clReleaseEvent(event);
   }

samples/sasum.c

@@ -67,13 +67,13 @@ int main(void) {
   clEnqueueWriteBuffer(queue, device_output, CL_TRUE, 0, 1*sizeof(float), host_output, 0, NULL, NULL);
 
   // Call the SASUM routine.
-  StatusCode status = CLBlastSasum(n,
-                                   device_output, 0,
-                                   device_input, 0, 1,
-                                   &queue, &event);
+  CLBlastStatusCode status = CLBlastSasum(n,
+                                          device_output, 0,
+                                          device_input, 0, 1,
+                                          &queue, &event);
 
   // Wait for completion
-  if (status == kSuccess) {
+  if (status == CLBlastSuccess) {
     clWaitForEvents(1, &event);
     clReleaseEvent(event);
   }

samples/sgemm.c

@@ -77,17 +77,18 @@ int main(void) {
   clEnqueueWriteBuffer(queue, device_c, CL_TRUE, 0, m*n*sizeof(float), host_c, 0, NULL, NULL);
 
   // Call the SGEMM routine.
-  StatusCode status = CLBlastSgemm(kRowMajor, kNo, kNo,
-                                   m, n, k,
-                                   alpha,
-                                   device_a, 0, a_ld,
-                                   device_b, 0, b_ld,
-                                   beta,
-                                   device_c, 0, c_ld,
-                                   &queue, &event);
+  CLBlastStatusCode status = CLBlastSgemm(CLBlastLayoutRowMajor,
+                                          CLBlastTransposeNo, CLBlastTransposeNo,
+                                          m, n, k,
+                                          alpha,
+                                          device_a, 0, a_ld,
+                                          device_b, 0, b_ld,
+                                          beta,
+                                          device_c, 0, c_ld,
+                                          &queue, &event);
 
   // Wait for completion
-  if (status == kSuccess) {
+  if (status == CLBlastSuccess) {
     clWaitForEvents(1, &event);
     clReleaseEvent(event);
   }

scripts/generator/generator/cpp.py

@@ -73,7 +73,7 @@ def clblast_c_h(routine):
     """The C API header (.h)"""
     result = NL + "// " + routine.description + ": " + routine.short_names() + NL
     for flavour in routine.flavours:
-        result += routine.routine_header_c(flavour, 31, " PUBLIC_API") + ";" + NL
+        result += routine.routine_header_c(flavour, 38, " PUBLIC_API") + ";" + NL
     return result
 
 
@@ -82,13 +82,15 @@ def clblast_c_cc(routine):
     result = NL + "// " + routine.name.upper() + NL
     for flavour in routine.flavours:
         template = "<" + flavour.template + ">" if routine.no_scalars() else ""
-        indent = " " * (45 + routine.length() + len(template))
-        result += routine.routine_header_c(flavour, 20, "") + " {" + NL
+        indent = " " * (16 + routine.length() + len(template))
+        result += routine.routine_header_c(flavour, 27, "") + " {" + NL
         result += "  try {" + NL
-        result += "    return static_cast<StatusCode>(clblast::" + routine.name.capitalize() + template + "("
+        result += "    return static_cast<CLBlastStatusCode>(" + NL
+        result += "      clblast::" + routine.name.capitalize() + template + "("
         result += ("," + NL + indent).join([a for a in routine.arguments_cast(flavour, indent)])
-        result += "," + NL + indent + "queue, event));" + NL
-        result += "  } catch (...) { return static_cast<StatusCode>(clblast::DispatchExceptionForC()); }" + NL
+        result += "," + NL + indent + "queue, event)" + NL
+        result += "    );" + NL
+        result += "  } catch (...) { return static_cast<CLBlastStatusCode>(clblast::DispatchExceptionForC()); }" + NL
         result += "}" + NL
     return result
 

scripts/generator/generator/doc.py

@@ -32,7 +32,7 @@ def generate(routine):
     result += "C API:" + NL
     result += "```" + NL
     for flavour in routine.flavours:
-        result += routine.routine_header_c(flavour, 20, "") + NL
+        result += routine.routine_header_c(flavour, 27, "") + NL
     result += "```" + NL + NL
 
     # Routine arguments

scripts/generator/generator/routine.py

@@ -349,6 +349,13 @@ class Routine:
             return [", ".join(definitions)]
         return []
 
+    def options_def_c(self):
+        """As above, but now for the C API"""
+        if self.options:
+            definitions = ["const CLBlast" + convert.option_to_clblast(o) + " " + o for o in self.options]
+            return [", ".join(definitions)]
+        return []
+
     def options_def_wrapper_clblas(self):
         """As above, but now using clBLAS data-types"""
         if self.options:
@@ -453,6 +460,17 @@ class Routine:
                 list(chain(*[self.buffer_def(b) for b in self.scalar_buffers_second()])) +
                 list(chain(*[self.scalar_def(s, flavour) for s in self.other_scalars()])))
 
+    def arguments_def_c(self, flavour):
+        """As above, but for the C API"""
+        return (self.options_def_c() + self.sizes_def() +
+                list(chain(*[self.buffer_def(b) for b in self.scalar_buffers_first()])) +
+                self.scalar_def("alpha", flavour) +
+                list(chain(*[self.buffer_def(b) for b in self.buffers_first()])) +
+                self.scalar_def("beta", flavour) +
+                list(chain(*[self.buffer_def(b) for b in self.buffers_second()])) +
+                list(chain(*[self.buffer_def(b) for b in self.scalar_buffers_second()])) +
+                list(chain(*[self.scalar_def(s, flavour) for s in self.other_scalars()])))
+
     def arguments_def_wrapper_clblas(self, flavour):
         """As above, but clBLAS wrapper plain data-types"""
         return (self.options_def_wrapper_clblas() + self.sizes_def() +
@@ -523,8 +541,8 @@ class Routine:
     def routine_header_c(self, flavour, spaces, extra_qualifier):
         """As above, but now for C"""
         indent = " " * (spaces + self.length())
-        result = "StatusCode" + extra_qualifier + " CLBlast" + flavour.name + self.name + "("
-        result += (",\n" + indent).join([a for a in self.arguments_def(flavour)])
+        result = "CLBlastStatusCode" + extra_qualifier + " CLBlast" + flavour.name + self.name + "("
+        result += (",\n" + indent).join([a for a in self.arguments_def_c(flavour)])
         result += ",\n" + indent + "cl_command_queue* queue, cl_event* event)"
         return result
 

test/correctness/tester.cpp

@@ -248,7 +248,7 @@ void Tester<T,U>::TestErrorCodes(const StatusCode clblas_status, const StatusCod
   }
 
   // Could not compile the CLBlast kernel properly
-  else if (clblast_status == StatusCode::kBuildProgramFailure ||
+  else if (clblast_status == StatusCode::kOpenCLBuildProgramFailure ||
            clblast_status == StatusCode::kNotImplemented) {
     PrintTestResult(kSkippedCompilation);
     ReportSkipped();