[SYCLomatic] implement cuCtxCreate_v3 and cuCtxCreate_v4 (#2240)

Signed-off-by: Huang, Andy <[email protected]>

Author: AndyCHHuang

clang/lib/DPCT/APINames.inc

@@ -1573,8 +1573,8 @@ ENTRY(cuDevicePrimaryCtxSetFlags, cuDevicePrimaryCtxSetFlags_v2, false, NO_FLAG,
 
 // Context Management
 ENTRY(cuCtxCreate, cuCtxCreate, true, NO_FLAG, P4, "comment")
-ENTRY(cuCtxCreate_v3, cuCtxCreate_v3, false, NO_FLAG, P4, "comment")
-ENTRY(cuCtxCreate_v4, cuCtxCreate_v4, false, NO_FLAG, P4, "comment")
+ENTRY(cuCtxCreate_v3, cuCtxCreate_v3, true, NO_FLAG, P4, "comment")
+ENTRY(cuCtxCreate_v4, cuCtxCreate_v4, true, NO_FLAG, P4, "comment")
 ENTRY(cuCtxDestroy, cuCtxDestroy_v2, true, NO_FLAG, P4, "DPCT1026/DPCT1027")
 ENTRY(cuCtxGetApiVersion, cuCtxGetApiVersion, true, NO_FLAG, P4, "comment")
 ENTRY(cuCtxGetCacheConfig, cuCtxGetCacheConfig, false, NO_FLAG, P4, "comment")
@@ -2304,4 +2304,4 @@ ENTRY(make_cuComplex, make_cuComplex, true, NO_FLAG, P4, "comment")
 ENTRY(make_cuDoubleComplex, make_cuDoubleComplex, true, NO_FLAG, P4, "comment")
 ENTRY(make_cuFloatComplex, make_cuFloatComplex, true, NO_FLAG, P4, "comment")
 ENTRY(__assert_fail, __assert_fail, true, NO_FLAG, P4, "Successful")
-ENTRY(__assertfail, __assertfail, true, NO_FLAG, P4, "Successful")
+ENTRY(__assertfail, __assertfail, true, NO_FLAG, P4, "Successful")

clang/lib/DPCT/APINamesDriver.inc

@@ -100,6 +100,38 @@ CONDITIONAL_FACTORY_ENTRY(
             "cuCtxCreate_v2", DEREF(0),
             CALL(MapNames::getDpctNamespace() + "push_device_for_curr_thread", ARG(2))))))
 
+CONDITIONAL_FACTORY_ENTRY(
+    checkIsUseNoQueueDevice(),
+    WARNING_FACTORY_ENTRY(
+        "cuCtxCreate_v3",
+        ASSIGNABLE_FACTORY(ASSIGN_FACTORY_ENTRY("cuCtxCreate_v3", DEREF(0),
+                                                ARG(4))),
+        Diagnostics::FUNC_CALL_REMOVED, std::string("cuCtxCreate_v3"),
+        std::string("it is redundant if it is migrated with option "
+                    "--helper-function-preference=no-queue-device which declares a global SYCL device "
+                    "and queue.")),
+    FEATURE_REQUEST_FACTORY(
+        HelperFeatureEnum::device_ext,
+        ASSIGNABLE_FACTORY(ASSIGN_FACTORY_ENTRY(
+            "cuCtxCreate_v3", DEREF(0),
+            CALL(MapNames::getDpctNamespace() + "select_device", ARG(4))))))
+
+CONDITIONAL_FACTORY_ENTRY(
+    checkIsUseNoQueueDevice(),
+    WARNING_FACTORY_ENTRY(
+        "cuCtxCreate_v4",
+        ASSIGNABLE_FACTORY(ASSIGN_FACTORY_ENTRY("cuCtxCreate_v4", DEREF(0),
+                                                ARG(3))),
+        Diagnostics::FUNC_CALL_REMOVED, std::string("cuCtxCreate_v4"),
+        std::string("it is redundant if it is migrated with option "
+                    "--helper-function-preference=no-queue-device which declares a global SYCL device "
+                    "and queue.")),
+    FEATURE_REQUEST_FACTORY(
+        HelperFeatureEnum::device_ext,
+        ASSIGNABLE_FACTORY(ASSIGN_FACTORY_ENTRY(
+            "cuCtxCreate_v4", DEREF(0),
+            CALL(MapNames::getDpctNamespace() + "select_device", ARG(3))))))
+
 CONDITIONAL_FACTORY_ENTRY(
     checkIsUseNoQueueDevice(),
     WARNING_FACTORY_ENTRY(

clang/lib/DPCT/ASTTraversal.cpp

@@ -14349,8 +14349,9 @@ REGISTER_RULE(DriverDeviceAPIRule, PassKind::PK_Migration)
 void DriverContextAPIRule::registerMatcher(ast_matchers::MatchFinder &MF) {
   auto contextAPI = [&]() {
     return hasAnyName(
-        "cuInit", "cuCtxCreate_v2", "cuCtxSetCurrent", "cuCtxGetCurrent",
-        "cuCtxSynchronize", "cuCtxDestroy_v2", "cuDevicePrimaryCtxRetain",
+        "cuInit", "cuCtxCreate_v2", "cuCtxCreate_v3", "cuCtxCreate_v4",
+        "cuCtxSetCurrent", "cuCtxGetCurrent", "cuCtxSynchronize",
+        "cuCtxDestroy_v2", "cuDevicePrimaryCtxRetain",
         "cuDevicePrimaryCtxRelease_v2", "cuDevicePrimaryCtxRelease",
         "cuCtxGetDevice", "cuCtxGetApiVersion", "cuCtxGetLimit",
         "cuCtxPushCurrent_v2", "cuCtxPopCurrent_v2");

clang/test/dpct/driver_context.cu

@@ -1,5 +1,5 @@
-// UNSUPPORTED: cuda-8.0, cuda-9.0, cuda-9.1, cuda-9.2, cuda-10.0, cuda-10.1, cuda-10.2
-// UNSUPPORTED: v8.0, v9.0, v9.1, v9.2, v10.0, v10.1, v10.2
+// UNSUPPORTED: cuda-8.0, cuda-9.0, cuda-9.1, cuda-9.2, cuda-10.0, cuda-10.1, cuda-10.2, cuda-11.0, cuda-11.1, cuda-11.2, cuda-11.3, cuda-11.4, cuda-11.5, cuda-11.6, cuda-11.7, cuda-11.8, cuda-12.0, cuda-12.1, cuda-12.2, cuda-12.3, cuda-12.4
+// UNSUPPORTED: v8.0, v9.0, v9.1, v9.2, v10.0, v10.1, v10.2, v11.0, v11.1, v11.2, v11.3, v11.4, v11.5, v11.6, v11.7, v11.8, v12.0, v12.1, v12.2, v12.3, v12.4
 // RUN: dpct --format-range=none -out-root %T/driver_context %s --cuda-include-path="%cuda-path/include" -- -std=c++14 -x cuda --cuda-host-only
 // RUN: FileCheck %s --match-full-lines --input-file %T/driver_context/driver_context.dp.cpp
 // RUN: %if build_lit %{icpx -c -fsycl %T/driver_context/driver_context.dp.cpp -o %T/driver_context/driver_context.dp.o %}
@@ -35,6 +35,16 @@ int main(){
   // CHECK: ctx = dpct::push_device_for_curr_thread(device);
   cuCtxCreate(&ctx, CU_CTX_LMEM_RESIZE_TO_MAX, device);
 
+  unsigned int flags = CU_CTX_MAP_HOST;
+  CUexecAffinityParam* paramsArray;
+  // CHECK: ctx = dpct::select_device(device);
+  cuCtxCreate_v3(&ctx, paramsArray, 1, flags, device);
+
+  CUctxCreateParams* ctxCreateParams;
+  // CHECK: ctx = dpct::select_device(device);
+  cuCtxCreate_v4(&ctx, ctxCreateParams, flags, device);
+
+
   // CHECK: ctx = dpct::select_device(device);
   cuDevicePrimaryCtxRetain(&ctx, device);