Add memset

CHANGELOG.md

@@ -11,6 +11,8 @@ project adheres to [Semantic Versioning](http://semver.org/).
 - Added `cu::Device::getArch()`
 - Added `cu::DeviceMemory` constructor to create non-owning slice of another
   `cu::DeviceMemory` object
+- Added `cu::DeviceMemory::memset()`
+- Added `cu::Stream::memsetAsync()`
 
 ### Changed
 

include/cudawrappers/cu.hpp

@@ -585,7 +585,19 @@ class DeviceMemory : public Wrapper<CUdeviceptr> {
                                          offset);
   }
 
-  void zero(size_t size) { checkCudaCall(cuMemsetD8(_obj, 0, size)); }
+  void memset(unsigned char value, size_t size) {
+    checkCudaCall(cuMemsetD8(_obj, value, size));
+  }
+
+  void memset(unsigned short value, size_t size) {
+    checkCudaCall(cuMemsetD16(_obj, value, size));
+  }
+
+  void memset(unsigned int value, size_t size) {
+    checkCudaCall(cuMemsetD32(_obj, value, size));
+  }
+
+  void zero(size_t size) { memset(static_cast<unsigned char>(0), size); }
 
   const void *parameter()
       const  // used to construct parameter list for launchKernel();
@@ -692,8 +704,20 @@ class Stream : public Wrapper<CUstream> {
     checkCudaCall(cuMemPrefetchAsync(devPtr, size, dstDevice, _obj));
   }
 
+  void memsetAsync(DeviceMemory &devPtr, unsigned char value, size_t size) {
+    checkCudaCall(cuMemsetD8Async(devPtr, value, size, _obj));
+  }
+
+  void memsetAsync(DeviceMemory &devPtr, unsigned short value, size_t size) {
+    checkCudaCall(cuMemsetD16Async(devPtr, value, size, _obj));
+  }
+
+  void memsetAsync(DeviceMemory &devPtr, unsigned int value, size_t size) {
+    checkCudaCall(cuMemsetD32Async(devPtr, value, size, _obj));
+  }
+
   void zero(DeviceMemory &devPtr, size_t size) {
-    checkCudaCall(cuMemsetD8Async(devPtr, 0, size, _obj));
+    memsetAsync(devPtr, static_cast<unsigned char>(0), size);
   }
 
   void launchKernel(Function &function, unsigned gridX, unsigned gridY,

tests/test_cu.cpp

@@ -1,4 +1,5 @@
 #include <array>
+#include <catch2/catch_template_test_macros.hpp>
 #include <catch2/catch_test_macros.hpp>
 #include <cstring>
 #include <iostream>
@@ -80,7 +81,7 @@ TEST_CASE("Test copying cu::DeviceMemory and cu::HostMemory using cu::Stream",
   }
 }
 
-TEST_CASE("Test zeroing cu::DeviceMemory", "[zero]") {
+TEST_CASE("Test cu::DeviceMemory", "[devicememory]") {
   cu::init();
   cu::Device device(0);
   cu::Context context(CU_CTX_SCHED_BLOCKING_SYNC, device);
@@ -134,7 +135,7 @@ TEST_CASE("Test zeroing cu::DeviceMemory", "[zero]") {
     CHECK(static_cast<bool>(memcmp(src, tgt, size)));
   }
 
-  SECTION("Test cu::RegisteredMemory") {
+  SECTION("Test cu::DeviceMemory memcpy asynchronously") {
     const size_t N = 3;
     const size_t size = N * sizeof(int);
 
@@ -204,6 +205,64 @@ TEST_CASE("Test zeroing cu::DeviceMemory", "[zero]") {
   }
 }
 
+using TestTypes = std::tuple<unsigned char, unsigned short, unsigned int>;
+TEMPLATE_LIST_TEST_CASE("Test memset", "[memset]", TestTypes) {
+  cu::init();
+  cu::Device device(0);
+  cu::Context context(CU_CTX_SCHED_BLOCKING_SYNC, device);
+
+  SECTION("Test memset cu::DeviceMemory asynchronously") {
+    const size_t N = 3;
+    const size_t size = N * sizeof(TestType);
+    cu::HostMemory a(size);
+    cu::HostMemory b(size);
+    TestType value = 0xAA;
+
+    // Populate the memory with values
+    TestType* const a_ptr = static_cast<TestType*>(a);
+    TestType* const b_ptr = static_cast<TestType*>(b);
+    for (int i = 0; i < N; i++) {
+      a_ptr[i] = 0;
+      b_ptr[i] = value;
+    }
+    cu::DeviceMemory mem(size);
+
+    cu::Stream stream;
+    stream.memcpyHtoDAsync(mem, a, size);
+    stream.memsetAsync(mem, value, N);
+    stream.memcpyDtoHAsync(b, mem, size);
+    stream.synchronize();
+
+    CHECK(static_cast<bool>(memcmp(a, b, size)));
+  }
+
+  SECTION("Test zeroing cu::DeviceMemory synchronously") {
+    const size_t N = 3;
+    const size_t size = N * sizeof(TestType);
+    cu::HostMemory a(size);
+    cu::HostMemory b(size);
+    TestType value = 0xAA;
+
+    // Populate the memory with values
+    TestType* const a_ptr = static_cast<TestType*>(a);
+    TestType* const b_ptr = static_cast<TestType*>(b);
+    for (int i = 0; i < N; i++) {
+      a_ptr[i] = 0;
+      b_ptr[i] = value;
+    }
+    cu::DeviceMemory mem(size);
+
+    cu::Stream stream;
+    stream.memcpyHtoDAsync(mem, a, size);
+    stream.synchronize();
+    mem.memset(value, N);
+    stream.memcpyDtoHAsync(b, mem, size);
+    stream.synchronize();
+
+    CHECK(static_cast<bool>(memcmp(a, b, size)));
+  }
+}
+
 TEST_CASE("Test cu::Stream", "[stream]") {
   cu::init();
   cu::Device device(0);