benchmark pinned memory (wip)

src/main.cpp

@@ -35,8 +35,10 @@ struct Gpu {
     // for benchmarks
     VkDeviceMemory stagingMem = nullptr;
     VkDeviceMemory deviceMem = nullptr;
+    VkDeviceMemory pinnedMem = nullptr;
     VkBuffer stagingBuf = nullptr;
     VkBuffer deviceBuf = nullptr;
+    VkBuffer pinnedBuf = nullptr;
 };
 
 // ---------- helpers ----------
@@ -182,6 +184,14 @@ Gpu initGpu(VkPhysicalDevice phy, uint32_t buffer_size) {
     gpu.deviceMem = allocateMem(gpu.device, phy, gpu.deviceBuf,
                                 VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
 
+    gpu.pinnedBuf = createBuffer(gpu.device, buffer_size,
+                                 VK_BUFFER_USAGE_TRANSFER_SRC_BIT |
+                                     VK_BUFFER_USAGE_TRANSFER_DST_BIT);
+    gpu.pinnedMem = allocateMem(gpu.device, phy, gpu.pinnedBuf,
+                                VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
+                                    VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                                    VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
+
     return gpu;
 }
 
@@ -189,8 +199,10 @@ void cleanupGpu(Gpu &gpu) {
 
     vkDestroyBuffer(gpu.device, gpu.stagingBuf, nullptr);
     vkDestroyBuffer(gpu.device, gpu.deviceBuf, nullptr);
+    vkDestroyBuffer(gpu.device, gpu.pinnedBuf, nullptr);
     vkFreeMemory(gpu.device, gpu.stagingMem, nullptr);
     vkFreeMemory(gpu.device, gpu.deviceMem, nullptr);
+    vkFreeMemory(gpu.device, gpu.pinnedMem, nullptr);
     vkDestroyCommandPool(gpu.device, gpu.pool, nullptr);
     vkDestroyDevice(gpu.device, nullptr);
 }
@@ -213,9 +225,9 @@ void reportBenchmark(const BenchmarkResult &result, const Gpu &gpu,
     std::cout << std::endl;
 }
 
-BenchmarkResult runBenchmark(Gpu &gpu, BenchmarkConfig config) {
+BenchmarkResult runStagedBenchmark(Gpu &gpu, BenchmarkConfig config) {
 
-    // ---- fill staging buffer ----
+    // fill staging buffer
     void *mapped;
     vkMapMemory(gpu.device, gpu.stagingMem, 0, config.buffer_size, 0, &mapped);
     std::memset(mapped, 0xAB, config.buffer_size);
@@ -242,6 +254,35 @@ BenchmarkResult runBenchmark(Gpu &gpu, BenchmarkConfig config) {
     return {tH2D, tD2H};
 }
 
+BenchmarkResult runPinnedBenchmark(Gpu &gpu, BenchmarkConfig config) {
+
+    // fill staging buffer
+    void *mapped;
+    vkMapMemory(gpu.device, gpu.pinnedMem, 0, config.buffer_size, 0, &mapped);
+    std::memset(mapped, 0xAB, config.buffer_size);
+    vkUnmapMemory(gpu.device, gpu.pinnedMem);
+
+    // warm-up, probably not significant
+    benchCopy(gpu.device, gpu.pool, gpu.queue, gpu.stagingBuf, gpu.deviceBuf,
+              config.buffer_size);
+
+    // ---- benchmark host->device ----
+    double tH2D = 0.0;
+    for (uint32_t i = 0; i < config.iterations; ++i)
+        tH2D += benchCopy(gpu.device, gpu.pool, gpu.queue, gpu.pinnedBuf,
+                          gpu.pinnedBuf, config.buffer_size);
+    tH2D /= config.iterations;
+
+    // ---- benchmark device->host ----
+    double tD2H = 0.0;
+    for (uint32_t i = 0; i < config.iterations; ++i)
+        tD2H += benchCopy(gpu.device, gpu.pool, gpu.queue, gpu.pinnedBuf,
+                          gpu.pinnedBuf, config.buffer_size);
+    tD2H /= config.iterations;
+
+    return {tH2D, tD2H};
+}
+
 auto main() -> int {
 
     VkApplicationInfo app{};
@@ -286,8 +327,15 @@ auto main() -> int {
     std::cout << "-------------------" << std::endl;
 
     for (auto &gpu : gpus) {
-        BenchmarkResult res = runBenchmark(gpu, config);
+        std::cout << "Running staged benchmark" << std::endl;
+        BenchmarkResult res = runStagedBenchmark(gpu, config);
         reportBenchmark(res, gpu, config);
+
+        std::cout << "Running pinned benchmark" << std::endl;
+        res = runPinnedBenchmark(gpu, config);
+        reportBenchmark(res, gpu, config);
+
+        std::cout << "--------------------" << std::endl;
     }
 
     for (auto gpu : gpus) {