From 7875e26ea220f9f44bdba1477ca89bf42bf9f9c0 Mon Sep 17 00:00:00 2001
From: Kasper Sauramo <kasper.sauramo@analogway.com>
Date: Mon, 10 Nov 2025 09:32:01 +0200
Subject: [PATCH] remove cached benchmark

it wasn't doing what I expected, probably
---
 src/main.cpp | 60 +---------------------------------------------------
 1 file changed, 1 insertion(+), 59 deletions(-)

diff --git a/src/main.cpp b/src/main.cpp
index 99befd8..c238942 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -1,11 +1,11 @@
 #include <vulkan/vulkan.h>
+#include <vulkan/vulkan_core.h>
 
 #include <chrono>
 #include <cstring>
 #include <iostream>
 #include <limits>
 #include <vector>
-#include <vulkan/vulkan_core.h>
 
 /// TODO: Pinned memory
 /// TODO: Plot by buffer size
@@ -34,13 +34,9 @@ struct Gpu {
     VkCommandPool pool = nullptr;
     // for benchmarks
     VkDeviceMemory stagingMem = nullptr;
-    VkDeviceMemory cachedMem = nullptr;
     VkDeviceMemory deviceMem = nullptr;
-    VkDeviceMemory pinnedMem = nullptr;
     VkBuffer stagingBuf = nullptr;
-    VkBuffer cachedBuf = nullptr;
     VkBuffer deviceBuf = nullptr;
-    VkBuffer pinnedBuf = nullptr;
 };
 
 // ---------- helpers ----------
@@ -186,22 +182,6 @@ Gpu initGpu(VkPhysicalDevice phy, uint32_t buffer_size) {
     gpu.deviceMem = allocateMem(gpu.device, phy, gpu.deviceBuf,
                                 VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
 
-    gpu.cachedBuf = createBuffer(gpu.device, buffer_size,
-                                 VK_BUFFER_USAGE_TRANSFER_DST_BIT |
-                                     VK_BUFFER_USAGE_TRANSFER_SRC_BIT);
-    gpu.cachedMem = allocateMem(gpu.device, phy, gpu.cachedBuf,
-                                VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
-                                    VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
-                                    VK_MEMORY_PROPERTY_HOST_CACHED_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;
 }
 
@@ -209,12 +189,8 @@ void cleanupGpu(Gpu &gpu) {
 
     vkDestroyBuffer(gpu.device, gpu.stagingBuf, nullptr);
     vkDestroyBuffer(gpu.device, gpu.deviceBuf, nullptr);
-    vkDestroyBuffer(gpu.device, gpu.cachedBuf, nullptr);
-    vkDestroyBuffer(gpu.device, gpu.pinnedBuf, nullptr);
     vkFreeMemory(gpu.device, gpu.stagingMem, nullptr);
     vkFreeMemory(gpu.device, gpu.deviceMem, nullptr);
-    vkFreeMemory(gpu.device, gpu.cachedMem, nullptr);
-    vkFreeMemory(gpu.device, gpu.pinnedMem, nullptr);
     vkDestroyCommandPool(gpu.device, gpu.pool, nullptr);
     vkDestroyDevice(gpu.device, nullptr);
 }
@@ -266,35 +242,6 @@ BenchmarkResult runStagedBenchmark(Gpu &gpu, BenchmarkConfig config) {
     return {tH2D, tD2H};
 }
 
-BenchmarkResult runCachedBenchmark(Gpu &gpu, BenchmarkConfig config) {
-
-    // fill staging buffer
-    void *mapped;
-    vkMapMemory(gpu.device, gpu.cachedMem, 0, config.buffer_size, 0, &mapped);
-    std::memset(mapped, 0xAB, config.buffer_size);
-    vkUnmapMemory(gpu.device, gpu.cachedMem);
-
-    // warm-up, probably not significant
-    benchCopy(gpu.device, gpu.pool, gpu.queue, gpu.cachedBuf, 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.cachedBuf,
-                          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.cachedBuf, config.buffer_size);
-    tD2H /= config.iterations;
-
-    return {tH2D, tD2H};
-}
-
 auto main() -> int {
 
     VkApplicationInfo app{};
@@ -342,11 +289,6 @@ auto main() -> int {
         std::cout << "Running staged benchmark" << std::endl;
         BenchmarkResult res = runStagedBenchmark(gpu, config);
         reportBenchmark(res, gpu, config);
-
-        std::cout << "Running cached benchmark" << std::endl;
-        res = runCachedBenchmark(gpu, config);
-        reportBenchmark(res, gpu, config);
-
         std::cout << "--------------------" << std::endl;
     }