kompute/test/TestAsyncOperations.cpp

#include "gtest/gtest.h"

#include <chrono>

#include "kompute/Kompute.hpp"

TEST(TestAsyncOperations, TestManagerAsync)
{
    // This test is built for NVIDIA 1650. It assumes:
    // * Queue family 0 and 2 have compute capabilities
    // * GPU is able to process parallel shader code across different families
    uint32_t size = 10;

    uint32_t numParallel = 2;

    std::string shader(R"(
        #version 450

        layout (local_size_x = 1) in;

        layout(set = 0, binding = 0) buffer b { float pb[]; };

        shared uint sharedTotal[1];

        void main() {
            uint index = gl_GlobalInvocationID.x;

            sharedTotal[0] = 0;

            for (int i = 0; i < 100000000; i++)
            {
                atomicAdd(sharedTotal[0], 1);
            }

            pb[index] = sharedTotal[0];
        }
    )");

    std::vector<float> data(size, 0.0);
    std::vector<float> resultSync(size, 100000000);
    std::vector<float> resultAsync(size, 100000000);

    kp::Manager mgr;

    std::vector<std::shared_ptr<kp::Tensor>> inputsSyncB;

    for (uint32_t i = 0; i < numParallel; i++) {
        inputsSyncB.push_back(std::make_shared<kp::Tensor>(kp::Tensor(data)));
    }

    mgr.evalOpDefault<kp::OpTensorCreate>(inputsSyncB);

    auto startSync = std::chrono::high_resolution_clock::now();

    for (uint32_t i = 0; i < numParallel; i++) {
        mgr.evalOpDefault<kp::OpAlgoBase<>>(
          { inputsSyncB[i] },
          std::vector<char>(shader.begin(), shader.end()));
    }

    auto endSync = std::chrono::high_resolution_clock::now();
    auto durationSync =
      std::chrono::duration_cast<std::chrono::microseconds>(endSync - startSync)
        .count();

    mgr.evalOpDefault<kp::OpTensorSyncLocal>(inputsSyncB);

    for (uint32_t i = 0; i < numParallel; i++) {
        EXPECT_EQ(inputsSyncB[i]->data(), resultSync);
    }

    kp::Manager mgrAsync(0, { 0, 2 });

    std::vector<std::shared_ptr<kp::Tensor>> inputsAsyncB;

    for (uint32_t i = 0; i < numParallel; i++) {
        inputsAsyncB.push_back(std::make_shared<kp::Tensor>(kp::Tensor(data)));
    }

    mgrAsync.evalOpDefault<kp::OpTensorCreate>(inputsAsyncB);

    for (uint32_t i = 0; i < numParallel; i++) {
        mgrAsync.createManagedSequence("async" + std::to_string(i), i);
    }

    auto startAsync = std::chrono::high_resolution_clock::now();

    for (uint32_t i = 0; i < numParallel; i++) {
        mgrAsync.evalOpAsync<kp::OpAlgoBase<>>(
          { inputsAsyncB[i] },
          "async" + std::to_string(i),
          std::vector<char>(shader.begin(), shader.end()));
    }

    for (uint32_t i = 0; i < numParallel; i++) {
        mgrAsync.evalOpAwait("async" + std::to_string(i));
    }

    auto endAsync = std::chrono::high_resolution_clock::now();
    auto durationAsync = std::chrono::duration_cast<std::chrono::microseconds>(
                           endAsync - startAsync)
                           .count();

    mgrAsync.evalOpDefault<kp::OpTensorSyncLocal>({ inputsAsyncB });

    for (uint32_t i = 0; i < numParallel; i++) {
        EXPECT_EQ(inputsAsyncB[i]->data(), resultAsync);
    }

    SPDLOG_ERROR("sync {}", durationSync);
    SPDLOG_ERROR("async {}", durationAsync);

    // The speedup should be at least 40%
    EXPECT_LT(durationAsync, durationSync * 0.6);
}