From af6182900c4b8724ea22d6b67f93d63570673d31 Mon Sep 17 00:00:00 2001
From: TeachRaccooon <m.y.melnich@gmail.com>
Date: Fri, 2 Aug 2024 16:05:42 -0400
Subject: [PATCH] Update

---
 CMakeLists.txt                 |  1 +
 RandLAPACK/CMakeLists.txt      | 10 +++++-----
 test/CMakeLists.txt            |  2 +-
 test/drivers/test_cqrrp_gpu.cu | 14 ++++++++++----
 4 files changed, 17 insertions(+), 10 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 77fd3288..39bf2a95 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -7,6 +7,7 @@ set(CMAKE_CUDA_STANDARD 20)
 set(CMAKE_CUDA_STANDARD_REQUIRED ON)
 set(CMAKE_CUDA_SEPARABLE_COMPILATION ON)
 set(CMAKE_CUDA_VISIBILITY_PRESET hidden)
+set(CMAKE_CUDA_ARCHITECTURES native)
 
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/CMake")
 list(APPEND CMAKE_MODULE_PATH ${CMAKE_SOURCE_DIR})
diff --git a/RandLAPACK/CMakeLists.txt b/RandLAPACK/CMakeLists.txt
index 3957616a..055e4a12 100644
--- a/RandLAPACK/CMakeLists.txt
+++ b/RandLAPACK/CMakeLists.txt
@@ -25,21 +25,23 @@ set(RandLAPACK_cxx_sources
     rl_cqrrp_gpu.hh
 )
 
-set(RandLAPACK_libs
+add_library(RandLAPACK INTERFACE)
+
+target_link_libraries(RandLAPACK INTERFACE 
     RandBLAS
     lapackpp
     blaspp
     Random123
 )
+
 if (RandLAPACK_HAS_OpenMP)
-    list(APPEND RandLAPACK_libs OpenMP::OpenMP_CXX)
+    target_link_libraries(RandLAPACK INTERFACE OpenMP::OpenMP_CXX)
 endif()
 
 configure_file(${CMAKE_CURRENT_SOURCE_DIR}/rl_config.hh.in rl_config.hh)
 install(FILES ${CMAKE_CURRENT_BINARY_DIR}/rl_config.hh
     DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/RandLAPACK)
 
-add_library(RandLAPACK INTERFACE)
 
 set(RandLAPACK_cxx_opts -Wall -Wextra)
 if (SANITIZE_ADDRESS)
@@ -69,8 +71,6 @@ target_include_directories(
     $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/RandLAPACK/gpu_functions>
 )
 
-target_link_libraries(RandLAPACK INTERFACE ${RandLAPACK_libs})
-
 install(
     FILES "${CMAKE_CURRENT_SOURCE_DIR}/../RandLAPACK.hh"
     DESTINATION include
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 3d0a3e2b..b26309ba 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -30,7 +30,7 @@ if (GTest_FOUND)
     add_executable(RandLAPACK_tests_gpu ${RandLAPACK_test_cu_srcs})
     add_executable(RandLAPACK_tests ${RandLAPACK_test_srcs})
 
-    target_link_libraries(RandLAPACK_tests_gpu RandLAPACK GTest::GTest GTest::Main)
+    target_link_libraries(RandLAPACK_tests_gpu RandLAPACK GTest::GTest GTest::Main CUDA::cusolver)
     set_property(TARGET RandLAPACK_tests_gpu PROPERTY CUDA_ARCHITECTURES native)
 
     target_link_libraries(RandLAPACK_tests RandLAPACK GTest::GTest GTest::Main)
diff --git a/test/drivers/test_cqrrp_gpu.cu b/test/drivers/test_cqrrp_gpu.cu
index 067c2102..8fa529ea 100644
--- a/test/drivers/test_cqrrp_gpu.cu
+++ b/test/drivers/test_cqrrp_gpu.cu
@@ -10,6 +10,7 @@
 #include <RandBLAS.hh>
 #include <fstream>
 #include <gtest/gtest.h>
+#include <chrono>
 
 // Use cuda kernels.
 #ifndef USE_CUDA
@@ -281,12 +282,16 @@ class TestCQRRP : public ::testing::Test
         blas::gemm(Layout::ColMajor, Op::NoTrans, Op::NoTrans, d, n, m, 1.0, S, d, all_data.A.data(), m, 0.0, all_data.A_sk, d);
         free(S);
         cudaMemcpy(all_data.A_sk_device, all_data.A_sk, d * n * sizeof(double), cudaMemcpyHostToDevice);
-
+	
         RandLAPACK::CQRRP_blocked_GPU<double, r123::Philox4x32> CQRRP_GPU(true, tol, block_size);
+	auto start = std::chrono::steady_clock::now();
         CQRRP_GPU.call(m, n, all_data.A_device, m, all_data.A_sk_device, d, all_data.tau_device, all_data.J_device);
-
+	auto stop = std::chrono::steady_clock::now();
+	auto diff = std::chrono::duration_cast<std::chrono::milliseconds>(stop - start).count();
         auto rank = CQRRP_GPU.rank;
         printf("RANK AS RETURNED BY CQRRP GPU %4ld\n", rank);
+	printf("  BLOCK SIZE = %ld\n", block_size);
+	printf("  TIME (MS) = %ld\n", diff);
 
         cudaFree(all_data.A_sk_device);
         free(all_data.A_sk);
@@ -377,7 +382,8 @@ TEST_F(TestCQRRP, CQRRP_GPU_benchmark_16k) {
     int64_t n = std::pow(2, 14);
     double d_factor = 1.25;
     int64_t b_sz_start = 32;
-    int64_t b_sz_end   = 4096;
+    int64_t b_sz_end   = 192;
+    int64_t b_sz_incr  = 8;
     double tol = std::pow(std::numeric_limits<double>::epsilon(), 0.85);
     auto state = RandBLAS::RNGState();
 
@@ -387,7 +393,7 @@ TEST_F(TestCQRRP, CQRRP_GPU_benchmark_16k) {
     cudaMemcpy(all_data.A_device, all_data.A.data(), m * n * sizeof(double), cudaMemcpyHostToDevice);
 
 #if !defined(__APPLE__)
-    for (;b_sz_start <= b_sz_end; b_sz_start *= 2) {
+    for (;b_sz_start <= b_sz_end; b_sz_start += b_sz_incr) {
         bench_CQRRP(d_factor, tol, b_sz_start, all_data, state);
     }
 #endif