diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index 73658a5fa..a97050044 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -34,7 +34,7 @@ endif()
 if (NOT UNAME_M)
     execute_process(COMMAND uname -m OUTPUT_VARIABLE UNAME_M)
 endif()
-message(STATUS "UNAME_S: ${UNAME_S}  UNAME_P: ${UNAME_P}  UNAME_M: ${UNAME_M}")
+#message(STATUS "UNAME_S: ${UNAME_S}  UNAME_P: ${UNAME_P}  UNAME_M: ${UNAME_M}")
 
 # Mac OS + Arm can report x86_64
 # ref: https://github.com/ggerganov/whisper.cpp/issues/66#issuecomment-1282546789
@@ -68,7 +68,7 @@ else()
             set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mfma")
         endif()
         set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mf16c")
-elseif (UNAME_S MATCHES "Linux")
+    elseif (UNAME_S MATCHES "Linux")
         message(STATUS "Linux detected")
 	execute_process(COMMAND grep "avx " /proc/cpuinfo OUTPUT_VARIABLE AVX1_M)
         if (AVX1_M MATCHES "avx")
@@ -90,26 +90,24 @@ elseif (UNAME_S MATCHES "Linux")
 	if (SSE3_M MATCHES "sse3")
             set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -msse3")
         endif()
-	message(STATUS "CMAKE_C_FLAGS: ${CMAKE_C_FLAGS}")
-elseif (UNAME_S MATCHES "Haiku")
+  elseif (UNAME_S MATCHES "Haiku")
 	message(STATUS "Haiku detected")
-	execute_process(COMMAND bash -c "sysinfo -cpu | grep -w AVX" OUTPUT_VARIABLE AVX1_M)
-        if (AVX1_M MATCHES "AVX")
+	execute_process(COMMAND sysinfo -cpu COMMAND grep "AVX " OUTPUT_VARIABLE AVX1_M)
+        if (AVX1_M MATCHES "avx")
             set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx")
         endif()
-	execute_process(COMMAND bash -c "sysinfo -cpu | grep -w AVX2" OUTPUT_VARIABLE AVX2_M)
-        if (AVX2_M MATCHES "AVX2")
+	execute_process(COMMAND sysinfo -cpu COMMAND grep "AVX2 " OUTPUT_VARIABLE AVX2_M)
+        if (AVX2_M MATCHES "avx2")
             set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx2")
         endif()
-	execute_process(COMMAND bash -c "sysinfo -cpu | grep -w FMA" OUTPUT_VARIABLE FMA_M)
-	if (FMA_M MATCHES "FMA")
+	execute_process(COMMAND sysinfo -cpu COMMAND grep "FMA " OUTPUT_VARIABLE FMA_M)
+	if (FMA_M MATCHES "fma")
             set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mfma")
         endif()
-	execute_process(COMMAND bash -c "sysinfo -cpu | grep -w F16C" OUTPUT_VARIABLE F16C_M)
-	if (F16C_M MATCHES "F16C")
+	execute_process(COMMAND sysinfo -cpu COMMAND grep "F16C " OUTPUT_VARIABLE F16C_M)
+	if (F16C_M MATCHES "f16c")
             set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mf16c")
         endif()
-	message(STATUS "CMAKE_C_FLAGS: ${CMAKE_C_FLAGS}")
     else()
         set(CMAKE_C_FLAGS  "${CMAKE_C_FLAGS} -mfma -mf16c -mavx -mavx2")
     endif()
@@ -133,6 +131,31 @@ if (APPLE AND NOT GGML_NO_ACCELERATE)
     endif()
 endif()
 
+if (GGML_OPENBLAS)
+    set(OPENBLAS_INCLUDE_SEARCH_PATHS
+        /usr/include
+        /usr/include/openblas
+        /usr/include/openblas-base
+        /usr/local/include
+        /usr/local/include/openblas
+        /usr/local/include/openblas-base
+        /opt/OpenBLAS/include
+        $ENV{OpenBLAS_HOME}
+        $ENV{OpenBLAS_HOME}/include
+        )
+    find_path(OPENBLAS_INC NAMES cblas.h PATHS ${OPENBLAS_INCLUDE_SEARCH_PATHS})
+    find_library(OPENBLAS_LIB NAMES openblas libopenblas)
+    if (OPENBLAS_LIB)
+        message(STATUS "OpenBLAS found")
+
+        set(GGML_EXTRA_LIBS  ${GGML_EXTRA_LIBS}  ${OPENBLAS_LIB})
+        set(GGML_EXTRA_INCS  ${GGML_EXTRA_INCS}  ${OPENBLAS_INC})
+	set(GGML_EXTRA_FLAGS ${GGML_EXTRA_FLAGS} -DGGML_USE_OPENBLAS)
+    else()
+        message(WARNING "OpenBLAS not found")
+    endif()
+endif()
+
 if (GGML_PERF)
     set(GGML_EXTRA_FLAGS ${GGML_EXTRA_FLAGS} -DGGML_PERF)
 endif()
@@ -145,6 +168,7 @@ target_include_directories(${TARGET} PUBLIC
     .
     ../include
     ../include/ggml
+    ${GGML_EXTRA_INCS}
     )
 
 if (MSVC)
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
index 856cc3ed8..ac7039fea 100644
--- a/tests/CMakeLists.txt
+++ b/tests/CMakeLists.txt
@@ -1,3 +1,93 @@
+# check systems
+if (NOT UNAME_S)
+    execute_process(COMMAND uname -s OUTPUT_VARIABLE UNAME_S)
+endif()
+if (NOT UNAME_P)
+    execute_process(COMMAND uname -p OUTPUT_VARIABLE UNAME_P)
+endif()
+if (NOT UNAME_M)
+    execute_process(COMMAND uname -m OUTPUT_VARIABLE UNAME_M)
+endif()
+#message(STATUS "UNAME_S: ${UNAME_S}  UNAME_P: ${UNAME_P}  UNAME_M: ${UNAME_M}")
+
+# Mac OS + Arm can report x86_64
+# ref: https://github.com/ggerganov/whisper.cpp/issues/66#issuecomment-1282546789
+if (UNAME_S MATCHES "Darwin")
+    if (NOT UNAME_P MATCHES "arm")
+        execute_process(COMMAND sysctl -n hw.optional.arm64 OUTPUT_VARIABLE SYSCTL_M)
+        if (SYSCTL_M MATCHES "1")
+            #set(UNAME_P "arm")
+            #set(UNAME_M "arm64")
+            message(WARNING "Your arch is announced as x86_64, but it seems to actually be ARM64. Not fixing that can lea
+d to bad performance. For more info see: https://github.com/ggerganov/whisper.cpp/issues/66\#issuecomment-#1282546789")
+        endif()
+    endif()
+endif()
+
+if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64")
+    message(STATUS "ARM detected")
+    #set(GGML_C_FLAGS "${GGML_C_FLAGS} -mcpu=apple-m1")
+else()
+    message(STATUS "x86 detected")
+    #set(GGML_C_FLAGS "${GGML_C_FLAGS} -mavx -mavx2 -mfma -mf16c")
+    if (UNAME_S MATCHES "Darwin")
+        execute_process(COMMAND sysctl machdep.cpu.features OUTPUT_VARIABLE AVX1_M)
+        if (AVX1_M MATCHES "AVX1.0")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mavx")
+        endif()
+        execute_process(COMMAND sysctl machdep.cpu.leaf7_features OUTPUT_VARIABLE AVX2_M)
+        if (AVX2_M MATCHES "AVX2")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mavx2")
+        endif()
+        if (AVX1_M MATCHES "FMA")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mfma")
+        endif()
+        set(GGML_C_FLAGS "${GGML_C_FLAGS} -mf16c")
+    elseif (UNAME_S MATCHES "Linux")
+        message(STATUS "Linux detected")
+        execute_process(COMMAND grep "avx " /proc/cpuinfo OUTPUT_VARIABLE AVX1_M)
+        if (AVX1_M MATCHES "avx")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mavx")
+        endif()
+        execute_process(COMMAND grep "avx2 " /proc/cpuinfo OUTPUT_VARIABLE AVX2_M)
+        if (AVX2_M MATCHES "avx2")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mavx2")
+        endif()
+        execute_process(COMMAND grep "fma " /proc/cpuinfo OUTPUT_VARIABLE FMA_M)
+        if (FMA_M MATCHES "fma")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mfma")
+        endif()
+        execute_process(COMMAND grep "f16c " /proc/cpuinfo OUTPUT_VARIABLE F16C_M)
+        if (F16C_M MATCHES "f16c")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mf16c")
+        endif()
+        execute_process(COMMAND grep "sse3 " /proc/cpuinfo OUTPUT_VARIABLE SSE3_M)
+        if (SSE3_M MATCHES "sse3")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -msse3")
+        endif()
+    elseif (UNAME_S MATCHES "Haiku")
+        message(STATUS "Haiku detected")
+	execute_process(COMMAND sysinfo -cpu COMMAND grep "AVX " OUTPUT_VARIABLE AVX1_M)
+        if (AVX1_M MATCHES "avx")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mavx")
+        endif()
+	execute_process(COMMAND sysinfo -cpu COMMAND grep "AVX2 " OUTPUT_VARIABLE AVX2_M)
+        if (AVX2_M MATCHES "avx2")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mavx2")
+        endif()
+	execute_process(COMMAND sysinfo -cpu COMMAND grep "FMA " OUTPUT_VARIABLE FMA_M)
+        if (FMA_M MATCHES "fma")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mfma")
+        endif()
+	execute_process(COMMAND sysinfo -cpu COMMAND grep "F16C " OUTPUT_VARIABLE F16C_M)
+        if (F16C_M MATCHES "f16c")
+            set(GGML_C_FLAGS "${GGML_C_FLAGS} -mf16c")
+        endif()
+    else()
+        set(GGML_C_FLAGS  "${GGML_C_FLAGS} -mfma -mf16c -mavx -mavx2")
+    endif()
+endif()
+
 # on APPLE - include Accelerate framework
 if (APPLE AND NOT GGML_NO_ACCELERATE)
     find_library(ACCELERATE_FRAMEWORK Accelerate)
@@ -11,6 +101,31 @@ if (APPLE AND NOT GGML_NO_ACCELERATE)
     endif()
 endif()
 
+if (GGML_OPENBLAS)
+    set(OPENBLAS_INCLUDE_SEARCH_PATHS
+        /usr/include
+        /usr/include/openblas
+        /usr/include/openblas-base
+        /usr/local/include
+        /usr/local/include/openblas
+        /usr/local/include/openblas-base
+        /opt/OpenBLAS/include
+        $ENV{OpenBLAS_HOME}
+        $ENV{OpenBLAS_HOME}/include
+        )
+    find_path(OPENBLAS_INC NAMES cblas.h PATHS ${OPENBLAS_INCLUDE_SEARCH_PATHS})
+    find_library(OPENBLAS_LIB NAMES openblas libopenblas)
+    if (OPENBLAS_LIB)
+        message(STATUS "OpenBLAS found")
+
+        set(GGML_EXTRA_LIBS  ${GGML_EXTRA_LIBS}  ${OPENBLAS_LIB})
+        set(GGML_EXTRA_INCS  ${GGML_EXTRA_INCS}  ${OPENBLAS_INC})
+        set(GGML_EXTRA_FLAGS ${GGML_EXTRA_FLAGS} -DGGML_USE_OPENBLAS)
+    else()
+        message(WARNING "OpenBLAS not found")
+    endif()
+endif()
+
 #
 # test-vec0
 
@@ -26,7 +141,8 @@ if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "x86")
     add_executable(${TEST_TARGET} ${TEST_TARGET}.c)
     target_link_libraries(${TEST_TARGET} PRIVATE ggml)
     add_test(NAME ${TEST_TARGET} COMMAND $<TARGET_FILE:${TEST_TARGET}>)
-    set_target_properties(${TEST_TARGET} PROPERTIES COMPILE_FLAGS "-mavx -mavx2 -mfma -mf16c")
+    #set_target_properties(${TEST_TARGET} PROPERTIES COMPILE_FLAGS "-mavx -mavx2 -mfma -mf16c")
+    set_target_properties(${TEST_TARGET} PROPERTIES COMPILE_FLAGS ${GGML_C_FLAGS})
 endif()
 
 #
@@ -51,7 +167,8 @@ add_test(NAME ${TEST_TARGET} COMMAND $<TARGET_FILE:${TEST_TARGET}>)
 
 set(TEST_TARGET test-mul-mat0)
 add_executable(${TEST_TARGET} ${TEST_TARGET}.c)
-target_link_libraries(${TEST_TARGET} PRIVATE ggml)
+target_link_libraries(${TEST_TARGET} PRIVATE ggml ${GGML_EXTRA_LIBS})
+target_compile_options(${TEST_TARGET} PRIVATE ${GGML_EXTRA_FLAGS})
 add_test(NAME ${TEST_TARGET} COMMAND $<TARGET_FILE:${TEST_TARGET}>)
 
 #
@@ -106,7 +223,7 @@ target_link_libraries(${TEST_TARGET} PRIVATE ggml)
 add_test(NAME ${TEST_TARGET} COMMAND $<TARGET_FILE:${TEST_TARGET}>)
 
 #
-# test-svd0 (arm)
+# test-svd0 (arm/x86)
 
 if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm" AND NOT GGML_NO_ACCELERATE)
     set(TEST_TARGET test-svd0)
@@ -114,5 +231,11 @@ if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm" AND NOT GGML_NO_ACCELERATE)
     target_link_libraries(${TEST_TARGET} PRIVATE ggml ${GGML_EXTRA_LIBS})
     target_compile_options(${TEST_TARGET} PRIVATE ${GGML_EXTRA_FLAGS})
     add_test(NAME ${TEST_TARGET} COMMAND $<TARGET_FILE:${TEST_TARGET}>)
+elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "x86" AND GGML_OPENBLAS)
+    set(TEST_TARGET test-svd0)
+    add_executable(${TEST_TARGET} ${TEST_TARGET}.c)
+    target_link_libraries(${TEST_TARGET} PRIVATE ggml ${GGML_EXTRA_LIBS})
+    target_compile_options(${TEST_TARGET} PRIVATE ${GGML_EXTRA_FLAGS})
+    add_test(NAME ${TEST_TARGET} COMMAND $<TARGET_FILE:${TEST_TARGET}>)
 endif()
 
diff --git a/tests/test-grad0.c b/tests/test-grad0.c
index 4814b5929..2bbcf2554 100644
--- a/tests/test-grad0.c
+++ b/tests/test-grad0.c
@@ -210,7 +210,17 @@ int main(int argc, const char ** argv) {
 
     int ne[4];
 
-    for (int iter = 0; iter < 1000; ++iter) {
+    // original loop: 1000
+    int niter = 1000;
+    const char *env = getenv("GGML_NLOOP");
+    if (env != NULL) {
+        niter = atoi(env);
+    }
+    if (argc > 1) {
+        niter = atoi(argv[1]);
+    }
+    for (int iter = 0; iter < niter; ++iter) {
+        printf("test-grad0: iter:%d/%d\n", iter, niter);
         struct ggml_context * ctx0 = ggml_init(params);
 
         get_random_dims(ne, 4);
diff --git a/tests/test-mul-mat0.c b/tests/test-mul-mat0.c
index 1215c4ca0..2a367c8e1 100644
--- a/tests/test-mul-mat0.c
+++ b/tests/test-mul-mat0.c
@@ -232,7 +232,17 @@ int main(int argc, const char ** argv) {
 
     int ne[4];
 
-    for (int iter = 0; iter < 500; ++iter) {
+    // original loop: 500
+    int niter = 500;
+    const char *env = getenv("GGML_NLOOP");
+    if (env != NULL) {
+        niter = atoi(env);
+    }
+    if (argc > 1) {
+        niter = atoi(argv[1]);
+    }
+    for (int iter = 0; iter < niter; ++iter) {
+        printf("test-mul-mat0: iter:%d/%d\n", iter, niter);
         struct ggml_context * ctx0 = ggml_init(params);
 
         get_random_dims(ne, 4);
diff --git a/tests/test-vec1.c b/tests/test-vec1.c
index 850c62252..34688debd 100644
--- a/tests/test-vec1.c
+++ b/tests/test-vec1.c
@@ -97,10 +97,17 @@ void mul_mat_vec_f32_2(
             __m256 b1 = _mm256_loadu_ps(src1 + j + 8);
             __m256 b2 = _mm256_loadu_ps(src1 + j + 16);
             __m256 b3 = _mm256_loadu_ps(src1 + j + 24);
+#if defined(__FMA__)
             sum0 = _mm256_fmadd_ps(a0, b0, sum0);
             sum1 = _mm256_fmadd_ps(a1, b1, sum1);
             sum2 = _mm256_fmadd_ps(a2, b2, sum2);
             sum3 = _mm256_fmadd_ps(a3, b3, sum3);
+#else
+            sum0 = _mm256_add_ps(_mm256_mul_ps(a0, b0), sum0);
+            sum1 = _mm256_add_ps(_mm256_mul_ps(a1, b1), sum1);
+            sum2 = _mm256_add_ps(_mm256_mul_ps(a2, b2), sum2);
+            sum3 = _mm256_add_ps(_mm256_mul_ps(a3, b3), sum3);
+#endif
         }
         dst[i] = reduce_vector8_0(_mm256_add_ps(_mm256_add_ps(sum0, sum1), _mm256_add_ps(sum2, sum3)));
 
@@ -314,7 +321,11 @@ void mul_mat_vec_f16_0(
         for (int j = 0; j < ncols8; j += 8) {
             __m256 a = _mm256_cvtph_ps(_mm_loadu_si128((__m128i*)(src0_row + j)));
             __m256 b = _mm256_cvtph_ps(_mm_loadu_si128((__m128i*)(src1 + j)));
+#if defined(__FMA__)
             sum = _mm256_fmadd_ps(a, b, sum);
+#else
+            sum = _mm256_add_ps(_mm256_mul_ps(a, b), sum);
+#endif
         }
         dst[i] = reduce_vector8_0(sum);
 
@@ -343,8 +354,13 @@ void mul_mat_vec_f16_1(
             __m256 a1 = _mm256_cvtph_ps(_mm_loadu_si128((__m128i*)(src0_row + j + 8)));
             __m256 b0 = _mm256_cvtph_ps(_mm_loadu_si128((__m128i*)(src1 + j)));
             __m256 b1 = _mm256_cvtph_ps(_mm_loadu_si128((__m128i*)(src1 + j + 8)));
+#if defined(__FMA__)
             sum0 = _mm256_fmadd_ps(a0, b0, sum0);
             sum1 = _mm256_fmadd_ps(a1, b1, sum1);
+#else
+            sum0 = _mm256_add_ps(_mm256_mul_ps(a0, b0), sum0);
+            sum1 = _mm256_add_ps(_mm256_mul_ps(a1, b1), sum1);
+#endif
         }
         dst[i] = reduce_vector8_0(sum0) + reduce_vector8_0(sum1);
 
@@ -379,10 +395,17 @@ void mul_mat_vec_f16_2(
             __m256 b1 = _mm256_cvtph_ps(_mm_loadu_si128((__m128i*)(src1 + j + 8)));
             __m256 b2 = _mm256_cvtph_ps(_mm_loadu_si128((__m128i*)(src1 + j + 16)));
             __m256 b3 = _mm256_cvtph_ps(_mm_loadu_si128((__m128i*)(src1 + j + 24)));
+#if defined(__FMA__)
             sum0 = _mm256_fmadd_ps(a0, b0, sum0);
             sum1 = _mm256_fmadd_ps(a1, b1, sum1);
             sum2 = _mm256_fmadd_ps(a2, b2, sum2);
             sum3 = _mm256_fmadd_ps(a3, b3, sum3);
+#else
+            sum0 = _mm256_add_ps(_mm256_mul_ps(a0, b0), sum0);
+            sum1 = _mm256_add_ps(_mm256_mul_ps(a1, b1), sum1);
+            sum2 = _mm256_add_ps(_mm256_mul_ps(a2, b2), sum2);
+            sum3 = _mm256_add_ps(_mm256_mul_ps(a3, b3), sum3);
+#endif
         }
         dst[i] = reduce_vector8_0(sum0) + reduce_vector8_0(sum1) + reduce_vector8_0(sum2) + reduce_vector8_0(sum3);
 
@@ -417,10 +440,17 @@ void mul_mat_vec_f16_3(
             __m256 b1 = _mm256_loadu_ps(src1 + j + 8);
             __m256 b2 = _mm256_loadu_ps(src1 + j + 16);
             __m256 b3 = _mm256_loadu_ps(src1 + j + 24);
+#if defined(__FMA__)
             sum0 = _mm256_fmadd_ps(a0, b0, sum0);
             sum1 = _mm256_fmadd_ps(a1, b1, sum1);
             sum2 = _mm256_fmadd_ps(a2, b2, sum2);
             sum3 = _mm256_fmadd_ps(a3, b3, sum3);
+#else
+            sum0 = _mm256_add_ps(_mm256_mul_ps(a0, b0), sum0);
+            sum1 = _mm256_add_ps(_mm256_mul_ps(a1, b1), sum1);
+            sum2 = _mm256_add_ps(_mm256_mul_ps(a2, b2), sum2);
+            sum3 = _mm256_add_ps(_mm256_mul_ps(a3, b3), sum3);
+#endif
         }
         dst[i] = reduce_vector8_0(sum0) + reduce_vector8_0(sum1) + reduce_vector8_0(sum2) + reduce_vector8_0(sum3);
 
diff --git a/tests/test2.c b/tests/test2.c
index 6ec52ab28..70f915820 100644
--- a/tests/test2.c
+++ b/tests/test2.c
@@ -20,7 +20,17 @@ int main(int argc, const char ** argv) {
     struct ggml_opt_params opt_params = ggml_opt_default_params(GGML_OPT_ADAM);
     opt_params.adam.alpha = 0.01f;
 
-    opt_params.n_threads = (argc > 1) ? atoi(argv[1]) : 8;
+    // original threads: 8
+    int nthreads = 8;
+    const char *env = getenv("GGML_NTHREADS");
+    if (env != NULL) {
+        nthreads = atoi(env);
+    }
+    if (argc > 1) {
+        nthreads = atoi(argv[1]);
+    }
+    opt_params.n_threads = nthreads;
+    printf("test2: n_threads:%d\n", opt_params.n_threads);
 
     const float xi[] = {  1.0f,  2.0f,  3.0f,  4.0f,  5.0f , 6.0f,  7.0f,  8.0f,  9.0f,  10.0f, };
           float yi[] = { 15.0f, 25.0f, 35.0f, 45.0f, 55.0f, 65.0f, 75.0f, 85.0f, 95.0f, 105.0f, };