enable fp16s and int8s on newer adreno/mali, actually enable int8 tests

src/gpu.cpp

@@ -1392,12 +1392,12 @@ int create_gpu_instance()
 
             if (gpu_info.support_VK_KHR_8bit_storage)
             {
-                gpu_info.support_int8_storage = query8BitStorageFeatures.storageBuffer8BitAccess && query8BitStorageFeatures.uniformAndStorageBuffer8BitAccess;
+                gpu_info.support_int8_storage = query8BitStorageFeatures.storageBuffer8BitAccess;
             }
             if (gpu_info.support_VK_KHR_16bit_storage && queryFeatures.features.shaderStorageImageExtendedFormats)
             {
                 // shaderStorageImageExtendedFormats enables r16f format in storage image
-                gpu_info.support_fp16_storage = query16BitStorageFeatures.storageBuffer16BitAccess && query16BitStorageFeatures.uniformAndStorageBuffer16BitAccess;
+                gpu_info.support_fp16_storage = query16BitStorageFeatures.storageBuffer16BitAccess;
             }
             if (gpu_info.support_VK_KHR_shader_float16_int8)
             {
@@ -1416,7 +1416,7 @@ int create_gpu_instance()
             //             vkGetPhysicalDeviceFeatures(physicalDevice, &features);
         }
 
-        if (physicalDeviceProperties.vendorID == 0x13b5)
+        if (physicalDeviceProperties.vendorID == 0x13b5 && physicalDeviceProperties.apiVersion < VK_MAKE_VERSION(1, 0, 82))
         {
             // the 16bit_storage implementation of arm mali driver is buggy :[
             gpu_info.support_fp16_storage = false;
@@ -1830,7 +1830,7 @@ VulkanDevice::VulkanDevice(int device_index)
     enabled8BitStorageFeatures.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR;
     enabled8BitStorageFeatures.pNext = 0;
     enabled8BitStorageFeatures.storageBuffer8BitAccess = info.support_int8_storage();
-    enabled8BitStorageFeatures.uniformAndStorageBuffer8BitAccess = info.support_int8_storage();
+    enabled8BitStorageFeatures.uniformAndStorageBuffer8BitAccess = VK_FALSE;
     enabled8BitStorageFeatures.storagePushConstant8 = VK_FALSE;
     if (support_VK_KHR_get_physical_device_properties2 && info.support_VK_KHR_8bit_storage())
     {
@@ -1843,7 +1843,7 @@ VulkanDevice::VulkanDevice(int device_index)
     enabled16BitStorageFeatures.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES_KHR;
     enabled16BitStorageFeatures.pNext = 0;
     enabled16BitStorageFeatures.storageBuffer16BitAccess = info.support_fp16_storage();
-    enabled16BitStorageFeatures.uniformAndStorageBuffer16BitAccess = info.support_fp16_storage();
+    enabled16BitStorageFeatures.uniformAndStorageBuffer16BitAccess = VK_FALSE;
     enabled16BitStorageFeatures.storagePushConstant16 = VK_FALSE;
     enabled16BitStorageFeatures.storageInputOutput16 = VK_FALSE;
     if (support_VK_KHR_get_physical_device_properties2 && info.support_VK_KHR_16bit_storage())

tests/test_convolution.cpp

@@ -208,7 +208,7 @@ static int test_convolution_int8(int w, int h, int c, int outch, int kernel, int
         fprintf(stderr, "test_convolution_int8 failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d requant=%d\n", w, h, c, outch, kernel, dilation, stride, pad, bias, requant);
     }
 
-    return 0;
+    return ret;
 }
 
 static int test_convolution_1()

tests/test_innerproduct.cpp

@@ -119,7 +119,7 @@ static int test_innerproduct_int8(const ncnn::Mat& a, int outch, int bias)
         fprintf(stderr, "test_innerproduct_int8 failed a.dims=%d a=(%d %d %d) outch=%d bias=%d\n", a.dims, a.w, a.h, a.c, outch, bias);
     }
 
-    return 0;
+    return ret;
 }
 
 static int test_innerproduct_3()