unified image shader storage type (#2231)

* drop bug_layout_binding_id_alias flag

src/allocator.cpp

@@ -375,13 +375,13 @@ VkDeviceMemory VkAllocator::allocate_dedicated_memory(size_t size, uint32_t memo
     return memory;
 }
 
-VkImage VkAllocator::create_image(VkImageType type, int width, int height, int depth, VkFormat format, VkImageTiling tiling, VkImageUsageFlags usage)
+VkImage VkAllocator::create_image(int width, int height, int depth, VkFormat format, VkImageTiling tiling, VkImageUsageFlags usage)
 {
     VkImageCreateInfo imageCreateInfo;
     imageCreateInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
     imageCreateInfo.pNext = 0;
     imageCreateInfo.flags = 0;
-    imageCreateInfo.imageType = type;
+    imageCreateInfo.imageType = VK_IMAGE_TYPE_3D;
     imageCreateInfo.format = format;
     imageCreateInfo.extent.width = width;
     imageCreateInfo.extent.height = height;
@@ -400,21 +400,21 @@ VkImage VkAllocator::create_image(VkImageType type, int width, int height, int d
     VkResult ret = vkCreateImage(vkdev->vkdevice(), &imageCreateInfo, 0, &image);
     if (ret != VK_SUCCESS)
     {
-        NCNN_LOGE("vkCreateImage failed %d %d %d %d %d %d %d %d", ret, type, width, height, depth, format, tiling, usage);
+        NCNN_LOGE("vkCreateImage failed %d %d %d %d %d %d %d", ret, width, height, depth, format, tiling, usage);
         return 0;
     }
 
     return image;
 }
 
-VkImageView VkAllocator::create_imageview(VkImageViewType type, VkImage image, VkFormat format)
+VkImageView VkAllocator::create_imageview(VkImage image, VkFormat format)
 {
     VkImageViewCreateInfo imageViewCreateInfo;
     imageViewCreateInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
     imageViewCreateInfo.pNext = 0;
     imageViewCreateInfo.flags = 0;
     imageViewCreateInfo.image = image;
-    imageViewCreateInfo.viewType = type;
+    imageViewCreateInfo.viewType = VK_IMAGE_VIEW_TYPE_3D;
     imageViewCreateInfo.format = format;
     imageViewCreateInfo.components.r = VK_COMPONENT_SWIZZLE_IDENTITY;
     imageViewCreateInfo.components.g = VK_COMPONENT_SWIZZLE_IDENTITY;
@@ -709,7 +709,7 @@ void VkBlobAllocator::fastFree(VkBufferMemory* ptr)
     delete ptr;
 }
 
-VkImageMemory* VkBlobAllocator::fastMalloc(int dims, int w, int h, int c, size_t elemsize, int elempack)
+VkImageMemory* VkBlobAllocator::fastMalloc(int w, int h, int c, size_t elemsize, int elempack)
 {
     if (elempack != 1 && elempack != 4 && elempack != 8)
     {
@@ -743,48 +743,16 @@ VkImageMemory* VkBlobAllocator::fastMalloc(int dims, int w, int h, int c, size_t
     // large elempack spills on image w
     if (elempack == 8) width *= 2;
 
-    VkImageType image_type;
-    VkImageViewType imageview_type;
-    if (dims == 1)
+    if (width > (int)vkdev->info.max_image_dimension_3d || height > (int)vkdev->info.max_image_dimension_3d || depth > (int)vkdev->info.max_image_dimension_3d)
     {
-        image_type = VK_IMAGE_TYPE_1D;
-        imageview_type = VK_IMAGE_VIEW_TYPE_1D;
-
-        if (width > (int)vkdev->info.max_image_dimension_1d)
-        {
-            NCNN_LOGE("image dimension too large %d > %d", width, (int)vkdev->info.max_image_dimension_1d);
-            return 0;
-        }
-    }
-    else if (dims == 2)
-    {
-        image_type = VK_IMAGE_TYPE_2D;
-        imageview_type = VK_IMAGE_VIEW_TYPE_2D;
-
-        if (width > (int)vkdev->info.max_image_dimension_2d || height > (int)vkdev->info.max_image_dimension_2d)
-        {
-            NCNN_LOGE("image dimension too large %d %d > %d", width, height, (int)vkdev->info.max_image_dimension_2d);
-            return 0;
-        }
-    }
-    else // if (dims == 3)
-    {
-        image_type = VK_IMAGE_TYPE_3D;
-        imageview_type = VK_IMAGE_VIEW_TYPE_3D;
-
-        if (width > (int)vkdev->info.max_image_dimension_3d || height > (int)vkdev->info.max_image_dimension_3d || depth > (int)vkdev->info.max_image_dimension_3d)
-        {
-            NCNN_LOGE("image dimension too large %d %d %d > %d", width, height, depth, (int)vkdev->info.max_image_dimension_3d);
-            return 0;
-        }
+        NCNN_LOGE("image dimension too large %d %d %d > %d", width, height, depth, (int)vkdev->info.max_image_dimension_3d);
+        return 0;
     }
 
     VkImageMemory* ptr = new VkImageMemory;
 
-    ptr->image = create_image(image_type, width, height, depth, format, VK_IMAGE_TILING_OPTIMAL, VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_STORAGE_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT);
+    ptr->image = create_image(width, height, depth, format, VK_IMAGE_TILING_OPTIMAL, VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_STORAGE_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT);
 
-    ptr->image_type = image_type;
-    ptr->imageview_type = imageview_type;
     ptr->width = width;
     ptr->height = height;
     ptr->depth = depth;
@@ -827,7 +795,7 @@ VkImageMemory* VkBlobAllocator::fastMalloc(int dims, int w, int h, int c, size_t
             // do not allow host access to optimal tiling image
             ptr->mapped_ptr = 0;
 
-            ptr->imageview = create_imageview(imageview_type, ptr->image, format);
+            ptr->imageview = create_imageview(ptr->image, format);
 
             ptr->access_flags = 0;
             ptr->image_layout = VK_IMAGE_LAYOUT_UNDEFINED;
@@ -894,7 +862,7 @@ VkImageMemory* VkBlobAllocator::fastMalloc(int dims, int w, int h, int c, size_t
     // do not allow host access to optimal tiling image
     ptr->mapped_ptr = 0;
 
-    ptr->imageview = create_imageview(imageview_type, ptr->image, format);
+    ptr->imageview = create_imageview(ptr->image, format);
 
     ptr->access_flags = 0;
     ptr->image_layout = VK_IMAGE_LAYOUT_UNDEFINED;
@@ -1244,7 +1212,7 @@ void VkWeightAllocator::fastFree(VkBufferMemory* ptr)
     delete ptr;
 }
 
-VkImageMemory* VkWeightAllocator::fastMalloc(int dims, int w, int h, int c, size_t elemsize, int elempack)
+VkImageMemory* VkWeightAllocator::fastMalloc(int w, int h, int c, size_t elemsize, int elempack)
 {
     if (elempack != 1 && elempack != 4 && elempack != 8 && elempack != 16 && elempack != 32 && elempack != 64)
     {
@@ -1287,48 +1255,16 @@ VkImageMemory* VkWeightAllocator::fastMalloc(int dims, int w, int h, int c, size
     if (elempack == 32) width *= 8;
     if (elempack == 64) width *= 16;
 
-    VkImageType image_type;
-    VkImageViewType imageview_type;
-    if (dims == 1)
+    if (width > (int)vkdev->info.max_image_dimension_3d || height > (int)vkdev->info.max_image_dimension_3d || depth > (int)vkdev->info.max_image_dimension_3d)
     {
-        image_type = VK_IMAGE_TYPE_1D;
-        imageview_type = VK_IMAGE_VIEW_TYPE_1D;
-
-        if (width > (int)vkdev->info.max_image_dimension_1d)
-        {
-            NCNN_LOGE("image dimension too large %d > %d", width, (int)vkdev->info.max_image_dimension_1d);
-            return 0;
-        }
-    }
-    else if (dims == 2)
-    {
-        image_type = VK_IMAGE_TYPE_2D;
-        imageview_type = VK_IMAGE_VIEW_TYPE_2D;
-
-        if (width > (int)vkdev->info.max_image_dimension_2d || height > (int)vkdev->info.max_image_dimension_2d)
-        {
-            NCNN_LOGE("image dimension too large %d %d > %d", width, height, (int)vkdev->info.max_image_dimension_2d);
-            return 0;
-        }
-    }
-    else // if (dims == 3)
-    {
-        image_type = VK_IMAGE_TYPE_3D;
-        imageview_type = VK_IMAGE_VIEW_TYPE_3D;
-
-        if (width > (int)vkdev->info.max_image_dimension_3d || height > (int)vkdev->info.max_image_dimension_3d || depth > (int)vkdev->info.max_image_dimension_3d)
-        {
-            NCNN_LOGE("image dimension too large %d %d %d > %d", width, height, depth, (int)vkdev->info.max_image_dimension_3d);
-            return 0;
-        }
+        NCNN_LOGE("image dimension too large %d %d %d > %d", width, height, depth, (int)vkdev->info.max_image_dimension_3d);
+        return 0;
     }
 
     VkImageMemory* ptr = new VkImageMemory;
 
-    ptr->image = create_image(image_type, width, height, depth, format, VK_IMAGE_TILING_OPTIMAL, VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT);
+    ptr->image = create_image(width, height, depth, format, VK_IMAGE_TILING_OPTIMAL, VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT);
 
-    ptr->image_type = image_type;
-    ptr->imageview_type = imageview_type;
     ptr->width = width;
     ptr->height = height;
     ptr->depth = depth;
@@ -1385,7 +1321,7 @@ VkImageMemory* VkWeightAllocator::fastMalloc(int dims, int w, int h, int c, size
             // do not allow host access to optimal tiling image
             ptr->mapped_ptr = 0;
 
-            ptr->imageview = create_imageview(imageview_type, ptr->image, format);
+            ptr->imageview = create_imageview(ptr->image, format);
 
             ptr->access_flags = 0;
             ptr->image_layout = VK_IMAGE_LAYOUT_UNDEFINED;
@@ -1426,7 +1362,7 @@ VkImageMemory* VkWeightAllocator::fastMalloc(int dims, int w, int h, int c, size
             // do not allow host access to optimal tiling image
             ptr->mapped_ptr = 0;
 
-            ptr->imageview = create_imageview(imageview_type, ptr->image, format);
+            ptr->imageview = create_imageview(ptr->image, format);
 
             ptr->access_flags = 0;
             ptr->image_layout = VK_IMAGE_LAYOUT_UNDEFINED;
@@ -1482,7 +1418,7 @@ VkImageMemory* VkWeightAllocator::fastMalloc(int dims, int w, int h, int c, size
     // do not allow host access to optimal tiling image
     ptr->mapped_ptr = 0;
 
-    ptr->imageview = create_imageview(imageview_type, ptr->image, format);
+    ptr->imageview = create_imageview(ptr->image, format);
 
     ptr->access_flags = 0;
     ptr->image_layout = VK_IMAGE_LAYOUT_UNDEFINED;
@@ -1612,36 +1548,16 @@ void VkStagingAllocator::fastFree(VkBufferMemory* ptr)
     buffer_budgets.push_back(ptr);
 }
 
-VkImageMemory* VkStagingAllocator::fastMalloc(int dims, int w, int h, int c, size_t elemsize, int /* elempack */)
+VkImageMemory* VkStagingAllocator::fastMalloc(int w, int h, int c, size_t elemsize, int /* elempack */)
 {
     // staging image is mainly used for storing small piece of dynamic parameters
     // we allocate host memory as a fake image, it's simple and good
 
     const size_t size = w * h * c * elemsize;
 
-    VkImageType image_type;
-    VkImageViewType imageview_type;
-    if (dims == 1)
-    {
-        image_type = VK_IMAGE_TYPE_1D;
-        imageview_type = VK_IMAGE_VIEW_TYPE_1D;
-    }
-    else if (dims == 2)
-    {
-        image_type = VK_IMAGE_TYPE_2D;
-        imageview_type = VK_IMAGE_VIEW_TYPE_2D;
-    }
-    else // if (dims == 3)
-    {
-        image_type = VK_IMAGE_TYPE_3D;
-        imageview_type = VK_IMAGE_VIEW_TYPE_3D;
-    }
-
     VkImageMemory* ptr = new VkImageMemory;
 
     ptr->image = 0;
-    ptr->image_type = image_type;
-    ptr->imageview_type = imageview_type;
     ptr->width = w;
     ptr->height = h;
     ptr->depth = c;
@@ -1746,7 +1662,7 @@ VkAndroidHardwareBufferImageAllocator::~VkAndroidHardwareBufferImageAllocator()
     }
 }
 
-VkImageMemory* VkAndroidHardwareBufferImageAllocator::fastMalloc(int /*dims*/, int /*w*/, int /*h*/, int /*c*/, size_t /*elemsize*/, int /*elempack*/)
+VkImageMemory* VkAndroidHardwareBufferImageAllocator::fastMalloc(int /*w*/, int /*h*/, int /*c*/, size_t /*elemsize*/, int /*elempack*/)
 {
     VkResult ret;
 

src/allocator.h

@@ -224,8 +224,6 @@ public:
     VkImageView imageview;
 
     // underlying info assigned by allocator
-    VkImageType image_type;
-    VkImageViewType imageview_type;
     int width;
     int height;
     int depth;
@@ -267,7 +265,7 @@ public:
     virtual int flush(VkBufferMemory* ptr);
     virtual int invalidate(VkBufferMemory* ptr);
 
-    virtual VkImageMemory* fastMalloc(int dims, int w, int h, int c, size_t elemsize, int elempack) = 0;
+    virtual VkImageMemory* fastMalloc(int w, int h, int c, size_t elemsize, int elempack) = 0;
     virtual void fastFree(VkImageMemory* ptr) = 0;
 
 public:
@@ -282,8 +280,8 @@ protected:
     VkDeviceMemory allocate_memory(size_t size, uint32_t memory_type_index);
     VkDeviceMemory allocate_dedicated_memory(size_t size, uint32_t memory_type_index, VkImage image, VkBuffer buffer);
 
-    VkImage create_image(VkImageType type, int width, int height, int depth, VkFormat format, VkImageTiling tiling, VkImageUsageFlags usage);
-    VkImageView create_imageview(VkImageViewType type, VkImage image, VkFormat format);
+    VkImage create_image(int width, int height, int depth, VkFormat format, VkImageTiling tiling, VkImageUsageFlags usage);
+    VkImageView create_imageview(VkImage image, VkFormat format);
 };
 
 class VkBlobAllocator : public VkAllocator
@@ -298,7 +296,7 @@ public:
 
     virtual VkBufferMemory* fastMalloc(size_t size);
     virtual void fastFree(VkBufferMemory* ptr);
-    virtual VkImageMemory* fastMalloc(int dims, int w, int h, int c, size_t elemsize, int elempack);
+    virtual VkImageMemory* fastMalloc(int w, int h, int c, size_t elemsize, int elempack);
     virtual void fastFree(VkImageMemory* ptr);
 
 protected:
@@ -324,7 +322,7 @@ public:
 public:
     virtual VkBufferMemory* fastMalloc(size_t size);
     virtual void fastFree(VkBufferMemory* ptr);
-    virtual VkImageMemory* fastMalloc(int dims, int w, int h, int c, size_t elemsize, int elempack);
+    virtual VkImageMemory* fastMalloc(int w, int h, int c, size_t elemsize, int elempack);
     virtual void fastFree(VkImageMemory* ptr);
 
 protected:
@@ -355,7 +353,7 @@ public:
 
     virtual VkBufferMemory* fastMalloc(size_t size);
     virtual void fastFree(VkBufferMemory* ptr);
-    virtual VkImageMemory* fastMalloc(int dims, int w, int h, int c, size_t elemsize, int elempack);
+    virtual VkImageMemory* fastMalloc(int w, int h, int c, size_t elemsize, int elempack);
     virtual void fastFree(VkImageMemory* ptr);
 
 protected:
@@ -372,7 +370,7 @@ public:
 public:
     virtual VkBufferMemory* fastMalloc(size_t size);
     virtual void fastFree(VkBufferMemory* ptr);
-    virtual VkImageMemory* fastMalloc(int /*dims*/, int /*w*/, int /*h*/, int /*c*/, size_t /*elemsize*/, int /*elempack*/)
+    virtual VkImageMemory* fastMalloc(int /*w*/, int /*h*/, int /*c*/, size_t /*elemsize*/, int /*elempack*/)
     {
         return 0;
     }
@@ -392,7 +390,7 @@ public:
     virtual ~VkAndroidHardwareBufferImageAllocator();
 
 public:
-    virtual VkImageMemory* fastMalloc(int dims, int w, int h, int c, size_t elemsize, int elempack);
+    virtual VkImageMemory* fastMalloc(int w, int h, int c, size_t elemsize, int elempack);
     virtual void fastFree(VkImageMemory* ptr);
     virtual VkBufferMemory* fastMalloc(size_t /*size*/)
     {

src/gpu.cpp

@@ -641,16 +641,9 @@ int create_gpu_instance()
         // 650 = 0x5143 0x6050002
 
         gpu_info.bug_storage_buffer_no_l1 = false;
-        gpu_info.bug_layout_binding_id_alias = false;
         gpu_info.bug_corrupted_online_pipeline_cache = false;
         gpu_info.bug_implicit_fp16_arithmetic = false;
 
-        if (physicalDeviceProperties.vendorID == 0x5143 && physicalDeviceProperties.apiVersion < VK_MAKE_VERSION(1, 0, 49))
-        {
-            // qcom adreno with old buggy driver cannot handle binding id alias
-            gpu_info.bug_layout_binding_id_alias = true;
-        }
-
         if (physicalDeviceProperties.vendorID == 0x5143 && physicalDeviceProperties.apiVersion < VK_MAKE_VERSION(1, 0, 66))
         {
             // qcom adreno with old buggy driver cannot share created pipeline properly
@@ -668,22 +661,6 @@ int create_gpu_instance()
             //             gpu_info.bug_storage_buffer_no_l1 = true;
         }
 
-        if (physicalDeviceProperties.vendorID == 0x13b5
-                && (physicalDeviceProperties.deviceID == 0x7500001
-                    || physicalDeviceProperties.deviceID == 0x8602000
-                    || physicalDeviceProperties.deviceID == 0x8800020))
-        {
-            // these arm mali midgard era driver cannot handle binding id alias
-            gpu_info.bug_layout_binding_id_alias = true;
-        }
-
-#if __APPLE__
-        {
-            // metal shader never accept binding id alias
-            gpu_info.bug_layout_binding_id_alias = true;
-        }
-#endif
-
         if (physicalDeviceProperties.vendorID == 0x13b5
                 && (physicalDeviceProperties.deviceID == 0x7500001
                     || physicalDeviceProperties.deviceID == 0x8602000
@@ -997,8 +974,8 @@ int create_gpu_instance()
                   gpu_info.graphics_queue_family_index, gpu_info.graphics_queue_count,
                   gpu_info.transfer_queue_family_index, gpu_info.transfer_queue_count);
 
-        NCNN_LOGE("[%u %s]  bugsbn1=%d  buglbia=%d  bugcopc=%d  bugihfa=%d", i, physicalDeviceProperties.deviceName,
-                  gpu_info.bug_storage_buffer_no_l1, gpu_info.bug_layout_binding_id_alias, gpu_info.bug_corrupted_online_pipeline_cache, gpu_info.bug_implicit_fp16_arithmetic);
+        NCNN_LOGE("[%u %s]  bugsbn1=%d  bugcopc=%d  bugihfa=%d", i, physicalDeviceProperties.deviceName,
+                  gpu_info.bug_storage_buffer_no_l1, gpu_info.bug_corrupted_online_pipeline_cache, gpu_info.bug_implicit_fp16_arithmetic);
 
         NCNN_LOGE("[%u %s]  fp16p=%d  fp16s=%d  fp16a=%d  int8s=%d  int8a=%d", i, physicalDeviceProperties.deviceName,
                   gpu_info.support_fp16_packed, gpu_info.support_fp16_storage, gpu_info.support_fp16_arithmetic,
@@ -2031,12 +2008,6 @@ void VulkanDevice::convert_packing(const VkMat& src, VkMat& dst, int dst_elempac
 
 void VulkanDevice::convert_packing(const VkImageMat& src, VkImageMat& dst, int dst_elempack, VkCompute& cmd, const Option& opt) const
 {
-    if (info.bug_layout_binding_id_alias)
-    {
-        NCNN_LOGE("cannot convert_packing i2i");
-        return;
-    }
-
     int cast_type_to_index = opt.use_fp16_storage ? 2 : opt.use_fp16_packed ? 1 : 0;
     int packing_type_to_index = dst_elempack == 1 ? 0 : dst_elempack == 4 ? 1 : 2;
 
@@ -2069,12 +2040,6 @@ void VulkanDevice::convert_packing(const VkImageMat& src, VkImageMat& dst, int d
 
 void VulkanDevice::convert_packing(const VkMat& src, VkImageMat& dst, int dst_elempack, VkCompute& cmd, const Option& opt) const
 {
-    if (info.bug_layout_binding_id_alias)
-    {
-        NCNN_LOGE("cannot convert_packing b2i");
-        return;
-    }
-
     int cast_type_to_index = opt.use_fp16_storage ? 2 : opt.use_fp16_packed ? 1 : 0;
     int packing_type_to_index = dst_elempack == 1 ? 0 : dst_elempack == 4 ? 1 : 2;
 
@@ -2107,12 +2072,6 @@ void VulkanDevice::convert_packing(const VkMat& src, VkImageMat& dst, int dst_el
 
 void VulkanDevice::convert_packing(const VkImageMat& src, VkMat& dst, int dst_elempack, VkCompute& cmd, const Option& opt) const
 {
-    if (info.bug_layout_binding_id_alias)
-    {
-        NCNN_LOGE("cannot convert_packing i2b");
-        return;
-    }
-
     int cast_type_to_index = opt.use_fp16_storage ? 2 : opt.use_fp16_packed ? 1 : 0;
     int packing_type_to_index = dst_elempack == 1 ? 0 : dst_elempack == 4 ? 1 : 2;
 
@@ -2359,12 +2318,6 @@ const ncnn::Packing_vulkan* VulkanDevice::get_utility_operator(int storage_type_
     opt.use_fp16_packed = (cast_type_from_index == 1 || cast_type_to_index == 1);
     opt.use_fp16_storage = (cast_type_from_index == 2 || cast_type_to_index == 2);
 
-    if (info.bug_layout_binding_id_alias && opt.use_image_storage)
-    {
-        NCNN_LOGE("cannot create uop with use_image_storage if bug_layout_binding_id_alias");
-        return 0;
-    }
-
     if (!info.support_fp16_packed && opt.use_fp16_packed)
     {
         NCNN_LOGE("cannot create uop with use_fp16_packed if not support_fp16_packed");
@@ -2424,8 +2377,6 @@ void VulkanDevice::destroy_utility_operator()
         for (int i1 = 0; i1 < 2; i1++)
         {
             opt.use_image_storage = (i0 == 1 || i1 == 1);
-            if (info.bug_layout_binding_id_alias && opt.use_image_storage)
-                continue;
 
             // from fp32-b/i | fp16p-b/i | fp16s-b/i
             // to fp32-b/i | fp16p-b/i | fp16s-b/i

src/gpu.h

@@ -129,7 +129,6 @@ public:
 
     // bug is not feature
     bool bug_storage_buffer_no_l1;
-    bool bug_layout_binding_id_alias;
     bool bug_corrupted_online_pipeline_cache;
 
     // but sometimes bug is a feature

src/layer/vulkan/shader/absval.comp

@@ -29,11 +29,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfp bottom_top_blob_data[]; };
@@ -58,19 +54,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afp v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld1(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld1(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afp v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -80,18 +64,7 @@ void main()
     v = abs(v);
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st1(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st1(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st1(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/absval_pack4.comp

@@ -29,11 +29,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec4 bottom_top_blob_data[]; };
@@ -58,19 +54,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec4 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld4(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld4(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec4 v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -80,18 +64,7 @@ void main()
     v = abs(v);
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st4(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st4(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st4(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/absval_pack8.comp

@@ -30,11 +30,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec8 bottom_top_blob_data[]; };
@@ -59,19 +55,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec8 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld8(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld8(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -82,18 +66,7 @@ void main()
     v[1] = abs(v[1]);
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st8(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st8(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st8(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/batchnorm.comp

@@ -29,14 +29,10 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
-layout (binding = 2) uniform unfp sampler1D a;
-layout (binding = 3) uniform unfp sampler1D b;
+layout (binding = 2) uniform unfp sampler3D a;
+layout (binding = 3) uniform unfp sampler3D b;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfp bottom_top_blob_data[]; };
 layout (binding = 1) readonly buffer a { sfp a_data[]; };
@@ -62,19 +58,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afp v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld1(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld1(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afp v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -85,8 +69,8 @@ void main()
     const int abi = gxyz[psc(dims) - 1];
 
 #if NCNN_image_shader
-    afp b = image1d_ld1(b, abi);
-    afp a = image1d_ld1(a, abi);
+    afp b = image3d_ld1(b, ivec3(abi, 0, 0));
+    afp a = image3d_ld1(a, ivec3(abi, 0, 0));
 #else
     afp b = buffer_ld1(b_data, abi);
     afp a = buffer_ld1(a_data, abi);
@@ -95,18 +79,7 @@ void main()
     v = b * v + a;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st1(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st1(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st1(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/batchnorm_pack4.comp

@@ -29,14 +29,10 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
-layout (binding = 2) uniform unfp sampler1D a;
-layout (binding = 3) uniform unfp sampler1D b;
+layout (binding = 2) uniform unfp sampler3D a;
+layout (binding = 3) uniform unfp sampler3D b;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec4 bottom_top_blob_data[]; };
 layout (binding = 1) readonly buffer a { sfpvec4 a_data[]; };
@@ -62,19 +58,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec4 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld4(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld4(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec4 v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -85,8 +69,8 @@ void main()
     const int abi = gxyz[psc(dims) - 1];
 
 #if NCNN_image_shader
-    afpvec4 b = image1d_ld4(b, abi);
-    afpvec4 a = image1d_ld4(a, abi);
+    afpvec4 b = image3d_ld4(b, ivec3(abi, 0, 0));
+    afpvec4 a = image3d_ld4(a, ivec3(abi, 0, 0));
 #else
     afpvec4 b = buffer_ld4(b_data, abi);
     afpvec4 a = buffer_ld4(a_data, abi);
@@ -95,18 +79,7 @@ void main()
     v = b * v + a;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st4(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st4(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st4(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/batchnorm_pack8.comp

@@ -30,14 +30,10 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
-layout (binding = 2) uniform unfp sampler1D a;
-layout (binding = 3) uniform unfp sampler1D b;
+layout (binding = 2) uniform unfp sampler3D a;
+layout (binding = 3) uniform unfp sampler3D b;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec8 bottom_top_blob_data[]; };
 layout (binding = 1) readonly buffer a { sfpvec8 a_data[]; };
@@ -63,19 +59,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec8 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld8(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld8(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -86,8 +70,8 @@ void main()
     const int abi = gxyz[psc(dims) - 1];
 
 #if NCNN_image_shader
-    afpvec8 b = image1d_ld8(b, abi);
-    afpvec8 a = image1d_ld8(a, abi);
+    afpvec8 b = image3d_ld8(b, ivec3(abi, 0, 0));
+    afpvec8 a = image3d_ld8(a, ivec3(abi, 0, 0));
 #else
     afpvec8 b = buffer_ld8(b_data, abi);
     afpvec8 a = buffer_ld8(a_data, abi);
@@ -97,18 +81,7 @@ void main()
     v[1] = b[1] * v[1] + a[1];
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st8(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st8(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st8(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/binaryop.comp

@@ -45,14 +45,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
-layout (binding = 0) uniform unfp sampler2D a_blob_2d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
-layout (binding = 1) uniform unfp sampler2D b_blob_2d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer a_blob { sfp a_blob_data[]; };
@@ -91,20 +85,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afp v1;
-
-    if (psc(outdims) == 1)
-    {
-        v1 = image1d_ld1(a_blob_1d, gx);
-    }
-    else if (psc(outdims) == 2)
-    {
-        v1 = image2d_ld1(a_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(outdims) == 3)
-    {
-        v1 = image3d_ld1(a_blob_3d, ivec3(gx, gy, gz));
-    }
+    afp v1 = image3d_ld1(a_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -129,18 +110,7 @@ void main()
         if (op_type == 8) res = b / v1;
 
 #if NCNN_image_shader
-        if (psc(outdims) == 1)
-        {
-            image1d_st1(top_blob_1d, gx, res);
-        }
-        else if (psc(outdims) == 2)
-        {
-            image2d_st1(top_blob_2d, ivec2(gx, gy), res);
-        }
-        else // if (psc(outdims) == 3)
-        {
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz), res);
-        }
+        image3d_st1(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
         buffer_st1(a_blob_data, gi, res);
 #endif
@@ -149,20 +119,7 @@ void main()
     {
         // type 7 13 19
 #if NCNN_image_shader
-        afp v2;
-
-        if (psc(outdims) == 1)
-        {
-            v2 = image1d_ld1(b_blob_1d, gx);
-        }
-        else if (psc(outdims) == 2)
-        {
-            v2 = image2d_ld1(b_blob_2d, ivec2(gx, gy));
-        }
-        else // if (psc(outdims) == 3)
-        {
-            v2 = image3d_ld1(b_blob_3d, ivec3(gx, gy, gz));
-        }
+        afp v2 = image3d_ld1(b_blob_3d, ivec3(gx, gy, gz));
 #else
         afp v2 = buffer_ld1(b_blob_data, gi);
 #endif
@@ -178,18 +135,7 @@ void main()
         if (op_type == 8) res = v2 / v1;
 
 #if NCNN_image_shader
-        if (psc(outdims) == 1)
-        {
-            image1d_st1(top_blob_1d, gx, res);
-        }
-        else if (psc(outdims) == 2)
-        {
-            image2d_st1(top_blob_2d, ivec2(gx, gy), res);
-        }
-        else // if (psc(outdims) == 3)
-        {
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz), res);
-        }
+        image3d_st1(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
         buffer_st1(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/binaryop_broadcast.comp

@@ -43,14 +43,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
-layout (binding = 0) uniform unfp sampler2D a_blob_2d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
-layout (binding = 1) uniform unfp sampler2D b_blob_2d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer a_blob { sfp a_blob_data[]; };
@@ -89,8 +83,12 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afp v1;
-    afp v2;
+    int ax = gx;
+    int ay = gy;
+    int az = gz;
+    int bx = gx;
+    int by = gy;
+    int bz = gz;
 
     if (psc(adims) == 3)
     {
@@ -99,37 +97,36 @@ void main()
             if (psc(bw) == 1 && psc(bh) == 1)
             {
                 // special type 1
-                v1 = image3d_ld1(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image3d_ld1(b_blob_3d, ivec3(0, 0, gz));
+                bx = 0;
+                by = 0;
             }
 
             if (psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(bc) == 1)
             {
                 // special type 2
-                v1 = image3d_ld1(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image3d_ld1(b_blob_3d, ivec3(gx, gy, 0));
+                bz = 0;
             }
 
             if (psc(aw) == 1 && psc(ah) == 1)
             {
                 // special type 3
-                v1 = image3d_ld1(a_blob_3d, ivec3(0, 0, gz));
-                v2 = image3d_ld1(b_blob_3d, ivec3(gx, gy, gz));
+                ax = 0;
+                ay = 0;
             }
 
             if (psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(ac) == 1)
             {
                 // special type 4
-                v1 = image3d_ld1(a_blob_3d, ivec3(gx, gy, 0));
-                v2 = image3d_ld1(b_blob_3d, ivec3(gx, gy, gz));
+                az = 0;
             }
         }
 
         if (psc(bdims) == 2)
         {
             // type 18
-            v1 = image3d_ld1(a_blob_3d, ivec3(gx, gy, gz));
-            v2 = image2d_ld1(b_blob_2d, ivec2(gy, gz));
+            bx = gy;
+            by = gz;
+            bz = 0;
         }
 
         if (psc(bdims) == 1)
@@ -137,14 +134,16 @@ void main()
             if (psc(bw) == 1)
             {
                 // type 16
-                v1 = image3d_ld1(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image1d_ld1(b_blob_1d, 0);
+                bx = 0;
+                by = 0;
+                bz = 0;
             }
             else
             {
                 // type 17
-                v1 = image3d_ld1(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image1d_ld1(b_blob_1d, gz);
+                bx = gz;
+                by = 0;
+                bz = 0;
             }
         }
     }
@@ -153,8 +152,9 @@ void main()
         if (psc(bdims) == 3)
         {
             // type 14
-            v1 = image2d_ld1(a_blob_2d, ivec2(gy, gz));
-            v2 = image3d_ld1(b_blob_3d, ivec3(gx, gy, gz));
+            ax = gy;
+            ay = gz;
+            az = 0;
         }
 
         if (psc(bdims) == 1)
@@ -162,67 +162,61 @@ void main()
             if (psc(bw) == 1)
             {
                 // type 11
-                v1 = image2d_ld1(a_blob_2d, ivec2(gx, gy));
-                v2 = image1d_ld1(b_blob_1d, 0);
+                bx = 0;
+                by = 0;
+                bz = 0;
             }
             else
             {
                 // type 12
-                v1 = image2d_ld1(a_blob_2d, ivec2(gx, gy));
-                v2 = image1d_ld1(b_blob_1d, gy);
+                bx = gy;
+                by = 0;
+                bz = 0;
             }
         }
     }
     else if (psc(adims) == 1)
     {
-        if (psc(bdims) == 3)
+        if (psc(aw) == 1)
         {
-            if (psc(aw) == 1)
-            {
-                // type 4
-                v1 = image1d_ld1(a_blob_1d, 0);
-                v2 = image3d_ld1(b_blob_3d, ivec3(gx, gy, gz));
-            }
-            else
-            {
-                // type 9
-                v1 = image1d_ld1(a_blob_1d, gz);
-                v2 = image3d_ld1(b_blob_3d, ivec3(gx, gy, gz));
-            }
+            // type 2 3 4
+            ax = 0;
+            ay = 0;
+            az = 0;
         }
-
-        if (psc(bdims) == 2)
+        else
         {
-            if (psc(aw) == 1)
+            if (psc(bdims) == 3)
             {
-                // type 3
-                v1 = image1d_ld1(a_blob_1d, 0);
-                v2 = image2d_ld1(b_blob_2d, ivec2(gx, gy));
+                // type 9
+                ax = gz;
+                ay = 0;
+                az = 0;
             }
-            else
+
+            if (psc(bdims) == 2)
             {
                 // type 8
-                v1 = image1d_ld1(a_blob_1d, gy);
-                v2 = image2d_ld1(b_blob_2d, ivec2(gx, gy));
+                ax = gy;
+                ay = 0;
+                az = 0;
             }
-        }
 
-        if (psc(bdims) == 1)
-        {
-            if (psc(aw) == 1)
-            {
-                // type 2
-                v1 = image1d_ld1(a_blob_1d, 0);
-                v2 = image1d_ld1(b_blob_1d, gx);
-            }
-            else // if (psc(bw) == 1)
+            if (psc(bdims) == 1)
             {
-                // type 6
-                v1 = image1d_ld1(a_blob_1d, gx);
-                v2 = image1d_ld1(b_blob_1d, 0);
+                if (psc(bw) == 1)
+                {
+                    // type 6
+                    bx = 0;
+                    by = 0;
+                    bz = 0;
+                }
             }
         }
     }
+
+    afp v1 = image3d_ld1(a_blob_3d, ivec3(ax, ay, az));
+    afp v2 = image3d_ld1(b_blob_3d, ivec3(bx, by, bz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -363,18 +357,7 @@ void main()
     if (op_type == 8) res = v2 / v1;
 
 #if NCNN_image_shader
-    if (psc(outdims) == 1)
-    {
-        image1d_st1(top_blob_1d, gx, res);
-    }
-    else if (psc(outdims) == 2)
-    {
-        image2d_st1(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(outdims) == 3)
-    {
-        image3d_st1(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st1(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st1(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/binaryop_broadcast_a1_pack4.comp

@@ -43,13 +43,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
-layout (binding = 1) uniform unfp sampler2D b_blob_2d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer a_blob { sfp a_blob_data[]; };
@@ -89,7 +84,6 @@ void main()
 
 #if NCNN_image_shader
     afpvec4 v1;
-    afpvec4 v2;
 
     if (psc(adims) == 3 && psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(ac) == 1)
     {
@@ -98,21 +92,10 @@ void main()
     }
     else
     {
-        v1 = afpvec4(image1d_ld1(a_blob_1d, 0));
+        v1 = afpvec4(image3d_ld1(a_blob_3d, ivec3(0, 0, 0)));
     }
 
-    if (psc(bdims) == 1)
-    {
-        v2 = image1d_ld4(b_blob_1d, gx);
-    }
-    else if (psc(bdims) == 2)
-    {
-        v2 = image2d_ld4(b_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(bdims) == 3)
-    {
-        v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec4 v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -142,18 +125,7 @@ void main()
     if (op_type == 8) res = v2 / v1;
 
 #if NCNN_image_shader
-    if (psc(outdims) == 1)
-    {
-        image1d_st4(top_blob_1d, gx, res);
-    }
-    else if (psc(outdims) == 2)
-    {
-        image2d_st4(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(outdims) == 3)
-    {
-        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st4(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/binaryop_broadcast_a1_pack8.comp

@@ -44,13 +44,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
-layout (binding = 1) uniform unfp sampler2D b_blob_2d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer a_blob { sfp a_blob_data[]; };
@@ -90,7 +85,6 @@ void main()
 
 #if NCNN_image_shader
     afpvec4 v1;
-    afpvec8 v2;
 
     if (psc(adims) == 3 && psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(ac) == 1)
     {
@@ -99,21 +93,10 @@ void main()
     }
     else
     {
-        v1 = afpvec4(image1d_ld1(a_blob_1d, 0));
+        v1 = afpvec4(image3d_ld1(a_blob_3d, ivec3(0, 0, 0)));
     }
 
-    if (psc(bdims) == 1)
-    {
-        v2 = image1d_ld8(b_blob_1d, gx);
-    }
-    else if (psc(bdims) == 2)
-    {
-        v2 = image2d_ld8(b_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(bdims) == 3)
-    {
-        v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec8 v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -179,18 +162,7 @@ void main()
     }
 
 #if NCNN_image_shader
-    if (psc(outdims) == 1)
-    {
-        image1d_st8(top_blob_1d, gx, res);
-    }
-    else if (psc(outdims) == 2)
-    {
-        image2d_st8(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(outdims) == 3)
-    {
-        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st8(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/binaryop_broadcast_b1_pack4.comp

@@ -43,13 +43,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
-layout (binding = 0) uniform unfp sampler2D a_blob_2d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer a_blob { sfpvec4 a_blob_data[]; };
@@ -88,7 +83,6 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec4 v1;
     afpvec4 v2;
 
     if (psc(bdims) == 3 && psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(bc) == 1)
@@ -98,21 +92,10 @@ void main()
     }
     else
     {
-        v2 = afpvec4(image1d_ld1(b_blob_1d, 0));
+        v2 = afpvec4(image3d_ld1(b_blob_3d, ivec3(0, 0, 0)));
     }
 
-    if (psc(adims) == 1)
-    {
-        v1 = image1d_ld4(a_blob_1d, gx);
-    }
-    else if (psc(adims) == 2)
-    {
-        v1 = image2d_ld4(a_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(adims) == 3)
-    {
-        v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec4 v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -142,18 +125,7 @@ void main()
     if (op_type == 8) res = v2 / v1;
 
 #if NCNN_image_shader
-    if (psc(outdims) == 1)
-    {
-        image1d_st4(top_blob_1d, gx, res);
-    }
-    else if (psc(outdims) == 2)
-    {
-        image2d_st4(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(outdims) == 3)
-    {
-        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st4(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/binaryop_broadcast_b1_pack8.comp

@@ -44,13 +44,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
-layout (binding = 0) uniform unfp sampler2D a_blob_2d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer a_blob { sfpvec8 a_blob_data[]; };
@@ -89,7 +84,6 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec8 v1;
     afpvec4 v2;
 
     if (psc(bdims) == 3 && psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(bc) == 1)
@@ -99,21 +93,10 @@ void main()
     }
     else
     {
-        v2 = afpvec4(image1d_ld1(b_blob_1d, 0));
+        v2 = afpvec4(image3d_ld1(b_blob_3d, ivec3(0, 0, 0)));
     }
 
-    if (psc(adims) == 1)
-    {
-        v1 = image1d_ld8(a_blob_1d, gx);
-    }
-    else if (psc(adims) == 2)
-    {
-        v1 = image2d_ld8(a_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(adims) == 3)
-    {
-        v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec8 v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -179,18 +162,7 @@ void main()
     }
 
 #if NCNN_image_shader
-    if (psc(outdims) == 1)
-    {
-        image1d_st8(top_blob_1d, gx, res);
-    }
-    else if (psc(outdims) == 2)
-    {
-        image2d_st8(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(outdims) == 3)
-    {
-        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st8(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/binaryop_broadcast_pack4.comp

@@ -43,14 +43,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
-layout (binding = 0) uniform unfp sampler2D a_blob_2d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
-layout (binding = 1) uniform unfp sampler2D b_blob_2d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer a_blob { sfpvec4 a_blob_data[]; };
@@ -89,8 +83,12 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec4 v1;
-    afpvec4 v2;
+    int ax = gx;
+    int ay = gy;
+    int az = gz;
+    int bx = gx;
+    int by = gy;
+    int bz = gz;
 
     if (psc(adims) == 3)
     {
@@ -99,37 +97,36 @@ void main()
             if (psc(bw) == 1 && psc(bh) == 1)
             {
                 // special type 1
-                v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image3d_ld4(b_blob_3d, ivec3(0, 0, gz));
+                bx = 0;
+                by = 0;
             }
 
             if (psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(bc) == 1)
             {
                 // special type 2
-                v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, 0));
+                bz = 0;
             }
 
             if (psc(aw) == 1 && psc(ah) == 1)
             {
                 // special type 3
-                v1 = image3d_ld4(a_blob_3d, ivec3(0, 0, gz));
-                v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
+                ax = 0;
+                ay = 0;
             }
 
             if (psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(ac) == 1)
             {
                 // special type 4
-                v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, 0));
-                v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
+                az = 0;
             }
         }
 
         if (psc(bdims) == 2)
         {
             // type 18
-            v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
-            v2 = image2d_ld4(b_blob_2d, ivec2(gy, gz));
+            bx = gy;
+            by = gz;
+            bz = 0;
         }
 
         if (psc(bdims) == 1)
@@ -137,14 +134,16 @@ void main()
             if (psc(bw) == 1)
             {
                 // type 16
-                v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image1d_ld4(b_blob_1d, 0);
+                bx = 0;
+                by = 0;
+                bz = 0;
             }
             else
             {
                 // type 17
-                v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image1d_ld4(b_blob_1d, gz);
+                bx = gz;
+                by = 0;
+                bz = 0;
             }
         }
     }
@@ -153,8 +152,9 @@ void main()
         if (psc(bdims) == 3)
         {
             // type 14
-            v1 = image2d_ld4(a_blob_2d, ivec2(gy, gz));
-            v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
+            ax = gy;
+            ay = gz;
+            az = 0;
         }
 
         if (psc(bdims) == 1)
@@ -162,67 +162,61 @@ void main()
             if (psc(bw) == 1)
             {
                 // type 11
-                v1 = image2d_ld4(a_blob_2d, ivec2(gx, gy));
-                v2 = image1d_ld4(b_blob_1d, 0);
+                bx = 0;
+                by = 0;
+                bz = 0;
             }
             else
             {
                 // type 12
-                v1 = image2d_ld4(a_blob_2d, ivec2(gx, gy));
-                v2 = image1d_ld4(b_blob_1d, gy);
+                bx = gy;
+                by = 0;
+                bz = 0;
             }
         }
     }
     else if (psc(adims) == 1)
     {
-        if (psc(bdims) == 3)
+        if (psc(aw) == 1)
         {
-            if (psc(aw) == 1)
-            {
-                // type 4
-                v1 = image1d_ld4(a_blob_1d, 0);
-                v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
-            }
-            else
-            {
-                // type 9
-                v1 = image1d_ld4(a_blob_1d, gz);
-                v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
-            }
+            // type 2 3 4
+            ax = 0;
+            ay = 0;
+            az = 0;
         }
-
-        if (psc(bdims) == 2)
+        else
         {
-            if (psc(aw) == 1)
+            if (psc(bdims) == 3)
             {
-                // type 3
-                v1 = image1d_ld4(a_blob_1d, 0);
-                v2 = image2d_ld4(b_blob_2d, ivec2(gx, gy));
+                // type 9
+                ax = gz;
+                ay = 0;
+                az = 0;
             }
-            else
+
+            if (psc(bdims) == 2)
             {
                 // type 8
-                v1 = image1d_ld4(a_blob_1d, gy);
-                v2 = image2d_ld4(b_blob_2d, ivec2(gx, gy));
+                ax = gy;
+                ay = 0;
+                az = 0;
             }
-        }
 
-        if (psc(bdims) == 1)
-        {
-            if (psc(aw) == 1)
+            if (psc(bdims) == 1)
             {
-                // type 2
-                v1 = image1d_ld4(a_blob_1d, 0);
-                v2 = image1d_ld4(b_blob_1d, gx);
-            }
-            else // if (psc(bw) == 1)
-            {
-                // type 6
-                v1 = image1d_ld4(a_blob_1d, gx);
-                v2 = image1d_ld4(b_blob_1d, 0);
+                if (psc(bw) == 1)
+                {
+                    // type 6
+                    bx = 0;
+                    by = 0;
+                    bz = 0;
+                }
             }
         }
     }
+
+    afpvec4 v1 = image3d_ld4(a_blob_3d, ivec3(ax, ay, az));
+    afpvec4 v2 = image3d_ld4(b_blob_3d, ivec3(bx, by, bz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -312,18 +306,7 @@ void main()
     if (op_type == 8) res = v2 / v1;
 
 #if NCNN_image_shader
-    if (psc(outdims) == 1)
-    {
-        image1d_st4(top_blob_1d, gx, res);
-    }
-    else if (psc(outdims) == 2)
-    {
-        image2d_st4(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(outdims) == 3)
-    {
-        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st4(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/binaryop_broadcast_pack8.comp

@@ -44,14 +44,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
-layout (binding = 0) uniform unfp sampler2D a_blob_2d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
-layout (binding = 1) uniform unfp sampler2D b_blob_2d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer a_blob { sfpvec8 a_blob_data[]; };
@@ -90,8 +84,12 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec8 v1;
-    afpvec8 v2;
+    int ax = gx;
+    int ay = gy;
+    int az = gz;
+    int bx = gx;
+    int by = gy;
+    int bz = gz;
 
     if (psc(adims) == 3)
     {
@@ -100,37 +98,36 @@ void main()
             if (psc(bw) == 1 && psc(bh) == 1)
             {
                 // special type 1
-                v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image3d_ld8(b_blob_3d, ivec3(0, 0, gz));
+                bx = 0;
+                by = 0;
             }
 
             if (psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(bc) == 1)
             {
                 // special type 2
-                v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, 0));
+                bz = 0;
             }
 
             if (psc(aw) == 1 && psc(ah) == 1)
             {
                 // special type 3
-                v1 = image3d_ld8(a_blob_3d, ivec3(0, 0, gz));
-                v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, gz));
+                ax = 0;
+                ay = 0;
             }
 
             if (psc(bw) == psc(aw) && psc(bh) == psc(ah) && psc(ac) == 1)
             {
                 // special type 4
-                v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, 0));
-                v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, gz));
+                az = 0;
             }
         }
 
         if (psc(bdims) == 2)
         {
             // type 18
-            v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, gz));
-            v2 = image2d_ld8(b_blob_2d, ivec2(gy, gz));
+            bx = gy;
+            by = gz;
+            bz = 0;
         }
 
         if (psc(bdims) == 1)
@@ -138,14 +135,16 @@ void main()
             if (psc(bw) == 1)
             {
                 // type 16
-                v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image1d_ld8(b_blob_1d, 0);
+                bx = 0;
+                by = 0;
+                bz = 0;
             }
             else
             {
                 // type 17
-                v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, gz));
-                v2 = image1d_ld8(b_blob_1d, gz);
+                bx = gz;
+                by = 0;
+                bz = 0;
             }
         }
     }
@@ -154,8 +153,9 @@ void main()
         if (psc(bdims) == 3)
         {
             // type 14
-            v1 = image2d_ld8(a_blob_2d, ivec2(gy, gz));
-            v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, gz));
+            ax = gy;
+            ay = gz;
+            az = 0;
         }
 
         if (psc(bdims) == 1)
@@ -163,67 +163,61 @@ void main()
             if (psc(bw) == 1)
             {
                 // type 11
-                v1 = image2d_ld8(a_blob_2d, ivec2(gx, gy));
-                v2 = image1d_ld8(b_blob_1d, 0);
+                bx = 0;
+                by = 0;
+                bz = 0;
             }
             else
             {
                 // type 12
-                v1 = image2d_ld8(a_blob_2d, ivec2(gx, gy));
-                v2 = image1d_ld8(b_blob_1d, gy);
+                bx = gy;
+                by = 0;
+                bz = 0;
             }
         }
     }
     else if (psc(adims) == 1)
     {
-        if (psc(bdims) == 3)
+        if (psc(aw) == 1)
         {
-            if (psc(aw) == 1)
-            {
-                // type 4
-                v1 = image1d_ld8(a_blob_1d, 0);
-                v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, gz));
-            }
-            else
-            {
-                // type 9
-                v1 = image1d_ld8(a_blob_1d, gz);
-                v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, gz));
-            }
+            // type 2 3 4
+            ax = 0;
+            ay = 0;
+            az = 0;
         }
-
-        if (psc(bdims) == 2)
+        else
         {
-            if (psc(aw) == 1)
+            if (psc(bdims) == 3)
             {
-                // type 3
-                v1 = image1d_ld8(a_blob_1d, 0);
-                v2 = image2d_ld8(b_blob_2d, ivec2(gx, gy));
+                // type 9
+                ax = gz;
+                ay = 0;
+                az = 0;
             }
-            else
+
+            if (psc(bdims) == 2)
             {
                 // type 8
-                v1 = image1d_ld8(a_blob_1d, gy);
-                v2 = image2d_ld8(b_blob_2d, ivec2(gx, gy));
+                ax = gy;
+                ay = 0;
+                az = 0;
             }
-        }
 
-        if (psc(bdims) == 1)
-        {
-            if (psc(aw) == 1)
+            if (psc(bdims) == 1)
             {
-                // type 2
-                v1 = image1d_ld8(a_blob_1d, 0);
-                v2 = image1d_ld8(b_blob_1d, gx);
-            }
-            else // if (psc(bw) == 1)
-            {
-                // type 6
-                v1 = image1d_ld8(a_blob_1d, gx);
-                v2 = image1d_ld8(b_blob_1d, 0);
+                if (psc(bw) == 1)
+                {
+                    // type 6
+                    bx = 0;
+                    by = 0;
+                    bz = 0;
+                }
             }
         }
     }
+
+    afpvec8 v1 = image3d_ld8(a_blob_3d, ivec3(ax, ay, az));
+    afpvec8 v2 = image3d_ld8(b_blob_3d, ivec3(bx, by, bz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -349,18 +343,7 @@ void main()
     }
 
 #if NCNN_image_shader
-    if (psc(outdims) == 1)
-    {
-        image1d_st8(top_blob_1d, gx, res);
-    }
-    else if (psc(outdims) == 2)
-    {
-        image2d_st8(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(outdims) == 3)
-    {
-        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st8(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/binaryop_pack4.comp

@@ -45,14 +45,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
-layout (binding = 0) uniform unfp sampler2D a_blob_2d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
-layout (binding = 1) uniform unfp sampler2D b_blob_2d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer a_blob { sfpvec4 a_blob_data[]; };
@@ -91,20 +85,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec4 v1;
-
-    if (psc(outdims) == 1)
-    {
-        v1 = image1d_ld4(a_blob_1d, gx);
-    }
-    else if (psc(outdims) == 2)
-    {
-        v1 = image2d_ld4(a_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(outdims) == 3)
-    {
-        v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec4 v1 = image3d_ld4(a_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -129,18 +110,7 @@ void main()
         if (op_type == 8) res = b / v1;
 
 #if NCNN_image_shader
-        if (psc(outdims) == 1)
-        {
-            image1d_st4(top_blob_1d, gx, res);
-        }
-        else if (psc(outdims) == 2)
-        {
-            image2d_st4(top_blob_2d, ivec2(gx, gy), res);
-        }
-        else // if (psc(outdims) == 3)
-        {
-            image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
-        }
+        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
         buffer_st4(a_blob_data, gi, res);
 #endif
@@ -149,20 +119,7 @@ void main()
     {
         // type 7 13 19
 #if NCNN_image_shader
-        afpvec4 v2;
-
-        if (psc(outdims) == 1)
-        {
-            v2 = image1d_ld4(b_blob_1d, gx);
-        }
-        else if (psc(outdims) == 2)
-        {
-            v2 = image2d_ld4(b_blob_2d, ivec2(gx, gy));
-        }
-        else // if (psc(outdims) == 3)
-        {
-            v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
-        }
+        afpvec4 v2 = image3d_ld4(b_blob_3d, ivec3(gx, gy, gz));
 #else
         afpvec4 v2 = buffer_ld4(b_blob_data, gi);
 #endif
@@ -178,18 +135,7 @@ void main()
         if (op_type == 8) res = v2 / v1;
 
 #if NCNN_image_shader
-        if (psc(outdims) == 1)
-        {
-            image1d_st4(top_blob_1d, gx, res);
-        }
-        else if (psc(outdims) == 2)
-        {
-            image2d_st4(top_blob_2d, ivec2(gx, gy), res);
-        }
-        else // if (psc(outdims) == 3)
-        {
-            image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
-        }
+        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
         buffer_st4(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/binaryop_pack8.comp

@@ -46,14 +46,8 @@ layout (constant_id = shape_constant_id_offset + 13) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 14) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D a_blob_1d;
-layout (binding = 0) uniform unfp sampler2D a_blob_2d;
 layout (binding = 0) uniform unfp sampler3D a_blob_3d;
-layout (binding = 1) uniform unfp sampler1D b_blob_1d;
-layout (binding = 1) uniform unfp sampler2D b_blob_2d;
 layout (binding = 1) uniform unfp sampler3D b_blob_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer a_blob { sfpvec8 a_blob_data[]; };
@@ -92,20 +86,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec8 v1;
-
-    if (psc(outdims) == 1)
-    {
-        v1 = image1d_ld8(a_blob_1d, gx);
-    }
-    else if (psc(outdims) == 2)
-    {
-        v1 = image2d_ld8(a_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(outdims) == 3)
-    {
-        v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec8 v1 = image3d_ld8(a_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 
@@ -166,18 +147,7 @@ void main()
         }
 
 #if NCNN_image_shader
-        if (psc(outdims) == 1)
-        {
-            image1d_st8(top_blob_1d, gx, res);
-        }
-        else if (psc(outdims) == 2)
-        {
-            image2d_st8(top_blob_2d, ivec2(gx, gy), res);
-        }
-        else // if (psc(outdims) == 3)
-        {
-            image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
-        }
+        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
         buffer_st8(a_blob_data, gi, res);
 #endif
@@ -186,20 +156,7 @@ void main()
     {
         // type 7 13 19
 #if NCNN_image_shader
-        afpvec8 v2;
-
-        if (psc(outdims) == 1)
-        {
-            v2 = image1d_ld8(b_blob_1d, gx);
-        }
-        else if (psc(outdims) == 2)
-        {
-            v2 = image2d_ld8(b_blob_2d, ivec2(gx, gy));
-        }
-        else // if (psc(outdims) == 3)
-        {
-            v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, gz));
-        }
+        afpvec8 v2 = image3d_ld8(b_blob_3d, ivec3(gx, gy, gz));
 #else
         afpvec8 v2 = buffer_ld8(b_blob_data, gi);
 #endif
@@ -251,18 +208,7 @@ void main()
         }
 
 #if NCNN_image_shader
-        if (psc(outdims) == 1)
-        {
-            image1d_st8(top_blob_1d, gx, res);
-        }
-        else if (psc(outdims) == 2)
-        {
-            image2d_st8(top_blob_2d, ivec2(gx, gy), res);
-        }
-        else // if (psc(outdims) == 3)
-        {
-            image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
-        }
+        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
         buffer_st8(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/cast_fp16_to_fp32.comp

@@ -35,11 +35,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, r32f) writeonly uniform highp image1D top_blob_1d;
-layout (binding = 1, r32f) writeonly uniform highp image2D top_blob_2d;
 layout (binding = 1, r32f) writeonly uniform highp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
@@ -71,18 +67,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp1(top_blob_1d, gx, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_cp1(top_blob_2d, ivec2(gx, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_cp1(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    image3d_cp1(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 

src/layer/vulkan/shader/cast_fp16_to_fp32_pack4.comp

@@ -35,11 +35,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, rgba32f) writeonly uniform highp image1D top_blob_1d;
-layout (binding = 1, rgba32f) writeonly uniform highp image2D top_blob_2d;
 layout (binding = 1, rgba32f) writeonly uniform highp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
@@ -71,18 +67,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp4(top_blob_1d, gx, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_cp4(top_blob_2d, ivec2(gx, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_cp4(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    image3d_cp4(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 

src/layer/vulkan/shader/cast_fp16_to_fp32_pack8.comp

@@ -36,11 +36,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, rgba32f) writeonly uniform highp image1D top_blob_1d;
-layout (binding = 1, rgba32f) writeonly uniform highp image2D top_blob_2d;
 layout (binding = 1, rgba32f) writeonly uniform highp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
@@ -72,18 +68,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp8(top_blob_1d, gx, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_cp8(top_blob_2d, ivec2(gx, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_cp8(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    image3d_cp8(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 

src/layer/vulkan/shader/cast_fp32_to_fp16.comp

@@ -35,11 +35,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform highp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform highp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform highp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { float bottom_blob_data[]; };
@@ -71,18 +67,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp1(top_blob_1d, gx, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_cp1(top_blob_2d, ivec2(gx, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_cp1(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    image3d_cp1(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 

src/layer/vulkan/shader/cast_fp32_to_fp16_pack4.comp

@@ -35,11 +35,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform highp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform highp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform highp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { vec4 bottom_blob_data[]; };
@@ -71,18 +67,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp4(top_blob_1d, gx, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_cp4(top_blob_2d, ivec2(gx, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_cp4(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    image3d_cp4(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 

src/layer/vulkan/shader/cast_fp32_to_fp16_pack8.comp

@@ -36,11 +36,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform highp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform highp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform highp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { mat2x4 bottom_blob_data[]; };
@@ -72,18 +68,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp8(top_blob_1d, gx, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_cp8(top_blob_2d, ivec2(gx, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_cp8(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    image3d_cp8(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
 

src/layer/vulkan/shader/clip.comp

@@ -32,11 +32,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfp bottom_top_blob_data[]; };
@@ -61,19 +57,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afp v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld1(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld1(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afp v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -83,18 +67,7 @@ void main()
     v = clamp(v, afp(const_min), afp(const_max));
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st1(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st1(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st1(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/clip_pack4.comp

@@ -32,11 +32,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec4 bottom_top_blob_data[]; };
@@ -61,19 +57,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec4 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld4(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld4(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec4 v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -83,18 +67,7 @@ void main()
     v = clamp(v, afp(const_min), afp(const_max));
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st4(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st4(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st4(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/clip_pack8.comp

@@ -33,11 +33,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec8 bottom_top_blob_data[]; };
@@ -62,19 +58,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec8 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld8(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld8(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -85,18 +69,7 @@ void main()
     v[1] = clamp(v[1], afp(const_min), afp(const_max));
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st8(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st8(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st8(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/concat.comp

@@ -37,11 +37,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
@@ -74,29 +70,15 @@ void main()
     if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
         return;
 
-#if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp1(top_blob_1d, gx + p.offset, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        if (axis == 0) image2d_cp1(top_blob_2d, ivec2(gx, gy + p.offset), bottom_blob_2d, ivec2(gx, gy));
-        if (axis == 1) image2d_cp1(top_blob_2d, ivec2(gx + p.offset, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        if (axis == 0) image3d_cp1(top_blob_3d, ivec3(gx, gy, gz + p.offset), bottom_blob_3d, ivec3(gx, gy, gz));
-        if (axis == 1) image3d_cp1(top_blob_3d, ivec3(gx, gy + p.offset, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-        if (axis == 2) image3d_cp1(top_blob_3d, ivec3(gx + p.offset, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
-#else
-    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
-
     ivec3 gxyz = ivec3(gx, gy, gz);
 
     gxyz[psc(dims) - 1 - axis] += p.offset;
 
+#if NCNN_image_shader
+    image3d_cp1(top_blob_3d, gxyz, bottom_blob_3d, ivec3(gx, gy, gz));
+#else
+    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
+
     int v_offset = gxyz.z * psc(outcstep) + gxyz.y * psc(outw) + gxyz.x;
 
     buffer_cp1(top_blob_data, v_offset, bottom_blob_data, gi);

src/layer/vulkan/shader/concat_pack4.comp

@@ -37,11 +37,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
@@ -74,29 +70,15 @@ void main()
     if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
         return;
 
-#if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp4(top_blob_1d, gx + p.offset, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        if (axis == 0) image2d_cp4(top_blob_2d, ivec2(gx, gy + p.offset), bottom_blob_2d, ivec2(gx, gy));
-        if (axis == 1) image2d_cp4(top_blob_2d, ivec2(gx + p.offset, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        if (axis == 0) image3d_cp4(top_blob_3d, ivec3(gx, gy, gz + p.offset), bottom_blob_3d, ivec3(gx, gy, gz));
-        if (axis == 1) image3d_cp4(top_blob_3d, ivec3(gx, gy + p.offset, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-        if (axis == 2) image3d_cp4(top_blob_3d, ivec3(gx + p.offset, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
-#else
-    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
-
     ivec3 gxyz = ivec3(gx, gy, gz);
 
     gxyz[psc(dims) - 1 - axis] += p.offset;
 
+#if NCNN_image_shader
+    image3d_cp4(top_blob_3d, gxyz, bottom_blob_3d, ivec3(gx, gy, gz));
+#else
+    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
+
     int v_offset = gxyz.z * psc(outcstep) + gxyz.y * psc(outw) + gxyz.x;
 
     buffer_cp4(top_blob_data, v_offset, bottom_blob_data, gi);

src/layer/vulkan/shader/concat_pack4to1.comp

@@ -37,11 +37,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
@@ -74,81 +70,65 @@ void main()
     if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
         return;
 
+    ivec3 gxyz = ivec3(gx, gy, gz);
+
+    gxyz[psc(dims) - 1] *= 4;
+    gxyz[psc(dims) - 1 - axis] += p.offset;
+
 #if NCNN_image_shader
+    afpvec4 v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
+
     if (psc(dims) == 1)
     {
-        afpvec4 v = image1d_ld4(bottom_blob_1d, gx);
-
-        int gx4 = gx * 4 + p.offset;
-
-        image1d_st1(top_blob_1d, gx4 + 0, v.r);
-        image1d_st1(top_blob_1d, gx4 + 1, v.g);
-        image1d_st1(top_blob_1d, gx4 + 2, v.b);
-        image1d_st1(top_blob_1d, gx4 + 3, v.a);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 0, 0, 0), v.r);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 1, 0, 0), v.g);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 2, 0, 0), v.b);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 3, 0, 0), v.a);
     }
     else if (psc(dims) == 2)
     {
-        afpvec4 v = image2d_ld4(bottom_blob_2d, ivec2(gx, gy));
-
         if (axis == 0)
         {
-            int gy4 = gy * 4 + p.offset;
-
-            image2d_st1(top_blob_2d, ivec2(gx, gy4 + 0), v.r);
-            image2d_st1(top_blob_2d, ivec2(gx, gy4 + 1), v.g);
-            image2d_st1(top_blob_2d, ivec2(gx, gy4 + 2), v.b);
-            image2d_st1(top_blob_2d, ivec2(gx, gy4 + 3), v.a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 0, 0), v.r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 1, 0), v.g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 2, 0), v.b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 3, 0), v.a);
         }
         if (axis == 1)
         {
-            int gx4 = gx * 4 + p.offset;
-
-            image2d_st1(top_blob_2d, ivec2(gx4 + 0, gy), v.r);
-            image2d_st1(top_blob_2d, ivec2(gx4 + 1, gy), v.g);
-            image2d_st1(top_blob_2d, ivec2(gx4 + 2, gy), v.b);
-            image2d_st1(top_blob_2d, ivec2(gx4 + 3, gy), v.a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 0, gxyz.y, 0), v.r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 1, gxyz.y, 0), v.g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 2, gxyz.y, 0), v.b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 3, gxyz.y, 0), v.a);
         }
     }
     else // if (psc(dims) == 3)
     {
-        afpvec4 v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
-
         if (axis == 0)
         {
-            int gz4 = gz * 4 + p.offset;
-
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz4 + 0), v.r);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz4 + 1), v.g);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz4 + 2), v.b);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz4 + 3), v.a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 0), v.r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 1), v.g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 2), v.b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 3), v.a);
         }
         if (axis == 1)
         {
-            int gy4 = gy * 4 + p.offset;
-
-            image3d_st1(top_blob_3d, ivec3(gx, gy4 + 0, gz), v.r);
-            image3d_st1(top_blob_3d, ivec3(gx, gy4 + 1, gz), v.g);
-            image3d_st1(top_blob_3d, ivec3(gx, gy4 + 2, gz), v.b);
-            image3d_st1(top_blob_3d, ivec3(gx, gy4 + 3, gz), v.a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 0, gxyz.z), v.r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 1, gxyz.z), v.g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 2, gxyz.z), v.b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 3, gxyz.z), v.a);
         }
         if (axis == 2)
         {
-            int gx4 = gx * 4 + p.offset;
-
-            image3d_st1(top_blob_3d, ivec3(gx4 + 0, gy, gz), v.r);
-            image3d_st1(top_blob_3d, ivec3(gx4 + 1, gy, gz), v.g);
-            image3d_st1(top_blob_3d, ivec3(gx4 + 2, gy, gz), v.b);
-            image3d_st1(top_blob_3d, ivec3(gx4 + 3, gy, gz), v.a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 0, gxyz.y, gxyz.z), v.r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 1, gxyz.y, gxyz.z), v.g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 2, gxyz.y, gxyz.z), v.b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 3, gxyz.y, gxyz.z), v.a);
         }
     }
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
-    ivec3 gxyz = ivec3(gx, gy, gz);
-
-    gxyz[psc(dims) - 1] *= 4;
-    gxyz[psc(dims) - 1 - axis] += p.offset;
-
     int v_offset_0 = gxyz.z * psc(outcstep) + gxyz.y * psc(outw) + gxyz.x;
 
     ivec3 gxyz4 = ivec3(1, psc(outw), psc(outcstep));

src/layer/vulkan/shader/concat_pack8.comp

@@ -38,11 +38,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
@@ -75,29 +71,15 @@ void main()
     if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
         return;
 
-#if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp8(top_blob_1d, gx + p.offset, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        if (axis == 0) image2d_cp8(top_blob_2d, ivec2(gx, gy + p.offset), bottom_blob_2d, ivec2(gx, gy));
-        if (axis == 1) image2d_cp8(top_blob_2d, ivec2(gx + p.offset, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        if (axis == 0) image3d_cp8(top_blob_3d, ivec3(gx, gy, gz + p.offset), bottom_blob_3d, ivec3(gx, gy, gz));
-        if (axis == 1) image3d_cp8(top_blob_3d, ivec3(gx, gy + p.offset, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-        if (axis == 2) image3d_cp8(top_blob_3d, ivec3(gx + p.offset, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
-#else
-    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
-
     ivec3 gxyz = ivec3(gx, gy, gz);
 
     gxyz[psc(dims) - 1 - axis] += p.offset;
 
+#if NCNN_image_shader
+    image3d_cp8(top_blob_3d, gxyz, bottom_blob_3d, ivec3(gx, gy, gz));
+#else
+    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
+
     int v_offset = gxyz.z * psc(outcstep) + gxyz.y * psc(outw) + gxyz.x;
 
     buffer_cp8(top_blob_data, v_offset, bottom_blob_data, gi);

src/layer/vulkan/shader/concat_pack8to1.comp

@@ -38,11 +38,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
@@ -75,105 +71,89 @@ void main()
     if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
         return;
 
+    ivec3 gxyz = ivec3(gx, gy, gz);
+
+    gxyz[psc(dims) - 1] *= 8;
+    gxyz[psc(dims) - 1 - axis] += p.offset;
+
 #if NCNN_image_shader
+    afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
+
     if (psc(dims) == 1)
     {
-        afpvec8 v = image1d_ld8(bottom_blob_1d, gx);
-
-        int gx8 = gx * 8 + p.offset;
-
-        image1d_st1(top_blob_1d, gx8 + 0, v[0].r);
-        image1d_st1(top_blob_1d, gx8 + 1, v[0].g);
-        image1d_st1(top_blob_1d, gx8 + 2, v[0].b);
-        image1d_st1(top_blob_1d, gx8 + 3, v[0].a);
-        image1d_st1(top_blob_1d, gx8 + 4, v[1].r);
-        image1d_st1(top_blob_1d, gx8 + 5, v[1].g);
-        image1d_st1(top_blob_1d, gx8 + 6, v[1].b);
-        image1d_st1(top_blob_1d, gx8 + 7, v[1].a);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 0, 0, 0), v[0].r);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 1, 0, 0), v[0].g);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 2, 0, 0), v[0].b);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 3, 0, 0), v[0].a);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 4, 0, 0), v[1].r);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 5, 0, 0), v[1].g);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 6, 0, 0), v[1].b);
+        image3d_st1(top_blob_3d, ivec3(gxyz.x + 7, 0, 0), v[1].a);
     }
     else if (psc(dims) == 2)
     {
-        afpvec8 v = image2d_ld8(bottom_blob_2d, ivec2(gx, gy));
-
         if (axis == 0)
         {
-            int gy8 = gy * 8 + p.offset;
-
-            image2d_st1(top_blob_2d, ivec2(gx, gy8 + 0), v[0].r);
-            image2d_st1(top_blob_2d, ivec2(gx, gy8 + 1), v[0].g);
-            image2d_st1(top_blob_2d, ivec2(gx, gy8 + 2), v[0].b);
-            image2d_st1(top_blob_2d, ivec2(gx, gy8 + 3), v[0].a);
-            image2d_st1(top_blob_2d, ivec2(gx, gy8 + 4), v[1].r);
-            image2d_st1(top_blob_2d, ivec2(gx, gy8 + 5), v[1].g);
-            image2d_st1(top_blob_2d, ivec2(gx, gy8 + 6), v[1].b);
-            image2d_st1(top_blob_2d, ivec2(gx, gy8 + 7), v[1].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 0, 0), v[0].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 1, 0), v[0].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 2, 0), v[0].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 3, 0), v[0].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 4, 0), v[1].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 5, 0), v[1].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 6, 0), v[1].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 7, 0), v[1].a);
         }
         if (axis == 1)
         {
-            int gx8 = gx * 8 + p.offset;
-
-            image2d_st1(top_blob_2d, ivec2(gx8 + 0, gy), v[0].r);
-            image2d_st1(top_blob_2d, ivec2(gx8 + 1, gy), v[0].g);
-            image2d_st1(top_blob_2d, ivec2(gx8 + 2, gy), v[0].b);
-            image2d_st1(top_blob_2d, ivec2(gx8 + 3, gy), v[0].a);
-            image2d_st1(top_blob_2d, ivec2(gx8 + 4, gy), v[1].r);
-            image2d_st1(top_blob_2d, ivec2(gx8 + 5, gy), v[1].g);
-            image2d_st1(top_blob_2d, ivec2(gx8 + 6, gy), v[1].b);
-            image2d_st1(top_blob_2d, ivec2(gx8 + 7, gy), v[1].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 0, gxyz.y, 0), v[0].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 1, gxyz.y, 0), v[0].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 2, gxyz.y, 0), v[0].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 3, gxyz.y, 0), v[0].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 4, gxyz.y, 0), v[1].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 5, gxyz.y, 0), v[1].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 6, gxyz.y, 0), v[1].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 7, gxyz.y, 0), v[1].a);
         }
     }
     else // if (psc(dims) == 3)
     {
-        afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
-
         if (axis == 0)
         {
-            int gz8 = gz * 8 + p.offset;
-
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz8 + 0), v[0].r);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz8 + 1), v[0].g);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz8 + 2), v[0].b);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz8 + 3), v[0].a);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz8 + 4), v[1].r);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz8 + 5), v[1].g);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz8 + 6), v[1].b);
-            image3d_st1(top_blob_3d, ivec3(gx, gy, gz8 + 7), v[1].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 0), v[0].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 1), v[0].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 2), v[0].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 3), v[0].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 4), v[1].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 5), v[1].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 6), v[1].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 7), v[1].a);
         }
         if (axis == 1)
         {
-            int gy8 = gy * 8 + p.offset;
-
-            image3d_st1(top_blob_3d, ivec3(gx, gy8 + 0, gz), v[0].r);
-            image3d_st1(top_blob_3d, ivec3(gx, gy8 + 1, gz), v[0].g);
-            image3d_st1(top_blob_3d, ivec3(gx, gy8 + 2, gz), v[0].b);
-            image3d_st1(top_blob_3d, ivec3(gx, gy8 + 3, gz), v[0].a);
-            image3d_st1(top_blob_3d, ivec3(gx, gy8 + 4, gz), v[1].r);
-            image3d_st1(top_blob_3d, ivec3(gx, gy8 + 5, gz), v[1].g);
-            image3d_st1(top_blob_3d, ivec3(gx, gy8 + 6, gz), v[1].b);
-            image3d_st1(top_blob_3d, ivec3(gx, gy8 + 7, gz), v[1].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 0, gxyz.z), v[0].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 1, gxyz.z), v[0].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 2, gxyz.z), v[0].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 3, gxyz.z), v[0].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 4, gxyz.z), v[1].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 5, gxyz.z), v[1].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 6, gxyz.z), v[1].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x, gxyz.y + 7, gxyz.z), v[1].a);
         }
         if (axis == 2)
         {
-            int gx8 = gx * 8 + p.offset;
-
-            image3d_st1(top_blob_3d, ivec3(gx8 + 0, gy, gz), v[0].r);
-            image3d_st1(top_blob_3d, ivec3(gx8 + 1, gy, gz), v[0].g);
-            image3d_st1(top_blob_3d, ivec3(gx8 + 2, gy, gz), v[0].b);
-            image3d_st1(top_blob_3d, ivec3(gx8 + 3, gy, gz), v[0].a);
-            image3d_st1(top_blob_3d, ivec3(gx8 + 4, gy, gz), v[1].r);
-            image3d_st1(top_blob_3d, ivec3(gx8 + 5, gy, gz), v[1].g);
-            image3d_st1(top_blob_3d, ivec3(gx8 + 6, gy, gz), v[1].b);
-            image3d_st1(top_blob_3d, ivec3(gx8 + 7, gy, gz), v[1].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 0, gxyz.y, gxyz.z), v[0].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 1, gxyz.y, gxyz.z), v[0].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 2, gxyz.y, gxyz.z), v[0].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 3, gxyz.y, gxyz.z), v[0].a);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 4, gxyz.y, gxyz.z), v[1].r);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 5, gxyz.y, gxyz.z), v[1].g);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 6, gxyz.y, gxyz.z), v[1].b);
+            image3d_st1(top_blob_3d, ivec3(gxyz.x + 7, gxyz.y, gxyz.z), v[1].a);
         }
     }
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
-    ivec3 gxyz = ivec3(gx, gy, gz);
-
-    gxyz[psc(dims) - 1] *= 8;
-    gxyz[psc(dims) - 1 - axis] += p.offset;
-
     int v_offset_0 = gxyz.z * psc(outcstep) + gxyz.y * psc(outw) + gxyz.x;
 
     ivec3 gxyz4 = ivec3(1, psc(outw), psc(outcstep));

src/layer/vulkan/shader/concat_pack8to4.comp

@@ -38,11 +38,7 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
@@ -75,70 +71,54 @@ void main()
     if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
         return;
 
+    ivec3 gxyz = ivec3(gx, gy, gz);
+
+    gxyz[psc(dims) - 1] *= 2;
+    gxyz[psc(dims) - 1 - axis] += p.offset;
+
 #if NCNN_image_shader
+    afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
+
     if (psc(dims) == 1)
     {
-        afpvec8 v = image1d_ld8(bottom_blob_1d, gx);
-
-        int gx2 = gx * 2 + p.offset;
-
-        image1d_st4(top_blob_1d, gx2 + 0, v[0]);
-        image1d_st4(top_blob_1d, gx2 + 1, v[1]);
+        image3d_st4(top_blob_3d, ivec3(gxyz.x + 0, 0, 0), v[0]);
+        image3d_st4(top_blob_3d, ivec3(gxyz.x + 1, 0, 0), v[1]);
 
     }
     else if (psc(dims) == 2)
     {
-        afpvec8 v = image2d_ld8(bottom_blob_2d, ivec2(gx, gy));
-
         if (axis == 0)
         {
-            int gy2 = gy * 2 + p.offset;
-
-            image2d_st4(top_blob_2d, ivec2(gx, gy2 + 0), v[0]);
-            image2d_st4(top_blob_2d, ivec2(gx, gy2 + 1), v[1]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y + 0, 0), v[0]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y + 1, 0), v[1]);
         }
         if (axis == 1)
         {
-            int gx2 = gx * 2 + p.offset;
-
-            image2d_st4(top_blob_2d, ivec2(gx2 + 0, gy), v[0]);
-            image2d_st4(top_blob_2d, ivec2(gx2 + 1, gy), v[1]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x + 0, gxyz.y, 0), v[0]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x + 1, gxyz.y, 0), v[1]);
         }
     }
     else // if (psc(dims) == 3)
     {
-        afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
-
         if (axis == 0)
         {
-            int gz2 = gz * 2 + p.offset;
-
-            image3d_st4(top_blob_3d, ivec3(gx, gy, gz2 + 0), v[0]);
-            image3d_st4(top_blob_3d, ivec3(gx, gy, gz2 + 1), v[1]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 0), v[0]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y, gxyz.z + 1), v[1]);
         }
         if (axis == 1)
         {
-            int gy2 = gy * 2 + p.offset;
-
-            image3d_st4(top_blob_3d, ivec3(gx, gy2 + 0, gz), v[0]);
-            image3d_st4(top_blob_3d, ivec3(gx, gy2 + 1, gz), v[1]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y + 0, gxyz.z), v[0]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x, gxyz.y + 1, gxyz.z), v[1]);
         }
         if (axis == 2)
         {
-            int gx2 = gx * 2 + p.offset;
-
-            image3d_st4(top_blob_3d, ivec3(gx2 + 0, gy, gz), v[0]);
-            image3d_st4(top_blob_3d, ivec3(gx2 + 1, gy, gz), v[1]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x + 0, gxyz.y, gxyz.z), v[0]);
+            image3d_st4(top_blob_3d, ivec3(gxyz.x + 1, gxyz.y, gxyz.z), v[1]);
         }
     }
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
-    ivec3 gxyz = ivec3(gx, gy, gz);
-
-    gxyz[psc(dims) - 1] *= 2;
-    gxyz[psc(dims) - 1 - axis] += p.offset;
-
     int v_offset_0 = gxyz.z * psc(outcstep) + gxyz.y * psc(outw) + gxyz.x;
 
     ivec3 gxyz4 = ivec3(1, psc(outw), psc(outcstep));

src/layer/vulkan/shader/convolution.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -86,7 +86,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_1x1s1d1.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 #if NCNN_fp16_packed
 layout (binding = 0) readonly buffer bottom_blob { vec4 bottom_blob_data[]; };
@@ -100,7 +100,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = afpvec4(image1d_ld1(bias_blob, gz));
+        sum = afpvec4(image3d_ld1(bias_blob, ivec3(gz, 0, 0)));
 #else
         sum = afpvec4(buffer_ld1(bias_data, gz));
 #endif

src/layer/vulkan/shader/convolution_pack1to4.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -86,7 +86,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack1to8.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack4.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -91,7 +91,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack4_1x1s1d1.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -103,7 +103,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        afpvec4 b = image1d_ld4(bias_blob, gz);
+        afpvec4 b = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         afpvec4 b = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack4_3x3s1d1_winograd23_transform_output.comp

@@ -40,7 +40,7 @@ layout (constant_id = shape_constant_id_offset + 6) const int outcstep = 0;
 #if NCNN_image_shader
 layout (binding = 0) uniform unfp sampler3D top_tm_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
-layout (binding = 2) uniform unfp sampler1D bias_blob;
+layout (binding = 2) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer top_tm_blob { sfpvec4 top_tm_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -129,7 +129,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        const afpvec4 bias_value = image1d_ld4(bias_blob, gz);
+        const afpvec4 bias_value = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         const afpvec4 bias_value = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack4to1.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -86,7 +86,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack4to8.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack8.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack8_1x1s1d1.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -99,7 +99,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        afpvec8 b = image1d_ld8(bias_blob, gz);
+        afpvec8 b = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         afpvec8 b = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack8_3x3s1d1_winograd23_transform_output.comp

@@ -41,7 +41,7 @@ layout (constant_id = shape_constant_id_offset + 6) const int outcstep = 0;
 #if NCNN_image_shader
 layout (binding = 0) uniform unfp sampler3D top_tm_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
-layout (binding = 2) uniform unfp sampler1D bias_blob;
+layout (binding = 2) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer top_tm_blob { sfpvec8 top_tm_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -130,7 +130,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        const afpvec8 bias_value = image1d_ld8(bias_blob, gz);
+        const afpvec8 bias_value = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         const afpvec8 bias_value = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack8to1.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolution_pack8to4.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise.comp

@@ -49,8 +49,8 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 #if NCNN_image_shader
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
-layout (binding = 2) uniform unfp sampler2D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 2) uniform unfp sampler3D weight_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif
@@ -108,7 +108,7 @@ void main()
 
         for (int x = 0; x < kernel_w; x++)
         {
-            sum += image2d_ld1(weight_blob, ivec2(wx, gz)) * image3d_ld1(bottom_blob, ivec3(sx, sy, gz));
+            sum += image3d_ld1(weight_blob, ivec3(wx, gz, 0)) * image3d_ld1(bottom_blob, ivec3(sx, sy, gz));
 
             sx += dilation_w;
             wx += 1;

src/layer/vulkan/shader/convolutiondepthwise_group.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise_group_pack1to4.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise_group_pack1to8.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise_group_pack4.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -92,7 +92,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise_group_pack4to1.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise_group_pack4to8.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise_group_pack8.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise_group_pack8to1.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise_group_pack8to4.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/convolutiondepthwise_pack4.comp

@@ -49,8 +49,8 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 #if NCNN_image_shader
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
-layout (binding = 2) uniform unfp sampler2D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 2) uniform unfp sampler3D weight_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif
@@ -110,7 +110,7 @@ void main()
         {
             afpvec4 v = image3d_ld4(bottom_blob, ivec3(sx, sy, gz));
 
-            afpvec4 k = image2d_ld4(weight_blob, ivec2(wx, gz));
+            afpvec4 k = image3d_ld4(weight_blob, ivec3(wx, gz, 0));
 
             sum += v * k;
 

src/layer/vulkan/shader/convolutiondepthwise_pack8.comp

@@ -50,8 +50,8 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 #if NCNN_image_shader
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
-layout (binding = 2) uniform unfp sampler2D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 2) uniform unfp sampler3D weight_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif
@@ -111,7 +111,7 @@ void main()
         {
             afpvec8 v = image3d_ld8(bottom_blob, ivec3(sx, sy, gz));
 
-            afpvec8 k = image2d_ld8(weight_blob, ivec2(wx, gz));
+            afpvec8 k = image3d_ld8(weight_blob, ivec3(wx, gz, 0));
 
             // sum += v * k;
             sum[0] += v[0] * k[0];

src/layer/vulkan/shader/deconvolution.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -86,7 +86,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolution_pack1to4.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -86,7 +86,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolution_pack1to8.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolution_pack4.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -91,7 +91,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolution_pack4to1.comp

@@ -49,7 +49,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -86,7 +86,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolution_pack4to8.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolution_pack8.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolution_pack8to1.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolution_pack8to4.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise.comp

@@ -49,8 +49,8 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 #if NCNN_image_shader
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
-layout (binding = 2) uniform unfp sampler2D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 2) uniform unfp sampler3D weight_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif
@@ -124,7 +124,7 @@ void main()
 
             int wx = y * kernel_w + x;
 
-            sum += image2d_ld1(weight_blob, ivec2(wx, gz)) * image3d_ld1(bottom_blob, ivec3(sx, sy, gz));
+            sum += image3d_ld1(weight_blob, ivec3(wx, gz, 0)) * image3d_ld1(bottom_blob, ivec3(sx, sy, gz));
         }
     }
 #else

src/layer/vulkan/shader/deconvolutiondepthwise_group.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise_group_pack1to4.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise_group_pack1to8.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise_group_pack4.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -92,7 +92,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise_group_pack4to1.comp

@@ -50,7 +50,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise_group_pack4to8.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise_group_pack8.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise_group_pack8to1.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld1(bias_blob, gz);
+        sum = image3d_ld1(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld1(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise_group_pack8to4.comp

@@ -51,7 +51,7 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 layout (binding = 2) uniform unfp sampler3D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif

src/layer/vulkan/shader/deconvolutiondepthwise_pack4.comp

@@ -49,8 +49,8 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 #if NCNN_image_shader
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
-layout (binding = 2) uniform unfp sampler2D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 2) uniform unfp sampler3D weight_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -87,7 +87,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld4(bias_blob, gz);
+        sum = image3d_ld4(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld4(bias_data, gz);
 #endif
@@ -126,7 +126,7 @@ void main()
 
             afpvec4 v = image3d_ld4(bottom_blob, ivec3(sx, sy, gz));
 
-            afpvec4 k = image2d_ld4(weight_blob, ivec2(wx, gz));
+            afpvec4 k = image3d_ld4(weight_blob, ivec3(wx, gz, 0));
 
             sum += v * k;
         }

src/layer/vulkan/shader/deconvolutiondepthwise_pack8.comp

@@ -50,8 +50,8 @@ layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 #if NCNN_image_shader
 layout (binding = 0) uniform unfp sampler3D bottom_blob;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
-layout (binding = 2) uniform unfp sampler2D weight_blob;
-layout (binding = 3) uniform unfp sampler1D bias_blob;
+layout (binding = 2) uniform unfp sampler3D weight_blob;
+layout (binding = 3) uniform unfp sampler3D bias_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -88,7 +88,7 @@ void main()
     if (bias_term == 1)
     {
 #if NCNN_image_shader
-        sum = image1d_ld8(bias_blob, gz);
+        sum = image3d_ld8(bias_blob, ivec3(gz, 0, 0));
 #else
         sum = buffer_ld8(bias_data, gz);
 #endif
@@ -127,7 +127,7 @@ void main()
 
             afpvec8 v = image3d_ld8(bottom_blob, ivec3(sx, sy, gz));
 
-            afpvec8 k = image2d_ld8(weight_blob, ivec2(wx, gz));
+            afpvec8 k = image3d_ld8(weight_blob, ivec3(wx, gz, 0));
 
             // sum += v * k;
             sum[0] += v[0] * k[0];

src/layer/vulkan/shader/deepcopy.comp

@@ -29,11 +29,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
@@ -59,18 +55,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp1(top_blob_1d, gx, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_cp1(top_blob_2d, ivec2(gx, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_cp1(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    image3d_cp1(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 

src/layer/vulkan/shader/deepcopy_pack4.comp

@@ -29,11 +29,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec4 bottom_blob_data[]; };
@@ -59,18 +55,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp4(top_blob_1d, gx, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_cp4(top_blob_2d, ivec2(gx, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_cp4(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    image3d_cp4(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 

src/layer/vulkan/shader/deepcopy_pack8.comp

@@ -30,11 +30,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
@@ -60,18 +56,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_cp8(top_blob_1d, gx, bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_cp8(top_blob_2d, ivec2(gx, gy), bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_cp8(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    image3d_cp8(top_blob_3d, ivec3(gx, gy, gz), bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 

src/layer/vulkan/shader/dropout.comp

@@ -31,11 +31,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfp bottom_top_blob_data[]; };
@@ -60,19 +56,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afp v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld1(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld1(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afp v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -82,18 +66,7 @@ void main()
     v *= afp(scale);
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st1(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st1(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st1(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/dropout_pack4.comp

@@ -31,11 +31,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec4 bottom_top_blob_data[]; };
@@ -60,19 +56,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec4 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld4(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld4(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec4 v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -82,18 +66,7 @@ void main()
     v *= afp(scale);
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st4(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st4(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st4(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/dropout_pack8.comp

@@ -32,11 +32,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec8 bottom_top_blob_data[]; };
@@ -61,19 +57,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec8 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld8(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld8(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -84,18 +68,7 @@ void main()
     v[1] = v[1] * afp(scale);
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st8(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st8(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st8(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/eltwise.comp

@@ -32,14 +32,8 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob1_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob1_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob1_3d;
-layout (binding = 1) uniform unfp sampler1D bottom_blob2_1d;
-layout (binding = 1) uniform unfp sampler2D bottom_blob2_2d;
 layout (binding = 1) uniform unfp sampler3D bottom_blob2_3d;
-layout (binding = 2, imfmtc1) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc1) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc1) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob1 { sfp bottom_blob1_data[]; };
@@ -69,23 +63,8 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afp v1;
-    afp v2;
-    if (psc(dims) == 1)
-    {
-        v1 = image1d_ld1(bottom_blob1_1d, gx);
-        v2 = image1d_ld1(bottom_blob2_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v1 = image2d_ld1(bottom_blob1_2d, ivec2(gx, gy));
-        v2 = image2d_ld1(bottom_blob2_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v1 = image3d_ld1(bottom_blob1_3d, ivec3(gx, gy, gz));
-        v2 = image3d_ld1(bottom_blob2_3d, ivec3(gx, gy, gz));
-    }
+    afp v1 = image3d_ld1(bottom_blob1_3d, ivec3(gx, gy, gz));
+    afp v2 = image3d_ld1(bottom_blob2_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -119,18 +98,7 @@ void main()
     }
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st1(top_blob_1d, gx, res);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st1(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st1(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st1(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st1(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/eltwise_pack4.comp

@@ -32,14 +32,8 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob1_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob1_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob1_3d;
-layout (binding = 1) uniform unfp sampler1D bottom_blob2_1d;
-layout (binding = 1) uniform unfp sampler2D bottom_blob2_2d;
 layout (binding = 1) uniform unfp sampler3D bottom_blob2_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob1 { sfpvec4 bottom_blob1_data[]; };
@@ -69,23 +63,8 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec4 v1;
-    afpvec4 v2;
-    if (psc(dims) == 1)
-    {
-        v1 = image1d_ld4(bottom_blob1_1d, gx);
-        v2 = image1d_ld4(bottom_blob2_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v1 = image2d_ld4(bottom_blob1_2d, ivec2(gx, gy));
-        v2 = image2d_ld4(bottom_blob2_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v1 = image3d_ld4(bottom_blob1_3d, ivec3(gx, gy, gz));
-        v2 = image3d_ld4(bottom_blob2_3d, ivec3(gx, gy, gz));
-    }
+    afpvec4 v1 = image3d_ld4(bottom_blob1_3d, ivec3(gx, gy, gz));
+    afpvec4 v2 = image3d_ld4(bottom_blob2_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -119,18 +98,7 @@ void main()
     }
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st4(top_blob_1d, gx, res);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st4(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st4(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/eltwise_pack8.comp

@@ -33,14 +33,8 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob1_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob1_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob1_3d;
-layout (binding = 1) uniform unfp sampler1D bottom_blob2_1d;
-layout (binding = 1) uniform unfp sampler2D bottom_blob2_2d;
 layout (binding = 1) uniform unfp sampler3D bottom_blob2_3d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 2, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 2, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) readonly buffer bottom_blob1 { sfpvec8 bottom_blob1_data[]; };
@@ -70,23 +64,8 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec8 v1;
-    afpvec8 v2;
-    if (psc(dims) == 1)
-    {
-        v1 = image1d_ld8(bottom_blob1_1d, gx);
-        v2 = image1d_ld8(bottom_blob2_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v1 = image2d_ld8(bottom_blob1_2d, ivec2(gx, gy));
-        v2 = image2d_ld8(bottom_blob2_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v1 = image3d_ld8(bottom_blob1_3d, ivec3(gx, gy, gz));
-        v2 = image3d_ld8(bottom_blob2_3d, ivec3(gx, gy, gz));
-    }
+    afpvec8 v1 = image3d_ld8(bottom_blob1_3d, ivec3(gx, gy, gz));
+    afpvec8 v2 = image3d_ld8(bottom_blob2_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -138,18 +117,7 @@ void main()
     }
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st8(top_blob_1d, gx, res);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st8(top_blob_2d, ivec2(gx, gy), res);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
-    }
+    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), res);
 #else
     buffer_st8(top_blob_data, gi, res);
 #endif

src/layer/vulkan/shader/flatten.comp

@@ -35,9 +35,8 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc1) writeonly uniform unfp image1D top_blob;
+layout (binding = 1, imfmtc1) writeonly uniform unfp image3D top_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfp top_blob_data[]; };
@@ -78,14 +77,14 @@ void main()
 
     if (psc(dims) == 2)
     {
-        v = image2d_ld1(bottom_blob_2d, ivec2(x, y));
+        v = image3d_ld1(bottom_blob_3d, ivec3(x, y, 0));
     }
     else // if (psc(dims) == 3)
     {
         v = image3d_ld1(bottom_blob_3d, ivec3(x, y, z));
     }
 
-    image1d_st1(top_blob, gx, v);
+    image3d_st1(top_blob, ivec3(gx, 0, 0), v);
 #else
     int v_offset = z * psc(cstep) + y * psc(w) + x;
 

src/layer/vulkan/shader/flatten_pack1to4.comp

@@ -35,9 +35,8 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob;
+layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec4 top_blob_data[]; };
@@ -77,10 +76,10 @@ void main()
         ivec4 y4 = i4 / psc(w);
         ivec4 x4 = i4 % psc(w);
 
-        v.r = image2d_ld1(bottom_blob_2d, ivec2(x4.r, y4.r));
-        v.g = image2d_ld1(bottom_blob_2d, ivec2(x4.g, y4.g));
-        v.b = image2d_ld1(bottom_blob_2d, ivec2(x4.b, y4.b));
-        v.a = image2d_ld1(bottom_blob_2d, ivec2(x4.a, y4.a));
+        v.r = image3d_ld1(bottom_blob_3d, ivec3(x4.r, y4.r, 0));
+        v.g = image3d_ld1(bottom_blob_3d, ivec3(x4.g, y4.g, 0));
+        v.b = image3d_ld1(bottom_blob_3d, ivec3(x4.b, y4.b, 0));
+        v.a = image3d_ld1(bottom_blob_3d, ivec3(x4.a, y4.a, 0));
     }
     else // if (psc(dims) == 3)
     {
@@ -96,7 +95,7 @@ void main()
         v.a = image3d_ld1(bottom_blob_3d, ivec3(x4.a, y4.a, z4.a));
     }
 
-    image1d_st4(top_blob, gx, v);
+    image3d_st4(top_blob, ivec3(gx, 0, 0), v);
 #else
     ivec4 v_offset;
 

src/layer/vulkan/shader/flatten_pack1to8.comp

@@ -36,9 +36,8 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob;
+layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 #else
 layout (binding = 0) readonly buffer bottom_blob { sfp bottom_blob_data[]; };
 layout (binding = 1) writeonly buffer top_blob { sfpvec8 top_blob_data[]; };
@@ -82,14 +81,14 @@ void main()
         ivec4 yy4 = ii4 / psc(w);
         ivec4 xx4 = ii4 % psc(w);
 
-        v[0].r = image2d_ld1(bottom_blob_2d, ivec2(x4.r, y4.r));
-        v[0].g = image2d_ld1(bottom_blob_2d, ivec2(x4.g, y4.g));
-        v[0].b = image2d_ld1(bottom_blob_2d, ivec2(x4.b, y4.b));
-        v[0].a = image2d_ld1(bottom_blob_2d, ivec2(x4.a, y4.a));
-        v[1].r = image2d_ld1(bottom_blob_2d, ivec2(xx4.r, yy4.r));
-        v[1].g = image2d_ld1(bottom_blob_2d, ivec2(xx4.g, yy4.g));
-        v[1].b = image2d_ld1(bottom_blob_2d, ivec2(xx4.b, yy4.b));
-        v[1].a = image2d_ld1(bottom_blob_2d, ivec2(xx4.a, yy4.a));
+        v[0].r = image3d_ld1(bottom_blob_3d, ivec3(x4.r, y4.r, 0));
+        v[0].g = image3d_ld1(bottom_blob_3d, ivec3(x4.g, y4.g, 0));
+        v[0].b = image3d_ld1(bottom_blob_3d, ivec3(x4.b, y4.b, 0));
+        v[0].a = image3d_ld1(bottom_blob_3d, ivec3(x4.a, y4.a, 0));
+        v[1].r = image3d_ld1(bottom_blob_3d, ivec3(xx4.r, yy4.r, 0));
+        v[1].g = image3d_ld1(bottom_blob_3d, ivec3(xx4.g, yy4.g, 0));
+        v[1].b = image3d_ld1(bottom_blob_3d, ivec3(xx4.b, yy4.b, 0));
+        v[1].a = image3d_ld1(bottom_blob_3d, ivec3(xx4.a, yy4.a, 0));
     }
     else // if (psc(dims) == 3)
     {
@@ -113,7 +112,7 @@ void main()
         v[1].a = image3d_ld1(bottom_blob_3d, ivec3(xx4.a, yy4.a, zz4.a));
     }
 
-    image1d_st8(top_blob, gx, v);
+    image3d_st8(top_blob, ivec3(gx, 0, 0), v);
 #else
     ivec4 v_offset;
     ivec4 vv_offset;

src/layer/vulkan/shader/flatten_pack4.comp

@@ -35,9 +35,8 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob;
+layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 #else
 #if NCNN_fp16_packed
 layout (binding = 0) readonly buffer bottom_blob { sfpvec2 bottom_blob_data[]; };
@@ -81,10 +80,10 @@ void main()
         ivec4 y4 = i4 / psc(w);
         ivec4 x4 = i4 % psc(w);
 
-        afpvec4 v0 = image2d_ld4(bottom_blob_2d, ivec2(x4.r, y4.r / 4));
-        afpvec4 v1 = image2d_ld4(bottom_blob_2d, ivec2(x4.g, y4.g / 4));
-        afpvec4 v2 = image2d_ld4(bottom_blob_2d, ivec2(x4.b, y4.b / 4));
-        afpvec4 v3 = image2d_ld4(bottom_blob_2d, ivec2(x4.a, y4.a / 4));
+        afpvec4 v0 = image3d_ld4(bottom_blob_3d, ivec3(x4.r, y4.r / 4, 0));
+        afpvec4 v1 = image3d_ld4(bottom_blob_3d, ivec3(x4.g, y4.g / 4, 0));
+        afpvec4 v2 = image3d_ld4(bottom_blob_3d, ivec3(x4.b, y4.b / 4, 0));
+        afpvec4 v3 = image3d_ld4(bottom_blob_3d, ivec3(x4.a, y4.a / 4, 0));
 
         v.r = v0[y4.r % 4];
         v.g = v1[y4.g % 4];
@@ -110,7 +109,7 @@ void main()
         v.a = v3[z4.a % 4];
     }
 
-    image1d_st4(top_blob, gx, v);
+    image3d_st4(top_blob, ivec3(gx, 0, 0), v);
 #else
 #if NCNN_fp16_packed
     ivec4 v_offset;

src/layer/vulkan/shader/flatten_pack4to8.comp

@@ -36,9 +36,8 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob;
+layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 #else
 #if NCNN_fp16_packed
 layout (binding = 0) readonly buffer bottom_blob { sfpvec2 bottom_blob_data[]; };
@@ -85,14 +84,14 @@ void main()
         ivec4 yy4 = ii4 / psc(w);
         ivec4 xx4 = ii4 % psc(w);
 
-        afpvec4 v0 = image2d_ld4(bottom_blob_2d, ivec2(x4.r, y4.r / 4));
-        afpvec4 v1 = image2d_ld4(bottom_blob_2d, ivec2(x4.g, y4.g / 4));
-        afpvec4 v2 = image2d_ld4(bottom_blob_2d, ivec2(x4.b, y4.b / 4));
-        afpvec4 v3 = image2d_ld4(bottom_blob_2d, ivec2(x4.a, y4.a / 4));
-        afpvec4 v4 = image2d_ld4(bottom_blob_2d, ivec2(xx4.r, yy4.r / 4));
-        afpvec4 v5 = image2d_ld4(bottom_blob_2d, ivec2(xx4.g, yy4.g / 4));
-        afpvec4 v6 = image2d_ld4(bottom_blob_2d, ivec2(xx4.b, yy4.b / 4));
-        afpvec4 v7 = image2d_ld4(bottom_blob_2d, ivec2(xx4.a, yy4.a / 4));
+        afpvec4 v0 = image3d_ld4(bottom_blob_3d, ivec3(x4.r, y4.r / 4, 0));
+        afpvec4 v1 = image3d_ld4(bottom_blob_3d, ivec3(x4.g, y4.g / 4, 0));
+        afpvec4 v2 = image3d_ld4(bottom_blob_3d, ivec3(x4.b, y4.b / 4, 0));
+        afpvec4 v3 = image3d_ld4(bottom_blob_3d, ivec3(x4.a, y4.a / 4, 0));
+        afpvec4 v4 = image3d_ld4(bottom_blob_3d, ivec3(xx4.r, yy4.r / 4, 0));
+        afpvec4 v5 = image3d_ld4(bottom_blob_3d, ivec3(xx4.g, yy4.g / 4, 0));
+        afpvec4 v6 = image3d_ld4(bottom_blob_3d, ivec3(xx4.b, yy4.b / 4, 0));
+        afpvec4 v7 = image3d_ld4(bottom_blob_3d, ivec3(xx4.a, yy4.a / 4, 0));
 
         v[0].r = v0[y4.r % 4];
         v[0].g = v1[y4.g % 4];
@@ -133,7 +132,7 @@ void main()
         v[1].a = v7[zz4.a % 4];
     }
 
-    image1d_st8(top_blob, gx, v);
+    image3d_st8(top_blob, ivec3(gx, 0, 0), v);
 #else
 #if NCNN_fp16_packed
     ivec4 v_offset;

src/layer/vulkan/shader/flatten_pack8.comp

@@ -36,9 +36,8 @@ layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
 layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob;
+layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob;
 #else
 #if NCNN_fp16_packed
 layout (binding = 0) readonly buffer bottom_blob { sfpvec2 bottom_blob_data[]; };
@@ -85,14 +84,14 @@ void main()
         ivec4 yy4 = ii4 / psc(w);
         ivec4 xx4 = ii4 % psc(w);
 
-        afpvec8 v0 = image2d_ld8(bottom_blob_2d, ivec2(x4.r, y4.r / 8));
-        afpvec8 v1 = image2d_ld8(bottom_blob_2d, ivec2(x4.g, y4.g / 8));
-        afpvec8 v2 = image2d_ld8(bottom_blob_2d, ivec2(x4.b, y4.b / 8));
-        afpvec8 v3 = image2d_ld8(bottom_blob_2d, ivec2(x4.a, y4.a / 8));
-        afpvec8 v4 = image2d_ld8(bottom_blob_2d, ivec2(xx4.r, yy4.r / 8));
-        afpvec8 v5 = image2d_ld8(bottom_blob_2d, ivec2(xx4.g, yy4.g / 8));
-        afpvec8 v6 = image2d_ld8(bottom_blob_2d, ivec2(xx4.b, yy4.b / 8));
-        afpvec8 v7 = image2d_ld8(bottom_blob_2d, ivec2(xx4.a, yy4.a / 8));
+        afpvec8 v0 = image3d_ld8(bottom_blob_3d, ivec3(x4.r, y4.r / 8, 0));
+        afpvec8 v1 = image3d_ld8(bottom_blob_3d, ivec3(x4.g, y4.g / 8, 0));
+        afpvec8 v2 = image3d_ld8(bottom_blob_3d, ivec3(x4.b, y4.b / 8, 0));
+        afpvec8 v3 = image3d_ld8(bottom_blob_3d, ivec3(x4.a, y4.a / 8, 0));
+        afpvec8 v4 = image3d_ld8(bottom_blob_3d, ivec3(xx4.r, yy4.r / 8, 0));
+        afpvec8 v5 = image3d_ld8(bottom_blob_3d, ivec3(xx4.g, yy4.g / 8, 0));
+        afpvec8 v6 = image3d_ld8(bottom_blob_3d, ivec3(xx4.b, yy4.b / 8, 0));
+        afpvec8 v7 = image3d_ld8(bottom_blob_3d, ivec3(xx4.a, yy4.a / 8, 0));
 
         v[0].r = v0[(y4.r % 8) / 4][y4.r % 4];
         v[0].g = v1[(y4.g % 8) / 4][y4.g % 4];
@@ -133,7 +132,7 @@ void main()
         v[1].a = v7[(zz4.a % 8) / 4][zz4.a % 4];
     }
 
-    image1d_st8(top_blob, gx, v);
+    image3d_st8(top_blob, ivec3(gx, 0, 0), v);
 #else
 #if NCNN_fp16_packed
     ivec4 v_offset;

src/layer/vulkan/shader/hardsigmoid.comp

@@ -61,19 +61,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afp v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld1(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld1(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afp v = image3d_ld1(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -83,18 +71,7 @@ void main()
     v = clamp(v * afp(alpha) + afp(beta), afp(0.f), afp(1.f));
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st1(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st1(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st1(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/hardsigmoid_pack4.comp

@@ -32,11 +32,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec4 bottom_top_blob_data[]; };
@@ -61,19 +57,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec4 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld4(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld4(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec4 v = image3d_ld4(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -83,18 +67,7 @@ void main()
     v = clamp(v * afp(alpha) + afp(beta), afp(0.f), afp(1.f));
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st4(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st4(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st4(bottom_top_blob_data, gi, v);
 #endif

src/layer/vulkan/shader/hardsigmoid_pack8.comp

@@ -33,11 +33,7 @@ layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
 layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
 
 #if NCNN_image_shader
-layout (binding = 0) uniform unfp sampler1D bottom_blob_1d;
-layout (binding = 0) uniform unfp sampler2D bottom_blob_2d;
 layout (binding = 0) uniform unfp sampler3D bottom_blob_3d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image1D top_blob_1d;
-layout (binding = 1, imfmtc4) writeonly uniform unfp image2D top_blob_2d;
 layout (binding = 1, imfmtc4) writeonly uniform unfp image3D top_blob_3d;
 #else
 layout (binding = 0) buffer bottom_top_blob { sfpvec8 bottom_top_blob_data[]; };
@@ -62,19 +58,7 @@ void main()
         return;
 
 #if NCNN_image_shader
-    afpvec8 v;
-    if (psc(dims) == 1)
-    {
-        v = image1d_ld8(bottom_blob_1d, gx);
-    }
-    else if (psc(dims) == 2)
-    {
-        v = image2d_ld8(bottom_blob_2d, ivec2(gx, gy));
-    }
-    else // if (psc(dims) == 3)
-    {
-        v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
-    }
+    afpvec8 v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
 #else
     const int gi = gz * psc(cstep) + gy * psc(w) + gx;
 
@@ -85,18 +69,7 @@ void main()
     v[1] = clamp(v[1] * afp(alpha) + afp(beta), afp(0.f), afp(1.f));
 
 #if NCNN_image_shader
-    if (psc(dims) == 1)
-    {
-        image1d_st8(top_blob_1d, gx, v);
-    }
-    else if (psc(dims) == 2)
-    {
-        image2d_st8(top_blob_2d, ivec2(gx, gy), v);
-    }
-    else // if (psc(dims) == 3)
-    {
-        image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
-    }
+    image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);
 #else
     buffer_st8(bottom_top_blob_data, gi, v);
 #endif