diff --git a/src/layer/vulkan/shader/unaryop.comp b/src/layer/vulkan/shader/unaryop.comp
index bd16cc17b171..524e911fe8fa 100644
--- a/src/layer/vulkan/shader/unaryop.comp
+++ b/src/layer/vulkan/shader/unaryop.comp
@@ -4,39 +4,25 @@
 #version 450
 
 layout (constant_id = 0) const int op_type = 0;
+layout (constant_id = 1) const uint n = 0;
 
-#define shape_constant_id_offset 1
-layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
-layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
-layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
-layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
-layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
-
-layout (binding = 0) buffer bottom_top_blob { sfp bottom_top_blob_data[]; };
+layout (binding = 0) buffer bottom_top_blob { sfpvec4 bottom_top_blob_data[]; };
 
 layout (push_constant) uniform parameter
 {
-    int dims;
-    int w;
-    int h;
-    int c;
-    int cstep;
+    uint n;
 } p;
 
 void main()
 {
-    int gx = int(gl_GlobalInvocationID.x);
-    int gy = int(gl_GlobalInvocationID.y);
-    int gz = int(gl_GlobalInvocationID.z);
+    const uint gi = gl_GlobalInvocationID.x;
 
-    if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
+    if (gi >= psc(n))
         return;
 
-    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
-
-    afp v = buffer_ld1(bottom_top_blob_data, gi);
+    afpvec4 v = buffer_ld4(bottom_top_blob_data, gi);
 
-    afp res;
+    afpvec4 res;
 
     if (op_type == 0) res = abs(v);
     if (op_type == 1) res = -v;
@@ -63,5 +49,5 @@ void main()
     if (op_type == 18) res = round(v);
     if (op_type == 19) res = trunc(v);
 
-    buffer_st1(bottom_top_blob_data, gi, res);
+    buffer_st4(bottom_top_blob_data, gi, res);
 }
diff --git a/src/layer/vulkan/shader/unaryop_pack4.comp b/src/layer/vulkan/shader/unaryop_pack4.comp
deleted file mode 100644
index 7c1f7c181137..000000000000
--- a/src/layer/vulkan/shader/unaryop_pack4.comp
+++ /dev/null
@@ -1,67 +0,0 @@
-// Copyright 2019 Tencent
-// SPDX-License-Identifier: BSD-3-Clause
-
-#version 450
-
-layout (constant_id = 0) const int op_type = 0;
-
-#define shape_constant_id_offset 1
-layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
-layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
-layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
-layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
-layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
-
-layout (binding = 0) buffer bottom_top_blob { sfpvec4 bottom_top_blob_data[]; };
-
-layout (push_constant) uniform parameter
-{
-    int dims;
-    int w;
-    int h;
-    int c;
-    int cstep;
-} p;
-
-void main()
-{
-    int gx = int(gl_GlobalInvocationID.x);
-    int gy = int(gl_GlobalInvocationID.y);
-    int gz = int(gl_GlobalInvocationID.z);
-
-    if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
-        return;
-
-    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
-
-    afpvec4 v = buffer_ld4(bottom_top_blob_data, gi);
-
-    afpvec4 res;
-
-    if (op_type == 0) res = abs(v);
-    if (op_type == 1) res = -v;
-    if (op_type == 2) res = floor(v);
-    if (op_type == 3) res = ceil(v);
-    if (op_type == 4) res = v * v;
-    if (op_type == 5) res = sqrt(v);
-    if (op_type == 6) res = inversesqrt(v);
-    if (op_type == 7) res = exp(v);
-    if (op_type == 8) res = log(v);
-    if (op_type == 9) res = sin(v);
-    if (op_type == 10) res = cos(v);
-    if (op_type == 11) res = tan(v);
-    if (op_type == 12) res = asin(v);
-    if (op_type == 13) res = acos(v);
-    if (op_type == 14) res = atan(v);
-    if (op_type == 15) res = afp(1.f) / v;
-#if NCNN_moltenvk
-    if (op_type == 16) res = afpvec4(tanh(vec4(v)));
-#else
-    if (op_type == 16) res = tanh(v);
-#endif
-    if (op_type == 17) res = log(v) * afp(0.434294481903);
-    if (op_type == 18) res = round(v);
-    if (op_type == 19) res = trunc(v);
-
-    buffer_st4(bottom_top_blob_data, gi, res);
-}
diff --git a/src/layer/vulkan/shader/unaryop_pack8.comp b/src/layer/vulkan/shader/unaryop_pack8.comp
deleted file mode 100644
index dbf56b04e94e..000000000000
--- a/src/layer/vulkan/shader/unaryop_pack8.comp
+++ /dev/null
@@ -1,148 +0,0 @@
-// Copyright 2020 Tencent
-// SPDX-License-Identifier: BSD-3-Clause
-
-#version 450
-
-layout (constant_id = 0) const int op_type = 0;
-
-#define shape_constant_id_offset 1
-layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
-layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
-layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
-layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
-layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
-
-layout (binding = 0) buffer bottom_top_blob { sfpvec8 bottom_top_blob_data[]; };
-
-layout (push_constant) uniform parameter
-{
-    int dims;
-    int w;
-    int h;
-    int c;
-    int cstep;
-} p;
-
-void main()
-{
-    int gx = int(gl_GlobalInvocationID.x);
-    int gy = int(gl_GlobalInvocationID.y);
-    int gz = int(gl_GlobalInvocationID.z);
-
-    if (gx >= psc(w) || gy >= psc(h) || gz >= psc(c))
-        return;
-
-    const int gi = gz * psc(cstep) + gy * psc(w) + gx;
-
-    afpvec8 v = buffer_ld8(bottom_top_blob_data, gi);
-
-    afpvec8 res;
-
-    if (op_type == 0)
-    {
-        res[0] = abs(v[0]);
-        res[1] = abs(v[1]);
-    }
-    if (op_type == 1)
-    {
-        res[0] = -v[0];
-        res[1] = -v[1];
-    }
-    if (op_type == 2)
-    {
-        res[0] = floor(v[0]);
-        res[1] = floor(v[1]);
-    }
-    if (op_type == 3)
-    {
-        res[0] = ceil(v[0]);
-        res[1] = ceil(v[1]);
-    }
-    if (op_type == 4)
-    {
-        res[0] = v[0] * v[0];
-        res[1] = v[1] * v[1];
-    }
-    if (op_type == 5)
-    {
-        res[0] = sqrt(v[0]);
-        res[1] = sqrt(v[1]);
-    }
-    if (op_type == 6)
-    {
-        res[0] = inversesqrt(v[0]);
-        res[1] = inversesqrt(v[1]);
-    }
-    if (op_type == 7)
-    {
-        res[0] = exp(v[0]);
-        res[1] = exp(v[1]);
-    }
-    if (op_type == 8)
-    {
-        res[0] = log(v[0]);
-        res[1] = log(v[1]);
-    }
-    if (op_type == 9)
-    {
-        res[0] = sin(v[0]);
-        res[1] = sin(v[1]);
-    }
-    if (op_type == 10)
-    {
-        res[0] = cos(v[0]);
-        res[1] = cos(v[1]);
-    }
-    if (op_type == 11)
-    {
-        res[0] = tan(v[0]);
-        res[1] = tan(v[1]);
-    }
-    if (op_type == 12)
-    {
-        res[0] = asin(v[0]);
-        res[1] = asin(v[1]);
-    }
-    if (op_type == 13)
-    {
-        res[0] = acos(v[0]);
-        res[1] = acos(v[1]);
-    }
-    if (op_type == 14)
-    {
-        res[0] = atan(v[0]);
-        res[1] = atan(v[1]);
-    }
-    if (op_type == 15)
-    {
-        res[0] = afp(1.f) / v[0];
-        res[1] = afp(1.f) / v[1];
-    }
-    if (op_type == 16)
-    {
-#if NCNN_moltenvk
-        res[0] = afpvec4(tanh(vec4(v[0])));
-        res[1] = afpvec4(tanh(vec4(v[1])));
-#else
-        res[0] = tanh(v[0]);
-        res[1] = tanh(v[1]);
-#endif
-    }
-    if (op_type == 17)
-    {
-        res[0] = log(v[0]) * afp(0.434294481903);
-        res[1] = log(v[1]) * afp(0.434294481903);
-    }
-    if (op_type == 18)
-    {
-        res[0] = round(v[0]);
-        res[1] = round(v[1]);
-    }
-    if (op_type == 19)
-    {
-        res[0] = trunc(v[0]);
-        res[1] = trunc(v[1]);
-    }
-
-    buffer_st8(bottom_top_blob_data, gi, res);
-}
diff --git a/src/layer/vulkan/unaryop_vulkan.cpp b/src/layer/vulkan/unaryop_vulkan.cpp
index 95654694c6a5..a191fe1b9561 100644
--- a/src/layer/vulkan/unaryop_vulkan.cpp
+++ b/src/layer/vulkan/unaryop_vulkan.cpp
@@ -12,92 +12,21 @@ UnaryOp_vulkan::UnaryOp_vulkan()
     support_vulkan = true;
 
     pipeline_unaryop = 0;
-    pipeline_unaryop_pack4 = 0;
-    pipeline_unaryop_pack8 = 0;
 }
 
 int UnaryOp_vulkan::create_pipeline(const Option& opt)
 {
     const Mat& shape = top_shapes.empty() ? Mat() : top_shapes[0];
 
-    int elempack = 1;
-    if (shape.dims == 1) elempack = opt.use_shader_pack8 && shape.w % 8 == 0 ? 8 : shape.w % 4 == 0 ? 4 : 1;
-    if (shape.dims == 2) elempack = opt.use_shader_pack8 && shape.h % 8 == 0 ? 8 : shape.h % 4 == 0 ? 4 : 1;
-    if (shape.dims == 3 || shape.dims == 4) elempack = opt.use_shader_pack8 && shape.c % 8 == 0 ? 8 : shape.c % 4 == 0 ? 4 : 1;
-
-    size_t elemsize;
-    if (opt.use_fp16_storage || opt.use_fp16_packed)
-    {
-        elemsize = elempack * 2u;
-    }
-    else
-    {
-        elemsize = elempack * 4u;
-    }
-
-    Mat shape_packed;
-    if (shape.dims == 1) shape_packed = Mat(shape.w / elempack, (void*)0, elemsize, elempack);
-    if (shape.dims == 2) shape_packed = Mat(shape.w, shape.h / elempack, (void*)0, elemsize, elempack);
-    if (shape.dims == 3) shape_packed = Mat(shape.w, shape.h, shape.c / elempack, (void*)0, elemsize, elempack);
-    if (shape.dims == 4) shape_packed = Mat(shape.w, shape.h, shape.d, shape.c / elempack, (void*)0, elemsize, elempack);
-
-    std::vector<vk_specialization_type> specializations(1 + 5);
+    std::vector<vk_specialization_type> specializations(2);
     specializations[0].i = op_type;
-    specializations[1 + 0].i = shape_packed.dims;
-    specializations[1 + 1].i = shape_packed.w;
-    specializations[1 + 2].i = shape_packed.h * shape_packed.d;
-    specializations[1 + 3].i = shape_packed.c;
-    specializations[1 + 4].i = shape_packed.cstep;
-
-    Mat local_size_xyz;
-    if (shape_packed.dims == 1)
-    {
-        local_size_xyz.w = std::min(64, shape_packed.w);
-        local_size_xyz.h = 1;
-        local_size_xyz.c = 1;
-    }
-    if (shape_packed.dims == 2)
-    {
-        local_size_xyz.w = std::min(8, shape_packed.w);
-        local_size_xyz.h = std::min(8, shape_packed.h);
-        local_size_xyz.c = 1;
-    }
-    if (shape_packed.dims == 3)
-    {
-        local_size_xyz.w = std::min(4, shape_packed.w);
-        local_size_xyz.h = std::min(4, shape_packed.h);
-        local_size_xyz.c = std::min(4, shape_packed.c);
-    }
-    if (shape_packed.dims == 4)
-    {
-        local_size_xyz.w = std::min(4, shape_packed.w);
-        local_size_xyz.h = std::min(4, shape_packed.h * shape_packed.d);
-        local_size_xyz.c = std::min(4, shape_packed.c);
-    }
-
-    // pack1
-    if (shape.dims == 0 || elempack == 1)
-    {
-        pipeline_unaryop = new Pipeline(vkdev);
-        pipeline_unaryop->set_optimal_local_size_xyz(local_size_xyz);
-        pipeline_unaryop->create(LayerShaderType::unaryop, opt, specializations);
-    }
-
-    // pack4
-    if (shape.dims == 0 || elempack == 4)
-    {
-        pipeline_unaryop_pack4 = new Pipeline(vkdev);
-        pipeline_unaryop_pack4->set_optimal_local_size_xyz(local_size_xyz);
-        pipeline_unaryop_pack4->create(LayerShaderType::unaryop_pack4, opt, specializations);
-    }
-
-    // pack8
-    if ((opt.use_shader_pack8 && shape.dims == 0) || elempack == 8)
-    {
-        pipeline_unaryop_pack8 = new Pipeline(vkdev);
-        pipeline_unaryop_pack8->set_optimal_local_size_xyz(local_size_xyz);
-        pipeline_unaryop_pack8->create(LayerShaderType::unaryop_pack8, opt, specializations);
-    }
+    specializations[1].u32 = shape.total() / 4;
+
+    const int local_size_x = vkdev->info.subgroup_size();
+
+    pipeline_unaryop = new Pipeline(vkdev);
+    pipeline_unaryop->set_optimal_local_size_xyz(local_size_x, 1, 1);
+    pipeline_unaryop->create(LayerShaderType::unaryop, opt, specializations);
 
     return 0;
 }
@@ -107,34 +36,24 @@ int UnaryOp_vulkan::destroy_pipeline(const Option& /*opt*/)
     delete pipeline_unaryop;
     pipeline_unaryop = 0;
 
-    delete pipeline_unaryop_pack4;
-    pipeline_unaryop_pack4 = 0;
-
-    delete pipeline_unaryop_pack8;
-    pipeline_unaryop_pack8 = 0;
-
     return 0;
 }
 
 int UnaryOp_vulkan::forward_inplace(VkMat& bottom_top_blob, VkCompute& cmd, const Option& /*opt*/) const
 {
-    int elempack = bottom_top_blob.elempack;
+    const size_t n = bottom_top_blob.total() * bottom_top_blob.elempack / 4;
 
     std::vector<VkMat> bindings(1);
     bindings[0] = bottom_top_blob;
 
-    std::vector<vk_constant_type> constants(5);
-    constants[0].i = bottom_top_blob.dims;
-    constants[1].i = bottom_top_blob.w;
-    constants[2].i = bottom_top_blob.h * bottom_top_blob.d;
-    constants[3].i = bottom_top_blob.c;
-    constants[4].i = bottom_top_blob.cstep;
-
-    const Pipeline* pipeline = elempack == 8 ? pipeline_unaryop_pack8
-                               : elempack == 4 ? pipeline_unaryop_pack4
-                               : pipeline_unaryop;
+    std::vector<vk_constant_type> constants(1);
+    constants[0].u32 = n;
 
-    cmd.record_pipeline(pipeline, bindings, constants, bottom_top_blob);
+    VkMat dispatcher;
+    dispatcher.w = n;
+    dispatcher.h = 1;
+    dispatcher.c = 1;
+    cmd.record_pipeline(pipeline_unaryop, bindings, constants, dispatcher);
 
     return 0;
 }
diff --git a/src/layer/vulkan/unaryop_vulkan.h b/src/layer/vulkan/unaryop_vulkan.h
index 8d10501e0659..9487eae7dea3 100644
--- a/src/layer/vulkan/unaryop_vulkan.h
+++ b/src/layer/vulkan/unaryop_vulkan.h
@@ -21,8 +21,6 @@ class UnaryOp_vulkan : public UnaryOp
 
 public:
     Pipeline* pipeline_unaryop;
-    Pipeline* pipeline_unaryop_pack4;
-    Pipeline* pipeline_unaryop_pack8;
 };
 
 } // namespace ncnn