[Pytorch][Vulkan] sum.dim_IntList with keepdim

aten/src/ATen/native/vulkan/glsl/sum_dim_keepdim.glsl

@@ -0,0 +1,70 @@
+#version 450 core
+#define PRECISION $precision
+#define FORMAT $format
+
+layout(std430) buffer;
+
+/* Qualifiers: layout - storage - precision - memory */
+layout(set = 0, binding = 0, FORMAT) uniform PRECISION restrict writeonly image3D uOutput;
+layout(set = 0, binding = 1) uniform PRECISION sampler3D uInput;
+layout(set = 0, binding = 2) uniform PRECISION restrict Block {
+  // dim_info.x: dim to sum
+  // dim_info.y: size of dim (in the input)
+  uvec2 dim_info;
+  int channel;
+}
+uBlock;
+
+/*
+ * Local Work Group Size
+ */
+layout(local_size_x_id = 0, local_size_y_id = 1, local_size_z_id = 2) in;
+
+/*
+ * Returns a new tensor with values summed along dimension dim.
+ * Output and input have same number of dimensions.
+ * summed dimension is of size 1.
+ */
+
+void main() {
+  const ivec3 pos = ivec3(gl_GlobalInvocationID);
+
+  int flattened_channels = int(ceil(uBlock.channel / 4.0));
+  vec4 out_texel = vec4(0, 0, 0, 0);
+
+  // Batch
+  if (uBlock.dim_info.x == 0) {
+    for (int batch = 0; batch < uBlock.dim_info.y; batch++) {
+      // src_n = batch
+      // src_c = pos.z
+      int src_z = batch * flattened_channels + pos.z;
+      out_texel += texelFetch(uInput, ivec3(pos.x, pos.y, src_z), 0);
+    }
+    imageStore(uOutput, pos, out_texel);
+  }
+
+  // Channel
+  else if (uBlock.dim_info.x == 1) {
+    for (int out_index = 0; out_index < 4; out_index++) {
+      for (int channel = 0; channel < uBlock.dim_info.y; channel++) {
+        // src_n = pos.z
+        // src_c = channel
+        int src_z = pos.z * flattened_channels + int(channel / 4);
+        vec4 v = texelFetch(uInput, ivec3(pos.x, pos.y, src_z), 0);
+        out_texel[out_index] += v[channel % 4];
+      }
+    }
+    imageStore(uOutput, pos, out_texel);
+  }
+
+  // Height, Width
+  else {
+    for (int hw = 0; hw < uBlock.dim_info.y; hw++) {
+      vec4 v = (uBlock.dim_info.x == 2)
+          ? texelFetch(uInput, ivec3(pos.x, hw, pos.z), 0) // Height
+          : texelFetch(uInput, ivec3(hw, pos.y, pos.z), 0); // Width
+      out_texel += v;
+    }
+    imageStore(uOutput, pos, out_texel);
+  }
+}

aten/src/ATen/native/vulkan/ops/Sum.cpp

@@ -40,7 +40,11 @@ Tensor sum_dim(
   // Create the output texture
   std::vector<int64_t> output_size = self.sizes().vec();
   uint32_t dim_size = output_size[dim];
-  output_size.erase(output_size.begin() + dim);
+  if (keepdim) {
+    output_size[dim] = 1;
+  } else {
+    output_size.erase(output_size.begin() + dim);
+  }
 
   ScalarType type = self.scalar_type();
   if (dtype.has_value()) {
@@ -74,7 +78,7 @@ Tensor sum_dim(
 
   context->submit_compute_job(
       // shader descriptor
-      VK_KERNEL(sum_dim),
+      keepdim ? VK_KERNEL(sum_dim_keepdim) : VK_KERNEL(sum_dim),
       // pipeline barrier
       pipeline_barrier,
       // global work group size
@@ -102,9 +106,6 @@ Tensor sum_dim_IntList(
   TORCH_CHECK(
       opt_dim.has_value(),
       "Vulkan sum.dim_IntList without a dim arg is not implemented");
-  TORCH_CHECK(
-      keepdim == false,
-      "Vulkan sum.dim_IntList with keepdim=true is not implemented");
 
   std::set<int64_t> dims_set;
   if (opt_dim.has_value()) {

aten/src/ATen/test/vulkan_api_test.cpp

@@ -3702,12 +3702,12 @@ TEST_F(VulkanAPITest, sub_to_scalar_wrapped) {
   ASSERT_TRUE(check);
 }
 
-void test_sum_dim(const at::IntArrayRef input_shape, const at::IntArrayRef dim_list) {
+void test_sum_dim(const at::IntArrayRef input_shape, const at::IntArrayRef dim_list, bool keepdim=false) {
   const auto in_cpu = at::rand(input_shape, at::device(at::kCPU).dtype(at::kFloat));
   const auto in_vulkan = in_cpu.vulkan();
 
-  const auto out_cpu = at::sum(in_cpu, dim_list);
-  const auto out_vulkan = at::sum(in_vulkan, dim_list);
+  const auto out_cpu = at::sum(in_cpu, dim_list, keepdim);
+  const auto out_vulkan = at::sum(in_vulkan, dim_list, keepdim);
 
   const auto check = almostEqual(out_cpu, out_vulkan.cpu());
   if (!check) {
@@ -3753,6 +3753,39 @@ TEST_F(VulkanAPITest, sum_dim_4d) {
   test_sum_dim({10, 7, 5, 6}, {3, 2, 1});
 }
 
+TEST_F(VulkanAPITest, sum_dim_keepdim_2d) {
+  test_sum_dim({5, 7}, {-1}, true);
+  test_sum_dim({5, 7}, {-2}, true);
+}
+
+TEST_F(VulkanAPITest, sum_dim_keepdim_3d) {
+  test_sum_dim({9, 5, 7}, {-1}, true);
+  test_sum_dim({5, 9, 7}, {-2}, true);
+  test_sum_dim({7, 9, 5}, {-3}, true);
+
+  test_sum_dim({9, 5, 7}, {0, 1}, true);
+  test_sum_dim({5, 9, 7}, {0, 2}, true);
+  test_sum_dim({7, 9, 5}, {1, 2}, true);
+}
+
+TEST_F(VulkanAPITest, sum_dim_keepdim_4d) {
+  test_sum_dim({9, 5, 7, 11}, {-1}, true);
+  test_sum_dim({5, 9, 11, 7}, {-2}, true);
+  test_sum_dim({7, 11, 9, 5}, {-3}, true);
+  test_sum_dim({11, 7, 9, 5}, {-4}, true);
+
+  test_sum_dim({9, 5, 7, 11}, {0, 1}, true);
+  test_sum_dim({5, 9, 11, 7}, {0, 2}, true);
+  test_sum_dim({7, 11, 9, 5}, {0, 3}, true);
+  test_sum_dim({11, 7, 9, 5}, {1, 2}, true);
+  test_sum_dim({9, 5, 7, 11}, {1, 3}, true);
+  test_sum_dim({5, 9, 11, 7}, {2, 3}, true);
+
+  test_sum_dim({7, 11, 9, 5}, {-1, -2, -3}, true);
+  test_sum_dim({11, 7, 9, 5}, {-1, -2, -4}, true);
+  test_sum_dim({9, 5, 7, 11}, {-2, -3, -4}, true);
+}
+
 TEST_F(VulkanAPITest, uniform) {
   float a_min = -8.2f;
   float a_max = -1.4f;