Update base for Update on "[ET-VK][ez] Implement helper functions to get fastest moving dim"

Add C++ and GLSL helpers to query the fastest moving dimension (the
dimension with stride 1 in buffer layout). This is useful for optimizing
memory access patterns in shaders, as iterating along the fastest moving
dimension maximizes cache locality.

The C++ `fastest_whcn_dim()` method accounts for block-transposed layouts by
returning `outer_packed_dim` instead of `packed_dim` when applicable. A
corresponding GLSL macro extracts this info from the hashed layout.

Differential Revision: [D92061369](https://our.internmc.facebook.com/intern/diff/D92061369/)

[ghstack-poisoned]

Author: ssjia

.github/workflows/pull.yml

@@ -1137,7 +1137,7 @@ jobs:
         ./cmake-out/backends/vulkan/test/custom_ops/q8csw_conv2d
         ./cmake-out/backends/vulkan/test/custom_ops/q4gsw_linear
         ./cmake-out/backends/vulkan/test/custom_ops/choose_qparams_per_row
-        ./cmake-out/backends/vulkan/test/custom_ops/qdq8ta_conv2d_activations
+        ./cmake-out/backends/vulkan/test/custom_ops/test_q8ta_qdq
         ./cmake-out/backends/vulkan/test/custom_ops/q8ta_q8ta_q8to_add
 
         # "Classic" Operator tests

backends/vulkan/test/custom_ops/CMakeLists.txt

@@ -97,7 +97,6 @@ if(TARGET vulkan_backend)
   add_operator_prototype(q8csw_conv2d)
   add_operator_prototype(q4gsw_linear)
   add_operator_prototype(choose_qparams_per_row)
-  add_operator_prototype(qdq8ta_conv2d_activations)
   add_operator_prototype(test_q8ta_qdq)
   add_operator_prototype(q8ta_q8csw_q8to_conv2d)
   add_operator_prototype(q8ta_q8csw_q8to_conv2d_dw)

backends/vulkan/test/custom_ops/qdq8ta_conv2d_activations.cpp

@@ -91,7 +91,6 @@ def define_common_targets(is_fbcode = False):
     define_custom_op_test_binary("q8csw_conv2d")
     define_custom_op_test_binary("choose_qparams_per_row")
     define_custom_op_test_binary("q4gsw_linear")
-    define_custom_op_test_binary("qdq8ta_conv2d_activations")
     define_custom_op_test_binary("test_q8ta_qdq")
     define_custom_op_test_binary("q8ta_q8csw_q8to_conv2d")
     define_custom_op_test_binary("q8ta_q8csw_q8to_conv2d_dw")