[ET-VK] benchmarking linear_qta8a_qga4w_qta8o

# Context
In order to benchmark the linear_qta8a_qga4w_qta8o operator, this diff adds functionality for int32 in the operator since Android does not support int8 operations. Functionally, this change should not be used, and this is only being provided in order to add support for int32 in android cases.

# Changes
Created a new benchmarking file for the linear_qta8a_qga4w_qta8o operator, which tests against various test cases. Also made a few changes to the existing operator with support for int32 along with a new variant.

Differential Revision: [D77384894](https://our.internmc.facebook.com/intern/diff/D77384894/)

ghstack-source-id: 292886977
Pull Request resolved: #12007

Author: morelos

backends/vulkan/runtime/graph/ops/glsl/linear_qta8a_qga4w_qta8o_coop.glsl

@@ -19,7 +19,7 @@
 #define NWORKERS 8
 
 ${define_required_extensions(DTYPE)}
-$if WEIGHT_STORAGE == "buffer":
+$if WEIGHT_STORAGE == "buffer" and WEIGHT_DTYPE == "uint8":
   ${define_required_extensions("uint8")}
 
 #extension GL_EXT_control_flow_attributes : require
@@ -28,7 +28,7 @@ layout(std430) buffer;
 
 ${layout_declare_tensor(B, "w", "t_out", DTYPE, OUT_STORAGE, is_scalar_array=False)}
 ${layout_declare_tensor(B, "r", "t_mat1", DTYPE, IN_STORAGE, is_scalar_array=False)}
-${layout_declare_tensor(B, "r", "t_qmat2", "uint8", WEIGHT_STORAGE, is_scalar_array=False)}
+${layout_declare_tensor(B, "r", "t_qmat2", WEIGHT_DTYPE, WEIGHT_STORAGE, is_scalar_array=False)}
 ${layout_declare_tensor(B, "r", "t_qparams", "float", PARAMS_STORAGE, is_scalar_array=False)}
 ${layout_declare_tensor(B, "r", "t_input_scale", "float", "buffer", is_scalar_array=True)}
 ${layout_declare_tensor(B, "r", "t_input_zero_point", "int", "buffer", is_scalar_array=True)}

backends/vulkan/runtime/graph/ops/glsl/linear_qta8a_qga4w_qta8o_coop.yaml

@@ -7,6 +7,7 @@
 linear_qta8a_qga4w_qta8o_coop:
   parameter_names_with_default_values:
     DTYPE: int8
+    WEIGHT_DTYPE: uint8
     OUT_STORAGE: texture3d
     IN_STORAGE: texture3d
     WEIGHT_STORAGE: texture2d
@@ -24,3 +25,6 @@ linear_qta8a_qga4w_qta8o_coop:
     - NAME: linear_qta8a_qga4w_qta8o_coop_buffer_texture2d_buffer_int8
       OUT_STORAGE: buffer
       WEIGHT_STORAGE: buffer
+    - NAME: linear_qta8a_qga4w_qta8o_coop_texture3d_texture3d_texture2d_int32
+      DTYPE: int32
+      WEIGHT_DTYPE: uint32

backends/vulkan/runtime/graph/ops/glsl/linear_qta8a_qga4w_qta8o_tiled.glsl

@@ -16,7 +16,7 @@
 #define TILE_ROWS ${TILE_ROWS}
 
 ${define_required_extensions(DTYPE)}
-$if WEIGHT_STORAGE == "buffer":
+$if WEIGHT_STORAGE == "buffer" and WEIGHT_DTYPE == "uint8":
   ${define_required_extensions("uint8")}
 
 #extension GL_EXT_control_flow_attributes : require
@@ -25,7 +25,7 @@ layout(std430) buffer;
 
 ${layout_declare_tensor(B, "w", "t_out", DTYPE, OUT_STORAGE, is_scalar_array=False)}
 ${layout_declare_tensor(B, "r", "t_mat1", DTYPE, IN_STORAGE, is_scalar_array=False)}
-${layout_declare_tensor(B, "r", "t_qmat2", "uint8", WEIGHT_STORAGE, is_scalar_array=False)}
+${layout_declare_tensor(B, "r", "t_qmat2", WEIGHT_DTYPE, WEIGHT_STORAGE, is_scalar_array=False)}
 ${layout_declare_tensor(B, "r", "t_qparams", "float", PARAMS_STORAGE, is_scalar_array=False)}
 ${layout_declare_tensor(B, "r", "t_input_scale", "float", "buffer", is_scalar_array=True)}
 ${layout_declare_tensor(B, "r", "t_input_zero_point", "int", "buffer", is_scalar_array=True)}

backends/vulkan/runtime/graph/ops/glsl/linear_qta8a_qga4w_qta8o_tiled.yaml

@@ -7,6 +7,7 @@
 linear_qta8a_qga4w_qta8o_tiled:
   parameter_names_with_default_values:
     DTYPE: int8
+    WEIGHT_DTYPE: uint8
     OUT_STORAGE: texture3d
     IN_STORAGE: texture3d
     WEIGHT_STORAGE: texture2d
@@ -24,3 +25,6 @@ linear_qta8a_qga4w_qta8o_tiled:
     - NAME: linear_qta8a_qga4w_qta8o_tiled_buffer_texture2d_buffer_int8
       OUT_STORAGE: buffer
       WEIGHT_STORAGE: buffer
+    - NAME: linear_qta8a_qga4w_qta8o_tiled_texture3d_texture3d_texture2d_int32
+      DTYPE: int32
+      WEIGHT_DTYPE: uint32