[ET-VK][ez] Refactor yaml configs for SDPA shaders

Title says it all! Use the new combos codegen API which makes it easier to express generating storage type combinations.

Differential Revision: [D86226138](https://our.internmc.facebook.com/intern/diff/D86226138/)

[ghstack-poisoned]

Author: ssjia

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

@@ -12,10 +12,14 @@ sdpa_compute_attn_weights_coop:
     TILE_K4: 1
     TILE_N4: 1
   generate_variant_forall:
+    combination:
+      parameter_names: [IO_STORAGE, K_CACHE_STORAGE]
+      combos:
+        - parameter_values: [texture3d, texture3d]
+        - parameter_values: [buffer, texture3d]
+        - parameter_values: [buffer, buffer]
     DTYPE:
       - VALUE: float
       - VALUE: half
   shader_variants:
-    - NAME: sdpa_compute_attn_weights_coop_texture3d_texture3d
-    - NAME: sdpa_compute_attn_weights_coop_buffer_texture3d
-      IO_STORAGE: buffer
+    - NAME: sdpa_compute_attn_weights_coop

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

@@ -13,10 +13,14 @@ sdpa_compute_attn_weights_tiled:
     TILE_K4: 1
     TILE_N4: 1
   generate_variant_forall:
+    combination:
+      parameter_names: [IO_STORAGE, K_CACHE_STORAGE]
+      combos:
+        - parameter_values: [texture3d, texture3d]
+        - parameter_values: [buffer, texture3d]
+        - parameter_values: [buffer, buffer]
     DTYPE:
       - VALUE: float
       - VALUE: half
   shader_variants:
-    - NAME: sdpa_compute_attn_weights_tiled_texture3d_texture3d
-    - NAME: sdpa_compute_attn_weights_tiled_buffer_texture3d
-      IO_STORAGE: buffer
+    - NAME: sdpa_compute_attn_weights_tiled

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

@@ -12,10 +12,14 @@ sdpa_compute_out_coop:
     TILE_K4: 1
     TILE_N4: 1
   generate_variant_forall:
+    combination:
+      parameter_names: [IO_STORAGE, V_CACHE_STORAGE]
+      combos:
+        - parameter_values: [texture3d, texture3d]
+        - parameter_values: [buffer, texture3d]
+        - parameter_values: [buffer, buffer]
     DTYPE:
       - VALUE: float
       - VALUE: half
   shader_variants:
-    - NAME: sdpa_compute_out_coop_texture3d_texture3d
-    - NAME: sdpa_compute_out_coop_buffer_texture3d
-      IO_STORAGE: buffer
+    - NAME: sdpa_compute_out_coop

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

@@ -13,10 +13,14 @@ sdpa_compute_out_tiled:
     TILE_K4: 1
     TILE_N4: 1
   generate_variant_forall:
+    combination:
+      parameter_names: [IO_STORAGE, V_CACHE_STORAGE]
+      combos:
+        - parameter_values: [texture3d, texture3d]
+        - parameter_values: [buffer, texture3d]
+        - parameter_values: [buffer, buffer]
     DTYPE:
       - VALUE: float
       - VALUE: half
   shader_variants:
-    - NAME: sdpa_compute_out_tiled_texture3d_texture3d
-    - NAME: sdpa_compute_out_tiled_buffer_texture3d
-      IO_STORAGE: buffer
+    - NAME: sdpa_compute_out_tiled

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

@@ -5,6 +5,8 @@
 #define IN_VEC4_T ${texel_load_type(DTYPE, INPUT_STORAGE)}
 #define T ${buffer_scalar_type(DTYPE)}
 
+$if OUTPUT_STORAGE == "buffer":
+  #define OUTPUT_BUFFER
 $if INPUT_STORAGE == "buffer":
   #define INPUT_BUFFER
 

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

@@ -10,10 +10,14 @@ sdpa_kv_cache_update:
     INPUT_STORAGE: texture3d
     OUTPUT_STORAGE: texture3d
   generate_variant_forall:
+    combination:
+      parameter_names: [OUTPUT_STORAGE, INPUT_STORAGE]
+      combos:
+        - parameter_values: [texture3d, texture3d]
+        - parameter_values: [texture3d, buffer]
+        - parameter_values: [buffer, buffer]
     DTYPE:
       - VALUE: half
       - VALUE: float
   shader_variants:
-    - NAME: sdpa_kv_cache_update_texture3d
-    - NAME: sdpa_kv_cache_update_buffer
-      INPUT_STORAGE: buffer
+    - NAME: sdpa_kv_cache_update

backends/vulkan/runtime/graph/ops/impl/SDPA.cpp

@@ -282,6 +282,7 @@ void add_sdpa_kv_cache_update_node(
     const ValueRef projected,
     const ValueRef cache) {
   std::string kernel_name("sdpa_kv_cache_update");
+  add_storage_type_suffix(kernel_name, graph.storage_type_of(cache));
   add_storage_type_suffix(kernel_name, graph.storage_type_of(projected));
   add_dtype_suffix(kernel_name, graph.dtype_of(projected));