[ExecuTorch][WebGPU] Generate extrema and unary shader variants

Pull Request resolved: #21450

**Generate extrema and unary shader families**

The extrema reductions and ten no-parameter unary kernels duplicated shader skeletons that could drift independently. This consolidates amax/amin behind one extrema template and abs/cos/exp/hardswish/neg/round/rsqrt/sin/sqrt/tanh behind one unary template while preserving the generated runtime payloads.

Key changes:
- Generate amax/amin from one extrema manifest.
- Generate ten unary payloads from one operator-expression manifest.
- Lock expanded bytes, registry entries, delegation, and boundary numerics.

The attempted Unary lifecycle migration is intentionally not part of the stack: its performance campaign did not produce an authoritative passing result, so the Unary builder, interface, and activation/sigmoid call sites are restored to their pre-migration bytes.

Co-authored-with: Claude Code.
ghstack-source-id: 411961475
@exported-using-ghexport

Differential Revision: [D113979760](https://our.internmc.facebook.com/intern/diff/D113979760/)

Author: JCNTH

backends/webgpu/runtime/WebGPUShaderRegistry.cpp

@@ -13,8 +13,6 @@
 #include <executorch/backends/webgpu/runtime/ops/adamw/adamw_step_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/add/binary_add_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/addmm/addmm_tiled_wgsl.h>
-#include <executorch/backends/webgpu/runtime/ops/amax/amax_wgsl.h>
-#include <executorch/backends/webgpu/runtime/ops/amin/amin_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/argmax/arg_reduce_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/avg_pool2d/avg_pool2d_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/batch_norm/batch_norm_wgsl.h>
@@ -48,6 +46,8 @@
 #include <executorch/backends/webgpu/runtime/ops/et_vk_sdpa/et_vk_sdpa_qk_entry_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/et_vk_sdpa/et_vk_sdpa_qk_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/expand_copy/expand_copy_wgsl.h>
+#include <executorch/backends/webgpu/runtime/ops/extrema/amax_wgsl.h>
+#include <executorch/backends/webgpu/runtime/ops/extrema/amin_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/fill/fill_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/flip/flip_wgsl.h>
 #include <executorch/backends/webgpu/runtime/ops/floor_divide/binary_floor_divide_wgsl.h>

backends/webgpu/runtime/ops/amax/Reduce.cpp

@@ -9,7 +9,7 @@
 #include <executorch/backends/webgpu/runtime/WebGPUGraph.h>
 #include <executorch/backends/webgpu/runtime/WebGPUUtils.h>
 #include <executorch/backends/webgpu/runtime/ops/OperatorRegistry.h>
-#include <executorch/backends/webgpu/runtime/ops/amax/amax_wgsl.h>
+#include <executorch/backends/webgpu/runtime/ops/extrema/amax_wgsl.h>
 
 #include <webgpu/webgpu.h>
 

backends/webgpu/runtime/ops/amin/Reduce.cpp

@@ -9,7 +9,7 @@
 #include <executorch/backends/webgpu/runtime/WebGPUGraph.h>
 #include <executorch/backends/webgpu/runtime/WebGPUUtils.h>
 #include <executorch/backends/webgpu/runtime/ops/OperatorRegistry.h>
-#include <executorch/backends/webgpu/runtime/ops/amin/amin_wgsl.h>
+#include <executorch/backends/webgpu/runtime/ops/extrema/amin_wgsl.h>
 
 #include <webgpu/webgpu.h>
 

backends/webgpu/runtime/ops/amin/amin.wgsl

@@ -12,7 +12,7 @@
 
 namespace executorch::backends::webgpu {
 
-// @generated from amax.wgsl - DO NOT EDIT.
+// @generated from extrema.wgsl - DO NOT EDIT.
 // wgsl-sha256: 35fc059d7c72caa17f9cb1128823ecfd8f75be4ce24b6cd4f9629a97b52f64c0
 inline constexpr const char* kAmaxWGSL = R"(
 @group(0) @binding(0) var<storage, read> input: array<f32>;

…ends/webgpu/runtime/ops/amax/amax_wgsl.h …s/webgpu/runtime/ops/extrema/amax_wgsl.hbackends/webgpu/runtime/ops/amax/amax_wgsl.h renamed to backends/webgpu/runtime/ops/extrema/amax_wgsl.h backends/webgpu/runtime/ops/amax/amax_wgsl.h renamed to backends/webgpu/runtime/ops/extrema/amax_wgsl.h

@@ -12,7 +12,7 @@
 
 namespace executorch::backends::webgpu {
 
-// @generated from amin.wgsl - DO NOT EDIT.
+// @generated from extrema.wgsl - DO NOT EDIT.
 // wgsl-sha256: 8cb6035ae4d34eb2a6cc973d93d9847905722e967239c96033fccfe3a1943cb2
 inline constexpr const char* kAminWGSL = R"(
 @group(0) @binding(0) var<storage, read> input: array<f32>;

…ends/webgpu/runtime/ops/amin/amin_wgsl.h …s/webgpu/runtime/ops/extrema/amin_wgsl.hbackends/webgpu/runtime/ops/amin/amin_wgsl.h renamed to backends/webgpu/runtime/ops/extrema/amin_wgsl.h backends/webgpu/runtime/ops/amin/amin_wgsl.h renamed to backends/webgpu/runtime/ops/extrema/amin_wgsl.h

@@ -32,7 +32,7 @@ fn main(
     var acc = input[base];
     var i = lid.x;
     while (i < params.reduce_size) {
-        acc = max(acc, input[base + i]);
+        acc = ${REDUCE_FN}(acc, input[base + i]);
         i = i + wg_size;
     }
     partials[lid.x] = acc;
@@ -42,7 +42,7 @@ fn main(
     if (lid.x == 0u) {
         var m = partials[0];
         for (var t = 1u; t < wg_size; t = t + 1u) {
-            m = max(m, partials[t]);
+            m = ${REDUCE_FN}(m, partials[t]);
         }
         output[row] = m;
     }

…ckends/webgpu/runtime/ops/amax/amax.wgsl …/webgpu/runtime/ops/extrema/extrema.wgslbackends/webgpu/runtime/ops/amax/amax.wgsl renamed to backends/webgpu/runtime/ops/extrema/extrema.wgsl backends/webgpu/runtime/ops/amax/amax.wgsl renamed to backends/webgpu/runtime/ops/extrema/extrema.wgsl

@@ -0,0 +1,13 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+extrema:
+  parameter_names_with_default_values:
+    REDUCE_FN: max
+  shader_variants:
+    - NAME: amax
+    - NAME: amin
+      REDUCE_FN: min

backends/webgpu/runtime/ops/extrema/extrema.yaml

@@ -12,7 +12,7 @@
 
 namespace executorch::backends::webgpu {
 
-// @generated from abs.wgsl - DO NOT EDIT.
+// @generated from unary.wgsl - DO NOT EDIT.
 // wgsl-sha256: 39d3c163fdf6a92286828f4b3217e00294e3ca5634a878ed5fd34e3b1cdf0a27
 inline constexpr const char* kAbsWGSL = R"(
 @group(0) @binding(0) var<storage, read> input: array<f32>;