kernels/optimized: add on-device verify_optimized_kernels binary

Standalone aarch64 binary that cross-checks opt_grid_sampler_2d_out and
opt_sum_dim_out against an fp32 reference derived from the portable
kernel (portable run on up-cast fp32 inputs, then down-cast to fp16).
Reference is independent of portable's own fp16 path, so the test stays
meaningful regardless of pytorch#19117's merge state.

Pass/fail uses numpy.testing.assert_allclose semantics:
  |a - b| <= abs_tol + rel_tol * |b|
Avoids the "relative error explodes at zero crossings" trap for
mean-zero reductions and bilinear samples near cancellation points.

Opt-in via -DEXECUTORCH_BUILD_OPTIMIZED_VERIFY=ON so default builds are
unaffected. Build + run:

  cmake -DEXECUTORCH_BUILD_OPTIMIZED_VERIFY=ON ...
  cmake --build <out> --target verify_optimized_kernels
  adb push <out>/kernels/optimized/verify_optimized_kernels /data/local/tmp/
  adb shell /data/local/tmp/verify_optimized_kernels

Exits 0 on all-pass; reports max_abs / max_rel(far) / near_zero / viol
per test case. 12 test cases across grid_sampler and sum, covering the
shapes the polycam depth model uses plus a few edge cases (odd channel
count, align_corners=1, multi-batch).

Author: jgibson2

kernels/optimized/CMakeLists.txt

@@ -98,6 +98,30 @@ gen_operators_lib(
   executorch_core
 )
 
+# On-device verifier for optimized grid_sampler_2d / sum.IntList_out.
+# Opt-in via -DEXECUTORCH_BUILD_OPTIMIZED_VERIFY=ON so it doesn't affect
+# default AAR / library builds. Cross-checks both ops against an fp32
+# reference derived from the portable kernel; non-zero exit on divergence.
+if(EXECUTORCH_BUILD_OPTIMIZED_VERIFY)
+  add_executable(
+    verify_optimized_kernels ${EXECUTORCH_ROOT}/kernels/optimized/verify.cpp
+  )
+  target_link_libraries(
+    verify_optimized_kernels
+    PRIVATE optimized_kernels portable_kernels executorch_core
+  )
+  target_compile_options(
+    verify_optimized_kernels PRIVATE ${_common_compile_options}
+  )
+  if(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64|arm64"
+     OR ANDROID_ABI STREQUAL "arm64-v8a"
+  )
+    target_compile_options(
+      verify_optimized_kernels PRIVATE -march=armv8.2-a+fp16
+    )
+  endif()
+endif()
+
 install(
   # eigen_blas doesn't export itself, so we have to do our own install to export
   # it.