FILTERBANK-ACCUMULATE microkernel test multiple rows

PiperOrigin-RevId: 469851386

3 files changed, 93 insertions, 17 deletions

diff --git a/test/filterbank-accumulate-microkernel-tester.h b/test/filterbank-accumulate-microkernel-tester.h
index 4a1a705a0..61e427893 100644
--- a/test/filterbank-accumulate-microkernel-tester.h
+++ b/test/filterbank-accumulate-microkernel-tester.h
@@ -56,12 +56,10 @@ class FilterbankAccumulateMicrokernelTester {
       std::generate(filterbank_widths.begin(), filterbank_widths.end(), std::ref(u8rng));
       const size_t num_channels = std::accumulate(filterbank_widths.cbegin(), filterbank_widths.cend(), 0);
 
-      std::vector<uint32_t> input(num_channels + XNN_EXTRA_BYTES / sizeof(int16_t));
+      std::vector<uint32_t> input(num_channels);
+      std::vector<uint16_t> weights(num_channels * 2);
       std::generate(input.begin(), input.end(), std::ref(u32rng));
-
-      std::vector<uint16_t> weights(num_channels * 2 + XNN_EXTRA_BYTES / sizeof(uint16_t));
       std::generate(weights.begin(), weights.end(), std::ref(u16rng));
-
       std::fill(output.begin(), output.end(), UINT64_C(0xCAFEB0BADEADBEAF));
 
       uint64_t weight_accumulator = 0;
diff --git a/test/u32-filterbank-accumulate.cc b/test/u32-filterbank-accumulate.cc
index 655c9e810..0669af136 100644
--- a/test/u32-filterbank-accumulate.cc
+++ b/test/u32-filterbank-accumulate.cc
@@ -1,4 +1,4 @@
-// Copyright 2019 Google LLC
+// Copyright 2022 Google LLC
 //
 // This source code is licensed under the BSD-style license found in the
 // LICENSE file in the root directory of this source tree.
@@ -17,14 +17,92 @@
 #include "filterbank-accumulate-microkernel-tester.h"
 
 
+#if XNN_ARCH_ARM
+  TEST(U32_FILTERBANK_ACCUMULATE__AARCH32_NEON_X1, rows_eq_1) {
+    TEST_REQUIRES_ARM_NEON;
+    FilterbankAccumulateMicrokernelTester()
+      .rows(1)
+      .Test(xnn_u32_filterbank_accumulate_ukernel__aarch32_neon_x1);
+  }
+
+  TEST(U32_FILTERBANK_ACCUMULATE__AARCH32_NEON_X1, rows_gt_1) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t rows = 2; rows < 10; rows++) {
+      FilterbankAccumulateMicrokernelTester()
+        .rows(rows)
+        .Test(xnn_u32_filterbank_accumulate_ukernel__aarch32_neon_x1);
+    }
+  }
+#endif  // XNN_ARCH_ARM
+
+
+#if XNN_ARCH_ARM
+  TEST(U32_FILTERBANK_ACCUMULATE__AARCH32_NEON_X2, rows_eq_1) {
+    TEST_REQUIRES_ARM_NEON;
+    FilterbankAccumulateMicrokernelTester()
+      .rows(1)
+      .Test(xnn_u32_filterbank_accumulate_ukernel__aarch32_neon_x2);
+  }
+
+  TEST(U32_FILTERBANK_ACCUMULATE__AARCH32_NEON_X2, rows_gt_1) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t rows = 2; rows < 10; rows++) {
+      FilterbankAccumulateMicrokernelTester()
+        .rows(rows)
+        .Test(xnn_u32_filterbank_accumulate_ukernel__aarch32_neon_x2);
+    }
+  }
+#endif  // XNN_ARCH_ARM
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(U32_FILTERBANK_ACCUMULATE__NEON_X1, rows_eq_1) {
+    TEST_REQUIRES_ARM_NEON;
+    FilterbankAccumulateMicrokernelTester()
+      .rows(1)
+      .Test(xnn_u32_filterbank_accumulate_ukernel__neon_x1);
+  }
+
+  TEST(U32_FILTERBANK_ACCUMULATE__NEON_X1, rows_gt_1) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t rows = 2; rows < 10; rows++) {
+      FilterbankAccumulateMicrokernelTester()
+        .rows(rows)
+        .Test(xnn_u32_filterbank_accumulate_ukernel__neon_x1);
+    }
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
+#if XNN_ARCH_ARM || XNN_ARCH_ARM64
+  TEST(U32_FILTERBANK_ACCUMULATE__NEON_X2, rows_eq_1) {
+    TEST_REQUIRES_ARM_NEON;
+    FilterbankAccumulateMicrokernelTester()
+      .rows(1)
+      .Test(xnn_u32_filterbank_accumulate_ukernel__neon_x2);
+  }
+
+  TEST(U32_FILTERBANK_ACCUMULATE__NEON_X2, rows_gt_1) {
+    TEST_REQUIRES_ARM_NEON;
+    for (size_t rows = 2; rows < 10; rows++) {
+      FilterbankAccumulateMicrokernelTester()
+        .rows(rows)
+        .Test(xnn_u32_filterbank_accumulate_ukernel__neon_x2);
+    }
+  }
+#endif  // XNN_ARCH_ARM || XNN_ARCH_ARM64
+
+
 TEST(U32_FILTERBANK_ACCUMULATE__SCALAR_X1, rows_eq_1) {
   FilterbankAccumulateMicrokernelTester()
     .rows(1)
     .Test(xnn_u32_filterbank_accumulate_ukernel__scalar_x1);
 }
 
-TEST(U32_FILTERBANK_ACCUMULATE__SCALAR_X1, rows_eq_2) {
-  FilterbankAccumulateMicrokernelTester()
-    .rows(2)
-    .Test(xnn_u32_filterbank_accumulate_ukernel__scalar_x1);
-}
+TEST(U32_FILTERBANK_ACCUMULATE__SCALAR_X1, rows_gt_1) {
+  for (size_t rows = 2; rows < 10; rows++) {
+    FilterbankAccumulateMicrokernelTester()
+      .rows(rows)
+      .Test(xnn_u32_filterbank_accumulate_ukernel__scalar_x1);
+  }
+}
\ No newline at end of file
diff --git a/tools/generate-filterbank-accumulate-test.py b/tools/generate-filterbank-accumulate-test.py
index 2139a897e..9d7dd92c5 100755
--- a/tools/generate-filterbank-accumulate-test.py
+++ b/tools/generate-filterbank-accumulate-test.py
@@ -47,15 +47,15 @@ TEST(${TEST_NAME}, rows_eq_1) {
     .Test(${", ".join(TEST_ARGS)});
 }
 
-TEST(${TEST_NAME}, rows_eq_2) {
+TEST(${TEST_NAME}, rows_gt_1) {
   $if ISA_CHECK:
     ${ISA_CHECK};
-  FilterbankAccumulateMicrokernelTester()
-    .rows(2)
-    .Test(${", ".join(TEST_ARGS)});
+  for (size_t rows = 2; rows < 10; rows++) {
+    FilterbankAccumulateMicrokernelTester()
+      .rows(rows)
+      .Test(${", ".join(TEST_ARGS)});
+  }
 }
-
-
 """
 
 
@@ -96,7 +96,7 @@ def main(args):
       raise ValueError("expected a list of micro-kernels in the spec")
 
     tests = """\
-// Copyright 2019 Google LLC
+// Copyright 2022 Google LLC
 //
 // This source code is licensed under the BSD-style license found in the
 // LICENSE file in the root directory of this source tree.