From 9f6404bac66d813ff084c3a1b8323e18933b7923 Mon Sep 17 00:00:00 2001
From: Allison Vacanti <alliepiper16@gmail.com>
Date: Tue, 16 Mar 2021 15:12:53 -0400
Subject: [PATCH] Pad range max for floating point types.

This avoids an annoying case where the max value is dropped due to
rounding errors.

Adds a few other missing test cases for `nvbench::range`, too.

Fixes #3.
---
 nvbench/range.cuh      |  4 +++
 testing/CMakeLists.txt |  1 +
 testing/range.cu       | 82 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 87 insertions(+)
 create mode 100644 testing/range.cu

diff --git a/nvbench/range.cuh b/nvbench/range.cuh
index 12d4d2d..f0e8255 100644
--- a/nvbench/range.cuh
+++ b/nvbench/range.cuh
@@ -38,6 +38,10 @@ template <typename InT,
           typename OutT = nvbench::detail::range_output_t<InT>>
 auto range(InT start, InT end, InT stride = InT{1})
 {
+  if constexpr (std::is_floating_point_v<InT>)
+  { // Pad end to account for floating point errors:
+    end += (stride / InT{2});
+  }
   using result_t = std::vector<OutT>;
   result_t result;
   for (; start <= end; start += stride)
diff --git a/testing/CMakeLists.txt b/testing/CMakeLists.txt
index a983dba..93dda8e 100644
--- a/testing/CMakeLists.txt
+++ b/testing/CMakeLists.txt
@@ -9,6 +9,7 @@ set(test_srcs
   int64_axis.cu
   named_values.cu
   option_parser.cu
+  range.cu
   runner.cu
   state.cu
   state_generator.cu
diff --git a/testing/range.cu b/testing/range.cu
new file mode 100644
index 0000000..35e158e
--- /dev/null
+++ b/testing/range.cu
@@ -0,0 +1,82 @@
+/*
+ *  Copyright 2021 NVIDIA Corporation
+ *
+ *  Licensed under the Apache License, Version 2.0 with the LLVM exception
+ *  (the "License"); you may not use this file except in compliance with
+ *  the License.
+ *
+ *  You may obtain a copy of the License at
+ *
+ *      http://llvm.org/foundation/relicensing/LICENSE.txt
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <nvbench/range.cuh>
+
+#include "test_asserts.cuh"
+
+void test_basic()
+{
+  ASSERT((nvbench::range(0, 6) ==
+          std::vector<nvbench::int64_t>{0, 1, 2, 3, 4, 5, 6}));
+  ASSERT((nvbench::range(0, 6, 1) ==
+          std::vector<nvbench::int64_t>{0, 1, 2, 3, 4, 5, 6}));
+  ASSERT(
+    (nvbench::range(0, 6, 2) == std::vector<nvbench::int64_t>{0, 2, 4, 6}));
+  ASSERT((nvbench::range(0, 6, 3) == std::vector<nvbench::int64_t>{0, 3, 6}));
+  ASSERT((nvbench::range(0, 6, 4) == std::vector<nvbench::int64_t>{0, 4}));
+  ASSERT((nvbench::range(0, 6, 5) == std::vector<nvbench::int64_t>{0, 5}));
+  ASSERT((nvbench::range(0, 6, 7) == std::vector<nvbench::int64_t>{0}));
+}
+
+void test_result_type()
+{
+  // All ints should turn into int64 by default:
+  ASSERT((std::is_same_v<decltype(nvbench::range(0ll, 1ll)),
+                         std::vector<nvbench::int64_t>>));
+  ASSERT((std::is_same_v<decltype(nvbench::range(0, 1)),
+                         std::vector<nvbench::int64_t>>));
+  ASSERT((std::is_same_v<decltype(nvbench::range(0u, 1u)),
+                         std::vector<nvbench::int64_t>>));
+
+  // All floats should turn into float64 by default:
+  ASSERT((std::is_same_v<decltype(nvbench::range(0., 1.)),
+                         std::vector<nvbench::float64_t>>));
+  ASSERT((std::is_same_v<decltype(nvbench::range(0.f, 1.f)),
+                         std::vector<nvbench::float64_t>>));
+
+  // Other types may be explicitly specified:
+  ASSERT((std::is_same_v<decltype(nvbench::range<nvbench::float32_t,
+                                                 nvbench::float32_t>(0.f, 1.f)),
+                         std::vector<nvbench::float32_t>>));
+  ASSERT((std::is_same_v<
+          decltype(nvbench::range<nvbench::int32_t, nvbench::int32_t>(0, 1)),
+          std::vector<nvbench::int32_t>>));
+}
+
+void test_fp_tolerance()
+{
+  // Make sure that the range is padded a bit for floats to prevent rounding
+  // errors from skipping `end`. This test will trigger failures without
+  // the padding.
+  const nvbench::float32_t start  = 0.1;
+  const nvbench::float32_t stride = 1e-4;
+  for (std::size_t size = 1; size < 1024; ++size)
+  {
+    const nvbench::float32_t end = start + stride * (size - 1);
+    ASSERT_MSG(nvbench::range(start, end, stride).size() == size,
+               "size={}", size);
+  }
+}
+
+int main()
+{
+  test_basic();
+  test_result_type();
+  test_fp_tolerance();
+}