From 6622c1dc61fbe650a0d8e9cd42446e9e5293ea4b Mon Sep 17 00:00:00 2001
From: Hannes Vogt <hannes@havogt.de>
Date: Tue, 18 Jun 2024 14:57:41 +0200
Subject: [PATCH] Introduce GT_PROMISE for __builtin_assume (#1785)

Credits to @lukasm91 for hinting to `__builtin_assume` in `deref`.

On more recent compilers (than 11.2 what we have in CI on daint), will
improve codegen to the level of #1779, but is save.

---------

Co-authored-by: Felix Thaler <felix.thaler@nummi.ch>
---
 include/gridtools/common/defs.hpp     | 20 +++++++++++++++++++-
 include/gridtools/fn/unstructured.hpp |  3 ++-
 2 files changed, 21 insertions(+), 2 deletions(-)

diff --git a/include/gridtools/common/defs.hpp b/include/gridtools/common/defs.hpp
index b46527f53..559dd2253 100644
--- a/include/gridtools/common/defs.hpp
+++ b/include/gridtools/common/defs.hpp
@@ -8,6 +8,7 @@
  * SPDX-License-Identifier: BSD-3-Clause
  */
 #pragma once
+#include <cassert>
 
 namespace gridtools {
     using int_t = int;
@@ -26,6 +27,22 @@ namespace gridtools {
 #endif
 #endif
 
+#if defined(__has_builtin)
+#if __has_builtin(__builtin_assume)
+#define GT_ASSUME(x) __builtin_assume(x)
+#else
+#define GT_ASSUME(x)
+#endif
+#else
+#define GT_ASSUME(x)
+#endif
+
+#ifdef NDEBUG
+#define GT_PROMISE(x) GT_ASSUME(x)
+#else
+#define GT_PROMISE(x) assert(x)
+#endif
+
 #ifdef __cpp_consteval
 #define GT_CONSTEVAL consteval
 #else
@@ -53,7 +70,8 @@ namespace gridtools {
 #endif
 
 #if defined(__NVCC__) && (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 1 && __CUDACC_VER_MINOR__ <= 4)
-// enables workaround for CTAD/constexpr issues in CUDA 12.1, 12.2, 12.3, 12.4 (https://github.com/GridTools/gridtools/issues/1766)
+// enables workaround for CTAD/constexpr issues in CUDA 12.1, 12.2, 12.3, 12.4
+// (https://github.com/GridTools/gridtools/issues/1766)
 #define GT_NVCC_WORKAROUND_1766 1
 #else
 #define GT_NVCC_WORKAROUND_1766 0
diff --git a/include/gridtools/fn/unstructured.hpp b/include/gridtools/fn/unstructured.hpp
index 136930c19..9844185a0 100644
--- a/include/gridtools/fn/unstructured.hpp
+++ b/include/gridtools/fn/unstructured.hpp
@@ -11,6 +11,7 @@
 
 #include <functional>
 
+#include "../common/defs.hpp"
 #include "../common/hymap.hpp"
 #include "../meta/logical.hpp"
 #include "../sid/concept.hpp"
@@ -76,7 +77,7 @@ namespace gridtools::fn {
 
         template <class Tag, class Ptr, class Strides, class Domain>
         GT_FUNCTION constexpr auto deref(iterator<Tag, Ptr, Strides, Domain> const &it) {
-            assert(can_deref(it));
+            GT_PROMISE(can_deref(it));
             decltype(auto) stride = host_device::at_key<Tag>(sid::get_stride<dim::horizontal>(it.m_strides));
             return *sid::shifted(it.m_ptr, stride, it.m_index);
         }