Reformat using clang-format 20

Author: stijnh

include/kernel_launcher/compiler.h

@@ -101,8 +101,9 @@ struct Compiler: ICompiler {
      */
     template<typename C>
     Compiler(C&& compiler) :
-        inner_(std::make_shared<typename std::decay<C>::type>(
-            std::forward<C>(compiler))) {}
+        inner_(
+            std::make_shared<typename std::decay<C>::type>(
+                std::forward<C>(compiler))) {}
 
     void compile_ptx(
         KernelDef def,

include/kernel_launcher/utils.h

@@ -2,6 +2,7 @@
 #define KERNEL_LAUNCHER_UTILS_H
 
 #include <cuda_runtime_api.h>
+#include <stdint.h>
 
 #include <functional>
 #include <iosfwd>
@@ -11,7 +12,6 @@
 #include <type_traits>
 #include <typeindex>
 #include <vector>
-#include <stdint.h>
 
 namespace kernel_launcher {
 

src/arg.cpp

@@ -37,7 +37,7 @@ KernelArg::KernelArg(KernelArg&& that) noexcept : KernelArg() {
 
 KernelArg::~KernelArg() {
     if (is_scalar() && !is_inline_scalar(type_)) {
-        delete[](char*) data_.large_scalar;
+        delete[] (char*)data_.large_scalar;
     }
 }
 

src/wisdom.cpp

@@ -655,9 +655,10 @@ WisdomSettings::WisdomSettings(
     std::string wisdom_dir,
     std::string capture_dir,
     std::vector<CaptureRule> capture_rules) :
-    WisdomSettings(std::make_shared<DefaultWisdomSettings>(
-        std::vector<std::string> {std::move(wisdom_dir)},
-        std::move(capture_dir),
-        std::move(capture_rules))) {}
+    WisdomSettings(
+        std::make_shared<DefaultWisdomSettings>(
+            std::vector<std::string> {std::move(wisdom_dir)},
+            std::move(capture_dir),
+            std::move(capture_rules))) {}
 
 }  // namespace kernel_launcher

test.cu

@@ -0,0 +1,16 @@
+namespace bar {
+namespace foo {
+
+#pragma kernel_tuner tune(block_size=32, 64, 128, 256)
+#pragma kernel_tuner tune(tile_factor=1, 2, 3, 4)
+#pragma kernel_tuner problem_size(n)
+#pragma kernel_tuner grid_divisor(block_size * tile_factor)
+template <typename T, int tile_factor>
+__global__ void baz(int n, T* input) {
+    if (threadIdx.x < 10) {
+        return a[threadIdx.x];
+    }
+}
+
+} // namespace foo
+} // namespace bar

tests/registry.cpp

@@ -44,8 +44,9 @@ TEST_CASE("KernelDescriptor", "[CUDA]") {
     KERNEL_LAUNCHER_CUDA_CHECK(cuCtxCreate(&ctx, 0, 0));
     auto a = KernelDescriptor(VectorAddDescriptor());
     auto b = KernelDescriptor(std::make_shared<MatrixMulDescriptor>(1));
-    auto c = KernelDescriptor(std::shared_ptr<IKernelDescriptor>(
-        std::make_shared<MatrixMulDescriptor>(1)));
+    auto c = KernelDescriptor(
+        std::shared_ptr<IKernelDescriptor>(
+            std::make_shared<MatrixMulDescriptor>(1)));
     auto d = KernelDescriptor(MatrixMulDescriptor(2));
 
     // A KernelDescriptor is based on the current CUDA context.