[libc] Efficiently implement aligned_alloc for AMDGPU (#146585)

Summary:
This patch uses the actual allocator interface to implement
`aligned_alloc`. We do this by simply rounding up the amount allocated.
Because of how index calculation works, any offset within an allocated
pointer will still map to the same chunk, so we can just adjust
internally and it will free all the same.

libc/test/integration/src/stdlib/gpu/CMakeLists.txt

@@ -32,6 +32,21 @@ if(NOT LIBC_TARGET_ARCHITECTURE_IS_NVPTX)
       --blocks 1024
   )
 
+  add_integration_test(
+    aligned_alloc
+    SUITE
+      stdlib-gpu-integration-tests
+    SRCS
+      aligned_alloc.cpp
+    DEPENDS
+      libc.src.stdlib.aligned_alloc
+      libc.src.stdlib.malloc
+      libc.src.stdlib.free
+    LOADER_ARGS
+      --threads 256
+      --blocks 128
+  )
+
   add_integration_test(
     malloc_stress
     SUITE

libc/test/integration/src/stdlib/gpu/aligned_alloc.cpp

@@ -0,0 +1,29 @@
+#include "test/IntegrationTest/test.h"
+
+#include "src/__support/GPU/utils.h"
+#include "src/stdlib/aligned_alloc.h" // Adjust path if needed
+#include "src/stdlib/free.h"
+
+using namespace LIBC_NAMESPACE;
+
+TEST_MAIN(int, char **, char **) {
+  // aligned_alloc with valid alignment and size
+  void *ptr = LIBC_NAMESPACE::aligned_alloc(32, 16);
+  EXPECT_NE(ptr, nullptr);
+  EXPECT_EQ(__builtin_is_aligned(ptr, 32), 0U);
+
+  LIBC_NAMESPACE::free(ptr);
+
+  // aligned_alloc fails if alignment is not power of two
+  void *bad_align = LIBC_NAMESPACE::aligned_alloc(30, 99);
+  EXPECT_EQ(bad_align, nullptr);
+
+  // aligned_alloc with a divergent size.
+  size_t alignment = 1 << (__gpu_lane_id() % 8 + 1);
+  void *div =
+      LIBC_NAMESPACE::aligned_alloc(alignment, (gpu::get_thread_id() + 1) * 4);
+  EXPECT_NE(div, nullptr);
+  EXPECT_EQ(__builtin_is_aligned(div, alignment), 0U);
+
+  return 0;
+}