Ensure kernel arguments to local address space allocated on 128 byte boundary

Buffers in local space, passed as kernel parameters, are allocated at runtime
by shamrock.

This patch ensures such allocated memory is aligned for maximum OpenCL data
type alignment (double16).

Signed-off-by: Gil Pitney <gil.pitney@linaro.org>

1 files changed, 5 insertions, 3 deletions

diff --git a/src/core/cpu/kernel.cpp b/src/core/cpu/kernel.cpp
index 49e4dcc..b89aafa 100644
--- a/src/core/cpu/kernel.cpp
+++ b/src/core/cpu/kernel.cpp
@@ -588,17 +588,19 @@ void *CPUKernelWorkGroup::callArgs(std::vector<void *> &locals_to_free)
         {
             case Kernel::Arg::Buffer:
             {
-                MemObject *buffer = *(MemObject **)arg->data();
-
                 if (arg->file() == Kernel::Arg::Local)
                 {
                     // Alloc a buffer and pass it to the kernel
-                    void *local_buffer = std::malloc(arg->allocAtKernelRuntime());
+                    // align for type double16 size.
+		    void *local_buffer = NULL;
+                    int retval = posix_memalign(&local_buffer, 128, arg->allocAtKernelRuntime());
                     locals_to_free.push_back(local_buffer);
                     *(void **)target = local_buffer;
                 }
                 else
                 {
+                    MemObject *buffer = *(MemObject **)arg->data();
+
                     if (!buffer)
                     {
                         // We can do that, just send NULL