prefetch: Implemented prefetch builtin (noop'ed)

Also, updated CL_DEVICE_LOCAL_MEM_SIZE and Buffer::Buffer() constructor
so that Khrnos basic "prefetch" test can pass.

Previously, it was trying to allocate huge global buffers (computed based
on the LOCAL_MEM_SIZE) and was failing.

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

3 files changed, 25 insertions, 22 deletions

diff --git a/include/cpu.h b/include/cpu.h
index 6fa0d90..c7c5976 100644
--- a/include/cpu.h
+++ b/include/cpu.h
@@ -31,23 +31,27 @@
 #include "clc.h"
 
 #define PREFETCH_VECTORIZE(PRIM_TYPE) \
-   _CLC_OVERLOAD _CLC_DECL void prefetch(const __global PRIM_TYPE    *p, size_t num_gentypes); \
-   _CLC_OVERLOAD _CLC_DECL void prefetch(const __global PRIM_TYPE##2 *p, size_t num_gentypes); \
-   _CLC_OVERLOAD _CLC_DECL void prefetch(const __global PRIM_TYPE##3 *p, size_t num_gentypes); \
-   _CLC_OVERLOAD _CLC_DECL void prefetch(const __global PRIM_TYPE##4 *p, size_t num_gentypes); \
-   _CLC_OVERLOAD _CLC_DECL void prefetch(const __global PRIM_TYPE##8 *p, size_t num_gentypes); \
-   _CLC_OVERLOAD _CLC_DECL void prefetch(const __global PRIM_TYPE##16 *p, size_t num_gentypes); \
-
-PREFETCH_VECTORIZE(char)
-PREFETCH_VECTORIZE(uchar)
-PREFETCH_VECTORIZE(short)
-PREFETCH_VECTORIZE(ushort)
-PREFETCH_VECTORIZE(int)
-PREFETCH_VECTORIZE(uint)
-PREFETCH_VECTORIZE(long)
-PREFETCH_VECTORIZE(ulong)
-PREFETCH_VECTORIZE(float)
-PREFETCH_VECTORIZE(double)
+  _CLC_OVERLOAD _CLC_INLINE void prefetch(const __global PRIM_TYPE    *p, size_t num_gentypes){} \
+   _CLC_OVERLOAD _CLC_INLINE void prefetch(const __global PRIM_TYPE##2 *p, size_t num_gentypes){} \
+   _CLC_OVERLOAD _CLC_INLINE void prefetch(const __global PRIM_TYPE##3 *p, size_t num_gentypes){} \
+   _CLC_OVERLOAD _CLC_INLINE void prefetch(const __global PRIM_TYPE##4 *p, size_t num_gentypes){} \
+   _CLC_OVERLOAD _CLC_INLINE void prefetch(const __global PRIM_TYPE##8 *p, size_t num_gentypes){} \
+   _CLC_OVERLOAD _CLC_INLINE void prefetch(const __global PRIM_TYPE##16 *p, size_t num_gentypes){} \
+
+
+#define PREFETCH_TYPES()           \
+        PREFETCH_VECTORIZE(char)   \
+        PREFETCH_VECTORIZE(uchar)  \
+        PREFETCH_VECTORIZE(short)  \
+        PREFETCH_VECTORIZE(ushort) \
+        PREFETCH_VECTORIZE(int)    \
+        PREFETCH_VECTORIZE(uint)   \
+        PREFETCH_VECTORIZE(long)   \
+        PREFETCH_VECTORIZE(ulong)  \
+        PREFETCH_VECTORIZE(float)  \
+        PREFETCH_VECTORIZE(double) \
+
+PREFETCH_TYPES()
 
 /*-----------------------------------------------------------------------------
 * This can be empty since our copy routines are currently synchronous. When 
diff --git a/src/core/cpu/device.cpp b/src/core/cpu/device.cpp
index b0732ac..2c85ea8 100644
--- a/src/core/cpu/device.cpp
+++ b/src/core/cpu/device.cpp
@@ -509,8 +509,11 @@ cl_int CPUDevice::info(cl_device_info param_name,
                                                "MemTotal:", 512*1024) * 1024);
             break;
 
-        case CL_DEVICE_MAX_MEM_ALLOC_SIZE:
         case CL_DEVICE_LOCAL_MEM_SIZE:
+            SIMPLE_ASSIGN(cl_ulong, 128 * 1024);
+	    break;
+
+        case CL_DEVICE_MAX_MEM_ALLOC_SIZE:
         case CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE:
             // TODO: 1 Gio seems to be enough for software acceleration
 
diff --git a/src/core/memobject.cpp b/src/core/memobject.cpp
index 5501ac1..3912740 100644
--- a/src/core/memobject.cpp
+++ b/src/core/memobject.cpp
@@ -408,11 +408,7 @@ Buffer::Buffer(Context *ctx, size_t size, void *host_ptr, cl_mem_flags flags,
         return;
     }
 
-#if defined(__arm__)
-    if (size > 512*1024*1024)
-#else
     if (size > 1*1024*1024*1024)
-#endif
     {
          *errcode_ret = CL_INVALID_BUFFER_SIZE;
          return;