summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--backend/src/libocl/include/ocl_enqueue.h35
-rw-r--r--backend/src/libocl/include/ocl_image.h2
-rw-r--r--backend/src/libocl/src/ocl_enqueue.cl54
3 files changed, 59 insertions, 32 deletions
diff --git a/backend/src/libocl/include/ocl_enqueue.h b/backend/src/libocl/include/ocl_enqueue.h
index 7ccab59f..8e9fd32b 100644
--- a/backend/src/libocl/include/ocl_enqueue.h
+++ b/backend/src/libocl/include/ocl_enqueue.h
@@ -43,14 +43,30 @@ struct Block_literal {
unsigned long int reserved; // NULL
unsigned long int size; // sizeof(struct Block_literal_1)
// optional helper functions
- void (*copy_helper)(void *dst, void *src); // IFF (1<<25)
- void (*dispose_helper)(void *src); // IFF (1<<25)
+ void *copy_helper; // IFF (1<<25)
+ void *dispose_helper; // IFF (1<<25)
// required ABI.2010.3.16
const char *signature; // IFF (1<<30)
} *descriptor;
// imported variables
};
+#if __clang_major__*10 + __clang_minor__ >= 50
+typedef struct ndrange_info_t ndrange_t;
+#endif
+
+#if __clang_major__*10 + __clang_minor__ >= 50
+#define BLOCK_TYPE void*
+#else
+#define BLOCK_TYPE __private void*
+#endif
+
+#if __clang_major__*10 + __clang_minor__ >= 40
+#define EVENT_TYPE clk_event_t*
+#else
+#define EVENT_TYPE __private clk_event_t*
+#endif
+
clk_event_t create_user_event(void);
void retain_event(clk_event_t event);
void release_event(clk_event_t event);
@@ -58,17 +74,16 @@ void set_user_event_status(clk_event_t event, int status);
bool is_valid_event(clk_event_t event);
void capture_event_profiling_info(clk_event_t event, int name, global void *value);
-uint __get_kernel_work_group_size_impl(__private void *block);
-uint __get_kernel_preferred_work_group_multiple_impl(__private void *block);
+uint __get_kernel_work_group_size_impl(BLOCK_TYPE block);
+uint __get_kernel_preferred_work_group_multiple_impl(BLOCK_TYPE block);
-OVERLOADABLE int enqueue_kernel(queue_t q, int flag, ndrange_t ndrange, void (^block)(void));
-OVERLOADABLE int enqueue_kernel(queue_t q, int flag, ndrange_t ndrange,
- uint num_events_in_wait_list, const clk_event_t *event_wait_list,
- clk_event_t *event_ret, void (^block)(void));
+int __enqueue_kernel_basic(queue_t q, int flag, ndrange_t ndrange, BLOCK_TYPE block);
+int __enqueue_kernel_basic_events(queue_t q, int flag, ndrange_t ndrange,
+ uint num_events_in_wait_list, const EVENT_TYPE event_wait_list,
+ EVENT_TYPE event_ret, BLOCK_TYPE block);
queue_t get_default_queue(void);
-int __gen_enqueue_kernel(queue_t q, int flag, ndrange_t ndrange, void (^block)(void), int size);
-int __gen_enqueue_kernel_slm(queue_t q, int flag, ndrange_t ndrange, __private void * block, int count, __private int* slm_sizes);
+int __gen_enqueue_kernel_slm(queue_t q, int flag, ndrange_t ndrange, BLOCK_TYPE block, int count, __private int* slm_sizes);
OVERLOADABLE ndrange_t ndrange_1D(size_t global_work_size);
OVERLOADABLE ndrange_t ndrange_1D(size_t global_work_size, size_t local_work_size);
diff --git a/backend/src/libocl/include/ocl_image.h b/backend/src/libocl/include/ocl_image.h
index 5a679aa8..f816e1af 100644
--- a/backend/src/libocl/include/ocl_image.h
+++ b/backend/src/libocl/include/ocl_image.h
@@ -23,6 +23,8 @@
#define int1 int
#define float1 float
+#pragma OPENCL EXTENSION cl_khr_3d_image_writes : enable
+
#define DECL_IMAGE_READ_SAMPLE_RETTYPE(IMG_TYPE, DATA_YPE, SUFFIX, N) \
OVERLOADABLE DATA_YPE read_image ## SUFFIX(IMG_TYPE cl_image, const sampler_t sampler, int##N coord); \
OVERLOADABLE DATA_YPE read_image ## SUFFIX(IMG_TYPE cl_image, const sampler_t sampler, float##N coord);
diff --git a/backend/src/libocl/src/ocl_enqueue.cl b/backend/src/libocl/src/ocl_enqueue.cl
index dc8fa3b1..1ae43aa4 100644
--- a/backend/src/libocl/src/ocl_enqueue.cl
+++ b/backend/src/libocl/src/ocl_enqueue.cl
@@ -30,7 +30,7 @@ ndrange_t __gen_ocl_set_ndrange_info(__private struct ndrange_info_t *info);
__private struct ndrange_info_t* __gen_ocl_get_ndrange_info(ndrange_t info);
__global int* __gen_ocl_get_enqueue_info_addr(void);
-OVERLOADABLE int enqueue_kernel(queue_t q, int flag, ndrange_t ndrange, void (^block)(void))
+int __enqueue_kernel_basic(queue_t q, int flag, ndrange_t ndrange, BLOCK_TYPE block)
{
int i;
__private struct Block_literal *literal = (__private struct Block_literal *)block;
@@ -40,8 +40,11 @@ OVERLOADABLE int enqueue_kernel(queue_t q, int flag, ndrange_t ndrange, void (^b
__global int* start_addr = __gen_ocl_get_enqueue_info_addr();
int offset = atomic_add(start_addr, size + sizeof(struct ndrange_info_t));
__global uchar* addr = (__global uchar*)start_addr + offset + sizeof(int);
+#if __clang_major__*10 + __clang_minor__ >= 50
+ __private struct ndrange_info_t *info = to_private(&ndrange);
+#else
__private struct ndrange_info_t *info = __gen_ocl_get_ndrange_info(ndrange);
-
+#endif
*((__global struct ndrange_info_t *)addr) = *info;
addr += sizeof(*info);
@@ -51,14 +54,14 @@ OVERLOADABLE int enqueue_kernel(queue_t q, int flag, ndrange_t ndrange, void (^b
return 0;
}
-OVERLOADABLE int enqueue_kernel(queue_t q, int flag, ndrange_t ndrange,
- uint num_events_in_wait_list, const clk_event_t *event_wait_list,
- clk_event_t *event_ret, void (^block)(void))
+int __enqueue_kernel_basic_events(queue_t q, int flag, ndrange_t ndrange,
+ uint num_events_in_wait_list, const EVENT_TYPE event_wait_list,
+ EVENT_TYPE event_ret, BLOCK_TYPE block)
{
- return enqueue_kernel(q, flag, ndrange, block);
+ return __enqueue_kernel_basic(q, flag, ndrange, block);
}
-int __gen_enqueue_kernel_slm(queue_t q, int flag, ndrange_t ndrange, __private void * block, int count, __private int* slm_sizes)
+int __gen_enqueue_kernel_slm(queue_t q, int flag, ndrange_t ndrange, BLOCK_TYPE block, int count, __private int* slm_sizes)
{
int i;
__private struct Block_literal* literal = (__private struct Block_literal *)block;
@@ -69,7 +72,11 @@ int __gen_enqueue_kernel_slm(queue_t q, int flag, ndrange_t ndrange, __private v
__global int* start_addr = __gen_ocl_get_enqueue_info_addr();
int offset = atomic_add(start_addr, size + sizeof(struct ndrange_info_t) + slm_size);
__global uchar* addr = (__global uchar*)start_addr + offset + sizeof(int);
+#if __clang_major__*10 + __clang_minor__ >= 50
+ __private struct ndrange_info_t *info = to_private(&ndrange);
+#else
__private struct ndrange_info_t *info = __gen_ocl_get_ndrange_info(ndrange);
+#endif
*((__global struct ndrange_info_t *)addr) = *info;
addr += sizeof(*info);
@@ -111,12 +118,12 @@ bool is_valid_event(clk_event_t event)
return 1;
}
-uint __get_kernel_work_group_size_impl(__private void *block)
+uint __get_kernel_work_group_size_impl(BLOCK_TYPE block)
{
return 256;
}
-uint __get_kernel_preferred_work_group_multiple_impl(__private void *block)
+uint __get_kernel_preferred_work_group_multiple_impl(BLOCK_TYPE block)
{
return 16;
}
@@ -127,13 +134,19 @@ void capture_event_profiling_info(clk_event_t event, int name, global void *valu
((__global ulong *)value)[0] = 0x3000;
((__global ulong *)value)[1] = 0x6000;
}
+
+#if __clang_major__*10 + __clang_minor__ >= 50
+#define RET_INFO return info;
+#else
+#define RET_INFO return __gen_ocl_set_ndrange_info(&info);
+#endif
+
OVERLOADABLE ndrange_t ndrange_1D(size_t global_work_size)
{
struct ndrange_info_t info;
info.type = 0x1;
info.global_work_size[0] = global_work_size;
- return __gen_ocl_set_ndrange_info(&info);
- //return ndrange;
+ RET_INFO;
}
OVERLOADABLE ndrange_t ndrange_1D(size_t global_work_size, size_t local_work_size)
@@ -142,8 +155,7 @@ OVERLOADABLE ndrange_t ndrange_1D(size_t global_work_size, size_t local_work_siz
info.type = 0x2;
info.global_work_size[0] = global_work_size;
info.local_work_size[0] = local_work_size;
- return __gen_ocl_set_ndrange_info(&info);
- // return ndrange;
+ RET_INFO;
}
@@ -154,8 +166,7 @@ OVERLOADABLE ndrange_t ndrange_1D(size_t global_work_offset, size_t global_work_
info.global_work_size[0] = global_work_size;
info.local_work_size[0] = local_work_size;
info.global_work_offset[0] = global_work_offset;
- return __gen_ocl_set_ndrange_info(&info);
- //return ndrange;
+ RET_INFO;
}
OVERLOADABLE ndrange_t ndrange_2D(const size_t global_work_size[2])
@@ -164,8 +175,7 @@ OVERLOADABLE ndrange_t ndrange_2D(const size_t global_work_size[2])
info.type = 0x11;
info.global_work_size[0] = global_work_size[0];
info.global_work_size[1] = global_work_size[1];
- return __gen_ocl_set_ndrange_info(&info);
- //return ndrange;
+ RET_INFO;
}
OVERLOADABLE ndrange_t ndrange_2D(const size_t global_work_size[2], const size_t local_work_size[2])
@@ -176,7 +186,7 @@ OVERLOADABLE ndrange_t ndrange_2D(const size_t global_work_size[2], const size_t
info.global_work_size[1] = global_work_size[1];
info.local_work_size[0] = local_work_size[0];
info.local_work_size[1] = local_work_size[1];
- return __gen_ocl_set_ndrange_info(&info);
+ RET_INFO;
}
@@ -190,7 +200,7 @@ OVERLOADABLE ndrange_t ndrange_2D(const size_t global_work_offset[2], const size
info.local_work_size[1] = local_work_size[1];
info.global_work_offset[0] = global_work_offset[0];
info.global_work_offset[1] = global_work_offset[1];
- return __gen_ocl_set_ndrange_info(&info);
+ RET_INFO;
}
OVERLOADABLE ndrange_t ndrange_3D(const size_t global_work_size[3])
@@ -200,7 +210,7 @@ OVERLOADABLE ndrange_t ndrange_3D(const size_t global_work_size[3])
info.global_work_size[0] = global_work_size[0];
info.global_work_size[1] = global_work_size[1];
info.global_work_size[2] = global_work_size[2];
- return __gen_ocl_set_ndrange_info(&info);
+ RET_INFO;
}
OVERLOADABLE ndrange_t ndrange_3D(const size_t global_work_size[3], const size_t local_work_size[3])
@@ -213,7 +223,7 @@ OVERLOADABLE ndrange_t ndrange_3D(const size_t global_work_size[3], const size_t
info.local_work_size[0] = local_work_size[0];
info.local_work_size[1] = local_work_size[1];
info.local_work_size[2] = local_work_size[2];
- return __gen_ocl_set_ndrange_info(&info);
+ RET_INFO;
}
OVERLOADABLE ndrange_t ndrange_3D(const size_t global_work_offset[3], const size_t global_work_size[3], const size_t local_work_size[3])
@@ -229,7 +239,7 @@ OVERLOADABLE ndrange_t ndrange_3D(const size_t global_work_offset[3], const size
info.global_work_offset[0] = global_work_offset[0];
info.global_work_offset[1] = global_work_offset[1];
info.global_work_offset[2] = global_work_offset[2];
- return __gen_ocl_set_ndrange_info(&info);
+ RET_INFO;
}
int enqueue_marker (queue_t queue, uint num_events_in_wait_list, const clk_event_t *event_wait_list, clk_event_t *event_ret)
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-21 14:27:15 +0000