the known issue is that utest will output lots of warning messages
since the total number of work-items is less than 64.
V2: use DEBUGP to wrap the code, and also refine the macro
Signed-off-by: Guo Yejun <yejun....@intel.com>
---
src/cl_api.c | 11 ++++++++++-
src/cl_command_queue_gen7.c | 6 +++---
src/cl_utils.h | 12 ++++++++++--
3 files changed, 23 insertions(+), 6 deletions(-)
diff --git a/src/cl_api.c b/src/cl_api.c
index 510941e..05ca889 100644
--- a/src/cl_api.c
+++ b/src/cl_api.c
@@ -2999,6 +2999,7 @@ clEnqueueNDRangeKernel(cl_command_queue command_queue,
fixed_local_sz[1] = 1;
} else {
uint j, maxDimSize = 64 /* from 64? */, maxGroupSize = 256;
//MAX_WORK_GROUP_SIZE may too large
+ size_t realGroupSize = 1;
for (i = 0; i< work_dim; i++) {
for (j = maxDimSize; j > 1; j--) {
if (global_work_size[i] % j == 0 && j <= maxGroupSize) {
@@ -3008,7 +3009,10 @@ clEnqueueNDRangeKernel(cl_command_queue command_queue,
break; //choose next work_dim
}
}
+ realGroupSize *= fixed_local_sz[i];
}
+ if (realGroupSize < 8 || realGroupSize % 8 != 0)
+ DEBUGP(DL_WARNING, "unable to find good values for local_work_size[i],
please provide local_work_size[] explicitly, you can find good values with
trial-and-error method.");
}
}
@@ -3016,8 +3020,13 @@ clEnqueueNDRangeKernel(cl_command_queue command_queue,
fixed_global_sz[0] = (global_work_size[0]+15) / 16 * 16;
fixed_global_sz[1] = (global_work_size[1]+15) / 16;
} else {
- for (i = 0; i < work_dim; ++i)
+ size_t total_work_items = 1;
+ for (i = 0; i < work_dim; ++i) {
fixed_global_sz[i] = global_work_size[i];
+ total_work_items *= fixed_global_sz[i];
+ }
+ if (total_work_items < 64)
+ DEBUGP(DL_WARNING, "too small work-items (see values in
global_work_size[]) might result in bad performance.");
}
if (global_work_offset != NULL)
for (i = 0; i < work_dim; ++i)
diff --git a/src/cl_command_queue_gen7.c b/src/cl_command_queue_gen7.c
index 44db7ed..791a7ca 100644
--- a/src/cl_command_queue_gen7.c
+++ b/src/cl_command_queue_gen7.c
@@ -369,21 +369,21 @@ cl_command_queue_ND_range_gen7(cl_command_queue queue,
/* Compute the number of HW threads we need */
if(UNLIKELY(err = cl_kernel_work_group_sz(ker, local_wk_sz, 3, &local_sz) !=
CL_SUCCESS)) {
- DEBUGP("Beignet: Work group size exceed Kernel's work group size.\n");
+ DEBUGP(DL_ERROR, "Work group size exceed Kernel's work group size.");
return err;
}
kernel.thread_n = thread_n = (local_sz + simd_sz - 1) / simd_sz;
kernel.curbe_sz = cst_sz;
if (scratch_sz > ker->program->ctx->device->scratch_mem_size) {
- DEBUGP("Beignet: Out of scratch memory %d.\n", scratch_sz);
+ DEBUGP(DL_ERROR, "Out of scratch memory %d.", scratch_sz);
return CL_OUT_OF_RESOURCES;
}
/* Curbe step 1: fill the constant urb buffer data shared by all threads */
if (ker->curbe) {
kernel.slm_sz = cl_curbe_fill(ker, work_dim, global_wk_off, global_wk_sz,
local_wk_sz, thread_n);
if (kernel.slm_sz > ker->program->ctx->device->local_mem_size) {
- DEBUGP("Beignet: Out of shared local memory %d.\n", kernel.slm_sz);
+ DEBUGP(DL_ERROR, "Out of shared local memory %d.", kernel.slm_sz);
return CL_OUT_OF_RESOURCES;
}
}
diff --git a/src/cl_utils.h b/src/cl_utils.h
index 7595158..a0e37b2 100644
--- a/src/cl_utils.h
+++ b/src/cl_utils.h
@@ -31,11 +31,19 @@
#define JOIN(X, Y) _DO_JOIN(X, Y)
#define _DO_JOIN(X, Y) _DO_JOIN2(X, Y)
#define _DO_JOIN2(X, Y) X##Y
+enum DEBUGP_LEVEL
+{
+ DL_INFO,
+ DL_WARNING,
+ DL_ERROR
+};
#ifdef NDEBUG
#define DEBUGP(...)
#else
- #define DEBUGP(fmt, ...) \
- fprintf(stderr, fmt, ##__VA_ARGS__)
+ //TODO: print or not with the value of level (DEBUGP_LEVEL)
+ #define DEBUGP(level, fmt, ...) \
+ fprintf(stderr, "Beignet: "#fmt, ##__VA_ARGS__); \
+ fprintf(stderr, "\n");
#endif
/* Check compile time errors */
--
1.9.1
_______________________________________________
Beignet mailing list
Beignet@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/beignet
