feat: add profiling macros for performance measurement in operations

Author: chraac

ggml/src/ggml-hexagon/htp/ops-utils.h

@@ -146,4 +146,11 @@ static inline void htp_dump_f16(char * pref, const __fp16 * x, uint32_t n) {
     }
 }
 
+#define PROFILER_START(name) const uint64_t name##_start_cycles = HAP_perf_get_qtimer_count()
+#define PROFILER_END(name, ...)                                                                                   \
+    do {                                                                                                          \
+        const uint64_t name##_end_cycles = HAP_perf_get_qtimer_count();                                           \
+        FARF(HIGH, __VA_ARGS__, (unsigned) HAP_perf_qtimer_count_to_us(name##_end_cycles - name##_start_cycles)); \
+    } while (0)
+
 #endif /* OPS_UTILS_H */

ggml/src/ggml-hexagon/htp/rope-ops.c

@@ -365,8 +365,7 @@ static void rope_job_f32_per_thread(struct rope_th_ctx * rope_ctx, int nth, int
         return;
     }
 
-    uint64_t t1, t2;
-    t1 = HAP_perf_get_qtimer_count();
+    PROFILER_START(rope_job_f32);
 
     int is_aligned = 1;
     int opt_path   = 0;
@@ -381,10 +380,8 @@ static void rope_job_f32_per_thread(struct rope_th_ctx * rope_ctx, int nth, int
 
     rope_hex_f32(rope_ctx, src0_start_row, src0_end_row, nth, ith, opt_path);
 
-    t2 = HAP_perf_get_qtimer_count();
-
-    FARF(HIGH, "rope-f32: %d/%d/%d: (%u:%u) usec %u\n", ith, nth, opt_path, src0_start_row, src0_end_row,
-         (unsigned) HAP_perf_qtimer_count_to_us(t2 - t1));
+    PROFILER_END(rope_job_f32, "rope-f32: %d/%d/%d: (%u:%u) usec %u\n", ith, nth, opt_path, src0_start_row,
+                 src0_end_row);
 }
 
 static void rope_job_dispatcher_f32(unsigned int n, unsigned int i, void * data) {