Remove profiler time hooks deps

Summary: This diff gets rid of the need for separate time hooks in the profiler code base. We switch over the profiler to use the time hook available in PAL. For Jarvis added an implementation for `et_pal_current_ticks` that is only brought in when compiling for Xtensa. If compiling for devserver in Jarvis, it'll fall back to the default POSIX implementation.

Reviewed By: larryliu0820, cccclai

Differential Revision: D47637865

fbshipit-source-id: 86c61c7f00bfa9be18c8257c115ecb1d0bb754a3

Author: tarun292

profiler/TARGETS

@@ -3,9 +3,6 @@
 
 load("@fbcode_macros//build_defs:python_binary.bzl", "python_binary")
 load("@fbcode_macros//build_defs:python_library.bzl", "python_library")
-load(":targets.bzl", "define_common_targets")
-
-define_common_targets()
 
 python_library(
     name = "parse_profiler_library",

profiler/profiler.bzl

@@ -26,6 +26,5 @@ def define_common_targets():
                 "//executorch/runtime/core:core",
                 "//executorch/runtime/core:evalue" + aten_suffix,
                 "//executorch/runtime/core:memory_allocator",
-                "//executorch/profiler:profiler",
             ],
         )

profiler/targets.bzl

@@ -22,9 +22,6 @@ def define_common_targets():
             "//executorch/...",
             "@EXECUTORCH_CLIENTS",
         ],
-        deps = [
-            "//executorch/profiler:profiler",
-        ],
         exported_deps = [
             "//executorch/runtime/platform:platform",
         ],
@@ -47,9 +44,6 @@ def define_common_targets():
             "hierarchical_allocator.h",
             "memory_allocator.h",
         ],
-        deps = [
-            "//executorch/profiler:profiler",
-        ],
         exported_deps = [
             ":core",
         ],

runtime/backend/targets.bzl

@@ -50,7 +50,6 @@ def define_common_targets():
             "//executorch/runtime/platform:platform",
             "//executorch/schema:extended_header",
             "//executorch/schema:schema",
-            "//executorch/profiler:profiler",
         ],
         preprocessor_flags = _program_preprocessor_flags(),
         exported_deps = ["//executorch/runtime/core:core"],
@@ -71,7 +70,6 @@ def define_common_targets():
                 "//executorch/runtime/backend:backend_registry",
                 "//executorch/kernels/prim_ops:prim_ops_registry" + aten_suffix,
                 "//executorch/runtime/kernel:kernel_runtime_context" + aten_suffix,
-                "//executorch/profiler:profiler",
                 "//executorch/schema:schema",
             ],
             exported_deps = [

runtime/core/targets.bzl

@@ -2,6 +2,7 @@
 
 #include <executorch/runtime/platform/assert.h>
 #include <executorch/runtime/platform/hooks.h>
+#include <executorch/runtime/platform/platform.h>
 #include <executorch/runtime/platform/profiler.h>
 #include <inttypes.h>
 
@@ -59,13 +60,13 @@ uint32_t begin_profiling(const char* name) {
   prof_arr[curr_counter].instruction_idx = state.instruction_idx;
   // Set start time at the last to ensure that we're not capturing
   // any of the overhead in this function.
-  prof_arr[curr_counter].start_time = get_curr_time();
+  prof_arr[curr_counter].start_time = et_pal_current_ticks();
   return curr_counter;
 }
 
 void end_profiling(uint32_t token_id) {
   ET_CHECK_MSG(token_id < MAX_PROFILE_EVENTS, "Invalid token id.");
-  prof_arr[token_id].end_time = get_curr_time();
+  prof_arr[token_id].end_time = et_pal_current_ticks();
 }
 
 void dump_profile_stats(prof_result_t* prof_result) {

runtime/executor/targets.bzl

@@ -10,6 +10,25 @@ def _select_pal(dict_):
         fail("Missing key for executorch.pal_default value '{}' in dict '{}'".format(pal_default, dict_))
     return dict_[pal_default]
 
+def profiling_enabled():
+    return native.read_config("executorch", "prof_enabled", "false") == "true"
+
+def get_profiling_flags():
+    profiling_flags = []
+    if profiling_enabled():
+        profiling_flags += ["-DPROFILING_ENABLED"]
+    prof_buf_size = native.read_config("executorch", "prof_buf_size", None)
+    if prof_buf_size != None:
+        if not profiling_enabled():
+            fail("Cannot set profiling buffer size without enabling profiling first.")
+        profiling_flags += ["-DMAX_PROFILE_EVENTS={}".format(prof_buf_size), "-DMAX_MEM_PROFILE_EVENTS={}".format(prof_buf_size)]
+    num_prof_blocks = native.read_config("executorch", "num_prof_blocks", None)
+    if num_prof_blocks != None:
+        if not profiling_enabled():
+            fail("Cannot configure number of profiling blocks without enabling profiling first.")
+        profiling_flags += ["-DMAX_PROFILE_BLOCKS={}".format(num_prof_blocks)]
+    return profiling_flags
+
 def define_common_targets():
     """Defines targets that should be shared between fbcode and xplat.
 
@@ -49,11 +68,11 @@ def define_common_targets():
             "profiler.cpp",
             "runtime.cpp",
         ],
+        exported_preprocessor_flags = get_profiling_flags(),
         exported_deps = [
             "//executorch/runtime/platform:pal_interface",
             ":compiler",
             ":platform_private",
-            "//executorch/profiler:profiler",
         ],
         visibility = [
             "//executorch/...",

runtime/platform/profiler.cpp

@@ -14,7 +14,6 @@ def define_common_targets():
         ],
         deps = [
             "//executorch/runtime/core:memory_allocator",
-            "//executorch/profiler:profiler",
             "//executorch/sdk/etdump:etdump_gen",
         ],
         preprocessor_flags = ["-DPROFILING_ENABLED"],

runtime/platform/targets.bzl

@@ -366,7 +366,6 @@ def executorch_generated_lib(
                 "//executorch/runtime/kernel:operator_registry",
                 "//executorch/kernels/prim_ops:prim_ops_registry" + aten_suffix,
                 "//executorch/runtime/core:evalue" + aten_suffix,
-                "//executorch/profiler:profiler",
                 "//executorch/codegen:macros",
             ] + deps,
             exported_deps = [