Remove dangling TENSORFLOW_C_LIB_PATH references in benchmarking code
This patch removes a couple dangling TENSORFLOW_C_LIB_PATH references
that I failed to catch in #150. Fixes #250, as I thought I hadn't
updated anything yet, but I was just seeing some dangling references.
diff --git a/compiler_opt/benchmark/benchmarking_utils.py b/compiler_opt/benchmark/benchmarking_utils.py
index 79bd55b..a212c3f 100644
--- a/compiler_opt/benchmark/benchmarking_utils.py
+++ b/compiler_opt/benchmark/benchmarking_utils.py
@@ -37,11 +37,10 @@
use_existing_build: Whether or not to do an incremental build
llvm_build_path: The path to where the LLVM build will go
llvm_source_path: The path to the root of the llvm-project repository
- tensorflow_c_lib_path: The path to the tensorflow c lib path
- Note: llvm_source_path and tensorflow_c_lib_path aren't necessary if you have
- set use_existing_build to true, you just need to make sure that the existing
- build is already set up to enable the necessary MLGO flags.
+ Note: llvm_source_path is not necessary if you have set use_existing_build to
+ true, you just need to make sure that the existing build is already set up to
+ enable the necessary MLGO flags.
"""
if not use_existing_build and os.path.exists(llvm_build_path):
shutil.rmtree(llvm_build_path)
diff --git a/docs/benchmarking.md b/docs/benchmarking.md
index eb8a9ac..966f006 100644
--- a/docs/benchmarking.md
+++ b/docs/benchmarking.md
@@ -53,8 +53,7 @@
--output_path=./results.json \
--perf_counter=INSTRUCTIONS \
--perf_counter=MEM_UOPS_RETIRED:ALL_LOADS \
- --perf_counter=MEM_UOPS_RETIRED:ALL_STORES \
- --tensorflow_c_lib_path=~/tensorflow
+ --perf_counter=MEM_UOPS_RETIRED:ALL_STORES
```
This will output a bunch of test information to `./results.json` that can then
@@ -90,8 +89,6 @@
performance counters in the libpfm format. There can only be up to three
performance counters specified due to underlying limitations in Google
benchmark.
-* `tensorflow_c_lib_path` - The path to the tensorflow c library if you aren't
-doing an incremental LLVM build.
* `tests_to_run` - This specifies the LLVM microbenchmarks to run relative to
the microbenchmarks library in the LLVM test suite build directory. The default
values for this flag should be pretty safe and produce good results.
@@ -122,8 +119,7 @@
--num_threads=32 \
--output_file=./output-chromium-testing.json \
--perf_counters=mem_uops_retired.all_loads \
- --perf_counters=mem_uops_retired.all_stores \
- --tensorflow_c_lib_path=~/tensorflow
+ --perf_counters=mem_uops_retired.all_stores
```
Several of the flags here are extremely similar to/the same as the flags
