aboutsummaryrefslogtreecommitdiff
path: root/llvm/docs/CommandGuide/llvm-profdata.rst
diff options
context:
space:
mode:
authorWei Mi <wmi@google.com>2020-07-08 15:19:44 -0700
committerWei Mi <wmi@google.com>2020-07-27 20:17:40 -0700
commita23f62343cb79a3306fa64545db1d61c2d76b9ca (patch)
tree8f6dcf7a8c865338265e1cd5a9fb4379210aed17 /llvm/docs/CommandGuide/llvm-profdata.rst
parent23d6525cbdc9de7cbfe7640d1e9e4f25a0c5dd85 (diff)
downloadllvm-a23f62343cb79a3306fa64545db1d61c2d76b9ca.zip
llvm-a23f62343cb79a3306fa64545db1d61c2d76b9ca.tar.gz
llvm-a23f62343cb79a3306fa64545db1d61c2d76b9ca.tar.bz2
Supplement instr profile with sample profile.
PGO profile is usually more precise than sample profile. However, PGO profile needs to be collected from loadtest and loadtest may not be representative enough to the production workload. Sample profile collected from production can be used as a supplement -- for functions cold in loadtest but warm/hot in production, we can scale up the related function in PGO profile if the function is warm or hot in sample profile. The implementation contains changes in compiler side and llvm-profdata side. Given an instr profile and a sample profile, for a function cold in PGO profile but warm/hot in sample profile, llvm-profdata will either mark all the counters in the profile to be -1 or scale up the max count in the function to be above hot threshold, depending on the zero counter ratio in the profile. The assumption is if there are too many counters being zero in the function profile, the profile is more likely to cause harm than good, then llvm-profdata will mark all the counters to be -1 indicating the function is hot but the profile is unaccountable. In compiler side, if a function profile with all -1 counters is seen, the function entry count will be set to be above hot threshold but its internal profile will be dropped. In the long run, it may be useful to let compiler support using PGO profile and sample profile at the same time, but that requires more careful design and more substantial changes to make two profiles work seamlessly. The patch here serves as a simple intermediate solution. Differential Revision: https://reviews.llvm.org/D81981
Diffstat (limited to 'llvm/docs/CommandGuide/llvm-profdata.rst')
-rw-r--r--llvm/docs/CommandGuide/llvm-profdata.rst24
1 files changed, 24 insertions, 0 deletions
diff --git a/llvm/docs/CommandGuide/llvm-profdata.rst b/llvm/docs/CommandGuide/llvm-profdata.rst
index 13a66dc..6472320 100644
--- a/llvm/docs/CommandGuide/llvm-profdata.rst
+++ b/llvm/docs/CommandGuide/llvm-profdata.rst
@@ -161,6 +161,30 @@ OPTIONS
coverage for the optimized target. This option can only be used with
sample-based profile in extbinary format.
+.. option:: -supplement-instr-with-sample=path_to_sample_profile
+
+ Supplement an instrumentation profile with sample profile. The sample profile
+ is the input of the flag. Output will be in instrumentation format (only works
+ with -instr).
+
+.. option:: -zero-counter-threshold=threshold_float_number
+
+ For the function which is cold in instr profile but hot in sample profile, if
+ the ratio of the number of zero counters divided by the the total number of
+ counters is above the threshold, the profile of the function will be regarded
+ as being harmful for performance and will be dropped.
+
+.. option:: -instr-prof-cold-threshold=threshold_int_number
+
+ User specified cold threshold for instr profile which will override the cold
+ threshold got from profile summary.
+
+.. option:: -suppl-min-size-threshold=threshold_int_number
+
+ If the size of a function is smaller than the threshold, assume it can be
+ inlined by PGO early inliner and it will not be adjusted based on sample
+ profile.
+
EXAMPLES
^^^^^^^^
Basic Usage