You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Here's the C1XX compiler memory consumption of all tests named *ranges*, as reported by VS 2022 17.6 Preview 1 x64 /d1reportMemorySummary. There are 4 tests that exceed 2 GB of RAM, which we should investigate reducing (either by pruning their tested type combinations, or by splitting the test up into multiple parts as we've done elsewhere).
This was brought to my attention by Russell Johnston in MS; while working with EDG to investigate and fix issues for the STL (:heart_eyes_cat:), the top 3 tests here were failing due to out-of-memory errors (and presumably the others in the ~1.8 to 2.2 GB range for C1XX were perilously close to OOM for EDG). This doesn't indicate anything pathological or undesirable about what EDG is doing; we should simply avoid writing such enormously expensive tests. This is problematic even for C1XX, since it means that x86-hosted test runs (internally) run close to OOM, and multithreaded test runs can come close to swapping (e.g. I have "only" 64 GB for 32 cores).
Here's the C1XX compiler memory consumption of all tests named
*ranges*
, as reported by VS 2022 17.6 Preview 1 x64/d1reportMemorySummary
. There are 4 tests that exceed 2 GB of RAM, which we should investigate reducing (either by pruning their tested type combinations, or by splitting the test up into multiple parts as we've done elsewhere).This was brought to my attention by Russell Johnston in MS; while working with EDG to investigate and fix issues for the STL (:heart_eyes_cat:), the top 3 tests here were failing due to out-of-memory errors (and presumably the others in the ~1.8 to 2.2 GB range for C1XX were perilously close to OOM for EDG). This doesn't indicate anything pathological or undesirable about what EDG is doing; we should simply avoid writing such enormously expensive tests. This is problematic even for C1XX, since it means that x86-hosted test runs (internally) run close to OOM, and multithreaded test runs can come close to swapping (e.g. I have "only" 64 GB for 32 cores).
Set Size (> 2 GB)
Set Size (< 2 GB)
Click to expand smaller (< 1 GB) tests:
Set Size (< 1 GB)
Click to expand repro incantations:
The text was updated successfully, but these errors were encountered: