[InstCombine] Make test resilient to metadata presence

[mtrofin]

Modernized it to using update_test_checks​ which addresses an ambgiuty in the previous test formulation, where a profile metadaat of value i32 1​ would have (incorrectly matched.

[mtrofin]

• [InstCombine] Make test resilient to metadata presence #157607 👈 (View in Graphite)
• [profcheck] Require unknown metadata have an origin parameter #157594 : 1 other dependent PR (#157599 )
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: Mircea Trofin (mtrofin)

Changes

Adding !prof​ annotations to Transforms/InstCombine/2004-09-20-BadLoadCombine2.ll​, and removed llvm-dis​ in favor of using -S​

The prof annotations don’t change the intent of the test, but make it resilient to the profcheck​ bot that would have otherwise introduced a branch_weights profile of i32 1, i32 1​, which would have matched the grep.

Added a profile presence test, too.

---

Full diff: https://github.com/llvm/llvm-project/pull/157607.diff

2 Files Affected:

• (modified) llvm/test/Transforms/InstCombine/2004-09-20-BadLoadCombine2.ll (+8-4)
• (modified) llvm/utils/profcheck-xfail.txt (-1)

diff --git a/llvm/test/Transforms/InstCombine/2004-09-20-BadLoadCombine2.ll b/llvm/test/Transforms/InstCombine/2004-09-20-BadLoadCombine2.ll
index f558e35ebe015..0b4c852664c2f 100644
--- a/llvm/test/Transforms/InstCombine/2004-09-20-BadLoadCombine2.ll
+++ b/llvm/test/Transforms/InstCombine/2004-09-20-BadLoadCombine2.ll
@@ -1,16 +1,17 @@
-; RUN: opt < %s -passes=instcombine,mem2reg,simplifycfg -simplifycfg-require-and-preserve-domtree=1 | \
-; RUN:   llvm-dis | grep -v store | not grep "i32 1"
+; RUN: opt %s -passes=instcombine,mem2reg,simplifycfg -simplifycfg-require-and-preserve-domtree=1 -S \
+; RUN:   | grep -v store | not grep "i32 1"
+; RUN: opt %s -passes=instcombine,mem2reg,simplifycfg -simplifycfg-require-and-preserve-domtree=1 -S | FileCheck %s
 
 ; Test to make sure that instcombine does not accidentally propagate the load
 ; into the PHI, which would break the program.
 
-define i32 @test(i1 %C) {
+define i32 @test(i1 %C) !prof !0 {
 entry:
         %X = alloca i32         ; <ptr> [#uses=3]
         %X2 = alloca i32                ; <ptr> [#uses=2]
         store i32 1, ptr %X
         store i32 2, ptr %X2
-        br i1 %C, label %cond_true.i, label %cond_continue.i
+        br i1 %C, label %cond_true.i, label %cond_continue.i, !prof !1
 
 cond_true.i:            ; preds = %entry
         br label %cond_continue.i
@@ -22,4 +23,7 @@ cond_continue.i:                ; preds = %cond_true.i, %entry
         ret i32 %tmp.3
 }
 
+; CHECK: %spec.select = select i1 %C, ptr %X, ptr %X2, !prof !1
 
+!0 = !{!"function_entry_count", i64 1000}
+!1 = !{!"branch_weights", i32 2, i32 7}
\ No newline at end of file
diff --git a/llvm/utils/profcheck-xfail.txt b/llvm/utils/profcheck-xfail.txt
index 258e9bcd0b34f..0f2a380a58dcb 100644
--- a/llvm/utils/profcheck-xfail.txt
+++ b/llvm/utils/profcheck-xfail.txt
@@ -831,7 +831,6 @@ Transforms/IndVarSimplify/invalidate-modified-lcssa-phi.ll
 Transforms/IndVarSimplify/pr45835.ll
 Transforms/IndVarSimplify/preserving-debugloc-rem-div.ll
 Transforms/Inline/optimization-remarks-hotness-threshold.ll
-Transforms/InstCombine/2004-09-20-BadLoadCombine2.ll
 Transforms/InstCombine/2004-09-20-BadLoadCombine.ll
 Transforms/InstCombine/2005-04-07-UDivSelectCrash.ll
 Transforms/InstCombine/2011-02-14-InfLoop.ll

[nikic]

Can you replace this grep test with standard FileCheck + update_test_checks?

[mtrofin]

Can you replace this grep test with standard FileCheck + update_test_checks?

Good point! Done.

[nikic]

LGTM, but do you actually still need the prof metadata for your original motivation after this change? I guess the grep was previously matching the wrong thing, which should no longer be a problem. If it's not needed I'd drop it again.

[mtrofin]

True - it'd get checked by profcheck.