[PATCH] D48067: [InstCombine] Replacing X86-specific rounding intrinsics with generic floor-ceil

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added inline comments.


================
Comment at: llvm/test/Transforms/InstCombine/X86/x86-avx.ll:34
+
+define <4 x double> @test_round_pd_deil(<4 x double> %a) {
+; CHECK-LABEL: @test_round_pd_deil(
----------------
ceil, not deil


================
Comment at: llvm/test/Transforms/InstCombine/X86/x86-sse41.ll:7
 ; CHECK-LABEL: @test_round_sd(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> %a, <2 x double> %b, i32 10)
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> [[A:%.*]], <2 x double> [[B:%.*]], i32 10)
 ; CHECK-NEXT:    ret <2 x double> [[TMP1]]
----------------
Can you regenerate this file right now in trunk (and commit), so this noise is gone?


https://reviews.llvm.org/D48067