[llvm-branch-commits] [llvm] [SLP] Price gathered lanes of vectorized entries as a shuffle (PR #240136)

Alexey Bataev via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Sun Nov 8 05:24:53 PST 2026


================
@@ -19,31 +19,30 @@ define void @test(ptr %srcGrid, ptr %dstGrid, i64 %iv) {
 ; CHECK-NEXT:    [[ARRAYIDX148:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 64
 ; CHECK-NEXT:    [[ARRAYIDX184:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 136
 ; CHECK-NEXT:    [[ARRAYIDX188:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 144
-; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[ARRAYIDX184]], align 8
-; CHECK-NEXT:    [[TMP4:%.*]] = load double, ptr [[ARRAYIDX188]], align 8
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> <double poison, double 0.000000e+00>, <2 x i32> <i32 0, i32 3>
-; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <2 x double> <double poison, double 0.000000e+00>, <2 x double> [[TMP1]], <2 x i32> <i32 3, i32 1>
-; CHECK-NEXT:    [[TMP5:%.*]] = fadd fast <2 x double> [[TMP3]], [[TMP9]]
-; CHECK-NEXT:    [[TMP6:%.*]] = fsub fast <2 x double> zeroinitializer, [[TMP5]]
-; CHECK-NEXT:    [[TMP7:%.*]] = fadd fast double 0.000000e+00, [[TMP4]]
-; CHECK-NEXT:    [[MOMENTUM_Z:%.*]] = fsub fast double 0.000000e+00, [[TMP7]]
 ; CHECK-NEXT:    [[ARRAYIDX675:%.*]] = getelementptr inbounds nuw [8 x i8], ptr [[DSTGRID]], i64 [[IV]]
 ; CHECK-NEXT:    [[ARRAYIDX787:%.*]] = getelementptr inbounds nuw i8, ptr [[ARRAYIDX675]], i64 32216
 ; CHECK-NEXT:    [[TMP8:%.*]] = load double, ptr [[ARRAYIDX148]], align 8
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[ARRAYIDX144]], align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = load double, ptr [[ARRAYIDX184]], align 8
----------------
alexey-bataev wrote:

Regression?

https://github.com/llvm/llvm-project/pull/240136


More information about the llvm-branch-commits mailing list