[llvm-branch-commits] [llvm] [AArch64][DAG] Improve codegen for clearing upper vector lanes (PR #237259)
via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Tue Nov 3 06:46:09 PST 2026
================
@@ -1,483 +1,298 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
-; RUN: llc < %s -mtriple=aarch64 | FileCheck %s --check-prefixes=CHECK,CHECK-SD
-; RUN: llc < %s -mtriple=aarch64 -global-isel | FileCheck %s --check-prefixes=CHECK,CHECK-GI
+; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 -O2 | FileCheck %s --check-prefixes=CHECK,SD,SD-FP16
+; RUN: llc < %s -mtriple=aarch64 -O2 | FileCheck %s --check-prefixes=CHECK,SD,SD-NOFP16
+; RUN: llc < %s -mtriple=aarch64 -O2 -mattr=+fullfp16 -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK,GI,GI-FP16
+; RUN: llc < %s -mtriple=aarch64 -O2 -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK,GI,GI-NOFP16
+; The canonical form uses <i32 0, i32 poison, i32 poison, ... > for the zero vector operand
+; We use equivalent zeroinitializer instead which has no functional change
-define <2 x double> @testv2f64_0(<2 x double> noundef %vec) {
-; CHECK-SD-LABEL: testv2f64_0:
-; CHECK-SD: // %bb.0: // %entry
-; CHECK-SD-NEXT: mov v0.d[1], xzr
-; CHECK-SD-NEXT: ret
+define <2 x double> @testv2f64(<2 x double> noundef %vec) {
+; SD-LABEL: testv2f64:
+; SD: // %bb.0: // %entry
+; SD-NEXT: mov v0.d[1], xzr
+; SD-NEXT: ret
----------------
CarolineConcatto wrote:
why?
https://github.com/llvm/llvm-project/pull/237259
More information about the llvm-branch-commits
mailing list