[PATCH] D89701: Fix TypeSize warning in redundant store elimination

Peter Waller via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Oct 19 07:30:43 PDT 2020


peterwaller-arm created this revision.
peterwaller-arm added reviewers: sdesmalen, david-arm, niravd, craig.topper, efriedma.
Herald added subscribers: llvm-commits, ecnelises, pengfei, hiraditya.
Herald added a project: LLVM.
peterwaller-arm requested review of this revision.

The modified code in visitSTORE was missing a scalable vector check, and still
using the now deprecated implicit cast of TypeSize to uint64_t through the
overloaded operator. This patch fixes these issues.

This brings the logic in line with the comment on the context line immediately
above the added precondition.

Add a test in Redundantstores.ll that the warning is not triggered.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D89701

Files:
  llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
  llvm/test/CodeGen/AArch64/Redundantstore.ll


Index: llvm/test/CodeGen/AArch64/Redundantstore.ll
===================================================================
--- llvm/test/CodeGen/AArch64/Redundantstore.ll
+++ llvm/test/CodeGen/AArch64/Redundantstore.ll
@@ -1,8 +1,11 @@
-; RUN: llc < %s -O3 -mtriple=aarch64-eabi | FileCheck %s 
+; RUN: llc < %s -O3 -mtriple=aarch64-eabi 2>&1 | FileCheck %s
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 @end_of_array = common global i8* null, align 8
 
+; The tests in this file should not produce a TypeSize warning.
+; CHECK-NOT: warning: {{.*}}TypeSize is not scalable
+
 ; CHECK-LABEL: @test
 ; CHECK: stur
 ; CHECK-NOT: stur
@@ -23,3 +26,23 @@
   ret i8* %0
 }
 
+; #include <arm_sve.h>
+; #include <stdint.h>
+;
+; void redundant_store(uint32_t *x) {
+;     *x = 1;
+;     *(svint32_t *)x = svdup_s32(0);
+; }
+
+; CHECK-LABEL: @redundant_store
+define void @redundant_store(i32* nocapture %x) local_unnamed_addr #0 {
+  %1 = bitcast i32* %x to <vscale x 4 x i32>*
+  store i32 1, i32* %x, align 4
+  %2 = tail call <vscale x 4 x i32> @llvm.aarch64.sve.dup.x.nxv4i32(i32 0)
+  store <vscale x 4 x i32> %2, <vscale x 4 x i32>* %1, align 16
+  ret void
+}
+
+declare <vscale x 4 x i32> @llvm.aarch64.sve.dup.x.nxv4i32(i32)
+
+attributes #0 = { "target-cpu"="generic" "target-features"="+neon,+sve,+v8.2a" }
Index: llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -17325,11 +17325,12 @@
           !ST1->getBasePtr().isUndef() &&
           // BaseIndexOffset and the code below requires knowing the size
           // of a vector, so bail out if MemoryVT is scalable.
+          !ST->getMemoryVT().isScalableVector() &&
           !ST1->getMemoryVT().isScalableVector()) {
         const BaseIndexOffset STBase = BaseIndexOffset::match(ST, DAG);
         const BaseIndexOffset ChainBase = BaseIndexOffset::match(ST1, DAG);
-        unsigned STBitSize = ST->getMemoryVT().getSizeInBits();
-        unsigned ChainBitSize = ST1->getMemoryVT().getSizeInBits();
+        unsigned STBitSize = ST->getMemoryVT().getFixedSizeInBits();
+        unsigned ChainBitSize = ST1->getMemoryVT().getFixedSizeInBits();
         // If this is a store who's preceding store to a subset of the current
         // location and no one other node is chained to that store we can
         // effectively drop the store. Do not remove stores to undef as they may


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D89701.299046.patch
Type: text/x-patch
Size: 2525 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20201019/1360113f/attachment.bin>


More information about the llvm-commits mailing list