[PATCH] D78480: [SelectionDAG] Fix legalization of non-byte-sized vector

[LemonBoy via Phabricator via llvm-commits]

LemonBoy updated this revision to Diff 259844.
LemonBoy added a comment.

Something less invasive that doesn't affect the codegen while fixing the problem.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D78480/new/

https://reviews.llvm.org/D78480

Files:
  llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
  llvm/test/CodeGen/X86/load-local-v4i5.ll


Index: llvm/test/CodeGen/X86/load-local-v4i5.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/X86/load-local-v4i5.ll
@@ -0,0 +1,77 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
+ at 0 = internal unnamed_addr constant [4 x i5] [i5 2, i5 0, i5 2, i5 -1], align 1
+
+; Function Attrs: nobuiltin nounwind
+define void @_start() {
+; CHECK-LABEL: _start:
+; CHECK:       # %bb.0: # %Entry
+; CHECK-NEXT:    movl {{.*}}(%rip), %eax
+; CHECK-NEXT:    movl %eax, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movb -{{[0-9]+}}(%rsp), %cl
+; CHECK-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edx
+; CHECK-NEXT:    movzbl -{{[0-9]+}}(%rsp), %esi
+; CHECK-NEXT:    andl $31, %eax
+; CHECK-NEXT:    andl $31, %esi
+; CHECK-NEXT:    shll $5, %esi
+; CHECK-NEXT:    orl %eax, %esi
+; CHECK-NEXT:    andl $31, %edx
+; CHECK-NEXT:    shll $10, %edx
+; CHECK-NEXT:    orl %esi, %edx
+; CHECK-NEXT:    movzbl %cl, %eax
+; CHECK-NEXT:    movl %eax, %ecx
+; CHECK-NEXT:    shll $15, %ecx
+; CHECK-NEXT:    orl %edx, %ecx
+; CHECK-NEXT:    movw %cx, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    shrl $16, %ecx
+; CHECK-NEXT:    andl $15, %ecx
+; CHECK-NEXT:    movb %cl, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movb %al, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    cmpb $31, %al
+; CHECK-NEXT:    je .LBB0_2
+; CHECK-NEXT:  # %bb.1: # %Then
+; CHECK-NEXT:    int3
+; CHECK-NEXT:  .LBB0_2: # %EndIf
+; CHECK-NEXT:    retq
+Entry:
+  %x = alloca [4 x i5], align 1
+  %y = alloca <4 x i5>, align 4
+  %z = alloca i5, align 1
+  %0 = bitcast [4 x i5]* %x to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 %0, i8* align 1 bitcast ([4 x i5]* @0 to i8*), i64 4, i1 false)
+  %1 = getelementptr inbounds [4 x i5], [4 x i5]* %x, i64 0, i64 0
+  %2 = load i5, i5* %1
+  %3 = insertelement <4 x i5> undef, i5 %2, i32 0
+  %4 = getelementptr inbounds [4 x i5], [4 x i5]* %x, i64 0, i64 1
+  %5 = load i5, i5* %4
+  %6 = insertelement <4 x i5> %3, i5 %5, i32 1
+  %7 = getelementptr inbounds [4 x i5], [4 x i5]* %x, i64 0, i64 2
+  %8 = load i5, i5* %7
+  %9 = insertelement <4 x i5> %6, i5 %8, i32 2
+  %10 = getelementptr inbounds [4 x i5], [4 x i5]* %x, i64 0, i64 3
+  %11 = load i5, i5* %10
+  %12 = insertelement <4 x i5> %9, i5 %11, i32 3
+  store <4 x i5> %12, <4 x i5>* %y, align 4
+  %13 = load <4 x i5>, <4 x i5>* %y
+  %14 = extractelement <4 x i5> %13, i32 3
+  store i5 %14, i5* %z, align 1
+  %15 = load i5, i5* %z, align 1
+  %16 = icmp ne i5 %15, -1
+  br i1 %16, label %Then, label %Else
+
+Then:                                             ; preds = %Entry
+  call void @llvm.debugtrap()
+  br label %EndIf
+
+Else:                                             ; preds = %Entry
+  br label %EndIf
+
+EndIf:                                            ; preds = %Else, %Then
+  ret void
+}
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* noalias nocapture writeonly, i8* noalias nocapture readonly, i64, i1 immarg)
+
+; Function Attrs: nounwind
+declare void @llvm.debugtrap()
Index: llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
+++ llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
@@ -747,11 +747,8 @@
             LD->getPointerInfo().getWithOffset(Offset), LD->getOriginalAlign(),
             LD->getMemOperand()->getFlags(), LD->getAAInfo());
       } else {
-        EVT LoadVT = WideVT;
-        while (RemainingBytes < LoadBytes) {
-          LoadBytes >>= 1; // Reduce the load size by half.
-          LoadVT = EVT::getIntegerVT(*DAG.getContext(), LoadBytes << 3);
-        }
+        LoadBytes = RemainingBytes;
+        EVT LoadVT = EVT::getIntegerVT(*DAG.getContext(), LoadBytes << 3);
         ScalarLoad =
             DAG.getExtLoad(ISD::EXTLOAD, dl, WideVT, Chain, BasePTR,
                            LD->getPointerInfo().getWithOffset(Offset), LoadVT,


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D78480.259844.patch
Type: text/x-patch
Size: 4060 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200424/a1ac46ec/attachment.bin>