[PATCH] D37729: [X86] Make sure we still emit zext for GR32 to GR64 when the source of the zext is AssertZext

[Craig Topper via Phabricator via llvm-commits]

craig.topper updated this revision to Diff 114760.
craig.topper added a comment.

Forgot to add full context to the diff the first time.


https://reviews.llvm.org/D37729

Files:
  lib/Target/X86/X86InstrCompiler.td
  test/CodeGen/X86/TruncAssertZext.ll


Index: test/CodeGen/X86/TruncAssertZext.ll
===================================================================
--- /dev/null
+++ test/CodeGen/X86/TruncAssertZext.ll
@@ -0,0 +1,42 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -O2 -mtriple=x86_64-unknown-unknown | FileCheck %s
+; Checks that a zeroing mov is inserted for the trunc/zext pair even when
+; the source of the zext is an AssertZext node
+; PR28540
+
+define i64 @foo() {
+; CHECK-LABEL: foo:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    movq $-1, %rax
+; CHECK-NEXT:    retq
+  ret i64 -1
+}
+
+define i64 @main() {
+; CHECK-LABEL: main:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:  .Lcfi0:
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    callq foo
+; CHECK-NEXT:    movabsq $-4294967041, %rcx # imm = 0xFFFFFFFF000000FF
+; CHECK-NEXT:    andq %rax, %rcx
+; CHECK-NEXT:    movl %ecx, %ecx
+; CHECK-NEXT:    leaq (,%rcx,8), %rax
+; CHECK-NEXT:    subq %rcx, %rax
+; CHECK-NEXT:    shrq $32, %rax
+; CHECK-NEXT:    popq %rcx
+; CHECK-NEXT:    retq
+  %b = call i64 @foo()
+  %or = and i64 %b, 18446744069414584575 ; this is 0xffffffff000000ff
+  %trunc = trunc i64 %or to i32
+  br label %l
+l:
+  %ext = zext i32 %trunc to i64
+  %mul = mul i64 %ext, 7
+  br label %m
+m: ; keeps dag combine from seeing the multiply and the shift together
+  %shr = lshr i64 %mul, 32
+  trunc i64 %or to i32 ; keeps the and alive so it doesn't simplify
+  ret i64 %shr
+}
Index: lib/Target/X86/X86InstrCompiler.td
===================================================================
--- lib/Target/X86/X86InstrCompiler.td
+++ lib/Target/X86/X86InstrCompiler.td
@@ -1247,12 +1247,14 @@
 // Any instruction that defines a 32-bit result leaves the high half of the
 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
 // be copying from a truncate. Any other 32-bit operation will zero-extend
-// up to 64 bits.
+// up to 64 bits. AssertSext/AssertZext aren't saying anything about the upper
+// 32 bits, they're probably just qualifying a CopyFromReg.
 def def32 : PatLeaf<(i32 GR32:$src), [{
   return N->getOpcode() != ISD::TRUNCATE &&
          N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
          N->getOpcode() != ISD::CopyFromReg &&
-         N->getOpcode() != ISD::AssertSext;
+         N->getOpcode() != ISD::AssertSext &&
+         N->getOpcode() != ISD::AssertZext;
 }]>;
 
 // In the case of a 32-bit def that is known to implicitly zero-extend,


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D37729.114760.patch
Type: text/x-patch
Size: 2512 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170912/f1248082/attachment.bin>