[MachineCopyPropagation] In isNopCopy, check the destination registers match in addition to the source registers.

Previously if the source match we asserted that the destination
matched. But GPR <-> mask register copies on X86 can violate this
since we use the same K-registers for multiple sizes.

Fixes this ISPC issue https://github.com/ispc/ispc/issues/1851

Differential Revision: https://reviews.llvm.org/D86507

2 files changed, 60 insertions, 3 deletions

diff --git a/llvm/lib/CodeGen/MachineCopyPropagation.cpp b/llvm/lib/CodeGen/MachineCopyPropagation.cpp
index 67d15129b904..e2d7827c5a2f 100644
--- a/llvm/lib/CodeGen/MachineCopyPropagation.cpp
+++ b/llvm/lib/CodeGen/MachineCopyPropagation.cpp
@@ -338,10 +338,8 @@ static bool isNopCopy(const MachineInstr &PreviousCopy, unsigned Src,
                       unsigned Def, const TargetRegisterInfo *TRI) {
   Register PreviousSrc = PreviousCopy.getOperand(1).getReg();
   Register PreviousDef = PreviousCopy.getOperand(0).getReg();
-  if (Src == PreviousSrc) {
-    assert(Def == PreviousDef);
+  if (Src == PreviousSrc && Def == PreviousDef)
     return true;
-  }
   if (!TRI->isSubRegister(PreviousSrc, Src))
     return false;
   unsigned SubIdx = TRI->getSubRegIndex(PreviousSrc, Src);
diff --git a/llvm/test/CodeGen/X86/machine-cp-mask-reg.mir b/llvm/test/CodeGen/X86/machine-cp-mask-reg.mir
new file mode 100644
index 000000000000..86a077e64764
--- /dev/null
+++ b/llvm/test/CodeGen/X86/machine-cp-mask-reg.mir
@@ -0,0 +1,59 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc %s -mtriple=x86_64-unknown-linux-gnu -mcpu=skx -run-pass=machine-cp -o - | FileCheck %s
+
+# machine-cp previously asserted trying to determine if the k0->eax copy below
+# could be combined with the k0->rax copy.
+
+--- |
+  ; ModuleID = 'test.ll'
+  source_filename = "test.ll"
+  target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+
+  define i8 @foo(<64 x i8> %x, i64* %y, i64 %z) #0 {
+    %a = icmp eq <64 x i8> %x, zeroinitializer
+    %b = bitcast <64 x i1> %a to i64
+    %c = add i64 %b, %z
+    store i64 %c, i64* %y, align 8
+    %d = extractelement <64 x i1> %a, i32 0
+    %e = zext i1 %d to i8
+    ret i8 %e
+  }
+
+  attributes #0 = { "target-cpu"="skx" }
+
+...
+---
+name:            foo
+alignment:       16
+tracksRegLiveness: true
+liveins:
+  - { reg: '$zmm0' }
+  - { reg: '$rdi' }
+  - { reg: '$rsi' }
+frameInfo:
+  maxAlignment:    1
+machineFunctionInfo: {}
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: $rdi, $rsi, $zmm0
+
+    ; CHECK-LABEL: name: foo
+    ; CHECK: liveins: $rdi, $rsi, $zmm0
+    ; CHECK: renamable $k0 = VPTESTNMBZrr killed renamable $zmm0, renamable $zmm0
+    ; CHECK: renamable $rax = COPY renamable $k0
+    ; CHECK: renamable $rsi = ADD64rr killed renamable $rsi, killed renamable $rax, implicit-def dead $eflags
+    ; CHECK: MOV64mr killed renamable $rdi, 1, $noreg, 0, $noreg, killed renamable $rsi :: (store 8 into %ir.y)
+    ; CHECK: renamable $eax = COPY killed renamable $k0
+    ; CHECK: renamable $al = AND8ri renamable $al, 1, implicit-def dead $eflags, implicit killed $eax, implicit-def $eax
+    ; CHECK: $al = KILL renamable $al, implicit killed $eax
+    ; CHECK: RET 0, $al
+    renamable $k0 = VPTESTNMBZrr killed renamable $zmm0, renamable $zmm0
+    renamable $rax = COPY renamable $k0
+    renamable $rsi = ADD64rr killed renamable $rsi, killed renamable $rax, implicit-def dead $eflags
+    MOV64mr killed renamable $rdi, 1, $noreg, 0, $noreg, killed renamable $rsi :: (store 8 into %ir.y)
+    renamable $eax = COPY killed renamable $k0
+    renamable $al = AND8ri renamable $al, 1, implicit-def dead $eflags, implicit killed $eax, implicit-def $eax
+    $al = KILL renamable $al, implicit killed $eax
+    RET 0, $al
+
+...