[PATCH] D13601: AMDGPU: Fix verifier error in SIFoldOperands

[Matt Arsenault via llvm-commits]

arsenm created this revision.
arsenm added a reviewer: tstellarAMD.
arsenm added a subscriber: llvm-commits.
Herald added a subscriber: arsenm.

There may be other use operands that also need their kill flags cleared.
    
This happens in a few tests when SIFoldOperands is moved after
PeepholeOptimizer.
    
PeepholeOptimizer rewrites cases that look like:
%vreg0 = ...
%vreg1 = COPY %vreg0
use %vreg1<kill>
%vreg2 = COPY %vreg0
use %vreg2<kill>
    
to use the earlier source to
%vreg0 = ...
use %vreg0
use %vreg0
    
Currently SIFoldOperands sees the copied registers, so there is
only one use. So far I haven't managed to come up with a test
that currently has multiple uses of a foldable VGPR -> VGPR copy.

http://reviews.llvm.org/D13601

Files:
  lib/Target/AMDGPU/SIFoldOperands.cpp
  test/CodeGen/AMDGPU/si-instr-info-correct-implicit-operands.ll

Index: test/CodeGen/AMDGPU/si-instr-info-correct-implicit-operands.ll
===================================================================
--- test/CodeGen/AMDGPU/si-instr-info-correct-implicit-operands.ll
+++ test/CodeGen/AMDGPU/si-instr-info-correct-implicit-operands.ll
@@ -3,7 +3,7 @@
 ; register operands in the correct order when modifying the opcode of an
 ; instruction to V_ADD_I32_e32.
 
-; CHECK: %19 = V_ADD_I32_e32 killed %13, killed %12, implicit-def %vcc, implicit %exec
+; CHECK: %19 = V_ADD_I32_e32 %13, %12, implicit-def %vcc, implicit %exec
 
 define void @test(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
 entry:
Index: lib/Target/AMDGPU/SIFoldOperands.cpp
===================================================================
--- lib/Target/AMDGPU/SIFoldOperands.cpp
+++ lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -366,7 +366,10 @@
           // Clear kill flags.
           if (!Fold.isImm()) {
             assert(Fold.OpToFold && Fold.OpToFold->isReg());
-            Fold.OpToFold->setIsKill(false);
+            // FIXME: Probably shouldn't bother trying to fold if not an
+            // SGPR. PeepholeOptimizer can eliminate redundant VGPR->VGPR
+            // copies.
+            MRI.clearKillFlags(Fold.OpToFold->getReg());
           }
           DEBUG(dbgs() << "Folded source from " << MI << " into OpNo " <<
                 Fold.UseOpNo << " of " << *Fold.UseMI << '\n');


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D13601.36979.patch
Type: text/x-patch
Size: 1408 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151009/c3a0fb9d/attachment.bin>