[PATCH] D65557: AMDGPU/LoadStoreOptimizer: Set the correct offset whem merging MMOs

[Tom Stellard via Phabricator via llvm-commits]

tstellar created this revision.
tstellar added a reviewer: arsenm.
Herald added subscribers: hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, wdng, jvesely, kzhuravl.
Herald added a project: LLVM.

This is a follow up to r367237.  MachineFunction::getMachineMemOperand()
adds the offset parameter to the existing offset instead of resetting it.
So we need to reset the offset to the correct value after calling this
function.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D65557

Files:
  llvm/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
  llvm/test/CodeGen/AMDGPU/merge-load-store.mir


Index: llvm/test/CodeGen/AMDGPU/merge-load-store.mir
===================================================================
--- llvm/test/CodeGen/AMDGPU/merge-load-store.mir
+++ llvm/test/CodeGen/AMDGPU/merge-load-store.mir
@@ -65,7 +65,8 @@
 
   attributes #0 = { convergent nounwind }
 
-  define amdgpu_kernel void @merge_mmos() { ret void }
+  define amdgpu_kernel void @merge_mmos(i32 addrspace(1)* %ptr_addr1) { ret void }
+
 ...
 ---
 name:            mem_dependency
@@ -170,6 +171,8 @@
 # CHECK: S_BUFFER_LOAD_DWORDX2_IMM %0, 0, 0, 0 :: (dereferenceable invariant load 8, align 4)
 # CHECK: BUFFER_LOAD_DWORDX2_OFFSET %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8, align 4)
 # CHECK: BUFFER_STORE_DWORDX2_OFFSET_exact killed %{{[0-9]+}}, %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 8, align 4)
+# CHECK: BUFFER_LOAD_DWORDX2_OFFSET %0, 0, 64, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8 from %ir.ptr_addr1 + 64, align 4
+# CHECK: BUFFER_STORE_DWORDX2_OFFSET_exact killed %{{[0-9]+}}, %0, 0, 64, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 8 into %ir.ptr_addr1 + 64, align 4
 name: merge_mmos
 tracksRegLiveness: true
 body: |
@@ -183,6 +186,11 @@
     %4:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %0, 0, 4, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4)
     BUFFER_STORE_DWORD_OFFSET_exact %3, %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4)
     BUFFER_STORE_DWORD_OFFSET_exact %4, %0, 0, 4, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4)
+    %5:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %0, 0, 64, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from %ir.ptr_addr1 + 64)
+    %6:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %0, 0, 68, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from %ir.ptr_addr1 + 68)
+    BUFFER_STORE_DWORD_OFFSET_exact %5, %0, 0, 64, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into %ir.ptr_addr1 + 64)
+    BUFFER_STORE_DWORD_OFFSET_exact %6, %0, 0, 68, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into %ir.ptr_addr1 + 68)
+
     S_ENDPGM 0
 
 ...
Index: llvm/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
+++ llvm/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
@@ -392,7 +392,12 @@
                                                    const MachineMemOperand *B) {
   unsigned MinOffset = std::min(A->getOffset(), B->getOffset());
   unsigned Size = A->getSize() + B->getSize();
-  return MF.getMachineMemOperand(A, MinOffset, Size);
+  // This function adds the offset parameter to the existing offset for A,
+  // so we pass 0 here as the offset and then manually set it to the correct
+  // value after the call.
+  MachineMemOperand *MMO = MF.getMachineMemOperand(A, 0, Size);
+  MMO->setOffset(MinOffset);
+  return MMO;
 }
 
 bool SILoadStoreOptimizer::offsetsCanBeCombined(CombineInfo &CI, bool Modify) {


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D65557.212727.patch
Type: text/x-patch
Size: 2961 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190801/33ca06ee/attachment.bin>