[PATCH] D61313: [AMDGPU] detect WaW hazards when moving/merging load/store instructions

[Rhys Perry via Phabricator via llvm-commits]

This revision was automatically updated to reflect the committed changes.
Closed by commit rL361008: [AMDGPU] detect WaW hazards when moving/merging load/store instructions (authored by pendingchaos, committed by ).
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

Changed prior to commit:
  https://reviews.llvm.org/D61313?vs=199761&id=200003#toc

Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61313/new/

https://reviews.llvm.org/D61313

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


Index: llvm/trunk/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
+++ llvm/trunk/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
@@ -281,6 +281,7 @@
     // registers are in SSA form.
     if (Use.isReg() &&
         ((Use.readsReg() && RegDefs.count(Use.getReg())) ||
+         (Use.isDef() && RegDefs.count(Use.getReg())) ||
          (Use.isDef() && TargetRegisterInfo::isPhysicalRegister(Use.getReg()) &&
           PhysRegUses.count(Use.getReg())))) {
       Insts.push_back(&MI);
Index: llvm/trunk/test/CodeGen/AMDGPU/merge-load-store.mir
===================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/merge-load-store.mir
+++ llvm/trunk/test/CodeGen/AMDGPU/merge-load-store.mir
@@ -59,6 +59,11 @@
   attributes #0 = { convergent nounwind }
   attributes #1 = { convergent nounwind readnone }
 
+  define amdgpu_kernel void @move_waw_hazards() #0 {
+    ret void
+  }
+
+  attributes #0 = { convergent nounwind }
 ...
 ---
 name:            mem_dependency
@@ -129,3 +134,32 @@
     S_SETPC_B64_return undef $sgpr30_sgpr31, implicit %6, implicit %7
 
 ...
+---
+# Make sure Write-after-Write hazards are correctly detected and the
+# instructions moved accordingly.
+# operations.
+# CHECK-LABEL: name: move_waw_hazards
+# CHECK: S_AND_B64
+# CHECK: S_CMP_EQ_U32
+name:            move_waw_hazards
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+
+    %3:sgpr_64 = COPY $sgpr0_sgpr1
+    %6:sreg_32_xm0_xexec = S_MOV_B32 0
+    %7:sreg_32_xm0 = S_MOV_B32 0
+    %8:sreg_64_xexec = REG_SEQUENCE killed %6, %subreg.sub0, %7, %subreg.sub1
+    %9:sreg_128 = S_LOAD_DWORDX4_IMM killed %8, 0, 0, 0 :: (invariant load 16, addrspace 6)
+    %31:sreg_64_xexec = S_BUFFER_LOAD_DWORDX2_IMM %9, 0, 0, 0 :: (dereferenceable invariant load 4)
+    %10:sreg_32_xm0_xexec = COPY %31.sub0
+    %11:sreg_32_xm0_xexec = COPY killed %31.sub1
+    %12:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM %9, 2, 0, 0 :: (dereferenceable invariant load 4)
+    %13:sreg_64 = V_CMP_NE_U32_e64 killed %11, 0, implicit $exec
+    %15:sreg_64 = V_CMP_NE_U32_e64 killed %12, 0, implicit $exec
+    %17:sreg_64_xexec = S_AND_B64 killed %13, killed %15, implicit-def dead $scc
+    S_CMP_EQ_U32 killed %10, 0, implicit-def $scc
+    %18:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM %9, 3, 0, 0 :: (dereferenceable invariant load 4)
+    S_ENDPGM 0
+...


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D61313.200003.patch
Type: text/x-patch
Size: 2511 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190517/f7b5816b/attachment.bin>