[PATCH] D28784: AMDGPU: Check users in fneg/fabs select combine

[Matt Arsenault via Phabricator via llvm-commits]

arsenm created this revision.
Herald added a reviewer: tstellarAMD.
Herald added subscribers: tony-tye, yaxunl, nhaehnle, wdng, kzhuravl.

https://reviews.llvm.org/D28784

Files:
  lib/Target/AMDGPU/AMDGPUISelLowering.cpp
  test/CodeGen/AMDGPU/select-fabs-fneg-extract-legacy.ll
  test/CodeGen/AMDGPU/select-fabs-fneg-extract.ll


Index: test/CodeGen/AMDGPU/select-fabs-fneg-extract.ll
===================================================================
--- test/CodeGen/AMDGPU/select-fabs-fneg-extract.ll
+++ test/CodeGen/AMDGPU/select-fabs-fneg-extract.ll
@@ -810,14 +810,12 @@
   ret void
 }
 
-; FIXME: This one should fold to rcp
 ; GCN-LABEL: {{^}}select_fneg_posk_src_rcp_f32:
 ; GCN: buffer_load_dword [[X:v[0-9]+]]
 
-; GCN: v_rcp_f32_e32 [[RCP:v[0-9]+]], [[X]]
-; GCN: v_cndmask_b32_e32 [[SELECT:v[0-9]+]], -2.0, [[RCP]], vcc
-; GCN: v_xor_b32_e32 [[NEG_SELECT:v[0-9]+]], 0x80000000, [[SELECT]]
-; GCN-NEXT: buffer_store_dword [[NEG_SELECT]]
+; GCN: v_rcp_f32_e64 [[RCP:v[0-9]+]], -[[X]]
+; GCN: v_cndmask_b32_e32 [[SELECT:v[0-9]+]], 2.0, [[RCP]], vcc
+; GCN-NEXT: buffer_store_dword [[SELECT]]
 define void @select_fneg_posk_src_rcp_f32(i32 %c) #0 {
   %x = load volatile float, float addrspace(1)* undef
   %y = load volatile float, float addrspace(1)* undef
Index: test/CodeGen/AMDGPU/select-fabs-fneg-extract-legacy.ll
===================================================================
--- test/CodeGen/AMDGPU/select-fabs-fneg-extract-legacy.ll
+++ test/CodeGen/AMDGPU/select-fabs-fneg-extract-legacy.ll
@@ -7,10 +7,9 @@
 ; GCN-LABEL: {{^}}select_fneg_posk_src_rcp_legacy_f32:
 ; GCN: buffer_load_dword [[X:v[0-9]+]]
 
-; GCN: v_rcp_legacy_f32_e32 [[RCP:v[0-9]+]], [[X]]
-; GCN: v_cndmask_b32_e32 [[SELECT:v[0-9]+]], -2.0, [[RCP]], vcc
-; GCN: v_xor_b32_e32 [[NEG_SELECT:v[0-9]+]], 0x80000000, [[SELECT]]
-; GCN-NEXT: buffer_store_dword [[NEG_SELECT]]
+; GCN: v_rcp_legacy_f32_e64 [[RCP:v[0-9]+]], -[[X]]
+; GCN: v_cndmask_b32_e32 [[SELECT:v[0-9]+]], 2.0, [[RCP]], vcc
+; GCN-NEXT: buffer_store_dword [[SELECT]]
 define void @select_fneg_posk_src_rcp_legacy_f32(i32 %c) #2 {
   %x = load volatile float, float addrspace(1)* undef
   %y = load volatile float, float addrspace(1)* undef
@@ -25,10 +24,9 @@
 ; GCN-LABEL: {{^}}select_fneg_posk_src_mul_legacy_f32:
 ; GCN: buffer_load_dword [[X:v[0-9]+]]
 
-; GCN: v_mul_legacy_f32_e32 [[MUL:v[0-9]+]], 4.0, [[X]]
-; GCN: v_cndmask_b32_e32 [[SELECT:v[0-9]+]], -2.0, [[MUL]], vcc
-; GCN: v_xor_b32_e32 [[NEG_SELECT:v[0-9]+]], 0x80000000, [[SELECT]]
-; GCN-NEXT: buffer_store_dword [[NEG_SELECT]]
+; GCN: v_mul_legacy_f32_e32 [[MUL:v[0-9]+]], -4.0, [[X]]
+; GCN: v_cndmask_b32_e32 [[SELECT:v[0-9]+]], 2.0, [[MUL]], vcc
+; GCN-NEXT: buffer_store_dword [[SELECT]]
 define void @select_fneg_posk_src_mul_legacy_f32(i32 %c) #2 {
   %x = load volatile float, float addrspace(1)* undef
   %cmp = icmp eq i32 %c, 0
Index: lib/Target/AMDGPU/AMDGPUISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -2793,8 +2793,10 @@
   SDValue RHS = N.getOperand(2);
 
   EVT VT = N.getValueType();
-  if ((LHS.getOpcode() == ISD::FABS && RHS.getOpcode() == ISD::FABS) ||
-      (LHS.getOpcode() == ISD::FNEG && RHS.getOpcode() == ISD::FNEG)) {
+  if (((LHS.getOpcode() == ISD::FABS && RHS.getOpcode() == ISD::FABS) ||
+       (LHS.getOpcode() == ISD::FNEG && RHS.getOpcode() == ISD::FNEG)) &&
+      (!LHS.hasOneUse() || !RHS.hasOneUse() ||
+       allUsesHaveSourceMods(N.getNode()))) {
     return distributeOpThroughSelect(DCI, LHS.getOpcode(),
                                      SDLoc(N), Cond, LHS, RHS);
   }
@@ -2807,7 +2809,8 @@
 
   // TODO: Support vector constants.
   ConstantFPSDNode *CRHS = dyn_cast<ConstantFPSDNode>(RHS);
-  if ((LHS.getOpcode() == ISD::FNEG || LHS.getOpcode() == ISD::FABS) && CRHS) {
+  if (CRHS && (LHS.getOpcode() == ISD::FNEG || LHS.getOpcode() == ISD::FABS) &&
+      allUsesHaveSourceMods(N.getNode())) {
     SDLoc SL(N);
     // If one side is an fneg/fabs and the other is a constant, we can push the
     // fneg/fabs down. If it's an fabs, the constant needs to be non-negative.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D28784.84604.patch
Type: text/x-patch
Size: 3859 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170116/bf904569/attachment.bin>