[llvm] r246078 - AMDGPU: Fix not moving users of s_bfe_i64 to VALU

[Matt Arsenault via llvm-commits]

Author: arsenm
Date: Wed Aug 26 15:47:58 2015
New Revision: 246078

URL: http://llvm.org/viewvc/llvm-project?rev=246078&view=rev
Log:
AMDGPU: Fix not moving users of s_bfe_i64 to VALU

This wouldn't propagate to users of the original BFE
and would hit a verifier error.

Modified:
    llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.cpp
    llvm/trunk/test/CodeGen/AMDGPU/sext-in-reg.ll

Modified: llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.cpp?rev=246078&r1=246077&r2=246078&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/SIInstrInfo.cpp Wed Aug 26 15:47:58 2015
@@ -2568,6 +2568,7 @@ void SIInstrInfo::splitScalar64BitBFE(Sm
       .addImm(AMDGPU::sub1);
 
     MRI.replaceRegWith(Dest.getReg(), ResultReg);
+    addUsersToMoveToVALUWorklist(ResultReg, MRI, Worklist);
     return;
   }
 
@@ -2586,6 +2587,7 @@ void SIInstrInfo::splitScalar64BitBFE(Sm
     .addImm(AMDGPU::sub1);
 
   MRI.replaceRegWith(Dest.getReg(), ResultReg);
+  addUsersToMoveToVALUWorklist(ResultReg, MRI, Worklist);
 }
 
 void SIInstrInfo::addUsersToMoveToVALUWorklist(

Modified: llvm/trunk/test/CodeGen/AMDGPU/sext-in-reg.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/sext-in-reg.ll?rev=246078&r1=246077&r2=246078&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/sext-in-reg.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/sext-in-reg.ll Wed Aug 26 15:47:58 2015
@@ -609,3 +609,53 @@ define void @sext_in_reg_i2_bfe_offset_1
   store i32 %bfe, i32 addrspace(1)* %out, align 4
   ret void
 }
+
+; Make sure we propagate the VALUness to users of a moved scalar BFE.
+
+; FUNC-LABEL: {{^}}v_sext_in_reg_i1_to_i64_move_use:
+; SI: buffer_load_dwordx2
+; SI: v_lshl_b64 v{{\[}}[[VAL_LO:[0-9]+]]:[[VAL_HI:[0-9]+]]{{\]}}
+; SI-DAG: v_bfe_i32 v[[LO:[0-9]+]], v[[VAL_LO]], 0, 1
+; SI-DAG: v_ashrrev_i32_e32 v[[HI:[0-9]+]], 31, v[[LO]]
+; SI-DAG: v_and_b32_e32 v[[RESULT_LO:[0-9]+]], s{{[0-9]+}}, v[[LO]]
+; SI-DAG: v_and_b32_e32 v[[RESULT_HI:[0-9]+]], s{{[0-9]+}}, v[[HI]]
+; SI: buffer_store_dwordx2 v{{\[}}[[RESULT_LO]]:[[RESULT_HI]]{{\]}}
+define void @v_sext_in_reg_i1_to_i64_move_use(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr, i64 %s.val) nounwind {
+  %tid = call i32 @llvm.r600.read.tidig.x()
+  %a.gep = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
+  %b.gep = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
+  %out.gep = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
+  %a = load i64, i64 addrspace(1)* %a.gep, align 8
+  %b = load i64, i64 addrspace(1)* %b.gep, align 8
+
+  %c = shl i64 %a, %b
+  %shl = shl i64 %c, 63
+  %ashr = ashr i64 %shl, 63
+
+  %and = and i64 %ashr, %s.val
+  store i64 %and, i64 addrspace(1)* %out.gep, align 8
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v_sext_in_reg_i32_to_i64_move_use:
+; SI: buffer_load_dwordx2
+; SI: v_lshl_b64 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}},
+; SI-DAG: v_ashrrev_i32_e32 v[[SHR:[0-9]+]], 31, v[[LO]]
+; SI-DAG: v_and_b32_e32 v[[RESULT_LO:[0-9]+]], s{{[0-9]+}}, v[[LO]]
+; SI-DAG: v_and_b32_e32 v[[RESULT_HI:[0-9]+]], s{{[0-9]+}}, v[[SHR]]
+; SI: buffer_store_dwordx2 v{{\[}}[[RESULT_LO]]:[[RESULT_HI]]{{\]}}
+define void @v_sext_in_reg_i32_to_i64_move_use(i64 addrspace(1)* %out, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr, i64 %s.val) nounwind {
+  %tid = call i32 @llvm.r600.read.tidig.x()
+  %a.gep = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
+  %b.gep = getelementptr i64, i64 addrspace(1)* %aptr, i32 %tid
+  %out.gep = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
+  %a = load i64, i64 addrspace(1)* %a.gep, align 8
+  %b = load i64, i64 addrspace(1)* %b.gep, align 8
+
+  %c = shl i64 %a, %b
+  %shl = shl i64 %c, 32
+  %ashr = ashr i64 %shl, 32
+  %and = and i64 %ashr, %s.val
+  store i64 %and, i64 addrspace(1)* %out.gep, align 8
+  ret void
+}