[llvm] r289231 - AMDGPU: Fix i128 mul

[Matt Arsenault via llvm-commits]

Author: arsenm
Date: Fri Dec  9 11:49:14 2016
New Revision: 289231

URL: http://llvm.org/viewvc/llvm-project?rev=289231&view=rev
Log:
AMDGPU: Fix i128 mul

Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
    llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
    llvm/trunk/test/CodeGen/AMDGPU/mul.ll

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp?rev=289231&r1=289230&r2=289231&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp Fri Dec  9 11:49:14 2016
@@ -2205,7 +2205,7 @@ void DAGTypeLegalizer::ExpandIntRes_MUL(
   else if (VT == MVT::i128)
     LC = RTLIB::MUL_I128;
 
-  if (LC == RTLIB::UNKNOWN_LIBCALL) {
+  if (LC == RTLIB::UNKNOWN_LIBCALL || !TLI.getLibcallName(LC)) {
     // We'll expand the multiplication by brute force because we have no other
     // options. This is a trivially-generalized version of the code from
     // Hacker's Delight (itself derived from Knuth's Algorithm M from section

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp?rev=289231&r1=289230&r2=289231&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp Fri Dec  9 11:49:14 2016
@@ -435,6 +435,10 @@ AMDGPUTargetLowering::AMDGPUTargetLoweri
   setOperationAction(ISD::SELECT, MVT::v4f32, Promote);
   AddPromotedToType(ISD::SELECT, MVT::v4f32, MVT::v4i32);
 
+  // There are no libcalls of any kind.
+  for (int I = 0; I < RTLIB::UNKNOWN_LIBCALL; ++I)
+    setLibcallName(static_cast<RTLIB::Libcall>(I), nullptr);
+
   setBooleanContents(ZeroOrNegativeOneBooleanContent);
   setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
 

Modified: llvm/trunk/test/CodeGen/AMDGPU/mul.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/mul.ll?rev=289231&r1=289230&r2=289231&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/mul.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/mul.ll Fri Dec  9 11:49:14 2016
@@ -198,3 +198,74 @@ endif:
   store i64 %3, i64 addrspace(1)* %out
   ret void
 }
+
+; FIXME: Load dwordx4
+; FUNC-LABEL: {{^}}s_mul_i128:
+; SI: s_load_dwordx2
+; SI: s_load_dwordx2
+; SI: s_load_dwordx2
+; SI: s_load_dwordx2
+
+; SI: v_mul_hi_u32
+; SI: v_mul_hi_u32
+; SI: s_mul_i32
+; SI: v_mul_hi_u32
+; SI: s_mul_i32
+; SI: s_mul_i32
+; SI: v_mul_hi_u32
+; SI: v_mul_hi_u32
+; SI: s_mul_i32
+; SI-DAG: s_mul_i32
+; SI-DAG: v_mul_hi_u32
+; SI: s_mul_i32
+; SI: s_mul_i32
+; SI: s_mul_i32
+; SI: s_mul_i32
+; SI: s_mul_i32
+
+; SI: buffer_store_dwordx4
+define void @s_mul_i128(i128 addrspace(1)* %out, i128 %a, i128 %b) nounwind #0 {
+  %mul = mul i128 %a, %b
+  store i128 %mul, i128 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v_mul_i128:
+; SI: {{buffer|flat}}_load_dwordx4
+; SI: {{buffer|flat}}_load_dwordx4
+
+; SI: v_mul_lo_i32
+; SI: v_mul_hi_u32
+; SI: v_mul_hi_u32
+; SI: v_mul_lo_i32
+; SI: v_mul_hi_u32
+; SI: v_mul_hi_u32
+; SI: v_mul_lo_i32
+; SI: v_mul_lo_i32
+; SI: v_add_i32_e32
+; SI: v_mul_hi_u32
+; SI: v_mul_lo_i32
+; SI: v_mul_hi_u32
+; SI: v_mul_lo_i32
+; SI: v_mul_lo_i32
+; SI: v_mul_lo_i32
+; SI: v_mul_lo_i32
+; SI: v_mul_lo_i32
+
+; SI: {{buffer|flat}}_store_dwordx4
+define void @v_mul_i128(i128 addrspace(1)* %out, i128 addrspace(1)* %aptr, i128 addrspace(1)* %bptr) #0 {
+  %tid = call i32 @llvm.r600.read.tidig.x()
+  %gep.a = getelementptr inbounds i128, i128 addrspace(1)* %aptr, i32 %tid
+  %gep.b = getelementptr inbounds i128, i128 addrspace(1)* %bptr, i32 %tid
+  %gep.out = getelementptr inbounds i128, i128 addrspace(1)* %bptr, i32 %tid
+  %a = load i128, i128 addrspace(1)* %gep.a
+  %b = load i128, i128 addrspace(1)* %gep.b
+  %mul = mul i128 %a, %b
+  store i128 %mul, i128 addrspace(1)* %gep.out
+  ret void
+}
+
+declare i32 @llvm.r600.read.tidig.x() #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readnone}