[Beignet] [PATCH] Add FCmpInst ord support.

[Yang Rong]

Because gen do not support isorder direct, use (src0 == src0) && (src1 == src1).
BTW: can't use !unordered.

Signed-off-by: Yang Rong <rong.r.yang at intel.com>
---
 backend/src/backend/gen_insn_selection.cpp | 11 ++++++++++-
 backend/src/ir/instruction.cpp             |  1 +
 backend/src/ir/instruction.hpp             |  2 ++
 backend/src/ir/instruction.hxx             |  1 +
 backend/src/llvm/llvm_gen_backend.cpp      |  1 +
 5 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/backend/src/backend/gen_insn_selection.cpp b/backend/src/backend/gen_insn_selection.cpp
index 5a6b9fd..385fc36 100644
--- a/backend/src/backend/gen_insn_selection.cpp
+++ b/backend/src/backend/gen_insn_selection.cpp
@@ -145,7 +145,8 @@ namespace gbe
       case OP_GT: return GEN_CONDITIONAL_G;
       case OP_EQ: return GEN_CONDITIONAL_EQ;
       case OP_NE: return GEN_CONDITIONAL_NEQ;
-      default: NOT_SUPPORTED; return 0u;
+      case OP_ORD:  return -1u;
+      default: NOT_SUPPORTED; return -1u;
     };
   }
 
@@ -1955,6 +1956,7 @@ namespace gbe
       // OK, we merge the instructions
       const ir::CompareInstruction &cmpInsn = cast<CompareInstruction>(cmp->insn);
       const ir::Opcode opcode = cmpInsn.getOpcode();
+      if(opcode == OP_ORD) return false;
       const uint32_t genCmp = getGenCompare(opcode);
 
       // Like for regular selects, we need a temporary since we cannot predicate
@@ -2543,6 +2545,13 @@ namespace gbe
           for(int i=0; i<3; i++)
             tmp[i] = sel.selReg(sel.reg(FAMILY_DWORD));
           sel.I64CMP(genCmp, src0, src1, tmp);
+        } else if(opcode == OP_ORD) {
+          const GenRegister tmp = sel.selReg(sel.reg(FAMILY_BOOL), TYPE_BOOL);
+          const GenRegister dstReg = sel.selReg(dst, TYPE_BOOL);
+          sel.CMP(GEN_CONDITIONAL_EQ, src0, src0);
+          sel.curr.flagIndex = uint16_t(tmp.reg());
+          sel.CMP(GEN_CONDITIONAL_EQ, src1, src1);
+          sel.AND(dstReg, dstReg, tmp);
         } else
           sel.CMP(genCmp, src0, src1);
       sel.pop();
diff --git a/backend/src/ir/instruction.cpp b/backend/src/ir/instruction.cpp
index da20d43..c624d7a 100644
--- a/backend/src/ir/instruction.cpp
+++ b/backend/src/ir/instruction.cpp
@@ -1552,6 +1552,7 @@ DECL_MEM_FN(GetImageInfoInstruction, uint32_t, getInfoType(void), getInfoType())
   DECL_EMIT_FUNCTION(LT)
   DECL_EMIT_FUNCTION(GE)
   DECL_EMIT_FUNCTION(GT)
+  DECL_EMIT_FUNCTION(ORD)
 
 #undef DECL_EMIT_FUNCTION
 
diff --git a/backend/src/ir/instruction.hpp b/backend/src/ir/instruction.hpp
index b1afd42..b7eebc0 100644
--- a/backend/src/ir/instruction.hpp
+++ b/backend/src/ir/instruction.hpp
@@ -634,6 +634,8 @@ namespace ir {
   Instruction GE(Type type, Register dst, Register src0, Register src1);
   /*! ge.type dst src0 src1 */
   Instruction GT(Type type, Register dst, Register src0, Register src1);
+  /*! ord.type dst src0 src1 */
+  Instruction ORD(Type type, Register dst, Register src0, Register src1);
   /*! BITCAST.{dstType <- srcType} dst src */
   Instruction BITCAST(Type dstType, Type srcType, Tuple dst, Tuple src, uint8_t dstNum, uint8_t srcNum);
   /*! cvt.{dstType <- srcType} dst src */
diff --git a/backend/src/ir/instruction.hxx b/backend/src/ir/instruction.hxx
index 83ecd1d..cf79e09 100644
--- a/backend/src/ir/instruction.hxx
+++ b/backend/src/ir/instruction.hxx
@@ -60,6 +60,7 @@ DECL_INSN(LE, CompareInstruction)
 DECL_INSN(LT, CompareInstruction)
 DECL_INSN(GE, CompareInstruction)
 DECL_INSN(GT, CompareInstruction)
+DECL_INSN(ORD, CompareInstruction)
 DECL_INSN(BITCAST, BitCastInstruction)
 DECL_INSN(CVT, ConvertInstruction)
 DECL_INSN(SAT_CVT, ConvertInstruction)
diff --git a/backend/src/llvm/llvm_gen_backend.cpp b/backend/src/llvm/llvm_gen_backend.cpp
index 23e5442..688c2b2 100644
--- a/backend/src/llvm/llvm_gen_backend.cpp
+++ b/backend/src/llvm/llvm_gen_backend.cpp
@@ -1596,6 +1596,7 @@ namespace gbe
       case ICmpInst::FCMP_ULT: ctx.LT(type, dst, src0, src1); break;
       case ICmpInst::FCMP_OGT:
       case ICmpInst::FCMP_UGT: ctx.GT(type, dst, src0, src1); break;
+      case ICmpInst::FCMP_ORD: ctx.ORD(type, dst, src0, src0); break;
       default: NOT_SUPPORTED;
     }
   }
-- 
1.8.1.2