[Beignet] [PATCH] Backend: for BDW and after, According to BSpec no need to split CMP when src is DW DF

[rander]

Signed-off-by: rander <rander.wang at intel.com>
---
 backend/src/backend/gen8_encoder.cpp | 30 ++++++++++++++++++++++++++++++
 backend/src/backend/gen8_encoder.hpp |  1 +
 backend/src/backend/gen_encoder.cpp  | 10 +++++-----
 backend/src/backend/gen_encoder.hpp  |  1 +
 4 files changed, 37 insertions(+), 5 deletions(-)

diff --git a/backend/src/backend/gen8_encoder.cpp b/backend/src/backend/gen8_encoder.cpp
index a33fbac..31ae6d1 100644
--- a/backend/src/backend/gen8_encoder.cpp
+++ b/backend/src/backend/gen8_encoder.cpp
@@ -37,7 +37,13 @@ static const uint32_t untypedRWMask[] = {
 
 namespace gbe
 {
+  extern bool compactAlu2(GenEncoder *p, uint32_t opcode, GenRegister dst, GenRegister src0, GenRegister src1, uint32_t condition, bool split);
   extern bool compactAlu3(GenEncoder *p, uint32_t opcode, GenRegister dst, GenRegister src0, GenRegister src1, GenRegister src2);
+  extern bool isSrcDstDiffSpan(GenRegister dst, GenRegister src);
+  extern bool isCrossMoreThan2(GenRegister reg);
+  extern bool isVectorOfLongs(GenRegister reg);
+  bool isVectorOfBytes(GenRegister reg);
+
   void Gen8Encoder::setHeader(GenNativeInstruction *insn) {
     Gen8NativeInstruction *gen8_insn = &insn->gen8_insn;
     if (this->curr.execWidth == 8)
@@ -883,4 +889,28 @@ namespace gbe
                    msg_length,
                    response_length);
    }
+
+    /* for BDW and after, no need to split CMP when src is DW DF*/
+      bool Gen8Encoder::needToSplitCmp(GenEncoder *p, GenRegister src0, GenRegister src1, GenRegister dst) {
+      if (p->curr.execWidth != 16) return false;
+      if (isVectorOfLongs(dst) == true) return true;
+      if (isCrossMoreThan2(dst) == true) return true;
+
+      if (src0.hstride == GEN_HORIZONTAL_STRIDE_0 &&
+              src1.hstride == GEN_HORIZONTAL_STRIDE_0)
+        return false;
+
+      if (isVectorOfBytes(src0) == true) return true;
+      if (isVectorOfBytes(src1) == true) return true;
+
+      if (isVectorOfLongs(src0) == true) return true;
+      if (isVectorOfLongs(src1) == true) return true;
+      if (isCrossMoreThan2(src0) == true) return true;
+      if (isCrossMoreThan2(src1) == true) return true;
+
+      if (isSrcDstDiffSpan(dst, src0) == true) return true;
+      if (isSrcDstDiffSpan(dst, src1) == true) return true;
+
+      return false;
+    }
 } /* End of the name space. */
diff --git a/backend/src/backend/gen8_encoder.hpp b/backend/src/backend/gen8_encoder.hpp
index fa62a8d..c1df3b1 100644
--- a/backend/src/backend/gen8_encoder.hpp
+++ b/backend/src/backend/gen8_encoder.hpp
@@ -83,6 +83,7 @@ namespace gbe
     virtual void OBREADA64(GenRegister dst, GenRegister header, uint32_t bti, uint32_t elemSize);
     /*! A64 OBlock write */
     virtual void OBWRITEA64(GenRegister header, uint32_t bti, uint32_t elemSize);
+    virtual bool needToSplitCmp(GenEncoder *p, GenRegister src0, GenRegister src1, GenRegister dst);
   };
 }
 #endif /* __GBE_GEN8_ENCODER_HPP__ */
diff --git a/backend/src/backend/gen_encoder.cpp b/backend/src/backend/gen_encoder.cpp
index 03ce0e2..fbc443b 100644
--- a/backend/src/backend/gen_encoder.cpp
+++ b/backend/src/backend/gen_encoder.cpp
@@ -59,7 +59,7 @@ namespace gbe
   //////////////////////////////////////////////////////////////////////////
   // Some helper functions to encode
   //////////////////////////////////////////////////////////////////////////
-  INLINE bool isVectorOfBytes(GenRegister reg) {
+  bool isVectorOfBytes(GenRegister reg) {
     if (reg.hstride != GEN_HORIZONTAL_STRIDE_0 &&
         (reg.type == GEN_TYPE_UB || reg.type == GEN_TYPE_B))
       return true;
@@ -67,7 +67,7 @@ namespace gbe
       return false;
   }
 
-  INLINE bool isVectorOfLongs(GenRegister reg) {
+  bool isVectorOfLongs(GenRegister reg) {
     if (reg.hstride != GEN_HORIZONTAL_STRIDE_0 &&
         (reg.type == GEN_TYPE_UL || reg.type == GEN_TYPE_L))
       return true;
@@ -75,7 +75,7 @@ namespace gbe
       return false;
   }
 
-  INLINE bool isCrossMoreThan2(GenRegister reg) {
+  bool isCrossMoreThan2(GenRegister reg) {
     if (reg.hstride == GEN_HORIZONTAL_STRIDE_0)
       return false;
 
@@ -87,7 +87,7 @@ namespace gbe
     return false;
   }
 
-  INLINE bool isSrcDstDiffSpan(GenRegister dst, GenRegister src) {
+   bool isSrcDstDiffSpan(GenRegister dst, GenRegister src) {
     if (src.hstride == GEN_HORIZONTAL_STRIDE_0) return false;
 
     GBE_ASSERT(dst.hstride != GEN_HORIZONTAL_STRIDE_0 && "dst register is uniform but src is not.");
@@ -172,7 +172,7 @@ namespace gbe
     return false;
   }
 
-  INLINE bool needToSplitCmp(GenEncoder *p, GenRegister src0, GenRegister src1, GenRegister dst) {
+  bool GenEncoder::needToSplitCmp(GenEncoder *p, GenRegister src0, GenRegister src1, GenRegister dst) {
     if (p->curr.execWidth != 16) return false;
     if (isVectorOfLongs(dst) == true) return true;
     if (isCrossMoreThan2(dst) == true) return true;
diff --git a/backend/src/backend/gen_encoder.hpp b/backend/src/backend/gen_encoder.hpp
index 3e45c81..b8c7a91 100644
--- a/backend/src/backend/gen_encoder.hpp
+++ b/backend/src/backend/gen_encoder.hpp
@@ -162,6 +162,7 @@ namespace gbe
     void BRD(GenRegister src);
     /*! Compare instructions */
     void CMP(uint32_t conditional, GenRegister src0, GenRegister src1, GenRegister dst = GenRegister::null());
+    virtual bool needToSplitCmp(GenEncoder *p, GenRegister src0, GenRegister src1, GenRegister dst);
     /*! Select with embedded compare (like sel.le ...) */
     void SEL_CMP(uint32_t conditional, GenRegister dst, GenRegister src0, GenRegister src1);
     /*! EOT is used to finish GPGPU threads */
-- 
2.7.4